1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28 /*
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 * pool.
32 */
33
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
37 #include <sys/zio.h>
38 #include <sys/zio_checksum.h>
39 #include <sys/dmu.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/zap.h>
42 #include <sys/zil.h>
43 #include <sys/ddt.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/metaslab.h>
46 #include <sys/metaslab_impl.h>
47 #include <sys/uberblock_impl.h>
48 #include <sys/txg.h>
49 #include <sys/avl.h>
50 #include <sys/dmu_traverse.h>
51 #include <sys/dmu_objset.h>
52 #include <sys/unique.h>
53 #include <sys/dsl_pool.h>
54 #include <sys/dsl_dataset.h>
55 #include <sys/dsl_dir.h>
56 #include <sys/dsl_prop.h>
57 #include <sys/dsl_synctask.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/arc.h>
60 #include <sys/callb.h>
61 #include <sys/systeminfo.h>
62 #include <sys/spa_boot.h>
63 #include <sys/zfs_ioctl.h>
64 #include <sys/dsl_scan.h>
65 #include <sys/zfeature.h>
66
67 #ifdef _KERNEL
68 #include <sys/bootprops.h>
69 #include <sys/callb.h>
70 #include <sys/cpupart.h>
71 #include <sys/pool.h>
72 #include <sys/sysdc.h>
73 #include <sys/zone.h>
74 #endif /* _KERNEL */
75
76 #include "zfs_prop.h"
77 #include "zfs_comutil.h"
78
79 typedef enum zti_modes {
80 zti_mode_fixed, /* value is # of threads (min 1) */
81 zti_mode_online_percent, /* value is % of online CPUs */
82 zti_mode_batch, /* cpu-intensive; value is ignored */
83 zti_mode_null, /* don't create a taskq */
84 zti_nmodes
85 } zti_modes_t;
86
87 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
88 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
89 #define ZTI_BATCH { zti_mode_batch, 0 }
90 #define ZTI_NULL { zti_mode_null, 0 }
91
92 #define ZTI_ONE ZTI_FIX(1)
93
94 typedef struct zio_taskq_info {
95 enum zti_modes zti_mode;
96 uint_t zti_value;
97 } zio_taskq_info_t;
98
99 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
100 "issue", "issue_high", "intr", "intr_high"
101 };
102
103 /*
104 * Define the taskq threads for the following I/O types:
105 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
106 */
107 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
108 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
109 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
111 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
112 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 };
116
117 static dsl_syncfunc_t spa_sync_version;
118 static dsl_syncfunc_t spa_sync_props;
119 static boolean_t spa_has_active_shared_spare(spa_t *spa);
120 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
121 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
122 char **ereport);
123 static void spa_vdev_resilver_done(spa_t *spa);
124
125 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
126 id_t zio_taskq_psrset_bind = PS_NONE;
127 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
128 uint_t zio_taskq_basedc = 80; /* base duty cycle */
129
130 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
131
132 /*
133 * This (illegal) pool name is used when temporarily importing a spa_t in order
134 * to get the vdev stats associated with the imported devices.
135 */
136 #define TRYIMPORT_NAME "$import"
137
138 /*
139 * ==========================================================================
140 * SPA properties routines
141 * ==========================================================================
142 */
143
144 /*
145 * Add a (source=src, propname=propval) list to an nvlist.
146 */
147 static void
148 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
149 uint64_t intval, zprop_source_t src)
150 {
151 const char *propname = zpool_prop_to_name(prop);
152 nvlist_t *propval;
153
154 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
155 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
156
157 if (strval != NULL)
158 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
159 else
160 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
161
162 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
163 nvlist_free(propval);
164 }
165
166 /*
167 * Get property values from the spa configuration.
168 */
169 static void
170 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
171 {
172 vdev_t *rvd = spa->spa_root_vdev;
173 dsl_pool_t *pool = spa->spa_dsl_pool;
174 uint64_t size;
175 uint64_t alloc;
176 uint64_t space;
177 uint64_t cap, version;
178 zprop_source_t src = ZPROP_SRC_NONE;
179 spa_config_dirent_t *dp;
180
181 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
182
183 if (rvd != NULL) {
184 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
185 size = metaslab_class_get_space(spa_normal_class(spa));
186 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
187 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
188 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
189 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
190 size - alloc, src);
191
192 space = 0;
193 for (int c = 0; c < rvd->vdev_children; c++) {
194 vdev_t *tvd = rvd->vdev_child[c];
195 space += tvd->vdev_max_asize - tvd->vdev_asize;
196 }
197 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
198 src);
199
200 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
201 (spa_mode(spa) == FREAD), src);
202
203 cap = (size == 0) ? 0 : (alloc * 100 / size);
204 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
205
206 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
207 ddt_get_pool_dedup_ratio(spa), src);
208
209 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
210 rvd->vdev_state, src);
211
212 version = spa_version(spa);
213 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
214 src = ZPROP_SRC_DEFAULT;
215 else
216 src = ZPROP_SRC_LOCAL;
217 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
218 }
219
220 if (pool != NULL) {
221 dsl_dir_t *freedir = pool->dp_free_dir;
222
223 /*
224 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
225 * when opening pools before this version freedir will be NULL.
226 */
227 if (freedir != NULL) {
228 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
229 freedir->dd_phys->dd_used_bytes, src);
230 } else {
231 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
232 NULL, 0, src);
233 }
234 }
235
236 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
237
238 if (spa->spa_comment != NULL) {
239 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
240 0, ZPROP_SRC_LOCAL);
241 }
242
243 if (spa->spa_root != NULL)
244 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
245 0, ZPROP_SRC_LOCAL);
246
247 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
248 if (dp->scd_path == NULL) {
249 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
250 "none", 0, ZPROP_SRC_LOCAL);
251 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
252 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
253 dp->scd_path, 0, ZPROP_SRC_LOCAL);
254 }
255 }
256 }
257
258 /*
259 * Get zpool property values.
260 */
261 int
262 spa_prop_get(spa_t *spa, nvlist_t **nvp)
263 {
264 objset_t *mos = spa->spa_meta_objset;
265 zap_cursor_t zc;
266 zap_attribute_t za;
267 int err;
268
269 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
270
271 mutex_enter(&spa->spa_props_lock);
272
273 /*
274 * Get properties from the spa config.
275 */
276 spa_prop_get_config(spa, nvp);
277
278 /* If no pool property object, no more prop to get. */
279 if (mos == NULL || spa->spa_pool_props_object == 0) {
280 mutex_exit(&spa->spa_props_lock);
281 return (0);
282 }
283
284 /*
285 * Get properties from the MOS pool property object.
286 */
287 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
288 (err = zap_cursor_retrieve(&zc, &za)) == 0;
289 zap_cursor_advance(&zc)) {
290 uint64_t intval = 0;
291 char *strval = NULL;
292 zprop_source_t src = ZPROP_SRC_DEFAULT;
293 zpool_prop_t prop;
294
295 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
296 continue;
297
298 switch (za.za_integer_length) {
299 case 8:
300 /* integer property */
301 if (za.za_first_integer !=
302 zpool_prop_default_numeric(prop))
303 src = ZPROP_SRC_LOCAL;
304
305 if (prop == ZPOOL_PROP_BOOTFS) {
306 dsl_pool_t *dp;
307 dsl_dataset_t *ds = NULL;
308
309 dp = spa_get_dsl(spa);
310 rw_enter(&dp->dp_config_rwlock, RW_READER);
311 if (err = dsl_dataset_hold_obj(dp,
312 za.za_first_integer, FTAG, &ds)) {
313 rw_exit(&dp->dp_config_rwlock);
314 break;
315 }
316
317 strval = kmem_alloc(
318 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
319 KM_SLEEP);
320 dsl_dataset_name(ds, strval);
321 dsl_dataset_rele(ds, FTAG);
322 rw_exit(&dp->dp_config_rwlock);
323 } else {
324 strval = NULL;
325 intval = za.za_first_integer;
326 }
327
328 spa_prop_add_list(*nvp, prop, strval, intval, src);
329
330 if (strval != NULL)
331 kmem_free(strval,
332 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
333
334 break;
335
336 case 1:
337 /* string property */
338 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
339 err = zap_lookup(mos, spa->spa_pool_props_object,
340 za.za_name, 1, za.za_num_integers, strval);
341 if (err) {
342 kmem_free(strval, za.za_num_integers);
343 break;
344 }
345 spa_prop_add_list(*nvp, prop, strval, 0, src);
346 kmem_free(strval, za.za_num_integers);
347 break;
348
349 default:
350 break;
351 }
352 }
353 zap_cursor_fini(&zc);
354 mutex_exit(&spa->spa_props_lock);
355 out:
356 if (err && err != ENOENT) {
357 nvlist_free(*nvp);
358 *nvp = NULL;
359 return (err);
360 }
361
362 return (0);
363 }
364
365 /*
366 * Validate the given pool properties nvlist and modify the list
367 * for the property values to be set.
368 */
369 static int
370 spa_prop_validate(spa_t *spa, nvlist_t *props)
371 {
372 nvpair_t *elem;
373 int error = 0, reset_bootfs = 0;
374 uint64_t objnum;
375 boolean_t has_feature = B_FALSE;
376
377 elem = NULL;
378 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
379 uint64_t intval;
380 char *strval, *slash, *check, *fname;
381 const char *propname = nvpair_name(elem);
382 zpool_prop_t prop = zpool_name_to_prop(propname);
383
384 switch (prop) {
385 case ZPROP_INVAL:
386 if (!zpool_prop_feature(propname)) {
387 error = EINVAL;
388 break;
389 }
390
391 /*
392 * Sanitize the input.
393 */
394 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
395 error = EINVAL;
396 break;
397 }
398
399 if (nvpair_value_uint64(elem, &intval) != 0) {
400 error = EINVAL;
401 break;
402 }
403
404 if (intval != 0) {
405 error = EINVAL;
406 break;
407 }
408
409 fname = strchr(propname, '@') + 1;
410 if (zfeature_lookup_name(fname, NULL) != 0) {
411 error = EINVAL;
412 break;
413 }
414
415 has_feature = B_TRUE;
416 break;
417
418 case ZPOOL_PROP_VERSION:
419 error = nvpair_value_uint64(elem, &intval);
420 if (!error &&
421 (intval < spa_version(spa) ||
422 intval > SPA_VERSION_BEFORE_FEATURES ||
423 has_feature))
424 error = EINVAL;
425 break;
426
427 case ZPOOL_PROP_DELEGATION:
428 case ZPOOL_PROP_AUTOREPLACE:
429 case ZPOOL_PROP_LISTSNAPS:
430 case ZPOOL_PROP_AUTOEXPAND:
431 error = nvpair_value_uint64(elem, &intval);
432 if (!error && intval > 1)
433 error = EINVAL;
434 break;
435
436 case ZPOOL_PROP_BOOTFS:
437 /*
438 * If the pool version is less than SPA_VERSION_BOOTFS,
439 * or the pool is still being created (version == 0),
440 * the bootfs property cannot be set.
441 */
442 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
443 error = ENOTSUP;
444 break;
445 }
446
447 /*
448 * Make sure the vdev config is bootable
449 */
450 if (!vdev_is_bootable(spa->spa_root_vdev)) {
451 error = ENOTSUP;
452 break;
453 }
454
455 reset_bootfs = 1;
456
457 error = nvpair_value_string(elem, &strval);
458
459 if (!error) {
460 objset_t *os;
461 uint64_t compress;
462
463 if (strval == NULL || strval[0] == '\0') {
464 objnum = zpool_prop_default_numeric(
465 ZPOOL_PROP_BOOTFS);
466 break;
467 }
468
469 if (error = dmu_objset_hold(strval, FTAG, &os))
470 break;
471
472 /* Must be ZPL and not gzip compressed. */
473
474 if (dmu_objset_type(os) != DMU_OST_ZFS) {
475 error = ENOTSUP;
476 } else if ((error = dsl_prop_get_integer(strval,
477 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
478 &compress, NULL)) == 0 &&
479 !BOOTFS_COMPRESS_VALID(compress)) {
480 error = ENOTSUP;
481 } else {
482 objnum = dmu_objset_id(os);
483 }
484 dmu_objset_rele(os, FTAG);
485 }
486 break;
487
488 case ZPOOL_PROP_FAILUREMODE:
489 error = nvpair_value_uint64(elem, &intval);
490 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
491 intval > ZIO_FAILURE_MODE_PANIC))
492 error = EINVAL;
493
494 /*
495 * This is a special case which only occurs when
496 * the pool has completely failed. This allows
497 * the user to change the in-core failmode property
498 * without syncing it out to disk (I/Os might
499 * currently be blocked). We do this by returning
500 * EIO to the caller (spa_prop_set) to trick it
501 * into thinking we encountered a property validation
502 * error.
503 */
504 if (!error && spa_suspended(spa)) {
505 spa->spa_failmode = intval;
506 error = EIO;
507 }
508 break;
509
510 case ZPOOL_PROP_CACHEFILE:
511 if ((error = nvpair_value_string(elem, &strval)) != 0)
512 break;
513
514 if (strval[0] == '\0')
515 break;
516
517 if (strcmp(strval, "none") == 0)
518 break;
519
520 if (strval[0] != '/') {
521 error = EINVAL;
522 break;
523 }
524
525 slash = strrchr(strval, '/');
526 ASSERT(slash != NULL);
527
528 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
529 strcmp(slash, "/..") == 0)
530 error = EINVAL;
531 break;
532
533 case ZPOOL_PROP_COMMENT:
534 if ((error = nvpair_value_string(elem, &strval)) != 0)
535 break;
536 for (check = strval; *check != '\0'; check++) {
537 /*
538 * The kernel doesn't have an easy isprint()
539 * check. For this kernel check, we merely
540 * check ASCII apart from DEL. Fix this if
541 * there is an easy-to-use kernel isprint().
542 */
543 if (*check >= 0x7f) {
544 error = EINVAL;
545 break;
546 }
547 check++;
548 }
549 if (strlen(strval) > ZPROP_MAX_COMMENT)
550 error = E2BIG;
551 break;
552
553 case ZPOOL_PROP_DEDUPDITTO:
554 if (spa_version(spa) < SPA_VERSION_DEDUP)
555 error = ENOTSUP;
556 else
557 error = nvpair_value_uint64(elem, &intval);
558 if (error == 0 &&
559 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
560 error = EINVAL;
561 break;
562 }
563
564 if (error)
565 break;
566 }
567
568 if (!error && reset_bootfs) {
569 error = nvlist_remove(props,
570 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
571
572 if (!error) {
573 error = nvlist_add_uint64(props,
574 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
575 }
576 }
577
578 return (error);
579 }
580
581 void
582 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
583 {
584 char *cachefile;
585 spa_config_dirent_t *dp;
586
587 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
588 &cachefile) != 0)
589 return;
590
591 dp = kmem_alloc(sizeof (spa_config_dirent_t),
592 KM_SLEEP);
593
594 if (cachefile[0] == '\0')
595 dp->scd_path = spa_strdup(spa_config_path);
596 else if (strcmp(cachefile, "none") == 0)
597 dp->scd_path = NULL;
598 else
599 dp->scd_path = spa_strdup(cachefile);
600
601 list_insert_head(&spa->spa_config_list, dp);
602 if (need_sync)
603 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
604 }
605
606 int
607 spa_prop_set(spa_t *spa, nvlist_t *nvp)
608 {
609 int error;
610 nvpair_t *elem = NULL;
611 boolean_t need_sync = B_FALSE;
612
613 if ((error = spa_prop_validate(spa, nvp)) != 0)
614 return (error);
615
616 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
617 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
618
619 if (prop == ZPOOL_PROP_CACHEFILE ||
620 prop == ZPOOL_PROP_ALTROOT ||
621 prop == ZPOOL_PROP_READONLY)
622 continue;
623
624 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
625 uint64_t ver;
626
627 if (prop == ZPOOL_PROP_VERSION) {
628 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
629 } else {
630 ASSERT(zpool_prop_feature(nvpair_name(elem)));
631 ver = SPA_VERSION_FEATURES;
632 need_sync = B_TRUE;
633 }
634
635 /* Save time if the version is already set. */
636 if (ver == spa_version(spa))
637 continue;
638
639 /*
640 * In addition to the pool directory object, we might
641 * create the pool properties object, the features for
642 * read object, the features for write object, or the
643 * feature descriptions object.
644 */
645 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
646 spa_sync_version, spa, &ver, 6);
647 if (error)
648 return (error);
649 continue;
650 }
651
652 need_sync = B_TRUE;
653 break;
654 }
655
656 if (need_sync) {
657 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
658 spa, nvp, 6));
659 }
660
661 return (0);
662 }
663
664 /*
665 * If the bootfs property value is dsobj, clear it.
666 */
667 void
668 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
669 {
670 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
671 VERIFY(zap_remove(spa->spa_meta_objset,
672 spa->spa_pool_props_object,
673 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
674 spa->spa_bootfs = 0;
675 }
676 }
677
678 /*
679 * Change the GUID for the pool. This is done so that we can later
680 * re-import a pool built from a clone of our own vdevs. We will modify
681 * the root vdev's guid, our own pool guid, and then mark all of our
682 * vdevs dirty. Note that we must make sure that all our vdevs are
683 * online when we do this, or else any vdevs that weren't present
684 * would be orphaned from our pool. We are also going to issue a
685 * sysevent to update any watchers.
686 */
687 int
688 spa_change_guid(spa_t *spa)
689 {
690 uint64_t oldguid, newguid;
691 uint64_t txg;
692
693 if (!(spa_mode_global & FWRITE))
694 return (EROFS);
695
696 txg = spa_vdev_enter(spa);
697
698 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
699 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
700
701 oldguid = spa_guid(spa);
702 newguid = spa_generate_guid(NULL);
703 ASSERT3U(oldguid, !=, newguid);
704
705 spa->spa_root_vdev->vdev_guid = newguid;
706 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
707
708 vdev_config_dirty(spa->spa_root_vdev);
709
710 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
711
712 return (spa_vdev_exit(spa, NULL, txg, 0));
713 }
714
715 /*
716 * ==========================================================================
717 * SPA state manipulation (open/create/destroy/import/export)
718 * ==========================================================================
719 */
720
721 static int
722 spa_error_entry_compare(const void *a, const void *b)
723 {
724 spa_error_entry_t *sa = (spa_error_entry_t *)a;
725 spa_error_entry_t *sb = (spa_error_entry_t *)b;
726 int ret;
727
728 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
729 sizeof (zbookmark_t));
730
731 if (ret < 0)
732 return (-1);
733 else if (ret > 0)
734 return (1);
735 else
736 return (0);
737 }
738
739 /*
740 * Utility function which retrieves copies of the current logs and
741 * re-initializes them in the process.
742 */
743 void
744 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
745 {
746 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
747
748 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
749 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
750
751 avl_create(&spa->spa_errlist_scrub,
752 spa_error_entry_compare, sizeof (spa_error_entry_t),
753 offsetof(spa_error_entry_t, se_avl));
754 avl_create(&spa->spa_errlist_last,
755 spa_error_entry_compare, sizeof (spa_error_entry_t),
756 offsetof(spa_error_entry_t, se_avl));
757 }
758
759 static taskq_t *
760 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
761 uint_t value)
762 {
763 uint_t flags = 0;
764 boolean_t batch = B_FALSE;
765
766 switch (mode) {
767 case zti_mode_null:
768 return (NULL); /* no taskq needed */
769
770 case zti_mode_fixed:
771 ASSERT3U(value, >=, 1);
772 value = MAX(value, 1);
773 break;
774
775 case zti_mode_batch:
776 batch = B_TRUE;
777 flags |= TASKQ_THREADS_CPU_PCT;
778 value = zio_taskq_batch_pct;
779 break;
780
781 case zti_mode_online_percent:
782 flags |= TASKQ_THREADS_CPU_PCT;
783 break;
784
785 default:
786 panic("unrecognized mode for %s taskq (%u:%u) in "
787 "spa_activate()",
788 name, mode, value);
789 break;
790 }
791
792 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
793 if (batch)
794 flags |= TASKQ_DC_BATCH;
795
796 return (taskq_create_sysdc(name, value, 50, INT_MAX,
797 spa->spa_proc, zio_taskq_basedc, flags));
798 }
799 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
800 spa->spa_proc, flags));
801 }
802
803 static void
804 spa_create_zio_taskqs(spa_t *spa)
805 {
806 for (int t = 0; t < ZIO_TYPES; t++) {
807 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
808 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
809 enum zti_modes mode = ztip->zti_mode;
810 uint_t value = ztip->zti_value;
811 char name[32];
812
813 (void) snprintf(name, sizeof (name),
814 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
815
816 spa->spa_zio_taskq[t][q] =
817 spa_taskq_create(spa, name, mode, value);
818 }
819 }
820 }
821
822 #ifdef _KERNEL
823 static void
824 spa_thread(void *arg)
825 {
826 callb_cpr_t cprinfo;
827
828 spa_t *spa = arg;
829 user_t *pu = PTOU(curproc);
830
831 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
832 spa->spa_name);
833
834 ASSERT(curproc != &p0);
835 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
836 "zpool-%s", spa->spa_name);
837 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
838
839 /* bind this thread to the requested psrset */
840 if (zio_taskq_psrset_bind != PS_NONE) {
841 pool_lock();
842 mutex_enter(&cpu_lock);
843 mutex_enter(&pidlock);
844 mutex_enter(&curproc->p_lock);
845
846 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
847 0, NULL, NULL) == 0) {
848 curthread->t_bind_pset = zio_taskq_psrset_bind;
849 } else {
850 cmn_err(CE_WARN,
851 "Couldn't bind process for zfs pool \"%s\" to "
852 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
853 }
854
855 mutex_exit(&curproc->p_lock);
856 mutex_exit(&pidlock);
857 mutex_exit(&cpu_lock);
858 pool_unlock();
859 }
860
861 if (zio_taskq_sysdc) {
862 sysdc_thread_enter(curthread, 100, 0);
863 }
864
865 spa->spa_proc = curproc;
866 spa->spa_did = curthread->t_did;
867
868 spa_create_zio_taskqs(spa);
869
870 mutex_enter(&spa->spa_proc_lock);
871 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
872
873 spa->spa_proc_state = SPA_PROC_ACTIVE;
874 cv_broadcast(&spa->spa_proc_cv);
875
876 CALLB_CPR_SAFE_BEGIN(&cprinfo);
877 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
878 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
879 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
880
881 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
882 spa->spa_proc_state = SPA_PROC_GONE;
883 spa->spa_proc = &p0;
884 cv_broadcast(&spa->spa_proc_cv);
885 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
886
887 mutex_enter(&curproc->p_lock);
888 lwp_exit();
889 }
890 #endif
891
892 /*
893 * Activate an uninitialized pool.
894 */
895 static void
896 spa_activate(spa_t *spa, int mode)
897 {
898 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
899
900 spa->spa_state = POOL_STATE_ACTIVE;
901 spa->spa_mode = mode;
902
903 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
904 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
905
906 /* Try to create a covering process */
907 mutex_enter(&spa->spa_proc_lock);
908 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
909 ASSERT(spa->spa_proc == &p0);
910 spa->spa_did = 0;
911
912 /* Only create a process if we're going to be around a while. */
913 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
914 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
915 NULL, 0) == 0) {
916 spa->spa_proc_state = SPA_PROC_CREATED;
917 while (spa->spa_proc_state == SPA_PROC_CREATED) {
918 cv_wait(&spa->spa_proc_cv,
919 &spa->spa_proc_lock);
920 }
921 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
922 ASSERT(spa->spa_proc != &p0);
923 ASSERT(spa->spa_did != 0);
924 } else {
925 #ifdef _KERNEL
926 cmn_err(CE_WARN,
927 "Couldn't create process for zfs pool \"%s\"\n",
928 spa->spa_name);
929 #endif
930 }
931 }
932 mutex_exit(&spa->spa_proc_lock);
933
934 /* If we didn't create a process, we need to create our taskqs. */
935 if (spa->spa_proc == &p0) {
936 spa_create_zio_taskqs(spa);
937 }
938
939 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
940 offsetof(vdev_t, vdev_config_dirty_node));
941 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
942 offsetof(vdev_t, vdev_state_dirty_node));
943
944 txg_list_create(&spa->spa_vdev_txg_list,
945 offsetof(struct vdev, vdev_txg_node));
946
947 avl_create(&spa->spa_errlist_scrub,
948 spa_error_entry_compare, sizeof (spa_error_entry_t),
949 offsetof(spa_error_entry_t, se_avl));
950 avl_create(&spa->spa_errlist_last,
951 spa_error_entry_compare, sizeof (spa_error_entry_t),
952 offsetof(spa_error_entry_t, se_avl));
953 }
954
955 /*
956 * Opposite of spa_activate().
957 */
958 static void
959 spa_deactivate(spa_t *spa)
960 {
961 ASSERT(spa->spa_sync_on == B_FALSE);
962 ASSERT(spa->spa_dsl_pool == NULL);
963 ASSERT(spa->spa_root_vdev == NULL);
964 ASSERT(spa->spa_async_zio_root == NULL);
965 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
966
967 txg_list_destroy(&spa->spa_vdev_txg_list);
968
969 list_destroy(&spa->spa_config_dirty_list);
970 list_destroy(&spa->spa_state_dirty_list);
971
972 for (int t = 0; t < ZIO_TYPES; t++) {
973 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
974 if (spa->spa_zio_taskq[t][q] != NULL)
975 taskq_destroy(spa->spa_zio_taskq[t][q]);
976 spa->spa_zio_taskq[t][q] = NULL;
977 }
978 }
979
980 metaslab_class_destroy(spa->spa_normal_class);
981 spa->spa_normal_class = NULL;
982
983 metaslab_class_destroy(spa->spa_log_class);
984 spa->spa_log_class = NULL;
985
986 /*
987 * If this was part of an import or the open otherwise failed, we may
988 * still have errors left in the queues. Empty them just in case.
989 */
990 spa_errlog_drain(spa);
991
992 avl_destroy(&spa->spa_errlist_scrub);
993 avl_destroy(&spa->spa_errlist_last);
994
995 spa->spa_state = POOL_STATE_UNINITIALIZED;
996
997 mutex_enter(&spa->spa_proc_lock);
998 if (spa->spa_proc_state != SPA_PROC_NONE) {
999 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1000 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1001 cv_broadcast(&spa->spa_proc_cv);
1002 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1003 ASSERT(spa->spa_proc != &p0);
1004 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1005 }
1006 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1007 spa->spa_proc_state = SPA_PROC_NONE;
1008 }
1009 ASSERT(spa->spa_proc == &p0);
1010 mutex_exit(&spa->spa_proc_lock);
1011
1012 /*
1013 * We want to make sure spa_thread() has actually exited the ZFS
1014 * module, so that the module can't be unloaded out from underneath
1015 * it.
1016 */
1017 if (spa->spa_did != 0) {
1018 thread_join(spa->spa_did);
1019 spa->spa_did = 0;
1020 }
1021 }
1022
1023 /*
1024 * Verify a pool configuration, and construct the vdev tree appropriately. This
1025 * will create all the necessary vdevs in the appropriate layout, with each vdev
1026 * in the CLOSED state. This will prep the pool before open/creation/import.
1027 * All vdev validation is done by the vdev_alloc() routine.
1028 */
1029 static int
1030 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1031 uint_t id, int atype)
1032 {
1033 nvlist_t **child;
1034 uint_t children;
1035 int error;
1036
1037 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1038 return (error);
1039
1040 if ((*vdp)->vdev_ops->vdev_op_leaf)
1041 return (0);
1042
1043 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1044 &child, &children);
1045
1046 if (error == ENOENT)
1047 return (0);
1048
1049 if (error) {
1050 vdev_free(*vdp);
1051 *vdp = NULL;
1052 return (EINVAL);
1053 }
1054
1055 for (int c = 0; c < children; c++) {
1056 vdev_t *vd;
1057 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1058 atype)) != 0) {
1059 vdev_free(*vdp);
1060 *vdp = NULL;
1061 return (error);
1062 }
1063 }
1064
1065 ASSERT(*vdp != NULL);
1066
1067 return (0);
1068 }
1069
1070 /*
1071 * Opposite of spa_load().
1072 */
1073 static void
1074 spa_unload(spa_t *spa)
1075 {
1076 int i;
1077
1078 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1079
1080 /*
1081 * Stop async tasks.
1082 */
1083 spa_async_suspend(spa);
1084
1085 /*
1086 * Stop syncing.
1087 */
1088 if (spa->spa_sync_on) {
1089 txg_sync_stop(spa->spa_dsl_pool);
1090 spa->spa_sync_on = B_FALSE;
1091 }
1092
1093 /*
1094 * Wait for any outstanding async I/O to complete.
1095 */
1096 if (spa->spa_async_zio_root != NULL) {
1097 (void) zio_wait(spa->spa_async_zio_root);
1098 spa->spa_async_zio_root = NULL;
1099 }
1100
1101 bpobj_close(&spa->spa_deferred_bpobj);
1102
1103 /*
1104 * Close the dsl pool.
1105 */
1106 if (spa->spa_dsl_pool) {
1107 dsl_pool_close(spa->spa_dsl_pool);
1108 spa->spa_dsl_pool = NULL;
1109 spa->spa_meta_objset = NULL;
1110 }
1111
1112 ddt_unload(spa);
1113
1114 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1115
1116 /*
1117 * Drop and purge level 2 cache
1118 */
1119 spa_l2cache_drop(spa);
1120
1121 /*
1122 * Close all vdevs.
1123 */
1124 if (spa->spa_root_vdev)
1125 vdev_free(spa->spa_root_vdev);
1126 ASSERT(spa->spa_root_vdev == NULL);
1127
1128 for (i = 0; i < spa->spa_spares.sav_count; i++)
1129 vdev_free(spa->spa_spares.sav_vdevs[i]);
1130 if (spa->spa_spares.sav_vdevs) {
1131 kmem_free(spa->spa_spares.sav_vdevs,
1132 spa->spa_spares.sav_count * sizeof (void *));
1133 spa->spa_spares.sav_vdevs = NULL;
1134 }
1135 if (spa->spa_spares.sav_config) {
1136 nvlist_free(spa->spa_spares.sav_config);
1137 spa->spa_spares.sav_config = NULL;
1138 }
1139 spa->spa_spares.sav_count = 0;
1140
1141 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1142 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1143 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1144 }
1145 if (spa->spa_l2cache.sav_vdevs) {
1146 kmem_free(spa->spa_l2cache.sav_vdevs,
1147 spa->spa_l2cache.sav_count * sizeof (void *));
1148 spa->spa_l2cache.sav_vdevs = NULL;
1149 }
1150 if (spa->spa_l2cache.sav_config) {
1151 nvlist_free(spa->spa_l2cache.sav_config);
1152 spa->spa_l2cache.sav_config = NULL;
1153 }
1154 spa->spa_l2cache.sav_count = 0;
1155
1156 spa->spa_async_suspended = 0;
1157
1158 if (spa->spa_comment != NULL) {
1159 spa_strfree(spa->spa_comment);
1160 spa->spa_comment = NULL;
1161 }
1162
1163 spa_config_exit(spa, SCL_ALL, FTAG);
1164 }
1165
1166 /*
1167 * Load (or re-load) the current list of vdevs describing the active spares for
1168 * this pool. When this is called, we have some form of basic information in
1169 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1170 * then re-generate a more complete list including status information.
1171 */
1172 static void
1173 spa_load_spares(spa_t *spa)
1174 {
1175 nvlist_t **spares;
1176 uint_t nspares;
1177 int i;
1178 vdev_t *vd, *tvd;
1179
1180 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1181
1182 /*
1183 * First, close and free any existing spare vdevs.
1184 */
1185 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1186 vd = spa->spa_spares.sav_vdevs[i];
1187
1188 /* Undo the call to spa_activate() below */
1189 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1190 B_FALSE)) != NULL && tvd->vdev_isspare)
1191 spa_spare_remove(tvd);
1192 vdev_close(vd);
1193 vdev_free(vd);
1194 }
1195
1196 if (spa->spa_spares.sav_vdevs)
1197 kmem_free(spa->spa_spares.sav_vdevs,
1198 spa->spa_spares.sav_count * sizeof (void *));
1199
1200 if (spa->spa_spares.sav_config == NULL)
1201 nspares = 0;
1202 else
1203 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1204 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1205
1206 spa->spa_spares.sav_count = (int)nspares;
1207 spa->spa_spares.sav_vdevs = NULL;
1208
1209 if (nspares == 0)
1210 return;
1211
1212 /*
1213 * Construct the array of vdevs, opening them to get status in the
1214 * process. For each spare, there is potentially two different vdev_t
1215 * structures associated with it: one in the list of spares (used only
1216 * for basic validation purposes) and one in the active vdev
1217 * configuration (if it's spared in). During this phase we open and
1218 * validate each vdev on the spare list. If the vdev also exists in the
1219 * active configuration, then we also mark this vdev as an active spare.
1220 */
1221 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1222 KM_SLEEP);
1223 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1224 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1225 VDEV_ALLOC_SPARE) == 0);
1226 ASSERT(vd != NULL);
1227
1228 spa->spa_spares.sav_vdevs[i] = vd;
1229
1230 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1231 B_FALSE)) != NULL) {
1232 if (!tvd->vdev_isspare)
1233 spa_spare_add(tvd);
1234
1235 /*
1236 * We only mark the spare active if we were successfully
1237 * able to load the vdev. Otherwise, importing a pool
1238 * with a bad active spare would result in strange
1239 * behavior, because multiple pool would think the spare
1240 * is actively in use.
1241 *
1242 * There is a vulnerability here to an equally bizarre
1243 * circumstance, where a dead active spare is later
1244 * brought back to life (onlined or otherwise). Given
1245 * the rarity of this scenario, and the extra complexity
1246 * it adds, we ignore the possibility.
1247 */
1248 if (!vdev_is_dead(tvd))
1249 spa_spare_activate(tvd);
1250 }
1251
1252 vd->vdev_top = vd;
1253 vd->vdev_aux = &spa->spa_spares;
1254
1255 if (vdev_open(vd) != 0)
1256 continue;
1257
1258 if (vdev_validate_aux(vd) == 0)
1259 spa_spare_add(vd);
1260 }
1261
1262 /*
1263 * Recompute the stashed list of spares, with status information
1264 * this time.
1265 */
1266 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1267 DATA_TYPE_NVLIST_ARRAY) == 0);
1268
1269 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1270 KM_SLEEP);
1271 for (i = 0; i < spa->spa_spares.sav_count; i++)
1272 spares[i] = vdev_config_generate(spa,
1273 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1274 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1275 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1276 for (i = 0; i < spa->spa_spares.sav_count; i++)
1277 nvlist_free(spares[i]);
1278 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1279 }
1280
1281 /*
1282 * Load (or re-load) the current list of vdevs describing the active l2cache for
1283 * this pool. When this is called, we have some form of basic information in
1284 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1285 * then re-generate a more complete list including status information.
1286 * Devices which are already active have their details maintained, and are
1287 * not re-opened.
1288 */
1289 static void
1290 spa_load_l2cache(spa_t *spa)
1291 {
1292 nvlist_t **l2cache;
1293 uint_t nl2cache;
1294 int i, j, oldnvdevs;
1295 uint64_t guid;
1296 vdev_t *vd, **oldvdevs, **newvdevs;
1297 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1298
1299 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1300
1301 if (sav->sav_config != NULL) {
1302 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1303 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1304 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1305 } else {
1306 nl2cache = 0;
1307 }
1308
1309 oldvdevs = sav->sav_vdevs;
1310 oldnvdevs = sav->sav_count;
1311 sav->sav_vdevs = NULL;
1312 sav->sav_count = 0;
1313
1314 /*
1315 * Process new nvlist of vdevs.
1316 */
1317 for (i = 0; i < nl2cache; i++) {
1318 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1319 &guid) == 0);
1320
1321 newvdevs[i] = NULL;
1322 for (j = 0; j < oldnvdevs; j++) {
1323 vd = oldvdevs[j];
1324 if (vd != NULL && guid == vd->vdev_guid) {
1325 /*
1326 * Retain previous vdev for add/remove ops.
1327 */
1328 newvdevs[i] = vd;
1329 oldvdevs[j] = NULL;
1330 break;
1331 }
1332 }
1333
1334 if (newvdevs[i] == NULL) {
1335 /*
1336 * Create new vdev
1337 */
1338 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1339 VDEV_ALLOC_L2CACHE) == 0);
1340 ASSERT(vd != NULL);
1341 newvdevs[i] = vd;
1342
1343 /*
1344 * Commit this vdev as an l2cache device,
1345 * even if it fails to open.
1346 */
1347 spa_l2cache_add(vd);
1348
1349 vd->vdev_top = vd;
1350 vd->vdev_aux = sav;
1351
1352 spa_l2cache_activate(vd);
1353
1354 if (vdev_open(vd) != 0)
1355 continue;
1356
1357 (void) vdev_validate_aux(vd);
1358
1359 if (!vdev_is_dead(vd))
1360 l2arc_add_vdev(spa, vd);
1361 }
1362 }
1363
1364 /*
1365 * Purge vdevs that were dropped
1366 */
1367 for (i = 0; i < oldnvdevs; i++) {
1368 uint64_t pool;
1369
1370 vd = oldvdevs[i];
1371 if (vd != NULL) {
1372 ASSERT(vd->vdev_isl2cache);
1373
1374 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1375 pool != 0ULL && l2arc_vdev_present(vd))
1376 l2arc_remove_vdev(vd);
1377 vdev_clear_stats(vd);
1378 vdev_free(vd);
1379 }
1380 }
1381
1382 if (oldvdevs)
1383 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1384
1385 if (sav->sav_config == NULL)
1386 goto out;
1387
1388 sav->sav_vdevs = newvdevs;
1389 sav->sav_count = (int)nl2cache;
1390
1391 /*
1392 * Recompute the stashed list of l2cache devices, with status
1393 * information this time.
1394 */
1395 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1396 DATA_TYPE_NVLIST_ARRAY) == 0);
1397
1398 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1399 for (i = 0; i < sav->sav_count; i++)
1400 l2cache[i] = vdev_config_generate(spa,
1401 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1402 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1403 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1404 out:
1405 for (i = 0; i < sav->sav_count; i++)
1406 nvlist_free(l2cache[i]);
1407 if (sav->sav_count)
1408 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1409 }
1410
1411 static int
1412 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1413 {
1414 dmu_buf_t *db;
1415 char *packed = NULL;
1416 size_t nvsize = 0;
1417 int error;
1418 *value = NULL;
1419
1420 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1421 nvsize = *(uint64_t *)db->db_data;
1422 dmu_buf_rele(db, FTAG);
1423
1424 packed = kmem_alloc(nvsize, KM_SLEEP);
1425 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1426 DMU_READ_PREFETCH);
1427 if (error == 0)
1428 error = nvlist_unpack(packed, nvsize, value, 0);
1429 kmem_free(packed, nvsize);
1430
1431 return (error);
1432 }
1433
1434 /*
1435 * Checks to see if the given vdev could not be opened, in which case we post a
1436 * sysevent to notify the autoreplace code that the device has been removed.
1437 */
1438 static void
1439 spa_check_removed(vdev_t *vd)
1440 {
1441 for (int c = 0; c < vd->vdev_children; c++)
1442 spa_check_removed(vd->vdev_child[c]);
1443
1444 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1445 zfs_post_autoreplace(vd->vdev_spa, vd);
1446 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1447 }
1448 }
1449
1450 /*
1451 * Validate the current config against the MOS config
1452 */
1453 static boolean_t
1454 spa_config_valid(spa_t *spa, nvlist_t *config)
1455 {
1456 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1457 nvlist_t *nv;
1458
1459 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1460
1461 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1462 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1463
1464 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1465
1466 /*
1467 * If we're doing a normal import, then build up any additional
1468 * diagnostic information about missing devices in this config.
1469 * We'll pass this up to the user for further processing.
1470 */
1471 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1472 nvlist_t **child, *nv;
1473 uint64_t idx = 0;
1474
1475 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1476 KM_SLEEP);
1477 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1478
1479 for (int c = 0; c < rvd->vdev_children; c++) {
1480 vdev_t *tvd = rvd->vdev_child[c];
1481 vdev_t *mtvd = mrvd->vdev_child[c];
1482
1483 if (tvd->vdev_ops == &vdev_missing_ops &&
1484 mtvd->vdev_ops != &vdev_missing_ops &&
1485 mtvd->vdev_islog)
1486 child[idx++] = vdev_config_generate(spa, mtvd,
1487 B_FALSE, 0);
1488 }
1489
1490 if (idx) {
1491 VERIFY(nvlist_add_nvlist_array(nv,
1492 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1493 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1494 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1495
1496 for (int i = 0; i < idx; i++)
1497 nvlist_free(child[i]);
1498 }
1499 nvlist_free(nv);
1500 kmem_free(child, rvd->vdev_children * sizeof (char **));
1501 }
1502
1503 /*
1504 * Compare the root vdev tree with the information we have
1505 * from the MOS config (mrvd). Check each top-level vdev
1506 * with the corresponding MOS config top-level (mtvd).
1507 */
1508 for (int c = 0; c < rvd->vdev_children; c++) {
1509 vdev_t *tvd = rvd->vdev_child[c];
1510 vdev_t *mtvd = mrvd->vdev_child[c];
1511
1512 /*
1513 * Resolve any "missing" vdevs in the current configuration.
1514 * If we find that the MOS config has more accurate information
1515 * about the top-level vdev then use that vdev instead.
1516 */
1517 if (tvd->vdev_ops == &vdev_missing_ops &&
1518 mtvd->vdev_ops != &vdev_missing_ops) {
1519
1520 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1521 continue;
1522
1523 /*
1524 * Device specific actions.
1525 */
1526 if (mtvd->vdev_islog) {
1527 spa_set_log_state(spa, SPA_LOG_CLEAR);
1528 } else {
1529 /*
1530 * XXX - once we have 'readonly' pool
1531 * support we should be able to handle
1532 * missing data devices by transitioning
1533 * the pool to readonly.
1534 */
1535 continue;
1536 }
1537
1538 /*
1539 * Swap the missing vdev with the data we were
1540 * able to obtain from the MOS config.
1541 */
1542 vdev_remove_child(rvd, tvd);
1543 vdev_remove_child(mrvd, mtvd);
1544
1545 vdev_add_child(rvd, mtvd);
1546 vdev_add_child(mrvd, tvd);
1547
1548 spa_config_exit(spa, SCL_ALL, FTAG);
1549 vdev_load(mtvd);
1550 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1551
1552 vdev_reopen(rvd);
1553 } else if (mtvd->vdev_islog) {
1554 /*
1555 * Load the slog device's state from the MOS config
1556 * since it's possible that the label does not
1557 * contain the most up-to-date information.
1558 */
1559 vdev_load_log_state(tvd, mtvd);
1560 vdev_reopen(tvd);
1561 }
1562 }
1563 vdev_free(mrvd);
1564 spa_config_exit(spa, SCL_ALL, FTAG);
1565
1566 /*
1567 * Ensure we were able to validate the config.
1568 */
1569 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1570 }
1571
1572 /*
1573 * Check for missing log devices
1574 */
1575 static int
1576 spa_check_logs(spa_t *spa)
1577 {
1578 switch (spa->spa_log_state) {
1579 case SPA_LOG_MISSING:
1580 /* need to recheck in case slog has been restored */
1581 case SPA_LOG_UNKNOWN:
1582 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1583 DS_FIND_CHILDREN)) {
1584 spa_set_log_state(spa, SPA_LOG_MISSING);
1585 return (1);
1586 }
1587 break;
1588 }
1589 return (0);
1590 }
1591
1592 static boolean_t
1593 spa_passivate_log(spa_t *spa)
1594 {
1595 vdev_t *rvd = spa->spa_root_vdev;
1596 boolean_t slog_found = B_FALSE;
1597
1598 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1599
1600 if (!spa_has_slogs(spa))
1601 return (B_FALSE);
1602
1603 for (int c = 0; c < rvd->vdev_children; c++) {
1604 vdev_t *tvd = rvd->vdev_child[c];
1605 metaslab_group_t *mg = tvd->vdev_mg;
1606
1607 if (tvd->vdev_islog) {
1608 metaslab_group_passivate(mg);
1609 slog_found = B_TRUE;
1610 }
1611 }
1612
1613 return (slog_found);
1614 }
1615
1616 static void
1617 spa_activate_log(spa_t *spa)
1618 {
1619 vdev_t *rvd = spa->spa_root_vdev;
1620
1621 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1622
1623 for (int c = 0; c < rvd->vdev_children; c++) {
1624 vdev_t *tvd = rvd->vdev_child[c];
1625 metaslab_group_t *mg = tvd->vdev_mg;
1626
1627 if (tvd->vdev_islog)
1628 metaslab_group_activate(mg);
1629 }
1630 }
1631
1632 int
1633 spa_offline_log(spa_t *spa)
1634 {
1635 int error = 0;
1636
1637 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1638 NULL, DS_FIND_CHILDREN)) == 0) {
1639
1640 /*
1641 * We successfully offlined the log device, sync out the
1642 * current txg so that the "stubby" block can be removed
1643 * by zil_sync().
1644 */
1645 txg_wait_synced(spa->spa_dsl_pool, 0);
1646 }
1647 return (error);
1648 }
1649
1650 static void
1651 spa_aux_check_removed(spa_aux_vdev_t *sav)
1652 {
1653 for (int i = 0; i < sav->sav_count; i++)
1654 spa_check_removed(sav->sav_vdevs[i]);
1655 }
1656
1657 void
1658 spa_claim_notify(zio_t *zio)
1659 {
1660 spa_t *spa = zio->io_spa;
1661
1662 if (zio->io_error)
1663 return;
1664
1665 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1666 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1667 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1668 mutex_exit(&spa->spa_props_lock);
1669 }
1670
1671 typedef struct spa_load_error {
1672 uint64_t sle_meta_count;
1673 uint64_t sle_data_count;
1674 } spa_load_error_t;
1675
1676 static void
1677 spa_load_verify_done(zio_t *zio)
1678 {
1679 blkptr_t *bp = zio->io_bp;
1680 spa_load_error_t *sle = zio->io_private;
1681 dmu_object_type_t type = BP_GET_TYPE(bp);
1682 int error = zio->io_error;
1683
1684 if (error) {
1685 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1686 type != DMU_OT_INTENT_LOG)
1687 atomic_add_64(&sle->sle_meta_count, 1);
1688 else
1689 atomic_add_64(&sle->sle_data_count, 1);
1690 }
1691 zio_data_buf_free(zio->io_data, zio->io_size);
1692 }
1693
1694 /*ARGSUSED*/
1695 static int
1696 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1697 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1698 {
1699 if (bp != NULL) {
1700 zio_t *rio = arg;
1701 size_t size = BP_GET_PSIZE(bp);
1702 void *data = zio_data_buf_alloc(size);
1703
1704 zio_nowait(zio_read(rio, spa, bp, data, size,
1705 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1706 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1707 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1708 }
1709 return (0);
1710 }
1711
1712 static int
1713 spa_load_verify(spa_t *spa)
1714 {
1715 zio_t *rio;
1716 spa_load_error_t sle = { 0 };
1717 zpool_rewind_policy_t policy;
1718 boolean_t verify_ok = B_FALSE;
1719 int error;
1720
1721 zpool_get_rewind_policy(spa->spa_config, &policy);
1722
1723 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1724 return (0);
1725
1726 rio = zio_root(spa, NULL, &sle,
1727 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1728
1729 error = traverse_pool(spa, spa->spa_verify_min_txg,
1730 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1731
1732 (void) zio_wait(rio);
1733
1734 spa->spa_load_meta_errors = sle.sle_meta_count;
1735 spa->spa_load_data_errors = sle.sle_data_count;
1736
1737 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1738 sle.sle_data_count <= policy.zrp_maxdata) {
1739 int64_t loss = 0;
1740
1741 verify_ok = B_TRUE;
1742 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1743 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1744
1745 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1746 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1747 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1748 VERIFY(nvlist_add_int64(spa->spa_load_info,
1749 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1750 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1751 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1752 } else {
1753 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1754 }
1755
1756 if (error) {
1757 if (error != ENXIO && error != EIO)
1758 error = EIO;
1759 return (error);
1760 }
1761
1762 return (verify_ok ? 0 : EIO);
1763 }
1764
1765 /*
1766 * Find a value in the pool props object.
1767 */
1768 static void
1769 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1770 {
1771 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1772 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1773 }
1774
1775 /*
1776 * Find a value in the pool directory object.
1777 */
1778 static int
1779 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1780 {
1781 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1782 name, sizeof (uint64_t), 1, val));
1783 }
1784
1785 static int
1786 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1787 {
1788 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1789 return (err);
1790 }
1791
1792 /*
1793 * Fix up config after a partly-completed split. This is done with the
1794 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1795 * pool have that entry in their config, but only the splitting one contains
1796 * a list of all the guids of the vdevs that are being split off.
1797 *
1798 * This function determines what to do with that list: either rejoin
1799 * all the disks to the pool, or complete the splitting process. To attempt
1800 * the rejoin, each disk that is offlined is marked online again, and
1801 * we do a reopen() call. If the vdev label for every disk that was
1802 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1803 * then we call vdev_split() on each disk, and complete the split.
1804 *
1805 * Otherwise we leave the config alone, with all the vdevs in place in
1806 * the original pool.
1807 */
1808 static void
1809 spa_try_repair(spa_t *spa, nvlist_t *config)
1810 {
1811 uint_t extracted;
1812 uint64_t *glist;
1813 uint_t i, gcount;
1814 nvlist_t *nvl;
1815 vdev_t **vd;
1816 boolean_t attempt_reopen;
1817
1818 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1819 return;
1820
1821 /* check that the config is complete */
1822 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1823 &glist, &gcount) != 0)
1824 return;
1825
1826 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1827
1828 /* attempt to online all the vdevs & validate */
1829 attempt_reopen = B_TRUE;
1830 for (i = 0; i < gcount; i++) {
1831 if (glist[i] == 0) /* vdev is hole */
1832 continue;
1833
1834 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1835 if (vd[i] == NULL) {
1836 /*
1837 * Don't bother attempting to reopen the disks;
1838 * just do the split.
1839 */
1840 attempt_reopen = B_FALSE;
1841 } else {
1842 /* attempt to re-online it */
1843 vd[i]->vdev_offline = B_FALSE;
1844 }
1845 }
1846
1847 if (attempt_reopen) {
1848 vdev_reopen(spa->spa_root_vdev);
1849
1850 /* check each device to see what state it's in */
1851 for (extracted = 0, i = 0; i < gcount; i++) {
1852 if (vd[i] != NULL &&
1853 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1854 break;
1855 ++extracted;
1856 }
1857 }
1858
1859 /*
1860 * If every disk has been moved to the new pool, or if we never
1861 * even attempted to look at them, then we split them off for
1862 * good.
1863 */
1864 if (!attempt_reopen || gcount == extracted) {
1865 for (i = 0; i < gcount; i++)
1866 if (vd[i] != NULL)
1867 vdev_split(vd[i]);
1868 vdev_reopen(spa->spa_root_vdev);
1869 }
1870
1871 kmem_free(vd, gcount * sizeof (vdev_t *));
1872 }
1873
1874 static int
1875 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1876 boolean_t mosconfig)
1877 {
1878 nvlist_t *config = spa->spa_config;
1879 char *ereport = FM_EREPORT_ZFS_POOL;
1880 char *comment;
1881 int error;
1882 uint64_t pool_guid;
1883 nvlist_t *nvl;
1884
1885 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1886 return (EINVAL);
1887
1888 ASSERT(spa->spa_comment == NULL);
1889 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1890 spa->spa_comment = spa_strdup(comment);
1891
1892 /*
1893 * Versioning wasn't explicitly added to the label until later, so if
1894 * it's not present treat it as the initial version.
1895 */
1896 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1897 &spa->spa_ubsync.ub_version) != 0)
1898 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1899
1900 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1901 &spa->spa_config_txg);
1902
1903 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1904 spa_guid_exists(pool_guid, 0)) {
1905 error = EEXIST;
1906 } else {
1907 spa->spa_config_guid = pool_guid;
1908
1909 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1910 &nvl) == 0) {
1911 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1912 KM_SLEEP) == 0);
1913 }
1914
1915 nvlist_free(spa->spa_load_info);
1916 spa->spa_load_info = fnvlist_alloc();
1917
1918 gethrestime(&spa->spa_loaded_ts);
1919 error = spa_load_impl(spa, pool_guid, config, state, type,
1920 mosconfig, &ereport);
1921 }
1922
1923 spa->spa_minref = refcount_count(&spa->spa_refcount);
1924 if (error) {
1925 if (error != EEXIST) {
1926 spa->spa_loaded_ts.tv_sec = 0;
1927 spa->spa_loaded_ts.tv_nsec = 0;
1928 }
1929 if (error != EBADF) {
1930 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1931 }
1932 }
1933 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1934 spa->spa_ena = 0;
1935
1936 return (error);
1937 }
1938
1939 /*
1940 * Load an existing storage pool, using the pool's builtin spa_config as a
1941 * source of configuration information.
1942 */
1943 static int
1944 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1945 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1946 char **ereport)
1947 {
1948 int error = 0;
1949 nvlist_t *nvroot = NULL;
1950 nvlist_t *label;
1951 vdev_t *rvd;
1952 uberblock_t *ub = &spa->spa_uberblock;
1953 uint64_t children, config_cache_txg = spa->spa_config_txg;
1954 int orig_mode = spa->spa_mode;
1955 int parse;
1956 uint64_t obj;
1957 boolean_t missing_feat_write = B_FALSE;
1958
1959 /*
1960 * If this is an untrusted config, access the pool in read-only mode.
1961 * This prevents things like resilvering recently removed devices.
1962 */
1963 if (!mosconfig)
1964 spa->spa_mode = FREAD;
1965
1966 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1967
1968 spa->spa_load_state = state;
1969
1970 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1971 return (EINVAL);
1972
1973 parse = (type == SPA_IMPORT_EXISTING ?
1974 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1975
1976 /*
1977 * Create "The Godfather" zio to hold all async IOs
1978 */
1979 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1980 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1981
1982 /*
1983 * Parse the configuration into a vdev tree. We explicitly set the
1984 * value that will be returned by spa_version() since parsing the
1985 * configuration requires knowing the version number.
1986 */
1987 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1988 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1989 spa_config_exit(spa, SCL_ALL, FTAG);
1990
1991 if (error != 0)
1992 return (error);
1993
1994 ASSERT(spa->spa_root_vdev == rvd);
1995
1996 if (type != SPA_IMPORT_ASSEMBLE) {
1997 ASSERT(spa_guid(spa) == pool_guid);
1998 }
1999
2000 /*
2001 * Try to open all vdevs, loading each label in the process.
2002 */
2003 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2004 error = vdev_open(rvd);
2005 spa_config_exit(spa, SCL_ALL, FTAG);
2006 if (error != 0)
2007 return (error);
2008
2009 /*
2010 * We need to validate the vdev labels against the configuration that
2011 * we have in hand, which is dependent on the setting of mosconfig. If
2012 * mosconfig is true then we're validating the vdev labels based on
2013 * that config. Otherwise, we're validating against the cached config
2014 * (zpool.cache) that was read when we loaded the zfs module, and then
2015 * later we will recursively call spa_load() and validate against
2016 * the vdev config.
2017 *
2018 * If we're assembling a new pool that's been split off from an
2019 * existing pool, the labels haven't yet been updated so we skip
2020 * validation for now.
2021 */
2022 if (type != SPA_IMPORT_ASSEMBLE) {
2023 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2024 error = vdev_validate(rvd, mosconfig);
2025 spa_config_exit(spa, SCL_ALL, FTAG);
2026
2027 if (error != 0)
2028 return (error);
2029
2030 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2031 return (ENXIO);
2032 }
2033
2034 /*
2035 * Find the best uberblock.
2036 */
2037 vdev_uberblock_load(rvd, ub, &label);
2038
2039 /*
2040 * If we weren't able to find a single valid uberblock, return failure.
2041 */
2042 if (ub->ub_txg == 0) {
2043 nvlist_free(label);
2044 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2045 }
2046
2047 /*
2048 * If the pool has an unsupported version we can't open it.
2049 */
2050 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2051 nvlist_free(label);
2052 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2053 }
2054
2055 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2056 nvlist_t *features;
2057
2058 /*
2059 * If we weren't able to find what's necessary for reading the
2060 * MOS in the label, return failure.
2061 */
2062 if (label == NULL || nvlist_lookup_nvlist(label,
2063 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2064 nvlist_free(label);
2065 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2066 ENXIO));
2067 }
2068
2069 /*
2070 * Update our in-core representation with the definitive values
2071 * from the label.
2072 */
2073 nvlist_free(spa->spa_label_features);
2074 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2075 }
2076
2077 nvlist_free(label);
2078
2079 /*
2080 * Look through entries in the label nvlist's features_for_read. If
2081 * there is a feature listed there which we don't understand then we
2082 * cannot open a pool.
2083 */
2084 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2085 nvlist_t *unsup_feat;
2086
2087 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2088 0);
2089
2090 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2091 NULL); nvp != NULL;
2092 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2093 if (!zfeature_is_supported(nvpair_name(nvp))) {
2094 VERIFY(nvlist_add_string(unsup_feat,
2095 nvpair_name(nvp), "") == 0);
2096 }
2097 }
2098
2099 if (!nvlist_empty(unsup_feat)) {
2100 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2101 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2102 nvlist_free(unsup_feat);
2103 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2104 ENOTSUP));
2105 }
2106
2107 nvlist_free(unsup_feat);
2108 }
2109
2110 /*
2111 * If the vdev guid sum doesn't match the uberblock, we have an
2112 * incomplete configuration. We first check to see if the pool
2113 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2114 * If it is, defer the vdev_guid_sum check till later so we
2115 * can handle missing vdevs.
2116 */
2117 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2118 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2119 rvd->vdev_guid_sum != ub->ub_guid_sum)
2120 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2121
2122 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2123 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2124 spa_try_repair(spa, config);
2125 spa_config_exit(spa, SCL_ALL, FTAG);
2126 nvlist_free(spa->spa_config_splitting);
2127 spa->spa_config_splitting = NULL;
2128 }
2129
2130 /*
2131 * Initialize internal SPA structures.
2132 */
2133 spa->spa_state = POOL_STATE_ACTIVE;
2134 spa->spa_ubsync = spa->spa_uberblock;
2135 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2136 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2137 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2138 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2139 spa->spa_claim_max_txg = spa->spa_first_txg;
2140 spa->spa_prev_software_version = ub->ub_software_version;
2141
2142 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2143 if (error)
2144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2145 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2146
2147 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2148 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2149
2150 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2151 boolean_t missing_feat_read = B_FALSE;
2152 nvlist_t *unsup_feat;
2153
2154 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2155 &spa->spa_feat_for_read_obj) != 0) {
2156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2157 }
2158
2159 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2160 &spa->spa_feat_for_write_obj) != 0) {
2161 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2162 }
2163
2164 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2165 &spa->spa_feat_desc_obj) != 0) {
2166 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2167 }
2168
2169 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2170 0);
2171
2172 if (!feature_is_supported(spa->spa_meta_objset,
2173 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2174 unsup_feat))
2175 missing_feat_read = B_TRUE;
2176
2177 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2178 if (!feature_is_supported(spa->spa_meta_objset,
2179 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2180 unsup_feat))
2181 missing_feat_write = B_TRUE;
2182 }
2183
2184 if (!nvlist_empty(unsup_feat)) {
2185 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2186 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2187 }
2188
2189 nvlist_free(unsup_feat);
2190
2191 if (!missing_feat_read) {
2192 fnvlist_add_boolean(spa->spa_load_info,
2193 ZPOOL_CONFIG_CAN_RDONLY);
2194 }
2195
2196 /*
2197 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2198 * twofold: to determine whether the pool is available for
2199 * import in read-write mode and (if it is not) whether the
2200 * pool is available for import in read-only mode. If the pool
2201 * is available for import in read-write mode, it is displayed
2202 * as available in userland; if it is not available for import
2203 * in read-only mode, it is displayed as unavailable in
2204 * userland. If the pool is available for import in read-only
2205 * mode but not read-write mode, it is displayed as unavailable
2206 * in userland with a special note that the pool is actually
2207 * available for open in read-only mode.
2208 *
2209 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2210 * missing a feature for write, we must first determine whether
2211 * the pool can be opened read-only before returning to
2212 * userland in order to know whether to display the
2213 * abovementioned note.
2214 */
2215 if (missing_feat_read || (missing_feat_write &&
2216 spa_writeable(spa))) {
2217 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2218 ENOTSUP));
2219 }
2220 }
2221
2222 spa->spa_is_initializing = B_TRUE;
2223 error = dsl_pool_open(spa->spa_dsl_pool);
2224 spa->spa_is_initializing = B_FALSE;
2225 if (error != 0)
2226 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2227
2228 if (!mosconfig) {
2229 uint64_t hostid;
2230 nvlist_t *policy = NULL, *nvconfig;
2231
2232 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2233 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2234
2235 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2236 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2237 char *hostname;
2238 unsigned long myhostid = 0;
2239
2240 VERIFY(nvlist_lookup_string(nvconfig,
2241 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2242
2243 #ifdef _KERNEL
2244 myhostid = zone_get_hostid(NULL);
2245 #else /* _KERNEL */
2246 /*
2247 * We're emulating the system's hostid in userland, so
2248 * we can't use zone_get_hostid().
2249 */
2250 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2251 #endif /* _KERNEL */
2252 if (hostid != 0 && myhostid != 0 &&
2253 hostid != myhostid) {
2254 nvlist_free(nvconfig);
2255 cmn_err(CE_WARN, "pool '%s' could not be "
2256 "loaded as it was last accessed by "
2257 "another system (host: %s hostid: 0x%lx). "
2258 "See: http://illumos.org/msg/ZFS-8000-EY",
2259 spa_name(spa), hostname,
2260 (unsigned long)hostid);
2261 return (EBADF);
2262 }
2263 }
2264 if (nvlist_lookup_nvlist(spa->spa_config,
2265 ZPOOL_REWIND_POLICY, &policy) == 0)
2266 VERIFY(nvlist_add_nvlist(nvconfig,
2267 ZPOOL_REWIND_POLICY, policy) == 0);
2268
2269 spa_config_set(spa, nvconfig);
2270 spa_unload(spa);
2271 spa_deactivate(spa);
2272 spa_activate(spa, orig_mode);
2273
2274 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2275 }
2276
2277 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2278 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2279 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2280 if (error != 0)
2281 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2282
2283 /*
2284 * Load the bit that tells us to use the new accounting function
2285 * (raid-z deflation). If we have an older pool, this will not
2286 * be present.
2287 */
2288 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2289 if (error != 0 && error != ENOENT)
2290 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2291
2292 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2293 &spa->spa_creation_version);
2294 if (error != 0 && error != ENOENT)
2295 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2296
2297 /*
2298 * Load the persistent error log. If we have an older pool, this will
2299 * not be present.
2300 */
2301 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2302 if (error != 0 && error != ENOENT)
2303 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2304
2305 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2306 &spa->spa_errlog_scrub);
2307 if (error != 0 && error != ENOENT)
2308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2309
2310 /*
2311 * Load the history object. If we have an older pool, this
2312 * will not be present.
2313 */
2314 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2315 if (error != 0 && error != ENOENT)
2316 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2317
2318 /*
2319 * If we're assembling the pool from the split-off vdevs of
2320 * an existing pool, we don't want to attach the spares & cache
2321 * devices.
2322 */
2323
2324 /*
2325 * Load any hot spares for this pool.
2326 */
2327 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2328 if (error != 0 && error != ENOENT)
2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2330 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2331 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2332 if (load_nvlist(spa, spa->spa_spares.sav_object,
2333 &spa->spa_spares.sav_config) != 0)
2334 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2335
2336 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2337 spa_load_spares(spa);
2338 spa_config_exit(spa, SCL_ALL, FTAG);
2339 } else if (error == 0) {
2340 spa->spa_spares.sav_sync = B_TRUE;
2341 }
2342
2343 /*
2344 * Load any level 2 ARC devices for this pool.
2345 */
2346 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2347 &spa->spa_l2cache.sav_object);
2348 if (error != 0 && error != ENOENT)
2349 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2350 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2351 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2352 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2353 &spa->spa_l2cache.sav_config) != 0)
2354 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355
2356 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2357 spa_load_l2cache(spa);
2358 spa_config_exit(spa, SCL_ALL, FTAG);
2359 } else if (error == 0) {
2360 spa->spa_l2cache.sav_sync = B_TRUE;
2361 }
2362
2363 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2364
2365 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2366 if (error && error != ENOENT)
2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2368
2369 if (error == 0) {
2370 uint64_t autoreplace;
2371
2372 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2373 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2374 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2375 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2376 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2377 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2378 &spa->spa_dedup_ditto);
2379
2380 spa->spa_autoreplace = (autoreplace != 0);
2381 }
2382
2383 /*
2384 * If the 'autoreplace' property is set, then post a resource notifying
2385 * the ZFS DE that it should not issue any faults for unopenable
2386 * devices. We also iterate over the vdevs, and post a sysevent for any
2387 * unopenable vdevs so that the normal autoreplace handler can take
2388 * over.
2389 */
2390 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2391 spa_check_removed(spa->spa_root_vdev);
2392 /*
2393 * For the import case, this is done in spa_import(), because
2394 * at this point we're using the spare definitions from
2395 * the MOS config, not necessarily from the userland config.
2396 */
2397 if (state != SPA_LOAD_IMPORT) {
2398 spa_aux_check_removed(&spa->spa_spares);
2399 spa_aux_check_removed(&spa->spa_l2cache);
2400 }
2401 }
2402
2403 /*
2404 * Load the vdev state for all toplevel vdevs.
2405 */
2406 vdev_load(rvd);
2407
2408 /*
2409 * Propagate the leaf DTLs we just loaded all the way up the tree.
2410 */
2411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2412 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2413 spa_config_exit(spa, SCL_ALL, FTAG);
2414
2415 /*
2416 * Load the DDTs (dedup tables).
2417 */
2418 error = ddt_load(spa);
2419 if (error != 0)
2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421
2422 spa_update_dspace(spa);
2423
2424 /*
2425 * Validate the config, using the MOS config to fill in any
2426 * information which might be missing. If we fail to validate
2427 * the config then declare the pool unfit for use. If we're
2428 * assembling a pool from a split, the log is not transferred
2429 * over.
2430 */
2431 if (type != SPA_IMPORT_ASSEMBLE) {
2432 nvlist_t *nvconfig;
2433
2434 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2436
2437 if (!spa_config_valid(spa, nvconfig)) {
2438 nvlist_free(nvconfig);
2439 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2440 ENXIO));
2441 }
2442 nvlist_free(nvconfig);
2443
2444 /*
2445 * Now that we've validated the config, check the state of the
2446 * root vdev. If it can't be opened, it indicates one or
2447 * more toplevel vdevs are faulted.
2448 */
2449 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2450 return (ENXIO);
2451
2452 if (spa_check_logs(spa)) {
2453 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2454 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2455 }
2456 }
2457
2458 if (missing_feat_write) {
2459 ASSERT(state == SPA_LOAD_TRYIMPORT);
2460
2461 /*
2462 * At this point, we know that we can open the pool in
2463 * read-only mode but not read-write mode. We now have enough
2464 * information and can return to userland.
2465 */
2466 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2467 }
2468
2469 /*
2470 * We've successfully opened the pool, verify that we're ready
2471 * to start pushing transactions.
2472 */
2473 if (state != SPA_LOAD_TRYIMPORT) {
2474 if (error = spa_load_verify(spa))
2475 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2476 error));
2477 }
2478
2479 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2480 spa->spa_load_max_txg == UINT64_MAX)) {
2481 dmu_tx_t *tx;
2482 int need_update = B_FALSE;
2483
2484 ASSERT(state != SPA_LOAD_TRYIMPORT);
2485
2486 /*
2487 * Claim log blocks that haven't been committed yet.
2488 * This must all happen in a single txg.
2489 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2490 * invoked from zil_claim_log_block()'s i/o done callback.
2491 * Price of rollback is that we abandon the log.
2492 */
2493 spa->spa_claiming = B_TRUE;
2494
2495 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2496 spa_first_txg(spa));
2497 (void) dmu_objset_find(spa_name(spa),
2498 zil_claim, tx, DS_FIND_CHILDREN);
2499 dmu_tx_commit(tx);
2500
2501 spa->spa_claiming = B_FALSE;
2502
2503 spa_set_log_state(spa, SPA_LOG_GOOD);
2504 spa->spa_sync_on = B_TRUE;
2505 txg_sync_start(spa->spa_dsl_pool);
2506
2507 /*
2508 * Wait for all claims to sync. We sync up to the highest
2509 * claimed log block birth time so that claimed log blocks
2510 * don't appear to be from the future. spa_claim_max_txg
2511 * will have been set for us by either zil_check_log_chain()
2512 * (invoked from spa_check_logs()) or zil_claim() above.
2513 */
2514 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2515
2516 /*
2517 * If the config cache is stale, or we have uninitialized
2518 * metaslabs (see spa_vdev_add()), then update the config.
2519 *
2520 * If this is a verbatim import, trust the current
2521 * in-core spa_config and update the disk labels.
2522 */
2523 if (config_cache_txg != spa->spa_config_txg ||
2524 state == SPA_LOAD_IMPORT ||
2525 state == SPA_LOAD_RECOVER ||
2526 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2527 need_update = B_TRUE;
2528
2529 for (int c = 0; c < rvd->vdev_children; c++)
2530 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2531 need_update = B_TRUE;
2532
2533 /*
2534 * Update the config cache asychronously in case we're the
2535 * root pool, in which case the config cache isn't writable yet.
2536 */
2537 if (need_update)
2538 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2539
2540 /*
2541 * Check all DTLs to see if anything needs resilvering.
2542 */
2543 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2544 vdev_resilver_needed(rvd, NULL, NULL))
2545 spa_async_request(spa, SPA_ASYNC_RESILVER);
2546
2547 /*
2548 * Log the fact that we booted up (so that we can detect if
2549 * we rebooted in the middle of an operation).
2550 */
2551 spa_history_log_version(spa, "open");
2552
2553 /*
2554 * Delete any inconsistent datasets.
2555 */
2556 (void) dmu_objset_find(spa_name(spa),
2557 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2558
2559 /*
2560 * Clean up any stale temporary dataset userrefs.
2561 */
2562 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2563 }
2564
2565 return (0);
2566 }
2567
2568 static int
2569 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2570 {
2571 int mode = spa->spa_mode;
2572
2573 spa_unload(spa);
2574 spa_deactivate(spa);
2575
2576 spa->spa_load_max_txg--;
2577
2578 spa_activate(spa, mode);
2579 spa_async_suspend(spa);
2580
2581 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2582 }
2583
2584 /*
2585 * If spa_load() fails this function will try loading prior txg's. If
2586 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2587 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2588 * function will not rewind the pool and will return the same error as
2589 * spa_load().
2590 */
2591 static int
2592 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2593 uint64_t max_request, int rewind_flags)
2594 {
2595 nvlist_t *loadinfo = NULL;
2596 nvlist_t *config = NULL;
2597 int load_error, rewind_error;
2598 uint64_t safe_rewind_txg;
2599 uint64_t min_txg;
2600
2601 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2602 spa->spa_load_max_txg = spa->spa_load_txg;
2603 spa_set_log_state(spa, SPA_LOG_CLEAR);
2604 } else {
2605 spa->spa_load_max_txg = max_request;
2606 }
2607
2608 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2609 mosconfig);
2610 if (load_error == 0)
2611 return (0);
2612
2613 if (spa->spa_root_vdev != NULL)
2614 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2615
2616 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2617 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2618
2619 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2620 nvlist_free(config);
2621 return (load_error);
2622 }
2623
2624 if (state == SPA_LOAD_RECOVER) {
2625 /* Price of rolling back is discarding txgs, including log */
2626 spa_set_log_state(spa, SPA_LOG_CLEAR);
2627 } else {
2628 /*
2629 * If we aren't rolling back save the load info from our first
2630 * import attempt so that we can restore it after attempting
2631 * to rewind.
2632 */
2633 loadinfo = spa->spa_load_info;
2634 spa->spa_load_info = fnvlist_alloc();
2635 }
2636
2637 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2638 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2639 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2640 TXG_INITIAL : safe_rewind_txg;
2641
2642 /*
2643 * Continue as long as we're finding errors, we're still within
2644 * the acceptable rewind range, and we're still finding uberblocks
2645 */
2646 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2647 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2648 if (spa->spa_load_max_txg < safe_rewind_txg)
2649 spa->spa_extreme_rewind = B_TRUE;
2650 rewind_error = spa_load_retry(spa, state, mosconfig);
2651 }
2652
2653 spa->spa_extreme_rewind = B_FALSE;
2654 spa->spa_load_max_txg = UINT64_MAX;
2655
2656 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2657 spa_config_set(spa, config);
2658
2659 if (state == SPA_LOAD_RECOVER) {
2660 ASSERT3P(loadinfo, ==, NULL);
2661 return (rewind_error);
2662 } else {
2663 /* Store the rewind info as part of the initial load info */
2664 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2665 spa->spa_load_info);
2666
2667 /* Restore the initial load info */
2668 fnvlist_free(spa->spa_load_info);
2669 spa->spa_load_info = loadinfo;
2670
2671 return (load_error);
2672 }
2673 }
2674
2675 /*
2676 * Pool Open/Import
2677 *
2678 * The import case is identical to an open except that the configuration is sent
2679 * down from userland, instead of grabbed from the configuration cache. For the
2680 * case of an open, the pool configuration will exist in the
2681 * POOL_STATE_UNINITIALIZED state.
2682 *
2683 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2684 * the same time open the pool, without having to keep around the spa_t in some
2685 * ambiguous state.
2686 */
2687 static int
2688 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2689 nvlist_t **config)
2690 {
2691 spa_t *spa;
2692 spa_load_state_t state = SPA_LOAD_OPEN;
2693 int error;
2694 int locked = B_FALSE;
2695
2696 *spapp = NULL;
2697
2698 /*
2699 * As disgusting as this is, we need to support recursive calls to this
2700 * function because dsl_dir_open() is called during spa_load(), and ends
2701 * up calling spa_open() again. The real fix is to figure out how to
2702 * avoid dsl_dir_open() calling this in the first place.
2703 */
2704 if (mutex_owner(&spa_namespace_lock) != curthread) {
2705 mutex_enter(&spa_namespace_lock);
2706 locked = B_TRUE;
2707 }
2708
2709 if ((spa = spa_lookup(pool)) == NULL) {
2710 if (locked)
2711 mutex_exit(&spa_namespace_lock);
2712 return (ENOENT);
2713 }
2714
2715 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2716 zpool_rewind_policy_t policy;
2717
2718 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2719 &policy);
2720 if (policy.zrp_request & ZPOOL_DO_REWIND)
2721 state = SPA_LOAD_RECOVER;
2722
2723 spa_activate(spa, spa_mode_global);
2724
2725 if (state != SPA_LOAD_RECOVER)
2726 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2727
2728 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2729 policy.zrp_request);
2730
2731 if (error == EBADF) {
2732 /*
2733 * If vdev_validate() returns failure (indicated by
2734 * EBADF), it indicates that one of the vdevs indicates
2735 * that the pool has been exported or destroyed. If
2736 * this is the case, the config cache is out of sync and
2737 * we should remove the pool from the namespace.
2738 */
2739 spa_unload(spa);
2740 spa_deactivate(spa);
2741 spa_config_sync(spa, B_TRUE, B_TRUE);
2742 spa_remove(spa);
2743 if (locked)
2744 mutex_exit(&spa_namespace_lock);
2745 return (ENOENT);
2746 }
2747
2748 if (error) {
2749 /*
2750 * We can't open the pool, but we still have useful
2751 * information: the state of each vdev after the
2752 * attempted vdev_open(). Return this to the user.
2753 */
2754 if (config != NULL && spa->spa_config) {
2755 VERIFY(nvlist_dup(spa->spa_config, config,
2756 KM_SLEEP) == 0);
2757 VERIFY(nvlist_add_nvlist(*config,
2758 ZPOOL_CONFIG_LOAD_INFO,
2759 spa->spa_load_info) == 0);
2760 }
2761 spa_unload(spa);
2762 spa_deactivate(spa);
2763 spa->spa_last_open_failed = error;
2764 if (locked)
2765 mutex_exit(&spa_namespace_lock);
2766 *spapp = NULL;
2767 return (error);
2768 }
2769 }
2770
2771 spa_open_ref(spa, tag);
2772
2773 if (config != NULL)
2774 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2775
2776 /*
2777 * If we've recovered the pool, pass back any information we
2778 * gathered while doing the load.
2779 */
2780 if (state == SPA_LOAD_RECOVER) {
2781 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2782 spa->spa_load_info) == 0);
2783 }
2784
2785 if (locked) {
2786 spa->spa_last_open_failed = 0;
2787 spa->spa_last_ubsync_txg = 0;
2788 spa->spa_load_txg = 0;
2789 mutex_exit(&spa_namespace_lock);
2790 }
2791
2792 *spapp = spa;
2793
2794 return (0);
2795 }
2796
2797 int
2798 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2799 nvlist_t **config)
2800 {
2801 return (spa_open_common(name, spapp, tag, policy, config));
2802 }
2803
2804 int
2805 spa_open(const char *name, spa_t **spapp, void *tag)
2806 {
2807 return (spa_open_common(name, spapp, tag, NULL, NULL));
2808 }
2809
2810 /*
2811 * Lookup the given spa_t, incrementing the inject count in the process,
2812 * preventing it from being exported or destroyed.
2813 */
2814 spa_t *
2815 spa_inject_addref(char *name)
2816 {
2817 spa_t *spa;
2818
2819 mutex_enter(&spa_namespace_lock);
2820 if ((spa = spa_lookup(name)) == NULL) {
2821 mutex_exit(&spa_namespace_lock);
2822 return (NULL);
2823 }
2824 spa->spa_inject_ref++;
2825 mutex_exit(&spa_namespace_lock);
2826
2827 return (spa);
2828 }
2829
2830 void
2831 spa_inject_delref(spa_t *spa)
2832 {
2833 mutex_enter(&spa_namespace_lock);
2834 spa->spa_inject_ref--;
2835 mutex_exit(&spa_namespace_lock);
2836 }
2837
2838 /*
2839 * Add spares device information to the nvlist.
2840 */
2841 static void
2842 spa_add_spares(spa_t *spa, nvlist_t *config)
2843 {
2844 nvlist_t **spares;
2845 uint_t i, nspares;
2846 nvlist_t *nvroot;
2847 uint64_t guid;
2848 vdev_stat_t *vs;
2849 uint_t vsc;
2850 uint64_t pool;
2851
2852 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2853
2854 if (spa->spa_spares.sav_count == 0)
2855 return;
2856
2857 VERIFY(nvlist_lookup_nvlist(config,
2858 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2859 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2860 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2861 if (nspares != 0) {
2862 VERIFY(nvlist_add_nvlist_array(nvroot,
2863 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2864 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2865 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2866
2867 /*
2868 * Go through and find any spares which have since been
2869 * repurposed as an active spare. If this is the case, update
2870 * their status appropriately.
2871 */
2872 for (i = 0; i < nspares; i++) {
2873 VERIFY(nvlist_lookup_uint64(spares[i],
2874 ZPOOL_CONFIG_GUID, &guid) == 0);
2875 if (spa_spare_exists(guid, &pool, NULL) &&
2876 pool != 0ULL) {
2877 VERIFY(nvlist_lookup_uint64_array(
2878 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2879 (uint64_t **)&vs, &vsc) == 0);
2880 vs->vs_state = VDEV_STATE_CANT_OPEN;
2881 vs->vs_aux = VDEV_AUX_SPARED;
2882 }
2883 }
2884 }
2885 }
2886
2887 /*
2888 * Add l2cache device information to the nvlist, including vdev stats.
2889 */
2890 static void
2891 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2892 {
2893 nvlist_t **l2cache;
2894 uint_t i, j, nl2cache;
2895 nvlist_t *nvroot;
2896 uint64_t guid;
2897 vdev_t *vd;
2898 vdev_stat_t *vs;
2899 uint_t vsc;
2900
2901 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2902
2903 if (spa->spa_l2cache.sav_count == 0)
2904 return;
2905
2906 VERIFY(nvlist_lookup_nvlist(config,
2907 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2908 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2909 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2910 if (nl2cache != 0) {
2911 VERIFY(nvlist_add_nvlist_array(nvroot,
2912 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2913 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2914 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2915
2916 /*
2917 * Update level 2 cache device stats.
2918 */
2919
2920 for (i = 0; i < nl2cache; i++) {
2921 VERIFY(nvlist_lookup_uint64(l2cache[i],
2922 ZPOOL_CONFIG_GUID, &guid) == 0);
2923
2924 vd = NULL;
2925 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2926 if (guid ==
2927 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2928 vd = spa->spa_l2cache.sav_vdevs[j];
2929 break;
2930 }
2931 }
2932 ASSERT(vd != NULL);
2933
2934 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2935 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2936 == 0);
2937 vdev_get_stats(vd, vs);
2938 }
2939 }
2940 }
2941
2942 static void
2943 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2944 {
2945 nvlist_t *features;
2946 zap_cursor_t zc;
2947 zap_attribute_t za;
2948
2949 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2950 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2951
2952 if (spa->spa_feat_for_read_obj != 0) {
2953 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2954 spa->spa_feat_for_read_obj);
2955 zap_cursor_retrieve(&zc, &za) == 0;
2956 zap_cursor_advance(&zc)) {
2957 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2958 za.za_num_integers == 1);
2959 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2960 za.za_first_integer));
2961 }
2962 zap_cursor_fini(&zc);
2963 }
2964
2965 if (spa->spa_feat_for_write_obj != 0) {
2966 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2967 spa->spa_feat_for_write_obj);
2968 zap_cursor_retrieve(&zc, &za) == 0;
2969 zap_cursor_advance(&zc)) {
2970 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2971 za.za_num_integers == 1);
2972 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2973 za.za_first_integer));
2974 }
2975 zap_cursor_fini(&zc);
2976 }
2977
2978 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
2979 features) == 0);
2980 nvlist_free(features);
2981 }
2982
2983 int
2984 spa_get_stats(const char *name, nvlist_t **config,
2985 char *altroot, size_t buflen)
2986 {
2987 int error;
2988 spa_t *spa;
2989
2990 *config = NULL;
2991 error = spa_open_common(name, &spa, FTAG, NULL, config);
2992
2993 if (spa != NULL) {
2994 /*
2995 * This still leaves a window of inconsistency where the spares
2996 * or l2cache devices could change and the config would be
2997 * self-inconsistent.
2998 */
2999 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3000
3001 if (*config != NULL) {
3002 uint64_t loadtimes[2];
3003
3004 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3005 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3006 VERIFY(nvlist_add_uint64_array(*config,
3007 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3008
3009 VERIFY(nvlist_add_uint64(*config,
3010 ZPOOL_CONFIG_ERRCOUNT,
3011 spa_get_errlog_size(spa)) == 0);
3012
3013 if (spa_suspended(spa))
3014 VERIFY(nvlist_add_uint64(*config,
3015 ZPOOL_CONFIG_SUSPENDED,
3016 spa->spa_failmode) == 0);
3017
3018 spa_add_spares(spa, *config);
3019 spa_add_l2cache(spa, *config);
3020 spa_add_feature_stats(spa, *config);
3021 }
3022 }
3023
3024 /*
3025 * We want to get the alternate root even for faulted pools, so we cheat
3026 * and call spa_lookup() directly.
3027 */
3028 if (altroot) {
3029 if (spa == NULL) {
3030 mutex_enter(&spa_namespace_lock);
3031 spa = spa_lookup(name);
3032 if (spa)
3033 spa_altroot(spa, altroot, buflen);
3034 else
3035 altroot[0] = '\0';
3036 spa = NULL;
3037 mutex_exit(&spa_namespace_lock);
3038 } else {
3039 spa_altroot(spa, altroot, buflen);
3040 }
3041 }
3042
3043 if (spa != NULL) {
3044 spa_config_exit(spa, SCL_CONFIG, FTAG);
3045 spa_close(spa, FTAG);
3046 }
3047
3048 return (error);
3049 }
3050
3051 /*
3052 * Validate that the auxiliary device array is well formed. We must have an
3053 * array of nvlists, each which describes a valid leaf vdev. If this is an
3054 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3055 * specified, as long as they are well-formed.
3056 */
3057 static int
3058 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3059 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3060 vdev_labeltype_t label)
3061 {
3062 nvlist_t **dev;
3063 uint_t i, ndev;
3064 vdev_t *vd;
3065 int error;
3066
3067 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3068
3069 /*
3070 * It's acceptable to have no devs specified.
3071 */
3072 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3073 return (0);
3074
3075 if (ndev == 0)
3076 return (EINVAL);
3077
3078 /*
3079 * Make sure the pool is formatted with a version that supports this
3080 * device type.
3081 */
3082 if (spa_version(spa) < version)
3083 return (ENOTSUP);
3084
3085 /*
3086 * Set the pending device list so we correctly handle device in-use
3087 * checking.
3088 */
3089 sav->sav_pending = dev;
3090 sav->sav_npending = ndev;
3091
3092 for (i = 0; i < ndev; i++) {
3093 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3094 mode)) != 0)
3095 goto out;
3096
3097 if (!vd->vdev_ops->vdev_op_leaf) {
3098 vdev_free(vd);
3099 error = EINVAL;
3100 goto out;
3101 }
3102
3103 /*
3104 * The L2ARC currently only supports disk devices in
3105 * kernel context. For user-level testing, we allow it.
3106 */
3107 #ifdef _KERNEL
3108 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3109 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3110 error = ENOTBLK;
3111 vdev_free(vd);
3112 goto out;
3113 }
3114 #endif
3115 vd->vdev_top = vd;
3116
3117 if ((error = vdev_open(vd)) == 0 &&
3118 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3119 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3120 vd->vdev_guid) == 0);
3121 }
3122
3123 vdev_free(vd);
3124
3125 if (error &&
3126 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3127 goto out;
3128 else
3129 error = 0;
3130 }
3131
3132 out:
3133 sav->sav_pending = NULL;
3134 sav->sav_npending = 0;
3135 return (error);
3136 }
3137
3138 static int
3139 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3140 {
3141 int error;
3142
3143 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3144
3145 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3146 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3147 VDEV_LABEL_SPARE)) != 0) {
3148 return (error);
3149 }
3150
3151 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3152 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3153 VDEV_LABEL_L2CACHE));
3154 }
3155
3156 static void
3157 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3158 const char *config)
3159 {
3160 int i;
3161
3162 if (sav->sav_config != NULL) {
3163 nvlist_t **olddevs;
3164 uint_t oldndevs;
3165 nvlist_t **newdevs;
3166
3167 /*
3168 * Generate new dev list by concatentating with the
3169 * current dev list.
3170 */
3171 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3172 &olddevs, &oldndevs) == 0);
3173
3174 newdevs = kmem_alloc(sizeof (void *) *
3175 (ndevs + oldndevs), KM_SLEEP);
3176 for (i = 0; i < oldndevs; i++)
3177 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3178 KM_SLEEP) == 0);
3179 for (i = 0; i < ndevs; i++)
3180 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3181 KM_SLEEP) == 0);
3182
3183 VERIFY(nvlist_remove(sav->sav_config, config,
3184 DATA_TYPE_NVLIST_ARRAY) == 0);
3185
3186 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3187 config, newdevs, ndevs + oldndevs) == 0);
3188 for (i = 0; i < oldndevs + ndevs; i++)
3189 nvlist_free(newdevs[i]);
3190 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3191 } else {
3192 /*
3193 * Generate a new dev list.
3194 */
3195 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3196 KM_SLEEP) == 0);
3197 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3198 devs, ndevs) == 0);
3199 }
3200 }
3201
3202 /*
3203 * Stop and drop level 2 ARC devices
3204 */
3205 void
3206 spa_l2cache_drop(spa_t *spa)
3207 {
3208 vdev_t *vd;
3209 int i;
3210 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3211
3212 for (i = 0; i < sav->sav_count; i++) {
3213 uint64_t pool;
3214
3215 vd = sav->sav_vdevs[i];
3216 ASSERT(vd != NULL);
3217
3218 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3219 pool != 0ULL && l2arc_vdev_present(vd))
3220 l2arc_remove_vdev(vd);
3221 }
3222 }
3223
3224 /*
3225 * Pool Creation
3226 */
3227 int
3228 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3229 nvlist_t *zplprops)
3230 {
3231 spa_t *spa;
3232 char *altroot = NULL;
3233 vdev_t *rvd;
3234 dsl_pool_t *dp;
3235 dmu_tx_t *tx;
3236 int error = 0;
3237 uint64_t txg = TXG_INITIAL;
3238 nvlist_t **spares, **l2cache;
3239 uint_t nspares, nl2cache;
3240 uint64_t version, obj;
3241 boolean_t has_features;
3242
3243 /*
3244 * If this pool already exists, return failure.
3245 */
3246 mutex_enter(&spa_namespace_lock);
3247 if (spa_lookup(pool) != NULL) {
3248 mutex_exit(&spa_namespace_lock);
3249 return (EEXIST);
3250 }
3251
3252 /*
3253 * Allocate a new spa_t structure.
3254 */
3255 (void) nvlist_lookup_string(props,
3256 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3257 spa = spa_add(pool, NULL, altroot);
3258 spa_activate(spa, spa_mode_global);
3259
3260 if (props && (error = spa_prop_validate(spa, props))) {
3261 spa_deactivate(spa);
3262 spa_remove(spa);
3263 mutex_exit(&spa_namespace_lock);
3264 return (error);
3265 }
3266
3267 has_features = B_FALSE;
3268 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3269 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3270 if (zpool_prop_feature(nvpair_name(elem)))
3271 has_features = B_TRUE;
3272 }
3273
3274 if (has_features || nvlist_lookup_uint64(props,
3275 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3276 version = SPA_VERSION;
3277 }
3278 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3279
3280 spa->spa_first_txg = txg;
3281 spa->spa_uberblock.ub_txg = txg - 1;
3282 spa->spa_uberblock.ub_version = version;
3283 spa->spa_ubsync = spa->spa_uberblock;
3284
3285 /*
3286 * Create "The Godfather" zio to hold all async IOs
3287 */
3288 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3289 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3290
3291 /*
3292 * Create the root vdev.
3293 */
3294 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3295
3296 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3297
3298 ASSERT(error != 0 || rvd != NULL);
3299 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3300
3301 if (error == 0 && !zfs_allocatable_devs(nvroot))
3302 error = EINVAL;
3303
3304 if (error == 0 &&
3305 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3306 (error = spa_validate_aux(spa, nvroot, txg,
3307 VDEV_ALLOC_ADD)) == 0) {
3308 for (int c = 0; c < rvd->vdev_children; c++) {
3309 vdev_metaslab_set_size(rvd->vdev_child[c]);
3310 vdev_expand(rvd->vdev_child[c], txg);
3311 }
3312 }
3313
3314 spa_config_exit(spa, SCL_ALL, FTAG);
3315
3316 if (error != 0) {
3317 spa_unload(spa);
3318 spa_deactivate(spa);
3319 spa_remove(spa);
3320 mutex_exit(&spa_namespace_lock);
3321 return (error);
3322 }
3323
3324 /*
3325 * Get the list of spares, if specified.
3326 */
3327 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3328 &spares, &nspares) == 0) {
3329 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3330 KM_SLEEP) == 0);
3331 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3332 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3333 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3334 spa_load_spares(spa);
3335 spa_config_exit(spa, SCL_ALL, FTAG);
3336 spa->spa_spares.sav_sync = B_TRUE;
3337 }
3338
3339 /*
3340 * Get the list of level 2 cache devices, if specified.
3341 */
3342 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3343 &l2cache, &nl2cache) == 0) {
3344 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3345 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3346 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3347 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3348 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3349 spa_load_l2cache(spa);
3350 spa_config_exit(spa, SCL_ALL, FTAG);
3351 spa->spa_l2cache.sav_sync = B_TRUE;
3352 }
3353
3354 spa->spa_is_initializing = B_TRUE;
3355 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3356 spa->spa_meta_objset = dp->dp_meta_objset;
3357 spa->spa_is_initializing = B_FALSE;
3358
3359 /*
3360 * Create DDTs (dedup tables).
3361 */
3362 ddt_create(spa);
3363
3364 spa_update_dspace(spa);
3365
3366 tx = dmu_tx_create_assigned(dp, txg);
3367
3368 /*
3369 * Create the pool config object.
3370 */
3371 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3372 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3373 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3374
3375 if (zap_add(spa->spa_meta_objset,
3376 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3377 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3378 cmn_err(CE_PANIC, "failed to add pool config");
3379 }
3380
3381 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3382 spa_feature_create_zap_objects(spa, tx);
3383
3384 if (zap_add(spa->spa_meta_objset,
3385 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3386 sizeof (uint64_t), 1, &version, tx) != 0) {
3387 cmn_err(CE_PANIC, "failed to add pool version");
3388 }
3389
3390 /* Newly created pools with the right version are always deflated. */
3391 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3392 spa->spa_deflate = TRUE;
3393 if (zap_add(spa->spa_meta_objset,
3394 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3395 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3396 cmn_err(CE_PANIC, "failed to add deflate");
3397 }
3398 }
3399
3400 /*
3401 * Create the deferred-free bpobj. Turn off compression
3402 * because sync-to-convergence takes longer if the blocksize
3403 * keeps changing.
3404 */
3405 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3406 dmu_object_set_compress(spa->spa_meta_objset, obj,
3407 ZIO_COMPRESS_OFF, tx);
3408 if (zap_add(spa->spa_meta_objset,
3409 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3410 sizeof (uint64_t), 1, &obj, tx) != 0) {
3411 cmn_err(CE_PANIC, "failed to add bpobj");
3412 }
3413 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3414 spa->spa_meta_objset, obj));
3415
3416 /*
3417 * Create the pool's history object.
3418 */
3419 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3420 spa_history_create_obj(spa, tx);
3421
3422 /*
3423 * Set pool properties.
3424 */
3425 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3426 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3427 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3428 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3429
3430 if (props != NULL) {
3431 spa_configfile_set(spa, props, B_FALSE);
3432 spa_sync_props(spa, props, tx);
3433 }
3434
3435 dmu_tx_commit(tx);
3436
3437 spa->spa_sync_on = B_TRUE;
3438 txg_sync_start(spa->spa_dsl_pool);
3439
3440 /*
3441 * We explicitly wait for the first transaction to complete so that our
3442 * bean counters are appropriately updated.
3443 */
3444 txg_wait_synced(spa->spa_dsl_pool, txg);
3445
3446 spa_config_sync(spa, B_FALSE, B_TRUE);
3447
3448 spa_history_log_version(spa, "create");
3449
3450 spa->spa_minref = refcount_count(&spa->spa_refcount);
3451
3452 mutex_exit(&spa_namespace_lock);
3453
3454 return (0);
3455 }
3456
3457 #ifdef _KERNEL
3458 /*
3459 * Get the root pool information from the root disk, then import the root pool
3460 * during the system boot up time.
3461 */
3462 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3463
3464 static nvlist_t *
3465 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3466 {
3467 nvlist_t *config;
3468 nvlist_t *nvtop, *nvroot;
3469 uint64_t pgid;
3470
3471 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3472 return (NULL);
3473
3474 /*
3475 * Add this top-level vdev to the child array.
3476 */
3477 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3478 &nvtop) == 0);
3479 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3480 &pgid) == 0);
3481 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3482
3483 /*
3484 * Put this pool's top-level vdevs into a root vdev.
3485 */
3486 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3487 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3488 VDEV_TYPE_ROOT) == 0);
3489 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3490 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3491 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3492 &nvtop, 1) == 0);
3493
3494 /*
3495 * Replace the existing vdev_tree with the new root vdev in
3496 * this pool's configuration (remove the old, add the new).
3497 */
3498 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3499 nvlist_free(nvroot);
3500 return (config);
3501 }
3502
3503 /*
3504 * Walk the vdev tree and see if we can find a device with "better"
3505 * configuration. A configuration is "better" if the label on that
3506 * device has a more recent txg.
3507 */
3508 static void
3509 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3510 {
3511 for (int c = 0; c < vd->vdev_children; c++)
3512 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3513
3514 if (vd->vdev_ops->vdev_op_leaf) {
3515 nvlist_t *label;
3516 uint64_t label_txg;
3517
3518 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3519 &label) != 0)
3520 return;
3521
3522 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3523 &label_txg) == 0);
3524
3525 /*
3526 * Do we have a better boot device?
3527 */
3528 if (label_txg > *txg) {
3529 *txg = label_txg;
3530 *avd = vd;
3531 }
3532 nvlist_free(label);
3533 }
3534 }
3535
3536 /*
3537 * Import a root pool.
3538 *
3539 * For x86. devpath_list will consist of devid and/or physpath name of
3540 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3541 * The GRUB "findroot" command will return the vdev we should boot.
3542 *
3543 * For Sparc, devpath_list consists the physpath name of the booting device
3544 * no matter the rootpool is a single device pool or a mirrored pool.
3545 * e.g.
3546 * "/pci@1f,0/ide@d/disk@0,0:a"
3547 */
3548 int
3549 spa_import_rootpool(char *devpath, char *devid)
3550 {
3551 spa_t *spa;
3552 vdev_t *rvd, *bvd, *avd = NULL;
3553 nvlist_t *config, *nvtop;
3554 uint64_t guid, txg;
3555 char *pname;
3556 int error;
3557
3558 /*
3559 * Read the label from the boot device and generate a configuration.
3560 */
3561 config = spa_generate_rootconf(devpath, devid, &guid);
3562 #if defined(_OBP) && defined(_KERNEL)
3563 if (config == NULL) {
3564 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3565 /* iscsi boot */
3566 get_iscsi_bootpath_phy(devpath);
3567 config = spa_generate_rootconf(devpath, devid, &guid);
3568 }
3569 }
3570 #endif
3571 if (config == NULL) {
3572 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3573 devpath);
3574 return (EIO);
3575 }
3576
3577 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3578 &pname) == 0);
3579 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3580
3581 mutex_enter(&spa_namespace_lock);
3582 if ((spa = spa_lookup(pname)) != NULL) {
3583 /*
3584 * Remove the existing root pool from the namespace so that we
3585 * can replace it with the correct config we just read in.
3586 */
3587 spa_remove(spa);
3588 }
3589
3590 spa = spa_add(pname, config, NULL);
3591 spa->spa_is_root = B_TRUE;
3592 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3593
3594 /*
3595 * Build up a vdev tree based on the boot device's label config.
3596 */
3597 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3598 &nvtop) == 0);
3599 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3600 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3601 VDEV_ALLOC_ROOTPOOL);
3602 spa_config_exit(spa, SCL_ALL, FTAG);
3603 if (error) {
3604 mutex_exit(&spa_namespace_lock);
3605 nvlist_free(config);
3606 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3607 pname);
3608 return (error);
3609 }
3610
3611 /*
3612 * Get the boot vdev.
3613 */
3614 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3615 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3616 (u_longlong_t)guid);
3617 error = ENOENT;
3618 goto out;
3619 }
3620
3621 /*
3622 * Determine if there is a better boot device.
3623 */
3624 avd = bvd;
3625 spa_alt_rootvdev(rvd, &avd, &txg);
3626 if (avd != bvd) {
3627 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3628 "try booting from '%s'", avd->vdev_path);
3629 error = EINVAL;
3630 goto out;
3631 }
3632
3633 /*
3634 * If the boot device is part of a spare vdev then ensure that
3635 * we're booting off the active spare.
3636 */
3637 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3638 !bvd->vdev_isspare) {
3639 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3640 "try booting from '%s'",
3641 bvd->vdev_parent->
3642 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3643 error = EINVAL;
3644 goto out;
3645 }
3646
3647 error = 0;
3648 out:
3649 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3650 vdev_free(rvd);
3651 spa_config_exit(spa, SCL_ALL, FTAG);
3652 mutex_exit(&spa_namespace_lock);
3653
3654 nvlist_free(config);
3655 return (error);
3656 }
3657
3658 #endif
3659
3660 /*
3661 * Import a non-root pool into the system.
3662 */
3663 int
3664 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3665 {
3666 spa_t *spa;
3667 char *altroot = NULL;
3668 spa_load_state_t state = SPA_LOAD_IMPORT;
3669 zpool_rewind_policy_t policy;
3670 uint64_t mode = spa_mode_global;
3671 uint64_t readonly = B_FALSE;
3672 int error;
3673 nvlist_t *nvroot;
3674 nvlist_t **spares, **l2cache;
3675 uint_t nspares, nl2cache;
3676
3677 /*
3678 * If a pool with this name exists, return failure.
3679 */
3680 mutex_enter(&spa_namespace_lock);
3681 if (spa_lookup(pool) != NULL) {
3682 mutex_exit(&spa_namespace_lock);
3683 return (EEXIST);
3684 }
3685
3686 /*
3687 * Create and initialize the spa structure.
3688 */
3689 (void) nvlist_lookup_string(props,
3690 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3691 (void) nvlist_lookup_uint64(props,
3692 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3693 if (readonly)
3694 mode = FREAD;
3695 spa = spa_add(pool, config, altroot);
3696 spa->spa_import_flags = flags;
3697
3698 /*
3699 * Verbatim import - Take a pool and insert it into the namespace
3700 * as if it had been loaded at boot.
3701 */
3702 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3703 if (props != NULL)
3704 spa_configfile_set(spa, props, B_FALSE);
3705
3706 spa_config_sync(spa, B_FALSE, B_TRUE);
3707
3708 mutex_exit(&spa_namespace_lock);
3709 spa_history_log_version(spa, "import");
3710
3711 return (0);
3712 }
3713
3714 spa_activate(spa, mode);
3715
3716 /*
3717 * Don't start async tasks until we know everything is healthy.
3718 */
3719 spa_async_suspend(spa);
3720
3721 zpool_get_rewind_policy(config, &policy);
3722 if (policy.zrp_request & ZPOOL_DO_REWIND)
3723 state = SPA_LOAD_RECOVER;
3724
3725 /*
3726 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3727 * because the user-supplied config is actually the one to trust when
3728 * doing an import.
3729 */
3730 if (state != SPA_LOAD_RECOVER)
3731 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3732
3733 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3734 policy.zrp_request);
3735
3736 /*
3737 * Propagate anything learned while loading the pool and pass it
3738 * back to caller (i.e. rewind info, missing devices, etc).
3739 */
3740 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3741 spa->spa_load_info) == 0);
3742
3743 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3744 /*
3745 * Toss any existing sparelist, as it doesn't have any validity
3746 * anymore, and conflicts with spa_has_spare().
3747 */
3748 if (spa->spa_spares.sav_config) {
3749 nvlist_free(spa->spa_spares.sav_config);
3750 spa->spa_spares.sav_config = NULL;
3751 spa_load_spares(spa);
3752 }
3753 if (spa->spa_l2cache.sav_config) {
3754 nvlist_free(spa->spa_l2cache.sav_config);
3755 spa->spa_l2cache.sav_config = NULL;
3756 spa_load_l2cache(spa);
3757 }
3758
3759 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3760 &nvroot) == 0);
3761 if (error == 0)
3762 error = spa_validate_aux(spa, nvroot, -1ULL,
3763 VDEV_ALLOC_SPARE);
3764 if (error == 0)
3765 error = spa_validate_aux(spa, nvroot, -1ULL,
3766 VDEV_ALLOC_L2CACHE);
3767 spa_config_exit(spa, SCL_ALL, FTAG);
3768
3769 if (props != NULL)
3770 spa_configfile_set(spa, props, B_FALSE);
3771
3772 if (error != 0 || (props && spa_writeable(spa) &&
3773 (error = spa_prop_set(spa, props)))) {
3774 spa_unload(spa);
3775 spa_deactivate(spa);
3776 spa_remove(spa);
3777 mutex_exit(&spa_namespace_lock);
3778 return (error);
3779 }
3780
3781 spa_async_resume(spa);
3782
3783 /*
3784 * Override any spares and level 2 cache devices as specified by
3785 * the user, as these may have correct device names/devids, etc.
3786 */
3787 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3788 &spares, &nspares) == 0) {
3789 if (spa->spa_spares.sav_config)
3790 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3791 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3792 else
3793 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3794 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3795 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3796 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3797 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3798 spa_load_spares(spa);
3799 spa_config_exit(spa, SCL_ALL, FTAG);
3800 spa->spa_spares.sav_sync = B_TRUE;
3801 }
3802 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3803 &l2cache, &nl2cache) == 0) {
3804 if (spa->spa_l2cache.sav_config)
3805 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3806 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3807 else
3808 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3809 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3810 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3811 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3812 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3813 spa_load_l2cache(spa);
3814 spa_config_exit(spa, SCL_ALL, FTAG);
3815 spa->spa_l2cache.sav_sync = B_TRUE;
3816 }
3817
3818 /*
3819 * Check for any removed devices.
3820 */
3821 if (spa->spa_autoreplace) {
3822 spa_aux_check_removed(&spa->spa_spares);
3823 spa_aux_check_removed(&spa->spa_l2cache);
3824 }
3825
3826 if (spa_writeable(spa)) {
3827 /*
3828 * Update the config cache to include the newly-imported pool.
3829 */
3830 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3831 }
3832
3833 /*
3834 * It's possible that the pool was expanded while it was exported.
3835 * We kick off an async task to handle this for us.
3836 */
3837 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3838
3839 mutex_exit(&spa_namespace_lock);
3840 spa_history_log_version(spa, "import");
3841
3842 return (0);
3843 }
3844
3845 nvlist_t *
3846 spa_tryimport(nvlist_t *tryconfig)
3847 {
3848 nvlist_t *config = NULL;
3849 char *poolname;
3850 spa_t *spa;
3851 uint64_t state;
3852 int error;
3853
3854 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3855 return (NULL);
3856
3857 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3858 return (NULL);
3859
3860 /*
3861 * Create and initialize the spa structure.
3862 */
3863 mutex_enter(&spa_namespace_lock);
3864 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3865 spa_activate(spa, FREAD);
3866
3867 /*
3868 * Pass off the heavy lifting to spa_load().
3869 * Pass TRUE for mosconfig because the user-supplied config
3870 * is actually the one to trust when doing an import.
3871 */
3872 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3873
3874 /*
3875 * If 'tryconfig' was at least parsable, return the current config.
3876 */
3877 if (spa->spa_root_vdev != NULL) {
3878 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3879 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3880 poolname) == 0);
3881 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3882 state) == 0);
3883 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3884 spa->spa_uberblock.ub_timestamp) == 0);
3885 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3886 spa->spa_load_info) == 0);
3887
3888 /*
3889 * If the bootfs property exists on this pool then we
3890 * copy it out so that external consumers can tell which
3891 * pools are bootable.
3892 */
3893 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3894 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3895
3896 /*
3897 * We have to play games with the name since the
3898 * pool was opened as TRYIMPORT_NAME.
3899 */
3900 if (dsl_dsobj_to_dsname(spa_name(spa),
3901 spa->spa_bootfs, tmpname) == 0) {
3902 char *cp;
3903 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3904
3905 cp = strchr(tmpname, '/');
3906 if (cp == NULL) {
3907 (void) strlcpy(dsname, tmpname,
3908 MAXPATHLEN);
3909 } else {
3910 (void) snprintf(dsname, MAXPATHLEN,
3911 "%s/%s", poolname, ++cp);
3912 }
3913 VERIFY(nvlist_add_string(config,
3914 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3915 kmem_free(dsname, MAXPATHLEN);
3916 }
3917 kmem_free(tmpname, MAXPATHLEN);
3918 }
3919
3920 /*
3921 * Add the list of hot spares and level 2 cache devices.
3922 */
3923 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3924 spa_add_spares(spa, config);
3925 spa_add_l2cache(spa, config);
3926 spa_config_exit(spa, SCL_CONFIG, FTAG);
3927 }
3928
3929 spa_unload(spa);
3930 spa_deactivate(spa);
3931 spa_remove(spa);
3932 mutex_exit(&spa_namespace_lock);
3933
3934 return (config);
3935 }
3936
3937 /*
3938 * Pool export/destroy
3939 *
3940 * The act of destroying or exporting a pool is very simple. We make sure there
3941 * is no more pending I/O and any references to the pool are gone. Then, we
3942 * update the pool state and sync all the labels to disk, removing the
3943 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3944 * we don't sync the labels or remove the configuration cache.
3945 */
3946 static int
3947 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3948 boolean_t force, boolean_t hardforce)
3949 {
3950 spa_t *spa;
3951
3952 if (oldconfig)
3953 *oldconfig = NULL;
3954
3955 if (!(spa_mode_global & FWRITE))
3956 return (EROFS);
3957
3958 mutex_enter(&spa_namespace_lock);
3959 if ((spa = spa_lookup(pool)) == NULL) {
3960 mutex_exit(&spa_namespace_lock);
3961 return (ENOENT);
3962 }
3963
3964 /*
3965 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3966 * reacquire the namespace lock, and see if we can export.
3967 */
3968 spa_open_ref(spa, FTAG);
3969 mutex_exit(&spa_namespace_lock);
3970 spa_async_suspend(spa);
3971 mutex_enter(&spa_namespace_lock);
3972 spa_close(spa, FTAG);
3973
3974 /*
3975 * The pool will be in core if it's openable,
3976 * in which case we can modify its state.
3977 */
3978 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3979 /*
3980 * Objsets may be open only because they're dirty, so we
3981 * have to force it to sync before checking spa_refcnt.
3982 */
3983 txg_wait_synced(spa->spa_dsl_pool, 0);
3984
3985 /*
3986 * A pool cannot be exported or destroyed if there are active
3987 * references. If we are resetting a pool, allow references by
3988 * fault injection handlers.
3989 */
3990 if (!spa_refcount_zero(spa) ||
3991 (spa->spa_inject_ref != 0 &&
3992 new_state != POOL_STATE_UNINITIALIZED)) {
3993 spa_async_resume(spa);
3994 mutex_exit(&spa_namespace_lock);
3995 return (EBUSY);
3996 }
3997
3998 /*
3999 * A pool cannot be exported if it has an active shared spare.
4000 * This is to prevent other pools stealing the active spare
4001 * from an exported pool. At user's own will, such pool can
4002 * be forcedly exported.
4003 */
4004 if (!force && new_state == POOL_STATE_EXPORTED &&
4005 spa_has_active_shared_spare(spa)) {
4006 spa_async_resume(spa);
4007 mutex_exit(&spa_namespace_lock);
4008 return (EXDEV);
4009 }
4010
4011 /*
4012 * We want this to be reflected on every label,
4013 * so mark them all dirty. spa_unload() will do the
4014 * final sync that pushes these changes out.
4015 */
4016 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4017 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4018 spa->spa_state = new_state;
4019 spa->spa_final_txg = spa_last_synced_txg(spa) +
4020 TXG_DEFER_SIZE + 1;
4021 vdev_config_dirty(spa->spa_root_vdev);
4022 spa_config_exit(spa, SCL_ALL, FTAG);
4023 }
4024 }
4025
4026 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4027
4028 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4029 spa_unload(spa);
4030 spa_deactivate(spa);
4031 }
4032
4033 if (oldconfig && spa->spa_config)
4034 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4035
4036 if (new_state != POOL_STATE_UNINITIALIZED) {
4037 if (!hardforce)
4038 spa_config_sync(spa, B_TRUE, B_TRUE);
4039 spa_remove(spa);
4040 }
4041 mutex_exit(&spa_namespace_lock);
4042
4043 return (0);
4044 }
4045
4046 /*
4047 * Destroy a storage pool.
4048 */
4049 int
4050 spa_destroy(char *pool)
4051 {
4052 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4053 B_FALSE, B_FALSE));
4054 }
4055
4056 /*
4057 * Export a storage pool.
4058 */
4059 int
4060 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4061 boolean_t hardforce)
4062 {
4063 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4064 force, hardforce));
4065 }
4066
4067 /*
4068 * Similar to spa_export(), this unloads the spa_t without actually removing it
4069 * from the namespace in any way.
4070 */
4071 int
4072 spa_reset(char *pool)
4073 {
4074 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4075 B_FALSE, B_FALSE));
4076 }
4077
4078 /*
4079 * ==========================================================================
4080 * Device manipulation
4081 * ==========================================================================
4082 */
4083
4084 /*
4085 * Add a device to a storage pool.
4086 */
4087 int
4088 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4089 {
4090 uint64_t txg, id;
4091 int error;
4092 vdev_t *rvd = spa->spa_root_vdev;
4093 vdev_t *vd, *tvd;
4094 nvlist_t **spares, **l2cache;
4095 uint_t nspares, nl2cache;
4096
4097 ASSERT(spa_writeable(spa));
4098
4099 txg = spa_vdev_enter(spa);
4100
4101 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4102 VDEV_ALLOC_ADD)) != 0)
4103 return (spa_vdev_exit(spa, NULL, txg, error));
4104
4105 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4106
4107 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4108 &nspares) != 0)
4109 nspares = 0;
4110
4111 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4112 &nl2cache) != 0)
4113 nl2cache = 0;
4114
4115 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4116 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4117
4118 if (vd->vdev_children != 0 &&
4119 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4120 return (spa_vdev_exit(spa, vd, txg, error));
4121
4122 /*
4123 * We must validate the spares and l2cache devices after checking the
4124 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4125 */
4126 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4127 return (spa_vdev_exit(spa, vd, txg, error));
4128
4129 /*
4130 * Transfer each new top-level vdev from vd to rvd.
4131 */
4132 for (int c = 0; c < vd->vdev_children; c++) {
4133
4134 /*
4135 * Set the vdev id to the first hole, if one exists.
4136 */
4137 for (id = 0; id < rvd->vdev_children; id++) {
4138 if (rvd->vdev_child[id]->vdev_ishole) {
4139 vdev_free(rvd->vdev_child[id]);
4140 break;
4141 }
4142 }
4143 tvd = vd->vdev_child[c];
4144 vdev_remove_child(vd, tvd);
4145 tvd->vdev_id = id;
4146 vdev_add_child(rvd, tvd);
4147 vdev_config_dirty(tvd);
4148 }
4149
4150 if (nspares != 0) {
4151 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4152 ZPOOL_CONFIG_SPARES);
4153 spa_load_spares(spa);
4154 spa->spa_spares.sav_sync = B_TRUE;
4155 }
4156
4157 if (nl2cache != 0) {
4158 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4159 ZPOOL_CONFIG_L2CACHE);
4160 spa_load_l2cache(spa);
4161 spa->spa_l2cache.sav_sync = B_TRUE;
4162 }
4163
4164 /*
4165 * We have to be careful when adding new vdevs to an existing pool.
4166 * If other threads start allocating from these vdevs before we
4167 * sync the config cache, and we lose power, then upon reboot we may
4168 * fail to open the pool because there are DVAs that the config cache
4169 * can't translate. Therefore, we first add the vdevs without
4170 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4171 * and then let spa_config_update() initialize the new metaslabs.
4172 *
4173 * spa_load() checks for added-but-not-initialized vdevs, so that
4174 * if we lose power at any point in this sequence, the remaining
4175 * steps will be completed the next time we load the pool.
4176 */
4177 (void) spa_vdev_exit(spa, vd, txg, 0);
4178
4179 mutex_enter(&spa_namespace_lock);
4180 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4181 mutex_exit(&spa_namespace_lock);
4182
4183 return (0);
4184 }
4185
4186 /*
4187 * Attach a device to a mirror. The arguments are the path to any device
4188 * in the mirror, and the nvroot for the new device. If the path specifies
4189 * a device that is not mirrored, we automatically insert the mirror vdev.
4190 *
4191 * If 'replacing' is specified, the new device is intended to replace the
4192 * existing device; in this case the two devices are made into their own
4193 * mirror using the 'replacing' vdev, which is functionally identical to
4194 * the mirror vdev (it actually reuses all the same ops) but has a few
4195 * extra rules: you can't attach to it after it's been created, and upon
4196 * completion of resilvering, the first disk (the one being replaced)
4197 * is automatically detached.
4198 */
4199 int
4200 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4201 {
4202 uint64_t txg, dtl_max_txg;
4203 vdev_t *rvd = spa->spa_root_vdev;
4204 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4205 vdev_ops_t *pvops;
4206 char *oldvdpath, *newvdpath;
4207 int newvd_isspare;
4208 int error;
4209
4210 ASSERT(spa_writeable(spa));
4211
4212 txg = spa_vdev_enter(spa);
4213
4214 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4215
4216 if (oldvd == NULL)
4217 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4218
4219 if (!oldvd->vdev_ops->vdev_op_leaf)
4220 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4221
4222 pvd = oldvd->vdev_parent;
4223
4224 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4225 VDEV_ALLOC_ATTACH)) != 0)
4226 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4227
4228 if (newrootvd->vdev_children != 1)
4229 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4230
4231 newvd = newrootvd->vdev_child[0];
4232
4233 if (!newvd->vdev_ops->vdev_op_leaf)
4234 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4235
4236 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4237 return (spa_vdev_exit(spa, newrootvd, txg, error));
4238
4239 /*
4240 * Spares can't replace logs
4241 */
4242 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4243 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4244
4245 if (!replacing) {
4246 /*
4247 * For attach, the only allowable parent is a mirror or the root
4248 * vdev.
4249 */
4250 if (pvd->vdev_ops != &vdev_mirror_ops &&
4251 pvd->vdev_ops != &vdev_root_ops)
4252 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4253
4254 pvops = &vdev_mirror_ops;
4255 } else {
4256 /*
4257 * Active hot spares can only be replaced by inactive hot
4258 * spares.
4259 */
4260 if (pvd->vdev_ops == &vdev_spare_ops &&
4261 oldvd->vdev_isspare &&
4262 !spa_has_spare(spa, newvd->vdev_guid))
4263 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4264
4265 /*
4266 * If the source is a hot spare, and the parent isn't already a
4267 * spare, then we want to create a new hot spare. Otherwise, we
4268 * want to create a replacing vdev. The user is not allowed to
4269 * attach to a spared vdev child unless the 'isspare' state is
4270 * the same (spare replaces spare, non-spare replaces
4271 * non-spare).
4272 */
4273 if (pvd->vdev_ops == &vdev_replacing_ops &&
4274 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4275 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4276 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4277 newvd->vdev_isspare != oldvd->vdev_isspare) {
4278 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4279 }
4280
4281 if (newvd->vdev_isspare)
4282 pvops = &vdev_spare_ops;
4283 else
4284 pvops = &vdev_replacing_ops;
4285 }
4286
4287 /*
4288 * Make sure the new device is big enough.
4289 */
4290 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4291 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4292
4293 /*
4294 * The new device cannot have a higher alignment requirement
4295 * than the top-level vdev.
4296 */
4297 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4298 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4299
4300 /*
4301 * If this is an in-place replacement, update oldvd's path and devid
4302 * to make it distinguishable from newvd, and unopenable from now on.
4303 */
4304 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4305 spa_strfree(oldvd->vdev_path);
4306 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4307 KM_SLEEP);
4308 (void) sprintf(oldvd->vdev_path, "%s/%s",
4309 newvd->vdev_path, "old");
4310 if (oldvd->vdev_devid != NULL) {
4311 spa_strfree(oldvd->vdev_devid);
4312 oldvd->vdev_devid = NULL;
4313 }
4314 }
4315
4316 /* mark the device being resilvered */
4317 newvd->vdev_resilvering = B_TRUE;
4318
4319 /*
4320 * If the parent is not a mirror, or if we're replacing, insert the new
4321 * mirror/replacing/spare vdev above oldvd.
4322 */
4323 if (pvd->vdev_ops != pvops)
4324 pvd = vdev_add_parent(oldvd, pvops);
4325
4326 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4327 ASSERT(pvd->vdev_ops == pvops);
4328 ASSERT(oldvd->vdev_parent == pvd);
4329
4330 /*
4331 * Extract the new device from its root and add it to pvd.
4332 */
4333 vdev_remove_child(newrootvd, newvd);
4334 newvd->vdev_id = pvd->vdev_children;
4335 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4336 vdev_add_child(pvd, newvd);
4337
4338 tvd = newvd->vdev_top;
4339 ASSERT(pvd->vdev_top == tvd);
4340 ASSERT(tvd->vdev_parent == rvd);
4341
4342 vdev_config_dirty(tvd);
4343
4344 /*
4345 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4346 * for any dmu_sync-ed blocks. It will propagate upward when
4347 * spa_vdev_exit() calls vdev_dtl_reassess().
4348 */
4349 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4350
4351 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4352 dtl_max_txg - TXG_INITIAL);
4353
4354 if (newvd->vdev_isspare) {
4355 spa_spare_activate(newvd);
4356 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4357 }
4358
4359 oldvdpath = spa_strdup(oldvd->vdev_path);
4360 newvdpath = spa_strdup(newvd->vdev_path);
4361 newvd_isspare = newvd->vdev_isspare;
4362
4363 /*
4364 * Mark newvd's DTL dirty in this txg.
4365 */
4366 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4367
4368 /*
4369 * Restart the resilver
4370 */
4371 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4372
4373 /*
4374 * Commit the config
4375 */
4376 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4377
4378 spa_history_log_internal(spa, "vdev attach", NULL,
4379 "%s vdev=%s %s vdev=%s",
4380 replacing && newvd_isspare ? "spare in" :
4381 replacing ? "replace" : "attach", newvdpath,
4382 replacing ? "for" : "to", oldvdpath);
4383
4384 spa_strfree(oldvdpath);
4385 spa_strfree(newvdpath);
4386
4387 if (spa->spa_bootfs)
4388 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4389
4390 return (0);
4391 }
4392
4393 /*
4394 * Detach a device from a mirror or replacing vdev.
4395 * If 'replace_done' is specified, only detach if the parent
4396 * is a replacing vdev.
4397 */
4398 int
4399 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4400 {
4401 uint64_t txg;
4402 int error;
4403 vdev_t *rvd = spa->spa_root_vdev;
4404 vdev_t *vd, *pvd, *cvd, *tvd;
4405 boolean_t unspare = B_FALSE;
4406 uint64_t unspare_guid;
4407 char *vdpath;
4408
4409 ASSERT(spa_writeable(spa));
4410
4411 txg = spa_vdev_enter(spa);
4412
4413 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4414
4415 if (vd == NULL)
4416 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4417
4418 if (!vd->vdev_ops->vdev_op_leaf)
4419 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4420
4421 pvd = vd->vdev_parent;
4422
4423 /*
4424 * If the parent/child relationship is not as expected, don't do it.
4425 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4426 * vdev that's replacing B with C. The user's intent in replacing
4427 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4428 * the replace by detaching C, the expected behavior is to end up
4429 * M(A,B). But suppose that right after deciding to detach C,
4430 * the replacement of B completes. We would have M(A,C), and then
4431 * ask to detach C, which would leave us with just A -- not what
4432 * the user wanted. To prevent this, we make sure that the
4433 * parent/child relationship hasn't changed -- in this example,
4434 * that C's parent is still the replacing vdev R.
4435 */
4436 if (pvd->vdev_guid != pguid && pguid != 0)
4437 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4438
4439 /*
4440 * Only 'replacing' or 'spare' vdevs can be replaced.
4441 */
4442 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4443 pvd->vdev_ops != &vdev_spare_ops)
4444 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4445
4446 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4447 spa_version(spa) >= SPA_VERSION_SPARES);
4448
4449 /*
4450 * Only mirror, replacing, and spare vdevs support detach.
4451 */
4452 if (pvd->vdev_ops != &vdev_replacing_ops &&
4453 pvd->vdev_ops != &vdev_mirror_ops &&
4454 pvd->vdev_ops != &vdev_spare_ops)
4455 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4456
4457 /*
4458 * If this device has the only valid copy of some data,
4459 * we cannot safely detach it.
4460 */
4461 if (vdev_dtl_required(vd))
4462 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4463
4464 ASSERT(pvd->vdev_children >= 2);
4465
4466 /*
4467 * If we are detaching the second disk from a replacing vdev, then
4468 * check to see if we changed the original vdev's path to have "/old"
4469 * at the end in spa_vdev_attach(). If so, undo that change now.
4470 */
4471 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4472 vd->vdev_path != NULL) {
4473 size_t len = strlen(vd->vdev_path);
4474
4475 for (int c = 0; c < pvd->vdev_children; c++) {
4476 cvd = pvd->vdev_child[c];
4477
4478 if (cvd == vd || cvd->vdev_path == NULL)
4479 continue;
4480
4481 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4482 strcmp(cvd->vdev_path + len, "/old") == 0) {
4483 spa_strfree(cvd->vdev_path);
4484 cvd->vdev_path = spa_strdup(vd->vdev_path);
4485 break;
4486 }
4487 }
4488 }
4489
4490 /*
4491 * If we are detaching the original disk from a spare, then it implies
4492 * that the spare should become a real disk, and be removed from the
4493 * active spare list for the pool.
4494 */
4495 if (pvd->vdev_ops == &vdev_spare_ops &&
4496 vd->vdev_id == 0 &&
4497 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4498 unspare = B_TRUE;
4499
4500 /*
4501 * Erase the disk labels so the disk can be used for other things.
4502 * This must be done after all other error cases are handled,
4503 * but before we disembowel vd (so we can still do I/O to it).
4504 * But if we can't do it, don't treat the error as fatal --
4505 * it may be that the unwritability of the disk is the reason
4506 * it's being detached!
4507 */
4508 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4509
4510 /*
4511 * Remove vd from its parent and compact the parent's children.
4512 */
4513 vdev_remove_child(pvd, vd);
4514 vdev_compact_children(pvd);
4515
4516 /*
4517 * Remember one of the remaining children so we can get tvd below.
4518 */
4519 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4520
4521 /*
4522 * If we need to remove the remaining child from the list of hot spares,
4523 * do it now, marking the vdev as no longer a spare in the process.
4524 * We must do this before vdev_remove_parent(), because that can
4525 * change the GUID if it creates a new toplevel GUID. For a similar
4526 * reason, we must remove the spare now, in the same txg as the detach;
4527 * otherwise someone could attach a new sibling, change the GUID, and
4528 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4529 */
4530 if (unspare) {
4531 ASSERT(cvd->vdev_isspare);
4532 spa_spare_remove(cvd);
4533 unspare_guid = cvd->vdev_guid;
4534 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4535 cvd->vdev_unspare = B_TRUE;
4536 }
4537
4538 /*
4539 * If the parent mirror/replacing vdev only has one child,
4540 * the parent is no longer needed. Remove it from the tree.
4541 */
4542 if (pvd->vdev_children == 1) {
4543 if (pvd->vdev_ops == &vdev_spare_ops)
4544 cvd->vdev_unspare = B_FALSE;
4545 vdev_remove_parent(cvd);
4546 cvd->vdev_resilvering = B_FALSE;
4547 }
4548
4549
4550 /*
4551 * We don't set tvd until now because the parent we just removed
4552 * may have been the previous top-level vdev.
4553 */
4554 tvd = cvd->vdev_top;
4555 ASSERT(tvd->vdev_parent == rvd);
4556
4557 /*
4558 * Reevaluate the parent vdev state.
4559 */
4560 vdev_propagate_state(cvd);
4561
4562 /*
4563 * If the 'autoexpand' property is set on the pool then automatically
4564 * try to expand the size of the pool. For example if the device we
4565 * just detached was smaller than the others, it may be possible to
4566 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4567 * first so that we can obtain the updated sizes of the leaf vdevs.
4568 */
4569 if (spa->spa_autoexpand) {
4570 vdev_reopen(tvd);
4571 vdev_expand(tvd, txg);
4572 }
4573
4574 vdev_config_dirty(tvd);
4575
4576 /*
4577 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4578 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4579 * But first make sure we're not on any *other* txg's DTL list, to
4580 * prevent vd from being accessed after it's freed.
4581 */
4582 vdpath = spa_strdup(vd->vdev_path);
4583 for (int t = 0; t < TXG_SIZE; t++)
4584 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4585 vd->vdev_detached = B_TRUE;
4586 vdev_dirty(tvd, VDD_DTL, vd, txg);
4587
4588 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4589
4590 /* hang on to the spa before we release the lock */
4591 spa_open_ref(spa, FTAG);
4592
4593 error = spa_vdev_exit(spa, vd, txg, 0);
4594
4595 spa_history_log_internal(spa, "detach", NULL,
4596 "vdev=%s", vdpath);
4597 spa_strfree(vdpath);
4598
4599 /*
4600 * If this was the removal of the original device in a hot spare vdev,
4601 * then we want to go through and remove the device from the hot spare
4602 * list of every other pool.
4603 */
4604 if (unspare) {
4605 spa_t *altspa = NULL;
4606
4607 mutex_enter(&spa_namespace_lock);
4608 while ((altspa = spa_next(altspa)) != NULL) {
4609 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4610 altspa == spa)
4611 continue;
4612
4613 spa_open_ref(altspa, FTAG);
4614 mutex_exit(&spa_namespace_lock);
4615 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4616 mutex_enter(&spa_namespace_lock);
4617 spa_close(altspa, FTAG);
4618 }
4619 mutex_exit(&spa_namespace_lock);
4620
4621 /* search the rest of the vdevs for spares to remove */
4622 spa_vdev_resilver_done(spa);
4623 }
4624
4625 /* all done with the spa; OK to release */
4626 mutex_enter(&spa_namespace_lock);
4627 spa_close(spa, FTAG);
4628 mutex_exit(&spa_namespace_lock);
4629
4630 return (error);
4631 }
4632
4633 /*
4634 * Split a set of devices from their mirrors, and create a new pool from them.
4635 */
4636 int
4637 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4638 nvlist_t *props, boolean_t exp)
4639 {
4640 int error = 0;
4641 uint64_t txg, *glist;
4642 spa_t *newspa;
4643 uint_t c, children, lastlog;
4644 nvlist_t **child, *nvl, *tmp;
4645 dmu_tx_t *tx;
4646 char *altroot = NULL;
4647 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4648 boolean_t activate_slog;
4649
4650 ASSERT(spa_writeable(spa));
4651
4652 txg = spa_vdev_enter(spa);
4653
4654 /* clear the log and flush everything up to now */
4655 activate_slog = spa_passivate_log(spa);
4656 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4657 error = spa_offline_log(spa);
4658 txg = spa_vdev_config_enter(spa);
4659
4660 if (activate_slog)
4661 spa_activate_log(spa);
4662
4663 if (error != 0)
4664 return (spa_vdev_exit(spa, NULL, txg, error));
4665
4666 /* check new spa name before going any further */
4667 if (spa_lookup(newname) != NULL)
4668 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4669
4670 /*
4671 * scan through all the children to ensure they're all mirrors
4672 */
4673 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4674 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4675 &children) != 0)
4676 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4677
4678 /* first, check to ensure we've got the right child count */
4679 rvd = spa->spa_root_vdev;
4680 lastlog = 0;
4681 for (c = 0; c < rvd->vdev_children; c++) {
4682 vdev_t *vd = rvd->vdev_child[c];
4683
4684 /* don't count the holes & logs as children */
4685 if (vd->vdev_islog || vd->vdev_ishole) {
4686 if (lastlog == 0)
4687 lastlog = c;
4688 continue;
4689 }
4690
4691 lastlog = 0;
4692 }
4693 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4694 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4695
4696 /* next, ensure no spare or cache devices are part of the split */
4697 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4698 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4699 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4700
4701 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4702 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4703
4704 /* then, loop over each vdev and validate it */
4705 for (c = 0; c < children; c++) {
4706 uint64_t is_hole = 0;
4707
4708 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4709 &is_hole);
4710
4711 if (is_hole != 0) {
4712 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4713 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4714 continue;
4715 } else {
4716 error = EINVAL;
4717 break;
4718 }
4719 }
4720
4721 /* which disk is going to be split? */
4722 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4723 &glist[c]) != 0) {
4724 error = EINVAL;
4725 break;
4726 }
4727
4728 /* look it up in the spa */
4729 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4730 if (vml[c] == NULL) {
4731 error = ENODEV;
4732 break;
4733 }
4734
4735 /* make sure there's nothing stopping the split */
4736 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4737 vml[c]->vdev_islog ||
4738 vml[c]->vdev_ishole ||
4739 vml[c]->vdev_isspare ||
4740 vml[c]->vdev_isl2cache ||
4741 !vdev_writeable(vml[c]) ||
4742 vml[c]->vdev_children != 0 ||
4743 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4744 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4745 error = EINVAL;
4746 break;
4747 }
4748
4749 if (vdev_dtl_required(vml[c])) {
4750 error = EBUSY;
4751 break;
4752 }
4753
4754 /* we need certain info from the top level */
4755 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4756 vml[c]->vdev_top->vdev_ms_array) == 0);
4757 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4758 vml[c]->vdev_top->vdev_ms_shift) == 0);
4759 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4760 vml[c]->vdev_top->vdev_asize) == 0);
4761 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4762 vml[c]->vdev_top->vdev_ashift) == 0);
4763 }
4764
4765 if (error != 0) {
4766 kmem_free(vml, children * sizeof (vdev_t *));
4767 kmem_free(glist, children * sizeof (uint64_t));
4768 return (spa_vdev_exit(spa, NULL, txg, error));
4769 }
4770
4771 /* stop writers from using the disks */
4772 for (c = 0; c < children; c++) {
4773 if (vml[c] != NULL)
4774 vml[c]->vdev_offline = B_TRUE;
4775 }
4776 vdev_reopen(spa->spa_root_vdev);
4777
4778 /*
4779 * Temporarily record the splitting vdevs in the spa config. This
4780 * will disappear once the config is regenerated.
4781 */
4782 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4783 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4784 glist, children) == 0);
4785 kmem_free(glist, children * sizeof (uint64_t));
4786
4787 mutex_enter(&spa->spa_props_lock);
4788 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4789 nvl) == 0);
4790 mutex_exit(&spa->spa_props_lock);
4791 spa->spa_config_splitting = nvl;
4792 vdev_config_dirty(spa->spa_root_vdev);
4793
4794 /* configure and create the new pool */
4795 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4796 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4797 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4798 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4799 spa_version(spa)) == 0);
4800 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4801 spa->spa_config_txg) == 0);
4802 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4803 spa_generate_guid(NULL)) == 0);
4804 (void) nvlist_lookup_string(props,
4805 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4806
4807 /* add the new pool to the namespace */
4808 newspa = spa_add(newname, config, altroot);
4809 newspa->spa_config_txg = spa->spa_config_txg;
4810 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4811
4812 /* release the spa config lock, retaining the namespace lock */
4813 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4814
4815 if (zio_injection_enabled)
4816 zio_handle_panic_injection(spa, FTAG, 1);
4817
4818 spa_activate(newspa, spa_mode_global);
4819 spa_async_suspend(newspa);
4820
4821 /* create the new pool from the disks of the original pool */
4822 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4823 if (error)
4824 goto out;
4825
4826 /* if that worked, generate a real config for the new pool */
4827 if (newspa->spa_root_vdev != NULL) {
4828 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4829 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4830 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4831 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4832 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4833 B_TRUE));
4834 }
4835
4836 /* set the props */
4837 if (props != NULL) {
4838 spa_configfile_set(newspa, props, B_FALSE);
4839 error = spa_prop_set(newspa, props);
4840 if (error)
4841 goto out;
4842 }
4843
4844 /* flush everything */
4845 txg = spa_vdev_config_enter(newspa);
4846 vdev_config_dirty(newspa->spa_root_vdev);
4847 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4848
4849 if (zio_injection_enabled)
4850 zio_handle_panic_injection(spa, FTAG, 2);
4851
4852 spa_async_resume(newspa);
4853
4854 /* finally, update the original pool's config */
4855 txg = spa_vdev_config_enter(spa);
4856 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4857 error = dmu_tx_assign(tx, TXG_WAIT);
4858 if (error != 0)
4859 dmu_tx_abort(tx);
4860 for (c = 0; c < children; c++) {
4861 if (vml[c] != NULL) {
4862 vdev_split(vml[c]);
4863 if (error == 0)
4864 spa_history_log_internal(spa, "detach", tx,
4865 "vdev=%s", vml[c]->vdev_path);
4866 vdev_free(vml[c]);
4867 }
4868 }
4869 vdev_config_dirty(spa->spa_root_vdev);
4870 spa->spa_config_splitting = NULL;
4871 nvlist_free(nvl);
4872 if (error == 0)
4873 dmu_tx_commit(tx);
4874 (void) spa_vdev_exit(spa, NULL, txg, 0);
4875
4876 if (zio_injection_enabled)
4877 zio_handle_panic_injection(spa, FTAG, 3);
4878
4879 /* split is complete; log a history record */
4880 spa_history_log_internal(newspa, "split", NULL,
4881 "from pool %s", spa_name(spa));
4882
4883 kmem_free(vml, children * sizeof (vdev_t *));
4884
4885 /* if we're not going to mount the filesystems in userland, export */
4886 if (exp)
4887 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4888 B_FALSE, B_FALSE);
4889
4890 return (error);
4891
4892 out:
4893 spa_unload(newspa);
4894 spa_deactivate(newspa);
4895 spa_remove(newspa);
4896
4897 txg = spa_vdev_config_enter(spa);
4898
4899 /* re-online all offlined disks */
4900 for (c = 0; c < children; c++) {
4901 if (vml[c] != NULL)
4902 vml[c]->vdev_offline = B_FALSE;
4903 }
4904 vdev_reopen(spa->spa_root_vdev);
4905
4906 nvlist_free(spa->spa_config_splitting);
4907 spa->spa_config_splitting = NULL;
4908 (void) spa_vdev_exit(spa, NULL, txg, error);
4909
4910 kmem_free(vml, children * sizeof (vdev_t *));
4911 return (error);
4912 }
4913
4914 static nvlist_t *
4915 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4916 {
4917 for (int i = 0; i < count; i++) {
4918 uint64_t guid;
4919
4920 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4921 &guid) == 0);
4922
4923 if (guid == target_guid)
4924 return (nvpp[i]);
4925 }
4926
4927 return (NULL);
4928 }
4929
4930 static void
4931 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4932 nvlist_t *dev_to_remove)
4933 {
4934 nvlist_t **newdev = NULL;
4935
4936 if (count > 1)
4937 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4938
4939 for (int i = 0, j = 0; i < count; i++) {
4940 if (dev[i] == dev_to_remove)
4941 continue;
4942 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4943 }
4944
4945 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4946 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4947
4948 for (int i = 0; i < count - 1; i++)
4949 nvlist_free(newdev[i]);
4950
4951 if (count > 1)
4952 kmem_free(newdev, (count - 1) * sizeof (void *));
4953 }
4954
4955 /*
4956 * Evacuate the device.
4957 */
4958 static int
4959 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4960 {
4961 uint64_t txg;
4962 int error = 0;
4963
4964 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4965 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4966 ASSERT(vd == vd->vdev_top);
4967
4968 /*
4969 * Evacuate the device. We don't hold the config lock as writer
4970 * since we need to do I/O but we do keep the
4971 * spa_namespace_lock held. Once this completes the device
4972 * should no longer have any blocks allocated on it.
4973 */
4974 if (vd->vdev_islog) {
4975 if (vd->vdev_stat.vs_alloc != 0)
4976 error = spa_offline_log(spa);
4977 } else {
4978 error = ENOTSUP;
4979 }
4980
4981 if (error)
4982 return (error);
4983
4984 /*
4985 * The evacuation succeeded. Remove any remaining MOS metadata
4986 * associated with this vdev, and wait for these changes to sync.
4987 */
4988 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4989 txg = spa_vdev_config_enter(spa);
4990 vd->vdev_removing = B_TRUE;
4991 vdev_dirty(vd, 0, NULL, txg);
4992 vdev_config_dirty(vd);
4993 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4994
4995 return (0);
4996 }
4997
4998 /*
4999 * Complete the removal by cleaning up the namespace.
5000 */
5001 static void
5002 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5003 {
5004 vdev_t *rvd = spa->spa_root_vdev;
5005 uint64_t id = vd->vdev_id;
5006 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5007
5008 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5009 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5010 ASSERT(vd == vd->vdev_top);
5011
5012 /*
5013 * Only remove any devices which are empty.
5014 */
5015 if (vd->vdev_stat.vs_alloc != 0)
5016 return;
5017
5018 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5019
5020 if (list_link_active(&vd->vdev_state_dirty_node))
5021 vdev_state_clean(vd);
5022 if (list_link_active(&vd->vdev_config_dirty_node))
5023 vdev_config_clean(vd);
5024
5025 vdev_free(vd);
5026
5027 if (last_vdev) {
5028 vdev_compact_children(rvd);
5029 } else {
5030 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5031 vdev_add_child(rvd, vd);
5032 }
5033 vdev_config_dirty(rvd);
5034
5035 /*
5036 * Reassess the health of our root vdev.
5037 */
5038 vdev_reopen(rvd);
5039 }
5040
5041 /*
5042 * Remove a device from the pool -
5043 *
5044 * Removing a device from the vdev namespace requires several steps
5045 * and can take a significant amount of time. As a result we use
5046 * the spa_vdev_config_[enter/exit] functions which allow us to
5047 * grab and release the spa_config_lock while still holding the namespace
5048 * lock. During each step the configuration is synced out.
5049 */
5050
5051 /*
5052 * Remove a device from the pool. Currently, this supports removing only hot
5053 * spares, slogs, and level 2 ARC devices.
5054 */
5055 int
5056 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5057 {
5058 vdev_t *vd;
5059 metaslab_group_t *mg;
5060 nvlist_t **spares, **l2cache, *nv;
5061 uint64_t txg = 0;
5062 uint_t nspares, nl2cache;
5063 int error = 0;
5064 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5065
5066 ASSERT(spa_writeable(spa));
5067
5068 if (!locked)
5069 txg = spa_vdev_enter(spa);
5070
5071 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5072
5073 if (spa->spa_spares.sav_vdevs != NULL &&
5074 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5075 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5076 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5077 /*
5078 * Only remove the hot spare if it's not currently in use
5079 * in this pool.
5080 */
5081 if (vd == NULL || unspare) {
5082 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5083 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5084 spa_load_spares(spa);
5085 spa->spa_spares.sav_sync = B_TRUE;
5086 } else {
5087 error = EBUSY;
5088 }
5089 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5090 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5091 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5092 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5093 /*
5094 * Cache devices can always be removed.
5095 */
5096 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5097 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5098 spa_load_l2cache(spa);
5099 spa->spa_l2cache.sav_sync = B_TRUE;
5100 } else if (vd != NULL && vd->vdev_islog) {
5101 ASSERT(!locked);
5102 ASSERT(vd == vd->vdev_top);
5103
5104 /*
5105 * XXX - Once we have bp-rewrite this should
5106 * become the common case.
5107 */
5108
5109 mg = vd->vdev_mg;
5110
5111 /*
5112 * Stop allocating from this vdev.
5113 */
5114 metaslab_group_passivate(mg);
5115
5116 /*
5117 * Wait for the youngest allocations and frees to sync,
5118 * and then wait for the deferral of those frees to finish.
5119 */
5120 spa_vdev_config_exit(spa, NULL,
5121 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5122
5123 /*
5124 * Attempt to evacuate the vdev.
5125 */
5126 error = spa_vdev_remove_evacuate(spa, vd);
5127
5128 txg = spa_vdev_config_enter(spa);
5129
5130 /*
5131 * If we couldn't evacuate the vdev, unwind.
5132 */
5133 if (error) {
5134 metaslab_group_activate(mg);
5135 return (spa_vdev_exit(spa, NULL, txg, error));
5136 }
5137
5138 /*
5139 * Clean up the vdev namespace.
5140 */
5141 spa_vdev_remove_from_namespace(spa, vd);
5142
5143 } else if (vd != NULL) {
5144 /*
5145 * Normal vdevs cannot be removed (yet).
5146 */
5147 error = ENOTSUP;
5148 } else {
5149 /*
5150 * There is no vdev of any kind with the specified guid.
5151 */
5152 error = ENOENT;
5153 }
5154
5155 if (!locked)
5156 return (spa_vdev_exit(spa, NULL, txg, error));
5157
5158 return (error);
5159 }
5160
5161 /*
5162 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5163 * current spared, so we can detach it.
5164 */
5165 static vdev_t *
5166 spa_vdev_resilver_done_hunt(vdev_t *vd)
5167 {
5168 vdev_t *newvd, *oldvd;
5169
5170 for (int c = 0; c < vd->vdev_children; c++) {
5171 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5172 if (oldvd != NULL)
5173 return (oldvd);
5174 }
5175
5176 /*
5177 * Check for a completed replacement. We always consider the first
5178 * vdev in the list to be the oldest vdev, and the last one to be
5179 * the newest (see spa_vdev_attach() for how that works). In
5180 * the case where the newest vdev is faulted, we will not automatically
5181 * remove it after a resilver completes. This is OK as it will require
5182 * user intervention to determine which disk the admin wishes to keep.
5183 */
5184 if (vd->vdev_ops == &vdev_replacing_ops) {
5185 ASSERT(vd->vdev_children > 1);
5186
5187 newvd = vd->vdev_child[vd->vdev_children - 1];
5188 oldvd = vd->vdev_child[0];
5189
5190 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5191 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5192 !vdev_dtl_required(oldvd))
5193 return (oldvd);
5194 }
5195
5196 /*
5197 * Check for a completed resilver with the 'unspare' flag set.
5198 */
5199 if (vd->vdev_ops == &vdev_spare_ops) {
5200 vdev_t *first = vd->vdev_child[0];
5201 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5202
5203 if (last->vdev_unspare) {
5204 oldvd = first;
5205 newvd = last;
5206 } else if (first->vdev_unspare) {
5207 oldvd = last;
5208 newvd = first;
5209 } else {
5210 oldvd = NULL;
5211 }
5212
5213 if (oldvd != NULL &&
5214 vdev_dtl_empty(newvd, DTL_MISSING) &&
5215 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5216 !vdev_dtl_required(oldvd))
5217 return (oldvd);
5218
5219 /*
5220 * If there are more than two spares attached to a disk,
5221 * and those spares are not required, then we want to
5222 * attempt to free them up now so that they can be used
5223 * by other pools. Once we're back down to a single
5224 * disk+spare, we stop removing them.
5225 */
5226 if (vd->vdev_children > 2) {
5227 newvd = vd->vdev_child[1];
5228
5229 if (newvd->vdev_isspare && last->vdev_isspare &&
5230 vdev_dtl_empty(last, DTL_MISSING) &&
5231 vdev_dtl_empty(last, DTL_OUTAGE) &&
5232 !vdev_dtl_required(newvd))
5233 return (newvd);
5234 }
5235 }
5236
5237 return (NULL);
5238 }
5239
5240 static void
5241 spa_vdev_resilver_done(spa_t *spa)
5242 {
5243 vdev_t *vd, *pvd, *ppvd;
5244 uint64_t guid, sguid, pguid, ppguid;
5245
5246 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5247
5248 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5249 pvd = vd->vdev_parent;
5250 ppvd = pvd->vdev_parent;
5251 guid = vd->vdev_guid;
5252 pguid = pvd->vdev_guid;
5253 ppguid = ppvd->vdev_guid;
5254 sguid = 0;
5255 /*
5256 * If we have just finished replacing a hot spared device, then
5257 * we need to detach the parent's first child (the original hot
5258 * spare) as well.
5259 */
5260 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5261 ppvd->vdev_children == 2) {
5262 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5263 sguid = ppvd->vdev_child[1]->vdev_guid;
5264 }
5265 spa_config_exit(spa, SCL_ALL, FTAG);
5266 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5267 return;
5268 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5269 return;
5270 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5271 }
5272
5273 spa_config_exit(spa, SCL_ALL, FTAG);
5274 }
5275
5276 /*
5277 * Update the stored path or FRU for this vdev.
5278 */
5279 int
5280 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5281 boolean_t ispath)
5282 {
5283 vdev_t *vd;
5284 boolean_t sync = B_FALSE;
5285
5286 ASSERT(spa_writeable(spa));
5287
5288 spa_vdev_state_enter(spa, SCL_ALL);
5289
5290 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5291 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5292
5293 if (!vd->vdev_ops->vdev_op_leaf)
5294 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5295
5296 if (ispath) {
5297 if (strcmp(value, vd->vdev_path) != 0) {
5298 spa_strfree(vd->vdev_path);
5299 vd->vdev_path = spa_strdup(value);
5300 sync = B_TRUE;
5301 }
5302 } else {
5303 if (vd->vdev_fru == NULL) {
5304 vd->vdev_fru = spa_strdup(value);
5305 sync = B_TRUE;
5306 } else if (strcmp(value, vd->vdev_fru) != 0) {
5307 spa_strfree(vd->vdev_fru);
5308 vd->vdev_fru = spa_strdup(value);
5309 sync = B_TRUE;
5310 }
5311 }
5312
5313 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5314 }
5315
5316 int
5317 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5318 {
5319 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5320 }
5321
5322 int
5323 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5324 {
5325 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5326 }
5327
5328 /*
5329 * ==========================================================================
5330 * SPA Scanning
5331 * ==========================================================================
5332 */
5333
5334 int
5335 spa_scan_stop(spa_t *spa)
5336 {
5337 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5338 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5339 return (EBUSY);
5340 return (dsl_scan_cancel(spa->spa_dsl_pool));
5341 }
5342
5343 int
5344 spa_scan(spa_t *spa, pool_scan_func_t func)
5345 {
5346 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5347
5348 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5349 return (ENOTSUP);
5350
5351 /*
5352 * If a resilver was requested, but there is no DTL on a
5353 * writeable leaf device, we have nothing to do.
5354 */
5355 if (func == POOL_SCAN_RESILVER &&
5356 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5357 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5358 return (0);
5359 }
5360
5361 return (dsl_scan(spa->spa_dsl_pool, func));
5362 }
5363
5364 /*
5365 * ==========================================================================
5366 * SPA async task processing
5367 * ==========================================================================
5368 */
5369
5370 static void
5371 spa_async_remove(spa_t *spa, vdev_t *vd)
5372 {
5373 if (vd->vdev_remove_wanted) {
5374 vd->vdev_remove_wanted = B_FALSE;
5375 vd->vdev_delayed_close = B_FALSE;
5376 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5377
5378 /*
5379 * We want to clear the stats, but we don't want to do a full
5380 * vdev_clear() as that will cause us to throw away
5381 * degraded/faulted state as well as attempt to reopen the
5382 * device, all of which is a waste.
5383 */
5384 vd->vdev_stat.vs_read_errors = 0;
5385 vd->vdev_stat.vs_write_errors = 0;
5386 vd->vdev_stat.vs_checksum_errors = 0;
5387
5388 vdev_state_dirty(vd->vdev_top);
5389 }
5390
5391 for (int c = 0; c < vd->vdev_children; c++)
5392 spa_async_remove(spa, vd->vdev_child[c]);
5393 }
5394
5395 static void
5396 spa_async_probe(spa_t *spa, vdev_t *vd)
5397 {
5398 if (vd->vdev_probe_wanted) {
5399 vd->vdev_probe_wanted = B_FALSE;
5400 vdev_reopen(vd); /* vdev_open() does the actual probe */
5401 }
5402
5403 for (int c = 0; c < vd->vdev_children; c++)
5404 spa_async_probe(spa, vd->vdev_child[c]);
5405 }
5406
5407 static void
5408 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5409 {
5410 sysevent_id_t eid;
5411 nvlist_t *attr;
5412 char *physpath;
5413
5414 if (!spa->spa_autoexpand)
5415 return;
5416
5417 for (int c = 0; c < vd->vdev_children; c++) {
5418 vdev_t *cvd = vd->vdev_child[c];
5419 spa_async_autoexpand(spa, cvd);
5420 }
5421
5422 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5423 return;
5424
5425 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5426 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5427
5428 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5429 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5430
5431 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5432 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5433
5434 nvlist_free(attr);
5435 kmem_free(physpath, MAXPATHLEN);
5436 }
5437
5438 static void
5439 spa_async_thread(spa_t *spa)
5440 {
5441 int tasks;
5442
5443 ASSERT(spa->spa_sync_on);
5444
5445 mutex_enter(&spa->spa_async_lock);
5446 tasks = spa->spa_async_tasks;
5447 spa->spa_async_tasks = 0;
5448 mutex_exit(&spa->spa_async_lock);
5449
5450 /*
5451 * See if the config needs to be updated.
5452 */
5453 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5454 uint64_t old_space, new_space;
5455
5456 mutex_enter(&spa_namespace_lock);
5457 old_space = metaslab_class_get_space(spa_normal_class(spa));
5458 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5459 new_space = metaslab_class_get_space(spa_normal_class(spa));
5460 mutex_exit(&spa_namespace_lock);
5461
5462 /*
5463 * If the pool grew as a result of the config update,
5464 * then log an internal history event.
5465 */
5466 if (new_space != old_space) {
5467 spa_history_log_internal(spa, "vdev online", NULL,
5468 "pool '%s' size: %llu(+%llu)",
5469 spa_name(spa), new_space, new_space - old_space);
5470 }
5471 }
5472
5473 /*
5474 * See if any devices need to be marked REMOVED.
5475 */
5476 if (tasks & SPA_ASYNC_REMOVE) {
5477 spa_vdev_state_enter(spa, SCL_NONE);
5478 spa_async_remove(spa, spa->spa_root_vdev);
5479 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5480 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5481 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5482 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5483 (void) spa_vdev_state_exit(spa, NULL, 0);
5484 }
5485
5486 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5487 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5488 spa_async_autoexpand(spa, spa->spa_root_vdev);
5489 spa_config_exit(spa, SCL_CONFIG, FTAG);
5490 }
5491
5492 /*
5493 * See if any devices need to be probed.
5494 */
5495 if (tasks & SPA_ASYNC_PROBE) {
5496 spa_vdev_state_enter(spa, SCL_NONE);
5497 spa_async_probe(spa, spa->spa_root_vdev);
5498 (void) spa_vdev_state_exit(spa, NULL, 0);
5499 }
5500
5501 /*
5502 * If any devices are done replacing, detach them.
5503 */
5504 if (tasks & SPA_ASYNC_RESILVER_DONE)
5505 spa_vdev_resilver_done(spa);
5506
5507 /*
5508 * Kick off a resilver.
5509 */
5510 if (tasks & SPA_ASYNC_RESILVER)
5511 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5512
5513 /*
5514 * Let the world know that we're done.
5515 */
5516 mutex_enter(&spa->spa_async_lock);
5517 spa->spa_async_thread = NULL;
5518 cv_broadcast(&spa->spa_async_cv);
5519 mutex_exit(&spa->spa_async_lock);
5520 thread_exit();
5521 }
5522
5523 void
5524 spa_async_suspend(spa_t *spa)
5525 {
5526 mutex_enter(&spa->spa_async_lock);
5527 spa->spa_async_suspended++;
5528 while (spa->spa_async_thread != NULL)
5529 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5530 mutex_exit(&spa->spa_async_lock);
5531 }
5532
5533 void
5534 spa_async_resume(spa_t *spa)
5535 {
5536 mutex_enter(&spa->spa_async_lock);
5537 ASSERT(spa->spa_async_suspended != 0);
5538 spa->spa_async_suspended--;
5539 mutex_exit(&spa->spa_async_lock);
5540 }
5541
5542 static void
5543 spa_async_dispatch(spa_t *spa)
5544 {
5545 mutex_enter(&spa->spa_async_lock);
5546 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5547 spa->spa_async_thread == NULL &&
5548 rootdir != NULL && !vn_is_readonly(rootdir))
5549 spa->spa_async_thread = thread_create(NULL, 0,
5550 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5551 mutex_exit(&spa->spa_async_lock);
5552 }
5553
5554 void
5555 spa_async_request(spa_t *spa, int task)
5556 {
5557 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5558 mutex_enter(&spa->spa_async_lock);
5559 spa->spa_async_tasks |= task;
5560 mutex_exit(&spa->spa_async_lock);
5561 }
5562
5563 /*
5564 * ==========================================================================
5565 * SPA syncing routines
5566 * ==========================================================================
5567 */
5568
5569 static int
5570 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5571 {
5572 bpobj_t *bpo = arg;
5573 bpobj_enqueue(bpo, bp, tx);
5574 return (0);
5575 }
5576
5577 static int
5578 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5579 {
5580 zio_t *zio = arg;
5581
5582 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5583 zio->io_flags));
5584 return (0);
5585 }
5586
5587 static void
5588 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5589 {
5590 char *packed = NULL;
5591 size_t bufsize;
5592 size_t nvsize = 0;
5593 dmu_buf_t *db;
5594
5595 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5596
5597 /*
5598 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5599 * information. This avoids the dbuf_will_dirty() path and
5600 * saves us a pre-read to get data we don't actually care about.
5601 */
5602 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5603 packed = kmem_alloc(bufsize, KM_SLEEP);
5604
5605 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5606 KM_SLEEP) == 0);
5607 bzero(packed + nvsize, bufsize - nvsize);
5608
5609 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5610
5611 kmem_free(packed, bufsize);
5612
5613 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5614 dmu_buf_will_dirty(db, tx);
5615 *(uint64_t *)db->db_data = nvsize;
5616 dmu_buf_rele(db, FTAG);
5617 }
5618
5619 static void
5620 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5621 const char *config, const char *entry)
5622 {
5623 nvlist_t *nvroot;
5624 nvlist_t **list;
5625 int i;
5626
5627 if (!sav->sav_sync)
5628 return;
5629
5630 /*
5631 * Update the MOS nvlist describing the list of available devices.
5632 * spa_validate_aux() will have already made sure this nvlist is
5633 * valid and the vdevs are labeled appropriately.
5634 */
5635 if (sav->sav_object == 0) {
5636 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5637 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5638 sizeof (uint64_t), tx);
5639 VERIFY(zap_update(spa->spa_meta_objset,
5640 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5641 &sav->sav_object, tx) == 0);
5642 }
5643
5644 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5645 if (sav->sav_count == 0) {
5646 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5647 } else {
5648 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5649 for (i = 0; i < sav->sav_count; i++)
5650 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5651 B_FALSE, VDEV_CONFIG_L2CACHE);
5652 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5653 sav->sav_count) == 0);
5654 for (i = 0; i < sav->sav_count; i++)
5655 nvlist_free(list[i]);
5656 kmem_free(list, sav->sav_count * sizeof (void *));
5657 }
5658
5659 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5660 nvlist_free(nvroot);
5661
5662 sav->sav_sync = B_FALSE;
5663 }
5664
5665 static void
5666 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5667 {
5668 nvlist_t *config;
5669
5670 if (list_is_empty(&spa->spa_config_dirty_list))
5671 return;
5672
5673 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5674
5675 config = spa_config_generate(spa, spa->spa_root_vdev,
5676 dmu_tx_get_txg(tx), B_FALSE);
5677
5678 spa_config_exit(spa, SCL_STATE, FTAG);
5679
5680 if (spa->spa_config_syncing)
5681 nvlist_free(spa->spa_config_syncing);
5682 spa->spa_config_syncing = config;
5683
5684 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5685 }
5686
5687 static void
5688 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5689 {
5690 spa_t *spa = arg1;
5691 uint64_t version = *(uint64_t *)arg2;
5692
5693 /*
5694 * Setting the version is special cased when first creating the pool.
5695 */
5696 ASSERT(tx->tx_txg != TXG_INITIAL);
5697
5698 ASSERT(version <= SPA_VERSION);
5699 ASSERT(version >= spa_version(spa));
5700
5701 spa->spa_uberblock.ub_version = version;
5702 vdev_config_dirty(spa->spa_root_vdev);
5703 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5704 }
5705
5706 /*
5707 * Set zpool properties.
5708 */
5709 static void
5710 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5711 {
5712 spa_t *spa = arg1;
5713 objset_t *mos = spa->spa_meta_objset;
5714 nvlist_t *nvp = arg2;
5715 nvpair_t *elem = NULL;
5716
5717 mutex_enter(&spa->spa_props_lock);
5718
5719 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5720 uint64_t intval;
5721 char *strval, *fname;
5722 zpool_prop_t prop;
5723 const char *propname;
5724 zprop_type_t proptype;
5725 zfeature_info_t *feature;
5726
5727 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5728 case ZPROP_INVAL:
5729 /*
5730 * We checked this earlier in spa_prop_validate().
5731 */
5732 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5733
5734 fname = strchr(nvpair_name(elem), '@') + 1;
5735 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5736
5737 spa_feature_enable(spa, feature, tx);
5738 spa_history_log_internal(spa, "set", tx,
5739 "%s=enabled", nvpair_name(elem));
5740 break;
5741
5742 case ZPOOL_PROP_VERSION:
5743 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5744 /*
5745 * The version is synced seperatly before other
5746 * properties and should be correct by now.
5747 */
5748 ASSERT3U(spa_version(spa), >=, intval);
5749 break;
5750
5751 case ZPOOL_PROP_ALTROOT:
5752 /*
5753 * 'altroot' is a non-persistent property. It should
5754 * have been set temporarily at creation or import time.
5755 */
5756 ASSERT(spa->spa_root != NULL);
5757 break;
5758
5759 case ZPOOL_PROP_READONLY:
5760 case ZPOOL_PROP_CACHEFILE:
5761 /*
5762 * 'readonly' and 'cachefile' are also non-persisitent
5763 * properties.
5764 */
5765 break;
5766 case ZPOOL_PROP_COMMENT:
5767 VERIFY(nvpair_value_string(elem, &strval) == 0);
5768 if (spa->spa_comment != NULL)
5769 spa_strfree(spa->spa_comment);
5770 spa->spa_comment = spa_strdup(strval);
5771 /*
5772 * We need to dirty the configuration on all the vdevs
5773 * so that their labels get updated. It's unnecessary
5774 * to do this for pool creation since the vdev's
5775 * configuratoin has already been dirtied.
5776 */
5777 if (tx->tx_txg != TXG_INITIAL)
5778 vdev_config_dirty(spa->spa_root_vdev);
5779 spa_history_log_internal(spa, "set", tx,
5780 "%s=%s", nvpair_name(elem), strval);
5781 break;
5782 default:
5783 /*
5784 * Set pool property values in the poolprops mos object.
5785 */
5786 if (spa->spa_pool_props_object == 0) {
5787 spa->spa_pool_props_object =
5788 zap_create_link(mos, DMU_OT_POOL_PROPS,
5789 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5790 tx);
5791 }
5792
5793 /* normalize the property name */
5794 propname = zpool_prop_to_name(prop);
5795 proptype = zpool_prop_get_type(prop);
5796
5797 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5798 ASSERT(proptype == PROP_TYPE_STRING);
5799 VERIFY(nvpair_value_string(elem, &strval) == 0);
5800 VERIFY(zap_update(mos,
5801 spa->spa_pool_props_object, propname,
5802 1, strlen(strval) + 1, strval, tx) == 0);
5803 spa_history_log_internal(spa, "set", tx,
5804 "%s=%s", nvpair_name(elem), strval);
5805 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5806 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5807
5808 if (proptype == PROP_TYPE_INDEX) {
5809 const char *unused;
5810 VERIFY(zpool_prop_index_to_string(
5811 prop, intval, &unused) == 0);
5812 }
5813 VERIFY(zap_update(mos,
5814 spa->spa_pool_props_object, propname,
5815 8, 1, &intval, tx) == 0);
5816 spa_history_log_internal(spa, "set", tx,
5817 "%s=%lld", nvpair_name(elem), intval);
5818 } else {
5819 ASSERT(0); /* not allowed */
5820 }
5821
5822 switch (prop) {
5823 case ZPOOL_PROP_DELEGATION:
5824 spa->spa_delegation = intval;
5825 break;
5826 case ZPOOL_PROP_BOOTFS:
5827 spa->spa_bootfs = intval;
5828 break;
5829 case ZPOOL_PROP_FAILUREMODE:
5830 spa->spa_failmode = intval;
5831 break;
5832 case ZPOOL_PROP_AUTOEXPAND:
5833 spa->spa_autoexpand = intval;
5834 if (tx->tx_txg != TXG_INITIAL)
5835 spa_async_request(spa,
5836 SPA_ASYNC_AUTOEXPAND);
5837 break;
5838 case ZPOOL_PROP_DEDUPDITTO:
5839 spa->spa_dedup_ditto = intval;
5840 break;
5841 default:
5842 break;
5843 }
5844 }
5845
5846 }
5847
5848 mutex_exit(&spa->spa_props_lock);
5849 }
5850
5851 /*
5852 * Perform one-time upgrade on-disk changes. spa_version() does not
5853 * reflect the new version this txg, so there must be no changes this
5854 * txg to anything that the upgrade code depends on after it executes.
5855 * Therefore this must be called after dsl_pool_sync() does the sync
5856 * tasks.
5857 */
5858 static void
5859 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5860 {
5861 dsl_pool_t *dp = spa->spa_dsl_pool;
5862
5863 ASSERT(spa->spa_sync_pass == 1);
5864
5865 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5866 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5867 dsl_pool_create_origin(dp, tx);
5868
5869 /* Keeping the origin open increases spa_minref */
5870 spa->spa_minref += 3;
5871 }
5872
5873 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5874 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5875 dsl_pool_upgrade_clones(dp, tx);
5876 }
5877
5878 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5879 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5880 dsl_pool_upgrade_dir_clones(dp, tx);
5881
5882 /* Keeping the freedir open increases spa_minref */
5883 spa->spa_minref += 3;
5884 }
5885
5886 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5887 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5888 spa_feature_create_zap_objects(spa, tx);
5889 }
5890 }
5891
5892 /*
5893 * Sync the specified transaction group. New blocks may be dirtied as
5894 * part of the process, so we iterate until it converges.
5895 */
5896 void
5897 spa_sync(spa_t *spa, uint64_t txg)
5898 {
5899 dsl_pool_t *dp = spa->spa_dsl_pool;
5900 objset_t *mos = spa->spa_meta_objset;
5901 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5902 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5903 vdev_t *rvd = spa->spa_root_vdev;
5904 vdev_t *vd;
5905 dmu_tx_t *tx;
5906 int error;
5907
5908 VERIFY(spa_writeable(spa));
5909
5910 /*
5911 * Lock out configuration changes.
5912 */
5913 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5914
5915 spa->spa_syncing_txg = txg;
5916 spa->spa_sync_pass = 0;
5917
5918 /*
5919 * If there are any pending vdev state changes, convert them
5920 * into config changes that go out with this transaction group.
5921 */
5922 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5923 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5924 /*
5925 * We need the write lock here because, for aux vdevs,
5926 * calling vdev_config_dirty() modifies sav_config.
5927 * This is ugly and will become unnecessary when we
5928 * eliminate the aux vdev wart by integrating all vdevs
5929 * into the root vdev tree.
5930 */
5931 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5932 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5933 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5934 vdev_state_clean(vd);
5935 vdev_config_dirty(vd);
5936 }
5937 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5938 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5939 }
5940 spa_config_exit(spa, SCL_STATE, FTAG);
5941
5942 tx = dmu_tx_create_assigned(dp, txg);
5943
5944 /*
5945 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5946 * set spa_deflate if we have no raid-z vdevs.
5947 */
5948 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5949 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5950 int i;
5951
5952 for (i = 0; i < rvd->vdev_children; i++) {
5953 vd = rvd->vdev_child[i];
5954 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5955 break;
5956 }
5957 if (i == rvd->vdev_children) {
5958 spa->spa_deflate = TRUE;
5959 VERIFY(0 == zap_add(spa->spa_meta_objset,
5960 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5961 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5962 }
5963 }
5964
5965 /*
5966 * If anything has changed in this txg, or if someone is waiting
5967 * for this txg to sync (eg, spa_vdev_remove()), push the
5968 * deferred frees from the previous txg. If not, leave them
5969 * alone so that we don't generate work on an otherwise idle
5970 * system.
5971 */
5972 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5973 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5974 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5975 ((dsl_scan_active(dp->dp_scan) ||
5976 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5977 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5978 VERIFY3U(bpobj_iterate(defer_bpo,
5979 spa_free_sync_cb, zio, tx), ==, 0);
5980 VERIFY3U(zio_wait(zio), ==, 0);
5981 }
5982
5983 /*
5984 * Iterate to convergence.
5985 */
5986 do {
5987 int pass = ++spa->spa_sync_pass;
5988
5989 spa_sync_config_object(spa, tx);
5990 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5991 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5992 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5993 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5994 spa_errlog_sync(spa, txg);
5995 dsl_pool_sync(dp, txg);
5996
5997 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5998 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5999 bplist_iterate(free_bpl, spa_free_sync_cb,
6000 zio, tx);
6001 VERIFY(zio_wait(zio) == 0);
6002 } else {
6003 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6004 defer_bpo, tx);
6005 }
6006
6007 ddt_sync(spa, txg);
6008 dsl_scan_sync(dp, tx);
6009
6010 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6011 vdev_sync(vd, txg);
6012
6013 if (pass == 1)
6014 spa_sync_upgrades(spa, tx);
6015
6016 } while (dmu_objset_is_dirty(mos, txg));
6017
6018 /*
6019 * Rewrite the vdev configuration (which includes the uberblock)
6020 * to commit the transaction group.
6021 *
6022 * If there are no dirty vdevs, we sync the uberblock to a few
6023 * random top-level vdevs that are known to be visible in the
6024 * config cache (see spa_vdev_add() for a complete description).
6025 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6026 */
6027 for (;;) {
6028 /*
6029 * We hold SCL_STATE to prevent vdev open/close/etc.
6030 * while we're attempting to write the vdev labels.
6031 */
6032 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6033
6034 if (list_is_empty(&spa->spa_config_dirty_list)) {
6035 vdev_t *svd[SPA_DVAS_PER_BP];
6036 int svdcount = 0;
6037 int children = rvd->vdev_children;
6038 int c0 = spa_get_random(children);
6039
6040 for (int c = 0; c < children; c++) {
6041 vd = rvd->vdev_child[(c0 + c) % children];
6042 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6043 continue;
6044 svd[svdcount++] = vd;
6045 if (svdcount == SPA_DVAS_PER_BP)
6046 break;
6047 }
6048 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6049 if (error != 0)
6050 error = vdev_config_sync(svd, svdcount, txg,
6051 B_TRUE);
6052 } else {
6053 error = vdev_config_sync(rvd->vdev_child,
6054 rvd->vdev_children, txg, B_FALSE);
6055 if (error != 0)
6056 error = vdev_config_sync(rvd->vdev_child,
6057 rvd->vdev_children, txg, B_TRUE);
6058 }
6059
6060 spa_config_exit(spa, SCL_STATE, FTAG);
6061
6062 if (error == 0)
6063 break;
6064 zio_suspend(spa, NULL);
6065 zio_resume_wait(spa);
6066 }
6067 dmu_tx_commit(tx);
6068
6069 /*
6070 * Clear the dirty config list.
6071 */
6072 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6073 vdev_config_clean(vd);
6074
6075 /*
6076 * Now that the new config has synced transactionally,
6077 * let it become visible to the config cache.
6078 */
6079 if (spa->spa_config_syncing != NULL) {
6080 spa_config_set(spa, spa->spa_config_syncing);
6081 spa->spa_config_txg = txg;
6082 spa->spa_config_syncing = NULL;
6083 }
6084
6085 spa->spa_ubsync = spa->spa_uberblock;
6086
6087 dsl_pool_sync_done(dp, txg);
6088
6089 /*
6090 * Update usable space statistics.
6091 */
6092 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6093 vdev_sync_done(vd, txg);
6094
6095 spa_update_dspace(spa);
6096
6097 /*
6098 * It had better be the case that we didn't dirty anything
6099 * since vdev_config_sync().
6100 */
6101 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6102 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6103 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6104
6105 spa->spa_sync_pass = 0;
6106
6107 spa_config_exit(spa, SCL_CONFIG, FTAG);
6108
6109 spa_handle_ignored_writes(spa);
6110
6111 /*
6112 * If any async tasks have been requested, kick them off.
6113 */
6114 spa_async_dispatch(spa);
6115 }
6116
6117 /*
6118 * Sync all pools. We don't want to hold the namespace lock across these
6119 * operations, so we take a reference on the spa_t and drop the lock during the
6120 * sync.
6121 */
6122 void
6123 spa_sync_allpools(void)
6124 {
6125 spa_t *spa = NULL;
6126 mutex_enter(&spa_namespace_lock);
6127 while ((spa = spa_next(spa)) != NULL) {
6128 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6129 !spa_writeable(spa) || spa_suspended(spa))
6130 continue;
6131 spa_open_ref(spa, FTAG);
6132 mutex_exit(&spa_namespace_lock);
6133 txg_wait_synced(spa_get_dsl(spa), 0);
6134 mutex_enter(&spa_namespace_lock);
6135 spa_close(spa, FTAG);
6136 }
6137 mutex_exit(&spa_namespace_lock);
6138 }
6139
6140 /*
6141 * ==========================================================================
6142 * Miscellaneous routines
6143 * ==========================================================================
6144 */
6145
6146 /*
6147 * Remove all pools in the system.
6148 */
6149 void
6150 spa_evict_all(void)
6151 {
6152 spa_t *spa;
6153
6154 /*
6155 * Remove all cached state. All pools should be closed now,
6156 * so every spa in the AVL tree should be unreferenced.
6157 */
6158 mutex_enter(&spa_namespace_lock);
6159 while ((spa = spa_next(NULL)) != NULL) {
6160 /*
6161 * Stop async tasks. The async thread may need to detach
6162 * a device that's been replaced, which requires grabbing
6163 * spa_namespace_lock, so we must drop it here.
6164 */
6165 spa_open_ref(spa, FTAG);
6166 mutex_exit(&spa_namespace_lock);
6167 spa_async_suspend(spa);
6168 mutex_enter(&spa_namespace_lock);
6169 spa_close(spa, FTAG);
6170
6171 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6172 spa_unload(spa);
6173 spa_deactivate(spa);
6174 }
6175 spa_remove(spa);
6176 }
6177 mutex_exit(&spa_namespace_lock);
6178 }
6179
6180 vdev_t *
6181 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6182 {
6183 vdev_t *vd;
6184 int i;
6185
6186 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6187 return (vd);
6188
6189 if (aux) {
6190 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6191 vd = spa->spa_l2cache.sav_vdevs[i];
6192 if (vd->vdev_guid == guid)
6193 return (vd);
6194 }
6195
6196 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6197 vd = spa->spa_spares.sav_vdevs[i];
6198 if (vd->vdev_guid == guid)
6199 return (vd);
6200 }
6201 }
6202
6203 return (NULL);
6204 }
6205
6206 void
6207 spa_upgrade(spa_t *spa, uint64_t version)
6208 {
6209 ASSERT(spa_writeable(spa));
6210
6211 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6212
6213 /*
6214 * This should only be called for a non-faulted pool, and since a
6215 * future version would result in an unopenable pool, this shouldn't be
6216 * possible.
6217 */
6218 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6219 ASSERT(version >= spa->spa_uberblock.ub_version);
6220
6221 spa->spa_uberblock.ub_version = version;
6222 vdev_config_dirty(spa->spa_root_vdev);
6223
6224 spa_config_exit(spa, SCL_ALL, FTAG);
6225
6226 txg_wait_synced(spa_get_dsl(spa), 0);
6227 }
6228
6229 boolean_t
6230 spa_has_spare(spa_t *spa, uint64_t guid)
6231 {
6232 int i;
6233 uint64_t spareguid;
6234 spa_aux_vdev_t *sav = &spa->spa_spares;
6235
6236 for (i = 0; i < sav->sav_count; i++)
6237 if (sav->sav_vdevs[i]->vdev_guid == guid)
6238 return (B_TRUE);
6239
6240 for (i = 0; i < sav->sav_npending; i++) {
6241 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6242 &spareguid) == 0 && spareguid == guid)
6243 return (B_TRUE);
6244 }
6245
6246 return (B_FALSE);
6247 }
6248
6249 /*
6250 * Check if a pool has an active shared spare device.
6251 * Note: reference count of an active spare is 2, as a spare and as a replace
6252 */
6253 static boolean_t
6254 spa_has_active_shared_spare(spa_t *spa)
6255 {
6256 int i, refcnt;
6257 uint64_t pool;
6258 spa_aux_vdev_t *sav = &spa->spa_spares;
6259
6260 for (i = 0; i < sav->sav_count; i++) {
6261 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6262 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6263 refcnt > 2)
6264 return (B_TRUE);
6265 }
6266
6267 return (B_FALSE);
6268 }
6269
6270 /*
6271 * Post a sysevent corresponding to the given event. The 'name' must be one of
6272 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6273 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6274 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6275 * or zdb as real changes.
6276 */
6277 void
6278 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6279 {
6280 #ifdef _KERNEL
6281 sysevent_t *ev;
6282 sysevent_attr_list_t *attr = NULL;
6283 sysevent_value_t value;
6284 sysevent_id_t eid;
6285
6286 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6287 SE_SLEEP);
6288
6289 value.value_type = SE_DATA_TYPE_STRING;
6290 value.value.sv_string = spa_name(spa);
6291 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6292 goto done;
6293
6294 value.value_type = SE_DATA_TYPE_UINT64;
6295 value.value.sv_uint64 = spa_guid(spa);
6296 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6297 goto done;
6298
6299 if (vd) {
6300 value.value_type = SE_DATA_TYPE_UINT64;
6301 value.value.sv_uint64 = vd->vdev_guid;
6302 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6303 SE_SLEEP) != 0)
6304 goto done;
6305
6306 if (vd->vdev_path) {
6307 value.value_type = SE_DATA_TYPE_STRING;
6308 value.value.sv_string = vd->vdev_path;
6309 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6310 &value, SE_SLEEP) != 0)
6311 goto done;
6312 }
6313 }
6314
6315 if (sysevent_attach_attributes(ev, attr) != 0)
6316 goto done;
6317 attr = NULL;
6318
6319 (void) log_sysevent(ev, SE_SLEEP, &eid);
6320
6321 done:
6322 if (attr)
6323 sysevent_free_attr(attr);
6324 sysevent_free(ev);
6325 #endif
6326 }