1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/spa_boot.h>
61#include <sys/zfs_ioctl.h>
62#include <sys/dsl_scan.h>
63#include <sys/zfeature.h>
64#include <sys/zvol.h>
65#include <sys/trim_map.h>
66
67#ifdef _KERNEL
68#include <sys/callb.h>
69#include <sys/cpupart.h>
70#include <sys/zone.h>
71#endif /* _KERNEL */
72
73#include "zfs_prop.h"
74#include "zfs_comutil.h"
75
76/* Check hostid on import? */
77static int check_hostid = 1;
78
79SYSCTL_DECL(_vfs_zfs);
80TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
81SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
82 "Check hostid on import?");
83
84typedef enum zti_modes {
85 zti_mode_fixed, /* value is # of threads (min 1) */
86 zti_mode_online_percent, /* value is % of online CPUs */
87 zti_mode_batch, /* cpu-intensive; value is ignored */
88 zti_mode_null, /* don't create a taskq */
89 zti_nmodes
90} zti_modes_t;
91
92#define ZTI_FIX(n) { zti_mode_fixed, (n) }
93#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
94#define ZTI_BATCH { zti_mode_batch, 0 }
95#define ZTI_NULL { zti_mode_null, 0 }
96
97#define ZTI_ONE ZTI_FIX(1)
98
99typedef struct zio_taskq_info {
100 enum zti_modes zti_mode;
101 uint_t zti_value;
102} zio_taskq_info_t;
103
104static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
105 "issue", "issue_high", "intr", "intr_high"
106};
107
108/*
109 * Define the taskq threads for the following I/O types:
110 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
111 */
112const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
113 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
116 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
117 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
119 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
120};
121
122static dsl_syncfunc_t spa_sync_version;
123static dsl_syncfunc_t spa_sync_props;
124static dsl_checkfunc_t spa_change_guid_check;
125static dsl_syncfunc_t spa_change_guid_sync;
126static boolean_t spa_has_active_shared_spare(spa_t *spa);
127static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
129 char **ereport);
130static void spa_vdev_resilver_done(spa_t *spa);
131
132uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
133#ifdef PSRSET_BIND
134id_t zio_taskq_psrset_bind = PS_NONE;
135#endif
136#ifdef SYSDC
137boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
138#endif
139uint_t zio_taskq_basedc = 80; /* base duty cycle */
140
141boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
142extern int zfs_sync_pass_deferred_free;
143
144/*
145 * This (illegal) pool name is used when temporarily importing a spa_t in order
146 * to get the vdev stats associated with the imported devices.
147 */
148#define TRYIMPORT_NAME "$import"
149
150/*
151 * ==========================================================================
152 * SPA properties routines
153 * ==========================================================================
154 */
155
156/*
157 * Add a (source=src, propname=propval) list to an nvlist.
158 */
159static void
160spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
161 uint64_t intval, zprop_source_t src)
162{
163 const char *propname = zpool_prop_to_name(prop);
164 nvlist_t *propval;
165
166 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
167 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
168
169 if (strval != NULL)
170 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
171 else
172 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
173
174 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
175 nvlist_free(propval);
176}
177
178/*
179 * Get property values from the spa configuration.
180 */
181static void
182spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
183{
184 vdev_t *rvd = spa->spa_root_vdev;
185 dsl_pool_t *pool = spa->spa_dsl_pool;
186 uint64_t size;
187 uint64_t alloc;
188 uint64_t space;
189 uint64_t cap, version;
190 zprop_source_t src = ZPROP_SRC_NONE;
191 spa_config_dirent_t *dp;
192
193 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
194
195 if (rvd != NULL) {
196 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
197 size = metaslab_class_get_space(spa_normal_class(spa));
198 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
199 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
200 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
201 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
202 size - alloc, src);
203
204 space = 0;
205 for (int c = 0; c < rvd->vdev_children; c++) {
206 vdev_t *tvd = rvd->vdev_child[c];
207 space += tvd->vdev_max_asize - tvd->vdev_asize;
208 }
209 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
210 src);
211
212 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
213 (spa_mode(spa) == FREAD), src);
214
215 cap = (size == 0) ? 0 : (alloc * 100 / size);
216 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
217
218 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
219 ddt_get_pool_dedup_ratio(spa), src);
220
221 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
222 rvd->vdev_state, src);
223
224 version = spa_version(spa);
225 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
226 src = ZPROP_SRC_DEFAULT;
227 else
228 src = ZPROP_SRC_LOCAL;
229 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
230 }
231
232 if (pool != NULL) {
233 dsl_dir_t *freedir = pool->dp_free_dir;
234
235 /*
236 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
237 * when opening pools before this version freedir will be NULL.
238 */
239 if (freedir != NULL) {
240 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
241 freedir->dd_phys->dd_used_bytes, src);
242 } else {
243 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
244 NULL, 0, src);
245 }
246 }
247
248 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
249
250 if (spa->spa_comment != NULL) {
251 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
252 0, ZPROP_SRC_LOCAL);
253 }
254
255 if (spa->spa_root != NULL)
256 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
257 0, ZPROP_SRC_LOCAL);
258
259 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
260 if (dp->scd_path == NULL) {
261 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
262 "none", 0, ZPROP_SRC_LOCAL);
263 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
264 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
265 dp->scd_path, 0, ZPROP_SRC_LOCAL);
266 }
267 }
268}
269
270/*
271 * Get zpool property values.
272 */
273int
274spa_prop_get(spa_t *spa, nvlist_t **nvp)
275{
276 objset_t *mos = spa->spa_meta_objset;
277 zap_cursor_t zc;
278 zap_attribute_t za;
279 int err;
280
281 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282
283 mutex_enter(&spa->spa_props_lock);
284
285 /*
286 * Get properties from the spa config.
287 */
288 spa_prop_get_config(spa, nvp);
289
290 /* If no pool property object, no more prop to get. */
291 if (mos == NULL || spa->spa_pool_props_object == 0) {
292 mutex_exit(&spa->spa_props_lock);
293 return (0);
294 }
295
296 /*
297 * Get properties from the MOS pool property object.
298 */
299 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
300 (err = zap_cursor_retrieve(&zc, &za)) == 0;
301 zap_cursor_advance(&zc)) {
302 uint64_t intval = 0;
303 char *strval = NULL;
304 zprop_source_t src = ZPROP_SRC_DEFAULT;
305 zpool_prop_t prop;
306
307 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
308 continue;
309
310 switch (za.za_integer_length) {
311 case 8:
312 /* integer property */
313 if (za.za_first_integer !=
314 zpool_prop_default_numeric(prop))
315 src = ZPROP_SRC_LOCAL;
316
317 if (prop == ZPOOL_PROP_BOOTFS) {
318 dsl_pool_t *dp;
319 dsl_dataset_t *ds = NULL;
320
321 dp = spa_get_dsl(spa);
322 rw_enter(&dp->dp_config_rwlock, RW_READER);
323 if (err = dsl_dataset_hold_obj(dp,
324 za.za_first_integer, FTAG, &ds)) {
325 rw_exit(&dp->dp_config_rwlock);
326 break;
327 }
328
329 strval = kmem_alloc(
330 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
331 KM_SLEEP);
332 dsl_dataset_name(ds, strval);
333 dsl_dataset_rele(ds, FTAG);
334 rw_exit(&dp->dp_config_rwlock);
335 } else {
336 strval = NULL;
337 intval = za.za_first_integer;
338 }
339
340 spa_prop_add_list(*nvp, prop, strval, intval, src);
341
342 if (strval != NULL)
343 kmem_free(strval,
344 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
345
346 break;
347
348 case 1:
349 /* string property */
350 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
351 err = zap_lookup(mos, spa->spa_pool_props_object,
352 za.za_name, 1, za.za_num_integers, strval);
353 if (err) {
354 kmem_free(strval, za.za_num_integers);
355 break;
356 }
357 spa_prop_add_list(*nvp, prop, strval, 0, src);
358 kmem_free(strval, za.za_num_integers);
359 break;
360
361 default:
362 break;
363 }
364 }
365 zap_cursor_fini(&zc);
366 mutex_exit(&spa->spa_props_lock);
367out:
368 if (err && err != ENOENT) {
369 nvlist_free(*nvp);
370 *nvp = NULL;
371 return (err);
372 }
373
374 return (0);
375}
376
377/*
378 * Validate the given pool properties nvlist and modify the list
379 * for the property values to be set.
380 */
381static int
382spa_prop_validate(spa_t *spa, nvlist_t *props)
383{
384 nvpair_t *elem;
385 int error = 0, reset_bootfs = 0;
386 uint64_t objnum;
387 boolean_t has_feature = B_FALSE;
388
389 elem = NULL;
390 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
391 uint64_t intval;
392 char *strval, *slash, *check, *fname;
393 const char *propname = nvpair_name(elem);
394 zpool_prop_t prop = zpool_name_to_prop(propname);
395
396 switch (prop) {
397 case ZPROP_INVAL:
398 if (!zpool_prop_feature(propname)) {
399 error = EINVAL;
400 break;
401 }
402
403 /*
404 * Sanitize the input.
405 */
406 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
407 error = EINVAL;
408 break;
409 }
410
411 if (nvpair_value_uint64(elem, &intval) != 0) {
412 error = EINVAL;
413 break;
414 }
415
416 if (intval != 0) {
417 error = EINVAL;
418 break;
419 }
420
421 fname = strchr(propname, '@') + 1;
422 if (zfeature_lookup_name(fname, NULL) != 0) {
423 error = EINVAL;
424 break;
425 }
426
427 has_feature = B_TRUE;
428 break;
429
430 case ZPOOL_PROP_VERSION:
431 error = nvpair_value_uint64(elem, &intval);
432 if (!error &&
433 (intval < spa_version(spa) ||
434 intval > SPA_VERSION_BEFORE_FEATURES ||
435 has_feature))
436 error = EINVAL;
437 break;
438
439 case ZPOOL_PROP_DELEGATION:
440 case ZPOOL_PROP_AUTOREPLACE:
441 case ZPOOL_PROP_LISTSNAPS:
442 case ZPOOL_PROP_AUTOEXPAND:
443 error = nvpair_value_uint64(elem, &intval);
444 if (!error && intval > 1)
445 error = EINVAL;
446 break;
447
448 case ZPOOL_PROP_BOOTFS:
449 /*
450 * If the pool version is less than SPA_VERSION_BOOTFS,
451 * or the pool is still being created (version == 0),
452 * the bootfs property cannot be set.
453 */
454 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
455 error = ENOTSUP;
456 break;
457 }
458
459 /*
460 * Make sure the vdev config is bootable
461 */
462 if (!vdev_is_bootable(spa->spa_root_vdev)) {
463 error = ENOTSUP;
464 break;
465 }
466
467 reset_bootfs = 1;
468
469 error = nvpair_value_string(elem, &strval);
470
471 if (!error) {
472 objset_t *os;
473 uint64_t compress;
474
475 if (strval == NULL || strval[0] == '\0') {
476 objnum = zpool_prop_default_numeric(
477 ZPOOL_PROP_BOOTFS);
478 break;
479 }
480
481 if (error = dmu_objset_hold(strval, FTAG, &os))
482 break;
483
484 /* Must be ZPL and not gzip compressed. */
485
486 if (dmu_objset_type(os) != DMU_OST_ZFS) {
487 error = ENOTSUP;
488 } else if ((error = dsl_prop_get_integer(strval,
489 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
490 &compress, NULL)) == 0 &&
491 !BOOTFS_COMPRESS_VALID(compress)) {
492 error = ENOTSUP;
493 } else {
494 objnum = dmu_objset_id(os);
495 }
496 dmu_objset_rele(os, FTAG);
497 }
498 break;
499
500 case ZPOOL_PROP_FAILUREMODE:
501 error = nvpair_value_uint64(elem, &intval);
502 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
503 intval > ZIO_FAILURE_MODE_PANIC))
504 error = EINVAL;
505
506 /*
507 * This is a special case which only occurs when
508 * the pool has completely failed. This allows
509 * the user to change the in-core failmode property
510 * without syncing it out to disk (I/Os might
511 * currently be blocked). We do this by returning
512 * EIO to the caller (spa_prop_set) to trick it
513 * into thinking we encountered a property validation
514 * error.
515 */
516 if (!error && spa_suspended(spa)) {
517 spa->spa_failmode = intval;
518 error = EIO;
519 }
520 break;
521
522 case ZPOOL_PROP_CACHEFILE:
523 if ((error = nvpair_value_string(elem, &strval)) != 0)
524 break;
525
526 if (strval[0] == '\0')
527 break;
528
529 if (strcmp(strval, "none") == 0)
530 break;
531
532 if (strval[0] != '/') {
533 error = EINVAL;
534 break;
535 }
536
537 slash = strrchr(strval, '/');
538 ASSERT(slash != NULL);
539
540 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
541 strcmp(slash, "/..") == 0)
542 error = EINVAL;
543 break;
544
545 case ZPOOL_PROP_COMMENT:
546 if ((error = nvpair_value_string(elem, &strval)) != 0)
547 break;
548 for (check = strval; *check != '\0'; check++) {
549 /*
550 * The kernel doesn't have an easy isprint()
551 * check. For this kernel check, we merely
552 * check ASCII apart from DEL. Fix this if
553 * there is an easy-to-use kernel isprint().
554 */
555 if (*check >= 0x7f) {
556 error = EINVAL;
557 break;
558 }
559 check++;
560 }
561 if (strlen(strval) > ZPROP_MAX_COMMENT)
562 error = E2BIG;
563 break;
564
565 case ZPOOL_PROP_DEDUPDITTO:
566 if (spa_version(spa) < SPA_VERSION_DEDUP)
567 error = ENOTSUP;
568 else
569 error = nvpair_value_uint64(elem, &intval);
570 if (error == 0 &&
571 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
572 error = EINVAL;
573 break;
574 }
575
576 if (error)
577 break;
578 }
579
580 if (!error && reset_bootfs) {
581 error = nvlist_remove(props,
582 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
583
584 if (!error) {
585 error = nvlist_add_uint64(props,
586 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
587 }
588 }
589
590 return (error);
591}
592
593void
594spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
595{
596 char *cachefile;
597 spa_config_dirent_t *dp;
598
599 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
600 &cachefile) != 0)
601 return;
602
603 dp = kmem_alloc(sizeof (spa_config_dirent_t),
604 KM_SLEEP);
605
606 if (cachefile[0] == '\0')
607 dp->scd_path = spa_strdup(spa_config_path);
608 else if (strcmp(cachefile, "none") == 0)
609 dp->scd_path = NULL;
610 else
611 dp->scd_path = spa_strdup(cachefile);
612
613 list_insert_head(&spa->spa_config_list, dp);
614 if (need_sync)
615 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
616}
617
618int
619spa_prop_set(spa_t *spa, nvlist_t *nvp)
620{
621 int error;
622 nvpair_t *elem = NULL;
623 boolean_t need_sync = B_FALSE;
624
625 if ((error = spa_prop_validate(spa, nvp)) != 0)
626 return (error);
627
628 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
629 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
630
631 if (prop == ZPOOL_PROP_CACHEFILE ||
632 prop == ZPOOL_PROP_ALTROOT ||
633 prop == ZPOOL_PROP_READONLY)
634 continue;
635
636 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
637 uint64_t ver;
638
639 if (prop == ZPOOL_PROP_VERSION) {
640 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
641 } else {
642 ASSERT(zpool_prop_feature(nvpair_name(elem)));
643 ver = SPA_VERSION_FEATURES;
644 need_sync = B_TRUE;
645 }
646
647 /* Save time if the version is already set. */
648 if (ver == spa_version(spa))
649 continue;
650
651 /*
652 * In addition to the pool directory object, we might
653 * create the pool properties object, the features for
654 * read object, the features for write object, or the
655 * feature descriptions object.
656 */
657 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
658 spa_sync_version, spa, &ver, 6);
659 if (error)
660 return (error);
661 continue;
662 }
663
664 need_sync = B_TRUE;
665 break;
666 }
667
668 if (need_sync) {
669 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
670 spa, nvp, 6));
671 }
672
673 return (0);
674}
675
676/*
677 * If the bootfs property value is dsobj, clear it.
678 */
679void
680spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
681{
682 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
683 VERIFY(zap_remove(spa->spa_meta_objset,
684 spa->spa_pool_props_object,
685 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
686 spa->spa_bootfs = 0;
687 }
688}
689
690/*ARGSUSED*/
691static int
692spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
693{
694 spa_t *spa = arg1;
695 uint64_t *newguid = arg2;
696 vdev_t *rvd = spa->spa_root_vdev;
697 uint64_t vdev_state;
698
699 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
700 vdev_state = rvd->vdev_state;
701 spa_config_exit(spa, SCL_STATE, FTAG);
702
703 if (vdev_state != VDEV_STATE_HEALTHY)
704 return (ENXIO);
705
706 ASSERT3U(spa_guid(spa), !=, *newguid);
707
708 return (0);
709}
710
711static void
712spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
713{
714 spa_t *spa = arg1;
715 uint64_t *newguid = arg2;
716 uint64_t oldguid;
717 vdev_t *rvd = spa->spa_root_vdev;
718
719 oldguid = spa_guid(spa);
720
721 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
722 rvd->vdev_guid = *newguid;
723 rvd->vdev_guid_sum += (*newguid - oldguid);
724 vdev_config_dirty(rvd);
725 spa_config_exit(spa, SCL_STATE, FTAG);
726
727#ifdef __FreeBSD__
728 /*
729 * TODO: until recent illumos logging changes are merged
730 * log reguid as pool property change
731 */
732 spa_history_log_internal(LOG_POOL_PROPSET, spa, tx,
733 "guid change old=%llu new=%llu", oldguid, *newguid);
734#else
735 spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld",
736 oldguid, *newguid);
737#endif
738}
739
740/*
741 * Change the GUID for the pool. This is done so that we can later
742 * re-import a pool built from a clone of our own vdevs. We will modify
743 * the root vdev's guid, our own pool guid, and then mark all of our
744 * vdevs dirty. Note that we must make sure that all our vdevs are
745 * online when we do this, or else any vdevs that weren't present
746 * would be orphaned from our pool. We are also going to issue a
747 * sysevent to update any watchers.
748 */
749int
750spa_change_guid(spa_t *spa)
751{
752 int error;
753 uint64_t guid;
754
755 mutex_enter(&spa_namespace_lock);
756 guid = spa_generate_guid(NULL);
757
758 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
759 spa_change_guid_sync, spa, &guid, 5);
760
761 if (error == 0) {
762 spa_config_sync(spa, B_FALSE, B_TRUE);
763 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
764 }
765
766 mutex_exit(&spa_namespace_lock);
767
768 return (error);
769}
770
771/*
772 * ==========================================================================
773 * SPA state manipulation (open/create/destroy/import/export)
774 * ==========================================================================
775 */
776
777static int
778spa_error_entry_compare(const void *a, const void *b)
779{
780 spa_error_entry_t *sa = (spa_error_entry_t *)a;
781 spa_error_entry_t *sb = (spa_error_entry_t *)b;
782 int ret;
783
784 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
785 sizeof (zbookmark_t));
786
787 if (ret < 0)
788 return (-1);
789 else if (ret > 0)
790 return (1);
791 else
792 return (0);
793}
794
795/*
796 * Utility function which retrieves copies of the current logs and
797 * re-initializes them in the process.
798 */
799void
800spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
801{
802 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
803
804 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
805 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
806
807 avl_create(&spa->spa_errlist_scrub,
808 spa_error_entry_compare, sizeof (spa_error_entry_t),
809 offsetof(spa_error_entry_t, se_avl));
810 avl_create(&spa->spa_errlist_last,
811 spa_error_entry_compare, sizeof (spa_error_entry_t),
812 offsetof(spa_error_entry_t, se_avl));
813}
814
815static taskq_t *
816spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
817 uint_t value)
818{
819 uint_t flags = TASKQ_PREPOPULATE;
820 boolean_t batch = B_FALSE;
821
822 switch (mode) {
823 case zti_mode_null:
824 return (NULL); /* no taskq needed */
825
826 case zti_mode_fixed:
827 ASSERT3U(value, >=, 1);
828 value = MAX(value, 1);
829 break;
830
831 case zti_mode_batch:
832 batch = B_TRUE;
833 flags |= TASKQ_THREADS_CPU_PCT;
834 value = zio_taskq_batch_pct;
835 break;
836
837 case zti_mode_online_percent:
838 flags |= TASKQ_THREADS_CPU_PCT;
839 break;
840
841 default:
842 panic("unrecognized mode for %s taskq (%u:%u) in "
843 "spa_activate()",
844 name, mode, value);
845 break;
846 }
847
848#ifdef SYSDC
849 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
850 if (batch)
851 flags |= TASKQ_DC_BATCH;
852
853 return (taskq_create_sysdc(name, value, 50, INT_MAX,
854 spa->spa_proc, zio_taskq_basedc, flags));
855 }
856#endif
857 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
858 spa->spa_proc, flags));
859}
860
861static void
862spa_create_zio_taskqs(spa_t *spa)
863{
864 for (int t = 0; t < ZIO_TYPES; t++) {
865 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
866 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
867 enum zti_modes mode = ztip->zti_mode;
868 uint_t value = ztip->zti_value;
869 char name[32];
870
871 (void) snprintf(name, sizeof (name),
872 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
873
874 spa->spa_zio_taskq[t][q] =
875 spa_taskq_create(spa, name, mode, value);
876 }
877 }
878}
879
880#ifdef _KERNEL
881#ifdef SPA_PROCESS
882static void
883spa_thread(void *arg)
884{
885 callb_cpr_t cprinfo;
886
887 spa_t *spa = arg;
888 user_t *pu = PTOU(curproc);
889
890 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
891 spa->spa_name);
892
893 ASSERT(curproc != &p0);
894 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
895 "zpool-%s", spa->spa_name);
896 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
897
898#ifdef PSRSET_BIND
899 /* bind this thread to the requested psrset */
900 if (zio_taskq_psrset_bind != PS_NONE) {
901 pool_lock();
902 mutex_enter(&cpu_lock);
903 mutex_enter(&pidlock);
904 mutex_enter(&curproc->p_lock);
905
906 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
907 0, NULL, NULL) == 0) {
908 curthread->t_bind_pset = zio_taskq_psrset_bind;
909 } else {
910 cmn_err(CE_WARN,
911 "Couldn't bind process for zfs pool \"%s\" to "
912 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
913 }
914
915 mutex_exit(&curproc->p_lock);
916 mutex_exit(&pidlock);
917 mutex_exit(&cpu_lock);
918 pool_unlock();
919 }
920#endif
921
922#ifdef SYSDC
923 if (zio_taskq_sysdc) {
924 sysdc_thread_enter(curthread, 100, 0);
925 }
926#endif
927
928 spa->spa_proc = curproc;
929 spa->spa_did = curthread->t_did;
930
931 spa_create_zio_taskqs(spa);
932
933 mutex_enter(&spa->spa_proc_lock);
934 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
935
936 spa->spa_proc_state = SPA_PROC_ACTIVE;
937 cv_broadcast(&spa->spa_proc_cv);
938
939 CALLB_CPR_SAFE_BEGIN(&cprinfo);
940 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
941 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
942 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
943
944 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
945 spa->spa_proc_state = SPA_PROC_GONE;
946 spa->spa_proc = &p0;
947 cv_broadcast(&spa->spa_proc_cv);
948 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
949
950 mutex_enter(&curproc->p_lock);
951 lwp_exit();
952}
953#endif /* SPA_PROCESS */
954#endif
955
956/*
957 * Activate an uninitialized pool.
958 */
959static void
960spa_activate(spa_t *spa, int mode)
961{
962 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
963
964 spa->spa_state = POOL_STATE_ACTIVE;
965 spa->spa_mode = mode;
966
967 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
968 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
969
970 /* Try to create a covering process */
971 mutex_enter(&spa->spa_proc_lock);
972 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
973 ASSERT(spa->spa_proc == &p0);
974 spa->spa_did = 0;
975
976#ifdef SPA_PROCESS
977 /* Only create a process if we're going to be around a while. */
978 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
979 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
980 NULL, 0) == 0) {
981 spa->spa_proc_state = SPA_PROC_CREATED;
982 while (spa->spa_proc_state == SPA_PROC_CREATED) {
983 cv_wait(&spa->spa_proc_cv,
984 &spa->spa_proc_lock);
985 }
986 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
987 ASSERT(spa->spa_proc != &p0);
988 ASSERT(spa->spa_did != 0);
989 } else {
990#ifdef _KERNEL
991 cmn_err(CE_WARN,
992 "Couldn't create process for zfs pool \"%s\"\n",
993 spa->spa_name);
994#endif
995 }
996 }
997#endif /* SPA_PROCESS */
998 mutex_exit(&spa->spa_proc_lock);
999
1000 /* If we didn't create a process, we need to create our taskqs. */
1001 ASSERT(spa->spa_proc == &p0);
1002 if (spa->spa_proc == &p0) {
1003 spa_create_zio_taskqs(spa);
1004 }
1005
1006 /*
1007 * Start TRIM thread.
1008 */
1009 trim_thread_create(spa);
1010
1011 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1012 offsetof(vdev_t, vdev_config_dirty_node));
1013 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1014 offsetof(vdev_t, vdev_state_dirty_node));
1015
1016 txg_list_create(&spa->spa_vdev_txg_list,
1017 offsetof(struct vdev, vdev_txg_node));
1018
1019 avl_create(&spa->spa_errlist_scrub,
1020 spa_error_entry_compare, sizeof (spa_error_entry_t),
1021 offsetof(spa_error_entry_t, se_avl));
1022 avl_create(&spa->spa_errlist_last,
1023 spa_error_entry_compare, sizeof (spa_error_entry_t),
1024 offsetof(spa_error_entry_t, se_avl));
1025}
1026
1027/*
1028 * Opposite of spa_activate().
1029 */
1030static void
1031spa_deactivate(spa_t *spa)
1032{
1033 ASSERT(spa->spa_sync_on == B_FALSE);
1034 ASSERT(spa->spa_dsl_pool == NULL);
1035 ASSERT(spa->spa_root_vdev == NULL);
1036 ASSERT(spa->spa_async_zio_root == NULL);
1037 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1038
1039 /*
1040 * Stop TRIM thread in case spa_unload() wasn't called directly
1041 * before spa_deactivate().
1042 */
1043 trim_thread_destroy(spa);
1044
1045 txg_list_destroy(&spa->spa_vdev_txg_list);
1046
1047 list_destroy(&spa->spa_config_dirty_list);
1048 list_destroy(&spa->spa_state_dirty_list);
1049
1050 for (int t = 0; t < ZIO_TYPES; t++) {
1051 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1052 if (spa->spa_zio_taskq[t][q] != NULL)
1053 taskq_destroy(spa->spa_zio_taskq[t][q]);
1054 spa->spa_zio_taskq[t][q] = NULL;
1055 }
1056 }
1057
1058 metaslab_class_destroy(spa->spa_normal_class);
1059 spa->spa_normal_class = NULL;
1060
1061 metaslab_class_destroy(spa->spa_log_class);
1062 spa->spa_log_class = NULL;
1063
1064 /*
1065 * If this was part of an import or the open otherwise failed, we may
1066 * still have errors left in the queues. Empty them just in case.
1067 */
1068 spa_errlog_drain(spa);
1069
1070 avl_destroy(&spa->spa_errlist_scrub);
1071 avl_destroy(&spa->spa_errlist_last);
1072
1073 spa->spa_state = POOL_STATE_UNINITIALIZED;
1074
1075 mutex_enter(&spa->spa_proc_lock);
1076 if (spa->spa_proc_state != SPA_PROC_NONE) {
1077 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1078 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1079 cv_broadcast(&spa->spa_proc_cv);
1080 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1081 ASSERT(spa->spa_proc != &p0);
1082 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1083 }
1084 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1085 spa->spa_proc_state = SPA_PROC_NONE;
1086 }
1087 ASSERT(spa->spa_proc == &p0);
1088 mutex_exit(&spa->spa_proc_lock);
1089
1090#ifdef SPA_PROCESS
1091 /*
1092 * We want to make sure spa_thread() has actually exited the ZFS
1093 * module, so that the module can't be unloaded out from underneath
1094 * it.
1095 */
1096 if (spa->spa_did != 0) {
1097 thread_join(spa->spa_did);
1098 spa->spa_did = 0;
1099 }
1100#endif /* SPA_PROCESS */
1101}
1102
1103/*
1104 * Verify a pool configuration, and construct the vdev tree appropriately. This
1105 * will create all the necessary vdevs in the appropriate layout, with each vdev
1106 * in the CLOSED state. This will prep the pool before open/creation/import.
1107 * All vdev validation is done by the vdev_alloc() routine.
1108 */
1109static int
1110spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1111 uint_t id, int atype)
1112{
1113 nvlist_t **child;
1114 uint_t children;
1115 int error;
1116
1117 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1118 return (error);
1119
1120 if ((*vdp)->vdev_ops->vdev_op_leaf)
1121 return (0);
1122
1123 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1124 &child, &children);
1125
1126 if (error == ENOENT)
1127 return (0);
1128
1129 if (error) {
1130 vdev_free(*vdp);
1131 *vdp = NULL;
1132 return (EINVAL);
1133 }
1134
1135 for (int c = 0; c < children; c++) {
1136 vdev_t *vd;
1137 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1138 atype)) != 0) {
1139 vdev_free(*vdp);
1140 *vdp = NULL;
1141 return (error);
1142 }
1143 }
1144
1145 ASSERT(*vdp != NULL);
1146
1147 return (0);
1148}
1149
1150/*
1151 * Opposite of spa_load().
1152 */
1153static void
1154spa_unload(spa_t *spa)
1155{
1156 int i;
1157
1158 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1159
1160 /*
1161 * Stop TRIM thread.
1162 */
1163 trim_thread_destroy(spa);
1164
1165 /*
1166 * Stop async tasks.
1167 */
1168 spa_async_suspend(spa);
1169
1170 /*
1171 * Stop syncing.
1172 */
1173 if (spa->spa_sync_on) {
1174 txg_sync_stop(spa->spa_dsl_pool);
1175 spa->spa_sync_on = B_FALSE;
1176 }
1177
1178 /*
1179 * Wait for any outstanding async I/O to complete.
1180 */
1181 if (spa->spa_async_zio_root != NULL) {
1182 (void) zio_wait(spa->spa_async_zio_root);
1183 spa->spa_async_zio_root = NULL;
1184 }
1185
1186 bpobj_close(&spa->spa_deferred_bpobj);
1187
1188 /*
1189 * Close the dsl pool.
1190 */
1191 if (spa->spa_dsl_pool) {
1192 dsl_pool_close(spa->spa_dsl_pool);
1193 spa->spa_dsl_pool = NULL;
1194 spa->spa_meta_objset = NULL;
1195 }
1196
1197 ddt_unload(spa);
1198
1199 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1200
1201 /*
1202 * Drop and purge level 2 cache
1203 */
1204 spa_l2cache_drop(spa);
1205
1206 /*
1207 * Close all vdevs.
1208 */
1209 if (spa->spa_root_vdev)
1210 vdev_free(spa->spa_root_vdev);
1211 ASSERT(spa->spa_root_vdev == NULL);
1212
1213 for (i = 0; i < spa->spa_spares.sav_count; i++)
1214 vdev_free(spa->spa_spares.sav_vdevs[i]);
1215 if (spa->spa_spares.sav_vdevs) {
1216 kmem_free(spa->spa_spares.sav_vdevs,
1217 spa->spa_spares.sav_count * sizeof (void *));
1218 spa->spa_spares.sav_vdevs = NULL;
1219 }
1220 if (spa->spa_spares.sav_config) {
1221 nvlist_free(spa->spa_spares.sav_config);
1222 spa->spa_spares.sav_config = NULL;
1223 }
1224 spa->spa_spares.sav_count = 0;
1225
1226 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1227 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1228 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1229 }
1230 if (spa->spa_l2cache.sav_vdevs) {
1231 kmem_free(spa->spa_l2cache.sav_vdevs,
1232 spa->spa_l2cache.sav_count * sizeof (void *));
1233 spa->spa_l2cache.sav_vdevs = NULL;
1234 }
1235 if (spa->spa_l2cache.sav_config) {
1236 nvlist_free(spa->spa_l2cache.sav_config);
1237 spa->spa_l2cache.sav_config = NULL;
1238 }
1239 spa->spa_l2cache.sav_count = 0;
1240
1241 spa->spa_async_suspended = 0;
1242
1243 if (spa->spa_comment != NULL) {
1244 spa_strfree(spa->spa_comment);
1245 spa->spa_comment = NULL;
1246 }
1247
1248 spa_config_exit(spa, SCL_ALL, FTAG);
1249}
1250
1251/*
1252 * Load (or re-load) the current list of vdevs describing the active spares for
1253 * this pool. When this is called, we have some form of basic information in
1254 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1255 * then re-generate a more complete list including status information.
1256 */
1257static void
1258spa_load_spares(spa_t *spa)
1259{
1260 nvlist_t **spares;
1261 uint_t nspares;
1262 int i;
1263 vdev_t *vd, *tvd;
1264
1265 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1266
1267 /*
1268 * First, close and free any existing spare vdevs.
1269 */
1270 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1271 vd = spa->spa_spares.sav_vdevs[i];
1272
1273 /* Undo the call to spa_activate() below */
1274 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1275 B_FALSE)) != NULL && tvd->vdev_isspare)
1276 spa_spare_remove(tvd);
1277 vdev_close(vd);
1278 vdev_free(vd);
1279 }
1280
1281 if (spa->spa_spares.sav_vdevs)
1282 kmem_free(spa->spa_spares.sav_vdevs,
1283 spa->spa_spares.sav_count * sizeof (void *));
1284
1285 if (spa->spa_spares.sav_config == NULL)
1286 nspares = 0;
1287 else
1288 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1289 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1290
1291 spa->spa_spares.sav_count = (int)nspares;
1292 spa->spa_spares.sav_vdevs = NULL;
1293
1294 if (nspares == 0)
1295 return;
1296
1297 /*
1298 * Construct the array of vdevs, opening them to get status in the
1299 * process. For each spare, there is potentially two different vdev_t
1300 * structures associated with it: one in the list of spares (used only
1301 * for basic validation purposes) and one in the active vdev
1302 * configuration (if it's spared in). During this phase we open and
1303 * validate each vdev on the spare list. If the vdev also exists in the
1304 * active configuration, then we also mark this vdev as an active spare.
1305 */
1306 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1307 KM_SLEEP);
1308 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1309 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1310 VDEV_ALLOC_SPARE) == 0);
1311 ASSERT(vd != NULL);
1312
1313 spa->spa_spares.sav_vdevs[i] = vd;
1314
1315 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1316 B_FALSE)) != NULL) {
1317 if (!tvd->vdev_isspare)
1318 spa_spare_add(tvd);
1319
1320 /*
1321 * We only mark the spare active if we were successfully
1322 * able to load the vdev. Otherwise, importing a pool
1323 * with a bad active spare would result in strange
1324 * behavior, because multiple pool would think the spare
1325 * is actively in use.
1326 *
1327 * There is a vulnerability here to an equally bizarre
1328 * circumstance, where a dead active spare is later
1329 * brought back to life (onlined or otherwise). Given
1330 * the rarity of this scenario, and the extra complexity
1331 * it adds, we ignore the possibility.
1332 */
1333 if (!vdev_is_dead(tvd))
1334 spa_spare_activate(tvd);
1335 }
1336
1337 vd->vdev_top = vd;
1338 vd->vdev_aux = &spa->spa_spares;
1339
1340 if (vdev_open(vd) != 0)
1341 continue;
1342
1343 if (vdev_validate_aux(vd) == 0)
1344 spa_spare_add(vd);
1345 }
1346
1347 /*
1348 * Recompute the stashed list of spares, with status information
1349 * this time.
1350 */
1351 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1352 DATA_TYPE_NVLIST_ARRAY) == 0);
1353
1354 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1355 KM_SLEEP);
1356 for (i = 0; i < spa->spa_spares.sav_count; i++)
1357 spares[i] = vdev_config_generate(spa,
1358 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1359 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1360 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1361 for (i = 0; i < spa->spa_spares.sav_count; i++)
1362 nvlist_free(spares[i]);
1363 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1364}
1365
1366/*
1367 * Load (or re-load) the current list of vdevs describing the active l2cache for
1368 * this pool. When this is called, we have some form of basic information in
1369 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1370 * then re-generate a more complete list including status information.
1371 * Devices which are already active have their details maintained, and are
1372 * not re-opened.
1373 */
1374static void
1375spa_load_l2cache(spa_t *spa)
1376{
1377 nvlist_t **l2cache;
1378 uint_t nl2cache;
1379 int i, j, oldnvdevs;
1380 uint64_t guid;
1381 vdev_t *vd, **oldvdevs, **newvdevs;
1382 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1383
1384 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1385
1386 if (sav->sav_config != NULL) {
1387 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1388 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1389 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1390 } else {
1391 nl2cache = 0;
1392 }
1393
1394 oldvdevs = sav->sav_vdevs;
1395 oldnvdevs = sav->sav_count;
1396 sav->sav_vdevs = NULL;
1397 sav->sav_count = 0;
1398
1399 /*
1400 * Process new nvlist of vdevs.
1401 */
1402 for (i = 0; i < nl2cache; i++) {
1403 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1404 &guid) == 0);
1405
1406 newvdevs[i] = NULL;
1407 for (j = 0; j < oldnvdevs; j++) {
1408 vd = oldvdevs[j];
1409 if (vd != NULL && guid == vd->vdev_guid) {
1410 /*
1411 * Retain previous vdev for add/remove ops.
1412 */
1413 newvdevs[i] = vd;
1414 oldvdevs[j] = NULL;
1415 break;
1416 }
1417 }
1418
1419 if (newvdevs[i] == NULL) {
1420 /*
1421 * Create new vdev
1422 */
1423 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1424 VDEV_ALLOC_L2CACHE) == 0);
1425 ASSERT(vd != NULL);
1426 newvdevs[i] = vd;
1427
1428 /*
1429 * Commit this vdev as an l2cache device,
1430 * even if it fails to open.
1431 */
1432 spa_l2cache_add(vd);
1433
1434 vd->vdev_top = vd;
1435 vd->vdev_aux = sav;
1436
1437 spa_l2cache_activate(vd);
1438
1439 if (vdev_open(vd) != 0)
1440 continue;
1441
1442 (void) vdev_validate_aux(vd);
1443
1444 if (!vdev_is_dead(vd))
1445 l2arc_add_vdev(spa, vd);
1446 }
1447 }
1448
1449 /*
1450 * Purge vdevs that were dropped
1451 */
1452 for (i = 0; i < oldnvdevs; i++) {
1453 uint64_t pool;
1454
1455 vd = oldvdevs[i];
1456 if (vd != NULL) {
1457 ASSERT(vd->vdev_isl2cache);
1458
1459 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1460 pool != 0ULL && l2arc_vdev_present(vd))
1461 l2arc_remove_vdev(vd);
1462 vdev_clear_stats(vd);
1463 vdev_free(vd);
1464 }
1465 }
1466
1467 if (oldvdevs)
1468 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1469
1470 if (sav->sav_config == NULL)
1471 goto out;
1472
1473 sav->sav_vdevs = newvdevs;
1474 sav->sav_count = (int)nl2cache;
1475
1476 /*
1477 * Recompute the stashed list of l2cache devices, with status
1478 * information this time.
1479 */
1480 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1481 DATA_TYPE_NVLIST_ARRAY) == 0);
1482
1483 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1484 for (i = 0; i < sav->sav_count; i++)
1485 l2cache[i] = vdev_config_generate(spa,
1486 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1487 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1488 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1489out:
1490 for (i = 0; i < sav->sav_count; i++)
1491 nvlist_free(l2cache[i]);
1492 if (sav->sav_count)
1493 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1494}
1495
1496static int
1497load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1498{
1499 dmu_buf_t *db;
1500 char *packed = NULL;
1501 size_t nvsize = 0;
1502 int error;
1503 *value = NULL;
1504
1505 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1506 nvsize = *(uint64_t *)db->db_data;
1507 dmu_buf_rele(db, FTAG);
1508
1509 packed = kmem_alloc(nvsize, KM_SLEEP);
1510 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1511 DMU_READ_PREFETCH);
1512 if (error == 0)
1513 error = nvlist_unpack(packed, nvsize, value, 0);
1514 kmem_free(packed, nvsize);
1515
1516 return (error);
1517}
1518
1519/*
1520 * Checks to see if the given vdev could not be opened, in which case we post a
1521 * sysevent to notify the autoreplace code that the device has been removed.
1522 */
1523static void
1524spa_check_removed(vdev_t *vd)
1525{
1526 for (int c = 0; c < vd->vdev_children; c++)
1527 spa_check_removed(vd->vdev_child[c]);
1528
1529 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1530 zfs_post_autoreplace(vd->vdev_spa, vd);
1531 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1532 }
1533}
1534
1535/*
1536 * Validate the current config against the MOS config
1537 */
1538static boolean_t
1539spa_config_valid(spa_t *spa, nvlist_t *config)
1540{
1541 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1542 nvlist_t *nv;
1543
1544 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1545
1546 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1547 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1548
1549 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1550
1551 /*
1552 * If we're doing a normal import, then build up any additional
1553 * diagnostic information about missing devices in this config.
1554 * We'll pass this up to the user for further processing.
1555 */
1556 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1557 nvlist_t **child, *nv;
1558 uint64_t idx = 0;
1559
1560 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1561 KM_SLEEP);
1562 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1563
1564 for (int c = 0; c < rvd->vdev_children; c++) {
1565 vdev_t *tvd = rvd->vdev_child[c];
1566 vdev_t *mtvd = mrvd->vdev_child[c];
1567
1568 if (tvd->vdev_ops == &vdev_missing_ops &&
1569 mtvd->vdev_ops != &vdev_missing_ops &&
1570 mtvd->vdev_islog)
1571 child[idx++] = vdev_config_generate(spa, mtvd,
1572 B_FALSE, 0);
1573 }
1574
1575 if (idx) {
1576 VERIFY(nvlist_add_nvlist_array(nv,
1577 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1578 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1579 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1580
1581 for (int i = 0; i < idx; i++)
1582 nvlist_free(child[i]);
1583 }
1584 nvlist_free(nv);
1585 kmem_free(child, rvd->vdev_children * sizeof (char **));
1586 }
1587
1588 /*
1589 * Compare the root vdev tree with the information we have
1590 * from the MOS config (mrvd). Check each top-level vdev
1591 * with the corresponding MOS config top-level (mtvd).
1592 */
1593 for (int c = 0; c < rvd->vdev_children; c++) {
1594 vdev_t *tvd = rvd->vdev_child[c];
1595 vdev_t *mtvd = mrvd->vdev_child[c];
1596
1597 /*
1598 * Resolve any "missing" vdevs in the current configuration.
1599 * If we find that the MOS config has more accurate information
1600 * about the top-level vdev then use that vdev instead.
1601 */
1602 if (tvd->vdev_ops == &vdev_missing_ops &&
1603 mtvd->vdev_ops != &vdev_missing_ops) {
1604
1605 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1606 continue;
1607
1608 /*
1609 * Device specific actions.
1610 */
1611 if (mtvd->vdev_islog) {
1612 spa_set_log_state(spa, SPA_LOG_CLEAR);
1613 } else {
1614 /*
1615 * XXX - once we have 'readonly' pool
1616 * support we should be able to handle
1617 * missing data devices by transitioning
1618 * the pool to readonly.
1619 */
1620 continue;
1621 }
1622
1623 /*
1624 * Swap the missing vdev with the data we were
1625 * able to obtain from the MOS config.
1626 */
1627 vdev_remove_child(rvd, tvd);
1628 vdev_remove_child(mrvd, mtvd);
1629
1630 vdev_add_child(rvd, mtvd);
1631 vdev_add_child(mrvd, tvd);
1632
1633 spa_config_exit(spa, SCL_ALL, FTAG);
1634 vdev_load(mtvd);
1635 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1636
1637 vdev_reopen(rvd);
1638 } else if (mtvd->vdev_islog) {
1639 /*
1640 * Load the slog device's state from the MOS config
1641 * since it's possible that the label does not
1642 * contain the most up-to-date information.
1643 */
1644 vdev_load_log_state(tvd, mtvd);
1645 vdev_reopen(tvd);
1646 }
1647 }
1648 vdev_free(mrvd);
1649 spa_config_exit(spa, SCL_ALL, FTAG);
1650
1651 /*
1652 * Ensure we were able to validate the config.
1653 */
1654 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1655}
1656
1657/*
1658 * Check for missing log devices
1659 */
1660static int
1661spa_check_logs(spa_t *spa)
1662{
1663 switch (spa->spa_log_state) {
1664 case SPA_LOG_MISSING:
1665 /* need to recheck in case slog has been restored */
1666 case SPA_LOG_UNKNOWN:
1667 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1668 DS_FIND_CHILDREN)) {
1669 spa_set_log_state(spa, SPA_LOG_MISSING);
1670 return (1);
1671 }
1672 break;
1673 }
1674 return (0);
1675}
1676
1677static boolean_t
1678spa_passivate_log(spa_t *spa)
1679{
1680 vdev_t *rvd = spa->spa_root_vdev;
1681 boolean_t slog_found = B_FALSE;
1682
1683 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1684
1685 if (!spa_has_slogs(spa))
1686 return (B_FALSE);
1687
1688 for (int c = 0; c < rvd->vdev_children; c++) {
1689 vdev_t *tvd = rvd->vdev_child[c];
1690 metaslab_group_t *mg = tvd->vdev_mg;
1691
1692 if (tvd->vdev_islog) {
1693 metaslab_group_passivate(mg);
1694 slog_found = B_TRUE;
1695 }
1696 }
1697
1698 return (slog_found);
1699}
1700
1701static void
1702spa_activate_log(spa_t *spa)
1703{
1704 vdev_t *rvd = spa->spa_root_vdev;
1705
1706 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1707
1708 for (int c = 0; c < rvd->vdev_children; c++) {
1709 vdev_t *tvd = rvd->vdev_child[c];
1710 metaslab_group_t *mg = tvd->vdev_mg;
1711
1712 if (tvd->vdev_islog)
1713 metaslab_group_activate(mg);
1714 }
1715}
1716
1717int
1718spa_offline_log(spa_t *spa)
1719{
1720 int error = 0;
1721
1722 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1723 NULL, DS_FIND_CHILDREN)) == 0) {
1724
1725 /*
1726 * We successfully offlined the log device, sync out the
1727 * current txg so that the "stubby" block can be removed
1728 * by zil_sync().
1729 */
1730 txg_wait_synced(spa->spa_dsl_pool, 0);
1731 }
1732 return (error);
1733}
1734
1735static void
1736spa_aux_check_removed(spa_aux_vdev_t *sav)
1737{
1738 int i;
1739
1740 for (i = 0; i < sav->sav_count; i++)
1741 spa_check_removed(sav->sav_vdevs[i]);
1742}
1743
1744void
1745spa_claim_notify(zio_t *zio)
1746{
1747 spa_t *spa = zio->io_spa;
1748
1749 if (zio->io_error)
1750 return;
1751
1752 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1753 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1754 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1755 mutex_exit(&spa->spa_props_lock);
1756}
1757
1758typedef struct spa_load_error {
1759 uint64_t sle_meta_count;
1760 uint64_t sle_data_count;
1761} spa_load_error_t;
1762
1763static void
1764spa_load_verify_done(zio_t *zio)
1765{
1766 blkptr_t *bp = zio->io_bp;
1767 spa_load_error_t *sle = zio->io_private;
1768 dmu_object_type_t type = BP_GET_TYPE(bp);
1769 int error = zio->io_error;
1770
1771 if (error) {
1772 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1773 type != DMU_OT_INTENT_LOG)
1774 atomic_add_64(&sle->sle_meta_count, 1);
1775 else
1776 atomic_add_64(&sle->sle_data_count, 1);
1777 }
1778 zio_data_buf_free(zio->io_data, zio->io_size);
1779}
1780
1781/*ARGSUSED*/
1782static int
1783spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1784 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1785{
1786 if (bp != NULL) {
1787 zio_t *rio = arg;
1788 size_t size = BP_GET_PSIZE(bp);
1789 void *data = zio_data_buf_alloc(size);
1790
1791 zio_nowait(zio_read(rio, spa, bp, data, size,
1792 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1793 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1794 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1795 }
1796 return (0);
1797}
1798
1799static int
1800spa_load_verify(spa_t *spa)
1801{
1802 zio_t *rio;
1803 spa_load_error_t sle = { 0 };
1804 zpool_rewind_policy_t policy;
1805 boolean_t verify_ok = B_FALSE;
1806 int error;
1807
1808 zpool_get_rewind_policy(spa->spa_config, &policy);
1809
1810 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1811 return (0);
1812
1813 rio = zio_root(spa, NULL, &sle,
1814 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1815
1816 error = traverse_pool(spa, spa->spa_verify_min_txg,
1817 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1818
1819 (void) zio_wait(rio);
1820
1821 spa->spa_load_meta_errors = sle.sle_meta_count;
1822 spa->spa_load_data_errors = sle.sle_data_count;
1823
1824 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1825 sle.sle_data_count <= policy.zrp_maxdata) {
1826 int64_t loss = 0;
1827
1828 verify_ok = B_TRUE;
1829 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1830 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1831
1832 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1833 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1834 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1835 VERIFY(nvlist_add_int64(spa->spa_load_info,
1836 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1837 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1838 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1839 } else {
1840 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1841 }
1842
1843 if (error) {
1844 if (error != ENXIO && error != EIO)
1845 error = EIO;
1846 return (error);
1847 }
1848
1849 return (verify_ok ? 0 : EIO);
1850}
1851
1852/*
1853 * Find a value in the pool props object.
1854 */
1855static void
1856spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1857{
1858 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1859 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1860}
1861
1862/*
1863 * Find a value in the pool directory object.
1864 */
1865static int
1866spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1867{
1868 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1869 name, sizeof (uint64_t), 1, val));
1870}
1871
1872static int
1873spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1874{
1875 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1876 return (err);
1877}
1878
1879/*
1880 * Fix up config after a partly-completed split. This is done with the
1881 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1882 * pool have that entry in their config, but only the splitting one contains
1883 * a list of all the guids of the vdevs that are being split off.
1884 *
1885 * This function determines what to do with that list: either rejoin
1886 * all the disks to the pool, or complete the splitting process. To attempt
1887 * the rejoin, each disk that is offlined is marked online again, and
1888 * we do a reopen() call. If the vdev label for every disk that was
1889 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1890 * then we call vdev_split() on each disk, and complete the split.
1891 *
1892 * Otherwise we leave the config alone, with all the vdevs in place in
1893 * the original pool.
1894 */
1895static void
1896spa_try_repair(spa_t *spa, nvlist_t *config)
1897{
1898 uint_t extracted;
1899 uint64_t *glist;
1900 uint_t i, gcount;
1901 nvlist_t *nvl;
1902 vdev_t **vd;
1903 boolean_t attempt_reopen;
1904
1905 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1906 return;
1907
1908 /* check that the config is complete */
1909 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1910 &glist, &gcount) != 0)
1911 return;
1912
1913 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1914
1915 /* attempt to online all the vdevs & validate */
1916 attempt_reopen = B_TRUE;
1917 for (i = 0; i < gcount; i++) {
1918 if (glist[i] == 0) /* vdev is hole */
1919 continue;
1920
1921 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1922 if (vd[i] == NULL) {
1923 /*
1924 * Don't bother attempting to reopen the disks;
1925 * just do the split.
1926 */
1927 attempt_reopen = B_FALSE;
1928 } else {
1929 /* attempt to re-online it */
1930 vd[i]->vdev_offline = B_FALSE;
1931 }
1932 }
1933
1934 if (attempt_reopen) {
1935 vdev_reopen(spa->spa_root_vdev);
1936
1937 /* check each device to see what state it's in */
1938 for (extracted = 0, i = 0; i < gcount; i++) {
1939 if (vd[i] != NULL &&
1940 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1941 break;
1942 ++extracted;
1943 }
1944 }
1945
1946 /*
1947 * If every disk has been moved to the new pool, or if we never
1948 * even attempted to look at them, then we split them off for
1949 * good.
1950 */
1951 if (!attempt_reopen || gcount == extracted) {
1952 for (i = 0; i < gcount; i++)
1953 if (vd[i] != NULL)
1954 vdev_split(vd[i]);
1955 vdev_reopen(spa->spa_root_vdev);
1956 }
1957
1958 kmem_free(vd, gcount * sizeof (vdev_t *));
1959}
1960
1961static int
1962spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1963 boolean_t mosconfig)
1964{
1965 nvlist_t *config = spa->spa_config;
1966 char *ereport = FM_EREPORT_ZFS_POOL;
1967 char *comment;
1968 int error;
1969 uint64_t pool_guid;
1970 nvlist_t *nvl;
1971
1972 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1973 return (EINVAL);
1974
1975 ASSERT(spa->spa_comment == NULL);
1976 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1977 spa->spa_comment = spa_strdup(comment);
1978
1979 /*
1980 * Versioning wasn't explicitly added to the label until later, so if
1981 * it's not present treat it as the initial version.
1982 */
1983 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1984 &spa->spa_ubsync.ub_version) != 0)
1985 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1986
1987 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1988 &spa->spa_config_txg);
1989
1990 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1991 spa_guid_exists(pool_guid, 0)) {
1992 error = EEXIST;
1993 } else {
1994 spa->spa_config_guid = pool_guid;
1995
1996 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1997 &nvl) == 0) {
1998 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1999 KM_SLEEP) == 0);
2000 }
2001
2002 nvlist_free(spa->spa_load_info);
2003 spa->spa_load_info = fnvlist_alloc();
2004
2005 gethrestime(&spa->spa_loaded_ts);
2006 error = spa_load_impl(spa, pool_guid, config, state, type,
2007 mosconfig, &ereport);
2008 }
2009
2010 spa->spa_minref = refcount_count(&spa->spa_refcount);
2011 if (error) {
2012 if (error != EEXIST) {
2013 spa->spa_loaded_ts.tv_sec = 0;
2014 spa->spa_loaded_ts.tv_nsec = 0;
2015 }
2016 if (error != EBADF) {
2017 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2018 }
2019 }
2020 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2021 spa->spa_ena = 0;
2022
2023 return (error);
2024}
2025
2026/*
2027 * Load an existing storage pool, using the pool's builtin spa_config as a
2028 * source of configuration information.
2029 */
2030static int
2031spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2032 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2033 char **ereport)
2034{
2035 int error = 0;
2036 nvlist_t *nvroot = NULL;
2037 nvlist_t *label;
2038 vdev_t *rvd;
2039 uberblock_t *ub = &spa->spa_uberblock;
2040 uint64_t children, config_cache_txg = spa->spa_config_txg;
2041 int orig_mode = spa->spa_mode;
2042 int parse;
2043 uint64_t obj;
2044 boolean_t missing_feat_write = B_FALSE;
2045
2046 /*
2047 * If this is an untrusted config, access the pool in read-only mode.
2048 * This prevents things like resilvering recently removed devices.
2049 */
2050 if (!mosconfig)
2051 spa->spa_mode = FREAD;
2052
2053 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2054
2055 spa->spa_load_state = state;
2056
2057 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2058 return (EINVAL);
2059
2060 parse = (type == SPA_IMPORT_EXISTING ?
2061 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2062
2063 /*
2064 * Create "The Godfather" zio to hold all async IOs
2065 */
2066 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2067 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2068
2069 /*
2070 * Parse the configuration into a vdev tree. We explicitly set the
2071 * value that will be returned by spa_version() since parsing the
2072 * configuration requires knowing the version number.
2073 */
2074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2075 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2076 spa_config_exit(spa, SCL_ALL, FTAG);
2077
2078 if (error != 0)
2079 return (error);
2080
2081 ASSERT(spa->spa_root_vdev == rvd);
2082
2083 if (type != SPA_IMPORT_ASSEMBLE) {
2084 ASSERT(spa_guid(spa) == pool_guid);
2085 }
2086
2087 /*
2088 * Try to open all vdevs, loading each label in the process.
2089 */
2090 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2091 error = vdev_open(rvd);
2092 spa_config_exit(spa, SCL_ALL, FTAG);
2093 if (error != 0)
2094 return (error);
2095
2096 /*
2097 * We need to validate the vdev labels against the configuration that
2098 * we have in hand, which is dependent on the setting of mosconfig. If
2099 * mosconfig is true then we're validating the vdev labels based on
2100 * that config. Otherwise, we're validating against the cached config
2101 * (zpool.cache) that was read when we loaded the zfs module, and then
2102 * later we will recursively call spa_load() and validate against
2103 * the vdev config.
2104 *
2105 * If we're assembling a new pool that's been split off from an
2106 * existing pool, the labels haven't yet been updated so we skip
2107 * validation for now.
2108 */
2109 if (type != SPA_IMPORT_ASSEMBLE) {
2110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2111 error = vdev_validate(rvd, mosconfig);
2112 spa_config_exit(spa, SCL_ALL, FTAG);
2113
2114 if (error != 0)
2115 return (error);
2116
2117 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2118 return (ENXIO);
2119 }
2120
2121 /*
2122 * Find the best uberblock.
2123 */
2124 vdev_uberblock_load(rvd, ub, &label);
2125
2126 /*
2127 * If we weren't able to find a single valid uberblock, return failure.
2128 */
2129 if (ub->ub_txg == 0) {
2130 nvlist_free(label);
2131 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2132 }
2133
2134 /*
2135 * If the pool has an unsupported version we can't open it.
2136 */
2137 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2138 nvlist_free(label);
2139 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2140 }
2141
2142 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2143 nvlist_t *features;
2144
2145 /*
2146 * If we weren't able to find what's necessary for reading the
2147 * MOS in the label, return failure.
2148 */
2149 if (label == NULL || nvlist_lookup_nvlist(label,
2150 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2151 nvlist_free(label);
2152 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2153 ENXIO));
2154 }
2155
2156 /*
2157 * Update our in-core representation with the definitive values
2158 * from the label.
2159 */
2160 nvlist_free(spa->spa_label_features);
2161 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2162 }
2163
2164 nvlist_free(label);
2165
2166 /*
2167 * Look through entries in the label nvlist's features_for_read. If
2168 * there is a feature listed there which we don't understand then we
2169 * cannot open a pool.
2170 */
2171 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2172 nvlist_t *unsup_feat;
2173
2174 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2175 0);
2176
2177 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2178 NULL); nvp != NULL;
2179 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2180 if (!zfeature_is_supported(nvpair_name(nvp))) {
2181 VERIFY(nvlist_add_string(unsup_feat,
2182 nvpair_name(nvp), "") == 0);
2183 }
2184 }
2185
2186 if (!nvlist_empty(unsup_feat)) {
2187 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2188 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2189 nvlist_free(unsup_feat);
2190 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2191 ENOTSUP));
2192 }
2193
2194 nvlist_free(unsup_feat);
2195 }
2196
2197 /*
2198 * If the vdev guid sum doesn't match the uberblock, we have an
2199 * incomplete configuration. We first check to see if the pool
2200 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2201 * If it is, defer the vdev_guid_sum check till later so we
2202 * can handle missing vdevs.
2203 */
2204 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2205 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2206 rvd->vdev_guid_sum != ub->ub_guid_sum)
2207 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2208
2209 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2210 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2211 spa_try_repair(spa, config);
2212 spa_config_exit(spa, SCL_ALL, FTAG);
2213 nvlist_free(spa->spa_config_splitting);
2214 spa->spa_config_splitting = NULL;
2215 }
2216
2217 /*
2218 * Initialize internal SPA structures.
2219 */
2220 spa->spa_state = POOL_STATE_ACTIVE;
2221 spa->spa_ubsync = spa->spa_uberblock;
2222 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2223 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2224 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2225 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2226 spa->spa_claim_max_txg = spa->spa_first_txg;
2227 spa->spa_prev_software_version = ub->ub_software_version;
2228
2229 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2230 if (error)
2231 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2232 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2233
2234 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2235 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2236
2237 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2238 boolean_t missing_feat_read = B_FALSE;
2239 nvlist_t *unsup_feat, *enabled_feat;
2240
2241 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2242 &spa->spa_feat_for_read_obj) != 0) {
2243 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2244 }
2245
2246 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2247 &spa->spa_feat_for_write_obj) != 0) {
2248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2249 }
2250
2251 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2252 &spa->spa_feat_desc_obj) != 0) {
2253 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2254 }
2255
2256 enabled_feat = fnvlist_alloc();
2257 unsup_feat = fnvlist_alloc();
2258
2259 if (!feature_is_supported(spa->spa_meta_objset,
2260 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2261 unsup_feat, enabled_feat))
2262 missing_feat_read = B_TRUE;
2263
2264 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2265 if (!feature_is_supported(spa->spa_meta_objset,
2266 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2267 unsup_feat, enabled_feat)) {
2268 missing_feat_write = B_TRUE;
2269 }
2270 }
2271
2272 fnvlist_add_nvlist(spa->spa_load_info,
2273 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2274
2275 if (!nvlist_empty(unsup_feat)) {
2276 fnvlist_add_nvlist(spa->spa_load_info,
2277 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2278 }
2279
2280 fnvlist_free(enabled_feat);
2281 fnvlist_free(unsup_feat);
2282
2283 if (!missing_feat_read) {
2284 fnvlist_add_boolean(spa->spa_load_info,
2285 ZPOOL_CONFIG_CAN_RDONLY);
2286 }
2287
2288 /*
2289 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2290 * twofold: to determine whether the pool is available for
2291 * import in read-write mode and (if it is not) whether the
2292 * pool is available for import in read-only mode. If the pool
2293 * is available for import in read-write mode, it is displayed
2294 * as available in userland; if it is not available for import
2295 * in read-only mode, it is displayed as unavailable in
2296 * userland. If the pool is available for import in read-only
2297 * mode but not read-write mode, it is displayed as unavailable
2298 * in userland with a special note that the pool is actually
2299 * available for open in read-only mode.
2300 *
2301 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2302 * missing a feature for write, we must first determine whether
2303 * the pool can be opened read-only before returning to
2304 * userland in order to know whether to display the
2305 * abovementioned note.
2306 */
2307 if (missing_feat_read || (missing_feat_write &&
2308 spa_writeable(spa))) {
2309 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2310 ENOTSUP));
2311 }
2312 }
2313
2314 spa->spa_is_initializing = B_TRUE;
2315 error = dsl_pool_open(spa->spa_dsl_pool);
2316 spa->spa_is_initializing = B_FALSE;
2317 if (error != 0)
2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2319
2320 if (!mosconfig) {
2321 uint64_t hostid;
2322 nvlist_t *policy = NULL, *nvconfig;
2323
2324 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2326
2327 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2328 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2329 char *hostname;
2330 unsigned long myhostid = 0;
2331
2332 VERIFY(nvlist_lookup_string(nvconfig,
2333 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2334
2335#ifdef _KERNEL
2336 myhostid = zone_get_hostid(NULL);
2337#else /* _KERNEL */
2338 /*
2339 * We're emulating the system's hostid in userland, so
2340 * we can't use zone_get_hostid().
2341 */
2342 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2343#endif /* _KERNEL */
2344 if (check_hostid && hostid != 0 && myhostid != 0 &&
2345 hostid != myhostid) {
2346 nvlist_free(nvconfig);
2347 cmn_err(CE_WARN, "pool '%s' could not be "
2348 "loaded as it was last accessed by "
2349 "another system (host: %s hostid: 0x%lx). "
2350 "See: http://illumos.org/msg/ZFS-8000-EY",
2351 spa_name(spa), hostname,
2352 (unsigned long)hostid);
2353 return (EBADF);
2354 }
2355 }
2356 if (nvlist_lookup_nvlist(spa->spa_config,
2357 ZPOOL_REWIND_POLICY, &policy) == 0)
2358 VERIFY(nvlist_add_nvlist(nvconfig,
2359 ZPOOL_REWIND_POLICY, policy) == 0);
2360
2361 spa_config_set(spa, nvconfig);
2362 spa_unload(spa);
2363 spa_deactivate(spa);
2364 spa_activate(spa, orig_mode);
2365
2366 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2367 }
2368
2369 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2371 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2372 if (error != 0)
2373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2374
2375 /*
2376 * Load the bit that tells us to use the new accounting function
2377 * (raid-z deflation). If we have an older pool, this will not
2378 * be present.
2379 */
2380 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2381 if (error != 0 && error != ENOENT)
2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2383
2384 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2385 &spa->spa_creation_version);
2386 if (error != 0 && error != ENOENT)
2387 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2388
2389 /*
2390 * Load the persistent error log. If we have an older pool, this will
2391 * not be present.
2392 */
2393 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2394 if (error != 0 && error != ENOENT)
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2396
2397 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2398 &spa->spa_errlog_scrub);
2399 if (error != 0 && error != ENOENT)
2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2401
2402 /*
2403 * Load the history object. If we have an older pool, this
2404 * will not be present.
2405 */
2406 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2407 if (error != 0 && error != ENOENT)
2408 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2409
2410 /*
2411 * If we're assembling the pool from the split-off vdevs of
2412 * an existing pool, we don't want to attach the spares & cache
2413 * devices.
2414 */
2415
2416 /*
2417 * Load any hot spares for this pool.
2418 */
2419 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2420 if (error != 0 && error != ENOENT)
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2423 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2424 if (load_nvlist(spa, spa->spa_spares.sav_object,
2425 &spa->spa_spares.sav_config) != 0)
2426 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427
2428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2429 spa_load_spares(spa);
2430 spa_config_exit(spa, SCL_ALL, FTAG);
2431 } else if (error == 0) {
2432 spa->spa_spares.sav_sync = B_TRUE;
2433 }
2434
2435 /*
2436 * Load any level 2 ARC devices for this pool.
2437 */
2438 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2439 &spa->spa_l2cache.sav_object);
2440 if (error != 0 && error != ENOENT)
2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2442 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2443 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2444 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2445 &spa->spa_l2cache.sav_config) != 0)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447
2448 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2449 spa_load_l2cache(spa);
2450 spa_config_exit(spa, SCL_ALL, FTAG);
2451 } else if (error == 0) {
2452 spa->spa_l2cache.sav_sync = B_TRUE;
2453 }
2454
2455 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2456
2457 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2458 if (error && error != ENOENT)
2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2460
2461 if (error == 0) {
2462 uint64_t autoreplace;
2463
2464 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2465 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2466 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2467 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2468 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2469 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2470 &spa->spa_dedup_ditto);
2471
2472 spa->spa_autoreplace = (autoreplace != 0);
2473 }
2474
2475 /*
2476 * If the 'autoreplace' property is set, then post a resource notifying
2477 * the ZFS DE that it should not issue any faults for unopenable
2478 * devices. We also iterate over the vdevs, and post a sysevent for any
2479 * unopenable vdevs so that the normal autoreplace handler can take
2480 * over.
2481 */
2482 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2483 spa_check_removed(spa->spa_root_vdev);
2484 /*
2485 * For the import case, this is done in spa_import(), because
2486 * at this point we're using the spare definitions from
2487 * the MOS config, not necessarily from the userland config.
2488 */
2489 if (state != SPA_LOAD_IMPORT) {
2490 spa_aux_check_removed(&spa->spa_spares);
2491 spa_aux_check_removed(&spa->spa_l2cache);
2492 }
2493 }
2494
2495 /*
2496 * Load the vdev state for all toplevel vdevs.
2497 */
2498 vdev_load(rvd);
2499
2500 /*
2501 * Propagate the leaf DTLs we just loaded all the way up the tree.
2502 */
2503 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2504 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2505 spa_config_exit(spa, SCL_ALL, FTAG);
2506
2507 /*
2508 * Load the DDTs (dedup tables).
2509 */
2510 error = ddt_load(spa);
2511 if (error != 0)
2512 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2513
2514 spa_update_dspace(spa);
2515
2516 /*
2517 * Validate the config, using the MOS config to fill in any
2518 * information which might be missing. If we fail to validate
2519 * the config then declare the pool unfit for use. If we're
2520 * assembling a pool from a split, the log is not transferred
2521 * over.
2522 */
2523 if (type != SPA_IMPORT_ASSEMBLE) {
2524 nvlist_t *nvconfig;
2525
2526 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528
2529 if (!spa_config_valid(spa, nvconfig)) {
2530 nvlist_free(nvconfig);
2531 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2532 ENXIO));
2533 }
2534 nvlist_free(nvconfig);
2535
2536 /*
2537 * Now that we've validated the config, check the state of the
2538 * root vdev. If it can't be opened, it indicates one or
2539 * more toplevel vdevs are faulted.
2540 */
2541 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2542 return (ENXIO);
2543
2544 if (spa_check_logs(spa)) {
2545 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2546 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2547 }
2548 }
2549
2550 if (missing_feat_write) {
2551 ASSERT(state == SPA_LOAD_TRYIMPORT);
2552
2553 /*
2554 * At this point, we know that we can open the pool in
2555 * read-only mode but not read-write mode. We now have enough
2556 * information and can return to userland.
2557 */
2558 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2559 }
2560
2561 /*
2562 * We've successfully opened the pool, verify that we're ready
2563 * to start pushing transactions.
2564 */
2565 if (state != SPA_LOAD_TRYIMPORT) {
2566 if (error = spa_load_verify(spa))
2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2568 error));
2569 }
2570
2571 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2572 spa->spa_load_max_txg == UINT64_MAX)) {
2573 dmu_tx_t *tx;
2574 int need_update = B_FALSE;
2575
2576 ASSERT(state != SPA_LOAD_TRYIMPORT);
2577
2578 /*
2579 * Claim log blocks that haven't been committed yet.
2580 * This must all happen in a single txg.
2581 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2582 * invoked from zil_claim_log_block()'s i/o done callback.
2583 * Price of rollback is that we abandon the log.
2584 */
2585 spa->spa_claiming = B_TRUE;
2586
2587 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2588 spa_first_txg(spa));
2589 (void) dmu_objset_find(spa_name(spa),
2590 zil_claim, tx, DS_FIND_CHILDREN);
2591 dmu_tx_commit(tx);
2592
2593 spa->spa_claiming = B_FALSE;
2594
2595 spa_set_log_state(spa, SPA_LOG_GOOD);
2596 spa->spa_sync_on = B_TRUE;
2597 txg_sync_start(spa->spa_dsl_pool);
2598
2599 /*
2600 * Wait for all claims to sync. We sync up to the highest
2601 * claimed log block birth time so that claimed log blocks
2602 * don't appear to be from the future. spa_claim_max_txg
2603 * will have been set for us by either zil_check_log_chain()
2604 * (invoked from spa_check_logs()) or zil_claim() above.
2605 */
2606 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2607
2608 /*
2609 * If the config cache is stale, or we have uninitialized
2610 * metaslabs (see spa_vdev_add()), then update the config.
2611 *
2612 * If this is a verbatim import, trust the current
2613 * in-core spa_config and update the disk labels.
2614 */
2615 if (config_cache_txg != spa->spa_config_txg ||
2616 state == SPA_LOAD_IMPORT ||
2617 state == SPA_LOAD_RECOVER ||
2618 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2619 need_update = B_TRUE;
2620
2621 for (int c = 0; c < rvd->vdev_children; c++)
2622 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2623 need_update = B_TRUE;
2624
2625 /*
2626 * Update the config cache asychronously in case we're the
2627 * root pool, in which case the config cache isn't writable yet.
2628 */
2629 if (need_update)
2630 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2631
2632 /*
2633 * Check all DTLs to see if anything needs resilvering.
2634 */
2635 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2636 vdev_resilver_needed(rvd, NULL, NULL))
2637 spa_async_request(spa, SPA_ASYNC_RESILVER);
2638
2639 /*
2640 * Delete any inconsistent datasets.
2641 */
2642 (void) dmu_objset_find(spa_name(spa),
2643 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2644
2645 /*
2646 * Clean up any stale temporary dataset userrefs.
2647 */
2648 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2649 }
2650
2651 return (0);
2652}
2653
2654static int
2655spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2656{
2657 int mode = spa->spa_mode;
2658
2659 spa_unload(spa);
2660 spa_deactivate(spa);
2661
2662 spa->spa_load_max_txg--;
2663
2664 spa_activate(spa, mode);
2665 spa_async_suspend(spa);
2666
2667 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2668}
2669
2670/*
2671 * If spa_load() fails this function will try loading prior txg's. If
2672 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2673 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2674 * function will not rewind the pool and will return the same error as
2675 * spa_load().
2676 */
2677static int
2678spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2679 uint64_t max_request, int rewind_flags)
2680{
2681 nvlist_t *loadinfo = NULL;
2682 nvlist_t *config = NULL;
2683 int load_error, rewind_error;
2684 uint64_t safe_rewind_txg;
2685 uint64_t min_txg;
2686
2687 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2688 spa->spa_load_max_txg = spa->spa_load_txg;
2689 spa_set_log_state(spa, SPA_LOG_CLEAR);
2690 } else {
2691 spa->spa_load_max_txg = max_request;
2692 }
2693
2694 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2695 mosconfig);
2696 if (load_error == 0)
2697 return (0);
2698
2699 if (spa->spa_root_vdev != NULL)
2700 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2701
2702 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2703 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2704
2705 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2706 nvlist_free(config);
2707 return (load_error);
2708 }
2709
2710 if (state == SPA_LOAD_RECOVER) {
2711 /* Price of rolling back is discarding txgs, including log */
2712 spa_set_log_state(spa, SPA_LOG_CLEAR);
2713 } else {
2714 /*
2715 * If we aren't rolling back save the load info from our first
2716 * import attempt so that we can restore it after attempting
2717 * to rewind.
2718 */
2719 loadinfo = spa->spa_load_info;
2720 spa->spa_load_info = fnvlist_alloc();
2721 }
2722
2723 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2724 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2725 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2726 TXG_INITIAL : safe_rewind_txg;
2727
2728 /*
2729 * Continue as long as we're finding errors, we're still within
2730 * the acceptable rewind range, and we're still finding uberblocks
2731 */
2732 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2733 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2734 if (spa->spa_load_max_txg < safe_rewind_txg)
2735 spa->spa_extreme_rewind = B_TRUE;
2736 rewind_error = spa_load_retry(spa, state, mosconfig);
2737 }
2738
2739 spa->spa_extreme_rewind = B_FALSE;
2740 spa->spa_load_max_txg = UINT64_MAX;
2741
2742 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2743 spa_config_set(spa, config);
2744
2745 if (state == SPA_LOAD_RECOVER) {
2746 ASSERT3P(loadinfo, ==, NULL);
2747 return (rewind_error);
2748 } else {
2749 /* Store the rewind info as part of the initial load info */
2750 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2751 spa->spa_load_info);
2752
2753 /* Restore the initial load info */
2754 fnvlist_free(spa->spa_load_info);
2755 spa->spa_load_info = loadinfo;
2756
2757 return (load_error);
2758 }
2759}
2760
2761/*
2762 * Pool Open/Import
2763 *
2764 * The import case is identical to an open except that the configuration is sent
2765 * down from userland, instead of grabbed from the configuration cache. For the
2766 * case of an open, the pool configuration will exist in the
2767 * POOL_STATE_UNINITIALIZED state.
2768 *
2769 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2770 * the same time open the pool, without having to keep around the spa_t in some
2771 * ambiguous state.
2772 */
2773static int
2774spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2775 nvlist_t **config)
2776{
2777 spa_t *spa;
2778 spa_load_state_t state = SPA_LOAD_OPEN;
2779 int error;
2780 int locked = B_FALSE;
2781 int firstopen = B_FALSE;
2782
2783 *spapp = NULL;
2784
2785 /*
2786 * As disgusting as this is, we need to support recursive calls to this
2787 * function because dsl_dir_open() is called during spa_load(), and ends
2788 * up calling spa_open() again. The real fix is to figure out how to
2789 * avoid dsl_dir_open() calling this in the first place.
2790 */
2791 if (mutex_owner(&spa_namespace_lock) != curthread) {
2792 mutex_enter(&spa_namespace_lock);
2793 locked = B_TRUE;
2794 }
2795
2796 if ((spa = spa_lookup(pool)) == NULL) {
2797 if (locked)
2798 mutex_exit(&spa_namespace_lock);
2799 return (ENOENT);
2800 }
2801
2802 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2803 zpool_rewind_policy_t policy;
2804
2805 firstopen = B_TRUE;
2806
2807 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2808 &policy);
2809 if (policy.zrp_request & ZPOOL_DO_REWIND)
2810 state = SPA_LOAD_RECOVER;
2811
2812 spa_activate(spa, spa_mode_global);
2813
2814 if (state != SPA_LOAD_RECOVER)
2815 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2816
2817 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2818 policy.zrp_request);
2819
2820 if (error == EBADF) {
2821 /*
2822 * If vdev_validate() returns failure (indicated by
2823 * EBADF), it indicates that one of the vdevs indicates
2824 * that the pool has been exported or destroyed. If
2825 * this is the case, the config cache is out of sync and
2826 * we should remove the pool from the namespace.
2827 */
2828 spa_unload(spa);
2829 spa_deactivate(spa);
2830 spa_config_sync(spa, B_TRUE, B_TRUE);
2831 spa_remove(spa);
2832 if (locked)
2833 mutex_exit(&spa_namespace_lock);
2834 return (ENOENT);
2835 }
2836
2837 if (error) {
2838 /*
2839 * We can't open the pool, but we still have useful
2840 * information: the state of each vdev after the
2841 * attempted vdev_open(). Return this to the user.
2842 */
2843 if (config != NULL && spa->spa_config) {
2844 VERIFY(nvlist_dup(spa->spa_config, config,
2845 KM_SLEEP) == 0);
2846 VERIFY(nvlist_add_nvlist(*config,
2847 ZPOOL_CONFIG_LOAD_INFO,
2848 spa->spa_load_info) == 0);
2849 }
2850 spa_unload(spa);
2851 spa_deactivate(spa);
2852 spa->spa_last_open_failed = error;
2853 if (locked)
2854 mutex_exit(&spa_namespace_lock);
2855 *spapp = NULL;
2856 return (error);
2857 }
2858 }
2859
2860 spa_open_ref(spa, tag);
2861
2862 if (config != NULL)
2863 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2864
2865 /*
2866 * If we've recovered the pool, pass back any information we
2867 * gathered while doing the load.
2868 */
2869 if (state == SPA_LOAD_RECOVER) {
2870 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2871 spa->spa_load_info) == 0);
2872 }
2873
2874 if (locked) {
2875 spa->spa_last_open_failed = 0;
2876 spa->spa_last_ubsync_txg = 0;
2877 spa->spa_load_txg = 0;
2878 mutex_exit(&spa_namespace_lock);
2879#ifdef __FreeBSD__
2880#ifdef _KERNEL
2881 if (firstopen)
2882 zvol_create_minors(pool);
2883#endif
2884#endif
2885 }
2886
2887 *spapp = spa;
2888
2889 return (0);
2890}
2891
2892int
2893spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2894 nvlist_t **config)
2895{
2896 return (spa_open_common(name, spapp, tag, policy, config));
2897}
2898
2899int
2900spa_open(const char *name, spa_t **spapp, void *tag)
2901{
2902 return (spa_open_common(name, spapp, tag, NULL, NULL));
2903}
2904
2905/*
2906 * Lookup the given spa_t, incrementing the inject count in the process,
2907 * preventing it from being exported or destroyed.
2908 */
2909spa_t *
2910spa_inject_addref(char *name)
2911{
2912 spa_t *spa;
2913
2914 mutex_enter(&spa_namespace_lock);
2915 if ((spa = spa_lookup(name)) == NULL) {
2916 mutex_exit(&spa_namespace_lock);
2917 return (NULL);
2918 }
2919 spa->spa_inject_ref++;
2920 mutex_exit(&spa_namespace_lock);
2921
2922 return (spa);
2923}
2924
2925void
2926spa_inject_delref(spa_t *spa)
2927{
2928 mutex_enter(&spa_namespace_lock);
2929 spa->spa_inject_ref--;
2930 mutex_exit(&spa_namespace_lock);
2931}
2932
2933/*
2934 * Add spares device information to the nvlist.
2935 */
2936static void
2937spa_add_spares(spa_t *spa, nvlist_t *config)
2938{
2939 nvlist_t **spares;
2940 uint_t i, nspares;
2941 nvlist_t *nvroot;
2942 uint64_t guid;
2943 vdev_stat_t *vs;
2944 uint_t vsc;
2945 uint64_t pool;
2946
2947 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2948
2949 if (spa->spa_spares.sav_count == 0)
2950 return;
2951
2952 VERIFY(nvlist_lookup_nvlist(config,
2953 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2954 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2955 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2956 if (nspares != 0) {
2957 VERIFY(nvlist_add_nvlist_array(nvroot,
2958 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2959 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2960 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2961
2962 /*
2963 * Go through and find any spares which have since been
2964 * repurposed as an active spare. If this is the case, update
2965 * their status appropriately.
2966 */
2967 for (i = 0; i < nspares; i++) {
2968 VERIFY(nvlist_lookup_uint64(spares[i],
2969 ZPOOL_CONFIG_GUID, &guid) == 0);
2970 if (spa_spare_exists(guid, &pool, NULL) &&
2971 pool != 0ULL) {
2972 VERIFY(nvlist_lookup_uint64_array(
2973 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2974 (uint64_t **)&vs, &vsc) == 0);
2975 vs->vs_state = VDEV_STATE_CANT_OPEN;
2976 vs->vs_aux = VDEV_AUX_SPARED;
2977 }
2978 }
2979 }
2980}
2981
2982/*
2983 * Add l2cache device information to the nvlist, including vdev stats.
2984 */
2985static void
2986spa_add_l2cache(spa_t *spa, nvlist_t *config)
2987{
2988 nvlist_t **l2cache;
2989 uint_t i, j, nl2cache;
2990 nvlist_t *nvroot;
2991 uint64_t guid;
2992 vdev_t *vd;
2993 vdev_stat_t *vs;
2994 uint_t vsc;
2995
2996 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2997
2998 if (spa->spa_l2cache.sav_count == 0)
2999 return;
3000
3001 VERIFY(nvlist_lookup_nvlist(config,
3002 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3003 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3004 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3005 if (nl2cache != 0) {
3006 VERIFY(nvlist_add_nvlist_array(nvroot,
3007 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3008 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3009 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3010
3011 /*
3012 * Update level 2 cache device stats.
3013 */
3014
3015 for (i = 0; i < nl2cache; i++) {
3016 VERIFY(nvlist_lookup_uint64(l2cache[i],
3017 ZPOOL_CONFIG_GUID, &guid) == 0);
3018
3019 vd = NULL;
3020 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3021 if (guid ==
3022 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3023 vd = spa->spa_l2cache.sav_vdevs[j];
3024 break;
3025 }
3026 }
3027 ASSERT(vd != NULL);
3028
3029 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3030 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3031 == 0);
3032 vdev_get_stats(vd, vs);
3033 }
3034 }
3035}
3036
3037static void
3038spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3039{
3040 nvlist_t *features;
3041 zap_cursor_t zc;
3042 zap_attribute_t za;
3043
3044 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3045 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3046
3047 if (spa->spa_feat_for_read_obj != 0) {
3048 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3049 spa->spa_feat_for_read_obj);
3050 zap_cursor_retrieve(&zc, &za) == 0;
3051 zap_cursor_advance(&zc)) {
3052 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3053 za.za_num_integers == 1);
3054 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3055 za.za_first_integer));
3056 }
3057 zap_cursor_fini(&zc);
3058 }
3059
3060 if (spa->spa_feat_for_write_obj != 0) {
3061 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3062 spa->spa_feat_for_write_obj);
3063 zap_cursor_retrieve(&zc, &za) == 0;
3064 zap_cursor_advance(&zc)) {
3065 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3066 za.za_num_integers == 1);
3067 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3068 za.za_first_integer));
3069 }
3070 zap_cursor_fini(&zc);
3071 }
3072
3073 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3074 features) == 0);
3075 nvlist_free(features);
3076}
3077
3078int
3079spa_get_stats(const char *name, nvlist_t **config,
3080 char *altroot, size_t buflen)
3081{
3082 int error;
3083 spa_t *spa;
3084
3085 *config = NULL;
3086 error = spa_open_common(name, &spa, FTAG, NULL, config);
3087
3088 if (spa != NULL) {
3089 /*
3090 * This still leaves a window of inconsistency where the spares
3091 * or l2cache devices could change and the config would be
3092 * self-inconsistent.
3093 */
3094 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3095
3096 if (*config != NULL) {
3097 uint64_t loadtimes[2];
3098
3099 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3100 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3101 VERIFY(nvlist_add_uint64_array(*config,
3102 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3103
3104 VERIFY(nvlist_add_uint64(*config,
3105 ZPOOL_CONFIG_ERRCOUNT,
3106 spa_get_errlog_size(spa)) == 0);
3107
3108 if (spa_suspended(spa))
3109 VERIFY(nvlist_add_uint64(*config,
3110 ZPOOL_CONFIG_SUSPENDED,
3111 spa->spa_failmode) == 0);
3112
3113 spa_add_spares(spa, *config);
3114 spa_add_l2cache(spa, *config);
3115 spa_add_feature_stats(spa, *config);
3116 }
3117 }
3118
3119 /*
3120 * We want to get the alternate root even for faulted pools, so we cheat
3121 * and call spa_lookup() directly.
3122 */
3123 if (altroot) {
3124 if (spa == NULL) {
3125 mutex_enter(&spa_namespace_lock);
3126 spa = spa_lookup(name);
3127 if (spa)
3128 spa_altroot(spa, altroot, buflen);
3129 else
3130 altroot[0] = '\0';
3131 spa = NULL;
3132 mutex_exit(&spa_namespace_lock);
3133 } else {
3134 spa_altroot(spa, altroot, buflen);
3135 }
3136 }
3137
3138 if (spa != NULL) {
3139 spa_config_exit(spa, SCL_CONFIG, FTAG);
3140 spa_close(spa, FTAG);
3141 }
3142
3143 return (error);
3144}
3145
3146/*
3147 * Validate that the auxiliary device array is well formed. We must have an
3148 * array of nvlists, each which describes a valid leaf vdev. If this is an
3149 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3150 * specified, as long as they are well-formed.
3151 */
3152static int
3153spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3154 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3155 vdev_labeltype_t label)
3156{
3157 nvlist_t **dev;
3158 uint_t i, ndev;
3159 vdev_t *vd;
3160 int error;
3161
3162 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3163
3164 /*
3165 * It's acceptable to have no devs specified.
3166 */
3167 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3168 return (0);
3169
3170 if (ndev == 0)
3171 return (EINVAL);
3172
3173 /*
3174 * Make sure the pool is formatted with a version that supports this
3175 * device type.
3176 */
3177 if (spa_version(spa) < version)
3178 return (ENOTSUP);
3179
3180 /*
3181 * Set the pending device list so we correctly handle device in-use
3182 * checking.
3183 */
3184 sav->sav_pending = dev;
3185 sav->sav_npending = ndev;
3186
3187 for (i = 0; i < ndev; i++) {
3188 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3189 mode)) != 0)
3190 goto out;
3191
3192 if (!vd->vdev_ops->vdev_op_leaf) {
3193 vdev_free(vd);
3194 error = EINVAL;
3195 goto out;
3196 }
3197
3198 /*
3199 * The L2ARC currently only supports disk devices in
3200 * kernel context. For user-level testing, we allow it.
3201 */
3202#ifdef _KERNEL
3203 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3204 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3205 error = ENOTBLK;
3206 vdev_free(vd);
3207 goto out;
3208 }
3209#endif
3210 vd->vdev_top = vd;
3211
3212 if ((error = vdev_open(vd)) == 0 &&
3213 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3214 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3215 vd->vdev_guid) == 0);
3216 }
3217
3218 vdev_free(vd);
3219
3220 if (error &&
3221 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3222 goto out;
3223 else
3224 error = 0;
3225 }
3226
3227out:
3228 sav->sav_pending = NULL;
3229 sav->sav_npending = 0;
3230 return (error);
3231}
3232
3233static int
3234spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3235{
3236 int error;
3237
3238 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3239
3240 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3241 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3242 VDEV_LABEL_SPARE)) != 0) {
3243 return (error);
3244 }
3245
3246 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3247 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3248 VDEV_LABEL_L2CACHE));
3249}
3250
3251static void
3252spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3253 const char *config)
3254{
3255 int i;
3256
3257 if (sav->sav_config != NULL) {
3258 nvlist_t **olddevs;
3259 uint_t oldndevs;
3260 nvlist_t **newdevs;
3261
3262 /*
3263 * Generate new dev list by concatentating with the
3264 * current dev list.
3265 */
3266 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3267 &olddevs, &oldndevs) == 0);
3268
3269 newdevs = kmem_alloc(sizeof (void *) *
3270 (ndevs + oldndevs), KM_SLEEP);
3271 for (i = 0; i < oldndevs; i++)
3272 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3273 KM_SLEEP) == 0);
3274 for (i = 0; i < ndevs; i++)
3275 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3276 KM_SLEEP) == 0);
3277
3278 VERIFY(nvlist_remove(sav->sav_config, config,
3279 DATA_TYPE_NVLIST_ARRAY) == 0);
3280
3281 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3282 config, newdevs, ndevs + oldndevs) == 0);
3283 for (i = 0; i < oldndevs + ndevs; i++)
3284 nvlist_free(newdevs[i]);
3285 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3286 } else {
3287 /*
3288 * Generate a new dev list.
3289 */
3290 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3291 KM_SLEEP) == 0);
3292 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3293 devs, ndevs) == 0);
3294 }
3295}
3296
3297/*
3298 * Stop and drop level 2 ARC devices
3299 */
3300void
3301spa_l2cache_drop(spa_t *spa)
3302{
3303 vdev_t *vd;
3304 int i;
3305 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3306
3307 for (i = 0; i < sav->sav_count; i++) {
3308 uint64_t pool;
3309
3310 vd = sav->sav_vdevs[i];
3311 ASSERT(vd != NULL);
3312
3313 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3314 pool != 0ULL && l2arc_vdev_present(vd))
3315 l2arc_remove_vdev(vd);
3316 }
3317}
3318
3319/*
3320 * Pool Creation
3321 */
3322int
3323spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3324 const char *history_str, nvlist_t *zplprops)
3325{
3326 spa_t *spa;
3327 char *altroot = NULL;
3328 vdev_t *rvd;
3329 dsl_pool_t *dp;
3330 dmu_tx_t *tx;
3331 int error = 0;
3332 uint64_t txg = TXG_INITIAL;
3333 nvlist_t **spares, **l2cache;
3334 uint_t nspares, nl2cache;
3335 uint64_t version, obj;
3336 boolean_t has_features;
3337
3338 /*
3339 * If this pool already exists, return failure.
3340 */
3341 mutex_enter(&spa_namespace_lock);
3342 if (spa_lookup(pool) != NULL) {
3343 mutex_exit(&spa_namespace_lock);
3344 return (EEXIST);
3345 }
3346
3347 /*
3348 * Allocate a new spa_t structure.
3349 */
3350 (void) nvlist_lookup_string(props,
3351 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3352 spa = spa_add(pool, NULL, altroot);
3353 spa_activate(spa, spa_mode_global);
3354
3355 if (props && (error = spa_prop_validate(spa, props))) {
3356 spa_deactivate(spa);
3357 spa_remove(spa);
3358 mutex_exit(&spa_namespace_lock);
3359 return (error);
3360 }
3361
3362 has_features = B_FALSE;
3363 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3364 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3365 if (zpool_prop_feature(nvpair_name(elem)))
3366 has_features = B_TRUE;
3367 }
3368
3369 if (has_features || nvlist_lookup_uint64(props,
3370 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3371 version = SPA_VERSION;
3372 }
3373 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3374
3375 spa->spa_first_txg = txg;
3376 spa->spa_uberblock.ub_txg = txg - 1;
3377 spa->spa_uberblock.ub_version = version;
3378 spa->spa_ubsync = spa->spa_uberblock;
3379
3380 /*
3381 * Create "The Godfather" zio to hold all async IOs
3382 */
3383 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3384 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3385
3386 /*
3387 * Create the root vdev.
3388 */
3389 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3390
3391 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3392
3393 ASSERT(error != 0 || rvd != NULL);
3394 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3395
3396 if (error == 0 && !zfs_allocatable_devs(nvroot))
3397 error = EINVAL;
3398
3399 if (error == 0 &&
3400 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3401 (error = spa_validate_aux(spa, nvroot, txg,
3402 VDEV_ALLOC_ADD)) == 0) {
3403 for (int c = 0; c < rvd->vdev_children; c++) {
3404 vdev_metaslab_set_size(rvd->vdev_child[c]);
3405 vdev_expand(rvd->vdev_child[c], txg);
3406 }
3407 }
3408
3409 spa_config_exit(spa, SCL_ALL, FTAG);
3410
3411 if (error != 0) {
3412 spa_unload(spa);
3413 spa_deactivate(spa);
3414 spa_remove(spa);
3415 mutex_exit(&spa_namespace_lock);
3416 return (error);
3417 }
3418
3419 /*
3420 * Get the list of spares, if specified.
3421 */
3422 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3423 &spares, &nspares) == 0) {
3424 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3425 KM_SLEEP) == 0);
3426 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3427 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3429 spa_load_spares(spa);
3430 spa_config_exit(spa, SCL_ALL, FTAG);
3431 spa->spa_spares.sav_sync = B_TRUE;
3432 }
3433
3434 /*
3435 * Get the list of level 2 cache devices, if specified.
3436 */
3437 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3438 &l2cache, &nl2cache) == 0) {
3439 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3440 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3441 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3442 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3444 spa_load_l2cache(spa);
3445 spa_config_exit(spa, SCL_ALL, FTAG);
3446 spa->spa_l2cache.sav_sync = B_TRUE;
3447 }
3448
3449 spa->spa_is_initializing = B_TRUE;
3450 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3451 spa->spa_meta_objset = dp->dp_meta_objset;
3452 spa->spa_is_initializing = B_FALSE;
3453
3454 /*
3455 * Create DDTs (dedup tables).
3456 */
3457 ddt_create(spa);
3458
3459 spa_update_dspace(spa);
3460
3461 tx = dmu_tx_create_assigned(dp, txg);
3462
3463 /*
3464 * Create the pool config object.
3465 */
3466 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3467 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3468 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3469
3470 if (zap_add(spa->spa_meta_objset,
3471 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3472 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3473 cmn_err(CE_PANIC, "failed to add pool config");
3474 }
3475
3476 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3477 spa_feature_create_zap_objects(spa, tx);
3478
3479 if (zap_add(spa->spa_meta_objset,
3480 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3481 sizeof (uint64_t), 1, &version, tx) != 0) {
3482 cmn_err(CE_PANIC, "failed to add pool version");
3483 }
3484
3485 /* Newly created pools with the right version are always deflated. */
3486 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3487 spa->spa_deflate = TRUE;
3488 if (zap_add(spa->spa_meta_objset,
3489 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3490 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3491 cmn_err(CE_PANIC, "failed to add deflate");
3492 }
3493 }
3494
3495 /*
3496 * Create the deferred-free bpobj. Turn off compression
3497 * because sync-to-convergence takes longer if the blocksize
3498 * keeps changing.
3499 */
3500 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3501 dmu_object_set_compress(spa->spa_meta_objset, obj,
3502 ZIO_COMPRESS_OFF, tx);
3503 if (zap_add(spa->spa_meta_objset,
3504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3505 sizeof (uint64_t), 1, &obj, tx) != 0) {
3506 cmn_err(CE_PANIC, "failed to add bpobj");
3507 }
3508 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3509 spa->spa_meta_objset, obj));
3510
3511 /*
3512 * Create the pool's history object.
3513 */
3514 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3515 spa_history_create_obj(spa, tx);
3516
3517 /*
3518 * Set pool properties.
3519 */
3520 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3521 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3522 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3523 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3524
3525 if (props != NULL) {
3526 spa_configfile_set(spa, props, B_FALSE);
3527 spa_sync_props(spa, props, tx);
3528 }
3529
3530 dmu_tx_commit(tx);
3531
3532 spa->spa_sync_on = B_TRUE;
3533 txg_sync_start(spa->spa_dsl_pool);
3534
3535 /*
3536 * We explicitly wait for the first transaction to complete so that our
3537 * bean counters are appropriately updated.
3538 */
3539 txg_wait_synced(spa->spa_dsl_pool, txg);
3540
3541 spa_config_sync(spa, B_FALSE, B_TRUE);
3542
3543 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3544 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3545 spa_history_log_version(spa, LOG_POOL_CREATE);
3546
3547 spa->spa_minref = refcount_count(&spa->spa_refcount);
3548
3549 mutex_exit(&spa_namespace_lock);
3550
3551 return (0);
3552}
3553
3554#ifdef _KERNEL
3555#if defined(sun)
3556/*
3557 * Get the root pool information from the root disk, then import the root pool
3558 * during the system boot up time.
3559 */
3560extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3561
3562static nvlist_t *
3563spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3564{
3565 nvlist_t *config;
3566 nvlist_t *nvtop, *nvroot;
3567 uint64_t pgid;
3568
3569 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3570 return (NULL);
3571
3572 /*
3573 * Add this top-level vdev to the child array.
3574 */
3575 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3576 &nvtop) == 0);
3577 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3578 &pgid) == 0);
3579 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3580
3581 /*
3582 * Put this pool's top-level vdevs into a root vdev.
3583 */
3584 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3585 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3586 VDEV_TYPE_ROOT) == 0);
3587 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3588 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3589 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3590 &nvtop, 1) == 0);
3591
3592 /*
3593 * Replace the existing vdev_tree with the new root vdev in
3594 * this pool's configuration (remove the old, add the new).
3595 */
3596 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3597 nvlist_free(nvroot);
3598 return (config);
3599}
3600
3601/*
3602 * Walk the vdev tree and see if we can find a device with "better"
3603 * configuration. A configuration is "better" if the label on that
3604 * device has a more recent txg.
3605 */
3606static void
3607spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3608{
3609 for (int c = 0; c < vd->vdev_children; c++)
3610 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3611
3612 if (vd->vdev_ops->vdev_op_leaf) {
3613 nvlist_t *label;
3614 uint64_t label_txg;
3615
3616 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3617 &label) != 0)
3618 return;
3619
3620 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3621 &label_txg) == 0);
3622
3623 /*
3624 * Do we have a better boot device?
3625 */
3626 if (label_txg > *txg) {
3627 *txg = label_txg;
3628 *avd = vd;
3629 }
3630 nvlist_free(label);
3631 }
3632}
3633
3634/*
3635 * Import a root pool.
3636 *
3637 * For x86. devpath_list will consist of devid and/or physpath name of
3638 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3639 * The GRUB "findroot" command will return the vdev we should boot.
3640 *
3641 * For Sparc, devpath_list consists the physpath name of the booting device
3642 * no matter the rootpool is a single device pool or a mirrored pool.
3643 * e.g.
3644 * "/pci@1f,0/ide@d/disk@0,0:a"
3645 */
3646int
3647spa_import_rootpool(char *devpath, char *devid)
3648{
3649 spa_t *spa;
3650 vdev_t *rvd, *bvd, *avd = NULL;
3651 nvlist_t *config, *nvtop;
3652 uint64_t guid, txg;
3653 char *pname;
3654 int error;
3655
3656 /*
3657 * Read the label from the boot device and generate a configuration.
3658 */
3659 config = spa_generate_rootconf(devpath, devid, &guid);
3660#if defined(_OBP) && defined(_KERNEL)
3661 if (config == NULL) {
3662 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3663 /* iscsi boot */
3664 get_iscsi_bootpath_phy(devpath);
3665 config = spa_generate_rootconf(devpath, devid, &guid);
3666 }
3667 }
3668#endif
3669 if (config == NULL) {
3670 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3671 devpath);
3672 return (EIO);
3673 }
3674
3675 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3676 &pname) == 0);
3677 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3678
3679 mutex_enter(&spa_namespace_lock);
3680 if ((spa = spa_lookup(pname)) != NULL) {
3681 /*
3682 * Remove the existing root pool from the namespace so that we
3683 * can replace it with the correct config we just read in.
3684 */
3685 spa_remove(spa);
3686 }
3687
3688 spa = spa_add(pname, config, NULL);
3689 spa->spa_is_root = B_TRUE;
3690 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3691
3692 /*
3693 * Build up a vdev tree based on the boot device's label config.
3694 */
3695 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3696 &nvtop) == 0);
3697 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3698 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3699 VDEV_ALLOC_ROOTPOOL);
3700 spa_config_exit(spa, SCL_ALL, FTAG);
3701 if (error) {
3702 mutex_exit(&spa_namespace_lock);
3703 nvlist_free(config);
3704 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3705 pname);
3706 return (error);
3707 }
3708
3709 /*
3710 * Get the boot vdev.
3711 */
3712 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3713 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3714 (u_longlong_t)guid);
3715 error = ENOENT;
3716 goto out;
3717 }
3718
3719 /*
3720 * Determine if there is a better boot device.
3721 */
3722 avd = bvd;
3723 spa_alt_rootvdev(rvd, &avd, &txg);
3724 if (avd != bvd) {
3725 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3726 "try booting from '%s'", avd->vdev_path);
3727 error = EINVAL;
3728 goto out;
3729 }
3730
3731 /*
3732 * If the boot device is part of a spare vdev then ensure that
3733 * we're booting off the active spare.
3734 */
3735 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3736 !bvd->vdev_isspare) {
3737 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3738 "try booting from '%s'",
3739 bvd->vdev_parent->
3740 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3741 error = EINVAL;
3742 goto out;
3743 }
3744
3745 error = 0;
3746 spa_history_log_version(spa, LOG_POOL_IMPORT);
3747out:
3748 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3749 vdev_free(rvd);
3750 spa_config_exit(spa, SCL_ALL, FTAG);
3751 mutex_exit(&spa_namespace_lock);
3752
3753 nvlist_free(config);
3754 return (error);
3755}
3756
3757#else
3758
3759extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3760 uint64_t *count);
3761
3762static nvlist_t *
3763spa_generate_rootconf(const char *name)
3764{
3765 nvlist_t **configs, **tops;
3766 nvlist_t *config;
3767 nvlist_t *best_cfg, *nvtop, *nvroot;
3768 uint64_t *holes;
3769 uint64_t best_txg;
3770 uint64_t nchildren;
3771 uint64_t pgid;
3772 uint64_t count;
3773 uint64_t i;
3774 uint_t nholes;
3775
3776 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3777 return (NULL);
3778
3779 ASSERT3U(count, !=, 0);
3780 best_txg = 0;
3781 for (i = 0; i < count; i++) {
3782 uint64_t txg;
3783
3784 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3785 &txg) == 0);
3786 if (txg > best_txg) {
3787 best_txg = txg;
3788 best_cfg = configs[i];
3789 }
3790 }
3791
3792 /*
3793 * Multi-vdev root pool configuration discovery is not supported yet.
3794 */
| 1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/spa_boot.h>
61#include <sys/zfs_ioctl.h>
62#include <sys/dsl_scan.h>
63#include <sys/zfeature.h>
64#include <sys/zvol.h>
65#include <sys/trim_map.h>
66
67#ifdef _KERNEL
68#include <sys/callb.h>
69#include <sys/cpupart.h>
70#include <sys/zone.h>
71#endif /* _KERNEL */
72
73#include "zfs_prop.h"
74#include "zfs_comutil.h"
75
76/* Check hostid on import? */
77static int check_hostid = 1;
78
79SYSCTL_DECL(_vfs_zfs);
80TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
81SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
82 "Check hostid on import?");
83
84typedef enum zti_modes {
85 zti_mode_fixed, /* value is # of threads (min 1) */
86 zti_mode_online_percent, /* value is % of online CPUs */
87 zti_mode_batch, /* cpu-intensive; value is ignored */
88 zti_mode_null, /* don't create a taskq */
89 zti_nmodes
90} zti_modes_t;
91
92#define ZTI_FIX(n) { zti_mode_fixed, (n) }
93#define ZTI_PCT(n) { zti_mode_online_percent, (n) }
94#define ZTI_BATCH { zti_mode_batch, 0 }
95#define ZTI_NULL { zti_mode_null, 0 }
96
97#define ZTI_ONE ZTI_FIX(1)
98
99typedef struct zio_taskq_info {
100 enum zti_modes zti_mode;
101 uint_t zti_value;
102} zio_taskq_info_t;
103
104static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
105 "issue", "issue_high", "intr", "intr_high"
106};
107
108/*
109 * Define the taskq threads for the following I/O types:
110 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
111 */
112const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
113 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
116 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
117 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
119 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
120};
121
122static dsl_syncfunc_t spa_sync_version;
123static dsl_syncfunc_t spa_sync_props;
124static dsl_checkfunc_t spa_change_guid_check;
125static dsl_syncfunc_t spa_change_guid_sync;
126static boolean_t spa_has_active_shared_spare(spa_t *spa);
127static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
129 char **ereport);
130static void spa_vdev_resilver_done(spa_t *spa);
131
132uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
133#ifdef PSRSET_BIND
134id_t zio_taskq_psrset_bind = PS_NONE;
135#endif
136#ifdef SYSDC
137boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
138#endif
139uint_t zio_taskq_basedc = 80; /* base duty cycle */
140
141boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
142extern int zfs_sync_pass_deferred_free;
143
144/*
145 * This (illegal) pool name is used when temporarily importing a spa_t in order
146 * to get the vdev stats associated with the imported devices.
147 */
148#define TRYIMPORT_NAME "$import"
149
150/*
151 * ==========================================================================
152 * SPA properties routines
153 * ==========================================================================
154 */
155
156/*
157 * Add a (source=src, propname=propval) list to an nvlist.
158 */
159static void
160spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
161 uint64_t intval, zprop_source_t src)
162{
163 const char *propname = zpool_prop_to_name(prop);
164 nvlist_t *propval;
165
166 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
167 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
168
169 if (strval != NULL)
170 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
171 else
172 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
173
174 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
175 nvlist_free(propval);
176}
177
178/*
179 * Get property values from the spa configuration.
180 */
181static void
182spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
183{
184 vdev_t *rvd = spa->spa_root_vdev;
185 dsl_pool_t *pool = spa->spa_dsl_pool;
186 uint64_t size;
187 uint64_t alloc;
188 uint64_t space;
189 uint64_t cap, version;
190 zprop_source_t src = ZPROP_SRC_NONE;
191 spa_config_dirent_t *dp;
192
193 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
194
195 if (rvd != NULL) {
196 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
197 size = metaslab_class_get_space(spa_normal_class(spa));
198 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
199 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
200 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
201 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
202 size - alloc, src);
203
204 space = 0;
205 for (int c = 0; c < rvd->vdev_children; c++) {
206 vdev_t *tvd = rvd->vdev_child[c];
207 space += tvd->vdev_max_asize - tvd->vdev_asize;
208 }
209 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
210 src);
211
212 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
213 (spa_mode(spa) == FREAD), src);
214
215 cap = (size == 0) ? 0 : (alloc * 100 / size);
216 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
217
218 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
219 ddt_get_pool_dedup_ratio(spa), src);
220
221 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
222 rvd->vdev_state, src);
223
224 version = spa_version(spa);
225 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
226 src = ZPROP_SRC_DEFAULT;
227 else
228 src = ZPROP_SRC_LOCAL;
229 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
230 }
231
232 if (pool != NULL) {
233 dsl_dir_t *freedir = pool->dp_free_dir;
234
235 /*
236 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
237 * when opening pools before this version freedir will be NULL.
238 */
239 if (freedir != NULL) {
240 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
241 freedir->dd_phys->dd_used_bytes, src);
242 } else {
243 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
244 NULL, 0, src);
245 }
246 }
247
248 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
249
250 if (spa->spa_comment != NULL) {
251 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
252 0, ZPROP_SRC_LOCAL);
253 }
254
255 if (spa->spa_root != NULL)
256 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
257 0, ZPROP_SRC_LOCAL);
258
259 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
260 if (dp->scd_path == NULL) {
261 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
262 "none", 0, ZPROP_SRC_LOCAL);
263 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
264 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
265 dp->scd_path, 0, ZPROP_SRC_LOCAL);
266 }
267 }
268}
269
270/*
271 * Get zpool property values.
272 */
273int
274spa_prop_get(spa_t *spa, nvlist_t **nvp)
275{
276 objset_t *mos = spa->spa_meta_objset;
277 zap_cursor_t zc;
278 zap_attribute_t za;
279 int err;
280
281 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282
283 mutex_enter(&spa->spa_props_lock);
284
285 /*
286 * Get properties from the spa config.
287 */
288 spa_prop_get_config(spa, nvp);
289
290 /* If no pool property object, no more prop to get. */
291 if (mos == NULL || spa->spa_pool_props_object == 0) {
292 mutex_exit(&spa->spa_props_lock);
293 return (0);
294 }
295
296 /*
297 * Get properties from the MOS pool property object.
298 */
299 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
300 (err = zap_cursor_retrieve(&zc, &za)) == 0;
301 zap_cursor_advance(&zc)) {
302 uint64_t intval = 0;
303 char *strval = NULL;
304 zprop_source_t src = ZPROP_SRC_DEFAULT;
305 zpool_prop_t prop;
306
307 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
308 continue;
309
310 switch (za.za_integer_length) {
311 case 8:
312 /* integer property */
313 if (za.za_first_integer !=
314 zpool_prop_default_numeric(prop))
315 src = ZPROP_SRC_LOCAL;
316
317 if (prop == ZPOOL_PROP_BOOTFS) {
318 dsl_pool_t *dp;
319 dsl_dataset_t *ds = NULL;
320
321 dp = spa_get_dsl(spa);
322 rw_enter(&dp->dp_config_rwlock, RW_READER);
323 if (err = dsl_dataset_hold_obj(dp,
324 za.za_first_integer, FTAG, &ds)) {
325 rw_exit(&dp->dp_config_rwlock);
326 break;
327 }
328
329 strval = kmem_alloc(
330 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
331 KM_SLEEP);
332 dsl_dataset_name(ds, strval);
333 dsl_dataset_rele(ds, FTAG);
334 rw_exit(&dp->dp_config_rwlock);
335 } else {
336 strval = NULL;
337 intval = za.za_first_integer;
338 }
339
340 spa_prop_add_list(*nvp, prop, strval, intval, src);
341
342 if (strval != NULL)
343 kmem_free(strval,
344 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
345
346 break;
347
348 case 1:
349 /* string property */
350 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
351 err = zap_lookup(mos, spa->spa_pool_props_object,
352 za.za_name, 1, za.za_num_integers, strval);
353 if (err) {
354 kmem_free(strval, za.za_num_integers);
355 break;
356 }
357 spa_prop_add_list(*nvp, prop, strval, 0, src);
358 kmem_free(strval, za.za_num_integers);
359 break;
360
361 default:
362 break;
363 }
364 }
365 zap_cursor_fini(&zc);
366 mutex_exit(&spa->spa_props_lock);
367out:
368 if (err && err != ENOENT) {
369 nvlist_free(*nvp);
370 *nvp = NULL;
371 return (err);
372 }
373
374 return (0);
375}
376
377/*
378 * Validate the given pool properties nvlist and modify the list
379 * for the property values to be set.
380 */
381static int
382spa_prop_validate(spa_t *spa, nvlist_t *props)
383{
384 nvpair_t *elem;
385 int error = 0, reset_bootfs = 0;
386 uint64_t objnum;
387 boolean_t has_feature = B_FALSE;
388
389 elem = NULL;
390 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
391 uint64_t intval;
392 char *strval, *slash, *check, *fname;
393 const char *propname = nvpair_name(elem);
394 zpool_prop_t prop = zpool_name_to_prop(propname);
395
396 switch (prop) {
397 case ZPROP_INVAL:
398 if (!zpool_prop_feature(propname)) {
399 error = EINVAL;
400 break;
401 }
402
403 /*
404 * Sanitize the input.
405 */
406 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
407 error = EINVAL;
408 break;
409 }
410
411 if (nvpair_value_uint64(elem, &intval) != 0) {
412 error = EINVAL;
413 break;
414 }
415
416 if (intval != 0) {
417 error = EINVAL;
418 break;
419 }
420
421 fname = strchr(propname, '@') + 1;
422 if (zfeature_lookup_name(fname, NULL) != 0) {
423 error = EINVAL;
424 break;
425 }
426
427 has_feature = B_TRUE;
428 break;
429
430 case ZPOOL_PROP_VERSION:
431 error = nvpair_value_uint64(elem, &intval);
432 if (!error &&
433 (intval < spa_version(spa) ||
434 intval > SPA_VERSION_BEFORE_FEATURES ||
435 has_feature))
436 error = EINVAL;
437 break;
438
439 case ZPOOL_PROP_DELEGATION:
440 case ZPOOL_PROP_AUTOREPLACE:
441 case ZPOOL_PROP_LISTSNAPS:
442 case ZPOOL_PROP_AUTOEXPAND:
443 error = nvpair_value_uint64(elem, &intval);
444 if (!error && intval > 1)
445 error = EINVAL;
446 break;
447
448 case ZPOOL_PROP_BOOTFS:
449 /*
450 * If the pool version is less than SPA_VERSION_BOOTFS,
451 * or the pool is still being created (version == 0),
452 * the bootfs property cannot be set.
453 */
454 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
455 error = ENOTSUP;
456 break;
457 }
458
459 /*
460 * Make sure the vdev config is bootable
461 */
462 if (!vdev_is_bootable(spa->spa_root_vdev)) {
463 error = ENOTSUP;
464 break;
465 }
466
467 reset_bootfs = 1;
468
469 error = nvpair_value_string(elem, &strval);
470
471 if (!error) {
472 objset_t *os;
473 uint64_t compress;
474
475 if (strval == NULL || strval[0] == '\0') {
476 objnum = zpool_prop_default_numeric(
477 ZPOOL_PROP_BOOTFS);
478 break;
479 }
480
481 if (error = dmu_objset_hold(strval, FTAG, &os))
482 break;
483
484 /* Must be ZPL and not gzip compressed. */
485
486 if (dmu_objset_type(os) != DMU_OST_ZFS) {
487 error = ENOTSUP;
488 } else if ((error = dsl_prop_get_integer(strval,
489 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
490 &compress, NULL)) == 0 &&
491 !BOOTFS_COMPRESS_VALID(compress)) {
492 error = ENOTSUP;
493 } else {
494 objnum = dmu_objset_id(os);
495 }
496 dmu_objset_rele(os, FTAG);
497 }
498 break;
499
500 case ZPOOL_PROP_FAILUREMODE:
501 error = nvpair_value_uint64(elem, &intval);
502 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
503 intval > ZIO_FAILURE_MODE_PANIC))
504 error = EINVAL;
505
506 /*
507 * This is a special case which only occurs when
508 * the pool has completely failed. This allows
509 * the user to change the in-core failmode property
510 * without syncing it out to disk (I/Os might
511 * currently be blocked). We do this by returning
512 * EIO to the caller (spa_prop_set) to trick it
513 * into thinking we encountered a property validation
514 * error.
515 */
516 if (!error && spa_suspended(spa)) {
517 spa->spa_failmode = intval;
518 error = EIO;
519 }
520 break;
521
522 case ZPOOL_PROP_CACHEFILE:
523 if ((error = nvpair_value_string(elem, &strval)) != 0)
524 break;
525
526 if (strval[0] == '\0')
527 break;
528
529 if (strcmp(strval, "none") == 0)
530 break;
531
532 if (strval[0] != '/') {
533 error = EINVAL;
534 break;
535 }
536
537 slash = strrchr(strval, '/');
538 ASSERT(slash != NULL);
539
540 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
541 strcmp(slash, "/..") == 0)
542 error = EINVAL;
543 break;
544
545 case ZPOOL_PROP_COMMENT:
546 if ((error = nvpair_value_string(elem, &strval)) != 0)
547 break;
548 for (check = strval; *check != '\0'; check++) {
549 /*
550 * The kernel doesn't have an easy isprint()
551 * check. For this kernel check, we merely
552 * check ASCII apart from DEL. Fix this if
553 * there is an easy-to-use kernel isprint().
554 */
555 if (*check >= 0x7f) {
556 error = EINVAL;
557 break;
558 }
559 check++;
560 }
561 if (strlen(strval) > ZPROP_MAX_COMMENT)
562 error = E2BIG;
563 break;
564
565 case ZPOOL_PROP_DEDUPDITTO:
566 if (spa_version(spa) < SPA_VERSION_DEDUP)
567 error = ENOTSUP;
568 else
569 error = nvpair_value_uint64(elem, &intval);
570 if (error == 0 &&
571 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
572 error = EINVAL;
573 break;
574 }
575
576 if (error)
577 break;
578 }
579
580 if (!error && reset_bootfs) {
581 error = nvlist_remove(props,
582 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
583
584 if (!error) {
585 error = nvlist_add_uint64(props,
586 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
587 }
588 }
589
590 return (error);
591}
592
593void
594spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
595{
596 char *cachefile;
597 spa_config_dirent_t *dp;
598
599 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
600 &cachefile) != 0)
601 return;
602
603 dp = kmem_alloc(sizeof (spa_config_dirent_t),
604 KM_SLEEP);
605
606 if (cachefile[0] == '\0')
607 dp->scd_path = spa_strdup(spa_config_path);
608 else if (strcmp(cachefile, "none") == 0)
609 dp->scd_path = NULL;
610 else
611 dp->scd_path = spa_strdup(cachefile);
612
613 list_insert_head(&spa->spa_config_list, dp);
614 if (need_sync)
615 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
616}
617
618int
619spa_prop_set(spa_t *spa, nvlist_t *nvp)
620{
621 int error;
622 nvpair_t *elem = NULL;
623 boolean_t need_sync = B_FALSE;
624
625 if ((error = spa_prop_validate(spa, nvp)) != 0)
626 return (error);
627
628 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
629 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
630
631 if (prop == ZPOOL_PROP_CACHEFILE ||
632 prop == ZPOOL_PROP_ALTROOT ||
633 prop == ZPOOL_PROP_READONLY)
634 continue;
635
636 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
637 uint64_t ver;
638
639 if (prop == ZPOOL_PROP_VERSION) {
640 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
641 } else {
642 ASSERT(zpool_prop_feature(nvpair_name(elem)));
643 ver = SPA_VERSION_FEATURES;
644 need_sync = B_TRUE;
645 }
646
647 /* Save time if the version is already set. */
648 if (ver == spa_version(spa))
649 continue;
650
651 /*
652 * In addition to the pool directory object, we might
653 * create the pool properties object, the features for
654 * read object, the features for write object, or the
655 * feature descriptions object.
656 */
657 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
658 spa_sync_version, spa, &ver, 6);
659 if (error)
660 return (error);
661 continue;
662 }
663
664 need_sync = B_TRUE;
665 break;
666 }
667
668 if (need_sync) {
669 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
670 spa, nvp, 6));
671 }
672
673 return (0);
674}
675
676/*
677 * If the bootfs property value is dsobj, clear it.
678 */
679void
680spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
681{
682 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
683 VERIFY(zap_remove(spa->spa_meta_objset,
684 spa->spa_pool_props_object,
685 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
686 spa->spa_bootfs = 0;
687 }
688}
689
690/*ARGSUSED*/
691static int
692spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
693{
694 spa_t *spa = arg1;
695 uint64_t *newguid = arg2;
696 vdev_t *rvd = spa->spa_root_vdev;
697 uint64_t vdev_state;
698
699 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
700 vdev_state = rvd->vdev_state;
701 spa_config_exit(spa, SCL_STATE, FTAG);
702
703 if (vdev_state != VDEV_STATE_HEALTHY)
704 return (ENXIO);
705
706 ASSERT3U(spa_guid(spa), !=, *newguid);
707
708 return (0);
709}
710
711static void
712spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
713{
714 spa_t *spa = arg1;
715 uint64_t *newguid = arg2;
716 uint64_t oldguid;
717 vdev_t *rvd = spa->spa_root_vdev;
718
719 oldguid = spa_guid(spa);
720
721 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
722 rvd->vdev_guid = *newguid;
723 rvd->vdev_guid_sum += (*newguid - oldguid);
724 vdev_config_dirty(rvd);
725 spa_config_exit(spa, SCL_STATE, FTAG);
726
727#ifdef __FreeBSD__
728 /*
729 * TODO: until recent illumos logging changes are merged
730 * log reguid as pool property change
731 */
732 spa_history_log_internal(LOG_POOL_PROPSET, spa, tx,
733 "guid change old=%llu new=%llu", oldguid, *newguid);
734#else
735 spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld",
736 oldguid, *newguid);
737#endif
738}
739
740/*
741 * Change the GUID for the pool. This is done so that we can later
742 * re-import a pool built from a clone of our own vdevs. We will modify
743 * the root vdev's guid, our own pool guid, and then mark all of our
744 * vdevs dirty. Note that we must make sure that all our vdevs are
745 * online when we do this, or else any vdevs that weren't present
746 * would be orphaned from our pool. We are also going to issue a
747 * sysevent to update any watchers.
748 */
749int
750spa_change_guid(spa_t *spa)
751{
752 int error;
753 uint64_t guid;
754
755 mutex_enter(&spa_namespace_lock);
756 guid = spa_generate_guid(NULL);
757
758 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
759 spa_change_guid_sync, spa, &guid, 5);
760
761 if (error == 0) {
762 spa_config_sync(spa, B_FALSE, B_TRUE);
763 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
764 }
765
766 mutex_exit(&spa_namespace_lock);
767
768 return (error);
769}
770
771/*
772 * ==========================================================================
773 * SPA state manipulation (open/create/destroy/import/export)
774 * ==========================================================================
775 */
776
777static int
778spa_error_entry_compare(const void *a, const void *b)
779{
780 spa_error_entry_t *sa = (spa_error_entry_t *)a;
781 spa_error_entry_t *sb = (spa_error_entry_t *)b;
782 int ret;
783
784 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
785 sizeof (zbookmark_t));
786
787 if (ret < 0)
788 return (-1);
789 else if (ret > 0)
790 return (1);
791 else
792 return (0);
793}
794
795/*
796 * Utility function which retrieves copies of the current logs and
797 * re-initializes them in the process.
798 */
799void
800spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
801{
802 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
803
804 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
805 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
806
807 avl_create(&spa->spa_errlist_scrub,
808 spa_error_entry_compare, sizeof (spa_error_entry_t),
809 offsetof(spa_error_entry_t, se_avl));
810 avl_create(&spa->spa_errlist_last,
811 spa_error_entry_compare, sizeof (spa_error_entry_t),
812 offsetof(spa_error_entry_t, se_avl));
813}
814
815static taskq_t *
816spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
817 uint_t value)
818{
819 uint_t flags = TASKQ_PREPOPULATE;
820 boolean_t batch = B_FALSE;
821
822 switch (mode) {
823 case zti_mode_null:
824 return (NULL); /* no taskq needed */
825
826 case zti_mode_fixed:
827 ASSERT3U(value, >=, 1);
828 value = MAX(value, 1);
829 break;
830
831 case zti_mode_batch:
832 batch = B_TRUE;
833 flags |= TASKQ_THREADS_CPU_PCT;
834 value = zio_taskq_batch_pct;
835 break;
836
837 case zti_mode_online_percent:
838 flags |= TASKQ_THREADS_CPU_PCT;
839 break;
840
841 default:
842 panic("unrecognized mode for %s taskq (%u:%u) in "
843 "spa_activate()",
844 name, mode, value);
845 break;
846 }
847
848#ifdef SYSDC
849 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
850 if (batch)
851 flags |= TASKQ_DC_BATCH;
852
853 return (taskq_create_sysdc(name, value, 50, INT_MAX,
854 spa->spa_proc, zio_taskq_basedc, flags));
855 }
856#endif
857 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
858 spa->spa_proc, flags));
859}
860
861static void
862spa_create_zio_taskqs(spa_t *spa)
863{
864 for (int t = 0; t < ZIO_TYPES; t++) {
865 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
866 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
867 enum zti_modes mode = ztip->zti_mode;
868 uint_t value = ztip->zti_value;
869 char name[32];
870
871 (void) snprintf(name, sizeof (name),
872 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
873
874 spa->spa_zio_taskq[t][q] =
875 spa_taskq_create(spa, name, mode, value);
876 }
877 }
878}
879
880#ifdef _KERNEL
881#ifdef SPA_PROCESS
882static void
883spa_thread(void *arg)
884{
885 callb_cpr_t cprinfo;
886
887 spa_t *spa = arg;
888 user_t *pu = PTOU(curproc);
889
890 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
891 spa->spa_name);
892
893 ASSERT(curproc != &p0);
894 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
895 "zpool-%s", spa->spa_name);
896 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
897
898#ifdef PSRSET_BIND
899 /* bind this thread to the requested psrset */
900 if (zio_taskq_psrset_bind != PS_NONE) {
901 pool_lock();
902 mutex_enter(&cpu_lock);
903 mutex_enter(&pidlock);
904 mutex_enter(&curproc->p_lock);
905
906 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
907 0, NULL, NULL) == 0) {
908 curthread->t_bind_pset = zio_taskq_psrset_bind;
909 } else {
910 cmn_err(CE_WARN,
911 "Couldn't bind process for zfs pool \"%s\" to "
912 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
913 }
914
915 mutex_exit(&curproc->p_lock);
916 mutex_exit(&pidlock);
917 mutex_exit(&cpu_lock);
918 pool_unlock();
919 }
920#endif
921
922#ifdef SYSDC
923 if (zio_taskq_sysdc) {
924 sysdc_thread_enter(curthread, 100, 0);
925 }
926#endif
927
928 spa->spa_proc = curproc;
929 spa->spa_did = curthread->t_did;
930
931 spa_create_zio_taskqs(spa);
932
933 mutex_enter(&spa->spa_proc_lock);
934 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
935
936 spa->spa_proc_state = SPA_PROC_ACTIVE;
937 cv_broadcast(&spa->spa_proc_cv);
938
939 CALLB_CPR_SAFE_BEGIN(&cprinfo);
940 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
941 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
942 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
943
944 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
945 spa->spa_proc_state = SPA_PROC_GONE;
946 spa->spa_proc = &p0;
947 cv_broadcast(&spa->spa_proc_cv);
948 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
949
950 mutex_enter(&curproc->p_lock);
951 lwp_exit();
952}
953#endif /* SPA_PROCESS */
954#endif
955
956/*
957 * Activate an uninitialized pool.
958 */
959static void
960spa_activate(spa_t *spa, int mode)
961{
962 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
963
964 spa->spa_state = POOL_STATE_ACTIVE;
965 spa->spa_mode = mode;
966
967 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
968 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
969
970 /* Try to create a covering process */
971 mutex_enter(&spa->spa_proc_lock);
972 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
973 ASSERT(spa->spa_proc == &p0);
974 spa->spa_did = 0;
975
976#ifdef SPA_PROCESS
977 /* Only create a process if we're going to be around a while. */
978 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
979 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
980 NULL, 0) == 0) {
981 spa->spa_proc_state = SPA_PROC_CREATED;
982 while (spa->spa_proc_state == SPA_PROC_CREATED) {
983 cv_wait(&spa->spa_proc_cv,
984 &spa->spa_proc_lock);
985 }
986 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
987 ASSERT(spa->spa_proc != &p0);
988 ASSERT(spa->spa_did != 0);
989 } else {
990#ifdef _KERNEL
991 cmn_err(CE_WARN,
992 "Couldn't create process for zfs pool \"%s\"\n",
993 spa->spa_name);
994#endif
995 }
996 }
997#endif /* SPA_PROCESS */
998 mutex_exit(&spa->spa_proc_lock);
999
1000 /* If we didn't create a process, we need to create our taskqs. */
1001 ASSERT(spa->spa_proc == &p0);
1002 if (spa->spa_proc == &p0) {
1003 spa_create_zio_taskqs(spa);
1004 }
1005
1006 /*
1007 * Start TRIM thread.
1008 */
1009 trim_thread_create(spa);
1010
1011 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1012 offsetof(vdev_t, vdev_config_dirty_node));
1013 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1014 offsetof(vdev_t, vdev_state_dirty_node));
1015
1016 txg_list_create(&spa->spa_vdev_txg_list,
1017 offsetof(struct vdev, vdev_txg_node));
1018
1019 avl_create(&spa->spa_errlist_scrub,
1020 spa_error_entry_compare, sizeof (spa_error_entry_t),
1021 offsetof(spa_error_entry_t, se_avl));
1022 avl_create(&spa->spa_errlist_last,
1023 spa_error_entry_compare, sizeof (spa_error_entry_t),
1024 offsetof(spa_error_entry_t, se_avl));
1025}
1026
1027/*
1028 * Opposite of spa_activate().
1029 */
1030static void
1031spa_deactivate(spa_t *spa)
1032{
1033 ASSERT(spa->spa_sync_on == B_FALSE);
1034 ASSERT(spa->spa_dsl_pool == NULL);
1035 ASSERT(spa->spa_root_vdev == NULL);
1036 ASSERT(spa->spa_async_zio_root == NULL);
1037 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1038
1039 /*
1040 * Stop TRIM thread in case spa_unload() wasn't called directly
1041 * before spa_deactivate().
1042 */
1043 trim_thread_destroy(spa);
1044
1045 txg_list_destroy(&spa->spa_vdev_txg_list);
1046
1047 list_destroy(&spa->spa_config_dirty_list);
1048 list_destroy(&spa->spa_state_dirty_list);
1049
1050 for (int t = 0; t < ZIO_TYPES; t++) {
1051 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1052 if (spa->spa_zio_taskq[t][q] != NULL)
1053 taskq_destroy(spa->spa_zio_taskq[t][q]);
1054 spa->spa_zio_taskq[t][q] = NULL;
1055 }
1056 }
1057
1058 metaslab_class_destroy(spa->spa_normal_class);
1059 spa->spa_normal_class = NULL;
1060
1061 metaslab_class_destroy(spa->spa_log_class);
1062 spa->spa_log_class = NULL;
1063
1064 /*
1065 * If this was part of an import or the open otherwise failed, we may
1066 * still have errors left in the queues. Empty them just in case.
1067 */
1068 spa_errlog_drain(spa);
1069
1070 avl_destroy(&spa->spa_errlist_scrub);
1071 avl_destroy(&spa->spa_errlist_last);
1072
1073 spa->spa_state = POOL_STATE_UNINITIALIZED;
1074
1075 mutex_enter(&spa->spa_proc_lock);
1076 if (spa->spa_proc_state != SPA_PROC_NONE) {
1077 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1078 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1079 cv_broadcast(&spa->spa_proc_cv);
1080 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1081 ASSERT(spa->spa_proc != &p0);
1082 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1083 }
1084 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1085 spa->spa_proc_state = SPA_PROC_NONE;
1086 }
1087 ASSERT(spa->spa_proc == &p0);
1088 mutex_exit(&spa->spa_proc_lock);
1089
1090#ifdef SPA_PROCESS
1091 /*
1092 * We want to make sure spa_thread() has actually exited the ZFS
1093 * module, so that the module can't be unloaded out from underneath
1094 * it.
1095 */
1096 if (spa->spa_did != 0) {
1097 thread_join(spa->spa_did);
1098 spa->spa_did = 0;
1099 }
1100#endif /* SPA_PROCESS */
1101}
1102
1103/*
1104 * Verify a pool configuration, and construct the vdev tree appropriately. This
1105 * will create all the necessary vdevs in the appropriate layout, with each vdev
1106 * in the CLOSED state. This will prep the pool before open/creation/import.
1107 * All vdev validation is done by the vdev_alloc() routine.
1108 */
1109static int
1110spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1111 uint_t id, int atype)
1112{
1113 nvlist_t **child;
1114 uint_t children;
1115 int error;
1116
1117 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1118 return (error);
1119
1120 if ((*vdp)->vdev_ops->vdev_op_leaf)
1121 return (0);
1122
1123 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1124 &child, &children);
1125
1126 if (error == ENOENT)
1127 return (0);
1128
1129 if (error) {
1130 vdev_free(*vdp);
1131 *vdp = NULL;
1132 return (EINVAL);
1133 }
1134
1135 for (int c = 0; c < children; c++) {
1136 vdev_t *vd;
1137 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1138 atype)) != 0) {
1139 vdev_free(*vdp);
1140 *vdp = NULL;
1141 return (error);
1142 }
1143 }
1144
1145 ASSERT(*vdp != NULL);
1146
1147 return (0);
1148}
1149
1150/*
1151 * Opposite of spa_load().
1152 */
1153static void
1154spa_unload(spa_t *spa)
1155{
1156 int i;
1157
1158 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1159
1160 /*
1161 * Stop TRIM thread.
1162 */
1163 trim_thread_destroy(spa);
1164
1165 /*
1166 * Stop async tasks.
1167 */
1168 spa_async_suspend(spa);
1169
1170 /*
1171 * Stop syncing.
1172 */
1173 if (spa->spa_sync_on) {
1174 txg_sync_stop(spa->spa_dsl_pool);
1175 spa->spa_sync_on = B_FALSE;
1176 }
1177
1178 /*
1179 * Wait for any outstanding async I/O to complete.
1180 */
1181 if (spa->spa_async_zio_root != NULL) {
1182 (void) zio_wait(spa->spa_async_zio_root);
1183 spa->spa_async_zio_root = NULL;
1184 }
1185
1186 bpobj_close(&spa->spa_deferred_bpobj);
1187
1188 /*
1189 * Close the dsl pool.
1190 */
1191 if (spa->spa_dsl_pool) {
1192 dsl_pool_close(spa->spa_dsl_pool);
1193 spa->spa_dsl_pool = NULL;
1194 spa->spa_meta_objset = NULL;
1195 }
1196
1197 ddt_unload(spa);
1198
1199 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1200
1201 /*
1202 * Drop and purge level 2 cache
1203 */
1204 spa_l2cache_drop(spa);
1205
1206 /*
1207 * Close all vdevs.
1208 */
1209 if (spa->spa_root_vdev)
1210 vdev_free(spa->spa_root_vdev);
1211 ASSERT(spa->spa_root_vdev == NULL);
1212
1213 for (i = 0; i < spa->spa_spares.sav_count; i++)
1214 vdev_free(spa->spa_spares.sav_vdevs[i]);
1215 if (spa->spa_spares.sav_vdevs) {
1216 kmem_free(spa->spa_spares.sav_vdevs,
1217 spa->spa_spares.sav_count * sizeof (void *));
1218 spa->spa_spares.sav_vdevs = NULL;
1219 }
1220 if (spa->spa_spares.sav_config) {
1221 nvlist_free(spa->spa_spares.sav_config);
1222 spa->spa_spares.sav_config = NULL;
1223 }
1224 spa->spa_spares.sav_count = 0;
1225
1226 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1227 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1228 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1229 }
1230 if (spa->spa_l2cache.sav_vdevs) {
1231 kmem_free(spa->spa_l2cache.sav_vdevs,
1232 spa->spa_l2cache.sav_count * sizeof (void *));
1233 spa->spa_l2cache.sav_vdevs = NULL;
1234 }
1235 if (spa->spa_l2cache.sav_config) {
1236 nvlist_free(spa->spa_l2cache.sav_config);
1237 spa->spa_l2cache.sav_config = NULL;
1238 }
1239 spa->spa_l2cache.sav_count = 0;
1240
1241 spa->spa_async_suspended = 0;
1242
1243 if (spa->spa_comment != NULL) {
1244 spa_strfree(spa->spa_comment);
1245 spa->spa_comment = NULL;
1246 }
1247
1248 spa_config_exit(spa, SCL_ALL, FTAG);
1249}
1250
1251/*
1252 * Load (or re-load) the current list of vdevs describing the active spares for
1253 * this pool. When this is called, we have some form of basic information in
1254 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1255 * then re-generate a more complete list including status information.
1256 */
1257static void
1258spa_load_spares(spa_t *spa)
1259{
1260 nvlist_t **spares;
1261 uint_t nspares;
1262 int i;
1263 vdev_t *vd, *tvd;
1264
1265 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1266
1267 /*
1268 * First, close and free any existing spare vdevs.
1269 */
1270 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1271 vd = spa->spa_spares.sav_vdevs[i];
1272
1273 /* Undo the call to spa_activate() below */
1274 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1275 B_FALSE)) != NULL && tvd->vdev_isspare)
1276 spa_spare_remove(tvd);
1277 vdev_close(vd);
1278 vdev_free(vd);
1279 }
1280
1281 if (spa->spa_spares.sav_vdevs)
1282 kmem_free(spa->spa_spares.sav_vdevs,
1283 spa->spa_spares.sav_count * sizeof (void *));
1284
1285 if (spa->spa_spares.sav_config == NULL)
1286 nspares = 0;
1287 else
1288 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1289 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1290
1291 spa->spa_spares.sav_count = (int)nspares;
1292 spa->spa_spares.sav_vdevs = NULL;
1293
1294 if (nspares == 0)
1295 return;
1296
1297 /*
1298 * Construct the array of vdevs, opening them to get status in the
1299 * process. For each spare, there is potentially two different vdev_t
1300 * structures associated with it: one in the list of spares (used only
1301 * for basic validation purposes) and one in the active vdev
1302 * configuration (if it's spared in). During this phase we open and
1303 * validate each vdev on the spare list. If the vdev also exists in the
1304 * active configuration, then we also mark this vdev as an active spare.
1305 */
1306 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1307 KM_SLEEP);
1308 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1309 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1310 VDEV_ALLOC_SPARE) == 0);
1311 ASSERT(vd != NULL);
1312
1313 spa->spa_spares.sav_vdevs[i] = vd;
1314
1315 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1316 B_FALSE)) != NULL) {
1317 if (!tvd->vdev_isspare)
1318 spa_spare_add(tvd);
1319
1320 /*
1321 * We only mark the spare active if we were successfully
1322 * able to load the vdev. Otherwise, importing a pool
1323 * with a bad active spare would result in strange
1324 * behavior, because multiple pool would think the spare
1325 * is actively in use.
1326 *
1327 * There is a vulnerability here to an equally bizarre
1328 * circumstance, where a dead active spare is later
1329 * brought back to life (onlined or otherwise). Given
1330 * the rarity of this scenario, and the extra complexity
1331 * it adds, we ignore the possibility.
1332 */
1333 if (!vdev_is_dead(tvd))
1334 spa_spare_activate(tvd);
1335 }
1336
1337 vd->vdev_top = vd;
1338 vd->vdev_aux = &spa->spa_spares;
1339
1340 if (vdev_open(vd) != 0)
1341 continue;
1342
1343 if (vdev_validate_aux(vd) == 0)
1344 spa_spare_add(vd);
1345 }
1346
1347 /*
1348 * Recompute the stashed list of spares, with status information
1349 * this time.
1350 */
1351 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1352 DATA_TYPE_NVLIST_ARRAY) == 0);
1353
1354 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1355 KM_SLEEP);
1356 for (i = 0; i < spa->spa_spares.sav_count; i++)
1357 spares[i] = vdev_config_generate(spa,
1358 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1359 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1360 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1361 for (i = 0; i < spa->spa_spares.sav_count; i++)
1362 nvlist_free(spares[i]);
1363 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1364}
1365
1366/*
1367 * Load (or re-load) the current list of vdevs describing the active l2cache for
1368 * this pool. When this is called, we have some form of basic information in
1369 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1370 * then re-generate a more complete list including status information.
1371 * Devices which are already active have their details maintained, and are
1372 * not re-opened.
1373 */
1374static void
1375spa_load_l2cache(spa_t *spa)
1376{
1377 nvlist_t **l2cache;
1378 uint_t nl2cache;
1379 int i, j, oldnvdevs;
1380 uint64_t guid;
1381 vdev_t *vd, **oldvdevs, **newvdevs;
1382 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1383
1384 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1385
1386 if (sav->sav_config != NULL) {
1387 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1388 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1389 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1390 } else {
1391 nl2cache = 0;
1392 }
1393
1394 oldvdevs = sav->sav_vdevs;
1395 oldnvdevs = sav->sav_count;
1396 sav->sav_vdevs = NULL;
1397 sav->sav_count = 0;
1398
1399 /*
1400 * Process new nvlist of vdevs.
1401 */
1402 for (i = 0; i < nl2cache; i++) {
1403 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1404 &guid) == 0);
1405
1406 newvdevs[i] = NULL;
1407 for (j = 0; j < oldnvdevs; j++) {
1408 vd = oldvdevs[j];
1409 if (vd != NULL && guid == vd->vdev_guid) {
1410 /*
1411 * Retain previous vdev for add/remove ops.
1412 */
1413 newvdevs[i] = vd;
1414 oldvdevs[j] = NULL;
1415 break;
1416 }
1417 }
1418
1419 if (newvdevs[i] == NULL) {
1420 /*
1421 * Create new vdev
1422 */
1423 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1424 VDEV_ALLOC_L2CACHE) == 0);
1425 ASSERT(vd != NULL);
1426 newvdevs[i] = vd;
1427
1428 /*
1429 * Commit this vdev as an l2cache device,
1430 * even if it fails to open.
1431 */
1432 spa_l2cache_add(vd);
1433
1434 vd->vdev_top = vd;
1435 vd->vdev_aux = sav;
1436
1437 spa_l2cache_activate(vd);
1438
1439 if (vdev_open(vd) != 0)
1440 continue;
1441
1442 (void) vdev_validate_aux(vd);
1443
1444 if (!vdev_is_dead(vd))
1445 l2arc_add_vdev(spa, vd);
1446 }
1447 }
1448
1449 /*
1450 * Purge vdevs that were dropped
1451 */
1452 for (i = 0; i < oldnvdevs; i++) {
1453 uint64_t pool;
1454
1455 vd = oldvdevs[i];
1456 if (vd != NULL) {
1457 ASSERT(vd->vdev_isl2cache);
1458
1459 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1460 pool != 0ULL && l2arc_vdev_present(vd))
1461 l2arc_remove_vdev(vd);
1462 vdev_clear_stats(vd);
1463 vdev_free(vd);
1464 }
1465 }
1466
1467 if (oldvdevs)
1468 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1469
1470 if (sav->sav_config == NULL)
1471 goto out;
1472
1473 sav->sav_vdevs = newvdevs;
1474 sav->sav_count = (int)nl2cache;
1475
1476 /*
1477 * Recompute the stashed list of l2cache devices, with status
1478 * information this time.
1479 */
1480 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1481 DATA_TYPE_NVLIST_ARRAY) == 0);
1482
1483 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1484 for (i = 0; i < sav->sav_count; i++)
1485 l2cache[i] = vdev_config_generate(spa,
1486 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1487 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1488 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1489out:
1490 for (i = 0; i < sav->sav_count; i++)
1491 nvlist_free(l2cache[i]);
1492 if (sav->sav_count)
1493 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1494}
1495
1496static int
1497load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1498{
1499 dmu_buf_t *db;
1500 char *packed = NULL;
1501 size_t nvsize = 0;
1502 int error;
1503 *value = NULL;
1504
1505 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1506 nvsize = *(uint64_t *)db->db_data;
1507 dmu_buf_rele(db, FTAG);
1508
1509 packed = kmem_alloc(nvsize, KM_SLEEP);
1510 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1511 DMU_READ_PREFETCH);
1512 if (error == 0)
1513 error = nvlist_unpack(packed, nvsize, value, 0);
1514 kmem_free(packed, nvsize);
1515
1516 return (error);
1517}
1518
1519/*
1520 * Checks to see if the given vdev could not be opened, in which case we post a
1521 * sysevent to notify the autoreplace code that the device has been removed.
1522 */
1523static void
1524spa_check_removed(vdev_t *vd)
1525{
1526 for (int c = 0; c < vd->vdev_children; c++)
1527 spa_check_removed(vd->vdev_child[c]);
1528
1529 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1530 zfs_post_autoreplace(vd->vdev_spa, vd);
1531 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1532 }
1533}
1534
1535/*
1536 * Validate the current config against the MOS config
1537 */
1538static boolean_t
1539spa_config_valid(spa_t *spa, nvlist_t *config)
1540{
1541 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1542 nvlist_t *nv;
1543
1544 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1545
1546 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1547 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1548
1549 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1550
1551 /*
1552 * If we're doing a normal import, then build up any additional
1553 * diagnostic information about missing devices in this config.
1554 * We'll pass this up to the user for further processing.
1555 */
1556 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1557 nvlist_t **child, *nv;
1558 uint64_t idx = 0;
1559
1560 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1561 KM_SLEEP);
1562 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1563
1564 for (int c = 0; c < rvd->vdev_children; c++) {
1565 vdev_t *tvd = rvd->vdev_child[c];
1566 vdev_t *mtvd = mrvd->vdev_child[c];
1567
1568 if (tvd->vdev_ops == &vdev_missing_ops &&
1569 mtvd->vdev_ops != &vdev_missing_ops &&
1570 mtvd->vdev_islog)
1571 child[idx++] = vdev_config_generate(spa, mtvd,
1572 B_FALSE, 0);
1573 }
1574
1575 if (idx) {
1576 VERIFY(nvlist_add_nvlist_array(nv,
1577 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1578 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1579 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1580
1581 for (int i = 0; i < idx; i++)
1582 nvlist_free(child[i]);
1583 }
1584 nvlist_free(nv);
1585 kmem_free(child, rvd->vdev_children * sizeof (char **));
1586 }
1587
1588 /*
1589 * Compare the root vdev tree with the information we have
1590 * from the MOS config (mrvd). Check each top-level vdev
1591 * with the corresponding MOS config top-level (mtvd).
1592 */
1593 for (int c = 0; c < rvd->vdev_children; c++) {
1594 vdev_t *tvd = rvd->vdev_child[c];
1595 vdev_t *mtvd = mrvd->vdev_child[c];
1596
1597 /*
1598 * Resolve any "missing" vdevs in the current configuration.
1599 * If we find that the MOS config has more accurate information
1600 * about the top-level vdev then use that vdev instead.
1601 */
1602 if (tvd->vdev_ops == &vdev_missing_ops &&
1603 mtvd->vdev_ops != &vdev_missing_ops) {
1604
1605 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1606 continue;
1607
1608 /*
1609 * Device specific actions.
1610 */
1611 if (mtvd->vdev_islog) {
1612 spa_set_log_state(spa, SPA_LOG_CLEAR);
1613 } else {
1614 /*
1615 * XXX - once we have 'readonly' pool
1616 * support we should be able to handle
1617 * missing data devices by transitioning
1618 * the pool to readonly.
1619 */
1620 continue;
1621 }
1622
1623 /*
1624 * Swap the missing vdev with the data we were
1625 * able to obtain from the MOS config.
1626 */
1627 vdev_remove_child(rvd, tvd);
1628 vdev_remove_child(mrvd, mtvd);
1629
1630 vdev_add_child(rvd, mtvd);
1631 vdev_add_child(mrvd, tvd);
1632
1633 spa_config_exit(spa, SCL_ALL, FTAG);
1634 vdev_load(mtvd);
1635 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1636
1637 vdev_reopen(rvd);
1638 } else if (mtvd->vdev_islog) {
1639 /*
1640 * Load the slog device's state from the MOS config
1641 * since it's possible that the label does not
1642 * contain the most up-to-date information.
1643 */
1644 vdev_load_log_state(tvd, mtvd);
1645 vdev_reopen(tvd);
1646 }
1647 }
1648 vdev_free(mrvd);
1649 spa_config_exit(spa, SCL_ALL, FTAG);
1650
1651 /*
1652 * Ensure we were able to validate the config.
1653 */
1654 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1655}
1656
1657/*
1658 * Check for missing log devices
1659 */
1660static int
1661spa_check_logs(spa_t *spa)
1662{
1663 switch (spa->spa_log_state) {
1664 case SPA_LOG_MISSING:
1665 /* need to recheck in case slog has been restored */
1666 case SPA_LOG_UNKNOWN:
1667 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1668 DS_FIND_CHILDREN)) {
1669 spa_set_log_state(spa, SPA_LOG_MISSING);
1670 return (1);
1671 }
1672 break;
1673 }
1674 return (0);
1675}
1676
1677static boolean_t
1678spa_passivate_log(spa_t *spa)
1679{
1680 vdev_t *rvd = spa->spa_root_vdev;
1681 boolean_t slog_found = B_FALSE;
1682
1683 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1684
1685 if (!spa_has_slogs(spa))
1686 return (B_FALSE);
1687
1688 for (int c = 0; c < rvd->vdev_children; c++) {
1689 vdev_t *tvd = rvd->vdev_child[c];
1690 metaslab_group_t *mg = tvd->vdev_mg;
1691
1692 if (tvd->vdev_islog) {
1693 metaslab_group_passivate(mg);
1694 slog_found = B_TRUE;
1695 }
1696 }
1697
1698 return (slog_found);
1699}
1700
1701static void
1702spa_activate_log(spa_t *spa)
1703{
1704 vdev_t *rvd = spa->spa_root_vdev;
1705
1706 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1707
1708 for (int c = 0; c < rvd->vdev_children; c++) {
1709 vdev_t *tvd = rvd->vdev_child[c];
1710 metaslab_group_t *mg = tvd->vdev_mg;
1711
1712 if (tvd->vdev_islog)
1713 metaslab_group_activate(mg);
1714 }
1715}
1716
1717int
1718spa_offline_log(spa_t *spa)
1719{
1720 int error = 0;
1721
1722 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1723 NULL, DS_FIND_CHILDREN)) == 0) {
1724
1725 /*
1726 * We successfully offlined the log device, sync out the
1727 * current txg so that the "stubby" block can be removed
1728 * by zil_sync().
1729 */
1730 txg_wait_synced(spa->spa_dsl_pool, 0);
1731 }
1732 return (error);
1733}
1734
1735static void
1736spa_aux_check_removed(spa_aux_vdev_t *sav)
1737{
1738 int i;
1739
1740 for (i = 0; i < sav->sav_count; i++)
1741 spa_check_removed(sav->sav_vdevs[i]);
1742}
1743
1744void
1745spa_claim_notify(zio_t *zio)
1746{
1747 spa_t *spa = zio->io_spa;
1748
1749 if (zio->io_error)
1750 return;
1751
1752 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1753 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1754 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1755 mutex_exit(&spa->spa_props_lock);
1756}
1757
1758typedef struct spa_load_error {
1759 uint64_t sle_meta_count;
1760 uint64_t sle_data_count;
1761} spa_load_error_t;
1762
1763static void
1764spa_load_verify_done(zio_t *zio)
1765{
1766 blkptr_t *bp = zio->io_bp;
1767 spa_load_error_t *sle = zio->io_private;
1768 dmu_object_type_t type = BP_GET_TYPE(bp);
1769 int error = zio->io_error;
1770
1771 if (error) {
1772 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1773 type != DMU_OT_INTENT_LOG)
1774 atomic_add_64(&sle->sle_meta_count, 1);
1775 else
1776 atomic_add_64(&sle->sle_data_count, 1);
1777 }
1778 zio_data_buf_free(zio->io_data, zio->io_size);
1779}
1780
1781/*ARGSUSED*/
1782static int
1783spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1784 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1785{
1786 if (bp != NULL) {
1787 zio_t *rio = arg;
1788 size_t size = BP_GET_PSIZE(bp);
1789 void *data = zio_data_buf_alloc(size);
1790
1791 zio_nowait(zio_read(rio, spa, bp, data, size,
1792 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1793 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1794 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1795 }
1796 return (0);
1797}
1798
1799static int
1800spa_load_verify(spa_t *spa)
1801{
1802 zio_t *rio;
1803 spa_load_error_t sle = { 0 };
1804 zpool_rewind_policy_t policy;
1805 boolean_t verify_ok = B_FALSE;
1806 int error;
1807
1808 zpool_get_rewind_policy(spa->spa_config, &policy);
1809
1810 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1811 return (0);
1812
1813 rio = zio_root(spa, NULL, &sle,
1814 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1815
1816 error = traverse_pool(spa, spa->spa_verify_min_txg,
1817 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1818
1819 (void) zio_wait(rio);
1820
1821 spa->spa_load_meta_errors = sle.sle_meta_count;
1822 spa->spa_load_data_errors = sle.sle_data_count;
1823
1824 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1825 sle.sle_data_count <= policy.zrp_maxdata) {
1826 int64_t loss = 0;
1827
1828 verify_ok = B_TRUE;
1829 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1830 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1831
1832 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1833 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1834 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1835 VERIFY(nvlist_add_int64(spa->spa_load_info,
1836 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1837 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1838 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1839 } else {
1840 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1841 }
1842
1843 if (error) {
1844 if (error != ENXIO && error != EIO)
1845 error = EIO;
1846 return (error);
1847 }
1848
1849 return (verify_ok ? 0 : EIO);
1850}
1851
1852/*
1853 * Find a value in the pool props object.
1854 */
1855static void
1856spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1857{
1858 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1859 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1860}
1861
1862/*
1863 * Find a value in the pool directory object.
1864 */
1865static int
1866spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1867{
1868 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1869 name, sizeof (uint64_t), 1, val));
1870}
1871
1872static int
1873spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1874{
1875 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1876 return (err);
1877}
1878
1879/*
1880 * Fix up config after a partly-completed split. This is done with the
1881 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1882 * pool have that entry in their config, but only the splitting one contains
1883 * a list of all the guids of the vdevs that are being split off.
1884 *
1885 * This function determines what to do with that list: either rejoin
1886 * all the disks to the pool, or complete the splitting process. To attempt
1887 * the rejoin, each disk that is offlined is marked online again, and
1888 * we do a reopen() call. If the vdev label for every disk that was
1889 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1890 * then we call vdev_split() on each disk, and complete the split.
1891 *
1892 * Otherwise we leave the config alone, with all the vdevs in place in
1893 * the original pool.
1894 */
1895static void
1896spa_try_repair(spa_t *spa, nvlist_t *config)
1897{
1898 uint_t extracted;
1899 uint64_t *glist;
1900 uint_t i, gcount;
1901 nvlist_t *nvl;
1902 vdev_t **vd;
1903 boolean_t attempt_reopen;
1904
1905 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1906 return;
1907
1908 /* check that the config is complete */
1909 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1910 &glist, &gcount) != 0)
1911 return;
1912
1913 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1914
1915 /* attempt to online all the vdevs & validate */
1916 attempt_reopen = B_TRUE;
1917 for (i = 0; i < gcount; i++) {
1918 if (glist[i] == 0) /* vdev is hole */
1919 continue;
1920
1921 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1922 if (vd[i] == NULL) {
1923 /*
1924 * Don't bother attempting to reopen the disks;
1925 * just do the split.
1926 */
1927 attempt_reopen = B_FALSE;
1928 } else {
1929 /* attempt to re-online it */
1930 vd[i]->vdev_offline = B_FALSE;
1931 }
1932 }
1933
1934 if (attempt_reopen) {
1935 vdev_reopen(spa->spa_root_vdev);
1936
1937 /* check each device to see what state it's in */
1938 for (extracted = 0, i = 0; i < gcount; i++) {
1939 if (vd[i] != NULL &&
1940 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1941 break;
1942 ++extracted;
1943 }
1944 }
1945
1946 /*
1947 * If every disk has been moved to the new pool, or if we never
1948 * even attempted to look at them, then we split them off for
1949 * good.
1950 */
1951 if (!attempt_reopen || gcount == extracted) {
1952 for (i = 0; i < gcount; i++)
1953 if (vd[i] != NULL)
1954 vdev_split(vd[i]);
1955 vdev_reopen(spa->spa_root_vdev);
1956 }
1957
1958 kmem_free(vd, gcount * sizeof (vdev_t *));
1959}
1960
1961static int
1962spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1963 boolean_t mosconfig)
1964{
1965 nvlist_t *config = spa->spa_config;
1966 char *ereport = FM_EREPORT_ZFS_POOL;
1967 char *comment;
1968 int error;
1969 uint64_t pool_guid;
1970 nvlist_t *nvl;
1971
1972 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1973 return (EINVAL);
1974
1975 ASSERT(spa->spa_comment == NULL);
1976 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1977 spa->spa_comment = spa_strdup(comment);
1978
1979 /*
1980 * Versioning wasn't explicitly added to the label until later, so if
1981 * it's not present treat it as the initial version.
1982 */
1983 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1984 &spa->spa_ubsync.ub_version) != 0)
1985 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1986
1987 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1988 &spa->spa_config_txg);
1989
1990 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1991 spa_guid_exists(pool_guid, 0)) {
1992 error = EEXIST;
1993 } else {
1994 spa->spa_config_guid = pool_guid;
1995
1996 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1997 &nvl) == 0) {
1998 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1999 KM_SLEEP) == 0);
2000 }
2001
2002 nvlist_free(spa->spa_load_info);
2003 spa->spa_load_info = fnvlist_alloc();
2004
2005 gethrestime(&spa->spa_loaded_ts);
2006 error = spa_load_impl(spa, pool_guid, config, state, type,
2007 mosconfig, &ereport);
2008 }
2009
2010 spa->spa_minref = refcount_count(&spa->spa_refcount);
2011 if (error) {
2012 if (error != EEXIST) {
2013 spa->spa_loaded_ts.tv_sec = 0;
2014 spa->spa_loaded_ts.tv_nsec = 0;
2015 }
2016 if (error != EBADF) {
2017 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2018 }
2019 }
2020 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2021 spa->spa_ena = 0;
2022
2023 return (error);
2024}
2025
2026/*
2027 * Load an existing storage pool, using the pool's builtin spa_config as a
2028 * source of configuration information.
2029 */
2030static int
2031spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2032 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2033 char **ereport)
2034{
2035 int error = 0;
2036 nvlist_t *nvroot = NULL;
2037 nvlist_t *label;
2038 vdev_t *rvd;
2039 uberblock_t *ub = &spa->spa_uberblock;
2040 uint64_t children, config_cache_txg = spa->spa_config_txg;
2041 int orig_mode = spa->spa_mode;
2042 int parse;
2043 uint64_t obj;
2044 boolean_t missing_feat_write = B_FALSE;
2045
2046 /*
2047 * If this is an untrusted config, access the pool in read-only mode.
2048 * This prevents things like resilvering recently removed devices.
2049 */
2050 if (!mosconfig)
2051 spa->spa_mode = FREAD;
2052
2053 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2054
2055 spa->spa_load_state = state;
2056
2057 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2058 return (EINVAL);
2059
2060 parse = (type == SPA_IMPORT_EXISTING ?
2061 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2062
2063 /*
2064 * Create "The Godfather" zio to hold all async IOs
2065 */
2066 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2067 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2068
2069 /*
2070 * Parse the configuration into a vdev tree. We explicitly set the
2071 * value that will be returned by spa_version() since parsing the
2072 * configuration requires knowing the version number.
2073 */
2074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2075 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2076 spa_config_exit(spa, SCL_ALL, FTAG);
2077
2078 if (error != 0)
2079 return (error);
2080
2081 ASSERT(spa->spa_root_vdev == rvd);
2082
2083 if (type != SPA_IMPORT_ASSEMBLE) {
2084 ASSERT(spa_guid(spa) == pool_guid);
2085 }
2086
2087 /*
2088 * Try to open all vdevs, loading each label in the process.
2089 */
2090 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2091 error = vdev_open(rvd);
2092 spa_config_exit(spa, SCL_ALL, FTAG);
2093 if (error != 0)
2094 return (error);
2095
2096 /*
2097 * We need to validate the vdev labels against the configuration that
2098 * we have in hand, which is dependent on the setting of mosconfig. If
2099 * mosconfig is true then we're validating the vdev labels based on
2100 * that config. Otherwise, we're validating against the cached config
2101 * (zpool.cache) that was read when we loaded the zfs module, and then
2102 * later we will recursively call spa_load() and validate against
2103 * the vdev config.
2104 *
2105 * If we're assembling a new pool that's been split off from an
2106 * existing pool, the labels haven't yet been updated so we skip
2107 * validation for now.
2108 */
2109 if (type != SPA_IMPORT_ASSEMBLE) {
2110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2111 error = vdev_validate(rvd, mosconfig);
2112 spa_config_exit(spa, SCL_ALL, FTAG);
2113
2114 if (error != 0)
2115 return (error);
2116
2117 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2118 return (ENXIO);
2119 }
2120
2121 /*
2122 * Find the best uberblock.
2123 */
2124 vdev_uberblock_load(rvd, ub, &label);
2125
2126 /*
2127 * If we weren't able to find a single valid uberblock, return failure.
2128 */
2129 if (ub->ub_txg == 0) {
2130 nvlist_free(label);
2131 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2132 }
2133
2134 /*
2135 * If the pool has an unsupported version we can't open it.
2136 */
2137 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2138 nvlist_free(label);
2139 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2140 }
2141
2142 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2143 nvlist_t *features;
2144
2145 /*
2146 * If we weren't able to find what's necessary for reading the
2147 * MOS in the label, return failure.
2148 */
2149 if (label == NULL || nvlist_lookup_nvlist(label,
2150 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2151 nvlist_free(label);
2152 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2153 ENXIO));
2154 }
2155
2156 /*
2157 * Update our in-core representation with the definitive values
2158 * from the label.
2159 */
2160 nvlist_free(spa->spa_label_features);
2161 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2162 }
2163
2164 nvlist_free(label);
2165
2166 /*
2167 * Look through entries in the label nvlist's features_for_read. If
2168 * there is a feature listed there which we don't understand then we
2169 * cannot open a pool.
2170 */
2171 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2172 nvlist_t *unsup_feat;
2173
2174 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2175 0);
2176
2177 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2178 NULL); nvp != NULL;
2179 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2180 if (!zfeature_is_supported(nvpair_name(nvp))) {
2181 VERIFY(nvlist_add_string(unsup_feat,
2182 nvpair_name(nvp), "") == 0);
2183 }
2184 }
2185
2186 if (!nvlist_empty(unsup_feat)) {
2187 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2188 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2189 nvlist_free(unsup_feat);
2190 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2191 ENOTSUP));
2192 }
2193
2194 nvlist_free(unsup_feat);
2195 }
2196
2197 /*
2198 * If the vdev guid sum doesn't match the uberblock, we have an
2199 * incomplete configuration. We first check to see if the pool
2200 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2201 * If it is, defer the vdev_guid_sum check till later so we
2202 * can handle missing vdevs.
2203 */
2204 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2205 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2206 rvd->vdev_guid_sum != ub->ub_guid_sum)
2207 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2208
2209 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2210 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2211 spa_try_repair(spa, config);
2212 spa_config_exit(spa, SCL_ALL, FTAG);
2213 nvlist_free(spa->spa_config_splitting);
2214 spa->spa_config_splitting = NULL;
2215 }
2216
2217 /*
2218 * Initialize internal SPA structures.
2219 */
2220 spa->spa_state = POOL_STATE_ACTIVE;
2221 spa->spa_ubsync = spa->spa_uberblock;
2222 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2223 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2224 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2225 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2226 spa->spa_claim_max_txg = spa->spa_first_txg;
2227 spa->spa_prev_software_version = ub->ub_software_version;
2228
2229 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2230 if (error)
2231 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2232 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2233
2234 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2235 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2236
2237 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2238 boolean_t missing_feat_read = B_FALSE;
2239 nvlist_t *unsup_feat, *enabled_feat;
2240
2241 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2242 &spa->spa_feat_for_read_obj) != 0) {
2243 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2244 }
2245
2246 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2247 &spa->spa_feat_for_write_obj) != 0) {
2248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2249 }
2250
2251 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2252 &spa->spa_feat_desc_obj) != 0) {
2253 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2254 }
2255
2256 enabled_feat = fnvlist_alloc();
2257 unsup_feat = fnvlist_alloc();
2258
2259 if (!feature_is_supported(spa->spa_meta_objset,
2260 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2261 unsup_feat, enabled_feat))
2262 missing_feat_read = B_TRUE;
2263
2264 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2265 if (!feature_is_supported(spa->spa_meta_objset,
2266 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2267 unsup_feat, enabled_feat)) {
2268 missing_feat_write = B_TRUE;
2269 }
2270 }
2271
2272 fnvlist_add_nvlist(spa->spa_load_info,
2273 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2274
2275 if (!nvlist_empty(unsup_feat)) {
2276 fnvlist_add_nvlist(spa->spa_load_info,
2277 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2278 }
2279
2280 fnvlist_free(enabled_feat);
2281 fnvlist_free(unsup_feat);
2282
2283 if (!missing_feat_read) {
2284 fnvlist_add_boolean(spa->spa_load_info,
2285 ZPOOL_CONFIG_CAN_RDONLY);
2286 }
2287
2288 /*
2289 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2290 * twofold: to determine whether the pool is available for
2291 * import in read-write mode and (if it is not) whether the
2292 * pool is available for import in read-only mode. If the pool
2293 * is available for import in read-write mode, it is displayed
2294 * as available in userland; if it is not available for import
2295 * in read-only mode, it is displayed as unavailable in
2296 * userland. If the pool is available for import in read-only
2297 * mode but not read-write mode, it is displayed as unavailable
2298 * in userland with a special note that the pool is actually
2299 * available for open in read-only mode.
2300 *
2301 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2302 * missing a feature for write, we must first determine whether
2303 * the pool can be opened read-only before returning to
2304 * userland in order to know whether to display the
2305 * abovementioned note.
2306 */
2307 if (missing_feat_read || (missing_feat_write &&
2308 spa_writeable(spa))) {
2309 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2310 ENOTSUP));
2311 }
2312 }
2313
2314 spa->spa_is_initializing = B_TRUE;
2315 error = dsl_pool_open(spa->spa_dsl_pool);
2316 spa->spa_is_initializing = B_FALSE;
2317 if (error != 0)
2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2319
2320 if (!mosconfig) {
2321 uint64_t hostid;
2322 nvlist_t *policy = NULL, *nvconfig;
2323
2324 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2326
2327 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2328 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2329 char *hostname;
2330 unsigned long myhostid = 0;
2331
2332 VERIFY(nvlist_lookup_string(nvconfig,
2333 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2334
2335#ifdef _KERNEL
2336 myhostid = zone_get_hostid(NULL);
2337#else /* _KERNEL */
2338 /*
2339 * We're emulating the system's hostid in userland, so
2340 * we can't use zone_get_hostid().
2341 */
2342 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2343#endif /* _KERNEL */
2344 if (check_hostid && hostid != 0 && myhostid != 0 &&
2345 hostid != myhostid) {
2346 nvlist_free(nvconfig);
2347 cmn_err(CE_WARN, "pool '%s' could not be "
2348 "loaded as it was last accessed by "
2349 "another system (host: %s hostid: 0x%lx). "
2350 "See: http://illumos.org/msg/ZFS-8000-EY",
2351 spa_name(spa), hostname,
2352 (unsigned long)hostid);
2353 return (EBADF);
2354 }
2355 }
2356 if (nvlist_lookup_nvlist(spa->spa_config,
2357 ZPOOL_REWIND_POLICY, &policy) == 0)
2358 VERIFY(nvlist_add_nvlist(nvconfig,
2359 ZPOOL_REWIND_POLICY, policy) == 0);
2360
2361 spa_config_set(spa, nvconfig);
2362 spa_unload(spa);
2363 spa_deactivate(spa);
2364 spa_activate(spa, orig_mode);
2365
2366 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2367 }
2368
2369 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2371 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2372 if (error != 0)
2373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2374
2375 /*
2376 * Load the bit that tells us to use the new accounting function
2377 * (raid-z deflation). If we have an older pool, this will not
2378 * be present.
2379 */
2380 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2381 if (error != 0 && error != ENOENT)
2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2383
2384 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2385 &spa->spa_creation_version);
2386 if (error != 0 && error != ENOENT)
2387 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2388
2389 /*
2390 * Load the persistent error log. If we have an older pool, this will
2391 * not be present.
2392 */
2393 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2394 if (error != 0 && error != ENOENT)
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2396
2397 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2398 &spa->spa_errlog_scrub);
2399 if (error != 0 && error != ENOENT)
2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2401
2402 /*
2403 * Load the history object. If we have an older pool, this
2404 * will not be present.
2405 */
2406 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2407 if (error != 0 && error != ENOENT)
2408 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2409
2410 /*
2411 * If we're assembling the pool from the split-off vdevs of
2412 * an existing pool, we don't want to attach the spares & cache
2413 * devices.
2414 */
2415
2416 /*
2417 * Load any hot spares for this pool.
2418 */
2419 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2420 if (error != 0 && error != ENOENT)
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2423 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2424 if (load_nvlist(spa, spa->spa_spares.sav_object,
2425 &spa->spa_spares.sav_config) != 0)
2426 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427
2428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2429 spa_load_spares(spa);
2430 spa_config_exit(spa, SCL_ALL, FTAG);
2431 } else if (error == 0) {
2432 spa->spa_spares.sav_sync = B_TRUE;
2433 }
2434
2435 /*
2436 * Load any level 2 ARC devices for this pool.
2437 */
2438 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2439 &spa->spa_l2cache.sav_object);
2440 if (error != 0 && error != ENOENT)
2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2442 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2443 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2444 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2445 &spa->spa_l2cache.sav_config) != 0)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447
2448 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2449 spa_load_l2cache(spa);
2450 spa_config_exit(spa, SCL_ALL, FTAG);
2451 } else if (error == 0) {
2452 spa->spa_l2cache.sav_sync = B_TRUE;
2453 }
2454
2455 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2456
2457 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2458 if (error && error != ENOENT)
2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2460
2461 if (error == 0) {
2462 uint64_t autoreplace;
2463
2464 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2465 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2466 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2467 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2468 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2469 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2470 &spa->spa_dedup_ditto);
2471
2472 spa->spa_autoreplace = (autoreplace != 0);
2473 }
2474
2475 /*
2476 * If the 'autoreplace' property is set, then post a resource notifying
2477 * the ZFS DE that it should not issue any faults for unopenable
2478 * devices. We also iterate over the vdevs, and post a sysevent for any
2479 * unopenable vdevs so that the normal autoreplace handler can take
2480 * over.
2481 */
2482 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2483 spa_check_removed(spa->spa_root_vdev);
2484 /*
2485 * For the import case, this is done in spa_import(), because
2486 * at this point we're using the spare definitions from
2487 * the MOS config, not necessarily from the userland config.
2488 */
2489 if (state != SPA_LOAD_IMPORT) {
2490 spa_aux_check_removed(&spa->spa_spares);
2491 spa_aux_check_removed(&spa->spa_l2cache);
2492 }
2493 }
2494
2495 /*
2496 * Load the vdev state for all toplevel vdevs.
2497 */
2498 vdev_load(rvd);
2499
2500 /*
2501 * Propagate the leaf DTLs we just loaded all the way up the tree.
2502 */
2503 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2504 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2505 spa_config_exit(spa, SCL_ALL, FTAG);
2506
2507 /*
2508 * Load the DDTs (dedup tables).
2509 */
2510 error = ddt_load(spa);
2511 if (error != 0)
2512 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2513
2514 spa_update_dspace(spa);
2515
2516 /*
2517 * Validate the config, using the MOS config to fill in any
2518 * information which might be missing. If we fail to validate
2519 * the config then declare the pool unfit for use. If we're
2520 * assembling a pool from a split, the log is not transferred
2521 * over.
2522 */
2523 if (type != SPA_IMPORT_ASSEMBLE) {
2524 nvlist_t *nvconfig;
2525
2526 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528
2529 if (!spa_config_valid(spa, nvconfig)) {
2530 nvlist_free(nvconfig);
2531 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2532 ENXIO));
2533 }
2534 nvlist_free(nvconfig);
2535
2536 /*
2537 * Now that we've validated the config, check the state of the
2538 * root vdev. If it can't be opened, it indicates one or
2539 * more toplevel vdevs are faulted.
2540 */
2541 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2542 return (ENXIO);
2543
2544 if (spa_check_logs(spa)) {
2545 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2546 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2547 }
2548 }
2549
2550 if (missing_feat_write) {
2551 ASSERT(state == SPA_LOAD_TRYIMPORT);
2552
2553 /*
2554 * At this point, we know that we can open the pool in
2555 * read-only mode but not read-write mode. We now have enough
2556 * information and can return to userland.
2557 */
2558 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2559 }
2560
2561 /*
2562 * We've successfully opened the pool, verify that we're ready
2563 * to start pushing transactions.
2564 */
2565 if (state != SPA_LOAD_TRYIMPORT) {
2566 if (error = spa_load_verify(spa))
2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2568 error));
2569 }
2570
2571 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2572 spa->spa_load_max_txg == UINT64_MAX)) {
2573 dmu_tx_t *tx;
2574 int need_update = B_FALSE;
2575
2576 ASSERT(state != SPA_LOAD_TRYIMPORT);
2577
2578 /*
2579 * Claim log blocks that haven't been committed yet.
2580 * This must all happen in a single txg.
2581 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2582 * invoked from zil_claim_log_block()'s i/o done callback.
2583 * Price of rollback is that we abandon the log.
2584 */
2585 spa->spa_claiming = B_TRUE;
2586
2587 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2588 spa_first_txg(spa));
2589 (void) dmu_objset_find(spa_name(spa),
2590 zil_claim, tx, DS_FIND_CHILDREN);
2591 dmu_tx_commit(tx);
2592
2593 spa->spa_claiming = B_FALSE;
2594
2595 spa_set_log_state(spa, SPA_LOG_GOOD);
2596 spa->spa_sync_on = B_TRUE;
2597 txg_sync_start(spa->spa_dsl_pool);
2598
2599 /*
2600 * Wait for all claims to sync. We sync up to the highest
2601 * claimed log block birth time so that claimed log blocks
2602 * don't appear to be from the future. spa_claim_max_txg
2603 * will have been set for us by either zil_check_log_chain()
2604 * (invoked from spa_check_logs()) or zil_claim() above.
2605 */
2606 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2607
2608 /*
2609 * If the config cache is stale, or we have uninitialized
2610 * metaslabs (see spa_vdev_add()), then update the config.
2611 *
2612 * If this is a verbatim import, trust the current
2613 * in-core spa_config and update the disk labels.
2614 */
2615 if (config_cache_txg != spa->spa_config_txg ||
2616 state == SPA_LOAD_IMPORT ||
2617 state == SPA_LOAD_RECOVER ||
2618 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2619 need_update = B_TRUE;
2620
2621 for (int c = 0; c < rvd->vdev_children; c++)
2622 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2623 need_update = B_TRUE;
2624
2625 /*
2626 * Update the config cache asychronously in case we're the
2627 * root pool, in which case the config cache isn't writable yet.
2628 */
2629 if (need_update)
2630 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2631
2632 /*
2633 * Check all DTLs to see if anything needs resilvering.
2634 */
2635 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2636 vdev_resilver_needed(rvd, NULL, NULL))
2637 spa_async_request(spa, SPA_ASYNC_RESILVER);
2638
2639 /*
2640 * Delete any inconsistent datasets.
2641 */
2642 (void) dmu_objset_find(spa_name(spa),
2643 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2644
2645 /*
2646 * Clean up any stale temporary dataset userrefs.
2647 */
2648 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2649 }
2650
2651 return (0);
2652}
2653
2654static int
2655spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2656{
2657 int mode = spa->spa_mode;
2658
2659 spa_unload(spa);
2660 spa_deactivate(spa);
2661
2662 spa->spa_load_max_txg--;
2663
2664 spa_activate(spa, mode);
2665 spa_async_suspend(spa);
2666
2667 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2668}
2669
2670/*
2671 * If spa_load() fails this function will try loading prior txg's. If
2672 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2673 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2674 * function will not rewind the pool and will return the same error as
2675 * spa_load().
2676 */
2677static int
2678spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2679 uint64_t max_request, int rewind_flags)
2680{
2681 nvlist_t *loadinfo = NULL;
2682 nvlist_t *config = NULL;
2683 int load_error, rewind_error;
2684 uint64_t safe_rewind_txg;
2685 uint64_t min_txg;
2686
2687 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2688 spa->spa_load_max_txg = spa->spa_load_txg;
2689 spa_set_log_state(spa, SPA_LOG_CLEAR);
2690 } else {
2691 spa->spa_load_max_txg = max_request;
2692 }
2693
2694 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2695 mosconfig);
2696 if (load_error == 0)
2697 return (0);
2698
2699 if (spa->spa_root_vdev != NULL)
2700 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2701
2702 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2703 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2704
2705 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2706 nvlist_free(config);
2707 return (load_error);
2708 }
2709
2710 if (state == SPA_LOAD_RECOVER) {
2711 /* Price of rolling back is discarding txgs, including log */
2712 spa_set_log_state(spa, SPA_LOG_CLEAR);
2713 } else {
2714 /*
2715 * If we aren't rolling back save the load info from our first
2716 * import attempt so that we can restore it after attempting
2717 * to rewind.
2718 */
2719 loadinfo = spa->spa_load_info;
2720 spa->spa_load_info = fnvlist_alloc();
2721 }
2722
2723 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2724 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2725 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2726 TXG_INITIAL : safe_rewind_txg;
2727
2728 /*
2729 * Continue as long as we're finding errors, we're still within
2730 * the acceptable rewind range, and we're still finding uberblocks
2731 */
2732 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2733 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2734 if (spa->spa_load_max_txg < safe_rewind_txg)
2735 spa->spa_extreme_rewind = B_TRUE;
2736 rewind_error = spa_load_retry(spa, state, mosconfig);
2737 }
2738
2739 spa->spa_extreme_rewind = B_FALSE;
2740 spa->spa_load_max_txg = UINT64_MAX;
2741
2742 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2743 spa_config_set(spa, config);
2744
2745 if (state == SPA_LOAD_RECOVER) {
2746 ASSERT3P(loadinfo, ==, NULL);
2747 return (rewind_error);
2748 } else {
2749 /* Store the rewind info as part of the initial load info */
2750 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2751 spa->spa_load_info);
2752
2753 /* Restore the initial load info */
2754 fnvlist_free(spa->spa_load_info);
2755 spa->spa_load_info = loadinfo;
2756
2757 return (load_error);
2758 }
2759}
2760
2761/*
2762 * Pool Open/Import
2763 *
2764 * The import case is identical to an open except that the configuration is sent
2765 * down from userland, instead of grabbed from the configuration cache. For the
2766 * case of an open, the pool configuration will exist in the
2767 * POOL_STATE_UNINITIALIZED state.
2768 *
2769 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2770 * the same time open the pool, without having to keep around the spa_t in some
2771 * ambiguous state.
2772 */
2773static int
2774spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2775 nvlist_t **config)
2776{
2777 spa_t *spa;
2778 spa_load_state_t state = SPA_LOAD_OPEN;
2779 int error;
2780 int locked = B_FALSE;
2781 int firstopen = B_FALSE;
2782
2783 *spapp = NULL;
2784
2785 /*
2786 * As disgusting as this is, we need to support recursive calls to this
2787 * function because dsl_dir_open() is called during spa_load(), and ends
2788 * up calling spa_open() again. The real fix is to figure out how to
2789 * avoid dsl_dir_open() calling this in the first place.
2790 */
2791 if (mutex_owner(&spa_namespace_lock) != curthread) {
2792 mutex_enter(&spa_namespace_lock);
2793 locked = B_TRUE;
2794 }
2795
2796 if ((spa = spa_lookup(pool)) == NULL) {
2797 if (locked)
2798 mutex_exit(&spa_namespace_lock);
2799 return (ENOENT);
2800 }
2801
2802 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2803 zpool_rewind_policy_t policy;
2804
2805 firstopen = B_TRUE;
2806
2807 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2808 &policy);
2809 if (policy.zrp_request & ZPOOL_DO_REWIND)
2810 state = SPA_LOAD_RECOVER;
2811
2812 spa_activate(spa, spa_mode_global);
2813
2814 if (state != SPA_LOAD_RECOVER)
2815 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2816
2817 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2818 policy.zrp_request);
2819
2820 if (error == EBADF) {
2821 /*
2822 * If vdev_validate() returns failure (indicated by
2823 * EBADF), it indicates that one of the vdevs indicates
2824 * that the pool has been exported or destroyed. If
2825 * this is the case, the config cache is out of sync and
2826 * we should remove the pool from the namespace.
2827 */
2828 spa_unload(spa);
2829 spa_deactivate(spa);
2830 spa_config_sync(spa, B_TRUE, B_TRUE);
2831 spa_remove(spa);
2832 if (locked)
2833 mutex_exit(&spa_namespace_lock);
2834 return (ENOENT);
2835 }
2836
2837 if (error) {
2838 /*
2839 * We can't open the pool, but we still have useful
2840 * information: the state of each vdev after the
2841 * attempted vdev_open(). Return this to the user.
2842 */
2843 if (config != NULL && spa->spa_config) {
2844 VERIFY(nvlist_dup(spa->spa_config, config,
2845 KM_SLEEP) == 0);
2846 VERIFY(nvlist_add_nvlist(*config,
2847 ZPOOL_CONFIG_LOAD_INFO,
2848 spa->spa_load_info) == 0);
2849 }
2850 spa_unload(spa);
2851 spa_deactivate(spa);
2852 spa->spa_last_open_failed = error;
2853 if (locked)
2854 mutex_exit(&spa_namespace_lock);
2855 *spapp = NULL;
2856 return (error);
2857 }
2858 }
2859
2860 spa_open_ref(spa, tag);
2861
2862 if (config != NULL)
2863 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2864
2865 /*
2866 * If we've recovered the pool, pass back any information we
2867 * gathered while doing the load.
2868 */
2869 if (state == SPA_LOAD_RECOVER) {
2870 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2871 spa->spa_load_info) == 0);
2872 }
2873
2874 if (locked) {
2875 spa->spa_last_open_failed = 0;
2876 spa->spa_last_ubsync_txg = 0;
2877 spa->spa_load_txg = 0;
2878 mutex_exit(&spa_namespace_lock);
2879#ifdef __FreeBSD__
2880#ifdef _KERNEL
2881 if (firstopen)
2882 zvol_create_minors(pool);
2883#endif
2884#endif
2885 }
2886
2887 *spapp = spa;
2888
2889 return (0);
2890}
2891
2892int
2893spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2894 nvlist_t **config)
2895{
2896 return (spa_open_common(name, spapp, tag, policy, config));
2897}
2898
2899int
2900spa_open(const char *name, spa_t **spapp, void *tag)
2901{
2902 return (spa_open_common(name, spapp, tag, NULL, NULL));
2903}
2904
2905/*
2906 * Lookup the given spa_t, incrementing the inject count in the process,
2907 * preventing it from being exported or destroyed.
2908 */
2909spa_t *
2910spa_inject_addref(char *name)
2911{
2912 spa_t *spa;
2913
2914 mutex_enter(&spa_namespace_lock);
2915 if ((spa = spa_lookup(name)) == NULL) {
2916 mutex_exit(&spa_namespace_lock);
2917 return (NULL);
2918 }
2919 spa->spa_inject_ref++;
2920 mutex_exit(&spa_namespace_lock);
2921
2922 return (spa);
2923}
2924
2925void
2926spa_inject_delref(spa_t *spa)
2927{
2928 mutex_enter(&spa_namespace_lock);
2929 spa->spa_inject_ref--;
2930 mutex_exit(&spa_namespace_lock);
2931}
2932
2933/*
2934 * Add spares device information to the nvlist.
2935 */
2936static void
2937spa_add_spares(spa_t *spa, nvlist_t *config)
2938{
2939 nvlist_t **spares;
2940 uint_t i, nspares;
2941 nvlist_t *nvroot;
2942 uint64_t guid;
2943 vdev_stat_t *vs;
2944 uint_t vsc;
2945 uint64_t pool;
2946
2947 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2948
2949 if (spa->spa_spares.sav_count == 0)
2950 return;
2951
2952 VERIFY(nvlist_lookup_nvlist(config,
2953 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2954 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2955 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2956 if (nspares != 0) {
2957 VERIFY(nvlist_add_nvlist_array(nvroot,
2958 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2959 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2960 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2961
2962 /*
2963 * Go through and find any spares which have since been
2964 * repurposed as an active spare. If this is the case, update
2965 * their status appropriately.
2966 */
2967 for (i = 0; i < nspares; i++) {
2968 VERIFY(nvlist_lookup_uint64(spares[i],
2969 ZPOOL_CONFIG_GUID, &guid) == 0);
2970 if (spa_spare_exists(guid, &pool, NULL) &&
2971 pool != 0ULL) {
2972 VERIFY(nvlist_lookup_uint64_array(
2973 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2974 (uint64_t **)&vs, &vsc) == 0);
2975 vs->vs_state = VDEV_STATE_CANT_OPEN;
2976 vs->vs_aux = VDEV_AUX_SPARED;
2977 }
2978 }
2979 }
2980}
2981
2982/*
2983 * Add l2cache device information to the nvlist, including vdev stats.
2984 */
2985static void
2986spa_add_l2cache(spa_t *spa, nvlist_t *config)
2987{
2988 nvlist_t **l2cache;
2989 uint_t i, j, nl2cache;
2990 nvlist_t *nvroot;
2991 uint64_t guid;
2992 vdev_t *vd;
2993 vdev_stat_t *vs;
2994 uint_t vsc;
2995
2996 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2997
2998 if (spa->spa_l2cache.sav_count == 0)
2999 return;
3000
3001 VERIFY(nvlist_lookup_nvlist(config,
3002 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3003 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3004 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3005 if (nl2cache != 0) {
3006 VERIFY(nvlist_add_nvlist_array(nvroot,
3007 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3008 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3009 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3010
3011 /*
3012 * Update level 2 cache device stats.
3013 */
3014
3015 for (i = 0; i < nl2cache; i++) {
3016 VERIFY(nvlist_lookup_uint64(l2cache[i],
3017 ZPOOL_CONFIG_GUID, &guid) == 0);
3018
3019 vd = NULL;
3020 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3021 if (guid ==
3022 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3023 vd = spa->spa_l2cache.sav_vdevs[j];
3024 break;
3025 }
3026 }
3027 ASSERT(vd != NULL);
3028
3029 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3030 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3031 == 0);
3032 vdev_get_stats(vd, vs);
3033 }
3034 }
3035}
3036
3037static void
3038spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3039{
3040 nvlist_t *features;
3041 zap_cursor_t zc;
3042 zap_attribute_t za;
3043
3044 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3045 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3046
3047 if (spa->spa_feat_for_read_obj != 0) {
3048 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3049 spa->spa_feat_for_read_obj);
3050 zap_cursor_retrieve(&zc, &za) == 0;
3051 zap_cursor_advance(&zc)) {
3052 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3053 za.za_num_integers == 1);
3054 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3055 za.za_first_integer));
3056 }
3057 zap_cursor_fini(&zc);
3058 }
3059
3060 if (spa->spa_feat_for_write_obj != 0) {
3061 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3062 spa->spa_feat_for_write_obj);
3063 zap_cursor_retrieve(&zc, &za) == 0;
3064 zap_cursor_advance(&zc)) {
3065 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3066 za.za_num_integers == 1);
3067 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3068 za.za_first_integer));
3069 }
3070 zap_cursor_fini(&zc);
3071 }
3072
3073 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3074 features) == 0);
3075 nvlist_free(features);
3076}
3077
3078int
3079spa_get_stats(const char *name, nvlist_t **config,
3080 char *altroot, size_t buflen)
3081{
3082 int error;
3083 spa_t *spa;
3084
3085 *config = NULL;
3086 error = spa_open_common(name, &spa, FTAG, NULL, config);
3087
3088 if (spa != NULL) {
3089 /*
3090 * This still leaves a window of inconsistency where the spares
3091 * or l2cache devices could change and the config would be
3092 * self-inconsistent.
3093 */
3094 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3095
3096 if (*config != NULL) {
3097 uint64_t loadtimes[2];
3098
3099 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3100 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3101 VERIFY(nvlist_add_uint64_array(*config,
3102 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3103
3104 VERIFY(nvlist_add_uint64(*config,
3105 ZPOOL_CONFIG_ERRCOUNT,
3106 spa_get_errlog_size(spa)) == 0);
3107
3108 if (spa_suspended(spa))
3109 VERIFY(nvlist_add_uint64(*config,
3110 ZPOOL_CONFIG_SUSPENDED,
3111 spa->spa_failmode) == 0);
3112
3113 spa_add_spares(spa, *config);
3114 spa_add_l2cache(spa, *config);
3115 spa_add_feature_stats(spa, *config);
3116 }
3117 }
3118
3119 /*
3120 * We want to get the alternate root even for faulted pools, so we cheat
3121 * and call spa_lookup() directly.
3122 */
3123 if (altroot) {
3124 if (spa == NULL) {
3125 mutex_enter(&spa_namespace_lock);
3126 spa = spa_lookup(name);
3127 if (spa)
3128 spa_altroot(spa, altroot, buflen);
3129 else
3130 altroot[0] = '\0';
3131 spa = NULL;
3132 mutex_exit(&spa_namespace_lock);
3133 } else {
3134 spa_altroot(spa, altroot, buflen);
3135 }
3136 }
3137
3138 if (spa != NULL) {
3139 spa_config_exit(spa, SCL_CONFIG, FTAG);
3140 spa_close(spa, FTAG);
3141 }
3142
3143 return (error);
3144}
3145
3146/*
3147 * Validate that the auxiliary device array is well formed. We must have an
3148 * array of nvlists, each which describes a valid leaf vdev. If this is an
3149 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3150 * specified, as long as they are well-formed.
3151 */
3152static int
3153spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3154 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3155 vdev_labeltype_t label)
3156{
3157 nvlist_t **dev;
3158 uint_t i, ndev;
3159 vdev_t *vd;
3160 int error;
3161
3162 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3163
3164 /*
3165 * It's acceptable to have no devs specified.
3166 */
3167 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3168 return (0);
3169
3170 if (ndev == 0)
3171 return (EINVAL);
3172
3173 /*
3174 * Make sure the pool is formatted with a version that supports this
3175 * device type.
3176 */
3177 if (spa_version(spa) < version)
3178 return (ENOTSUP);
3179
3180 /*
3181 * Set the pending device list so we correctly handle device in-use
3182 * checking.
3183 */
3184 sav->sav_pending = dev;
3185 sav->sav_npending = ndev;
3186
3187 for (i = 0; i < ndev; i++) {
3188 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3189 mode)) != 0)
3190 goto out;
3191
3192 if (!vd->vdev_ops->vdev_op_leaf) {
3193 vdev_free(vd);
3194 error = EINVAL;
3195 goto out;
3196 }
3197
3198 /*
3199 * The L2ARC currently only supports disk devices in
3200 * kernel context. For user-level testing, we allow it.
3201 */
3202#ifdef _KERNEL
3203 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3204 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3205 error = ENOTBLK;
3206 vdev_free(vd);
3207 goto out;
3208 }
3209#endif
3210 vd->vdev_top = vd;
3211
3212 if ((error = vdev_open(vd)) == 0 &&
3213 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3214 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3215 vd->vdev_guid) == 0);
3216 }
3217
3218 vdev_free(vd);
3219
3220 if (error &&
3221 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3222 goto out;
3223 else
3224 error = 0;
3225 }
3226
3227out:
3228 sav->sav_pending = NULL;
3229 sav->sav_npending = 0;
3230 return (error);
3231}
3232
3233static int
3234spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3235{
3236 int error;
3237
3238 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3239
3240 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3241 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3242 VDEV_LABEL_SPARE)) != 0) {
3243 return (error);
3244 }
3245
3246 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3247 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3248 VDEV_LABEL_L2CACHE));
3249}
3250
3251static void
3252spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3253 const char *config)
3254{
3255 int i;
3256
3257 if (sav->sav_config != NULL) {
3258 nvlist_t **olddevs;
3259 uint_t oldndevs;
3260 nvlist_t **newdevs;
3261
3262 /*
3263 * Generate new dev list by concatentating with the
3264 * current dev list.
3265 */
3266 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3267 &olddevs, &oldndevs) == 0);
3268
3269 newdevs = kmem_alloc(sizeof (void *) *
3270 (ndevs + oldndevs), KM_SLEEP);
3271 for (i = 0; i < oldndevs; i++)
3272 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3273 KM_SLEEP) == 0);
3274 for (i = 0; i < ndevs; i++)
3275 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3276 KM_SLEEP) == 0);
3277
3278 VERIFY(nvlist_remove(sav->sav_config, config,
3279 DATA_TYPE_NVLIST_ARRAY) == 0);
3280
3281 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3282 config, newdevs, ndevs + oldndevs) == 0);
3283 for (i = 0; i < oldndevs + ndevs; i++)
3284 nvlist_free(newdevs[i]);
3285 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3286 } else {
3287 /*
3288 * Generate a new dev list.
3289 */
3290 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3291 KM_SLEEP) == 0);
3292 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3293 devs, ndevs) == 0);
3294 }
3295}
3296
3297/*
3298 * Stop and drop level 2 ARC devices
3299 */
3300void
3301spa_l2cache_drop(spa_t *spa)
3302{
3303 vdev_t *vd;
3304 int i;
3305 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3306
3307 for (i = 0; i < sav->sav_count; i++) {
3308 uint64_t pool;
3309
3310 vd = sav->sav_vdevs[i];
3311 ASSERT(vd != NULL);
3312
3313 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3314 pool != 0ULL && l2arc_vdev_present(vd))
3315 l2arc_remove_vdev(vd);
3316 }
3317}
3318
3319/*
3320 * Pool Creation
3321 */
3322int
3323spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3324 const char *history_str, nvlist_t *zplprops)
3325{
3326 spa_t *spa;
3327 char *altroot = NULL;
3328 vdev_t *rvd;
3329 dsl_pool_t *dp;
3330 dmu_tx_t *tx;
3331 int error = 0;
3332 uint64_t txg = TXG_INITIAL;
3333 nvlist_t **spares, **l2cache;
3334 uint_t nspares, nl2cache;
3335 uint64_t version, obj;
3336 boolean_t has_features;
3337
3338 /*
3339 * If this pool already exists, return failure.
3340 */
3341 mutex_enter(&spa_namespace_lock);
3342 if (spa_lookup(pool) != NULL) {
3343 mutex_exit(&spa_namespace_lock);
3344 return (EEXIST);
3345 }
3346
3347 /*
3348 * Allocate a new spa_t structure.
3349 */
3350 (void) nvlist_lookup_string(props,
3351 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3352 spa = spa_add(pool, NULL, altroot);
3353 spa_activate(spa, spa_mode_global);
3354
3355 if (props && (error = spa_prop_validate(spa, props))) {
3356 spa_deactivate(spa);
3357 spa_remove(spa);
3358 mutex_exit(&spa_namespace_lock);
3359 return (error);
3360 }
3361
3362 has_features = B_FALSE;
3363 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3364 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3365 if (zpool_prop_feature(nvpair_name(elem)))
3366 has_features = B_TRUE;
3367 }
3368
3369 if (has_features || nvlist_lookup_uint64(props,
3370 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3371 version = SPA_VERSION;
3372 }
3373 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3374
3375 spa->spa_first_txg = txg;
3376 spa->spa_uberblock.ub_txg = txg - 1;
3377 spa->spa_uberblock.ub_version = version;
3378 spa->spa_ubsync = spa->spa_uberblock;
3379
3380 /*
3381 * Create "The Godfather" zio to hold all async IOs
3382 */
3383 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3384 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3385
3386 /*
3387 * Create the root vdev.
3388 */
3389 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3390
3391 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3392
3393 ASSERT(error != 0 || rvd != NULL);
3394 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3395
3396 if (error == 0 && !zfs_allocatable_devs(nvroot))
3397 error = EINVAL;
3398
3399 if (error == 0 &&
3400 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3401 (error = spa_validate_aux(spa, nvroot, txg,
3402 VDEV_ALLOC_ADD)) == 0) {
3403 for (int c = 0; c < rvd->vdev_children; c++) {
3404 vdev_metaslab_set_size(rvd->vdev_child[c]);
3405 vdev_expand(rvd->vdev_child[c], txg);
3406 }
3407 }
3408
3409 spa_config_exit(spa, SCL_ALL, FTAG);
3410
3411 if (error != 0) {
3412 spa_unload(spa);
3413 spa_deactivate(spa);
3414 spa_remove(spa);
3415 mutex_exit(&spa_namespace_lock);
3416 return (error);
3417 }
3418
3419 /*
3420 * Get the list of spares, if specified.
3421 */
3422 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3423 &spares, &nspares) == 0) {
3424 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3425 KM_SLEEP) == 0);
3426 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3427 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3429 spa_load_spares(spa);
3430 spa_config_exit(spa, SCL_ALL, FTAG);
3431 spa->spa_spares.sav_sync = B_TRUE;
3432 }
3433
3434 /*
3435 * Get the list of level 2 cache devices, if specified.
3436 */
3437 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3438 &l2cache, &nl2cache) == 0) {
3439 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3440 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3441 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3442 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3444 spa_load_l2cache(spa);
3445 spa_config_exit(spa, SCL_ALL, FTAG);
3446 spa->spa_l2cache.sav_sync = B_TRUE;
3447 }
3448
3449 spa->spa_is_initializing = B_TRUE;
3450 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3451 spa->spa_meta_objset = dp->dp_meta_objset;
3452 spa->spa_is_initializing = B_FALSE;
3453
3454 /*
3455 * Create DDTs (dedup tables).
3456 */
3457 ddt_create(spa);
3458
3459 spa_update_dspace(spa);
3460
3461 tx = dmu_tx_create_assigned(dp, txg);
3462
3463 /*
3464 * Create the pool config object.
3465 */
3466 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3467 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3468 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3469
3470 if (zap_add(spa->spa_meta_objset,
3471 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3472 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3473 cmn_err(CE_PANIC, "failed to add pool config");
3474 }
3475
3476 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3477 spa_feature_create_zap_objects(spa, tx);
3478
3479 if (zap_add(spa->spa_meta_objset,
3480 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3481 sizeof (uint64_t), 1, &version, tx) != 0) {
3482 cmn_err(CE_PANIC, "failed to add pool version");
3483 }
3484
3485 /* Newly created pools with the right version are always deflated. */
3486 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3487 spa->spa_deflate = TRUE;
3488 if (zap_add(spa->spa_meta_objset,
3489 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3490 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3491 cmn_err(CE_PANIC, "failed to add deflate");
3492 }
3493 }
3494
3495 /*
3496 * Create the deferred-free bpobj. Turn off compression
3497 * because sync-to-convergence takes longer if the blocksize
3498 * keeps changing.
3499 */
3500 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3501 dmu_object_set_compress(spa->spa_meta_objset, obj,
3502 ZIO_COMPRESS_OFF, tx);
3503 if (zap_add(spa->spa_meta_objset,
3504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3505 sizeof (uint64_t), 1, &obj, tx) != 0) {
3506 cmn_err(CE_PANIC, "failed to add bpobj");
3507 }
3508 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3509 spa->spa_meta_objset, obj));
3510
3511 /*
3512 * Create the pool's history object.
3513 */
3514 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3515 spa_history_create_obj(spa, tx);
3516
3517 /*
3518 * Set pool properties.
3519 */
3520 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3521 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3522 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3523 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3524
3525 if (props != NULL) {
3526 spa_configfile_set(spa, props, B_FALSE);
3527 spa_sync_props(spa, props, tx);
3528 }
3529
3530 dmu_tx_commit(tx);
3531
3532 spa->spa_sync_on = B_TRUE;
3533 txg_sync_start(spa->spa_dsl_pool);
3534
3535 /*
3536 * We explicitly wait for the first transaction to complete so that our
3537 * bean counters are appropriately updated.
3538 */
3539 txg_wait_synced(spa->spa_dsl_pool, txg);
3540
3541 spa_config_sync(spa, B_FALSE, B_TRUE);
3542
3543 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3544 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3545 spa_history_log_version(spa, LOG_POOL_CREATE);
3546
3547 spa->spa_minref = refcount_count(&spa->spa_refcount);
3548
3549 mutex_exit(&spa_namespace_lock);
3550
3551 return (0);
3552}
3553
3554#ifdef _KERNEL
3555#if defined(sun)
3556/*
3557 * Get the root pool information from the root disk, then import the root pool
3558 * during the system boot up time.
3559 */
3560extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3561
3562static nvlist_t *
3563spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3564{
3565 nvlist_t *config;
3566 nvlist_t *nvtop, *nvroot;
3567 uint64_t pgid;
3568
3569 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3570 return (NULL);
3571
3572 /*
3573 * Add this top-level vdev to the child array.
3574 */
3575 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3576 &nvtop) == 0);
3577 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3578 &pgid) == 0);
3579 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3580
3581 /*
3582 * Put this pool's top-level vdevs into a root vdev.
3583 */
3584 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3585 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3586 VDEV_TYPE_ROOT) == 0);
3587 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3588 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3589 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3590 &nvtop, 1) == 0);
3591
3592 /*
3593 * Replace the existing vdev_tree with the new root vdev in
3594 * this pool's configuration (remove the old, add the new).
3595 */
3596 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3597 nvlist_free(nvroot);
3598 return (config);
3599}
3600
3601/*
3602 * Walk the vdev tree and see if we can find a device with "better"
3603 * configuration. A configuration is "better" if the label on that
3604 * device has a more recent txg.
3605 */
3606static void
3607spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3608{
3609 for (int c = 0; c < vd->vdev_children; c++)
3610 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3611
3612 if (vd->vdev_ops->vdev_op_leaf) {
3613 nvlist_t *label;
3614 uint64_t label_txg;
3615
3616 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3617 &label) != 0)
3618 return;
3619
3620 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3621 &label_txg) == 0);
3622
3623 /*
3624 * Do we have a better boot device?
3625 */
3626 if (label_txg > *txg) {
3627 *txg = label_txg;
3628 *avd = vd;
3629 }
3630 nvlist_free(label);
3631 }
3632}
3633
3634/*
3635 * Import a root pool.
3636 *
3637 * For x86. devpath_list will consist of devid and/or physpath name of
3638 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3639 * The GRUB "findroot" command will return the vdev we should boot.
3640 *
3641 * For Sparc, devpath_list consists the physpath name of the booting device
3642 * no matter the rootpool is a single device pool or a mirrored pool.
3643 * e.g.
3644 * "/pci@1f,0/ide@d/disk@0,0:a"
3645 */
3646int
3647spa_import_rootpool(char *devpath, char *devid)
3648{
3649 spa_t *spa;
3650 vdev_t *rvd, *bvd, *avd = NULL;
3651 nvlist_t *config, *nvtop;
3652 uint64_t guid, txg;
3653 char *pname;
3654 int error;
3655
3656 /*
3657 * Read the label from the boot device and generate a configuration.
3658 */
3659 config = spa_generate_rootconf(devpath, devid, &guid);
3660#if defined(_OBP) && defined(_KERNEL)
3661 if (config == NULL) {
3662 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3663 /* iscsi boot */
3664 get_iscsi_bootpath_phy(devpath);
3665 config = spa_generate_rootconf(devpath, devid, &guid);
3666 }
3667 }
3668#endif
3669 if (config == NULL) {
3670 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3671 devpath);
3672 return (EIO);
3673 }
3674
3675 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3676 &pname) == 0);
3677 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3678
3679 mutex_enter(&spa_namespace_lock);
3680 if ((spa = spa_lookup(pname)) != NULL) {
3681 /*
3682 * Remove the existing root pool from the namespace so that we
3683 * can replace it with the correct config we just read in.
3684 */
3685 spa_remove(spa);
3686 }
3687
3688 spa = spa_add(pname, config, NULL);
3689 spa->spa_is_root = B_TRUE;
3690 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3691
3692 /*
3693 * Build up a vdev tree based on the boot device's label config.
3694 */
3695 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3696 &nvtop) == 0);
3697 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3698 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3699 VDEV_ALLOC_ROOTPOOL);
3700 spa_config_exit(spa, SCL_ALL, FTAG);
3701 if (error) {
3702 mutex_exit(&spa_namespace_lock);
3703 nvlist_free(config);
3704 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3705 pname);
3706 return (error);
3707 }
3708
3709 /*
3710 * Get the boot vdev.
3711 */
3712 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3713 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3714 (u_longlong_t)guid);
3715 error = ENOENT;
3716 goto out;
3717 }
3718
3719 /*
3720 * Determine if there is a better boot device.
3721 */
3722 avd = bvd;
3723 spa_alt_rootvdev(rvd, &avd, &txg);
3724 if (avd != bvd) {
3725 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3726 "try booting from '%s'", avd->vdev_path);
3727 error = EINVAL;
3728 goto out;
3729 }
3730
3731 /*
3732 * If the boot device is part of a spare vdev then ensure that
3733 * we're booting off the active spare.
3734 */
3735 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3736 !bvd->vdev_isspare) {
3737 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3738 "try booting from '%s'",
3739 bvd->vdev_parent->
3740 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3741 error = EINVAL;
3742 goto out;
3743 }
3744
3745 error = 0;
3746 spa_history_log_version(spa, LOG_POOL_IMPORT);
3747out:
3748 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3749 vdev_free(rvd);
3750 spa_config_exit(spa, SCL_ALL, FTAG);
3751 mutex_exit(&spa_namespace_lock);
3752
3753 nvlist_free(config);
3754 return (error);
3755}
3756
3757#else
3758
3759extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3760 uint64_t *count);
3761
3762static nvlist_t *
3763spa_generate_rootconf(const char *name)
3764{
3765 nvlist_t **configs, **tops;
3766 nvlist_t *config;
3767 nvlist_t *best_cfg, *nvtop, *nvroot;
3768 uint64_t *holes;
3769 uint64_t best_txg;
3770 uint64_t nchildren;
3771 uint64_t pgid;
3772 uint64_t count;
3773 uint64_t i;
3774 uint_t nholes;
3775
3776 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3777 return (NULL);
3778
3779 ASSERT3U(count, !=, 0);
3780 best_txg = 0;
3781 for (i = 0; i < count; i++) {
3782 uint64_t txg;
3783
3784 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3785 &txg) == 0);
3786 if (txg > best_txg) {
3787 best_txg = txg;
3788 best_cfg = configs[i];
3789 }
3790 }
3791
3792 /*
3793 * Multi-vdev root pool configuration discovery is not supported yet.
3794 */
|
3798 holes = NULL;
3799 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3800 &holes, &nholes);
3801
3802 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3803 for (i = 0; i < nchildren; i++) {
3804 if (i >= count)
3805 break;
3806 if (configs[i] == NULL)
3807 continue;
3808 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3809 &nvtop) == 0);
3810 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3811 }
3812 for (i = 0; holes != NULL && i < nholes; i++) {
3813 if (i >= nchildren)
3814 continue;
3815 if (tops[holes[i]] != NULL)
3816 continue;
3817 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3818 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3819 VDEV_TYPE_HOLE) == 0);
3820 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3821 holes[i]) == 0);
3822 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3823 0) == 0);
3824 }
3825 for (i = 0; i < nchildren; i++) {
3826 if (tops[i] != NULL)
3827 continue;
3828 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3829 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3830 VDEV_TYPE_MISSING) == 0);
3831 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3832 i) == 0);
3833 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3834 0) == 0);
3835 }
3836
3837 /*
3838 * Create pool config based on the best vdev config.
3839 */
3840 nvlist_dup(best_cfg, &config, KM_SLEEP);
3841
3842 /*
3843 * Put this pool's top-level vdevs into a root vdev.
3844 */
3845 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3846 &pgid) == 0);
3847 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3848 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3849 VDEV_TYPE_ROOT) == 0);
3850 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3851 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3852 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3853 tops, nchildren) == 0);
3854
3855 /*
3856 * Replace the existing vdev_tree with the new root vdev in
3857 * this pool's configuration (remove the old, add the new).
3858 */
3859 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3860
3861 /*
3862 * Drop vdev config elements that should not be present at pool level.
3863 */
3864 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3865 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3866
3867 for (i = 0; i < count; i++)
3868 nvlist_free(configs[i]);
3869 kmem_free(configs, count * sizeof(void *));
3870 for (i = 0; i < nchildren; i++)
3871 nvlist_free(tops[i]);
3872 kmem_free(tops, nchildren * sizeof(void *));
3873 nvlist_free(nvroot);
3874 return (config);
3875}
3876
3877int
3878spa_import_rootpool(const char *name)
3879{
3880 spa_t *spa;
3881 vdev_t *rvd, *bvd, *avd = NULL;
3882 nvlist_t *config, *nvtop;
3883 uint64_t txg;
3884 char *pname;
3885 int error;
3886
3887 /*
3888 * Read the label from the boot device and generate a configuration.
3889 */
3890 config = spa_generate_rootconf(name);
3891
3892 mutex_enter(&spa_namespace_lock);
3893 if (config != NULL) {
3894 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3895 &pname) == 0 && strcmp(name, pname) == 0);
3896 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
3897 == 0);
3898
3899 if ((spa = spa_lookup(pname)) != NULL) {
3900 /*
3901 * Remove the existing root pool from the namespace so
3902 * that we can replace it with the correct config
3903 * we just read in.
3904 */
3905 spa_remove(spa);
3906 }
3907 spa = spa_add(pname, config, NULL);
3908
3909 /*
3910 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
3911 * via spa_version().
3912 */
3913 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3914 &spa->spa_ubsync.ub_version) != 0)
3915 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3916 } else if ((spa = spa_lookup(name)) == NULL) {
3917 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
3918 name);
3919 return (EIO);
3920 } else {
3921 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
3922 }
3923 spa->spa_is_root = B_TRUE;
3924 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3925
3926 /*
3927 * Build up a vdev tree based on the boot device's label config.
3928 */
3929 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3930 &nvtop) == 0);
3931 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3932 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3933 VDEV_ALLOC_ROOTPOOL);
3934 spa_config_exit(spa, SCL_ALL, FTAG);
3935 if (error) {
3936 mutex_exit(&spa_namespace_lock);
3937 nvlist_free(config);
3938 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3939 pname);
3940 return (error);
3941 }
3942
3943 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3944 vdev_free(rvd);
3945 spa_config_exit(spa, SCL_ALL, FTAG);
3946 mutex_exit(&spa_namespace_lock);
3947
3948 nvlist_free(config);
3949 return (0);
3950}
3951
3952#endif /* sun */
3953#endif
3954
3955/*
3956 * Import a non-root pool into the system.
3957 */
3958int
3959spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3960{
3961 spa_t *spa;
3962 char *altroot = NULL;
3963 spa_load_state_t state = SPA_LOAD_IMPORT;
3964 zpool_rewind_policy_t policy;
3965 uint64_t mode = spa_mode_global;
3966 uint64_t readonly = B_FALSE;
3967 int error;
3968 nvlist_t *nvroot;
3969 nvlist_t **spares, **l2cache;
3970 uint_t nspares, nl2cache;
3971
3972 /*
3973 * If a pool with this name exists, return failure.
3974 */
3975 mutex_enter(&spa_namespace_lock);
3976 if (spa_lookup(pool) != NULL) {
3977 mutex_exit(&spa_namespace_lock);
3978 return (EEXIST);
3979 }
3980
3981 /*
3982 * Create and initialize the spa structure.
3983 */
3984 (void) nvlist_lookup_string(props,
3985 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3986 (void) nvlist_lookup_uint64(props,
3987 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3988 if (readonly)
3989 mode = FREAD;
3990 spa = spa_add(pool, config, altroot);
3991 spa->spa_import_flags = flags;
3992
3993 /*
3994 * Verbatim import - Take a pool and insert it into the namespace
3995 * as if it had been loaded at boot.
3996 */
3997 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3998 if (props != NULL)
3999 spa_configfile_set(spa, props, B_FALSE);
4000
4001 spa_config_sync(spa, B_FALSE, B_TRUE);
4002
4003 mutex_exit(&spa_namespace_lock);
4004 spa_history_log_version(spa, LOG_POOL_IMPORT);
4005
4006 return (0);
4007 }
4008
4009 spa_activate(spa, mode);
4010
4011 /*
4012 * Don't start async tasks until we know everything is healthy.
4013 */
4014 spa_async_suspend(spa);
4015
4016 zpool_get_rewind_policy(config, &policy);
4017 if (policy.zrp_request & ZPOOL_DO_REWIND)
4018 state = SPA_LOAD_RECOVER;
4019
4020 /*
4021 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4022 * because the user-supplied config is actually the one to trust when
4023 * doing an import.
4024 */
4025 if (state != SPA_LOAD_RECOVER)
4026 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4027
4028 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4029 policy.zrp_request);
4030
4031 /*
4032 * Propagate anything learned while loading the pool and pass it
4033 * back to caller (i.e. rewind info, missing devices, etc).
4034 */
4035 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4036 spa->spa_load_info) == 0);
4037
4038 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4039 /*
4040 * Toss any existing sparelist, as it doesn't have any validity
4041 * anymore, and conflicts with spa_has_spare().
4042 */
4043 if (spa->spa_spares.sav_config) {
4044 nvlist_free(spa->spa_spares.sav_config);
4045 spa->spa_spares.sav_config = NULL;
4046 spa_load_spares(spa);
4047 }
4048 if (spa->spa_l2cache.sav_config) {
4049 nvlist_free(spa->spa_l2cache.sav_config);
4050 spa->spa_l2cache.sav_config = NULL;
4051 spa_load_l2cache(spa);
4052 }
4053
4054 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4055 &nvroot) == 0);
4056 if (error == 0)
4057 error = spa_validate_aux(spa, nvroot, -1ULL,
4058 VDEV_ALLOC_SPARE);
4059 if (error == 0)
4060 error = spa_validate_aux(spa, nvroot, -1ULL,
4061 VDEV_ALLOC_L2CACHE);
4062 spa_config_exit(spa, SCL_ALL, FTAG);
4063
4064 if (props != NULL)
4065 spa_configfile_set(spa, props, B_FALSE);
4066
4067 if (error != 0 || (props && spa_writeable(spa) &&
4068 (error = spa_prop_set(spa, props)))) {
4069 spa_unload(spa);
4070 spa_deactivate(spa);
4071 spa_remove(spa);
4072 mutex_exit(&spa_namespace_lock);
4073 return (error);
4074 }
4075
4076 spa_async_resume(spa);
4077
4078 /*
4079 * Override any spares and level 2 cache devices as specified by
4080 * the user, as these may have correct device names/devids, etc.
4081 */
4082 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4083 &spares, &nspares) == 0) {
4084 if (spa->spa_spares.sav_config)
4085 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4086 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4087 else
4088 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4089 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4090 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4091 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4092 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4093 spa_load_spares(spa);
4094 spa_config_exit(spa, SCL_ALL, FTAG);
4095 spa->spa_spares.sav_sync = B_TRUE;
4096 }
4097 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4098 &l2cache, &nl2cache) == 0) {
4099 if (spa->spa_l2cache.sav_config)
4100 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4101 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4102 else
4103 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4104 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4105 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4106 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4107 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4108 spa_load_l2cache(spa);
4109 spa_config_exit(spa, SCL_ALL, FTAG);
4110 spa->spa_l2cache.sav_sync = B_TRUE;
4111 }
4112
4113 /*
4114 * Check for any removed devices.
4115 */
4116 if (spa->spa_autoreplace) {
4117 spa_aux_check_removed(&spa->spa_spares);
4118 spa_aux_check_removed(&spa->spa_l2cache);
4119 }
4120
4121 if (spa_writeable(spa)) {
4122 /*
4123 * Update the config cache to include the newly-imported pool.
4124 */
4125 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4126 }
4127
4128 /*
4129 * It's possible that the pool was expanded while it was exported.
4130 * We kick off an async task to handle this for us.
4131 */
4132 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4133
4134 mutex_exit(&spa_namespace_lock);
4135 spa_history_log_version(spa, LOG_POOL_IMPORT);
4136
4137#ifdef __FreeBSD__
4138#ifdef _KERNEL
4139 zvol_create_minors(pool);
4140#endif
4141#endif
4142 return (0);
4143}
4144
4145nvlist_t *
4146spa_tryimport(nvlist_t *tryconfig)
4147{
4148 nvlist_t *config = NULL;
4149 char *poolname;
4150 spa_t *spa;
4151 uint64_t state;
4152 int error;
4153
4154 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4155 return (NULL);
4156
4157 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4158 return (NULL);
4159
4160 /*
4161 * Create and initialize the spa structure.
4162 */
4163 mutex_enter(&spa_namespace_lock);
4164 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4165 spa_activate(spa, FREAD);
4166
4167 /*
4168 * Pass off the heavy lifting to spa_load().
4169 * Pass TRUE for mosconfig because the user-supplied config
4170 * is actually the one to trust when doing an import.
4171 */
4172 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4173
4174 /*
4175 * If 'tryconfig' was at least parsable, return the current config.
4176 */
4177 if (spa->spa_root_vdev != NULL) {
4178 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4179 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4180 poolname) == 0);
4181 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4182 state) == 0);
4183 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4184 spa->spa_uberblock.ub_timestamp) == 0);
4185 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4186 spa->spa_load_info) == 0);
4187
4188 /*
4189 * If the bootfs property exists on this pool then we
4190 * copy it out so that external consumers can tell which
4191 * pools are bootable.
4192 */
4193 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4194 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4195
4196 /*
4197 * We have to play games with the name since the
4198 * pool was opened as TRYIMPORT_NAME.
4199 */
4200 if (dsl_dsobj_to_dsname(spa_name(spa),
4201 spa->spa_bootfs, tmpname) == 0) {
4202 char *cp;
4203 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4204
4205 cp = strchr(tmpname, '/');
4206 if (cp == NULL) {
4207 (void) strlcpy(dsname, tmpname,
4208 MAXPATHLEN);
4209 } else {
4210 (void) snprintf(dsname, MAXPATHLEN,
4211 "%s/%s", poolname, ++cp);
4212 }
4213 VERIFY(nvlist_add_string(config,
4214 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4215 kmem_free(dsname, MAXPATHLEN);
4216 }
4217 kmem_free(tmpname, MAXPATHLEN);
4218 }
4219
4220 /*
4221 * Add the list of hot spares and level 2 cache devices.
4222 */
4223 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4224 spa_add_spares(spa, config);
4225 spa_add_l2cache(spa, config);
4226 spa_config_exit(spa, SCL_CONFIG, FTAG);
4227 }
4228
4229 spa_unload(spa);
4230 spa_deactivate(spa);
4231 spa_remove(spa);
4232 mutex_exit(&spa_namespace_lock);
4233
4234 return (config);
4235}
4236
4237/*
4238 * Pool export/destroy
4239 *
4240 * The act of destroying or exporting a pool is very simple. We make sure there
4241 * is no more pending I/O and any references to the pool are gone. Then, we
4242 * update the pool state and sync all the labels to disk, removing the
4243 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4244 * we don't sync the labels or remove the configuration cache.
4245 */
4246static int
4247spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4248 boolean_t force, boolean_t hardforce)
4249{
4250 spa_t *spa;
4251
4252 if (oldconfig)
4253 *oldconfig = NULL;
4254
4255 if (!(spa_mode_global & FWRITE))
4256 return (EROFS);
4257
4258 mutex_enter(&spa_namespace_lock);
4259 if ((spa = spa_lookup(pool)) == NULL) {
4260 mutex_exit(&spa_namespace_lock);
4261 return (ENOENT);
4262 }
4263
4264 /*
4265 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4266 * reacquire the namespace lock, and see if we can export.
4267 */
4268 spa_open_ref(spa, FTAG);
4269 mutex_exit(&spa_namespace_lock);
4270 spa_async_suspend(spa);
4271 mutex_enter(&spa_namespace_lock);
4272 spa_close(spa, FTAG);
4273
4274 /*
4275 * The pool will be in core if it's openable,
4276 * in which case we can modify its state.
4277 */
4278 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4279 /*
4280 * Objsets may be open only because they're dirty, so we
4281 * have to force it to sync before checking spa_refcnt.
4282 */
4283 txg_wait_synced(spa->spa_dsl_pool, 0);
4284
4285 /*
4286 * A pool cannot be exported or destroyed if there are active
4287 * references. If we are resetting a pool, allow references by
4288 * fault injection handlers.
4289 */
4290 if (!spa_refcount_zero(spa) ||
4291 (spa->spa_inject_ref != 0 &&
4292 new_state != POOL_STATE_UNINITIALIZED)) {
4293 spa_async_resume(spa);
4294 mutex_exit(&spa_namespace_lock);
4295 return (EBUSY);
4296 }
4297
4298 /*
4299 * A pool cannot be exported if it has an active shared spare.
4300 * This is to prevent other pools stealing the active spare
4301 * from an exported pool. At user's own will, such pool can
4302 * be forcedly exported.
4303 */
4304 if (!force && new_state == POOL_STATE_EXPORTED &&
4305 spa_has_active_shared_spare(spa)) {
4306 spa_async_resume(spa);
4307 mutex_exit(&spa_namespace_lock);
4308 return (EXDEV);
4309 }
4310
4311 /*
4312 * We want this to be reflected on every label,
4313 * so mark them all dirty. spa_unload() will do the
4314 * final sync that pushes these changes out.
4315 */
4316 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4317 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4318 spa->spa_state = new_state;
4319 spa->spa_final_txg = spa_last_synced_txg(spa) +
4320 TXG_DEFER_SIZE + 1;
4321 vdev_config_dirty(spa->spa_root_vdev);
4322 spa_config_exit(spa, SCL_ALL, FTAG);
4323 }
4324 }
4325
4326 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4327
4328 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4329 spa_unload(spa);
4330 spa_deactivate(spa);
4331 }
4332
4333 if (oldconfig && spa->spa_config)
4334 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4335
4336 if (new_state != POOL_STATE_UNINITIALIZED) {
4337 if (!hardforce)
4338 spa_config_sync(spa, B_TRUE, B_TRUE);
4339 spa_remove(spa);
4340 }
4341 mutex_exit(&spa_namespace_lock);
4342
4343 return (0);
4344}
4345
4346/*
4347 * Destroy a storage pool.
4348 */
4349int
4350spa_destroy(char *pool)
4351{
4352 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4353 B_FALSE, B_FALSE));
4354}
4355
4356/*
4357 * Export a storage pool.
4358 */
4359int
4360spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4361 boolean_t hardforce)
4362{
4363 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4364 force, hardforce));
4365}
4366
4367/*
4368 * Similar to spa_export(), this unloads the spa_t without actually removing it
4369 * from the namespace in any way.
4370 */
4371int
4372spa_reset(char *pool)
4373{
4374 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4375 B_FALSE, B_FALSE));
4376}
4377
4378/*
4379 * ==========================================================================
4380 * Device manipulation
4381 * ==========================================================================
4382 */
4383
4384/*
4385 * Add a device to a storage pool.
4386 */
4387int
4388spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4389{
4390 uint64_t txg, id;
4391 int error;
4392 vdev_t *rvd = spa->spa_root_vdev;
4393 vdev_t *vd, *tvd;
4394 nvlist_t **spares, **l2cache;
4395 uint_t nspares, nl2cache;
4396
4397 ASSERT(spa_writeable(spa));
4398
4399 txg = spa_vdev_enter(spa);
4400
4401 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4402 VDEV_ALLOC_ADD)) != 0)
4403 return (spa_vdev_exit(spa, NULL, txg, error));
4404
4405 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4406
4407 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4408 &nspares) != 0)
4409 nspares = 0;
4410
4411 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4412 &nl2cache) != 0)
4413 nl2cache = 0;
4414
4415 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4416 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4417
4418 if (vd->vdev_children != 0 &&
4419 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4420 return (spa_vdev_exit(spa, vd, txg, error));
4421
4422 /*
4423 * We must validate the spares and l2cache devices after checking the
4424 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4425 */
4426 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4427 return (spa_vdev_exit(spa, vd, txg, error));
4428
4429 /*
4430 * Transfer each new top-level vdev from vd to rvd.
4431 */
4432 for (int c = 0; c < vd->vdev_children; c++) {
4433
4434 /*
4435 * Set the vdev id to the first hole, if one exists.
4436 */
4437 for (id = 0; id < rvd->vdev_children; id++) {
4438 if (rvd->vdev_child[id]->vdev_ishole) {
4439 vdev_free(rvd->vdev_child[id]);
4440 break;
4441 }
4442 }
4443 tvd = vd->vdev_child[c];
4444 vdev_remove_child(vd, tvd);
4445 tvd->vdev_id = id;
4446 vdev_add_child(rvd, tvd);
4447 vdev_config_dirty(tvd);
4448 }
4449
4450 if (nspares != 0) {
4451 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4452 ZPOOL_CONFIG_SPARES);
4453 spa_load_spares(spa);
4454 spa->spa_spares.sav_sync = B_TRUE;
4455 }
4456
4457 if (nl2cache != 0) {
4458 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4459 ZPOOL_CONFIG_L2CACHE);
4460 spa_load_l2cache(spa);
4461 spa->spa_l2cache.sav_sync = B_TRUE;
4462 }
4463
4464 /*
4465 * We have to be careful when adding new vdevs to an existing pool.
4466 * If other threads start allocating from these vdevs before we
4467 * sync the config cache, and we lose power, then upon reboot we may
4468 * fail to open the pool because there are DVAs that the config cache
4469 * can't translate. Therefore, we first add the vdevs without
4470 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4471 * and then let spa_config_update() initialize the new metaslabs.
4472 *
4473 * spa_load() checks for added-but-not-initialized vdevs, so that
4474 * if we lose power at any point in this sequence, the remaining
4475 * steps will be completed the next time we load the pool.
4476 */
4477 (void) spa_vdev_exit(spa, vd, txg, 0);
4478
4479 mutex_enter(&spa_namespace_lock);
4480 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4481 mutex_exit(&spa_namespace_lock);
4482
4483 return (0);
4484}
4485
4486/*
4487 * Attach a device to a mirror. The arguments are the path to any device
4488 * in the mirror, and the nvroot for the new device. If the path specifies
4489 * a device that is not mirrored, we automatically insert the mirror vdev.
4490 *
4491 * If 'replacing' is specified, the new device is intended to replace the
4492 * existing device; in this case the two devices are made into their own
4493 * mirror using the 'replacing' vdev, which is functionally identical to
4494 * the mirror vdev (it actually reuses all the same ops) but has a few
4495 * extra rules: you can't attach to it after it's been created, and upon
4496 * completion of resilvering, the first disk (the one being replaced)
4497 * is automatically detached.
4498 */
4499int
4500spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4501{
4502 uint64_t txg, dtl_max_txg;
4503 vdev_t *rvd = spa->spa_root_vdev;
4504 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4505 vdev_ops_t *pvops;
4506 char *oldvdpath, *newvdpath;
4507 int newvd_isspare;
4508 int error;
4509
4510 ASSERT(spa_writeable(spa));
4511
4512 txg = spa_vdev_enter(spa);
4513
4514 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4515
4516 if (oldvd == NULL)
4517 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4518
4519 if (!oldvd->vdev_ops->vdev_op_leaf)
4520 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4521
4522 pvd = oldvd->vdev_parent;
4523
4524 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4525 VDEV_ALLOC_ATTACH)) != 0)
4526 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4527
4528 if (newrootvd->vdev_children != 1)
4529 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4530
4531 newvd = newrootvd->vdev_child[0];
4532
4533 if (!newvd->vdev_ops->vdev_op_leaf)
4534 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4535
4536 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4537 return (spa_vdev_exit(spa, newrootvd, txg, error));
4538
4539 /*
4540 * Spares can't replace logs
4541 */
4542 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4543 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4544
4545 if (!replacing) {
4546 /*
4547 * For attach, the only allowable parent is a mirror or the root
4548 * vdev.
4549 */
4550 if (pvd->vdev_ops != &vdev_mirror_ops &&
4551 pvd->vdev_ops != &vdev_root_ops)
4552 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4553
4554 pvops = &vdev_mirror_ops;
4555 } else {
4556 /*
4557 * Active hot spares can only be replaced by inactive hot
4558 * spares.
4559 */
4560 if (pvd->vdev_ops == &vdev_spare_ops &&
4561 oldvd->vdev_isspare &&
4562 !spa_has_spare(spa, newvd->vdev_guid))
4563 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4564
4565 /*
4566 * If the source is a hot spare, and the parent isn't already a
4567 * spare, then we want to create a new hot spare. Otherwise, we
4568 * want to create a replacing vdev. The user is not allowed to
4569 * attach to a spared vdev child unless the 'isspare' state is
4570 * the same (spare replaces spare, non-spare replaces
4571 * non-spare).
4572 */
4573 if (pvd->vdev_ops == &vdev_replacing_ops &&
4574 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4575 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4576 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4577 newvd->vdev_isspare != oldvd->vdev_isspare) {
4578 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4579 }
4580
4581 if (newvd->vdev_isspare)
4582 pvops = &vdev_spare_ops;
4583 else
4584 pvops = &vdev_replacing_ops;
4585 }
4586
4587 /*
4588 * Make sure the new device is big enough.
4589 */
4590 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4591 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4592
4593 /*
4594 * The new device cannot have a higher alignment requirement
4595 * than the top-level vdev.
4596 */
4597 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4598 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4599
4600 /*
4601 * If this is an in-place replacement, update oldvd's path and devid
4602 * to make it distinguishable from newvd, and unopenable from now on.
4603 */
4604 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4605 spa_strfree(oldvd->vdev_path);
4606 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4607 KM_SLEEP);
4608 (void) sprintf(oldvd->vdev_path, "%s/%s",
4609 newvd->vdev_path, "old");
4610 if (oldvd->vdev_devid != NULL) {
4611 spa_strfree(oldvd->vdev_devid);
4612 oldvd->vdev_devid = NULL;
4613 }
4614 }
4615
4616 /* mark the device being resilvered */
4617 newvd->vdev_resilvering = B_TRUE;
4618
4619 /*
4620 * If the parent is not a mirror, or if we're replacing, insert the new
4621 * mirror/replacing/spare vdev above oldvd.
4622 */
4623 if (pvd->vdev_ops != pvops)
4624 pvd = vdev_add_parent(oldvd, pvops);
4625
4626 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4627 ASSERT(pvd->vdev_ops == pvops);
4628 ASSERT(oldvd->vdev_parent == pvd);
4629
4630 /*
4631 * Extract the new device from its root and add it to pvd.
4632 */
4633 vdev_remove_child(newrootvd, newvd);
4634 newvd->vdev_id = pvd->vdev_children;
4635 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4636 vdev_add_child(pvd, newvd);
4637
4638 tvd = newvd->vdev_top;
4639 ASSERT(pvd->vdev_top == tvd);
4640 ASSERT(tvd->vdev_parent == rvd);
4641
4642 vdev_config_dirty(tvd);
4643
4644 /*
4645 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4646 * for any dmu_sync-ed blocks. It will propagate upward when
4647 * spa_vdev_exit() calls vdev_dtl_reassess().
4648 */
4649 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4650
4651 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4652 dtl_max_txg - TXG_INITIAL);
4653
4654 if (newvd->vdev_isspare) {
4655 spa_spare_activate(newvd);
4656 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4657 }
4658
4659 oldvdpath = spa_strdup(oldvd->vdev_path);
4660 newvdpath = spa_strdup(newvd->vdev_path);
4661 newvd_isspare = newvd->vdev_isspare;
4662
4663 /*
4664 * Mark newvd's DTL dirty in this txg.
4665 */
4666 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4667
4668 /*
4669 * Restart the resilver
4670 */
4671 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4672
4673 /*
4674 * Commit the config
4675 */
4676 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4677
4678 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4679 "%s vdev=%s %s vdev=%s",
4680 replacing && newvd_isspare ? "spare in" :
4681 replacing ? "replace" : "attach", newvdpath,
4682 replacing ? "for" : "to", oldvdpath);
4683
4684 spa_strfree(oldvdpath);
4685 spa_strfree(newvdpath);
4686
4687 if (spa->spa_bootfs)
4688 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4689
4690 return (0);
4691}
4692
4693/*
4694 * Detach a device from a mirror or replacing vdev.
4695 * If 'replace_done' is specified, only detach if the parent
4696 * is a replacing vdev.
4697 */
4698int
4699spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4700{
4701 uint64_t txg;
4702 int error;
4703 vdev_t *rvd = spa->spa_root_vdev;
4704 vdev_t *vd, *pvd, *cvd, *tvd;
4705 boolean_t unspare = B_FALSE;
4706 uint64_t unspare_guid;
4707 char *vdpath;
4708
4709 ASSERT(spa_writeable(spa));
4710
4711 txg = spa_vdev_enter(spa);
4712
4713 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4714
4715 if (vd == NULL)
4716 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4717
4718 if (!vd->vdev_ops->vdev_op_leaf)
4719 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4720
4721 pvd = vd->vdev_parent;
4722
4723 /*
4724 * If the parent/child relationship is not as expected, don't do it.
4725 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4726 * vdev that's replacing B with C. The user's intent in replacing
4727 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4728 * the replace by detaching C, the expected behavior is to end up
4729 * M(A,B). But suppose that right after deciding to detach C,
4730 * the replacement of B completes. We would have M(A,C), and then
4731 * ask to detach C, which would leave us with just A -- not what
4732 * the user wanted. To prevent this, we make sure that the
4733 * parent/child relationship hasn't changed -- in this example,
4734 * that C's parent is still the replacing vdev R.
4735 */
4736 if (pvd->vdev_guid != pguid && pguid != 0)
4737 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4738
4739 /*
4740 * Only 'replacing' or 'spare' vdevs can be replaced.
4741 */
4742 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4743 pvd->vdev_ops != &vdev_spare_ops)
4744 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4745
4746 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4747 spa_version(spa) >= SPA_VERSION_SPARES);
4748
4749 /*
4750 * Only mirror, replacing, and spare vdevs support detach.
4751 */
4752 if (pvd->vdev_ops != &vdev_replacing_ops &&
4753 pvd->vdev_ops != &vdev_mirror_ops &&
4754 pvd->vdev_ops != &vdev_spare_ops)
4755 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4756
4757 /*
4758 * If this device has the only valid copy of some data,
4759 * we cannot safely detach it.
4760 */
4761 if (vdev_dtl_required(vd))
4762 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4763
4764 ASSERT(pvd->vdev_children >= 2);
4765
4766 /*
4767 * If we are detaching the second disk from a replacing vdev, then
4768 * check to see if we changed the original vdev's path to have "/old"
4769 * at the end in spa_vdev_attach(). If so, undo that change now.
4770 */
4771 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4772 vd->vdev_path != NULL) {
4773 size_t len = strlen(vd->vdev_path);
4774
4775 for (int c = 0; c < pvd->vdev_children; c++) {
4776 cvd = pvd->vdev_child[c];
4777
4778 if (cvd == vd || cvd->vdev_path == NULL)
4779 continue;
4780
4781 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4782 strcmp(cvd->vdev_path + len, "/old") == 0) {
4783 spa_strfree(cvd->vdev_path);
4784 cvd->vdev_path = spa_strdup(vd->vdev_path);
4785 break;
4786 }
4787 }
4788 }
4789
4790 /*
4791 * If we are detaching the original disk from a spare, then it implies
4792 * that the spare should become a real disk, and be removed from the
4793 * active spare list for the pool.
4794 */
4795 if (pvd->vdev_ops == &vdev_spare_ops &&
4796 vd->vdev_id == 0 &&
4797 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4798 unspare = B_TRUE;
4799
4800 /*
4801 * Erase the disk labels so the disk can be used for other things.
4802 * This must be done after all other error cases are handled,
4803 * but before we disembowel vd (so we can still do I/O to it).
4804 * But if we can't do it, don't treat the error as fatal --
4805 * it may be that the unwritability of the disk is the reason
4806 * it's being detached!
4807 */
4808 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4809
4810 /*
4811 * Remove vd from its parent and compact the parent's children.
4812 */
4813 vdev_remove_child(pvd, vd);
4814 vdev_compact_children(pvd);
4815
4816 /*
4817 * Remember one of the remaining children so we can get tvd below.
4818 */
4819 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4820
4821 /*
4822 * If we need to remove the remaining child from the list of hot spares,
4823 * do it now, marking the vdev as no longer a spare in the process.
4824 * We must do this before vdev_remove_parent(), because that can
4825 * change the GUID if it creates a new toplevel GUID. For a similar
4826 * reason, we must remove the spare now, in the same txg as the detach;
4827 * otherwise someone could attach a new sibling, change the GUID, and
4828 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4829 */
4830 if (unspare) {
4831 ASSERT(cvd->vdev_isspare);
4832 spa_spare_remove(cvd);
4833 unspare_guid = cvd->vdev_guid;
4834 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4835 cvd->vdev_unspare = B_TRUE;
4836 }
4837
4838 /*
4839 * If the parent mirror/replacing vdev only has one child,
4840 * the parent is no longer needed. Remove it from the tree.
4841 */
4842 if (pvd->vdev_children == 1) {
4843 if (pvd->vdev_ops == &vdev_spare_ops)
4844 cvd->vdev_unspare = B_FALSE;
4845 vdev_remove_parent(cvd);
4846 cvd->vdev_resilvering = B_FALSE;
4847 }
4848
4849
4850 /*
4851 * We don't set tvd until now because the parent we just removed
4852 * may have been the previous top-level vdev.
4853 */
4854 tvd = cvd->vdev_top;
4855 ASSERT(tvd->vdev_parent == rvd);
4856
4857 /*
4858 * Reevaluate the parent vdev state.
4859 */
4860 vdev_propagate_state(cvd);
4861
4862 /*
4863 * If the 'autoexpand' property is set on the pool then automatically
4864 * try to expand the size of the pool. For example if the device we
4865 * just detached was smaller than the others, it may be possible to
4866 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4867 * first so that we can obtain the updated sizes of the leaf vdevs.
4868 */
4869 if (spa->spa_autoexpand) {
4870 vdev_reopen(tvd);
4871 vdev_expand(tvd, txg);
4872 }
4873
4874 vdev_config_dirty(tvd);
4875
4876 /*
4877 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4878 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4879 * But first make sure we're not on any *other* txg's DTL list, to
4880 * prevent vd from being accessed after it's freed.
4881 */
4882 vdpath = spa_strdup(vd->vdev_path);
4883 for (int t = 0; t < TXG_SIZE; t++)
4884 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4885 vd->vdev_detached = B_TRUE;
4886 vdev_dirty(tvd, VDD_DTL, vd, txg);
4887
4888 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4889
4890 /* hang on to the spa before we release the lock */
4891 spa_open_ref(spa, FTAG);
4892
4893 error = spa_vdev_exit(spa, vd, txg, 0);
4894
4895 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4896 "vdev=%s", vdpath);
4897 spa_strfree(vdpath);
4898
4899 /*
4900 * If this was the removal of the original device in a hot spare vdev,
4901 * then we want to go through and remove the device from the hot spare
4902 * list of every other pool.
4903 */
4904 if (unspare) {
4905 spa_t *altspa = NULL;
4906
4907 mutex_enter(&spa_namespace_lock);
4908 while ((altspa = spa_next(altspa)) != NULL) {
4909 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4910 altspa == spa)
4911 continue;
4912
4913 spa_open_ref(altspa, FTAG);
4914 mutex_exit(&spa_namespace_lock);
4915 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4916 mutex_enter(&spa_namespace_lock);
4917 spa_close(altspa, FTAG);
4918 }
4919 mutex_exit(&spa_namespace_lock);
4920
4921 /* search the rest of the vdevs for spares to remove */
4922 spa_vdev_resilver_done(spa);
4923 }
4924
4925 /* all done with the spa; OK to release */
4926 mutex_enter(&spa_namespace_lock);
4927 spa_close(spa, FTAG);
4928 mutex_exit(&spa_namespace_lock);
4929
4930 return (error);
4931}
4932
4933/*
4934 * Split a set of devices from their mirrors, and create a new pool from them.
4935 */
4936int
4937spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4938 nvlist_t *props, boolean_t exp)
4939{
4940 int error = 0;
4941 uint64_t txg, *glist;
4942 spa_t *newspa;
4943 uint_t c, children, lastlog;
4944 nvlist_t **child, *nvl, *tmp;
4945 dmu_tx_t *tx;
4946 char *altroot = NULL;
4947 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4948 boolean_t activate_slog;
4949
4950 ASSERT(spa_writeable(spa));
4951
4952 txg = spa_vdev_enter(spa);
4953
4954 /* clear the log and flush everything up to now */
4955 activate_slog = spa_passivate_log(spa);
4956 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4957 error = spa_offline_log(spa);
4958 txg = spa_vdev_config_enter(spa);
4959
4960 if (activate_slog)
4961 spa_activate_log(spa);
4962
4963 if (error != 0)
4964 return (spa_vdev_exit(spa, NULL, txg, error));
4965
4966 /* check new spa name before going any further */
4967 if (spa_lookup(newname) != NULL)
4968 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4969
4970 /*
4971 * scan through all the children to ensure they're all mirrors
4972 */
4973 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4974 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4975 &children) != 0)
4976 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4977
4978 /* first, check to ensure we've got the right child count */
4979 rvd = spa->spa_root_vdev;
4980 lastlog = 0;
4981 for (c = 0; c < rvd->vdev_children; c++) {
4982 vdev_t *vd = rvd->vdev_child[c];
4983
4984 /* don't count the holes & logs as children */
4985 if (vd->vdev_islog || vd->vdev_ishole) {
4986 if (lastlog == 0)
4987 lastlog = c;
4988 continue;
4989 }
4990
4991 lastlog = 0;
4992 }
4993 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4994 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4995
4996 /* next, ensure no spare or cache devices are part of the split */
4997 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4998 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4999 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5000
5001 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5002 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5003
5004 /* then, loop over each vdev and validate it */
5005 for (c = 0; c < children; c++) {
5006 uint64_t is_hole = 0;
5007
5008 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5009 &is_hole);
5010
5011 if (is_hole != 0) {
5012 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5013 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5014 continue;
5015 } else {
5016 error = EINVAL;
5017 break;
5018 }
5019 }
5020
5021 /* which disk is going to be split? */
5022 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5023 &glist[c]) != 0) {
5024 error = EINVAL;
5025 break;
5026 }
5027
5028 /* look it up in the spa */
5029 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5030 if (vml[c] == NULL) {
5031 error = ENODEV;
5032 break;
5033 }
5034
5035 /* make sure there's nothing stopping the split */
5036 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5037 vml[c]->vdev_islog ||
5038 vml[c]->vdev_ishole ||
5039 vml[c]->vdev_isspare ||
5040 vml[c]->vdev_isl2cache ||
5041 !vdev_writeable(vml[c]) ||
5042 vml[c]->vdev_children != 0 ||
5043 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5044 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5045 error = EINVAL;
5046 break;
5047 }
5048
5049 if (vdev_dtl_required(vml[c])) {
5050 error = EBUSY;
5051 break;
5052 }
5053
5054 /* we need certain info from the top level */
5055 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5056 vml[c]->vdev_top->vdev_ms_array) == 0);
5057 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5058 vml[c]->vdev_top->vdev_ms_shift) == 0);
5059 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5060 vml[c]->vdev_top->vdev_asize) == 0);
5061 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5062 vml[c]->vdev_top->vdev_ashift) == 0);
5063 }
5064
5065 if (error != 0) {
5066 kmem_free(vml, children * sizeof (vdev_t *));
5067 kmem_free(glist, children * sizeof (uint64_t));
5068 return (spa_vdev_exit(spa, NULL, txg, error));
5069 }
5070
5071 /* stop writers from using the disks */
5072 for (c = 0; c < children; c++) {
5073 if (vml[c] != NULL)
5074 vml[c]->vdev_offline = B_TRUE;
5075 }
5076 vdev_reopen(spa->spa_root_vdev);
5077
5078 /*
5079 * Temporarily record the splitting vdevs in the spa config. This
5080 * will disappear once the config is regenerated.
5081 */
5082 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5083 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5084 glist, children) == 0);
5085 kmem_free(glist, children * sizeof (uint64_t));
5086
5087 mutex_enter(&spa->spa_props_lock);
5088 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5089 nvl) == 0);
5090 mutex_exit(&spa->spa_props_lock);
5091 spa->spa_config_splitting = nvl;
5092 vdev_config_dirty(spa->spa_root_vdev);
5093
5094 /* configure and create the new pool */
5095 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5096 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5097 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5098 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5099 spa_version(spa)) == 0);
5100 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5101 spa->spa_config_txg) == 0);
5102 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5103 spa_generate_guid(NULL)) == 0);
5104 (void) nvlist_lookup_string(props,
5105 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5106
5107 /* add the new pool to the namespace */
5108 newspa = spa_add(newname, config, altroot);
5109 newspa->spa_config_txg = spa->spa_config_txg;
5110 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5111
5112 /* release the spa config lock, retaining the namespace lock */
5113 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5114
5115 if (zio_injection_enabled)
5116 zio_handle_panic_injection(spa, FTAG, 1);
5117
5118 spa_activate(newspa, spa_mode_global);
5119 spa_async_suspend(newspa);
5120
5121#ifndef sun
5122 /* mark that we are creating new spa by splitting */
5123 newspa->spa_splitting_newspa = B_TRUE;
5124#endif
5125 /* create the new pool from the disks of the original pool */
5126 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5127#ifndef sun
5128 newspa->spa_splitting_newspa = B_FALSE;
5129#endif
5130 if (error)
5131 goto out;
5132
5133 /* if that worked, generate a real config for the new pool */
5134 if (newspa->spa_root_vdev != NULL) {
5135 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5136 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5137 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5138 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5139 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5140 B_TRUE));
5141 }
5142
5143 /* set the props */
5144 if (props != NULL) {
5145 spa_configfile_set(newspa, props, B_FALSE);
5146 error = spa_prop_set(newspa, props);
5147 if (error)
5148 goto out;
5149 }
5150
5151 /* flush everything */
5152 txg = spa_vdev_config_enter(newspa);
5153 vdev_config_dirty(newspa->spa_root_vdev);
5154 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5155
5156 if (zio_injection_enabled)
5157 zio_handle_panic_injection(spa, FTAG, 2);
5158
5159 spa_async_resume(newspa);
5160
5161 /* finally, update the original pool's config */
5162 txg = spa_vdev_config_enter(spa);
5163 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5164 error = dmu_tx_assign(tx, TXG_WAIT);
5165 if (error != 0)
5166 dmu_tx_abort(tx);
5167 for (c = 0; c < children; c++) {
5168 if (vml[c] != NULL) {
5169 vdev_split(vml[c]);
5170 if (error == 0)
5171 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5172 spa, tx, "vdev=%s",
5173 vml[c]->vdev_path);
5174 vdev_free(vml[c]);
5175 }
5176 }
5177 vdev_config_dirty(spa->spa_root_vdev);
5178 spa->spa_config_splitting = NULL;
5179 nvlist_free(nvl);
5180 if (error == 0)
5181 dmu_tx_commit(tx);
5182 (void) spa_vdev_exit(spa, NULL, txg, 0);
5183
5184 if (zio_injection_enabled)
5185 zio_handle_panic_injection(spa, FTAG, 3);
5186
5187 /* split is complete; log a history record */
5188 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5189 "split new pool %s from pool %s", newname, spa_name(spa));
5190
5191 kmem_free(vml, children * sizeof (vdev_t *));
5192
5193 /* if we're not going to mount the filesystems in userland, export */
5194 if (exp)
5195 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5196 B_FALSE, B_FALSE);
5197
5198 return (error);
5199
5200out:
5201 spa_unload(newspa);
5202 spa_deactivate(newspa);
5203 spa_remove(newspa);
5204
5205 txg = spa_vdev_config_enter(spa);
5206
5207 /* re-online all offlined disks */
5208 for (c = 0; c < children; c++) {
5209 if (vml[c] != NULL)
5210 vml[c]->vdev_offline = B_FALSE;
5211 }
5212 vdev_reopen(spa->spa_root_vdev);
5213
5214 nvlist_free(spa->spa_config_splitting);
5215 spa->spa_config_splitting = NULL;
5216 (void) spa_vdev_exit(spa, NULL, txg, error);
5217
5218 kmem_free(vml, children * sizeof (vdev_t *));
5219 return (error);
5220}
5221
5222static nvlist_t *
5223spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5224{
5225 for (int i = 0; i < count; i++) {
5226 uint64_t guid;
5227
5228 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5229 &guid) == 0);
5230
5231 if (guid == target_guid)
5232 return (nvpp[i]);
5233 }
5234
5235 return (NULL);
5236}
5237
5238static void
5239spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5240 nvlist_t *dev_to_remove)
5241{
5242 nvlist_t **newdev = NULL;
5243
5244 if (count > 1)
5245 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5246
5247 for (int i = 0, j = 0; i < count; i++) {
5248 if (dev[i] == dev_to_remove)
5249 continue;
5250 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5251 }
5252
5253 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5254 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5255
5256 for (int i = 0; i < count - 1; i++)
5257 nvlist_free(newdev[i]);
5258
5259 if (count > 1)
5260 kmem_free(newdev, (count - 1) * sizeof (void *));
5261}
5262
5263/*
5264 * Evacuate the device.
5265 */
5266static int
5267spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5268{
5269 uint64_t txg;
5270 int error = 0;
5271
5272 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5273 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5274 ASSERT(vd == vd->vdev_top);
5275
5276 /*
5277 * Evacuate the device. We don't hold the config lock as writer
5278 * since we need to do I/O but we do keep the
5279 * spa_namespace_lock held. Once this completes the device
5280 * should no longer have any blocks allocated on it.
5281 */
5282 if (vd->vdev_islog) {
5283 if (vd->vdev_stat.vs_alloc != 0)
5284 error = spa_offline_log(spa);
5285 } else {
5286 error = ENOTSUP;
5287 }
5288
5289 if (error)
5290 return (error);
5291
5292 /*
5293 * The evacuation succeeded. Remove any remaining MOS metadata
5294 * associated with this vdev, and wait for these changes to sync.
5295 */
5296 ASSERT0(vd->vdev_stat.vs_alloc);
5297 txg = spa_vdev_config_enter(spa);
5298 vd->vdev_removing = B_TRUE;
5299 vdev_dirty(vd, 0, NULL, txg);
5300 vdev_config_dirty(vd);
5301 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5302
5303 return (0);
5304}
5305
5306/*
5307 * Complete the removal by cleaning up the namespace.
5308 */
5309static void
5310spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5311{
5312 vdev_t *rvd = spa->spa_root_vdev;
5313 uint64_t id = vd->vdev_id;
5314 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5315
5316 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5317 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5318 ASSERT(vd == vd->vdev_top);
5319
5320 /*
5321 * Only remove any devices which are empty.
5322 */
5323 if (vd->vdev_stat.vs_alloc != 0)
5324 return;
5325
5326 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5327
5328 if (list_link_active(&vd->vdev_state_dirty_node))
5329 vdev_state_clean(vd);
5330 if (list_link_active(&vd->vdev_config_dirty_node))
5331 vdev_config_clean(vd);
5332
5333 vdev_free(vd);
5334
5335 if (last_vdev) {
5336 vdev_compact_children(rvd);
5337 } else {
5338 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5339 vdev_add_child(rvd, vd);
5340 }
5341 vdev_config_dirty(rvd);
5342
5343 /*
5344 * Reassess the health of our root vdev.
5345 */
5346 vdev_reopen(rvd);
5347}
5348
5349/*
5350 * Remove a device from the pool -
5351 *
5352 * Removing a device from the vdev namespace requires several steps
5353 * and can take a significant amount of time. As a result we use
5354 * the spa_vdev_config_[enter/exit] functions which allow us to
5355 * grab and release the spa_config_lock while still holding the namespace
5356 * lock. During each step the configuration is synced out.
5357 */
5358
5359/*
5360 * Remove a device from the pool. Currently, this supports removing only hot
5361 * spares, slogs, and level 2 ARC devices.
5362 */
5363int
5364spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5365{
5366 vdev_t *vd;
5367 metaslab_group_t *mg;
5368 nvlist_t **spares, **l2cache, *nv;
5369 uint64_t txg = 0;
5370 uint_t nspares, nl2cache;
5371 int error = 0;
5372 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5373
5374 ASSERT(spa_writeable(spa));
5375
5376 if (!locked)
5377 txg = spa_vdev_enter(spa);
5378
5379 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5380
5381 if (spa->spa_spares.sav_vdevs != NULL &&
5382 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5383 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5384 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5385 /*
5386 * Only remove the hot spare if it's not currently in use
5387 * in this pool.
5388 */
5389 if (vd == NULL || unspare) {
5390 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5391 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5392 spa_load_spares(spa);
5393 spa->spa_spares.sav_sync = B_TRUE;
5394 } else {
5395 error = EBUSY;
5396 }
5397 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5398 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5399 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5400 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5401 /*
5402 * Cache devices can always be removed.
5403 */
5404 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5405 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5406 spa_load_l2cache(spa);
5407 spa->spa_l2cache.sav_sync = B_TRUE;
5408 } else if (vd != NULL && vd->vdev_islog) {
5409 ASSERT(!locked);
5410 ASSERT(vd == vd->vdev_top);
5411
5412 /*
5413 * XXX - Once we have bp-rewrite this should
5414 * become the common case.
5415 */
5416
5417 mg = vd->vdev_mg;
5418
5419 /*
5420 * Stop allocating from this vdev.
5421 */
5422 metaslab_group_passivate(mg);
5423
5424 /*
5425 * Wait for the youngest allocations and frees to sync,
5426 * and then wait for the deferral of those frees to finish.
5427 */
5428 spa_vdev_config_exit(spa, NULL,
5429 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5430
5431 /*
5432 * Attempt to evacuate the vdev.
5433 */
5434 error = spa_vdev_remove_evacuate(spa, vd);
5435
5436 txg = spa_vdev_config_enter(spa);
5437
5438 /*
5439 * If we couldn't evacuate the vdev, unwind.
5440 */
5441 if (error) {
5442 metaslab_group_activate(mg);
5443 return (spa_vdev_exit(spa, NULL, txg, error));
5444 }
5445
5446 /*
5447 * Clean up the vdev namespace.
5448 */
5449 spa_vdev_remove_from_namespace(spa, vd);
5450
5451 } else if (vd != NULL) {
5452 /*
5453 * Normal vdevs cannot be removed (yet).
5454 */
5455 error = ENOTSUP;
5456 } else {
5457 /*
5458 * There is no vdev of any kind with the specified guid.
5459 */
5460 error = ENOENT;
5461 }
5462
5463 if (!locked)
5464 return (spa_vdev_exit(spa, NULL, txg, error));
5465
5466 return (error);
5467}
5468
5469/*
5470 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5471 * current spared, so we can detach it.
5472 */
5473static vdev_t *
5474spa_vdev_resilver_done_hunt(vdev_t *vd)
5475{
5476 vdev_t *newvd, *oldvd;
5477
5478 for (int c = 0; c < vd->vdev_children; c++) {
5479 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5480 if (oldvd != NULL)
5481 return (oldvd);
5482 }
5483
5484 /*
5485 * Check for a completed replacement. We always consider the first
5486 * vdev in the list to be the oldest vdev, and the last one to be
5487 * the newest (see spa_vdev_attach() for how that works). In
5488 * the case where the newest vdev is faulted, we will not automatically
5489 * remove it after a resilver completes. This is OK as it will require
5490 * user intervention to determine which disk the admin wishes to keep.
5491 */
5492 if (vd->vdev_ops == &vdev_replacing_ops) {
5493 ASSERT(vd->vdev_children > 1);
5494
5495 newvd = vd->vdev_child[vd->vdev_children - 1];
5496 oldvd = vd->vdev_child[0];
5497
5498 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5499 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5500 !vdev_dtl_required(oldvd))
5501 return (oldvd);
5502 }
5503
5504 /*
5505 * Check for a completed resilver with the 'unspare' flag set.
5506 */
5507 if (vd->vdev_ops == &vdev_spare_ops) {
5508 vdev_t *first = vd->vdev_child[0];
5509 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5510
5511 if (last->vdev_unspare) {
5512 oldvd = first;
5513 newvd = last;
5514 } else if (first->vdev_unspare) {
5515 oldvd = last;
5516 newvd = first;
5517 } else {
5518 oldvd = NULL;
5519 }
5520
5521 if (oldvd != NULL &&
5522 vdev_dtl_empty(newvd, DTL_MISSING) &&
5523 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5524 !vdev_dtl_required(oldvd))
5525 return (oldvd);
5526
5527 /*
5528 * If there are more than two spares attached to a disk,
5529 * and those spares are not required, then we want to
5530 * attempt to free them up now so that they can be used
5531 * by other pools. Once we're back down to a single
5532 * disk+spare, we stop removing them.
5533 */
5534 if (vd->vdev_children > 2) {
5535 newvd = vd->vdev_child[1];
5536
5537 if (newvd->vdev_isspare && last->vdev_isspare &&
5538 vdev_dtl_empty(last, DTL_MISSING) &&
5539 vdev_dtl_empty(last, DTL_OUTAGE) &&
5540 !vdev_dtl_required(newvd))
5541 return (newvd);
5542 }
5543 }
5544
5545 return (NULL);
5546}
5547
5548static void
5549spa_vdev_resilver_done(spa_t *spa)
5550{
5551 vdev_t *vd, *pvd, *ppvd;
5552 uint64_t guid, sguid, pguid, ppguid;
5553
5554 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5555
5556 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5557 pvd = vd->vdev_parent;
5558 ppvd = pvd->vdev_parent;
5559 guid = vd->vdev_guid;
5560 pguid = pvd->vdev_guid;
5561 ppguid = ppvd->vdev_guid;
5562 sguid = 0;
5563 /*
5564 * If we have just finished replacing a hot spared device, then
5565 * we need to detach the parent's first child (the original hot
5566 * spare) as well.
5567 */
5568 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5569 ppvd->vdev_children == 2) {
5570 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5571 sguid = ppvd->vdev_child[1]->vdev_guid;
5572 }
5573 spa_config_exit(spa, SCL_ALL, FTAG);
5574 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5575 return;
5576 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5577 return;
5578 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5579 }
5580
5581 spa_config_exit(spa, SCL_ALL, FTAG);
5582}
5583
5584/*
5585 * Update the stored path or FRU for this vdev.
5586 */
5587int
5588spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5589 boolean_t ispath)
5590{
5591 vdev_t *vd;
5592 boolean_t sync = B_FALSE;
5593
5594 ASSERT(spa_writeable(spa));
5595
5596 spa_vdev_state_enter(spa, SCL_ALL);
5597
5598 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5599 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5600
5601 if (!vd->vdev_ops->vdev_op_leaf)
5602 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5603
5604 if (ispath) {
5605 if (strcmp(value, vd->vdev_path) != 0) {
5606 spa_strfree(vd->vdev_path);
5607 vd->vdev_path = spa_strdup(value);
5608 sync = B_TRUE;
5609 }
5610 } else {
5611 if (vd->vdev_fru == NULL) {
5612 vd->vdev_fru = spa_strdup(value);
5613 sync = B_TRUE;
5614 } else if (strcmp(value, vd->vdev_fru) != 0) {
5615 spa_strfree(vd->vdev_fru);
5616 vd->vdev_fru = spa_strdup(value);
5617 sync = B_TRUE;
5618 }
5619 }
5620
5621 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5622}
5623
5624int
5625spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5626{
5627 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5628}
5629
5630int
5631spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5632{
5633 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5634}
5635
5636/*
5637 * ==========================================================================
5638 * SPA Scanning
5639 * ==========================================================================
5640 */
5641
5642int
5643spa_scan_stop(spa_t *spa)
5644{
5645 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5646 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5647 return (EBUSY);
5648 return (dsl_scan_cancel(spa->spa_dsl_pool));
5649}
5650
5651int
5652spa_scan(spa_t *spa, pool_scan_func_t func)
5653{
5654 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5655
5656 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5657 return (ENOTSUP);
5658
5659 /*
5660 * If a resilver was requested, but there is no DTL on a
5661 * writeable leaf device, we have nothing to do.
5662 */
5663 if (func == POOL_SCAN_RESILVER &&
5664 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5665 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5666 return (0);
5667 }
5668
5669 return (dsl_scan(spa->spa_dsl_pool, func));
5670}
5671
5672/*
5673 * ==========================================================================
5674 * SPA async task processing
5675 * ==========================================================================
5676 */
5677
5678static void
5679spa_async_remove(spa_t *spa, vdev_t *vd)
5680{
5681 if (vd->vdev_remove_wanted) {
5682 vd->vdev_remove_wanted = B_FALSE;
5683 vd->vdev_delayed_close = B_FALSE;
5684 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5685
5686 /*
5687 * We want to clear the stats, but we don't want to do a full
5688 * vdev_clear() as that will cause us to throw away
5689 * degraded/faulted state as well as attempt to reopen the
5690 * device, all of which is a waste.
5691 */
5692 vd->vdev_stat.vs_read_errors = 0;
5693 vd->vdev_stat.vs_write_errors = 0;
5694 vd->vdev_stat.vs_checksum_errors = 0;
5695
5696 vdev_state_dirty(vd->vdev_top);
5697 }
5698
5699 for (int c = 0; c < vd->vdev_children; c++)
5700 spa_async_remove(spa, vd->vdev_child[c]);
5701}
5702
5703static void
5704spa_async_probe(spa_t *spa, vdev_t *vd)
5705{
5706 if (vd->vdev_probe_wanted) {
5707 vd->vdev_probe_wanted = B_FALSE;
5708 vdev_reopen(vd); /* vdev_open() does the actual probe */
5709 }
5710
5711 for (int c = 0; c < vd->vdev_children; c++)
5712 spa_async_probe(spa, vd->vdev_child[c]);
5713}
5714
5715static void
5716spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5717{
5718 sysevent_id_t eid;
5719 nvlist_t *attr;
5720 char *physpath;
5721
5722 if (!spa->spa_autoexpand)
5723 return;
5724
5725 for (int c = 0; c < vd->vdev_children; c++) {
5726 vdev_t *cvd = vd->vdev_child[c];
5727 spa_async_autoexpand(spa, cvd);
5728 }
5729
5730 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5731 return;
5732
5733 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5734 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5735
5736 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5737 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5738
5739 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5740 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5741
5742 nvlist_free(attr);
5743 kmem_free(physpath, MAXPATHLEN);
5744}
5745
5746static void
5747spa_async_thread(void *arg)
5748{
5749 spa_t *spa = arg;
5750 int tasks;
5751
5752 ASSERT(spa->spa_sync_on);
5753
5754 mutex_enter(&spa->spa_async_lock);
5755 tasks = spa->spa_async_tasks;
5756 spa->spa_async_tasks = 0;
5757 mutex_exit(&spa->spa_async_lock);
5758
5759 /*
5760 * See if the config needs to be updated.
5761 */
5762 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5763 uint64_t old_space, new_space;
5764
5765 mutex_enter(&spa_namespace_lock);
5766 old_space = metaslab_class_get_space(spa_normal_class(spa));
5767 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5768 new_space = metaslab_class_get_space(spa_normal_class(spa));
5769 mutex_exit(&spa_namespace_lock);
5770
5771 /*
5772 * If the pool grew as a result of the config update,
5773 * then log an internal history event.
5774 */
5775 if (new_space != old_space) {
5776 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5777 spa, NULL,
5778 "pool '%s' size: %llu(+%llu)",
5779 spa_name(spa), new_space, new_space - old_space);
5780 }
5781 }
5782
5783 /*
5784 * See if any devices need to be marked REMOVED.
5785 */
5786 if (tasks & SPA_ASYNC_REMOVE) {
5787 spa_vdev_state_enter(spa, SCL_NONE);
5788 spa_async_remove(spa, spa->spa_root_vdev);
5789 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5790 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5791 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5792 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5793 (void) spa_vdev_state_exit(spa, NULL, 0);
5794 }
5795
5796 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5797 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5798 spa_async_autoexpand(spa, spa->spa_root_vdev);
5799 spa_config_exit(spa, SCL_CONFIG, FTAG);
5800 }
5801
5802 /*
5803 * See if any devices need to be probed.
5804 */
5805 if (tasks & SPA_ASYNC_PROBE) {
5806 spa_vdev_state_enter(spa, SCL_NONE);
5807 spa_async_probe(spa, spa->spa_root_vdev);
5808 (void) spa_vdev_state_exit(spa, NULL, 0);
5809 }
5810
5811 /*
5812 * If any devices are done replacing, detach them.
5813 */
5814 if (tasks & SPA_ASYNC_RESILVER_DONE)
5815 spa_vdev_resilver_done(spa);
5816
5817 /*
5818 * Kick off a resilver.
5819 */
5820 if (tasks & SPA_ASYNC_RESILVER)
5821 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5822
5823 /*
5824 * Let the world know that we're done.
5825 */
5826 mutex_enter(&spa->spa_async_lock);
5827 spa->spa_async_thread = NULL;
5828 cv_broadcast(&spa->spa_async_cv);
5829 mutex_exit(&spa->spa_async_lock);
5830 thread_exit();
5831}
5832
5833void
5834spa_async_suspend(spa_t *spa)
5835{
5836 mutex_enter(&spa->spa_async_lock);
5837 spa->spa_async_suspended++;
5838 while (spa->spa_async_thread != NULL)
5839 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5840 mutex_exit(&spa->spa_async_lock);
5841}
5842
5843void
5844spa_async_resume(spa_t *spa)
5845{
5846 mutex_enter(&spa->spa_async_lock);
5847 ASSERT(spa->spa_async_suspended != 0);
5848 spa->spa_async_suspended--;
5849 mutex_exit(&spa->spa_async_lock);
5850}
5851
5852static void
5853spa_async_dispatch(spa_t *spa)
5854{
5855 mutex_enter(&spa->spa_async_lock);
5856 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5857 spa->spa_async_thread == NULL &&
5858 rootdir != NULL && !vn_is_readonly(rootdir))
5859 spa->spa_async_thread = thread_create(NULL, 0,
5860 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5861 mutex_exit(&spa->spa_async_lock);
5862}
5863
5864void
5865spa_async_request(spa_t *spa, int task)
5866{
5867 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5868 mutex_enter(&spa->spa_async_lock);
5869 spa->spa_async_tasks |= task;
5870 mutex_exit(&spa->spa_async_lock);
5871}
5872
5873/*
5874 * ==========================================================================
5875 * SPA syncing routines
5876 * ==========================================================================
5877 */
5878
5879static int
5880bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5881{
5882 bpobj_t *bpo = arg;
5883 bpobj_enqueue(bpo, bp, tx);
5884 return (0);
5885}
5886
5887static int
5888spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5889{
5890 zio_t *zio = arg;
5891
5892 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5893 BP_GET_PSIZE(bp), zio->io_flags));
5894 return (0);
5895}
5896
5897static void
5898spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5899{
5900 char *packed = NULL;
5901 size_t bufsize;
5902 size_t nvsize = 0;
5903 dmu_buf_t *db;
5904
5905 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5906
5907 /*
5908 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5909 * information. This avoids the dbuf_will_dirty() path and
5910 * saves us a pre-read to get data we don't actually care about.
5911 */
5912 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5913 packed = kmem_alloc(bufsize, KM_SLEEP);
5914
5915 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5916 KM_SLEEP) == 0);
5917 bzero(packed + nvsize, bufsize - nvsize);
5918
5919 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5920
5921 kmem_free(packed, bufsize);
5922
5923 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5924 dmu_buf_will_dirty(db, tx);
5925 *(uint64_t *)db->db_data = nvsize;
5926 dmu_buf_rele(db, FTAG);
5927}
5928
5929static void
5930spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5931 const char *config, const char *entry)
5932{
5933 nvlist_t *nvroot;
5934 nvlist_t **list;
5935 int i;
5936
5937 if (!sav->sav_sync)
5938 return;
5939
5940 /*
5941 * Update the MOS nvlist describing the list of available devices.
5942 * spa_validate_aux() will have already made sure this nvlist is
5943 * valid and the vdevs are labeled appropriately.
5944 */
5945 if (sav->sav_object == 0) {
5946 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5947 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5948 sizeof (uint64_t), tx);
5949 VERIFY(zap_update(spa->spa_meta_objset,
5950 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5951 &sav->sav_object, tx) == 0);
5952 }
5953
5954 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5955 if (sav->sav_count == 0) {
5956 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5957 } else {
5958 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5959 for (i = 0; i < sav->sav_count; i++)
5960 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5961 B_FALSE, VDEV_CONFIG_L2CACHE);
5962 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5963 sav->sav_count) == 0);
5964 for (i = 0; i < sav->sav_count; i++)
5965 nvlist_free(list[i]);
5966 kmem_free(list, sav->sav_count * sizeof (void *));
5967 }
5968
5969 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5970 nvlist_free(nvroot);
5971
5972 sav->sav_sync = B_FALSE;
5973}
5974
5975static void
5976spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5977{
5978 nvlist_t *config;
5979
5980 if (list_is_empty(&spa->spa_config_dirty_list))
5981 return;
5982
5983 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5984
5985 config = spa_config_generate(spa, spa->spa_root_vdev,
5986 dmu_tx_get_txg(tx), B_FALSE);
5987
5988 /*
5989 * If we're upgrading the spa version then make sure that
5990 * the config object gets updated with the correct version.
5991 */
5992 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5993 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5994 spa->spa_uberblock.ub_version);
5995
5996 spa_config_exit(spa, SCL_STATE, FTAG);
5997
5998 if (spa->spa_config_syncing)
5999 nvlist_free(spa->spa_config_syncing);
6000 spa->spa_config_syncing = config;
6001
6002 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6003}
6004
6005static void
6006spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
6007{
6008 spa_t *spa = arg1;
6009 uint64_t version = *(uint64_t *)arg2;
6010
6011 /*
6012 * Setting the version is special cased when first creating the pool.
6013 */
6014 ASSERT(tx->tx_txg != TXG_INITIAL);
6015
6016 ASSERT(version <= SPA_VERSION);
6017 ASSERT(version >= spa_version(spa));
6018
6019 spa->spa_uberblock.ub_version = version;
6020 vdev_config_dirty(spa->spa_root_vdev);
6021}
6022
6023/*
6024 * Set zpool properties.
6025 */
6026static void
6027spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
6028{
6029 spa_t *spa = arg1;
6030 objset_t *mos = spa->spa_meta_objset;
6031 nvlist_t *nvp = arg2;
6032 nvpair_t *elem = NULL;
6033
6034 mutex_enter(&spa->spa_props_lock);
6035
6036 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6037 uint64_t intval;
6038 char *strval, *fname;
6039 zpool_prop_t prop;
6040 const char *propname;
6041 zprop_type_t proptype;
6042 zfeature_info_t *feature;
6043
6044 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6045 case ZPROP_INVAL:
6046 /*
6047 * We checked this earlier in spa_prop_validate().
6048 */
6049 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6050
6051 fname = strchr(nvpair_name(elem), '@') + 1;
6052 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
6053
6054 spa_feature_enable(spa, feature, tx);
6055 break;
6056
6057 case ZPOOL_PROP_VERSION:
6058 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6059 /*
6060 * The version is synced seperatly before other
6061 * properties and should be correct by now.
6062 */
6063 ASSERT3U(spa_version(spa), >=, intval);
6064 break;
6065
6066 case ZPOOL_PROP_ALTROOT:
6067 /*
6068 * 'altroot' is a non-persistent property. It should
6069 * have been set temporarily at creation or import time.
6070 */
6071 ASSERT(spa->spa_root != NULL);
6072 break;
6073
6074 case ZPOOL_PROP_READONLY:
6075 case ZPOOL_PROP_CACHEFILE:
6076 /*
6077 * 'readonly' and 'cachefile' are also non-persisitent
6078 * properties.
6079 */
6080 break;
6081 case ZPOOL_PROP_COMMENT:
6082 VERIFY(nvpair_value_string(elem, &strval) == 0);
6083 if (spa->spa_comment != NULL)
6084 spa_strfree(spa->spa_comment);
6085 spa->spa_comment = spa_strdup(strval);
6086 /*
6087 * We need to dirty the configuration on all the vdevs
6088 * so that their labels get updated. It's unnecessary
6089 * to do this for pool creation since the vdev's
6090 * configuratoin has already been dirtied.
6091 */
6092 if (tx->tx_txg != TXG_INITIAL)
6093 vdev_config_dirty(spa->spa_root_vdev);
6094 break;
6095 default:
6096 /*
6097 * Set pool property values in the poolprops mos object.
6098 */
6099 if (spa->spa_pool_props_object == 0) {
6100 spa->spa_pool_props_object =
6101 zap_create_link(mos, DMU_OT_POOL_PROPS,
6102 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6103 tx);
6104 }
6105
6106 /* normalize the property name */
6107 propname = zpool_prop_to_name(prop);
6108 proptype = zpool_prop_get_type(prop);
6109
6110 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6111 ASSERT(proptype == PROP_TYPE_STRING);
6112 VERIFY(nvpair_value_string(elem, &strval) == 0);
6113 VERIFY(zap_update(mos,
6114 spa->spa_pool_props_object, propname,
6115 1, strlen(strval) + 1, strval, tx) == 0);
6116
6117 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6118 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6119
6120 if (proptype == PROP_TYPE_INDEX) {
6121 const char *unused;
6122 VERIFY(zpool_prop_index_to_string(
6123 prop, intval, &unused) == 0);
6124 }
6125 VERIFY(zap_update(mos,
6126 spa->spa_pool_props_object, propname,
6127 8, 1, &intval, tx) == 0);
6128 } else {
6129 ASSERT(0); /* not allowed */
6130 }
6131
6132 switch (prop) {
6133 case ZPOOL_PROP_DELEGATION:
6134 spa->spa_delegation = intval;
6135 break;
6136 case ZPOOL_PROP_BOOTFS:
6137 spa->spa_bootfs = intval;
6138 break;
6139 case ZPOOL_PROP_FAILUREMODE:
6140 spa->spa_failmode = intval;
6141 break;
6142 case ZPOOL_PROP_AUTOEXPAND:
6143 spa->spa_autoexpand = intval;
6144 if (tx->tx_txg != TXG_INITIAL)
6145 spa_async_request(spa,
6146 SPA_ASYNC_AUTOEXPAND);
6147 break;
6148 case ZPOOL_PROP_DEDUPDITTO:
6149 spa->spa_dedup_ditto = intval;
6150 break;
6151 default:
6152 break;
6153 }
6154 }
6155
6156 /* log internal history if this is not a zpool create */
6157 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6158 tx->tx_txg != TXG_INITIAL) {
6159 spa_history_log_internal(LOG_POOL_PROPSET,
6160 spa, tx, "%s %lld %s",
6161 nvpair_name(elem), intval, spa_name(spa));
6162 }
6163 }
6164
6165 mutex_exit(&spa->spa_props_lock);
6166}
6167
6168/*
6169 * Perform one-time upgrade on-disk changes. spa_version() does not
6170 * reflect the new version this txg, so there must be no changes this
6171 * txg to anything that the upgrade code depends on after it executes.
6172 * Therefore this must be called after dsl_pool_sync() does the sync
6173 * tasks.
6174 */
6175static void
6176spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6177{
6178 dsl_pool_t *dp = spa->spa_dsl_pool;
6179
6180 ASSERT(spa->spa_sync_pass == 1);
6181
6182 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6183 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6184 dsl_pool_create_origin(dp, tx);
6185
6186 /* Keeping the origin open increases spa_minref */
6187 spa->spa_minref += 3;
6188 }
6189
6190 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6191 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6192 dsl_pool_upgrade_clones(dp, tx);
6193 }
6194
6195 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6196 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6197 dsl_pool_upgrade_dir_clones(dp, tx);
6198
6199 /* Keeping the freedir open increases spa_minref */
6200 spa->spa_minref += 3;
6201 }
6202
6203 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6204 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6205 spa_feature_create_zap_objects(spa, tx);
6206 }
6207}
6208
6209/*
6210 * Sync the specified transaction group. New blocks may be dirtied as
6211 * part of the process, so we iterate until it converges.
6212 */
6213void
6214spa_sync(spa_t *spa, uint64_t txg)
6215{
6216 dsl_pool_t *dp = spa->spa_dsl_pool;
6217 objset_t *mos = spa->spa_meta_objset;
6218 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6219 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6220 vdev_t *rvd = spa->spa_root_vdev;
6221 vdev_t *vd;
6222 dmu_tx_t *tx;
6223 int error;
6224
6225 VERIFY(spa_writeable(spa));
6226
6227 /*
6228 * Lock out configuration changes.
6229 */
6230 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6231
6232 spa->spa_syncing_txg = txg;
6233 spa->spa_sync_pass = 0;
6234
6235 /*
6236 * If there are any pending vdev state changes, convert them
6237 * into config changes that go out with this transaction group.
6238 */
6239 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6240 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6241 /*
6242 * We need the write lock here because, for aux vdevs,
6243 * calling vdev_config_dirty() modifies sav_config.
6244 * This is ugly and will become unnecessary when we
6245 * eliminate the aux vdev wart by integrating all vdevs
6246 * into the root vdev tree.
6247 */
6248 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6249 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6250 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6251 vdev_state_clean(vd);
6252 vdev_config_dirty(vd);
6253 }
6254 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6255 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6256 }
6257 spa_config_exit(spa, SCL_STATE, FTAG);
6258
6259 tx = dmu_tx_create_assigned(dp, txg);
6260
6261 /*
6262 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6263 * set spa_deflate if we have no raid-z vdevs.
6264 */
6265 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6266 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6267 int i;
6268
6269 for (i = 0; i < rvd->vdev_children; i++) {
6270 vd = rvd->vdev_child[i];
6271 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6272 break;
6273 }
6274 if (i == rvd->vdev_children) {
6275 spa->spa_deflate = TRUE;
6276 VERIFY(0 == zap_add(spa->spa_meta_objset,
6277 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6278 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6279 }
6280 }
6281
6282 /*
6283 * If anything has changed in this txg, or if someone is waiting
6284 * for this txg to sync (eg, spa_vdev_remove()), push the
6285 * deferred frees from the previous txg. If not, leave them
6286 * alone so that we don't generate work on an otherwise idle
6287 * system.
6288 */
6289 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6290 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6291 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6292 ((dsl_scan_active(dp->dp_scan) ||
6293 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6294 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6295 VERIFY3U(bpobj_iterate(defer_bpo,
6296 spa_free_sync_cb, zio, tx), ==, 0);
6297 VERIFY0(zio_wait(zio));
6298 }
6299
6300 /*
6301 * Iterate to convergence.
6302 */
6303 do {
6304 int pass = ++spa->spa_sync_pass;
6305
6306 spa_sync_config_object(spa, tx);
6307 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6308 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6309 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6310 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6311 spa_errlog_sync(spa, txg);
6312 dsl_pool_sync(dp, txg);
6313
6314 if (pass < zfs_sync_pass_deferred_free) {
6315 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6316 bplist_iterate(free_bpl, spa_free_sync_cb,
6317 zio, tx);
6318 VERIFY(zio_wait(zio) == 0);
6319 } else {
6320 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6321 defer_bpo, tx);
6322 }
6323
6324 ddt_sync(spa, txg);
6325 dsl_scan_sync(dp, tx);
6326
6327 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6328 vdev_sync(vd, txg);
6329
6330 if (pass == 1)
6331 spa_sync_upgrades(spa, tx);
6332
6333 } while (dmu_objset_is_dirty(mos, txg));
6334
6335 /*
6336 * Rewrite the vdev configuration (which includes the uberblock)
6337 * to commit the transaction group.
6338 *
6339 * If there are no dirty vdevs, we sync the uberblock to a few
6340 * random top-level vdevs that are known to be visible in the
6341 * config cache (see spa_vdev_add() for a complete description).
6342 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6343 */
6344 for (;;) {
6345 /*
6346 * We hold SCL_STATE to prevent vdev open/close/etc.
6347 * while we're attempting to write the vdev labels.
6348 */
6349 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6350
6351 if (list_is_empty(&spa->spa_config_dirty_list)) {
6352 vdev_t *svd[SPA_DVAS_PER_BP];
6353 int svdcount = 0;
6354 int children = rvd->vdev_children;
6355 int c0 = spa_get_random(children);
6356
6357 for (int c = 0; c < children; c++) {
6358 vd = rvd->vdev_child[(c0 + c) % children];
6359 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6360 continue;
6361 svd[svdcount++] = vd;
6362 if (svdcount == SPA_DVAS_PER_BP)
6363 break;
6364 }
6365 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6366 if (error != 0)
6367 error = vdev_config_sync(svd, svdcount, txg,
6368 B_TRUE);
6369 } else {
6370 error = vdev_config_sync(rvd->vdev_child,
6371 rvd->vdev_children, txg, B_FALSE);
6372 if (error != 0)
6373 error = vdev_config_sync(rvd->vdev_child,
6374 rvd->vdev_children, txg, B_TRUE);
6375 }
6376
6377 if (error == 0)
6378 spa->spa_last_synced_guid = rvd->vdev_guid;
6379
6380 spa_config_exit(spa, SCL_STATE, FTAG);
6381
6382 if (error == 0)
6383 break;
6384 zio_suspend(spa, NULL);
6385 zio_resume_wait(spa);
6386 }
6387 dmu_tx_commit(tx);
6388
6389 /*
6390 * Clear the dirty config list.
6391 */
6392 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6393 vdev_config_clean(vd);
6394
6395 /*
6396 * Now that the new config has synced transactionally,
6397 * let it become visible to the config cache.
6398 */
6399 if (spa->spa_config_syncing != NULL) {
6400 spa_config_set(spa, spa->spa_config_syncing);
6401 spa->spa_config_txg = txg;
6402 spa->spa_config_syncing = NULL;
6403 }
6404
6405 spa->spa_ubsync = spa->spa_uberblock;
6406
6407 dsl_pool_sync_done(dp, txg);
6408
6409 /*
6410 * Update usable space statistics.
6411 */
6412 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6413 vdev_sync_done(vd, txg);
6414
6415 spa_update_dspace(spa);
6416
6417 /*
6418 * It had better be the case that we didn't dirty anything
6419 * since vdev_config_sync().
6420 */
6421 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6422 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6423 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6424
6425 spa->spa_sync_pass = 0;
6426
6427 spa_config_exit(spa, SCL_CONFIG, FTAG);
6428
6429 spa_handle_ignored_writes(spa);
6430
6431 /*
6432 * If any async tasks have been requested, kick them off.
6433 */
6434 spa_async_dispatch(spa);
6435}
6436
6437/*
6438 * Sync all pools. We don't want to hold the namespace lock across these
6439 * operations, so we take a reference on the spa_t and drop the lock during the
6440 * sync.
6441 */
6442void
6443spa_sync_allpools(void)
6444{
6445 spa_t *spa = NULL;
6446 mutex_enter(&spa_namespace_lock);
6447 while ((spa = spa_next(spa)) != NULL) {
6448 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6449 !spa_writeable(spa) || spa_suspended(spa))
6450 continue;
6451 spa_open_ref(spa, FTAG);
6452 mutex_exit(&spa_namespace_lock);
6453 txg_wait_synced(spa_get_dsl(spa), 0);
6454 mutex_enter(&spa_namespace_lock);
6455 spa_close(spa, FTAG);
6456 }
6457 mutex_exit(&spa_namespace_lock);
6458}
6459
6460/*
6461 * ==========================================================================
6462 * Miscellaneous routines
6463 * ==========================================================================
6464 */
6465
6466/*
6467 * Remove all pools in the system.
6468 */
6469void
6470spa_evict_all(void)
6471{
6472 spa_t *spa;
6473
6474 /*
6475 * Remove all cached state. All pools should be closed now,
6476 * so every spa in the AVL tree should be unreferenced.
6477 */
6478 mutex_enter(&spa_namespace_lock);
6479 while ((spa = spa_next(NULL)) != NULL) {
6480 /*
6481 * Stop async tasks. The async thread may need to detach
6482 * a device that's been replaced, which requires grabbing
6483 * spa_namespace_lock, so we must drop it here.
6484 */
6485 spa_open_ref(spa, FTAG);
6486 mutex_exit(&spa_namespace_lock);
6487 spa_async_suspend(spa);
6488 mutex_enter(&spa_namespace_lock);
6489 spa_close(spa, FTAG);
6490
6491 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6492 spa_unload(spa);
6493 spa_deactivate(spa);
6494 }
6495 spa_remove(spa);
6496 }
6497 mutex_exit(&spa_namespace_lock);
6498}
6499
6500vdev_t *
6501spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6502{
6503 vdev_t *vd;
6504 int i;
6505
6506 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6507 return (vd);
6508
6509 if (aux) {
6510 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6511 vd = spa->spa_l2cache.sav_vdevs[i];
6512 if (vd->vdev_guid == guid)
6513 return (vd);
6514 }
6515
6516 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6517 vd = spa->spa_spares.sav_vdevs[i];
6518 if (vd->vdev_guid == guid)
6519 return (vd);
6520 }
6521 }
6522
6523 return (NULL);
6524}
6525
6526void
6527spa_upgrade(spa_t *spa, uint64_t version)
6528{
6529 ASSERT(spa_writeable(spa));
6530
6531 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6532
6533 /*
6534 * This should only be called for a non-faulted pool, and since a
6535 * future version would result in an unopenable pool, this shouldn't be
6536 * possible.
6537 */
6538 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6539 ASSERT(version >= spa->spa_uberblock.ub_version);
6540
6541 spa->spa_uberblock.ub_version = version;
6542 vdev_config_dirty(spa->spa_root_vdev);
6543
6544 spa_config_exit(spa, SCL_ALL, FTAG);
6545
6546 txg_wait_synced(spa_get_dsl(spa), 0);
6547}
6548
6549boolean_t
6550spa_has_spare(spa_t *spa, uint64_t guid)
6551{
6552 int i;
6553 uint64_t spareguid;
6554 spa_aux_vdev_t *sav = &spa->spa_spares;
6555
6556 for (i = 0; i < sav->sav_count; i++)
6557 if (sav->sav_vdevs[i]->vdev_guid == guid)
6558 return (B_TRUE);
6559
6560 for (i = 0; i < sav->sav_npending; i++) {
6561 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6562 &spareguid) == 0 && spareguid == guid)
6563 return (B_TRUE);
6564 }
6565
6566 return (B_FALSE);
6567}
6568
6569/*
6570 * Check if a pool has an active shared spare device.
6571 * Note: reference count of an active spare is 2, as a spare and as a replace
6572 */
6573static boolean_t
6574spa_has_active_shared_spare(spa_t *spa)
6575{
6576 int i, refcnt;
6577 uint64_t pool;
6578 spa_aux_vdev_t *sav = &spa->spa_spares;
6579
6580 for (i = 0; i < sav->sav_count; i++) {
6581 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6582 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6583 refcnt > 2)
6584 return (B_TRUE);
6585 }
6586
6587 return (B_FALSE);
6588}
6589
6590/*
6591 * Post a sysevent corresponding to the given event. The 'name' must be one of
6592 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6593 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6594 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6595 * or zdb as real changes.
6596 */
6597void
6598spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6599{
6600#ifdef _KERNEL
6601 sysevent_t *ev;
6602 sysevent_attr_list_t *attr = NULL;
6603 sysevent_value_t value;
6604 sysevent_id_t eid;
6605
6606 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6607 SE_SLEEP);
6608
6609 value.value_type = SE_DATA_TYPE_STRING;
6610 value.value.sv_string = spa_name(spa);
6611 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6612 goto done;
6613
6614 value.value_type = SE_DATA_TYPE_UINT64;
6615 value.value.sv_uint64 = spa_guid(spa);
6616 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6617 goto done;
6618
6619 if (vd) {
6620 value.value_type = SE_DATA_TYPE_UINT64;
6621 value.value.sv_uint64 = vd->vdev_guid;
6622 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6623 SE_SLEEP) != 0)
6624 goto done;
6625
6626 if (vd->vdev_path) {
6627 value.value_type = SE_DATA_TYPE_STRING;
6628 value.value.sv_string = vd->vdev_path;
6629 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6630 &value, SE_SLEEP) != 0)
6631 goto done;
6632 }
6633 }
6634
6635 if (sysevent_attach_attributes(ev, attr) != 0)
6636 goto done;
6637 attr = NULL;
6638
6639 (void) log_sysevent(ev, SE_SLEEP, &eid);
6640
6641done:
6642 if (attr)
6643 sysevent_free_attr(attr);
6644 sysevent_free(ev);
6645#endif
6646}
| 3797 holes = NULL;
3798 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3799 &holes, &nholes);
3800
3801 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3802 for (i = 0; i < nchildren; i++) {
3803 if (i >= count)
3804 break;
3805 if (configs[i] == NULL)
3806 continue;
3807 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3808 &nvtop) == 0);
3809 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3810 }
3811 for (i = 0; holes != NULL && i < nholes; i++) {
3812 if (i >= nchildren)
3813 continue;
3814 if (tops[holes[i]] != NULL)
3815 continue;
3816 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3817 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3818 VDEV_TYPE_HOLE) == 0);
3819 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3820 holes[i]) == 0);
3821 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3822 0) == 0);
3823 }
3824 for (i = 0; i < nchildren; i++) {
3825 if (tops[i] != NULL)
3826 continue;
3827 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3828 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3829 VDEV_TYPE_MISSING) == 0);
3830 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3831 i) == 0);
3832 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3833 0) == 0);
3834 }
3835
3836 /*
3837 * Create pool config based on the best vdev config.
3838 */
3839 nvlist_dup(best_cfg, &config, KM_SLEEP);
3840
3841 /*
3842 * Put this pool's top-level vdevs into a root vdev.
3843 */
3844 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3845 &pgid) == 0);
3846 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3847 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3848 VDEV_TYPE_ROOT) == 0);
3849 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3850 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3851 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3852 tops, nchildren) == 0);
3853
3854 /*
3855 * Replace the existing vdev_tree with the new root vdev in
3856 * this pool's configuration (remove the old, add the new).
3857 */
3858 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3859
3860 /*
3861 * Drop vdev config elements that should not be present at pool level.
3862 */
3863 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3864 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3865
3866 for (i = 0; i < count; i++)
3867 nvlist_free(configs[i]);
3868 kmem_free(configs, count * sizeof(void *));
3869 for (i = 0; i < nchildren; i++)
3870 nvlist_free(tops[i]);
3871 kmem_free(tops, nchildren * sizeof(void *));
3872 nvlist_free(nvroot);
3873 return (config);
3874}
3875
3876int
3877spa_import_rootpool(const char *name)
3878{
3879 spa_t *spa;
3880 vdev_t *rvd, *bvd, *avd = NULL;
3881 nvlist_t *config, *nvtop;
3882 uint64_t txg;
3883 char *pname;
3884 int error;
3885
3886 /*
3887 * Read the label from the boot device and generate a configuration.
3888 */
3889 config = spa_generate_rootconf(name);
3890
3891 mutex_enter(&spa_namespace_lock);
3892 if (config != NULL) {
3893 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3894 &pname) == 0 && strcmp(name, pname) == 0);
3895 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
3896 == 0);
3897
3898 if ((spa = spa_lookup(pname)) != NULL) {
3899 /*
3900 * Remove the existing root pool from the namespace so
3901 * that we can replace it with the correct config
3902 * we just read in.
3903 */
3904 spa_remove(spa);
3905 }
3906 spa = spa_add(pname, config, NULL);
3907
3908 /*
3909 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
3910 * via spa_version().
3911 */
3912 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3913 &spa->spa_ubsync.ub_version) != 0)
3914 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3915 } else if ((spa = spa_lookup(name)) == NULL) {
3916 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
3917 name);
3918 return (EIO);
3919 } else {
3920 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
3921 }
3922 spa->spa_is_root = B_TRUE;
3923 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3924
3925 /*
3926 * Build up a vdev tree based on the boot device's label config.
3927 */
3928 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3929 &nvtop) == 0);
3930 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3931 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3932 VDEV_ALLOC_ROOTPOOL);
3933 spa_config_exit(spa, SCL_ALL, FTAG);
3934 if (error) {
3935 mutex_exit(&spa_namespace_lock);
3936 nvlist_free(config);
3937 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3938 pname);
3939 return (error);
3940 }
3941
3942 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3943 vdev_free(rvd);
3944 spa_config_exit(spa, SCL_ALL, FTAG);
3945 mutex_exit(&spa_namespace_lock);
3946
3947 nvlist_free(config);
3948 return (0);
3949}
3950
3951#endif /* sun */
3952#endif
3953
3954/*
3955 * Import a non-root pool into the system.
3956 */
3957int
3958spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3959{
3960 spa_t *spa;
3961 char *altroot = NULL;
3962 spa_load_state_t state = SPA_LOAD_IMPORT;
3963 zpool_rewind_policy_t policy;
3964 uint64_t mode = spa_mode_global;
3965 uint64_t readonly = B_FALSE;
3966 int error;
3967 nvlist_t *nvroot;
3968 nvlist_t **spares, **l2cache;
3969 uint_t nspares, nl2cache;
3970
3971 /*
3972 * If a pool with this name exists, return failure.
3973 */
3974 mutex_enter(&spa_namespace_lock);
3975 if (spa_lookup(pool) != NULL) {
3976 mutex_exit(&spa_namespace_lock);
3977 return (EEXIST);
3978 }
3979
3980 /*
3981 * Create and initialize the spa structure.
3982 */
3983 (void) nvlist_lookup_string(props,
3984 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3985 (void) nvlist_lookup_uint64(props,
3986 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3987 if (readonly)
3988 mode = FREAD;
3989 spa = spa_add(pool, config, altroot);
3990 spa->spa_import_flags = flags;
3991
3992 /*
3993 * Verbatim import - Take a pool and insert it into the namespace
3994 * as if it had been loaded at boot.
3995 */
3996 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3997 if (props != NULL)
3998 spa_configfile_set(spa, props, B_FALSE);
3999
4000 spa_config_sync(spa, B_FALSE, B_TRUE);
4001
4002 mutex_exit(&spa_namespace_lock);
4003 spa_history_log_version(spa, LOG_POOL_IMPORT);
4004
4005 return (0);
4006 }
4007
4008 spa_activate(spa, mode);
4009
4010 /*
4011 * Don't start async tasks until we know everything is healthy.
4012 */
4013 spa_async_suspend(spa);
4014
4015 zpool_get_rewind_policy(config, &policy);
4016 if (policy.zrp_request & ZPOOL_DO_REWIND)
4017 state = SPA_LOAD_RECOVER;
4018
4019 /*
4020 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4021 * because the user-supplied config is actually the one to trust when
4022 * doing an import.
4023 */
4024 if (state != SPA_LOAD_RECOVER)
4025 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4026
4027 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4028 policy.zrp_request);
4029
4030 /*
4031 * Propagate anything learned while loading the pool and pass it
4032 * back to caller (i.e. rewind info, missing devices, etc).
4033 */
4034 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4035 spa->spa_load_info) == 0);
4036
4037 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4038 /*
4039 * Toss any existing sparelist, as it doesn't have any validity
4040 * anymore, and conflicts with spa_has_spare().
4041 */
4042 if (spa->spa_spares.sav_config) {
4043 nvlist_free(spa->spa_spares.sav_config);
4044 spa->spa_spares.sav_config = NULL;
4045 spa_load_spares(spa);
4046 }
4047 if (spa->spa_l2cache.sav_config) {
4048 nvlist_free(spa->spa_l2cache.sav_config);
4049 spa->spa_l2cache.sav_config = NULL;
4050 spa_load_l2cache(spa);
4051 }
4052
4053 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4054 &nvroot) == 0);
4055 if (error == 0)
4056 error = spa_validate_aux(spa, nvroot, -1ULL,
4057 VDEV_ALLOC_SPARE);
4058 if (error == 0)
4059 error = spa_validate_aux(spa, nvroot, -1ULL,
4060 VDEV_ALLOC_L2CACHE);
4061 spa_config_exit(spa, SCL_ALL, FTAG);
4062
4063 if (props != NULL)
4064 spa_configfile_set(spa, props, B_FALSE);
4065
4066 if (error != 0 || (props && spa_writeable(spa) &&
4067 (error = spa_prop_set(spa, props)))) {
4068 spa_unload(spa);
4069 spa_deactivate(spa);
4070 spa_remove(spa);
4071 mutex_exit(&spa_namespace_lock);
4072 return (error);
4073 }
4074
4075 spa_async_resume(spa);
4076
4077 /*
4078 * Override any spares and level 2 cache devices as specified by
4079 * the user, as these may have correct device names/devids, etc.
4080 */
4081 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4082 &spares, &nspares) == 0) {
4083 if (spa->spa_spares.sav_config)
4084 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4085 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4086 else
4087 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4088 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4089 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4090 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4091 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4092 spa_load_spares(spa);
4093 spa_config_exit(spa, SCL_ALL, FTAG);
4094 spa->spa_spares.sav_sync = B_TRUE;
4095 }
4096 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4097 &l2cache, &nl2cache) == 0) {
4098 if (spa->spa_l2cache.sav_config)
4099 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4100 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4101 else
4102 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4103 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4104 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4105 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4106 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4107 spa_load_l2cache(spa);
4108 spa_config_exit(spa, SCL_ALL, FTAG);
4109 spa->spa_l2cache.sav_sync = B_TRUE;
4110 }
4111
4112 /*
4113 * Check for any removed devices.
4114 */
4115 if (spa->spa_autoreplace) {
4116 spa_aux_check_removed(&spa->spa_spares);
4117 spa_aux_check_removed(&spa->spa_l2cache);
4118 }
4119
4120 if (spa_writeable(spa)) {
4121 /*
4122 * Update the config cache to include the newly-imported pool.
4123 */
4124 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4125 }
4126
4127 /*
4128 * It's possible that the pool was expanded while it was exported.
4129 * We kick off an async task to handle this for us.
4130 */
4131 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4132
4133 mutex_exit(&spa_namespace_lock);
4134 spa_history_log_version(spa, LOG_POOL_IMPORT);
4135
4136#ifdef __FreeBSD__
4137#ifdef _KERNEL
4138 zvol_create_minors(pool);
4139#endif
4140#endif
4141 return (0);
4142}
4143
4144nvlist_t *
4145spa_tryimport(nvlist_t *tryconfig)
4146{
4147 nvlist_t *config = NULL;
4148 char *poolname;
4149 spa_t *spa;
4150 uint64_t state;
4151 int error;
4152
4153 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4154 return (NULL);
4155
4156 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4157 return (NULL);
4158
4159 /*
4160 * Create and initialize the spa structure.
4161 */
4162 mutex_enter(&spa_namespace_lock);
4163 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4164 spa_activate(spa, FREAD);
4165
4166 /*
4167 * Pass off the heavy lifting to spa_load().
4168 * Pass TRUE for mosconfig because the user-supplied config
4169 * is actually the one to trust when doing an import.
4170 */
4171 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4172
4173 /*
4174 * If 'tryconfig' was at least parsable, return the current config.
4175 */
4176 if (spa->spa_root_vdev != NULL) {
4177 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4178 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4179 poolname) == 0);
4180 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4181 state) == 0);
4182 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4183 spa->spa_uberblock.ub_timestamp) == 0);
4184 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4185 spa->spa_load_info) == 0);
4186
4187 /*
4188 * If the bootfs property exists on this pool then we
4189 * copy it out so that external consumers can tell which
4190 * pools are bootable.
4191 */
4192 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4193 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4194
4195 /*
4196 * We have to play games with the name since the
4197 * pool was opened as TRYIMPORT_NAME.
4198 */
4199 if (dsl_dsobj_to_dsname(spa_name(spa),
4200 spa->spa_bootfs, tmpname) == 0) {
4201 char *cp;
4202 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4203
4204 cp = strchr(tmpname, '/');
4205 if (cp == NULL) {
4206 (void) strlcpy(dsname, tmpname,
4207 MAXPATHLEN);
4208 } else {
4209 (void) snprintf(dsname, MAXPATHLEN,
4210 "%s/%s", poolname, ++cp);
4211 }
4212 VERIFY(nvlist_add_string(config,
4213 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4214 kmem_free(dsname, MAXPATHLEN);
4215 }
4216 kmem_free(tmpname, MAXPATHLEN);
4217 }
4218
4219 /*
4220 * Add the list of hot spares and level 2 cache devices.
4221 */
4222 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4223 spa_add_spares(spa, config);
4224 spa_add_l2cache(spa, config);
4225 spa_config_exit(spa, SCL_CONFIG, FTAG);
4226 }
4227
4228 spa_unload(spa);
4229 spa_deactivate(spa);
4230 spa_remove(spa);
4231 mutex_exit(&spa_namespace_lock);
4232
4233 return (config);
4234}
4235
4236/*
4237 * Pool export/destroy
4238 *
4239 * The act of destroying or exporting a pool is very simple. We make sure there
4240 * is no more pending I/O and any references to the pool are gone. Then, we
4241 * update the pool state and sync all the labels to disk, removing the
4242 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4243 * we don't sync the labels or remove the configuration cache.
4244 */
4245static int
4246spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4247 boolean_t force, boolean_t hardforce)
4248{
4249 spa_t *spa;
4250
4251 if (oldconfig)
4252 *oldconfig = NULL;
4253
4254 if (!(spa_mode_global & FWRITE))
4255 return (EROFS);
4256
4257 mutex_enter(&spa_namespace_lock);
4258 if ((spa = spa_lookup(pool)) == NULL) {
4259 mutex_exit(&spa_namespace_lock);
4260 return (ENOENT);
4261 }
4262
4263 /*
4264 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4265 * reacquire the namespace lock, and see if we can export.
4266 */
4267 spa_open_ref(spa, FTAG);
4268 mutex_exit(&spa_namespace_lock);
4269 spa_async_suspend(spa);
4270 mutex_enter(&spa_namespace_lock);
4271 spa_close(spa, FTAG);
4272
4273 /*
4274 * The pool will be in core if it's openable,
4275 * in which case we can modify its state.
4276 */
4277 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4278 /*
4279 * Objsets may be open only because they're dirty, so we
4280 * have to force it to sync before checking spa_refcnt.
4281 */
4282 txg_wait_synced(spa->spa_dsl_pool, 0);
4283
4284 /*
4285 * A pool cannot be exported or destroyed if there are active
4286 * references. If we are resetting a pool, allow references by
4287 * fault injection handlers.
4288 */
4289 if (!spa_refcount_zero(spa) ||
4290 (spa->spa_inject_ref != 0 &&
4291 new_state != POOL_STATE_UNINITIALIZED)) {
4292 spa_async_resume(spa);
4293 mutex_exit(&spa_namespace_lock);
4294 return (EBUSY);
4295 }
4296
4297 /*
4298 * A pool cannot be exported if it has an active shared spare.
4299 * This is to prevent other pools stealing the active spare
4300 * from an exported pool. At user's own will, such pool can
4301 * be forcedly exported.
4302 */
4303 if (!force && new_state == POOL_STATE_EXPORTED &&
4304 spa_has_active_shared_spare(spa)) {
4305 spa_async_resume(spa);
4306 mutex_exit(&spa_namespace_lock);
4307 return (EXDEV);
4308 }
4309
4310 /*
4311 * We want this to be reflected on every label,
4312 * so mark them all dirty. spa_unload() will do the
4313 * final sync that pushes these changes out.
4314 */
4315 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4317 spa->spa_state = new_state;
4318 spa->spa_final_txg = spa_last_synced_txg(spa) +
4319 TXG_DEFER_SIZE + 1;
4320 vdev_config_dirty(spa->spa_root_vdev);
4321 spa_config_exit(spa, SCL_ALL, FTAG);
4322 }
4323 }
4324
4325 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4326
4327 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4328 spa_unload(spa);
4329 spa_deactivate(spa);
4330 }
4331
4332 if (oldconfig && spa->spa_config)
4333 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4334
4335 if (new_state != POOL_STATE_UNINITIALIZED) {
4336 if (!hardforce)
4337 spa_config_sync(spa, B_TRUE, B_TRUE);
4338 spa_remove(spa);
4339 }
4340 mutex_exit(&spa_namespace_lock);
4341
4342 return (0);
4343}
4344
4345/*
4346 * Destroy a storage pool.
4347 */
4348int
4349spa_destroy(char *pool)
4350{
4351 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4352 B_FALSE, B_FALSE));
4353}
4354
4355/*
4356 * Export a storage pool.
4357 */
4358int
4359spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4360 boolean_t hardforce)
4361{
4362 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4363 force, hardforce));
4364}
4365
4366/*
4367 * Similar to spa_export(), this unloads the spa_t without actually removing it
4368 * from the namespace in any way.
4369 */
4370int
4371spa_reset(char *pool)
4372{
4373 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4374 B_FALSE, B_FALSE));
4375}
4376
4377/*
4378 * ==========================================================================
4379 * Device manipulation
4380 * ==========================================================================
4381 */
4382
4383/*
4384 * Add a device to a storage pool.
4385 */
4386int
4387spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4388{
4389 uint64_t txg, id;
4390 int error;
4391 vdev_t *rvd = spa->spa_root_vdev;
4392 vdev_t *vd, *tvd;
4393 nvlist_t **spares, **l2cache;
4394 uint_t nspares, nl2cache;
4395
4396 ASSERT(spa_writeable(spa));
4397
4398 txg = spa_vdev_enter(spa);
4399
4400 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4401 VDEV_ALLOC_ADD)) != 0)
4402 return (spa_vdev_exit(spa, NULL, txg, error));
4403
4404 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4405
4406 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4407 &nspares) != 0)
4408 nspares = 0;
4409
4410 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4411 &nl2cache) != 0)
4412 nl2cache = 0;
4413
4414 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4415 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4416
4417 if (vd->vdev_children != 0 &&
4418 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4419 return (spa_vdev_exit(spa, vd, txg, error));
4420
4421 /*
4422 * We must validate the spares and l2cache devices after checking the
4423 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4424 */
4425 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4426 return (spa_vdev_exit(spa, vd, txg, error));
4427
4428 /*
4429 * Transfer each new top-level vdev from vd to rvd.
4430 */
4431 for (int c = 0; c < vd->vdev_children; c++) {
4432
4433 /*
4434 * Set the vdev id to the first hole, if one exists.
4435 */
4436 for (id = 0; id < rvd->vdev_children; id++) {
4437 if (rvd->vdev_child[id]->vdev_ishole) {
4438 vdev_free(rvd->vdev_child[id]);
4439 break;
4440 }
4441 }
4442 tvd = vd->vdev_child[c];
4443 vdev_remove_child(vd, tvd);
4444 tvd->vdev_id = id;
4445 vdev_add_child(rvd, tvd);
4446 vdev_config_dirty(tvd);
4447 }
4448
4449 if (nspares != 0) {
4450 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4451 ZPOOL_CONFIG_SPARES);
4452 spa_load_spares(spa);
4453 spa->spa_spares.sav_sync = B_TRUE;
4454 }
4455
4456 if (nl2cache != 0) {
4457 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4458 ZPOOL_CONFIG_L2CACHE);
4459 spa_load_l2cache(spa);
4460 spa->spa_l2cache.sav_sync = B_TRUE;
4461 }
4462
4463 /*
4464 * We have to be careful when adding new vdevs to an existing pool.
4465 * If other threads start allocating from these vdevs before we
4466 * sync the config cache, and we lose power, then upon reboot we may
4467 * fail to open the pool because there are DVAs that the config cache
4468 * can't translate. Therefore, we first add the vdevs without
4469 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4470 * and then let spa_config_update() initialize the new metaslabs.
4471 *
4472 * spa_load() checks for added-but-not-initialized vdevs, so that
4473 * if we lose power at any point in this sequence, the remaining
4474 * steps will be completed the next time we load the pool.
4475 */
4476 (void) spa_vdev_exit(spa, vd, txg, 0);
4477
4478 mutex_enter(&spa_namespace_lock);
4479 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4480 mutex_exit(&spa_namespace_lock);
4481
4482 return (0);
4483}
4484
4485/*
4486 * Attach a device to a mirror. The arguments are the path to any device
4487 * in the mirror, and the nvroot for the new device. If the path specifies
4488 * a device that is not mirrored, we automatically insert the mirror vdev.
4489 *
4490 * If 'replacing' is specified, the new device is intended to replace the
4491 * existing device; in this case the two devices are made into their own
4492 * mirror using the 'replacing' vdev, which is functionally identical to
4493 * the mirror vdev (it actually reuses all the same ops) but has a few
4494 * extra rules: you can't attach to it after it's been created, and upon
4495 * completion of resilvering, the first disk (the one being replaced)
4496 * is automatically detached.
4497 */
4498int
4499spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4500{
4501 uint64_t txg, dtl_max_txg;
4502 vdev_t *rvd = spa->spa_root_vdev;
4503 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4504 vdev_ops_t *pvops;
4505 char *oldvdpath, *newvdpath;
4506 int newvd_isspare;
4507 int error;
4508
4509 ASSERT(spa_writeable(spa));
4510
4511 txg = spa_vdev_enter(spa);
4512
4513 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4514
4515 if (oldvd == NULL)
4516 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4517
4518 if (!oldvd->vdev_ops->vdev_op_leaf)
4519 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4520
4521 pvd = oldvd->vdev_parent;
4522
4523 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4524 VDEV_ALLOC_ATTACH)) != 0)
4525 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4526
4527 if (newrootvd->vdev_children != 1)
4528 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4529
4530 newvd = newrootvd->vdev_child[0];
4531
4532 if (!newvd->vdev_ops->vdev_op_leaf)
4533 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4534
4535 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4536 return (spa_vdev_exit(spa, newrootvd, txg, error));
4537
4538 /*
4539 * Spares can't replace logs
4540 */
4541 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4542 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4543
4544 if (!replacing) {
4545 /*
4546 * For attach, the only allowable parent is a mirror or the root
4547 * vdev.
4548 */
4549 if (pvd->vdev_ops != &vdev_mirror_ops &&
4550 pvd->vdev_ops != &vdev_root_ops)
4551 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4552
4553 pvops = &vdev_mirror_ops;
4554 } else {
4555 /*
4556 * Active hot spares can only be replaced by inactive hot
4557 * spares.
4558 */
4559 if (pvd->vdev_ops == &vdev_spare_ops &&
4560 oldvd->vdev_isspare &&
4561 !spa_has_spare(spa, newvd->vdev_guid))
4562 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4563
4564 /*
4565 * If the source is a hot spare, and the parent isn't already a
4566 * spare, then we want to create a new hot spare. Otherwise, we
4567 * want to create a replacing vdev. The user is not allowed to
4568 * attach to a spared vdev child unless the 'isspare' state is
4569 * the same (spare replaces spare, non-spare replaces
4570 * non-spare).
4571 */
4572 if (pvd->vdev_ops == &vdev_replacing_ops &&
4573 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4574 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4575 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4576 newvd->vdev_isspare != oldvd->vdev_isspare) {
4577 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4578 }
4579
4580 if (newvd->vdev_isspare)
4581 pvops = &vdev_spare_ops;
4582 else
4583 pvops = &vdev_replacing_ops;
4584 }
4585
4586 /*
4587 * Make sure the new device is big enough.
4588 */
4589 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4590 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4591
4592 /*
4593 * The new device cannot have a higher alignment requirement
4594 * than the top-level vdev.
4595 */
4596 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4597 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4598
4599 /*
4600 * If this is an in-place replacement, update oldvd's path and devid
4601 * to make it distinguishable from newvd, and unopenable from now on.
4602 */
4603 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4604 spa_strfree(oldvd->vdev_path);
4605 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4606 KM_SLEEP);
4607 (void) sprintf(oldvd->vdev_path, "%s/%s",
4608 newvd->vdev_path, "old");
4609 if (oldvd->vdev_devid != NULL) {
4610 spa_strfree(oldvd->vdev_devid);
4611 oldvd->vdev_devid = NULL;
4612 }
4613 }
4614
4615 /* mark the device being resilvered */
4616 newvd->vdev_resilvering = B_TRUE;
4617
4618 /*
4619 * If the parent is not a mirror, or if we're replacing, insert the new
4620 * mirror/replacing/spare vdev above oldvd.
4621 */
4622 if (pvd->vdev_ops != pvops)
4623 pvd = vdev_add_parent(oldvd, pvops);
4624
4625 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4626 ASSERT(pvd->vdev_ops == pvops);
4627 ASSERT(oldvd->vdev_parent == pvd);
4628
4629 /*
4630 * Extract the new device from its root and add it to pvd.
4631 */
4632 vdev_remove_child(newrootvd, newvd);
4633 newvd->vdev_id = pvd->vdev_children;
4634 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4635 vdev_add_child(pvd, newvd);
4636
4637 tvd = newvd->vdev_top;
4638 ASSERT(pvd->vdev_top == tvd);
4639 ASSERT(tvd->vdev_parent == rvd);
4640
4641 vdev_config_dirty(tvd);
4642
4643 /*
4644 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4645 * for any dmu_sync-ed blocks. It will propagate upward when
4646 * spa_vdev_exit() calls vdev_dtl_reassess().
4647 */
4648 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4649
4650 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4651 dtl_max_txg - TXG_INITIAL);
4652
4653 if (newvd->vdev_isspare) {
4654 spa_spare_activate(newvd);
4655 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4656 }
4657
4658 oldvdpath = spa_strdup(oldvd->vdev_path);
4659 newvdpath = spa_strdup(newvd->vdev_path);
4660 newvd_isspare = newvd->vdev_isspare;
4661
4662 /*
4663 * Mark newvd's DTL dirty in this txg.
4664 */
4665 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4666
4667 /*
4668 * Restart the resilver
4669 */
4670 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4671
4672 /*
4673 * Commit the config
4674 */
4675 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4676
4677 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4678 "%s vdev=%s %s vdev=%s",
4679 replacing && newvd_isspare ? "spare in" :
4680 replacing ? "replace" : "attach", newvdpath,
4681 replacing ? "for" : "to", oldvdpath);
4682
4683 spa_strfree(oldvdpath);
4684 spa_strfree(newvdpath);
4685
4686 if (spa->spa_bootfs)
4687 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4688
4689 return (0);
4690}
4691
4692/*
4693 * Detach a device from a mirror or replacing vdev.
4694 * If 'replace_done' is specified, only detach if the parent
4695 * is a replacing vdev.
4696 */
4697int
4698spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4699{
4700 uint64_t txg;
4701 int error;
4702 vdev_t *rvd = spa->spa_root_vdev;
4703 vdev_t *vd, *pvd, *cvd, *tvd;
4704 boolean_t unspare = B_FALSE;
4705 uint64_t unspare_guid;
4706 char *vdpath;
4707
4708 ASSERT(spa_writeable(spa));
4709
4710 txg = spa_vdev_enter(spa);
4711
4712 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4713
4714 if (vd == NULL)
4715 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4716
4717 if (!vd->vdev_ops->vdev_op_leaf)
4718 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4719
4720 pvd = vd->vdev_parent;
4721
4722 /*
4723 * If the parent/child relationship is not as expected, don't do it.
4724 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4725 * vdev that's replacing B with C. The user's intent in replacing
4726 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4727 * the replace by detaching C, the expected behavior is to end up
4728 * M(A,B). But suppose that right after deciding to detach C,
4729 * the replacement of B completes. We would have M(A,C), and then
4730 * ask to detach C, which would leave us with just A -- not what
4731 * the user wanted. To prevent this, we make sure that the
4732 * parent/child relationship hasn't changed -- in this example,
4733 * that C's parent is still the replacing vdev R.
4734 */
4735 if (pvd->vdev_guid != pguid && pguid != 0)
4736 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4737
4738 /*
4739 * Only 'replacing' or 'spare' vdevs can be replaced.
4740 */
4741 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4742 pvd->vdev_ops != &vdev_spare_ops)
4743 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4744
4745 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4746 spa_version(spa) >= SPA_VERSION_SPARES);
4747
4748 /*
4749 * Only mirror, replacing, and spare vdevs support detach.
4750 */
4751 if (pvd->vdev_ops != &vdev_replacing_ops &&
4752 pvd->vdev_ops != &vdev_mirror_ops &&
4753 pvd->vdev_ops != &vdev_spare_ops)
4754 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4755
4756 /*
4757 * If this device has the only valid copy of some data,
4758 * we cannot safely detach it.
4759 */
4760 if (vdev_dtl_required(vd))
4761 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4762
4763 ASSERT(pvd->vdev_children >= 2);
4764
4765 /*
4766 * If we are detaching the second disk from a replacing vdev, then
4767 * check to see if we changed the original vdev's path to have "/old"
4768 * at the end in spa_vdev_attach(). If so, undo that change now.
4769 */
4770 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4771 vd->vdev_path != NULL) {
4772 size_t len = strlen(vd->vdev_path);
4773
4774 for (int c = 0; c < pvd->vdev_children; c++) {
4775 cvd = pvd->vdev_child[c];
4776
4777 if (cvd == vd || cvd->vdev_path == NULL)
4778 continue;
4779
4780 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4781 strcmp(cvd->vdev_path + len, "/old") == 0) {
4782 spa_strfree(cvd->vdev_path);
4783 cvd->vdev_path = spa_strdup(vd->vdev_path);
4784 break;
4785 }
4786 }
4787 }
4788
4789 /*
4790 * If we are detaching the original disk from a spare, then it implies
4791 * that the spare should become a real disk, and be removed from the
4792 * active spare list for the pool.
4793 */
4794 if (pvd->vdev_ops == &vdev_spare_ops &&
4795 vd->vdev_id == 0 &&
4796 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4797 unspare = B_TRUE;
4798
4799 /*
4800 * Erase the disk labels so the disk can be used for other things.
4801 * This must be done after all other error cases are handled,
4802 * but before we disembowel vd (so we can still do I/O to it).
4803 * But if we can't do it, don't treat the error as fatal --
4804 * it may be that the unwritability of the disk is the reason
4805 * it's being detached!
4806 */
4807 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4808
4809 /*
4810 * Remove vd from its parent and compact the parent's children.
4811 */
4812 vdev_remove_child(pvd, vd);
4813 vdev_compact_children(pvd);
4814
4815 /*
4816 * Remember one of the remaining children so we can get tvd below.
4817 */
4818 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4819
4820 /*
4821 * If we need to remove the remaining child from the list of hot spares,
4822 * do it now, marking the vdev as no longer a spare in the process.
4823 * We must do this before vdev_remove_parent(), because that can
4824 * change the GUID if it creates a new toplevel GUID. For a similar
4825 * reason, we must remove the spare now, in the same txg as the detach;
4826 * otherwise someone could attach a new sibling, change the GUID, and
4827 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4828 */
4829 if (unspare) {
4830 ASSERT(cvd->vdev_isspare);
4831 spa_spare_remove(cvd);
4832 unspare_guid = cvd->vdev_guid;
4833 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4834 cvd->vdev_unspare = B_TRUE;
4835 }
4836
4837 /*
4838 * If the parent mirror/replacing vdev only has one child,
4839 * the parent is no longer needed. Remove it from the tree.
4840 */
4841 if (pvd->vdev_children == 1) {
4842 if (pvd->vdev_ops == &vdev_spare_ops)
4843 cvd->vdev_unspare = B_FALSE;
4844 vdev_remove_parent(cvd);
4845 cvd->vdev_resilvering = B_FALSE;
4846 }
4847
4848
4849 /*
4850 * We don't set tvd until now because the parent we just removed
4851 * may have been the previous top-level vdev.
4852 */
4853 tvd = cvd->vdev_top;
4854 ASSERT(tvd->vdev_parent == rvd);
4855
4856 /*
4857 * Reevaluate the parent vdev state.
4858 */
4859 vdev_propagate_state(cvd);
4860
4861 /*
4862 * If the 'autoexpand' property is set on the pool then automatically
4863 * try to expand the size of the pool. For example if the device we
4864 * just detached was smaller than the others, it may be possible to
4865 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4866 * first so that we can obtain the updated sizes of the leaf vdevs.
4867 */
4868 if (spa->spa_autoexpand) {
4869 vdev_reopen(tvd);
4870 vdev_expand(tvd, txg);
4871 }
4872
4873 vdev_config_dirty(tvd);
4874
4875 /*
4876 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4877 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4878 * But first make sure we're not on any *other* txg's DTL list, to
4879 * prevent vd from being accessed after it's freed.
4880 */
4881 vdpath = spa_strdup(vd->vdev_path);
4882 for (int t = 0; t < TXG_SIZE; t++)
4883 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4884 vd->vdev_detached = B_TRUE;
4885 vdev_dirty(tvd, VDD_DTL, vd, txg);
4886
4887 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4888
4889 /* hang on to the spa before we release the lock */
4890 spa_open_ref(spa, FTAG);
4891
4892 error = spa_vdev_exit(spa, vd, txg, 0);
4893
4894 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4895 "vdev=%s", vdpath);
4896 spa_strfree(vdpath);
4897
4898 /*
4899 * If this was the removal of the original device in a hot spare vdev,
4900 * then we want to go through and remove the device from the hot spare
4901 * list of every other pool.
4902 */
4903 if (unspare) {
4904 spa_t *altspa = NULL;
4905
4906 mutex_enter(&spa_namespace_lock);
4907 while ((altspa = spa_next(altspa)) != NULL) {
4908 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4909 altspa == spa)
4910 continue;
4911
4912 spa_open_ref(altspa, FTAG);
4913 mutex_exit(&spa_namespace_lock);
4914 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4915 mutex_enter(&spa_namespace_lock);
4916 spa_close(altspa, FTAG);
4917 }
4918 mutex_exit(&spa_namespace_lock);
4919
4920 /* search the rest of the vdevs for spares to remove */
4921 spa_vdev_resilver_done(spa);
4922 }
4923
4924 /* all done with the spa; OK to release */
4925 mutex_enter(&spa_namespace_lock);
4926 spa_close(spa, FTAG);
4927 mutex_exit(&spa_namespace_lock);
4928
4929 return (error);
4930}
4931
4932/*
4933 * Split a set of devices from their mirrors, and create a new pool from them.
4934 */
4935int
4936spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4937 nvlist_t *props, boolean_t exp)
4938{
4939 int error = 0;
4940 uint64_t txg, *glist;
4941 spa_t *newspa;
4942 uint_t c, children, lastlog;
4943 nvlist_t **child, *nvl, *tmp;
4944 dmu_tx_t *tx;
4945 char *altroot = NULL;
4946 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4947 boolean_t activate_slog;
4948
4949 ASSERT(spa_writeable(spa));
4950
4951 txg = spa_vdev_enter(spa);
4952
4953 /* clear the log and flush everything up to now */
4954 activate_slog = spa_passivate_log(spa);
4955 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4956 error = spa_offline_log(spa);
4957 txg = spa_vdev_config_enter(spa);
4958
4959 if (activate_slog)
4960 spa_activate_log(spa);
4961
4962 if (error != 0)
4963 return (spa_vdev_exit(spa, NULL, txg, error));
4964
4965 /* check new spa name before going any further */
4966 if (spa_lookup(newname) != NULL)
4967 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4968
4969 /*
4970 * scan through all the children to ensure they're all mirrors
4971 */
4972 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4973 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4974 &children) != 0)
4975 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4976
4977 /* first, check to ensure we've got the right child count */
4978 rvd = spa->spa_root_vdev;
4979 lastlog = 0;
4980 for (c = 0; c < rvd->vdev_children; c++) {
4981 vdev_t *vd = rvd->vdev_child[c];
4982
4983 /* don't count the holes & logs as children */
4984 if (vd->vdev_islog || vd->vdev_ishole) {
4985 if (lastlog == 0)
4986 lastlog = c;
4987 continue;
4988 }
4989
4990 lastlog = 0;
4991 }
4992 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4993 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4994
4995 /* next, ensure no spare or cache devices are part of the split */
4996 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4997 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4998 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4999
5000 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5001 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5002
5003 /* then, loop over each vdev and validate it */
5004 for (c = 0; c < children; c++) {
5005 uint64_t is_hole = 0;
5006
5007 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5008 &is_hole);
5009
5010 if (is_hole != 0) {
5011 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5012 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5013 continue;
5014 } else {
5015 error = EINVAL;
5016 break;
5017 }
5018 }
5019
5020 /* which disk is going to be split? */
5021 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5022 &glist[c]) != 0) {
5023 error = EINVAL;
5024 break;
5025 }
5026
5027 /* look it up in the spa */
5028 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5029 if (vml[c] == NULL) {
5030 error = ENODEV;
5031 break;
5032 }
5033
5034 /* make sure there's nothing stopping the split */
5035 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5036 vml[c]->vdev_islog ||
5037 vml[c]->vdev_ishole ||
5038 vml[c]->vdev_isspare ||
5039 vml[c]->vdev_isl2cache ||
5040 !vdev_writeable(vml[c]) ||
5041 vml[c]->vdev_children != 0 ||
5042 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5043 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5044 error = EINVAL;
5045 break;
5046 }
5047
5048 if (vdev_dtl_required(vml[c])) {
5049 error = EBUSY;
5050 break;
5051 }
5052
5053 /* we need certain info from the top level */
5054 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5055 vml[c]->vdev_top->vdev_ms_array) == 0);
5056 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5057 vml[c]->vdev_top->vdev_ms_shift) == 0);
5058 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5059 vml[c]->vdev_top->vdev_asize) == 0);
5060 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5061 vml[c]->vdev_top->vdev_ashift) == 0);
5062 }
5063
5064 if (error != 0) {
5065 kmem_free(vml, children * sizeof (vdev_t *));
5066 kmem_free(glist, children * sizeof (uint64_t));
5067 return (spa_vdev_exit(spa, NULL, txg, error));
5068 }
5069
5070 /* stop writers from using the disks */
5071 for (c = 0; c < children; c++) {
5072 if (vml[c] != NULL)
5073 vml[c]->vdev_offline = B_TRUE;
5074 }
5075 vdev_reopen(spa->spa_root_vdev);
5076
5077 /*
5078 * Temporarily record the splitting vdevs in the spa config. This
5079 * will disappear once the config is regenerated.
5080 */
5081 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5082 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5083 glist, children) == 0);
5084 kmem_free(glist, children * sizeof (uint64_t));
5085
5086 mutex_enter(&spa->spa_props_lock);
5087 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5088 nvl) == 0);
5089 mutex_exit(&spa->spa_props_lock);
5090 spa->spa_config_splitting = nvl;
5091 vdev_config_dirty(spa->spa_root_vdev);
5092
5093 /* configure and create the new pool */
5094 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5095 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5096 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5097 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5098 spa_version(spa)) == 0);
5099 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5100 spa->spa_config_txg) == 0);
5101 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5102 spa_generate_guid(NULL)) == 0);
5103 (void) nvlist_lookup_string(props,
5104 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5105
5106 /* add the new pool to the namespace */
5107 newspa = spa_add(newname, config, altroot);
5108 newspa->spa_config_txg = spa->spa_config_txg;
5109 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5110
5111 /* release the spa config lock, retaining the namespace lock */
5112 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5113
5114 if (zio_injection_enabled)
5115 zio_handle_panic_injection(spa, FTAG, 1);
5116
5117 spa_activate(newspa, spa_mode_global);
5118 spa_async_suspend(newspa);
5119
5120#ifndef sun
5121 /* mark that we are creating new spa by splitting */
5122 newspa->spa_splitting_newspa = B_TRUE;
5123#endif
5124 /* create the new pool from the disks of the original pool */
5125 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5126#ifndef sun
5127 newspa->spa_splitting_newspa = B_FALSE;
5128#endif
5129 if (error)
5130 goto out;
5131
5132 /* if that worked, generate a real config for the new pool */
5133 if (newspa->spa_root_vdev != NULL) {
5134 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5135 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5136 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5137 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5138 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5139 B_TRUE));
5140 }
5141
5142 /* set the props */
5143 if (props != NULL) {
5144 spa_configfile_set(newspa, props, B_FALSE);
5145 error = spa_prop_set(newspa, props);
5146 if (error)
5147 goto out;
5148 }
5149
5150 /* flush everything */
5151 txg = spa_vdev_config_enter(newspa);
5152 vdev_config_dirty(newspa->spa_root_vdev);
5153 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5154
5155 if (zio_injection_enabled)
5156 zio_handle_panic_injection(spa, FTAG, 2);
5157
5158 spa_async_resume(newspa);
5159
5160 /* finally, update the original pool's config */
5161 txg = spa_vdev_config_enter(spa);
5162 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5163 error = dmu_tx_assign(tx, TXG_WAIT);
5164 if (error != 0)
5165 dmu_tx_abort(tx);
5166 for (c = 0; c < children; c++) {
5167 if (vml[c] != NULL) {
5168 vdev_split(vml[c]);
5169 if (error == 0)
5170 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5171 spa, tx, "vdev=%s",
5172 vml[c]->vdev_path);
5173 vdev_free(vml[c]);
5174 }
5175 }
5176 vdev_config_dirty(spa->spa_root_vdev);
5177 spa->spa_config_splitting = NULL;
5178 nvlist_free(nvl);
5179 if (error == 0)
5180 dmu_tx_commit(tx);
5181 (void) spa_vdev_exit(spa, NULL, txg, 0);
5182
5183 if (zio_injection_enabled)
5184 zio_handle_panic_injection(spa, FTAG, 3);
5185
5186 /* split is complete; log a history record */
5187 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5188 "split new pool %s from pool %s", newname, spa_name(spa));
5189
5190 kmem_free(vml, children * sizeof (vdev_t *));
5191
5192 /* if we're not going to mount the filesystems in userland, export */
5193 if (exp)
5194 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5195 B_FALSE, B_FALSE);
5196
5197 return (error);
5198
5199out:
5200 spa_unload(newspa);
5201 spa_deactivate(newspa);
5202 spa_remove(newspa);
5203
5204 txg = spa_vdev_config_enter(spa);
5205
5206 /* re-online all offlined disks */
5207 for (c = 0; c < children; c++) {
5208 if (vml[c] != NULL)
5209 vml[c]->vdev_offline = B_FALSE;
5210 }
5211 vdev_reopen(spa->spa_root_vdev);
5212
5213 nvlist_free(spa->spa_config_splitting);
5214 spa->spa_config_splitting = NULL;
5215 (void) spa_vdev_exit(spa, NULL, txg, error);
5216
5217 kmem_free(vml, children * sizeof (vdev_t *));
5218 return (error);
5219}
5220
5221static nvlist_t *
5222spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5223{
5224 for (int i = 0; i < count; i++) {
5225 uint64_t guid;
5226
5227 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5228 &guid) == 0);
5229
5230 if (guid == target_guid)
5231 return (nvpp[i]);
5232 }
5233
5234 return (NULL);
5235}
5236
5237static void
5238spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5239 nvlist_t *dev_to_remove)
5240{
5241 nvlist_t **newdev = NULL;
5242
5243 if (count > 1)
5244 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5245
5246 for (int i = 0, j = 0; i < count; i++) {
5247 if (dev[i] == dev_to_remove)
5248 continue;
5249 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5250 }
5251
5252 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5253 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5254
5255 for (int i = 0; i < count - 1; i++)
5256 nvlist_free(newdev[i]);
5257
5258 if (count > 1)
5259 kmem_free(newdev, (count - 1) * sizeof (void *));
5260}
5261
5262/*
5263 * Evacuate the device.
5264 */
5265static int
5266spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5267{
5268 uint64_t txg;
5269 int error = 0;
5270
5271 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5272 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5273 ASSERT(vd == vd->vdev_top);
5274
5275 /*
5276 * Evacuate the device. We don't hold the config lock as writer
5277 * since we need to do I/O but we do keep the
5278 * spa_namespace_lock held. Once this completes the device
5279 * should no longer have any blocks allocated on it.
5280 */
5281 if (vd->vdev_islog) {
5282 if (vd->vdev_stat.vs_alloc != 0)
5283 error = spa_offline_log(spa);
5284 } else {
5285 error = ENOTSUP;
5286 }
5287
5288 if (error)
5289 return (error);
5290
5291 /*
5292 * The evacuation succeeded. Remove any remaining MOS metadata
5293 * associated with this vdev, and wait for these changes to sync.
5294 */
5295 ASSERT0(vd->vdev_stat.vs_alloc);
5296 txg = spa_vdev_config_enter(spa);
5297 vd->vdev_removing = B_TRUE;
5298 vdev_dirty(vd, 0, NULL, txg);
5299 vdev_config_dirty(vd);
5300 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5301
5302 return (0);
5303}
5304
5305/*
5306 * Complete the removal by cleaning up the namespace.
5307 */
5308static void
5309spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5310{
5311 vdev_t *rvd = spa->spa_root_vdev;
5312 uint64_t id = vd->vdev_id;
5313 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5314
5315 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5316 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5317 ASSERT(vd == vd->vdev_top);
5318
5319 /*
5320 * Only remove any devices which are empty.
5321 */
5322 if (vd->vdev_stat.vs_alloc != 0)
5323 return;
5324
5325 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5326
5327 if (list_link_active(&vd->vdev_state_dirty_node))
5328 vdev_state_clean(vd);
5329 if (list_link_active(&vd->vdev_config_dirty_node))
5330 vdev_config_clean(vd);
5331
5332 vdev_free(vd);
5333
5334 if (last_vdev) {
5335 vdev_compact_children(rvd);
5336 } else {
5337 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5338 vdev_add_child(rvd, vd);
5339 }
5340 vdev_config_dirty(rvd);
5341
5342 /*
5343 * Reassess the health of our root vdev.
5344 */
5345 vdev_reopen(rvd);
5346}
5347
5348/*
5349 * Remove a device from the pool -
5350 *
5351 * Removing a device from the vdev namespace requires several steps
5352 * and can take a significant amount of time. As a result we use
5353 * the spa_vdev_config_[enter/exit] functions which allow us to
5354 * grab and release the spa_config_lock while still holding the namespace
5355 * lock. During each step the configuration is synced out.
5356 */
5357
5358/*
5359 * Remove a device from the pool. Currently, this supports removing only hot
5360 * spares, slogs, and level 2 ARC devices.
5361 */
5362int
5363spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5364{
5365 vdev_t *vd;
5366 metaslab_group_t *mg;
5367 nvlist_t **spares, **l2cache, *nv;
5368 uint64_t txg = 0;
5369 uint_t nspares, nl2cache;
5370 int error = 0;
5371 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5372
5373 ASSERT(spa_writeable(spa));
5374
5375 if (!locked)
5376 txg = spa_vdev_enter(spa);
5377
5378 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5379
5380 if (spa->spa_spares.sav_vdevs != NULL &&
5381 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5382 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5383 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5384 /*
5385 * Only remove the hot spare if it's not currently in use
5386 * in this pool.
5387 */
5388 if (vd == NULL || unspare) {
5389 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5390 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5391 spa_load_spares(spa);
5392 spa->spa_spares.sav_sync = B_TRUE;
5393 } else {
5394 error = EBUSY;
5395 }
5396 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5397 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5398 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5399 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5400 /*
5401 * Cache devices can always be removed.
5402 */
5403 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5404 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5405 spa_load_l2cache(spa);
5406 spa->spa_l2cache.sav_sync = B_TRUE;
5407 } else if (vd != NULL && vd->vdev_islog) {
5408 ASSERT(!locked);
5409 ASSERT(vd == vd->vdev_top);
5410
5411 /*
5412 * XXX - Once we have bp-rewrite this should
5413 * become the common case.
5414 */
5415
5416 mg = vd->vdev_mg;
5417
5418 /*
5419 * Stop allocating from this vdev.
5420 */
5421 metaslab_group_passivate(mg);
5422
5423 /*
5424 * Wait for the youngest allocations and frees to sync,
5425 * and then wait for the deferral of those frees to finish.
5426 */
5427 spa_vdev_config_exit(spa, NULL,
5428 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5429
5430 /*
5431 * Attempt to evacuate the vdev.
5432 */
5433 error = spa_vdev_remove_evacuate(spa, vd);
5434
5435 txg = spa_vdev_config_enter(spa);
5436
5437 /*
5438 * If we couldn't evacuate the vdev, unwind.
5439 */
5440 if (error) {
5441 metaslab_group_activate(mg);
5442 return (spa_vdev_exit(spa, NULL, txg, error));
5443 }
5444
5445 /*
5446 * Clean up the vdev namespace.
5447 */
5448 spa_vdev_remove_from_namespace(spa, vd);
5449
5450 } else if (vd != NULL) {
5451 /*
5452 * Normal vdevs cannot be removed (yet).
5453 */
5454 error = ENOTSUP;
5455 } else {
5456 /*
5457 * There is no vdev of any kind with the specified guid.
5458 */
5459 error = ENOENT;
5460 }
5461
5462 if (!locked)
5463 return (spa_vdev_exit(spa, NULL, txg, error));
5464
5465 return (error);
5466}
5467
5468/*
5469 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5470 * current spared, so we can detach it.
5471 */
5472static vdev_t *
5473spa_vdev_resilver_done_hunt(vdev_t *vd)
5474{
5475 vdev_t *newvd, *oldvd;
5476
5477 for (int c = 0; c < vd->vdev_children; c++) {
5478 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5479 if (oldvd != NULL)
5480 return (oldvd);
5481 }
5482
5483 /*
5484 * Check for a completed replacement. We always consider the first
5485 * vdev in the list to be the oldest vdev, and the last one to be
5486 * the newest (see spa_vdev_attach() for how that works). In
5487 * the case where the newest vdev is faulted, we will not automatically
5488 * remove it after a resilver completes. This is OK as it will require
5489 * user intervention to determine which disk the admin wishes to keep.
5490 */
5491 if (vd->vdev_ops == &vdev_replacing_ops) {
5492 ASSERT(vd->vdev_children > 1);
5493
5494 newvd = vd->vdev_child[vd->vdev_children - 1];
5495 oldvd = vd->vdev_child[0];
5496
5497 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5498 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5499 !vdev_dtl_required(oldvd))
5500 return (oldvd);
5501 }
5502
5503 /*
5504 * Check for a completed resilver with the 'unspare' flag set.
5505 */
5506 if (vd->vdev_ops == &vdev_spare_ops) {
5507 vdev_t *first = vd->vdev_child[0];
5508 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5509
5510 if (last->vdev_unspare) {
5511 oldvd = first;
5512 newvd = last;
5513 } else if (first->vdev_unspare) {
5514 oldvd = last;
5515 newvd = first;
5516 } else {
5517 oldvd = NULL;
5518 }
5519
5520 if (oldvd != NULL &&
5521 vdev_dtl_empty(newvd, DTL_MISSING) &&
5522 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5523 !vdev_dtl_required(oldvd))
5524 return (oldvd);
5525
5526 /*
5527 * If there are more than two spares attached to a disk,
5528 * and those spares are not required, then we want to
5529 * attempt to free them up now so that they can be used
5530 * by other pools. Once we're back down to a single
5531 * disk+spare, we stop removing them.
5532 */
5533 if (vd->vdev_children > 2) {
5534 newvd = vd->vdev_child[1];
5535
5536 if (newvd->vdev_isspare && last->vdev_isspare &&
5537 vdev_dtl_empty(last, DTL_MISSING) &&
5538 vdev_dtl_empty(last, DTL_OUTAGE) &&
5539 !vdev_dtl_required(newvd))
5540 return (newvd);
5541 }
5542 }
5543
5544 return (NULL);
5545}
5546
5547static void
5548spa_vdev_resilver_done(spa_t *spa)
5549{
5550 vdev_t *vd, *pvd, *ppvd;
5551 uint64_t guid, sguid, pguid, ppguid;
5552
5553 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5554
5555 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5556 pvd = vd->vdev_parent;
5557 ppvd = pvd->vdev_parent;
5558 guid = vd->vdev_guid;
5559 pguid = pvd->vdev_guid;
5560 ppguid = ppvd->vdev_guid;
5561 sguid = 0;
5562 /*
5563 * If we have just finished replacing a hot spared device, then
5564 * we need to detach the parent's first child (the original hot
5565 * spare) as well.
5566 */
5567 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5568 ppvd->vdev_children == 2) {
5569 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5570 sguid = ppvd->vdev_child[1]->vdev_guid;
5571 }
5572 spa_config_exit(spa, SCL_ALL, FTAG);
5573 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5574 return;
5575 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5576 return;
5577 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5578 }
5579
5580 spa_config_exit(spa, SCL_ALL, FTAG);
5581}
5582
5583/*
5584 * Update the stored path or FRU for this vdev.
5585 */
5586int
5587spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5588 boolean_t ispath)
5589{
5590 vdev_t *vd;
5591 boolean_t sync = B_FALSE;
5592
5593 ASSERT(spa_writeable(spa));
5594
5595 spa_vdev_state_enter(spa, SCL_ALL);
5596
5597 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5598 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5599
5600 if (!vd->vdev_ops->vdev_op_leaf)
5601 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5602
5603 if (ispath) {
5604 if (strcmp(value, vd->vdev_path) != 0) {
5605 spa_strfree(vd->vdev_path);
5606 vd->vdev_path = spa_strdup(value);
5607 sync = B_TRUE;
5608 }
5609 } else {
5610 if (vd->vdev_fru == NULL) {
5611 vd->vdev_fru = spa_strdup(value);
5612 sync = B_TRUE;
5613 } else if (strcmp(value, vd->vdev_fru) != 0) {
5614 spa_strfree(vd->vdev_fru);
5615 vd->vdev_fru = spa_strdup(value);
5616 sync = B_TRUE;
5617 }
5618 }
5619
5620 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5621}
5622
5623int
5624spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5625{
5626 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5627}
5628
5629int
5630spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5631{
5632 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5633}
5634
5635/*
5636 * ==========================================================================
5637 * SPA Scanning
5638 * ==========================================================================
5639 */
5640
5641int
5642spa_scan_stop(spa_t *spa)
5643{
5644 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5645 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5646 return (EBUSY);
5647 return (dsl_scan_cancel(spa->spa_dsl_pool));
5648}
5649
5650int
5651spa_scan(spa_t *spa, pool_scan_func_t func)
5652{
5653 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5654
5655 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5656 return (ENOTSUP);
5657
5658 /*
5659 * If a resilver was requested, but there is no DTL on a
5660 * writeable leaf device, we have nothing to do.
5661 */
5662 if (func == POOL_SCAN_RESILVER &&
5663 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5664 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5665 return (0);
5666 }
5667
5668 return (dsl_scan(spa->spa_dsl_pool, func));
5669}
5670
5671/*
5672 * ==========================================================================
5673 * SPA async task processing
5674 * ==========================================================================
5675 */
5676
5677static void
5678spa_async_remove(spa_t *spa, vdev_t *vd)
5679{
5680 if (vd->vdev_remove_wanted) {
5681 vd->vdev_remove_wanted = B_FALSE;
5682 vd->vdev_delayed_close = B_FALSE;
5683 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5684
5685 /*
5686 * We want to clear the stats, but we don't want to do a full
5687 * vdev_clear() as that will cause us to throw away
5688 * degraded/faulted state as well as attempt to reopen the
5689 * device, all of which is a waste.
5690 */
5691 vd->vdev_stat.vs_read_errors = 0;
5692 vd->vdev_stat.vs_write_errors = 0;
5693 vd->vdev_stat.vs_checksum_errors = 0;
5694
5695 vdev_state_dirty(vd->vdev_top);
5696 }
5697
5698 for (int c = 0; c < vd->vdev_children; c++)
5699 spa_async_remove(spa, vd->vdev_child[c]);
5700}
5701
5702static void
5703spa_async_probe(spa_t *spa, vdev_t *vd)
5704{
5705 if (vd->vdev_probe_wanted) {
5706 vd->vdev_probe_wanted = B_FALSE;
5707 vdev_reopen(vd); /* vdev_open() does the actual probe */
5708 }
5709
5710 for (int c = 0; c < vd->vdev_children; c++)
5711 spa_async_probe(spa, vd->vdev_child[c]);
5712}
5713
5714static void
5715spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5716{
5717 sysevent_id_t eid;
5718 nvlist_t *attr;
5719 char *physpath;
5720
5721 if (!spa->spa_autoexpand)
5722 return;
5723
5724 for (int c = 0; c < vd->vdev_children; c++) {
5725 vdev_t *cvd = vd->vdev_child[c];
5726 spa_async_autoexpand(spa, cvd);
5727 }
5728
5729 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5730 return;
5731
5732 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5733 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5734
5735 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5736 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5737
5738 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5739 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5740
5741 nvlist_free(attr);
5742 kmem_free(physpath, MAXPATHLEN);
5743}
5744
5745static void
5746spa_async_thread(void *arg)
5747{
5748 spa_t *spa = arg;
5749 int tasks;
5750
5751 ASSERT(spa->spa_sync_on);
5752
5753 mutex_enter(&spa->spa_async_lock);
5754 tasks = spa->spa_async_tasks;
5755 spa->spa_async_tasks = 0;
5756 mutex_exit(&spa->spa_async_lock);
5757
5758 /*
5759 * See if the config needs to be updated.
5760 */
5761 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5762 uint64_t old_space, new_space;
5763
5764 mutex_enter(&spa_namespace_lock);
5765 old_space = metaslab_class_get_space(spa_normal_class(spa));
5766 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5767 new_space = metaslab_class_get_space(spa_normal_class(spa));
5768 mutex_exit(&spa_namespace_lock);
5769
5770 /*
5771 * If the pool grew as a result of the config update,
5772 * then log an internal history event.
5773 */
5774 if (new_space != old_space) {
5775 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5776 spa, NULL,
5777 "pool '%s' size: %llu(+%llu)",
5778 spa_name(spa), new_space, new_space - old_space);
5779 }
5780 }
5781
5782 /*
5783 * See if any devices need to be marked REMOVED.
5784 */
5785 if (tasks & SPA_ASYNC_REMOVE) {
5786 spa_vdev_state_enter(spa, SCL_NONE);
5787 spa_async_remove(spa, spa->spa_root_vdev);
5788 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5789 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5790 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5791 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5792 (void) spa_vdev_state_exit(spa, NULL, 0);
5793 }
5794
5795 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5796 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5797 spa_async_autoexpand(spa, spa->spa_root_vdev);
5798 spa_config_exit(spa, SCL_CONFIG, FTAG);
5799 }
5800
5801 /*
5802 * See if any devices need to be probed.
5803 */
5804 if (tasks & SPA_ASYNC_PROBE) {
5805 spa_vdev_state_enter(spa, SCL_NONE);
5806 spa_async_probe(spa, spa->spa_root_vdev);
5807 (void) spa_vdev_state_exit(spa, NULL, 0);
5808 }
5809
5810 /*
5811 * If any devices are done replacing, detach them.
5812 */
5813 if (tasks & SPA_ASYNC_RESILVER_DONE)
5814 spa_vdev_resilver_done(spa);
5815
5816 /*
5817 * Kick off a resilver.
5818 */
5819 if (tasks & SPA_ASYNC_RESILVER)
5820 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5821
5822 /*
5823 * Let the world know that we're done.
5824 */
5825 mutex_enter(&spa->spa_async_lock);
5826 spa->spa_async_thread = NULL;
5827 cv_broadcast(&spa->spa_async_cv);
5828 mutex_exit(&spa->spa_async_lock);
5829 thread_exit();
5830}
5831
5832void
5833spa_async_suspend(spa_t *spa)
5834{
5835 mutex_enter(&spa->spa_async_lock);
5836 spa->spa_async_suspended++;
5837 while (spa->spa_async_thread != NULL)
5838 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5839 mutex_exit(&spa->spa_async_lock);
5840}
5841
5842void
5843spa_async_resume(spa_t *spa)
5844{
5845 mutex_enter(&spa->spa_async_lock);
5846 ASSERT(spa->spa_async_suspended != 0);
5847 spa->spa_async_suspended--;
5848 mutex_exit(&spa->spa_async_lock);
5849}
5850
5851static void
5852spa_async_dispatch(spa_t *spa)
5853{
5854 mutex_enter(&spa->spa_async_lock);
5855 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5856 spa->spa_async_thread == NULL &&
5857 rootdir != NULL && !vn_is_readonly(rootdir))
5858 spa->spa_async_thread = thread_create(NULL, 0,
5859 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5860 mutex_exit(&spa->spa_async_lock);
5861}
5862
5863void
5864spa_async_request(spa_t *spa, int task)
5865{
5866 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5867 mutex_enter(&spa->spa_async_lock);
5868 spa->spa_async_tasks |= task;
5869 mutex_exit(&spa->spa_async_lock);
5870}
5871
5872/*
5873 * ==========================================================================
5874 * SPA syncing routines
5875 * ==========================================================================
5876 */
5877
5878static int
5879bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5880{
5881 bpobj_t *bpo = arg;
5882 bpobj_enqueue(bpo, bp, tx);
5883 return (0);
5884}
5885
5886static int
5887spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5888{
5889 zio_t *zio = arg;
5890
5891 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5892 BP_GET_PSIZE(bp), zio->io_flags));
5893 return (0);
5894}
5895
5896static void
5897spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5898{
5899 char *packed = NULL;
5900 size_t bufsize;
5901 size_t nvsize = 0;
5902 dmu_buf_t *db;
5903
5904 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5905
5906 /*
5907 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5908 * information. This avoids the dbuf_will_dirty() path and
5909 * saves us a pre-read to get data we don't actually care about.
5910 */
5911 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5912 packed = kmem_alloc(bufsize, KM_SLEEP);
5913
5914 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5915 KM_SLEEP) == 0);
5916 bzero(packed + nvsize, bufsize - nvsize);
5917
5918 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5919
5920 kmem_free(packed, bufsize);
5921
5922 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5923 dmu_buf_will_dirty(db, tx);
5924 *(uint64_t *)db->db_data = nvsize;
5925 dmu_buf_rele(db, FTAG);
5926}
5927
5928static void
5929spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5930 const char *config, const char *entry)
5931{
5932 nvlist_t *nvroot;
5933 nvlist_t **list;
5934 int i;
5935
5936 if (!sav->sav_sync)
5937 return;
5938
5939 /*
5940 * Update the MOS nvlist describing the list of available devices.
5941 * spa_validate_aux() will have already made sure this nvlist is
5942 * valid and the vdevs are labeled appropriately.
5943 */
5944 if (sav->sav_object == 0) {
5945 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5946 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5947 sizeof (uint64_t), tx);
5948 VERIFY(zap_update(spa->spa_meta_objset,
5949 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5950 &sav->sav_object, tx) == 0);
5951 }
5952
5953 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5954 if (sav->sav_count == 0) {
5955 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5956 } else {
5957 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5958 for (i = 0; i < sav->sav_count; i++)
5959 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5960 B_FALSE, VDEV_CONFIG_L2CACHE);
5961 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5962 sav->sav_count) == 0);
5963 for (i = 0; i < sav->sav_count; i++)
5964 nvlist_free(list[i]);
5965 kmem_free(list, sav->sav_count * sizeof (void *));
5966 }
5967
5968 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5969 nvlist_free(nvroot);
5970
5971 sav->sav_sync = B_FALSE;
5972}
5973
5974static void
5975spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5976{
5977 nvlist_t *config;
5978
5979 if (list_is_empty(&spa->spa_config_dirty_list))
5980 return;
5981
5982 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5983
5984 config = spa_config_generate(spa, spa->spa_root_vdev,
5985 dmu_tx_get_txg(tx), B_FALSE);
5986
5987 /*
5988 * If we're upgrading the spa version then make sure that
5989 * the config object gets updated with the correct version.
5990 */
5991 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5992 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5993 spa->spa_uberblock.ub_version);
5994
5995 spa_config_exit(spa, SCL_STATE, FTAG);
5996
5997 if (spa->spa_config_syncing)
5998 nvlist_free(spa->spa_config_syncing);
5999 spa->spa_config_syncing = config;
6000
6001 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6002}
6003
6004static void
6005spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
6006{
6007 spa_t *spa = arg1;
6008 uint64_t version = *(uint64_t *)arg2;
6009
6010 /*
6011 * Setting the version is special cased when first creating the pool.
6012 */
6013 ASSERT(tx->tx_txg != TXG_INITIAL);
6014
6015 ASSERT(version <= SPA_VERSION);
6016 ASSERT(version >= spa_version(spa));
6017
6018 spa->spa_uberblock.ub_version = version;
6019 vdev_config_dirty(spa->spa_root_vdev);
6020}
6021
6022/*
6023 * Set zpool properties.
6024 */
6025static void
6026spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
6027{
6028 spa_t *spa = arg1;
6029 objset_t *mos = spa->spa_meta_objset;
6030 nvlist_t *nvp = arg2;
6031 nvpair_t *elem = NULL;
6032
6033 mutex_enter(&spa->spa_props_lock);
6034
6035 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6036 uint64_t intval;
6037 char *strval, *fname;
6038 zpool_prop_t prop;
6039 const char *propname;
6040 zprop_type_t proptype;
6041 zfeature_info_t *feature;
6042
6043 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6044 case ZPROP_INVAL:
6045 /*
6046 * We checked this earlier in spa_prop_validate().
6047 */
6048 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6049
6050 fname = strchr(nvpair_name(elem), '@') + 1;
6051 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
6052
6053 spa_feature_enable(spa, feature, tx);
6054 break;
6055
6056 case ZPOOL_PROP_VERSION:
6057 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6058 /*
6059 * The version is synced seperatly before other
6060 * properties and should be correct by now.
6061 */
6062 ASSERT3U(spa_version(spa), >=, intval);
6063 break;
6064
6065 case ZPOOL_PROP_ALTROOT:
6066 /*
6067 * 'altroot' is a non-persistent property. It should
6068 * have been set temporarily at creation or import time.
6069 */
6070 ASSERT(spa->spa_root != NULL);
6071 break;
6072
6073 case ZPOOL_PROP_READONLY:
6074 case ZPOOL_PROP_CACHEFILE:
6075 /*
6076 * 'readonly' and 'cachefile' are also non-persisitent
6077 * properties.
6078 */
6079 break;
6080 case ZPOOL_PROP_COMMENT:
6081 VERIFY(nvpair_value_string(elem, &strval) == 0);
6082 if (spa->spa_comment != NULL)
6083 spa_strfree(spa->spa_comment);
6084 spa->spa_comment = spa_strdup(strval);
6085 /*
6086 * We need to dirty the configuration on all the vdevs
6087 * so that their labels get updated. It's unnecessary
6088 * to do this for pool creation since the vdev's
6089 * configuratoin has already been dirtied.
6090 */
6091 if (tx->tx_txg != TXG_INITIAL)
6092 vdev_config_dirty(spa->spa_root_vdev);
6093 break;
6094 default:
6095 /*
6096 * Set pool property values in the poolprops mos object.
6097 */
6098 if (spa->spa_pool_props_object == 0) {
6099 spa->spa_pool_props_object =
6100 zap_create_link(mos, DMU_OT_POOL_PROPS,
6101 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6102 tx);
6103 }
6104
6105 /* normalize the property name */
6106 propname = zpool_prop_to_name(prop);
6107 proptype = zpool_prop_get_type(prop);
6108
6109 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6110 ASSERT(proptype == PROP_TYPE_STRING);
6111 VERIFY(nvpair_value_string(elem, &strval) == 0);
6112 VERIFY(zap_update(mos,
6113 spa->spa_pool_props_object, propname,
6114 1, strlen(strval) + 1, strval, tx) == 0);
6115
6116 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6117 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6118
6119 if (proptype == PROP_TYPE_INDEX) {
6120 const char *unused;
6121 VERIFY(zpool_prop_index_to_string(
6122 prop, intval, &unused) == 0);
6123 }
6124 VERIFY(zap_update(mos,
6125 spa->spa_pool_props_object, propname,
6126 8, 1, &intval, tx) == 0);
6127 } else {
6128 ASSERT(0); /* not allowed */
6129 }
6130
6131 switch (prop) {
6132 case ZPOOL_PROP_DELEGATION:
6133 spa->spa_delegation = intval;
6134 break;
6135 case ZPOOL_PROP_BOOTFS:
6136 spa->spa_bootfs = intval;
6137 break;
6138 case ZPOOL_PROP_FAILUREMODE:
6139 spa->spa_failmode = intval;
6140 break;
6141 case ZPOOL_PROP_AUTOEXPAND:
6142 spa->spa_autoexpand = intval;
6143 if (tx->tx_txg != TXG_INITIAL)
6144 spa_async_request(spa,
6145 SPA_ASYNC_AUTOEXPAND);
6146 break;
6147 case ZPOOL_PROP_DEDUPDITTO:
6148 spa->spa_dedup_ditto = intval;
6149 break;
6150 default:
6151 break;
6152 }
6153 }
6154
6155 /* log internal history if this is not a zpool create */
6156 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6157 tx->tx_txg != TXG_INITIAL) {
6158 spa_history_log_internal(LOG_POOL_PROPSET,
6159 spa, tx, "%s %lld %s",
6160 nvpair_name(elem), intval, spa_name(spa));
6161 }
6162 }
6163
6164 mutex_exit(&spa->spa_props_lock);
6165}
6166
6167/*
6168 * Perform one-time upgrade on-disk changes. spa_version() does not
6169 * reflect the new version this txg, so there must be no changes this
6170 * txg to anything that the upgrade code depends on after it executes.
6171 * Therefore this must be called after dsl_pool_sync() does the sync
6172 * tasks.
6173 */
6174static void
6175spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6176{
6177 dsl_pool_t *dp = spa->spa_dsl_pool;
6178
6179 ASSERT(spa->spa_sync_pass == 1);
6180
6181 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6182 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6183 dsl_pool_create_origin(dp, tx);
6184
6185 /* Keeping the origin open increases spa_minref */
6186 spa->spa_minref += 3;
6187 }
6188
6189 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6190 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6191 dsl_pool_upgrade_clones(dp, tx);
6192 }
6193
6194 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6195 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6196 dsl_pool_upgrade_dir_clones(dp, tx);
6197
6198 /* Keeping the freedir open increases spa_minref */
6199 spa->spa_minref += 3;
6200 }
6201
6202 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6203 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6204 spa_feature_create_zap_objects(spa, tx);
6205 }
6206}
6207
6208/*
6209 * Sync the specified transaction group. New blocks may be dirtied as
6210 * part of the process, so we iterate until it converges.
6211 */
6212void
6213spa_sync(spa_t *spa, uint64_t txg)
6214{
6215 dsl_pool_t *dp = spa->spa_dsl_pool;
6216 objset_t *mos = spa->spa_meta_objset;
6217 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6218 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6219 vdev_t *rvd = spa->spa_root_vdev;
6220 vdev_t *vd;
6221 dmu_tx_t *tx;
6222 int error;
6223
6224 VERIFY(spa_writeable(spa));
6225
6226 /*
6227 * Lock out configuration changes.
6228 */
6229 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6230
6231 spa->spa_syncing_txg = txg;
6232 spa->spa_sync_pass = 0;
6233
6234 /*
6235 * If there are any pending vdev state changes, convert them
6236 * into config changes that go out with this transaction group.
6237 */
6238 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6239 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6240 /*
6241 * We need the write lock here because, for aux vdevs,
6242 * calling vdev_config_dirty() modifies sav_config.
6243 * This is ugly and will become unnecessary when we
6244 * eliminate the aux vdev wart by integrating all vdevs
6245 * into the root vdev tree.
6246 */
6247 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6248 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6249 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6250 vdev_state_clean(vd);
6251 vdev_config_dirty(vd);
6252 }
6253 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6254 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6255 }
6256 spa_config_exit(spa, SCL_STATE, FTAG);
6257
6258 tx = dmu_tx_create_assigned(dp, txg);
6259
6260 /*
6261 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6262 * set spa_deflate if we have no raid-z vdevs.
6263 */
6264 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6265 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6266 int i;
6267
6268 for (i = 0; i < rvd->vdev_children; i++) {
6269 vd = rvd->vdev_child[i];
6270 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6271 break;
6272 }
6273 if (i == rvd->vdev_children) {
6274 spa->spa_deflate = TRUE;
6275 VERIFY(0 == zap_add(spa->spa_meta_objset,
6276 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6277 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6278 }
6279 }
6280
6281 /*
6282 * If anything has changed in this txg, or if someone is waiting
6283 * for this txg to sync (eg, spa_vdev_remove()), push the
6284 * deferred frees from the previous txg. If not, leave them
6285 * alone so that we don't generate work on an otherwise idle
6286 * system.
6287 */
6288 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6289 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6290 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6291 ((dsl_scan_active(dp->dp_scan) ||
6292 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6293 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6294 VERIFY3U(bpobj_iterate(defer_bpo,
6295 spa_free_sync_cb, zio, tx), ==, 0);
6296 VERIFY0(zio_wait(zio));
6297 }
6298
6299 /*
6300 * Iterate to convergence.
6301 */
6302 do {
6303 int pass = ++spa->spa_sync_pass;
6304
6305 spa_sync_config_object(spa, tx);
6306 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6307 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6308 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6309 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6310 spa_errlog_sync(spa, txg);
6311 dsl_pool_sync(dp, txg);
6312
6313 if (pass < zfs_sync_pass_deferred_free) {
6314 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6315 bplist_iterate(free_bpl, spa_free_sync_cb,
6316 zio, tx);
6317 VERIFY(zio_wait(zio) == 0);
6318 } else {
6319 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6320 defer_bpo, tx);
6321 }
6322
6323 ddt_sync(spa, txg);
6324 dsl_scan_sync(dp, tx);
6325
6326 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6327 vdev_sync(vd, txg);
6328
6329 if (pass == 1)
6330 spa_sync_upgrades(spa, tx);
6331
6332 } while (dmu_objset_is_dirty(mos, txg));
6333
6334 /*
6335 * Rewrite the vdev configuration (which includes the uberblock)
6336 * to commit the transaction group.
6337 *
6338 * If there are no dirty vdevs, we sync the uberblock to a few
6339 * random top-level vdevs that are known to be visible in the
6340 * config cache (see spa_vdev_add() for a complete description).
6341 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6342 */
6343 for (;;) {
6344 /*
6345 * We hold SCL_STATE to prevent vdev open/close/etc.
6346 * while we're attempting to write the vdev labels.
6347 */
6348 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6349
6350 if (list_is_empty(&spa->spa_config_dirty_list)) {
6351 vdev_t *svd[SPA_DVAS_PER_BP];
6352 int svdcount = 0;
6353 int children = rvd->vdev_children;
6354 int c0 = spa_get_random(children);
6355
6356 for (int c = 0; c < children; c++) {
6357 vd = rvd->vdev_child[(c0 + c) % children];
6358 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6359 continue;
6360 svd[svdcount++] = vd;
6361 if (svdcount == SPA_DVAS_PER_BP)
6362 break;
6363 }
6364 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6365 if (error != 0)
6366 error = vdev_config_sync(svd, svdcount, txg,
6367 B_TRUE);
6368 } else {
6369 error = vdev_config_sync(rvd->vdev_child,
6370 rvd->vdev_children, txg, B_FALSE);
6371 if (error != 0)
6372 error = vdev_config_sync(rvd->vdev_child,
6373 rvd->vdev_children, txg, B_TRUE);
6374 }
6375
6376 if (error == 0)
6377 spa->spa_last_synced_guid = rvd->vdev_guid;
6378
6379 spa_config_exit(spa, SCL_STATE, FTAG);
6380
6381 if (error == 0)
6382 break;
6383 zio_suspend(spa, NULL);
6384 zio_resume_wait(spa);
6385 }
6386 dmu_tx_commit(tx);
6387
6388 /*
6389 * Clear the dirty config list.
6390 */
6391 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6392 vdev_config_clean(vd);
6393
6394 /*
6395 * Now that the new config has synced transactionally,
6396 * let it become visible to the config cache.
6397 */
6398 if (spa->spa_config_syncing != NULL) {
6399 spa_config_set(spa, spa->spa_config_syncing);
6400 spa->spa_config_txg = txg;
6401 spa->spa_config_syncing = NULL;
6402 }
6403
6404 spa->spa_ubsync = spa->spa_uberblock;
6405
6406 dsl_pool_sync_done(dp, txg);
6407
6408 /*
6409 * Update usable space statistics.
6410 */
6411 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6412 vdev_sync_done(vd, txg);
6413
6414 spa_update_dspace(spa);
6415
6416 /*
6417 * It had better be the case that we didn't dirty anything
6418 * since vdev_config_sync().
6419 */
6420 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6421 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6422 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6423
6424 spa->spa_sync_pass = 0;
6425
6426 spa_config_exit(spa, SCL_CONFIG, FTAG);
6427
6428 spa_handle_ignored_writes(spa);
6429
6430 /*
6431 * If any async tasks have been requested, kick them off.
6432 */
6433 spa_async_dispatch(spa);
6434}
6435
6436/*
6437 * Sync all pools. We don't want to hold the namespace lock across these
6438 * operations, so we take a reference on the spa_t and drop the lock during the
6439 * sync.
6440 */
6441void
6442spa_sync_allpools(void)
6443{
6444 spa_t *spa = NULL;
6445 mutex_enter(&spa_namespace_lock);
6446 while ((spa = spa_next(spa)) != NULL) {
6447 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6448 !spa_writeable(spa) || spa_suspended(spa))
6449 continue;
6450 spa_open_ref(spa, FTAG);
6451 mutex_exit(&spa_namespace_lock);
6452 txg_wait_synced(spa_get_dsl(spa), 0);
6453 mutex_enter(&spa_namespace_lock);
6454 spa_close(spa, FTAG);
6455 }
6456 mutex_exit(&spa_namespace_lock);
6457}
6458
6459/*
6460 * ==========================================================================
6461 * Miscellaneous routines
6462 * ==========================================================================
6463 */
6464
6465/*
6466 * Remove all pools in the system.
6467 */
6468void
6469spa_evict_all(void)
6470{
6471 spa_t *spa;
6472
6473 /*
6474 * Remove all cached state. All pools should be closed now,
6475 * so every spa in the AVL tree should be unreferenced.
6476 */
6477 mutex_enter(&spa_namespace_lock);
6478 while ((spa = spa_next(NULL)) != NULL) {
6479 /*
6480 * Stop async tasks. The async thread may need to detach
6481 * a device that's been replaced, which requires grabbing
6482 * spa_namespace_lock, so we must drop it here.
6483 */
6484 spa_open_ref(spa, FTAG);
6485 mutex_exit(&spa_namespace_lock);
6486 spa_async_suspend(spa);
6487 mutex_enter(&spa_namespace_lock);
6488 spa_close(spa, FTAG);
6489
6490 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6491 spa_unload(spa);
6492 spa_deactivate(spa);
6493 }
6494 spa_remove(spa);
6495 }
6496 mutex_exit(&spa_namespace_lock);
6497}
6498
6499vdev_t *
6500spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6501{
6502 vdev_t *vd;
6503 int i;
6504
6505 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6506 return (vd);
6507
6508 if (aux) {
6509 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6510 vd = spa->spa_l2cache.sav_vdevs[i];
6511 if (vd->vdev_guid == guid)
6512 return (vd);
6513 }
6514
6515 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6516 vd = spa->spa_spares.sav_vdevs[i];
6517 if (vd->vdev_guid == guid)
6518 return (vd);
6519 }
6520 }
6521
6522 return (NULL);
6523}
6524
6525void
6526spa_upgrade(spa_t *spa, uint64_t version)
6527{
6528 ASSERT(spa_writeable(spa));
6529
6530 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6531
6532 /*
6533 * This should only be called for a non-faulted pool, and since a
6534 * future version would result in an unopenable pool, this shouldn't be
6535 * possible.
6536 */
6537 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6538 ASSERT(version >= spa->spa_uberblock.ub_version);
6539
6540 spa->spa_uberblock.ub_version = version;
6541 vdev_config_dirty(spa->spa_root_vdev);
6542
6543 spa_config_exit(spa, SCL_ALL, FTAG);
6544
6545 txg_wait_synced(spa_get_dsl(spa), 0);
6546}
6547
6548boolean_t
6549spa_has_spare(spa_t *spa, uint64_t guid)
6550{
6551 int i;
6552 uint64_t spareguid;
6553 spa_aux_vdev_t *sav = &spa->spa_spares;
6554
6555 for (i = 0; i < sav->sav_count; i++)
6556 if (sav->sav_vdevs[i]->vdev_guid == guid)
6557 return (B_TRUE);
6558
6559 for (i = 0; i < sav->sav_npending; i++) {
6560 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6561 &spareguid) == 0 && spareguid == guid)
6562 return (B_TRUE);
6563 }
6564
6565 return (B_FALSE);
6566}
6567
6568/*
6569 * Check if a pool has an active shared spare device.
6570 * Note: reference count of an active spare is 2, as a spare and as a replace
6571 */
6572static boolean_t
6573spa_has_active_shared_spare(spa_t *spa)
6574{
6575 int i, refcnt;
6576 uint64_t pool;
6577 spa_aux_vdev_t *sav = &spa->spa_spares;
6578
6579 for (i = 0; i < sav->sav_count; i++) {
6580 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6581 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6582 refcnt > 2)
6583 return (B_TRUE);
6584 }
6585
6586 return (B_FALSE);
6587}
6588
6589/*
6590 * Post a sysevent corresponding to the given event. The 'name' must be one of
6591 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6592 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6593 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6594 * or zdb as real changes.
6595 */
6596void
6597spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6598{
6599#ifdef _KERNEL
6600 sysevent_t *ev;
6601 sysevent_attr_list_t *attr = NULL;
6602 sysevent_value_t value;
6603 sysevent_id_t eid;
6604
6605 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6606 SE_SLEEP);
6607
6608 value.value_type = SE_DATA_TYPE_STRING;
6609 value.value.sv_string = spa_name(spa);
6610 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6611 goto done;
6612
6613 value.value_type = SE_DATA_TYPE_UINT64;
6614 value.value.sv_uint64 = spa_guid(spa);
6615 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6616 goto done;
6617
6618 if (vd) {
6619 value.value_type = SE_DATA_TYPE_UINT64;
6620 value.value.sv_uint64 = vd->vdev_guid;
6621 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6622 SE_SLEEP) != 0)
6623 goto done;
6624
6625 if (vd->vdev_path) {
6626 value.value_type = SE_DATA_TYPE_STRING;
6627 value.value.sv_string = vd->vdev_path;
6628 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6629 &value, SE_SLEEP) != 0)
6630 goto done;
6631 }
6632 }
6633
6634 if (sysevent_attach_attributes(ev, attr) != 0)
6635 goto done;
6636 attr = NULL;
6637
6638 (void) log_sysevent(ev, SE_SLEEP, &eid);
6639
6640done:
6641 if (attr)
6642 sysevent_free_attr(attr);
6643 sysevent_free(ev);
6644#endif
6645}
|