1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright 2017 Joyent, Inc.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
34 */
35
36/*
37 * SPA: Storage Pool Allocator
38 *
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
41 * pool.
42 */
43
44#include <sys/zfs_context.h>
45#include <sys/fm/fs/zfs.h>
46#include <sys/spa_impl.h>
47#include <sys/zio.h>
48#include <sys/zio_checksum.h>
49#include <sys/dmu.h>
50#include <sys/dmu_tx.h>
51#include <sys/zap.h>
52#include <sys/zil.h>
53#include <sys/ddt.h>
54#include <sys/vdev_impl.h>
55#include <sys/vdev_removal.h>
56#include <sys/vdev_indirect_mapping.h>
57#include <sys/vdev_indirect_births.h>
58#include <sys/metaslab.h>
59#include <sys/metaslab_impl.h>
60#include <sys/uberblock_impl.h>
61#include <sys/txg.h>
62#include <sys/avl.h>
63#include <sys/bpobj.h>
64#include <sys/dmu_traverse.h>
65#include <sys/dmu_objset.h>
66#include <sys/unique.h>
67#include <sys/dsl_pool.h>
68#include <sys/dsl_dataset.h>
69#include <sys/dsl_dir.h>
70#include <sys/dsl_prop.h>
71#include <sys/dsl_synctask.h>
72#include <sys/fs/zfs.h>
73#include <sys/arc.h>
74#include <sys/callb.h>
75#include <sys/spa_boot.h>
76#include <sys/zfs_ioctl.h>
77#include <sys/dsl_scan.h>
78#include <sys/dmu_send.h>
79#include <sys/dsl_destroy.h>
80#include <sys/dsl_userhold.h>
81#include <sys/zfeature.h>
82#include <sys/zvol.h>
83#include <sys/trim_map.h>
84#include <sys/abd.h>
85
86#ifdef _KERNEL
87#include <sys/callb.h>
88#include <sys/cpupart.h>
89#include <sys/zone.h>
90#endif /* _KERNEL */
91
92#include "zfs_prop.h"
93#include "zfs_comutil.h"
94
95/* Check hostid on import? */
96static int check_hostid = 1;
97
98/*
99 * The interval, in seconds, at which failed configuration cache file writes
100 * should be retried.
101 */
102int zfs_ccw_retry_interval = 300;
103
104SYSCTL_DECL(_vfs_zfs);
105SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
106 "Check hostid on import?");
107TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
108SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
109 &zfs_ccw_retry_interval, 0,
110 "Configuration cache file write, retry after failure, interval (seconds)");
111
112typedef enum zti_modes {
113 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
114 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
115 ZTI_MODE_NULL, /* don't create a taskq */
116 ZTI_NMODES
117} zti_modes_t;
118
119#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
120#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
121#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
122
123#define ZTI_N(n) ZTI_P(n, 1)
124#define ZTI_ONE ZTI_N(1)
125
126typedef struct zio_taskq_info {
127 zti_modes_t zti_mode;
128 uint_t zti_value;
129 uint_t zti_count;
130} zio_taskq_info_t;
131
132static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
133 "issue", "issue_high", "intr", "intr_high"
134};
135
136/*
137 * This table defines the taskq settings for each ZFS I/O type. When
138 * initializing a pool, we use this table to create an appropriately sized
139 * taskq. Some operations are low volume and therefore have a small, static
140 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
141 * macros. Other operations process a large amount of data; the ZTI_BATCH
142 * macro causes us to create a taskq oriented for throughput. Some operations
143 * are so high frequency and short-lived that the taskq itself can become a a
144 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
145 * additional degree of parallelism specified by the number of threads per-
146 * taskq and the number of taskqs; when dispatching an event in this case, the
147 * particular taskq is chosen at random.
148 *
149 * The different taskq priorities are to handle the different contexts (issue
150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
151 * need to be handled with minimum delay.
152 */
153const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
156 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
157 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
158 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
161};
162
163static void spa_sync_version(void *arg, dmu_tx_t *tx);
164static void spa_sync_props(void *arg, dmu_tx_t *tx);
165static boolean_t spa_has_active_shared_spare(spa_t *spa);
166static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
167static void spa_vdev_resilver_done(spa_t *spa);
168
169uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
170#ifdef PSRSET_BIND
171id_t zio_taskq_psrset_bind = PS_NONE;
172#endif
173#ifdef SYSDC
174boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
175uint_t zio_taskq_basedc = 80; /* base duty cycle */
176#endif
177
178boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
179extern int zfs_sync_pass_deferred_free;
180
181/*
182 * Report any spa_load_verify errors found, but do not fail spa_load.
183 * This is used by zdb to analyze non-idle pools.
184 */
185boolean_t spa_load_verify_dryrun = B_FALSE;
186
187/*
188 * This (illegal) pool name is used when temporarily importing a spa_t in order
189 * to get the vdev stats associated with the imported devices.
190 */
191#define TRYIMPORT_NAME "$import"
192
193/*
194 * For debugging purposes: print out vdev tree during pool import.
195 */
196int spa_load_print_vdev_tree = B_FALSE;
197
198/*
199 * A non-zero value for zfs_max_missing_tvds means that we allow importing
200 * pools with missing top-level vdevs. This is strictly intended for advanced
201 * pool recovery cases since missing data is almost inevitable. Pools with
202 * missing devices can only be imported read-only for safety reasons, and their
203 * fail-mode will be automatically set to "continue".
204 *
205 * With 1 missing vdev we should be able to import the pool and mount all
206 * datasets. User data that was not modified after the missing device has been
207 * added should be recoverable. This means that snapshots created prior to the
208 * addition of that device should be completely intact.
209 *
210 * With 2 missing vdevs, some datasets may fail to mount since there are
211 * dataset statistics that are stored as regular metadata. Some data might be
212 * recoverable if those vdevs were added recently.
213 *
214 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
215 * may be missing entirely. Chances of data recovery are very low. Note that
216 * there are also risks of performing an inadvertent rewind as we might be
217 * missing all the vdevs with the latest uberblocks.
218 */
219uint64_t zfs_max_missing_tvds = 0;
220
221/*
222 * The parameters below are similar to zfs_max_missing_tvds but are only
223 * intended for a preliminary open of the pool with an untrusted config which
224 * might be incomplete or out-dated.
225 *
226 * We are more tolerant for pools opened from a cachefile since we could have
227 * an out-dated cachefile where a device removal was not registered.
228 * We could have set the limit arbitrarily high but in the case where devices
229 * are really missing we would want to return the proper error codes; we chose
230 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
231 * and we get a chance to retrieve the trusted config.
232 */
233uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
234
235/*
236 * In the case where config was assembled by scanning device paths (/dev/dsks
237 * by default) we are less tolerant since all the existing devices should have
238 * been detected and we want spa_load to return the right error codes.
239 */
240uint64_t zfs_max_missing_tvds_scan = 0;
241
242
243SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN,
244 &spa_load_print_vdev_tree, 0,
245 "print out vdev tree during pool import");
246SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN,
247 &zfs_max_missing_tvds, 0,
248 "allow importing pools with missing top-level vdevs");
249SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN,
250 &zfs_max_missing_tvds_cachefile, 0,
251 "allow importing pools with missing top-level vdevs in cache file");
252SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN,
253 &zfs_max_missing_tvds_scan, 0,
254 "allow importing pools with missing top-level vdevs during scan");
255
256/*
257 * Debugging aid that pauses spa_sync() towards the end.
258 */
259boolean_t zfs_pause_spa_sync = B_FALSE;
260
261/*
262 * ==========================================================================
263 * SPA properties routines
264 * ==========================================================================
265 */
266
267/*
268 * Add a (source=src, propname=propval) list to an nvlist.
269 */
270static void
271spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
272 uint64_t intval, zprop_source_t src)
273{
274 const char *propname = zpool_prop_to_name(prop);
275 nvlist_t *propval;
276
277 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
278 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
279
280 if (strval != NULL)
281 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
282 else
283 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
284
285 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
286 nvlist_free(propval);
287}
288
289/*
290 * Get property values from the spa configuration.
291 */
292static void
293spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
294{
295 vdev_t *rvd = spa->spa_root_vdev;
296 dsl_pool_t *pool = spa->spa_dsl_pool;
297 uint64_t size, alloc, cap, version;
298 zprop_source_t src = ZPROP_SRC_NONE;
299 spa_config_dirent_t *dp;
300 metaslab_class_t *mc = spa_normal_class(spa);
301
302 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
303
304 if (rvd != NULL) {
305 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
306 size = metaslab_class_get_space(spa_normal_class(spa));
307 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
308 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
309 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
310 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
311 size - alloc, src);
312 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
313 spa->spa_checkpoint_info.sci_dspace, src);
314
315 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
316 metaslab_class_fragmentation(mc), src);
317 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
318 metaslab_class_expandable_space(mc), src);
319 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
320 (spa_mode(spa) == FREAD), src);
321
322 cap = (size == 0) ? 0 : (alloc * 100 / size);
323 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
324
325 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
326 ddt_get_pool_dedup_ratio(spa), src);
327
328 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
329 rvd->vdev_state, src);
330
331 version = spa_version(spa);
332 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
333 src = ZPROP_SRC_DEFAULT;
334 else
335 src = ZPROP_SRC_LOCAL;
336 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
337 }
338
339 if (pool != NULL) {
340 /*
341 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
342 * when opening pools before this version freedir will be NULL.
343 */
344 if (pool->dp_free_dir != NULL) {
345 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
346 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
347 src);
348 } else {
349 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
350 NULL, 0, src);
351 }
352
353 if (pool->dp_leak_dir != NULL) {
354 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
355 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
356 src);
357 } else {
358 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
359 NULL, 0, src);
360 }
361 }
362
363 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
364
365 if (spa->spa_comment != NULL) {
366 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
367 0, ZPROP_SRC_LOCAL);
368 }
369
370 if (spa->spa_root != NULL)
371 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
372 0, ZPROP_SRC_LOCAL);
373
374 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
375 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
376 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
377 } else {
378 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
379 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
380 }
381
382 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
383 if (dp->scd_path == NULL) {
384 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
385 "none", 0, ZPROP_SRC_LOCAL);
386 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
387 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
388 dp->scd_path, 0, ZPROP_SRC_LOCAL);
389 }
390 }
391}
392
393/*
394 * Get zpool property values.
395 */
396int
397spa_prop_get(spa_t *spa, nvlist_t **nvp)
398{
399 objset_t *mos = spa->spa_meta_objset;
400 zap_cursor_t zc;
401 zap_attribute_t za;
402 int err;
403
404 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
405
406 mutex_enter(&spa->spa_props_lock);
407
408 /*
409 * Get properties from the spa config.
410 */
411 spa_prop_get_config(spa, nvp);
412
413 /* If no pool property object, no more prop to get. */
414 if (mos == NULL || spa->spa_pool_props_object == 0) {
415 mutex_exit(&spa->spa_props_lock);
416 return (0);
417 }
418
419 /*
420 * Get properties from the MOS pool property object.
421 */
422 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
423 (err = zap_cursor_retrieve(&zc, &za)) == 0;
424 zap_cursor_advance(&zc)) {
425 uint64_t intval = 0;
426 char *strval = NULL;
427 zprop_source_t src = ZPROP_SRC_DEFAULT;
428 zpool_prop_t prop;
429
430 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
431 continue;
432
433 switch (za.za_integer_length) {
434 case 8:
435 /* integer property */
436 if (za.za_first_integer !=
437 zpool_prop_default_numeric(prop))
438 src = ZPROP_SRC_LOCAL;
439
440 if (prop == ZPOOL_PROP_BOOTFS) {
441 dsl_pool_t *dp;
442 dsl_dataset_t *ds = NULL;
443
444 dp = spa_get_dsl(spa);
445 dsl_pool_config_enter(dp, FTAG);
446 if (err = dsl_dataset_hold_obj(dp,
447 za.za_first_integer, FTAG, &ds)) {
448 dsl_pool_config_exit(dp, FTAG);
449 break;
450 }
451
452 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
453 KM_SLEEP);
454 dsl_dataset_name(ds, strval);
455 dsl_dataset_rele(ds, FTAG);
456 dsl_pool_config_exit(dp, FTAG);
457 } else {
458 strval = NULL;
459 intval = za.za_first_integer;
460 }
461
462 spa_prop_add_list(*nvp, prop, strval, intval, src);
463
464 if (strval != NULL)
465 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
466
467 break;
468
469 case 1:
470 /* string property */
471 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
472 err = zap_lookup(mos, spa->spa_pool_props_object,
473 za.za_name, 1, za.za_num_integers, strval);
474 if (err) {
475 kmem_free(strval, za.za_num_integers);
476 break;
477 }
478 spa_prop_add_list(*nvp, prop, strval, 0, src);
479 kmem_free(strval, za.za_num_integers);
480 break;
481
482 default:
483 break;
484 }
485 }
486 zap_cursor_fini(&zc);
487 mutex_exit(&spa->spa_props_lock);
488out:
489 if (err && err != ENOENT) {
490 nvlist_free(*nvp);
491 *nvp = NULL;
492 return (err);
493 }
494
495 return (0);
496}
497
498/*
499 * Validate the given pool properties nvlist and modify the list
500 * for the property values to be set.
501 */
502static int
503spa_prop_validate(spa_t *spa, nvlist_t *props)
504{
505 nvpair_t *elem;
506 int error = 0, reset_bootfs = 0;
507 uint64_t objnum = 0;
508 boolean_t has_feature = B_FALSE;
509
510 elem = NULL;
511 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
512 uint64_t intval;
513 char *strval, *slash, *check, *fname;
514 const char *propname = nvpair_name(elem);
515 zpool_prop_t prop = zpool_name_to_prop(propname);
516
517 switch (prop) {
518 case ZPOOL_PROP_INVAL:
519 if (!zpool_prop_feature(propname)) {
520 error = SET_ERROR(EINVAL);
521 break;
522 }
523
524 /*
525 * Sanitize the input.
526 */
527 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
528 error = SET_ERROR(EINVAL);
529 break;
530 }
531
532 if (nvpair_value_uint64(elem, &intval) != 0) {
533 error = SET_ERROR(EINVAL);
534 break;
535 }
536
537 if (intval != 0) {
538 error = SET_ERROR(EINVAL);
539 break;
540 }
541
542 fname = strchr(propname, '@') + 1;
543 if (zfeature_lookup_name(fname, NULL) != 0) {
544 error = SET_ERROR(EINVAL);
545 break;
546 }
547
548 has_feature = B_TRUE;
549 break;
550
551 case ZPOOL_PROP_VERSION:
552 error = nvpair_value_uint64(elem, &intval);
553 if (!error &&
554 (intval < spa_version(spa) ||
555 intval > SPA_VERSION_BEFORE_FEATURES ||
556 has_feature))
557 error = SET_ERROR(EINVAL);
558 break;
559
560 case ZPOOL_PROP_DELEGATION:
561 case ZPOOL_PROP_AUTOREPLACE:
562 case ZPOOL_PROP_LISTSNAPS:
563 case ZPOOL_PROP_AUTOEXPAND:
564 error = nvpair_value_uint64(elem, &intval);
565 if (!error && intval > 1)
566 error = SET_ERROR(EINVAL);
567 break;
568
569 case ZPOOL_PROP_BOOTFS:
570 /*
571 * If the pool version is less than SPA_VERSION_BOOTFS,
572 * or the pool is still being created (version == 0),
573 * the bootfs property cannot be set.
574 */
575 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
576 error = SET_ERROR(ENOTSUP);
577 break;
578 }
579
580 /*
581 * Make sure the vdev config is bootable
582 */
583 if (!vdev_is_bootable(spa->spa_root_vdev)) {
584 error = SET_ERROR(ENOTSUP);
585 break;
586 }
587
588 reset_bootfs = 1;
589
590 error = nvpair_value_string(elem, &strval);
591
592 if (!error) {
593 objset_t *os;
594 uint64_t propval;
595
596 if (strval == NULL || strval[0] == '\0') {
597 objnum = zpool_prop_default_numeric(
598 ZPOOL_PROP_BOOTFS);
599 break;
600 }
601
602 if (error = dmu_objset_hold(strval, FTAG, &os))
603 break;
604
605 /*
606 * Must be ZPL, and its property settings
607 * must be supported by GRUB (compression
608 * is not gzip, and large blocks are not used).
609 */
610
611 if (dmu_objset_type(os) != DMU_OST_ZFS) {
612 error = SET_ERROR(ENOTSUP);
613 } else if ((error =
614 dsl_prop_get_int_ds(dmu_objset_ds(os),
615 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
616 &propval)) == 0 &&
617 !BOOTFS_COMPRESS_VALID(propval)) {
618 error = SET_ERROR(ENOTSUP);
619 } else {
620 objnum = dmu_objset_id(os);
621 }
622 dmu_objset_rele(os, FTAG);
623 }
624 break;
625
626 case ZPOOL_PROP_FAILUREMODE:
627 error = nvpair_value_uint64(elem, &intval);
628 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
629 intval > ZIO_FAILURE_MODE_PANIC))
630 error = SET_ERROR(EINVAL);
631
632 /*
633 * This is a special case which only occurs when
634 * the pool has completely failed. This allows
635 * the user to change the in-core failmode property
636 * without syncing it out to disk (I/Os might
637 * currently be blocked). We do this by returning
638 * EIO to the caller (spa_prop_set) to trick it
639 * into thinking we encountered a property validation
640 * error.
641 */
642 if (!error && spa_suspended(spa)) {
643 spa->spa_failmode = intval;
644 error = SET_ERROR(EIO);
645 }
646 break;
647
648 case ZPOOL_PROP_CACHEFILE:
649 if ((error = nvpair_value_string(elem, &strval)) != 0)
650 break;
651
652 if (strval[0] == '\0')
653 break;
654
655 if (strcmp(strval, "none") == 0)
656 break;
657
658 if (strval[0] != '/') {
659 error = SET_ERROR(EINVAL);
660 break;
661 }
662
663 slash = strrchr(strval, '/');
664 ASSERT(slash != NULL);
665
666 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
667 strcmp(slash, "/..") == 0)
668 error = SET_ERROR(EINVAL);
669 break;
670
671 case ZPOOL_PROP_COMMENT:
672 if ((error = nvpair_value_string(elem, &strval)) != 0)
673 break;
674 for (check = strval; *check != '\0'; check++) {
675 /*
676 * The kernel doesn't have an easy isprint()
677 * check. For this kernel check, we merely
678 * check ASCII apart from DEL. Fix this if
679 * there is an easy-to-use kernel isprint().
680 */
681 if (*check >= 0x7f) {
682 error = SET_ERROR(EINVAL);
683 break;
684 }
685 }
686 if (strlen(strval) > ZPROP_MAX_COMMENT)
687 error = E2BIG;
688 break;
689
690 case ZPOOL_PROP_DEDUPDITTO:
691 if (spa_version(spa) < SPA_VERSION_DEDUP)
692 error = SET_ERROR(ENOTSUP);
693 else
694 error = nvpair_value_uint64(elem, &intval);
695 if (error == 0 &&
696 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
697 error = SET_ERROR(EINVAL);
698 break;
699 }
700
701 if (error)
702 break;
703 }
704
705 if (!error && reset_bootfs) {
706 error = nvlist_remove(props,
707 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
708
709 if (!error) {
710 error = nvlist_add_uint64(props,
711 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
712 }
713 }
714
715 return (error);
716}
717
718void
719spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
720{
721 char *cachefile;
722 spa_config_dirent_t *dp;
723
724 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
725 &cachefile) != 0)
726 return;
727
728 dp = kmem_alloc(sizeof (spa_config_dirent_t),
729 KM_SLEEP);
730
731 if (cachefile[0] == '\0')
732 dp->scd_path = spa_strdup(spa_config_path);
733 else if (strcmp(cachefile, "none") == 0)
734 dp->scd_path = NULL;
735 else
736 dp->scd_path = spa_strdup(cachefile);
737
738 list_insert_head(&spa->spa_config_list, dp);
739 if (need_sync)
740 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
741}
742
743int
744spa_prop_set(spa_t *spa, nvlist_t *nvp)
745{
746 int error;
747 nvpair_t *elem = NULL;
748 boolean_t need_sync = B_FALSE;
749
750 if ((error = spa_prop_validate(spa, nvp)) != 0)
751 return (error);
752
753 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
754 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
755
756 if (prop == ZPOOL_PROP_CACHEFILE ||
757 prop == ZPOOL_PROP_ALTROOT ||
758 prop == ZPOOL_PROP_READONLY)
759 continue;
760
761 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
762 uint64_t ver;
763
764 if (prop == ZPOOL_PROP_VERSION) {
765 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
766 } else {
767 ASSERT(zpool_prop_feature(nvpair_name(elem)));
768 ver = SPA_VERSION_FEATURES;
769 need_sync = B_TRUE;
770 }
771
772 /* Save time if the version is already set. */
773 if (ver == spa_version(spa))
774 continue;
775
776 /*
777 * In addition to the pool directory object, we might
778 * create the pool properties object, the features for
779 * read object, the features for write object, or the
780 * feature descriptions object.
781 */
782 error = dsl_sync_task(spa->spa_name, NULL,
783 spa_sync_version, &ver,
784 6, ZFS_SPACE_CHECK_RESERVED);
785 if (error)
786 return (error);
787 continue;
788 }
789
790 need_sync = B_TRUE;
791 break;
792 }
793
794 if (need_sync) {
795 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
796 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
797 }
798
799 return (0);
800}
801
802/*
803 * If the bootfs property value is dsobj, clear it.
804 */
805void
806spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
807{
808 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
809 VERIFY(zap_remove(spa->spa_meta_objset,
810 spa->spa_pool_props_object,
811 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
812 spa->spa_bootfs = 0;
813 }
814}
815
816/*ARGSUSED*/
817static int
818spa_change_guid_check(void *arg, dmu_tx_t *tx)
819{
820 uint64_t *newguid = arg;
821 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
822 vdev_t *rvd = spa->spa_root_vdev;
823 uint64_t vdev_state;
824
825 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
826 int error = (spa_has_checkpoint(spa)) ?
827 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
828 return (SET_ERROR(error));
829 }
830
831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
832 vdev_state = rvd->vdev_state;
833 spa_config_exit(spa, SCL_STATE, FTAG);
834
835 if (vdev_state != VDEV_STATE_HEALTHY)
836 return (SET_ERROR(ENXIO));
837
838 ASSERT3U(spa_guid(spa), !=, *newguid);
839
840 return (0);
841}
842
843static void
844spa_change_guid_sync(void *arg, dmu_tx_t *tx)
845{
846 uint64_t *newguid = arg;
847 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
848 uint64_t oldguid;
849 vdev_t *rvd = spa->spa_root_vdev;
850
851 oldguid = spa_guid(spa);
852
853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
854 rvd->vdev_guid = *newguid;
855 rvd->vdev_guid_sum += (*newguid - oldguid);
856 vdev_config_dirty(rvd);
857 spa_config_exit(spa, SCL_STATE, FTAG);
858
859 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
860 oldguid, *newguid);
861}
862
863/*
864 * Change the GUID for the pool. This is done so that we can later
865 * re-import a pool built from a clone of our own vdevs. We will modify
866 * the root vdev's guid, our own pool guid, and then mark all of our
867 * vdevs dirty. Note that we must make sure that all our vdevs are
868 * online when we do this, or else any vdevs that weren't present
869 * would be orphaned from our pool. We are also going to issue a
870 * sysevent to update any watchers.
871 */
872int
873spa_change_guid(spa_t *spa)
874{
875 int error;
876 uint64_t guid;
877
878 mutex_enter(&spa->spa_vdev_top_lock);
879 mutex_enter(&spa_namespace_lock);
880 guid = spa_generate_guid(NULL);
881
882 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
883 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
884
885 if (error == 0) {
886 spa_write_cachefile(spa, B_FALSE, B_TRUE);
887 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
888 }
889
890 mutex_exit(&spa_namespace_lock);
891 mutex_exit(&spa->spa_vdev_top_lock);
892
893 return (error);
894}
895
896/*
897 * ==========================================================================
898 * SPA state manipulation (open/create/destroy/import/export)
899 * ==========================================================================
900 */
901
902static int
903spa_error_entry_compare(const void *a, const void *b)
904{
905 spa_error_entry_t *sa = (spa_error_entry_t *)a;
906 spa_error_entry_t *sb = (spa_error_entry_t *)b;
907 int ret;
908
909 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
910 sizeof (zbookmark_phys_t));
911
912 if (ret < 0)
913 return (-1);
914 else if (ret > 0)
915 return (1);
916 else
917 return (0);
918}
919
920/*
921 * Utility function which retrieves copies of the current logs and
922 * re-initializes them in the process.
923 */
924void
925spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
926{
927 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
928
929 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
930 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
931
932 avl_create(&spa->spa_errlist_scrub,
933 spa_error_entry_compare, sizeof (spa_error_entry_t),
934 offsetof(spa_error_entry_t, se_avl));
935 avl_create(&spa->spa_errlist_last,
936 spa_error_entry_compare, sizeof (spa_error_entry_t),
937 offsetof(spa_error_entry_t, se_avl));
938}
939
940static void
941spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
942{
943 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
944 enum zti_modes mode = ztip->zti_mode;
945 uint_t value = ztip->zti_value;
946 uint_t count = ztip->zti_count;
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
948 char name[32];
949 uint_t flags = 0;
950 boolean_t batch = B_FALSE;
951
952 if (mode == ZTI_MODE_NULL) {
953 tqs->stqs_count = 0;
954 tqs->stqs_taskq = NULL;
955 return;
956 }
957
958 ASSERT3U(count, >, 0);
959
960 tqs->stqs_count = count;
961 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
962
963 switch (mode) {
964 case ZTI_MODE_FIXED:
965 ASSERT3U(value, >=, 1);
966 value = MAX(value, 1);
967 break;
968
969 case ZTI_MODE_BATCH:
970 batch = B_TRUE;
971 flags |= TASKQ_THREADS_CPU_PCT;
972 value = zio_taskq_batch_pct;
973 break;
974
975 default:
976 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
977 "spa_activate()",
978 zio_type_name[t], zio_taskq_types[q], mode, value);
979 break;
980 }
981
982 for (uint_t i = 0; i < count; i++) {
983 taskq_t *tq;
984
985 if (count > 1) {
986 (void) snprintf(name, sizeof (name), "%s_%s_%u",
987 zio_type_name[t], zio_taskq_types[q], i);
988 } else {
989 (void) snprintf(name, sizeof (name), "%s_%s",
990 zio_type_name[t], zio_taskq_types[q]);
991 }
992
993#ifdef SYSDC
994 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
995 if (batch)
996 flags |= TASKQ_DC_BATCH;
997
998 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
999 spa->spa_proc, zio_taskq_basedc, flags);
1000 } else {
1001#endif
1002 pri_t pri = maxclsyspri;
1003 /*
1004 * The write issue taskq can be extremely CPU
1005 * intensive. Run it at slightly lower priority
1006 * than the other taskqs.
1007 * FreeBSD notes:
1008 * - numerically higher priorities are lower priorities;
1009 * - if priorities divided by four (RQ_PPQ) are equal
1010 * then a difference between them is insignificant.
1011 */
1012 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1013#ifdef illumos
1014 pri--;
1015#else
1016 pri += 4;
1017#endif
1018
1019 tq = taskq_create_proc(name, value, pri, 50,
1020 INT_MAX, spa->spa_proc, flags);
1021#ifdef SYSDC
1022 }
1023#endif
1024
1025 tqs->stqs_taskq[i] = tq;
1026 }
1027}
1028
1029static void
1030spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1031{
1032 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1033
1034 if (tqs->stqs_taskq == NULL) {
1035 ASSERT0(tqs->stqs_count);
1036 return;
1037 }
1038
1039 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1040 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1041 taskq_destroy(tqs->stqs_taskq[i]);
1042 }
1043
1044 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1045 tqs->stqs_taskq = NULL;
1046}
1047
1048/*
1049 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1050 * Note that a type may have multiple discrete taskqs to avoid lock contention
1051 * on the taskq itself. In that case we choose which taskq at random by using
1052 * the low bits of gethrtime().
1053 */
1054void
1055spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1056 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1057{
1058 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1059 taskq_t *tq;
1060
1061 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1062 ASSERT3U(tqs->stqs_count, !=, 0);
1063
1064 if (tqs->stqs_count == 1) {
1065 tq = tqs->stqs_taskq[0];
1066 } else {
1067#ifdef _KERNEL
1068 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
1069#else
1070 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1071#endif
1072 }
1073
1074 taskq_dispatch_ent(tq, func, arg, flags, ent);
1075}
1076
1077static void
1078spa_create_zio_taskqs(spa_t *spa)
1079{
1080 for (int t = 0; t < ZIO_TYPES; t++) {
1081 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1082 spa_taskqs_init(spa, t, q);
1083 }
1084 }
1085}
1086
1087#ifdef _KERNEL
1088#ifdef SPA_PROCESS
1089static void
1090spa_thread(void *arg)
1091{
1092 callb_cpr_t cprinfo;
1093
1094 spa_t *spa = arg;
1095 user_t *pu = PTOU(curproc);
1096
1097 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1098 spa->spa_name);
1099
1100 ASSERT(curproc != &p0);
1101 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1102 "zpool-%s", spa->spa_name);
1103 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1104
1105#ifdef PSRSET_BIND
1106 /* bind this thread to the requested psrset */
1107 if (zio_taskq_psrset_bind != PS_NONE) {
1108 pool_lock();
1109 mutex_enter(&cpu_lock);
1110 mutex_enter(&pidlock);
1111 mutex_enter(&curproc->p_lock);
1112
1113 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1114 0, NULL, NULL) == 0) {
1115 curthread->t_bind_pset = zio_taskq_psrset_bind;
1116 } else {
1117 cmn_err(CE_WARN,
1118 "Couldn't bind process for zfs pool \"%s\" to "
1119 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1120 }
1121
1122 mutex_exit(&curproc->p_lock);
1123 mutex_exit(&pidlock);
1124 mutex_exit(&cpu_lock);
1125 pool_unlock();
1126 }
1127#endif
1128
1129#ifdef SYSDC
1130 if (zio_taskq_sysdc) {
1131 sysdc_thread_enter(curthread, 100, 0);
1132 }
1133#endif
1134
1135 spa->spa_proc = curproc;
1136 spa->spa_did = curthread->t_did;
1137
1138 spa_create_zio_taskqs(spa);
1139
1140 mutex_enter(&spa->spa_proc_lock);
1141 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1142
1143 spa->spa_proc_state = SPA_PROC_ACTIVE;
1144 cv_broadcast(&spa->spa_proc_cv);
1145
1146 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1147 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1148 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1149 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1150
1151 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1152 spa->spa_proc_state = SPA_PROC_GONE;
1153 spa->spa_proc = &p0;
1154 cv_broadcast(&spa->spa_proc_cv);
1155 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1156
1157 mutex_enter(&curproc->p_lock);
1158 lwp_exit();
1159}
1160#endif /* SPA_PROCESS */
1161#endif
1162
1163/*
1164 * Activate an uninitialized pool.
1165 */
1166static void
1167spa_activate(spa_t *spa, int mode)
1168{
1169 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1170
1171 spa->spa_state = POOL_STATE_ACTIVE;
1172 spa->spa_mode = mode;
1173
1174 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1175 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1176
1177 /* Try to create a covering process */
1178 mutex_enter(&spa->spa_proc_lock);
1179 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1180 ASSERT(spa->spa_proc == &p0);
1181 spa->spa_did = 0;
1182
1183#ifdef SPA_PROCESS
1184 /* Only create a process if we're going to be around a while. */
1185 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1186 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1187 NULL, 0) == 0) {
1188 spa->spa_proc_state = SPA_PROC_CREATED;
1189 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1190 cv_wait(&spa->spa_proc_cv,
1191 &spa->spa_proc_lock);
1192 }
1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1194 ASSERT(spa->spa_proc != &p0);
1195 ASSERT(spa->spa_did != 0);
1196 } else {
1197#ifdef _KERNEL
1198 cmn_err(CE_WARN,
1199 "Couldn't create process for zfs pool \"%s\"\n",
1200 spa->spa_name);
1201#endif
1202 }
1203 }
1204#endif /* SPA_PROCESS */
1205 mutex_exit(&spa->spa_proc_lock);
1206
1207 /* If we didn't create a process, we need to create our taskqs. */
1208 ASSERT(spa->spa_proc == &p0);
1209 if (spa->spa_proc == &p0) {
1210 spa_create_zio_taskqs(spa);
1211 }
1212
1213 /*
1214 * Start TRIM thread.
1215 */
1216 trim_thread_create(spa);
1217
1218 for (size_t i = 0; i < TXG_SIZE; i++)
1219 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0);
1220
1221 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1222 offsetof(vdev_t, vdev_config_dirty_node));
1223 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1224 offsetof(objset_t, os_evicting_node));
1225 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1226 offsetof(vdev_t, vdev_state_dirty_node));
1227
1228 txg_list_create(&spa->spa_vdev_txg_list, spa,
1229 offsetof(struct vdev, vdev_txg_node));
1230
1231 avl_create(&spa->spa_errlist_scrub,
1232 spa_error_entry_compare, sizeof (spa_error_entry_t),
1233 offsetof(spa_error_entry_t, se_avl));
1234 avl_create(&spa->spa_errlist_last,
1235 spa_error_entry_compare, sizeof (spa_error_entry_t),
1236 offsetof(spa_error_entry_t, se_avl));
1237}
1238
1239/*
1240 * Opposite of spa_activate().
1241 */
1242static void
1243spa_deactivate(spa_t *spa)
1244{
1245 ASSERT(spa->spa_sync_on == B_FALSE);
1246 ASSERT(spa->spa_dsl_pool == NULL);
1247 ASSERT(spa->spa_root_vdev == NULL);
1248 ASSERT(spa->spa_async_zio_root == NULL);
1249 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1250
1251 /*
1252 * Stop TRIM thread in case spa_unload() wasn't called directly
1253 * before spa_deactivate().
1254 */
1255 trim_thread_destroy(spa);
1256
1257 spa_evicting_os_wait(spa);
1258
1259 txg_list_destroy(&spa->spa_vdev_txg_list);
1260
1261 list_destroy(&spa->spa_config_dirty_list);
1262 list_destroy(&spa->spa_evicting_os_list);
1263 list_destroy(&spa->spa_state_dirty_list);
1264
1265 for (int t = 0; t < ZIO_TYPES; t++) {
1266 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1267 spa_taskqs_fini(spa, t, q);
1268 }
1269 }
1270
1271 for (size_t i = 0; i < TXG_SIZE; i++) {
1272 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1273 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1274 spa->spa_txg_zio[i] = NULL;
1275 }
1276
1277 metaslab_class_destroy(spa->spa_normal_class);
1278 spa->spa_normal_class = NULL;
1279
1280 metaslab_class_destroy(spa->spa_log_class);
1281 spa->spa_log_class = NULL;
1282
1283 /*
1284 * If this was part of an import or the open otherwise failed, we may
1285 * still have errors left in the queues. Empty them just in case.
1286 */
1287 spa_errlog_drain(spa);
1288
1289 avl_destroy(&spa->spa_errlist_scrub);
1290 avl_destroy(&spa->spa_errlist_last);
1291
1292 spa->spa_state = POOL_STATE_UNINITIALIZED;
1293
1294 mutex_enter(&spa->spa_proc_lock);
1295 if (spa->spa_proc_state != SPA_PROC_NONE) {
1296 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1297 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1298 cv_broadcast(&spa->spa_proc_cv);
1299 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1300 ASSERT(spa->spa_proc != &p0);
1301 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1302 }
1303 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1304 spa->spa_proc_state = SPA_PROC_NONE;
1305 }
1306 ASSERT(spa->spa_proc == &p0);
1307 mutex_exit(&spa->spa_proc_lock);
1308
1309#ifdef SPA_PROCESS
1310 /*
1311 * We want to make sure spa_thread() has actually exited the ZFS
1312 * module, so that the module can't be unloaded out from underneath
1313 * it.
1314 */
1315 if (spa->spa_did != 0) {
1316 thread_join(spa->spa_did);
1317 spa->spa_did = 0;
1318 }
1319#endif /* SPA_PROCESS */
1320}
1321
1322/*
1323 * Verify a pool configuration, and construct the vdev tree appropriately. This
1324 * will create all the necessary vdevs in the appropriate layout, with each vdev
1325 * in the CLOSED state. This will prep the pool before open/creation/import.
1326 * All vdev validation is done by the vdev_alloc() routine.
1327 */
1328static int
1329spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1330 uint_t id, int atype)
1331{
1332 nvlist_t **child;
1333 uint_t children;
1334 int error;
1335
1336 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1337 return (error);
1338
1339 if ((*vdp)->vdev_ops->vdev_op_leaf)
1340 return (0);
1341
1342 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1343 &child, &children);
1344
1345 if (error == ENOENT)
1346 return (0);
1347
1348 if (error) {
1349 vdev_free(*vdp);
1350 *vdp = NULL;
1351 return (SET_ERROR(EINVAL));
1352 }
1353
1354 for (int c = 0; c < children; c++) {
1355 vdev_t *vd;
1356 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1357 atype)) != 0) {
1358 vdev_free(*vdp);
1359 *vdp = NULL;
1360 return (error);
1361 }
1362 }
1363
1364 ASSERT(*vdp != NULL);
1365
1366 return (0);
1367}
1368
1369/*
1370 * Opposite of spa_load().
1371 */
1372static void
1373spa_unload(spa_t *spa)
1374{
1375 int i;
1376
1377 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1378
1379 spa_load_note(spa, "UNLOADING");
1380
1381 /*
1382 * Stop TRIM thread.
1383 */
1384 trim_thread_destroy(spa);
1385
1386 /*
1387 * Stop async tasks.
1388 */
1389 spa_async_suspend(spa);
1390
1391 /*
1392 * Stop syncing.
1393 */
1394 if (spa->spa_sync_on) {
1395 txg_sync_stop(spa->spa_dsl_pool);
1396 spa->spa_sync_on = B_FALSE;
1397 }
1398
1399 /*
1400 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1401 * to call it earlier, before we wait for async i/o to complete.
1402 * This ensures that there is no async metaslab prefetching, by
1403 * calling taskq_wait(mg_taskq).
1404 */
1405 if (spa->spa_root_vdev != NULL) {
1406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1407 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1408 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1409 spa_config_exit(spa, SCL_ALL, FTAG);
1410 }
1411
1412 /*
1413 * Wait for any outstanding async I/O to complete.
1414 */
1415 if (spa->spa_async_zio_root != NULL) {
1416 for (int i = 0; i < max_ncpus; i++)
1417 (void) zio_wait(spa->spa_async_zio_root[i]);
1418 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1419 spa->spa_async_zio_root = NULL;
1420 }
1421
1422 if (spa->spa_vdev_removal != NULL) {
1423 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1424 spa->spa_vdev_removal = NULL;
1425 }
1426
1427 if (spa->spa_condense_zthr != NULL) {
1428 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1429 zthr_destroy(spa->spa_condense_zthr);
1430 spa->spa_condense_zthr = NULL;
1431 }
1432
1433 if (spa->spa_checkpoint_discard_zthr != NULL) {
1434 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1435 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1436 spa->spa_checkpoint_discard_zthr = NULL;
1437 }
1438
1439 spa_condense_fini(spa);
1440
1441 bpobj_close(&spa->spa_deferred_bpobj);
1442
1443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1444
1445 /*
1446 * Close all vdevs.
1447 */
1448 if (spa->spa_root_vdev)
1449 vdev_free(spa->spa_root_vdev);
1450 ASSERT(spa->spa_root_vdev == NULL);
1451
1452 /*
1453 * Close the dsl pool.
1454 */
1455 if (spa->spa_dsl_pool) {
1456 dsl_pool_close(spa->spa_dsl_pool);
1457 spa->spa_dsl_pool = NULL;
1458 spa->spa_meta_objset = NULL;
1459 }
1460
1461 ddt_unload(spa);
1462
1463 /*
1464 * Drop and purge level 2 cache
1465 */
1466 spa_l2cache_drop(spa);
1467
1468 for (i = 0; i < spa->spa_spares.sav_count; i++)
1469 vdev_free(spa->spa_spares.sav_vdevs[i]);
1470 if (spa->spa_spares.sav_vdevs) {
1471 kmem_free(spa->spa_spares.sav_vdevs,
1472 spa->spa_spares.sav_count * sizeof (void *));
1473 spa->spa_spares.sav_vdevs = NULL;
1474 }
1475 if (spa->spa_spares.sav_config) {
1476 nvlist_free(spa->spa_spares.sav_config);
1477 spa->spa_spares.sav_config = NULL;
1478 }
1479 spa->spa_spares.sav_count = 0;
1480
1481 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1482 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1483 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1484 }
1485 if (spa->spa_l2cache.sav_vdevs) {
1486 kmem_free(spa->spa_l2cache.sav_vdevs,
1487 spa->spa_l2cache.sav_count * sizeof (void *));
1488 spa->spa_l2cache.sav_vdevs = NULL;
1489 }
1490 if (spa->spa_l2cache.sav_config) {
1491 nvlist_free(spa->spa_l2cache.sav_config);
1492 spa->spa_l2cache.sav_config = NULL;
1493 }
1494 spa->spa_l2cache.sav_count = 0;
1495
1496 spa->spa_async_suspended = 0;
1497
1498 spa->spa_indirect_vdevs_loaded = B_FALSE;
1499
1500 if (spa->spa_comment != NULL) {
1501 spa_strfree(spa->spa_comment);
1502 spa->spa_comment = NULL;
1503 }
1504
1505 spa_config_exit(spa, SCL_ALL, FTAG);
1506}
1507
1508/*
1509 * Load (or re-load) the current list of vdevs describing the active spares for
1510 * this pool. When this is called, we have some form of basic information in
1511 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1512 * then re-generate a more complete list including status information.
1513 */
1514void
1515spa_load_spares(spa_t *spa)
1516{
1517 nvlist_t **spares;
1518 uint_t nspares;
1519 int i;
1520 vdev_t *vd, *tvd;
1521
1522#ifndef _KERNEL
1523 /*
1524 * zdb opens both the current state of the pool and the
1525 * checkpointed state (if present), with a different spa_t.
1526 *
1527 * As spare vdevs are shared among open pools, we skip loading
1528 * them when we load the checkpointed state of the pool.
1529 */
1530 if (!spa_writeable(spa))
1531 return;
1532#endif
1533
1534 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1535
1536 /*
1537 * First, close and free any existing spare vdevs.
1538 */
1539 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1540 vd = spa->spa_spares.sav_vdevs[i];
1541
1542 /* Undo the call to spa_activate() below */
1543 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1544 B_FALSE)) != NULL && tvd->vdev_isspare)
1545 spa_spare_remove(tvd);
1546 vdev_close(vd);
1547 vdev_free(vd);
1548 }
1549
1550 if (spa->spa_spares.sav_vdevs)
1551 kmem_free(spa->spa_spares.sav_vdevs,
1552 spa->spa_spares.sav_count * sizeof (void *));
1553
1554 if (spa->spa_spares.sav_config == NULL)
1555 nspares = 0;
1556 else
1557 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1558 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1559
1560 spa->spa_spares.sav_count = (int)nspares;
1561 spa->spa_spares.sav_vdevs = NULL;
1562
1563 if (nspares == 0)
1564 return;
1565
1566 /*
1567 * Construct the array of vdevs, opening them to get status in the
1568 * process. For each spare, there is potentially two different vdev_t
1569 * structures associated with it: one in the list of spares (used only
1570 * for basic validation purposes) and one in the active vdev
1571 * configuration (if it's spared in). During this phase we open and
1572 * validate each vdev on the spare list. If the vdev also exists in the
1573 * active configuration, then we also mark this vdev as an active spare.
1574 */
1575 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1576 KM_SLEEP);
1577 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1578 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1579 VDEV_ALLOC_SPARE) == 0);
1580 ASSERT(vd != NULL);
1581
1582 spa->spa_spares.sav_vdevs[i] = vd;
1583
1584 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1585 B_FALSE)) != NULL) {
1586 if (!tvd->vdev_isspare)
1587 spa_spare_add(tvd);
1588
1589 /*
1590 * We only mark the spare active if we were successfully
1591 * able to load the vdev. Otherwise, importing a pool
1592 * with a bad active spare would result in strange
1593 * behavior, because multiple pool would think the spare
1594 * is actively in use.
1595 *
1596 * There is a vulnerability here to an equally bizarre
1597 * circumstance, where a dead active spare is later
1598 * brought back to life (onlined or otherwise). Given
1599 * the rarity of this scenario, and the extra complexity
1600 * it adds, we ignore the possibility.
1601 */
1602 if (!vdev_is_dead(tvd))
1603 spa_spare_activate(tvd);
1604 }
1605
1606 vd->vdev_top = vd;
1607 vd->vdev_aux = &spa->spa_spares;
1608
1609 if (vdev_open(vd) != 0)
1610 continue;
1611
1612 if (vdev_validate_aux(vd) == 0)
1613 spa_spare_add(vd);
1614 }
1615
1616 /*
1617 * Recompute the stashed list of spares, with status information
1618 * this time.
1619 */
1620 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1621 DATA_TYPE_NVLIST_ARRAY) == 0);
1622
1623 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1624 KM_SLEEP);
1625 for (i = 0; i < spa->spa_spares.sav_count; i++)
1626 spares[i] = vdev_config_generate(spa,
1627 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1628 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1629 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1630 for (i = 0; i < spa->spa_spares.sav_count; i++)
1631 nvlist_free(spares[i]);
1632 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1633}
1634
1635/*
1636 * Load (or re-load) the current list of vdevs describing the active l2cache for
1637 * this pool. When this is called, we have some form of basic information in
1638 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1639 * then re-generate a more complete list including status information.
1640 * Devices which are already active have their details maintained, and are
1641 * not re-opened.
1642 */
1643void
1644spa_load_l2cache(spa_t *spa)
1645{
1646 nvlist_t **l2cache;
1647 uint_t nl2cache;
1648 int i, j, oldnvdevs;
1649 uint64_t guid;
1650 vdev_t *vd, **oldvdevs, **newvdevs;
1651 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1652
1653#ifndef _KERNEL
1654 /*
1655 * zdb opens both the current state of the pool and the
1656 * checkpointed state (if present), with a different spa_t.
1657 *
1658 * As L2 caches are part of the ARC which is shared among open
1659 * pools, we skip loading them when we load the checkpointed
1660 * state of the pool.
1661 */
1662 if (!spa_writeable(spa))
1663 return;
1664#endif
1665
1666 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1667
1668 if (sav->sav_config != NULL) {
1669 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1670 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1671 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1672 } else {
1673 nl2cache = 0;
1674 newvdevs = NULL;
1675 }
1676
1677 oldvdevs = sav->sav_vdevs;
1678 oldnvdevs = sav->sav_count;
1679 sav->sav_vdevs = NULL;
1680 sav->sav_count = 0;
1681
1682 /*
1683 * Process new nvlist of vdevs.
1684 */
1685 for (i = 0; i < nl2cache; i++) {
1686 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1687 &guid) == 0);
1688
1689 newvdevs[i] = NULL;
1690 for (j = 0; j < oldnvdevs; j++) {
1691 vd = oldvdevs[j];
1692 if (vd != NULL && guid == vd->vdev_guid) {
1693 /*
1694 * Retain previous vdev for add/remove ops.
1695 */
1696 newvdevs[i] = vd;
1697 oldvdevs[j] = NULL;
1698 break;
1699 }
1700 }
1701
1702 if (newvdevs[i] == NULL) {
1703 /*
1704 * Create new vdev
1705 */
1706 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1707 VDEV_ALLOC_L2CACHE) == 0);
1708 ASSERT(vd != NULL);
1709 newvdevs[i] = vd;
1710
1711 /*
1712 * Commit this vdev as an l2cache device,
1713 * even if it fails to open.
1714 */
1715 spa_l2cache_add(vd);
1716
1717 vd->vdev_top = vd;
1718 vd->vdev_aux = sav;
1719
1720 spa_l2cache_activate(vd);
1721
1722 if (vdev_open(vd) != 0)
1723 continue;
1724
1725 (void) vdev_validate_aux(vd);
1726
1727 if (!vdev_is_dead(vd))
1728 l2arc_add_vdev(spa, vd);
1729 }
1730 }
1731
1732 /*
1733 * Purge vdevs that were dropped
1734 */
1735 for (i = 0; i < oldnvdevs; i++) {
1736 uint64_t pool;
1737
1738 vd = oldvdevs[i];
1739 if (vd != NULL) {
1740 ASSERT(vd->vdev_isl2cache);
1741
1742 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1743 pool != 0ULL && l2arc_vdev_present(vd))
1744 l2arc_remove_vdev(vd);
1745 vdev_clear_stats(vd);
1746 vdev_free(vd);
1747 }
1748 }
1749
1750 if (oldvdevs)
1751 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1752
1753 if (sav->sav_config == NULL)
1754 goto out;
1755
1756 sav->sav_vdevs = newvdevs;
1757 sav->sav_count = (int)nl2cache;
1758
1759 /*
1760 * Recompute the stashed list of l2cache devices, with status
1761 * information this time.
1762 */
1763 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1764 DATA_TYPE_NVLIST_ARRAY) == 0);
1765
1766 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1767 for (i = 0; i < sav->sav_count; i++)
1768 l2cache[i] = vdev_config_generate(spa,
1769 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1770 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1771 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1772out:
1773 for (i = 0; i < sav->sav_count; i++)
1774 nvlist_free(l2cache[i]);
1775 if (sav->sav_count)
1776 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1777}
1778
1779static int
1780load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1781{
1782 dmu_buf_t *db;
1783 char *packed = NULL;
1784 size_t nvsize = 0;
1785 int error;
1786 *value = NULL;
1787
1788 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1789 if (error != 0)
1790 return (error);
1791
1792 nvsize = *(uint64_t *)db->db_data;
1793 dmu_buf_rele(db, FTAG);
1794
1795 packed = kmem_alloc(nvsize, KM_SLEEP);
1796 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1797 DMU_READ_PREFETCH);
1798 if (error == 0)
1799 error = nvlist_unpack(packed, nvsize, value, 0);
1800 kmem_free(packed, nvsize);
1801
1802 return (error);
1803}
1804
1805/*
1806 * Concrete top-level vdevs that are not missing and are not logs. At every
1807 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1808 */
1809static uint64_t
1810spa_healthy_core_tvds(spa_t *spa)
1811{
1812 vdev_t *rvd = spa->spa_root_vdev;
1813 uint64_t tvds = 0;
1814
1815 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1816 vdev_t *vd = rvd->vdev_child[i];
1817 if (vd->vdev_islog)
1818 continue;
1819 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1820 tvds++;
1821 }
1822
1823 return (tvds);
1824}
1825
1826/*
1827 * Checks to see if the given vdev could not be opened, in which case we post a
1828 * sysevent to notify the autoreplace code that the device has been removed.
1829 */
1830static void
1831spa_check_removed(vdev_t *vd)
1832{
1833 for (uint64_t c = 0; c < vd->vdev_children; c++)
1834 spa_check_removed(vd->vdev_child[c]);
1835
1836 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1837 vdev_is_concrete(vd)) {
1838 zfs_post_autoreplace(vd->vdev_spa, vd);
1839 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1840 }
1841}
1842
1843static int
1844spa_check_for_missing_logs(spa_t *spa)
1845{
1846 vdev_t *rvd = spa->spa_root_vdev;
1847
1848 /*
1849 * If we're doing a normal import, then build up any additional
1850 * diagnostic information about missing log devices.
1851 * We'll pass this up to the user for further processing.
1852 */
1853 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1854 nvlist_t **child, *nv;
1855 uint64_t idx = 0;
1856
1857 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1858 KM_SLEEP);
1859 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1860
1861 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1862 vdev_t *tvd = rvd->vdev_child[c];
1863
1864 /*
1865 * We consider a device as missing only if it failed
1866 * to open (i.e. offline or faulted is not considered
1867 * as missing).
1868 */
1869 if (tvd->vdev_islog &&
1870 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1871 child[idx++] = vdev_config_generate(spa, tvd,
1872 B_FALSE, VDEV_CONFIG_MISSING);
1873 }
1874 }
1875
1876 if (idx > 0) {
1877 fnvlist_add_nvlist_array(nv,
1878 ZPOOL_CONFIG_CHILDREN, child, idx);
1879 fnvlist_add_nvlist(spa->spa_load_info,
1880 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1881
1882 for (uint64_t i = 0; i < idx; i++)
1883 nvlist_free(child[i]);
1884 }
1885 nvlist_free(nv);
1886 kmem_free(child, rvd->vdev_children * sizeof (char **));
1887
1888 if (idx > 0) {
1889 spa_load_failed(spa, "some log devices are missing");
1890 vdev_dbgmsg_print_tree(rvd, 2);
1891 return (SET_ERROR(ENXIO));
1892 }
1893 } else {
1894 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1895 vdev_t *tvd = rvd->vdev_child[c];
1896
1897 if (tvd->vdev_islog &&
1898 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1899 spa_set_log_state(spa, SPA_LOG_CLEAR);
1900 spa_load_note(spa, "some log devices are "
1901 "missing, ZIL is dropped.");
1902 vdev_dbgmsg_print_tree(rvd, 2);
1903 break;
1904 }
1905 }
1906 }
1907
1908 return (0);
1909}
1910
1911/*
1912 * Check for missing log devices
1913 */
1914static boolean_t
1915spa_check_logs(spa_t *spa)
1916{
1917 boolean_t rv = B_FALSE;
1918 dsl_pool_t *dp = spa_get_dsl(spa);
1919
1920 switch (spa->spa_log_state) {
1921 case SPA_LOG_MISSING:
1922 /* need to recheck in case slog has been restored */
1923 case SPA_LOG_UNKNOWN:
1924 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1925 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1926 if (rv)
1927 spa_set_log_state(spa, SPA_LOG_MISSING);
1928 break;
1929 }
1930 return (rv);
1931}
1932
1933static boolean_t
1934spa_passivate_log(spa_t *spa)
1935{
1936 vdev_t *rvd = spa->spa_root_vdev;
1937 boolean_t slog_found = B_FALSE;
1938
1939 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1940
1941 if (!spa_has_slogs(spa))
1942 return (B_FALSE);
1943
1944 for (int c = 0; c < rvd->vdev_children; c++) {
1945 vdev_t *tvd = rvd->vdev_child[c];
1946 metaslab_group_t *mg = tvd->vdev_mg;
1947
1948 if (tvd->vdev_islog) {
1949 metaslab_group_passivate(mg);
1950 slog_found = B_TRUE;
1951 }
1952 }
1953
1954 return (slog_found);
1955}
1956
1957static void
1958spa_activate_log(spa_t *spa)
1959{
1960 vdev_t *rvd = spa->spa_root_vdev;
1961
1962 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1963
1964 for (int c = 0; c < rvd->vdev_children; c++) {
1965 vdev_t *tvd = rvd->vdev_child[c];
1966 metaslab_group_t *mg = tvd->vdev_mg;
1967
1968 if (tvd->vdev_islog)
1969 metaslab_group_activate(mg);
1970 }
1971}
1972
1973int
1974spa_reset_logs(spa_t *spa)
1975{
1976 int error;
1977
1978 error = dmu_objset_find(spa_name(spa), zil_reset,
1979 NULL, DS_FIND_CHILDREN);
1980 if (error == 0) {
1981 /*
1982 * We successfully offlined the log device, sync out the
1983 * current txg so that the "stubby" block can be removed
1984 * by zil_sync().
1985 */
1986 txg_wait_synced(spa->spa_dsl_pool, 0);
1987 }
1988 return (error);
1989}
1990
1991static void
1992spa_aux_check_removed(spa_aux_vdev_t *sav)
1993{
1994 int i;
1995
1996 for (i = 0; i < sav->sav_count; i++)
1997 spa_check_removed(sav->sav_vdevs[i]);
1998}
1999
2000void
2001spa_claim_notify(zio_t *zio)
2002{
2003 spa_t *spa = zio->io_spa;
2004
2005 if (zio->io_error)
2006 return;
2007
2008 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2009 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2010 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2011 mutex_exit(&spa->spa_props_lock);
2012}
2013
2014typedef struct spa_load_error {
2015 uint64_t sle_meta_count;
2016 uint64_t sle_data_count;
2017} spa_load_error_t;
2018
2019static void
2020spa_load_verify_done(zio_t *zio)
2021{
2022 blkptr_t *bp = zio->io_bp;
2023 spa_load_error_t *sle = zio->io_private;
2024 dmu_object_type_t type = BP_GET_TYPE(bp);
2025 int error = zio->io_error;
2026 spa_t *spa = zio->io_spa;
2027
2028 abd_free(zio->io_abd);
2029 if (error) {
2030 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2031 type != DMU_OT_INTENT_LOG)
2032 atomic_inc_64(&sle->sle_meta_count);
2033 else
2034 atomic_inc_64(&sle->sle_data_count);
2035 }
2036
2037 mutex_enter(&spa->spa_scrub_lock);
2038 spa->spa_scrub_inflight--;
2039 cv_broadcast(&spa->spa_scrub_io_cv);
2040 mutex_exit(&spa->spa_scrub_lock);
2041}
2042
2043/*
2044 * Maximum number of concurrent scrub i/os to create while verifying
2045 * a pool while importing it.
2046 */
2047int spa_load_verify_maxinflight = 10000;
2048boolean_t spa_load_verify_metadata = B_TRUE;
2049boolean_t spa_load_verify_data = B_TRUE;
2050
2051SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
2052 &spa_load_verify_maxinflight, 0,
2053 "Maximum number of concurrent scrub I/Os to create while verifying a "
2054 "pool while importing it");
2055
2056SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
2057 &spa_load_verify_metadata, 0,
2058 "Check metadata on import?");
2059
2060SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
2061 &spa_load_verify_data, 0,
2062 "Check user data on import?");
2063
2064/*ARGSUSED*/
2065static int
2066spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2067 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2068{
2069 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2070 return (0);
2071 /*
2072 * Note: normally this routine will not be called if
2073 * spa_load_verify_metadata is not set. However, it may be useful
2074 * to manually set the flag after the traversal has begun.
2075 */
2076 if (!spa_load_verify_metadata)
2077 return (0);
2078 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2079 return (0);
2080
2081 zio_t *rio = arg;
2082 size_t size = BP_GET_PSIZE(bp);
2083
2084 mutex_enter(&spa->spa_scrub_lock);
2085 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2086 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2087 spa->spa_scrub_inflight++;
2088 mutex_exit(&spa->spa_scrub_lock);
2089
2090 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2091 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2092 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2093 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2094 return (0);
2095}
2096
2097/* ARGSUSED */
2098int
2099verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2100{
2101 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2102 return (SET_ERROR(ENAMETOOLONG));
2103
2104 return (0);
2105}
2106
2107static int
2108spa_load_verify(spa_t *spa)
2109{
2110 zio_t *rio;
2111 spa_load_error_t sle = { 0 };
| 1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright 2017 Joyent, Inc.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
34 */
35
36/*
37 * SPA: Storage Pool Allocator
38 *
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
41 * pool.
42 */
43
44#include <sys/zfs_context.h>
45#include <sys/fm/fs/zfs.h>
46#include <sys/spa_impl.h>
47#include <sys/zio.h>
48#include <sys/zio_checksum.h>
49#include <sys/dmu.h>
50#include <sys/dmu_tx.h>
51#include <sys/zap.h>
52#include <sys/zil.h>
53#include <sys/ddt.h>
54#include <sys/vdev_impl.h>
55#include <sys/vdev_removal.h>
56#include <sys/vdev_indirect_mapping.h>
57#include <sys/vdev_indirect_births.h>
58#include <sys/metaslab.h>
59#include <sys/metaslab_impl.h>
60#include <sys/uberblock_impl.h>
61#include <sys/txg.h>
62#include <sys/avl.h>
63#include <sys/bpobj.h>
64#include <sys/dmu_traverse.h>
65#include <sys/dmu_objset.h>
66#include <sys/unique.h>
67#include <sys/dsl_pool.h>
68#include <sys/dsl_dataset.h>
69#include <sys/dsl_dir.h>
70#include <sys/dsl_prop.h>
71#include <sys/dsl_synctask.h>
72#include <sys/fs/zfs.h>
73#include <sys/arc.h>
74#include <sys/callb.h>
75#include <sys/spa_boot.h>
76#include <sys/zfs_ioctl.h>
77#include <sys/dsl_scan.h>
78#include <sys/dmu_send.h>
79#include <sys/dsl_destroy.h>
80#include <sys/dsl_userhold.h>
81#include <sys/zfeature.h>
82#include <sys/zvol.h>
83#include <sys/trim_map.h>
84#include <sys/abd.h>
85
86#ifdef _KERNEL
87#include <sys/callb.h>
88#include <sys/cpupart.h>
89#include <sys/zone.h>
90#endif /* _KERNEL */
91
92#include "zfs_prop.h"
93#include "zfs_comutil.h"
94
95/* Check hostid on import? */
96static int check_hostid = 1;
97
98/*
99 * The interval, in seconds, at which failed configuration cache file writes
100 * should be retried.
101 */
102int zfs_ccw_retry_interval = 300;
103
104SYSCTL_DECL(_vfs_zfs);
105SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
106 "Check hostid on import?");
107TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
108SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
109 &zfs_ccw_retry_interval, 0,
110 "Configuration cache file write, retry after failure, interval (seconds)");
111
112typedef enum zti_modes {
113 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
114 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
115 ZTI_MODE_NULL, /* don't create a taskq */
116 ZTI_NMODES
117} zti_modes_t;
118
119#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
120#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
121#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
122
123#define ZTI_N(n) ZTI_P(n, 1)
124#define ZTI_ONE ZTI_N(1)
125
126typedef struct zio_taskq_info {
127 zti_modes_t zti_mode;
128 uint_t zti_value;
129 uint_t zti_count;
130} zio_taskq_info_t;
131
132static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
133 "issue", "issue_high", "intr", "intr_high"
134};
135
136/*
137 * This table defines the taskq settings for each ZFS I/O type. When
138 * initializing a pool, we use this table to create an appropriately sized
139 * taskq. Some operations are low volume and therefore have a small, static
140 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
141 * macros. Other operations process a large amount of data; the ZTI_BATCH
142 * macro causes us to create a taskq oriented for throughput. Some operations
143 * are so high frequency and short-lived that the taskq itself can become a a
144 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
145 * additional degree of parallelism specified by the number of threads per-
146 * taskq and the number of taskqs; when dispatching an event in this case, the
147 * particular taskq is chosen at random.
148 *
149 * The different taskq priorities are to handle the different contexts (issue
150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
151 * need to be handled with minimum delay.
152 */
153const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
156 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
157 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
158 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
161};
162
163static void spa_sync_version(void *arg, dmu_tx_t *tx);
164static void spa_sync_props(void *arg, dmu_tx_t *tx);
165static boolean_t spa_has_active_shared_spare(spa_t *spa);
166static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
167static void spa_vdev_resilver_done(spa_t *spa);
168
169uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
170#ifdef PSRSET_BIND
171id_t zio_taskq_psrset_bind = PS_NONE;
172#endif
173#ifdef SYSDC
174boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
175uint_t zio_taskq_basedc = 80; /* base duty cycle */
176#endif
177
178boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
179extern int zfs_sync_pass_deferred_free;
180
181/*
182 * Report any spa_load_verify errors found, but do not fail spa_load.
183 * This is used by zdb to analyze non-idle pools.
184 */
185boolean_t spa_load_verify_dryrun = B_FALSE;
186
187/*
188 * This (illegal) pool name is used when temporarily importing a spa_t in order
189 * to get the vdev stats associated with the imported devices.
190 */
191#define TRYIMPORT_NAME "$import"
192
193/*
194 * For debugging purposes: print out vdev tree during pool import.
195 */
196int spa_load_print_vdev_tree = B_FALSE;
197
198/*
199 * A non-zero value for zfs_max_missing_tvds means that we allow importing
200 * pools with missing top-level vdevs. This is strictly intended for advanced
201 * pool recovery cases since missing data is almost inevitable. Pools with
202 * missing devices can only be imported read-only for safety reasons, and their
203 * fail-mode will be automatically set to "continue".
204 *
205 * With 1 missing vdev we should be able to import the pool and mount all
206 * datasets. User data that was not modified after the missing device has been
207 * added should be recoverable. This means that snapshots created prior to the
208 * addition of that device should be completely intact.
209 *
210 * With 2 missing vdevs, some datasets may fail to mount since there are
211 * dataset statistics that are stored as regular metadata. Some data might be
212 * recoverable if those vdevs were added recently.
213 *
214 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
215 * may be missing entirely. Chances of data recovery are very low. Note that
216 * there are also risks of performing an inadvertent rewind as we might be
217 * missing all the vdevs with the latest uberblocks.
218 */
219uint64_t zfs_max_missing_tvds = 0;
220
221/*
222 * The parameters below are similar to zfs_max_missing_tvds but are only
223 * intended for a preliminary open of the pool with an untrusted config which
224 * might be incomplete or out-dated.
225 *
226 * We are more tolerant for pools opened from a cachefile since we could have
227 * an out-dated cachefile where a device removal was not registered.
228 * We could have set the limit arbitrarily high but in the case where devices
229 * are really missing we would want to return the proper error codes; we chose
230 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
231 * and we get a chance to retrieve the trusted config.
232 */
233uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
234
235/*
236 * In the case where config was assembled by scanning device paths (/dev/dsks
237 * by default) we are less tolerant since all the existing devices should have
238 * been detected and we want spa_load to return the right error codes.
239 */
240uint64_t zfs_max_missing_tvds_scan = 0;
241
242
243SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN,
244 &spa_load_print_vdev_tree, 0,
245 "print out vdev tree during pool import");
246SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN,
247 &zfs_max_missing_tvds, 0,
248 "allow importing pools with missing top-level vdevs");
249SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN,
250 &zfs_max_missing_tvds_cachefile, 0,
251 "allow importing pools with missing top-level vdevs in cache file");
252SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN,
253 &zfs_max_missing_tvds_scan, 0,
254 "allow importing pools with missing top-level vdevs during scan");
255
256/*
257 * Debugging aid that pauses spa_sync() towards the end.
258 */
259boolean_t zfs_pause_spa_sync = B_FALSE;
260
261/*
262 * ==========================================================================
263 * SPA properties routines
264 * ==========================================================================
265 */
266
267/*
268 * Add a (source=src, propname=propval) list to an nvlist.
269 */
270static void
271spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
272 uint64_t intval, zprop_source_t src)
273{
274 const char *propname = zpool_prop_to_name(prop);
275 nvlist_t *propval;
276
277 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
278 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
279
280 if (strval != NULL)
281 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
282 else
283 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
284
285 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
286 nvlist_free(propval);
287}
288
289/*
290 * Get property values from the spa configuration.
291 */
292static void
293spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
294{
295 vdev_t *rvd = spa->spa_root_vdev;
296 dsl_pool_t *pool = spa->spa_dsl_pool;
297 uint64_t size, alloc, cap, version;
298 zprop_source_t src = ZPROP_SRC_NONE;
299 spa_config_dirent_t *dp;
300 metaslab_class_t *mc = spa_normal_class(spa);
301
302 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
303
304 if (rvd != NULL) {
305 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
306 size = metaslab_class_get_space(spa_normal_class(spa));
307 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
308 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
309 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
310 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
311 size - alloc, src);
312 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
313 spa->spa_checkpoint_info.sci_dspace, src);
314
315 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
316 metaslab_class_fragmentation(mc), src);
317 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
318 metaslab_class_expandable_space(mc), src);
319 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
320 (spa_mode(spa) == FREAD), src);
321
322 cap = (size == 0) ? 0 : (alloc * 100 / size);
323 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
324
325 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
326 ddt_get_pool_dedup_ratio(spa), src);
327
328 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
329 rvd->vdev_state, src);
330
331 version = spa_version(spa);
332 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
333 src = ZPROP_SRC_DEFAULT;
334 else
335 src = ZPROP_SRC_LOCAL;
336 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
337 }
338
339 if (pool != NULL) {
340 /*
341 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
342 * when opening pools before this version freedir will be NULL.
343 */
344 if (pool->dp_free_dir != NULL) {
345 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
346 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
347 src);
348 } else {
349 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
350 NULL, 0, src);
351 }
352
353 if (pool->dp_leak_dir != NULL) {
354 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
355 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
356 src);
357 } else {
358 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
359 NULL, 0, src);
360 }
361 }
362
363 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
364
365 if (spa->spa_comment != NULL) {
366 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
367 0, ZPROP_SRC_LOCAL);
368 }
369
370 if (spa->spa_root != NULL)
371 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
372 0, ZPROP_SRC_LOCAL);
373
374 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
375 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
376 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
377 } else {
378 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
379 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
380 }
381
382 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
383 if (dp->scd_path == NULL) {
384 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
385 "none", 0, ZPROP_SRC_LOCAL);
386 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
387 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
388 dp->scd_path, 0, ZPROP_SRC_LOCAL);
389 }
390 }
391}
392
393/*
394 * Get zpool property values.
395 */
396int
397spa_prop_get(spa_t *spa, nvlist_t **nvp)
398{
399 objset_t *mos = spa->spa_meta_objset;
400 zap_cursor_t zc;
401 zap_attribute_t za;
402 int err;
403
404 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
405
406 mutex_enter(&spa->spa_props_lock);
407
408 /*
409 * Get properties from the spa config.
410 */
411 spa_prop_get_config(spa, nvp);
412
413 /* If no pool property object, no more prop to get. */
414 if (mos == NULL || spa->spa_pool_props_object == 0) {
415 mutex_exit(&spa->spa_props_lock);
416 return (0);
417 }
418
419 /*
420 * Get properties from the MOS pool property object.
421 */
422 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
423 (err = zap_cursor_retrieve(&zc, &za)) == 0;
424 zap_cursor_advance(&zc)) {
425 uint64_t intval = 0;
426 char *strval = NULL;
427 zprop_source_t src = ZPROP_SRC_DEFAULT;
428 zpool_prop_t prop;
429
430 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
431 continue;
432
433 switch (za.za_integer_length) {
434 case 8:
435 /* integer property */
436 if (za.za_first_integer !=
437 zpool_prop_default_numeric(prop))
438 src = ZPROP_SRC_LOCAL;
439
440 if (prop == ZPOOL_PROP_BOOTFS) {
441 dsl_pool_t *dp;
442 dsl_dataset_t *ds = NULL;
443
444 dp = spa_get_dsl(spa);
445 dsl_pool_config_enter(dp, FTAG);
446 if (err = dsl_dataset_hold_obj(dp,
447 za.za_first_integer, FTAG, &ds)) {
448 dsl_pool_config_exit(dp, FTAG);
449 break;
450 }
451
452 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
453 KM_SLEEP);
454 dsl_dataset_name(ds, strval);
455 dsl_dataset_rele(ds, FTAG);
456 dsl_pool_config_exit(dp, FTAG);
457 } else {
458 strval = NULL;
459 intval = za.za_first_integer;
460 }
461
462 spa_prop_add_list(*nvp, prop, strval, intval, src);
463
464 if (strval != NULL)
465 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
466
467 break;
468
469 case 1:
470 /* string property */
471 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
472 err = zap_lookup(mos, spa->spa_pool_props_object,
473 za.za_name, 1, za.za_num_integers, strval);
474 if (err) {
475 kmem_free(strval, za.za_num_integers);
476 break;
477 }
478 spa_prop_add_list(*nvp, prop, strval, 0, src);
479 kmem_free(strval, za.za_num_integers);
480 break;
481
482 default:
483 break;
484 }
485 }
486 zap_cursor_fini(&zc);
487 mutex_exit(&spa->spa_props_lock);
488out:
489 if (err && err != ENOENT) {
490 nvlist_free(*nvp);
491 *nvp = NULL;
492 return (err);
493 }
494
495 return (0);
496}
497
498/*
499 * Validate the given pool properties nvlist and modify the list
500 * for the property values to be set.
501 */
502static int
503spa_prop_validate(spa_t *spa, nvlist_t *props)
504{
505 nvpair_t *elem;
506 int error = 0, reset_bootfs = 0;
507 uint64_t objnum = 0;
508 boolean_t has_feature = B_FALSE;
509
510 elem = NULL;
511 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
512 uint64_t intval;
513 char *strval, *slash, *check, *fname;
514 const char *propname = nvpair_name(elem);
515 zpool_prop_t prop = zpool_name_to_prop(propname);
516
517 switch (prop) {
518 case ZPOOL_PROP_INVAL:
519 if (!zpool_prop_feature(propname)) {
520 error = SET_ERROR(EINVAL);
521 break;
522 }
523
524 /*
525 * Sanitize the input.
526 */
527 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
528 error = SET_ERROR(EINVAL);
529 break;
530 }
531
532 if (nvpair_value_uint64(elem, &intval) != 0) {
533 error = SET_ERROR(EINVAL);
534 break;
535 }
536
537 if (intval != 0) {
538 error = SET_ERROR(EINVAL);
539 break;
540 }
541
542 fname = strchr(propname, '@') + 1;
543 if (zfeature_lookup_name(fname, NULL) != 0) {
544 error = SET_ERROR(EINVAL);
545 break;
546 }
547
548 has_feature = B_TRUE;
549 break;
550
551 case ZPOOL_PROP_VERSION:
552 error = nvpair_value_uint64(elem, &intval);
553 if (!error &&
554 (intval < spa_version(spa) ||
555 intval > SPA_VERSION_BEFORE_FEATURES ||
556 has_feature))
557 error = SET_ERROR(EINVAL);
558 break;
559
560 case ZPOOL_PROP_DELEGATION:
561 case ZPOOL_PROP_AUTOREPLACE:
562 case ZPOOL_PROP_LISTSNAPS:
563 case ZPOOL_PROP_AUTOEXPAND:
564 error = nvpair_value_uint64(elem, &intval);
565 if (!error && intval > 1)
566 error = SET_ERROR(EINVAL);
567 break;
568
569 case ZPOOL_PROP_BOOTFS:
570 /*
571 * If the pool version is less than SPA_VERSION_BOOTFS,
572 * or the pool is still being created (version == 0),
573 * the bootfs property cannot be set.
574 */
575 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
576 error = SET_ERROR(ENOTSUP);
577 break;
578 }
579
580 /*
581 * Make sure the vdev config is bootable
582 */
583 if (!vdev_is_bootable(spa->spa_root_vdev)) {
584 error = SET_ERROR(ENOTSUP);
585 break;
586 }
587
588 reset_bootfs = 1;
589
590 error = nvpair_value_string(elem, &strval);
591
592 if (!error) {
593 objset_t *os;
594 uint64_t propval;
595
596 if (strval == NULL || strval[0] == '\0') {
597 objnum = zpool_prop_default_numeric(
598 ZPOOL_PROP_BOOTFS);
599 break;
600 }
601
602 if (error = dmu_objset_hold(strval, FTAG, &os))
603 break;
604
605 /*
606 * Must be ZPL, and its property settings
607 * must be supported by GRUB (compression
608 * is not gzip, and large blocks are not used).
609 */
610
611 if (dmu_objset_type(os) != DMU_OST_ZFS) {
612 error = SET_ERROR(ENOTSUP);
613 } else if ((error =
614 dsl_prop_get_int_ds(dmu_objset_ds(os),
615 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
616 &propval)) == 0 &&
617 !BOOTFS_COMPRESS_VALID(propval)) {
618 error = SET_ERROR(ENOTSUP);
619 } else {
620 objnum = dmu_objset_id(os);
621 }
622 dmu_objset_rele(os, FTAG);
623 }
624 break;
625
626 case ZPOOL_PROP_FAILUREMODE:
627 error = nvpair_value_uint64(elem, &intval);
628 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
629 intval > ZIO_FAILURE_MODE_PANIC))
630 error = SET_ERROR(EINVAL);
631
632 /*
633 * This is a special case which only occurs when
634 * the pool has completely failed. This allows
635 * the user to change the in-core failmode property
636 * without syncing it out to disk (I/Os might
637 * currently be blocked). We do this by returning
638 * EIO to the caller (spa_prop_set) to trick it
639 * into thinking we encountered a property validation
640 * error.
641 */
642 if (!error && spa_suspended(spa)) {
643 spa->spa_failmode = intval;
644 error = SET_ERROR(EIO);
645 }
646 break;
647
648 case ZPOOL_PROP_CACHEFILE:
649 if ((error = nvpair_value_string(elem, &strval)) != 0)
650 break;
651
652 if (strval[0] == '\0')
653 break;
654
655 if (strcmp(strval, "none") == 0)
656 break;
657
658 if (strval[0] != '/') {
659 error = SET_ERROR(EINVAL);
660 break;
661 }
662
663 slash = strrchr(strval, '/');
664 ASSERT(slash != NULL);
665
666 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
667 strcmp(slash, "/..") == 0)
668 error = SET_ERROR(EINVAL);
669 break;
670
671 case ZPOOL_PROP_COMMENT:
672 if ((error = nvpair_value_string(elem, &strval)) != 0)
673 break;
674 for (check = strval; *check != '\0'; check++) {
675 /*
676 * The kernel doesn't have an easy isprint()
677 * check. For this kernel check, we merely
678 * check ASCII apart from DEL. Fix this if
679 * there is an easy-to-use kernel isprint().
680 */
681 if (*check >= 0x7f) {
682 error = SET_ERROR(EINVAL);
683 break;
684 }
685 }
686 if (strlen(strval) > ZPROP_MAX_COMMENT)
687 error = E2BIG;
688 break;
689
690 case ZPOOL_PROP_DEDUPDITTO:
691 if (spa_version(spa) < SPA_VERSION_DEDUP)
692 error = SET_ERROR(ENOTSUP);
693 else
694 error = nvpair_value_uint64(elem, &intval);
695 if (error == 0 &&
696 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
697 error = SET_ERROR(EINVAL);
698 break;
699 }
700
701 if (error)
702 break;
703 }
704
705 if (!error && reset_bootfs) {
706 error = nvlist_remove(props,
707 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
708
709 if (!error) {
710 error = nvlist_add_uint64(props,
711 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
712 }
713 }
714
715 return (error);
716}
717
718void
719spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
720{
721 char *cachefile;
722 spa_config_dirent_t *dp;
723
724 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
725 &cachefile) != 0)
726 return;
727
728 dp = kmem_alloc(sizeof (spa_config_dirent_t),
729 KM_SLEEP);
730
731 if (cachefile[0] == '\0')
732 dp->scd_path = spa_strdup(spa_config_path);
733 else if (strcmp(cachefile, "none") == 0)
734 dp->scd_path = NULL;
735 else
736 dp->scd_path = spa_strdup(cachefile);
737
738 list_insert_head(&spa->spa_config_list, dp);
739 if (need_sync)
740 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
741}
742
743int
744spa_prop_set(spa_t *spa, nvlist_t *nvp)
745{
746 int error;
747 nvpair_t *elem = NULL;
748 boolean_t need_sync = B_FALSE;
749
750 if ((error = spa_prop_validate(spa, nvp)) != 0)
751 return (error);
752
753 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
754 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
755
756 if (prop == ZPOOL_PROP_CACHEFILE ||
757 prop == ZPOOL_PROP_ALTROOT ||
758 prop == ZPOOL_PROP_READONLY)
759 continue;
760
761 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
762 uint64_t ver;
763
764 if (prop == ZPOOL_PROP_VERSION) {
765 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
766 } else {
767 ASSERT(zpool_prop_feature(nvpair_name(elem)));
768 ver = SPA_VERSION_FEATURES;
769 need_sync = B_TRUE;
770 }
771
772 /* Save time if the version is already set. */
773 if (ver == spa_version(spa))
774 continue;
775
776 /*
777 * In addition to the pool directory object, we might
778 * create the pool properties object, the features for
779 * read object, the features for write object, or the
780 * feature descriptions object.
781 */
782 error = dsl_sync_task(spa->spa_name, NULL,
783 spa_sync_version, &ver,
784 6, ZFS_SPACE_CHECK_RESERVED);
785 if (error)
786 return (error);
787 continue;
788 }
789
790 need_sync = B_TRUE;
791 break;
792 }
793
794 if (need_sync) {
795 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
796 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
797 }
798
799 return (0);
800}
801
802/*
803 * If the bootfs property value is dsobj, clear it.
804 */
805void
806spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
807{
808 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
809 VERIFY(zap_remove(spa->spa_meta_objset,
810 spa->spa_pool_props_object,
811 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
812 spa->spa_bootfs = 0;
813 }
814}
815
816/*ARGSUSED*/
817static int
818spa_change_guid_check(void *arg, dmu_tx_t *tx)
819{
820 uint64_t *newguid = arg;
821 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
822 vdev_t *rvd = spa->spa_root_vdev;
823 uint64_t vdev_state;
824
825 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
826 int error = (spa_has_checkpoint(spa)) ?
827 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
828 return (SET_ERROR(error));
829 }
830
831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
832 vdev_state = rvd->vdev_state;
833 spa_config_exit(spa, SCL_STATE, FTAG);
834
835 if (vdev_state != VDEV_STATE_HEALTHY)
836 return (SET_ERROR(ENXIO));
837
838 ASSERT3U(spa_guid(spa), !=, *newguid);
839
840 return (0);
841}
842
843static void
844spa_change_guid_sync(void *arg, dmu_tx_t *tx)
845{
846 uint64_t *newguid = arg;
847 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
848 uint64_t oldguid;
849 vdev_t *rvd = spa->spa_root_vdev;
850
851 oldguid = spa_guid(spa);
852
853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
854 rvd->vdev_guid = *newguid;
855 rvd->vdev_guid_sum += (*newguid - oldguid);
856 vdev_config_dirty(rvd);
857 spa_config_exit(spa, SCL_STATE, FTAG);
858
859 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
860 oldguid, *newguid);
861}
862
863/*
864 * Change the GUID for the pool. This is done so that we can later
865 * re-import a pool built from a clone of our own vdevs. We will modify
866 * the root vdev's guid, our own pool guid, and then mark all of our
867 * vdevs dirty. Note that we must make sure that all our vdevs are
868 * online when we do this, or else any vdevs that weren't present
869 * would be orphaned from our pool. We are also going to issue a
870 * sysevent to update any watchers.
871 */
872int
873spa_change_guid(spa_t *spa)
874{
875 int error;
876 uint64_t guid;
877
878 mutex_enter(&spa->spa_vdev_top_lock);
879 mutex_enter(&spa_namespace_lock);
880 guid = spa_generate_guid(NULL);
881
882 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
883 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
884
885 if (error == 0) {
886 spa_write_cachefile(spa, B_FALSE, B_TRUE);
887 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
888 }
889
890 mutex_exit(&spa_namespace_lock);
891 mutex_exit(&spa->spa_vdev_top_lock);
892
893 return (error);
894}
895
896/*
897 * ==========================================================================
898 * SPA state manipulation (open/create/destroy/import/export)
899 * ==========================================================================
900 */
901
902static int
903spa_error_entry_compare(const void *a, const void *b)
904{
905 spa_error_entry_t *sa = (spa_error_entry_t *)a;
906 spa_error_entry_t *sb = (spa_error_entry_t *)b;
907 int ret;
908
909 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
910 sizeof (zbookmark_phys_t));
911
912 if (ret < 0)
913 return (-1);
914 else if (ret > 0)
915 return (1);
916 else
917 return (0);
918}
919
920/*
921 * Utility function which retrieves copies of the current logs and
922 * re-initializes them in the process.
923 */
924void
925spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
926{
927 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
928
929 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
930 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
931
932 avl_create(&spa->spa_errlist_scrub,
933 spa_error_entry_compare, sizeof (spa_error_entry_t),
934 offsetof(spa_error_entry_t, se_avl));
935 avl_create(&spa->spa_errlist_last,
936 spa_error_entry_compare, sizeof (spa_error_entry_t),
937 offsetof(spa_error_entry_t, se_avl));
938}
939
940static void
941spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
942{
943 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
944 enum zti_modes mode = ztip->zti_mode;
945 uint_t value = ztip->zti_value;
946 uint_t count = ztip->zti_count;
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
948 char name[32];
949 uint_t flags = 0;
950 boolean_t batch = B_FALSE;
951
952 if (mode == ZTI_MODE_NULL) {
953 tqs->stqs_count = 0;
954 tqs->stqs_taskq = NULL;
955 return;
956 }
957
958 ASSERT3U(count, >, 0);
959
960 tqs->stqs_count = count;
961 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
962
963 switch (mode) {
964 case ZTI_MODE_FIXED:
965 ASSERT3U(value, >=, 1);
966 value = MAX(value, 1);
967 break;
968
969 case ZTI_MODE_BATCH:
970 batch = B_TRUE;
971 flags |= TASKQ_THREADS_CPU_PCT;
972 value = zio_taskq_batch_pct;
973 break;
974
975 default:
976 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
977 "spa_activate()",
978 zio_type_name[t], zio_taskq_types[q], mode, value);
979 break;
980 }
981
982 for (uint_t i = 0; i < count; i++) {
983 taskq_t *tq;
984
985 if (count > 1) {
986 (void) snprintf(name, sizeof (name), "%s_%s_%u",
987 zio_type_name[t], zio_taskq_types[q], i);
988 } else {
989 (void) snprintf(name, sizeof (name), "%s_%s",
990 zio_type_name[t], zio_taskq_types[q]);
991 }
992
993#ifdef SYSDC
994 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
995 if (batch)
996 flags |= TASKQ_DC_BATCH;
997
998 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
999 spa->spa_proc, zio_taskq_basedc, flags);
1000 } else {
1001#endif
1002 pri_t pri = maxclsyspri;
1003 /*
1004 * The write issue taskq can be extremely CPU
1005 * intensive. Run it at slightly lower priority
1006 * than the other taskqs.
1007 * FreeBSD notes:
1008 * - numerically higher priorities are lower priorities;
1009 * - if priorities divided by four (RQ_PPQ) are equal
1010 * then a difference between them is insignificant.
1011 */
1012 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1013#ifdef illumos
1014 pri--;
1015#else
1016 pri += 4;
1017#endif
1018
1019 tq = taskq_create_proc(name, value, pri, 50,
1020 INT_MAX, spa->spa_proc, flags);
1021#ifdef SYSDC
1022 }
1023#endif
1024
1025 tqs->stqs_taskq[i] = tq;
1026 }
1027}
1028
1029static void
1030spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1031{
1032 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1033
1034 if (tqs->stqs_taskq == NULL) {
1035 ASSERT0(tqs->stqs_count);
1036 return;
1037 }
1038
1039 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1040 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1041 taskq_destroy(tqs->stqs_taskq[i]);
1042 }
1043
1044 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1045 tqs->stqs_taskq = NULL;
1046}
1047
1048/*
1049 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1050 * Note that a type may have multiple discrete taskqs to avoid lock contention
1051 * on the taskq itself. In that case we choose which taskq at random by using
1052 * the low bits of gethrtime().
1053 */
1054void
1055spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1056 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1057{
1058 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1059 taskq_t *tq;
1060
1061 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1062 ASSERT3U(tqs->stqs_count, !=, 0);
1063
1064 if (tqs->stqs_count == 1) {
1065 tq = tqs->stqs_taskq[0];
1066 } else {
1067#ifdef _KERNEL
1068 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
1069#else
1070 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1071#endif
1072 }
1073
1074 taskq_dispatch_ent(tq, func, arg, flags, ent);
1075}
1076
1077static void
1078spa_create_zio_taskqs(spa_t *spa)
1079{
1080 for (int t = 0; t < ZIO_TYPES; t++) {
1081 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1082 spa_taskqs_init(spa, t, q);
1083 }
1084 }
1085}
1086
1087#ifdef _KERNEL
1088#ifdef SPA_PROCESS
1089static void
1090spa_thread(void *arg)
1091{
1092 callb_cpr_t cprinfo;
1093
1094 spa_t *spa = arg;
1095 user_t *pu = PTOU(curproc);
1096
1097 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1098 spa->spa_name);
1099
1100 ASSERT(curproc != &p0);
1101 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1102 "zpool-%s", spa->spa_name);
1103 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1104
1105#ifdef PSRSET_BIND
1106 /* bind this thread to the requested psrset */
1107 if (zio_taskq_psrset_bind != PS_NONE) {
1108 pool_lock();
1109 mutex_enter(&cpu_lock);
1110 mutex_enter(&pidlock);
1111 mutex_enter(&curproc->p_lock);
1112
1113 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1114 0, NULL, NULL) == 0) {
1115 curthread->t_bind_pset = zio_taskq_psrset_bind;
1116 } else {
1117 cmn_err(CE_WARN,
1118 "Couldn't bind process for zfs pool \"%s\" to "
1119 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1120 }
1121
1122 mutex_exit(&curproc->p_lock);
1123 mutex_exit(&pidlock);
1124 mutex_exit(&cpu_lock);
1125 pool_unlock();
1126 }
1127#endif
1128
1129#ifdef SYSDC
1130 if (zio_taskq_sysdc) {
1131 sysdc_thread_enter(curthread, 100, 0);
1132 }
1133#endif
1134
1135 spa->spa_proc = curproc;
1136 spa->spa_did = curthread->t_did;
1137
1138 spa_create_zio_taskqs(spa);
1139
1140 mutex_enter(&spa->spa_proc_lock);
1141 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1142
1143 spa->spa_proc_state = SPA_PROC_ACTIVE;
1144 cv_broadcast(&spa->spa_proc_cv);
1145
1146 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1147 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1148 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1149 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1150
1151 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1152 spa->spa_proc_state = SPA_PROC_GONE;
1153 spa->spa_proc = &p0;
1154 cv_broadcast(&spa->spa_proc_cv);
1155 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1156
1157 mutex_enter(&curproc->p_lock);
1158 lwp_exit();
1159}
1160#endif /* SPA_PROCESS */
1161#endif
1162
1163/*
1164 * Activate an uninitialized pool.
1165 */
1166static void
1167spa_activate(spa_t *spa, int mode)
1168{
1169 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1170
1171 spa->spa_state = POOL_STATE_ACTIVE;
1172 spa->spa_mode = mode;
1173
1174 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1175 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1176
1177 /* Try to create a covering process */
1178 mutex_enter(&spa->spa_proc_lock);
1179 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1180 ASSERT(spa->spa_proc == &p0);
1181 spa->spa_did = 0;
1182
1183#ifdef SPA_PROCESS
1184 /* Only create a process if we're going to be around a while. */
1185 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1186 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1187 NULL, 0) == 0) {
1188 spa->spa_proc_state = SPA_PROC_CREATED;
1189 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1190 cv_wait(&spa->spa_proc_cv,
1191 &spa->spa_proc_lock);
1192 }
1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1194 ASSERT(spa->spa_proc != &p0);
1195 ASSERT(spa->spa_did != 0);
1196 } else {
1197#ifdef _KERNEL
1198 cmn_err(CE_WARN,
1199 "Couldn't create process for zfs pool \"%s\"\n",
1200 spa->spa_name);
1201#endif
1202 }
1203 }
1204#endif /* SPA_PROCESS */
1205 mutex_exit(&spa->spa_proc_lock);
1206
1207 /* If we didn't create a process, we need to create our taskqs. */
1208 ASSERT(spa->spa_proc == &p0);
1209 if (spa->spa_proc == &p0) {
1210 spa_create_zio_taskqs(spa);
1211 }
1212
1213 /*
1214 * Start TRIM thread.
1215 */
1216 trim_thread_create(spa);
1217
1218 for (size_t i = 0; i < TXG_SIZE; i++)
1219 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0);
1220
1221 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1222 offsetof(vdev_t, vdev_config_dirty_node));
1223 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1224 offsetof(objset_t, os_evicting_node));
1225 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1226 offsetof(vdev_t, vdev_state_dirty_node));
1227
1228 txg_list_create(&spa->spa_vdev_txg_list, spa,
1229 offsetof(struct vdev, vdev_txg_node));
1230
1231 avl_create(&spa->spa_errlist_scrub,
1232 spa_error_entry_compare, sizeof (spa_error_entry_t),
1233 offsetof(spa_error_entry_t, se_avl));
1234 avl_create(&spa->spa_errlist_last,
1235 spa_error_entry_compare, sizeof (spa_error_entry_t),
1236 offsetof(spa_error_entry_t, se_avl));
1237}
1238
1239/*
1240 * Opposite of spa_activate().
1241 */
1242static void
1243spa_deactivate(spa_t *spa)
1244{
1245 ASSERT(spa->spa_sync_on == B_FALSE);
1246 ASSERT(spa->spa_dsl_pool == NULL);
1247 ASSERT(spa->spa_root_vdev == NULL);
1248 ASSERT(spa->spa_async_zio_root == NULL);
1249 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1250
1251 /*
1252 * Stop TRIM thread in case spa_unload() wasn't called directly
1253 * before spa_deactivate().
1254 */
1255 trim_thread_destroy(spa);
1256
1257 spa_evicting_os_wait(spa);
1258
1259 txg_list_destroy(&spa->spa_vdev_txg_list);
1260
1261 list_destroy(&spa->spa_config_dirty_list);
1262 list_destroy(&spa->spa_evicting_os_list);
1263 list_destroy(&spa->spa_state_dirty_list);
1264
1265 for (int t = 0; t < ZIO_TYPES; t++) {
1266 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1267 spa_taskqs_fini(spa, t, q);
1268 }
1269 }
1270
1271 for (size_t i = 0; i < TXG_SIZE; i++) {
1272 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1273 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1274 spa->spa_txg_zio[i] = NULL;
1275 }
1276
1277 metaslab_class_destroy(spa->spa_normal_class);
1278 spa->spa_normal_class = NULL;
1279
1280 metaslab_class_destroy(spa->spa_log_class);
1281 spa->spa_log_class = NULL;
1282
1283 /*
1284 * If this was part of an import or the open otherwise failed, we may
1285 * still have errors left in the queues. Empty them just in case.
1286 */
1287 spa_errlog_drain(spa);
1288
1289 avl_destroy(&spa->spa_errlist_scrub);
1290 avl_destroy(&spa->spa_errlist_last);
1291
1292 spa->spa_state = POOL_STATE_UNINITIALIZED;
1293
1294 mutex_enter(&spa->spa_proc_lock);
1295 if (spa->spa_proc_state != SPA_PROC_NONE) {
1296 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1297 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1298 cv_broadcast(&spa->spa_proc_cv);
1299 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1300 ASSERT(spa->spa_proc != &p0);
1301 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1302 }
1303 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1304 spa->spa_proc_state = SPA_PROC_NONE;
1305 }
1306 ASSERT(spa->spa_proc == &p0);
1307 mutex_exit(&spa->spa_proc_lock);
1308
1309#ifdef SPA_PROCESS
1310 /*
1311 * We want to make sure spa_thread() has actually exited the ZFS
1312 * module, so that the module can't be unloaded out from underneath
1313 * it.
1314 */
1315 if (spa->spa_did != 0) {
1316 thread_join(spa->spa_did);
1317 spa->spa_did = 0;
1318 }
1319#endif /* SPA_PROCESS */
1320}
1321
1322/*
1323 * Verify a pool configuration, and construct the vdev tree appropriately. This
1324 * will create all the necessary vdevs in the appropriate layout, with each vdev
1325 * in the CLOSED state. This will prep the pool before open/creation/import.
1326 * All vdev validation is done by the vdev_alloc() routine.
1327 */
1328static int
1329spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1330 uint_t id, int atype)
1331{
1332 nvlist_t **child;
1333 uint_t children;
1334 int error;
1335
1336 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1337 return (error);
1338
1339 if ((*vdp)->vdev_ops->vdev_op_leaf)
1340 return (0);
1341
1342 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1343 &child, &children);
1344
1345 if (error == ENOENT)
1346 return (0);
1347
1348 if (error) {
1349 vdev_free(*vdp);
1350 *vdp = NULL;
1351 return (SET_ERROR(EINVAL));
1352 }
1353
1354 for (int c = 0; c < children; c++) {
1355 vdev_t *vd;
1356 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1357 atype)) != 0) {
1358 vdev_free(*vdp);
1359 *vdp = NULL;
1360 return (error);
1361 }
1362 }
1363
1364 ASSERT(*vdp != NULL);
1365
1366 return (0);
1367}
1368
1369/*
1370 * Opposite of spa_load().
1371 */
1372static void
1373spa_unload(spa_t *spa)
1374{
1375 int i;
1376
1377 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1378
1379 spa_load_note(spa, "UNLOADING");
1380
1381 /*
1382 * Stop TRIM thread.
1383 */
1384 trim_thread_destroy(spa);
1385
1386 /*
1387 * Stop async tasks.
1388 */
1389 spa_async_suspend(spa);
1390
1391 /*
1392 * Stop syncing.
1393 */
1394 if (spa->spa_sync_on) {
1395 txg_sync_stop(spa->spa_dsl_pool);
1396 spa->spa_sync_on = B_FALSE;
1397 }
1398
1399 /*
1400 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1401 * to call it earlier, before we wait for async i/o to complete.
1402 * This ensures that there is no async metaslab prefetching, by
1403 * calling taskq_wait(mg_taskq).
1404 */
1405 if (spa->spa_root_vdev != NULL) {
1406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1407 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1408 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1409 spa_config_exit(spa, SCL_ALL, FTAG);
1410 }
1411
1412 /*
1413 * Wait for any outstanding async I/O to complete.
1414 */
1415 if (spa->spa_async_zio_root != NULL) {
1416 for (int i = 0; i < max_ncpus; i++)
1417 (void) zio_wait(spa->spa_async_zio_root[i]);
1418 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1419 spa->spa_async_zio_root = NULL;
1420 }
1421
1422 if (spa->spa_vdev_removal != NULL) {
1423 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1424 spa->spa_vdev_removal = NULL;
1425 }
1426
1427 if (spa->spa_condense_zthr != NULL) {
1428 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1429 zthr_destroy(spa->spa_condense_zthr);
1430 spa->spa_condense_zthr = NULL;
1431 }
1432
1433 if (spa->spa_checkpoint_discard_zthr != NULL) {
1434 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1435 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1436 spa->spa_checkpoint_discard_zthr = NULL;
1437 }
1438
1439 spa_condense_fini(spa);
1440
1441 bpobj_close(&spa->spa_deferred_bpobj);
1442
1443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1444
1445 /*
1446 * Close all vdevs.
1447 */
1448 if (spa->spa_root_vdev)
1449 vdev_free(spa->spa_root_vdev);
1450 ASSERT(spa->spa_root_vdev == NULL);
1451
1452 /*
1453 * Close the dsl pool.
1454 */
1455 if (spa->spa_dsl_pool) {
1456 dsl_pool_close(spa->spa_dsl_pool);
1457 spa->spa_dsl_pool = NULL;
1458 spa->spa_meta_objset = NULL;
1459 }
1460
1461 ddt_unload(spa);
1462
1463 /*
1464 * Drop and purge level 2 cache
1465 */
1466 spa_l2cache_drop(spa);
1467
1468 for (i = 0; i < spa->spa_spares.sav_count; i++)
1469 vdev_free(spa->spa_spares.sav_vdevs[i]);
1470 if (spa->spa_spares.sav_vdevs) {
1471 kmem_free(spa->spa_spares.sav_vdevs,
1472 spa->spa_spares.sav_count * sizeof (void *));
1473 spa->spa_spares.sav_vdevs = NULL;
1474 }
1475 if (spa->spa_spares.sav_config) {
1476 nvlist_free(spa->spa_spares.sav_config);
1477 spa->spa_spares.sav_config = NULL;
1478 }
1479 spa->spa_spares.sav_count = 0;
1480
1481 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1482 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1483 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1484 }
1485 if (spa->spa_l2cache.sav_vdevs) {
1486 kmem_free(spa->spa_l2cache.sav_vdevs,
1487 spa->spa_l2cache.sav_count * sizeof (void *));
1488 spa->spa_l2cache.sav_vdevs = NULL;
1489 }
1490 if (spa->spa_l2cache.sav_config) {
1491 nvlist_free(spa->spa_l2cache.sav_config);
1492 spa->spa_l2cache.sav_config = NULL;
1493 }
1494 spa->spa_l2cache.sav_count = 0;
1495
1496 spa->spa_async_suspended = 0;
1497
1498 spa->spa_indirect_vdevs_loaded = B_FALSE;
1499
1500 if (spa->spa_comment != NULL) {
1501 spa_strfree(spa->spa_comment);
1502 spa->spa_comment = NULL;
1503 }
1504
1505 spa_config_exit(spa, SCL_ALL, FTAG);
1506}
1507
1508/*
1509 * Load (or re-load) the current list of vdevs describing the active spares for
1510 * this pool. When this is called, we have some form of basic information in
1511 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1512 * then re-generate a more complete list including status information.
1513 */
1514void
1515spa_load_spares(spa_t *spa)
1516{
1517 nvlist_t **spares;
1518 uint_t nspares;
1519 int i;
1520 vdev_t *vd, *tvd;
1521
1522#ifndef _KERNEL
1523 /*
1524 * zdb opens both the current state of the pool and the
1525 * checkpointed state (if present), with a different spa_t.
1526 *
1527 * As spare vdevs are shared among open pools, we skip loading
1528 * them when we load the checkpointed state of the pool.
1529 */
1530 if (!spa_writeable(spa))
1531 return;
1532#endif
1533
1534 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1535
1536 /*
1537 * First, close and free any existing spare vdevs.
1538 */
1539 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1540 vd = spa->spa_spares.sav_vdevs[i];
1541
1542 /* Undo the call to spa_activate() below */
1543 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1544 B_FALSE)) != NULL && tvd->vdev_isspare)
1545 spa_spare_remove(tvd);
1546 vdev_close(vd);
1547 vdev_free(vd);
1548 }
1549
1550 if (spa->spa_spares.sav_vdevs)
1551 kmem_free(spa->spa_spares.sav_vdevs,
1552 spa->spa_spares.sav_count * sizeof (void *));
1553
1554 if (spa->spa_spares.sav_config == NULL)
1555 nspares = 0;
1556 else
1557 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1558 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1559
1560 spa->spa_spares.sav_count = (int)nspares;
1561 spa->spa_spares.sav_vdevs = NULL;
1562
1563 if (nspares == 0)
1564 return;
1565
1566 /*
1567 * Construct the array of vdevs, opening them to get status in the
1568 * process. For each spare, there is potentially two different vdev_t
1569 * structures associated with it: one in the list of spares (used only
1570 * for basic validation purposes) and one in the active vdev
1571 * configuration (if it's spared in). During this phase we open and
1572 * validate each vdev on the spare list. If the vdev also exists in the
1573 * active configuration, then we also mark this vdev as an active spare.
1574 */
1575 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1576 KM_SLEEP);
1577 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1578 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1579 VDEV_ALLOC_SPARE) == 0);
1580 ASSERT(vd != NULL);
1581
1582 spa->spa_spares.sav_vdevs[i] = vd;
1583
1584 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1585 B_FALSE)) != NULL) {
1586 if (!tvd->vdev_isspare)
1587 spa_spare_add(tvd);
1588
1589 /*
1590 * We only mark the spare active if we were successfully
1591 * able to load the vdev. Otherwise, importing a pool
1592 * with a bad active spare would result in strange
1593 * behavior, because multiple pool would think the spare
1594 * is actively in use.
1595 *
1596 * There is a vulnerability here to an equally bizarre
1597 * circumstance, where a dead active spare is later
1598 * brought back to life (onlined or otherwise). Given
1599 * the rarity of this scenario, and the extra complexity
1600 * it adds, we ignore the possibility.
1601 */
1602 if (!vdev_is_dead(tvd))
1603 spa_spare_activate(tvd);
1604 }
1605
1606 vd->vdev_top = vd;
1607 vd->vdev_aux = &spa->spa_spares;
1608
1609 if (vdev_open(vd) != 0)
1610 continue;
1611
1612 if (vdev_validate_aux(vd) == 0)
1613 spa_spare_add(vd);
1614 }
1615
1616 /*
1617 * Recompute the stashed list of spares, with status information
1618 * this time.
1619 */
1620 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1621 DATA_TYPE_NVLIST_ARRAY) == 0);
1622
1623 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1624 KM_SLEEP);
1625 for (i = 0; i < spa->spa_spares.sav_count; i++)
1626 spares[i] = vdev_config_generate(spa,
1627 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1628 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1629 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1630 for (i = 0; i < spa->spa_spares.sav_count; i++)
1631 nvlist_free(spares[i]);
1632 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1633}
1634
1635/*
1636 * Load (or re-load) the current list of vdevs describing the active l2cache for
1637 * this pool. When this is called, we have some form of basic information in
1638 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1639 * then re-generate a more complete list including status information.
1640 * Devices which are already active have their details maintained, and are
1641 * not re-opened.
1642 */
1643void
1644spa_load_l2cache(spa_t *spa)
1645{
1646 nvlist_t **l2cache;
1647 uint_t nl2cache;
1648 int i, j, oldnvdevs;
1649 uint64_t guid;
1650 vdev_t *vd, **oldvdevs, **newvdevs;
1651 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1652
1653#ifndef _KERNEL
1654 /*
1655 * zdb opens both the current state of the pool and the
1656 * checkpointed state (if present), with a different spa_t.
1657 *
1658 * As L2 caches are part of the ARC which is shared among open
1659 * pools, we skip loading them when we load the checkpointed
1660 * state of the pool.
1661 */
1662 if (!spa_writeable(spa))
1663 return;
1664#endif
1665
1666 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1667
1668 if (sav->sav_config != NULL) {
1669 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1670 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1671 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1672 } else {
1673 nl2cache = 0;
1674 newvdevs = NULL;
1675 }
1676
1677 oldvdevs = sav->sav_vdevs;
1678 oldnvdevs = sav->sav_count;
1679 sav->sav_vdevs = NULL;
1680 sav->sav_count = 0;
1681
1682 /*
1683 * Process new nvlist of vdevs.
1684 */
1685 for (i = 0; i < nl2cache; i++) {
1686 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1687 &guid) == 0);
1688
1689 newvdevs[i] = NULL;
1690 for (j = 0; j < oldnvdevs; j++) {
1691 vd = oldvdevs[j];
1692 if (vd != NULL && guid == vd->vdev_guid) {
1693 /*
1694 * Retain previous vdev for add/remove ops.
1695 */
1696 newvdevs[i] = vd;
1697 oldvdevs[j] = NULL;
1698 break;
1699 }
1700 }
1701
1702 if (newvdevs[i] == NULL) {
1703 /*
1704 * Create new vdev
1705 */
1706 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1707 VDEV_ALLOC_L2CACHE) == 0);
1708 ASSERT(vd != NULL);
1709 newvdevs[i] = vd;
1710
1711 /*
1712 * Commit this vdev as an l2cache device,
1713 * even if it fails to open.
1714 */
1715 spa_l2cache_add(vd);
1716
1717 vd->vdev_top = vd;
1718 vd->vdev_aux = sav;
1719
1720 spa_l2cache_activate(vd);
1721
1722 if (vdev_open(vd) != 0)
1723 continue;
1724
1725 (void) vdev_validate_aux(vd);
1726
1727 if (!vdev_is_dead(vd))
1728 l2arc_add_vdev(spa, vd);
1729 }
1730 }
1731
1732 /*
1733 * Purge vdevs that were dropped
1734 */
1735 for (i = 0; i < oldnvdevs; i++) {
1736 uint64_t pool;
1737
1738 vd = oldvdevs[i];
1739 if (vd != NULL) {
1740 ASSERT(vd->vdev_isl2cache);
1741
1742 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1743 pool != 0ULL && l2arc_vdev_present(vd))
1744 l2arc_remove_vdev(vd);
1745 vdev_clear_stats(vd);
1746 vdev_free(vd);
1747 }
1748 }
1749
1750 if (oldvdevs)
1751 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1752
1753 if (sav->sav_config == NULL)
1754 goto out;
1755
1756 sav->sav_vdevs = newvdevs;
1757 sav->sav_count = (int)nl2cache;
1758
1759 /*
1760 * Recompute the stashed list of l2cache devices, with status
1761 * information this time.
1762 */
1763 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1764 DATA_TYPE_NVLIST_ARRAY) == 0);
1765
1766 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1767 for (i = 0; i < sav->sav_count; i++)
1768 l2cache[i] = vdev_config_generate(spa,
1769 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1770 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1771 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1772out:
1773 for (i = 0; i < sav->sav_count; i++)
1774 nvlist_free(l2cache[i]);
1775 if (sav->sav_count)
1776 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1777}
1778
1779static int
1780load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1781{
1782 dmu_buf_t *db;
1783 char *packed = NULL;
1784 size_t nvsize = 0;
1785 int error;
1786 *value = NULL;
1787
1788 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1789 if (error != 0)
1790 return (error);
1791
1792 nvsize = *(uint64_t *)db->db_data;
1793 dmu_buf_rele(db, FTAG);
1794
1795 packed = kmem_alloc(nvsize, KM_SLEEP);
1796 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1797 DMU_READ_PREFETCH);
1798 if (error == 0)
1799 error = nvlist_unpack(packed, nvsize, value, 0);
1800 kmem_free(packed, nvsize);
1801
1802 return (error);
1803}
1804
1805/*
1806 * Concrete top-level vdevs that are not missing and are not logs. At every
1807 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1808 */
1809static uint64_t
1810spa_healthy_core_tvds(spa_t *spa)
1811{
1812 vdev_t *rvd = spa->spa_root_vdev;
1813 uint64_t tvds = 0;
1814
1815 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1816 vdev_t *vd = rvd->vdev_child[i];
1817 if (vd->vdev_islog)
1818 continue;
1819 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1820 tvds++;
1821 }
1822
1823 return (tvds);
1824}
1825
1826/*
1827 * Checks to see if the given vdev could not be opened, in which case we post a
1828 * sysevent to notify the autoreplace code that the device has been removed.
1829 */
1830static void
1831spa_check_removed(vdev_t *vd)
1832{
1833 for (uint64_t c = 0; c < vd->vdev_children; c++)
1834 spa_check_removed(vd->vdev_child[c]);
1835
1836 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1837 vdev_is_concrete(vd)) {
1838 zfs_post_autoreplace(vd->vdev_spa, vd);
1839 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1840 }
1841}
1842
1843static int
1844spa_check_for_missing_logs(spa_t *spa)
1845{
1846 vdev_t *rvd = spa->spa_root_vdev;
1847
1848 /*
1849 * If we're doing a normal import, then build up any additional
1850 * diagnostic information about missing log devices.
1851 * We'll pass this up to the user for further processing.
1852 */
1853 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1854 nvlist_t **child, *nv;
1855 uint64_t idx = 0;
1856
1857 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1858 KM_SLEEP);
1859 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1860
1861 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1862 vdev_t *tvd = rvd->vdev_child[c];
1863
1864 /*
1865 * We consider a device as missing only if it failed
1866 * to open (i.e. offline or faulted is not considered
1867 * as missing).
1868 */
1869 if (tvd->vdev_islog &&
1870 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1871 child[idx++] = vdev_config_generate(spa, tvd,
1872 B_FALSE, VDEV_CONFIG_MISSING);
1873 }
1874 }
1875
1876 if (idx > 0) {
1877 fnvlist_add_nvlist_array(nv,
1878 ZPOOL_CONFIG_CHILDREN, child, idx);
1879 fnvlist_add_nvlist(spa->spa_load_info,
1880 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1881
1882 for (uint64_t i = 0; i < idx; i++)
1883 nvlist_free(child[i]);
1884 }
1885 nvlist_free(nv);
1886 kmem_free(child, rvd->vdev_children * sizeof (char **));
1887
1888 if (idx > 0) {
1889 spa_load_failed(spa, "some log devices are missing");
1890 vdev_dbgmsg_print_tree(rvd, 2);
1891 return (SET_ERROR(ENXIO));
1892 }
1893 } else {
1894 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1895 vdev_t *tvd = rvd->vdev_child[c];
1896
1897 if (tvd->vdev_islog &&
1898 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1899 spa_set_log_state(spa, SPA_LOG_CLEAR);
1900 spa_load_note(spa, "some log devices are "
1901 "missing, ZIL is dropped.");
1902 vdev_dbgmsg_print_tree(rvd, 2);
1903 break;
1904 }
1905 }
1906 }
1907
1908 return (0);
1909}
1910
1911/*
1912 * Check for missing log devices
1913 */
1914static boolean_t
1915spa_check_logs(spa_t *spa)
1916{
1917 boolean_t rv = B_FALSE;
1918 dsl_pool_t *dp = spa_get_dsl(spa);
1919
1920 switch (spa->spa_log_state) {
1921 case SPA_LOG_MISSING:
1922 /* need to recheck in case slog has been restored */
1923 case SPA_LOG_UNKNOWN:
1924 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1925 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1926 if (rv)
1927 spa_set_log_state(spa, SPA_LOG_MISSING);
1928 break;
1929 }
1930 return (rv);
1931}
1932
1933static boolean_t
1934spa_passivate_log(spa_t *spa)
1935{
1936 vdev_t *rvd = spa->spa_root_vdev;
1937 boolean_t slog_found = B_FALSE;
1938
1939 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1940
1941 if (!spa_has_slogs(spa))
1942 return (B_FALSE);
1943
1944 for (int c = 0; c < rvd->vdev_children; c++) {
1945 vdev_t *tvd = rvd->vdev_child[c];
1946 metaslab_group_t *mg = tvd->vdev_mg;
1947
1948 if (tvd->vdev_islog) {
1949 metaslab_group_passivate(mg);
1950 slog_found = B_TRUE;
1951 }
1952 }
1953
1954 return (slog_found);
1955}
1956
1957static void
1958spa_activate_log(spa_t *spa)
1959{
1960 vdev_t *rvd = spa->spa_root_vdev;
1961
1962 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1963
1964 for (int c = 0; c < rvd->vdev_children; c++) {
1965 vdev_t *tvd = rvd->vdev_child[c];
1966 metaslab_group_t *mg = tvd->vdev_mg;
1967
1968 if (tvd->vdev_islog)
1969 metaslab_group_activate(mg);
1970 }
1971}
1972
1973int
1974spa_reset_logs(spa_t *spa)
1975{
1976 int error;
1977
1978 error = dmu_objset_find(spa_name(spa), zil_reset,
1979 NULL, DS_FIND_CHILDREN);
1980 if (error == 0) {
1981 /*
1982 * We successfully offlined the log device, sync out the
1983 * current txg so that the "stubby" block can be removed
1984 * by zil_sync().
1985 */
1986 txg_wait_synced(spa->spa_dsl_pool, 0);
1987 }
1988 return (error);
1989}
1990
1991static void
1992spa_aux_check_removed(spa_aux_vdev_t *sav)
1993{
1994 int i;
1995
1996 for (i = 0; i < sav->sav_count; i++)
1997 spa_check_removed(sav->sav_vdevs[i]);
1998}
1999
2000void
2001spa_claim_notify(zio_t *zio)
2002{
2003 spa_t *spa = zio->io_spa;
2004
2005 if (zio->io_error)
2006 return;
2007
2008 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2009 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2010 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2011 mutex_exit(&spa->spa_props_lock);
2012}
2013
2014typedef struct spa_load_error {
2015 uint64_t sle_meta_count;
2016 uint64_t sle_data_count;
2017} spa_load_error_t;
2018
2019static void
2020spa_load_verify_done(zio_t *zio)
2021{
2022 blkptr_t *bp = zio->io_bp;
2023 spa_load_error_t *sle = zio->io_private;
2024 dmu_object_type_t type = BP_GET_TYPE(bp);
2025 int error = zio->io_error;
2026 spa_t *spa = zio->io_spa;
2027
2028 abd_free(zio->io_abd);
2029 if (error) {
2030 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2031 type != DMU_OT_INTENT_LOG)
2032 atomic_inc_64(&sle->sle_meta_count);
2033 else
2034 atomic_inc_64(&sle->sle_data_count);
2035 }
2036
2037 mutex_enter(&spa->spa_scrub_lock);
2038 spa->spa_scrub_inflight--;
2039 cv_broadcast(&spa->spa_scrub_io_cv);
2040 mutex_exit(&spa->spa_scrub_lock);
2041}
2042
2043/*
2044 * Maximum number of concurrent scrub i/os to create while verifying
2045 * a pool while importing it.
2046 */
2047int spa_load_verify_maxinflight = 10000;
2048boolean_t spa_load_verify_metadata = B_TRUE;
2049boolean_t spa_load_verify_data = B_TRUE;
2050
2051SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
2052 &spa_load_verify_maxinflight, 0,
2053 "Maximum number of concurrent scrub I/Os to create while verifying a "
2054 "pool while importing it");
2055
2056SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
2057 &spa_load_verify_metadata, 0,
2058 "Check metadata on import?");
2059
2060SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
2061 &spa_load_verify_data, 0,
2062 "Check user data on import?");
2063
2064/*ARGSUSED*/
2065static int
2066spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2067 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2068{
2069 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2070 return (0);
2071 /*
2072 * Note: normally this routine will not be called if
2073 * spa_load_verify_metadata is not set. However, it may be useful
2074 * to manually set the flag after the traversal has begun.
2075 */
2076 if (!spa_load_verify_metadata)
2077 return (0);
2078 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2079 return (0);
2080
2081 zio_t *rio = arg;
2082 size_t size = BP_GET_PSIZE(bp);
2083
2084 mutex_enter(&spa->spa_scrub_lock);
2085 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2086 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2087 spa->spa_scrub_inflight++;
2088 mutex_exit(&spa->spa_scrub_lock);
2089
2090 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2091 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2092 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2093 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2094 return (0);
2095}
2096
2097/* ARGSUSED */
2098int
2099verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2100{
2101 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2102 return (SET_ERROR(ENAMETOOLONG));
2103
2104 return (0);
2105}
2106
2107static int
2108spa_load_verify(spa_t *spa)
2109{
2110 zio_t *rio;
2111 spa_load_error_t sle = { 0 };
|
2112 zpool_rewind_policy_t policy;
| 2112 zpool_load_policy_t policy;
|
2113 boolean_t verify_ok = B_FALSE;
2114 int error = 0;
2115
| 2113 boolean_t verify_ok = B_FALSE;
2114 int error = 0;
2115
|
2116 zpool_get_rewind_policy(spa->spa_config, &policy);
| 2116 zpool_get_load_policy(spa->spa_config, &policy);
|
2117
| 2117
|
2118 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
| 2118 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
|
2119 return (0);
2120
2121 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2122 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2123 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2124 DS_FIND_CHILDREN);
2125 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2126 if (error != 0)
2127 return (error);
2128
2129 rio = zio_root(spa, NULL, &sle,
2130 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2131
2132 if (spa_load_verify_metadata) {
2133 if (spa->spa_extreme_rewind) {
2134 spa_load_note(spa, "performing a complete scan of the "
2135 "pool since extreme rewind is on. This may take "
2136 "a very long time.\n (spa_load_verify_data=%u, "
2137 "spa_load_verify_metadata=%u)",
2138 spa_load_verify_data, spa_load_verify_metadata);
2139 }
2140 error = traverse_pool(spa, spa->spa_verify_min_txg,
2141 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2142 spa_load_verify_cb, rio);
2143 }
2144
2145 (void) zio_wait(rio);
2146
2147 spa->spa_load_meta_errors = sle.sle_meta_count;
2148 spa->spa_load_data_errors = sle.sle_data_count;
2149
2150 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2151 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2152 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2153 (u_longlong_t)sle.sle_data_count);
2154 }
2155
2156 if (spa_load_verify_dryrun ||
| 2119 return (0);
2120
2121 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2122 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2123 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2124 DS_FIND_CHILDREN);
2125 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2126 if (error != 0)
2127 return (error);
2128
2129 rio = zio_root(spa, NULL, &sle,
2130 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2131
2132 if (spa_load_verify_metadata) {
2133 if (spa->spa_extreme_rewind) {
2134 spa_load_note(spa, "performing a complete scan of the "
2135 "pool since extreme rewind is on. This may take "
2136 "a very long time.\n (spa_load_verify_data=%u, "
2137 "spa_load_verify_metadata=%u)",
2138 spa_load_verify_data, spa_load_verify_metadata);
2139 }
2140 error = traverse_pool(spa, spa->spa_verify_min_txg,
2141 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2142 spa_load_verify_cb, rio);
2143 }
2144
2145 (void) zio_wait(rio);
2146
2147 spa->spa_load_meta_errors = sle.sle_meta_count;
2148 spa->spa_load_data_errors = sle.sle_data_count;
2149
2150 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2151 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2152 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2153 (u_longlong_t)sle.sle_data_count);
2154 }
2155
2156 if (spa_load_verify_dryrun ||
|
2157 (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2158 sle.sle_data_count <= policy.zrp_maxdata)) {
| 2157 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2158 sle.sle_data_count <= policy.zlp_maxdata)) {
|
2159 int64_t loss = 0;
2160
2161 verify_ok = B_TRUE;
2162 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2163 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2164
2165 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2166 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2167 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2168 VERIFY(nvlist_add_int64(spa->spa_load_info,
2169 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2170 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2171 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2172 } else {
2173 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2174 }
2175
2176 if (spa_load_verify_dryrun)
2177 return (0);
2178
2179 if (error) {
2180 if (error != ENXIO && error != EIO)
2181 error = SET_ERROR(EIO);
2182 return (error);
2183 }
2184
2185 return (verify_ok ? 0 : EIO);
2186}
2187
2188/*
2189 * Find a value in the pool props object.
2190 */
2191static void
2192spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2193{
2194 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2195 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2196}
2197
2198/*
2199 * Find a value in the pool directory object.
2200 */
2201static int
2202spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2203{
2204 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2205 name, sizeof (uint64_t), 1, val);
2206
2207 if (error != 0 && (error != ENOENT || log_enoent)) {
2208 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2209 "[error=%d]", name, error);
2210 }
2211
2212 return (error);
2213}
2214
2215static int
2216spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2217{
2218 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2219 return (SET_ERROR(err));
2220}
2221
2222static void
2223spa_spawn_aux_threads(spa_t *spa)
2224{
2225 ASSERT(spa_writeable(spa));
2226
2227 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2228
2229 spa_start_indirect_condensing_thread(spa);
2230
2231 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2232 spa->spa_checkpoint_discard_zthr =
2233 zthr_create(spa_checkpoint_discard_thread_check,
2234 spa_checkpoint_discard_thread, spa);
2235}
2236
2237/*
2238 * Fix up config after a partly-completed split. This is done with the
2239 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2240 * pool have that entry in their config, but only the splitting one contains
2241 * a list of all the guids of the vdevs that are being split off.
2242 *
2243 * This function determines what to do with that list: either rejoin
2244 * all the disks to the pool, or complete the splitting process. To attempt
2245 * the rejoin, each disk that is offlined is marked online again, and
2246 * we do a reopen() call. If the vdev label for every disk that was
2247 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2248 * then we call vdev_split() on each disk, and complete the split.
2249 *
2250 * Otherwise we leave the config alone, with all the vdevs in place in
2251 * the original pool.
2252 */
2253static void
2254spa_try_repair(spa_t *spa, nvlist_t *config)
2255{
2256 uint_t extracted;
2257 uint64_t *glist;
2258 uint_t i, gcount;
2259 nvlist_t *nvl;
2260 vdev_t **vd;
2261 boolean_t attempt_reopen;
2262
2263 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2264 return;
2265
2266 /* check that the config is complete */
2267 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2268 &glist, &gcount) != 0)
2269 return;
2270
2271 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2272
2273 /* attempt to online all the vdevs & validate */
2274 attempt_reopen = B_TRUE;
2275 for (i = 0; i < gcount; i++) {
2276 if (glist[i] == 0) /* vdev is hole */
2277 continue;
2278
2279 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2280 if (vd[i] == NULL) {
2281 /*
2282 * Don't bother attempting to reopen the disks;
2283 * just do the split.
2284 */
2285 attempt_reopen = B_FALSE;
2286 } else {
2287 /* attempt to re-online it */
2288 vd[i]->vdev_offline = B_FALSE;
2289 }
2290 }
2291
2292 if (attempt_reopen) {
2293 vdev_reopen(spa->spa_root_vdev);
2294
2295 /* check each device to see what state it's in */
2296 for (extracted = 0, i = 0; i < gcount; i++) {
2297 if (vd[i] != NULL &&
2298 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2299 break;
2300 ++extracted;
2301 }
2302 }
2303
2304 /*
2305 * If every disk has been moved to the new pool, or if we never
2306 * even attempted to look at them, then we split them off for
2307 * good.
2308 */
2309 if (!attempt_reopen || gcount == extracted) {
2310 for (i = 0; i < gcount; i++)
2311 if (vd[i] != NULL)
2312 vdev_split(vd[i]);
2313 vdev_reopen(spa->spa_root_vdev);
2314 }
2315
2316 kmem_free(vd, gcount * sizeof (vdev_t *));
2317}
2318
2319static int
2320spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2321{
2322 char *ereport = FM_EREPORT_ZFS_POOL;
2323 int error;
2324
2325 spa->spa_load_state = state;
2326
2327 gethrestime(&spa->spa_loaded_ts);
2328 error = spa_load_impl(spa, type, &ereport);
2329
2330 /*
2331 * Don't count references from objsets that are already closed
2332 * and are making their way through the eviction process.
2333 */
2334 spa_evicting_os_wait(spa);
2335 spa->spa_minref = refcount_count(&spa->spa_refcount);
2336 if (error) {
2337 if (error != EEXIST) {
2338 spa->spa_loaded_ts.tv_sec = 0;
2339 spa->spa_loaded_ts.tv_nsec = 0;
2340 }
2341 if (error != EBADF) {
2342 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2343 }
2344 }
2345 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2346 spa->spa_ena = 0;
2347
2348 return (error);
2349}
2350
2351/*
2352 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2353 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2354 * spa's per-vdev ZAP list.
2355 */
2356static uint64_t
2357vdev_count_verify_zaps(vdev_t *vd)
2358{
2359 spa_t *spa = vd->vdev_spa;
2360 uint64_t total = 0;
2361 if (vd->vdev_top_zap != 0) {
2362 total++;
2363 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2364 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2365 }
2366 if (vd->vdev_leaf_zap != 0) {
2367 total++;
2368 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2369 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2370 }
2371
2372 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2373 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2374 }
2375
2376 return (total);
2377}
2378
2379static int
2380spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2381{
2382 uint64_t hostid;
2383 char *hostname;
2384 uint64_t myhostid = 0;
2385
2386 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2387 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2388 hostname = fnvlist_lookup_string(mos_config,
2389 ZPOOL_CONFIG_HOSTNAME);
2390
2391 myhostid = zone_get_hostid(NULL);
2392
2393 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2394 cmn_err(CE_WARN, "pool '%s' could not be "
2395 "loaded as it was last accessed by "
2396 "another system (host: %s hostid: 0x%llx). "
2397 "See: http://illumos.org/msg/ZFS-8000-EY",
2398 spa_name(spa), hostname, (u_longlong_t)hostid);
2399 spa_load_failed(spa, "hostid verification failed: pool "
2400 "last accessed by host: %s (hostid: 0x%llx)",
2401 hostname, (u_longlong_t)hostid);
2402 return (SET_ERROR(EBADF));
2403 }
2404 }
2405
2406 return (0);
2407}
2408
2409static int
2410spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2411{
2412 int error = 0;
2413 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2414 int parse;
2415 vdev_t *rvd;
2416 uint64_t pool_guid;
2417 char *comment;
2418
2419 /*
2420 * Versioning wasn't explicitly added to the label until later, so if
2421 * it's not present treat it as the initial version.
2422 */
2423 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2424 &spa->spa_ubsync.ub_version) != 0)
2425 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2426
2427 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2428 spa_load_failed(spa, "invalid config provided: '%s' missing",
2429 ZPOOL_CONFIG_POOL_GUID);
2430 return (SET_ERROR(EINVAL));
2431 }
2432
2433 /*
2434 * If we are doing an import, ensure that the pool is not already
2435 * imported by checking if its pool guid already exists in the
2436 * spa namespace.
2437 *
2438 * The only case that we allow an already imported pool to be
2439 * imported again, is when the pool is checkpointed and we want to
2440 * look at its checkpointed state from userland tools like zdb.
2441 */
2442#ifdef _KERNEL
2443 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2444 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2445 spa_guid_exists(pool_guid, 0)) {
2446#else
2447 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2448 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2449 spa_guid_exists(pool_guid, 0) &&
2450 !spa_importing_readonly_checkpoint(spa)) {
2451#endif
2452 spa_load_failed(spa, "a pool with guid %llu is already open",
2453 (u_longlong_t)pool_guid);
2454 return (SET_ERROR(EEXIST));
2455 }
2456
2457 spa->spa_config_guid = pool_guid;
2458
2459 nvlist_free(spa->spa_load_info);
2460 spa->spa_load_info = fnvlist_alloc();
2461
2462 ASSERT(spa->spa_comment == NULL);
2463 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2464 spa->spa_comment = spa_strdup(comment);
2465
2466 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2467 &spa->spa_config_txg);
2468
2469 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2470 spa->spa_config_splitting = fnvlist_dup(nvl);
2471
2472 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2473 spa_load_failed(spa, "invalid config provided: '%s' missing",
2474 ZPOOL_CONFIG_VDEV_TREE);
2475 return (SET_ERROR(EINVAL));
2476 }
2477
2478 /*
2479 * Create "The Godfather" zio to hold all async IOs
2480 */
2481 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2482 KM_SLEEP);
2483 for (int i = 0; i < max_ncpus; i++) {
2484 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2485 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2486 ZIO_FLAG_GODFATHER);
2487 }
2488
2489 /*
2490 * Parse the configuration into a vdev tree. We explicitly set the
2491 * value that will be returned by spa_version() since parsing the
2492 * configuration requires knowing the version number.
2493 */
2494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2495 parse = (type == SPA_IMPORT_EXISTING ?
2496 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2497 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2498 spa_config_exit(spa, SCL_ALL, FTAG);
2499
2500 if (error != 0) {
2501 spa_load_failed(spa, "unable to parse config [error=%d]",
2502 error);
2503 return (error);
2504 }
2505
2506 ASSERT(spa->spa_root_vdev == rvd);
2507 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2508 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2509
2510 if (type != SPA_IMPORT_ASSEMBLE) {
2511 ASSERT(spa_guid(spa) == pool_guid);
2512 }
2513
2514 return (0);
2515}
2516
2517/*
2518 * Recursively open all vdevs in the vdev tree. This function is called twice:
2519 * first with the untrusted config, then with the trusted config.
2520 */
2521static int
2522spa_ld_open_vdevs(spa_t *spa)
2523{
2524 int error = 0;
2525
2526 /*
2527 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2528 * missing/unopenable for the root vdev to be still considered openable.
2529 */
2530 if (spa->spa_trust_config) {
2531 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2532 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2533 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2534 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2535 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2536 } else {
2537 spa->spa_missing_tvds_allowed = 0;
2538 }
2539
2540 spa->spa_missing_tvds_allowed =
2541 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2542
2543 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2544 error = vdev_open(spa->spa_root_vdev);
2545 spa_config_exit(spa, SCL_ALL, FTAG);
2546
2547 if (spa->spa_missing_tvds != 0) {
2548 spa_load_note(spa, "vdev tree has %lld missing top-level "
2549 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2550 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2551 /*
2552 * Although theoretically we could allow users to open
2553 * incomplete pools in RW mode, we'd need to add a lot
2554 * of extra logic (e.g. adjust pool space to account
2555 * for missing vdevs).
2556 * This limitation also prevents users from accidentally
2557 * opening the pool in RW mode during data recovery and
2558 * damaging it further.
2559 */
2560 spa_load_note(spa, "pools with missing top-level "
2561 "vdevs can only be opened in read-only mode.");
2562 error = SET_ERROR(ENXIO);
2563 } else {
2564 spa_load_note(spa, "current settings allow for maximum "
2565 "%lld missing top-level vdevs at this stage.",
2566 (u_longlong_t)spa->spa_missing_tvds_allowed);
2567 }
2568 }
2569 if (error != 0) {
2570 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2571 error);
2572 }
2573 if (spa->spa_missing_tvds != 0 || error != 0)
2574 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2575
2576 return (error);
2577}
2578
2579/*
2580 * We need to validate the vdev labels against the configuration that
2581 * we have in hand. This function is called twice: first with an untrusted
2582 * config, then with a trusted config. The validation is more strict when the
2583 * config is trusted.
2584 */
2585static int
2586spa_ld_validate_vdevs(spa_t *spa)
2587{
2588 int error = 0;
2589 vdev_t *rvd = spa->spa_root_vdev;
2590
2591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2592 error = vdev_validate(rvd);
2593 spa_config_exit(spa, SCL_ALL, FTAG);
2594
2595 if (error != 0) {
2596 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2597 return (error);
2598 }
2599
2600 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2601 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2602 "some vdevs");
2603 vdev_dbgmsg_print_tree(rvd, 2);
2604 return (SET_ERROR(ENXIO));
2605 }
2606
2607 return (0);
2608}
2609
2610static void
2611spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2612{
2613 spa->spa_state = POOL_STATE_ACTIVE;
2614 spa->spa_ubsync = spa->spa_uberblock;
2615 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2616 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2617 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2618 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2619 spa->spa_claim_max_txg = spa->spa_first_txg;
2620 spa->spa_prev_software_version = ub->ub_software_version;
2621}
2622
2623static int
2624spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2625{
2626 vdev_t *rvd = spa->spa_root_vdev;
2627 nvlist_t *label;
2628 uberblock_t *ub = &spa->spa_uberblock;
2629
2630 /*
2631 * If we are opening the checkpointed state of the pool by
2632 * rewinding to it, at this point we will have written the
2633 * checkpointed uberblock to the vdev labels, so searching
2634 * the labels will find the right uberblock. However, if
2635 * we are opening the checkpointed state read-only, we have
2636 * not modified the labels. Therefore, we must ignore the
2637 * labels and continue using the spa_uberblock that was set
2638 * by spa_ld_checkpoint_rewind.
2639 *
2640 * Note that it would be fine to ignore the labels when
2641 * rewinding (opening writeable) as well. However, if we
2642 * crash just after writing the labels, we will end up
2643 * searching the labels. Doing so in the common case means
2644 * that this code path gets exercised normally, rather than
2645 * just in the edge case.
2646 */
2647 if (ub->ub_checkpoint_txg != 0 &&
2648 spa_importing_readonly_checkpoint(spa)) {
2649 spa_ld_select_uberblock_done(spa, ub);
2650 return (0);
2651 }
2652
2653 /*
2654 * Find the best uberblock.
2655 */
2656 vdev_uberblock_load(rvd, ub, &label);
2657
2658 /*
2659 * If we weren't able to find a single valid uberblock, return failure.
2660 */
2661 if (ub->ub_txg == 0) {
2662 nvlist_free(label);
2663 spa_load_failed(spa, "no valid uberblock found");
2664 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2665 }
2666
2667 spa_load_note(spa, "using uberblock with txg=%llu",
2668 (u_longlong_t)ub->ub_txg);
2669
2670 /*
2671 * If the pool has an unsupported version we can't open it.
2672 */
2673 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2674 nvlist_free(label);
2675 spa_load_failed(spa, "version %llu is not supported",
2676 (u_longlong_t)ub->ub_version);
2677 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2678 }
2679
2680 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2681 nvlist_t *features;
2682
2683 /*
2684 * If we weren't able to find what's necessary for reading the
2685 * MOS in the label, return failure.
2686 */
2687 if (label == NULL) {
2688 spa_load_failed(spa, "label config unavailable");
2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2690 ENXIO));
2691 }
2692
2693 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2694 &features) != 0) {
2695 nvlist_free(label);
2696 spa_load_failed(spa, "invalid label: '%s' missing",
2697 ZPOOL_CONFIG_FEATURES_FOR_READ);
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2699 ENXIO));
2700 }
2701
2702 /*
2703 * Update our in-core representation with the definitive values
2704 * from the label.
2705 */
2706 nvlist_free(spa->spa_label_features);
2707 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2708 }
2709
2710 nvlist_free(label);
2711
2712 /*
2713 * Look through entries in the label nvlist's features_for_read. If
2714 * there is a feature listed there which we don't understand then we
2715 * cannot open a pool.
2716 */
2717 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2718 nvlist_t *unsup_feat;
2719
2720 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2721 0);
2722
2723 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2724 NULL); nvp != NULL;
2725 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2726 if (!zfeature_is_supported(nvpair_name(nvp))) {
2727 VERIFY(nvlist_add_string(unsup_feat,
2728 nvpair_name(nvp), "") == 0);
2729 }
2730 }
2731
2732 if (!nvlist_empty(unsup_feat)) {
2733 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2734 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2735 nvlist_free(unsup_feat);
2736 spa_load_failed(spa, "some features are unsupported");
2737 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2738 ENOTSUP));
2739 }
2740
2741 nvlist_free(unsup_feat);
2742 }
2743
2744 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2745 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2746 spa_try_repair(spa, spa->spa_config);
2747 spa_config_exit(spa, SCL_ALL, FTAG);
2748 nvlist_free(spa->spa_config_splitting);
2749 spa->spa_config_splitting = NULL;
2750 }
2751
2752 /*
2753 * Initialize internal SPA structures.
2754 */
2755 spa_ld_select_uberblock_done(spa, ub);
2756
2757 return (0);
2758}
2759
2760static int
2761spa_ld_open_rootbp(spa_t *spa)
2762{
2763 int error = 0;
2764 vdev_t *rvd = spa->spa_root_vdev;
2765
2766 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2767 if (error != 0) {
2768 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2769 "[error=%d]", error);
2770 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2771 }
2772 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2773
2774 return (0);
2775}
2776
2777static int
2778spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2779 boolean_t reloading)
2780{
2781 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2782 nvlist_t *nv, *mos_config, *policy;
2783 int error = 0, copy_error;
2784 uint64_t healthy_tvds, healthy_tvds_mos;
2785 uint64_t mos_config_txg;
2786
2787 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2788 != 0)
2789 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2790
2791 /*
2792 * If we're assembling a pool from a split, the config provided is
2793 * already trusted so there is nothing to do.
2794 */
2795 if (type == SPA_IMPORT_ASSEMBLE)
2796 return (0);
2797
2798 healthy_tvds = spa_healthy_core_tvds(spa);
2799
2800 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2801 != 0) {
2802 spa_load_failed(spa, "unable to retrieve MOS config");
2803 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2804 }
2805
2806 /*
2807 * If we are doing an open, pool owner wasn't verified yet, thus do
2808 * the verification here.
2809 */
2810 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2811 error = spa_verify_host(spa, mos_config);
2812 if (error != 0) {
2813 nvlist_free(mos_config);
2814 return (error);
2815 }
2816 }
2817
2818 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2819
2820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2821
2822 /*
2823 * Build a new vdev tree from the trusted config
2824 */
2825 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2826
2827 /*
2828 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2829 * obtained by scanning /dev/dsk, then it will have the right vdev
2830 * paths. We update the trusted MOS config with this information.
2831 * We first try to copy the paths with vdev_copy_path_strict, which
2832 * succeeds only when both configs have exactly the same vdev tree.
2833 * If that fails, we fall back to a more flexible method that has a
2834 * best effort policy.
2835 */
2836 copy_error = vdev_copy_path_strict(rvd, mrvd);
2837 if (copy_error != 0 || spa_load_print_vdev_tree) {
2838 spa_load_note(spa, "provided vdev tree:");
2839 vdev_dbgmsg_print_tree(rvd, 2);
2840 spa_load_note(spa, "MOS vdev tree:");
2841 vdev_dbgmsg_print_tree(mrvd, 2);
2842 }
2843 if (copy_error != 0) {
2844 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2845 "back to vdev_copy_path_relaxed");
2846 vdev_copy_path_relaxed(rvd, mrvd);
2847 }
2848
2849 vdev_close(rvd);
2850 vdev_free(rvd);
2851 spa->spa_root_vdev = mrvd;
2852 rvd = mrvd;
2853 spa_config_exit(spa, SCL_ALL, FTAG);
2854
2855 /*
2856 * We will use spa_config if we decide to reload the spa or if spa_load
2857 * fails and we rewind. We must thus regenerate the config using the
| 2159 int64_t loss = 0;
2160
2161 verify_ok = B_TRUE;
2162 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2163 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2164
2165 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2166 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2167 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2168 VERIFY(nvlist_add_int64(spa->spa_load_info,
2169 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2170 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2171 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2172 } else {
2173 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2174 }
2175
2176 if (spa_load_verify_dryrun)
2177 return (0);
2178
2179 if (error) {
2180 if (error != ENXIO && error != EIO)
2181 error = SET_ERROR(EIO);
2182 return (error);
2183 }
2184
2185 return (verify_ok ? 0 : EIO);
2186}
2187
2188/*
2189 * Find a value in the pool props object.
2190 */
2191static void
2192spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2193{
2194 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2195 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2196}
2197
2198/*
2199 * Find a value in the pool directory object.
2200 */
2201static int
2202spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2203{
2204 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2205 name, sizeof (uint64_t), 1, val);
2206
2207 if (error != 0 && (error != ENOENT || log_enoent)) {
2208 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2209 "[error=%d]", name, error);
2210 }
2211
2212 return (error);
2213}
2214
2215static int
2216spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2217{
2218 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2219 return (SET_ERROR(err));
2220}
2221
2222static void
2223spa_spawn_aux_threads(spa_t *spa)
2224{
2225 ASSERT(spa_writeable(spa));
2226
2227 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2228
2229 spa_start_indirect_condensing_thread(spa);
2230
2231 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2232 spa->spa_checkpoint_discard_zthr =
2233 zthr_create(spa_checkpoint_discard_thread_check,
2234 spa_checkpoint_discard_thread, spa);
2235}
2236
2237/*
2238 * Fix up config after a partly-completed split. This is done with the
2239 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2240 * pool have that entry in their config, but only the splitting one contains
2241 * a list of all the guids of the vdevs that are being split off.
2242 *
2243 * This function determines what to do with that list: either rejoin
2244 * all the disks to the pool, or complete the splitting process. To attempt
2245 * the rejoin, each disk that is offlined is marked online again, and
2246 * we do a reopen() call. If the vdev label for every disk that was
2247 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2248 * then we call vdev_split() on each disk, and complete the split.
2249 *
2250 * Otherwise we leave the config alone, with all the vdevs in place in
2251 * the original pool.
2252 */
2253static void
2254spa_try_repair(spa_t *spa, nvlist_t *config)
2255{
2256 uint_t extracted;
2257 uint64_t *glist;
2258 uint_t i, gcount;
2259 nvlist_t *nvl;
2260 vdev_t **vd;
2261 boolean_t attempt_reopen;
2262
2263 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2264 return;
2265
2266 /* check that the config is complete */
2267 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2268 &glist, &gcount) != 0)
2269 return;
2270
2271 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2272
2273 /* attempt to online all the vdevs & validate */
2274 attempt_reopen = B_TRUE;
2275 for (i = 0; i < gcount; i++) {
2276 if (glist[i] == 0) /* vdev is hole */
2277 continue;
2278
2279 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2280 if (vd[i] == NULL) {
2281 /*
2282 * Don't bother attempting to reopen the disks;
2283 * just do the split.
2284 */
2285 attempt_reopen = B_FALSE;
2286 } else {
2287 /* attempt to re-online it */
2288 vd[i]->vdev_offline = B_FALSE;
2289 }
2290 }
2291
2292 if (attempt_reopen) {
2293 vdev_reopen(spa->spa_root_vdev);
2294
2295 /* check each device to see what state it's in */
2296 for (extracted = 0, i = 0; i < gcount; i++) {
2297 if (vd[i] != NULL &&
2298 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2299 break;
2300 ++extracted;
2301 }
2302 }
2303
2304 /*
2305 * If every disk has been moved to the new pool, or if we never
2306 * even attempted to look at them, then we split them off for
2307 * good.
2308 */
2309 if (!attempt_reopen || gcount == extracted) {
2310 for (i = 0; i < gcount; i++)
2311 if (vd[i] != NULL)
2312 vdev_split(vd[i]);
2313 vdev_reopen(spa->spa_root_vdev);
2314 }
2315
2316 kmem_free(vd, gcount * sizeof (vdev_t *));
2317}
2318
2319static int
2320spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2321{
2322 char *ereport = FM_EREPORT_ZFS_POOL;
2323 int error;
2324
2325 spa->spa_load_state = state;
2326
2327 gethrestime(&spa->spa_loaded_ts);
2328 error = spa_load_impl(spa, type, &ereport);
2329
2330 /*
2331 * Don't count references from objsets that are already closed
2332 * and are making their way through the eviction process.
2333 */
2334 spa_evicting_os_wait(spa);
2335 spa->spa_minref = refcount_count(&spa->spa_refcount);
2336 if (error) {
2337 if (error != EEXIST) {
2338 spa->spa_loaded_ts.tv_sec = 0;
2339 spa->spa_loaded_ts.tv_nsec = 0;
2340 }
2341 if (error != EBADF) {
2342 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2343 }
2344 }
2345 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2346 spa->spa_ena = 0;
2347
2348 return (error);
2349}
2350
2351/*
2352 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2353 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2354 * spa's per-vdev ZAP list.
2355 */
2356static uint64_t
2357vdev_count_verify_zaps(vdev_t *vd)
2358{
2359 spa_t *spa = vd->vdev_spa;
2360 uint64_t total = 0;
2361 if (vd->vdev_top_zap != 0) {
2362 total++;
2363 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2364 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2365 }
2366 if (vd->vdev_leaf_zap != 0) {
2367 total++;
2368 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2369 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2370 }
2371
2372 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2373 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2374 }
2375
2376 return (total);
2377}
2378
2379static int
2380spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2381{
2382 uint64_t hostid;
2383 char *hostname;
2384 uint64_t myhostid = 0;
2385
2386 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2387 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2388 hostname = fnvlist_lookup_string(mos_config,
2389 ZPOOL_CONFIG_HOSTNAME);
2390
2391 myhostid = zone_get_hostid(NULL);
2392
2393 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2394 cmn_err(CE_WARN, "pool '%s' could not be "
2395 "loaded as it was last accessed by "
2396 "another system (host: %s hostid: 0x%llx). "
2397 "See: http://illumos.org/msg/ZFS-8000-EY",
2398 spa_name(spa), hostname, (u_longlong_t)hostid);
2399 spa_load_failed(spa, "hostid verification failed: pool "
2400 "last accessed by host: %s (hostid: 0x%llx)",
2401 hostname, (u_longlong_t)hostid);
2402 return (SET_ERROR(EBADF));
2403 }
2404 }
2405
2406 return (0);
2407}
2408
2409static int
2410spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2411{
2412 int error = 0;
2413 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2414 int parse;
2415 vdev_t *rvd;
2416 uint64_t pool_guid;
2417 char *comment;
2418
2419 /*
2420 * Versioning wasn't explicitly added to the label until later, so if
2421 * it's not present treat it as the initial version.
2422 */
2423 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2424 &spa->spa_ubsync.ub_version) != 0)
2425 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2426
2427 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2428 spa_load_failed(spa, "invalid config provided: '%s' missing",
2429 ZPOOL_CONFIG_POOL_GUID);
2430 return (SET_ERROR(EINVAL));
2431 }
2432
2433 /*
2434 * If we are doing an import, ensure that the pool is not already
2435 * imported by checking if its pool guid already exists in the
2436 * spa namespace.
2437 *
2438 * The only case that we allow an already imported pool to be
2439 * imported again, is when the pool is checkpointed and we want to
2440 * look at its checkpointed state from userland tools like zdb.
2441 */
2442#ifdef _KERNEL
2443 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2444 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2445 spa_guid_exists(pool_guid, 0)) {
2446#else
2447 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2448 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2449 spa_guid_exists(pool_guid, 0) &&
2450 !spa_importing_readonly_checkpoint(spa)) {
2451#endif
2452 spa_load_failed(spa, "a pool with guid %llu is already open",
2453 (u_longlong_t)pool_guid);
2454 return (SET_ERROR(EEXIST));
2455 }
2456
2457 spa->spa_config_guid = pool_guid;
2458
2459 nvlist_free(spa->spa_load_info);
2460 spa->spa_load_info = fnvlist_alloc();
2461
2462 ASSERT(spa->spa_comment == NULL);
2463 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2464 spa->spa_comment = spa_strdup(comment);
2465
2466 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2467 &spa->spa_config_txg);
2468
2469 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2470 spa->spa_config_splitting = fnvlist_dup(nvl);
2471
2472 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2473 spa_load_failed(spa, "invalid config provided: '%s' missing",
2474 ZPOOL_CONFIG_VDEV_TREE);
2475 return (SET_ERROR(EINVAL));
2476 }
2477
2478 /*
2479 * Create "The Godfather" zio to hold all async IOs
2480 */
2481 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2482 KM_SLEEP);
2483 for (int i = 0; i < max_ncpus; i++) {
2484 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2485 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2486 ZIO_FLAG_GODFATHER);
2487 }
2488
2489 /*
2490 * Parse the configuration into a vdev tree. We explicitly set the
2491 * value that will be returned by spa_version() since parsing the
2492 * configuration requires knowing the version number.
2493 */
2494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2495 parse = (type == SPA_IMPORT_EXISTING ?
2496 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2497 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2498 spa_config_exit(spa, SCL_ALL, FTAG);
2499
2500 if (error != 0) {
2501 spa_load_failed(spa, "unable to parse config [error=%d]",
2502 error);
2503 return (error);
2504 }
2505
2506 ASSERT(spa->spa_root_vdev == rvd);
2507 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2508 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2509
2510 if (type != SPA_IMPORT_ASSEMBLE) {
2511 ASSERT(spa_guid(spa) == pool_guid);
2512 }
2513
2514 return (0);
2515}
2516
2517/*
2518 * Recursively open all vdevs in the vdev tree. This function is called twice:
2519 * first with the untrusted config, then with the trusted config.
2520 */
2521static int
2522spa_ld_open_vdevs(spa_t *spa)
2523{
2524 int error = 0;
2525
2526 /*
2527 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2528 * missing/unopenable for the root vdev to be still considered openable.
2529 */
2530 if (spa->spa_trust_config) {
2531 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2532 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2533 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2534 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2535 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2536 } else {
2537 spa->spa_missing_tvds_allowed = 0;
2538 }
2539
2540 spa->spa_missing_tvds_allowed =
2541 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2542
2543 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2544 error = vdev_open(spa->spa_root_vdev);
2545 spa_config_exit(spa, SCL_ALL, FTAG);
2546
2547 if (spa->spa_missing_tvds != 0) {
2548 spa_load_note(spa, "vdev tree has %lld missing top-level "
2549 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2550 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2551 /*
2552 * Although theoretically we could allow users to open
2553 * incomplete pools in RW mode, we'd need to add a lot
2554 * of extra logic (e.g. adjust pool space to account
2555 * for missing vdevs).
2556 * This limitation also prevents users from accidentally
2557 * opening the pool in RW mode during data recovery and
2558 * damaging it further.
2559 */
2560 spa_load_note(spa, "pools with missing top-level "
2561 "vdevs can only be opened in read-only mode.");
2562 error = SET_ERROR(ENXIO);
2563 } else {
2564 spa_load_note(spa, "current settings allow for maximum "
2565 "%lld missing top-level vdevs at this stage.",
2566 (u_longlong_t)spa->spa_missing_tvds_allowed);
2567 }
2568 }
2569 if (error != 0) {
2570 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2571 error);
2572 }
2573 if (spa->spa_missing_tvds != 0 || error != 0)
2574 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2575
2576 return (error);
2577}
2578
2579/*
2580 * We need to validate the vdev labels against the configuration that
2581 * we have in hand. This function is called twice: first with an untrusted
2582 * config, then with a trusted config. The validation is more strict when the
2583 * config is trusted.
2584 */
2585static int
2586spa_ld_validate_vdevs(spa_t *spa)
2587{
2588 int error = 0;
2589 vdev_t *rvd = spa->spa_root_vdev;
2590
2591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2592 error = vdev_validate(rvd);
2593 spa_config_exit(spa, SCL_ALL, FTAG);
2594
2595 if (error != 0) {
2596 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2597 return (error);
2598 }
2599
2600 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2601 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2602 "some vdevs");
2603 vdev_dbgmsg_print_tree(rvd, 2);
2604 return (SET_ERROR(ENXIO));
2605 }
2606
2607 return (0);
2608}
2609
2610static void
2611spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2612{
2613 spa->spa_state = POOL_STATE_ACTIVE;
2614 spa->spa_ubsync = spa->spa_uberblock;
2615 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2616 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2617 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2618 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2619 spa->spa_claim_max_txg = spa->spa_first_txg;
2620 spa->spa_prev_software_version = ub->ub_software_version;
2621}
2622
2623static int
2624spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2625{
2626 vdev_t *rvd = spa->spa_root_vdev;
2627 nvlist_t *label;
2628 uberblock_t *ub = &spa->spa_uberblock;
2629
2630 /*
2631 * If we are opening the checkpointed state of the pool by
2632 * rewinding to it, at this point we will have written the
2633 * checkpointed uberblock to the vdev labels, so searching
2634 * the labels will find the right uberblock. However, if
2635 * we are opening the checkpointed state read-only, we have
2636 * not modified the labels. Therefore, we must ignore the
2637 * labels and continue using the spa_uberblock that was set
2638 * by spa_ld_checkpoint_rewind.
2639 *
2640 * Note that it would be fine to ignore the labels when
2641 * rewinding (opening writeable) as well. However, if we
2642 * crash just after writing the labels, we will end up
2643 * searching the labels. Doing so in the common case means
2644 * that this code path gets exercised normally, rather than
2645 * just in the edge case.
2646 */
2647 if (ub->ub_checkpoint_txg != 0 &&
2648 spa_importing_readonly_checkpoint(spa)) {
2649 spa_ld_select_uberblock_done(spa, ub);
2650 return (0);
2651 }
2652
2653 /*
2654 * Find the best uberblock.
2655 */
2656 vdev_uberblock_load(rvd, ub, &label);
2657
2658 /*
2659 * If we weren't able to find a single valid uberblock, return failure.
2660 */
2661 if (ub->ub_txg == 0) {
2662 nvlist_free(label);
2663 spa_load_failed(spa, "no valid uberblock found");
2664 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2665 }
2666
2667 spa_load_note(spa, "using uberblock with txg=%llu",
2668 (u_longlong_t)ub->ub_txg);
2669
2670 /*
2671 * If the pool has an unsupported version we can't open it.
2672 */
2673 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2674 nvlist_free(label);
2675 spa_load_failed(spa, "version %llu is not supported",
2676 (u_longlong_t)ub->ub_version);
2677 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2678 }
2679
2680 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2681 nvlist_t *features;
2682
2683 /*
2684 * If we weren't able to find what's necessary for reading the
2685 * MOS in the label, return failure.
2686 */
2687 if (label == NULL) {
2688 spa_load_failed(spa, "label config unavailable");
2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2690 ENXIO));
2691 }
2692
2693 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2694 &features) != 0) {
2695 nvlist_free(label);
2696 spa_load_failed(spa, "invalid label: '%s' missing",
2697 ZPOOL_CONFIG_FEATURES_FOR_READ);
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2699 ENXIO));
2700 }
2701
2702 /*
2703 * Update our in-core representation with the definitive values
2704 * from the label.
2705 */
2706 nvlist_free(spa->spa_label_features);
2707 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2708 }
2709
2710 nvlist_free(label);
2711
2712 /*
2713 * Look through entries in the label nvlist's features_for_read. If
2714 * there is a feature listed there which we don't understand then we
2715 * cannot open a pool.
2716 */
2717 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2718 nvlist_t *unsup_feat;
2719
2720 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2721 0);
2722
2723 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2724 NULL); nvp != NULL;
2725 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2726 if (!zfeature_is_supported(nvpair_name(nvp))) {
2727 VERIFY(nvlist_add_string(unsup_feat,
2728 nvpair_name(nvp), "") == 0);
2729 }
2730 }
2731
2732 if (!nvlist_empty(unsup_feat)) {
2733 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2734 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2735 nvlist_free(unsup_feat);
2736 spa_load_failed(spa, "some features are unsupported");
2737 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2738 ENOTSUP));
2739 }
2740
2741 nvlist_free(unsup_feat);
2742 }
2743
2744 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2745 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2746 spa_try_repair(spa, spa->spa_config);
2747 spa_config_exit(spa, SCL_ALL, FTAG);
2748 nvlist_free(spa->spa_config_splitting);
2749 spa->spa_config_splitting = NULL;
2750 }
2751
2752 /*
2753 * Initialize internal SPA structures.
2754 */
2755 spa_ld_select_uberblock_done(spa, ub);
2756
2757 return (0);
2758}
2759
2760static int
2761spa_ld_open_rootbp(spa_t *spa)
2762{
2763 int error = 0;
2764 vdev_t *rvd = spa->spa_root_vdev;
2765
2766 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2767 if (error != 0) {
2768 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2769 "[error=%d]", error);
2770 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2771 }
2772 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2773
2774 return (0);
2775}
2776
2777static int
2778spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2779 boolean_t reloading)
2780{
2781 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2782 nvlist_t *nv, *mos_config, *policy;
2783 int error = 0, copy_error;
2784 uint64_t healthy_tvds, healthy_tvds_mos;
2785 uint64_t mos_config_txg;
2786
2787 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2788 != 0)
2789 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2790
2791 /*
2792 * If we're assembling a pool from a split, the config provided is
2793 * already trusted so there is nothing to do.
2794 */
2795 if (type == SPA_IMPORT_ASSEMBLE)
2796 return (0);
2797
2798 healthy_tvds = spa_healthy_core_tvds(spa);
2799
2800 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2801 != 0) {
2802 spa_load_failed(spa, "unable to retrieve MOS config");
2803 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2804 }
2805
2806 /*
2807 * If we are doing an open, pool owner wasn't verified yet, thus do
2808 * the verification here.
2809 */
2810 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2811 error = spa_verify_host(spa, mos_config);
2812 if (error != 0) {
2813 nvlist_free(mos_config);
2814 return (error);
2815 }
2816 }
2817
2818 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2819
2820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2821
2822 /*
2823 * Build a new vdev tree from the trusted config
2824 */
2825 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2826
2827 /*
2828 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2829 * obtained by scanning /dev/dsk, then it will have the right vdev
2830 * paths. We update the trusted MOS config with this information.
2831 * We first try to copy the paths with vdev_copy_path_strict, which
2832 * succeeds only when both configs have exactly the same vdev tree.
2833 * If that fails, we fall back to a more flexible method that has a
2834 * best effort policy.
2835 */
2836 copy_error = vdev_copy_path_strict(rvd, mrvd);
2837 if (copy_error != 0 || spa_load_print_vdev_tree) {
2838 spa_load_note(spa, "provided vdev tree:");
2839 vdev_dbgmsg_print_tree(rvd, 2);
2840 spa_load_note(spa, "MOS vdev tree:");
2841 vdev_dbgmsg_print_tree(mrvd, 2);
2842 }
2843 if (copy_error != 0) {
2844 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2845 "back to vdev_copy_path_relaxed");
2846 vdev_copy_path_relaxed(rvd, mrvd);
2847 }
2848
2849 vdev_close(rvd);
2850 vdev_free(rvd);
2851 spa->spa_root_vdev = mrvd;
2852 rvd = mrvd;
2853 spa_config_exit(spa, SCL_ALL, FTAG);
2854
2855 /*
2856 * We will use spa_config if we decide to reload the spa or if spa_load
2857 * fails and we rewind. We must thus regenerate the config using the
|
2858 * MOS information with the updated paths. Rewind policy is an import
2859 * setting and is not in the MOS. We copy it over to our new, trusted
2860 * config.
| 2858 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
2859 * pass settings on how to load the pool and is not stored in the MOS.
2860 * We copy it over to our new, trusted config.
|
2861 */
2862 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2863 ZPOOL_CONFIG_POOL_TXG);
2864 nvlist_free(mos_config);
2865 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
| 2861 */
2862 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2863 ZPOOL_CONFIG_POOL_TXG);
2864 nvlist_free(mos_config);
2865 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
|
2866 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_REWIND_POLICY,
| 2866 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
|
2867 &policy) == 0)
| 2867 &policy) == 0)
|
2868 fnvlist_add_nvlist(mos_config, ZPOOL_REWIND_POLICY, policy);
| 2868 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
|
2869 spa_config_set(spa, mos_config);
2870 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2871
2872 /*
2873 * Now that we got the config from the MOS, we should be more strict
2874 * in checking blkptrs and can make assumptions about the consistency
2875 * of the vdev tree. spa_trust_config must be set to true before opening
2876 * vdevs in order for them to be writeable.
2877 */
2878 spa->spa_trust_config = B_TRUE;
2879
2880 /*
2881 * Open and validate the new vdev tree
2882 */
2883 error = spa_ld_open_vdevs(spa);
2884 if (error != 0)
2885 return (error);
2886
2887 error = spa_ld_validate_vdevs(spa);
2888 if (error != 0)
2889 return (error);
2890
2891 if (copy_error != 0 || spa_load_print_vdev_tree) {
2892 spa_load_note(spa, "final vdev tree:");
2893 vdev_dbgmsg_print_tree(rvd, 2);
2894 }
2895
2896 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2897 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2898 /*
2899 * Sanity check to make sure that we are indeed loading the
2900 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2901 * in the config provided and they happened to be the only ones
2902 * to have the latest uberblock, we could involuntarily perform
2903 * an extreme rewind.
2904 */
2905 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2906 if (healthy_tvds_mos - healthy_tvds >=
2907 SPA_SYNC_MIN_VDEVS) {
2908 spa_load_note(spa, "config provided misses too many "
2909 "top-level vdevs compared to MOS (%lld vs %lld). ",
2910 (u_longlong_t)healthy_tvds,
2911 (u_longlong_t)healthy_tvds_mos);
2912 spa_load_note(spa, "vdev tree:");
2913 vdev_dbgmsg_print_tree(rvd, 2);
2914 if (reloading) {
2915 spa_load_failed(spa, "config was already "
2916 "provided from MOS. Aborting.");
2917 return (spa_vdev_err(rvd,
2918 VDEV_AUX_CORRUPT_DATA, EIO));
2919 }
2920 spa_load_note(spa, "spa must be reloaded using MOS "
2921 "config");
2922 return (SET_ERROR(EAGAIN));
2923 }
2924 }
2925
2926 error = spa_check_for_missing_logs(spa);
2927 if (error != 0)
2928 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2929
2930 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2931 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2932 "guid sum (%llu != %llu)",
2933 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2934 (u_longlong_t)rvd->vdev_guid_sum);
2935 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2936 ENXIO));
2937 }
2938
2939 return (0);
2940}
2941
2942static int
2943spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2944{
2945 int error = 0;
2946 vdev_t *rvd = spa->spa_root_vdev;
2947
2948 /*
2949 * Everything that we read before spa_remove_init() must be stored
2950 * on concreted vdevs. Therefore we do this as early as possible.
2951 */
2952 error = spa_remove_init(spa);
2953 if (error != 0) {
2954 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2955 error);
2956 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2957 }
2958
2959 /*
2960 * Retrieve information needed to condense indirect vdev mappings.
2961 */
2962 error = spa_condense_init(spa);
2963 if (error != 0) {
2964 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2965 error);
2966 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2967 }
2968
2969 return (0);
2970}
2971
2972static int
2973spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2974{
2975 int error = 0;
2976 vdev_t *rvd = spa->spa_root_vdev;
2977
2978 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2979 boolean_t missing_feat_read = B_FALSE;
2980 nvlist_t *unsup_feat, *enabled_feat;
2981
2982 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2983 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2985 }
2986
2987 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2988 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
2989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2990 }
2991
2992 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2993 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
2994 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2995 }
2996
2997 enabled_feat = fnvlist_alloc();
2998 unsup_feat = fnvlist_alloc();
2999
3000 if (!spa_features_check(spa, B_FALSE,
3001 unsup_feat, enabled_feat))
3002 missing_feat_read = B_TRUE;
3003
3004 if (spa_writeable(spa) ||
3005 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3006 if (!spa_features_check(spa, B_TRUE,
3007 unsup_feat, enabled_feat)) {
3008 *missing_feat_writep = B_TRUE;
3009 }
3010 }
3011
3012 fnvlist_add_nvlist(spa->spa_load_info,
3013 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3014
3015 if (!nvlist_empty(unsup_feat)) {
3016 fnvlist_add_nvlist(spa->spa_load_info,
3017 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3018 }
3019
3020 fnvlist_free(enabled_feat);
3021 fnvlist_free(unsup_feat);
3022
3023 if (!missing_feat_read) {
3024 fnvlist_add_boolean(spa->spa_load_info,
3025 ZPOOL_CONFIG_CAN_RDONLY);
3026 }
3027
3028 /*
3029 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3030 * twofold: to determine whether the pool is available for
3031 * import in read-write mode and (if it is not) whether the
3032 * pool is available for import in read-only mode. If the pool
3033 * is available for import in read-write mode, it is displayed
3034 * as available in userland; if it is not available for import
3035 * in read-only mode, it is displayed as unavailable in
3036 * userland. If the pool is available for import in read-only
3037 * mode but not read-write mode, it is displayed as unavailable
3038 * in userland with a special note that the pool is actually
3039 * available for open in read-only mode.
3040 *
3041 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3042 * missing a feature for write, we must first determine whether
3043 * the pool can be opened read-only before returning to
3044 * userland in order to know whether to display the
3045 * abovementioned note.
3046 */
3047 if (missing_feat_read || (*missing_feat_writep &&
3048 spa_writeable(spa))) {
3049 spa_load_failed(spa, "pool uses unsupported features");
3050 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3051 ENOTSUP));
3052 }
3053
3054 /*
3055 * Load refcounts for ZFS features from disk into an in-memory
3056 * cache during SPA initialization.
3057 */
3058 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3059 uint64_t refcount;
3060
3061 error = feature_get_refcount_from_disk(spa,
3062 &spa_feature_table[i], &refcount);
3063 if (error == 0) {
3064 spa->spa_feat_refcount_cache[i] = refcount;
3065 } else if (error == ENOTSUP) {
3066 spa->spa_feat_refcount_cache[i] =
3067 SPA_FEATURE_DISABLED;
3068 } else {
3069 spa_load_failed(spa, "error getting refcount "
3070 "for feature %s [error=%d]",
3071 spa_feature_table[i].fi_guid, error);
3072 return (spa_vdev_err(rvd,
3073 VDEV_AUX_CORRUPT_DATA, EIO));
3074 }
3075 }
3076 }
3077
3078 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3079 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3080 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3082 }
3083
3084 return (0);
3085}
3086
3087static int
3088spa_ld_load_special_directories(spa_t *spa)
3089{
3090 int error = 0;
3091 vdev_t *rvd = spa->spa_root_vdev;
3092
3093 spa->spa_is_initializing = B_TRUE;
3094 error = dsl_pool_open(spa->spa_dsl_pool);
3095 spa->spa_is_initializing = B_FALSE;
3096 if (error != 0) {
3097 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3099 }
3100
3101 return (0);
3102}
3103
3104static int
3105spa_ld_get_props(spa_t *spa)
3106{
3107 int error = 0;
3108 uint64_t obj;
3109 vdev_t *rvd = spa->spa_root_vdev;
3110
3111 /* Grab the secret checksum salt from the MOS. */
3112 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3113 DMU_POOL_CHECKSUM_SALT, 1,
3114 sizeof (spa->spa_cksum_salt.zcs_bytes),
3115 spa->spa_cksum_salt.zcs_bytes);
3116 if (error == ENOENT) {
3117 /* Generate a new salt for subsequent use */
3118 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3119 sizeof (spa->spa_cksum_salt.zcs_bytes));
3120 } else if (error != 0) {
3121 spa_load_failed(spa, "unable to retrieve checksum salt from "
3122 "MOS [error=%d]", error);
3123 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3124 }
3125
3126 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3127 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3128 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3129 if (error != 0) {
3130 spa_load_failed(spa, "error opening deferred-frees bpobj "
3131 "[error=%d]", error);
3132 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3133 }
3134
3135 /*
3136 * Load the bit that tells us to use the new accounting function
3137 * (raid-z deflation). If we have an older pool, this will not
3138 * be present.
3139 */
3140 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3141 if (error != 0 && error != ENOENT)
3142 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3143
3144 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3145 &spa->spa_creation_version, B_FALSE);
3146 if (error != 0 && error != ENOENT)
3147 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3148
3149 /*
3150 * Load the persistent error log. If we have an older pool, this will
3151 * not be present.
3152 */
3153 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3154 B_FALSE);
3155 if (error != 0 && error != ENOENT)
3156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3157
3158 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3159 &spa->spa_errlog_scrub, B_FALSE);
3160 if (error != 0 && error != ENOENT)
3161 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3162
3163 /*
3164 * Load the history object. If we have an older pool, this
3165 * will not be present.
3166 */
3167 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3168 if (error != 0 && error != ENOENT)
3169 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3170
3171 /*
3172 * Load the per-vdev ZAP map. If we have an older pool, this will not
3173 * be present; in this case, defer its creation to a later time to
3174 * avoid dirtying the MOS this early / out of sync context. See
3175 * spa_sync_config_object.
3176 */
3177
3178 /* The sentinel is only available in the MOS config. */
3179 nvlist_t *mos_config;
3180 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3181 spa_load_failed(spa, "unable to retrieve MOS config");
3182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3183 }
3184
3185 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3186 &spa->spa_all_vdev_zaps, B_FALSE);
3187
3188 if (error == ENOENT) {
3189 VERIFY(!nvlist_exists(mos_config,
3190 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3191 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3192 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3193 } else if (error != 0) {
3194 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3195 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3196 /*
3197 * An older version of ZFS overwrote the sentinel value, so
3198 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3199 * destruction to later; see spa_sync_config_object.
3200 */
3201 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3202 /*
3203 * We're assuming that no vdevs have had their ZAPs created
3204 * before this. Better be sure of it.
3205 */
3206 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3207 }
3208 nvlist_free(mos_config);
3209
3210 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3211
3212 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3213 B_FALSE);
3214 if (error && error != ENOENT)
3215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3216
3217 if (error == 0) {
3218 uint64_t autoreplace;
3219
3220 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3221 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3222 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3223 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3224 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3225 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3226 &spa->spa_dedup_ditto);
3227
3228 spa->spa_autoreplace = (autoreplace != 0);
3229 }
3230
3231 /*
3232 * If we are importing a pool with missing top-level vdevs,
3233 * we enforce that the pool doesn't panic or get suspended on
3234 * error since the likelihood of missing data is extremely high.
3235 */
3236 if (spa->spa_missing_tvds > 0 &&
3237 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3238 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3239 spa_load_note(spa, "forcing failmode to 'continue' "
3240 "as some top level vdevs are missing");
3241 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3242 }
3243
3244 return (0);
3245}
3246
3247static int
3248spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3249{
3250 int error = 0;
3251 vdev_t *rvd = spa->spa_root_vdev;
3252
3253 /*
3254 * If we're assembling the pool from the split-off vdevs of
3255 * an existing pool, we don't want to attach the spares & cache
3256 * devices.
3257 */
3258
3259 /*
3260 * Load any hot spares for this pool.
3261 */
3262 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3263 B_FALSE);
3264 if (error != 0 && error != ENOENT)
3265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3266 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3267 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3268 if (load_nvlist(spa, spa->spa_spares.sav_object,
3269 &spa->spa_spares.sav_config) != 0) {
3270 spa_load_failed(spa, "error loading spares nvlist");
3271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3272 }
3273
3274 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3275 spa_load_spares(spa);
3276 spa_config_exit(spa, SCL_ALL, FTAG);
3277 } else if (error == 0) {
3278 spa->spa_spares.sav_sync = B_TRUE;
3279 }
3280
3281 /*
3282 * Load any level 2 ARC devices for this pool.
3283 */
3284 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3285 &spa->spa_l2cache.sav_object, B_FALSE);
3286 if (error != 0 && error != ENOENT)
3287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3288 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3289 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3290 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3291 &spa->spa_l2cache.sav_config) != 0) {
3292 spa_load_failed(spa, "error loading l2cache nvlist");
3293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3294 }
3295
3296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3297 spa_load_l2cache(spa);
3298 spa_config_exit(spa, SCL_ALL, FTAG);
3299 } else if (error == 0) {
3300 spa->spa_l2cache.sav_sync = B_TRUE;
3301 }
3302
3303 return (0);
3304}
3305
3306static int
3307spa_ld_load_vdev_metadata(spa_t *spa)
3308{
3309 int error = 0;
3310 vdev_t *rvd = spa->spa_root_vdev;
3311
3312 /*
3313 * If the 'autoreplace' property is set, then post a resource notifying
3314 * the ZFS DE that it should not issue any faults for unopenable
3315 * devices. We also iterate over the vdevs, and post a sysevent for any
3316 * unopenable vdevs so that the normal autoreplace handler can take
3317 * over.
3318 */
3319 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3320 spa_check_removed(spa->spa_root_vdev);
3321 /*
3322 * For the import case, this is done in spa_import(), because
3323 * at this point we're using the spare definitions from
3324 * the MOS config, not necessarily from the userland config.
3325 */
3326 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3327 spa_aux_check_removed(&spa->spa_spares);
3328 spa_aux_check_removed(&spa->spa_l2cache);
3329 }
3330 }
3331
3332 /*
3333 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3334 */
3335 error = vdev_load(rvd);
3336 if (error != 0) {
3337 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3338 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3339 }
3340
3341 /*
3342 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3343 */
3344 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3345 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3346 spa_config_exit(spa, SCL_ALL, FTAG);
3347
3348 return (0);
3349}
3350
3351static int
3352spa_ld_load_dedup_tables(spa_t *spa)
3353{
3354 int error = 0;
3355 vdev_t *rvd = spa->spa_root_vdev;
3356
3357 error = ddt_load(spa);
3358 if (error != 0) {
3359 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3361 }
3362
3363 return (0);
3364}
3365
3366static int
3367spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3368{
3369 vdev_t *rvd = spa->spa_root_vdev;
3370
3371 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3372 boolean_t missing = spa_check_logs(spa);
3373 if (missing) {
3374 if (spa->spa_missing_tvds != 0) {
3375 spa_load_note(spa, "spa_check_logs failed "
3376 "so dropping the logs");
3377 } else {
3378 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3379 spa_load_failed(spa, "spa_check_logs failed");
3380 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3381 ENXIO));
3382 }
3383 }
3384 }
3385
3386 return (0);
3387}
3388
3389static int
3390spa_ld_verify_pool_data(spa_t *spa)
3391{
3392 int error = 0;
3393 vdev_t *rvd = spa->spa_root_vdev;
3394
3395 /*
3396 * We've successfully opened the pool, verify that we're ready
3397 * to start pushing transactions.
3398 */
3399 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3400 error = spa_load_verify(spa);
3401 if (error != 0) {
3402 spa_load_failed(spa, "spa_load_verify failed "
3403 "[error=%d]", error);
3404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3405 error));
3406 }
3407 }
3408
3409 return (0);
3410}
3411
3412static void
3413spa_ld_claim_log_blocks(spa_t *spa)
3414{
3415 dmu_tx_t *tx;
3416 dsl_pool_t *dp = spa_get_dsl(spa);
3417
3418 /*
3419 * Claim log blocks that haven't been committed yet.
3420 * This must all happen in a single txg.
3421 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3422 * invoked from zil_claim_log_block()'s i/o done callback.
3423 * Price of rollback is that we abandon the log.
3424 */
3425 spa->spa_claiming = B_TRUE;
3426
3427 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3428 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3429 zil_claim, tx, DS_FIND_CHILDREN);
3430 dmu_tx_commit(tx);
3431
3432 spa->spa_claiming = B_FALSE;
3433
3434 spa_set_log_state(spa, SPA_LOG_GOOD);
3435}
3436
3437static void
3438spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3439 boolean_t update_config_cache)
3440{
3441 vdev_t *rvd = spa->spa_root_vdev;
3442 int need_update = B_FALSE;
3443
3444 /*
3445 * If the config cache is stale, or we have uninitialized
3446 * metaslabs (see spa_vdev_add()), then update the config.
3447 *
3448 * If this is a verbatim import, trust the current
3449 * in-core spa_config and update the disk labels.
3450 */
3451 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3452 spa->spa_load_state == SPA_LOAD_IMPORT ||
3453 spa->spa_load_state == SPA_LOAD_RECOVER ||
3454 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3455 need_update = B_TRUE;
3456
3457 for (int c = 0; c < rvd->vdev_children; c++)
3458 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3459 need_update = B_TRUE;
3460
3461 /*
3462 * Update the config cache asychronously in case we're the
3463 * root pool, in which case the config cache isn't writable yet.
3464 */
3465 if (need_update)
3466 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3467}
3468
3469static void
3470spa_ld_prepare_for_reload(spa_t *spa)
3471{
3472 int mode = spa->spa_mode;
3473 int async_suspended = spa->spa_async_suspended;
3474
3475 spa_unload(spa);
3476 spa_deactivate(spa);
3477 spa_activate(spa, mode);
3478
3479 /*
3480 * We save the value of spa_async_suspended as it gets reset to 0 by
3481 * spa_unload(). We want to restore it back to the original value before
3482 * returning as we might be calling spa_async_resume() later.
3483 */
3484 spa->spa_async_suspended = async_suspended;
3485}
3486
3487static int
3488spa_ld_read_checkpoint_txg(spa_t *spa)
3489{
3490 uberblock_t checkpoint;
3491 int error = 0;
3492
3493 ASSERT0(spa->spa_checkpoint_txg);
3494 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3495
3496 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3497 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3498 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3499
3500 if (error == ENOENT)
3501 return (0);
3502
3503 if (error != 0)
3504 return (error);
3505
3506 ASSERT3U(checkpoint.ub_txg, !=, 0);
3507 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3508 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3509 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3510 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3511
3512 return (0);
3513}
3514
3515static int
3516spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3517{
3518 int error = 0;
3519
3520 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3521 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3522
3523 /*
3524 * Never trust the config that is provided unless we are assembling
3525 * a pool following a split.
3526 * This means don't trust blkptrs and the vdev tree in general. This
3527 * also effectively puts the spa in read-only mode since
3528 * spa_writeable() checks for spa_trust_config to be true.
3529 * We will later load a trusted config from the MOS.
3530 */
3531 if (type != SPA_IMPORT_ASSEMBLE)
3532 spa->spa_trust_config = B_FALSE;
3533
3534 /*
3535 * Parse the config provided to create a vdev tree.
3536 */
3537 error = spa_ld_parse_config(spa, type);
3538 if (error != 0)
3539 return (error);
3540
3541 /*
3542 * Now that we have the vdev tree, try to open each vdev. This involves
3543 * opening the underlying physical device, retrieving its geometry and
3544 * probing the vdev with a dummy I/O. The state of each vdev will be set
3545 * based on the success of those operations. After this we'll be ready
3546 * to read from the vdevs.
3547 */
3548 error = spa_ld_open_vdevs(spa);
3549 if (error != 0)
3550 return (error);
3551
3552 /*
3553 * Read the label of each vdev and make sure that the GUIDs stored
3554 * there match the GUIDs in the config provided.
3555 * If we're assembling a new pool that's been split off from an
3556 * existing pool, the labels haven't yet been updated so we skip
3557 * validation for now.
3558 */
3559 if (type != SPA_IMPORT_ASSEMBLE) {
3560 error = spa_ld_validate_vdevs(spa);
3561 if (error != 0)
3562 return (error);
3563 }
3564
3565 /*
3566 * Read all vdev labels to find the best uberblock (i.e. latest,
3567 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3568 * get the list of features required to read blkptrs in the MOS from
3569 * the vdev label with the best uberblock and verify that our version
3570 * of zfs supports them all.
3571 */
3572 error = spa_ld_select_uberblock(spa, type);
3573 if (error != 0)
3574 return (error);
3575
3576 /*
3577 * Pass that uberblock to the dsl_pool layer which will open the root
3578 * blkptr. This blkptr points to the latest version of the MOS and will
3579 * allow us to read its contents.
3580 */
3581 error = spa_ld_open_rootbp(spa);
3582 if (error != 0)
3583 return (error);
3584
3585 return (0);
3586}
3587
3588static int
3589spa_ld_checkpoint_rewind(spa_t *spa)
3590{
3591 uberblock_t checkpoint;
3592 int error = 0;
3593
3594 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3595 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3596
3597 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3598 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3599 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3600
3601 if (error != 0) {
3602 spa_load_failed(spa, "unable to retrieve checkpointed "
3603 "uberblock from the MOS config [error=%d]", error);
3604
3605 if (error == ENOENT)
3606 error = ZFS_ERR_NO_CHECKPOINT;
3607
3608 return (error);
3609 }
3610
3611 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3612 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3613
3614 /*
3615 * We need to update the txg and timestamp of the checkpointed
3616 * uberblock to be higher than the latest one. This ensures that
3617 * the checkpointed uberblock is selected if we were to close and
3618 * reopen the pool right after we've written it in the vdev labels.
3619 * (also see block comment in vdev_uberblock_compare)
3620 */
3621 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3622 checkpoint.ub_timestamp = gethrestime_sec();
3623
3624 /*
3625 * Set current uberblock to be the checkpointed uberblock.
3626 */
3627 spa->spa_uberblock = checkpoint;
3628
3629 /*
3630 * If we are doing a normal rewind, then the pool is open for
3631 * writing and we sync the "updated" checkpointed uberblock to
3632 * disk. Once this is done, we've basically rewound the whole
3633 * pool and there is no way back.
3634 *
3635 * There are cases when we don't want to attempt and sync the
3636 * checkpointed uberblock to disk because we are opening a
3637 * pool as read-only. Specifically, verifying the checkpointed
3638 * state with zdb, and importing the checkpointed state to get
3639 * a "preview" of its content.
3640 */
3641 if (spa_writeable(spa)) {
3642 vdev_t *rvd = spa->spa_root_vdev;
3643
3644 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3645 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3646 int svdcount = 0;
3647 int children = rvd->vdev_children;
3648 int c0 = spa_get_random(children);
3649
3650 for (int c = 0; c < children; c++) {
3651 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3652
3653 /* Stop when revisiting the first vdev */
3654 if (c > 0 && svd[0] == vd)
3655 break;
3656
3657 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3658 !vdev_is_concrete(vd))
3659 continue;
3660
3661 svd[svdcount++] = vd;
3662 if (svdcount == SPA_SYNC_MIN_VDEVS)
3663 break;
3664 }
3665 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3666 if (error == 0)
3667 spa->spa_last_synced_guid = rvd->vdev_guid;
3668 spa_config_exit(spa, SCL_ALL, FTAG);
3669
3670 if (error != 0) {
3671 spa_load_failed(spa, "failed to write checkpointed "
3672 "uberblock to the vdev labels [error=%d]", error);
3673 return (error);
3674 }
3675 }
3676
3677 return (0);
3678}
3679
3680static int
3681spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3682 boolean_t *update_config_cache)
3683{
3684 int error;
3685
3686 /*
3687 * Parse the config for pool, open and validate vdevs,
3688 * select an uberblock, and use that uberblock to open
3689 * the MOS.
3690 */
3691 error = spa_ld_mos_init(spa, type);
3692 if (error != 0)
3693 return (error);
3694
3695 /*
3696 * Retrieve the trusted config stored in the MOS and use it to create
3697 * a new, exact version of the vdev tree, then reopen all vdevs.
3698 */
3699 error = spa_ld_trusted_config(spa, type, B_FALSE);
3700 if (error == EAGAIN) {
3701 if (update_config_cache != NULL)
3702 *update_config_cache = B_TRUE;
3703
3704 /*
3705 * Redo the loading process with the trusted config if it is
3706 * too different from the untrusted config.
3707 */
3708 spa_ld_prepare_for_reload(spa);
3709 spa_load_note(spa, "RELOADING");
3710 error = spa_ld_mos_init(spa, type);
3711 if (error != 0)
3712 return (error);
3713
3714 error = spa_ld_trusted_config(spa, type, B_TRUE);
3715 if (error != 0)
3716 return (error);
3717
3718 } else if (error != 0) {
3719 return (error);
3720 }
3721
3722 return (0);
3723}
3724
3725/*
3726 * Load an existing storage pool, using the config provided. This config
3727 * describes which vdevs are part of the pool and is later validated against
3728 * partial configs present in each vdev's label and an entire copy of the
3729 * config stored in the MOS.
3730 */
3731static int
3732spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3733{
3734 int error = 0;
3735 boolean_t missing_feat_write = B_FALSE;
3736 boolean_t checkpoint_rewind =
3737 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3738 boolean_t update_config_cache = B_FALSE;
3739
3740 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3741 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3742
3743 spa_load_note(spa, "LOADING");
3744
3745 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3746 if (error != 0)
3747 return (error);
3748
3749 /*
3750 * If we are rewinding to the checkpoint then we need to repeat
3751 * everything we've done so far in this function but this time
3752 * selecting the checkpointed uberblock and using that to open
3753 * the MOS.
3754 */
3755 if (checkpoint_rewind) {
3756 /*
3757 * If we are rewinding to the checkpoint update config cache
3758 * anyway.
3759 */
3760 update_config_cache = B_TRUE;
3761
3762 /*
3763 * Extract the checkpointed uberblock from the current MOS
3764 * and use this as the pool's uberblock from now on. If the
3765 * pool is imported as writeable we also write the checkpoint
3766 * uberblock to the labels, making the rewind permanent.
3767 */
3768 error = spa_ld_checkpoint_rewind(spa);
3769 if (error != 0)
3770 return (error);
3771
3772 /*
3773 * Redo the loading process process again with the
3774 * checkpointed uberblock.
3775 */
3776 spa_ld_prepare_for_reload(spa);
3777 spa_load_note(spa, "LOADING checkpointed uberblock");
3778 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3779 if (error != 0)
3780 return (error);
3781 }
3782
3783 /*
3784 * Retrieve the checkpoint txg if the pool has a checkpoint.
3785 */
3786 error = spa_ld_read_checkpoint_txg(spa);
3787 if (error != 0)
3788 return (error);
3789
3790 /*
3791 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3792 * from the pool and their contents were re-mapped to other vdevs. Note
3793 * that everything that we read before this step must have been
3794 * rewritten on concrete vdevs after the last device removal was
3795 * initiated. Otherwise we could be reading from indirect vdevs before
3796 * we have loaded their mappings.
3797 */
3798 error = spa_ld_open_indirect_vdev_metadata(spa);
3799 if (error != 0)
3800 return (error);
3801
3802 /*
3803 * Retrieve the full list of active features from the MOS and check if
3804 * they are all supported.
3805 */
3806 error = spa_ld_check_features(spa, &missing_feat_write);
3807 if (error != 0)
3808 return (error);
3809
3810 /*
3811 * Load several special directories from the MOS needed by the dsl_pool
3812 * layer.
3813 */
3814 error = spa_ld_load_special_directories(spa);
3815 if (error != 0)
3816 return (error);
3817
3818 /*
3819 * Retrieve pool properties from the MOS.
3820 */
3821 error = spa_ld_get_props(spa);
3822 if (error != 0)
3823 return (error);
3824
3825 /*
3826 * Retrieve the list of auxiliary devices - cache devices and spares -
3827 * and open them.
3828 */
3829 error = spa_ld_open_aux_vdevs(spa, type);
3830 if (error != 0)
3831 return (error);
3832
3833 /*
3834 * Load the metadata for all vdevs. Also check if unopenable devices
3835 * should be autoreplaced.
3836 */
3837 error = spa_ld_load_vdev_metadata(spa);
3838 if (error != 0)
3839 return (error);
3840
3841 error = spa_ld_load_dedup_tables(spa);
3842 if (error != 0)
3843 return (error);
3844
3845 /*
3846 * Verify the logs now to make sure we don't have any unexpected errors
3847 * when we claim log blocks later.
3848 */
3849 error = spa_ld_verify_logs(spa, type, ereport);
3850 if (error != 0)
3851 return (error);
3852
3853 if (missing_feat_write) {
3854 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3855
3856 /*
3857 * At this point, we know that we can open the pool in
3858 * read-only mode but not read-write mode. We now have enough
3859 * information and can return to userland.
3860 */
3861 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3862 ENOTSUP));
3863 }
3864
3865 /*
3866 * Traverse the last txgs to make sure the pool was left off in a safe
3867 * state. When performing an extreme rewind, we verify the whole pool,
3868 * which can take a very long time.
3869 */
3870 error = spa_ld_verify_pool_data(spa);
3871 if (error != 0)
3872 return (error);
3873
3874 /*
3875 * Calculate the deflated space for the pool. This must be done before
3876 * we write anything to the pool because we'd need to update the space
3877 * accounting using the deflated sizes.
3878 */
3879 spa_update_dspace(spa);
3880
3881 /*
3882 * We have now retrieved all the information we needed to open the
3883 * pool. If we are importing the pool in read-write mode, a few
3884 * additional steps must be performed to finish the import.
3885 */
3886 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3887 spa->spa_load_max_txg == UINT64_MAX)) {
3888 uint64_t config_cache_txg = spa->spa_config_txg;
3889
3890 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3891
3892 /*
3893 * In case of a checkpoint rewind, log the original txg
3894 * of the checkpointed uberblock.
3895 */
3896 if (checkpoint_rewind) {
3897 spa_history_log_internal(spa, "checkpoint rewind",
3898 NULL, "rewound state to txg=%llu",
3899 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3900 }
3901
3902 /*
3903 * Traverse the ZIL and claim all blocks.
3904 */
3905 spa_ld_claim_log_blocks(spa);
3906
3907 /*
3908 * Kick-off the syncing thread.
3909 */
3910 spa->spa_sync_on = B_TRUE;
3911 txg_sync_start(spa->spa_dsl_pool);
3912
3913 /*
3914 * Wait for all claims to sync. We sync up to the highest
3915 * claimed log block birth time so that claimed log blocks
3916 * don't appear to be from the future. spa_claim_max_txg
3917 * will have been set for us by ZIL traversal operations
3918 * performed above.
3919 */
3920 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3921
3922 /*
3923 * Check if we need to request an update of the config. On the
3924 * next sync, we would update the config stored in vdev labels
3925 * and the cachefile (by default /etc/zfs/zpool.cache).
3926 */
3927 spa_ld_check_for_config_update(spa, config_cache_txg,
3928 update_config_cache);
3929
3930 /*
3931 * Check all DTLs to see if anything needs resilvering.
3932 */
3933 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3934 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3935 spa_async_request(spa, SPA_ASYNC_RESILVER);
3936
3937 /*
3938 * Log the fact that we booted up (so that we can detect if
3939 * we rebooted in the middle of an operation).
3940 */
3941 spa_history_log_version(spa, "open");
3942
3943 /*
3944 * Delete any inconsistent datasets.
3945 */
3946 (void) dmu_objset_find(spa_name(spa),
3947 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3948
3949 /*
3950 * Clean up any stale temporary dataset userrefs.
3951 */
3952 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3953
3954 spa_restart_removal(spa);
3955
3956 spa_spawn_aux_threads(spa);
3957 }
3958
3959 spa_load_note(spa, "LOADED");
3960
3961 return (0);
3962}
3963
3964static int
3965spa_load_retry(spa_t *spa, spa_load_state_t state)
3966{
3967 int mode = spa->spa_mode;
3968
3969 spa_unload(spa);
3970 spa_deactivate(spa);
3971
3972 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3973
3974 spa_activate(spa, mode);
3975 spa_async_suspend(spa);
3976
3977 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3978 (u_longlong_t)spa->spa_load_max_txg);
3979
3980 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
3981}
3982
3983/*
3984 * If spa_load() fails this function will try loading prior txg's. If
3985 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3986 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3987 * function will not rewind the pool and will return the same error as
3988 * spa_load().
3989 */
3990static int
3991spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
3992 int rewind_flags)
3993{
3994 nvlist_t *loadinfo = NULL;
3995 nvlist_t *config = NULL;
3996 int load_error, rewind_error;
3997 uint64_t safe_rewind_txg;
3998 uint64_t min_txg;
3999
4000 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4001 spa->spa_load_max_txg = spa->spa_load_txg;
4002 spa_set_log_state(spa, SPA_LOG_CLEAR);
4003 } else {
4004 spa->spa_load_max_txg = max_request;
4005 if (max_request != UINT64_MAX)
4006 spa->spa_extreme_rewind = B_TRUE;
4007 }
4008
4009 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4010 if (load_error == 0)
4011 return (0);
4012 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4013 /*
4014 * When attempting checkpoint-rewind on a pool with no
4015 * checkpoint, we should not attempt to load uberblocks
4016 * from previous txgs when spa_load fails.
4017 */
4018 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4019 return (load_error);
4020 }
4021
4022 if (spa->spa_root_vdev != NULL)
4023 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4024
4025 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4026 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4027
4028 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4029 nvlist_free(config);
4030 return (load_error);
4031 }
4032
4033 if (state == SPA_LOAD_RECOVER) {
4034 /* Price of rolling back is discarding txgs, including log */
4035 spa_set_log_state(spa, SPA_LOG_CLEAR);
4036 } else {
4037 /*
4038 * If we aren't rolling back save the load info from our first
4039 * import attempt so that we can restore it after attempting
4040 * to rewind.
4041 */
4042 loadinfo = spa->spa_load_info;
4043 spa->spa_load_info = fnvlist_alloc();
4044 }
4045
4046 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4047 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4048 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4049 TXG_INITIAL : safe_rewind_txg;
4050
4051 /*
4052 * Continue as long as we're finding errors, we're still within
4053 * the acceptable rewind range, and we're still finding uberblocks
4054 */
4055 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4056 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4057 if (spa->spa_load_max_txg < safe_rewind_txg)
4058 spa->spa_extreme_rewind = B_TRUE;
4059 rewind_error = spa_load_retry(spa, state);
4060 }
4061
4062 spa->spa_extreme_rewind = B_FALSE;
4063 spa->spa_load_max_txg = UINT64_MAX;
4064
4065 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4066 spa_config_set(spa, config);
4067 else
4068 nvlist_free(config);
4069
4070 if (state == SPA_LOAD_RECOVER) {
4071 ASSERT3P(loadinfo, ==, NULL);
4072 return (rewind_error);
4073 } else {
4074 /* Store the rewind info as part of the initial load info */
4075 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4076 spa->spa_load_info);
4077
4078 /* Restore the initial load info */
4079 fnvlist_free(spa->spa_load_info);
4080 spa->spa_load_info = loadinfo;
4081
4082 return (load_error);
4083 }
4084}
4085
4086/*
4087 * Pool Open/Import
4088 *
4089 * The import case is identical to an open except that the configuration is sent
4090 * down from userland, instead of grabbed from the configuration cache. For the
4091 * case of an open, the pool configuration will exist in the
4092 * POOL_STATE_UNINITIALIZED state.
4093 *
4094 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4095 * the same time open the pool, without having to keep around the spa_t in some
4096 * ambiguous state.
4097 */
4098static int
4099spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4100 nvlist_t **config)
4101{
4102 spa_t *spa;
4103 spa_load_state_t state = SPA_LOAD_OPEN;
4104 int error;
4105 int locked = B_FALSE;
4106 int firstopen = B_FALSE;
4107
4108 *spapp = NULL;
4109
4110 /*
4111 * As disgusting as this is, we need to support recursive calls to this
4112 * function because dsl_dir_open() is called during spa_load(), and ends
4113 * up calling spa_open() again. The real fix is to figure out how to
4114 * avoid dsl_dir_open() calling this in the first place.
4115 */
4116 if (mutex_owner(&spa_namespace_lock) != curthread) {
4117 mutex_enter(&spa_namespace_lock);
4118 locked = B_TRUE;
4119 }
4120
4121 if ((spa = spa_lookup(pool)) == NULL) {
4122 if (locked)
4123 mutex_exit(&spa_namespace_lock);
4124 return (SET_ERROR(ENOENT));
4125 }
4126
4127 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
| 2869 spa_config_set(spa, mos_config);
2870 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2871
2872 /*
2873 * Now that we got the config from the MOS, we should be more strict
2874 * in checking blkptrs and can make assumptions about the consistency
2875 * of the vdev tree. spa_trust_config must be set to true before opening
2876 * vdevs in order for them to be writeable.
2877 */
2878 spa->spa_trust_config = B_TRUE;
2879
2880 /*
2881 * Open and validate the new vdev tree
2882 */
2883 error = spa_ld_open_vdevs(spa);
2884 if (error != 0)
2885 return (error);
2886
2887 error = spa_ld_validate_vdevs(spa);
2888 if (error != 0)
2889 return (error);
2890
2891 if (copy_error != 0 || spa_load_print_vdev_tree) {
2892 spa_load_note(spa, "final vdev tree:");
2893 vdev_dbgmsg_print_tree(rvd, 2);
2894 }
2895
2896 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2897 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2898 /*
2899 * Sanity check to make sure that we are indeed loading the
2900 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2901 * in the config provided and they happened to be the only ones
2902 * to have the latest uberblock, we could involuntarily perform
2903 * an extreme rewind.
2904 */
2905 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2906 if (healthy_tvds_mos - healthy_tvds >=
2907 SPA_SYNC_MIN_VDEVS) {
2908 spa_load_note(spa, "config provided misses too many "
2909 "top-level vdevs compared to MOS (%lld vs %lld). ",
2910 (u_longlong_t)healthy_tvds,
2911 (u_longlong_t)healthy_tvds_mos);
2912 spa_load_note(spa, "vdev tree:");
2913 vdev_dbgmsg_print_tree(rvd, 2);
2914 if (reloading) {
2915 spa_load_failed(spa, "config was already "
2916 "provided from MOS. Aborting.");
2917 return (spa_vdev_err(rvd,
2918 VDEV_AUX_CORRUPT_DATA, EIO));
2919 }
2920 spa_load_note(spa, "spa must be reloaded using MOS "
2921 "config");
2922 return (SET_ERROR(EAGAIN));
2923 }
2924 }
2925
2926 error = spa_check_for_missing_logs(spa);
2927 if (error != 0)
2928 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2929
2930 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2931 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2932 "guid sum (%llu != %llu)",
2933 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2934 (u_longlong_t)rvd->vdev_guid_sum);
2935 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2936 ENXIO));
2937 }
2938
2939 return (0);
2940}
2941
2942static int
2943spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2944{
2945 int error = 0;
2946 vdev_t *rvd = spa->spa_root_vdev;
2947
2948 /*
2949 * Everything that we read before spa_remove_init() must be stored
2950 * on concreted vdevs. Therefore we do this as early as possible.
2951 */
2952 error = spa_remove_init(spa);
2953 if (error != 0) {
2954 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2955 error);
2956 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2957 }
2958
2959 /*
2960 * Retrieve information needed to condense indirect vdev mappings.
2961 */
2962 error = spa_condense_init(spa);
2963 if (error != 0) {
2964 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2965 error);
2966 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2967 }
2968
2969 return (0);
2970}
2971
2972static int
2973spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2974{
2975 int error = 0;
2976 vdev_t *rvd = spa->spa_root_vdev;
2977
2978 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2979 boolean_t missing_feat_read = B_FALSE;
2980 nvlist_t *unsup_feat, *enabled_feat;
2981
2982 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2983 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2985 }
2986
2987 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2988 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
2989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2990 }
2991
2992 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2993 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
2994 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2995 }
2996
2997 enabled_feat = fnvlist_alloc();
2998 unsup_feat = fnvlist_alloc();
2999
3000 if (!spa_features_check(spa, B_FALSE,
3001 unsup_feat, enabled_feat))
3002 missing_feat_read = B_TRUE;
3003
3004 if (spa_writeable(spa) ||
3005 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3006 if (!spa_features_check(spa, B_TRUE,
3007 unsup_feat, enabled_feat)) {
3008 *missing_feat_writep = B_TRUE;
3009 }
3010 }
3011
3012 fnvlist_add_nvlist(spa->spa_load_info,
3013 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3014
3015 if (!nvlist_empty(unsup_feat)) {
3016 fnvlist_add_nvlist(spa->spa_load_info,
3017 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3018 }
3019
3020 fnvlist_free(enabled_feat);
3021 fnvlist_free(unsup_feat);
3022
3023 if (!missing_feat_read) {
3024 fnvlist_add_boolean(spa->spa_load_info,
3025 ZPOOL_CONFIG_CAN_RDONLY);
3026 }
3027
3028 /*
3029 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3030 * twofold: to determine whether the pool is available for
3031 * import in read-write mode and (if it is not) whether the
3032 * pool is available for import in read-only mode. If the pool
3033 * is available for import in read-write mode, it is displayed
3034 * as available in userland; if it is not available for import
3035 * in read-only mode, it is displayed as unavailable in
3036 * userland. If the pool is available for import in read-only
3037 * mode but not read-write mode, it is displayed as unavailable
3038 * in userland with a special note that the pool is actually
3039 * available for open in read-only mode.
3040 *
3041 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3042 * missing a feature for write, we must first determine whether
3043 * the pool can be opened read-only before returning to
3044 * userland in order to know whether to display the
3045 * abovementioned note.
3046 */
3047 if (missing_feat_read || (*missing_feat_writep &&
3048 spa_writeable(spa))) {
3049 spa_load_failed(spa, "pool uses unsupported features");
3050 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3051 ENOTSUP));
3052 }
3053
3054 /*
3055 * Load refcounts for ZFS features from disk into an in-memory
3056 * cache during SPA initialization.
3057 */
3058 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3059 uint64_t refcount;
3060
3061 error = feature_get_refcount_from_disk(spa,
3062 &spa_feature_table[i], &refcount);
3063 if (error == 0) {
3064 spa->spa_feat_refcount_cache[i] = refcount;
3065 } else if (error == ENOTSUP) {
3066 spa->spa_feat_refcount_cache[i] =
3067 SPA_FEATURE_DISABLED;
3068 } else {
3069 spa_load_failed(spa, "error getting refcount "
3070 "for feature %s [error=%d]",
3071 spa_feature_table[i].fi_guid, error);
3072 return (spa_vdev_err(rvd,
3073 VDEV_AUX_CORRUPT_DATA, EIO));
3074 }
3075 }
3076 }
3077
3078 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3079 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3080 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3082 }
3083
3084 return (0);
3085}
3086
3087static int
3088spa_ld_load_special_directories(spa_t *spa)
3089{
3090 int error = 0;
3091 vdev_t *rvd = spa->spa_root_vdev;
3092
3093 spa->spa_is_initializing = B_TRUE;
3094 error = dsl_pool_open(spa->spa_dsl_pool);
3095 spa->spa_is_initializing = B_FALSE;
3096 if (error != 0) {
3097 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3099 }
3100
3101 return (0);
3102}
3103
3104static int
3105spa_ld_get_props(spa_t *spa)
3106{
3107 int error = 0;
3108 uint64_t obj;
3109 vdev_t *rvd = spa->spa_root_vdev;
3110
3111 /* Grab the secret checksum salt from the MOS. */
3112 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3113 DMU_POOL_CHECKSUM_SALT, 1,
3114 sizeof (spa->spa_cksum_salt.zcs_bytes),
3115 spa->spa_cksum_salt.zcs_bytes);
3116 if (error == ENOENT) {
3117 /* Generate a new salt for subsequent use */
3118 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3119 sizeof (spa->spa_cksum_salt.zcs_bytes));
3120 } else if (error != 0) {
3121 spa_load_failed(spa, "unable to retrieve checksum salt from "
3122 "MOS [error=%d]", error);
3123 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3124 }
3125
3126 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3127 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3128 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3129 if (error != 0) {
3130 spa_load_failed(spa, "error opening deferred-frees bpobj "
3131 "[error=%d]", error);
3132 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3133 }
3134
3135 /*
3136 * Load the bit that tells us to use the new accounting function
3137 * (raid-z deflation). If we have an older pool, this will not
3138 * be present.
3139 */
3140 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3141 if (error != 0 && error != ENOENT)
3142 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3143
3144 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3145 &spa->spa_creation_version, B_FALSE);
3146 if (error != 0 && error != ENOENT)
3147 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3148
3149 /*
3150 * Load the persistent error log. If we have an older pool, this will
3151 * not be present.
3152 */
3153 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3154 B_FALSE);
3155 if (error != 0 && error != ENOENT)
3156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3157
3158 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3159 &spa->spa_errlog_scrub, B_FALSE);
3160 if (error != 0 && error != ENOENT)
3161 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3162
3163 /*
3164 * Load the history object. If we have an older pool, this
3165 * will not be present.
3166 */
3167 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3168 if (error != 0 && error != ENOENT)
3169 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3170
3171 /*
3172 * Load the per-vdev ZAP map. If we have an older pool, this will not
3173 * be present; in this case, defer its creation to a later time to
3174 * avoid dirtying the MOS this early / out of sync context. See
3175 * spa_sync_config_object.
3176 */
3177
3178 /* The sentinel is only available in the MOS config. */
3179 nvlist_t *mos_config;
3180 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3181 spa_load_failed(spa, "unable to retrieve MOS config");
3182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3183 }
3184
3185 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3186 &spa->spa_all_vdev_zaps, B_FALSE);
3187
3188 if (error == ENOENT) {
3189 VERIFY(!nvlist_exists(mos_config,
3190 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3191 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3192 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3193 } else if (error != 0) {
3194 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3195 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3196 /*
3197 * An older version of ZFS overwrote the sentinel value, so
3198 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3199 * destruction to later; see spa_sync_config_object.
3200 */
3201 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3202 /*
3203 * We're assuming that no vdevs have had their ZAPs created
3204 * before this. Better be sure of it.
3205 */
3206 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3207 }
3208 nvlist_free(mos_config);
3209
3210 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3211
3212 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3213 B_FALSE);
3214 if (error && error != ENOENT)
3215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3216
3217 if (error == 0) {
3218 uint64_t autoreplace;
3219
3220 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3221 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3222 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3223 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3224 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3225 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3226 &spa->spa_dedup_ditto);
3227
3228 spa->spa_autoreplace = (autoreplace != 0);
3229 }
3230
3231 /*
3232 * If we are importing a pool with missing top-level vdevs,
3233 * we enforce that the pool doesn't panic or get suspended on
3234 * error since the likelihood of missing data is extremely high.
3235 */
3236 if (spa->spa_missing_tvds > 0 &&
3237 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3238 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3239 spa_load_note(spa, "forcing failmode to 'continue' "
3240 "as some top level vdevs are missing");
3241 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3242 }
3243
3244 return (0);
3245}
3246
3247static int
3248spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3249{
3250 int error = 0;
3251 vdev_t *rvd = spa->spa_root_vdev;
3252
3253 /*
3254 * If we're assembling the pool from the split-off vdevs of
3255 * an existing pool, we don't want to attach the spares & cache
3256 * devices.
3257 */
3258
3259 /*
3260 * Load any hot spares for this pool.
3261 */
3262 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3263 B_FALSE);
3264 if (error != 0 && error != ENOENT)
3265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3266 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3267 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3268 if (load_nvlist(spa, spa->spa_spares.sav_object,
3269 &spa->spa_spares.sav_config) != 0) {
3270 spa_load_failed(spa, "error loading spares nvlist");
3271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3272 }
3273
3274 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3275 spa_load_spares(spa);
3276 spa_config_exit(spa, SCL_ALL, FTAG);
3277 } else if (error == 0) {
3278 spa->spa_spares.sav_sync = B_TRUE;
3279 }
3280
3281 /*
3282 * Load any level 2 ARC devices for this pool.
3283 */
3284 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3285 &spa->spa_l2cache.sav_object, B_FALSE);
3286 if (error != 0 && error != ENOENT)
3287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3288 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3289 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3290 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3291 &spa->spa_l2cache.sav_config) != 0) {
3292 spa_load_failed(spa, "error loading l2cache nvlist");
3293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3294 }
3295
3296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3297 spa_load_l2cache(spa);
3298 spa_config_exit(spa, SCL_ALL, FTAG);
3299 } else if (error == 0) {
3300 spa->spa_l2cache.sav_sync = B_TRUE;
3301 }
3302
3303 return (0);
3304}
3305
3306static int
3307spa_ld_load_vdev_metadata(spa_t *spa)
3308{
3309 int error = 0;
3310 vdev_t *rvd = spa->spa_root_vdev;
3311
3312 /*
3313 * If the 'autoreplace' property is set, then post a resource notifying
3314 * the ZFS DE that it should not issue any faults for unopenable
3315 * devices. We also iterate over the vdevs, and post a sysevent for any
3316 * unopenable vdevs so that the normal autoreplace handler can take
3317 * over.
3318 */
3319 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3320 spa_check_removed(spa->spa_root_vdev);
3321 /*
3322 * For the import case, this is done in spa_import(), because
3323 * at this point we're using the spare definitions from
3324 * the MOS config, not necessarily from the userland config.
3325 */
3326 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3327 spa_aux_check_removed(&spa->spa_spares);
3328 spa_aux_check_removed(&spa->spa_l2cache);
3329 }
3330 }
3331
3332 /*
3333 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3334 */
3335 error = vdev_load(rvd);
3336 if (error != 0) {
3337 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3338 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3339 }
3340
3341 /*
3342 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3343 */
3344 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3345 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3346 spa_config_exit(spa, SCL_ALL, FTAG);
3347
3348 return (0);
3349}
3350
3351static int
3352spa_ld_load_dedup_tables(spa_t *spa)
3353{
3354 int error = 0;
3355 vdev_t *rvd = spa->spa_root_vdev;
3356
3357 error = ddt_load(spa);
3358 if (error != 0) {
3359 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3361 }
3362
3363 return (0);
3364}
3365
3366static int
3367spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3368{
3369 vdev_t *rvd = spa->spa_root_vdev;
3370
3371 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3372 boolean_t missing = spa_check_logs(spa);
3373 if (missing) {
3374 if (spa->spa_missing_tvds != 0) {
3375 spa_load_note(spa, "spa_check_logs failed "
3376 "so dropping the logs");
3377 } else {
3378 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3379 spa_load_failed(spa, "spa_check_logs failed");
3380 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3381 ENXIO));
3382 }
3383 }
3384 }
3385
3386 return (0);
3387}
3388
3389static int
3390spa_ld_verify_pool_data(spa_t *spa)
3391{
3392 int error = 0;
3393 vdev_t *rvd = spa->spa_root_vdev;
3394
3395 /*
3396 * We've successfully opened the pool, verify that we're ready
3397 * to start pushing transactions.
3398 */
3399 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3400 error = spa_load_verify(spa);
3401 if (error != 0) {
3402 spa_load_failed(spa, "spa_load_verify failed "
3403 "[error=%d]", error);
3404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3405 error));
3406 }
3407 }
3408
3409 return (0);
3410}
3411
3412static void
3413spa_ld_claim_log_blocks(spa_t *spa)
3414{
3415 dmu_tx_t *tx;
3416 dsl_pool_t *dp = spa_get_dsl(spa);
3417
3418 /*
3419 * Claim log blocks that haven't been committed yet.
3420 * This must all happen in a single txg.
3421 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3422 * invoked from zil_claim_log_block()'s i/o done callback.
3423 * Price of rollback is that we abandon the log.
3424 */
3425 spa->spa_claiming = B_TRUE;
3426
3427 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3428 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3429 zil_claim, tx, DS_FIND_CHILDREN);
3430 dmu_tx_commit(tx);
3431
3432 spa->spa_claiming = B_FALSE;
3433
3434 spa_set_log_state(spa, SPA_LOG_GOOD);
3435}
3436
3437static void
3438spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3439 boolean_t update_config_cache)
3440{
3441 vdev_t *rvd = spa->spa_root_vdev;
3442 int need_update = B_FALSE;
3443
3444 /*
3445 * If the config cache is stale, or we have uninitialized
3446 * metaslabs (see spa_vdev_add()), then update the config.
3447 *
3448 * If this is a verbatim import, trust the current
3449 * in-core spa_config and update the disk labels.
3450 */
3451 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3452 spa->spa_load_state == SPA_LOAD_IMPORT ||
3453 spa->spa_load_state == SPA_LOAD_RECOVER ||
3454 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3455 need_update = B_TRUE;
3456
3457 for (int c = 0; c < rvd->vdev_children; c++)
3458 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3459 need_update = B_TRUE;
3460
3461 /*
3462 * Update the config cache asychronously in case we're the
3463 * root pool, in which case the config cache isn't writable yet.
3464 */
3465 if (need_update)
3466 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3467}
3468
3469static void
3470spa_ld_prepare_for_reload(spa_t *spa)
3471{
3472 int mode = spa->spa_mode;
3473 int async_suspended = spa->spa_async_suspended;
3474
3475 spa_unload(spa);
3476 spa_deactivate(spa);
3477 spa_activate(spa, mode);
3478
3479 /*
3480 * We save the value of spa_async_suspended as it gets reset to 0 by
3481 * spa_unload(). We want to restore it back to the original value before
3482 * returning as we might be calling spa_async_resume() later.
3483 */
3484 spa->spa_async_suspended = async_suspended;
3485}
3486
3487static int
3488spa_ld_read_checkpoint_txg(spa_t *spa)
3489{
3490 uberblock_t checkpoint;
3491 int error = 0;
3492
3493 ASSERT0(spa->spa_checkpoint_txg);
3494 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3495
3496 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3497 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3498 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3499
3500 if (error == ENOENT)
3501 return (0);
3502
3503 if (error != 0)
3504 return (error);
3505
3506 ASSERT3U(checkpoint.ub_txg, !=, 0);
3507 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3508 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3509 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3510 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3511
3512 return (0);
3513}
3514
3515static int
3516spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3517{
3518 int error = 0;
3519
3520 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3521 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3522
3523 /*
3524 * Never trust the config that is provided unless we are assembling
3525 * a pool following a split.
3526 * This means don't trust blkptrs and the vdev tree in general. This
3527 * also effectively puts the spa in read-only mode since
3528 * spa_writeable() checks for spa_trust_config to be true.
3529 * We will later load a trusted config from the MOS.
3530 */
3531 if (type != SPA_IMPORT_ASSEMBLE)
3532 spa->spa_trust_config = B_FALSE;
3533
3534 /*
3535 * Parse the config provided to create a vdev tree.
3536 */
3537 error = spa_ld_parse_config(spa, type);
3538 if (error != 0)
3539 return (error);
3540
3541 /*
3542 * Now that we have the vdev tree, try to open each vdev. This involves
3543 * opening the underlying physical device, retrieving its geometry and
3544 * probing the vdev with a dummy I/O. The state of each vdev will be set
3545 * based on the success of those operations. After this we'll be ready
3546 * to read from the vdevs.
3547 */
3548 error = spa_ld_open_vdevs(spa);
3549 if (error != 0)
3550 return (error);
3551
3552 /*
3553 * Read the label of each vdev and make sure that the GUIDs stored
3554 * there match the GUIDs in the config provided.
3555 * If we're assembling a new pool that's been split off from an
3556 * existing pool, the labels haven't yet been updated so we skip
3557 * validation for now.
3558 */
3559 if (type != SPA_IMPORT_ASSEMBLE) {
3560 error = spa_ld_validate_vdevs(spa);
3561 if (error != 0)
3562 return (error);
3563 }
3564
3565 /*
3566 * Read all vdev labels to find the best uberblock (i.e. latest,
3567 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3568 * get the list of features required to read blkptrs in the MOS from
3569 * the vdev label with the best uberblock and verify that our version
3570 * of zfs supports them all.
3571 */
3572 error = spa_ld_select_uberblock(spa, type);
3573 if (error != 0)
3574 return (error);
3575
3576 /*
3577 * Pass that uberblock to the dsl_pool layer which will open the root
3578 * blkptr. This blkptr points to the latest version of the MOS and will
3579 * allow us to read its contents.
3580 */
3581 error = spa_ld_open_rootbp(spa);
3582 if (error != 0)
3583 return (error);
3584
3585 return (0);
3586}
3587
3588static int
3589spa_ld_checkpoint_rewind(spa_t *spa)
3590{
3591 uberblock_t checkpoint;
3592 int error = 0;
3593
3594 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3595 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3596
3597 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3598 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3599 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3600
3601 if (error != 0) {
3602 spa_load_failed(spa, "unable to retrieve checkpointed "
3603 "uberblock from the MOS config [error=%d]", error);
3604
3605 if (error == ENOENT)
3606 error = ZFS_ERR_NO_CHECKPOINT;
3607
3608 return (error);
3609 }
3610
3611 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3612 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3613
3614 /*
3615 * We need to update the txg and timestamp of the checkpointed
3616 * uberblock to be higher than the latest one. This ensures that
3617 * the checkpointed uberblock is selected if we were to close and
3618 * reopen the pool right after we've written it in the vdev labels.
3619 * (also see block comment in vdev_uberblock_compare)
3620 */
3621 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3622 checkpoint.ub_timestamp = gethrestime_sec();
3623
3624 /*
3625 * Set current uberblock to be the checkpointed uberblock.
3626 */
3627 spa->spa_uberblock = checkpoint;
3628
3629 /*
3630 * If we are doing a normal rewind, then the pool is open for
3631 * writing and we sync the "updated" checkpointed uberblock to
3632 * disk. Once this is done, we've basically rewound the whole
3633 * pool and there is no way back.
3634 *
3635 * There are cases when we don't want to attempt and sync the
3636 * checkpointed uberblock to disk because we are opening a
3637 * pool as read-only. Specifically, verifying the checkpointed
3638 * state with zdb, and importing the checkpointed state to get
3639 * a "preview" of its content.
3640 */
3641 if (spa_writeable(spa)) {
3642 vdev_t *rvd = spa->spa_root_vdev;
3643
3644 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3645 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3646 int svdcount = 0;
3647 int children = rvd->vdev_children;
3648 int c0 = spa_get_random(children);
3649
3650 for (int c = 0; c < children; c++) {
3651 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3652
3653 /* Stop when revisiting the first vdev */
3654 if (c > 0 && svd[0] == vd)
3655 break;
3656
3657 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3658 !vdev_is_concrete(vd))
3659 continue;
3660
3661 svd[svdcount++] = vd;
3662 if (svdcount == SPA_SYNC_MIN_VDEVS)
3663 break;
3664 }
3665 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3666 if (error == 0)
3667 spa->spa_last_synced_guid = rvd->vdev_guid;
3668 spa_config_exit(spa, SCL_ALL, FTAG);
3669
3670 if (error != 0) {
3671 spa_load_failed(spa, "failed to write checkpointed "
3672 "uberblock to the vdev labels [error=%d]", error);
3673 return (error);
3674 }
3675 }
3676
3677 return (0);
3678}
3679
3680static int
3681spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3682 boolean_t *update_config_cache)
3683{
3684 int error;
3685
3686 /*
3687 * Parse the config for pool, open and validate vdevs,
3688 * select an uberblock, and use that uberblock to open
3689 * the MOS.
3690 */
3691 error = spa_ld_mos_init(spa, type);
3692 if (error != 0)
3693 return (error);
3694
3695 /*
3696 * Retrieve the trusted config stored in the MOS and use it to create
3697 * a new, exact version of the vdev tree, then reopen all vdevs.
3698 */
3699 error = spa_ld_trusted_config(spa, type, B_FALSE);
3700 if (error == EAGAIN) {
3701 if (update_config_cache != NULL)
3702 *update_config_cache = B_TRUE;
3703
3704 /*
3705 * Redo the loading process with the trusted config if it is
3706 * too different from the untrusted config.
3707 */
3708 spa_ld_prepare_for_reload(spa);
3709 spa_load_note(spa, "RELOADING");
3710 error = spa_ld_mos_init(spa, type);
3711 if (error != 0)
3712 return (error);
3713
3714 error = spa_ld_trusted_config(spa, type, B_TRUE);
3715 if (error != 0)
3716 return (error);
3717
3718 } else if (error != 0) {
3719 return (error);
3720 }
3721
3722 return (0);
3723}
3724
3725/*
3726 * Load an existing storage pool, using the config provided. This config
3727 * describes which vdevs are part of the pool and is later validated against
3728 * partial configs present in each vdev's label and an entire copy of the
3729 * config stored in the MOS.
3730 */
3731static int
3732spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3733{
3734 int error = 0;
3735 boolean_t missing_feat_write = B_FALSE;
3736 boolean_t checkpoint_rewind =
3737 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3738 boolean_t update_config_cache = B_FALSE;
3739
3740 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3741 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3742
3743 spa_load_note(spa, "LOADING");
3744
3745 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3746 if (error != 0)
3747 return (error);
3748
3749 /*
3750 * If we are rewinding to the checkpoint then we need to repeat
3751 * everything we've done so far in this function but this time
3752 * selecting the checkpointed uberblock and using that to open
3753 * the MOS.
3754 */
3755 if (checkpoint_rewind) {
3756 /*
3757 * If we are rewinding to the checkpoint update config cache
3758 * anyway.
3759 */
3760 update_config_cache = B_TRUE;
3761
3762 /*
3763 * Extract the checkpointed uberblock from the current MOS
3764 * and use this as the pool's uberblock from now on. If the
3765 * pool is imported as writeable we also write the checkpoint
3766 * uberblock to the labels, making the rewind permanent.
3767 */
3768 error = spa_ld_checkpoint_rewind(spa);
3769 if (error != 0)
3770 return (error);
3771
3772 /*
3773 * Redo the loading process process again with the
3774 * checkpointed uberblock.
3775 */
3776 spa_ld_prepare_for_reload(spa);
3777 spa_load_note(spa, "LOADING checkpointed uberblock");
3778 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3779 if (error != 0)
3780 return (error);
3781 }
3782
3783 /*
3784 * Retrieve the checkpoint txg if the pool has a checkpoint.
3785 */
3786 error = spa_ld_read_checkpoint_txg(spa);
3787 if (error != 0)
3788 return (error);
3789
3790 /*
3791 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3792 * from the pool and their contents were re-mapped to other vdevs. Note
3793 * that everything that we read before this step must have been
3794 * rewritten on concrete vdevs after the last device removal was
3795 * initiated. Otherwise we could be reading from indirect vdevs before
3796 * we have loaded their mappings.
3797 */
3798 error = spa_ld_open_indirect_vdev_metadata(spa);
3799 if (error != 0)
3800 return (error);
3801
3802 /*
3803 * Retrieve the full list of active features from the MOS and check if
3804 * they are all supported.
3805 */
3806 error = spa_ld_check_features(spa, &missing_feat_write);
3807 if (error != 0)
3808 return (error);
3809
3810 /*
3811 * Load several special directories from the MOS needed by the dsl_pool
3812 * layer.
3813 */
3814 error = spa_ld_load_special_directories(spa);
3815 if (error != 0)
3816 return (error);
3817
3818 /*
3819 * Retrieve pool properties from the MOS.
3820 */
3821 error = spa_ld_get_props(spa);
3822 if (error != 0)
3823 return (error);
3824
3825 /*
3826 * Retrieve the list of auxiliary devices - cache devices and spares -
3827 * and open them.
3828 */
3829 error = spa_ld_open_aux_vdevs(spa, type);
3830 if (error != 0)
3831 return (error);
3832
3833 /*
3834 * Load the metadata for all vdevs. Also check if unopenable devices
3835 * should be autoreplaced.
3836 */
3837 error = spa_ld_load_vdev_metadata(spa);
3838 if (error != 0)
3839 return (error);
3840
3841 error = spa_ld_load_dedup_tables(spa);
3842 if (error != 0)
3843 return (error);
3844
3845 /*
3846 * Verify the logs now to make sure we don't have any unexpected errors
3847 * when we claim log blocks later.
3848 */
3849 error = spa_ld_verify_logs(spa, type, ereport);
3850 if (error != 0)
3851 return (error);
3852
3853 if (missing_feat_write) {
3854 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3855
3856 /*
3857 * At this point, we know that we can open the pool in
3858 * read-only mode but not read-write mode. We now have enough
3859 * information and can return to userland.
3860 */
3861 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3862 ENOTSUP));
3863 }
3864
3865 /*
3866 * Traverse the last txgs to make sure the pool was left off in a safe
3867 * state. When performing an extreme rewind, we verify the whole pool,
3868 * which can take a very long time.
3869 */
3870 error = spa_ld_verify_pool_data(spa);
3871 if (error != 0)
3872 return (error);
3873
3874 /*
3875 * Calculate the deflated space for the pool. This must be done before
3876 * we write anything to the pool because we'd need to update the space
3877 * accounting using the deflated sizes.
3878 */
3879 spa_update_dspace(spa);
3880
3881 /*
3882 * We have now retrieved all the information we needed to open the
3883 * pool. If we are importing the pool in read-write mode, a few
3884 * additional steps must be performed to finish the import.
3885 */
3886 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3887 spa->spa_load_max_txg == UINT64_MAX)) {
3888 uint64_t config_cache_txg = spa->spa_config_txg;
3889
3890 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3891
3892 /*
3893 * In case of a checkpoint rewind, log the original txg
3894 * of the checkpointed uberblock.
3895 */
3896 if (checkpoint_rewind) {
3897 spa_history_log_internal(spa, "checkpoint rewind",
3898 NULL, "rewound state to txg=%llu",
3899 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3900 }
3901
3902 /*
3903 * Traverse the ZIL and claim all blocks.
3904 */
3905 spa_ld_claim_log_blocks(spa);
3906
3907 /*
3908 * Kick-off the syncing thread.
3909 */
3910 spa->spa_sync_on = B_TRUE;
3911 txg_sync_start(spa->spa_dsl_pool);
3912
3913 /*
3914 * Wait for all claims to sync. We sync up to the highest
3915 * claimed log block birth time so that claimed log blocks
3916 * don't appear to be from the future. spa_claim_max_txg
3917 * will have been set for us by ZIL traversal operations
3918 * performed above.
3919 */
3920 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3921
3922 /*
3923 * Check if we need to request an update of the config. On the
3924 * next sync, we would update the config stored in vdev labels
3925 * and the cachefile (by default /etc/zfs/zpool.cache).
3926 */
3927 spa_ld_check_for_config_update(spa, config_cache_txg,
3928 update_config_cache);
3929
3930 /*
3931 * Check all DTLs to see if anything needs resilvering.
3932 */
3933 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3934 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3935 spa_async_request(spa, SPA_ASYNC_RESILVER);
3936
3937 /*
3938 * Log the fact that we booted up (so that we can detect if
3939 * we rebooted in the middle of an operation).
3940 */
3941 spa_history_log_version(spa, "open");
3942
3943 /*
3944 * Delete any inconsistent datasets.
3945 */
3946 (void) dmu_objset_find(spa_name(spa),
3947 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3948
3949 /*
3950 * Clean up any stale temporary dataset userrefs.
3951 */
3952 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3953
3954 spa_restart_removal(spa);
3955
3956 spa_spawn_aux_threads(spa);
3957 }
3958
3959 spa_load_note(spa, "LOADED");
3960
3961 return (0);
3962}
3963
3964static int
3965spa_load_retry(spa_t *spa, spa_load_state_t state)
3966{
3967 int mode = spa->spa_mode;
3968
3969 spa_unload(spa);
3970 spa_deactivate(spa);
3971
3972 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3973
3974 spa_activate(spa, mode);
3975 spa_async_suspend(spa);
3976
3977 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3978 (u_longlong_t)spa->spa_load_max_txg);
3979
3980 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
3981}
3982
3983/*
3984 * If spa_load() fails this function will try loading prior txg's. If
3985 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3986 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3987 * function will not rewind the pool and will return the same error as
3988 * spa_load().
3989 */
3990static int
3991spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
3992 int rewind_flags)
3993{
3994 nvlist_t *loadinfo = NULL;
3995 nvlist_t *config = NULL;
3996 int load_error, rewind_error;
3997 uint64_t safe_rewind_txg;
3998 uint64_t min_txg;
3999
4000 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4001 spa->spa_load_max_txg = spa->spa_load_txg;
4002 spa_set_log_state(spa, SPA_LOG_CLEAR);
4003 } else {
4004 spa->spa_load_max_txg = max_request;
4005 if (max_request != UINT64_MAX)
4006 spa->spa_extreme_rewind = B_TRUE;
4007 }
4008
4009 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4010 if (load_error == 0)
4011 return (0);
4012 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4013 /*
4014 * When attempting checkpoint-rewind on a pool with no
4015 * checkpoint, we should not attempt to load uberblocks
4016 * from previous txgs when spa_load fails.
4017 */
4018 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4019 return (load_error);
4020 }
4021
4022 if (spa->spa_root_vdev != NULL)
4023 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4024
4025 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4026 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4027
4028 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4029 nvlist_free(config);
4030 return (load_error);
4031 }
4032
4033 if (state == SPA_LOAD_RECOVER) {
4034 /* Price of rolling back is discarding txgs, including log */
4035 spa_set_log_state(spa, SPA_LOG_CLEAR);
4036 } else {
4037 /*
4038 * If we aren't rolling back save the load info from our first
4039 * import attempt so that we can restore it after attempting
4040 * to rewind.
4041 */
4042 loadinfo = spa->spa_load_info;
4043 spa->spa_load_info = fnvlist_alloc();
4044 }
4045
4046 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4047 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4048 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4049 TXG_INITIAL : safe_rewind_txg;
4050
4051 /*
4052 * Continue as long as we're finding errors, we're still within
4053 * the acceptable rewind range, and we're still finding uberblocks
4054 */
4055 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4056 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4057 if (spa->spa_load_max_txg < safe_rewind_txg)
4058 spa->spa_extreme_rewind = B_TRUE;
4059 rewind_error = spa_load_retry(spa, state);
4060 }
4061
4062 spa->spa_extreme_rewind = B_FALSE;
4063 spa->spa_load_max_txg = UINT64_MAX;
4064
4065 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4066 spa_config_set(spa, config);
4067 else
4068 nvlist_free(config);
4069
4070 if (state == SPA_LOAD_RECOVER) {
4071 ASSERT3P(loadinfo, ==, NULL);
4072 return (rewind_error);
4073 } else {
4074 /* Store the rewind info as part of the initial load info */
4075 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4076 spa->spa_load_info);
4077
4078 /* Restore the initial load info */
4079 fnvlist_free(spa->spa_load_info);
4080 spa->spa_load_info = loadinfo;
4081
4082 return (load_error);
4083 }
4084}
4085
4086/*
4087 * Pool Open/Import
4088 *
4089 * The import case is identical to an open except that the configuration is sent
4090 * down from userland, instead of grabbed from the configuration cache. For the
4091 * case of an open, the pool configuration will exist in the
4092 * POOL_STATE_UNINITIALIZED state.
4093 *
4094 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4095 * the same time open the pool, without having to keep around the spa_t in some
4096 * ambiguous state.
4097 */
4098static int
4099spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4100 nvlist_t **config)
4101{
4102 spa_t *spa;
4103 spa_load_state_t state = SPA_LOAD_OPEN;
4104 int error;
4105 int locked = B_FALSE;
4106 int firstopen = B_FALSE;
4107
4108 *spapp = NULL;
4109
4110 /*
4111 * As disgusting as this is, we need to support recursive calls to this
4112 * function because dsl_dir_open() is called during spa_load(), and ends
4113 * up calling spa_open() again. The real fix is to figure out how to
4114 * avoid dsl_dir_open() calling this in the first place.
4115 */
4116 if (mutex_owner(&spa_namespace_lock) != curthread) {
4117 mutex_enter(&spa_namespace_lock);
4118 locked = B_TRUE;
4119 }
4120
4121 if ((spa = spa_lookup(pool)) == NULL) {
4122 if (locked)
4123 mutex_exit(&spa_namespace_lock);
4124 return (SET_ERROR(ENOENT));
4125 }
4126
4127 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
|
4128 zpool_rewind_policy_t policy;
| 4128 zpool_load_policy_t policy;
|
4129
4130 firstopen = B_TRUE;
4131
| 4129
4130 firstopen = B_TRUE;
4131
|
4132 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
| 4132 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
|
4133 &policy);
| 4133 &policy);
|
4134 if (policy.zrp_request & ZPOOL_DO_REWIND)
| 4134 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
|
4135 state = SPA_LOAD_RECOVER;
4136
4137 spa_activate(spa, spa_mode_global);
4138
4139 if (state != SPA_LOAD_RECOVER)
4140 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4141 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4142
4143 zfs_dbgmsg("spa_open_common: opening %s", pool);
| 4135 state = SPA_LOAD_RECOVER;
4136
4137 spa_activate(spa, spa_mode_global);
4138
4139 if (state != SPA_LOAD_RECOVER)
4140 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4141 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4142
4143 zfs_dbgmsg("spa_open_common: opening %s", pool);
|
4144 error = spa_load_best(spa, state, policy.zrp_txg,
4145 policy.zrp_request);
| 4144 error = spa_load_best(spa, state, policy.zlp_txg,
4145 policy.zlp_rewind);
|
4146
4147 if (error == EBADF) {
4148 /*
4149 * If vdev_validate() returns failure (indicated by
4150 * EBADF), it indicates that one of the vdevs indicates
4151 * that the pool has been exported or destroyed. If
4152 * this is the case, the config cache is out of sync and
4153 * we should remove the pool from the namespace.
4154 */
4155 spa_unload(spa);
4156 spa_deactivate(spa);
4157 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4158 spa_remove(spa);
4159 if (locked)
4160 mutex_exit(&spa_namespace_lock);
4161 return (SET_ERROR(ENOENT));
4162 }
4163
4164 if (error) {
4165 /*
4166 * We can't open the pool, but we still have useful
4167 * information: the state of each vdev after the
4168 * attempted vdev_open(). Return this to the user.
4169 */
4170 if (config != NULL && spa->spa_config) {
4171 VERIFY(nvlist_dup(spa->spa_config, config,
4172 KM_SLEEP) == 0);
4173 VERIFY(nvlist_add_nvlist(*config,
4174 ZPOOL_CONFIG_LOAD_INFO,
4175 spa->spa_load_info) == 0);
4176 }
4177 spa_unload(spa);
4178 spa_deactivate(spa);
4179 spa->spa_last_open_failed = error;
4180 if (locked)
4181 mutex_exit(&spa_namespace_lock);
4182 *spapp = NULL;
4183 return (error);
4184 }
4185 }
4186
4187 spa_open_ref(spa, tag);
4188
4189 if (config != NULL)
4190 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4191
4192 /*
4193 * If we've recovered the pool, pass back any information we
4194 * gathered while doing the load.
4195 */
4196 if (state == SPA_LOAD_RECOVER) {
4197 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4198 spa->spa_load_info) == 0);
4199 }
4200
4201 if (locked) {
4202 spa->spa_last_open_failed = 0;
4203 spa->spa_last_ubsync_txg = 0;
4204 spa->spa_load_txg = 0;
4205 mutex_exit(&spa_namespace_lock);
4206#ifdef __FreeBSD__
4207#ifdef _KERNEL
4208 if (firstopen)
4209 zvol_create_minors(spa->spa_name);
4210#endif
4211#endif
4212 }
4213
4214 *spapp = spa;
4215
4216 return (0);
4217}
4218
4219int
4220spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4221 nvlist_t **config)
4222{
4223 return (spa_open_common(name, spapp, tag, policy, config));
4224}
4225
4226int
4227spa_open(const char *name, spa_t **spapp, void *tag)
4228{
4229 return (spa_open_common(name, spapp, tag, NULL, NULL));
4230}
4231
4232/*
4233 * Lookup the given spa_t, incrementing the inject count in the process,
4234 * preventing it from being exported or destroyed.
4235 */
4236spa_t *
4237spa_inject_addref(char *name)
4238{
4239 spa_t *spa;
4240
4241 mutex_enter(&spa_namespace_lock);
4242 if ((spa = spa_lookup(name)) == NULL) {
4243 mutex_exit(&spa_namespace_lock);
4244 return (NULL);
4245 }
4246 spa->spa_inject_ref++;
4247 mutex_exit(&spa_namespace_lock);
4248
4249 return (spa);
4250}
4251
4252void
4253spa_inject_delref(spa_t *spa)
4254{
4255 mutex_enter(&spa_namespace_lock);
4256 spa->spa_inject_ref--;
4257 mutex_exit(&spa_namespace_lock);
4258}
4259
4260/*
4261 * Add spares device information to the nvlist.
4262 */
4263static void
4264spa_add_spares(spa_t *spa, nvlist_t *config)
4265{
4266 nvlist_t **spares;
4267 uint_t i, nspares;
4268 nvlist_t *nvroot;
4269 uint64_t guid;
4270 vdev_stat_t *vs;
4271 uint_t vsc;
4272 uint64_t pool;
4273
4274 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4275
4276 if (spa->spa_spares.sav_count == 0)
4277 return;
4278
4279 VERIFY(nvlist_lookup_nvlist(config,
4280 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4281 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4282 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4283 if (nspares != 0) {
4284 VERIFY(nvlist_add_nvlist_array(nvroot,
4285 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4286 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4287 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4288
4289 /*
4290 * Go through and find any spares which have since been
4291 * repurposed as an active spare. If this is the case, update
4292 * their status appropriately.
4293 */
4294 for (i = 0; i < nspares; i++) {
4295 VERIFY(nvlist_lookup_uint64(spares[i],
4296 ZPOOL_CONFIG_GUID, &guid) == 0);
4297 if (spa_spare_exists(guid, &pool, NULL) &&
4298 pool != 0ULL) {
4299 VERIFY(nvlist_lookup_uint64_array(
4300 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4301 (uint64_t **)&vs, &vsc) == 0);
4302 vs->vs_state = VDEV_STATE_CANT_OPEN;
4303 vs->vs_aux = VDEV_AUX_SPARED;
4304 }
4305 }
4306 }
4307}
4308
4309/*
4310 * Add l2cache device information to the nvlist, including vdev stats.
4311 */
4312static void
4313spa_add_l2cache(spa_t *spa, nvlist_t *config)
4314{
4315 nvlist_t **l2cache;
4316 uint_t i, j, nl2cache;
4317 nvlist_t *nvroot;
4318 uint64_t guid;
4319 vdev_t *vd;
4320 vdev_stat_t *vs;
4321 uint_t vsc;
4322
4323 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4324
4325 if (spa->spa_l2cache.sav_count == 0)
4326 return;
4327
4328 VERIFY(nvlist_lookup_nvlist(config,
4329 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4330 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4331 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4332 if (nl2cache != 0) {
4333 VERIFY(nvlist_add_nvlist_array(nvroot,
4334 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4335 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4336 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4337
4338 /*
4339 * Update level 2 cache device stats.
4340 */
4341
4342 for (i = 0; i < nl2cache; i++) {
4343 VERIFY(nvlist_lookup_uint64(l2cache[i],
4344 ZPOOL_CONFIG_GUID, &guid) == 0);
4345
4346 vd = NULL;
4347 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4348 if (guid ==
4349 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4350 vd = spa->spa_l2cache.sav_vdevs[j];
4351 break;
4352 }
4353 }
4354 ASSERT(vd != NULL);
4355
4356 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4357 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4358 == 0);
4359 vdev_get_stats(vd, vs);
4360 }
4361 }
4362}
4363
4364static void
4365spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4366{
4367 nvlist_t *features;
4368 zap_cursor_t zc;
4369 zap_attribute_t za;
4370
4371 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4372 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4373
4374 /* We may be unable to read features if pool is suspended. */
4375 if (spa_suspended(spa))
4376 goto out;
4377
4378 if (spa->spa_feat_for_read_obj != 0) {
4379 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4380 spa->spa_feat_for_read_obj);
4381 zap_cursor_retrieve(&zc, &za) == 0;
4382 zap_cursor_advance(&zc)) {
4383 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4384 za.za_num_integers == 1);
4385 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4386 za.za_first_integer));
4387 }
4388 zap_cursor_fini(&zc);
4389 }
4390
4391 if (spa->spa_feat_for_write_obj != 0) {
4392 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4393 spa->spa_feat_for_write_obj);
4394 zap_cursor_retrieve(&zc, &za) == 0;
4395 zap_cursor_advance(&zc)) {
4396 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4397 za.za_num_integers == 1);
4398 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4399 za.za_first_integer));
4400 }
4401 zap_cursor_fini(&zc);
4402 }
4403
4404out:
4405 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4406 features) == 0);
4407 nvlist_free(features);
4408}
4409
4410int
4411spa_get_stats(const char *name, nvlist_t **config,
4412 char *altroot, size_t buflen)
4413{
4414 int error;
4415 spa_t *spa;
4416
4417 *config = NULL;
4418 error = spa_open_common(name, &spa, FTAG, NULL, config);
4419
4420 if (spa != NULL) {
4421 /*
4422 * This still leaves a window of inconsistency where the spares
4423 * or l2cache devices could change and the config would be
4424 * self-inconsistent.
4425 */
4426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4427
4428 if (*config != NULL) {
4429 uint64_t loadtimes[2];
4430
4431 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4432 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4433 VERIFY(nvlist_add_uint64_array(*config,
4434 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4435
4436 VERIFY(nvlist_add_uint64(*config,
4437 ZPOOL_CONFIG_ERRCOUNT,
4438 spa_get_errlog_size(spa)) == 0);
4439
4440 if (spa_suspended(spa))
4441 VERIFY(nvlist_add_uint64(*config,
4442 ZPOOL_CONFIG_SUSPENDED,
4443 spa->spa_failmode) == 0);
4444
4445 spa_add_spares(spa, *config);
4446 spa_add_l2cache(spa, *config);
4447 spa_add_feature_stats(spa, *config);
4448 }
4449 }
4450
4451 /*
4452 * We want to get the alternate root even for faulted pools, so we cheat
4453 * and call spa_lookup() directly.
4454 */
4455 if (altroot) {
4456 if (spa == NULL) {
4457 mutex_enter(&spa_namespace_lock);
4458 spa = spa_lookup(name);
4459 if (spa)
4460 spa_altroot(spa, altroot, buflen);
4461 else
4462 altroot[0] = '\0';
4463 spa = NULL;
4464 mutex_exit(&spa_namespace_lock);
4465 } else {
4466 spa_altroot(spa, altroot, buflen);
4467 }
4468 }
4469
4470 if (spa != NULL) {
4471 spa_config_exit(spa, SCL_CONFIG, FTAG);
4472 spa_close(spa, FTAG);
4473 }
4474
4475 return (error);
4476}
4477
4478/*
4479 * Validate that the auxiliary device array is well formed. We must have an
4480 * array of nvlists, each which describes a valid leaf vdev. If this is an
4481 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4482 * specified, as long as they are well-formed.
4483 */
4484static int
4485spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4486 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4487 vdev_labeltype_t label)
4488{
4489 nvlist_t **dev;
4490 uint_t i, ndev;
4491 vdev_t *vd;
4492 int error;
4493
4494 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4495
4496 /*
4497 * It's acceptable to have no devs specified.
4498 */
4499 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4500 return (0);
4501
4502 if (ndev == 0)
4503 return (SET_ERROR(EINVAL));
4504
4505 /*
4506 * Make sure the pool is formatted with a version that supports this
4507 * device type.
4508 */
4509 if (spa_version(spa) < version)
4510 return (SET_ERROR(ENOTSUP));
4511
4512 /*
4513 * Set the pending device list so we correctly handle device in-use
4514 * checking.
4515 */
4516 sav->sav_pending = dev;
4517 sav->sav_npending = ndev;
4518
4519 for (i = 0; i < ndev; i++) {
4520 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4521 mode)) != 0)
4522 goto out;
4523
4524 if (!vd->vdev_ops->vdev_op_leaf) {
4525 vdev_free(vd);
4526 error = SET_ERROR(EINVAL);
4527 goto out;
4528 }
4529
4530 /*
4531 * The L2ARC currently only supports disk devices in
4532 * kernel context. For user-level testing, we allow it.
4533 */
4534#ifdef _KERNEL
4535 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4536 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4537 error = SET_ERROR(ENOTBLK);
4538 vdev_free(vd);
4539 goto out;
4540 }
4541#endif
4542 vd->vdev_top = vd;
4543
4544 if ((error = vdev_open(vd)) == 0 &&
4545 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4546 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4547 vd->vdev_guid) == 0);
4548 }
4549
4550 vdev_free(vd);
4551
4552 if (error &&
4553 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4554 goto out;
4555 else
4556 error = 0;
4557 }
4558
4559out:
4560 sav->sav_pending = NULL;
4561 sav->sav_npending = 0;
4562 return (error);
4563}
4564
4565static int
4566spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4567{
4568 int error;
4569
4570 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4571
4572 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4573 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4574 VDEV_LABEL_SPARE)) != 0) {
4575 return (error);
4576 }
4577
4578 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4579 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4580 VDEV_LABEL_L2CACHE));
4581}
4582
4583static void
4584spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4585 const char *config)
4586{
4587 int i;
4588
4589 if (sav->sav_config != NULL) {
4590 nvlist_t **olddevs;
4591 uint_t oldndevs;
4592 nvlist_t **newdevs;
4593
4594 /*
4595 * Generate new dev list by concatentating with the
4596 * current dev list.
4597 */
4598 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4599 &olddevs, &oldndevs) == 0);
4600
4601 newdevs = kmem_alloc(sizeof (void *) *
4602 (ndevs + oldndevs), KM_SLEEP);
4603 for (i = 0; i < oldndevs; i++)
4604 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4605 KM_SLEEP) == 0);
4606 for (i = 0; i < ndevs; i++)
4607 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4608 KM_SLEEP) == 0);
4609
4610 VERIFY(nvlist_remove(sav->sav_config, config,
4611 DATA_TYPE_NVLIST_ARRAY) == 0);
4612
4613 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4614 config, newdevs, ndevs + oldndevs) == 0);
4615 for (i = 0; i < oldndevs + ndevs; i++)
4616 nvlist_free(newdevs[i]);
4617 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4618 } else {
4619 /*
4620 * Generate a new dev list.
4621 */
4622 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4623 KM_SLEEP) == 0);
4624 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4625 devs, ndevs) == 0);
4626 }
4627}
4628
4629/*
4630 * Stop and drop level 2 ARC devices
4631 */
4632void
4633spa_l2cache_drop(spa_t *spa)
4634{
4635 vdev_t *vd;
4636 int i;
4637 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4638
4639 for (i = 0; i < sav->sav_count; i++) {
4640 uint64_t pool;
4641
4642 vd = sav->sav_vdevs[i];
4643 ASSERT(vd != NULL);
4644
4645 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4646 pool != 0ULL && l2arc_vdev_present(vd))
4647 l2arc_remove_vdev(vd);
4648 }
4649}
4650
4651/*
4652 * Pool Creation
4653 */
4654int
4655spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4656 nvlist_t *zplprops)
4657{
4658 spa_t *spa;
4659 char *altroot = NULL;
4660 vdev_t *rvd;
4661 dsl_pool_t *dp;
4662 dmu_tx_t *tx;
4663 int error = 0;
4664 uint64_t txg = TXG_INITIAL;
4665 nvlist_t **spares, **l2cache;
4666 uint_t nspares, nl2cache;
4667 uint64_t version, obj;
4668 boolean_t has_features;
4669
4670 /*
4671 * If this pool already exists, return failure.
4672 */
4673 mutex_enter(&spa_namespace_lock);
4674 if (spa_lookup(pool) != NULL) {
4675 mutex_exit(&spa_namespace_lock);
4676 return (SET_ERROR(EEXIST));
4677 }
4678
4679 /*
4680 * Allocate a new spa_t structure.
4681 */
4682 (void) nvlist_lookup_string(props,
4683 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4684 spa = spa_add(pool, NULL, altroot);
4685 spa_activate(spa, spa_mode_global);
4686
4687 if (props && (error = spa_prop_validate(spa, props))) {
4688 spa_deactivate(spa);
4689 spa_remove(spa);
4690 mutex_exit(&spa_namespace_lock);
4691 return (error);
4692 }
4693
4694 has_features = B_FALSE;
4695 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4696 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4697 if (zpool_prop_feature(nvpair_name(elem)))
4698 has_features = B_TRUE;
4699 }
4700
4701 if (has_features || nvlist_lookup_uint64(props,
4702 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4703 version = SPA_VERSION;
4704 }
4705 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4706
4707 spa->spa_first_txg = txg;
4708 spa->spa_uberblock.ub_txg = txg - 1;
4709 spa->spa_uberblock.ub_version = version;
4710 spa->spa_ubsync = spa->spa_uberblock;
4711 spa->spa_load_state = SPA_LOAD_CREATE;
4712 spa->spa_removing_phys.sr_state = DSS_NONE;
4713 spa->spa_removing_phys.sr_removing_vdev = -1;
4714 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4715
4716 /*
4717 * Create "The Godfather" zio to hold all async IOs
4718 */
4719 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4720 KM_SLEEP);
4721 for (int i = 0; i < max_ncpus; i++) {
4722 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4723 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4724 ZIO_FLAG_GODFATHER);
4725 }
4726
4727 /*
4728 * Create the root vdev.
4729 */
4730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4731
4732 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4733
4734 ASSERT(error != 0 || rvd != NULL);
4735 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4736
4737 if (error == 0 && !zfs_allocatable_devs(nvroot))
4738 error = SET_ERROR(EINVAL);
4739
4740 if (error == 0 &&
4741 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4742 (error = spa_validate_aux(spa, nvroot, txg,
4743 VDEV_ALLOC_ADD)) == 0) {
4744 for (int c = 0; c < rvd->vdev_children; c++) {
4745 vdev_ashift_optimize(rvd->vdev_child[c]);
4746 vdev_metaslab_set_size(rvd->vdev_child[c]);
4747 vdev_expand(rvd->vdev_child[c], txg);
4748 }
4749 }
4750
4751 spa_config_exit(spa, SCL_ALL, FTAG);
4752
4753 if (error != 0) {
4754 spa_unload(spa);
4755 spa_deactivate(spa);
4756 spa_remove(spa);
4757 mutex_exit(&spa_namespace_lock);
4758 return (error);
4759 }
4760
4761 /*
4762 * Get the list of spares, if specified.
4763 */
4764 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4765 &spares, &nspares) == 0) {
4766 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4767 KM_SLEEP) == 0);
4768 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4769 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4770 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4771 spa_load_spares(spa);
4772 spa_config_exit(spa, SCL_ALL, FTAG);
4773 spa->spa_spares.sav_sync = B_TRUE;
4774 }
4775
4776 /*
4777 * Get the list of level 2 cache devices, if specified.
4778 */
4779 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4780 &l2cache, &nl2cache) == 0) {
4781 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4782 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4783 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4784 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4786 spa_load_l2cache(spa);
4787 spa_config_exit(spa, SCL_ALL, FTAG);
4788 spa->spa_l2cache.sav_sync = B_TRUE;
4789 }
4790
4791 spa->spa_is_initializing = B_TRUE;
4792 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4793 spa->spa_meta_objset = dp->dp_meta_objset;
4794 spa->spa_is_initializing = B_FALSE;
4795
4796 /*
4797 * Create DDTs (dedup tables).
4798 */
4799 ddt_create(spa);
4800
4801 spa_update_dspace(spa);
4802
4803 tx = dmu_tx_create_assigned(dp, txg);
4804
4805 /*
4806 * Create the pool config object.
4807 */
4808 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4809 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4810 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4811
4812 if (zap_add(spa->spa_meta_objset,
4813 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4814 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4815 cmn_err(CE_PANIC, "failed to add pool config");
4816 }
4817
4818 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4819 spa_feature_create_zap_objects(spa, tx);
4820
4821 if (zap_add(spa->spa_meta_objset,
4822 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4823 sizeof (uint64_t), 1, &version, tx) != 0) {
4824 cmn_err(CE_PANIC, "failed to add pool version");
4825 }
4826
4827 /* Newly created pools with the right version are always deflated. */
4828 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4829 spa->spa_deflate = TRUE;
4830 if (zap_add(spa->spa_meta_objset,
4831 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4832 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4833 cmn_err(CE_PANIC, "failed to add deflate");
4834 }
4835 }
4836
4837 /*
4838 * Create the deferred-free bpobj. Turn off compression
4839 * because sync-to-convergence takes longer if the blocksize
4840 * keeps changing.
4841 */
4842 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4843 dmu_object_set_compress(spa->spa_meta_objset, obj,
4844 ZIO_COMPRESS_OFF, tx);
4845 if (zap_add(spa->spa_meta_objset,
4846 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4847 sizeof (uint64_t), 1, &obj, tx) != 0) {
4848 cmn_err(CE_PANIC, "failed to add bpobj");
4849 }
4850 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4851 spa->spa_meta_objset, obj));
4852
4853 /*
4854 * Create the pool's history object.
4855 */
4856 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4857 spa_history_create_obj(spa, tx);
4858
4859 /*
4860 * Generate some random noise for salted checksums to operate on.
4861 */
4862 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4863 sizeof (spa->spa_cksum_salt.zcs_bytes));
4864
4865 /*
4866 * Set pool properties.
4867 */
4868 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4869 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4870 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4871 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4872
4873 if (props != NULL) {
4874 spa_configfile_set(spa, props, B_FALSE);
4875 spa_sync_props(props, tx);
4876 }
4877
4878 dmu_tx_commit(tx);
4879
4880 spa->spa_sync_on = B_TRUE;
4881 txg_sync_start(spa->spa_dsl_pool);
4882
4883 /*
4884 * We explicitly wait for the first transaction to complete so that our
4885 * bean counters are appropriately updated.
4886 */
4887 txg_wait_synced(spa->spa_dsl_pool, txg);
4888
4889 spa_spawn_aux_threads(spa);
4890
4891 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4892 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4893
4894 spa_history_log_version(spa, "create");
4895
4896 /*
4897 * Don't count references from objsets that are already closed
4898 * and are making their way through the eviction process.
4899 */
4900 spa_evicting_os_wait(spa);
4901 spa->spa_minref = refcount_count(&spa->spa_refcount);
4902 spa->spa_load_state = SPA_LOAD_NONE;
4903
4904 mutex_exit(&spa_namespace_lock);
4905
4906 return (0);
4907}
4908
4909#ifdef _KERNEL
4910#ifdef illumos
4911/*
4912 * Get the root pool information from the root disk, then import the root pool
4913 * during the system boot up time.
4914 */
4915extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4916
4917static nvlist_t *
4918spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4919{
4920 nvlist_t *config;
4921 nvlist_t *nvtop, *nvroot;
4922 uint64_t pgid;
4923
4924 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
4925 return (NULL);
4926
4927 /*
4928 * Add this top-level vdev to the child array.
4929 */
4930 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4931 &nvtop) == 0);
4932 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4933 &pgid) == 0);
4934 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
4935
4936 /*
4937 * Put this pool's top-level vdevs into a root vdev.
4938 */
4939 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4940 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4941 VDEV_TYPE_ROOT) == 0);
4942 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4943 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4944 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4945 &nvtop, 1) == 0);
4946
4947 /*
4948 * Replace the existing vdev_tree with the new root vdev in
4949 * this pool's configuration (remove the old, add the new).
4950 */
4951 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4952 nvlist_free(nvroot);
4953 return (config);
4954}
4955
4956/*
4957 * Walk the vdev tree and see if we can find a device with "better"
4958 * configuration. A configuration is "better" if the label on that
4959 * device has a more recent txg.
4960 */
4961static void
4962spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
4963{
4964 for (int c = 0; c < vd->vdev_children; c++)
4965 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
4966
4967 if (vd->vdev_ops->vdev_op_leaf) {
4968 nvlist_t *label;
4969 uint64_t label_txg;
4970
4971 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4972 &label) != 0)
4973 return;
4974
4975 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4976 &label_txg) == 0);
4977
4978 /*
4979 * Do we have a better boot device?
4980 */
4981 if (label_txg > *txg) {
4982 *txg = label_txg;
4983 *avd = vd;
4984 }
4985 nvlist_free(label);
4986 }
4987}
4988
4989/*
4990 * Import a root pool.
4991 *
4992 * For x86. devpath_list will consist of devid and/or physpath name of
4993 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4994 * The GRUB "findroot" command will return the vdev we should boot.
4995 *
4996 * For Sparc, devpath_list consists the physpath name of the booting device
4997 * no matter the rootpool is a single device pool or a mirrored pool.
4998 * e.g.
4999 * "/pci@1f,0/ide@d/disk@0,0:a"
5000 */
5001int
5002spa_import_rootpool(char *devpath, char *devid)
5003{
5004 spa_t *spa;
5005 vdev_t *rvd, *bvd, *avd = NULL;
5006 nvlist_t *config, *nvtop;
5007 uint64_t guid, txg;
5008 char *pname;
5009 int error;
5010
5011 /*
5012 * Read the label from the boot device and generate a configuration.
5013 */
5014 config = spa_generate_rootconf(devpath, devid, &guid);
5015#if defined(_OBP) && defined(_KERNEL)
5016 if (config == NULL) {
5017 if (strstr(devpath, "/iscsi/ssd") != NULL) {
5018 /* iscsi boot */
5019 get_iscsi_bootpath_phy(devpath);
5020 config = spa_generate_rootconf(devpath, devid, &guid);
5021 }
5022 }
5023#endif
5024 if (config == NULL) {
5025 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5026 devpath);
5027 return (SET_ERROR(EIO));
5028 }
5029
5030 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5031 &pname) == 0);
5032 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5033
5034 mutex_enter(&spa_namespace_lock);
5035 if ((spa = spa_lookup(pname)) != NULL) {
5036 /*
5037 * Remove the existing root pool from the namespace so that we
5038 * can replace it with the correct config we just read in.
5039 */
5040 spa_remove(spa);
5041 }
5042
5043 spa = spa_add(pname, config, NULL);
5044 spa->spa_is_root = B_TRUE;
5045 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5046 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5047 &spa->spa_ubsync.ub_version) != 0)
5048 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5049
5050 /*
5051 * Build up a vdev tree based on the boot device's label config.
5052 */
5053 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5054 &nvtop) == 0);
5055 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5056 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5057 VDEV_ALLOC_ROOTPOOL);
5058 spa_config_exit(spa, SCL_ALL, FTAG);
5059 if (error) {
5060 mutex_exit(&spa_namespace_lock);
5061 nvlist_free(config);
5062 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5063 pname);
5064 return (error);
5065 }
5066
5067 /*
5068 * Get the boot vdev.
5069 */
5070 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5071 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5072 (u_longlong_t)guid);
5073 error = SET_ERROR(ENOENT);
5074 goto out;
5075 }
5076
5077 /*
5078 * Determine if there is a better boot device.
5079 */
5080 avd = bvd;
5081 spa_alt_rootvdev(rvd, &avd, &txg);
5082 if (avd != bvd) {
5083 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5084 "try booting from '%s'", avd->vdev_path);
5085 error = SET_ERROR(EINVAL);
5086 goto out;
5087 }
5088
5089 /*
5090 * If the boot device is part of a spare vdev then ensure that
5091 * we're booting off the active spare.
5092 */
5093 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5094 !bvd->vdev_isspare) {
5095 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5096 "try booting from '%s'",
5097 bvd->vdev_parent->
5098 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5099 error = SET_ERROR(EINVAL);
5100 goto out;
5101 }
5102
5103 error = 0;
5104out:
5105 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5106 vdev_free(rvd);
5107 spa_config_exit(spa, SCL_ALL, FTAG);
5108 mutex_exit(&spa_namespace_lock);
5109
5110 nvlist_free(config);
5111 return (error);
5112}
5113
5114#else /* !illumos */
5115
5116extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
5117 uint64_t *count);
5118
5119static nvlist_t *
5120spa_generate_rootconf(const char *name)
5121{
5122 nvlist_t **configs, **tops;
5123 nvlist_t *config;
5124 nvlist_t *best_cfg, *nvtop, *nvroot;
5125 uint64_t *holes;
5126 uint64_t best_txg;
5127 uint64_t nchildren;
5128 uint64_t pgid;
5129 uint64_t count;
5130 uint64_t i;
5131 uint_t nholes;
5132
5133 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
5134 return (NULL);
5135
5136 ASSERT3U(count, !=, 0);
5137 best_txg = 0;
5138 for (i = 0; i < count; i++) {
5139 uint64_t txg;
5140
5141 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
5142 &txg) == 0);
5143 if (txg > best_txg) {
5144 best_txg = txg;
5145 best_cfg = configs[i];
5146 }
5147 }
5148
5149 nchildren = 1;
5150 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
5151 holes = NULL;
5152 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
5153 &holes, &nholes);
5154
5155 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
5156 for (i = 0; i < nchildren; i++) {
5157 if (i >= count)
5158 break;
5159 if (configs[i] == NULL)
5160 continue;
5161 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
5162 &nvtop) == 0);
5163 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
5164 }
5165 for (i = 0; holes != NULL && i < nholes; i++) {
5166 if (i >= nchildren)
5167 continue;
5168 if (tops[holes[i]] != NULL)
5169 continue;
5170 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
5171 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
5172 VDEV_TYPE_HOLE) == 0);
5173 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
5174 holes[i]) == 0);
5175 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
5176 0) == 0);
5177 }
5178 for (i = 0; i < nchildren; i++) {
5179 if (tops[i] != NULL)
5180 continue;
5181 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
5182 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
5183 VDEV_TYPE_MISSING) == 0);
5184 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
5185 i) == 0);
5186 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
5187 0) == 0);
5188 }
5189
5190 /*
5191 * Create pool config based on the best vdev config.
5192 */
5193 nvlist_dup(best_cfg, &config, KM_SLEEP);
5194
5195 /*
5196 * Put this pool's top-level vdevs into a root vdev.
5197 */
5198 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5199 &pgid) == 0);
5200 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5201 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5202 VDEV_TYPE_ROOT) == 0);
5203 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5204 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5205 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5206 tops, nchildren) == 0);
5207
5208 /*
5209 * Replace the existing vdev_tree with the new root vdev in
5210 * this pool's configuration (remove the old, add the new).
5211 */
5212 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5213
5214 /*
5215 * Drop vdev config elements that should not be present at pool level.
5216 */
5217 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
5218 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
5219
5220 for (i = 0; i < count; i++)
5221 nvlist_free(configs[i]);
5222 kmem_free(configs, count * sizeof(void *));
5223 for (i = 0; i < nchildren; i++)
5224 nvlist_free(tops[i]);
5225 kmem_free(tops, nchildren * sizeof(void *));
5226 nvlist_free(nvroot);
5227 return (config);
5228}
5229
5230int
5231spa_import_rootpool(const char *name)
5232{
5233 spa_t *spa;
5234 vdev_t *rvd, *bvd, *avd = NULL;
5235 nvlist_t *config, *nvtop;
5236 uint64_t txg;
5237 char *pname;
5238 int error;
5239
5240 /*
5241 * Read the label from the boot device and generate a configuration.
5242 */
5243 config = spa_generate_rootconf(name);
5244
5245 mutex_enter(&spa_namespace_lock);
5246 if (config != NULL) {
5247 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5248 &pname) == 0 && strcmp(name, pname) == 0);
5249 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
5250 == 0);
5251
5252 if ((spa = spa_lookup(pname)) != NULL) {
5253 /*
5254 * The pool could already be imported,
5255 * e.g., after reboot -r.
5256 */
5257 if (spa->spa_state == POOL_STATE_ACTIVE) {
5258 mutex_exit(&spa_namespace_lock);
5259 nvlist_free(config);
5260 return (0);
5261 }
5262
5263 /*
5264 * Remove the existing root pool from the namespace so
5265 * that we can replace it with the correct config
5266 * we just read in.
5267 */
5268 spa_remove(spa);
5269 }
5270 spa = spa_add(pname, config, NULL);
5271
5272 /*
5273 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
5274 * via spa_version().
5275 */
5276 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5277 &spa->spa_ubsync.ub_version) != 0)
5278 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5279 } else if ((spa = spa_lookup(name)) == NULL) {
5280 mutex_exit(&spa_namespace_lock);
5281 nvlist_free(config);
5282 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
5283 name);
5284 return (EIO);
5285 } else {
5286 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
5287 }
5288 spa->spa_is_root = B_TRUE;
5289 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5290
5291 /*
5292 * Build up a vdev tree based on the boot device's label config.
5293 */
5294 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5295 &nvtop) == 0);
5296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5297 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5298 VDEV_ALLOC_ROOTPOOL);
5299 spa_config_exit(spa, SCL_ALL, FTAG);
5300 if (error) {
5301 mutex_exit(&spa_namespace_lock);
5302 nvlist_free(config);
5303 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5304 pname);
5305 return (error);
5306 }
5307
5308 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5309 vdev_free(rvd);
5310 spa_config_exit(spa, SCL_ALL, FTAG);
5311 mutex_exit(&spa_namespace_lock);
5312
5313 nvlist_free(config);
5314 return (0);
5315}
5316
5317#endif /* illumos */
5318#endif /* _KERNEL */
5319
5320/*
5321 * Import a non-root pool into the system.
5322 */
5323int
5324spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5325{
5326 spa_t *spa;
5327 char *altroot = NULL;
5328 spa_load_state_t state = SPA_LOAD_IMPORT;
| 4146
4147 if (error == EBADF) {
4148 /*
4149 * If vdev_validate() returns failure (indicated by
4150 * EBADF), it indicates that one of the vdevs indicates
4151 * that the pool has been exported or destroyed. If
4152 * this is the case, the config cache is out of sync and
4153 * we should remove the pool from the namespace.
4154 */
4155 spa_unload(spa);
4156 spa_deactivate(spa);
4157 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4158 spa_remove(spa);
4159 if (locked)
4160 mutex_exit(&spa_namespace_lock);
4161 return (SET_ERROR(ENOENT));
4162 }
4163
4164 if (error) {
4165 /*
4166 * We can't open the pool, but we still have useful
4167 * information: the state of each vdev after the
4168 * attempted vdev_open(). Return this to the user.
4169 */
4170 if (config != NULL && spa->spa_config) {
4171 VERIFY(nvlist_dup(spa->spa_config, config,
4172 KM_SLEEP) == 0);
4173 VERIFY(nvlist_add_nvlist(*config,
4174 ZPOOL_CONFIG_LOAD_INFO,
4175 spa->spa_load_info) == 0);
4176 }
4177 spa_unload(spa);
4178 spa_deactivate(spa);
4179 spa->spa_last_open_failed = error;
4180 if (locked)
4181 mutex_exit(&spa_namespace_lock);
4182 *spapp = NULL;
4183 return (error);
4184 }
4185 }
4186
4187 spa_open_ref(spa, tag);
4188
4189 if (config != NULL)
4190 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4191
4192 /*
4193 * If we've recovered the pool, pass back any information we
4194 * gathered while doing the load.
4195 */
4196 if (state == SPA_LOAD_RECOVER) {
4197 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4198 spa->spa_load_info) == 0);
4199 }
4200
4201 if (locked) {
4202 spa->spa_last_open_failed = 0;
4203 spa->spa_last_ubsync_txg = 0;
4204 spa->spa_load_txg = 0;
4205 mutex_exit(&spa_namespace_lock);
4206#ifdef __FreeBSD__
4207#ifdef _KERNEL
4208 if (firstopen)
4209 zvol_create_minors(spa->spa_name);
4210#endif
4211#endif
4212 }
4213
4214 *spapp = spa;
4215
4216 return (0);
4217}
4218
4219int
4220spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4221 nvlist_t **config)
4222{
4223 return (spa_open_common(name, spapp, tag, policy, config));
4224}
4225
4226int
4227spa_open(const char *name, spa_t **spapp, void *tag)
4228{
4229 return (spa_open_common(name, spapp, tag, NULL, NULL));
4230}
4231
4232/*
4233 * Lookup the given spa_t, incrementing the inject count in the process,
4234 * preventing it from being exported or destroyed.
4235 */
4236spa_t *
4237spa_inject_addref(char *name)
4238{
4239 spa_t *spa;
4240
4241 mutex_enter(&spa_namespace_lock);
4242 if ((spa = spa_lookup(name)) == NULL) {
4243 mutex_exit(&spa_namespace_lock);
4244 return (NULL);
4245 }
4246 spa->spa_inject_ref++;
4247 mutex_exit(&spa_namespace_lock);
4248
4249 return (spa);
4250}
4251
4252void
4253spa_inject_delref(spa_t *spa)
4254{
4255 mutex_enter(&spa_namespace_lock);
4256 spa->spa_inject_ref--;
4257 mutex_exit(&spa_namespace_lock);
4258}
4259
4260/*
4261 * Add spares device information to the nvlist.
4262 */
4263static void
4264spa_add_spares(spa_t *spa, nvlist_t *config)
4265{
4266 nvlist_t **spares;
4267 uint_t i, nspares;
4268 nvlist_t *nvroot;
4269 uint64_t guid;
4270 vdev_stat_t *vs;
4271 uint_t vsc;
4272 uint64_t pool;
4273
4274 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4275
4276 if (spa->spa_spares.sav_count == 0)
4277 return;
4278
4279 VERIFY(nvlist_lookup_nvlist(config,
4280 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4281 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4282 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4283 if (nspares != 0) {
4284 VERIFY(nvlist_add_nvlist_array(nvroot,
4285 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4286 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4287 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4288
4289 /*
4290 * Go through and find any spares which have since been
4291 * repurposed as an active spare. If this is the case, update
4292 * their status appropriately.
4293 */
4294 for (i = 0; i < nspares; i++) {
4295 VERIFY(nvlist_lookup_uint64(spares[i],
4296 ZPOOL_CONFIG_GUID, &guid) == 0);
4297 if (spa_spare_exists(guid, &pool, NULL) &&
4298 pool != 0ULL) {
4299 VERIFY(nvlist_lookup_uint64_array(
4300 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4301 (uint64_t **)&vs, &vsc) == 0);
4302 vs->vs_state = VDEV_STATE_CANT_OPEN;
4303 vs->vs_aux = VDEV_AUX_SPARED;
4304 }
4305 }
4306 }
4307}
4308
4309/*
4310 * Add l2cache device information to the nvlist, including vdev stats.
4311 */
4312static void
4313spa_add_l2cache(spa_t *spa, nvlist_t *config)
4314{
4315 nvlist_t **l2cache;
4316 uint_t i, j, nl2cache;
4317 nvlist_t *nvroot;
4318 uint64_t guid;
4319 vdev_t *vd;
4320 vdev_stat_t *vs;
4321 uint_t vsc;
4322
4323 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4324
4325 if (spa->spa_l2cache.sav_count == 0)
4326 return;
4327
4328 VERIFY(nvlist_lookup_nvlist(config,
4329 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4330 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4331 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4332 if (nl2cache != 0) {
4333 VERIFY(nvlist_add_nvlist_array(nvroot,
4334 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4335 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4336 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4337
4338 /*
4339 * Update level 2 cache device stats.
4340 */
4341
4342 for (i = 0; i < nl2cache; i++) {
4343 VERIFY(nvlist_lookup_uint64(l2cache[i],
4344 ZPOOL_CONFIG_GUID, &guid) == 0);
4345
4346 vd = NULL;
4347 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4348 if (guid ==
4349 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4350 vd = spa->spa_l2cache.sav_vdevs[j];
4351 break;
4352 }
4353 }
4354 ASSERT(vd != NULL);
4355
4356 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4357 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4358 == 0);
4359 vdev_get_stats(vd, vs);
4360 }
4361 }
4362}
4363
4364static void
4365spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4366{
4367 nvlist_t *features;
4368 zap_cursor_t zc;
4369 zap_attribute_t za;
4370
4371 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4372 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4373
4374 /* We may be unable to read features if pool is suspended. */
4375 if (spa_suspended(spa))
4376 goto out;
4377
4378 if (spa->spa_feat_for_read_obj != 0) {
4379 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4380 spa->spa_feat_for_read_obj);
4381 zap_cursor_retrieve(&zc, &za) == 0;
4382 zap_cursor_advance(&zc)) {
4383 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4384 za.za_num_integers == 1);
4385 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4386 za.za_first_integer));
4387 }
4388 zap_cursor_fini(&zc);
4389 }
4390
4391 if (spa->spa_feat_for_write_obj != 0) {
4392 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4393 spa->spa_feat_for_write_obj);
4394 zap_cursor_retrieve(&zc, &za) == 0;
4395 zap_cursor_advance(&zc)) {
4396 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4397 za.za_num_integers == 1);
4398 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4399 za.za_first_integer));
4400 }
4401 zap_cursor_fini(&zc);
4402 }
4403
4404out:
4405 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4406 features) == 0);
4407 nvlist_free(features);
4408}
4409
4410int
4411spa_get_stats(const char *name, nvlist_t **config,
4412 char *altroot, size_t buflen)
4413{
4414 int error;
4415 spa_t *spa;
4416
4417 *config = NULL;
4418 error = spa_open_common(name, &spa, FTAG, NULL, config);
4419
4420 if (spa != NULL) {
4421 /*
4422 * This still leaves a window of inconsistency where the spares
4423 * or l2cache devices could change and the config would be
4424 * self-inconsistent.
4425 */
4426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4427
4428 if (*config != NULL) {
4429 uint64_t loadtimes[2];
4430
4431 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4432 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4433 VERIFY(nvlist_add_uint64_array(*config,
4434 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4435
4436 VERIFY(nvlist_add_uint64(*config,
4437 ZPOOL_CONFIG_ERRCOUNT,
4438 spa_get_errlog_size(spa)) == 0);
4439
4440 if (spa_suspended(spa))
4441 VERIFY(nvlist_add_uint64(*config,
4442 ZPOOL_CONFIG_SUSPENDED,
4443 spa->spa_failmode) == 0);
4444
4445 spa_add_spares(spa, *config);
4446 spa_add_l2cache(spa, *config);
4447 spa_add_feature_stats(spa, *config);
4448 }
4449 }
4450
4451 /*
4452 * We want to get the alternate root even for faulted pools, so we cheat
4453 * and call spa_lookup() directly.
4454 */
4455 if (altroot) {
4456 if (spa == NULL) {
4457 mutex_enter(&spa_namespace_lock);
4458 spa = spa_lookup(name);
4459 if (spa)
4460 spa_altroot(spa, altroot, buflen);
4461 else
4462 altroot[0] = '\0';
4463 spa = NULL;
4464 mutex_exit(&spa_namespace_lock);
4465 } else {
4466 spa_altroot(spa, altroot, buflen);
4467 }
4468 }
4469
4470 if (spa != NULL) {
4471 spa_config_exit(spa, SCL_CONFIG, FTAG);
4472 spa_close(spa, FTAG);
4473 }
4474
4475 return (error);
4476}
4477
4478/*
4479 * Validate that the auxiliary device array is well formed. We must have an
4480 * array of nvlists, each which describes a valid leaf vdev. If this is an
4481 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4482 * specified, as long as they are well-formed.
4483 */
4484static int
4485spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4486 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4487 vdev_labeltype_t label)
4488{
4489 nvlist_t **dev;
4490 uint_t i, ndev;
4491 vdev_t *vd;
4492 int error;
4493
4494 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4495
4496 /*
4497 * It's acceptable to have no devs specified.
4498 */
4499 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4500 return (0);
4501
4502 if (ndev == 0)
4503 return (SET_ERROR(EINVAL));
4504
4505 /*
4506 * Make sure the pool is formatted with a version that supports this
4507 * device type.
4508 */
4509 if (spa_version(spa) < version)
4510 return (SET_ERROR(ENOTSUP));
4511
4512 /*
4513 * Set the pending device list so we correctly handle device in-use
4514 * checking.
4515 */
4516 sav->sav_pending = dev;
4517 sav->sav_npending = ndev;
4518
4519 for (i = 0; i < ndev; i++) {
4520 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4521 mode)) != 0)
4522 goto out;
4523
4524 if (!vd->vdev_ops->vdev_op_leaf) {
4525 vdev_free(vd);
4526 error = SET_ERROR(EINVAL);
4527 goto out;
4528 }
4529
4530 /*
4531 * The L2ARC currently only supports disk devices in
4532 * kernel context. For user-level testing, we allow it.
4533 */
4534#ifdef _KERNEL
4535 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4536 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4537 error = SET_ERROR(ENOTBLK);
4538 vdev_free(vd);
4539 goto out;
4540 }
4541#endif
4542 vd->vdev_top = vd;
4543
4544 if ((error = vdev_open(vd)) == 0 &&
4545 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4546 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4547 vd->vdev_guid) == 0);
4548 }
4549
4550 vdev_free(vd);
4551
4552 if (error &&
4553 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4554 goto out;
4555 else
4556 error = 0;
4557 }
4558
4559out:
4560 sav->sav_pending = NULL;
4561 sav->sav_npending = 0;
4562 return (error);
4563}
4564
4565static int
4566spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4567{
4568 int error;
4569
4570 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4571
4572 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4573 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4574 VDEV_LABEL_SPARE)) != 0) {
4575 return (error);
4576 }
4577
4578 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4579 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4580 VDEV_LABEL_L2CACHE));
4581}
4582
4583static void
4584spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4585 const char *config)
4586{
4587 int i;
4588
4589 if (sav->sav_config != NULL) {
4590 nvlist_t **olddevs;
4591 uint_t oldndevs;
4592 nvlist_t **newdevs;
4593
4594 /*
4595 * Generate new dev list by concatentating with the
4596 * current dev list.
4597 */
4598 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4599 &olddevs, &oldndevs) == 0);
4600
4601 newdevs = kmem_alloc(sizeof (void *) *
4602 (ndevs + oldndevs), KM_SLEEP);
4603 for (i = 0; i < oldndevs; i++)
4604 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4605 KM_SLEEP) == 0);
4606 for (i = 0; i < ndevs; i++)
4607 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4608 KM_SLEEP) == 0);
4609
4610 VERIFY(nvlist_remove(sav->sav_config, config,
4611 DATA_TYPE_NVLIST_ARRAY) == 0);
4612
4613 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4614 config, newdevs, ndevs + oldndevs) == 0);
4615 for (i = 0; i < oldndevs + ndevs; i++)
4616 nvlist_free(newdevs[i]);
4617 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4618 } else {
4619 /*
4620 * Generate a new dev list.
4621 */
4622 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4623 KM_SLEEP) == 0);
4624 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4625 devs, ndevs) == 0);
4626 }
4627}
4628
4629/*
4630 * Stop and drop level 2 ARC devices
4631 */
4632void
4633spa_l2cache_drop(spa_t *spa)
4634{
4635 vdev_t *vd;
4636 int i;
4637 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4638
4639 for (i = 0; i < sav->sav_count; i++) {
4640 uint64_t pool;
4641
4642 vd = sav->sav_vdevs[i];
4643 ASSERT(vd != NULL);
4644
4645 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4646 pool != 0ULL && l2arc_vdev_present(vd))
4647 l2arc_remove_vdev(vd);
4648 }
4649}
4650
4651/*
4652 * Pool Creation
4653 */
4654int
4655spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4656 nvlist_t *zplprops)
4657{
4658 spa_t *spa;
4659 char *altroot = NULL;
4660 vdev_t *rvd;
4661 dsl_pool_t *dp;
4662 dmu_tx_t *tx;
4663 int error = 0;
4664 uint64_t txg = TXG_INITIAL;
4665 nvlist_t **spares, **l2cache;
4666 uint_t nspares, nl2cache;
4667 uint64_t version, obj;
4668 boolean_t has_features;
4669
4670 /*
4671 * If this pool already exists, return failure.
4672 */
4673 mutex_enter(&spa_namespace_lock);
4674 if (spa_lookup(pool) != NULL) {
4675 mutex_exit(&spa_namespace_lock);
4676 return (SET_ERROR(EEXIST));
4677 }
4678
4679 /*
4680 * Allocate a new spa_t structure.
4681 */
4682 (void) nvlist_lookup_string(props,
4683 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4684 spa = spa_add(pool, NULL, altroot);
4685 spa_activate(spa, spa_mode_global);
4686
4687 if (props && (error = spa_prop_validate(spa, props))) {
4688 spa_deactivate(spa);
4689 spa_remove(spa);
4690 mutex_exit(&spa_namespace_lock);
4691 return (error);
4692 }
4693
4694 has_features = B_FALSE;
4695 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4696 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4697 if (zpool_prop_feature(nvpair_name(elem)))
4698 has_features = B_TRUE;
4699 }
4700
4701 if (has_features || nvlist_lookup_uint64(props,
4702 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4703 version = SPA_VERSION;
4704 }
4705 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4706
4707 spa->spa_first_txg = txg;
4708 spa->spa_uberblock.ub_txg = txg - 1;
4709 spa->spa_uberblock.ub_version = version;
4710 spa->spa_ubsync = spa->spa_uberblock;
4711 spa->spa_load_state = SPA_LOAD_CREATE;
4712 spa->spa_removing_phys.sr_state = DSS_NONE;
4713 spa->spa_removing_phys.sr_removing_vdev = -1;
4714 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4715
4716 /*
4717 * Create "The Godfather" zio to hold all async IOs
4718 */
4719 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4720 KM_SLEEP);
4721 for (int i = 0; i < max_ncpus; i++) {
4722 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4723 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4724 ZIO_FLAG_GODFATHER);
4725 }
4726
4727 /*
4728 * Create the root vdev.
4729 */
4730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4731
4732 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4733
4734 ASSERT(error != 0 || rvd != NULL);
4735 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4736
4737 if (error == 0 && !zfs_allocatable_devs(nvroot))
4738 error = SET_ERROR(EINVAL);
4739
4740 if (error == 0 &&
4741 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4742 (error = spa_validate_aux(spa, nvroot, txg,
4743 VDEV_ALLOC_ADD)) == 0) {
4744 for (int c = 0; c < rvd->vdev_children; c++) {
4745 vdev_ashift_optimize(rvd->vdev_child[c]);
4746 vdev_metaslab_set_size(rvd->vdev_child[c]);
4747 vdev_expand(rvd->vdev_child[c], txg);
4748 }
4749 }
4750
4751 spa_config_exit(spa, SCL_ALL, FTAG);
4752
4753 if (error != 0) {
4754 spa_unload(spa);
4755 spa_deactivate(spa);
4756 spa_remove(spa);
4757 mutex_exit(&spa_namespace_lock);
4758 return (error);
4759 }
4760
4761 /*
4762 * Get the list of spares, if specified.
4763 */
4764 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4765 &spares, &nspares) == 0) {
4766 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4767 KM_SLEEP) == 0);
4768 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4769 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4770 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4771 spa_load_spares(spa);
4772 spa_config_exit(spa, SCL_ALL, FTAG);
4773 spa->spa_spares.sav_sync = B_TRUE;
4774 }
4775
4776 /*
4777 * Get the list of level 2 cache devices, if specified.
4778 */
4779 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4780 &l2cache, &nl2cache) == 0) {
4781 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4782 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4783 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4784 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4786 spa_load_l2cache(spa);
4787 spa_config_exit(spa, SCL_ALL, FTAG);
4788 spa->spa_l2cache.sav_sync = B_TRUE;
4789 }
4790
4791 spa->spa_is_initializing = B_TRUE;
4792 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4793 spa->spa_meta_objset = dp->dp_meta_objset;
4794 spa->spa_is_initializing = B_FALSE;
4795
4796 /*
4797 * Create DDTs (dedup tables).
4798 */
4799 ddt_create(spa);
4800
4801 spa_update_dspace(spa);
4802
4803 tx = dmu_tx_create_assigned(dp, txg);
4804
4805 /*
4806 * Create the pool config object.
4807 */
4808 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4809 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4810 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4811
4812 if (zap_add(spa->spa_meta_objset,
4813 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4814 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4815 cmn_err(CE_PANIC, "failed to add pool config");
4816 }
4817
4818 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4819 spa_feature_create_zap_objects(spa, tx);
4820
4821 if (zap_add(spa->spa_meta_objset,
4822 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4823 sizeof (uint64_t), 1, &version, tx) != 0) {
4824 cmn_err(CE_PANIC, "failed to add pool version");
4825 }
4826
4827 /* Newly created pools with the right version are always deflated. */
4828 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4829 spa->spa_deflate = TRUE;
4830 if (zap_add(spa->spa_meta_objset,
4831 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4832 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4833 cmn_err(CE_PANIC, "failed to add deflate");
4834 }
4835 }
4836
4837 /*
4838 * Create the deferred-free bpobj. Turn off compression
4839 * because sync-to-convergence takes longer if the blocksize
4840 * keeps changing.
4841 */
4842 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4843 dmu_object_set_compress(spa->spa_meta_objset, obj,
4844 ZIO_COMPRESS_OFF, tx);
4845 if (zap_add(spa->spa_meta_objset,
4846 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4847 sizeof (uint64_t), 1, &obj, tx) != 0) {
4848 cmn_err(CE_PANIC, "failed to add bpobj");
4849 }
4850 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4851 spa->spa_meta_objset, obj));
4852
4853 /*
4854 * Create the pool's history object.
4855 */
4856 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4857 spa_history_create_obj(spa, tx);
4858
4859 /*
4860 * Generate some random noise for salted checksums to operate on.
4861 */
4862 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4863 sizeof (spa->spa_cksum_salt.zcs_bytes));
4864
4865 /*
4866 * Set pool properties.
4867 */
4868 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4869 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4870 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4871 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4872
4873 if (props != NULL) {
4874 spa_configfile_set(spa, props, B_FALSE);
4875 spa_sync_props(props, tx);
4876 }
4877
4878 dmu_tx_commit(tx);
4879
4880 spa->spa_sync_on = B_TRUE;
4881 txg_sync_start(spa->spa_dsl_pool);
4882
4883 /*
4884 * We explicitly wait for the first transaction to complete so that our
4885 * bean counters are appropriately updated.
4886 */
4887 txg_wait_synced(spa->spa_dsl_pool, txg);
4888
4889 spa_spawn_aux_threads(spa);
4890
4891 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4892 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4893
4894 spa_history_log_version(spa, "create");
4895
4896 /*
4897 * Don't count references from objsets that are already closed
4898 * and are making their way through the eviction process.
4899 */
4900 spa_evicting_os_wait(spa);
4901 spa->spa_minref = refcount_count(&spa->spa_refcount);
4902 spa->spa_load_state = SPA_LOAD_NONE;
4903
4904 mutex_exit(&spa_namespace_lock);
4905
4906 return (0);
4907}
4908
4909#ifdef _KERNEL
4910#ifdef illumos
4911/*
4912 * Get the root pool information from the root disk, then import the root pool
4913 * during the system boot up time.
4914 */
4915extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4916
4917static nvlist_t *
4918spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4919{
4920 nvlist_t *config;
4921 nvlist_t *nvtop, *nvroot;
4922 uint64_t pgid;
4923
4924 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
4925 return (NULL);
4926
4927 /*
4928 * Add this top-level vdev to the child array.
4929 */
4930 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4931 &nvtop) == 0);
4932 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4933 &pgid) == 0);
4934 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
4935
4936 /*
4937 * Put this pool's top-level vdevs into a root vdev.
4938 */
4939 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4940 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4941 VDEV_TYPE_ROOT) == 0);
4942 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4943 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4944 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4945 &nvtop, 1) == 0);
4946
4947 /*
4948 * Replace the existing vdev_tree with the new root vdev in
4949 * this pool's configuration (remove the old, add the new).
4950 */
4951 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4952 nvlist_free(nvroot);
4953 return (config);
4954}
4955
4956/*
4957 * Walk the vdev tree and see if we can find a device with "better"
4958 * configuration. A configuration is "better" if the label on that
4959 * device has a more recent txg.
4960 */
4961static void
4962spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
4963{
4964 for (int c = 0; c < vd->vdev_children; c++)
4965 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
4966
4967 if (vd->vdev_ops->vdev_op_leaf) {
4968 nvlist_t *label;
4969 uint64_t label_txg;
4970
4971 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4972 &label) != 0)
4973 return;
4974
4975 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4976 &label_txg) == 0);
4977
4978 /*
4979 * Do we have a better boot device?
4980 */
4981 if (label_txg > *txg) {
4982 *txg = label_txg;
4983 *avd = vd;
4984 }
4985 nvlist_free(label);
4986 }
4987}
4988
4989/*
4990 * Import a root pool.
4991 *
4992 * For x86. devpath_list will consist of devid and/or physpath name of
4993 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4994 * The GRUB "findroot" command will return the vdev we should boot.
4995 *
4996 * For Sparc, devpath_list consists the physpath name of the booting device
4997 * no matter the rootpool is a single device pool or a mirrored pool.
4998 * e.g.
4999 * "/pci@1f,0/ide@d/disk@0,0:a"
5000 */
5001int
5002spa_import_rootpool(char *devpath, char *devid)
5003{
5004 spa_t *spa;
5005 vdev_t *rvd, *bvd, *avd = NULL;
5006 nvlist_t *config, *nvtop;
5007 uint64_t guid, txg;
5008 char *pname;
5009 int error;
5010
5011 /*
5012 * Read the label from the boot device and generate a configuration.
5013 */
5014 config = spa_generate_rootconf(devpath, devid, &guid);
5015#if defined(_OBP) && defined(_KERNEL)
5016 if (config == NULL) {
5017 if (strstr(devpath, "/iscsi/ssd") != NULL) {
5018 /* iscsi boot */
5019 get_iscsi_bootpath_phy(devpath);
5020 config = spa_generate_rootconf(devpath, devid, &guid);
5021 }
5022 }
5023#endif
5024 if (config == NULL) {
5025 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5026 devpath);
5027 return (SET_ERROR(EIO));
5028 }
5029
5030 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5031 &pname) == 0);
5032 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5033
5034 mutex_enter(&spa_namespace_lock);
5035 if ((spa = spa_lookup(pname)) != NULL) {
5036 /*
5037 * Remove the existing root pool from the namespace so that we
5038 * can replace it with the correct config we just read in.
5039 */
5040 spa_remove(spa);
5041 }
5042
5043 spa = spa_add(pname, config, NULL);
5044 spa->spa_is_root = B_TRUE;
5045 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5046 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5047 &spa->spa_ubsync.ub_version) != 0)
5048 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5049
5050 /*
5051 * Build up a vdev tree based on the boot device's label config.
5052 */
5053 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5054 &nvtop) == 0);
5055 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5056 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5057 VDEV_ALLOC_ROOTPOOL);
5058 spa_config_exit(spa, SCL_ALL, FTAG);
5059 if (error) {
5060 mutex_exit(&spa_namespace_lock);
5061 nvlist_free(config);
5062 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5063 pname);
5064 return (error);
5065 }
5066
5067 /*
5068 * Get the boot vdev.
5069 */
5070 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5071 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5072 (u_longlong_t)guid);
5073 error = SET_ERROR(ENOENT);
5074 goto out;
5075 }
5076
5077 /*
5078 * Determine if there is a better boot device.
5079 */
5080 avd = bvd;
5081 spa_alt_rootvdev(rvd, &avd, &txg);
5082 if (avd != bvd) {
5083 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5084 "try booting from '%s'", avd->vdev_path);
5085 error = SET_ERROR(EINVAL);
5086 goto out;
5087 }
5088
5089 /*
5090 * If the boot device is part of a spare vdev then ensure that
5091 * we're booting off the active spare.
5092 */
5093 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5094 !bvd->vdev_isspare) {
5095 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5096 "try booting from '%s'",
5097 bvd->vdev_parent->
5098 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5099 error = SET_ERROR(EINVAL);
5100 goto out;
5101 }
5102
5103 error = 0;
5104out:
5105 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5106 vdev_free(rvd);
5107 spa_config_exit(spa, SCL_ALL, FTAG);
5108 mutex_exit(&spa_namespace_lock);
5109
5110 nvlist_free(config);
5111 return (error);
5112}
5113
5114#else /* !illumos */
5115
5116extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
5117 uint64_t *count);
5118
5119static nvlist_t *
5120spa_generate_rootconf(const char *name)
5121{
5122 nvlist_t **configs, **tops;
5123 nvlist_t *config;
5124 nvlist_t *best_cfg, *nvtop, *nvroot;
5125 uint64_t *holes;
5126 uint64_t best_txg;
5127 uint64_t nchildren;
5128 uint64_t pgid;
5129 uint64_t count;
5130 uint64_t i;
5131 uint_t nholes;
5132
5133 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
5134 return (NULL);
5135
5136 ASSERT3U(count, !=, 0);
5137 best_txg = 0;
5138 for (i = 0; i < count; i++) {
5139 uint64_t txg;
5140
5141 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
5142 &txg) == 0);
5143 if (txg > best_txg) {
5144 best_txg = txg;
5145 best_cfg = configs[i];
5146 }
5147 }
5148
5149 nchildren = 1;
5150 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
5151 holes = NULL;
5152 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
5153 &holes, &nholes);
5154
5155 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
5156 for (i = 0; i < nchildren; i++) {
5157 if (i >= count)
5158 break;
5159 if (configs[i] == NULL)
5160 continue;
5161 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
5162 &nvtop) == 0);
5163 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
5164 }
5165 for (i = 0; holes != NULL && i < nholes; i++) {
5166 if (i >= nchildren)
5167 continue;
5168 if (tops[holes[i]] != NULL)
5169 continue;
5170 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
5171 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
5172 VDEV_TYPE_HOLE) == 0);
5173 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
5174 holes[i]) == 0);
5175 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
5176 0) == 0);
5177 }
5178 for (i = 0; i < nchildren; i++) {
5179 if (tops[i] != NULL)
5180 continue;
5181 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
5182 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
5183 VDEV_TYPE_MISSING) == 0);
5184 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
5185 i) == 0);
5186 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
5187 0) == 0);
5188 }
5189
5190 /*
5191 * Create pool config based on the best vdev config.
5192 */
5193 nvlist_dup(best_cfg, &config, KM_SLEEP);
5194
5195 /*
5196 * Put this pool's top-level vdevs into a root vdev.
5197 */
5198 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5199 &pgid) == 0);
5200 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5201 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5202 VDEV_TYPE_ROOT) == 0);
5203 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5204 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5205 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5206 tops, nchildren) == 0);
5207
5208 /*
5209 * Replace the existing vdev_tree with the new root vdev in
5210 * this pool's configuration (remove the old, add the new).
5211 */
5212 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5213
5214 /*
5215 * Drop vdev config elements that should not be present at pool level.
5216 */
5217 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
5218 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
5219
5220 for (i = 0; i < count; i++)
5221 nvlist_free(configs[i]);
5222 kmem_free(configs, count * sizeof(void *));
5223 for (i = 0; i < nchildren; i++)
5224 nvlist_free(tops[i]);
5225 kmem_free(tops, nchildren * sizeof(void *));
5226 nvlist_free(nvroot);
5227 return (config);
5228}
5229
5230int
5231spa_import_rootpool(const char *name)
5232{
5233 spa_t *spa;
5234 vdev_t *rvd, *bvd, *avd = NULL;
5235 nvlist_t *config, *nvtop;
5236 uint64_t txg;
5237 char *pname;
5238 int error;
5239
5240 /*
5241 * Read the label from the boot device and generate a configuration.
5242 */
5243 config = spa_generate_rootconf(name);
5244
5245 mutex_enter(&spa_namespace_lock);
5246 if (config != NULL) {
5247 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5248 &pname) == 0 && strcmp(name, pname) == 0);
5249 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
5250 == 0);
5251
5252 if ((spa = spa_lookup(pname)) != NULL) {
5253 /*
5254 * The pool could already be imported,
5255 * e.g., after reboot -r.
5256 */
5257 if (spa->spa_state == POOL_STATE_ACTIVE) {
5258 mutex_exit(&spa_namespace_lock);
5259 nvlist_free(config);
5260 return (0);
5261 }
5262
5263 /*
5264 * Remove the existing root pool from the namespace so
5265 * that we can replace it with the correct config
5266 * we just read in.
5267 */
5268 spa_remove(spa);
5269 }
5270 spa = spa_add(pname, config, NULL);
5271
5272 /*
5273 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
5274 * via spa_version().
5275 */
5276 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5277 &spa->spa_ubsync.ub_version) != 0)
5278 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5279 } else if ((spa = spa_lookup(name)) == NULL) {
5280 mutex_exit(&spa_namespace_lock);
5281 nvlist_free(config);
5282 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
5283 name);
5284 return (EIO);
5285 } else {
5286 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
5287 }
5288 spa->spa_is_root = B_TRUE;
5289 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5290
5291 /*
5292 * Build up a vdev tree based on the boot device's label config.
5293 */
5294 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5295 &nvtop) == 0);
5296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5297 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5298 VDEV_ALLOC_ROOTPOOL);
5299 spa_config_exit(spa, SCL_ALL, FTAG);
5300 if (error) {
5301 mutex_exit(&spa_namespace_lock);
5302 nvlist_free(config);
5303 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5304 pname);
5305 return (error);
5306 }
5307
5308 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5309 vdev_free(rvd);
5310 spa_config_exit(spa, SCL_ALL, FTAG);
5311 mutex_exit(&spa_namespace_lock);
5312
5313 nvlist_free(config);
5314 return (0);
5315}
5316
5317#endif /* illumos */
5318#endif /* _KERNEL */
5319
5320/*
5321 * Import a non-root pool into the system.
5322 */
5323int
5324spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5325{
5326 spa_t *spa;
5327 char *altroot = NULL;
5328 spa_load_state_t state = SPA_LOAD_IMPORT;
|
5329 zpool_rewind_policy_t policy;
| 5329 zpool_load_policy_t policy;
|
5330 uint64_t mode = spa_mode_global;
5331 uint64_t readonly = B_FALSE;
5332 int error;
5333 nvlist_t *nvroot;
5334 nvlist_t **spares, **l2cache;
5335 uint_t nspares, nl2cache;
5336
5337 /*
5338 * If a pool with this name exists, return failure.
5339 */
5340 mutex_enter(&spa_namespace_lock);
5341 if (spa_lookup(pool) != NULL) {
5342 mutex_exit(&spa_namespace_lock);
5343 return (SET_ERROR(EEXIST));
5344 }
5345
5346 /*
5347 * Create and initialize the spa structure.
5348 */
5349 (void) nvlist_lookup_string(props,
5350 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5351 (void) nvlist_lookup_uint64(props,
5352 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5353 if (readonly)
5354 mode = FREAD;
5355 spa = spa_add(pool, config, altroot);
5356 spa->spa_import_flags = flags;
5357
5358 /*
5359 * Verbatim import - Take a pool and insert it into the namespace
5360 * as if it had been loaded at boot.
5361 */
5362 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5363 if (props != NULL)
5364 spa_configfile_set(spa, props, B_FALSE);
5365
5366 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5367 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5368 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5369 mutex_exit(&spa_namespace_lock);
5370 return (0);
5371 }
5372
5373 spa_activate(spa, mode);
5374
5375 /*
5376 * Don't start async tasks until we know everything is healthy.
5377 */
5378 spa_async_suspend(spa);
5379
| 5330 uint64_t mode = spa_mode_global;
5331 uint64_t readonly = B_FALSE;
5332 int error;
5333 nvlist_t *nvroot;
5334 nvlist_t **spares, **l2cache;
5335 uint_t nspares, nl2cache;
5336
5337 /*
5338 * If a pool with this name exists, return failure.
5339 */
5340 mutex_enter(&spa_namespace_lock);
5341 if (spa_lookup(pool) != NULL) {
5342 mutex_exit(&spa_namespace_lock);
5343 return (SET_ERROR(EEXIST));
5344 }
5345
5346 /*
5347 * Create and initialize the spa structure.
5348 */
5349 (void) nvlist_lookup_string(props,
5350 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5351 (void) nvlist_lookup_uint64(props,
5352 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5353 if (readonly)
5354 mode = FREAD;
5355 spa = spa_add(pool, config, altroot);
5356 spa->spa_import_flags = flags;
5357
5358 /*
5359 * Verbatim import - Take a pool and insert it into the namespace
5360 * as if it had been loaded at boot.
5361 */
5362 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5363 if (props != NULL)
5364 spa_configfile_set(spa, props, B_FALSE);
5365
5366 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5367 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5368 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5369 mutex_exit(&spa_namespace_lock);
5370 return (0);
5371 }
5372
5373 spa_activate(spa, mode);
5374
5375 /*
5376 * Don't start async tasks until we know everything is healthy.
5377 */
5378 spa_async_suspend(spa);
5379
|
5380 zpool_get_rewind_policy(config, &policy);
5381 if (policy.zrp_request & ZPOOL_DO_REWIND)
| 5380 zpool_get_load_policy(config, &policy);
5381 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
|
5382 state = SPA_LOAD_RECOVER;
5383
5384 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5385
5386 if (state != SPA_LOAD_RECOVER) {
5387 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5388 zfs_dbgmsg("spa_import: importing %s", pool);
5389 } else {
5390 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
| 5382 state = SPA_LOAD_RECOVER;
5383
5384 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5385
5386 if (state != SPA_LOAD_RECOVER) {
5387 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5388 zfs_dbgmsg("spa_import: importing %s", pool);
5389 } else {
5390 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
|
5391 "(RECOVERY MODE)", pool, (longlong_t)policy.zrp_txg);
| 5391 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
|
5392 }
| 5392 }
|
5393 error = spa_load_best(spa, state, policy.zrp_txg, policy.zrp_request);
| 5393 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
|
5394
5395 /*
5396 * Propagate anything learned while loading the pool and pass it
5397 * back to caller (i.e. rewind info, missing devices, etc).
5398 */
5399 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5400 spa->spa_load_info) == 0);
5401
5402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5403 /*
5404 * Toss any existing sparelist, as it doesn't have any validity
5405 * anymore, and conflicts with spa_has_spare().
5406 */
5407 if (spa->spa_spares.sav_config) {
5408 nvlist_free(spa->spa_spares.sav_config);
5409 spa->spa_spares.sav_config = NULL;
5410 spa_load_spares(spa);
5411 }
5412 if (spa->spa_l2cache.sav_config) {
5413 nvlist_free(spa->spa_l2cache.sav_config);
5414 spa->spa_l2cache.sav_config = NULL;
5415 spa_load_l2cache(spa);
5416 }
5417
5418 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5419 &nvroot) == 0);
5420 if (error == 0)
5421 error = spa_validate_aux(spa, nvroot, -1ULL,
5422 VDEV_ALLOC_SPARE);
5423 if (error == 0)
5424 error = spa_validate_aux(spa, nvroot, -1ULL,
5425 VDEV_ALLOC_L2CACHE);
5426 spa_config_exit(spa, SCL_ALL, FTAG);
5427
5428 if (props != NULL)
5429 spa_configfile_set(spa, props, B_FALSE);
5430
5431 if (error != 0 || (props && spa_writeable(spa) &&
5432 (error = spa_prop_set(spa, props)))) {
5433 spa_unload(spa);
5434 spa_deactivate(spa);
5435 spa_remove(spa);
5436 mutex_exit(&spa_namespace_lock);
5437 return (error);
5438 }
5439
5440 spa_async_resume(spa);
5441
5442 /*
5443 * Override any spares and level 2 cache devices as specified by
5444 * the user, as these may have correct device names/devids, etc.
5445 */
5446 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5447 &spares, &nspares) == 0) {
5448 if (spa->spa_spares.sav_config)
5449 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5450 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5451 else
5452 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5453 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5454 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5455 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5456 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5457 spa_load_spares(spa);
5458 spa_config_exit(spa, SCL_ALL, FTAG);
5459 spa->spa_spares.sav_sync = B_TRUE;
5460 }
5461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5462 &l2cache, &nl2cache) == 0) {
5463 if (spa->spa_l2cache.sav_config)
5464 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5465 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5466 else
5467 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5468 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5469 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5470 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5471 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5472 spa_load_l2cache(spa);
5473 spa_config_exit(spa, SCL_ALL, FTAG);
5474 spa->spa_l2cache.sav_sync = B_TRUE;
5475 }
5476
5477 /*
5478 * Check for any removed devices.
5479 */
5480 if (spa->spa_autoreplace) {
5481 spa_aux_check_removed(&spa->spa_spares);
5482 spa_aux_check_removed(&spa->spa_l2cache);
5483 }
5484
5485 if (spa_writeable(spa)) {
5486 /*
5487 * Update the config cache to include the newly-imported pool.
5488 */
5489 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5490 }
5491
5492 /*
5493 * It's possible that the pool was expanded while it was exported.
5494 * We kick off an async task to handle this for us.
5495 */
5496 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5497
5498 spa_history_log_version(spa, "import");
5499
5500 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5501
5502 mutex_exit(&spa_namespace_lock);
5503
5504#ifdef __FreeBSD__
5505#ifdef _KERNEL
5506 zvol_create_minors(pool);
5507#endif
5508#endif
5509 return (0);
5510}
5511
5512nvlist_t *
5513spa_tryimport(nvlist_t *tryconfig)
5514{
5515 nvlist_t *config = NULL;
5516 char *poolname, *cachefile;
5517 spa_t *spa;
5518 uint64_t state;
5519 int error;
| 5394
5395 /*
5396 * Propagate anything learned while loading the pool and pass it
5397 * back to caller (i.e. rewind info, missing devices, etc).
5398 */
5399 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5400 spa->spa_load_info) == 0);
5401
5402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5403 /*
5404 * Toss any existing sparelist, as it doesn't have any validity
5405 * anymore, and conflicts with spa_has_spare().
5406 */
5407 if (spa->spa_spares.sav_config) {
5408 nvlist_free(spa->spa_spares.sav_config);
5409 spa->spa_spares.sav_config = NULL;
5410 spa_load_spares(spa);
5411 }
5412 if (spa->spa_l2cache.sav_config) {
5413 nvlist_free(spa->spa_l2cache.sav_config);
5414 spa->spa_l2cache.sav_config = NULL;
5415 spa_load_l2cache(spa);
5416 }
5417
5418 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5419 &nvroot) == 0);
5420 if (error == 0)
5421 error = spa_validate_aux(spa, nvroot, -1ULL,
5422 VDEV_ALLOC_SPARE);
5423 if (error == 0)
5424 error = spa_validate_aux(spa, nvroot, -1ULL,
5425 VDEV_ALLOC_L2CACHE);
5426 spa_config_exit(spa, SCL_ALL, FTAG);
5427
5428 if (props != NULL)
5429 spa_configfile_set(spa, props, B_FALSE);
5430
5431 if (error != 0 || (props && spa_writeable(spa) &&
5432 (error = spa_prop_set(spa, props)))) {
5433 spa_unload(spa);
5434 spa_deactivate(spa);
5435 spa_remove(spa);
5436 mutex_exit(&spa_namespace_lock);
5437 return (error);
5438 }
5439
5440 spa_async_resume(spa);
5441
5442 /*
5443 * Override any spares and level 2 cache devices as specified by
5444 * the user, as these may have correct device names/devids, etc.
5445 */
5446 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5447 &spares, &nspares) == 0) {
5448 if (spa->spa_spares.sav_config)
5449 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5450 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5451 else
5452 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5453 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5454 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5455 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5456 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5457 spa_load_spares(spa);
5458 spa_config_exit(spa, SCL_ALL, FTAG);
5459 spa->spa_spares.sav_sync = B_TRUE;
5460 }
5461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5462 &l2cache, &nl2cache) == 0) {
5463 if (spa->spa_l2cache.sav_config)
5464 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5465 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5466 else
5467 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5468 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5469 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5470 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5471 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5472 spa_load_l2cache(spa);
5473 spa_config_exit(spa, SCL_ALL, FTAG);
5474 spa->spa_l2cache.sav_sync = B_TRUE;
5475 }
5476
5477 /*
5478 * Check for any removed devices.
5479 */
5480 if (spa->spa_autoreplace) {
5481 spa_aux_check_removed(&spa->spa_spares);
5482 spa_aux_check_removed(&spa->spa_l2cache);
5483 }
5484
5485 if (spa_writeable(spa)) {
5486 /*
5487 * Update the config cache to include the newly-imported pool.
5488 */
5489 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5490 }
5491
5492 /*
5493 * It's possible that the pool was expanded while it was exported.
5494 * We kick off an async task to handle this for us.
5495 */
5496 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5497
5498 spa_history_log_version(spa, "import");
5499
5500 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5501
5502 mutex_exit(&spa_namespace_lock);
5503
5504#ifdef __FreeBSD__
5505#ifdef _KERNEL
5506 zvol_create_minors(pool);
5507#endif
5508#endif
5509 return (0);
5510}
5511
5512nvlist_t *
5513spa_tryimport(nvlist_t *tryconfig)
5514{
5515 nvlist_t *config = NULL;
5516 char *poolname, *cachefile;
5517 spa_t *spa;
5518 uint64_t state;
5519 int error;
|
5520 zpool_rewind_policy_t policy;
| 5520 zpool_load_policy_t policy;
|
5521
5522 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5523 return (NULL);
5524
5525 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5526 return (NULL);
5527
5528 /*
5529 * Create and initialize the spa structure.
5530 */
5531 mutex_enter(&spa_namespace_lock);
5532 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5533 spa_activate(spa, FREAD);
5534
5535 /*
| 5521
5522 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5523 return (NULL);
5524
5525 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5526 return (NULL);
5527
5528 /*
5529 * Create and initialize the spa structure.
5530 */
5531 mutex_enter(&spa_namespace_lock);
5532 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5533 spa_activate(spa, FREAD);
5534
5535 /*
|
5536 * Rewind pool if a max txg was provided. Note that even though we
5537 * retrieve the complete rewind policy, only the rewind txg is relevant
5538 * for tryimport.
| 5536 * Rewind pool if a max txg was provided.
|
5539 */
| 5537 */
|
5540 zpool_get_rewind_policy(spa->spa_config, &policy);
5541 if (policy.zrp_txg != UINT64_MAX) {
5542 spa->spa_load_max_txg = policy.zrp_txg;
| 5538 zpool_get_load_policy(spa->spa_config, &policy);
5539 if (policy.zlp_txg != UINT64_MAX) {
5540 spa->spa_load_max_txg = policy.zlp_txg;
|
5543 spa->spa_extreme_rewind = B_TRUE;
5544 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
| 5541 spa->spa_extreme_rewind = B_TRUE;
5542 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
|
5545 poolname, (longlong_t)policy.zrp_txg);
| 5543 poolname, (longlong_t)policy.zlp_txg);
|
5546 } else {
5547 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5548 }
5549
5550 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5551 == 0) {
5552 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5553 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5554 } else {
5555 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5556 }
5557
5558 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5559
5560 /*
5561 * If 'tryconfig' was at least parsable, return the current config.
5562 */
5563 if (spa->spa_root_vdev != NULL) {
5564 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5565 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5566 poolname) == 0);
5567 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5568 state) == 0);
5569 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5570 spa->spa_uberblock.ub_timestamp) == 0);
5571 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5572 spa->spa_load_info) == 0);
5573
5574 /*
5575 * If the bootfs property exists on this pool then we
5576 * copy it out so that external consumers can tell which
5577 * pools are bootable.
5578 */
5579 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5580 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5581
5582 /*
5583 * We have to play games with the name since the
5584 * pool was opened as TRYIMPORT_NAME.
5585 */
5586 if (dsl_dsobj_to_dsname(spa_name(spa),
5587 spa->spa_bootfs, tmpname) == 0) {
5588 char *cp;
5589 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5590
5591 cp = strchr(tmpname, '/');
5592 if (cp == NULL) {
5593 (void) strlcpy(dsname, tmpname,
5594 MAXPATHLEN);
5595 } else {
5596 (void) snprintf(dsname, MAXPATHLEN,
5597 "%s/%s", poolname, ++cp);
5598 }
5599 VERIFY(nvlist_add_string(config,
5600 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5601 kmem_free(dsname, MAXPATHLEN);
5602 }
5603 kmem_free(tmpname, MAXPATHLEN);
5604 }
5605
5606 /*
5607 * Add the list of hot spares and level 2 cache devices.
5608 */
5609 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5610 spa_add_spares(spa, config);
5611 spa_add_l2cache(spa, config);
5612 spa_config_exit(spa, SCL_CONFIG, FTAG);
5613 }
5614
5615 spa_unload(spa);
5616 spa_deactivate(spa);
5617 spa_remove(spa);
5618 mutex_exit(&spa_namespace_lock);
5619
5620 return (config);
5621}
5622
5623/*
5624 * Pool export/destroy
5625 *
5626 * The act of destroying or exporting a pool is very simple. We make sure there
5627 * is no more pending I/O and any references to the pool are gone. Then, we
5628 * update the pool state and sync all the labels to disk, removing the
5629 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5630 * we don't sync the labels or remove the configuration cache.
5631 */
5632static int
5633spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5634 boolean_t force, boolean_t hardforce)
5635{
5636 spa_t *spa;
5637
5638 if (oldconfig)
5639 *oldconfig = NULL;
5640
5641 if (!(spa_mode_global & FWRITE))
5642 return (SET_ERROR(EROFS));
5643
5644 mutex_enter(&spa_namespace_lock);
5645 if ((spa = spa_lookup(pool)) == NULL) {
5646 mutex_exit(&spa_namespace_lock);
5647 return (SET_ERROR(ENOENT));
5648 }
5649
5650 /*
5651 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5652 * reacquire the namespace lock, and see if we can export.
5653 */
5654 spa_open_ref(spa, FTAG);
5655 mutex_exit(&spa_namespace_lock);
5656 spa_async_suspend(spa);
5657 mutex_enter(&spa_namespace_lock);
5658 spa_close(spa, FTAG);
5659
5660 /*
5661 * The pool will be in core if it's openable,
5662 * in which case we can modify its state.
5663 */
5664 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5665 /*
5666 * Objsets may be open only because they're dirty, so we
5667 * have to force it to sync before checking spa_refcnt.
5668 */
5669 txg_wait_synced(spa->spa_dsl_pool, 0);
5670 spa_evicting_os_wait(spa);
5671
5672 /*
5673 * A pool cannot be exported or destroyed if there are active
5674 * references. If we are resetting a pool, allow references by
5675 * fault injection handlers.
5676 */
5677 if (!spa_refcount_zero(spa) ||
5678 (spa->spa_inject_ref != 0 &&
5679 new_state != POOL_STATE_UNINITIALIZED)) {
5680 spa_async_resume(spa);
5681 mutex_exit(&spa_namespace_lock);
5682 return (SET_ERROR(EBUSY));
5683 }
5684
5685 /*
5686 * A pool cannot be exported if it has an active shared spare.
5687 * This is to prevent other pools stealing the active spare
5688 * from an exported pool. At user's own will, such pool can
5689 * be forcedly exported.
5690 */
5691 if (!force && new_state == POOL_STATE_EXPORTED &&
5692 spa_has_active_shared_spare(spa)) {
5693 spa_async_resume(spa);
5694 mutex_exit(&spa_namespace_lock);
5695 return (SET_ERROR(EXDEV));
5696 }
5697
5698 /*
5699 * We want this to be reflected on every label,
5700 * so mark them all dirty. spa_unload() will do the
5701 * final sync that pushes these changes out.
5702 */
5703 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5704 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5705 spa->spa_state = new_state;
5706 spa->spa_final_txg = spa_last_synced_txg(spa) +
5707 TXG_DEFER_SIZE + 1;
5708 vdev_config_dirty(spa->spa_root_vdev);
5709 spa_config_exit(spa, SCL_ALL, FTAG);
5710 }
5711 }
5712
5713 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5714
5715 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5716 spa_unload(spa);
5717 spa_deactivate(spa);
5718 }
5719
5720 if (oldconfig && spa->spa_config)
5721 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5722
5723 if (new_state != POOL_STATE_UNINITIALIZED) {
5724 if (!hardforce)
5725 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5726 spa_remove(spa);
5727 }
5728 mutex_exit(&spa_namespace_lock);
5729
5730 return (0);
5731}
5732
5733/*
5734 * Destroy a storage pool.
5735 */
5736int
5737spa_destroy(char *pool)
5738{
5739 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5740 B_FALSE, B_FALSE));
5741}
5742
5743/*
5744 * Export a storage pool.
5745 */
5746int
5747spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5748 boolean_t hardforce)
5749{
5750 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5751 force, hardforce));
5752}
5753
5754/*
5755 * Similar to spa_export(), this unloads the spa_t without actually removing it
5756 * from the namespace in any way.
5757 */
5758int
5759spa_reset(char *pool)
5760{
5761 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5762 B_FALSE, B_FALSE));
5763}
5764
5765/*
5766 * ==========================================================================
5767 * Device manipulation
5768 * ==========================================================================
5769 */
5770
5771/*
5772 * Add a device to a storage pool.
5773 */
5774int
5775spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5776{
5777 uint64_t txg, id;
5778 int error;
5779 vdev_t *rvd = spa->spa_root_vdev;
5780 vdev_t *vd, *tvd;
5781 nvlist_t **spares, **l2cache;
5782 uint_t nspares, nl2cache;
5783
5784 ASSERT(spa_writeable(spa));
5785
5786 txg = spa_vdev_enter(spa);
5787
5788 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5789 VDEV_ALLOC_ADD)) != 0)
5790 return (spa_vdev_exit(spa, NULL, txg, error));
5791
5792 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5793
5794 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5795 &nspares) != 0)
5796 nspares = 0;
5797
5798 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5799 &nl2cache) != 0)
5800 nl2cache = 0;
5801
5802 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5803 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5804
5805 if (vd->vdev_children != 0 &&
5806 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5807 return (spa_vdev_exit(spa, vd, txg, error));
5808
5809 /*
5810 * We must validate the spares and l2cache devices after checking the
5811 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5812 */
5813 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5814 return (spa_vdev_exit(spa, vd, txg, error));
5815
5816 /*
5817 * If we are in the middle of a device removal, we can only add
5818 * devices which match the existing devices in the pool.
5819 * If we are in the middle of a removal, or have some indirect
5820 * vdevs, we can not add raidz toplevels.
5821 */
5822 if (spa->spa_vdev_removal != NULL ||
5823 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5824 for (int c = 0; c < vd->vdev_children; c++) {
5825 tvd = vd->vdev_child[c];
5826 if (spa->spa_vdev_removal != NULL &&
5827 tvd->vdev_ashift !=
5828 spa->spa_vdev_removal->svr_vdev->vdev_ashift) {
5829 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5830 }
5831 /* Fail if top level vdev is raidz */
5832 if (tvd->vdev_ops == &vdev_raidz_ops) {
5833 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5834 }
5835 /*
5836 * Need the top level mirror to be
5837 * a mirror of leaf vdevs only
5838 */
5839 if (tvd->vdev_ops == &vdev_mirror_ops) {
5840 for (uint64_t cid = 0;
5841 cid < tvd->vdev_children; cid++) {
5842 vdev_t *cvd = tvd->vdev_child[cid];
5843 if (!cvd->vdev_ops->vdev_op_leaf) {
5844 return (spa_vdev_exit(spa, vd,
5845 txg, EINVAL));
5846 }
5847 }
5848 }
5849 }
5850 }
5851
5852 for (int c = 0; c < vd->vdev_children; c++) {
5853
5854 /*
5855 * Set the vdev id to the first hole, if one exists.
5856 */
5857 for (id = 0; id < rvd->vdev_children; id++) {
5858 if (rvd->vdev_child[id]->vdev_ishole) {
5859 vdev_free(rvd->vdev_child[id]);
5860 break;
5861 }
5862 }
5863 tvd = vd->vdev_child[c];
5864 vdev_remove_child(vd, tvd);
5865 tvd->vdev_id = id;
5866 vdev_add_child(rvd, tvd);
5867 vdev_config_dirty(tvd);
5868 }
5869
5870 if (nspares != 0) {
5871 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5872 ZPOOL_CONFIG_SPARES);
5873 spa_load_spares(spa);
5874 spa->spa_spares.sav_sync = B_TRUE;
5875 }
5876
5877 if (nl2cache != 0) {
5878 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5879 ZPOOL_CONFIG_L2CACHE);
5880 spa_load_l2cache(spa);
5881 spa->spa_l2cache.sav_sync = B_TRUE;
5882 }
5883
5884 /*
5885 * We have to be careful when adding new vdevs to an existing pool.
5886 * If other threads start allocating from these vdevs before we
5887 * sync the config cache, and we lose power, then upon reboot we may
5888 * fail to open the pool because there are DVAs that the config cache
5889 * can't translate. Therefore, we first add the vdevs without
5890 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5891 * and then let spa_config_update() initialize the new metaslabs.
5892 *
5893 * spa_load() checks for added-but-not-initialized vdevs, so that
5894 * if we lose power at any point in this sequence, the remaining
5895 * steps will be completed the next time we load the pool.
5896 */
5897 (void) spa_vdev_exit(spa, vd, txg, 0);
5898
5899 mutex_enter(&spa_namespace_lock);
5900 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5901 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5902 mutex_exit(&spa_namespace_lock);
5903
5904 return (0);
5905}
5906
5907/*
5908 * Attach a device to a mirror. The arguments are the path to any device
5909 * in the mirror, and the nvroot for the new device. If the path specifies
5910 * a device that is not mirrored, we automatically insert the mirror vdev.
5911 *
5912 * If 'replacing' is specified, the new device is intended to replace the
5913 * existing device; in this case the two devices are made into their own
5914 * mirror using the 'replacing' vdev, which is functionally identical to
5915 * the mirror vdev (it actually reuses all the same ops) but has a few
5916 * extra rules: you can't attach to it after it's been created, and upon
5917 * completion of resilvering, the first disk (the one being replaced)
5918 * is automatically detached.
5919 */
5920int
5921spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5922{
5923 uint64_t txg, dtl_max_txg;
5924 vdev_t *rvd = spa->spa_root_vdev;
5925 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5926 vdev_ops_t *pvops;
5927 char *oldvdpath, *newvdpath;
5928 int newvd_isspare;
5929 int error;
5930
5931 ASSERT(spa_writeable(spa));
5932
5933 txg = spa_vdev_enter(spa);
5934
5935 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5936
5937 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5938 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5939 error = (spa_has_checkpoint(spa)) ?
5940 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5941 return (spa_vdev_exit(spa, NULL, txg, error));
5942 }
5943
5944 if (spa->spa_vdev_removal != NULL ||
5945 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5946 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5947 }
5948
5949 if (oldvd == NULL)
5950 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5951
5952 if (!oldvd->vdev_ops->vdev_op_leaf)
5953 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5954
5955 pvd = oldvd->vdev_parent;
5956
5957 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
5958 VDEV_ALLOC_ATTACH)) != 0)
5959 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5960
5961 if (newrootvd->vdev_children != 1)
5962 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5963
5964 newvd = newrootvd->vdev_child[0];
5965
5966 if (!newvd->vdev_ops->vdev_op_leaf)
5967 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5968
5969 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
5970 return (spa_vdev_exit(spa, newrootvd, txg, error));
5971
5972 /*
5973 * Spares can't replace logs
5974 */
5975 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
5976 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5977
5978 if (!replacing) {
5979 /*
5980 * For attach, the only allowable parent is a mirror or the root
5981 * vdev.
5982 */
5983 if (pvd->vdev_ops != &vdev_mirror_ops &&
5984 pvd->vdev_ops != &vdev_root_ops)
5985 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5986
5987 pvops = &vdev_mirror_ops;
5988 } else {
5989 /*
5990 * Active hot spares can only be replaced by inactive hot
5991 * spares.
5992 */
5993 if (pvd->vdev_ops == &vdev_spare_ops &&
5994 oldvd->vdev_isspare &&
5995 !spa_has_spare(spa, newvd->vdev_guid))
5996 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5997
5998 /*
5999 * If the source is a hot spare, and the parent isn't already a
6000 * spare, then we want to create a new hot spare. Otherwise, we
6001 * want to create a replacing vdev. The user is not allowed to
6002 * attach to a spared vdev child unless the 'isspare' state is
6003 * the same (spare replaces spare, non-spare replaces
6004 * non-spare).
6005 */
6006 if (pvd->vdev_ops == &vdev_replacing_ops &&
6007 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6008 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6009 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6010 newvd->vdev_isspare != oldvd->vdev_isspare) {
6011 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6012 }
6013
6014 if (newvd->vdev_isspare)
6015 pvops = &vdev_spare_ops;
6016 else
6017 pvops = &vdev_replacing_ops;
6018 }
6019
6020 /*
6021 * Make sure the new device is big enough.
6022 */
6023 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6024 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6025
6026 /*
6027 * The new device cannot have a higher alignment requirement
6028 * than the top-level vdev.
6029 */
6030 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6031 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6032
6033 /*
6034 * If this is an in-place replacement, update oldvd's path and devid
6035 * to make it distinguishable from newvd, and unopenable from now on.
6036 */
6037 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6038 spa_strfree(oldvd->vdev_path);
6039 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6040 KM_SLEEP);
6041 (void) sprintf(oldvd->vdev_path, "%s/%s",
6042 newvd->vdev_path, "old");
6043 if (oldvd->vdev_devid != NULL) {
6044 spa_strfree(oldvd->vdev_devid);
6045 oldvd->vdev_devid = NULL;
6046 }
6047 }
6048
6049 /* mark the device being resilvered */
6050 newvd->vdev_resilver_txg = txg;
6051
6052 /*
6053 * If the parent is not a mirror, or if we're replacing, insert the new
6054 * mirror/replacing/spare vdev above oldvd.
6055 */
6056 if (pvd->vdev_ops != pvops)
6057 pvd = vdev_add_parent(oldvd, pvops);
6058
6059 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6060 ASSERT(pvd->vdev_ops == pvops);
6061 ASSERT(oldvd->vdev_parent == pvd);
6062
6063 /*
6064 * Extract the new device from its root and add it to pvd.
6065 */
6066 vdev_remove_child(newrootvd, newvd);
6067 newvd->vdev_id = pvd->vdev_children;
6068 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6069 vdev_add_child(pvd, newvd);
6070
6071 tvd = newvd->vdev_top;
6072 ASSERT(pvd->vdev_top == tvd);
6073 ASSERT(tvd->vdev_parent == rvd);
6074
6075 vdev_config_dirty(tvd);
6076
6077 /*
6078 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6079 * for any dmu_sync-ed blocks. It will propagate upward when
6080 * spa_vdev_exit() calls vdev_dtl_reassess().
6081 */
6082 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6083
6084 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6085 dtl_max_txg - TXG_INITIAL);
6086
6087 if (newvd->vdev_isspare) {
6088 spa_spare_activate(newvd);
6089 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6090 }
6091
6092 oldvdpath = spa_strdup(oldvd->vdev_path);
6093 newvdpath = spa_strdup(newvd->vdev_path);
6094 newvd_isspare = newvd->vdev_isspare;
6095
6096 /*
6097 * Mark newvd's DTL dirty in this txg.
6098 */
6099 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6100
6101 /*
6102 * Schedule the resilver to restart in the future. We do this to
6103 * ensure that dmu_sync-ed blocks have been stitched into the
6104 * respective datasets.
6105 */
6106 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6107
6108 if (spa->spa_bootfs)
6109 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6110
6111 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6112
6113 /*
6114 * Commit the config
6115 */
6116 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6117
6118 spa_history_log_internal(spa, "vdev attach", NULL,
6119 "%s vdev=%s %s vdev=%s",
6120 replacing && newvd_isspare ? "spare in" :
6121 replacing ? "replace" : "attach", newvdpath,
6122 replacing ? "for" : "to", oldvdpath);
6123
6124 spa_strfree(oldvdpath);
6125 spa_strfree(newvdpath);
6126
6127 return (0);
6128}
6129
6130/*
6131 * Detach a device from a mirror or replacing vdev.
6132 *
6133 * If 'replace_done' is specified, only detach if the parent
6134 * is a replacing vdev.
6135 */
6136int
6137spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6138{
6139 uint64_t txg;
6140 int error;
6141 vdev_t *rvd = spa->spa_root_vdev;
6142 vdev_t *vd, *pvd, *cvd, *tvd;
6143 boolean_t unspare = B_FALSE;
6144 uint64_t unspare_guid = 0;
6145 char *vdpath;
6146
6147 ASSERT(spa_writeable(spa));
6148
6149 txg = spa_vdev_enter(spa);
6150
6151 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6152
6153 /*
6154 * Besides being called directly from the userland through the
6155 * ioctl interface, spa_vdev_detach() can be potentially called
6156 * at the end of spa_vdev_resilver_done().
6157 *
6158 * In the regular case, when we have a checkpoint this shouldn't
6159 * happen as we never empty the DTLs of a vdev during the scrub
6160 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6161 * should never get here when we have a checkpoint.
6162 *
6163 * That said, even in a case when we checkpoint the pool exactly
6164 * as spa_vdev_resilver_done() calls this function everything
6165 * should be fine as the resilver will return right away.
6166 */
6167 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6168 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6169 error = (spa_has_checkpoint(spa)) ?
6170 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6171 return (spa_vdev_exit(spa, NULL, txg, error));
6172 }
6173
6174 if (vd == NULL)
6175 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6176
6177 if (!vd->vdev_ops->vdev_op_leaf)
6178 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6179
6180 pvd = vd->vdev_parent;
6181
6182 /*
6183 * If the parent/child relationship is not as expected, don't do it.
6184 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6185 * vdev that's replacing B with C. The user's intent in replacing
6186 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6187 * the replace by detaching C, the expected behavior is to end up
6188 * M(A,B). But suppose that right after deciding to detach C,
6189 * the replacement of B completes. We would have M(A,C), and then
6190 * ask to detach C, which would leave us with just A -- not what
6191 * the user wanted. To prevent this, we make sure that the
6192 * parent/child relationship hasn't changed -- in this example,
6193 * that C's parent is still the replacing vdev R.
6194 */
6195 if (pvd->vdev_guid != pguid && pguid != 0)
6196 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6197
6198 /*
6199 * Only 'replacing' or 'spare' vdevs can be replaced.
6200 */
6201 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6202 pvd->vdev_ops != &vdev_spare_ops)
6203 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6204
6205 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6206 spa_version(spa) >= SPA_VERSION_SPARES);
6207
6208 /*
6209 * Only mirror, replacing, and spare vdevs support detach.
6210 */
6211 if (pvd->vdev_ops != &vdev_replacing_ops &&
6212 pvd->vdev_ops != &vdev_mirror_ops &&
6213 pvd->vdev_ops != &vdev_spare_ops)
6214 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6215
6216 /*
6217 * If this device has the only valid copy of some data,
6218 * we cannot safely detach it.
6219 */
6220 if (vdev_dtl_required(vd))
6221 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6222
6223 ASSERT(pvd->vdev_children >= 2);
6224
6225 /*
6226 * If we are detaching the second disk from a replacing vdev, then
6227 * check to see if we changed the original vdev's path to have "/old"
6228 * at the end in spa_vdev_attach(). If so, undo that change now.
6229 */
6230 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6231 vd->vdev_path != NULL) {
6232 size_t len = strlen(vd->vdev_path);
6233
6234 for (int c = 0; c < pvd->vdev_children; c++) {
6235 cvd = pvd->vdev_child[c];
6236
6237 if (cvd == vd || cvd->vdev_path == NULL)
6238 continue;
6239
6240 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6241 strcmp(cvd->vdev_path + len, "/old") == 0) {
6242 spa_strfree(cvd->vdev_path);
6243 cvd->vdev_path = spa_strdup(vd->vdev_path);
6244 break;
6245 }
6246 }
6247 }
6248
6249 /*
6250 * If we are detaching the original disk from a spare, then it implies
6251 * that the spare should become a real disk, and be removed from the
6252 * active spare list for the pool.
6253 */
6254 if (pvd->vdev_ops == &vdev_spare_ops &&
6255 vd->vdev_id == 0 &&
6256 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6257 unspare = B_TRUE;
6258
6259 /*
6260 * Erase the disk labels so the disk can be used for other things.
6261 * This must be done after all other error cases are handled,
6262 * but before we disembowel vd (so we can still do I/O to it).
6263 * But if we can't do it, don't treat the error as fatal --
6264 * it may be that the unwritability of the disk is the reason
6265 * it's being detached!
6266 */
6267 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6268
6269 /*
6270 * Remove vd from its parent and compact the parent's children.
6271 */
6272 vdev_remove_child(pvd, vd);
6273 vdev_compact_children(pvd);
6274
6275 /*
6276 * Remember one of the remaining children so we can get tvd below.
6277 */
6278 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6279
6280 /*
6281 * If we need to remove the remaining child from the list of hot spares,
6282 * do it now, marking the vdev as no longer a spare in the process.
6283 * We must do this before vdev_remove_parent(), because that can
6284 * change the GUID if it creates a new toplevel GUID. For a similar
6285 * reason, we must remove the spare now, in the same txg as the detach;
6286 * otherwise someone could attach a new sibling, change the GUID, and
6287 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6288 */
6289 if (unspare) {
6290 ASSERT(cvd->vdev_isspare);
6291 spa_spare_remove(cvd);
6292 unspare_guid = cvd->vdev_guid;
6293 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6294 cvd->vdev_unspare = B_TRUE;
6295 }
6296
6297 /*
6298 * If the parent mirror/replacing vdev only has one child,
6299 * the parent is no longer needed. Remove it from the tree.
6300 */
6301 if (pvd->vdev_children == 1) {
6302 if (pvd->vdev_ops == &vdev_spare_ops)
6303 cvd->vdev_unspare = B_FALSE;
6304 vdev_remove_parent(cvd);
6305 }
6306
6307
6308 /*
6309 * We don't set tvd until now because the parent we just removed
6310 * may have been the previous top-level vdev.
6311 */
6312 tvd = cvd->vdev_top;
6313 ASSERT(tvd->vdev_parent == rvd);
6314
6315 /*
6316 * Reevaluate the parent vdev state.
6317 */
6318 vdev_propagate_state(cvd);
6319
6320 /*
6321 * If the 'autoexpand' property is set on the pool then automatically
6322 * try to expand the size of the pool. For example if the device we
6323 * just detached was smaller than the others, it may be possible to
6324 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6325 * first so that we can obtain the updated sizes of the leaf vdevs.
6326 */
6327 if (spa->spa_autoexpand) {
6328 vdev_reopen(tvd);
6329 vdev_expand(tvd, txg);
6330 }
6331
6332 vdev_config_dirty(tvd);
6333
6334 /*
6335 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6336 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6337 * But first make sure we're not on any *other* txg's DTL list, to
6338 * prevent vd from being accessed after it's freed.
6339 */
6340 vdpath = spa_strdup(vd->vdev_path);
6341 for (int t = 0; t < TXG_SIZE; t++)
6342 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6343 vd->vdev_detached = B_TRUE;
6344 vdev_dirty(tvd, VDD_DTL, vd, txg);
6345
6346 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6347
6348 /* hang on to the spa before we release the lock */
6349 spa_open_ref(spa, FTAG);
6350
6351 error = spa_vdev_exit(spa, vd, txg, 0);
6352
6353 spa_history_log_internal(spa, "detach", NULL,
6354 "vdev=%s", vdpath);
6355 spa_strfree(vdpath);
6356
6357 /*
6358 * If this was the removal of the original device in a hot spare vdev,
6359 * then we want to go through and remove the device from the hot spare
6360 * list of every other pool.
6361 */
6362 if (unspare) {
6363 spa_t *altspa = NULL;
6364
6365 mutex_enter(&spa_namespace_lock);
6366 while ((altspa = spa_next(altspa)) != NULL) {
6367 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6368 altspa == spa)
6369 continue;
6370
6371 spa_open_ref(altspa, FTAG);
6372 mutex_exit(&spa_namespace_lock);
6373 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6374 mutex_enter(&spa_namespace_lock);
6375 spa_close(altspa, FTAG);
6376 }
6377 mutex_exit(&spa_namespace_lock);
6378
6379 /* search the rest of the vdevs for spares to remove */
6380 spa_vdev_resilver_done(spa);
6381 }
6382
6383 /* all done with the spa; OK to release */
6384 mutex_enter(&spa_namespace_lock);
6385 spa_close(spa, FTAG);
6386 mutex_exit(&spa_namespace_lock);
6387
6388 return (error);
6389}
6390
6391/*
6392 * Split a set of devices from their mirrors, and create a new pool from them.
6393 */
6394int
6395spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6396 nvlist_t *props, boolean_t exp)
6397{
6398 int error = 0;
6399 uint64_t txg, *glist;
6400 spa_t *newspa;
6401 uint_t c, children, lastlog;
6402 nvlist_t **child, *nvl, *tmp;
6403 dmu_tx_t *tx;
6404 char *altroot = NULL;
6405 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6406 boolean_t activate_slog;
6407
6408 ASSERT(spa_writeable(spa));
6409
6410 txg = spa_vdev_enter(spa);
6411
6412 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6413 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6414 error = (spa_has_checkpoint(spa)) ?
6415 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6416 return (spa_vdev_exit(spa, NULL, txg, error));
6417 }
6418
6419 /* clear the log and flush everything up to now */
6420 activate_slog = spa_passivate_log(spa);
6421 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6422 error = spa_reset_logs(spa);
6423 txg = spa_vdev_config_enter(spa);
6424
6425 if (activate_slog)
6426 spa_activate_log(spa);
6427
6428 if (error != 0)
6429 return (spa_vdev_exit(spa, NULL, txg, error));
6430
6431 /* check new spa name before going any further */
6432 if (spa_lookup(newname) != NULL)
6433 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6434
6435 /*
6436 * scan through all the children to ensure they're all mirrors
6437 */
6438 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6439 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6440 &children) != 0)
6441 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6442
6443 /* first, check to ensure we've got the right child count */
6444 rvd = spa->spa_root_vdev;
6445 lastlog = 0;
6446 for (c = 0; c < rvd->vdev_children; c++) {
6447 vdev_t *vd = rvd->vdev_child[c];
6448
6449 /* don't count the holes & logs as children */
6450 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6451 if (lastlog == 0)
6452 lastlog = c;
6453 continue;
6454 }
6455
6456 lastlog = 0;
6457 }
6458 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6459 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6460
6461 /* next, ensure no spare or cache devices are part of the split */
6462 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6463 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6464 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6465
6466 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6467 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6468
6469 /* then, loop over each vdev and validate it */
6470 for (c = 0; c < children; c++) {
6471 uint64_t is_hole = 0;
6472
6473 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6474 &is_hole);
6475
6476 if (is_hole != 0) {
6477 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6478 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6479 continue;
6480 } else {
6481 error = SET_ERROR(EINVAL);
6482 break;
6483 }
6484 }
6485
6486 /* which disk is going to be split? */
6487 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6488 &glist[c]) != 0) {
6489 error = SET_ERROR(EINVAL);
6490 break;
6491 }
6492
6493 /* look it up in the spa */
6494 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6495 if (vml[c] == NULL) {
6496 error = SET_ERROR(ENODEV);
6497 break;
6498 }
6499
6500 /* make sure there's nothing stopping the split */
6501 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6502 vml[c]->vdev_islog ||
6503 !vdev_is_concrete(vml[c]) ||
6504 vml[c]->vdev_isspare ||
6505 vml[c]->vdev_isl2cache ||
6506 !vdev_writeable(vml[c]) ||
6507 vml[c]->vdev_children != 0 ||
6508 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6509 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6510 error = SET_ERROR(EINVAL);
6511 break;
6512 }
6513
6514 if (vdev_dtl_required(vml[c])) {
6515 error = SET_ERROR(EBUSY);
6516 break;
6517 }
6518
6519 /* we need certain info from the top level */
6520 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6521 vml[c]->vdev_top->vdev_ms_array) == 0);
6522 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6523 vml[c]->vdev_top->vdev_ms_shift) == 0);
6524 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6525 vml[c]->vdev_top->vdev_asize) == 0);
6526 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6527 vml[c]->vdev_top->vdev_ashift) == 0);
6528
6529 /* transfer per-vdev ZAPs */
6530 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6531 VERIFY0(nvlist_add_uint64(child[c],
6532 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6533
6534 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6535 VERIFY0(nvlist_add_uint64(child[c],
6536 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6537 vml[c]->vdev_parent->vdev_top_zap));
6538 }
6539
6540 if (error != 0) {
6541 kmem_free(vml, children * sizeof (vdev_t *));
6542 kmem_free(glist, children * sizeof (uint64_t));
6543 return (spa_vdev_exit(spa, NULL, txg, error));
6544 }
6545
6546 /* stop writers from using the disks */
6547 for (c = 0; c < children; c++) {
6548 if (vml[c] != NULL)
6549 vml[c]->vdev_offline = B_TRUE;
6550 }
6551 vdev_reopen(spa->spa_root_vdev);
6552
6553 /*
6554 * Temporarily record the splitting vdevs in the spa config. This
6555 * will disappear once the config is regenerated.
6556 */
6557 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6558 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6559 glist, children) == 0);
6560 kmem_free(glist, children * sizeof (uint64_t));
6561
6562 mutex_enter(&spa->spa_props_lock);
6563 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6564 nvl) == 0);
6565 mutex_exit(&spa->spa_props_lock);
6566 spa->spa_config_splitting = nvl;
6567 vdev_config_dirty(spa->spa_root_vdev);
6568
6569 /* configure and create the new pool */
6570 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6571 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6572 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6573 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6574 spa_version(spa)) == 0);
6575 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6576 spa->spa_config_txg) == 0);
6577 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6578 spa_generate_guid(NULL)) == 0);
6579 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6580 (void) nvlist_lookup_string(props,
6581 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6582
6583 /* add the new pool to the namespace */
6584 newspa = spa_add(newname, config, altroot);
6585 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6586 newspa->spa_config_txg = spa->spa_config_txg;
6587 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6588
6589 /* release the spa config lock, retaining the namespace lock */
6590 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6591
6592 if (zio_injection_enabled)
6593 zio_handle_panic_injection(spa, FTAG, 1);
6594
6595 spa_activate(newspa, spa_mode_global);
6596 spa_async_suspend(newspa);
6597
6598#ifndef illumos
6599 /* mark that we are creating new spa by splitting */
6600 newspa->spa_splitting_newspa = B_TRUE;
6601#endif
6602 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6603
6604 /* create the new pool from the disks of the original pool */
6605 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6606#ifndef illumos
6607 newspa->spa_splitting_newspa = B_FALSE;
6608#endif
6609 if (error)
6610 goto out;
6611
6612 /* if that worked, generate a real config for the new pool */
6613 if (newspa->spa_root_vdev != NULL) {
6614 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6615 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6616 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6617 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6618 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6619 B_TRUE));
6620 }
6621
6622 /* set the props */
6623 if (props != NULL) {
6624 spa_configfile_set(newspa, props, B_FALSE);
6625 error = spa_prop_set(newspa, props);
6626 if (error)
6627 goto out;
6628 }
6629
6630 /* flush everything */
6631 txg = spa_vdev_config_enter(newspa);
6632 vdev_config_dirty(newspa->spa_root_vdev);
6633 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6634
6635 if (zio_injection_enabled)
6636 zio_handle_panic_injection(spa, FTAG, 2);
6637
6638 spa_async_resume(newspa);
6639
6640 /* finally, update the original pool's config */
6641 txg = spa_vdev_config_enter(spa);
6642 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6643 error = dmu_tx_assign(tx, TXG_WAIT);
6644 if (error != 0)
6645 dmu_tx_abort(tx);
6646 for (c = 0; c < children; c++) {
6647 if (vml[c] != NULL) {
6648 vdev_split(vml[c]);
6649 if (error == 0)
6650 spa_history_log_internal(spa, "detach", tx,
6651 "vdev=%s", vml[c]->vdev_path);
6652
6653 vdev_free(vml[c]);
6654 }
6655 }
6656 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6657 vdev_config_dirty(spa->spa_root_vdev);
6658 spa->spa_config_splitting = NULL;
6659 nvlist_free(nvl);
6660 if (error == 0)
6661 dmu_tx_commit(tx);
6662 (void) spa_vdev_exit(spa, NULL, txg, 0);
6663
6664 if (zio_injection_enabled)
6665 zio_handle_panic_injection(spa, FTAG, 3);
6666
6667 /* split is complete; log a history record */
6668 spa_history_log_internal(newspa, "split", NULL,
6669 "from pool %s", spa_name(spa));
6670
6671 kmem_free(vml, children * sizeof (vdev_t *));
6672
6673 /* if we're not going to mount the filesystems in userland, export */
6674 if (exp)
6675 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6676 B_FALSE, B_FALSE);
6677
6678 return (error);
6679
6680out:
6681 spa_unload(newspa);
6682 spa_deactivate(newspa);
6683 spa_remove(newspa);
6684
6685 txg = spa_vdev_config_enter(spa);
6686
6687 /* re-online all offlined disks */
6688 for (c = 0; c < children; c++) {
6689 if (vml[c] != NULL)
6690 vml[c]->vdev_offline = B_FALSE;
6691 }
6692 vdev_reopen(spa->spa_root_vdev);
6693
6694 nvlist_free(spa->spa_config_splitting);
6695 spa->spa_config_splitting = NULL;
6696 (void) spa_vdev_exit(spa, NULL, txg, error);
6697
6698 kmem_free(vml, children * sizeof (vdev_t *));
6699 return (error);
6700}
6701
6702/*
6703 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6704 * currently spared, so we can detach it.
6705 */
6706static vdev_t *
6707spa_vdev_resilver_done_hunt(vdev_t *vd)
6708{
6709 vdev_t *newvd, *oldvd;
6710
6711 for (int c = 0; c < vd->vdev_children; c++) {
6712 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6713 if (oldvd != NULL)
6714 return (oldvd);
6715 }
6716
6717 /*
6718 * Check for a completed replacement. We always consider the first
6719 * vdev in the list to be the oldest vdev, and the last one to be
6720 * the newest (see spa_vdev_attach() for how that works). In
6721 * the case where the newest vdev is faulted, we will not automatically
6722 * remove it after a resilver completes. This is OK as it will require
6723 * user intervention to determine which disk the admin wishes to keep.
6724 */
6725 if (vd->vdev_ops == &vdev_replacing_ops) {
6726 ASSERT(vd->vdev_children > 1);
6727
6728 newvd = vd->vdev_child[vd->vdev_children - 1];
6729 oldvd = vd->vdev_child[0];
6730
6731 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6732 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6733 !vdev_dtl_required(oldvd))
6734 return (oldvd);
6735 }
6736
6737 /*
6738 * Check for a completed resilver with the 'unspare' flag set.
6739 */
6740 if (vd->vdev_ops == &vdev_spare_ops) {
6741 vdev_t *first = vd->vdev_child[0];
6742 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6743
6744 if (last->vdev_unspare) {
6745 oldvd = first;
6746 newvd = last;
6747 } else if (first->vdev_unspare) {
6748 oldvd = last;
6749 newvd = first;
6750 } else {
6751 oldvd = NULL;
6752 }
6753
6754 if (oldvd != NULL &&
6755 vdev_dtl_empty(newvd, DTL_MISSING) &&
6756 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6757 !vdev_dtl_required(oldvd))
6758 return (oldvd);
6759
6760 /*
6761 * If there are more than two spares attached to a disk,
6762 * and those spares are not required, then we want to
6763 * attempt to free them up now so that they can be used
6764 * by other pools. Once we're back down to a single
6765 * disk+spare, we stop removing them.
6766 */
6767 if (vd->vdev_children > 2) {
6768 newvd = vd->vdev_child[1];
6769
6770 if (newvd->vdev_isspare && last->vdev_isspare &&
6771 vdev_dtl_empty(last, DTL_MISSING) &&
6772 vdev_dtl_empty(last, DTL_OUTAGE) &&
6773 !vdev_dtl_required(newvd))
6774 return (newvd);
6775 }
6776 }
6777
6778 return (NULL);
6779}
6780
6781static void
6782spa_vdev_resilver_done(spa_t *spa)
6783{
6784 vdev_t *vd, *pvd, *ppvd;
6785 uint64_t guid, sguid, pguid, ppguid;
6786
6787 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6788
6789 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6790 pvd = vd->vdev_parent;
6791 ppvd = pvd->vdev_parent;
6792 guid = vd->vdev_guid;
6793 pguid = pvd->vdev_guid;
6794 ppguid = ppvd->vdev_guid;
6795 sguid = 0;
6796 /*
6797 * If we have just finished replacing a hot spared device, then
6798 * we need to detach the parent's first child (the original hot
6799 * spare) as well.
6800 */
6801 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6802 ppvd->vdev_children == 2) {
6803 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6804 sguid = ppvd->vdev_child[1]->vdev_guid;
6805 }
6806 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6807
6808 spa_config_exit(spa, SCL_ALL, FTAG);
6809 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6810 return;
6811 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6812 return;
6813 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6814 }
6815
6816 spa_config_exit(spa, SCL_ALL, FTAG);
6817}
6818
6819/*
6820 * Update the stored path or FRU for this vdev.
6821 */
6822int
6823spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
6824 boolean_t ispath)
6825{
6826 vdev_t *vd;
6827 boolean_t sync = B_FALSE;
6828
6829 ASSERT(spa_writeable(spa));
6830
6831 spa_vdev_state_enter(spa, SCL_ALL);
6832
6833 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
6834 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6835
6836 if (!vd->vdev_ops->vdev_op_leaf)
6837 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6838
6839 if (ispath) {
6840 if (strcmp(value, vd->vdev_path) != 0) {
6841 spa_strfree(vd->vdev_path);
6842 vd->vdev_path = spa_strdup(value);
6843 sync = B_TRUE;
6844 }
6845 } else {
6846 if (vd->vdev_fru == NULL) {
6847 vd->vdev_fru = spa_strdup(value);
6848 sync = B_TRUE;
6849 } else if (strcmp(value, vd->vdev_fru) != 0) {
6850 spa_strfree(vd->vdev_fru);
6851 vd->vdev_fru = spa_strdup(value);
6852 sync = B_TRUE;
6853 }
6854 }
6855
6856 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6857}
6858
6859int
6860spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6861{
6862 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6863}
6864
6865int
6866spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6867{
6868 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6869}
6870
6871/*
6872 * ==========================================================================
6873 * SPA Scanning
6874 * ==========================================================================
6875 */
6876int
6877spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
6878{
6879 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6880
6881 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6882 return (SET_ERROR(EBUSY));
6883
6884 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
6885}
6886
6887int
6888spa_scan_stop(spa_t *spa)
6889{
6890 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6891 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6892 return (SET_ERROR(EBUSY));
6893 return (dsl_scan_cancel(spa->spa_dsl_pool));
6894}
6895
6896int
6897spa_scan(spa_t *spa, pool_scan_func_t func)
6898{
6899 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6900
6901 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6902 return (SET_ERROR(ENOTSUP));
6903
6904 /*
6905 * If a resilver was requested, but there is no DTL on a
6906 * writeable leaf device, we have nothing to do.
6907 */
6908 if (func == POOL_SCAN_RESILVER &&
6909 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6910 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6911 return (0);
6912 }
6913
6914 return (dsl_scan(spa->spa_dsl_pool, func));
6915}
6916
6917/*
6918 * ==========================================================================
6919 * SPA async task processing
6920 * ==========================================================================
6921 */
6922
6923static void
6924spa_async_remove(spa_t *spa, vdev_t *vd)
6925{
6926 if (vd->vdev_remove_wanted) {
6927 vd->vdev_remove_wanted = B_FALSE;
6928 vd->vdev_delayed_close = B_FALSE;
6929 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6930
6931 /*
6932 * We want to clear the stats, but we don't want to do a full
6933 * vdev_clear() as that will cause us to throw away
6934 * degraded/faulted state as well as attempt to reopen the
6935 * device, all of which is a waste.
6936 */
6937 vd->vdev_stat.vs_read_errors = 0;
6938 vd->vdev_stat.vs_write_errors = 0;
6939 vd->vdev_stat.vs_checksum_errors = 0;
6940
6941 vdev_state_dirty(vd->vdev_top);
6942 /* Tell userspace that the vdev is gone. */
6943 zfs_post_remove(spa, vd);
6944 }
6945
6946 for (int c = 0; c < vd->vdev_children; c++)
6947 spa_async_remove(spa, vd->vdev_child[c]);
6948}
6949
6950static void
6951spa_async_probe(spa_t *spa, vdev_t *vd)
6952{
6953 if (vd->vdev_probe_wanted) {
6954 vd->vdev_probe_wanted = B_FALSE;
6955 vdev_reopen(vd); /* vdev_open() does the actual probe */
6956 }
6957
6958 for (int c = 0; c < vd->vdev_children; c++)
6959 spa_async_probe(spa, vd->vdev_child[c]);
6960}
6961
6962static void
6963spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6964{
6965 sysevent_id_t eid;
6966 nvlist_t *attr;
6967 char *physpath;
6968
6969 if (!spa->spa_autoexpand)
6970 return;
6971
6972 for (int c = 0; c < vd->vdev_children; c++) {
6973 vdev_t *cvd = vd->vdev_child[c];
6974 spa_async_autoexpand(spa, cvd);
6975 }
6976
6977 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6978 return;
6979
6980 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6981 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6982
6983 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6984 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6985
6986 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6987 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6988
6989 nvlist_free(attr);
6990 kmem_free(physpath, MAXPATHLEN);
6991}
6992
6993static void
6994spa_async_thread(void *arg)
6995{
6996 spa_t *spa = (spa_t *)arg;
6997 int tasks;
6998
6999 ASSERT(spa->spa_sync_on);
7000
7001 mutex_enter(&spa->spa_async_lock);
7002 tasks = spa->spa_async_tasks;
7003 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
7004 mutex_exit(&spa->spa_async_lock);
7005
7006 /*
7007 * See if the config needs to be updated.
7008 */
7009 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7010 uint64_t old_space, new_space;
7011
7012 mutex_enter(&spa_namespace_lock);
7013 old_space = metaslab_class_get_space(spa_normal_class(spa));
7014 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7015 new_space = metaslab_class_get_space(spa_normal_class(spa));
7016 mutex_exit(&spa_namespace_lock);
7017
7018 /*
7019 * If the pool grew as a result of the config update,
7020 * then log an internal history event.
7021 */
7022 if (new_space != old_space) {
7023 spa_history_log_internal(spa, "vdev online", NULL,
7024 "pool '%s' size: %llu(+%llu)",
7025 spa_name(spa), new_space, new_space - old_space);
7026 }
7027 }
7028
7029 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7030 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7031 spa_async_autoexpand(spa, spa->spa_root_vdev);
7032 spa_config_exit(spa, SCL_CONFIG, FTAG);
7033 }
7034
7035 /*
7036 * See if any devices need to be probed.
7037 */
7038 if (tasks & SPA_ASYNC_PROBE) {
7039 spa_vdev_state_enter(spa, SCL_NONE);
7040 spa_async_probe(spa, spa->spa_root_vdev);
7041 (void) spa_vdev_state_exit(spa, NULL, 0);
7042 }
7043
7044 /*
7045 * If any devices are done replacing, detach them.
7046 */
7047 if (tasks & SPA_ASYNC_RESILVER_DONE)
7048 spa_vdev_resilver_done(spa);
7049
7050 /*
7051 * Kick off a resilver.
7052 */
7053 if (tasks & SPA_ASYNC_RESILVER)
7054 dsl_resilver_restart(spa->spa_dsl_pool, 0);
7055
7056 /*
7057 * Let the world know that we're done.
7058 */
7059 mutex_enter(&spa->spa_async_lock);
7060 spa->spa_async_thread = NULL;
7061 cv_broadcast(&spa->spa_async_cv);
7062 mutex_exit(&spa->spa_async_lock);
7063 thread_exit();
7064}
7065
7066static void
7067spa_async_thread_vd(void *arg)
7068{
7069 spa_t *spa = arg;
7070 int tasks;
7071
7072 mutex_enter(&spa->spa_async_lock);
7073 tasks = spa->spa_async_tasks;
7074retry:
7075 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
7076 mutex_exit(&spa->spa_async_lock);
7077
7078 /*
7079 * See if any devices need to be marked REMOVED.
7080 */
7081 if (tasks & SPA_ASYNC_REMOVE) {
7082 spa_vdev_state_enter(spa, SCL_NONE);
7083 spa_async_remove(spa, spa->spa_root_vdev);
7084 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7085 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7086 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7087 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7088 (void) spa_vdev_state_exit(spa, NULL, 0);
7089 }
7090
7091 /*
7092 * Let the world know that we're done.
7093 */
7094 mutex_enter(&spa->spa_async_lock);
7095 tasks = spa->spa_async_tasks;
7096 if ((tasks & SPA_ASYNC_REMOVE) != 0)
7097 goto retry;
7098 spa->spa_async_thread_vd = NULL;
7099 cv_broadcast(&spa->spa_async_cv);
7100 mutex_exit(&spa->spa_async_lock);
7101 thread_exit();
7102}
7103
7104void
7105spa_async_suspend(spa_t *spa)
7106{
7107 mutex_enter(&spa->spa_async_lock);
7108 spa->spa_async_suspended++;
7109 while (spa->spa_async_thread != NULL ||
7110 spa->spa_async_thread_vd != NULL)
7111 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
7112 mutex_exit(&spa->spa_async_lock);
7113
7114 spa_vdev_remove_suspend(spa);
7115
7116 zthr_t *condense_thread = spa->spa_condense_zthr;
7117 if (condense_thread != NULL && zthr_isrunning(condense_thread))
7118 VERIFY0(zthr_cancel(condense_thread));
7119
7120 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7121 if (discard_thread != NULL && zthr_isrunning(discard_thread))
7122 VERIFY0(zthr_cancel(discard_thread));
7123}
7124
7125void
7126spa_async_resume(spa_t *spa)
7127{
7128 mutex_enter(&spa->spa_async_lock);
7129 ASSERT(spa->spa_async_suspended != 0);
7130 spa->spa_async_suspended--;
7131 mutex_exit(&spa->spa_async_lock);
7132 spa_restart_removal(spa);
7133
7134 zthr_t *condense_thread = spa->spa_condense_zthr;
7135 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
7136 zthr_resume(condense_thread);
7137
7138 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7139 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
7140 zthr_resume(discard_thread);
7141}
7142
7143static boolean_t
7144spa_async_tasks_pending(spa_t *spa)
7145{
7146 uint_t non_config_tasks;
7147 uint_t config_task;
7148 boolean_t config_task_suspended;
7149
7150 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
7151 SPA_ASYNC_REMOVE);
7152 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
7153 if (spa->spa_ccw_fail_time == 0) {
7154 config_task_suspended = B_FALSE;
7155 } else {
7156 config_task_suspended =
7157 (gethrtime() - spa->spa_ccw_fail_time) <
7158 (zfs_ccw_retry_interval * NANOSEC);
7159 }
7160
7161 return (non_config_tasks || (config_task && !config_task_suspended));
7162}
7163
7164static void
7165spa_async_dispatch(spa_t *spa)
7166{
7167 mutex_enter(&spa->spa_async_lock);
7168 if (spa_async_tasks_pending(spa) &&
7169 !spa->spa_async_suspended &&
7170 spa->spa_async_thread == NULL &&
7171 rootdir != NULL)
7172 spa->spa_async_thread = thread_create(NULL, 0,
7173 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
7174 mutex_exit(&spa->spa_async_lock);
7175}
7176
7177static void
7178spa_async_dispatch_vd(spa_t *spa)
7179{
7180 mutex_enter(&spa->spa_async_lock);
7181 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
7182 !spa->spa_async_suspended &&
7183 spa->spa_async_thread_vd == NULL &&
7184 rootdir != NULL)
7185 spa->spa_async_thread_vd = thread_create(NULL, 0,
7186 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
7187 mutex_exit(&spa->spa_async_lock);
7188}
7189
7190void
7191spa_async_request(spa_t *spa, int task)
7192{
7193 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
7194 mutex_enter(&spa->spa_async_lock);
7195 spa->spa_async_tasks |= task;
7196 mutex_exit(&spa->spa_async_lock);
7197 spa_async_dispatch_vd(spa);
7198}
7199
7200/*
7201 * ==========================================================================
7202 * SPA syncing routines
7203 * ==========================================================================
7204 */
7205
7206static int
7207bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7208{
7209 bpobj_t *bpo = arg;
7210 bpobj_enqueue(bpo, bp, tx);
7211 return (0);
7212}
7213
7214static int
7215spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7216{
7217 zio_t *zio = arg;
7218
7219 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7220 BP_GET_PSIZE(bp), zio->io_flags));
7221 return (0);
7222}
7223
7224/*
7225 * Note: this simple function is not inlined to make it easier to dtrace the
7226 * amount of time spent syncing frees.
7227 */
7228static void
7229spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7230{
7231 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7232 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7233 VERIFY(zio_wait(zio) == 0);
7234}
7235
7236/*
7237 * Note: this simple function is not inlined to make it easier to dtrace the
7238 * amount of time spent syncing deferred frees.
7239 */
7240static void
7241spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7242{
7243 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7244 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7245 spa_free_sync_cb, zio, tx), ==, 0);
7246 VERIFY0(zio_wait(zio));
7247}
7248
7249
7250static void
7251spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7252{
7253 char *packed = NULL;
7254 size_t bufsize;
7255 size_t nvsize = 0;
7256 dmu_buf_t *db;
7257
7258 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7259
7260 /*
7261 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7262 * information. This avoids the dmu_buf_will_dirty() path and
7263 * saves us a pre-read to get data we don't actually care about.
7264 */
7265 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7266 packed = kmem_alloc(bufsize, KM_SLEEP);
7267
7268 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7269 KM_SLEEP) == 0);
7270 bzero(packed + nvsize, bufsize - nvsize);
7271
7272 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7273
7274 kmem_free(packed, bufsize);
7275
7276 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7277 dmu_buf_will_dirty(db, tx);
7278 *(uint64_t *)db->db_data = nvsize;
7279 dmu_buf_rele(db, FTAG);
7280}
7281
7282static void
7283spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7284 const char *config, const char *entry)
7285{
7286 nvlist_t *nvroot;
7287 nvlist_t **list;
7288 int i;
7289
7290 if (!sav->sav_sync)
7291 return;
7292
7293 /*
7294 * Update the MOS nvlist describing the list of available devices.
7295 * spa_validate_aux() will have already made sure this nvlist is
7296 * valid and the vdevs are labeled appropriately.
7297 */
7298 if (sav->sav_object == 0) {
7299 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7300 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7301 sizeof (uint64_t), tx);
7302 VERIFY(zap_update(spa->spa_meta_objset,
7303 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7304 &sav->sav_object, tx) == 0);
7305 }
7306
7307 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7308 if (sav->sav_count == 0) {
7309 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7310 } else {
7311 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7312 for (i = 0; i < sav->sav_count; i++)
7313 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7314 B_FALSE, VDEV_CONFIG_L2CACHE);
7315 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7316 sav->sav_count) == 0);
7317 for (i = 0; i < sav->sav_count; i++)
7318 nvlist_free(list[i]);
7319 kmem_free(list, sav->sav_count * sizeof (void *));
7320 }
7321
7322 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7323 nvlist_free(nvroot);
7324
7325 sav->sav_sync = B_FALSE;
7326}
7327
7328/*
7329 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7330 * The all-vdev ZAP must be empty.
7331 */
7332static void
7333spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7334{
7335 spa_t *spa = vd->vdev_spa;
7336 if (vd->vdev_top_zap != 0) {
7337 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7338 vd->vdev_top_zap, tx));
7339 }
7340 if (vd->vdev_leaf_zap != 0) {
7341 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7342 vd->vdev_leaf_zap, tx));
7343 }
7344 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7345 spa_avz_build(vd->vdev_child[i], avz, tx);
7346 }
7347}
7348
7349static void
7350spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7351{
7352 nvlist_t *config;
7353
7354 /*
7355 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7356 * its config may not be dirty but we still need to build per-vdev ZAPs.
7357 * Similarly, if the pool is being assembled (e.g. after a split), we
7358 * need to rebuild the AVZ although the config may not be dirty.
7359 */
7360 if (list_is_empty(&spa->spa_config_dirty_list) &&
7361 spa->spa_avz_action == AVZ_ACTION_NONE)
7362 return;
7363
7364 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7365
7366 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7367 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7368 spa->spa_all_vdev_zaps != 0);
7369
7370 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7371 /* Make and build the new AVZ */
7372 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7373 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7374 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7375
7376 /* Diff old AVZ with new one */
7377 zap_cursor_t zc;
7378 zap_attribute_t za;
7379
7380 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7381 spa->spa_all_vdev_zaps);
7382 zap_cursor_retrieve(&zc, &za) == 0;
7383 zap_cursor_advance(&zc)) {
7384 uint64_t vdzap = za.za_first_integer;
7385 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7386 vdzap) == ENOENT) {
7387 /*
7388 * ZAP is listed in old AVZ but not in new one;
7389 * destroy it
7390 */
7391 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7392 tx));
7393 }
7394 }
7395
7396 zap_cursor_fini(&zc);
7397
7398 /* Destroy the old AVZ */
7399 VERIFY0(zap_destroy(spa->spa_meta_objset,
7400 spa->spa_all_vdev_zaps, tx));
7401
7402 /* Replace the old AVZ in the dir obj with the new one */
7403 VERIFY0(zap_update(spa->spa_meta_objset,
7404 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7405 sizeof (new_avz), 1, &new_avz, tx));
7406
7407 spa->spa_all_vdev_zaps = new_avz;
7408 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7409 zap_cursor_t zc;
7410 zap_attribute_t za;
7411
7412 /* Walk through the AVZ and destroy all listed ZAPs */
7413 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7414 spa->spa_all_vdev_zaps);
7415 zap_cursor_retrieve(&zc, &za) == 0;
7416 zap_cursor_advance(&zc)) {
7417 uint64_t zap = za.za_first_integer;
7418 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7419 }
7420
7421 zap_cursor_fini(&zc);
7422
7423 /* Destroy and unlink the AVZ itself */
7424 VERIFY0(zap_destroy(spa->spa_meta_objset,
7425 spa->spa_all_vdev_zaps, tx));
7426 VERIFY0(zap_remove(spa->spa_meta_objset,
7427 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7428 spa->spa_all_vdev_zaps = 0;
7429 }
7430
7431 if (spa->spa_all_vdev_zaps == 0) {
7432 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7433 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7434 DMU_POOL_VDEV_ZAP_MAP, tx);
7435 }
7436 spa->spa_avz_action = AVZ_ACTION_NONE;
7437
7438 /* Create ZAPs for vdevs that don't have them. */
7439 vdev_construct_zaps(spa->spa_root_vdev, tx);
7440
7441 config = spa_config_generate(spa, spa->spa_root_vdev,
7442 dmu_tx_get_txg(tx), B_FALSE);
7443
7444 /*
7445 * If we're upgrading the spa version then make sure that
7446 * the config object gets updated with the correct version.
7447 */
7448 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7449 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7450 spa->spa_uberblock.ub_version);
7451
7452 spa_config_exit(spa, SCL_STATE, FTAG);
7453
7454 nvlist_free(spa->spa_config_syncing);
7455 spa->spa_config_syncing = config;
7456
7457 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7458}
7459
7460static void
7461spa_sync_version(void *arg, dmu_tx_t *tx)
7462{
7463 uint64_t *versionp = arg;
7464 uint64_t version = *versionp;
7465 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7466
7467 /*
7468 * Setting the version is special cased when first creating the pool.
7469 */
7470 ASSERT(tx->tx_txg != TXG_INITIAL);
7471
7472 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7473 ASSERT(version >= spa_version(spa));
7474
7475 spa->spa_uberblock.ub_version = version;
7476 vdev_config_dirty(spa->spa_root_vdev);
7477 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7478}
7479
7480/*
7481 * Set zpool properties.
7482 */
7483static void
7484spa_sync_props(void *arg, dmu_tx_t *tx)
7485{
7486 nvlist_t *nvp = arg;
7487 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7488 objset_t *mos = spa->spa_meta_objset;
7489 nvpair_t *elem = NULL;
7490
7491 mutex_enter(&spa->spa_props_lock);
7492
7493 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7494 uint64_t intval;
7495 char *strval, *fname;
7496 zpool_prop_t prop;
7497 const char *propname;
7498 zprop_type_t proptype;
7499 spa_feature_t fid;
7500
7501 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7502 case ZPOOL_PROP_INVAL:
7503 /*
7504 * We checked this earlier in spa_prop_validate().
7505 */
7506 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7507
7508 fname = strchr(nvpair_name(elem), '@') + 1;
7509 VERIFY0(zfeature_lookup_name(fname, &fid));
7510
7511 spa_feature_enable(spa, fid, tx);
7512 spa_history_log_internal(spa, "set", tx,
7513 "%s=enabled", nvpair_name(elem));
7514 break;
7515
7516 case ZPOOL_PROP_VERSION:
7517 intval = fnvpair_value_uint64(elem);
7518 /*
7519 * The version is synced seperatly before other
7520 * properties and should be correct by now.
7521 */
7522 ASSERT3U(spa_version(spa), >=, intval);
7523 break;
7524
7525 case ZPOOL_PROP_ALTROOT:
7526 /*
7527 * 'altroot' is a non-persistent property. It should
7528 * have been set temporarily at creation or import time.
7529 */
7530 ASSERT(spa->spa_root != NULL);
7531 break;
7532
7533 case ZPOOL_PROP_READONLY:
7534 case ZPOOL_PROP_CACHEFILE:
7535 /*
7536 * 'readonly' and 'cachefile' are also non-persisitent
7537 * properties.
7538 */
7539 break;
7540 case ZPOOL_PROP_COMMENT:
7541 strval = fnvpair_value_string(elem);
7542 if (spa->spa_comment != NULL)
7543 spa_strfree(spa->spa_comment);
7544 spa->spa_comment = spa_strdup(strval);
7545 /*
7546 * We need to dirty the configuration on all the vdevs
7547 * so that their labels get updated. It's unnecessary
7548 * to do this for pool creation since the vdev's
7549 * configuratoin has already been dirtied.
7550 */
7551 if (tx->tx_txg != TXG_INITIAL)
7552 vdev_config_dirty(spa->spa_root_vdev);
7553 spa_history_log_internal(spa, "set", tx,
7554 "%s=%s", nvpair_name(elem), strval);
7555 break;
7556 default:
7557 /*
7558 * Set pool property values in the poolprops mos object.
7559 */
7560 if (spa->spa_pool_props_object == 0) {
7561 spa->spa_pool_props_object =
7562 zap_create_link(mos, DMU_OT_POOL_PROPS,
7563 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7564 tx);
7565 }
7566
7567 /* normalize the property name */
7568 propname = zpool_prop_to_name(prop);
7569 proptype = zpool_prop_get_type(prop);
7570
7571 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7572 ASSERT(proptype == PROP_TYPE_STRING);
7573 strval = fnvpair_value_string(elem);
7574 VERIFY0(zap_update(mos,
7575 spa->spa_pool_props_object, propname,
7576 1, strlen(strval) + 1, strval, tx));
7577 spa_history_log_internal(spa, "set", tx,
7578 "%s=%s", nvpair_name(elem), strval);
7579 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7580 intval = fnvpair_value_uint64(elem);
7581
7582 if (proptype == PROP_TYPE_INDEX) {
7583 const char *unused;
7584 VERIFY0(zpool_prop_index_to_string(
7585 prop, intval, &unused));
7586 }
7587 VERIFY0(zap_update(mos,
7588 spa->spa_pool_props_object, propname,
7589 8, 1, &intval, tx));
7590 spa_history_log_internal(spa, "set", tx,
7591 "%s=%lld", nvpair_name(elem), intval);
7592 } else {
7593 ASSERT(0); /* not allowed */
7594 }
7595
7596 switch (prop) {
7597 case ZPOOL_PROP_DELEGATION:
7598 spa->spa_delegation = intval;
7599 break;
7600 case ZPOOL_PROP_BOOTFS:
7601 spa->spa_bootfs = intval;
7602 break;
7603 case ZPOOL_PROP_FAILUREMODE:
7604 spa->spa_failmode = intval;
7605 break;
7606 case ZPOOL_PROP_AUTOEXPAND:
7607 spa->spa_autoexpand = intval;
7608 if (tx->tx_txg != TXG_INITIAL)
7609 spa_async_request(spa,
7610 SPA_ASYNC_AUTOEXPAND);
7611 break;
7612 case ZPOOL_PROP_DEDUPDITTO:
7613 spa->spa_dedup_ditto = intval;
7614 break;
7615 default:
7616 break;
7617 }
7618 }
7619
7620 }
7621
7622 mutex_exit(&spa->spa_props_lock);
7623}
7624
7625/*
7626 * Perform one-time upgrade on-disk changes. spa_version() does not
7627 * reflect the new version this txg, so there must be no changes this
7628 * txg to anything that the upgrade code depends on after it executes.
7629 * Therefore this must be called after dsl_pool_sync() does the sync
7630 * tasks.
7631 */
7632static void
7633spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7634{
7635 dsl_pool_t *dp = spa->spa_dsl_pool;
7636
7637 ASSERT(spa->spa_sync_pass == 1);
7638
7639 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7640
7641 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7642 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7643 dsl_pool_create_origin(dp, tx);
7644
7645 /* Keeping the origin open increases spa_minref */
7646 spa->spa_minref += 3;
7647 }
7648
7649 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7650 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7651 dsl_pool_upgrade_clones(dp, tx);
7652 }
7653
7654 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7655 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7656 dsl_pool_upgrade_dir_clones(dp, tx);
7657
7658 /* Keeping the freedir open increases spa_minref */
7659 spa->spa_minref += 3;
7660 }
7661
7662 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7663 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7664 spa_feature_create_zap_objects(spa, tx);
7665 }
7666
7667 /*
7668 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7669 * when possibility to use lz4 compression for metadata was added
7670 * Old pools that have this feature enabled must be upgraded to have
7671 * this feature active
7672 */
7673 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7674 boolean_t lz4_en = spa_feature_is_enabled(spa,
7675 SPA_FEATURE_LZ4_COMPRESS);
7676 boolean_t lz4_ac = spa_feature_is_active(spa,
7677 SPA_FEATURE_LZ4_COMPRESS);
7678
7679 if (lz4_en && !lz4_ac)
7680 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7681 }
7682
7683 /*
7684 * If we haven't written the salt, do so now. Note that the
7685 * feature may not be activated yet, but that's fine since
7686 * the presence of this ZAP entry is backwards compatible.
7687 */
7688 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7689 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7690 VERIFY0(zap_add(spa->spa_meta_objset,
7691 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7692 sizeof (spa->spa_cksum_salt.zcs_bytes),
7693 spa->spa_cksum_salt.zcs_bytes, tx));
7694 }
7695
7696 rrw_exit(&dp->dp_config_rwlock, FTAG);
7697}
7698
7699static void
7700vdev_indirect_state_sync_verify(vdev_t *vd)
7701{
7702 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7703 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7704
7705 if (vd->vdev_ops == &vdev_indirect_ops) {
7706 ASSERT(vim != NULL);
7707 ASSERT(vib != NULL);
7708 }
7709
7710 if (vdev_obsolete_sm_object(vd) != 0) {
7711 ASSERT(vd->vdev_obsolete_sm != NULL);
7712 ASSERT(vd->vdev_removing ||
7713 vd->vdev_ops == &vdev_indirect_ops);
7714 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7715 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7716
7717 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7718 space_map_object(vd->vdev_obsolete_sm));
7719 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7720 space_map_allocated(vd->vdev_obsolete_sm));
7721 }
7722 ASSERT(vd->vdev_obsolete_segments != NULL);
7723
7724 /*
7725 * Since frees / remaps to an indirect vdev can only
7726 * happen in syncing context, the obsolete segments
7727 * tree must be empty when we start syncing.
7728 */
7729 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7730}
7731
7732/*
7733 * Sync the specified transaction group. New blocks may be dirtied as
7734 * part of the process, so we iterate until it converges.
7735 */
7736void
7737spa_sync(spa_t *spa, uint64_t txg)
7738{
7739 dsl_pool_t *dp = spa->spa_dsl_pool;
7740 objset_t *mos = spa->spa_meta_objset;
7741 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7742 vdev_t *rvd = spa->spa_root_vdev;
7743 vdev_t *vd;
7744 dmu_tx_t *tx;
7745 int error;
7746 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7747 zfs_vdev_queue_depth_pct / 100;
7748
7749 VERIFY(spa_writeable(spa));
7750
7751 /*
7752 * Wait for i/os issued in open context that need to complete
7753 * before this txg syncs.
7754 */
7755 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK]));
7756 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0);
7757
7758 /*
7759 * Lock out configuration changes.
7760 */
7761 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7762
7763 spa->spa_syncing_txg = txg;
7764 spa->spa_sync_pass = 0;
7765
7766 mutex_enter(&spa->spa_alloc_lock);
7767 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7768 mutex_exit(&spa->spa_alloc_lock);
7769
7770 /*
7771 * If there are any pending vdev state changes, convert them
7772 * into config changes that go out with this transaction group.
7773 */
7774 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7775 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7776 /*
7777 * We need the write lock here because, for aux vdevs,
7778 * calling vdev_config_dirty() modifies sav_config.
7779 * This is ugly and will become unnecessary when we
7780 * eliminate the aux vdev wart by integrating all vdevs
7781 * into the root vdev tree.
7782 */
7783 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7784 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7785 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7786 vdev_state_clean(vd);
7787 vdev_config_dirty(vd);
7788 }
7789 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7790 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7791 }
7792 spa_config_exit(spa, SCL_STATE, FTAG);
7793
7794 tx = dmu_tx_create_assigned(dp, txg);
7795
7796 spa->spa_sync_starttime = gethrtime();
7797#ifdef illumos
7798 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7799 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7800#else /* !illumos */
7801#ifdef _KERNEL
7802 callout_schedule(&spa->spa_deadman_cycid,
7803 hz * spa->spa_deadman_synctime / NANOSEC);
7804#endif
7805#endif /* illumos */
7806
7807 /*
7808 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7809 * set spa_deflate if we have no raid-z vdevs.
7810 */
7811 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
7812 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
7813 int i;
7814
7815 for (i = 0; i < rvd->vdev_children; i++) {
7816 vd = rvd->vdev_child[i];
7817 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
7818 break;
7819 }
7820 if (i == rvd->vdev_children) {
7821 spa->spa_deflate = TRUE;
7822 VERIFY(0 == zap_add(spa->spa_meta_objset,
7823 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
7824 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
7825 }
7826 }
7827
7828 /*
7829 * Set the top-level vdev's max queue depth. Evaluate each
7830 * top-level's async write queue depth in case it changed.
7831 * The max queue depth will not change in the middle of syncing
7832 * out this txg.
7833 */
7834 uint64_t queue_depth_total = 0;
7835 for (int c = 0; c < rvd->vdev_children; c++) {
7836 vdev_t *tvd = rvd->vdev_child[c];
7837 metaslab_group_t *mg = tvd->vdev_mg;
7838
7839 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
7840 !metaslab_group_initialized(mg))
7841 continue;
7842
7843 /*
7844 * It is safe to do a lock-free check here because only async
7845 * allocations look at mg_max_alloc_queue_depth, and async
7846 * allocations all happen from spa_sync().
7847 */
7848 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
7849 mg->mg_max_alloc_queue_depth = max_queue_depth;
7850 queue_depth_total += mg->mg_max_alloc_queue_depth;
7851 }
7852 metaslab_class_t *mc = spa_normal_class(spa);
7853 ASSERT0(refcount_count(&mc->mc_alloc_slots));
7854 mc->mc_alloc_max_slots = queue_depth_total;
7855 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
7856
7857 ASSERT3U(mc->mc_alloc_max_slots, <=,
7858 max_queue_depth * rvd->vdev_children);
7859
7860 for (int c = 0; c < rvd->vdev_children; c++) {
7861 vdev_t *vd = rvd->vdev_child[c];
7862 vdev_indirect_state_sync_verify(vd);
7863
7864 if (vdev_indirect_should_condense(vd)) {
7865 spa_condense_indirect_start_sync(vd, tx);
7866 break;
7867 }
7868 }
7869
7870 /*
7871 * Iterate to convergence.
7872 */
7873 do {
7874 int pass = ++spa->spa_sync_pass;
7875
7876 spa_sync_config_object(spa, tx);
7877 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
7878 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
7879 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
7880 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
7881 spa_errlog_sync(spa, txg);
7882 dsl_pool_sync(dp, txg);
7883
7884 if (pass < zfs_sync_pass_deferred_free) {
7885 spa_sync_frees(spa, free_bpl, tx);
7886 } else {
7887 /*
7888 * We can not defer frees in pass 1, because
7889 * we sync the deferred frees later in pass 1.
7890 */
7891 ASSERT3U(pass, >, 1);
7892 bplist_iterate(free_bpl, bpobj_enqueue_cb,
7893 &spa->spa_deferred_bpobj, tx);
7894 }
7895
7896 ddt_sync(spa, txg);
7897 dsl_scan_sync(dp, tx);
7898
7899 if (spa->spa_vdev_removal != NULL)
7900 svr_sync(spa, tx);
7901
7902 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
7903 != NULL)
7904 vdev_sync(vd, txg);
7905
7906 if (pass == 1) {
7907 spa_sync_upgrades(spa, tx);
7908 ASSERT3U(txg, >=,
7909 spa->spa_uberblock.ub_rootbp.blk_birth);
7910 /*
7911 * Note: We need to check if the MOS is dirty
7912 * because we could have marked the MOS dirty
7913 * without updating the uberblock (e.g. if we
7914 * have sync tasks but no dirty user data). We
7915 * need to check the uberblock's rootbp because
7916 * it is updated if we have synced out dirty
7917 * data (though in this case the MOS will most
7918 * likely also be dirty due to second order
7919 * effects, we don't want to rely on that here).
7920 */
7921 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7922 !dmu_objset_is_dirty(mos, txg)) {
7923 /*
7924 * Nothing changed on the first pass,
7925 * therefore this TXG is a no-op. Avoid
7926 * syncing deferred frees, so that we
7927 * can keep this TXG as a no-op.
7928 */
7929 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7930 txg));
7931 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7932 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7933 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
7934 txg));
7935 break;
7936 }
7937 spa_sync_deferred_frees(spa, tx);
7938 }
7939
7940 } while (dmu_objset_is_dirty(mos, txg));
7941
7942 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7943 /*
7944 * Make sure that the number of ZAPs for all the vdevs matches
7945 * the number of ZAPs in the per-vdev ZAP list. This only gets
7946 * called if the config is dirty; otherwise there may be
7947 * outstanding AVZ operations that weren't completed in
7948 * spa_sync_config_object.
7949 */
7950 uint64_t all_vdev_zap_entry_count;
7951 ASSERT0(zap_count(spa->spa_meta_objset,
7952 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7953 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7954 all_vdev_zap_entry_count);
7955 }
7956
7957 if (spa->spa_vdev_removal != NULL) {
7958 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
7959 }
7960
7961 /*
7962 * Rewrite the vdev configuration (which includes the uberblock)
7963 * to commit the transaction group.
7964 *
7965 * If there are no dirty vdevs, we sync the uberblock to a few
7966 * random top-level vdevs that are known to be visible in the
7967 * config cache (see spa_vdev_add() for a complete description).
7968 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7969 */
7970 for (;;) {
7971 /*
7972 * We hold SCL_STATE to prevent vdev open/close/etc.
7973 * while we're attempting to write the vdev labels.
7974 */
7975 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7976
7977 if (list_is_empty(&spa->spa_config_dirty_list)) {
7978 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
7979 int svdcount = 0;
7980 int children = rvd->vdev_children;
7981 int c0 = spa_get_random(children);
7982
7983 for (int c = 0; c < children; c++) {
7984 vd = rvd->vdev_child[(c0 + c) % children];
7985
7986 /* Stop when revisiting the first vdev */
7987 if (c > 0 && svd[0] == vd)
7988 break;
7989
7990 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
7991 !vdev_is_concrete(vd))
7992 continue;
7993
7994 svd[svdcount++] = vd;
7995 if (svdcount == SPA_SYNC_MIN_VDEVS)
7996 break;
7997 }
7998 error = vdev_config_sync(svd, svdcount, txg);
7999 } else {
8000 error = vdev_config_sync(rvd->vdev_child,
8001 rvd->vdev_children, txg);
8002 }
8003
8004 if (error == 0)
8005 spa->spa_last_synced_guid = rvd->vdev_guid;
8006
8007 spa_config_exit(spa, SCL_STATE, FTAG);
8008
8009 if (error == 0)
8010 break;
8011 zio_suspend(spa, NULL);
8012 zio_resume_wait(spa);
8013 }
8014 dmu_tx_commit(tx);
8015
8016#ifdef illumos
8017 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
8018#else /* !illumos */
8019#ifdef _KERNEL
8020 callout_drain(&spa->spa_deadman_cycid);
8021#endif
8022#endif /* illumos */
8023
8024 /*
8025 * Clear the dirty config list.
8026 */
8027 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
8028 vdev_config_clean(vd);
8029
8030 /*
8031 * Now that the new config has synced transactionally,
8032 * let it become visible to the config cache.
8033 */
8034 if (spa->spa_config_syncing != NULL) {
8035 spa_config_set(spa, spa->spa_config_syncing);
8036 spa->spa_config_txg = txg;
8037 spa->spa_config_syncing = NULL;
8038 }
8039
8040 dsl_pool_sync_done(dp, txg);
8041
8042 mutex_enter(&spa->spa_alloc_lock);
8043 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
8044 mutex_exit(&spa->spa_alloc_lock);
8045
8046 /*
8047 * Update usable space statistics.
8048 */
8049 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
8050 vdev_sync_done(vd, txg);
8051
8052 spa_update_dspace(spa);
8053
8054 /*
8055 * It had better be the case that we didn't dirty anything
8056 * since vdev_config_sync().
8057 */
8058 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8059 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8060 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
8061
8062 while (zfs_pause_spa_sync)
8063 delay(1);
8064
8065 spa->spa_sync_pass = 0;
8066
8067 /*
8068 * Update the last synced uberblock here. We want to do this at
8069 * the end of spa_sync() so that consumers of spa_last_synced_txg()
8070 * will be guaranteed that all the processing associated with
8071 * that txg has been completed.
8072 */
8073 spa->spa_ubsync = spa->spa_uberblock;
8074 spa_config_exit(spa, SCL_CONFIG, FTAG);
8075
8076 spa_handle_ignored_writes(spa);
8077
8078 /*
8079 * If any async tasks have been requested, kick them off.
8080 */
8081 spa_async_dispatch(spa);
8082 spa_async_dispatch_vd(spa);
8083}
8084
8085/*
8086 * Sync all pools. We don't want to hold the namespace lock across these
8087 * operations, so we take a reference on the spa_t and drop the lock during the
8088 * sync.
8089 */
8090void
8091spa_sync_allpools(void)
8092{
8093 spa_t *spa = NULL;
8094 mutex_enter(&spa_namespace_lock);
8095 while ((spa = spa_next(spa)) != NULL) {
8096 if (spa_state(spa) != POOL_STATE_ACTIVE ||
8097 !spa_writeable(spa) || spa_suspended(spa))
8098 continue;
8099 spa_open_ref(spa, FTAG);
8100 mutex_exit(&spa_namespace_lock);
8101 txg_wait_synced(spa_get_dsl(spa), 0);
8102 mutex_enter(&spa_namespace_lock);
8103 spa_close(spa, FTAG);
8104 }
8105 mutex_exit(&spa_namespace_lock);
8106}
8107
8108/*
8109 * ==========================================================================
8110 * Miscellaneous routines
8111 * ==========================================================================
8112 */
8113
8114/*
8115 * Remove all pools in the system.
8116 */
8117void
8118spa_evict_all(void)
8119{
8120 spa_t *spa;
8121
8122 /*
8123 * Remove all cached state. All pools should be closed now,
8124 * so every spa in the AVL tree should be unreferenced.
8125 */
8126 mutex_enter(&spa_namespace_lock);
8127 while ((spa = spa_next(NULL)) != NULL) {
8128 /*
8129 * Stop async tasks. The async thread may need to detach
8130 * a device that's been replaced, which requires grabbing
8131 * spa_namespace_lock, so we must drop it here.
8132 */
8133 spa_open_ref(spa, FTAG);
8134 mutex_exit(&spa_namespace_lock);
8135 spa_async_suspend(spa);
8136 mutex_enter(&spa_namespace_lock);
8137 spa_close(spa, FTAG);
8138
8139 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
8140 spa_unload(spa);
8141 spa_deactivate(spa);
8142 }
8143 spa_remove(spa);
8144 }
8145 mutex_exit(&spa_namespace_lock);
8146}
8147
8148vdev_t *
8149spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
8150{
8151 vdev_t *vd;
8152 int i;
8153
8154 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
8155 return (vd);
8156
8157 if (aux) {
8158 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
8159 vd = spa->spa_l2cache.sav_vdevs[i];
8160 if (vd->vdev_guid == guid)
8161 return (vd);
8162 }
8163
8164 for (i = 0; i < spa->spa_spares.sav_count; i++) {
8165 vd = spa->spa_spares.sav_vdevs[i];
8166 if (vd->vdev_guid == guid)
8167 return (vd);
8168 }
8169 }
8170
8171 return (NULL);
8172}
8173
8174void
8175spa_upgrade(spa_t *spa, uint64_t version)
8176{
8177 ASSERT(spa_writeable(spa));
8178
8179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
8180
8181 /*
8182 * This should only be called for a non-faulted pool, and since a
8183 * future version would result in an unopenable pool, this shouldn't be
8184 * possible.
8185 */
8186 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
8187 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
8188
8189 spa->spa_uberblock.ub_version = version;
8190 vdev_config_dirty(spa->spa_root_vdev);
8191
8192 spa_config_exit(spa, SCL_ALL, FTAG);
8193
8194 txg_wait_synced(spa_get_dsl(spa), 0);
8195}
8196
8197boolean_t
8198spa_has_spare(spa_t *spa, uint64_t guid)
8199{
8200 int i;
8201 uint64_t spareguid;
8202 spa_aux_vdev_t *sav = &spa->spa_spares;
8203
8204 for (i = 0; i < sav->sav_count; i++)
8205 if (sav->sav_vdevs[i]->vdev_guid == guid)
8206 return (B_TRUE);
8207
8208 for (i = 0; i < sav->sav_npending; i++) {
8209 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
8210 &spareguid) == 0 && spareguid == guid)
8211 return (B_TRUE);
8212 }
8213
8214 return (B_FALSE);
8215}
8216
8217/*
8218 * Check if a pool has an active shared spare device.
8219 * Note: reference count of an active spare is 2, as a spare and as a replace
8220 */
8221static boolean_t
8222spa_has_active_shared_spare(spa_t *spa)
8223{
8224 int i, refcnt;
8225 uint64_t pool;
8226 spa_aux_vdev_t *sav = &spa->spa_spares;
8227
8228 for (i = 0; i < sav->sav_count; i++) {
8229 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8230 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8231 refcnt > 2)
8232 return (B_TRUE);
8233 }
8234
8235 return (B_FALSE);
8236}
8237
8238sysevent_t *
8239spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8240{
8241 sysevent_t *ev = NULL;
8242#ifdef _KERNEL
8243 sysevent_attr_list_t *attr = NULL;
8244 sysevent_value_t value;
8245
8246 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8247 SE_SLEEP);
8248 ASSERT(ev != NULL);
8249
8250 value.value_type = SE_DATA_TYPE_STRING;
8251 value.value.sv_string = spa_name(spa);
8252 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8253 goto done;
8254
8255 value.value_type = SE_DATA_TYPE_UINT64;
8256 value.value.sv_uint64 = spa_guid(spa);
8257 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8258 goto done;
8259
8260 if (vd) {
8261 value.value_type = SE_DATA_TYPE_UINT64;
8262 value.value.sv_uint64 = vd->vdev_guid;
8263 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8264 SE_SLEEP) != 0)
8265 goto done;
8266
8267 if (vd->vdev_path) {
8268 value.value_type = SE_DATA_TYPE_STRING;
8269 value.value.sv_string = vd->vdev_path;
8270 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8271 &value, SE_SLEEP) != 0)
8272 goto done;
8273 }
8274 }
8275
8276 if (hist_nvl != NULL) {
8277 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8278 }
8279
8280 if (sysevent_attach_attributes(ev, attr) != 0)
8281 goto done;
8282 attr = NULL;
8283
8284done:
8285 if (attr)
8286 sysevent_free_attr(attr);
8287
8288#endif
8289 return (ev);
8290}
8291
8292void
8293spa_event_post(sysevent_t *ev)
8294{
8295#ifdef _KERNEL
8296 sysevent_id_t eid;
8297
8298 (void) log_sysevent(ev, SE_SLEEP, &eid);
8299 sysevent_free(ev);
8300#endif
8301}
8302
8303void
8304spa_event_discard(sysevent_t *ev)
8305{
8306#ifdef _KERNEL
8307 sysevent_free(ev);
8308#endif
8309}
8310
8311/*
8312 * Post a sysevent corresponding to the given event. The 'name' must be one of
8313 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8314 * filled in from the spa and (optionally) the vdev and history nvl. This
8315 * doesn't do anything in the userland libzpool, as we don't want consumers to
8316 * misinterpret ztest or zdb as real changes.
8317 */
8318void
8319spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8320{
8321 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
8322}
| 5544 } else {
5545 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5546 }
5547
5548 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5549 == 0) {
5550 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5551 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5552 } else {
5553 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5554 }
5555
5556 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5557
5558 /*
5559 * If 'tryconfig' was at least parsable, return the current config.
5560 */
5561 if (spa->spa_root_vdev != NULL) {
5562 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5563 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5564 poolname) == 0);
5565 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5566 state) == 0);
5567 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5568 spa->spa_uberblock.ub_timestamp) == 0);
5569 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5570 spa->spa_load_info) == 0);
5571
5572 /*
5573 * If the bootfs property exists on this pool then we
5574 * copy it out so that external consumers can tell which
5575 * pools are bootable.
5576 */
5577 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5578 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5579
5580 /*
5581 * We have to play games with the name since the
5582 * pool was opened as TRYIMPORT_NAME.
5583 */
5584 if (dsl_dsobj_to_dsname(spa_name(spa),
5585 spa->spa_bootfs, tmpname) == 0) {
5586 char *cp;
5587 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5588
5589 cp = strchr(tmpname, '/');
5590 if (cp == NULL) {
5591 (void) strlcpy(dsname, tmpname,
5592 MAXPATHLEN);
5593 } else {
5594 (void) snprintf(dsname, MAXPATHLEN,
5595 "%s/%s", poolname, ++cp);
5596 }
5597 VERIFY(nvlist_add_string(config,
5598 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5599 kmem_free(dsname, MAXPATHLEN);
5600 }
5601 kmem_free(tmpname, MAXPATHLEN);
5602 }
5603
5604 /*
5605 * Add the list of hot spares and level 2 cache devices.
5606 */
5607 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5608 spa_add_spares(spa, config);
5609 spa_add_l2cache(spa, config);
5610 spa_config_exit(spa, SCL_CONFIG, FTAG);
5611 }
5612
5613 spa_unload(spa);
5614 spa_deactivate(spa);
5615 spa_remove(spa);
5616 mutex_exit(&spa_namespace_lock);
5617
5618 return (config);
5619}
5620
5621/*
5622 * Pool export/destroy
5623 *
5624 * The act of destroying or exporting a pool is very simple. We make sure there
5625 * is no more pending I/O and any references to the pool are gone. Then, we
5626 * update the pool state and sync all the labels to disk, removing the
5627 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5628 * we don't sync the labels or remove the configuration cache.
5629 */
5630static int
5631spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5632 boolean_t force, boolean_t hardforce)
5633{
5634 spa_t *spa;
5635
5636 if (oldconfig)
5637 *oldconfig = NULL;
5638
5639 if (!(spa_mode_global & FWRITE))
5640 return (SET_ERROR(EROFS));
5641
5642 mutex_enter(&spa_namespace_lock);
5643 if ((spa = spa_lookup(pool)) == NULL) {
5644 mutex_exit(&spa_namespace_lock);
5645 return (SET_ERROR(ENOENT));
5646 }
5647
5648 /*
5649 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5650 * reacquire the namespace lock, and see if we can export.
5651 */
5652 spa_open_ref(spa, FTAG);
5653 mutex_exit(&spa_namespace_lock);
5654 spa_async_suspend(spa);
5655 mutex_enter(&spa_namespace_lock);
5656 spa_close(spa, FTAG);
5657
5658 /*
5659 * The pool will be in core if it's openable,
5660 * in which case we can modify its state.
5661 */
5662 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5663 /*
5664 * Objsets may be open only because they're dirty, so we
5665 * have to force it to sync before checking spa_refcnt.
5666 */
5667 txg_wait_synced(spa->spa_dsl_pool, 0);
5668 spa_evicting_os_wait(spa);
5669
5670 /*
5671 * A pool cannot be exported or destroyed if there are active
5672 * references. If we are resetting a pool, allow references by
5673 * fault injection handlers.
5674 */
5675 if (!spa_refcount_zero(spa) ||
5676 (spa->spa_inject_ref != 0 &&
5677 new_state != POOL_STATE_UNINITIALIZED)) {
5678 spa_async_resume(spa);
5679 mutex_exit(&spa_namespace_lock);
5680 return (SET_ERROR(EBUSY));
5681 }
5682
5683 /*
5684 * A pool cannot be exported if it has an active shared spare.
5685 * This is to prevent other pools stealing the active spare
5686 * from an exported pool. At user's own will, such pool can
5687 * be forcedly exported.
5688 */
5689 if (!force && new_state == POOL_STATE_EXPORTED &&
5690 spa_has_active_shared_spare(spa)) {
5691 spa_async_resume(spa);
5692 mutex_exit(&spa_namespace_lock);
5693 return (SET_ERROR(EXDEV));
5694 }
5695
5696 /*
5697 * We want this to be reflected on every label,
5698 * so mark them all dirty. spa_unload() will do the
5699 * final sync that pushes these changes out.
5700 */
5701 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5702 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5703 spa->spa_state = new_state;
5704 spa->spa_final_txg = spa_last_synced_txg(spa) +
5705 TXG_DEFER_SIZE + 1;
5706 vdev_config_dirty(spa->spa_root_vdev);
5707 spa_config_exit(spa, SCL_ALL, FTAG);
5708 }
5709 }
5710
5711 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5712
5713 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5714 spa_unload(spa);
5715 spa_deactivate(spa);
5716 }
5717
5718 if (oldconfig && spa->spa_config)
5719 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5720
5721 if (new_state != POOL_STATE_UNINITIALIZED) {
5722 if (!hardforce)
5723 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5724 spa_remove(spa);
5725 }
5726 mutex_exit(&spa_namespace_lock);
5727
5728 return (0);
5729}
5730
5731/*
5732 * Destroy a storage pool.
5733 */
5734int
5735spa_destroy(char *pool)
5736{
5737 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5738 B_FALSE, B_FALSE));
5739}
5740
5741/*
5742 * Export a storage pool.
5743 */
5744int
5745spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5746 boolean_t hardforce)
5747{
5748 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5749 force, hardforce));
5750}
5751
5752/*
5753 * Similar to spa_export(), this unloads the spa_t without actually removing it
5754 * from the namespace in any way.
5755 */
5756int
5757spa_reset(char *pool)
5758{
5759 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5760 B_FALSE, B_FALSE));
5761}
5762
5763/*
5764 * ==========================================================================
5765 * Device manipulation
5766 * ==========================================================================
5767 */
5768
5769/*
5770 * Add a device to a storage pool.
5771 */
5772int
5773spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5774{
5775 uint64_t txg, id;
5776 int error;
5777 vdev_t *rvd = spa->spa_root_vdev;
5778 vdev_t *vd, *tvd;
5779 nvlist_t **spares, **l2cache;
5780 uint_t nspares, nl2cache;
5781
5782 ASSERT(spa_writeable(spa));
5783
5784 txg = spa_vdev_enter(spa);
5785
5786 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5787 VDEV_ALLOC_ADD)) != 0)
5788 return (spa_vdev_exit(spa, NULL, txg, error));
5789
5790 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5791
5792 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5793 &nspares) != 0)
5794 nspares = 0;
5795
5796 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5797 &nl2cache) != 0)
5798 nl2cache = 0;
5799
5800 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5801 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5802
5803 if (vd->vdev_children != 0 &&
5804 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5805 return (spa_vdev_exit(spa, vd, txg, error));
5806
5807 /*
5808 * We must validate the spares and l2cache devices after checking the
5809 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5810 */
5811 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5812 return (spa_vdev_exit(spa, vd, txg, error));
5813
5814 /*
5815 * If we are in the middle of a device removal, we can only add
5816 * devices which match the existing devices in the pool.
5817 * If we are in the middle of a removal, or have some indirect
5818 * vdevs, we can not add raidz toplevels.
5819 */
5820 if (spa->spa_vdev_removal != NULL ||
5821 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5822 for (int c = 0; c < vd->vdev_children; c++) {
5823 tvd = vd->vdev_child[c];
5824 if (spa->spa_vdev_removal != NULL &&
5825 tvd->vdev_ashift !=
5826 spa->spa_vdev_removal->svr_vdev->vdev_ashift) {
5827 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5828 }
5829 /* Fail if top level vdev is raidz */
5830 if (tvd->vdev_ops == &vdev_raidz_ops) {
5831 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5832 }
5833 /*
5834 * Need the top level mirror to be
5835 * a mirror of leaf vdevs only
5836 */
5837 if (tvd->vdev_ops == &vdev_mirror_ops) {
5838 for (uint64_t cid = 0;
5839 cid < tvd->vdev_children; cid++) {
5840 vdev_t *cvd = tvd->vdev_child[cid];
5841 if (!cvd->vdev_ops->vdev_op_leaf) {
5842 return (spa_vdev_exit(spa, vd,
5843 txg, EINVAL));
5844 }
5845 }
5846 }
5847 }
5848 }
5849
5850 for (int c = 0; c < vd->vdev_children; c++) {
5851
5852 /*
5853 * Set the vdev id to the first hole, if one exists.
5854 */
5855 for (id = 0; id < rvd->vdev_children; id++) {
5856 if (rvd->vdev_child[id]->vdev_ishole) {
5857 vdev_free(rvd->vdev_child[id]);
5858 break;
5859 }
5860 }
5861 tvd = vd->vdev_child[c];
5862 vdev_remove_child(vd, tvd);
5863 tvd->vdev_id = id;
5864 vdev_add_child(rvd, tvd);
5865 vdev_config_dirty(tvd);
5866 }
5867
5868 if (nspares != 0) {
5869 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5870 ZPOOL_CONFIG_SPARES);
5871 spa_load_spares(spa);
5872 spa->spa_spares.sav_sync = B_TRUE;
5873 }
5874
5875 if (nl2cache != 0) {
5876 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5877 ZPOOL_CONFIG_L2CACHE);
5878 spa_load_l2cache(spa);
5879 spa->spa_l2cache.sav_sync = B_TRUE;
5880 }
5881
5882 /*
5883 * We have to be careful when adding new vdevs to an existing pool.
5884 * If other threads start allocating from these vdevs before we
5885 * sync the config cache, and we lose power, then upon reboot we may
5886 * fail to open the pool because there are DVAs that the config cache
5887 * can't translate. Therefore, we first add the vdevs without
5888 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5889 * and then let spa_config_update() initialize the new metaslabs.
5890 *
5891 * spa_load() checks for added-but-not-initialized vdevs, so that
5892 * if we lose power at any point in this sequence, the remaining
5893 * steps will be completed the next time we load the pool.
5894 */
5895 (void) spa_vdev_exit(spa, vd, txg, 0);
5896
5897 mutex_enter(&spa_namespace_lock);
5898 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5899 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5900 mutex_exit(&spa_namespace_lock);
5901
5902 return (0);
5903}
5904
5905/*
5906 * Attach a device to a mirror. The arguments are the path to any device
5907 * in the mirror, and the nvroot for the new device. If the path specifies
5908 * a device that is not mirrored, we automatically insert the mirror vdev.
5909 *
5910 * If 'replacing' is specified, the new device is intended to replace the
5911 * existing device; in this case the two devices are made into their own
5912 * mirror using the 'replacing' vdev, which is functionally identical to
5913 * the mirror vdev (it actually reuses all the same ops) but has a few
5914 * extra rules: you can't attach to it after it's been created, and upon
5915 * completion of resilvering, the first disk (the one being replaced)
5916 * is automatically detached.
5917 */
5918int
5919spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5920{
5921 uint64_t txg, dtl_max_txg;
5922 vdev_t *rvd = spa->spa_root_vdev;
5923 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5924 vdev_ops_t *pvops;
5925 char *oldvdpath, *newvdpath;
5926 int newvd_isspare;
5927 int error;
5928
5929 ASSERT(spa_writeable(spa));
5930
5931 txg = spa_vdev_enter(spa);
5932
5933 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5934
5935 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5936 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5937 error = (spa_has_checkpoint(spa)) ?
5938 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5939 return (spa_vdev_exit(spa, NULL, txg, error));
5940 }
5941
5942 if (spa->spa_vdev_removal != NULL ||
5943 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5944 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5945 }
5946
5947 if (oldvd == NULL)
5948 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5949
5950 if (!oldvd->vdev_ops->vdev_op_leaf)
5951 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5952
5953 pvd = oldvd->vdev_parent;
5954
5955 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
5956 VDEV_ALLOC_ATTACH)) != 0)
5957 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5958
5959 if (newrootvd->vdev_children != 1)
5960 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5961
5962 newvd = newrootvd->vdev_child[0];
5963
5964 if (!newvd->vdev_ops->vdev_op_leaf)
5965 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5966
5967 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
5968 return (spa_vdev_exit(spa, newrootvd, txg, error));
5969
5970 /*
5971 * Spares can't replace logs
5972 */
5973 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
5974 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5975
5976 if (!replacing) {
5977 /*
5978 * For attach, the only allowable parent is a mirror or the root
5979 * vdev.
5980 */
5981 if (pvd->vdev_ops != &vdev_mirror_ops &&
5982 pvd->vdev_ops != &vdev_root_ops)
5983 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5984
5985 pvops = &vdev_mirror_ops;
5986 } else {
5987 /*
5988 * Active hot spares can only be replaced by inactive hot
5989 * spares.
5990 */
5991 if (pvd->vdev_ops == &vdev_spare_ops &&
5992 oldvd->vdev_isspare &&
5993 !spa_has_spare(spa, newvd->vdev_guid))
5994 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5995
5996 /*
5997 * If the source is a hot spare, and the parent isn't already a
5998 * spare, then we want to create a new hot spare. Otherwise, we
5999 * want to create a replacing vdev. The user is not allowed to
6000 * attach to a spared vdev child unless the 'isspare' state is
6001 * the same (spare replaces spare, non-spare replaces
6002 * non-spare).
6003 */
6004 if (pvd->vdev_ops == &vdev_replacing_ops &&
6005 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6006 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6007 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6008 newvd->vdev_isspare != oldvd->vdev_isspare) {
6009 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6010 }
6011
6012 if (newvd->vdev_isspare)
6013 pvops = &vdev_spare_ops;
6014 else
6015 pvops = &vdev_replacing_ops;
6016 }
6017
6018 /*
6019 * Make sure the new device is big enough.
6020 */
6021 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6022 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6023
6024 /*
6025 * The new device cannot have a higher alignment requirement
6026 * than the top-level vdev.
6027 */
6028 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6029 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6030
6031 /*
6032 * If this is an in-place replacement, update oldvd's path and devid
6033 * to make it distinguishable from newvd, and unopenable from now on.
6034 */
6035 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6036 spa_strfree(oldvd->vdev_path);
6037 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6038 KM_SLEEP);
6039 (void) sprintf(oldvd->vdev_path, "%s/%s",
6040 newvd->vdev_path, "old");
6041 if (oldvd->vdev_devid != NULL) {
6042 spa_strfree(oldvd->vdev_devid);
6043 oldvd->vdev_devid = NULL;
6044 }
6045 }
6046
6047 /* mark the device being resilvered */
6048 newvd->vdev_resilver_txg = txg;
6049
6050 /*
6051 * If the parent is not a mirror, or if we're replacing, insert the new
6052 * mirror/replacing/spare vdev above oldvd.
6053 */
6054 if (pvd->vdev_ops != pvops)
6055 pvd = vdev_add_parent(oldvd, pvops);
6056
6057 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6058 ASSERT(pvd->vdev_ops == pvops);
6059 ASSERT(oldvd->vdev_parent == pvd);
6060
6061 /*
6062 * Extract the new device from its root and add it to pvd.
6063 */
6064 vdev_remove_child(newrootvd, newvd);
6065 newvd->vdev_id = pvd->vdev_children;
6066 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6067 vdev_add_child(pvd, newvd);
6068
6069 tvd = newvd->vdev_top;
6070 ASSERT(pvd->vdev_top == tvd);
6071 ASSERT(tvd->vdev_parent == rvd);
6072
6073 vdev_config_dirty(tvd);
6074
6075 /*
6076 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6077 * for any dmu_sync-ed blocks. It will propagate upward when
6078 * spa_vdev_exit() calls vdev_dtl_reassess().
6079 */
6080 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6081
6082 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6083 dtl_max_txg - TXG_INITIAL);
6084
6085 if (newvd->vdev_isspare) {
6086 spa_spare_activate(newvd);
6087 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6088 }
6089
6090 oldvdpath = spa_strdup(oldvd->vdev_path);
6091 newvdpath = spa_strdup(newvd->vdev_path);
6092 newvd_isspare = newvd->vdev_isspare;
6093
6094 /*
6095 * Mark newvd's DTL dirty in this txg.
6096 */
6097 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6098
6099 /*
6100 * Schedule the resilver to restart in the future. We do this to
6101 * ensure that dmu_sync-ed blocks have been stitched into the
6102 * respective datasets.
6103 */
6104 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6105
6106 if (spa->spa_bootfs)
6107 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6108
6109 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6110
6111 /*
6112 * Commit the config
6113 */
6114 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6115
6116 spa_history_log_internal(spa, "vdev attach", NULL,
6117 "%s vdev=%s %s vdev=%s",
6118 replacing && newvd_isspare ? "spare in" :
6119 replacing ? "replace" : "attach", newvdpath,
6120 replacing ? "for" : "to", oldvdpath);
6121
6122 spa_strfree(oldvdpath);
6123 spa_strfree(newvdpath);
6124
6125 return (0);
6126}
6127
6128/*
6129 * Detach a device from a mirror or replacing vdev.
6130 *
6131 * If 'replace_done' is specified, only detach if the parent
6132 * is a replacing vdev.
6133 */
6134int
6135spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6136{
6137 uint64_t txg;
6138 int error;
6139 vdev_t *rvd = spa->spa_root_vdev;
6140 vdev_t *vd, *pvd, *cvd, *tvd;
6141 boolean_t unspare = B_FALSE;
6142 uint64_t unspare_guid = 0;
6143 char *vdpath;
6144
6145 ASSERT(spa_writeable(spa));
6146
6147 txg = spa_vdev_enter(spa);
6148
6149 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6150
6151 /*
6152 * Besides being called directly from the userland through the
6153 * ioctl interface, spa_vdev_detach() can be potentially called
6154 * at the end of spa_vdev_resilver_done().
6155 *
6156 * In the regular case, when we have a checkpoint this shouldn't
6157 * happen as we never empty the DTLs of a vdev during the scrub
6158 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6159 * should never get here when we have a checkpoint.
6160 *
6161 * That said, even in a case when we checkpoint the pool exactly
6162 * as spa_vdev_resilver_done() calls this function everything
6163 * should be fine as the resilver will return right away.
6164 */
6165 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6166 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6167 error = (spa_has_checkpoint(spa)) ?
6168 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6169 return (spa_vdev_exit(spa, NULL, txg, error));
6170 }
6171
6172 if (vd == NULL)
6173 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6174
6175 if (!vd->vdev_ops->vdev_op_leaf)
6176 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6177
6178 pvd = vd->vdev_parent;
6179
6180 /*
6181 * If the parent/child relationship is not as expected, don't do it.
6182 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6183 * vdev that's replacing B with C. The user's intent in replacing
6184 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6185 * the replace by detaching C, the expected behavior is to end up
6186 * M(A,B). But suppose that right after deciding to detach C,
6187 * the replacement of B completes. We would have M(A,C), and then
6188 * ask to detach C, which would leave us with just A -- not what
6189 * the user wanted. To prevent this, we make sure that the
6190 * parent/child relationship hasn't changed -- in this example,
6191 * that C's parent is still the replacing vdev R.
6192 */
6193 if (pvd->vdev_guid != pguid && pguid != 0)
6194 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6195
6196 /*
6197 * Only 'replacing' or 'spare' vdevs can be replaced.
6198 */
6199 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6200 pvd->vdev_ops != &vdev_spare_ops)
6201 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6202
6203 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6204 spa_version(spa) >= SPA_VERSION_SPARES);
6205
6206 /*
6207 * Only mirror, replacing, and spare vdevs support detach.
6208 */
6209 if (pvd->vdev_ops != &vdev_replacing_ops &&
6210 pvd->vdev_ops != &vdev_mirror_ops &&
6211 pvd->vdev_ops != &vdev_spare_ops)
6212 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6213
6214 /*
6215 * If this device has the only valid copy of some data,
6216 * we cannot safely detach it.
6217 */
6218 if (vdev_dtl_required(vd))
6219 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6220
6221 ASSERT(pvd->vdev_children >= 2);
6222
6223 /*
6224 * If we are detaching the second disk from a replacing vdev, then
6225 * check to see if we changed the original vdev's path to have "/old"
6226 * at the end in spa_vdev_attach(). If so, undo that change now.
6227 */
6228 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6229 vd->vdev_path != NULL) {
6230 size_t len = strlen(vd->vdev_path);
6231
6232 for (int c = 0; c < pvd->vdev_children; c++) {
6233 cvd = pvd->vdev_child[c];
6234
6235 if (cvd == vd || cvd->vdev_path == NULL)
6236 continue;
6237
6238 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6239 strcmp(cvd->vdev_path + len, "/old") == 0) {
6240 spa_strfree(cvd->vdev_path);
6241 cvd->vdev_path = spa_strdup(vd->vdev_path);
6242 break;
6243 }
6244 }
6245 }
6246
6247 /*
6248 * If we are detaching the original disk from a spare, then it implies
6249 * that the spare should become a real disk, and be removed from the
6250 * active spare list for the pool.
6251 */
6252 if (pvd->vdev_ops == &vdev_spare_ops &&
6253 vd->vdev_id == 0 &&
6254 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6255 unspare = B_TRUE;
6256
6257 /*
6258 * Erase the disk labels so the disk can be used for other things.
6259 * This must be done after all other error cases are handled,
6260 * but before we disembowel vd (so we can still do I/O to it).
6261 * But if we can't do it, don't treat the error as fatal --
6262 * it may be that the unwritability of the disk is the reason
6263 * it's being detached!
6264 */
6265 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6266
6267 /*
6268 * Remove vd from its parent and compact the parent's children.
6269 */
6270 vdev_remove_child(pvd, vd);
6271 vdev_compact_children(pvd);
6272
6273 /*
6274 * Remember one of the remaining children so we can get tvd below.
6275 */
6276 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6277
6278 /*
6279 * If we need to remove the remaining child from the list of hot spares,
6280 * do it now, marking the vdev as no longer a spare in the process.
6281 * We must do this before vdev_remove_parent(), because that can
6282 * change the GUID if it creates a new toplevel GUID. For a similar
6283 * reason, we must remove the spare now, in the same txg as the detach;
6284 * otherwise someone could attach a new sibling, change the GUID, and
6285 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6286 */
6287 if (unspare) {
6288 ASSERT(cvd->vdev_isspare);
6289 spa_spare_remove(cvd);
6290 unspare_guid = cvd->vdev_guid;
6291 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6292 cvd->vdev_unspare = B_TRUE;
6293 }
6294
6295 /*
6296 * If the parent mirror/replacing vdev only has one child,
6297 * the parent is no longer needed. Remove it from the tree.
6298 */
6299 if (pvd->vdev_children == 1) {
6300 if (pvd->vdev_ops == &vdev_spare_ops)
6301 cvd->vdev_unspare = B_FALSE;
6302 vdev_remove_parent(cvd);
6303 }
6304
6305
6306 /*
6307 * We don't set tvd until now because the parent we just removed
6308 * may have been the previous top-level vdev.
6309 */
6310 tvd = cvd->vdev_top;
6311 ASSERT(tvd->vdev_parent == rvd);
6312
6313 /*
6314 * Reevaluate the parent vdev state.
6315 */
6316 vdev_propagate_state(cvd);
6317
6318 /*
6319 * If the 'autoexpand' property is set on the pool then automatically
6320 * try to expand the size of the pool. For example if the device we
6321 * just detached was smaller than the others, it may be possible to
6322 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6323 * first so that we can obtain the updated sizes of the leaf vdevs.
6324 */
6325 if (spa->spa_autoexpand) {
6326 vdev_reopen(tvd);
6327 vdev_expand(tvd, txg);
6328 }
6329
6330 vdev_config_dirty(tvd);
6331
6332 /*
6333 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6334 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6335 * But first make sure we're not on any *other* txg's DTL list, to
6336 * prevent vd from being accessed after it's freed.
6337 */
6338 vdpath = spa_strdup(vd->vdev_path);
6339 for (int t = 0; t < TXG_SIZE; t++)
6340 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6341 vd->vdev_detached = B_TRUE;
6342 vdev_dirty(tvd, VDD_DTL, vd, txg);
6343
6344 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6345
6346 /* hang on to the spa before we release the lock */
6347 spa_open_ref(spa, FTAG);
6348
6349 error = spa_vdev_exit(spa, vd, txg, 0);
6350
6351 spa_history_log_internal(spa, "detach", NULL,
6352 "vdev=%s", vdpath);
6353 spa_strfree(vdpath);
6354
6355 /*
6356 * If this was the removal of the original device in a hot spare vdev,
6357 * then we want to go through and remove the device from the hot spare
6358 * list of every other pool.
6359 */
6360 if (unspare) {
6361 spa_t *altspa = NULL;
6362
6363 mutex_enter(&spa_namespace_lock);
6364 while ((altspa = spa_next(altspa)) != NULL) {
6365 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6366 altspa == spa)
6367 continue;
6368
6369 spa_open_ref(altspa, FTAG);
6370 mutex_exit(&spa_namespace_lock);
6371 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6372 mutex_enter(&spa_namespace_lock);
6373 spa_close(altspa, FTAG);
6374 }
6375 mutex_exit(&spa_namespace_lock);
6376
6377 /* search the rest of the vdevs for spares to remove */
6378 spa_vdev_resilver_done(spa);
6379 }
6380
6381 /* all done with the spa; OK to release */
6382 mutex_enter(&spa_namespace_lock);
6383 spa_close(spa, FTAG);
6384 mutex_exit(&spa_namespace_lock);
6385
6386 return (error);
6387}
6388
6389/*
6390 * Split a set of devices from their mirrors, and create a new pool from them.
6391 */
6392int
6393spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6394 nvlist_t *props, boolean_t exp)
6395{
6396 int error = 0;
6397 uint64_t txg, *glist;
6398 spa_t *newspa;
6399 uint_t c, children, lastlog;
6400 nvlist_t **child, *nvl, *tmp;
6401 dmu_tx_t *tx;
6402 char *altroot = NULL;
6403 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6404 boolean_t activate_slog;
6405
6406 ASSERT(spa_writeable(spa));
6407
6408 txg = spa_vdev_enter(spa);
6409
6410 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6411 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6412 error = (spa_has_checkpoint(spa)) ?
6413 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6414 return (spa_vdev_exit(spa, NULL, txg, error));
6415 }
6416
6417 /* clear the log and flush everything up to now */
6418 activate_slog = spa_passivate_log(spa);
6419 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6420 error = spa_reset_logs(spa);
6421 txg = spa_vdev_config_enter(spa);
6422
6423 if (activate_slog)
6424 spa_activate_log(spa);
6425
6426 if (error != 0)
6427 return (spa_vdev_exit(spa, NULL, txg, error));
6428
6429 /* check new spa name before going any further */
6430 if (spa_lookup(newname) != NULL)
6431 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6432
6433 /*
6434 * scan through all the children to ensure they're all mirrors
6435 */
6436 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6437 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6438 &children) != 0)
6439 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6440
6441 /* first, check to ensure we've got the right child count */
6442 rvd = spa->spa_root_vdev;
6443 lastlog = 0;
6444 for (c = 0; c < rvd->vdev_children; c++) {
6445 vdev_t *vd = rvd->vdev_child[c];
6446
6447 /* don't count the holes & logs as children */
6448 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6449 if (lastlog == 0)
6450 lastlog = c;
6451 continue;
6452 }
6453
6454 lastlog = 0;
6455 }
6456 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6457 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6458
6459 /* next, ensure no spare or cache devices are part of the split */
6460 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6461 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6462 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6463
6464 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6465 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6466
6467 /* then, loop over each vdev and validate it */
6468 for (c = 0; c < children; c++) {
6469 uint64_t is_hole = 0;
6470
6471 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6472 &is_hole);
6473
6474 if (is_hole != 0) {
6475 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6476 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6477 continue;
6478 } else {
6479 error = SET_ERROR(EINVAL);
6480 break;
6481 }
6482 }
6483
6484 /* which disk is going to be split? */
6485 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6486 &glist[c]) != 0) {
6487 error = SET_ERROR(EINVAL);
6488 break;
6489 }
6490
6491 /* look it up in the spa */
6492 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6493 if (vml[c] == NULL) {
6494 error = SET_ERROR(ENODEV);
6495 break;
6496 }
6497
6498 /* make sure there's nothing stopping the split */
6499 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6500 vml[c]->vdev_islog ||
6501 !vdev_is_concrete(vml[c]) ||
6502 vml[c]->vdev_isspare ||
6503 vml[c]->vdev_isl2cache ||
6504 !vdev_writeable(vml[c]) ||
6505 vml[c]->vdev_children != 0 ||
6506 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6507 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6508 error = SET_ERROR(EINVAL);
6509 break;
6510 }
6511
6512 if (vdev_dtl_required(vml[c])) {
6513 error = SET_ERROR(EBUSY);
6514 break;
6515 }
6516
6517 /* we need certain info from the top level */
6518 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6519 vml[c]->vdev_top->vdev_ms_array) == 0);
6520 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6521 vml[c]->vdev_top->vdev_ms_shift) == 0);
6522 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6523 vml[c]->vdev_top->vdev_asize) == 0);
6524 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6525 vml[c]->vdev_top->vdev_ashift) == 0);
6526
6527 /* transfer per-vdev ZAPs */
6528 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6529 VERIFY0(nvlist_add_uint64(child[c],
6530 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6531
6532 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6533 VERIFY0(nvlist_add_uint64(child[c],
6534 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6535 vml[c]->vdev_parent->vdev_top_zap));
6536 }
6537
6538 if (error != 0) {
6539 kmem_free(vml, children * sizeof (vdev_t *));
6540 kmem_free(glist, children * sizeof (uint64_t));
6541 return (spa_vdev_exit(spa, NULL, txg, error));
6542 }
6543
6544 /* stop writers from using the disks */
6545 for (c = 0; c < children; c++) {
6546 if (vml[c] != NULL)
6547 vml[c]->vdev_offline = B_TRUE;
6548 }
6549 vdev_reopen(spa->spa_root_vdev);
6550
6551 /*
6552 * Temporarily record the splitting vdevs in the spa config. This
6553 * will disappear once the config is regenerated.
6554 */
6555 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6556 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6557 glist, children) == 0);
6558 kmem_free(glist, children * sizeof (uint64_t));
6559
6560 mutex_enter(&spa->spa_props_lock);
6561 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6562 nvl) == 0);
6563 mutex_exit(&spa->spa_props_lock);
6564 spa->spa_config_splitting = nvl;
6565 vdev_config_dirty(spa->spa_root_vdev);
6566
6567 /* configure and create the new pool */
6568 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6569 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6570 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6571 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6572 spa_version(spa)) == 0);
6573 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6574 spa->spa_config_txg) == 0);
6575 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6576 spa_generate_guid(NULL)) == 0);
6577 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6578 (void) nvlist_lookup_string(props,
6579 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6580
6581 /* add the new pool to the namespace */
6582 newspa = spa_add(newname, config, altroot);
6583 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6584 newspa->spa_config_txg = spa->spa_config_txg;
6585 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6586
6587 /* release the spa config lock, retaining the namespace lock */
6588 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6589
6590 if (zio_injection_enabled)
6591 zio_handle_panic_injection(spa, FTAG, 1);
6592
6593 spa_activate(newspa, spa_mode_global);
6594 spa_async_suspend(newspa);
6595
6596#ifndef illumos
6597 /* mark that we are creating new spa by splitting */
6598 newspa->spa_splitting_newspa = B_TRUE;
6599#endif
6600 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6601
6602 /* create the new pool from the disks of the original pool */
6603 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6604#ifndef illumos
6605 newspa->spa_splitting_newspa = B_FALSE;
6606#endif
6607 if (error)
6608 goto out;
6609
6610 /* if that worked, generate a real config for the new pool */
6611 if (newspa->spa_root_vdev != NULL) {
6612 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6613 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6614 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6615 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6616 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6617 B_TRUE));
6618 }
6619
6620 /* set the props */
6621 if (props != NULL) {
6622 spa_configfile_set(newspa, props, B_FALSE);
6623 error = spa_prop_set(newspa, props);
6624 if (error)
6625 goto out;
6626 }
6627
6628 /* flush everything */
6629 txg = spa_vdev_config_enter(newspa);
6630 vdev_config_dirty(newspa->spa_root_vdev);
6631 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6632
6633 if (zio_injection_enabled)
6634 zio_handle_panic_injection(spa, FTAG, 2);
6635
6636 spa_async_resume(newspa);
6637
6638 /* finally, update the original pool's config */
6639 txg = spa_vdev_config_enter(spa);
6640 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6641 error = dmu_tx_assign(tx, TXG_WAIT);
6642 if (error != 0)
6643 dmu_tx_abort(tx);
6644 for (c = 0; c < children; c++) {
6645 if (vml[c] != NULL) {
6646 vdev_split(vml[c]);
6647 if (error == 0)
6648 spa_history_log_internal(spa, "detach", tx,
6649 "vdev=%s", vml[c]->vdev_path);
6650
6651 vdev_free(vml[c]);
6652 }
6653 }
6654 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6655 vdev_config_dirty(spa->spa_root_vdev);
6656 spa->spa_config_splitting = NULL;
6657 nvlist_free(nvl);
6658 if (error == 0)
6659 dmu_tx_commit(tx);
6660 (void) spa_vdev_exit(spa, NULL, txg, 0);
6661
6662 if (zio_injection_enabled)
6663 zio_handle_panic_injection(spa, FTAG, 3);
6664
6665 /* split is complete; log a history record */
6666 spa_history_log_internal(newspa, "split", NULL,
6667 "from pool %s", spa_name(spa));
6668
6669 kmem_free(vml, children * sizeof (vdev_t *));
6670
6671 /* if we're not going to mount the filesystems in userland, export */
6672 if (exp)
6673 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6674 B_FALSE, B_FALSE);
6675
6676 return (error);
6677
6678out:
6679 spa_unload(newspa);
6680 spa_deactivate(newspa);
6681 spa_remove(newspa);
6682
6683 txg = spa_vdev_config_enter(spa);
6684
6685 /* re-online all offlined disks */
6686 for (c = 0; c < children; c++) {
6687 if (vml[c] != NULL)
6688 vml[c]->vdev_offline = B_FALSE;
6689 }
6690 vdev_reopen(spa->spa_root_vdev);
6691
6692 nvlist_free(spa->spa_config_splitting);
6693 spa->spa_config_splitting = NULL;
6694 (void) spa_vdev_exit(spa, NULL, txg, error);
6695
6696 kmem_free(vml, children * sizeof (vdev_t *));
6697 return (error);
6698}
6699
6700/*
6701 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6702 * currently spared, so we can detach it.
6703 */
6704static vdev_t *
6705spa_vdev_resilver_done_hunt(vdev_t *vd)
6706{
6707 vdev_t *newvd, *oldvd;
6708
6709 for (int c = 0; c < vd->vdev_children; c++) {
6710 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6711 if (oldvd != NULL)
6712 return (oldvd);
6713 }
6714
6715 /*
6716 * Check for a completed replacement. We always consider the first
6717 * vdev in the list to be the oldest vdev, and the last one to be
6718 * the newest (see spa_vdev_attach() for how that works). In
6719 * the case where the newest vdev is faulted, we will not automatically
6720 * remove it after a resilver completes. This is OK as it will require
6721 * user intervention to determine which disk the admin wishes to keep.
6722 */
6723 if (vd->vdev_ops == &vdev_replacing_ops) {
6724 ASSERT(vd->vdev_children > 1);
6725
6726 newvd = vd->vdev_child[vd->vdev_children - 1];
6727 oldvd = vd->vdev_child[0];
6728
6729 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6730 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6731 !vdev_dtl_required(oldvd))
6732 return (oldvd);
6733 }
6734
6735 /*
6736 * Check for a completed resilver with the 'unspare' flag set.
6737 */
6738 if (vd->vdev_ops == &vdev_spare_ops) {
6739 vdev_t *first = vd->vdev_child[0];
6740 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6741
6742 if (last->vdev_unspare) {
6743 oldvd = first;
6744 newvd = last;
6745 } else if (first->vdev_unspare) {
6746 oldvd = last;
6747 newvd = first;
6748 } else {
6749 oldvd = NULL;
6750 }
6751
6752 if (oldvd != NULL &&
6753 vdev_dtl_empty(newvd, DTL_MISSING) &&
6754 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6755 !vdev_dtl_required(oldvd))
6756 return (oldvd);
6757
6758 /*
6759 * If there are more than two spares attached to a disk,
6760 * and those spares are not required, then we want to
6761 * attempt to free them up now so that they can be used
6762 * by other pools. Once we're back down to a single
6763 * disk+spare, we stop removing them.
6764 */
6765 if (vd->vdev_children > 2) {
6766 newvd = vd->vdev_child[1];
6767
6768 if (newvd->vdev_isspare && last->vdev_isspare &&
6769 vdev_dtl_empty(last, DTL_MISSING) &&
6770 vdev_dtl_empty(last, DTL_OUTAGE) &&
6771 !vdev_dtl_required(newvd))
6772 return (newvd);
6773 }
6774 }
6775
6776 return (NULL);
6777}
6778
6779static void
6780spa_vdev_resilver_done(spa_t *spa)
6781{
6782 vdev_t *vd, *pvd, *ppvd;
6783 uint64_t guid, sguid, pguid, ppguid;
6784
6785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6786
6787 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6788 pvd = vd->vdev_parent;
6789 ppvd = pvd->vdev_parent;
6790 guid = vd->vdev_guid;
6791 pguid = pvd->vdev_guid;
6792 ppguid = ppvd->vdev_guid;
6793 sguid = 0;
6794 /*
6795 * If we have just finished replacing a hot spared device, then
6796 * we need to detach the parent's first child (the original hot
6797 * spare) as well.
6798 */
6799 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6800 ppvd->vdev_children == 2) {
6801 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6802 sguid = ppvd->vdev_child[1]->vdev_guid;
6803 }
6804 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6805
6806 spa_config_exit(spa, SCL_ALL, FTAG);
6807 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6808 return;
6809 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6810 return;
6811 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6812 }
6813
6814 spa_config_exit(spa, SCL_ALL, FTAG);
6815}
6816
6817/*
6818 * Update the stored path or FRU for this vdev.
6819 */
6820int
6821spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
6822 boolean_t ispath)
6823{
6824 vdev_t *vd;
6825 boolean_t sync = B_FALSE;
6826
6827 ASSERT(spa_writeable(spa));
6828
6829 spa_vdev_state_enter(spa, SCL_ALL);
6830
6831 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
6832 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6833
6834 if (!vd->vdev_ops->vdev_op_leaf)
6835 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6836
6837 if (ispath) {
6838 if (strcmp(value, vd->vdev_path) != 0) {
6839 spa_strfree(vd->vdev_path);
6840 vd->vdev_path = spa_strdup(value);
6841 sync = B_TRUE;
6842 }
6843 } else {
6844 if (vd->vdev_fru == NULL) {
6845 vd->vdev_fru = spa_strdup(value);
6846 sync = B_TRUE;
6847 } else if (strcmp(value, vd->vdev_fru) != 0) {
6848 spa_strfree(vd->vdev_fru);
6849 vd->vdev_fru = spa_strdup(value);
6850 sync = B_TRUE;
6851 }
6852 }
6853
6854 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6855}
6856
6857int
6858spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6859{
6860 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6861}
6862
6863int
6864spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6865{
6866 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6867}
6868
6869/*
6870 * ==========================================================================
6871 * SPA Scanning
6872 * ==========================================================================
6873 */
6874int
6875spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
6876{
6877 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6878
6879 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6880 return (SET_ERROR(EBUSY));
6881
6882 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
6883}
6884
6885int
6886spa_scan_stop(spa_t *spa)
6887{
6888 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6889 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6890 return (SET_ERROR(EBUSY));
6891 return (dsl_scan_cancel(spa->spa_dsl_pool));
6892}
6893
6894int
6895spa_scan(spa_t *spa, pool_scan_func_t func)
6896{
6897 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6898
6899 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6900 return (SET_ERROR(ENOTSUP));
6901
6902 /*
6903 * If a resilver was requested, but there is no DTL on a
6904 * writeable leaf device, we have nothing to do.
6905 */
6906 if (func == POOL_SCAN_RESILVER &&
6907 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6908 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6909 return (0);
6910 }
6911
6912 return (dsl_scan(spa->spa_dsl_pool, func));
6913}
6914
6915/*
6916 * ==========================================================================
6917 * SPA async task processing
6918 * ==========================================================================
6919 */
6920
6921static void
6922spa_async_remove(spa_t *spa, vdev_t *vd)
6923{
6924 if (vd->vdev_remove_wanted) {
6925 vd->vdev_remove_wanted = B_FALSE;
6926 vd->vdev_delayed_close = B_FALSE;
6927 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6928
6929 /*
6930 * We want to clear the stats, but we don't want to do a full
6931 * vdev_clear() as that will cause us to throw away
6932 * degraded/faulted state as well as attempt to reopen the
6933 * device, all of which is a waste.
6934 */
6935 vd->vdev_stat.vs_read_errors = 0;
6936 vd->vdev_stat.vs_write_errors = 0;
6937 vd->vdev_stat.vs_checksum_errors = 0;
6938
6939 vdev_state_dirty(vd->vdev_top);
6940 /* Tell userspace that the vdev is gone. */
6941 zfs_post_remove(spa, vd);
6942 }
6943
6944 for (int c = 0; c < vd->vdev_children; c++)
6945 spa_async_remove(spa, vd->vdev_child[c]);
6946}
6947
6948static void
6949spa_async_probe(spa_t *spa, vdev_t *vd)
6950{
6951 if (vd->vdev_probe_wanted) {
6952 vd->vdev_probe_wanted = B_FALSE;
6953 vdev_reopen(vd); /* vdev_open() does the actual probe */
6954 }
6955
6956 for (int c = 0; c < vd->vdev_children; c++)
6957 spa_async_probe(spa, vd->vdev_child[c]);
6958}
6959
6960static void
6961spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6962{
6963 sysevent_id_t eid;
6964 nvlist_t *attr;
6965 char *physpath;
6966
6967 if (!spa->spa_autoexpand)
6968 return;
6969
6970 for (int c = 0; c < vd->vdev_children; c++) {
6971 vdev_t *cvd = vd->vdev_child[c];
6972 spa_async_autoexpand(spa, cvd);
6973 }
6974
6975 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6976 return;
6977
6978 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6979 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6980
6981 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6982 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6983
6984 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6985 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6986
6987 nvlist_free(attr);
6988 kmem_free(physpath, MAXPATHLEN);
6989}
6990
6991static void
6992spa_async_thread(void *arg)
6993{
6994 spa_t *spa = (spa_t *)arg;
6995 int tasks;
6996
6997 ASSERT(spa->spa_sync_on);
6998
6999 mutex_enter(&spa->spa_async_lock);
7000 tasks = spa->spa_async_tasks;
7001 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
7002 mutex_exit(&spa->spa_async_lock);
7003
7004 /*
7005 * See if the config needs to be updated.
7006 */
7007 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7008 uint64_t old_space, new_space;
7009
7010 mutex_enter(&spa_namespace_lock);
7011 old_space = metaslab_class_get_space(spa_normal_class(spa));
7012 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7013 new_space = metaslab_class_get_space(spa_normal_class(spa));
7014 mutex_exit(&spa_namespace_lock);
7015
7016 /*
7017 * If the pool grew as a result of the config update,
7018 * then log an internal history event.
7019 */
7020 if (new_space != old_space) {
7021 spa_history_log_internal(spa, "vdev online", NULL,
7022 "pool '%s' size: %llu(+%llu)",
7023 spa_name(spa), new_space, new_space - old_space);
7024 }
7025 }
7026
7027 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7028 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7029 spa_async_autoexpand(spa, spa->spa_root_vdev);
7030 spa_config_exit(spa, SCL_CONFIG, FTAG);
7031 }
7032
7033 /*
7034 * See if any devices need to be probed.
7035 */
7036 if (tasks & SPA_ASYNC_PROBE) {
7037 spa_vdev_state_enter(spa, SCL_NONE);
7038 spa_async_probe(spa, spa->spa_root_vdev);
7039 (void) spa_vdev_state_exit(spa, NULL, 0);
7040 }
7041
7042 /*
7043 * If any devices are done replacing, detach them.
7044 */
7045 if (tasks & SPA_ASYNC_RESILVER_DONE)
7046 spa_vdev_resilver_done(spa);
7047
7048 /*
7049 * Kick off a resilver.
7050 */
7051 if (tasks & SPA_ASYNC_RESILVER)
7052 dsl_resilver_restart(spa->spa_dsl_pool, 0);
7053
7054 /*
7055 * Let the world know that we're done.
7056 */
7057 mutex_enter(&spa->spa_async_lock);
7058 spa->spa_async_thread = NULL;
7059 cv_broadcast(&spa->spa_async_cv);
7060 mutex_exit(&spa->spa_async_lock);
7061 thread_exit();
7062}
7063
7064static void
7065spa_async_thread_vd(void *arg)
7066{
7067 spa_t *spa = arg;
7068 int tasks;
7069
7070 mutex_enter(&spa->spa_async_lock);
7071 tasks = spa->spa_async_tasks;
7072retry:
7073 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
7074 mutex_exit(&spa->spa_async_lock);
7075
7076 /*
7077 * See if any devices need to be marked REMOVED.
7078 */
7079 if (tasks & SPA_ASYNC_REMOVE) {
7080 spa_vdev_state_enter(spa, SCL_NONE);
7081 spa_async_remove(spa, spa->spa_root_vdev);
7082 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7083 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7084 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7085 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7086 (void) spa_vdev_state_exit(spa, NULL, 0);
7087 }
7088
7089 /*
7090 * Let the world know that we're done.
7091 */
7092 mutex_enter(&spa->spa_async_lock);
7093 tasks = spa->spa_async_tasks;
7094 if ((tasks & SPA_ASYNC_REMOVE) != 0)
7095 goto retry;
7096 spa->spa_async_thread_vd = NULL;
7097 cv_broadcast(&spa->spa_async_cv);
7098 mutex_exit(&spa->spa_async_lock);
7099 thread_exit();
7100}
7101
7102void
7103spa_async_suspend(spa_t *spa)
7104{
7105 mutex_enter(&spa->spa_async_lock);
7106 spa->spa_async_suspended++;
7107 while (spa->spa_async_thread != NULL ||
7108 spa->spa_async_thread_vd != NULL)
7109 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
7110 mutex_exit(&spa->spa_async_lock);
7111
7112 spa_vdev_remove_suspend(spa);
7113
7114 zthr_t *condense_thread = spa->spa_condense_zthr;
7115 if (condense_thread != NULL && zthr_isrunning(condense_thread))
7116 VERIFY0(zthr_cancel(condense_thread));
7117
7118 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7119 if (discard_thread != NULL && zthr_isrunning(discard_thread))
7120 VERIFY0(zthr_cancel(discard_thread));
7121}
7122
7123void
7124spa_async_resume(spa_t *spa)
7125{
7126 mutex_enter(&spa->spa_async_lock);
7127 ASSERT(spa->spa_async_suspended != 0);
7128 spa->spa_async_suspended--;
7129 mutex_exit(&spa->spa_async_lock);
7130 spa_restart_removal(spa);
7131
7132 zthr_t *condense_thread = spa->spa_condense_zthr;
7133 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
7134 zthr_resume(condense_thread);
7135
7136 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7137 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
7138 zthr_resume(discard_thread);
7139}
7140
7141static boolean_t
7142spa_async_tasks_pending(spa_t *spa)
7143{
7144 uint_t non_config_tasks;
7145 uint_t config_task;
7146 boolean_t config_task_suspended;
7147
7148 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
7149 SPA_ASYNC_REMOVE);
7150 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
7151 if (spa->spa_ccw_fail_time == 0) {
7152 config_task_suspended = B_FALSE;
7153 } else {
7154 config_task_suspended =
7155 (gethrtime() - spa->spa_ccw_fail_time) <
7156 (zfs_ccw_retry_interval * NANOSEC);
7157 }
7158
7159 return (non_config_tasks || (config_task && !config_task_suspended));
7160}
7161
7162static void
7163spa_async_dispatch(spa_t *spa)
7164{
7165 mutex_enter(&spa->spa_async_lock);
7166 if (spa_async_tasks_pending(spa) &&
7167 !spa->spa_async_suspended &&
7168 spa->spa_async_thread == NULL &&
7169 rootdir != NULL)
7170 spa->spa_async_thread = thread_create(NULL, 0,
7171 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
7172 mutex_exit(&spa->spa_async_lock);
7173}
7174
7175static void
7176spa_async_dispatch_vd(spa_t *spa)
7177{
7178 mutex_enter(&spa->spa_async_lock);
7179 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
7180 !spa->spa_async_suspended &&
7181 spa->spa_async_thread_vd == NULL &&
7182 rootdir != NULL)
7183 spa->spa_async_thread_vd = thread_create(NULL, 0,
7184 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
7185 mutex_exit(&spa->spa_async_lock);
7186}
7187
7188void
7189spa_async_request(spa_t *spa, int task)
7190{
7191 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
7192 mutex_enter(&spa->spa_async_lock);
7193 spa->spa_async_tasks |= task;
7194 mutex_exit(&spa->spa_async_lock);
7195 spa_async_dispatch_vd(spa);
7196}
7197
7198/*
7199 * ==========================================================================
7200 * SPA syncing routines
7201 * ==========================================================================
7202 */
7203
7204static int
7205bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7206{
7207 bpobj_t *bpo = arg;
7208 bpobj_enqueue(bpo, bp, tx);
7209 return (0);
7210}
7211
7212static int
7213spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7214{
7215 zio_t *zio = arg;
7216
7217 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7218 BP_GET_PSIZE(bp), zio->io_flags));
7219 return (0);
7220}
7221
7222/*
7223 * Note: this simple function is not inlined to make it easier to dtrace the
7224 * amount of time spent syncing frees.
7225 */
7226static void
7227spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7228{
7229 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7230 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7231 VERIFY(zio_wait(zio) == 0);
7232}
7233
7234/*
7235 * Note: this simple function is not inlined to make it easier to dtrace the
7236 * amount of time spent syncing deferred frees.
7237 */
7238static void
7239spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7240{
7241 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7242 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7243 spa_free_sync_cb, zio, tx), ==, 0);
7244 VERIFY0(zio_wait(zio));
7245}
7246
7247
7248static void
7249spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7250{
7251 char *packed = NULL;
7252 size_t bufsize;
7253 size_t nvsize = 0;
7254 dmu_buf_t *db;
7255
7256 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7257
7258 /*
7259 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7260 * information. This avoids the dmu_buf_will_dirty() path and
7261 * saves us a pre-read to get data we don't actually care about.
7262 */
7263 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7264 packed = kmem_alloc(bufsize, KM_SLEEP);
7265
7266 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7267 KM_SLEEP) == 0);
7268 bzero(packed + nvsize, bufsize - nvsize);
7269
7270 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7271
7272 kmem_free(packed, bufsize);
7273
7274 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7275 dmu_buf_will_dirty(db, tx);
7276 *(uint64_t *)db->db_data = nvsize;
7277 dmu_buf_rele(db, FTAG);
7278}
7279
7280static void
7281spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7282 const char *config, const char *entry)
7283{
7284 nvlist_t *nvroot;
7285 nvlist_t **list;
7286 int i;
7287
7288 if (!sav->sav_sync)
7289 return;
7290
7291 /*
7292 * Update the MOS nvlist describing the list of available devices.
7293 * spa_validate_aux() will have already made sure this nvlist is
7294 * valid and the vdevs are labeled appropriately.
7295 */
7296 if (sav->sav_object == 0) {
7297 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7298 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7299 sizeof (uint64_t), tx);
7300 VERIFY(zap_update(spa->spa_meta_objset,
7301 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7302 &sav->sav_object, tx) == 0);
7303 }
7304
7305 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7306 if (sav->sav_count == 0) {
7307 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7308 } else {
7309 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7310 for (i = 0; i < sav->sav_count; i++)
7311 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7312 B_FALSE, VDEV_CONFIG_L2CACHE);
7313 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7314 sav->sav_count) == 0);
7315 for (i = 0; i < sav->sav_count; i++)
7316 nvlist_free(list[i]);
7317 kmem_free(list, sav->sav_count * sizeof (void *));
7318 }
7319
7320 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7321 nvlist_free(nvroot);
7322
7323 sav->sav_sync = B_FALSE;
7324}
7325
7326/*
7327 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7328 * The all-vdev ZAP must be empty.
7329 */
7330static void
7331spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7332{
7333 spa_t *spa = vd->vdev_spa;
7334 if (vd->vdev_top_zap != 0) {
7335 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7336 vd->vdev_top_zap, tx));
7337 }
7338 if (vd->vdev_leaf_zap != 0) {
7339 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7340 vd->vdev_leaf_zap, tx));
7341 }
7342 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7343 spa_avz_build(vd->vdev_child[i], avz, tx);
7344 }
7345}
7346
7347static void
7348spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7349{
7350 nvlist_t *config;
7351
7352 /*
7353 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7354 * its config may not be dirty but we still need to build per-vdev ZAPs.
7355 * Similarly, if the pool is being assembled (e.g. after a split), we
7356 * need to rebuild the AVZ although the config may not be dirty.
7357 */
7358 if (list_is_empty(&spa->spa_config_dirty_list) &&
7359 spa->spa_avz_action == AVZ_ACTION_NONE)
7360 return;
7361
7362 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7363
7364 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7365 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7366 spa->spa_all_vdev_zaps != 0);
7367
7368 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7369 /* Make and build the new AVZ */
7370 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7371 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7372 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7373
7374 /* Diff old AVZ with new one */
7375 zap_cursor_t zc;
7376 zap_attribute_t za;
7377
7378 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7379 spa->spa_all_vdev_zaps);
7380 zap_cursor_retrieve(&zc, &za) == 0;
7381 zap_cursor_advance(&zc)) {
7382 uint64_t vdzap = za.za_first_integer;
7383 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7384 vdzap) == ENOENT) {
7385 /*
7386 * ZAP is listed in old AVZ but not in new one;
7387 * destroy it
7388 */
7389 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7390 tx));
7391 }
7392 }
7393
7394 zap_cursor_fini(&zc);
7395
7396 /* Destroy the old AVZ */
7397 VERIFY0(zap_destroy(spa->spa_meta_objset,
7398 spa->spa_all_vdev_zaps, tx));
7399
7400 /* Replace the old AVZ in the dir obj with the new one */
7401 VERIFY0(zap_update(spa->spa_meta_objset,
7402 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7403 sizeof (new_avz), 1, &new_avz, tx));
7404
7405 spa->spa_all_vdev_zaps = new_avz;
7406 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7407 zap_cursor_t zc;
7408 zap_attribute_t za;
7409
7410 /* Walk through the AVZ and destroy all listed ZAPs */
7411 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7412 spa->spa_all_vdev_zaps);
7413 zap_cursor_retrieve(&zc, &za) == 0;
7414 zap_cursor_advance(&zc)) {
7415 uint64_t zap = za.za_first_integer;
7416 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7417 }
7418
7419 zap_cursor_fini(&zc);
7420
7421 /* Destroy and unlink the AVZ itself */
7422 VERIFY0(zap_destroy(spa->spa_meta_objset,
7423 spa->spa_all_vdev_zaps, tx));
7424 VERIFY0(zap_remove(spa->spa_meta_objset,
7425 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7426 spa->spa_all_vdev_zaps = 0;
7427 }
7428
7429 if (spa->spa_all_vdev_zaps == 0) {
7430 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7431 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7432 DMU_POOL_VDEV_ZAP_MAP, tx);
7433 }
7434 spa->spa_avz_action = AVZ_ACTION_NONE;
7435
7436 /* Create ZAPs for vdevs that don't have them. */
7437 vdev_construct_zaps(spa->spa_root_vdev, tx);
7438
7439 config = spa_config_generate(spa, spa->spa_root_vdev,
7440 dmu_tx_get_txg(tx), B_FALSE);
7441
7442 /*
7443 * If we're upgrading the spa version then make sure that
7444 * the config object gets updated with the correct version.
7445 */
7446 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7447 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7448 spa->spa_uberblock.ub_version);
7449
7450 spa_config_exit(spa, SCL_STATE, FTAG);
7451
7452 nvlist_free(spa->spa_config_syncing);
7453 spa->spa_config_syncing = config;
7454
7455 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7456}
7457
7458static void
7459spa_sync_version(void *arg, dmu_tx_t *tx)
7460{
7461 uint64_t *versionp = arg;
7462 uint64_t version = *versionp;
7463 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7464
7465 /*
7466 * Setting the version is special cased when first creating the pool.
7467 */
7468 ASSERT(tx->tx_txg != TXG_INITIAL);
7469
7470 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7471 ASSERT(version >= spa_version(spa));
7472
7473 spa->spa_uberblock.ub_version = version;
7474 vdev_config_dirty(spa->spa_root_vdev);
7475 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7476}
7477
7478/*
7479 * Set zpool properties.
7480 */
7481static void
7482spa_sync_props(void *arg, dmu_tx_t *tx)
7483{
7484 nvlist_t *nvp = arg;
7485 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7486 objset_t *mos = spa->spa_meta_objset;
7487 nvpair_t *elem = NULL;
7488
7489 mutex_enter(&spa->spa_props_lock);
7490
7491 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7492 uint64_t intval;
7493 char *strval, *fname;
7494 zpool_prop_t prop;
7495 const char *propname;
7496 zprop_type_t proptype;
7497 spa_feature_t fid;
7498
7499 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7500 case ZPOOL_PROP_INVAL:
7501 /*
7502 * We checked this earlier in spa_prop_validate().
7503 */
7504 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7505
7506 fname = strchr(nvpair_name(elem), '@') + 1;
7507 VERIFY0(zfeature_lookup_name(fname, &fid));
7508
7509 spa_feature_enable(spa, fid, tx);
7510 spa_history_log_internal(spa, "set", tx,
7511 "%s=enabled", nvpair_name(elem));
7512 break;
7513
7514 case ZPOOL_PROP_VERSION:
7515 intval = fnvpair_value_uint64(elem);
7516 /*
7517 * The version is synced seperatly before other
7518 * properties and should be correct by now.
7519 */
7520 ASSERT3U(spa_version(spa), >=, intval);
7521 break;
7522
7523 case ZPOOL_PROP_ALTROOT:
7524 /*
7525 * 'altroot' is a non-persistent property. It should
7526 * have been set temporarily at creation or import time.
7527 */
7528 ASSERT(spa->spa_root != NULL);
7529 break;
7530
7531 case ZPOOL_PROP_READONLY:
7532 case ZPOOL_PROP_CACHEFILE:
7533 /*
7534 * 'readonly' and 'cachefile' are also non-persisitent
7535 * properties.
7536 */
7537 break;
7538 case ZPOOL_PROP_COMMENT:
7539 strval = fnvpair_value_string(elem);
7540 if (spa->spa_comment != NULL)
7541 spa_strfree(spa->spa_comment);
7542 spa->spa_comment = spa_strdup(strval);
7543 /*
7544 * We need to dirty the configuration on all the vdevs
7545 * so that their labels get updated. It's unnecessary
7546 * to do this for pool creation since the vdev's
7547 * configuratoin has already been dirtied.
7548 */
7549 if (tx->tx_txg != TXG_INITIAL)
7550 vdev_config_dirty(spa->spa_root_vdev);
7551 spa_history_log_internal(spa, "set", tx,
7552 "%s=%s", nvpair_name(elem), strval);
7553 break;
7554 default:
7555 /*
7556 * Set pool property values in the poolprops mos object.
7557 */
7558 if (spa->spa_pool_props_object == 0) {
7559 spa->spa_pool_props_object =
7560 zap_create_link(mos, DMU_OT_POOL_PROPS,
7561 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7562 tx);
7563 }
7564
7565 /* normalize the property name */
7566 propname = zpool_prop_to_name(prop);
7567 proptype = zpool_prop_get_type(prop);
7568
7569 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7570 ASSERT(proptype == PROP_TYPE_STRING);
7571 strval = fnvpair_value_string(elem);
7572 VERIFY0(zap_update(mos,
7573 spa->spa_pool_props_object, propname,
7574 1, strlen(strval) + 1, strval, tx));
7575 spa_history_log_internal(spa, "set", tx,
7576 "%s=%s", nvpair_name(elem), strval);
7577 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7578 intval = fnvpair_value_uint64(elem);
7579
7580 if (proptype == PROP_TYPE_INDEX) {
7581 const char *unused;
7582 VERIFY0(zpool_prop_index_to_string(
7583 prop, intval, &unused));
7584 }
7585 VERIFY0(zap_update(mos,
7586 spa->spa_pool_props_object, propname,
7587 8, 1, &intval, tx));
7588 spa_history_log_internal(spa, "set", tx,
7589 "%s=%lld", nvpair_name(elem), intval);
7590 } else {
7591 ASSERT(0); /* not allowed */
7592 }
7593
7594 switch (prop) {
7595 case ZPOOL_PROP_DELEGATION:
7596 spa->spa_delegation = intval;
7597 break;
7598 case ZPOOL_PROP_BOOTFS:
7599 spa->spa_bootfs = intval;
7600 break;
7601 case ZPOOL_PROP_FAILUREMODE:
7602 spa->spa_failmode = intval;
7603 break;
7604 case ZPOOL_PROP_AUTOEXPAND:
7605 spa->spa_autoexpand = intval;
7606 if (tx->tx_txg != TXG_INITIAL)
7607 spa_async_request(spa,
7608 SPA_ASYNC_AUTOEXPAND);
7609 break;
7610 case ZPOOL_PROP_DEDUPDITTO:
7611 spa->spa_dedup_ditto = intval;
7612 break;
7613 default:
7614 break;
7615 }
7616 }
7617
7618 }
7619
7620 mutex_exit(&spa->spa_props_lock);
7621}
7622
7623/*
7624 * Perform one-time upgrade on-disk changes. spa_version() does not
7625 * reflect the new version this txg, so there must be no changes this
7626 * txg to anything that the upgrade code depends on after it executes.
7627 * Therefore this must be called after dsl_pool_sync() does the sync
7628 * tasks.
7629 */
7630static void
7631spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7632{
7633 dsl_pool_t *dp = spa->spa_dsl_pool;
7634
7635 ASSERT(spa->spa_sync_pass == 1);
7636
7637 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7638
7639 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7640 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7641 dsl_pool_create_origin(dp, tx);
7642
7643 /* Keeping the origin open increases spa_minref */
7644 spa->spa_minref += 3;
7645 }
7646
7647 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7648 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7649 dsl_pool_upgrade_clones(dp, tx);
7650 }
7651
7652 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7653 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7654 dsl_pool_upgrade_dir_clones(dp, tx);
7655
7656 /* Keeping the freedir open increases spa_minref */
7657 spa->spa_minref += 3;
7658 }
7659
7660 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7661 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7662 spa_feature_create_zap_objects(spa, tx);
7663 }
7664
7665 /*
7666 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7667 * when possibility to use lz4 compression for metadata was added
7668 * Old pools that have this feature enabled must be upgraded to have
7669 * this feature active
7670 */
7671 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7672 boolean_t lz4_en = spa_feature_is_enabled(spa,
7673 SPA_FEATURE_LZ4_COMPRESS);
7674 boolean_t lz4_ac = spa_feature_is_active(spa,
7675 SPA_FEATURE_LZ4_COMPRESS);
7676
7677 if (lz4_en && !lz4_ac)
7678 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7679 }
7680
7681 /*
7682 * If we haven't written the salt, do so now. Note that the
7683 * feature may not be activated yet, but that's fine since
7684 * the presence of this ZAP entry is backwards compatible.
7685 */
7686 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7687 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7688 VERIFY0(zap_add(spa->spa_meta_objset,
7689 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7690 sizeof (spa->spa_cksum_salt.zcs_bytes),
7691 spa->spa_cksum_salt.zcs_bytes, tx));
7692 }
7693
7694 rrw_exit(&dp->dp_config_rwlock, FTAG);
7695}
7696
7697static void
7698vdev_indirect_state_sync_verify(vdev_t *vd)
7699{
7700 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7701 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7702
7703 if (vd->vdev_ops == &vdev_indirect_ops) {
7704 ASSERT(vim != NULL);
7705 ASSERT(vib != NULL);
7706 }
7707
7708 if (vdev_obsolete_sm_object(vd) != 0) {
7709 ASSERT(vd->vdev_obsolete_sm != NULL);
7710 ASSERT(vd->vdev_removing ||
7711 vd->vdev_ops == &vdev_indirect_ops);
7712 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7713 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7714
7715 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7716 space_map_object(vd->vdev_obsolete_sm));
7717 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7718 space_map_allocated(vd->vdev_obsolete_sm));
7719 }
7720 ASSERT(vd->vdev_obsolete_segments != NULL);
7721
7722 /*
7723 * Since frees / remaps to an indirect vdev can only
7724 * happen in syncing context, the obsolete segments
7725 * tree must be empty when we start syncing.
7726 */
7727 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7728}
7729
7730/*
7731 * Sync the specified transaction group. New blocks may be dirtied as
7732 * part of the process, so we iterate until it converges.
7733 */
7734void
7735spa_sync(spa_t *spa, uint64_t txg)
7736{
7737 dsl_pool_t *dp = spa->spa_dsl_pool;
7738 objset_t *mos = spa->spa_meta_objset;
7739 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7740 vdev_t *rvd = spa->spa_root_vdev;
7741 vdev_t *vd;
7742 dmu_tx_t *tx;
7743 int error;
7744 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7745 zfs_vdev_queue_depth_pct / 100;
7746
7747 VERIFY(spa_writeable(spa));
7748
7749 /*
7750 * Wait for i/os issued in open context that need to complete
7751 * before this txg syncs.
7752 */
7753 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK]));
7754 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0);
7755
7756 /*
7757 * Lock out configuration changes.
7758 */
7759 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7760
7761 spa->spa_syncing_txg = txg;
7762 spa->spa_sync_pass = 0;
7763
7764 mutex_enter(&spa->spa_alloc_lock);
7765 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7766 mutex_exit(&spa->spa_alloc_lock);
7767
7768 /*
7769 * If there are any pending vdev state changes, convert them
7770 * into config changes that go out with this transaction group.
7771 */
7772 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7773 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7774 /*
7775 * We need the write lock here because, for aux vdevs,
7776 * calling vdev_config_dirty() modifies sav_config.
7777 * This is ugly and will become unnecessary when we
7778 * eliminate the aux vdev wart by integrating all vdevs
7779 * into the root vdev tree.
7780 */
7781 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7782 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7783 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7784 vdev_state_clean(vd);
7785 vdev_config_dirty(vd);
7786 }
7787 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7788 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7789 }
7790 spa_config_exit(spa, SCL_STATE, FTAG);
7791
7792 tx = dmu_tx_create_assigned(dp, txg);
7793
7794 spa->spa_sync_starttime = gethrtime();
7795#ifdef illumos
7796 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7797 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7798#else /* !illumos */
7799#ifdef _KERNEL
7800 callout_schedule(&spa->spa_deadman_cycid,
7801 hz * spa->spa_deadman_synctime / NANOSEC);
7802#endif
7803#endif /* illumos */
7804
7805 /*
7806 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7807 * set spa_deflate if we have no raid-z vdevs.
7808 */
7809 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
7810 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
7811 int i;
7812
7813 for (i = 0; i < rvd->vdev_children; i++) {
7814 vd = rvd->vdev_child[i];
7815 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
7816 break;
7817 }
7818 if (i == rvd->vdev_children) {
7819 spa->spa_deflate = TRUE;
7820 VERIFY(0 == zap_add(spa->spa_meta_objset,
7821 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
7822 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
7823 }
7824 }
7825
7826 /*
7827 * Set the top-level vdev's max queue depth. Evaluate each
7828 * top-level's async write queue depth in case it changed.
7829 * The max queue depth will not change in the middle of syncing
7830 * out this txg.
7831 */
7832 uint64_t queue_depth_total = 0;
7833 for (int c = 0; c < rvd->vdev_children; c++) {
7834 vdev_t *tvd = rvd->vdev_child[c];
7835 metaslab_group_t *mg = tvd->vdev_mg;
7836
7837 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
7838 !metaslab_group_initialized(mg))
7839 continue;
7840
7841 /*
7842 * It is safe to do a lock-free check here because only async
7843 * allocations look at mg_max_alloc_queue_depth, and async
7844 * allocations all happen from spa_sync().
7845 */
7846 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
7847 mg->mg_max_alloc_queue_depth = max_queue_depth;
7848 queue_depth_total += mg->mg_max_alloc_queue_depth;
7849 }
7850 metaslab_class_t *mc = spa_normal_class(spa);
7851 ASSERT0(refcount_count(&mc->mc_alloc_slots));
7852 mc->mc_alloc_max_slots = queue_depth_total;
7853 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
7854
7855 ASSERT3U(mc->mc_alloc_max_slots, <=,
7856 max_queue_depth * rvd->vdev_children);
7857
7858 for (int c = 0; c < rvd->vdev_children; c++) {
7859 vdev_t *vd = rvd->vdev_child[c];
7860 vdev_indirect_state_sync_verify(vd);
7861
7862 if (vdev_indirect_should_condense(vd)) {
7863 spa_condense_indirect_start_sync(vd, tx);
7864 break;
7865 }
7866 }
7867
7868 /*
7869 * Iterate to convergence.
7870 */
7871 do {
7872 int pass = ++spa->spa_sync_pass;
7873
7874 spa_sync_config_object(spa, tx);
7875 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
7876 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
7877 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
7878 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
7879 spa_errlog_sync(spa, txg);
7880 dsl_pool_sync(dp, txg);
7881
7882 if (pass < zfs_sync_pass_deferred_free) {
7883 spa_sync_frees(spa, free_bpl, tx);
7884 } else {
7885 /*
7886 * We can not defer frees in pass 1, because
7887 * we sync the deferred frees later in pass 1.
7888 */
7889 ASSERT3U(pass, >, 1);
7890 bplist_iterate(free_bpl, bpobj_enqueue_cb,
7891 &spa->spa_deferred_bpobj, tx);
7892 }
7893
7894 ddt_sync(spa, txg);
7895 dsl_scan_sync(dp, tx);
7896
7897 if (spa->spa_vdev_removal != NULL)
7898 svr_sync(spa, tx);
7899
7900 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
7901 != NULL)
7902 vdev_sync(vd, txg);
7903
7904 if (pass == 1) {
7905 spa_sync_upgrades(spa, tx);
7906 ASSERT3U(txg, >=,
7907 spa->spa_uberblock.ub_rootbp.blk_birth);
7908 /*
7909 * Note: We need to check if the MOS is dirty
7910 * because we could have marked the MOS dirty
7911 * without updating the uberblock (e.g. if we
7912 * have sync tasks but no dirty user data). We
7913 * need to check the uberblock's rootbp because
7914 * it is updated if we have synced out dirty
7915 * data (though in this case the MOS will most
7916 * likely also be dirty due to second order
7917 * effects, we don't want to rely on that here).
7918 */
7919 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7920 !dmu_objset_is_dirty(mos, txg)) {
7921 /*
7922 * Nothing changed on the first pass,
7923 * therefore this TXG is a no-op. Avoid
7924 * syncing deferred frees, so that we
7925 * can keep this TXG as a no-op.
7926 */
7927 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7928 txg));
7929 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7930 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7931 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
7932 txg));
7933 break;
7934 }
7935 spa_sync_deferred_frees(spa, tx);
7936 }
7937
7938 } while (dmu_objset_is_dirty(mos, txg));
7939
7940 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7941 /*
7942 * Make sure that the number of ZAPs for all the vdevs matches
7943 * the number of ZAPs in the per-vdev ZAP list. This only gets
7944 * called if the config is dirty; otherwise there may be
7945 * outstanding AVZ operations that weren't completed in
7946 * spa_sync_config_object.
7947 */
7948 uint64_t all_vdev_zap_entry_count;
7949 ASSERT0(zap_count(spa->spa_meta_objset,
7950 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7951 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7952 all_vdev_zap_entry_count);
7953 }
7954
7955 if (spa->spa_vdev_removal != NULL) {
7956 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
7957 }
7958
7959 /*
7960 * Rewrite the vdev configuration (which includes the uberblock)
7961 * to commit the transaction group.
7962 *
7963 * If there are no dirty vdevs, we sync the uberblock to a few
7964 * random top-level vdevs that are known to be visible in the
7965 * config cache (see spa_vdev_add() for a complete description).
7966 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7967 */
7968 for (;;) {
7969 /*
7970 * We hold SCL_STATE to prevent vdev open/close/etc.
7971 * while we're attempting to write the vdev labels.
7972 */
7973 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7974
7975 if (list_is_empty(&spa->spa_config_dirty_list)) {
7976 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
7977 int svdcount = 0;
7978 int children = rvd->vdev_children;
7979 int c0 = spa_get_random(children);
7980
7981 for (int c = 0; c < children; c++) {
7982 vd = rvd->vdev_child[(c0 + c) % children];
7983
7984 /* Stop when revisiting the first vdev */
7985 if (c > 0 && svd[0] == vd)
7986 break;
7987
7988 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
7989 !vdev_is_concrete(vd))
7990 continue;
7991
7992 svd[svdcount++] = vd;
7993 if (svdcount == SPA_SYNC_MIN_VDEVS)
7994 break;
7995 }
7996 error = vdev_config_sync(svd, svdcount, txg);
7997 } else {
7998 error = vdev_config_sync(rvd->vdev_child,
7999 rvd->vdev_children, txg);
8000 }
8001
8002 if (error == 0)
8003 spa->spa_last_synced_guid = rvd->vdev_guid;
8004
8005 spa_config_exit(spa, SCL_STATE, FTAG);
8006
8007 if (error == 0)
8008 break;
8009 zio_suspend(spa, NULL);
8010 zio_resume_wait(spa);
8011 }
8012 dmu_tx_commit(tx);
8013
8014#ifdef illumos
8015 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
8016#else /* !illumos */
8017#ifdef _KERNEL
8018 callout_drain(&spa->spa_deadman_cycid);
8019#endif
8020#endif /* illumos */
8021
8022 /*
8023 * Clear the dirty config list.
8024 */
8025 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
8026 vdev_config_clean(vd);
8027
8028 /*
8029 * Now that the new config has synced transactionally,
8030 * let it become visible to the config cache.
8031 */
8032 if (spa->spa_config_syncing != NULL) {
8033 spa_config_set(spa, spa->spa_config_syncing);
8034 spa->spa_config_txg = txg;
8035 spa->spa_config_syncing = NULL;
8036 }
8037
8038 dsl_pool_sync_done(dp, txg);
8039
8040 mutex_enter(&spa->spa_alloc_lock);
8041 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
8042 mutex_exit(&spa->spa_alloc_lock);
8043
8044 /*
8045 * Update usable space statistics.
8046 */
8047 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
8048 vdev_sync_done(vd, txg);
8049
8050 spa_update_dspace(spa);
8051
8052 /*
8053 * It had better be the case that we didn't dirty anything
8054 * since vdev_config_sync().
8055 */
8056 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8057 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8058 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
8059
8060 while (zfs_pause_spa_sync)
8061 delay(1);
8062
8063 spa->spa_sync_pass = 0;
8064
8065 /*
8066 * Update the last synced uberblock here. We want to do this at
8067 * the end of spa_sync() so that consumers of spa_last_synced_txg()
8068 * will be guaranteed that all the processing associated with
8069 * that txg has been completed.
8070 */
8071 spa->spa_ubsync = spa->spa_uberblock;
8072 spa_config_exit(spa, SCL_CONFIG, FTAG);
8073
8074 spa_handle_ignored_writes(spa);
8075
8076 /*
8077 * If any async tasks have been requested, kick them off.
8078 */
8079 spa_async_dispatch(spa);
8080 spa_async_dispatch_vd(spa);
8081}
8082
8083/*
8084 * Sync all pools. We don't want to hold the namespace lock across these
8085 * operations, so we take a reference on the spa_t and drop the lock during the
8086 * sync.
8087 */
8088void
8089spa_sync_allpools(void)
8090{
8091 spa_t *spa = NULL;
8092 mutex_enter(&spa_namespace_lock);
8093 while ((spa = spa_next(spa)) != NULL) {
8094 if (spa_state(spa) != POOL_STATE_ACTIVE ||
8095 !spa_writeable(spa) || spa_suspended(spa))
8096 continue;
8097 spa_open_ref(spa, FTAG);
8098 mutex_exit(&spa_namespace_lock);
8099 txg_wait_synced(spa_get_dsl(spa), 0);
8100 mutex_enter(&spa_namespace_lock);
8101 spa_close(spa, FTAG);
8102 }
8103 mutex_exit(&spa_namespace_lock);
8104}
8105
8106/*
8107 * ==========================================================================
8108 * Miscellaneous routines
8109 * ==========================================================================
8110 */
8111
8112/*
8113 * Remove all pools in the system.
8114 */
8115void
8116spa_evict_all(void)
8117{
8118 spa_t *spa;
8119
8120 /*
8121 * Remove all cached state. All pools should be closed now,
8122 * so every spa in the AVL tree should be unreferenced.
8123 */
8124 mutex_enter(&spa_namespace_lock);
8125 while ((spa = spa_next(NULL)) != NULL) {
8126 /*
8127 * Stop async tasks. The async thread may need to detach
8128 * a device that's been replaced, which requires grabbing
8129 * spa_namespace_lock, so we must drop it here.
8130 */
8131 spa_open_ref(spa, FTAG);
8132 mutex_exit(&spa_namespace_lock);
8133 spa_async_suspend(spa);
8134 mutex_enter(&spa_namespace_lock);
8135 spa_close(spa, FTAG);
8136
8137 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
8138 spa_unload(spa);
8139 spa_deactivate(spa);
8140 }
8141 spa_remove(spa);
8142 }
8143 mutex_exit(&spa_namespace_lock);
8144}
8145
8146vdev_t *
8147spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
8148{
8149 vdev_t *vd;
8150 int i;
8151
8152 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
8153 return (vd);
8154
8155 if (aux) {
8156 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
8157 vd = spa->spa_l2cache.sav_vdevs[i];
8158 if (vd->vdev_guid == guid)
8159 return (vd);
8160 }
8161
8162 for (i = 0; i < spa->spa_spares.sav_count; i++) {
8163 vd = spa->spa_spares.sav_vdevs[i];
8164 if (vd->vdev_guid == guid)
8165 return (vd);
8166 }
8167 }
8168
8169 return (NULL);
8170}
8171
8172void
8173spa_upgrade(spa_t *spa, uint64_t version)
8174{
8175 ASSERT(spa_writeable(spa));
8176
8177 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
8178
8179 /*
8180 * This should only be called for a non-faulted pool, and since a
8181 * future version would result in an unopenable pool, this shouldn't be
8182 * possible.
8183 */
8184 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
8185 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
8186
8187 spa->spa_uberblock.ub_version = version;
8188 vdev_config_dirty(spa->spa_root_vdev);
8189
8190 spa_config_exit(spa, SCL_ALL, FTAG);
8191
8192 txg_wait_synced(spa_get_dsl(spa), 0);
8193}
8194
8195boolean_t
8196spa_has_spare(spa_t *spa, uint64_t guid)
8197{
8198 int i;
8199 uint64_t spareguid;
8200 spa_aux_vdev_t *sav = &spa->spa_spares;
8201
8202 for (i = 0; i < sav->sav_count; i++)
8203 if (sav->sav_vdevs[i]->vdev_guid == guid)
8204 return (B_TRUE);
8205
8206 for (i = 0; i < sav->sav_npending; i++) {
8207 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
8208 &spareguid) == 0 && spareguid == guid)
8209 return (B_TRUE);
8210 }
8211
8212 return (B_FALSE);
8213}
8214
8215/*
8216 * Check if a pool has an active shared spare device.
8217 * Note: reference count of an active spare is 2, as a spare and as a replace
8218 */
8219static boolean_t
8220spa_has_active_shared_spare(spa_t *spa)
8221{
8222 int i, refcnt;
8223 uint64_t pool;
8224 spa_aux_vdev_t *sav = &spa->spa_spares;
8225
8226 for (i = 0; i < sav->sav_count; i++) {
8227 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8228 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8229 refcnt > 2)
8230 return (B_TRUE);
8231 }
8232
8233 return (B_FALSE);
8234}
8235
8236sysevent_t *
8237spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8238{
8239 sysevent_t *ev = NULL;
8240#ifdef _KERNEL
8241 sysevent_attr_list_t *attr = NULL;
8242 sysevent_value_t value;
8243
8244 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8245 SE_SLEEP);
8246 ASSERT(ev != NULL);
8247
8248 value.value_type = SE_DATA_TYPE_STRING;
8249 value.value.sv_string = spa_name(spa);
8250 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8251 goto done;
8252
8253 value.value_type = SE_DATA_TYPE_UINT64;
8254 value.value.sv_uint64 = spa_guid(spa);
8255 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8256 goto done;
8257
8258 if (vd) {
8259 value.value_type = SE_DATA_TYPE_UINT64;
8260 value.value.sv_uint64 = vd->vdev_guid;
8261 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8262 SE_SLEEP) != 0)
8263 goto done;
8264
8265 if (vd->vdev_path) {
8266 value.value_type = SE_DATA_TYPE_STRING;
8267 value.value.sv_string = vd->vdev_path;
8268 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8269 &value, SE_SLEEP) != 0)
8270 goto done;
8271 }
8272 }
8273
8274 if (hist_nvl != NULL) {
8275 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8276 }
8277
8278 if (sysevent_attach_attributes(ev, attr) != 0)
8279 goto done;
8280 attr = NULL;
8281
8282done:
8283 if (attr)
8284 sysevent_free_attr(attr);
8285
8286#endif
8287 return (ev);
8288}
8289
8290void
8291spa_event_post(sysevent_t *ev)
8292{
8293#ifdef _KERNEL
8294 sysevent_id_t eid;
8295
8296 (void) log_sysevent(ev, SE_SLEEP, &eid);
8297 sysevent_free(ev);
8298#endif
8299}
8300
8301void
8302spa_event_discard(sysevent_t *ev)
8303{
8304#ifdef _KERNEL
8305 sysevent_free(ev);
8306#endif
8307}
8308
8309/*
8310 * Post a sysevent corresponding to the given event. The 'name' must be one of
8311 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8312 * filled in from the spa and (optionally) the vdev and history nvl. This
8313 * doesn't do anything in the userland libzpool, as we don't want consumers to
8314 * misinterpret ztest or zdb as real changes.
8315 */
8316void
8317spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8318{
8319 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
8320}
|