1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. 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 #include <sys/dsl_destroy.h>
67
68 #ifdef _KERNEL
69 #include <sys/bootprops.h>
70 #include <sys/callb.h>
71 #include <sys/cpupart.h>
72 #include <sys/pool.h>
73 #include <sys/sysdc.h>
74 #include <sys/zone.h>
75 #endif /* _KERNEL */
76
77 #include "zfs_prop.h"
78 #include "zfs_comutil.h"
79
80 /*
81 * The interval at which failed configuration cache file writes
82 * should be retried.
83 */
84 static int zfs_ccw_retry_interval = 300;
85
86 typedef enum zti_modes {
87 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
88 ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */
89 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
90 ZTI_MODE_NULL, /* don't create a taskq */
91 ZTI_NMODES
92 } zti_modes_t;
93
94 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
95 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
96 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
97 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
98
99 #define ZTI_N(n) ZTI_P(n, 1)
100 #define ZTI_ONE ZTI_N(1)
101
102 typedef struct zio_taskq_info {
103 zti_modes_t zti_mode;
104 uint_t zti_value;
105 uint_t zti_count;
106 } zio_taskq_info_t;
107
108 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109 "issue", "issue_high", "intr", "intr_high"
110 };
111
112 /*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
132 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
133 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
134 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
135 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
137 };
138
139 static void spa_sync_version(void *arg, dmu_tx_t *tx);
140 static void spa_sync_props(void *arg, dmu_tx_t *tx);
141 static boolean_t spa_has_active_shared_spare(spa_t *spa);
142 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144 char **ereport);
145 static void spa_vdev_resilver_done(spa_t *spa);
146
147 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
148 id_t zio_taskq_psrset_bind = PS_NONE;
149 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 uint_t zio_taskq_basedc = 80; /* base duty cycle */
151
152 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 extern int zfs_sync_pass_deferred_free;
154
155 /*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159 #define TRYIMPORT_NAME "$import"
160
161 /*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167 /*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170 static void
171 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172 uint64_t intval, zprop_source_t src)
173 {
174 const char *propname = zpool_prop_to_name(prop);
175 nvlist_t *propval;
176
177 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180 if (strval != NULL)
181 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182 else
183 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186 nvlist_free(propval);
187 }
188
189 /*
190 * Get property values from the spa configuration.
191 */
192 static void
193 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194 {
195 vdev_t *rvd = spa->spa_root_vdev;
196 dsl_pool_t *pool = spa->spa_dsl_pool;
197 uint64_t size;
198 uint64_t alloc;
199 uint64_t space;
200 uint64_t cap, version;
201 zprop_source_t src = ZPROP_SRC_NONE;
202 spa_config_dirent_t *dp;
203
204 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206 if (rvd != NULL) {
207 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208 size = metaslab_class_get_space(spa_normal_class(spa));
209 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 size - alloc, src);
214
215 space = 0;
216 for (int c = 0; c < rvd->vdev_children; c++) {
217 vdev_t *tvd = rvd->vdev_child[c];
218 space += tvd->vdev_max_asize - tvd->vdev_asize;
219 }
220 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221 src);
222
223 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 (spa_mode(spa) == FREAD), src);
225
226 cap = (size == 0) ? 0 : (alloc * 100 / size);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228
229 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 ddt_get_pool_dedup_ratio(spa), src);
231
232 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 rvd->vdev_state, src);
234
235 version = spa_version(spa);
236 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 src = ZPROP_SRC_DEFAULT;
238 else
239 src = ZPROP_SRC_LOCAL;
240 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 }
242
243 if (pool != NULL) {
244 dsl_dir_t *freedir = pool->dp_free_dir;
245
246 /*
247 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248 * when opening pools before this version freedir will be NULL.
249 */
250 if (freedir != NULL) {
251 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252 freedir->dd_phys->dd_used_bytes, src);
253 } else {
254 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 NULL, 0, src);
256 }
257 }
258
259 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260
261 if (spa->spa_comment != NULL) {
262 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 0, ZPROP_SRC_LOCAL);
264 }
265
266 if (spa->spa_root != NULL)
267 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268 0, ZPROP_SRC_LOCAL);
269
270 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271 if (dp->scd_path == NULL) {
272 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 "none", 0, ZPROP_SRC_LOCAL);
274 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276 dp->scd_path, 0, ZPROP_SRC_LOCAL);
277 }
278 }
279 }
280
281 /*
282 * Get zpool property values.
283 */
284 int
285 spa_prop_get(spa_t *spa, nvlist_t **nvp)
286 {
287 objset_t *mos = spa->spa_meta_objset;
288 zap_cursor_t zc;
289 zap_attribute_t za;
290 int err;
291
292 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293
294 mutex_enter(&spa->spa_props_lock);
295
296 /*
297 * Get properties from the spa config.
298 */
299 spa_prop_get_config(spa, nvp);
300
301 /* If no pool property object, no more prop to get. */
302 if (mos == NULL || spa->spa_pool_props_object == 0) {
303 mutex_exit(&spa->spa_props_lock);
304 return (0);
305 }
306
307 /*
308 * Get properties from the MOS pool property object.
309 */
310 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311 (err = zap_cursor_retrieve(&zc, &za)) == 0;
312 zap_cursor_advance(&zc)) {
313 uint64_t intval = 0;
314 char *strval = NULL;
315 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 zpool_prop_t prop;
317
318 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 continue;
320
321 switch (za.za_integer_length) {
322 case 8:
323 /* integer property */
324 if (za.za_first_integer !=
325 zpool_prop_default_numeric(prop))
326 src = ZPROP_SRC_LOCAL;
327
328 if (prop == ZPOOL_PROP_BOOTFS) {
329 dsl_pool_t *dp;
330 dsl_dataset_t *ds = NULL;
331
332 dp = spa_get_dsl(spa);
333 dsl_pool_config_enter(dp, FTAG);
334 if (err = dsl_dataset_hold_obj(dp,
335 za.za_first_integer, FTAG, &ds)) {
336 dsl_pool_config_exit(dp, FTAG);
337 break;
338 }
339
340 strval = kmem_alloc(
341 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 KM_SLEEP);
343 dsl_dataset_name(ds, strval);
344 dsl_dataset_rele(ds, FTAG);
345 dsl_pool_config_exit(dp, FTAG);
346 } else {
347 strval = NULL;
348 intval = za.za_first_integer;
349 }
350
351 spa_prop_add_list(*nvp, prop, strval, intval, src);
352
353 if (strval != NULL)
354 kmem_free(strval,
355 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356
357 break;
358
359 case 1:
360 /* string property */
361 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362 err = zap_lookup(mos, spa->spa_pool_props_object,
363 za.za_name, 1, za.za_num_integers, strval);
364 if (err) {
365 kmem_free(strval, za.za_num_integers);
366 break;
367 }
368 spa_prop_add_list(*nvp, prop, strval, 0, src);
369 kmem_free(strval, za.za_num_integers);
370 break;
371
372 default:
373 break;
374 }
375 }
376 zap_cursor_fini(&zc);
377 mutex_exit(&spa->spa_props_lock);
378 out:
379 if (err && err != ENOENT) {
380 nvlist_free(*nvp);
381 *nvp = NULL;
382 return (err);
383 }
384
385 return (0);
386 }
387
388 /*
389 * Validate the given pool properties nvlist and modify the list
390 * for the property values to be set.
391 */
392 static int
393 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 {
395 nvpair_t *elem;
396 int error = 0, reset_bootfs = 0;
397 uint64_t objnum = 0;
398 boolean_t has_feature = B_FALSE;
399
400 elem = NULL;
401 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 uint64_t intval;
403 char *strval, *slash, *check, *fname;
404 const char *propname = nvpair_name(elem);
405 zpool_prop_t prop = zpool_name_to_prop(propname);
406
407 switch (prop) {
408 case ZPROP_INVAL:
409 if (!zpool_prop_feature(propname)) {
410 error = SET_ERROR(EINVAL);
411 break;
412 }
413
414 /*
415 * Sanitize the input.
416 */
417 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418 error = SET_ERROR(EINVAL);
419 break;
420 }
421
422 if (nvpair_value_uint64(elem, &intval) != 0) {
423 error = SET_ERROR(EINVAL);
424 break;
425 }
426
427 if (intval != 0) {
428 error = SET_ERROR(EINVAL);
429 break;
430 }
431
432 fname = strchr(propname, '@') + 1;
433 if (zfeature_lookup_name(fname, NULL) != 0) {
434 error = SET_ERROR(EINVAL);
435 break;
436 }
437
438 has_feature = B_TRUE;
439 break;
440
441 case ZPOOL_PROP_VERSION:
442 error = nvpair_value_uint64(elem, &intval);
443 if (!error &&
444 (intval < spa_version(spa) ||
445 intval > SPA_VERSION_BEFORE_FEATURES ||
446 has_feature))
447 error = SET_ERROR(EINVAL);
448 break;
449
450 case ZPOOL_PROP_DELEGATION:
451 case ZPOOL_PROP_AUTOREPLACE:
452 case ZPOOL_PROP_LISTSNAPS:
453 case ZPOOL_PROP_AUTOEXPAND:
454 error = nvpair_value_uint64(elem, &intval);
455 if (!error && intval > 1)
456 error = SET_ERROR(EINVAL);
457 break;
458
459 case ZPOOL_PROP_BOOTFS:
460 /*
461 * If the pool version is less than SPA_VERSION_BOOTFS,
462 * or the pool is still being created (version == 0),
463 * the bootfs property cannot be set.
464 */
465 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466 error = SET_ERROR(ENOTSUP);
467 break;
468 }
469
470 /*
471 * Make sure the vdev config is bootable
472 */
473 if (!vdev_is_bootable(spa->spa_root_vdev)) {
474 error = SET_ERROR(ENOTSUP);
475 break;
476 }
477
478 reset_bootfs = 1;
479
480 error = nvpair_value_string(elem, &strval);
481
482 if (!error) {
483 objset_t *os;
484 uint64_t compress;
485
486 if (strval == NULL || strval[0] == '\0') {
487 objnum = zpool_prop_default_numeric(
488 ZPOOL_PROP_BOOTFS);
489 break;
490 }
491
492 if (error = dmu_objset_hold(strval, FTAG, &os))
493 break;
494
495 /* Must be ZPL and not gzip compressed. */
496
497 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 error = SET_ERROR(ENOTSUP);
499 } else if ((error =
500 dsl_prop_get_int_ds(dmu_objset_ds(os),
501 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502 &compress)) == 0 &&
503 !BOOTFS_COMPRESS_VALID(compress)) {
504 error = SET_ERROR(ENOTSUP);
505 } else {
506 objnum = dmu_objset_id(os);
507 }
508 dmu_objset_rele(os, FTAG);
509 }
510 break;
511
512 case ZPOOL_PROP_FAILUREMODE:
513 error = nvpair_value_uint64(elem, &intval);
514 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515 intval > ZIO_FAILURE_MODE_PANIC))
516 error = SET_ERROR(EINVAL);
517
518 /*
519 * This is a special case which only occurs when
520 * the pool has completely failed. This allows
521 * the user to change the in-core failmode property
522 * without syncing it out to disk (I/Os might
523 * currently be blocked). We do this by returning
524 * EIO to the caller (spa_prop_set) to trick it
525 * into thinking we encountered a property validation
526 * error.
527 */
528 if (!error && spa_suspended(spa)) {
529 spa->spa_failmode = intval;
530 error = SET_ERROR(EIO);
531 }
532 break;
533
534 case ZPOOL_PROP_CACHEFILE:
535 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 break;
537
538 if (strval[0] == '\0')
539 break;
540
541 if (strcmp(strval, "none") == 0)
542 break;
543
544 if (strval[0] != '/') {
545 error = SET_ERROR(EINVAL);
546 break;
547 }
548
549 slash = strrchr(strval, '/');
550 ASSERT(slash != NULL);
551
552 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553 strcmp(slash, "/..") == 0)
554 error = SET_ERROR(EINVAL);
555 break;
556
557 case ZPOOL_PROP_COMMENT:
558 if ((error = nvpair_value_string(elem, &strval)) != 0)
559 break;
560 for (check = strval; *check != '\0'; check++) {
561 /*
562 * The kernel doesn't have an easy isprint()
563 * check. For this kernel check, we merely
564 * check ASCII apart from DEL. Fix this if
565 * there is an easy-to-use kernel isprint().
566 */
567 if (*check >= 0x7f) {
568 error = SET_ERROR(EINVAL);
569 break;
570 }
571 check++;
572 }
573 if (strlen(strval) > ZPROP_MAX_COMMENT)
574 error = E2BIG;
575 break;
576
577 case ZPOOL_PROP_DEDUPDITTO:
578 if (spa_version(spa) < SPA_VERSION_DEDUP)
579 error = SET_ERROR(ENOTSUP);
580 else
581 error = nvpair_value_uint64(elem, &intval);
582 if (error == 0 &&
583 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584 error = SET_ERROR(EINVAL);
585 break;
586 }
587
588 if (error)
589 break;
590 }
591
592 if (!error && reset_bootfs) {
593 error = nvlist_remove(props,
594 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595
596 if (!error) {
597 error = nvlist_add_uint64(props,
598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599 }
600 }
601
602 return (error);
603 }
604
605 void
606 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607 {
608 char *cachefile;
609 spa_config_dirent_t *dp;
610
611 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612 &cachefile) != 0)
613 return;
614
615 dp = kmem_alloc(sizeof (spa_config_dirent_t),
616 KM_SLEEP);
617
618 if (cachefile[0] == '\0')
619 dp->scd_path = spa_strdup(spa_config_path);
620 else if (strcmp(cachefile, "none") == 0)
621 dp->scd_path = NULL;
622 else
623 dp->scd_path = spa_strdup(cachefile);
624
625 list_insert_head(&spa->spa_config_list, dp);
626 if (need_sync)
627 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628 }
629
630 int
631 spa_prop_set(spa_t *spa, nvlist_t *nvp)
632 {
633 int error;
634 nvpair_t *elem = NULL;
635 boolean_t need_sync = B_FALSE;
636
637 if ((error = spa_prop_validate(spa, nvp)) != 0)
638 return (error);
639
640 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642
643 if (prop == ZPOOL_PROP_CACHEFILE ||
644 prop == ZPOOL_PROP_ALTROOT ||
645 prop == ZPOOL_PROP_READONLY)
646 continue;
647
648 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649 uint64_t ver;
650
651 if (prop == ZPOOL_PROP_VERSION) {
652 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653 } else {
654 ASSERT(zpool_prop_feature(nvpair_name(elem)));
655 ver = SPA_VERSION_FEATURES;
656 need_sync = B_TRUE;
657 }
658
659 /* Save time if the version is already set. */
660 if (ver == spa_version(spa))
661 continue;
662
663 /*
664 * In addition to the pool directory object, we might
665 * create the pool properties object, the features for
666 * read object, the features for write object, or the
667 * feature descriptions object.
668 */
669 error = dsl_sync_task(spa->spa_name, NULL,
670 spa_sync_version, &ver, 6);
671 if (error)
672 return (error);
673 continue;
674 }
675
676 need_sync = B_TRUE;
677 break;
678 }
679
680 if (need_sync) {
681 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682 nvp, 6));
683 }
684
685 return (0);
686 }
687
688 /*
689 * If the bootfs property value is dsobj, clear it.
690 */
691 void
692 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693 {
694 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695 VERIFY(zap_remove(spa->spa_meta_objset,
696 spa->spa_pool_props_object,
697 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 spa->spa_bootfs = 0;
699 }
700 }
701
702 /*ARGSUSED*/
703 static int
704 spa_change_guid_check(void *arg, dmu_tx_t *tx)
705 {
706 uint64_t *newguid = arg;
707 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708 vdev_t *rvd = spa->spa_root_vdev;
709 uint64_t vdev_state;
710
711 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712 vdev_state = rvd->vdev_state;
713 spa_config_exit(spa, SCL_STATE, FTAG);
714
715 if (vdev_state != VDEV_STATE_HEALTHY)
716 return (SET_ERROR(ENXIO));
717
718 ASSERT3U(spa_guid(spa), !=, *newguid);
719
720 return (0);
721 }
722
723 static void
724 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725 {
726 uint64_t *newguid = arg;
727 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728 uint64_t oldguid;
729 vdev_t *rvd = spa->spa_root_vdev;
730
731 oldguid = spa_guid(spa);
732
733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 rvd->vdev_guid = *newguid;
735 rvd->vdev_guid_sum += (*newguid - oldguid);
736 vdev_config_dirty(rvd);
737 spa_config_exit(spa, SCL_STATE, FTAG);
738
739 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740 oldguid, *newguid);
741 }
742
743 /*
744 * Change the GUID for the pool. This is done so that we can later
745 * re-import a pool built from a clone of our own vdevs. We will modify
746 * the root vdev's guid, our own pool guid, and then mark all of our
747 * vdevs dirty. Note that we must make sure that all our vdevs are
748 * online when we do this, or else any vdevs that weren't present
749 * would be orphaned from our pool. We are also going to issue a
750 * sysevent to update any watchers.
751 */
752 int
753 spa_change_guid(spa_t *spa)
754 {
755 int error;
756 uint64_t guid;
757
758 mutex_enter(&spa_namespace_lock);
759 guid = spa_generate_guid(NULL);
760
761 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
762 spa_change_guid_sync, &guid, 5);
763
764 if (error == 0) {
765 spa_config_sync(spa, B_FALSE, B_TRUE);
766 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
767 }
768
769 mutex_exit(&spa_namespace_lock);
770
771 return (error);
772 }
773
774 /*
775 * ==========================================================================
776 * SPA state manipulation (open/create/destroy/import/export)
777 * ==========================================================================
778 */
779
780 static int
781 spa_error_entry_compare(const void *a, const void *b)
782 {
783 spa_error_entry_t *sa = (spa_error_entry_t *)a;
784 spa_error_entry_t *sb = (spa_error_entry_t *)b;
785 int ret;
786
787 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
788 sizeof (zbookmark_t));
789
790 if (ret < 0)
791 return (-1);
792 else if (ret > 0)
793 return (1);
794 else
795 return (0);
796 }
797
798 /*
799 * Utility function which retrieves copies of the current logs and
800 * re-initializes them in the process.
801 */
802 void
803 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
804 {
805 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
806
807 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
808 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
809
810 avl_create(&spa->spa_errlist_scrub,
811 spa_error_entry_compare, sizeof (spa_error_entry_t),
812 offsetof(spa_error_entry_t, se_avl));
813 avl_create(&spa->spa_errlist_last,
814 spa_error_entry_compare, sizeof (spa_error_entry_t),
815 offsetof(spa_error_entry_t, se_avl));
816 }
817
818 static void
819 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
820 {
821 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
822 enum zti_modes mode = ztip->zti_mode;
823 uint_t value = ztip->zti_value;
824 uint_t count = ztip->zti_count;
825 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
826 char name[32];
827 uint_t flags = 0;
828 boolean_t batch = B_FALSE;
829
830 if (mode == ZTI_MODE_NULL) {
831 tqs->stqs_count = 0;
832 tqs->stqs_taskq = NULL;
833 return;
834 }
835
836 ASSERT3U(count, >, 0);
837
838 tqs->stqs_count = count;
839 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
840
841 for (uint_t i = 0; i < count; i++) {
842 taskq_t *tq;
843
844 switch (mode) {
845 case ZTI_MODE_FIXED:
846 ASSERT3U(value, >=, 1);
847 value = MAX(value, 1);
848 break;
849
850 case ZTI_MODE_BATCH:
851 batch = B_TRUE;
852 flags |= TASKQ_THREADS_CPU_PCT;
853 value = zio_taskq_batch_pct;
854 break;
855
856 case ZTI_MODE_ONLINE_PERCENT:
857 flags |= TASKQ_THREADS_CPU_PCT;
858 break;
859
860 default:
861 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
862 "spa_activate()",
863 zio_type_name[t], zio_taskq_types[q], mode, value);
864 break;
865 }
866
867 if (count > 1) {
868 (void) snprintf(name, sizeof (name), "%s_%s_%u",
869 zio_type_name[t], zio_taskq_types[q], i);
870 } else {
871 (void) snprintf(name, sizeof (name), "%s_%s",
872 zio_type_name[t], zio_taskq_types[q]);
873 }
874
875 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
876 if (batch)
877 flags |= TASKQ_DC_BATCH;
878
879 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
880 spa->spa_proc, zio_taskq_basedc, flags);
881 } else {
882 tq = taskq_create_proc(name, value, maxclsyspri, 50,
883 INT_MAX, spa->spa_proc, flags);
884 }
885
886 tqs->stqs_taskq[i] = tq;
887 }
888 }
889
890 static void
891 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
892 {
893 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
894
895 if (tqs->stqs_taskq == NULL) {
896 ASSERT0(tqs->stqs_count);
897 return;
898 }
899
900 for (uint_t i = 0; i < tqs->stqs_count; i++) {
901 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
902 taskq_destroy(tqs->stqs_taskq[i]);
903 }
904
905 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
906 tqs->stqs_taskq = NULL;
907 }
908
909 /*
910 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
911 * Note that a type may have multiple discrete taskqs to avoid lock contention
912 * on the taskq itself. In that case we choose which taskq at random by using
913 * the low bits of gethrtime().
914 */
915 void
916 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
917 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
918 {
919 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
920 taskq_t *tq;
921
922 ASSERT3P(tqs->stqs_taskq, !=, NULL);
923 ASSERT3U(tqs->stqs_count, !=, 0);
924
925 if (tqs->stqs_count == 1) {
926 tq = tqs->stqs_taskq[0];
927 } else {
928 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
929 }
930
931 taskq_dispatch_ent(tq, func, arg, flags, ent);
932 }
933
934 static void
935 spa_create_zio_taskqs(spa_t *spa)
936 {
937 for (int t = 0; t < ZIO_TYPES; t++) {
938 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
939 spa_taskqs_init(spa, t, q);
940 }
941 }
942 }
943
944 #ifdef _KERNEL
945 static void
946 spa_thread(void *arg)
947 {
948 callb_cpr_t cprinfo;
949
950 spa_t *spa = arg;
951 user_t *pu = PTOU(curproc);
952
953 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
954 spa->spa_name);
955
956 ASSERT(curproc != &p0);
957 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
958 "zpool-%s", spa->spa_name);
959 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
960
961 /* bind this thread to the requested psrset */
962 if (zio_taskq_psrset_bind != PS_NONE) {
963 pool_lock();
964 mutex_enter(&cpu_lock);
965 mutex_enter(&pidlock);
966 mutex_enter(&curproc->p_lock);
967
968 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
969 0, NULL, NULL) == 0) {
970 curthread->t_bind_pset = zio_taskq_psrset_bind;
971 } else {
972 cmn_err(CE_WARN,
973 "Couldn't bind process for zfs pool \"%s\" to "
974 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
975 }
976
977 mutex_exit(&curproc->p_lock);
978 mutex_exit(&pidlock);
979 mutex_exit(&cpu_lock);
980 pool_unlock();
981 }
982
983 if (zio_taskq_sysdc) {
984 sysdc_thread_enter(curthread, 100, 0);
985 }
986
987 spa->spa_proc = curproc;
988 spa->spa_did = curthread->t_did;
989
990 spa_create_zio_taskqs(spa);
991
992 mutex_enter(&spa->spa_proc_lock);
993 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
994
995 spa->spa_proc_state = SPA_PROC_ACTIVE;
996 cv_broadcast(&spa->spa_proc_cv);
997
998 CALLB_CPR_SAFE_BEGIN(&cprinfo);
999 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1000 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1001 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1002
1003 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1004 spa->spa_proc_state = SPA_PROC_GONE;
1005 spa->spa_proc = &p0;
1006 cv_broadcast(&spa->spa_proc_cv);
1007 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1008
1009 mutex_enter(&curproc->p_lock);
1010 lwp_exit();
1011 }
1012 #endif
1013
1014 /*
1015 * Activate an uninitialized pool.
1016 */
1017 static void
1018 spa_activate(spa_t *spa, int mode)
1019 {
1020 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1021
1022 spa->spa_state = POOL_STATE_ACTIVE;
1023 spa->spa_mode = mode;
1024
1025 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1026 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1027
1028 /* Try to create a covering process */
1029 mutex_enter(&spa->spa_proc_lock);
1030 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1031 ASSERT(spa->spa_proc == &p0);
1032 spa->spa_did = 0;
1033
1034 /* Only create a process if we're going to be around a while. */
1035 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1036 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1037 NULL, 0) == 0) {
1038 spa->spa_proc_state = SPA_PROC_CREATED;
1039 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1040 cv_wait(&spa->spa_proc_cv,
1041 &spa->spa_proc_lock);
1042 }
1043 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1044 ASSERT(spa->spa_proc != &p0);
1045 ASSERT(spa->spa_did != 0);
1046 } else {
1047 #ifdef _KERNEL
1048 cmn_err(CE_WARN,
1049 "Couldn't create process for zfs pool \"%s\"\n",
1050 spa->spa_name);
1051 #endif
1052 }
1053 }
1054 mutex_exit(&spa->spa_proc_lock);
1055
1056 /* If we didn't create a process, we need to create our taskqs. */
1057 if (spa->spa_proc == &p0) {
1058 spa_create_zio_taskqs(spa);
1059 }
1060
1061 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1062 offsetof(vdev_t, vdev_config_dirty_node));
1063 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1064 offsetof(vdev_t, vdev_state_dirty_node));
1065
1066 txg_list_create(&spa->spa_vdev_txg_list,
1067 offsetof(struct vdev, vdev_txg_node));
1068
1069 avl_create(&spa->spa_errlist_scrub,
1070 spa_error_entry_compare, sizeof (spa_error_entry_t),
1071 offsetof(spa_error_entry_t, se_avl));
1072 avl_create(&spa->spa_errlist_last,
1073 spa_error_entry_compare, sizeof (spa_error_entry_t),
1074 offsetof(spa_error_entry_t, se_avl));
1075 }
1076
1077 /*
1078 * Opposite of spa_activate().
1079 */
1080 static void
1081 spa_deactivate(spa_t *spa)
1082 {
1083 ASSERT(spa->spa_sync_on == B_FALSE);
1084 ASSERT(spa->spa_dsl_pool == NULL);
1085 ASSERT(spa->spa_root_vdev == NULL);
1086 ASSERT(spa->spa_async_zio_root == NULL);
1087 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1088
1089 txg_list_destroy(&spa->spa_vdev_txg_list);
1090
1091 list_destroy(&spa->spa_config_dirty_list);
1092 list_destroy(&spa->spa_state_dirty_list);
1093
1094 for (int t = 0; t < ZIO_TYPES; t++) {
1095 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1096 spa_taskqs_fini(spa, t, q);
1097 }
1098 }
1099
1100 metaslab_class_destroy(spa->spa_normal_class);
1101 spa->spa_normal_class = NULL;
1102
1103 metaslab_class_destroy(spa->spa_log_class);
1104 spa->spa_log_class = NULL;
1105
1106 /*
1107 * If this was part of an import or the open otherwise failed, we may
1108 * still have errors left in the queues. Empty them just in case.
1109 */
1110 spa_errlog_drain(spa);
1111
1112 avl_destroy(&spa->spa_errlist_scrub);
1113 avl_destroy(&spa->spa_errlist_last);
1114
1115 spa->spa_state = POOL_STATE_UNINITIALIZED;
1116
1117 mutex_enter(&spa->spa_proc_lock);
1118 if (spa->spa_proc_state != SPA_PROC_NONE) {
1119 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1120 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1121 cv_broadcast(&spa->spa_proc_cv);
1122 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1123 ASSERT(spa->spa_proc != &p0);
1124 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1125 }
1126 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1127 spa->spa_proc_state = SPA_PROC_NONE;
1128 }
1129 ASSERT(spa->spa_proc == &p0);
1130 mutex_exit(&spa->spa_proc_lock);
1131
1132 /*
1133 * We want to make sure spa_thread() has actually exited the ZFS
1134 * module, so that the module can't be unloaded out from underneath
1135 * it.
1136 */
1137 if (spa->spa_did != 0) {
1138 thread_join(spa->spa_did);
1139 spa->spa_did = 0;
1140 }
1141 }
1142
1143 /*
1144 * Verify a pool configuration, and construct the vdev tree appropriately. This
1145 * will create all the necessary vdevs in the appropriate layout, with each vdev
1146 * in the CLOSED state. This will prep the pool before open/creation/import.
1147 * All vdev validation is done by the vdev_alloc() routine.
1148 */
1149 static int
1150 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1151 uint_t id, int atype)
1152 {
1153 nvlist_t **child;
1154 uint_t children;
1155 int error;
1156
1157 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1158 return (error);
1159
1160 if ((*vdp)->vdev_ops->vdev_op_leaf)
1161 return (0);
1162
1163 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1164 &child, &children);
1165
1166 if (error == ENOENT)
1167 return (0);
1168
1169 if (error) {
1170 vdev_free(*vdp);
1171 *vdp = NULL;
1172 return (SET_ERROR(EINVAL));
1173 }
1174
1175 for (int c = 0; c < children; c++) {
1176 vdev_t *vd;
1177 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1178 atype)) != 0) {
1179 vdev_free(*vdp);
1180 *vdp = NULL;
1181 return (error);
1182 }
1183 }
1184
1185 ASSERT(*vdp != NULL);
1186
1187 return (0);
1188 }
1189
1190 /*
1191 * Opposite of spa_load().
1192 */
1193 static void
1194 spa_unload(spa_t *spa)
1195 {
1196 int i;
1197
1198 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1199
1200 /*
1201 * Stop async tasks.
1202 */
1203 spa_async_suspend(spa);
1204
1205 /*
1206 * Stop syncing.
1207 */
1208 if (spa->spa_sync_on) {
1209 txg_sync_stop(spa->spa_dsl_pool);
1210 spa->spa_sync_on = B_FALSE;
1211 }
1212
1213 /*
1214 * Wait for any outstanding async I/O to complete.
1215 */
1216 if (spa->spa_async_zio_root != NULL) {
1217 (void) zio_wait(spa->spa_async_zio_root);
1218 spa->spa_async_zio_root = NULL;
1219 }
1220
1221 bpobj_close(&spa->spa_deferred_bpobj);
1222
1223 /*
1224 * Close the dsl pool.
1225 */
1226 if (spa->spa_dsl_pool) {
1227 dsl_pool_close(spa->spa_dsl_pool);
1228 spa->spa_dsl_pool = NULL;
1229 spa->spa_meta_objset = NULL;
1230 }
1231
1232 ddt_unload(spa);
1233
1234 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1235
1236 /*
1237 * Drop and purge level 2 cache
1238 */
1239 spa_l2cache_drop(spa);
1240
1241 /*
1242 * Close all vdevs.
1243 */
1244 if (spa->spa_root_vdev)
1245 vdev_free(spa->spa_root_vdev);
1246 ASSERT(spa->spa_root_vdev == NULL);
1247
1248 for (i = 0; i < spa->spa_spares.sav_count; i++)
1249 vdev_free(spa->spa_spares.sav_vdevs[i]);
1250 if (spa->spa_spares.sav_vdevs) {
1251 kmem_free(spa->spa_spares.sav_vdevs,
1252 spa->spa_spares.sav_count * sizeof (void *));
1253 spa->spa_spares.sav_vdevs = NULL;
1254 }
1255 if (spa->spa_spares.sav_config) {
1256 nvlist_free(spa->spa_spares.sav_config);
1257 spa->spa_spares.sav_config = NULL;
1258 }
1259 spa->spa_spares.sav_count = 0;
1260
1261 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1262 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1263 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1264 }
1265 if (spa->spa_l2cache.sav_vdevs) {
1266 kmem_free(spa->spa_l2cache.sav_vdevs,
1267 spa->spa_l2cache.sav_count * sizeof (void *));
1268 spa->spa_l2cache.sav_vdevs = NULL;
1269 }
1270 if (spa->spa_l2cache.sav_config) {
1271 nvlist_free(spa->spa_l2cache.sav_config);
1272 spa->spa_l2cache.sav_config = NULL;
1273 }
1274 spa->spa_l2cache.sav_count = 0;
1275
1276 spa->spa_async_suspended = 0;
1277
1278 if (spa->spa_comment != NULL) {
1279 spa_strfree(spa->spa_comment);
1280 spa->spa_comment = NULL;
1281 }
1282
1283 spa_config_exit(spa, SCL_ALL, FTAG);
1284 }
1285
1286 /*
1287 * Load (or re-load) the current list of vdevs describing the active spares for
1288 * this pool. When this is called, we have some form of basic information in
1289 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1290 * then re-generate a more complete list including status information.
1291 */
1292 static void
1293 spa_load_spares(spa_t *spa)
1294 {
1295 nvlist_t **spares;
1296 uint_t nspares;
1297 int i;
1298 vdev_t *vd, *tvd;
1299
1300 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1301
1302 /*
1303 * First, close and free any existing spare vdevs.
1304 */
1305 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1306 vd = spa->spa_spares.sav_vdevs[i];
1307
1308 /* Undo the call to spa_activate() below */
1309 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1310 B_FALSE)) != NULL && tvd->vdev_isspare)
1311 spa_spare_remove(tvd);
1312 vdev_close(vd);
1313 vdev_free(vd);
1314 }
1315
1316 if (spa->spa_spares.sav_vdevs)
1317 kmem_free(spa->spa_spares.sav_vdevs,
1318 spa->spa_spares.sav_count * sizeof (void *));
1319
1320 if (spa->spa_spares.sav_config == NULL)
1321 nspares = 0;
1322 else
1323 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1324 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1325
1326 spa->spa_spares.sav_count = (int)nspares;
1327 spa->spa_spares.sav_vdevs = NULL;
1328
1329 if (nspares == 0)
1330 return;
1331
1332 /*
1333 * Construct the array of vdevs, opening them to get status in the
1334 * process. For each spare, there is potentially two different vdev_t
1335 * structures associated with it: one in the list of spares (used only
1336 * for basic validation purposes) and one in the active vdev
1337 * configuration (if it's spared in). During this phase we open and
1338 * validate each vdev on the spare list. If the vdev also exists in the
1339 * active configuration, then we also mark this vdev as an active spare.
1340 */
1341 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1342 KM_SLEEP);
1343 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1344 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1345 VDEV_ALLOC_SPARE) == 0);
1346 ASSERT(vd != NULL);
1347
1348 spa->spa_spares.sav_vdevs[i] = vd;
1349
1350 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1351 B_FALSE)) != NULL) {
1352 if (!tvd->vdev_isspare)
1353 spa_spare_add(tvd);
1354
1355 /*
1356 * We only mark the spare active if we were successfully
1357 * able to load the vdev. Otherwise, importing a pool
1358 * with a bad active spare would result in strange
1359 * behavior, because multiple pool would think the spare
1360 * is actively in use.
1361 *
1362 * There is a vulnerability here to an equally bizarre
1363 * circumstance, where a dead active spare is later
1364 * brought back to life (onlined or otherwise). Given
1365 * the rarity of this scenario, and the extra complexity
1366 * it adds, we ignore the possibility.
1367 */
1368 if (!vdev_is_dead(tvd))
1369 spa_spare_activate(tvd);
1370 }
1371
1372 vd->vdev_top = vd;
1373 vd->vdev_aux = &spa->spa_spares;
1374
1375 if (vdev_open(vd) != 0)
1376 continue;
1377
1378 if (vdev_validate_aux(vd) == 0)
1379 spa_spare_add(vd);
1380 }
1381
1382 /*
1383 * Recompute the stashed list of spares, with status information
1384 * this time.
1385 */
1386 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1387 DATA_TYPE_NVLIST_ARRAY) == 0);
1388
1389 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1390 KM_SLEEP);
1391 for (i = 0; i < spa->spa_spares.sav_count; i++)
1392 spares[i] = vdev_config_generate(spa,
1393 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1394 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1395 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1396 for (i = 0; i < spa->spa_spares.sav_count; i++)
1397 nvlist_free(spares[i]);
1398 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1399 }
1400
1401 /*
1402 * Load (or re-load) the current list of vdevs describing the active l2cache for
1403 * this pool. When this is called, we have some form of basic information in
1404 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1405 * then re-generate a more complete list including status information.
1406 * Devices which are already active have their details maintained, and are
1407 * not re-opened.
1408 */
1409 static void
1410 spa_load_l2cache(spa_t *spa)
1411 {
1412 nvlist_t **l2cache;
1413 uint_t nl2cache;
1414 int i, j, oldnvdevs;
1415 uint64_t guid;
1416 vdev_t *vd, **oldvdevs, **newvdevs;
1417 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1418
1419 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1420
1421 if (sav->sav_config != NULL) {
1422 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1423 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1424 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1425 } else {
1426 nl2cache = 0;
1427 newvdevs = NULL;
1428 }
1429
1430 oldvdevs = sav->sav_vdevs;
1431 oldnvdevs = sav->sav_count;
1432 sav->sav_vdevs = NULL;
1433 sav->sav_count = 0;
1434
1435 /*
1436 * Process new nvlist of vdevs.
1437 */
1438 for (i = 0; i < nl2cache; i++) {
1439 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1440 &guid) == 0);
1441
1442 newvdevs[i] = NULL;
1443 for (j = 0; j < oldnvdevs; j++) {
1444 vd = oldvdevs[j];
1445 if (vd != NULL && guid == vd->vdev_guid) {
1446 /*
1447 * Retain previous vdev for add/remove ops.
1448 */
1449 newvdevs[i] = vd;
1450 oldvdevs[j] = NULL;
1451 break;
1452 }
1453 }
1454
1455 if (newvdevs[i] == NULL) {
1456 /*
1457 * Create new vdev
1458 */
1459 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1460 VDEV_ALLOC_L2CACHE) == 0);
1461 ASSERT(vd != NULL);
1462 newvdevs[i] = vd;
1463
1464 /*
1465 * Commit this vdev as an l2cache device,
1466 * even if it fails to open.
1467 */
1468 spa_l2cache_add(vd);
1469
1470 vd->vdev_top = vd;
1471 vd->vdev_aux = sav;
1472
1473 spa_l2cache_activate(vd);
1474
1475 if (vdev_open(vd) != 0)
1476 continue;
1477
1478 (void) vdev_validate_aux(vd);
1479
1480 if (!vdev_is_dead(vd))
1481 l2arc_add_vdev(spa, vd);
1482 }
1483 }
1484
1485 /*
1486 * Purge vdevs that were dropped
1487 */
1488 for (i = 0; i < oldnvdevs; i++) {
1489 uint64_t pool;
1490
1491 vd = oldvdevs[i];
1492 if (vd != NULL) {
1493 ASSERT(vd->vdev_isl2cache);
1494
1495 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1496 pool != 0ULL && l2arc_vdev_present(vd))
1497 l2arc_remove_vdev(vd);
1498 vdev_clear_stats(vd);
1499 vdev_free(vd);
1500 }
1501 }
1502
1503 if (oldvdevs)
1504 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1505
1506 if (sav->sav_config == NULL)
1507 goto out;
1508
1509 sav->sav_vdevs = newvdevs;
1510 sav->sav_count = (int)nl2cache;
1511
1512 /*
1513 * Recompute the stashed list of l2cache devices, with status
1514 * information this time.
1515 */
1516 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1517 DATA_TYPE_NVLIST_ARRAY) == 0);
1518
1519 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1520 for (i = 0; i < sav->sav_count; i++)
1521 l2cache[i] = vdev_config_generate(spa,
1522 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1523 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1524 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1525 out:
1526 for (i = 0; i < sav->sav_count; i++)
1527 nvlist_free(l2cache[i]);
1528 if (sav->sav_count)
1529 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1530 }
1531
1532 static int
1533 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1534 {
1535 dmu_buf_t *db;
1536 char *packed = NULL;
1537 size_t nvsize = 0;
1538 int error;
1539 *value = NULL;
1540
1541 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1542 nvsize = *(uint64_t *)db->db_data;
1543 dmu_buf_rele(db, FTAG);
1544
1545 packed = kmem_alloc(nvsize, KM_SLEEP);
1546 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1547 DMU_READ_PREFETCH);
1548 if (error == 0)
1549 error = nvlist_unpack(packed, nvsize, value, 0);
1550 kmem_free(packed, nvsize);
1551
1552 return (error);
1553 }
1554
1555 /*
1556 * Checks to see if the given vdev could not be opened, in which case we post a
1557 * sysevent to notify the autoreplace code that the device has been removed.
1558 */
1559 static void
1560 spa_check_removed(vdev_t *vd)
1561 {
1562 for (int c = 0; c < vd->vdev_children; c++)
1563 spa_check_removed(vd->vdev_child[c]);
1564
1565 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1566 !vd->vdev_ishole) {
1567 zfs_post_autoreplace(vd->vdev_spa, vd);
1568 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1569 }
1570 }
1571
1572 /*
1573 * Validate the current config against the MOS config
1574 */
1575 static boolean_t
1576 spa_config_valid(spa_t *spa, nvlist_t *config)
1577 {
1578 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1579 nvlist_t *nv;
1580
1581 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1582
1583 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1584 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1585
1586 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1587
1588 /*
1589 * If we're doing a normal import, then build up any additional
1590 * diagnostic information about missing devices in this config.
1591 * We'll pass this up to the user for further processing.
1592 */
1593 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1594 nvlist_t **child, *nv;
1595 uint64_t idx = 0;
1596
1597 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1598 KM_SLEEP);
1599 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1600
1601 for (int c = 0; c < rvd->vdev_children; c++) {
1602 vdev_t *tvd = rvd->vdev_child[c];
1603 vdev_t *mtvd = mrvd->vdev_child[c];
1604
1605 if (tvd->vdev_ops == &vdev_missing_ops &&
1606 mtvd->vdev_ops != &vdev_missing_ops &&
1607 mtvd->vdev_islog)
1608 child[idx++] = vdev_config_generate(spa, mtvd,
1609 B_FALSE, 0);
1610 }
1611
1612 if (idx) {
1613 VERIFY(nvlist_add_nvlist_array(nv,
1614 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1615 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1616 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1617
1618 for (int i = 0; i < idx; i++)
1619 nvlist_free(child[i]);
1620 }
1621 nvlist_free(nv);
1622 kmem_free(child, rvd->vdev_children * sizeof (char **));
1623 }
1624
1625 /*
1626 * Compare the root vdev tree with the information we have
1627 * from the MOS config (mrvd). Check each top-level vdev
1628 * with the corresponding MOS config top-level (mtvd).
1629 */
1630 for (int c = 0; c < rvd->vdev_children; c++) {
1631 vdev_t *tvd = rvd->vdev_child[c];
1632 vdev_t *mtvd = mrvd->vdev_child[c];
1633
1634 /*
1635 * Resolve any "missing" vdevs in the current configuration.
1636 * If we find that the MOS config has more accurate information
1637 * about the top-level vdev then use that vdev instead.
1638 */
1639 if (tvd->vdev_ops == &vdev_missing_ops &&
1640 mtvd->vdev_ops != &vdev_missing_ops) {
1641
1642 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1643 continue;
1644
1645 /*
1646 * Device specific actions.
1647 */
1648 if (mtvd->vdev_islog) {
1649 spa_set_log_state(spa, SPA_LOG_CLEAR);
1650 } else {
1651 /*
1652 * XXX - once we have 'readonly' pool
1653 * support we should be able to handle
1654 * missing data devices by transitioning
1655 * the pool to readonly.
1656 */
1657 continue;
1658 }
1659
1660 /*
1661 * Swap the missing vdev with the data we were
1662 * able to obtain from the MOS config.
1663 */
1664 vdev_remove_child(rvd, tvd);
1665 vdev_remove_child(mrvd, mtvd);
1666
1667 vdev_add_child(rvd, mtvd);
1668 vdev_add_child(mrvd, tvd);
1669
1670 spa_config_exit(spa, SCL_ALL, FTAG);
1671 vdev_load(mtvd);
1672 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1673
1674 vdev_reopen(rvd);
1675 } else if (mtvd->vdev_islog) {
1676 /*
1677 * Load the slog device's state from the MOS config
1678 * since it's possible that the label does not
1679 * contain the most up-to-date information.
1680 */
1681 vdev_load_log_state(tvd, mtvd);
1682 vdev_reopen(tvd);
1683 }
1684 }
1685 vdev_free(mrvd);
1686 spa_config_exit(spa, SCL_ALL, FTAG);
1687
1688 /*
1689 * Ensure we were able to validate the config.
1690 */
1691 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1692 }
1693
1694 /*
1695 * Check for missing log devices
1696 */
1697 static boolean_t
1698 spa_check_logs(spa_t *spa)
1699 {
1700 boolean_t rv = B_FALSE;
1701
1702 switch (spa->spa_log_state) {
1703 case SPA_LOG_MISSING:
1704 /* need to recheck in case slog has been restored */
1705 case SPA_LOG_UNKNOWN:
1706 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1707 NULL, DS_FIND_CHILDREN) != 0);
1708 if (rv)
1709 spa_set_log_state(spa, SPA_LOG_MISSING);
1710 break;
1711 }
1712 return (rv);
1713 }
1714
1715 static boolean_t
1716 spa_passivate_log(spa_t *spa)
1717 {
1718 vdev_t *rvd = spa->spa_root_vdev;
1719 boolean_t slog_found = B_FALSE;
1720
1721 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1722
1723 if (!spa_has_slogs(spa))
1724 return (B_FALSE);
1725
1726 for (int c = 0; c < rvd->vdev_children; c++) {
1727 vdev_t *tvd = rvd->vdev_child[c];
1728 metaslab_group_t *mg = tvd->vdev_mg;
1729
1730 if (tvd->vdev_islog) {
1731 metaslab_group_passivate(mg);
1732 slog_found = B_TRUE;
1733 }
1734 }
1735
1736 return (slog_found);
1737 }
1738
1739 static void
1740 spa_activate_log(spa_t *spa)
1741 {
1742 vdev_t *rvd = spa->spa_root_vdev;
1743
1744 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1745
1746 for (int c = 0; c < rvd->vdev_children; c++) {
1747 vdev_t *tvd = rvd->vdev_child[c];
1748 metaslab_group_t *mg = tvd->vdev_mg;
1749
1750 if (tvd->vdev_islog)
1751 metaslab_group_activate(mg);
1752 }
1753 }
1754
1755 int
1756 spa_offline_log(spa_t *spa)
1757 {
1758 int error;
1759
1760 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1761 NULL, DS_FIND_CHILDREN);
1762 if (error == 0) {
1763 /*
1764 * We successfully offlined the log device, sync out the
1765 * current txg so that the "stubby" block can be removed
1766 * by zil_sync().
1767 */
1768 txg_wait_synced(spa->spa_dsl_pool, 0);
1769 }
1770 return (error);
1771 }
1772
1773 static void
1774 spa_aux_check_removed(spa_aux_vdev_t *sav)
1775 {
1776 for (int i = 0; i < sav->sav_count; i++)
1777 spa_check_removed(sav->sav_vdevs[i]);
1778 }
1779
1780 void
1781 spa_claim_notify(zio_t *zio)
1782 {
1783 spa_t *spa = zio->io_spa;
1784
1785 if (zio->io_error)
1786 return;
1787
1788 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1789 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1790 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1791 mutex_exit(&spa->spa_props_lock);
1792 }
1793
1794 typedef struct spa_load_error {
1795 uint64_t sle_meta_count;
1796 uint64_t sle_data_count;
1797 } spa_load_error_t;
1798
1799 static void
1800 spa_load_verify_done(zio_t *zio)
1801 {
1802 blkptr_t *bp = zio->io_bp;
1803 spa_load_error_t *sle = zio->io_private;
1804 dmu_object_type_t type = BP_GET_TYPE(bp);
1805 int error = zio->io_error;
1806
1807 if (error) {
1808 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1809 type != DMU_OT_INTENT_LOG)
1810 atomic_add_64(&sle->sle_meta_count, 1);
1811 else
1812 atomic_add_64(&sle->sle_data_count, 1);
1813 }
1814 zio_data_buf_free(zio->io_data, zio->io_size);
1815 }
1816
1817 /*ARGSUSED*/
1818 static int
1819 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1820 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1821 {
1822 if (bp != NULL) {
1823 zio_t *rio = arg;
1824 size_t size = BP_GET_PSIZE(bp);
1825 void *data = zio_data_buf_alloc(size);
1826
1827 zio_nowait(zio_read(rio, spa, bp, data, size,
1828 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1829 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1830 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1831 }
1832 return (0);
1833 }
1834
1835 static int
1836 spa_load_verify(spa_t *spa)
1837 {
1838 zio_t *rio;
1839 spa_load_error_t sle = { 0 };
1840 zpool_rewind_policy_t policy;
1841 boolean_t verify_ok = B_FALSE;
1842 int error;
1843
1844 zpool_get_rewind_policy(spa->spa_config, &policy);
1845
1846 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1847 return (0);
1848
1849 rio = zio_root(spa, NULL, &sle,
1850 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1851
1852 error = traverse_pool(spa, spa->spa_verify_min_txg,
1853 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1854
1855 (void) zio_wait(rio);
1856
1857 spa->spa_load_meta_errors = sle.sle_meta_count;
1858 spa->spa_load_data_errors = sle.sle_data_count;
1859
1860 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1861 sle.sle_data_count <= policy.zrp_maxdata) {
1862 int64_t loss = 0;
1863
1864 verify_ok = B_TRUE;
1865 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1866 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1867
1868 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1869 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1870 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1871 VERIFY(nvlist_add_int64(spa->spa_load_info,
1872 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1873 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1874 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1875 } else {
1876 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1877 }
1878
1879 if (error) {
1880 if (error != ENXIO && error != EIO)
1881 error = SET_ERROR(EIO);
1882 return (error);
1883 }
1884
1885 return (verify_ok ? 0 : EIO);
1886 }
1887
1888 /*
1889 * Find a value in the pool props object.
1890 */
1891 static void
1892 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1893 {
1894 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1895 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1896 }
1897
1898 /*
1899 * Find a value in the pool directory object.
1900 */
1901 static int
1902 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1903 {
1904 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1905 name, sizeof (uint64_t), 1, val));
1906 }
1907
1908 static int
1909 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1910 {
1911 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1912 return (err);
1913 }
1914
1915 /*
1916 * Fix up config after a partly-completed split. This is done with the
1917 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1918 * pool have that entry in their config, but only the splitting one contains
1919 * a list of all the guids of the vdevs that are being split off.
1920 *
1921 * This function determines what to do with that list: either rejoin
1922 * all the disks to the pool, or complete the splitting process. To attempt
1923 * the rejoin, each disk that is offlined is marked online again, and
1924 * we do a reopen() call. If the vdev label for every disk that was
1925 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1926 * then we call vdev_split() on each disk, and complete the split.
1927 *
1928 * Otherwise we leave the config alone, with all the vdevs in place in
1929 * the original pool.
1930 */
1931 static void
1932 spa_try_repair(spa_t *spa, nvlist_t *config)
1933 {
1934 uint_t extracted;
1935 uint64_t *glist;
1936 uint_t i, gcount;
1937 nvlist_t *nvl;
1938 vdev_t **vd;
1939 boolean_t attempt_reopen;
1940
1941 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1942 return;
1943
1944 /* check that the config is complete */
1945 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1946 &glist, &gcount) != 0)
1947 return;
1948
1949 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1950
1951 /* attempt to online all the vdevs & validate */
1952 attempt_reopen = B_TRUE;
1953 for (i = 0; i < gcount; i++) {
1954 if (glist[i] == 0) /* vdev is hole */
1955 continue;
1956
1957 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1958 if (vd[i] == NULL) {
1959 /*
1960 * Don't bother attempting to reopen the disks;
1961 * just do the split.
1962 */
1963 attempt_reopen = B_FALSE;
1964 } else {
1965 /* attempt to re-online it */
1966 vd[i]->vdev_offline = B_FALSE;
1967 }
1968 }
1969
1970 if (attempt_reopen) {
1971 vdev_reopen(spa->spa_root_vdev);
1972
1973 /* check each device to see what state it's in */
1974 for (extracted = 0, i = 0; i < gcount; i++) {
1975 if (vd[i] != NULL &&
1976 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1977 break;
1978 ++extracted;
1979 }
1980 }
1981
1982 /*
1983 * If every disk has been moved to the new pool, or if we never
1984 * even attempted to look at them, then we split them off for
1985 * good.
1986 */
1987 if (!attempt_reopen || gcount == extracted) {
1988 for (i = 0; i < gcount; i++)
1989 if (vd[i] != NULL)
1990 vdev_split(vd[i]);
1991 vdev_reopen(spa->spa_root_vdev);
1992 }
1993
1994 kmem_free(vd, gcount * sizeof (vdev_t *));
1995 }
1996
1997 static int
1998 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1999 boolean_t mosconfig)
2000 {
2001 nvlist_t *config = spa->spa_config;
2002 char *ereport = FM_EREPORT_ZFS_POOL;
2003 char *comment;
2004 int error;
2005 uint64_t pool_guid;
2006 nvlist_t *nvl;
2007
2008 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2009 return (SET_ERROR(EINVAL));
2010
2011 ASSERT(spa->spa_comment == NULL);
2012 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2013 spa->spa_comment = spa_strdup(comment);
2014
2015 /*
2016 * Versioning wasn't explicitly added to the label until later, so if
2017 * it's not present treat it as the initial version.
2018 */
2019 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2020 &spa->spa_ubsync.ub_version) != 0)
2021 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2022
2023 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2024 &spa->spa_config_txg);
2025
2026 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2027 spa_guid_exists(pool_guid, 0)) {
2028 error = SET_ERROR(EEXIST);
2029 } else {
2030 spa->spa_config_guid = pool_guid;
2031
2032 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2033 &nvl) == 0) {
2034 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2035 KM_SLEEP) == 0);
2036 }
2037
2038 nvlist_free(spa->spa_load_info);
2039 spa->spa_load_info = fnvlist_alloc();
2040
2041 gethrestime(&spa->spa_loaded_ts);
2042 error = spa_load_impl(spa, pool_guid, config, state, type,
2043 mosconfig, &ereport);
2044 }
2045
2046 spa->spa_minref = refcount_count(&spa->spa_refcount);
2047 if (error) {
2048 if (error != EEXIST) {
2049 spa->spa_loaded_ts.tv_sec = 0;
2050 spa->spa_loaded_ts.tv_nsec = 0;
2051 }
2052 if (error != EBADF) {
2053 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2054 }
2055 }
2056 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2057 spa->spa_ena = 0;
2058
2059 return (error);
2060 }
2061
2062 /*
2063 * Load an existing storage pool, using the pool's builtin spa_config as a
2064 * source of configuration information.
2065 */
2066 static int
2067 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2068 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2069 char **ereport)
2070 {
2071 int error = 0;
2072 nvlist_t *nvroot = NULL;
2073 nvlist_t *label;
2074 vdev_t *rvd;
2075 uberblock_t *ub = &spa->spa_uberblock;
2076 uint64_t children, config_cache_txg = spa->spa_config_txg;
2077 int orig_mode = spa->spa_mode;
2078 int parse;
2079 uint64_t obj;
2080 boolean_t missing_feat_write = B_FALSE;
2081
2082 /*
2083 * If this is an untrusted config, access the pool in read-only mode.
2084 * This prevents things like resilvering recently removed devices.
2085 */
2086 if (!mosconfig)
2087 spa->spa_mode = FREAD;
2088
2089 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2090
2091 spa->spa_load_state = state;
2092
2093 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2094 return (SET_ERROR(EINVAL));
2095
2096 parse = (type == SPA_IMPORT_EXISTING ?
2097 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2098
2099 /*
2100 * Create "The Godfather" zio to hold all async IOs
2101 */
2102 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2103 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2104
2105 /*
2106 * Parse the configuration into a vdev tree. We explicitly set the
2107 * value that will be returned by spa_version() since parsing the
2108 * configuration requires knowing the version number.
2109 */
2110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2111 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2112 spa_config_exit(spa, SCL_ALL, FTAG);
2113
2114 if (error != 0)
2115 return (error);
2116
2117 ASSERT(spa->spa_root_vdev == rvd);
2118
2119 if (type != SPA_IMPORT_ASSEMBLE) {
2120 ASSERT(spa_guid(spa) == pool_guid);
2121 }
2122
2123 /*
2124 * Try to open all vdevs, loading each label in the process.
2125 */
2126 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2127 error = vdev_open(rvd);
2128 spa_config_exit(spa, SCL_ALL, FTAG);
2129 if (error != 0)
2130 return (error);
2131
2132 /*
2133 * We need to validate the vdev labels against the configuration that
2134 * we have in hand, which is dependent on the setting of mosconfig. If
2135 * mosconfig is true then we're validating the vdev labels based on
2136 * that config. Otherwise, we're validating against the cached config
2137 * (zpool.cache) that was read when we loaded the zfs module, and then
2138 * later we will recursively call spa_load() and validate against
2139 * the vdev config.
2140 *
2141 * If we're assembling a new pool that's been split off from an
2142 * existing pool, the labels haven't yet been updated so we skip
2143 * validation for now.
2144 */
2145 if (type != SPA_IMPORT_ASSEMBLE) {
2146 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2147 error = vdev_validate(rvd, mosconfig);
2148 spa_config_exit(spa, SCL_ALL, FTAG);
2149
2150 if (error != 0)
2151 return (error);
2152
2153 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2154 return (SET_ERROR(ENXIO));
2155 }
2156
2157 /*
2158 * Find the best uberblock.
2159 */
2160 vdev_uberblock_load(rvd, ub, &label);
2161
2162 /*
2163 * If we weren't able to find a single valid uberblock, return failure.
2164 */
2165 if (ub->ub_txg == 0) {
2166 nvlist_free(label);
2167 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2168 }
2169
2170 /*
2171 * If the pool has an unsupported version we can't open it.
2172 */
2173 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2174 nvlist_free(label);
2175 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2176 }
2177
2178 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2179 nvlist_t *features;
2180
2181 /*
2182 * If we weren't able to find what's necessary for reading the
2183 * MOS in the label, return failure.
2184 */
2185 if (label == NULL || nvlist_lookup_nvlist(label,
2186 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2187 nvlist_free(label);
2188 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2189 ENXIO));
2190 }
2191
2192 /*
2193 * Update our in-core representation with the definitive values
2194 * from the label.
2195 */
2196 nvlist_free(spa->spa_label_features);
2197 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2198 }
2199
2200 nvlist_free(label);
2201
2202 /*
2203 * Look through entries in the label nvlist's features_for_read. If
2204 * there is a feature listed there which we don't understand then we
2205 * cannot open a pool.
2206 */
2207 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2208 nvlist_t *unsup_feat;
2209
2210 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2211 0);
2212
2213 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2214 NULL); nvp != NULL;
2215 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2216 if (!zfeature_is_supported(nvpair_name(nvp))) {
2217 VERIFY(nvlist_add_string(unsup_feat,
2218 nvpair_name(nvp), "") == 0);
2219 }
2220 }
2221
2222 if (!nvlist_empty(unsup_feat)) {
2223 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2224 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2225 nvlist_free(unsup_feat);
2226 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2227 ENOTSUP));
2228 }
2229
2230 nvlist_free(unsup_feat);
2231 }
2232
2233 /*
2234 * If the vdev guid sum doesn't match the uberblock, we have an
2235 * incomplete configuration. We first check to see if the pool
2236 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2237 * If it is, defer the vdev_guid_sum check till later so we
2238 * can handle missing vdevs.
2239 */
2240 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2241 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2242 rvd->vdev_guid_sum != ub->ub_guid_sum)
2243 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2244
2245 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2246 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2247 spa_try_repair(spa, config);
2248 spa_config_exit(spa, SCL_ALL, FTAG);
2249 nvlist_free(spa->spa_config_splitting);
2250 spa->spa_config_splitting = NULL;
2251 }
2252
2253 /*
2254 * Initialize internal SPA structures.
2255 */
2256 spa->spa_state = POOL_STATE_ACTIVE;
2257 spa->spa_ubsync = spa->spa_uberblock;
2258 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2259 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2260 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2261 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2262 spa->spa_claim_max_txg = spa->spa_first_txg;
2263 spa->spa_prev_software_version = ub->ub_software_version;
2264
2265 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2266 if (error)
2267 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2268 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2269
2270 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2272
2273 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2274 boolean_t missing_feat_read = B_FALSE;
2275 nvlist_t *unsup_feat, *enabled_feat;
2276
2277 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2278 &spa->spa_feat_for_read_obj) != 0) {
2279 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280 }
2281
2282 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2283 &spa->spa_feat_for_write_obj) != 0) {
2284 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2285 }
2286
2287 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2288 &spa->spa_feat_desc_obj) != 0) {
2289 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2290 }
2291
2292 enabled_feat = fnvlist_alloc();
2293 unsup_feat = fnvlist_alloc();
2294
2295 if (!feature_is_supported(spa->spa_meta_objset,
2296 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2297 unsup_feat, enabled_feat))
2298 missing_feat_read = B_TRUE;
2299
2300 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2301 if (!feature_is_supported(spa->spa_meta_objset,
2302 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2303 unsup_feat, enabled_feat)) {
2304 missing_feat_write = B_TRUE;
2305 }
2306 }
2307
2308 fnvlist_add_nvlist(spa->spa_load_info,
2309 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2310
2311 if (!nvlist_empty(unsup_feat)) {
2312 fnvlist_add_nvlist(spa->spa_load_info,
2313 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2314 }
2315
2316 fnvlist_free(enabled_feat);
2317 fnvlist_free(unsup_feat);
2318
2319 if (!missing_feat_read) {
2320 fnvlist_add_boolean(spa->spa_load_info,
2321 ZPOOL_CONFIG_CAN_RDONLY);
2322 }
2323
2324 /*
2325 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2326 * twofold: to determine whether the pool is available for
2327 * import in read-write mode and (if it is not) whether the
2328 * pool is available for import in read-only mode. If the pool
2329 * is available for import in read-write mode, it is displayed
2330 * as available in userland; if it is not available for import
2331 * in read-only mode, it is displayed as unavailable in
2332 * userland. If the pool is available for import in read-only
2333 * mode but not read-write mode, it is displayed as unavailable
2334 * in userland with a special note that the pool is actually
2335 * available for open in read-only mode.
2336 *
2337 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2338 * missing a feature for write, we must first determine whether
2339 * the pool can be opened read-only before returning to
2340 * userland in order to know whether to display the
2341 * abovementioned note.
2342 */
2343 if (missing_feat_read || (missing_feat_write &&
2344 spa_writeable(spa))) {
2345 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2346 ENOTSUP));
2347 }
2348 }
2349
2350 spa->spa_is_initializing = B_TRUE;
2351 error = dsl_pool_open(spa->spa_dsl_pool);
2352 spa->spa_is_initializing = B_FALSE;
2353 if (error != 0)
2354 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355
2356 if (!mosconfig) {
2357 uint64_t hostid;
2358 nvlist_t *policy = NULL, *nvconfig;
2359
2360 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2361 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2362
2363 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2364 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2365 char *hostname;
2366 unsigned long myhostid = 0;
2367
2368 VERIFY(nvlist_lookup_string(nvconfig,
2369 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2370
2371 #ifdef _KERNEL
2372 myhostid = zone_get_hostid(NULL);
2373 #else /* _KERNEL */
2374 /*
2375 * We're emulating the system's hostid in userland, so
2376 * we can't use zone_get_hostid().
2377 */
2378 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2379 #endif /* _KERNEL */
2380 if (hostid != 0 && myhostid != 0 &&
2381 hostid != myhostid) {
2382 nvlist_free(nvconfig);
2383 cmn_err(CE_WARN, "pool '%s' could not be "
2384 "loaded as it was last accessed by "
2385 "another system (host: %s hostid: 0x%lx). "
2386 "See: http://illumos.org/msg/ZFS-8000-EY",
2387 spa_name(spa), hostname,
2388 (unsigned long)hostid);
2389 return (SET_ERROR(EBADF));
2390 }
2391 }
2392 if (nvlist_lookup_nvlist(spa->spa_config,
2393 ZPOOL_REWIND_POLICY, &policy) == 0)
2394 VERIFY(nvlist_add_nvlist(nvconfig,
2395 ZPOOL_REWIND_POLICY, policy) == 0);
2396
2397 spa_config_set(spa, nvconfig);
2398 spa_unload(spa);
2399 spa_deactivate(spa);
2400 spa_activate(spa, orig_mode);
2401
2402 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2403 }
2404
2405 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2406 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2407 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2408 if (error != 0)
2409 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2410
2411 /*
2412 * Load the bit that tells us to use the new accounting function
2413 * (raid-z deflation). If we have an older pool, this will not
2414 * be present.
2415 */
2416 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2417 if (error != 0 && error != ENOENT)
2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2419
2420 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2421 &spa->spa_creation_version);
2422 if (error != 0 && error != ENOENT)
2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2424
2425 /*
2426 * Load the persistent error log. If we have an older pool, this will
2427 * not be present.
2428 */
2429 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2430 if (error != 0 && error != ENOENT)
2431 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432
2433 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2434 &spa->spa_errlog_scrub);
2435 if (error != 0 && error != ENOENT)
2436 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2437
2438 /*
2439 * Load the history object. If we have an older pool, this
2440 * will not be present.
2441 */
2442 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2443 if (error != 0 && error != ENOENT)
2444 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445
2446 /*
2447 * If we're assembling the pool from the split-off vdevs of
2448 * an existing pool, we don't want to attach the spares & cache
2449 * devices.
2450 */
2451
2452 /*
2453 * Load any hot spares for this pool.
2454 */
2455 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2456 if (error != 0 && error != ENOENT)
2457 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2458 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2459 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2460 if (load_nvlist(spa, spa->spa_spares.sav_object,
2461 &spa->spa_spares.sav_config) != 0)
2462 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2463
2464 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2465 spa_load_spares(spa);
2466 spa_config_exit(spa, SCL_ALL, FTAG);
2467 } else if (error == 0) {
2468 spa->spa_spares.sav_sync = B_TRUE;
2469 }
2470
2471 /*
2472 * Load any level 2 ARC devices for this pool.
2473 */
2474 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2475 &spa->spa_l2cache.sav_object);
2476 if (error != 0 && error != ENOENT)
2477 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2479 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2480 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2481 &spa->spa_l2cache.sav_config) != 0)
2482 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2483
2484 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2485 spa_load_l2cache(spa);
2486 spa_config_exit(spa, SCL_ALL, FTAG);
2487 } else if (error == 0) {
2488 spa->spa_l2cache.sav_sync = B_TRUE;
2489 }
2490
2491 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2492
2493 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2494 if (error && error != ENOENT)
2495 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2496
2497 if (error == 0) {
2498 uint64_t autoreplace;
2499
2500 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2501 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2502 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2503 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2504 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2505 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2506 &spa->spa_dedup_ditto);
2507
2508 spa->spa_autoreplace = (autoreplace != 0);
2509 }
2510
2511 /*
2512 * If the 'autoreplace' property is set, then post a resource notifying
2513 * the ZFS DE that it should not issue any faults for unopenable
2514 * devices. We also iterate over the vdevs, and post a sysevent for any
2515 * unopenable vdevs so that the normal autoreplace handler can take
2516 * over.
2517 */
2518 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2519 spa_check_removed(spa->spa_root_vdev);
2520 /*
2521 * For the import case, this is done in spa_import(), because
2522 * at this point we're using the spare definitions from
2523 * the MOS config, not necessarily from the userland config.
2524 */
2525 if (state != SPA_LOAD_IMPORT) {
2526 spa_aux_check_removed(&spa->spa_spares);
2527 spa_aux_check_removed(&spa->spa_l2cache);
2528 }
2529 }
2530
2531 /*
2532 * Load the vdev state for all toplevel vdevs.
2533 */
2534 vdev_load(rvd);
2535
2536 /*
2537 * Propagate the leaf DTLs we just loaded all the way up the tree.
2538 */
2539 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2540 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2541 spa_config_exit(spa, SCL_ALL, FTAG);
2542
2543 /*
2544 * Load the DDTs (dedup tables).
2545 */
2546 error = ddt_load(spa);
2547 if (error != 0)
2548 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2549
2550 spa_update_dspace(spa);
2551
2552 /*
2553 * Validate the config, using the MOS config to fill in any
2554 * information which might be missing. If we fail to validate
2555 * the config then declare the pool unfit for use. If we're
2556 * assembling a pool from a split, the log is not transferred
2557 * over.
2558 */
2559 if (type != SPA_IMPORT_ASSEMBLE) {
2560 nvlist_t *nvconfig;
2561
2562 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2563 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2564
2565 if (!spa_config_valid(spa, nvconfig)) {
2566 nvlist_free(nvconfig);
2567 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2568 ENXIO));
2569 }
2570 nvlist_free(nvconfig);
2571
2572 /*
2573 * Now that we've validated the config, check the state of the
2574 * root vdev. If it can't be opened, it indicates one or
2575 * more toplevel vdevs are faulted.
2576 */
2577 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2578 return (SET_ERROR(ENXIO));
2579
2580 if (spa_check_logs(spa)) {
2581 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2582 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2583 }
2584 }
2585
2586 if (missing_feat_write) {
2587 ASSERT(state == SPA_LOAD_TRYIMPORT);
2588
2589 /*
2590 * At this point, we know that we can open the pool in
2591 * read-only mode but not read-write mode. We now have enough
2592 * information and can return to userland.
2593 */
2594 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2595 }
2596
2597 /*
2598 * We've successfully opened the pool, verify that we're ready
2599 * to start pushing transactions.
2600 */
2601 if (state != SPA_LOAD_TRYIMPORT) {
2602 if (error = spa_load_verify(spa))
2603 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2604 error));
2605 }
2606
2607 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2608 spa->spa_load_max_txg == UINT64_MAX)) {
2609 dmu_tx_t *tx;
2610 int need_update = B_FALSE;
2611
2612 ASSERT(state != SPA_LOAD_TRYIMPORT);
2613
2614 /*
2615 * Claim log blocks that haven't been committed yet.
2616 * This must all happen in a single txg.
2617 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2618 * invoked from zil_claim_log_block()'s i/o done callback.
2619 * Price of rollback is that we abandon the log.
2620 */
2621 spa->spa_claiming = B_TRUE;
2622
2623 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2624 spa_first_txg(spa));
2625 (void) dmu_objset_find(spa_name(spa),
2626 zil_claim, tx, DS_FIND_CHILDREN);
2627 dmu_tx_commit(tx);
2628
2629 spa->spa_claiming = B_FALSE;
2630
2631 spa_set_log_state(spa, SPA_LOG_GOOD);
2632 spa->spa_sync_on = B_TRUE;
2633 txg_sync_start(spa->spa_dsl_pool);
2634
2635 /*
2636 * Wait for all claims to sync. We sync up to the highest
2637 * claimed log block birth time so that claimed log blocks
2638 * don't appear to be from the future. spa_claim_max_txg
2639 * will have been set for us by either zil_check_log_chain()
2640 * (invoked from spa_check_logs()) or zil_claim() above.
2641 */
2642 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2643
2644 /*
2645 * If the config cache is stale, or we have uninitialized
2646 * metaslabs (see spa_vdev_add()), then update the config.
2647 *
2648 * If this is a verbatim import, trust the current
2649 * in-core spa_config and update the disk labels.
2650 */
2651 if (config_cache_txg != spa->spa_config_txg ||
2652 state == SPA_LOAD_IMPORT ||
2653 state == SPA_LOAD_RECOVER ||
2654 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2655 need_update = B_TRUE;
2656
2657 for (int c = 0; c < rvd->vdev_children; c++)
2658 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2659 need_update = B_TRUE;
2660
2661 /*
2662 * Update the config cache asychronously in case we're the
2663 * root pool, in which case the config cache isn't writable yet.
2664 */
2665 if (need_update)
2666 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2667
2668 /*
2669 * Check all DTLs to see if anything needs resilvering.
2670 */
2671 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2672 vdev_resilver_needed(rvd, NULL, NULL))
2673 spa_async_request(spa, SPA_ASYNC_RESILVER);
2674
2675 /*
2676 * Log the fact that we booted up (so that we can detect if
2677 * we rebooted in the middle of an operation).
2678 */
2679 spa_history_log_version(spa, "open");
2680
2681 /*
2682 * Delete any inconsistent datasets.
2683 */
2684 (void) dmu_objset_find(spa_name(spa),
2685 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2686
2687 /*
2688 * Clean up any stale temporary dataset userrefs.
2689 */
2690 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2691 }
2692
2693 return (0);
2694 }
2695
2696 static int
2697 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2698 {
2699 int mode = spa->spa_mode;
2700
2701 spa_unload(spa);
2702 spa_deactivate(spa);
2703
2704 spa->spa_load_max_txg--;
2705
2706 spa_activate(spa, mode);
2707 spa_async_suspend(spa);
2708
2709 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2710 }
2711
2712 /*
2713 * If spa_load() fails this function will try loading prior txg's. If
2714 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2715 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2716 * function will not rewind the pool and will return the same error as
2717 * spa_load().
2718 */
2719 static int
2720 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2721 uint64_t max_request, int rewind_flags)
2722 {
2723 nvlist_t *loadinfo = NULL;
2724 nvlist_t *config = NULL;
2725 int load_error, rewind_error;
2726 uint64_t safe_rewind_txg;
2727 uint64_t min_txg;
2728
2729 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2730 spa->spa_load_max_txg = spa->spa_load_txg;
2731 spa_set_log_state(spa, SPA_LOG_CLEAR);
2732 } else {
2733 spa->spa_load_max_txg = max_request;
2734 }
2735
2736 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2737 mosconfig);
2738 if (load_error == 0)
2739 return (0);
2740
2741 if (spa->spa_root_vdev != NULL)
2742 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2743
2744 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2745 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2746
2747 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2748 nvlist_free(config);
2749 return (load_error);
2750 }
2751
2752 if (state == SPA_LOAD_RECOVER) {
2753 /* Price of rolling back is discarding txgs, including log */
2754 spa_set_log_state(spa, SPA_LOG_CLEAR);
2755 } else {
2756 /*
2757 * If we aren't rolling back save the load info from our first
2758 * import attempt so that we can restore it after attempting
2759 * to rewind.
2760 */
2761 loadinfo = spa->spa_load_info;
2762 spa->spa_load_info = fnvlist_alloc();
2763 }
2764
2765 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2766 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2767 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2768 TXG_INITIAL : safe_rewind_txg;
2769
2770 /*
2771 * Continue as long as we're finding errors, we're still within
2772 * the acceptable rewind range, and we're still finding uberblocks
2773 */
2774 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2775 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2776 if (spa->spa_load_max_txg < safe_rewind_txg)
2777 spa->spa_extreme_rewind = B_TRUE;
2778 rewind_error = spa_load_retry(spa, state, mosconfig);
2779 }
2780
2781 spa->spa_extreme_rewind = B_FALSE;
2782 spa->spa_load_max_txg = UINT64_MAX;
2783
2784 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2785 spa_config_set(spa, config);
2786
2787 if (state == SPA_LOAD_RECOVER) {
2788 ASSERT3P(loadinfo, ==, NULL);
2789 return (rewind_error);
2790 } else {
2791 /* Store the rewind info as part of the initial load info */
2792 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2793 spa->spa_load_info);
2794
2795 /* Restore the initial load info */
2796 fnvlist_free(spa->spa_load_info);
2797 spa->spa_load_info = loadinfo;
2798
2799 return (load_error);
2800 }
2801 }
2802
2803 /*
2804 * Pool Open/Import
2805 *
2806 * The import case is identical to an open except that the configuration is sent
2807 * down from userland, instead of grabbed from the configuration cache. For the
2808 * case of an open, the pool configuration will exist in the
2809 * POOL_STATE_UNINITIALIZED state.
2810 *
2811 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2812 * the same time open the pool, without having to keep around the spa_t in some
2813 * ambiguous state.
2814 */
2815 static int
2816 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2817 nvlist_t **config)
2818 {
2819 spa_t *spa;
2820 spa_load_state_t state = SPA_LOAD_OPEN;
2821 int error;
2822 int locked = B_FALSE;
2823
2824 *spapp = NULL;
2825
2826 /*
2827 * As disgusting as this is, we need to support recursive calls to this
2828 * function because dsl_dir_open() is called during spa_load(), and ends
2829 * up calling spa_open() again. The real fix is to figure out how to
2830 * avoid dsl_dir_open() calling this in the first place.
2831 */
2832 if (mutex_owner(&spa_namespace_lock) != curthread) {
2833 mutex_enter(&spa_namespace_lock);
2834 locked = B_TRUE;
2835 }
2836
2837 if ((spa = spa_lookup(pool)) == NULL) {
2838 if (locked)
2839 mutex_exit(&spa_namespace_lock);
2840 return (SET_ERROR(ENOENT));
2841 }
2842
2843 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2844 zpool_rewind_policy_t policy;
2845
2846 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2847 &policy);
2848 if (policy.zrp_request & ZPOOL_DO_REWIND)
2849 state = SPA_LOAD_RECOVER;
2850
2851 spa_activate(spa, spa_mode_global);
2852
2853 if (state != SPA_LOAD_RECOVER)
2854 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2855
2856 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2857 policy.zrp_request);
2858
2859 if (error == EBADF) {
2860 /*
2861 * If vdev_validate() returns failure (indicated by
2862 * EBADF), it indicates that one of the vdevs indicates
2863 * that the pool has been exported or destroyed. If
2864 * this is the case, the config cache is out of sync and
2865 * we should remove the pool from the namespace.
2866 */
2867 spa_unload(spa);
2868 spa_deactivate(spa);
2869 spa_config_sync(spa, B_TRUE, B_TRUE);
2870 spa_remove(spa);
2871 if (locked)
2872 mutex_exit(&spa_namespace_lock);
2873 return (SET_ERROR(ENOENT));
2874 }
2875
2876 if (error) {
2877 /*
2878 * We can't open the pool, but we still have useful
2879 * information: the state of each vdev after the
2880 * attempted vdev_open(). Return this to the user.
2881 */
2882 if (config != NULL && spa->spa_config) {
2883 VERIFY(nvlist_dup(spa->spa_config, config,
2884 KM_SLEEP) == 0);
2885 VERIFY(nvlist_add_nvlist(*config,
2886 ZPOOL_CONFIG_LOAD_INFO,
2887 spa->spa_load_info) == 0);
2888 }
2889 spa_unload(spa);
2890 spa_deactivate(spa);
2891 spa->spa_last_open_failed = error;
2892 if (locked)
2893 mutex_exit(&spa_namespace_lock);
2894 *spapp = NULL;
2895 return (error);
2896 }
2897 }
2898
2899 spa_open_ref(spa, tag);
2900
2901 if (config != NULL)
2902 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2903
2904 /*
2905 * If we've recovered the pool, pass back any information we
2906 * gathered while doing the load.
2907 */
2908 if (state == SPA_LOAD_RECOVER) {
2909 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2910 spa->spa_load_info) == 0);
2911 }
2912
2913 if (locked) {
2914 spa->spa_last_open_failed = 0;
2915 spa->spa_last_ubsync_txg = 0;
2916 spa->spa_load_txg = 0;
2917 mutex_exit(&spa_namespace_lock);
2918 }
2919
2920 *spapp = spa;
2921
2922 return (0);
2923 }
2924
2925 int
2926 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2927 nvlist_t **config)
2928 {
2929 return (spa_open_common(name, spapp, tag, policy, config));
2930 }
2931
2932 int
2933 spa_open(const char *name, spa_t **spapp, void *tag)
2934 {
2935 return (spa_open_common(name, spapp, tag, NULL, NULL));
2936 }
2937
2938 /*
2939 * Lookup the given spa_t, incrementing the inject count in the process,
2940 * preventing it from being exported or destroyed.
2941 */
2942 spa_t *
2943 spa_inject_addref(char *name)
2944 {
2945 spa_t *spa;
2946
2947 mutex_enter(&spa_namespace_lock);
2948 if ((spa = spa_lookup(name)) == NULL) {
2949 mutex_exit(&spa_namespace_lock);
2950 return (NULL);
2951 }
2952 spa->spa_inject_ref++;
2953 mutex_exit(&spa_namespace_lock);
2954
2955 return (spa);
2956 }
2957
2958 void
2959 spa_inject_delref(spa_t *spa)
2960 {
2961 mutex_enter(&spa_namespace_lock);
2962 spa->spa_inject_ref--;
2963 mutex_exit(&spa_namespace_lock);
2964 }
2965
2966 /*
2967 * Add spares device information to the nvlist.
2968 */
2969 static void
2970 spa_add_spares(spa_t *spa, nvlist_t *config)
2971 {
2972 nvlist_t **spares;
2973 uint_t i, nspares;
2974 nvlist_t *nvroot;
2975 uint64_t guid;
2976 vdev_stat_t *vs;
2977 uint_t vsc;
2978 uint64_t pool;
2979
2980 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2981
2982 if (spa->spa_spares.sav_count == 0)
2983 return;
2984
2985 VERIFY(nvlist_lookup_nvlist(config,
2986 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2987 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2988 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2989 if (nspares != 0) {
2990 VERIFY(nvlist_add_nvlist_array(nvroot,
2991 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2992 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2993 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2994
2995 /*
2996 * Go through and find any spares which have since been
2997 * repurposed as an active spare. If this is the case, update
2998 * their status appropriately.
2999 */
3000 for (i = 0; i < nspares; i++) {
3001 VERIFY(nvlist_lookup_uint64(spares[i],
3002 ZPOOL_CONFIG_GUID, &guid) == 0);
3003 if (spa_spare_exists(guid, &pool, NULL) &&
3004 pool != 0ULL) {
3005 VERIFY(nvlist_lookup_uint64_array(
3006 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3007 (uint64_t **)&vs, &vsc) == 0);
3008 vs->vs_state = VDEV_STATE_CANT_OPEN;
3009 vs->vs_aux = VDEV_AUX_SPARED;
3010 }
3011 }
3012 }
3013 }
3014
3015 /*
3016 * Add l2cache device information to the nvlist, including vdev stats.
3017 */
3018 static void
3019 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3020 {
3021 nvlist_t **l2cache;
3022 uint_t i, j, nl2cache;
3023 nvlist_t *nvroot;
3024 uint64_t guid;
3025 vdev_t *vd;
3026 vdev_stat_t *vs;
3027 uint_t vsc;
3028
3029 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3030
3031 if (spa->spa_l2cache.sav_count == 0)
3032 return;
3033
3034 VERIFY(nvlist_lookup_nvlist(config,
3035 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3036 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3037 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3038 if (nl2cache != 0) {
3039 VERIFY(nvlist_add_nvlist_array(nvroot,
3040 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3041 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3042 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3043
3044 /*
3045 * Update level 2 cache device stats.
3046 */
3047
3048 for (i = 0; i < nl2cache; i++) {
3049 VERIFY(nvlist_lookup_uint64(l2cache[i],
3050 ZPOOL_CONFIG_GUID, &guid) == 0);
3051
3052 vd = NULL;
3053 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3054 if (guid ==
3055 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3056 vd = spa->spa_l2cache.sav_vdevs[j];
3057 break;
3058 }
3059 }
3060 ASSERT(vd != NULL);
3061
3062 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3063 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3064 == 0);
3065 vdev_get_stats(vd, vs);
3066 }
3067 }
3068 }
3069
3070 static void
3071 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3072 {
3073 nvlist_t *features;
3074 zap_cursor_t zc;
3075 zap_attribute_t za;
3076
3077 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3078 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3079
3080 if (spa->spa_feat_for_read_obj != 0) {
3081 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3082 spa->spa_feat_for_read_obj);
3083 zap_cursor_retrieve(&zc, &za) == 0;
3084 zap_cursor_advance(&zc)) {
3085 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3086 za.za_num_integers == 1);
3087 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3088 za.za_first_integer));
3089 }
3090 zap_cursor_fini(&zc);
3091 }
3092
3093 if (spa->spa_feat_for_write_obj != 0) {
3094 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3095 spa->spa_feat_for_write_obj);
3096 zap_cursor_retrieve(&zc, &za) == 0;
3097 zap_cursor_advance(&zc)) {
3098 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3099 za.za_num_integers == 1);
3100 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3101 za.za_first_integer));
3102 }
3103 zap_cursor_fini(&zc);
3104 }
3105
3106 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3107 features) == 0);
3108 nvlist_free(features);
3109 }
3110
3111 int
3112 spa_get_stats(const char *name, nvlist_t **config,
3113 char *altroot, size_t buflen)
3114 {
3115 int error;
3116 spa_t *spa;
3117
3118 *config = NULL;
3119 error = spa_open_common(name, &spa, FTAG, NULL, config);
3120
3121 if (spa != NULL) {
3122 /*
3123 * This still leaves a window of inconsistency where the spares
3124 * or l2cache devices could change and the config would be
3125 * self-inconsistent.
3126 */
3127 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3128
3129 if (*config != NULL) {
3130 uint64_t loadtimes[2];
3131
3132 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3133 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3134 VERIFY(nvlist_add_uint64_array(*config,
3135 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3136
3137 VERIFY(nvlist_add_uint64(*config,
3138 ZPOOL_CONFIG_ERRCOUNT,
3139 spa_get_errlog_size(spa)) == 0);
3140
3141 if (spa_suspended(spa))
3142 VERIFY(nvlist_add_uint64(*config,
3143 ZPOOL_CONFIG_SUSPENDED,
3144 spa->spa_failmode) == 0);
3145
3146 spa_add_spares(spa, *config);
3147 spa_add_l2cache(spa, *config);
3148 spa_add_feature_stats(spa, *config);
3149 }
3150 }
3151
3152 /*
3153 * We want to get the alternate root even for faulted pools, so we cheat
3154 * and call spa_lookup() directly.
3155 */
3156 if (altroot) {
3157 if (spa == NULL) {
3158 mutex_enter(&spa_namespace_lock);
3159 spa = spa_lookup(name);
3160 if (spa)
3161 spa_altroot(spa, altroot, buflen);
3162 else
3163 altroot[0] = '\0';
3164 spa = NULL;
3165 mutex_exit(&spa_namespace_lock);
3166 } else {
3167 spa_altroot(spa, altroot, buflen);
3168 }
3169 }
3170
3171 if (spa != NULL) {
3172 spa_config_exit(spa, SCL_CONFIG, FTAG);
3173 spa_close(spa, FTAG);
3174 }
3175
3176 return (error);
3177 }
3178
3179 /*
3180 * Validate that the auxiliary device array is well formed. We must have an
3181 * array of nvlists, each which describes a valid leaf vdev. If this is an
3182 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3183 * specified, as long as they are well-formed.
3184 */
3185 static int
3186 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3187 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3188 vdev_labeltype_t label)
3189 {
3190 nvlist_t **dev;
3191 uint_t i, ndev;
3192 vdev_t *vd;
3193 int error;
3194
3195 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3196
3197 /*
3198 * It's acceptable to have no devs specified.
3199 */
3200 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3201 return (0);
3202
3203 if (ndev == 0)
3204 return (SET_ERROR(EINVAL));
3205
3206 /*
3207 * Make sure the pool is formatted with a version that supports this
3208 * device type.
3209 */
3210 if (spa_version(spa) < version)
3211 return (SET_ERROR(ENOTSUP));
3212
3213 /*
3214 * Set the pending device list so we correctly handle device in-use
3215 * checking.
3216 */
3217 sav->sav_pending = dev;
3218 sav->sav_npending = ndev;
3219
3220 for (i = 0; i < ndev; i++) {
3221 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3222 mode)) != 0)
3223 goto out;
3224
3225 if (!vd->vdev_ops->vdev_op_leaf) {
3226 vdev_free(vd);
3227 error = SET_ERROR(EINVAL);
3228 goto out;
3229 }
3230
3231 /*
3232 * The L2ARC currently only supports disk devices in
3233 * kernel context. For user-level testing, we allow it.
3234 */
3235 #ifdef _KERNEL
3236 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3237 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3238 error = SET_ERROR(ENOTBLK);
3239 vdev_free(vd);
3240 goto out;
3241 }
3242 #endif
3243 vd->vdev_top = vd;
3244
3245 if ((error = vdev_open(vd)) == 0 &&
3246 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3247 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3248 vd->vdev_guid) == 0);
3249 }
3250
3251 vdev_free(vd);
3252
3253 if (error &&
3254 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3255 goto out;
3256 else
3257 error = 0;
3258 }
3259
3260 out:
3261 sav->sav_pending = NULL;
3262 sav->sav_npending = 0;
3263 return (error);
3264 }
3265
3266 static int
3267 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3268 {
3269 int error;
3270
3271 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3272
3273 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3274 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3275 VDEV_LABEL_SPARE)) != 0) {
3276 return (error);
3277 }
3278
3279 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3280 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3281 VDEV_LABEL_L2CACHE));
3282 }
3283
3284 static void
3285 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3286 const char *config)
3287 {
3288 int i;
3289
3290 if (sav->sav_config != NULL) {
3291 nvlist_t **olddevs;
3292 uint_t oldndevs;
3293 nvlist_t **newdevs;
3294
3295 /*
3296 * Generate new dev list by concatentating with the
3297 * current dev list.
3298 */
3299 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3300 &olddevs, &oldndevs) == 0);
3301
3302 newdevs = kmem_alloc(sizeof (void *) *
3303 (ndevs + oldndevs), KM_SLEEP);
3304 for (i = 0; i < oldndevs; i++)
3305 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3306 KM_SLEEP) == 0);
3307 for (i = 0; i < ndevs; i++)
3308 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3309 KM_SLEEP) == 0);
3310
3311 VERIFY(nvlist_remove(sav->sav_config, config,
3312 DATA_TYPE_NVLIST_ARRAY) == 0);
3313
3314 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3315 config, newdevs, ndevs + oldndevs) == 0);
3316 for (i = 0; i < oldndevs + ndevs; i++)
3317 nvlist_free(newdevs[i]);
3318 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3319 } else {
3320 /*
3321 * Generate a new dev list.
3322 */
3323 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3324 KM_SLEEP) == 0);
3325 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3326 devs, ndevs) == 0);
3327 }
3328 }
3329
3330 /*
3331 * Stop and drop level 2 ARC devices
3332 */
3333 void
3334 spa_l2cache_drop(spa_t *spa)
3335 {
3336 vdev_t *vd;
3337 int i;
3338 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3339
3340 for (i = 0; i < sav->sav_count; i++) {
3341 uint64_t pool;
3342
3343 vd = sav->sav_vdevs[i];
3344 ASSERT(vd != NULL);
3345
3346 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3347 pool != 0ULL && l2arc_vdev_present(vd))
3348 l2arc_remove_vdev(vd);
3349 }
3350 }
3351
3352 /*
3353 * Pool Creation
3354 */
3355 int
3356 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3357 nvlist_t *zplprops)
3358 {
3359 spa_t *spa;
3360 char *altroot = NULL;
3361 vdev_t *rvd;
3362 dsl_pool_t *dp;
3363 dmu_tx_t *tx;
3364 int error = 0;
3365 uint64_t txg = TXG_INITIAL;
3366 nvlist_t **spares, **l2cache;
3367 uint_t nspares, nl2cache;
3368 uint64_t version, obj;
3369 boolean_t has_features;
3370
3371 /*
3372 * If this pool already exists, return failure.
3373 */
3374 mutex_enter(&spa_namespace_lock);
3375 if (spa_lookup(pool) != NULL) {
3376 mutex_exit(&spa_namespace_lock);
3377 return (SET_ERROR(EEXIST));
3378 }
3379
3380 /*
3381 * Allocate a new spa_t structure.
3382 */
3383 (void) nvlist_lookup_string(props,
3384 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3385 spa = spa_add(pool, NULL, altroot);
3386 spa_activate(spa, spa_mode_global);
3387
3388 if (props && (error = spa_prop_validate(spa, props))) {
3389 spa_deactivate(spa);
3390 spa_remove(spa);
3391 mutex_exit(&spa_namespace_lock);
3392 return (error);
3393 }
3394
3395 has_features = B_FALSE;
3396 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3397 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3398 if (zpool_prop_feature(nvpair_name(elem)))
3399 has_features = B_TRUE;
3400 }
3401
3402 if (has_features || nvlist_lookup_uint64(props,
3403 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3404 version = SPA_VERSION;
3405 }
3406 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3407
3408 spa->spa_first_txg = txg;
3409 spa->spa_uberblock.ub_txg = txg - 1;
3410 spa->spa_uberblock.ub_version = version;
3411 spa->spa_ubsync = spa->spa_uberblock;
3412
3413 /*
3414 * Create "The Godfather" zio to hold all async IOs
3415 */
3416 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3417 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3418
3419 /*
3420 * Create the root vdev.
3421 */
3422 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3423
3424 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3425
3426 ASSERT(error != 0 || rvd != NULL);
3427 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3428
3429 if (error == 0 && !zfs_allocatable_devs(nvroot))
3430 error = SET_ERROR(EINVAL);
3431
3432 if (error == 0 &&
3433 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3434 (error = spa_validate_aux(spa, nvroot, txg,
3435 VDEV_ALLOC_ADD)) == 0) {
3436 for (int c = 0; c < rvd->vdev_children; c++) {
3437 vdev_metaslab_set_size(rvd->vdev_child[c]);
3438 vdev_expand(rvd->vdev_child[c], txg);
3439 }
3440 }
3441
3442 spa_config_exit(spa, SCL_ALL, FTAG);
3443
3444 if (error != 0) {
3445 spa_unload(spa);
3446 spa_deactivate(spa);
3447 spa_remove(spa);
3448 mutex_exit(&spa_namespace_lock);
3449 return (error);
3450 }
3451
3452 /*
3453 * Get the list of spares, if specified.
3454 */
3455 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3456 &spares, &nspares) == 0) {
3457 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3458 KM_SLEEP) == 0);
3459 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3460 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3461 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3462 spa_load_spares(spa);
3463 spa_config_exit(spa, SCL_ALL, FTAG);
3464 spa->spa_spares.sav_sync = B_TRUE;
3465 }
3466
3467 /*
3468 * Get the list of level 2 cache devices, if specified.
3469 */
3470 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3471 &l2cache, &nl2cache) == 0) {
3472 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3473 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3474 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3475 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3476 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3477 spa_load_l2cache(spa);
3478 spa_config_exit(spa, SCL_ALL, FTAG);
3479 spa->spa_l2cache.sav_sync = B_TRUE;
3480 }
3481
3482 spa->spa_is_initializing = B_TRUE;
3483 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3484 spa->spa_meta_objset = dp->dp_meta_objset;
3485 spa->spa_is_initializing = B_FALSE;
3486
3487 /*
3488 * Create DDTs (dedup tables).
3489 */
3490 ddt_create(spa);
3491
3492 spa_update_dspace(spa);
3493
3494 tx = dmu_tx_create_assigned(dp, txg);
3495
3496 /*
3497 * Create the pool config object.
3498 */
3499 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3500 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3501 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3502
3503 if (zap_add(spa->spa_meta_objset,
3504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3505 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3506 cmn_err(CE_PANIC, "failed to add pool config");
3507 }
3508
3509 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3510 spa_feature_create_zap_objects(spa, tx);
3511
3512 if (zap_add(spa->spa_meta_objset,
3513 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3514 sizeof (uint64_t), 1, &version, tx) != 0) {
3515 cmn_err(CE_PANIC, "failed to add pool version");
3516 }
3517
3518 /* Newly created pools with the right version are always deflated. */
3519 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3520 spa->spa_deflate = TRUE;
3521 if (zap_add(spa->spa_meta_objset,
3522 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3523 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3524 cmn_err(CE_PANIC, "failed to add deflate");
3525 }
3526 }
3527
3528 /*
3529 * Create the deferred-free bpobj. Turn off compression
3530 * because sync-to-convergence takes longer if the blocksize
3531 * keeps changing.
3532 */
3533 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3534 dmu_object_set_compress(spa->spa_meta_objset, obj,
3535 ZIO_COMPRESS_OFF, tx);
3536 if (zap_add(spa->spa_meta_objset,
3537 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3538 sizeof (uint64_t), 1, &obj, tx) != 0) {
3539 cmn_err(CE_PANIC, "failed to add bpobj");
3540 }
3541 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3542 spa->spa_meta_objset, obj));
3543
3544 /*
3545 * Create the pool's history object.
3546 */
3547 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3548 spa_history_create_obj(spa, tx);
3549
3550 /*
3551 * Set pool properties.
3552 */
3553 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3554 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3555 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3556 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3557
3558 if (props != NULL) {
3559 spa_configfile_set(spa, props, B_FALSE);
3560 spa_sync_props(props, tx);
3561 }
3562
3563 dmu_tx_commit(tx);
3564
3565 spa->spa_sync_on = B_TRUE;
3566 txg_sync_start(spa->spa_dsl_pool);
3567
3568 /*
3569 * We explicitly wait for the first transaction to complete so that our
3570 * bean counters are appropriately updated.
3571 */
3572 txg_wait_synced(spa->spa_dsl_pool, txg);
3573
3574 spa_config_sync(spa, B_FALSE, B_TRUE);
3575
3576 spa_history_log_version(spa, "create");
3577
3578 spa->spa_minref = refcount_count(&spa->spa_refcount);
3579
3580 mutex_exit(&spa_namespace_lock);
3581
3582 return (0);
3583 }
3584
3585 #ifdef _KERNEL
3586 /*
3587 * Get the root pool information from the root disk, then import the root pool
3588 * during the system boot up time.
3589 */
3590 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3591
3592 static nvlist_t *
3593 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3594 {
3595 nvlist_t *config;
3596 nvlist_t *nvtop, *nvroot;
3597 uint64_t pgid;
3598
3599 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3600 return (NULL);
3601
3602 /*
3603 * Add this top-level vdev to the child array.
3604 */
3605 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3606 &nvtop) == 0);
3607 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3608 &pgid) == 0);
3609 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3610
3611 /*
3612 * Put this pool's top-level vdevs into a root vdev.
3613 */
3614 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3615 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3616 VDEV_TYPE_ROOT) == 0);
3617 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3618 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3619 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3620 &nvtop, 1) == 0);
3621
3622 /*
3623 * Replace the existing vdev_tree with the new root vdev in
3624 * this pool's configuration (remove the old, add the new).
3625 */
3626 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3627 nvlist_free(nvroot);
3628 return (config);
3629 }
3630
3631 /*
3632 * Walk the vdev tree and see if we can find a device with "better"
3633 * configuration. A configuration is "better" if the label on that
3634 * device has a more recent txg.
3635 */
3636 static void
3637 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3638 {
3639 for (int c = 0; c < vd->vdev_children; c++)
3640 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3641
3642 if (vd->vdev_ops->vdev_op_leaf) {
3643 nvlist_t *label;
3644 uint64_t label_txg;
3645
3646 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3647 &label) != 0)
3648 return;
3649
3650 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3651 &label_txg) == 0);
3652
3653 /*
3654 * Do we have a better boot device?
3655 */
3656 if (label_txg > *txg) {
3657 *txg = label_txg;
3658 *avd = vd;
3659 }
3660 nvlist_free(label);
3661 }
3662 }
3663
3664 /*
3665 * Import a root pool.
3666 *
3667 * For x86. devpath_list will consist of devid and/or physpath name of
3668 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3669 * The GRUB "findroot" command will return the vdev we should boot.
3670 *
3671 * For Sparc, devpath_list consists the physpath name of the booting device
3672 * no matter the rootpool is a single device pool or a mirrored pool.
3673 * e.g.
3674 * "/pci@1f,0/ide@d/disk@0,0:a"
3675 */
3676 int
3677 spa_import_rootpool(char *devpath, char *devid)
3678 {
3679 spa_t *spa;
3680 vdev_t *rvd, *bvd, *avd = NULL;
3681 nvlist_t *config, *nvtop;
3682 uint64_t guid, txg;
3683 char *pname;
3684 int error;
3685
3686 /*
3687 * Read the label from the boot device and generate a configuration.
3688 */
3689 config = spa_generate_rootconf(devpath, devid, &guid);
3690 #if defined(_OBP) && defined(_KERNEL)
3691 if (config == NULL) {
3692 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3693 /* iscsi boot */
3694 get_iscsi_bootpath_phy(devpath);
3695 config = spa_generate_rootconf(devpath, devid, &guid);
3696 }
3697 }
3698 #endif
3699 if (config == NULL) {
3700 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3701 devpath);
3702 return (SET_ERROR(EIO));
3703 }
3704
3705 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3706 &pname) == 0);
3707 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3708
3709 mutex_enter(&spa_namespace_lock);
3710 if ((spa = spa_lookup(pname)) != NULL) {
3711 /*
3712 * Remove the existing root pool from the namespace so that we
3713 * can replace it with the correct config we just read in.
3714 */
3715 spa_remove(spa);
3716 }
3717
3718 spa = spa_add(pname, config, NULL);
3719 spa->spa_is_root = B_TRUE;
3720 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3721
3722 /*
3723 * Build up a vdev tree based on the boot device's label config.
3724 */
3725 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3726 &nvtop) == 0);
3727 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3728 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3729 VDEV_ALLOC_ROOTPOOL);
3730 spa_config_exit(spa, SCL_ALL, FTAG);
3731 if (error) {
3732 mutex_exit(&spa_namespace_lock);
3733 nvlist_free(config);
3734 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3735 pname);
3736 return (error);
3737 }
3738
3739 /*
3740 * Get the boot vdev.
3741 */
3742 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3743 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3744 (u_longlong_t)guid);
3745 error = SET_ERROR(ENOENT);
3746 goto out;
3747 }
3748
3749 /*
3750 * Determine if there is a better boot device.
3751 */
3752 avd = bvd;
3753 spa_alt_rootvdev(rvd, &avd, &txg);
3754 if (avd != bvd) {
3755 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3756 "try booting from '%s'", avd->vdev_path);
3757 error = SET_ERROR(EINVAL);
3758 goto out;
3759 }
3760
3761 /*
3762 * If the boot device is part of a spare vdev then ensure that
3763 * we're booting off the active spare.
3764 */
3765 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3766 !bvd->vdev_isspare) {
3767 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3768 "try booting from '%s'",
3769 bvd->vdev_parent->
3770 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3771 error = SET_ERROR(EINVAL);
3772 goto out;
3773 }
3774
3775 error = 0;
3776 out:
3777 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3778 vdev_free(rvd);
3779 spa_config_exit(spa, SCL_ALL, FTAG);
3780 mutex_exit(&spa_namespace_lock);
3781
3782 nvlist_free(config);
3783 return (error);
3784 }
3785
3786 #endif
3787
3788 /*
3789 * Import a non-root pool into the system.
3790 */
3791 int
3792 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3793 {
3794 spa_t *spa;
3795 char *altroot = NULL;
3796 spa_load_state_t state = SPA_LOAD_IMPORT;
3797 zpool_rewind_policy_t policy;
3798 uint64_t mode = spa_mode_global;
3799 uint64_t readonly = B_FALSE;
3800 int error;
3801 nvlist_t *nvroot;
3802 nvlist_t **spares, **l2cache;
3803 uint_t nspares, nl2cache;
3804
3805 /*
3806 * If a pool with this name exists, return failure.
3807 */
3808 mutex_enter(&spa_namespace_lock);
3809 if (spa_lookup(pool) != NULL) {
3810 mutex_exit(&spa_namespace_lock);
3811 return (SET_ERROR(EEXIST));
3812 }
3813
3814 /*
3815 * Create and initialize the spa structure.
3816 */
3817 (void) nvlist_lookup_string(props,
3818 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3819 (void) nvlist_lookup_uint64(props,
3820 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3821 if (readonly)
3822 mode = FREAD;
3823 spa = spa_add(pool, config, altroot);
3824 spa->spa_import_flags = flags;
3825
3826 /*
3827 * Verbatim import - Take a pool and insert it into the namespace
3828 * as if it had been loaded at boot.
3829 */
3830 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3831 if (props != NULL)
3832 spa_configfile_set(spa, props, B_FALSE);
3833
3834 spa_config_sync(spa, B_FALSE, B_TRUE);
3835
3836 mutex_exit(&spa_namespace_lock);
3837 spa_history_log_version(spa, "import");
3838
3839 return (0);
3840 }
3841
3842 spa_activate(spa, mode);
3843
3844 /*
3845 * Don't start async tasks until we know everything is healthy.
3846 */
3847 spa_async_suspend(spa);
3848
3849 zpool_get_rewind_policy(config, &policy);
3850 if (policy.zrp_request & ZPOOL_DO_REWIND)
3851 state = SPA_LOAD_RECOVER;
3852
3853 /*
3854 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3855 * because the user-supplied config is actually the one to trust when
3856 * doing an import.
3857 */
3858 if (state != SPA_LOAD_RECOVER)
3859 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3860
3861 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3862 policy.zrp_request);
3863
3864 /*
3865 * Propagate anything learned while loading the pool and pass it
3866 * back to caller (i.e. rewind info, missing devices, etc).
3867 */
3868 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3869 spa->spa_load_info) == 0);
3870
3871 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3872 /*
3873 * Toss any existing sparelist, as it doesn't have any validity
3874 * anymore, and conflicts with spa_has_spare().
3875 */
3876 if (spa->spa_spares.sav_config) {
3877 nvlist_free(spa->spa_spares.sav_config);
3878 spa->spa_spares.sav_config = NULL;
3879 spa_load_spares(spa);
3880 }
3881 if (spa->spa_l2cache.sav_config) {
3882 nvlist_free(spa->spa_l2cache.sav_config);
3883 spa->spa_l2cache.sav_config = NULL;
3884 spa_load_l2cache(spa);
3885 }
3886
3887 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3888 &nvroot) == 0);
3889 if (error == 0)
3890 error = spa_validate_aux(spa, nvroot, -1ULL,
3891 VDEV_ALLOC_SPARE);
3892 if (error == 0)
3893 error = spa_validate_aux(spa, nvroot, -1ULL,
3894 VDEV_ALLOC_L2CACHE);
3895 spa_config_exit(spa, SCL_ALL, FTAG);
3896
3897 if (props != NULL)
3898 spa_configfile_set(spa, props, B_FALSE);
3899
3900 if (error != 0 || (props && spa_writeable(spa) &&
3901 (error = spa_prop_set(spa, props)))) {
3902 spa_unload(spa);
3903 spa_deactivate(spa);
3904 spa_remove(spa);
3905 mutex_exit(&spa_namespace_lock);
3906 return (error);
3907 }
3908
3909 spa_async_resume(spa);
3910
3911 /*
3912 * Override any spares and level 2 cache devices as specified by
3913 * the user, as these may have correct device names/devids, etc.
3914 */
3915 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3916 &spares, &nspares) == 0) {
3917 if (spa->spa_spares.sav_config)
3918 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3919 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3920 else
3921 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3922 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3923 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3924 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3925 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3926 spa_load_spares(spa);
3927 spa_config_exit(spa, SCL_ALL, FTAG);
3928 spa->spa_spares.sav_sync = B_TRUE;
3929 }
3930 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3931 &l2cache, &nl2cache) == 0) {
3932 if (spa->spa_l2cache.sav_config)
3933 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3934 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3935 else
3936 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3937 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3938 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3939 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3940 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3941 spa_load_l2cache(spa);
3942 spa_config_exit(spa, SCL_ALL, FTAG);
3943 spa->spa_l2cache.sav_sync = B_TRUE;
3944 }
3945
3946 /*
3947 * Check for any removed devices.
3948 */
3949 if (spa->spa_autoreplace) {
3950 spa_aux_check_removed(&spa->spa_spares);
3951 spa_aux_check_removed(&spa->spa_l2cache);
3952 }
3953
3954 if (spa_writeable(spa)) {
3955 /*
3956 * Update the config cache to include the newly-imported pool.
3957 */
3958 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3959 }
3960
3961 /*
3962 * It's possible that the pool was expanded while it was exported.
3963 * We kick off an async task to handle this for us.
3964 */
3965 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3966
3967 mutex_exit(&spa_namespace_lock);
3968 spa_history_log_version(spa, "import");
3969
3970 return (0);
3971 }
3972
3973 nvlist_t *
3974 spa_tryimport(nvlist_t *tryconfig)
3975 {
3976 nvlist_t *config = NULL;
3977 char *poolname;
3978 spa_t *spa;
3979 uint64_t state;
3980 int error;
3981
3982 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3983 return (NULL);
3984
3985 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3986 return (NULL);
3987
3988 /*
3989 * Create and initialize the spa structure.
3990 */
3991 mutex_enter(&spa_namespace_lock);
3992 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3993 spa_activate(spa, FREAD);
3994
3995 /*
3996 * Pass off the heavy lifting to spa_load().
3997 * Pass TRUE for mosconfig because the user-supplied config
3998 * is actually the one to trust when doing an import.
3999 */
4000 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4001
4002 /*
4003 * If 'tryconfig' was at least parsable, return the current config.
4004 */
4005 if (spa->spa_root_vdev != NULL) {
4006 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4007 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4008 poolname) == 0);
4009 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4010 state) == 0);
4011 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4012 spa->spa_uberblock.ub_timestamp) == 0);
4013 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4014 spa->spa_load_info) == 0);
4015
4016 /*
4017 * If the bootfs property exists on this pool then we
4018 * copy it out so that external consumers can tell which
4019 * pools are bootable.
4020 */
4021 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4022 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4023
4024 /*
4025 * We have to play games with the name since the
4026 * pool was opened as TRYIMPORT_NAME.
4027 */
4028 if (dsl_dsobj_to_dsname(spa_name(spa),
4029 spa->spa_bootfs, tmpname) == 0) {
4030 char *cp;
4031 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4032
4033 cp = strchr(tmpname, '/');
4034 if (cp == NULL) {
4035 (void) strlcpy(dsname, tmpname,
4036 MAXPATHLEN);
4037 } else {
4038 (void) snprintf(dsname, MAXPATHLEN,
4039 "%s/%s", poolname, ++cp);
4040 }
4041 VERIFY(nvlist_add_string(config,
4042 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4043 kmem_free(dsname, MAXPATHLEN);
4044 }
4045 kmem_free(tmpname, MAXPATHLEN);
4046 }
4047
4048 /*
4049 * Add the list of hot spares and level 2 cache devices.
4050 */
4051 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4052 spa_add_spares(spa, config);
4053 spa_add_l2cache(spa, config);
4054 spa_config_exit(spa, SCL_CONFIG, FTAG);
4055 }
4056
4057 spa_unload(spa);
4058 spa_deactivate(spa);
4059 spa_remove(spa);
4060 mutex_exit(&spa_namespace_lock);
4061
4062 return (config);
4063 }
4064
4065 /*
4066 * Pool export/destroy
4067 *
4068 * The act of destroying or exporting a pool is very simple. We make sure there
4069 * is no more pending I/O and any references to the pool are gone. Then, we
4070 * update the pool state and sync all the labels to disk, removing the
4071 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4072 * we don't sync the labels or remove the configuration cache.
4073 */
4074 static int
4075 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4076 boolean_t force, boolean_t hardforce)
4077 {
4078 spa_t *spa;
4079
4080 if (oldconfig)
4081 *oldconfig = NULL;
4082
4083 if (!(spa_mode_global & FWRITE))
4084 return (SET_ERROR(EROFS));
4085
4086 mutex_enter(&spa_namespace_lock);
4087 if ((spa = spa_lookup(pool)) == NULL) {
4088 mutex_exit(&spa_namespace_lock);
4089 return (SET_ERROR(ENOENT));
4090 }
4091
4092 /*
4093 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4094 * reacquire the namespace lock, and see if we can export.
4095 */
4096 spa_open_ref(spa, FTAG);
4097 mutex_exit(&spa_namespace_lock);
4098 spa_async_suspend(spa);
4099 mutex_enter(&spa_namespace_lock);
4100 spa_close(spa, FTAG);
4101
4102 /*
4103 * The pool will be in core if it's openable,
4104 * in which case we can modify its state.
4105 */
4106 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4107 /*
4108 * Objsets may be open only because they're dirty, so we
4109 * have to force it to sync before checking spa_refcnt.
4110 */
4111 txg_wait_synced(spa->spa_dsl_pool, 0);
4112
4113 /*
4114 * A pool cannot be exported or destroyed if there are active
4115 * references. If we are resetting a pool, allow references by
4116 * fault injection handlers.
4117 */
4118 if (!spa_refcount_zero(spa) ||
4119 (spa->spa_inject_ref != 0 &&
4120 new_state != POOL_STATE_UNINITIALIZED)) {
4121 spa_async_resume(spa);
4122 mutex_exit(&spa_namespace_lock);
4123 return (SET_ERROR(EBUSY));
4124 }
4125
4126 /*
4127 * A pool cannot be exported if it has an active shared spare.
4128 * This is to prevent other pools stealing the active spare
4129 * from an exported pool. At user's own will, such pool can
4130 * be forcedly exported.
4131 */
4132 if (!force && new_state == POOL_STATE_EXPORTED &&
4133 spa_has_active_shared_spare(spa)) {
4134 spa_async_resume(spa);
4135 mutex_exit(&spa_namespace_lock);
4136 return (SET_ERROR(EXDEV));
4137 }
4138
4139 /*
4140 * We want this to be reflected on every label,
4141 * so mark them all dirty. spa_unload() will do the
4142 * final sync that pushes these changes out.
4143 */
4144 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4145 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4146 spa->spa_state = new_state;
4147 spa->spa_final_txg = spa_last_synced_txg(spa) +
4148 TXG_DEFER_SIZE + 1;
4149 vdev_config_dirty(spa->spa_root_vdev);
4150 spa_config_exit(spa, SCL_ALL, FTAG);
4151 }
4152 }
4153
4154 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4155
4156 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4157 spa_unload(spa);
4158 spa_deactivate(spa);
4159 }
4160
4161 if (oldconfig && spa->spa_config)
4162 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4163
4164 if (new_state != POOL_STATE_UNINITIALIZED) {
4165 if (!hardforce)
4166 spa_config_sync(spa, B_TRUE, B_TRUE);
4167 spa_remove(spa);
4168 }
4169 mutex_exit(&spa_namespace_lock);
4170
4171 return (0);
4172 }
4173
4174 /*
4175 * Destroy a storage pool.
4176 */
4177 int
4178 spa_destroy(char *pool)
4179 {
4180 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4181 B_FALSE, B_FALSE));
4182 }
4183
4184 /*
4185 * Export a storage pool.
4186 */
4187 int
4188 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4189 boolean_t hardforce)
4190 {
4191 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4192 force, hardforce));
4193 }
4194
4195 /*
4196 * Similar to spa_export(), this unloads the spa_t without actually removing it
4197 * from the namespace in any way.
4198 */
4199 int
4200 spa_reset(char *pool)
4201 {
4202 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4203 B_FALSE, B_FALSE));
4204 }
4205
4206 /*
4207 * ==========================================================================
4208 * Device manipulation
4209 * ==========================================================================
4210 */
4211
4212 /*
4213 * Add a device to a storage pool.
4214 */
4215 int
4216 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4217 {
4218 uint64_t txg, id;
4219 int error;
4220 vdev_t *rvd = spa->spa_root_vdev;
4221 vdev_t *vd, *tvd;
4222 nvlist_t **spares, **l2cache;
4223 uint_t nspares, nl2cache;
4224
4225 ASSERT(spa_writeable(spa));
4226
4227 txg = spa_vdev_enter(spa);
4228
4229 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4230 VDEV_ALLOC_ADD)) != 0)
4231 return (spa_vdev_exit(spa, NULL, txg, error));
4232
4233 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4234
4235 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4236 &nspares) != 0)
4237 nspares = 0;
4238
4239 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4240 &nl2cache) != 0)
4241 nl2cache = 0;
4242
4243 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4244 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4245
4246 if (vd->vdev_children != 0 &&
4247 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4248 return (spa_vdev_exit(spa, vd, txg, error));
4249
4250 /*
4251 * We must validate the spares and l2cache devices after checking the
4252 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4253 */
4254 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4255 return (spa_vdev_exit(spa, vd, txg, error));
4256
4257 /*
4258 * Transfer each new top-level vdev from vd to rvd.
4259 */
4260 for (int c = 0; c < vd->vdev_children; c++) {
4261
4262 /*
4263 * Set the vdev id to the first hole, if one exists.
4264 */
4265 for (id = 0; id < rvd->vdev_children; id++) {
4266 if (rvd->vdev_child[id]->vdev_ishole) {
4267 vdev_free(rvd->vdev_child[id]);
4268 break;
4269 }
4270 }
4271 tvd = vd->vdev_child[c];
4272 vdev_remove_child(vd, tvd);
4273 tvd->vdev_id = id;
4274 vdev_add_child(rvd, tvd);
4275 vdev_config_dirty(tvd);
4276 }
4277
4278 if (nspares != 0) {
4279 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4280 ZPOOL_CONFIG_SPARES);
4281 spa_load_spares(spa);
4282 spa->spa_spares.sav_sync = B_TRUE;
4283 }
4284
4285 if (nl2cache != 0) {
4286 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4287 ZPOOL_CONFIG_L2CACHE);
4288 spa_load_l2cache(spa);
4289 spa->spa_l2cache.sav_sync = B_TRUE;
4290 }
4291
4292 /*
4293 * We have to be careful when adding new vdevs to an existing pool.
4294 * If other threads start allocating from these vdevs before we
4295 * sync the config cache, and we lose power, then upon reboot we may
4296 * fail to open the pool because there are DVAs that the config cache
4297 * can't translate. Therefore, we first add the vdevs without
4298 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4299 * and then let spa_config_update() initialize the new metaslabs.
4300 *
4301 * spa_load() checks for added-but-not-initialized vdevs, so that
4302 * if we lose power at any point in this sequence, the remaining
4303 * steps will be completed the next time we load the pool.
4304 */
4305 (void) spa_vdev_exit(spa, vd, txg, 0);
4306
4307 mutex_enter(&spa_namespace_lock);
4308 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4309 mutex_exit(&spa_namespace_lock);
4310
4311 return (0);
4312 }
4313
4314 /*
4315 * Attach a device to a mirror. The arguments are the path to any device
4316 * in the mirror, and the nvroot for the new device. If the path specifies
4317 * a device that is not mirrored, we automatically insert the mirror vdev.
4318 *
4319 * If 'replacing' is specified, the new device is intended to replace the
4320 * existing device; in this case the two devices are made into their own
4321 * mirror using the 'replacing' vdev, which is functionally identical to
4322 * the mirror vdev (it actually reuses all the same ops) but has a few
4323 * extra rules: you can't attach to it after it's been created, and upon
4324 * completion of resilvering, the first disk (the one being replaced)
4325 * is automatically detached.
4326 */
4327 int
4328 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4329 {
4330 uint64_t txg, dtl_max_txg;
4331 vdev_t *rvd = spa->spa_root_vdev;
4332 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4333 vdev_ops_t *pvops;
4334 char *oldvdpath, *newvdpath;
4335 int newvd_isspare;
4336 int error;
4337
4338 ASSERT(spa_writeable(spa));
4339
4340 txg = spa_vdev_enter(spa);
4341
4342 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4343
4344 if (oldvd == NULL)
4345 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4346
4347 if (!oldvd->vdev_ops->vdev_op_leaf)
4348 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4349
4350 pvd = oldvd->vdev_parent;
4351
4352 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4353 VDEV_ALLOC_ATTACH)) != 0)
4354 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4355
4356 if (newrootvd->vdev_children != 1)
4357 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4358
4359 newvd = newrootvd->vdev_child[0];
4360
4361 if (!newvd->vdev_ops->vdev_op_leaf)
4362 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4363
4364 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4365 return (spa_vdev_exit(spa, newrootvd, txg, error));
4366
4367 /*
4368 * Spares can't replace logs
4369 */
4370 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4371 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4372
4373 if (!replacing) {
4374 /*
4375 * For attach, the only allowable parent is a mirror or the root
4376 * vdev.
4377 */
4378 if (pvd->vdev_ops != &vdev_mirror_ops &&
4379 pvd->vdev_ops != &vdev_root_ops)
4380 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4381
4382 pvops = &vdev_mirror_ops;
4383 } else {
4384 /*
4385 * Active hot spares can only be replaced by inactive hot
4386 * spares.
4387 */
4388 if (pvd->vdev_ops == &vdev_spare_ops &&
4389 oldvd->vdev_isspare &&
4390 !spa_has_spare(spa, newvd->vdev_guid))
4391 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4392
4393 /*
4394 * If the source is a hot spare, and the parent isn't already a
4395 * spare, then we want to create a new hot spare. Otherwise, we
4396 * want to create a replacing vdev. The user is not allowed to
4397 * attach to a spared vdev child unless the 'isspare' state is
4398 * the same (spare replaces spare, non-spare replaces
4399 * non-spare).
4400 */
4401 if (pvd->vdev_ops == &vdev_replacing_ops &&
4402 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4403 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4404 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4405 newvd->vdev_isspare != oldvd->vdev_isspare) {
4406 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4407 }
4408
4409 if (newvd->vdev_isspare)
4410 pvops = &vdev_spare_ops;
4411 else
4412 pvops = &vdev_replacing_ops;
4413 }
4414
4415 /*
4416 * Make sure the new device is big enough.
4417 */
4418 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4419 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4420
4421 /*
4422 * The new device cannot have a higher alignment requirement
4423 * than the top-level vdev.
4424 */
4425 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4426 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4427
4428 /*
4429 * If this is an in-place replacement, update oldvd's path and devid
4430 * to make it distinguishable from newvd, and unopenable from now on.
4431 */
4432 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4433 spa_strfree(oldvd->vdev_path);
4434 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4435 KM_SLEEP);
4436 (void) sprintf(oldvd->vdev_path, "%s/%s",
4437 newvd->vdev_path, "old");
4438 if (oldvd->vdev_devid != NULL) {
4439 spa_strfree(oldvd->vdev_devid);
4440 oldvd->vdev_devid = NULL;
4441 }
4442 }
4443
4444 /* mark the device being resilvered */
4445 newvd->vdev_resilvering = B_TRUE;
4446
4447 /*
4448 * If the parent is not a mirror, or if we're replacing, insert the new
4449 * mirror/replacing/spare vdev above oldvd.
4450 */
4451 if (pvd->vdev_ops != pvops)
4452 pvd = vdev_add_parent(oldvd, pvops);
4453
4454 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4455 ASSERT(pvd->vdev_ops == pvops);
4456 ASSERT(oldvd->vdev_parent == pvd);
4457
4458 /*
4459 * Extract the new device from its root and add it to pvd.
4460 */
4461 vdev_remove_child(newrootvd, newvd);
4462 newvd->vdev_id = pvd->vdev_children;
4463 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4464 vdev_add_child(pvd, newvd);
4465
4466 tvd = newvd->vdev_top;
4467 ASSERT(pvd->vdev_top == tvd);
4468 ASSERT(tvd->vdev_parent == rvd);
4469
4470 vdev_config_dirty(tvd);
4471
4472 /*
4473 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4474 * for any dmu_sync-ed blocks. It will propagate upward when
4475 * spa_vdev_exit() calls vdev_dtl_reassess().
4476 */
4477 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4478
4479 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4480 dtl_max_txg - TXG_INITIAL);
4481
4482 if (newvd->vdev_isspare) {
4483 spa_spare_activate(newvd);
4484 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4485 }
4486
4487 oldvdpath = spa_strdup(oldvd->vdev_path);
4488 newvdpath = spa_strdup(newvd->vdev_path);
4489 newvd_isspare = newvd->vdev_isspare;
4490
4491 /*
4492 * Mark newvd's DTL dirty in this txg.
4493 */
4494 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4495
4496 /*
4497 * Restart the resilver
4498 */
4499 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4500
4501 /*
4502 * Commit the config
4503 */
4504 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4505
4506 spa_history_log_internal(spa, "vdev attach", NULL,
4507 "%s vdev=%s %s vdev=%s",
4508 replacing && newvd_isspare ? "spare in" :
4509 replacing ? "replace" : "attach", newvdpath,
4510 replacing ? "for" : "to", oldvdpath);
4511
4512 spa_strfree(oldvdpath);
4513 spa_strfree(newvdpath);
4514
4515 if (spa->spa_bootfs)
4516 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4517
4518 return (0);
4519 }
4520
4521 /*
4522 * Detach a device from a mirror or replacing vdev.
4523 * If 'replace_done' is specified, only detach if the parent
4524 * is a replacing vdev.
4525 */
4526 int
4527 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4528 {
4529 uint64_t txg;
4530 int error;
4531 vdev_t *rvd = spa->spa_root_vdev;
4532 vdev_t *vd, *pvd, *cvd, *tvd;
4533 boolean_t unspare = B_FALSE;
4534 uint64_t unspare_guid = 0;
4535 char *vdpath;
4536
4537 ASSERT(spa_writeable(spa));
4538
4539 txg = spa_vdev_enter(spa);
4540
4541 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4542
4543 if (vd == NULL)
4544 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4545
4546 if (!vd->vdev_ops->vdev_op_leaf)
4547 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4548
4549 pvd = vd->vdev_parent;
4550
4551 /*
4552 * If the parent/child relationship is not as expected, don't do it.
4553 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4554 * vdev that's replacing B with C. The user's intent in replacing
4555 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4556 * the replace by detaching C, the expected behavior is to end up
4557 * M(A,B). But suppose that right after deciding to detach C,
4558 * the replacement of B completes. We would have M(A,C), and then
4559 * ask to detach C, which would leave us with just A -- not what
4560 * the user wanted. To prevent this, we make sure that the
4561 * parent/child relationship hasn't changed -- in this example,
4562 * that C's parent is still the replacing vdev R.
4563 */
4564 if (pvd->vdev_guid != pguid && pguid != 0)
4565 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4566
4567 /*
4568 * Only 'replacing' or 'spare' vdevs can be replaced.
4569 */
4570 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4571 pvd->vdev_ops != &vdev_spare_ops)
4572 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4573
4574 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4575 spa_version(spa) >= SPA_VERSION_SPARES);
4576
4577 /*
4578 * Only mirror, replacing, and spare vdevs support detach.
4579 */
4580 if (pvd->vdev_ops != &vdev_replacing_ops &&
4581 pvd->vdev_ops != &vdev_mirror_ops &&
4582 pvd->vdev_ops != &vdev_spare_ops)
4583 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4584
4585 /*
4586 * If this device has the only valid copy of some data,
4587 * we cannot safely detach it.
4588 */
4589 if (vdev_dtl_required(vd))
4590 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4591
4592 ASSERT(pvd->vdev_children >= 2);
4593
4594 /*
4595 * If we are detaching the second disk from a replacing vdev, then
4596 * check to see if we changed the original vdev's path to have "/old"
4597 * at the end in spa_vdev_attach(). If so, undo that change now.
4598 */
4599 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4600 vd->vdev_path != NULL) {
4601 size_t len = strlen(vd->vdev_path);
4602
4603 for (int c = 0; c < pvd->vdev_children; c++) {
4604 cvd = pvd->vdev_child[c];
4605
4606 if (cvd == vd || cvd->vdev_path == NULL)
4607 continue;
4608
4609 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4610 strcmp(cvd->vdev_path + len, "/old") == 0) {
4611 spa_strfree(cvd->vdev_path);
4612 cvd->vdev_path = spa_strdup(vd->vdev_path);
4613 break;
4614 }
4615 }
4616 }
4617
4618 /*
4619 * If we are detaching the original disk from a spare, then it implies
4620 * that the spare should become a real disk, and be removed from the
4621 * active spare list for the pool.
4622 */
4623 if (pvd->vdev_ops == &vdev_spare_ops &&
4624 vd->vdev_id == 0 &&
4625 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4626 unspare = B_TRUE;
4627
4628 /*
4629 * Erase the disk labels so the disk can be used for other things.
4630 * This must be done after all other error cases are handled,
4631 * but before we disembowel vd (so we can still do I/O to it).
4632 * But if we can't do it, don't treat the error as fatal --
4633 * it may be that the unwritability of the disk is the reason
4634 * it's being detached!
4635 */
4636 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4637
4638 /*
4639 * Remove vd from its parent and compact the parent's children.
4640 */
4641 vdev_remove_child(pvd, vd);
4642 vdev_compact_children(pvd);
4643
4644 /*
4645 * Remember one of the remaining children so we can get tvd below.
4646 */
4647 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4648
4649 /*
4650 * If we need to remove the remaining child from the list of hot spares,
4651 * do it now, marking the vdev as no longer a spare in the process.
4652 * We must do this before vdev_remove_parent(), because that can
4653 * change the GUID if it creates a new toplevel GUID. For a similar
4654 * reason, we must remove the spare now, in the same txg as the detach;
4655 * otherwise someone could attach a new sibling, change the GUID, and
4656 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4657 */
4658 if (unspare) {
4659 ASSERT(cvd->vdev_isspare);
4660 spa_spare_remove(cvd);
4661 unspare_guid = cvd->vdev_guid;
4662 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4663 cvd->vdev_unspare = B_TRUE;
4664 }
4665
4666 /*
4667 * If the parent mirror/replacing vdev only has one child,
4668 * the parent is no longer needed. Remove it from the tree.
4669 */
4670 if (pvd->vdev_children == 1) {
4671 if (pvd->vdev_ops == &vdev_spare_ops)
4672 cvd->vdev_unspare = B_FALSE;
4673 vdev_remove_parent(cvd);
4674 cvd->vdev_resilvering = B_FALSE;
4675 }
4676
4677
4678 /*
4679 * We don't set tvd until now because the parent we just removed
4680 * may have been the previous top-level vdev.
4681 */
4682 tvd = cvd->vdev_top;
4683 ASSERT(tvd->vdev_parent == rvd);
4684
4685 /*
4686 * Reevaluate the parent vdev state.
4687 */
4688 vdev_propagate_state(cvd);
4689
4690 /*
4691 * If the 'autoexpand' property is set on the pool then automatically
4692 * try to expand the size of the pool. For example if the device we
4693 * just detached was smaller than the others, it may be possible to
4694 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4695 * first so that we can obtain the updated sizes of the leaf vdevs.
4696 */
4697 if (spa->spa_autoexpand) {
4698 vdev_reopen(tvd);
4699 vdev_expand(tvd, txg);
4700 }
4701
4702 vdev_config_dirty(tvd);
4703
4704 /*
4705 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4706 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4707 * But first make sure we're not on any *other* txg's DTL list, to
4708 * prevent vd from being accessed after it's freed.
4709 */
4710 vdpath = spa_strdup(vd->vdev_path);
4711 for (int t = 0; t < TXG_SIZE; t++)
4712 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4713 vd->vdev_detached = B_TRUE;
4714 vdev_dirty(tvd, VDD_DTL, vd, txg);
4715
4716 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4717
4718 /* hang on to the spa before we release the lock */
4719 spa_open_ref(spa, FTAG);
4720
4721 error = spa_vdev_exit(spa, vd, txg, 0);
4722
4723 spa_history_log_internal(spa, "detach", NULL,
4724 "vdev=%s", vdpath);
4725 spa_strfree(vdpath);
4726
4727 /*
4728 * If this was the removal of the original device in a hot spare vdev,
4729 * then we want to go through and remove the device from the hot spare
4730 * list of every other pool.
4731 */
4732 if (unspare) {
4733 spa_t *altspa = NULL;
4734
4735 mutex_enter(&spa_namespace_lock);
4736 while ((altspa = spa_next(altspa)) != NULL) {
4737 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4738 altspa == spa)
4739 continue;
4740
4741 spa_open_ref(altspa, FTAG);
4742 mutex_exit(&spa_namespace_lock);
4743 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4744 mutex_enter(&spa_namespace_lock);
4745 spa_close(altspa, FTAG);
4746 }
4747 mutex_exit(&spa_namespace_lock);
4748
4749 /* search the rest of the vdevs for spares to remove */
4750 spa_vdev_resilver_done(spa);
4751 }
4752
4753 /* all done with the spa; OK to release */
4754 mutex_enter(&spa_namespace_lock);
4755 spa_close(spa, FTAG);
4756 mutex_exit(&spa_namespace_lock);
4757
4758 return (error);
4759 }
4760
4761 /*
4762 * Split a set of devices from their mirrors, and create a new pool from them.
4763 */
4764 int
4765 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4766 nvlist_t *props, boolean_t exp)
4767 {
4768 int error = 0;
4769 uint64_t txg, *glist;
4770 spa_t *newspa;
4771 uint_t c, children, lastlog;
4772 nvlist_t **child, *nvl, *tmp;
4773 dmu_tx_t *tx;
4774 char *altroot = NULL;
4775 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4776 boolean_t activate_slog;
4777
4778 ASSERT(spa_writeable(spa));
4779
4780 txg = spa_vdev_enter(spa);
4781
4782 /* clear the log and flush everything up to now */
4783 activate_slog = spa_passivate_log(spa);
4784 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4785 error = spa_offline_log(spa);
4786 txg = spa_vdev_config_enter(spa);
4787
4788 if (activate_slog)
4789 spa_activate_log(spa);
4790
4791 if (error != 0)
4792 return (spa_vdev_exit(spa, NULL, txg, error));
4793
4794 /* check new spa name before going any further */
4795 if (spa_lookup(newname) != NULL)
4796 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4797
4798 /*
4799 * scan through all the children to ensure they're all mirrors
4800 */
4801 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4802 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4803 &children) != 0)
4804 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4805
4806 /* first, check to ensure we've got the right child count */
4807 rvd = spa->spa_root_vdev;
4808 lastlog = 0;
4809 for (c = 0; c < rvd->vdev_children; c++) {
4810 vdev_t *vd = rvd->vdev_child[c];
4811
4812 /* don't count the holes & logs as children */
4813 if (vd->vdev_islog || vd->vdev_ishole) {
4814 if (lastlog == 0)
4815 lastlog = c;
4816 continue;
4817 }
4818
4819 lastlog = 0;
4820 }
4821 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4822 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4823
4824 /* next, ensure no spare or cache devices are part of the split */
4825 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4826 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4827 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4828
4829 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4830 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4831
4832 /* then, loop over each vdev and validate it */
4833 for (c = 0; c < children; c++) {
4834 uint64_t is_hole = 0;
4835
4836 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4837 &is_hole);
4838
4839 if (is_hole != 0) {
4840 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4841 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4842 continue;
4843 } else {
4844 error = SET_ERROR(EINVAL);
4845 break;
4846 }
4847 }
4848
4849 /* which disk is going to be split? */
4850 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4851 &glist[c]) != 0) {
4852 error = SET_ERROR(EINVAL);
4853 break;
4854 }
4855
4856 /* look it up in the spa */
4857 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4858 if (vml[c] == NULL) {
4859 error = SET_ERROR(ENODEV);
4860 break;
4861 }
4862
4863 /* make sure there's nothing stopping the split */
4864 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4865 vml[c]->vdev_islog ||
4866 vml[c]->vdev_ishole ||
4867 vml[c]->vdev_isspare ||
4868 vml[c]->vdev_isl2cache ||
4869 !vdev_writeable(vml[c]) ||
4870 vml[c]->vdev_children != 0 ||
4871 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4872 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4873 error = SET_ERROR(EINVAL);
4874 break;
4875 }
4876
4877 if (vdev_dtl_required(vml[c])) {
4878 error = SET_ERROR(EBUSY);
4879 break;
4880 }
4881
4882 /* we need certain info from the top level */
4883 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4884 vml[c]->vdev_top->vdev_ms_array) == 0);
4885 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4886 vml[c]->vdev_top->vdev_ms_shift) == 0);
4887 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4888 vml[c]->vdev_top->vdev_asize) == 0);
4889 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4890 vml[c]->vdev_top->vdev_ashift) == 0);
4891 }
4892
4893 if (error != 0) {
4894 kmem_free(vml, children * sizeof (vdev_t *));
4895 kmem_free(glist, children * sizeof (uint64_t));
4896 return (spa_vdev_exit(spa, NULL, txg, error));
4897 }
4898
4899 /* stop writers from using the disks */
4900 for (c = 0; c < children; c++) {
4901 if (vml[c] != NULL)
4902 vml[c]->vdev_offline = B_TRUE;
4903 }
4904 vdev_reopen(spa->spa_root_vdev);
4905
4906 /*
4907 * Temporarily record the splitting vdevs in the spa config. This
4908 * will disappear once the config is regenerated.
4909 */
4910 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4911 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4912 glist, children) == 0);
4913 kmem_free(glist, children * sizeof (uint64_t));
4914
4915 mutex_enter(&spa->spa_props_lock);
4916 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4917 nvl) == 0);
4918 mutex_exit(&spa->spa_props_lock);
4919 spa->spa_config_splitting = nvl;
4920 vdev_config_dirty(spa->spa_root_vdev);
4921
4922 /* configure and create the new pool */
4923 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4924 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4925 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4926 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4927 spa_version(spa)) == 0);
4928 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4929 spa->spa_config_txg) == 0);
4930 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4931 spa_generate_guid(NULL)) == 0);
4932 (void) nvlist_lookup_string(props,
4933 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4934
4935 /* add the new pool to the namespace */
4936 newspa = spa_add(newname, config, altroot);
4937 newspa->spa_config_txg = spa->spa_config_txg;
4938 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4939
4940 /* release the spa config lock, retaining the namespace lock */
4941 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4942
4943 if (zio_injection_enabled)
4944 zio_handle_panic_injection(spa, FTAG, 1);
4945
4946 spa_activate(newspa, spa_mode_global);
4947 spa_async_suspend(newspa);
4948
4949 /* create the new pool from the disks of the original pool */
4950 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4951 if (error)
4952 goto out;
4953
4954 /* if that worked, generate a real config for the new pool */
4955 if (newspa->spa_root_vdev != NULL) {
4956 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4957 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4958 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4959 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4960 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4961 B_TRUE));
4962 }
4963
4964 /* set the props */
4965 if (props != NULL) {
4966 spa_configfile_set(newspa, props, B_FALSE);
4967 error = spa_prop_set(newspa, props);
4968 if (error)
4969 goto out;
4970 }
4971
4972 /* flush everything */
4973 txg = spa_vdev_config_enter(newspa);
4974 vdev_config_dirty(newspa->spa_root_vdev);
4975 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4976
4977 if (zio_injection_enabled)
4978 zio_handle_panic_injection(spa, FTAG, 2);
4979
4980 spa_async_resume(newspa);
4981
4982 /* finally, update the original pool's config */
4983 txg = spa_vdev_config_enter(spa);
4984 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4985 error = dmu_tx_assign(tx, TXG_WAIT);
4986 if (error != 0)
4987 dmu_tx_abort(tx);
4988 for (c = 0; c < children; c++) {
4989 if (vml[c] != NULL) {
4990 vdev_split(vml[c]);
4991 if (error == 0)
4992 spa_history_log_internal(spa, "detach", tx,
4993 "vdev=%s", vml[c]->vdev_path);
4994 vdev_free(vml[c]);
4995 }
4996 }
4997 vdev_config_dirty(spa->spa_root_vdev);
4998 spa->spa_config_splitting = NULL;
4999 nvlist_free(nvl);
5000 if (error == 0)
5001 dmu_tx_commit(tx);
5002 (void) spa_vdev_exit(spa, NULL, txg, 0);
5003
5004 if (zio_injection_enabled)
5005 zio_handle_panic_injection(spa, FTAG, 3);
5006
5007 /* split is complete; log a history record */
5008 spa_history_log_internal(newspa, "split", NULL,
5009 "from pool %s", spa_name(spa));
5010
5011 kmem_free(vml, children * sizeof (vdev_t *));
5012
5013 /* if we're not going to mount the filesystems in userland, export */
5014 if (exp)
5015 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5016 B_FALSE, B_FALSE);
5017
5018 return (error);
5019
5020 out:
5021 spa_unload(newspa);
5022 spa_deactivate(newspa);
5023 spa_remove(newspa);
5024
5025 txg = spa_vdev_config_enter(spa);
5026
5027 /* re-online all offlined disks */
5028 for (c = 0; c < children; c++) {
5029 if (vml[c] != NULL)
5030 vml[c]->vdev_offline = B_FALSE;
5031 }
5032 vdev_reopen(spa->spa_root_vdev);
5033
5034 nvlist_free(spa->spa_config_splitting);
5035 spa->spa_config_splitting = NULL;
5036 (void) spa_vdev_exit(spa, NULL, txg, error);
5037
5038 kmem_free(vml, children * sizeof (vdev_t *));
5039 return (error);
5040 }
5041
5042 static nvlist_t *
5043 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5044 {
5045 for (int i = 0; i < count; i++) {
5046 uint64_t guid;
5047
5048 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5049 &guid) == 0);
5050
5051 if (guid == target_guid)
5052 return (nvpp[i]);
5053 }
5054
5055 return (NULL);
5056 }
5057
5058 static void
5059 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5060 nvlist_t *dev_to_remove)
5061 {
5062 nvlist_t **newdev = NULL;
5063
5064 if (count > 1)
5065 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5066
5067 for (int i = 0, j = 0; i < count; i++) {
5068 if (dev[i] == dev_to_remove)
5069 continue;
5070 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5071 }
5072
5073 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5074 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5075
5076 for (int i = 0; i < count - 1; i++)
5077 nvlist_free(newdev[i]);
5078
5079 if (count > 1)
5080 kmem_free(newdev, (count - 1) * sizeof (void *));
5081 }
5082
5083 /*
5084 * Evacuate the device.
5085 */
5086 static int
5087 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5088 {
5089 uint64_t txg;
5090 int error = 0;
5091
5092 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5093 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5094 ASSERT(vd == vd->vdev_top);
5095
5096 /*
5097 * Evacuate the device. We don't hold the config lock as writer
5098 * since we need to do I/O but we do keep the
5099 * spa_namespace_lock held. Once this completes the device
5100 * should no longer have any blocks allocated on it.
5101 */
5102 if (vd->vdev_islog) {
5103 if (vd->vdev_stat.vs_alloc != 0)
5104 error = spa_offline_log(spa);
5105 } else {
5106 error = SET_ERROR(ENOTSUP);
5107 }
5108
5109 if (error)
5110 return (error);
5111
5112 /*
5113 * The evacuation succeeded. Remove any remaining MOS metadata
5114 * associated with this vdev, and wait for these changes to sync.
5115 */
5116 ASSERT0(vd->vdev_stat.vs_alloc);
5117 txg = spa_vdev_config_enter(spa);
5118 vd->vdev_removing = B_TRUE;
5119 vdev_dirty(vd, 0, NULL, txg);
5120 vdev_config_dirty(vd);
5121 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5122
5123 return (0);
5124 }
5125
5126 /*
5127 * Complete the removal by cleaning up the namespace.
5128 */
5129 static void
5130 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5131 {
5132 vdev_t *rvd = spa->spa_root_vdev;
5133 uint64_t id = vd->vdev_id;
5134 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5135
5136 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5137 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5138 ASSERT(vd == vd->vdev_top);
5139
5140 /*
5141 * Only remove any devices which are empty.
5142 */
5143 if (vd->vdev_stat.vs_alloc != 0)
5144 return;
5145
5146 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5147
5148 if (list_link_active(&vd->vdev_state_dirty_node))
5149 vdev_state_clean(vd);
5150 if (list_link_active(&vd->vdev_config_dirty_node))
5151 vdev_config_clean(vd);
5152
5153 vdev_free(vd);
5154
5155 if (last_vdev) {
5156 vdev_compact_children(rvd);
5157 } else {
5158 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5159 vdev_add_child(rvd, vd);
5160 }
5161 vdev_config_dirty(rvd);
5162
5163 /*
5164 * Reassess the health of our root vdev.
5165 */
5166 vdev_reopen(rvd);
5167 }
5168
5169 /*
5170 * Remove a device from the pool -
5171 *
5172 * Removing a device from the vdev namespace requires several steps
5173 * and can take a significant amount of time. As a result we use
5174 * the spa_vdev_config_[enter/exit] functions which allow us to
5175 * grab and release the spa_config_lock while still holding the namespace
5176 * lock. During each step the configuration is synced out.
5177 */
5178
5179 /*
5180 * Remove a device from the pool. Currently, this supports removing only hot
5181 * spares, slogs, and level 2 ARC devices.
5182 */
5183 int
5184 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5185 {
5186 vdev_t *vd;
5187 metaslab_group_t *mg;
5188 nvlist_t **spares, **l2cache, *nv;
5189 uint64_t txg = 0;
5190 uint_t nspares, nl2cache;
5191 int error = 0;
5192 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5193
5194 ASSERT(spa_writeable(spa));
5195
5196 if (!locked)
5197 txg = spa_vdev_enter(spa);
5198
5199 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5200
5201 if (spa->spa_spares.sav_vdevs != NULL &&
5202 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5203 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5204 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5205 /*
5206 * Only remove the hot spare if it's not currently in use
5207 * in this pool.
5208 */
5209 if (vd == NULL || unspare) {
5210 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5211 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5212 spa_load_spares(spa);
5213 spa->spa_spares.sav_sync = B_TRUE;
5214 } else {
5215 error = SET_ERROR(EBUSY);
5216 }
5217 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5218 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5219 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5220 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5221 /*
5222 * Cache devices can always be removed.
5223 */
5224 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5225 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5226 spa_load_l2cache(spa);
5227 spa->spa_l2cache.sav_sync = B_TRUE;
5228 } else if (vd != NULL && vd->vdev_islog) {
5229 ASSERT(!locked);
5230 ASSERT(vd == vd->vdev_top);
5231
5232 /*
5233 * XXX - Once we have bp-rewrite this should
5234 * become the common case.
5235 */
5236
5237 mg = vd->vdev_mg;
5238
5239 /*
5240 * Stop allocating from this vdev.
5241 */
5242 metaslab_group_passivate(mg);
5243
5244 /*
5245 * Wait for the youngest allocations and frees to sync,
5246 * and then wait for the deferral of those frees to finish.
5247 */
5248 spa_vdev_config_exit(spa, NULL,
5249 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5250
5251 /*
5252 * Attempt to evacuate the vdev.
5253 */
5254 error = spa_vdev_remove_evacuate(spa, vd);
5255
5256 txg = spa_vdev_config_enter(spa);
5257
5258 /*
5259 * If we couldn't evacuate the vdev, unwind.
5260 */
5261 if (error) {
5262 metaslab_group_activate(mg);
5263 return (spa_vdev_exit(spa, NULL, txg, error));
5264 }
5265
5266 /*
5267 * Clean up the vdev namespace.
5268 */
5269 spa_vdev_remove_from_namespace(spa, vd);
5270
5271 } else if (vd != NULL) {
5272 /*
5273 * Normal vdevs cannot be removed (yet).
5274 */
5275 error = SET_ERROR(ENOTSUP);
5276 } else {
5277 /*
5278 * There is no vdev of any kind with the specified guid.
5279 */
5280 error = SET_ERROR(ENOENT);
5281 }
5282
5283 if (!locked)
5284 return (spa_vdev_exit(spa, NULL, txg, error));
5285
5286 return (error);
5287 }
5288
5289 /*
5290 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5291 * current spared, so we can detach it.
5292 */
5293 static vdev_t *
5294 spa_vdev_resilver_done_hunt(vdev_t *vd)
5295 {
5296 vdev_t *newvd, *oldvd;
5297
5298 for (int c = 0; c < vd->vdev_children; c++) {
5299 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5300 if (oldvd != NULL)
5301 return (oldvd);
5302 }
5303
5304 /*
5305 * Check for a completed replacement. We always consider the first
5306 * vdev in the list to be the oldest vdev, and the last one to be
5307 * the newest (see spa_vdev_attach() for how that works). In
5308 * the case where the newest vdev is faulted, we will not automatically
5309 * remove it after a resilver completes. This is OK as it will require
5310 * user intervention to determine which disk the admin wishes to keep.
5311 */
5312 if (vd->vdev_ops == &vdev_replacing_ops) {
5313 ASSERT(vd->vdev_children > 1);
5314
5315 newvd = vd->vdev_child[vd->vdev_children - 1];
5316 oldvd = vd->vdev_child[0];
5317
5318 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5319 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5320 !vdev_dtl_required(oldvd))
5321 return (oldvd);
5322 }
5323
5324 /*
5325 * Check for a completed resilver with the 'unspare' flag set.
5326 */
5327 if (vd->vdev_ops == &vdev_spare_ops) {
5328 vdev_t *first = vd->vdev_child[0];
5329 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5330
5331 if (last->vdev_unspare) {
5332 oldvd = first;
5333 newvd = last;
5334 } else if (first->vdev_unspare) {
5335 oldvd = last;
5336 newvd = first;
5337 } else {
5338 oldvd = NULL;
5339 }
5340
5341 if (oldvd != NULL &&
5342 vdev_dtl_empty(newvd, DTL_MISSING) &&
5343 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5344 !vdev_dtl_required(oldvd))
5345 return (oldvd);
5346
5347 /*
5348 * If there are more than two spares attached to a disk,
5349 * and those spares are not required, then we want to
5350 * attempt to free them up now so that they can be used
5351 * by other pools. Once we're back down to a single
5352 * disk+spare, we stop removing them.
5353 */
5354 if (vd->vdev_children > 2) {
5355 newvd = vd->vdev_child[1];
5356
5357 if (newvd->vdev_isspare && last->vdev_isspare &&
5358 vdev_dtl_empty(last, DTL_MISSING) &&
5359 vdev_dtl_empty(last, DTL_OUTAGE) &&
5360 !vdev_dtl_required(newvd))
5361 return (newvd);
5362 }
5363 }
5364
5365 return (NULL);
5366 }
5367
5368 static void
5369 spa_vdev_resilver_done(spa_t *spa)
5370 {
5371 vdev_t *vd, *pvd, *ppvd;
5372 uint64_t guid, sguid, pguid, ppguid;
5373
5374 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5375
5376 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5377 pvd = vd->vdev_parent;
5378 ppvd = pvd->vdev_parent;
5379 guid = vd->vdev_guid;
5380 pguid = pvd->vdev_guid;
5381 ppguid = ppvd->vdev_guid;
5382 sguid = 0;
5383 /*
5384 * If we have just finished replacing a hot spared device, then
5385 * we need to detach the parent's first child (the original hot
5386 * spare) as well.
5387 */
5388 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5389 ppvd->vdev_children == 2) {
5390 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5391 sguid = ppvd->vdev_child[1]->vdev_guid;
5392 }
5393 spa_config_exit(spa, SCL_ALL, FTAG);
5394 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5395 return;
5396 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5397 return;
5398 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5399 }
5400
5401 spa_config_exit(spa, SCL_ALL, FTAG);
5402 }
5403
5404 /*
5405 * Update the stored path or FRU for this vdev.
5406 */
5407 int
5408 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5409 boolean_t ispath)
5410 {
5411 vdev_t *vd;
5412 boolean_t sync = B_FALSE;
5413
5414 ASSERT(spa_writeable(spa));
5415
5416 spa_vdev_state_enter(spa, SCL_ALL);
5417
5418 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5419 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5420
5421 if (!vd->vdev_ops->vdev_op_leaf)
5422 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5423
5424 if (ispath) {
5425 if (strcmp(value, vd->vdev_path) != 0) {
5426 spa_strfree(vd->vdev_path);
5427 vd->vdev_path = spa_strdup(value);
5428 sync = B_TRUE;
5429 }
5430 } else {
5431 if (vd->vdev_fru == NULL) {
5432 vd->vdev_fru = spa_strdup(value);
5433 sync = B_TRUE;
5434 } else if (strcmp(value, vd->vdev_fru) != 0) {
5435 spa_strfree(vd->vdev_fru);
5436 vd->vdev_fru = spa_strdup(value);
5437 sync = B_TRUE;
5438 }
5439 }
5440
5441 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5442 }
5443
5444 int
5445 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5446 {
5447 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5448 }
5449
5450 int
5451 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5452 {
5453 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5454 }
5455
5456 /*
5457 * ==========================================================================
5458 * SPA Scanning
5459 * ==========================================================================
5460 */
5461
5462 int
5463 spa_scan_stop(spa_t *spa)
5464 {
5465 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5466 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5467 return (SET_ERROR(EBUSY));
5468 return (dsl_scan_cancel(spa->spa_dsl_pool));
5469 }
5470
5471 int
5472 spa_scan(spa_t *spa, pool_scan_func_t func)
5473 {
5474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5475
5476 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5477 return (SET_ERROR(ENOTSUP));
5478
5479 /*
5480 * If a resilver was requested, but there is no DTL on a
5481 * writeable leaf device, we have nothing to do.
5482 */
5483 if (func == POOL_SCAN_RESILVER &&
5484 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5485 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5486 return (0);
5487 }
5488
5489 return (dsl_scan(spa->spa_dsl_pool, func));
5490 }
5491
5492 /*
5493 * ==========================================================================
5494 * SPA async task processing
5495 * ==========================================================================
5496 */
5497
5498 static void
5499 spa_async_remove(spa_t *spa, vdev_t *vd)
5500 {
5501 if (vd->vdev_remove_wanted) {
5502 vd->vdev_remove_wanted = B_FALSE;
5503 vd->vdev_delayed_close = B_FALSE;
5504 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5505
5506 /*
5507 * We want to clear the stats, but we don't want to do a full
5508 * vdev_clear() as that will cause us to throw away
5509 * degraded/faulted state as well as attempt to reopen the
5510 * device, all of which is a waste.
5511 */
5512 vd->vdev_stat.vs_read_errors = 0;
5513 vd->vdev_stat.vs_write_errors = 0;
5514 vd->vdev_stat.vs_checksum_errors = 0;
5515
5516 vdev_state_dirty(vd->vdev_top);
5517 }
5518
5519 for (int c = 0; c < vd->vdev_children; c++)
5520 spa_async_remove(spa, vd->vdev_child[c]);
5521 }
5522
5523 static void
5524 spa_async_probe(spa_t *spa, vdev_t *vd)
5525 {
5526 if (vd->vdev_probe_wanted) {
5527 vd->vdev_probe_wanted = B_FALSE;
5528 vdev_reopen(vd); /* vdev_open() does the actual probe */
5529 }
5530
5531 for (int c = 0; c < vd->vdev_children; c++)
5532 spa_async_probe(spa, vd->vdev_child[c]);
5533 }
5534
5535 static void
5536 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5537 {
5538 sysevent_id_t eid;
5539 nvlist_t *attr;
5540 char *physpath;
5541
5542 if (!spa->spa_autoexpand)
5543 return;
5544
5545 for (int c = 0; c < vd->vdev_children; c++) {
5546 vdev_t *cvd = vd->vdev_child[c];
5547 spa_async_autoexpand(spa, cvd);
5548 }
5549
5550 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5551 return;
5552
5553 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5554 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5555
5556 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5557 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5558
5559 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5560 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5561
5562 nvlist_free(attr);
5563 kmem_free(physpath, MAXPATHLEN);
5564 }
5565
5566 static void
5567 spa_async_thread(spa_t *spa)
5568 {
5569 int tasks;
5570
5571 ASSERT(spa->spa_sync_on);
5572
5573 mutex_enter(&spa->spa_async_lock);
5574 tasks = spa->spa_async_tasks;
5575 spa->spa_async_tasks = 0;
5576 mutex_exit(&spa->spa_async_lock);
5577
5578 /*
5579 * See if the config needs to be updated.
5580 */
5581 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5582 uint64_t old_space, new_space;
5583
5584 mutex_enter(&spa_namespace_lock);
5585 old_space = metaslab_class_get_space(spa_normal_class(spa));
5586 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5587 new_space = metaslab_class_get_space(spa_normal_class(spa));
5588 mutex_exit(&spa_namespace_lock);
5589
5590 /*
5591 * If the pool grew as a result of the config update,
5592 * then log an internal history event.
5593 */
5594 if (new_space != old_space) {
5595 spa_history_log_internal(spa, "vdev online", NULL,
5596 "pool '%s' size: %llu(+%llu)",
5597 spa_name(spa), new_space, new_space - old_space);
5598 }
5599 }
5600
5601 /*
5602 * See if any devices need to be marked REMOVED.
5603 */
5604 if (tasks & SPA_ASYNC_REMOVE) {
5605 spa_vdev_state_enter(spa, SCL_NONE);
5606 spa_async_remove(spa, spa->spa_root_vdev);
5607 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5608 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5609 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5610 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5611 (void) spa_vdev_state_exit(spa, NULL, 0);
5612 }
5613
5614 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5615 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5616 spa_async_autoexpand(spa, spa->spa_root_vdev);
5617 spa_config_exit(spa, SCL_CONFIG, FTAG);
5618 }
5619
5620 /*
5621 * See if any devices need to be probed.
5622 */
5623 if (tasks & SPA_ASYNC_PROBE) {
5624 spa_vdev_state_enter(spa, SCL_NONE);
5625 spa_async_probe(spa, spa->spa_root_vdev);
5626 (void) spa_vdev_state_exit(spa, NULL, 0);
5627 }
5628
5629 /*
5630 * If any devices are done replacing, detach them.
5631 */
5632 if (tasks & SPA_ASYNC_RESILVER_DONE)
5633 spa_vdev_resilver_done(spa);
5634
5635 /*
5636 * Kick off a resilver.
5637 */
5638 if (tasks & SPA_ASYNC_RESILVER)
5639 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5640
5641 /*
5642 * Let the world know that we're done.
5643 */
5644 mutex_enter(&spa->spa_async_lock);
5645 spa->spa_async_thread = NULL;
5646 cv_broadcast(&spa->spa_async_cv);
5647 mutex_exit(&spa->spa_async_lock);
5648 thread_exit();
5649 }
5650
5651 void
5652 spa_async_suspend(spa_t *spa)
5653 {
5654 mutex_enter(&spa->spa_async_lock);
5655 spa->spa_async_suspended++;
5656 while (spa->spa_async_thread != NULL)
5657 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5658 mutex_exit(&spa->spa_async_lock);
5659 }
5660
5661 void
5662 spa_async_resume(spa_t *spa)
5663 {
5664 mutex_enter(&spa->spa_async_lock);
5665 ASSERT(spa->spa_async_suspended != 0);
5666 spa->spa_async_suspended--;
5667 mutex_exit(&spa->spa_async_lock);
5668 }
5669
5670 static int
5671 spa_async_tasks_pending(spa_t *spa)
5672 {
5673 u_int non_config_tasks;
5674 u_int config_task;
5675 boolean_t config_task_suspended;
5676
5677 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5678 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5679 if (spa->spa_ccw_fail_time == 0)
5680 config_task_suspended = B_FALSE;
5681 else
5682 config_task_suspended =
5683 (ddi_get_lbolt64() - spa->spa_ccw_fail_time)
5684 < (zfs_ccw_retry_interval * hz);
5685
5686 return (non_config_tasks || (config_task && !config_task_suspended));
5687 }
5688
5689 static void
5690 spa_async_dispatch(spa_t *spa)
5691 {
5692 mutex_enter(&spa->spa_async_lock);
5693 if (spa_async_tasks_pending(spa) &&
5694 !spa->spa_async_suspended &&
5695 spa->spa_async_thread == NULL &&
5696 rootdir != NULL)
5697 spa->spa_async_thread = thread_create(NULL, 0,
5698 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5699 mutex_exit(&spa->spa_async_lock);
5700 }
5701
5702 void
5703 spa_async_request(spa_t *spa, int task)
5704 {
5705 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5706 mutex_enter(&spa->spa_async_lock);
5707 spa->spa_async_tasks |= task;
5708 mutex_exit(&spa->spa_async_lock);
5709 }
5710
5711 /*
5712 * ==========================================================================
5713 * SPA syncing routines
5714 * ==========================================================================
5715 */
5716
5717 static int
5718 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5719 {
5720 bpobj_t *bpo = arg;
5721 bpobj_enqueue(bpo, bp, tx);
5722 return (0);
5723 }
5724
5725 static int
5726 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5727 {
5728 zio_t *zio = arg;
5729
5730 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5731 zio->io_flags));
5732 return (0);
5733 }
5734
5735 static void
5736 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5737 {
5738 char *packed = NULL;
5739 size_t bufsize;
5740 size_t nvsize = 0;
5741 dmu_buf_t *db;
5742
5743 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5744
5745 /*
5746 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5747 * information. This avoids the dbuf_will_dirty() path and
5748 * saves us a pre-read to get data we don't actually care about.
5749 */
5750 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5751 packed = kmem_alloc(bufsize, KM_SLEEP);
5752
5753 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5754 KM_SLEEP) == 0);
5755 bzero(packed + nvsize, bufsize - nvsize);
5756
5757 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5758
5759 kmem_free(packed, bufsize);
5760
5761 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5762 dmu_buf_will_dirty(db, tx);
5763 *(uint64_t *)db->db_data = nvsize;
5764 dmu_buf_rele(db, FTAG);
5765 }
5766
5767 static void
5768 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5769 const char *config, const char *entry)
5770 {
5771 nvlist_t *nvroot;
5772 nvlist_t **list;
5773 int i;
5774
5775 if (!sav->sav_sync)
5776 return;
5777
5778 /*
5779 * Update the MOS nvlist describing the list of available devices.
5780 * spa_validate_aux() will have already made sure this nvlist is
5781 * valid and the vdevs are labeled appropriately.
5782 */
5783 if (sav->sav_object == 0) {
5784 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5785 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5786 sizeof (uint64_t), tx);
5787 VERIFY(zap_update(spa->spa_meta_objset,
5788 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5789 &sav->sav_object, tx) == 0);
5790 }
5791
5792 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5793 if (sav->sav_count == 0) {
5794 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5795 } else {
5796 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5797 for (i = 0; i < sav->sav_count; i++)
5798 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5799 B_FALSE, VDEV_CONFIG_L2CACHE);
5800 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5801 sav->sav_count) == 0);
5802 for (i = 0; i < sav->sav_count; i++)
5803 nvlist_free(list[i]);
5804 kmem_free(list, sav->sav_count * sizeof (void *));
5805 }
5806
5807 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5808 nvlist_free(nvroot);
5809
5810 sav->sav_sync = B_FALSE;
5811 }
5812
5813 static void
5814 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5815 {
5816 nvlist_t *config;
5817
5818 if (list_is_empty(&spa->spa_config_dirty_list))
5819 return;
5820
5821 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5822
5823 config = spa_config_generate(spa, spa->spa_root_vdev,
5824 dmu_tx_get_txg(tx), B_FALSE);
5825
5826 /*
5827 * If we're upgrading the spa version then make sure that
5828 * the config object gets updated with the correct version.
5829 */
5830 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5831 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5832 spa->spa_uberblock.ub_version);
5833
5834 spa_config_exit(spa, SCL_STATE, FTAG);
5835
5836 if (spa->spa_config_syncing)
5837 nvlist_free(spa->spa_config_syncing);
5838 spa->spa_config_syncing = config;
5839
5840 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5841 }
5842
5843 static void
5844 spa_sync_version(void *arg, dmu_tx_t *tx)
5845 {
5846 uint64_t *versionp = arg;
5847 uint64_t version = *versionp;
5848 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5849
5850 /*
5851 * Setting the version is special cased when first creating the pool.
5852 */
5853 ASSERT(tx->tx_txg != TXG_INITIAL);
5854
5855 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5856 ASSERT(version >= spa_version(spa));
5857
5858 spa->spa_uberblock.ub_version = version;
5859 vdev_config_dirty(spa->spa_root_vdev);
5860 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5861 }
5862
5863 /*
5864 * Set zpool properties.
5865 */
5866 static void
5867 spa_sync_props(void *arg, dmu_tx_t *tx)
5868 {
5869 nvlist_t *nvp = arg;
5870 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5871 objset_t *mos = spa->spa_meta_objset;
5872 nvpair_t *elem = NULL;
5873
5874 mutex_enter(&spa->spa_props_lock);
5875
5876 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5877 uint64_t intval;
5878 char *strval, *fname;
5879 zpool_prop_t prop;
5880 const char *propname;
5881 zprop_type_t proptype;
5882 zfeature_info_t *feature;
5883
5884 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5885 case ZPROP_INVAL:
5886 /*
5887 * We checked this earlier in spa_prop_validate().
5888 */
5889 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5890
5891 fname = strchr(nvpair_name(elem), '@') + 1;
5892 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5893
5894 spa_feature_enable(spa, feature, tx);
5895 spa_history_log_internal(spa, "set", tx,
5896 "%s=enabled", nvpair_name(elem));
5897 break;
5898
5899 case ZPOOL_PROP_VERSION:
5900 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5901 /*
5902 * The version is synced seperatly before other
5903 * properties and should be correct by now.
5904 */
5905 ASSERT3U(spa_version(spa), >=, intval);
5906 break;
5907
5908 case ZPOOL_PROP_ALTROOT:
5909 /*
5910 * 'altroot' is a non-persistent property. It should
5911 * have been set temporarily at creation or import time.
5912 */
5913 ASSERT(spa->spa_root != NULL);
5914 break;
5915
5916 case ZPOOL_PROP_READONLY:
5917 case ZPOOL_PROP_CACHEFILE:
5918 /*
5919 * 'readonly' and 'cachefile' are also non-persisitent
5920 * properties.
5921 */
5922 break;
5923 case ZPOOL_PROP_COMMENT:
5924 VERIFY(nvpair_value_string(elem, &strval) == 0);
5925 if (spa->spa_comment != NULL)
5926 spa_strfree(spa->spa_comment);
5927 spa->spa_comment = spa_strdup(strval);
5928 /*
5929 * We need to dirty the configuration on all the vdevs
5930 * so that their labels get updated. It's unnecessary
5931 * to do this for pool creation since the vdev's
5932 * configuratoin has already been dirtied.
5933 */
5934 if (tx->tx_txg != TXG_INITIAL)
5935 vdev_config_dirty(spa->spa_root_vdev);
5936 spa_history_log_internal(spa, "set", tx,
5937 "%s=%s", nvpair_name(elem), strval);
5938 break;
5939 default:
5940 /*
5941 * Set pool property values in the poolprops mos object.
5942 */
5943 if (spa->spa_pool_props_object == 0) {
5944 spa->spa_pool_props_object =
5945 zap_create_link(mos, DMU_OT_POOL_PROPS,
5946 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5947 tx);
5948 }
5949
5950 /* normalize the property name */
5951 propname = zpool_prop_to_name(prop);
5952 proptype = zpool_prop_get_type(prop);
5953
5954 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5955 ASSERT(proptype == PROP_TYPE_STRING);
5956 VERIFY(nvpair_value_string(elem, &strval) == 0);
5957 VERIFY(zap_update(mos,
5958 spa->spa_pool_props_object, propname,
5959 1, strlen(strval) + 1, strval, tx) == 0);
5960 spa_history_log_internal(spa, "set", tx,
5961 "%s=%s", nvpair_name(elem), strval);
5962 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5963 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5964
5965 if (proptype == PROP_TYPE_INDEX) {
5966 const char *unused;
5967 VERIFY(zpool_prop_index_to_string(
5968 prop, intval, &unused) == 0);
5969 }
5970 VERIFY(zap_update(mos,
5971 spa->spa_pool_props_object, propname,
5972 8, 1, &intval, tx) == 0);
5973 spa_history_log_internal(spa, "set", tx,
5974 "%s=%lld", nvpair_name(elem), intval);
5975 } else {
5976 ASSERT(0); /* not allowed */
5977 }
5978
5979 switch (prop) {
5980 case ZPOOL_PROP_DELEGATION:
5981 spa->spa_delegation = intval;
5982 break;
5983 case ZPOOL_PROP_BOOTFS:
5984 spa->spa_bootfs = intval;
5985 break;
5986 case ZPOOL_PROP_FAILUREMODE:
5987 spa->spa_failmode = intval;
5988 break;
5989 case ZPOOL_PROP_AUTOEXPAND:
5990 spa->spa_autoexpand = intval;
5991 if (tx->tx_txg != TXG_INITIAL)
5992 spa_async_request(spa,
5993 SPA_ASYNC_AUTOEXPAND);
5994 break;
5995 case ZPOOL_PROP_DEDUPDITTO:
5996 spa->spa_dedup_ditto = intval;
5997 break;
5998 default:
5999 break;
6000 }
6001 }
6002
6003 }
6004
6005 mutex_exit(&spa->spa_props_lock);
6006 }
6007
6008 /*
6009 * Perform one-time upgrade on-disk changes. spa_version() does not
6010 * reflect the new version this txg, so there must be no changes this
6011 * txg to anything that the upgrade code depends on after it executes.
6012 * Therefore this must be called after dsl_pool_sync() does the sync
6013 * tasks.
6014 */
6015 static void
6016 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6017 {
6018 dsl_pool_t *dp = spa->spa_dsl_pool;
6019
6020 ASSERT(spa->spa_sync_pass == 1);
6021
6022 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6023
6024 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6025 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6026 dsl_pool_create_origin(dp, tx);
6027
6028 /* Keeping the origin open increases spa_minref */
6029 spa->spa_minref += 3;
6030 }
6031
6032 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6033 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6034 dsl_pool_upgrade_clones(dp, tx);
6035 }
6036
6037 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6038 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6039 dsl_pool_upgrade_dir_clones(dp, tx);
6040
6041 /* Keeping the freedir open increases spa_minref */
6042 spa->spa_minref += 3;
6043 }
6044
6045 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6046 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6047 spa_feature_create_zap_objects(spa, tx);
6048 }
6049 rrw_exit(&dp->dp_config_rwlock, FTAG);
6050 }
6051
6052 /*
6053 * Sync the specified transaction group. New blocks may be dirtied as
6054 * part of the process, so we iterate until it converges.
6055 */
6056 void
6057 spa_sync(spa_t *spa, uint64_t txg)
6058 {
6059 dsl_pool_t *dp = spa->spa_dsl_pool;
6060 objset_t *mos = spa->spa_meta_objset;
6061 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6062 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6063 vdev_t *rvd = spa->spa_root_vdev;
6064 vdev_t *vd;
6065 dmu_tx_t *tx;
6066 int error;
6067
6068 VERIFY(spa_writeable(spa));
6069
6070 /*
6071 * Lock out configuration changes.
6072 */
6073 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6074
6075 spa->spa_syncing_txg = txg;
6076 spa->spa_sync_pass = 0;
6077
6078 /*
6079 * If there are any pending vdev state changes, convert them
6080 * into config changes that go out with this transaction group.
6081 */
6082 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6083 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6084 /*
6085 * We need the write lock here because, for aux vdevs,
6086 * calling vdev_config_dirty() modifies sav_config.
6087 * This is ugly and will become unnecessary when we
6088 * eliminate the aux vdev wart by integrating all vdevs
6089 * into the root vdev tree.
6090 */
6091 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6092 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6093 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6094 vdev_state_clean(vd);
6095 vdev_config_dirty(vd);
6096 }
6097 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6098 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6099 }
6100 spa_config_exit(spa, SCL_STATE, FTAG);
6101
6102 tx = dmu_tx_create_assigned(dp, txg);
6103
6104 spa->spa_sync_starttime = gethrtime();
6105 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6106 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6107
6108 /*
6109 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6110 * set spa_deflate if we have no raid-z vdevs.
6111 */
6112 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6113 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6114 int i;
6115
6116 for (i = 0; i < rvd->vdev_children; i++) {
6117 vd = rvd->vdev_child[i];
6118 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6119 break;
6120 }
6121 if (i == rvd->vdev_children) {
6122 spa->spa_deflate = TRUE;
6123 VERIFY(0 == zap_add(spa->spa_meta_objset,
6124 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6125 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6126 }
6127 }
6128
6129 /*
6130 * If anything has changed in this txg, or if someone is waiting
6131 * for this txg to sync (eg, spa_vdev_remove()), push the
6132 * deferred frees from the previous txg. If not, leave them
6133 * alone so that we don't generate work on an otherwise idle
6134 * system.
6135 */
6136 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6137 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6138 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6139 ((dsl_scan_active(dp->dp_scan) ||
6140 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6141 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6142 VERIFY3U(bpobj_iterate(defer_bpo,
6143 spa_free_sync_cb, zio, tx), ==, 0);
6144 VERIFY0(zio_wait(zio));
6145 }
6146
6147 /*
6148 * Iterate to convergence.
6149 */
6150 do {
6151 int pass = ++spa->spa_sync_pass;
6152
6153 spa_sync_config_object(spa, tx);
6154 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6155 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6156 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6157 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6158 spa_errlog_sync(spa, txg);
6159 dsl_pool_sync(dp, txg);
6160
6161 if (pass < zfs_sync_pass_deferred_free) {
6162 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6163 bplist_iterate(free_bpl, spa_free_sync_cb,
6164 zio, tx);
6165 VERIFY(zio_wait(zio) == 0);
6166 } else {
6167 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6168 defer_bpo, tx);
6169 }
6170
6171 ddt_sync(spa, txg);
6172 dsl_scan_sync(dp, tx);
6173
6174 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6175 vdev_sync(vd, txg);
6176
6177 if (pass == 1)
6178 spa_sync_upgrades(spa, tx);
6179
6180 } while (dmu_objset_is_dirty(mos, txg));
6181
6182 /*
6183 * Rewrite the vdev configuration (which includes the uberblock)
6184 * to commit the transaction group.
6185 *
6186 * If there are no dirty vdevs, we sync the uberblock to a few
6187 * random top-level vdevs that are known to be visible in the
6188 * config cache (see spa_vdev_add() for a complete description).
6189 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6190 */
6191 for (;;) {
6192 /*
6193 * We hold SCL_STATE to prevent vdev open/close/etc.
6194 * while we're attempting to write the vdev labels.
6195 */
6196 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6197
6198 if (list_is_empty(&spa->spa_config_dirty_list)) {
6199 vdev_t *svd[SPA_DVAS_PER_BP];
6200 int svdcount = 0;
6201 int children = rvd->vdev_children;
6202 int c0 = spa_get_random(children);
6203
6204 for (int c = 0; c < children; c++) {
6205 vd = rvd->vdev_child[(c0 + c) % children];
6206 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6207 continue;
6208 svd[svdcount++] = vd;
6209 if (svdcount == SPA_DVAS_PER_BP)
6210 break;
6211 }
6212 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6213 if (error != 0)
6214 error = vdev_config_sync(svd, svdcount, txg,
6215 B_TRUE);
6216 } else {
6217 error = vdev_config_sync(rvd->vdev_child,
6218 rvd->vdev_children, txg, B_FALSE);
6219 if (error != 0)
6220 error = vdev_config_sync(rvd->vdev_child,
6221 rvd->vdev_children, txg, B_TRUE);
6222 }
6223
6224 if (error == 0)
6225 spa->spa_last_synced_guid = rvd->vdev_guid;
6226
6227 spa_config_exit(spa, SCL_STATE, FTAG);
6228
6229 if (error == 0)
6230 break;
6231 zio_suspend(spa, NULL);
6232 zio_resume_wait(spa);
6233 }
6234 dmu_tx_commit(tx);
6235
6236 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6237
6238 /*
6239 * Clear the dirty config list.
6240 */
6241 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6242 vdev_config_clean(vd);
6243
6244 /*
6245 * Now that the new config has synced transactionally,
6246 * let it become visible to the config cache.
6247 */
6248 if (spa->spa_config_syncing != NULL) {
6249 spa_config_set(spa, spa->spa_config_syncing);
6250 spa->spa_config_txg = txg;
6251 spa->spa_config_syncing = NULL;
6252 }
6253
6254 spa->spa_ubsync = spa->spa_uberblock;
6255
6256 dsl_pool_sync_done(dp, txg);
6257
6258 /*
6259 * Update usable space statistics.
6260 */
6261 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6262 vdev_sync_done(vd, txg);
6263
6264 spa_update_dspace(spa);
6265
6266 /*
6267 * It had better be the case that we didn't dirty anything
6268 * since vdev_config_sync().
6269 */
6270 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6271 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6272 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6273
6274 spa->spa_sync_pass = 0;
6275
6276 spa_config_exit(spa, SCL_CONFIG, FTAG);
6277
6278 spa_handle_ignored_writes(spa);
6279
6280 /*
6281 * If any async tasks have been requested, kick them off.
6282 */
6283 spa_async_dispatch(spa);
6284 }
6285
6286 /*
6287 * Sync all pools. We don't want to hold the namespace lock across these
6288 * operations, so we take a reference on the spa_t and drop the lock during the
6289 * sync.
6290 */
6291 void
6292 spa_sync_allpools(void)
6293 {
6294 spa_t *spa = NULL;
6295 mutex_enter(&spa_namespace_lock);
6296 while ((spa = spa_next(spa)) != NULL) {
6297 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6298 !spa_writeable(spa) || spa_suspended(spa))
6299 continue;
6300 spa_open_ref(spa, FTAG);
6301 mutex_exit(&spa_namespace_lock);
6302 txg_wait_synced(spa_get_dsl(spa), 0);
6303 mutex_enter(&spa_namespace_lock);
6304 spa_close(spa, FTAG);
6305 }
6306 mutex_exit(&spa_namespace_lock);
6307 }
6308
6309 /*
6310 * ==========================================================================
6311 * Miscellaneous routines
6312 * ==========================================================================
6313 */
6314
6315 /*
6316 * Remove all pools in the system.
6317 */
6318 void
6319 spa_evict_all(void)
6320 {
6321 spa_t *spa;
6322
6323 /*
6324 * Remove all cached state. All pools should be closed now,
6325 * so every spa in the AVL tree should be unreferenced.
6326 */
6327 mutex_enter(&spa_namespace_lock);
6328 while ((spa = spa_next(NULL)) != NULL) {
6329 /*
6330 * Stop async tasks. The async thread may need to detach
6331 * a device that's been replaced, which requires grabbing
6332 * spa_namespace_lock, so we must drop it here.
6333 */
6334 spa_open_ref(spa, FTAG);
6335 mutex_exit(&spa_namespace_lock);
6336 spa_async_suspend(spa);
6337 mutex_enter(&spa_namespace_lock);
6338 spa_close(spa, FTAG);
6339
6340 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6341 spa_unload(spa);
6342 spa_deactivate(spa);
6343 }
6344 spa_remove(spa);
6345 }
6346 mutex_exit(&spa_namespace_lock);
6347 }
6348
6349 vdev_t *
6350 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6351 {
6352 vdev_t *vd;
6353 int i;
6354
6355 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6356 return (vd);
6357
6358 if (aux) {
6359 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6360 vd = spa->spa_l2cache.sav_vdevs[i];
6361 if (vd->vdev_guid == guid)
6362 return (vd);
6363 }
6364
6365 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6366 vd = spa->spa_spares.sav_vdevs[i];
6367 if (vd->vdev_guid == guid)
6368 return (vd);
6369 }
6370 }
6371
6372 return (NULL);
6373 }
6374
6375 void
6376 spa_upgrade(spa_t *spa, uint64_t version)
6377 {
6378 ASSERT(spa_writeable(spa));
6379
6380 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6381
6382 /*
6383 * This should only be called for a non-faulted pool, and since a
6384 * future version would result in an unopenable pool, this shouldn't be
6385 * possible.
6386 */
6387 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6388 ASSERT(version >= spa->spa_uberblock.ub_version);
6389
6390 spa->spa_uberblock.ub_version = version;
6391 vdev_config_dirty(spa->spa_root_vdev);
6392
6393 spa_config_exit(spa, SCL_ALL, FTAG);
6394
6395 txg_wait_synced(spa_get_dsl(spa), 0);
6396 }
6397
6398 boolean_t
6399 spa_has_spare(spa_t *spa, uint64_t guid)
6400 {
6401 int i;
6402 uint64_t spareguid;
6403 spa_aux_vdev_t *sav = &spa->spa_spares;
6404
6405 for (i = 0; i < sav->sav_count; i++)
6406 if (sav->sav_vdevs[i]->vdev_guid == guid)
6407 return (B_TRUE);
6408
6409 for (i = 0; i < sav->sav_npending; i++) {
6410 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6411 &spareguid) == 0 && spareguid == guid)
6412 return (B_TRUE);
6413 }
6414
6415 return (B_FALSE);
6416 }
6417
6418 /*
6419 * Check if a pool has an active shared spare device.
6420 * Note: reference count of an active spare is 2, as a spare and as a replace
6421 */
6422 static boolean_t
6423 spa_has_active_shared_spare(spa_t *spa)
6424 {
6425 int i, refcnt;
6426 uint64_t pool;
6427 spa_aux_vdev_t *sav = &spa->spa_spares;
6428
6429 for (i = 0; i < sav->sav_count; i++) {
6430 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6431 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6432 refcnt > 2)
6433 return (B_TRUE);
6434 }
6435
6436 return (B_FALSE);
6437 }
6438
6439 /*
6440 * Post a sysevent corresponding to the given event. The 'name' must be one of
6441 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6442 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6443 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6444 * or zdb as real changes.
6445 */
6446 void
6447 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6448 {
6449 #ifdef _KERNEL
6450 sysevent_t *ev;
6451 sysevent_attr_list_t *attr = NULL;
6452 sysevent_value_t value;
6453 sysevent_id_t eid;
6454
6455 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6456 SE_SLEEP);
6457
6458 value.value_type = SE_DATA_TYPE_STRING;
6459 value.value.sv_string = spa_name(spa);
6460 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6461 goto done;
6462
6463 value.value_type = SE_DATA_TYPE_UINT64;
6464 value.value.sv_uint64 = spa_guid(spa);
6465 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6466 goto done;
6467
6468 if (vd) {
6469 value.value_type = SE_DATA_TYPE_UINT64;
6470 value.value.sv_uint64 = vd->vdev_guid;
6471 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6472 SE_SLEEP) != 0)
6473 goto done;
6474
6475 if (vd->vdev_path) {
6476 value.value_type = SE_DATA_TYPE_STRING;
6477 value.value.sv_string = vd->vdev_path;
6478 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6479 &value, SE_SLEEP) != 0)
6480 goto done;
6481 }
6482 }
6483
6484 if (sysevent_attach_attributes(ev, attr) != 0)
6485 goto done;
6486 attr = NULL;
6487
6488 (void) log_sysevent(ev, SE_SLEEP, &eid);
6489
6490 done:
6491 if (attr)
6492 sysevent_free_attr(attr);
6493 sysevent_free(ev);
6494 #endif
6495 }