]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/spa.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Fix stack inline
[mirror_zfs.git] / module / zfs / spa.c
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 */
25
26 /*
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
29 * pool.
30 */
31
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
35 #include <sys/zio.h>
36 #include <sys/zio_checksum.h>
37 #include <sys/dmu.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/zap.h>
40 #include <sys/zil.h>
41 #include <sys/ddt.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/metaslab.h>
44 #include <sys/metaslab_impl.h>
45 #include <sys/uberblock_impl.h>
46 #include <sys/txg.h>
47 #include <sys/avl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/arc.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/spa_boot.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/dsl_scan.h>
63
64 #ifdef _KERNEL
65 #include <sys/bootprops.h>
66 #include <sys/callb.h>
67 #include <sys/cpupart.h>
68 #include <sys/pool.h>
69 #include <sys/sysdc.h>
70 #include <sys/zone.h>
71 #endif /* _KERNEL */
72
73 #include "zfs_prop.h"
74 #include "zfs_comutil.h"
75
76 typedef enum zti_modes {
77 zti_mode_fixed, /* value is # of threads (min 1) */
78 zti_mode_online_percent, /* value is % of online CPUs */
79 zti_mode_batch, /* cpu-intensive; value is ignored */
80 zti_mode_null, /* don't create a taskq */
81 zti_nmodes
82 } zti_modes_t;
83
84 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
85 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
86 #define ZTI_BATCH { zti_mode_batch, 0 }
87 #define ZTI_NULL { zti_mode_null, 0 }
88
89 #define ZTI_ONE ZTI_FIX(1)
90
91 typedef struct zio_taskq_info {
92 enum zti_modes zti_mode;
93 uint_t zti_value;
94 } zio_taskq_info_t;
95
96 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97 "issue", "issue_high", "intr", "intr_high"
98 };
99
100 /*
101 * Define the taskq threads for the following I/O types:
102 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
103 */
104 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
106 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
107 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
108 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
109 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
112 };
113
114 static dsl_syncfunc_t spa_sync_props;
115 static boolean_t spa_has_active_shared_spare(spa_t *spa);
116 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
118 char **ereport);
119 static void spa_vdev_resilver_done(spa_t *spa);
120
121 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
122 id_t zio_taskq_psrset_bind = PS_NONE;
123 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
124 uint_t zio_taskq_basedc = 80; /* base duty cycle */
125
126 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
127
128 /*
129 * This (illegal) pool name is used when temporarily importing a spa_t in order
130 * to get the vdev stats associated with the imported devices.
131 */
132 #define TRYIMPORT_NAME "$import"
133
134 /*
135 * ==========================================================================
136 * SPA properties routines
137 * ==========================================================================
138 */
139
140 /*
141 * Add a (source=src, propname=propval) list to an nvlist.
142 */
143 static void
144 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
145 uint64_t intval, zprop_source_t src)
146 {
147 const char *propname = zpool_prop_to_name(prop);
148 nvlist_t *propval;
149
150 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
151 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
152
153 if (strval != NULL)
154 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
155 else
156 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
157
158 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
159 nvlist_free(propval);
160 }
161
162 /*
163 * Get property values from the spa configuration.
164 */
165 static void
166 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
167 {
168 uint64_t size;
169 uint64_t alloc;
170 uint64_t cap, version;
171 zprop_source_t src = ZPROP_SRC_NONE;
172 spa_config_dirent_t *dp;
173
174 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
175
176 if (spa->spa_root_vdev != NULL) {
177 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
178 size = metaslab_class_get_space(spa_normal_class(spa));
179 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
180 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
181 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
182 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
183 size - alloc, src);
184 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
185 (spa_mode(spa) == FREAD), src);
186
187 cap = (size == 0) ? 0 : (alloc * 100 / size);
188 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
189
190 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
191 ddt_get_pool_dedup_ratio(spa), src);
192
193 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
194 spa->spa_root_vdev->vdev_state, src);
195
196 version = spa_version(spa);
197 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
198 src = ZPROP_SRC_DEFAULT;
199 else
200 src = ZPROP_SRC_LOCAL;
201 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
202 }
203
204 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
205
206 if (spa->spa_root != NULL)
207 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
208 0, ZPROP_SRC_LOCAL);
209
210 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
211 if (dp->scd_path == NULL) {
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213 "none", 0, ZPROP_SRC_LOCAL);
214 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
215 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
216 dp->scd_path, 0, ZPROP_SRC_LOCAL);
217 }
218 }
219 }
220
221 /*
222 * Get zpool property values.
223 */
224 int
225 spa_prop_get(spa_t *spa, nvlist_t **nvp)
226 {
227 objset_t *mos = spa->spa_meta_objset;
228 zap_cursor_t zc;
229 zap_attribute_t za;
230 int err;
231
232 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
233
234 mutex_enter(&spa->spa_props_lock);
235
236 /*
237 * Get properties from the spa config.
238 */
239 spa_prop_get_config(spa, nvp);
240
241 /* If no pool property object, no more prop to get. */
242 if (mos == NULL || spa->spa_pool_props_object == 0) {
243 mutex_exit(&spa->spa_props_lock);
244 return (0);
245 }
246
247 /*
248 * Get properties from the MOS pool property object.
249 */
250 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
251 (err = zap_cursor_retrieve(&zc, &za)) == 0;
252 zap_cursor_advance(&zc)) {
253 uint64_t intval = 0;
254 char *strval = NULL;
255 zprop_source_t src = ZPROP_SRC_DEFAULT;
256 zpool_prop_t prop;
257
258 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
259 continue;
260
261 switch (za.za_integer_length) {
262 case 8:
263 /* integer property */
264 if (za.za_first_integer !=
265 zpool_prop_default_numeric(prop))
266 src = ZPROP_SRC_LOCAL;
267
268 if (prop == ZPOOL_PROP_BOOTFS) {
269 dsl_pool_t *dp;
270 dsl_dataset_t *ds = NULL;
271
272 dp = spa_get_dsl(spa);
273 rw_enter(&dp->dp_config_rwlock, RW_READER);
274 if ((err = dsl_dataset_hold_obj(dp,
275 za.za_first_integer, FTAG, &ds))) {
276 rw_exit(&dp->dp_config_rwlock);
277 break;
278 }
279
280 strval = kmem_alloc(
281 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
282 KM_SLEEP);
283 dsl_dataset_name(ds, strval);
284 dsl_dataset_rele(ds, FTAG);
285 rw_exit(&dp->dp_config_rwlock);
286 } else {
287 strval = NULL;
288 intval = za.za_first_integer;
289 }
290
291 spa_prop_add_list(*nvp, prop, strval, intval, src);
292
293 if (strval != NULL)
294 kmem_free(strval,
295 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
296
297 break;
298
299 case 1:
300 /* string property */
301 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
302 err = zap_lookup(mos, spa->spa_pool_props_object,
303 za.za_name, 1, za.za_num_integers, strval);
304 if (err) {
305 kmem_free(strval, za.za_num_integers);
306 break;
307 }
308 spa_prop_add_list(*nvp, prop, strval, 0, src);
309 kmem_free(strval, za.za_num_integers);
310 break;
311
312 default:
313 break;
314 }
315 }
316 zap_cursor_fini(&zc);
317 mutex_exit(&spa->spa_props_lock);
318 out:
319 if (err && err != ENOENT) {
320 nvlist_free(*nvp);
321 *nvp = NULL;
322 return (err);
323 }
324
325 return (0);
326 }
327
328 /*
329 * Validate the given pool properties nvlist and modify the list
330 * for the property values to be set.
331 */
332 static int
333 spa_prop_validate(spa_t *spa, nvlist_t *props)
334 {
335 nvpair_t *elem;
336 int error = 0, reset_bootfs = 0;
337 uint64_t objnum = 0;
338
339 elem = NULL;
340 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
341 zpool_prop_t prop;
342 char *propname, *strval;
343 uint64_t intval;
344 objset_t *os;
345 char *slash;
346
347 propname = nvpair_name(elem);
348
349 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
350 return (EINVAL);
351
352 switch (prop) {
353 case ZPOOL_PROP_VERSION:
354 error = nvpair_value_uint64(elem, &intval);
355 if (!error &&
356 (intval < spa_version(spa) || intval > SPA_VERSION))
357 error = EINVAL;
358 break;
359
360 case ZPOOL_PROP_DELEGATION:
361 case ZPOOL_PROP_AUTOREPLACE:
362 case ZPOOL_PROP_LISTSNAPS:
363 case ZPOOL_PROP_AUTOEXPAND:
364 error = nvpair_value_uint64(elem, &intval);
365 if (!error && intval > 1)
366 error = EINVAL;
367 break;
368
369 case ZPOOL_PROP_BOOTFS:
370 /*
371 * If the pool version is less than SPA_VERSION_BOOTFS,
372 * or the pool is still being created (version == 0),
373 * the bootfs property cannot be set.
374 */
375 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
376 error = ENOTSUP;
377 break;
378 }
379
380 /*
381 * Make sure the vdev config is bootable
382 */
383 if (!vdev_is_bootable(spa->spa_root_vdev)) {
384 error = ENOTSUP;
385 break;
386 }
387
388 reset_bootfs = 1;
389
390 error = nvpair_value_string(elem, &strval);
391
392 if (!error) {
393 uint64_t compress;
394
395 if (strval == NULL || strval[0] == '\0') {
396 objnum = zpool_prop_default_numeric(
397 ZPOOL_PROP_BOOTFS);
398 break;
399 }
400
401 if ((error = dmu_objset_hold(strval,FTAG,&os)))
402 break;
403
404 /* Must be ZPL and not gzip compressed. */
405
406 if (dmu_objset_type(os) != DMU_OST_ZFS) {
407 error = ENOTSUP;
408 } else if ((error = dsl_prop_get_integer(strval,
409 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
410 &compress, NULL)) == 0 &&
411 !BOOTFS_COMPRESS_VALID(compress)) {
412 error = ENOTSUP;
413 } else {
414 objnum = dmu_objset_id(os);
415 }
416 dmu_objset_rele(os, FTAG);
417 }
418 break;
419
420 case ZPOOL_PROP_FAILUREMODE:
421 error = nvpair_value_uint64(elem, &intval);
422 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
423 intval > ZIO_FAILURE_MODE_PANIC))
424 error = EINVAL;
425
426 /*
427 * This is a special case which only occurs when
428 * the pool has completely failed. This allows
429 * the user to change the in-core failmode property
430 * without syncing it out to disk (I/Os might
431 * currently be blocked). We do this by returning
432 * EIO to the caller (spa_prop_set) to trick it
433 * into thinking we encountered a property validation
434 * error.
435 */
436 if (!error && spa_suspended(spa)) {
437 spa->spa_failmode = intval;
438 error = EIO;
439 }
440 break;
441
442 case ZPOOL_PROP_CACHEFILE:
443 if ((error = nvpair_value_string(elem, &strval)) != 0)
444 break;
445
446 if (strval[0] == '\0')
447 break;
448
449 if (strcmp(strval, "none") == 0)
450 break;
451
452 if (strval[0] != '/') {
453 error = EINVAL;
454 break;
455 }
456
457 slash = strrchr(strval, '/');
458 ASSERT(slash != NULL);
459
460 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
461 strcmp(slash, "/..") == 0)
462 error = EINVAL;
463 break;
464
465 case ZPOOL_PROP_DEDUPDITTO:
466 if (spa_version(spa) < SPA_VERSION_DEDUP)
467 error = ENOTSUP;
468 else
469 error = nvpair_value_uint64(elem, &intval);
470 if (error == 0 &&
471 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
472 error = EINVAL;
473 break;
474
475 default:
476 break;
477 }
478
479 if (error)
480 break;
481 }
482
483 if (!error && reset_bootfs) {
484 error = nvlist_remove(props,
485 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
486
487 if (!error) {
488 error = nvlist_add_uint64(props,
489 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
490 }
491 }
492
493 return (error);
494 }
495
496 void
497 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
498 {
499 char *cachefile;
500 spa_config_dirent_t *dp;
501
502 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
503 &cachefile) != 0)
504 return;
505
506 dp = kmem_alloc(sizeof (spa_config_dirent_t),
507 KM_SLEEP);
508
509 if (cachefile[0] == '\0')
510 dp->scd_path = spa_strdup(spa_config_path);
511 else if (strcmp(cachefile, "none") == 0)
512 dp->scd_path = NULL;
513 else
514 dp->scd_path = spa_strdup(cachefile);
515
516 list_insert_head(&spa->spa_config_list, dp);
517 if (need_sync)
518 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
519 }
520
521 int
522 spa_prop_set(spa_t *spa, nvlist_t *nvp)
523 {
524 int error;
525 nvpair_t *elem;
526 boolean_t need_sync = B_FALSE;
527 zpool_prop_t prop;
528
529 if ((error = spa_prop_validate(spa, nvp)) != 0)
530 return (error);
531
532 elem = NULL;
533 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
534 if ((prop = zpool_name_to_prop(
535 nvpair_name(elem))) == ZPROP_INVAL)
536 return (EINVAL);
537
538 if (prop == ZPOOL_PROP_CACHEFILE ||
539 prop == ZPOOL_PROP_ALTROOT ||
540 prop == ZPOOL_PROP_READONLY)
541 continue;
542
543 need_sync = B_TRUE;
544 break;
545 }
546
547 if (need_sync)
548 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
549 spa, nvp, 3));
550 else
551 return (0);
552 }
553
554 /*
555 * If the bootfs property value is dsobj, clear it.
556 */
557 void
558 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
559 {
560 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
561 VERIFY(zap_remove(spa->spa_meta_objset,
562 spa->spa_pool_props_object,
563 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
564 spa->spa_bootfs = 0;
565 }
566 }
567
568 /*
569 * ==========================================================================
570 * SPA state manipulation (open/create/destroy/import/export)
571 * ==========================================================================
572 */
573
574 static int
575 spa_error_entry_compare(const void *a, const void *b)
576 {
577 spa_error_entry_t *sa = (spa_error_entry_t *)a;
578 spa_error_entry_t *sb = (spa_error_entry_t *)b;
579 int ret;
580
581 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
582 sizeof (zbookmark_t));
583
584 if (ret < 0)
585 return (-1);
586 else if (ret > 0)
587 return (1);
588 else
589 return (0);
590 }
591
592 /*
593 * Utility function which retrieves copies of the current logs and
594 * re-initializes them in the process.
595 */
596 void
597 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
598 {
599 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
600
601 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
602 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
603
604 avl_create(&spa->spa_errlist_scrub,
605 spa_error_entry_compare, sizeof (spa_error_entry_t),
606 offsetof(spa_error_entry_t, se_avl));
607 avl_create(&spa->spa_errlist_last,
608 spa_error_entry_compare, sizeof (spa_error_entry_t),
609 offsetof(spa_error_entry_t, se_avl));
610 }
611
612 static taskq_t *
613 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
614 uint_t value)
615 {
616 uint_t flags = TASKQ_PREPOPULATE;
617 boolean_t batch = B_FALSE;
618
619 switch (mode) {
620 case zti_mode_null:
621 return (NULL); /* no taskq needed */
622
623 case zti_mode_fixed:
624 ASSERT3U(value, >=, 1);
625 value = MAX(value, 1);
626 break;
627
628 case zti_mode_batch:
629 batch = B_TRUE;
630 flags |= TASKQ_THREADS_CPU_PCT;
631 value = zio_taskq_batch_pct;
632 break;
633
634 case zti_mode_online_percent:
635 flags |= TASKQ_THREADS_CPU_PCT;
636 break;
637
638 default:
639 panic("unrecognized mode for %s taskq (%u:%u) in "
640 "spa_activate()",
641 name, mode, value);
642 break;
643 }
644
645 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
646 if (batch)
647 flags |= TASKQ_DC_BATCH;
648
649 return (taskq_create_sysdc(name, value, 50, INT_MAX,
650 spa->spa_proc, zio_taskq_basedc, flags));
651 }
652 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
653 spa->spa_proc, flags));
654 }
655
656 static void
657 spa_create_zio_taskqs(spa_t *spa)
658 {
659 int t, q;
660
661 for (t = 0; t < ZIO_TYPES; t++) {
662 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
663 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
664 enum zti_modes mode = ztip->zti_mode;
665 uint_t value = ztip->zti_value;
666 char name[32];
667
668 (void) snprintf(name, sizeof (name),
669 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
670
671 spa->spa_zio_taskq[t][q] =
672 spa_taskq_create(spa, name, mode, value);
673 }
674 }
675 }
676
677 #ifdef _KERNEL
678 static void
679 spa_thread(void *arg)
680 {
681 callb_cpr_t cprinfo;
682
683 spa_t *spa = arg;
684 user_t *pu = PTOU(curproc);
685
686 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
687 spa->spa_name);
688
689 ASSERT(curproc != &p0);
690 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
691 "zpool-%s", spa->spa_name);
692 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
693
694 /* bind this thread to the requested psrset */
695 if (zio_taskq_psrset_bind != PS_NONE) {
696 pool_lock();
697 mutex_enter(&cpu_lock);
698 mutex_enter(&pidlock);
699 mutex_enter(&curproc->p_lock);
700
701 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
702 0, NULL, NULL) == 0) {
703 curthread->t_bind_pset = zio_taskq_psrset_bind;
704 } else {
705 cmn_err(CE_WARN,
706 "Couldn't bind process for zfs pool \"%s\" to "
707 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
708 }
709
710 mutex_exit(&curproc->p_lock);
711 mutex_exit(&pidlock);
712 mutex_exit(&cpu_lock);
713 pool_unlock();
714 }
715
716 if (zio_taskq_sysdc) {
717 sysdc_thread_enter(curthread, 100, 0);
718 }
719
720 spa->spa_proc = curproc;
721 spa->spa_did = curthread->t_did;
722
723 spa_create_zio_taskqs(spa);
724
725 mutex_enter(&spa->spa_proc_lock);
726 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
727
728 spa->spa_proc_state = SPA_PROC_ACTIVE;
729 cv_broadcast(&spa->spa_proc_cv);
730
731 CALLB_CPR_SAFE_BEGIN(&cprinfo);
732 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
733 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
734 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
735
736 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
737 spa->spa_proc_state = SPA_PROC_GONE;
738 spa->spa_proc = &p0;
739 cv_broadcast(&spa->spa_proc_cv);
740 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
741
742 mutex_enter(&curproc->p_lock);
743 lwp_exit();
744 }
745 #endif
746
747 /*
748 * Activate an uninitialized pool.
749 */
750 static void
751 spa_activate(spa_t *spa, int mode)
752 {
753 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
754
755 spa->spa_state = POOL_STATE_ACTIVE;
756 spa->spa_mode = mode;
757
758 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
759 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
760
761 /* Try to create a covering process */
762 mutex_enter(&spa->spa_proc_lock);
763 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
764 ASSERT(spa->spa_proc == &p0);
765 spa->spa_did = 0;
766
767 /* Only create a process if we're going to be around a while. */
768 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
769 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
770 NULL, 0) == 0) {
771 spa->spa_proc_state = SPA_PROC_CREATED;
772 while (spa->spa_proc_state == SPA_PROC_CREATED) {
773 cv_wait(&spa->spa_proc_cv,
774 &spa->spa_proc_lock);
775 }
776 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
777 ASSERT(spa->spa_proc != &p0);
778 ASSERT(spa->spa_did != 0);
779 } else {
780 #ifdef _KERNEL
781 cmn_err(CE_WARN,
782 "Couldn't create process for zfs pool \"%s\"\n",
783 spa->spa_name);
784 #endif
785 }
786 }
787 mutex_exit(&spa->spa_proc_lock);
788
789 /* If we didn't create a process, we need to create our taskqs. */
790 if (spa->spa_proc == &p0) {
791 spa_create_zio_taskqs(spa);
792 }
793
794 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
795 offsetof(vdev_t, vdev_config_dirty_node));
796 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
797 offsetof(vdev_t, vdev_state_dirty_node));
798
799 txg_list_create(&spa->spa_vdev_txg_list,
800 offsetof(struct vdev, vdev_txg_node));
801
802 avl_create(&spa->spa_errlist_scrub,
803 spa_error_entry_compare, sizeof (spa_error_entry_t),
804 offsetof(spa_error_entry_t, se_avl));
805 avl_create(&spa->spa_errlist_last,
806 spa_error_entry_compare, sizeof (spa_error_entry_t),
807 offsetof(spa_error_entry_t, se_avl));
808 }
809
810 /*
811 * Opposite of spa_activate().
812 */
813 static void
814 spa_deactivate(spa_t *spa)
815 {
816 int t, q;
817
818 ASSERT(spa->spa_sync_on == B_FALSE);
819 ASSERT(spa->spa_dsl_pool == NULL);
820 ASSERT(spa->spa_root_vdev == NULL);
821 ASSERT(spa->spa_async_zio_root == NULL);
822 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
823
824 txg_list_destroy(&spa->spa_vdev_txg_list);
825
826 list_destroy(&spa->spa_config_dirty_list);
827 list_destroy(&spa->spa_state_dirty_list);
828
829 for (t = 0; t < ZIO_TYPES; t++) {
830 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
831 if (spa->spa_zio_taskq[t][q] != NULL)
832 taskq_destroy(spa->spa_zio_taskq[t][q]);
833 spa->spa_zio_taskq[t][q] = NULL;
834 }
835 }
836
837 metaslab_class_destroy(spa->spa_normal_class);
838 spa->spa_normal_class = NULL;
839
840 metaslab_class_destroy(spa->spa_log_class);
841 spa->spa_log_class = NULL;
842
843 /*
844 * If this was part of an import or the open otherwise failed, we may
845 * still have errors left in the queues. Empty them just in case.
846 */
847 spa_errlog_drain(spa);
848
849 avl_destroy(&spa->spa_errlist_scrub);
850 avl_destroy(&spa->spa_errlist_last);
851
852 spa->spa_state = POOL_STATE_UNINITIALIZED;
853
854 mutex_enter(&spa->spa_proc_lock);
855 if (spa->spa_proc_state != SPA_PROC_NONE) {
856 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
857 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
858 cv_broadcast(&spa->spa_proc_cv);
859 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
860 ASSERT(spa->spa_proc != &p0);
861 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
862 }
863 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
864 spa->spa_proc_state = SPA_PROC_NONE;
865 }
866 ASSERT(spa->spa_proc == &p0);
867 mutex_exit(&spa->spa_proc_lock);
868
869 /*
870 * We want to make sure spa_thread() has actually exited the ZFS
871 * module, so that the module can't be unloaded out from underneath
872 * it.
873 */
874 if (spa->spa_did != 0) {
875 thread_join(spa->spa_did);
876 spa->spa_did = 0;
877 }
878 }
879
880 /*
881 * Verify a pool configuration, and construct the vdev tree appropriately. This
882 * will create all the necessary vdevs in the appropriate layout, with each vdev
883 * in the CLOSED state. This will prep the pool before open/creation/import.
884 * All vdev validation is done by the vdev_alloc() routine.
885 */
886 static int
887 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
888 uint_t id, int atype)
889 {
890 nvlist_t **child;
891 uint_t children;
892 int error;
893 int c;
894
895 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
896 return (error);
897
898 if ((*vdp)->vdev_ops->vdev_op_leaf)
899 return (0);
900
901 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
902 &child, &children);
903
904 if (error == ENOENT)
905 return (0);
906
907 if (error) {
908 vdev_free(*vdp);
909 *vdp = NULL;
910 return (EINVAL);
911 }
912
913 for (c = 0; c < children; c++) {
914 vdev_t *vd;
915 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
916 atype)) != 0) {
917 vdev_free(*vdp);
918 *vdp = NULL;
919 return (error);
920 }
921 }
922
923 ASSERT(*vdp != NULL);
924
925 return (0);
926 }
927
928 /*
929 * Opposite of spa_load().
930 */
931 static void
932 spa_unload(spa_t *spa)
933 {
934 int i;
935
936 ASSERT(MUTEX_HELD(&spa_namespace_lock));
937
938 /*
939 * Stop async tasks.
940 */
941 spa_async_suspend(spa);
942
943 /*
944 * Stop syncing.
945 */
946 if (spa->spa_sync_on) {
947 txg_sync_stop(spa->spa_dsl_pool);
948 spa->spa_sync_on = B_FALSE;
949 }
950
951 /*
952 * Wait for any outstanding async I/O to complete.
953 */
954 if (spa->spa_async_zio_root != NULL) {
955 (void) zio_wait(spa->spa_async_zio_root);
956 spa->spa_async_zio_root = NULL;
957 }
958
959 bpobj_close(&spa->spa_deferred_bpobj);
960
961 /*
962 * Close the dsl pool.
963 */
964 if (spa->spa_dsl_pool) {
965 dsl_pool_close(spa->spa_dsl_pool);
966 spa->spa_dsl_pool = NULL;
967 spa->spa_meta_objset = NULL;
968 }
969
970 ddt_unload(spa);
971
972 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
973
974 /*
975 * Drop and purge level 2 cache
976 */
977 spa_l2cache_drop(spa);
978
979 /*
980 * Close all vdevs.
981 */
982 if (spa->spa_root_vdev)
983 vdev_free(spa->spa_root_vdev);
984 ASSERT(spa->spa_root_vdev == NULL);
985
986 for (i = 0; i < spa->spa_spares.sav_count; i++)
987 vdev_free(spa->spa_spares.sav_vdevs[i]);
988 if (spa->spa_spares.sav_vdevs) {
989 kmem_free(spa->spa_spares.sav_vdevs,
990 spa->spa_spares.sav_count * sizeof (void *));
991 spa->spa_spares.sav_vdevs = NULL;
992 }
993 if (spa->spa_spares.sav_config) {
994 nvlist_free(spa->spa_spares.sav_config);
995 spa->spa_spares.sav_config = NULL;
996 }
997 spa->spa_spares.sav_count = 0;
998
999 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1000 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1001 if (spa->spa_l2cache.sav_vdevs) {
1002 kmem_free(spa->spa_l2cache.sav_vdevs,
1003 spa->spa_l2cache.sav_count * sizeof (void *));
1004 spa->spa_l2cache.sav_vdevs = NULL;
1005 }
1006 if (spa->spa_l2cache.sav_config) {
1007 nvlist_free(spa->spa_l2cache.sav_config);
1008 spa->spa_l2cache.sav_config = NULL;
1009 }
1010 spa->spa_l2cache.sav_count = 0;
1011
1012 spa->spa_async_suspended = 0;
1013
1014 spa_config_exit(spa, SCL_ALL, FTAG);
1015 }
1016
1017 /*
1018 * Load (or re-load) the current list of vdevs describing the active spares for
1019 * this pool. When this is called, we have some form of basic information in
1020 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1021 * then re-generate a more complete list including status information.
1022 */
1023 static void
1024 spa_load_spares(spa_t *spa)
1025 {
1026 nvlist_t **spares;
1027 uint_t nspares;
1028 int i;
1029 vdev_t *vd, *tvd;
1030
1031 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1032
1033 /*
1034 * First, close and free any existing spare vdevs.
1035 */
1036 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1037 vd = spa->spa_spares.sav_vdevs[i];
1038
1039 /* Undo the call to spa_activate() below */
1040 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1041 B_FALSE)) != NULL && tvd->vdev_isspare)
1042 spa_spare_remove(tvd);
1043 vdev_close(vd);
1044 vdev_free(vd);
1045 }
1046
1047 if (spa->spa_spares.sav_vdevs)
1048 kmem_free(spa->spa_spares.sav_vdevs,
1049 spa->spa_spares.sav_count * sizeof (void *));
1050
1051 if (spa->spa_spares.sav_config == NULL)
1052 nspares = 0;
1053 else
1054 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1055 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1056
1057 spa->spa_spares.sav_count = (int)nspares;
1058 spa->spa_spares.sav_vdevs = NULL;
1059
1060 if (nspares == 0)
1061 return;
1062
1063 /*
1064 * Construct the array of vdevs, opening them to get status in the
1065 * process. For each spare, there is potentially two different vdev_t
1066 * structures associated with it: one in the list of spares (used only
1067 * for basic validation purposes) and one in the active vdev
1068 * configuration (if it's spared in). During this phase we open and
1069 * validate each vdev on the spare list. If the vdev also exists in the
1070 * active configuration, then we also mark this vdev as an active spare.
1071 */
1072 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1073 KM_SLEEP);
1074 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1075 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1076 VDEV_ALLOC_SPARE) == 0);
1077 ASSERT(vd != NULL);
1078
1079 spa->spa_spares.sav_vdevs[i] = vd;
1080
1081 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1082 B_FALSE)) != NULL) {
1083 if (!tvd->vdev_isspare)
1084 spa_spare_add(tvd);
1085
1086 /*
1087 * We only mark the spare active if we were successfully
1088 * able to load the vdev. Otherwise, importing a pool
1089 * with a bad active spare would result in strange
1090 * behavior, because multiple pool would think the spare
1091 * is actively in use.
1092 *
1093 * There is a vulnerability here to an equally bizarre
1094 * circumstance, where a dead active spare is later
1095 * brought back to life (onlined or otherwise). Given
1096 * the rarity of this scenario, and the extra complexity
1097 * it adds, we ignore the possibility.
1098 */
1099 if (!vdev_is_dead(tvd))
1100 spa_spare_activate(tvd);
1101 }
1102
1103 vd->vdev_top = vd;
1104 vd->vdev_aux = &spa->spa_spares;
1105
1106 if (vdev_open(vd) != 0)
1107 continue;
1108
1109 if (vdev_validate_aux(vd) == 0)
1110 spa_spare_add(vd);
1111 }
1112
1113 /*
1114 * Recompute the stashed list of spares, with status information
1115 * this time.
1116 */
1117 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1118 DATA_TYPE_NVLIST_ARRAY) == 0);
1119
1120 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1121 KM_SLEEP);
1122 for (i = 0; i < spa->spa_spares.sav_count; i++)
1123 spares[i] = vdev_config_generate(spa,
1124 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1125 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1126 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1127 for (i = 0; i < spa->spa_spares.sav_count; i++)
1128 nvlist_free(spares[i]);
1129 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1130 }
1131
1132 /*
1133 * Load (or re-load) the current list of vdevs describing the active l2cache for
1134 * this pool. When this is called, we have some form of basic information in
1135 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1136 * then re-generate a more complete list including status information.
1137 * Devices which are already active have their details maintained, and are
1138 * not re-opened.
1139 */
1140 static void
1141 spa_load_l2cache(spa_t *spa)
1142 {
1143 nvlist_t **l2cache;
1144 uint_t nl2cache;
1145 int i, j, oldnvdevs;
1146 uint64_t guid;
1147 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1148 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1149
1150 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1151
1152 if (sav->sav_config != NULL) {
1153 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1154 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1155 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1156 } else {
1157 nl2cache = 0;
1158 }
1159
1160 oldvdevs = sav->sav_vdevs;
1161 oldnvdevs = sav->sav_count;
1162 sav->sav_vdevs = NULL;
1163 sav->sav_count = 0;
1164
1165 /*
1166 * Process new nvlist of vdevs.
1167 */
1168 for (i = 0; i < nl2cache; i++) {
1169 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1170 &guid) == 0);
1171
1172 newvdevs[i] = NULL;
1173 for (j = 0; j < oldnvdevs; j++) {
1174 vd = oldvdevs[j];
1175 if (vd != NULL && guid == vd->vdev_guid) {
1176 /*
1177 * Retain previous vdev for add/remove ops.
1178 */
1179 newvdevs[i] = vd;
1180 oldvdevs[j] = NULL;
1181 break;
1182 }
1183 }
1184
1185 if (newvdevs[i] == NULL) {
1186 /*
1187 * Create new vdev
1188 */
1189 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1190 VDEV_ALLOC_L2CACHE) == 0);
1191 ASSERT(vd != NULL);
1192 newvdevs[i] = vd;
1193
1194 /*
1195 * Commit this vdev as an l2cache device,
1196 * even if it fails to open.
1197 */
1198 spa_l2cache_add(vd);
1199
1200 vd->vdev_top = vd;
1201 vd->vdev_aux = sav;
1202
1203 spa_l2cache_activate(vd);
1204
1205 if (vdev_open(vd) != 0)
1206 continue;
1207
1208 (void) vdev_validate_aux(vd);
1209
1210 if (!vdev_is_dead(vd))
1211 l2arc_add_vdev(spa, vd);
1212 }
1213 }
1214
1215 /*
1216 * Purge vdevs that were dropped
1217 */
1218 for (i = 0; i < oldnvdevs; i++) {
1219 uint64_t pool;
1220
1221 vd = oldvdevs[i];
1222 if (vd != NULL) {
1223 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1224 pool != 0ULL && l2arc_vdev_present(vd))
1225 l2arc_remove_vdev(vd);
1226 (void) vdev_close(vd);
1227 spa_l2cache_remove(vd);
1228 }
1229 }
1230
1231 if (oldvdevs)
1232 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1233
1234 if (sav->sav_config == NULL)
1235 goto out;
1236
1237 sav->sav_vdevs = newvdevs;
1238 sav->sav_count = (int)nl2cache;
1239
1240 /*
1241 * Recompute the stashed list of l2cache devices, with status
1242 * information this time.
1243 */
1244 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1245 DATA_TYPE_NVLIST_ARRAY) == 0);
1246
1247 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1248 for (i = 0; i < sav->sav_count; i++)
1249 l2cache[i] = vdev_config_generate(spa,
1250 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1251 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1252 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1253 out:
1254 for (i = 0; i < sav->sav_count; i++)
1255 nvlist_free(l2cache[i]);
1256 if (sav->sav_count)
1257 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1258 }
1259
1260 static int
1261 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1262 {
1263 dmu_buf_t *db;
1264 char *packed = NULL;
1265 size_t nvsize = 0;
1266 int error;
1267 *value = NULL;
1268
1269 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1270 nvsize = *(uint64_t *)db->db_data;
1271 dmu_buf_rele(db, FTAG);
1272
1273 packed = kmem_alloc(nvsize, KM_SLEEP);
1274 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1275 DMU_READ_PREFETCH);
1276 if (error == 0)
1277 error = nvlist_unpack(packed, nvsize, value, 0);
1278 kmem_free(packed, nvsize);
1279
1280 return (error);
1281 }
1282
1283 /*
1284 * Checks to see if the given vdev could not be opened, in which case we post a
1285 * sysevent to notify the autoreplace code that the device has been removed.
1286 */
1287 static void
1288 spa_check_removed(vdev_t *vd)
1289 {
1290 int c;
1291
1292 for (c = 0; c < vd->vdev_children; c++)
1293 spa_check_removed(vd->vdev_child[c]);
1294
1295 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1296 zfs_post_autoreplace(vd->vdev_spa, vd);
1297 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1298 }
1299 }
1300
1301 /*
1302 * Validate the current config against the MOS config
1303 */
1304 static boolean_t
1305 spa_config_valid(spa_t *spa, nvlist_t *config)
1306 {
1307 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1308 nvlist_t *nv;
1309 int c, i;
1310
1311 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1312
1313 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1314 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1315
1316 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1317
1318 /*
1319 * If we're doing a normal import, then build up any additional
1320 * diagnostic information about missing devices in this config.
1321 * We'll pass this up to the user for further processing.
1322 */
1323 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1324 nvlist_t **child, *nv;
1325 uint64_t idx = 0;
1326
1327 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1328 KM_SLEEP);
1329 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1330
1331 for (c = 0; c < rvd->vdev_children; c++) {
1332 vdev_t *tvd = rvd->vdev_child[c];
1333 vdev_t *mtvd = mrvd->vdev_child[c];
1334
1335 if (tvd->vdev_ops == &vdev_missing_ops &&
1336 mtvd->vdev_ops != &vdev_missing_ops &&
1337 mtvd->vdev_islog)
1338 child[idx++] = vdev_config_generate(spa, mtvd,
1339 B_FALSE, 0);
1340 }
1341
1342 if (idx) {
1343 VERIFY(nvlist_add_nvlist_array(nv,
1344 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1345 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1346 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1347
1348 for (i = 0; i < idx; i++)
1349 nvlist_free(child[i]);
1350 }
1351 nvlist_free(nv);
1352 kmem_free(child, rvd->vdev_children * sizeof (char **));
1353 }
1354
1355 /*
1356 * Compare the root vdev tree with the information we have
1357 * from the MOS config (mrvd). Check each top-level vdev
1358 * with the corresponding MOS config top-level (mtvd).
1359 */
1360 for (c = 0; c < rvd->vdev_children; c++) {
1361 vdev_t *tvd = rvd->vdev_child[c];
1362 vdev_t *mtvd = mrvd->vdev_child[c];
1363
1364 /*
1365 * Resolve any "missing" vdevs in the current configuration.
1366 * If we find that the MOS config has more accurate information
1367 * about the top-level vdev then use that vdev instead.
1368 */
1369 if (tvd->vdev_ops == &vdev_missing_ops &&
1370 mtvd->vdev_ops != &vdev_missing_ops) {
1371
1372 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1373 continue;
1374
1375 /*
1376 * Device specific actions.
1377 */
1378 if (mtvd->vdev_islog) {
1379 spa_set_log_state(spa, SPA_LOG_CLEAR);
1380 } else {
1381 /*
1382 * XXX - once we have 'readonly' pool
1383 * support we should be able to handle
1384 * missing data devices by transitioning
1385 * the pool to readonly.
1386 */
1387 continue;
1388 }
1389
1390 /*
1391 * Swap the missing vdev with the data we were
1392 * able to obtain from the MOS config.
1393 */
1394 vdev_remove_child(rvd, tvd);
1395 vdev_remove_child(mrvd, mtvd);
1396
1397 vdev_add_child(rvd, mtvd);
1398 vdev_add_child(mrvd, tvd);
1399
1400 spa_config_exit(spa, SCL_ALL, FTAG);
1401 vdev_load(mtvd);
1402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1403
1404 vdev_reopen(rvd);
1405 } else if (mtvd->vdev_islog) {
1406 /*
1407 * Load the slog device's state from the MOS config
1408 * since it's possible that the label does not
1409 * contain the most up-to-date information.
1410 */
1411 vdev_load_log_state(tvd, mtvd);
1412 vdev_reopen(tvd);
1413 }
1414 }
1415 vdev_free(mrvd);
1416 spa_config_exit(spa, SCL_ALL, FTAG);
1417
1418 /*
1419 * Ensure we were able to validate the config.
1420 */
1421 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1422 }
1423
1424 /*
1425 * Check for missing log devices
1426 */
1427 static int
1428 spa_check_logs(spa_t *spa)
1429 {
1430 switch (spa->spa_log_state) {
1431 default:
1432 break;
1433 case SPA_LOG_MISSING:
1434 /* need to recheck in case slog has been restored */
1435 case SPA_LOG_UNKNOWN:
1436 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1437 DS_FIND_CHILDREN)) {
1438 spa_set_log_state(spa, SPA_LOG_MISSING);
1439 return (1);
1440 }
1441 break;
1442 }
1443 return (0);
1444 }
1445
1446 static boolean_t
1447 spa_passivate_log(spa_t *spa)
1448 {
1449 vdev_t *rvd = spa->spa_root_vdev;
1450 boolean_t slog_found = B_FALSE;
1451 int c;
1452
1453 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1454
1455 if (!spa_has_slogs(spa))
1456 return (B_FALSE);
1457
1458 for (c = 0; c < rvd->vdev_children; c++) {
1459 vdev_t *tvd = rvd->vdev_child[c];
1460 metaslab_group_t *mg = tvd->vdev_mg;
1461
1462 if (tvd->vdev_islog) {
1463 metaslab_group_passivate(mg);
1464 slog_found = B_TRUE;
1465 }
1466 }
1467
1468 return (slog_found);
1469 }
1470
1471 static void
1472 spa_activate_log(spa_t *spa)
1473 {
1474 vdev_t *rvd = spa->spa_root_vdev;
1475 int c;
1476
1477 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1478
1479 for (c = 0; c < rvd->vdev_children; c++) {
1480 vdev_t *tvd = rvd->vdev_child[c];
1481 metaslab_group_t *mg = tvd->vdev_mg;
1482
1483 if (tvd->vdev_islog)
1484 metaslab_group_activate(mg);
1485 }
1486 }
1487
1488 int
1489 spa_offline_log(spa_t *spa)
1490 {
1491 int error = 0;
1492
1493 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1494 NULL, DS_FIND_CHILDREN)) == 0) {
1495
1496 /*
1497 * We successfully offlined the log device, sync out the
1498 * current txg so that the "stubby" block can be removed
1499 * by zil_sync().
1500 */
1501 txg_wait_synced(spa->spa_dsl_pool, 0);
1502 }
1503 return (error);
1504 }
1505
1506 static void
1507 spa_aux_check_removed(spa_aux_vdev_t *sav)
1508 {
1509 int i;
1510
1511 for (i = 0; i < sav->sav_count; i++)
1512 spa_check_removed(sav->sav_vdevs[i]);
1513 }
1514
1515 void
1516 spa_claim_notify(zio_t *zio)
1517 {
1518 spa_t *spa = zio->io_spa;
1519
1520 if (zio->io_error)
1521 return;
1522
1523 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1524 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1525 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1526 mutex_exit(&spa->spa_props_lock);
1527 }
1528
1529 typedef struct spa_load_error {
1530 uint64_t sle_meta_count;
1531 uint64_t sle_data_count;
1532 } spa_load_error_t;
1533
1534 static void
1535 spa_load_verify_done(zio_t *zio)
1536 {
1537 blkptr_t *bp = zio->io_bp;
1538 spa_load_error_t *sle = zio->io_private;
1539 dmu_object_type_t type = BP_GET_TYPE(bp);
1540 int error = zio->io_error;
1541
1542 if (error) {
1543 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1544 type != DMU_OT_INTENT_LOG)
1545 atomic_add_64(&sle->sle_meta_count, 1);
1546 else
1547 atomic_add_64(&sle->sle_data_count, 1);
1548 }
1549 zio_data_buf_free(zio->io_data, zio->io_size);
1550 }
1551
1552 /*ARGSUSED*/
1553 static int
1554 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1555 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1556 {
1557 if (bp != NULL) {
1558 zio_t *rio = arg;
1559 size_t size = BP_GET_PSIZE(bp);
1560 void *data = zio_data_buf_alloc(size);
1561
1562 zio_nowait(zio_read(rio, spa, bp, data, size,
1563 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1564 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1565 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1566 }
1567 return (0);
1568 }
1569
1570 static int
1571 spa_load_verify(spa_t *spa)
1572 {
1573 zio_t *rio;
1574 spa_load_error_t sle = { 0 };
1575 zpool_rewind_policy_t policy;
1576 boolean_t verify_ok = B_FALSE;
1577 int error;
1578
1579 zpool_get_rewind_policy(spa->spa_config, &policy);
1580
1581 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1582 return (0);
1583
1584 rio = zio_root(spa, NULL, &sle,
1585 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1586
1587 error = traverse_pool(spa, spa->spa_verify_min_txg,
1588 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1589
1590 (void) zio_wait(rio);
1591
1592 spa->spa_load_meta_errors = sle.sle_meta_count;
1593 spa->spa_load_data_errors = sle.sle_data_count;
1594
1595 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1596 sle.sle_data_count <= policy.zrp_maxdata) {
1597 int64_t loss = 0;
1598
1599 verify_ok = B_TRUE;
1600 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1601 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1602
1603 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1604 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1605 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1606 VERIFY(nvlist_add_int64(spa->spa_load_info,
1607 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1608 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1609 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1610 } else {
1611 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1612 }
1613
1614 if (error) {
1615 if (error != ENXIO && error != EIO)
1616 error = EIO;
1617 return (error);
1618 }
1619
1620 return (verify_ok ? 0 : EIO);
1621 }
1622
1623 /*
1624 * Find a value in the pool props object.
1625 */
1626 static void
1627 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1628 {
1629 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1630 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1631 }
1632
1633 /*
1634 * Find a value in the pool directory object.
1635 */
1636 static int
1637 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1638 {
1639 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1640 name, sizeof (uint64_t), 1, val));
1641 }
1642
1643 static int
1644 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1645 {
1646 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1647 return (err);
1648 }
1649
1650 /*
1651 * Fix up config after a partly-completed split. This is done with the
1652 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1653 * pool have that entry in their config, but only the splitting one contains
1654 * a list of all the guids of the vdevs that are being split off.
1655 *
1656 * This function determines what to do with that list: either rejoin
1657 * all the disks to the pool, or complete the splitting process. To attempt
1658 * the rejoin, each disk that is offlined is marked online again, and
1659 * we do a reopen() call. If the vdev label for every disk that was
1660 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1661 * then we call vdev_split() on each disk, and complete the split.
1662 *
1663 * Otherwise we leave the config alone, with all the vdevs in place in
1664 * the original pool.
1665 */
1666 static void
1667 spa_try_repair(spa_t *spa, nvlist_t *config)
1668 {
1669 uint_t extracted;
1670 uint64_t *glist;
1671 uint_t i, gcount;
1672 nvlist_t *nvl;
1673 vdev_t **vd;
1674 boolean_t attempt_reopen;
1675
1676 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1677 return;
1678
1679 /* check that the config is complete */
1680 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1681 &glist, &gcount) != 0)
1682 return;
1683
1684 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1685
1686 /* attempt to online all the vdevs & validate */
1687 attempt_reopen = B_TRUE;
1688 for (i = 0; i < gcount; i++) {
1689 if (glist[i] == 0) /* vdev is hole */
1690 continue;
1691
1692 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1693 if (vd[i] == NULL) {
1694 /*
1695 * Don't bother attempting to reopen the disks;
1696 * just do the split.
1697 */
1698 attempt_reopen = B_FALSE;
1699 } else {
1700 /* attempt to re-online it */
1701 vd[i]->vdev_offline = B_FALSE;
1702 }
1703 }
1704
1705 if (attempt_reopen) {
1706 vdev_reopen(spa->spa_root_vdev);
1707
1708 /* check each device to see what state it's in */
1709 for (extracted = 0, i = 0; i < gcount; i++) {
1710 if (vd[i] != NULL &&
1711 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1712 break;
1713 ++extracted;
1714 }
1715 }
1716
1717 /*
1718 * If every disk has been moved to the new pool, or if we never
1719 * even attempted to look at them, then we split them off for
1720 * good.
1721 */
1722 if (!attempt_reopen || gcount == extracted) {
1723 for (i = 0; i < gcount; i++)
1724 if (vd[i] != NULL)
1725 vdev_split(vd[i]);
1726 vdev_reopen(spa->spa_root_vdev);
1727 }
1728
1729 kmem_free(vd, gcount * sizeof (vdev_t *));
1730 }
1731
1732 static int
1733 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1734 boolean_t mosconfig)
1735 {
1736 nvlist_t *config = spa->spa_config;
1737 char *ereport = FM_EREPORT_ZFS_POOL;
1738 int error;
1739 uint64_t pool_guid;
1740 nvlist_t *nvl;
1741
1742 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1743 return (EINVAL);
1744
1745 /*
1746 * Versioning wasn't explicitly added to the label until later, so if
1747 * it's not present treat it as the initial version.
1748 */
1749 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1750 &spa->spa_ubsync.ub_version) != 0)
1751 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1752
1753 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1754 &spa->spa_config_txg);
1755
1756 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1757 spa_guid_exists(pool_guid, 0)) {
1758 error = EEXIST;
1759 } else {
1760 spa->spa_load_guid = pool_guid;
1761
1762 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1763 &nvl) == 0) {
1764 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1765 KM_SLEEP) == 0);
1766 }
1767
1768 gethrestime(&spa->spa_loaded_ts);
1769 error = spa_load_impl(spa, pool_guid, config, state, type,
1770 mosconfig, &ereport);
1771 }
1772
1773 spa->spa_minref = refcount_count(&spa->spa_refcount);
1774 if (error) {
1775 if (error != EEXIST) {
1776 spa->spa_loaded_ts.tv_sec = 0;
1777 spa->spa_loaded_ts.tv_nsec = 0;
1778 }
1779 if (error != EBADF) {
1780 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1781 }
1782 }
1783 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1784 spa->spa_ena = 0;
1785
1786 return (error);
1787 }
1788
1789 /*
1790 * Load an existing storage pool, using the pool's builtin spa_config as a
1791 * source of configuration information.
1792 */
1793 __attribute__((always_inline))
1794 static inline int
1795 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1796 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1797 char **ereport)
1798 {
1799 int error = 0;
1800 nvlist_t *nvroot = NULL;
1801 vdev_t *rvd;
1802 uberblock_t *ub = &spa->spa_uberblock;
1803 uint64_t children, config_cache_txg = spa->spa_config_txg;
1804 int orig_mode = spa->spa_mode;
1805 int parse;
1806 uint64_t obj;
1807
1808 /*
1809 * If this is an untrusted config, access the pool in read-only mode.
1810 * This prevents things like resilvering recently removed devices.
1811 */
1812 if (!mosconfig)
1813 spa->spa_mode = FREAD;
1814
1815 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1816
1817 spa->spa_load_state = state;
1818
1819 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1820 return (EINVAL);
1821
1822 parse = (type == SPA_IMPORT_EXISTING ?
1823 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1824
1825 /*
1826 * Create "The Godfather" zio to hold all async IOs
1827 */
1828 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1829 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1830
1831 /*
1832 * Parse the configuration into a vdev tree. We explicitly set the
1833 * value that will be returned by spa_version() since parsing the
1834 * configuration requires knowing the version number.
1835 */
1836 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1837 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1838 spa_config_exit(spa, SCL_ALL, FTAG);
1839
1840 if (error != 0)
1841 return (error);
1842
1843 ASSERT(spa->spa_root_vdev == rvd);
1844
1845 if (type != SPA_IMPORT_ASSEMBLE) {
1846 ASSERT(spa_guid(spa) == pool_guid);
1847 }
1848
1849 /*
1850 * Try to open all vdevs, loading each label in the process.
1851 */
1852 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1853 error = vdev_open(rvd);
1854 spa_config_exit(spa, SCL_ALL, FTAG);
1855 if (error != 0)
1856 return (error);
1857
1858 /*
1859 * We need to validate the vdev labels against the configuration that
1860 * we have in hand, which is dependent on the setting of mosconfig. If
1861 * mosconfig is true then we're validating the vdev labels based on
1862 * that config. Otherwise, we're validating against the cached config
1863 * (zpool.cache) that was read when we loaded the zfs module, and then
1864 * later we will recursively call spa_load() and validate against
1865 * the vdev config.
1866 *
1867 * If we're assembling a new pool that's been split off from an
1868 * existing pool, the labels haven't yet been updated so we skip
1869 * validation for now.
1870 */
1871 if (type != SPA_IMPORT_ASSEMBLE) {
1872 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1873 error = vdev_validate(rvd);
1874 spa_config_exit(spa, SCL_ALL, FTAG);
1875
1876 if (error != 0)
1877 return (error);
1878
1879 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1880 return (ENXIO);
1881 }
1882
1883 /*
1884 * Find the best uberblock.
1885 */
1886 vdev_uberblock_load(NULL, rvd, ub);
1887
1888 /*
1889 * If we weren't able to find a single valid uberblock, return failure.
1890 */
1891 if (ub->ub_txg == 0)
1892 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1893
1894 /*
1895 * If the pool is newer than the code, we can't open it.
1896 */
1897 if (ub->ub_version > SPA_VERSION)
1898 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1899
1900 /*
1901 * If the vdev guid sum doesn't match the uberblock, we have an
1902 * incomplete configuration. We first check to see if the pool
1903 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1904 * If it is, defer the vdev_guid_sum check till later so we
1905 * can handle missing vdevs.
1906 */
1907 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1908 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1909 rvd->vdev_guid_sum != ub->ub_guid_sum)
1910 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1911
1912 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1913 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1914 spa_try_repair(spa, config);
1915 spa_config_exit(spa, SCL_ALL, FTAG);
1916 nvlist_free(spa->spa_config_splitting);
1917 spa->spa_config_splitting = NULL;
1918 }
1919
1920 /*
1921 * Initialize internal SPA structures.
1922 */
1923 spa->spa_state = POOL_STATE_ACTIVE;
1924 spa->spa_ubsync = spa->spa_uberblock;
1925 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1926 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1927 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1928 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1929 spa->spa_claim_max_txg = spa->spa_first_txg;
1930 spa->spa_prev_software_version = ub->ub_software_version;
1931
1932 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1933 if (error)
1934 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1935 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1936
1937 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1938 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1939
1940 if (!mosconfig) {
1941 uint64_t hostid;
1942 nvlist_t *policy = NULL, *nvconfig;
1943
1944 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1945 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1946
1947 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1948 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1949 char *hostname;
1950 unsigned long myhostid = 0;
1951
1952 VERIFY(nvlist_lookup_string(nvconfig,
1953 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1954
1955 #ifdef _KERNEL
1956 myhostid = zone_get_hostid(NULL);
1957 #else /* _KERNEL */
1958 /*
1959 * We're emulating the system's hostid in userland, so
1960 * we can't use zone_get_hostid().
1961 */
1962 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1963 #endif /* _KERNEL */
1964 if (hostid != 0 && myhostid != 0 &&
1965 hostid != myhostid) {
1966 nvlist_free(nvconfig);
1967 cmn_err(CE_WARN, "pool '%s' could not be "
1968 "loaded as it was last accessed by "
1969 "another system (host: %s hostid: 0x%lx). "
1970 "See: http://www.sun.com/msg/ZFS-8000-EY",
1971 spa_name(spa), hostname,
1972 (unsigned long)hostid);
1973 return (EBADF);
1974 }
1975 }
1976 if (nvlist_lookup_nvlist(spa->spa_config,
1977 ZPOOL_REWIND_POLICY, &policy) == 0)
1978 VERIFY(nvlist_add_nvlist(nvconfig,
1979 ZPOOL_REWIND_POLICY, policy) == 0);
1980
1981 spa_config_set(spa, nvconfig);
1982 spa_unload(spa);
1983 spa_deactivate(spa);
1984 spa_activate(spa, orig_mode);
1985
1986 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1987 }
1988
1989 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1990 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1991 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1992 if (error != 0)
1993 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1994
1995 /*
1996 * Load the bit that tells us to use the new accounting function
1997 * (raid-z deflation). If we have an older pool, this will not
1998 * be present.
1999 */
2000 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2001 if (error != 0 && error != ENOENT)
2002 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2003
2004 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2005 &spa->spa_creation_version);
2006 if (error != 0 && error != ENOENT)
2007 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2008
2009 /*
2010 * Load the persistent error log. If we have an older pool, this will
2011 * not be present.
2012 */
2013 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2014 if (error != 0 && error != ENOENT)
2015 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2016
2017 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2018 &spa->spa_errlog_scrub);
2019 if (error != 0 && error != ENOENT)
2020 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2021
2022 /*
2023 * Load the history object. If we have an older pool, this
2024 * will not be present.
2025 */
2026 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2027 if (error != 0 && error != ENOENT)
2028 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2029
2030 /*
2031 * If we're assembling the pool from the split-off vdevs of
2032 * an existing pool, we don't want to attach the spares & cache
2033 * devices.
2034 */
2035
2036 /*
2037 * Load any hot spares for this pool.
2038 */
2039 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2040 if (error != 0 && error != ENOENT)
2041 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2042 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2043 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2044 if (load_nvlist(spa, spa->spa_spares.sav_object,
2045 &spa->spa_spares.sav_config) != 0)
2046 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2047
2048 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2049 spa_load_spares(spa);
2050 spa_config_exit(spa, SCL_ALL, FTAG);
2051 } else if (error == 0) {
2052 spa->spa_spares.sav_sync = B_TRUE;
2053 }
2054
2055 /*
2056 * Load any level 2 ARC devices for this pool.
2057 */
2058 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2059 &spa->spa_l2cache.sav_object);
2060 if (error != 0 && error != ENOENT)
2061 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2062 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2063 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2064 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2065 &spa->spa_l2cache.sav_config) != 0)
2066 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2067
2068 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2069 spa_load_l2cache(spa);
2070 spa_config_exit(spa, SCL_ALL, FTAG);
2071 } else if (error == 0) {
2072 spa->spa_l2cache.sav_sync = B_TRUE;
2073 }
2074
2075 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2076
2077 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2078 if (error && error != ENOENT)
2079 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2080
2081 if (error == 0) {
2082 uint64_t autoreplace;
2083
2084 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2085 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2086 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2087 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2088 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2089 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2090 &spa->spa_dedup_ditto);
2091
2092 spa->spa_autoreplace = (autoreplace != 0);
2093 }
2094
2095 /*
2096 * If the 'autoreplace' property is set, then post a resource notifying
2097 * the ZFS DE that it should not issue any faults for unopenable
2098 * devices. We also iterate over the vdevs, and post a sysevent for any
2099 * unopenable vdevs so that the normal autoreplace handler can take
2100 * over.
2101 */
2102 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2103 spa_check_removed(spa->spa_root_vdev);
2104 /*
2105 * For the import case, this is done in spa_import(), because
2106 * at this point we're using the spare definitions from
2107 * the MOS config, not necessarily from the userland config.
2108 */
2109 if (state != SPA_LOAD_IMPORT) {
2110 spa_aux_check_removed(&spa->spa_spares);
2111 spa_aux_check_removed(&spa->spa_l2cache);
2112 }
2113 }
2114
2115 /*
2116 * Load the vdev state for all toplevel vdevs.
2117 */
2118 vdev_load(rvd);
2119
2120 /*
2121 * Propagate the leaf DTLs we just loaded all the way up the tree.
2122 */
2123 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2124 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2125 spa_config_exit(spa, SCL_ALL, FTAG);
2126
2127 /*
2128 * Load the DDTs (dedup tables).
2129 */
2130 error = ddt_load(spa);
2131 if (error != 0)
2132 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2133
2134 spa_update_dspace(spa);
2135
2136 /*
2137 * Validate the config, using the MOS config to fill in any
2138 * information which might be missing. If we fail to validate
2139 * the config then declare the pool unfit for use. If we're
2140 * assembling a pool from a split, the log is not transferred
2141 * over.
2142 */
2143 if (type != SPA_IMPORT_ASSEMBLE) {
2144 nvlist_t *nvconfig;
2145
2146 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2147 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2148
2149 if (!spa_config_valid(spa, nvconfig)) {
2150 nvlist_free(nvconfig);
2151 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2152 ENXIO));
2153 }
2154 nvlist_free(nvconfig);
2155
2156 /*
2157 * Now that we've validate the config, check the state of the
2158 * root vdev. If it can't be opened, it indicates one or
2159 * more toplevel vdevs are faulted.
2160 */
2161 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162 return (ENXIO);
2163
2164 if (spa_check_logs(spa)) {
2165 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2166 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2167 }
2168 }
2169
2170 /*
2171 * We've successfully opened the pool, verify that we're ready
2172 * to start pushing transactions.
2173 */
2174 if (state != SPA_LOAD_TRYIMPORT) {
2175 if ((error = spa_load_verify(spa)))
2176 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2177 error));
2178 }
2179
2180 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2181 spa->spa_load_max_txg == UINT64_MAX)) {
2182 dmu_tx_t *tx;
2183 int need_update = B_FALSE;
2184 int c;
2185
2186 ASSERT(state != SPA_LOAD_TRYIMPORT);
2187
2188 /*
2189 * Claim log blocks that haven't been committed yet.
2190 * This must all happen in a single txg.
2191 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2192 * invoked from zil_claim_log_block()'s i/o done callback.
2193 * Price of rollback is that we abandon the log.
2194 */
2195 spa->spa_claiming = B_TRUE;
2196
2197 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2198 spa_first_txg(spa));
2199 (void) dmu_objset_find(spa_name(spa),
2200 zil_claim, tx, DS_FIND_CHILDREN);
2201 dmu_tx_commit(tx);
2202
2203 spa->spa_claiming = B_FALSE;
2204
2205 spa_set_log_state(spa, SPA_LOG_GOOD);
2206 spa->spa_sync_on = B_TRUE;
2207 txg_sync_start(spa->spa_dsl_pool);
2208
2209 /*
2210 * Wait for all claims to sync. We sync up to the highest
2211 * claimed log block birth time so that claimed log blocks
2212 * don't appear to be from the future. spa_claim_max_txg
2213 * will have been set for us by either zil_check_log_chain()
2214 * (invoked from spa_check_logs()) or zil_claim() above.
2215 */
2216 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2217
2218 /*
2219 * If the config cache is stale, or we have uninitialized
2220 * metaslabs (see spa_vdev_add()), then update the config.
2221 *
2222 * If this is a verbatim import, trust the current
2223 * in-core spa_config and update the disk labels.
2224 */
2225 if (config_cache_txg != spa->spa_config_txg ||
2226 state == SPA_LOAD_IMPORT ||
2227 state == SPA_LOAD_RECOVER ||
2228 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2229 need_update = B_TRUE;
2230
2231 for (c = 0; c < rvd->vdev_children; c++)
2232 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2233 need_update = B_TRUE;
2234
2235 /*
2236 * Update the config cache asychronously in case we're the
2237 * root pool, in which case the config cache isn't writable yet.
2238 */
2239 if (need_update)
2240 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2241
2242 /*
2243 * Check all DTLs to see if anything needs resilvering.
2244 */
2245 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2246 vdev_resilver_needed(rvd, NULL, NULL))
2247 spa_async_request(spa, SPA_ASYNC_RESILVER);
2248
2249 /*
2250 * Delete any inconsistent datasets.
2251 */
2252 (void) dmu_objset_find(spa_name(spa),
2253 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2254
2255 /*
2256 * Clean up any stale temporary dataset userrefs.
2257 */
2258 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2259 }
2260
2261 return (0);
2262 }
2263
2264 static int
2265 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2266 {
2267 int mode = spa->spa_mode;
2268
2269 spa_unload(spa);
2270 spa_deactivate(spa);
2271
2272 spa->spa_load_max_txg--;
2273
2274 spa_activate(spa, mode);
2275 spa_async_suspend(spa);
2276
2277 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2278 }
2279
2280 static int
2281 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2282 uint64_t max_request, int rewind_flags)
2283 {
2284 nvlist_t *config = NULL;
2285 int load_error, rewind_error;
2286 uint64_t safe_rewind_txg;
2287 uint64_t min_txg;
2288
2289 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2290 spa->spa_load_max_txg = spa->spa_load_txg;
2291 spa_set_log_state(spa, SPA_LOG_CLEAR);
2292 } else {
2293 spa->spa_load_max_txg = max_request;
2294 }
2295
2296 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2297 mosconfig);
2298 if (load_error == 0)
2299 return (0);
2300
2301 if (spa->spa_root_vdev != NULL)
2302 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2303
2304 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2305 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2306
2307 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2308 nvlist_free(config);
2309 return (load_error);
2310 }
2311
2312 /* Price of rolling back is discarding txgs, including log */
2313 if (state == SPA_LOAD_RECOVER)
2314 spa_set_log_state(spa, SPA_LOG_CLEAR);
2315
2316 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2317 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2318 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2319 TXG_INITIAL : safe_rewind_txg;
2320
2321 /*
2322 * Continue as long as we're finding errors, we're still within
2323 * the acceptable rewind range, and we're still finding uberblocks
2324 */
2325 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2326 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2327 if (spa->spa_load_max_txg < safe_rewind_txg)
2328 spa->spa_extreme_rewind = B_TRUE;
2329 rewind_error = spa_load_retry(spa, state, mosconfig);
2330 }
2331
2332 spa->spa_extreme_rewind = B_FALSE;
2333 spa->spa_load_max_txg = UINT64_MAX;
2334
2335 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2336 spa_config_set(spa, config);
2337
2338 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2339 }
2340
2341 /*
2342 * Pool Open/Import
2343 *
2344 * The import case is identical to an open except that the configuration is sent
2345 * down from userland, instead of grabbed from the configuration cache. For the
2346 * case of an open, the pool configuration will exist in the
2347 * POOL_STATE_UNINITIALIZED state.
2348 *
2349 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2350 * the same time open the pool, without having to keep around the spa_t in some
2351 * ambiguous state.
2352 */
2353 static int
2354 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2355 nvlist_t **config)
2356 {
2357 spa_t *spa;
2358 spa_load_state_t state = SPA_LOAD_OPEN;
2359 int error;
2360 int locked = B_FALSE;
2361
2362 *spapp = NULL;
2363
2364 /*
2365 * As disgusting as this is, we need to support recursive calls to this
2366 * function because dsl_dir_open() is called during spa_load(), and ends
2367 * up calling spa_open() again. The real fix is to figure out how to
2368 * avoid dsl_dir_open() calling this in the first place.
2369 */
2370 if (mutex_owner(&spa_namespace_lock) != curthread) {
2371 mutex_enter(&spa_namespace_lock);
2372 locked = B_TRUE;
2373 }
2374
2375 if ((spa = spa_lookup(pool)) == NULL) {
2376 if (locked)
2377 mutex_exit(&spa_namespace_lock);
2378 return (ENOENT);
2379 }
2380
2381 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2382 zpool_rewind_policy_t policy;
2383
2384 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2385 &policy);
2386 if (policy.zrp_request & ZPOOL_DO_REWIND)
2387 state = SPA_LOAD_RECOVER;
2388
2389 spa_activate(spa, spa_mode_global);
2390
2391 if (state != SPA_LOAD_RECOVER)
2392 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2393
2394 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2395 policy.zrp_request);
2396
2397 if (error == EBADF) {
2398 /*
2399 * If vdev_validate() returns failure (indicated by
2400 * EBADF), it indicates that one of the vdevs indicates
2401 * that the pool has been exported or destroyed. If
2402 * this is the case, the config cache is out of sync and
2403 * we should remove the pool from the namespace.
2404 */
2405 spa_unload(spa);
2406 spa_deactivate(spa);
2407 spa_config_sync(spa, B_TRUE, B_TRUE);
2408 spa_remove(spa);
2409 if (locked)
2410 mutex_exit(&spa_namespace_lock);
2411 return (ENOENT);
2412 }
2413
2414 if (error) {
2415 /*
2416 * We can't open the pool, but we still have useful
2417 * information: the state of each vdev after the
2418 * attempted vdev_open(). Return this to the user.
2419 */
2420 if (config != NULL && spa->spa_config) {
2421 VERIFY(nvlist_dup(spa->spa_config, config,
2422 KM_SLEEP) == 0);
2423 VERIFY(nvlist_add_nvlist(*config,
2424 ZPOOL_CONFIG_LOAD_INFO,
2425 spa->spa_load_info) == 0);
2426 }
2427 spa_unload(spa);
2428 spa_deactivate(spa);
2429 spa->spa_last_open_failed = error;
2430 if (locked)
2431 mutex_exit(&spa_namespace_lock);
2432 *spapp = NULL;
2433 return (error);
2434 }
2435 }
2436
2437 spa_open_ref(spa, tag);
2438
2439 if (config != NULL)
2440 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2441
2442 /*
2443 * If we've recovered the pool, pass back any information we
2444 * gathered while doing the load.
2445 */
2446 if (state == SPA_LOAD_RECOVER) {
2447 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2448 spa->spa_load_info) == 0);
2449 }
2450
2451 if (locked) {
2452 spa->spa_last_open_failed = 0;
2453 spa->spa_last_ubsync_txg = 0;
2454 spa->spa_load_txg = 0;
2455 mutex_exit(&spa_namespace_lock);
2456 }
2457
2458 *spapp = spa;
2459
2460 return (0);
2461 }
2462
2463 int
2464 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2465 nvlist_t **config)
2466 {
2467 return (spa_open_common(name, spapp, tag, policy, config));
2468 }
2469
2470 int
2471 spa_open(const char *name, spa_t **spapp, void *tag)
2472 {
2473 return (spa_open_common(name, spapp, tag, NULL, NULL));
2474 }
2475
2476 /*
2477 * Lookup the given spa_t, incrementing the inject count in the process,
2478 * preventing it from being exported or destroyed.
2479 */
2480 spa_t *
2481 spa_inject_addref(char *name)
2482 {
2483 spa_t *spa;
2484
2485 mutex_enter(&spa_namespace_lock);
2486 if ((spa = spa_lookup(name)) == NULL) {
2487 mutex_exit(&spa_namespace_lock);
2488 return (NULL);
2489 }
2490 spa->spa_inject_ref++;
2491 mutex_exit(&spa_namespace_lock);
2492
2493 return (spa);
2494 }
2495
2496 void
2497 spa_inject_delref(spa_t *spa)
2498 {
2499 mutex_enter(&spa_namespace_lock);
2500 spa->spa_inject_ref--;
2501 mutex_exit(&spa_namespace_lock);
2502 }
2503
2504 /*
2505 * Add spares device information to the nvlist.
2506 */
2507 static void
2508 spa_add_spares(spa_t *spa, nvlist_t *config)
2509 {
2510 nvlist_t **spares;
2511 uint_t i, nspares;
2512 nvlist_t *nvroot;
2513 uint64_t guid;
2514 vdev_stat_t *vs;
2515 uint_t vsc;
2516 uint64_t pool;
2517
2518 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2519
2520 if (spa->spa_spares.sav_count == 0)
2521 return;
2522
2523 VERIFY(nvlist_lookup_nvlist(config,
2524 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2525 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2526 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2527 if (nspares != 0) {
2528 VERIFY(nvlist_add_nvlist_array(nvroot,
2529 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2530 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2531 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2532
2533 /*
2534 * Go through and find any spares which have since been
2535 * repurposed as an active spare. If this is the case, update
2536 * their status appropriately.
2537 */
2538 for (i = 0; i < nspares; i++) {
2539 VERIFY(nvlist_lookup_uint64(spares[i],
2540 ZPOOL_CONFIG_GUID, &guid) == 0);
2541 if (spa_spare_exists(guid, &pool, NULL) &&
2542 pool != 0ULL) {
2543 VERIFY(nvlist_lookup_uint64_array(
2544 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2545 (uint64_t **)&vs, &vsc) == 0);
2546 vs->vs_state = VDEV_STATE_CANT_OPEN;
2547 vs->vs_aux = VDEV_AUX_SPARED;
2548 }
2549 }
2550 }
2551 }
2552
2553 /*
2554 * Add l2cache device information to the nvlist, including vdev stats.
2555 */
2556 static void
2557 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2558 {
2559 nvlist_t **l2cache;
2560 uint_t i, j, nl2cache;
2561 nvlist_t *nvroot;
2562 uint64_t guid;
2563 vdev_t *vd;
2564 vdev_stat_t *vs;
2565 uint_t vsc;
2566
2567 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2568
2569 if (spa->spa_l2cache.sav_count == 0)
2570 return;
2571
2572 VERIFY(nvlist_lookup_nvlist(config,
2573 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2574 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2575 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2576 if (nl2cache != 0) {
2577 VERIFY(nvlist_add_nvlist_array(nvroot,
2578 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2579 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2580 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2581
2582 /*
2583 * Update level 2 cache device stats.
2584 */
2585
2586 for (i = 0; i < nl2cache; i++) {
2587 VERIFY(nvlist_lookup_uint64(l2cache[i],
2588 ZPOOL_CONFIG_GUID, &guid) == 0);
2589
2590 vd = NULL;
2591 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2592 if (guid ==
2593 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2594 vd = spa->spa_l2cache.sav_vdevs[j];
2595 break;
2596 }
2597 }
2598 ASSERT(vd != NULL);
2599
2600 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2601 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2602 == 0);
2603 vdev_get_stats(vd, vs);
2604 }
2605 }
2606 }
2607
2608 int
2609 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2610 {
2611 int error;
2612 spa_t *spa;
2613
2614 *config = NULL;
2615 error = spa_open_common(name, &spa, FTAG, NULL, config);
2616
2617 if (spa != NULL) {
2618 /*
2619 * This still leaves a window of inconsistency where the spares
2620 * or l2cache devices could change and the config would be
2621 * self-inconsistent.
2622 */
2623 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2624
2625 if (*config != NULL) {
2626 uint64_t loadtimes[2];
2627
2628 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2629 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2630 VERIFY(nvlist_add_uint64_array(*config,
2631 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2632
2633 VERIFY(nvlist_add_uint64(*config,
2634 ZPOOL_CONFIG_ERRCOUNT,
2635 spa_get_errlog_size(spa)) == 0);
2636
2637 if (spa_suspended(spa))
2638 VERIFY(nvlist_add_uint64(*config,
2639 ZPOOL_CONFIG_SUSPENDED,
2640 spa->spa_failmode) == 0);
2641
2642 spa_add_spares(spa, *config);
2643 spa_add_l2cache(spa, *config);
2644 }
2645 }
2646
2647 /*
2648 * We want to get the alternate root even for faulted pools, so we cheat
2649 * and call spa_lookup() directly.
2650 */
2651 if (altroot) {
2652 if (spa == NULL) {
2653 mutex_enter(&spa_namespace_lock);
2654 spa = spa_lookup(name);
2655 if (spa)
2656 spa_altroot(spa, altroot, buflen);
2657 else
2658 altroot[0] = '\0';
2659 spa = NULL;
2660 mutex_exit(&spa_namespace_lock);
2661 } else {
2662 spa_altroot(spa, altroot, buflen);
2663 }
2664 }
2665
2666 if (spa != NULL) {
2667 spa_config_exit(spa, SCL_CONFIG, FTAG);
2668 spa_close(spa, FTAG);
2669 }
2670
2671 return (error);
2672 }
2673
2674 /*
2675 * Validate that the auxiliary device array is well formed. We must have an
2676 * array of nvlists, each which describes a valid leaf vdev. If this is an
2677 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2678 * specified, as long as they are well-formed.
2679 */
2680 static int
2681 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2682 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2683 vdev_labeltype_t label)
2684 {
2685 nvlist_t **dev;
2686 uint_t i, ndev;
2687 vdev_t *vd;
2688 int error;
2689
2690 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2691
2692 /*
2693 * It's acceptable to have no devs specified.
2694 */
2695 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2696 return (0);
2697
2698 if (ndev == 0)
2699 return (EINVAL);
2700
2701 /*
2702 * Make sure the pool is formatted with a version that supports this
2703 * device type.
2704 */
2705 if (spa_version(spa) < version)
2706 return (ENOTSUP);
2707
2708 /*
2709 * Set the pending device list so we correctly handle device in-use
2710 * checking.
2711 */
2712 sav->sav_pending = dev;
2713 sav->sav_npending = ndev;
2714
2715 for (i = 0; i < ndev; i++) {
2716 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2717 mode)) != 0)
2718 goto out;
2719
2720 if (!vd->vdev_ops->vdev_op_leaf) {
2721 vdev_free(vd);
2722 error = EINVAL;
2723 goto out;
2724 }
2725
2726 /*
2727 * The L2ARC currently only supports disk devices in
2728 * kernel context. For user-level testing, we allow it.
2729 */
2730 #ifdef _KERNEL
2731 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2732 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2733 error = ENOTBLK;
2734 goto out;
2735 }
2736 #endif
2737 vd->vdev_top = vd;
2738
2739 if ((error = vdev_open(vd)) == 0 &&
2740 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2741 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2742 vd->vdev_guid) == 0);
2743 }
2744
2745 vdev_free(vd);
2746
2747 if (error &&
2748 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2749 goto out;
2750 else
2751 error = 0;
2752 }
2753
2754 out:
2755 sav->sav_pending = NULL;
2756 sav->sav_npending = 0;
2757 return (error);
2758 }
2759
2760 static int
2761 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2762 {
2763 int error;
2764
2765 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2766
2767 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2768 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2769 VDEV_LABEL_SPARE)) != 0) {
2770 return (error);
2771 }
2772
2773 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2774 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2775 VDEV_LABEL_L2CACHE));
2776 }
2777
2778 static void
2779 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2780 const char *config)
2781 {
2782 int i;
2783
2784 if (sav->sav_config != NULL) {
2785 nvlist_t **olddevs;
2786 uint_t oldndevs;
2787 nvlist_t **newdevs;
2788
2789 /*
2790 * Generate new dev list by concatentating with the
2791 * current dev list.
2792 */
2793 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2794 &olddevs, &oldndevs) == 0);
2795
2796 newdevs = kmem_alloc(sizeof (void *) *
2797 (ndevs + oldndevs), KM_SLEEP);
2798 for (i = 0; i < oldndevs; i++)
2799 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2800 KM_SLEEP) == 0);
2801 for (i = 0; i < ndevs; i++)
2802 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2803 KM_SLEEP) == 0);
2804
2805 VERIFY(nvlist_remove(sav->sav_config, config,
2806 DATA_TYPE_NVLIST_ARRAY) == 0);
2807
2808 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2809 config, newdevs, ndevs + oldndevs) == 0);
2810 for (i = 0; i < oldndevs + ndevs; i++)
2811 nvlist_free(newdevs[i]);
2812 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2813 } else {
2814 /*
2815 * Generate a new dev list.
2816 */
2817 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2818 KM_SLEEP) == 0);
2819 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2820 devs, ndevs) == 0);
2821 }
2822 }
2823
2824 /*
2825 * Stop and drop level 2 ARC devices
2826 */
2827 void
2828 spa_l2cache_drop(spa_t *spa)
2829 {
2830 vdev_t *vd;
2831 int i;
2832 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2833
2834 for (i = 0; i < sav->sav_count; i++) {
2835 uint64_t pool;
2836
2837 vd = sav->sav_vdevs[i];
2838 ASSERT(vd != NULL);
2839
2840 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2841 pool != 0ULL && l2arc_vdev_present(vd))
2842 l2arc_remove_vdev(vd);
2843 if (vd->vdev_isl2cache)
2844 spa_l2cache_remove(vd);
2845 vdev_clear_stats(vd);
2846 (void) vdev_close(vd);
2847 }
2848 }
2849
2850 /*
2851 * Pool Creation
2852 */
2853 int
2854 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2855 const char *history_str, nvlist_t *zplprops)
2856 {
2857 spa_t *spa;
2858 char *altroot = NULL;
2859 vdev_t *rvd;
2860 dsl_pool_t *dp;
2861 dmu_tx_t *tx;
2862 int error = 0;
2863 uint64_t txg = TXG_INITIAL;
2864 nvlist_t **spares, **l2cache;
2865 uint_t nspares, nl2cache;
2866 uint64_t version, obj;
2867 int c;
2868
2869 /*
2870 * If this pool already exists, return failure.
2871 */
2872 mutex_enter(&spa_namespace_lock);
2873 if (spa_lookup(pool) != NULL) {
2874 mutex_exit(&spa_namespace_lock);
2875 return (EEXIST);
2876 }
2877
2878 /*
2879 * Allocate a new spa_t structure.
2880 */
2881 (void) nvlist_lookup_string(props,
2882 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2883 spa = spa_add(pool, NULL, altroot);
2884 spa_activate(spa, spa_mode_global);
2885
2886 if (props && (error = spa_prop_validate(spa, props))) {
2887 spa_deactivate(spa);
2888 spa_remove(spa);
2889 mutex_exit(&spa_namespace_lock);
2890 return (error);
2891 }
2892
2893 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2894 &version) != 0)
2895 version = SPA_VERSION;
2896 ASSERT(version <= SPA_VERSION);
2897
2898 spa->spa_first_txg = txg;
2899 spa->spa_uberblock.ub_txg = txg - 1;
2900 spa->spa_uberblock.ub_version = version;
2901 spa->spa_ubsync = spa->spa_uberblock;
2902
2903 /*
2904 * Create "The Godfather" zio to hold all async IOs
2905 */
2906 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2907 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2908
2909 /*
2910 * Create the root vdev.
2911 */
2912 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2913
2914 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2915
2916 ASSERT(error != 0 || rvd != NULL);
2917 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2918
2919 if (error == 0 && !zfs_allocatable_devs(nvroot))
2920 error = EINVAL;
2921
2922 if (error == 0 &&
2923 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2924 (error = spa_validate_aux(spa, nvroot, txg,
2925 VDEV_ALLOC_ADD)) == 0) {
2926 for (c = 0; c < rvd->vdev_children; c++) {
2927 vdev_metaslab_set_size(rvd->vdev_child[c]);
2928 vdev_expand(rvd->vdev_child[c], txg);
2929 }
2930 }
2931
2932 spa_config_exit(spa, SCL_ALL, FTAG);
2933
2934 if (error != 0) {
2935 spa_unload(spa);
2936 spa_deactivate(spa);
2937 spa_remove(spa);
2938 mutex_exit(&spa_namespace_lock);
2939 return (error);
2940 }
2941
2942 /*
2943 * Get the list of spares, if specified.
2944 */
2945 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2946 &spares, &nspares) == 0) {
2947 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2948 KM_SLEEP) == 0);
2949 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2950 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2951 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2952 spa_load_spares(spa);
2953 spa_config_exit(spa, SCL_ALL, FTAG);
2954 spa->spa_spares.sav_sync = B_TRUE;
2955 }
2956
2957 /*
2958 * Get the list of level 2 cache devices, if specified.
2959 */
2960 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2961 &l2cache, &nl2cache) == 0) {
2962 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2963 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2964 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2965 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2966 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2967 spa_load_l2cache(spa);
2968 spa_config_exit(spa, SCL_ALL, FTAG);
2969 spa->spa_l2cache.sav_sync = B_TRUE;
2970 }
2971
2972 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2973 spa->spa_meta_objset = dp->dp_meta_objset;
2974
2975 /*
2976 * Create DDTs (dedup tables).
2977 */
2978 ddt_create(spa);
2979
2980 spa_update_dspace(spa);
2981
2982 tx = dmu_tx_create_assigned(dp, txg);
2983
2984 /*
2985 * Create the pool config object.
2986 */
2987 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2988 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2989 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2990
2991 if (zap_add(spa->spa_meta_objset,
2992 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2993 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2994 cmn_err(CE_PANIC, "failed to add pool config");
2995 }
2996
2997 if (zap_add(spa->spa_meta_objset,
2998 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2999 sizeof (uint64_t), 1, &version, tx) != 0) {
3000 cmn_err(CE_PANIC, "failed to add pool version");
3001 }
3002
3003 /* Newly created pools with the right version are always deflated. */
3004 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3005 spa->spa_deflate = TRUE;
3006 if (zap_add(spa->spa_meta_objset,
3007 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3008 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3009 cmn_err(CE_PANIC, "failed to add deflate");
3010 }
3011 }
3012
3013 /*
3014 * Create the deferred-free bpobj. Turn off compression
3015 * because sync-to-convergence takes longer if the blocksize
3016 * keeps changing.
3017 */
3018 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3019 dmu_object_set_compress(spa->spa_meta_objset, obj,
3020 ZIO_COMPRESS_OFF, tx);
3021 if (zap_add(spa->spa_meta_objset,
3022 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3023 sizeof (uint64_t), 1, &obj, tx) != 0) {
3024 cmn_err(CE_PANIC, "failed to add bpobj");
3025 }
3026 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3027 spa->spa_meta_objset, obj));
3028
3029 /*
3030 * Create the pool's history object.
3031 */
3032 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3033 spa_history_create_obj(spa, tx);
3034
3035 /*
3036 * Set pool properties.
3037 */
3038 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3039 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3040 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3041 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3042
3043 if (props != NULL) {
3044 spa_configfile_set(spa, props, B_FALSE);
3045 spa_sync_props(spa, props, tx);
3046 }
3047
3048 dmu_tx_commit(tx);
3049
3050 spa->spa_sync_on = B_TRUE;
3051 txg_sync_start(spa->spa_dsl_pool);
3052
3053 /*
3054 * We explicitly wait for the first transaction to complete so that our
3055 * bean counters are appropriately updated.
3056 */
3057 txg_wait_synced(spa->spa_dsl_pool, txg);
3058
3059 spa_config_sync(spa, B_FALSE, B_TRUE);
3060
3061 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3062 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3063 spa_history_log_version(spa, LOG_POOL_CREATE);
3064
3065 spa->spa_minref = refcount_count(&spa->spa_refcount);
3066
3067 mutex_exit(&spa_namespace_lock);
3068
3069 return (0);
3070 }
3071
3072 #ifdef _KERNEL
3073 /*
3074 * Get the root pool information from the root disk, then import the root pool
3075 * during the system boot up time.
3076 */
3077 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3078
3079 static nvlist_t *
3080 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3081 {
3082 nvlist_t *config;
3083 nvlist_t *nvtop, *nvroot;
3084 uint64_t pgid;
3085
3086 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3087 return (NULL);
3088
3089 /*
3090 * Add this top-level vdev to the child array.
3091 */
3092 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3093 &nvtop) == 0);
3094 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3095 &pgid) == 0);
3096 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3097
3098 /*
3099 * Put this pool's top-level vdevs into a root vdev.
3100 */
3101 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3102 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3103 VDEV_TYPE_ROOT) == 0);
3104 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3105 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3106 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3107 &nvtop, 1) == 0);
3108
3109 /*
3110 * Replace the existing vdev_tree with the new root vdev in
3111 * this pool's configuration (remove the old, add the new).
3112 */
3113 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3114 nvlist_free(nvroot);
3115 return (config);
3116 }
3117
3118 /*
3119 * Walk the vdev tree and see if we can find a device with "better"
3120 * configuration. A configuration is "better" if the label on that
3121 * device has a more recent txg.
3122 */
3123 static void
3124 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3125 {
3126 int c;
3127
3128 for (c = 0; c < vd->vdev_children; c++)
3129 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3130
3131 if (vd->vdev_ops->vdev_op_leaf) {
3132 nvlist_t *label;
3133 uint64_t label_txg;
3134
3135 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3136 &label) != 0)
3137 return;
3138
3139 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3140 &label_txg) == 0);
3141
3142 /*
3143 * Do we have a better boot device?
3144 */
3145 if (label_txg > *txg) {
3146 *txg = label_txg;
3147 *avd = vd;
3148 }
3149 nvlist_free(label);
3150 }
3151 }
3152
3153 /*
3154 * Import a root pool.
3155 *
3156 * For x86. devpath_list will consist of devid and/or physpath name of
3157 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3158 * The GRUB "findroot" command will return the vdev we should boot.
3159 *
3160 * For Sparc, devpath_list consists the physpath name of the booting device
3161 * no matter the rootpool is a single device pool or a mirrored pool.
3162 * e.g.
3163 * "/pci@1f,0/ide@d/disk@0,0:a"
3164 */
3165 int
3166 spa_import_rootpool(char *devpath, char *devid)
3167 {
3168 spa_t *spa;
3169 vdev_t *rvd, *bvd, *avd = NULL;
3170 nvlist_t *config, *nvtop;
3171 uint64_t guid, txg;
3172 char *pname;
3173 int error;
3174
3175 /*
3176 * Read the label from the boot device and generate a configuration.
3177 */
3178 config = spa_generate_rootconf(devpath, devid, &guid);
3179 #if defined(_OBP) && defined(_KERNEL)
3180 if (config == NULL) {
3181 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3182 /* iscsi boot */
3183 get_iscsi_bootpath_phy(devpath);
3184 config = spa_generate_rootconf(devpath, devid, &guid);
3185 }
3186 }
3187 #endif
3188 if (config == NULL) {
3189 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3190 devpath);
3191 return (EIO);
3192 }
3193
3194 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3195 &pname) == 0);
3196 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3197
3198 mutex_enter(&spa_namespace_lock);
3199 if ((spa = spa_lookup(pname)) != NULL) {
3200 /*
3201 * Remove the existing root pool from the namespace so that we
3202 * can replace it with the correct config we just read in.
3203 */
3204 spa_remove(spa);
3205 }
3206
3207 spa = spa_add(pname, config, NULL);
3208 spa->spa_is_root = B_TRUE;
3209 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3210
3211 /*
3212 * Build up a vdev tree based on the boot device's label config.
3213 */
3214 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3215 &nvtop) == 0);
3216 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3217 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3218 VDEV_ALLOC_ROOTPOOL);
3219 spa_config_exit(spa, SCL_ALL, FTAG);
3220 if (error) {
3221 mutex_exit(&spa_namespace_lock);
3222 nvlist_free(config);
3223 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3224 pname);
3225 return (error);
3226 }
3227
3228 /*
3229 * Get the boot vdev.
3230 */
3231 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3232 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3233 (u_longlong_t)guid);
3234 error = ENOENT;
3235 goto out;
3236 }
3237
3238 /*
3239 * Determine if there is a better boot device.
3240 */
3241 avd = bvd;
3242 spa_alt_rootvdev(rvd, &avd, &txg);
3243 if (avd != bvd) {
3244 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3245 "try booting from '%s'", avd->vdev_path);
3246 error = EINVAL;
3247 goto out;
3248 }
3249
3250 /*
3251 * If the boot device is part of a spare vdev then ensure that
3252 * we're booting off the active spare.
3253 */
3254 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3255 !bvd->vdev_isspare) {
3256 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3257 "try booting from '%s'",
3258 bvd->vdev_parent->
3259 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3260 error = EINVAL;
3261 goto out;
3262 }
3263
3264 error = 0;
3265 spa_history_log_version(spa, LOG_POOL_IMPORT);
3266 out:
3267 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3268 vdev_free(rvd);
3269 spa_config_exit(spa, SCL_ALL, FTAG);
3270 mutex_exit(&spa_namespace_lock);
3271
3272 nvlist_free(config);
3273 return (error);
3274 }
3275
3276 #endif
3277
3278 /*
3279 * Import a non-root pool into the system.
3280 */
3281 int
3282 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3283 {
3284 spa_t *spa;
3285 char *altroot = NULL;
3286 spa_load_state_t state = SPA_LOAD_IMPORT;
3287 zpool_rewind_policy_t policy;
3288 uint64_t mode = spa_mode_global;
3289 uint64_t readonly = B_FALSE;
3290 int error;
3291 nvlist_t *nvroot;
3292 nvlist_t **spares, **l2cache;
3293 uint_t nspares, nl2cache;
3294
3295 /*
3296 * If a pool with this name exists, return failure.
3297 */
3298 mutex_enter(&spa_namespace_lock);
3299 if (spa_lookup(pool) != NULL) {
3300 mutex_exit(&spa_namespace_lock);
3301 return (EEXIST);
3302 }
3303
3304 /*
3305 * Create and initialize the spa structure.
3306 */
3307 (void) nvlist_lookup_string(props,
3308 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3309 (void) nvlist_lookup_uint64(props,
3310 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3311 if (readonly)
3312 mode = FREAD;
3313 spa = spa_add(pool, config, altroot);
3314 spa->spa_import_flags = flags;
3315
3316 /*
3317 * Verbatim import - Take a pool and insert it into the namespace
3318 * as if it had been loaded at boot.
3319 */
3320 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3321 if (props != NULL)
3322 spa_configfile_set(spa, props, B_FALSE);
3323
3324 spa_config_sync(spa, B_FALSE, B_TRUE);
3325
3326 mutex_exit(&spa_namespace_lock);
3327 spa_history_log_version(spa, LOG_POOL_IMPORT);
3328
3329 return (0);
3330 }
3331
3332 spa_activate(spa, mode);
3333
3334 /*
3335 * Don't start async tasks until we know everything is healthy.
3336 */
3337 spa_async_suspend(spa);
3338
3339 zpool_get_rewind_policy(config, &policy);
3340 if (policy.zrp_request & ZPOOL_DO_REWIND)
3341 state = SPA_LOAD_RECOVER;
3342
3343 /*
3344 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3345 * because the user-supplied config is actually the one to trust when
3346 * doing an import.
3347 */
3348 if (state != SPA_LOAD_RECOVER)
3349 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3350
3351 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3352 policy.zrp_request);
3353
3354 /*
3355 * Propagate anything learned while loading the pool and pass it
3356 * back to caller (i.e. rewind info, missing devices, etc).
3357 */
3358 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3359 spa->spa_load_info) == 0);
3360
3361 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3362 /*
3363 * Toss any existing sparelist, as it doesn't have any validity
3364 * anymore, and conflicts with spa_has_spare().
3365 */
3366 if (spa->spa_spares.sav_config) {
3367 nvlist_free(spa->spa_spares.sav_config);
3368 spa->spa_spares.sav_config = NULL;
3369 spa_load_spares(spa);
3370 }
3371 if (spa->spa_l2cache.sav_config) {
3372 nvlist_free(spa->spa_l2cache.sav_config);
3373 spa->spa_l2cache.sav_config = NULL;
3374 spa_load_l2cache(spa);
3375 }
3376
3377 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3378 &nvroot) == 0);
3379 if (error == 0)
3380 error = spa_validate_aux(spa, nvroot, -1ULL,
3381 VDEV_ALLOC_SPARE);
3382 if (error == 0)
3383 error = spa_validate_aux(spa, nvroot, -1ULL,
3384 VDEV_ALLOC_L2CACHE);
3385 spa_config_exit(spa, SCL_ALL, FTAG);
3386
3387 if (props != NULL)
3388 spa_configfile_set(spa, props, B_FALSE);
3389
3390 if (error != 0 || (props && spa_writeable(spa) &&
3391 (error = spa_prop_set(spa, props)))) {
3392 spa_unload(spa);
3393 spa_deactivate(spa);
3394 spa_remove(spa);
3395 mutex_exit(&spa_namespace_lock);
3396 return (error);
3397 }
3398
3399 spa_async_resume(spa);
3400
3401 /*
3402 * Override any spares and level 2 cache devices as specified by
3403 * the user, as these may have correct device names/devids, etc.
3404 */
3405 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3406 &spares, &nspares) == 0) {
3407 if (spa->spa_spares.sav_config)
3408 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3409 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3410 else
3411 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3412 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3413 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3414 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3415 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3416 spa_load_spares(spa);
3417 spa_config_exit(spa, SCL_ALL, FTAG);
3418 spa->spa_spares.sav_sync = B_TRUE;
3419 }
3420 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3421 &l2cache, &nl2cache) == 0) {
3422 if (spa->spa_l2cache.sav_config)
3423 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3424 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3425 else
3426 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3427 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3428 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3429 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3430 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3431 spa_load_l2cache(spa);
3432 spa_config_exit(spa, SCL_ALL, FTAG);
3433 spa->spa_l2cache.sav_sync = B_TRUE;
3434 }
3435
3436 /*
3437 * Check for any removed devices.
3438 */
3439 if (spa->spa_autoreplace) {
3440 spa_aux_check_removed(&spa->spa_spares);
3441 spa_aux_check_removed(&spa->spa_l2cache);
3442 }
3443
3444 if (spa_writeable(spa)) {
3445 /*
3446 * Update the config cache to include the newly-imported pool.
3447 */
3448 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3449 }
3450
3451 /*
3452 * It's possible that the pool was expanded while it was exported.
3453 * We kick off an async task to handle this for us.
3454 */
3455 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3456
3457 mutex_exit(&spa_namespace_lock);
3458 spa_history_log_version(spa, LOG_POOL_IMPORT);
3459
3460 return (0);
3461 }
3462
3463 nvlist_t *
3464 spa_tryimport(nvlist_t *tryconfig)
3465 {
3466 nvlist_t *config = NULL;
3467 char *poolname;
3468 spa_t *spa;
3469 uint64_t state;
3470 int error;
3471
3472 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3473 return (NULL);
3474
3475 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3476 return (NULL);
3477
3478 /*
3479 * Create and initialize the spa structure.
3480 */
3481 mutex_enter(&spa_namespace_lock);
3482 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3483 spa_activate(spa, FREAD);
3484
3485 /*
3486 * Pass off the heavy lifting to spa_load().
3487 * Pass TRUE for mosconfig because the user-supplied config
3488 * is actually the one to trust when doing an import.
3489 */
3490 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3491
3492 /*
3493 * If 'tryconfig' was at least parsable, return the current config.
3494 */
3495 if (spa->spa_root_vdev != NULL) {
3496 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3497 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3498 poolname) == 0);
3499 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3500 state) == 0);
3501 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3502 spa->spa_uberblock.ub_timestamp) == 0);
3503
3504 /*
3505 * If the bootfs property exists on this pool then we
3506 * copy it out so that external consumers can tell which
3507 * pools are bootable.
3508 */
3509 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3510 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3511
3512 /*
3513 * We have to play games with the name since the
3514 * pool was opened as TRYIMPORT_NAME.
3515 */
3516 if (dsl_dsobj_to_dsname(spa_name(spa),
3517 spa->spa_bootfs, tmpname) == 0) {
3518 char *cp;
3519 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3520
3521 cp = strchr(tmpname, '/');
3522 if (cp == NULL) {
3523 (void) strlcpy(dsname, tmpname,
3524 MAXPATHLEN);
3525 } else {
3526 (void) snprintf(dsname, MAXPATHLEN,
3527 "%s/%s", poolname, ++cp);
3528 }
3529 VERIFY(nvlist_add_string(config,
3530 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3531 kmem_free(dsname, MAXPATHLEN);
3532 }
3533 kmem_free(tmpname, MAXPATHLEN);
3534 }
3535
3536 /*
3537 * Add the list of hot spares and level 2 cache devices.
3538 */
3539 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3540 spa_add_spares(spa, config);
3541 spa_add_l2cache(spa, config);
3542 spa_config_exit(spa, SCL_CONFIG, FTAG);
3543 }
3544
3545 spa_unload(spa);
3546 spa_deactivate(spa);
3547 spa_remove(spa);
3548 mutex_exit(&spa_namespace_lock);
3549
3550 return (config);
3551 }
3552
3553 /*
3554 * Pool export/destroy
3555 *
3556 * The act of destroying or exporting a pool is very simple. We make sure there
3557 * is no more pending I/O and any references to the pool are gone. Then, we
3558 * update the pool state and sync all the labels to disk, removing the
3559 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3560 * we don't sync the labels or remove the configuration cache.
3561 */
3562 static int
3563 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3564 boolean_t force, boolean_t hardforce)
3565 {
3566 spa_t *spa;
3567
3568 if (oldconfig)
3569 *oldconfig = NULL;
3570
3571 if (!(spa_mode_global & FWRITE))
3572 return (EROFS);
3573
3574 mutex_enter(&spa_namespace_lock);
3575 if ((spa = spa_lookup(pool)) == NULL) {
3576 mutex_exit(&spa_namespace_lock);
3577 return (ENOENT);
3578 }
3579
3580 /*
3581 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3582 * reacquire the namespace lock, and see if we can export.
3583 */
3584 spa_open_ref(spa, FTAG);
3585 mutex_exit(&spa_namespace_lock);
3586 spa_async_suspend(spa);
3587 mutex_enter(&spa_namespace_lock);
3588 spa_close(spa, FTAG);
3589
3590 /*
3591 * The pool will be in core if it's openable,
3592 * in which case we can modify its state.
3593 */
3594 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3595 /*
3596 * Objsets may be open only because they're dirty, so we
3597 * have to force it to sync before checking spa_refcnt.
3598 */
3599 txg_wait_synced(spa->spa_dsl_pool, 0);
3600
3601 /*
3602 * A pool cannot be exported or destroyed if there are active
3603 * references. If we are resetting a pool, allow references by
3604 * fault injection handlers.
3605 */
3606 if (!spa_refcount_zero(spa) ||
3607 (spa->spa_inject_ref != 0 &&
3608 new_state != POOL_STATE_UNINITIALIZED)) {
3609 spa_async_resume(spa);
3610 mutex_exit(&spa_namespace_lock);
3611 return (EBUSY);
3612 }
3613
3614 /*
3615 * A pool cannot be exported if it has an active shared spare.
3616 * This is to prevent other pools stealing the active spare
3617 * from an exported pool. At user's own will, such pool can
3618 * be forcedly exported.
3619 */
3620 if (!force && new_state == POOL_STATE_EXPORTED &&
3621 spa_has_active_shared_spare(spa)) {
3622 spa_async_resume(spa);
3623 mutex_exit(&spa_namespace_lock);
3624 return (EXDEV);
3625 }
3626
3627 /*
3628 * We want this to be reflected on every label,
3629 * so mark them all dirty. spa_unload() will do the
3630 * final sync that pushes these changes out.
3631 */
3632 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3633 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3634 spa->spa_state = new_state;
3635 spa->spa_final_txg = spa_last_synced_txg(spa) +
3636 TXG_DEFER_SIZE + 1;
3637 vdev_config_dirty(spa->spa_root_vdev);
3638 spa_config_exit(spa, SCL_ALL, FTAG);
3639 }
3640 }
3641
3642 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3643
3644 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3645 spa_unload(spa);
3646 spa_deactivate(spa);
3647 }
3648
3649 if (oldconfig && spa->spa_config)
3650 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3651
3652 if (new_state != POOL_STATE_UNINITIALIZED) {
3653 if (!hardforce)
3654 spa_config_sync(spa, B_TRUE, B_TRUE);
3655 spa_remove(spa);
3656 }
3657 mutex_exit(&spa_namespace_lock);
3658
3659 return (0);
3660 }
3661
3662 /*
3663 * Destroy a storage pool.
3664 */
3665 int
3666 spa_destroy(char *pool)
3667 {
3668 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3669 B_FALSE, B_FALSE));
3670 }
3671
3672 /*
3673 * Export a storage pool.
3674 */
3675 int
3676 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3677 boolean_t hardforce)
3678 {
3679 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3680 force, hardforce));
3681 }
3682
3683 /*
3684 * Similar to spa_export(), this unloads the spa_t without actually removing it
3685 * from the namespace in any way.
3686 */
3687 int
3688 spa_reset(char *pool)
3689 {
3690 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3691 B_FALSE, B_FALSE));
3692 }
3693
3694 /*
3695 * ==========================================================================
3696 * Device manipulation
3697 * ==========================================================================
3698 */
3699
3700 /*
3701 * Add a device to a storage pool.
3702 */
3703 int
3704 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3705 {
3706 uint64_t txg, id;
3707 int error;
3708 vdev_t *rvd = spa->spa_root_vdev;
3709 vdev_t *vd, *tvd;
3710 nvlist_t **spares, **l2cache;
3711 uint_t nspares, nl2cache;
3712 int c;
3713
3714 ASSERT(spa_writeable(spa));
3715
3716 txg = spa_vdev_enter(spa);
3717
3718 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3719 VDEV_ALLOC_ADD)) != 0)
3720 return (spa_vdev_exit(spa, NULL, txg, error));
3721
3722 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3723
3724 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3725 &nspares) != 0)
3726 nspares = 0;
3727
3728 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3729 &nl2cache) != 0)
3730 nl2cache = 0;
3731
3732 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3733 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3734
3735 if (vd->vdev_children != 0 &&
3736 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3737 return (spa_vdev_exit(spa, vd, txg, error));
3738
3739 /*
3740 * We must validate the spares and l2cache devices after checking the
3741 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3742 */
3743 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3744 return (spa_vdev_exit(spa, vd, txg, error));
3745
3746 /*
3747 * Transfer each new top-level vdev from vd to rvd.
3748 */
3749 for (c = 0; c < vd->vdev_children; c++) {
3750
3751 /*
3752 * Set the vdev id to the first hole, if one exists.
3753 */
3754 for (id = 0; id < rvd->vdev_children; id++) {
3755 if (rvd->vdev_child[id]->vdev_ishole) {
3756 vdev_free(rvd->vdev_child[id]);
3757 break;
3758 }
3759 }
3760 tvd = vd->vdev_child[c];
3761 vdev_remove_child(vd, tvd);
3762 tvd->vdev_id = id;
3763 vdev_add_child(rvd, tvd);
3764 vdev_config_dirty(tvd);
3765 }
3766
3767 if (nspares != 0) {
3768 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3769 ZPOOL_CONFIG_SPARES);
3770 spa_load_spares(spa);
3771 spa->spa_spares.sav_sync = B_TRUE;
3772 }
3773
3774 if (nl2cache != 0) {
3775 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3776 ZPOOL_CONFIG_L2CACHE);
3777 spa_load_l2cache(spa);
3778 spa->spa_l2cache.sav_sync = B_TRUE;
3779 }
3780
3781 /*
3782 * We have to be careful when adding new vdevs to an existing pool.
3783 * If other threads start allocating from these vdevs before we
3784 * sync the config cache, and we lose power, then upon reboot we may
3785 * fail to open the pool because there are DVAs that the config cache
3786 * can't translate. Therefore, we first add the vdevs without
3787 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3788 * and then let spa_config_update() initialize the new metaslabs.
3789 *
3790 * spa_load() checks for added-but-not-initialized vdevs, so that
3791 * if we lose power at any point in this sequence, the remaining
3792 * steps will be completed the next time we load the pool.
3793 */
3794 (void) spa_vdev_exit(spa, vd, txg, 0);
3795
3796 mutex_enter(&spa_namespace_lock);
3797 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3798 mutex_exit(&spa_namespace_lock);
3799
3800 return (0);
3801 }
3802
3803 /*
3804 * Attach a device to a mirror. The arguments are the path to any device
3805 * in the mirror, and the nvroot for the new device. If the path specifies
3806 * a device that is not mirrored, we automatically insert the mirror vdev.
3807 *
3808 * If 'replacing' is specified, the new device is intended to replace the
3809 * existing device; in this case the two devices are made into their own
3810 * mirror using the 'replacing' vdev, which is functionally identical to
3811 * the mirror vdev (it actually reuses all the same ops) but has a few
3812 * extra rules: you can't attach to it after it's been created, and upon
3813 * completion of resilvering, the first disk (the one being replaced)
3814 * is automatically detached.
3815 */
3816 int
3817 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3818 {
3819 uint64_t txg, dtl_max_txg;
3820 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3821 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3822 vdev_ops_t *pvops;
3823 char *oldvdpath, *newvdpath;
3824 int newvd_isspare;
3825 int error;
3826
3827 ASSERT(spa_writeable(spa));
3828
3829 txg = spa_vdev_enter(spa);
3830
3831 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3832
3833 if (oldvd == NULL)
3834 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3835
3836 if (!oldvd->vdev_ops->vdev_op_leaf)
3837 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3838
3839 pvd = oldvd->vdev_parent;
3840
3841 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3842 VDEV_ALLOC_ADD)) != 0)
3843 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3844
3845 if (newrootvd->vdev_children != 1)
3846 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3847
3848 newvd = newrootvd->vdev_child[0];
3849
3850 if (!newvd->vdev_ops->vdev_op_leaf)
3851 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3852
3853 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3854 return (spa_vdev_exit(spa, newrootvd, txg, error));
3855
3856 /*
3857 * Spares can't replace logs
3858 */
3859 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3860 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3861
3862 if (!replacing) {
3863 /*
3864 * For attach, the only allowable parent is a mirror or the root
3865 * vdev.
3866 */
3867 if (pvd->vdev_ops != &vdev_mirror_ops &&
3868 pvd->vdev_ops != &vdev_root_ops)
3869 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3870
3871 pvops = &vdev_mirror_ops;
3872 } else {
3873 /*
3874 * Active hot spares can only be replaced by inactive hot
3875 * spares.
3876 */
3877 if (pvd->vdev_ops == &vdev_spare_ops &&
3878 oldvd->vdev_isspare &&
3879 !spa_has_spare(spa, newvd->vdev_guid))
3880 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3881
3882 /*
3883 * If the source is a hot spare, and the parent isn't already a
3884 * spare, then we want to create a new hot spare. Otherwise, we
3885 * want to create a replacing vdev. The user is not allowed to
3886 * attach to a spared vdev child unless the 'isspare' state is
3887 * the same (spare replaces spare, non-spare replaces
3888 * non-spare).
3889 */
3890 if (pvd->vdev_ops == &vdev_replacing_ops &&
3891 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3892 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3893 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3894 newvd->vdev_isspare != oldvd->vdev_isspare) {
3895 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3896 }
3897
3898 if (newvd->vdev_isspare)
3899 pvops = &vdev_spare_ops;
3900 else
3901 pvops = &vdev_replacing_ops;
3902 }
3903
3904 /*
3905 * Make sure the new device is big enough.
3906 */
3907 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3908 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3909
3910 /*
3911 * The new device cannot have a higher alignment requirement
3912 * than the top-level vdev.
3913 */
3914 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3915 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3916
3917 /*
3918 * If this is an in-place replacement, update oldvd's path and devid
3919 * to make it distinguishable from newvd, and unopenable from now on.
3920 */
3921 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3922 spa_strfree(oldvd->vdev_path);
3923 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3924 KM_SLEEP);
3925 (void) sprintf(oldvd->vdev_path, "%s/%s",
3926 newvd->vdev_path, "old");
3927 if (oldvd->vdev_devid != NULL) {
3928 spa_strfree(oldvd->vdev_devid);
3929 oldvd->vdev_devid = NULL;
3930 }
3931 }
3932
3933 /* mark the device being resilvered */
3934 newvd->vdev_resilvering = B_TRUE;
3935
3936 /*
3937 * If the parent is not a mirror, or if we're replacing, insert the new
3938 * mirror/replacing/spare vdev above oldvd.
3939 */
3940 if (pvd->vdev_ops != pvops)
3941 pvd = vdev_add_parent(oldvd, pvops);
3942
3943 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3944 ASSERT(pvd->vdev_ops == pvops);
3945 ASSERT(oldvd->vdev_parent == pvd);
3946
3947 /*
3948 * Extract the new device from its root and add it to pvd.
3949 */
3950 vdev_remove_child(newrootvd, newvd);
3951 newvd->vdev_id = pvd->vdev_children;
3952 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3953 vdev_add_child(pvd, newvd);
3954
3955 tvd = newvd->vdev_top;
3956 ASSERT(pvd->vdev_top == tvd);
3957 ASSERT(tvd->vdev_parent == rvd);
3958
3959 vdev_config_dirty(tvd);
3960
3961 /*
3962 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3963 * for any dmu_sync-ed blocks. It will propagate upward when
3964 * spa_vdev_exit() calls vdev_dtl_reassess().
3965 */
3966 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3967
3968 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3969 dtl_max_txg - TXG_INITIAL);
3970
3971 if (newvd->vdev_isspare) {
3972 spa_spare_activate(newvd);
3973 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3974 }
3975
3976 oldvdpath = spa_strdup(oldvd->vdev_path);
3977 newvdpath = spa_strdup(newvd->vdev_path);
3978 newvd_isspare = newvd->vdev_isspare;
3979
3980 /*
3981 * Mark newvd's DTL dirty in this txg.
3982 */
3983 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3984
3985 /*
3986 * Restart the resilver
3987 */
3988 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3989
3990 /*
3991 * Commit the config
3992 */
3993 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3994
3995 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3996 "%s vdev=%s %s vdev=%s",
3997 replacing && newvd_isspare ? "spare in" :
3998 replacing ? "replace" : "attach", newvdpath,
3999 replacing ? "for" : "to", oldvdpath);
4000
4001 spa_strfree(oldvdpath);
4002 spa_strfree(newvdpath);
4003
4004 if (spa->spa_bootfs)
4005 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4006
4007 return (0);
4008 }
4009
4010 /*
4011 * Detach a device from a mirror or replacing vdev.
4012 * If 'replace_done' is specified, only detach if the parent
4013 * is a replacing vdev.
4014 */
4015 int
4016 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4017 {
4018 uint64_t txg;
4019 int error;
4020 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4021 vdev_t *vd, *pvd, *cvd, *tvd;
4022 boolean_t unspare = B_FALSE;
4023 uint64_t unspare_guid = 0;
4024 char *vdpath;
4025 int c, t;
4026
4027 ASSERT(spa_writeable(spa));
4028
4029 txg = spa_vdev_enter(spa);
4030
4031 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4032
4033 if (vd == NULL)
4034 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4035
4036 if (!vd->vdev_ops->vdev_op_leaf)
4037 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4038
4039 pvd = vd->vdev_parent;
4040
4041 /*
4042 * If the parent/child relationship is not as expected, don't do it.
4043 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4044 * vdev that's replacing B with C. The user's intent in replacing
4045 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4046 * the replace by detaching C, the expected behavior is to end up
4047 * M(A,B). But suppose that right after deciding to detach C,
4048 * the replacement of B completes. We would have M(A,C), and then
4049 * ask to detach C, which would leave us with just A -- not what
4050 * the user wanted. To prevent this, we make sure that the
4051 * parent/child relationship hasn't changed -- in this example,
4052 * that C's parent is still the replacing vdev R.
4053 */
4054 if (pvd->vdev_guid != pguid && pguid != 0)
4055 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4056
4057 /*
4058 * Only 'replacing' or 'spare' vdevs can be replaced.
4059 */
4060 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4061 pvd->vdev_ops != &vdev_spare_ops)
4062 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4063
4064 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4065 spa_version(spa) >= SPA_VERSION_SPARES);
4066
4067 /*
4068 * Only mirror, replacing, and spare vdevs support detach.
4069 */
4070 if (pvd->vdev_ops != &vdev_replacing_ops &&
4071 pvd->vdev_ops != &vdev_mirror_ops &&
4072 pvd->vdev_ops != &vdev_spare_ops)
4073 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4074
4075 /*
4076 * If this device has the only valid copy of some data,
4077 * we cannot safely detach it.
4078 */
4079 if (vdev_dtl_required(vd))
4080 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4081
4082 ASSERT(pvd->vdev_children >= 2);
4083
4084 /*
4085 * If we are detaching the second disk from a replacing vdev, then
4086 * check to see if we changed the original vdev's path to have "/old"
4087 * at the end in spa_vdev_attach(). If so, undo that change now.
4088 */
4089 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4090 vd->vdev_path != NULL) {
4091 size_t len = strlen(vd->vdev_path);
4092
4093 for (c = 0; c < pvd->vdev_children; c++) {
4094 cvd = pvd->vdev_child[c];
4095
4096 if (cvd == vd || cvd->vdev_path == NULL)
4097 continue;
4098
4099 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4100 strcmp(cvd->vdev_path + len, "/old") == 0) {
4101 spa_strfree(cvd->vdev_path);
4102 cvd->vdev_path = spa_strdup(vd->vdev_path);
4103 break;
4104 }
4105 }
4106 }
4107
4108 /*
4109 * If we are detaching the original disk from a spare, then it implies
4110 * that the spare should become a real disk, and be removed from the
4111 * active spare list for the pool.
4112 */
4113 if (pvd->vdev_ops == &vdev_spare_ops &&
4114 vd->vdev_id == 0 &&
4115 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4116 unspare = B_TRUE;
4117
4118 /*
4119 * Erase the disk labels so the disk can be used for other things.
4120 * This must be done after all other error cases are handled,
4121 * but before we disembowel vd (so we can still do I/O to it).
4122 * But if we can't do it, don't treat the error as fatal --
4123 * it may be that the unwritability of the disk is the reason
4124 * it's being detached!
4125 */
4126 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4127
4128 /*
4129 * Remove vd from its parent and compact the parent's children.
4130 */
4131 vdev_remove_child(pvd, vd);
4132 vdev_compact_children(pvd);
4133
4134 /*
4135 * Remember one of the remaining children so we can get tvd below.
4136 */
4137 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4138
4139 /*
4140 * If we need to remove the remaining child from the list of hot spares,
4141 * do it now, marking the vdev as no longer a spare in the process.
4142 * We must do this before vdev_remove_parent(), because that can
4143 * change the GUID if it creates a new toplevel GUID. For a similar
4144 * reason, we must remove the spare now, in the same txg as the detach;
4145 * otherwise someone could attach a new sibling, change the GUID, and
4146 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4147 */
4148 if (unspare) {
4149 ASSERT(cvd->vdev_isspare);
4150 spa_spare_remove(cvd);
4151 unspare_guid = cvd->vdev_guid;
4152 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4153 cvd->vdev_unspare = B_TRUE;
4154 }
4155
4156 /*
4157 * If the parent mirror/replacing vdev only has one child,
4158 * the parent is no longer needed. Remove it from the tree.
4159 */
4160 if (pvd->vdev_children == 1) {
4161 if (pvd->vdev_ops == &vdev_spare_ops)
4162 cvd->vdev_unspare = B_FALSE;
4163 vdev_remove_parent(cvd);
4164 cvd->vdev_resilvering = B_FALSE;
4165 }
4166
4167
4168 /*
4169 * We don't set tvd until now because the parent we just removed
4170 * may have been the previous top-level vdev.
4171 */
4172 tvd = cvd->vdev_top;
4173 ASSERT(tvd->vdev_parent == rvd);
4174
4175 /*
4176 * Reevaluate the parent vdev state.
4177 */
4178 vdev_propagate_state(cvd);
4179
4180 /*
4181 * If the 'autoexpand' property is set on the pool then automatically
4182 * try to expand the size of the pool. For example if the device we
4183 * just detached was smaller than the others, it may be possible to
4184 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4185 * first so that we can obtain the updated sizes of the leaf vdevs.
4186 */
4187 if (spa->spa_autoexpand) {
4188 vdev_reopen(tvd);
4189 vdev_expand(tvd, txg);
4190 }
4191
4192 vdev_config_dirty(tvd);
4193
4194 /*
4195 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4196 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4197 * But first make sure we're not on any *other* txg's DTL list, to
4198 * prevent vd from being accessed after it's freed.
4199 */
4200 vdpath = spa_strdup(vd->vdev_path);
4201 for (t = 0; t < TXG_SIZE; t++)
4202 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4203 vd->vdev_detached = B_TRUE;
4204 vdev_dirty(tvd, VDD_DTL, vd, txg);
4205
4206 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4207
4208 /* hang on to the spa before we release the lock */
4209 spa_open_ref(spa, FTAG);
4210
4211 error = spa_vdev_exit(spa, vd, txg, 0);
4212
4213 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4214 "vdev=%s", vdpath);
4215 spa_strfree(vdpath);
4216
4217 /*
4218 * If this was the removal of the original device in a hot spare vdev,
4219 * then we want to go through and remove the device from the hot spare
4220 * list of every other pool.
4221 */
4222 if (unspare) {
4223 spa_t *altspa = NULL;
4224
4225 mutex_enter(&spa_namespace_lock);
4226 while ((altspa = spa_next(altspa)) != NULL) {
4227 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4228 altspa == spa)
4229 continue;
4230
4231 spa_open_ref(altspa, FTAG);
4232 mutex_exit(&spa_namespace_lock);
4233 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4234 mutex_enter(&spa_namespace_lock);
4235 spa_close(altspa, FTAG);
4236 }
4237 mutex_exit(&spa_namespace_lock);
4238
4239 /* search the rest of the vdevs for spares to remove */
4240 spa_vdev_resilver_done(spa);
4241 }
4242
4243 /* all done with the spa; OK to release */
4244 mutex_enter(&spa_namespace_lock);
4245 spa_close(spa, FTAG);
4246 mutex_exit(&spa_namespace_lock);
4247
4248 return (error);
4249 }
4250
4251 /*
4252 * Split a set of devices from their mirrors, and create a new pool from them.
4253 */
4254 int
4255 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4256 nvlist_t *props, boolean_t exp)
4257 {
4258 int error = 0;
4259 uint64_t txg, *glist;
4260 spa_t *newspa;
4261 uint_t c, children, lastlog;
4262 nvlist_t **child, *nvl, *tmp;
4263 dmu_tx_t *tx;
4264 char *altroot = NULL;
4265 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4266 boolean_t activate_slog;
4267
4268 ASSERT(spa_writeable(spa));
4269
4270 txg = spa_vdev_enter(spa);
4271
4272 /* clear the log and flush everything up to now */
4273 activate_slog = spa_passivate_log(spa);
4274 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4275 error = spa_offline_log(spa);
4276 txg = spa_vdev_config_enter(spa);
4277
4278 if (activate_slog)
4279 spa_activate_log(spa);
4280
4281 if (error != 0)
4282 return (spa_vdev_exit(spa, NULL, txg, error));
4283
4284 /* check new spa name before going any further */
4285 if (spa_lookup(newname) != NULL)
4286 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4287
4288 /*
4289 * scan through all the children to ensure they're all mirrors
4290 */
4291 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4292 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4293 &children) != 0)
4294 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4295
4296 /* first, check to ensure we've got the right child count */
4297 rvd = spa->spa_root_vdev;
4298 lastlog = 0;
4299 for (c = 0; c < rvd->vdev_children; c++) {
4300 vdev_t *vd = rvd->vdev_child[c];
4301
4302 /* don't count the holes & logs as children */
4303 if (vd->vdev_islog || vd->vdev_ishole) {
4304 if (lastlog == 0)
4305 lastlog = c;
4306 continue;
4307 }
4308
4309 lastlog = 0;
4310 }
4311 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4312 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4313
4314 /* next, ensure no spare or cache devices are part of the split */
4315 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4316 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4317 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4318
4319 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4320 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4321
4322 /* then, loop over each vdev and validate it */
4323 for (c = 0; c < children; c++) {
4324 uint64_t is_hole = 0;
4325
4326 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4327 &is_hole);
4328
4329 if (is_hole != 0) {
4330 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4331 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4332 continue;
4333 } else {
4334 error = EINVAL;
4335 break;
4336 }
4337 }
4338
4339 /* which disk is going to be split? */
4340 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4341 &glist[c]) != 0) {
4342 error = EINVAL;
4343 break;
4344 }
4345
4346 /* look it up in the spa */
4347 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4348 if (vml[c] == NULL) {
4349 error = ENODEV;
4350 break;
4351 }
4352
4353 /* make sure there's nothing stopping the split */
4354 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4355 vml[c]->vdev_islog ||
4356 vml[c]->vdev_ishole ||
4357 vml[c]->vdev_isspare ||
4358 vml[c]->vdev_isl2cache ||
4359 !vdev_writeable(vml[c]) ||
4360 vml[c]->vdev_children != 0 ||
4361 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4362 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4363 error = EINVAL;
4364 break;
4365 }
4366
4367 if (vdev_dtl_required(vml[c])) {
4368 error = EBUSY;
4369 break;
4370 }
4371
4372 /* we need certain info from the top level */
4373 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4374 vml[c]->vdev_top->vdev_ms_array) == 0);
4375 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4376 vml[c]->vdev_top->vdev_ms_shift) == 0);
4377 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4378 vml[c]->vdev_top->vdev_asize) == 0);
4379 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4380 vml[c]->vdev_top->vdev_ashift) == 0);
4381 }
4382
4383 if (error != 0) {
4384 kmem_free(vml, children * sizeof (vdev_t *));
4385 kmem_free(glist, children * sizeof (uint64_t));
4386 return (spa_vdev_exit(spa, NULL, txg, error));
4387 }
4388
4389 /* stop writers from using the disks */
4390 for (c = 0; c < children; c++) {
4391 if (vml[c] != NULL)
4392 vml[c]->vdev_offline = B_TRUE;
4393 }
4394 vdev_reopen(spa->spa_root_vdev);
4395
4396 /*
4397 * Temporarily record the splitting vdevs in the spa config. This
4398 * will disappear once the config is regenerated.
4399 */
4400 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4401 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4402 glist, children) == 0);
4403 kmem_free(glist, children * sizeof (uint64_t));
4404
4405 mutex_enter(&spa->spa_props_lock);
4406 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4407 nvl) == 0);
4408 mutex_exit(&spa->spa_props_lock);
4409 spa->spa_config_splitting = nvl;
4410 vdev_config_dirty(spa->spa_root_vdev);
4411
4412 /* configure and create the new pool */
4413 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4415 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4417 spa_version(spa)) == 0);
4418 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4419 spa->spa_config_txg) == 0);
4420 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4421 spa_generate_guid(NULL)) == 0);
4422 (void) nvlist_lookup_string(props,
4423 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4424
4425 /* add the new pool to the namespace */
4426 newspa = spa_add(newname, config, altroot);
4427 newspa->spa_config_txg = spa->spa_config_txg;
4428 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4429
4430 /* release the spa config lock, retaining the namespace lock */
4431 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4432
4433 if (zio_injection_enabled)
4434 zio_handle_panic_injection(spa, FTAG, 1);
4435
4436 spa_activate(newspa, spa_mode_global);
4437 spa_async_suspend(newspa);
4438
4439 /* create the new pool from the disks of the original pool */
4440 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4441 if (error)
4442 goto out;
4443
4444 /* if that worked, generate a real config for the new pool */
4445 if (newspa->spa_root_vdev != NULL) {
4446 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4447 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4448 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4449 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4450 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4451 B_TRUE));
4452 }
4453
4454 /* set the props */
4455 if (props != NULL) {
4456 spa_configfile_set(newspa, props, B_FALSE);
4457 error = spa_prop_set(newspa, props);
4458 if (error)
4459 goto out;
4460 }
4461
4462 /* flush everything */
4463 txg = spa_vdev_config_enter(newspa);
4464 vdev_config_dirty(newspa->spa_root_vdev);
4465 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4466
4467 if (zio_injection_enabled)
4468 zio_handle_panic_injection(spa, FTAG, 2);
4469
4470 spa_async_resume(newspa);
4471
4472 /* finally, update the original pool's config */
4473 txg = spa_vdev_config_enter(spa);
4474 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4475 error = dmu_tx_assign(tx, TXG_WAIT);
4476 if (error != 0)
4477 dmu_tx_abort(tx);
4478 for (c = 0; c < children; c++) {
4479 if (vml[c] != NULL) {
4480 vdev_split(vml[c]);
4481 if (error == 0)
4482 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4483 spa, tx, "vdev=%s",
4484 vml[c]->vdev_path);
4485 vdev_free(vml[c]);
4486 }
4487 }
4488 vdev_config_dirty(spa->spa_root_vdev);
4489 spa->spa_config_splitting = NULL;
4490 nvlist_free(nvl);
4491 if (error == 0)
4492 dmu_tx_commit(tx);
4493 (void) spa_vdev_exit(spa, NULL, txg, 0);
4494
4495 if (zio_injection_enabled)
4496 zio_handle_panic_injection(spa, FTAG, 3);
4497
4498 /* split is complete; log a history record */
4499 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4500 "split new pool %s from pool %s", newname, spa_name(spa));
4501
4502 kmem_free(vml, children * sizeof (vdev_t *));
4503
4504 /* if we're not going to mount the filesystems in userland, export */
4505 if (exp)
4506 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4507 B_FALSE, B_FALSE);
4508
4509 return (error);
4510
4511 out:
4512 spa_unload(newspa);
4513 spa_deactivate(newspa);
4514 spa_remove(newspa);
4515
4516 txg = spa_vdev_config_enter(spa);
4517
4518 /* re-online all offlined disks */
4519 for (c = 0; c < children; c++) {
4520 if (vml[c] != NULL)
4521 vml[c]->vdev_offline = B_FALSE;
4522 }
4523 vdev_reopen(spa->spa_root_vdev);
4524
4525 nvlist_free(spa->spa_config_splitting);
4526 spa->spa_config_splitting = NULL;
4527 (void) spa_vdev_exit(spa, NULL, txg, error);
4528
4529 kmem_free(vml, children * sizeof (vdev_t *));
4530 return (error);
4531 }
4532
4533 static nvlist_t *
4534 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4535 {
4536 int i;
4537
4538 for (i = 0; i < count; i++) {
4539 uint64_t guid;
4540
4541 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4542 &guid) == 0);
4543
4544 if (guid == target_guid)
4545 return (nvpp[i]);
4546 }
4547
4548 return (NULL);
4549 }
4550
4551 static void
4552 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4553 nvlist_t *dev_to_remove)
4554 {
4555 nvlist_t **newdev = NULL;
4556 int i, j;
4557
4558 if (count > 1)
4559 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4560
4561 for (i = 0, j = 0; i < count; i++) {
4562 if (dev[i] == dev_to_remove)
4563 continue;
4564 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4565 }
4566
4567 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4568 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4569
4570 for (i = 0; i < count - 1; i++)
4571 nvlist_free(newdev[i]);
4572
4573 if (count > 1)
4574 kmem_free(newdev, (count - 1) * sizeof (void *));
4575 }
4576
4577 /*
4578 * Evacuate the device.
4579 */
4580 static int
4581 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4582 {
4583 uint64_t txg;
4584 int error = 0;
4585
4586 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4587 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4588 ASSERT(vd == vd->vdev_top);
4589
4590 /*
4591 * Evacuate the device. We don't hold the config lock as writer
4592 * since we need to do I/O but we do keep the
4593 * spa_namespace_lock held. Once this completes the device
4594 * should no longer have any blocks allocated on it.
4595 */
4596 if (vd->vdev_islog) {
4597 if (vd->vdev_stat.vs_alloc != 0)
4598 error = spa_offline_log(spa);
4599 } else {
4600 error = ENOTSUP;
4601 }
4602
4603 if (error)
4604 return (error);
4605
4606 /*
4607 * The evacuation succeeded. Remove any remaining MOS metadata
4608 * associated with this vdev, and wait for these changes to sync.
4609 */
4610 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4611 txg = spa_vdev_config_enter(spa);
4612 vd->vdev_removing = B_TRUE;
4613 vdev_dirty(vd, 0, NULL, txg);
4614 vdev_config_dirty(vd);
4615 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4616
4617 return (0);
4618 }
4619
4620 /*
4621 * Complete the removal by cleaning up the namespace.
4622 */
4623 static void
4624 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4625 {
4626 vdev_t *rvd = spa->spa_root_vdev;
4627 uint64_t id = vd->vdev_id;
4628 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4629
4630 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4631 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4632 ASSERT(vd == vd->vdev_top);
4633
4634 /*
4635 * Only remove any devices which are empty.
4636 */
4637 if (vd->vdev_stat.vs_alloc != 0)
4638 return;
4639
4640 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4641
4642 if (list_link_active(&vd->vdev_state_dirty_node))
4643 vdev_state_clean(vd);
4644 if (list_link_active(&vd->vdev_config_dirty_node))
4645 vdev_config_clean(vd);
4646
4647 vdev_free(vd);
4648
4649 if (last_vdev) {
4650 vdev_compact_children(rvd);
4651 } else {
4652 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4653 vdev_add_child(rvd, vd);
4654 }
4655 vdev_config_dirty(rvd);
4656
4657 /*
4658 * Reassess the health of our root vdev.
4659 */
4660 vdev_reopen(rvd);
4661 }
4662
4663 /*
4664 * Remove a device from the pool -
4665 *
4666 * Removing a device from the vdev namespace requires several steps
4667 * and can take a significant amount of time. As a result we use
4668 * the spa_vdev_config_[enter/exit] functions which allow us to
4669 * grab and release the spa_config_lock while still holding the namespace
4670 * lock. During each step the configuration is synced out.
4671 */
4672
4673 /*
4674 * Remove a device from the pool. Currently, this supports removing only hot
4675 * spares, slogs, and level 2 ARC devices.
4676 */
4677 int
4678 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4679 {
4680 vdev_t *vd;
4681 metaslab_group_t *mg;
4682 nvlist_t **spares, **l2cache, *nv;
4683 uint64_t txg = 0;
4684 uint_t nspares, nl2cache;
4685 int error = 0;
4686 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4687
4688 ASSERT(spa_writeable(spa));
4689
4690 if (!locked)
4691 txg = spa_vdev_enter(spa);
4692
4693 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4694
4695 if (spa->spa_spares.sav_vdevs != NULL &&
4696 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4697 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4698 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4699 /*
4700 * Only remove the hot spare if it's not currently in use
4701 * in this pool.
4702 */
4703 if (vd == NULL || unspare) {
4704 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4705 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4706 spa_load_spares(spa);
4707 spa->spa_spares.sav_sync = B_TRUE;
4708 } else {
4709 error = EBUSY;
4710 }
4711 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4712 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4713 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4714 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4715 /*
4716 * Cache devices can always be removed.
4717 */
4718 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4719 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4720 spa_load_l2cache(spa);
4721 spa->spa_l2cache.sav_sync = B_TRUE;
4722 } else if (vd != NULL && vd->vdev_islog) {
4723 ASSERT(!locked);
4724 ASSERT(vd == vd->vdev_top);
4725
4726 /*
4727 * XXX - Once we have bp-rewrite this should
4728 * become the common case.
4729 */
4730
4731 mg = vd->vdev_mg;
4732
4733 /*
4734 * Stop allocating from this vdev.
4735 */
4736 metaslab_group_passivate(mg);
4737
4738 /*
4739 * Wait for the youngest allocations and frees to sync,
4740 * and then wait for the deferral of those frees to finish.
4741 */
4742 spa_vdev_config_exit(spa, NULL,
4743 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4744
4745 /*
4746 * Attempt to evacuate the vdev.
4747 */
4748 error = spa_vdev_remove_evacuate(spa, vd);
4749
4750 txg = spa_vdev_config_enter(spa);
4751
4752 /*
4753 * If we couldn't evacuate the vdev, unwind.
4754 */
4755 if (error) {
4756 metaslab_group_activate(mg);
4757 return (spa_vdev_exit(spa, NULL, txg, error));
4758 }
4759
4760 /*
4761 * Clean up the vdev namespace.
4762 */
4763 spa_vdev_remove_from_namespace(spa, vd);
4764
4765 } else if (vd != NULL) {
4766 /*
4767 * Normal vdevs cannot be removed (yet).
4768 */
4769 error = ENOTSUP;
4770 } else {
4771 /*
4772 * There is no vdev of any kind with the specified guid.
4773 */
4774 error = ENOENT;
4775 }
4776
4777 if (!locked)
4778 return (spa_vdev_exit(spa, NULL, txg, error));
4779
4780 return (error);
4781 }
4782
4783 /*
4784 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4785 * current spared, so we can detach it.
4786 */
4787 static vdev_t *
4788 spa_vdev_resilver_done_hunt(vdev_t *vd)
4789 {
4790 vdev_t *newvd, *oldvd;
4791 int c;
4792
4793 for (c = 0; c < vd->vdev_children; c++) {
4794 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4795 if (oldvd != NULL)
4796 return (oldvd);
4797 }
4798
4799 /*
4800 * Check for a completed replacement. We always consider the first
4801 * vdev in the list to be the oldest vdev, and the last one to be
4802 * the newest (see spa_vdev_attach() for how that works). In
4803 * the case where the newest vdev is faulted, we will not automatically
4804 * remove it after a resilver completes. This is OK as it will require
4805 * user intervention to determine which disk the admin wishes to keep.
4806 */
4807 if (vd->vdev_ops == &vdev_replacing_ops) {
4808 ASSERT(vd->vdev_children > 1);
4809
4810 newvd = vd->vdev_child[vd->vdev_children - 1];
4811 oldvd = vd->vdev_child[0];
4812
4813 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4814 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4815 !vdev_dtl_required(oldvd))
4816 return (oldvd);
4817 }
4818
4819 /*
4820 * Check for a completed resilver with the 'unspare' flag set.
4821 */
4822 if (vd->vdev_ops == &vdev_spare_ops) {
4823 vdev_t *first = vd->vdev_child[0];
4824 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4825
4826 if (last->vdev_unspare) {
4827 oldvd = first;
4828 newvd = last;
4829 } else if (first->vdev_unspare) {
4830 oldvd = last;
4831 newvd = first;
4832 } else {
4833 oldvd = NULL;
4834 }
4835
4836 if (oldvd != NULL &&
4837 vdev_dtl_empty(newvd, DTL_MISSING) &&
4838 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4839 !vdev_dtl_required(oldvd))
4840 return (oldvd);
4841
4842 /*
4843 * If there are more than two spares attached to a disk,
4844 * and those spares are not required, then we want to
4845 * attempt to free them up now so that they can be used
4846 * by other pools. Once we're back down to a single
4847 * disk+spare, we stop removing them.
4848 */
4849 if (vd->vdev_children > 2) {
4850 newvd = vd->vdev_child[1];
4851
4852 if (newvd->vdev_isspare && last->vdev_isspare &&
4853 vdev_dtl_empty(last, DTL_MISSING) &&
4854 vdev_dtl_empty(last, DTL_OUTAGE) &&
4855 !vdev_dtl_required(newvd))
4856 return (newvd);
4857 }
4858 }
4859
4860 return (NULL);
4861 }
4862
4863 static void
4864 spa_vdev_resilver_done(spa_t *spa)
4865 {
4866 vdev_t *vd, *pvd, *ppvd;
4867 uint64_t guid, sguid, pguid, ppguid;
4868
4869 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4870
4871 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4872 pvd = vd->vdev_parent;
4873 ppvd = pvd->vdev_parent;
4874 guid = vd->vdev_guid;
4875 pguid = pvd->vdev_guid;
4876 ppguid = ppvd->vdev_guid;
4877 sguid = 0;
4878 /*
4879 * If we have just finished replacing a hot spared device, then
4880 * we need to detach the parent's first child (the original hot
4881 * spare) as well.
4882 */
4883 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4884 ppvd->vdev_children == 2) {
4885 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4886 sguid = ppvd->vdev_child[1]->vdev_guid;
4887 }
4888 spa_config_exit(spa, SCL_ALL, FTAG);
4889 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4890 return;
4891 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4892 return;
4893 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4894 }
4895
4896 spa_config_exit(spa, SCL_ALL, FTAG);
4897 }
4898
4899 /*
4900 * Update the stored path or FRU for this vdev.
4901 */
4902 int
4903 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4904 boolean_t ispath)
4905 {
4906 vdev_t *vd;
4907 boolean_t sync = B_FALSE;
4908
4909 ASSERT(spa_writeable(spa));
4910
4911 spa_vdev_state_enter(spa, SCL_ALL);
4912
4913 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4914 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4915
4916 if (!vd->vdev_ops->vdev_op_leaf)
4917 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4918
4919 if (ispath) {
4920 if (strcmp(value, vd->vdev_path) != 0) {
4921 spa_strfree(vd->vdev_path);
4922 vd->vdev_path = spa_strdup(value);
4923 sync = B_TRUE;
4924 }
4925 } else {
4926 if (vd->vdev_fru == NULL) {
4927 vd->vdev_fru = spa_strdup(value);
4928 sync = B_TRUE;
4929 } else if (strcmp(value, vd->vdev_fru) != 0) {
4930 spa_strfree(vd->vdev_fru);
4931 vd->vdev_fru = spa_strdup(value);
4932 sync = B_TRUE;
4933 }
4934 }
4935
4936 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4937 }
4938
4939 int
4940 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4941 {
4942 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4943 }
4944
4945 int
4946 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4947 {
4948 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4949 }
4950
4951 /*
4952 * ==========================================================================
4953 * SPA Scanning
4954 * ==========================================================================
4955 */
4956
4957 int
4958 spa_scan_stop(spa_t *spa)
4959 {
4960 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4961 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4962 return (EBUSY);
4963 return (dsl_scan_cancel(spa->spa_dsl_pool));
4964 }
4965
4966 int
4967 spa_scan(spa_t *spa, pool_scan_func_t func)
4968 {
4969 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4970
4971 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4972 return (ENOTSUP);
4973
4974 /*
4975 * If a resilver was requested, but there is no DTL on a
4976 * writeable leaf device, we have nothing to do.
4977 */
4978 if (func == POOL_SCAN_RESILVER &&
4979 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4980 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4981 return (0);
4982 }
4983
4984 return (dsl_scan(spa->spa_dsl_pool, func));
4985 }
4986
4987 /*
4988 * ==========================================================================
4989 * SPA async task processing
4990 * ==========================================================================
4991 */
4992
4993 static void
4994 spa_async_remove(spa_t *spa, vdev_t *vd)
4995 {
4996 int c;
4997
4998 if (vd->vdev_remove_wanted) {
4999 vd->vdev_remove_wanted = B_FALSE;
5000 vd->vdev_delayed_close = B_FALSE;
5001 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5002
5003 /*
5004 * We want to clear the stats, but we don't want to do a full
5005 * vdev_clear() as that will cause us to throw away
5006 * degraded/faulted state as well as attempt to reopen the
5007 * device, all of which is a waste.
5008 */
5009 vd->vdev_stat.vs_read_errors = 0;
5010 vd->vdev_stat.vs_write_errors = 0;
5011 vd->vdev_stat.vs_checksum_errors = 0;
5012
5013 vdev_state_dirty(vd->vdev_top);
5014 }
5015
5016 for (c = 0; c < vd->vdev_children; c++)
5017 spa_async_remove(spa, vd->vdev_child[c]);
5018 }
5019
5020 static void
5021 spa_async_probe(spa_t *spa, vdev_t *vd)
5022 {
5023 int c;
5024
5025 if (vd->vdev_probe_wanted) {
5026 vd->vdev_probe_wanted = B_FALSE;
5027 vdev_reopen(vd); /* vdev_open() does the actual probe */
5028 }
5029
5030 for (c = 0; c < vd->vdev_children; c++)
5031 spa_async_probe(spa, vd->vdev_child[c]);
5032 }
5033
5034 static void
5035 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5036 {
5037 sysevent_id_t eid;
5038 nvlist_t *attr;
5039 char *physpath;
5040 int c;
5041
5042 if (!spa->spa_autoexpand)
5043 return;
5044
5045 for (c = 0; c < vd->vdev_children; c++) {
5046 vdev_t *cvd = vd->vdev_child[c];
5047 spa_async_autoexpand(spa, cvd);
5048 }
5049
5050 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5051 return;
5052
5053 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5054 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5055
5056 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5057 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5058
5059 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5060 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5061
5062 nvlist_free(attr);
5063 kmem_free(physpath, MAXPATHLEN);
5064 }
5065
5066 static void
5067 spa_async_thread(spa_t *spa)
5068 {
5069 int tasks, i;
5070
5071 ASSERT(spa->spa_sync_on);
5072
5073 mutex_enter(&spa->spa_async_lock);
5074 tasks = spa->spa_async_tasks;
5075 spa->spa_async_tasks = 0;
5076 mutex_exit(&spa->spa_async_lock);
5077
5078 /*
5079 * See if the config needs to be updated.
5080 */
5081 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5082 uint64_t old_space, new_space;
5083
5084 mutex_enter(&spa_namespace_lock);
5085 old_space = metaslab_class_get_space(spa_normal_class(spa));
5086 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5087 new_space = metaslab_class_get_space(spa_normal_class(spa));
5088 mutex_exit(&spa_namespace_lock);
5089
5090 /*
5091 * If the pool grew as a result of the config update,
5092 * then log an internal history event.
5093 */
5094 if (new_space != old_space) {
5095 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5096 spa, NULL,
5097 "pool '%s' size: %llu(+%llu)",
5098 spa_name(spa), new_space, new_space - old_space);
5099 }
5100 }
5101
5102 /*
5103 * See if any devices need to be marked REMOVED.
5104 */
5105 if (tasks & SPA_ASYNC_REMOVE) {
5106 spa_vdev_state_enter(spa, SCL_NONE);
5107 spa_async_remove(spa, spa->spa_root_vdev);
5108 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5109 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5110 for (i = 0; i < spa->spa_spares.sav_count; i++)
5111 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5112 (void) spa_vdev_state_exit(spa, NULL, 0);
5113 }
5114
5115 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5116 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5117 spa_async_autoexpand(spa, spa->spa_root_vdev);
5118 spa_config_exit(spa, SCL_CONFIG, FTAG);
5119 }
5120
5121 /*
5122 * See if any devices need to be probed.
5123 */
5124 if (tasks & SPA_ASYNC_PROBE) {
5125 spa_vdev_state_enter(spa, SCL_NONE);
5126 spa_async_probe(spa, spa->spa_root_vdev);
5127 (void) spa_vdev_state_exit(spa, NULL, 0);
5128 }
5129
5130 /*
5131 * If any devices are done replacing, detach them.
5132 */
5133 if (tasks & SPA_ASYNC_RESILVER_DONE)
5134 spa_vdev_resilver_done(spa);
5135
5136 /*
5137 * Kick off a resilver.
5138 */
5139 if (tasks & SPA_ASYNC_RESILVER)
5140 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5141
5142 /*
5143 * Let the world know that we're done.
5144 */
5145 mutex_enter(&spa->spa_async_lock);
5146 spa->spa_async_thread = NULL;
5147 cv_broadcast(&spa->spa_async_cv);
5148 mutex_exit(&spa->spa_async_lock);
5149 thread_exit();
5150 }
5151
5152 void
5153 spa_async_suspend(spa_t *spa)
5154 {
5155 mutex_enter(&spa->spa_async_lock);
5156 spa->spa_async_suspended++;
5157 while (spa->spa_async_thread != NULL)
5158 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5159 mutex_exit(&spa->spa_async_lock);
5160 }
5161
5162 void
5163 spa_async_resume(spa_t *spa)
5164 {
5165 mutex_enter(&spa->spa_async_lock);
5166 ASSERT(spa->spa_async_suspended != 0);
5167 spa->spa_async_suspended--;
5168 mutex_exit(&spa->spa_async_lock);
5169 }
5170
5171 static void
5172 spa_async_dispatch(spa_t *spa)
5173 {
5174 mutex_enter(&spa->spa_async_lock);
5175 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5176 spa->spa_async_thread == NULL &&
5177 rootdir != NULL && !vn_is_readonly(rootdir))
5178 spa->spa_async_thread = thread_create(NULL, 0,
5179 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5180 mutex_exit(&spa->spa_async_lock);
5181 }
5182
5183 void
5184 spa_async_request(spa_t *spa, int task)
5185 {
5186 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5187 mutex_enter(&spa->spa_async_lock);
5188 spa->spa_async_tasks |= task;
5189 mutex_exit(&spa->spa_async_lock);
5190 }
5191
5192 /*
5193 * ==========================================================================
5194 * SPA syncing routines
5195 * ==========================================================================
5196 */
5197
5198 static int
5199 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5200 {
5201 bpobj_t *bpo = arg;
5202 bpobj_enqueue(bpo, bp, tx);
5203 return (0);
5204 }
5205
5206 static int
5207 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5208 {
5209 zio_t *zio = arg;
5210
5211 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5212 zio->io_flags));
5213 return (0);
5214 }
5215
5216 static void
5217 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5218 {
5219 char *packed = NULL;
5220 size_t bufsize;
5221 size_t nvsize = 0;
5222 dmu_buf_t *db;
5223
5224 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5225
5226 /*
5227 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5228 * information. This avoids the dbuf_will_dirty() path and
5229 * saves us a pre-read to get data we don't actually care about.
5230 */
5231 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5232 packed = kmem_alloc(bufsize, KM_SLEEP);
5233
5234 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5235 KM_SLEEP) == 0);
5236 bzero(packed + nvsize, bufsize - nvsize);
5237
5238 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5239
5240 kmem_free(packed, bufsize);
5241
5242 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5243 dmu_buf_will_dirty(db, tx);
5244 *(uint64_t *)db->db_data = nvsize;
5245 dmu_buf_rele(db, FTAG);
5246 }
5247
5248 static void
5249 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5250 const char *config, const char *entry)
5251 {
5252 nvlist_t *nvroot;
5253 nvlist_t **list;
5254 int i;
5255
5256 if (!sav->sav_sync)
5257 return;
5258
5259 /*
5260 * Update the MOS nvlist describing the list of available devices.
5261 * spa_validate_aux() will have already made sure this nvlist is
5262 * valid and the vdevs are labeled appropriately.
5263 */
5264 if (sav->sav_object == 0) {
5265 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5266 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5267 sizeof (uint64_t), tx);
5268 VERIFY(zap_update(spa->spa_meta_objset,
5269 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5270 &sav->sav_object, tx) == 0);
5271 }
5272
5273 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5274 if (sav->sav_count == 0) {
5275 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5276 } else {
5277 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5278 for (i = 0; i < sav->sav_count; i++)
5279 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5280 B_FALSE, VDEV_CONFIG_L2CACHE);
5281 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5282 sav->sav_count) == 0);
5283 for (i = 0; i < sav->sav_count; i++)
5284 nvlist_free(list[i]);
5285 kmem_free(list, sav->sav_count * sizeof (void *));
5286 }
5287
5288 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5289 nvlist_free(nvroot);
5290
5291 sav->sav_sync = B_FALSE;
5292 }
5293
5294 static void
5295 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5296 {
5297 nvlist_t *config;
5298
5299 if (list_is_empty(&spa->spa_config_dirty_list))
5300 return;
5301
5302 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5303
5304 config = spa_config_generate(spa, spa->spa_root_vdev,
5305 dmu_tx_get_txg(tx), B_FALSE);
5306
5307 spa_config_exit(spa, SCL_STATE, FTAG);
5308
5309 if (spa->spa_config_syncing)
5310 nvlist_free(spa->spa_config_syncing);
5311 spa->spa_config_syncing = config;
5312
5313 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5314 }
5315
5316 /*
5317 * Set zpool properties.
5318 */
5319 static void
5320 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5321 {
5322 spa_t *spa = arg1;
5323 objset_t *mos = spa->spa_meta_objset;
5324 nvlist_t *nvp = arg2;
5325 nvpair_t *elem;
5326 uint64_t intval;
5327 char *strval;
5328 zpool_prop_t prop;
5329 const char *propname;
5330 zprop_type_t proptype;
5331
5332 mutex_enter(&spa->spa_props_lock);
5333
5334 elem = NULL;
5335 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5336 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5337 case ZPOOL_PROP_VERSION:
5338 /*
5339 * Only set version for non-zpool-creation cases
5340 * (set/import). spa_create() needs special care
5341 * for version setting.
5342 */
5343 if (tx->tx_txg != TXG_INITIAL) {
5344 VERIFY(nvpair_value_uint64(elem,
5345 &intval) == 0);
5346 ASSERT(intval <= SPA_VERSION);
5347 ASSERT(intval >= spa_version(spa));
5348 spa->spa_uberblock.ub_version = intval;
5349 vdev_config_dirty(spa->spa_root_vdev);
5350 }
5351 break;
5352
5353 case ZPOOL_PROP_ALTROOT:
5354 /*
5355 * 'altroot' is a non-persistent property. It should
5356 * have been set temporarily at creation or import time.
5357 */
5358 ASSERT(spa->spa_root != NULL);
5359 break;
5360
5361 case ZPOOL_PROP_READONLY:
5362 case ZPOOL_PROP_CACHEFILE:
5363 /*
5364 * 'readonly' and 'cachefile' are also non-persisitent
5365 * properties.
5366 */
5367 break;
5368 default:
5369 /*
5370 * Set pool property values in the poolprops mos object.
5371 */
5372 if (spa->spa_pool_props_object == 0) {
5373 VERIFY((spa->spa_pool_props_object =
5374 zap_create(mos, DMU_OT_POOL_PROPS,
5375 DMU_OT_NONE, 0, tx)) > 0);
5376
5377 VERIFY(zap_update(mos,
5378 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5379 8, 1, &spa->spa_pool_props_object, tx)
5380 == 0);
5381 }
5382
5383 /* normalize the property name */
5384 propname = zpool_prop_to_name(prop);
5385 proptype = zpool_prop_get_type(prop);
5386
5387 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5388 ASSERT(proptype == PROP_TYPE_STRING);
5389 VERIFY(nvpair_value_string(elem, &strval) == 0);
5390 VERIFY(zap_update(mos,
5391 spa->spa_pool_props_object, propname,
5392 1, strlen(strval) + 1, strval, tx) == 0);
5393
5394 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5395 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5396
5397 if (proptype == PROP_TYPE_INDEX) {
5398 const char *unused;
5399 VERIFY(zpool_prop_index_to_string(
5400 prop, intval, &unused) == 0);
5401 }
5402 VERIFY(zap_update(mos,
5403 spa->spa_pool_props_object, propname,
5404 8, 1, &intval, tx) == 0);
5405 } else {
5406 ASSERT(0); /* not allowed */
5407 }
5408
5409 switch (prop) {
5410 case ZPOOL_PROP_DELEGATION:
5411 spa->spa_delegation = intval;
5412 break;
5413 case ZPOOL_PROP_BOOTFS:
5414 spa->spa_bootfs = intval;
5415 break;
5416 case ZPOOL_PROP_FAILUREMODE:
5417 spa->spa_failmode = intval;
5418 break;
5419 case ZPOOL_PROP_AUTOEXPAND:
5420 spa->spa_autoexpand = intval;
5421 if (tx->tx_txg != TXG_INITIAL)
5422 spa_async_request(spa,
5423 SPA_ASYNC_AUTOEXPAND);
5424 break;
5425 case ZPOOL_PROP_DEDUPDITTO:
5426 spa->spa_dedup_ditto = intval;
5427 break;
5428 default:
5429 break;
5430 }
5431 }
5432
5433 /* log internal history if this is not a zpool create */
5434 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5435 tx->tx_txg != TXG_INITIAL) {
5436 spa_history_log_internal(LOG_POOL_PROPSET,
5437 spa, tx, "%s %lld %s",
5438 nvpair_name(elem), intval, spa_name(spa));
5439 }
5440 }
5441
5442 mutex_exit(&spa->spa_props_lock);
5443 }
5444
5445 /*
5446 * Perform one-time upgrade on-disk changes. spa_version() does not
5447 * reflect the new version this txg, so there must be no changes this
5448 * txg to anything that the upgrade code depends on after it executes.
5449 * Therefore this must be called after dsl_pool_sync() does the sync
5450 * tasks.
5451 */
5452 static void
5453 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5454 {
5455 dsl_pool_t *dp = spa->spa_dsl_pool;
5456
5457 ASSERT(spa->spa_sync_pass == 1);
5458
5459 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5460 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5461 dsl_pool_create_origin(dp, tx);
5462
5463 /* Keeping the origin open increases spa_minref */
5464 spa->spa_minref += 3;
5465 }
5466
5467 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5468 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5469 dsl_pool_upgrade_clones(dp, tx);
5470 }
5471
5472 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5473 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5474 dsl_pool_upgrade_dir_clones(dp, tx);
5475
5476 /* Keeping the freedir open increases spa_minref */
5477 spa->spa_minref += 3;
5478 }
5479 }
5480
5481 /*
5482 * Sync the specified transaction group. New blocks may be dirtied as
5483 * part of the process, so we iterate until it converges.
5484 */
5485 void
5486 spa_sync(spa_t *spa, uint64_t txg)
5487 {
5488 dsl_pool_t *dp = spa->spa_dsl_pool;
5489 objset_t *mos = spa->spa_meta_objset;
5490 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5491 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5492 vdev_t *rvd = spa->spa_root_vdev;
5493 vdev_t *vd;
5494 dmu_tx_t *tx;
5495 int error;
5496 int c;
5497
5498 VERIFY(spa_writeable(spa));
5499
5500 /*
5501 * Lock out configuration changes.
5502 */
5503 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5504
5505 spa->spa_syncing_txg = txg;
5506 spa->spa_sync_pass = 0;
5507
5508 /*
5509 * If there are any pending vdev state changes, convert them
5510 * into config changes that go out with this transaction group.
5511 */
5512 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5513 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5514 /*
5515 * We need the write lock here because, for aux vdevs,
5516 * calling vdev_config_dirty() modifies sav_config.
5517 * This is ugly and will become unnecessary when we
5518 * eliminate the aux vdev wart by integrating all vdevs
5519 * into the root vdev tree.
5520 */
5521 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5522 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5523 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5524 vdev_state_clean(vd);
5525 vdev_config_dirty(vd);
5526 }
5527 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5528 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5529 }
5530 spa_config_exit(spa, SCL_STATE, FTAG);
5531
5532 tx = dmu_tx_create_assigned(dp, txg);
5533
5534 /*
5535 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5536 * set spa_deflate if we have no raid-z vdevs.
5537 */
5538 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5539 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5540 int i;
5541
5542 for (i = 0; i < rvd->vdev_children; i++) {
5543 vd = rvd->vdev_child[i];
5544 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5545 break;
5546 }
5547 if (i == rvd->vdev_children) {
5548 spa->spa_deflate = TRUE;
5549 VERIFY(0 == zap_add(spa->spa_meta_objset,
5550 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5551 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5552 }
5553 }
5554
5555 /*
5556 * If anything has changed in this txg, or if someone is waiting
5557 * for this txg to sync (eg, spa_vdev_remove()), push the
5558 * deferred frees from the previous txg. If not, leave them
5559 * alone so that we don't generate work on an otherwise idle
5560 * system.
5561 */
5562 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5563 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5564 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5565 ((dsl_scan_active(dp->dp_scan) ||
5566 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5567 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5568 VERIFY3U(bpobj_iterate(defer_bpo,
5569 spa_free_sync_cb, zio, tx), ==, 0);
5570 VERIFY3U(zio_wait(zio), ==, 0);
5571 }
5572
5573 /*
5574 * Iterate to convergence.
5575 */
5576 do {
5577 int pass = ++spa->spa_sync_pass;
5578
5579 spa_sync_config_object(spa, tx);
5580 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5581 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5582 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5583 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5584 spa_errlog_sync(spa, txg);
5585 dsl_pool_sync(dp, txg);
5586
5587 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5588 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5589 bplist_iterate(free_bpl, spa_free_sync_cb,
5590 zio, tx);
5591 VERIFY(zio_wait(zio) == 0);
5592 } else {
5593 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5594 defer_bpo, tx);
5595 }
5596
5597 ddt_sync(spa, txg);
5598 dsl_scan_sync(dp, tx);
5599
5600 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5601 vdev_sync(vd, txg);
5602
5603 if (pass == 1)
5604 spa_sync_upgrades(spa, tx);
5605
5606 } while (dmu_objset_is_dirty(mos, txg));
5607
5608 /*
5609 * Rewrite the vdev configuration (which includes the uberblock)
5610 * to commit the transaction group.
5611 *
5612 * If there are no dirty vdevs, we sync the uberblock to a few
5613 * random top-level vdevs that are known to be visible in the
5614 * config cache (see spa_vdev_add() for a complete description).
5615 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5616 */
5617 for (;;) {
5618 /*
5619 * We hold SCL_STATE to prevent vdev open/close/etc.
5620 * while we're attempting to write the vdev labels.
5621 */
5622 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5623
5624 if (list_is_empty(&spa->spa_config_dirty_list)) {
5625 vdev_t *svd[SPA_DVAS_PER_BP];
5626 int svdcount = 0;
5627 int children = rvd->vdev_children;
5628 int c0 = spa_get_random(children);
5629
5630 for (c = 0; c < children; c++) {
5631 vd = rvd->vdev_child[(c0 + c) % children];
5632 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5633 continue;
5634 svd[svdcount++] = vd;
5635 if (svdcount == SPA_DVAS_PER_BP)
5636 break;
5637 }
5638 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5639 if (error != 0)
5640 error = vdev_config_sync(svd, svdcount, txg,
5641 B_TRUE);
5642 } else {
5643 error = vdev_config_sync(rvd->vdev_child,
5644 rvd->vdev_children, txg, B_FALSE);
5645 if (error != 0)
5646 error = vdev_config_sync(rvd->vdev_child,
5647 rvd->vdev_children, txg, B_TRUE);
5648 }
5649
5650 spa_config_exit(spa, SCL_STATE, FTAG);
5651
5652 if (error == 0)
5653 break;
5654 zio_suspend(spa, NULL);
5655 zio_resume_wait(spa);
5656 }
5657 dmu_tx_commit(tx);
5658
5659 /*
5660 * Clear the dirty config list.
5661 */
5662 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5663 vdev_config_clean(vd);
5664
5665 /*
5666 * Now that the new config has synced transactionally,
5667 * let it become visible to the config cache.
5668 */
5669 if (spa->spa_config_syncing != NULL) {
5670 spa_config_set(spa, spa->spa_config_syncing);
5671 spa->spa_config_txg = txg;
5672 spa->spa_config_syncing = NULL;
5673 }
5674
5675 spa->spa_ubsync = spa->spa_uberblock;
5676
5677 dsl_pool_sync_done(dp, txg);
5678
5679 /*
5680 * Update usable space statistics.
5681 */
5682 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5683 vdev_sync_done(vd, txg);
5684
5685 spa_update_dspace(spa);
5686
5687 /*
5688 * It had better be the case that we didn't dirty anything
5689 * since vdev_config_sync().
5690 */
5691 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5692 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5693 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5694
5695 spa->spa_sync_pass = 0;
5696
5697 spa_config_exit(spa, SCL_CONFIG, FTAG);
5698
5699 spa_handle_ignored_writes(spa);
5700
5701 /*
5702 * If any async tasks have been requested, kick them off.
5703 */
5704 spa_async_dispatch(spa);
5705 }
5706
5707 /*
5708 * Sync all pools. We don't want to hold the namespace lock across these
5709 * operations, so we take a reference on the spa_t and drop the lock during the
5710 * sync.
5711 */
5712 void
5713 spa_sync_allpools(void)
5714 {
5715 spa_t *spa = NULL;
5716 mutex_enter(&spa_namespace_lock);
5717 while ((spa = spa_next(spa)) != NULL) {
5718 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5719 !spa_writeable(spa) || spa_suspended(spa))
5720 continue;
5721 spa_open_ref(spa, FTAG);
5722 mutex_exit(&spa_namespace_lock);
5723 txg_wait_synced(spa_get_dsl(spa), 0);
5724 mutex_enter(&spa_namespace_lock);
5725 spa_close(spa, FTAG);
5726 }
5727 mutex_exit(&spa_namespace_lock);
5728 }
5729
5730 /*
5731 * ==========================================================================
5732 * Miscellaneous routines
5733 * ==========================================================================
5734 */
5735
5736 /*
5737 * Remove all pools in the system.
5738 */
5739 void
5740 spa_evict_all(void)
5741 {
5742 spa_t *spa;
5743
5744 /*
5745 * Remove all cached state. All pools should be closed now,
5746 * so every spa in the AVL tree should be unreferenced.
5747 */
5748 mutex_enter(&spa_namespace_lock);
5749 while ((spa = spa_next(NULL)) != NULL) {
5750 /*
5751 * Stop async tasks. The async thread may need to detach
5752 * a device that's been replaced, which requires grabbing
5753 * spa_namespace_lock, so we must drop it here.
5754 */
5755 spa_open_ref(spa, FTAG);
5756 mutex_exit(&spa_namespace_lock);
5757 spa_async_suspend(spa);
5758 mutex_enter(&spa_namespace_lock);
5759 spa_close(spa, FTAG);
5760
5761 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5762 spa_unload(spa);
5763 spa_deactivate(spa);
5764 }
5765 spa_remove(spa);
5766 }
5767 mutex_exit(&spa_namespace_lock);
5768 }
5769
5770 vdev_t *
5771 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5772 {
5773 vdev_t *vd;
5774 int i;
5775
5776 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5777 return (vd);
5778
5779 if (aux) {
5780 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5781 vd = spa->spa_l2cache.sav_vdevs[i];
5782 if (vd->vdev_guid == guid)
5783 return (vd);
5784 }
5785
5786 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5787 vd = spa->spa_spares.sav_vdevs[i];
5788 if (vd->vdev_guid == guid)
5789 return (vd);
5790 }
5791 }
5792
5793 return (NULL);
5794 }
5795
5796 void
5797 spa_upgrade(spa_t *spa, uint64_t version)
5798 {
5799 ASSERT(spa_writeable(spa));
5800
5801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5802
5803 /*
5804 * This should only be called for a non-faulted pool, and since a
5805 * future version would result in an unopenable pool, this shouldn't be
5806 * possible.
5807 */
5808 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5809 ASSERT(version >= spa->spa_uberblock.ub_version);
5810
5811 spa->spa_uberblock.ub_version = version;
5812 vdev_config_dirty(spa->spa_root_vdev);
5813
5814 spa_config_exit(spa, SCL_ALL, FTAG);
5815
5816 txg_wait_synced(spa_get_dsl(spa), 0);
5817 }
5818
5819 boolean_t
5820 spa_has_spare(spa_t *spa, uint64_t guid)
5821 {
5822 int i;
5823 uint64_t spareguid;
5824 spa_aux_vdev_t *sav = &spa->spa_spares;
5825
5826 for (i = 0; i < sav->sav_count; i++)
5827 if (sav->sav_vdevs[i]->vdev_guid == guid)
5828 return (B_TRUE);
5829
5830 for (i = 0; i < sav->sav_npending; i++) {
5831 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5832 &spareguid) == 0 && spareguid == guid)
5833 return (B_TRUE);
5834 }
5835
5836 return (B_FALSE);
5837 }
5838
5839 /*
5840 * Check if a pool has an active shared spare device.
5841 * Note: reference count of an active spare is 2, as a spare and as a replace
5842 */
5843 static boolean_t
5844 spa_has_active_shared_spare(spa_t *spa)
5845 {
5846 int i, refcnt;
5847 uint64_t pool;
5848 spa_aux_vdev_t *sav = &spa->spa_spares;
5849
5850 for (i = 0; i < sav->sav_count; i++) {
5851 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5852 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5853 refcnt > 2)
5854 return (B_TRUE);
5855 }
5856
5857 return (B_FALSE);
5858 }
5859
5860 /*
5861 * Post a sysevent corresponding to the given event. The 'name' must be one of
5862 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5863 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5864 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5865 * or zdb as real changes.
5866 */
5867 void
5868 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5869 {
5870 #ifdef _KERNEL
5871 sysevent_t *ev;
5872 sysevent_attr_list_t *attr = NULL;
5873 sysevent_value_t value;
5874 sysevent_id_t eid;
5875
5876 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5877 SE_SLEEP);
5878
5879 value.value_type = SE_DATA_TYPE_STRING;
5880 value.value.sv_string = spa_name(spa);
5881 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5882 goto done;
5883
5884 value.value_type = SE_DATA_TYPE_UINT64;
5885 value.value.sv_uint64 = spa_guid(spa);
5886 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5887 goto done;
5888
5889 if (vd) {
5890 value.value_type = SE_DATA_TYPE_UINT64;
5891 value.value.sv_uint64 = vd->vdev_guid;
5892 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5893 SE_SLEEP) != 0)
5894 goto done;
5895
5896 if (vd->vdev_path) {
5897 value.value_type = SE_DATA_TYPE_STRING;
5898 value.value.sv_string = vd->vdev_path;
5899 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5900 &value, SE_SLEEP) != 0)
5901 goto done;
5902 }
5903 }
5904
5905 if (sysevent_attach_attributes(ev, attr) != 0)
5906 goto done;
5907 attr = NULL;
5908
5909 (void) log_sysevent(ev, SE_SLEEP, &eid);
5910
5911 done:
5912 if (attr)
5913 sysevent_free_attr(attr);
5914 sysevent_free(ev);
5915 #endif
5916 }
ZFS On Linux mirror for PVE