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) 2012 by Delphix. All rights reserved.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/spa_boot.h>
61#include <sys/zfs_ioctl.h>
62#include <sys/dsl_scan.h>
63#include <sys/zfeature.h>
64#include <sys/zvol.h>
65
66#ifdef _KERNEL
67#include <sys/callb.h>
68#include <sys/cpupart.h>
69#include <sys/zone.h>
70#endif /* _KERNEL */
71
72#include "zfs_prop.h"
73#include "zfs_comutil.h"
74
75/* Check hostid on import? */
76static int check_hostid = 1;
77
78SYSCTL_DECL(_vfs_zfs);
79TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
80SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
81 "Check hostid on import?");
82
83typedef enum zti_modes {
84 zti_mode_fixed, /* value is # of threads (min 1) */
85 zti_mode_online_percent, /* value is % of online CPUs */
86 zti_mode_batch, /* cpu-intensive; value is ignored */
87 zti_mode_null, /* don't create a taskq */
88 zti_nmodes
89} zti_modes_t;
90
91#define ZTI_FIX(n) { zti_mode_fixed, (n) }
92#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
93#define ZTI_BATCH { zti_mode_batch, 0 }
94#define ZTI_NULL { zti_mode_null, 0 }
95
96#define ZTI_ONE ZTI_FIX(1)
97
98typedef struct zio_taskq_info {
99 enum zti_modes zti_mode;
100 uint_t zti_value;
101} zio_taskq_info_t;
102
103static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
104 "issue", "issue_high", "intr", "intr_high"
105};
106
107/*
108 * Define the taskq threads for the following I/O types:
109 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
110 */
111const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
112 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
115 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
116 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
119};
120
121static dsl_syncfunc_t spa_sync_version;
122static dsl_syncfunc_t spa_sync_props;
123static boolean_t spa_has_active_shared_spare(spa_t *spa);
124static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
125 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
126 char **ereport);
127static void spa_vdev_resilver_done(spa_t *spa);
128
129uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
130#ifdef PSRSET_BIND
131id_t zio_taskq_psrset_bind = PS_NONE;
132#endif
133#ifdef SYSDC
134boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
135#endif
136uint_t zio_taskq_basedc = 80; /* base duty cycle */
137
138boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
139
140/*
141 * This (illegal) pool name is used when temporarily importing a spa_t in order
142 * to get the vdev stats associated with the imported devices.
143 */
144#define TRYIMPORT_NAME "$import"
145
146/*
147 * ==========================================================================
148 * SPA properties routines
149 * ==========================================================================
150 */
151
152/*
153 * Add a (source=src, propname=propval) list to an nvlist.
154 */
155static void
156spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
157 uint64_t intval, zprop_source_t src)
158{
159 const char *propname = zpool_prop_to_name(prop);
160 nvlist_t *propval;
161
162 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
163 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
164
165 if (strval != NULL)
166 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
167 else
168 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
169
170 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
171 nvlist_free(propval);
172}
173
174/*
175 * Get property values from the spa configuration.
176 */
177static void
178spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
179{
180 vdev_t *rvd = spa->spa_root_vdev;
181 dsl_pool_t *pool = spa->spa_dsl_pool;
182 uint64_t size;
183 uint64_t alloc;
184 uint64_t space;
185 uint64_t cap, version;
186 zprop_source_t src = ZPROP_SRC_NONE;
187 spa_config_dirent_t *dp;
188
189 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
190
191 if (rvd != NULL) {
192 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
193 size = metaslab_class_get_space(spa_normal_class(spa));
194 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
195 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
196 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
197 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
198 size - alloc, src);
199
200 space = 0;
201 for (int c = 0; c < rvd->vdev_children; c++) {
202 vdev_t *tvd = rvd->vdev_child[c];
203 space += tvd->vdev_max_asize - tvd->vdev_asize;
204 }
205 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
206 src);
207
208 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
209 (spa_mode(spa) == FREAD), src);
210
211 cap = (size == 0) ? 0 : (alloc * 100 / size);
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
213
214 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
215 ddt_get_pool_dedup_ratio(spa), src);
216
217 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
218 rvd->vdev_state, src);
219
220 version = spa_version(spa);
221 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
222 src = ZPROP_SRC_DEFAULT;
223 else
224 src = ZPROP_SRC_LOCAL;
225 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
226 }
227
228 if (pool != NULL) {
229 dsl_dir_t *freedir = pool->dp_free_dir;
230
231 /*
232 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
233 * when opening pools before this version freedir will be NULL.
234 */
235 if (freedir != NULL) {
236 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
237 freedir->dd_phys->dd_used_bytes, src);
238 } else {
239 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
240 NULL, 0, src);
241 }
242 }
243
244 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
245
246 if (spa->spa_comment != NULL) {
247 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
248 0, ZPROP_SRC_LOCAL);
249 }
250
251 if (spa->spa_root != NULL)
252 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
253 0, ZPROP_SRC_LOCAL);
254
255 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
256 if (dp->scd_path == NULL) {
257 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
258 "none", 0, ZPROP_SRC_LOCAL);
259 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
260 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
261 dp->scd_path, 0, ZPROP_SRC_LOCAL);
262 }
263 }
264}
265
266/*
267 * Get zpool property values.
268 */
269int
270spa_prop_get(spa_t *spa, nvlist_t **nvp)
271{
272 objset_t *mos = spa->spa_meta_objset;
273 zap_cursor_t zc;
274 zap_attribute_t za;
275 int err;
276
277 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
278
279 mutex_enter(&spa->spa_props_lock);
280
281 /*
282 * Get properties from the spa config.
283 */
284 spa_prop_get_config(spa, nvp);
285
286 /* If no pool property object, no more prop to get. */
287 if (mos == NULL || spa->spa_pool_props_object == 0) {
288 mutex_exit(&spa->spa_props_lock);
289 return (0);
290 }
291
292 /*
293 * Get properties from the MOS pool property object.
294 */
295 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
296 (err = zap_cursor_retrieve(&zc, &za)) == 0;
297 zap_cursor_advance(&zc)) {
298 uint64_t intval = 0;
299 char *strval = NULL;
300 zprop_source_t src = ZPROP_SRC_DEFAULT;
301 zpool_prop_t prop;
302
303 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
304 continue;
305
306 switch (za.za_integer_length) {
307 case 8:
308 /* integer property */
309 if (za.za_first_integer !=
310 zpool_prop_default_numeric(prop))
311 src = ZPROP_SRC_LOCAL;
312
313 if (prop == ZPOOL_PROP_BOOTFS) {
314 dsl_pool_t *dp;
315 dsl_dataset_t *ds = NULL;
316
317 dp = spa_get_dsl(spa);
318 rw_enter(&dp->dp_config_rwlock, RW_READER);
319 if (err = dsl_dataset_hold_obj(dp,
320 za.za_first_integer, FTAG, &ds)) {
321 rw_exit(&dp->dp_config_rwlock);
322 break;
323 }
324
325 strval = kmem_alloc(
326 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
327 KM_SLEEP);
328 dsl_dataset_name(ds, strval);
329 dsl_dataset_rele(ds, FTAG);
330 rw_exit(&dp->dp_config_rwlock);
331 } else {
332 strval = NULL;
333 intval = za.za_first_integer;
334 }
335
336 spa_prop_add_list(*nvp, prop, strval, intval, src);
337
338 if (strval != NULL)
339 kmem_free(strval,
340 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
341
342 break;
343
344 case 1:
345 /* string property */
346 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
347 err = zap_lookup(mos, spa->spa_pool_props_object,
348 za.za_name, 1, za.za_num_integers, strval);
349 if (err) {
350 kmem_free(strval, za.za_num_integers);
351 break;
352 }
353 spa_prop_add_list(*nvp, prop, strval, 0, src);
354 kmem_free(strval, za.za_num_integers);
355 break;
356
357 default:
358 break;
359 }
360 }
361 zap_cursor_fini(&zc);
362 mutex_exit(&spa->spa_props_lock);
363out:
364 if (err && err != ENOENT) {
365 nvlist_free(*nvp);
366 *nvp = NULL;
367 return (err);
368 }
369
370 return (0);
371}
372
373/*
374 * Validate the given pool properties nvlist and modify the list
375 * for the property values to be set.
376 */
377static int
378spa_prop_validate(spa_t *spa, nvlist_t *props)
379{
380 nvpair_t *elem;
381 int error = 0, reset_bootfs = 0;
382 uint64_t objnum;
383 boolean_t has_feature = B_FALSE;
384
385 elem = NULL;
386 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
387 uint64_t intval;
388 char *strval, *slash, *check, *fname;
389 const char *propname = nvpair_name(elem);
390 zpool_prop_t prop = zpool_name_to_prop(propname);
391
392 switch (prop) {
393 case ZPROP_INVAL:
394 if (!zpool_prop_feature(propname)) {
395 error = EINVAL;
396 break;
397 }
398
399 /*
400 * Sanitize the input.
401 */
402 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
403 error = EINVAL;
404 break;
405 }
406
407 if (nvpair_value_uint64(elem, &intval) != 0) {
408 error = EINVAL;
409 break;
410 }
411
412 if (intval != 0) {
413 error = EINVAL;
414 break;
415 }
416
417 fname = strchr(propname, '@') + 1;
418 if (zfeature_lookup_name(fname, NULL) != 0) {
419 error = EINVAL;
420 break;
421 }
422
423 has_feature = B_TRUE;
424 break;
425
426 case ZPOOL_PROP_VERSION:
427 error = nvpair_value_uint64(elem, &intval);
428 if (!error &&
429 (intval < spa_version(spa) ||
430 intval > SPA_VERSION_BEFORE_FEATURES ||
431 has_feature))
432 error = EINVAL;
433 break;
434
435 case ZPOOL_PROP_DELEGATION:
436 case ZPOOL_PROP_AUTOREPLACE:
437 case ZPOOL_PROP_LISTSNAPS:
438 case ZPOOL_PROP_AUTOEXPAND:
439 error = nvpair_value_uint64(elem, &intval);
440 if (!error && intval > 1)
441 error = EINVAL;
442 break;
443
444 case ZPOOL_PROP_BOOTFS:
445 /*
446 * If the pool version is less than SPA_VERSION_BOOTFS,
447 * or the pool is still being created (version == 0),
448 * the bootfs property cannot be set.
449 */
450 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
451 error = ENOTSUP;
452 break;
453 }
454
455 /*
456 * Make sure the vdev config is bootable
457 */
458 if (!vdev_is_bootable(spa->spa_root_vdev)) {
459 error = ENOTSUP;
460 break;
461 }
462
463 reset_bootfs = 1;
464
465 error = nvpair_value_string(elem, &strval);
466
467 if (!error) {
468 objset_t *os;
469 uint64_t compress;
470
471 if (strval == NULL || strval[0] == '\0') {
472 objnum = zpool_prop_default_numeric(
473 ZPOOL_PROP_BOOTFS);
474 break;
475 }
476
477 if (error = dmu_objset_hold(strval, FTAG, &os))
478 break;
479
480 /* Must be ZPL and not gzip compressed. */
481
482 if (dmu_objset_type(os) != DMU_OST_ZFS) {
483 error = ENOTSUP;
484 } else if ((error = dsl_prop_get_integer(strval,
485 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
486 &compress, NULL)) == 0 &&
487 !BOOTFS_COMPRESS_VALID(compress)) {
488 error = ENOTSUP;
489 } else {
490 objnum = dmu_objset_id(os);
491 }
492 dmu_objset_rele(os, FTAG);
493 }
494 break;
495
496 case ZPOOL_PROP_FAILUREMODE:
497 error = nvpair_value_uint64(elem, &intval);
498 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
499 intval > ZIO_FAILURE_MODE_PANIC))
500 error = EINVAL;
501
502 /*
503 * This is a special case which only occurs when
504 * the pool has completely failed. This allows
505 * the user to change the in-core failmode property
506 * without syncing it out to disk (I/Os might
507 * currently be blocked). We do this by returning
508 * EIO to the caller (spa_prop_set) to trick it
509 * into thinking we encountered a property validation
510 * error.
511 */
512 if (!error && spa_suspended(spa)) {
513 spa->spa_failmode = intval;
514 error = EIO;
515 }
516 break;
517
518 case ZPOOL_PROP_CACHEFILE:
519 if ((error = nvpair_value_string(elem, &strval)) != 0)
520 break;
521
522 if (strval[0] == '\0')
523 break;
524
525 if (strcmp(strval, "none") == 0)
526 break;
527
528 if (strval[0] != '/') {
529 error = EINVAL;
530 break;
531 }
532
533 slash = strrchr(strval, '/');
534 ASSERT(slash != NULL);
535
536 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
537 strcmp(slash, "/..") == 0)
538 error = EINVAL;
539 break;
540
541 case ZPOOL_PROP_COMMENT:
542 if ((error = nvpair_value_string(elem, &strval)) != 0)
543 break;
544 for (check = strval; *check != '\0'; check++) {
545 /*
546 * The kernel doesn't have an easy isprint()
547 * check. For this kernel check, we merely
548 * check ASCII apart from DEL. Fix this if
549 * there is an easy-to-use kernel isprint().
550 */
551 if (*check >= 0x7f) {
552 error = EINVAL;
553 break;
554 }
555 check++;
556 }
557 if (strlen(strval) > ZPROP_MAX_COMMENT)
558 error = E2BIG;
559 break;
560
561 case ZPOOL_PROP_DEDUPDITTO:
562 if (spa_version(spa) < SPA_VERSION_DEDUP)
563 error = ENOTSUP;
564 else
565 error = nvpair_value_uint64(elem, &intval);
566 if (error == 0 &&
567 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
568 error = EINVAL;
569 break;
570 }
571
572 if (error)
573 break;
574 }
575
576 if (!error && reset_bootfs) {
577 error = nvlist_remove(props,
578 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
579
580 if (!error) {
581 error = nvlist_add_uint64(props,
582 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
583 }
584 }
585
586 return (error);
587}
588
589void
590spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
591{
592 char *cachefile;
593 spa_config_dirent_t *dp;
594
595 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
596 &cachefile) != 0)
597 return;
598
599 dp = kmem_alloc(sizeof (spa_config_dirent_t),
600 KM_SLEEP);
601
602 if (cachefile[0] == '\0')
603 dp->scd_path = spa_strdup(spa_config_path);
604 else if (strcmp(cachefile, "none") == 0)
605 dp->scd_path = NULL;
606 else
607 dp->scd_path = spa_strdup(cachefile);
608
609 list_insert_head(&spa->spa_config_list, dp);
610 if (need_sync)
611 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
612}
613
614int
615spa_prop_set(spa_t *spa, nvlist_t *nvp)
616{
617 int error;
618 nvpair_t *elem = NULL;
619 boolean_t need_sync = B_FALSE;
620
621 if ((error = spa_prop_validate(spa, nvp)) != 0)
622 return (error);
623
624 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
625 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
626
627 if (prop == ZPOOL_PROP_CACHEFILE ||
628 prop == ZPOOL_PROP_ALTROOT ||
629 prop == ZPOOL_PROP_READONLY)
630 continue;
631
632 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
633 uint64_t ver;
634
635 if (prop == ZPOOL_PROP_VERSION) {
636 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
637 } else {
638 ASSERT(zpool_prop_feature(nvpair_name(elem)));
639 ver = SPA_VERSION_FEATURES;
640 need_sync = B_TRUE;
641 }
642
643 /* Save time if the version is already set. */
644 if (ver == spa_version(spa))
645 continue;
646
647 /*
648 * In addition to the pool directory object, we might
649 * create the pool properties object, the features for
650 * read object, the features for write object, or the
651 * feature descriptions object.
652 */
653 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
654 spa_sync_version, spa, &ver, 6);
655 if (error)
656 return (error);
657 continue;
658 }
659
660 need_sync = B_TRUE;
661 break;
662 }
663
664 if (need_sync) {
665 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
666 spa, nvp, 6));
667 }
668
669 return (0);
670}
671
672/*
673 * If the bootfs property value is dsobj, clear it.
674 */
675void
676spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
677{
678 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
679 VERIFY(zap_remove(spa->spa_meta_objset,
680 spa->spa_pool_props_object,
681 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
682 spa->spa_bootfs = 0;
683 }
684}
685
686/*
687 * Change the GUID for the pool. This is done so that we can later
688 * re-import a pool built from a clone of our own vdevs. We will modify
689 * the root vdev's guid, our own pool guid, and then mark all of our
690 * vdevs dirty. Note that we must make sure that all our vdevs are
691 * online when we do this, or else any vdevs that weren't present
692 * would be orphaned from our pool. We are also going to issue a
693 * sysevent to update any watchers.
694 */
695int
696spa_change_guid(spa_t *spa)
697{
698 uint64_t oldguid, newguid;
699 uint64_t txg;
700
701 if (!(spa_mode_global & FWRITE))
702 return (EROFS);
703
704 txg = spa_vdev_enter(spa);
705
706 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
707 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
708
709 oldguid = spa_guid(spa);
710 newguid = spa_generate_guid(NULL);
711 ASSERT3U(oldguid, !=, newguid);
712
713 spa->spa_root_vdev->vdev_guid = newguid;
714 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
715
716 vdev_config_dirty(spa->spa_root_vdev);
717
718 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
719
720 return (spa_vdev_exit(spa, NULL, txg, 0));
721}
722
723/*
724 * ==========================================================================
725 * SPA state manipulation (open/create/destroy/import/export)
726 * ==========================================================================
727 */
728
729static int
730spa_error_entry_compare(const void *a, const void *b)
731{
732 spa_error_entry_t *sa = (spa_error_entry_t *)a;
733 spa_error_entry_t *sb = (spa_error_entry_t *)b;
734 int ret;
735
736 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
737 sizeof (zbookmark_t));
738
739 if (ret < 0)
740 return (-1);
741 else if (ret > 0)
742 return (1);
743 else
744 return (0);
745}
746
747/*
748 * Utility function which retrieves copies of the current logs and
749 * re-initializes them in the process.
750 */
751void
752spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
753{
754 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
755
756 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
757 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
758
759 avl_create(&spa->spa_errlist_scrub,
760 spa_error_entry_compare, sizeof (spa_error_entry_t),
761 offsetof(spa_error_entry_t, se_avl));
762 avl_create(&spa->spa_errlist_last,
763 spa_error_entry_compare, sizeof (spa_error_entry_t),
764 offsetof(spa_error_entry_t, se_avl));
765}
766
767static taskq_t *
768spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
769 uint_t value)
770{
771 uint_t flags = TASKQ_PREPOPULATE;
772 boolean_t batch = B_FALSE;
773
774 switch (mode) {
775 case zti_mode_null:
776 return (NULL); /* no taskq needed */
777
778 case zti_mode_fixed:
779 ASSERT3U(value, >=, 1);
780 value = MAX(value, 1);
781 break;
782
783 case zti_mode_batch:
784 batch = B_TRUE;
785 flags |= TASKQ_THREADS_CPU_PCT;
786 value = zio_taskq_batch_pct;
787 break;
788
789 case zti_mode_online_percent:
790 flags |= TASKQ_THREADS_CPU_PCT;
791 break;
792
793 default:
794 panic("unrecognized mode for %s taskq (%u:%u) in "
795 "spa_activate()",
796 name, mode, value);
797 break;
798 }
799
800#ifdef SYSDC
801 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
802 if (batch)
803 flags |= TASKQ_DC_BATCH;
804
805 return (taskq_create_sysdc(name, value, 50, INT_MAX,
806 spa->spa_proc, zio_taskq_basedc, flags));
807 }
808#endif
809 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
810 spa->spa_proc, flags));
811}
812
813static void
814spa_create_zio_taskqs(spa_t *spa)
815{
816 for (int t = 0; t < ZIO_TYPES; t++) {
817 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
818 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
819 enum zti_modes mode = ztip->zti_mode;
820 uint_t value = ztip->zti_value;
821 char name[32];
822
823 (void) snprintf(name, sizeof (name),
824 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
825
826 spa->spa_zio_taskq[t][q] =
827 spa_taskq_create(spa, name, mode, value);
828 }
829 }
830}
831
832#ifdef _KERNEL
833#ifdef SPA_PROCESS
834static void
835spa_thread(void *arg)
836{
837 callb_cpr_t cprinfo;
838
839 spa_t *spa = arg;
840 user_t *pu = PTOU(curproc);
841
842 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
843 spa->spa_name);
844
845 ASSERT(curproc != &p0);
846 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
847 "zpool-%s", spa->spa_name);
848 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
849
850#ifdef PSRSET_BIND
851 /* bind this thread to the requested psrset */
852 if (zio_taskq_psrset_bind != PS_NONE) {
853 pool_lock();
854 mutex_enter(&cpu_lock);
855 mutex_enter(&pidlock);
856 mutex_enter(&curproc->p_lock);
857
858 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
859 0, NULL, NULL) == 0) {
860 curthread->t_bind_pset = zio_taskq_psrset_bind;
861 } else {
862 cmn_err(CE_WARN,
863 "Couldn't bind process for zfs pool \"%s\" to "
864 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
865 }
866
867 mutex_exit(&curproc->p_lock);
868 mutex_exit(&pidlock);
869 mutex_exit(&cpu_lock);
870 pool_unlock();
871 }
872#endif
873
874#ifdef SYSDC
875 if (zio_taskq_sysdc) {
876 sysdc_thread_enter(curthread, 100, 0);
877 }
878#endif
879
880 spa->spa_proc = curproc;
881 spa->spa_did = curthread->t_did;
882
883 spa_create_zio_taskqs(spa);
884
885 mutex_enter(&spa->spa_proc_lock);
886 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
887
888 spa->spa_proc_state = SPA_PROC_ACTIVE;
889 cv_broadcast(&spa->spa_proc_cv);
890
891 CALLB_CPR_SAFE_BEGIN(&cprinfo);
892 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
893 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
894 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
895
896 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
897 spa->spa_proc_state = SPA_PROC_GONE;
898 spa->spa_proc = &p0;
899 cv_broadcast(&spa->spa_proc_cv);
900 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
901
902 mutex_enter(&curproc->p_lock);
903 lwp_exit();
904}
905#endif /* SPA_PROCESS */
906#endif
907
908/*
909 * Activate an uninitialized pool.
910 */
911static void
912spa_activate(spa_t *spa, int mode)
913{
914 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
915
916 spa->spa_state = POOL_STATE_ACTIVE;
917 spa->spa_mode = mode;
918
919 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
920 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
921
922 /* Try to create a covering process */
923 mutex_enter(&spa->spa_proc_lock);
924 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
925 ASSERT(spa->spa_proc == &p0);
926 spa->spa_did = 0;
927
928#ifdef SPA_PROCESS
929 /* Only create a process if we're going to be around a while. */
930 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
931 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
932 NULL, 0) == 0) {
933 spa->spa_proc_state = SPA_PROC_CREATED;
934 while (spa->spa_proc_state == SPA_PROC_CREATED) {
935 cv_wait(&spa->spa_proc_cv,
936 &spa->spa_proc_lock);
937 }
938 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
939 ASSERT(spa->spa_proc != &p0);
940 ASSERT(spa->spa_did != 0);
941 } else {
942#ifdef _KERNEL
943 cmn_err(CE_WARN,
944 "Couldn't create process for zfs pool \"%s\"\n",
945 spa->spa_name);
946#endif
947 }
948 }
949#endif /* SPA_PROCESS */
950 mutex_exit(&spa->spa_proc_lock);
951
952 /* If we didn't create a process, we need to create our taskqs. */
953 ASSERT(spa->spa_proc == &p0);
954 if (spa->spa_proc == &p0) {
955 spa_create_zio_taskqs(spa);
956 }
957
958 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
959 offsetof(vdev_t, vdev_config_dirty_node));
960 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
961 offsetof(vdev_t, vdev_state_dirty_node));
962
963 txg_list_create(&spa->spa_vdev_txg_list,
964 offsetof(struct vdev, vdev_txg_node));
965
966 avl_create(&spa->spa_errlist_scrub,
967 spa_error_entry_compare, sizeof (spa_error_entry_t),
968 offsetof(spa_error_entry_t, se_avl));
969 avl_create(&spa->spa_errlist_last,
970 spa_error_entry_compare, sizeof (spa_error_entry_t),
971 offsetof(spa_error_entry_t, se_avl));
972}
973
974/*
975 * Opposite of spa_activate().
976 */
977static void
978spa_deactivate(spa_t *spa)
979{
980 ASSERT(spa->spa_sync_on == B_FALSE);
981 ASSERT(spa->spa_dsl_pool == NULL);
982 ASSERT(spa->spa_root_vdev == NULL);
983 ASSERT(spa->spa_async_zio_root == NULL);
984 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
985
986 txg_list_destroy(&spa->spa_vdev_txg_list);
987
988 list_destroy(&spa->spa_config_dirty_list);
989 list_destroy(&spa->spa_state_dirty_list);
990
991 for (int t = 0; t < ZIO_TYPES; t++) {
992 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
993 if (spa->spa_zio_taskq[t][q] != NULL)
994 taskq_destroy(spa->spa_zio_taskq[t][q]);
995 spa->spa_zio_taskq[t][q] = NULL;
996 }
997 }
998
999 metaslab_class_destroy(spa->spa_normal_class);
1000 spa->spa_normal_class = NULL;
1001
1002 metaslab_class_destroy(spa->spa_log_class);
1003 spa->spa_log_class = NULL;
1004
1005 /*
1006 * If this was part of an import or the open otherwise failed, we may
1007 * still have errors left in the queues. Empty them just in case.
1008 */
1009 spa_errlog_drain(spa);
1010
1011 avl_destroy(&spa->spa_errlist_scrub);
1012 avl_destroy(&spa->spa_errlist_last);
1013
1014 spa->spa_state = POOL_STATE_UNINITIALIZED;
1015
1016 mutex_enter(&spa->spa_proc_lock);
1017 if (spa->spa_proc_state != SPA_PROC_NONE) {
1018 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1019 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1020 cv_broadcast(&spa->spa_proc_cv);
1021 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1022 ASSERT(spa->spa_proc != &p0);
1023 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1024 }
1025 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1026 spa->spa_proc_state = SPA_PROC_NONE;
1027 }
1028 ASSERT(spa->spa_proc == &p0);
1029 mutex_exit(&spa->spa_proc_lock);
1030
1031#ifdef SPA_PROCESS
1032 /*
1033 * We want to make sure spa_thread() has actually exited the ZFS
1034 * module, so that the module can't be unloaded out from underneath
1035 * it.
1036 */
1037 if (spa->spa_did != 0) {
1038 thread_join(spa->spa_did);
1039 spa->spa_did = 0;
1040 }
1041#endif /* SPA_PROCESS */
1042}
1043
1044/*
1045 * Verify a pool configuration, and construct the vdev tree appropriately. This
1046 * will create all the necessary vdevs in the appropriate layout, with each vdev
1047 * in the CLOSED state. This will prep the pool before open/creation/import.
1048 * All vdev validation is done by the vdev_alloc() routine.
1049 */
1050static int
1051spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1052 uint_t id, int atype)
1053{
1054 nvlist_t **child;
1055 uint_t children;
1056 int error;
1057
1058 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1059 return (error);
1060
1061 if ((*vdp)->vdev_ops->vdev_op_leaf)
1062 return (0);
1063
1064 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1065 &child, &children);
1066
1067 if (error == ENOENT)
1068 return (0);
1069
1070 if (error) {
1071 vdev_free(*vdp);
1072 *vdp = NULL;
1073 return (EINVAL);
1074 }
1075
1076 for (int c = 0; c < children; c++) {
1077 vdev_t *vd;
1078 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1079 atype)) != 0) {
1080 vdev_free(*vdp);
1081 *vdp = NULL;
1082 return (error);
1083 }
1084 }
1085
1086 ASSERT(*vdp != NULL);
1087
1088 return (0);
1089}
1090
1091/*
1092 * Opposite of spa_load().
1093 */
1094static void
1095spa_unload(spa_t *spa)
1096{
1097 int i;
1098
1099 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1100
1101 /*
1102 * Stop async tasks.
1103 */
1104 spa_async_suspend(spa);
1105
1106 /*
1107 * Stop syncing.
1108 */
1109 if (spa->spa_sync_on) {
1110 txg_sync_stop(spa->spa_dsl_pool);
1111 spa->spa_sync_on = B_FALSE;
1112 }
1113
1114 /*
1115 * Wait for any outstanding async I/O to complete.
1116 */
1117 if (spa->spa_async_zio_root != NULL) {
1118 (void) zio_wait(spa->spa_async_zio_root);
1119 spa->spa_async_zio_root = NULL;
1120 }
1121
1122 bpobj_close(&spa->spa_deferred_bpobj);
1123
1124 /*
1125 * Close the dsl pool.
1126 */
1127 if (spa->spa_dsl_pool) {
1128 dsl_pool_close(spa->spa_dsl_pool);
1129 spa->spa_dsl_pool = NULL;
1130 spa->spa_meta_objset = NULL;
1131 }
1132
1133 ddt_unload(spa);
1134
1135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1136
1137 /*
1138 * Drop and purge level 2 cache
1139 */
1140 spa_l2cache_drop(spa);
1141
1142 /*
1143 * Close all vdevs.
1144 */
1145 if (spa->spa_root_vdev)
1146 vdev_free(spa->spa_root_vdev);
1147 ASSERT(spa->spa_root_vdev == NULL);
1148
1149 for (i = 0; i < spa->spa_spares.sav_count; i++)
1150 vdev_free(spa->spa_spares.sav_vdevs[i]);
1151 if (spa->spa_spares.sav_vdevs) {
1152 kmem_free(spa->spa_spares.sav_vdevs,
1153 spa->spa_spares.sav_count * sizeof (void *));
1154 spa->spa_spares.sav_vdevs = NULL;
1155 }
1156 if (spa->spa_spares.sav_config) {
1157 nvlist_free(spa->spa_spares.sav_config);
1158 spa->spa_spares.sav_config = NULL;
1159 }
1160 spa->spa_spares.sav_count = 0;
1161
1162 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1163 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1164 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1165 }
1166 if (spa->spa_l2cache.sav_vdevs) {
1167 kmem_free(spa->spa_l2cache.sav_vdevs,
1168 spa->spa_l2cache.sav_count * sizeof (void *));
1169 spa->spa_l2cache.sav_vdevs = NULL;
1170 }
1171 if (spa->spa_l2cache.sav_config) {
1172 nvlist_free(spa->spa_l2cache.sav_config);
1173 spa->spa_l2cache.sav_config = NULL;
1174 }
1175 spa->spa_l2cache.sav_count = 0;
1176
1177 spa->spa_async_suspended = 0;
1178
1179 if (spa->spa_comment != NULL) {
1180 spa_strfree(spa->spa_comment);
1181 spa->spa_comment = NULL;
1182 }
1183
1184 spa_config_exit(spa, SCL_ALL, FTAG);
1185}
1186
1187/*
1188 * Load (or re-load) the current list of vdevs describing the active spares for
1189 * this pool. When this is called, we have some form of basic information in
1190 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1191 * then re-generate a more complete list including status information.
1192 */
1193static void
1194spa_load_spares(spa_t *spa)
1195{
1196 nvlist_t **spares;
1197 uint_t nspares;
1198 int i;
1199 vdev_t *vd, *tvd;
1200
1201 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1202
1203 /*
1204 * First, close and free any existing spare vdevs.
1205 */
1206 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1207 vd = spa->spa_spares.sav_vdevs[i];
1208
1209 /* Undo the call to spa_activate() below */
1210 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1211 B_FALSE)) != NULL && tvd->vdev_isspare)
1212 spa_spare_remove(tvd);
1213 vdev_close(vd);
1214 vdev_free(vd);
1215 }
1216
1217 if (spa->spa_spares.sav_vdevs)
1218 kmem_free(spa->spa_spares.sav_vdevs,
1219 spa->spa_spares.sav_count * sizeof (void *));
1220
1221 if (spa->spa_spares.sav_config == NULL)
1222 nspares = 0;
1223 else
1224 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1225 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1226
1227 spa->spa_spares.sav_count = (int)nspares;
1228 spa->spa_spares.sav_vdevs = NULL;
1229
1230 if (nspares == 0)
1231 return;
1232
1233 /*
1234 * Construct the array of vdevs, opening them to get status in the
1235 * process. For each spare, there is potentially two different vdev_t
1236 * structures associated with it: one in the list of spares (used only
1237 * for basic validation purposes) and one in the active vdev
1238 * configuration (if it's spared in). During this phase we open and
1239 * validate each vdev on the spare list. If the vdev also exists in the
1240 * active configuration, then we also mark this vdev as an active spare.
1241 */
1242 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1243 KM_SLEEP);
1244 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1245 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1246 VDEV_ALLOC_SPARE) == 0);
1247 ASSERT(vd != NULL);
1248
1249 spa->spa_spares.sav_vdevs[i] = vd;
1250
1251 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1252 B_FALSE)) != NULL) {
1253 if (!tvd->vdev_isspare)
1254 spa_spare_add(tvd);
1255
1256 /*
1257 * We only mark the spare active if we were successfully
1258 * able to load the vdev. Otherwise, importing a pool
1259 * with a bad active spare would result in strange
1260 * behavior, because multiple pool would think the spare
1261 * is actively in use.
1262 *
1263 * There is a vulnerability here to an equally bizarre
1264 * circumstance, where a dead active spare is later
1265 * brought back to life (onlined or otherwise). Given
1266 * the rarity of this scenario, and the extra complexity
1267 * it adds, we ignore the possibility.
1268 */
1269 if (!vdev_is_dead(tvd))
1270 spa_spare_activate(tvd);
1271 }
1272
1273 vd->vdev_top = vd;
1274 vd->vdev_aux = &spa->spa_spares;
1275
1276 if (vdev_open(vd) != 0)
1277 continue;
1278
1279 if (vdev_validate_aux(vd) == 0)
1280 spa_spare_add(vd);
1281 }
1282
1283 /*
1284 * Recompute the stashed list of spares, with status information
1285 * this time.
1286 */
1287 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1288 DATA_TYPE_NVLIST_ARRAY) == 0);
1289
1290 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1291 KM_SLEEP);
1292 for (i = 0; i < spa->spa_spares.sav_count; i++)
1293 spares[i] = vdev_config_generate(spa,
1294 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1295 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1296 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1297 for (i = 0; i < spa->spa_spares.sav_count; i++)
1298 nvlist_free(spares[i]);
1299 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1300}
1301
1302/*
1303 * Load (or re-load) the current list of vdevs describing the active l2cache for
1304 * this pool. When this is called, we have some form of basic information in
1305 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1306 * then re-generate a more complete list including status information.
1307 * Devices which are already active have their details maintained, and are
1308 * not re-opened.
1309 */
1310static void
1311spa_load_l2cache(spa_t *spa)
1312{
1313 nvlist_t **l2cache;
1314 uint_t nl2cache;
1315 int i, j, oldnvdevs;
1316 uint64_t guid;
1317 vdev_t *vd, **oldvdevs, **newvdevs;
1318 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1319
1320 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1321
1322 if (sav->sav_config != NULL) {
1323 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1324 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1325 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1326 } else {
1327 nl2cache = 0;
1328 }
1329
1330 oldvdevs = sav->sav_vdevs;
1331 oldnvdevs = sav->sav_count;
1332 sav->sav_vdevs = NULL;
1333 sav->sav_count = 0;
1334
1335 /*
1336 * Process new nvlist of vdevs.
1337 */
1338 for (i = 0; i < nl2cache; i++) {
1339 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1340 &guid) == 0);
1341
1342 newvdevs[i] = NULL;
1343 for (j = 0; j < oldnvdevs; j++) {
1344 vd = oldvdevs[j];
1345 if (vd != NULL && guid == vd->vdev_guid) {
1346 /*
1347 * Retain previous vdev for add/remove ops.
1348 */
1349 newvdevs[i] = vd;
1350 oldvdevs[j] = NULL;
1351 break;
1352 }
1353 }
1354
1355 if (newvdevs[i] == NULL) {
1356 /*
1357 * Create new vdev
1358 */
1359 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1360 VDEV_ALLOC_L2CACHE) == 0);
1361 ASSERT(vd != NULL);
1362 newvdevs[i] = vd;
1363
1364 /*
1365 * Commit this vdev as an l2cache device,
1366 * even if it fails to open.
1367 */
1368 spa_l2cache_add(vd);
1369
1370 vd->vdev_top = vd;
1371 vd->vdev_aux = sav;
1372
1373 spa_l2cache_activate(vd);
1374
1375 if (vdev_open(vd) != 0)
1376 continue;
1377
1378 (void) vdev_validate_aux(vd);
1379
1380 if (!vdev_is_dead(vd))
1381 l2arc_add_vdev(spa, vd);
1382 }
1383 }
1384
1385 /*
1386 * Purge vdevs that were dropped
1387 */
1388 for (i = 0; i < oldnvdevs; i++) {
1389 uint64_t pool;
1390
1391 vd = oldvdevs[i];
1392 if (vd != NULL) {
1393 ASSERT(vd->vdev_isl2cache);
1394
1395 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1396 pool != 0ULL && l2arc_vdev_present(vd))
1397 l2arc_remove_vdev(vd);
1398 vdev_clear_stats(vd);
1399 vdev_free(vd);
1400 }
1401 }
1402
1403 if (oldvdevs)
1404 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1405
1406 if (sav->sav_config == NULL)
1407 goto out;
1408
1409 sav->sav_vdevs = newvdevs;
1410 sav->sav_count = (int)nl2cache;
1411
1412 /*
1413 * Recompute the stashed list of l2cache devices, with status
1414 * information this time.
1415 */
1416 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1417 DATA_TYPE_NVLIST_ARRAY) == 0);
1418
1419 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1420 for (i = 0; i < sav->sav_count; i++)
1421 l2cache[i] = vdev_config_generate(spa,
1422 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1423 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1424 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1425out:
1426 for (i = 0; i < sav->sav_count; i++)
1427 nvlist_free(l2cache[i]);
1428 if (sav->sav_count)
1429 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1430}
1431
1432static int
1433load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1434{
1435 dmu_buf_t *db;
1436 char *packed = NULL;
1437 size_t nvsize = 0;
1438 int error;
1439 *value = NULL;
1440
1441 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1442 nvsize = *(uint64_t *)db->db_data;
1443 dmu_buf_rele(db, FTAG);
1444
1445 packed = kmem_alloc(nvsize, KM_SLEEP);
1446 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1447 DMU_READ_PREFETCH);
1448 if (error == 0)
1449 error = nvlist_unpack(packed, nvsize, value, 0);
1450 kmem_free(packed, nvsize);
1451
1452 return (error);
1453}
1454
1455/*
1456 * Checks to see if the given vdev could not be opened, in which case we post a
1457 * sysevent to notify the autoreplace code that the device has been removed.
1458 */
1459static void
1460spa_check_removed(vdev_t *vd)
1461{
1462 for (int c = 0; c < vd->vdev_children; c++)
1463 spa_check_removed(vd->vdev_child[c]);
1464
1465 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1466 zfs_post_autoreplace(vd->vdev_spa, vd);
1467 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1468 }
1469}
1470
1471/*
1472 * Validate the current config against the MOS config
1473 */
1474static boolean_t
1475spa_config_valid(spa_t *spa, nvlist_t *config)
1476{
1477 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1478 nvlist_t *nv;
1479
1480 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1481
1482 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1483 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1484
1485 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1486
1487 /*
1488 * If we're doing a normal import, then build up any additional
1489 * diagnostic information about missing devices in this config.
1490 * We'll pass this up to the user for further processing.
1491 */
1492 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1493 nvlist_t **child, *nv;
1494 uint64_t idx = 0;
1495
1496 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1497 KM_SLEEP);
1498 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1499
1500 for (int c = 0; c < rvd->vdev_children; c++) {
1501 vdev_t *tvd = rvd->vdev_child[c];
1502 vdev_t *mtvd = mrvd->vdev_child[c];
1503
1504 if (tvd->vdev_ops == &vdev_missing_ops &&
1505 mtvd->vdev_ops != &vdev_missing_ops &&
1506 mtvd->vdev_islog)
1507 child[idx++] = vdev_config_generate(spa, mtvd,
1508 B_FALSE, 0);
1509 }
1510
1511 if (idx) {
1512 VERIFY(nvlist_add_nvlist_array(nv,
1513 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1514 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1515 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1516
1517 for (int i = 0; i < idx; i++)
1518 nvlist_free(child[i]);
1519 }
1520 nvlist_free(nv);
1521 kmem_free(child, rvd->vdev_children * sizeof (char **));
1522 }
1523
1524 /*
1525 * Compare the root vdev tree with the information we have
1526 * from the MOS config (mrvd). Check each top-level vdev
1527 * with the corresponding MOS config top-level (mtvd).
1528 */
1529 for (int c = 0; c < rvd->vdev_children; c++) {
1530 vdev_t *tvd = rvd->vdev_child[c];
1531 vdev_t *mtvd = mrvd->vdev_child[c];
1532
1533 /*
1534 * Resolve any "missing" vdevs in the current configuration.
1535 * If we find that the MOS config has more accurate information
1536 * about the top-level vdev then use that vdev instead.
1537 */
1538 if (tvd->vdev_ops == &vdev_missing_ops &&
1539 mtvd->vdev_ops != &vdev_missing_ops) {
1540
1541 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1542 continue;
1543
1544 /*
1545 * Device specific actions.
1546 */
1547 if (mtvd->vdev_islog) {
1548 spa_set_log_state(spa, SPA_LOG_CLEAR);
1549 } else {
1550 /*
1551 * XXX - once we have 'readonly' pool
1552 * support we should be able to handle
1553 * missing data devices by transitioning
1554 * the pool to readonly.
1555 */
1556 continue;
1557 }
1558
1559 /*
1560 * Swap the missing vdev with the data we were
1561 * able to obtain from the MOS config.
1562 */
1563 vdev_remove_child(rvd, tvd);
1564 vdev_remove_child(mrvd, mtvd);
1565
1566 vdev_add_child(rvd, mtvd);
1567 vdev_add_child(mrvd, tvd);
1568
1569 spa_config_exit(spa, SCL_ALL, FTAG);
1570 vdev_load(mtvd);
1571 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1572
1573 vdev_reopen(rvd);
1574 } else if (mtvd->vdev_islog) {
1575 /*
1576 * Load the slog device's state from the MOS config
1577 * since it's possible that the label does not
1578 * contain the most up-to-date information.
1579 */
1580 vdev_load_log_state(tvd, mtvd);
1581 vdev_reopen(tvd);
1582 }
1583 }
1584 vdev_free(mrvd);
1585 spa_config_exit(spa, SCL_ALL, FTAG);
1586
1587 /*
1588 * Ensure we were able to validate the config.
1589 */
1590 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1591}
1592
1593/*
1594 * Check for missing log devices
1595 */
1596static int
1597spa_check_logs(spa_t *spa)
1598{
1599 switch (spa->spa_log_state) {
1600 case SPA_LOG_MISSING:
1601 /* need to recheck in case slog has been restored */
1602 case SPA_LOG_UNKNOWN:
1603 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1604 DS_FIND_CHILDREN)) {
1605 spa_set_log_state(spa, SPA_LOG_MISSING);
1606 return (1);
1607 }
1608 break;
1609 }
1610 return (0);
1611}
1612
1613static boolean_t
1614spa_passivate_log(spa_t *spa)
1615{
1616 vdev_t *rvd = spa->spa_root_vdev;
1617 boolean_t slog_found = B_FALSE;
1618
1619 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1620
1621 if (!spa_has_slogs(spa))
1622 return (B_FALSE);
1623
1624 for (int c = 0; c < rvd->vdev_children; c++) {
1625 vdev_t *tvd = rvd->vdev_child[c];
1626 metaslab_group_t *mg = tvd->vdev_mg;
1627
1628 if (tvd->vdev_islog) {
1629 metaslab_group_passivate(mg);
1630 slog_found = B_TRUE;
1631 }
1632 }
1633
1634 return (slog_found);
1635}
1636
1637static void
1638spa_activate_log(spa_t *spa)
1639{
1640 vdev_t *rvd = spa->spa_root_vdev;
1641
1642 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1643
1644 for (int c = 0; c < rvd->vdev_children; c++) {
1645 vdev_t *tvd = rvd->vdev_child[c];
1646 metaslab_group_t *mg = tvd->vdev_mg;
1647
1648 if (tvd->vdev_islog)
1649 metaslab_group_activate(mg);
1650 }
1651}
1652
1653int
1654spa_offline_log(spa_t *spa)
1655{
1656 int error = 0;
1657
1658 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1659 NULL, DS_FIND_CHILDREN)) == 0) {
1660
1661 /*
1662 * We successfully offlined the log device, sync out the
1663 * current txg so that the "stubby" block can be removed
1664 * by zil_sync().
1665 */
1666 txg_wait_synced(spa->spa_dsl_pool, 0);
1667 }
1668 return (error);
1669}
1670
1671static void
1672spa_aux_check_removed(spa_aux_vdev_t *sav)
1673{
1674 int i;
1675
1676 for (i = 0; i < sav->sav_count; i++)
1677 spa_check_removed(sav->sav_vdevs[i]);
1678}
1679
1680void
1681spa_claim_notify(zio_t *zio)
1682{
1683 spa_t *spa = zio->io_spa;
1684
1685 if (zio->io_error)
1686 return;
1687
1688 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1689 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1690 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1691 mutex_exit(&spa->spa_props_lock);
1692}
1693
1694typedef struct spa_load_error {
1695 uint64_t sle_meta_count;
1696 uint64_t sle_data_count;
1697} spa_load_error_t;
1698
1699static void
1700spa_load_verify_done(zio_t *zio)
1701{
1702 blkptr_t *bp = zio->io_bp;
1703 spa_load_error_t *sle = zio->io_private;
1704 dmu_object_type_t type = BP_GET_TYPE(bp);
1705 int error = zio->io_error;
1706
1707 if (error) {
1708 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1709 type != DMU_OT_INTENT_LOG)
1710 atomic_add_64(&sle->sle_meta_count, 1);
1711 else
1712 atomic_add_64(&sle->sle_data_count, 1);
1713 }
1714 zio_data_buf_free(zio->io_data, zio->io_size);
1715}
1716
1717/*ARGSUSED*/
1718static int
1719spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1720 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1721{
1722 if (bp != NULL) {
1723 zio_t *rio = arg;
1724 size_t size = BP_GET_PSIZE(bp);
1725 void *data = zio_data_buf_alloc(size);
1726
1727 zio_nowait(zio_read(rio, spa, bp, data, size,
1728 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1729 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1730 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1731 }
1732 return (0);
1733}
1734
1735static int
1736spa_load_verify(spa_t *spa)
1737{
1738 zio_t *rio;
1739 spa_load_error_t sle = { 0 };
1740 zpool_rewind_policy_t policy;
1741 boolean_t verify_ok = B_FALSE;
1742 int error;
1743
1744 zpool_get_rewind_policy(spa->spa_config, &policy);
1745
1746 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1747 return (0);
1748
1749 rio = zio_root(spa, NULL, &sle,
1750 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1751
1752 error = traverse_pool(spa, spa->spa_verify_min_txg,
1753 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1754
1755 (void) zio_wait(rio);
1756
1757 spa->spa_load_meta_errors = sle.sle_meta_count;
1758 spa->spa_load_data_errors = sle.sle_data_count;
1759
1760 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1761 sle.sle_data_count <= policy.zrp_maxdata) {
1762 int64_t loss = 0;
1763
1764 verify_ok = B_TRUE;
1765 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1766 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1767
1768 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1769 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1770 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1771 VERIFY(nvlist_add_int64(spa->spa_load_info,
1772 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1773 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1774 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1775 } else {
1776 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1777 }
1778
1779 if (error) {
1780 if (error != ENXIO && error != EIO)
1781 error = EIO;
1782 return (error);
1783 }
1784
1785 return (verify_ok ? 0 : EIO);
1786}
1787
1788/*
1789 * Find a value in the pool props object.
1790 */
1791static void
1792spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1793{
1794 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1795 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1796}
1797
1798/*
1799 * Find a value in the pool directory object.
1800 */
1801static int
1802spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1803{
1804 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1805 name, sizeof (uint64_t), 1, val));
1806}
1807
1808static int
1809spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1810{
1811 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1812 return (err);
1813}
1814
1815/*
1816 * Fix up config after a partly-completed split. This is done with the
1817 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1818 * pool have that entry in their config, but only the splitting one contains
1819 * a list of all the guids of the vdevs that are being split off.
1820 *
1821 * This function determines what to do with that list: either rejoin
1822 * all the disks to the pool, or complete the splitting process. To attempt
1823 * the rejoin, each disk that is offlined is marked online again, and
1824 * we do a reopen() call. If the vdev label for every disk that was
1825 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1826 * then we call vdev_split() on each disk, and complete the split.
1827 *
1828 * Otherwise we leave the config alone, with all the vdevs in place in
1829 * the original pool.
1830 */
1831static void
1832spa_try_repair(spa_t *spa, nvlist_t *config)
1833{
1834 uint_t extracted;
1835 uint64_t *glist;
1836 uint_t i, gcount;
1837 nvlist_t *nvl;
1838 vdev_t **vd;
1839 boolean_t attempt_reopen;
1840
1841 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1842 return;
1843
1844 /* check that the config is complete */
1845 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1846 &glist, &gcount) != 0)
1847 return;
1848
1849 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1850
1851 /* attempt to online all the vdevs & validate */
1852 attempt_reopen = B_TRUE;
1853 for (i = 0; i < gcount; i++) {
1854 if (glist[i] == 0) /* vdev is hole */
1855 continue;
1856
1857 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1858 if (vd[i] == NULL) {
1859 /*
1860 * Don't bother attempting to reopen the disks;
1861 * just do the split.
1862 */
1863 attempt_reopen = B_FALSE;
1864 } else {
1865 /* attempt to re-online it */
1866 vd[i]->vdev_offline = B_FALSE;
1867 }
1868 }
1869
1870 if (attempt_reopen) {
1871 vdev_reopen(spa->spa_root_vdev);
1872
1873 /* check each device to see what state it's in */
1874 for (extracted = 0, i = 0; i < gcount; i++) {
1875 if (vd[i] != NULL &&
1876 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1877 break;
1878 ++extracted;
1879 }
1880 }
1881
1882 /*
1883 * If every disk has been moved to the new pool, or if we never
1884 * even attempted to look at them, then we split them off for
1885 * good.
1886 */
1887 if (!attempt_reopen || gcount == extracted) {
1888 for (i = 0; i < gcount; i++)
1889 if (vd[i] != NULL)
1890 vdev_split(vd[i]);
1891 vdev_reopen(spa->spa_root_vdev);
1892 }
1893
1894 kmem_free(vd, gcount * sizeof (vdev_t *));
1895}
1896
1897static int
1898spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1899 boolean_t mosconfig)
1900{
1901 nvlist_t *config = spa->spa_config;
1902 char *ereport = FM_EREPORT_ZFS_POOL;
1903 char *comment;
1904 int error;
1905 uint64_t pool_guid;
1906 nvlist_t *nvl;
1907
1908 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1909 return (EINVAL);
1910
1911 ASSERT(spa->spa_comment == NULL);
1912 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1913 spa->spa_comment = spa_strdup(comment);
1914
1915 /*
1916 * Versioning wasn't explicitly added to the label until later, so if
1917 * it's not present treat it as the initial version.
1918 */
1919 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1920 &spa->spa_ubsync.ub_version) != 0)
1921 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1922
1923 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1924 &spa->spa_config_txg);
1925
1926 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1927 spa_guid_exists(pool_guid, 0)) {
1928 error = EEXIST;
1929 } else {
1930 spa->spa_config_guid = pool_guid;
1931
1932 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1933 &nvl) == 0) {
1934 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1935 KM_SLEEP) == 0);
1936 }
1937
1938 nvlist_free(spa->spa_load_info);
1939 spa->spa_load_info = fnvlist_alloc();
1940
1941 gethrestime(&spa->spa_loaded_ts);
1942 error = spa_load_impl(spa, pool_guid, config, state, type,
1943 mosconfig, &ereport);
1944 }
1945
1946 spa->spa_minref = refcount_count(&spa->spa_refcount);
1947 if (error) {
1948 if (error != EEXIST) {
1949 spa->spa_loaded_ts.tv_sec = 0;
1950 spa->spa_loaded_ts.tv_nsec = 0;
1951 }
1952 if (error != EBADF) {
1953 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1954 }
1955 }
1956 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1957 spa->spa_ena = 0;
1958
1959 return (error);
1960}
1961
1962/*
1963 * Load an existing storage pool, using the pool's builtin spa_config as a
1964 * source of configuration information.
1965 */
1966static int
1967spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1968 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1969 char **ereport)
1970{
1971 int error = 0;
1972 nvlist_t *nvroot = NULL;
1973 nvlist_t *label;
1974 vdev_t *rvd;
1975 uberblock_t *ub = &spa->spa_uberblock;
1976 uint64_t children, config_cache_txg = spa->spa_config_txg;
1977 int orig_mode = spa->spa_mode;
1978 int parse;
1979 uint64_t obj;
1980 boolean_t missing_feat_write = B_FALSE;
1981
1982 /*
1983 * If this is an untrusted config, access the pool in read-only mode.
1984 * This prevents things like resilvering recently removed devices.
1985 */
1986 if (!mosconfig)
1987 spa->spa_mode = FREAD;
1988
1989 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1990
1991 spa->spa_load_state = state;
1992
1993 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1994 return (EINVAL);
1995
1996 parse = (type == SPA_IMPORT_EXISTING ?
1997 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1998
1999 /*
2000 * Create "The Godfather" zio to hold all async IOs
2001 */
2002 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2003 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2004
2005 /*
2006 * Parse the configuration into a vdev tree. We explicitly set the
2007 * value that will be returned by spa_version() since parsing the
2008 * configuration requires knowing the version number.
2009 */
2010 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2011 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2012 spa_config_exit(spa, SCL_ALL, FTAG);
2013
2014 if (error != 0)
2015 return (error);
2016
2017 ASSERT(spa->spa_root_vdev == rvd);
2018
2019 if (type != SPA_IMPORT_ASSEMBLE) {
2020 ASSERT(spa_guid(spa) == pool_guid);
2021 }
2022
2023 /*
2024 * Try to open all vdevs, loading each label in the process.
2025 */
2026 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2027 error = vdev_open(rvd);
2028 spa_config_exit(spa, SCL_ALL, FTAG);
2029 if (error != 0)
2030 return (error);
2031
2032 /*
2033 * We need to validate the vdev labels against the configuration that
2034 * we have in hand, which is dependent on the setting of mosconfig. If
2035 * mosconfig is true then we're validating the vdev labels based on
2036 * that config. Otherwise, we're validating against the cached config
2037 * (zpool.cache) that was read when we loaded the zfs module, and then
2038 * later we will recursively call spa_load() and validate against
2039 * the vdev config.
2040 *
2041 * If we're assembling a new pool that's been split off from an
2042 * existing pool, the labels haven't yet been updated so we skip
2043 * validation for now.
2044 */
2045 if (type != SPA_IMPORT_ASSEMBLE) {
2046 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2047 error = vdev_validate(rvd, mosconfig);
2048 spa_config_exit(spa, SCL_ALL, FTAG);
2049
2050 if (error != 0)
2051 return (error);
2052
2053 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2054 return (ENXIO);
2055 }
2056
2057 /*
2058 * Find the best uberblock.
2059 */
2060 vdev_uberblock_load(rvd, ub, &label);
2061
2062 /*
2063 * If we weren't able to find a single valid uberblock, return failure.
2064 */
2065 if (ub->ub_txg == 0) {
2066 nvlist_free(label);
2067 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2068 }
2069
2070 /*
2071 * If the pool has an unsupported version we can't open it.
2072 */
2073 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2074 nvlist_free(label);
2075 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2076 }
2077
2078 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2079 nvlist_t *features;
2080
2081 /*
2082 * If we weren't able to find what's necessary for reading the
2083 * MOS in the label, return failure.
2084 */
2085 if (label == NULL || nvlist_lookup_nvlist(label,
2086 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2087 nvlist_free(label);
2088 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2089 ENXIO));
2090 }
2091
2092 /*
2093 * Update our in-core representation with the definitive values
2094 * from the label.
2095 */
2096 nvlist_free(spa->spa_label_features);
2097 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2098 }
2099
2100 nvlist_free(label);
2101
2102 /*
2103 * Look through entries in the label nvlist's features_for_read. If
2104 * there is a feature listed there which we don't understand then we
2105 * cannot open a pool.
2106 */
2107 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2108 nvlist_t *unsup_feat;
2109
2110 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2111 0);
2112
2113 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2114 NULL); nvp != NULL;
2115 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2116 if (!zfeature_is_supported(nvpair_name(nvp))) {
2117 VERIFY(nvlist_add_string(unsup_feat,
2118 nvpair_name(nvp), "") == 0);
2119 }
2120 }
2121
2122 if (!nvlist_empty(unsup_feat)) {
2123 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2124 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2125 nvlist_free(unsup_feat);
2126 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2127 ENOTSUP));
2128 }
2129
2130 nvlist_free(unsup_feat);
2131 }
2132
2133 /*
2134 * If the vdev guid sum doesn't match the uberblock, we have an
2135 * incomplete configuration. We first check to see if the pool
2136 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2137 * If it is, defer the vdev_guid_sum check till later so we
2138 * can handle missing vdevs.
2139 */
2140 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2141 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2142 rvd->vdev_guid_sum != ub->ub_guid_sum)
2143 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2144
2145 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2146 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2147 spa_try_repair(spa, config);
2148 spa_config_exit(spa, SCL_ALL, FTAG);
2149 nvlist_free(spa->spa_config_splitting);
2150 spa->spa_config_splitting = NULL;
2151 }
2152
2153 /*
2154 * Initialize internal SPA structures.
2155 */
2156 spa->spa_state = POOL_STATE_ACTIVE;
2157 spa->spa_ubsync = spa->spa_uberblock;
2158 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2159 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2160 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2161 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2162 spa->spa_claim_max_txg = spa->spa_first_txg;
2163 spa->spa_prev_software_version = ub->ub_software_version;
2164
2165 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2166 if (error)
2167 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2168 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2169
2170 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2171 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2172
2173 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2174 boolean_t missing_feat_read = B_FALSE;
| 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) 2012 by Delphix. All rights reserved.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/spa_boot.h>
61#include <sys/zfs_ioctl.h>
62#include <sys/dsl_scan.h>
63#include <sys/zfeature.h>
64#include <sys/zvol.h>
65
66#ifdef _KERNEL
67#include <sys/callb.h>
68#include <sys/cpupart.h>
69#include <sys/zone.h>
70#endif /* _KERNEL */
71
72#include "zfs_prop.h"
73#include "zfs_comutil.h"
74
75/* Check hostid on import? */
76static int check_hostid = 1;
77
78SYSCTL_DECL(_vfs_zfs);
79TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
80SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
81 "Check hostid on import?");
82
83typedef enum zti_modes {
84 zti_mode_fixed, /* value is # of threads (min 1) */
85 zti_mode_online_percent, /* value is % of online CPUs */
86 zti_mode_batch, /* cpu-intensive; value is ignored */
87 zti_mode_null, /* don't create a taskq */
88 zti_nmodes
89} zti_modes_t;
90
91#define ZTI_FIX(n) { zti_mode_fixed, (n) }
92#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
93#define ZTI_BATCH { zti_mode_batch, 0 }
94#define ZTI_NULL { zti_mode_null, 0 }
95
96#define ZTI_ONE ZTI_FIX(1)
97
98typedef struct zio_taskq_info {
99 enum zti_modes zti_mode;
100 uint_t zti_value;
101} zio_taskq_info_t;
102
103static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
104 "issue", "issue_high", "intr", "intr_high"
105};
106
107/*
108 * Define the taskq threads for the following I/O types:
109 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
110 */
111const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
112 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
115 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
116 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
119};
120
121static dsl_syncfunc_t spa_sync_version;
122static dsl_syncfunc_t spa_sync_props;
123static boolean_t spa_has_active_shared_spare(spa_t *spa);
124static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
125 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
126 char **ereport);
127static void spa_vdev_resilver_done(spa_t *spa);
128
129uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
130#ifdef PSRSET_BIND
131id_t zio_taskq_psrset_bind = PS_NONE;
132#endif
133#ifdef SYSDC
134boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
135#endif
136uint_t zio_taskq_basedc = 80; /* base duty cycle */
137
138boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
139
140/*
141 * This (illegal) pool name is used when temporarily importing a spa_t in order
142 * to get the vdev stats associated with the imported devices.
143 */
144#define TRYIMPORT_NAME "$import"
145
146/*
147 * ==========================================================================
148 * SPA properties routines
149 * ==========================================================================
150 */
151
152/*
153 * Add a (source=src, propname=propval) list to an nvlist.
154 */
155static void
156spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
157 uint64_t intval, zprop_source_t src)
158{
159 const char *propname = zpool_prop_to_name(prop);
160 nvlist_t *propval;
161
162 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
163 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
164
165 if (strval != NULL)
166 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
167 else
168 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
169
170 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
171 nvlist_free(propval);
172}
173
174/*
175 * Get property values from the spa configuration.
176 */
177static void
178spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
179{
180 vdev_t *rvd = spa->spa_root_vdev;
181 dsl_pool_t *pool = spa->spa_dsl_pool;
182 uint64_t size;
183 uint64_t alloc;
184 uint64_t space;
185 uint64_t cap, version;
186 zprop_source_t src = ZPROP_SRC_NONE;
187 spa_config_dirent_t *dp;
188
189 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
190
191 if (rvd != NULL) {
192 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
193 size = metaslab_class_get_space(spa_normal_class(spa));
194 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
195 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
196 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
197 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
198 size - alloc, src);
199
200 space = 0;
201 for (int c = 0; c < rvd->vdev_children; c++) {
202 vdev_t *tvd = rvd->vdev_child[c];
203 space += tvd->vdev_max_asize - tvd->vdev_asize;
204 }
205 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
206 src);
207
208 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
209 (spa_mode(spa) == FREAD), src);
210
211 cap = (size == 0) ? 0 : (alloc * 100 / size);
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
213
214 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
215 ddt_get_pool_dedup_ratio(spa), src);
216
217 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
218 rvd->vdev_state, src);
219
220 version = spa_version(spa);
221 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
222 src = ZPROP_SRC_DEFAULT;
223 else
224 src = ZPROP_SRC_LOCAL;
225 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
226 }
227
228 if (pool != NULL) {
229 dsl_dir_t *freedir = pool->dp_free_dir;
230
231 /*
232 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
233 * when opening pools before this version freedir will be NULL.
234 */
235 if (freedir != NULL) {
236 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
237 freedir->dd_phys->dd_used_bytes, src);
238 } else {
239 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
240 NULL, 0, src);
241 }
242 }
243
244 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
245
246 if (spa->spa_comment != NULL) {
247 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
248 0, ZPROP_SRC_LOCAL);
249 }
250
251 if (spa->spa_root != NULL)
252 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
253 0, ZPROP_SRC_LOCAL);
254
255 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
256 if (dp->scd_path == NULL) {
257 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
258 "none", 0, ZPROP_SRC_LOCAL);
259 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
260 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
261 dp->scd_path, 0, ZPROP_SRC_LOCAL);
262 }
263 }
264}
265
266/*
267 * Get zpool property values.
268 */
269int
270spa_prop_get(spa_t *spa, nvlist_t **nvp)
271{
272 objset_t *mos = spa->spa_meta_objset;
273 zap_cursor_t zc;
274 zap_attribute_t za;
275 int err;
276
277 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
278
279 mutex_enter(&spa->spa_props_lock);
280
281 /*
282 * Get properties from the spa config.
283 */
284 spa_prop_get_config(spa, nvp);
285
286 /* If no pool property object, no more prop to get. */
287 if (mos == NULL || spa->spa_pool_props_object == 0) {
288 mutex_exit(&spa->spa_props_lock);
289 return (0);
290 }
291
292 /*
293 * Get properties from the MOS pool property object.
294 */
295 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
296 (err = zap_cursor_retrieve(&zc, &za)) == 0;
297 zap_cursor_advance(&zc)) {
298 uint64_t intval = 0;
299 char *strval = NULL;
300 zprop_source_t src = ZPROP_SRC_DEFAULT;
301 zpool_prop_t prop;
302
303 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
304 continue;
305
306 switch (za.za_integer_length) {
307 case 8:
308 /* integer property */
309 if (za.za_first_integer !=
310 zpool_prop_default_numeric(prop))
311 src = ZPROP_SRC_LOCAL;
312
313 if (prop == ZPOOL_PROP_BOOTFS) {
314 dsl_pool_t *dp;
315 dsl_dataset_t *ds = NULL;
316
317 dp = spa_get_dsl(spa);
318 rw_enter(&dp->dp_config_rwlock, RW_READER);
319 if (err = dsl_dataset_hold_obj(dp,
320 za.za_first_integer, FTAG, &ds)) {
321 rw_exit(&dp->dp_config_rwlock);
322 break;
323 }
324
325 strval = kmem_alloc(
326 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
327 KM_SLEEP);
328 dsl_dataset_name(ds, strval);
329 dsl_dataset_rele(ds, FTAG);
330 rw_exit(&dp->dp_config_rwlock);
331 } else {
332 strval = NULL;
333 intval = za.za_first_integer;
334 }
335
336 spa_prop_add_list(*nvp, prop, strval, intval, src);
337
338 if (strval != NULL)
339 kmem_free(strval,
340 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
341
342 break;
343
344 case 1:
345 /* string property */
346 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
347 err = zap_lookup(mos, spa->spa_pool_props_object,
348 za.za_name, 1, za.za_num_integers, strval);
349 if (err) {
350 kmem_free(strval, za.za_num_integers);
351 break;
352 }
353 spa_prop_add_list(*nvp, prop, strval, 0, src);
354 kmem_free(strval, za.za_num_integers);
355 break;
356
357 default:
358 break;
359 }
360 }
361 zap_cursor_fini(&zc);
362 mutex_exit(&spa->spa_props_lock);
363out:
364 if (err && err != ENOENT) {
365 nvlist_free(*nvp);
366 *nvp = NULL;
367 return (err);
368 }
369
370 return (0);
371}
372
373/*
374 * Validate the given pool properties nvlist and modify the list
375 * for the property values to be set.
376 */
377static int
378spa_prop_validate(spa_t *spa, nvlist_t *props)
379{
380 nvpair_t *elem;
381 int error = 0, reset_bootfs = 0;
382 uint64_t objnum;
383 boolean_t has_feature = B_FALSE;
384
385 elem = NULL;
386 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
387 uint64_t intval;
388 char *strval, *slash, *check, *fname;
389 const char *propname = nvpair_name(elem);
390 zpool_prop_t prop = zpool_name_to_prop(propname);
391
392 switch (prop) {
393 case ZPROP_INVAL:
394 if (!zpool_prop_feature(propname)) {
395 error = EINVAL;
396 break;
397 }
398
399 /*
400 * Sanitize the input.
401 */
402 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
403 error = EINVAL;
404 break;
405 }
406
407 if (nvpair_value_uint64(elem, &intval) != 0) {
408 error = EINVAL;
409 break;
410 }
411
412 if (intval != 0) {
413 error = EINVAL;
414 break;
415 }
416
417 fname = strchr(propname, '@') + 1;
418 if (zfeature_lookup_name(fname, NULL) != 0) {
419 error = EINVAL;
420 break;
421 }
422
423 has_feature = B_TRUE;
424 break;
425
426 case ZPOOL_PROP_VERSION:
427 error = nvpair_value_uint64(elem, &intval);
428 if (!error &&
429 (intval < spa_version(spa) ||
430 intval > SPA_VERSION_BEFORE_FEATURES ||
431 has_feature))
432 error = EINVAL;
433 break;
434
435 case ZPOOL_PROP_DELEGATION:
436 case ZPOOL_PROP_AUTOREPLACE:
437 case ZPOOL_PROP_LISTSNAPS:
438 case ZPOOL_PROP_AUTOEXPAND:
439 error = nvpair_value_uint64(elem, &intval);
440 if (!error && intval > 1)
441 error = EINVAL;
442 break;
443
444 case ZPOOL_PROP_BOOTFS:
445 /*
446 * If the pool version is less than SPA_VERSION_BOOTFS,
447 * or the pool is still being created (version == 0),
448 * the bootfs property cannot be set.
449 */
450 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
451 error = ENOTSUP;
452 break;
453 }
454
455 /*
456 * Make sure the vdev config is bootable
457 */
458 if (!vdev_is_bootable(spa->spa_root_vdev)) {
459 error = ENOTSUP;
460 break;
461 }
462
463 reset_bootfs = 1;
464
465 error = nvpair_value_string(elem, &strval);
466
467 if (!error) {
468 objset_t *os;
469 uint64_t compress;
470
471 if (strval == NULL || strval[0] == '\0') {
472 objnum = zpool_prop_default_numeric(
473 ZPOOL_PROP_BOOTFS);
474 break;
475 }
476
477 if (error = dmu_objset_hold(strval, FTAG, &os))
478 break;
479
480 /* Must be ZPL and not gzip compressed. */
481
482 if (dmu_objset_type(os) != DMU_OST_ZFS) {
483 error = ENOTSUP;
484 } else if ((error = dsl_prop_get_integer(strval,
485 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
486 &compress, NULL)) == 0 &&
487 !BOOTFS_COMPRESS_VALID(compress)) {
488 error = ENOTSUP;
489 } else {
490 objnum = dmu_objset_id(os);
491 }
492 dmu_objset_rele(os, FTAG);
493 }
494 break;
495
496 case ZPOOL_PROP_FAILUREMODE:
497 error = nvpair_value_uint64(elem, &intval);
498 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
499 intval > ZIO_FAILURE_MODE_PANIC))
500 error = EINVAL;
501
502 /*
503 * This is a special case which only occurs when
504 * the pool has completely failed. This allows
505 * the user to change the in-core failmode property
506 * without syncing it out to disk (I/Os might
507 * currently be blocked). We do this by returning
508 * EIO to the caller (spa_prop_set) to trick it
509 * into thinking we encountered a property validation
510 * error.
511 */
512 if (!error && spa_suspended(spa)) {
513 spa->spa_failmode = intval;
514 error = EIO;
515 }
516 break;
517
518 case ZPOOL_PROP_CACHEFILE:
519 if ((error = nvpair_value_string(elem, &strval)) != 0)
520 break;
521
522 if (strval[0] == '\0')
523 break;
524
525 if (strcmp(strval, "none") == 0)
526 break;
527
528 if (strval[0] != '/') {
529 error = EINVAL;
530 break;
531 }
532
533 slash = strrchr(strval, '/');
534 ASSERT(slash != NULL);
535
536 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
537 strcmp(slash, "/..") == 0)
538 error = EINVAL;
539 break;
540
541 case ZPOOL_PROP_COMMENT:
542 if ((error = nvpair_value_string(elem, &strval)) != 0)
543 break;
544 for (check = strval; *check != '\0'; check++) {
545 /*
546 * The kernel doesn't have an easy isprint()
547 * check. For this kernel check, we merely
548 * check ASCII apart from DEL. Fix this if
549 * there is an easy-to-use kernel isprint().
550 */
551 if (*check >= 0x7f) {
552 error = EINVAL;
553 break;
554 }
555 check++;
556 }
557 if (strlen(strval) > ZPROP_MAX_COMMENT)
558 error = E2BIG;
559 break;
560
561 case ZPOOL_PROP_DEDUPDITTO:
562 if (spa_version(spa) < SPA_VERSION_DEDUP)
563 error = ENOTSUP;
564 else
565 error = nvpair_value_uint64(elem, &intval);
566 if (error == 0 &&
567 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
568 error = EINVAL;
569 break;
570 }
571
572 if (error)
573 break;
574 }
575
576 if (!error && reset_bootfs) {
577 error = nvlist_remove(props,
578 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
579
580 if (!error) {
581 error = nvlist_add_uint64(props,
582 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
583 }
584 }
585
586 return (error);
587}
588
589void
590spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
591{
592 char *cachefile;
593 spa_config_dirent_t *dp;
594
595 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
596 &cachefile) != 0)
597 return;
598
599 dp = kmem_alloc(sizeof (spa_config_dirent_t),
600 KM_SLEEP);
601
602 if (cachefile[0] == '\0')
603 dp->scd_path = spa_strdup(spa_config_path);
604 else if (strcmp(cachefile, "none") == 0)
605 dp->scd_path = NULL;
606 else
607 dp->scd_path = spa_strdup(cachefile);
608
609 list_insert_head(&spa->spa_config_list, dp);
610 if (need_sync)
611 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
612}
613
614int
615spa_prop_set(spa_t *spa, nvlist_t *nvp)
616{
617 int error;
618 nvpair_t *elem = NULL;
619 boolean_t need_sync = B_FALSE;
620
621 if ((error = spa_prop_validate(spa, nvp)) != 0)
622 return (error);
623
624 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
625 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
626
627 if (prop == ZPOOL_PROP_CACHEFILE ||
628 prop == ZPOOL_PROP_ALTROOT ||
629 prop == ZPOOL_PROP_READONLY)
630 continue;
631
632 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
633 uint64_t ver;
634
635 if (prop == ZPOOL_PROP_VERSION) {
636 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
637 } else {
638 ASSERT(zpool_prop_feature(nvpair_name(elem)));
639 ver = SPA_VERSION_FEATURES;
640 need_sync = B_TRUE;
641 }
642
643 /* Save time if the version is already set. */
644 if (ver == spa_version(spa))
645 continue;
646
647 /*
648 * In addition to the pool directory object, we might
649 * create the pool properties object, the features for
650 * read object, the features for write object, or the
651 * feature descriptions object.
652 */
653 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
654 spa_sync_version, spa, &ver, 6);
655 if (error)
656 return (error);
657 continue;
658 }
659
660 need_sync = B_TRUE;
661 break;
662 }
663
664 if (need_sync) {
665 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
666 spa, nvp, 6));
667 }
668
669 return (0);
670}
671
672/*
673 * If the bootfs property value is dsobj, clear it.
674 */
675void
676spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
677{
678 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
679 VERIFY(zap_remove(spa->spa_meta_objset,
680 spa->spa_pool_props_object,
681 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
682 spa->spa_bootfs = 0;
683 }
684}
685
686/*
687 * Change the GUID for the pool. This is done so that we can later
688 * re-import a pool built from a clone of our own vdevs. We will modify
689 * the root vdev's guid, our own pool guid, and then mark all of our
690 * vdevs dirty. Note that we must make sure that all our vdevs are
691 * online when we do this, or else any vdevs that weren't present
692 * would be orphaned from our pool. We are also going to issue a
693 * sysevent to update any watchers.
694 */
695int
696spa_change_guid(spa_t *spa)
697{
698 uint64_t oldguid, newguid;
699 uint64_t txg;
700
701 if (!(spa_mode_global & FWRITE))
702 return (EROFS);
703
704 txg = spa_vdev_enter(spa);
705
706 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
707 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
708
709 oldguid = spa_guid(spa);
710 newguid = spa_generate_guid(NULL);
711 ASSERT3U(oldguid, !=, newguid);
712
713 spa->spa_root_vdev->vdev_guid = newguid;
714 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
715
716 vdev_config_dirty(spa->spa_root_vdev);
717
718 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
719
720 return (spa_vdev_exit(spa, NULL, txg, 0));
721}
722
723/*
724 * ==========================================================================
725 * SPA state manipulation (open/create/destroy/import/export)
726 * ==========================================================================
727 */
728
729static int
730spa_error_entry_compare(const void *a, const void *b)
731{
732 spa_error_entry_t *sa = (spa_error_entry_t *)a;
733 spa_error_entry_t *sb = (spa_error_entry_t *)b;
734 int ret;
735
736 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
737 sizeof (zbookmark_t));
738
739 if (ret < 0)
740 return (-1);
741 else if (ret > 0)
742 return (1);
743 else
744 return (0);
745}
746
747/*
748 * Utility function which retrieves copies of the current logs and
749 * re-initializes them in the process.
750 */
751void
752spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
753{
754 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
755
756 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
757 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
758
759 avl_create(&spa->spa_errlist_scrub,
760 spa_error_entry_compare, sizeof (spa_error_entry_t),
761 offsetof(spa_error_entry_t, se_avl));
762 avl_create(&spa->spa_errlist_last,
763 spa_error_entry_compare, sizeof (spa_error_entry_t),
764 offsetof(spa_error_entry_t, se_avl));
765}
766
767static taskq_t *
768spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
769 uint_t value)
770{
771 uint_t flags = TASKQ_PREPOPULATE;
772 boolean_t batch = B_FALSE;
773
774 switch (mode) {
775 case zti_mode_null:
776 return (NULL); /* no taskq needed */
777
778 case zti_mode_fixed:
779 ASSERT3U(value, >=, 1);
780 value = MAX(value, 1);
781 break;
782
783 case zti_mode_batch:
784 batch = B_TRUE;
785 flags |= TASKQ_THREADS_CPU_PCT;
786 value = zio_taskq_batch_pct;
787 break;
788
789 case zti_mode_online_percent:
790 flags |= TASKQ_THREADS_CPU_PCT;
791 break;
792
793 default:
794 panic("unrecognized mode for %s taskq (%u:%u) in "
795 "spa_activate()",
796 name, mode, value);
797 break;
798 }
799
800#ifdef SYSDC
801 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
802 if (batch)
803 flags |= TASKQ_DC_BATCH;
804
805 return (taskq_create_sysdc(name, value, 50, INT_MAX,
806 spa->spa_proc, zio_taskq_basedc, flags));
807 }
808#endif
809 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
810 spa->spa_proc, flags));
811}
812
813static void
814spa_create_zio_taskqs(spa_t *spa)
815{
816 for (int t = 0; t < ZIO_TYPES; t++) {
817 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
818 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
819 enum zti_modes mode = ztip->zti_mode;
820 uint_t value = ztip->zti_value;
821 char name[32];
822
823 (void) snprintf(name, sizeof (name),
824 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
825
826 spa->spa_zio_taskq[t][q] =
827 spa_taskq_create(spa, name, mode, value);
828 }
829 }
830}
831
832#ifdef _KERNEL
833#ifdef SPA_PROCESS
834static void
835spa_thread(void *arg)
836{
837 callb_cpr_t cprinfo;
838
839 spa_t *spa = arg;
840 user_t *pu = PTOU(curproc);
841
842 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
843 spa->spa_name);
844
845 ASSERT(curproc != &p0);
846 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
847 "zpool-%s", spa->spa_name);
848 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
849
850#ifdef PSRSET_BIND
851 /* bind this thread to the requested psrset */
852 if (zio_taskq_psrset_bind != PS_NONE) {
853 pool_lock();
854 mutex_enter(&cpu_lock);
855 mutex_enter(&pidlock);
856 mutex_enter(&curproc->p_lock);
857
858 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
859 0, NULL, NULL) == 0) {
860 curthread->t_bind_pset = zio_taskq_psrset_bind;
861 } else {
862 cmn_err(CE_WARN,
863 "Couldn't bind process for zfs pool \"%s\" to "
864 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
865 }
866
867 mutex_exit(&curproc->p_lock);
868 mutex_exit(&pidlock);
869 mutex_exit(&cpu_lock);
870 pool_unlock();
871 }
872#endif
873
874#ifdef SYSDC
875 if (zio_taskq_sysdc) {
876 sysdc_thread_enter(curthread, 100, 0);
877 }
878#endif
879
880 spa->spa_proc = curproc;
881 spa->spa_did = curthread->t_did;
882
883 spa_create_zio_taskqs(spa);
884
885 mutex_enter(&spa->spa_proc_lock);
886 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
887
888 spa->spa_proc_state = SPA_PROC_ACTIVE;
889 cv_broadcast(&spa->spa_proc_cv);
890
891 CALLB_CPR_SAFE_BEGIN(&cprinfo);
892 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
893 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
894 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
895
896 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
897 spa->spa_proc_state = SPA_PROC_GONE;
898 spa->spa_proc = &p0;
899 cv_broadcast(&spa->spa_proc_cv);
900 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
901
902 mutex_enter(&curproc->p_lock);
903 lwp_exit();
904}
905#endif /* SPA_PROCESS */
906#endif
907
908/*
909 * Activate an uninitialized pool.
910 */
911static void
912spa_activate(spa_t *spa, int mode)
913{
914 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
915
916 spa->spa_state = POOL_STATE_ACTIVE;
917 spa->spa_mode = mode;
918
919 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
920 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
921
922 /* Try to create a covering process */
923 mutex_enter(&spa->spa_proc_lock);
924 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
925 ASSERT(spa->spa_proc == &p0);
926 spa->spa_did = 0;
927
928#ifdef SPA_PROCESS
929 /* Only create a process if we're going to be around a while. */
930 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
931 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
932 NULL, 0) == 0) {
933 spa->spa_proc_state = SPA_PROC_CREATED;
934 while (spa->spa_proc_state == SPA_PROC_CREATED) {
935 cv_wait(&spa->spa_proc_cv,
936 &spa->spa_proc_lock);
937 }
938 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
939 ASSERT(spa->spa_proc != &p0);
940 ASSERT(spa->spa_did != 0);
941 } else {
942#ifdef _KERNEL
943 cmn_err(CE_WARN,
944 "Couldn't create process for zfs pool \"%s\"\n",
945 spa->spa_name);
946#endif
947 }
948 }
949#endif /* SPA_PROCESS */
950 mutex_exit(&spa->spa_proc_lock);
951
952 /* If we didn't create a process, we need to create our taskqs. */
953 ASSERT(spa->spa_proc == &p0);
954 if (spa->spa_proc == &p0) {
955 spa_create_zio_taskqs(spa);
956 }
957
958 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
959 offsetof(vdev_t, vdev_config_dirty_node));
960 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
961 offsetof(vdev_t, vdev_state_dirty_node));
962
963 txg_list_create(&spa->spa_vdev_txg_list,
964 offsetof(struct vdev, vdev_txg_node));
965
966 avl_create(&spa->spa_errlist_scrub,
967 spa_error_entry_compare, sizeof (spa_error_entry_t),
968 offsetof(spa_error_entry_t, se_avl));
969 avl_create(&spa->spa_errlist_last,
970 spa_error_entry_compare, sizeof (spa_error_entry_t),
971 offsetof(spa_error_entry_t, se_avl));
972}
973
974/*
975 * Opposite of spa_activate().
976 */
977static void
978spa_deactivate(spa_t *spa)
979{
980 ASSERT(spa->spa_sync_on == B_FALSE);
981 ASSERT(spa->spa_dsl_pool == NULL);
982 ASSERT(spa->spa_root_vdev == NULL);
983 ASSERT(spa->spa_async_zio_root == NULL);
984 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
985
986 txg_list_destroy(&spa->spa_vdev_txg_list);
987
988 list_destroy(&spa->spa_config_dirty_list);
989 list_destroy(&spa->spa_state_dirty_list);
990
991 for (int t = 0; t < ZIO_TYPES; t++) {
992 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
993 if (spa->spa_zio_taskq[t][q] != NULL)
994 taskq_destroy(spa->spa_zio_taskq[t][q]);
995 spa->spa_zio_taskq[t][q] = NULL;
996 }
997 }
998
999 metaslab_class_destroy(spa->spa_normal_class);
1000 spa->spa_normal_class = NULL;
1001
1002 metaslab_class_destroy(spa->spa_log_class);
1003 spa->spa_log_class = NULL;
1004
1005 /*
1006 * If this was part of an import or the open otherwise failed, we may
1007 * still have errors left in the queues. Empty them just in case.
1008 */
1009 spa_errlog_drain(spa);
1010
1011 avl_destroy(&spa->spa_errlist_scrub);
1012 avl_destroy(&spa->spa_errlist_last);
1013
1014 spa->spa_state = POOL_STATE_UNINITIALIZED;
1015
1016 mutex_enter(&spa->spa_proc_lock);
1017 if (spa->spa_proc_state != SPA_PROC_NONE) {
1018 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1019 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1020 cv_broadcast(&spa->spa_proc_cv);
1021 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1022 ASSERT(spa->spa_proc != &p0);
1023 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1024 }
1025 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1026 spa->spa_proc_state = SPA_PROC_NONE;
1027 }
1028 ASSERT(spa->spa_proc == &p0);
1029 mutex_exit(&spa->spa_proc_lock);
1030
1031#ifdef SPA_PROCESS
1032 /*
1033 * We want to make sure spa_thread() has actually exited the ZFS
1034 * module, so that the module can't be unloaded out from underneath
1035 * it.
1036 */
1037 if (spa->spa_did != 0) {
1038 thread_join(spa->spa_did);
1039 spa->spa_did = 0;
1040 }
1041#endif /* SPA_PROCESS */
1042}
1043
1044/*
1045 * Verify a pool configuration, and construct the vdev tree appropriately. This
1046 * will create all the necessary vdevs in the appropriate layout, with each vdev
1047 * in the CLOSED state. This will prep the pool before open/creation/import.
1048 * All vdev validation is done by the vdev_alloc() routine.
1049 */
1050static int
1051spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1052 uint_t id, int atype)
1053{
1054 nvlist_t **child;
1055 uint_t children;
1056 int error;
1057
1058 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1059 return (error);
1060
1061 if ((*vdp)->vdev_ops->vdev_op_leaf)
1062 return (0);
1063
1064 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1065 &child, &children);
1066
1067 if (error == ENOENT)
1068 return (0);
1069
1070 if (error) {
1071 vdev_free(*vdp);
1072 *vdp = NULL;
1073 return (EINVAL);
1074 }
1075
1076 for (int c = 0; c < children; c++) {
1077 vdev_t *vd;
1078 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1079 atype)) != 0) {
1080 vdev_free(*vdp);
1081 *vdp = NULL;
1082 return (error);
1083 }
1084 }
1085
1086 ASSERT(*vdp != NULL);
1087
1088 return (0);
1089}
1090
1091/*
1092 * Opposite of spa_load().
1093 */
1094static void
1095spa_unload(spa_t *spa)
1096{
1097 int i;
1098
1099 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1100
1101 /*
1102 * Stop async tasks.
1103 */
1104 spa_async_suspend(spa);
1105
1106 /*
1107 * Stop syncing.
1108 */
1109 if (spa->spa_sync_on) {
1110 txg_sync_stop(spa->spa_dsl_pool);
1111 spa->spa_sync_on = B_FALSE;
1112 }
1113
1114 /*
1115 * Wait for any outstanding async I/O to complete.
1116 */
1117 if (spa->spa_async_zio_root != NULL) {
1118 (void) zio_wait(spa->spa_async_zio_root);
1119 spa->spa_async_zio_root = NULL;
1120 }
1121
1122 bpobj_close(&spa->spa_deferred_bpobj);
1123
1124 /*
1125 * Close the dsl pool.
1126 */
1127 if (spa->spa_dsl_pool) {
1128 dsl_pool_close(spa->spa_dsl_pool);
1129 spa->spa_dsl_pool = NULL;
1130 spa->spa_meta_objset = NULL;
1131 }
1132
1133 ddt_unload(spa);
1134
1135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1136
1137 /*
1138 * Drop and purge level 2 cache
1139 */
1140 spa_l2cache_drop(spa);
1141
1142 /*
1143 * Close all vdevs.
1144 */
1145 if (spa->spa_root_vdev)
1146 vdev_free(spa->spa_root_vdev);
1147 ASSERT(spa->spa_root_vdev == NULL);
1148
1149 for (i = 0; i < spa->spa_spares.sav_count; i++)
1150 vdev_free(spa->spa_spares.sav_vdevs[i]);
1151 if (spa->spa_spares.sav_vdevs) {
1152 kmem_free(spa->spa_spares.sav_vdevs,
1153 spa->spa_spares.sav_count * sizeof (void *));
1154 spa->spa_spares.sav_vdevs = NULL;
1155 }
1156 if (spa->spa_spares.sav_config) {
1157 nvlist_free(spa->spa_spares.sav_config);
1158 spa->spa_spares.sav_config = NULL;
1159 }
1160 spa->spa_spares.sav_count = 0;
1161
1162 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1163 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1164 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1165 }
1166 if (spa->spa_l2cache.sav_vdevs) {
1167 kmem_free(spa->spa_l2cache.sav_vdevs,
1168 spa->spa_l2cache.sav_count * sizeof (void *));
1169 spa->spa_l2cache.sav_vdevs = NULL;
1170 }
1171 if (spa->spa_l2cache.sav_config) {
1172 nvlist_free(spa->spa_l2cache.sav_config);
1173 spa->spa_l2cache.sav_config = NULL;
1174 }
1175 spa->spa_l2cache.sav_count = 0;
1176
1177 spa->spa_async_suspended = 0;
1178
1179 if (spa->spa_comment != NULL) {
1180 spa_strfree(spa->spa_comment);
1181 spa->spa_comment = NULL;
1182 }
1183
1184 spa_config_exit(spa, SCL_ALL, FTAG);
1185}
1186
1187/*
1188 * Load (or re-load) the current list of vdevs describing the active spares for
1189 * this pool. When this is called, we have some form of basic information in
1190 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1191 * then re-generate a more complete list including status information.
1192 */
1193static void
1194spa_load_spares(spa_t *spa)
1195{
1196 nvlist_t **spares;
1197 uint_t nspares;
1198 int i;
1199 vdev_t *vd, *tvd;
1200
1201 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1202
1203 /*
1204 * First, close and free any existing spare vdevs.
1205 */
1206 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1207 vd = spa->spa_spares.sav_vdevs[i];
1208
1209 /* Undo the call to spa_activate() below */
1210 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1211 B_FALSE)) != NULL && tvd->vdev_isspare)
1212 spa_spare_remove(tvd);
1213 vdev_close(vd);
1214 vdev_free(vd);
1215 }
1216
1217 if (spa->spa_spares.sav_vdevs)
1218 kmem_free(spa->spa_spares.sav_vdevs,
1219 spa->spa_spares.sav_count * sizeof (void *));
1220
1221 if (spa->spa_spares.sav_config == NULL)
1222 nspares = 0;
1223 else
1224 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1225 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1226
1227 spa->spa_spares.sav_count = (int)nspares;
1228 spa->spa_spares.sav_vdevs = NULL;
1229
1230 if (nspares == 0)
1231 return;
1232
1233 /*
1234 * Construct the array of vdevs, opening them to get status in the
1235 * process. For each spare, there is potentially two different vdev_t
1236 * structures associated with it: one in the list of spares (used only
1237 * for basic validation purposes) and one in the active vdev
1238 * configuration (if it's spared in). During this phase we open and
1239 * validate each vdev on the spare list. If the vdev also exists in the
1240 * active configuration, then we also mark this vdev as an active spare.
1241 */
1242 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1243 KM_SLEEP);
1244 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1245 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1246 VDEV_ALLOC_SPARE) == 0);
1247 ASSERT(vd != NULL);
1248
1249 spa->spa_spares.sav_vdevs[i] = vd;
1250
1251 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1252 B_FALSE)) != NULL) {
1253 if (!tvd->vdev_isspare)
1254 spa_spare_add(tvd);
1255
1256 /*
1257 * We only mark the spare active if we were successfully
1258 * able to load the vdev. Otherwise, importing a pool
1259 * with a bad active spare would result in strange
1260 * behavior, because multiple pool would think the spare
1261 * is actively in use.
1262 *
1263 * There is a vulnerability here to an equally bizarre
1264 * circumstance, where a dead active spare is later
1265 * brought back to life (onlined or otherwise). Given
1266 * the rarity of this scenario, and the extra complexity
1267 * it adds, we ignore the possibility.
1268 */
1269 if (!vdev_is_dead(tvd))
1270 spa_spare_activate(tvd);
1271 }
1272
1273 vd->vdev_top = vd;
1274 vd->vdev_aux = &spa->spa_spares;
1275
1276 if (vdev_open(vd) != 0)
1277 continue;
1278
1279 if (vdev_validate_aux(vd) == 0)
1280 spa_spare_add(vd);
1281 }
1282
1283 /*
1284 * Recompute the stashed list of spares, with status information
1285 * this time.
1286 */
1287 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1288 DATA_TYPE_NVLIST_ARRAY) == 0);
1289
1290 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1291 KM_SLEEP);
1292 for (i = 0; i < spa->spa_spares.sav_count; i++)
1293 spares[i] = vdev_config_generate(spa,
1294 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1295 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1296 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1297 for (i = 0; i < spa->spa_spares.sav_count; i++)
1298 nvlist_free(spares[i]);
1299 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1300}
1301
1302/*
1303 * Load (or re-load) the current list of vdevs describing the active l2cache for
1304 * this pool. When this is called, we have some form of basic information in
1305 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1306 * then re-generate a more complete list including status information.
1307 * Devices which are already active have their details maintained, and are
1308 * not re-opened.
1309 */
1310static void
1311spa_load_l2cache(spa_t *spa)
1312{
1313 nvlist_t **l2cache;
1314 uint_t nl2cache;
1315 int i, j, oldnvdevs;
1316 uint64_t guid;
1317 vdev_t *vd, **oldvdevs, **newvdevs;
1318 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1319
1320 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1321
1322 if (sav->sav_config != NULL) {
1323 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1324 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1325 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1326 } else {
1327 nl2cache = 0;
1328 }
1329
1330 oldvdevs = sav->sav_vdevs;
1331 oldnvdevs = sav->sav_count;
1332 sav->sav_vdevs = NULL;
1333 sav->sav_count = 0;
1334
1335 /*
1336 * Process new nvlist of vdevs.
1337 */
1338 for (i = 0; i < nl2cache; i++) {
1339 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1340 &guid) == 0);
1341
1342 newvdevs[i] = NULL;
1343 for (j = 0; j < oldnvdevs; j++) {
1344 vd = oldvdevs[j];
1345 if (vd != NULL && guid == vd->vdev_guid) {
1346 /*
1347 * Retain previous vdev for add/remove ops.
1348 */
1349 newvdevs[i] = vd;
1350 oldvdevs[j] = NULL;
1351 break;
1352 }
1353 }
1354
1355 if (newvdevs[i] == NULL) {
1356 /*
1357 * Create new vdev
1358 */
1359 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1360 VDEV_ALLOC_L2CACHE) == 0);
1361 ASSERT(vd != NULL);
1362 newvdevs[i] = vd;
1363
1364 /*
1365 * Commit this vdev as an l2cache device,
1366 * even if it fails to open.
1367 */
1368 spa_l2cache_add(vd);
1369
1370 vd->vdev_top = vd;
1371 vd->vdev_aux = sav;
1372
1373 spa_l2cache_activate(vd);
1374
1375 if (vdev_open(vd) != 0)
1376 continue;
1377
1378 (void) vdev_validate_aux(vd);
1379
1380 if (!vdev_is_dead(vd))
1381 l2arc_add_vdev(spa, vd);
1382 }
1383 }
1384
1385 /*
1386 * Purge vdevs that were dropped
1387 */
1388 for (i = 0; i < oldnvdevs; i++) {
1389 uint64_t pool;
1390
1391 vd = oldvdevs[i];
1392 if (vd != NULL) {
1393 ASSERT(vd->vdev_isl2cache);
1394
1395 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1396 pool != 0ULL && l2arc_vdev_present(vd))
1397 l2arc_remove_vdev(vd);
1398 vdev_clear_stats(vd);
1399 vdev_free(vd);
1400 }
1401 }
1402
1403 if (oldvdevs)
1404 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1405
1406 if (sav->sav_config == NULL)
1407 goto out;
1408
1409 sav->sav_vdevs = newvdevs;
1410 sav->sav_count = (int)nl2cache;
1411
1412 /*
1413 * Recompute the stashed list of l2cache devices, with status
1414 * information this time.
1415 */
1416 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1417 DATA_TYPE_NVLIST_ARRAY) == 0);
1418
1419 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1420 for (i = 0; i < sav->sav_count; i++)
1421 l2cache[i] = vdev_config_generate(spa,
1422 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1423 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1424 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1425out:
1426 for (i = 0; i < sav->sav_count; i++)
1427 nvlist_free(l2cache[i]);
1428 if (sav->sav_count)
1429 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1430}
1431
1432static int
1433load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1434{
1435 dmu_buf_t *db;
1436 char *packed = NULL;
1437 size_t nvsize = 0;
1438 int error;
1439 *value = NULL;
1440
1441 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1442 nvsize = *(uint64_t *)db->db_data;
1443 dmu_buf_rele(db, FTAG);
1444
1445 packed = kmem_alloc(nvsize, KM_SLEEP);
1446 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1447 DMU_READ_PREFETCH);
1448 if (error == 0)
1449 error = nvlist_unpack(packed, nvsize, value, 0);
1450 kmem_free(packed, nvsize);
1451
1452 return (error);
1453}
1454
1455/*
1456 * Checks to see if the given vdev could not be opened, in which case we post a
1457 * sysevent to notify the autoreplace code that the device has been removed.
1458 */
1459static void
1460spa_check_removed(vdev_t *vd)
1461{
1462 for (int c = 0; c < vd->vdev_children; c++)
1463 spa_check_removed(vd->vdev_child[c]);
1464
1465 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1466 zfs_post_autoreplace(vd->vdev_spa, vd);
1467 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1468 }
1469}
1470
1471/*
1472 * Validate the current config against the MOS config
1473 */
1474static boolean_t
1475spa_config_valid(spa_t *spa, nvlist_t *config)
1476{
1477 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1478 nvlist_t *nv;
1479
1480 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1481
1482 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1483 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1484
1485 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1486
1487 /*
1488 * If we're doing a normal import, then build up any additional
1489 * diagnostic information about missing devices in this config.
1490 * We'll pass this up to the user for further processing.
1491 */
1492 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1493 nvlist_t **child, *nv;
1494 uint64_t idx = 0;
1495
1496 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1497 KM_SLEEP);
1498 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1499
1500 for (int c = 0; c < rvd->vdev_children; c++) {
1501 vdev_t *tvd = rvd->vdev_child[c];
1502 vdev_t *mtvd = mrvd->vdev_child[c];
1503
1504 if (tvd->vdev_ops == &vdev_missing_ops &&
1505 mtvd->vdev_ops != &vdev_missing_ops &&
1506 mtvd->vdev_islog)
1507 child[idx++] = vdev_config_generate(spa, mtvd,
1508 B_FALSE, 0);
1509 }
1510
1511 if (idx) {
1512 VERIFY(nvlist_add_nvlist_array(nv,
1513 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1514 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1515 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1516
1517 for (int i = 0; i < idx; i++)
1518 nvlist_free(child[i]);
1519 }
1520 nvlist_free(nv);
1521 kmem_free(child, rvd->vdev_children * sizeof (char **));
1522 }
1523
1524 /*
1525 * Compare the root vdev tree with the information we have
1526 * from the MOS config (mrvd). Check each top-level vdev
1527 * with the corresponding MOS config top-level (mtvd).
1528 */
1529 for (int c = 0; c < rvd->vdev_children; c++) {
1530 vdev_t *tvd = rvd->vdev_child[c];
1531 vdev_t *mtvd = mrvd->vdev_child[c];
1532
1533 /*
1534 * Resolve any "missing" vdevs in the current configuration.
1535 * If we find that the MOS config has more accurate information
1536 * about the top-level vdev then use that vdev instead.
1537 */
1538 if (tvd->vdev_ops == &vdev_missing_ops &&
1539 mtvd->vdev_ops != &vdev_missing_ops) {
1540
1541 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1542 continue;
1543
1544 /*
1545 * Device specific actions.
1546 */
1547 if (mtvd->vdev_islog) {
1548 spa_set_log_state(spa, SPA_LOG_CLEAR);
1549 } else {
1550 /*
1551 * XXX - once we have 'readonly' pool
1552 * support we should be able to handle
1553 * missing data devices by transitioning
1554 * the pool to readonly.
1555 */
1556 continue;
1557 }
1558
1559 /*
1560 * Swap the missing vdev with the data we were
1561 * able to obtain from the MOS config.
1562 */
1563 vdev_remove_child(rvd, tvd);
1564 vdev_remove_child(mrvd, mtvd);
1565
1566 vdev_add_child(rvd, mtvd);
1567 vdev_add_child(mrvd, tvd);
1568
1569 spa_config_exit(spa, SCL_ALL, FTAG);
1570 vdev_load(mtvd);
1571 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1572
1573 vdev_reopen(rvd);
1574 } else if (mtvd->vdev_islog) {
1575 /*
1576 * Load the slog device's state from the MOS config
1577 * since it's possible that the label does not
1578 * contain the most up-to-date information.
1579 */
1580 vdev_load_log_state(tvd, mtvd);
1581 vdev_reopen(tvd);
1582 }
1583 }
1584 vdev_free(mrvd);
1585 spa_config_exit(spa, SCL_ALL, FTAG);
1586
1587 /*
1588 * Ensure we were able to validate the config.
1589 */
1590 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1591}
1592
1593/*
1594 * Check for missing log devices
1595 */
1596static int
1597spa_check_logs(spa_t *spa)
1598{
1599 switch (spa->spa_log_state) {
1600 case SPA_LOG_MISSING:
1601 /* need to recheck in case slog has been restored */
1602 case SPA_LOG_UNKNOWN:
1603 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1604 DS_FIND_CHILDREN)) {
1605 spa_set_log_state(spa, SPA_LOG_MISSING);
1606 return (1);
1607 }
1608 break;
1609 }
1610 return (0);
1611}
1612
1613static boolean_t
1614spa_passivate_log(spa_t *spa)
1615{
1616 vdev_t *rvd = spa->spa_root_vdev;
1617 boolean_t slog_found = B_FALSE;
1618
1619 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1620
1621 if (!spa_has_slogs(spa))
1622 return (B_FALSE);
1623
1624 for (int c = 0; c < rvd->vdev_children; c++) {
1625 vdev_t *tvd = rvd->vdev_child[c];
1626 metaslab_group_t *mg = tvd->vdev_mg;
1627
1628 if (tvd->vdev_islog) {
1629 metaslab_group_passivate(mg);
1630 slog_found = B_TRUE;
1631 }
1632 }
1633
1634 return (slog_found);
1635}
1636
1637static void
1638spa_activate_log(spa_t *spa)
1639{
1640 vdev_t *rvd = spa->spa_root_vdev;
1641
1642 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1643
1644 for (int c = 0; c < rvd->vdev_children; c++) {
1645 vdev_t *tvd = rvd->vdev_child[c];
1646 metaslab_group_t *mg = tvd->vdev_mg;
1647
1648 if (tvd->vdev_islog)
1649 metaslab_group_activate(mg);
1650 }
1651}
1652
1653int
1654spa_offline_log(spa_t *spa)
1655{
1656 int error = 0;
1657
1658 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1659 NULL, DS_FIND_CHILDREN)) == 0) {
1660
1661 /*
1662 * We successfully offlined the log device, sync out the
1663 * current txg so that the "stubby" block can be removed
1664 * by zil_sync().
1665 */
1666 txg_wait_synced(spa->spa_dsl_pool, 0);
1667 }
1668 return (error);
1669}
1670
1671static void
1672spa_aux_check_removed(spa_aux_vdev_t *sav)
1673{
1674 int i;
1675
1676 for (i = 0; i < sav->sav_count; i++)
1677 spa_check_removed(sav->sav_vdevs[i]);
1678}
1679
1680void
1681spa_claim_notify(zio_t *zio)
1682{
1683 spa_t *spa = zio->io_spa;
1684
1685 if (zio->io_error)
1686 return;
1687
1688 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1689 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1690 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1691 mutex_exit(&spa->spa_props_lock);
1692}
1693
1694typedef struct spa_load_error {
1695 uint64_t sle_meta_count;
1696 uint64_t sle_data_count;
1697} spa_load_error_t;
1698
1699static void
1700spa_load_verify_done(zio_t *zio)
1701{
1702 blkptr_t *bp = zio->io_bp;
1703 spa_load_error_t *sle = zio->io_private;
1704 dmu_object_type_t type = BP_GET_TYPE(bp);
1705 int error = zio->io_error;
1706
1707 if (error) {
1708 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1709 type != DMU_OT_INTENT_LOG)
1710 atomic_add_64(&sle->sle_meta_count, 1);
1711 else
1712 atomic_add_64(&sle->sle_data_count, 1);
1713 }
1714 zio_data_buf_free(zio->io_data, zio->io_size);
1715}
1716
1717/*ARGSUSED*/
1718static int
1719spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1720 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1721{
1722 if (bp != NULL) {
1723 zio_t *rio = arg;
1724 size_t size = BP_GET_PSIZE(bp);
1725 void *data = zio_data_buf_alloc(size);
1726
1727 zio_nowait(zio_read(rio, spa, bp, data, size,
1728 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1729 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1730 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1731 }
1732 return (0);
1733}
1734
1735static int
1736spa_load_verify(spa_t *spa)
1737{
1738 zio_t *rio;
1739 spa_load_error_t sle = { 0 };
1740 zpool_rewind_policy_t policy;
1741 boolean_t verify_ok = B_FALSE;
1742 int error;
1743
1744 zpool_get_rewind_policy(spa->spa_config, &policy);
1745
1746 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1747 return (0);
1748
1749 rio = zio_root(spa, NULL, &sle,
1750 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1751
1752 error = traverse_pool(spa, spa->spa_verify_min_txg,
1753 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1754
1755 (void) zio_wait(rio);
1756
1757 spa->spa_load_meta_errors = sle.sle_meta_count;
1758 spa->spa_load_data_errors = sle.sle_data_count;
1759
1760 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1761 sle.sle_data_count <= policy.zrp_maxdata) {
1762 int64_t loss = 0;
1763
1764 verify_ok = B_TRUE;
1765 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1766 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1767
1768 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1769 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1770 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1771 VERIFY(nvlist_add_int64(spa->spa_load_info,
1772 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1773 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1774 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1775 } else {
1776 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1777 }
1778
1779 if (error) {
1780 if (error != ENXIO && error != EIO)
1781 error = EIO;
1782 return (error);
1783 }
1784
1785 return (verify_ok ? 0 : EIO);
1786}
1787
1788/*
1789 * Find a value in the pool props object.
1790 */
1791static void
1792spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1793{
1794 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1795 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1796}
1797
1798/*
1799 * Find a value in the pool directory object.
1800 */
1801static int
1802spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1803{
1804 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1805 name, sizeof (uint64_t), 1, val));
1806}
1807
1808static int
1809spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1810{
1811 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1812 return (err);
1813}
1814
1815/*
1816 * Fix up config after a partly-completed split. This is done with the
1817 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1818 * pool have that entry in their config, but only the splitting one contains
1819 * a list of all the guids of the vdevs that are being split off.
1820 *
1821 * This function determines what to do with that list: either rejoin
1822 * all the disks to the pool, or complete the splitting process. To attempt
1823 * the rejoin, each disk that is offlined is marked online again, and
1824 * we do a reopen() call. If the vdev label for every disk that was
1825 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1826 * then we call vdev_split() on each disk, and complete the split.
1827 *
1828 * Otherwise we leave the config alone, with all the vdevs in place in
1829 * the original pool.
1830 */
1831static void
1832spa_try_repair(spa_t *spa, nvlist_t *config)
1833{
1834 uint_t extracted;
1835 uint64_t *glist;
1836 uint_t i, gcount;
1837 nvlist_t *nvl;
1838 vdev_t **vd;
1839 boolean_t attempt_reopen;
1840
1841 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1842 return;
1843
1844 /* check that the config is complete */
1845 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1846 &glist, &gcount) != 0)
1847 return;
1848
1849 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1850
1851 /* attempt to online all the vdevs & validate */
1852 attempt_reopen = B_TRUE;
1853 for (i = 0; i < gcount; i++) {
1854 if (glist[i] == 0) /* vdev is hole */
1855 continue;
1856
1857 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1858 if (vd[i] == NULL) {
1859 /*
1860 * Don't bother attempting to reopen the disks;
1861 * just do the split.
1862 */
1863 attempt_reopen = B_FALSE;
1864 } else {
1865 /* attempt to re-online it */
1866 vd[i]->vdev_offline = B_FALSE;
1867 }
1868 }
1869
1870 if (attempt_reopen) {
1871 vdev_reopen(spa->spa_root_vdev);
1872
1873 /* check each device to see what state it's in */
1874 for (extracted = 0, i = 0; i < gcount; i++) {
1875 if (vd[i] != NULL &&
1876 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1877 break;
1878 ++extracted;
1879 }
1880 }
1881
1882 /*
1883 * If every disk has been moved to the new pool, or if we never
1884 * even attempted to look at them, then we split them off for
1885 * good.
1886 */
1887 if (!attempt_reopen || gcount == extracted) {
1888 for (i = 0; i < gcount; i++)
1889 if (vd[i] != NULL)
1890 vdev_split(vd[i]);
1891 vdev_reopen(spa->spa_root_vdev);
1892 }
1893
1894 kmem_free(vd, gcount * sizeof (vdev_t *));
1895}
1896
1897static int
1898spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1899 boolean_t mosconfig)
1900{
1901 nvlist_t *config = spa->spa_config;
1902 char *ereport = FM_EREPORT_ZFS_POOL;
1903 char *comment;
1904 int error;
1905 uint64_t pool_guid;
1906 nvlist_t *nvl;
1907
1908 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1909 return (EINVAL);
1910
1911 ASSERT(spa->spa_comment == NULL);
1912 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1913 spa->spa_comment = spa_strdup(comment);
1914
1915 /*
1916 * Versioning wasn't explicitly added to the label until later, so if
1917 * it's not present treat it as the initial version.
1918 */
1919 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1920 &spa->spa_ubsync.ub_version) != 0)
1921 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1922
1923 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1924 &spa->spa_config_txg);
1925
1926 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1927 spa_guid_exists(pool_guid, 0)) {
1928 error = EEXIST;
1929 } else {
1930 spa->spa_config_guid = pool_guid;
1931
1932 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1933 &nvl) == 0) {
1934 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1935 KM_SLEEP) == 0);
1936 }
1937
1938 nvlist_free(spa->spa_load_info);
1939 spa->spa_load_info = fnvlist_alloc();
1940
1941 gethrestime(&spa->spa_loaded_ts);
1942 error = spa_load_impl(spa, pool_guid, config, state, type,
1943 mosconfig, &ereport);
1944 }
1945
1946 spa->spa_minref = refcount_count(&spa->spa_refcount);
1947 if (error) {
1948 if (error != EEXIST) {
1949 spa->spa_loaded_ts.tv_sec = 0;
1950 spa->spa_loaded_ts.tv_nsec = 0;
1951 }
1952 if (error != EBADF) {
1953 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1954 }
1955 }
1956 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1957 spa->spa_ena = 0;
1958
1959 return (error);
1960}
1961
1962/*
1963 * Load an existing storage pool, using the pool's builtin spa_config as a
1964 * source of configuration information.
1965 */
1966static int
1967spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1968 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1969 char **ereport)
1970{
1971 int error = 0;
1972 nvlist_t *nvroot = NULL;
1973 nvlist_t *label;
1974 vdev_t *rvd;
1975 uberblock_t *ub = &spa->spa_uberblock;
1976 uint64_t children, config_cache_txg = spa->spa_config_txg;
1977 int orig_mode = spa->spa_mode;
1978 int parse;
1979 uint64_t obj;
1980 boolean_t missing_feat_write = B_FALSE;
1981
1982 /*
1983 * If this is an untrusted config, access the pool in read-only mode.
1984 * This prevents things like resilvering recently removed devices.
1985 */
1986 if (!mosconfig)
1987 spa->spa_mode = FREAD;
1988
1989 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1990
1991 spa->spa_load_state = state;
1992
1993 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1994 return (EINVAL);
1995
1996 parse = (type == SPA_IMPORT_EXISTING ?
1997 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1998
1999 /*
2000 * Create "The Godfather" zio to hold all async IOs
2001 */
2002 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2003 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2004
2005 /*
2006 * Parse the configuration into a vdev tree. We explicitly set the
2007 * value that will be returned by spa_version() since parsing the
2008 * configuration requires knowing the version number.
2009 */
2010 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2011 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2012 spa_config_exit(spa, SCL_ALL, FTAG);
2013
2014 if (error != 0)
2015 return (error);
2016
2017 ASSERT(spa->spa_root_vdev == rvd);
2018
2019 if (type != SPA_IMPORT_ASSEMBLE) {
2020 ASSERT(spa_guid(spa) == pool_guid);
2021 }
2022
2023 /*
2024 * Try to open all vdevs, loading each label in the process.
2025 */
2026 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2027 error = vdev_open(rvd);
2028 spa_config_exit(spa, SCL_ALL, FTAG);
2029 if (error != 0)
2030 return (error);
2031
2032 /*
2033 * We need to validate the vdev labels against the configuration that
2034 * we have in hand, which is dependent on the setting of mosconfig. If
2035 * mosconfig is true then we're validating the vdev labels based on
2036 * that config. Otherwise, we're validating against the cached config
2037 * (zpool.cache) that was read when we loaded the zfs module, and then
2038 * later we will recursively call spa_load() and validate against
2039 * the vdev config.
2040 *
2041 * If we're assembling a new pool that's been split off from an
2042 * existing pool, the labels haven't yet been updated so we skip
2043 * validation for now.
2044 */
2045 if (type != SPA_IMPORT_ASSEMBLE) {
2046 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2047 error = vdev_validate(rvd, mosconfig);
2048 spa_config_exit(spa, SCL_ALL, FTAG);
2049
2050 if (error != 0)
2051 return (error);
2052
2053 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2054 return (ENXIO);
2055 }
2056
2057 /*
2058 * Find the best uberblock.
2059 */
2060 vdev_uberblock_load(rvd, ub, &label);
2061
2062 /*
2063 * If we weren't able to find a single valid uberblock, return failure.
2064 */
2065 if (ub->ub_txg == 0) {
2066 nvlist_free(label);
2067 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2068 }
2069
2070 /*
2071 * If the pool has an unsupported version we can't open it.
2072 */
2073 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2074 nvlist_free(label);
2075 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2076 }
2077
2078 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2079 nvlist_t *features;
2080
2081 /*
2082 * If we weren't able to find what's necessary for reading the
2083 * MOS in the label, return failure.
2084 */
2085 if (label == NULL || nvlist_lookup_nvlist(label,
2086 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2087 nvlist_free(label);
2088 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2089 ENXIO));
2090 }
2091
2092 /*
2093 * Update our in-core representation with the definitive values
2094 * from the label.
2095 */
2096 nvlist_free(spa->spa_label_features);
2097 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2098 }
2099
2100 nvlist_free(label);
2101
2102 /*
2103 * Look through entries in the label nvlist's features_for_read. If
2104 * there is a feature listed there which we don't understand then we
2105 * cannot open a pool.
2106 */
2107 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2108 nvlist_t *unsup_feat;
2109
2110 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2111 0);
2112
2113 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2114 NULL); nvp != NULL;
2115 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2116 if (!zfeature_is_supported(nvpair_name(nvp))) {
2117 VERIFY(nvlist_add_string(unsup_feat,
2118 nvpair_name(nvp), "") == 0);
2119 }
2120 }
2121
2122 if (!nvlist_empty(unsup_feat)) {
2123 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2124 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2125 nvlist_free(unsup_feat);
2126 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2127 ENOTSUP));
2128 }
2129
2130 nvlist_free(unsup_feat);
2131 }
2132
2133 /*
2134 * If the vdev guid sum doesn't match the uberblock, we have an
2135 * incomplete configuration. We first check to see if the pool
2136 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2137 * If it is, defer the vdev_guid_sum check till later so we
2138 * can handle missing vdevs.
2139 */
2140 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2141 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2142 rvd->vdev_guid_sum != ub->ub_guid_sum)
2143 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2144
2145 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2146 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2147 spa_try_repair(spa, config);
2148 spa_config_exit(spa, SCL_ALL, FTAG);
2149 nvlist_free(spa->spa_config_splitting);
2150 spa->spa_config_splitting = NULL;
2151 }
2152
2153 /*
2154 * Initialize internal SPA structures.
2155 */
2156 spa->spa_state = POOL_STATE_ACTIVE;
2157 spa->spa_ubsync = spa->spa_uberblock;
2158 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2159 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2160 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2161 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2162 spa->spa_claim_max_txg = spa->spa_first_txg;
2163 spa->spa_prev_software_version = ub->ub_software_version;
2164
2165 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2166 if (error)
2167 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2168 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2169
2170 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2171 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2172
2173 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2174 boolean_t missing_feat_read = B_FALSE;
|
2213
2214 if (!missing_feat_read) {
2215 fnvlist_add_boolean(spa->spa_load_info,
2216 ZPOOL_CONFIG_CAN_RDONLY);
2217 }
2218
2219 /*
2220 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2221 * twofold: to determine whether the pool is available for
2222 * import in read-write mode and (if it is not) whether the
2223 * pool is available for import in read-only mode. If the pool
2224 * is available for import in read-write mode, it is displayed
2225 * as available in userland; if it is not available for import
2226 * in read-only mode, it is displayed as unavailable in
2227 * userland. If the pool is available for import in read-only
2228 * mode but not read-write mode, it is displayed as unavailable
2229 * in userland with a special note that the pool is actually
2230 * available for open in read-only mode.
2231 *
2232 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2233 * missing a feature for write, we must first determine whether
2234 * the pool can be opened read-only before returning to
2235 * userland in order to know whether to display the
2236 * abovementioned note.
2237 */
2238 if (missing_feat_read || (missing_feat_write &&
2239 spa_writeable(spa))) {
2240 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2241 ENOTSUP));
2242 }
2243 }
2244
2245 spa->spa_is_initializing = B_TRUE;
2246 error = dsl_pool_open(spa->spa_dsl_pool);
2247 spa->spa_is_initializing = B_FALSE;
2248 if (error != 0)
2249 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2250
2251 if (!mosconfig) {
2252 uint64_t hostid;
2253 nvlist_t *policy = NULL, *nvconfig;
2254
2255 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2256 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2257
2258 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2259 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2260 char *hostname;
2261 unsigned long myhostid = 0;
2262
2263 VERIFY(nvlist_lookup_string(nvconfig,
2264 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2265
2266#ifdef _KERNEL
2267 myhostid = zone_get_hostid(NULL);
2268#else /* _KERNEL */
2269 /*
2270 * We're emulating the system's hostid in userland, so
2271 * we can't use zone_get_hostid().
2272 */
2273 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2274#endif /* _KERNEL */
2275 if (check_hostid && hostid != 0 && myhostid != 0 &&
2276 hostid != myhostid) {
2277 nvlist_free(nvconfig);
2278 cmn_err(CE_WARN, "pool '%s' could not be "
2279 "loaded as it was last accessed by "
2280 "another system (host: %s hostid: 0x%lx). "
2281 "See: http://illumos.org/msg/ZFS-8000-EY",
2282 spa_name(spa), hostname,
2283 (unsigned long)hostid);
2284 return (EBADF);
2285 }
2286 }
2287 if (nvlist_lookup_nvlist(spa->spa_config,
2288 ZPOOL_REWIND_POLICY, &policy) == 0)
2289 VERIFY(nvlist_add_nvlist(nvconfig,
2290 ZPOOL_REWIND_POLICY, policy) == 0);
2291
2292 spa_config_set(spa, nvconfig);
2293 spa_unload(spa);
2294 spa_deactivate(spa);
2295 spa_activate(spa, orig_mode);
2296
2297 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2298 }
2299
2300 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2301 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2302 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2303 if (error != 0)
2304 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2305
2306 /*
2307 * Load the bit that tells us to use the new accounting function
2308 * (raid-z deflation). If we have an older pool, this will not
2309 * be present.
2310 */
2311 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2312 if (error != 0 && error != ENOENT)
2313 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2314
2315 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2316 &spa->spa_creation_version);
2317 if (error != 0 && error != ENOENT)
2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2319
2320 /*
2321 * Load the persistent error log. If we have an older pool, this will
2322 * not be present.
2323 */
2324 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2325 if (error != 0 && error != ENOENT)
2326 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2327
2328 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2329 &spa->spa_errlog_scrub);
2330 if (error != 0 && error != ENOENT)
2331 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2332
2333 /*
2334 * Load the history object. If we have an older pool, this
2335 * will not be present.
2336 */
2337 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2338 if (error != 0 && error != ENOENT)
2339 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340
2341 /*
2342 * If we're assembling the pool from the split-off vdevs of
2343 * an existing pool, we don't want to attach the spares & cache
2344 * devices.
2345 */
2346
2347 /*
2348 * Load any hot spares for this pool.
2349 */
2350 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2351 if (error != 0 && error != ENOENT)
2352 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2353 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2354 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2355 if (load_nvlist(spa, spa->spa_spares.sav_object,
2356 &spa->spa_spares.sav_config) != 0)
2357 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2358
2359 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2360 spa_load_spares(spa);
2361 spa_config_exit(spa, SCL_ALL, FTAG);
2362 } else if (error == 0) {
2363 spa->spa_spares.sav_sync = B_TRUE;
2364 }
2365
2366 /*
2367 * Load any level 2 ARC devices for this pool.
2368 */
2369 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2370 &spa->spa_l2cache.sav_object);
2371 if (error != 0 && error != ENOENT)
2372 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2373 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2374 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2375 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2376 &spa->spa_l2cache.sav_config) != 0)
2377 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2378
2379 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2380 spa_load_l2cache(spa);
2381 spa_config_exit(spa, SCL_ALL, FTAG);
2382 } else if (error == 0) {
2383 spa->spa_l2cache.sav_sync = B_TRUE;
2384 }
2385
2386 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2387
2388 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2389 if (error && error != ENOENT)
2390 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2391
2392 if (error == 0) {
2393 uint64_t autoreplace;
2394
2395 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2396 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2397 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2398 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2399 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2400 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2401 &spa->spa_dedup_ditto);
2402
2403 spa->spa_autoreplace = (autoreplace != 0);
2404 }
2405
2406 /*
2407 * If the 'autoreplace' property is set, then post a resource notifying
2408 * the ZFS DE that it should not issue any faults for unopenable
2409 * devices. We also iterate over the vdevs, and post a sysevent for any
2410 * unopenable vdevs so that the normal autoreplace handler can take
2411 * over.
2412 */
2413 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2414 spa_check_removed(spa->spa_root_vdev);
2415 /*
2416 * For the import case, this is done in spa_import(), because
2417 * at this point we're using the spare definitions from
2418 * the MOS config, not necessarily from the userland config.
2419 */
2420 if (state != SPA_LOAD_IMPORT) {
2421 spa_aux_check_removed(&spa->spa_spares);
2422 spa_aux_check_removed(&spa->spa_l2cache);
2423 }
2424 }
2425
2426 /*
2427 * Load the vdev state for all toplevel vdevs.
2428 */
2429 vdev_load(rvd);
2430
2431 /*
2432 * Propagate the leaf DTLs we just loaded all the way up the tree.
2433 */
2434 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2435 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2436 spa_config_exit(spa, SCL_ALL, FTAG);
2437
2438 /*
2439 * Load the DDTs (dedup tables).
2440 */
2441 error = ddt_load(spa);
2442 if (error != 0)
2443 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2444
2445 spa_update_dspace(spa);
2446
2447 /*
2448 * Validate the config, using the MOS config to fill in any
2449 * information which might be missing. If we fail to validate
2450 * the config then declare the pool unfit for use. If we're
2451 * assembling a pool from a split, the log is not transferred
2452 * over.
2453 */
2454 if (type != SPA_IMPORT_ASSEMBLE) {
2455 nvlist_t *nvconfig;
2456
2457 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2459
2460 if (!spa_config_valid(spa, nvconfig)) {
2461 nvlist_free(nvconfig);
2462 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2463 ENXIO));
2464 }
2465 nvlist_free(nvconfig);
2466
2467 /*
2468 * Now that we've validated the config, check the state of the
2469 * root vdev. If it can't be opened, it indicates one or
2470 * more toplevel vdevs are faulted.
2471 */
2472 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2473 return (ENXIO);
2474
2475 if (spa_check_logs(spa)) {
2476 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2477 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2478 }
2479 }
2480
2481 if (missing_feat_write) {
2482 ASSERT(state == SPA_LOAD_TRYIMPORT);
2483
2484 /*
2485 * At this point, we know that we can open the pool in
2486 * read-only mode but not read-write mode. We now have enough
2487 * information and can return to userland.
2488 */
2489 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2490 }
2491
2492 /*
2493 * We've successfully opened the pool, verify that we're ready
2494 * to start pushing transactions.
2495 */
2496 if (state != SPA_LOAD_TRYIMPORT) {
2497 if (error = spa_load_verify(spa))
2498 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2499 error));
2500 }
2501
2502 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2503 spa->spa_load_max_txg == UINT64_MAX)) {
2504 dmu_tx_t *tx;
2505 int need_update = B_FALSE;
2506
2507 ASSERT(state != SPA_LOAD_TRYIMPORT);
2508
2509 /*
2510 * Claim log blocks that haven't been committed yet.
2511 * This must all happen in a single txg.
2512 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2513 * invoked from zil_claim_log_block()'s i/o done callback.
2514 * Price of rollback is that we abandon the log.
2515 */
2516 spa->spa_claiming = B_TRUE;
2517
2518 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2519 spa_first_txg(spa));
2520 (void) dmu_objset_find(spa_name(spa),
2521 zil_claim, tx, DS_FIND_CHILDREN);
2522 dmu_tx_commit(tx);
2523
2524 spa->spa_claiming = B_FALSE;
2525
2526 spa_set_log_state(spa, SPA_LOG_GOOD);
2527 spa->spa_sync_on = B_TRUE;
2528 txg_sync_start(spa->spa_dsl_pool);
2529
2530 /*
2531 * Wait for all claims to sync. We sync up to the highest
2532 * claimed log block birth time so that claimed log blocks
2533 * don't appear to be from the future. spa_claim_max_txg
2534 * will have been set for us by either zil_check_log_chain()
2535 * (invoked from spa_check_logs()) or zil_claim() above.
2536 */
2537 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2538
2539 /*
2540 * If the config cache is stale, or we have uninitialized
2541 * metaslabs (see spa_vdev_add()), then update the config.
2542 *
2543 * If this is a verbatim import, trust the current
2544 * in-core spa_config and update the disk labels.
2545 */
2546 if (config_cache_txg != spa->spa_config_txg ||
2547 state == SPA_LOAD_IMPORT ||
2548 state == SPA_LOAD_RECOVER ||
2549 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2550 need_update = B_TRUE;
2551
2552 for (int c = 0; c < rvd->vdev_children; c++)
2553 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2554 need_update = B_TRUE;
2555
2556 /*
2557 * Update the config cache asychronously in case we're the
2558 * root pool, in which case the config cache isn't writable yet.
2559 */
2560 if (need_update)
2561 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2562
2563 /*
2564 * Check all DTLs to see if anything needs resilvering.
2565 */
2566 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2567 vdev_resilver_needed(rvd, NULL, NULL))
2568 spa_async_request(spa, SPA_ASYNC_RESILVER);
2569
2570 /*
2571 * Delete any inconsistent datasets.
2572 */
2573 (void) dmu_objset_find(spa_name(spa),
2574 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2575
2576 /*
2577 * Clean up any stale temporary dataset userrefs.
2578 */
2579 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2580 }
2581
2582 return (0);
2583}
2584
2585static int
2586spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2587{
2588 int mode = spa->spa_mode;
2589
2590 spa_unload(spa);
2591 spa_deactivate(spa);
2592
2593 spa->spa_load_max_txg--;
2594
2595 spa_activate(spa, mode);
2596 spa_async_suspend(spa);
2597
2598 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2599}
2600
2601/*
2602 * If spa_load() fails this function will try loading prior txg's. If
2603 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2604 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2605 * function will not rewind the pool and will return the same error as
2606 * spa_load().
2607 */
2608static int
2609spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2610 uint64_t max_request, int rewind_flags)
2611{
2612 nvlist_t *loadinfo = NULL;
2613 nvlist_t *config = NULL;
2614 int load_error, rewind_error;
2615 uint64_t safe_rewind_txg;
2616 uint64_t min_txg;
2617
2618 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2619 spa->spa_load_max_txg = spa->spa_load_txg;
2620 spa_set_log_state(spa, SPA_LOG_CLEAR);
2621 } else {
2622 spa->spa_load_max_txg = max_request;
2623 }
2624
2625 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2626 mosconfig);
2627 if (load_error == 0)
2628 return (0);
2629
2630 if (spa->spa_root_vdev != NULL)
2631 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2632
2633 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2634 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2635
2636 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2637 nvlist_free(config);
2638 return (load_error);
2639 }
2640
2641 if (state == SPA_LOAD_RECOVER) {
2642 /* Price of rolling back is discarding txgs, including log */
2643 spa_set_log_state(spa, SPA_LOG_CLEAR);
2644 } else {
2645 /*
2646 * If we aren't rolling back save the load info from our first
2647 * import attempt so that we can restore it after attempting
2648 * to rewind.
2649 */
2650 loadinfo = spa->spa_load_info;
2651 spa->spa_load_info = fnvlist_alloc();
2652 }
2653
2654 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2655 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2656 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2657 TXG_INITIAL : safe_rewind_txg;
2658
2659 /*
2660 * Continue as long as we're finding errors, we're still within
2661 * the acceptable rewind range, and we're still finding uberblocks
2662 */
2663 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2664 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2665 if (spa->spa_load_max_txg < safe_rewind_txg)
2666 spa->spa_extreme_rewind = B_TRUE;
2667 rewind_error = spa_load_retry(spa, state, mosconfig);
2668 }
2669
2670 spa->spa_extreme_rewind = B_FALSE;
2671 spa->spa_load_max_txg = UINT64_MAX;
2672
2673 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2674 spa_config_set(spa, config);
2675
2676 if (state == SPA_LOAD_RECOVER) {
2677 ASSERT3P(loadinfo, ==, NULL);
2678 return (rewind_error);
2679 } else {
2680 /* Store the rewind info as part of the initial load info */
2681 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2682 spa->spa_load_info);
2683
2684 /* Restore the initial load info */
2685 fnvlist_free(spa->spa_load_info);
2686 spa->spa_load_info = loadinfo;
2687
2688 return (load_error);
2689 }
2690}
2691
2692/*
2693 * Pool Open/Import
2694 *
2695 * The import case is identical to an open except that the configuration is sent
2696 * down from userland, instead of grabbed from the configuration cache. For the
2697 * case of an open, the pool configuration will exist in the
2698 * POOL_STATE_UNINITIALIZED state.
2699 *
2700 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2701 * the same time open the pool, without having to keep around the spa_t in some
2702 * ambiguous state.
2703 */
2704static int
2705spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2706 nvlist_t **config)
2707{
2708 spa_t *spa;
2709 spa_load_state_t state = SPA_LOAD_OPEN;
2710 int error;
2711 int locked = B_FALSE;
2712 int firstopen = B_FALSE;
2713
2714 *spapp = NULL;
2715
2716 /*
2717 * As disgusting as this is, we need to support recursive calls to this
2718 * function because dsl_dir_open() is called during spa_load(), and ends
2719 * up calling spa_open() again. The real fix is to figure out how to
2720 * avoid dsl_dir_open() calling this in the first place.
2721 */
2722 if (mutex_owner(&spa_namespace_lock) != curthread) {
2723 mutex_enter(&spa_namespace_lock);
2724 locked = B_TRUE;
2725 }
2726
2727 if ((spa = spa_lookup(pool)) == NULL) {
2728 if (locked)
2729 mutex_exit(&spa_namespace_lock);
2730 return (ENOENT);
2731 }
2732
2733 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2734 zpool_rewind_policy_t policy;
2735
2736 firstopen = B_TRUE;
2737
2738 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2739 &policy);
2740 if (policy.zrp_request & ZPOOL_DO_REWIND)
2741 state = SPA_LOAD_RECOVER;
2742
2743 spa_activate(spa, spa_mode_global);
2744
2745 if (state != SPA_LOAD_RECOVER)
2746 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2747
2748 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2749 policy.zrp_request);
2750
2751 if (error == EBADF) {
2752 /*
2753 * If vdev_validate() returns failure (indicated by
2754 * EBADF), it indicates that one of the vdevs indicates
2755 * that the pool has been exported or destroyed. If
2756 * this is the case, the config cache is out of sync and
2757 * we should remove the pool from the namespace.
2758 */
2759 spa_unload(spa);
2760 spa_deactivate(spa);
2761 spa_config_sync(spa, B_TRUE, B_TRUE);
2762 spa_remove(spa);
2763 if (locked)
2764 mutex_exit(&spa_namespace_lock);
2765 return (ENOENT);
2766 }
2767
2768 if (error) {
2769 /*
2770 * We can't open the pool, but we still have useful
2771 * information: the state of each vdev after the
2772 * attempted vdev_open(). Return this to the user.
2773 */
2774 if (config != NULL && spa->spa_config) {
2775 VERIFY(nvlist_dup(spa->spa_config, config,
2776 KM_SLEEP) == 0);
2777 VERIFY(nvlist_add_nvlist(*config,
2778 ZPOOL_CONFIG_LOAD_INFO,
2779 spa->spa_load_info) == 0);
2780 }
2781 spa_unload(spa);
2782 spa_deactivate(spa);
2783 spa->spa_last_open_failed = error;
2784 if (locked)
2785 mutex_exit(&spa_namespace_lock);
2786 *spapp = NULL;
2787 return (error);
2788 }
2789 }
2790
2791 spa_open_ref(spa, tag);
2792
2793 if (config != NULL)
2794 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2795
2796 /*
2797 * If we've recovered the pool, pass back any information we
2798 * gathered while doing the load.
2799 */
2800 if (state == SPA_LOAD_RECOVER) {
2801 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2802 spa->spa_load_info) == 0);
2803 }
2804
2805 if (locked) {
2806 spa->spa_last_open_failed = 0;
2807 spa->spa_last_ubsync_txg = 0;
2808 spa->spa_load_txg = 0;
2809 mutex_exit(&spa_namespace_lock);
2810#ifdef __FreeBSD__
2811#ifdef _KERNEL
2812 if (firstopen)
2813 zvol_create_minors(pool);
2814#endif
2815#endif
2816 }
2817
2818 *spapp = spa;
2819
2820 return (0);
2821}
2822
2823int
2824spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2825 nvlist_t **config)
2826{
2827 return (spa_open_common(name, spapp, tag, policy, config));
2828}
2829
2830int
2831spa_open(const char *name, spa_t **spapp, void *tag)
2832{
2833 return (spa_open_common(name, spapp, tag, NULL, NULL));
2834}
2835
2836/*
2837 * Lookup the given spa_t, incrementing the inject count in the process,
2838 * preventing it from being exported or destroyed.
2839 */
2840spa_t *
2841spa_inject_addref(char *name)
2842{
2843 spa_t *spa;
2844
2845 mutex_enter(&spa_namespace_lock);
2846 if ((spa = spa_lookup(name)) == NULL) {
2847 mutex_exit(&spa_namespace_lock);
2848 return (NULL);
2849 }
2850 spa->spa_inject_ref++;
2851 mutex_exit(&spa_namespace_lock);
2852
2853 return (spa);
2854}
2855
2856void
2857spa_inject_delref(spa_t *spa)
2858{
2859 mutex_enter(&spa_namespace_lock);
2860 spa->spa_inject_ref--;
2861 mutex_exit(&spa_namespace_lock);
2862}
2863
2864/*
2865 * Add spares device information to the nvlist.
2866 */
2867static void
2868spa_add_spares(spa_t *spa, nvlist_t *config)
2869{
2870 nvlist_t **spares;
2871 uint_t i, nspares;
2872 nvlist_t *nvroot;
2873 uint64_t guid;
2874 vdev_stat_t *vs;
2875 uint_t vsc;
2876 uint64_t pool;
2877
2878 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2879
2880 if (spa->spa_spares.sav_count == 0)
2881 return;
2882
2883 VERIFY(nvlist_lookup_nvlist(config,
2884 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2885 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2886 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2887 if (nspares != 0) {
2888 VERIFY(nvlist_add_nvlist_array(nvroot,
2889 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2890 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2891 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2892
2893 /*
2894 * Go through and find any spares which have since been
2895 * repurposed as an active spare. If this is the case, update
2896 * their status appropriately.
2897 */
2898 for (i = 0; i < nspares; i++) {
2899 VERIFY(nvlist_lookup_uint64(spares[i],
2900 ZPOOL_CONFIG_GUID, &guid) == 0);
2901 if (spa_spare_exists(guid, &pool, NULL) &&
2902 pool != 0ULL) {
2903 VERIFY(nvlist_lookup_uint64_array(
2904 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2905 (uint64_t **)&vs, &vsc) == 0);
2906 vs->vs_state = VDEV_STATE_CANT_OPEN;
2907 vs->vs_aux = VDEV_AUX_SPARED;
2908 }
2909 }
2910 }
2911}
2912
2913/*
2914 * Add l2cache device information to the nvlist, including vdev stats.
2915 */
2916static void
2917spa_add_l2cache(spa_t *spa, nvlist_t *config)
2918{
2919 nvlist_t **l2cache;
2920 uint_t i, j, nl2cache;
2921 nvlist_t *nvroot;
2922 uint64_t guid;
2923 vdev_t *vd;
2924 vdev_stat_t *vs;
2925 uint_t vsc;
2926
2927 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2928
2929 if (spa->spa_l2cache.sav_count == 0)
2930 return;
2931
2932 VERIFY(nvlist_lookup_nvlist(config,
2933 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2934 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2935 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2936 if (nl2cache != 0) {
2937 VERIFY(nvlist_add_nvlist_array(nvroot,
2938 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2939 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2940 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2941
2942 /*
2943 * Update level 2 cache device stats.
2944 */
2945
2946 for (i = 0; i < nl2cache; i++) {
2947 VERIFY(nvlist_lookup_uint64(l2cache[i],
2948 ZPOOL_CONFIG_GUID, &guid) == 0);
2949
2950 vd = NULL;
2951 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2952 if (guid ==
2953 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2954 vd = spa->spa_l2cache.sav_vdevs[j];
2955 break;
2956 }
2957 }
2958 ASSERT(vd != NULL);
2959
2960 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2961 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2962 == 0);
2963 vdev_get_stats(vd, vs);
2964 }
2965 }
2966}
2967
2968static void
2969spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2970{
2971 nvlist_t *features;
2972 zap_cursor_t zc;
2973 zap_attribute_t za;
2974
2975 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2976 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2977
2978 if (spa->spa_feat_for_read_obj != 0) {
2979 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2980 spa->spa_feat_for_read_obj);
2981 zap_cursor_retrieve(&zc, &za) == 0;
2982 zap_cursor_advance(&zc)) {
2983 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2984 za.za_num_integers == 1);
2985 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2986 za.za_first_integer));
2987 }
2988 zap_cursor_fini(&zc);
2989 }
2990
2991 if (spa->spa_feat_for_write_obj != 0) {
2992 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2993 spa->spa_feat_for_write_obj);
2994 zap_cursor_retrieve(&zc, &za) == 0;
2995 zap_cursor_advance(&zc)) {
2996 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2997 za.za_num_integers == 1);
2998 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2999 za.za_first_integer));
3000 }
3001 zap_cursor_fini(&zc);
3002 }
3003
3004 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3005 features) == 0);
3006 nvlist_free(features);
3007}
3008
3009int
3010spa_get_stats(const char *name, nvlist_t **config,
3011 char *altroot, size_t buflen)
3012{
3013 int error;
3014 spa_t *spa;
3015
3016 *config = NULL;
3017 error = spa_open_common(name, &spa, FTAG, NULL, config);
3018
3019 if (spa != NULL) {
3020 /*
3021 * This still leaves a window of inconsistency where the spares
3022 * or l2cache devices could change and the config would be
3023 * self-inconsistent.
3024 */
3025 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3026
3027 if (*config != NULL) {
3028 uint64_t loadtimes[2];
3029
3030 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3031 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3032 VERIFY(nvlist_add_uint64_array(*config,
3033 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3034
3035 VERIFY(nvlist_add_uint64(*config,
3036 ZPOOL_CONFIG_ERRCOUNT,
3037 spa_get_errlog_size(spa)) == 0);
3038
3039 if (spa_suspended(spa))
3040 VERIFY(nvlist_add_uint64(*config,
3041 ZPOOL_CONFIG_SUSPENDED,
3042 spa->spa_failmode) == 0);
3043
3044 spa_add_spares(spa, *config);
3045 spa_add_l2cache(spa, *config);
3046 spa_add_feature_stats(spa, *config);
3047 }
3048 }
3049
3050 /*
3051 * We want to get the alternate root even for faulted pools, so we cheat
3052 * and call spa_lookup() directly.
3053 */
3054 if (altroot) {
3055 if (spa == NULL) {
3056 mutex_enter(&spa_namespace_lock);
3057 spa = spa_lookup(name);
3058 if (spa)
3059 spa_altroot(spa, altroot, buflen);
3060 else
3061 altroot[0] = '\0';
3062 spa = NULL;
3063 mutex_exit(&spa_namespace_lock);
3064 } else {
3065 spa_altroot(spa, altroot, buflen);
3066 }
3067 }
3068
3069 if (spa != NULL) {
3070 spa_config_exit(spa, SCL_CONFIG, FTAG);
3071 spa_close(spa, FTAG);
3072 }
3073
3074 return (error);
3075}
3076
3077/*
3078 * Validate that the auxiliary device array is well formed. We must have an
3079 * array of nvlists, each which describes a valid leaf vdev. If this is an
3080 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3081 * specified, as long as they are well-formed.
3082 */
3083static int
3084spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3085 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3086 vdev_labeltype_t label)
3087{
3088 nvlist_t **dev;
3089 uint_t i, ndev;
3090 vdev_t *vd;
3091 int error;
3092
3093 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3094
3095 /*
3096 * It's acceptable to have no devs specified.
3097 */
3098 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3099 return (0);
3100
3101 if (ndev == 0)
3102 return (EINVAL);
3103
3104 /*
3105 * Make sure the pool is formatted with a version that supports this
3106 * device type.
3107 */
3108 if (spa_version(spa) < version)
3109 return (ENOTSUP);
3110
3111 /*
3112 * Set the pending device list so we correctly handle device in-use
3113 * checking.
3114 */
3115 sav->sav_pending = dev;
3116 sav->sav_npending = ndev;
3117
3118 for (i = 0; i < ndev; i++) {
3119 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3120 mode)) != 0)
3121 goto out;
3122
3123 if (!vd->vdev_ops->vdev_op_leaf) {
3124 vdev_free(vd);
3125 error = EINVAL;
3126 goto out;
3127 }
3128
3129 /*
3130 * The L2ARC currently only supports disk devices in
3131 * kernel context. For user-level testing, we allow it.
3132 */
3133#ifdef _KERNEL
3134 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3135 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3136 error = ENOTBLK;
3137 vdev_free(vd);
3138 goto out;
3139 }
3140#endif
3141 vd->vdev_top = vd;
3142
3143 if ((error = vdev_open(vd)) == 0 &&
3144 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3145 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3146 vd->vdev_guid) == 0);
3147 }
3148
3149 vdev_free(vd);
3150
3151 if (error &&
3152 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3153 goto out;
3154 else
3155 error = 0;
3156 }
3157
3158out:
3159 sav->sav_pending = NULL;
3160 sav->sav_npending = 0;
3161 return (error);
3162}
3163
3164static int
3165spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3166{
3167 int error;
3168
3169 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3170
3171 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3172 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3173 VDEV_LABEL_SPARE)) != 0) {
3174 return (error);
3175 }
3176
3177 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3178 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3179 VDEV_LABEL_L2CACHE));
3180}
3181
3182static void
3183spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3184 const char *config)
3185{
3186 int i;
3187
3188 if (sav->sav_config != NULL) {
3189 nvlist_t **olddevs;
3190 uint_t oldndevs;
3191 nvlist_t **newdevs;
3192
3193 /*
3194 * Generate new dev list by concatentating with the
3195 * current dev list.
3196 */
3197 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3198 &olddevs, &oldndevs) == 0);
3199
3200 newdevs = kmem_alloc(sizeof (void *) *
3201 (ndevs + oldndevs), KM_SLEEP);
3202 for (i = 0; i < oldndevs; i++)
3203 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3204 KM_SLEEP) == 0);
3205 for (i = 0; i < ndevs; i++)
3206 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3207 KM_SLEEP) == 0);
3208
3209 VERIFY(nvlist_remove(sav->sav_config, config,
3210 DATA_TYPE_NVLIST_ARRAY) == 0);
3211
3212 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3213 config, newdevs, ndevs + oldndevs) == 0);
3214 for (i = 0; i < oldndevs + ndevs; i++)
3215 nvlist_free(newdevs[i]);
3216 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3217 } else {
3218 /*
3219 * Generate a new dev list.
3220 */
3221 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3222 KM_SLEEP) == 0);
3223 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3224 devs, ndevs) == 0);
3225 }
3226}
3227
3228/*
3229 * Stop and drop level 2 ARC devices
3230 */
3231void
3232spa_l2cache_drop(spa_t *spa)
3233{
3234 vdev_t *vd;
3235 int i;
3236 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3237
3238 for (i = 0; i < sav->sav_count; i++) {
3239 uint64_t pool;
3240
3241 vd = sav->sav_vdevs[i];
3242 ASSERT(vd != NULL);
3243
3244 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3245 pool != 0ULL && l2arc_vdev_present(vd))
3246 l2arc_remove_vdev(vd);
3247 }
3248}
3249
3250/*
3251 * Pool Creation
3252 */
3253int
3254spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3255 const char *history_str, nvlist_t *zplprops)
3256{
3257 spa_t *spa;
3258 char *altroot = NULL;
3259 vdev_t *rvd;
3260 dsl_pool_t *dp;
3261 dmu_tx_t *tx;
3262 int error = 0;
3263 uint64_t txg = TXG_INITIAL;
3264 nvlist_t **spares, **l2cache;
3265 uint_t nspares, nl2cache;
3266 uint64_t version, obj;
3267 boolean_t has_features;
3268
3269 /*
3270 * If this pool already exists, return failure.
3271 */
3272 mutex_enter(&spa_namespace_lock);
3273 if (spa_lookup(pool) != NULL) {
3274 mutex_exit(&spa_namespace_lock);
3275 return (EEXIST);
3276 }
3277
3278 /*
3279 * Allocate a new spa_t structure.
3280 */
3281 (void) nvlist_lookup_string(props,
3282 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3283 spa = spa_add(pool, NULL, altroot);
3284 spa_activate(spa, spa_mode_global);
3285
3286 if (props && (error = spa_prop_validate(spa, props))) {
3287 spa_deactivate(spa);
3288 spa_remove(spa);
3289 mutex_exit(&spa_namespace_lock);
3290 return (error);
3291 }
3292
3293 has_features = B_FALSE;
3294 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3295 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3296 if (zpool_prop_feature(nvpair_name(elem)))
3297 has_features = B_TRUE;
3298 }
3299
3300 if (has_features || nvlist_lookup_uint64(props,
3301 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3302 version = SPA_VERSION;
3303 }
3304 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3305
3306 spa->spa_first_txg = txg;
3307 spa->spa_uberblock.ub_txg = txg - 1;
3308 spa->spa_uberblock.ub_version = version;
3309 spa->spa_ubsync = spa->spa_uberblock;
3310
3311 /*
3312 * Create "The Godfather" zio to hold all async IOs
3313 */
3314 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3315 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3316
3317 /*
3318 * Create the root vdev.
3319 */
3320 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3321
3322 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3323
3324 ASSERT(error != 0 || rvd != NULL);
3325 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3326
3327 if (error == 0 && !zfs_allocatable_devs(nvroot))
3328 error = EINVAL;
3329
3330 if (error == 0 &&
3331 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3332 (error = spa_validate_aux(spa, nvroot, txg,
3333 VDEV_ALLOC_ADD)) == 0) {
3334 for (int c = 0; c < rvd->vdev_children; c++) {
3335 vdev_metaslab_set_size(rvd->vdev_child[c]);
3336 vdev_expand(rvd->vdev_child[c], txg);
3337 }
3338 }
3339
3340 spa_config_exit(spa, SCL_ALL, FTAG);
3341
3342 if (error != 0) {
3343 spa_unload(spa);
3344 spa_deactivate(spa);
3345 spa_remove(spa);
3346 mutex_exit(&spa_namespace_lock);
3347 return (error);
3348 }
3349
3350 /*
3351 * Get the list of spares, if specified.
3352 */
3353 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3354 &spares, &nspares) == 0) {
3355 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3356 KM_SLEEP) == 0);
3357 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3358 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3359 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3360 spa_load_spares(spa);
3361 spa_config_exit(spa, SCL_ALL, FTAG);
3362 spa->spa_spares.sav_sync = B_TRUE;
3363 }
3364
3365 /*
3366 * Get the list of level 2 cache devices, if specified.
3367 */
3368 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3369 &l2cache, &nl2cache) == 0) {
3370 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3371 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3372 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3373 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3374 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3375 spa_load_l2cache(spa);
3376 spa_config_exit(spa, SCL_ALL, FTAG);
3377 spa->spa_l2cache.sav_sync = B_TRUE;
3378 }
3379
3380 spa->spa_is_initializing = B_TRUE;
3381 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3382 spa->spa_meta_objset = dp->dp_meta_objset;
3383 spa->spa_is_initializing = B_FALSE;
3384
3385 /*
3386 * Create DDTs (dedup tables).
3387 */
3388 ddt_create(spa);
3389
3390 spa_update_dspace(spa);
3391
3392 tx = dmu_tx_create_assigned(dp, txg);
3393
3394 /*
3395 * Create the pool config object.
3396 */
3397 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3398 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3399 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3400
3401 if (zap_add(spa->spa_meta_objset,
3402 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3403 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3404 cmn_err(CE_PANIC, "failed to add pool config");
3405 }
3406
3407 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3408 spa_feature_create_zap_objects(spa, tx);
3409
3410 if (zap_add(spa->spa_meta_objset,
3411 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3412 sizeof (uint64_t), 1, &version, tx) != 0) {
3413 cmn_err(CE_PANIC, "failed to add pool version");
3414 }
3415
3416 /* Newly created pools with the right version are always deflated. */
3417 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3418 spa->spa_deflate = TRUE;
3419 if (zap_add(spa->spa_meta_objset,
3420 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3421 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3422 cmn_err(CE_PANIC, "failed to add deflate");
3423 }
3424 }
3425
3426 /*
3427 * Create the deferred-free bpobj. Turn off compression
3428 * because sync-to-convergence takes longer if the blocksize
3429 * keeps changing.
3430 */
3431 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3432 dmu_object_set_compress(spa->spa_meta_objset, obj,
3433 ZIO_COMPRESS_OFF, tx);
3434 if (zap_add(spa->spa_meta_objset,
3435 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3436 sizeof (uint64_t), 1, &obj, tx) != 0) {
3437 cmn_err(CE_PANIC, "failed to add bpobj");
3438 }
3439 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3440 spa->spa_meta_objset, obj));
3441
3442 /*
3443 * Create the pool's history object.
3444 */
3445 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3446 spa_history_create_obj(spa, tx);
3447
3448 /*
3449 * Set pool properties.
3450 */
3451 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3452 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3453 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3454 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3455
3456 if (props != NULL) {
3457 spa_configfile_set(spa, props, B_FALSE);
3458 spa_sync_props(spa, props, tx);
3459 }
3460
3461 dmu_tx_commit(tx);
3462
3463 spa->spa_sync_on = B_TRUE;
3464 txg_sync_start(spa->spa_dsl_pool);
3465
3466 /*
3467 * We explicitly wait for the first transaction to complete so that our
3468 * bean counters are appropriately updated.
3469 */
3470 txg_wait_synced(spa->spa_dsl_pool, txg);
3471
3472 spa_config_sync(spa, B_FALSE, B_TRUE);
3473
3474 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3475 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3476 spa_history_log_version(spa, LOG_POOL_CREATE);
3477
3478 spa->spa_minref = refcount_count(&spa->spa_refcount);
3479
3480 mutex_exit(&spa_namespace_lock);
3481
3482 return (0);
3483}
3484
3485#if defined(sun)
3486#ifdef _KERNEL
3487/*
3488 * Get the root pool information from the root disk, then import the root pool
3489 * during the system boot up time.
3490 */
3491extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3492
3493static nvlist_t *
3494spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3495{
3496 nvlist_t *config;
3497 nvlist_t *nvtop, *nvroot;
3498 uint64_t pgid;
3499
3500 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3501 return (NULL);
3502
3503 /*
3504 * Add this top-level vdev to the child array.
3505 */
3506 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3507 &nvtop) == 0);
3508 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3509 &pgid) == 0);
3510 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3511
3512 /*
3513 * Put this pool's top-level vdevs into a root vdev.
3514 */
3515 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3516 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3517 VDEV_TYPE_ROOT) == 0);
3518 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3519 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3520 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3521 &nvtop, 1) == 0);
3522
3523 /*
3524 * Replace the existing vdev_tree with the new root vdev in
3525 * this pool's configuration (remove the old, add the new).
3526 */
3527 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3528 nvlist_free(nvroot);
3529 return (config);
3530}
3531
3532/*
3533 * Walk the vdev tree and see if we can find a device with "better"
3534 * configuration. A configuration is "better" if the label on that
3535 * device has a more recent txg.
3536 */
3537static void
3538spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3539{
3540 for (int c = 0; c < vd->vdev_children; c++)
3541 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3542
3543 if (vd->vdev_ops->vdev_op_leaf) {
3544 nvlist_t *label;
3545 uint64_t label_txg;
3546
3547 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3548 &label) != 0)
3549 return;
3550
3551 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3552 &label_txg) == 0);
3553
3554 /*
3555 * Do we have a better boot device?
3556 */
3557 if (label_txg > *txg) {
3558 *txg = label_txg;
3559 *avd = vd;
3560 }
3561 nvlist_free(label);
3562 }
3563}
3564
3565/*
3566 * Import a root pool.
3567 *
3568 * For x86. devpath_list will consist of devid and/or physpath name of
3569 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3570 * The GRUB "findroot" command will return the vdev we should boot.
3571 *
3572 * For Sparc, devpath_list consists the physpath name of the booting device
3573 * no matter the rootpool is a single device pool or a mirrored pool.
3574 * e.g.
3575 * "/pci@1f,0/ide@d/disk@0,0:a"
3576 */
3577int
3578spa_import_rootpool(char *devpath, char *devid)
3579{
3580 spa_t *spa;
3581 vdev_t *rvd, *bvd, *avd = NULL;
3582 nvlist_t *config, *nvtop;
3583 uint64_t guid, txg;
3584 char *pname;
3585 int error;
3586
3587 /*
3588 * Read the label from the boot device and generate a configuration.
3589 */
3590 config = spa_generate_rootconf(devpath, devid, &guid);
3591#if defined(_OBP) && defined(_KERNEL)
3592 if (config == NULL) {
3593 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3594 /* iscsi boot */
3595 get_iscsi_bootpath_phy(devpath);
3596 config = spa_generate_rootconf(devpath, devid, &guid);
3597 }
3598 }
3599#endif
3600 if (config == NULL) {
3601 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3602 devpath);
3603 return (EIO);
3604 }
3605
3606 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3607 &pname) == 0);
3608 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3609
3610 mutex_enter(&spa_namespace_lock);
3611 if ((spa = spa_lookup(pname)) != NULL) {
3612 /*
3613 * Remove the existing root pool from the namespace so that we
3614 * can replace it with the correct config we just read in.
3615 */
3616 spa_remove(spa);
3617 }
3618
3619 spa = spa_add(pname, config, NULL);
3620 spa->spa_is_root = B_TRUE;
3621 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3622
3623 /*
3624 * Build up a vdev tree based on the boot device's label config.
3625 */
3626 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3627 &nvtop) == 0);
3628 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3629 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3630 VDEV_ALLOC_ROOTPOOL);
3631 spa_config_exit(spa, SCL_ALL, FTAG);
3632 if (error) {
3633 mutex_exit(&spa_namespace_lock);
3634 nvlist_free(config);
3635 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3636 pname);
3637 return (error);
3638 }
3639
3640 /*
3641 * Get the boot vdev.
3642 */
3643 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3644 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3645 (u_longlong_t)guid);
3646 error = ENOENT;
3647 goto out;
3648 }
3649
3650 /*
3651 * Determine if there is a better boot device.
3652 */
3653 avd = bvd;
3654 spa_alt_rootvdev(rvd, &avd, &txg);
3655 if (avd != bvd) {
3656 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3657 "try booting from '%s'", avd->vdev_path);
3658 error = EINVAL;
3659 goto out;
3660 }
3661
3662 /*
3663 * If the boot device is part of a spare vdev then ensure that
3664 * we're booting off the active spare.
3665 */
3666 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3667 !bvd->vdev_isspare) {
3668 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3669 "try booting from '%s'",
3670 bvd->vdev_parent->
3671 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3672 error = EINVAL;
3673 goto out;
3674 }
3675
3676 error = 0;
3677 spa_history_log_version(spa, LOG_POOL_IMPORT);
3678out:
3679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3680 vdev_free(rvd);
3681 spa_config_exit(spa, SCL_ALL, FTAG);
3682 mutex_exit(&spa_namespace_lock);
3683
3684 nvlist_free(config);
3685 return (error);
3686}
3687
3688#endif
3689#endif /* sun */
3690
3691/*
3692 * Import a non-root pool into the system.
3693 */
3694int
3695spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3696{
3697 spa_t *spa;
3698 char *altroot = NULL;
3699 spa_load_state_t state = SPA_LOAD_IMPORT;
3700 zpool_rewind_policy_t policy;
3701 uint64_t mode = spa_mode_global;
3702 uint64_t readonly = B_FALSE;
3703 int error;
3704 nvlist_t *nvroot;
3705 nvlist_t **spares, **l2cache;
3706 uint_t nspares, nl2cache;
3707
3708 /*
3709 * If a pool with this name exists, return failure.
3710 */
3711 mutex_enter(&spa_namespace_lock);
3712 if (spa_lookup(pool) != NULL) {
3713 mutex_exit(&spa_namespace_lock);
3714 return (EEXIST);
3715 }
3716
3717 /*
3718 * Create and initialize the spa structure.
3719 */
3720 (void) nvlist_lookup_string(props,
3721 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3722 (void) nvlist_lookup_uint64(props,
3723 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3724 if (readonly)
3725 mode = FREAD;
3726 spa = spa_add(pool, config, altroot);
3727 spa->spa_import_flags = flags;
3728
3729 /*
3730 * Verbatim import - Take a pool and insert it into the namespace
3731 * as if it had been loaded at boot.
3732 */
3733 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3734 if (props != NULL)
3735 spa_configfile_set(spa, props, B_FALSE);
3736
3737 spa_config_sync(spa, B_FALSE, B_TRUE);
3738
3739 mutex_exit(&spa_namespace_lock);
3740 spa_history_log_version(spa, LOG_POOL_IMPORT);
3741
3742 return (0);
3743 }
3744
3745 spa_activate(spa, mode);
3746
3747 /*
3748 * Don't start async tasks until we know everything is healthy.
3749 */
3750 spa_async_suspend(spa);
3751
3752 zpool_get_rewind_policy(config, &policy);
3753 if (policy.zrp_request & ZPOOL_DO_REWIND)
3754 state = SPA_LOAD_RECOVER;
3755
3756 /*
3757 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3758 * because the user-supplied config is actually the one to trust when
3759 * doing an import.
3760 */
3761 if (state != SPA_LOAD_RECOVER)
3762 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3763
3764 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3765 policy.zrp_request);
3766
3767 /*
3768 * Propagate anything learned while loading the pool and pass it
3769 * back to caller (i.e. rewind info, missing devices, etc).
3770 */
3771 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3772 spa->spa_load_info) == 0);
3773
3774 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3775 /*
3776 * Toss any existing sparelist, as it doesn't have any validity
3777 * anymore, and conflicts with spa_has_spare().
3778 */
3779 if (spa->spa_spares.sav_config) {
3780 nvlist_free(spa->spa_spares.sav_config);
3781 spa->spa_spares.sav_config = NULL;
3782 spa_load_spares(spa);
3783 }
3784 if (spa->spa_l2cache.sav_config) {
3785 nvlist_free(spa->spa_l2cache.sav_config);
3786 spa->spa_l2cache.sav_config = NULL;
3787 spa_load_l2cache(spa);
3788 }
3789
3790 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3791 &nvroot) == 0);
3792 if (error == 0)
3793 error = spa_validate_aux(spa, nvroot, -1ULL,
3794 VDEV_ALLOC_SPARE);
3795 if (error == 0)
3796 error = spa_validate_aux(spa, nvroot, -1ULL,
3797 VDEV_ALLOC_L2CACHE);
3798 spa_config_exit(spa, SCL_ALL, FTAG);
3799
3800 if (props != NULL)
3801 spa_configfile_set(spa, props, B_FALSE);
3802
3803 if (error != 0 || (props && spa_writeable(spa) &&
3804 (error = spa_prop_set(spa, props)))) {
3805 spa_unload(spa);
3806 spa_deactivate(spa);
3807 spa_remove(spa);
3808 mutex_exit(&spa_namespace_lock);
3809 return (error);
3810 }
3811
3812 spa_async_resume(spa);
3813
3814 /*
3815 * Override any spares and level 2 cache devices as specified by
3816 * the user, as these may have correct device names/devids, etc.
3817 */
3818 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3819 &spares, &nspares) == 0) {
3820 if (spa->spa_spares.sav_config)
3821 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3822 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3823 else
3824 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3825 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3826 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3827 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3828 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3829 spa_load_spares(spa);
3830 spa_config_exit(spa, SCL_ALL, FTAG);
3831 spa->spa_spares.sav_sync = B_TRUE;
3832 }
3833 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3834 &l2cache, &nl2cache) == 0) {
3835 if (spa->spa_l2cache.sav_config)
3836 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3837 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3838 else
3839 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3840 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3841 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3842 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3843 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3844 spa_load_l2cache(spa);
3845 spa_config_exit(spa, SCL_ALL, FTAG);
3846 spa->spa_l2cache.sav_sync = B_TRUE;
3847 }
3848
3849 /*
3850 * Check for any removed devices.
3851 */
3852 if (spa->spa_autoreplace) {
3853 spa_aux_check_removed(&spa->spa_spares);
3854 spa_aux_check_removed(&spa->spa_l2cache);
3855 }
3856
3857 if (spa_writeable(spa)) {
3858 /*
3859 * Update the config cache to include the newly-imported pool.
3860 */
3861 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3862 }
3863
3864 /*
3865 * It's possible that the pool was expanded while it was exported.
3866 * We kick off an async task to handle this for us.
3867 */
3868 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3869
3870 mutex_exit(&spa_namespace_lock);
3871 spa_history_log_version(spa, LOG_POOL_IMPORT);
3872
3873#ifdef __FreeBSD__
3874#ifdef _KERNEL
3875 zvol_create_minors(pool);
3876#endif
3877#endif
3878 return (0);
3879}
3880
3881nvlist_t *
3882spa_tryimport(nvlist_t *tryconfig)
3883{
3884 nvlist_t *config = NULL;
3885 char *poolname;
3886 spa_t *spa;
3887 uint64_t state;
3888 int error;
3889
3890 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3891 return (NULL);
3892
3893 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3894 return (NULL);
3895
3896 /*
3897 * Create and initialize the spa structure.
3898 */
3899 mutex_enter(&spa_namespace_lock);
3900 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3901 spa_activate(spa, FREAD);
3902
3903 /*
3904 * Pass off the heavy lifting to spa_load().
3905 * Pass TRUE for mosconfig because the user-supplied config
3906 * is actually the one to trust when doing an import.
3907 */
3908 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3909
3910 /*
3911 * If 'tryconfig' was at least parsable, return the current config.
3912 */
3913 if (spa->spa_root_vdev != NULL) {
3914 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3915 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3916 poolname) == 0);
3917 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3918 state) == 0);
3919 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3920 spa->spa_uberblock.ub_timestamp) == 0);
3921 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3922 spa->spa_load_info) == 0);
3923
3924 /*
3925 * If the bootfs property exists on this pool then we
3926 * copy it out so that external consumers can tell which
3927 * pools are bootable.
3928 */
3929 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3930 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3931
3932 /*
3933 * We have to play games with the name since the
3934 * pool was opened as TRYIMPORT_NAME.
3935 */
3936 if (dsl_dsobj_to_dsname(spa_name(spa),
3937 spa->spa_bootfs, tmpname) == 0) {
3938 char *cp;
3939 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3940
3941 cp = strchr(tmpname, '/');
3942 if (cp == NULL) {
3943 (void) strlcpy(dsname, tmpname,
3944 MAXPATHLEN);
3945 } else {
3946 (void) snprintf(dsname, MAXPATHLEN,
3947 "%s/%s", poolname, ++cp);
3948 }
3949 VERIFY(nvlist_add_string(config,
3950 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3951 kmem_free(dsname, MAXPATHLEN);
3952 }
3953 kmem_free(tmpname, MAXPATHLEN);
3954 }
3955
3956 /*
3957 * Add the list of hot spares and level 2 cache devices.
3958 */
3959 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3960 spa_add_spares(spa, config);
3961 spa_add_l2cache(spa, config);
3962 spa_config_exit(spa, SCL_CONFIG, FTAG);
3963 }
3964
3965 spa_unload(spa);
3966 spa_deactivate(spa);
3967 spa_remove(spa);
3968 mutex_exit(&spa_namespace_lock);
3969
3970 return (config);
3971}
3972
3973/*
3974 * Pool export/destroy
3975 *
3976 * The act of destroying or exporting a pool is very simple. We make sure there
3977 * is no more pending I/O and any references to the pool are gone. Then, we
3978 * update the pool state and sync all the labels to disk, removing the
3979 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3980 * we don't sync the labels or remove the configuration cache.
3981 */
3982static int
3983spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3984 boolean_t force, boolean_t hardforce)
3985{
3986 spa_t *spa;
3987
3988 if (oldconfig)
3989 *oldconfig = NULL;
3990
3991 if (!(spa_mode_global & FWRITE))
3992 return (EROFS);
3993
3994 mutex_enter(&spa_namespace_lock);
3995 if ((spa = spa_lookup(pool)) == NULL) {
3996 mutex_exit(&spa_namespace_lock);
3997 return (ENOENT);
3998 }
3999
4000 /*
4001 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4002 * reacquire the namespace lock, and see if we can export.
4003 */
4004 spa_open_ref(spa, FTAG);
4005 mutex_exit(&spa_namespace_lock);
4006 spa_async_suspend(spa);
4007 mutex_enter(&spa_namespace_lock);
4008 spa_close(spa, FTAG);
4009
4010 /*
4011 * The pool will be in core if it's openable,
4012 * in which case we can modify its state.
4013 */
4014 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4015 /*
4016 * Objsets may be open only because they're dirty, so we
4017 * have to force it to sync before checking spa_refcnt.
4018 */
4019 txg_wait_synced(spa->spa_dsl_pool, 0);
4020
4021 /*
4022 * A pool cannot be exported or destroyed if there are active
4023 * references. If we are resetting a pool, allow references by
4024 * fault injection handlers.
4025 */
4026 if (!spa_refcount_zero(spa) ||
4027 (spa->spa_inject_ref != 0 &&
4028 new_state != POOL_STATE_UNINITIALIZED)) {
4029 spa_async_resume(spa);
4030 mutex_exit(&spa_namespace_lock);
4031 return (EBUSY);
4032 }
4033
4034 /*
4035 * A pool cannot be exported if it has an active shared spare.
4036 * This is to prevent other pools stealing the active spare
4037 * from an exported pool. At user's own will, such pool can
4038 * be forcedly exported.
4039 */
4040 if (!force && new_state == POOL_STATE_EXPORTED &&
4041 spa_has_active_shared_spare(spa)) {
4042 spa_async_resume(spa);
4043 mutex_exit(&spa_namespace_lock);
4044 return (EXDEV);
4045 }
4046
4047 /*
4048 * We want this to be reflected on every label,
4049 * so mark them all dirty. spa_unload() will do the
4050 * final sync that pushes these changes out.
4051 */
4052 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4053 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4054 spa->spa_state = new_state;
4055 spa->spa_final_txg = spa_last_synced_txg(spa) +
4056 TXG_DEFER_SIZE + 1;
4057 vdev_config_dirty(spa->spa_root_vdev);
4058 spa_config_exit(spa, SCL_ALL, FTAG);
4059 }
4060 }
4061
4062 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4063
4064 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4065 spa_unload(spa);
4066 spa_deactivate(spa);
4067 }
4068
4069 if (oldconfig && spa->spa_config)
4070 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4071
4072 if (new_state != POOL_STATE_UNINITIALIZED) {
4073 if (!hardforce)
4074 spa_config_sync(spa, B_TRUE, B_TRUE);
4075 spa_remove(spa);
4076 }
4077 mutex_exit(&spa_namespace_lock);
4078
4079 return (0);
4080}
4081
4082/*
4083 * Destroy a storage pool.
4084 */
4085int
4086spa_destroy(char *pool)
4087{
4088 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4089 B_FALSE, B_FALSE));
4090}
4091
4092/*
4093 * Export a storage pool.
4094 */
4095int
4096spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4097 boolean_t hardforce)
4098{
4099 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4100 force, hardforce));
4101}
4102
4103/*
4104 * Similar to spa_export(), this unloads the spa_t without actually removing it
4105 * from the namespace in any way.
4106 */
4107int
4108spa_reset(char *pool)
4109{
4110 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4111 B_FALSE, B_FALSE));
4112}
4113
4114/*
4115 * ==========================================================================
4116 * Device manipulation
4117 * ==========================================================================
4118 */
4119
4120/*
4121 * Add a device to a storage pool.
4122 */
4123int
4124spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4125{
4126 uint64_t txg, id;
4127 int error;
4128 vdev_t *rvd = spa->spa_root_vdev;
4129 vdev_t *vd, *tvd;
4130 nvlist_t **spares, **l2cache;
4131 uint_t nspares, nl2cache;
4132
4133 ASSERT(spa_writeable(spa));
4134
4135 txg = spa_vdev_enter(spa);
4136
4137 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4138 VDEV_ALLOC_ADD)) != 0)
4139 return (spa_vdev_exit(spa, NULL, txg, error));
4140
4141 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4142
4143 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4144 &nspares) != 0)
4145 nspares = 0;
4146
4147 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4148 &nl2cache) != 0)
4149 nl2cache = 0;
4150
4151 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4152 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4153
4154 if (vd->vdev_children != 0 &&
4155 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4156 return (spa_vdev_exit(spa, vd, txg, error));
4157
4158 /*
4159 * We must validate the spares and l2cache devices after checking the
4160 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4161 */
4162 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4163 return (spa_vdev_exit(spa, vd, txg, error));
4164
4165 /*
4166 * Transfer each new top-level vdev from vd to rvd.
4167 */
4168 for (int c = 0; c < vd->vdev_children; c++) {
4169
4170 /*
4171 * Set the vdev id to the first hole, if one exists.
4172 */
4173 for (id = 0; id < rvd->vdev_children; id++) {
4174 if (rvd->vdev_child[id]->vdev_ishole) {
4175 vdev_free(rvd->vdev_child[id]);
4176 break;
4177 }
4178 }
4179 tvd = vd->vdev_child[c];
4180 vdev_remove_child(vd, tvd);
4181 tvd->vdev_id = id;
4182 vdev_add_child(rvd, tvd);
4183 vdev_config_dirty(tvd);
4184 }
4185
4186 if (nspares != 0) {
4187 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4188 ZPOOL_CONFIG_SPARES);
4189 spa_load_spares(spa);
4190 spa->spa_spares.sav_sync = B_TRUE;
4191 }
4192
4193 if (nl2cache != 0) {
4194 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4195 ZPOOL_CONFIG_L2CACHE);
4196 spa_load_l2cache(spa);
4197 spa->spa_l2cache.sav_sync = B_TRUE;
4198 }
4199
4200 /*
4201 * We have to be careful when adding new vdevs to an existing pool.
4202 * If other threads start allocating from these vdevs before we
4203 * sync the config cache, and we lose power, then upon reboot we may
4204 * fail to open the pool because there are DVAs that the config cache
4205 * can't translate. Therefore, we first add the vdevs without
4206 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4207 * and then let spa_config_update() initialize the new metaslabs.
4208 *
4209 * spa_load() checks for added-but-not-initialized vdevs, so that
4210 * if we lose power at any point in this sequence, the remaining
4211 * steps will be completed the next time we load the pool.
4212 */
4213 (void) spa_vdev_exit(spa, vd, txg, 0);
4214
4215 mutex_enter(&spa_namespace_lock);
4216 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4217 mutex_exit(&spa_namespace_lock);
4218
4219 return (0);
4220}
4221
4222/*
4223 * Attach a device to a mirror. The arguments are the path to any device
4224 * in the mirror, and the nvroot for the new device. If the path specifies
4225 * a device that is not mirrored, we automatically insert the mirror vdev.
4226 *
4227 * If 'replacing' is specified, the new device is intended to replace the
4228 * existing device; in this case the two devices are made into their own
4229 * mirror using the 'replacing' vdev, which is functionally identical to
4230 * the mirror vdev (it actually reuses all the same ops) but has a few
4231 * extra rules: you can't attach to it after it's been created, and upon
4232 * completion of resilvering, the first disk (the one being replaced)
4233 * is automatically detached.
4234 */
4235int
4236spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4237{
4238 uint64_t txg, dtl_max_txg;
4239 vdev_t *rvd = spa->spa_root_vdev;
4240 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4241 vdev_ops_t *pvops;
4242 char *oldvdpath, *newvdpath;
4243 int newvd_isspare;
4244 int error;
4245
4246 ASSERT(spa_writeable(spa));
4247
4248 txg = spa_vdev_enter(spa);
4249
4250 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4251
4252 if (oldvd == NULL)
4253 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4254
4255 if (!oldvd->vdev_ops->vdev_op_leaf)
4256 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4257
4258 pvd = oldvd->vdev_parent;
4259
4260 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4261 VDEV_ALLOC_ATTACH)) != 0)
4262 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4263
4264 if (newrootvd->vdev_children != 1)
4265 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4266
4267 newvd = newrootvd->vdev_child[0];
4268
4269 if (!newvd->vdev_ops->vdev_op_leaf)
4270 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4271
4272 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4273 return (spa_vdev_exit(spa, newrootvd, txg, error));
4274
4275 /*
4276 * Spares can't replace logs
4277 */
4278 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4279 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4280
4281 if (!replacing) {
4282 /*
4283 * For attach, the only allowable parent is a mirror or the root
4284 * vdev.
4285 */
4286 if (pvd->vdev_ops != &vdev_mirror_ops &&
4287 pvd->vdev_ops != &vdev_root_ops)
4288 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4289
4290 pvops = &vdev_mirror_ops;
4291 } else {
4292 /*
4293 * Active hot spares can only be replaced by inactive hot
4294 * spares.
4295 */
4296 if (pvd->vdev_ops == &vdev_spare_ops &&
4297 oldvd->vdev_isspare &&
4298 !spa_has_spare(spa, newvd->vdev_guid))
4299 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4300
4301 /*
4302 * If the source is a hot spare, and the parent isn't already a
4303 * spare, then we want to create a new hot spare. Otherwise, we
4304 * want to create a replacing vdev. The user is not allowed to
4305 * attach to a spared vdev child unless the 'isspare' state is
4306 * the same (spare replaces spare, non-spare replaces
4307 * non-spare).
4308 */
4309 if (pvd->vdev_ops == &vdev_replacing_ops &&
4310 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4311 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4312 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4313 newvd->vdev_isspare != oldvd->vdev_isspare) {
4314 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4315 }
4316
4317 if (newvd->vdev_isspare)
4318 pvops = &vdev_spare_ops;
4319 else
4320 pvops = &vdev_replacing_ops;
4321 }
4322
4323 /*
4324 * Make sure the new device is big enough.
4325 */
4326 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4327 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4328
4329 /*
4330 * The new device cannot have a higher alignment requirement
4331 * than the top-level vdev.
4332 */
4333 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4334 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4335
4336 /*
4337 * If this is an in-place replacement, update oldvd's path and devid
4338 * to make it distinguishable from newvd, and unopenable from now on.
4339 */
4340 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4341 spa_strfree(oldvd->vdev_path);
4342 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4343 KM_SLEEP);
4344 (void) sprintf(oldvd->vdev_path, "%s/%s",
4345 newvd->vdev_path, "old");
4346 if (oldvd->vdev_devid != NULL) {
4347 spa_strfree(oldvd->vdev_devid);
4348 oldvd->vdev_devid = NULL;
4349 }
4350 }
4351
4352 /* mark the device being resilvered */
4353 newvd->vdev_resilvering = B_TRUE;
4354
4355 /*
4356 * If the parent is not a mirror, or if we're replacing, insert the new
4357 * mirror/replacing/spare vdev above oldvd.
4358 */
4359 if (pvd->vdev_ops != pvops)
4360 pvd = vdev_add_parent(oldvd, pvops);
4361
4362 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4363 ASSERT(pvd->vdev_ops == pvops);
4364 ASSERT(oldvd->vdev_parent == pvd);
4365
4366 /*
4367 * Extract the new device from its root and add it to pvd.
4368 */
4369 vdev_remove_child(newrootvd, newvd);
4370 newvd->vdev_id = pvd->vdev_children;
4371 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4372 vdev_add_child(pvd, newvd);
4373
4374 tvd = newvd->vdev_top;
4375 ASSERT(pvd->vdev_top == tvd);
4376 ASSERT(tvd->vdev_parent == rvd);
4377
4378 vdev_config_dirty(tvd);
4379
4380 /*
4381 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4382 * for any dmu_sync-ed blocks. It will propagate upward when
4383 * spa_vdev_exit() calls vdev_dtl_reassess().
4384 */
4385 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4386
4387 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4388 dtl_max_txg - TXG_INITIAL);
4389
4390 if (newvd->vdev_isspare) {
4391 spa_spare_activate(newvd);
4392 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4393 }
4394
4395 oldvdpath = spa_strdup(oldvd->vdev_path);
4396 newvdpath = spa_strdup(newvd->vdev_path);
4397 newvd_isspare = newvd->vdev_isspare;
4398
4399 /*
4400 * Mark newvd's DTL dirty in this txg.
4401 */
4402 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4403
4404 /*
4405 * Restart the resilver
4406 */
4407 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4408
4409 /*
4410 * Commit the config
4411 */
4412 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4413
4414 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4415 "%s vdev=%s %s vdev=%s",
4416 replacing && newvd_isspare ? "spare in" :
4417 replacing ? "replace" : "attach", newvdpath,
4418 replacing ? "for" : "to", oldvdpath);
4419
4420 spa_strfree(oldvdpath);
4421 spa_strfree(newvdpath);
4422
4423 if (spa->spa_bootfs)
4424 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4425
4426 return (0);
4427}
4428
4429/*
4430 * Detach a device from a mirror or replacing vdev.
4431 * If 'replace_done' is specified, only detach if the parent
4432 * is a replacing vdev.
4433 */
4434int
4435spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4436{
4437 uint64_t txg;
4438 int error;
4439 vdev_t *rvd = spa->spa_root_vdev;
4440 vdev_t *vd, *pvd, *cvd, *tvd;
4441 boolean_t unspare = B_FALSE;
4442 uint64_t unspare_guid;
4443 char *vdpath;
4444
4445 ASSERT(spa_writeable(spa));
4446
4447 txg = spa_vdev_enter(spa);
4448
4449 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4450
4451 if (vd == NULL)
4452 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4453
4454 if (!vd->vdev_ops->vdev_op_leaf)
4455 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4456
4457 pvd = vd->vdev_parent;
4458
4459 /*
4460 * If the parent/child relationship is not as expected, don't do it.
4461 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4462 * vdev that's replacing B with C. The user's intent in replacing
4463 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4464 * the replace by detaching C, the expected behavior is to end up
4465 * M(A,B). But suppose that right after deciding to detach C,
4466 * the replacement of B completes. We would have M(A,C), and then
4467 * ask to detach C, which would leave us with just A -- not what
4468 * the user wanted. To prevent this, we make sure that the
4469 * parent/child relationship hasn't changed -- in this example,
4470 * that C's parent is still the replacing vdev R.
4471 */
4472 if (pvd->vdev_guid != pguid && pguid != 0)
4473 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4474
4475 /*
4476 * Only 'replacing' or 'spare' vdevs can be replaced.
4477 */
4478 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4479 pvd->vdev_ops != &vdev_spare_ops)
4480 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4481
4482 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4483 spa_version(spa) >= SPA_VERSION_SPARES);
4484
4485 /*
4486 * Only mirror, replacing, and spare vdevs support detach.
4487 */
4488 if (pvd->vdev_ops != &vdev_replacing_ops &&
4489 pvd->vdev_ops != &vdev_mirror_ops &&
4490 pvd->vdev_ops != &vdev_spare_ops)
4491 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4492
4493 /*
4494 * If this device has the only valid copy of some data,
4495 * we cannot safely detach it.
4496 */
4497 if (vdev_dtl_required(vd))
4498 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4499
4500 ASSERT(pvd->vdev_children >= 2);
4501
4502 /*
4503 * If we are detaching the second disk from a replacing vdev, then
4504 * check to see if we changed the original vdev's path to have "/old"
4505 * at the end in spa_vdev_attach(). If so, undo that change now.
4506 */
4507 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4508 vd->vdev_path != NULL) {
4509 size_t len = strlen(vd->vdev_path);
4510
4511 for (int c = 0; c < pvd->vdev_children; c++) {
4512 cvd = pvd->vdev_child[c];
4513
4514 if (cvd == vd || cvd->vdev_path == NULL)
4515 continue;
4516
4517 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4518 strcmp(cvd->vdev_path + len, "/old") == 0) {
4519 spa_strfree(cvd->vdev_path);
4520 cvd->vdev_path = spa_strdup(vd->vdev_path);
4521 break;
4522 }
4523 }
4524 }
4525
4526 /*
4527 * If we are detaching the original disk from a spare, then it implies
4528 * that the spare should become a real disk, and be removed from the
4529 * active spare list for the pool.
4530 */
4531 if (pvd->vdev_ops == &vdev_spare_ops &&
4532 vd->vdev_id == 0 &&
4533 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4534 unspare = B_TRUE;
4535
4536 /*
4537 * Erase the disk labels so the disk can be used for other things.
4538 * This must be done after all other error cases are handled,
4539 * but before we disembowel vd (so we can still do I/O to it).
4540 * But if we can't do it, don't treat the error as fatal --
4541 * it may be that the unwritability of the disk is the reason
4542 * it's being detached!
4543 */
4544 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4545
4546 /*
4547 * Remove vd from its parent and compact the parent's children.
4548 */
4549 vdev_remove_child(pvd, vd);
4550 vdev_compact_children(pvd);
4551
4552 /*
4553 * Remember one of the remaining children so we can get tvd below.
4554 */
4555 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4556
4557 /*
4558 * If we need to remove the remaining child from the list of hot spares,
4559 * do it now, marking the vdev as no longer a spare in the process.
4560 * We must do this before vdev_remove_parent(), because that can
4561 * change the GUID if it creates a new toplevel GUID. For a similar
4562 * reason, we must remove the spare now, in the same txg as the detach;
4563 * otherwise someone could attach a new sibling, change the GUID, and
4564 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4565 */
4566 if (unspare) {
4567 ASSERT(cvd->vdev_isspare);
4568 spa_spare_remove(cvd);
4569 unspare_guid = cvd->vdev_guid;
4570 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4571 cvd->vdev_unspare = B_TRUE;
4572 }
4573
4574 /*
4575 * If the parent mirror/replacing vdev only has one child,
4576 * the parent is no longer needed. Remove it from the tree.
4577 */
4578 if (pvd->vdev_children == 1) {
4579 if (pvd->vdev_ops == &vdev_spare_ops)
4580 cvd->vdev_unspare = B_FALSE;
4581 vdev_remove_parent(cvd);
4582 cvd->vdev_resilvering = B_FALSE;
4583 }
4584
4585
4586 /*
4587 * We don't set tvd until now because the parent we just removed
4588 * may have been the previous top-level vdev.
4589 */
4590 tvd = cvd->vdev_top;
4591 ASSERT(tvd->vdev_parent == rvd);
4592
4593 /*
4594 * Reevaluate the parent vdev state.
4595 */
4596 vdev_propagate_state(cvd);
4597
4598 /*
4599 * If the 'autoexpand' property is set on the pool then automatically
4600 * try to expand the size of the pool. For example if the device we
4601 * just detached was smaller than the others, it may be possible to
4602 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4603 * first so that we can obtain the updated sizes of the leaf vdevs.
4604 */
4605 if (spa->spa_autoexpand) {
4606 vdev_reopen(tvd);
4607 vdev_expand(tvd, txg);
4608 }
4609
4610 vdev_config_dirty(tvd);
4611
4612 /*
4613 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4614 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4615 * But first make sure we're not on any *other* txg's DTL list, to
4616 * prevent vd from being accessed after it's freed.
4617 */
4618 vdpath = spa_strdup(vd->vdev_path);
4619 for (int t = 0; t < TXG_SIZE; t++)
4620 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4621 vd->vdev_detached = B_TRUE;
4622 vdev_dirty(tvd, VDD_DTL, vd, txg);
4623
4624 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4625
4626 /* hang on to the spa before we release the lock */
4627 spa_open_ref(spa, FTAG);
4628
4629 error = spa_vdev_exit(spa, vd, txg, 0);
4630
4631 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4632 "vdev=%s", vdpath);
4633 spa_strfree(vdpath);
4634
4635 /*
4636 * If this was the removal of the original device in a hot spare vdev,
4637 * then we want to go through and remove the device from the hot spare
4638 * list of every other pool.
4639 */
4640 if (unspare) {
4641 spa_t *altspa = NULL;
4642
4643 mutex_enter(&spa_namespace_lock);
4644 while ((altspa = spa_next(altspa)) != NULL) {
4645 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4646 altspa == spa)
4647 continue;
4648
4649 spa_open_ref(altspa, FTAG);
4650 mutex_exit(&spa_namespace_lock);
4651 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4652 mutex_enter(&spa_namespace_lock);
4653 spa_close(altspa, FTAG);
4654 }
4655 mutex_exit(&spa_namespace_lock);
4656
4657 /* search the rest of the vdevs for spares to remove */
4658 spa_vdev_resilver_done(spa);
4659 }
4660
4661 /* all done with the spa; OK to release */
4662 mutex_enter(&spa_namespace_lock);
4663 spa_close(spa, FTAG);
4664 mutex_exit(&spa_namespace_lock);
4665
4666 return (error);
4667}
4668
4669/*
4670 * Split a set of devices from their mirrors, and create a new pool from them.
4671 */
4672int
4673spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4674 nvlist_t *props, boolean_t exp)
4675{
4676 int error = 0;
4677 uint64_t txg, *glist;
4678 spa_t *newspa;
4679 uint_t c, children, lastlog;
4680 nvlist_t **child, *nvl, *tmp;
4681 dmu_tx_t *tx;
4682 char *altroot = NULL;
4683 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4684 boolean_t activate_slog;
4685
4686 ASSERT(spa_writeable(spa));
4687
4688 txg = spa_vdev_enter(spa);
4689
4690 /* clear the log and flush everything up to now */
4691 activate_slog = spa_passivate_log(spa);
4692 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4693 error = spa_offline_log(spa);
4694 txg = spa_vdev_config_enter(spa);
4695
4696 if (activate_slog)
4697 spa_activate_log(spa);
4698
4699 if (error != 0)
4700 return (spa_vdev_exit(spa, NULL, txg, error));
4701
4702 /* check new spa name before going any further */
4703 if (spa_lookup(newname) != NULL)
4704 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4705
4706 /*
4707 * scan through all the children to ensure they're all mirrors
4708 */
4709 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4710 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4711 &children) != 0)
4712 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4713
4714 /* first, check to ensure we've got the right child count */
4715 rvd = spa->spa_root_vdev;
4716 lastlog = 0;
4717 for (c = 0; c < rvd->vdev_children; c++) {
4718 vdev_t *vd = rvd->vdev_child[c];
4719
4720 /* don't count the holes & logs as children */
4721 if (vd->vdev_islog || vd->vdev_ishole) {
4722 if (lastlog == 0)
4723 lastlog = c;
4724 continue;
4725 }
4726
4727 lastlog = 0;
4728 }
4729 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4730 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4731
4732 /* next, ensure no spare or cache devices are part of the split */
4733 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4734 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4735 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4736
4737 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4738 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4739
4740 /* then, loop over each vdev and validate it */
4741 for (c = 0; c < children; c++) {
4742 uint64_t is_hole = 0;
4743
4744 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4745 &is_hole);
4746
4747 if (is_hole != 0) {
4748 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4749 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4750 continue;
4751 } else {
4752 error = EINVAL;
4753 break;
4754 }
4755 }
4756
4757 /* which disk is going to be split? */
4758 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4759 &glist[c]) != 0) {
4760 error = EINVAL;
4761 break;
4762 }
4763
4764 /* look it up in the spa */
4765 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4766 if (vml[c] == NULL) {
4767 error = ENODEV;
4768 break;
4769 }
4770
4771 /* make sure there's nothing stopping the split */
4772 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4773 vml[c]->vdev_islog ||
4774 vml[c]->vdev_ishole ||
4775 vml[c]->vdev_isspare ||
4776 vml[c]->vdev_isl2cache ||
4777 !vdev_writeable(vml[c]) ||
4778 vml[c]->vdev_children != 0 ||
4779 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4780 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4781 error = EINVAL;
4782 break;
4783 }
4784
4785 if (vdev_dtl_required(vml[c])) {
4786 error = EBUSY;
4787 break;
4788 }
4789
4790 /* we need certain info from the top level */
4791 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4792 vml[c]->vdev_top->vdev_ms_array) == 0);
4793 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4794 vml[c]->vdev_top->vdev_ms_shift) == 0);
4795 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4796 vml[c]->vdev_top->vdev_asize) == 0);
4797 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4798 vml[c]->vdev_top->vdev_ashift) == 0);
4799 }
4800
4801 if (error != 0) {
4802 kmem_free(vml, children * sizeof (vdev_t *));
4803 kmem_free(glist, children * sizeof (uint64_t));
4804 return (spa_vdev_exit(spa, NULL, txg, error));
4805 }
4806
4807 /* stop writers from using the disks */
4808 for (c = 0; c < children; c++) {
4809 if (vml[c] != NULL)
4810 vml[c]->vdev_offline = B_TRUE;
4811 }
4812 vdev_reopen(spa->spa_root_vdev);
4813
4814 /*
4815 * Temporarily record the splitting vdevs in the spa config. This
4816 * will disappear once the config is regenerated.
4817 */
4818 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4819 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4820 glist, children) == 0);
4821 kmem_free(glist, children * sizeof (uint64_t));
4822
4823 mutex_enter(&spa->spa_props_lock);
4824 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4825 nvl) == 0);
4826 mutex_exit(&spa->spa_props_lock);
4827 spa->spa_config_splitting = nvl;
4828 vdev_config_dirty(spa->spa_root_vdev);
4829
4830 /* configure and create the new pool */
4831 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4832 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4833 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4834 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4835 spa_version(spa)) == 0);
4836 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4837 spa->spa_config_txg) == 0);
4838 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4839 spa_generate_guid(NULL)) == 0);
4840 (void) nvlist_lookup_string(props,
4841 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4842
4843 /* add the new pool to the namespace */
4844 newspa = spa_add(newname, config, altroot);
4845 newspa->spa_config_txg = spa->spa_config_txg;
4846 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4847
4848 /* release the spa config lock, retaining the namespace lock */
4849 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4850
4851 if (zio_injection_enabled)
4852 zio_handle_panic_injection(spa, FTAG, 1);
4853
4854 spa_activate(newspa, spa_mode_global);
4855 spa_async_suspend(newspa);
4856
4857#ifndef sun
4858 /* mark that we are creating new spa by splitting */
4859 newspa->spa_splitting_newspa = B_TRUE;
4860#endif
4861 /* create the new pool from the disks of the original pool */
4862 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4863#ifndef sun
4864 newspa->spa_splitting_newspa = B_FALSE;
4865#endif
4866 if (error)
4867 goto out;
4868
4869 /* if that worked, generate a real config for the new pool */
4870 if (newspa->spa_root_vdev != NULL) {
4871 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4872 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4873 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4874 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4875 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4876 B_TRUE));
4877 }
4878
4879 /* set the props */
4880 if (props != NULL) {
4881 spa_configfile_set(newspa, props, B_FALSE);
4882 error = spa_prop_set(newspa, props);
4883 if (error)
4884 goto out;
4885 }
4886
4887 /* flush everything */
4888 txg = spa_vdev_config_enter(newspa);
4889 vdev_config_dirty(newspa->spa_root_vdev);
4890 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4891
4892 if (zio_injection_enabled)
4893 zio_handle_panic_injection(spa, FTAG, 2);
4894
4895 spa_async_resume(newspa);
4896
4897 /* finally, update the original pool's config */
4898 txg = spa_vdev_config_enter(spa);
4899 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4900 error = dmu_tx_assign(tx, TXG_WAIT);
4901 if (error != 0)
4902 dmu_tx_abort(tx);
4903 for (c = 0; c < children; c++) {
4904 if (vml[c] != NULL) {
4905 vdev_split(vml[c]);
4906 if (error == 0)
4907 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4908 spa, tx, "vdev=%s",
4909 vml[c]->vdev_path);
4910 vdev_free(vml[c]);
4911 }
4912 }
4913 vdev_config_dirty(spa->spa_root_vdev);
4914 spa->spa_config_splitting = NULL;
4915 nvlist_free(nvl);
4916 if (error == 0)
4917 dmu_tx_commit(tx);
4918 (void) spa_vdev_exit(spa, NULL, txg, 0);
4919
4920 if (zio_injection_enabled)
4921 zio_handle_panic_injection(spa, FTAG, 3);
4922
4923 /* split is complete; log a history record */
4924 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4925 "split new pool %s from pool %s", newname, spa_name(spa));
4926
4927 kmem_free(vml, children * sizeof (vdev_t *));
4928
4929 /* if we're not going to mount the filesystems in userland, export */
4930 if (exp)
4931 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4932 B_FALSE, B_FALSE);
4933
4934 return (error);
4935
4936out:
4937 spa_unload(newspa);
4938 spa_deactivate(newspa);
4939 spa_remove(newspa);
4940
4941 txg = spa_vdev_config_enter(spa);
4942
4943 /* re-online all offlined disks */
4944 for (c = 0; c < children; c++) {
4945 if (vml[c] != NULL)
4946 vml[c]->vdev_offline = B_FALSE;
4947 }
4948 vdev_reopen(spa->spa_root_vdev);
4949
4950 nvlist_free(spa->spa_config_splitting);
4951 spa->spa_config_splitting = NULL;
4952 (void) spa_vdev_exit(spa, NULL, txg, error);
4953
4954 kmem_free(vml, children * sizeof (vdev_t *));
4955 return (error);
4956}
4957
4958static nvlist_t *
4959spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4960{
4961 for (int i = 0; i < count; i++) {
4962 uint64_t guid;
4963
4964 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4965 &guid) == 0);
4966
4967 if (guid == target_guid)
4968 return (nvpp[i]);
4969 }
4970
4971 return (NULL);
4972}
4973
4974static void
4975spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4976 nvlist_t *dev_to_remove)
4977{
4978 nvlist_t **newdev = NULL;
4979
4980 if (count > 1)
4981 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4982
4983 for (int i = 0, j = 0; i < count; i++) {
4984 if (dev[i] == dev_to_remove)
4985 continue;
4986 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4987 }
4988
4989 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4990 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4991
4992 for (int i = 0; i < count - 1; i++)
4993 nvlist_free(newdev[i]);
4994
4995 if (count > 1)
4996 kmem_free(newdev, (count - 1) * sizeof (void *));
4997}
4998
4999/*
5000 * Evacuate the device.
5001 */
5002static int
5003spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5004{
5005 uint64_t txg;
5006 int error = 0;
5007
5008 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5009 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5010 ASSERT(vd == vd->vdev_top);
5011
5012 /*
5013 * Evacuate the device. We don't hold the config lock as writer
5014 * since we need to do I/O but we do keep the
5015 * spa_namespace_lock held. Once this completes the device
5016 * should no longer have any blocks allocated on it.
5017 */
5018 if (vd->vdev_islog) {
5019 if (vd->vdev_stat.vs_alloc != 0)
5020 error = spa_offline_log(spa);
5021 } else {
5022 error = ENOTSUP;
5023 }
5024
5025 if (error)
5026 return (error);
5027
5028 /*
5029 * The evacuation succeeded. Remove any remaining MOS metadata
5030 * associated with this vdev, and wait for these changes to sync.
5031 */
5032 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
5033 txg = spa_vdev_config_enter(spa);
5034 vd->vdev_removing = B_TRUE;
5035 vdev_dirty(vd, 0, NULL, txg);
5036 vdev_config_dirty(vd);
5037 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5038
5039 return (0);
5040}
5041
5042/*
5043 * Complete the removal by cleaning up the namespace.
5044 */
5045static void
5046spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5047{
5048 vdev_t *rvd = spa->spa_root_vdev;
5049 uint64_t id = vd->vdev_id;
5050 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5051
5052 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5053 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5054 ASSERT(vd == vd->vdev_top);
5055
5056 /*
5057 * Only remove any devices which are empty.
5058 */
5059 if (vd->vdev_stat.vs_alloc != 0)
5060 return;
5061
5062 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5063
5064 if (list_link_active(&vd->vdev_state_dirty_node))
5065 vdev_state_clean(vd);
5066 if (list_link_active(&vd->vdev_config_dirty_node))
5067 vdev_config_clean(vd);
5068
5069 vdev_free(vd);
5070
5071 if (last_vdev) {
5072 vdev_compact_children(rvd);
5073 } else {
5074 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5075 vdev_add_child(rvd, vd);
5076 }
5077 vdev_config_dirty(rvd);
5078
5079 /*
5080 * Reassess the health of our root vdev.
5081 */
5082 vdev_reopen(rvd);
5083}
5084
5085/*
5086 * Remove a device from the pool -
5087 *
5088 * Removing a device from the vdev namespace requires several steps
5089 * and can take a significant amount of time. As a result we use
5090 * the spa_vdev_config_[enter/exit] functions which allow us to
5091 * grab and release the spa_config_lock while still holding the namespace
5092 * lock. During each step the configuration is synced out.
5093 */
5094
5095/*
5096 * Remove a device from the pool. Currently, this supports removing only hot
5097 * spares, slogs, and level 2 ARC devices.
5098 */
5099int
5100spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5101{
5102 vdev_t *vd;
5103 metaslab_group_t *mg;
5104 nvlist_t **spares, **l2cache, *nv;
5105 uint64_t txg = 0;
5106 uint_t nspares, nl2cache;
5107 int error = 0;
5108 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5109
5110 ASSERT(spa_writeable(spa));
5111
5112 if (!locked)
5113 txg = spa_vdev_enter(spa);
5114
5115 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5116
5117 if (spa->spa_spares.sav_vdevs != NULL &&
5118 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5119 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5120 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5121 /*
5122 * Only remove the hot spare if it's not currently in use
5123 * in this pool.
5124 */
5125 if (vd == NULL || unspare) {
5126 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5127 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5128 spa_load_spares(spa);
5129 spa->spa_spares.sav_sync = B_TRUE;
5130 } else {
5131 error = EBUSY;
5132 }
5133 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5134 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5135 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5136 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5137 /*
5138 * Cache devices can always be removed.
5139 */
5140 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5141 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5142 spa_load_l2cache(spa);
5143 spa->spa_l2cache.sav_sync = B_TRUE;
5144 } else if (vd != NULL && vd->vdev_islog) {
5145 ASSERT(!locked);
5146 ASSERT(vd == vd->vdev_top);
5147
5148 /*
5149 * XXX - Once we have bp-rewrite this should
5150 * become the common case.
5151 */
5152
5153 mg = vd->vdev_mg;
5154
5155 /*
5156 * Stop allocating from this vdev.
5157 */
5158 metaslab_group_passivate(mg);
5159
5160 /*
5161 * Wait for the youngest allocations and frees to sync,
5162 * and then wait for the deferral of those frees to finish.
5163 */
5164 spa_vdev_config_exit(spa, NULL,
5165 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5166
5167 /*
5168 * Attempt to evacuate the vdev.
5169 */
5170 error = spa_vdev_remove_evacuate(spa, vd);
5171
5172 txg = spa_vdev_config_enter(spa);
5173
5174 /*
5175 * If we couldn't evacuate the vdev, unwind.
5176 */
5177 if (error) {
5178 metaslab_group_activate(mg);
5179 return (spa_vdev_exit(spa, NULL, txg, error));
5180 }
5181
5182 /*
5183 * Clean up the vdev namespace.
5184 */
5185 spa_vdev_remove_from_namespace(spa, vd);
5186
5187 } else if (vd != NULL) {
5188 /*
5189 * Normal vdevs cannot be removed (yet).
5190 */
5191 error = ENOTSUP;
5192 } else {
5193 /*
5194 * There is no vdev of any kind with the specified guid.
5195 */
5196 error = ENOENT;
5197 }
5198
5199 if (!locked)
5200 return (spa_vdev_exit(spa, NULL, txg, error));
5201
5202 return (error);
5203}
5204
5205/*
5206 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5207 * current spared, so we can detach it.
5208 */
5209static vdev_t *
5210spa_vdev_resilver_done_hunt(vdev_t *vd)
5211{
5212 vdev_t *newvd, *oldvd;
5213
5214 for (int c = 0; c < vd->vdev_children; c++) {
5215 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5216 if (oldvd != NULL)
5217 return (oldvd);
5218 }
5219
5220 /*
5221 * Check for a completed replacement. We always consider the first
5222 * vdev in the list to be the oldest vdev, and the last one to be
5223 * the newest (see spa_vdev_attach() for how that works). In
5224 * the case where the newest vdev is faulted, we will not automatically
5225 * remove it after a resilver completes. This is OK as it will require
5226 * user intervention to determine which disk the admin wishes to keep.
5227 */
5228 if (vd->vdev_ops == &vdev_replacing_ops) {
5229 ASSERT(vd->vdev_children > 1);
5230
5231 newvd = vd->vdev_child[vd->vdev_children - 1];
5232 oldvd = vd->vdev_child[0];
5233
5234 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5235 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5236 !vdev_dtl_required(oldvd))
5237 return (oldvd);
5238 }
5239
5240 /*
5241 * Check for a completed resilver with the 'unspare' flag set.
5242 */
5243 if (vd->vdev_ops == &vdev_spare_ops) {
5244 vdev_t *first = vd->vdev_child[0];
5245 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5246
5247 if (last->vdev_unspare) {
5248 oldvd = first;
5249 newvd = last;
5250 } else if (first->vdev_unspare) {
5251 oldvd = last;
5252 newvd = first;
5253 } else {
5254 oldvd = NULL;
5255 }
5256
5257 if (oldvd != NULL &&
5258 vdev_dtl_empty(newvd, DTL_MISSING) &&
5259 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5260 !vdev_dtl_required(oldvd))
5261 return (oldvd);
5262
5263 /*
5264 * If there are more than two spares attached to a disk,
5265 * and those spares are not required, then we want to
5266 * attempt to free them up now so that they can be used
5267 * by other pools. Once we're back down to a single
5268 * disk+spare, we stop removing them.
5269 */
5270 if (vd->vdev_children > 2) {
5271 newvd = vd->vdev_child[1];
5272
5273 if (newvd->vdev_isspare && last->vdev_isspare &&
5274 vdev_dtl_empty(last, DTL_MISSING) &&
5275 vdev_dtl_empty(last, DTL_OUTAGE) &&
5276 !vdev_dtl_required(newvd))
5277 return (newvd);
5278 }
5279 }
5280
5281 return (NULL);
5282}
5283
5284static void
5285spa_vdev_resilver_done(spa_t *spa)
5286{
5287 vdev_t *vd, *pvd, *ppvd;
5288 uint64_t guid, sguid, pguid, ppguid;
5289
5290 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5291
5292 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5293 pvd = vd->vdev_parent;
5294 ppvd = pvd->vdev_parent;
5295 guid = vd->vdev_guid;
5296 pguid = pvd->vdev_guid;
5297 ppguid = ppvd->vdev_guid;
5298 sguid = 0;
5299 /*
5300 * If we have just finished replacing a hot spared device, then
5301 * we need to detach the parent's first child (the original hot
5302 * spare) as well.
5303 */
5304 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5305 ppvd->vdev_children == 2) {
5306 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5307 sguid = ppvd->vdev_child[1]->vdev_guid;
5308 }
5309 spa_config_exit(spa, SCL_ALL, FTAG);
5310 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5311 return;
5312 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5313 return;
5314 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5315 }
5316
5317 spa_config_exit(spa, SCL_ALL, FTAG);
5318}
5319
5320/*
5321 * Update the stored path or FRU for this vdev.
5322 */
5323int
5324spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5325 boolean_t ispath)
5326{
5327 vdev_t *vd;
5328 boolean_t sync = B_FALSE;
5329
5330 ASSERT(spa_writeable(spa));
5331
5332 spa_vdev_state_enter(spa, SCL_ALL);
5333
5334 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5335 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5336
5337 if (!vd->vdev_ops->vdev_op_leaf)
5338 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5339
5340 if (ispath) {
5341 if (strcmp(value, vd->vdev_path) != 0) {
5342 spa_strfree(vd->vdev_path);
5343 vd->vdev_path = spa_strdup(value);
5344 sync = B_TRUE;
5345 }
5346 } else {
5347 if (vd->vdev_fru == NULL) {
5348 vd->vdev_fru = spa_strdup(value);
5349 sync = B_TRUE;
5350 } else if (strcmp(value, vd->vdev_fru) != 0) {
5351 spa_strfree(vd->vdev_fru);
5352 vd->vdev_fru = spa_strdup(value);
5353 sync = B_TRUE;
5354 }
5355 }
5356
5357 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5358}
5359
5360int
5361spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5362{
5363 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5364}
5365
5366int
5367spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5368{
5369 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5370}
5371
5372/*
5373 * ==========================================================================
5374 * SPA Scanning
5375 * ==========================================================================
5376 */
5377
5378int
5379spa_scan_stop(spa_t *spa)
5380{
5381 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5382 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5383 return (EBUSY);
5384 return (dsl_scan_cancel(spa->spa_dsl_pool));
5385}
5386
5387int
5388spa_scan(spa_t *spa, pool_scan_func_t func)
5389{
5390 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5391
5392 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5393 return (ENOTSUP);
5394
5395 /*
5396 * If a resilver was requested, but there is no DTL on a
5397 * writeable leaf device, we have nothing to do.
5398 */
5399 if (func == POOL_SCAN_RESILVER &&
5400 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5401 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5402 return (0);
5403 }
5404
5405 return (dsl_scan(spa->spa_dsl_pool, func));
5406}
5407
5408/*
5409 * ==========================================================================
5410 * SPA async task processing
5411 * ==========================================================================
5412 */
5413
5414static void
5415spa_async_remove(spa_t *spa, vdev_t *vd)
5416{
5417 if (vd->vdev_remove_wanted) {
5418 vd->vdev_remove_wanted = B_FALSE;
5419 vd->vdev_delayed_close = B_FALSE;
5420 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5421
5422 /*
5423 * We want to clear the stats, but we don't want to do a full
5424 * vdev_clear() as that will cause us to throw away
5425 * degraded/faulted state as well as attempt to reopen the
5426 * device, all of which is a waste.
5427 */
5428 vd->vdev_stat.vs_read_errors = 0;
5429 vd->vdev_stat.vs_write_errors = 0;
5430 vd->vdev_stat.vs_checksum_errors = 0;
5431
5432 vdev_state_dirty(vd->vdev_top);
5433 }
5434
5435 for (int c = 0; c < vd->vdev_children; c++)
5436 spa_async_remove(spa, vd->vdev_child[c]);
5437}
5438
5439static void
5440spa_async_probe(spa_t *spa, vdev_t *vd)
5441{
5442 if (vd->vdev_probe_wanted) {
5443 vd->vdev_probe_wanted = B_FALSE;
5444 vdev_reopen(vd); /* vdev_open() does the actual probe */
5445 }
5446
5447 for (int c = 0; c < vd->vdev_children; c++)
5448 spa_async_probe(spa, vd->vdev_child[c]);
5449}
5450
5451static void
5452spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5453{
5454 sysevent_id_t eid;
5455 nvlist_t *attr;
5456 char *physpath;
5457
5458 if (!spa->spa_autoexpand)
5459 return;
5460
5461 for (int c = 0; c < vd->vdev_children; c++) {
5462 vdev_t *cvd = vd->vdev_child[c];
5463 spa_async_autoexpand(spa, cvd);
5464 }
5465
5466 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5467 return;
5468
5469 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5470 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5471
5472 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5473 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5474
5475 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5476 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5477
5478 nvlist_free(attr);
5479 kmem_free(physpath, MAXPATHLEN);
5480}
5481
5482static void
5483spa_async_thread(void *arg)
5484{
5485 spa_t *spa = arg;
5486 int tasks;
5487
5488 ASSERT(spa->spa_sync_on);
5489
5490 mutex_enter(&spa->spa_async_lock);
5491 tasks = spa->spa_async_tasks;
5492 spa->spa_async_tasks = 0;
5493 mutex_exit(&spa->spa_async_lock);
5494
5495 /*
5496 * See if the config needs to be updated.
5497 */
5498 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5499 uint64_t old_space, new_space;
5500
5501 mutex_enter(&spa_namespace_lock);
5502 old_space = metaslab_class_get_space(spa_normal_class(spa));
5503 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5504 new_space = metaslab_class_get_space(spa_normal_class(spa));
5505 mutex_exit(&spa_namespace_lock);
5506
5507 /*
5508 * If the pool grew as a result of the config update,
5509 * then log an internal history event.
5510 */
5511 if (new_space != old_space) {
5512 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5513 spa, NULL,
5514 "pool '%s' size: %llu(+%llu)",
5515 spa_name(spa), new_space, new_space - old_space);
5516 }
5517 }
5518
5519 /*
5520 * See if any devices need to be marked REMOVED.
5521 */
5522 if (tasks & SPA_ASYNC_REMOVE) {
5523 spa_vdev_state_enter(spa, SCL_NONE);
5524 spa_async_remove(spa, spa->spa_root_vdev);
5525 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5526 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5527 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5528 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5529 (void) spa_vdev_state_exit(spa, NULL, 0);
5530 }
5531
5532 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5533 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5534 spa_async_autoexpand(spa, spa->spa_root_vdev);
5535 spa_config_exit(spa, SCL_CONFIG, FTAG);
5536 }
5537
5538 /*
5539 * See if any devices need to be probed.
5540 */
5541 if (tasks & SPA_ASYNC_PROBE) {
5542 spa_vdev_state_enter(spa, SCL_NONE);
5543 spa_async_probe(spa, spa->spa_root_vdev);
5544 (void) spa_vdev_state_exit(spa, NULL, 0);
5545 }
5546
5547 /*
5548 * If any devices are done replacing, detach them.
5549 */
5550 if (tasks & SPA_ASYNC_RESILVER_DONE)
5551 spa_vdev_resilver_done(spa);
5552
5553 /*
5554 * Kick off a resilver.
5555 */
5556 if (tasks & SPA_ASYNC_RESILVER)
5557 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5558
5559 /*
5560 * Let the world know that we're done.
5561 */
5562 mutex_enter(&spa->spa_async_lock);
5563 spa->spa_async_thread = NULL;
5564 cv_broadcast(&spa->spa_async_cv);
5565 mutex_exit(&spa->spa_async_lock);
5566 thread_exit();
5567}
5568
5569void
5570spa_async_suspend(spa_t *spa)
5571{
5572 mutex_enter(&spa->spa_async_lock);
5573 spa->spa_async_suspended++;
5574 while (spa->spa_async_thread != NULL)
5575 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5576 mutex_exit(&spa->spa_async_lock);
5577}
5578
5579void
5580spa_async_resume(spa_t *spa)
5581{
5582 mutex_enter(&spa->spa_async_lock);
5583 ASSERT(spa->spa_async_suspended != 0);
5584 spa->spa_async_suspended--;
5585 mutex_exit(&spa->spa_async_lock);
5586}
5587
5588static void
5589spa_async_dispatch(spa_t *spa)
5590{
5591 mutex_enter(&spa->spa_async_lock);
5592 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5593 spa->spa_async_thread == NULL &&
5594 rootdir != NULL && !vn_is_readonly(rootdir))
5595 spa->spa_async_thread = thread_create(NULL, 0,
5596 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5597 mutex_exit(&spa->spa_async_lock);
5598}
5599
5600void
5601spa_async_request(spa_t *spa, int task)
5602{
5603 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5604 mutex_enter(&spa->spa_async_lock);
5605 spa->spa_async_tasks |= task;
5606 mutex_exit(&spa->spa_async_lock);
5607}
5608
5609/*
5610 * ==========================================================================
5611 * SPA syncing routines
5612 * ==========================================================================
5613 */
5614
5615static int
5616bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5617{
5618 bpobj_t *bpo = arg;
5619 bpobj_enqueue(bpo, bp, tx);
5620 return (0);
5621}
5622
5623static int
5624spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5625{
5626 zio_t *zio = arg;
5627
5628 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5629 zio->io_flags));
5630 return (0);
5631}
5632
5633static void
5634spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5635{
5636 char *packed = NULL;
5637 size_t bufsize;
5638 size_t nvsize = 0;
5639 dmu_buf_t *db;
5640
5641 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5642
5643 /*
5644 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5645 * information. This avoids the dbuf_will_dirty() path and
5646 * saves us a pre-read to get data we don't actually care about.
5647 */
5648 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5649 packed = kmem_alloc(bufsize, KM_SLEEP);
5650
5651 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5652 KM_SLEEP) == 0);
5653 bzero(packed + nvsize, bufsize - nvsize);
5654
5655 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5656
5657 kmem_free(packed, bufsize);
5658
5659 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5660 dmu_buf_will_dirty(db, tx);
5661 *(uint64_t *)db->db_data = nvsize;
5662 dmu_buf_rele(db, FTAG);
5663}
5664
5665static void
5666spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5667 const char *config, const char *entry)
5668{
5669 nvlist_t *nvroot;
5670 nvlist_t **list;
5671 int i;
5672
5673 if (!sav->sav_sync)
5674 return;
5675
5676 /*
5677 * Update the MOS nvlist describing the list of available devices.
5678 * spa_validate_aux() will have already made sure this nvlist is
5679 * valid and the vdevs are labeled appropriately.
5680 */
5681 if (sav->sav_object == 0) {
5682 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5683 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5684 sizeof (uint64_t), tx);
5685 VERIFY(zap_update(spa->spa_meta_objset,
5686 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5687 &sav->sav_object, tx) == 0);
5688 }
5689
5690 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5691 if (sav->sav_count == 0) {
5692 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5693 } else {
5694 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5695 for (i = 0; i < sav->sav_count; i++)
5696 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5697 B_FALSE, VDEV_CONFIG_L2CACHE);
5698 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5699 sav->sav_count) == 0);
5700 for (i = 0; i < sav->sav_count; i++)
5701 nvlist_free(list[i]);
5702 kmem_free(list, sav->sav_count * sizeof (void *));
5703 }
5704
5705 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5706 nvlist_free(nvroot);
5707
5708 sav->sav_sync = B_FALSE;
5709}
5710
5711static void
5712spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5713{
5714 nvlist_t *config;
5715
5716 if (list_is_empty(&spa->spa_config_dirty_list))
5717 return;
5718
5719 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5720
5721 config = spa_config_generate(spa, spa->spa_root_vdev,
5722 dmu_tx_get_txg(tx), B_FALSE);
5723
5724 spa_config_exit(spa, SCL_STATE, FTAG);
5725
5726 if (spa->spa_config_syncing)
5727 nvlist_free(spa->spa_config_syncing);
5728 spa->spa_config_syncing = config;
5729
5730 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5731}
5732
5733static void
5734spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5735{
5736 spa_t *spa = arg1;
5737 uint64_t version = *(uint64_t *)arg2;
5738
5739 /*
5740 * Setting the version is special cased when first creating the pool.
5741 */
5742 ASSERT(tx->tx_txg != TXG_INITIAL);
5743
5744 ASSERT(version <= SPA_VERSION);
5745 ASSERT(version >= spa_version(spa));
5746
5747 spa->spa_uberblock.ub_version = version;
5748 vdev_config_dirty(spa->spa_root_vdev);
5749}
5750
5751/*
5752 * Set zpool properties.
5753 */
5754static void
5755spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5756{
5757 spa_t *spa = arg1;
5758 objset_t *mos = spa->spa_meta_objset;
5759 nvlist_t *nvp = arg2;
5760 nvpair_t *elem = NULL;
5761
5762 mutex_enter(&spa->spa_props_lock);
5763
5764 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5765 uint64_t intval;
5766 char *strval, *fname;
5767 zpool_prop_t prop;
5768 const char *propname;
5769 zprop_type_t proptype;
5770 zfeature_info_t *feature;
5771
5772 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5773 case ZPROP_INVAL:
5774 /*
5775 * We checked this earlier in spa_prop_validate().
5776 */
5777 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5778
5779 fname = strchr(nvpair_name(elem), '@') + 1;
5780 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5781
5782 spa_feature_enable(spa, feature, tx);
5783 break;
5784
5785 case ZPOOL_PROP_VERSION:
5786 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5787 /*
5788 * The version is synced seperatly before other
5789 * properties and should be correct by now.
5790 */
5791 ASSERT3U(spa_version(spa), >=, intval);
5792 break;
5793
5794 case ZPOOL_PROP_ALTROOT:
5795 /*
5796 * 'altroot' is a non-persistent property. It should
5797 * have been set temporarily at creation or import time.
5798 */
5799 ASSERT(spa->spa_root != NULL);
5800 break;
5801
5802 case ZPOOL_PROP_READONLY:
5803 case ZPOOL_PROP_CACHEFILE:
5804 /*
5805 * 'readonly' and 'cachefile' are also non-persisitent
5806 * properties.
5807 */
5808 break;
5809 case ZPOOL_PROP_COMMENT:
5810 VERIFY(nvpair_value_string(elem, &strval) == 0);
5811 if (spa->spa_comment != NULL)
5812 spa_strfree(spa->spa_comment);
5813 spa->spa_comment = spa_strdup(strval);
5814 /*
5815 * We need to dirty the configuration on all the vdevs
5816 * so that their labels get updated. It's unnecessary
5817 * to do this for pool creation since the vdev's
5818 * configuratoin has already been dirtied.
5819 */
5820 if (tx->tx_txg != TXG_INITIAL)
5821 vdev_config_dirty(spa->spa_root_vdev);
5822 break;
5823 default:
5824 /*
5825 * Set pool property values in the poolprops mos object.
5826 */
5827 if (spa->spa_pool_props_object == 0) {
5828 spa->spa_pool_props_object =
5829 zap_create_link(mos, DMU_OT_POOL_PROPS,
5830 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5831 tx);
5832 }
5833
5834 /* normalize the property name */
5835 propname = zpool_prop_to_name(prop);
5836 proptype = zpool_prop_get_type(prop);
5837
5838 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5839 ASSERT(proptype == PROP_TYPE_STRING);
5840 VERIFY(nvpair_value_string(elem, &strval) == 0);
5841 VERIFY(zap_update(mos,
5842 spa->spa_pool_props_object, propname,
5843 1, strlen(strval) + 1, strval, tx) == 0);
5844
5845 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5846 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5847
5848 if (proptype == PROP_TYPE_INDEX) {
5849 const char *unused;
5850 VERIFY(zpool_prop_index_to_string(
5851 prop, intval, &unused) == 0);
5852 }
5853 VERIFY(zap_update(mos,
5854 spa->spa_pool_props_object, propname,
5855 8, 1, &intval, tx) == 0);
5856 } else {
5857 ASSERT(0); /* not allowed */
5858 }
5859
5860 switch (prop) {
5861 case ZPOOL_PROP_DELEGATION:
5862 spa->spa_delegation = intval;
5863 break;
5864 case ZPOOL_PROP_BOOTFS:
5865 spa->spa_bootfs = intval;
5866 break;
5867 case ZPOOL_PROP_FAILUREMODE:
5868 spa->spa_failmode = intval;
5869 break;
5870 case ZPOOL_PROP_AUTOEXPAND:
5871 spa->spa_autoexpand = intval;
5872 if (tx->tx_txg != TXG_INITIAL)
5873 spa_async_request(spa,
5874 SPA_ASYNC_AUTOEXPAND);
5875 break;
5876 case ZPOOL_PROP_DEDUPDITTO:
5877 spa->spa_dedup_ditto = intval;
5878 break;
5879 default:
5880 break;
5881 }
5882 }
5883
5884 /* log internal history if this is not a zpool create */
5885 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5886 tx->tx_txg != TXG_INITIAL) {
5887 spa_history_log_internal(LOG_POOL_PROPSET,
5888 spa, tx, "%s %lld %s",
5889 nvpair_name(elem), intval, spa_name(spa));
5890 }
5891 }
5892
5893 mutex_exit(&spa->spa_props_lock);
5894}
5895
5896/*
5897 * Perform one-time upgrade on-disk changes. spa_version() does not
5898 * reflect the new version this txg, so there must be no changes this
5899 * txg to anything that the upgrade code depends on after it executes.
5900 * Therefore this must be called after dsl_pool_sync() does the sync
5901 * tasks.
5902 */
5903static void
5904spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5905{
5906 dsl_pool_t *dp = spa->spa_dsl_pool;
5907
5908 ASSERT(spa->spa_sync_pass == 1);
5909
5910 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5911 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5912 dsl_pool_create_origin(dp, tx);
5913
5914 /* Keeping the origin open increases spa_minref */
5915 spa->spa_minref += 3;
5916 }
5917
5918 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5919 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5920 dsl_pool_upgrade_clones(dp, tx);
5921 }
5922
5923 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5924 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5925 dsl_pool_upgrade_dir_clones(dp, tx);
5926
5927 /* Keeping the freedir open increases spa_minref */
5928 spa->spa_minref += 3;
5929 }
5930
5931 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5932 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5933 spa_feature_create_zap_objects(spa, tx);
5934 }
5935}
5936
5937/*
5938 * Sync the specified transaction group. New blocks may be dirtied as
5939 * part of the process, so we iterate until it converges.
5940 */
5941void
5942spa_sync(spa_t *spa, uint64_t txg)
5943{
5944 dsl_pool_t *dp = spa->spa_dsl_pool;
5945 objset_t *mos = spa->spa_meta_objset;
5946 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5947 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5948 vdev_t *rvd = spa->spa_root_vdev;
5949 vdev_t *vd;
5950 dmu_tx_t *tx;
5951 int error;
5952
5953 VERIFY(spa_writeable(spa));
5954
5955 /*
5956 * Lock out configuration changes.
5957 */
5958 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5959
5960 spa->spa_syncing_txg = txg;
5961 spa->spa_sync_pass = 0;
5962
5963 /*
5964 * If there are any pending vdev state changes, convert them
5965 * into config changes that go out with this transaction group.
5966 */
5967 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5968 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5969 /*
5970 * We need the write lock here because, for aux vdevs,
5971 * calling vdev_config_dirty() modifies sav_config.
5972 * This is ugly and will become unnecessary when we
5973 * eliminate the aux vdev wart by integrating all vdevs
5974 * into the root vdev tree.
5975 */
5976 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5977 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5978 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5979 vdev_state_clean(vd);
5980 vdev_config_dirty(vd);
5981 }
5982 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5983 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5984 }
5985 spa_config_exit(spa, SCL_STATE, FTAG);
5986
5987 tx = dmu_tx_create_assigned(dp, txg);
5988
5989 /*
5990 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5991 * set spa_deflate if we have no raid-z vdevs.
5992 */
5993 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5994 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5995 int i;
5996
5997 for (i = 0; i < rvd->vdev_children; i++) {
5998 vd = rvd->vdev_child[i];
5999 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6000 break;
6001 }
6002 if (i == rvd->vdev_children) {
6003 spa->spa_deflate = TRUE;
6004 VERIFY(0 == zap_add(spa->spa_meta_objset,
6005 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6006 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6007 }
6008 }
6009
6010 /*
6011 * If anything has changed in this txg, or if someone is waiting
6012 * for this txg to sync (eg, spa_vdev_remove()), push the
6013 * deferred frees from the previous txg. If not, leave them
6014 * alone so that we don't generate work on an otherwise idle
6015 * system.
6016 */
6017 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6018 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6019 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6020 ((dsl_scan_active(dp->dp_scan) ||
6021 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6022 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6023 VERIFY3U(bpobj_iterate(defer_bpo,
6024 spa_free_sync_cb, zio, tx), ==, 0);
6025 VERIFY3U(zio_wait(zio), ==, 0);
6026 }
6027
6028 /*
6029 * Iterate to convergence.
6030 */
6031 do {
6032 int pass = ++spa->spa_sync_pass;
6033
6034 spa_sync_config_object(spa, tx);
6035 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6036 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6037 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6038 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6039 spa_errlog_sync(spa, txg);
6040 dsl_pool_sync(dp, txg);
6041
6042 if (pass <= SYNC_PASS_DEFERRED_FREE) {
6043 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6044 bplist_iterate(free_bpl, spa_free_sync_cb,
6045 zio, tx);
6046 VERIFY(zio_wait(zio) == 0);
6047 } else {
6048 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6049 defer_bpo, tx);
6050 }
6051
6052 ddt_sync(spa, txg);
6053 dsl_scan_sync(dp, tx);
6054
6055 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6056 vdev_sync(vd, txg);
6057
6058 if (pass == 1)
6059 spa_sync_upgrades(spa, tx);
6060
6061 } while (dmu_objset_is_dirty(mos, txg));
6062
6063 /*
6064 * Rewrite the vdev configuration (which includes the uberblock)
6065 * to commit the transaction group.
6066 *
6067 * If there are no dirty vdevs, we sync the uberblock to a few
6068 * random top-level vdevs that are known to be visible in the
6069 * config cache (see spa_vdev_add() for a complete description).
6070 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6071 */
6072 for (;;) {
6073 /*
6074 * We hold SCL_STATE to prevent vdev open/close/etc.
6075 * while we're attempting to write the vdev labels.
6076 */
6077 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6078
6079 if (list_is_empty(&spa->spa_config_dirty_list)) {
6080 vdev_t *svd[SPA_DVAS_PER_BP];
6081 int svdcount = 0;
6082 int children = rvd->vdev_children;
6083 int c0 = spa_get_random(children);
6084
6085 for (int c = 0; c < children; c++) {
6086 vd = rvd->vdev_child[(c0 + c) % children];
6087 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6088 continue;
6089 svd[svdcount++] = vd;
6090 if (svdcount == SPA_DVAS_PER_BP)
6091 break;
6092 }
6093 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6094 if (error != 0)
6095 error = vdev_config_sync(svd, svdcount, txg,
6096 B_TRUE);
6097 } else {
6098 error = vdev_config_sync(rvd->vdev_child,
6099 rvd->vdev_children, txg, B_FALSE);
6100 if (error != 0)
6101 error = vdev_config_sync(rvd->vdev_child,
6102 rvd->vdev_children, txg, B_TRUE);
6103 }
6104
6105 spa_config_exit(spa, SCL_STATE, FTAG);
6106
6107 if (error == 0)
6108 break;
6109 zio_suspend(spa, NULL);
6110 zio_resume_wait(spa);
6111 }
6112 dmu_tx_commit(tx);
6113
6114 /*
6115 * Clear the dirty config list.
6116 */
6117 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6118 vdev_config_clean(vd);
6119
6120 /*
6121 * Now that the new config has synced transactionally,
6122 * let it become visible to the config cache.
6123 */
6124 if (spa->spa_config_syncing != NULL) {
6125 spa_config_set(spa, spa->spa_config_syncing);
6126 spa->spa_config_txg = txg;
6127 spa->spa_config_syncing = NULL;
6128 }
6129
6130 spa->spa_ubsync = spa->spa_uberblock;
6131
6132 dsl_pool_sync_done(dp, txg);
6133
6134 /*
6135 * Update usable space statistics.
6136 */
6137 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6138 vdev_sync_done(vd, txg);
6139
6140 spa_update_dspace(spa);
6141
6142 /*
6143 * It had better be the case that we didn't dirty anything
6144 * since vdev_config_sync().
6145 */
6146 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6147 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6148 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6149
6150 spa->spa_sync_pass = 0;
6151
6152 spa_config_exit(spa, SCL_CONFIG, FTAG);
6153
6154 spa_handle_ignored_writes(spa);
6155
6156 /*
6157 * If any async tasks have been requested, kick them off.
6158 */
6159 spa_async_dispatch(spa);
6160}
6161
6162/*
6163 * Sync all pools. We don't want to hold the namespace lock across these
6164 * operations, so we take a reference on the spa_t and drop the lock during the
6165 * sync.
6166 */
6167void
6168spa_sync_allpools(void)
6169{
6170 spa_t *spa = NULL;
6171 mutex_enter(&spa_namespace_lock);
6172 while ((spa = spa_next(spa)) != NULL) {
6173 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6174 !spa_writeable(spa) || spa_suspended(spa))
6175 continue;
6176 spa_open_ref(spa, FTAG);
6177 mutex_exit(&spa_namespace_lock);
6178 txg_wait_synced(spa_get_dsl(spa), 0);
6179 mutex_enter(&spa_namespace_lock);
6180 spa_close(spa, FTAG);
6181 }
6182 mutex_exit(&spa_namespace_lock);
6183}
6184
6185/*
6186 * ==========================================================================
6187 * Miscellaneous routines
6188 * ==========================================================================
6189 */
6190
6191/*
6192 * Remove all pools in the system.
6193 */
6194void
6195spa_evict_all(void)
6196{
6197 spa_t *spa;
6198
6199 /*
6200 * Remove all cached state. All pools should be closed now,
6201 * so every spa in the AVL tree should be unreferenced.
6202 */
6203 mutex_enter(&spa_namespace_lock);
6204 while ((spa = spa_next(NULL)) != NULL) {
6205 /*
6206 * Stop async tasks. The async thread may need to detach
6207 * a device that's been replaced, which requires grabbing
6208 * spa_namespace_lock, so we must drop it here.
6209 */
6210 spa_open_ref(spa, FTAG);
6211 mutex_exit(&spa_namespace_lock);
6212 spa_async_suspend(spa);
6213 mutex_enter(&spa_namespace_lock);
6214 spa_close(spa, FTAG);
6215
6216 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6217 spa_unload(spa);
6218 spa_deactivate(spa);
6219 }
6220 spa_remove(spa);
6221 }
6222 mutex_exit(&spa_namespace_lock);
6223}
6224
6225vdev_t *
6226spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6227{
6228 vdev_t *vd;
6229 int i;
6230
6231 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6232 return (vd);
6233
6234 if (aux) {
6235 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6236 vd = spa->spa_l2cache.sav_vdevs[i];
6237 if (vd->vdev_guid == guid)
6238 return (vd);
6239 }
6240
6241 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6242 vd = spa->spa_spares.sav_vdevs[i];
6243 if (vd->vdev_guid == guid)
6244 return (vd);
6245 }
6246 }
6247
6248 return (NULL);
6249}
6250
6251void
6252spa_upgrade(spa_t *spa, uint64_t version)
6253{
6254 ASSERT(spa_writeable(spa));
6255
6256 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6257
6258 /*
6259 * This should only be called for a non-faulted pool, and since a
6260 * future version would result in an unopenable pool, this shouldn't be
6261 * possible.
6262 */
6263 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6264 ASSERT(version >= spa->spa_uberblock.ub_version);
6265
6266 spa->spa_uberblock.ub_version = version;
6267 vdev_config_dirty(spa->spa_root_vdev);
6268
6269 spa_config_exit(spa, SCL_ALL, FTAG);
6270
6271 txg_wait_synced(spa_get_dsl(spa), 0);
6272}
6273
6274boolean_t
6275spa_has_spare(spa_t *spa, uint64_t guid)
6276{
6277 int i;
6278 uint64_t spareguid;
6279 spa_aux_vdev_t *sav = &spa->spa_spares;
6280
6281 for (i = 0; i < sav->sav_count; i++)
6282 if (sav->sav_vdevs[i]->vdev_guid == guid)
6283 return (B_TRUE);
6284
6285 for (i = 0; i < sav->sav_npending; i++) {
6286 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6287 &spareguid) == 0 && spareguid == guid)
6288 return (B_TRUE);
6289 }
6290
6291 return (B_FALSE);
6292}
6293
6294/*
6295 * Check if a pool has an active shared spare device.
6296 * Note: reference count of an active spare is 2, as a spare and as a replace
6297 */
6298static boolean_t
6299spa_has_active_shared_spare(spa_t *spa)
6300{
6301 int i, refcnt;
6302 uint64_t pool;
6303 spa_aux_vdev_t *sav = &spa->spa_spares;
6304
6305 for (i = 0; i < sav->sav_count; i++) {
6306 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6307 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6308 refcnt > 2)
6309 return (B_TRUE);
6310 }
6311
6312 return (B_FALSE);
6313}
6314
6315/*
6316 * Post a sysevent corresponding to the given event. The 'name' must be one of
6317 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6318 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6319 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6320 * or zdb as real changes.
6321 */
6322void
6323spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6324{
6325#ifdef _KERNEL
6326 sysevent_t *ev;
6327 sysevent_attr_list_t *attr = NULL;
6328 sysevent_value_t value;
6329 sysevent_id_t eid;
6330
6331 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6332 SE_SLEEP);
6333
6334 value.value_type = SE_DATA_TYPE_STRING;
6335 value.value.sv_string = spa_name(spa);
6336 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6337 goto done;
6338
6339 value.value_type = SE_DATA_TYPE_UINT64;
6340 value.value.sv_uint64 = spa_guid(spa);
6341 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6342 goto done;
6343
6344 if (vd) {
6345 value.value_type = SE_DATA_TYPE_UINT64;
6346 value.value.sv_uint64 = vd->vdev_guid;
6347 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6348 SE_SLEEP) != 0)
6349 goto done;
6350
6351 if (vd->vdev_path) {
6352 value.value_type = SE_DATA_TYPE_STRING;
6353 value.value.sv_string = vd->vdev_path;
6354 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6355 &value, SE_SLEEP) != 0)
6356 goto done;
6357 }
6358 }
6359
6360 if (sysevent_attach_attributes(ev, attr) != 0)
6361 goto done;
6362 attr = NULL;
6363
6364 (void) log_sysevent(ev, SE_SLEEP, &eid);
6365
6366done:
6367 if (attr)
6368 sysevent_free_attr(attr);
6369 sysevent_free(ev);
6370#endif
6371}
| 2218
2219 if (!missing_feat_read) {
2220 fnvlist_add_boolean(spa->spa_load_info,
2221 ZPOOL_CONFIG_CAN_RDONLY);
2222 }
2223
2224 /*
2225 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2226 * twofold: to determine whether the pool is available for
2227 * import in read-write mode and (if it is not) whether the
2228 * pool is available for import in read-only mode. If the pool
2229 * is available for import in read-write mode, it is displayed
2230 * as available in userland; if it is not available for import
2231 * in read-only mode, it is displayed as unavailable in
2232 * userland. If the pool is available for import in read-only
2233 * mode but not read-write mode, it is displayed as unavailable
2234 * in userland with a special note that the pool is actually
2235 * available for open in read-only mode.
2236 *
2237 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2238 * missing a feature for write, we must first determine whether
2239 * the pool can be opened read-only before returning to
2240 * userland in order to know whether to display the
2241 * abovementioned note.
2242 */
2243 if (missing_feat_read || (missing_feat_write &&
2244 spa_writeable(spa))) {
2245 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2246 ENOTSUP));
2247 }
2248 }
2249
2250 spa->spa_is_initializing = B_TRUE;
2251 error = dsl_pool_open(spa->spa_dsl_pool);
2252 spa->spa_is_initializing = B_FALSE;
2253 if (error != 0)
2254 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2255
2256 if (!mosconfig) {
2257 uint64_t hostid;
2258 nvlist_t *policy = NULL, *nvconfig;
2259
2260 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2261 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2262
2263 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2264 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2265 char *hostname;
2266 unsigned long myhostid = 0;
2267
2268 VERIFY(nvlist_lookup_string(nvconfig,
2269 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2270
2271#ifdef _KERNEL
2272 myhostid = zone_get_hostid(NULL);
2273#else /* _KERNEL */
2274 /*
2275 * We're emulating the system's hostid in userland, so
2276 * we can't use zone_get_hostid().
2277 */
2278 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2279#endif /* _KERNEL */
2280 if (check_hostid && hostid != 0 && myhostid != 0 &&
2281 hostid != myhostid) {
2282 nvlist_free(nvconfig);
2283 cmn_err(CE_WARN, "pool '%s' could not be "
2284 "loaded as it was last accessed by "
2285 "another system (host: %s hostid: 0x%lx). "
2286 "See: http://illumos.org/msg/ZFS-8000-EY",
2287 spa_name(spa), hostname,
2288 (unsigned long)hostid);
2289 return (EBADF);
2290 }
2291 }
2292 if (nvlist_lookup_nvlist(spa->spa_config,
2293 ZPOOL_REWIND_POLICY, &policy) == 0)
2294 VERIFY(nvlist_add_nvlist(nvconfig,
2295 ZPOOL_REWIND_POLICY, policy) == 0);
2296
2297 spa_config_set(spa, nvconfig);
2298 spa_unload(spa);
2299 spa_deactivate(spa);
2300 spa_activate(spa, orig_mode);
2301
2302 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2303 }
2304
2305 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2306 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2307 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2308 if (error != 0)
2309 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2310
2311 /*
2312 * Load the bit that tells us to use the new accounting function
2313 * (raid-z deflation). If we have an older pool, this will not
2314 * be present.
2315 */
2316 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2317 if (error != 0 && error != ENOENT)
2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2319
2320 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2321 &spa->spa_creation_version);
2322 if (error != 0 && error != ENOENT)
2323 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2324
2325 /*
2326 * Load the persistent error log. If we have an older pool, this will
2327 * not be present.
2328 */
2329 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2330 if (error != 0 && error != ENOENT)
2331 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2332
2333 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2334 &spa->spa_errlog_scrub);
2335 if (error != 0 && error != ENOENT)
2336 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2337
2338 /*
2339 * Load the history object. If we have an older pool, this
2340 * will not be present.
2341 */
2342 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2343 if (error != 0 && error != ENOENT)
2344 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2345
2346 /*
2347 * If we're assembling the pool from the split-off vdevs of
2348 * an existing pool, we don't want to attach the spares & cache
2349 * devices.
2350 */
2351
2352 /*
2353 * Load any hot spares for this pool.
2354 */
2355 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2356 if (error != 0 && error != ENOENT)
2357 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2358 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2359 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2360 if (load_nvlist(spa, spa->spa_spares.sav_object,
2361 &spa->spa_spares.sav_config) != 0)
2362 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2363
2364 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2365 spa_load_spares(spa);
2366 spa_config_exit(spa, SCL_ALL, FTAG);
2367 } else if (error == 0) {
2368 spa->spa_spares.sav_sync = B_TRUE;
2369 }
2370
2371 /*
2372 * Load any level 2 ARC devices for this pool.
2373 */
2374 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2375 &spa->spa_l2cache.sav_object);
2376 if (error != 0 && error != ENOENT)
2377 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2378 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2379 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2380 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2381 &spa->spa_l2cache.sav_config) != 0)
2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2383
2384 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2385 spa_load_l2cache(spa);
2386 spa_config_exit(spa, SCL_ALL, FTAG);
2387 } else if (error == 0) {
2388 spa->spa_l2cache.sav_sync = B_TRUE;
2389 }
2390
2391 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2392
2393 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2394 if (error && error != ENOENT)
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2396
2397 if (error == 0) {
2398 uint64_t autoreplace;
2399
2400 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2401 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2402 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2403 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2404 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2405 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2406 &spa->spa_dedup_ditto);
2407
2408 spa->spa_autoreplace = (autoreplace != 0);
2409 }
2410
2411 /*
2412 * If the 'autoreplace' property is set, then post a resource notifying
2413 * the ZFS DE that it should not issue any faults for unopenable
2414 * devices. We also iterate over the vdevs, and post a sysevent for any
2415 * unopenable vdevs so that the normal autoreplace handler can take
2416 * over.
2417 */
2418 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2419 spa_check_removed(spa->spa_root_vdev);
2420 /*
2421 * For the import case, this is done in spa_import(), because
2422 * at this point we're using the spare definitions from
2423 * the MOS config, not necessarily from the userland config.
2424 */
2425 if (state != SPA_LOAD_IMPORT) {
2426 spa_aux_check_removed(&spa->spa_spares);
2427 spa_aux_check_removed(&spa->spa_l2cache);
2428 }
2429 }
2430
2431 /*
2432 * Load the vdev state for all toplevel vdevs.
2433 */
2434 vdev_load(rvd);
2435
2436 /*
2437 * Propagate the leaf DTLs we just loaded all the way up the tree.
2438 */
2439 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2440 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2441 spa_config_exit(spa, SCL_ALL, FTAG);
2442
2443 /*
2444 * Load the DDTs (dedup tables).
2445 */
2446 error = ddt_load(spa);
2447 if (error != 0)
2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2449
2450 spa_update_dspace(spa);
2451
2452 /*
2453 * Validate the config, using the MOS config to fill in any
2454 * information which might be missing. If we fail to validate
2455 * the config then declare the pool unfit for use. If we're
2456 * assembling a pool from a split, the log is not transferred
2457 * over.
2458 */
2459 if (type != SPA_IMPORT_ASSEMBLE) {
2460 nvlist_t *nvconfig;
2461
2462 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2464
2465 if (!spa_config_valid(spa, nvconfig)) {
2466 nvlist_free(nvconfig);
2467 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2468 ENXIO));
2469 }
2470 nvlist_free(nvconfig);
2471
2472 /*
2473 * Now that we've validated the config, check the state of the
2474 * root vdev. If it can't be opened, it indicates one or
2475 * more toplevel vdevs are faulted.
2476 */
2477 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2478 return (ENXIO);
2479
2480 if (spa_check_logs(spa)) {
2481 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2482 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2483 }
2484 }
2485
2486 if (missing_feat_write) {
2487 ASSERT(state == SPA_LOAD_TRYIMPORT);
2488
2489 /*
2490 * At this point, we know that we can open the pool in
2491 * read-only mode but not read-write mode. We now have enough
2492 * information and can return to userland.
2493 */
2494 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2495 }
2496
2497 /*
2498 * We've successfully opened the pool, verify that we're ready
2499 * to start pushing transactions.
2500 */
2501 if (state != SPA_LOAD_TRYIMPORT) {
2502 if (error = spa_load_verify(spa))
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2504 error));
2505 }
2506
2507 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2508 spa->spa_load_max_txg == UINT64_MAX)) {
2509 dmu_tx_t *tx;
2510 int need_update = B_FALSE;
2511
2512 ASSERT(state != SPA_LOAD_TRYIMPORT);
2513
2514 /*
2515 * Claim log blocks that haven't been committed yet.
2516 * This must all happen in a single txg.
2517 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2518 * invoked from zil_claim_log_block()'s i/o done callback.
2519 * Price of rollback is that we abandon the log.
2520 */
2521 spa->spa_claiming = B_TRUE;
2522
2523 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2524 spa_first_txg(spa));
2525 (void) dmu_objset_find(spa_name(spa),
2526 zil_claim, tx, DS_FIND_CHILDREN);
2527 dmu_tx_commit(tx);
2528
2529 spa->spa_claiming = B_FALSE;
2530
2531 spa_set_log_state(spa, SPA_LOG_GOOD);
2532 spa->spa_sync_on = B_TRUE;
2533 txg_sync_start(spa->spa_dsl_pool);
2534
2535 /*
2536 * Wait for all claims to sync. We sync up to the highest
2537 * claimed log block birth time so that claimed log blocks
2538 * don't appear to be from the future. spa_claim_max_txg
2539 * will have been set for us by either zil_check_log_chain()
2540 * (invoked from spa_check_logs()) or zil_claim() above.
2541 */
2542 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2543
2544 /*
2545 * If the config cache is stale, or we have uninitialized
2546 * metaslabs (see spa_vdev_add()), then update the config.
2547 *
2548 * If this is a verbatim import, trust the current
2549 * in-core spa_config and update the disk labels.
2550 */
2551 if (config_cache_txg != spa->spa_config_txg ||
2552 state == SPA_LOAD_IMPORT ||
2553 state == SPA_LOAD_RECOVER ||
2554 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2555 need_update = B_TRUE;
2556
2557 for (int c = 0; c < rvd->vdev_children; c++)
2558 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2559 need_update = B_TRUE;
2560
2561 /*
2562 * Update the config cache asychronously in case we're the
2563 * root pool, in which case the config cache isn't writable yet.
2564 */
2565 if (need_update)
2566 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2567
2568 /*
2569 * Check all DTLs to see if anything needs resilvering.
2570 */
2571 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2572 vdev_resilver_needed(rvd, NULL, NULL))
2573 spa_async_request(spa, SPA_ASYNC_RESILVER);
2574
2575 /*
2576 * Delete any inconsistent datasets.
2577 */
2578 (void) dmu_objset_find(spa_name(spa),
2579 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2580
2581 /*
2582 * Clean up any stale temporary dataset userrefs.
2583 */
2584 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2585 }
2586
2587 return (0);
2588}
2589
2590static int
2591spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2592{
2593 int mode = spa->spa_mode;
2594
2595 spa_unload(spa);
2596 spa_deactivate(spa);
2597
2598 spa->spa_load_max_txg--;
2599
2600 spa_activate(spa, mode);
2601 spa_async_suspend(spa);
2602
2603 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2604}
2605
2606/*
2607 * If spa_load() fails this function will try loading prior txg's. If
2608 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2609 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2610 * function will not rewind the pool and will return the same error as
2611 * spa_load().
2612 */
2613static int
2614spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2615 uint64_t max_request, int rewind_flags)
2616{
2617 nvlist_t *loadinfo = NULL;
2618 nvlist_t *config = NULL;
2619 int load_error, rewind_error;
2620 uint64_t safe_rewind_txg;
2621 uint64_t min_txg;
2622
2623 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2624 spa->spa_load_max_txg = spa->spa_load_txg;
2625 spa_set_log_state(spa, SPA_LOG_CLEAR);
2626 } else {
2627 spa->spa_load_max_txg = max_request;
2628 }
2629
2630 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2631 mosconfig);
2632 if (load_error == 0)
2633 return (0);
2634
2635 if (spa->spa_root_vdev != NULL)
2636 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2637
2638 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2639 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2640
2641 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2642 nvlist_free(config);
2643 return (load_error);
2644 }
2645
2646 if (state == SPA_LOAD_RECOVER) {
2647 /* Price of rolling back is discarding txgs, including log */
2648 spa_set_log_state(spa, SPA_LOG_CLEAR);
2649 } else {
2650 /*
2651 * If we aren't rolling back save the load info from our first
2652 * import attempt so that we can restore it after attempting
2653 * to rewind.
2654 */
2655 loadinfo = spa->spa_load_info;
2656 spa->spa_load_info = fnvlist_alloc();
2657 }
2658
2659 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2660 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2661 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2662 TXG_INITIAL : safe_rewind_txg;
2663
2664 /*
2665 * Continue as long as we're finding errors, we're still within
2666 * the acceptable rewind range, and we're still finding uberblocks
2667 */
2668 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2669 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2670 if (spa->spa_load_max_txg < safe_rewind_txg)
2671 spa->spa_extreme_rewind = B_TRUE;
2672 rewind_error = spa_load_retry(spa, state, mosconfig);
2673 }
2674
2675 spa->spa_extreme_rewind = B_FALSE;
2676 spa->spa_load_max_txg = UINT64_MAX;
2677
2678 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2679 spa_config_set(spa, config);
2680
2681 if (state == SPA_LOAD_RECOVER) {
2682 ASSERT3P(loadinfo, ==, NULL);
2683 return (rewind_error);
2684 } else {
2685 /* Store the rewind info as part of the initial load info */
2686 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2687 spa->spa_load_info);
2688
2689 /* Restore the initial load info */
2690 fnvlist_free(spa->spa_load_info);
2691 spa->spa_load_info = loadinfo;
2692
2693 return (load_error);
2694 }
2695}
2696
2697/*
2698 * Pool Open/Import
2699 *
2700 * The import case is identical to an open except that the configuration is sent
2701 * down from userland, instead of grabbed from the configuration cache. For the
2702 * case of an open, the pool configuration will exist in the
2703 * POOL_STATE_UNINITIALIZED state.
2704 *
2705 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2706 * the same time open the pool, without having to keep around the spa_t in some
2707 * ambiguous state.
2708 */
2709static int
2710spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2711 nvlist_t **config)
2712{
2713 spa_t *spa;
2714 spa_load_state_t state = SPA_LOAD_OPEN;
2715 int error;
2716 int locked = B_FALSE;
2717 int firstopen = B_FALSE;
2718
2719 *spapp = NULL;
2720
2721 /*
2722 * As disgusting as this is, we need to support recursive calls to this
2723 * function because dsl_dir_open() is called during spa_load(), and ends
2724 * up calling spa_open() again. The real fix is to figure out how to
2725 * avoid dsl_dir_open() calling this in the first place.
2726 */
2727 if (mutex_owner(&spa_namespace_lock) != curthread) {
2728 mutex_enter(&spa_namespace_lock);
2729 locked = B_TRUE;
2730 }
2731
2732 if ((spa = spa_lookup(pool)) == NULL) {
2733 if (locked)
2734 mutex_exit(&spa_namespace_lock);
2735 return (ENOENT);
2736 }
2737
2738 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2739 zpool_rewind_policy_t policy;
2740
2741 firstopen = B_TRUE;
2742
2743 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2744 &policy);
2745 if (policy.zrp_request & ZPOOL_DO_REWIND)
2746 state = SPA_LOAD_RECOVER;
2747
2748 spa_activate(spa, spa_mode_global);
2749
2750 if (state != SPA_LOAD_RECOVER)
2751 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2752
2753 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2754 policy.zrp_request);
2755
2756 if (error == EBADF) {
2757 /*
2758 * If vdev_validate() returns failure (indicated by
2759 * EBADF), it indicates that one of the vdevs indicates
2760 * that the pool has been exported or destroyed. If
2761 * this is the case, the config cache is out of sync and
2762 * we should remove the pool from the namespace.
2763 */
2764 spa_unload(spa);
2765 spa_deactivate(spa);
2766 spa_config_sync(spa, B_TRUE, B_TRUE);
2767 spa_remove(spa);
2768 if (locked)
2769 mutex_exit(&spa_namespace_lock);
2770 return (ENOENT);
2771 }
2772
2773 if (error) {
2774 /*
2775 * We can't open the pool, but we still have useful
2776 * information: the state of each vdev after the
2777 * attempted vdev_open(). Return this to the user.
2778 */
2779 if (config != NULL && spa->spa_config) {
2780 VERIFY(nvlist_dup(spa->spa_config, config,
2781 KM_SLEEP) == 0);
2782 VERIFY(nvlist_add_nvlist(*config,
2783 ZPOOL_CONFIG_LOAD_INFO,
2784 spa->spa_load_info) == 0);
2785 }
2786 spa_unload(spa);
2787 spa_deactivate(spa);
2788 spa->spa_last_open_failed = error;
2789 if (locked)
2790 mutex_exit(&spa_namespace_lock);
2791 *spapp = NULL;
2792 return (error);
2793 }
2794 }
2795
2796 spa_open_ref(spa, tag);
2797
2798 if (config != NULL)
2799 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2800
2801 /*
2802 * If we've recovered the pool, pass back any information we
2803 * gathered while doing the load.
2804 */
2805 if (state == SPA_LOAD_RECOVER) {
2806 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2807 spa->spa_load_info) == 0);
2808 }
2809
2810 if (locked) {
2811 spa->spa_last_open_failed = 0;
2812 spa->spa_last_ubsync_txg = 0;
2813 spa->spa_load_txg = 0;
2814 mutex_exit(&spa_namespace_lock);
2815#ifdef __FreeBSD__
2816#ifdef _KERNEL
2817 if (firstopen)
2818 zvol_create_minors(pool);
2819#endif
2820#endif
2821 }
2822
2823 *spapp = spa;
2824
2825 return (0);
2826}
2827
2828int
2829spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2830 nvlist_t **config)
2831{
2832 return (spa_open_common(name, spapp, tag, policy, config));
2833}
2834
2835int
2836spa_open(const char *name, spa_t **spapp, void *tag)
2837{
2838 return (spa_open_common(name, spapp, tag, NULL, NULL));
2839}
2840
2841/*
2842 * Lookup the given spa_t, incrementing the inject count in the process,
2843 * preventing it from being exported or destroyed.
2844 */
2845spa_t *
2846spa_inject_addref(char *name)
2847{
2848 spa_t *spa;
2849
2850 mutex_enter(&spa_namespace_lock);
2851 if ((spa = spa_lookup(name)) == NULL) {
2852 mutex_exit(&spa_namespace_lock);
2853 return (NULL);
2854 }
2855 spa->spa_inject_ref++;
2856 mutex_exit(&spa_namespace_lock);
2857
2858 return (spa);
2859}
2860
2861void
2862spa_inject_delref(spa_t *spa)
2863{
2864 mutex_enter(&spa_namespace_lock);
2865 spa->spa_inject_ref--;
2866 mutex_exit(&spa_namespace_lock);
2867}
2868
2869/*
2870 * Add spares device information to the nvlist.
2871 */
2872static void
2873spa_add_spares(spa_t *spa, nvlist_t *config)
2874{
2875 nvlist_t **spares;
2876 uint_t i, nspares;
2877 nvlist_t *nvroot;
2878 uint64_t guid;
2879 vdev_stat_t *vs;
2880 uint_t vsc;
2881 uint64_t pool;
2882
2883 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2884
2885 if (spa->spa_spares.sav_count == 0)
2886 return;
2887
2888 VERIFY(nvlist_lookup_nvlist(config,
2889 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2890 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2891 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2892 if (nspares != 0) {
2893 VERIFY(nvlist_add_nvlist_array(nvroot,
2894 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2895 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2896 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2897
2898 /*
2899 * Go through and find any spares which have since been
2900 * repurposed as an active spare. If this is the case, update
2901 * their status appropriately.
2902 */
2903 for (i = 0; i < nspares; i++) {
2904 VERIFY(nvlist_lookup_uint64(spares[i],
2905 ZPOOL_CONFIG_GUID, &guid) == 0);
2906 if (spa_spare_exists(guid, &pool, NULL) &&
2907 pool != 0ULL) {
2908 VERIFY(nvlist_lookup_uint64_array(
2909 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2910 (uint64_t **)&vs, &vsc) == 0);
2911 vs->vs_state = VDEV_STATE_CANT_OPEN;
2912 vs->vs_aux = VDEV_AUX_SPARED;
2913 }
2914 }
2915 }
2916}
2917
2918/*
2919 * Add l2cache device information to the nvlist, including vdev stats.
2920 */
2921static void
2922spa_add_l2cache(spa_t *spa, nvlist_t *config)
2923{
2924 nvlist_t **l2cache;
2925 uint_t i, j, nl2cache;
2926 nvlist_t *nvroot;
2927 uint64_t guid;
2928 vdev_t *vd;
2929 vdev_stat_t *vs;
2930 uint_t vsc;
2931
2932 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2933
2934 if (spa->spa_l2cache.sav_count == 0)
2935 return;
2936
2937 VERIFY(nvlist_lookup_nvlist(config,
2938 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2939 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2940 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2941 if (nl2cache != 0) {
2942 VERIFY(nvlist_add_nvlist_array(nvroot,
2943 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2944 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2945 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2946
2947 /*
2948 * Update level 2 cache device stats.
2949 */
2950
2951 for (i = 0; i < nl2cache; i++) {
2952 VERIFY(nvlist_lookup_uint64(l2cache[i],
2953 ZPOOL_CONFIG_GUID, &guid) == 0);
2954
2955 vd = NULL;
2956 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2957 if (guid ==
2958 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2959 vd = spa->spa_l2cache.sav_vdevs[j];
2960 break;
2961 }
2962 }
2963 ASSERT(vd != NULL);
2964
2965 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2966 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2967 == 0);
2968 vdev_get_stats(vd, vs);
2969 }
2970 }
2971}
2972
2973static void
2974spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2975{
2976 nvlist_t *features;
2977 zap_cursor_t zc;
2978 zap_attribute_t za;
2979
2980 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2981 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2982
2983 if (spa->spa_feat_for_read_obj != 0) {
2984 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2985 spa->spa_feat_for_read_obj);
2986 zap_cursor_retrieve(&zc, &za) == 0;
2987 zap_cursor_advance(&zc)) {
2988 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2989 za.za_num_integers == 1);
2990 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2991 za.za_first_integer));
2992 }
2993 zap_cursor_fini(&zc);
2994 }
2995
2996 if (spa->spa_feat_for_write_obj != 0) {
2997 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2998 spa->spa_feat_for_write_obj);
2999 zap_cursor_retrieve(&zc, &za) == 0;
3000 zap_cursor_advance(&zc)) {
3001 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3002 za.za_num_integers == 1);
3003 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3004 za.za_first_integer));
3005 }
3006 zap_cursor_fini(&zc);
3007 }
3008
3009 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3010 features) == 0);
3011 nvlist_free(features);
3012}
3013
3014int
3015spa_get_stats(const char *name, nvlist_t **config,
3016 char *altroot, size_t buflen)
3017{
3018 int error;
3019 spa_t *spa;
3020
3021 *config = NULL;
3022 error = spa_open_common(name, &spa, FTAG, NULL, config);
3023
3024 if (spa != NULL) {
3025 /*
3026 * This still leaves a window of inconsistency where the spares
3027 * or l2cache devices could change and the config would be
3028 * self-inconsistent.
3029 */
3030 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3031
3032 if (*config != NULL) {
3033 uint64_t loadtimes[2];
3034
3035 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3036 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3037 VERIFY(nvlist_add_uint64_array(*config,
3038 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3039
3040 VERIFY(nvlist_add_uint64(*config,
3041 ZPOOL_CONFIG_ERRCOUNT,
3042 spa_get_errlog_size(spa)) == 0);
3043
3044 if (spa_suspended(spa))
3045 VERIFY(nvlist_add_uint64(*config,
3046 ZPOOL_CONFIG_SUSPENDED,
3047 spa->spa_failmode) == 0);
3048
3049 spa_add_spares(spa, *config);
3050 spa_add_l2cache(spa, *config);
3051 spa_add_feature_stats(spa, *config);
3052 }
3053 }
3054
3055 /*
3056 * We want to get the alternate root even for faulted pools, so we cheat
3057 * and call spa_lookup() directly.
3058 */
3059 if (altroot) {
3060 if (spa == NULL) {
3061 mutex_enter(&spa_namespace_lock);
3062 spa = spa_lookup(name);
3063 if (spa)
3064 spa_altroot(spa, altroot, buflen);
3065 else
3066 altroot[0] = '\0';
3067 spa = NULL;
3068 mutex_exit(&spa_namespace_lock);
3069 } else {
3070 spa_altroot(spa, altroot, buflen);
3071 }
3072 }
3073
3074 if (spa != NULL) {
3075 spa_config_exit(spa, SCL_CONFIG, FTAG);
3076 spa_close(spa, FTAG);
3077 }
3078
3079 return (error);
3080}
3081
3082/*
3083 * Validate that the auxiliary device array is well formed. We must have an
3084 * array of nvlists, each which describes a valid leaf vdev. If this is an
3085 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3086 * specified, as long as they are well-formed.
3087 */
3088static int
3089spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3090 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3091 vdev_labeltype_t label)
3092{
3093 nvlist_t **dev;
3094 uint_t i, ndev;
3095 vdev_t *vd;
3096 int error;
3097
3098 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3099
3100 /*
3101 * It's acceptable to have no devs specified.
3102 */
3103 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3104 return (0);
3105
3106 if (ndev == 0)
3107 return (EINVAL);
3108
3109 /*
3110 * Make sure the pool is formatted with a version that supports this
3111 * device type.
3112 */
3113 if (spa_version(spa) < version)
3114 return (ENOTSUP);
3115
3116 /*
3117 * Set the pending device list so we correctly handle device in-use
3118 * checking.
3119 */
3120 sav->sav_pending = dev;
3121 sav->sav_npending = ndev;
3122
3123 for (i = 0; i < ndev; i++) {
3124 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3125 mode)) != 0)
3126 goto out;
3127
3128 if (!vd->vdev_ops->vdev_op_leaf) {
3129 vdev_free(vd);
3130 error = EINVAL;
3131 goto out;
3132 }
3133
3134 /*
3135 * The L2ARC currently only supports disk devices in
3136 * kernel context. For user-level testing, we allow it.
3137 */
3138#ifdef _KERNEL
3139 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3140 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3141 error = ENOTBLK;
3142 vdev_free(vd);
3143 goto out;
3144 }
3145#endif
3146 vd->vdev_top = vd;
3147
3148 if ((error = vdev_open(vd)) == 0 &&
3149 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3150 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3151 vd->vdev_guid) == 0);
3152 }
3153
3154 vdev_free(vd);
3155
3156 if (error &&
3157 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3158 goto out;
3159 else
3160 error = 0;
3161 }
3162
3163out:
3164 sav->sav_pending = NULL;
3165 sav->sav_npending = 0;
3166 return (error);
3167}
3168
3169static int
3170spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3171{
3172 int error;
3173
3174 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3175
3176 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3177 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3178 VDEV_LABEL_SPARE)) != 0) {
3179 return (error);
3180 }
3181
3182 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3183 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3184 VDEV_LABEL_L2CACHE));
3185}
3186
3187static void
3188spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3189 const char *config)
3190{
3191 int i;
3192
3193 if (sav->sav_config != NULL) {
3194 nvlist_t **olddevs;
3195 uint_t oldndevs;
3196 nvlist_t **newdevs;
3197
3198 /*
3199 * Generate new dev list by concatentating with the
3200 * current dev list.
3201 */
3202 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3203 &olddevs, &oldndevs) == 0);
3204
3205 newdevs = kmem_alloc(sizeof (void *) *
3206 (ndevs + oldndevs), KM_SLEEP);
3207 for (i = 0; i < oldndevs; i++)
3208 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3209 KM_SLEEP) == 0);
3210 for (i = 0; i < ndevs; i++)
3211 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3212 KM_SLEEP) == 0);
3213
3214 VERIFY(nvlist_remove(sav->sav_config, config,
3215 DATA_TYPE_NVLIST_ARRAY) == 0);
3216
3217 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3218 config, newdevs, ndevs + oldndevs) == 0);
3219 for (i = 0; i < oldndevs + ndevs; i++)
3220 nvlist_free(newdevs[i]);
3221 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3222 } else {
3223 /*
3224 * Generate a new dev list.
3225 */
3226 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3227 KM_SLEEP) == 0);
3228 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3229 devs, ndevs) == 0);
3230 }
3231}
3232
3233/*
3234 * Stop and drop level 2 ARC devices
3235 */
3236void
3237spa_l2cache_drop(spa_t *spa)
3238{
3239 vdev_t *vd;
3240 int i;
3241 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3242
3243 for (i = 0; i < sav->sav_count; i++) {
3244 uint64_t pool;
3245
3246 vd = sav->sav_vdevs[i];
3247 ASSERT(vd != NULL);
3248
3249 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3250 pool != 0ULL && l2arc_vdev_present(vd))
3251 l2arc_remove_vdev(vd);
3252 }
3253}
3254
3255/*
3256 * Pool Creation
3257 */
3258int
3259spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3260 const char *history_str, nvlist_t *zplprops)
3261{
3262 spa_t *spa;
3263 char *altroot = NULL;
3264 vdev_t *rvd;
3265 dsl_pool_t *dp;
3266 dmu_tx_t *tx;
3267 int error = 0;
3268 uint64_t txg = TXG_INITIAL;
3269 nvlist_t **spares, **l2cache;
3270 uint_t nspares, nl2cache;
3271 uint64_t version, obj;
3272 boolean_t has_features;
3273
3274 /*
3275 * If this pool already exists, return failure.
3276 */
3277 mutex_enter(&spa_namespace_lock);
3278 if (spa_lookup(pool) != NULL) {
3279 mutex_exit(&spa_namespace_lock);
3280 return (EEXIST);
3281 }
3282
3283 /*
3284 * Allocate a new spa_t structure.
3285 */
3286 (void) nvlist_lookup_string(props,
3287 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3288 spa = spa_add(pool, NULL, altroot);
3289 spa_activate(spa, spa_mode_global);
3290
3291 if (props && (error = spa_prop_validate(spa, props))) {
3292 spa_deactivate(spa);
3293 spa_remove(spa);
3294 mutex_exit(&spa_namespace_lock);
3295 return (error);
3296 }
3297
3298 has_features = B_FALSE;
3299 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3300 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3301 if (zpool_prop_feature(nvpair_name(elem)))
3302 has_features = B_TRUE;
3303 }
3304
3305 if (has_features || nvlist_lookup_uint64(props,
3306 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3307 version = SPA_VERSION;
3308 }
3309 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3310
3311 spa->spa_first_txg = txg;
3312 spa->spa_uberblock.ub_txg = txg - 1;
3313 spa->spa_uberblock.ub_version = version;
3314 spa->spa_ubsync = spa->spa_uberblock;
3315
3316 /*
3317 * Create "The Godfather" zio to hold all async IOs
3318 */
3319 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3320 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3321
3322 /*
3323 * Create the root vdev.
3324 */
3325 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3326
3327 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3328
3329 ASSERT(error != 0 || rvd != NULL);
3330 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3331
3332 if (error == 0 && !zfs_allocatable_devs(nvroot))
3333 error = EINVAL;
3334
3335 if (error == 0 &&
3336 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3337 (error = spa_validate_aux(spa, nvroot, txg,
3338 VDEV_ALLOC_ADD)) == 0) {
3339 for (int c = 0; c < rvd->vdev_children; c++) {
3340 vdev_metaslab_set_size(rvd->vdev_child[c]);
3341 vdev_expand(rvd->vdev_child[c], txg);
3342 }
3343 }
3344
3345 spa_config_exit(spa, SCL_ALL, FTAG);
3346
3347 if (error != 0) {
3348 spa_unload(spa);
3349 spa_deactivate(spa);
3350 spa_remove(spa);
3351 mutex_exit(&spa_namespace_lock);
3352 return (error);
3353 }
3354
3355 /*
3356 * Get the list of spares, if specified.
3357 */
3358 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3359 &spares, &nspares) == 0) {
3360 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3361 KM_SLEEP) == 0);
3362 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3363 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3364 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3365 spa_load_spares(spa);
3366 spa_config_exit(spa, SCL_ALL, FTAG);
3367 spa->spa_spares.sav_sync = B_TRUE;
3368 }
3369
3370 /*
3371 * Get the list of level 2 cache devices, if specified.
3372 */
3373 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3374 &l2cache, &nl2cache) == 0) {
3375 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3376 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3377 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3378 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3379 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3380 spa_load_l2cache(spa);
3381 spa_config_exit(spa, SCL_ALL, FTAG);
3382 spa->spa_l2cache.sav_sync = B_TRUE;
3383 }
3384
3385 spa->spa_is_initializing = B_TRUE;
3386 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3387 spa->spa_meta_objset = dp->dp_meta_objset;
3388 spa->spa_is_initializing = B_FALSE;
3389
3390 /*
3391 * Create DDTs (dedup tables).
3392 */
3393 ddt_create(spa);
3394
3395 spa_update_dspace(spa);
3396
3397 tx = dmu_tx_create_assigned(dp, txg);
3398
3399 /*
3400 * Create the pool config object.
3401 */
3402 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3403 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3404 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3405
3406 if (zap_add(spa->spa_meta_objset,
3407 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3408 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3409 cmn_err(CE_PANIC, "failed to add pool config");
3410 }
3411
3412 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3413 spa_feature_create_zap_objects(spa, tx);
3414
3415 if (zap_add(spa->spa_meta_objset,
3416 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3417 sizeof (uint64_t), 1, &version, tx) != 0) {
3418 cmn_err(CE_PANIC, "failed to add pool version");
3419 }
3420
3421 /* Newly created pools with the right version are always deflated. */
3422 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3423 spa->spa_deflate = TRUE;
3424 if (zap_add(spa->spa_meta_objset,
3425 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3426 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3427 cmn_err(CE_PANIC, "failed to add deflate");
3428 }
3429 }
3430
3431 /*
3432 * Create the deferred-free bpobj. Turn off compression
3433 * because sync-to-convergence takes longer if the blocksize
3434 * keeps changing.
3435 */
3436 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3437 dmu_object_set_compress(spa->spa_meta_objset, obj,
3438 ZIO_COMPRESS_OFF, tx);
3439 if (zap_add(spa->spa_meta_objset,
3440 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3441 sizeof (uint64_t), 1, &obj, tx) != 0) {
3442 cmn_err(CE_PANIC, "failed to add bpobj");
3443 }
3444 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3445 spa->spa_meta_objset, obj));
3446
3447 /*
3448 * Create the pool's history object.
3449 */
3450 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3451 spa_history_create_obj(spa, tx);
3452
3453 /*
3454 * Set pool properties.
3455 */
3456 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3457 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3458 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3459 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3460
3461 if (props != NULL) {
3462 spa_configfile_set(spa, props, B_FALSE);
3463 spa_sync_props(spa, props, tx);
3464 }
3465
3466 dmu_tx_commit(tx);
3467
3468 spa->spa_sync_on = B_TRUE;
3469 txg_sync_start(spa->spa_dsl_pool);
3470
3471 /*
3472 * We explicitly wait for the first transaction to complete so that our
3473 * bean counters are appropriately updated.
3474 */
3475 txg_wait_synced(spa->spa_dsl_pool, txg);
3476
3477 spa_config_sync(spa, B_FALSE, B_TRUE);
3478
3479 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3480 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3481 spa_history_log_version(spa, LOG_POOL_CREATE);
3482
3483 spa->spa_minref = refcount_count(&spa->spa_refcount);
3484
3485 mutex_exit(&spa_namespace_lock);
3486
3487 return (0);
3488}
3489
3490#if defined(sun)
3491#ifdef _KERNEL
3492/*
3493 * Get the root pool information from the root disk, then import the root pool
3494 * during the system boot up time.
3495 */
3496extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3497
3498static nvlist_t *
3499spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3500{
3501 nvlist_t *config;
3502 nvlist_t *nvtop, *nvroot;
3503 uint64_t pgid;
3504
3505 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3506 return (NULL);
3507
3508 /*
3509 * Add this top-level vdev to the child array.
3510 */
3511 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3512 &nvtop) == 0);
3513 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3514 &pgid) == 0);
3515 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3516
3517 /*
3518 * Put this pool's top-level vdevs into a root vdev.
3519 */
3520 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3521 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3522 VDEV_TYPE_ROOT) == 0);
3523 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3524 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3525 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3526 &nvtop, 1) == 0);
3527
3528 /*
3529 * Replace the existing vdev_tree with the new root vdev in
3530 * this pool's configuration (remove the old, add the new).
3531 */
3532 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3533 nvlist_free(nvroot);
3534 return (config);
3535}
3536
3537/*
3538 * Walk the vdev tree and see if we can find a device with "better"
3539 * configuration. A configuration is "better" if the label on that
3540 * device has a more recent txg.
3541 */
3542static void
3543spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3544{
3545 for (int c = 0; c < vd->vdev_children; c++)
3546 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3547
3548 if (vd->vdev_ops->vdev_op_leaf) {
3549 nvlist_t *label;
3550 uint64_t label_txg;
3551
3552 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3553 &label) != 0)
3554 return;
3555
3556 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3557 &label_txg) == 0);
3558
3559 /*
3560 * Do we have a better boot device?
3561 */
3562 if (label_txg > *txg) {
3563 *txg = label_txg;
3564 *avd = vd;
3565 }
3566 nvlist_free(label);
3567 }
3568}
3569
3570/*
3571 * Import a root pool.
3572 *
3573 * For x86. devpath_list will consist of devid and/or physpath name of
3574 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3575 * The GRUB "findroot" command will return the vdev we should boot.
3576 *
3577 * For Sparc, devpath_list consists the physpath name of the booting device
3578 * no matter the rootpool is a single device pool or a mirrored pool.
3579 * e.g.
3580 * "/pci@1f,0/ide@d/disk@0,0:a"
3581 */
3582int
3583spa_import_rootpool(char *devpath, char *devid)
3584{
3585 spa_t *spa;
3586 vdev_t *rvd, *bvd, *avd = NULL;
3587 nvlist_t *config, *nvtop;
3588 uint64_t guid, txg;
3589 char *pname;
3590 int error;
3591
3592 /*
3593 * Read the label from the boot device and generate a configuration.
3594 */
3595 config = spa_generate_rootconf(devpath, devid, &guid);
3596#if defined(_OBP) && defined(_KERNEL)
3597 if (config == NULL) {
3598 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3599 /* iscsi boot */
3600 get_iscsi_bootpath_phy(devpath);
3601 config = spa_generate_rootconf(devpath, devid, &guid);
3602 }
3603 }
3604#endif
3605 if (config == NULL) {
3606 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3607 devpath);
3608 return (EIO);
3609 }
3610
3611 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3612 &pname) == 0);
3613 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3614
3615 mutex_enter(&spa_namespace_lock);
3616 if ((spa = spa_lookup(pname)) != NULL) {
3617 /*
3618 * Remove the existing root pool from the namespace so that we
3619 * can replace it with the correct config we just read in.
3620 */
3621 spa_remove(spa);
3622 }
3623
3624 spa = spa_add(pname, config, NULL);
3625 spa->spa_is_root = B_TRUE;
3626 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3627
3628 /*
3629 * Build up a vdev tree based on the boot device's label config.
3630 */
3631 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3632 &nvtop) == 0);
3633 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3634 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3635 VDEV_ALLOC_ROOTPOOL);
3636 spa_config_exit(spa, SCL_ALL, FTAG);
3637 if (error) {
3638 mutex_exit(&spa_namespace_lock);
3639 nvlist_free(config);
3640 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3641 pname);
3642 return (error);
3643 }
3644
3645 /*
3646 * Get the boot vdev.
3647 */
3648 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3649 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3650 (u_longlong_t)guid);
3651 error = ENOENT;
3652 goto out;
3653 }
3654
3655 /*
3656 * Determine if there is a better boot device.
3657 */
3658 avd = bvd;
3659 spa_alt_rootvdev(rvd, &avd, &txg);
3660 if (avd != bvd) {
3661 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3662 "try booting from '%s'", avd->vdev_path);
3663 error = EINVAL;
3664 goto out;
3665 }
3666
3667 /*
3668 * If the boot device is part of a spare vdev then ensure that
3669 * we're booting off the active spare.
3670 */
3671 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3672 !bvd->vdev_isspare) {
3673 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3674 "try booting from '%s'",
3675 bvd->vdev_parent->
3676 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3677 error = EINVAL;
3678 goto out;
3679 }
3680
3681 error = 0;
3682 spa_history_log_version(spa, LOG_POOL_IMPORT);
3683out:
3684 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3685 vdev_free(rvd);
3686 spa_config_exit(spa, SCL_ALL, FTAG);
3687 mutex_exit(&spa_namespace_lock);
3688
3689 nvlist_free(config);
3690 return (error);
3691}
3692
3693#endif
3694#endif /* sun */
3695
3696/*
3697 * Import a non-root pool into the system.
3698 */
3699int
3700spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3701{
3702 spa_t *spa;
3703 char *altroot = NULL;
3704 spa_load_state_t state = SPA_LOAD_IMPORT;
3705 zpool_rewind_policy_t policy;
3706 uint64_t mode = spa_mode_global;
3707 uint64_t readonly = B_FALSE;
3708 int error;
3709 nvlist_t *nvroot;
3710 nvlist_t **spares, **l2cache;
3711 uint_t nspares, nl2cache;
3712
3713 /*
3714 * If a pool with this name exists, return failure.
3715 */
3716 mutex_enter(&spa_namespace_lock);
3717 if (spa_lookup(pool) != NULL) {
3718 mutex_exit(&spa_namespace_lock);
3719 return (EEXIST);
3720 }
3721
3722 /*
3723 * Create and initialize the spa structure.
3724 */
3725 (void) nvlist_lookup_string(props,
3726 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3727 (void) nvlist_lookup_uint64(props,
3728 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3729 if (readonly)
3730 mode = FREAD;
3731 spa = spa_add(pool, config, altroot);
3732 spa->spa_import_flags = flags;
3733
3734 /*
3735 * Verbatim import - Take a pool and insert it into the namespace
3736 * as if it had been loaded at boot.
3737 */
3738 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3739 if (props != NULL)
3740 spa_configfile_set(spa, props, B_FALSE);
3741
3742 spa_config_sync(spa, B_FALSE, B_TRUE);
3743
3744 mutex_exit(&spa_namespace_lock);
3745 spa_history_log_version(spa, LOG_POOL_IMPORT);
3746
3747 return (0);
3748 }
3749
3750 spa_activate(spa, mode);
3751
3752 /*
3753 * Don't start async tasks until we know everything is healthy.
3754 */
3755 spa_async_suspend(spa);
3756
3757 zpool_get_rewind_policy(config, &policy);
3758 if (policy.zrp_request & ZPOOL_DO_REWIND)
3759 state = SPA_LOAD_RECOVER;
3760
3761 /*
3762 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3763 * because the user-supplied config is actually the one to trust when
3764 * doing an import.
3765 */
3766 if (state != SPA_LOAD_RECOVER)
3767 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3768
3769 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3770 policy.zrp_request);
3771
3772 /*
3773 * Propagate anything learned while loading the pool and pass it
3774 * back to caller (i.e. rewind info, missing devices, etc).
3775 */
3776 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3777 spa->spa_load_info) == 0);
3778
3779 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3780 /*
3781 * Toss any existing sparelist, as it doesn't have any validity
3782 * anymore, and conflicts with spa_has_spare().
3783 */
3784 if (spa->spa_spares.sav_config) {
3785 nvlist_free(spa->spa_spares.sav_config);
3786 spa->spa_spares.sav_config = NULL;
3787 spa_load_spares(spa);
3788 }
3789 if (spa->spa_l2cache.sav_config) {
3790 nvlist_free(spa->spa_l2cache.sav_config);
3791 spa->spa_l2cache.sav_config = NULL;
3792 spa_load_l2cache(spa);
3793 }
3794
3795 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3796 &nvroot) == 0);
3797 if (error == 0)
3798 error = spa_validate_aux(spa, nvroot, -1ULL,
3799 VDEV_ALLOC_SPARE);
3800 if (error == 0)
3801 error = spa_validate_aux(spa, nvroot, -1ULL,
3802 VDEV_ALLOC_L2CACHE);
3803 spa_config_exit(spa, SCL_ALL, FTAG);
3804
3805 if (props != NULL)
3806 spa_configfile_set(spa, props, B_FALSE);
3807
3808 if (error != 0 || (props && spa_writeable(spa) &&
3809 (error = spa_prop_set(spa, props)))) {
3810 spa_unload(spa);
3811 spa_deactivate(spa);
3812 spa_remove(spa);
3813 mutex_exit(&spa_namespace_lock);
3814 return (error);
3815 }
3816
3817 spa_async_resume(spa);
3818
3819 /*
3820 * Override any spares and level 2 cache devices as specified by
3821 * the user, as these may have correct device names/devids, etc.
3822 */
3823 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3824 &spares, &nspares) == 0) {
3825 if (spa->spa_spares.sav_config)
3826 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3827 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3828 else
3829 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3830 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3831 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3832 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3833 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3834 spa_load_spares(spa);
3835 spa_config_exit(spa, SCL_ALL, FTAG);
3836 spa->spa_spares.sav_sync = B_TRUE;
3837 }
3838 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3839 &l2cache, &nl2cache) == 0) {
3840 if (spa->spa_l2cache.sav_config)
3841 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3842 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3843 else
3844 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3845 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3846 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3847 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3848 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3849 spa_load_l2cache(spa);
3850 spa_config_exit(spa, SCL_ALL, FTAG);
3851 spa->spa_l2cache.sav_sync = B_TRUE;
3852 }
3853
3854 /*
3855 * Check for any removed devices.
3856 */
3857 if (spa->spa_autoreplace) {
3858 spa_aux_check_removed(&spa->spa_spares);
3859 spa_aux_check_removed(&spa->spa_l2cache);
3860 }
3861
3862 if (spa_writeable(spa)) {
3863 /*
3864 * Update the config cache to include the newly-imported pool.
3865 */
3866 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3867 }
3868
3869 /*
3870 * It's possible that the pool was expanded while it was exported.
3871 * We kick off an async task to handle this for us.
3872 */
3873 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3874
3875 mutex_exit(&spa_namespace_lock);
3876 spa_history_log_version(spa, LOG_POOL_IMPORT);
3877
3878#ifdef __FreeBSD__
3879#ifdef _KERNEL
3880 zvol_create_minors(pool);
3881#endif
3882#endif
3883 return (0);
3884}
3885
3886nvlist_t *
3887spa_tryimport(nvlist_t *tryconfig)
3888{
3889 nvlist_t *config = NULL;
3890 char *poolname;
3891 spa_t *spa;
3892 uint64_t state;
3893 int error;
3894
3895 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3896 return (NULL);
3897
3898 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3899 return (NULL);
3900
3901 /*
3902 * Create and initialize the spa structure.
3903 */
3904 mutex_enter(&spa_namespace_lock);
3905 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3906 spa_activate(spa, FREAD);
3907
3908 /*
3909 * Pass off the heavy lifting to spa_load().
3910 * Pass TRUE for mosconfig because the user-supplied config
3911 * is actually the one to trust when doing an import.
3912 */
3913 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3914
3915 /*
3916 * If 'tryconfig' was at least parsable, return the current config.
3917 */
3918 if (spa->spa_root_vdev != NULL) {
3919 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3920 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3921 poolname) == 0);
3922 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3923 state) == 0);
3924 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3925 spa->spa_uberblock.ub_timestamp) == 0);
3926 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3927 spa->spa_load_info) == 0);
3928
3929 /*
3930 * If the bootfs property exists on this pool then we
3931 * copy it out so that external consumers can tell which
3932 * pools are bootable.
3933 */
3934 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3935 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3936
3937 /*
3938 * We have to play games with the name since the
3939 * pool was opened as TRYIMPORT_NAME.
3940 */
3941 if (dsl_dsobj_to_dsname(spa_name(spa),
3942 spa->spa_bootfs, tmpname) == 0) {
3943 char *cp;
3944 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3945
3946 cp = strchr(tmpname, '/');
3947 if (cp == NULL) {
3948 (void) strlcpy(dsname, tmpname,
3949 MAXPATHLEN);
3950 } else {
3951 (void) snprintf(dsname, MAXPATHLEN,
3952 "%s/%s", poolname, ++cp);
3953 }
3954 VERIFY(nvlist_add_string(config,
3955 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3956 kmem_free(dsname, MAXPATHLEN);
3957 }
3958 kmem_free(tmpname, MAXPATHLEN);
3959 }
3960
3961 /*
3962 * Add the list of hot spares and level 2 cache devices.
3963 */
3964 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3965 spa_add_spares(spa, config);
3966 spa_add_l2cache(spa, config);
3967 spa_config_exit(spa, SCL_CONFIG, FTAG);
3968 }
3969
3970 spa_unload(spa);
3971 spa_deactivate(spa);
3972 spa_remove(spa);
3973 mutex_exit(&spa_namespace_lock);
3974
3975 return (config);
3976}
3977
3978/*
3979 * Pool export/destroy
3980 *
3981 * The act of destroying or exporting a pool is very simple. We make sure there
3982 * is no more pending I/O and any references to the pool are gone. Then, we
3983 * update the pool state and sync all the labels to disk, removing the
3984 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3985 * we don't sync the labels or remove the configuration cache.
3986 */
3987static int
3988spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3989 boolean_t force, boolean_t hardforce)
3990{
3991 spa_t *spa;
3992
3993 if (oldconfig)
3994 *oldconfig = NULL;
3995
3996 if (!(spa_mode_global & FWRITE))
3997 return (EROFS);
3998
3999 mutex_enter(&spa_namespace_lock);
4000 if ((spa = spa_lookup(pool)) == NULL) {
4001 mutex_exit(&spa_namespace_lock);
4002 return (ENOENT);
4003 }
4004
4005 /*
4006 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4007 * reacquire the namespace lock, and see if we can export.
4008 */
4009 spa_open_ref(spa, FTAG);
4010 mutex_exit(&spa_namespace_lock);
4011 spa_async_suspend(spa);
4012 mutex_enter(&spa_namespace_lock);
4013 spa_close(spa, FTAG);
4014
4015 /*
4016 * The pool will be in core if it's openable,
4017 * in which case we can modify its state.
4018 */
4019 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4020 /*
4021 * Objsets may be open only because they're dirty, so we
4022 * have to force it to sync before checking spa_refcnt.
4023 */
4024 txg_wait_synced(spa->spa_dsl_pool, 0);
4025
4026 /*
4027 * A pool cannot be exported or destroyed if there are active
4028 * references. If we are resetting a pool, allow references by
4029 * fault injection handlers.
4030 */
4031 if (!spa_refcount_zero(spa) ||
4032 (spa->spa_inject_ref != 0 &&
4033 new_state != POOL_STATE_UNINITIALIZED)) {
4034 spa_async_resume(spa);
4035 mutex_exit(&spa_namespace_lock);
4036 return (EBUSY);
4037 }
4038
4039 /*
4040 * A pool cannot be exported if it has an active shared spare.
4041 * This is to prevent other pools stealing the active spare
4042 * from an exported pool. At user's own will, such pool can
4043 * be forcedly exported.
4044 */
4045 if (!force && new_state == POOL_STATE_EXPORTED &&
4046 spa_has_active_shared_spare(spa)) {
4047 spa_async_resume(spa);
4048 mutex_exit(&spa_namespace_lock);
4049 return (EXDEV);
4050 }
4051
4052 /*
4053 * We want this to be reflected on every label,
4054 * so mark them all dirty. spa_unload() will do the
4055 * final sync that pushes these changes out.
4056 */
4057 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4058 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4059 spa->spa_state = new_state;
4060 spa->spa_final_txg = spa_last_synced_txg(spa) +
4061 TXG_DEFER_SIZE + 1;
4062 vdev_config_dirty(spa->spa_root_vdev);
4063 spa_config_exit(spa, SCL_ALL, FTAG);
4064 }
4065 }
4066
4067 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4068
4069 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4070 spa_unload(spa);
4071 spa_deactivate(spa);
4072 }
4073
4074 if (oldconfig && spa->spa_config)
4075 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4076
4077 if (new_state != POOL_STATE_UNINITIALIZED) {
4078 if (!hardforce)
4079 spa_config_sync(spa, B_TRUE, B_TRUE);
4080 spa_remove(spa);
4081 }
4082 mutex_exit(&spa_namespace_lock);
4083
4084 return (0);
4085}
4086
4087/*
4088 * Destroy a storage pool.
4089 */
4090int
4091spa_destroy(char *pool)
4092{
4093 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4094 B_FALSE, B_FALSE));
4095}
4096
4097/*
4098 * Export a storage pool.
4099 */
4100int
4101spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4102 boolean_t hardforce)
4103{
4104 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4105 force, hardforce));
4106}
4107
4108/*
4109 * Similar to spa_export(), this unloads the spa_t without actually removing it
4110 * from the namespace in any way.
4111 */
4112int
4113spa_reset(char *pool)
4114{
4115 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4116 B_FALSE, B_FALSE));
4117}
4118
4119/*
4120 * ==========================================================================
4121 * Device manipulation
4122 * ==========================================================================
4123 */
4124
4125/*
4126 * Add a device to a storage pool.
4127 */
4128int
4129spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4130{
4131 uint64_t txg, id;
4132 int error;
4133 vdev_t *rvd = spa->spa_root_vdev;
4134 vdev_t *vd, *tvd;
4135 nvlist_t **spares, **l2cache;
4136 uint_t nspares, nl2cache;
4137
4138 ASSERT(spa_writeable(spa));
4139
4140 txg = spa_vdev_enter(spa);
4141
4142 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4143 VDEV_ALLOC_ADD)) != 0)
4144 return (spa_vdev_exit(spa, NULL, txg, error));
4145
4146 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4147
4148 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4149 &nspares) != 0)
4150 nspares = 0;
4151
4152 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4153 &nl2cache) != 0)
4154 nl2cache = 0;
4155
4156 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4157 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4158
4159 if (vd->vdev_children != 0 &&
4160 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4161 return (spa_vdev_exit(spa, vd, txg, error));
4162
4163 /*
4164 * We must validate the spares and l2cache devices after checking the
4165 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4166 */
4167 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4168 return (spa_vdev_exit(spa, vd, txg, error));
4169
4170 /*
4171 * Transfer each new top-level vdev from vd to rvd.
4172 */
4173 for (int c = 0; c < vd->vdev_children; c++) {
4174
4175 /*
4176 * Set the vdev id to the first hole, if one exists.
4177 */
4178 for (id = 0; id < rvd->vdev_children; id++) {
4179 if (rvd->vdev_child[id]->vdev_ishole) {
4180 vdev_free(rvd->vdev_child[id]);
4181 break;
4182 }
4183 }
4184 tvd = vd->vdev_child[c];
4185 vdev_remove_child(vd, tvd);
4186 tvd->vdev_id = id;
4187 vdev_add_child(rvd, tvd);
4188 vdev_config_dirty(tvd);
4189 }
4190
4191 if (nspares != 0) {
4192 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4193 ZPOOL_CONFIG_SPARES);
4194 spa_load_spares(spa);
4195 spa->spa_spares.sav_sync = B_TRUE;
4196 }
4197
4198 if (nl2cache != 0) {
4199 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4200 ZPOOL_CONFIG_L2CACHE);
4201 spa_load_l2cache(spa);
4202 spa->spa_l2cache.sav_sync = B_TRUE;
4203 }
4204
4205 /*
4206 * We have to be careful when adding new vdevs to an existing pool.
4207 * If other threads start allocating from these vdevs before we
4208 * sync the config cache, and we lose power, then upon reboot we may
4209 * fail to open the pool because there are DVAs that the config cache
4210 * can't translate. Therefore, we first add the vdevs without
4211 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4212 * and then let spa_config_update() initialize the new metaslabs.
4213 *
4214 * spa_load() checks for added-but-not-initialized vdevs, so that
4215 * if we lose power at any point in this sequence, the remaining
4216 * steps will be completed the next time we load the pool.
4217 */
4218 (void) spa_vdev_exit(spa, vd, txg, 0);
4219
4220 mutex_enter(&spa_namespace_lock);
4221 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4222 mutex_exit(&spa_namespace_lock);
4223
4224 return (0);
4225}
4226
4227/*
4228 * Attach a device to a mirror. The arguments are the path to any device
4229 * in the mirror, and the nvroot for the new device. If the path specifies
4230 * a device that is not mirrored, we automatically insert the mirror vdev.
4231 *
4232 * If 'replacing' is specified, the new device is intended to replace the
4233 * existing device; in this case the two devices are made into their own
4234 * mirror using the 'replacing' vdev, which is functionally identical to
4235 * the mirror vdev (it actually reuses all the same ops) but has a few
4236 * extra rules: you can't attach to it after it's been created, and upon
4237 * completion of resilvering, the first disk (the one being replaced)
4238 * is automatically detached.
4239 */
4240int
4241spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4242{
4243 uint64_t txg, dtl_max_txg;
4244 vdev_t *rvd = spa->spa_root_vdev;
4245 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4246 vdev_ops_t *pvops;
4247 char *oldvdpath, *newvdpath;
4248 int newvd_isspare;
4249 int error;
4250
4251 ASSERT(spa_writeable(spa));
4252
4253 txg = spa_vdev_enter(spa);
4254
4255 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4256
4257 if (oldvd == NULL)
4258 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4259
4260 if (!oldvd->vdev_ops->vdev_op_leaf)
4261 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4262
4263 pvd = oldvd->vdev_parent;
4264
4265 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4266 VDEV_ALLOC_ATTACH)) != 0)
4267 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4268
4269 if (newrootvd->vdev_children != 1)
4270 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4271
4272 newvd = newrootvd->vdev_child[0];
4273
4274 if (!newvd->vdev_ops->vdev_op_leaf)
4275 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4276
4277 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4278 return (spa_vdev_exit(spa, newrootvd, txg, error));
4279
4280 /*
4281 * Spares can't replace logs
4282 */
4283 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4284 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4285
4286 if (!replacing) {
4287 /*
4288 * For attach, the only allowable parent is a mirror or the root
4289 * vdev.
4290 */
4291 if (pvd->vdev_ops != &vdev_mirror_ops &&
4292 pvd->vdev_ops != &vdev_root_ops)
4293 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4294
4295 pvops = &vdev_mirror_ops;
4296 } else {
4297 /*
4298 * Active hot spares can only be replaced by inactive hot
4299 * spares.
4300 */
4301 if (pvd->vdev_ops == &vdev_spare_ops &&
4302 oldvd->vdev_isspare &&
4303 !spa_has_spare(spa, newvd->vdev_guid))
4304 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4305
4306 /*
4307 * If the source is a hot spare, and the parent isn't already a
4308 * spare, then we want to create a new hot spare. Otherwise, we
4309 * want to create a replacing vdev. The user is not allowed to
4310 * attach to a spared vdev child unless the 'isspare' state is
4311 * the same (spare replaces spare, non-spare replaces
4312 * non-spare).
4313 */
4314 if (pvd->vdev_ops == &vdev_replacing_ops &&
4315 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4316 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4317 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4318 newvd->vdev_isspare != oldvd->vdev_isspare) {
4319 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4320 }
4321
4322 if (newvd->vdev_isspare)
4323 pvops = &vdev_spare_ops;
4324 else
4325 pvops = &vdev_replacing_ops;
4326 }
4327
4328 /*
4329 * Make sure the new device is big enough.
4330 */
4331 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4332 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4333
4334 /*
4335 * The new device cannot have a higher alignment requirement
4336 * than the top-level vdev.
4337 */
4338 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4339 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4340
4341 /*
4342 * If this is an in-place replacement, update oldvd's path and devid
4343 * to make it distinguishable from newvd, and unopenable from now on.
4344 */
4345 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4346 spa_strfree(oldvd->vdev_path);
4347 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4348 KM_SLEEP);
4349 (void) sprintf(oldvd->vdev_path, "%s/%s",
4350 newvd->vdev_path, "old");
4351 if (oldvd->vdev_devid != NULL) {
4352 spa_strfree(oldvd->vdev_devid);
4353 oldvd->vdev_devid = NULL;
4354 }
4355 }
4356
4357 /* mark the device being resilvered */
4358 newvd->vdev_resilvering = B_TRUE;
4359
4360 /*
4361 * If the parent is not a mirror, or if we're replacing, insert the new
4362 * mirror/replacing/spare vdev above oldvd.
4363 */
4364 if (pvd->vdev_ops != pvops)
4365 pvd = vdev_add_parent(oldvd, pvops);
4366
4367 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4368 ASSERT(pvd->vdev_ops == pvops);
4369 ASSERT(oldvd->vdev_parent == pvd);
4370
4371 /*
4372 * Extract the new device from its root and add it to pvd.
4373 */
4374 vdev_remove_child(newrootvd, newvd);
4375 newvd->vdev_id = pvd->vdev_children;
4376 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4377 vdev_add_child(pvd, newvd);
4378
4379 tvd = newvd->vdev_top;
4380 ASSERT(pvd->vdev_top == tvd);
4381 ASSERT(tvd->vdev_parent == rvd);
4382
4383 vdev_config_dirty(tvd);
4384
4385 /*
4386 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4387 * for any dmu_sync-ed blocks. It will propagate upward when
4388 * spa_vdev_exit() calls vdev_dtl_reassess().
4389 */
4390 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4391
4392 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4393 dtl_max_txg - TXG_INITIAL);
4394
4395 if (newvd->vdev_isspare) {
4396 spa_spare_activate(newvd);
4397 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4398 }
4399
4400 oldvdpath = spa_strdup(oldvd->vdev_path);
4401 newvdpath = spa_strdup(newvd->vdev_path);
4402 newvd_isspare = newvd->vdev_isspare;
4403
4404 /*
4405 * Mark newvd's DTL dirty in this txg.
4406 */
4407 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4408
4409 /*
4410 * Restart the resilver
4411 */
4412 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4413
4414 /*
4415 * Commit the config
4416 */
4417 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4418
4419 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4420 "%s vdev=%s %s vdev=%s",
4421 replacing && newvd_isspare ? "spare in" :
4422 replacing ? "replace" : "attach", newvdpath,
4423 replacing ? "for" : "to", oldvdpath);
4424
4425 spa_strfree(oldvdpath);
4426 spa_strfree(newvdpath);
4427
4428 if (spa->spa_bootfs)
4429 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4430
4431 return (0);
4432}
4433
4434/*
4435 * Detach a device from a mirror or replacing vdev.
4436 * If 'replace_done' is specified, only detach if the parent
4437 * is a replacing vdev.
4438 */
4439int
4440spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4441{
4442 uint64_t txg;
4443 int error;
4444 vdev_t *rvd = spa->spa_root_vdev;
4445 vdev_t *vd, *pvd, *cvd, *tvd;
4446 boolean_t unspare = B_FALSE;
4447 uint64_t unspare_guid;
4448 char *vdpath;
4449
4450 ASSERT(spa_writeable(spa));
4451
4452 txg = spa_vdev_enter(spa);
4453
4454 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4455
4456 if (vd == NULL)
4457 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4458
4459 if (!vd->vdev_ops->vdev_op_leaf)
4460 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4461
4462 pvd = vd->vdev_parent;
4463
4464 /*
4465 * If the parent/child relationship is not as expected, don't do it.
4466 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4467 * vdev that's replacing B with C. The user's intent in replacing
4468 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4469 * the replace by detaching C, the expected behavior is to end up
4470 * M(A,B). But suppose that right after deciding to detach C,
4471 * the replacement of B completes. We would have M(A,C), and then
4472 * ask to detach C, which would leave us with just A -- not what
4473 * the user wanted. To prevent this, we make sure that the
4474 * parent/child relationship hasn't changed -- in this example,
4475 * that C's parent is still the replacing vdev R.
4476 */
4477 if (pvd->vdev_guid != pguid && pguid != 0)
4478 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4479
4480 /*
4481 * Only 'replacing' or 'spare' vdevs can be replaced.
4482 */
4483 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4484 pvd->vdev_ops != &vdev_spare_ops)
4485 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4486
4487 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4488 spa_version(spa) >= SPA_VERSION_SPARES);
4489
4490 /*
4491 * Only mirror, replacing, and spare vdevs support detach.
4492 */
4493 if (pvd->vdev_ops != &vdev_replacing_ops &&
4494 pvd->vdev_ops != &vdev_mirror_ops &&
4495 pvd->vdev_ops != &vdev_spare_ops)
4496 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4497
4498 /*
4499 * If this device has the only valid copy of some data,
4500 * we cannot safely detach it.
4501 */
4502 if (vdev_dtl_required(vd))
4503 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4504
4505 ASSERT(pvd->vdev_children >= 2);
4506
4507 /*
4508 * If we are detaching the second disk from a replacing vdev, then
4509 * check to see if we changed the original vdev's path to have "/old"
4510 * at the end in spa_vdev_attach(). If so, undo that change now.
4511 */
4512 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4513 vd->vdev_path != NULL) {
4514 size_t len = strlen(vd->vdev_path);
4515
4516 for (int c = 0; c < pvd->vdev_children; c++) {
4517 cvd = pvd->vdev_child[c];
4518
4519 if (cvd == vd || cvd->vdev_path == NULL)
4520 continue;
4521
4522 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4523 strcmp(cvd->vdev_path + len, "/old") == 0) {
4524 spa_strfree(cvd->vdev_path);
4525 cvd->vdev_path = spa_strdup(vd->vdev_path);
4526 break;
4527 }
4528 }
4529 }
4530
4531 /*
4532 * If we are detaching the original disk from a spare, then it implies
4533 * that the spare should become a real disk, and be removed from the
4534 * active spare list for the pool.
4535 */
4536 if (pvd->vdev_ops == &vdev_spare_ops &&
4537 vd->vdev_id == 0 &&
4538 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4539 unspare = B_TRUE;
4540
4541 /*
4542 * Erase the disk labels so the disk can be used for other things.
4543 * This must be done after all other error cases are handled,
4544 * but before we disembowel vd (so we can still do I/O to it).
4545 * But if we can't do it, don't treat the error as fatal --
4546 * it may be that the unwritability of the disk is the reason
4547 * it's being detached!
4548 */
4549 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4550
4551 /*
4552 * Remove vd from its parent and compact the parent's children.
4553 */
4554 vdev_remove_child(pvd, vd);
4555 vdev_compact_children(pvd);
4556
4557 /*
4558 * Remember one of the remaining children so we can get tvd below.
4559 */
4560 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4561
4562 /*
4563 * If we need to remove the remaining child from the list of hot spares,
4564 * do it now, marking the vdev as no longer a spare in the process.
4565 * We must do this before vdev_remove_parent(), because that can
4566 * change the GUID if it creates a new toplevel GUID. For a similar
4567 * reason, we must remove the spare now, in the same txg as the detach;
4568 * otherwise someone could attach a new sibling, change the GUID, and
4569 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4570 */
4571 if (unspare) {
4572 ASSERT(cvd->vdev_isspare);
4573 spa_spare_remove(cvd);
4574 unspare_guid = cvd->vdev_guid;
4575 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4576 cvd->vdev_unspare = B_TRUE;
4577 }
4578
4579 /*
4580 * If the parent mirror/replacing vdev only has one child,
4581 * the parent is no longer needed. Remove it from the tree.
4582 */
4583 if (pvd->vdev_children == 1) {
4584 if (pvd->vdev_ops == &vdev_spare_ops)
4585 cvd->vdev_unspare = B_FALSE;
4586 vdev_remove_parent(cvd);
4587 cvd->vdev_resilvering = B_FALSE;
4588 }
4589
4590
4591 /*
4592 * We don't set tvd until now because the parent we just removed
4593 * may have been the previous top-level vdev.
4594 */
4595 tvd = cvd->vdev_top;
4596 ASSERT(tvd->vdev_parent == rvd);
4597
4598 /*
4599 * Reevaluate the parent vdev state.
4600 */
4601 vdev_propagate_state(cvd);
4602
4603 /*
4604 * If the 'autoexpand' property is set on the pool then automatically
4605 * try to expand the size of the pool. For example if the device we
4606 * just detached was smaller than the others, it may be possible to
4607 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4608 * first so that we can obtain the updated sizes of the leaf vdevs.
4609 */
4610 if (spa->spa_autoexpand) {
4611 vdev_reopen(tvd);
4612 vdev_expand(tvd, txg);
4613 }
4614
4615 vdev_config_dirty(tvd);
4616
4617 /*
4618 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4619 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4620 * But first make sure we're not on any *other* txg's DTL list, to
4621 * prevent vd from being accessed after it's freed.
4622 */
4623 vdpath = spa_strdup(vd->vdev_path);
4624 for (int t = 0; t < TXG_SIZE; t++)
4625 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4626 vd->vdev_detached = B_TRUE;
4627 vdev_dirty(tvd, VDD_DTL, vd, txg);
4628
4629 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4630
4631 /* hang on to the spa before we release the lock */
4632 spa_open_ref(spa, FTAG);
4633
4634 error = spa_vdev_exit(spa, vd, txg, 0);
4635
4636 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4637 "vdev=%s", vdpath);
4638 spa_strfree(vdpath);
4639
4640 /*
4641 * If this was the removal of the original device in a hot spare vdev,
4642 * then we want to go through and remove the device from the hot spare
4643 * list of every other pool.
4644 */
4645 if (unspare) {
4646 spa_t *altspa = NULL;
4647
4648 mutex_enter(&spa_namespace_lock);
4649 while ((altspa = spa_next(altspa)) != NULL) {
4650 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4651 altspa == spa)
4652 continue;
4653
4654 spa_open_ref(altspa, FTAG);
4655 mutex_exit(&spa_namespace_lock);
4656 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4657 mutex_enter(&spa_namespace_lock);
4658 spa_close(altspa, FTAG);
4659 }
4660 mutex_exit(&spa_namespace_lock);
4661
4662 /* search the rest of the vdevs for spares to remove */
4663 spa_vdev_resilver_done(spa);
4664 }
4665
4666 /* all done with the spa; OK to release */
4667 mutex_enter(&spa_namespace_lock);
4668 spa_close(spa, FTAG);
4669 mutex_exit(&spa_namespace_lock);
4670
4671 return (error);
4672}
4673
4674/*
4675 * Split a set of devices from their mirrors, and create a new pool from them.
4676 */
4677int
4678spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4679 nvlist_t *props, boolean_t exp)
4680{
4681 int error = 0;
4682 uint64_t txg, *glist;
4683 spa_t *newspa;
4684 uint_t c, children, lastlog;
4685 nvlist_t **child, *nvl, *tmp;
4686 dmu_tx_t *tx;
4687 char *altroot = NULL;
4688 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4689 boolean_t activate_slog;
4690
4691 ASSERT(spa_writeable(spa));
4692
4693 txg = spa_vdev_enter(spa);
4694
4695 /* clear the log and flush everything up to now */
4696 activate_slog = spa_passivate_log(spa);
4697 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4698 error = spa_offline_log(spa);
4699 txg = spa_vdev_config_enter(spa);
4700
4701 if (activate_slog)
4702 spa_activate_log(spa);
4703
4704 if (error != 0)
4705 return (spa_vdev_exit(spa, NULL, txg, error));
4706
4707 /* check new spa name before going any further */
4708 if (spa_lookup(newname) != NULL)
4709 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4710
4711 /*
4712 * scan through all the children to ensure they're all mirrors
4713 */
4714 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4715 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4716 &children) != 0)
4717 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4718
4719 /* first, check to ensure we've got the right child count */
4720 rvd = spa->spa_root_vdev;
4721 lastlog = 0;
4722 for (c = 0; c < rvd->vdev_children; c++) {
4723 vdev_t *vd = rvd->vdev_child[c];
4724
4725 /* don't count the holes & logs as children */
4726 if (vd->vdev_islog || vd->vdev_ishole) {
4727 if (lastlog == 0)
4728 lastlog = c;
4729 continue;
4730 }
4731
4732 lastlog = 0;
4733 }
4734 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4735 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4736
4737 /* next, ensure no spare or cache devices are part of the split */
4738 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4739 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4740 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4741
4742 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4743 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4744
4745 /* then, loop over each vdev and validate it */
4746 for (c = 0; c < children; c++) {
4747 uint64_t is_hole = 0;
4748
4749 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4750 &is_hole);
4751
4752 if (is_hole != 0) {
4753 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4754 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4755 continue;
4756 } else {
4757 error = EINVAL;
4758 break;
4759 }
4760 }
4761
4762 /* which disk is going to be split? */
4763 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4764 &glist[c]) != 0) {
4765 error = EINVAL;
4766 break;
4767 }
4768
4769 /* look it up in the spa */
4770 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4771 if (vml[c] == NULL) {
4772 error = ENODEV;
4773 break;
4774 }
4775
4776 /* make sure there's nothing stopping the split */
4777 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4778 vml[c]->vdev_islog ||
4779 vml[c]->vdev_ishole ||
4780 vml[c]->vdev_isspare ||
4781 vml[c]->vdev_isl2cache ||
4782 !vdev_writeable(vml[c]) ||
4783 vml[c]->vdev_children != 0 ||
4784 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4785 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4786 error = EINVAL;
4787 break;
4788 }
4789
4790 if (vdev_dtl_required(vml[c])) {
4791 error = EBUSY;
4792 break;
4793 }
4794
4795 /* we need certain info from the top level */
4796 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4797 vml[c]->vdev_top->vdev_ms_array) == 0);
4798 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4799 vml[c]->vdev_top->vdev_ms_shift) == 0);
4800 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4801 vml[c]->vdev_top->vdev_asize) == 0);
4802 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4803 vml[c]->vdev_top->vdev_ashift) == 0);
4804 }
4805
4806 if (error != 0) {
4807 kmem_free(vml, children * sizeof (vdev_t *));
4808 kmem_free(glist, children * sizeof (uint64_t));
4809 return (spa_vdev_exit(spa, NULL, txg, error));
4810 }
4811
4812 /* stop writers from using the disks */
4813 for (c = 0; c < children; c++) {
4814 if (vml[c] != NULL)
4815 vml[c]->vdev_offline = B_TRUE;
4816 }
4817 vdev_reopen(spa->spa_root_vdev);
4818
4819 /*
4820 * Temporarily record the splitting vdevs in the spa config. This
4821 * will disappear once the config is regenerated.
4822 */
4823 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4824 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4825 glist, children) == 0);
4826 kmem_free(glist, children * sizeof (uint64_t));
4827
4828 mutex_enter(&spa->spa_props_lock);
4829 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4830 nvl) == 0);
4831 mutex_exit(&spa->spa_props_lock);
4832 spa->spa_config_splitting = nvl;
4833 vdev_config_dirty(spa->spa_root_vdev);
4834
4835 /* configure and create the new pool */
4836 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4837 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4838 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4839 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4840 spa_version(spa)) == 0);
4841 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4842 spa->spa_config_txg) == 0);
4843 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4844 spa_generate_guid(NULL)) == 0);
4845 (void) nvlist_lookup_string(props,
4846 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4847
4848 /* add the new pool to the namespace */
4849 newspa = spa_add(newname, config, altroot);
4850 newspa->spa_config_txg = spa->spa_config_txg;
4851 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4852
4853 /* release the spa config lock, retaining the namespace lock */
4854 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4855
4856 if (zio_injection_enabled)
4857 zio_handle_panic_injection(spa, FTAG, 1);
4858
4859 spa_activate(newspa, spa_mode_global);
4860 spa_async_suspend(newspa);
4861
4862#ifndef sun
4863 /* mark that we are creating new spa by splitting */
4864 newspa->spa_splitting_newspa = B_TRUE;
4865#endif
4866 /* create the new pool from the disks of the original pool */
4867 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4868#ifndef sun
4869 newspa->spa_splitting_newspa = B_FALSE;
4870#endif
4871 if (error)
4872 goto out;
4873
4874 /* if that worked, generate a real config for the new pool */
4875 if (newspa->spa_root_vdev != NULL) {
4876 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4877 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4878 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4879 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4880 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4881 B_TRUE));
4882 }
4883
4884 /* set the props */
4885 if (props != NULL) {
4886 spa_configfile_set(newspa, props, B_FALSE);
4887 error = spa_prop_set(newspa, props);
4888 if (error)
4889 goto out;
4890 }
4891
4892 /* flush everything */
4893 txg = spa_vdev_config_enter(newspa);
4894 vdev_config_dirty(newspa->spa_root_vdev);
4895 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4896
4897 if (zio_injection_enabled)
4898 zio_handle_panic_injection(spa, FTAG, 2);
4899
4900 spa_async_resume(newspa);
4901
4902 /* finally, update the original pool's config */
4903 txg = spa_vdev_config_enter(spa);
4904 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4905 error = dmu_tx_assign(tx, TXG_WAIT);
4906 if (error != 0)
4907 dmu_tx_abort(tx);
4908 for (c = 0; c < children; c++) {
4909 if (vml[c] != NULL) {
4910 vdev_split(vml[c]);
4911 if (error == 0)
4912 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4913 spa, tx, "vdev=%s",
4914 vml[c]->vdev_path);
4915 vdev_free(vml[c]);
4916 }
4917 }
4918 vdev_config_dirty(spa->spa_root_vdev);
4919 spa->spa_config_splitting = NULL;
4920 nvlist_free(nvl);
4921 if (error == 0)
4922 dmu_tx_commit(tx);
4923 (void) spa_vdev_exit(spa, NULL, txg, 0);
4924
4925 if (zio_injection_enabled)
4926 zio_handle_panic_injection(spa, FTAG, 3);
4927
4928 /* split is complete; log a history record */
4929 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4930 "split new pool %s from pool %s", newname, spa_name(spa));
4931
4932 kmem_free(vml, children * sizeof (vdev_t *));
4933
4934 /* if we're not going to mount the filesystems in userland, export */
4935 if (exp)
4936 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4937 B_FALSE, B_FALSE);
4938
4939 return (error);
4940
4941out:
4942 spa_unload(newspa);
4943 spa_deactivate(newspa);
4944 spa_remove(newspa);
4945
4946 txg = spa_vdev_config_enter(spa);
4947
4948 /* re-online all offlined disks */
4949 for (c = 0; c < children; c++) {
4950 if (vml[c] != NULL)
4951 vml[c]->vdev_offline = B_FALSE;
4952 }
4953 vdev_reopen(spa->spa_root_vdev);
4954
4955 nvlist_free(spa->spa_config_splitting);
4956 spa->spa_config_splitting = NULL;
4957 (void) spa_vdev_exit(spa, NULL, txg, error);
4958
4959 kmem_free(vml, children * sizeof (vdev_t *));
4960 return (error);
4961}
4962
4963static nvlist_t *
4964spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4965{
4966 for (int i = 0; i < count; i++) {
4967 uint64_t guid;
4968
4969 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4970 &guid) == 0);
4971
4972 if (guid == target_guid)
4973 return (nvpp[i]);
4974 }
4975
4976 return (NULL);
4977}
4978
4979static void
4980spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4981 nvlist_t *dev_to_remove)
4982{
4983 nvlist_t **newdev = NULL;
4984
4985 if (count > 1)
4986 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4987
4988 for (int i = 0, j = 0; i < count; i++) {
4989 if (dev[i] == dev_to_remove)
4990 continue;
4991 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4992 }
4993
4994 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4995 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4996
4997 for (int i = 0; i < count - 1; i++)
4998 nvlist_free(newdev[i]);
4999
5000 if (count > 1)
5001 kmem_free(newdev, (count - 1) * sizeof (void *));
5002}
5003
5004/*
5005 * Evacuate the device.
5006 */
5007static int
5008spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5009{
5010 uint64_t txg;
5011 int error = 0;
5012
5013 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5014 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5015 ASSERT(vd == vd->vdev_top);
5016
5017 /*
5018 * Evacuate the device. We don't hold the config lock as writer
5019 * since we need to do I/O but we do keep the
5020 * spa_namespace_lock held. Once this completes the device
5021 * should no longer have any blocks allocated on it.
5022 */
5023 if (vd->vdev_islog) {
5024 if (vd->vdev_stat.vs_alloc != 0)
5025 error = spa_offline_log(spa);
5026 } else {
5027 error = ENOTSUP;
5028 }
5029
5030 if (error)
5031 return (error);
5032
5033 /*
5034 * The evacuation succeeded. Remove any remaining MOS metadata
5035 * associated with this vdev, and wait for these changes to sync.
5036 */
5037 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
5038 txg = spa_vdev_config_enter(spa);
5039 vd->vdev_removing = B_TRUE;
5040 vdev_dirty(vd, 0, NULL, txg);
5041 vdev_config_dirty(vd);
5042 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5043
5044 return (0);
5045}
5046
5047/*
5048 * Complete the removal by cleaning up the namespace.
5049 */
5050static void
5051spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5052{
5053 vdev_t *rvd = spa->spa_root_vdev;
5054 uint64_t id = vd->vdev_id;
5055 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5056
5057 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5058 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5059 ASSERT(vd == vd->vdev_top);
5060
5061 /*
5062 * Only remove any devices which are empty.
5063 */
5064 if (vd->vdev_stat.vs_alloc != 0)
5065 return;
5066
5067 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5068
5069 if (list_link_active(&vd->vdev_state_dirty_node))
5070 vdev_state_clean(vd);
5071 if (list_link_active(&vd->vdev_config_dirty_node))
5072 vdev_config_clean(vd);
5073
5074 vdev_free(vd);
5075
5076 if (last_vdev) {
5077 vdev_compact_children(rvd);
5078 } else {
5079 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5080 vdev_add_child(rvd, vd);
5081 }
5082 vdev_config_dirty(rvd);
5083
5084 /*
5085 * Reassess the health of our root vdev.
5086 */
5087 vdev_reopen(rvd);
5088}
5089
5090/*
5091 * Remove a device from the pool -
5092 *
5093 * Removing a device from the vdev namespace requires several steps
5094 * and can take a significant amount of time. As a result we use
5095 * the spa_vdev_config_[enter/exit] functions which allow us to
5096 * grab and release the spa_config_lock while still holding the namespace
5097 * lock. During each step the configuration is synced out.
5098 */
5099
5100/*
5101 * Remove a device from the pool. Currently, this supports removing only hot
5102 * spares, slogs, and level 2 ARC devices.
5103 */
5104int
5105spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5106{
5107 vdev_t *vd;
5108 metaslab_group_t *mg;
5109 nvlist_t **spares, **l2cache, *nv;
5110 uint64_t txg = 0;
5111 uint_t nspares, nl2cache;
5112 int error = 0;
5113 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5114
5115 ASSERT(spa_writeable(spa));
5116
5117 if (!locked)
5118 txg = spa_vdev_enter(spa);
5119
5120 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5121
5122 if (spa->spa_spares.sav_vdevs != NULL &&
5123 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5124 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5125 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5126 /*
5127 * Only remove the hot spare if it's not currently in use
5128 * in this pool.
5129 */
5130 if (vd == NULL || unspare) {
5131 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5132 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5133 spa_load_spares(spa);
5134 spa->spa_spares.sav_sync = B_TRUE;
5135 } else {
5136 error = EBUSY;
5137 }
5138 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5139 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5140 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5141 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5142 /*
5143 * Cache devices can always be removed.
5144 */
5145 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5146 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5147 spa_load_l2cache(spa);
5148 spa->spa_l2cache.sav_sync = B_TRUE;
5149 } else if (vd != NULL && vd->vdev_islog) {
5150 ASSERT(!locked);
5151 ASSERT(vd == vd->vdev_top);
5152
5153 /*
5154 * XXX - Once we have bp-rewrite this should
5155 * become the common case.
5156 */
5157
5158 mg = vd->vdev_mg;
5159
5160 /*
5161 * Stop allocating from this vdev.
5162 */
5163 metaslab_group_passivate(mg);
5164
5165 /*
5166 * Wait for the youngest allocations and frees to sync,
5167 * and then wait for the deferral of those frees to finish.
5168 */
5169 spa_vdev_config_exit(spa, NULL,
5170 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5171
5172 /*
5173 * Attempt to evacuate the vdev.
5174 */
5175 error = spa_vdev_remove_evacuate(spa, vd);
5176
5177 txg = spa_vdev_config_enter(spa);
5178
5179 /*
5180 * If we couldn't evacuate the vdev, unwind.
5181 */
5182 if (error) {
5183 metaslab_group_activate(mg);
5184 return (spa_vdev_exit(spa, NULL, txg, error));
5185 }
5186
5187 /*
5188 * Clean up the vdev namespace.
5189 */
5190 spa_vdev_remove_from_namespace(spa, vd);
5191
5192 } else if (vd != NULL) {
5193 /*
5194 * Normal vdevs cannot be removed (yet).
5195 */
5196 error = ENOTSUP;
5197 } else {
5198 /*
5199 * There is no vdev of any kind with the specified guid.
5200 */
5201 error = ENOENT;
5202 }
5203
5204 if (!locked)
5205 return (spa_vdev_exit(spa, NULL, txg, error));
5206
5207 return (error);
5208}
5209
5210/*
5211 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5212 * current spared, so we can detach it.
5213 */
5214static vdev_t *
5215spa_vdev_resilver_done_hunt(vdev_t *vd)
5216{
5217 vdev_t *newvd, *oldvd;
5218
5219 for (int c = 0; c < vd->vdev_children; c++) {
5220 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5221 if (oldvd != NULL)
5222 return (oldvd);
5223 }
5224
5225 /*
5226 * Check for a completed replacement. We always consider the first
5227 * vdev in the list to be the oldest vdev, and the last one to be
5228 * the newest (see spa_vdev_attach() for how that works). In
5229 * the case where the newest vdev is faulted, we will not automatically
5230 * remove it after a resilver completes. This is OK as it will require
5231 * user intervention to determine which disk the admin wishes to keep.
5232 */
5233 if (vd->vdev_ops == &vdev_replacing_ops) {
5234 ASSERT(vd->vdev_children > 1);
5235
5236 newvd = vd->vdev_child[vd->vdev_children - 1];
5237 oldvd = vd->vdev_child[0];
5238
5239 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5240 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5241 !vdev_dtl_required(oldvd))
5242 return (oldvd);
5243 }
5244
5245 /*
5246 * Check for a completed resilver with the 'unspare' flag set.
5247 */
5248 if (vd->vdev_ops == &vdev_spare_ops) {
5249 vdev_t *first = vd->vdev_child[0];
5250 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5251
5252 if (last->vdev_unspare) {
5253 oldvd = first;
5254 newvd = last;
5255 } else if (first->vdev_unspare) {
5256 oldvd = last;
5257 newvd = first;
5258 } else {
5259 oldvd = NULL;
5260 }
5261
5262 if (oldvd != NULL &&
5263 vdev_dtl_empty(newvd, DTL_MISSING) &&
5264 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5265 !vdev_dtl_required(oldvd))
5266 return (oldvd);
5267
5268 /*
5269 * If there are more than two spares attached to a disk,
5270 * and those spares are not required, then we want to
5271 * attempt to free them up now so that they can be used
5272 * by other pools. Once we're back down to a single
5273 * disk+spare, we stop removing them.
5274 */
5275 if (vd->vdev_children > 2) {
5276 newvd = vd->vdev_child[1];
5277
5278 if (newvd->vdev_isspare && last->vdev_isspare &&
5279 vdev_dtl_empty(last, DTL_MISSING) &&
5280 vdev_dtl_empty(last, DTL_OUTAGE) &&
5281 !vdev_dtl_required(newvd))
5282 return (newvd);
5283 }
5284 }
5285
5286 return (NULL);
5287}
5288
5289static void
5290spa_vdev_resilver_done(spa_t *spa)
5291{
5292 vdev_t *vd, *pvd, *ppvd;
5293 uint64_t guid, sguid, pguid, ppguid;
5294
5295 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5296
5297 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5298 pvd = vd->vdev_parent;
5299 ppvd = pvd->vdev_parent;
5300 guid = vd->vdev_guid;
5301 pguid = pvd->vdev_guid;
5302 ppguid = ppvd->vdev_guid;
5303 sguid = 0;
5304 /*
5305 * If we have just finished replacing a hot spared device, then
5306 * we need to detach the parent's first child (the original hot
5307 * spare) as well.
5308 */
5309 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5310 ppvd->vdev_children == 2) {
5311 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5312 sguid = ppvd->vdev_child[1]->vdev_guid;
5313 }
5314 spa_config_exit(spa, SCL_ALL, FTAG);
5315 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5316 return;
5317 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5318 return;
5319 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5320 }
5321
5322 spa_config_exit(spa, SCL_ALL, FTAG);
5323}
5324
5325/*
5326 * Update the stored path or FRU for this vdev.
5327 */
5328int
5329spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5330 boolean_t ispath)
5331{
5332 vdev_t *vd;
5333 boolean_t sync = B_FALSE;
5334
5335 ASSERT(spa_writeable(spa));
5336
5337 spa_vdev_state_enter(spa, SCL_ALL);
5338
5339 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5340 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5341
5342 if (!vd->vdev_ops->vdev_op_leaf)
5343 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5344
5345 if (ispath) {
5346 if (strcmp(value, vd->vdev_path) != 0) {
5347 spa_strfree(vd->vdev_path);
5348 vd->vdev_path = spa_strdup(value);
5349 sync = B_TRUE;
5350 }
5351 } else {
5352 if (vd->vdev_fru == NULL) {
5353 vd->vdev_fru = spa_strdup(value);
5354 sync = B_TRUE;
5355 } else if (strcmp(value, vd->vdev_fru) != 0) {
5356 spa_strfree(vd->vdev_fru);
5357 vd->vdev_fru = spa_strdup(value);
5358 sync = B_TRUE;
5359 }
5360 }
5361
5362 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5363}
5364
5365int
5366spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5367{
5368 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5369}
5370
5371int
5372spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5373{
5374 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5375}
5376
5377/*
5378 * ==========================================================================
5379 * SPA Scanning
5380 * ==========================================================================
5381 */
5382
5383int
5384spa_scan_stop(spa_t *spa)
5385{
5386 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5387 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5388 return (EBUSY);
5389 return (dsl_scan_cancel(spa->spa_dsl_pool));
5390}
5391
5392int
5393spa_scan(spa_t *spa, pool_scan_func_t func)
5394{
5395 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5396
5397 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5398 return (ENOTSUP);
5399
5400 /*
5401 * If a resilver was requested, but there is no DTL on a
5402 * writeable leaf device, we have nothing to do.
5403 */
5404 if (func == POOL_SCAN_RESILVER &&
5405 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5406 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5407 return (0);
5408 }
5409
5410 return (dsl_scan(spa->spa_dsl_pool, func));
5411}
5412
5413/*
5414 * ==========================================================================
5415 * SPA async task processing
5416 * ==========================================================================
5417 */
5418
5419static void
5420spa_async_remove(spa_t *spa, vdev_t *vd)
5421{
5422 if (vd->vdev_remove_wanted) {
5423 vd->vdev_remove_wanted = B_FALSE;
5424 vd->vdev_delayed_close = B_FALSE;
5425 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5426
5427 /*
5428 * We want to clear the stats, but we don't want to do a full
5429 * vdev_clear() as that will cause us to throw away
5430 * degraded/faulted state as well as attempt to reopen the
5431 * device, all of which is a waste.
5432 */
5433 vd->vdev_stat.vs_read_errors = 0;
5434 vd->vdev_stat.vs_write_errors = 0;
5435 vd->vdev_stat.vs_checksum_errors = 0;
5436
5437 vdev_state_dirty(vd->vdev_top);
5438 }
5439
5440 for (int c = 0; c < vd->vdev_children; c++)
5441 spa_async_remove(spa, vd->vdev_child[c]);
5442}
5443
5444static void
5445spa_async_probe(spa_t *spa, vdev_t *vd)
5446{
5447 if (vd->vdev_probe_wanted) {
5448 vd->vdev_probe_wanted = B_FALSE;
5449 vdev_reopen(vd); /* vdev_open() does the actual probe */
5450 }
5451
5452 for (int c = 0; c < vd->vdev_children; c++)
5453 spa_async_probe(spa, vd->vdev_child[c]);
5454}
5455
5456static void
5457spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5458{
5459 sysevent_id_t eid;
5460 nvlist_t *attr;
5461 char *physpath;
5462
5463 if (!spa->spa_autoexpand)
5464 return;
5465
5466 for (int c = 0; c < vd->vdev_children; c++) {
5467 vdev_t *cvd = vd->vdev_child[c];
5468 spa_async_autoexpand(spa, cvd);
5469 }
5470
5471 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5472 return;
5473
5474 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5475 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5476
5477 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5478 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5479
5480 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5481 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5482
5483 nvlist_free(attr);
5484 kmem_free(physpath, MAXPATHLEN);
5485}
5486
5487static void
5488spa_async_thread(void *arg)
5489{
5490 spa_t *spa = arg;
5491 int tasks;
5492
5493 ASSERT(spa->spa_sync_on);
5494
5495 mutex_enter(&spa->spa_async_lock);
5496 tasks = spa->spa_async_tasks;
5497 spa->spa_async_tasks = 0;
5498 mutex_exit(&spa->spa_async_lock);
5499
5500 /*
5501 * See if the config needs to be updated.
5502 */
5503 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5504 uint64_t old_space, new_space;
5505
5506 mutex_enter(&spa_namespace_lock);
5507 old_space = metaslab_class_get_space(spa_normal_class(spa));
5508 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5509 new_space = metaslab_class_get_space(spa_normal_class(spa));
5510 mutex_exit(&spa_namespace_lock);
5511
5512 /*
5513 * If the pool grew as a result of the config update,
5514 * then log an internal history event.
5515 */
5516 if (new_space != old_space) {
5517 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5518 spa, NULL,
5519 "pool '%s' size: %llu(+%llu)",
5520 spa_name(spa), new_space, new_space - old_space);
5521 }
5522 }
5523
5524 /*
5525 * See if any devices need to be marked REMOVED.
5526 */
5527 if (tasks & SPA_ASYNC_REMOVE) {
5528 spa_vdev_state_enter(spa, SCL_NONE);
5529 spa_async_remove(spa, spa->spa_root_vdev);
5530 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5531 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5532 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5533 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5534 (void) spa_vdev_state_exit(spa, NULL, 0);
5535 }
5536
5537 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5538 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5539 spa_async_autoexpand(spa, spa->spa_root_vdev);
5540 spa_config_exit(spa, SCL_CONFIG, FTAG);
5541 }
5542
5543 /*
5544 * See if any devices need to be probed.
5545 */
5546 if (tasks & SPA_ASYNC_PROBE) {
5547 spa_vdev_state_enter(spa, SCL_NONE);
5548 spa_async_probe(spa, spa->spa_root_vdev);
5549 (void) spa_vdev_state_exit(spa, NULL, 0);
5550 }
5551
5552 /*
5553 * If any devices are done replacing, detach them.
5554 */
5555 if (tasks & SPA_ASYNC_RESILVER_DONE)
5556 spa_vdev_resilver_done(spa);
5557
5558 /*
5559 * Kick off a resilver.
5560 */
5561 if (tasks & SPA_ASYNC_RESILVER)
5562 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5563
5564 /*
5565 * Let the world know that we're done.
5566 */
5567 mutex_enter(&spa->spa_async_lock);
5568 spa->spa_async_thread = NULL;
5569 cv_broadcast(&spa->spa_async_cv);
5570 mutex_exit(&spa->spa_async_lock);
5571 thread_exit();
5572}
5573
5574void
5575spa_async_suspend(spa_t *spa)
5576{
5577 mutex_enter(&spa->spa_async_lock);
5578 spa->spa_async_suspended++;
5579 while (spa->spa_async_thread != NULL)
5580 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5581 mutex_exit(&spa->spa_async_lock);
5582}
5583
5584void
5585spa_async_resume(spa_t *spa)
5586{
5587 mutex_enter(&spa->spa_async_lock);
5588 ASSERT(spa->spa_async_suspended != 0);
5589 spa->spa_async_suspended--;
5590 mutex_exit(&spa->spa_async_lock);
5591}
5592
5593static void
5594spa_async_dispatch(spa_t *spa)
5595{
5596 mutex_enter(&spa->spa_async_lock);
5597 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5598 spa->spa_async_thread == NULL &&
5599 rootdir != NULL && !vn_is_readonly(rootdir))
5600 spa->spa_async_thread = thread_create(NULL, 0,
5601 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5602 mutex_exit(&spa->spa_async_lock);
5603}
5604
5605void
5606spa_async_request(spa_t *spa, int task)
5607{
5608 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5609 mutex_enter(&spa->spa_async_lock);
5610 spa->spa_async_tasks |= task;
5611 mutex_exit(&spa->spa_async_lock);
5612}
5613
5614/*
5615 * ==========================================================================
5616 * SPA syncing routines
5617 * ==========================================================================
5618 */
5619
5620static int
5621bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5622{
5623 bpobj_t *bpo = arg;
5624 bpobj_enqueue(bpo, bp, tx);
5625 return (0);
5626}
5627
5628static int
5629spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5630{
5631 zio_t *zio = arg;
5632
5633 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5634 zio->io_flags));
5635 return (0);
5636}
5637
5638static void
5639spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5640{
5641 char *packed = NULL;
5642 size_t bufsize;
5643 size_t nvsize = 0;
5644 dmu_buf_t *db;
5645
5646 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5647
5648 /*
5649 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5650 * information. This avoids the dbuf_will_dirty() path and
5651 * saves us a pre-read to get data we don't actually care about.
5652 */
5653 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5654 packed = kmem_alloc(bufsize, KM_SLEEP);
5655
5656 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5657 KM_SLEEP) == 0);
5658 bzero(packed + nvsize, bufsize - nvsize);
5659
5660 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5661
5662 kmem_free(packed, bufsize);
5663
5664 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5665 dmu_buf_will_dirty(db, tx);
5666 *(uint64_t *)db->db_data = nvsize;
5667 dmu_buf_rele(db, FTAG);
5668}
5669
5670static void
5671spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5672 const char *config, const char *entry)
5673{
5674 nvlist_t *nvroot;
5675 nvlist_t **list;
5676 int i;
5677
5678 if (!sav->sav_sync)
5679 return;
5680
5681 /*
5682 * Update the MOS nvlist describing the list of available devices.
5683 * spa_validate_aux() will have already made sure this nvlist is
5684 * valid and the vdevs are labeled appropriately.
5685 */
5686 if (sav->sav_object == 0) {
5687 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5688 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5689 sizeof (uint64_t), tx);
5690 VERIFY(zap_update(spa->spa_meta_objset,
5691 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5692 &sav->sav_object, tx) == 0);
5693 }
5694
5695 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5696 if (sav->sav_count == 0) {
5697 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5698 } else {
5699 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5700 for (i = 0; i < sav->sav_count; i++)
5701 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5702 B_FALSE, VDEV_CONFIG_L2CACHE);
5703 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5704 sav->sav_count) == 0);
5705 for (i = 0; i < sav->sav_count; i++)
5706 nvlist_free(list[i]);
5707 kmem_free(list, sav->sav_count * sizeof (void *));
5708 }
5709
5710 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5711 nvlist_free(nvroot);
5712
5713 sav->sav_sync = B_FALSE;
5714}
5715
5716static void
5717spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5718{
5719 nvlist_t *config;
5720
5721 if (list_is_empty(&spa->spa_config_dirty_list))
5722 return;
5723
5724 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5725
5726 config = spa_config_generate(spa, spa->spa_root_vdev,
5727 dmu_tx_get_txg(tx), B_FALSE);
5728
5729 spa_config_exit(spa, SCL_STATE, FTAG);
5730
5731 if (spa->spa_config_syncing)
5732 nvlist_free(spa->spa_config_syncing);
5733 spa->spa_config_syncing = config;
5734
5735 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5736}
5737
5738static void
5739spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5740{
5741 spa_t *spa = arg1;
5742 uint64_t version = *(uint64_t *)arg2;
5743
5744 /*
5745 * Setting the version is special cased when first creating the pool.
5746 */
5747 ASSERT(tx->tx_txg != TXG_INITIAL);
5748
5749 ASSERT(version <= SPA_VERSION);
5750 ASSERT(version >= spa_version(spa));
5751
5752 spa->spa_uberblock.ub_version = version;
5753 vdev_config_dirty(spa->spa_root_vdev);
5754}
5755
5756/*
5757 * Set zpool properties.
5758 */
5759static void
5760spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5761{
5762 spa_t *spa = arg1;
5763 objset_t *mos = spa->spa_meta_objset;
5764 nvlist_t *nvp = arg2;
5765 nvpair_t *elem = NULL;
5766
5767 mutex_enter(&spa->spa_props_lock);
5768
5769 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5770 uint64_t intval;
5771 char *strval, *fname;
5772 zpool_prop_t prop;
5773 const char *propname;
5774 zprop_type_t proptype;
5775 zfeature_info_t *feature;
5776
5777 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5778 case ZPROP_INVAL:
5779 /*
5780 * We checked this earlier in spa_prop_validate().
5781 */
5782 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5783
5784 fname = strchr(nvpair_name(elem), '@') + 1;
5785 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5786
5787 spa_feature_enable(spa, feature, tx);
5788 break;
5789
5790 case ZPOOL_PROP_VERSION:
5791 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5792 /*
5793 * The version is synced seperatly before other
5794 * properties and should be correct by now.
5795 */
5796 ASSERT3U(spa_version(spa), >=, intval);
5797 break;
5798
5799 case ZPOOL_PROP_ALTROOT:
5800 /*
5801 * 'altroot' is a non-persistent property. It should
5802 * have been set temporarily at creation or import time.
5803 */
5804 ASSERT(spa->spa_root != NULL);
5805 break;
5806
5807 case ZPOOL_PROP_READONLY:
5808 case ZPOOL_PROP_CACHEFILE:
5809 /*
5810 * 'readonly' and 'cachefile' are also non-persisitent
5811 * properties.
5812 */
5813 break;
5814 case ZPOOL_PROP_COMMENT:
5815 VERIFY(nvpair_value_string(elem, &strval) == 0);
5816 if (spa->spa_comment != NULL)
5817 spa_strfree(spa->spa_comment);
5818 spa->spa_comment = spa_strdup(strval);
5819 /*
5820 * We need to dirty the configuration on all the vdevs
5821 * so that their labels get updated. It's unnecessary
5822 * to do this for pool creation since the vdev's
5823 * configuratoin has already been dirtied.
5824 */
5825 if (tx->tx_txg != TXG_INITIAL)
5826 vdev_config_dirty(spa->spa_root_vdev);
5827 break;
5828 default:
5829 /*
5830 * Set pool property values in the poolprops mos object.
5831 */
5832 if (spa->spa_pool_props_object == 0) {
5833 spa->spa_pool_props_object =
5834 zap_create_link(mos, DMU_OT_POOL_PROPS,
5835 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5836 tx);
5837 }
5838
5839 /* normalize the property name */
5840 propname = zpool_prop_to_name(prop);
5841 proptype = zpool_prop_get_type(prop);
5842
5843 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5844 ASSERT(proptype == PROP_TYPE_STRING);
5845 VERIFY(nvpair_value_string(elem, &strval) == 0);
5846 VERIFY(zap_update(mos,
5847 spa->spa_pool_props_object, propname,
5848 1, strlen(strval) + 1, strval, tx) == 0);
5849
5850 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5851 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5852
5853 if (proptype == PROP_TYPE_INDEX) {
5854 const char *unused;
5855 VERIFY(zpool_prop_index_to_string(
5856 prop, intval, &unused) == 0);
5857 }
5858 VERIFY(zap_update(mos,
5859 spa->spa_pool_props_object, propname,
5860 8, 1, &intval, tx) == 0);
5861 } else {
5862 ASSERT(0); /* not allowed */
5863 }
5864
5865 switch (prop) {
5866 case ZPOOL_PROP_DELEGATION:
5867 spa->spa_delegation = intval;
5868 break;
5869 case ZPOOL_PROP_BOOTFS:
5870 spa->spa_bootfs = intval;
5871 break;
5872 case ZPOOL_PROP_FAILUREMODE:
5873 spa->spa_failmode = intval;
5874 break;
5875 case ZPOOL_PROP_AUTOEXPAND:
5876 spa->spa_autoexpand = intval;
5877 if (tx->tx_txg != TXG_INITIAL)
5878 spa_async_request(spa,
5879 SPA_ASYNC_AUTOEXPAND);
5880 break;
5881 case ZPOOL_PROP_DEDUPDITTO:
5882 spa->spa_dedup_ditto = intval;
5883 break;
5884 default:
5885 break;
5886 }
5887 }
5888
5889 /* log internal history if this is not a zpool create */
5890 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5891 tx->tx_txg != TXG_INITIAL) {
5892 spa_history_log_internal(LOG_POOL_PROPSET,
5893 spa, tx, "%s %lld %s",
5894 nvpair_name(elem), intval, spa_name(spa));
5895 }
5896 }
5897
5898 mutex_exit(&spa->spa_props_lock);
5899}
5900
5901/*
5902 * Perform one-time upgrade on-disk changes. spa_version() does not
5903 * reflect the new version this txg, so there must be no changes this
5904 * txg to anything that the upgrade code depends on after it executes.
5905 * Therefore this must be called after dsl_pool_sync() does the sync
5906 * tasks.
5907 */
5908static void
5909spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5910{
5911 dsl_pool_t *dp = spa->spa_dsl_pool;
5912
5913 ASSERT(spa->spa_sync_pass == 1);
5914
5915 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5916 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5917 dsl_pool_create_origin(dp, tx);
5918
5919 /* Keeping the origin open increases spa_minref */
5920 spa->spa_minref += 3;
5921 }
5922
5923 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5924 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5925 dsl_pool_upgrade_clones(dp, tx);
5926 }
5927
5928 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5929 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5930 dsl_pool_upgrade_dir_clones(dp, tx);
5931
5932 /* Keeping the freedir open increases spa_minref */
5933 spa->spa_minref += 3;
5934 }
5935
5936 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5937 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5938 spa_feature_create_zap_objects(spa, tx);
5939 }
5940}
5941
5942/*
5943 * Sync the specified transaction group. New blocks may be dirtied as
5944 * part of the process, so we iterate until it converges.
5945 */
5946void
5947spa_sync(spa_t *spa, uint64_t txg)
5948{
5949 dsl_pool_t *dp = spa->spa_dsl_pool;
5950 objset_t *mos = spa->spa_meta_objset;
5951 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5952 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5953 vdev_t *rvd = spa->spa_root_vdev;
5954 vdev_t *vd;
5955 dmu_tx_t *tx;
5956 int error;
5957
5958 VERIFY(spa_writeable(spa));
5959
5960 /*
5961 * Lock out configuration changes.
5962 */
5963 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5964
5965 spa->spa_syncing_txg = txg;
5966 spa->spa_sync_pass = 0;
5967
5968 /*
5969 * If there are any pending vdev state changes, convert them
5970 * into config changes that go out with this transaction group.
5971 */
5972 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5973 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5974 /*
5975 * We need the write lock here because, for aux vdevs,
5976 * calling vdev_config_dirty() modifies sav_config.
5977 * This is ugly and will become unnecessary when we
5978 * eliminate the aux vdev wart by integrating all vdevs
5979 * into the root vdev tree.
5980 */
5981 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5982 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5983 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5984 vdev_state_clean(vd);
5985 vdev_config_dirty(vd);
5986 }
5987 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5988 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5989 }
5990 spa_config_exit(spa, SCL_STATE, FTAG);
5991
5992 tx = dmu_tx_create_assigned(dp, txg);
5993
5994 /*
5995 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5996 * set spa_deflate if we have no raid-z vdevs.
5997 */
5998 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5999 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6000 int i;
6001
6002 for (i = 0; i < rvd->vdev_children; i++) {
6003 vd = rvd->vdev_child[i];
6004 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6005 break;
6006 }
6007 if (i == rvd->vdev_children) {
6008 spa->spa_deflate = TRUE;
6009 VERIFY(0 == zap_add(spa->spa_meta_objset,
6010 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6011 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6012 }
6013 }
6014
6015 /*
6016 * If anything has changed in this txg, or if someone is waiting
6017 * for this txg to sync (eg, spa_vdev_remove()), push the
6018 * deferred frees from the previous txg. If not, leave them
6019 * alone so that we don't generate work on an otherwise idle
6020 * system.
6021 */
6022 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6023 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6024 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6025 ((dsl_scan_active(dp->dp_scan) ||
6026 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6027 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6028 VERIFY3U(bpobj_iterate(defer_bpo,
6029 spa_free_sync_cb, zio, tx), ==, 0);
6030 VERIFY3U(zio_wait(zio), ==, 0);
6031 }
6032
6033 /*
6034 * Iterate to convergence.
6035 */
6036 do {
6037 int pass = ++spa->spa_sync_pass;
6038
6039 spa_sync_config_object(spa, tx);
6040 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6041 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6042 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6043 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6044 spa_errlog_sync(spa, txg);
6045 dsl_pool_sync(dp, txg);
6046
6047 if (pass <= SYNC_PASS_DEFERRED_FREE) {
6048 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6049 bplist_iterate(free_bpl, spa_free_sync_cb,
6050 zio, tx);
6051 VERIFY(zio_wait(zio) == 0);
6052 } else {
6053 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6054 defer_bpo, tx);
6055 }
6056
6057 ddt_sync(spa, txg);
6058 dsl_scan_sync(dp, tx);
6059
6060 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6061 vdev_sync(vd, txg);
6062
6063 if (pass == 1)
6064 spa_sync_upgrades(spa, tx);
6065
6066 } while (dmu_objset_is_dirty(mos, txg));
6067
6068 /*
6069 * Rewrite the vdev configuration (which includes the uberblock)
6070 * to commit the transaction group.
6071 *
6072 * If there are no dirty vdevs, we sync the uberblock to a few
6073 * random top-level vdevs that are known to be visible in the
6074 * config cache (see spa_vdev_add() for a complete description).
6075 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6076 */
6077 for (;;) {
6078 /*
6079 * We hold SCL_STATE to prevent vdev open/close/etc.
6080 * while we're attempting to write the vdev labels.
6081 */
6082 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6083
6084 if (list_is_empty(&spa->spa_config_dirty_list)) {
6085 vdev_t *svd[SPA_DVAS_PER_BP];
6086 int svdcount = 0;
6087 int children = rvd->vdev_children;
6088 int c0 = spa_get_random(children);
6089
6090 for (int c = 0; c < children; c++) {
6091 vd = rvd->vdev_child[(c0 + c) % children];
6092 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6093 continue;
6094 svd[svdcount++] = vd;
6095 if (svdcount == SPA_DVAS_PER_BP)
6096 break;
6097 }
6098 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6099 if (error != 0)
6100 error = vdev_config_sync(svd, svdcount, txg,
6101 B_TRUE);
6102 } else {
6103 error = vdev_config_sync(rvd->vdev_child,
6104 rvd->vdev_children, txg, B_FALSE);
6105 if (error != 0)
6106 error = vdev_config_sync(rvd->vdev_child,
6107 rvd->vdev_children, txg, B_TRUE);
6108 }
6109
6110 spa_config_exit(spa, SCL_STATE, FTAG);
6111
6112 if (error == 0)
6113 break;
6114 zio_suspend(spa, NULL);
6115 zio_resume_wait(spa);
6116 }
6117 dmu_tx_commit(tx);
6118
6119 /*
6120 * Clear the dirty config list.
6121 */
6122 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6123 vdev_config_clean(vd);
6124
6125 /*
6126 * Now that the new config has synced transactionally,
6127 * let it become visible to the config cache.
6128 */
6129 if (spa->spa_config_syncing != NULL) {
6130 spa_config_set(spa, spa->spa_config_syncing);
6131 spa->spa_config_txg = txg;
6132 spa->spa_config_syncing = NULL;
6133 }
6134
6135 spa->spa_ubsync = spa->spa_uberblock;
6136
6137 dsl_pool_sync_done(dp, txg);
6138
6139 /*
6140 * Update usable space statistics.
6141 */
6142 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6143 vdev_sync_done(vd, txg);
6144
6145 spa_update_dspace(spa);
6146
6147 /*
6148 * It had better be the case that we didn't dirty anything
6149 * since vdev_config_sync().
6150 */
6151 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6152 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6153 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6154
6155 spa->spa_sync_pass = 0;
6156
6157 spa_config_exit(spa, SCL_CONFIG, FTAG);
6158
6159 spa_handle_ignored_writes(spa);
6160
6161 /*
6162 * If any async tasks have been requested, kick them off.
6163 */
6164 spa_async_dispatch(spa);
6165}
6166
6167/*
6168 * Sync all pools. We don't want to hold the namespace lock across these
6169 * operations, so we take a reference on the spa_t and drop the lock during the
6170 * sync.
6171 */
6172void
6173spa_sync_allpools(void)
6174{
6175 spa_t *spa = NULL;
6176 mutex_enter(&spa_namespace_lock);
6177 while ((spa = spa_next(spa)) != NULL) {
6178 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6179 !spa_writeable(spa) || spa_suspended(spa))
6180 continue;
6181 spa_open_ref(spa, FTAG);
6182 mutex_exit(&spa_namespace_lock);
6183 txg_wait_synced(spa_get_dsl(spa), 0);
6184 mutex_enter(&spa_namespace_lock);
6185 spa_close(spa, FTAG);
6186 }
6187 mutex_exit(&spa_namespace_lock);
6188}
6189
6190/*
6191 * ==========================================================================
6192 * Miscellaneous routines
6193 * ==========================================================================
6194 */
6195
6196/*
6197 * Remove all pools in the system.
6198 */
6199void
6200spa_evict_all(void)
6201{
6202 spa_t *spa;
6203
6204 /*
6205 * Remove all cached state. All pools should be closed now,
6206 * so every spa in the AVL tree should be unreferenced.
6207 */
6208 mutex_enter(&spa_namespace_lock);
6209 while ((spa = spa_next(NULL)) != NULL) {
6210 /*
6211 * Stop async tasks. The async thread may need to detach
6212 * a device that's been replaced, which requires grabbing
6213 * spa_namespace_lock, so we must drop it here.
6214 */
6215 spa_open_ref(spa, FTAG);
6216 mutex_exit(&spa_namespace_lock);
6217 spa_async_suspend(spa);
6218 mutex_enter(&spa_namespace_lock);
6219 spa_close(spa, FTAG);
6220
6221 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6222 spa_unload(spa);
6223 spa_deactivate(spa);
6224 }
6225 spa_remove(spa);
6226 }
6227 mutex_exit(&spa_namespace_lock);
6228}
6229
6230vdev_t *
6231spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6232{
6233 vdev_t *vd;
6234 int i;
6235
6236 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6237 return (vd);
6238
6239 if (aux) {
6240 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6241 vd = spa->spa_l2cache.sav_vdevs[i];
6242 if (vd->vdev_guid == guid)
6243 return (vd);
6244 }
6245
6246 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6247 vd = spa->spa_spares.sav_vdevs[i];
6248 if (vd->vdev_guid == guid)
6249 return (vd);
6250 }
6251 }
6252
6253 return (NULL);
6254}
6255
6256void
6257spa_upgrade(spa_t *spa, uint64_t version)
6258{
6259 ASSERT(spa_writeable(spa));
6260
6261 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6262
6263 /*
6264 * This should only be called for a non-faulted pool, and since a
6265 * future version would result in an unopenable pool, this shouldn't be
6266 * possible.
6267 */
6268 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6269 ASSERT(version >= spa->spa_uberblock.ub_version);
6270
6271 spa->spa_uberblock.ub_version = version;
6272 vdev_config_dirty(spa->spa_root_vdev);
6273
6274 spa_config_exit(spa, SCL_ALL, FTAG);
6275
6276 txg_wait_synced(spa_get_dsl(spa), 0);
6277}
6278
6279boolean_t
6280spa_has_spare(spa_t *spa, uint64_t guid)
6281{
6282 int i;
6283 uint64_t spareguid;
6284 spa_aux_vdev_t *sav = &spa->spa_spares;
6285
6286 for (i = 0; i < sav->sav_count; i++)
6287 if (sav->sav_vdevs[i]->vdev_guid == guid)
6288 return (B_TRUE);
6289
6290 for (i = 0; i < sav->sav_npending; i++) {
6291 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6292 &spareguid) == 0 && spareguid == guid)
6293 return (B_TRUE);
6294 }
6295
6296 return (B_FALSE);
6297}
6298
6299/*
6300 * Check if a pool has an active shared spare device.
6301 * Note: reference count of an active spare is 2, as a spare and as a replace
6302 */
6303static boolean_t
6304spa_has_active_shared_spare(spa_t *spa)
6305{
6306 int i, refcnt;
6307 uint64_t pool;
6308 spa_aux_vdev_t *sav = &spa->spa_spares;
6309
6310 for (i = 0; i < sav->sav_count; i++) {
6311 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6312 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6313 refcnt > 2)
6314 return (B_TRUE);
6315 }
6316
6317 return (B_FALSE);
6318}
6319
6320/*
6321 * Post a sysevent corresponding to the given event. The 'name' must be one of
6322 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6323 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6324 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6325 * or zdb as real changes.
6326 */
6327void
6328spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6329{
6330#ifdef _KERNEL
6331 sysevent_t *ev;
6332 sysevent_attr_list_t *attr = NULL;
6333 sysevent_value_t value;
6334 sysevent_id_t eid;
6335
6336 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6337 SE_SLEEP);
6338
6339 value.value_type = SE_DATA_TYPE_STRING;
6340 value.value.sv_string = spa_name(spa);
6341 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6342 goto done;
6343
6344 value.value_type = SE_DATA_TYPE_UINT64;
6345 value.value.sv_uint64 = spa_guid(spa);
6346 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6347 goto done;
6348
6349 if (vd) {
6350 value.value_type = SE_DATA_TYPE_UINT64;
6351 value.value.sv_uint64 = vd->vdev_guid;
6352 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6353 SE_SLEEP) != 0)
6354 goto done;
6355
6356 if (vd->vdev_path) {
6357 value.value_type = SE_DATA_TYPE_STRING;
6358 value.value.sv_string = vd->vdev_path;
6359 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6360 &value, SE_SLEEP) != 0)
6361 goto done;
6362 }
6363 }
6364
6365 if (sysevent_attach_attributes(ev, attr) != 0)
6366 goto done;
6367 attr = NULL;
6368
6369 (void) log_sysevent(ev, SE_SLEEP, &eid);
6370
6371done:
6372 if (attr)
6373 sysevent_free_attr(attr);
6374 sysevent_free(ev);
6375#endif
6376}
|