1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011 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/zvol.h>
64
65#ifdef _KERNEL
66#include <sys/callb.h>
67#include <sys/cpupart.h>
68#include <sys/zone.h>
69#endif /* _KERNEL */
70
71#include "zfs_prop.h"
72#include "zfs_comutil.h"
73
74/* Check hostid on import? */
75static int check_hostid = 1;
76
77SYSCTL_DECL(_vfs_zfs);
78TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
79SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
80 "Check hostid on import?");
81
82typedef enum zti_modes {
83 zti_mode_fixed, /* value is # of threads (min 1) */
84 zti_mode_online_percent, /* value is % of online CPUs */
85 zti_mode_batch, /* cpu-intensive; value is ignored */
86 zti_mode_null, /* don't create a taskq */
87 zti_nmodes
88} zti_modes_t;
89
90#define ZTI_FIX(n) { zti_mode_fixed, (n) }
91#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
92#define ZTI_BATCH { zti_mode_batch, 0 }
93#define ZTI_NULL { zti_mode_null, 0 }
94
95#define ZTI_ONE ZTI_FIX(1)
96
97typedef struct zio_taskq_info {
98 enum zti_modes zti_mode;
99 uint_t zti_value;
100} zio_taskq_info_t;
101
102static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
103 "issue", "issue_high", "intr", "intr_high"
104};
105
106/*
107 * Define the taskq threads for the following I/O types:
108 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
109 */
110const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
111 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
112 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
114 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
115 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
116 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118};
119
120static dsl_syncfunc_t spa_sync_props;
121static boolean_t spa_has_active_shared_spare(spa_t *spa);
122static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
123 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
124 char **ereport);
125static void spa_vdev_resilver_done(spa_t *spa);
126
127uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
128#ifdef PSRSET_BIND
129id_t zio_taskq_psrset_bind = PS_NONE;
130#endif
131#ifdef SYSDC
132boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
133#endif
134uint_t zio_taskq_basedc = 80; /* base duty cycle */
135
136boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
137
138/*
139 * This (illegal) pool name is used when temporarily importing a spa_t in order
140 * to get the vdev stats associated with the imported devices.
141 */
142#define TRYIMPORT_NAME "$import"
143
144/*
145 * ==========================================================================
146 * SPA properties routines
147 * ==========================================================================
148 */
149
150/*
151 * Add a (source=src, propname=propval) list to an nvlist.
152 */
153static void
154spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
155 uint64_t intval, zprop_source_t src)
156{
157 const char *propname = zpool_prop_to_name(prop);
158 nvlist_t *propval;
159
160 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
161 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
162
163 if (strval != NULL)
164 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
165 else
166 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
167
168 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
169 nvlist_free(propval);
170}
171
172/*
173 * Get property values from the spa configuration.
174 */
175static void
176spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
177{
178 uint64_t size;
179 uint64_t alloc;
180 uint64_t cap, version;
181 zprop_source_t src = ZPROP_SRC_NONE;
182 spa_config_dirent_t *dp;
183
184 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
185
186 if (spa->spa_root_vdev != NULL) {
187 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
188 size = metaslab_class_get_space(spa_normal_class(spa));
189 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
190 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
192 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
193 size - alloc, src);
194 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
195 (spa_mode(spa) == FREAD), src);
196
197 cap = (size == 0) ? 0 : (alloc * 100 / size);
198 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
199
200 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
201 ddt_get_pool_dedup_ratio(spa), src);
202
203 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
204 spa->spa_root_vdev->vdev_state, src);
205
206 version = spa_version(spa);
207 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
208 src = ZPROP_SRC_DEFAULT;
209 else
210 src = ZPROP_SRC_LOCAL;
211 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
212 }
213
214 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
215
216 if (spa->spa_comment != NULL) {
217 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
218 0, ZPROP_SRC_LOCAL);
219 }
220
221 if (spa->spa_root != NULL)
222 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
223 0, ZPROP_SRC_LOCAL);
224
225 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
226 if (dp->scd_path == NULL) {
227 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
228 "none", 0, ZPROP_SRC_LOCAL);
229 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
230 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
231 dp->scd_path, 0, ZPROP_SRC_LOCAL);
232 }
233 }
234}
235
236/*
237 * Get zpool property values.
238 */
239int
240spa_prop_get(spa_t *spa, nvlist_t **nvp)
241{
242 objset_t *mos = spa->spa_meta_objset;
243 zap_cursor_t zc;
244 zap_attribute_t za;
245 int err;
246
247 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
248
249 mutex_enter(&spa->spa_props_lock);
250
251 /*
252 * Get properties from the spa config.
253 */
254 spa_prop_get_config(spa, nvp);
255
256 /* If no pool property object, no more prop to get. */
257 if (mos == NULL || spa->spa_pool_props_object == 0) {
258 mutex_exit(&spa->spa_props_lock);
259 return (0);
260 }
261
262 /*
263 * Get properties from the MOS pool property object.
264 */
265 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
266 (err = zap_cursor_retrieve(&zc, &za)) == 0;
267 zap_cursor_advance(&zc)) {
268 uint64_t intval = 0;
269 char *strval = NULL;
270 zprop_source_t src = ZPROP_SRC_DEFAULT;
271 zpool_prop_t prop;
272
273 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
274 continue;
275
276 switch (za.za_integer_length) {
277 case 8:
278 /* integer property */
279 if (za.za_first_integer !=
280 zpool_prop_default_numeric(prop))
281 src = ZPROP_SRC_LOCAL;
282
283 if (prop == ZPOOL_PROP_BOOTFS) {
284 dsl_pool_t *dp;
285 dsl_dataset_t *ds = NULL;
286
287 dp = spa_get_dsl(spa);
288 rw_enter(&dp->dp_config_rwlock, RW_READER);
289 if (err = dsl_dataset_hold_obj(dp,
290 za.za_first_integer, FTAG, &ds)) {
291 rw_exit(&dp->dp_config_rwlock);
292 break;
293 }
294
295 strval = kmem_alloc(
296 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
297 KM_SLEEP);
298 dsl_dataset_name(ds, strval);
299 dsl_dataset_rele(ds, FTAG);
300 rw_exit(&dp->dp_config_rwlock);
301 } else {
302 strval = NULL;
303 intval = za.za_first_integer;
304 }
305
306 spa_prop_add_list(*nvp, prop, strval, intval, src);
307
308 if (strval != NULL)
309 kmem_free(strval,
310 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
311
312 break;
313
314 case 1:
315 /* string property */
316 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
317 err = zap_lookup(mos, spa->spa_pool_props_object,
318 za.za_name, 1, za.za_num_integers, strval);
319 if (err) {
320 kmem_free(strval, za.za_num_integers);
321 break;
322 }
323 spa_prop_add_list(*nvp, prop, strval, 0, src);
324 kmem_free(strval, za.za_num_integers);
325 break;
326
327 default:
328 break;
329 }
330 }
331 zap_cursor_fini(&zc);
332 mutex_exit(&spa->spa_props_lock);
333out:
334 if (err && err != ENOENT) {
335 nvlist_free(*nvp);
336 *nvp = NULL;
337 return (err);
338 }
339
340 return (0);
341}
342
343/*
344 * Validate the given pool properties nvlist and modify the list
345 * for the property values to be set.
346 */
347static int
348spa_prop_validate(spa_t *spa, nvlist_t *props)
349{
350 nvpair_t *elem;
351 int error = 0, reset_bootfs = 0;
352 uint64_t objnum;
353
354 elem = NULL;
355 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
356 zpool_prop_t prop;
357 char *propname, *strval;
358 uint64_t intval;
359 objset_t *os;
360 char *slash, *check;
361
362 propname = nvpair_name(elem);
363
364 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
365 return (EINVAL);
366
367 switch (prop) {
368 case ZPOOL_PROP_VERSION:
369 error = nvpair_value_uint64(elem, &intval);
370 if (!error &&
371 (intval < spa_version(spa) || intval > SPA_VERSION))
372 error = EINVAL;
373 break;
374
375 case ZPOOL_PROP_DELEGATION:
376 case ZPOOL_PROP_AUTOREPLACE:
377 case ZPOOL_PROP_LISTSNAPS:
378 case ZPOOL_PROP_AUTOEXPAND:
379 error = nvpair_value_uint64(elem, &intval);
380 if (!error && intval > 1)
381 error = EINVAL;
382 break;
383
384 case ZPOOL_PROP_BOOTFS:
385 /*
386 * If the pool version is less than SPA_VERSION_BOOTFS,
387 * or the pool is still being created (version == 0),
388 * the bootfs property cannot be set.
389 */
390 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
391 error = ENOTSUP;
392 break;
393 }
394
395 /*
396 * Make sure the vdev config is bootable
397 */
398 if (!vdev_is_bootable(spa->spa_root_vdev)) {
399 error = ENOTSUP;
400 break;
401 }
402
403 reset_bootfs = 1;
404
405 error = nvpair_value_string(elem, &strval);
406
407 if (!error) {
408 uint64_t compress;
409
410 if (strval == NULL || strval[0] == '\0') {
411 objnum = zpool_prop_default_numeric(
412 ZPOOL_PROP_BOOTFS);
413 break;
414 }
415
416 if (error = dmu_objset_hold(strval, FTAG, &os))
417 break;
418
419 /* Must be ZPL and not gzip compressed. */
420
421 if (dmu_objset_type(os) != DMU_OST_ZFS) {
422 error = ENOTSUP;
423 } else if ((error = dsl_prop_get_integer(strval,
424 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
425 &compress, NULL)) == 0 &&
426 !BOOTFS_COMPRESS_VALID(compress)) {
427 error = ENOTSUP;
428 } else {
429 objnum = dmu_objset_id(os);
430 }
431 dmu_objset_rele(os, FTAG);
432 }
433 break;
434
435 case ZPOOL_PROP_FAILUREMODE:
436 error = nvpair_value_uint64(elem, &intval);
437 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
438 intval > ZIO_FAILURE_MODE_PANIC))
439 error = EINVAL;
440
441 /*
442 * This is a special case which only occurs when
443 * the pool has completely failed. This allows
444 * the user to change the in-core failmode property
445 * without syncing it out to disk (I/Os might
446 * currently be blocked). We do this by returning
447 * EIO to the caller (spa_prop_set) to trick it
448 * into thinking we encountered a property validation
449 * error.
450 */
451 if (!error && spa_suspended(spa)) {
452 spa->spa_failmode = intval;
453 error = EIO;
454 }
455 break;
456
457 case ZPOOL_PROP_CACHEFILE:
458 if ((error = nvpair_value_string(elem, &strval)) != 0)
459 break;
460
461 if (strval[0] == '\0')
462 break;
463
464 if (strcmp(strval, "none") == 0)
465 break;
466
467 if (strval[0] != '/') {
468 error = EINVAL;
469 break;
470 }
471
472 slash = strrchr(strval, '/');
473 ASSERT(slash != NULL);
474
475 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
476 strcmp(slash, "/..") == 0)
477 error = EINVAL;
478 break;
479
480 case ZPOOL_PROP_COMMENT:
481 if ((error = nvpair_value_string(elem, &strval)) != 0)
482 break;
483 for (check = strval; *check != '\0'; check++) {
484 /*
485 * The kernel doesn't have an easy isprint()
486 * check. For this kernel check, we merely
487 * check ASCII apart from DEL. Fix this if
488 * there is an easy-to-use kernel isprint().
489 */
490 if (*check >= 0x7f) {
491 error = EINVAL;
492 break;
493 }
494 check++;
495 }
496 if (strlen(strval) > ZPROP_MAX_COMMENT)
497 error = E2BIG;
498 break;
499
500 case ZPOOL_PROP_DEDUPDITTO:
501 if (spa_version(spa) < SPA_VERSION_DEDUP)
502 error = ENOTSUP;
503 else
504 error = nvpair_value_uint64(elem, &intval);
505 if (error == 0 &&
506 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
507 error = EINVAL;
508 break;
509 }
510
511 if (error)
512 break;
513 }
514
515 if (!error && reset_bootfs) {
516 error = nvlist_remove(props,
517 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
518
519 if (!error) {
520 error = nvlist_add_uint64(props,
521 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
522 }
523 }
524
525 return (error);
526}
527
528void
529spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
530{
531 char *cachefile;
532 spa_config_dirent_t *dp;
533
534 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
535 &cachefile) != 0)
536 return;
537
538 dp = kmem_alloc(sizeof (spa_config_dirent_t),
539 KM_SLEEP);
540
541 if (cachefile[0] == '\0')
542 dp->scd_path = spa_strdup(spa_config_path);
543 else if (strcmp(cachefile, "none") == 0)
544 dp->scd_path = NULL;
545 else
546 dp->scd_path = spa_strdup(cachefile);
547
548 list_insert_head(&spa->spa_config_list, dp);
549 if (need_sync)
550 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
551}
552
553int
554spa_prop_set(spa_t *spa, nvlist_t *nvp)
555{
556 int error;
557 nvpair_t *elem;
558 boolean_t need_sync = B_FALSE;
559 zpool_prop_t prop;
560
561 if ((error = spa_prop_validate(spa, nvp)) != 0)
562 return (error);
563
564 elem = NULL;
565 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
566 if ((prop = zpool_name_to_prop(
567 nvpair_name(elem))) == ZPROP_INVAL)
568 return (EINVAL);
569
570 if (prop == ZPOOL_PROP_CACHEFILE ||
571 prop == ZPOOL_PROP_ALTROOT ||
572 prop == ZPOOL_PROP_READONLY)
573 continue;
574
575 need_sync = B_TRUE;
576 break;
577 }
578
579 if (need_sync)
580 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
581 spa, nvp, 3));
582 else
583 return (0);
584}
585
586/*
587 * If the bootfs property value is dsobj, clear it.
588 */
589void
590spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
591{
592 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
593 VERIFY(zap_remove(spa->spa_meta_objset,
594 spa->spa_pool_props_object,
595 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
596 spa->spa_bootfs = 0;
597 }
598}
599
600/*
601 * Change the GUID for the pool. This is done so that we can later
602 * re-import a pool built from a clone of our own vdevs. We will modify
603 * the root vdev's guid, our own pool guid, and then mark all of our
604 * vdevs dirty. Note that we must make sure that all our vdevs are
605 * online when we do this, or else any vdevs that weren't present
606 * would be orphaned from our pool. We are also going to issue a
607 * sysevent to update any watchers.
608 */
609int
610spa_change_guid(spa_t *spa)
611{
612 uint64_t oldguid, newguid;
613 uint64_t txg;
614
615 if (!(spa_mode_global & FWRITE))
616 return (EROFS);
617
618 txg = spa_vdev_enter(spa);
619
620 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
621 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
622
623 oldguid = spa_guid(spa);
624 newguid = spa_generate_guid(NULL);
625 ASSERT3U(oldguid, !=, newguid);
626
627 spa->spa_root_vdev->vdev_guid = newguid;
628 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
629
630 vdev_config_dirty(spa->spa_root_vdev);
631
632 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
633
634 return (spa_vdev_exit(spa, NULL, txg, 0));
635}
636
637/*
638 * ==========================================================================
639 * SPA state manipulation (open/create/destroy/import/export)
640 * ==========================================================================
641 */
642
643static int
644spa_error_entry_compare(const void *a, const void *b)
645{
646 spa_error_entry_t *sa = (spa_error_entry_t *)a;
647 spa_error_entry_t *sb = (spa_error_entry_t *)b;
648 int ret;
649
650 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
651 sizeof (zbookmark_t));
652
653 if (ret < 0)
654 return (-1);
655 else if (ret > 0)
656 return (1);
657 else
658 return (0);
659}
660
661/*
662 * Utility function which retrieves copies of the current logs and
663 * re-initializes them in the process.
664 */
665void
666spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
667{
668 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
669
670 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
671 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
672
673 avl_create(&spa->spa_errlist_scrub,
674 spa_error_entry_compare, sizeof (spa_error_entry_t),
675 offsetof(spa_error_entry_t, se_avl));
676 avl_create(&spa->spa_errlist_last,
677 spa_error_entry_compare, sizeof (spa_error_entry_t),
678 offsetof(spa_error_entry_t, se_avl));
679}
680
681static taskq_t *
682spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
683 uint_t value)
684{
685 uint_t flags = TASKQ_PREPOPULATE;
686 boolean_t batch = B_FALSE;
687
688 switch (mode) {
689 case zti_mode_null:
690 return (NULL); /* no taskq needed */
691
692 case zti_mode_fixed:
693 ASSERT3U(value, >=, 1);
694 value = MAX(value, 1);
695 break;
696
697 case zti_mode_batch:
698 batch = B_TRUE;
699 flags |= TASKQ_THREADS_CPU_PCT;
700 value = zio_taskq_batch_pct;
701 break;
702
703 case zti_mode_online_percent:
704 flags |= TASKQ_THREADS_CPU_PCT;
705 break;
706
707 default:
708 panic("unrecognized mode for %s taskq (%u:%u) in "
709 "spa_activate()",
710 name, mode, value);
711 break;
712 }
713
714#ifdef SYSDC
715 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
716 if (batch)
717 flags |= TASKQ_DC_BATCH;
718
719 return (taskq_create_sysdc(name, value, 50, INT_MAX,
720 spa->spa_proc, zio_taskq_basedc, flags));
721 }
722#endif
723 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
724 spa->spa_proc, flags));
725}
726
727static void
728spa_create_zio_taskqs(spa_t *spa)
729{
730 for (int t = 0; t < ZIO_TYPES; t++) {
731 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
732 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
733 enum zti_modes mode = ztip->zti_mode;
734 uint_t value = ztip->zti_value;
735 char name[32];
736
737 (void) snprintf(name, sizeof (name),
738 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
739
740 spa->spa_zio_taskq[t][q] =
741 spa_taskq_create(spa, name, mode, value);
742 }
743 }
744}
745
746#ifdef _KERNEL
747#ifdef SPA_PROCESS
748static void
749spa_thread(void *arg)
750{
751 callb_cpr_t cprinfo;
752
753 spa_t *spa = arg;
754 user_t *pu = PTOU(curproc);
755
756 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
757 spa->spa_name);
758
759 ASSERT(curproc != &p0);
760 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
761 "zpool-%s", spa->spa_name);
762 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
763
764#ifdef PSRSET_BIND
765 /* bind this thread to the requested psrset */
766 if (zio_taskq_psrset_bind != PS_NONE) {
767 pool_lock();
768 mutex_enter(&cpu_lock);
769 mutex_enter(&pidlock);
770 mutex_enter(&curproc->p_lock);
771
772 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
773 0, NULL, NULL) == 0) {
774 curthread->t_bind_pset = zio_taskq_psrset_bind;
775 } else {
776 cmn_err(CE_WARN,
777 "Couldn't bind process for zfs pool \"%s\" to "
778 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
779 }
780
781 mutex_exit(&curproc->p_lock);
782 mutex_exit(&pidlock);
783 mutex_exit(&cpu_lock);
784 pool_unlock();
785 }
786#endif
787
788#ifdef SYSDC
789 if (zio_taskq_sysdc) {
790 sysdc_thread_enter(curthread, 100, 0);
791 }
792#endif
793
794 spa->spa_proc = curproc;
795 spa->spa_did = curthread->t_did;
796
797 spa_create_zio_taskqs(spa);
798
799 mutex_enter(&spa->spa_proc_lock);
800 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
801
802 spa->spa_proc_state = SPA_PROC_ACTIVE;
803 cv_broadcast(&spa->spa_proc_cv);
804
805 CALLB_CPR_SAFE_BEGIN(&cprinfo);
806 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
807 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
808 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
809
810 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
811 spa->spa_proc_state = SPA_PROC_GONE;
812 spa->spa_proc = &p0;
813 cv_broadcast(&spa->spa_proc_cv);
814 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
815
816 mutex_enter(&curproc->p_lock);
817 lwp_exit();
818}
819#endif /* SPA_PROCESS */
820#endif
821
822/*
823 * Activate an uninitialized pool.
824 */
825static void
826spa_activate(spa_t *spa, int mode)
827{
828 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
829
830 spa->spa_state = POOL_STATE_ACTIVE;
831 spa->spa_mode = mode;
832
833 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
834 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
835
836 /* Try to create a covering process */
837 mutex_enter(&spa->spa_proc_lock);
838 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
839 ASSERT(spa->spa_proc == &p0);
840 spa->spa_did = 0;
841
842#ifdef SPA_PROCESS
843 /* Only create a process if we're going to be around a while. */
844 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
845 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
846 NULL, 0) == 0) {
847 spa->spa_proc_state = SPA_PROC_CREATED;
848 while (spa->spa_proc_state == SPA_PROC_CREATED) {
849 cv_wait(&spa->spa_proc_cv,
850 &spa->spa_proc_lock);
851 }
852 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
853 ASSERT(spa->spa_proc != &p0);
854 ASSERT(spa->spa_did != 0);
855 } else {
856#ifdef _KERNEL
857 cmn_err(CE_WARN,
858 "Couldn't create process for zfs pool \"%s\"\n",
859 spa->spa_name);
860#endif
861 }
862 }
863#endif /* SPA_PROCESS */
864 mutex_exit(&spa->spa_proc_lock);
865
866 /* If we didn't create a process, we need to create our taskqs. */
867 ASSERT(spa->spa_proc == &p0);
868 if (spa->spa_proc == &p0) {
869 spa_create_zio_taskqs(spa);
870 }
871
872 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
873 offsetof(vdev_t, vdev_config_dirty_node));
874 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
875 offsetof(vdev_t, vdev_state_dirty_node));
876
877 txg_list_create(&spa->spa_vdev_txg_list,
878 offsetof(struct vdev, vdev_txg_node));
879
880 avl_create(&spa->spa_errlist_scrub,
881 spa_error_entry_compare, sizeof (spa_error_entry_t),
882 offsetof(spa_error_entry_t, se_avl));
883 avl_create(&spa->spa_errlist_last,
884 spa_error_entry_compare, sizeof (spa_error_entry_t),
885 offsetof(spa_error_entry_t, se_avl));
886}
887
888/*
889 * Opposite of spa_activate().
890 */
891static void
892spa_deactivate(spa_t *spa)
893{
894 ASSERT(spa->spa_sync_on == B_FALSE);
895 ASSERT(spa->spa_dsl_pool == NULL);
896 ASSERT(spa->spa_root_vdev == NULL);
897 ASSERT(spa->spa_async_zio_root == NULL);
898 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
899
900 txg_list_destroy(&spa->spa_vdev_txg_list);
901
902 list_destroy(&spa->spa_config_dirty_list);
903 list_destroy(&spa->spa_state_dirty_list);
904
905 for (int t = 0; t < ZIO_TYPES; t++) {
906 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
907 if (spa->spa_zio_taskq[t][q] != NULL)
908 taskq_destroy(spa->spa_zio_taskq[t][q]);
909 spa->spa_zio_taskq[t][q] = NULL;
910 }
911 }
912
913 metaslab_class_destroy(spa->spa_normal_class);
914 spa->spa_normal_class = NULL;
915
916 metaslab_class_destroy(spa->spa_log_class);
917 spa->spa_log_class = NULL;
918
919 /*
920 * If this was part of an import or the open otherwise failed, we may
921 * still have errors left in the queues. Empty them just in case.
922 */
923 spa_errlog_drain(spa);
924
925 avl_destroy(&spa->spa_errlist_scrub);
926 avl_destroy(&spa->spa_errlist_last);
927
928 spa->spa_state = POOL_STATE_UNINITIALIZED;
929
930 mutex_enter(&spa->spa_proc_lock);
931 if (spa->spa_proc_state != SPA_PROC_NONE) {
932 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
933 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
934 cv_broadcast(&spa->spa_proc_cv);
935 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
936 ASSERT(spa->spa_proc != &p0);
937 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
938 }
939 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
940 spa->spa_proc_state = SPA_PROC_NONE;
941 }
942 ASSERT(spa->spa_proc == &p0);
943 mutex_exit(&spa->spa_proc_lock);
944
945#ifdef SPA_PROCESS
946 /*
947 * We want to make sure spa_thread() has actually exited the ZFS
948 * module, so that the module can't be unloaded out from underneath
949 * it.
950 */
951 if (spa->spa_did != 0) {
952 thread_join(spa->spa_did);
953 spa->spa_did = 0;
954 }
955#endif /* SPA_PROCESS */
956}
957
958/*
959 * Verify a pool configuration, and construct the vdev tree appropriately. This
960 * will create all the necessary vdevs in the appropriate layout, with each vdev
961 * in the CLOSED state. This will prep the pool before open/creation/import.
962 * All vdev validation is done by the vdev_alloc() routine.
963 */
964static int
965spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
966 uint_t id, int atype)
967{
968 nvlist_t **child;
969 uint_t children;
970 int error;
971
972 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
973 return (error);
974
975 if ((*vdp)->vdev_ops->vdev_op_leaf)
976 return (0);
977
978 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
979 &child, &children);
980
981 if (error == ENOENT)
982 return (0);
983
984 if (error) {
985 vdev_free(*vdp);
986 *vdp = NULL;
987 return (EINVAL);
988 }
989
990 for (int c = 0; c < children; c++) {
991 vdev_t *vd;
992 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
993 atype)) != 0) {
994 vdev_free(*vdp);
995 *vdp = NULL;
996 return (error);
997 }
998 }
999
1000 ASSERT(*vdp != NULL);
1001
1002 return (0);
1003}
1004
1005/*
1006 * Opposite of spa_load().
1007 */
1008static void
1009spa_unload(spa_t *spa)
1010{
1011 int i;
1012
1013 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1014
1015 /*
1016 * Stop async tasks.
1017 */
1018 spa_async_suspend(spa);
1019
1020 /*
1021 * Stop syncing.
1022 */
1023 if (spa->spa_sync_on) {
1024 txg_sync_stop(spa->spa_dsl_pool);
1025 spa->spa_sync_on = B_FALSE;
1026 }
1027
1028 /*
1029 * Wait for any outstanding async I/O to complete.
1030 */
1031 if (spa->spa_async_zio_root != NULL) {
1032 (void) zio_wait(spa->spa_async_zio_root);
1033 spa->spa_async_zio_root = NULL;
1034 }
1035
1036 bpobj_close(&spa->spa_deferred_bpobj);
1037
1038 /*
1039 * Close the dsl pool.
1040 */
1041 if (spa->spa_dsl_pool) {
1042 dsl_pool_close(spa->spa_dsl_pool);
1043 spa->spa_dsl_pool = NULL;
1044 spa->spa_meta_objset = NULL;
1045 }
1046
1047 ddt_unload(spa);
1048
1049 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1050
1051 /*
1052 * Drop and purge level 2 cache
1053 */
1054 spa_l2cache_drop(spa);
1055
1056 /*
1057 * Close all vdevs.
1058 */
1059 if (spa->spa_root_vdev)
1060 vdev_free(spa->spa_root_vdev);
1061 ASSERT(spa->spa_root_vdev == NULL);
1062
1063 for (i = 0; i < spa->spa_spares.sav_count; i++)
1064 vdev_free(spa->spa_spares.sav_vdevs[i]);
1065 if (spa->spa_spares.sav_vdevs) {
1066 kmem_free(spa->spa_spares.sav_vdevs,
1067 spa->spa_spares.sav_count * sizeof (void *));
1068 spa->spa_spares.sav_vdevs = NULL;
1069 }
1070 if (spa->spa_spares.sav_config) {
1071 nvlist_free(spa->spa_spares.sav_config);
1072 spa->spa_spares.sav_config = NULL;
1073 }
1074 spa->spa_spares.sav_count = 0;
1075
| 1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011 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/zvol.h>
64
65#ifdef _KERNEL
66#include <sys/callb.h>
67#include <sys/cpupart.h>
68#include <sys/zone.h>
69#endif /* _KERNEL */
70
71#include "zfs_prop.h"
72#include "zfs_comutil.h"
73
74/* Check hostid on import? */
75static int check_hostid = 1;
76
77SYSCTL_DECL(_vfs_zfs);
78TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
79SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
80 "Check hostid on import?");
81
82typedef enum zti_modes {
83 zti_mode_fixed, /* value is # of threads (min 1) */
84 zti_mode_online_percent, /* value is % of online CPUs */
85 zti_mode_batch, /* cpu-intensive; value is ignored */
86 zti_mode_null, /* don't create a taskq */
87 zti_nmodes
88} zti_modes_t;
89
90#define ZTI_FIX(n) { zti_mode_fixed, (n) }
91#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
92#define ZTI_BATCH { zti_mode_batch, 0 }
93#define ZTI_NULL { zti_mode_null, 0 }
94
95#define ZTI_ONE ZTI_FIX(1)
96
97typedef struct zio_taskq_info {
98 enum zti_modes zti_mode;
99 uint_t zti_value;
100} zio_taskq_info_t;
101
102static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
103 "issue", "issue_high", "intr", "intr_high"
104};
105
106/*
107 * Define the taskq threads for the following I/O types:
108 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
109 */
110const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
111 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
112 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
114 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
115 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
116 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118};
119
120static dsl_syncfunc_t spa_sync_props;
121static boolean_t spa_has_active_shared_spare(spa_t *spa);
122static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
123 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
124 char **ereport);
125static void spa_vdev_resilver_done(spa_t *spa);
126
127uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
128#ifdef PSRSET_BIND
129id_t zio_taskq_psrset_bind = PS_NONE;
130#endif
131#ifdef SYSDC
132boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
133#endif
134uint_t zio_taskq_basedc = 80; /* base duty cycle */
135
136boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
137
138/*
139 * This (illegal) pool name is used when temporarily importing a spa_t in order
140 * to get the vdev stats associated with the imported devices.
141 */
142#define TRYIMPORT_NAME "$import"
143
144/*
145 * ==========================================================================
146 * SPA properties routines
147 * ==========================================================================
148 */
149
150/*
151 * Add a (source=src, propname=propval) list to an nvlist.
152 */
153static void
154spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
155 uint64_t intval, zprop_source_t src)
156{
157 const char *propname = zpool_prop_to_name(prop);
158 nvlist_t *propval;
159
160 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
161 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
162
163 if (strval != NULL)
164 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
165 else
166 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
167
168 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
169 nvlist_free(propval);
170}
171
172/*
173 * Get property values from the spa configuration.
174 */
175static void
176spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
177{
178 uint64_t size;
179 uint64_t alloc;
180 uint64_t cap, version;
181 zprop_source_t src = ZPROP_SRC_NONE;
182 spa_config_dirent_t *dp;
183
184 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
185
186 if (spa->spa_root_vdev != NULL) {
187 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
188 size = metaslab_class_get_space(spa_normal_class(spa));
189 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
190 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
192 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
193 size - alloc, src);
194 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
195 (spa_mode(spa) == FREAD), src);
196
197 cap = (size == 0) ? 0 : (alloc * 100 / size);
198 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
199
200 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
201 ddt_get_pool_dedup_ratio(spa), src);
202
203 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
204 spa->spa_root_vdev->vdev_state, src);
205
206 version = spa_version(spa);
207 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
208 src = ZPROP_SRC_DEFAULT;
209 else
210 src = ZPROP_SRC_LOCAL;
211 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
212 }
213
214 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
215
216 if (spa->spa_comment != NULL) {
217 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
218 0, ZPROP_SRC_LOCAL);
219 }
220
221 if (spa->spa_root != NULL)
222 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
223 0, ZPROP_SRC_LOCAL);
224
225 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
226 if (dp->scd_path == NULL) {
227 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
228 "none", 0, ZPROP_SRC_LOCAL);
229 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
230 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
231 dp->scd_path, 0, ZPROP_SRC_LOCAL);
232 }
233 }
234}
235
236/*
237 * Get zpool property values.
238 */
239int
240spa_prop_get(spa_t *spa, nvlist_t **nvp)
241{
242 objset_t *mos = spa->spa_meta_objset;
243 zap_cursor_t zc;
244 zap_attribute_t za;
245 int err;
246
247 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
248
249 mutex_enter(&spa->spa_props_lock);
250
251 /*
252 * Get properties from the spa config.
253 */
254 spa_prop_get_config(spa, nvp);
255
256 /* If no pool property object, no more prop to get. */
257 if (mos == NULL || spa->spa_pool_props_object == 0) {
258 mutex_exit(&spa->spa_props_lock);
259 return (0);
260 }
261
262 /*
263 * Get properties from the MOS pool property object.
264 */
265 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
266 (err = zap_cursor_retrieve(&zc, &za)) == 0;
267 zap_cursor_advance(&zc)) {
268 uint64_t intval = 0;
269 char *strval = NULL;
270 zprop_source_t src = ZPROP_SRC_DEFAULT;
271 zpool_prop_t prop;
272
273 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
274 continue;
275
276 switch (za.za_integer_length) {
277 case 8:
278 /* integer property */
279 if (za.za_first_integer !=
280 zpool_prop_default_numeric(prop))
281 src = ZPROP_SRC_LOCAL;
282
283 if (prop == ZPOOL_PROP_BOOTFS) {
284 dsl_pool_t *dp;
285 dsl_dataset_t *ds = NULL;
286
287 dp = spa_get_dsl(spa);
288 rw_enter(&dp->dp_config_rwlock, RW_READER);
289 if (err = dsl_dataset_hold_obj(dp,
290 za.za_first_integer, FTAG, &ds)) {
291 rw_exit(&dp->dp_config_rwlock);
292 break;
293 }
294
295 strval = kmem_alloc(
296 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
297 KM_SLEEP);
298 dsl_dataset_name(ds, strval);
299 dsl_dataset_rele(ds, FTAG);
300 rw_exit(&dp->dp_config_rwlock);
301 } else {
302 strval = NULL;
303 intval = za.za_first_integer;
304 }
305
306 spa_prop_add_list(*nvp, prop, strval, intval, src);
307
308 if (strval != NULL)
309 kmem_free(strval,
310 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
311
312 break;
313
314 case 1:
315 /* string property */
316 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
317 err = zap_lookup(mos, spa->spa_pool_props_object,
318 za.za_name, 1, za.za_num_integers, strval);
319 if (err) {
320 kmem_free(strval, za.za_num_integers);
321 break;
322 }
323 spa_prop_add_list(*nvp, prop, strval, 0, src);
324 kmem_free(strval, za.za_num_integers);
325 break;
326
327 default:
328 break;
329 }
330 }
331 zap_cursor_fini(&zc);
332 mutex_exit(&spa->spa_props_lock);
333out:
334 if (err && err != ENOENT) {
335 nvlist_free(*nvp);
336 *nvp = NULL;
337 return (err);
338 }
339
340 return (0);
341}
342
343/*
344 * Validate the given pool properties nvlist and modify the list
345 * for the property values to be set.
346 */
347static int
348spa_prop_validate(spa_t *spa, nvlist_t *props)
349{
350 nvpair_t *elem;
351 int error = 0, reset_bootfs = 0;
352 uint64_t objnum;
353
354 elem = NULL;
355 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
356 zpool_prop_t prop;
357 char *propname, *strval;
358 uint64_t intval;
359 objset_t *os;
360 char *slash, *check;
361
362 propname = nvpair_name(elem);
363
364 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
365 return (EINVAL);
366
367 switch (prop) {
368 case ZPOOL_PROP_VERSION:
369 error = nvpair_value_uint64(elem, &intval);
370 if (!error &&
371 (intval < spa_version(spa) || intval > SPA_VERSION))
372 error = EINVAL;
373 break;
374
375 case ZPOOL_PROP_DELEGATION:
376 case ZPOOL_PROP_AUTOREPLACE:
377 case ZPOOL_PROP_LISTSNAPS:
378 case ZPOOL_PROP_AUTOEXPAND:
379 error = nvpair_value_uint64(elem, &intval);
380 if (!error && intval > 1)
381 error = EINVAL;
382 break;
383
384 case ZPOOL_PROP_BOOTFS:
385 /*
386 * If the pool version is less than SPA_VERSION_BOOTFS,
387 * or the pool is still being created (version == 0),
388 * the bootfs property cannot be set.
389 */
390 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
391 error = ENOTSUP;
392 break;
393 }
394
395 /*
396 * Make sure the vdev config is bootable
397 */
398 if (!vdev_is_bootable(spa->spa_root_vdev)) {
399 error = ENOTSUP;
400 break;
401 }
402
403 reset_bootfs = 1;
404
405 error = nvpair_value_string(elem, &strval);
406
407 if (!error) {
408 uint64_t compress;
409
410 if (strval == NULL || strval[0] == '\0') {
411 objnum = zpool_prop_default_numeric(
412 ZPOOL_PROP_BOOTFS);
413 break;
414 }
415
416 if (error = dmu_objset_hold(strval, FTAG, &os))
417 break;
418
419 /* Must be ZPL and not gzip compressed. */
420
421 if (dmu_objset_type(os) != DMU_OST_ZFS) {
422 error = ENOTSUP;
423 } else if ((error = dsl_prop_get_integer(strval,
424 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
425 &compress, NULL)) == 0 &&
426 !BOOTFS_COMPRESS_VALID(compress)) {
427 error = ENOTSUP;
428 } else {
429 objnum = dmu_objset_id(os);
430 }
431 dmu_objset_rele(os, FTAG);
432 }
433 break;
434
435 case ZPOOL_PROP_FAILUREMODE:
436 error = nvpair_value_uint64(elem, &intval);
437 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
438 intval > ZIO_FAILURE_MODE_PANIC))
439 error = EINVAL;
440
441 /*
442 * This is a special case which only occurs when
443 * the pool has completely failed. This allows
444 * the user to change the in-core failmode property
445 * without syncing it out to disk (I/Os might
446 * currently be blocked). We do this by returning
447 * EIO to the caller (spa_prop_set) to trick it
448 * into thinking we encountered a property validation
449 * error.
450 */
451 if (!error && spa_suspended(spa)) {
452 spa->spa_failmode = intval;
453 error = EIO;
454 }
455 break;
456
457 case ZPOOL_PROP_CACHEFILE:
458 if ((error = nvpair_value_string(elem, &strval)) != 0)
459 break;
460
461 if (strval[0] == '\0')
462 break;
463
464 if (strcmp(strval, "none") == 0)
465 break;
466
467 if (strval[0] != '/') {
468 error = EINVAL;
469 break;
470 }
471
472 slash = strrchr(strval, '/');
473 ASSERT(slash != NULL);
474
475 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
476 strcmp(slash, "/..") == 0)
477 error = EINVAL;
478 break;
479
480 case ZPOOL_PROP_COMMENT:
481 if ((error = nvpair_value_string(elem, &strval)) != 0)
482 break;
483 for (check = strval; *check != '\0'; check++) {
484 /*
485 * The kernel doesn't have an easy isprint()
486 * check. For this kernel check, we merely
487 * check ASCII apart from DEL. Fix this if
488 * there is an easy-to-use kernel isprint().
489 */
490 if (*check >= 0x7f) {
491 error = EINVAL;
492 break;
493 }
494 check++;
495 }
496 if (strlen(strval) > ZPROP_MAX_COMMENT)
497 error = E2BIG;
498 break;
499
500 case ZPOOL_PROP_DEDUPDITTO:
501 if (spa_version(spa) < SPA_VERSION_DEDUP)
502 error = ENOTSUP;
503 else
504 error = nvpair_value_uint64(elem, &intval);
505 if (error == 0 &&
506 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
507 error = EINVAL;
508 break;
509 }
510
511 if (error)
512 break;
513 }
514
515 if (!error && reset_bootfs) {
516 error = nvlist_remove(props,
517 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
518
519 if (!error) {
520 error = nvlist_add_uint64(props,
521 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
522 }
523 }
524
525 return (error);
526}
527
528void
529spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
530{
531 char *cachefile;
532 spa_config_dirent_t *dp;
533
534 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
535 &cachefile) != 0)
536 return;
537
538 dp = kmem_alloc(sizeof (spa_config_dirent_t),
539 KM_SLEEP);
540
541 if (cachefile[0] == '\0')
542 dp->scd_path = spa_strdup(spa_config_path);
543 else if (strcmp(cachefile, "none") == 0)
544 dp->scd_path = NULL;
545 else
546 dp->scd_path = spa_strdup(cachefile);
547
548 list_insert_head(&spa->spa_config_list, dp);
549 if (need_sync)
550 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
551}
552
553int
554spa_prop_set(spa_t *spa, nvlist_t *nvp)
555{
556 int error;
557 nvpair_t *elem;
558 boolean_t need_sync = B_FALSE;
559 zpool_prop_t prop;
560
561 if ((error = spa_prop_validate(spa, nvp)) != 0)
562 return (error);
563
564 elem = NULL;
565 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
566 if ((prop = zpool_name_to_prop(
567 nvpair_name(elem))) == ZPROP_INVAL)
568 return (EINVAL);
569
570 if (prop == ZPOOL_PROP_CACHEFILE ||
571 prop == ZPOOL_PROP_ALTROOT ||
572 prop == ZPOOL_PROP_READONLY)
573 continue;
574
575 need_sync = B_TRUE;
576 break;
577 }
578
579 if (need_sync)
580 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
581 spa, nvp, 3));
582 else
583 return (0);
584}
585
586/*
587 * If the bootfs property value is dsobj, clear it.
588 */
589void
590spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
591{
592 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
593 VERIFY(zap_remove(spa->spa_meta_objset,
594 spa->spa_pool_props_object,
595 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
596 spa->spa_bootfs = 0;
597 }
598}
599
600/*
601 * Change the GUID for the pool. This is done so that we can later
602 * re-import a pool built from a clone of our own vdevs. We will modify
603 * the root vdev's guid, our own pool guid, and then mark all of our
604 * vdevs dirty. Note that we must make sure that all our vdevs are
605 * online when we do this, or else any vdevs that weren't present
606 * would be orphaned from our pool. We are also going to issue a
607 * sysevent to update any watchers.
608 */
609int
610spa_change_guid(spa_t *spa)
611{
612 uint64_t oldguid, newguid;
613 uint64_t txg;
614
615 if (!(spa_mode_global & FWRITE))
616 return (EROFS);
617
618 txg = spa_vdev_enter(spa);
619
620 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
621 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
622
623 oldguid = spa_guid(spa);
624 newguid = spa_generate_guid(NULL);
625 ASSERT3U(oldguid, !=, newguid);
626
627 spa->spa_root_vdev->vdev_guid = newguid;
628 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
629
630 vdev_config_dirty(spa->spa_root_vdev);
631
632 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
633
634 return (spa_vdev_exit(spa, NULL, txg, 0));
635}
636
637/*
638 * ==========================================================================
639 * SPA state manipulation (open/create/destroy/import/export)
640 * ==========================================================================
641 */
642
643static int
644spa_error_entry_compare(const void *a, const void *b)
645{
646 spa_error_entry_t *sa = (spa_error_entry_t *)a;
647 spa_error_entry_t *sb = (spa_error_entry_t *)b;
648 int ret;
649
650 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
651 sizeof (zbookmark_t));
652
653 if (ret < 0)
654 return (-1);
655 else if (ret > 0)
656 return (1);
657 else
658 return (0);
659}
660
661/*
662 * Utility function which retrieves copies of the current logs and
663 * re-initializes them in the process.
664 */
665void
666spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
667{
668 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
669
670 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
671 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
672
673 avl_create(&spa->spa_errlist_scrub,
674 spa_error_entry_compare, sizeof (spa_error_entry_t),
675 offsetof(spa_error_entry_t, se_avl));
676 avl_create(&spa->spa_errlist_last,
677 spa_error_entry_compare, sizeof (spa_error_entry_t),
678 offsetof(spa_error_entry_t, se_avl));
679}
680
681static taskq_t *
682spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
683 uint_t value)
684{
685 uint_t flags = TASKQ_PREPOPULATE;
686 boolean_t batch = B_FALSE;
687
688 switch (mode) {
689 case zti_mode_null:
690 return (NULL); /* no taskq needed */
691
692 case zti_mode_fixed:
693 ASSERT3U(value, >=, 1);
694 value = MAX(value, 1);
695 break;
696
697 case zti_mode_batch:
698 batch = B_TRUE;
699 flags |= TASKQ_THREADS_CPU_PCT;
700 value = zio_taskq_batch_pct;
701 break;
702
703 case zti_mode_online_percent:
704 flags |= TASKQ_THREADS_CPU_PCT;
705 break;
706
707 default:
708 panic("unrecognized mode for %s taskq (%u:%u) in "
709 "spa_activate()",
710 name, mode, value);
711 break;
712 }
713
714#ifdef SYSDC
715 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
716 if (batch)
717 flags |= TASKQ_DC_BATCH;
718
719 return (taskq_create_sysdc(name, value, 50, INT_MAX,
720 spa->spa_proc, zio_taskq_basedc, flags));
721 }
722#endif
723 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
724 spa->spa_proc, flags));
725}
726
727static void
728spa_create_zio_taskqs(spa_t *spa)
729{
730 for (int t = 0; t < ZIO_TYPES; t++) {
731 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
732 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
733 enum zti_modes mode = ztip->zti_mode;
734 uint_t value = ztip->zti_value;
735 char name[32];
736
737 (void) snprintf(name, sizeof (name),
738 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
739
740 spa->spa_zio_taskq[t][q] =
741 spa_taskq_create(spa, name, mode, value);
742 }
743 }
744}
745
746#ifdef _KERNEL
747#ifdef SPA_PROCESS
748static void
749spa_thread(void *arg)
750{
751 callb_cpr_t cprinfo;
752
753 spa_t *spa = arg;
754 user_t *pu = PTOU(curproc);
755
756 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
757 spa->spa_name);
758
759 ASSERT(curproc != &p0);
760 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
761 "zpool-%s", spa->spa_name);
762 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
763
764#ifdef PSRSET_BIND
765 /* bind this thread to the requested psrset */
766 if (zio_taskq_psrset_bind != PS_NONE) {
767 pool_lock();
768 mutex_enter(&cpu_lock);
769 mutex_enter(&pidlock);
770 mutex_enter(&curproc->p_lock);
771
772 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
773 0, NULL, NULL) == 0) {
774 curthread->t_bind_pset = zio_taskq_psrset_bind;
775 } else {
776 cmn_err(CE_WARN,
777 "Couldn't bind process for zfs pool \"%s\" to "
778 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
779 }
780
781 mutex_exit(&curproc->p_lock);
782 mutex_exit(&pidlock);
783 mutex_exit(&cpu_lock);
784 pool_unlock();
785 }
786#endif
787
788#ifdef SYSDC
789 if (zio_taskq_sysdc) {
790 sysdc_thread_enter(curthread, 100, 0);
791 }
792#endif
793
794 spa->spa_proc = curproc;
795 spa->spa_did = curthread->t_did;
796
797 spa_create_zio_taskqs(spa);
798
799 mutex_enter(&spa->spa_proc_lock);
800 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
801
802 spa->spa_proc_state = SPA_PROC_ACTIVE;
803 cv_broadcast(&spa->spa_proc_cv);
804
805 CALLB_CPR_SAFE_BEGIN(&cprinfo);
806 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
807 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
808 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
809
810 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
811 spa->spa_proc_state = SPA_PROC_GONE;
812 spa->spa_proc = &p0;
813 cv_broadcast(&spa->spa_proc_cv);
814 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
815
816 mutex_enter(&curproc->p_lock);
817 lwp_exit();
818}
819#endif /* SPA_PROCESS */
820#endif
821
822/*
823 * Activate an uninitialized pool.
824 */
825static void
826spa_activate(spa_t *spa, int mode)
827{
828 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
829
830 spa->spa_state = POOL_STATE_ACTIVE;
831 spa->spa_mode = mode;
832
833 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
834 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
835
836 /* Try to create a covering process */
837 mutex_enter(&spa->spa_proc_lock);
838 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
839 ASSERT(spa->spa_proc == &p0);
840 spa->spa_did = 0;
841
842#ifdef SPA_PROCESS
843 /* Only create a process if we're going to be around a while. */
844 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
845 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
846 NULL, 0) == 0) {
847 spa->spa_proc_state = SPA_PROC_CREATED;
848 while (spa->spa_proc_state == SPA_PROC_CREATED) {
849 cv_wait(&spa->spa_proc_cv,
850 &spa->spa_proc_lock);
851 }
852 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
853 ASSERT(spa->spa_proc != &p0);
854 ASSERT(spa->spa_did != 0);
855 } else {
856#ifdef _KERNEL
857 cmn_err(CE_WARN,
858 "Couldn't create process for zfs pool \"%s\"\n",
859 spa->spa_name);
860#endif
861 }
862 }
863#endif /* SPA_PROCESS */
864 mutex_exit(&spa->spa_proc_lock);
865
866 /* If we didn't create a process, we need to create our taskqs. */
867 ASSERT(spa->spa_proc == &p0);
868 if (spa->spa_proc == &p0) {
869 spa_create_zio_taskqs(spa);
870 }
871
872 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
873 offsetof(vdev_t, vdev_config_dirty_node));
874 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
875 offsetof(vdev_t, vdev_state_dirty_node));
876
877 txg_list_create(&spa->spa_vdev_txg_list,
878 offsetof(struct vdev, vdev_txg_node));
879
880 avl_create(&spa->spa_errlist_scrub,
881 spa_error_entry_compare, sizeof (spa_error_entry_t),
882 offsetof(spa_error_entry_t, se_avl));
883 avl_create(&spa->spa_errlist_last,
884 spa_error_entry_compare, sizeof (spa_error_entry_t),
885 offsetof(spa_error_entry_t, se_avl));
886}
887
888/*
889 * Opposite of spa_activate().
890 */
891static void
892spa_deactivate(spa_t *spa)
893{
894 ASSERT(spa->spa_sync_on == B_FALSE);
895 ASSERT(spa->spa_dsl_pool == NULL);
896 ASSERT(spa->spa_root_vdev == NULL);
897 ASSERT(spa->spa_async_zio_root == NULL);
898 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
899
900 txg_list_destroy(&spa->spa_vdev_txg_list);
901
902 list_destroy(&spa->spa_config_dirty_list);
903 list_destroy(&spa->spa_state_dirty_list);
904
905 for (int t = 0; t < ZIO_TYPES; t++) {
906 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
907 if (spa->spa_zio_taskq[t][q] != NULL)
908 taskq_destroy(spa->spa_zio_taskq[t][q]);
909 spa->spa_zio_taskq[t][q] = NULL;
910 }
911 }
912
913 metaslab_class_destroy(spa->spa_normal_class);
914 spa->spa_normal_class = NULL;
915
916 metaslab_class_destroy(spa->spa_log_class);
917 spa->spa_log_class = NULL;
918
919 /*
920 * If this was part of an import or the open otherwise failed, we may
921 * still have errors left in the queues. Empty them just in case.
922 */
923 spa_errlog_drain(spa);
924
925 avl_destroy(&spa->spa_errlist_scrub);
926 avl_destroy(&spa->spa_errlist_last);
927
928 spa->spa_state = POOL_STATE_UNINITIALIZED;
929
930 mutex_enter(&spa->spa_proc_lock);
931 if (spa->spa_proc_state != SPA_PROC_NONE) {
932 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
933 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
934 cv_broadcast(&spa->spa_proc_cv);
935 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
936 ASSERT(spa->spa_proc != &p0);
937 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
938 }
939 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
940 spa->spa_proc_state = SPA_PROC_NONE;
941 }
942 ASSERT(spa->spa_proc == &p0);
943 mutex_exit(&spa->spa_proc_lock);
944
945#ifdef SPA_PROCESS
946 /*
947 * We want to make sure spa_thread() has actually exited the ZFS
948 * module, so that the module can't be unloaded out from underneath
949 * it.
950 */
951 if (spa->spa_did != 0) {
952 thread_join(spa->spa_did);
953 spa->spa_did = 0;
954 }
955#endif /* SPA_PROCESS */
956}
957
958/*
959 * Verify a pool configuration, and construct the vdev tree appropriately. This
960 * will create all the necessary vdevs in the appropriate layout, with each vdev
961 * in the CLOSED state. This will prep the pool before open/creation/import.
962 * All vdev validation is done by the vdev_alloc() routine.
963 */
964static int
965spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
966 uint_t id, int atype)
967{
968 nvlist_t **child;
969 uint_t children;
970 int error;
971
972 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
973 return (error);
974
975 if ((*vdp)->vdev_ops->vdev_op_leaf)
976 return (0);
977
978 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
979 &child, &children);
980
981 if (error == ENOENT)
982 return (0);
983
984 if (error) {
985 vdev_free(*vdp);
986 *vdp = NULL;
987 return (EINVAL);
988 }
989
990 for (int c = 0; c < children; c++) {
991 vdev_t *vd;
992 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
993 atype)) != 0) {
994 vdev_free(*vdp);
995 *vdp = NULL;
996 return (error);
997 }
998 }
999
1000 ASSERT(*vdp != NULL);
1001
1002 return (0);
1003}
1004
1005/*
1006 * Opposite of spa_load().
1007 */
1008static void
1009spa_unload(spa_t *spa)
1010{
1011 int i;
1012
1013 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1014
1015 /*
1016 * Stop async tasks.
1017 */
1018 spa_async_suspend(spa);
1019
1020 /*
1021 * Stop syncing.
1022 */
1023 if (spa->spa_sync_on) {
1024 txg_sync_stop(spa->spa_dsl_pool);
1025 spa->spa_sync_on = B_FALSE;
1026 }
1027
1028 /*
1029 * Wait for any outstanding async I/O to complete.
1030 */
1031 if (spa->spa_async_zio_root != NULL) {
1032 (void) zio_wait(spa->spa_async_zio_root);
1033 spa->spa_async_zio_root = NULL;
1034 }
1035
1036 bpobj_close(&spa->spa_deferred_bpobj);
1037
1038 /*
1039 * Close the dsl pool.
1040 */
1041 if (spa->spa_dsl_pool) {
1042 dsl_pool_close(spa->spa_dsl_pool);
1043 spa->spa_dsl_pool = NULL;
1044 spa->spa_meta_objset = NULL;
1045 }
1046
1047 ddt_unload(spa);
1048
1049 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1050
1051 /*
1052 * Drop and purge level 2 cache
1053 */
1054 spa_l2cache_drop(spa);
1055
1056 /*
1057 * Close all vdevs.
1058 */
1059 if (spa->spa_root_vdev)
1060 vdev_free(spa->spa_root_vdev);
1061 ASSERT(spa->spa_root_vdev == NULL);
1062
1063 for (i = 0; i < spa->spa_spares.sav_count; i++)
1064 vdev_free(spa->spa_spares.sav_vdevs[i]);
1065 if (spa->spa_spares.sav_vdevs) {
1066 kmem_free(spa->spa_spares.sav_vdevs,
1067 spa->spa_spares.sav_count * sizeof (void *));
1068 spa->spa_spares.sav_vdevs = NULL;
1069 }
1070 if (spa->spa_spares.sav_config) {
1071 nvlist_free(spa->spa_spares.sav_config);
1072 spa->spa_spares.sav_config = NULL;
1073 }
1074 spa->spa_spares.sav_count = 0;
1075
|
1076 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
| 1076 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1077 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
|
1077 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
| 1078 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
|
| 1079 }
|
1078 if (spa->spa_l2cache.sav_vdevs) {
1079 kmem_free(spa->spa_l2cache.sav_vdevs,
1080 spa->spa_l2cache.sav_count * sizeof (void *));
1081 spa->spa_l2cache.sav_vdevs = NULL;
1082 }
1083 if (spa->spa_l2cache.sav_config) {
1084 nvlist_free(spa->spa_l2cache.sav_config);
1085 spa->spa_l2cache.sav_config = NULL;
1086 }
1087 spa->spa_l2cache.sav_count = 0;
1088
1089 spa->spa_async_suspended = 0;
1090
1091 if (spa->spa_comment != NULL) {
1092 spa_strfree(spa->spa_comment);
1093 spa->spa_comment = NULL;
1094 }
1095
1096 spa_config_exit(spa, SCL_ALL, FTAG);
1097}
1098
1099/*
1100 * Load (or re-load) the current list of vdevs describing the active spares for
1101 * this pool. When this is called, we have some form of basic information in
1102 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1103 * then re-generate a more complete list including status information.
1104 */
1105static void
1106spa_load_spares(spa_t *spa)
1107{
1108 nvlist_t **spares;
1109 uint_t nspares;
1110 int i;
1111 vdev_t *vd, *tvd;
1112
1113 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1114
1115 /*
1116 * First, close and free any existing spare vdevs.
1117 */
1118 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1119 vd = spa->spa_spares.sav_vdevs[i];
1120
1121 /* Undo the call to spa_activate() below */
1122 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1123 B_FALSE)) != NULL && tvd->vdev_isspare)
1124 spa_spare_remove(tvd);
1125 vdev_close(vd);
1126 vdev_free(vd);
1127 }
1128
1129 if (spa->spa_spares.sav_vdevs)
1130 kmem_free(spa->spa_spares.sav_vdevs,
1131 spa->spa_spares.sav_count * sizeof (void *));
1132
1133 if (spa->spa_spares.sav_config == NULL)
1134 nspares = 0;
1135 else
1136 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1137 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1138
1139 spa->spa_spares.sav_count = (int)nspares;
1140 spa->spa_spares.sav_vdevs = NULL;
1141
1142 if (nspares == 0)
1143 return;
1144
1145 /*
1146 * Construct the array of vdevs, opening them to get status in the
1147 * process. For each spare, there is potentially two different vdev_t
1148 * structures associated with it: one in the list of spares (used only
1149 * for basic validation purposes) and one in the active vdev
1150 * configuration (if it's spared in). During this phase we open and
1151 * validate each vdev on the spare list. If the vdev also exists in the
1152 * active configuration, then we also mark this vdev as an active spare.
1153 */
1154 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1155 KM_SLEEP);
1156 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1157 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1158 VDEV_ALLOC_SPARE) == 0);
1159 ASSERT(vd != NULL);
1160
1161 spa->spa_spares.sav_vdevs[i] = vd;
1162
1163 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1164 B_FALSE)) != NULL) {
1165 if (!tvd->vdev_isspare)
1166 spa_spare_add(tvd);
1167
1168 /*
1169 * We only mark the spare active if we were successfully
1170 * able to load the vdev. Otherwise, importing a pool
1171 * with a bad active spare would result in strange
1172 * behavior, because multiple pool would think the spare
1173 * is actively in use.
1174 *
1175 * There is a vulnerability here to an equally bizarre
1176 * circumstance, where a dead active spare is later
1177 * brought back to life (onlined or otherwise). Given
1178 * the rarity of this scenario, and the extra complexity
1179 * it adds, we ignore the possibility.
1180 */
1181 if (!vdev_is_dead(tvd))
1182 spa_spare_activate(tvd);
1183 }
1184
1185 vd->vdev_top = vd;
1186 vd->vdev_aux = &spa->spa_spares;
1187
1188 if (vdev_open(vd) != 0)
1189 continue;
1190
1191 if (vdev_validate_aux(vd) == 0)
1192 spa_spare_add(vd);
1193 }
1194
1195 /*
1196 * Recompute the stashed list of spares, with status information
1197 * this time.
1198 */
1199 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1200 DATA_TYPE_NVLIST_ARRAY) == 0);
1201
1202 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1203 KM_SLEEP);
1204 for (i = 0; i < spa->spa_spares.sav_count; i++)
1205 spares[i] = vdev_config_generate(spa,
1206 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1207 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1208 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1209 for (i = 0; i < spa->spa_spares.sav_count; i++)
1210 nvlist_free(spares[i]);
1211 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1212}
1213
1214/*
1215 * Load (or re-load) the current list of vdevs describing the active l2cache for
1216 * this pool. When this is called, we have some form of basic information in
1217 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1218 * then re-generate a more complete list including status information.
1219 * Devices which are already active have their details maintained, and are
1220 * not re-opened.
1221 */
1222static void
1223spa_load_l2cache(spa_t *spa)
1224{
1225 nvlist_t **l2cache;
1226 uint_t nl2cache;
1227 int i, j, oldnvdevs;
1228 uint64_t guid;
1229 vdev_t *vd, **oldvdevs, **newvdevs;
1230 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1231
1232 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1233
1234 if (sav->sav_config != NULL) {
1235 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1236 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1237 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1238 } else {
1239 nl2cache = 0;
1240 }
1241
1242 oldvdevs = sav->sav_vdevs;
1243 oldnvdevs = sav->sav_count;
1244 sav->sav_vdevs = NULL;
1245 sav->sav_count = 0;
1246
1247 /*
1248 * Process new nvlist of vdevs.
1249 */
1250 for (i = 0; i < nl2cache; i++) {
1251 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1252 &guid) == 0);
1253
1254 newvdevs[i] = NULL;
1255 for (j = 0; j < oldnvdevs; j++) {
1256 vd = oldvdevs[j];
1257 if (vd != NULL && guid == vd->vdev_guid) {
1258 /*
1259 * Retain previous vdev for add/remove ops.
1260 */
1261 newvdevs[i] = vd;
1262 oldvdevs[j] = NULL;
1263 break;
1264 }
1265 }
1266
1267 if (newvdevs[i] == NULL) {
1268 /*
1269 * Create new vdev
1270 */
1271 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1272 VDEV_ALLOC_L2CACHE) == 0);
1273 ASSERT(vd != NULL);
1274 newvdevs[i] = vd;
1275
1276 /*
1277 * Commit this vdev as an l2cache device,
1278 * even if it fails to open.
1279 */
1280 spa_l2cache_add(vd);
1281
1282 vd->vdev_top = vd;
1283 vd->vdev_aux = sav;
1284
1285 spa_l2cache_activate(vd);
1286
1287 if (vdev_open(vd) != 0)
1288 continue;
1289
1290 (void) vdev_validate_aux(vd);
1291
1292 if (!vdev_is_dead(vd))
1293 l2arc_add_vdev(spa, vd);
1294 }
1295 }
1296
1297 /*
1298 * Purge vdevs that were dropped
1299 */
1300 for (i = 0; i < oldnvdevs; i++) {
1301 uint64_t pool;
1302
1303 vd = oldvdevs[i];
1304 if (vd != NULL) {
| 1080 if (spa->spa_l2cache.sav_vdevs) {
1081 kmem_free(spa->spa_l2cache.sav_vdevs,
1082 spa->spa_l2cache.sav_count * sizeof (void *));
1083 spa->spa_l2cache.sav_vdevs = NULL;
1084 }
1085 if (spa->spa_l2cache.sav_config) {
1086 nvlist_free(spa->spa_l2cache.sav_config);
1087 spa->spa_l2cache.sav_config = NULL;
1088 }
1089 spa->spa_l2cache.sav_count = 0;
1090
1091 spa->spa_async_suspended = 0;
1092
1093 if (spa->spa_comment != NULL) {
1094 spa_strfree(spa->spa_comment);
1095 spa->spa_comment = NULL;
1096 }
1097
1098 spa_config_exit(spa, SCL_ALL, FTAG);
1099}
1100
1101/*
1102 * Load (or re-load) the current list of vdevs describing the active spares for
1103 * this pool. When this is called, we have some form of basic information in
1104 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1105 * then re-generate a more complete list including status information.
1106 */
1107static void
1108spa_load_spares(spa_t *spa)
1109{
1110 nvlist_t **spares;
1111 uint_t nspares;
1112 int i;
1113 vdev_t *vd, *tvd;
1114
1115 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1116
1117 /*
1118 * First, close and free any existing spare vdevs.
1119 */
1120 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1121 vd = spa->spa_spares.sav_vdevs[i];
1122
1123 /* Undo the call to spa_activate() below */
1124 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1125 B_FALSE)) != NULL && tvd->vdev_isspare)
1126 spa_spare_remove(tvd);
1127 vdev_close(vd);
1128 vdev_free(vd);
1129 }
1130
1131 if (spa->spa_spares.sav_vdevs)
1132 kmem_free(spa->spa_spares.sav_vdevs,
1133 spa->spa_spares.sav_count * sizeof (void *));
1134
1135 if (spa->spa_spares.sav_config == NULL)
1136 nspares = 0;
1137 else
1138 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1139 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1140
1141 spa->spa_spares.sav_count = (int)nspares;
1142 spa->spa_spares.sav_vdevs = NULL;
1143
1144 if (nspares == 0)
1145 return;
1146
1147 /*
1148 * Construct the array of vdevs, opening them to get status in the
1149 * process. For each spare, there is potentially two different vdev_t
1150 * structures associated with it: one in the list of spares (used only
1151 * for basic validation purposes) and one in the active vdev
1152 * configuration (if it's spared in). During this phase we open and
1153 * validate each vdev on the spare list. If the vdev also exists in the
1154 * active configuration, then we also mark this vdev as an active spare.
1155 */
1156 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1157 KM_SLEEP);
1158 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1159 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1160 VDEV_ALLOC_SPARE) == 0);
1161 ASSERT(vd != NULL);
1162
1163 spa->spa_spares.sav_vdevs[i] = vd;
1164
1165 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1166 B_FALSE)) != NULL) {
1167 if (!tvd->vdev_isspare)
1168 spa_spare_add(tvd);
1169
1170 /*
1171 * We only mark the spare active if we were successfully
1172 * able to load the vdev. Otherwise, importing a pool
1173 * with a bad active spare would result in strange
1174 * behavior, because multiple pool would think the spare
1175 * is actively in use.
1176 *
1177 * There is a vulnerability here to an equally bizarre
1178 * circumstance, where a dead active spare is later
1179 * brought back to life (onlined or otherwise). Given
1180 * the rarity of this scenario, and the extra complexity
1181 * it adds, we ignore the possibility.
1182 */
1183 if (!vdev_is_dead(tvd))
1184 spa_spare_activate(tvd);
1185 }
1186
1187 vd->vdev_top = vd;
1188 vd->vdev_aux = &spa->spa_spares;
1189
1190 if (vdev_open(vd) != 0)
1191 continue;
1192
1193 if (vdev_validate_aux(vd) == 0)
1194 spa_spare_add(vd);
1195 }
1196
1197 /*
1198 * Recompute the stashed list of spares, with status information
1199 * this time.
1200 */
1201 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1202 DATA_TYPE_NVLIST_ARRAY) == 0);
1203
1204 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1205 KM_SLEEP);
1206 for (i = 0; i < spa->spa_spares.sav_count; i++)
1207 spares[i] = vdev_config_generate(spa,
1208 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1209 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1210 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1211 for (i = 0; i < spa->spa_spares.sav_count; i++)
1212 nvlist_free(spares[i]);
1213 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1214}
1215
1216/*
1217 * Load (or re-load) the current list of vdevs describing the active l2cache for
1218 * this pool. When this is called, we have some form of basic information in
1219 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1220 * then re-generate a more complete list including status information.
1221 * Devices which are already active have their details maintained, and are
1222 * not re-opened.
1223 */
1224static void
1225spa_load_l2cache(spa_t *spa)
1226{
1227 nvlist_t **l2cache;
1228 uint_t nl2cache;
1229 int i, j, oldnvdevs;
1230 uint64_t guid;
1231 vdev_t *vd, **oldvdevs, **newvdevs;
1232 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1233
1234 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1235
1236 if (sav->sav_config != NULL) {
1237 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1238 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1239 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1240 } else {
1241 nl2cache = 0;
1242 }
1243
1244 oldvdevs = sav->sav_vdevs;
1245 oldnvdevs = sav->sav_count;
1246 sav->sav_vdevs = NULL;
1247 sav->sav_count = 0;
1248
1249 /*
1250 * Process new nvlist of vdevs.
1251 */
1252 for (i = 0; i < nl2cache; i++) {
1253 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1254 &guid) == 0);
1255
1256 newvdevs[i] = NULL;
1257 for (j = 0; j < oldnvdevs; j++) {
1258 vd = oldvdevs[j];
1259 if (vd != NULL && guid == vd->vdev_guid) {
1260 /*
1261 * Retain previous vdev for add/remove ops.
1262 */
1263 newvdevs[i] = vd;
1264 oldvdevs[j] = NULL;
1265 break;
1266 }
1267 }
1268
1269 if (newvdevs[i] == NULL) {
1270 /*
1271 * Create new vdev
1272 */
1273 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1274 VDEV_ALLOC_L2CACHE) == 0);
1275 ASSERT(vd != NULL);
1276 newvdevs[i] = vd;
1277
1278 /*
1279 * Commit this vdev as an l2cache device,
1280 * even if it fails to open.
1281 */
1282 spa_l2cache_add(vd);
1283
1284 vd->vdev_top = vd;
1285 vd->vdev_aux = sav;
1286
1287 spa_l2cache_activate(vd);
1288
1289 if (vdev_open(vd) != 0)
1290 continue;
1291
1292 (void) vdev_validate_aux(vd);
1293
1294 if (!vdev_is_dead(vd))
1295 l2arc_add_vdev(spa, vd);
1296 }
1297 }
1298
1299 /*
1300 * Purge vdevs that were dropped
1301 */
1302 for (i = 0; i < oldnvdevs; i++) {
1303 uint64_t pool;
1304
1305 vd = oldvdevs[i];
1306 if (vd != NULL) {
|
| 1307 ASSERT(vd->vdev_isl2cache);
1308
|
1305 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1306 pool != 0ULL && l2arc_vdev_present(vd))
1307 l2arc_remove_vdev(vd);
| 1309 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1310 pool != 0ULL && l2arc_vdev_present(vd))
1311 l2arc_remove_vdev(vd);
|
1308 (void) vdev_close(vd);
1309 spa_l2cache_remove(vd);
| 1312 vdev_clear_stats(vd);
1313 vdev_free(vd);
|
1310 }
1311 }
1312
1313 if (oldvdevs)
1314 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1315
1316 if (sav->sav_config == NULL)
1317 goto out;
1318
1319 sav->sav_vdevs = newvdevs;
1320 sav->sav_count = (int)nl2cache;
1321
1322 /*
1323 * Recompute the stashed list of l2cache devices, with status
1324 * information this time.
1325 */
1326 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1327 DATA_TYPE_NVLIST_ARRAY) == 0);
1328
1329 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1330 for (i = 0; i < sav->sav_count; i++)
1331 l2cache[i] = vdev_config_generate(spa,
1332 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1333 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1334 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1335out:
1336 for (i = 0; i < sav->sav_count; i++)
1337 nvlist_free(l2cache[i]);
1338 if (sav->sav_count)
1339 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1340}
1341
1342static int
1343load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1344{
1345 dmu_buf_t *db;
1346 char *packed = NULL;
1347 size_t nvsize = 0;
1348 int error;
1349 *value = NULL;
1350
1351 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1352 nvsize = *(uint64_t *)db->db_data;
1353 dmu_buf_rele(db, FTAG);
1354
1355 packed = kmem_alloc(nvsize, KM_SLEEP);
1356 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1357 DMU_READ_PREFETCH);
1358 if (error == 0)
1359 error = nvlist_unpack(packed, nvsize, value, 0);
1360 kmem_free(packed, nvsize);
1361
1362 return (error);
1363}
1364
1365/*
1366 * Checks to see if the given vdev could not be opened, in which case we post a
1367 * sysevent to notify the autoreplace code that the device has been removed.
1368 */
1369static void
1370spa_check_removed(vdev_t *vd)
1371{
1372 for (int c = 0; c < vd->vdev_children; c++)
1373 spa_check_removed(vd->vdev_child[c]);
1374
1375 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1376 zfs_post_autoreplace(vd->vdev_spa, vd);
1377 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1378 }
1379}
1380
1381/*
1382 * Validate the current config against the MOS config
1383 */
1384static boolean_t
1385spa_config_valid(spa_t *spa, nvlist_t *config)
1386{
1387 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1388 nvlist_t *nv;
1389
1390 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1391
1392 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1393 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1394
1395 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1396
1397 /*
1398 * If we're doing a normal import, then build up any additional
1399 * diagnostic information about missing devices in this config.
1400 * We'll pass this up to the user for further processing.
1401 */
1402 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1403 nvlist_t **child, *nv;
1404 uint64_t idx = 0;
1405
1406 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1407 KM_SLEEP);
1408 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1409
1410 for (int c = 0; c < rvd->vdev_children; c++) {
1411 vdev_t *tvd = rvd->vdev_child[c];
1412 vdev_t *mtvd = mrvd->vdev_child[c];
1413
1414 if (tvd->vdev_ops == &vdev_missing_ops &&
1415 mtvd->vdev_ops != &vdev_missing_ops &&
1416 mtvd->vdev_islog)
1417 child[idx++] = vdev_config_generate(spa, mtvd,
1418 B_FALSE, 0);
1419 }
1420
1421 if (idx) {
1422 VERIFY(nvlist_add_nvlist_array(nv,
1423 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1424 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1425 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1426
1427 for (int i = 0; i < idx; i++)
1428 nvlist_free(child[i]);
1429 }
1430 nvlist_free(nv);
1431 kmem_free(child, rvd->vdev_children * sizeof (char **));
1432 }
1433
1434 /*
1435 * Compare the root vdev tree with the information we have
1436 * from the MOS config (mrvd). Check each top-level vdev
1437 * with the corresponding MOS config top-level (mtvd).
1438 */
1439 for (int c = 0; c < rvd->vdev_children; c++) {
1440 vdev_t *tvd = rvd->vdev_child[c];
1441 vdev_t *mtvd = mrvd->vdev_child[c];
1442
1443 /*
1444 * Resolve any "missing" vdevs in the current configuration.
1445 * If we find that the MOS config has more accurate information
1446 * about the top-level vdev then use that vdev instead.
1447 */
1448 if (tvd->vdev_ops == &vdev_missing_ops &&
1449 mtvd->vdev_ops != &vdev_missing_ops) {
1450
1451 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1452 continue;
1453
1454 /*
1455 * Device specific actions.
1456 */
1457 if (mtvd->vdev_islog) {
1458 spa_set_log_state(spa, SPA_LOG_CLEAR);
1459 } else {
1460 /*
1461 * XXX - once we have 'readonly' pool
1462 * support we should be able to handle
1463 * missing data devices by transitioning
1464 * the pool to readonly.
1465 */
1466 continue;
1467 }
1468
1469 /*
1470 * Swap the missing vdev with the data we were
1471 * able to obtain from the MOS config.
1472 */
1473 vdev_remove_child(rvd, tvd);
1474 vdev_remove_child(mrvd, mtvd);
1475
1476 vdev_add_child(rvd, mtvd);
1477 vdev_add_child(mrvd, tvd);
1478
1479 spa_config_exit(spa, SCL_ALL, FTAG);
1480 vdev_load(mtvd);
1481 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1482
1483 vdev_reopen(rvd);
1484 } else if (mtvd->vdev_islog) {
1485 /*
1486 * Load the slog device's state from the MOS config
1487 * since it's possible that the label does not
1488 * contain the most up-to-date information.
1489 */
1490 vdev_load_log_state(tvd, mtvd);
1491 vdev_reopen(tvd);
1492 }
1493 }
1494 vdev_free(mrvd);
1495 spa_config_exit(spa, SCL_ALL, FTAG);
1496
1497 /*
1498 * Ensure we were able to validate the config.
1499 */
1500 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1501}
1502
1503/*
1504 * Check for missing log devices
1505 */
1506static int
1507spa_check_logs(spa_t *spa)
1508{
1509 switch (spa->spa_log_state) {
1510 case SPA_LOG_MISSING:
1511 /* need to recheck in case slog has been restored */
1512 case SPA_LOG_UNKNOWN:
1513 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1514 DS_FIND_CHILDREN)) {
1515 spa_set_log_state(spa, SPA_LOG_MISSING);
1516 return (1);
1517 }
1518 break;
1519 }
1520 return (0);
1521}
1522
1523static boolean_t
1524spa_passivate_log(spa_t *spa)
1525{
1526 vdev_t *rvd = spa->spa_root_vdev;
1527 boolean_t slog_found = B_FALSE;
1528
1529 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1530
1531 if (!spa_has_slogs(spa))
1532 return (B_FALSE);
1533
1534 for (int c = 0; c < rvd->vdev_children; c++) {
1535 vdev_t *tvd = rvd->vdev_child[c];
1536 metaslab_group_t *mg = tvd->vdev_mg;
1537
1538 if (tvd->vdev_islog) {
1539 metaslab_group_passivate(mg);
1540 slog_found = B_TRUE;
1541 }
1542 }
1543
1544 return (slog_found);
1545}
1546
1547static void
1548spa_activate_log(spa_t *spa)
1549{
1550 vdev_t *rvd = spa->spa_root_vdev;
1551
1552 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1553
1554 for (int c = 0; c < rvd->vdev_children; c++) {
1555 vdev_t *tvd = rvd->vdev_child[c];
1556 metaslab_group_t *mg = tvd->vdev_mg;
1557
1558 if (tvd->vdev_islog)
1559 metaslab_group_activate(mg);
1560 }
1561}
1562
1563int
1564spa_offline_log(spa_t *spa)
1565{
1566 int error = 0;
1567
1568 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1569 NULL, DS_FIND_CHILDREN)) == 0) {
1570
1571 /*
1572 * We successfully offlined the log device, sync out the
1573 * current txg so that the "stubby" block can be removed
1574 * by zil_sync().
1575 */
1576 txg_wait_synced(spa->spa_dsl_pool, 0);
1577 }
1578 return (error);
1579}
1580
1581static void
1582spa_aux_check_removed(spa_aux_vdev_t *sav)
1583{
1584 int i;
1585
1586 for (i = 0; i < sav->sav_count; i++)
1587 spa_check_removed(sav->sav_vdevs[i]);
1588}
1589
1590void
1591spa_claim_notify(zio_t *zio)
1592{
1593 spa_t *spa = zio->io_spa;
1594
1595 if (zio->io_error)
1596 return;
1597
1598 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1599 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1600 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1601 mutex_exit(&spa->spa_props_lock);
1602}
1603
1604typedef struct spa_load_error {
1605 uint64_t sle_meta_count;
1606 uint64_t sle_data_count;
1607} spa_load_error_t;
1608
1609static void
1610spa_load_verify_done(zio_t *zio)
1611{
1612 blkptr_t *bp = zio->io_bp;
1613 spa_load_error_t *sle = zio->io_private;
1614 dmu_object_type_t type = BP_GET_TYPE(bp);
1615 int error = zio->io_error;
1616
1617 if (error) {
1618 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1619 type != DMU_OT_INTENT_LOG)
1620 atomic_add_64(&sle->sle_meta_count, 1);
1621 else
1622 atomic_add_64(&sle->sle_data_count, 1);
1623 }
1624 zio_data_buf_free(zio->io_data, zio->io_size);
1625}
1626
1627/*ARGSUSED*/
1628static int
1629spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1630 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1631{
1632 if (bp != NULL) {
1633 zio_t *rio = arg;
1634 size_t size = BP_GET_PSIZE(bp);
1635 void *data = zio_data_buf_alloc(size);
1636
1637 zio_nowait(zio_read(rio, spa, bp, data, size,
1638 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1639 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1640 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1641 }
1642 return (0);
1643}
1644
1645static int
1646spa_load_verify(spa_t *spa)
1647{
1648 zio_t *rio;
1649 spa_load_error_t sle = { 0 };
1650 zpool_rewind_policy_t policy;
1651 boolean_t verify_ok = B_FALSE;
1652 int error;
1653
1654 zpool_get_rewind_policy(spa->spa_config, &policy);
1655
1656 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1657 return (0);
1658
1659 rio = zio_root(spa, NULL, &sle,
1660 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1661
1662 error = traverse_pool(spa, spa->spa_verify_min_txg,
1663 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1664
1665 (void) zio_wait(rio);
1666
1667 spa->spa_load_meta_errors = sle.sle_meta_count;
1668 spa->spa_load_data_errors = sle.sle_data_count;
1669
1670 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1671 sle.sle_data_count <= policy.zrp_maxdata) {
1672 int64_t loss = 0;
1673
1674 verify_ok = B_TRUE;
1675 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1676 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1677
1678 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1679 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1680 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1681 VERIFY(nvlist_add_int64(spa->spa_load_info,
1682 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1683 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1684 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1685 } else {
1686 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1687 }
1688
1689 if (error) {
1690 if (error != ENXIO && error != EIO)
1691 error = EIO;
1692 return (error);
1693 }
1694
1695 return (verify_ok ? 0 : EIO);
1696}
1697
1698/*
1699 * Find a value in the pool props object.
1700 */
1701static void
1702spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1703{
1704 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1705 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1706}
1707
1708/*
1709 * Find a value in the pool directory object.
1710 */
1711static int
1712spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1713{
1714 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1715 name, sizeof (uint64_t), 1, val));
1716}
1717
1718static int
1719spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1720{
1721 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1722 return (err);
1723}
1724
1725/*
1726 * Fix up config after a partly-completed split. This is done with the
1727 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1728 * pool have that entry in their config, but only the splitting one contains
1729 * a list of all the guids of the vdevs that are being split off.
1730 *
1731 * This function determines what to do with that list: either rejoin
1732 * all the disks to the pool, or complete the splitting process. To attempt
1733 * the rejoin, each disk that is offlined is marked online again, and
1734 * we do a reopen() call. If the vdev label for every disk that was
1735 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1736 * then we call vdev_split() on each disk, and complete the split.
1737 *
1738 * Otherwise we leave the config alone, with all the vdevs in place in
1739 * the original pool.
1740 */
1741static void
1742spa_try_repair(spa_t *spa, nvlist_t *config)
1743{
1744 uint_t extracted;
1745 uint64_t *glist;
1746 uint_t i, gcount;
1747 nvlist_t *nvl;
1748 vdev_t **vd;
1749 boolean_t attempt_reopen;
1750
1751 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1752 return;
1753
1754 /* check that the config is complete */
1755 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1756 &glist, &gcount) != 0)
1757 return;
1758
1759 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1760
1761 /* attempt to online all the vdevs & validate */
1762 attempt_reopen = B_TRUE;
1763 for (i = 0; i < gcount; i++) {
1764 if (glist[i] == 0) /* vdev is hole */
1765 continue;
1766
1767 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1768 if (vd[i] == NULL) {
1769 /*
1770 * Don't bother attempting to reopen the disks;
1771 * just do the split.
1772 */
1773 attempt_reopen = B_FALSE;
1774 } else {
1775 /* attempt to re-online it */
1776 vd[i]->vdev_offline = B_FALSE;
1777 }
1778 }
1779
1780 if (attempt_reopen) {
1781 vdev_reopen(spa->spa_root_vdev);
1782
1783 /* check each device to see what state it's in */
1784 for (extracted = 0, i = 0; i < gcount; i++) {
1785 if (vd[i] != NULL &&
1786 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1787 break;
1788 ++extracted;
1789 }
1790 }
1791
1792 /*
1793 * If every disk has been moved to the new pool, or if we never
1794 * even attempted to look at them, then we split them off for
1795 * good.
1796 */
1797 if (!attempt_reopen || gcount == extracted) {
1798 for (i = 0; i < gcount; i++)
1799 if (vd[i] != NULL)
1800 vdev_split(vd[i]);
1801 vdev_reopen(spa->spa_root_vdev);
1802 }
1803
1804 kmem_free(vd, gcount * sizeof (vdev_t *));
1805}
1806
1807static int
1808spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1809 boolean_t mosconfig)
1810{
1811 nvlist_t *config = spa->spa_config;
1812 char *ereport = FM_EREPORT_ZFS_POOL;
1813 char *comment;
1814 int error;
1815 uint64_t pool_guid;
1816 nvlist_t *nvl;
1817
1818 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1819 return (EINVAL);
1820
1821 ASSERT(spa->spa_comment == NULL);
1822 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1823 spa->spa_comment = spa_strdup(comment);
1824
1825 /*
1826 * Versioning wasn't explicitly added to the label until later, so if
1827 * it's not present treat it as the initial version.
1828 */
1829 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1830 &spa->spa_ubsync.ub_version) != 0)
1831 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1832
1833 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1834 &spa->spa_config_txg);
1835
1836 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1837 spa_guid_exists(pool_guid, 0)) {
1838 error = EEXIST;
1839 } else {
1840 spa->spa_config_guid = pool_guid;
1841
1842 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1843 &nvl) == 0) {
1844 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1845 KM_SLEEP) == 0);
1846 }
1847
1848 gethrestime(&spa->spa_loaded_ts);
1849 error = spa_load_impl(spa, pool_guid, config, state, type,
1850 mosconfig, &ereport);
1851 }
1852
1853 spa->spa_minref = refcount_count(&spa->spa_refcount);
1854 if (error) {
1855 if (error != EEXIST) {
1856 spa->spa_loaded_ts.tv_sec = 0;
1857 spa->spa_loaded_ts.tv_nsec = 0;
1858 }
1859 if (error != EBADF) {
1860 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1861 }
1862 }
1863 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1864 spa->spa_ena = 0;
1865
1866 return (error);
1867}
1868
1869/*
1870 * Load an existing storage pool, using the pool's builtin spa_config as a
1871 * source of configuration information.
1872 */
1873static int
1874spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1875 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1876 char **ereport)
1877{
1878 int error = 0;
1879 nvlist_t *nvroot = NULL;
1880 vdev_t *rvd;
1881 uberblock_t *ub = &spa->spa_uberblock;
1882 uint64_t children, config_cache_txg = spa->spa_config_txg;
1883 int orig_mode = spa->spa_mode;
1884 int parse;
1885 uint64_t obj;
1886
1887 /*
1888 * If this is an untrusted config, access the pool in read-only mode.
1889 * This prevents things like resilvering recently removed devices.
1890 */
1891 if (!mosconfig)
1892 spa->spa_mode = FREAD;
1893
1894 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1895
1896 spa->spa_load_state = state;
1897
1898 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1899 return (EINVAL);
1900
1901 parse = (type == SPA_IMPORT_EXISTING ?
1902 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1903
1904 /*
1905 * Create "The Godfather" zio to hold all async IOs
1906 */
1907 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1908 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1909
1910 /*
1911 * Parse the configuration into a vdev tree. We explicitly set the
1912 * value that will be returned by spa_version() since parsing the
1913 * configuration requires knowing the version number.
1914 */
1915 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1916 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1917 spa_config_exit(spa, SCL_ALL, FTAG);
1918
1919 if (error != 0)
1920 return (error);
1921
1922 ASSERT(spa->spa_root_vdev == rvd);
1923
1924 if (type != SPA_IMPORT_ASSEMBLE) {
1925 ASSERT(spa_guid(spa) == pool_guid);
1926 }
1927
1928 /*
1929 * Try to open all vdevs, loading each label in the process.
1930 */
1931 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1932 error = vdev_open(rvd);
1933 spa_config_exit(spa, SCL_ALL, FTAG);
1934 if (error != 0)
1935 return (error);
1936
1937 /*
1938 * We need to validate the vdev labels against the configuration that
1939 * we have in hand, which is dependent on the setting of mosconfig. If
1940 * mosconfig is true then we're validating the vdev labels based on
1941 * that config. Otherwise, we're validating against the cached config
1942 * (zpool.cache) that was read when we loaded the zfs module, and then
1943 * later we will recursively call spa_load() and validate against
1944 * the vdev config.
1945 *
1946 * If we're assembling a new pool that's been split off from an
1947 * existing pool, the labels haven't yet been updated so we skip
1948 * validation for now.
1949 */
1950 if (type != SPA_IMPORT_ASSEMBLE) {
1951 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
| 1314 }
1315 }
1316
1317 if (oldvdevs)
1318 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1319
1320 if (sav->sav_config == NULL)
1321 goto out;
1322
1323 sav->sav_vdevs = newvdevs;
1324 sav->sav_count = (int)nl2cache;
1325
1326 /*
1327 * Recompute the stashed list of l2cache devices, with status
1328 * information this time.
1329 */
1330 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1331 DATA_TYPE_NVLIST_ARRAY) == 0);
1332
1333 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1334 for (i = 0; i < sav->sav_count; i++)
1335 l2cache[i] = vdev_config_generate(spa,
1336 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1337 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1338 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1339out:
1340 for (i = 0; i < sav->sav_count; i++)
1341 nvlist_free(l2cache[i]);
1342 if (sav->sav_count)
1343 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1344}
1345
1346static int
1347load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1348{
1349 dmu_buf_t *db;
1350 char *packed = NULL;
1351 size_t nvsize = 0;
1352 int error;
1353 *value = NULL;
1354
1355 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1356 nvsize = *(uint64_t *)db->db_data;
1357 dmu_buf_rele(db, FTAG);
1358
1359 packed = kmem_alloc(nvsize, KM_SLEEP);
1360 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1361 DMU_READ_PREFETCH);
1362 if (error == 0)
1363 error = nvlist_unpack(packed, nvsize, value, 0);
1364 kmem_free(packed, nvsize);
1365
1366 return (error);
1367}
1368
1369/*
1370 * Checks to see if the given vdev could not be opened, in which case we post a
1371 * sysevent to notify the autoreplace code that the device has been removed.
1372 */
1373static void
1374spa_check_removed(vdev_t *vd)
1375{
1376 for (int c = 0; c < vd->vdev_children; c++)
1377 spa_check_removed(vd->vdev_child[c]);
1378
1379 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1380 zfs_post_autoreplace(vd->vdev_spa, vd);
1381 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1382 }
1383}
1384
1385/*
1386 * Validate the current config against the MOS config
1387 */
1388static boolean_t
1389spa_config_valid(spa_t *spa, nvlist_t *config)
1390{
1391 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1392 nvlist_t *nv;
1393
1394 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1395
1396 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1397 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1398
1399 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1400
1401 /*
1402 * If we're doing a normal import, then build up any additional
1403 * diagnostic information about missing devices in this config.
1404 * We'll pass this up to the user for further processing.
1405 */
1406 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1407 nvlist_t **child, *nv;
1408 uint64_t idx = 0;
1409
1410 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1411 KM_SLEEP);
1412 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1413
1414 for (int c = 0; c < rvd->vdev_children; c++) {
1415 vdev_t *tvd = rvd->vdev_child[c];
1416 vdev_t *mtvd = mrvd->vdev_child[c];
1417
1418 if (tvd->vdev_ops == &vdev_missing_ops &&
1419 mtvd->vdev_ops != &vdev_missing_ops &&
1420 mtvd->vdev_islog)
1421 child[idx++] = vdev_config_generate(spa, mtvd,
1422 B_FALSE, 0);
1423 }
1424
1425 if (idx) {
1426 VERIFY(nvlist_add_nvlist_array(nv,
1427 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1428 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1429 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1430
1431 for (int i = 0; i < idx; i++)
1432 nvlist_free(child[i]);
1433 }
1434 nvlist_free(nv);
1435 kmem_free(child, rvd->vdev_children * sizeof (char **));
1436 }
1437
1438 /*
1439 * Compare the root vdev tree with the information we have
1440 * from the MOS config (mrvd). Check each top-level vdev
1441 * with the corresponding MOS config top-level (mtvd).
1442 */
1443 for (int c = 0; c < rvd->vdev_children; c++) {
1444 vdev_t *tvd = rvd->vdev_child[c];
1445 vdev_t *mtvd = mrvd->vdev_child[c];
1446
1447 /*
1448 * Resolve any "missing" vdevs in the current configuration.
1449 * If we find that the MOS config has more accurate information
1450 * about the top-level vdev then use that vdev instead.
1451 */
1452 if (tvd->vdev_ops == &vdev_missing_ops &&
1453 mtvd->vdev_ops != &vdev_missing_ops) {
1454
1455 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1456 continue;
1457
1458 /*
1459 * Device specific actions.
1460 */
1461 if (mtvd->vdev_islog) {
1462 spa_set_log_state(spa, SPA_LOG_CLEAR);
1463 } else {
1464 /*
1465 * XXX - once we have 'readonly' pool
1466 * support we should be able to handle
1467 * missing data devices by transitioning
1468 * the pool to readonly.
1469 */
1470 continue;
1471 }
1472
1473 /*
1474 * Swap the missing vdev with the data we were
1475 * able to obtain from the MOS config.
1476 */
1477 vdev_remove_child(rvd, tvd);
1478 vdev_remove_child(mrvd, mtvd);
1479
1480 vdev_add_child(rvd, mtvd);
1481 vdev_add_child(mrvd, tvd);
1482
1483 spa_config_exit(spa, SCL_ALL, FTAG);
1484 vdev_load(mtvd);
1485 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1486
1487 vdev_reopen(rvd);
1488 } else if (mtvd->vdev_islog) {
1489 /*
1490 * Load the slog device's state from the MOS config
1491 * since it's possible that the label does not
1492 * contain the most up-to-date information.
1493 */
1494 vdev_load_log_state(tvd, mtvd);
1495 vdev_reopen(tvd);
1496 }
1497 }
1498 vdev_free(mrvd);
1499 spa_config_exit(spa, SCL_ALL, FTAG);
1500
1501 /*
1502 * Ensure we were able to validate the config.
1503 */
1504 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1505}
1506
1507/*
1508 * Check for missing log devices
1509 */
1510static int
1511spa_check_logs(spa_t *spa)
1512{
1513 switch (spa->spa_log_state) {
1514 case SPA_LOG_MISSING:
1515 /* need to recheck in case slog has been restored */
1516 case SPA_LOG_UNKNOWN:
1517 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1518 DS_FIND_CHILDREN)) {
1519 spa_set_log_state(spa, SPA_LOG_MISSING);
1520 return (1);
1521 }
1522 break;
1523 }
1524 return (0);
1525}
1526
1527static boolean_t
1528spa_passivate_log(spa_t *spa)
1529{
1530 vdev_t *rvd = spa->spa_root_vdev;
1531 boolean_t slog_found = B_FALSE;
1532
1533 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1534
1535 if (!spa_has_slogs(spa))
1536 return (B_FALSE);
1537
1538 for (int c = 0; c < rvd->vdev_children; c++) {
1539 vdev_t *tvd = rvd->vdev_child[c];
1540 metaslab_group_t *mg = tvd->vdev_mg;
1541
1542 if (tvd->vdev_islog) {
1543 metaslab_group_passivate(mg);
1544 slog_found = B_TRUE;
1545 }
1546 }
1547
1548 return (slog_found);
1549}
1550
1551static void
1552spa_activate_log(spa_t *spa)
1553{
1554 vdev_t *rvd = spa->spa_root_vdev;
1555
1556 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1557
1558 for (int c = 0; c < rvd->vdev_children; c++) {
1559 vdev_t *tvd = rvd->vdev_child[c];
1560 metaslab_group_t *mg = tvd->vdev_mg;
1561
1562 if (tvd->vdev_islog)
1563 metaslab_group_activate(mg);
1564 }
1565}
1566
1567int
1568spa_offline_log(spa_t *spa)
1569{
1570 int error = 0;
1571
1572 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1573 NULL, DS_FIND_CHILDREN)) == 0) {
1574
1575 /*
1576 * We successfully offlined the log device, sync out the
1577 * current txg so that the "stubby" block can be removed
1578 * by zil_sync().
1579 */
1580 txg_wait_synced(spa->spa_dsl_pool, 0);
1581 }
1582 return (error);
1583}
1584
1585static void
1586spa_aux_check_removed(spa_aux_vdev_t *sav)
1587{
1588 int i;
1589
1590 for (i = 0; i < sav->sav_count; i++)
1591 spa_check_removed(sav->sav_vdevs[i]);
1592}
1593
1594void
1595spa_claim_notify(zio_t *zio)
1596{
1597 spa_t *spa = zio->io_spa;
1598
1599 if (zio->io_error)
1600 return;
1601
1602 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1603 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1604 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1605 mutex_exit(&spa->spa_props_lock);
1606}
1607
1608typedef struct spa_load_error {
1609 uint64_t sle_meta_count;
1610 uint64_t sle_data_count;
1611} spa_load_error_t;
1612
1613static void
1614spa_load_verify_done(zio_t *zio)
1615{
1616 blkptr_t *bp = zio->io_bp;
1617 spa_load_error_t *sle = zio->io_private;
1618 dmu_object_type_t type = BP_GET_TYPE(bp);
1619 int error = zio->io_error;
1620
1621 if (error) {
1622 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1623 type != DMU_OT_INTENT_LOG)
1624 atomic_add_64(&sle->sle_meta_count, 1);
1625 else
1626 atomic_add_64(&sle->sle_data_count, 1);
1627 }
1628 zio_data_buf_free(zio->io_data, zio->io_size);
1629}
1630
1631/*ARGSUSED*/
1632static int
1633spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1634 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1635{
1636 if (bp != NULL) {
1637 zio_t *rio = arg;
1638 size_t size = BP_GET_PSIZE(bp);
1639 void *data = zio_data_buf_alloc(size);
1640
1641 zio_nowait(zio_read(rio, spa, bp, data, size,
1642 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1643 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1644 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1645 }
1646 return (0);
1647}
1648
1649static int
1650spa_load_verify(spa_t *spa)
1651{
1652 zio_t *rio;
1653 spa_load_error_t sle = { 0 };
1654 zpool_rewind_policy_t policy;
1655 boolean_t verify_ok = B_FALSE;
1656 int error;
1657
1658 zpool_get_rewind_policy(spa->spa_config, &policy);
1659
1660 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1661 return (0);
1662
1663 rio = zio_root(spa, NULL, &sle,
1664 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1665
1666 error = traverse_pool(spa, spa->spa_verify_min_txg,
1667 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1668
1669 (void) zio_wait(rio);
1670
1671 spa->spa_load_meta_errors = sle.sle_meta_count;
1672 spa->spa_load_data_errors = sle.sle_data_count;
1673
1674 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1675 sle.sle_data_count <= policy.zrp_maxdata) {
1676 int64_t loss = 0;
1677
1678 verify_ok = B_TRUE;
1679 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1680 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1681
1682 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1683 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1684 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1685 VERIFY(nvlist_add_int64(spa->spa_load_info,
1686 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1687 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1688 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1689 } else {
1690 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1691 }
1692
1693 if (error) {
1694 if (error != ENXIO && error != EIO)
1695 error = EIO;
1696 return (error);
1697 }
1698
1699 return (verify_ok ? 0 : EIO);
1700}
1701
1702/*
1703 * Find a value in the pool props object.
1704 */
1705static void
1706spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1707{
1708 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1709 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1710}
1711
1712/*
1713 * Find a value in the pool directory object.
1714 */
1715static int
1716spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1717{
1718 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1719 name, sizeof (uint64_t), 1, val));
1720}
1721
1722static int
1723spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1724{
1725 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1726 return (err);
1727}
1728
1729/*
1730 * Fix up config after a partly-completed split. This is done with the
1731 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1732 * pool have that entry in their config, but only the splitting one contains
1733 * a list of all the guids of the vdevs that are being split off.
1734 *
1735 * This function determines what to do with that list: either rejoin
1736 * all the disks to the pool, or complete the splitting process. To attempt
1737 * the rejoin, each disk that is offlined is marked online again, and
1738 * we do a reopen() call. If the vdev label for every disk that was
1739 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1740 * then we call vdev_split() on each disk, and complete the split.
1741 *
1742 * Otherwise we leave the config alone, with all the vdevs in place in
1743 * the original pool.
1744 */
1745static void
1746spa_try_repair(spa_t *spa, nvlist_t *config)
1747{
1748 uint_t extracted;
1749 uint64_t *glist;
1750 uint_t i, gcount;
1751 nvlist_t *nvl;
1752 vdev_t **vd;
1753 boolean_t attempt_reopen;
1754
1755 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1756 return;
1757
1758 /* check that the config is complete */
1759 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1760 &glist, &gcount) != 0)
1761 return;
1762
1763 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1764
1765 /* attempt to online all the vdevs & validate */
1766 attempt_reopen = B_TRUE;
1767 for (i = 0; i < gcount; i++) {
1768 if (glist[i] == 0) /* vdev is hole */
1769 continue;
1770
1771 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1772 if (vd[i] == NULL) {
1773 /*
1774 * Don't bother attempting to reopen the disks;
1775 * just do the split.
1776 */
1777 attempt_reopen = B_FALSE;
1778 } else {
1779 /* attempt to re-online it */
1780 vd[i]->vdev_offline = B_FALSE;
1781 }
1782 }
1783
1784 if (attempt_reopen) {
1785 vdev_reopen(spa->spa_root_vdev);
1786
1787 /* check each device to see what state it's in */
1788 for (extracted = 0, i = 0; i < gcount; i++) {
1789 if (vd[i] != NULL &&
1790 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1791 break;
1792 ++extracted;
1793 }
1794 }
1795
1796 /*
1797 * If every disk has been moved to the new pool, or if we never
1798 * even attempted to look at them, then we split them off for
1799 * good.
1800 */
1801 if (!attempt_reopen || gcount == extracted) {
1802 for (i = 0; i < gcount; i++)
1803 if (vd[i] != NULL)
1804 vdev_split(vd[i]);
1805 vdev_reopen(spa->spa_root_vdev);
1806 }
1807
1808 kmem_free(vd, gcount * sizeof (vdev_t *));
1809}
1810
1811static int
1812spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1813 boolean_t mosconfig)
1814{
1815 nvlist_t *config = spa->spa_config;
1816 char *ereport = FM_EREPORT_ZFS_POOL;
1817 char *comment;
1818 int error;
1819 uint64_t pool_guid;
1820 nvlist_t *nvl;
1821
1822 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1823 return (EINVAL);
1824
1825 ASSERT(spa->spa_comment == NULL);
1826 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1827 spa->spa_comment = spa_strdup(comment);
1828
1829 /*
1830 * Versioning wasn't explicitly added to the label until later, so if
1831 * it's not present treat it as the initial version.
1832 */
1833 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1834 &spa->spa_ubsync.ub_version) != 0)
1835 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1836
1837 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1838 &spa->spa_config_txg);
1839
1840 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1841 spa_guid_exists(pool_guid, 0)) {
1842 error = EEXIST;
1843 } else {
1844 spa->spa_config_guid = pool_guid;
1845
1846 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1847 &nvl) == 0) {
1848 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1849 KM_SLEEP) == 0);
1850 }
1851
1852 gethrestime(&spa->spa_loaded_ts);
1853 error = spa_load_impl(spa, pool_guid, config, state, type,
1854 mosconfig, &ereport);
1855 }
1856
1857 spa->spa_minref = refcount_count(&spa->spa_refcount);
1858 if (error) {
1859 if (error != EEXIST) {
1860 spa->spa_loaded_ts.tv_sec = 0;
1861 spa->spa_loaded_ts.tv_nsec = 0;
1862 }
1863 if (error != EBADF) {
1864 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1865 }
1866 }
1867 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1868 spa->spa_ena = 0;
1869
1870 return (error);
1871}
1872
1873/*
1874 * Load an existing storage pool, using the pool's builtin spa_config as a
1875 * source of configuration information.
1876 */
1877static int
1878spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1879 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1880 char **ereport)
1881{
1882 int error = 0;
1883 nvlist_t *nvroot = NULL;
1884 vdev_t *rvd;
1885 uberblock_t *ub = &spa->spa_uberblock;
1886 uint64_t children, config_cache_txg = spa->spa_config_txg;
1887 int orig_mode = spa->spa_mode;
1888 int parse;
1889 uint64_t obj;
1890
1891 /*
1892 * If this is an untrusted config, access the pool in read-only mode.
1893 * This prevents things like resilvering recently removed devices.
1894 */
1895 if (!mosconfig)
1896 spa->spa_mode = FREAD;
1897
1898 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1899
1900 spa->spa_load_state = state;
1901
1902 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1903 return (EINVAL);
1904
1905 parse = (type == SPA_IMPORT_EXISTING ?
1906 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1907
1908 /*
1909 * Create "The Godfather" zio to hold all async IOs
1910 */
1911 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1912 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1913
1914 /*
1915 * Parse the configuration into a vdev tree. We explicitly set the
1916 * value that will be returned by spa_version() since parsing the
1917 * configuration requires knowing the version number.
1918 */
1919 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1920 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1921 spa_config_exit(spa, SCL_ALL, FTAG);
1922
1923 if (error != 0)
1924 return (error);
1925
1926 ASSERT(spa->spa_root_vdev == rvd);
1927
1928 if (type != SPA_IMPORT_ASSEMBLE) {
1929 ASSERT(spa_guid(spa) == pool_guid);
1930 }
1931
1932 /*
1933 * Try to open all vdevs, loading each label in the process.
1934 */
1935 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1936 error = vdev_open(rvd);
1937 spa_config_exit(spa, SCL_ALL, FTAG);
1938 if (error != 0)
1939 return (error);
1940
1941 /*
1942 * We need to validate the vdev labels against the configuration that
1943 * we have in hand, which is dependent on the setting of mosconfig. If
1944 * mosconfig is true then we're validating the vdev labels based on
1945 * that config. Otherwise, we're validating against the cached config
1946 * (zpool.cache) that was read when we loaded the zfs module, and then
1947 * later we will recursively call spa_load() and validate against
1948 * the vdev config.
1949 *
1950 * If we're assembling a new pool that's been split off from an
1951 * existing pool, the labels haven't yet been updated so we skip
1952 * validation for now.
1953 */
1954 if (type != SPA_IMPORT_ASSEMBLE) {
1955 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
|
1952 error = vdev_validate(rvd);
| 1956 error = vdev_validate(rvd, mosconfig);
|
1953 spa_config_exit(spa, SCL_ALL, FTAG);
1954
1955 if (error != 0)
1956 return (error);
1957
1958 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1959 return (ENXIO);
1960 }
1961
1962 /*
1963 * Find the best uberblock.
1964 */
1965 vdev_uberblock_load(NULL, rvd, ub);
1966
1967 /*
1968 * If we weren't able to find a single valid uberblock, return failure.
1969 */
1970 if (ub->ub_txg == 0)
1971 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1972
1973 /*
1974 * If the pool is newer than the code, we can't open it.
1975 */
1976 if (ub->ub_version > SPA_VERSION)
1977 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1978
1979 /*
1980 * If the vdev guid sum doesn't match the uberblock, we have an
1981 * incomplete configuration. We first check to see if the pool
1982 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1983 * If it is, defer the vdev_guid_sum check till later so we
1984 * can handle missing vdevs.
1985 */
1986 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1987 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1988 rvd->vdev_guid_sum != ub->ub_guid_sum)
1989 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1990
1991 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1992 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1993 spa_try_repair(spa, config);
1994 spa_config_exit(spa, SCL_ALL, FTAG);
1995 nvlist_free(spa->spa_config_splitting);
1996 spa->spa_config_splitting = NULL;
1997 }
1998
1999 /*
2000 * Initialize internal SPA structures.
2001 */
2002 spa->spa_state = POOL_STATE_ACTIVE;
2003 spa->spa_ubsync = spa->spa_uberblock;
2004 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2005 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2006 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2007 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2008 spa->spa_claim_max_txg = spa->spa_first_txg;
2009 spa->spa_prev_software_version = ub->ub_software_version;
2010
2011 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2012 if (error)
2013 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2014 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2015
2016 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2017 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2018
2019 if (!mosconfig) {
2020 uint64_t hostid;
2021 nvlist_t *policy = NULL, *nvconfig;
2022
2023 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2024 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2025
2026 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2027 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2028 char *hostname;
2029 unsigned long myhostid = 0;
2030
2031 VERIFY(nvlist_lookup_string(nvconfig,
2032 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2033
2034#ifdef _KERNEL
2035 myhostid = zone_get_hostid(NULL);
2036#else /* _KERNEL */
2037 /*
2038 * We're emulating the system's hostid in userland, so
2039 * we can't use zone_get_hostid().
2040 */
2041 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2042#endif /* _KERNEL */
2043 if (check_hostid && hostid != 0 && myhostid != 0 &&
2044 hostid != myhostid) {
2045 nvlist_free(nvconfig);
2046 cmn_err(CE_WARN, "pool '%s' could not be "
2047 "loaded as it was last accessed by "
2048 "another system (host: %s hostid: 0x%lx). "
2049 "See: http://www.sun.com/msg/ZFS-8000-EY",
2050 spa_name(spa), hostname,
2051 (unsigned long)hostid);
2052 return (EBADF);
2053 }
2054 }
2055 if (nvlist_lookup_nvlist(spa->spa_config,
2056 ZPOOL_REWIND_POLICY, &policy) == 0)
2057 VERIFY(nvlist_add_nvlist(nvconfig,
2058 ZPOOL_REWIND_POLICY, policy) == 0);
2059
2060 spa_config_set(spa, nvconfig);
2061 spa_unload(spa);
2062 spa_deactivate(spa);
2063 spa_activate(spa, orig_mode);
2064
2065 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2066 }
2067
2068 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2069 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2070 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2071 if (error != 0)
2072 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2073
2074 /*
2075 * Load the bit that tells us to use the new accounting function
2076 * (raid-z deflation). If we have an older pool, this will not
2077 * be present.
2078 */
2079 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2080 if (error != 0 && error != ENOENT)
2081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2082
2083 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2084 &spa->spa_creation_version);
2085 if (error != 0 && error != ENOENT)
2086 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2087
2088 /*
2089 * Load the persistent error log. If we have an older pool, this will
2090 * not be present.
2091 */
2092 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2093 if (error != 0 && error != ENOENT)
2094 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2095
2096 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2097 &spa->spa_errlog_scrub);
2098 if (error != 0 && error != ENOENT)
2099 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2100
2101 /*
2102 * Load the history object. If we have an older pool, this
2103 * will not be present.
2104 */
2105 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2106 if (error != 0 && error != ENOENT)
2107 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2108
2109 /*
2110 * If we're assembling the pool from the split-off vdevs of
2111 * an existing pool, we don't want to attach the spares & cache
2112 * devices.
2113 */
2114
2115 /*
2116 * Load any hot spares for this pool.
2117 */
2118 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2119 if (error != 0 && error != ENOENT)
2120 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2121 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2122 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2123 if (load_nvlist(spa, spa->spa_spares.sav_object,
2124 &spa->spa_spares.sav_config) != 0)
2125 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2126
2127 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2128 spa_load_spares(spa);
2129 spa_config_exit(spa, SCL_ALL, FTAG);
2130 } else if (error == 0) {
2131 spa->spa_spares.sav_sync = B_TRUE;
2132 }
2133
2134 /*
2135 * Load any level 2 ARC devices for this pool.
2136 */
2137 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2138 &spa->spa_l2cache.sav_object);
2139 if (error != 0 && error != ENOENT)
2140 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2141 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2142 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2143 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2144 &spa->spa_l2cache.sav_config) != 0)
2145 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2146
2147 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2148 spa_load_l2cache(spa);
2149 spa_config_exit(spa, SCL_ALL, FTAG);
2150 } else if (error == 0) {
2151 spa->spa_l2cache.sav_sync = B_TRUE;
2152 }
2153
2154 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2155
2156 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2157 if (error && error != ENOENT)
2158 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2159
2160 if (error == 0) {
2161 uint64_t autoreplace;
2162
2163 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2164 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2165 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2166 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2167 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2168 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2169 &spa->spa_dedup_ditto);
2170
2171 spa->spa_autoreplace = (autoreplace != 0);
2172 }
2173
2174 /*
2175 * If the 'autoreplace' property is set, then post a resource notifying
2176 * the ZFS DE that it should not issue any faults for unopenable
2177 * devices. We also iterate over the vdevs, and post a sysevent for any
2178 * unopenable vdevs so that the normal autoreplace handler can take
2179 * over.
2180 */
2181 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2182 spa_check_removed(spa->spa_root_vdev);
2183 /*
2184 * For the import case, this is done in spa_import(), because
2185 * at this point we're using the spare definitions from
2186 * the MOS config, not necessarily from the userland config.
2187 */
2188 if (state != SPA_LOAD_IMPORT) {
2189 spa_aux_check_removed(&spa->spa_spares);
2190 spa_aux_check_removed(&spa->spa_l2cache);
2191 }
2192 }
2193
2194 /*
2195 * Load the vdev state for all toplevel vdevs.
2196 */
2197 vdev_load(rvd);
2198
2199 /*
2200 * Propagate the leaf DTLs we just loaded all the way up the tree.
2201 */
2202 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2203 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2204 spa_config_exit(spa, SCL_ALL, FTAG);
2205
2206 /*
2207 * Load the DDTs (dedup tables).
2208 */
2209 error = ddt_load(spa);
2210 if (error != 0)
2211 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2212
2213 spa_update_dspace(spa);
2214
2215 /*
2216 * Validate the config, using the MOS config to fill in any
2217 * information which might be missing. If we fail to validate
2218 * the config then declare the pool unfit for use. If we're
2219 * assembling a pool from a split, the log is not transferred
2220 * over.
2221 */
2222 if (type != SPA_IMPORT_ASSEMBLE) {
2223 nvlist_t *nvconfig;
2224
2225 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2226 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2227
2228 if (!spa_config_valid(spa, nvconfig)) {
2229 nvlist_free(nvconfig);
2230 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2231 ENXIO));
2232 }
2233 nvlist_free(nvconfig);
2234
2235 /*
2236 * Now that we've validate the config, check the state of the
2237 * root vdev. If it can't be opened, it indicates one or
2238 * more toplevel vdevs are faulted.
2239 */
2240 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2241 return (ENXIO);
2242
2243 if (spa_check_logs(spa)) {
2244 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2245 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2246 }
2247 }
2248
2249 /*
2250 * We've successfully opened the pool, verify that we're ready
2251 * to start pushing transactions.
2252 */
2253 if (state != SPA_LOAD_TRYIMPORT) {
2254 if (error = spa_load_verify(spa))
2255 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2256 error));
2257 }
2258
2259 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2260 spa->spa_load_max_txg == UINT64_MAX)) {
2261 dmu_tx_t *tx;
2262 int need_update = B_FALSE;
2263
2264 ASSERT(state != SPA_LOAD_TRYIMPORT);
2265
2266 /*
2267 * Claim log blocks that haven't been committed yet.
2268 * This must all happen in a single txg.
2269 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2270 * invoked from zil_claim_log_block()'s i/o done callback.
2271 * Price of rollback is that we abandon the log.
2272 */
2273 spa->spa_claiming = B_TRUE;
2274
2275 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2276 spa_first_txg(spa));
2277 (void) dmu_objset_find(spa_name(spa),
2278 zil_claim, tx, DS_FIND_CHILDREN);
2279 dmu_tx_commit(tx);
2280
2281 spa->spa_claiming = B_FALSE;
2282
2283 spa_set_log_state(spa, SPA_LOG_GOOD);
2284 spa->spa_sync_on = B_TRUE;
2285 txg_sync_start(spa->spa_dsl_pool);
2286
2287 /*
2288 * Wait for all claims to sync. We sync up to the highest
2289 * claimed log block birth time so that claimed log blocks
2290 * don't appear to be from the future. spa_claim_max_txg
2291 * will have been set for us by either zil_check_log_chain()
2292 * (invoked from spa_check_logs()) or zil_claim() above.
2293 */
2294 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2295
2296 /*
2297 * If the config cache is stale, or we have uninitialized
2298 * metaslabs (see spa_vdev_add()), then update the config.
2299 *
2300 * If this is a verbatim import, trust the current
2301 * in-core spa_config and update the disk labels.
2302 */
2303 if (config_cache_txg != spa->spa_config_txg ||
2304 state == SPA_LOAD_IMPORT ||
2305 state == SPA_LOAD_RECOVER ||
2306 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2307 need_update = B_TRUE;
2308
2309 for (int c = 0; c < rvd->vdev_children; c++)
2310 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2311 need_update = B_TRUE;
2312
2313 /*
2314 * Update the config cache asychronously in case we're the
2315 * root pool, in which case the config cache isn't writable yet.
2316 */
2317 if (need_update)
2318 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2319
2320 /*
2321 * Check all DTLs to see if anything needs resilvering.
2322 */
2323 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2324 vdev_resilver_needed(rvd, NULL, NULL))
2325 spa_async_request(spa, SPA_ASYNC_RESILVER);
2326
2327 /*
2328 * Delete any inconsistent datasets.
2329 */
2330 (void) dmu_objset_find(spa_name(spa),
2331 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2332
2333 /*
2334 * Clean up any stale temporary dataset userrefs.
2335 */
2336 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2337 }
2338
2339 return (0);
2340}
2341
2342static int
2343spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2344{
2345 int mode = spa->spa_mode;
2346
2347 spa_unload(spa);
2348 spa_deactivate(spa);
2349
2350 spa->spa_load_max_txg--;
2351
2352 spa_activate(spa, mode);
2353 spa_async_suspend(spa);
2354
2355 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2356}
2357
2358static int
2359spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2360 uint64_t max_request, int rewind_flags)
2361{
2362 nvlist_t *config = NULL;
2363 int load_error, rewind_error;
2364 uint64_t safe_rewind_txg;
2365 uint64_t min_txg;
2366
2367 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2368 spa->spa_load_max_txg = spa->spa_load_txg;
2369 spa_set_log_state(spa, SPA_LOG_CLEAR);
2370 } else {
2371 spa->spa_load_max_txg = max_request;
2372 }
2373
2374 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2375 mosconfig);
2376 if (load_error == 0)
2377 return (0);
2378
2379 if (spa->spa_root_vdev != NULL)
2380 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2381
2382 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2383 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2384
2385 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2386 nvlist_free(config);
2387 return (load_error);
2388 }
2389
2390 /* Price of rolling back is discarding txgs, including log */
2391 if (state == SPA_LOAD_RECOVER)
2392 spa_set_log_state(spa, SPA_LOG_CLEAR);
2393
2394 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2395 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2396 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2397 TXG_INITIAL : safe_rewind_txg;
2398
2399 /*
2400 * Continue as long as we're finding errors, we're still within
2401 * the acceptable rewind range, and we're still finding uberblocks
2402 */
2403 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2404 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2405 if (spa->spa_load_max_txg < safe_rewind_txg)
2406 spa->spa_extreme_rewind = B_TRUE;
2407 rewind_error = spa_load_retry(spa, state, mosconfig);
2408 }
2409
2410 spa->spa_extreme_rewind = B_FALSE;
2411 spa->spa_load_max_txg = UINT64_MAX;
2412
2413 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2414 spa_config_set(spa, config);
2415
2416 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2417}
2418
2419/*
2420 * Pool Open/Import
2421 *
2422 * The import case is identical to an open except that the configuration is sent
2423 * down from userland, instead of grabbed from the configuration cache. For the
2424 * case of an open, the pool configuration will exist in the
2425 * POOL_STATE_UNINITIALIZED state.
2426 *
2427 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2428 * the same time open the pool, without having to keep around the spa_t in some
2429 * ambiguous state.
2430 */
2431static int
2432spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2433 nvlist_t **config)
2434{
2435 spa_t *spa;
2436 spa_load_state_t state = SPA_LOAD_OPEN;
2437 int error;
2438 int locked = B_FALSE;
2439 int firstopen = B_FALSE;
2440
2441 *spapp = NULL;
2442
2443 /*
2444 * As disgusting as this is, we need to support recursive calls to this
2445 * function because dsl_dir_open() is called during spa_load(), and ends
2446 * up calling spa_open() again. The real fix is to figure out how to
2447 * avoid dsl_dir_open() calling this in the first place.
2448 */
2449 if (mutex_owner(&spa_namespace_lock) != curthread) {
2450 mutex_enter(&spa_namespace_lock);
2451 locked = B_TRUE;
2452 }
2453
2454 if ((spa = spa_lookup(pool)) == NULL) {
2455 if (locked)
2456 mutex_exit(&spa_namespace_lock);
2457 return (ENOENT);
2458 }
2459
2460 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2461 zpool_rewind_policy_t policy;
2462
2463 firstopen = B_TRUE;
2464
2465 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2466 &policy);
2467 if (policy.zrp_request & ZPOOL_DO_REWIND)
2468 state = SPA_LOAD_RECOVER;
2469
2470 spa_activate(spa, spa_mode_global);
2471
2472 if (state != SPA_LOAD_RECOVER)
2473 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2474
2475 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2476 policy.zrp_request);
2477
2478 if (error == EBADF) {
2479 /*
2480 * If vdev_validate() returns failure (indicated by
2481 * EBADF), it indicates that one of the vdevs indicates
2482 * that the pool has been exported or destroyed. If
2483 * this is the case, the config cache is out of sync and
2484 * we should remove the pool from the namespace.
2485 */
2486 spa_unload(spa);
2487 spa_deactivate(spa);
2488 spa_config_sync(spa, B_TRUE, B_TRUE);
2489 spa_remove(spa);
2490 if (locked)
2491 mutex_exit(&spa_namespace_lock);
2492 return (ENOENT);
2493 }
2494
2495 if (error) {
2496 /*
2497 * We can't open the pool, but we still have useful
2498 * information: the state of each vdev after the
2499 * attempted vdev_open(). Return this to the user.
2500 */
2501 if (config != NULL && spa->spa_config) {
2502 VERIFY(nvlist_dup(spa->spa_config, config,
2503 KM_SLEEP) == 0);
2504 VERIFY(nvlist_add_nvlist(*config,
2505 ZPOOL_CONFIG_LOAD_INFO,
2506 spa->spa_load_info) == 0);
2507 }
2508 spa_unload(spa);
2509 spa_deactivate(spa);
2510 spa->spa_last_open_failed = error;
2511 if (locked)
2512 mutex_exit(&spa_namespace_lock);
2513 *spapp = NULL;
2514 return (error);
2515 }
2516 }
2517
2518 spa_open_ref(spa, tag);
2519
2520 if (config != NULL)
2521 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2522
2523 /*
2524 * If we've recovered the pool, pass back any information we
2525 * gathered while doing the load.
2526 */
2527 if (state == SPA_LOAD_RECOVER) {
2528 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2529 spa->spa_load_info) == 0);
2530 }
2531
2532 if (locked) {
2533 spa->spa_last_open_failed = 0;
2534 spa->spa_last_ubsync_txg = 0;
2535 spa->spa_load_txg = 0;
2536 mutex_exit(&spa_namespace_lock);
2537#ifdef __FreeBSD__
2538#ifdef _KERNEL
2539 if (firstopen)
2540 zvol_create_minors(pool);
2541#endif
2542#endif
2543 }
2544
2545 *spapp = spa;
2546
2547 return (0);
2548}
2549
2550int
2551spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2552 nvlist_t **config)
2553{
2554 return (spa_open_common(name, spapp, tag, policy, config));
2555}
2556
2557int
2558spa_open(const char *name, spa_t **spapp, void *tag)
2559{
2560 return (spa_open_common(name, spapp, tag, NULL, NULL));
2561}
2562
2563/*
2564 * Lookup the given spa_t, incrementing the inject count in the process,
2565 * preventing it from being exported or destroyed.
2566 */
2567spa_t *
2568spa_inject_addref(char *name)
2569{
2570 spa_t *spa;
2571
2572 mutex_enter(&spa_namespace_lock);
2573 if ((spa = spa_lookup(name)) == NULL) {
2574 mutex_exit(&spa_namespace_lock);
2575 return (NULL);
2576 }
2577 spa->spa_inject_ref++;
2578 mutex_exit(&spa_namespace_lock);
2579
2580 return (spa);
2581}
2582
2583void
2584spa_inject_delref(spa_t *spa)
2585{
2586 mutex_enter(&spa_namespace_lock);
2587 spa->spa_inject_ref--;
2588 mutex_exit(&spa_namespace_lock);
2589}
2590
2591/*
2592 * Add spares device information to the nvlist.
2593 */
2594static void
2595spa_add_spares(spa_t *spa, nvlist_t *config)
2596{
2597 nvlist_t **spares;
2598 uint_t i, nspares;
2599 nvlist_t *nvroot;
2600 uint64_t guid;
2601 vdev_stat_t *vs;
2602 uint_t vsc;
2603 uint64_t pool;
2604
2605 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2606
2607 if (spa->spa_spares.sav_count == 0)
2608 return;
2609
2610 VERIFY(nvlist_lookup_nvlist(config,
2611 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2612 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2613 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2614 if (nspares != 0) {
2615 VERIFY(nvlist_add_nvlist_array(nvroot,
2616 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2617 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2618 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2619
2620 /*
2621 * Go through and find any spares which have since been
2622 * repurposed as an active spare. If this is the case, update
2623 * their status appropriately.
2624 */
2625 for (i = 0; i < nspares; i++) {
2626 VERIFY(nvlist_lookup_uint64(spares[i],
2627 ZPOOL_CONFIG_GUID, &guid) == 0);
2628 if (spa_spare_exists(guid, &pool, NULL) &&
2629 pool != 0ULL) {
2630 VERIFY(nvlist_lookup_uint64_array(
2631 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2632 (uint64_t **)&vs, &vsc) == 0);
2633 vs->vs_state = VDEV_STATE_CANT_OPEN;
2634 vs->vs_aux = VDEV_AUX_SPARED;
2635 }
2636 }
2637 }
2638}
2639
2640/*
2641 * Add l2cache device information to the nvlist, including vdev stats.
2642 */
2643static void
2644spa_add_l2cache(spa_t *spa, nvlist_t *config)
2645{
2646 nvlist_t **l2cache;
2647 uint_t i, j, nl2cache;
2648 nvlist_t *nvroot;
2649 uint64_t guid;
2650 vdev_t *vd;
2651 vdev_stat_t *vs;
2652 uint_t vsc;
2653
2654 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2655
2656 if (spa->spa_l2cache.sav_count == 0)
2657 return;
2658
2659 VERIFY(nvlist_lookup_nvlist(config,
2660 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2661 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2662 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2663 if (nl2cache != 0) {
2664 VERIFY(nvlist_add_nvlist_array(nvroot,
2665 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2666 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2667 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2668
2669 /*
2670 * Update level 2 cache device stats.
2671 */
2672
2673 for (i = 0; i < nl2cache; i++) {
2674 VERIFY(nvlist_lookup_uint64(l2cache[i],
2675 ZPOOL_CONFIG_GUID, &guid) == 0);
2676
2677 vd = NULL;
2678 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2679 if (guid ==
2680 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2681 vd = spa->spa_l2cache.sav_vdevs[j];
2682 break;
2683 }
2684 }
2685 ASSERT(vd != NULL);
2686
2687 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2688 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2689 == 0);
2690 vdev_get_stats(vd, vs);
2691 }
2692 }
2693}
2694
2695int
2696spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2697{
2698 int error;
2699 spa_t *spa;
2700
2701 *config = NULL;
2702 error = spa_open_common(name, &spa, FTAG, NULL, config);
2703
2704 if (spa != NULL) {
2705 /*
2706 * This still leaves a window of inconsistency where the spares
2707 * or l2cache devices could change and the config would be
2708 * self-inconsistent.
2709 */
2710 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2711
2712 if (*config != NULL) {
2713 uint64_t loadtimes[2];
2714
2715 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2716 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2717 VERIFY(nvlist_add_uint64_array(*config,
2718 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2719
2720 VERIFY(nvlist_add_uint64(*config,
2721 ZPOOL_CONFIG_ERRCOUNT,
2722 spa_get_errlog_size(spa)) == 0);
2723
2724 if (spa_suspended(spa))
2725 VERIFY(nvlist_add_uint64(*config,
2726 ZPOOL_CONFIG_SUSPENDED,
2727 spa->spa_failmode) == 0);
2728
2729 spa_add_spares(spa, *config);
2730 spa_add_l2cache(spa, *config);
2731 }
2732 }
2733
2734 /*
2735 * We want to get the alternate root even for faulted pools, so we cheat
2736 * and call spa_lookup() directly.
2737 */
2738 if (altroot) {
2739 if (spa == NULL) {
2740 mutex_enter(&spa_namespace_lock);
2741 spa = spa_lookup(name);
2742 if (spa)
2743 spa_altroot(spa, altroot, buflen);
2744 else
2745 altroot[0] = '\0';
2746 spa = NULL;
2747 mutex_exit(&spa_namespace_lock);
2748 } else {
2749 spa_altroot(spa, altroot, buflen);
2750 }
2751 }
2752
2753 if (spa != NULL) {
2754 spa_config_exit(spa, SCL_CONFIG, FTAG);
2755 spa_close(spa, FTAG);
2756 }
2757
2758 return (error);
2759}
2760
2761/*
2762 * Validate that the auxiliary device array is well formed. We must have an
2763 * array of nvlists, each which describes a valid leaf vdev. If this is an
2764 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2765 * specified, as long as they are well-formed.
2766 */
2767static int
2768spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2769 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2770 vdev_labeltype_t label)
2771{
2772 nvlist_t **dev;
2773 uint_t i, ndev;
2774 vdev_t *vd;
2775 int error;
2776
2777 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2778
2779 /*
2780 * It's acceptable to have no devs specified.
2781 */
2782 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2783 return (0);
2784
2785 if (ndev == 0)
2786 return (EINVAL);
2787
2788 /*
2789 * Make sure the pool is formatted with a version that supports this
2790 * device type.
2791 */
2792 if (spa_version(spa) < version)
2793 return (ENOTSUP);
2794
2795 /*
2796 * Set the pending device list so we correctly handle device in-use
2797 * checking.
2798 */
2799 sav->sav_pending = dev;
2800 sav->sav_npending = ndev;
2801
2802 for (i = 0; i < ndev; i++) {
2803 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2804 mode)) != 0)
2805 goto out;
2806
2807 if (!vd->vdev_ops->vdev_op_leaf) {
2808 vdev_free(vd);
2809 error = EINVAL;
2810 goto out;
2811 }
2812
2813 /*
2814 * The L2ARC currently only supports disk devices in
2815 * kernel context. For user-level testing, we allow it.
2816 */
2817#ifdef _KERNEL
2818 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2819 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2820 error = ENOTBLK;
| 1957 spa_config_exit(spa, SCL_ALL, FTAG);
1958
1959 if (error != 0)
1960 return (error);
1961
1962 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1963 return (ENXIO);
1964 }
1965
1966 /*
1967 * Find the best uberblock.
1968 */
1969 vdev_uberblock_load(NULL, rvd, ub);
1970
1971 /*
1972 * If we weren't able to find a single valid uberblock, return failure.
1973 */
1974 if (ub->ub_txg == 0)
1975 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1976
1977 /*
1978 * If the pool is newer than the code, we can't open it.
1979 */
1980 if (ub->ub_version > SPA_VERSION)
1981 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1982
1983 /*
1984 * If the vdev guid sum doesn't match the uberblock, we have an
1985 * incomplete configuration. We first check to see if the pool
1986 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1987 * If it is, defer the vdev_guid_sum check till later so we
1988 * can handle missing vdevs.
1989 */
1990 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1991 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1992 rvd->vdev_guid_sum != ub->ub_guid_sum)
1993 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1994
1995 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1996 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1997 spa_try_repair(spa, config);
1998 spa_config_exit(spa, SCL_ALL, FTAG);
1999 nvlist_free(spa->spa_config_splitting);
2000 spa->spa_config_splitting = NULL;
2001 }
2002
2003 /*
2004 * Initialize internal SPA structures.
2005 */
2006 spa->spa_state = POOL_STATE_ACTIVE;
2007 spa->spa_ubsync = spa->spa_uberblock;
2008 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2009 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2010 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2011 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2012 spa->spa_claim_max_txg = spa->spa_first_txg;
2013 spa->spa_prev_software_version = ub->ub_software_version;
2014
2015 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2016 if (error)
2017 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2018 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2019
2020 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2021 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2022
2023 if (!mosconfig) {
2024 uint64_t hostid;
2025 nvlist_t *policy = NULL, *nvconfig;
2026
2027 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2028 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2029
2030 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2031 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2032 char *hostname;
2033 unsigned long myhostid = 0;
2034
2035 VERIFY(nvlist_lookup_string(nvconfig,
2036 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2037
2038#ifdef _KERNEL
2039 myhostid = zone_get_hostid(NULL);
2040#else /* _KERNEL */
2041 /*
2042 * We're emulating the system's hostid in userland, so
2043 * we can't use zone_get_hostid().
2044 */
2045 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2046#endif /* _KERNEL */
2047 if (check_hostid && hostid != 0 && myhostid != 0 &&
2048 hostid != myhostid) {
2049 nvlist_free(nvconfig);
2050 cmn_err(CE_WARN, "pool '%s' could not be "
2051 "loaded as it was last accessed by "
2052 "another system (host: %s hostid: 0x%lx). "
2053 "See: http://www.sun.com/msg/ZFS-8000-EY",
2054 spa_name(spa), hostname,
2055 (unsigned long)hostid);
2056 return (EBADF);
2057 }
2058 }
2059 if (nvlist_lookup_nvlist(spa->spa_config,
2060 ZPOOL_REWIND_POLICY, &policy) == 0)
2061 VERIFY(nvlist_add_nvlist(nvconfig,
2062 ZPOOL_REWIND_POLICY, policy) == 0);
2063
2064 spa_config_set(spa, nvconfig);
2065 spa_unload(spa);
2066 spa_deactivate(spa);
2067 spa_activate(spa, orig_mode);
2068
2069 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2070 }
2071
2072 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2073 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2075 if (error != 0)
2076 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2077
2078 /*
2079 * Load the bit that tells us to use the new accounting function
2080 * (raid-z deflation). If we have an older pool, this will not
2081 * be present.
2082 */
2083 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2084 if (error != 0 && error != ENOENT)
2085 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2086
2087 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2088 &spa->spa_creation_version);
2089 if (error != 0 && error != ENOENT)
2090 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2091
2092 /*
2093 * Load the persistent error log. If we have an older pool, this will
2094 * not be present.
2095 */
2096 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2097 if (error != 0 && error != ENOENT)
2098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2099
2100 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2101 &spa->spa_errlog_scrub);
2102 if (error != 0 && error != ENOENT)
2103 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2104
2105 /*
2106 * Load the history object. If we have an older pool, this
2107 * will not be present.
2108 */
2109 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2110 if (error != 0 && error != ENOENT)
2111 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2112
2113 /*
2114 * If we're assembling the pool from the split-off vdevs of
2115 * an existing pool, we don't want to attach the spares & cache
2116 * devices.
2117 */
2118
2119 /*
2120 * Load any hot spares for this pool.
2121 */
2122 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2123 if (error != 0 && error != ENOENT)
2124 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2125 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2126 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2127 if (load_nvlist(spa, spa->spa_spares.sav_object,
2128 &spa->spa_spares.sav_config) != 0)
2129 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2130
2131 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2132 spa_load_spares(spa);
2133 spa_config_exit(spa, SCL_ALL, FTAG);
2134 } else if (error == 0) {
2135 spa->spa_spares.sav_sync = B_TRUE;
2136 }
2137
2138 /*
2139 * Load any level 2 ARC devices for this pool.
2140 */
2141 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2142 &spa->spa_l2cache.sav_object);
2143 if (error != 0 && error != ENOENT)
2144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2145 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2146 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2147 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2148 &spa->spa_l2cache.sav_config) != 0)
2149 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2150
2151 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2152 spa_load_l2cache(spa);
2153 spa_config_exit(spa, SCL_ALL, FTAG);
2154 } else if (error == 0) {
2155 spa->spa_l2cache.sav_sync = B_TRUE;
2156 }
2157
2158 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2159
2160 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2161 if (error && error != ENOENT)
2162 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2163
2164 if (error == 0) {
2165 uint64_t autoreplace;
2166
2167 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2168 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2169 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2170 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2171 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2172 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2173 &spa->spa_dedup_ditto);
2174
2175 spa->spa_autoreplace = (autoreplace != 0);
2176 }
2177
2178 /*
2179 * If the 'autoreplace' property is set, then post a resource notifying
2180 * the ZFS DE that it should not issue any faults for unopenable
2181 * devices. We also iterate over the vdevs, and post a sysevent for any
2182 * unopenable vdevs so that the normal autoreplace handler can take
2183 * over.
2184 */
2185 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2186 spa_check_removed(spa->spa_root_vdev);
2187 /*
2188 * For the import case, this is done in spa_import(), because
2189 * at this point we're using the spare definitions from
2190 * the MOS config, not necessarily from the userland config.
2191 */
2192 if (state != SPA_LOAD_IMPORT) {
2193 spa_aux_check_removed(&spa->spa_spares);
2194 spa_aux_check_removed(&spa->spa_l2cache);
2195 }
2196 }
2197
2198 /*
2199 * Load the vdev state for all toplevel vdevs.
2200 */
2201 vdev_load(rvd);
2202
2203 /*
2204 * Propagate the leaf DTLs we just loaded all the way up the tree.
2205 */
2206 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2207 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2208 spa_config_exit(spa, SCL_ALL, FTAG);
2209
2210 /*
2211 * Load the DDTs (dedup tables).
2212 */
2213 error = ddt_load(spa);
2214 if (error != 0)
2215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2216
2217 spa_update_dspace(spa);
2218
2219 /*
2220 * Validate the config, using the MOS config to fill in any
2221 * information which might be missing. If we fail to validate
2222 * the config then declare the pool unfit for use. If we're
2223 * assembling a pool from a split, the log is not transferred
2224 * over.
2225 */
2226 if (type != SPA_IMPORT_ASSEMBLE) {
2227 nvlist_t *nvconfig;
2228
2229 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2230 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2231
2232 if (!spa_config_valid(spa, nvconfig)) {
2233 nvlist_free(nvconfig);
2234 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2235 ENXIO));
2236 }
2237 nvlist_free(nvconfig);
2238
2239 /*
2240 * Now that we've validate the config, check the state of the
2241 * root vdev. If it can't be opened, it indicates one or
2242 * more toplevel vdevs are faulted.
2243 */
2244 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2245 return (ENXIO);
2246
2247 if (spa_check_logs(spa)) {
2248 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2249 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2250 }
2251 }
2252
2253 /*
2254 * We've successfully opened the pool, verify that we're ready
2255 * to start pushing transactions.
2256 */
2257 if (state != SPA_LOAD_TRYIMPORT) {
2258 if (error = spa_load_verify(spa))
2259 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2260 error));
2261 }
2262
2263 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2264 spa->spa_load_max_txg == UINT64_MAX)) {
2265 dmu_tx_t *tx;
2266 int need_update = B_FALSE;
2267
2268 ASSERT(state != SPA_LOAD_TRYIMPORT);
2269
2270 /*
2271 * Claim log blocks that haven't been committed yet.
2272 * This must all happen in a single txg.
2273 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2274 * invoked from zil_claim_log_block()'s i/o done callback.
2275 * Price of rollback is that we abandon the log.
2276 */
2277 spa->spa_claiming = B_TRUE;
2278
2279 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2280 spa_first_txg(spa));
2281 (void) dmu_objset_find(spa_name(spa),
2282 zil_claim, tx, DS_FIND_CHILDREN);
2283 dmu_tx_commit(tx);
2284
2285 spa->spa_claiming = B_FALSE;
2286
2287 spa_set_log_state(spa, SPA_LOG_GOOD);
2288 spa->spa_sync_on = B_TRUE;
2289 txg_sync_start(spa->spa_dsl_pool);
2290
2291 /*
2292 * Wait for all claims to sync. We sync up to the highest
2293 * claimed log block birth time so that claimed log blocks
2294 * don't appear to be from the future. spa_claim_max_txg
2295 * will have been set for us by either zil_check_log_chain()
2296 * (invoked from spa_check_logs()) or zil_claim() above.
2297 */
2298 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2299
2300 /*
2301 * If the config cache is stale, or we have uninitialized
2302 * metaslabs (see spa_vdev_add()), then update the config.
2303 *
2304 * If this is a verbatim import, trust the current
2305 * in-core spa_config and update the disk labels.
2306 */
2307 if (config_cache_txg != spa->spa_config_txg ||
2308 state == SPA_LOAD_IMPORT ||
2309 state == SPA_LOAD_RECOVER ||
2310 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2311 need_update = B_TRUE;
2312
2313 for (int c = 0; c < rvd->vdev_children; c++)
2314 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2315 need_update = B_TRUE;
2316
2317 /*
2318 * Update the config cache asychronously in case we're the
2319 * root pool, in which case the config cache isn't writable yet.
2320 */
2321 if (need_update)
2322 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2323
2324 /*
2325 * Check all DTLs to see if anything needs resilvering.
2326 */
2327 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2328 vdev_resilver_needed(rvd, NULL, NULL))
2329 spa_async_request(spa, SPA_ASYNC_RESILVER);
2330
2331 /*
2332 * Delete any inconsistent datasets.
2333 */
2334 (void) dmu_objset_find(spa_name(spa),
2335 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2336
2337 /*
2338 * Clean up any stale temporary dataset userrefs.
2339 */
2340 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2341 }
2342
2343 return (0);
2344}
2345
2346static int
2347spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2348{
2349 int mode = spa->spa_mode;
2350
2351 spa_unload(spa);
2352 spa_deactivate(spa);
2353
2354 spa->spa_load_max_txg--;
2355
2356 spa_activate(spa, mode);
2357 spa_async_suspend(spa);
2358
2359 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2360}
2361
2362static int
2363spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2364 uint64_t max_request, int rewind_flags)
2365{
2366 nvlist_t *config = NULL;
2367 int load_error, rewind_error;
2368 uint64_t safe_rewind_txg;
2369 uint64_t min_txg;
2370
2371 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2372 spa->spa_load_max_txg = spa->spa_load_txg;
2373 spa_set_log_state(spa, SPA_LOG_CLEAR);
2374 } else {
2375 spa->spa_load_max_txg = max_request;
2376 }
2377
2378 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2379 mosconfig);
2380 if (load_error == 0)
2381 return (0);
2382
2383 if (spa->spa_root_vdev != NULL)
2384 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2385
2386 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2387 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2388
2389 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2390 nvlist_free(config);
2391 return (load_error);
2392 }
2393
2394 /* Price of rolling back is discarding txgs, including log */
2395 if (state == SPA_LOAD_RECOVER)
2396 spa_set_log_state(spa, SPA_LOG_CLEAR);
2397
2398 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2399 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2400 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2401 TXG_INITIAL : safe_rewind_txg;
2402
2403 /*
2404 * Continue as long as we're finding errors, we're still within
2405 * the acceptable rewind range, and we're still finding uberblocks
2406 */
2407 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2408 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2409 if (spa->spa_load_max_txg < safe_rewind_txg)
2410 spa->spa_extreme_rewind = B_TRUE;
2411 rewind_error = spa_load_retry(spa, state, mosconfig);
2412 }
2413
2414 spa->spa_extreme_rewind = B_FALSE;
2415 spa->spa_load_max_txg = UINT64_MAX;
2416
2417 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2418 spa_config_set(spa, config);
2419
2420 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2421}
2422
2423/*
2424 * Pool Open/Import
2425 *
2426 * The import case is identical to an open except that the configuration is sent
2427 * down from userland, instead of grabbed from the configuration cache. For the
2428 * case of an open, the pool configuration will exist in the
2429 * POOL_STATE_UNINITIALIZED state.
2430 *
2431 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2432 * the same time open the pool, without having to keep around the spa_t in some
2433 * ambiguous state.
2434 */
2435static int
2436spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2437 nvlist_t **config)
2438{
2439 spa_t *spa;
2440 spa_load_state_t state = SPA_LOAD_OPEN;
2441 int error;
2442 int locked = B_FALSE;
2443 int firstopen = B_FALSE;
2444
2445 *spapp = NULL;
2446
2447 /*
2448 * As disgusting as this is, we need to support recursive calls to this
2449 * function because dsl_dir_open() is called during spa_load(), and ends
2450 * up calling spa_open() again. The real fix is to figure out how to
2451 * avoid dsl_dir_open() calling this in the first place.
2452 */
2453 if (mutex_owner(&spa_namespace_lock) != curthread) {
2454 mutex_enter(&spa_namespace_lock);
2455 locked = B_TRUE;
2456 }
2457
2458 if ((spa = spa_lookup(pool)) == NULL) {
2459 if (locked)
2460 mutex_exit(&spa_namespace_lock);
2461 return (ENOENT);
2462 }
2463
2464 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2465 zpool_rewind_policy_t policy;
2466
2467 firstopen = B_TRUE;
2468
2469 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2470 &policy);
2471 if (policy.zrp_request & ZPOOL_DO_REWIND)
2472 state = SPA_LOAD_RECOVER;
2473
2474 spa_activate(spa, spa_mode_global);
2475
2476 if (state != SPA_LOAD_RECOVER)
2477 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2478
2479 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2480 policy.zrp_request);
2481
2482 if (error == EBADF) {
2483 /*
2484 * If vdev_validate() returns failure (indicated by
2485 * EBADF), it indicates that one of the vdevs indicates
2486 * that the pool has been exported or destroyed. If
2487 * this is the case, the config cache is out of sync and
2488 * we should remove the pool from the namespace.
2489 */
2490 spa_unload(spa);
2491 spa_deactivate(spa);
2492 spa_config_sync(spa, B_TRUE, B_TRUE);
2493 spa_remove(spa);
2494 if (locked)
2495 mutex_exit(&spa_namespace_lock);
2496 return (ENOENT);
2497 }
2498
2499 if (error) {
2500 /*
2501 * We can't open the pool, but we still have useful
2502 * information: the state of each vdev after the
2503 * attempted vdev_open(). Return this to the user.
2504 */
2505 if (config != NULL && spa->spa_config) {
2506 VERIFY(nvlist_dup(spa->spa_config, config,
2507 KM_SLEEP) == 0);
2508 VERIFY(nvlist_add_nvlist(*config,
2509 ZPOOL_CONFIG_LOAD_INFO,
2510 spa->spa_load_info) == 0);
2511 }
2512 spa_unload(spa);
2513 spa_deactivate(spa);
2514 spa->spa_last_open_failed = error;
2515 if (locked)
2516 mutex_exit(&spa_namespace_lock);
2517 *spapp = NULL;
2518 return (error);
2519 }
2520 }
2521
2522 spa_open_ref(spa, tag);
2523
2524 if (config != NULL)
2525 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2526
2527 /*
2528 * If we've recovered the pool, pass back any information we
2529 * gathered while doing the load.
2530 */
2531 if (state == SPA_LOAD_RECOVER) {
2532 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2533 spa->spa_load_info) == 0);
2534 }
2535
2536 if (locked) {
2537 spa->spa_last_open_failed = 0;
2538 spa->spa_last_ubsync_txg = 0;
2539 spa->spa_load_txg = 0;
2540 mutex_exit(&spa_namespace_lock);
2541#ifdef __FreeBSD__
2542#ifdef _KERNEL
2543 if (firstopen)
2544 zvol_create_minors(pool);
2545#endif
2546#endif
2547 }
2548
2549 *spapp = spa;
2550
2551 return (0);
2552}
2553
2554int
2555spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2556 nvlist_t **config)
2557{
2558 return (spa_open_common(name, spapp, tag, policy, config));
2559}
2560
2561int
2562spa_open(const char *name, spa_t **spapp, void *tag)
2563{
2564 return (spa_open_common(name, spapp, tag, NULL, NULL));
2565}
2566
2567/*
2568 * Lookup the given spa_t, incrementing the inject count in the process,
2569 * preventing it from being exported or destroyed.
2570 */
2571spa_t *
2572spa_inject_addref(char *name)
2573{
2574 spa_t *spa;
2575
2576 mutex_enter(&spa_namespace_lock);
2577 if ((spa = spa_lookup(name)) == NULL) {
2578 mutex_exit(&spa_namespace_lock);
2579 return (NULL);
2580 }
2581 spa->spa_inject_ref++;
2582 mutex_exit(&spa_namespace_lock);
2583
2584 return (spa);
2585}
2586
2587void
2588spa_inject_delref(spa_t *spa)
2589{
2590 mutex_enter(&spa_namespace_lock);
2591 spa->spa_inject_ref--;
2592 mutex_exit(&spa_namespace_lock);
2593}
2594
2595/*
2596 * Add spares device information to the nvlist.
2597 */
2598static void
2599spa_add_spares(spa_t *spa, nvlist_t *config)
2600{
2601 nvlist_t **spares;
2602 uint_t i, nspares;
2603 nvlist_t *nvroot;
2604 uint64_t guid;
2605 vdev_stat_t *vs;
2606 uint_t vsc;
2607 uint64_t pool;
2608
2609 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2610
2611 if (spa->spa_spares.sav_count == 0)
2612 return;
2613
2614 VERIFY(nvlist_lookup_nvlist(config,
2615 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2616 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2617 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2618 if (nspares != 0) {
2619 VERIFY(nvlist_add_nvlist_array(nvroot,
2620 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2621 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2622 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2623
2624 /*
2625 * Go through and find any spares which have since been
2626 * repurposed as an active spare. If this is the case, update
2627 * their status appropriately.
2628 */
2629 for (i = 0; i < nspares; i++) {
2630 VERIFY(nvlist_lookup_uint64(spares[i],
2631 ZPOOL_CONFIG_GUID, &guid) == 0);
2632 if (spa_spare_exists(guid, &pool, NULL) &&
2633 pool != 0ULL) {
2634 VERIFY(nvlist_lookup_uint64_array(
2635 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2636 (uint64_t **)&vs, &vsc) == 0);
2637 vs->vs_state = VDEV_STATE_CANT_OPEN;
2638 vs->vs_aux = VDEV_AUX_SPARED;
2639 }
2640 }
2641 }
2642}
2643
2644/*
2645 * Add l2cache device information to the nvlist, including vdev stats.
2646 */
2647static void
2648spa_add_l2cache(spa_t *spa, nvlist_t *config)
2649{
2650 nvlist_t **l2cache;
2651 uint_t i, j, nl2cache;
2652 nvlist_t *nvroot;
2653 uint64_t guid;
2654 vdev_t *vd;
2655 vdev_stat_t *vs;
2656 uint_t vsc;
2657
2658 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2659
2660 if (spa->spa_l2cache.sav_count == 0)
2661 return;
2662
2663 VERIFY(nvlist_lookup_nvlist(config,
2664 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2665 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2666 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2667 if (nl2cache != 0) {
2668 VERIFY(nvlist_add_nvlist_array(nvroot,
2669 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2670 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2671 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2672
2673 /*
2674 * Update level 2 cache device stats.
2675 */
2676
2677 for (i = 0; i < nl2cache; i++) {
2678 VERIFY(nvlist_lookup_uint64(l2cache[i],
2679 ZPOOL_CONFIG_GUID, &guid) == 0);
2680
2681 vd = NULL;
2682 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2683 if (guid ==
2684 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2685 vd = spa->spa_l2cache.sav_vdevs[j];
2686 break;
2687 }
2688 }
2689 ASSERT(vd != NULL);
2690
2691 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2692 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2693 == 0);
2694 vdev_get_stats(vd, vs);
2695 }
2696 }
2697}
2698
2699int
2700spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2701{
2702 int error;
2703 spa_t *spa;
2704
2705 *config = NULL;
2706 error = spa_open_common(name, &spa, FTAG, NULL, config);
2707
2708 if (spa != NULL) {
2709 /*
2710 * This still leaves a window of inconsistency where the spares
2711 * or l2cache devices could change and the config would be
2712 * self-inconsistent.
2713 */
2714 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2715
2716 if (*config != NULL) {
2717 uint64_t loadtimes[2];
2718
2719 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2720 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2721 VERIFY(nvlist_add_uint64_array(*config,
2722 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2723
2724 VERIFY(nvlist_add_uint64(*config,
2725 ZPOOL_CONFIG_ERRCOUNT,
2726 spa_get_errlog_size(spa)) == 0);
2727
2728 if (spa_suspended(spa))
2729 VERIFY(nvlist_add_uint64(*config,
2730 ZPOOL_CONFIG_SUSPENDED,
2731 spa->spa_failmode) == 0);
2732
2733 spa_add_spares(spa, *config);
2734 spa_add_l2cache(spa, *config);
2735 }
2736 }
2737
2738 /*
2739 * We want to get the alternate root even for faulted pools, so we cheat
2740 * and call spa_lookup() directly.
2741 */
2742 if (altroot) {
2743 if (spa == NULL) {
2744 mutex_enter(&spa_namespace_lock);
2745 spa = spa_lookup(name);
2746 if (spa)
2747 spa_altroot(spa, altroot, buflen);
2748 else
2749 altroot[0] = '\0';
2750 spa = NULL;
2751 mutex_exit(&spa_namespace_lock);
2752 } else {
2753 spa_altroot(spa, altroot, buflen);
2754 }
2755 }
2756
2757 if (spa != NULL) {
2758 spa_config_exit(spa, SCL_CONFIG, FTAG);
2759 spa_close(spa, FTAG);
2760 }
2761
2762 return (error);
2763}
2764
2765/*
2766 * Validate that the auxiliary device array is well formed. We must have an
2767 * array of nvlists, each which describes a valid leaf vdev. If this is an
2768 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2769 * specified, as long as they are well-formed.
2770 */
2771static int
2772spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2773 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2774 vdev_labeltype_t label)
2775{
2776 nvlist_t **dev;
2777 uint_t i, ndev;
2778 vdev_t *vd;
2779 int error;
2780
2781 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2782
2783 /*
2784 * It's acceptable to have no devs specified.
2785 */
2786 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2787 return (0);
2788
2789 if (ndev == 0)
2790 return (EINVAL);
2791
2792 /*
2793 * Make sure the pool is formatted with a version that supports this
2794 * device type.
2795 */
2796 if (spa_version(spa) < version)
2797 return (ENOTSUP);
2798
2799 /*
2800 * Set the pending device list so we correctly handle device in-use
2801 * checking.
2802 */
2803 sav->sav_pending = dev;
2804 sav->sav_npending = ndev;
2805
2806 for (i = 0; i < ndev; i++) {
2807 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2808 mode)) != 0)
2809 goto out;
2810
2811 if (!vd->vdev_ops->vdev_op_leaf) {
2812 vdev_free(vd);
2813 error = EINVAL;
2814 goto out;
2815 }
2816
2817 /*
2818 * The L2ARC currently only supports disk devices in
2819 * kernel context. For user-level testing, we allow it.
2820 */
2821#ifdef _KERNEL
2822 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2823 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2824 error = ENOTBLK;
|
| 2825 vdev_free(vd);
|
2821 goto out;
2822 }
2823#endif
2824 vd->vdev_top = vd;
2825
2826 if ((error = vdev_open(vd)) == 0 &&
2827 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2828 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2829 vd->vdev_guid) == 0);
2830 }
2831
2832 vdev_free(vd);
2833
2834 if (error &&
2835 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2836 goto out;
2837 else
2838 error = 0;
2839 }
2840
2841out:
2842 sav->sav_pending = NULL;
2843 sav->sav_npending = 0;
2844 return (error);
2845}
2846
2847static int
2848spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2849{
2850 int error;
2851
2852 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2853
2854 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2855 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2856 VDEV_LABEL_SPARE)) != 0) {
2857 return (error);
2858 }
2859
2860 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2861 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2862 VDEV_LABEL_L2CACHE));
2863}
2864
2865static void
2866spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2867 const char *config)
2868{
2869 int i;
2870
2871 if (sav->sav_config != NULL) {
2872 nvlist_t **olddevs;
2873 uint_t oldndevs;
2874 nvlist_t **newdevs;
2875
2876 /*
2877 * Generate new dev list by concatentating with the
2878 * current dev list.
2879 */
2880 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2881 &olddevs, &oldndevs) == 0);
2882
2883 newdevs = kmem_alloc(sizeof (void *) *
2884 (ndevs + oldndevs), KM_SLEEP);
2885 for (i = 0; i < oldndevs; i++)
2886 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2887 KM_SLEEP) == 0);
2888 for (i = 0; i < ndevs; i++)
2889 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2890 KM_SLEEP) == 0);
2891
2892 VERIFY(nvlist_remove(sav->sav_config, config,
2893 DATA_TYPE_NVLIST_ARRAY) == 0);
2894
2895 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2896 config, newdevs, ndevs + oldndevs) == 0);
2897 for (i = 0; i < oldndevs + ndevs; i++)
2898 nvlist_free(newdevs[i]);
2899 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2900 } else {
2901 /*
2902 * Generate a new dev list.
2903 */
2904 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2905 KM_SLEEP) == 0);
2906 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2907 devs, ndevs) == 0);
2908 }
2909}
2910
2911/*
2912 * Stop and drop level 2 ARC devices
2913 */
2914void
2915spa_l2cache_drop(spa_t *spa)
2916{
2917 vdev_t *vd;
2918 int i;
2919 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2920
2921 for (i = 0; i < sav->sav_count; i++) {
2922 uint64_t pool;
2923
2924 vd = sav->sav_vdevs[i];
2925 ASSERT(vd != NULL);
2926
2927 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2928 pool != 0ULL && l2arc_vdev_present(vd))
2929 l2arc_remove_vdev(vd);
| 2826 goto out;
2827 }
2828#endif
2829 vd->vdev_top = vd;
2830
2831 if ((error = vdev_open(vd)) == 0 &&
2832 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2833 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2834 vd->vdev_guid) == 0);
2835 }
2836
2837 vdev_free(vd);
2838
2839 if (error &&
2840 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2841 goto out;
2842 else
2843 error = 0;
2844 }
2845
2846out:
2847 sav->sav_pending = NULL;
2848 sav->sav_npending = 0;
2849 return (error);
2850}
2851
2852static int
2853spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2854{
2855 int error;
2856
2857 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2858
2859 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2860 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2861 VDEV_LABEL_SPARE)) != 0) {
2862 return (error);
2863 }
2864
2865 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2866 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2867 VDEV_LABEL_L2CACHE));
2868}
2869
2870static void
2871spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2872 const char *config)
2873{
2874 int i;
2875
2876 if (sav->sav_config != NULL) {
2877 nvlist_t **olddevs;
2878 uint_t oldndevs;
2879 nvlist_t **newdevs;
2880
2881 /*
2882 * Generate new dev list by concatentating with the
2883 * current dev list.
2884 */
2885 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2886 &olddevs, &oldndevs) == 0);
2887
2888 newdevs = kmem_alloc(sizeof (void *) *
2889 (ndevs + oldndevs), KM_SLEEP);
2890 for (i = 0; i < oldndevs; i++)
2891 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2892 KM_SLEEP) == 0);
2893 for (i = 0; i < ndevs; i++)
2894 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2895 KM_SLEEP) == 0);
2896
2897 VERIFY(nvlist_remove(sav->sav_config, config,
2898 DATA_TYPE_NVLIST_ARRAY) == 0);
2899
2900 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2901 config, newdevs, ndevs + oldndevs) == 0);
2902 for (i = 0; i < oldndevs + ndevs; i++)
2903 nvlist_free(newdevs[i]);
2904 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2905 } else {
2906 /*
2907 * Generate a new dev list.
2908 */
2909 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2910 KM_SLEEP) == 0);
2911 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2912 devs, ndevs) == 0);
2913 }
2914}
2915
2916/*
2917 * Stop and drop level 2 ARC devices
2918 */
2919void
2920spa_l2cache_drop(spa_t *spa)
2921{
2922 vdev_t *vd;
2923 int i;
2924 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2925
2926 for (i = 0; i < sav->sav_count; i++) {
2927 uint64_t pool;
2928
2929 vd = sav->sav_vdevs[i];
2930 ASSERT(vd != NULL);
2931
2932 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2933 pool != 0ULL && l2arc_vdev_present(vd))
2934 l2arc_remove_vdev(vd);
|
2930 if (vd->vdev_isl2cache)
2931 spa_l2cache_remove(vd);
2932 vdev_clear_stats(vd);
2933 (void) vdev_close(vd);
| |
2934 }
2935}
2936
2937/*
2938 * Pool Creation
2939 */
2940int
2941spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2942 const char *history_str, nvlist_t *zplprops)
2943{
2944 spa_t *spa;
2945 char *altroot = NULL;
2946 vdev_t *rvd;
2947 dsl_pool_t *dp;
2948 dmu_tx_t *tx;
2949 int error = 0;
2950 uint64_t txg = TXG_INITIAL;
2951 nvlist_t **spares, **l2cache;
2952 uint_t nspares, nl2cache;
2953 uint64_t version, obj;
2954
2955 /*
2956 * If this pool already exists, return failure.
2957 */
2958 mutex_enter(&spa_namespace_lock);
2959 if (spa_lookup(pool) != NULL) {
2960 mutex_exit(&spa_namespace_lock);
2961 return (EEXIST);
2962 }
2963
2964 /*
2965 * Allocate a new spa_t structure.
2966 */
2967 (void) nvlist_lookup_string(props,
2968 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2969 spa = spa_add(pool, NULL, altroot);
2970 spa_activate(spa, spa_mode_global);
2971
2972 if (props && (error = spa_prop_validate(spa, props))) {
2973 spa_deactivate(spa);
2974 spa_remove(spa);
2975 mutex_exit(&spa_namespace_lock);
2976 return (error);
2977 }
2978
2979 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2980 &version) != 0)
2981 version = SPA_VERSION;
2982 ASSERT(version <= SPA_VERSION);
2983
2984 spa->spa_first_txg = txg;
2985 spa->spa_uberblock.ub_txg = txg - 1;
2986 spa->spa_uberblock.ub_version = version;
2987 spa->spa_ubsync = spa->spa_uberblock;
2988
2989 /*
2990 * Create "The Godfather" zio to hold all async IOs
2991 */
2992 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2993 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2994
2995 /*
2996 * Create the root vdev.
2997 */
2998 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2999
3000 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3001
3002 ASSERT(error != 0 || rvd != NULL);
3003 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3004
3005 if (error == 0 && !zfs_allocatable_devs(nvroot))
3006 error = EINVAL;
3007
3008 if (error == 0 &&
3009 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3010 (error = spa_validate_aux(spa, nvroot, txg,
3011 VDEV_ALLOC_ADD)) == 0) {
3012 for (int c = 0; c < rvd->vdev_children; c++) {
3013 vdev_metaslab_set_size(rvd->vdev_child[c]);
3014 vdev_expand(rvd->vdev_child[c], txg);
3015 }
3016 }
3017
3018 spa_config_exit(spa, SCL_ALL, FTAG);
3019
3020 if (error != 0) {
3021 spa_unload(spa);
3022 spa_deactivate(spa);
3023 spa_remove(spa);
3024 mutex_exit(&spa_namespace_lock);
3025 return (error);
3026 }
3027
3028 /*
3029 * Get the list of spares, if specified.
3030 */
3031 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3032 &spares, &nspares) == 0) {
3033 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3034 KM_SLEEP) == 0);
3035 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3036 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3037 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3038 spa_load_spares(spa);
3039 spa_config_exit(spa, SCL_ALL, FTAG);
3040 spa->spa_spares.sav_sync = B_TRUE;
3041 }
3042
3043 /*
3044 * Get the list of level 2 cache devices, if specified.
3045 */
3046 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3047 &l2cache, &nl2cache) == 0) {
3048 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3049 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3050 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3051 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3052 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3053 spa_load_l2cache(spa);
3054 spa_config_exit(spa, SCL_ALL, FTAG);
3055 spa->spa_l2cache.sav_sync = B_TRUE;
3056 }
3057
3058 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3059 spa->spa_meta_objset = dp->dp_meta_objset;
3060
3061 /*
3062 * Create DDTs (dedup tables).
3063 */
3064 ddt_create(spa);
3065
3066 spa_update_dspace(spa);
3067
3068 tx = dmu_tx_create_assigned(dp, txg);
3069
3070 /*
3071 * Create the pool config object.
3072 */
3073 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3074 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3075 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3076
3077 if (zap_add(spa->spa_meta_objset,
3078 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3079 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3080 cmn_err(CE_PANIC, "failed to add pool config");
3081 }
3082
3083 if (zap_add(spa->spa_meta_objset,
3084 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3085 sizeof (uint64_t), 1, &version, tx) != 0) {
3086 cmn_err(CE_PANIC, "failed to add pool version");
3087 }
3088
3089 /* Newly created pools with the right version are always deflated. */
3090 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3091 spa->spa_deflate = TRUE;
3092 if (zap_add(spa->spa_meta_objset,
3093 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3094 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3095 cmn_err(CE_PANIC, "failed to add deflate");
3096 }
3097 }
3098
3099 /*
3100 * Create the deferred-free bpobj. Turn off compression
3101 * because sync-to-convergence takes longer if the blocksize
3102 * keeps changing.
3103 */
3104 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3105 dmu_object_set_compress(spa->spa_meta_objset, obj,
3106 ZIO_COMPRESS_OFF, tx);
3107 if (zap_add(spa->spa_meta_objset,
3108 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3109 sizeof (uint64_t), 1, &obj, tx) != 0) {
3110 cmn_err(CE_PANIC, "failed to add bpobj");
3111 }
3112 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3113 spa->spa_meta_objset, obj));
3114
3115 /*
3116 * Create the pool's history object.
3117 */
3118 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3119 spa_history_create_obj(spa, tx);
3120
3121 /*
3122 * Set pool properties.
3123 */
3124 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3125 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3126 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3127 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3128
3129 if (props != NULL) {
3130 spa_configfile_set(spa, props, B_FALSE);
3131 spa_sync_props(spa, props, tx);
3132 }
3133
3134 dmu_tx_commit(tx);
3135
3136 spa->spa_sync_on = B_TRUE;
3137 txg_sync_start(spa->spa_dsl_pool);
3138
3139 /*
3140 * We explicitly wait for the first transaction to complete so that our
3141 * bean counters are appropriately updated.
3142 */
3143 txg_wait_synced(spa->spa_dsl_pool, txg);
3144
3145 spa_config_sync(spa, B_FALSE, B_TRUE);
3146
3147 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3148 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3149 spa_history_log_version(spa, LOG_POOL_CREATE);
3150
3151 spa->spa_minref = refcount_count(&spa->spa_refcount);
3152
3153 mutex_exit(&spa_namespace_lock);
3154
3155 return (0);
3156}
3157
3158#if defined(sun)
3159#ifdef _KERNEL
3160/*
3161 * Get the root pool information from the root disk, then import the root pool
3162 * during the system boot up time.
3163 */
3164extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3165
3166static nvlist_t *
3167spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3168{
3169 nvlist_t *config;
3170 nvlist_t *nvtop, *nvroot;
3171 uint64_t pgid;
3172
3173 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3174 return (NULL);
3175
3176 /*
3177 * Add this top-level vdev to the child array.
3178 */
3179 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3180 &nvtop) == 0);
3181 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3182 &pgid) == 0);
3183 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3184
3185 /*
3186 * Put this pool's top-level vdevs into a root vdev.
3187 */
3188 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3189 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3190 VDEV_TYPE_ROOT) == 0);
3191 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3192 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3193 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3194 &nvtop, 1) == 0);
3195
3196 /*
3197 * Replace the existing vdev_tree with the new root vdev in
3198 * this pool's configuration (remove the old, add the new).
3199 */
3200 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3201 nvlist_free(nvroot);
3202 return (config);
3203}
3204
3205/*
3206 * Walk the vdev tree and see if we can find a device with "better"
3207 * configuration. A configuration is "better" if the label on that
3208 * device has a more recent txg.
3209 */
3210static void
3211spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3212{
3213 for (int c = 0; c < vd->vdev_children; c++)
3214 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3215
3216 if (vd->vdev_ops->vdev_op_leaf) {
3217 nvlist_t *label;
3218 uint64_t label_txg;
3219
3220 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3221 &label) != 0)
3222 return;
3223
3224 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3225 &label_txg) == 0);
3226
3227 /*
3228 * Do we have a better boot device?
3229 */
3230 if (label_txg > *txg) {
3231 *txg = label_txg;
3232 *avd = vd;
3233 }
3234 nvlist_free(label);
3235 }
3236}
3237
3238/*
3239 * Import a root pool.
3240 *
3241 * For x86. devpath_list will consist of devid and/or physpath name of
3242 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3243 * The GRUB "findroot" command will return the vdev we should boot.
3244 *
3245 * For Sparc, devpath_list consists the physpath name of the booting device
3246 * no matter the rootpool is a single device pool or a mirrored pool.
3247 * e.g.
3248 * "/pci@1f,0/ide@d/disk@0,0:a"
3249 */
3250int
3251spa_import_rootpool(char *devpath, char *devid)
3252{
3253 spa_t *spa;
3254 vdev_t *rvd, *bvd, *avd = NULL;
3255 nvlist_t *config, *nvtop;
3256 uint64_t guid, txg;
3257 char *pname;
3258 int error;
3259
3260 /*
3261 * Read the label from the boot device and generate a configuration.
3262 */
3263 config = spa_generate_rootconf(devpath, devid, &guid);
3264#if defined(_OBP) && defined(_KERNEL)
3265 if (config == NULL) {
3266 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3267 /* iscsi boot */
3268 get_iscsi_bootpath_phy(devpath);
3269 config = spa_generate_rootconf(devpath, devid, &guid);
3270 }
3271 }
3272#endif
3273 if (config == NULL) {
3274 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3275 devpath);
3276 return (EIO);
3277 }
3278
3279 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3280 &pname) == 0);
3281 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3282
3283 mutex_enter(&spa_namespace_lock);
3284 if ((spa = spa_lookup(pname)) != NULL) {
3285 /*
3286 * Remove the existing root pool from the namespace so that we
3287 * can replace it with the correct config we just read in.
3288 */
3289 spa_remove(spa);
3290 }
3291
3292 spa = spa_add(pname, config, NULL);
3293 spa->spa_is_root = B_TRUE;
3294 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3295
3296 /*
3297 * Build up a vdev tree based on the boot device's label config.
3298 */
3299 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3300 &nvtop) == 0);
3301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3302 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3303 VDEV_ALLOC_ROOTPOOL);
3304 spa_config_exit(spa, SCL_ALL, FTAG);
3305 if (error) {
3306 mutex_exit(&spa_namespace_lock);
3307 nvlist_free(config);
3308 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3309 pname);
3310 return (error);
3311 }
3312
3313 /*
3314 * Get the boot vdev.
3315 */
3316 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3317 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3318 (u_longlong_t)guid);
3319 error = ENOENT;
3320 goto out;
3321 }
3322
3323 /*
3324 * Determine if there is a better boot device.
3325 */
3326 avd = bvd;
3327 spa_alt_rootvdev(rvd, &avd, &txg);
3328 if (avd != bvd) {
3329 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3330 "try booting from '%s'", avd->vdev_path);
3331 error = EINVAL;
3332 goto out;
3333 }
3334
3335 /*
3336 * If the boot device is part of a spare vdev then ensure that
3337 * we're booting off the active spare.
3338 */
3339 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3340 !bvd->vdev_isspare) {
3341 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3342 "try booting from '%s'",
3343 bvd->vdev_parent->
3344 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3345 error = EINVAL;
3346 goto out;
3347 }
3348
3349 error = 0;
3350 spa_history_log_version(spa, LOG_POOL_IMPORT);
3351out:
3352 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3353 vdev_free(rvd);
3354 spa_config_exit(spa, SCL_ALL, FTAG);
3355 mutex_exit(&spa_namespace_lock);
3356
3357 nvlist_free(config);
3358 return (error);
3359}
3360
3361#endif
3362#endif /* sun */
3363
3364/*
3365 * Import a non-root pool into the system.
3366 */
3367int
3368spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3369{
3370 spa_t *spa;
3371 char *altroot = NULL;
3372 spa_load_state_t state = SPA_LOAD_IMPORT;
3373 zpool_rewind_policy_t policy;
3374 uint64_t mode = spa_mode_global;
3375 uint64_t readonly = B_FALSE;
3376 int error;
3377 nvlist_t *nvroot;
3378 nvlist_t **spares, **l2cache;
3379 uint_t nspares, nl2cache;
3380
3381 /*
3382 * If a pool with this name exists, return failure.
3383 */
3384 mutex_enter(&spa_namespace_lock);
3385 if (spa_lookup(pool) != NULL) {
3386 mutex_exit(&spa_namespace_lock);
3387 return (EEXIST);
3388 }
3389
3390 /*
3391 * Create and initialize the spa structure.
3392 */
3393 (void) nvlist_lookup_string(props,
3394 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3395 (void) nvlist_lookup_uint64(props,
3396 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3397 if (readonly)
3398 mode = FREAD;
3399 spa = spa_add(pool, config, altroot);
3400 spa->spa_import_flags = flags;
3401
3402 /*
3403 * Verbatim import - Take a pool and insert it into the namespace
3404 * as if it had been loaded at boot.
3405 */
3406 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3407 if (props != NULL)
3408 spa_configfile_set(spa, props, B_FALSE);
3409
3410 spa_config_sync(spa, B_FALSE, B_TRUE);
3411
3412 mutex_exit(&spa_namespace_lock);
3413 spa_history_log_version(spa, LOG_POOL_IMPORT);
3414
3415 return (0);
3416 }
3417
3418 spa_activate(spa, mode);
3419
3420 /*
3421 * Don't start async tasks until we know everything is healthy.
3422 */
3423 spa_async_suspend(spa);
3424
3425 zpool_get_rewind_policy(config, &policy);
3426 if (policy.zrp_request & ZPOOL_DO_REWIND)
3427 state = SPA_LOAD_RECOVER;
3428
3429 /*
3430 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3431 * because the user-supplied config is actually the one to trust when
3432 * doing an import.
3433 */
3434 if (state != SPA_LOAD_RECOVER)
3435 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3436
3437 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3438 policy.zrp_request);
3439
3440 /*
3441 * Propagate anything learned while loading the pool and pass it
3442 * back to caller (i.e. rewind info, missing devices, etc).
3443 */
3444 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3445 spa->spa_load_info) == 0);
3446
3447 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3448 /*
3449 * Toss any existing sparelist, as it doesn't have any validity
3450 * anymore, and conflicts with spa_has_spare().
3451 */
3452 if (spa->spa_spares.sav_config) {
3453 nvlist_free(spa->spa_spares.sav_config);
3454 spa->spa_spares.sav_config = NULL;
3455 spa_load_spares(spa);
3456 }
3457 if (spa->spa_l2cache.sav_config) {
3458 nvlist_free(spa->spa_l2cache.sav_config);
3459 spa->spa_l2cache.sav_config = NULL;
3460 spa_load_l2cache(spa);
3461 }
3462
3463 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3464 &nvroot) == 0);
3465 if (error == 0)
3466 error = spa_validate_aux(spa, nvroot, -1ULL,
3467 VDEV_ALLOC_SPARE);
3468 if (error == 0)
3469 error = spa_validate_aux(spa, nvroot, -1ULL,
3470 VDEV_ALLOC_L2CACHE);
3471 spa_config_exit(spa, SCL_ALL, FTAG);
3472
3473 if (props != NULL)
3474 spa_configfile_set(spa, props, B_FALSE);
3475
3476 if (error != 0 || (props && spa_writeable(spa) &&
3477 (error = spa_prop_set(spa, props)))) {
3478 spa_unload(spa);
3479 spa_deactivate(spa);
3480 spa_remove(spa);
3481 mutex_exit(&spa_namespace_lock);
3482 return (error);
3483 }
3484
3485 spa_async_resume(spa);
3486
3487 /*
3488 * Override any spares and level 2 cache devices as specified by
3489 * the user, as these may have correct device names/devids, etc.
3490 */
3491 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3492 &spares, &nspares) == 0) {
3493 if (spa->spa_spares.sav_config)
3494 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3495 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3496 else
3497 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3498 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3499 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3500 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3501 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3502 spa_load_spares(spa);
3503 spa_config_exit(spa, SCL_ALL, FTAG);
3504 spa->spa_spares.sav_sync = B_TRUE;
3505 }
3506 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3507 &l2cache, &nl2cache) == 0) {
3508 if (spa->spa_l2cache.sav_config)
3509 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3510 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3511 else
3512 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3513 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3514 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3515 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3516 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3517 spa_load_l2cache(spa);
3518 spa_config_exit(spa, SCL_ALL, FTAG);
3519 spa->spa_l2cache.sav_sync = B_TRUE;
3520 }
3521
3522 /*
3523 * Check for any removed devices.
3524 */
3525 if (spa->spa_autoreplace) {
3526 spa_aux_check_removed(&spa->spa_spares);
3527 spa_aux_check_removed(&spa->spa_l2cache);
3528 }
3529
3530 if (spa_writeable(spa)) {
3531 /*
3532 * Update the config cache to include the newly-imported pool.
3533 */
3534 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3535 }
3536
3537 /*
3538 * It's possible that the pool was expanded while it was exported.
3539 * We kick off an async task to handle this for us.
3540 */
3541 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3542
3543 mutex_exit(&spa_namespace_lock);
3544 spa_history_log_version(spa, LOG_POOL_IMPORT);
3545
3546#ifdef __FreeBSD__
3547#ifdef _KERNEL
3548 zvol_create_minors(pool);
3549#endif
3550#endif
3551 return (0);
3552}
3553
3554nvlist_t *
3555spa_tryimport(nvlist_t *tryconfig)
3556{
3557 nvlist_t *config = NULL;
3558 char *poolname;
3559 spa_t *spa;
3560 uint64_t state;
3561 int error;
3562
3563 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3564 return (NULL);
3565
3566 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3567 return (NULL);
3568
3569 /*
3570 * Create and initialize the spa structure.
3571 */
3572 mutex_enter(&spa_namespace_lock);
3573 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3574 spa_activate(spa, FREAD);
3575
3576 /*
3577 * Pass off the heavy lifting to spa_load().
3578 * Pass TRUE for mosconfig because the user-supplied config
3579 * is actually the one to trust when doing an import.
3580 */
3581 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3582
3583 /*
3584 * If 'tryconfig' was at least parsable, return the current config.
3585 */
3586 if (spa->spa_root_vdev != NULL) {
3587 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3588 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3589 poolname) == 0);
3590 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3591 state) == 0);
3592 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3593 spa->spa_uberblock.ub_timestamp) == 0);
3594
3595 /*
3596 * If the bootfs property exists on this pool then we
3597 * copy it out so that external consumers can tell which
3598 * pools are bootable.
3599 */
3600 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3601 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3602
3603 /*
3604 * We have to play games with the name since the
3605 * pool was opened as TRYIMPORT_NAME.
3606 */
3607 if (dsl_dsobj_to_dsname(spa_name(spa),
3608 spa->spa_bootfs, tmpname) == 0) {
3609 char *cp;
3610 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3611
3612 cp = strchr(tmpname, '/');
3613 if (cp == NULL) {
3614 (void) strlcpy(dsname, tmpname,
3615 MAXPATHLEN);
3616 } else {
3617 (void) snprintf(dsname, MAXPATHLEN,
3618 "%s/%s", poolname, ++cp);
3619 }
3620 VERIFY(nvlist_add_string(config,
3621 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3622 kmem_free(dsname, MAXPATHLEN);
3623 }
3624 kmem_free(tmpname, MAXPATHLEN);
3625 }
3626
3627 /*
3628 * Add the list of hot spares and level 2 cache devices.
3629 */
3630 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3631 spa_add_spares(spa, config);
3632 spa_add_l2cache(spa, config);
3633 spa_config_exit(spa, SCL_CONFIG, FTAG);
3634 }
3635
3636 spa_unload(spa);
3637 spa_deactivate(spa);
3638 spa_remove(spa);
3639 mutex_exit(&spa_namespace_lock);
3640
3641 return (config);
3642}
3643
3644/*
3645 * Pool export/destroy
3646 *
3647 * The act of destroying or exporting a pool is very simple. We make sure there
3648 * is no more pending I/O and any references to the pool are gone. Then, we
3649 * update the pool state and sync all the labels to disk, removing the
3650 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3651 * we don't sync the labels or remove the configuration cache.
3652 */
3653static int
3654spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3655 boolean_t force, boolean_t hardforce)
3656{
3657 spa_t *spa;
3658
3659 if (oldconfig)
3660 *oldconfig = NULL;
3661
3662 if (!(spa_mode_global & FWRITE))
3663 return (EROFS);
3664
3665 mutex_enter(&spa_namespace_lock);
3666 if ((spa = spa_lookup(pool)) == NULL) {
3667 mutex_exit(&spa_namespace_lock);
3668 return (ENOENT);
3669 }
3670
3671 /*
3672 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3673 * reacquire the namespace lock, and see if we can export.
3674 */
3675 spa_open_ref(spa, FTAG);
3676 mutex_exit(&spa_namespace_lock);
3677 spa_async_suspend(spa);
3678 mutex_enter(&spa_namespace_lock);
3679 spa_close(spa, FTAG);
3680
3681 /*
3682 * The pool will be in core if it's openable,
3683 * in which case we can modify its state.
3684 */
3685 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3686 /*
3687 * Objsets may be open only because they're dirty, so we
3688 * have to force it to sync before checking spa_refcnt.
3689 */
3690 txg_wait_synced(spa->spa_dsl_pool, 0);
3691
3692 /*
3693 * A pool cannot be exported or destroyed if there are active
3694 * references. If we are resetting a pool, allow references by
3695 * fault injection handlers.
3696 */
3697 if (!spa_refcount_zero(spa) ||
3698 (spa->spa_inject_ref != 0 &&
3699 new_state != POOL_STATE_UNINITIALIZED)) {
3700 spa_async_resume(spa);
3701 mutex_exit(&spa_namespace_lock);
3702 return (EBUSY);
3703 }
3704
3705 /*
3706 * A pool cannot be exported if it has an active shared spare.
3707 * This is to prevent other pools stealing the active spare
3708 * from an exported pool. At user's own will, such pool can
3709 * be forcedly exported.
3710 */
3711 if (!force && new_state == POOL_STATE_EXPORTED &&
3712 spa_has_active_shared_spare(spa)) {
3713 spa_async_resume(spa);
3714 mutex_exit(&spa_namespace_lock);
3715 return (EXDEV);
3716 }
3717
3718 /*
3719 * We want this to be reflected on every label,
3720 * so mark them all dirty. spa_unload() will do the
3721 * final sync that pushes these changes out.
3722 */
3723 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3724 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3725 spa->spa_state = new_state;
3726 spa->spa_final_txg = spa_last_synced_txg(spa) +
3727 TXG_DEFER_SIZE + 1;
3728 vdev_config_dirty(spa->spa_root_vdev);
3729 spa_config_exit(spa, SCL_ALL, FTAG);
3730 }
3731 }
3732
3733 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3734
3735 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3736 spa_unload(spa);
3737 spa_deactivate(spa);
3738 }
3739
3740 if (oldconfig && spa->spa_config)
3741 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3742
3743 if (new_state != POOL_STATE_UNINITIALIZED) {
3744 if (!hardforce)
3745 spa_config_sync(spa, B_TRUE, B_TRUE);
3746 spa_remove(spa);
3747 }
3748 mutex_exit(&spa_namespace_lock);
3749
3750 return (0);
3751}
3752
3753/*
3754 * Destroy a storage pool.
3755 */
3756int
3757spa_destroy(char *pool)
3758{
3759 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3760 B_FALSE, B_FALSE));
3761}
3762
3763/*
3764 * Export a storage pool.
3765 */
3766int
3767spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3768 boolean_t hardforce)
3769{
3770 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3771 force, hardforce));
3772}
3773
3774/*
3775 * Similar to spa_export(), this unloads the spa_t without actually removing it
3776 * from the namespace in any way.
3777 */
3778int
3779spa_reset(char *pool)
3780{
3781 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3782 B_FALSE, B_FALSE));
3783}
3784
3785/*
3786 * ==========================================================================
3787 * Device manipulation
3788 * ==========================================================================
3789 */
3790
3791/*
3792 * Add a device to a storage pool.
3793 */
3794int
3795spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3796{
3797 uint64_t txg, id;
3798 int error;
3799 vdev_t *rvd = spa->spa_root_vdev;
3800 vdev_t *vd, *tvd;
3801 nvlist_t **spares, **l2cache;
3802 uint_t nspares, nl2cache;
3803
3804 ASSERT(spa_writeable(spa));
3805
3806 txg = spa_vdev_enter(spa);
3807
3808 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3809 VDEV_ALLOC_ADD)) != 0)
3810 return (spa_vdev_exit(spa, NULL, txg, error));
3811
3812 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3813
3814 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3815 &nspares) != 0)
3816 nspares = 0;
3817
3818 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3819 &nl2cache) != 0)
3820 nl2cache = 0;
3821
3822 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3823 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3824
3825 if (vd->vdev_children != 0 &&
3826 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3827 return (spa_vdev_exit(spa, vd, txg, error));
3828
3829 /*
3830 * We must validate the spares and l2cache devices after checking the
3831 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3832 */
3833 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3834 return (spa_vdev_exit(spa, vd, txg, error));
3835
3836 /*
3837 * Transfer each new top-level vdev from vd to rvd.
3838 */
3839 for (int c = 0; c < vd->vdev_children; c++) {
3840
3841 /*
3842 * Set the vdev id to the first hole, if one exists.
3843 */
3844 for (id = 0; id < rvd->vdev_children; id++) {
3845 if (rvd->vdev_child[id]->vdev_ishole) {
3846 vdev_free(rvd->vdev_child[id]);
3847 break;
3848 }
3849 }
3850 tvd = vd->vdev_child[c];
3851 vdev_remove_child(vd, tvd);
3852 tvd->vdev_id = id;
3853 vdev_add_child(rvd, tvd);
3854 vdev_config_dirty(tvd);
3855 }
3856
3857 if (nspares != 0) {
3858 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3859 ZPOOL_CONFIG_SPARES);
3860 spa_load_spares(spa);
3861 spa->spa_spares.sav_sync = B_TRUE;
3862 }
3863
3864 if (nl2cache != 0) {
3865 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3866 ZPOOL_CONFIG_L2CACHE);
3867 spa_load_l2cache(spa);
3868 spa->spa_l2cache.sav_sync = B_TRUE;
3869 }
3870
3871 /*
3872 * We have to be careful when adding new vdevs to an existing pool.
3873 * If other threads start allocating from these vdevs before we
3874 * sync the config cache, and we lose power, then upon reboot we may
3875 * fail to open the pool because there are DVAs that the config cache
3876 * can't translate. Therefore, we first add the vdevs without
3877 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3878 * and then let spa_config_update() initialize the new metaslabs.
3879 *
3880 * spa_load() checks for added-but-not-initialized vdevs, so that
3881 * if we lose power at any point in this sequence, the remaining
3882 * steps will be completed the next time we load the pool.
3883 */
3884 (void) spa_vdev_exit(spa, vd, txg, 0);
3885
3886 mutex_enter(&spa_namespace_lock);
3887 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3888 mutex_exit(&spa_namespace_lock);
3889
3890 return (0);
3891}
3892
3893/*
3894 * Attach a device to a mirror. The arguments are the path to any device
3895 * in the mirror, and the nvroot for the new device. If the path specifies
3896 * a device that is not mirrored, we automatically insert the mirror vdev.
3897 *
3898 * If 'replacing' is specified, the new device is intended to replace the
3899 * existing device; in this case the two devices are made into their own
3900 * mirror using the 'replacing' vdev, which is functionally identical to
3901 * the mirror vdev (it actually reuses all the same ops) but has a few
3902 * extra rules: you can't attach to it after it's been created, and upon
3903 * completion of resilvering, the first disk (the one being replaced)
3904 * is automatically detached.
3905 */
3906int
3907spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3908{
3909 uint64_t txg, dtl_max_txg;
3910 vdev_t *rvd = spa->spa_root_vdev;
3911 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3912 vdev_ops_t *pvops;
3913 char *oldvdpath, *newvdpath;
3914 int newvd_isspare;
3915 int error;
3916
3917 ASSERT(spa_writeable(spa));
3918
3919 txg = spa_vdev_enter(spa);
3920
3921 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3922
3923 if (oldvd == NULL)
3924 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3925
3926 if (!oldvd->vdev_ops->vdev_op_leaf)
3927 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3928
3929 pvd = oldvd->vdev_parent;
3930
3931 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
| 2935 }
2936}
2937
2938/*
2939 * Pool Creation
2940 */
2941int
2942spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2943 const char *history_str, nvlist_t *zplprops)
2944{
2945 spa_t *spa;
2946 char *altroot = NULL;
2947 vdev_t *rvd;
2948 dsl_pool_t *dp;
2949 dmu_tx_t *tx;
2950 int error = 0;
2951 uint64_t txg = TXG_INITIAL;
2952 nvlist_t **spares, **l2cache;
2953 uint_t nspares, nl2cache;
2954 uint64_t version, obj;
2955
2956 /*
2957 * If this pool already exists, return failure.
2958 */
2959 mutex_enter(&spa_namespace_lock);
2960 if (spa_lookup(pool) != NULL) {
2961 mutex_exit(&spa_namespace_lock);
2962 return (EEXIST);
2963 }
2964
2965 /*
2966 * Allocate a new spa_t structure.
2967 */
2968 (void) nvlist_lookup_string(props,
2969 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2970 spa = spa_add(pool, NULL, altroot);
2971 spa_activate(spa, spa_mode_global);
2972
2973 if (props && (error = spa_prop_validate(spa, props))) {
2974 spa_deactivate(spa);
2975 spa_remove(spa);
2976 mutex_exit(&spa_namespace_lock);
2977 return (error);
2978 }
2979
2980 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2981 &version) != 0)
2982 version = SPA_VERSION;
2983 ASSERT(version <= SPA_VERSION);
2984
2985 spa->spa_first_txg = txg;
2986 spa->spa_uberblock.ub_txg = txg - 1;
2987 spa->spa_uberblock.ub_version = version;
2988 spa->spa_ubsync = spa->spa_uberblock;
2989
2990 /*
2991 * Create "The Godfather" zio to hold all async IOs
2992 */
2993 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2994 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2995
2996 /*
2997 * Create the root vdev.
2998 */
2999 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3000
3001 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3002
3003 ASSERT(error != 0 || rvd != NULL);
3004 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3005
3006 if (error == 0 && !zfs_allocatable_devs(nvroot))
3007 error = EINVAL;
3008
3009 if (error == 0 &&
3010 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3011 (error = spa_validate_aux(spa, nvroot, txg,
3012 VDEV_ALLOC_ADD)) == 0) {
3013 for (int c = 0; c < rvd->vdev_children; c++) {
3014 vdev_metaslab_set_size(rvd->vdev_child[c]);
3015 vdev_expand(rvd->vdev_child[c], txg);
3016 }
3017 }
3018
3019 spa_config_exit(spa, SCL_ALL, FTAG);
3020
3021 if (error != 0) {
3022 spa_unload(spa);
3023 spa_deactivate(spa);
3024 spa_remove(spa);
3025 mutex_exit(&spa_namespace_lock);
3026 return (error);
3027 }
3028
3029 /*
3030 * Get the list of spares, if specified.
3031 */
3032 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3033 &spares, &nspares) == 0) {
3034 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3035 KM_SLEEP) == 0);
3036 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3037 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3038 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3039 spa_load_spares(spa);
3040 spa_config_exit(spa, SCL_ALL, FTAG);
3041 spa->spa_spares.sav_sync = B_TRUE;
3042 }
3043
3044 /*
3045 * Get the list of level 2 cache devices, if specified.
3046 */
3047 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3048 &l2cache, &nl2cache) == 0) {
3049 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3050 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3051 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3052 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3053 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3054 spa_load_l2cache(spa);
3055 spa_config_exit(spa, SCL_ALL, FTAG);
3056 spa->spa_l2cache.sav_sync = B_TRUE;
3057 }
3058
3059 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3060 spa->spa_meta_objset = dp->dp_meta_objset;
3061
3062 /*
3063 * Create DDTs (dedup tables).
3064 */
3065 ddt_create(spa);
3066
3067 spa_update_dspace(spa);
3068
3069 tx = dmu_tx_create_assigned(dp, txg);
3070
3071 /*
3072 * Create the pool config object.
3073 */
3074 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3075 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3076 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3077
3078 if (zap_add(spa->spa_meta_objset,
3079 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3080 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3081 cmn_err(CE_PANIC, "failed to add pool config");
3082 }
3083
3084 if (zap_add(spa->spa_meta_objset,
3085 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3086 sizeof (uint64_t), 1, &version, tx) != 0) {
3087 cmn_err(CE_PANIC, "failed to add pool version");
3088 }
3089
3090 /* Newly created pools with the right version are always deflated. */
3091 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3092 spa->spa_deflate = TRUE;
3093 if (zap_add(spa->spa_meta_objset,
3094 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3095 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3096 cmn_err(CE_PANIC, "failed to add deflate");
3097 }
3098 }
3099
3100 /*
3101 * Create the deferred-free bpobj. Turn off compression
3102 * because sync-to-convergence takes longer if the blocksize
3103 * keeps changing.
3104 */
3105 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3106 dmu_object_set_compress(spa->spa_meta_objset, obj,
3107 ZIO_COMPRESS_OFF, tx);
3108 if (zap_add(spa->spa_meta_objset,
3109 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3110 sizeof (uint64_t), 1, &obj, tx) != 0) {
3111 cmn_err(CE_PANIC, "failed to add bpobj");
3112 }
3113 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3114 spa->spa_meta_objset, obj));
3115
3116 /*
3117 * Create the pool's history object.
3118 */
3119 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3120 spa_history_create_obj(spa, tx);
3121
3122 /*
3123 * Set pool properties.
3124 */
3125 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3126 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3127 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3128 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3129
3130 if (props != NULL) {
3131 spa_configfile_set(spa, props, B_FALSE);
3132 spa_sync_props(spa, props, tx);
3133 }
3134
3135 dmu_tx_commit(tx);
3136
3137 spa->spa_sync_on = B_TRUE;
3138 txg_sync_start(spa->spa_dsl_pool);
3139
3140 /*
3141 * We explicitly wait for the first transaction to complete so that our
3142 * bean counters are appropriately updated.
3143 */
3144 txg_wait_synced(spa->spa_dsl_pool, txg);
3145
3146 spa_config_sync(spa, B_FALSE, B_TRUE);
3147
3148 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3149 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3150 spa_history_log_version(spa, LOG_POOL_CREATE);
3151
3152 spa->spa_minref = refcount_count(&spa->spa_refcount);
3153
3154 mutex_exit(&spa_namespace_lock);
3155
3156 return (0);
3157}
3158
3159#if defined(sun)
3160#ifdef _KERNEL
3161/*
3162 * Get the root pool information from the root disk, then import the root pool
3163 * during the system boot up time.
3164 */
3165extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3166
3167static nvlist_t *
3168spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3169{
3170 nvlist_t *config;
3171 nvlist_t *nvtop, *nvroot;
3172 uint64_t pgid;
3173
3174 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3175 return (NULL);
3176
3177 /*
3178 * Add this top-level vdev to the child array.
3179 */
3180 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3181 &nvtop) == 0);
3182 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3183 &pgid) == 0);
3184 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3185
3186 /*
3187 * Put this pool's top-level vdevs into a root vdev.
3188 */
3189 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3190 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3191 VDEV_TYPE_ROOT) == 0);
3192 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3193 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3194 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3195 &nvtop, 1) == 0);
3196
3197 /*
3198 * Replace the existing vdev_tree with the new root vdev in
3199 * this pool's configuration (remove the old, add the new).
3200 */
3201 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3202 nvlist_free(nvroot);
3203 return (config);
3204}
3205
3206/*
3207 * Walk the vdev tree and see if we can find a device with "better"
3208 * configuration. A configuration is "better" if the label on that
3209 * device has a more recent txg.
3210 */
3211static void
3212spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3213{
3214 for (int c = 0; c < vd->vdev_children; c++)
3215 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3216
3217 if (vd->vdev_ops->vdev_op_leaf) {
3218 nvlist_t *label;
3219 uint64_t label_txg;
3220
3221 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3222 &label) != 0)
3223 return;
3224
3225 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3226 &label_txg) == 0);
3227
3228 /*
3229 * Do we have a better boot device?
3230 */
3231 if (label_txg > *txg) {
3232 *txg = label_txg;
3233 *avd = vd;
3234 }
3235 nvlist_free(label);
3236 }
3237}
3238
3239/*
3240 * Import a root pool.
3241 *
3242 * For x86. devpath_list will consist of devid and/or physpath name of
3243 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3244 * The GRUB "findroot" command will return the vdev we should boot.
3245 *
3246 * For Sparc, devpath_list consists the physpath name of the booting device
3247 * no matter the rootpool is a single device pool or a mirrored pool.
3248 * e.g.
3249 * "/pci@1f,0/ide@d/disk@0,0:a"
3250 */
3251int
3252spa_import_rootpool(char *devpath, char *devid)
3253{
3254 spa_t *spa;
3255 vdev_t *rvd, *bvd, *avd = NULL;
3256 nvlist_t *config, *nvtop;
3257 uint64_t guid, txg;
3258 char *pname;
3259 int error;
3260
3261 /*
3262 * Read the label from the boot device and generate a configuration.
3263 */
3264 config = spa_generate_rootconf(devpath, devid, &guid);
3265#if defined(_OBP) && defined(_KERNEL)
3266 if (config == NULL) {
3267 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3268 /* iscsi boot */
3269 get_iscsi_bootpath_phy(devpath);
3270 config = spa_generate_rootconf(devpath, devid, &guid);
3271 }
3272 }
3273#endif
3274 if (config == NULL) {
3275 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3276 devpath);
3277 return (EIO);
3278 }
3279
3280 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3281 &pname) == 0);
3282 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3283
3284 mutex_enter(&spa_namespace_lock);
3285 if ((spa = spa_lookup(pname)) != NULL) {
3286 /*
3287 * Remove the existing root pool from the namespace so that we
3288 * can replace it with the correct config we just read in.
3289 */
3290 spa_remove(spa);
3291 }
3292
3293 spa = spa_add(pname, config, NULL);
3294 spa->spa_is_root = B_TRUE;
3295 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3296
3297 /*
3298 * Build up a vdev tree based on the boot device's label config.
3299 */
3300 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3301 &nvtop) == 0);
3302 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3303 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3304 VDEV_ALLOC_ROOTPOOL);
3305 spa_config_exit(spa, SCL_ALL, FTAG);
3306 if (error) {
3307 mutex_exit(&spa_namespace_lock);
3308 nvlist_free(config);
3309 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3310 pname);
3311 return (error);
3312 }
3313
3314 /*
3315 * Get the boot vdev.
3316 */
3317 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3318 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3319 (u_longlong_t)guid);
3320 error = ENOENT;
3321 goto out;
3322 }
3323
3324 /*
3325 * Determine if there is a better boot device.
3326 */
3327 avd = bvd;
3328 spa_alt_rootvdev(rvd, &avd, &txg);
3329 if (avd != bvd) {
3330 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3331 "try booting from '%s'", avd->vdev_path);
3332 error = EINVAL;
3333 goto out;
3334 }
3335
3336 /*
3337 * If the boot device is part of a spare vdev then ensure that
3338 * we're booting off the active spare.
3339 */
3340 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3341 !bvd->vdev_isspare) {
3342 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3343 "try booting from '%s'",
3344 bvd->vdev_parent->
3345 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3346 error = EINVAL;
3347 goto out;
3348 }
3349
3350 error = 0;
3351 spa_history_log_version(spa, LOG_POOL_IMPORT);
3352out:
3353 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3354 vdev_free(rvd);
3355 spa_config_exit(spa, SCL_ALL, FTAG);
3356 mutex_exit(&spa_namespace_lock);
3357
3358 nvlist_free(config);
3359 return (error);
3360}
3361
3362#endif
3363#endif /* sun */
3364
3365/*
3366 * Import a non-root pool into the system.
3367 */
3368int
3369spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3370{
3371 spa_t *spa;
3372 char *altroot = NULL;
3373 spa_load_state_t state = SPA_LOAD_IMPORT;
3374 zpool_rewind_policy_t policy;
3375 uint64_t mode = spa_mode_global;
3376 uint64_t readonly = B_FALSE;
3377 int error;
3378 nvlist_t *nvroot;
3379 nvlist_t **spares, **l2cache;
3380 uint_t nspares, nl2cache;
3381
3382 /*
3383 * If a pool with this name exists, return failure.
3384 */
3385 mutex_enter(&spa_namespace_lock);
3386 if (spa_lookup(pool) != NULL) {
3387 mutex_exit(&spa_namespace_lock);
3388 return (EEXIST);
3389 }
3390
3391 /*
3392 * Create and initialize the spa structure.
3393 */
3394 (void) nvlist_lookup_string(props,
3395 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3396 (void) nvlist_lookup_uint64(props,
3397 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3398 if (readonly)
3399 mode = FREAD;
3400 spa = spa_add(pool, config, altroot);
3401 spa->spa_import_flags = flags;
3402
3403 /*
3404 * Verbatim import - Take a pool and insert it into the namespace
3405 * as if it had been loaded at boot.
3406 */
3407 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3408 if (props != NULL)
3409 spa_configfile_set(spa, props, B_FALSE);
3410
3411 spa_config_sync(spa, B_FALSE, B_TRUE);
3412
3413 mutex_exit(&spa_namespace_lock);
3414 spa_history_log_version(spa, LOG_POOL_IMPORT);
3415
3416 return (0);
3417 }
3418
3419 spa_activate(spa, mode);
3420
3421 /*
3422 * Don't start async tasks until we know everything is healthy.
3423 */
3424 spa_async_suspend(spa);
3425
3426 zpool_get_rewind_policy(config, &policy);
3427 if (policy.zrp_request & ZPOOL_DO_REWIND)
3428 state = SPA_LOAD_RECOVER;
3429
3430 /*
3431 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3432 * because the user-supplied config is actually the one to trust when
3433 * doing an import.
3434 */
3435 if (state != SPA_LOAD_RECOVER)
3436 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3437
3438 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3439 policy.zrp_request);
3440
3441 /*
3442 * Propagate anything learned while loading the pool and pass it
3443 * back to caller (i.e. rewind info, missing devices, etc).
3444 */
3445 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3446 spa->spa_load_info) == 0);
3447
3448 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3449 /*
3450 * Toss any existing sparelist, as it doesn't have any validity
3451 * anymore, and conflicts with spa_has_spare().
3452 */
3453 if (spa->spa_spares.sav_config) {
3454 nvlist_free(spa->spa_spares.sav_config);
3455 spa->spa_spares.sav_config = NULL;
3456 spa_load_spares(spa);
3457 }
3458 if (spa->spa_l2cache.sav_config) {
3459 nvlist_free(spa->spa_l2cache.sav_config);
3460 spa->spa_l2cache.sav_config = NULL;
3461 spa_load_l2cache(spa);
3462 }
3463
3464 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3465 &nvroot) == 0);
3466 if (error == 0)
3467 error = spa_validate_aux(spa, nvroot, -1ULL,
3468 VDEV_ALLOC_SPARE);
3469 if (error == 0)
3470 error = spa_validate_aux(spa, nvroot, -1ULL,
3471 VDEV_ALLOC_L2CACHE);
3472 spa_config_exit(spa, SCL_ALL, FTAG);
3473
3474 if (props != NULL)
3475 spa_configfile_set(spa, props, B_FALSE);
3476
3477 if (error != 0 || (props && spa_writeable(spa) &&
3478 (error = spa_prop_set(spa, props)))) {
3479 spa_unload(spa);
3480 spa_deactivate(spa);
3481 spa_remove(spa);
3482 mutex_exit(&spa_namespace_lock);
3483 return (error);
3484 }
3485
3486 spa_async_resume(spa);
3487
3488 /*
3489 * Override any spares and level 2 cache devices as specified by
3490 * the user, as these may have correct device names/devids, etc.
3491 */
3492 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3493 &spares, &nspares) == 0) {
3494 if (spa->spa_spares.sav_config)
3495 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3496 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3497 else
3498 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3499 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3500 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3501 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3502 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3503 spa_load_spares(spa);
3504 spa_config_exit(spa, SCL_ALL, FTAG);
3505 spa->spa_spares.sav_sync = B_TRUE;
3506 }
3507 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3508 &l2cache, &nl2cache) == 0) {
3509 if (spa->spa_l2cache.sav_config)
3510 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3511 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3512 else
3513 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3514 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3515 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3516 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3517 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3518 spa_load_l2cache(spa);
3519 spa_config_exit(spa, SCL_ALL, FTAG);
3520 spa->spa_l2cache.sav_sync = B_TRUE;
3521 }
3522
3523 /*
3524 * Check for any removed devices.
3525 */
3526 if (spa->spa_autoreplace) {
3527 spa_aux_check_removed(&spa->spa_spares);
3528 spa_aux_check_removed(&spa->spa_l2cache);
3529 }
3530
3531 if (spa_writeable(spa)) {
3532 /*
3533 * Update the config cache to include the newly-imported pool.
3534 */
3535 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3536 }
3537
3538 /*
3539 * It's possible that the pool was expanded while it was exported.
3540 * We kick off an async task to handle this for us.
3541 */
3542 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3543
3544 mutex_exit(&spa_namespace_lock);
3545 spa_history_log_version(spa, LOG_POOL_IMPORT);
3546
3547#ifdef __FreeBSD__
3548#ifdef _KERNEL
3549 zvol_create_minors(pool);
3550#endif
3551#endif
3552 return (0);
3553}
3554
3555nvlist_t *
3556spa_tryimport(nvlist_t *tryconfig)
3557{
3558 nvlist_t *config = NULL;
3559 char *poolname;
3560 spa_t *spa;
3561 uint64_t state;
3562 int error;
3563
3564 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3565 return (NULL);
3566
3567 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3568 return (NULL);
3569
3570 /*
3571 * Create and initialize the spa structure.
3572 */
3573 mutex_enter(&spa_namespace_lock);
3574 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3575 spa_activate(spa, FREAD);
3576
3577 /*
3578 * Pass off the heavy lifting to spa_load().
3579 * Pass TRUE for mosconfig because the user-supplied config
3580 * is actually the one to trust when doing an import.
3581 */
3582 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3583
3584 /*
3585 * If 'tryconfig' was at least parsable, return the current config.
3586 */
3587 if (spa->spa_root_vdev != NULL) {
3588 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3589 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3590 poolname) == 0);
3591 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3592 state) == 0);
3593 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3594 spa->spa_uberblock.ub_timestamp) == 0);
3595
3596 /*
3597 * If the bootfs property exists on this pool then we
3598 * copy it out so that external consumers can tell which
3599 * pools are bootable.
3600 */
3601 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3602 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3603
3604 /*
3605 * We have to play games with the name since the
3606 * pool was opened as TRYIMPORT_NAME.
3607 */
3608 if (dsl_dsobj_to_dsname(spa_name(spa),
3609 spa->spa_bootfs, tmpname) == 0) {
3610 char *cp;
3611 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3612
3613 cp = strchr(tmpname, '/');
3614 if (cp == NULL) {
3615 (void) strlcpy(dsname, tmpname,
3616 MAXPATHLEN);
3617 } else {
3618 (void) snprintf(dsname, MAXPATHLEN,
3619 "%s/%s", poolname, ++cp);
3620 }
3621 VERIFY(nvlist_add_string(config,
3622 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3623 kmem_free(dsname, MAXPATHLEN);
3624 }
3625 kmem_free(tmpname, MAXPATHLEN);
3626 }
3627
3628 /*
3629 * Add the list of hot spares and level 2 cache devices.
3630 */
3631 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3632 spa_add_spares(spa, config);
3633 spa_add_l2cache(spa, config);
3634 spa_config_exit(spa, SCL_CONFIG, FTAG);
3635 }
3636
3637 spa_unload(spa);
3638 spa_deactivate(spa);
3639 spa_remove(spa);
3640 mutex_exit(&spa_namespace_lock);
3641
3642 return (config);
3643}
3644
3645/*
3646 * Pool export/destroy
3647 *
3648 * The act of destroying or exporting a pool is very simple. We make sure there
3649 * is no more pending I/O and any references to the pool are gone. Then, we
3650 * update the pool state and sync all the labels to disk, removing the
3651 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3652 * we don't sync the labels or remove the configuration cache.
3653 */
3654static int
3655spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3656 boolean_t force, boolean_t hardforce)
3657{
3658 spa_t *spa;
3659
3660 if (oldconfig)
3661 *oldconfig = NULL;
3662
3663 if (!(spa_mode_global & FWRITE))
3664 return (EROFS);
3665
3666 mutex_enter(&spa_namespace_lock);
3667 if ((spa = spa_lookup(pool)) == NULL) {
3668 mutex_exit(&spa_namespace_lock);
3669 return (ENOENT);
3670 }
3671
3672 /*
3673 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3674 * reacquire the namespace lock, and see if we can export.
3675 */
3676 spa_open_ref(spa, FTAG);
3677 mutex_exit(&spa_namespace_lock);
3678 spa_async_suspend(spa);
3679 mutex_enter(&spa_namespace_lock);
3680 spa_close(spa, FTAG);
3681
3682 /*
3683 * The pool will be in core if it's openable,
3684 * in which case we can modify its state.
3685 */
3686 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3687 /*
3688 * Objsets may be open only because they're dirty, so we
3689 * have to force it to sync before checking spa_refcnt.
3690 */
3691 txg_wait_synced(spa->spa_dsl_pool, 0);
3692
3693 /*
3694 * A pool cannot be exported or destroyed if there are active
3695 * references. If we are resetting a pool, allow references by
3696 * fault injection handlers.
3697 */
3698 if (!spa_refcount_zero(spa) ||
3699 (spa->spa_inject_ref != 0 &&
3700 new_state != POOL_STATE_UNINITIALIZED)) {
3701 spa_async_resume(spa);
3702 mutex_exit(&spa_namespace_lock);
3703 return (EBUSY);
3704 }
3705
3706 /*
3707 * A pool cannot be exported if it has an active shared spare.
3708 * This is to prevent other pools stealing the active spare
3709 * from an exported pool. At user's own will, such pool can
3710 * be forcedly exported.
3711 */
3712 if (!force && new_state == POOL_STATE_EXPORTED &&
3713 spa_has_active_shared_spare(spa)) {
3714 spa_async_resume(spa);
3715 mutex_exit(&spa_namespace_lock);
3716 return (EXDEV);
3717 }
3718
3719 /*
3720 * We want this to be reflected on every label,
3721 * so mark them all dirty. spa_unload() will do the
3722 * final sync that pushes these changes out.
3723 */
3724 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3725 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3726 spa->spa_state = new_state;
3727 spa->spa_final_txg = spa_last_synced_txg(spa) +
3728 TXG_DEFER_SIZE + 1;
3729 vdev_config_dirty(spa->spa_root_vdev);
3730 spa_config_exit(spa, SCL_ALL, FTAG);
3731 }
3732 }
3733
3734 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3735
3736 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3737 spa_unload(spa);
3738 spa_deactivate(spa);
3739 }
3740
3741 if (oldconfig && spa->spa_config)
3742 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3743
3744 if (new_state != POOL_STATE_UNINITIALIZED) {
3745 if (!hardforce)
3746 spa_config_sync(spa, B_TRUE, B_TRUE);
3747 spa_remove(spa);
3748 }
3749 mutex_exit(&spa_namespace_lock);
3750
3751 return (0);
3752}
3753
3754/*
3755 * Destroy a storage pool.
3756 */
3757int
3758spa_destroy(char *pool)
3759{
3760 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3761 B_FALSE, B_FALSE));
3762}
3763
3764/*
3765 * Export a storage pool.
3766 */
3767int
3768spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3769 boolean_t hardforce)
3770{
3771 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3772 force, hardforce));
3773}
3774
3775/*
3776 * Similar to spa_export(), this unloads the spa_t without actually removing it
3777 * from the namespace in any way.
3778 */
3779int
3780spa_reset(char *pool)
3781{
3782 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3783 B_FALSE, B_FALSE));
3784}
3785
3786/*
3787 * ==========================================================================
3788 * Device manipulation
3789 * ==========================================================================
3790 */
3791
3792/*
3793 * Add a device to a storage pool.
3794 */
3795int
3796spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3797{
3798 uint64_t txg, id;
3799 int error;
3800 vdev_t *rvd = spa->spa_root_vdev;
3801 vdev_t *vd, *tvd;
3802 nvlist_t **spares, **l2cache;
3803 uint_t nspares, nl2cache;
3804
3805 ASSERT(spa_writeable(spa));
3806
3807 txg = spa_vdev_enter(spa);
3808
3809 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3810 VDEV_ALLOC_ADD)) != 0)
3811 return (spa_vdev_exit(spa, NULL, txg, error));
3812
3813 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3814
3815 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3816 &nspares) != 0)
3817 nspares = 0;
3818
3819 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3820 &nl2cache) != 0)
3821 nl2cache = 0;
3822
3823 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3824 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3825
3826 if (vd->vdev_children != 0 &&
3827 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3828 return (spa_vdev_exit(spa, vd, txg, error));
3829
3830 /*
3831 * We must validate the spares and l2cache devices after checking the
3832 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3833 */
3834 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3835 return (spa_vdev_exit(spa, vd, txg, error));
3836
3837 /*
3838 * Transfer each new top-level vdev from vd to rvd.
3839 */
3840 for (int c = 0; c < vd->vdev_children; c++) {
3841
3842 /*
3843 * Set the vdev id to the first hole, if one exists.
3844 */
3845 for (id = 0; id < rvd->vdev_children; id++) {
3846 if (rvd->vdev_child[id]->vdev_ishole) {
3847 vdev_free(rvd->vdev_child[id]);
3848 break;
3849 }
3850 }
3851 tvd = vd->vdev_child[c];
3852 vdev_remove_child(vd, tvd);
3853 tvd->vdev_id = id;
3854 vdev_add_child(rvd, tvd);
3855 vdev_config_dirty(tvd);
3856 }
3857
3858 if (nspares != 0) {
3859 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3860 ZPOOL_CONFIG_SPARES);
3861 spa_load_spares(spa);
3862 spa->spa_spares.sav_sync = B_TRUE;
3863 }
3864
3865 if (nl2cache != 0) {
3866 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3867 ZPOOL_CONFIG_L2CACHE);
3868 spa_load_l2cache(spa);
3869 spa->spa_l2cache.sav_sync = B_TRUE;
3870 }
3871
3872 /*
3873 * We have to be careful when adding new vdevs to an existing pool.
3874 * If other threads start allocating from these vdevs before we
3875 * sync the config cache, and we lose power, then upon reboot we may
3876 * fail to open the pool because there are DVAs that the config cache
3877 * can't translate. Therefore, we first add the vdevs without
3878 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3879 * and then let spa_config_update() initialize the new metaslabs.
3880 *
3881 * spa_load() checks for added-but-not-initialized vdevs, so that
3882 * if we lose power at any point in this sequence, the remaining
3883 * steps will be completed the next time we load the pool.
3884 */
3885 (void) spa_vdev_exit(spa, vd, txg, 0);
3886
3887 mutex_enter(&spa_namespace_lock);
3888 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3889 mutex_exit(&spa_namespace_lock);
3890
3891 return (0);
3892}
3893
3894/*
3895 * Attach a device to a mirror. The arguments are the path to any device
3896 * in the mirror, and the nvroot for the new device. If the path specifies
3897 * a device that is not mirrored, we automatically insert the mirror vdev.
3898 *
3899 * If 'replacing' is specified, the new device is intended to replace the
3900 * existing device; in this case the two devices are made into their own
3901 * mirror using the 'replacing' vdev, which is functionally identical to
3902 * the mirror vdev (it actually reuses all the same ops) but has a few
3903 * extra rules: you can't attach to it after it's been created, and upon
3904 * completion of resilvering, the first disk (the one being replaced)
3905 * is automatically detached.
3906 */
3907int
3908spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3909{
3910 uint64_t txg, dtl_max_txg;
3911 vdev_t *rvd = spa->spa_root_vdev;
3912 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3913 vdev_ops_t *pvops;
3914 char *oldvdpath, *newvdpath;
3915 int newvd_isspare;
3916 int error;
3917
3918 ASSERT(spa_writeable(spa));
3919
3920 txg = spa_vdev_enter(spa);
3921
3922 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3923
3924 if (oldvd == NULL)
3925 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3926
3927 if (!oldvd->vdev_ops->vdev_op_leaf)
3928 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3929
3930 pvd = oldvd->vdev_parent;
3931
3932 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
|
3932 VDEV_ALLOC_ADD)) != 0)
| 3933 VDEV_ALLOC_ATTACH)) != 0)
|
3933 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3934
3935 if (newrootvd->vdev_children != 1)
3936 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3937
3938 newvd = newrootvd->vdev_child[0];
3939
3940 if (!newvd->vdev_ops->vdev_op_leaf)
3941 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3942
3943 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3944 return (spa_vdev_exit(spa, newrootvd, txg, error));
3945
3946 /*
3947 * Spares can't replace logs
3948 */
3949 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3950 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3951
3952 if (!replacing) {
3953 /*
3954 * For attach, the only allowable parent is a mirror or the root
3955 * vdev.
3956 */
3957 if (pvd->vdev_ops != &vdev_mirror_ops &&
3958 pvd->vdev_ops != &vdev_root_ops)
3959 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3960
3961 pvops = &vdev_mirror_ops;
3962 } else {
3963 /*
3964 * Active hot spares can only be replaced by inactive hot
3965 * spares.
3966 */
3967 if (pvd->vdev_ops == &vdev_spare_ops &&
3968 oldvd->vdev_isspare &&
3969 !spa_has_spare(spa, newvd->vdev_guid))
3970 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3971
3972 /*
3973 * If the source is a hot spare, and the parent isn't already a
3974 * spare, then we want to create a new hot spare. Otherwise, we
3975 * want to create a replacing vdev. The user is not allowed to
3976 * attach to a spared vdev child unless the 'isspare' state is
3977 * the same (spare replaces spare, non-spare replaces
3978 * non-spare).
3979 */
3980 if (pvd->vdev_ops == &vdev_replacing_ops &&
3981 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3982 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3983 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3984 newvd->vdev_isspare != oldvd->vdev_isspare) {
3985 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3986 }
3987
3988 if (newvd->vdev_isspare)
3989 pvops = &vdev_spare_ops;
3990 else
3991 pvops = &vdev_replacing_ops;
3992 }
3993
3994 /*
3995 * Make sure the new device is big enough.
3996 */
3997 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3998 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3999
4000 /*
4001 * The new device cannot have a higher alignment requirement
4002 * than the top-level vdev.
4003 */
4004 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4005 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4006
4007 /*
4008 * If this is an in-place replacement, update oldvd's path and devid
4009 * to make it distinguishable from newvd, and unopenable from now on.
4010 */
4011 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4012 spa_strfree(oldvd->vdev_path);
4013 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4014 KM_SLEEP);
4015 (void) sprintf(oldvd->vdev_path, "%s/%s",
4016 newvd->vdev_path, "old");
4017 if (oldvd->vdev_devid != NULL) {
4018 spa_strfree(oldvd->vdev_devid);
4019 oldvd->vdev_devid = NULL;
4020 }
4021 }
4022
4023 /* mark the device being resilvered */
4024 newvd->vdev_resilvering = B_TRUE;
4025
4026 /*
4027 * If the parent is not a mirror, or if we're replacing, insert the new
4028 * mirror/replacing/spare vdev above oldvd.
4029 */
4030 if (pvd->vdev_ops != pvops)
4031 pvd = vdev_add_parent(oldvd, pvops);
4032
4033 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4034 ASSERT(pvd->vdev_ops == pvops);
4035 ASSERT(oldvd->vdev_parent == pvd);
4036
4037 /*
4038 * Extract the new device from its root and add it to pvd.
4039 */
4040 vdev_remove_child(newrootvd, newvd);
4041 newvd->vdev_id = pvd->vdev_children;
4042 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4043 vdev_add_child(pvd, newvd);
4044
4045 tvd = newvd->vdev_top;
4046 ASSERT(pvd->vdev_top == tvd);
4047 ASSERT(tvd->vdev_parent == rvd);
4048
4049 vdev_config_dirty(tvd);
4050
4051 /*
4052 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4053 * for any dmu_sync-ed blocks. It will propagate upward when
4054 * spa_vdev_exit() calls vdev_dtl_reassess().
4055 */
4056 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4057
4058 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4059 dtl_max_txg - TXG_INITIAL);
4060
4061 if (newvd->vdev_isspare) {
4062 spa_spare_activate(newvd);
4063 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4064 }
4065
4066 oldvdpath = spa_strdup(oldvd->vdev_path);
4067 newvdpath = spa_strdup(newvd->vdev_path);
4068 newvd_isspare = newvd->vdev_isspare;
4069
4070 /*
4071 * Mark newvd's DTL dirty in this txg.
4072 */
4073 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4074
4075 /*
4076 * Restart the resilver
4077 */
4078 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4079
4080 /*
4081 * Commit the config
4082 */
4083 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4084
4085 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4086 "%s vdev=%s %s vdev=%s",
4087 replacing && newvd_isspare ? "spare in" :
4088 replacing ? "replace" : "attach", newvdpath,
4089 replacing ? "for" : "to", oldvdpath);
4090
4091 spa_strfree(oldvdpath);
4092 spa_strfree(newvdpath);
4093
4094 if (spa->spa_bootfs)
4095 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4096
4097 return (0);
4098}
4099
4100/*
4101 * Detach a device from a mirror or replacing vdev.
4102 * If 'replace_done' is specified, only detach if the parent
4103 * is a replacing vdev.
4104 */
4105int
4106spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4107{
4108 uint64_t txg;
4109 int error;
4110 vdev_t *rvd = spa->spa_root_vdev;
4111 vdev_t *vd, *pvd, *cvd, *tvd;
4112 boolean_t unspare = B_FALSE;
4113 uint64_t unspare_guid;
4114 char *vdpath;
4115
4116 ASSERT(spa_writeable(spa));
4117
4118 txg = spa_vdev_enter(spa);
4119
4120 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4121
4122 if (vd == NULL)
4123 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4124
4125 if (!vd->vdev_ops->vdev_op_leaf)
4126 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4127
4128 pvd = vd->vdev_parent;
4129
4130 /*
4131 * If the parent/child relationship is not as expected, don't do it.
4132 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4133 * vdev that's replacing B with C. The user's intent in replacing
4134 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4135 * the replace by detaching C, the expected behavior is to end up
4136 * M(A,B). But suppose that right after deciding to detach C,
4137 * the replacement of B completes. We would have M(A,C), and then
4138 * ask to detach C, which would leave us with just A -- not what
4139 * the user wanted. To prevent this, we make sure that the
4140 * parent/child relationship hasn't changed -- in this example,
4141 * that C's parent is still the replacing vdev R.
4142 */
4143 if (pvd->vdev_guid != pguid && pguid != 0)
4144 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4145
4146 /*
4147 * Only 'replacing' or 'spare' vdevs can be replaced.
4148 */
4149 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4150 pvd->vdev_ops != &vdev_spare_ops)
4151 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4152
4153 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4154 spa_version(spa) >= SPA_VERSION_SPARES);
4155
4156 /*
4157 * Only mirror, replacing, and spare vdevs support detach.
4158 */
4159 if (pvd->vdev_ops != &vdev_replacing_ops &&
4160 pvd->vdev_ops != &vdev_mirror_ops &&
4161 pvd->vdev_ops != &vdev_spare_ops)
4162 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4163
4164 /*
4165 * If this device has the only valid copy of some data,
4166 * we cannot safely detach it.
4167 */
4168 if (vdev_dtl_required(vd))
4169 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4170
4171 ASSERT(pvd->vdev_children >= 2);
4172
4173 /*
4174 * If we are detaching the second disk from a replacing vdev, then
4175 * check to see if we changed the original vdev's path to have "/old"
4176 * at the end in spa_vdev_attach(). If so, undo that change now.
4177 */
4178 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4179 vd->vdev_path != NULL) {
4180 size_t len = strlen(vd->vdev_path);
4181
4182 for (int c = 0; c < pvd->vdev_children; c++) {
4183 cvd = pvd->vdev_child[c];
4184
4185 if (cvd == vd || cvd->vdev_path == NULL)
4186 continue;
4187
4188 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4189 strcmp(cvd->vdev_path + len, "/old") == 0) {
4190 spa_strfree(cvd->vdev_path);
4191 cvd->vdev_path = spa_strdup(vd->vdev_path);
4192 break;
4193 }
4194 }
4195 }
4196
4197 /*
4198 * If we are detaching the original disk from a spare, then it implies
4199 * that the spare should become a real disk, and be removed from the
4200 * active spare list for the pool.
4201 */
4202 if (pvd->vdev_ops == &vdev_spare_ops &&
4203 vd->vdev_id == 0 &&
4204 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4205 unspare = B_TRUE;
4206
4207 /*
4208 * Erase the disk labels so the disk can be used for other things.
4209 * This must be done after all other error cases are handled,
4210 * but before we disembowel vd (so we can still do I/O to it).
4211 * But if we can't do it, don't treat the error as fatal --
4212 * it may be that the unwritability of the disk is the reason
4213 * it's being detached!
4214 */
4215 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4216
4217 /*
4218 * Remove vd from its parent and compact the parent's children.
4219 */
4220 vdev_remove_child(pvd, vd);
4221 vdev_compact_children(pvd);
4222
4223 /*
4224 * Remember one of the remaining children so we can get tvd below.
4225 */
4226 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4227
4228 /*
4229 * If we need to remove the remaining child from the list of hot spares,
4230 * do it now, marking the vdev as no longer a spare in the process.
4231 * We must do this before vdev_remove_parent(), because that can
4232 * change the GUID if it creates a new toplevel GUID. For a similar
4233 * reason, we must remove the spare now, in the same txg as the detach;
4234 * otherwise someone could attach a new sibling, change the GUID, and
4235 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4236 */
4237 if (unspare) {
4238 ASSERT(cvd->vdev_isspare);
4239 spa_spare_remove(cvd);
4240 unspare_guid = cvd->vdev_guid;
4241 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4242 cvd->vdev_unspare = B_TRUE;
4243 }
4244
4245 /*
4246 * If the parent mirror/replacing vdev only has one child,
4247 * the parent is no longer needed. Remove it from the tree.
4248 */
4249 if (pvd->vdev_children == 1) {
4250 if (pvd->vdev_ops == &vdev_spare_ops)
4251 cvd->vdev_unspare = B_FALSE;
4252 vdev_remove_parent(cvd);
4253 cvd->vdev_resilvering = B_FALSE;
4254 }
4255
4256
4257 /*
4258 * We don't set tvd until now because the parent we just removed
4259 * may have been the previous top-level vdev.
4260 */
4261 tvd = cvd->vdev_top;
4262 ASSERT(tvd->vdev_parent == rvd);
4263
4264 /*
4265 * Reevaluate the parent vdev state.
4266 */
4267 vdev_propagate_state(cvd);
4268
4269 /*
4270 * If the 'autoexpand' property is set on the pool then automatically
4271 * try to expand the size of the pool. For example if the device we
4272 * just detached was smaller than the others, it may be possible to
4273 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4274 * first so that we can obtain the updated sizes of the leaf vdevs.
4275 */
4276 if (spa->spa_autoexpand) {
4277 vdev_reopen(tvd);
4278 vdev_expand(tvd, txg);
4279 }
4280
4281 vdev_config_dirty(tvd);
4282
4283 /*
4284 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4285 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4286 * But first make sure we're not on any *other* txg's DTL list, to
4287 * prevent vd from being accessed after it's freed.
4288 */
4289 vdpath = spa_strdup(vd->vdev_path);
4290 for (int t = 0; t < TXG_SIZE; t++)
4291 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4292 vd->vdev_detached = B_TRUE;
4293 vdev_dirty(tvd, VDD_DTL, vd, txg);
4294
4295 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4296
4297 /* hang on to the spa before we release the lock */
4298 spa_open_ref(spa, FTAG);
4299
4300 error = spa_vdev_exit(spa, vd, txg, 0);
4301
4302 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4303 "vdev=%s", vdpath);
4304 spa_strfree(vdpath);
4305
4306 /*
4307 * If this was the removal of the original device in a hot spare vdev,
4308 * then we want to go through and remove the device from the hot spare
4309 * list of every other pool.
4310 */
4311 if (unspare) {
4312 spa_t *altspa = NULL;
4313
4314 mutex_enter(&spa_namespace_lock);
4315 while ((altspa = spa_next(altspa)) != NULL) {
4316 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4317 altspa == spa)
4318 continue;
4319
4320 spa_open_ref(altspa, FTAG);
4321 mutex_exit(&spa_namespace_lock);
4322 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4323 mutex_enter(&spa_namespace_lock);
4324 spa_close(altspa, FTAG);
4325 }
4326 mutex_exit(&spa_namespace_lock);
4327
4328 /* search the rest of the vdevs for spares to remove */
4329 spa_vdev_resilver_done(spa);
4330 }
4331
4332 /* all done with the spa; OK to release */
4333 mutex_enter(&spa_namespace_lock);
4334 spa_close(spa, FTAG);
4335 mutex_exit(&spa_namespace_lock);
4336
4337 return (error);
4338}
4339
4340/*
4341 * Split a set of devices from their mirrors, and create a new pool from them.
4342 */
4343int
4344spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4345 nvlist_t *props, boolean_t exp)
4346{
4347 int error = 0;
4348 uint64_t txg, *glist;
4349 spa_t *newspa;
4350 uint_t c, children, lastlog;
4351 nvlist_t **child, *nvl, *tmp;
4352 dmu_tx_t *tx;
4353 char *altroot = NULL;
4354 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4355 boolean_t activate_slog;
4356
4357 ASSERT(spa_writeable(spa));
4358
4359 txg = spa_vdev_enter(spa);
4360
4361 /* clear the log and flush everything up to now */
4362 activate_slog = spa_passivate_log(spa);
4363 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4364 error = spa_offline_log(spa);
4365 txg = spa_vdev_config_enter(spa);
4366
4367 if (activate_slog)
4368 spa_activate_log(spa);
4369
4370 if (error != 0)
4371 return (spa_vdev_exit(spa, NULL, txg, error));
4372
4373 /* check new spa name before going any further */
4374 if (spa_lookup(newname) != NULL)
4375 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4376
4377 /*
4378 * scan through all the children to ensure they're all mirrors
4379 */
4380 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4381 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4382 &children) != 0)
4383 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4384
4385 /* first, check to ensure we've got the right child count */
4386 rvd = spa->spa_root_vdev;
4387 lastlog = 0;
4388 for (c = 0; c < rvd->vdev_children; c++) {
4389 vdev_t *vd = rvd->vdev_child[c];
4390
4391 /* don't count the holes & logs as children */
4392 if (vd->vdev_islog || vd->vdev_ishole) {
4393 if (lastlog == 0)
4394 lastlog = c;
4395 continue;
4396 }
4397
4398 lastlog = 0;
4399 }
4400 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4401 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4402
4403 /* next, ensure no spare or cache devices are part of the split */
4404 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4405 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4406 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4407
4408 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4409 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4410
4411 /* then, loop over each vdev and validate it */
4412 for (c = 0; c < children; c++) {
4413 uint64_t is_hole = 0;
4414
4415 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4416 &is_hole);
4417
4418 if (is_hole != 0) {
4419 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4420 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4421 continue;
4422 } else {
4423 error = EINVAL;
4424 break;
4425 }
4426 }
4427
4428 /* which disk is going to be split? */
4429 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4430 &glist[c]) != 0) {
4431 error = EINVAL;
4432 break;
4433 }
4434
4435 /* look it up in the spa */
4436 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4437 if (vml[c] == NULL) {
4438 error = ENODEV;
4439 break;
4440 }
4441
4442 /* make sure there's nothing stopping the split */
4443 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4444 vml[c]->vdev_islog ||
4445 vml[c]->vdev_ishole ||
4446 vml[c]->vdev_isspare ||
4447 vml[c]->vdev_isl2cache ||
4448 !vdev_writeable(vml[c]) ||
4449 vml[c]->vdev_children != 0 ||
4450 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4451 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4452 error = EINVAL;
4453 break;
4454 }
4455
4456 if (vdev_dtl_required(vml[c])) {
4457 error = EBUSY;
4458 break;
4459 }
4460
4461 /* we need certain info from the top level */
4462 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4463 vml[c]->vdev_top->vdev_ms_array) == 0);
4464 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4465 vml[c]->vdev_top->vdev_ms_shift) == 0);
4466 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4467 vml[c]->vdev_top->vdev_asize) == 0);
4468 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4469 vml[c]->vdev_top->vdev_ashift) == 0);
4470 }
4471
4472 if (error != 0) {
4473 kmem_free(vml, children * sizeof (vdev_t *));
4474 kmem_free(glist, children * sizeof (uint64_t));
4475 return (spa_vdev_exit(spa, NULL, txg, error));
4476 }
4477
4478 /* stop writers from using the disks */
4479 for (c = 0; c < children; c++) {
4480 if (vml[c] != NULL)
4481 vml[c]->vdev_offline = B_TRUE;
4482 }
4483 vdev_reopen(spa->spa_root_vdev);
4484
4485 /*
4486 * Temporarily record the splitting vdevs in the spa config. This
4487 * will disappear once the config is regenerated.
4488 */
4489 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4490 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4491 glist, children) == 0);
4492 kmem_free(glist, children * sizeof (uint64_t));
4493
4494 mutex_enter(&spa->spa_props_lock);
4495 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4496 nvl) == 0);
4497 mutex_exit(&spa->spa_props_lock);
4498 spa->spa_config_splitting = nvl;
4499 vdev_config_dirty(spa->spa_root_vdev);
4500
4501 /* configure and create the new pool */
4502 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4503 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4504 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4505 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4506 spa_version(spa)) == 0);
4507 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4508 spa->spa_config_txg) == 0);
4509 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4510 spa_generate_guid(NULL)) == 0);
4511 (void) nvlist_lookup_string(props,
4512 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4513
4514 /* add the new pool to the namespace */
4515 newspa = spa_add(newname, config, altroot);
4516 newspa->spa_config_txg = spa->spa_config_txg;
4517 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4518
4519 /* release the spa config lock, retaining the namespace lock */
4520 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4521
4522 if (zio_injection_enabled)
4523 zio_handle_panic_injection(spa, FTAG, 1);
4524
4525 spa_activate(newspa, spa_mode_global);
4526 spa_async_suspend(newspa);
4527
4528#ifndef sun
4529 /* mark that we are creating new spa by splitting */
4530 newspa->spa_splitting_newspa = B_TRUE;
4531#endif
4532 /* create the new pool from the disks of the original pool */
4533 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4534#ifndef sun
4535 newspa->spa_splitting_newspa = B_FALSE;
4536#endif
4537 if (error)
4538 goto out;
4539
4540 /* if that worked, generate a real config for the new pool */
4541 if (newspa->spa_root_vdev != NULL) {
4542 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4543 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4544 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4545 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4546 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4547 B_TRUE));
4548 }
4549
4550 /* set the props */
4551 if (props != NULL) {
4552 spa_configfile_set(newspa, props, B_FALSE);
4553 error = spa_prop_set(newspa, props);
4554 if (error)
4555 goto out;
4556 }
4557
4558 /* flush everything */
4559 txg = spa_vdev_config_enter(newspa);
4560 vdev_config_dirty(newspa->spa_root_vdev);
4561 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4562
4563 if (zio_injection_enabled)
4564 zio_handle_panic_injection(spa, FTAG, 2);
4565
4566 spa_async_resume(newspa);
4567
4568 /* finally, update the original pool's config */
4569 txg = spa_vdev_config_enter(spa);
4570 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4571 error = dmu_tx_assign(tx, TXG_WAIT);
4572 if (error != 0)
4573 dmu_tx_abort(tx);
4574 for (c = 0; c < children; c++) {
4575 if (vml[c] != NULL) {
4576 vdev_split(vml[c]);
4577 if (error == 0)
4578 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4579 spa, tx, "vdev=%s",
4580 vml[c]->vdev_path);
4581 vdev_free(vml[c]);
4582 }
4583 }
4584 vdev_config_dirty(spa->spa_root_vdev);
4585 spa->spa_config_splitting = NULL;
4586 nvlist_free(nvl);
4587 if (error == 0)
4588 dmu_tx_commit(tx);
4589 (void) spa_vdev_exit(spa, NULL, txg, 0);
4590
4591 if (zio_injection_enabled)
4592 zio_handle_panic_injection(spa, FTAG, 3);
4593
4594 /* split is complete; log a history record */
4595 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4596 "split new pool %s from pool %s", newname, spa_name(spa));
4597
4598 kmem_free(vml, children * sizeof (vdev_t *));
4599
4600 /* if we're not going to mount the filesystems in userland, export */
4601 if (exp)
4602 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4603 B_FALSE, B_FALSE);
4604
4605 return (error);
4606
4607out:
4608 spa_unload(newspa);
4609 spa_deactivate(newspa);
4610 spa_remove(newspa);
4611
4612 txg = spa_vdev_config_enter(spa);
4613
4614 /* re-online all offlined disks */
4615 for (c = 0; c < children; c++) {
4616 if (vml[c] != NULL)
4617 vml[c]->vdev_offline = B_FALSE;
4618 }
4619 vdev_reopen(spa->spa_root_vdev);
4620
4621 nvlist_free(spa->spa_config_splitting);
4622 spa->spa_config_splitting = NULL;
4623 (void) spa_vdev_exit(spa, NULL, txg, error);
4624
4625 kmem_free(vml, children * sizeof (vdev_t *));
4626 return (error);
4627}
4628
4629static nvlist_t *
4630spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4631{
4632 for (int i = 0; i < count; i++) {
4633 uint64_t guid;
4634
4635 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4636 &guid) == 0);
4637
4638 if (guid == target_guid)
4639 return (nvpp[i]);
4640 }
4641
4642 return (NULL);
4643}
4644
4645static void
4646spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4647 nvlist_t *dev_to_remove)
4648{
4649 nvlist_t **newdev = NULL;
4650
4651 if (count > 1)
4652 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4653
4654 for (int i = 0, j = 0; i < count; i++) {
4655 if (dev[i] == dev_to_remove)
4656 continue;
4657 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4658 }
4659
4660 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4661 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4662
4663 for (int i = 0; i < count - 1; i++)
4664 nvlist_free(newdev[i]);
4665
4666 if (count > 1)
4667 kmem_free(newdev, (count - 1) * sizeof (void *));
4668}
4669
4670/*
4671 * Evacuate the device.
4672 */
4673static int
4674spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4675{
4676 uint64_t txg;
4677 int error = 0;
4678
4679 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4680 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4681 ASSERT(vd == vd->vdev_top);
4682
4683 /*
4684 * Evacuate the device. We don't hold the config lock as writer
4685 * since we need to do I/O but we do keep the
4686 * spa_namespace_lock held. Once this completes the device
4687 * should no longer have any blocks allocated on it.
4688 */
4689 if (vd->vdev_islog) {
4690 if (vd->vdev_stat.vs_alloc != 0)
4691 error = spa_offline_log(spa);
4692 } else {
4693 error = ENOTSUP;
4694 }
4695
4696 if (error)
4697 return (error);
4698
4699 /*
4700 * The evacuation succeeded. Remove any remaining MOS metadata
4701 * associated with this vdev, and wait for these changes to sync.
4702 */
4703 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4704 txg = spa_vdev_config_enter(spa);
4705 vd->vdev_removing = B_TRUE;
4706 vdev_dirty(vd, 0, NULL, txg);
4707 vdev_config_dirty(vd);
4708 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4709
4710 return (0);
4711}
4712
4713/*
4714 * Complete the removal by cleaning up the namespace.
4715 */
4716static void
4717spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4718{
4719 vdev_t *rvd = spa->spa_root_vdev;
4720 uint64_t id = vd->vdev_id;
4721 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4722
4723 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4724 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4725 ASSERT(vd == vd->vdev_top);
4726
4727 /*
4728 * Only remove any devices which are empty.
4729 */
4730 if (vd->vdev_stat.vs_alloc != 0)
4731 return;
4732
4733 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4734
4735 if (list_link_active(&vd->vdev_state_dirty_node))
4736 vdev_state_clean(vd);
4737 if (list_link_active(&vd->vdev_config_dirty_node))
4738 vdev_config_clean(vd);
4739
4740 vdev_free(vd);
4741
4742 if (last_vdev) {
4743 vdev_compact_children(rvd);
4744 } else {
4745 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4746 vdev_add_child(rvd, vd);
4747 }
4748 vdev_config_dirty(rvd);
4749
4750 /*
4751 * Reassess the health of our root vdev.
4752 */
4753 vdev_reopen(rvd);
4754}
4755
4756/*
4757 * Remove a device from the pool -
4758 *
4759 * Removing a device from the vdev namespace requires several steps
4760 * and can take a significant amount of time. As a result we use
4761 * the spa_vdev_config_[enter/exit] functions which allow us to
4762 * grab and release the spa_config_lock while still holding the namespace
4763 * lock. During each step the configuration is synced out.
4764 */
4765
4766/*
4767 * Remove a device from the pool. Currently, this supports removing only hot
4768 * spares, slogs, and level 2 ARC devices.
4769 */
4770int
4771spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4772{
4773 vdev_t *vd;
4774 metaslab_group_t *mg;
4775 nvlist_t **spares, **l2cache, *nv;
4776 uint64_t txg = 0;
4777 uint_t nspares, nl2cache;
4778 int error = 0;
4779 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4780
4781 ASSERT(spa_writeable(spa));
4782
4783 if (!locked)
4784 txg = spa_vdev_enter(spa);
4785
4786 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4787
4788 if (spa->spa_spares.sav_vdevs != NULL &&
4789 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4790 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4791 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4792 /*
4793 * Only remove the hot spare if it's not currently in use
4794 * in this pool.
4795 */
4796 if (vd == NULL || unspare) {
4797 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4798 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4799 spa_load_spares(spa);
4800 spa->spa_spares.sav_sync = B_TRUE;
4801 } else {
4802 error = EBUSY;
4803 }
4804 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4805 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4806 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4807 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4808 /*
4809 * Cache devices can always be removed.
4810 */
4811 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4812 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4813 spa_load_l2cache(spa);
4814 spa->spa_l2cache.sav_sync = B_TRUE;
4815 } else if (vd != NULL && vd->vdev_islog) {
4816 ASSERT(!locked);
4817 ASSERT(vd == vd->vdev_top);
4818
4819 /*
4820 * XXX - Once we have bp-rewrite this should
4821 * become the common case.
4822 */
4823
4824 mg = vd->vdev_mg;
4825
4826 /*
4827 * Stop allocating from this vdev.
4828 */
4829 metaslab_group_passivate(mg);
4830
4831 /*
4832 * Wait for the youngest allocations and frees to sync,
4833 * and then wait for the deferral of those frees to finish.
4834 */
4835 spa_vdev_config_exit(spa, NULL,
4836 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4837
4838 /*
4839 * Attempt to evacuate the vdev.
4840 */
4841 error = spa_vdev_remove_evacuate(spa, vd);
4842
4843 txg = spa_vdev_config_enter(spa);
4844
4845 /*
4846 * If we couldn't evacuate the vdev, unwind.
4847 */
4848 if (error) {
4849 metaslab_group_activate(mg);
4850 return (spa_vdev_exit(spa, NULL, txg, error));
4851 }
4852
4853 /*
4854 * Clean up the vdev namespace.
4855 */
4856 spa_vdev_remove_from_namespace(spa, vd);
4857
4858 } else if (vd != NULL) {
4859 /*
4860 * Normal vdevs cannot be removed (yet).
4861 */
4862 error = ENOTSUP;
4863 } else {
4864 /*
4865 * There is no vdev of any kind with the specified guid.
4866 */
4867 error = ENOENT;
4868 }
4869
4870 if (!locked)
4871 return (spa_vdev_exit(spa, NULL, txg, error));
4872
4873 return (error);
4874}
4875
4876/*
4877 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4878 * current spared, so we can detach it.
4879 */
4880static vdev_t *
4881spa_vdev_resilver_done_hunt(vdev_t *vd)
4882{
4883 vdev_t *newvd, *oldvd;
4884
4885 for (int c = 0; c < vd->vdev_children; c++) {
4886 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4887 if (oldvd != NULL)
4888 return (oldvd);
4889 }
4890
4891 /*
4892 * Check for a completed replacement. We always consider the first
4893 * vdev in the list to be the oldest vdev, and the last one to be
4894 * the newest (see spa_vdev_attach() for how that works). In
4895 * the case where the newest vdev is faulted, we will not automatically
4896 * remove it after a resilver completes. This is OK as it will require
4897 * user intervention to determine which disk the admin wishes to keep.
4898 */
4899 if (vd->vdev_ops == &vdev_replacing_ops) {
4900 ASSERT(vd->vdev_children > 1);
4901
4902 newvd = vd->vdev_child[vd->vdev_children - 1];
4903 oldvd = vd->vdev_child[0];
4904
4905 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4906 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4907 !vdev_dtl_required(oldvd))
4908 return (oldvd);
4909 }
4910
4911 /*
4912 * Check for a completed resilver with the 'unspare' flag set.
4913 */
4914 if (vd->vdev_ops == &vdev_spare_ops) {
4915 vdev_t *first = vd->vdev_child[0];
4916 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4917
4918 if (last->vdev_unspare) {
4919 oldvd = first;
4920 newvd = last;
4921 } else if (first->vdev_unspare) {
4922 oldvd = last;
4923 newvd = first;
4924 } else {
4925 oldvd = NULL;
4926 }
4927
4928 if (oldvd != NULL &&
4929 vdev_dtl_empty(newvd, DTL_MISSING) &&
4930 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4931 !vdev_dtl_required(oldvd))
4932 return (oldvd);
4933
4934 /*
4935 * If there are more than two spares attached to a disk,
4936 * and those spares are not required, then we want to
4937 * attempt to free them up now so that they can be used
4938 * by other pools. Once we're back down to a single
4939 * disk+spare, we stop removing them.
4940 */
4941 if (vd->vdev_children > 2) {
4942 newvd = vd->vdev_child[1];
4943
4944 if (newvd->vdev_isspare && last->vdev_isspare &&
4945 vdev_dtl_empty(last, DTL_MISSING) &&
4946 vdev_dtl_empty(last, DTL_OUTAGE) &&
4947 !vdev_dtl_required(newvd))
4948 return (newvd);
4949 }
4950 }
4951
4952 return (NULL);
4953}
4954
4955static void
4956spa_vdev_resilver_done(spa_t *spa)
4957{
4958 vdev_t *vd, *pvd, *ppvd;
4959 uint64_t guid, sguid, pguid, ppguid;
4960
4961 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4962
4963 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4964 pvd = vd->vdev_parent;
4965 ppvd = pvd->vdev_parent;
4966 guid = vd->vdev_guid;
4967 pguid = pvd->vdev_guid;
4968 ppguid = ppvd->vdev_guid;
4969 sguid = 0;
4970 /*
4971 * If we have just finished replacing a hot spared device, then
4972 * we need to detach the parent's first child (the original hot
4973 * spare) as well.
4974 */
4975 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4976 ppvd->vdev_children == 2) {
4977 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4978 sguid = ppvd->vdev_child[1]->vdev_guid;
4979 }
4980 spa_config_exit(spa, SCL_ALL, FTAG);
4981 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4982 return;
4983 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4984 return;
4985 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4986 }
4987
4988 spa_config_exit(spa, SCL_ALL, FTAG);
4989}
4990
4991/*
4992 * Update the stored path or FRU for this vdev.
4993 */
4994int
4995spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4996 boolean_t ispath)
4997{
4998 vdev_t *vd;
4999 boolean_t sync = B_FALSE;
5000
5001 ASSERT(spa_writeable(spa));
5002
5003 spa_vdev_state_enter(spa, SCL_ALL);
5004
5005 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5006 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5007
5008 if (!vd->vdev_ops->vdev_op_leaf)
5009 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5010
5011 if (ispath) {
5012 if (strcmp(value, vd->vdev_path) != 0) {
5013 spa_strfree(vd->vdev_path);
5014 vd->vdev_path = spa_strdup(value);
5015 sync = B_TRUE;
5016 }
5017 } else {
5018 if (vd->vdev_fru == NULL) {
5019 vd->vdev_fru = spa_strdup(value);
5020 sync = B_TRUE;
5021 } else if (strcmp(value, vd->vdev_fru) != 0) {
5022 spa_strfree(vd->vdev_fru);
5023 vd->vdev_fru = spa_strdup(value);
5024 sync = B_TRUE;
5025 }
5026 }
5027
5028 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5029}
5030
5031int
5032spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5033{
5034 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5035}
5036
5037int
5038spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5039{
5040 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5041}
5042
5043/*
5044 * ==========================================================================
5045 * SPA Scanning
5046 * ==========================================================================
5047 */
5048
5049int
5050spa_scan_stop(spa_t *spa)
5051{
5052 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5053 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5054 return (EBUSY);
5055 return (dsl_scan_cancel(spa->spa_dsl_pool));
5056}
5057
5058int
5059spa_scan(spa_t *spa, pool_scan_func_t func)
5060{
5061 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5062
5063 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5064 return (ENOTSUP);
5065
5066 /*
5067 * If a resilver was requested, but there is no DTL on a
5068 * writeable leaf device, we have nothing to do.
5069 */
5070 if (func == POOL_SCAN_RESILVER &&
5071 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5072 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5073 return (0);
5074 }
5075
5076 return (dsl_scan(spa->spa_dsl_pool, func));
5077}
5078
5079/*
5080 * ==========================================================================
5081 * SPA async task processing
5082 * ==========================================================================
5083 */
5084
5085static void
5086spa_async_remove(spa_t *spa, vdev_t *vd)
5087{
5088 if (vd->vdev_remove_wanted) {
5089 vd->vdev_remove_wanted = B_FALSE;
5090 vd->vdev_delayed_close = B_FALSE;
5091 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5092
5093 /*
5094 * We want to clear the stats, but we don't want to do a full
5095 * vdev_clear() as that will cause us to throw away
5096 * degraded/faulted state as well as attempt to reopen the
5097 * device, all of which is a waste.
5098 */
5099 vd->vdev_stat.vs_read_errors = 0;
5100 vd->vdev_stat.vs_write_errors = 0;
5101 vd->vdev_stat.vs_checksum_errors = 0;
5102
5103 vdev_state_dirty(vd->vdev_top);
5104 }
5105
5106 for (int c = 0; c < vd->vdev_children; c++)
5107 spa_async_remove(spa, vd->vdev_child[c]);
5108}
5109
5110static void
5111spa_async_probe(spa_t *spa, vdev_t *vd)
5112{
5113 if (vd->vdev_probe_wanted) {
5114 vd->vdev_probe_wanted = B_FALSE;
5115 vdev_reopen(vd); /* vdev_open() does the actual probe */
5116 }
5117
5118 for (int c = 0; c < vd->vdev_children; c++)
5119 spa_async_probe(spa, vd->vdev_child[c]);
5120}
5121
5122static void
5123spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5124{
5125 sysevent_id_t eid;
5126 nvlist_t *attr;
5127 char *physpath;
5128
5129 if (!spa->spa_autoexpand)
5130 return;
5131
5132 for (int c = 0; c < vd->vdev_children; c++) {
5133 vdev_t *cvd = vd->vdev_child[c];
5134 spa_async_autoexpand(spa, cvd);
5135 }
5136
5137 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5138 return;
5139
5140 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5141 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5142
5143 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5144 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5145
5146 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5147 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5148
5149 nvlist_free(attr);
5150 kmem_free(physpath, MAXPATHLEN);
5151}
5152
5153static void
5154spa_async_thread(void *arg)
5155{
5156 spa_t *spa = arg;
5157 int tasks;
5158
5159 ASSERT(spa->spa_sync_on);
5160
5161 mutex_enter(&spa->spa_async_lock);
5162 tasks = spa->spa_async_tasks;
5163 spa->spa_async_tasks = 0;
5164 mutex_exit(&spa->spa_async_lock);
5165
5166 /*
5167 * See if the config needs to be updated.
5168 */
5169 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5170 uint64_t old_space, new_space;
5171
5172 mutex_enter(&spa_namespace_lock);
5173 old_space = metaslab_class_get_space(spa_normal_class(spa));
5174 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5175 new_space = metaslab_class_get_space(spa_normal_class(spa));
5176 mutex_exit(&spa_namespace_lock);
5177
5178 /*
5179 * If the pool grew as a result of the config update,
5180 * then log an internal history event.
5181 */
5182 if (new_space != old_space) {
5183 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5184 spa, NULL,
5185 "pool '%s' size: %llu(+%llu)",
5186 spa_name(spa), new_space, new_space - old_space);
5187 }
5188 }
5189
5190 /*
5191 * See if any devices need to be marked REMOVED.
5192 */
5193 if (tasks & SPA_ASYNC_REMOVE) {
5194 spa_vdev_state_enter(spa, SCL_NONE);
5195 spa_async_remove(spa, spa->spa_root_vdev);
5196 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5197 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5198 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5199 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5200 (void) spa_vdev_state_exit(spa, NULL, 0);
5201 }
5202
5203 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5204 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5205 spa_async_autoexpand(spa, spa->spa_root_vdev);
5206 spa_config_exit(spa, SCL_CONFIG, FTAG);
5207 }
5208
5209 /*
5210 * See if any devices need to be probed.
5211 */
5212 if (tasks & SPA_ASYNC_PROBE) {
5213 spa_vdev_state_enter(spa, SCL_NONE);
5214 spa_async_probe(spa, spa->spa_root_vdev);
5215 (void) spa_vdev_state_exit(spa, NULL, 0);
5216 }
5217
5218 /*
5219 * If any devices are done replacing, detach them.
5220 */
5221 if (tasks & SPA_ASYNC_RESILVER_DONE)
5222 spa_vdev_resilver_done(spa);
5223
5224 /*
5225 * Kick off a resilver.
5226 */
5227 if (tasks & SPA_ASYNC_RESILVER)
5228 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5229
5230 /*
5231 * Let the world know that we're done.
5232 */
5233 mutex_enter(&spa->spa_async_lock);
5234 spa->spa_async_thread = NULL;
5235 cv_broadcast(&spa->spa_async_cv);
5236 mutex_exit(&spa->spa_async_lock);
5237 thread_exit();
5238}
5239
5240void
5241spa_async_suspend(spa_t *spa)
5242{
5243 mutex_enter(&spa->spa_async_lock);
5244 spa->spa_async_suspended++;
5245 while (spa->spa_async_thread != NULL)
5246 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5247 mutex_exit(&spa->spa_async_lock);
5248}
5249
5250void
5251spa_async_resume(spa_t *spa)
5252{
5253 mutex_enter(&spa->spa_async_lock);
5254 ASSERT(spa->spa_async_suspended != 0);
5255 spa->spa_async_suspended--;
5256 mutex_exit(&spa->spa_async_lock);
5257}
5258
5259static void
5260spa_async_dispatch(spa_t *spa)
5261{
5262 mutex_enter(&spa->spa_async_lock);
5263 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5264 spa->spa_async_thread == NULL &&
5265 rootdir != NULL && !vn_is_readonly(rootdir))
5266 spa->spa_async_thread = thread_create(NULL, 0,
5267 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5268 mutex_exit(&spa->spa_async_lock);
5269}
5270
5271void
5272spa_async_request(spa_t *spa, int task)
5273{
5274 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5275 mutex_enter(&spa->spa_async_lock);
5276 spa->spa_async_tasks |= task;
5277 mutex_exit(&spa->spa_async_lock);
5278}
5279
5280/*
5281 * ==========================================================================
5282 * SPA syncing routines
5283 * ==========================================================================
5284 */
5285
5286static int
5287bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5288{
5289 bpobj_t *bpo = arg;
5290 bpobj_enqueue(bpo, bp, tx);
5291 return (0);
5292}
5293
5294static int
5295spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5296{
5297 zio_t *zio = arg;
5298
5299 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5300 zio->io_flags));
5301 return (0);
5302}
5303
5304static void
5305spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5306{
5307 char *packed = NULL;
5308 size_t bufsize;
5309 size_t nvsize = 0;
5310 dmu_buf_t *db;
5311
5312 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5313
5314 /*
5315 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5316 * information. This avoids the dbuf_will_dirty() path and
5317 * saves us a pre-read to get data we don't actually care about.
5318 */
5319 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5320 packed = kmem_alloc(bufsize, KM_SLEEP);
5321
5322 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5323 KM_SLEEP) == 0);
5324 bzero(packed + nvsize, bufsize - nvsize);
5325
5326 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5327
5328 kmem_free(packed, bufsize);
5329
5330 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5331 dmu_buf_will_dirty(db, tx);
5332 *(uint64_t *)db->db_data = nvsize;
5333 dmu_buf_rele(db, FTAG);
5334}
5335
5336static void
5337spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5338 const char *config, const char *entry)
5339{
5340 nvlist_t *nvroot;
5341 nvlist_t **list;
5342 int i;
5343
5344 if (!sav->sav_sync)
5345 return;
5346
5347 /*
5348 * Update the MOS nvlist describing the list of available devices.
5349 * spa_validate_aux() will have already made sure this nvlist is
5350 * valid and the vdevs are labeled appropriately.
5351 */
5352 if (sav->sav_object == 0) {
5353 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5354 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5355 sizeof (uint64_t), tx);
5356 VERIFY(zap_update(spa->spa_meta_objset,
5357 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5358 &sav->sav_object, tx) == 0);
5359 }
5360
5361 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5362 if (sav->sav_count == 0) {
5363 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5364 } else {
5365 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5366 for (i = 0; i < sav->sav_count; i++)
5367 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5368 B_FALSE, VDEV_CONFIG_L2CACHE);
5369 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5370 sav->sav_count) == 0);
5371 for (i = 0; i < sav->sav_count; i++)
5372 nvlist_free(list[i]);
5373 kmem_free(list, sav->sav_count * sizeof (void *));
5374 }
5375
5376 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5377 nvlist_free(nvroot);
5378
5379 sav->sav_sync = B_FALSE;
5380}
5381
5382static void
5383spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5384{
5385 nvlist_t *config;
5386
5387 if (list_is_empty(&spa->spa_config_dirty_list))
5388 return;
5389
5390 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5391
5392 config = spa_config_generate(spa, spa->spa_root_vdev,
5393 dmu_tx_get_txg(tx), B_FALSE);
5394
5395 spa_config_exit(spa, SCL_STATE, FTAG);
5396
5397 if (spa->spa_config_syncing)
5398 nvlist_free(spa->spa_config_syncing);
5399 spa->spa_config_syncing = config;
5400
5401 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5402}
5403
5404/*
5405 * Set zpool properties.
5406 */
5407static void
5408spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5409{
5410 spa_t *spa = arg1;
5411 objset_t *mos = spa->spa_meta_objset;
5412 nvlist_t *nvp = arg2;
5413 nvpair_t *elem;
5414 uint64_t intval;
5415 char *strval;
5416 zpool_prop_t prop;
5417 const char *propname;
5418 zprop_type_t proptype;
5419
5420 mutex_enter(&spa->spa_props_lock);
5421
5422 elem = NULL;
5423 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5424 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5425 case ZPOOL_PROP_VERSION:
5426 /*
5427 * Only set version for non-zpool-creation cases
5428 * (set/import). spa_create() needs special care
5429 * for version setting.
5430 */
5431 if (tx->tx_txg != TXG_INITIAL) {
5432 VERIFY(nvpair_value_uint64(elem,
5433 &intval) == 0);
5434 ASSERT(intval <= SPA_VERSION);
5435 ASSERT(intval >= spa_version(spa));
5436 spa->spa_uberblock.ub_version = intval;
5437 vdev_config_dirty(spa->spa_root_vdev);
5438 }
5439 break;
5440
5441 case ZPOOL_PROP_ALTROOT:
5442 /*
5443 * 'altroot' is a non-persistent property. It should
5444 * have been set temporarily at creation or import time.
5445 */
5446 ASSERT(spa->spa_root != NULL);
5447 break;
5448
5449 case ZPOOL_PROP_READONLY:
5450 case ZPOOL_PROP_CACHEFILE:
5451 /*
5452 * 'readonly' and 'cachefile' are also non-persisitent
5453 * properties.
5454 */
5455 break;
5456 case ZPOOL_PROP_COMMENT:
5457 VERIFY(nvpair_value_string(elem, &strval) == 0);
5458 if (spa->spa_comment != NULL)
5459 spa_strfree(spa->spa_comment);
5460 spa->spa_comment = spa_strdup(strval);
5461 /*
5462 * We need to dirty the configuration on all the vdevs
5463 * so that their labels get updated. It's unnecessary
5464 * to do this for pool creation since the vdev's
5465 * configuratoin has already been dirtied.
5466 */
5467 if (tx->tx_txg != TXG_INITIAL)
5468 vdev_config_dirty(spa->spa_root_vdev);
5469 break;
5470 default:
5471 /*
5472 * Set pool property values in the poolprops mos object.
5473 */
5474 if (spa->spa_pool_props_object == 0) {
5475 VERIFY((spa->spa_pool_props_object =
5476 zap_create(mos, DMU_OT_POOL_PROPS,
5477 DMU_OT_NONE, 0, tx)) > 0);
5478
5479 VERIFY(zap_update(mos,
5480 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5481 8, 1, &spa->spa_pool_props_object, tx)
5482 == 0);
5483 }
5484
5485 /* normalize the property name */
5486 propname = zpool_prop_to_name(prop);
5487 proptype = zpool_prop_get_type(prop);
5488
5489 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5490 ASSERT(proptype == PROP_TYPE_STRING);
5491 VERIFY(nvpair_value_string(elem, &strval) == 0);
5492 VERIFY(zap_update(mos,
5493 spa->spa_pool_props_object, propname,
5494 1, strlen(strval) + 1, strval, tx) == 0);
5495
5496 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5497 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5498
5499 if (proptype == PROP_TYPE_INDEX) {
5500 const char *unused;
5501 VERIFY(zpool_prop_index_to_string(
5502 prop, intval, &unused) == 0);
5503 }
5504 VERIFY(zap_update(mos,
5505 spa->spa_pool_props_object, propname,
5506 8, 1, &intval, tx) == 0);
5507 } else {
5508 ASSERT(0); /* not allowed */
5509 }
5510
5511 switch (prop) {
5512 case ZPOOL_PROP_DELEGATION:
5513 spa->spa_delegation = intval;
5514 break;
5515 case ZPOOL_PROP_BOOTFS:
5516 spa->spa_bootfs = intval;
5517 break;
5518 case ZPOOL_PROP_FAILUREMODE:
5519 spa->spa_failmode = intval;
5520 break;
5521 case ZPOOL_PROP_AUTOEXPAND:
5522 spa->spa_autoexpand = intval;
5523 if (tx->tx_txg != TXG_INITIAL)
5524 spa_async_request(spa,
5525 SPA_ASYNC_AUTOEXPAND);
5526 break;
5527 case ZPOOL_PROP_DEDUPDITTO:
5528 spa->spa_dedup_ditto = intval;
5529 break;
5530 default:
5531 break;
5532 }
5533 }
5534
5535 /* log internal history if this is not a zpool create */
5536 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5537 tx->tx_txg != TXG_INITIAL) {
5538 spa_history_log_internal(LOG_POOL_PROPSET,
5539 spa, tx, "%s %lld %s",
5540 nvpair_name(elem), intval, spa_name(spa));
5541 }
5542 }
5543
5544 mutex_exit(&spa->spa_props_lock);
5545}
5546
5547/*
5548 * Perform one-time upgrade on-disk changes. spa_version() does not
5549 * reflect the new version this txg, so there must be no changes this
5550 * txg to anything that the upgrade code depends on after it executes.
5551 * Therefore this must be called after dsl_pool_sync() does the sync
5552 * tasks.
5553 */
5554static void
5555spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5556{
5557 dsl_pool_t *dp = spa->spa_dsl_pool;
5558
5559 ASSERT(spa->spa_sync_pass == 1);
5560
5561 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5562 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5563 dsl_pool_create_origin(dp, tx);
5564
5565 /* Keeping the origin open increases spa_minref */
5566 spa->spa_minref += 3;
5567 }
5568
5569 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5570 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5571 dsl_pool_upgrade_clones(dp, tx);
5572 }
5573
5574 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5575 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5576 dsl_pool_upgrade_dir_clones(dp, tx);
5577
5578 /* Keeping the freedir open increases spa_minref */
5579 spa->spa_minref += 3;
5580 }
5581}
5582
5583/*
5584 * Sync the specified transaction group. New blocks may be dirtied as
5585 * part of the process, so we iterate until it converges.
5586 */
5587void
5588spa_sync(spa_t *spa, uint64_t txg)
5589{
5590 dsl_pool_t *dp = spa->spa_dsl_pool;
5591 objset_t *mos = spa->spa_meta_objset;
5592 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5593 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5594 vdev_t *rvd = spa->spa_root_vdev;
5595 vdev_t *vd;
5596 dmu_tx_t *tx;
5597 int error;
5598
5599 VERIFY(spa_writeable(spa));
5600
5601 /*
5602 * Lock out configuration changes.
5603 */
5604 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5605
5606 spa->spa_syncing_txg = txg;
5607 spa->spa_sync_pass = 0;
5608
5609 /*
5610 * If there are any pending vdev state changes, convert them
5611 * into config changes that go out with this transaction group.
5612 */
5613 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5614 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5615 /*
5616 * We need the write lock here because, for aux vdevs,
5617 * calling vdev_config_dirty() modifies sav_config.
5618 * This is ugly and will become unnecessary when we
5619 * eliminate the aux vdev wart by integrating all vdevs
5620 * into the root vdev tree.
5621 */
5622 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5623 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5624 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5625 vdev_state_clean(vd);
5626 vdev_config_dirty(vd);
5627 }
5628 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5629 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5630 }
5631 spa_config_exit(spa, SCL_STATE, FTAG);
5632
5633 tx = dmu_tx_create_assigned(dp, txg);
5634
5635 /*
5636 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5637 * set spa_deflate if we have no raid-z vdevs.
5638 */
5639 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5640 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5641 int i;
5642
5643 for (i = 0; i < rvd->vdev_children; i++) {
5644 vd = rvd->vdev_child[i];
5645 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5646 break;
5647 }
5648 if (i == rvd->vdev_children) {
5649 spa->spa_deflate = TRUE;
5650 VERIFY(0 == zap_add(spa->spa_meta_objset,
5651 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5652 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5653 }
5654 }
5655
5656 /*
5657 * If anything has changed in this txg, or if someone is waiting
5658 * for this txg to sync (eg, spa_vdev_remove()), push the
5659 * deferred frees from the previous txg. If not, leave them
5660 * alone so that we don't generate work on an otherwise idle
5661 * system.
5662 */
5663 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5664 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5665 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5666 ((dsl_scan_active(dp->dp_scan) ||
5667 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5668 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5669 VERIFY3U(bpobj_iterate(defer_bpo,
5670 spa_free_sync_cb, zio, tx), ==, 0);
5671 VERIFY3U(zio_wait(zio), ==, 0);
5672 }
5673
5674 /*
5675 * Iterate to convergence.
5676 */
5677 do {
5678 int pass = ++spa->spa_sync_pass;
5679
5680 spa_sync_config_object(spa, tx);
5681 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5682 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5683 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5684 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5685 spa_errlog_sync(spa, txg);
5686 dsl_pool_sync(dp, txg);
5687
5688 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5689 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5690 bplist_iterate(free_bpl, spa_free_sync_cb,
5691 zio, tx);
5692 VERIFY(zio_wait(zio) == 0);
5693 } else {
5694 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5695 defer_bpo, tx);
5696 }
5697
5698 ddt_sync(spa, txg);
5699 dsl_scan_sync(dp, tx);
5700
5701 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5702 vdev_sync(vd, txg);
5703
5704 if (pass == 1)
5705 spa_sync_upgrades(spa, tx);
5706
5707 } while (dmu_objset_is_dirty(mos, txg));
5708
5709 /*
5710 * Rewrite the vdev configuration (which includes the uberblock)
5711 * to commit the transaction group.
5712 *
5713 * If there are no dirty vdevs, we sync the uberblock to a few
5714 * random top-level vdevs that are known to be visible in the
5715 * config cache (see spa_vdev_add() for a complete description).
5716 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5717 */
5718 for (;;) {
5719 /*
5720 * We hold SCL_STATE to prevent vdev open/close/etc.
5721 * while we're attempting to write the vdev labels.
5722 */
5723 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5724
5725 if (list_is_empty(&spa->spa_config_dirty_list)) {
5726 vdev_t *svd[SPA_DVAS_PER_BP];
5727 int svdcount = 0;
5728 int children = rvd->vdev_children;
5729 int c0 = spa_get_random(children);
5730
5731 for (int c = 0; c < children; c++) {
5732 vd = rvd->vdev_child[(c0 + c) % children];
5733 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5734 continue;
5735 svd[svdcount++] = vd;
5736 if (svdcount == SPA_DVAS_PER_BP)
5737 break;
5738 }
5739 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5740 if (error != 0)
5741 error = vdev_config_sync(svd, svdcount, txg,
5742 B_TRUE);
5743 } else {
5744 error = vdev_config_sync(rvd->vdev_child,
5745 rvd->vdev_children, txg, B_FALSE);
5746 if (error != 0)
5747 error = vdev_config_sync(rvd->vdev_child,
5748 rvd->vdev_children, txg, B_TRUE);
5749 }
5750
5751 spa_config_exit(spa, SCL_STATE, FTAG);
5752
5753 if (error == 0)
5754 break;
5755 zio_suspend(spa, NULL);
5756 zio_resume_wait(spa);
5757 }
5758 dmu_tx_commit(tx);
5759
5760 /*
5761 * Clear the dirty config list.
5762 */
5763 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5764 vdev_config_clean(vd);
5765
5766 /*
5767 * Now that the new config has synced transactionally,
5768 * let it become visible to the config cache.
5769 */
5770 if (spa->spa_config_syncing != NULL) {
5771 spa_config_set(spa, spa->spa_config_syncing);
5772 spa->spa_config_txg = txg;
5773 spa->spa_config_syncing = NULL;
5774 }
5775
5776 spa->spa_ubsync = spa->spa_uberblock;
5777
5778 dsl_pool_sync_done(dp, txg);
5779
5780 /*
5781 * Update usable space statistics.
5782 */
5783 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5784 vdev_sync_done(vd, txg);
5785
5786 spa_update_dspace(spa);
5787
5788 /*
5789 * It had better be the case that we didn't dirty anything
5790 * since vdev_config_sync().
5791 */
5792 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5793 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5794 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5795
5796 spa->spa_sync_pass = 0;
5797
5798 spa_config_exit(spa, SCL_CONFIG, FTAG);
5799
5800 spa_handle_ignored_writes(spa);
5801
5802 /*
5803 * If any async tasks have been requested, kick them off.
5804 */
5805 spa_async_dispatch(spa);
5806}
5807
5808/*
5809 * Sync all pools. We don't want to hold the namespace lock across these
5810 * operations, so we take a reference on the spa_t and drop the lock during the
5811 * sync.
5812 */
5813void
5814spa_sync_allpools(void)
5815{
5816 spa_t *spa = NULL;
5817 mutex_enter(&spa_namespace_lock);
5818 while ((spa = spa_next(spa)) != NULL) {
5819 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5820 !spa_writeable(spa) || spa_suspended(spa))
5821 continue;
5822 spa_open_ref(spa, FTAG);
5823 mutex_exit(&spa_namespace_lock);
5824 txg_wait_synced(spa_get_dsl(spa), 0);
5825 mutex_enter(&spa_namespace_lock);
5826 spa_close(spa, FTAG);
5827 }
5828 mutex_exit(&spa_namespace_lock);
5829}
5830
5831/*
5832 * ==========================================================================
5833 * Miscellaneous routines
5834 * ==========================================================================
5835 */
5836
5837/*
5838 * Remove all pools in the system.
5839 */
5840void
5841spa_evict_all(void)
5842{
5843 spa_t *spa;
5844
5845 /*
5846 * Remove all cached state. All pools should be closed now,
5847 * so every spa in the AVL tree should be unreferenced.
5848 */
5849 mutex_enter(&spa_namespace_lock);
5850 while ((spa = spa_next(NULL)) != NULL) {
5851 /*
5852 * Stop async tasks. The async thread may need to detach
5853 * a device that's been replaced, which requires grabbing
5854 * spa_namespace_lock, so we must drop it here.
5855 */
5856 spa_open_ref(spa, FTAG);
5857 mutex_exit(&spa_namespace_lock);
5858 spa_async_suspend(spa);
5859 mutex_enter(&spa_namespace_lock);
5860 spa_close(spa, FTAG);
5861
5862 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5863 spa_unload(spa);
5864 spa_deactivate(spa);
5865 }
5866 spa_remove(spa);
5867 }
5868 mutex_exit(&spa_namespace_lock);
5869}
5870
5871vdev_t *
5872spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5873{
5874 vdev_t *vd;
5875 int i;
5876
5877 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5878 return (vd);
5879
5880 if (aux) {
5881 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5882 vd = spa->spa_l2cache.sav_vdevs[i];
5883 if (vd->vdev_guid == guid)
5884 return (vd);
5885 }
5886
5887 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5888 vd = spa->spa_spares.sav_vdevs[i];
5889 if (vd->vdev_guid == guid)
5890 return (vd);
5891 }
5892 }
5893
5894 return (NULL);
5895}
5896
5897void
5898spa_upgrade(spa_t *spa, uint64_t version)
5899{
5900 ASSERT(spa_writeable(spa));
5901
5902 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5903
5904 /*
5905 * This should only be called for a non-faulted pool, and since a
5906 * future version would result in an unopenable pool, this shouldn't be
5907 * possible.
5908 */
5909 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5910 ASSERT(version >= spa->spa_uberblock.ub_version);
5911
5912 spa->spa_uberblock.ub_version = version;
5913 vdev_config_dirty(spa->spa_root_vdev);
5914
5915 spa_config_exit(spa, SCL_ALL, FTAG);
5916
5917 txg_wait_synced(spa_get_dsl(spa), 0);
5918}
5919
5920boolean_t
5921spa_has_spare(spa_t *spa, uint64_t guid)
5922{
5923 int i;
5924 uint64_t spareguid;
5925 spa_aux_vdev_t *sav = &spa->spa_spares;
5926
5927 for (i = 0; i < sav->sav_count; i++)
5928 if (sav->sav_vdevs[i]->vdev_guid == guid)
5929 return (B_TRUE);
5930
5931 for (i = 0; i < sav->sav_npending; i++) {
5932 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5933 &spareguid) == 0 && spareguid == guid)
5934 return (B_TRUE);
5935 }
5936
5937 return (B_FALSE);
5938}
5939
5940/*
5941 * Check if a pool has an active shared spare device.
5942 * Note: reference count of an active spare is 2, as a spare and as a replace
5943 */
5944static boolean_t
5945spa_has_active_shared_spare(spa_t *spa)
5946{
5947 int i, refcnt;
5948 uint64_t pool;
5949 spa_aux_vdev_t *sav = &spa->spa_spares;
5950
5951 for (i = 0; i < sav->sav_count; i++) {
5952 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5953 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5954 refcnt > 2)
5955 return (B_TRUE);
5956 }
5957
5958 return (B_FALSE);
5959}
5960
5961/*
5962 * Post a sysevent corresponding to the given event. The 'name' must be one of
5963 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5964 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5965 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5966 * or zdb as real changes.
5967 */
5968void
5969spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5970{
5971#ifdef _KERNEL
5972 sysevent_t *ev;
5973 sysevent_attr_list_t *attr = NULL;
5974 sysevent_value_t value;
5975 sysevent_id_t eid;
5976
5977 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5978 SE_SLEEP);
5979
5980 value.value_type = SE_DATA_TYPE_STRING;
5981 value.value.sv_string = spa_name(spa);
5982 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5983 goto done;
5984
5985 value.value_type = SE_DATA_TYPE_UINT64;
5986 value.value.sv_uint64 = spa_guid(spa);
5987 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5988 goto done;
5989
5990 if (vd) {
5991 value.value_type = SE_DATA_TYPE_UINT64;
5992 value.value.sv_uint64 = vd->vdev_guid;
5993 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5994 SE_SLEEP) != 0)
5995 goto done;
5996
5997 if (vd->vdev_path) {
5998 value.value_type = SE_DATA_TYPE_STRING;
5999 value.value.sv_string = vd->vdev_path;
6000 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6001 &value, SE_SLEEP) != 0)
6002 goto done;
6003 }
6004 }
6005
6006 if (sysevent_attach_attributes(ev, attr) != 0)
6007 goto done;
6008 attr = NULL;
6009
6010 (void) log_sysevent(ev, SE_SLEEP, &eid);
6011
6012done:
6013 if (attr)
6014 sysevent_free_attr(attr);
6015 sysevent_free(ev);
6016#endif
6017}
| 3934 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3935
3936 if (newrootvd->vdev_children != 1)
3937 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3938
3939 newvd = newrootvd->vdev_child[0];
3940
3941 if (!newvd->vdev_ops->vdev_op_leaf)
3942 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3943
3944 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3945 return (spa_vdev_exit(spa, newrootvd, txg, error));
3946
3947 /*
3948 * Spares can't replace logs
3949 */
3950 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3951 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3952
3953 if (!replacing) {
3954 /*
3955 * For attach, the only allowable parent is a mirror or the root
3956 * vdev.
3957 */
3958 if (pvd->vdev_ops != &vdev_mirror_ops &&
3959 pvd->vdev_ops != &vdev_root_ops)
3960 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3961
3962 pvops = &vdev_mirror_ops;
3963 } else {
3964 /*
3965 * Active hot spares can only be replaced by inactive hot
3966 * spares.
3967 */
3968 if (pvd->vdev_ops == &vdev_spare_ops &&
3969 oldvd->vdev_isspare &&
3970 !spa_has_spare(spa, newvd->vdev_guid))
3971 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3972
3973 /*
3974 * If the source is a hot spare, and the parent isn't already a
3975 * spare, then we want to create a new hot spare. Otherwise, we
3976 * want to create a replacing vdev. The user is not allowed to
3977 * attach to a spared vdev child unless the 'isspare' state is
3978 * the same (spare replaces spare, non-spare replaces
3979 * non-spare).
3980 */
3981 if (pvd->vdev_ops == &vdev_replacing_ops &&
3982 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3983 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3984 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3985 newvd->vdev_isspare != oldvd->vdev_isspare) {
3986 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3987 }
3988
3989 if (newvd->vdev_isspare)
3990 pvops = &vdev_spare_ops;
3991 else
3992 pvops = &vdev_replacing_ops;
3993 }
3994
3995 /*
3996 * Make sure the new device is big enough.
3997 */
3998 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3999 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4000
4001 /*
4002 * The new device cannot have a higher alignment requirement
4003 * than the top-level vdev.
4004 */
4005 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4006 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4007
4008 /*
4009 * If this is an in-place replacement, update oldvd's path and devid
4010 * to make it distinguishable from newvd, and unopenable from now on.
4011 */
4012 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4013 spa_strfree(oldvd->vdev_path);
4014 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4015 KM_SLEEP);
4016 (void) sprintf(oldvd->vdev_path, "%s/%s",
4017 newvd->vdev_path, "old");
4018 if (oldvd->vdev_devid != NULL) {
4019 spa_strfree(oldvd->vdev_devid);
4020 oldvd->vdev_devid = NULL;
4021 }
4022 }
4023
4024 /* mark the device being resilvered */
4025 newvd->vdev_resilvering = B_TRUE;
4026
4027 /*
4028 * If the parent is not a mirror, or if we're replacing, insert the new
4029 * mirror/replacing/spare vdev above oldvd.
4030 */
4031 if (pvd->vdev_ops != pvops)
4032 pvd = vdev_add_parent(oldvd, pvops);
4033
4034 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4035 ASSERT(pvd->vdev_ops == pvops);
4036 ASSERT(oldvd->vdev_parent == pvd);
4037
4038 /*
4039 * Extract the new device from its root and add it to pvd.
4040 */
4041 vdev_remove_child(newrootvd, newvd);
4042 newvd->vdev_id = pvd->vdev_children;
4043 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4044 vdev_add_child(pvd, newvd);
4045
4046 tvd = newvd->vdev_top;
4047 ASSERT(pvd->vdev_top == tvd);
4048 ASSERT(tvd->vdev_parent == rvd);
4049
4050 vdev_config_dirty(tvd);
4051
4052 /*
4053 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4054 * for any dmu_sync-ed blocks. It will propagate upward when
4055 * spa_vdev_exit() calls vdev_dtl_reassess().
4056 */
4057 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4058
4059 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4060 dtl_max_txg - TXG_INITIAL);
4061
4062 if (newvd->vdev_isspare) {
4063 spa_spare_activate(newvd);
4064 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4065 }
4066
4067 oldvdpath = spa_strdup(oldvd->vdev_path);
4068 newvdpath = spa_strdup(newvd->vdev_path);
4069 newvd_isspare = newvd->vdev_isspare;
4070
4071 /*
4072 * Mark newvd's DTL dirty in this txg.
4073 */
4074 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4075
4076 /*
4077 * Restart the resilver
4078 */
4079 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4080
4081 /*
4082 * Commit the config
4083 */
4084 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4085
4086 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4087 "%s vdev=%s %s vdev=%s",
4088 replacing && newvd_isspare ? "spare in" :
4089 replacing ? "replace" : "attach", newvdpath,
4090 replacing ? "for" : "to", oldvdpath);
4091
4092 spa_strfree(oldvdpath);
4093 spa_strfree(newvdpath);
4094
4095 if (spa->spa_bootfs)
4096 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4097
4098 return (0);
4099}
4100
4101/*
4102 * Detach a device from a mirror or replacing vdev.
4103 * If 'replace_done' is specified, only detach if the parent
4104 * is a replacing vdev.
4105 */
4106int
4107spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4108{
4109 uint64_t txg;
4110 int error;
4111 vdev_t *rvd = spa->spa_root_vdev;
4112 vdev_t *vd, *pvd, *cvd, *tvd;
4113 boolean_t unspare = B_FALSE;
4114 uint64_t unspare_guid;
4115 char *vdpath;
4116
4117 ASSERT(spa_writeable(spa));
4118
4119 txg = spa_vdev_enter(spa);
4120
4121 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4122
4123 if (vd == NULL)
4124 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4125
4126 if (!vd->vdev_ops->vdev_op_leaf)
4127 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4128
4129 pvd = vd->vdev_parent;
4130
4131 /*
4132 * If the parent/child relationship is not as expected, don't do it.
4133 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4134 * vdev that's replacing B with C. The user's intent in replacing
4135 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4136 * the replace by detaching C, the expected behavior is to end up
4137 * M(A,B). But suppose that right after deciding to detach C,
4138 * the replacement of B completes. We would have M(A,C), and then
4139 * ask to detach C, which would leave us with just A -- not what
4140 * the user wanted. To prevent this, we make sure that the
4141 * parent/child relationship hasn't changed -- in this example,
4142 * that C's parent is still the replacing vdev R.
4143 */
4144 if (pvd->vdev_guid != pguid && pguid != 0)
4145 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4146
4147 /*
4148 * Only 'replacing' or 'spare' vdevs can be replaced.
4149 */
4150 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4151 pvd->vdev_ops != &vdev_spare_ops)
4152 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4153
4154 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4155 spa_version(spa) >= SPA_VERSION_SPARES);
4156
4157 /*
4158 * Only mirror, replacing, and spare vdevs support detach.
4159 */
4160 if (pvd->vdev_ops != &vdev_replacing_ops &&
4161 pvd->vdev_ops != &vdev_mirror_ops &&
4162 pvd->vdev_ops != &vdev_spare_ops)
4163 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4164
4165 /*
4166 * If this device has the only valid copy of some data,
4167 * we cannot safely detach it.
4168 */
4169 if (vdev_dtl_required(vd))
4170 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4171
4172 ASSERT(pvd->vdev_children >= 2);
4173
4174 /*
4175 * If we are detaching the second disk from a replacing vdev, then
4176 * check to see if we changed the original vdev's path to have "/old"
4177 * at the end in spa_vdev_attach(). If so, undo that change now.
4178 */
4179 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4180 vd->vdev_path != NULL) {
4181 size_t len = strlen(vd->vdev_path);
4182
4183 for (int c = 0; c < pvd->vdev_children; c++) {
4184 cvd = pvd->vdev_child[c];
4185
4186 if (cvd == vd || cvd->vdev_path == NULL)
4187 continue;
4188
4189 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4190 strcmp(cvd->vdev_path + len, "/old") == 0) {
4191 spa_strfree(cvd->vdev_path);
4192 cvd->vdev_path = spa_strdup(vd->vdev_path);
4193 break;
4194 }
4195 }
4196 }
4197
4198 /*
4199 * If we are detaching the original disk from a spare, then it implies
4200 * that the spare should become a real disk, and be removed from the
4201 * active spare list for the pool.
4202 */
4203 if (pvd->vdev_ops == &vdev_spare_ops &&
4204 vd->vdev_id == 0 &&
4205 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4206 unspare = B_TRUE;
4207
4208 /*
4209 * Erase the disk labels so the disk can be used for other things.
4210 * This must be done after all other error cases are handled,
4211 * but before we disembowel vd (so we can still do I/O to it).
4212 * But if we can't do it, don't treat the error as fatal --
4213 * it may be that the unwritability of the disk is the reason
4214 * it's being detached!
4215 */
4216 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4217
4218 /*
4219 * Remove vd from its parent and compact the parent's children.
4220 */
4221 vdev_remove_child(pvd, vd);
4222 vdev_compact_children(pvd);
4223
4224 /*
4225 * Remember one of the remaining children so we can get tvd below.
4226 */
4227 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4228
4229 /*
4230 * If we need to remove the remaining child from the list of hot spares,
4231 * do it now, marking the vdev as no longer a spare in the process.
4232 * We must do this before vdev_remove_parent(), because that can
4233 * change the GUID if it creates a new toplevel GUID. For a similar
4234 * reason, we must remove the spare now, in the same txg as the detach;
4235 * otherwise someone could attach a new sibling, change the GUID, and
4236 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4237 */
4238 if (unspare) {
4239 ASSERT(cvd->vdev_isspare);
4240 spa_spare_remove(cvd);
4241 unspare_guid = cvd->vdev_guid;
4242 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4243 cvd->vdev_unspare = B_TRUE;
4244 }
4245
4246 /*
4247 * If the parent mirror/replacing vdev only has one child,
4248 * the parent is no longer needed. Remove it from the tree.
4249 */
4250 if (pvd->vdev_children == 1) {
4251 if (pvd->vdev_ops == &vdev_spare_ops)
4252 cvd->vdev_unspare = B_FALSE;
4253 vdev_remove_parent(cvd);
4254 cvd->vdev_resilvering = B_FALSE;
4255 }
4256
4257
4258 /*
4259 * We don't set tvd until now because the parent we just removed
4260 * may have been the previous top-level vdev.
4261 */
4262 tvd = cvd->vdev_top;
4263 ASSERT(tvd->vdev_parent == rvd);
4264
4265 /*
4266 * Reevaluate the parent vdev state.
4267 */
4268 vdev_propagate_state(cvd);
4269
4270 /*
4271 * If the 'autoexpand' property is set on the pool then automatically
4272 * try to expand the size of the pool. For example if the device we
4273 * just detached was smaller than the others, it may be possible to
4274 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4275 * first so that we can obtain the updated sizes of the leaf vdevs.
4276 */
4277 if (spa->spa_autoexpand) {
4278 vdev_reopen(tvd);
4279 vdev_expand(tvd, txg);
4280 }
4281
4282 vdev_config_dirty(tvd);
4283
4284 /*
4285 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4286 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4287 * But first make sure we're not on any *other* txg's DTL list, to
4288 * prevent vd from being accessed after it's freed.
4289 */
4290 vdpath = spa_strdup(vd->vdev_path);
4291 for (int t = 0; t < TXG_SIZE; t++)
4292 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4293 vd->vdev_detached = B_TRUE;
4294 vdev_dirty(tvd, VDD_DTL, vd, txg);
4295
4296 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4297
4298 /* hang on to the spa before we release the lock */
4299 spa_open_ref(spa, FTAG);
4300
4301 error = spa_vdev_exit(spa, vd, txg, 0);
4302
4303 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4304 "vdev=%s", vdpath);
4305 spa_strfree(vdpath);
4306
4307 /*
4308 * If this was the removal of the original device in a hot spare vdev,
4309 * then we want to go through and remove the device from the hot spare
4310 * list of every other pool.
4311 */
4312 if (unspare) {
4313 spa_t *altspa = NULL;
4314
4315 mutex_enter(&spa_namespace_lock);
4316 while ((altspa = spa_next(altspa)) != NULL) {
4317 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4318 altspa == spa)
4319 continue;
4320
4321 spa_open_ref(altspa, FTAG);
4322 mutex_exit(&spa_namespace_lock);
4323 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4324 mutex_enter(&spa_namespace_lock);
4325 spa_close(altspa, FTAG);
4326 }
4327 mutex_exit(&spa_namespace_lock);
4328
4329 /* search the rest of the vdevs for spares to remove */
4330 spa_vdev_resilver_done(spa);
4331 }
4332
4333 /* all done with the spa; OK to release */
4334 mutex_enter(&spa_namespace_lock);
4335 spa_close(spa, FTAG);
4336 mutex_exit(&spa_namespace_lock);
4337
4338 return (error);
4339}
4340
4341/*
4342 * Split a set of devices from their mirrors, and create a new pool from them.
4343 */
4344int
4345spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4346 nvlist_t *props, boolean_t exp)
4347{
4348 int error = 0;
4349 uint64_t txg, *glist;
4350 spa_t *newspa;
4351 uint_t c, children, lastlog;
4352 nvlist_t **child, *nvl, *tmp;
4353 dmu_tx_t *tx;
4354 char *altroot = NULL;
4355 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4356 boolean_t activate_slog;
4357
4358 ASSERT(spa_writeable(spa));
4359
4360 txg = spa_vdev_enter(spa);
4361
4362 /* clear the log and flush everything up to now */
4363 activate_slog = spa_passivate_log(spa);
4364 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4365 error = spa_offline_log(spa);
4366 txg = spa_vdev_config_enter(spa);
4367
4368 if (activate_slog)
4369 spa_activate_log(spa);
4370
4371 if (error != 0)
4372 return (spa_vdev_exit(spa, NULL, txg, error));
4373
4374 /* check new spa name before going any further */
4375 if (spa_lookup(newname) != NULL)
4376 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4377
4378 /*
4379 * scan through all the children to ensure they're all mirrors
4380 */
4381 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4382 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4383 &children) != 0)
4384 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4385
4386 /* first, check to ensure we've got the right child count */
4387 rvd = spa->spa_root_vdev;
4388 lastlog = 0;
4389 for (c = 0; c < rvd->vdev_children; c++) {
4390 vdev_t *vd = rvd->vdev_child[c];
4391
4392 /* don't count the holes & logs as children */
4393 if (vd->vdev_islog || vd->vdev_ishole) {
4394 if (lastlog == 0)
4395 lastlog = c;
4396 continue;
4397 }
4398
4399 lastlog = 0;
4400 }
4401 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4402 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4403
4404 /* next, ensure no spare or cache devices are part of the split */
4405 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4406 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4407 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4408
4409 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4410 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4411
4412 /* then, loop over each vdev and validate it */
4413 for (c = 0; c < children; c++) {
4414 uint64_t is_hole = 0;
4415
4416 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4417 &is_hole);
4418
4419 if (is_hole != 0) {
4420 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4421 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4422 continue;
4423 } else {
4424 error = EINVAL;
4425 break;
4426 }
4427 }
4428
4429 /* which disk is going to be split? */
4430 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4431 &glist[c]) != 0) {
4432 error = EINVAL;
4433 break;
4434 }
4435
4436 /* look it up in the spa */
4437 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4438 if (vml[c] == NULL) {
4439 error = ENODEV;
4440 break;
4441 }
4442
4443 /* make sure there's nothing stopping the split */
4444 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4445 vml[c]->vdev_islog ||
4446 vml[c]->vdev_ishole ||
4447 vml[c]->vdev_isspare ||
4448 vml[c]->vdev_isl2cache ||
4449 !vdev_writeable(vml[c]) ||
4450 vml[c]->vdev_children != 0 ||
4451 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4452 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4453 error = EINVAL;
4454 break;
4455 }
4456
4457 if (vdev_dtl_required(vml[c])) {
4458 error = EBUSY;
4459 break;
4460 }
4461
4462 /* we need certain info from the top level */
4463 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4464 vml[c]->vdev_top->vdev_ms_array) == 0);
4465 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4466 vml[c]->vdev_top->vdev_ms_shift) == 0);
4467 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4468 vml[c]->vdev_top->vdev_asize) == 0);
4469 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4470 vml[c]->vdev_top->vdev_ashift) == 0);
4471 }
4472
4473 if (error != 0) {
4474 kmem_free(vml, children * sizeof (vdev_t *));
4475 kmem_free(glist, children * sizeof (uint64_t));
4476 return (spa_vdev_exit(spa, NULL, txg, error));
4477 }
4478
4479 /* stop writers from using the disks */
4480 for (c = 0; c < children; c++) {
4481 if (vml[c] != NULL)
4482 vml[c]->vdev_offline = B_TRUE;
4483 }
4484 vdev_reopen(spa->spa_root_vdev);
4485
4486 /*
4487 * Temporarily record the splitting vdevs in the spa config. This
4488 * will disappear once the config is regenerated.
4489 */
4490 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4491 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4492 glist, children) == 0);
4493 kmem_free(glist, children * sizeof (uint64_t));
4494
4495 mutex_enter(&spa->spa_props_lock);
4496 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4497 nvl) == 0);
4498 mutex_exit(&spa->spa_props_lock);
4499 spa->spa_config_splitting = nvl;
4500 vdev_config_dirty(spa->spa_root_vdev);
4501
4502 /* configure and create the new pool */
4503 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4504 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4505 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4506 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4507 spa_version(spa)) == 0);
4508 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4509 spa->spa_config_txg) == 0);
4510 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4511 spa_generate_guid(NULL)) == 0);
4512 (void) nvlist_lookup_string(props,
4513 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4514
4515 /* add the new pool to the namespace */
4516 newspa = spa_add(newname, config, altroot);
4517 newspa->spa_config_txg = spa->spa_config_txg;
4518 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4519
4520 /* release the spa config lock, retaining the namespace lock */
4521 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4522
4523 if (zio_injection_enabled)
4524 zio_handle_panic_injection(spa, FTAG, 1);
4525
4526 spa_activate(newspa, spa_mode_global);
4527 spa_async_suspend(newspa);
4528
4529#ifndef sun
4530 /* mark that we are creating new spa by splitting */
4531 newspa->spa_splitting_newspa = B_TRUE;
4532#endif
4533 /* create the new pool from the disks of the original pool */
4534 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4535#ifndef sun
4536 newspa->spa_splitting_newspa = B_FALSE;
4537#endif
4538 if (error)
4539 goto out;
4540
4541 /* if that worked, generate a real config for the new pool */
4542 if (newspa->spa_root_vdev != NULL) {
4543 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4544 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4545 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4546 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4547 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4548 B_TRUE));
4549 }
4550
4551 /* set the props */
4552 if (props != NULL) {
4553 spa_configfile_set(newspa, props, B_FALSE);
4554 error = spa_prop_set(newspa, props);
4555 if (error)
4556 goto out;
4557 }
4558
4559 /* flush everything */
4560 txg = spa_vdev_config_enter(newspa);
4561 vdev_config_dirty(newspa->spa_root_vdev);
4562 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4563
4564 if (zio_injection_enabled)
4565 zio_handle_panic_injection(spa, FTAG, 2);
4566
4567 spa_async_resume(newspa);
4568
4569 /* finally, update the original pool's config */
4570 txg = spa_vdev_config_enter(spa);
4571 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4572 error = dmu_tx_assign(tx, TXG_WAIT);
4573 if (error != 0)
4574 dmu_tx_abort(tx);
4575 for (c = 0; c < children; c++) {
4576 if (vml[c] != NULL) {
4577 vdev_split(vml[c]);
4578 if (error == 0)
4579 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4580 spa, tx, "vdev=%s",
4581 vml[c]->vdev_path);
4582 vdev_free(vml[c]);
4583 }
4584 }
4585 vdev_config_dirty(spa->spa_root_vdev);
4586 spa->spa_config_splitting = NULL;
4587 nvlist_free(nvl);
4588 if (error == 0)
4589 dmu_tx_commit(tx);
4590 (void) spa_vdev_exit(spa, NULL, txg, 0);
4591
4592 if (zio_injection_enabled)
4593 zio_handle_panic_injection(spa, FTAG, 3);
4594
4595 /* split is complete; log a history record */
4596 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4597 "split new pool %s from pool %s", newname, spa_name(spa));
4598
4599 kmem_free(vml, children * sizeof (vdev_t *));
4600
4601 /* if we're not going to mount the filesystems in userland, export */
4602 if (exp)
4603 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4604 B_FALSE, B_FALSE);
4605
4606 return (error);
4607
4608out:
4609 spa_unload(newspa);
4610 spa_deactivate(newspa);
4611 spa_remove(newspa);
4612
4613 txg = spa_vdev_config_enter(spa);
4614
4615 /* re-online all offlined disks */
4616 for (c = 0; c < children; c++) {
4617 if (vml[c] != NULL)
4618 vml[c]->vdev_offline = B_FALSE;
4619 }
4620 vdev_reopen(spa->spa_root_vdev);
4621
4622 nvlist_free(spa->spa_config_splitting);
4623 spa->spa_config_splitting = NULL;
4624 (void) spa_vdev_exit(spa, NULL, txg, error);
4625
4626 kmem_free(vml, children * sizeof (vdev_t *));
4627 return (error);
4628}
4629
4630static nvlist_t *
4631spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4632{
4633 for (int i = 0; i < count; i++) {
4634 uint64_t guid;
4635
4636 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4637 &guid) == 0);
4638
4639 if (guid == target_guid)
4640 return (nvpp[i]);
4641 }
4642
4643 return (NULL);
4644}
4645
4646static void
4647spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4648 nvlist_t *dev_to_remove)
4649{
4650 nvlist_t **newdev = NULL;
4651
4652 if (count > 1)
4653 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4654
4655 for (int i = 0, j = 0; i < count; i++) {
4656 if (dev[i] == dev_to_remove)
4657 continue;
4658 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4659 }
4660
4661 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4662 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4663
4664 for (int i = 0; i < count - 1; i++)
4665 nvlist_free(newdev[i]);
4666
4667 if (count > 1)
4668 kmem_free(newdev, (count - 1) * sizeof (void *));
4669}
4670
4671/*
4672 * Evacuate the device.
4673 */
4674static int
4675spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4676{
4677 uint64_t txg;
4678 int error = 0;
4679
4680 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4681 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4682 ASSERT(vd == vd->vdev_top);
4683
4684 /*
4685 * Evacuate the device. We don't hold the config lock as writer
4686 * since we need to do I/O but we do keep the
4687 * spa_namespace_lock held. Once this completes the device
4688 * should no longer have any blocks allocated on it.
4689 */
4690 if (vd->vdev_islog) {
4691 if (vd->vdev_stat.vs_alloc != 0)
4692 error = spa_offline_log(spa);
4693 } else {
4694 error = ENOTSUP;
4695 }
4696
4697 if (error)
4698 return (error);
4699
4700 /*
4701 * The evacuation succeeded. Remove any remaining MOS metadata
4702 * associated with this vdev, and wait for these changes to sync.
4703 */
4704 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4705 txg = spa_vdev_config_enter(spa);
4706 vd->vdev_removing = B_TRUE;
4707 vdev_dirty(vd, 0, NULL, txg);
4708 vdev_config_dirty(vd);
4709 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4710
4711 return (0);
4712}
4713
4714/*
4715 * Complete the removal by cleaning up the namespace.
4716 */
4717static void
4718spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4719{
4720 vdev_t *rvd = spa->spa_root_vdev;
4721 uint64_t id = vd->vdev_id;
4722 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4723
4724 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4725 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4726 ASSERT(vd == vd->vdev_top);
4727
4728 /*
4729 * Only remove any devices which are empty.
4730 */
4731 if (vd->vdev_stat.vs_alloc != 0)
4732 return;
4733
4734 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4735
4736 if (list_link_active(&vd->vdev_state_dirty_node))
4737 vdev_state_clean(vd);
4738 if (list_link_active(&vd->vdev_config_dirty_node))
4739 vdev_config_clean(vd);
4740
4741 vdev_free(vd);
4742
4743 if (last_vdev) {
4744 vdev_compact_children(rvd);
4745 } else {
4746 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4747 vdev_add_child(rvd, vd);
4748 }
4749 vdev_config_dirty(rvd);
4750
4751 /*
4752 * Reassess the health of our root vdev.
4753 */
4754 vdev_reopen(rvd);
4755}
4756
4757/*
4758 * Remove a device from the pool -
4759 *
4760 * Removing a device from the vdev namespace requires several steps
4761 * and can take a significant amount of time. As a result we use
4762 * the spa_vdev_config_[enter/exit] functions which allow us to
4763 * grab and release the spa_config_lock while still holding the namespace
4764 * lock. During each step the configuration is synced out.
4765 */
4766
4767/*
4768 * Remove a device from the pool. Currently, this supports removing only hot
4769 * spares, slogs, and level 2 ARC devices.
4770 */
4771int
4772spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4773{
4774 vdev_t *vd;
4775 metaslab_group_t *mg;
4776 nvlist_t **spares, **l2cache, *nv;
4777 uint64_t txg = 0;
4778 uint_t nspares, nl2cache;
4779 int error = 0;
4780 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4781
4782 ASSERT(spa_writeable(spa));
4783
4784 if (!locked)
4785 txg = spa_vdev_enter(spa);
4786
4787 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4788
4789 if (spa->spa_spares.sav_vdevs != NULL &&
4790 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4791 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4792 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4793 /*
4794 * Only remove the hot spare if it's not currently in use
4795 * in this pool.
4796 */
4797 if (vd == NULL || unspare) {
4798 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4799 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4800 spa_load_spares(spa);
4801 spa->spa_spares.sav_sync = B_TRUE;
4802 } else {
4803 error = EBUSY;
4804 }
4805 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4806 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4807 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4808 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4809 /*
4810 * Cache devices can always be removed.
4811 */
4812 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4813 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4814 spa_load_l2cache(spa);
4815 spa->spa_l2cache.sav_sync = B_TRUE;
4816 } else if (vd != NULL && vd->vdev_islog) {
4817 ASSERT(!locked);
4818 ASSERT(vd == vd->vdev_top);
4819
4820 /*
4821 * XXX - Once we have bp-rewrite this should
4822 * become the common case.
4823 */
4824
4825 mg = vd->vdev_mg;
4826
4827 /*
4828 * Stop allocating from this vdev.
4829 */
4830 metaslab_group_passivate(mg);
4831
4832 /*
4833 * Wait for the youngest allocations and frees to sync,
4834 * and then wait for the deferral of those frees to finish.
4835 */
4836 spa_vdev_config_exit(spa, NULL,
4837 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4838
4839 /*
4840 * Attempt to evacuate the vdev.
4841 */
4842 error = spa_vdev_remove_evacuate(spa, vd);
4843
4844 txg = spa_vdev_config_enter(spa);
4845
4846 /*
4847 * If we couldn't evacuate the vdev, unwind.
4848 */
4849 if (error) {
4850 metaslab_group_activate(mg);
4851 return (spa_vdev_exit(spa, NULL, txg, error));
4852 }
4853
4854 /*
4855 * Clean up the vdev namespace.
4856 */
4857 spa_vdev_remove_from_namespace(spa, vd);
4858
4859 } else if (vd != NULL) {
4860 /*
4861 * Normal vdevs cannot be removed (yet).
4862 */
4863 error = ENOTSUP;
4864 } else {
4865 /*
4866 * There is no vdev of any kind with the specified guid.
4867 */
4868 error = ENOENT;
4869 }
4870
4871 if (!locked)
4872 return (spa_vdev_exit(spa, NULL, txg, error));
4873
4874 return (error);
4875}
4876
4877/*
4878 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4879 * current spared, so we can detach it.
4880 */
4881static vdev_t *
4882spa_vdev_resilver_done_hunt(vdev_t *vd)
4883{
4884 vdev_t *newvd, *oldvd;
4885
4886 for (int c = 0; c < vd->vdev_children; c++) {
4887 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4888 if (oldvd != NULL)
4889 return (oldvd);
4890 }
4891
4892 /*
4893 * Check for a completed replacement. We always consider the first
4894 * vdev in the list to be the oldest vdev, and the last one to be
4895 * the newest (see spa_vdev_attach() for how that works). In
4896 * the case where the newest vdev is faulted, we will not automatically
4897 * remove it after a resilver completes. This is OK as it will require
4898 * user intervention to determine which disk the admin wishes to keep.
4899 */
4900 if (vd->vdev_ops == &vdev_replacing_ops) {
4901 ASSERT(vd->vdev_children > 1);
4902
4903 newvd = vd->vdev_child[vd->vdev_children - 1];
4904 oldvd = vd->vdev_child[0];
4905
4906 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4907 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4908 !vdev_dtl_required(oldvd))
4909 return (oldvd);
4910 }
4911
4912 /*
4913 * Check for a completed resilver with the 'unspare' flag set.
4914 */
4915 if (vd->vdev_ops == &vdev_spare_ops) {
4916 vdev_t *first = vd->vdev_child[0];
4917 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4918
4919 if (last->vdev_unspare) {
4920 oldvd = first;
4921 newvd = last;
4922 } else if (first->vdev_unspare) {
4923 oldvd = last;
4924 newvd = first;
4925 } else {
4926 oldvd = NULL;
4927 }
4928
4929 if (oldvd != NULL &&
4930 vdev_dtl_empty(newvd, DTL_MISSING) &&
4931 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4932 !vdev_dtl_required(oldvd))
4933 return (oldvd);
4934
4935 /*
4936 * If there are more than two spares attached to a disk,
4937 * and those spares are not required, then we want to
4938 * attempt to free them up now so that they can be used
4939 * by other pools. Once we're back down to a single
4940 * disk+spare, we stop removing them.
4941 */
4942 if (vd->vdev_children > 2) {
4943 newvd = vd->vdev_child[1];
4944
4945 if (newvd->vdev_isspare && last->vdev_isspare &&
4946 vdev_dtl_empty(last, DTL_MISSING) &&
4947 vdev_dtl_empty(last, DTL_OUTAGE) &&
4948 !vdev_dtl_required(newvd))
4949 return (newvd);
4950 }
4951 }
4952
4953 return (NULL);
4954}
4955
4956static void
4957spa_vdev_resilver_done(spa_t *spa)
4958{
4959 vdev_t *vd, *pvd, *ppvd;
4960 uint64_t guid, sguid, pguid, ppguid;
4961
4962 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4963
4964 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4965 pvd = vd->vdev_parent;
4966 ppvd = pvd->vdev_parent;
4967 guid = vd->vdev_guid;
4968 pguid = pvd->vdev_guid;
4969 ppguid = ppvd->vdev_guid;
4970 sguid = 0;
4971 /*
4972 * If we have just finished replacing a hot spared device, then
4973 * we need to detach the parent's first child (the original hot
4974 * spare) as well.
4975 */
4976 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4977 ppvd->vdev_children == 2) {
4978 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4979 sguid = ppvd->vdev_child[1]->vdev_guid;
4980 }
4981 spa_config_exit(spa, SCL_ALL, FTAG);
4982 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4983 return;
4984 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4985 return;
4986 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4987 }
4988
4989 spa_config_exit(spa, SCL_ALL, FTAG);
4990}
4991
4992/*
4993 * Update the stored path or FRU for this vdev.
4994 */
4995int
4996spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4997 boolean_t ispath)
4998{
4999 vdev_t *vd;
5000 boolean_t sync = B_FALSE;
5001
5002 ASSERT(spa_writeable(spa));
5003
5004 spa_vdev_state_enter(spa, SCL_ALL);
5005
5006 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5007 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5008
5009 if (!vd->vdev_ops->vdev_op_leaf)
5010 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5011
5012 if (ispath) {
5013 if (strcmp(value, vd->vdev_path) != 0) {
5014 spa_strfree(vd->vdev_path);
5015 vd->vdev_path = spa_strdup(value);
5016 sync = B_TRUE;
5017 }
5018 } else {
5019 if (vd->vdev_fru == NULL) {
5020 vd->vdev_fru = spa_strdup(value);
5021 sync = B_TRUE;
5022 } else if (strcmp(value, vd->vdev_fru) != 0) {
5023 spa_strfree(vd->vdev_fru);
5024 vd->vdev_fru = spa_strdup(value);
5025 sync = B_TRUE;
5026 }
5027 }
5028
5029 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5030}
5031
5032int
5033spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5034{
5035 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5036}
5037
5038int
5039spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5040{
5041 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5042}
5043
5044/*
5045 * ==========================================================================
5046 * SPA Scanning
5047 * ==========================================================================
5048 */
5049
5050int
5051spa_scan_stop(spa_t *spa)
5052{
5053 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5054 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5055 return (EBUSY);
5056 return (dsl_scan_cancel(spa->spa_dsl_pool));
5057}
5058
5059int
5060spa_scan(spa_t *spa, pool_scan_func_t func)
5061{
5062 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5063
5064 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5065 return (ENOTSUP);
5066
5067 /*
5068 * If a resilver was requested, but there is no DTL on a
5069 * writeable leaf device, we have nothing to do.
5070 */
5071 if (func == POOL_SCAN_RESILVER &&
5072 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5073 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5074 return (0);
5075 }
5076
5077 return (dsl_scan(spa->spa_dsl_pool, func));
5078}
5079
5080/*
5081 * ==========================================================================
5082 * SPA async task processing
5083 * ==========================================================================
5084 */
5085
5086static void
5087spa_async_remove(spa_t *spa, vdev_t *vd)
5088{
5089 if (vd->vdev_remove_wanted) {
5090 vd->vdev_remove_wanted = B_FALSE;
5091 vd->vdev_delayed_close = B_FALSE;
5092 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5093
5094 /*
5095 * We want to clear the stats, but we don't want to do a full
5096 * vdev_clear() as that will cause us to throw away
5097 * degraded/faulted state as well as attempt to reopen the
5098 * device, all of which is a waste.
5099 */
5100 vd->vdev_stat.vs_read_errors = 0;
5101 vd->vdev_stat.vs_write_errors = 0;
5102 vd->vdev_stat.vs_checksum_errors = 0;
5103
5104 vdev_state_dirty(vd->vdev_top);
5105 }
5106
5107 for (int c = 0; c < vd->vdev_children; c++)
5108 spa_async_remove(spa, vd->vdev_child[c]);
5109}
5110
5111static void
5112spa_async_probe(spa_t *spa, vdev_t *vd)
5113{
5114 if (vd->vdev_probe_wanted) {
5115 vd->vdev_probe_wanted = B_FALSE;
5116 vdev_reopen(vd); /* vdev_open() does the actual probe */
5117 }
5118
5119 for (int c = 0; c < vd->vdev_children; c++)
5120 spa_async_probe(spa, vd->vdev_child[c]);
5121}
5122
5123static void
5124spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5125{
5126 sysevent_id_t eid;
5127 nvlist_t *attr;
5128 char *physpath;
5129
5130 if (!spa->spa_autoexpand)
5131 return;
5132
5133 for (int c = 0; c < vd->vdev_children; c++) {
5134 vdev_t *cvd = vd->vdev_child[c];
5135 spa_async_autoexpand(spa, cvd);
5136 }
5137
5138 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5139 return;
5140
5141 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5142 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5143
5144 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5145 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5146
5147 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5148 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5149
5150 nvlist_free(attr);
5151 kmem_free(physpath, MAXPATHLEN);
5152}
5153
5154static void
5155spa_async_thread(void *arg)
5156{
5157 spa_t *spa = arg;
5158 int tasks;
5159
5160 ASSERT(spa->spa_sync_on);
5161
5162 mutex_enter(&spa->spa_async_lock);
5163 tasks = spa->spa_async_tasks;
5164 spa->spa_async_tasks = 0;
5165 mutex_exit(&spa->spa_async_lock);
5166
5167 /*
5168 * See if the config needs to be updated.
5169 */
5170 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5171 uint64_t old_space, new_space;
5172
5173 mutex_enter(&spa_namespace_lock);
5174 old_space = metaslab_class_get_space(spa_normal_class(spa));
5175 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5176 new_space = metaslab_class_get_space(spa_normal_class(spa));
5177 mutex_exit(&spa_namespace_lock);
5178
5179 /*
5180 * If the pool grew as a result of the config update,
5181 * then log an internal history event.
5182 */
5183 if (new_space != old_space) {
5184 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5185 spa, NULL,
5186 "pool '%s' size: %llu(+%llu)",
5187 spa_name(spa), new_space, new_space - old_space);
5188 }
5189 }
5190
5191 /*
5192 * See if any devices need to be marked REMOVED.
5193 */
5194 if (tasks & SPA_ASYNC_REMOVE) {
5195 spa_vdev_state_enter(spa, SCL_NONE);
5196 spa_async_remove(spa, spa->spa_root_vdev);
5197 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5198 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5199 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5200 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5201 (void) spa_vdev_state_exit(spa, NULL, 0);
5202 }
5203
5204 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5205 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5206 spa_async_autoexpand(spa, spa->spa_root_vdev);
5207 spa_config_exit(spa, SCL_CONFIG, FTAG);
5208 }
5209
5210 /*
5211 * See if any devices need to be probed.
5212 */
5213 if (tasks & SPA_ASYNC_PROBE) {
5214 spa_vdev_state_enter(spa, SCL_NONE);
5215 spa_async_probe(spa, spa->spa_root_vdev);
5216 (void) spa_vdev_state_exit(spa, NULL, 0);
5217 }
5218
5219 /*
5220 * If any devices are done replacing, detach them.
5221 */
5222 if (tasks & SPA_ASYNC_RESILVER_DONE)
5223 spa_vdev_resilver_done(spa);
5224
5225 /*
5226 * Kick off a resilver.
5227 */
5228 if (tasks & SPA_ASYNC_RESILVER)
5229 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5230
5231 /*
5232 * Let the world know that we're done.
5233 */
5234 mutex_enter(&spa->spa_async_lock);
5235 spa->spa_async_thread = NULL;
5236 cv_broadcast(&spa->spa_async_cv);
5237 mutex_exit(&spa->spa_async_lock);
5238 thread_exit();
5239}
5240
5241void
5242spa_async_suspend(spa_t *spa)
5243{
5244 mutex_enter(&spa->spa_async_lock);
5245 spa->spa_async_suspended++;
5246 while (spa->spa_async_thread != NULL)
5247 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5248 mutex_exit(&spa->spa_async_lock);
5249}
5250
5251void
5252spa_async_resume(spa_t *spa)
5253{
5254 mutex_enter(&spa->spa_async_lock);
5255 ASSERT(spa->spa_async_suspended != 0);
5256 spa->spa_async_suspended--;
5257 mutex_exit(&spa->spa_async_lock);
5258}
5259
5260static void
5261spa_async_dispatch(spa_t *spa)
5262{
5263 mutex_enter(&spa->spa_async_lock);
5264 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5265 spa->spa_async_thread == NULL &&
5266 rootdir != NULL && !vn_is_readonly(rootdir))
5267 spa->spa_async_thread = thread_create(NULL, 0,
5268 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5269 mutex_exit(&spa->spa_async_lock);
5270}
5271
5272void
5273spa_async_request(spa_t *spa, int task)
5274{
5275 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5276 mutex_enter(&spa->spa_async_lock);
5277 spa->spa_async_tasks |= task;
5278 mutex_exit(&spa->spa_async_lock);
5279}
5280
5281/*
5282 * ==========================================================================
5283 * SPA syncing routines
5284 * ==========================================================================
5285 */
5286
5287static int
5288bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5289{
5290 bpobj_t *bpo = arg;
5291 bpobj_enqueue(bpo, bp, tx);
5292 return (0);
5293}
5294
5295static int
5296spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5297{
5298 zio_t *zio = arg;
5299
5300 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5301 zio->io_flags));
5302 return (0);
5303}
5304
5305static void
5306spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5307{
5308 char *packed = NULL;
5309 size_t bufsize;
5310 size_t nvsize = 0;
5311 dmu_buf_t *db;
5312
5313 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5314
5315 /*
5316 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5317 * information. This avoids the dbuf_will_dirty() path and
5318 * saves us a pre-read to get data we don't actually care about.
5319 */
5320 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5321 packed = kmem_alloc(bufsize, KM_SLEEP);
5322
5323 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5324 KM_SLEEP) == 0);
5325 bzero(packed + nvsize, bufsize - nvsize);
5326
5327 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5328
5329 kmem_free(packed, bufsize);
5330
5331 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5332 dmu_buf_will_dirty(db, tx);
5333 *(uint64_t *)db->db_data = nvsize;
5334 dmu_buf_rele(db, FTAG);
5335}
5336
5337static void
5338spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5339 const char *config, const char *entry)
5340{
5341 nvlist_t *nvroot;
5342 nvlist_t **list;
5343 int i;
5344
5345 if (!sav->sav_sync)
5346 return;
5347
5348 /*
5349 * Update the MOS nvlist describing the list of available devices.
5350 * spa_validate_aux() will have already made sure this nvlist is
5351 * valid and the vdevs are labeled appropriately.
5352 */
5353 if (sav->sav_object == 0) {
5354 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5355 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5356 sizeof (uint64_t), tx);
5357 VERIFY(zap_update(spa->spa_meta_objset,
5358 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5359 &sav->sav_object, tx) == 0);
5360 }
5361
5362 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5363 if (sav->sav_count == 0) {
5364 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5365 } else {
5366 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5367 for (i = 0; i < sav->sav_count; i++)
5368 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5369 B_FALSE, VDEV_CONFIG_L2CACHE);
5370 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5371 sav->sav_count) == 0);
5372 for (i = 0; i < sav->sav_count; i++)
5373 nvlist_free(list[i]);
5374 kmem_free(list, sav->sav_count * sizeof (void *));
5375 }
5376
5377 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5378 nvlist_free(nvroot);
5379
5380 sav->sav_sync = B_FALSE;
5381}
5382
5383static void
5384spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5385{
5386 nvlist_t *config;
5387
5388 if (list_is_empty(&spa->spa_config_dirty_list))
5389 return;
5390
5391 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5392
5393 config = spa_config_generate(spa, spa->spa_root_vdev,
5394 dmu_tx_get_txg(tx), B_FALSE);
5395
5396 spa_config_exit(spa, SCL_STATE, FTAG);
5397
5398 if (spa->spa_config_syncing)
5399 nvlist_free(spa->spa_config_syncing);
5400 spa->spa_config_syncing = config;
5401
5402 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5403}
5404
5405/*
5406 * Set zpool properties.
5407 */
5408static void
5409spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5410{
5411 spa_t *spa = arg1;
5412 objset_t *mos = spa->spa_meta_objset;
5413 nvlist_t *nvp = arg2;
5414 nvpair_t *elem;
5415 uint64_t intval;
5416 char *strval;
5417 zpool_prop_t prop;
5418 const char *propname;
5419 zprop_type_t proptype;
5420
5421 mutex_enter(&spa->spa_props_lock);
5422
5423 elem = NULL;
5424 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5425 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5426 case ZPOOL_PROP_VERSION:
5427 /*
5428 * Only set version for non-zpool-creation cases
5429 * (set/import). spa_create() needs special care
5430 * for version setting.
5431 */
5432 if (tx->tx_txg != TXG_INITIAL) {
5433 VERIFY(nvpair_value_uint64(elem,
5434 &intval) == 0);
5435 ASSERT(intval <= SPA_VERSION);
5436 ASSERT(intval >= spa_version(spa));
5437 spa->spa_uberblock.ub_version = intval;
5438 vdev_config_dirty(spa->spa_root_vdev);
5439 }
5440 break;
5441
5442 case ZPOOL_PROP_ALTROOT:
5443 /*
5444 * 'altroot' is a non-persistent property. It should
5445 * have been set temporarily at creation or import time.
5446 */
5447 ASSERT(spa->spa_root != NULL);
5448 break;
5449
5450 case ZPOOL_PROP_READONLY:
5451 case ZPOOL_PROP_CACHEFILE:
5452 /*
5453 * 'readonly' and 'cachefile' are also non-persisitent
5454 * properties.
5455 */
5456 break;
5457 case ZPOOL_PROP_COMMENT:
5458 VERIFY(nvpair_value_string(elem, &strval) == 0);
5459 if (spa->spa_comment != NULL)
5460 spa_strfree(spa->spa_comment);
5461 spa->spa_comment = spa_strdup(strval);
5462 /*
5463 * We need to dirty the configuration on all the vdevs
5464 * so that their labels get updated. It's unnecessary
5465 * to do this for pool creation since the vdev's
5466 * configuratoin has already been dirtied.
5467 */
5468 if (tx->tx_txg != TXG_INITIAL)
5469 vdev_config_dirty(spa->spa_root_vdev);
5470 break;
5471 default:
5472 /*
5473 * Set pool property values in the poolprops mos object.
5474 */
5475 if (spa->spa_pool_props_object == 0) {
5476 VERIFY((spa->spa_pool_props_object =
5477 zap_create(mos, DMU_OT_POOL_PROPS,
5478 DMU_OT_NONE, 0, tx)) > 0);
5479
5480 VERIFY(zap_update(mos,
5481 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5482 8, 1, &spa->spa_pool_props_object, tx)
5483 == 0);
5484 }
5485
5486 /* normalize the property name */
5487 propname = zpool_prop_to_name(prop);
5488 proptype = zpool_prop_get_type(prop);
5489
5490 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5491 ASSERT(proptype == PROP_TYPE_STRING);
5492 VERIFY(nvpair_value_string(elem, &strval) == 0);
5493 VERIFY(zap_update(mos,
5494 spa->spa_pool_props_object, propname,
5495 1, strlen(strval) + 1, strval, tx) == 0);
5496
5497 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5498 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5499
5500 if (proptype == PROP_TYPE_INDEX) {
5501 const char *unused;
5502 VERIFY(zpool_prop_index_to_string(
5503 prop, intval, &unused) == 0);
5504 }
5505 VERIFY(zap_update(mos,
5506 spa->spa_pool_props_object, propname,
5507 8, 1, &intval, tx) == 0);
5508 } else {
5509 ASSERT(0); /* not allowed */
5510 }
5511
5512 switch (prop) {
5513 case ZPOOL_PROP_DELEGATION:
5514 spa->spa_delegation = intval;
5515 break;
5516 case ZPOOL_PROP_BOOTFS:
5517 spa->spa_bootfs = intval;
5518 break;
5519 case ZPOOL_PROP_FAILUREMODE:
5520 spa->spa_failmode = intval;
5521 break;
5522 case ZPOOL_PROP_AUTOEXPAND:
5523 spa->spa_autoexpand = intval;
5524 if (tx->tx_txg != TXG_INITIAL)
5525 spa_async_request(spa,
5526 SPA_ASYNC_AUTOEXPAND);
5527 break;
5528 case ZPOOL_PROP_DEDUPDITTO:
5529 spa->spa_dedup_ditto = intval;
5530 break;
5531 default:
5532 break;
5533 }
5534 }
5535
5536 /* log internal history if this is not a zpool create */
5537 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5538 tx->tx_txg != TXG_INITIAL) {
5539 spa_history_log_internal(LOG_POOL_PROPSET,
5540 spa, tx, "%s %lld %s",
5541 nvpair_name(elem), intval, spa_name(spa));
5542 }
5543 }
5544
5545 mutex_exit(&spa->spa_props_lock);
5546}
5547
5548/*
5549 * Perform one-time upgrade on-disk changes. spa_version() does not
5550 * reflect the new version this txg, so there must be no changes this
5551 * txg to anything that the upgrade code depends on after it executes.
5552 * Therefore this must be called after dsl_pool_sync() does the sync
5553 * tasks.
5554 */
5555static void
5556spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5557{
5558 dsl_pool_t *dp = spa->spa_dsl_pool;
5559
5560 ASSERT(spa->spa_sync_pass == 1);
5561
5562 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5563 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5564 dsl_pool_create_origin(dp, tx);
5565
5566 /* Keeping the origin open increases spa_minref */
5567 spa->spa_minref += 3;
5568 }
5569
5570 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5571 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5572 dsl_pool_upgrade_clones(dp, tx);
5573 }
5574
5575 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5576 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5577 dsl_pool_upgrade_dir_clones(dp, tx);
5578
5579 /* Keeping the freedir open increases spa_minref */
5580 spa->spa_minref += 3;
5581 }
5582}
5583
5584/*
5585 * Sync the specified transaction group. New blocks may be dirtied as
5586 * part of the process, so we iterate until it converges.
5587 */
5588void
5589spa_sync(spa_t *spa, uint64_t txg)
5590{
5591 dsl_pool_t *dp = spa->spa_dsl_pool;
5592 objset_t *mos = spa->spa_meta_objset;
5593 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5594 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5595 vdev_t *rvd = spa->spa_root_vdev;
5596 vdev_t *vd;
5597 dmu_tx_t *tx;
5598 int error;
5599
5600 VERIFY(spa_writeable(spa));
5601
5602 /*
5603 * Lock out configuration changes.
5604 */
5605 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5606
5607 spa->spa_syncing_txg = txg;
5608 spa->spa_sync_pass = 0;
5609
5610 /*
5611 * If there are any pending vdev state changes, convert them
5612 * into config changes that go out with this transaction group.
5613 */
5614 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5615 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5616 /*
5617 * We need the write lock here because, for aux vdevs,
5618 * calling vdev_config_dirty() modifies sav_config.
5619 * This is ugly and will become unnecessary when we
5620 * eliminate the aux vdev wart by integrating all vdevs
5621 * into the root vdev tree.
5622 */
5623 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5624 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5625 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5626 vdev_state_clean(vd);
5627 vdev_config_dirty(vd);
5628 }
5629 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5630 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5631 }
5632 spa_config_exit(spa, SCL_STATE, FTAG);
5633
5634 tx = dmu_tx_create_assigned(dp, txg);
5635
5636 /*
5637 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5638 * set spa_deflate if we have no raid-z vdevs.
5639 */
5640 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5641 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5642 int i;
5643
5644 for (i = 0; i < rvd->vdev_children; i++) {
5645 vd = rvd->vdev_child[i];
5646 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5647 break;
5648 }
5649 if (i == rvd->vdev_children) {
5650 spa->spa_deflate = TRUE;
5651 VERIFY(0 == zap_add(spa->spa_meta_objset,
5652 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5653 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5654 }
5655 }
5656
5657 /*
5658 * If anything has changed in this txg, or if someone is waiting
5659 * for this txg to sync (eg, spa_vdev_remove()), push the
5660 * deferred frees from the previous txg. If not, leave them
5661 * alone so that we don't generate work on an otherwise idle
5662 * system.
5663 */
5664 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5665 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5666 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5667 ((dsl_scan_active(dp->dp_scan) ||
5668 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5669 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5670 VERIFY3U(bpobj_iterate(defer_bpo,
5671 spa_free_sync_cb, zio, tx), ==, 0);
5672 VERIFY3U(zio_wait(zio), ==, 0);
5673 }
5674
5675 /*
5676 * Iterate to convergence.
5677 */
5678 do {
5679 int pass = ++spa->spa_sync_pass;
5680
5681 spa_sync_config_object(spa, tx);
5682 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5683 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5684 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5685 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5686 spa_errlog_sync(spa, txg);
5687 dsl_pool_sync(dp, txg);
5688
5689 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5690 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5691 bplist_iterate(free_bpl, spa_free_sync_cb,
5692 zio, tx);
5693 VERIFY(zio_wait(zio) == 0);
5694 } else {
5695 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5696 defer_bpo, tx);
5697 }
5698
5699 ddt_sync(spa, txg);
5700 dsl_scan_sync(dp, tx);
5701
5702 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5703 vdev_sync(vd, txg);
5704
5705 if (pass == 1)
5706 spa_sync_upgrades(spa, tx);
5707
5708 } while (dmu_objset_is_dirty(mos, txg));
5709
5710 /*
5711 * Rewrite the vdev configuration (which includes the uberblock)
5712 * to commit the transaction group.
5713 *
5714 * If there are no dirty vdevs, we sync the uberblock to a few
5715 * random top-level vdevs that are known to be visible in the
5716 * config cache (see spa_vdev_add() for a complete description).
5717 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5718 */
5719 for (;;) {
5720 /*
5721 * We hold SCL_STATE to prevent vdev open/close/etc.
5722 * while we're attempting to write the vdev labels.
5723 */
5724 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5725
5726 if (list_is_empty(&spa->spa_config_dirty_list)) {
5727 vdev_t *svd[SPA_DVAS_PER_BP];
5728 int svdcount = 0;
5729 int children = rvd->vdev_children;
5730 int c0 = spa_get_random(children);
5731
5732 for (int c = 0; c < children; c++) {
5733 vd = rvd->vdev_child[(c0 + c) % children];
5734 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5735 continue;
5736 svd[svdcount++] = vd;
5737 if (svdcount == SPA_DVAS_PER_BP)
5738 break;
5739 }
5740 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5741 if (error != 0)
5742 error = vdev_config_sync(svd, svdcount, txg,
5743 B_TRUE);
5744 } else {
5745 error = vdev_config_sync(rvd->vdev_child,
5746 rvd->vdev_children, txg, B_FALSE);
5747 if (error != 0)
5748 error = vdev_config_sync(rvd->vdev_child,
5749 rvd->vdev_children, txg, B_TRUE);
5750 }
5751
5752 spa_config_exit(spa, SCL_STATE, FTAG);
5753
5754 if (error == 0)
5755 break;
5756 zio_suspend(spa, NULL);
5757 zio_resume_wait(spa);
5758 }
5759 dmu_tx_commit(tx);
5760
5761 /*
5762 * Clear the dirty config list.
5763 */
5764 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5765 vdev_config_clean(vd);
5766
5767 /*
5768 * Now that the new config has synced transactionally,
5769 * let it become visible to the config cache.
5770 */
5771 if (spa->spa_config_syncing != NULL) {
5772 spa_config_set(spa, spa->spa_config_syncing);
5773 spa->spa_config_txg = txg;
5774 spa->spa_config_syncing = NULL;
5775 }
5776
5777 spa->spa_ubsync = spa->spa_uberblock;
5778
5779 dsl_pool_sync_done(dp, txg);
5780
5781 /*
5782 * Update usable space statistics.
5783 */
5784 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5785 vdev_sync_done(vd, txg);
5786
5787 spa_update_dspace(spa);
5788
5789 /*
5790 * It had better be the case that we didn't dirty anything
5791 * since vdev_config_sync().
5792 */
5793 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5794 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5795 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5796
5797 spa->spa_sync_pass = 0;
5798
5799 spa_config_exit(spa, SCL_CONFIG, FTAG);
5800
5801 spa_handle_ignored_writes(spa);
5802
5803 /*
5804 * If any async tasks have been requested, kick them off.
5805 */
5806 spa_async_dispatch(spa);
5807}
5808
5809/*
5810 * Sync all pools. We don't want to hold the namespace lock across these
5811 * operations, so we take a reference on the spa_t and drop the lock during the
5812 * sync.
5813 */
5814void
5815spa_sync_allpools(void)
5816{
5817 spa_t *spa = NULL;
5818 mutex_enter(&spa_namespace_lock);
5819 while ((spa = spa_next(spa)) != NULL) {
5820 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5821 !spa_writeable(spa) || spa_suspended(spa))
5822 continue;
5823 spa_open_ref(spa, FTAG);
5824 mutex_exit(&spa_namespace_lock);
5825 txg_wait_synced(spa_get_dsl(spa), 0);
5826 mutex_enter(&spa_namespace_lock);
5827 spa_close(spa, FTAG);
5828 }
5829 mutex_exit(&spa_namespace_lock);
5830}
5831
5832/*
5833 * ==========================================================================
5834 * Miscellaneous routines
5835 * ==========================================================================
5836 */
5837
5838/*
5839 * Remove all pools in the system.
5840 */
5841void
5842spa_evict_all(void)
5843{
5844 spa_t *spa;
5845
5846 /*
5847 * Remove all cached state. All pools should be closed now,
5848 * so every spa in the AVL tree should be unreferenced.
5849 */
5850 mutex_enter(&spa_namespace_lock);
5851 while ((spa = spa_next(NULL)) != NULL) {
5852 /*
5853 * Stop async tasks. The async thread may need to detach
5854 * a device that's been replaced, which requires grabbing
5855 * spa_namespace_lock, so we must drop it here.
5856 */
5857 spa_open_ref(spa, FTAG);
5858 mutex_exit(&spa_namespace_lock);
5859 spa_async_suspend(spa);
5860 mutex_enter(&spa_namespace_lock);
5861 spa_close(spa, FTAG);
5862
5863 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5864 spa_unload(spa);
5865 spa_deactivate(spa);
5866 }
5867 spa_remove(spa);
5868 }
5869 mutex_exit(&spa_namespace_lock);
5870}
5871
5872vdev_t *
5873spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5874{
5875 vdev_t *vd;
5876 int i;
5877
5878 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5879 return (vd);
5880
5881 if (aux) {
5882 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5883 vd = spa->spa_l2cache.sav_vdevs[i];
5884 if (vd->vdev_guid == guid)
5885 return (vd);
5886 }
5887
5888 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5889 vd = spa->spa_spares.sav_vdevs[i];
5890 if (vd->vdev_guid == guid)
5891 return (vd);
5892 }
5893 }
5894
5895 return (NULL);
5896}
5897
5898void
5899spa_upgrade(spa_t *spa, uint64_t version)
5900{
5901 ASSERT(spa_writeable(spa));
5902
5903 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5904
5905 /*
5906 * This should only be called for a non-faulted pool, and since a
5907 * future version would result in an unopenable pool, this shouldn't be
5908 * possible.
5909 */
5910 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5911 ASSERT(version >= spa->spa_uberblock.ub_version);
5912
5913 spa->spa_uberblock.ub_version = version;
5914 vdev_config_dirty(spa->spa_root_vdev);
5915
5916 spa_config_exit(spa, SCL_ALL, FTAG);
5917
5918 txg_wait_synced(spa_get_dsl(spa), 0);
5919}
5920
5921boolean_t
5922spa_has_spare(spa_t *spa, uint64_t guid)
5923{
5924 int i;
5925 uint64_t spareguid;
5926 spa_aux_vdev_t *sav = &spa->spa_spares;
5927
5928 for (i = 0; i < sav->sav_count; i++)
5929 if (sav->sav_vdevs[i]->vdev_guid == guid)
5930 return (B_TRUE);
5931
5932 for (i = 0; i < sav->sav_npending; i++) {
5933 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5934 &spareguid) == 0 && spareguid == guid)
5935 return (B_TRUE);
5936 }
5937
5938 return (B_FALSE);
5939}
5940
5941/*
5942 * Check if a pool has an active shared spare device.
5943 * Note: reference count of an active spare is 2, as a spare and as a replace
5944 */
5945static boolean_t
5946spa_has_active_shared_spare(spa_t *spa)
5947{
5948 int i, refcnt;
5949 uint64_t pool;
5950 spa_aux_vdev_t *sav = &spa->spa_spares;
5951
5952 for (i = 0; i < sav->sav_count; i++) {
5953 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5954 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5955 refcnt > 2)
5956 return (B_TRUE);
5957 }
5958
5959 return (B_FALSE);
5960}
5961
5962/*
5963 * Post a sysevent corresponding to the given event. The 'name' must be one of
5964 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5965 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5966 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5967 * or zdb as real changes.
5968 */
5969void
5970spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5971{
5972#ifdef _KERNEL
5973 sysevent_t *ev;
5974 sysevent_attr_list_t *attr = NULL;
5975 sysevent_value_t value;
5976 sysevent_id_t eid;
5977
5978 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5979 SE_SLEEP);
5980
5981 value.value_type = SE_DATA_TYPE_STRING;
5982 value.value.sv_string = spa_name(spa);
5983 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5984 goto done;
5985
5986 value.value_type = SE_DATA_TYPE_UINT64;
5987 value.value.sv_uint64 = spa_guid(spa);
5988 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5989 goto done;
5990
5991 if (vd) {
5992 value.value_type = SE_DATA_TYPE_UINT64;
5993 value.value.sv_uint64 = vd->vdev_guid;
5994 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5995 SE_SLEEP) != 0)
5996 goto done;
5997
5998 if (vd->vdev_path) {
5999 value.value_type = SE_DATA_TYPE_STRING;
6000 value.value.sv_string = vd->vdev_path;
6001 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6002 &value, SE_SLEEP) != 0)
6003 goto done;
6004 }
6005 }
6006
6007 if (sysevent_attach_attributes(ev, attr) != 0)
6008 goto done;
6009 attr = NULL;
6010
6011 (void) log_sysevent(ev, SE_SLEEP, &eid);
6012
6013done:
6014 if (attr)
6015 sysevent_free_attr(attr);
6016 sysevent_free(ev);
6017#endif
6018}
|