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 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36 #include <sys/zfs_context.h>
37 #include <sys/fm/fs/zfs.h>
38 #include <sys/spa_impl.h>
39 #include <sys/zio.h>
40 #include <sys/zio_checksum.h>
41 #include <sys/dmu.h>
42 #include <sys/dmu_tx.h>
43 #include <sys/zap.h>
44 #include <sys/zil.h>
45 #include <sys/ddt.h>
46 #include <sys/vdev_impl.h>
47 #include <sys/metaslab.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/uberblock_impl.h>
50 #include <sys/txg.h>
51 #include <sys/avl.h>
52 #include <sys/dmu_traverse.h>
53 #include <sys/dmu_objset.h>
54 #include <sys/unique.h>
55 #include <sys/dsl_pool.h>
56 #include <sys/dsl_dataset.h>
57 #include <sys/dsl_dir.h>
58 #include <sys/dsl_prop.h>
59 #include <sys/dsl_synctask.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/arc.h>
62 #include <sys/callb.h>
63 #include <sys/systeminfo.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/zfeature.h>
68 #include <sys/dsl_destroy.h>
69
70 #ifdef _KERNEL
71 #include <sys/bootprops.h>
72 #include <sys/callb.h>
73 #include <sys/cpupart.h>
74 #include <sys/pool.h>
75 #include <sys/sysdc.h>
76 #include <sys/zone.h>
77 #endif /* _KERNEL */
78
79 #include "zfs_prop.h"
80 #include "zfs_comutil.h"
81
82 /*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86 static int zfs_ccw_retry_interval = 300;
87
88 typedef enum zti_modes {
89 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
90 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
91 ZTI_MODE_NULL, /* don't create a taskq */
92 ZTI_NMODES
93 } zti_modes_t;
94
95 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
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 = 75; /* 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->spa_vdev_top_lock);
759 mutex_enter(&spa_namespace_lock);
760 guid = spa_generate_guid(NULL);
761
762 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763 spa_change_guid_sync, &guid, 5);
764
765 if (error == 0) {
766 spa_config_sync(spa, B_FALSE, B_TRUE);
767 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768 }
769
770 mutex_exit(&spa_namespace_lock);
771 mutex_exit(&spa->spa_vdev_top_lock);
772
773 return (error);
774 }
775
776 /*
777 * ==========================================================================
778 * SPA state manipulation (open/create/destroy/import/export)
779 * ==========================================================================
780 */
781
782 static int
783 spa_error_entry_compare(const void *a, const void *b)
784 {
785 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 int ret;
788
789 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 sizeof (zbookmark_t));
791
792 if (ret < 0)
793 return (-1);
794 else if (ret > 0)
795 return (1);
796 else
797 return (0);
798 }
799
800 /*
801 * Utility function which retrieves copies of the current logs and
802 * re-initializes them in the process.
803 */
804 void
805 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 {
807 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808
809 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811
812 avl_create(&spa->spa_errlist_scrub,
813 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 offsetof(spa_error_entry_t, se_avl));
815 avl_create(&spa->spa_errlist_last,
816 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 offsetof(spa_error_entry_t, se_avl));
818 }
819
820 static void
821 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 {
823 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 enum zti_modes mode = ztip->zti_mode;
825 uint_t value = ztip->zti_value;
826 uint_t count = ztip->zti_count;
827 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 char name[32];
829 uint_t flags = 0;
830 boolean_t batch = B_FALSE;
831
832 if (mode == ZTI_MODE_NULL) {
833 tqs->stqs_count = 0;
834 tqs->stqs_taskq = NULL;
835 return;
836 }
837
838 ASSERT3U(count, >, 0);
839
840 tqs->stqs_count = count;
841 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842
843 switch (mode) {
844 case ZTI_MODE_FIXED:
845 ASSERT3U(value, >=, 1);
846 value = MAX(value, 1);
847 break;
848
849 case ZTI_MODE_BATCH:
850 batch = B_TRUE;
851 flags |= TASKQ_THREADS_CPU_PCT;
852 value = zio_taskq_batch_pct;
853 break;
854
855 default:
856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 "spa_activate()",
858 zio_type_name[t], zio_taskq_types[q], mode, value);
859 break;
860 }
861
862 for (uint_t i = 0; i < count; i++) {
863 taskq_t *tq;
864
865 if (count > 1) {
866 (void) snprintf(name, sizeof (name), "%s_%s_%u",
867 zio_type_name[t], zio_taskq_types[q], i);
868 } else {
869 (void) snprintf(name, sizeof (name), "%s_%s",
870 zio_type_name[t], zio_taskq_types[q]);
871 }
872
873 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874 if (batch)
875 flags |= TASKQ_DC_BATCH;
876
877 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878 spa->spa_proc, zio_taskq_basedc, flags);
879 } else {
880 pri_t pri = maxclsyspri;
881 /*
882 * The write issue taskq can be extremely CPU
883 * intensive. Run it at slightly lower priority
884 * than the other taskqs.
885 */
886 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887 pri--;
888
889 tq = taskq_create_proc(name, value, pri, 50,
890 INT_MAX, spa->spa_proc, flags);
891 }
892
893 tqs->stqs_taskq[i] = tq;
894 }
895 }
896
897 static void
898 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899 {
900 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901
902 if (tqs->stqs_taskq == NULL) {
903 ASSERT0(tqs->stqs_count);
904 return;
905 }
906
907 for (uint_t i = 0; i < tqs->stqs_count; i++) {
908 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909 taskq_destroy(tqs->stqs_taskq[i]);
910 }
911
912 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913 tqs->stqs_taskq = NULL;
914 }
915
916 /*
917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 * on the taskq itself. In that case we choose which taskq at random by using
920 * the low bits of gethrtime().
921 */
922 void
923 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925 {
926 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927 taskq_t *tq;
928
929 ASSERT3P(tqs->stqs_taskq, !=, NULL);
930 ASSERT3U(tqs->stqs_count, !=, 0);
931
932 if (tqs->stqs_count == 1) {
933 tq = tqs->stqs_taskq[0];
934 } else {
935 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936 }
937
938 taskq_dispatch_ent(tq, func, arg, flags, ent);
939 }
940
941 static void
942 spa_create_zio_taskqs(spa_t *spa)
943 {
944 for (int t = 0; t < ZIO_TYPES; t++) {
945 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946 spa_taskqs_init(spa, t, q);
947 }
948 }
949 }
950
951 #ifdef _KERNEL
952 static void
953 spa_thread(void *arg)
954 {
955 callb_cpr_t cprinfo;
956
957 spa_t *spa = arg;
958 user_t *pu = PTOU(curproc);
959
960 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961 spa->spa_name);
962
963 ASSERT(curproc != &p0);
964 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965 "zpool-%s", spa->spa_name);
966 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967
968 /* bind this thread to the requested psrset */
969 if (zio_taskq_psrset_bind != PS_NONE) {
970 pool_lock();
971 mutex_enter(&cpu_lock);
972 mutex_enter(&pidlock);
973 mutex_enter(&curproc->p_lock);
974
975 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976 0, NULL, NULL) == 0) {
977 curthread->t_bind_pset = zio_taskq_psrset_bind;
978 } else {
979 cmn_err(CE_WARN,
980 "Couldn't bind process for zfs pool \"%s\" to "
981 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982 }
983
984 mutex_exit(&curproc->p_lock);
985 mutex_exit(&pidlock);
986 mutex_exit(&cpu_lock);
987 pool_unlock();
988 }
989
990 if (zio_taskq_sysdc) {
991 sysdc_thread_enter(curthread, 100, 0);
992 }
993
994 spa->spa_proc = curproc;
995 spa->spa_did = curthread->t_did;
996
997 spa_create_zio_taskqs(spa);
998
999 mutex_enter(&spa->spa_proc_lock);
1000 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001
1002 spa->spa_proc_state = SPA_PROC_ACTIVE;
1003 cv_broadcast(&spa->spa_proc_cv);
1004
1005 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009
1010 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011 spa->spa_proc_state = SPA_PROC_GONE;
1012 spa->spa_proc = &p0;
1013 cv_broadcast(&spa->spa_proc_cv);
1014 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1015
1016 mutex_enter(&curproc->p_lock);
1017 lwp_exit();
1018 }
1019 #endif
1020
1021 /*
1022 * Activate an uninitialized pool.
1023 */
1024 static void
1025 spa_activate(spa_t *spa, int mode)
1026 {
1027 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028
1029 spa->spa_state = POOL_STATE_ACTIVE;
1030 spa->spa_mode = mode;
1031
1032 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034
1035 /* Try to create a covering process */
1036 mutex_enter(&spa->spa_proc_lock);
1037 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038 ASSERT(spa->spa_proc == &p0);
1039 spa->spa_did = 0;
1040
1041 /* Only create a process if we're going to be around a while. */
1042 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044 NULL, 0) == 0) {
1045 spa->spa_proc_state = SPA_PROC_CREATED;
1046 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047 cv_wait(&spa->spa_proc_cv,
1048 &spa->spa_proc_lock);
1049 }
1050 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051 ASSERT(spa->spa_proc != &p0);
1052 ASSERT(spa->spa_did != 0);
1053 } else {
1054 #ifdef _KERNEL
1055 cmn_err(CE_WARN,
1056 "Couldn't create process for zfs pool \"%s\"\n",
1057 spa->spa_name);
1058 #endif
1059 }
1060 }
1061 mutex_exit(&spa->spa_proc_lock);
1062
1063 /* If we didn't create a process, we need to create our taskqs. */
1064 if (spa->spa_proc == &p0) {
1065 spa_create_zio_taskqs(spa);
1066 }
1067
1068 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069 offsetof(vdev_t, vdev_config_dirty_node));
1070 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071 offsetof(vdev_t, vdev_state_dirty_node));
1072
1073 txg_list_create(&spa->spa_vdev_txg_list,
1074 offsetof(struct vdev, vdev_txg_node));
1075
1076 avl_create(&spa->spa_errlist_scrub,
1077 spa_error_entry_compare, sizeof (spa_error_entry_t),
1078 offsetof(spa_error_entry_t, se_avl));
1079 avl_create(&spa->spa_errlist_last,
1080 spa_error_entry_compare, sizeof (spa_error_entry_t),
1081 offsetof(spa_error_entry_t, se_avl));
1082 }
1083
1084 /*
1085 * Opposite of spa_activate().
1086 */
1087 static void
1088 spa_deactivate(spa_t *spa)
1089 {
1090 ASSERT(spa->spa_sync_on == B_FALSE);
1091 ASSERT(spa->spa_dsl_pool == NULL);
1092 ASSERT(spa->spa_root_vdev == NULL);
1093 ASSERT(spa->spa_async_zio_root == NULL);
1094 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095
1096 txg_list_destroy(&spa->spa_vdev_txg_list);
1097
1098 list_destroy(&spa->spa_config_dirty_list);
1099 list_destroy(&spa->spa_state_dirty_list);
1100
1101 for (int t = 0; t < ZIO_TYPES; t++) {
1102 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103 spa_taskqs_fini(spa, t, q);
1104 }
1105 }
1106
1107 metaslab_class_destroy(spa->spa_normal_class);
1108 spa->spa_normal_class = NULL;
1109
1110 metaslab_class_destroy(spa->spa_log_class);
1111 spa->spa_log_class = NULL;
1112
1113 /*
1114 * If this was part of an import or the open otherwise failed, we may
1115 * still have errors left in the queues. Empty them just in case.
1116 */
1117 spa_errlog_drain(spa);
1118
1119 avl_destroy(&spa->spa_errlist_scrub);
1120 avl_destroy(&spa->spa_errlist_last);
1121
1122 spa->spa_state = POOL_STATE_UNINITIALIZED;
1123
1124 mutex_enter(&spa->spa_proc_lock);
1125 if (spa->spa_proc_state != SPA_PROC_NONE) {
1126 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128 cv_broadcast(&spa->spa_proc_cv);
1129 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130 ASSERT(spa->spa_proc != &p0);
1131 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132 }
1133 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134 spa->spa_proc_state = SPA_PROC_NONE;
1135 }
1136 ASSERT(spa->spa_proc == &p0);
1137 mutex_exit(&spa->spa_proc_lock);
1138
1139 /*
1140 * We want to make sure spa_thread() has actually exited the ZFS
1141 * module, so that the module can't be unloaded out from underneath
1142 * it.
1143 */
1144 if (spa->spa_did != 0) {
1145 thread_join(spa->spa_did);
1146 spa->spa_did = 0;
1147 }
1148 }
1149
1150 /*
1151 * Verify a pool configuration, and construct the vdev tree appropriately. This
1152 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 * in the CLOSED state. This will prep the pool before open/creation/import.
1154 * All vdev validation is done by the vdev_alloc() routine.
1155 */
1156 static int
1157 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158 uint_t id, int atype)
1159 {
1160 nvlist_t **child;
1161 uint_t children;
1162 int error;
1163
1164 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165 return (error);
1166
1167 if ((*vdp)->vdev_ops->vdev_op_leaf)
1168 return (0);
1169
1170 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171 &child, &children);
1172
1173 if (error == ENOENT)
1174 return (0);
1175
1176 if (error) {
1177 vdev_free(*vdp);
1178 *vdp = NULL;
1179 return (SET_ERROR(EINVAL));
1180 }
1181
1182 for (int c = 0; c < children; c++) {
1183 vdev_t *vd;
1184 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185 atype)) != 0) {
1186 vdev_free(*vdp);
1187 *vdp = NULL;
1188 return (error);
1189 }
1190 }
1191
1192 ASSERT(*vdp != NULL);
1193
1194 return (0);
1195 }
1196
1197 /*
1198 * Opposite of spa_load().
1199 */
1200 static void
1201 spa_unload(spa_t *spa)
1202 {
1203 int i;
1204
1205 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206
1207 /*
1208 * Stop async tasks.
1209 */
1210 spa_async_suspend(spa);
1211
1212 /*
1213 * Stop syncing.
1214 */
1215 if (spa->spa_sync_on) {
1216 txg_sync_stop(spa->spa_dsl_pool);
1217 spa->spa_sync_on = B_FALSE;
1218 }
1219
1220 /*
1221 * Wait for any outstanding async I/O to complete.
1222 */
1223 if (spa->spa_async_zio_root != NULL) {
1224 (void) zio_wait(spa->spa_async_zio_root);
1225 spa->spa_async_zio_root = NULL;
1226 }
1227
1228 bpobj_close(&spa->spa_deferred_bpobj);
1229
1230 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231
1232 /*
1233 * Close all vdevs.
1234 */
1235 if (spa->spa_root_vdev)
1236 vdev_free(spa->spa_root_vdev);
1237 ASSERT(spa->spa_root_vdev == NULL);
1238
1239 /*
1240 * Close the dsl pool.
1241 */
1242 if (spa->spa_dsl_pool) {
1243 dsl_pool_close(spa->spa_dsl_pool);
1244 spa->spa_dsl_pool = NULL;
1245 spa->spa_meta_objset = NULL;
1246 }
1247
1248 ddt_unload(spa);
1249
1250
1251 /*
1252 * Drop and purge level 2 cache
1253 */
1254 spa_l2cache_drop(spa);
1255
1256 for (i = 0; i < spa->spa_spares.sav_count; i++)
1257 vdev_free(spa->spa_spares.sav_vdevs[i]);
1258 if (spa->spa_spares.sav_vdevs) {
1259 kmem_free(spa->spa_spares.sav_vdevs,
1260 spa->spa_spares.sav_count * sizeof (void *));
1261 spa->spa_spares.sav_vdevs = NULL;
1262 }
1263 if (spa->spa_spares.sav_config) {
1264 nvlist_free(spa->spa_spares.sav_config);
1265 spa->spa_spares.sav_config = NULL;
1266 }
1267 spa->spa_spares.sav_count = 0;
1268
1269 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272 }
1273 if (spa->spa_l2cache.sav_vdevs) {
1274 kmem_free(spa->spa_l2cache.sav_vdevs,
1275 spa->spa_l2cache.sav_count * sizeof (void *));
1276 spa->spa_l2cache.sav_vdevs = NULL;
1277 }
1278 if (spa->spa_l2cache.sav_config) {
1279 nvlist_free(spa->spa_l2cache.sav_config);
1280 spa->spa_l2cache.sav_config = NULL;
1281 }
1282 spa->spa_l2cache.sav_count = 0;
1283
1284 spa->spa_async_suspended = 0;
1285
1286 if (spa->spa_comment != NULL) {
1287 spa_strfree(spa->spa_comment);
1288 spa->spa_comment = NULL;
1289 }
1290
1291 spa_config_exit(spa, SCL_ALL, FTAG);
1292 }
1293
1294 /*
1295 * Load (or re-load) the current list of vdevs describing the active spares for
1296 * this pool. When this is called, we have some form of basic information in
1297 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1298 * then re-generate a more complete list including status information.
1299 */
1300 static void
1301 spa_load_spares(spa_t *spa)
1302 {
1303 nvlist_t **spares;
1304 uint_t nspares;
1305 int i;
1306 vdev_t *vd, *tvd;
1307
1308 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309
1310 /*
1311 * First, close and free any existing spare vdevs.
1312 */
1313 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314 vd = spa->spa_spares.sav_vdevs[i];
1315
1316 /* Undo the call to spa_activate() below */
1317 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318 B_FALSE)) != NULL && tvd->vdev_isspare)
1319 spa_spare_remove(tvd);
1320 vdev_close(vd);
1321 vdev_free(vd);
1322 }
1323
1324 if (spa->spa_spares.sav_vdevs)
1325 kmem_free(spa->spa_spares.sav_vdevs,
1326 spa->spa_spares.sav_count * sizeof (void *));
1327
1328 if (spa->spa_spares.sav_config == NULL)
1329 nspares = 0;
1330 else
1331 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333
1334 spa->spa_spares.sav_count = (int)nspares;
1335 spa->spa_spares.sav_vdevs = NULL;
1336
1337 if (nspares == 0)
1338 return;
1339
1340 /*
1341 * Construct the array of vdevs, opening them to get status in the
1342 * process. For each spare, there is potentially two different vdev_t
1343 * structures associated with it: one in the list of spares (used only
1344 * for basic validation purposes) and one in the active vdev
1345 * configuration (if it's spared in). During this phase we open and
1346 * validate each vdev on the spare list. If the vdev also exists in the
1347 * active configuration, then we also mark this vdev as an active spare.
1348 */
1349 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350 KM_SLEEP);
1351 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353 VDEV_ALLOC_SPARE) == 0);
1354 ASSERT(vd != NULL);
1355
1356 spa->spa_spares.sav_vdevs[i] = vd;
1357
1358 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359 B_FALSE)) != NULL) {
1360 if (!tvd->vdev_isspare)
1361 spa_spare_add(tvd);
1362
1363 /*
1364 * We only mark the spare active if we were successfully
1365 * able to load the vdev. Otherwise, importing a pool
1366 * with a bad active spare would result in strange
1367 * behavior, because multiple pool would think the spare
1368 * is actively in use.
1369 *
1370 * There is a vulnerability here to an equally bizarre
1371 * circumstance, where a dead active spare is later
1372 * brought back to life (onlined or otherwise). Given
1373 * the rarity of this scenario, and the extra complexity
1374 * it adds, we ignore the possibility.
1375 */
1376 if (!vdev_is_dead(tvd))
1377 spa_spare_activate(tvd);
1378 }
1379
1380 vd->vdev_top = vd;
1381 vd->vdev_aux = &spa->spa_spares;
1382
1383 if (vdev_open(vd) != 0)
1384 continue;
1385
1386 if (vdev_validate_aux(vd) == 0)
1387 spa_spare_add(vd);
1388 }
1389
1390 /*
1391 * Recompute the stashed list of spares, with status information
1392 * this time.
1393 */
1394 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395 DATA_TYPE_NVLIST_ARRAY) == 0);
1396
1397 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398 KM_SLEEP);
1399 for (i = 0; i < spa->spa_spares.sav_count; i++)
1400 spares[i] = vdev_config_generate(spa,
1401 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404 for (i = 0; i < spa->spa_spares.sav_count; i++)
1405 nvlist_free(spares[i]);
1406 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407 }
1408
1409 /*
1410 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 * this pool. When this is called, we have some form of basic information in
1412 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1413 * then re-generate a more complete list including status information.
1414 * Devices which are already active have their details maintained, and are
1415 * not re-opened.
1416 */
1417 static void
1418 spa_load_l2cache(spa_t *spa)
1419 {
1420 nvlist_t **l2cache;
1421 uint_t nl2cache;
1422 int i, j, oldnvdevs;
1423 uint64_t guid;
1424 vdev_t *vd, **oldvdevs, **newvdevs;
1425 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426
1427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428
1429 if (sav->sav_config != NULL) {
1430 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433 } else {
1434 nl2cache = 0;
1435 newvdevs = NULL;
1436 }
1437
1438 oldvdevs = sav->sav_vdevs;
1439 oldnvdevs = sav->sav_count;
1440 sav->sav_vdevs = NULL;
1441 sav->sav_count = 0;
1442
1443 /*
1444 * Process new nvlist of vdevs.
1445 */
1446 for (i = 0; i < nl2cache; i++) {
1447 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448 &guid) == 0);
1449
1450 newvdevs[i] = NULL;
1451 for (j = 0; j < oldnvdevs; j++) {
1452 vd = oldvdevs[j];
1453 if (vd != NULL && guid == vd->vdev_guid) {
1454 /*
1455 * Retain previous vdev for add/remove ops.
1456 */
1457 newvdevs[i] = vd;
1458 oldvdevs[j] = NULL;
1459 break;
1460 }
1461 }
1462
1463 if (newvdevs[i] == NULL) {
1464 /*
1465 * Create new vdev
1466 */
1467 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468 VDEV_ALLOC_L2CACHE) == 0);
1469 ASSERT(vd != NULL);
1470 newvdevs[i] = vd;
1471
1472 /*
1473 * Commit this vdev as an l2cache device,
1474 * even if it fails to open.
1475 */
1476 spa_l2cache_add(vd);
1477
1478 vd->vdev_top = vd;
1479 vd->vdev_aux = sav;
1480
1481 spa_l2cache_activate(vd);
1482
1483 if (vdev_open(vd) != 0)
1484 continue;
1485
1486 (void) vdev_validate_aux(vd);
1487
1488 if (!vdev_is_dead(vd))
1489 l2arc_add_vdev(spa, vd);
1490 }
1491 }
1492
1493 /*
1494 * Purge vdevs that were dropped
1495 */
1496 for (i = 0; i < oldnvdevs; i++) {
1497 uint64_t pool;
1498
1499 vd = oldvdevs[i];
1500 if (vd != NULL) {
1501 ASSERT(vd->vdev_isl2cache);
1502
1503 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1504 pool != 0ULL && l2arc_vdev_present(vd))
1505 l2arc_remove_vdev(vd);
1506 vdev_clear_stats(vd);
1507 vdev_free(vd);
1508 }
1509 }
1510
1511 if (oldvdevs)
1512 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1513
1514 if (sav->sav_config == NULL)
1515 goto out;
1516
1517 sav->sav_vdevs = newvdevs;
1518 sav->sav_count = (int)nl2cache;
1519
1520 /*
1521 * Recompute the stashed list of l2cache devices, with status
1522 * information this time.
1523 */
1524 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1525 DATA_TYPE_NVLIST_ARRAY) == 0);
1526
1527 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1528 for (i = 0; i < sav->sav_count; i++)
1529 l2cache[i] = vdev_config_generate(spa,
1530 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1531 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1532 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1533 out:
1534 for (i = 0; i < sav->sav_count; i++)
1535 nvlist_free(l2cache[i]);
1536 if (sav->sav_count)
1537 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1538 }
1539
1540 static int
1541 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1542 {
1543 dmu_buf_t *db;
1544 char *packed = NULL;
1545 size_t nvsize = 0;
1546 int error;
1547 *value = NULL;
1548
1549 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1550 nvsize = *(uint64_t *)db->db_data;
1551 dmu_buf_rele(db, FTAG);
1552
1553 packed = kmem_alloc(nvsize, KM_SLEEP);
1554 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1555 DMU_READ_PREFETCH);
1556 if (error == 0)
1557 error = nvlist_unpack(packed, nvsize, value, 0);
1558 kmem_free(packed, nvsize);
1559
1560 return (error);
1561 }
1562
1563 /*
1564 * Checks to see if the given vdev could not be opened, in which case we post a
1565 * sysevent to notify the autoreplace code that the device has been removed.
1566 */
1567 static void
1568 spa_check_removed(vdev_t *vd)
1569 {
1570 for (int c = 0; c < vd->vdev_children; c++)
1571 spa_check_removed(vd->vdev_child[c]);
1572
1573 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1574 !vd->vdev_ishole) {
1575 zfs_post_autoreplace(vd->vdev_spa, vd);
1576 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1577 }
1578 }
1579
1580 /*
1581 * Validate the current config against the MOS config
1582 */
1583 static boolean_t
1584 spa_config_valid(spa_t *spa, nvlist_t *config)
1585 {
1586 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1587 nvlist_t *nv;
1588
1589 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1590
1591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1592 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1593
1594 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1595
1596 /*
1597 * If we're doing a normal import, then build up any additional
1598 * diagnostic information about missing devices in this config.
1599 * We'll pass this up to the user for further processing.
1600 */
1601 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1602 nvlist_t **child, *nv;
1603 uint64_t idx = 0;
1604
1605 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1606 KM_SLEEP);
1607 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1608
1609 for (int c = 0; c < rvd->vdev_children; c++) {
1610 vdev_t *tvd = rvd->vdev_child[c];
1611 vdev_t *mtvd = mrvd->vdev_child[c];
1612
1613 if (tvd->vdev_ops == &vdev_missing_ops &&
1614 mtvd->vdev_ops != &vdev_missing_ops &&
1615 mtvd->vdev_islog)
1616 child[idx++] = vdev_config_generate(spa, mtvd,
1617 B_FALSE, 0);
1618 }
1619
1620 if (idx) {
1621 VERIFY(nvlist_add_nvlist_array(nv,
1622 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1623 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1624 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1625
1626 for (int i = 0; i < idx; i++)
1627 nvlist_free(child[i]);
1628 }
1629 nvlist_free(nv);
1630 kmem_free(child, rvd->vdev_children * sizeof (char **));
1631 }
1632
1633 /*
1634 * Compare the root vdev tree with the information we have
1635 * from the MOS config (mrvd). Check each top-level vdev
1636 * with the corresponding MOS config top-level (mtvd).
1637 */
1638 for (int c = 0; c < rvd->vdev_children; c++) {
1639 vdev_t *tvd = rvd->vdev_child[c];
1640 vdev_t *mtvd = mrvd->vdev_child[c];
1641
1642 /*
1643 * Resolve any "missing" vdevs in the current configuration.
1644 * If we find that the MOS config has more accurate information
1645 * about the top-level vdev then use that vdev instead.
1646 */
1647 if (tvd->vdev_ops == &vdev_missing_ops &&
1648 mtvd->vdev_ops != &vdev_missing_ops) {
1649
1650 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1651 continue;
1652
1653 /*
1654 * Device specific actions.
1655 */
1656 if (mtvd->vdev_islog) {
1657 spa_set_log_state(spa, SPA_LOG_CLEAR);
1658 } else {
1659 /*
1660 * XXX - once we have 'readonly' pool
1661 * support we should be able to handle
1662 * missing data devices by transitioning
1663 * the pool to readonly.
1664 */
1665 continue;
1666 }
1667
1668 /*
1669 * Swap the missing vdev with the data we were
1670 * able to obtain from the MOS config.
1671 */
1672 vdev_remove_child(rvd, tvd);
1673 vdev_remove_child(mrvd, mtvd);
1674
1675 vdev_add_child(rvd, mtvd);
1676 vdev_add_child(mrvd, tvd);
1677
1678 spa_config_exit(spa, SCL_ALL, FTAG);
1679 vdev_load(mtvd);
1680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681
1682 vdev_reopen(rvd);
1683 } else if (mtvd->vdev_islog) {
1684 /*
1685 * Load the slog device's state from the MOS config
1686 * since it's possible that the label does not
1687 * contain the most up-to-date information.
1688 */
1689 vdev_load_log_state(tvd, mtvd);
1690 vdev_reopen(tvd);
1691 }
1692 }
1693 vdev_free(mrvd);
1694 spa_config_exit(spa, SCL_ALL, FTAG);
1695
1696 /*
1697 * Ensure we were able to validate the config.
1698 */
1699 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1700 }
1701
1702 /*
1703 * Check for missing log devices
1704 */
1705 static boolean_t
1706 spa_check_logs(spa_t *spa)
1707 {
1708 boolean_t rv = B_FALSE;
1709
1710 switch (spa->spa_log_state) {
1711 case SPA_LOG_MISSING:
1712 /* need to recheck in case slog has been restored */
1713 case SPA_LOG_UNKNOWN:
1714 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1715 NULL, DS_FIND_CHILDREN) != 0);
1716 if (rv)
1717 spa_set_log_state(spa, SPA_LOG_MISSING);
1718 break;
1719 }
1720 return (rv);
1721 }
1722
1723 static boolean_t
1724 spa_passivate_log(spa_t *spa)
1725 {
1726 vdev_t *rvd = spa->spa_root_vdev;
1727 boolean_t slog_found = B_FALSE;
1728
1729 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1730
1731 if (!spa_has_slogs(spa))
1732 return (B_FALSE);
1733
1734 for (int c = 0; c < rvd->vdev_children; c++) {
1735 vdev_t *tvd = rvd->vdev_child[c];
1736 metaslab_group_t *mg = tvd->vdev_mg;
1737
1738 if (tvd->vdev_islog) {
1739 metaslab_group_passivate(mg);
1740 slog_found = B_TRUE;
1741 }
1742 }
1743
1744 return (slog_found);
1745 }
1746
1747 static void
1748 spa_activate_log(spa_t *spa)
1749 {
1750 vdev_t *rvd = spa->spa_root_vdev;
1751
1752 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1753
1754 for (int c = 0; c < rvd->vdev_children; c++) {
1755 vdev_t *tvd = rvd->vdev_child[c];
1756 metaslab_group_t *mg = tvd->vdev_mg;
1757
1758 if (tvd->vdev_islog)
1759 metaslab_group_activate(mg);
1760 }
1761 }
1762
1763 int
1764 spa_offline_log(spa_t *spa)
1765 {
1766 int error;
1767
1768 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1769 NULL, DS_FIND_CHILDREN);
1770 if (error == 0) {
1771 /*
1772 * We successfully offlined the log device, sync out the
1773 * current txg so that the "stubby" block can be removed
1774 * by zil_sync().
1775 */
1776 txg_wait_synced(spa->spa_dsl_pool, 0);
1777 }
1778 return (error);
1779 }
1780
1781 static void
1782 spa_aux_check_removed(spa_aux_vdev_t *sav)
1783 {
1784 for (int i = 0; i < sav->sav_count; i++)
1785 spa_check_removed(sav->sav_vdevs[i]);
1786 }
1787
1788 void
1789 spa_claim_notify(zio_t *zio)
1790 {
1791 spa_t *spa = zio->io_spa;
1792
1793 if (zio->io_error)
1794 return;
1795
1796 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1797 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1798 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1799 mutex_exit(&spa->spa_props_lock);
1800 }
1801
1802 typedef struct spa_load_error {
1803 uint64_t sle_meta_count;
1804 uint64_t sle_data_count;
1805 } spa_load_error_t;
1806
1807 static void
1808 spa_load_verify_done(zio_t *zio)
1809 {
1810 blkptr_t *bp = zio->io_bp;
1811 spa_load_error_t *sle = zio->io_private;
1812 dmu_object_type_t type = BP_GET_TYPE(bp);
1813 int error = zio->io_error;
1814
1815 if (error) {
1816 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1817 type != DMU_OT_INTENT_LOG)
1818 atomic_add_64(&sle->sle_meta_count, 1);
1819 else
1820 atomic_add_64(&sle->sle_data_count, 1);
1821 }
1822 zio_data_buf_free(zio->io_data, zio->io_size);
1823 }
1824
1825 /*ARGSUSED*/
1826 static int
1827 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1828 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1829 {
1830 if (bp != NULL) {
1831 zio_t *rio = arg;
1832 size_t size = BP_GET_PSIZE(bp);
1833 void *data = zio_data_buf_alloc(size);
1834
1835 zio_nowait(zio_read(rio, spa, bp, data, size,
1836 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1837 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1838 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1839 }
1840 return (0);
1841 }
1842
1843 static int
1844 spa_load_verify(spa_t *spa)
1845 {
1846 zio_t *rio;
1847 spa_load_error_t sle = { 0 };
1848 zpool_rewind_policy_t policy;
1849 boolean_t verify_ok = B_FALSE;
1850 int error;
1851
1852 zpool_get_rewind_policy(spa->spa_config, &policy);
1853
1854 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1855 return (0);
1856
1857 rio = zio_root(spa, NULL, &sle,
1858 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1859
1860 error = traverse_pool(spa, spa->spa_verify_min_txg,
1861 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1862
1863 (void) zio_wait(rio);
1864
1865 spa->spa_load_meta_errors = sle.sle_meta_count;
1866 spa->spa_load_data_errors = sle.sle_data_count;
1867
1868 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1869 sle.sle_data_count <= policy.zrp_maxdata) {
1870 int64_t loss = 0;
1871
1872 verify_ok = B_TRUE;
1873 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1874 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1875
1876 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1877 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1879 VERIFY(nvlist_add_int64(spa->spa_load_info,
1880 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1881 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1882 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1883 } else {
1884 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1885 }
1886
1887 if (error) {
1888 if (error != ENXIO && error != EIO)
1889 error = SET_ERROR(EIO);
1890 return (error);
1891 }
1892
1893 return (verify_ok ? 0 : EIO);
1894 }
1895
1896 /*
1897 * Find a value in the pool props object.
1898 */
1899 static void
1900 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1901 {
1902 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1903 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1904 }
1905
1906 /*
1907 * Find a value in the pool directory object.
1908 */
1909 static int
1910 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1911 {
1912 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1913 name, sizeof (uint64_t), 1, val));
1914 }
1915
1916 static int
1917 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1918 {
1919 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1920 return (err);
1921 }
1922
1923 /*
1924 * Fix up config after a partly-completed split. This is done with the
1925 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1926 * pool have that entry in their config, but only the splitting one contains
1927 * a list of all the guids of the vdevs that are being split off.
1928 *
1929 * This function determines what to do with that list: either rejoin
1930 * all the disks to the pool, or complete the splitting process. To attempt
1931 * the rejoin, each disk that is offlined is marked online again, and
1932 * we do a reopen() call. If the vdev label for every disk that was
1933 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1934 * then we call vdev_split() on each disk, and complete the split.
1935 *
1936 * Otherwise we leave the config alone, with all the vdevs in place in
1937 * the original pool.
1938 */
1939 static void
1940 spa_try_repair(spa_t *spa, nvlist_t *config)
1941 {
1942 uint_t extracted;
1943 uint64_t *glist;
1944 uint_t i, gcount;
1945 nvlist_t *nvl;
1946 vdev_t **vd;
1947 boolean_t attempt_reopen;
1948
1949 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1950 return;
1951
1952 /* check that the config is complete */
1953 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1954 &glist, &gcount) != 0)
1955 return;
1956
1957 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1958
1959 /* attempt to online all the vdevs & validate */
1960 attempt_reopen = B_TRUE;
1961 for (i = 0; i < gcount; i++) {
1962 if (glist[i] == 0) /* vdev is hole */
1963 continue;
1964
1965 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1966 if (vd[i] == NULL) {
1967 /*
1968 * Don't bother attempting to reopen the disks;
1969 * just do the split.
1970 */
1971 attempt_reopen = B_FALSE;
1972 } else {
1973 /* attempt to re-online it */
1974 vd[i]->vdev_offline = B_FALSE;
1975 }
1976 }
1977
1978 if (attempt_reopen) {
1979 vdev_reopen(spa->spa_root_vdev);
1980
1981 /* check each device to see what state it's in */
1982 for (extracted = 0, i = 0; i < gcount; i++) {
1983 if (vd[i] != NULL &&
1984 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1985 break;
1986 ++extracted;
1987 }
1988 }
1989
1990 /*
1991 * If every disk has been moved to the new pool, or if we never
1992 * even attempted to look at them, then we split them off for
1993 * good.
1994 */
1995 if (!attempt_reopen || gcount == extracted) {
1996 for (i = 0; i < gcount; i++)
1997 if (vd[i] != NULL)
1998 vdev_split(vd[i]);
1999 vdev_reopen(spa->spa_root_vdev);
2000 }
2001
2002 kmem_free(vd, gcount * sizeof (vdev_t *));
2003 }
2004
2005 static int
2006 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2007 boolean_t mosconfig)
2008 {
2009 nvlist_t *config = spa->spa_config;
2010 char *ereport = FM_EREPORT_ZFS_POOL;
2011 char *comment;
2012 int error;
2013 uint64_t pool_guid;
2014 nvlist_t *nvl;
2015
2016 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2017 return (SET_ERROR(EINVAL));
2018
2019 ASSERT(spa->spa_comment == NULL);
2020 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2021 spa->spa_comment = spa_strdup(comment);
2022
2023 /*
2024 * Versioning wasn't explicitly added to the label until later, so if
2025 * it's not present treat it as the initial version.
2026 */
2027 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2028 &spa->spa_ubsync.ub_version) != 0)
2029 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2030
2031 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2032 &spa->spa_config_txg);
2033
2034 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2035 spa_guid_exists(pool_guid, 0)) {
2036 error = SET_ERROR(EEXIST);
2037 } else {
2038 spa->spa_config_guid = pool_guid;
2039
2040 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2041 &nvl) == 0) {
2042 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2043 KM_SLEEP) == 0);
2044 }
2045
2046 nvlist_free(spa->spa_load_info);
2047 spa->spa_load_info = fnvlist_alloc();
2048
2049 gethrestime(&spa->spa_loaded_ts);
2050 error = spa_load_impl(spa, pool_guid, config, state, type,
2051 mosconfig, &ereport);
2052 }
2053
2054 spa->spa_minref = refcount_count(&spa->spa_refcount);
2055 if (error) {
2056 if (error != EEXIST) {
2057 spa->spa_loaded_ts.tv_sec = 0;
2058 spa->spa_loaded_ts.tv_nsec = 0;
2059 }
2060 if (error != EBADF) {
2061 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2062 }
2063 }
2064 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2065 spa->spa_ena = 0;
2066
2067 return (error);
2068 }
2069
2070 /*
2071 * Load an existing storage pool, using the pool's builtin spa_config as a
2072 * source of configuration information.
2073 */
2074 static int
2075 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2076 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2077 char **ereport)
2078 {
2079 int error = 0;
2080 nvlist_t *nvroot = NULL;
2081 nvlist_t *label;
2082 vdev_t *rvd;
2083 uberblock_t *ub = &spa->spa_uberblock;
2084 uint64_t children, config_cache_txg = spa->spa_config_txg;
2085 int orig_mode = spa->spa_mode;
2086 int parse;
2087 uint64_t obj;
2088 boolean_t missing_feat_write = B_FALSE;
2089
2090 /*
2091 * If this is an untrusted config, access the pool in read-only mode.
2092 * This prevents things like resilvering recently removed devices.
2093 */
2094 if (!mosconfig)
2095 spa->spa_mode = FREAD;
2096
2097 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2098
2099 spa->spa_load_state = state;
2100
2101 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2102 return (SET_ERROR(EINVAL));
2103
2104 parse = (type == SPA_IMPORT_EXISTING ?
2105 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2106
2107 /*
2108 * Create "The Godfather" zio to hold all async IOs
2109 */
2110 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2111 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2112
2113 /*
2114 * Parse the configuration into a vdev tree. We explicitly set the
2115 * value that will be returned by spa_version() since parsing the
2116 * configuration requires knowing the version number.
2117 */
2118 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2119 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2120 spa_config_exit(spa, SCL_ALL, FTAG);
2121
2122 if (error != 0)
2123 return (error);
2124
2125 ASSERT(spa->spa_root_vdev == rvd);
2126
2127 if (type != SPA_IMPORT_ASSEMBLE) {
2128 ASSERT(spa_guid(spa) == pool_guid);
2129 }
2130
2131 /*
2132 * Try to open all vdevs, loading each label in the process.
2133 */
2134 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2135 error = vdev_open(rvd);
2136 spa_config_exit(spa, SCL_ALL, FTAG);
2137 if (error != 0)
2138 return (error);
2139
2140 /*
2141 * We need to validate the vdev labels against the configuration that
2142 * we have in hand, which is dependent on the setting of mosconfig. If
2143 * mosconfig is true then we're validating the vdev labels based on
2144 * that config. Otherwise, we're validating against the cached config
2145 * (zpool.cache) that was read when we loaded the zfs module, and then
2146 * later we will recursively call spa_load() and validate against
2147 * the vdev config.
2148 *
2149 * If we're assembling a new pool that's been split off from an
2150 * existing pool, the labels haven't yet been updated so we skip
2151 * validation for now.
2152 */
2153 if (type != SPA_IMPORT_ASSEMBLE) {
2154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155 error = vdev_validate(rvd, mosconfig);
2156 spa_config_exit(spa, SCL_ALL, FTAG);
2157
2158 if (error != 0)
2159 return (error);
2160
2161 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162 return (SET_ERROR(ENXIO));
2163 }
2164
2165 /*
2166 * Find the best uberblock.
2167 */
2168 vdev_uberblock_load(rvd, ub, &label);
2169
2170 /*
2171 * If we weren't able to find a single valid uberblock, return failure.
2172 */
2173 if (ub->ub_txg == 0) {
2174 nvlist_free(label);
2175 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2176 }
2177
2178 /*
2179 * If the pool has an unsupported version we can't open it.
2180 */
2181 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2182 nvlist_free(label);
2183 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2184 }
2185
2186 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2187 nvlist_t *features;
2188
2189 /*
2190 * If we weren't able to find what's necessary for reading the
2191 * MOS in the label, return failure.
2192 */
2193 if (label == NULL || nvlist_lookup_nvlist(label,
2194 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2195 nvlist_free(label);
2196 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2197 ENXIO));
2198 }
2199
2200 /*
2201 * Update our in-core representation with the definitive values
2202 * from the label.
2203 */
2204 nvlist_free(spa->spa_label_features);
2205 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2206 }
2207
2208 nvlist_free(label);
2209
2210 /*
2211 * Look through entries in the label nvlist's features_for_read. If
2212 * there is a feature listed there which we don't understand then we
2213 * cannot open a pool.
2214 */
2215 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2216 nvlist_t *unsup_feat;
2217
2218 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2219 0);
2220
2221 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2222 NULL); nvp != NULL;
2223 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2224 if (!zfeature_is_supported(nvpair_name(nvp))) {
2225 VERIFY(nvlist_add_string(unsup_feat,
2226 nvpair_name(nvp), "") == 0);
2227 }
2228 }
2229
2230 if (!nvlist_empty(unsup_feat)) {
2231 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2232 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2233 nvlist_free(unsup_feat);
2234 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2235 ENOTSUP));
2236 }
2237
2238 nvlist_free(unsup_feat);
2239 }
2240
2241 /*
2242 * If the vdev guid sum doesn't match the uberblock, we have an
2243 * incomplete configuration. We first check to see if the pool
2244 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2245 * If it is, defer the vdev_guid_sum check till later so we
2246 * can handle missing vdevs.
2247 */
2248 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2249 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2250 rvd->vdev_guid_sum != ub->ub_guid_sum)
2251 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2252
2253 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2254 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2255 spa_try_repair(spa, config);
2256 spa_config_exit(spa, SCL_ALL, FTAG);
2257 nvlist_free(spa->spa_config_splitting);
2258 spa->spa_config_splitting = NULL;
2259 }
2260
2261 /*
2262 * Initialize internal SPA structures.
2263 */
2264 spa->spa_state = POOL_STATE_ACTIVE;
2265 spa->spa_ubsync = spa->spa_uberblock;
2266 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2267 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2268 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2269 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2270 spa->spa_claim_max_txg = spa->spa_first_txg;
2271 spa->spa_prev_software_version = ub->ub_software_version;
2272
2273 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2274 if (error)
2275 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2277
2278 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2279 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280
2281 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2282 boolean_t missing_feat_read = B_FALSE;
2283 nvlist_t *unsup_feat, *enabled_feat;
2284
2285 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2286 &spa->spa_feat_for_read_obj) != 0) {
2287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2288 }
2289
2290 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2291 &spa->spa_feat_for_write_obj) != 0) {
2292 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2293 }
2294
2295 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2296 &spa->spa_feat_desc_obj) != 0) {
2297 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2298 }
2299
2300 enabled_feat = fnvlist_alloc();
2301 unsup_feat = fnvlist_alloc();
2302
2303 if (!spa_features_check(spa, B_FALSE,
2304 unsup_feat, enabled_feat))
2305 missing_feat_read = B_TRUE;
2306
2307 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2308 if (!spa_features_check(spa, B_TRUE,
2309 unsup_feat, enabled_feat)) {
2310 missing_feat_write = B_TRUE;
2311 }
2312 }
2313
2314 fnvlist_add_nvlist(spa->spa_load_info,
2315 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2316
2317 if (!nvlist_empty(unsup_feat)) {
2318 fnvlist_add_nvlist(spa->spa_load_info,
2319 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2320 }
2321
2322 fnvlist_free(enabled_feat);
2323 fnvlist_free(unsup_feat);
2324
2325 if (!missing_feat_read) {
2326 fnvlist_add_boolean(spa->spa_load_info,
2327 ZPOOL_CONFIG_CAN_RDONLY);
2328 }
2329
2330 /*
2331 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2332 * twofold: to determine whether the pool is available for
2333 * import in read-write mode and (if it is not) whether the
2334 * pool is available for import in read-only mode. If the pool
2335 * is available for import in read-write mode, it is displayed
2336 * as available in userland; if it is not available for import
2337 * in read-only mode, it is displayed as unavailable in
2338 * userland. If the pool is available for import in read-only
2339 * mode but not read-write mode, it is displayed as unavailable
2340 * in userland with a special note that the pool is actually
2341 * available for open in read-only mode.
2342 *
2343 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2344 * missing a feature for write, we must first determine whether
2345 * the pool can be opened read-only before returning to
2346 * userland in order to know whether to display the
2347 * abovementioned note.
2348 */
2349 if (missing_feat_read || (missing_feat_write &&
2350 spa_writeable(spa))) {
2351 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2352 ENOTSUP));
2353 }
2354 }
2355
2356 spa->spa_is_initializing = B_TRUE;
2357 error = dsl_pool_open(spa->spa_dsl_pool);
2358 spa->spa_is_initializing = B_FALSE;
2359 if (error != 0)
2360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2361
2362 if (!mosconfig) {
2363 uint64_t hostid;
2364 nvlist_t *policy = NULL, *nvconfig;
2365
2366 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2368
2369 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2370 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2371 char *hostname;
2372 unsigned long myhostid = 0;
2373
2374 VERIFY(nvlist_lookup_string(nvconfig,
2375 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2376
2377 #ifdef _KERNEL
2378 myhostid = zone_get_hostid(NULL);
2379 #else /* _KERNEL */
2380 /*
2381 * We're emulating the system's hostid in userland, so
2382 * we can't use zone_get_hostid().
2383 */
2384 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2385 #endif /* _KERNEL */
2386 if (hostid != 0 && myhostid != 0 &&
2387 hostid != myhostid) {
2388 nvlist_free(nvconfig);
2389 cmn_err(CE_WARN, "pool '%s' could not be "
2390 "loaded as it was last accessed by "
2391 "another system (host: %s hostid: 0x%lx). "
2392 "See: http://illumos.org/msg/ZFS-8000-EY",
2393 spa_name(spa), hostname,
2394 (unsigned long)hostid);
2395 return (SET_ERROR(EBADF));
2396 }
2397 }
2398 if (nvlist_lookup_nvlist(spa->spa_config,
2399 ZPOOL_REWIND_POLICY, &policy) == 0)
2400 VERIFY(nvlist_add_nvlist(nvconfig,
2401 ZPOOL_REWIND_POLICY, policy) == 0);
2402
2403 spa_config_set(spa, nvconfig);
2404 spa_unload(spa);
2405 spa_deactivate(spa);
2406 spa_activate(spa, orig_mode);
2407
2408 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2409 }
2410
2411 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2413 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2414 if (error != 0)
2415 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2416
2417 /*
2418 * Load the bit that tells us to use the new accounting function
2419 * (raid-z deflation). If we have an older pool, this will not
2420 * be present.
2421 */
2422 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2423 if (error != 0 && error != ENOENT)
2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2425
2426 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2427 &spa->spa_creation_version);
2428 if (error != 0 && error != ENOENT)
2429 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2430
2431 /*
2432 * Load the persistent error log. If we have an older pool, this will
2433 * not be present.
2434 */
2435 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2436 if (error != 0 && error != ENOENT)
2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2438
2439 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2440 &spa->spa_errlog_scrub);
2441 if (error != 0 && error != ENOENT)
2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443
2444 /*
2445 * Load the history object. If we have an older pool, this
2446 * will not be present.
2447 */
2448 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2449 if (error != 0 && error != ENOENT)
2450 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2451
2452 /*
2453 * If we're assembling the pool from the split-off vdevs of
2454 * an existing pool, we don't want to attach the spares & cache
2455 * devices.
2456 */
2457
2458 /*
2459 * Load any hot spares for this pool.
2460 */
2461 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2462 if (error != 0 && error != ENOENT)
2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2464 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2465 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2466 if (load_nvlist(spa, spa->spa_spares.sav_object,
2467 &spa->spa_spares.sav_config) != 0)
2468 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2469
2470 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2471 spa_load_spares(spa);
2472 spa_config_exit(spa, SCL_ALL, FTAG);
2473 } else if (error == 0) {
2474 spa->spa_spares.sav_sync = B_TRUE;
2475 }
2476
2477 /*
2478 * Load any level 2 ARC devices for this pool.
2479 */
2480 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2481 &spa->spa_l2cache.sav_object);
2482 if (error != 0 && error != ENOENT)
2483 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2484 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2485 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2486 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2487 &spa->spa_l2cache.sav_config) != 0)
2488 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2489
2490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2491 spa_load_l2cache(spa);
2492 spa_config_exit(spa, SCL_ALL, FTAG);
2493 } else if (error == 0) {
2494 spa->spa_l2cache.sav_sync = B_TRUE;
2495 }
2496
2497 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2498
2499 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2500 if (error && error != ENOENT)
2501 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2502
2503 if (error == 0) {
2504 uint64_t autoreplace;
2505
2506 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2507 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2508 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2509 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2510 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2511 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2512 &spa->spa_dedup_ditto);
2513
2514 spa->spa_autoreplace = (autoreplace != 0);
2515 }
2516
2517 /*
2518 * If the 'autoreplace' property is set, then post a resource notifying
2519 * the ZFS DE that it should not issue any faults for unopenable
2520 * devices. We also iterate over the vdevs, and post a sysevent for any
2521 * unopenable vdevs so that the normal autoreplace handler can take
2522 * over.
2523 */
2524 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2525 spa_check_removed(spa->spa_root_vdev);
2526 /*
2527 * For the import case, this is done in spa_import(), because
2528 * at this point we're using the spare definitions from
2529 * the MOS config, not necessarily from the userland config.
2530 */
2531 if (state != SPA_LOAD_IMPORT) {
2532 spa_aux_check_removed(&spa->spa_spares);
2533 spa_aux_check_removed(&spa->spa_l2cache);
2534 }
2535 }
2536
2537 /*
2538 * Load the vdev state for all toplevel vdevs.
2539 */
2540 vdev_load(rvd);
2541
2542 /*
2543 * Propagate the leaf DTLs we just loaded all the way up the tree.
2544 */
2545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2546 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2547 spa_config_exit(spa, SCL_ALL, FTAG);
2548
2549 /*
2550 * Load the DDTs (dedup tables).
2551 */
2552 error = ddt_load(spa);
2553 if (error != 0)
2554 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555
2556 spa_update_dspace(spa);
2557
2558 /*
2559 * Validate the config, using the MOS config to fill in any
2560 * information which might be missing. If we fail to validate
2561 * the config then declare the pool unfit for use. If we're
2562 * assembling a pool from a split, the log is not transferred
2563 * over.
2564 */
2565 if (type != SPA_IMPORT_ASSEMBLE) {
2566 nvlist_t *nvconfig;
2567
2568 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2570
2571 if (!spa_config_valid(spa, nvconfig)) {
2572 nvlist_free(nvconfig);
2573 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2574 ENXIO));
2575 }
2576 nvlist_free(nvconfig);
2577
2578 /*
2579 * Now that we've validated the config, check the state of the
2580 * root vdev. If it can't be opened, it indicates one or
2581 * more toplevel vdevs are faulted.
2582 */
2583 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2584 return (SET_ERROR(ENXIO));
2585
2586 if (spa_check_logs(spa)) {
2587 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2588 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2589 }
2590 }
2591
2592 if (missing_feat_write) {
2593 ASSERT(state == SPA_LOAD_TRYIMPORT);
2594
2595 /*
2596 * At this point, we know that we can open the pool in
2597 * read-only mode but not read-write mode. We now have enough
2598 * information and can return to userland.
2599 */
2600 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2601 }
2602
2603 /*
2604 * We've successfully opened the pool, verify that we're ready
2605 * to start pushing transactions.
2606 */
2607 if (state != SPA_LOAD_TRYIMPORT) {
2608 if (error = spa_load_verify(spa))
2609 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2610 error));
2611 }
2612
2613 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2614 spa->spa_load_max_txg == UINT64_MAX)) {
2615 dmu_tx_t *tx;
2616 int need_update = B_FALSE;
2617
2618 ASSERT(state != SPA_LOAD_TRYIMPORT);
2619
2620 /*
2621 * Claim log blocks that haven't been committed yet.
2622 * This must all happen in a single txg.
2623 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2624 * invoked from zil_claim_log_block()'s i/o done callback.
2625 * Price of rollback is that we abandon the log.
2626 */
2627 spa->spa_claiming = B_TRUE;
2628
2629 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2630 spa_first_txg(spa));
2631 (void) dmu_objset_find(spa_name(spa),
2632 zil_claim, tx, DS_FIND_CHILDREN);
2633 dmu_tx_commit(tx);
2634
2635 spa->spa_claiming = B_FALSE;
2636
2637 spa_set_log_state(spa, SPA_LOG_GOOD);
2638 spa->spa_sync_on = B_TRUE;
2639 txg_sync_start(spa->spa_dsl_pool);
2640
2641 /*
2642 * Wait for all claims to sync. We sync up to the highest
2643 * claimed log block birth time so that claimed log blocks
2644 * don't appear to be from the future. spa_claim_max_txg
2645 * will have been set for us by either zil_check_log_chain()
2646 * (invoked from spa_check_logs()) or zil_claim() above.
2647 */
2648 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2649
2650 /*
2651 * If the config cache is stale, or we have uninitialized
2652 * metaslabs (see spa_vdev_add()), then update the config.
2653 *
2654 * If this is a verbatim import, trust the current
2655 * in-core spa_config and update the disk labels.
2656 */
2657 if (config_cache_txg != spa->spa_config_txg ||
2658 state == SPA_LOAD_IMPORT ||
2659 state == SPA_LOAD_RECOVER ||
2660 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2661 need_update = B_TRUE;
2662
2663 for (int c = 0; c < rvd->vdev_children; c++)
2664 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2665 need_update = B_TRUE;
2666
2667 /*
2668 * Update the config cache asychronously in case we're the
2669 * root pool, in which case the config cache isn't writable yet.
2670 */
2671 if (need_update)
2672 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2673
2674 /*
2675 * Check all DTLs to see if anything needs resilvering.
2676 */
2677 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2678 vdev_resilver_needed(rvd, NULL, NULL))
2679 spa_async_request(spa, SPA_ASYNC_RESILVER);
2680
2681 /*
2682 * Log the fact that we booted up (so that we can detect if
2683 * we rebooted in the middle of an operation).
2684 */
2685 spa_history_log_version(spa, "open");
2686
2687 /*
2688 * Delete any inconsistent datasets.
2689 */
2690 (void) dmu_objset_find(spa_name(spa),
2691 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2692
2693 /*
2694 * Clean up any stale temporary dataset userrefs.
2695 */
2696 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2697 }
2698
2699 return (0);
2700 }
2701
2702 static int
2703 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2704 {
2705 int mode = spa->spa_mode;
2706
2707 spa_unload(spa);
2708 spa_deactivate(spa);
2709
2710 spa->spa_load_max_txg--;
2711
2712 spa_activate(spa, mode);
2713 spa_async_suspend(spa);
2714
2715 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2716 }
2717
2718 /*
2719 * If spa_load() fails this function will try loading prior txg's. If
2720 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2721 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2722 * function will not rewind the pool and will return the same error as
2723 * spa_load().
2724 */
2725 static int
2726 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2727 uint64_t max_request, int rewind_flags)
2728 {
2729 nvlist_t *loadinfo = NULL;
2730 nvlist_t *config = NULL;
2731 int load_error, rewind_error;
2732 uint64_t safe_rewind_txg;
2733 uint64_t min_txg;
2734
2735 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2736 spa->spa_load_max_txg = spa->spa_load_txg;
2737 spa_set_log_state(spa, SPA_LOG_CLEAR);
2738 } else {
2739 spa->spa_load_max_txg = max_request;
2740 }
2741
2742 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2743 mosconfig);
2744 if (load_error == 0)
2745 return (0);
2746
2747 if (spa->spa_root_vdev != NULL)
2748 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2749
2750 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2751 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2752
2753 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2754 nvlist_free(config);
2755 return (load_error);
2756 }
2757
2758 if (state == SPA_LOAD_RECOVER) {
2759 /* Price of rolling back is discarding txgs, including log */
2760 spa_set_log_state(spa, SPA_LOG_CLEAR);
2761 } else {
2762 /*
2763 * If we aren't rolling back save the load info from our first
2764 * import attempt so that we can restore it after attempting
2765 * to rewind.
2766 */
2767 loadinfo = spa->spa_load_info;
2768 spa->spa_load_info = fnvlist_alloc();
2769 }
2770
2771 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2772 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2773 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2774 TXG_INITIAL : safe_rewind_txg;
2775
2776 /*
2777 * Continue as long as we're finding errors, we're still within
2778 * the acceptable rewind range, and we're still finding uberblocks
2779 */
2780 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2781 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2782 if (spa->spa_load_max_txg < safe_rewind_txg)
2783 spa->spa_extreme_rewind = B_TRUE;
2784 rewind_error = spa_load_retry(spa, state, mosconfig);
2785 }
2786
2787 spa->spa_extreme_rewind = B_FALSE;
2788 spa->spa_load_max_txg = UINT64_MAX;
2789
2790 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2791 spa_config_set(spa, config);
2792
2793 if (state == SPA_LOAD_RECOVER) {
2794 ASSERT3P(loadinfo, ==, NULL);
2795 return (rewind_error);
2796 } else {
2797 /* Store the rewind info as part of the initial load info */
2798 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2799 spa->spa_load_info);
2800
2801 /* Restore the initial load info */
2802 fnvlist_free(spa->spa_load_info);
2803 spa->spa_load_info = loadinfo;
2804
2805 return (load_error);
2806 }
2807 }
2808
2809 /*
2810 * Pool Open/Import
2811 *
2812 * The import case is identical to an open except that the configuration is sent
2813 * down from userland, instead of grabbed from the configuration cache. For the
2814 * case of an open, the pool configuration will exist in the
2815 * POOL_STATE_UNINITIALIZED state.
2816 *
2817 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2818 * the same time open the pool, without having to keep around the spa_t in some
2819 * ambiguous state.
2820 */
2821 static int
2822 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2823 nvlist_t **config)
2824 {
2825 spa_t *spa;
2826 spa_load_state_t state = SPA_LOAD_OPEN;
2827 int error;
2828 int locked = B_FALSE;
2829
2830 *spapp = NULL;
2831
2832 /*
2833 * As disgusting as this is, we need to support recursive calls to this
2834 * function because dsl_dir_open() is called during spa_load(), and ends
2835 * up calling spa_open() again. The real fix is to figure out how to
2836 * avoid dsl_dir_open() calling this in the first place.
2837 */
2838 if (mutex_owner(&spa_namespace_lock) != curthread) {
2839 mutex_enter(&spa_namespace_lock);
2840 locked = B_TRUE;
2841 }
2842
2843 if ((spa = spa_lookup(pool)) == NULL) {
2844 if (locked)
2845 mutex_exit(&spa_namespace_lock);
2846 return (SET_ERROR(ENOENT));
2847 }
2848
2849 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2850 zpool_rewind_policy_t policy;
2851
2852 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2853 &policy);
2854 if (policy.zrp_request & ZPOOL_DO_REWIND)
2855 state = SPA_LOAD_RECOVER;
2856
2857 spa_activate(spa, spa_mode_global);
2858
2859 if (state != SPA_LOAD_RECOVER)
2860 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2861
2862 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2863 policy.zrp_request);
2864
2865 if (error == EBADF) {
2866 /*
2867 * If vdev_validate() returns failure (indicated by
2868 * EBADF), it indicates that one of the vdevs indicates
2869 * that the pool has been exported or destroyed. If
2870 * this is the case, the config cache is out of sync and
2871 * we should remove the pool from the namespace.
2872 */
2873 spa_unload(spa);
2874 spa_deactivate(spa);
2875 spa_config_sync(spa, B_TRUE, B_TRUE);
2876 spa_remove(spa);
2877 if (locked)
2878 mutex_exit(&spa_namespace_lock);
2879 return (SET_ERROR(ENOENT));
2880 }
2881
2882 if (error) {
2883 /*
2884 * We can't open the pool, but we still have useful
2885 * information: the state of each vdev after the
2886 * attempted vdev_open(). Return this to the user.
2887 */
2888 if (config != NULL && spa->spa_config) {
2889 VERIFY(nvlist_dup(spa->spa_config, config,
2890 KM_SLEEP) == 0);
2891 VERIFY(nvlist_add_nvlist(*config,
2892 ZPOOL_CONFIG_LOAD_INFO,
2893 spa->spa_load_info) == 0);
2894 }
2895 spa_unload(spa);
2896 spa_deactivate(spa);
2897 spa->spa_last_open_failed = error;
2898 if (locked)
2899 mutex_exit(&spa_namespace_lock);
2900 *spapp = NULL;
2901 return (error);
2902 }
2903 }
2904
2905 spa_open_ref(spa, tag);
2906
2907 if (config != NULL)
2908 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2909
2910 /*
2911 * If we've recovered the pool, pass back any information we
2912 * gathered while doing the load.
2913 */
2914 if (state == SPA_LOAD_RECOVER) {
2915 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2916 spa->spa_load_info) == 0);
2917 }
2918
2919 if (locked) {
2920 spa->spa_last_open_failed = 0;
2921 spa->spa_last_ubsync_txg = 0;
2922 spa->spa_load_txg = 0;
2923 mutex_exit(&spa_namespace_lock);
2924 }
2925
2926 *spapp = spa;
2927
2928 return (0);
2929 }
2930
2931 int
2932 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2933 nvlist_t **config)
2934 {
2935 return (spa_open_common(name, spapp, tag, policy, config));
2936 }
2937
2938 int
2939 spa_open(const char *name, spa_t **spapp, void *tag)
2940 {
2941 return (spa_open_common(name, spapp, tag, NULL, NULL));
2942 }
2943
2944 /*
2945 * Lookup the given spa_t, incrementing the inject count in the process,
2946 * preventing it from being exported or destroyed.
2947 */
2948 spa_t *
2949 spa_inject_addref(char *name)
2950 {
2951 spa_t *spa;
2952
2953 mutex_enter(&spa_namespace_lock);
2954 if ((spa = spa_lookup(name)) == NULL) {
2955 mutex_exit(&spa_namespace_lock);
2956 return (NULL);
2957 }
2958 spa->spa_inject_ref++;
2959 mutex_exit(&spa_namespace_lock);
2960
2961 return (spa);
2962 }
2963
2964 void
2965 spa_inject_delref(spa_t *spa)
2966 {
2967 mutex_enter(&spa_namespace_lock);
2968 spa->spa_inject_ref--;
2969 mutex_exit(&spa_namespace_lock);
2970 }
2971
2972 /*
2973 * Add spares device information to the nvlist.
2974 */
2975 static void
2976 spa_add_spares(spa_t *spa, nvlist_t *config)
2977 {
2978 nvlist_t **spares;
2979 uint_t i, nspares;
2980 nvlist_t *nvroot;
2981 uint64_t guid;
2982 vdev_stat_t *vs;
2983 uint_t vsc;
2984 uint64_t pool;
2985
2986 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2987
2988 if (spa->spa_spares.sav_count == 0)
2989 return;
2990
2991 VERIFY(nvlist_lookup_nvlist(config,
2992 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2993 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2994 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2995 if (nspares != 0) {
2996 VERIFY(nvlist_add_nvlist_array(nvroot,
2997 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2998 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2999 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3000
3001 /*
3002 * Go through and find any spares which have since been
3003 * repurposed as an active spare. If this is the case, update
3004 * their status appropriately.
3005 */
3006 for (i = 0; i < nspares; i++) {
3007 VERIFY(nvlist_lookup_uint64(spares[i],
3008 ZPOOL_CONFIG_GUID, &guid) == 0);
3009 if (spa_spare_exists(guid, &pool, NULL) &&
3010 pool != 0ULL) {
3011 VERIFY(nvlist_lookup_uint64_array(
3012 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3013 (uint64_t **)&vs, &vsc) == 0);
3014 vs->vs_state = VDEV_STATE_CANT_OPEN;
3015 vs->vs_aux = VDEV_AUX_SPARED;
3016 }
3017 }
3018 }
3019 }
3020
3021 /*
3022 * Add l2cache device information to the nvlist, including vdev stats.
3023 */
3024 static void
3025 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3026 {
3027 nvlist_t **l2cache;
3028 uint_t i, j, nl2cache;
3029 nvlist_t *nvroot;
3030 uint64_t guid;
3031 vdev_t *vd;
3032 vdev_stat_t *vs;
3033 uint_t vsc;
3034
3035 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3036
3037 if (spa->spa_l2cache.sav_count == 0)
3038 return;
3039
3040 VERIFY(nvlist_lookup_nvlist(config,
3041 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3042 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3043 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3044 if (nl2cache != 0) {
3045 VERIFY(nvlist_add_nvlist_array(nvroot,
3046 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3047 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3048 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3049
3050 /*
3051 * Update level 2 cache device stats.
3052 */
3053
3054 for (i = 0; i < nl2cache; i++) {
3055 VERIFY(nvlist_lookup_uint64(l2cache[i],
3056 ZPOOL_CONFIG_GUID, &guid) == 0);
3057
3058 vd = NULL;
3059 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3060 if (guid ==
3061 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3062 vd = spa->spa_l2cache.sav_vdevs[j];
3063 break;
3064 }
3065 }
3066 ASSERT(vd != NULL);
3067
3068 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3069 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3070 == 0);
3071 vdev_get_stats(vd, vs);
3072 }
3073 }
3074 }
3075
3076 static void
3077 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3078 {
3079 nvlist_t *features;
3080 zap_cursor_t zc;
3081 zap_attribute_t za;
3082
3083 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3084 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3085
3086 if (spa->spa_feat_for_read_obj != 0) {
3087 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3088 spa->spa_feat_for_read_obj);
3089 zap_cursor_retrieve(&zc, &za) == 0;
3090 zap_cursor_advance(&zc)) {
3091 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3092 za.za_num_integers == 1);
3093 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3094 za.za_first_integer));
3095 }
3096 zap_cursor_fini(&zc);
3097 }
3098
3099 if (spa->spa_feat_for_write_obj != 0) {
3100 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3101 spa->spa_feat_for_write_obj);
3102 zap_cursor_retrieve(&zc, &za) == 0;
3103 zap_cursor_advance(&zc)) {
3104 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3105 za.za_num_integers == 1);
3106 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3107 za.za_first_integer));
3108 }
3109 zap_cursor_fini(&zc);
3110 }
3111
3112 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3113 features) == 0);
3114 nvlist_free(features);
3115 }
3116
3117 int
3118 spa_get_stats(const char *name, nvlist_t **config,
3119 char *altroot, size_t buflen)
3120 {
3121 int error;
3122 spa_t *spa;
3123
3124 *config = NULL;
3125 error = spa_open_common(name, &spa, FTAG, NULL, config);
3126
3127 if (spa != NULL) {
3128 /*
3129 * This still leaves a window of inconsistency where the spares
3130 * or l2cache devices could change and the config would be
3131 * self-inconsistent.
3132 */
3133 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3134
3135 if (*config != NULL) {
3136 uint64_t loadtimes[2];
3137
3138 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3139 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3140 VERIFY(nvlist_add_uint64_array(*config,
3141 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3142
3143 VERIFY(nvlist_add_uint64(*config,
3144 ZPOOL_CONFIG_ERRCOUNT,
3145 spa_get_errlog_size(spa)) == 0);
3146
3147 if (spa_suspended(spa))
3148 VERIFY(nvlist_add_uint64(*config,
3149 ZPOOL_CONFIG_SUSPENDED,
3150 spa->spa_failmode) == 0);
3151
3152 spa_add_spares(spa, *config);
3153 spa_add_l2cache(spa, *config);
3154 spa_add_feature_stats(spa, *config);
3155 }
3156 }
3157
3158 /*
3159 * We want to get the alternate root even for faulted pools, so we cheat
3160 * and call spa_lookup() directly.
3161 */
3162 if (altroot) {
3163 if (spa == NULL) {
3164 mutex_enter(&spa_namespace_lock);
3165 spa = spa_lookup(name);
3166 if (spa)
3167 spa_altroot(spa, altroot, buflen);
3168 else
3169 altroot[0] = '\0';
3170 spa = NULL;
3171 mutex_exit(&spa_namespace_lock);
3172 } else {
3173 spa_altroot(spa, altroot, buflen);
3174 }
3175 }
3176
3177 if (spa != NULL) {
3178 spa_config_exit(spa, SCL_CONFIG, FTAG);
3179 spa_close(spa, FTAG);
3180 }
3181
3182 return (error);
3183 }
3184
3185 /*
3186 * Validate that the auxiliary device array is well formed. We must have an
3187 * array of nvlists, each which describes a valid leaf vdev. If this is an
3188 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3189 * specified, as long as they are well-formed.
3190 */
3191 static int
3192 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3193 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3194 vdev_labeltype_t label)
3195 {
3196 nvlist_t **dev;
3197 uint_t i, ndev;
3198 vdev_t *vd;
3199 int error;
3200
3201 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3202
3203 /*
3204 * It's acceptable to have no devs specified.
3205 */
3206 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3207 return (0);
3208
3209 if (ndev == 0)
3210 return (SET_ERROR(EINVAL));
3211
3212 /*
3213 * Make sure the pool is formatted with a version that supports this
3214 * device type.
3215 */
3216 if (spa_version(spa) < version)
3217 return (SET_ERROR(ENOTSUP));
3218
3219 /*
3220 * Set the pending device list so we correctly handle device in-use
3221 * checking.
3222 */
3223 sav->sav_pending = dev;
3224 sav->sav_npending = ndev;
3225
3226 for (i = 0; i < ndev; i++) {
3227 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3228 mode)) != 0)
3229 goto out;
3230
3231 if (!vd->vdev_ops->vdev_op_leaf) {
3232 vdev_free(vd);
3233 error = SET_ERROR(EINVAL);
3234 goto out;
3235 }
3236
3237 /*
3238 * The L2ARC currently only supports disk devices in
3239 * kernel context. For user-level testing, we allow it.
3240 */
3241 #ifdef _KERNEL
3242 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3243 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3244 error = SET_ERROR(ENOTBLK);
3245 vdev_free(vd);
3246 goto out;
3247 }
3248 #endif
3249 vd->vdev_top = vd;
3250
3251 if ((error = vdev_open(vd)) == 0 &&
3252 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3253 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3254 vd->vdev_guid) == 0);
3255 }
3256
3257 vdev_free(vd);
3258
3259 if (error &&
3260 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3261 goto out;
3262 else
3263 error = 0;
3264 }
3265
3266 out:
3267 sav->sav_pending = NULL;
3268 sav->sav_npending = 0;
3269 return (error);
3270 }
3271
3272 static int
3273 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3274 {
3275 int error;
3276
3277 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3278
3279 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3280 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3281 VDEV_LABEL_SPARE)) != 0) {
3282 return (error);
3283 }
3284
3285 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3286 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3287 VDEV_LABEL_L2CACHE));
3288 }
3289
3290 static void
3291 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3292 const char *config)
3293 {
3294 int i;
3295
3296 if (sav->sav_config != NULL) {
3297 nvlist_t **olddevs;
3298 uint_t oldndevs;
3299 nvlist_t **newdevs;
3300
3301 /*
3302 * Generate new dev list by concatentating with the
3303 * current dev list.
3304 */
3305 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3306 &olddevs, &oldndevs) == 0);
3307
3308 newdevs = kmem_alloc(sizeof (void *) *
3309 (ndevs + oldndevs), KM_SLEEP);
3310 for (i = 0; i < oldndevs; i++)
3311 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3312 KM_SLEEP) == 0);
3313 for (i = 0; i < ndevs; i++)
3314 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3315 KM_SLEEP) == 0);
3316
3317 VERIFY(nvlist_remove(sav->sav_config, config,
3318 DATA_TYPE_NVLIST_ARRAY) == 0);
3319
3320 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3321 config, newdevs, ndevs + oldndevs) == 0);
3322 for (i = 0; i < oldndevs + ndevs; i++)
3323 nvlist_free(newdevs[i]);
3324 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3325 } else {
3326 /*
3327 * Generate a new dev list.
3328 */
3329 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3330 KM_SLEEP) == 0);
3331 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3332 devs, ndevs) == 0);
3333 }
3334 }
3335
3336 /*
3337 * Stop and drop level 2 ARC devices
3338 */
3339 void
3340 spa_l2cache_drop(spa_t *spa)
3341 {
3342 vdev_t *vd;
3343 int i;
3344 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3345
3346 for (i = 0; i < sav->sav_count; i++) {
3347 uint64_t pool;
3348
3349 vd = sav->sav_vdevs[i];
3350 ASSERT(vd != NULL);
3351
3352 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3353 pool != 0ULL && l2arc_vdev_present(vd))
3354 l2arc_remove_vdev(vd);
3355 }
3356 }
3357
3358 /*
3359 * Pool Creation
3360 */
3361 int
3362 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3363 nvlist_t *zplprops)
3364 {
3365 spa_t *spa;
3366 char *altroot = NULL;
3367 vdev_t *rvd;
3368 dsl_pool_t *dp;
3369 dmu_tx_t *tx;
3370 int error = 0;
3371 uint64_t txg = TXG_INITIAL;
3372 nvlist_t **spares, **l2cache;
3373 uint_t nspares, nl2cache;
3374 uint64_t version, obj;
3375 boolean_t has_features;
3376
3377 /*
3378 * If this pool already exists, return failure.
3379 */
3380 mutex_enter(&spa_namespace_lock);
3381 if (spa_lookup(pool) != NULL) {
3382 mutex_exit(&spa_namespace_lock);
3383 return (SET_ERROR(EEXIST));
3384 }
3385
3386 /*
3387 * Allocate a new spa_t structure.
3388 */
3389 (void) nvlist_lookup_string(props,
3390 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3391 spa = spa_add(pool, NULL, altroot);
3392 spa_activate(spa, spa_mode_global);
3393
3394 if (props && (error = spa_prop_validate(spa, props))) {
3395 spa_deactivate(spa);
3396 spa_remove(spa);
3397 mutex_exit(&spa_namespace_lock);
3398 return (error);
3399 }
3400
3401 has_features = B_FALSE;
3402 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3403 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3404 if (zpool_prop_feature(nvpair_name(elem)))
3405 has_features = B_TRUE;
3406 }
3407
3408 if (has_features || nvlist_lookup_uint64(props,
3409 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3410 version = SPA_VERSION;
3411 }
3412 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3413
3414 spa->spa_first_txg = txg;
3415 spa->spa_uberblock.ub_txg = txg - 1;
3416 spa->spa_uberblock.ub_version = version;
3417 spa->spa_ubsync = spa->spa_uberblock;
3418
3419 /*
3420 * Create "The Godfather" zio to hold all async IOs
3421 */
3422 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3423 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3424
3425 /*
3426 * Create the root vdev.
3427 */
3428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3429
3430 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3431
3432 ASSERT(error != 0 || rvd != NULL);
3433 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3434
3435 if (error == 0 && !zfs_allocatable_devs(nvroot))
3436 error = SET_ERROR(EINVAL);
3437
3438 if (error == 0 &&
3439 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3440 (error = spa_validate_aux(spa, nvroot, txg,
3441 VDEV_ALLOC_ADD)) == 0) {
3442 for (int c = 0; c < rvd->vdev_children; c++) {
3443 vdev_metaslab_set_size(rvd->vdev_child[c]);
3444 vdev_expand(rvd->vdev_child[c], txg);
3445 }
3446 }
3447
3448 spa_config_exit(spa, SCL_ALL, FTAG);
3449
3450 if (error != 0) {
3451 spa_unload(spa);
3452 spa_deactivate(spa);
3453 spa_remove(spa);
3454 mutex_exit(&spa_namespace_lock);
3455 return (error);
3456 }
3457
3458 /*
3459 * Get the list of spares, if specified.
3460 */
3461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3462 &spares, &nspares) == 0) {
3463 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3464 KM_SLEEP) == 0);
3465 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3466 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3468 spa_load_spares(spa);
3469 spa_config_exit(spa, SCL_ALL, FTAG);
3470 spa->spa_spares.sav_sync = B_TRUE;
3471 }
3472
3473 /*
3474 * Get the list of level 2 cache devices, if specified.
3475 */
3476 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3477 &l2cache, &nl2cache) == 0) {
3478 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3479 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3480 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3481 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3482 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3483 spa_load_l2cache(spa);
3484 spa_config_exit(spa, SCL_ALL, FTAG);
3485 spa->spa_l2cache.sav_sync = B_TRUE;
3486 }
3487
3488 spa->spa_is_initializing = B_TRUE;
3489 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3490 spa->spa_meta_objset = dp->dp_meta_objset;
3491 spa->spa_is_initializing = B_FALSE;
3492
3493 /*
3494 * Create DDTs (dedup tables).
3495 */
3496 ddt_create(spa);
3497
3498 spa_update_dspace(spa);
3499
3500 tx = dmu_tx_create_assigned(dp, txg);
3501
3502 /*
3503 * Create the pool config object.
3504 */
3505 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3506 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3507 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3508
3509 if (zap_add(spa->spa_meta_objset,
3510 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3511 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3512 cmn_err(CE_PANIC, "failed to add pool config");
3513 }
3514
3515 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3516 spa_feature_create_zap_objects(spa, tx);
3517
3518 if (zap_add(spa->spa_meta_objset,
3519 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3520 sizeof (uint64_t), 1, &version, tx) != 0) {
3521 cmn_err(CE_PANIC, "failed to add pool version");
3522 }
3523
3524 /* Newly created pools with the right version are always deflated. */
3525 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3526 spa->spa_deflate = TRUE;
3527 if (zap_add(spa->spa_meta_objset,
3528 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3529 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3530 cmn_err(CE_PANIC, "failed to add deflate");
3531 }
3532 }
3533
3534 /*
3535 * Create the deferred-free bpobj. Turn off compression
3536 * because sync-to-convergence takes longer if the blocksize
3537 * keeps changing.
3538 */
3539 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3540 dmu_object_set_compress(spa->spa_meta_objset, obj,
3541 ZIO_COMPRESS_OFF, tx);
3542 if (zap_add(spa->spa_meta_objset,
3543 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3544 sizeof (uint64_t), 1, &obj, tx) != 0) {
3545 cmn_err(CE_PANIC, "failed to add bpobj");
3546 }
3547 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3548 spa->spa_meta_objset, obj));
3549
3550 /*
3551 * Create the pool's history object.
3552 */
3553 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3554 spa_history_create_obj(spa, tx);
3555
3556 /*
3557 * Set pool properties.
3558 */
3559 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3560 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3561 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3562 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3563
3564 if (props != NULL) {
3565 spa_configfile_set(spa, props, B_FALSE);
3566 spa_sync_props(props, tx);
3567 }
3568
3569 dmu_tx_commit(tx);
3570
3571 spa->spa_sync_on = B_TRUE;
3572 txg_sync_start(spa->spa_dsl_pool);
3573
3574 /*
3575 * We explicitly wait for the first transaction to complete so that our
3576 * bean counters are appropriately updated.
3577 */
3578 txg_wait_synced(spa->spa_dsl_pool, txg);
3579
3580 spa_config_sync(spa, B_FALSE, B_TRUE);
3581
3582 spa_history_log_version(spa, "create");
3583
3584 spa->spa_minref = refcount_count(&spa->spa_refcount);
3585
3586 mutex_exit(&spa_namespace_lock);
3587
3588 return (0);
3589 }
3590
3591 #ifdef _KERNEL
3592 /*
3593 * Get the root pool information from the root disk, then import the root pool
3594 * during the system boot up time.
3595 */
3596 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3597
3598 static nvlist_t *
3599 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3600 {
3601 nvlist_t *config;
3602 nvlist_t *nvtop, *nvroot;
3603 uint64_t pgid;
3604
3605 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3606 return (NULL);
3607
3608 /*
3609 * Add this top-level vdev to the child array.
3610 */
3611 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3612 &nvtop) == 0);
3613 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3614 &pgid) == 0);
3615 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3616
3617 /*
3618 * Put this pool's top-level vdevs into a root vdev.
3619 */
3620 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3621 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3622 VDEV_TYPE_ROOT) == 0);
3623 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3624 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3625 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3626 &nvtop, 1) == 0);
3627
3628 /*
3629 * Replace the existing vdev_tree with the new root vdev in
3630 * this pool's configuration (remove the old, add the new).
3631 */
3632 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3633 nvlist_free(nvroot);
3634 return (config);
3635 }
3636
3637 /*
3638 * Walk the vdev tree and see if we can find a device with "better"
3639 * configuration. A configuration is "better" if the label on that
3640 * device has a more recent txg.
3641 */
3642 static void
3643 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3644 {
3645 for (int c = 0; c < vd->vdev_children; c++)
3646 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3647
3648 if (vd->vdev_ops->vdev_op_leaf) {
3649 nvlist_t *label;
3650 uint64_t label_txg;
3651
3652 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3653 &label) != 0)
3654 return;
3655
3656 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3657 &label_txg) == 0);
3658
3659 /*
3660 * Do we have a better boot device?
3661 */
3662 if (label_txg > *txg) {
3663 *txg = label_txg;
3664 *avd = vd;
3665 }
3666 nvlist_free(label);
3667 }
3668 }
3669
3670 /*
3671 * Import a root pool.
3672 *
3673 * For x86. devpath_list will consist of devid and/or physpath name of
3674 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3675 * The GRUB "findroot" command will return the vdev we should boot.
3676 *
3677 * For Sparc, devpath_list consists the physpath name of the booting device
3678 * no matter the rootpool is a single device pool or a mirrored pool.
3679 * e.g.
3680 * "/pci@1f,0/ide@d/disk@0,0:a"
3681 */
3682 int
3683 spa_import_rootpool(char *devpath, char *devid)
3684 {
3685 spa_t *spa;
3686 vdev_t *rvd, *bvd, *avd = NULL;
3687 nvlist_t *config, *nvtop;
3688 uint64_t guid, txg;
3689 char *pname;
3690 int error;
3691
3692 /*
3693 * Read the label from the boot device and generate a configuration.
3694 */
3695 config = spa_generate_rootconf(devpath, devid, &guid);
3696 #if defined(_OBP) && defined(_KERNEL)
3697 if (config == NULL) {
3698 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3699 /* iscsi boot */
3700 get_iscsi_bootpath_phy(devpath);
3701 config = spa_generate_rootconf(devpath, devid, &guid);
3702 }
3703 }
3704 #endif
3705 if (config == NULL) {
3706 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3707 devpath);
3708 return (SET_ERROR(EIO));
3709 }
3710
3711 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3712 &pname) == 0);
3713 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3714
3715 mutex_enter(&spa_namespace_lock);
3716 if ((spa = spa_lookup(pname)) != NULL) {
3717 /*
3718 * Remove the existing root pool from the namespace so that we
3719 * can replace it with the correct config we just read in.
3720 */
3721 spa_remove(spa);
3722 }
3723
3724 spa = spa_add(pname, config, NULL);
3725 spa->spa_is_root = B_TRUE;
3726 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3727
3728 /*
3729 * Build up a vdev tree based on the boot device's label config.
3730 */
3731 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3732 &nvtop) == 0);
3733 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3734 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3735 VDEV_ALLOC_ROOTPOOL);
3736 spa_config_exit(spa, SCL_ALL, FTAG);
3737 if (error) {
3738 mutex_exit(&spa_namespace_lock);
3739 nvlist_free(config);
3740 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3741 pname);
3742 return (error);
3743 }
3744
3745 /*
3746 * Get the boot vdev.
3747 */
3748 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3749 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3750 (u_longlong_t)guid);
3751 error = SET_ERROR(ENOENT);
3752 goto out;
3753 }
3754
3755 /*
3756 * Determine if there is a better boot device.
3757 */
3758 avd = bvd;
3759 spa_alt_rootvdev(rvd, &avd, &txg);
3760 if (avd != bvd) {
3761 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3762 "try booting from '%s'", avd->vdev_path);
3763 error = SET_ERROR(EINVAL);
3764 goto out;
3765 }
3766
3767 /*
3768 * If the boot device is part of a spare vdev then ensure that
3769 * we're booting off the active spare.
3770 */
3771 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3772 !bvd->vdev_isspare) {
3773 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3774 "try booting from '%s'",
3775 bvd->vdev_parent->
3776 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3777 error = SET_ERROR(EINVAL);
3778 goto out;
3779 }
3780
3781 error = 0;
3782 out:
3783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3784 vdev_free(rvd);
3785 spa_config_exit(spa, SCL_ALL, FTAG);
3786 mutex_exit(&spa_namespace_lock);
3787
3788 nvlist_free(config);
3789 return (error);
3790 }
3791
3792 #endif
3793
3794 /*
3795 * Import a non-root pool into the system.
3796 */
3797 int
3798 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3799 {
3800 spa_t *spa;
3801 char *altroot = NULL;
3802 spa_load_state_t state = SPA_LOAD_IMPORT;
3803 zpool_rewind_policy_t policy;
3804 uint64_t mode = spa_mode_global;
3805 uint64_t readonly = B_FALSE;
3806 int error;
3807 nvlist_t *nvroot;
3808 nvlist_t **spares, **l2cache;
3809 uint_t nspares, nl2cache;
3810
3811 /*
3812 * If a pool with this name exists, return failure.
3813 */
3814 mutex_enter(&spa_namespace_lock);
3815 if (spa_lookup(pool) != NULL) {
3816 mutex_exit(&spa_namespace_lock);
3817 return (SET_ERROR(EEXIST));
3818 }
3819
3820 /*
3821 * Create and initialize the spa structure.
3822 */
3823 (void) nvlist_lookup_string(props,
3824 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3825 (void) nvlist_lookup_uint64(props,
3826 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3827 if (readonly)
3828 mode = FREAD;
3829 spa = spa_add(pool, config, altroot);
3830 spa->spa_import_flags = flags;
3831
3832 /*
3833 * Verbatim import - Take a pool and insert it into the namespace
3834 * as if it had been loaded at boot.
3835 */
3836 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3837 if (props != NULL)
3838 spa_configfile_set(spa, props, B_FALSE);
3839
3840 spa_config_sync(spa, B_FALSE, B_TRUE);
3841
3842 mutex_exit(&spa_namespace_lock);
3843 return (0);
3844 }
3845
3846 spa_activate(spa, mode);
3847
3848 /*
3849 * Don't start async tasks until we know everything is healthy.
3850 */
3851 spa_async_suspend(spa);
3852
3853 zpool_get_rewind_policy(config, &policy);
3854 if (policy.zrp_request & ZPOOL_DO_REWIND)
3855 state = SPA_LOAD_RECOVER;
3856
3857 /*
3858 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3859 * because the user-supplied config is actually the one to trust when
3860 * doing an import.
3861 */
3862 if (state != SPA_LOAD_RECOVER)
3863 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3864
3865 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3866 policy.zrp_request);
3867
3868 /*
3869 * Propagate anything learned while loading the pool and pass it
3870 * back to caller (i.e. rewind info, missing devices, etc).
3871 */
3872 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3873 spa->spa_load_info) == 0);
3874
3875 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3876 /*
3877 * Toss any existing sparelist, as it doesn't have any validity
3878 * anymore, and conflicts with spa_has_spare().
3879 */
3880 if (spa->spa_spares.sav_config) {
3881 nvlist_free(spa->spa_spares.sav_config);
3882 spa->spa_spares.sav_config = NULL;
3883 spa_load_spares(spa);
3884 }
3885 if (spa->spa_l2cache.sav_config) {
3886 nvlist_free(spa->spa_l2cache.sav_config);
3887 spa->spa_l2cache.sav_config = NULL;
3888 spa_load_l2cache(spa);
3889 }
3890
3891 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3892 &nvroot) == 0);
3893 if (error == 0)
3894 error = spa_validate_aux(spa, nvroot, -1ULL,
3895 VDEV_ALLOC_SPARE);
3896 if (error == 0)
3897 error = spa_validate_aux(spa, nvroot, -1ULL,
3898 VDEV_ALLOC_L2CACHE);
3899 spa_config_exit(spa, SCL_ALL, FTAG);
3900
3901 if (props != NULL)
3902 spa_configfile_set(spa, props, B_FALSE);
3903
3904 if (error != 0 || (props && spa_writeable(spa) &&
3905 (error = spa_prop_set(spa, props)))) {
3906 spa_unload(spa);
3907 spa_deactivate(spa);
3908 spa_remove(spa);
3909 mutex_exit(&spa_namespace_lock);
3910 return (error);
3911 }
3912
3913 spa_async_resume(spa);
3914
3915 /*
3916 * Override any spares and level 2 cache devices as specified by
3917 * the user, as these may have correct device names/devids, etc.
3918 */
3919 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3920 &spares, &nspares) == 0) {
3921 if (spa->spa_spares.sav_config)
3922 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3923 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3924 else
3925 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3926 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3927 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3928 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3929 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3930 spa_load_spares(spa);
3931 spa_config_exit(spa, SCL_ALL, FTAG);
3932 spa->spa_spares.sav_sync = B_TRUE;
3933 }
3934 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3935 &l2cache, &nl2cache) == 0) {
3936 if (spa->spa_l2cache.sav_config)
3937 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3938 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3939 else
3940 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3941 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3942 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3943 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3944 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3945 spa_load_l2cache(spa);
3946 spa_config_exit(spa, SCL_ALL, FTAG);
3947 spa->spa_l2cache.sav_sync = B_TRUE;
3948 }
3949
3950 /*
3951 * Check for any removed devices.
3952 */
3953 if (spa->spa_autoreplace) {
3954 spa_aux_check_removed(&spa->spa_spares);
3955 spa_aux_check_removed(&spa->spa_l2cache);
3956 }
3957
3958 if (spa_writeable(spa)) {
3959 /*
3960 * Update the config cache to include the newly-imported pool.
3961 */
3962 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3963 }
3964
3965 /*
3966 * It's possible that the pool was expanded while it was exported.
3967 * We kick off an async task to handle this for us.
3968 */
3969 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3970
3971 mutex_exit(&spa_namespace_lock);
3972 spa_history_log_version(spa, "import");
3973
3974 return (0);
3975 }
3976
3977 nvlist_t *
3978 spa_tryimport(nvlist_t *tryconfig)
3979 {
3980 nvlist_t *config = NULL;
3981 char *poolname;
3982 spa_t *spa;
3983 uint64_t state;
3984 int error;
3985
3986 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3987 return (NULL);
3988
3989 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3990 return (NULL);
3991
3992 /*
3993 * Create and initialize the spa structure.
3994 */
3995 mutex_enter(&spa_namespace_lock);
3996 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3997 spa_activate(spa, FREAD);
3998
3999 /*
4000 * Pass off the heavy lifting to spa_load().
4001 * Pass TRUE for mosconfig because the user-supplied config
4002 * is actually the one to trust when doing an import.
4003 */
4004 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4005
4006 /*
4007 * If 'tryconfig' was at least parsable, return the current config.
4008 */
4009 if (spa->spa_root_vdev != NULL) {
4010 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4011 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4012 poolname) == 0);
4013 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4014 state) == 0);
4015 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4016 spa->spa_uberblock.ub_timestamp) == 0);
4017 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4018 spa->spa_load_info) == 0);
4019
4020 /*
4021 * If the bootfs property exists on this pool then we
4022 * copy it out so that external consumers can tell which
4023 * pools are bootable.
4024 */
4025 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4026 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4027
4028 /*
4029 * We have to play games with the name since the
4030 * pool was opened as TRYIMPORT_NAME.
4031 */
4032 if (dsl_dsobj_to_dsname(spa_name(spa),
4033 spa->spa_bootfs, tmpname) == 0) {
4034 char *cp;
4035 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4036
4037 cp = strchr(tmpname, '/');
4038 if (cp == NULL) {
4039 (void) strlcpy(dsname, tmpname,
4040 MAXPATHLEN);
4041 } else {
4042 (void) snprintf(dsname, MAXPATHLEN,
4043 "%s/%s", poolname, ++cp);
4044 }
4045 VERIFY(nvlist_add_string(config,
4046 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4047 kmem_free(dsname, MAXPATHLEN);
4048 }
4049 kmem_free(tmpname, MAXPATHLEN);
4050 }
4051
4052 /*
4053 * Add the list of hot spares and level 2 cache devices.
4054 */
4055 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4056 spa_add_spares(spa, config);
4057 spa_add_l2cache(spa, config);
4058 spa_config_exit(spa, SCL_CONFIG, FTAG);
4059 }
4060
4061 spa_unload(spa);
4062 spa_deactivate(spa);
4063 spa_remove(spa);
4064 mutex_exit(&spa_namespace_lock);
4065
4066 return (config);
4067 }
4068
4069 /*
4070 * Pool export/destroy
4071 *
4072 * The act of destroying or exporting a pool is very simple. We make sure there
4073 * is no more pending I/O and any references to the pool are gone. Then, we
4074 * update the pool state and sync all the labels to disk, removing the
4075 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4076 * we don't sync the labels or remove the configuration cache.
4077 */
4078 static int
4079 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4080 boolean_t force, boolean_t hardforce)
4081 {
4082 spa_t *spa;
4083
4084 if (oldconfig)
4085 *oldconfig = NULL;
4086
4087 if (!(spa_mode_global & FWRITE))
4088 return (SET_ERROR(EROFS));
4089
4090 mutex_enter(&spa_namespace_lock);
4091 if ((spa = spa_lookup(pool)) == NULL) {
4092 mutex_exit(&spa_namespace_lock);
4093 return (SET_ERROR(ENOENT));
4094 }
4095
4096 /*
4097 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4098 * reacquire the namespace lock, and see if we can export.
4099 */
4100 spa_open_ref(spa, FTAG);
4101 mutex_exit(&spa_namespace_lock);
4102 spa_async_suspend(spa);
4103 mutex_enter(&spa_namespace_lock);
4104 spa_close(spa, FTAG);
4105
4106 /*
4107 * The pool will be in core if it's openable,
4108 * in which case we can modify its state.
4109 */
4110 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4111 /*
4112 * Objsets may be open only because they're dirty, so we
4113 * have to force it to sync before checking spa_refcnt.
4114 */
4115 txg_wait_synced(spa->spa_dsl_pool, 0);
4116
4117 /*
4118 * A pool cannot be exported or destroyed if there are active
4119 * references. If we are resetting a pool, allow references by
4120 * fault injection handlers.
4121 */
4122 if (!spa_refcount_zero(spa) ||
4123 (spa->spa_inject_ref != 0 &&
4124 new_state != POOL_STATE_UNINITIALIZED)) {
4125 spa_async_resume(spa);
4126 mutex_exit(&spa_namespace_lock);
4127 return (SET_ERROR(EBUSY));
4128 }
4129
4130 /*
4131 * A pool cannot be exported if it has an active shared spare.
4132 * This is to prevent other pools stealing the active spare
4133 * from an exported pool. At user's own will, such pool can
4134 * be forcedly exported.
4135 */
4136 if (!force && new_state == POOL_STATE_EXPORTED &&
4137 spa_has_active_shared_spare(spa)) {
4138 spa_async_resume(spa);
4139 mutex_exit(&spa_namespace_lock);
4140 return (SET_ERROR(EXDEV));
4141 }
4142
4143 /*
4144 * We want this to be reflected on every label,
4145 * so mark them all dirty. spa_unload() will do the
4146 * final sync that pushes these changes out.
4147 */
4148 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4149 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4150 spa->spa_state = new_state;
4151 spa->spa_final_txg = spa_last_synced_txg(spa) +
4152 TXG_DEFER_SIZE + 1;
4153 vdev_config_dirty(spa->spa_root_vdev);
4154 spa_config_exit(spa, SCL_ALL, FTAG);
4155 }
4156 }
4157
4158 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4159
4160 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4161 spa_unload(spa);
4162 spa_deactivate(spa);
4163 }
4164
4165 if (oldconfig && spa->spa_config)
4166 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4167
4168 if (new_state != POOL_STATE_UNINITIALIZED) {
4169 if (!hardforce)
4170 spa_config_sync(spa, B_TRUE, B_TRUE);
4171 spa_remove(spa);
4172 }
4173 mutex_exit(&spa_namespace_lock);
4174
4175 return (0);
4176 }
4177
4178 /*
4179 * Destroy a storage pool.
4180 */
4181 int
4182 spa_destroy(char *pool)
4183 {
4184 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4185 B_FALSE, B_FALSE));
4186 }
4187
4188 /*
4189 * Export a storage pool.
4190 */
4191 int
4192 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4193 boolean_t hardforce)
4194 {
4195 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4196 force, hardforce));
4197 }
4198
4199 /*
4200 * Similar to spa_export(), this unloads the spa_t without actually removing it
4201 * from the namespace in any way.
4202 */
4203 int
4204 spa_reset(char *pool)
4205 {
4206 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4207 B_FALSE, B_FALSE));
4208 }
4209
4210 /*
4211 * ==========================================================================
4212 * Device manipulation
4213 * ==========================================================================
4214 */
4215
4216 /*
4217 * Add a device to a storage pool.
4218 */
4219 int
4220 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4221 {
4222 uint64_t txg, id;
4223 int error;
4224 vdev_t *rvd = spa->spa_root_vdev;
4225 vdev_t *vd, *tvd;
4226 nvlist_t **spares, **l2cache;
4227 uint_t nspares, nl2cache;
4228
4229 ASSERT(spa_writeable(spa));
4230
4231 txg = spa_vdev_enter(spa);
4232
4233 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4234 VDEV_ALLOC_ADD)) != 0)
4235 return (spa_vdev_exit(spa, NULL, txg, error));
4236
4237 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4238
4239 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4240 &nspares) != 0)
4241 nspares = 0;
4242
4243 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4244 &nl2cache) != 0)
4245 nl2cache = 0;
4246
4247 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4248 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4249
4250 if (vd->vdev_children != 0 &&
4251 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4252 return (spa_vdev_exit(spa, vd, txg, error));
4253
4254 /*
4255 * We must validate the spares and l2cache devices after checking the
4256 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4257 */
4258 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4259 return (spa_vdev_exit(spa, vd, txg, error));
4260
4261 /*
4262 * Transfer each new top-level vdev from vd to rvd.
4263 */
4264 for (int c = 0; c < vd->vdev_children; c++) {
4265
4266 /*
4267 * Set the vdev id to the first hole, if one exists.
4268 */
4269 for (id = 0; id < rvd->vdev_children; id++) {
4270 if (rvd->vdev_child[id]->vdev_ishole) {
4271 vdev_free(rvd->vdev_child[id]);
4272 break;
4273 }
4274 }
4275 tvd = vd->vdev_child[c];
4276 vdev_remove_child(vd, tvd);
4277 tvd->vdev_id = id;
4278 vdev_add_child(rvd, tvd);
4279 vdev_config_dirty(tvd);
4280 }
4281
4282 if (nspares != 0) {
4283 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4284 ZPOOL_CONFIG_SPARES);
4285 spa_load_spares(spa);
4286 spa->spa_spares.sav_sync = B_TRUE;
4287 }
4288
4289 if (nl2cache != 0) {
4290 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4291 ZPOOL_CONFIG_L2CACHE);
4292 spa_load_l2cache(spa);
4293 spa->spa_l2cache.sav_sync = B_TRUE;
4294 }
4295
4296 /*
4297 * We have to be careful when adding new vdevs to an existing pool.
4298 * If other threads start allocating from these vdevs before we
4299 * sync the config cache, and we lose power, then upon reboot we may
4300 * fail to open the pool because there are DVAs that the config cache
4301 * can't translate. Therefore, we first add the vdevs without
4302 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4303 * and then let spa_config_update() initialize the new metaslabs.
4304 *
4305 * spa_load() checks for added-but-not-initialized vdevs, so that
4306 * if we lose power at any point in this sequence, the remaining
4307 * steps will be completed the next time we load the pool.
4308 */
4309 (void) spa_vdev_exit(spa, vd, txg, 0);
4310
4311 mutex_enter(&spa_namespace_lock);
4312 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4313 mutex_exit(&spa_namespace_lock);
4314
4315 return (0);
4316 }
4317
4318 /*
4319 * Attach a device to a mirror. The arguments are the path to any device
4320 * in the mirror, and the nvroot for the new device. If the path specifies
4321 * a device that is not mirrored, we automatically insert the mirror vdev.
4322 *
4323 * If 'replacing' is specified, the new device is intended to replace the
4324 * existing device; in this case the two devices are made into their own
4325 * mirror using the 'replacing' vdev, which is functionally identical to
4326 * the mirror vdev (it actually reuses all the same ops) but has a few
4327 * extra rules: you can't attach to it after it's been created, and upon
4328 * completion of resilvering, the first disk (the one being replaced)
4329 * is automatically detached.
4330 */
4331 int
4332 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4333 {
4334 uint64_t txg, dtl_max_txg;
4335 vdev_t *rvd = spa->spa_root_vdev;
4336 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4337 vdev_ops_t *pvops;
4338 char *oldvdpath, *newvdpath;
4339 int newvd_isspare;
4340 int error;
4341
4342 ASSERT(spa_writeable(spa));
4343
4344 txg = spa_vdev_enter(spa);
4345
4346 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4347
4348 if (oldvd == NULL)
4349 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4350
4351 if (!oldvd->vdev_ops->vdev_op_leaf)
4352 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4353
4354 pvd = oldvd->vdev_parent;
4355
4356 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4357 VDEV_ALLOC_ATTACH)) != 0)
4358 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4359
4360 if (newrootvd->vdev_children != 1)
4361 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4362
4363 newvd = newrootvd->vdev_child[0];
4364
4365 if (!newvd->vdev_ops->vdev_op_leaf)
4366 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4367
4368 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4369 return (spa_vdev_exit(spa, newrootvd, txg, error));
4370
4371 /*
4372 * Spares can't replace logs
4373 */
4374 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4375 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4376
4377 if (!replacing) {
4378 /*
4379 * For attach, the only allowable parent is a mirror or the root
4380 * vdev.
4381 */
4382 if (pvd->vdev_ops != &vdev_mirror_ops &&
4383 pvd->vdev_ops != &vdev_root_ops)
4384 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4385
4386 pvops = &vdev_mirror_ops;
4387 } else {
4388 /*
4389 * Active hot spares can only be replaced by inactive hot
4390 * spares.
4391 */
4392 if (pvd->vdev_ops == &vdev_spare_ops &&
4393 oldvd->vdev_isspare &&
4394 !spa_has_spare(spa, newvd->vdev_guid))
4395 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4396
4397 /*
4398 * If the source is a hot spare, and the parent isn't already a
4399 * spare, then we want to create a new hot spare. Otherwise, we
4400 * want to create a replacing vdev. The user is not allowed to
4401 * attach to a spared vdev child unless the 'isspare' state is
4402 * the same (spare replaces spare, non-spare replaces
4403 * non-spare).
4404 */
4405 if (pvd->vdev_ops == &vdev_replacing_ops &&
4406 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4407 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4408 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4409 newvd->vdev_isspare != oldvd->vdev_isspare) {
4410 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4411 }
4412
4413 if (newvd->vdev_isspare)
4414 pvops = &vdev_spare_ops;
4415 else
4416 pvops = &vdev_replacing_ops;
4417 }
4418
4419 /*
4420 * Make sure the new device is big enough.
4421 */
4422 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4423 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4424
4425 /*
4426 * The new device cannot have a higher alignment requirement
4427 * than the top-level vdev.
4428 */
4429 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4430 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4431
4432 /*
4433 * If this is an in-place replacement, update oldvd's path and devid
4434 * to make it distinguishable from newvd, and unopenable from now on.
4435 */
4436 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4437 spa_strfree(oldvd->vdev_path);
4438 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4439 KM_SLEEP);
4440 (void) sprintf(oldvd->vdev_path, "%s/%s",
4441 newvd->vdev_path, "old");
4442 if (oldvd->vdev_devid != NULL) {
4443 spa_strfree(oldvd->vdev_devid);
4444 oldvd->vdev_devid = NULL;
4445 }
4446 }
4447
4448 /* mark the device being resilvered */
4449 newvd->vdev_resilver_txg = txg;
4450
4451 /*
4452 * If the parent is not a mirror, or if we're replacing, insert the new
4453 * mirror/replacing/spare vdev above oldvd.
4454 */
4455 if (pvd->vdev_ops != pvops)
4456 pvd = vdev_add_parent(oldvd, pvops);
4457
4458 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4459 ASSERT(pvd->vdev_ops == pvops);
4460 ASSERT(oldvd->vdev_parent == pvd);
4461
4462 /*
4463 * Extract the new device from its root and add it to pvd.
4464 */
4465 vdev_remove_child(newrootvd, newvd);
4466 newvd->vdev_id = pvd->vdev_children;
4467 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4468 vdev_add_child(pvd, newvd);
4469
4470 tvd = newvd->vdev_top;
4471 ASSERT(pvd->vdev_top == tvd);
4472 ASSERT(tvd->vdev_parent == rvd);
4473
4474 vdev_config_dirty(tvd);
4475
4476 /*
4477 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4478 * for any dmu_sync-ed blocks. It will propagate upward when
4479 * spa_vdev_exit() calls vdev_dtl_reassess().
4480 */
4481 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4482
4483 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4484 dtl_max_txg - TXG_INITIAL);
4485
4486 if (newvd->vdev_isspare) {
4487 spa_spare_activate(newvd);
4488 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4489 }
4490
4491 oldvdpath = spa_strdup(oldvd->vdev_path);
4492 newvdpath = spa_strdup(newvd->vdev_path);
4493 newvd_isspare = newvd->vdev_isspare;
4494
4495 /*
4496 * Mark newvd's DTL dirty in this txg.
4497 */
4498 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4499
4500 /*
4501 * Schedule the resilver to restart in the future. We do this to
4502 * ensure that dmu_sync-ed blocks have been stitched into the
4503 * respective datasets.
4504 */
4505 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4506
4507 /*
4508 * Commit the config
4509 */
4510 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4511
4512 spa_history_log_internal(spa, "vdev attach", NULL,
4513 "%s vdev=%s %s vdev=%s",
4514 replacing && newvd_isspare ? "spare in" :
4515 replacing ? "replace" : "attach", newvdpath,
4516 replacing ? "for" : "to", oldvdpath);
4517
4518 spa_strfree(oldvdpath);
4519 spa_strfree(newvdpath);
4520
4521 if (spa->spa_bootfs)
4522 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4523
4524 return (0);
4525 }
4526
4527 /*
4528 * Detach a device from a mirror or replacing vdev.
4529 *
4530 * If 'replace_done' is specified, only detach if the parent
4531 * is a replacing vdev.
4532 */
4533 int
4534 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4535 {
4536 uint64_t txg;
4537 int error;
4538 vdev_t *rvd = spa->spa_root_vdev;
4539 vdev_t *vd, *pvd, *cvd, *tvd;
4540 boolean_t unspare = B_FALSE;
4541 uint64_t unspare_guid = 0;
4542 char *vdpath;
4543
4544 ASSERT(spa_writeable(spa));
4545
4546 txg = spa_vdev_enter(spa);
4547
4548 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4549
4550 if (vd == NULL)
4551 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4552
4553 if (!vd->vdev_ops->vdev_op_leaf)
4554 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4555
4556 pvd = vd->vdev_parent;
4557
4558 /*
4559 * If the parent/child relationship is not as expected, don't do it.
4560 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4561 * vdev that's replacing B with C. The user's intent in replacing
4562 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4563 * the replace by detaching C, the expected behavior is to end up
4564 * M(A,B). But suppose that right after deciding to detach C,
4565 * the replacement of B completes. We would have M(A,C), and then
4566 * ask to detach C, which would leave us with just A -- not what
4567 * the user wanted. To prevent this, we make sure that the
4568 * parent/child relationship hasn't changed -- in this example,
4569 * that C's parent is still the replacing vdev R.
4570 */
4571 if (pvd->vdev_guid != pguid && pguid != 0)
4572 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4573
4574 /*
4575 * Only 'replacing' or 'spare' vdevs can be replaced.
4576 */
4577 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4578 pvd->vdev_ops != &vdev_spare_ops)
4579 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4580
4581 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4582 spa_version(spa) >= SPA_VERSION_SPARES);
4583
4584 /*
4585 * Only mirror, replacing, and spare vdevs support detach.
4586 */
4587 if (pvd->vdev_ops != &vdev_replacing_ops &&
4588 pvd->vdev_ops != &vdev_mirror_ops &&
4589 pvd->vdev_ops != &vdev_spare_ops)
4590 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4591
4592 /*
4593 * If this device has the only valid copy of some data,
4594 * we cannot safely detach it.
4595 */
4596 if (vdev_dtl_required(vd))
4597 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4598
4599 ASSERT(pvd->vdev_children >= 2);
4600
4601 /*
4602 * If we are detaching the second disk from a replacing vdev, then
4603 * check to see if we changed the original vdev's path to have "/old"
4604 * at the end in spa_vdev_attach(). If so, undo that change now.
4605 */
4606 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4607 vd->vdev_path != NULL) {
4608 size_t len = strlen(vd->vdev_path);
4609
4610 for (int c = 0; c < pvd->vdev_children; c++) {
4611 cvd = pvd->vdev_child[c];
4612
4613 if (cvd == vd || cvd->vdev_path == NULL)
4614 continue;
4615
4616 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4617 strcmp(cvd->vdev_path + len, "/old") == 0) {
4618 spa_strfree(cvd->vdev_path);
4619 cvd->vdev_path = spa_strdup(vd->vdev_path);
4620 break;
4621 }
4622 }
4623 }
4624
4625 /*
4626 * If we are detaching the original disk from a spare, then it implies
4627 * that the spare should become a real disk, and be removed from the
4628 * active spare list for the pool.
4629 */
4630 if (pvd->vdev_ops == &vdev_spare_ops &&
4631 vd->vdev_id == 0 &&
4632 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4633 unspare = B_TRUE;
4634
4635 /*
4636 * Erase the disk labels so the disk can be used for other things.
4637 * This must be done after all other error cases are handled,
4638 * but before we disembowel vd (so we can still do I/O to it).
4639 * But if we can't do it, don't treat the error as fatal --
4640 * it may be that the unwritability of the disk is the reason
4641 * it's being detached!
4642 */
4643 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4644
4645 /*
4646 * Remove vd from its parent and compact the parent's children.
4647 */
4648 vdev_remove_child(pvd, vd);
4649 vdev_compact_children(pvd);
4650
4651 /*
4652 * Remember one of the remaining children so we can get tvd below.
4653 */
4654 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4655
4656 /*
4657 * If we need to remove the remaining child from the list of hot spares,
4658 * do it now, marking the vdev as no longer a spare in the process.
4659 * We must do this before vdev_remove_parent(), because that can
4660 * change the GUID if it creates a new toplevel GUID. For a similar
4661 * reason, we must remove the spare now, in the same txg as the detach;
4662 * otherwise someone could attach a new sibling, change the GUID, and
4663 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4664 */
4665 if (unspare) {
4666 ASSERT(cvd->vdev_isspare);
4667 spa_spare_remove(cvd);
4668 unspare_guid = cvd->vdev_guid;
4669 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4670 cvd->vdev_unspare = B_TRUE;
4671 }
4672
4673 /*
4674 * If the parent mirror/replacing vdev only has one child,
4675 * the parent is no longer needed. Remove it from the tree.
4676 */
4677 if (pvd->vdev_children == 1) {
4678 if (pvd->vdev_ops == &vdev_spare_ops)
4679 cvd->vdev_unspare = B_FALSE;
4680 vdev_remove_parent(cvd);
4681 }
4682
4683
4684 /*
4685 * We don't set tvd until now because the parent we just removed
4686 * may have been the previous top-level vdev.
4687 */
4688 tvd = cvd->vdev_top;
4689 ASSERT(tvd->vdev_parent == rvd);
4690
4691 /*
4692 * Reevaluate the parent vdev state.
4693 */
4694 vdev_propagate_state(cvd);
4695
4696 /*
4697 * If the 'autoexpand' property is set on the pool then automatically
4698 * try to expand the size of the pool. For example if the device we
4699 * just detached was smaller than the others, it may be possible to
4700 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4701 * first so that we can obtain the updated sizes of the leaf vdevs.
4702 */
4703 if (spa->spa_autoexpand) {
4704 vdev_reopen(tvd);
4705 vdev_expand(tvd, txg);
4706 }
4707
4708 vdev_config_dirty(tvd);
4709
4710 /*
4711 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4712 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4713 * But first make sure we're not on any *other* txg's DTL list, to
4714 * prevent vd from being accessed after it's freed.
4715 */
4716 vdpath = spa_strdup(vd->vdev_path);
4717 for (int t = 0; t < TXG_SIZE; t++)
4718 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4719 vd->vdev_detached = B_TRUE;
4720 vdev_dirty(tvd, VDD_DTL, vd, txg);
4721
4722 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4723
4724 /* hang on to the spa before we release the lock */
4725 spa_open_ref(spa, FTAG);
4726
4727 error = spa_vdev_exit(spa, vd, txg, 0);
4728
4729 spa_history_log_internal(spa, "detach", NULL,
4730 "vdev=%s", vdpath);
4731 spa_strfree(vdpath);
4732
4733 /*
4734 * If this was the removal of the original device in a hot spare vdev,
4735 * then we want to go through and remove the device from the hot spare
4736 * list of every other pool.
4737 */
4738 if (unspare) {
4739 spa_t *altspa = NULL;
4740
4741 mutex_enter(&spa_namespace_lock);
4742 while ((altspa = spa_next(altspa)) != NULL) {
4743 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4744 altspa == spa)
4745 continue;
4746
4747 spa_open_ref(altspa, FTAG);
4748 mutex_exit(&spa_namespace_lock);
4749 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4750 mutex_enter(&spa_namespace_lock);
4751 spa_close(altspa, FTAG);
4752 }
4753 mutex_exit(&spa_namespace_lock);
4754
4755 /* search the rest of the vdevs for spares to remove */
4756 spa_vdev_resilver_done(spa);
4757 }
4758
4759 /* all done with the spa; OK to release */
4760 mutex_enter(&spa_namespace_lock);
4761 spa_close(spa, FTAG);
4762 mutex_exit(&spa_namespace_lock);
4763
4764 return (error);
4765 }
4766
4767 /*
4768 * Split a set of devices from their mirrors, and create a new pool from them.
4769 */
4770 int
4771 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4772 nvlist_t *props, boolean_t exp)
4773 {
4774 int error = 0;
4775 uint64_t txg, *glist;
4776 spa_t *newspa;
4777 uint_t c, children, lastlog;
4778 nvlist_t **child, *nvl, *tmp;
4779 dmu_tx_t *tx;
4780 char *altroot = NULL;
4781 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4782 boolean_t activate_slog;
4783
4784 ASSERT(spa_writeable(spa));
4785
4786 txg = spa_vdev_enter(spa);
4787
4788 /* clear the log and flush everything up to now */
4789 activate_slog = spa_passivate_log(spa);
4790 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4791 error = spa_offline_log(spa);
4792 txg = spa_vdev_config_enter(spa);
4793
4794 if (activate_slog)
4795 spa_activate_log(spa);
4796
4797 if (error != 0)
4798 return (spa_vdev_exit(spa, NULL, txg, error));
4799
4800 /* check new spa name before going any further */
4801 if (spa_lookup(newname) != NULL)
4802 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4803
4804 /*
4805 * scan through all the children to ensure they're all mirrors
4806 */
4807 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4808 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4809 &children) != 0)
4810 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4811
4812 /* first, check to ensure we've got the right child count */
4813 rvd = spa->spa_root_vdev;
4814 lastlog = 0;
4815 for (c = 0; c < rvd->vdev_children; c++) {
4816 vdev_t *vd = rvd->vdev_child[c];
4817
4818 /* don't count the holes & logs as children */
4819 if (vd->vdev_islog || vd->vdev_ishole) {
4820 if (lastlog == 0)
4821 lastlog = c;
4822 continue;
4823 }
4824
4825 lastlog = 0;
4826 }
4827 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4828 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4829
4830 /* next, ensure no spare or cache devices are part of the split */
4831 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4832 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4833 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4834
4835 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4836 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4837
4838 /* then, loop over each vdev and validate it */
4839 for (c = 0; c < children; c++) {
4840 uint64_t is_hole = 0;
4841
4842 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4843 &is_hole);
4844
4845 if (is_hole != 0) {
4846 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4847 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4848 continue;
4849 } else {
4850 error = SET_ERROR(EINVAL);
4851 break;
4852 }
4853 }
4854
4855 /* which disk is going to be split? */
4856 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4857 &glist[c]) != 0) {
4858 error = SET_ERROR(EINVAL);
4859 break;
4860 }
4861
4862 /* look it up in the spa */
4863 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4864 if (vml[c] == NULL) {
4865 error = SET_ERROR(ENODEV);
4866 break;
4867 }
4868
4869 /* make sure there's nothing stopping the split */
4870 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4871 vml[c]->vdev_islog ||
4872 vml[c]->vdev_ishole ||
4873 vml[c]->vdev_isspare ||
4874 vml[c]->vdev_isl2cache ||
4875 !vdev_writeable(vml[c]) ||
4876 vml[c]->vdev_children != 0 ||
4877 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4878 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4879 error = SET_ERROR(EINVAL);
4880 break;
4881 }
4882
4883 if (vdev_dtl_required(vml[c])) {
4884 error = SET_ERROR(EBUSY);
4885 break;
4886 }
4887
4888 /* we need certain info from the top level */
4889 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4890 vml[c]->vdev_top->vdev_ms_array) == 0);
4891 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4892 vml[c]->vdev_top->vdev_ms_shift) == 0);
4893 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4894 vml[c]->vdev_top->vdev_asize) == 0);
4895 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4896 vml[c]->vdev_top->vdev_ashift) == 0);
4897 }
4898
4899 if (error != 0) {
4900 kmem_free(vml, children * sizeof (vdev_t *));
4901 kmem_free(glist, children * sizeof (uint64_t));
4902 return (spa_vdev_exit(spa, NULL, txg, error));
4903 }
4904
4905 /* stop writers from using the disks */
4906 for (c = 0; c < children; c++) {
4907 if (vml[c] != NULL)
4908 vml[c]->vdev_offline = B_TRUE;
4909 }
4910 vdev_reopen(spa->spa_root_vdev);
4911
4912 /*
4913 * Temporarily record the splitting vdevs in the spa config. This
4914 * will disappear once the config is regenerated.
4915 */
4916 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4917 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4918 glist, children) == 0);
4919 kmem_free(glist, children * sizeof (uint64_t));
4920
4921 mutex_enter(&spa->spa_props_lock);
4922 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4923 nvl) == 0);
4924 mutex_exit(&spa->spa_props_lock);
4925 spa->spa_config_splitting = nvl;
4926 vdev_config_dirty(spa->spa_root_vdev);
4927
4928 /* configure and create the new pool */
4929 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4930 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4931 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4932 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4933 spa_version(spa)) == 0);
4934 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4935 spa->spa_config_txg) == 0);
4936 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4937 spa_generate_guid(NULL)) == 0);
4938 (void) nvlist_lookup_string(props,
4939 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4940
4941 /* add the new pool to the namespace */
4942 newspa = spa_add(newname, config, altroot);
4943 newspa->spa_config_txg = spa->spa_config_txg;
4944 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4945
4946 /* release the spa config lock, retaining the namespace lock */
4947 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4948
4949 if (zio_injection_enabled)
4950 zio_handle_panic_injection(spa, FTAG, 1);
4951
4952 spa_activate(newspa, spa_mode_global);
4953 spa_async_suspend(newspa);
4954
4955 /* create the new pool from the disks of the original pool */
4956 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4957 if (error)
4958 goto out;
4959
4960 /* if that worked, generate a real config for the new pool */
4961 if (newspa->spa_root_vdev != NULL) {
4962 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4963 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4964 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4965 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4966 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4967 B_TRUE));
4968 }
4969
4970 /* set the props */
4971 if (props != NULL) {
4972 spa_configfile_set(newspa, props, B_FALSE);
4973 error = spa_prop_set(newspa, props);
4974 if (error)
4975 goto out;
4976 }
4977
4978 /* flush everything */
4979 txg = spa_vdev_config_enter(newspa);
4980 vdev_config_dirty(newspa->spa_root_vdev);
4981 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4982
4983 if (zio_injection_enabled)
4984 zio_handle_panic_injection(spa, FTAG, 2);
4985
4986 spa_async_resume(newspa);
4987
4988 /* finally, update the original pool's config */
4989 txg = spa_vdev_config_enter(spa);
4990 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4991 error = dmu_tx_assign(tx, TXG_WAIT);
4992 if (error != 0)
4993 dmu_tx_abort(tx);
4994 for (c = 0; c < children; c++) {
4995 if (vml[c] != NULL) {
4996 vdev_split(vml[c]);
4997 if (error == 0)
4998 spa_history_log_internal(spa, "detach", tx,
4999 "vdev=%s", vml[c]->vdev_path);
5000 vdev_free(vml[c]);
5001 }
5002 }
5003 vdev_config_dirty(spa->spa_root_vdev);
5004 spa->spa_config_splitting = NULL;
5005 nvlist_free(nvl);
5006 if (error == 0)
5007 dmu_tx_commit(tx);
5008 (void) spa_vdev_exit(spa, NULL, txg, 0);
5009
5010 if (zio_injection_enabled)
5011 zio_handle_panic_injection(spa, FTAG, 3);
5012
5013 /* split is complete; log a history record */
5014 spa_history_log_internal(newspa, "split", NULL,
5015 "from pool %s", spa_name(spa));
5016
5017 kmem_free(vml, children * sizeof (vdev_t *));
5018
5019 /* if we're not going to mount the filesystems in userland, export */
5020 if (exp)
5021 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5022 B_FALSE, B_FALSE);
5023
5024 return (error);
5025
5026 out:
5027 spa_unload(newspa);
5028 spa_deactivate(newspa);
5029 spa_remove(newspa);
5030
5031 txg = spa_vdev_config_enter(spa);
5032
5033 /* re-online all offlined disks */
5034 for (c = 0; c < children; c++) {
5035 if (vml[c] != NULL)
5036 vml[c]->vdev_offline = B_FALSE;
5037 }
5038 vdev_reopen(spa->spa_root_vdev);
5039
5040 nvlist_free(spa->spa_config_splitting);
5041 spa->spa_config_splitting = NULL;
5042 (void) spa_vdev_exit(spa, NULL, txg, error);
5043
5044 kmem_free(vml, children * sizeof (vdev_t *));
5045 return (error);
5046 }
5047
5048 static nvlist_t *
5049 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5050 {
5051 for (int i = 0; i < count; i++) {
5052 uint64_t guid;
5053
5054 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5055 &guid) == 0);
5056
5057 if (guid == target_guid)
5058 return (nvpp[i]);
5059 }
5060
5061 return (NULL);
5062 }
5063
5064 static void
5065 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5066 nvlist_t *dev_to_remove)
5067 {
5068 nvlist_t **newdev = NULL;
5069
5070 if (count > 1)
5071 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5072
5073 for (int i = 0, j = 0; i < count; i++) {
5074 if (dev[i] == dev_to_remove)
5075 continue;
5076 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5077 }
5078
5079 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5080 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5081
5082 for (int i = 0; i < count - 1; i++)
5083 nvlist_free(newdev[i]);
5084
5085 if (count > 1)
5086 kmem_free(newdev, (count - 1) * sizeof (void *));
5087 }
5088
5089 /*
5090 * Evacuate the device.
5091 */
5092 static int
5093 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5094 {
5095 uint64_t txg;
5096 int error = 0;
5097
5098 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5099 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5100 ASSERT(vd == vd->vdev_top);
5101
5102 /*
5103 * Evacuate the device. We don't hold the config lock as writer
5104 * since we need to do I/O but we do keep the
5105 * spa_namespace_lock held. Once this completes the device
5106 * should no longer have any blocks allocated on it.
5107 */
5108 if (vd->vdev_islog) {
5109 if (vd->vdev_stat.vs_alloc != 0)
5110 error = spa_offline_log(spa);
5111 } else {
5112 error = SET_ERROR(ENOTSUP);
5113 }
5114
5115 if (error)
5116 return (error);
5117
5118 /*
5119 * The evacuation succeeded. Remove any remaining MOS metadata
5120 * associated with this vdev, and wait for these changes to sync.
5121 */
5122 ASSERT0(vd->vdev_stat.vs_alloc);
5123 txg = spa_vdev_config_enter(spa);
5124 vd->vdev_removing = B_TRUE;
5125 vdev_dirty_leaves(vd, VDD_DTL, txg);
5126 vdev_config_dirty(vd);
5127 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5128
5129 return (0);
5130 }
5131
5132 /*
5133 * Complete the removal by cleaning up the namespace.
5134 */
5135 static void
5136 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5137 {
5138 vdev_t *rvd = spa->spa_root_vdev;
5139 uint64_t id = vd->vdev_id;
5140 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5141
5142 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5143 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5144 ASSERT(vd == vd->vdev_top);
5145
5146 /*
5147 * Only remove any devices which are empty.
5148 */
5149 if (vd->vdev_stat.vs_alloc != 0)
5150 return;
5151
5152 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5153
5154 if (list_link_active(&vd->vdev_state_dirty_node))
5155 vdev_state_clean(vd);
5156 if (list_link_active(&vd->vdev_config_dirty_node))
5157 vdev_config_clean(vd);
5158
5159 vdev_free(vd);
5160
5161 if (last_vdev) {
5162 vdev_compact_children(rvd);
5163 } else {
5164 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5165 vdev_add_child(rvd, vd);
5166 }
5167 vdev_config_dirty(rvd);
5168
5169 /*
5170 * Reassess the health of our root vdev.
5171 */
5172 vdev_reopen(rvd);
5173 }
5174
5175 /*
5176 * Remove a device from the pool -
5177 *
5178 * Removing a device from the vdev namespace requires several steps
5179 * and can take a significant amount of time. As a result we use
5180 * the spa_vdev_config_[enter/exit] functions which allow us to
5181 * grab and release the spa_config_lock while still holding the namespace
5182 * lock. During each step the configuration is synced out.
5183 *
5184 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5185 * devices.
5186 */
5187 int
5188 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5189 {
5190 vdev_t *vd;
5191 metaslab_group_t *mg;
5192 nvlist_t **spares, **l2cache, *nv;
5193 uint64_t txg = 0;
5194 uint_t nspares, nl2cache;
5195 int error = 0;
5196 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5197
5198 ASSERT(spa_writeable(spa));
5199
5200 if (!locked)
5201 txg = spa_vdev_enter(spa);
5202
5203 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5204
5205 if (spa->spa_spares.sav_vdevs != NULL &&
5206 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5207 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5208 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5209 /*
5210 * Only remove the hot spare if it's not currently in use
5211 * in this pool.
5212 */
5213 if (vd == NULL || unspare) {
5214 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5215 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5216 spa_load_spares(spa);
5217 spa->spa_spares.sav_sync = B_TRUE;
5218 } else {
5219 error = SET_ERROR(EBUSY);
5220 }
5221 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5222 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5223 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5224 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5225 /*
5226 * Cache devices can always be removed.
5227 */
5228 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5229 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5230 spa_load_l2cache(spa);
5231 spa->spa_l2cache.sav_sync = B_TRUE;
5232 } else if (vd != NULL && vd->vdev_islog) {
5233 ASSERT(!locked);
5234 ASSERT(vd == vd->vdev_top);
5235
5236 /*
5237 * XXX - Once we have bp-rewrite this should
5238 * become the common case.
5239 */
5240
5241 mg = vd->vdev_mg;
5242
5243 /*
5244 * Stop allocating from this vdev.
5245 */
5246 metaslab_group_passivate(mg);
5247
5248 /*
5249 * Wait for the youngest allocations and frees to sync,
5250 * and then wait for the deferral of those frees to finish.
5251 */
5252 spa_vdev_config_exit(spa, NULL,
5253 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5254
5255 /*
5256 * Attempt to evacuate the vdev.
5257 */
5258 error = spa_vdev_remove_evacuate(spa, vd);
5259
5260 txg = spa_vdev_config_enter(spa);
5261
5262 /*
5263 * If we couldn't evacuate the vdev, unwind.
5264 */
5265 if (error) {
5266 metaslab_group_activate(mg);
5267 return (spa_vdev_exit(spa, NULL, txg, error));
5268 }
5269
5270 /*
5271 * Clean up the vdev namespace.
5272 */
5273 spa_vdev_remove_from_namespace(spa, vd);
5274
5275 } else if (vd != NULL) {
5276 /*
5277 * Normal vdevs cannot be removed (yet).
5278 */
5279 error = SET_ERROR(ENOTSUP);
5280 } else {
5281 /*
5282 * There is no vdev of any kind with the specified guid.
5283 */
5284 error = SET_ERROR(ENOENT);
5285 }
5286
5287 if (!locked)
5288 return (spa_vdev_exit(spa, NULL, txg, error));
5289
5290 return (error);
5291 }
5292
5293 /*
5294 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5295 * currently spared, so we can detach it.
5296 */
5297 static vdev_t *
5298 spa_vdev_resilver_done_hunt(vdev_t *vd)
5299 {
5300 vdev_t *newvd, *oldvd;
5301
5302 for (int c = 0; c < vd->vdev_children; c++) {
5303 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5304 if (oldvd != NULL)
5305 return (oldvd);
5306 }
5307
5308 /*
5309 * Check for a completed replacement. We always consider the first
5310 * vdev in the list to be the oldest vdev, and the last one to be
5311 * the newest (see spa_vdev_attach() for how that works). In
5312 * the case where the newest vdev is faulted, we will not automatically
5313 * remove it after a resilver completes. This is OK as it will require
5314 * user intervention to determine which disk the admin wishes to keep.
5315 */
5316 if (vd->vdev_ops == &vdev_replacing_ops) {
5317 ASSERT(vd->vdev_children > 1);
5318
5319 newvd = vd->vdev_child[vd->vdev_children - 1];
5320 oldvd = vd->vdev_child[0];
5321
5322 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5323 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5324 !vdev_dtl_required(oldvd))
5325 return (oldvd);
5326 }
5327
5328 /*
5329 * Check for a completed resilver with the 'unspare' flag set.
5330 */
5331 if (vd->vdev_ops == &vdev_spare_ops) {
5332 vdev_t *first = vd->vdev_child[0];
5333 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5334
5335 if (last->vdev_unspare) {
5336 oldvd = first;
5337 newvd = last;
5338 } else if (first->vdev_unspare) {
5339 oldvd = last;
5340 newvd = first;
5341 } else {
5342 oldvd = NULL;
5343 }
5344
5345 if (oldvd != NULL &&
5346 vdev_dtl_empty(newvd, DTL_MISSING) &&
5347 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5348 !vdev_dtl_required(oldvd))
5349 return (oldvd);
5350
5351 /*
5352 * If there are more than two spares attached to a disk,
5353 * and those spares are not required, then we want to
5354 * attempt to free them up now so that they can be used
5355 * by other pools. Once we're back down to a single
5356 * disk+spare, we stop removing them.
5357 */
5358 if (vd->vdev_children > 2) {
5359 newvd = vd->vdev_child[1];
5360
5361 if (newvd->vdev_isspare && last->vdev_isspare &&
5362 vdev_dtl_empty(last, DTL_MISSING) &&
5363 vdev_dtl_empty(last, DTL_OUTAGE) &&
5364 !vdev_dtl_required(newvd))
5365 return (newvd);
5366 }
5367 }
5368
5369 return (NULL);
5370 }
5371
5372 static void
5373 spa_vdev_resilver_done(spa_t *spa)
5374 {
5375 vdev_t *vd, *pvd, *ppvd;
5376 uint64_t guid, sguid, pguid, ppguid;
5377
5378 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5379
5380 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5381 pvd = vd->vdev_parent;
5382 ppvd = pvd->vdev_parent;
5383 guid = vd->vdev_guid;
5384 pguid = pvd->vdev_guid;
5385 ppguid = ppvd->vdev_guid;
5386 sguid = 0;
5387 /*
5388 * If we have just finished replacing a hot spared device, then
5389 * we need to detach the parent's first child (the original hot
5390 * spare) as well.
5391 */
5392 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5393 ppvd->vdev_children == 2) {
5394 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5395 sguid = ppvd->vdev_child[1]->vdev_guid;
5396 }
5397 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5398
5399 spa_config_exit(spa, SCL_ALL, FTAG);
5400 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5401 return;
5402 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5403 return;
5404 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5405 }
5406
5407 spa_config_exit(spa, SCL_ALL, FTAG);
5408 }
5409
5410 /*
5411 * Update the stored path or FRU for this vdev.
5412 */
5413 int
5414 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5415 boolean_t ispath)
5416 {
5417 vdev_t *vd;
5418 boolean_t sync = B_FALSE;
5419
5420 ASSERT(spa_writeable(spa));
5421
5422 spa_vdev_state_enter(spa, SCL_ALL);
5423
5424 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5425 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5426
5427 if (!vd->vdev_ops->vdev_op_leaf)
5428 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5429
5430 if (ispath) {
5431 if (strcmp(value, vd->vdev_path) != 0) {
5432 spa_strfree(vd->vdev_path);
5433 vd->vdev_path = spa_strdup(value);
5434 sync = B_TRUE;
5435 }
5436 } else {
5437 if (vd->vdev_fru == NULL) {
5438 vd->vdev_fru = spa_strdup(value);
5439 sync = B_TRUE;
5440 } else if (strcmp(value, vd->vdev_fru) != 0) {
5441 spa_strfree(vd->vdev_fru);
5442 vd->vdev_fru = spa_strdup(value);
5443 sync = B_TRUE;
5444 }
5445 }
5446
5447 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5448 }
5449
5450 int
5451 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5452 {
5453 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5454 }
5455
5456 int
5457 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5458 {
5459 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5460 }
5461
5462 /*
5463 * ==========================================================================
5464 * SPA Scanning
5465 * ==========================================================================
5466 */
5467
5468 int
5469 spa_scan_stop(spa_t *spa)
5470 {
5471 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5472 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5473 return (SET_ERROR(EBUSY));
5474 return (dsl_scan_cancel(spa->spa_dsl_pool));
5475 }
5476
5477 int
5478 spa_scan(spa_t *spa, pool_scan_func_t func)
5479 {
5480 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5481
5482 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5483 return (SET_ERROR(ENOTSUP));
5484
5485 /*
5486 * If a resilver was requested, but there is no DTL on a
5487 * writeable leaf device, we have nothing to do.
5488 */
5489 if (func == POOL_SCAN_RESILVER &&
5490 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5491 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5492 return (0);
5493 }
5494
5495 return (dsl_scan(spa->spa_dsl_pool, func));
5496 }
5497
5498 /*
5499 * ==========================================================================
5500 * SPA async task processing
5501 * ==========================================================================
5502 */
5503
5504 static void
5505 spa_async_remove(spa_t *spa, vdev_t *vd)
5506 {
5507 if (vd->vdev_remove_wanted) {
5508 vd->vdev_remove_wanted = B_FALSE;
5509 vd->vdev_delayed_close = B_FALSE;
5510 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5511
5512 /*
5513 * We want to clear the stats, but we don't want to do a full
5514 * vdev_clear() as that will cause us to throw away
5515 * degraded/faulted state as well as attempt to reopen the
5516 * device, all of which is a waste.
5517 */
5518 vd->vdev_stat.vs_read_errors = 0;
5519 vd->vdev_stat.vs_write_errors = 0;
5520 vd->vdev_stat.vs_checksum_errors = 0;
5521
5522 vdev_state_dirty(vd->vdev_top);
5523 }
5524
5525 for (int c = 0; c < vd->vdev_children; c++)
5526 spa_async_remove(spa, vd->vdev_child[c]);
5527 }
5528
5529 static void
5530 spa_async_probe(spa_t *spa, vdev_t *vd)
5531 {
5532 if (vd->vdev_probe_wanted) {
5533 vd->vdev_probe_wanted = B_FALSE;
5534 vdev_reopen(vd); /* vdev_open() does the actual probe */
5535 }
5536
5537 for (int c = 0; c < vd->vdev_children; c++)
5538 spa_async_probe(spa, vd->vdev_child[c]);
5539 }
5540
5541 static void
5542 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5543 {
5544 sysevent_id_t eid;
5545 nvlist_t *attr;
5546 char *physpath;
5547
5548 if (!spa->spa_autoexpand)
5549 return;
5550
5551 for (int c = 0; c < vd->vdev_children; c++) {
5552 vdev_t *cvd = vd->vdev_child[c];
5553 spa_async_autoexpand(spa, cvd);
5554 }
5555
5556 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5557 return;
5558
5559 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5560 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5561
5562 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5563 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5564
5565 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5566 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5567
5568 nvlist_free(attr);
5569 kmem_free(physpath, MAXPATHLEN);
5570 }
5571
5572 static void
5573 spa_async_thread(spa_t *spa)
5574 {
5575 int tasks;
5576
5577 ASSERT(spa->spa_sync_on);
5578
5579 mutex_enter(&spa->spa_async_lock);
5580 tasks = spa->spa_async_tasks;
5581 spa->spa_async_tasks = 0;
5582 mutex_exit(&spa->spa_async_lock);
5583
5584 /*
5585 * See if the config needs to be updated.
5586 */
5587 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5588 uint64_t old_space, new_space;
5589
5590 mutex_enter(&spa_namespace_lock);
5591 old_space = metaslab_class_get_space(spa_normal_class(spa));
5592 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5593 new_space = metaslab_class_get_space(spa_normal_class(spa));
5594 mutex_exit(&spa_namespace_lock);
5595
5596 /*
5597 * If the pool grew as a result of the config update,
5598 * then log an internal history event.
5599 */
5600 if (new_space != old_space) {
5601 spa_history_log_internal(spa, "vdev online", NULL,
5602 "pool '%s' size: %llu(+%llu)",
5603 spa_name(spa), new_space, new_space - old_space);
5604 }
5605 }
5606
5607 /*
5608 * See if any devices need to be marked REMOVED.
5609 */
5610 if (tasks & SPA_ASYNC_REMOVE) {
5611 spa_vdev_state_enter(spa, SCL_NONE);
5612 spa_async_remove(spa, spa->spa_root_vdev);
5613 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5614 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5615 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5616 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5617 (void) spa_vdev_state_exit(spa, NULL, 0);
5618 }
5619
5620 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5621 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5622 spa_async_autoexpand(spa, spa->spa_root_vdev);
5623 spa_config_exit(spa, SCL_CONFIG, FTAG);
5624 }
5625
5626 /*
5627 * See if any devices need to be probed.
5628 */
5629 if (tasks & SPA_ASYNC_PROBE) {
5630 spa_vdev_state_enter(spa, SCL_NONE);
5631 spa_async_probe(spa, spa->spa_root_vdev);
5632 (void) spa_vdev_state_exit(spa, NULL, 0);
5633 }
5634
5635 /*
5636 * If any devices are done replacing, detach them.
5637 */
5638 if (tasks & SPA_ASYNC_RESILVER_DONE)
5639 spa_vdev_resilver_done(spa);
5640
5641 /*
5642 * Kick off a resilver.
5643 */
5644 if (tasks & SPA_ASYNC_RESILVER)
5645 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5646
5647 /*
5648 * Let the world know that we're done.
5649 */
5650 mutex_enter(&spa->spa_async_lock);
5651 spa->spa_async_thread = NULL;
5652 cv_broadcast(&spa->spa_async_cv);
5653 mutex_exit(&spa->spa_async_lock);
5654 thread_exit();
5655 }
5656
5657 void
5658 spa_async_suspend(spa_t *spa)
5659 {
5660 mutex_enter(&spa->spa_async_lock);
5661 spa->spa_async_suspended++;
5662 while (spa->spa_async_thread != NULL)
5663 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5664 mutex_exit(&spa->spa_async_lock);
5665 }
5666
5667 void
5668 spa_async_resume(spa_t *spa)
5669 {
5670 mutex_enter(&spa->spa_async_lock);
5671 ASSERT(spa->spa_async_suspended != 0);
5672 spa->spa_async_suspended--;
5673 mutex_exit(&spa->spa_async_lock);
5674 }
5675
5676 static boolean_t
5677 spa_async_tasks_pending(spa_t *spa)
5678 {
5679 uint_t non_config_tasks;
5680 uint_t config_task;
5681 boolean_t config_task_suspended;
5682
5683 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5684 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5685 if (spa->spa_ccw_fail_time == 0) {
5686 config_task_suspended = B_FALSE;
5687 } else {
5688 config_task_suspended =
5689 (gethrtime() - spa->spa_ccw_fail_time) <
5690 (zfs_ccw_retry_interval * NANOSEC);
5691 }
5692
5693 return (non_config_tasks || (config_task && !config_task_suspended));
5694 }
5695
5696 static void
5697 spa_async_dispatch(spa_t *spa)
5698 {
5699 mutex_enter(&spa->spa_async_lock);
5700 if (spa_async_tasks_pending(spa) &&
5701 !spa->spa_async_suspended &&
5702 spa->spa_async_thread == NULL &&
5703 rootdir != NULL)
5704 spa->spa_async_thread = thread_create(NULL, 0,
5705 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5706 mutex_exit(&spa->spa_async_lock);
5707 }
5708
5709 void
5710 spa_async_request(spa_t *spa, int task)
5711 {
5712 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5713 mutex_enter(&spa->spa_async_lock);
5714 spa->spa_async_tasks |= task;
5715 mutex_exit(&spa->spa_async_lock);
5716 }
5717
5718 /*
5719 * ==========================================================================
5720 * SPA syncing routines
5721 * ==========================================================================
5722 */
5723
5724 static int
5725 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5726 {
5727 bpobj_t *bpo = arg;
5728 bpobj_enqueue(bpo, bp, tx);
5729 return (0);
5730 }
5731
5732 static int
5733 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5734 {
5735 zio_t *zio = arg;
5736
5737 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5738 zio->io_flags));
5739 return (0);
5740 }
5741
5742 /*
5743 * Note: this simple function is not inlined to make it easier to dtrace the
5744 * amount of time spent syncing frees.
5745 */
5746 static void
5747 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5748 {
5749 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5750 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5751 VERIFY(zio_wait(zio) == 0);
5752 }
5753
5754 /*
5755 * Note: this simple function is not inlined to make it easier to dtrace the
5756 * amount of time spent syncing deferred frees.
5757 */
5758 static void
5759 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5760 {
5761 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5762 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5763 spa_free_sync_cb, zio, tx), ==, 0);
5764 VERIFY0(zio_wait(zio));
5765 }
5766
5767
5768 static void
5769 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5770 {
5771 char *packed = NULL;
5772 size_t bufsize;
5773 size_t nvsize = 0;
5774 dmu_buf_t *db;
5775
5776 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5777
5778 /*
5779 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5780 * information. This avoids the dbuf_will_dirty() path and
5781 * saves us a pre-read to get data we don't actually care about.
5782 */
5783 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5784 packed = kmem_alloc(bufsize, KM_SLEEP);
5785
5786 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5787 KM_SLEEP) == 0);
5788 bzero(packed + nvsize, bufsize - nvsize);
5789
5790 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5791
5792 kmem_free(packed, bufsize);
5793
5794 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5795 dmu_buf_will_dirty(db, tx);
5796 *(uint64_t *)db->db_data = nvsize;
5797 dmu_buf_rele(db, FTAG);
5798 }
5799
5800 static void
5801 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5802 const char *config, const char *entry)
5803 {
5804 nvlist_t *nvroot;
5805 nvlist_t **list;
5806 int i;
5807
5808 if (!sav->sav_sync)
5809 return;
5810
5811 /*
5812 * Update the MOS nvlist describing the list of available devices.
5813 * spa_validate_aux() will have already made sure this nvlist is
5814 * valid and the vdevs are labeled appropriately.
5815 */
5816 if (sav->sav_object == 0) {
5817 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5818 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5819 sizeof (uint64_t), tx);
5820 VERIFY(zap_update(spa->spa_meta_objset,
5821 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5822 &sav->sav_object, tx) == 0);
5823 }
5824
5825 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5826 if (sav->sav_count == 0) {
5827 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5828 } else {
5829 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5830 for (i = 0; i < sav->sav_count; i++)
5831 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5832 B_FALSE, VDEV_CONFIG_L2CACHE);
5833 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5834 sav->sav_count) == 0);
5835 for (i = 0; i < sav->sav_count; i++)
5836 nvlist_free(list[i]);
5837 kmem_free(list, sav->sav_count * sizeof (void *));
5838 }
5839
5840 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5841 nvlist_free(nvroot);
5842
5843 sav->sav_sync = B_FALSE;
5844 }
5845
5846 static void
5847 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5848 {
5849 nvlist_t *config;
5850
5851 if (list_is_empty(&spa->spa_config_dirty_list))
5852 return;
5853
5854 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5855
5856 config = spa_config_generate(spa, spa->spa_root_vdev,
5857 dmu_tx_get_txg(tx), B_FALSE);
5858
5859 /*
5860 * If we're upgrading the spa version then make sure that
5861 * the config object gets updated with the correct version.
5862 */
5863 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5864 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5865 spa->spa_uberblock.ub_version);
5866
5867 spa_config_exit(spa, SCL_STATE, FTAG);
5868
5869 if (spa->spa_config_syncing)
5870 nvlist_free(spa->spa_config_syncing);
5871 spa->spa_config_syncing = config;
5872
5873 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5874 }
5875
5876 static void
5877 spa_sync_version(void *arg, dmu_tx_t *tx)
5878 {
5879 uint64_t *versionp = arg;
5880 uint64_t version = *versionp;
5881 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5882
5883 /*
5884 * Setting the version is special cased when first creating the pool.
5885 */
5886 ASSERT(tx->tx_txg != TXG_INITIAL);
5887
5888 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5889 ASSERT(version >= spa_version(spa));
5890
5891 spa->spa_uberblock.ub_version = version;
5892 vdev_config_dirty(spa->spa_root_vdev);
5893 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5894 }
5895
5896 /*
5897 * Set zpool properties.
5898 */
5899 static void
5900 spa_sync_props(void *arg, dmu_tx_t *tx)
5901 {
5902 nvlist_t *nvp = arg;
5903 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5904 objset_t *mos = spa->spa_meta_objset;
5905 nvpair_t *elem = NULL;
5906
5907 mutex_enter(&spa->spa_props_lock);
5908
5909 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5910 uint64_t intval;
5911 char *strval, *fname;
5912 zpool_prop_t prop;
5913 const char *propname;
5914 zprop_type_t proptype;
5915 spa_feature_t fid;
5916
5917 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5918 case ZPROP_INVAL:
5919 /*
5920 * We checked this earlier in spa_prop_validate().
5921 */
5922 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5923
5924 fname = strchr(nvpair_name(elem), '@') + 1;
5925 VERIFY0(zfeature_lookup_name(fname, &fid));
5926
5927 spa_feature_enable(spa, fid, tx);
5928 spa_history_log_internal(spa, "set", tx,
5929 "%s=enabled", nvpair_name(elem));
5930 break;
5931
5932 case ZPOOL_PROP_VERSION:
5933 intval = fnvpair_value_uint64(elem);
5934 /*
5935 * The version is synced seperatly before other
5936 * properties and should be correct by now.
5937 */
5938 ASSERT3U(spa_version(spa), >=, intval);
5939 break;
5940
5941 case ZPOOL_PROP_ALTROOT:
5942 /*
5943 * 'altroot' is a non-persistent property. It should
5944 * have been set temporarily at creation or import time.
5945 */
5946 ASSERT(spa->spa_root != NULL);
5947 break;
5948
5949 case ZPOOL_PROP_READONLY:
5950 case ZPOOL_PROP_CACHEFILE:
5951 /*
5952 * 'readonly' and 'cachefile' are also non-persisitent
5953 * properties.
5954 */
5955 break;
5956 case ZPOOL_PROP_COMMENT:
5957 strval = fnvpair_value_string(elem);
5958 if (spa->spa_comment != NULL)
5959 spa_strfree(spa->spa_comment);
5960 spa->spa_comment = spa_strdup(strval);
5961 /*
5962 * We need to dirty the configuration on all the vdevs
5963 * so that their labels get updated. It's unnecessary
5964 * to do this for pool creation since the vdev's
5965 * configuratoin has already been dirtied.
5966 */
5967 if (tx->tx_txg != TXG_INITIAL)
5968 vdev_config_dirty(spa->spa_root_vdev);
5969 spa_history_log_internal(spa, "set", tx,
5970 "%s=%s", nvpair_name(elem), strval);
5971 break;
5972 default:
5973 /*
5974 * Set pool property values in the poolprops mos object.
5975 */
5976 if (spa->spa_pool_props_object == 0) {
5977 spa->spa_pool_props_object =
5978 zap_create_link(mos, DMU_OT_POOL_PROPS,
5979 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5980 tx);
5981 }
5982
5983 /* normalize the property name */
5984 propname = zpool_prop_to_name(prop);
5985 proptype = zpool_prop_get_type(prop);
5986
5987 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5988 ASSERT(proptype == PROP_TYPE_STRING);
5989 strval = fnvpair_value_string(elem);
5990 VERIFY0(zap_update(mos,
5991 spa->spa_pool_props_object, propname,
5992 1, strlen(strval) + 1, strval, tx));
5993 spa_history_log_internal(spa, "set", tx,
5994 "%s=%s", nvpair_name(elem), strval);
5995 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5996 intval = fnvpair_value_uint64(elem);
5997
5998 if (proptype == PROP_TYPE_INDEX) {
5999 const char *unused;
6000 VERIFY0(zpool_prop_index_to_string(
6001 prop, intval, &unused));
6002 }
6003 VERIFY0(zap_update(mos,
6004 spa->spa_pool_props_object, propname,
6005 8, 1, &intval, tx));
6006 spa_history_log_internal(spa, "set", tx,
6007 "%s=%lld", nvpair_name(elem), intval);
6008 } else {
6009 ASSERT(0); /* not allowed */
6010 }
6011
6012 switch (prop) {
6013 case ZPOOL_PROP_DELEGATION:
6014 spa->spa_delegation = intval;
6015 break;
6016 case ZPOOL_PROP_BOOTFS:
6017 spa->spa_bootfs = intval;
6018 break;
6019 case ZPOOL_PROP_FAILUREMODE:
6020 spa->spa_failmode = intval;
6021 break;
6022 case ZPOOL_PROP_AUTOEXPAND:
6023 spa->spa_autoexpand = intval;
6024 if (tx->tx_txg != TXG_INITIAL)
6025 spa_async_request(spa,
6026 SPA_ASYNC_AUTOEXPAND);
6027 break;
6028 case ZPOOL_PROP_DEDUPDITTO:
6029 spa->spa_dedup_ditto = intval;
6030 break;
6031 default:
6032 break;
6033 }
6034 }
6035
6036 }
6037
6038 mutex_exit(&spa->spa_props_lock);
6039 }
6040
6041 /*
6042 * Perform one-time upgrade on-disk changes. spa_version() does not
6043 * reflect the new version this txg, so there must be no changes this
6044 * txg to anything that the upgrade code depends on after it executes.
6045 * Therefore this must be called after dsl_pool_sync() does the sync
6046 * tasks.
6047 */
6048 static void
6049 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6050 {
6051 dsl_pool_t *dp = spa->spa_dsl_pool;
6052
6053 ASSERT(spa->spa_sync_pass == 1);
6054
6055 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6056
6057 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6058 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6059 dsl_pool_create_origin(dp, tx);
6060
6061 /* Keeping the origin open increases spa_minref */
6062 spa->spa_minref += 3;
6063 }
6064
6065 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6066 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6067 dsl_pool_upgrade_clones(dp, tx);
6068 }
6069
6070 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6071 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6072 dsl_pool_upgrade_dir_clones(dp, tx);
6073
6074 /* Keeping the freedir open increases spa_minref */
6075 spa->spa_minref += 3;
6076 }
6077
6078 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6079 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6080 spa_feature_create_zap_objects(spa, tx);
6081 }
6082 rrw_exit(&dp->dp_config_rwlock, FTAG);
6083 }
6084
6085 /*
6086 * Sync the specified transaction group. New blocks may be dirtied as
6087 * part of the process, so we iterate until it converges.
6088 */
6089 void
6090 spa_sync(spa_t *spa, uint64_t txg)
6091 {
6092 dsl_pool_t *dp = spa->spa_dsl_pool;
6093 objset_t *mos = spa->spa_meta_objset;
6094 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6095 vdev_t *rvd = spa->spa_root_vdev;
6096 vdev_t *vd;
6097 dmu_tx_t *tx;
6098 int error;
6099
6100 VERIFY(spa_writeable(spa));
6101
6102 /*
6103 * Lock out configuration changes.
6104 */
6105 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6106
6107 spa->spa_syncing_txg = txg;
6108 spa->spa_sync_pass = 0;
6109
6110 /*
6111 * If there are any pending vdev state changes, convert them
6112 * into config changes that go out with this transaction group.
6113 */
6114 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6115 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6116 /*
6117 * We need the write lock here because, for aux vdevs,
6118 * calling vdev_config_dirty() modifies sav_config.
6119 * This is ugly and will become unnecessary when we
6120 * eliminate the aux vdev wart by integrating all vdevs
6121 * into the root vdev tree.
6122 */
6123 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6124 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6125 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6126 vdev_state_clean(vd);
6127 vdev_config_dirty(vd);
6128 }
6129 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6130 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6131 }
6132 spa_config_exit(spa, SCL_STATE, FTAG);
6133
6134 tx = dmu_tx_create_assigned(dp, txg);
6135
6136 spa->spa_sync_starttime = gethrtime();
6137 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6138 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6139
6140 /*
6141 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6142 * set spa_deflate if we have no raid-z vdevs.
6143 */
6144 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6145 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6146 int i;
6147
6148 for (i = 0; i < rvd->vdev_children; i++) {
6149 vd = rvd->vdev_child[i];
6150 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6151 break;
6152 }
6153 if (i == rvd->vdev_children) {
6154 spa->spa_deflate = TRUE;
6155 VERIFY(0 == zap_add(spa->spa_meta_objset,
6156 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6157 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6158 }
6159 }
6160
6161 /*
6162 * If anything has changed in this txg, or if someone is waiting
6163 * for this txg to sync (eg, spa_vdev_remove()), push the
6164 * deferred frees from the previous txg. If not, leave them
6165 * alone so that we don't generate work on an otherwise idle
6166 * system.
6167 */
6168 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6169 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6170 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6171 ((dsl_scan_active(dp->dp_scan) ||
6172 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6173 spa_sync_deferred_frees(spa, tx);
6174 }
6175
6176 /*
6177 * Iterate to convergence.
6178 */
6179 do {
6180 int pass = ++spa->spa_sync_pass;
6181
6182 spa_sync_config_object(spa, tx);
6183 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6184 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6185 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6186 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6187 spa_errlog_sync(spa, txg);
6188 dsl_pool_sync(dp, txg);
6189
6190 if (pass < zfs_sync_pass_deferred_free) {
6191 spa_sync_frees(spa, free_bpl, tx);
6192 } else {
6193 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6194 &spa->spa_deferred_bpobj, tx);
6195 }
6196
6197 ddt_sync(spa, txg);
6198 dsl_scan_sync(dp, tx);
6199
6200 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6201 vdev_sync(vd, txg);
6202
6203 if (pass == 1)
6204 spa_sync_upgrades(spa, tx);
6205
6206 } while (dmu_objset_is_dirty(mos, txg));
6207
6208 /*
6209 * Rewrite the vdev configuration (which includes the uberblock)
6210 * to commit the transaction group.
6211 *
6212 * If there are no dirty vdevs, we sync the uberblock to a few
6213 * random top-level vdevs that are known to be visible in the
6214 * config cache (see spa_vdev_add() for a complete description).
6215 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6216 */
6217 for (;;) {
6218 /*
6219 * We hold SCL_STATE to prevent vdev open/close/etc.
6220 * while we're attempting to write the vdev labels.
6221 */
6222 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6223
6224 if (list_is_empty(&spa->spa_config_dirty_list)) {
6225 vdev_t *svd[SPA_DVAS_PER_BP];
6226 int svdcount = 0;
6227 int children = rvd->vdev_children;
6228 int c0 = spa_get_random(children);
6229
6230 for (int c = 0; c < children; c++) {
6231 vd = rvd->vdev_child[(c0 + c) % children];
6232 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6233 continue;
6234 svd[svdcount++] = vd;
6235 if (svdcount == SPA_DVAS_PER_BP)
6236 break;
6237 }
6238 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6239 if (error != 0)
6240 error = vdev_config_sync(svd, svdcount, txg,
6241 B_TRUE);
6242 } else {
6243 error = vdev_config_sync(rvd->vdev_child,
6244 rvd->vdev_children, txg, B_FALSE);
6245 if (error != 0)
6246 error = vdev_config_sync(rvd->vdev_child,
6247 rvd->vdev_children, txg, B_TRUE);
6248 }
6249
6250 if (error == 0)
6251 spa->spa_last_synced_guid = rvd->vdev_guid;
6252
6253 spa_config_exit(spa, SCL_STATE, FTAG);
6254
6255 if (error == 0)
6256 break;
6257 zio_suspend(spa, NULL);
6258 zio_resume_wait(spa);
6259 }
6260 dmu_tx_commit(tx);
6261
6262 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6263
6264 /*
6265 * Clear the dirty config list.
6266 */
6267 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6268 vdev_config_clean(vd);
6269
6270 /*
6271 * Now that the new config has synced transactionally,
6272 * let it become visible to the config cache.
6273 */
6274 if (spa->spa_config_syncing != NULL) {
6275 spa_config_set(spa, spa->spa_config_syncing);
6276 spa->spa_config_txg = txg;
6277 spa->spa_config_syncing = NULL;
6278 }
6279
6280 spa->spa_ubsync = spa->spa_uberblock;
6281
6282 dsl_pool_sync_done(dp, txg);
6283
6284 /*
6285 * Update usable space statistics.
6286 */
6287 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6288 vdev_sync_done(vd, txg);
6289
6290 spa_update_dspace(spa);
6291
6292 /*
6293 * It had better be the case that we didn't dirty anything
6294 * since vdev_config_sync().
6295 */
6296 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6297 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6298 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6299
6300 spa->spa_sync_pass = 0;
6301
6302 spa_config_exit(spa, SCL_CONFIG, FTAG);
6303
6304 spa_handle_ignored_writes(spa);
6305
6306 /*
6307 * If any async tasks have been requested, kick them off.
6308 */
6309 spa_async_dispatch(spa);
6310 }
6311
6312 /*
6313 * Sync all pools. We don't want to hold the namespace lock across these
6314 * operations, so we take a reference on the spa_t and drop the lock during the
6315 * sync.
6316 */
6317 void
6318 spa_sync_allpools(void)
6319 {
6320 spa_t *spa = NULL;
6321 mutex_enter(&spa_namespace_lock);
6322 while ((spa = spa_next(spa)) != NULL) {
6323 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6324 !spa_writeable(spa) || spa_suspended(spa))
6325 continue;
6326 spa_open_ref(spa, FTAG);
6327 mutex_exit(&spa_namespace_lock);
6328 txg_wait_synced(spa_get_dsl(spa), 0);
6329 mutex_enter(&spa_namespace_lock);
6330 spa_close(spa, FTAG);
6331 }
6332 mutex_exit(&spa_namespace_lock);
6333 }
6334
6335 /*
6336 * ==========================================================================
6337 * Miscellaneous routines
6338 * ==========================================================================
6339 */
6340
6341 /*
6342 * Remove all pools in the system.
6343 */
6344 void
6345 spa_evict_all(void)
6346 {
6347 spa_t *spa;
6348
6349 /*
6350 * Remove all cached state. All pools should be closed now,
6351 * so every spa in the AVL tree should be unreferenced.
6352 */
6353 mutex_enter(&spa_namespace_lock);
6354 while ((spa = spa_next(NULL)) != NULL) {
6355 /*
6356 * Stop async tasks. The async thread may need to detach
6357 * a device that's been replaced, which requires grabbing
6358 * spa_namespace_lock, so we must drop it here.
6359 */
6360 spa_open_ref(spa, FTAG);
6361 mutex_exit(&spa_namespace_lock);
6362 spa_async_suspend(spa);
6363 mutex_enter(&spa_namespace_lock);
6364 spa_close(spa, FTAG);
6365
6366 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6367 spa_unload(spa);
6368 spa_deactivate(spa);
6369 }
6370 spa_remove(spa);
6371 }
6372 mutex_exit(&spa_namespace_lock);
6373 }
6374
6375 vdev_t *
6376 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6377 {
6378 vdev_t *vd;
6379 int i;
6380
6381 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6382 return (vd);
6383
6384 if (aux) {
6385 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6386 vd = spa->spa_l2cache.sav_vdevs[i];
6387 if (vd->vdev_guid == guid)
6388 return (vd);
6389 }
6390
6391 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6392 vd = spa->spa_spares.sav_vdevs[i];
6393 if (vd->vdev_guid == guid)
6394 return (vd);
6395 }
6396 }
6397
6398 return (NULL);
6399 }
6400
6401 void
6402 spa_upgrade(spa_t *spa, uint64_t version)
6403 {
6404 ASSERT(spa_writeable(spa));
6405
6406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6407
6408 /*
6409 * This should only be called for a non-faulted pool, and since a
6410 * future version would result in an unopenable pool, this shouldn't be
6411 * possible.
6412 */
6413 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6414 ASSERT(version >= spa->spa_uberblock.ub_version);
6415
6416 spa->spa_uberblock.ub_version = version;
6417 vdev_config_dirty(spa->spa_root_vdev);
6418
6419 spa_config_exit(spa, SCL_ALL, FTAG);
6420
6421 txg_wait_synced(spa_get_dsl(spa), 0);
6422 }
6423
6424 boolean_t
6425 spa_has_spare(spa_t *spa, uint64_t guid)
6426 {
6427 int i;
6428 uint64_t spareguid;
6429 spa_aux_vdev_t *sav = &spa->spa_spares;
6430
6431 for (i = 0; i < sav->sav_count; i++)
6432 if (sav->sav_vdevs[i]->vdev_guid == guid)
6433 return (B_TRUE);
6434
6435 for (i = 0; i < sav->sav_npending; i++) {
6436 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6437 &spareguid) == 0 && spareguid == guid)
6438 return (B_TRUE);
6439 }
6440
6441 return (B_FALSE);
6442 }
6443
6444 /*
6445 * Check if a pool has an active shared spare device.
6446 * Note: reference count of an active spare is 2, as a spare and as a replace
6447 */
6448 static boolean_t
6449 spa_has_active_shared_spare(spa_t *spa)
6450 {
6451 int i, refcnt;
6452 uint64_t pool;
6453 spa_aux_vdev_t *sav = &spa->spa_spares;
6454
6455 for (i = 0; i < sav->sav_count; i++) {
6456 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6457 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6458 refcnt > 2)
6459 return (B_TRUE);
6460 }
6461
6462 return (B_FALSE);
6463 }
6464
6465 /*
6466 * Post a sysevent corresponding to the given event. The 'name' must be one of
6467 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6468 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6469 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6470 * or zdb as real changes.
6471 */
6472 void
6473 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6474 {
6475 #ifdef _KERNEL
6476 sysevent_t *ev;
6477 sysevent_attr_list_t *attr = NULL;
6478 sysevent_value_t value;
6479 sysevent_id_t eid;
6480
6481 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6482 SE_SLEEP);
6483
6484 value.value_type = SE_DATA_TYPE_STRING;
6485 value.value.sv_string = spa_name(spa);
6486 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6487 goto done;
6488
6489 value.value_type = SE_DATA_TYPE_UINT64;
6490 value.value.sv_uint64 = spa_guid(spa);
6491 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6492 goto done;
6493
6494 if (vd) {
6495 value.value_type = SE_DATA_TYPE_UINT64;
6496 value.value.sv_uint64 = vd->vdev_guid;
6497 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6498 SE_SLEEP) != 0)
6499 goto done;
6500
6501 if (vd->vdev_path) {
6502 value.value_type = SE_DATA_TYPE_STRING;
6503 value.value.sv_string = vd->vdev_path;
6504 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6505 &value, SE_SLEEP) != 0)
6506 goto done;
6507 }
6508 }
6509
6510 if (sysevent_attach_attributes(ev, attr) != 0)
6511 goto done;
6512 attr = NULL;
6513
6514 (void) log_sysevent(ev, SE_SLEEP, &eid);
6515
6516 done:
6517 if (attr)
6518 sysevent_free_attr(attr);
6519 sysevent_free(ev);
6520 #endif
6521 }