]> 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 gcc c90 compliance warnings
[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 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;
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
476 if (error)
477 break;
478 }
479
480 if (!error && reset_bootfs) {
481 error = nvlist_remove(props,
482 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
483
484 if (!error) {
485 error = nvlist_add_uint64(props,
486 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
487 }
488 }
489
490 return (error);
491 }
492
493 void
494 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
495 {
496 char *cachefile;
497 spa_config_dirent_t *dp;
498
499 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
500 &cachefile) != 0)
501 return;
502
503 dp = kmem_alloc(sizeof (spa_config_dirent_t),
504 KM_SLEEP);
505
506 if (cachefile[0] == '\0')
507 dp->scd_path = spa_strdup(spa_config_path);
508 else if (strcmp(cachefile, "none") == 0)
509 dp->scd_path = NULL;
510 else
511 dp->scd_path = spa_strdup(cachefile);
512
513 list_insert_head(&spa->spa_config_list, dp);
514 if (need_sync)
515 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
516 }
517
518 int
519 spa_prop_set(spa_t *spa, nvlist_t *nvp)
520 {
521 int error;
522 nvpair_t *elem;
523 boolean_t need_sync = B_FALSE;
524 zpool_prop_t prop;
525
526 if ((error = spa_prop_validate(spa, nvp)) != 0)
527 return (error);
528
529 elem = NULL;
530 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
531 if ((prop = zpool_name_to_prop(
532 nvpair_name(elem))) == ZPROP_INVAL)
533 return (EINVAL);
534
535 if (prop == ZPOOL_PROP_CACHEFILE ||
536 prop == ZPOOL_PROP_ALTROOT ||
537 prop == ZPOOL_PROP_READONLY)
538 continue;
539
540 need_sync = B_TRUE;
541 break;
542 }
543
544 if (need_sync)
545 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
546 spa, nvp, 3));
547 else
548 return (0);
549 }
550
551 /*
552 * If the bootfs property value is dsobj, clear it.
553 */
554 void
555 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
556 {
557 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
558 VERIFY(zap_remove(spa->spa_meta_objset,
559 spa->spa_pool_props_object,
560 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
561 spa->spa_bootfs = 0;
562 }
563 }
564
565 /*
566 * ==========================================================================
567 * SPA state manipulation (open/create/destroy/import/export)
568 * ==========================================================================
569 */
570
571 static int
572 spa_error_entry_compare(const void *a, const void *b)
573 {
574 spa_error_entry_t *sa = (spa_error_entry_t *)a;
575 spa_error_entry_t *sb = (spa_error_entry_t *)b;
576 int ret;
577
578 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
579 sizeof (zbookmark_t));
580
581 if (ret < 0)
582 return (-1);
583 else if (ret > 0)
584 return (1);
585 else
586 return (0);
587 }
588
589 /*
590 * Utility function which retrieves copies of the current logs and
591 * re-initializes them in the process.
592 */
593 void
594 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
595 {
596 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
597
598 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
599 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
600
601 avl_create(&spa->spa_errlist_scrub,
602 spa_error_entry_compare, sizeof (spa_error_entry_t),
603 offsetof(spa_error_entry_t, se_avl));
604 avl_create(&spa->spa_errlist_last,
605 spa_error_entry_compare, sizeof (spa_error_entry_t),
606 offsetof(spa_error_entry_t, se_avl));
607 }
608
609 static taskq_t *
610 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
611 uint_t value)
612 {
613 uint_t flags = TASKQ_PREPOPULATE;
614 boolean_t batch = B_FALSE;
615
616 switch (mode) {
617 case zti_mode_null:
618 return (NULL); /* no taskq needed */
619
620 case zti_mode_fixed:
621 ASSERT3U(value, >=, 1);
622 value = MAX(value, 1);
623 break;
624
625 case zti_mode_batch:
626 batch = B_TRUE;
627 flags |= TASKQ_THREADS_CPU_PCT;
628 value = zio_taskq_batch_pct;
629 break;
630
631 case zti_mode_online_percent:
632 flags |= TASKQ_THREADS_CPU_PCT;
633 break;
634
635 default:
636 panic("unrecognized mode for %s taskq (%u:%u) in "
637 "spa_activate()",
638 name, mode, value);
639 break;
640 }
641
642 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
643 if (batch)
644 flags |= TASKQ_DC_BATCH;
645
646 return (taskq_create_sysdc(name, value, 50, INT_MAX,
647 spa->spa_proc, zio_taskq_basedc, flags));
648 }
649 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
650 spa->spa_proc, flags));
651 }
652
653 static void
654 spa_create_zio_taskqs(spa_t *spa)
655 {
656 int t, q;
657
658 for (t = 0; t < ZIO_TYPES; t++) {
659 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
660 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
661 enum zti_modes mode = ztip->zti_mode;
662 uint_t value = ztip->zti_value;
663 char name[32];
664
665 (void) snprintf(name, sizeof (name),
666 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
667
668 spa->spa_zio_taskq[t][q] =
669 spa_taskq_create(spa, name, mode, value);
670 }
671 }
672 }
673
674 #ifdef _KERNEL
675 static void
676 spa_thread(void *arg)
677 {
678 callb_cpr_t cprinfo;
679
680 spa_t *spa = arg;
681 user_t *pu = PTOU(curproc);
682
683 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
684 spa->spa_name);
685
686 ASSERT(curproc != &p0);
687 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
688 "zpool-%s", spa->spa_name);
689 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
690
691 /* bind this thread to the requested psrset */
692 if (zio_taskq_psrset_bind != PS_NONE) {
693 pool_lock();
694 mutex_enter(&cpu_lock);
695 mutex_enter(&pidlock);
696 mutex_enter(&curproc->p_lock);
697
698 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
699 0, NULL, NULL) == 0) {
700 curthread->t_bind_pset = zio_taskq_psrset_bind;
701 } else {
702 cmn_err(CE_WARN,
703 "Couldn't bind process for zfs pool \"%s\" to "
704 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
705 }
706
707 mutex_exit(&curproc->p_lock);
708 mutex_exit(&pidlock);
709 mutex_exit(&cpu_lock);
710 pool_unlock();
711 }
712
713 if (zio_taskq_sysdc) {
714 sysdc_thread_enter(curthread, 100, 0);
715 }
716
717 spa->spa_proc = curproc;
718 spa->spa_did = curthread->t_did;
719
720 spa_create_zio_taskqs(spa);
721
722 mutex_enter(&spa->spa_proc_lock);
723 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
724
725 spa->spa_proc_state = SPA_PROC_ACTIVE;
726 cv_broadcast(&spa->spa_proc_cv);
727
728 CALLB_CPR_SAFE_BEGIN(&cprinfo);
729 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
730 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
731 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
732
733 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
734 spa->spa_proc_state = SPA_PROC_GONE;
735 spa->spa_proc = &p0;
736 cv_broadcast(&spa->spa_proc_cv);
737 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
738
739 mutex_enter(&curproc->p_lock);
740 lwp_exit();
741 }
742 #endif
743
744 /*
745 * Activate an uninitialized pool.
746 */
747 static void
748 spa_activate(spa_t *spa, int mode)
749 {
750 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
751
752 spa->spa_state = POOL_STATE_ACTIVE;
753 spa->spa_mode = mode;
754
755 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
756 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
757
758 /* Try to create a covering process */
759 mutex_enter(&spa->spa_proc_lock);
760 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
761 ASSERT(spa->spa_proc == &p0);
762 spa->spa_did = 0;
763
764 /* Only create a process if we're going to be around a while. */
765 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
766 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
767 NULL, 0) == 0) {
768 spa->spa_proc_state = SPA_PROC_CREATED;
769 while (spa->spa_proc_state == SPA_PROC_CREATED) {
770 cv_wait(&spa->spa_proc_cv,
771 &spa->spa_proc_lock);
772 }
773 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
774 ASSERT(spa->spa_proc != &p0);
775 ASSERT(spa->spa_did != 0);
776 } else {
777 #ifdef _KERNEL
778 cmn_err(CE_WARN,
779 "Couldn't create process for zfs pool \"%s\"\n",
780 spa->spa_name);
781 #endif
782 }
783 }
784 mutex_exit(&spa->spa_proc_lock);
785
786 /* If we didn't create a process, we need to create our taskqs. */
787 if (spa->spa_proc == &p0) {
788 spa_create_zio_taskqs(spa);
789 }
790
791 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
792 offsetof(vdev_t, vdev_config_dirty_node));
793 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
794 offsetof(vdev_t, vdev_state_dirty_node));
795
796 txg_list_create(&spa->spa_vdev_txg_list,
797 offsetof(struct vdev, vdev_txg_node));
798
799 avl_create(&spa->spa_errlist_scrub,
800 spa_error_entry_compare, sizeof (spa_error_entry_t),
801 offsetof(spa_error_entry_t, se_avl));
802 avl_create(&spa->spa_errlist_last,
803 spa_error_entry_compare, sizeof (spa_error_entry_t),
804 offsetof(spa_error_entry_t, se_avl));
805 }
806
807 /*
808 * Opposite of spa_activate().
809 */
810 static void
811 spa_deactivate(spa_t *spa)
812 {
813 int t, q;
814
815 ASSERT(spa->spa_sync_on == B_FALSE);
816 ASSERT(spa->spa_dsl_pool == NULL);
817 ASSERT(spa->spa_root_vdev == NULL);
818 ASSERT(spa->spa_async_zio_root == NULL);
819 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
820
821 txg_list_destroy(&spa->spa_vdev_txg_list);
822
823 list_destroy(&spa->spa_config_dirty_list);
824 list_destroy(&spa->spa_state_dirty_list);
825
826 for (t = 0; t < ZIO_TYPES; t++) {
827 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
828 if (spa->spa_zio_taskq[t][q] != NULL)
829 taskq_destroy(spa->spa_zio_taskq[t][q]);
830 spa->spa_zio_taskq[t][q] = NULL;
831 }
832 }
833
834 metaslab_class_destroy(spa->spa_normal_class);
835 spa->spa_normal_class = NULL;
836
837 metaslab_class_destroy(spa->spa_log_class);
838 spa->spa_log_class = NULL;
839
840 /*
841 * If this was part of an import or the open otherwise failed, we may
842 * still have errors left in the queues. Empty them just in case.
843 */
844 spa_errlog_drain(spa);
845
846 avl_destroy(&spa->spa_errlist_scrub);
847 avl_destroy(&spa->spa_errlist_last);
848
849 spa->spa_state = POOL_STATE_UNINITIALIZED;
850
851 mutex_enter(&spa->spa_proc_lock);
852 if (spa->spa_proc_state != SPA_PROC_NONE) {
853 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
854 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
855 cv_broadcast(&spa->spa_proc_cv);
856 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
857 ASSERT(spa->spa_proc != &p0);
858 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
859 }
860 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
861 spa->spa_proc_state = SPA_PROC_NONE;
862 }
863 ASSERT(spa->spa_proc == &p0);
864 mutex_exit(&spa->spa_proc_lock);
865
866 /*
867 * We want to make sure spa_thread() has actually exited the ZFS
868 * module, so that the module can't be unloaded out from underneath
869 * it.
870 */
871 if (spa->spa_did != 0) {
872 thread_join(spa->spa_did);
873 spa->spa_did = 0;
874 }
875 }
876
877 /*
878 * Verify a pool configuration, and construct the vdev tree appropriately. This
879 * will create all the necessary vdevs in the appropriate layout, with each vdev
880 * in the CLOSED state. This will prep the pool before open/creation/import.
881 * All vdev validation is done by the vdev_alloc() routine.
882 */
883 static int
884 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
885 uint_t id, int atype)
886 {
887 nvlist_t **child;
888 uint_t children;
889 int error;
890 int c;
891
892 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
893 return (error);
894
895 if ((*vdp)->vdev_ops->vdev_op_leaf)
896 return (0);
897
898 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
899 &child, &children);
900
901 if (error == ENOENT)
902 return (0);
903
904 if (error) {
905 vdev_free(*vdp);
906 *vdp = NULL;
907 return (EINVAL);
908 }
909
910 for (c = 0; c < children; c++) {
911 vdev_t *vd;
912 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
913 atype)) != 0) {
914 vdev_free(*vdp);
915 *vdp = NULL;
916 return (error);
917 }
918 }
919
920 ASSERT(*vdp != NULL);
921
922 return (0);
923 }
924
925 /*
926 * Opposite of spa_load().
927 */
928 static void
929 spa_unload(spa_t *spa)
930 {
931 int i;
932
933 ASSERT(MUTEX_HELD(&spa_namespace_lock));
934
935 /*
936 * Stop async tasks.
937 */
938 spa_async_suspend(spa);
939
940 /*
941 * Stop syncing.
942 */
943 if (spa->spa_sync_on) {
944 txg_sync_stop(spa->spa_dsl_pool);
945 spa->spa_sync_on = B_FALSE;
946 }
947
948 /*
949 * Wait for any outstanding async I/O to complete.
950 */
951 if (spa->spa_async_zio_root != NULL) {
952 (void) zio_wait(spa->spa_async_zio_root);
953 spa->spa_async_zio_root = NULL;
954 }
955
956 bpobj_close(&spa->spa_deferred_bpobj);
957
958 /*
959 * Close the dsl pool.
960 */
961 if (spa->spa_dsl_pool) {
962 dsl_pool_close(spa->spa_dsl_pool);
963 spa->spa_dsl_pool = NULL;
964 spa->spa_meta_objset = NULL;
965 }
966
967 ddt_unload(spa);
968
969 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
970
971 /*
972 * Drop and purge level 2 cache
973 */
974 spa_l2cache_drop(spa);
975
976 /*
977 * Close all vdevs.
978 */
979 if (spa->spa_root_vdev)
980 vdev_free(spa->spa_root_vdev);
981 ASSERT(spa->spa_root_vdev == NULL);
982
983 for (i = 0; i < spa->spa_spares.sav_count; i++)
984 vdev_free(spa->spa_spares.sav_vdevs[i]);
985 if (spa->spa_spares.sav_vdevs) {
986 kmem_free(spa->spa_spares.sav_vdevs,
987 spa->spa_spares.sav_count * sizeof (void *));
988 spa->spa_spares.sav_vdevs = NULL;
989 }
990 if (spa->spa_spares.sav_config) {
991 nvlist_free(spa->spa_spares.sav_config);
992 spa->spa_spares.sav_config = NULL;
993 }
994 spa->spa_spares.sav_count = 0;
995
996 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
997 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
998 if (spa->spa_l2cache.sav_vdevs) {
999 kmem_free(spa->spa_l2cache.sav_vdevs,
1000 spa->spa_l2cache.sav_count * sizeof (void *));
1001 spa->spa_l2cache.sav_vdevs = NULL;
1002 }
1003 if (spa->spa_l2cache.sav_config) {
1004 nvlist_free(spa->spa_l2cache.sav_config);
1005 spa->spa_l2cache.sav_config = NULL;
1006 }
1007 spa->spa_l2cache.sav_count = 0;
1008
1009 spa->spa_async_suspended = 0;
1010
1011 spa_config_exit(spa, SCL_ALL, FTAG);
1012 }
1013
1014 /*
1015 * Load (or re-load) the current list of vdevs describing the active spares for
1016 * this pool. When this is called, we have some form of basic information in
1017 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1018 * then re-generate a more complete list including status information.
1019 */
1020 static void
1021 spa_load_spares(spa_t *spa)
1022 {
1023 nvlist_t **spares;
1024 uint_t nspares;
1025 int i;
1026 vdev_t *vd, *tvd;
1027
1028 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1029
1030 /*
1031 * First, close and free any existing spare vdevs.
1032 */
1033 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1034 vd = spa->spa_spares.sav_vdevs[i];
1035
1036 /* Undo the call to spa_activate() below */
1037 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1038 B_FALSE)) != NULL && tvd->vdev_isspare)
1039 spa_spare_remove(tvd);
1040 vdev_close(vd);
1041 vdev_free(vd);
1042 }
1043
1044 if (spa->spa_spares.sav_vdevs)
1045 kmem_free(spa->spa_spares.sav_vdevs,
1046 spa->spa_spares.sav_count * sizeof (void *));
1047
1048 if (spa->spa_spares.sav_config == NULL)
1049 nspares = 0;
1050 else
1051 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1052 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1053
1054 spa->spa_spares.sav_count = (int)nspares;
1055 spa->spa_spares.sav_vdevs = NULL;
1056
1057 if (nspares == 0)
1058 return;
1059
1060 /*
1061 * Construct the array of vdevs, opening them to get status in the
1062 * process. For each spare, there is potentially two different vdev_t
1063 * structures associated with it: one in the list of spares (used only
1064 * for basic validation purposes) and one in the active vdev
1065 * configuration (if it's spared in). During this phase we open and
1066 * validate each vdev on the spare list. If the vdev also exists in the
1067 * active configuration, then we also mark this vdev as an active spare.
1068 */
1069 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1070 KM_SLEEP);
1071 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1072 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1073 VDEV_ALLOC_SPARE) == 0);
1074 ASSERT(vd != NULL);
1075
1076 spa->spa_spares.sav_vdevs[i] = vd;
1077
1078 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1079 B_FALSE)) != NULL) {
1080 if (!tvd->vdev_isspare)
1081 spa_spare_add(tvd);
1082
1083 /*
1084 * We only mark the spare active if we were successfully
1085 * able to load the vdev. Otherwise, importing a pool
1086 * with a bad active spare would result in strange
1087 * behavior, because multiple pool would think the spare
1088 * is actively in use.
1089 *
1090 * There is a vulnerability here to an equally bizarre
1091 * circumstance, where a dead active spare is later
1092 * brought back to life (onlined or otherwise). Given
1093 * the rarity of this scenario, and the extra complexity
1094 * it adds, we ignore the possibility.
1095 */
1096 if (!vdev_is_dead(tvd))
1097 spa_spare_activate(tvd);
1098 }
1099
1100 vd->vdev_top = vd;
1101 vd->vdev_aux = &spa->spa_spares;
1102
1103 if (vdev_open(vd) != 0)
1104 continue;
1105
1106 if (vdev_validate_aux(vd) == 0)
1107 spa_spare_add(vd);
1108 }
1109
1110 /*
1111 * Recompute the stashed list of spares, with status information
1112 * this time.
1113 */
1114 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1115 DATA_TYPE_NVLIST_ARRAY) == 0);
1116
1117 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1118 KM_SLEEP);
1119 for (i = 0; i < spa->spa_spares.sav_count; i++)
1120 spares[i] = vdev_config_generate(spa,
1121 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1122 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1123 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1124 for (i = 0; i < spa->spa_spares.sav_count; i++)
1125 nvlist_free(spares[i]);
1126 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1127 }
1128
1129 /*
1130 * Load (or re-load) the current list of vdevs describing the active l2cache for
1131 * this pool. When this is called, we have some form of basic information in
1132 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1133 * then re-generate a more complete list including status information.
1134 * Devices which are already active have their details maintained, and are
1135 * not re-opened.
1136 */
1137 static void
1138 spa_load_l2cache(spa_t *spa)
1139 {
1140 nvlist_t **l2cache;
1141 uint_t nl2cache;
1142 int i, j, oldnvdevs;
1143 uint64_t guid;
1144 vdev_t *vd, **oldvdevs, **newvdevs;
1145 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1146
1147 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1148
1149 if (sav->sav_config != NULL) {
1150 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1151 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1152 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1153 } else {
1154 nl2cache = 0;
1155 }
1156
1157 oldvdevs = sav->sav_vdevs;
1158 oldnvdevs = sav->sav_count;
1159 sav->sav_vdevs = NULL;
1160 sav->sav_count = 0;
1161
1162 /*
1163 * Process new nvlist of vdevs.
1164 */
1165 for (i = 0; i < nl2cache; i++) {
1166 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1167 &guid) == 0);
1168
1169 newvdevs[i] = NULL;
1170 for (j = 0; j < oldnvdevs; j++) {
1171 vd = oldvdevs[j];
1172 if (vd != NULL && guid == vd->vdev_guid) {
1173 /*
1174 * Retain previous vdev for add/remove ops.
1175 */
1176 newvdevs[i] = vd;
1177 oldvdevs[j] = NULL;
1178 break;
1179 }
1180 }
1181
1182 if (newvdevs[i] == NULL) {
1183 /*
1184 * Create new vdev
1185 */
1186 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1187 VDEV_ALLOC_L2CACHE) == 0);
1188 ASSERT(vd != NULL);
1189 newvdevs[i] = vd;
1190
1191 /*
1192 * Commit this vdev as an l2cache device,
1193 * even if it fails to open.
1194 */
1195 spa_l2cache_add(vd);
1196
1197 vd->vdev_top = vd;
1198 vd->vdev_aux = sav;
1199
1200 spa_l2cache_activate(vd);
1201
1202 if (vdev_open(vd) != 0)
1203 continue;
1204
1205 (void) vdev_validate_aux(vd);
1206
1207 if (!vdev_is_dead(vd))
1208 l2arc_add_vdev(spa, vd);
1209 }
1210 }
1211
1212 /*
1213 * Purge vdevs that were dropped
1214 */
1215 for (i = 0; i < oldnvdevs; i++) {
1216 uint64_t pool;
1217
1218 vd = oldvdevs[i];
1219 if (vd != NULL) {
1220 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1221 pool != 0ULL && l2arc_vdev_present(vd))
1222 l2arc_remove_vdev(vd);
1223 (void) vdev_close(vd);
1224 spa_l2cache_remove(vd);
1225 }
1226 }
1227
1228 if (oldvdevs)
1229 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1230
1231 if (sav->sav_config == NULL)
1232 goto out;
1233
1234 sav->sav_vdevs = newvdevs;
1235 sav->sav_count = (int)nl2cache;
1236
1237 /*
1238 * Recompute the stashed list of l2cache devices, with status
1239 * information this time.
1240 */
1241 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1242 DATA_TYPE_NVLIST_ARRAY) == 0);
1243
1244 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1245 for (i = 0; i < sav->sav_count; i++)
1246 l2cache[i] = vdev_config_generate(spa,
1247 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1248 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1249 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1250 out:
1251 for (i = 0; i < sav->sav_count; i++)
1252 nvlist_free(l2cache[i]);
1253 if (sav->sav_count)
1254 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1255 }
1256
1257 static int
1258 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1259 {
1260 dmu_buf_t *db;
1261 char *packed = NULL;
1262 size_t nvsize = 0;
1263 int error;
1264 *value = NULL;
1265
1266 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1267 nvsize = *(uint64_t *)db->db_data;
1268 dmu_buf_rele(db, FTAG);
1269
1270 packed = kmem_alloc(nvsize, KM_SLEEP);
1271 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1272 DMU_READ_PREFETCH);
1273 if (error == 0)
1274 error = nvlist_unpack(packed, nvsize, value, 0);
1275 kmem_free(packed, nvsize);
1276
1277 return (error);
1278 }
1279
1280 /*
1281 * Checks to see if the given vdev could not be opened, in which case we post a
1282 * sysevent to notify the autoreplace code that the device has been removed.
1283 */
1284 static void
1285 spa_check_removed(vdev_t *vd)
1286 {
1287 int c;
1288
1289 for (c = 0; c < vd->vdev_children; c++)
1290 spa_check_removed(vd->vdev_child[c]);
1291
1292 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1293 zfs_post_autoreplace(vd->vdev_spa, vd);
1294 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1295 }
1296 }
1297
1298 /*
1299 * Validate the current config against the MOS config
1300 */
1301 static boolean_t
1302 spa_config_valid(spa_t *spa, nvlist_t *config)
1303 {
1304 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1305 nvlist_t *nv;
1306 int c, i;
1307
1308 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1309
1310 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1311 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1312
1313 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1314
1315 /*
1316 * If we're doing a normal import, then build up any additional
1317 * diagnostic information about missing devices in this config.
1318 * We'll pass this up to the user for further processing.
1319 */
1320 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1321 nvlist_t **child, *nv;
1322 uint64_t idx = 0;
1323
1324 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1325 KM_SLEEP);
1326 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1327
1328 for (c = 0; c < rvd->vdev_children; c++) {
1329 vdev_t *tvd = rvd->vdev_child[c];
1330 vdev_t *mtvd = mrvd->vdev_child[c];
1331
1332 if (tvd->vdev_ops == &vdev_missing_ops &&
1333 mtvd->vdev_ops != &vdev_missing_ops &&
1334 mtvd->vdev_islog)
1335 child[idx++] = vdev_config_generate(spa, mtvd,
1336 B_FALSE, 0);
1337 }
1338
1339 if (idx) {
1340 VERIFY(nvlist_add_nvlist_array(nv,
1341 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1342 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1343 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1344
1345 for (i = 0; i < idx; i++)
1346 nvlist_free(child[i]);
1347 }
1348 nvlist_free(nv);
1349 kmem_free(child, rvd->vdev_children * sizeof (char **));
1350 }
1351
1352 /*
1353 * Compare the root vdev tree with the information we have
1354 * from the MOS config (mrvd). Check each top-level vdev
1355 * with the corresponding MOS config top-level (mtvd).
1356 */
1357 for (c = 0; c < rvd->vdev_children; c++) {
1358 vdev_t *tvd = rvd->vdev_child[c];
1359 vdev_t *mtvd = mrvd->vdev_child[c];
1360
1361 /*
1362 * Resolve any "missing" vdevs in the current configuration.
1363 * If we find that the MOS config has more accurate information
1364 * about the top-level vdev then use that vdev instead.
1365 */
1366 if (tvd->vdev_ops == &vdev_missing_ops &&
1367 mtvd->vdev_ops != &vdev_missing_ops) {
1368
1369 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1370 continue;
1371
1372 /*
1373 * Device specific actions.
1374 */
1375 if (mtvd->vdev_islog) {
1376 spa_set_log_state(spa, SPA_LOG_CLEAR);
1377 } else {
1378 /*
1379 * XXX - once we have 'readonly' pool
1380 * support we should be able to handle
1381 * missing data devices by transitioning
1382 * the pool to readonly.
1383 */
1384 continue;
1385 }
1386
1387 /*
1388 * Swap the missing vdev with the data we were
1389 * able to obtain from the MOS config.
1390 */
1391 vdev_remove_child(rvd, tvd);
1392 vdev_remove_child(mrvd, mtvd);
1393
1394 vdev_add_child(rvd, mtvd);
1395 vdev_add_child(mrvd, tvd);
1396
1397 spa_config_exit(spa, SCL_ALL, FTAG);
1398 vdev_load(mtvd);
1399 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1400
1401 vdev_reopen(rvd);
1402 } else if (mtvd->vdev_islog) {
1403 /*
1404 * Load the slog device's state from the MOS config
1405 * since it's possible that the label does not
1406 * contain the most up-to-date information.
1407 */
1408 vdev_load_log_state(tvd, mtvd);
1409 vdev_reopen(tvd);
1410 }
1411 }
1412 vdev_free(mrvd);
1413 spa_config_exit(spa, SCL_ALL, FTAG);
1414
1415 /*
1416 * Ensure we were able to validate the config.
1417 */
1418 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1419 }
1420
1421 /*
1422 * Check for missing log devices
1423 */
1424 static int
1425 spa_check_logs(spa_t *spa)
1426 {
1427 switch (spa->spa_log_state) {
1428 case SPA_LOG_MISSING:
1429 /* need to recheck in case slog has been restored */
1430 case SPA_LOG_UNKNOWN:
1431 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1432 DS_FIND_CHILDREN)) {
1433 spa_set_log_state(spa, SPA_LOG_MISSING);
1434 return (1);
1435 }
1436 break;
1437 }
1438 return (0);
1439 }
1440
1441 static boolean_t
1442 spa_passivate_log(spa_t *spa)
1443 {
1444 vdev_t *rvd = spa->spa_root_vdev;
1445 boolean_t slog_found = B_FALSE;
1446 int c;
1447
1448 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1449
1450 if (!spa_has_slogs(spa))
1451 return (B_FALSE);
1452
1453 for (c = 0; c < rvd->vdev_children; c++) {
1454 vdev_t *tvd = rvd->vdev_child[c];
1455 metaslab_group_t *mg = tvd->vdev_mg;
1456
1457 if (tvd->vdev_islog) {
1458 metaslab_group_passivate(mg);
1459 slog_found = B_TRUE;
1460 }
1461 }
1462
1463 return (slog_found);
1464 }
1465
1466 static void
1467 spa_activate_log(spa_t *spa)
1468 {
1469 vdev_t *rvd = spa->spa_root_vdev;
1470 int c;
1471
1472 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1473
1474 for (c = 0; c < rvd->vdev_children; c++) {
1475 vdev_t *tvd = rvd->vdev_child[c];
1476 metaslab_group_t *mg = tvd->vdev_mg;
1477
1478 if (tvd->vdev_islog)
1479 metaslab_group_activate(mg);
1480 }
1481 }
1482
1483 int
1484 spa_offline_log(spa_t *spa)
1485 {
1486 int error = 0;
1487
1488 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1489 NULL, DS_FIND_CHILDREN)) == 0) {
1490
1491 /*
1492 * We successfully offlined the log device, sync out the
1493 * current txg so that the "stubby" block can be removed
1494 * by zil_sync().
1495 */
1496 txg_wait_synced(spa->spa_dsl_pool, 0);
1497 }
1498 return (error);
1499 }
1500
1501 static void
1502 spa_aux_check_removed(spa_aux_vdev_t *sav)
1503 {
1504 int i;
1505
1506 for (i = 0; i < sav->sav_count; i++)
1507 spa_check_removed(sav->sav_vdevs[i]);
1508 }
1509
1510 void
1511 spa_claim_notify(zio_t *zio)
1512 {
1513 spa_t *spa = zio->io_spa;
1514
1515 if (zio->io_error)
1516 return;
1517
1518 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1519 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1520 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1521 mutex_exit(&spa->spa_props_lock);
1522 }
1523
1524 typedef struct spa_load_error {
1525 uint64_t sle_meta_count;
1526 uint64_t sle_data_count;
1527 } spa_load_error_t;
1528
1529 static void
1530 spa_load_verify_done(zio_t *zio)
1531 {
1532 blkptr_t *bp = zio->io_bp;
1533 spa_load_error_t *sle = zio->io_private;
1534 dmu_object_type_t type = BP_GET_TYPE(bp);
1535 int error = zio->io_error;
1536
1537 if (error) {
1538 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1539 type != DMU_OT_INTENT_LOG)
1540 atomic_add_64(&sle->sle_meta_count, 1);
1541 else
1542 atomic_add_64(&sle->sle_data_count, 1);
1543 }
1544 zio_data_buf_free(zio->io_data, zio->io_size);
1545 }
1546
1547 /*ARGSUSED*/
1548 static int
1549 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1550 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1551 {
1552 if (bp != NULL) {
1553 zio_t *rio = arg;
1554 size_t size = BP_GET_PSIZE(bp);
1555 void *data = zio_data_buf_alloc(size);
1556
1557 zio_nowait(zio_read(rio, spa, bp, data, size,
1558 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1559 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1560 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1561 }
1562 return (0);
1563 }
1564
1565 static int
1566 spa_load_verify(spa_t *spa)
1567 {
1568 zio_t *rio;
1569 spa_load_error_t sle = { 0 };
1570 zpool_rewind_policy_t policy;
1571 boolean_t verify_ok = B_FALSE;
1572 int error;
1573
1574 zpool_get_rewind_policy(spa->spa_config, &policy);
1575
1576 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1577 return (0);
1578
1579 rio = zio_root(spa, NULL, &sle,
1580 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1581
1582 error = traverse_pool(spa, spa->spa_verify_min_txg,
1583 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1584
1585 (void) zio_wait(rio);
1586
1587 spa->spa_load_meta_errors = sle.sle_meta_count;
1588 spa->spa_load_data_errors = sle.sle_data_count;
1589
1590 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1591 sle.sle_data_count <= policy.zrp_maxdata) {
1592 int64_t loss = 0;
1593
1594 verify_ok = B_TRUE;
1595 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1596 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1597
1598 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1599 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1600 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1601 VERIFY(nvlist_add_int64(spa->spa_load_info,
1602 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1603 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1604 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1605 } else {
1606 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1607 }
1608
1609 if (error) {
1610 if (error != ENXIO && error != EIO)
1611 error = EIO;
1612 return (error);
1613 }
1614
1615 return (verify_ok ? 0 : EIO);
1616 }
1617
1618 /*
1619 * Find a value in the pool props object.
1620 */
1621 static void
1622 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1623 {
1624 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1625 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1626 }
1627
1628 /*
1629 * Find a value in the pool directory object.
1630 */
1631 static int
1632 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1633 {
1634 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1635 name, sizeof (uint64_t), 1, val));
1636 }
1637
1638 static int
1639 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1640 {
1641 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1642 return (err);
1643 }
1644
1645 /*
1646 * Fix up config after a partly-completed split. This is done with the
1647 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1648 * pool have that entry in their config, but only the splitting one contains
1649 * a list of all the guids of the vdevs that are being split off.
1650 *
1651 * This function determines what to do with that list: either rejoin
1652 * all the disks to the pool, or complete the splitting process. To attempt
1653 * the rejoin, each disk that is offlined is marked online again, and
1654 * we do a reopen() call. If the vdev label for every disk that was
1655 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1656 * then we call vdev_split() on each disk, and complete the split.
1657 *
1658 * Otherwise we leave the config alone, with all the vdevs in place in
1659 * the original pool.
1660 */
1661 static void
1662 spa_try_repair(spa_t *spa, nvlist_t *config)
1663 {
1664 uint_t extracted;
1665 uint64_t *glist;
1666 uint_t i, gcount;
1667 nvlist_t *nvl;
1668 vdev_t **vd;
1669 boolean_t attempt_reopen;
1670
1671 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1672 return;
1673
1674 /* check that the config is complete */
1675 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1676 &glist, &gcount) != 0)
1677 return;
1678
1679 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1680
1681 /* attempt to online all the vdevs & validate */
1682 attempt_reopen = B_TRUE;
1683 for (i = 0; i < gcount; i++) {
1684 if (glist[i] == 0) /* vdev is hole */
1685 continue;
1686
1687 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1688 if (vd[i] == NULL) {
1689 /*
1690 * Don't bother attempting to reopen the disks;
1691 * just do the split.
1692 */
1693 attempt_reopen = B_FALSE;
1694 } else {
1695 /* attempt to re-online it */
1696 vd[i]->vdev_offline = B_FALSE;
1697 }
1698 }
1699
1700 if (attempt_reopen) {
1701 vdev_reopen(spa->spa_root_vdev);
1702
1703 /* check each device to see what state it's in */
1704 for (extracted = 0, i = 0; i < gcount; i++) {
1705 if (vd[i] != NULL &&
1706 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1707 break;
1708 ++extracted;
1709 }
1710 }
1711
1712 /*
1713 * If every disk has been moved to the new pool, or if we never
1714 * even attempted to look at them, then we split them off for
1715 * good.
1716 */
1717 if (!attempt_reopen || gcount == extracted) {
1718 for (i = 0; i < gcount; i++)
1719 if (vd[i] != NULL)
1720 vdev_split(vd[i]);
1721 vdev_reopen(spa->spa_root_vdev);
1722 }
1723
1724 kmem_free(vd, gcount * sizeof (vdev_t *));
1725 }
1726
1727 static int
1728 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1729 boolean_t mosconfig)
1730 {
1731 nvlist_t *config = spa->spa_config;
1732 char *ereport = FM_EREPORT_ZFS_POOL;
1733 int error;
1734 uint64_t pool_guid;
1735 nvlist_t *nvl;
1736
1737 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1738 return (EINVAL);
1739
1740 /*
1741 * Versioning wasn't explicitly added to the label until later, so if
1742 * it's not present treat it as the initial version.
1743 */
1744 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1745 &spa->spa_ubsync.ub_version) != 0)
1746 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1747
1748 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1749 &spa->spa_config_txg);
1750
1751 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1752 spa_guid_exists(pool_guid, 0)) {
1753 error = EEXIST;
1754 } else {
1755 spa->spa_load_guid = pool_guid;
1756
1757 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1758 &nvl) == 0) {
1759 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1760 KM_SLEEP) == 0);
1761 }
1762
1763 gethrestime(&spa->spa_loaded_ts);
1764 error = spa_load_impl(spa, pool_guid, config, state, type,
1765 mosconfig, &ereport);
1766 }
1767
1768 spa->spa_minref = refcount_count(&spa->spa_refcount);
1769 if (error) {
1770 if (error != EEXIST) {
1771 spa->spa_loaded_ts.tv_sec = 0;
1772 spa->spa_loaded_ts.tv_nsec = 0;
1773 }
1774 if (error != EBADF) {
1775 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1776 }
1777 }
1778 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1779 spa->spa_ena = 0;
1780
1781 return (error);
1782 }
1783
1784 /*
1785 * Load an existing storage pool, using the pool's builtin spa_config as a
1786 * source of configuration information.
1787 */
1788 static int
1789 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1790 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1791 char **ereport)
1792 {
1793 int error = 0;
1794 nvlist_t *nvroot = NULL;
1795 vdev_t *rvd;
1796 uberblock_t *ub = &spa->spa_uberblock;
1797 uint64_t children, config_cache_txg = spa->spa_config_txg;
1798 int orig_mode = spa->spa_mode;
1799 int parse;
1800 uint64_t obj;
1801
1802 /*
1803 * If this is an untrusted config, access the pool in read-only mode.
1804 * This prevents things like resilvering recently removed devices.
1805 */
1806 if (!mosconfig)
1807 spa->spa_mode = FREAD;
1808
1809 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1810
1811 spa->spa_load_state = state;
1812
1813 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1814 return (EINVAL);
1815
1816 parse = (type == SPA_IMPORT_EXISTING ?
1817 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1818
1819 /*
1820 * Create "The Godfather" zio to hold all async IOs
1821 */
1822 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1823 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1824
1825 /*
1826 * Parse the configuration into a vdev tree. We explicitly set the
1827 * value that will be returned by spa_version() since parsing the
1828 * configuration requires knowing the version number.
1829 */
1830 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1831 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1832 spa_config_exit(spa, SCL_ALL, FTAG);
1833
1834 if (error != 0)
1835 return (error);
1836
1837 ASSERT(spa->spa_root_vdev == rvd);
1838
1839 if (type != SPA_IMPORT_ASSEMBLE) {
1840 ASSERT(spa_guid(spa) == pool_guid);
1841 }
1842
1843 /*
1844 * Try to open all vdevs, loading each label in the process.
1845 */
1846 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1847 error = vdev_open(rvd);
1848 spa_config_exit(spa, SCL_ALL, FTAG);
1849 if (error != 0)
1850 return (error);
1851
1852 /*
1853 * We need to validate the vdev labels against the configuration that
1854 * we have in hand, which is dependent on the setting of mosconfig. If
1855 * mosconfig is true then we're validating the vdev labels based on
1856 * that config. Otherwise, we're validating against the cached config
1857 * (zpool.cache) that was read when we loaded the zfs module, and then
1858 * later we will recursively call spa_load() and validate against
1859 * the vdev config.
1860 *
1861 * If we're assembling a new pool that's been split off from an
1862 * existing pool, the labels haven't yet been updated so we skip
1863 * validation for now.
1864 */
1865 if (type != SPA_IMPORT_ASSEMBLE) {
1866 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1867 error = vdev_validate(rvd);
1868 spa_config_exit(spa, SCL_ALL, FTAG);
1869
1870 if (error != 0)
1871 return (error);
1872
1873 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1874 return (ENXIO);
1875 }
1876
1877 /*
1878 * Find the best uberblock.
1879 */
1880 vdev_uberblock_load(NULL, rvd, ub);
1881
1882 /*
1883 * If we weren't able to find a single valid uberblock, return failure.
1884 */
1885 if (ub->ub_txg == 0)
1886 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1887
1888 /*
1889 * If the pool is newer than the code, we can't open it.
1890 */
1891 if (ub->ub_version > SPA_VERSION)
1892 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1893
1894 /*
1895 * If the vdev guid sum doesn't match the uberblock, we have an
1896 * incomplete configuration. We first check to see if the pool
1897 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1898 * If it is, defer the vdev_guid_sum check till later so we
1899 * can handle missing vdevs.
1900 */
1901 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1902 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1903 rvd->vdev_guid_sum != ub->ub_guid_sum)
1904 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1905
1906 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1907 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1908 spa_try_repair(spa, config);
1909 spa_config_exit(spa, SCL_ALL, FTAG);
1910 nvlist_free(spa->spa_config_splitting);
1911 spa->spa_config_splitting = NULL;
1912 }
1913
1914 /*
1915 * Initialize internal SPA structures.
1916 */
1917 spa->spa_state = POOL_STATE_ACTIVE;
1918 spa->spa_ubsync = spa->spa_uberblock;
1919 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1920 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1921 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1922 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1923 spa->spa_claim_max_txg = spa->spa_first_txg;
1924 spa->spa_prev_software_version = ub->ub_software_version;
1925
1926 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1927 if (error)
1928 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1929 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1930
1931 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1932 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1933
1934 if (!mosconfig) {
1935 uint64_t hostid;
1936 nvlist_t *policy = NULL, *nvconfig;
1937
1938 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1939 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1940
1941 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1942 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1943 char *hostname;
1944 unsigned long myhostid = 0;
1945
1946 VERIFY(nvlist_lookup_string(nvconfig,
1947 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1948
1949 #ifdef _KERNEL
1950 myhostid = zone_get_hostid(NULL);
1951 #else /* _KERNEL */
1952 /*
1953 * We're emulating the system's hostid in userland, so
1954 * we can't use zone_get_hostid().
1955 */
1956 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1957 #endif /* _KERNEL */
1958 if (hostid != 0 && myhostid != 0 &&
1959 hostid != myhostid) {
1960 nvlist_free(nvconfig);
1961 cmn_err(CE_WARN, "pool '%s' could not be "
1962 "loaded as it was last accessed by "
1963 "another system (host: %s hostid: 0x%lx). "
1964 "See: http://www.sun.com/msg/ZFS-8000-EY",
1965 spa_name(spa), hostname,
1966 (unsigned long)hostid);
1967 return (EBADF);
1968 }
1969 }
1970 if (nvlist_lookup_nvlist(spa->spa_config,
1971 ZPOOL_REWIND_POLICY, &policy) == 0)
1972 VERIFY(nvlist_add_nvlist(nvconfig,
1973 ZPOOL_REWIND_POLICY, policy) == 0);
1974
1975 spa_config_set(spa, nvconfig);
1976 spa_unload(spa);
1977 spa_deactivate(spa);
1978 spa_activate(spa, orig_mode);
1979
1980 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1981 }
1982
1983 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1985 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1986 if (error != 0)
1987 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1988
1989 /*
1990 * Load the bit that tells us to use the new accounting function
1991 * (raid-z deflation). If we have an older pool, this will not
1992 * be present.
1993 */
1994 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1995 if (error != 0 && error != ENOENT)
1996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1997
1998 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
1999 &spa->spa_creation_version);
2000 if (error != 0 && error != ENOENT)
2001 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2002
2003 /*
2004 * Load the persistent error log. If we have an older pool, this will
2005 * not be present.
2006 */
2007 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2008 if (error != 0 && error != ENOENT)
2009 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2010
2011 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2012 &spa->spa_errlog_scrub);
2013 if (error != 0 && error != ENOENT)
2014 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2015
2016 /*
2017 * Load the history object. If we have an older pool, this
2018 * will not be present.
2019 */
2020 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2021 if (error != 0 && error != ENOENT)
2022 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2023
2024 /*
2025 * If we're assembling the pool from the split-off vdevs of
2026 * an existing pool, we don't want to attach the spares & cache
2027 * devices.
2028 */
2029
2030 /*
2031 * Load any hot spares for this pool.
2032 */
2033 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2034 if (error != 0 && error != ENOENT)
2035 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2036 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2037 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2038 if (load_nvlist(spa, spa->spa_spares.sav_object,
2039 &spa->spa_spares.sav_config) != 0)
2040 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2041
2042 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2043 spa_load_spares(spa);
2044 spa_config_exit(spa, SCL_ALL, FTAG);
2045 } else if (error == 0) {
2046 spa->spa_spares.sav_sync = B_TRUE;
2047 }
2048
2049 /*
2050 * Load any level 2 ARC devices for this pool.
2051 */
2052 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2053 &spa->spa_l2cache.sav_object);
2054 if (error != 0 && error != ENOENT)
2055 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2056 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2057 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2058 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2059 &spa->spa_l2cache.sav_config) != 0)
2060 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2061
2062 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2063 spa_load_l2cache(spa);
2064 spa_config_exit(spa, SCL_ALL, FTAG);
2065 } else if (error == 0) {
2066 spa->spa_l2cache.sav_sync = B_TRUE;
2067 }
2068
2069 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2070
2071 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2072 if (error && error != ENOENT)
2073 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074
2075 if (error == 0) {
2076 uint64_t autoreplace;
2077
2078 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2079 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2080 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2081 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2082 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2083 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2084 &spa->spa_dedup_ditto);
2085
2086 spa->spa_autoreplace = (autoreplace != 0);
2087 }
2088
2089 /*
2090 * If the 'autoreplace' property is set, then post a resource notifying
2091 * the ZFS DE that it should not issue any faults for unopenable
2092 * devices. We also iterate over the vdevs, and post a sysevent for any
2093 * unopenable vdevs so that the normal autoreplace handler can take
2094 * over.
2095 */
2096 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2097 spa_check_removed(spa->spa_root_vdev);
2098 /*
2099 * For the import case, this is done in spa_import(), because
2100 * at this point we're using the spare definitions from
2101 * the MOS config, not necessarily from the userland config.
2102 */
2103 if (state != SPA_LOAD_IMPORT) {
2104 spa_aux_check_removed(&spa->spa_spares);
2105 spa_aux_check_removed(&spa->spa_l2cache);
2106 }
2107 }
2108
2109 /*
2110 * Load the vdev state for all toplevel vdevs.
2111 */
2112 vdev_load(rvd);
2113
2114 /*
2115 * Propagate the leaf DTLs we just loaded all the way up the tree.
2116 */
2117 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2118 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2119 spa_config_exit(spa, SCL_ALL, FTAG);
2120
2121 /*
2122 * Load the DDTs (dedup tables).
2123 */
2124 error = ddt_load(spa);
2125 if (error != 0)
2126 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2127
2128 spa_update_dspace(spa);
2129
2130 /*
2131 * Validate the config, using the MOS config to fill in any
2132 * information which might be missing. If we fail to validate
2133 * the config then declare the pool unfit for use. If we're
2134 * assembling a pool from a split, the log is not transferred
2135 * over.
2136 */
2137 if (type != SPA_IMPORT_ASSEMBLE) {
2138 nvlist_t *nvconfig;
2139
2140 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2141 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2142
2143 if (!spa_config_valid(spa, nvconfig)) {
2144 nvlist_free(nvconfig);
2145 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2146 ENXIO));
2147 }
2148 nvlist_free(nvconfig);
2149
2150 /*
2151 * Now that we've validate the config, check the state of the
2152 * root vdev. If it can't be opened, it indicates one or
2153 * more toplevel vdevs are faulted.
2154 */
2155 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2156 return (ENXIO);
2157
2158 if (spa_check_logs(spa)) {
2159 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2160 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2161 }
2162 }
2163
2164 /*
2165 * We've successfully opened the pool, verify that we're ready
2166 * to start pushing transactions.
2167 */
2168 if (state != SPA_LOAD_TRYIMPORT) {
2169 if (error = spa_load_verify(spa))
2170 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2171 error));
2172 }
2173
2174 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2175 spa->spa_load_max_txg == UINT64_MAX)) {
2176 dmu_tx_t *tx;
2177 int need_update = B_FALSE;
2178 int c;
2179
2180 ASSERT(state != SPA_LOAD_TRYIMPORT);
2181
2182 /*
2183 * Claim log blocks that haven't been committed yet.
2184 * This must all happen in a single txg.
2185 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2186 * invoked from zil_claim_log_block()'s i/o done callback.
2187 * Price of rollback is that we abandon the log.
2188 */
2189 spa->spa_claiming = B_TRUE;
2190
2191 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2192 spa_first_txg(spa));
2193 (void) dmu_objset_find(spa_name(spa),
2194 zil_claim, tx, DS_FIND_CHILDREN);
2195 dmu_tx_commit(tx);
2196
2197 spa->spa_claiming = B_FALSE;
2198
2199 spa_set_log_state(spa, SPA_LOG_GOOD);
2200 spa->spa_sync_on = B_TRUE;
2201 txg_sync_start(spa->spa_dsl_pool);
2202
2203 /*
2204 * Wait for all claims to sync. We sync up to the highest
2205 * claimed log block birth time so that claimed log blocks
2206 * don't appear to be from the future. spa_claim_max_txg
2207 * will have been set for us by either zil_check_log_chain()
2208 * (invoked from spa_check_logs()) or zil_claim() above.
2209 */
2210 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2211
2212 /*
2213 * If the config cache is stale, or we have uninitialized
2214 * metaslabs (see spa_vdev_add()), then update the config.
2215 *
2216 * If this is a verbatim import, trust the current
2217 * in-core spa_config and update the disk labels.
2218 */
2219 if (config_cache_txg != spa->spa_config_txg ||
2220 state == SPA_LOAD_IMPORT ||
2221 state == SPA_LOAD_RECOVER ||
2222 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2223 need_update = B_TRUE;
2224
2225 for (c = 0; c < rvd->vdev_children; c++)
2226 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2227 need_update = B_TRUE;
2228
2229 /*
2230 * Update the config cache asychronously in case we're the
2231 * root pool, in which case the config cache isn't writable yet.
2232 */
2233 if (need_update)
2234 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2235
2236 /*
2237 * Check all DTLs to see if anything needs resilvering.
2238 */
2239 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2240 vdev_resilver_needed(rvd, NULL, NULL))
2241 spa_async_request(spa, SPA_ASYNC_RESILVER);
2242
2243 /*
2244 * Delete any inconsistent datasets.
2245 */
2246 (void) dmu_objset_find(spa_name(spa),
2247 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2248
2249 /*
2250 * Clean up any stale temporary dataset userrefs.
2251 */
2252 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2253 }
2254
2255 return (0);
2256 }
2257
2258 static int
2259 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2260 {
2261 int mode = spa->spa_mode;
2262
2263 spa_unload(spa);
2264 spa_deactivate(spa);
2265
2266 spa->spa_load_max_txg--;
2267
2268 spa_activate(spa, mode);
2269 spa_async_suspend(spa);
2270
2271 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2272 }
2273
2274 static int
2275 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2276 uint64_t max_request, int rewind_flags)
2277 {
2278 nvlist_t *config = NULL;
2279 int load_error, rewind_error;
2280 uint64_t safe_rewind_txg;
2281 uint64_t min_txg;
2282
2283 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2284 spa->spa_load_max_txg = spa->spa_load_txg;
2285 spa_set_log_state(spa, SPA_LOG_CLEAR);
2286 } else {
2287 spa->spa_load_max_txg = max_request;
2288 }
2289
2290 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2291 mosconfig);
2292 if (load_error == 0)
2293 return (0);
2294
2295 if (spa->spa_root_vdev != NULL)
2296 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2297
2298 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2299 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2300
2301 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2302 nvlist_free(config);
2303 return (load_error);
2304 }
2305
2306 /* Price of rolling back is discarding txgs, including log */
2307 if (state == SPA_LOAD_RECOVER)
2308 spa_set_log_state(spa, SPA_LOG_CLEAR);
2309
2310 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2311 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2312 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2313 TXG_INITIAL : safe_rewind_txg;
2314
2315 /*
2316 * Continue as long as we're finding errors, we're still within
2317 * the acceptable rewind range, and we're still finding uberblocks
2318 */
2319 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2320 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2321 if (spa->spa_load_max_txg < safe_rewind_txg)
2322 spa->spa_extreme_rewind = B_TRUE;
2323 rewind_error = spa_load_retry(spa, state, mosconfig);
2324 }
2325
2326 spa->spa_extreme_rewind = B_FALSE;
2327 spa->spa_load_max_txg = UINT64_MAX;
2328
2329 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2330 spa_config_set(spa, config);
2331
2332 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2333 }
2334
2335 /*
2336 * Pool Open/Import
2337 *
2338 * The import case is identical to an open except that the configuration is sent
2339 * down from userland, instead of grabbed from the configuration cache. For the
2340 * case of an open, the pool configuration will exist in the
2341 * POOL_STATE_UNINITIALIZED state.
2342 *
2343 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2344 * the same time open the pool, without having to keep around the spa_t in some
2345 * ambiguous state.
2346 */
2347 static int
2348 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2349 nvlist_t **config)
2350 {
2351 spa_t *spa;
2352 spa_load_state_t state = SPA_LOAD_OPEN;
2353 int error;
2354 int locked = B_FALSE;
2355
2356 *spapp = NULL;
2357
2358 /*
2359 * As disgusting as this is, we need to support recursive calls to this
2360 * function because dsl_dir_open() is called during spa_load(), and ends
2361 * up calling spa_open() again. The real fix is to figure out how to
2362 * avoid dsl_dir_open() calling this in the first place.
2363 */
2364 if (mutex_owner(&spa_namespace_lock) != curthread) {
2365 mutex_enter(&spa_namespace_lock);
2366 locked = B_TRUE;
2367 }
2368
2369 if ((spa = spa_lookup(pool)) == NULL) {
2370 if (locked)
2371 mutex_exit(&spa_namespace_lock);
2372 return (ENOENT);
2373 }
2374
2375 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2376 zpool_rewind_policy_t policy;
2377
2378 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2379 &policy);
2380 if (policy.zrp_request & ZPOOL_DO_REWIND)
2381 state = SPA_LOAD_RECOVER;
2382
2383 spa_activate(spa, spa_mode_global);
2384
2385 if (state != SPA_LOAD_RECOVER)
2386 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2387
2388 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2389 policy.zrp_request);
2390
2391 if (error == EBADF) {
2392 /*
2393 * If vdev_validate() returns failure (indicated by
2394 * EBADF), it indicates that one of the vdevs indicates
2395 * that the pool has been exported or destroyed. If
2396 * this is the case, the config cache is out of sync and
2397 * we should remove the pool from the namespace.
2398 */
2399 spa_unload(spa);
2400 spa_deactivate(spa);
2401 spa_config_sync(spa, B_TRUE, B_TRUE);
2402 spa_remove(spa);
2403 if (locked)
2404 mutex_exit(&spa_namespace_lock);
2405 return (ENOENT);
2406 }
2407
2408 if (error) {
2409 /*
2410 * We can't open the pool, but we still have useful
2411 * information: the state of each vdev after the
2412 * attempted vdev_open(). Return this to the user.
2413 */
2414 if (config != NULL && spa->spa_config) {
2415 VERIFY(nvlist_dup(spa->spa_config, config,
2416 KM_SLEEP) == 0);
2417 VERIFY(nvlist_add_nvlist(*config,
2418 ZPOOL_CONFIG_LOAD_INFO,
2419 spa->spa_load_info) == 0);
2420 }
2421 spa_unload(spa);
2422 spa_deactivate(spa);
2423 spa->spa_last_open_failed = error;
2424 if (locked)
2425 mutex_exit(&spa_namespace_lock);
2426 *spapp = NULL;
2427 return (error);
2428 }
2429 }
2430
2431 spa_open_ref(spa, tag);
2432
2433 if (config != NULL)
2434 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2435
2436 /*
2437 * If we've recovered the pool, pass back any information we
2438 * gathered while doing the load.
2439 */
2440 if (state == SPA_LOAD_RECOVER) {
2441 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2442 spa->spa_load_info) == 0);
2443 }
2444
2445 if (locked) {
2446 spa->spa_last_open_failed = 0;
2447 spa->spa_last_ubsync_txg = 0;
2448 spa->spa_load_txg = 0;
2449 mutex_exit(&spa_namespace_lock);
2450 }
2451
2452 *spapp = spa;
2453
2454 return (0);
2455 }
2456
2457 int
2458 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2459 nvlist_t **config)
2460 {
2461 return (spa_open_common(name, spapp, tag, policy, config));
2462 }
2463
2464 int
2465 spa_open(const char *name, spa_t **spapp, void *tag)
2466 {
2467 return (spa_open_common(name, spapp, tag, NULL, NULL));
2468 }
2469
2470 /*
2471 * Lookup the given spa_t, incrementing the inject count in the process,
2472 * preventing it from being exported or destroyed.
2473 */
2474 spa_t *
2475 spa_inject_addref(char *name)
2476 {
2477 spa_t *spa;
2478
2479 mutex_enter(&spa_namespace_lock);
2480 if ((spa = spa_lookup(name)) == NULL) {
2481 mutex_exit(&spa_namespace_lock);
2482 return (NULL);
2483 }
2484 spa->spa_inject_ref++;
2485 mutex_exit(&spa_namespace_lock);
2486
2487 return (spa);
2488 }
2489
2490 void
2491 spa_inject_delref(spa_t *spa)
2492 {
2493 mutex_enter(&spa_namespace_lock);
2494 spa->spa_inject_ref--;
2495 mutex_exit(&spa_namespace_lock);
2496 }
2497
2498 /*
2499 * Add spares device information to the nvlist.
2500 */
2501 static void
2502 spa_add_spares(spa_t *spa, nvlist_t *config)
2503 {
2504 nvlist_t **spares;
2505 uint_t i, nspares;
2506 nvlist_t *nvroot;
2507 uint64_t guid;
2508 vdev_stat_t *vs;
2509 uint_t vsc;
2510 uint64_t pool;
2511
2512 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2513
2514 if (spa->spa_spares.sav_count == 0)
2515 return;
2516
2517 VERIFY(nvlist_lookup_nvlist(config,
2518 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2519 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2520 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2521 if (nspares != 0) {
2522 VERIFY(nvlist_add_nvlist_array(nvroot,
2523 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2524 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2525 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2526
2527 /*
2528 * Go through and find any spares which have since been
2529 * repurposed as an active spare. If this is the case, update
2530 * their status appropriately.
2531 */
2532 for (i = 0; i < nspares; i++) {
2533 VERIFY(nvlist_lookup_uint64(spares[i],
2534 ZPOOL_CONFIG_GUID, &guid) == 0);
2535 if (spa_spare_exists(guid, &pool, NULL) &&
2536 pool != 0ULL) {
2537 VERIFY(nvlist_lookup_uint64_array(
2538 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2539 (uint64_t **)&vs, &vsc) == 0);
2540 vs->vs_state = VDEV_STATE_CANT_OPEN;
2541 vs->vs_aux = VDEV_AUX_SPARED;
2542 }
2543 }
2544 }
2545 }
2546
2547 /*
2548 * Add l2cache device information to the nvlist, including vdev stats.
2549 */
2550 static void
2551 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2552 {
2553 nvlist_t **l2cache;
2554 uint_t i, j, nl2cache;
2555 nvlist_t *nvroot;
2556 uint64_t guid;
2557 vdev_t *vd;
2558 vdev_stat_t *vs;
2559 uint_t vsc;
2560
2561 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2562
2563 if (spa->spa_l2cache.sav_count == 0)
2564 return;
2565
2566 VERIFY(nvlist_lookup_nvlist(config,
2567 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2568 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2569 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2570 if (nl2cache != 0) {
2571 VERIFY(nvlist_add_nvlist_array(nvroot,
2572 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2573 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2574 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2575
2576 /*
2577 * Update level 2 cache device stats.
2578 */
2579
2580 for (i = 0; i < nl2cache; i++) {
2581 VERIFY(nvlist_lookup_uint64(l2cache[i],
2582 ZPOOL_CONFIG_GUID, &guid) == 0);
2583
2584 vd = NULL;
2585 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2586 if (guid ==
2587 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2588 vd = spa->spa_l2cache.sav_vdevs[j];
2589 break;
2590 }
2591 }
2592 ASSERT(vd != NULL);
2593
2594 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2595 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2596 == 0);
2597 vdev_get_stats(vd, vs);
2598 }
2599 }
2600 }
2601
2602 int
2603 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2604 {
2605 int error;
2606 spa_t *spa;
2607
2608 *config = NULL;
2609 error = spa_open_common(name, &spa, FTAG, NULL, config);
2610
2611 if (spa != NULL) {
2612 /*
2613 * This still leaves a window of inconsistency where the spares
2614 * or l2cache devices could change and the config would be
2615 * self-inconsistent.
2616 */
2617 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2618
2619 if (*config != NULL) {
2620 uint64_t loadtimes[2];
2621
2622 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2623 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2624 VERIFY(nvlist_add_uint64_array(*config,
2625 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2626
2627 VERIFY(nvlist_add_uint64(*config,
2628 ZPOOL_CONFIG_ERRCOUNT,
2629 spa_get_errlog_size(spa)) == 0);
2630
2631 if (spa_suspended(spa))
2632 VERIFY(nvlist_add_uint64(*config,
2633 ZPOOL_CONFIG_SUSPENDED,
2634 spa->spa_failmode) == 0);
2635
2636 spa_add_spares(spa, *config);
2637 spa_add_l2cache(spa, *config);
2638 }
2639 }
2640
2641 /*
2642 * We want to get the alternate root even for faulted pools, so we cheat
2643 * and call spa_lookup() directly.
2644 */
2645 if (altroot) {
2646 if (spa == NULL) {
2647 mutex_enter(&spa_namespace_lock);
2648 spa = spa_lookup(name);
2649 if (spa)
2650 spa_altroot(spa, altroot, buflen);
2651 else
2652 altroot[0] = '\0';
2653 spa = NULL;
2654 mutex_exit(&spa_namespace_lock);
2655 } else {
2656 spa_altroot(spa, altroot, buflen);
2657 }
2658 }
2659
2660 if (spa != NULL) {
2661 spa_config_exit(spa, SCL_CONFIG, FTAG);
2662 spa_close(spa, FTAG);
2663 }
2664
2665 return (error);
2666 }
2667
2668 /*
2669 * Validate that the auxiliary device array is well formed. We must have an
2670 * array of nvlists, each which describes a valid leaf vdev. If this is an
2671 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2672 * specified, as long as they are well-formed.
2673 */
2674 static int
2675 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2676 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2677 vdev_labeltype_t label)
2678 {
2679 nvlist_t **dev;
2680 uint_t i, ndev;
2681 vdev_t *vd;
2682 int error;
2683
2684 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2685
2686 /*
2687 * It's acceptable to have no devs specified.
2688 */
2689 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2690 return (0);
2691
2692 if (ndev == 0)
2693 return (EINVAL);
2694
2695 /*
2696 * Make sure the pool is formatted with a version that supports this
2697 * device type.
2698 */
2699 if (spa_version(spa) < version)
2700 return (ENOTSUP);
2701
2702 /*
2703 * Set the pending device list so we correctly handle device in-use
2704 * checking.
2705 */
2706 sav->sav_pending = dev;
2707 sav->sav_npending = ndev;
2708
2709 for (i = 0; i < ndev; i++) {
2710 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2711 mode)) != 0)
2712 goto out;
2713
2714 if (!vd->vdev_ops->vdev_op_leaf) {
2715 vdev_free(vd);
2716 error = EINVAL;
2717 goto out;
2718 }
2719
2720 /*
2721 * The L2ARC currently only supports disk devices in
2722 * kernel context. For user-level testing, we allow it.
2723 */
2724 #ifdef _KERNEL
2725 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2726 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2727 error = ENOTBLK;
2728 goto out;
2729 }
2730 #endif
2731 vd->vdev_top = vd;
2732
2733 if ((error = vdev_open(vd)) == 0 &&
2734 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2735 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2736 vd->vdev_guid) == 0);
2737 }
2738
2739 vdev_free(vd);
2740
2741 if (error &&
2742 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2743 goto out;
2744 else
2745 error = 0;
2746 }
2747
2748 out:
2749 sav->sav_pending = NULL;
2750 sav->sav_npending = 0;
2751 return (error);
2752 }
2753
2754 static int
2755 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2756 {
2757 int error;
2758
2759 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2760
2761 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2762 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2763 VDEV_LABEL_SPARE)) != 0) {
2764 return (error);
2765 }
2766
2767 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2768 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2769 VDEV_LABEL_L2CACHE));
2770 }
2771
2772 static void
2773 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2774 const char *config)
2775 {
2776 int i;
2777
2778 if (sav->sav_config != NULL) {
2779 nvlist_t **olddevs;
2780 uint_t oldndevs;
2781 nvlist_t **newdevs;
2782
2783 /*
2784 * Generate new dev list by concatentating with the
2785 * current dev list.
2786 */
2787 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2788 &olddevs, &oldndevs) == 0);
2789
2790 newdevs = kmem_alloc(sizeof (void *) *
2791 (ndevs + oldndevs), KM_SLEEP);
2792 for (i = 0; i < oldndevs; i++)
2793 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2794 KM_SLEEP) == 0);
2795 for (i = 0; i < ndevs; i++)
2796 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2797 KM_SLEEP) == 0);
2798
2799 VERIFY(nvlist_remove(sav->sav_config, config,
2800 DATA_TYPE_NVLIST_ARRAY) == 0);
2801
2802 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2803 config, newdevs, ndevs + oldndevs) == 0);
2804 for (i = 0; i < oldndevs + ndevs; i++)
2805 nvlist_free(newdevs[i]);
2806 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2807 } else {
2808 /*
2809 * Generate a new dev list.
2810 */
2811 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2812 KM_SLEEP) == 0);
2813 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2814 devs, ndevs) == 0);
2815 }
2816 }
2817
2818 /*
2819 * Stop and drop level 2 ARC devices
2820 */
2821 void
2822 spa_l2cache_drop(spa_t *spa)
2823 {
2824 vdev_t *vd;
2825 int i;
2826 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2827
2828 for (i = 0; i < sav->sav_count; i++) {
2829 uint64_t pool;
2830
2831 vd = sav->sav_vdevs[i];
2832 ASSERT(vd != NULL);
2833
2834 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2835 pool != 0ULL && l2arc_vdev_present(vd))
2836 l2arc_remove_vdev(vd);
2837 if (vd->vdev_isl2cache)
2838 spa_l2cache_remove(vd);
2839 vdev_clear_stats(vd);
2840 (void) vdev_close(vd);
2841 }
2842 }
2843
2844 /*
2845 * Pool Creation
2846 */
2847 int
2848 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2849 const char *history_str, nvlist_t *zplprops)
2850 {
2851 spa_t *spa;
2852 char *altroot = NULL;
2853 vdev_t *rvd;
2854 dsl_pool_t *dp;
2855 dmu_tx_t *tx;
2856 int error = 0;
2857 uint64_t txg = TXG_INITIAL;
2858 nvlist_t **spares, **l2cache;
2859 uint_t nspares, nl2cache;
2860 uint64_t version, obj;
2861 int c;
2862
2863 /*
2864 * If this pool already exists, return failure.
2865 */
2866 mutex_enter(&spa_namespace_lock);
2867 if (spa_lookup(pool) != NULL) {
2868 mutex_exit(&spa_namespace_lock);
2869 return (EEXIST);
2870 }
2871
2872 /*
2873 * Allocate a new spa_t structure.
2874 */
2875 (void) nvlist_lookup_string(props,
2876 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2877 spa = spa_add(pool, NULL, altroot);
2878 spa_activate(spa, spa_mode_global);
2879
2880 if (props && (error = spa_prop_validate(spa, props))) {
2881 spa_deactivate(spa);
2882 spa_remove(spa);
2883 mutex_exit(&spa_namespace_lock);
2884 return (error);
2885 }
2886
2887 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2888 &version) != 0)
2889 version = SPA_VERSION;
2890 ASSERT(version <= SPA_VERSION);
2891
2892 spa->spa_first_txg = txg;
2893 spa->spa_uberblock.ub_txg = txg - 1;
2894 spa->spa_uberblock.ub_version = version;
2895 spa->spa_ubsync = spa->spa_uberblock;
2896
2897 /*
2898 * Create "The Godfather" zio to hold all async IOs
2899 */
2900 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2901 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2902
2903 /*
2904 * Create the root vdev.
2905 */
2906 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2907
2908 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2909
2910 ASSERT(error != 0 || rvd != NULL);
2911 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2912
2913 if (error == 0 && !zfs_allocatable_devs(nvroot))
2914 error = EINVAL;
2915
2916 if (error == 0 &&
2917 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2918 (error = spa_validate_aux(spa, nvroot, txg,
2919 VDEV_ALLOC_ADD)) == 0) {
2920 for (c = 0; c < rvd->vdev_children; c++) {
2921 vdev_metaslab_set_size(rvd->vdev_child[c]);
2922 vdev_expand(rvd->vdev_child[c], txg);
2923 }
2924 }
2925
2926 spa_config_exit(spa, SCL_ALL, FTAG);
2927
2928 if (error != 0) {
2929 spa_unload(spa);
2930 spa_deactivate(spa);
2931 spa_remove(spa);
2932 mutex_exit(&spa_namespace_lock);
2933 return (error);
2934 }
2935
2936 /*
2937 * Get the list of spares, if specified.
2938 */
2939 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2940 &spares, &nspares) == 0) {
2941 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2942 KM_SLEEP) == 0);
2943 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2944 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2945 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2946 spa_load_spares(spa);
2947 spa_config_exit(spa, SCL_ALL, FTAG);
2948 spa->spa_spares.sav_sync = B_TRUE;
2949 }
2950
2951 /*
2952 * Get the list of level 2 cache devices, if specified.
2953 */
2954 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2955 &l2cache, &nl2cache) == 0) {
2956 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2957 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2958 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2959 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2960 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2961 spa_load_l2cache(spa);
2962 spa_config_exit(spa, SCL_ALL, FTAG);
2963 spa->spa_l2cache.sav_sync = B_TRUE;
2964 }
2965
2966 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2967 spa->spa_meta_objset = dp->dp_meta_objset;
2968
2969 /*
2970 * Create DDTs (dedup tables).
2971 */
2972 ddt_create(spa);
2973
2974 spa_update_dspace(spa);
2975
2976 tx = dmu_tx_create_assigned(dp, txg);
2977
2978 /*
2979 * Create the pool config object.
2980 */
2981 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2982 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2983 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2984
2985 if (zap_add(spa->spa_meta_objset,
2986 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2987 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2988 cmn_err(CE_PANIC, "failed to add pool config");
2989 }
2990
2991 if (zap_add(spa->spa_meta_objset,
2992 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2993 sizeof (uint64_t), 1, &version, tx) != 0) {
2994 cmn_err(CE_PANIC, "failed to add pool version");
2995 }
2996
2997 /* Newly created pools with the right version are always deflated. */
2998 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2999 spa->spa_deflate = TRUE;
3000 if (zap_add(spa->spa_meta_objset,
3001 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3002 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3003 cmn_err(CE_PANIC, "failed to add deflate");
3004 }
3005 }
3006
3007 /*
3008 * Create the deferred-free bpobj. Turn off compression
3009 * because sync-to-convergence takes longer if the blocksize
3010 * keeps changing.
3011 */
3012 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3013 dmu_object_set_compress(spa->spa_meta_objset, obj,
3014 ZIO_COMPRESS_OFF, tx);
3015 if (zap_add(spa->spa_meta_objset,
3016 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3017 sizeof (uint64_t), 1, &obj, tx) != 0) {
3018 cmn_err(CE_PANIC, "failed to add bpobj");
3019 }
3020 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3021 spa->spa_meta_objset, obj));
3022
3023 /*
3024 * Create the pool's history object.
3025 */
3026 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3027 spa_history_create_obj(spa, tx);
3028
3029 /*
3030 * Set pool properties.
3031 */
3032 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3033 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3034 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3035 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3036
3037 if (props != NULL) {
3038 spa_configfile_set(spa, props, B_FALSE);
3039 spa_sync_props(spa, props, tx);
3040 }
3041
3042 dmu_tx_commit(tx);
3043
3044 spa->spa_sync_on = B_TRUE;
3045 txg_sync_start(spa->spa_dsl_pool);
3046
3047 /*
3048 * We explicitly wait for the first transaction to complete so that our
3049 * bean counters are appropriately updated.
3050 */
3051 txg_wait_synced(spa->spa_dsl_pool, txg);
3052
3053 spa_config_sync(spa, B_FALSE, B_TRUE);
3054
3055 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3056 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3057 spa_history_log_version(spa, LOG_POOL_CREATE);
3058
3059 spa->spa_minref = refcount_count(&spa->spa_refcount);
3060
3061 mutex_exit(&spa_namespace_lock);
3062
3063 return (0);
3064 }
3065
3066 #ifdef _KERNEL
3067 /*
3068 * Get the root pool information from the root disk, then import the root pool
3069 * during the system boot up time.
3070 */
3071 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3072
3073 static nvlist_t *
3074 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3075 {
3076 nvlist_t *config;
3077 nvlist_t *nvtop, *nvroot;
3078 uint64_t pgid;
3079
3080 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3081 return (NULL);
3082
3083 /*
3084 * Add this top-level vdev to the child array.
3085 */
3086 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3087 &nvtop) == 0);
3088 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3089 &pgid) == 0);
3090 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3091
3092 /*
3093 * Put this pool's top-level vdevs into a root vdev.
3094 */
3095 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3096 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3097 VDEV_TYPE_ROOT) == 0);
3098 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3099 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3100 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3101 &nvtop, 1) == 0);
3102
3103 /*
3104 * Replace the existing vdev_tree with the new root vdev in
3105 * this pool's configuration (remove the old, add the new).
3106 */
3107 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3108 nvlist_free(nvroot);
3109 return (config);
3110 }
3111
3112 /*
3113 * Walk the vdev tree and see if we can find a device with "better"
3114 * configuration. A configuration is "better" if the label on that
3115 * device has a more recent txg.
3116 */
3117 static void
3118 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3119 {
3120 int c;
3121
3122 for (c = 0; c < vd->vdev_children; c++)
3123 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3124
3125 if (vd->vdev_ops->vdev_op_leaf) {
3126 nvlist_t *label;
3127 uint64_t label_txg;
3128
3129 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3130 &label) != 0)
3131 return;
3132
3133 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3134 &label_txg) == 0);
3135
3136 /*
3137 * Do we have a better boot device?
3138 */
3139 if (label_txg > *txg) {
3140 *txg = label_txg;
3141 *avd = vd;
3142 }
3143 nvlist_free(label);
3144 }
3145 }
3146
3147 /*
3148 * Import a root pool.
3149 *
3150 * For x86. devpath_list will consist of devid and/or physpath name of
3151 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3152 * The GRUB "findroot" command will return the vdev we should boot.
3153 *
3154 * For Sparc, devpath_list consists the physpath name of the booting device
3155 * no matter the rootpool is a single device pool or a mirrored pool.
3156 * e.g.
3157 * "/pci@1f,0/ide@d/disk@0,0:a"
3158 */
3159 int
3160 spa_import_rootpool(char *devpath, char *devid)
3161 {
3162 spa_t *spa;
3163 vdev_t *rvd, *bvd, *avd = NULL;
3164 nvlist_t *config, *nvtop;
3165 uint64_t guid, txg;
3166 char *pname;
3167 int error;
3168
3169 /*
3170 * Read the label from the boot device and generate a configuration.
3171 */
3172 config = spa_generate_rootconf(devpath, devid, &guid);
3173 #if defined(_OBP) && defined(_KERNEL)
3174 if (config == NULL) {
3175 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3176 /* iscsi boot */
3177 get_iscsi_bootpath_phy(devpath);
3178 config = spa_generate_rootconf(devpath, devid, &guid);
3179 }
3180 }
3181 #endif
3182 if (config == NULL) {
3183 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3184 devpath);
3185 return (EIO);
3186 }
3187
3188 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3189 &pname) == 0);
3190 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3191
3192 mutex_enter(&spa_namespace_lock);
3193 if ((spa = spa_lookup(pname)) != NULL) {
3194 /*
3195 * Remove the existing root pool from the namespace so that we
3196 * can replace it with the correct config we just read in.
3197 */
3198 spa_remove(spa);
3199 }
3200
3201 spa = spa_add(pname, config, NULL);
3202 spa->spa_is_root = B_TRUE;
3203 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3204
3205 /*
3206 * Build up a vdev tree based on the boot device's label config.
3207 */
3208 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3209 &nvtop) == 0);
3210 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3211 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3212 VDEV_ALLOC_ROOTPOOL);
3213 spa_config_exit(spa, SCL_ALL, FTAG);
3214 if (error) {
3215 mutex_exit(&spa_namespace_lock);
3216 nvlist_free(config);
3217 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3218 pname);
3219 return (error);
3220 }
3221
3222 /*
3223 * Get the boot vdev.
3224 */
3225 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3226 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3227 (u_longlong_t)guid);
3228 error = ENOENT;
3229 goto out;
3230 }
3231
3232 /*
3233 * Determine if there is a better boot device.
3234 */
3235 avd = bvd;
3236 spa_alt_rootvdev(rvd, &avd, &txg);
3237 if (avd != bvd) {
3238 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3239 "try booting from '%s'", avd->vdev_path);
3240 error = EINVAL;
3241 goto out;
3242 }
3243
3244 /*
3245 * If the boot device is part of a spare vdev then ensure that
3246 * we're booting off the active spare.
3247 */
3248 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3249 !bvd->vdev_isspare) {
3250 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3251 "try booting from '%s'",
3252 bvd->vdev_parent->
3253 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3254 error = EINVAL;
3255 goto out;
3256 }
3257
3258 error = 0;
3259 spa_history_log_version(spa, LOG_POOL_IMPORT);
3260 out:
3261 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3262 vdev_free(rvd);
3263 spa_config_exit(spa, SCL_ALL, FTAG);
3264 mutex_exit(&spa_namespace_lock);
3265
3266 nvlist_free(config);
3267 return (error);
3268 }
3269
3270 #endif
3271
3272 /*
3273 * Import a non-root pool into the system.
3274 */
3275 int
3276 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3277 {
3278 spa_t *spa;
3279 char *altroot = NULL;
3280 spa_load_state_t state = SPA_LOAD_IMPORT;
3281 zpool_rewind_policy_t policy;
3282 uint64_t mode = spa_mode_global;
3283 uint64_t readonly = B_FALSE;
3284 int error;
3285 nvlist_t *nvroot;
3286 nvlist_t **spares, **l2cache;
3287 uint_t nspares, nl2cache;
3288
3289 /*
3290 * If a pool with this name exists, return failure.
3291 */
3292 mutex_enter(&spa_namespace_lock);
3293 if (spa_lookup(pool) != NULL) {
3294 mutex_exit(&spa_namespace_lock);
3295 return (EEXIST);
3296 }
3297
3298 /*
3299 * Create and initialize the spa structure.
3300 */
3301 (void) nvlist_lookup_string(props,
3302 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3303 (void) nvlist_lookup_uint64(props,
3304 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3305 if (readonly)
3306 mode = FREAD;
3307 spa = spa_add(pool, config, altroot);
3308 spa->spa_import_flags = flags;
3309
3310 /*
3311 * Verbatim import - Take a pool and insert it into the namespace
3312 * as if it had been loaded at boot.
3313 */
3314 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3315 if (props != NULL)
3316 spa_configfile_set(spa, props, B_FALSE);
3317
3318 spa_config_sync(spa, B_FALSE, B_TRUE);
3319
3320 mutex_exit(&spa_namespace_lock);
3321 spa_history_log_version(spa, LOG_POOL_IMPORT);
3322
3323 return (0);
3324 }
3325
3326 spa_activate(spa, mode);
3327
3328 /*
3329 * Don't start async tasks until we know everything is healthy.
3330 */
3331 spa_async_suspend(spa);
3332
3333 zpool_get_rewind_policy(config, &policy);
3334 if (policy.zrp_request & ZPOOL_DO_REWIND)
3335 state = SPA_LOAD_RECOVER;
3336
3337 /*
3338 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3339 * because the user-supplied config is actually the one to trust when
3340 * doing an import.
3341 */
3342 if (state != SPA_LOAD_RECOVER)
3343 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3344
3345 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3346 policy.zrp_request);
3347
3348 /*
3349 * Propagate anything learned while loading the pool and pass it
3350 * back to caller (i.e. rewind info, missing devices, etc).
3351 */
3352 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3353 spa->spa_load_info) == 0);
3354
3355 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3356 /*
3357 * Toss any existing sparelist, as it doesn't have any validity
3358 * anymore, and conflicts with spa_has_spare().
3359 */
3360 if (spa->spa_spares.sav_config) {
3361 nvlist_free(spa->spa_spares.sav_config);
3362 spa->spa_spares.sav_config = NULL;
3363 spa_load_spares(spa);
3364 }
3365 if (spa->spa_l2cache.sav_config) {
3366 nvlist_free(spa->spa_l2cache.sav_config);
3367 spa->spa_l2cache.sav_config = NULL;
3368 spa_load_l2cache(spa);
3369 }
3370
3371 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3372 &nvroot) == 0);
3373 if (error == 0)
3374 error = spa_validate_aux(spa, nvroot, -1ULL,
3375 VDEV_ALLOC_SPARE);
3376 if (error == 0)
3377 error = spa_validate_aux(spa, nvroot, -1ULL,
3378 VDEV_ALLOC_L2CACHE);
3379 spa_config_exit(spa, SCL_ALL, FTAG);
3380
3381 if (props != NULL)
3382 spa_configfile_set(spa, props, B_FALSE);
3383
3384 if (error != 0 || (props && spa_writeable(spa) &&
3385 (error = spa_prop_set(spa, props)))) {
3386 spa_unload(spa);
3387 spa_deactivate(spa);
3388 spa_remove(spa);
3389 mutex_exit(&spa_namespace_lock);
3390 return (error);
3391 }
3392
3393 spa_async_resume(spa);
3394
3395 /*
3396 * Override any spares and level 2 cache devices as specified by
3397 * the user, as these may have correct device names/devids, etc.
3398 */
3399 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3400 &spares, &nspares) == 0) {
3401 if (spa->spa_spares.sav_config)
3402 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3403 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3404 else
3405 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3406 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3407 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3408 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3409 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3410 spa_load_spares(spa);
3411 spa_config_exit(spa, SCL_ALL, FTAG);
3412 spa->spa_spares.sav_sync = B_TRUE;
3413 }
3414 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3415 &l2cache, &nl2cache) == 0) {
3416 if (spa->spa_l2cache.sav_config)
3417 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3418 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3419 else
3420 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3421 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3422 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3423 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3424 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3425 spa_load_l2cache(spa);
3426 spa_config_exit(spa, SCL_ALL, FTAG);
3427 spa->spa_l2cache.sav_sync = B_TRUE;
3428 }
3429
3430 /*
3431 * Check for any removed devices.
3432 */
3433 if (spa->spa_autoreplace) {
3434 spa_aux_check_removed(&spa->spa_spares);
3435 spa_aux_check_removed(&spa->spa_l2cache);
3436 }
3437
3438 if (spa_writeable(spa)) {
3439 /*
3440 * Update the config cache to include the newly-imported pool.
3441 */
3442 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3443 }
3444
3445 /*
3446 * It's possible that the pool was expanded while it was exported.
3447 * We kick off an async task to handle this for us.
3448 */
3449 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3450
3451 mutex_exit(&spa_namespace_lock);
3452 spa_history_log_version(spa, LOG_POOL_IMPORT);
3453
3454 return (0);
3455 }
3456
3457 nvlist_t *
3458 spa_tryimport(nvlist_t *tryconfig)
3459 {
3460 nvlist_t *config = NULL;
3461 char *poolname;
3462 spa_t *spa;
3463 uint64_t state;
3464 int error;
3465
3466 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3467 return (NULL);
3468
3469 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3470 return (NULL);
3471
3472 /*
3473 * Create and initialize the spa structure.
3474 */
3475 mutex_enter(&spa_namespace_lock);
3476 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3477 spa_activate(spa, FREAD);
3478
3479 /*
3480 * Pass off the heavy lifting to spa_load().
3481 * Pass TRUE for mosconfig because the user-supplied config
3482 * is actually the one to trust when doing an import.
3483 */
3484 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3485
3486 /*
3487 * If 'tryconfig' was at least parsable, return the current config.
3488 */
3489 if (spa->spa_root_vdev != NULL) {
3490 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3491 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3492 poolname) == 0);
3493 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3494 state) == 0);
3495 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3496 spa->spa_uberblock.ub_timestamp) == 0);
3497
3498 /*
3499 * If the bootfs property exists on this pool then we
3500 * copy it out so that external consumers can tell which
3501 * pools are bootable.
3502 */
3503 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3504 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3505
3506 /*
3507 * We have to play games with the name since the
3508 * pool was opened as TRYIMPORT_NAME.
3509 */
3510 if (dsl_dsobj_to_dsname(spa_name(spa),
3511 spa->spa_bootfs, tmpname) == 0) {
3512 char *cp;
3513 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3514
3515 cp = strchr(tmpname, '/');
3516 if (cp == NULL) {
3517 (void) strlcpy(dsname, tmpname,
3518 MAXPATHLEN);
3519 } else {
3520 (void) snprintf(dsname, MAXPATHLEN,
3521 "%s/%s", poolname, ++cp);
3522 }
3523 VERIFY(nvlist_add_string(config,
3524 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3525 kmem_free(dsname, MAXPATHLEN);
3526 }
3527 kmem_free(tmpname, MAXPATHLEN);
3528 }
3529
3530 /*
3531 * Add the list of hot spares and level 2 cache devices.
3532 */
3533 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3534 spa_add_spares(spa, config);
3535 spa_add_l2cache(spa, config);
3536 spa_config_exit(spa, SCL_CONFIG, FTAG);
3537 }
3538
3539 spa_unload(spa);
3540 spa_deactivate(spa);
3541 spa_remove(spa);
3542 mutex_exit(&spa_namespace_lock);
3543
3544 return (config);
3545 }
3546
3547 /*
3548 * Pool export/destroy
3549 *
3550 * The act of destroying or exporting a pool is very simple. We make sure there
3551 * is no more pending I/O and any references to the pool are gone. Then, we
3552 * update the pool state and sync all the labels to disk, removing the
3553 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3554 * we don't sync the labels or remove the configuration cache.
3555 */
3556 static int
3557 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3558 boolean_t force, boolean_t hardforce)
3559 {
3560 spa_t *spa;
3561
3562 if (oldconfig)
3563 *oldconfig = NULL;
3564
3565 if (!(spa_mode_global & FWRITE))
3566 return (EROFS);
3567
3568 mutex_enter(&spa_namespace_lock);
3569 if ((spa = spa_lookup(pool)) == NULL) {
3570 mutex_exit(&spa_namespace_lock);
3571 return (ENOENT);
3572 }
3573
3574 /*
3575 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3576 * reacquire the namespace lock, and see if we can export.
3577 */
3578 spa_open_ref(spa, FTAG);
3579 mutex_exit(&spa_namespace_lock);
3580 spa_async_suspend(spa);
3581 mutex_enter(&spa_namespace_lock);
3582 spa_close(spa, FTAG);
3583
3584 /*
3585 * The pool will be in core if it's openable,
3586 * in which case we can modify its state.
3587 */
3588 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3589 /*
3590 * Objsets may be open only because they're dirty, so we
3591 * have to force it to sync before checking spa_refcnt.
3592 */
3593 txg_wait_synced(spa->spa_dsl_pool, 0);
3594
3595 /*
3596 * A pool cannot be exported or destroyed if there are active
3597 * references. If we are resetting a pool, allow references by
3598 * fault injection handlers.
3599 */
3600 if (!spa_refcount_zero(spa) ||
3601 (spa->spa_inject_ref != 0 &&
3602 new_state != POOL_STATE_UNINITIALIZED)) {
3603 spa_async_resume(spa);
3604 mutex_exit(&spa_namespace_lock);
3605 return (EBUSY);
3606 }
3607
3608 /*
3609 * A pool cannot be exported if it has an active shared spare.
3610 * This is to prevent other pools stealing the active spare
3611 * from an exported pool. At user's own will, such pool can
3612 * be forcedly exported.
3613 */
3614 if (!force && new_state == POOL_STATE_EXPORTED &&
3615 spa_has_active_shared_spare(spa)) {
3616 spa_async_resume(spa);
3617 mutex_exit(&spa_namespace_lock);
3618 return (EXDEV);
3619 }
3620
3621 /*
3622 * We want this to be reflected on every label,
3623 * so mark them all dirty. spa_unload() will do the
3624 * final sync that pushes these changes out.
3625 */
3626 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3627 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3628 spa->spa_state = new_state;
3629 spa->spa_final_txg = spa_last_synced_txg(spa) +
3630 TXG_DEFER_SIZE + 1;
3631 vdev_config_dirty(spa->spa_root_vdev);
3632 spa_config_exit(spa, SCL_ALL, FTAG);
3633 }
3634 }
3635
3636 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3637
3638 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3639 spa_unload(spa);
3640 spa_deactivate(spa);
3641 }
3642
3643 if (oldconfig && spa->spa_config)
3644 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3645
3646 if (new_state != POOL_STATE_UNINITIALIZED) {
3647 if (!hardforce)
3648 spa_config_sync(spa, B_TRUE, B_TRUE);
3649 spa_remove(spa);
3650 }
3651 mutex_exit(&spa_namespace_lock);
3652
3653 return (0);
3654 }
3655
3656 /*
3657 * Destroy a storage pool.
3658 */
3659 int
3660 spa_destroy(char *pool)
3661 {
3662 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3663 B_FALSE, B_FALSE));
3664 }
3665
3666 /*
3667 * Export a storage pool.
3668 */
3669 int
3670 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3671 boolean_t hardforce)
3672 {
3673 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3674 force, hardforce));
3675 }
3676
3677 /*
3678 * Similar to spa_export(), this unloads the spa_t without actually removing it
3679 * from the namespace in any way.
3680 */
3681 int
3682 spa_reset(char *pool)
3683 {
3684 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3685 B_FALSE, B_FALSE));
3686 }
3687
3688 /*
3689 * ==========================================================================
3690 * Device manipulation
3691 * ==========================================================================
3692 */
3693
3694 /*
3695 * Add a device to a storage pool.
3696 */
3697 int
3698 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3699 {
3700 uint64_t txg, id;
3701 int error;
3702 vdev_t *rvd = spa->spa_root_vdev;
3703 vdev_t *vd, *tvd;
3704 nvlist_t **spares, **l2cache;
3705 uint_t nspares, nl2cache;
3706 int c;
3707
3708 ASSERT(spa_writeable(spa));
3709
3710 txg = spa_vdev_enter(spa);
3711
3712 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3713 VDEV_ALLOC_ADD)) != 0)
3714 return (spa_vdev_exit(spa, NULL, txg, error));
3715
3716 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3717
3718 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3719 &nspares) != 0)
3720 nspares = 0;
3721
3722 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3723 &nl2cache) != 0)
3724 nl2cache = 0;
3725
3726 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3727 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3728
3729 if (vd->vdev_children != 0 &&
3730 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3731 return (spa_vdev_exit(spa, vd, txg, error));
3732
3733 /*
3734 * We must validate the spares and l2cache devices after checking the
3735 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3736 */
3737 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3738 return (spa_vdev_exit(spa, vd, txg, error));
3739
3740 /*
3741 * Transfer each new top-level vdev from vd to rvd.
3742 */
3743 for (c = 0; c < vd->vdev_children; c++) {
3744
3745 /*
3746 * Set the vdev id to the first hole, if one exists.
3747 */
3748 for (id = 0; id < rvd->vdev_children; id++) {
3749 if (rvd->vdev_child[id]->vdev_ishole) {
3750 vdev_free(rvd->vdev_child[id]);
3751 break;
3752 }
3753 }
3754 tvd = vd->vdev_child[c];
3755 vdev_remove_child(vd, tvd);
3756 tvd->vdev_id = id;
3757 vdev_add_child(rvd, tvd);
3758 vdev_config_dirty(tvd);
3759 }
3760
3761 if (nspares != 0) {
3762 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3763 ZPOOL_CONFIG_SPARES);
3764 spa_load_spares(spa);
3765 spa->spa_spares.sav_sync = B_TRUE;
3766 }
3767
3768 if (nl2cache != 0) {
3769 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3770 ZPOOL_CONFIG_L2CACHE);
3771 spa_load_l2cache(spa);
3772 spa->spa_l2cache.sav_sync = B_TRUE;
3773 }
3774
3775 /*
3776 * We have to be careful when adding new vdevs to an existing pool.
3777 * If other threads start allocating from these vdevs before we
3778 * sync the config cache, and we lose power, then upon reboot we may
3779 * fail to open the pool because there are DVAs that the config cache
3780 * can't translate. Therefore, we first add the vdevs without
3781 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3782 * and then let spa_config_update() initialize the new metaslabs.
3783 *
3784 * spa_load() checks for added-but-not-initialized vdevs, so that
3785 * if we lose power at any point in this sequence, the remaining
3786 * steps will be completed the next time we load the pool.
3787 */
3788 (void) spa_vdev_exit(spa, vd, txg, 0);
3789
3790 mutex_enter(&spa_namespace_lock);
3791 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3792 mutex_exit(&spa_namespace_lock);
3793
3794 return (0);
3795 }
3796
3797 /*
3798 * Attach a device to a mirror. The arguments are the path to any device
3799 * in the mirror, and the nvroot for the new device. If the path specifies
3800 * a device that is not mirrored, we automatically insert the mirror vdev.
3801 *
3802 * If 'replacing' is specified, the new device is intended to replace the
3803 * existing device; in this case the two devices are made into their own
3804 * mirror using the 'replacing' vdev, which is functionally identical to
3805 * the mirror vdev (it actually reuses all the same ops) but has a few
3806 * extra rules: you can't attach to it after it's been created, and upon
3807 * completion of resilvering, the first disk (the one being replaced)
3808 * is automatically detached.
3809 */
3810 int
3811 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3812 {
3813 uint64_t txg, dtl_max_txg;
3814 vdev_t *rvd = spa->spa_root_vdev;
3815 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3816 vdev_ops_t *pvops;
3817 char *oldvdpath, *newvdpath;
3818 int newvd_isspare;
3819 int error;
3820
3821 ASSERT(spa_writeable(spa));
3822
3823 txg = spa_vdev_enter(spa);
3824
3825 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3826
3827 if (oldvd == NULL)
3828 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3829
3830 if (!oldvd->vdev_ops->vdev_op_leaf)
3831 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3832
3833 pvd = oldvd->vdev_parent;
3834
3835 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3836 VDEV_ALLOC_ADD)) != 0)
3837 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3838
3839 if (newrootvd->vdev_children != 1)
3840 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3841
3842 newvd = newrootvd->vdev_child[0];
3843
3844 if (!newvd->vdev_ops->vdev_op_leaf)
3845 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3846
3847 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3848 return (spa_vdev_exit(spa, newrootvd, txg, error));
3849
3850 /*
3851 * Spares can't replace logs
3852 */
3853 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3854 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3855
3856 if (!replacing) {
3857 /*
3858 * For attach, the only allowable parent is a mirror or the root
3859 * vdev.
3860 */
3861 if (pvd->vdev_ops != &vdev_mirror_ops &&
3862 pvd->vdev_ops != &vdev_root_ops)
3863 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3864
3865 pvops = &vdev_mirror_ops;
3866 } else {
3867 /*
3868 * Active hot spares can only be replaced by inactive hot
3869 * spares.
3870 */
3871 if (pvd->vdev_ops == &vdev_spare_ops &&
3872 oldvd->vdev_isspare &&
3873 !spa_has_spare(spa, newvd->vdev_guid))
3874 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3875
3876 /*
3877 * If the source is a hot spare, and the parent isn't already a
3878 * spare, then we want to create a new hot spare. Otherwise, we
3879 * want to create a replacing vdev. The user is not allowed to
3880 * attach to a spared vdev child unless the 'isspare' state is
3881 * the same (spare replaces spare, non-spare replaces
3882 * non-spare).
3883 */
3884 if (pvd->vdev_ops == &vdev_replacing_ops &&
3885 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3886 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3887 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3888 newvd->vdev_isspare != oldvd->vdev_isspare) {
3889 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3890 }
3891
3892 if (newvd->vdev_isspare)
3893 pvops = &vdev_spare_ops;
3894 else
3895 pvops = &vdev_replacing_ops;
3896 }
3897
3898 /*
3899 * Make sure the new device is big enough.
3900 */
3901 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3902 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3903
3904 /*
3905 * The new device cannot have a higher alignment requirement
3906 * than the top-level vdev.
3907 */
3908 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3909 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3910
3911 /*
3912 * If this is an in-place replacement, update oldvd's path and devid
3913 * to make it distinguishable from newvd, and unopenable from now on.
3914 */
3915 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3916 spa_strfree(oldvd->vdev_path);
3917 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3918 KM_SLEEP);
3919 (void) sprintf(oldvd->vdev_path, "%s/%s",
3920 newvd->vdev_path, "old");
3921 if (oldvd->vdev_devid != NULL) {
3922 spa_strfree(oldvd->vdev_devid);
3923 oldvd->vdev_devid = NULL;
3924 }
3925 }
3926
3927 /* mark the device being resilvered */
3928 newvd->vdev_resilvering = B_TRUE;
3929
3930 /*
3931 * If the parent is not a mirror, or if we're replacing, insert the new
3932 * mirror/replacing/spare vdev above oldvd.
3933 */
3934 if (pvd->vdev_ops != pvops)
3935 pvd = vdev_add_parent(oldvd, pvops);
3936
3937 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3938 ASSERT(pvd->vdev_ops == pvops);
3939 ASSERT(oldvd->vdev_parent == pvd);
3940
3941 /*
3942 * Extract the new device from its root and add it to pvd.
3943 */
3944 vdev_remove_child(newrootvd, newvd);
3945 newvd->vdev_id = pvd->vdev_children;
3946 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3947 vdev_add_child(pvd, newvd);
3948
3949 tvd = newvd->vdev_top;
3950 ASSERT(pvd->vdev_top == tvd);
3951 ASSERT(tvd->vdev_parent == rvd);
3952
3953 vdev_config_dirty(tvd);
3954
3955 /*
3956 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3957 * for any dmu_sync-ed blocks. It will propagate upward when
3958 * spa_vdev_exit() calls vdev_dtl_reassess().
3959 */
3960 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3961
3962 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3963 dtl_max_txg - TXG_INITIAL);
3964
3965 if (newvd->vdev_isspare) {
3966 spa_spare_activate(newvd);
3967 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3968 }
3969
3970 oldvdpath = spa_strdup(oldvd->vdev_path);
3971 newvdpath = spa_strdup(newvd->vdev_path);
3972 newvd_isspare = newvd->vdev_isspare;
3973
3974 /*
3975 * Mark newvd's DTL dirty in this txg.
3976 */
3977 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3978
3979 /*
3980 * Restart the resilver
3981 */
3982 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3983
3984 /*
3985 * Commit the config
3986 */
3987 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3988
3989 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3990 "%s vdev=%s %s vdev=%s",
3991 replacing && newvd_isspare ? "spare in" :
3992 replacing ? "replace" : "attach", newvdpath,
3993 replacing ? "for" : "to", oldvdpath);
3994
3995 spa_strfree(oldvdpath);
3996 spa_strfree(newvdpath);
3997
3998 if (spa->spa_bootfs)
3999 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4000
4001 return (0);
4002 }
4003
4004 /*
4005 * Detach a device from a mirror or replacing vdev.
4006 * If 'replace_done' is specified, only detach if the parent
4007 * is a replacing vdev.
4008 */
4009 int
4010 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4011 {
4012 uint64_t txg;
4013 int error;
4014 vdev_t *rvd = spa->spa_root_vdev;
4015 vdev_t *vd, *pvd, *cvd, *tvd;
4016 boolean_t unspare = B_FALSE;
4017 uint64_t unspare_guid;
4018 char *vdpath;
4019 int c, t;
4020
4021 ASSERT(spa_writeable(spa));
4022
4023 txg = spa_vdev_enter(spa);
4024
4025 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4026
4027 if (vd == NULL)
4028 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4029
4030 if (!vd->vdev_ops->vdev_op_leaf)
4031 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4032
4033 pvd = vd->vdev_parent;
4034
4035 /*
4036 * If the parent/child relationship is not as expected, don't do it.
4037 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4038 * vdev that's replacing B with C. The user's intent in replacing
4039 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4040 * the replace by detaching C, the expected behavior is to end up
4041 * M(A,B). But suppose that right after deciding to detach C,
4042 * the replacement of B completes. We would have M(A,C), and then
4043 * ask to detach C, which would leave us with just A -- not what
4044 * the user wanted. To prevent this, we make sure that the
4045 * parent/child relationship hasn't changed -- in this example,
4046 * that C's parent is still the replacing vdev R.
4047 */
4048 if (pvd->vdev_guid != pguid && pguid != 0)
4049 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4050
4051 /*
4052 * Only 'replacing' or 'spare' vdevs can be replaced.
4053 */
4054 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4055 pvd->vdev_ops != &vdev_spare_ops)
4056 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4057
4058 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4059 spa_version(spa) >= SPA_VERSION_SPARES);
4060
4061 /*
4062 * Only mirror, replacing, and spare vdevs support detach.
4063 */
4064 if (pvd->vdev_ops != &vdev_replacing_ops &&
4065 pvd->vdev_ops != &vdev_mirror_ops &&
4066 pvd->vdev_ops != &vdev_spare_ops)
4067 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4068
4069 /*
4070 * If this device has the only valid copy of some data,
4071 * we cannot safely detach it.
4072 */
4073 if (vdev_dtl_required(vd))
4074 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4075
4076 ASSERT(pvd->vdev_children >= 2);
4077
4078 /*
4079 * If we are detaching the second disk from a replacing vdev, then
4080 * check to see if we changed the original vdev's path to have "/old"
4081 * at the end in spa_vdev_attach(). If so, undo that change now.
4082 */
4083 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4084 vd->vdev_path != NULL) {
4085 size_t len = strlen(vd->vdev_path);
4086
4087 for (c = 0; c < pvd->vdev_children; c++) {
4088 cvd = pvd->vdev_child[c];
4089
4090 if (cvd == vd || cvd->vdev_path == NULL)
4091 continue;
4092
4093 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4094 strcmp(cvd->vdev_path + len, "/old") == 0) {
4095 spa_strfree(cvd->vdev_path);
4096 cvd->vdev_path = spa_strdup(vd->vdev_path);
4097 break;
4098 }
4099 }
4100 }
4101
4102 /*
4103 * If we are detaching the original disk from a spare, then it implies
4104 * that the spare should become a real disk, and be removed from the
4105 * active spare list for the pool.
4106 */
4107 if (pvd->vdev_ops == &vdev_spare_ops &&
4108 vd->vdev_id == 0 &&
4109 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4110 unspare = B_TRUE;
4111
4112 /*
4113 * Erase the disk labels so the disk can be used for other things.
4114 * This must be done after all other error cases are handled,
4115 * but before we disembowel vd (so we can still do I/O to it).
4116 * But if we can't do it, don't treat the error as fatal --
4117 * it may be that the unwritability of the disk is the reason
4118 * it's being detached!
4119 */
4120 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4121
4122 /*
4123 * Remove vd from its parent and compact the parent's children.
4124 */
4125 vdev_remove_child(pvd, vd);
4126 vdev_compact_children(pvd);
4127
4128 /*
4129 * Remember one of the remaining children so we can get tvd below.
4130 */
4131 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4132
4133 /*
4134 * If we need to remove the remaining child from the list of hot spares,
4135 * do it now, marking the vdev as no longer a spare in the process.
4136 * We must do this before vdev_remove_parent(), because that can
4137 * change the GUID if it creates a new toplevel GUID. For a similar
4138 * reason, we must remove the spare now, in the same txg as the detach;
4139 * otherwise someone could attach a new sibling, change the GUID, and
4140 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4141 */
4142 if (unspare) {
4143 ASSERT(cvd->vdev_isspare);
4144 spa_spare_remove(cvd);
4145 unspare_guid = cvd->vdev_guid;
4146 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4147 cvd->vdev_unspare = B_TRUE;
4148 }
4149
4150 /*
4151 * If the parent mirror/replacing vdev only has one child,
4152 * the parent is no longer needed. Remove it from the tree.
4153 */
4154 if (pvd->vdev_children == 1) {
4155 if (pvd->vdev_ops == &vdev_spare_ops)
4156 cvd->vdev_unspare = B_FALSE;
4157 vdev_remove_parent(cvd);
4158 cvd->vdev_resilvering = B_FALSE;
4159 }
4160
4161
4162 /*
4163 * We don't set tvd until now because the parent we just removed
4164 * may have been the previous top-level vdev.
4165 */
4166 tvd = cvd->vdev_top;
4167 ASSERT(tvd->vdev_parent == rvd);
4168
4169 /*
4170 * Reevaluate the parent vdev state.
4171 */
4172 vdev_propagate_state(cvd);
4173
4174 /*
4175 * If the 'autoexpand' property is set on the pool then automatically
4176 * try to expand the size of the pool. For example if the device we
4177 * just detached was smaller than the others, it may be possible to
4178 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4179 * first so that we can obtain the updated sizes of the leaf vdevs.
4180 */
4181 if (spa->spa_autoexpand) {
4182 vdev_reopen(tvd);
4183 vdev_expand(tvd, txg);
4184 }
4185
4186 vdev_config_dirty(tvd);
4187
4188 /*
4189 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4190 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4191 * But first make sure we're not on any *other* txg's DTL list, to
4192 * prevent vd from being accessed after it's freed.
4193 */
4194 vdpath = spa_strdup(vd->vdev_path);
4195 for (t = 0; t < TXG_SIZE; t++)
4196 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4197 vd->vdev_detached = B_TRUE;
4198 vdev_dirty(tvd, VDD_DTL, vd, txg);
4199
4200 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4201
4202 /* hang on to the spa before we release the lock */
4203 spa_open_ref(spa, FTAG);
4204
4205 error = spa_vdev_exit(spa, vd, txg, 0);
4206
4207 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4208 "vdev=%s", vdpath);
4209 spa_strfree(vdpath);
4210
4211 /*
4212 * If this was the removal of the original device in a hot spare vdev,
4213 * then we want to go through and remove the device from the hot spare
4214 * list of every other pool.
4215 */
4216 if (unspare) {
4217 spa_t *altspa = NULL;
4218
4219 mutex_enter(&spa_namespace_lock);
4220 while ((altspa = spa_next(altspa)) != NULL) {
4221 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4222 altspa == spa)
4223 continue;
4224
4225 spa_open_ref(altspa, FTAG);
4226 mutex_exit(&spa_namespace_lock);
4227 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4228 mutex_enter(&spa_namespace_lock);
4229 spa_close(altspa, FTAG);
4230 }
4231 mutex_exit(&spa_namespace_lock);
4232
4233 /* search the rest of the vdevs for spares to remove */
4234 spa_vdev_resilver_done(spa);
4235 }
4236
4237 /* all done with the spa; OK to release */
4238 mutex_enter(&spa_namespace_lock);
4239 spa_close(spa, FTAG);
4240 mutex_exit(&spa_namespace_lock);
4241
4242 return (error);
4243 }
4244
4245 /*
4246 * Split a set of devices from their mirrors, and create a new pool from them.
4247 */
4248 int
4249 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4250 nvlist_t *props, boolean_t exp)
4251 {
4252 int error = 0;
4253 uint64_t txg, *glist;
4254 spa_t *newspa;
4255 uint_t c, children, lastlog;
4256 nvlist_t **child, *nvl, *tmp;
4257 dmu_tx_t *tx;
4258 char *altroot = NULL;
4259 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4260 boolean_t activate_slog;
4261
4262 ASSERT(spa_writeable(spa));
4263
4264 txg = spa_vdev_enter(spa);
4265
4266 /* clear the log and flush everything up to now */
4267 activate_slog = spa_passivate_log(spa);
4268 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4269 error = spa_offline_log(spa);
4270 txg = spa_vdev_config_enter(spa);
4271
4272 if (activate_slog)
4273 spa_activate_log(spa);
4274
4275 if (error != 0)
4276 return (spa_vdev_exit(spa, NULL, txg, error));
4277
4278 /* check new spa name before going any further */
4279 if (spa_lookup(newname) != NULL)
4280 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4281
4282 /*
4283 * scan through all the children to ensure they're all mirrors
4284 */
4285 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4286 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4287 &children) != 0)
4288 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4289
4290 /* first, check to ensure we've got the right child count */
4291 rvd = spa->spa_root_vdev;
4292 lastlog = 0;
4293 for (c = 0; c < rvd->vdev_children; c++) {
4294 vdev_t *vd = rvd->vdev_child[c];
4295
4296 /* don't count the holes & logs as children */
4297 if (vd->vdev_islog || vd->vdev_ishole) {
4298 if (lastlog == 0)
4299 lastlog = c;
4300 continue;
4301 }
4302
4303 lastlog = 0;
4304 }
4305 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4306 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4307
4308 /* next, ensure no spare or cache devices are part of the split */
4309 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4310 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4311 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4312
4313 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4314 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4315
4316 /* then, loop over each vdev and validate it */
4317 for (c = 0; c < children; c++) {
4318 uint64_t is_hole = 0;
4319
4320 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4321 &is_hole);
4322
4323 if (is_hole != 0) {
4324 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4325 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4326 continue;
4327 } else {
4328 error = EINVAL;
4329 break;
4330 }
4331 }
4332
4333 /* which disk is going to be split? */
4334 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4335 &glist[c]) != 0) {
4336 error = EINVAL;
4337 break;
4338 }
4339
4340 /* look it up in the spa */
4341 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4342 if (vml[c] == NULL) {
4343 error = ENODEV;
4344 break;
4345 }
4346
4347 /* make sure there's nothing stopping the split */
4348 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4349 vml[c]->vdev_islog ||
4350 vml[c]->vdev_ishole ||
4351 vml[c]->vdev_isspare ||
4352 vml[c]->vdev_isl2cache ||
4353 !vdev_writeable(vml[c]) ||
4354 vml[c]->vdev_children != 0 ||
4355 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4356 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4357 error = EINVAL;
4358 break;
4359 }
4360
4361 if (vdev_dtl_required(vml[c])) {
4362 error = EBUSY;
4363 break;
4364 }
4365
4366 /* we need certain info from the top level */
4367 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4368 vml[c]->vdev_top->vdev_ms_array) == 0);
4369 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4370 vml[c]->vdev_top->vdev_ms_shift) == 0);
4371 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4372 vml[c]->vdev_top->vdev_asize) == 0);
4373 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4374 vml[c]->vdev_top->vdev_ashift) == 0);
4375 }
4376
4377 if (error != 0) {
4378 kmem_free(vml, children * sizeof (vdev_t *));
4379 kmem_free(glist, children * sizeof (uint64_t));
4380 return (spa_vdev_exit(spa, NULL, txg, error));
4381 }
4382
4383 /* stop writers from using the disks */
4384 for (c = 0; c < children; c++) {
4385 if (vml[c] != NULL)
4386 vml[c]->vdev_offline = B_TRUE;
4387 }
4388 vdev_reopen(spa->spa_root_vdev);
4389
4390 /*
4391 * Temporarily record the splitting vdevs in the spa config. This
4392 * will disappear once the config is regenerated.
4393 */
4394 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4395 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4396 glist, children) == 0);
4397 kmem_free(glist, children * sizeof (uint64_t));
4398
4399 mutex_enter(&spa->spa_props_lock);
4400 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4401 nvl) == 0);
4402 mutex_exit(&spa->spa_props_lock);
4403 spa->spa_config_splitting = nvl;
4404 vdev_config_dirty(spa->spa_root_vdev);
4405
4406 /* configure and create the new pool */
4407 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4408 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4409 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4410 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4411 spa_version(spa)) == 0);
4412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4413 spa->spa_config_txg) == 0);
4414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4415 spa_generate_guid(NULL)) == 0);
4416 (void) nvlist_lookup_string(props,
4417 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4418
4419 /* add the new pool to the namespace */
4420 newspa = spa_add(newname, config, altroot);
4421 newspa->spa_config_txg = spa->spa_config_txg;
4422 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4423
4424 /* release the spa config lock, retaining the namespace lock */
4425 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4426
4427 if (zio_injection_enabled)
4428 zio_handle_panic_injection(spa, FTAG, 1);
4429
4430 spa_activate(newspa, spa_mode_global);
4431 spa_async_suspend(newspa);
4432
4433 /* create the new pool from the disks of the original pool */
4434 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4435 if (error)
4436 goto out;
4437
4438 /* if that worked, generate a real config for the new pool */
4439 if (newspa->spa_root_vdev != NULL) {
4440 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4441 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4442 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4443 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4444 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4445 B_TRUE));
4446 }
4447
4448 /* set the props */
4449 if (props != NULL) {
4450 spa_configfile_set(newspa, props, B_FALSE);
4451 error = spa_prop_set(newspa, props);
4452 if (error)
4453 goto out;
4454 }
4455
4456 /* flush everything */
4457 txg = spa_vdev_config_enter(newspa);
4458 vdev_config_dirty(newspa->spa_root_vdev);
4459 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4460
4461 if (zio_injection_enabled)
4462 zio_handle_panic_injection(spa, FTAG, 2);
4463
4464 spa_async_resume(newspa);
4465
4466 /* finally, update the original pool's config */
4467 txg = spa_vdev_config_enter(spa);
4468 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4469 error = dmu_tx_assign(tx, TXG_WAIT);
4470 if (error != 0)
4471 dmu_tx_abort(tx);
4472 for (c = 0; c < children; c++) {
4473 if (vml[c] != NULL) {
4474 vdev_split(vml[c]);
4475 if (error == 0)
4476 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4477 spa, tx, "vdev=%s",
4478 vml[c]->vdev_path);
4479 vdev_free(vml[c]);
4480 }
4481 }
4482 vdev_config_dirty(spa->spa_root_vdev);
4483 spa->spa_config_splitting = NULL;
4484 nvlist_free(nvl);
4485 if (error == 0)
4486 dmu_tx_commit(tx);
4487 (void) spa_vdev_exit(spa, NULL, txg, 0);
4488
4489 if (zio_injection_enabled)
4490 zio_handle_panic_injection(spa, FTAG, 3);
4491
4492 /* split is complete; log a history record */
4493 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4494 "split new pool %s from pool %s", newname, spa_name(spa));
4495
4496 kmem_free(vml, children * sizeof (vdev_t *));
4497
4498 /* if we're not going to mount the filesystems in userland, export */
4499 if (exp)
4500 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4501 B_FALSE, B_FALSE);
4502
4503 return (error);
4504
4505 out:
4506 spa_unload(newspa);
4507 spa_deactivate(newspa);
4508 spa_remove(newspa);
4509
4510 txg = spa_vdev_config_enter(spa);
4511
4512 /* re-online all offlined disks */
4513 for (c = 0; c < children; c++) {
4514 if (vml[c] != NULL)
4515 vml[c]->vdev_offline = B_FALSE;
4516 }
4517 vdev_reopen(spa->spa_root_vdev);
4518
4519 nvlist_free(spa->spa_config_splitting);
4520 spa->spa_config_splitting = NULL;
4521 (void) spa_vdev_exit(spa, NULL, txg, error);
4522
4523 kmem_free(vml, children * sizeof (vdev_t *));
4524 return (error);
4525 }
4526
4527 static nvlist_t *
4528 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4529 {
4530 int i;
4531
4532 for (i = 0; i < count; i++) {
4533 uint64_t guid;
4534
4535 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4536 &guid) == 0);
4537
4538 if (guid == target_guid)
4539 return (nvpp[i]);
4540 }
4541
4542 return (NULL);
4543 }
4544
4545 static void
4546 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4547 nvlist_t *dev_to_remove)
4548 {
4549 nvlist_t **newdev = NULL;
4550 int i, j;
4551
4552 if (count > 1)
4553 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4554
4555 for (i = 0, j = 0; i < count; i++) {
4556 if (dev[i] == dev_to_remove)
4557 continue;
4558 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4559 }
4560
4561 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4562 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4563
4564 for (i = 0; i < count - 1; i++)
4565 nvlist_free(newdev[i]);
4566
4567 if (count > 1)
4568 kmem_free(newdev, (count - 1) * sizeof (void *));
4569 }
4570
4571 /*
4572 * Evacuate the device.
4573 */
4574 static int
4575 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4576 {
4577 uint64_t txg;
4578 int error = 0;
4579
4580 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4581 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4582 ASSERT(vd == vd->vdev_top);
4583
4584 /*
4585 * Evacuate the device. We don't hold the config lock as writer
4586 * since we need to do I/O but we do keep the
4587 * spa_namespace_lock held. Once this completes the device
4588 * should no longer have any blocks allocated on it.
4589 */
4590 if (vd->vdev_islog) {
4591 if (vd->vdev_stat.vs_alloc != 0)
4592 error = spa_offline_log(spa);
4593 } else {
4594 error = ENOTSUP;
4595 }
4596
4597 if (error)
4598 return (error);
4599
4600 /*
4601 * The evacuation succeeded. Remove any remaining MOS metadata
4602 * associated with this vdev, and wait for these changes to sync.
4603 */
4604 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4605 txg = spa_vdev_config_enter(spa);
4606 vd->vdev_removing = B_TRUE;
4607 vdev_dirty(vd, 0, NULL, txg);
4608 vdev_config_dirty(vd);
4609 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4610
4611 return (0);
4612 }
4613
4614 /*
4615 * Complete the removal by cleaning up the namespace.
4616 */
4617 static void
4618 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4619 {
4620 vdev_t *rvd = spa->spa_root_vdev;
4621 uint64_t id = vd->vdev_id;
4622 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4623
4624 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4625 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4626 ASSERT(vd == vd->vdev_top);
4627
4628 /*
4629 * Only remove any devices which are empty.
4630 */
4631 if (vd->vdev_stat.vs_alloc != 0)
4632 return;
4633
4634 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4635
4636 if (list_link_active(&vd->vdev_state_dirty_node))
4637 vdev_state_clean(vd);
4638 if (list_link_active(&vd->vdev_config_dirty_node))
4639 vdev_config_clean(vd);
4640
4641 vdev_free(vd);
4642
4643 if (last_vdev) {
4644 vdev_compact_children(rvd);
4645 } else {
4646 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4647 vdev_add_child(rvd, vd);
4648 }
4649 vdev_config_dirty(rvd);
4650
4651 /*
4652 * Reassess the health of our root vdev.
4653 */
4654 vdev_reopen(rvd);
4655 }
4656
4657 /*
4658 * Remove a device from the pool -
4659 *
4660 * Removing a device from the vdev namespace requires several steps
4661 * and can take a significant amount of time. As a result we use
4662 * the spa_vdev_config_[enter/exit] functions which allow us to
4663 * grab and release the spa_config_lock while still holding the namespace
4664 * lock. During each step the configuration is synced out.
4665 */
4666
4667 /*
4668 * Remove a device from the pool. Currently, this supports removing only hot
4669 * spares, slogs, and level 2 ARC devices.
4670 */
4671 int
4672 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4673 {
4674 vdev_t *vd;
4675 metaslab_group_t *mg;
4676 nvlist_t **spares, **l2cache, *nv;
4677 uint64_t txg = 0;
4678 uint_t nspares, nl2cache;
4679 int error = 0;
4680 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4681
4682 ASSERT(spa_writeable(spa));
4683
4684 if (!locked)
4685 txg = spa_vdev_enter(spa);
4686
4687 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4688
4689 if (spa->spa_spares.sav_vdevs != NULL &&
4690 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4691 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4692 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4693 /*
4694 * Only remove the hot spare if it's not currently in use
4695 * in this pool.
4696 */
4697 if (vd == NULL || unspare) {
4698 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4699 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4700 spa_load_spares(spa);
4701 spa->spa_spares.sav_sync = B_TRUE;
4702 } else {
4703 error = EBUSY;
4704 }
4705 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4706 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4707 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4708 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4709 /*
4710 * Cache devices can always be removed.
4711 */
4712 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4713 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4714 spa_load_l2cache(spa);
4715 spa->spa_l2cache.sav_sync = B_TRUE;
4716 } else if (vd != NULL && vd->vdev_islog) {
4717 ASSERT(!locked);
4718 ASSERT(vd == vd->vdev_top);
4719
4720 /*
4721 * XXX - Once we have bp-rewrite this should
4722 * become the common case.
4723 */
4724
4725 mg = vd->vdev_mg;
4726
4727 /*
4728 * Stop allocating from this vdev.
4729 */
4730 metaslab_group_passivate(mg);
4731
4732 /*
4733 * Wait for the youngest allocations and frees to sync,
4734 * and then wait for the deferral of those frees to finish.
4735 */
4736 spa_vdev_config_exit(spa, NULL,
4737 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4738
4739 /*
4740 * Attempt to evacuate the vdev.
4741 */
4742 error = spa_vdev_remove_evacuate(spa, vd);
4743
4744 txg = spa_vdev_config_enter(spa);
4745
4746 /*
4747 * If we couldn't evacuate the vdev, unwind.
4748 */
4749 if (error) {
4750 metaslab_group_activate(mg);
4751 return (spa_vdev_exit(spa, NULL, txg, error));
4752 }
4753
4754 /*
4755 * Clean up the vdev namespace.
4756 */
4757 spa_vdev_remove_from_namespace(spa, vd);
4758
4759 } else if (vd != NULL) {
4760 /*
4761 * Normal vdevs cannot be removed (yet).
4762 */
4763 error = ENOTSUP;
4764 } else {
4765 /*
4766 * There is no vdev of any kind with the specified guid.
4767 */
4768 error = ENOENT;
4769 }
4770
4771 if (!locked)
4772 return (spa_vdev_exit(spa, NULL, txg, error));
4773
4774 return (error);
4775 }
4776
4777 /*
4778 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4779 * current spared, so we can detach it.
4780 */
4781 static vdev_t *
4782 spa_vdev_resilver_done_hunt(vdev_t *vd)
4783 {
4784 vdev_t *newvd, *oldvd;
4785 int c;
4786
4787 for (c = 0; c < vd->vdev_children; c++) {
4788 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4789 if (oldvd != NULL)
4790 return (oldvd);
4791 }
4792
4793 /*
4794 * Check for a completed replacement. We always consider the first
4795 * vdev in the list to be the oldest vdev, and the last one to be
4796 * the newest (see spa_vdev_attach() for how that works). In
4797 * the case where the newest vdev is faulted, we will not automatically
4798 * remove it after a resilver completes. This is OK as it will require
4799 * user intervention to determine which disk the admin wishes to keep.
4800 */
4801 if (vd->vdev_ops == &vdev_replacing_ops) {
4802 ASSERT(vd->vdev_children > 1);
4803
4804 newvd = vd->vdev_child[vd->vdev_children - 1];
4805 oldvd = vd->vdev_child[0];
4806
4807 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4808 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4809 !vdev_dtl_required(oldvd))
4810 return (oldvd);
4811 }
4812
4813 /*
4814 * Check for a completed resilver with the 'unspare' flag set.
4815 */
4816 if (vd->vdev_ops == &vdev_spare_ops) {
4817 vdev_t *first = vd->vdev_child[0];
4818 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4819
4820 if (last->vdev_unspare) {
4821 oldvd = first;
4822 newvd = last;
4823 } else if (first->vdev_unspare) {
4824 oldvd = last;
4825 newvd = first;
4826 } else {
4827 oldvd = NULL;
4828 }
4829
4830 if (oldvd != NULL &&
4831 vdev_dtl_empty(newvd, DTL_MISSING) &&
4832 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4833 !vdev_dtl_required(oldvd))
4834 return (oldvd);
4835
4836 /*
4837 * If there are more than two spares attached to a disk,
4838 * and those spares are not required, then we want to
4839 * attempt to free them up now so that they can be used
4840 * by other pools. Once we're back down to a single
4841 * disk+spare, we stop removing them.
4842 */
4843 if (vd->vdev_children > 2) {
4844 newvd = vd->vdev_child[1];
4845
4846 if (newvd->vdev_isspare && last->vdev_isspare &&
4847 vdev_dtl_empty(last, DTL_MISSING) &&
4848 vdev_dtl_empty(last, DTL_OUTAGE) &&
4849 !vdev_dtl_required(newvd))
4850 return (newvd);
4851 }
4852 }
4853
4854 return (NULL);
4855 }
4856
4857 static void
4858 spa_vdev_resilver_done(spa_t *spa)
4859 {
4860 vdev_t *vd, *pvd, *ppvd;
4861 uint64_t guid, sguid, pguid, ppguid;
4862
4863 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4864
4865 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4866 pvd = vd->vdev_parent;
4867 ppvd = pvd->vdev_parent;
4868 guid = vd->vdev_guid;
4869 pguid = pvd->vdev_guid;
4870 ppguid = ppvd->vdev_guid;
4871 sguid = 0;
4872 /*
4873 * If we have just finished replacing a hot spared device, then
4874 * we need to detach the parent's first child (the original hot
4875 * spare) as well.
4876 */
4877 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4878 ppvd->vdev_children == 2) {
4879 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4880 sguid = ppvd->vdev_child[1]->vdev_guid;
4881 }
4882 spa_config_exit(spa, SCL_ALL, FTAG);
4883 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4884 return;
4885 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4886 return;
4887 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4888 }
4889
4890 spa_config_exit(spa, SCL_ALL, FTAG);
4891 }
4892
4893 /*
4894 * Update the stored path or FRU for this vdev.
4895 */
4896 int
4897 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4898 boolean_t ispath)
4899 {
4900 vdev_t *vd;
4901 boolean_t sync = B_FALSE;
4902
4903 ASSERT(spa_writeable(spa));
4904
4905 spa_vdev_state_enter(spa, SCL_ALL);
4906
4907 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4908 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4909
4910 if (!vd->vdev_ops->vdev_op_leaf)
4911 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4912
4913 if (ispath) {
4914 if (strcmp(value, vd->vdev_path) != 0) {
4915 spa_strfree(vd->vdev_path);
4916 vd->vdev_path = spa_strdup(value);
4917 sync = B_TRUE;
4918 }
4919 } else {
4920 if (vd->vdev_fru == NULL) {
4921 vd->vdev_fru = spa_strdup(value);
4922 sync = B_TRUE;
4923 } else if (strcmp(value, vd->vdev_fru) != 0) {
4924 spa_strfree(vd->vdev_fru);
4925 vd->vdev_fru = spa_strdup(value);
4926 sync = B_TRUE;
4927 }
4928 }
4929
4930 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4931 }
4932
4933 int
4934 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4935 {
4936 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4937 }
4938
4939 int
4940 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4941 {
4942 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4943 }
4944
4945 /*
4946 * ==========================================================================
4947 * SPA Scanning
4948 * ==========================================================================
4949 */
4950
4951 int
4952 spa_scan_stop(spa_t *spa)
4953 {
4954 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4955 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4956 return (EBUSY);
4957 return (dsl_scan_cancel(spa->spa_dsl_pool));
4958 }
4959
4960 int
4961 spa_scan(spa_t *spa, pool_scan_func_t func)
4962 {
4963 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4964
4965 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4966 return (ENOTSUP);
4967
4968 /*
4969 * If a resilver was requested, but there is no DTL on a
4970 * writeable leaf device, we have nothing to do.
4971 */
4972 if (func == POOL_SCAN_RESILVER &&
4973 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4974 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4975 return (0);
4976 }
4977
4978 return (dsl_scan(spa->spa_dsl_pool, func));
4979 }
4980
4981 /*
4982 * ==========================================================================
4983 * SPA async task processing
4984 * ==========================================================================
4985 */
4986
4987 static void
4988 spa_async_remove(spa_t *spa, vdev_t *vd)
4989 {
4990 int c;
4991
4992 if (vd->vdev_remove_wanted) {
4993 vd->vdev_remove_wanted = B_FALSE;
4994 vd->vdev_delayed_close = B_FALSE;
4995 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4996
4997 /*
4998 * We want to clear the stats, but we don't want to do a full
4999 * vdev_clear() as that will cause us to throw away
5000 * degraded/faulted state as well as attempt to reopen the
5001 * device, all of which is a waste.
5002 */
5003 vd->vdev_stat.vs_read_errors = 0;
5004 vd->vdev_stat.vs_write_errors = 0;
5005 vd->vdev_stat.vs_checksum_errors = 0;
5006
5007 vdev_state_dirty(vd->vdev_top);
5008 }
5009
5010 for (c = 0; c < vd->vdev_children; c++)
5011 spa_async_remove(spa, vd->vdev_child[c]);
5012 }
5013
5014 static void
5015 spa_async_probe(spa_t *spa, vdev_t *vd)
5016 {
5017 int c;
5018
5019 if (vd->vdev_probe_wanted) {
5020 vd->vdev_probe_wanted = B_FALSE;
5021 vdev_reopen(vd); /* vdev_open() does the actual probe */
5022 }
5023
5024 for (c = 0; c < vd->vdev_children; c++)
5025 spa_async_probe(spa, vd->vdev_child[c]);
5026 }
5027
5028 static void
5029 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5030 {
5031 sysevent_id_t eid;
5032 nvlist_t *attr;
5033 char *physpath;
5034 int c;
5035
5036 if (!spa->spa_autoexpand)
5037 return;
5038
5039 for (c = 0; c < vd->vdev_children; c++) {
5040 vdev_t *cvd = vd->vdev_child[c];
5041 spa_async_autoexpand(spa, cvd);
5042 }
5043
5044 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5045 return;
5046
5047 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5048 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5049
5050 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5051 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5052
5053 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5054 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5055
5056 nvlist_free(attr);
5057 kmem_free(physpath, MAXPATHLEN);
5058 }
5059
5060 static void
5061 spa_async_thread(spa_t *spa)
5062 {
5063 int tasks, i;
5064
5065 ASSERT(spa->spa_sync_on);
5066
5067 mutex_enter(&spa->spa_async_lock);
5068 tasks = spa->spa_async_tasks;
5069 spa->spa_async_tasks = 0;
5070 mutex_exit(&spa->spa_async_lock);
5071
5072 /*
5073 * See if the config needs to be updated.
5074 */
5075 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5076 uint64_t old_space, new_space;
5077
5078 mutex_enter(&spa_namespace_lock);
5079 old_space = metaslab_class_get_space(spa_normal_class(spa));
5080 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5081 new_space = metaslab_class_get_space(spa_normal_class(spa));
5082 mutex_exit(&spa_namespace_lock);
5083
5084 /*
5085 * If the pool grew as a result of the config update,
5086 * then log an internal history event.
5087 */
5088 if (new_space != old_space) {
5089 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5090 spa, NULL,
5091 "pool '%s' size: %llu(+%llu)",
5092 spa_name(spa), new_space, new_space - old_space);
5093 }
5094 }
5095
5096 /*
5097 * See if any devices need to be marked REMOVED.
5098 */
5099 if (tasks & SPA_ASYNC_REMOVE) {
5100 spa_vdev_state_enter(spa, SCL_NONE);
5101 spa_async_remove(spa, spa->spa_root_vdev);
5102 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5103 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5104 for (i = 0; i < spa->spa_spares.sav_count; i++)
5105 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5106 (void) spa_vdev_state_exit(spa, NULL, 0);
5107 }
5108
5109 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5110 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5111 spa_async_autoexpand(spa, spa->spa_root_vdev);
5112 spa_config_exit(spa, SCL_CONFIG, FTAG);
5113 }
5114
5115 /*
5116 * See if any devices need to be probed.
5117 */
5118 if (tasks & SPA_ASYNC_PROBE) {
5119 spa_vdev_state_enter(spa, SCL_NONE);
5120 spa_async_probe(spa, spa->spa_root_vdev);
5121 (void) spa_vdev_state_exit(spa, NULL, 0);
5122 }
5123
5124 /*
5125 * If any devices are done replacing, detach them.
5126 */
5127 if (tasks & SPA_ASYNC_RESILVER_DONE)
5128 spa_vdev_resilver_done(spa);
5129
5130 /*
5131 * Kick off a resilver.
5132 */
5133 if (tasks & SPA_ASYNC_RESILVER)
5134 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5135
5136 /*
5137 * Let the world know that we're done.
5138 */
5139 mutex_enter(&spa->spa_async_lock);
5140 spa->spa_async_thread = NULL;
5141 cv_broadcast(&spa->spa_async_cv);
5142 mutex_exit(&spa->spa_async_lock);
5143 thread_exit();
5144 }
5145
5146 void
5147 spa_async_suspend(spa_t *spa)
5148 {
5149 mutex_enter(&spa->spa_async_lock);
5150 spa->spa_async_suspended++;
5151 while (spa->spa_async_thread != NULL)
5152 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5153 mutex_exit(&spa->spa_async_lock);
5154 }
5155
5156 void
5157 spa_async_resume(spa_t *spa)
5158 {
5159 mutex_enter(&spa->spa_async_lock);
5160 ASSERT(spa->spa_async_suspended != 0);
5161 spa->spa_async_suspended--;
5162 mutex_exit(&spa->spa_async_lock);
5163 }
5164
5165 static void
5166 spa_async_dispatch(spa_t *spa)
5167 {
5168 mutex_enter(&spa->spa_async_lock);
5169 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5170 spa->spa_async_thread == NULL &&
5171 rootdir != NULL && !vn_is_readonly(rootdir))
5172 spa->spa_async_thread = thread_create(NULL, 0,
5173 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5174 mutex_exit(&spa->spa_async_lock);
5175 }
5176
5177 void
5178 spa_async_request(spa_t *spa, int task)
5179 {
5180 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5181 mutex_enter(&spa->spa_async_lock);
5182 spa->spa_async_tasks |= task;
5183 mutex_exit(&spa->spa_async_lock);
5184 }
5185
5186 /*
5187 * ==========================================================================
5188 * SPA syncing routines
5189 * ==========================================================================
5190 */
5191
5192 static int
5193 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5194 {
5195 bpobj_t *bpo = arg;
5196 bpobj_enqueue(bpo, bp, tx);
5197 return (0);
5198 }
5199
5200 static int
5201 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5202 {
5203 zio_t *zio = arg;
5204
5205 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5206 zio->io_flags));
5207 return (0);
5208 }
5209
5210 static void
5211 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5212 {
5213 char *packed = NULL;
5214 size_t bufsize;
5215 size_t nvsize = 0;
5216 dmu_buf_t *db;
5217
5218 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5219
5220 /*
5221 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5222 * information. This avoids the dbuf_will_dirty() path and
5223 * saves us a pre-read to get data we don't actually care about.
5224 */
5225 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5226 packed = kmem_alloc(bufsize, KM_SLEEP);
5227
5228 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5229 KM_SLEEP) == 0);
5230 bzero(packed + nvsize, bufsize - nvsize);
5231
5232 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5233
5234 kmem_free(packed, bufsize);
5235
5236 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5237 dmu_buf_will_dirty(db, tx);
5238 *(uint64_t *)db->db_data = nvsize;
5239 dmu_buf_rele(db, FTAG);
5240 }
5241
5242 static void
5243 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5244 const char *config, const char *entry)
5245 {
5246 nvlist_t *nvroot;
5247 nvlist_t **list;
5248 int i;
5249
5250 if (!sav->sav_sync)
5251 return;
5252
5253 /*
5254 * Update the MOS nvlist describing the list of available devices.
5255 * spa_validate_aux() will have already made sure this nvlist is
5256 * valid and the vdevs are labeled appropriately.
5257 */
5258 if (sav->sav_object == 0) {
5259 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5260 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5261 sizeof (uint64_t), tx);
5262 VERIFY(zap_update(spa->spa_meta_objset,
5263 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5264 &sav->sav_object, tx) == 0);
5265 }
5266
5267 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5268 if (sav->sav_count == 0) {
5269 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5270 } else {
5271 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5272 for (i = 0; i < sav->sav_count; i++)
5273 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5274 B_FALSE, VDEV_CONFIG_L2CACHE);
5275 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5276 sav->sav_count) == 0);
5277 for (i = 0; i < sav->sav_count; i++)
5278 nvlist_free(list[i]);
5279 kmem_free(list, sav->sav_count * sizeof (void *));
5280 }
5281
5282 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5283 nvlist_free(nvroot);
5284
5285 sav->sav_sync = B_FALSE;
5286 }
5287
5288 static void
5289 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5290 {
5291 nvlist_t *config;
5292
5293 if (list_is_empty(&spa->spa_config_dirty_list))
5294 return;
5295
5296 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5297
5298 config = spa_config_generate(spa, spa->spa_root_vdev,
5299 dmu_tx_get_txg(tx), B_FALSE);
5300
5301 spa_config_exit(spa, SCL_STATE, FTAG);
5302
5303 if (spa->spa_config_syncing)
5304 nvlist_free(spa->spa_config_syncing);
5305 spa->spa_config_syncing = config;
5306
5307 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5308 }
5309
5310 /*
5311 * Set zpool properties.
5312 */
5313 static void
5314 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5315 {
5316 spa_t *spa = arg1;
5317 objset_t *mos = spa->spa_meta_objset;
5318 nvlist_t *nvp = arg2;
5319 nvpair_t *elem;
5320 uint64_t intval;
5321 char *strval;
5322 zpool_prop_t prop;
5323 const char *propname;
5324 zprop_type_t proptype;
5325
5326 mutex_enter(&spa->spa_props_lock);
5327
5328 elem = NULL;
5329 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5330 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5331 case ZPOOL_PROP_VERSION:
5332 /*
5333 * Only set version for non-zpool-creation cases
5334 * (set/import). spa_create() needs special care
5335 * for version setting.
5336 */
5337 if (tx->tx_txg != TXG_INITIAL) {
5338 VERIFY(nvpair_value_uint64(elem,
5339 &intval) == 0);
5340 ASSERT(intval <= SPA_VERSION);
5341 ASSERT(intval >= spa_version(spa));
5342 spa->spa_uberblock.ub_version = intval;
5343 vdev_config_dirty(spa->spa_root_vdev);
5344 }
5345 break;
5346
5347 case ZPOOL_PROP_ALTROOT:
5348 /*
5349 * 'altroot' is a non-persistent property. It should
5350 * have been set temporarily at creation or import time.
5351 */
5352 ASSERT(spa->spa_root != NULL);
5353 break;
5354
5355 case ZPOOL_PROP_READONLY:
5356 case ZPOOL_PROP_CACHEFILE:
5357 /*
5358 * 'readonly' and 'cachefile' are also non-persisitent
5359 * properties.
5360 */
5361 break;
5362 default:
5363 /*
5364 * Set pool property values in the poolprops mos object.
5365 */
5366 if (spa->spa_pool_props_object == 0) {
5367 VERIFY((spa->spa_pool_props_object =
5368 zap_create(mos, DMU_OT_POOL_PROPS,
5369 DMU_OT_NONE, 0, tx)) > 0);
5370
5371 VERIFY(zap_update(mos,
5372 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5373 8, 1, &spa->spa_pool_props_object, tx)
5374 == 0);
5375 }
5376
5377 /* normalize the property name */
5378 propname = zpool_prop_to_name(prop);
5379 proptype = zpool_prop_get_type(prop);
5380
5381 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5382 ASSERT(proptype == PROP_TYPE_STRING);
5383 VERIFY(nvpair_value_string(elem, &strval) == 0);
5384 VERIFY(zap_update(mos,
5385 spa->spa_pool_props_object, propname,
5386 1, strlen(strval) + 1, strval, tx) == 0);
5387
5388 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5389 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5390
5391 if (proptype == PROP_TYPE_INDEX) {
5392 const char *unused;
5393 VERIFY(zpool_prop_index_to_string(
5394 prop, intval, &unused) == 0);
5395 }
5396 VERIFY(zap_update(mos,
5397 spa->spa_pool_props_object, propname,
5398 8, 1, &intval, tx) == 0);
5399 } else {
5400 ASSERT(0); /* not allowed */
5401 }
5402
5403 switch (prop) {
5404 case ZPOOL_PROP_DELEGATION:
5405 spa->spa_delegation = intval;
5406 break;
5407 case ZPOOL_PROP_BOOTFS:
5408 spa->spa_bootfs = intval;
5409 break;
5410 case ZPOOL_PROP_FAILUREMODE:
5411 spa->spa_failmode = intval;
5412 break;
5413 case ZPOOL_PROP_AUTOEXPAND:
5414 spa->spa_autoexpand = intval;
5415 if (tx->tx_txg != TXG_INITIAL)
5416 spa_async_request(spa,
5417 SPA_ASYNC_AUTOEXPAND);
5418 break;
5419 case ZPOOL_PROP_DEDUPDITTO:
5420 spa->spa_dedup_ditto = intval;
5421 break;
5422 default:
5423 break;
5424 }
5425 }
5426
5427 /* log internal history if this is not a zpool create */
5428 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5429 tx->tx_txg != TXG_INITIAL) {
5430 spa_history_log_internal(LOG_POOL_PROPSET,
5431 spa, tx, "%s %lld %s",
5432 nvpair_name(elem), intval, spa_name(spa));
5433 }
5434 }
5435
5436 mutex_exit(&spa->spa_props_lock);
5437 }
5438
5439 /*
5440 * Perform one-time upgrade on-disk changes. spa_version() does not
5441 * reflect the new version this txg, so there must be no changes this
5442 * txg to anything that the upgrade code depends on after it executes.
5443 * Therefore this must be called after dsl_pool_sync() does the sync
5444 * tasks.
5445 */
5446 static void
5447 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5448 {
5449 dsl_pool_t *dp = spa->spa_dsl_pool;
5450
5451 ASSERT(spa->spa_sync_pass == 1);
5452
5453 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5454 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5455 dsl_pool_create_origin(dp, tx);
5456
5457 /* Keeping the origin open increases spa_minref */
5458 spa->spa_minref += 3;
5459 }
5460
5461 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5462 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5463 dsl_pool_upgrade_clones(dp, tx);
5464 }
5465
5466 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5467 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5468 dsl_pool_upgrade_dir_clones(dp, tx);
5469
5470 /* Keeping the freedir open increases spa_minref */
5471 spa->spa_minref += 3;
5472 }
5473 }
5474
5475 /*
5476 * Sync the specified transaction group. New blocks may be dirtied as
5477 * part of the process, so we iterate until it converges.
5478 */
5479 void
5480 spa_sync(spa_t *spa, uint64_t txg)
5481 {
5482 dsl_pool_t *dp = spa->spa_dsl_pool;
5483 objset_t *mos = spa->spa_meta_objset;
5484 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5485 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5486 vdev_t *rvd = spa->spa_root_vdev;
5487 vdev_t *vd;
5488 dmu_tx_t *tx;
5489 int error;
5490 int c;
5491
5492 VERIFY(spa_writeable(spa));
5493
5494 /*
5495 * Lock out configuration changes.
5496 */
5497 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5498
5499 spa->spa_syncing_txg = txg;
5500 spa->spa_sync_pass = 0;
5501
5502 /*
5503 * If there are any pending vdev state changes, convert them
5504 * into config changes that go out with this transaction group.
5505 */
5506 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5507 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5508 /*
5509 * We need the write lock here because, for aux vdevs,
5510 * calling vdev_config_dirty() modifies sav_config.
5511 * This is ugly and will become unnecessary when we
5512 * eliminate the aux vdev wart by integrating all vdevs
5513 * into the root vdev tree.
5514 */
5515 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5516 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5517 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5518 vdev_state_clean(vd);
5519 vdev_config_dirty(vd);
5520 }
5521 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5522 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5523 }
5524 spa_config_exit(spa, SCL_STATE, FTAG);
5525
5526 tx = dmu_tx_create_assigned(dp, txg);
5527
5528 /*
5529 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5530 * set spa_deflate if we have no raid-z vdevs.
5531 */
5532 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5533 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5534 int i;
5535
5536 for (i = 0; i < rvd->vdev_children; i++) {
5537 vd = rvd->vdev_child[i];
5538 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5539 break;
5540 }
5541 if (i == rvd->vdev_children) {
5542 spa->spa_deflate = TRUE;
5543 VERIFY(0 == zap_add(spa->spa_meta_objset,
5544 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5545 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5546 }
5547 }
5548
5549 /*
5550 * If anything has changed in this txg, or if someone is waiting
5551 * for this txg to sync (eg, spa_vdev_remove()), push the
5552 * deferred frees from the previous txg. If not, leave them
5553 * alone so that we don't generate work on an otherwise idle
5554 * system.
5555 */
5556 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5557 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5558 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5559 ((dsl_scan_active(dp->dp_scan) ||
5560 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5561 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5562 VERIFY3U(bpobj_iterate(defer_bpo,
5563 spa_free_sync_cb, zio, tx), ==, 0);
5564 VERIFY3U(zio_wait(zio), ==, 0);
5565 }
5566
5567 /*
5568 * Iterate to convergence.
5569 */
5570 do {
5571 int pass = ++spa->spa_sync_pass;
5572
5573 spa_sync_config_object(spa, tx);
5574 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5575 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5576 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5577 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5578 spa_errlog_sync(spa, txg);
5579 dsl_pool_sync(dp, txg);
5580
5581 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5582 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5583 bplist_iterate(free_bpl, spa_free_sync_cb,
5584 zio, tx);
5585 VERIFY(zio_wait(zio) == 0);
5586 } else {
5587 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5588 defer_bpo, tx);
5589 }
5590
5591 ddt_sync(spa, txg);
5592 dsl_scan_sync(dp, tx);
5593
5594 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5595 vdev_sync(vd, txg);
5596
5597 if (pass == 1)
5598 spa_sync_upgrades(spa, tx);
5599
5600 } while (dmu_objset_is_dirty(mos, txg));
5601
5602 /*
5603 * Rewrite the vdev configuration (which includes the uberblock)
5604 * to commit the transaction group.
5605 *
5606 * If there are no dirty vdevs, we sync the uberblock to a few
5607 * random top-level vdevs that are known to be visible in the
5608 * config cache (see spa_vdev_add() for a complete description).
5609 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5610 */
5611 for (;;) {
5612 /*
5613 * We hold SCL_STATE to prevent vdev open/close/etc.
5614 * while we're attempting to write the vdev labels.
5615 */
5616 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5617
5618 if (list_is_empty(&spa->spa_config_dirty_list)) {
5619 vdev_t *svd[SPA_DVAS_PER_BP];
5620 int svdcount = 0;
5621 int children = rvd->vdev_children;
5622 int c0 = spa_get_random(children);
5623
5624 for (c = 0; c < children; c++) {
5625 vd = rvd->vdev_child[(c0 + c) % children];
5626 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5627 continue;
5628 svd[svdcount++] = vd;
5629 if (svdcount == SPA_DVAS_PER_BP)
5630 break;
5631 }
5632 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5633 if (error != 0)
5634 error = vdev_config_sync(svd, svdcount, txg,
5635 B_TRUE);
5636 } else {
5637 error = vdev_config_sync(rvd->vdev_child,
5638 rvd->vdev_children, txg, B_FALSE);
5639 if (error != 0)
5640 error = vdev_config_sync(rvd->vdev_child,
5641 rvd->vdev_children, txg, B_TRUE);
5642 }
5643
5644 spa_config_exit(spa, SCL_STATE, FTAG);
5645
5646 if (error == 0)
5647 break;
5648 zio_suspend(spa, NULL);
5649 zio_resume_wait(spa);
5650 }
5651 dmu_tx_commit(tx);
5652
5653 /*
5654 * Clear the dirty config list.
5655 */
5656 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5657 vdev_config_clean(vd);
5658
5659 /*
5660 * Now that the new config has synced transactionally,
5661 * let it become visible to the config cache.
5662 */
5663 if (spa->spa_config_syncing != NULL) {
5664 spa_config_set(spa, spa->spa_config_syncing);
5665 spa->spa_config_txg = txg;
5666 spa->spa_config_syncing = NULL;
5667 }
5668
5669 spa->spa_ubsync = spa->spa_uberblock;
5670
5671 dsl_pool_sync_done(dp, txg);
5672
5673 /*
5674 * Update usable space statistics.
5675 */
5676 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5677 vdev_sync_done(vd, txg);
5678
5679 spa_update_dspace(spa);
5680
5681 /*
5682 * It had better be the case that we didn't dirty anything
5683 * since vdev_config_sync().
5684 */
5685 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5686 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5687 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5688
5689 spa->spa_sync_pass = 0;
5690
5691 spa_config_exit(spa, SCL_CONFIG, FTAG);
5692
5693 spa_handle_ignored_writes(spa);
5694
5695 /*
5696 * If any async tasks have been requested, kick them off.
5697 */
5698 spa_async_dispatch(spa);
5699 }
5700
5701 /*
5702 * Sync all pools. We don't want to hold the namespace lock across these
5703 * operations, so we take a reference on the spa_t and drop the lock during the
5704 * sync.
5705 */
5706 void
5707 spa_sync_allpools(void)
5708 {
5709 spa_t *spa = NULL;
5710 mutex_enter(&spa_namespace_lock);
5711 while ((spa = spa_next(spa)) != NULL) {
5712 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5713 !spa_writeable(spa) || spa_suspended(spa))
5714 continue;
5715 spa_open_ref(spa, FTAG);
5716 mutex_exit(&spa_namespace_lock);
5717 txg_wait_synced(spa_get_dsl(spa), 0);
5718 mutex_enter(&spa_namespace_lock);
5719 spa_close(spa, FTAG);
5720 }
5721 mutex_exit(&spa_namespace_lock);
5722 }
5723
5724 /*
5725 * ==========================================================================
5726 * Miscellaneous routines
5727 * ==========================================================================
5728 */
5729
5730 /*
5731 * Remove all pools in the system.
5732 */
5733 void
5734 spa_evict_all(void)
5735 {
5736 spa_t *spa;
5737
5738 /*
5739 * Remove all cached state. All pools should be closed now,
5740 * so every spa in the AVL tree should be unreferenced.
5741 */
5742 mutex_enter(&spa_namespace_lock);
5743 while ((spa = spa_next(NULL)) != NULL) {
5744 /*
5745 * Stop async tasks. The async thread may need to detach
5746 * a device that's been replaced, which requires grabbing
5747 * spa_namespace_lock, so we must drop it here.
5748 */
5749 spa_open_ref(spa, FTAG);
5750 mutex_exit(&spa_namespace_lock);
5751 spa_async_suspend(spa);
5752 mutex_enter(&spa_namespace_lock);
5753 spa_close(spa, FTAG);
5754
5755 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5756 spa_unload(spa);
5757 spa_deactivate(spa);
5758 }
5759 spa_remove(spa);
5760 }
5761 mutex_exit(&spa_namespace_lock);
5762 }
5763
5764 vdev_t *
5765 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5766 {
5767 vdev_t *vd;
5768 int i;
5769
5770 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5771 return (vd);
5772
5773 if (aux) {
5774 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5775 vd = spa->spa_l2cache.sav_vdevs[i];
5776 if (vd->vdev_guid == guid)
5777 return (vd);
5778 }
5779
5780 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5781 vd = spa->spa_spares.sav_vdevs[i];
5782 if (vd->vdev_guid == guid)
5783 return (vd);
5784 }
5785 }
5786
5787 return (NULL);
5788 }
5789
5790 void
5791 spa_upgrade(spa_t *spa, uint64_t version)
5792 {
5793 ASSERT(spa_writeable(spa));
5794
5795 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5796
5797 /*
5798 * This should only be called for a non-faulted pool, and since a
5799 * future version would result in an unopenable pool, this shouldn't be
5800 * possible.
5801 */
5802 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5803 ASSERT(version >= spa->spa_uberblock.ub_version);
5804
5805 spa->spa_uberblock.ub_version = version;
5806 vdev_config_dirty(spa->spa_root_vdev);
5807
5808 spa_config_exit(spa, SCL_ALL, FTAG);
5809
5810 txg_wait_synced(spa_get_dsl(spa), 0);
5811 }
5812
5813 boolean_t
5814 spa_has_spare(spa_t *spa, uint64_t guid)
5815 {
5816 int i;
5817 uint64_t spareguid;
5818 spa_aux_vdev_t *sav = &spa->spa_spares;
5819
5820 for (i = 0; i < sav->sav_count; i++)
5821 if (sav->sav_vdevs[i]->vdev_guid == guid)
5822 return (B_TRUE);
5823
5824 for (i = 0; i < sav->sav_npending; i++) {
5825 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5826 &spareguid) == 0 && spareguid == guid)
5827 return (B_TRUE);
5828 }
5829
5830 return (B_FALSE);
5831 }
5832
5833 /*
5834 * Check if a pool has an active shared spare device.
5835 * Note: reference count of an active spare is 2, as a spare and as a replace
5836 */
5837 static boolean_t
5838 spa_has_active_shared_spare(spa_t *spa)
5839 {
5840 int i, refcnt;
5841 uint64_t pool;
5842 spa_aux_vdev_t *sav = &spa->spa_spares;
5843
5844 for (i = 0; i < sav->sav_count; i++) {
5845 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5846 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5847 refcnt > 2)
5848 return (B_TRUE);
5849 }
5850
5851 return (B_FALSE);
5852 }
5853
5854 /*
5855 * Post a sysevent corresponding to the given event. The 'name' must be one of
5856 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5857 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5858 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5859 * or zdb as real changes.
5860 */
5861 void
5862 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5863 {
5864 #ifdef _KERNEL
5865 sysevent_t *ev;
5866 sysevent_attr_list_t *attr = NULL;
5867 sysevent_value_t value;
5868 sysevent_id_t eid;
5869
5870 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5871 SE_SLEEP);
5872
5873 value.value_type = SE_DATA_TYPE_STRING;
5874 value.value.sv_string = spa_name(spa);
5875 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5876 goto done;
5877
5878 value.value_type = SE_DATA_TYPE_UINT64;
5879 value.value.sv_uint64 = spa_guid(spa);
5880 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5881 goto done;
5882
5883 if (vd) {
5884 value.value_type = SE_DATA_TYPE_UINT64;
5885 value.value.sv_uint64 = vd->vdev_guid;
5886 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5887 SE_SLEEP) != 0)
5888 goto done;
5889
5890 if (vd->vdev_path) {
5891 value.value_type = SE_DATA_TYPE_STRING;
5892 value.value.sv_string = vd->vdev_path;
5893 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5894 &value, SE_SLEEP) != 0)
5895 goto done;
5896 }
5897 }
5898
5899 if (sysevent_attach_attributes(ev, attr) != 0)
5900 goto done;
5901 attr = NULL;
5902
5903 (void) log_sysevent(ev, SE_SLEEP, &eid);
5904
5905 done:
5906 if (attr)
5907 sysevent_free_attr(attr);
5908 sysevent_free(ev);
5909 #endif
5910 }
ZFS On Linux mirror for PVE