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_ONLINE_PERCENT, /* value is % of online CPUs */
91 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
92 ZTI_MODE_NULL, /* don't create a taskq */
93 ZTI_NMODES
94 } zti_modes_t;
95
96 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
97 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
98 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
99 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
100
101 #define ZTI_N(n) ZTI_P(n, 1)
102 #define ZTI_ONE ZTI_N(1)
103
104 typedef struct zio_taskq_info {
105 zti_modes_t zti_mode;
106 uint_t zti_value;
107 uint_t zti_count;
108 } zio_taskq_info_t;
109
110 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
111 "issue", "issue_high", "intr", "intr_high"
112 };
113
114 /*
115 * This table defines the taskq settings for each ZFS I/O type. When
116 * initializing a pool, we use this table to create an appropriately sized
117 * taskq. Some operations are low volume and therefore have a small, static
118 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
119 * macros. Other operations process a large amount of data; the ZTI_BATCH
120 * macro causes us to create a taskq oriented for throughput. Some operations
121 * are so high frequency and short-lived that the taskq itself can become a a
122 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
123 * additional degree of parallelism specified by the number of threads per-
124 * taskq and the number of taskqs; when dispatching an event in this case, the
125 * particular taskq is chosen at random.
126 *
127 * The different taskq priorities are to handle the different contexts (issue
128 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
129 * need to be handled with minimum delay.
130 */
131 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
132 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
133 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
134 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
135 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
136 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
137 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
138 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
139 };
140
141 static void spa_sync_version(void *arg, dmu_tx_t *tx);
142 static void spa_sync_props(void *arg, dmu_tx_t *tx);
143 static boolean_t spa_has_active_shared_spare(spa_t *spa);
144 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
145 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
146 char **ereport);
147 static void spa_vdev_resilver_done(spa_t *spa);
148
149 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
150 id_t zio_taskq_psrset_bind = PS_NONE;
151 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
152 uint_t zio_taskq_basedc = 80; /* base duty cycle */
153
154 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
155 extern int zfs_sync_pass_deferred_free;
156
157 /*
158 * This (illegal) pool name is used when temporarily importing a spa_t in order
159 * to get the vdev stats associated with the imported devices.
160 */
161 #define TRYIMPORT_NAME "$import"
162
163 /*
164 * ==========================================================================
165 * SPA properties routines
166 * ==========================================================================
167 */
168
169 /*
170 * Add a (source=src, propname=propval) list to an nvlist.
171 */
172 static void
173 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
174 uint64_t intval, zprop_source_t src)
175 {
176 const char *propname = zpool_prop_to_name(prop);
177 nvlist_t *propval;
178
179 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
180 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
181
182 if (strval != NULL)
183 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
184 else
185 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
186
187 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
188 nvlist_free(propval);
189 }
190
191 /*
192 * Get property values from the spa configuration.
193 */
194 static void
195 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
196 {
197 vdev_t *rvd = spa->spa_root_vdev;
198 dsl_pool_t *pool = spa->spa_dsl_pool;
199 uint64_t size;
200 uint64_t alloc;
201 uint64_t space;
202 uint64_t cap, version;
203 zprop_source_t src = ZPROP_SRC_NONE;
204 spa_config_dirent_t *dp;
205
206 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
207
208 if (rvd != NULL) {
209 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
210 size = metaslab_class_get_space(spa_normal_class(spa));
211 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
212 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
213 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
214 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
215 size - alloc, src);
216
217 space = 0;
218 for (int c = 0; c < rvd->vdev_children; c++) {
219 vdev_t *tvd = rvd->vdev_child[c];
220 space += tvd->vdev_max_asize - tvd->vdev_asize;
221 }
222 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
223 src);
224
225 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
226 (spa_mode(spa) == FREAD), src);
227
228 cap = (size == 0) ? 0 : (alloc * 100 / size);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
230
231 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
232 ddt_get_pool_dedup_ratio(spa), src);
233
234 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
235 rvd->vdev_state, src);
236
237 version = spa_version(spa);
238 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
239 src = ZPROP_SRC_DEFAULT;
240 else
241 src = ZPROP_SRC_LOCAL;
242 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
243 }
244
245 if (pool != NULL) {
246 dsl_dir_t *freedir = pool->dp_free_dir;
247
248 /*
249 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
250 * when opening pools before this version freedir will be NULL.
251 */
252 if (freedir != NULL) {
253 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
254 freedir->dd_phys->dd_used_bytes, src);
255 } else {
256 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
257 NULL, 0, src);
258 }
259 }
260
261 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
262
263 if (spa->spa_comment != NULL) {
264 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
265 0, ZPROP_SRC_LOCAL);
266 }
267
268 if (spa->spa_root != NULL)
269 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
270 0, ZPROP_SRC_LOCAL);
271
272 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
273 if (dp->scd_path == NULL) {
274 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
275 "none", 0, ZPROP_SRC_LOCAL);
276 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
277 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
278 dp->scd_path, 0, ZPROP_SRC_LOCAL);
279 }
280 }
281 }
282
283 /*
284 * Get zpool property values.
285 */
286 int
287 spa_prop_get(spa_t *spa, nvlist_t **nvp)
288 {
289 objset_t *mos = spa->spa_meta_objset;
290 zap_cursor_t zc;
291 zap_attribute_t za;
292 int err;
293
294 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
295
296 mutex_enter(&spa->spa_props_lock);
297
298 /*
299 * Get properties from the spa config.
300 */
301 spa_prop_get_config(spa, nvp);
302
303 /* If no pool property object, no more prop to get. */
304 if (mos == NULL || spa->spa_pool_props_object == 0) {
305 mutex_exit(&spa->spa_props_lock);
306 return (0);
307 }
308
309 /*
310 * Get properties from the MOS pool property object.
311 */
312 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
313 (err = zap_cursor_retrieve(&zc, &za)) == 0;
314 zap_cursor_advance(&zc)) {
315 uint64_t intval = 0;
316 char *strval = NULL;
317 zprop_source_t src = ZPROP_SRC_DEFAULT;
318 zpool_prop_t prop;
319
320 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
321 continue;
322
323 switch (za.za_integer_length) {
324 case 8:
325 /* integer property */
326 if (za.za_first_integer !=
327 zpool_prop_default_numeric(prop))
328 src = ZPROP_SRC_LOCAL;
329
330 if (prop == ZPOOL_PROP_BOOTFS) {
331 dsl_pool_t *dp;
332 dsl_dataset_t *ds = NULL;
333
334 dp = spa_get_dsl(spa);
335 dsl_pool_config_enter(dp, FTAG);
336 if (err = dsl_dataset_hold_obj(dp,
337 za.za_first_integer, FTAG, &ds)) {
338 dsl_pool_config_exit(dp, FTAG);
339 break;
340 }
341
342 strval = kmem_alloc(
343 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
344 KM_SLEEP);
345 dsl_dataset_name(ds, strval);
346 dsl_dataset_rele(ds, FTAG);
347 dsl_pool_config_exit(dp, FTAG);
348 } else {
349 strval = NULL;
350 intval = za.za_first_integer;
351 }
352
353 spa_prop_add_list(*nvp, prop, strval, intval, src);
354
355 if (strval != NULL)
356 kmem_free(strval,
357 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
358
359 break;
360
361 case 1:
362 /* string property */
363 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
364 err = zap_lookup(mos, spa->spa_pool_props_object,
365 za.za_name, 1, za.za_num_integers, strval);
366 if (err) {
367 kmem_free(strval, za.za_num_integers);
368 break;
369 }
370 spa_prop_add_list(*nvp, prop, strval, 0, src);
371 kmem_free(strval, za.za_num_integers);
372 break;
373
374 default:
375 break;
376 }
377 }
378 zap_cursor_fini(&zc);
379 mutex_exit(&spa->spa_props_lock);
380 out:
381 if (err && err != ENOENT) {
382 nvlist_free(*nvp);
383 *nvp = NULL;
384 return (err);
385 }
386
387 return (0);
388 }
389
390 /*
391 * Validate the given pool properties nvlist and modify the list
392 * for the property values to be set.
393 */
394 static int
395 spa_prop_validate(spa_t *spa, nvlist_t *props)
396 {
397 nvpair_t *elem;
398 int error = 0, reset_bootfs = 0;
399 uint64_t objnum = 0;
400 boolean_t has_feature = B_FALSE;
401
402 elem = NULL;
403 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
404 uint64_t intval;
405 char *strval, *slash, *check, *fname;
406 const char *propname = nvpair_name(elem);
407 zpool_prop_t prop = zpool_name_to_prop(propname);
408
409 switch (prop) {
410 case ZPROP_INVAL:
411 if (!zpool_prop_feature(propname)) {
412 error = SET_ERROR(EINVAL);
413 break;
414 }
415
416 /*
417 * Sanitize the input.
418 */
419 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
420 error = SET_ERROR(EINVAL);
421 break;
422 }
423
424 if (nvpair_value_uint64(elem, &intval) != 0) {
425 error = SET_ERROR(EINVAL);
426 break;
427 }
428
429 if (intval != 0) {
430 error = SET_ERROR(EINVAL);
431 break;
432 }
433
434 fname = strchr(propname, '@') + 1;
435 if (zfeature_lookup_name(fname, NULL) != 0) {
436 error = SET_ERROR(EINVAL);
437 break;
438 }
439
440 has_feature = B_TRUE;
441 break;
442
443 case ZPOOL_PROP_VERSION:
444 error = nvpair_value_uint64(elem, &intval);
445 if (!error &&
446 (intval < spa_version(spa) ||
447 intval > SPA_VERSION_BEFORE_FEATURES ||
448 has_feature))
449 error = SET_ERROR(EINVAL);
450 break;
451
452 case ZPOOL_PROP_DELEGATION:
453 case ZPOOL_PROP_AUTOREPLACE:
454 case ZPOOL_PROP_LISTSNAPS:
455 case ZPOOL_PROP_AUTOEXPAND:
456 error = nvpair_value_uint64(elem, &intval);
457 if (!error && intval > 1)
458 error = SET_ERROR(EINVAL);
459 break;
460
461 case ZPOOL_PROP_BOOTFS:
462 /*
463 * If the pool version is less than SPA_VERSION_BOOTFS,
464 * or the pool is still being created (version == 0),
465 * the bootfs property cannot be set.
466 */
467 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
468 error = SET_ERROR(ENOTSUP);
469 break;
470 }
471
472 /*
473 * Make sure the vdev config is bootable
474 */
475 if (!vdev_is_bootable(spa->spa_root_vdev)) {
476 error = SET_ERROR(ENOTSUP);
477 break;
478 }
479
480 reset_bootfs = 1;
481
482 error = nvpair_value_string(elem, &strval);
483
484 if (!error) {
485 objset_t *os;
486 uint64_t compress;
487
488 if (strval == NULL || strval[0] == '\0') {
489 objnum = zpool_prop_default_numeric(
490 ZPOOL_PROP_BOOTFS);
491 break;
492 }
493
494 if (error = dmu_objset_hold(strval, FTAG, &os))
495 break;
496
497 /* Must be ZPL and not gzip compressed. */
498
499 if (dmu_objset_type(os) != DMU_OST_ZFS) {
500 error = SET_ERROR(ENOTSUP);
501 } else if ((error =
502 dsl_prop_get_int_ds(dmu_objset_ds(os),
503 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
504 &compress)) == 0 &&
505 !BOOTFS_COMPRESS_VALID(compress)) {
506 error = SET_ERROR(ENOTSUP);
507 } else {
508 objnum = dmu_objset_id(os);
509 }
510 dmu_objset_rele(os, FTAG);
511 }
512 break;
513
514 case ZPOOL_PROP_FAILUREMODE:
515 error = nvpair_value_uint64(elem, &intval);
516 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
517 intval > ZIO_FAILURE_MODE_PANIC))
518 error = SET_ERROR(EINVAL);
519
520 /*
521 * This is a special case which only occurs when
522 * the pool has completely failed. This allows
523 * the user to change the in-core failmode property
524 * without syncing it out to disk (I/Os might
525 * currently be blocked). We do this by returning
526 * EIO to the caller (spa_prop_set) to trick it
527 * into thinking we encountered a property validation
528 * error.
529 */
530 if (!error && spa_suspended(spa)) {
531 spa->spa_failmode = intval;
532 error = SET_ERROR(EIO);
533 }
534 break;
535
536 case ZPOOL_PROP_CACHEFILE:
537 if ((error = nvpair_value_string(elem, &strval)) != 0)
538 break;
539
540 if (strval[0] == '\0')
541 break;
542
543 if (strcmp(strval, "none") == 0)
544 break;
545
546 if (strval[0] != '/') {
547 error = SET_ERROR(EINVAL);
548 break;
549 }
550
551 slash = strrchr(strval, '/');
552 ASSERT(slash != NULL);
553
554 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
555 strcmp(slash, "/..") == 0)
556 error = SET_ERROR(EINVAL);
557 break;
558
559 case ZPOOL_PROP_COMMENT:
560 if ((error = nvpair_value_string(elem, &strval)) != 0)
561 break;
562 for (check = strval; *check != '\0'; check++) {
563 /*
564 * The kernel doesn't have an easy isprint()
565 * check. For this kernel check, we merely
566 * check ASCII apart from DEL. Fix this if
567 * there is an easy-to-use kernel isprint().
568 */
569 if (*check >= 0x7f) {
570 error = SET_ERROR(EINVAL);
571 break;
572 }
573 check++;
574 }
575 if (strlen(strval) > ZPROP_MAX_COMMENT)
576 error = E2BIG;
577 break;
578
579 case ZPOOL_PROP_DEDUPDITTO:
580 if (spa_version(spa) < SPA_VERSION_DEDUP)
581 error = SET_ERROR(ENOTSUP);
582 else
583 error = nvpair_value_uint64(elem, &intval);
584 if (error == 0 &&
585 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
586 error = SET_ERROR(EINVAL);
587 break;
588 }
589
590 if (error)
591 break;
592 }
593
594 if (!error && reset_bootfs) {
595 error = nvlist_remove(props,
596 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
597
598 if (!error) {
599 error = nvlist_add_uint64(props,
600 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
601 }
602 }
603
604 return (error);
605 }
606
607 void
608 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
609 {
610 char *cachefile;
611 spa_config_dirent_t *dp;
612
613 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
614 &cachefile) != 0)
615 return;
616
617 dp = kmem_alloc(sizeof (spa_config_dirent_t),
618 KM_SLEEP);
619
620 if (cachefile[0] == '\0')
621 dp->scd_path = spa_strdup(spa_config_path);
622 else if (strcmp(cachefile, "none") == 0)
623 dp->scd_path = NULL;
624 else
625 dp->scd_path = spa_strdup(cachefile);
626
627 list_insert_head(&spa->spa_config_list, dp);
628 if (need_sync)
629 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
630 }
631
632 int
633 spa_prop_set(spa_t *spa, nvlist_t *nvp)
634 {
635 int error;
636 nvpair_t *elem = NULL;
637 boolean_t need_sync = B_FALSE;
638
639 if ((error = spa_prop_validate(spa, nvp)) != 0)
640 return (error);
641
642 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
643 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
644
645 if (prop == ZPOOL_PROP_CACHEFILE ||
646 prop == ZPOOL_PROP_ALTROOT ||
647 prop == ZPOOL_PROP_READONLY)
648 continue;
649
650 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
651 uint64_t ver;
652
653 if (prop == ZPOOL_PROP_VERSION) {
654 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
655 } else {
656 ASSERT(zpool_prop_feature(nvpair_name(elem)));
657 ver = SPA_VERSION_FEATURES;
658 need_sync = B_TRUE;
659 }
660
661 /* Save time if the version is already set. */
662 if (ver == spa_version(spa))
663 continue;
664
665 /*
666 * In addition to the pool directory object, we might
667 * create the pool properties object, the features for
668 * read object, the features for write object, or the
669 * feature descriptions object.
670 */
671 error = dsl_sync_task(spa->spa_name, NULL,
672 spa_sync_version, &ver, 6);
673 if (error)
674 return (error);
675 continue;
676 }
677
678 need_sync = B_TRUE;
679 break;
680 }
681
682 if (need_sync) {
683 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
684 nvp, 6));
685 }
686
687 return (0);
688 }
689
690 /*
691 * If the bootfs property value is dsobj, clear it.
692 */
693 void
694 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
695 {
696 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
697 VERIFY(zap_remove(spa->spa_meta_objset,
698 spa->spa_pool_props_object,
699 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
700 spa->spa_bootfs = 0;
701 }
702 }
703
704 /*ARGSUSED*/
705 static int
706 spa_change_guid_check(void *arg, dmu_tx_t *tx)
707 {
708 uint64_t *newguid = arg;
709 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
710 vdev_t *rvd = spa->spa_root_vdev;
711 uint64_t vdev_state;
712
713 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
714 vdev_state = rvd->vdev_state;
715 spa_config_exit(spa, SCL_STATE, FTAG);
716
717 if (vdev_state != VDEV_STATE_HEALTHY)
718 return (SET_ERROR(ENXIO));
719
720 ASSERT3U(spa_guid(spa), !=, *newguid);
721
722 return (0);
723 }
724
725 static void
726 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
727 {
728 uint64_t *newguid = arg;
729 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730 uint64_t oldguid;
731 vdev_t *rvd = spa->spa_root_vdev;
732
733 oldguid = spa_guid(spa);
734
735 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
736 rvd->vdev_guid = *newguid;
737 rvd->vdev_guid_sum += (*newguid - oldguid);
738 vdev_config_dirty(rvd);
739 spa_config_exit(spa, SCL_STATE, FTAG);
740
741 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
742 oldguid, *newguid);
743 }
744
745 /*
746 * Change the GUID for the pool. This is done so that we can later
747 * re-import a pool built from a clone of our own vdevs. We will modify
748 * the root vdev's guid, our own pool guid, and then mark all of our
749 * vdevs dirty. Note that we must make sure that all our vdevs are
750 * online when we do this, or else any vdevs that weren't present
751 * would be orphaned from our pool. We are also going to issue a
752 * sysevent to update any watchers.
753 */
754 int
755 spa_change_guid(spa_t *spa)
756 {
757 int error;
758 uint64_t guid;
759
760 mutex_enter(&spa_namespace_lock);
761 guid = spa_generate_guid(NULL);
762
763 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
764 spa_change_guid_sync, &guid, 5);
765
766 if (error == 0) {
767 spa_config_sync(spa, B_FALSE, B_TRUE);
768 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
769 }
770
771 mutex_exit(&spa_namespace_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 for (uint_t i = 0; i < count; i++) {
844 taskq_t *tq;
845
846 switch (mode) {
847 case ZTI_MODE_FIXED:
848 ASSERT3U(value, >=, 1);
849 value = MAX(value, 1);
850 break;
851
852 case ZTI_MODE_BATCH:
853 batch = B_TRUE;
854 flags |= TASKQ_THREADS_CPU_PCT;
855 value = zio_taskq_batch_pct;
856 break;
857
858 case ZTI_MODE_ONLINE_PERCENT:
859 flags |= TASKQ_THREADS_CPU_PCT;
860 break;
861
862 default:
863 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
864 "spa_activate()",
865 zio_type_name[t], zio_taskq_types[q], mode, value);
866 break;
867 }
868
869 if (count > 1) {
870 (void) snprintf(name, sizeof (name), "%s_%s_%u",
871 zio_type_name[t], zio_taskq_types[q], i);
872 } else {
873 (void) snprintf(name, sizeof (name), "%s_%s",
874 zio_type_name[t], zio_taskq_types[q]);
875 }
876
877 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
878 if (batch)
879 flags |= TASKQ_DC_BATCH;
880
881 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
882 spa->spa_proc, zio_taskq_basedc, flags);
883 } else {
884 tq = taskq_create_proc(name, value, maxclsyspri, 50,
885 INT_MAX, spa->spa_proc, flags);
886 }
887
888 tqs->stqs_taskq[i] = tq;
889 }
890 }
891
892 static void
893 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
894 {
895 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
896
897 if (tqs->stqs_taskq == NULL) {
898 ASSERT0(tqs->stqs_count);
899 return;
900 }
901
902 for (uint_t i = 0; i < tqs->stqs_count; i++) {
903 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
904 taskq_destroy(tqs->stqs_taskq[i]);
905 }
906
907 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
908 tqs->stqs_taskq = NULL;
909 }
910
911 /*
912 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
913 * Note that a type may have multiple discrete taskqs to avoid lock contention
914 * on the taskq itself. In that case we choose which taskq at random by using
915 * the low bits of gethrtime().
916 */
917 void
918 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
919 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
920 {
921 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
922 taskq_t *tq;
923
924 ASSERT3P(tqs->stqs_taskq, !=, NULL);
925 ASSERT3U(tqs->stqs_count, !=, 0);
926
927 if (tqs->stqs_count == 1) {
928 tq = tqs->stqs_taskq[0];
929 } else {
930 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
931 }
932
933 taskq_dispatch_ent(tq, func, arg, flags, ent);
934 }
935
936 static void
937 spa_create_zio_taskqs(spa_t *spa)
938 {
939 for (int t = 0; t < ZIO_TYPES; t++) {
940 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
941 spa_taskqs_init(spa, t, q);
942 }
943 }
944 }
945
946 #ifdef _KERNEL
947 static void
948 spa_thread(void *arg)
949 {
950 callb_cpr_t cprinfo;
951
952 spa_t *spa = arg;
953 user_t *pu = PTOU(curproc);
954
955 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
956 spa->spa_name);
957
958 ASSERT(curproc != &p0);
959 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
960 "zpool-%s", spa->spa_name);
961 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
962
963 /* bind this thread to the requested psrset */
964 if (zio_taskq_psrset_bind != PS_NONE) {
965 pool_lock();
966 mutex_enter(&cpu_lock);
967 mutex_enter(&pidlock);
968 mutex_enter(&curproc->p_lock);
969
970 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
971 0, NULL, NULL) == 0) {
972 curthread->t_bind_pset = zio_taskq_psrset_bind;
973 } else {
974 cmn_err(CE_WARN,
975 "Couldn't bind process for zfs pool \"%s\" to "
976 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
977 }
978
979 mutex_exit(&curproc->p_lock);
980 mutex_exit(&pidlock);
981 mutex_exit(&cpu_lock);
982 pool_unlock();
983 }
984
985 if (zio_taskq_sysdc) {
986 sysdc_thread_enter(curthread, 100, 0);
987 }
988
989 spa->spa_proc = curproc;
990 spa->spa_did = curthread->t_did;
991
992 spa_create_zio_taskqs(spa);
993
994 mutex_enter(&spa->spa_proc_lock);
995 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
996
997 spa->spa_proc_state = SPA_PROC_ACTIVE;
998 cv_broadcast(&spa->spa_proc_cv);
999
1000 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1001 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1002 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1003 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1004
1005 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1006 spa->spa_proc_state = SPA_PROC_GONE;
1007 spa->spa_proc = &p0;
1008 cv_broadcast(&spa->spa_proc_cv);
1009 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1010
1011 mutex_enter(&curproc->p_lock);
1012 lwp_exit();
1013 }
1014 #endif
1015
1016 /*
1017 * Activate an uninitialized pool.
1018 */
1019 static void
1020 spa_activate(spa_t *spa, int mode)
1021 {
1022 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1023
1024 spa->spa_state = POOL_STATE_ACTIVE;
1025 spa->spa_mode = mode;
1026
1027 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1028 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1029
1030 /* Try to create a covering process */
1031 mutex_enter(&spa->spa_proc_lock);
1032 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1033 ASSERT(spa->spa_proc == &p0);
1034 spa->spa_did = 0;
1035
1036 /* Only create a process if we're going to be around a while. */
1037 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1038 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1039 NULL, 0) == 0) {
1040 spa->spa_proc_state = SPA_PROC_CREATED;
1041 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1042 cv_wait(&spa->spa_proc_cv,
1043 &spa->spa_proc_lock);
1044 }
1045 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1046 ASSERT(spa->spa_proc != &p0);
1047 ASSERT(spa->spa_did != 0);
1048 } else {
1049 #ifdef _KERNEL
1050 cmn_err(CE_WARN,
1051 "Couldn't create process for zfs pool \"%s\"\n",
1052 spa->spa_name);
1053 #endif
1054 }
1055 }
1056 mutex_exit(&spa->spa_proc_lock);
1057
1058 /* If we didn't create a process, we need to create our taskqs. */
1059 if (spa->spa_proc == &p0) {
1060 spa_create_zio_taskqs(spa);
1061 }
1062
1063 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1064 offsetof(vdev_t, vdev_config_dirty_node));
1065 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1066 offsetof(vdev_t, vdev_state_dirty_node));
1067
1068 txg_list_create(&spa->spa_vdev_txg_list,
1069 offsetof(struct vdev, vdev_txg_node));
1070
1071 avl_create(&spa->spa_errlist_scrub,
1072 spa_error_entry_compare, sizeof (spa_error_entry_t),
1073 offsetof(spa_error_entry_t, se_avl));
1074 avl_create(&spa->spa_errlist_last,
1075 spa_error_entry_compare, sizeof (spa_error_entry_t),
1076 offsetof(spa_error_entry_t, se_avl));
1077 }
1078
1079 /*
1080 * Opposite of spa_activate().
1081 */
1082 static void
1083 spa_deactivate(spa_t *spa)
1084 {
1085 ASSERT(spa->spa_sync_on == B_FALSE);
1086 ASSERT(spa->spa_dsl_pool == NULL);
1087 ASSERT(spa->spa_root_vdev == NULL);
1088 ASSERT(spa->spa_async_zio_root == NULL);
1089 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1090
1091 txg_list_destroy(&spa->spa_vdev_txg_list);
1092
1093 list_destroy(&spa->spa_config_dirty_list);
1094 list_destroy(&spa->spa_state_dirty_list);
1095
1096 for (int t = 0; t < ZIO_TYPES; t++) {
1097 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1098 spa_taskqs_fini(spa, t, q);
1099 }
1100 }
1101
1102 metaslab_class_destroy(spa->spa_normal_class);
1103 spa->spa_normal_class = NULL;
1104
1105 metaslab_class_destroy(spa->spa_log_class);
1106 spa->spa_log_class = NULL;
1107
1108 /*
1109 * If this was part of an import or the open otherwise failed, we may
1110 * still have errors left in the queues. Empty them just in case.
1111 */
1112 spa_errlog_drain(spa);
1113
1114 avl_destroy(&spa->spa_errlist_scrub);
1115 avl_destroy(&spa->spa_errlist_last);
1116
1117 spa->spa_state = POOL_STATE_UNINITIALIZED;
1118
1119 mutex_enter(&spa->spa_proc_lock);
1120 if (spa->spa_proc_state != SPA_PROC_NONE) {
1121 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1122 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1123 cv_broadcast(&spa->spa_proc_cv);
1124 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1125 ASSERT(spa->spa_proc != &p0);
1126 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1127 }
1128 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1129 spa->spa_proc_state = SPA_PROC_NONE;
1130 }
1131 ASSERT(spa->spa_proc == &p0);
1132 mutex_exit(&spa->spa_proc_lock);
1133
1134 /*
1135 * We want to make sure spa_thread() has actually exited the ZFS
1136 * module, so that the module can't be unloaded out from underneath
1137 * it.
1138 */
1139 if (spa->spa_did != 0) {
1140 thread_join(spa->spa_did);
1141 spa->spa_did = 0;
1142 }
1143 }
1144
1145 /*
1146 * Verify a pool configuration, and construct the vdev tree appropriately. This
1147 * will create all the necessary vdevs in the appropriate layout, with each vdev
1148 * in the CLOSED state. This will prep the pool before open/creation/import.
1149 * All vdev validation is done by the vdev_alloc() routine.
1150 */
1151 static int
1152 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1153 uint_t id, int atype)
1154 {
1155 nvlist_t **child;
1156 uint_t children;
1157 int error;
1158
1159 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1160 return (error);
1161
1162 if ((*vdp)->vdev_ops->vdev_op_leaf)
1163 return (0);
1164
1165 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1166 &child, &children);
1167
1168 if (error == ENOENT)
1169 return (0);
1170
1171 if (error) {
1172 vdev_free(*vdp);
1173 *vdp = NULL;
1174 return (SET_ERROR(EINVAL));
1175 }
1176
1177 for (int c = 0; c < children; c++) {
1178 vdev_t *vd;
1179 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1180 atype)) != 0) {
1181 vdev_free(*vdp);
1182 *vdp = NULL;
1183 return (error);
1184 }
1185 }
1186
1187 ASSERT(*vdp != NULL);
1188
1189 return (0);
1190 }
1191
1192 /*
1193 * Opposite of spa_load().
1194 */
1195 static void
1196 spa_unload(spa_t *spa)
1197 {
1198 int i;
1199
1200 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1201
1202 /*
1203 * Stop async tasks.
1204 */
1205 spa_async_suspend(spa);
1206
1207 /*
1208 * Stop syncing.
1209 */
1210 if (spa->spa_sync_on) {
1211 txg_sync_stop(spa->spa_dsl_pool);
1212 spa->spa_sync_on = B_FALSE;
1213 }
1214
1215 /*
1216 * Wait for any outstanding async I/O to complete.
1217 */
1218 if (spa->spa_async_zio_root != NULL) {
1219 (void) zio_wait(spa->spa_async_zio_root);
1220 spa->spa_async_zio_root = NULL;
1221 }
1222
1223 bpobj_close(&spa->spa_deferred_bpobj);
1224
1225 /*
1226 * Close the dsl pool.
1227 */
1228 if (spa->spa_dsl_pool) {
1229 dsl_pool_close(spa->spa_dsl_pool);
1230 spa->spa_dsl_pool = NULL;
1231 spa->spa_meta_objset = NULL;
1232 }
1233
1234 ddt_unload(spa);
1235
1236 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1237
1238 /*
1239 * Drop and purge level 2 cache
1240 */
1241 spa_l2cache_drop(spa);
1242
1243 /*
1244 * Close all vdevs.
1245 */
1246 if (spa->spa_root_vdev)
1247 vdev_free(spa->spa_root_vdev);
1248 ASSERT(spa->spa_root_vdev == NULL);
1249
1250 for (i = 0; i < spa->spa_spares.sav_count; i++)
1251 vdev_free(spa->spa_spares.sav_vdevs[i]);
1252 if (spa->spa_spares.sav_vdevs) {
1253 kmem_free(spa->spa_spares.sav_vdevs,
1254 spa->spa_spares.sav_count * sizeof (void *));
1255 spa->spa_spares.sav_vdevs = NULL;
1256 }
1257 if (spa->spa_spares.sav_config) {
1258 nvlist_free(spa->spa_spares.sav_config);
1259 spa->spa_spares.sav_config = NULL;
1260 }
1261 spa->spa_spares.sav_count = 0;
1262
1263 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1264 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1265 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1266 }
1267 if (spa->spa_l2cache.sav_vdevs) {
1268 kmem_free(spa->spa_l2cache.sav_vdevs,
1269 spa->spa_l2cache.sav_count * sizeof (void *));
1270 spa->spa_l2cache.sav_vdevs = NULL;
1271 }
1272 if (spa->spa_l2cache.sav_config) {
1273 nvlist_free(spa->spa_l2cache.sav_config);
1274 spa->spa_l2cache.sav_config = NULL;
1275 }
1276 spa->spa_l2cache.sav_count = 0;
1277
1278 spa->spa_async_suspended = 0;
1279
1280 if (spa->spa_comment != NULL) {
1281 spa_strfree(spa->spa_comment);
1282 spa->spa_comment = NULL;
1283 }
1284
1285 spa_config_exit(spa, SCL_ALL, FTAG);
1286 }
1287
1288 /*
1289 * Load (or re-load) the current list of vdevs describing the active spares for
1290 * this pool. When this is called, we have some form of basic information in
1291 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1292 * then re-generate a more complete list including status information.
1293 */
1294 static void
1295 spa_load_spares(spa_t *spa)
1296 {
1297 nvlist_t **spares;
1298 uint_t nspares;
1299 int i;
1300 vdev_t *vd, *tvd;
1301
1302 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1303
1304 /*
1305 * First, close and free any existing spare vdevs.
1306 */
1307 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1308 vd = spa->spa_spares.sav_vdevs[i];
1309
1310 /* Undo the call to spa_activate() below */
1311 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1312 B_FALSE)) != NULL && tvd->vdev_isspare)
1313 spa_spare_remove(tvd);
1314 vdev_close(vd);
1315 vdev_free(vd);
1316 }
1317
1318 if (spa->spa_spares.sav_vdevs)
1319 kmem_free(spa->spa_spares.sav_vdevs,
1320 spa->spa_spares.sav_count * sizeof (void *));
1321
1322 if (spa->spa_spares.sav_config == NULL)
1323 nspares = 0;
1324 else
1325 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1326 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1327
1328 spa->spa_spares.sav_count = (int)nspares;
1329 spa->spa_spares.sav_vdevs = NULL;
1330
1331 if (nspares == 0)
1332 return;
1333
1334 /*
1335 * Construct the array of vdevs, opening them to get status in the
1336 * process. For each spare, there is potentially two different vdev_t
1337 * structures associated with it: one in the list of spares (used only
1338 * for basic validation purposes) and one in the active vdev
1339 * configuration (if it's spared in). During this phase we open and
1340 * validate each vdev on the spare list. If the vdev also exists in the
1341 * active configuration, then we also mark this vdev as an active spare.
1342 */
1343 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1344 KM_SLEEP);
1345 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1346 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1347 VDEV_ALLOC_SPARE) == 0);
1348 ASSERT(vd != NULL);
1349
1350 spa->spa_spares.sav_vdevs[i] = vd;
1351
1352 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1353 B_FALSE)) != NULL) {
1354 if (!tvd->vdev_isspare)
1355 spa_spare_add(tvd);
1356
1357 /*
1358 * We only mark the spare active if we were successfully
1359 * able to load the vdev. Otherwise, importing a pool
1360 * with a bad active spare would result in strange
1361 * behavior, because multiple pool would think the spare
1362 * is actively in use.
1363 *
1364 * There is a vulnerability here to an equally bizarre
1365 * circumstance, where a dead active spare is later
1366 * brought back to life (onlined or otherwise). Given
1367 * the rarity of this scenario, and the extra complexity
1368 * it adds, we ignore the possibility.
1369 */
1370 if (!vdev_is_dead(tvd))
1371 spa_spare_activate(tvd);
1372 }
1373
1374 vd->vdev_top = vd;
1375 vd->vdev_aux = &spa->spa_spares;
1376
1377 if (vdev_open(vd) != 0)
1378 continue;
1379
1380 if (vdev_validate_aux(vd) == 0)
1381 spa_spare_add(vd);
1382 }
1383
1384 /*
1385 * Recompute the stashed list of spares, with status information
1386 * this time.
1387 */
1388 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1389 DATA_TYPE_NVLIST_ARRAY) == 0);
1390
1391 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1392 KM_SLEEP);
1393 for (i = 0; i < spa->spa_spares.sav_count; i++)
1394 spares[i] = vdev_config_generate(spa,
1395 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1396 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1397 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1398 for (i = 0; i < spa->spa_spares.sav_count; i++)
1399 nvlist_free(spares[i]);
1400 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1401 }
1402
1403 /*
1404 * Load (or re-load) the current list of vdevs describing the active l2cache for
1405 * this pool. When this is called, we have some form of basic information in
1406 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1407 * then re-generate a more complete list including status information.
1408 * Devices which are already active have their details maintained, and are
1409 * not re-opened.
1410 */
1411 static void
1412 spa_load_l2cache(spa_t *spa)
1413 {
1414 nvlist_t **l2cache;
1415 uint_t nl2cache;
1416 int i, j, oldnvdevs;
1417 uint64_t guid;
1418 vdev_t *vd, **oldvdevs, **newvdevs;
1419 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1420
1421 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1422
1423 if (sav->sav_config != NULL) {
1424 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1425 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1426 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1427 } else {
1428 nl2cache = 0;
1429 newvdevs = NULL;
1430 }
1431
1432 oldvdevs = sav->sav_vdevs;
1433 oldnvdevs = sav->sav_count;
1434 sav->sav_vdevs = NULL;
1435 sav->sav_count = 0;
1436
1437 /*
1438 * Process new nvlist of vdevs.
1439 */
1440 for (i = 0; i < nl2cache; i++) {
1441 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1442 &guid) == 0);
1443
1444 newvdevs[i] = NULL;
1445 for (j = 0; j < oldnvdevs; j++) {
1446 vd = oldvdevs[j];
1447 if (vd != NULL && guid == vd->vdev_guid) {
1448 /*
1449 * Retain previous vdev for add/remove ops.
1450 */
1451 newvdevs[i] = vd;
1452 oldvdevs[j] = NULL;
1453 break;
1454 }
1455 }
1456
1457 if (newvdevs[i] == NULL) {
1458 /*
1459 * Create new vdev
1460 */
1461 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1462 VDEV_ALLOC_L2CACHE) == 0);
1463 ASSERT(vd != NULL);
1464 newvdevs[i] = vd;
1465
1466 /*
1467 * Commit this vdev as an l2cache device,
1468 * even if it fails to open.
1469 */
1470 spa_l2cache_add(vd);
1471
1472 vd->vdev_top = vd;
1473 vd->vdev_aux = sav;
1474
1475 spa_l2cache_activate(vd);
1476
1477 if (vdev_open(vd) != 0)
1478 continue;
1479
1480 (void) vdev_validate_aux(vd);
1481
1482 if (!vdev_is_dead(vd)) {
1483 boolean_t persist = B_FALSE;
1484
1485 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
1486 /*
1487 * Only allow to do the L2ARC rebuild
1488 * when not doing a spa try-load.
1489 */
1490 (void) nvlist_lookup_boolean_value(
1491 l2cache[i],
1492 ZPOOL_CONFIG_L2CACHE_PERSISTENT,
1493 &persist);
1494 }
1495 l2arc_add_vdev(spa, vd, persist);
1496 }
1497 }
1498 }
1499
1500 /*
1501 * Purge vdevs that were dropped
1502 */
1503 for (i = 0; i < oldnvdevs; i++) {
1504 uint64_t pool;
1505
1506 vd = oldvdevs[i];
1507 if (vd != NULL) {
1508 ASSERT(vd->vdev_isl2cache);
1509
1510 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1511 pool != 0ULL && l2arc_vdev_present(vd))
1512 l2arc_remove_vdev(vd);
1513 vdev_clear_stats(vd);
1514 vdev_free(vd);
1515 }
1516 }
1517
1518 if (oldvdevs)
1519 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1520
1521 if (sav->sav_config == NULL)
1522 goto out;
1523
1524 sav->sav_vdevs = newvdevs;
1525 sav->sav_count = (int)nl2cache;
1526
1527 /*
1528 * Recompute the stashed list of l2cache devices, with status
1529 * information this time.
1530 */
1531 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1532 DATA_TYPE_NVLIST_ARRAY) == 0);
1533
1534 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1535 for (i = 0; i < sav->sav_count; i++)
1536 l2cache[i] = vdev_config_generate(spa,
1537 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1538 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1539 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1540 out:
1541 for (i = 0; i < sav->sav_count; i++)
1542 nvlist_free(l2cache[i]);
1543 if (sav->sav_count)
1544 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1545 }
1546
1547 static int
1548 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1549 {
1550 dmu_buf_t *db;
1551 char *packed = NULL;
1552 size_t nvsize = 0;
1553 int error;
1554 *value = NULL;
1555
1556 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1557 nvsize = *(uint64_t *)db->db_data;
1558 dmu_buf_rele(db, FTAG);
1559
1560 packed = kmem_alloc(nvsize, KM_SLEEP);
1561 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1562 DMU_READ_PREFETCH);
1563 if (error == 0)
1564 error = nvlist_unpack(packed, nvsize, value, 0);
1565 kmem_free(packed, nvsize);
1566
1567 return (error);
1568 }
1569
1570 /*
1571 * Checks to see if the given vdev could not be opened, in which case we post a
1572 * sysevent to notify the autoreplace code that the device has been removed.
1573 */
1574 static void
1575 spa_check_removed(vdev_t *vd)
1576 {
1577 for (int c = 0; c < vd->vdev_children; c++)
1578 spa_check_removed(vd->vdev_child[c]);
1579
1580 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1581 !vd->vdev_ishole) {
1582 zfs_post_autoreplace(vd->vdev_spa, vd);
1583 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1584 }
1585 }
1586
1587 /*
1588 * Validate the current config against the MOS config
1589 */
1590 static boolean_t
1591 spa_config_valid(spa_t *spa, nvlist_t *config)
1592 {
1593 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1594 nvlist_t *nv;
1595
1596 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1597
1598 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1599 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1600
1601 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1602
1603 /*
1604 * If we're doing a normal import, then build up any additional
1605 * diagnostic information about missing devices in this config.
1606 * We'll pass this up to the user for further processing.
1607 */
1608 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1609 nvlist_t **child, *nv;
1610 uint64_t idx = 0;
1611
1612 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1613 KM_SLEEP);
1614 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1615
1616 for (int c = 0; c < rvd->vdev_children; c++) {
1617 vdev_t *tvd = rvd->vdev_child[c];
1618 vdev_t *mtvd = mrvd->vdev_child[c];
1619
1620 if (tvd->vdev_ops == &vdev_missing_ops &&
1621 mtvd->vdev_ops != &vdev_missing_ops &&
1622 mtvd->vdev_islog)
1623 child[idx++] = vdev_config_generate(spa, mtvd,
1624 B_FALSE, 0);
1625 }
1626
1627 if (idx) {
1628 VERIFY(nvlist_add_nvlist_array(nv,
1629 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1630 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1631 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1632
1633 for (int i = 0; i < idx; i++)
1634 nvlist_free(child[i]);
1635 }
1636 nvlist_free(nv);
1637 kmem_free(child, rvd->vdev_children * sizeof (char **));
1638 }
1639
1640 /*
1641 * Compare the root vdev tree with the information we have
1642 * from the MOS config (mrvd). Check each top-level vdev
1643 * with the corresponding MOS config top-level (mtvd).
1644 */
1645 for (int c = 0; c < rvd->vdev_children; c++) {
1646 vdev_t *tvd = rvd->vdev_child[c];
1647 vdev_t *mtvd = mrvd->vdev_child[c];
1648
1649 /*
1650 * Resolve any "missing" vdevs in the current configuration.
1651 * If we find that the MOS config has more accurate information
1652 * about the top-level vdev then use that vdev instead.
1653 */
1654 if (tvd->vdev_ops == &vdev_missing_ops &&
1655 mtvd->vdev_ops != &vdev_missing_ops) {
1656
1657 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1658 continue;
1659
1660 /*
1661 * Device specific actions.
1662 */
1663 if (mtvd->vdev_islog) {
1664 spa_set_log_state(spa, SPA_LOG_CLEAR);
1665 } else {
1666 /*
1667 * XXX - once we have 'readonly' pool
1668 * support we should be able to handle
1669 * missing data devices by transitioning
1670 * the pool to readonly.
1671 */
1672 continue;
1673 }
1674
1675 /*
1676 * Swap the missing vdev with the data we were
1677 * able to obtain from the MOS config.
1678 */
1679 vdev_remove_child(rvd, tvd);
1680 vdev_remove_child(mrvd, mtvd);
1681
1682 vdev_add_child(rvd, mtvd);
1683 vdev_add_child(mrvd, tvd);
1684
1685 spa_config_exit(spa, SCL_ALL, FTAG);
1686 vdev_load(mtvd);
1687 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1688
1689 vdev_reopen(rvd);
1690 } else if (mtvd->vdev_islog) {
1691 /*
1692 * Load the slog device's state from the MOS config
1693 * since it's possible that the label does not
1694 * contain the most up-to-date information.
1695 */
1696 vdev_load_log_state(tvd, mtvd);
1697 vdev_reopen(tvd);
1698 }
1699 }
1700 vdev_free(mrvd);
1701 spa_config_exit(spa, SCL_ALL, FTAG);
1702
1703 /*
1704 * Ensure we were able to validate the config.
1705 */
1706 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1707 }
1708
1709 /*
1710 * Check for missing log devices
1711 */
1712 static boolean_t
1713 spa_check_logs(spa_t *spa)
1714 {
1715 boolean_t rv = B_FALSE;
1716
1717 switch (spa->spa_log_state) {
1718 case SPA_LOG_MISSING:
1719 /* need to recheck in case slog has been restored */
1720 case SPA_LOG_UNKNOWN:
1721 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1722 NULL, DS_FIND_CHILDREN) != 0);
1723 if (rv)
1724 spa_set_log_state(spa, SPA_LOG_MISSING);
1725 break;
1726 }
1727 return (rv);
1728 }
1729
1730 static boolean_t
1731 spa_passivate_log(spa_t *spa)
1732 {
1733 vdev_t *rvd = spa->spa_root_vdev;
1734 boolean_t slog_found = B_FALSE;
1735
1736 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1737
1738 if (!spa_has_slogs(spa))
1739 return (B_FALSE);
1740
1741 for (int c = 0; c < rvd->vdev_children; c++) {
1742 vdev_t *tvd = rvd->vdev_child[c];
1743 metaslab_group_t *mg = tvd->vdev_mg;
1744
1745 if (tvd->vdev_islog) {
1746 metaslab_group_passivate(mg);
1747 slog_found = B_TRUE;
1748 }
1749 }
1750
1751 return (slog_found);
1752 }
1753
1754 static void
1755 spa_activate_log(spa_t *spa)
1756 {
1757 vdev_t *rvd = spa->spa_root_vdev;
1758
1759 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1760
1761 for (int c = 0; c < rvd->vdev_children; c++) {
1762 vdev_t *tvd = rvd->vdev_child[c];
1763 metaslab_group_t *mg = tvd->vdev_mg;
1764
1765 if (tvd->vdev_islog)
1766 metaslab_group_activate(mg);
1767 }
1768 }
1769
1770 int
1771 spa_offline_log(spa_t *spa)
1772 {
1773 int error;
1774
1775 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1776 NULL, DS_FIND_CHILDREN);
1777 if (error == 0) {
1778 /*
1779 * We successfully offlined the log device, sync out the
1780 * current txg so that the "stubby" block can be removed
1781 * by zil_sync().
1782 */
1783 txg_wait_synced(spa->spa_dsl_pool, 0);
1784 }
1785 return (error);
1786 }
1787
1788 static void
1789 spa_aux_check_removed(spa_aux_vdev_t *sav)
1790 {
1791 for (int i = 0; i < sav->sav_count; i++)
1792 spa_check_removed(sav->sav_vdevs[i]);
1793 }
1794
1795 void
1796 spa_claim_notify(zio_t *zio)
1797 {
1798 spa_t *spa = zio->io_spa;
1799
1800 if (zio->io_error)
1801 return;
1802
1803 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1804 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1805 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1806 mutex_exit(&spa->spa_props_lock);
1807 }
1808
1809 typedef struct spa_load_error {
1810 uint64_t sle_meta_count;
1811 uint64_t sle_data_count;
1812 } spa_load_error_t;
1813
1814 static void
1815 spa_load_verify_done(zio_t *zio)
1816 {
1817 blkptr_t *bp = zio->io_bp;
1818 spa_load_error_t *sle = zio->io_private;
1819 dmu_object_type_t type = BP_GET_TYPE(bp);
1820 int error = zio->io_error;
1821
1822 if (error) {
1823 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1824 type != DMU_OT_INTENT_LOG)
1825 atomic_add_64(&sle->sle_meta_count, 1);
1826 else
1827 atomic_add_64(&sle->sle_data_count, 1);
1828 }
1829 zio_data_buf_free(zio->io_data, zio->io_size);
1830 }
1831
1832 /*ARGSUSED*/
1833 static int
1834 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1835 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1836 {
1837 if (bp != NULL) {
1838 zio_t *rio = arg;
1839 size_t size = BP_GET_PSIZE(bp);
1840 void *data = zio_data_buf_alloc(size);
1841
1842 zio_nowait(zio_read(rio, spa, bp, data, size,
1843 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1844 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1845 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1846 }
1847 return (0);
1848 }
1849
1850 static int
1851 spa_load_verify(spa_t *spa)
1852 {
1853 zio_t *rio;
1854 spa_load_error_t sle = { 0 };
1855 zpool_rewind_policy_t policy;
1856 boolean_t verify_ok = B_FALSE;
1857 int error;
1858
1859 zpool_get_rewind_policy(spa->spa_config, &policy);
1860
1861 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1862 return (0);
1863
1864 rio = zio_root(spa, NULL, &sle,
1865 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1866
1867 error = traverse_pool(spa, spa->spa_verify_min_txg,
1868 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1869
1870 (void) zio_wait(rio);
1871
1872 spa->spa_load_meta_errors = sle.sle_meta_count;
1873 spa->spa_load_data_errors = sle.sle_data_count;
1874
1875 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1876 sle.sle_data_count <= policy.zrp_maxdata) {
1877 int64_t loss = 0;
1878
1879 verify_ok = B_TRUE;
1880 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1881 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1882
1883 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1884 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1885 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1886 VERIFY(nvlist_add_int64(spa->spa_load_info,
1887 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1888 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1889 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1890 } else {
1891 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1892 }
1893
1894 if (error) {
1895 if (error != ENXIO && error != EIO)
1896 error = SET_ERROR(EIO);
1897 return (error);
1898 }
1899
1900 return (verify_ok ? 0 : EIO);
1901 }
1902
1903 /*
1904 * Find a value in the pool props object.
1905 */
1906 static void
1907 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1908 {
1909 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1910 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1911 }
1912
1913 /*
1914 * Find a value in the pool directory object.
1915 */
1916 static int
1917 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1918 {
1919 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1920 name, sizeof (uint64_t), 1, val));
1921 }
1922
1923 static int
1924 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1925 {
1926 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1927 return (err);
1928 }
1929
1930 /*
1931 * Fix up config after a partly-completed split. This is done with the
1932 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1933 * pool have that entry in their config, but only the splitting one contains
1934 * a list of all the guids of the vdevs that are being split off.
1935 *
1936 * This function determines what to do with that list: either rejoin
1937 * all the disks to the pool, or complete the splitting process. To attempt
1938 * the rejoin, each disk that is offlined is marked online again, and
1939 * we do a reopen() call. If the vdev label for every disk that was
1940 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1941 * then we call vdev_split() on each disk, and complete the split.
1942 *
1943 * Otherwise we leave the config alone, with all the vdevs in place in
1944 * the original pool.
1945 */
1946 static void
1947 spa_try_repair(spa_t *spa, nvlist_t *config)
1948 {
1949 uint_t extracted;
1950 uint64_t *glist;
1951 uint_t i, gcount;
1952 nvlist_t *nvl;
1953 vdev_t **vd;
1954 boolean_t attempt_reopen;
1955
1956 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1957 return;
1958
1959 /* check that the config is complete */
1960 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1961 &glist, &gcount) != 0)
1962 return;
1963
1964 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1965
1966 /* attempt to online all the vdevs & validate */
1967 attempt_reopen = B_TRUE;
1968 for (i = 0; i < gcount; i++) {
1969 if (glist[i] == 0) /* vdev is hole */
1970 continue;
1971
1972 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1973 if (vd[i] == NULL) {
1974 /*
1975 * Don't bother attempting to reopen the disks;
1976 * just do the split.
1977 */
1978 attempt_reopen = B_FALSE;
1979 } else {
1980 /* attempt to re-online it */
1981 vd[i]->vdev_offline = B_FALSE;
1982 }
1983 }
1984
1985 if (attempt_reopen) {
1986 vdev_reopen(spa->spa_root_vdev);
1987
1988 /* check each device to see what state it's in */
1989 for (extracted = 0, i = 0; i < gcount; i++) {
1990 if (vd[i] != NULL &&
1991 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1992 break;
1993 ++extracted;
1994 }
1995 }
1996
1997 /*
1998 * If every disk has been moved to the new pool, or if we never
1999 * even attempted to look at them, then we split them off for
2000 * good.
2001 */
2002 if (!attempt_reopen || gcount == extracted) {
2003 for (i = 0; i < gcount; i++)
2004 if (vd[i] != NULL)
2005 vdev_split(vd[i]);
2006 vdev_reopen(spa->spa_root_vdev);
2007 }
2008
2009 kmem_free(vd, gcount * sizeof (vdev_t *));
2010 }
2011
2012 static int
2013 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2014 boolean_t mosconfig)
2015 {
2016 nvlist_t *config = spa->spa_config;
2017 char *ereport = FM_EREPORT_ZFS_POOL;
2018 char *comment;
2019 int error;
2020 uint64_t pool_guid;
2021 nvlist_t *nvl;
2022
2023 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2024 return (SET_ERROR(EINVAL));
2025
2026 ASSERT(spa->spa_comment == NULL);
2027 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2028 spa->spa_comment = spa_strdup(comment);
2029
2030 /*
2031 * Versioning wasn't explicitly added to the label until later, so if
2032 * it's not present treat it as the initial version.
2033 */
2034 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2035 &spa->spa_ubsync.ub_version) != 0)
2036 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2037
2038 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2039 &spa->spa_config_txg);
2040
2041 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2042 spa_guid_exists(pool_guid, 0)) {
2043 error = SET_ERROR(EEXIST);
2044 } else {
2045 spa->spa_config_guid = pool_guid;
2046
2047 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2048 &nvl) == 0) {
2049 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2050 KM_SLEEP) == 0);
2051 }
2052
2053 nvlist_free(spa->spa_load_info);
2054 spa->spa_load_info = fnvlist_alloc();
2055
2056 gethrestime(&spa->spa_loaded_ts);
2057 error = spa_load_impl(spa, pool_guid, config, state, type,
2058 mosconfig, &ereport);
2059 }
2060
2061 spa->spa_minref = refcount_count(&spa->spa_refcount);
2062 if (error) {
2063 if (error != EEXIST) {
2064 spa->spa_loaded_ts.tv_sec = 0;
2065 spa->spa_loaded_ts.tv_nsec = 0;
2066 }
2067 if (error != EBADF) {
2068 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2069 }
2070 }
2071 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2072 spa->spa_ena = 0;
2073
2074 return (error);
2075 }
2076
2077 /*
2078 * Load an existing storage pool, using the pool's builtin spa_config as a
2079 * source of configuration information.
2080 */
2081 static int
2082 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2083 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2084 char **ereport)
2085 {
2086 int error = 0;
2087 nvlist_t *nvroot = NULL;
2088 nvlist_t *label;
2089 vdev_t *rvd;
2090 uberblock_t *ub = &spa->spa_uberblock;
2091 uint64_t children, config_cache_txg = spa->spa_config_txg;
2092 int orig_mode = spa->spa_mode;
2093 int parse;
2094 uint64_t obj;
2095 boolean_t missing_feat_write = B_FALSE;
2096
2097 /*
2098 * If this is an untrusted config, access the pool in read-only mode.
2099 * This prevents things like resilvering recently removed devices.
2100 */
2101 if (!mosconfig)
2102 spa->spa_mode = FREAD;
2103
2104 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2105
2106 spa->spa_load_state = state;
2107
2108 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2109 return (SET_ERROR(EINVAL));
2110
2111 parse = (type == SPA_IMPORT_EXISTING ?
2112 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2113
2114 /*
2115 * Create "The Godfather" zio to hold all async IOs
2116 */
2117 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2118 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2119
2120 /*
2121 * Parse the configuration into a vdev tree. We explicitly set the
2122 * value that will be returned by spa_version() since parsing the
2123 * configuration requires knowing the version number.
2124 */
2125 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2126 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2127 spa_config_exit(spa, SCL_ALL, FTAG);
2128
2129 if (error != 0)
2130 return (error);
2131
2132 ASSERT(spa->spa_root_vdev == rvd);
2133
2134 if (type != SPA_IMPORT_ASSEMBLE) {
2135 ASSERT(spa_guid(spa) == pool_guid);
2136 }
2137
2138 /*
2139 * Try to open all vdevs, loading each label in the process.
2140 */
2141 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2142 error = vdev_open(rvd);
2143 spa_config_exit(spa, SCL_ALL, FTAG);
2144 if (error != 0)
2145 return (error);
2146
2147 /*
2148 * We need to validate the vdev labels against the configuration that
2149 * we have in hand, which is dependent on the setting of mosconfig. If
2150 * mosconfig is true then we're validating the vdev labels based on
2151 * that config. Otherwise, we're validating against the cached config
2152 * (zpool.cache) that was read when we loaded the zfs module, and then
2153 * later we will recursively call spa_load() and validate against
2154 * the vdev config.
2155 *
2156 * If we're assembling a new pool that's been split off from an
2157 * existing pool, the labels haven't yet been updated so we skip
2158 * validation for now.
2159 */
2160 if (type != SPA_IMPORT_ASSEMBLE) {
2161 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2162 error = vdev_validate(rvd, mosconfig);
2163 spa_config_exit(spa, SCL_ALL, FTAG);
2164
2165 if (error != 0)
2166 return (error);
2167
2168 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2169 return (SET_ERROR(ENXIO));
2170 }
2171
2172 /*
2173 * Find the best uberblock.
2174 */
2175 vdev_uberblock_load(rvd, ub, &label);
2176
2177 /*
2178 * If we weren't able to find a single valid uberblock, return failure.
2179 */
2180 if (ub->ub_txg == 0) {
2181 nvlist_free(label);
2182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2183 }
2184
2185 /*
2186 * If the pool has an unsupported version we can't open it.
2187 */
2188 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2189 nvlist_free(label);
2190 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2191 }
2192
2193 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2194 nvlist_t *features;
2195
2196 /*
2197 * If we weren't able to find what's necessary for reading the
2198 * MOS in the label, return failure.
2199 */
2200 if (label == NULL || nvlist_lookup_nvlist(label,
2201 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2202 nvlist_free(label);
2203 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2204 ENXIO));
2205 }
2206
2207 /*
2208 * Update our in-core representation with the definitive values
2209 * from the label.
2210 */
2211 nvlist_free(spa->spa_label_features);
2212 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2213 }
2214
2215 nvlist_free(label);
2216
2217 /*
2218 * Look through entries in the label nvlist's features_for_read. If
2219 * there is a feature listed there which we don't understand then we
2220 * cannot open a pool.
2221 */
2222 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2223 nvlist_t *unsup_feat;
2224
2225 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2226 0);
2227
2228 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2229 NULL); nvp != NULL;
2230 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2231 if (!zfeature_is_supported(nvpair_name(nvp))) {
2232 VERIFY(nvlist_add_string(unsup_feat,
2233 nvpair_name(nvp), "") == 0);
2234 }
2235 }
2236
2237 if (!nvlist_empty(unsup_feat)) {
2238 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2239 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2240 nvlist_free(unsup_feat);
2241 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2242 ENOTSUP));
2243 }
2244
2245 nvlist_free(unsup_feat);
2246 }
2247
2248 /*
2249 * If the vdev guid sum doesn't match the uberblock, we have an
2250 * incomplete configuration. We first check to see if the pool
2251 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2252 * If it is, defer the vdev_guid_sum check till later so we
2253 * can handle missing vdevs.
2254 */
2255 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2256 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2257 rvd->vdev_guid_sum != ub->ub_guid_sum)
2258 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2259
2260 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2261 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2262 spa_try_repair(spa, config);
2263 spa_config_exit(spa, SCL_ALL, FTAG);
2264 nvlist_free(spa->spa_config_splitting);
2265 spa->spa_config_splitting = NULL;
2266 }
2267
2268 /*
2269 * Initialize internal SPA structures.
2270 */
2271 spa->spa_state = POOL_STATE_ACTIVE;
2272 spa->spa_ubsync = spa->spa_uberblock;
2273 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2274 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2275 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2276 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2277 spa->spa_claim_max_txg = spa->spa_first_txg;
2278 spa->spa_prev_software_version = ub->ub_software_version;
2279
2280 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2281 if (error)
2282 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2283 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2284
2285 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2286 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2287
2288 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2289 boolean_t missing_feat_read = B_FALSE;
2290 nvlist_t *unsup_feat, *enabled_feat;
2291
2292 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2293 &spa->spa_feat_for_read_obj) != 0) {
2294 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2295 }
2296
2297 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2298 &spa->spa_feat_for_write_obj) != 0) {
2299 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2300 }
2301
2302 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2303 &spa->spa_feat_desc_obj) != 0) {
2304 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2305 }
2306
2307 enabled_feat = fnvlist_alloc();
2308 unsup_feat = fnvlist_alloc();
2309
2310 if (!feature_is_supported(spa->spa_meta_objset,
2311 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2312 unsup_feat, enabled_feat))
2313 missing_feat_read = B_TRUE;
2314
2315 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2316 if (!feature_is_supported(spa->spa_meta_objset,
2317 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2318 unsup_feat, enabled_feat)) {
2319 missing_feat_write = B_TRUE;
2320 }
2321 }
2322
2323 fnvlist_add_nvlist(spa->spa_load_info,
2324 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2325
2326 if (!nvlist_empty(unsup_feat)) {
2327 fnvlist_add_nvlist(spa->spa_load_info,
2328 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2329 }
2330
2331 fnvlist_free(enabled_feat);
2332 fnvlist_free(unsup_feat);
2333
2334 if (!missing_feat_read) {
2335 fnvlist_add_boolean(spa->spa_load_info,
2336 ZPOOL_CONFIG_CAN_RDONLY);
2337 }
2338
2339 /*
2340 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2341 * twofold: to determine whether the pool is available for
2342 * import in read-write mode and (if it is not) whether the
2343 * pool is available for import in read-only mode. If the pool
2344 * is available for import in read-write mode, it is displayed
2345 * as available in userland; if it is not available for import
2346 * in read-only mode, it is displayed as unavailable in
2347 * userland. If the pool is available for import in read-only
2348 * mode but not read-write mode, it is displayed as unavailable
2349 * in userland with a special note that the pool is actually
2350 * available for open in read-only mode.
2351 *
2352 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2353 * missing a feature for write, we must first determine whether
2354 * the pool can be opened read-only before returning to
2355 * userland in order to know whether to display the
2356 * abovementioned note.
2357 */
2358 if (missing_feat_read || (missing_feat_write &&
2359 spa_writeable(spa))) {
2360 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2361 ENOTSUP));
2362 }
2363 }
2364
2365 spa->spa_is_initializing = B_TRUE;
2366 error = dsl_pool_open(spa->spa_dsl_pool);
2367 spa->spa_is_initializing = B_FALSE;
2368 if (error != 0)
2369 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370
2371 if (!mosconfig) {
2372 uint64_t hostid;
2373 nvlist_t *policy = NULL, *nvconfig;
2374
2375 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2376 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2377
2378 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2379 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2380 char *hostname;
2381 unsigned long myhostid = 0;
2382
2383 VERIFY(nvlist_lookup_string(nvconfig,
2384 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2385
2386 #ifdef _KERNEL
2387 myhostid = zone_get_hostid(NULL);
2388 #else /* _KERNEL */
2389 /*
2390 * We're emulating the system's hostid in userland, so
2391 * we can't use zone_get_hostid().
2392 */
2393 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2394 #endif /* _KERNEL */
2395 if (hostid != 0 && myhostid != 0 &&
2396 hostid != myhostid) {
2397 nvlist_free(nvconfig);
2398 cmn_err(CE_WARN, "pool '%s' could not be "
2399 "loaded as it was last accessed by "
2400 "another system (host: %s hostid: 0x%lx). "
2401 "See: http://illumos.org/msg/ZFS-8000-EY",
2402 spa_name(spa), hostname,
2403 (unsigned long)hostid);
2404 return (SET_ERROR(EBADF));
2405 }
2406 }
2407 if (nvlist_lookup_nvlist(spa->spa_config,
2408 ZPOOL_REWIND_POLICY, &policy) == 0)
2409 VERIFY(nvlist_add_nvlist(nvconfig,
2410 ZPOOL_REWIND_POLICY, policy) == 0);
2411
2412 spa_config_set(spa, nvconfig);
2413 spa_unload(spa);
2414 spa_deactivate(spa);
2415 spa_activate(spa, orig_mode);
2416
2417 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2418 }
2419
2420 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2423 if (error != 0)
2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2425
2426 /*
2427 * Load the bit that tells us to use the new accounting function
2428 * (raid-z deflation). If we have an older pool, this will not
2429 * be present.
2430 */
2431 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2432 if (error != 0 && error != ENOENT)
2433 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2434
2435 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2436 &spa->spa_creation_version);
2437 if (error != 0 && error != ENOENT)
2438 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2439
2440 /*
2441 * Load the persistent error log. If we have an older pool, this will
2442 * not be present.
2443 */
2444 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2445 if (error != 0 && error != ENOENT)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447
2448 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2449 &spa->spa_errlog_scrub);
2450 if (error != 0 && error != ENOENT)
2451 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452
2453 /*
2454 * Load the history object. If we have an older pool, this
2455 * will not be present.
2456 */
2457 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2458 if (error != 0 && error != ENOENT)
2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2460
2461 /*
2462 * If we're assembling the pool from the split-off vdevs of
2463 * an existing pool, we don't want to attach the spares & cache
2464 * devices.
2465 */
2466
2467 /*
2468 * Load any hot spares for this pool.
2469 */
2470 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2471 if (error != 0 && error != ENOENT)
2472 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2473 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2474 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2475 if (load_nvlist(spa, spa->spa_spares.sav_object,
2476 &spa->spa_spares.sav_config) != 0)
2477 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478
2479 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2480 spa_load_spares(spa);
2481 spa_config_exit(spa, SCL_ALL, FTAG);
2482 } else if (error == 0) {
2483 spa->spa_spares.sav_sync = B_TRUE;
2484 }
2485
2486 /*
2487 * Load any level 2 ARC devices for this pool.
2488 */
2489 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2490 &spa->spa_l2cache.sav_object);
2491 if (error != 0 && error != ENOENT)
2492 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2493 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2494 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2495 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2496 &spa->spa_l2cache.sav_config) != 0)
2497 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2498
2499 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2500 spa_load_l2cache(spa);
2501 spa_config_exit(spa, SCL_ALL, FTAG);
2502 } else if (error == 0) {
2503 spa->spa_l2cache.sav_sync = B_TRUE;
2504 }
2505
2506 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2507
2508 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2509 if (error && error != ENOENT)
2510 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2511
2512 if (error == 0) {
2513 uint64_t autoreplace;
2514
2515 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2516 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2517 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2518 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2519 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2520 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2521 &spa->spa_dedup_ditto);
2522
2523 spa->spa_autoreplace = (autoreplace != 0);
2524 }
2525
2526 /*
2527 * If the 'autoreplace' property is set, then post a resource notifying
2528 * the ZFS DE that it should not issue any faults for unopenable
2529 * devices. We also iterate over the vdevs, and post a sysevent for any
2530 * unopenable vdevs so that the normal autoreplace handler can take
2531 * over.
2532 */
2533 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2534 spa_check_removed(spa->spa_root_vdev);
2535 /*
2536 * For the import case, this is done in spa_import(), because
2537 * at this point we're using the spare definitions from
2538 * the MOS config, not necessarily from the userland config.
2539 */
2540 if (state != SPA_LOAD_IMPORT) {
2541 spa_aux_check_removed(&spa->spa_spares);
2542 spa_aux_check_removed(&spa->spa_l2cache);
2543 }
2544 }
2545
2546 /*
2547 * Load the vdev state for all toplevel vdevs.
2548 */
2549 vdev_load(rvd);
2550
2551 /*
2552 * Propagate the leaf DTLs we just loaded all the way up the tree.
2553 */
2554 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2555 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2556 spa_config_exit(spa, SCL_ALL, FTAG);
2557
2558 /*
2559 * Load the DDTs (dedup tables).
2560 */
2561 error = ddt_load(spa);
2562 if (error != 0)
2563 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2564
2565 spa_update_dspace(spa);
2566
2567 /*
2568 * Validate the config, using the MOS config to fill in any
2569 * information which might be missing. If we fail to validate
2570 * the config then declare the pool unfit for use. If we're
2571 * assembling a pool from a split, the log is not transferred
2572 * over.
2573 */
2574 if (type != SPA_IMPORT_ASSEMBLE) {
2575 nvlist_t *nvconfig;
2576
2577 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2578 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2579
2580 if (!spa_config_valid(spa, nvconfig)) {
2581 nvlist_free(nvconfig);
2582 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2583 ENXIO));
2584 }
2585 nvlist_free(nvconfig);
2586
2587 /*
2588 * Now that we've validated the config, check the state of the
2589 * root vdev. If it can't be opened, it indicates one or
2590 * more toplevel vdevs are faulted.
2591 */
2592 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2593 return (SET_ERROR(ENXIO));
2594
2595 if (spa_check_logs(spa)) {
2596 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2597 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2598 }
2599 }
2600
2601 if (missing_feat_write) {
2602 ASSERT(state == SPA_LOAD_TRYIMPORT);
2603
2604 /*
2605 * At this point, we know that we can open the pool in
2606 * read-only mode but not read-write mode. We now have enough
2607 * information and can return to userland.
2608 */
2609 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2610 }
2611
2612 /*
2613 * We've successfully opened the pool, verify that we're ready
2614 * to start pushing transactions.
2615 */
2616 if (state != SPA_LOAD_TRYIMPORT) {
2617 if (error = spa_load_verify(spa))
2618 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2619 error));
2620 }
2621
2622 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2623 spa->spa_load_max_txg == UINT64_MAX)) {
2624 dmu_tx_t *tx;
2625 int need_update = B_FALSE;
2626
2627 ASSERT(state != SPA_LOAD_TRYIMPORT);
2628
2629 /*
2630 * Claim log blocks that haven't been committed yet.
2631 * This must all happen in a single txg.
2632 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2633 * invoked from zil_claim_log_block()'s i/o done callback.
2634 * Price of rollback is that we abandon the log.
2635 */
2636 spa->spa_claiming = B_TRUE;
2637
2638 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2639 spa_first_txg(spa));
2640 (void) dmu_objset_find(spa_name(spa),
2641 zil_claim, tx, DS_FIND_CHILDREN);
2642 dmu_tx_commit(tx);
2643
2644 spa->spa_claiming = B_FALSE;
2645
2646 spa_set_log_state(spa, SPA_LOG_GOOD);
2647 spa->spa_sync_on = B_TRUE;
2648 txg_sync_start(spa->spa_dsl_pool);
2649
2650 /*
2651 * Wait for all claims to sync. We sync up to the highest
2652 * claimed log block birth time so that claimed log blocks
2653 * don't appear to be from the future. spa_claim_max_txg
2654 * will have been set for us by either zil_check_log_chain()
2655 * (invoked from spa_check_logs()) or zil_claim() above.
2656 */
2657 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2658
2659 /*
2660 * If the config cache is stale, or we have uninitialized
2661 * metaslabs (see spa_vdev_add()), then update the config.
2662 *
2663 * If this is a verbatim import, trust the current
2664 * in-core spa_config and update the disk labels.
2665 */
2666 if (config_cache_txg != spa->spa_config_txg ||
2667 state == SPA_LOAD_IMPORT ||
2668 state == SPA_LOAD_RECOVER ||
2669 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2670 need_update = B_TRUE;
2671
2672 for (int c = 0; c < rvd->vdev_children; c++)
2673 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2674 need_update = B_TRUE;
2675
2676 /*
2677 * Update the config cache asychronously in case we're the
2678 * root pool, in which case the config cache isn't writable yet.
2679 */
2680 if (need_update)
2681 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2682
2683 /*
2684 * Check all DTLs to see if anything needs resilvering.
2685 */
2686 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2687 vdev_resilver_needed(rvd, NULL, NULL))
2688 spa_async_request(spa, SPA_ASYNC_RESILVER);
2689
2690 /*
2691 * Log the fact that we booted up (so that we can detect if
2692 * we rebooted in the middle of an operation).
2693 */
2694 spa_history_log_version(spa, "open");
2695
2696 /*
2697 * Delete any inconsistent datasets.
2698 */
2699 (void) dmu_objset_find(spa_name(spa),
2700 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2701
2702 /*
2703 * Clean up any stale temporary dataset userrefs.
2704 */
2705 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2706 }
2707
2708 return (0);
2709 }
2710
2711 static int
2712 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2713 {
2714 int mode = spa->spa_mode;
2715
2716 spa_unload(spa);
2717 spa_deactivate(spa);
2718
2719 spa->spa_load_max_txg--;
2720
2721 spa_activate(spa, mode);
2722 spa_async_suspend(spa);
2723
2724 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2725 }
2726
2727 /*
2728 * If spa_load() fails this function will try loading prior txg's. If
2729 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2730 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2731 * function will not rewind the pool and will return the same error as
2732 * spa_load().
2733 */
2734 static int
2735 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2736 uint64_t max_request, int rewind_flags)
2737 {
2738 nvlist_t *loadinfo = NULL;
2739 nvlist_t *config = NULL;
2740 int load_error, rewind_error;
2741 uint64_t safe_rewind_txg;
2742 uint64_t min_txg;
2743
2744 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2745 spa->spa_load_max_txg = spa->spa_load_txg;
2746 spa_set_log_state(spa, SPA_LOG_CLEAR);
2747 } else {
2748 spa->spa_load_max_txg = max_request;
2749 }
2750
2751 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2752 mosconfig);
2753 if (load_error == 0)
2754 return (0);
2755
2756 if (spa->spa_root_vdev != NULL)
2757 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2758
2759 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2760 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2761
2762 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2763 nvlist_free(config);
2764 return (load_error);
2765 }
2766
2767 if (state == SPA_LOAD_RECOVER) {
2768 /* Price of rolling back is discarding txgs, including log */
2769 spa_set_log_state(spa, SPA_LOG_CLEAR);
2770 } else {
2771 /*
2772 * If we aren't rolling back save the load info from our first
2773 * import attempt so that we can restore it after attempting
2774 * to rewind.
2775 */
2776 loadinfo = spa->spa_load_info;
2777 spa->spa_load_info = fnvlist_alloc();
2778 }
2779
2780 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2781 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2782 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2783 TXG_INITIAL : safe_rewind_txg;
2784
2785 /*
2786 * Continue as long as we're finding errors, we're still within
2787 * the acceptable rewind range, and we're still finding uberblocks
2788 */
2789 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2790 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2791 if (spa->spa_load_max_txg < safe_rewind_txg)
2792 spa->spa_extreme_rewind = B_TRUE;
2793 rewind_error = spa_load_retry(spa, state, mosconfig);
2794 }
2795
2796 spa->spa_extreme_rewind = B_FALSE;
2797 spa->spa_load_max_txg = UINT64_MAX;
2798
2799 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2800 spa_config_set(spa, config);
2801
2802 if (state == SPA_LOAD_RECOVER) {
2803 ASSERT3P(loadinfo, ==, NULL);
2804 return (rewind_error);
2805 } else {
2806 /* Store the rewind info as part of the initial load info */
2807 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2808 spa->spa_load_info);
2809
2810 /* Restore the initial load info */
2811 fnvlist_free(spa->spa_load_info);
2812 spa->spa_load_info = loadinfo;
2813
2814 return (load_error);
2815 }
2816 }
2817
2818 /*
2819 * Pool Open/Import
2820 *
2821 * The import case is identical to an open except that the configuration is sent
2822 * down from userland, instead of grabbed from the configuration cache. For the
2823 * case of an open, the pool configuration will exist in the
2824 * POOL_STATE_UNINITIALIZED state.
2825 *
2826 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2827 * the same time open the pool, without having to keep around the spa_t in some
2828 * ambiguous state.
2829 */
2830 static int
2831 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2832 nvlist_t **config)
2833 {
2834 spa_t *spa;
2835 spa_load_state_t state = SPA_LOAD_OPEN;
2836 int error;
2837 int locked = B_FALSE;
2838
2839 *spapp = NULL;
2840
2841 /*
2842 * As disgusting as this is, we need to support recursive calls to this
2843 * function because dsl_dir_open() is called during spa_load(), and ends
2844 * up calling spa_open() again. The real fix is to figure out how to
2845 * avoid dsl_dir_open() calling this in the first place.
2846 */
2847 if (mutex_owner(&spa_namespace_lock) != curthread) {
2848 mutex_enter(&spa_namespace_lock);
2849 locked = B_TRUE;
2850 }
2851
2852 if ((spa = spa_lookup(pool)) == NULL) {
2853 if (locked)
2854 mutex_exit(&spa_namespace_lock);
2855 return (SET_ERROR(ENOENT));
2856 }
2857
2858 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2859 zpool_rewind_policy_t policy;
2860
2861 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2862 &policy);
2863 if (policy.zrp_request & ZPOOL_DO_REWIND)
2864 state = SPA_LOAD_RECOVER;
2865
2866 spa_activate(spa, spa_mode_global);
2867
2868 if (state != SPA_LOAD_RECOVER)
2869 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2870
2871 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2872 policy.zrp_request);
2873
2874 if (error == EBADF) {
2875 /*
2876 * If vdev_validate() returns failure (indicated by
2877 * EBADF), it indicates that one of the vdevs indicates
2878 * that the pool has been exported or destroyed. If
2879 * this is the case, the config cache is out of sync and
2880 * we should remove the pool from the namespace.
2881 */
2882 spa_unload(spa);
2883 spa_deactivate(spa);
2884 spa_config_sync(spa, B_TRUE, B_TRUE);
2885 spa_remove(spa);
2886 if (locked)
2887 mutex_exit(&spa_namespace_lock);
2888 return (SET_ERROR(ENOENT));
2889 }
2890
2891 if (error) {
2892 /*
2893 * We can't open the pool, but we still have useful
2894 * information: the state of each vdev after the
2895 * attempted vdev_open(). Return this to the user.
2896 */
2897 if (config != NULL && spa->spa_config) {
2898 VERIFY(nvlist_dup(spa->spa_config, config,
2899 KM_SLEEP) == 0);
2900 VERIFY(nvlist_add_nvlist(*config,
2901 ZPOOL_CONFIG_LOAD_INFO,
2902 spa->spa_load_info) == 0);
2903 }
2904 spa_unload(spa);
2905 spa_deactivate(spa);
2906 spa->spa_last_open_failed = error;
2907 if (locked)
2908 mutex_exit(&spa_namespace_lock);
2909 *spapp = NULL;
2910 return (error);
2911 }
2912 }
2913
2914 spa_open_ref(spa, tag);
2915
2916 if (config != NULL)
2917 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2918
2919 /*
2920 * If we've recovered the pool, pass back any information we
2921 * gathered while doing the load.
2922 */
2923 if (state == SPA_LOAD_RECOVER) {
2924 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2925 spa->spa_load_info) == 0);
2926 }
2927
2928 if (locked) {
2929 spa->spa_last_open_failed = 0;
2930 spa->spa_last_ubsync_txg = 0;
2931 spa->spa_load_txg = 0;
2932 mutex_exit(&spa_namespace_lock);
2933 }
2934
2935 *spapp = spa;
2936
2937 return (0);
2938 }
2939
2940 int
2941 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2942 nvlist_t **config)
2943 {
2944 return (spa_open_common(name, spapp, tag, policy, config));
2945 }
2946
2947 int
2948 spa_open(const char *name, spa_t **spapp, void *tag)
2949 {
2950 return (spa_open_common(name, spapp, tag, NULL, NULL));
2951 }
2952
2953 /*
2954 * Lookup the given spa_t, incrementing the inject count in the process,
2955 * preventing it from being exported or destroyed.
2956 */
2957 spa_t *
2958 spa_inject_addref(char *name)
2959 {
2960 spa_t *spa;
2961
2962 mutex_enter(&spa_namespace_lock);
2963 if ((spa = spa_lookup(name)) == NULL) {
2964 mutex_exit(&spa_namespace_lock);
2965 return (NULL);
2966 }
2967 spa->spa_inject_ref++;
2968 mutex_exit(&spa_namespace_lock);
2969
2970 return (spa);
2971 }
2972
2973 void
2974 spa_inject_delref(spa_t *spa)
2975 {
2976 mutex_enter(&spa_namespace_lock);
2977 spa->spa_inject_ref--;
2978 mutex_exit(&spa_namespace_lock);
2979 }
2980
2981 /*
2982 * Add spares device information to the nvlist.
2983 */
2984 static void
2985 spa_add_spares(spa_t *spa, nvlist_t *config)
2986 {
2987 nvlist_t **spares;
2988 uint_t i, nspares;
2989 nvlist_t *nvroot;
2990 uint64_t guid;
2991 vdev_stat_t *vs;
2992 uint_t vsc;
2993 uint64_t pool;
2994
2995 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2996
2997 if (spa->spa_spares.sav_count == 0)
2998 return;
2999
3000 VERIFY(nvlist_lookup_nvlist(config,
3001 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3002 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3003 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3004 if (nspares != 0) {
3005 VERIFY(nvlist_add_nvlist_array(nvroot,
3006 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3007 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3008 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3009
3010 /*
3011 * Go through and find any spares which have since been
3012 * repurposed as an active spare. If this is the case, update
3013 * their status appropriately.
3014 */
3015 for (i = 0; i < nspares; i++) {
3016 VERIFY(nvlist_lookup_uint64(spares[i],
3017 ZPOOL_CONFIG_GUID, &guid) == 0);
3018 if (spa_spare_exists(guid, &pool, NULL) &&
3019 pool != 0ULL) {
3020 VERIFY(nvlist_lookup_uint64_array(
3021 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3022 (uint64_t **)&vs, &vsc) == 0);
3023 vs->vs_state = VDEV_STATE_CANT_OPEN;
3024 vs->vs_aux = VDEV_AUX_SPARED;
3025 }
3026 }
3027 }
3028 }
3029
3030 /*
3031 * Add l2cache device information to the nvlist, including vdev stats.
3032 */
3033 static void
3034 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3035 {
3036 nvlist_t **l2cache;
3037 uint_t i, j, nl2cache;
3038 nvlist_t *nvroot;
3039 uint64_t guid;
3040 vdev_t *vd;
3041 vdev_stat_t *vs;
3042 uint_t vsc;
3043
3044 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3045
3046 if (spa->spa_l2cache.sav_count == 0)
3047 return;
3048
3049 VERIFY(nvlist_lookup_nvlist(config,
3050 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3051 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3052 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3053 if (nl2cache != 0) {
3054 VERIFY(nvlist_add_nvlist_array(nvroot,
3055 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3056 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3057 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3058
3059 /*
3060 * Update level 2 cache device stats.
3061 */
3062
3063 for (i = 0; i < nl2cache; i++) {
3064 VERIFY(nvlist_lookup_uint64(l2cache[i],
3065 ZPOOL_CONFIG_GUID, &guid) == 0);
3066
3067 vd = NULL;
3068 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3069 if (guid ==
3070 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3071 vd = spa->spa_l2cache.sav_vdevs[j];
3072 break;
3073 }
3074 }
3075 ASSERT(vd != NULL);
3076
3077 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3078 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3079 == 0);
3080 vdev_get_stats(vd, vs);
3081 }
3082 }
3083 }
3084
3085 static void
3086 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3087 {
3088 nvlist_t *features;
3089 zap_cursor_t zc;
3090 zap_attribute_t za;
3091
3092 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3093 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3094
3095 if (spa->spa_feat_for_read_obj != 0) {
3096 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3097 spa->spa_feat_for_read_obj);
3098 zap_cursor_retrieve(&zc, &za) == 0;
3099 zap_cursor_advance(&zc)) {
3100 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3101 za.za_num_integers == 1);
3102 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3103 za.za_first_integer));
3104 }
3105 zap_cursor_fini(&zc);
3106 }
3107
3108 if (spa->spa_feat_for_write_obj != 0) {
3109 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3110 spa->spa_feat_for_write_obj);
3111 zap_cursor_retrieve(&zc, &za) == 0;
3112 zap_cursor_advance(&zc)) {
3113 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3114 za.za_num_integers == 1);
3115 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3116 za.za_first_integer));
3117 }
3118 zap_cursor_fini(&zc);
3119 }
3120
3121 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3122 features) == 0);
3123 nvlist_free(features);
3124 }
3125
3126 int
3127 spa_get_stats(const char *name, nvlist_t **config,
3128 char *altroot, size_t buflen)
3129 {
3130 int error;
3131 spa_t *spa;
3132
3133 *config = NULL;
3134 error = spa_open_common(name, &spa, FTAG, NULL, config);
3135
3136 if (spa != NULL) {
3137 /*
3138 * This still leaves a window of inconsistency where the spares
3139 * or l2cache devices could change and the config would be
3140 * self-inconsistent.
3141 */
3142 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3143
3144 if (*config != NULL) {
3145 uint64_t loadtimes[2];
3146
3147 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3148 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3149 VERIFY(nvlist_add_uint64_array(*config,
3150 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3151
3152 VERIFY(nvlist_add_uint64(*config,
3153 ZPOOL_CONFIG_ERRCOUNT,
3154 spa_get_errlog_size(spa)) == 0);
3155
3156 if (spa_suspended(spa))
3157 VERIFY(nvlist_add_uint64(*config,
3158 ZPOOL_CONFIG_SUSPENDED,
3159 spa->spa_failmode) == 0);
3160
3161 spa_add_spares(spa, *config);
3162 spa_add_l2cache(spa, *config);
3163 spa_add_feature_stats(spa, *config);
3164 }
3165 }
3166
3167 /*
3168 * We want to get the alternate root even for faulted pools, so we cheat
3169 * and call spa_lookup() directly.
3170 */
3171 if (altroot) {
3172 if (spa == NULL) {
3173 mutex_enter(&spa_namespace_lock);
3174 spa = spa_lookup(name);
3175 if (spa)
3176 spa_altroot(spa, altroot, buflen);
3177 else
3178 altroot[0] = '\0';
3179 spa = NULL;
3180 mutex_exit(&spa_namespace_lock);
3181 } else {
3182 spa_altroot(spa, altroot, buflen);
3183 }
3184 }
3185
3186 if (spa != NULL) {
3187 spa_config_exit(spa, SCL_CONFIG, FTAG);
3188 spa_close(spa, FTAG);
3189 }
3190
3191 return (error);
3192 }
3193
3194 /*
3195 * Validate that the auxiliary device array is well formed. We must have an
3196 * array of nvlists, each which describes a valid leaf vdev. If this is an
3197 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3198 * specified, as long as they are well-formed.
3199 */
3200 static int
3201 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3202 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3203 vdev_labeltype_t label)
3204 {
3205 nvlist_t **dev;
3206 uint_t i, ndev;
3207 vdev_t *vd;
3208 int error;
3209
3210 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3211
3212 /*
3213 * It's acceptable to have no devs specified.
3214 */
3215 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3216 return (0);
3217
3218 if (ndev == 0)
3219 return (SET_ERROR(EINVAL));
3220
3221 /*
3222 * Make sure the pool is formatted with a version that supports this
3223 * device type.
3224 */
3225 if (spa_version(spa) < version)
3226 return (SET_ERROR(ENOTSUP));
3227
3228 /*
3229 * Set the pending device list so we correctly handle device in-use
3230 * checking.
3231 */
3232 sav->sav_pending = dev;
3233 sav->sav_npending = ndev;
3234
3235 for (i = 0; i < ndev; i++) {
3236 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3237 mode)) != 0)
3238 goto out;
3239
3240 if (!vd->vdev_ops->vdev_op_leaf) {
3241 vdev_free(vd);
3242 error = SET_ERROR(EINVAL);
3243 goto out;
3244 }
3245
3246 /*
3247 * The L2ARC currently only supports disk devices in
3248 * kernel context. For user-level testing, we allow it.
3249 */
3250 #ifdef _KERNEL
3251 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3252 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3253 error = SET_ERROR(ENOTBLK);
3254 vdev_free(vd);
3255 goto out;
3256 }
3257 #endif
3258 vd->vdev_top = vd;
3259
3260 if ((error = vdev_open(vd)) == 0 &&
3261 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3262 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3263 vd->vdev_guid) == 0);
3264 }
3265
3266 vdev_free(vd);
3267
3268 if (error &&
3269 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3270 goto out;
3271 else
3272 error = 0;
3273 }
3274
3275 out:
3276 sav->sav_pending = NULL;
3277 sav->sav_npending = 0;
3278 return (error);
3279 }
3280
3281 static int
3282 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3283 {
3284 int error;
3285
3286 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3287
3288 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3289 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3290 VDEV_LABEL_SPARE)) != 0) {
3291 return (error);
3292 }
3293
3294 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3295 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3296 VDEV_LABEL_L2CACHE));
3297 }
3298
3299 static void
3300 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3301 const char *config)
3302 {
3303 int i;
3304
3305 if (sav->sav_config != NULL) {
3306 nvlist_t **olddevs;
3307 uint_t oldndevs;
3308 nvlist_t **newdevs;
3309
3310 /*
3311 * Generate new dev list by concatentating with the
3312 * current dev list.
3313 */
3314 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3315 &olddevs, &oldndevs) == 0);
3316
3317 newdevs = kmem_alloc(sizeof (void *) *
3318 (ndevs + oldndevs), KM_SLEEP);
3319 for (i = 0; i < oldndevs; i++)
3320 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3321 KM_SLEEP) == 0);
3322 for (i = 0; i < ndevs; i++)
3323 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3324 KM_SLEEP) == 0);
3325
3326 VERIFY(nvlist_remove(sav->sav_config, config,
3327 DATA_TYPE_NVLIST_ARRAY) == 0);
3328
3329 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3330 config, newdevs, ndevs + oldndevs) == 0);
3331 for (i = 0; i < oldndevs + ndevs; i++)
3332 nvlist_free(newdevs[i]);
3333 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3334 } else {
3335 /*
3336 * Generate a new dev list.
3337 */
3338 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3339 KM_SLEEP) == 0);
3340 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3341 devs, ndevs) == 0);
3342 }
3343 }
3344
3345 /*
3346 * Stop and drop level 2 ARC devices
3347 */
3348 void
3349 spa_l2cache_drop(spa_t *spa)
3350 {
3351 vdev_t *vd;
3352 int i;
3353 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3354
3355 for (i = 0; i < sav->sav_count; i++) {
3356 uint64_t pool;
3357
3358 vd = sav->sav_vdevs[i];
3359 ASSERT(vd != NULL);
3360
3361 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3362 pool != 0ULL && l2arc_vdev_present(vd))
3363 l2arc_remove_vdev(vd);
3364 }
3365 }
3366
3367 /*
3368 * Pool Creation
3369 */
3370 int
3371 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3372 nvlist_t *zplprops)
3373 {
3374 spa_t *spa;
3375 char *altroot = NULL;
3376 vdev_t *rvd;
3377 dsl_pool_t *dp;
3378 dmu_tx_t *tx;
3379 int error = 0;
3380 uint64_t txg = TXG_INITIAL;
3381 nvlist_t **spares, **l2cache;
3382 uint_t nspares, nl2cache;
3383 uint64_t version, obj;
3384 boolean_t has_features;
3385
3386 /*
3387 * If this pool already exists, return failure.
3388 */
3389 mutex_enter(&spa_namespace_lock);
3390 if (spa_lookup(pool) != NULL) {
3391 mutex_exit(&spa_namespace_lock);
3392 return (SET_ERROR(EEXIST));
3393 }
3394
3395 /*
3396 * Allocate a new spa_t structure.
3397 */
3398 (void) nvlist_lookup_string(props,
3399 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3400 spa = spa_add(pool, NULL, altroot);
3401 spa_activate(spa, spa_mode_global);
3402
3403 if (props && (error = spa_prop_validate(spa, props))) {
3404 spa_deactivate(spa);
3405 spa_remove(spa);
3406 mutex_exit(&spa_namespace_lock);
3407 return (error);
3408 }
3409
3410 has_features = B_FALSE;
3411 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3412 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3413 if (zpool_prop_feature(nvpair_name(elem)))
3414 has_features = B_TRUE;
3415 }
3416
3417 if (has_features || nvlist_lookup_uint64(props,
3418 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3419 version = SPA_VERSION;
3420 }
3421 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3422
3423 spa->spa_first_txg = txg;
3424 spa->spa_uberblock.ub_txg = txg - 1;
3425 spa->spa_uberblock.ub_version = version;
3426 spa->spa_ubsync = spa->spa_uberblock;
3427
3428 /*
3429 * Create "The Godfather" zio to hold all async IOs
3430 */
3431 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3432 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3433
3434 /*
3435 * Create the root vdev.
3436 */
3437 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3438
3439 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3440
3441 ASSERT(error != 0 || rvd != NULL);
3442 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3443
3444 if (error == 0 && !zfs_allocatable_devs(nvroot))
3445 error = SET_ERROR(EINVAL);
3446
3447 if (error == 0 &&
3448 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3449 (error = spa_validate_aux(spa, nvroot, txg,
3450 VDEV_ALLOC_ADD)) == 0) {
3451 for (int c = 0; c < rvd->vdev_children; c++) {
3452 vdev_metaslab_set_size(rvd->vdev_child[c]);
3453 vdev_expand(rvd->vdev_child[c], txg);
3454 }
3455 }
3456
3457 spa_config_exit(spa, SCL_ALL, FTAG);
3458
3459 if (error != 0) {
3460 spa_unload(spa);
3461 spa_deactivate(spa);
3462 spa_remove(spa);
3463 mutex_exit(&spa_namespace_lock);
3464 return (error);
3465 }
3466
3467 /*
3468 * Get the list of spares, if specified.
3469 */
3470 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3471 &spares, &nspares) == 0) {
3472 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3473 KM_SLEEP) == 0);
3474 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3475 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3476 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3477 spa_load_spares(spa);
3478 spa_config_exit(spa, SCL_ALL, FTAG);
3479 spa->spa_spares.sav_sync = B_TRUE;
3480 }
3481
3482 /*
3483 * Get the list of level 2 cache devices, if specified.
3484 */
3485 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3486 &l2cache, &nl2cache) == 0) {
3487 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3488 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3489 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3490 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3491 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3492 spa_load_l2cache(spa);
3493 spa_config_exit(spa, SCL_ALL, FTAG);
3494 spa->spa_l2cache.sav_sync = B_TRUE;
3495 }
3496
3497 spa->spa_is_initializing = B_TRUE;
3498 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3499 spa->spa_meta_objset = dp->dp_meta_objset;
3500 spa->spa_is_initializing = B_FALSE;
3501
3502 /*
3503 * Create DDTs (dedup tables).
3504 */
3505 ddt_create(spa);
3506
3507 spa_update_dspace(spa);
3508
3509 tx = dmu_tx_create_assigned(dp, txg);
3510
3511 /*
3512 * Create the pool config object.
3513 */
3514 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3515 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3516 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3517
3518 if (zap_add(spa->spa_meta_objset,
3519 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3520 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3521 cmn_err(CE_PANIC, "failed to add pool config");
3522 }
3523
3524 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3525 spa_feature_create_zap_objects(spa, tx);
3526
3527 if (zap_add(spa->spa_meta_objset,
3528 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3529 sizeof (uint64_t), 1, &version, tx) != 0) {
3530 cmn_err(CE_PANIC, "failed to add pool version");
3531 }
3532
3533 /* Newly created pools with the right version are always deflated. */
3534 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3535 spa->spa_deflate = TRUE;
3536 if (zap_add(spa->spa_meta_objset,
3537 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3538 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3539 cmn_err(CE_PANIC, "failed to add deflate");
3540 }
3541 }
3542
3543 /*
3544 * Create the deferred-free bpobj. Turn off compression
3545 * because sync-to-convergence takes longer if the blocksize
3546 * keeps changing.
3547 */
3548 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3549 dmu_object_set_compress(spa->spa_meta_objset, obj,
3550 ZIO_COMPRESS_OFF, tx);
3551 if (zap_add(spa->spa_meta_objset,
3552 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3553 sizeof (uint64_t), 1, &obj, tx) != 0) {
3554 cmn_err(CE_PANIC, "failed to add bpobj");
3555 }
3556 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3557 spa->spa_meta_objset, obj));
3558
3559 /*
3560 * Create the pool's history object.
3561 */
3562 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3563 spa_history_create_obj(spa, tx);
3564
3565 /*
3566 * Set pool properties.
3567 */
3568 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3569 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3570 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3571 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3572
3573 if (props != NULL) {
3574 spa_configfile_set(spa, props, B_FALSE);
3575 spa_sync_props(props, tx);
3576 }
3577
3578 dmu_tx_commit(tx);
3579
3580 spa->spa_sync_on = B_TRUE;
3581 txg_sync_start(spa->spa_dsl_pool);
3582
3583 /*
3584 * We explicitly wait for the first transaction to complete so that our
3585 * bean counters are appropriately updated.
3586 */
3587 txg_wait_synced(spa->spa_dsl_pool, txg);
3588
3589 spa_config_sync(spa, B_FALSE, B_TRUE);
3590
3591 spa_history_log_version(spa, "create");
3592
3593 spa->spa_minref = refcount_count(&spa->spa_refcount);
3594
3595 mutex_exit(&spa_namespace_lock);
3596
3597 return (0);
3598 }
3599
3600 #ifdef _KERNEL
3601 /*
3602 * Get the root pool information from the root disk, then import the root pool
3603 * during the system boot up time.
3604 */
3605 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3606
3607 static nvlist_t *
3608 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3609 {
3610 nvlist_t *config;
3611 nvlist_t *nvtop, *nvroot;
3612 uint64_t pgid;
3613
3614 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3615 return (NULL);
3616
3617 /*
3618 * Add this top-level vdev to the child array.
3619 */
3620 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3621 &nvtop) == 0);
3622 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3623 &pgid) == 0);
3624 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3625
3626 /*
3627 * Put this pool's top-level vdevs into a root vdev.
3628 */
3629 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3630 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3631 VDEV_TYPE_ROOT) == 0);
3632 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3633 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3634 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3635 &nvtop, 1) == 0);
3636
3637 /*
3638 * Replace the existing vdev_tree with the new root vdev in
3639 * this pool's configuration (remove the old, add the new).
3640 */
3641 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3642 nvlist_free(nvroot);
3643 return (config);
3644 }
3645
3646 /*
3647 * Walk the vdev tree and see if we can find a device with "better"
3648 * configuration. A configuration is "better" if the label on that
3649 * device has a more recent txg.
3650 */
3651 static void
3652 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3653 {
3654 for (int c = 0; c < vd->vdev_children; c++)
3655 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3656
3657 if (vd->vdev_ops->vdev_op_leaf) {
3658 nvlist_t *label;
3659 uint64_t label_txg;
3660
3661 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3662 &label) != 0)
3663 return;
3664
3665 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3666 &label_txg) == 0);
3667
3668 /*
3669 * Do we have a better boot device?
3670 */
3671 if (label_txg > *txg) {
3672 *txg = label_txg;
3673 *avd = vd;
3674 }
3675 nvlist_free(label);
3676 }
3677 }
3678
3679 /*
3680 * Import a root pool.
3681 *
3682 * For x86. devpath_list will consist of devid and/or physpath name of
3683 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3684 * The GRUB "findroot" command will return the vdev we should boot.
3685 *
3686 * For Sparc, devpath_list consists the physpath name of the booting device
3687 * no matter the rootpool is a single device pool or a mirrored pool.
3688 * e.g.
3689 * "/pci@1f,0/ide@d/disk@0,0:a"
3690 */
3691 int
3692 spa_import_rootpool(char *devpath, char *devid)
3693 {
3694 spa_t *spa;
3695 vdev_t *rvd, *bvd, *avd = NULL;
3696 nvlist_t *config, *nvtop;
3697 uint64_t guid, txg;
3698 char *pname;
3699 int error;
3700
3701 /*
3702 * Read the label from the boot device and generate a configuration.
3703 */
3704 config = spa_generate_rootconf(devpath, devid, &guid);
3705 #if defined(_OBP) && defined(_KERNEL)
3706 if (config == NULL) {
3707 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3708 /* iscsi boot */
3709 get_iscsi_bootpath_phy(devpath);
3710 config = spa_generate_rootconf(devpath, devid, &guid);
3711 }
3712 }
3713 #endif
3714 if (config == NULL) {
3715 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3716 devpath);
3717 return (SET_ERROR(EIO));
3718 }
3719
3720 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3721 &pname) == 0);
3722 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3723
3724 mutex_enter(&spa_namespace_lock);
3725 if ((spa = spa_lookup(pname)) != NULL) {
3726 /*
3727 * Remove the existing root pool from the namespace so that we
3728 * can replace it with the correct config we just read in.
3729 */
3730 spa_remove(spa);
3731 }
3732
3733 spa = spa_add(pname, config, NULL);
3734 spa->spa_is_root = B_TRUE;
3735 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3736
3737 /*
3738 * Build up a vdev tree based on the boot device's label config.
3739 */
3740 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3741 &nvtop) == 0);
3742 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3743 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3744 VDEV_ALLOC_ROOTPOOL);
3745 spa_config_exit(spa, SCL_ALL, FTAG);
3746 if (error) {
3747 mutex_exit(&spa_namespace_lock);
3748 nvlist_free(config);
3749 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3750 pname);
3751 return (error);
3752 }
3753
3754 /*
3755 * Get the boot vdev.
3756 */
3757 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3758 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3759 (u_longlong_t)guid);
3760 error = SET_ERROR(ENOENT);
3761 goto out;
3762 }
3763
3764 /*
3765 * Determine if there is a better boot device.
3766 */
3767 avd = bvd;
3768 spa_alt_rootvdev(rvd, &avd, &txg);
3769 if (avd != bvd) {
3770 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3771 "try booting from '%s'", avd->vdev_path);
3772 error = SET_ERROR(EINVAL);
3773 goto out;
3774 }
3775
3776 /*
3777 * If the boot device is part of a spare vdev then ensure that
3778 * we're booting off the active spare.
3779 */
3780 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3781 !bvd->vdev_isspare) {
3782 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3783 "try booting from '%s'",
3784 bvd->vdev_parent->
3785 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3786 error = SET_ERROR(EINVAL);
3787 goto out;
3788 }
3789
3790 error = 0;
3791 out:
3792 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3793 vdev_free(rvd);
3794 spa_config_exit(spa, SCL_ALL, FTAG);
3795 mutex_exit(&spa_namespace_lock);
3796
3797 nvlist_free(config);
3798 return (error);
3799 }
3800
3801 #endif
3802
3803 /*
3804 * Import a non-root pool into the system.
3805 */
3806 int
3807 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3808 {
3809 spa_t *spa;
3810 char *altroot = NULL;
3811 spa_load_state_t state = SPA_LOAD_IMPORT;
3812 zpool_rewind_policy_t policy;
3813 uint64_t mode = spa_mode_global;
3814 uint64_t readonly = B_FALSE;
3815 int error;
3816 nvlist_t *nvroot;
3817 nvlist_t **spares, **l2cache;
3818 uint_t nspares, nl2cache;
3819
3820 /*
3821 * If a pool with this name exists, return failure.
3822 */
3823 mutex_enter(&spa_namespace_lock);
3824 if (spa_lookup(pool) != NULL) {
3825 mutex_exit(&spa_namespace_lock);
3826 return (SET_ERROR(EEXIST));
3827 }
3828
3829 /*
3830 * Create and initialize the spa structure.
3831 */
3832 (void) nvlist_lookup_string(props,
3833 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3834 (void) nvlist_lookup_uint64(props,
3835 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3836 if (readonly)
3837 mode = FREAD;
3838 spa = spa_add(pool, config, altroot);
3839 spa->spa_import_flags = flags;
3840
3841 /*
3842 * Verbatim import - Take a pool and insert it into the namespace
3843 * as if it had been loaded at boot.
3844 */
3845 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3846 if (props != NULL)
3847 spa_configfile_set(spa, props, B_FALSE);
3848
3849 spa_config_sync(spa, B_FALSE, B_TRUE);
3850
3851 mutex_exit(&spa_namespace_lock);
3852 spa_history_log_version(spa, "import");
3853
3854 return (0);
3855 }
3856
3857 spa_activate(spa, mode);
3858
3859 /*
3860 * Don't start async tasks until we know everything is healthy.
3861 */
3862 spa_async_suspend(spa);
3863
3864 zpool_get_rewind_policy(config, &policy);
3865 if (policy.zrp_request & ZPOOL_DO_REWIND)
3866 state = SPA_LOAD_RECOVER;
3867
3868 /*
3869 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3870 * because the user-supplied config is actually the one to trust when
3871 * doing an import.
3872 */
3873 if (state != SPA_LOAD_RECOVER)
3874 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3875
3876 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3877 policy.zrp_request);
3878
3879 /*
3880 * Propagate anything learned while loading the pool and pass it
3881 * back to caller (i.e. rewind info, missing devices, etc).
3882 */
3883 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3884 spa->spa_load_info) == 0);
3885
3886 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3887 /*
3888 * Toss any existing sparelist, as it doesn't have any validity
3889 * anymore, and conflicts with spa_has_spare().
3890 */
3891 if (spa->spa_spares.sav_config) {
3892 nvlist_free(spa->spa_spares.sav_config);
3893 spa->spa_spares.sav_config = NULL;
3894 spa_load_spares(spa);
3895 }
3896
3897 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3898 &nvroot) == 0);
3899 if (error == 0)
3900 error = spa_validate_aux(spa, nvroot, -1ULL,
3901 VDEV_ALLOC_SPARE);
3902 if (error == 0)
3903 error = spa_validate_aux(spa, nvroot, -1ULL,
3904 VDEV_ALLOC_L2CACHE);
3905 spa_config_exit(spa, SCL_ALL, FTAG);
3906
3907 if (props != NULL)
3908 spa_configfile_set(spa, props, B_FALSE);
3909
3910 if (error != 0 || (props && spa_writeable(spa) &&
3911 (error = spa_prop_set(spa, props)))) {
3912 spa_unload(spa);
3913 spa_deactivate(spa);
3914 spa_remove(spa);
3915 mutex_exit(&spa_namespace_lock);
3916 return (error);
3917 }
3918
3919 spa_async_resume(spa);
3920
3921 /*
3922 * Override any spares and level 2 cache devices as specified by
3923 * the user, as these may have correct device names/devids, etc.
3924 */
3925 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3926 &spares, &nspares) == 0) {
3927 if (spa->spa_spares.sav_config)
3928 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3929 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3930 else
3931 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3932 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3933 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3934 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3935 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3936 spa_load_spares(spa);
3937 spa_config_exit(spa, SCL_ALL, FTAG);
3938 spa->spa_spares.sav_sync = B_TRUE;
3939 }
3940 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3941 &l2cache, &nl2cache) == 0) {
3942 if (spa->spa_l2cache.sav_config)
3943 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3944 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3945 else
3946 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3947 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3948 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3949 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3950 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3951 spa_load_l2cache(spa);
3952 spa_config_exit(spa, SCL_ALL, FTAG);
3953 spa->spa_l2cache.sav_sync = B_TRUE;
3954 }
3955
3956 /*
3957 * Check for any removed devices.
3958 */
3959 if (spa->spa_autoreplace) {
3960 spa_aux_check_removed(&spa->spa_spares);
3961 spa_aux_check_removed(&spa->spa_l2cache);
3962 }
3963
3964 if (spa_writeable(spa)) {
3965 /*
3966 * Update the config cache to include the newly-imported pool.
3967 */
3968 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3969 }
3970
3971 /*
3972 * It's possible that the pool was expanded while it was exported.
3973 * We kick off an async task to handle this for us.
3974 */
3975 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3976
3977 mutex_exit(&spa_namespace_lock);
3978 spa_history_log_version(spa, "import");
3979
3980 return (0);
3981 }
3982
3983 nvlist_t *
3984 spa_tryimport(nvlist_t *tryconfig)
3985 {
3986 nvlist_t *config = NULL;
3987 char *poolname;
3988 spa_t *spa;
3989 uint64_t state;
3990 int error;
3991
3992 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3993 return (NULL);
3994
3995 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3996 return (NULL);
3997
3998 /*
3999 * Create and initialize the spa structure.
4000 */
4001 mutex_enter(&spa_namespace_lock);
4002 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4003 spa_activate(spa, FREAD);
4004
4005 /*
4006 * Pass off the heavy lifting to spa_load().
4007 * Pass TRUE for mosconfig because the user-supplied config
4008 * is actually the one to trust when doing an import.
4009 */
4010 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4011
4012 /*
4013 * If 'tryconfig' was at least parsable, return the current config.
4014 */
4015 if (spa->spa_root_vdev != NULL) {
4016 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4017 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4018 poolname) == 0);
4019 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4020 state) == 0);
4021 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4022 spa->spa_uberblock.ub_timestamp) == 0);
4023 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4024 spa->spa_load_info) == 0);
4025
4026 /*
4027 * If the bootfs property exists on this pool then we
4028 * copy it out so that external consumers can tell which
4029 * pools are bootable.
4030 */
4031 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4032 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4033
4034 /*
4035 * We have to play games with the name since the
4036 * pool was opened as TRYIMPORT_NAME.
4037 */
4038 if (dsl_dsobj_to_dsname(spa_name(spa),
4039 spa->spa_bootfs, tmpname) == 0) {
4040 char *cp;
4041 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4042
4043 cp = strchr(tmpname, '/');
4044 if (cp == NULL) {
4045 (void) strlcpy(dsname, tmpname,
4046 MAXPATHLEN);
4047 } else {
4048 (void) snprintf(dsname, MAXPATHLEN,
4049 "%s/%s", poolname, ++cp);
4050 }
4051 VERIFY(nvlist_add_string(config,
4052 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4053 kmem_free(dsname, MAXPATHLEN);
4054 }
4055 kmem_free(tmpname, MAXPATHLEN);
4056 }
4057
4058 /*
4059 * Add the list of hot spares and level 2 cache devices.
4060 */
4061 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4062 spa_add_spares(spa, config);
4063 spa_add_l2cache(spa, config);
4064 spa_config_exit(spa, SCL_CONFIG, FTAG);
4065 }
4066
4067 spa_unload(spa);
4068 spa_deactivate(spa);
4069 spa_remove(spa);
4070 mutex_exit(&spa_namespace_lock);
4071
4072 return (config);
4073 }
4074
4075 /*
4076 * Pool export/destroy
4077 *
4078 * The act of destroying or exporting a pool is very simple. We make sure there
4079 * is no more pending I/O and any references to the pool are gone. Then, we
4080 * update the pool state and sync all the labels to disk, removing the
4081 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4082 * we don't sync the labels or remove the configuration cache.
4083 */
4084 static int
4085 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4086 boolean_t force, boolean_t hardforce)
4087 {
4088 spa_t *spa;
4089
4090 if (oldconfig)
4091 *oldconfig = NULL;
4092
4093 if (!(spa_mode_global & FWRITE))
4094 return (SET_ERROR(EROFS));
4095
4096 mutex_enter(&spa_namespace_lock);
4097 if ((spa = spa_lookup(pool)) == NULL) {
4098 mutex_exit(&spa_namespace_lock);
4099 return (SET_ERROR(ENOENT));
4100 }
4101
4102 /*
4103 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4104 * reacquire the namespace lock, and see if we can export.
4105 */
4106 spa_open_ref(spa, FTAG);
4107 mutex_exit(&spa_namespace_lock);
4108 spa_async_suspend(spa);
4109 mutex_enter(&spa_namespace_lock);
4110 spa_close(spa, FTAG);
4111
4112 /*
4113 * The pool will be in core if it's openable,
4114 * in which case we can modify its state.
4115 */
4116 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4117 /*
4118 * Objsets may be open only because they're dirty, so we
4119 * have to force it to sync before checking spa_refcnt.
4120 */
4121 txg_wait_synced(spa->spa_dsl_pool, 0);
4122
4123 /*
4124 * A pool cannot be exported or destroyed if there are active
4125 * references. If we are resetting a pool, allow references by
4126 * fault injection handlers.
4127 */
4128 if (!spa_refcount_zero(spa) ||
4129 (spa->spa_inject_ref != 0 &&
4130 new_state != POOL_STATE_UNINITIALIZED)) {
4131 spa_async_resume(spa);
4132 mutex_exit(&spa_namespace_lock);
4133 return (SET_ERROR(EBUSY));
4134 }
4135
4136 /*
4137 * A pool cannot be exported if it has an active shared spare.
4138 * This is to prevent other pools stealing the active spare
4139 * from an exported pool. At user's own will, such pool can
4140 * be forcedly exported.
4141 */
4142 if (!force && new_state == POOL_STATE_EXPORTED &&
4143 spa_has_active_shared_spare(spa)) {
4144 spa_async_resume(spa);
4145 mutex_exit(&spa_namespace_lock);
4146 return (SET_ERROR(EXDEV));
4147 }
4148
4149 /*
4150 * We want this to be reflected on every label,
4151 * so mark them all dirty. spa_unload() will do the
4152 * final sync that pushes these changes out.
4153 */
4154 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4155 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4156 spa->spa_state = new_state;
4157 spa->spa_final_txg = spa_last_synced_txg(spa) +
4158 TXG_DEFER_SIZE + 1;
4159 vdev_config_dirty(spa->spa_root_vdev);
4160 spa_config_exit(spa, SCL_ALL, FTAG);
4161 }
4162 }
4163
4164 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4165
4166 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4167 spa_unload(spa);
4168 spa_deactivate(spa);
4169 }
4170
4171 if (oldconfig && spa->spa_config)
4172 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4173
4174 if (new_state != POOL_STATE_UNINITIALIZED) {
4175 if (!hardforce)
4176 spa_config_sync(spa, B_TRUE, B_TRUE);
4177 spa_remove(spa);
4178 }
4179 mutex_exit(&spa_namespace_lock);
4180
4181 return (0);
4182 }
4183
4184 /*
4185 * Destroy a storage pool.
4186 */
4187 int
4188 spa_destroy(char *pool)
4189 {
4190 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4191 B_FALSE, B_FALSE));
4192 }
4193
4194 /*
4195 * Export a storage pool.
4196 */
4197 int
4198 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4199 boolean_t hardforce)
4200 {
4201 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4202 force, hardforce));
4203 }
4204
4205 /*
4206 * Similar to spa_export(), this unloads the spa_t without actually removing it
4207 * from the namespace in any way.
4208 */
4209 int
4210 spa_reset(char *pool)
4211 {
4212 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4213 B_FALSE, B_FALSE));
4214 }
4215
4216 /*
4217 * ==========================================================================
4218 * Device manipulation
4219 * ==========================================================================
4220 */
4221
4222 /*
4223 * Add a device to a storage pool.
4224 */
4225 int
4226 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4227 {
4228 uint64_t txg, id;
4229 int error;
4230 vdev_t *rvd = spa->spa_root_vdev;
4231 vdev_t *vd, *tvd;
4232 nvlist_t **spares, **l2cache;
4233 uint_t nspares, nl2cache;
4234
4235 ASSERT(spa_writeable(spa));
4236
4237 txg = spa_vdev_enter(spa);
4238
4239 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4240 VDEV_ALLOC_ADD)) != 0)
4241 return (spa_vdev_exit(spa, NULL, txg, error));
4242
4243 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4244
4245 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4246 &nspares) != 0)
4247 nspares = 0;
4248
4249 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4250 &nl2cache) != 0)
4251 nl2cache = 0;
4252
4253 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4254 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4255
4256 if (vd->vdev_children != 0 &&
4257 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4258 return (spa_vdev_exit(spa, vd, txg, error));
4259
4260 /*
4261 * We must validate the spares and l2cache devices after checking the
4262 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4263 */
4264 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4265 return (spa_vdev_exit(spa, vd, txg, error));
4266
4267 /*
4268 * Transfer each new top-level vdev from vd to rvd.
4269 */
4270 for (int c = 0; c < vd->vdev_children; c++) {
4271
4272 /*
4273 * Set the vdev id to the first hole, if one exists.
4274 */
4275 for (id = 0; id < rvd->vdev_children; id++) {
4276 if (rvd->vdev_child[id]->vdev_ishole) {
4277 vdev_free(rvd->vdev_child[id]);
4278 break;
4279 }
4280 }
4281 tvd = vd->vdev_child[c];
4282 vdev_remove_child(vd, tvd);
4283 tvd->vdev_id = id;
4284 vdev_add_child(rvd, tvd);
4285 vdev_config_dirty(tvd);
4286 }
4287
4288 if (nspares != 0) {
4289 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4290 ZPOOL_CONFIG_SPARES);
4291 spa_load_spares(spa);
4292 spa->spa_spares.sav_sync = B_TRUE;
4293 }
4294
4295 if (nl2cache != 0) {
4296 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4297 ZPOOL_CONFIG_L2CACHE);
4298 spa_load_l2cache(spa);
4299 spa->spa_l2cache.sav_sync = B_TRUE;
4300 }
4301
4302 /*
4303 * We have to be careful when adding new vdevs to an existing pool.
4304 * If other threads start allocating from these vdevs before we
4305 * sync the config cache, and we lose power, then upon reboot we may
4306 * fail to open the pool because there are DVAs that the config cache
4307 * can't translate. Therefore, we first add the vdevs without
4308 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4309 * and then let spa_config_update() initialize the new metaslabs.
4310 *
4311 * spa_load() checks for added-but-not-initialized vdevs, so that
4312 * if we lose power at any point in this sequence, the remaining
4313 * steps will be completed the next time we load the pool.
4314 */
4315 (void) spa_vdev_exit(spa, vd, txg, 0);
4316
4317 mutex_enter(&spa_namespace_lock);
4318 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4319 mutex_exit(&spa_namespace_lock);
4320
4321 return (0);
4322 }
4323
4324 /*
4325 * Attach a device to a mirror. The arguments are the path to any device
4326 * in the mirror, and the nvroot for the new device. If the path specifies
4327 * a device that is not mirrored, we automatically insert the mirror vdev.
4328 *
4329 * If 'replacing' is specified, the new device is intended to replace the
4330 * existing device; in this case the two devices are made into their own
4331 * mirror using the 'replacing' vdev, which is functionally identical to
4332 * the mirror vdev (it actually reuses all the same ops) but has a few
4333 * extra rules: you can't attach to it after it's been created, and upon
4334 * completion of resilvering, the first disk (the one being replaced)
4335 * is automatically detached.
4336 */
4337 int
4338 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4339 {
4340 uint64_t txg, dtl_max_txg;
4341 vdev_t *rvd = spa->spa_root_vdev;
4342 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4343 vdev_ops_t *pvops;
4344 char *oldvdpath, *newvdpath;
4345 int newvd_isspare;
4346 int error;
4347
4348 ASSERT(spa_writeable(spa));
4349
4350 txg = spa_vdev_enter(spa);
4351
4352 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4353
4354 if (oldvd == NULL)
4355 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4356
4357 if (!oldvd->vdev_ops->vdev_op_leaf)
4358 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4359
4360 pvd = oldvd->vdev_parent;
4361
4362 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4363 VDEV_ALLOC_ATTACH)) != 0)
4364 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4365
4366 if (newrootvd->vdev_children != 1)
4367 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4368
4369 newvd = newrootvd->vdev_child[0];
4370
4371 if (!newvd->vdev_ops->vdev_op_leaf)
4372 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4373
4374 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4375 return (spa_vdev_exit(spa, newrootvd, txg, error));
4376
4377 /*
4378 * Spares can't replace logs
4379 */
4380 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4381 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4382
4383 if (!replacing) {
4384 /*
4385 * For attach, the only allowable parent is a mirror or the root
4386 * vdev.
4387 */
4388 if (pvd->vdev_ops != &vdev_mirror_ops &&
4389 pvd->vdev_ops != &vdev_root_ops)
4390 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4391
4392 pvops = &vdev_mirror_ops;
4393 } else {
4394 /*
4395 * Active hot spares can only be replaced by inactive hot
4396 * spares.
4397 */
4398 if (pvd->vdev_ops == &vdev_spare_ops &&
4399 oldvd->vdev_isspare &&
4400 !spa_has_spare(spa, newvd->vdev_guid))
4401 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4402
4403 /*
4404 * If the source is a hot spare, and the parent isn't already a
4405 * spare, then we want to create a new hot spare. Otherwise, we
4406 * want to create a replacing vdev. The user is not allowed to
4407 * attach to a spared vdev child unless the 'isspare' state is
4408 * the same (spare replaces spare, non-spare replaces
4409 * non-spare).
4410 */
4411 if (pvd->vdev_ops == &vdev_replacing_ops &&
4412 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4413 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4414 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4415 newvd->vdev_isspare != oldvd->vdev_isspare) {
4416 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4417 }
4418
4419 if (newvd->vdev_isspare)
4420 pvops = &vdev_spare_ops;
4421 else
4422 pvops = &vdev_replacing_ops;
4423 }
4424
4425 /*
4426 * Make sure the new device is big enough.
4427 */
4428 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4429 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4430
4431 /*
4432 * The new device cannot have a higher alignment requirement
4433 * than the top-level vdev.
4434 */
4435 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4436 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4437
4438 /*
4439 * If this is an in-place replacement, update oldvd's path and devid
4440 * to make it distinguishable from newvd, and unopenable from now on.
4441 */
4442 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4443 spa_strfree(oldvd->vdev_path);
4444 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4445 KM_SLEEP);
4446 (void) sprintf(oldvd->vdev_path, "%s/%s",
4447 newvd->vdev_path, "old");
4448 if (oldvd->vdev_devid != NULL) {
4449 spa_strfree(oldvd->vdev_devid);
4450 oldvd->vdev_devid = NULL;
4451 }
4452 }
4453
4454 /* mark the device being resilvered */
4455 newvd->vdev_resilvering = B_TRUE;
4456
4457 /*
4458 * If the parent is not a mirror, or if we're replacing, insert the new
4459 * mirror/replacing/spare vdev above oldvd.
4460 */
4461 if (pvd->vdev_ops != pvops)
4462 pvd = vdev_add_parent(oldvd, pvops);
4463
4464 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4465 ASSERT(pvd->vdev_ops == pvops);
4466 ASSERT(oldvd->vdev_parent == pvd);
4467
4468 /*
4469 * Extract the new device from its root and add it to pvd.
4470 */
4471 vdev_remove_child(newrootvd, newvd);
4472 newvd->vdev_id = pvd->vdev_children;
4473 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4474 vdev_add_child(pvd, newvd);
4475
4476 tvd = newvd->vdev_top;
4477 ASSERT(pvd->vdev_top == tvd);
4478 ASSERT(tvd->vdev_parent == rvd);
4479
4480 vdev_config_dirty(tvd);
4481
4482 /*
4483 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4484 * for any dmu_sync-ed blocks. It will propagate upward when
4485 * spa_vdev_exit() calls vdev_dtl_reassess().
4486 */
4487 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4488
4489 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4490 dtl_max_txg - TXG_INITIAL);
4491
4492 if (newvd->vdev_isspare) {
4493 spa_spare_activate(newvd);
4494 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4495 }
4496
4497 oldvdpath = spa_strdup(oldvd->vdev_path);
4498 newvdpath = spa_strdup(newvd->vdev_path);
4499 newvd_isspare = newvd->vdev_isspare;
4500
4501 /*
4502 * Mark newvd's DTL dirty in this txg.
4503 */
4504 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4505
4506 /*
4507 * Restart the resilver
4508 */
4509 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4510
4511 /*
4512 * Commit the config
4513 */
4514 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4515
4516 spa_history_log_internal(spa, "vdev attach", NULL,
4517 "%s vdev=%s %s vdev=%s",
4518 replacing && newvd_isspare ? "spare in" :
4519 replacing ? "replace" : "attach", newvdpath,
4520 replacing ? "for" : "to", oldvdpath);
4521
4522 spa_strfree(oldvdpath);
4523 spa_strfree(newvdpath);
4524
4525 if (spa->spa_bootfs)
4526 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4527
4528 return (0);
4529 }
4530
4531 /*
4532 * Detach a device from a mirror or replacing vdev.
4533 *
4534 * If 'replace_done' is specified, only detach if the parent
4535 * is a replacing vdev.
4536 */
4537 int
4538 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4539 {
4540 uint64_t txg;
4541 int error;
4542 vdev_t *rvd = spa->spa_root_vdev;
4543 vdev_t *vd, *pvd, *cvd, *tvd;
4544 boolean_t unspare = B_FALSE;
4545 uint64_t unspare_guid = 0;
4546 char *vdpath;
4547
4548 ASSERT(spa_writeable(spa));
4549
4550 txg = spa_vdev_enter(spa);
4551
4552 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4553
4554 if (vd == NULL)
4555 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4556
4557 if (!vd->vdev_ops->vdev_op_leaf)
4558 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4559
4560 pvd = vd->vdev_parent;
4561
4562 /*
4563 * If the parent/child relationship is not as expected, don't do it.
4564 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4565 * vdev that's replacing B with C. The user's intent in replacing
4566 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4567 * the replace by detaching C, the expected behavior is to end up
4568 * M(A,B). But suppose that right after deciding to detach C,
4569 * the replacement of B completes. We would have M(A,C), and then
4570 * ask to detach C, which would leave us with just A -- not what
4571 * the user wanted. To prevent this, we make sure that the
4572 * parent/child relationship hasn't changed -- in this example,
4573 * that C's parent is still the replacing vdev R.
4574 */
4575 if (pvd->vdev_guid != pguid && pguid != 0)
4576 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4577
4578 /*
4579 * Only 'replacing' or 'spare' vdevs can be replaced.
4580 */
4581 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4582 pvd->vdev_ops != &vdev_spare_ops)
4583 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4584
4585 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4586 spa_version(spa) >= SPA_VERSION_SPARES);
4587
4588 /*
4589 * Only mirror, replacing, and spare vdevs support detach.
4590 */
4591 if (pvd->vdev_ops != &vdev_replacing_ops &&
4592 pvd->vdev_ops != &vdev_mirror_ops &&
4593 pvd->vdev_ops != &vdev_spare_ops)
4594 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4595
4596 /*
4597 * If this device has the only valid copy of some data,
4598 * we cannot safely detach it.
4599 */
4600 if (vdev_dtl_required(vd))
4601 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4602
4603 ASSERT(pvd->vdev_children >= 2);
4604
4605 /*
4606 * If we are detaching the second disk from a replacing vdev, then
4607 * check to see if we changed the original vdev's path to have "/old"
4608 * at the end in spa_vdev_attach(). If so, undo that change now.
4609 */
4610 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4611 vd->vdev_path != NULL) {
4612 size_t len = strlen(vd->vdev_path);
4613
4614 for (int c = 0; c < pvd->vdev_children; c++) {
4615 cvd = pvd->vdev_child[c];
4616
4617 if (cvd == vd || cvd->vdev_path == NULL)
4618 continue;
4619
4620 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4621 strcmp(cvd->vdev_path + len, "/old") == 0) {
4622 spa_strfree(cvd->vdev_path);
4623 cvd->vdev_path = spa_strdup(vd->vdev_path);
4624 break;
4625 }
4626 }
4627 }
4628
4629 /*
4630 * If we are detaching the original disk from a spare, then it implies
4631 * that the spare should become a real disk, and be removed from the
4632 * active spare list for the pool.
4633 */
4634 if (pvd->vdev_ops == &vdev_spare_ops &&
4635 vd->vdev_id == 0 &&
4636 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4637 unspare = B_TRUE;
4638
4639 /*
4640 * Erase the disk labels so the disk can be used for other things.
4641 * This must be done after all other error cases are handled,
4642 * but before we disembowel vd (so we can still do I/O to it).
4643 * But if we can't do it, don't treat the error as fatal --
4644 * it may be that the unwritability of the disk is the reason
4645 * it's being detached!
4646 */
4647 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4648
4649 /*
4650 * Remove vd from its parent and compact the parent's children.
4651 */
4652 vdev_remove_child(pvd, vd);
4653 vdev_compact_children(pvd);
4654
4655 /*
4656 * Remember one of the remaining children so we can get tvd below.
4657 */
4658 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4659
4660 /*
4661 * If we need to remove the remaining child from the list of hot spares,
4662 * do it now, marking the vdev as no longer a spare in the process.
4663 * We must do this before vdev_remove_parent(), because that can
4664 * change the GUID if it creates a new toplevel GUID. For a similar
4665 * reason, we must remove the spare now, in the same txg as the detach;
4666 * otherwise someone could attach a new sibling, change the GUID, and
4667 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4668 */
4669 if (unspare) {
4670 ASSERT(cvd->vdev_isspare);
4671 spa_spare_remove(cvd);
4672 unspare_guid = cvd->vdev_guid;
4673 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4674 cvd->vdev_unspare = B_TRUE;
4675 }
4676
4677 /*
4678 * If the parent mirror/replacing vdev only has one child,
4679 * the parent is no longer needed. Remove it from the tree.
4680 */
4681 if (pvd->vdev_children == 1) {
4682 if (pvd->vdev_ops == &vdev_spare_ops)
4683 cvd->vdev_unspare = B_FALSE;
4684 vdev_remove_parent(cvd);
4685 cvd->vdev_resilvering = B_FALSE;
4686 }
4687
4688
4689 /*
4690 * We don't set tvd until now because the parent we just removed
4691 * may have been the previous top-level vdev.
4692 */
4693 tvd = cvd->vdev_top;
4694 ASSERT(tvd->vdev_parent == rvd);
4695
4696 /*
4697 * Reevaluate the parent vdev state.
4698 */
4699 vdev_propagate_state(cvd);
4700
4701 /*
4702 * If the 'autoexpand' property is set on the pool then automatically
4703 * try to expand the size of the pool. For example if the device we
4704 * just detached was smaller than the others, it may be possible to
4705 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4706 * first so that we can obtain the updated sizes of the leaf vdevs.
4707 */
4708 if (spa->spa_autoexpand) {
4709 vdev_reopen(tvd);
4710 vdev_expand(tvd, txg);
4711 }
4712
4713 vdev_config_dirty(tvd);
4714
4715 /*
4716 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4717 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4718 * But first make sure we're not on any *other* txg's DTL list, to
4719 * prevent vd from being accessed after it's freed.
4720 */
4721 vdpath = spa_strdup(vd->vdev_path);
4722 for (int t = 0; t < TXG_SIZE; t++)
4723 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4724 vd->vdev_detached = B_TRUE;
4725 vdev_dirty(tvd, VDD_DTL, vd, txg);
4726
4727 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4728
4729 /* hang on to the spa before we release the lock */
4730 spa_open_ref(spa, FTAG);
4731
4732 error = spa_vdev_exit(spa, vd, txg, 0);
4733
4734 spa_history_log_internal(spa, "detach", NULL,
4735 "vdev=%s", vdpath);
4736 spa_strfree(vdpath);
4737
4738 /*
4739 * If this was the removal of the original device in a hot spare vdev,
4740 * then we want to go through and remove the device from the hot spare
4741 * list of every other pool.
4742 */
4743 if (unspare) {
4744 spa_t *altspa = NULL;
4745
4746 mutex_enter(&spa_namespace_lock);
4747 while ((altspa = spa_next(altspa)) != NULL) {
4748 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4749 altspa == spa)
4750 continue;
4751
4752 spa_open_ref(altspa, FTAG);
4753 mutex_exit(&spa_namespace_lock);
4754 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4755 mutex_enter(&spa_namespace_lock);
4756 spa_close(altspa, FTAG);
4757 }
4758 mutex_exit(&spa_namespace_lock);
4759
4760 /* search the rest of the vdevs for spares to remove */
4761 spa_vdev_resilver_done(spa);
4762 }
4763
4764 /* all done with the spa; OK to release */
4765 mutex_enter(&spa_namespace_lock);
4766 spa_close(spa, FTAG);
4767 mutex_exit(&spa_namespace_lock);
4768
4769 return (error);
4770 }
4771
4772 /*
4773 * Split a set of devices from their mirrors, and create a new pool from them.
4774 */
4775 int
4776 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4777 nvlist_t *props, boolean_t exp)
4778 {
4779 int error = 0;
4780 uint64_t txg, *glist;
4781 spa_t *newspa;
4782 uint_t c, children, lastlog;
4783 nvlist_t **child, *nvl, *tmp;
4784 dmu_tx_t *tx;
4785 char *altroot = NULL;
4786 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4787 boolean_t activate_slog;
4788
4789 ASSERT(spa_writeable(spa));
4790
4791 txg = spa_vdev_enter(spa);
4792
4793 /* clear the log and flush everything up to now */
4794 activate_slog = spa_passivate_log(spa);
4795 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4796 error = spa_offline_log(spa);
4797 txg = spa_vdev_config_enter(spa);
4798
4799 if (activate_slog)
4800 spa_activate_log(spa);
4801
4802 if (error != 0)
4803 return (spa_vdev_exit(spa, NULL, txg, error));
4804
4805 /* check new spa name before going any further */
4806 if (spa_lookup(newname) != NULL)
4807 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4808
4809 /*
4810 * scan through all the children to ensure they're all mirrors
4811 */
4812 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4813 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4814 &children) != 0)
4815 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4816
4817 /* first, check to ensure we've got the right child count */
4818 rvd = spa->spa_root_vdev;
4819 lastlog = 0;
4820 for (c = 0; c < rvd->vdev_children; c++) {
4821 vdev_t *vd = rvd->vdev_child[c];
4822
4823 /* don't count the holes & logs as children */
4824 if (vd->vdev_islog || vd->vdev_ishole) {
4825 if (lastlog == 0)
4826 lastlog = c;
4827 continue;
4828 }
4829
4830 lastlog = 0;
4831 }
4832 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4833 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4834
4835 /* next, ensure no spare or cache devices are part of the split */
4836 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4837 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4838 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4839
4840 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4841 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4842
4843 /* then, loop over each vdev and validate it */
4844 for (c = 0; c < children; c++) {
4845 uint64_t is_hole = 0;
4846
4847 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4848 &is_hole);
4849
4850 if (is_hole != 0) {
4851 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4852 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4853 continue;
4854 } else {
4855 error = SET_ERROR(EINVAL);
4856 break;
4857 }
4858 }
4859
4860 /* which disk is going to be split? */
4861 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4862 &glist[c]) != 0) {
4863 error = SET_ERROR(EINVAL);
4864 break;
4865 }
4866
4867 /* look it up in the spa */
4868 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4869 if (vml[c] == NULL) {
4870 error = SET_ERROR(ENODEV);
4871 break;
4872 }
4873
4874 /* make sure there's nothing stopping the split */
4875 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4876 vml[c]->vdev_islog ||
4877 vml[c]->vdev_ishole ||
4878 vml[c]->vdev_isspare ||
4879 vml[c]->vdev_isl2cache ||
4880 !vdev_writeable(vml[c]) ||
4881 vml[c]->vdev_children != 0 ||
4882 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4883 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4884 error = SET_ERROR(EINVAL);
4885 break;
4886 }
4887
4888 if (vdev_dtl_required(vml[c])) {
4889 error = SET_ERROR(EBUSY);
4890 break;
4891 }
4892
4893 /* we need certain info from the top level */
4894 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4895 vml[c]->vdev_top->vdev_ms_array) == 0);
4896 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4897 vml[c]->vdev_top->vdev_ms_shift) == 0);
4898 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4899 vml[c]->vdev_top->vdev_asize) == 0);
4900 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4901 vml[c]->vdev_top->vdev_ashift) == 0);
4902 }
4903
4904 if (error != 0) {
4905 kmem_free(vml, children * sizeof (vdev_t *));
4906 kmem_free(glist, children * sizeof (uint64_t));
4907 return (spa_vdev_exit(spa, NULL, txg, error));
4908 }
4909
4910 /* stop writers from using the disks */
4911 for (c = 0; c < children; c++) {
4912 if (vml[c] != NULL)
4913 vml[c]->vdev_offline = B_TRUE;
4914 }
4915 vdev_reopen(spa->spa_root_vdev);
4916
4917 /*
4918 * Temporarily record the splitting vdevs in the spa config. This
4919 * will disappear once the config is regenerated.
4920 */
4921 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4922 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4923 glist, children) == 0);
4924 kmem_free(glist, children * sizeof (uint64_t));
4925
4926 mutex_enter(&spa->spa_props_lock);
4927 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4928 nvl) == 0);
4929 mutex_exit(&spa->spa_props_lock);
4930 spa->spa_config_splitting = nvl;
4931 vdev_config_dirty(spa->spa_root_vdev);
4932
4933 /* configure and create the new pool */
4934 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4935 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4936 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4937 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4938 spa_version(spa)) == 0);
4939 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4940 spa->spa_config_txg) == 0);
4941 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4942 spa_generate_guid(NULL)) == 0);
4943 (void) nvlist_lookup_string(props,
4944 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4945
4946 /* add the new pool to the namespace */
4947 newspa = spa_add(newname, config, altroot);
4948 newspa->spa_config_txg = spa->spa_config_txg;
4949 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4950
4951 /* release the spa config lock, retaining the namespace lock */
4952 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4953
4954 if (zio_injection_enabled)
4955 zio_handle_panic_injection(spa, FTAG, 1);
4956
4957 spa_activate(newspa, spa_mode_global);
4958 spa_async_suspend(newspa);
4959
4960 /* create the new pool from the disks of the original pool */
4961 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4962 if (error)
4963 goto out;
4964
4965 /* if that worked, generate a real config for the new pool */
4966 if (newspa->spa_root_vdev != NULL) {
4967 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4968 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4969 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4970 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4971 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4972 B_TRUE));
4973 }
4974
4975 /* set the props */
4976 if (props != NULL) {
4977 spa_configfile_set(newspa, props, B_FALSE);
4978 error = spa_prop_set(newspa, props);
4979 if (error)
4980 goto out;
4981 }
4982
4983 /* flush everything */
4984 txg = spa_vdev_config_enter(newspa);
4985 vdev_config_dirty(newspa->spa_root_vdev);
4986 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4987
4988 if (zio_injection_enabled)
4989 zio_handle_panic_injection(spa, FTAG, 2);
4990
4991 spa_async_resume(newspa);
4992
4993 /* finally, update the original pool's config */
4994 txg = spa_vdev_config_enter(spa);
4995 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4996 error = dmu_tx_assign(tx, TXG_WAIT);
4997 if (error != 0)
4998 dmu_tx_abort(tx);
4999 for (c = 0; c < children; c++) {
5000 if (vml[c] != NULL) {
5001 vdev_split(vml[c]);
5002 if (error == 0)
5003 spa_history_log_internal(spa, "detach", tx,
5004 "vdev=%s", vml[c]->vdev_path);
5005 vdev_free(vml[c]);
5006 }
5007 }
5008 vdev_config_dirty(spa->spa_root_vdev);
5009 spa->spa_config_splitting = NULL;
5010 nvlist_free(nvl);
5011 if (error == 0)
5012 dmu_tx_commit(tx);
5013 (void) spa_vdev_exit(spa, NULL, txg, 0);
5014
5015 if (zio_injection_enabled)
5016 zio_handle_panic_injection(spa, FTAG, 3);
5017
5018 /* split is complete; log a history record */
5019 spa_history_log_internal(newspa, "split", NULL,
5020 "from pool %s", spa_name(spa));
5021
5022 kmem_free(vml, children * sizeof (vdev_t *));
5023
5024 /* if we're not going to mount the filesystems in userland, export */
5025 if (exp)
5026 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5027 B_FALSE, B_FALSE);
5028
5029 return (error);
5030
5031 out:
5032 spa_unload(newspa);
5033 spa_deactivate(newspa);
5034 spa_remove(newspa);
5035
5036 txg = spa_vdev_config_enter(spa);
5037
5038 /* re-online all offlined disks */
5039 for (c = 0; c < children; c++) {
5040 if (vml[c] != NULL)
5041 vml[c]->vdev_offline = B_FALSE;
5042 }
5043 vdev_reopen(spa->spa_root_vdev);
5044
5045 nvlist_free(spa->spa_config_splitting);
5046 spa->spa_config_splitting = NULL;
5047 (void) spa_vdev_exit(spa, NULL, txg, error);
5048
5049 kmem_free(vml, children * sizeof (vdev_t *));
5050 return (error);
5051 }
5052
5053 static nvlist_t *
5054 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5055 {
5056 for (int i = 0; i < count; i++) {
5057 uint64_t guid;
5058
5059 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5060 &guid) == 0);
5061
5062 if (guid == target_guid)
5063 return (nvpp[i]);
5064 }
5065
5066 return (NULL);
5067 }
5068
5069 static void
5070 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5071 nvlist_t *dev_to_remove)
5072 {
5073 nvlist_t **newdev = NULL;
5074
5075 if (count > 1)
5076 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5077
5078 for (int i = 0, j = 0; i < count; i++) {
5079 if (dev[i] == dev_to_remove)
5080 continue;
5081 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5082 }
5083
5084 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5085 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5086
5087 for (int i = 0; i < count - 1; i++)
5088 nvlist_free(newdev[i]);
5089
5090 if (count > 1)
5091 kmem_free(newdev, (count - 1) * sizeof (void *));
5092 }
5093
5094 /*
5095 * Evacuate the device.
5096 */
5097 static int
5098 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5099 {
5100 uint64_t txg;
5101 int error = 0;
5102
5103 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5104 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5105 ASSERT(vd == vd->vdev_top);
5106
5107 /*
5108 * Evacuate the device. We don't hold the config lock as writer
5109 * since we need to do I/O but we do keep the
5110 * spa_namespace_lock held. Once this completes the device
5111 * should no longer have any blocks allocated on it.
5112 */
5113 if (vd->vdev_islog) {
5114 if (vd->vdev_stat.vs_alloc != 0)
5115 error = spa_offline_log(spa);
5116 } else {
5117 error = SET_ERROR(ENOTSUP);
5118 }
5119
5120 if (error)
5121 return (error);
5122
5123 /*
5124 * The evacuation succeeded. Remove any remaining MOS metadata
5125 * associated with this vdev, and wait for these changes to sync.
5126 */
5127 ASSERT0(vd->vdev_stat.vs_alloc);
5128 txg = spa_vdev_config_enter(spa);
5129 vd->vdev_removing = B_TRUE;
5130 vdev_dirty(vd, 0, NULL, txg);
5131 vdev_config_dirty(vd);
5132 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5133
5134 return (0);
5135 }
5136
5137 /*
5138 * Complete the removal by cleaning up the namespace.
5139 */
5140 static void
5141 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5142 {
5143 vdev_t *rvd = spa->spa_root_vdev;
5144 uint64_t id = vd->vdev_id;
5145 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5146
5147 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5148 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5149 ASSERT(vd == vd->vdev_top);
5150
5151 /*
5152 * Only remove any devices which are empty.
5153 */
5154 if (vd->vdev_stat.vs_alloc != 0)
5155 return;
5156
5157 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5158
5159 if (list_link_active(&vd->vdev_state_dirty_node))
5160 vdev_state_clean(vd);
5161 if (list_link_active(&vd->vdev_config_dirty_node))
5162 vdev_config_clean(vd);
5163
5164 vdev_free(vd);
5165
5166 if (last_vdev) {
5167 vdev_compact_children(rvd);
5168 } else {
5169 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5170 vdev_add_child(rvd, vd);
5171 }
5172 vdev_config_dirty(rvd);
5173
5174 /*
5175 * Reassess the health of our root vdev.
5176 */
5177 vdev_reopen(rvd);
5178 }
5179
5180 /*
5181 * Remove a device from the pool -
5182 *
5183 * Removing a device from the vdev namespace requires several steps
5184 * and can take a significant amount of time. As a result we use
5185 * the spa_vdev_config_[enter/exit] functions which allow us to
5186 * grab and release the spa_config_lock while still holding the namespace
5187 * lock. During each step the configuration is synced out.
5188 *
5189 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5190 * devices.
5191 */
5192 int
5193 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5194 {
5195 vdev_t *vd;
5196 metaslab_group_t *mg;
5197 nvlist_t **spares, **l2cache, *nv;
5198 uint64_t txg = 0;
5199 uint_t nspares, nl2cache;
5200 int error = 0;
5201 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5202
5203 ASSERT(spa_writeable(spa));
5204
5205 if (!locked)
5206 txg = spa_vdev_enter(spa);
5207
5208 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5209
5210 if (spa->spa_spares.sav_vdevs != NULL &&
5211 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5212 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5213 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5214 /*
5215 * Only remove the hot spare if it's not currently in use
5216 * in this pool.
5217 */
5218 if (vd == NULL || unspare) {
5219 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5220 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5221 spa_load_spares(spa);
5222 spa->spa_spares.sav_sync = B_TRUE;
5223 } else {
5224 error = SET_ERROR(EBUSY);
5225 }
5226 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5227 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5228 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5229 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5230 /*
5231 * Cache devices can always be removed.
5232 */
5233 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5234 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5235 spa_load_l2cache(spa);
5236 spa->spa_l2cache.sav_sync = B_TRUE;
5237 } else if (vd != NULL && vd->vdev_islog) {
5238 ASSERT(!locked);
5239 ASSERT(vd == vd->vdev_top);
5240
5241 /*
5242 * XXX - Once we have bp-rewrite this should
5243 * become the common case.
5244 */
5245
5246 mg = vd->vdev_mg;
5247
5248 /*
5249 * Stop allocating from this vdev.
5250 */
5251 metaslab_group_passivate(mg);
5252
5253 /*
5254 * Wait for the youngest allocations and frees to sync,
5255 * and then wait for the deferral of those frees to finish.
5256 */
5257 spa_vdev_config_exit(spa, NULL,
5258 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5259
5260 /*
5261 * Attempt to evacuate the vdev.
5262 */
5263 error = spa_vdev_remove_evacuate(spa, vd);
5264
5265 txg = spa_vdev_config_enter(spa);
5266
5267 /*
5268 * If we couldn't evacuate the vdev, unwind.
5269 */
5270 if (error) {
5271 metaslab_group_activate(mg);
5272 return (spa_vdev_exit(spa, NULL, txg, error));
5273 }
5274
5275 /*
5276 * Clean up the vdev namespace.
5277 */
5278 spa_vdev_remove_from_namespace(spa, vd);
5279
5280 } else if (vd != NULL) {
5281 /*
5282 * Normal vdevs cannot be removed (yet).
5283 */
5284 error = SET_ERROR(ENOTSUP);
5285 } else {
5286 /*
5287 * There is no vdev of any kind with the specified guid.
5288 */
5289 error = SET_ERROR(ENOENT);
5290 }
5291
5292 if (!locked)
5293 return (spa_vdev_exit(spa, NULL, txg, error));
5294
5295 return (error);
5296 }
5297
5298 /*
5299 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5300 * currently spared, so we can detach it.
5301 */
5302 static vdev_t *
5303 spa_vdev_resilver_done_hunt(vdev_t *vd)
5304 {
5305 vdev_t *newvd, *oldvd;
5306
5307 for (int c = 0; c < vd->vdev_children; c++) {
5308 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5309 if (oldvd != NULL)
5310 return (oldvd);
5311 }
5312
5313 /*
5314 * Check for a completed replacement. We always consider the first
5315 * vdev in the list to be the oldest vdev, and the last one to be
5316 * the newest (see spa_vdev_attach() for how that works). In
5317 * the case where the newest vdev is faulted, we will not automatically
5318 * remove it after a resilver completes. This is OK as it will require
5319 * user intervention to determine which disk the admin wishes to keep.
5320 */
5321 if (vd->vdev_ops == &vdev_replacing_ops) {
5322 ASSERT(vd->vdev_children > 1);
5323
5324 newvd = vd->vdev_child[vd->vdev_children - 1];
5325 oldvd = vd->vdev_child[0];
5326
5327 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5328 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5329 !vdev_dtl_required(oldvd))
5330 return (oldvd);
5331 }
5332
5333 /*
5334 * Check for a completed resilver with the 'unspare' flag set.
5335 */
5336 if (vd->vdev_ops == &vdev_spare_ops) {
5337 vdev_t *first = vd->vdev_child[0];
5338 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5339
5340 if (last->vdev_unspare) {
5341 oldvd = first;
5342 newvd = last;
5343 } else if (first->vdev_unspare) {
5344 oldvd = last;
5345 newvd = first;
5346 } else {
5347 oldvd = NULL;
5348 }
5349
5350 if (oldvd != NULL &&
5351 vdev_dtl_empty(newvd, DTL_MISSING) &&
5352 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5353 !vdev_dtl_required(oldvd))
5354 return (oldvd);
5355
5356 /*
5357 * If there are more than two spares attached to a disk,
5358 * and those spares are not required, then we want to
5359 * attempt to free them up now so that they can be used
5360 * by other pools. Once we're back down to a single
5361 * disk+spare, we stop removing them.
5362 */
5363 if (vd->vdev_children > 2) {
5364 newvd = vd->vdev_child[1];
5365
5366 if (newvd->vdev_isspare && last->vdev_isspare &&
5367 vdev_dtl_empty(last, DTL_MISSING) &&
5368 vdev_dtl_empty(last, DTL_OUTAGE) &&
5369 !vdev_dtl_required(newvd))
5370 return (newvd);
5371 }
5372 }
5373
5374 return (NULL);
5375 }
5376
5377 static void
5378 spa_vdev_resilver_done(spa_t *spa)
5379 {
5380 vdev_t *vd, *pvd, *ppvd;
5381 uint64_t guid, sguid, pguid, ppguid;
5382
5383 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5384
5385 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5386 pvd = vd->vdev_parent;
5387 ppvd = pvd->vdev_parent;
5388 guid = vd->vdev_guid;
5389 pguid = pvd->vdev_guid;
5390 ppguid = ppvd->vdev_guid;
5391 sguid = 0;
5392 /*
5393 * If we have just finished replacing a hot spared device, then
5394 * we need to detach the parent's first child (the original hot
5395 * spare) as well.
5396 */
5397 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5398 ppvd->vdev_children == 2) {
5399 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5400 sguid = ppvd->vdev_child[1]->vdev_guid;
5401 }
5402 spa_config_exit(spa, SCL_ALL, FTAG);
5403 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5404 return;
5405 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5406 return;
5407 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5408 }
5409
5410 spa_config_exit(spa, SCL_ALL, FTAG);
5411 }
5412
5413 /*
5414 * Update the stored path or FRU for this vdev.
5415 */
5416 int
5417 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5418 boolean_t ispath)
5419 {
5420 vdev_t *vd;
5421 boolean_t sync = B_FALSE;
5422
5423 ASSERT(spa_writeable(spa));
5424
5425 spa_vdev_state_enter(spa, SCL_ALL);
5426
5427 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5428 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5429
5430 if (!vd->vdev_ops->vdev_op_leaf)
5431 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5432
5433 if (ispath) {
5434 if (strcmp(value, vd->vdev_path) != 0) {
5435 spa_strfree(vd->vdev_path);
5436 vd->vdev_path = spa_strdup(value);
5437 sync = B_TRUE;
5438 }
5439 } else {
5440 if (vd->vdev_fru == NULL) {
5441 vd->vdev_fru = spa_strdup(value);
5442 sync = B_TRUE;
5443 } else if (strcmp(value, vd->vdev_fru) != 0) {
5444 spa_strfree(vd->vdev_fru);
5445 vd->vdev_fru = spa_strdup(value);
5446 sync = B_TRUE;
5447 }
5448 }
5449
5450 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5451 }
5452
5453 int
5454 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5455 {
5456 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5457 }
5458
5459 int
5460 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5461 {
5462 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5463 }
5464
5465 /*
5466 * ==========================================================================
5467 * SPA Scanning
5468 * ==========================================================================
5469 */
5470
5471 int
5472 spa_scan_stop(spa_t *spa)
5473 {
5474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5475 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5476 return (SET_ERROR(EBUSY));
5477 return (dsl_scan_cancel(spa->spa_dsl_pool));
5478 }
5479
5480 int
5481 spa_scan(spa_t *spa, pool_scan_func_t func)
5482 {
5483 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5484
5485 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5486 return (SET_ERROR(ENOTSUP));
5487
5488 /*
5489 * If a resilver was requested, but there is no DTL on a
5490 * writeable leaf device, we have nothing to do.
5491 */
5492 if (func == POOL_SCAN_RESILVER &&
5493 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5494 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5495 return (0);
5496 }
5497
5498 return (dsl_scan(spa->spa_dsl_pool, func));
5499 }
5500
5501 /*
5502 * ==========================================================================
5503 * SPA async task processing
5504 * ==========================================================================
5505 */
5506
5507 static void
5508 spa_async_remove(spa_t *spa, vdev_t *vd)
5509 {
5510 if (vd->vdev_remove_wanted) {
5511 vd->vdev_remove_wanted = B_FALSE;
5512 vd->vdev_delayed_close = B_FALSE;
5513 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5514
5515 /*
5516 * We want to clear the stats, but we don't want to do a full
5517 * vdev_clear() as that will cause us to throw away
5518 * degraded/faulted state as well as attempt to reopen the
5519 * device, all of which is a waste.
5520 */
5521 vd->vdev_stat.vs_read_errors = 0;
5522 vd->vdev_stat.vs_write_errors = 0;
5523 vd->vdev_stat.vs_checksum_errors = 0;
5524
5525 vdev_state_dirty(vd->vdev_top);
5526 }
5527
5528 for (int c = 0; c < vd->vdev_children; c++)
5529 spa_async_remove(spa, vd->vdev_child[c]);
5530 }
5531
5532 static void
5533 spa_async_probe(spa_t *spa, vdev_t *vd)
5534 {
5535 if (vd->vdev_probe_wanted) {
5536 vd->vdev_probe_wanted = B_FALSE;
5537 vdev_reopen(vd); /* vdev_open() does the actual probe */
5538 }
5539
5540 for (int c = 0; c < vd->vdev_children; c++)
5541 spa_async_probe(spa, vd->vdev_child[c]);
5542 }
5543
5544 static void
5545 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5546 {
5547 sysevent_id_t eid;
5548 nvlist_t *attr;
5549 char *physpath;
5550
5551 if (!spa->spa_autoexpand)
5552 return;
5553
5554 for (int c = 0; c < vd->vdev_children; c++) {
5555 vdev_t *cvd = vd->vdev_child[c];
5556 spa_async_autoexpand(spa, cvd);
5557 }
5558
5559 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5560 return;
5561
5562 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5563 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5564
5565 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5566 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5567
5568 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5569 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5570
5571 nvlist_free(attr);
5572 kmem_free(physpath, MAXPATHLEN);
5573 }
5574
5575 static void
5576 spa_async_thread(spa_t *spa)
5577 {
5578 int tasks;
5579
5580 ASSERT(spa->spa_sync_on);
5581
5582 mutex_enter(&spa->spa_async_lock);
5583 tasks = spa->spa_async_tasks;
5584 spa->spa_async_tasks = 0;
5585 mutex_exit(&spa->spa_async_lock);
5586
5587 /*
5588 * See if the config needs to be updated.
5589 */
5590 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5591 uint64_t old_space, new_space;
5592
5593 mutex_enter(&spa_namespace_lock);
5594 old_space = metaslab_class_get_space(spa_normal_class(spa));
5595 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5596 new_space = metaslab_class_get_space(spa_normal_class(spa));
5597 mutex_exit(&spa_namespace_lock);
5598
5599 /*
5600 * If the pool grew as a result of the config update,
5601 * then log an internal history event.
5602 */
5603 if (new_space != old_space) {
5604 spa_history_log_internal(spa, "vdev online", NULL,
5605 "pool '%s' size: %llu(+%llu)",
5606 spa_name(spa), new_space, new_space - old_space);
5607 }
5608 }
5609
5610 /*
5611 * See if any devices need to be marked REMOVED.
5612 */
5613 if (tasks & SPA_ASYNC_REMOVE) {
5614 spa_vdev_state_enter(spa, SCL_NONE);
5615 spa_async_remove(spa, spa->spa_root_vdev);
5616 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5617 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5618 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5619 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5620 (void) spa_vdev_state_exit(spa, NULL, 0);
5621 }
5622
5623 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5624 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5625 spa_async_autoexpand(spa, spa->spa_root_vdev);
5626 spa_config_exit(spa, SCL_CONFIG, FTAG);
5627 }
5628
5629 /*
5630 * See if any devices need to be probed.
5631 */
5632 if (tasks & SPA_ASYNC_PROBE) {
5633 spa_vdev_state_enter(spa, SCL_NONE);
5634 spa_async_probe(spa, spa->spa_root_vdev);
5635 (void) spa_vdev_state_exit(spa, NULL, 0);
5636 }
5637
5638 /*
5639 * If any devices are done replacing, detach them.
5640 */
5641 if (tasks & SPA_ASYNC_RESILVER_DONE)
5642 spa_vdev_resilver_done(spa);
5643
5644 /*
5645 * Kick off a resilver.
5646 */
5647 if (tasks & SPA_ASYNC_RESILVER)
5648 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5649
5650 /*
5651 * Let the world know that we're done.
5652 */
5653 mutex_enter(&spa->spa_async_lock);
5654 spa->spa_async_thread = NULL;
5655 cv_broadcast(&spa->spa_async_cv);
5656 mutex_exit(&spa->spa_async_lock);
5657 thread_exit();
5658 }
5659
5660 void
5661 spa_async_suspend(spa_t *spa)
5662 {
5663 mutex_enter(&spa->spa_async_lock);
5664 spa->spa_async_suspended++;
5665 while (spa->spa_async_thread != NULL)
5666 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5667 mutex_exit(&spa->spa_async_lock);
5668 }
5669
5670 void
5671 spa_async_resume(spa_t *spa)
5672 {
5673 mutex_enter(&spa->spa_async_lock);
5674 ASSERT(spa->spa_async_suspended != 0);
5675 spa->spa_async_suspended--;
5676 mutex_exit(&spa->spa_async_lock);
5677 }
5678
5679 static boolean_t
5680 spa_async_tasks_pending(spa_t *spa)
5681 {
5682 uint_t non_config_tasks;
5683 uint_t config_task;
5684 boolean_t config_task_suspended;
5685
5686 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5687 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5688 if (spa->spa_ccw_fail_time == 0) {
5689 config_task_suspended = B_FALSE;
5690 } else {
5691 config_task_suspended =
5692 (gethrtime() - spa->spa_ccw_fail_time) <
5693 (zfs_ccw_retry_interval * NANOSEC);
5694 }
5695
5696 return (non_config_tasks || (config_task && !config_task_suspended));
5697 }
5698
5699 static void
5700 spa_async_dispatch(spa_t *spa)
5701 {
5702 mutex_enter(&spa->spa_async_lock);
5703 if (spa_async_tasks_pending(spa) &&
5704 !spa->spa_async_suspended &&
5705 spa->spa_async_thread == NULL &&
5706 rootdir != NULL)
5707 spa->spa_async_thread = thread_create(NULL, 0,
5708 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5709 mutex_exit(&spa->spa_async_lock);
5710 }
5711
5712 void
5713 spa_async_request(spa_t *spa, int task)
5714 {
5715 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5716 mutex_enter(&spa->spa_async_lock);
5717 spa->spa_async_tasks |= task;
5718 mutex_exit(&spa->spa_async_lock);
5719 }
5720
5721 /*
5722 * ==========================================================================
5723 * SPA syncing routines
5724 * ==========================================================================
5725 */
5726
5727 static int
5728 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5729 {
5730 bpobj_t *bpo = arg;
5731 bpobj_enqueue(bpo, bp, tx);
5732 return (0);
5733 }
5734
5735 static int
5736 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5737 {
5738 zio_t *zio = arg;
5739
5740 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5741 zio->io_flags));
5742 return (0);
5743 }
5744
5745 static void
5746 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5747 {
5748 char *packed = NULL;
5749 size_t bufsize;
5750 size_t nvsize = 0;
5751 dmu_buf_t *db;
5752
5753 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5754
5755 /*
5756 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5757 * information. This avoids the dbuf_will_dirty() path and
5758 * saves us a pre-read to get data we don't actually care about.
5759 */
5760 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5761 packed = kmem_alloc(bufsize, KM_SLEEP);
5762
5763 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5764 KM_SLEEP) == 0);
5765 bzero(packed + nvsize, bufsize - nvsize);
5766
5767 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5768
5769 kmem_free(packed, bufsize);
5770
5771 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5772 dmu_buf_will_dirty(db, tx);
5773 *(uint64_t *)db->db_data = nvsize;
5774 dmu_buf_rele(db, FTAG);
5775 }
5776
5777 static void
5778 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5779 const char *config, const char *entry)
5780 {
5781 nvlist_t *nvroot;
5782 nvlist_t **list;
5783 int i;
5784
5785 if (!sav->sav_sync)
5786 return;
5787
5788 /*
5789 * Update the MOS nvlist describing the list of available devices.
5790 * spa_validate_aux() will have already made sure this nvlist is
5791 * valid and the vdevs are labeled appropriately.
5792 */
5793 if (sav->sav_object == 0) {
5794 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5795 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5796 sizeof (uint64_t), tx);
5797 VERIFY(zap_update(spa->spa_meta_objset,
5798 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5799 &sav->sav_object, tx) == 0);
5800 }
5801
5802 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5803 if (sav->sav_count == 0) {
5804 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5805 } else {
5806 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5807 for (i = 0; i < sav->sav_count; i++)
5808 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5809 B_FALSE, VDEV_CONFIG_L2CACHE);
5810 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5811 sav->sav_count) == 0);
5812 for (i = 0; i < sav->sav_count; i++)
5813 nvlist_free(list[i]);
5814 kmem_free(list, sav->sav_count * sizeof (void *));
5815 }
5816
5817 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5818 nvlist_free(nvroot);
5819
5820 sav->sav_sync = B_FALSE;
5821 }
5822
5823 static void
5824 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5825 {
5826 nvlist_t *config;
5827
5828 if (list_is_empty(&spa->spa_config_dirty_list))
5829 return;
5830
5831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5832
5833 config = spa_config_generate(spa, spa->spa_root_vdev,
5834 dmu_tx_get_txg(tx), B_FALSE);
5835
5836 /*
5837 * If we're upgrading the spa version then make sure that
5838 * the config object gets updated with the correct version.
5839 */
5840 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5841 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5842 spa->spa_uberblock.ub_version);
5843
5844 spa_config_exit(spa, SCL_STATE, FTAG);
5845
5846 if (spa->spa_config_syncing)
5847 nvlist_free(spa->spa_config_syncing);
5848 spa->spa_config_syncing = config;
5849
5850 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5851 }
5852
5853 static void
5854 spa_sync_version(void *arg, dmu_tx_t *tx)
5855 {
5856 uint64_t *versionp = arg;
5857 uint64_t version = *versionp;
5858 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5859
5860 /*
5861 * Setting the version is special cased when first creating the pool.
5862 */
5863 ASSERT(tx->tx_txg != TXG_INITIAL);
5864
5865 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5866 ASSERT(version >= spa_version(spa));
5867
5868 spa->spa_uberblock.ub_version = version;
5869 vdev_config_dirty(spa->spa_root_vdev);
5870 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5871 }
5872
5873 /*
5874 * Set zpool properties.
5875 */
5876 static void
5877 spa_sync_props(void *arg, dmu_tx_t *tx)
5878 {
5879 nvlist_t *nvp = arg;
5880 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5881 objset_t *mos = spa->spa_meta_objset;
5882 nvpair_t *elem = NULL;
5883
5884 mutex_enter(&spa->spa_props_lock);
5885
5886 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5887 uint64_t intval;
5888 char *strval, *fname;
5889 zpool_prop_t prop;
5890 const char *propname;
5891 zprop_type_t proptype;
5892 zfeature_info_t *feature;
5893
5894 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5895 case ZPROP_INVAL:
5896 /*
5897 * We checked this earlier in spa_prop_validate().
5898 */
5899 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5900
5901 fname = strchr(nvpair_name(elem), '@') + 1;
5902 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5903
5904 spa_feature_enable(spa, feature, tx);
5905 spa_history_log_internal(spa, "set", tx,
5906 "%s=enabled", nvpair_name(elem));
5907 break;
5908
5909 case ZPOOL_PROP_VERSION:
5910 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5911 /*
5912 * The version is synced seperatly before other
5913 * properties and should be correct by now.
5914 */
5915 ASSERT3U(spa_version(spa), >=, intval);
5916 break;
5917
5918 case ZPOOL_PROP_ALTROOT:
5919 /*
5920 * 'altroot' is a non-persistent property. It should
5921 * have been set temporarily at creation or import time.
5922 */
5923 ASSERT(spa->spa_root != NULL);
5924 break;
5925
5926 case ZPOOL_PROP_READONLY:
5927 case ZPOOL_PROP_CACHEFILE:
5928 /*
5929 * 'readonly' and 'cachefile' are also non-persisitent
5930 * properties.
5931 */
5932 break;
5933 case ZPOOL_PROP_COMMENT:
5934 VERIFY(nvpair_value_string(elem, &strval) == 0);
5935 if (spa->spa_comment != NULL)
5936 spa_strfree(spa->spa_comment);
5937 spa->spa_comment = spa_strdup(strval);
5938 /*
5939 * We need to dirty the configuration on all the vdevs
5940 * so that their labels get updated. It's unnecessary
5941 * to do this for pool creation since the vdev's
5942 * configuratoin has already been dirtied.
5943 */
5944 if (tx->tx_txg != TXG_INITIAL)
5945 vdev_config_dirty(spa->spa_root_vdev);
5946 spa_history_log_internal(spa, "set", tx,
5947 "%s=%s", nvpair_name(elem), strval);
5948 break;
5949 default:
5950 /*
5951 * Set pool property values in the poolprops mos object.
5952 */
5953 if (spa->spa_pool_props_object == 0) {
5954 spa->spa_pool_props_object =
5955 zap_create_link(mos, DMU_OT_POOL_PROPS,
5956 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5957 tx);
5958 }
5959
5960 /* normalize the property name */
5961 propname = zpool_prop_to_name(prop);
5962 proptype = zpool_prop_get_type(prop);
5963
5964 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5965 ASSERT(proptype == PROP_TYPE_STRING);
5966 VERIFY(nvpair_value_string(elem, &strval) == 0);
5967 VERIFY(zap_update(mos,
5968 spa->spa_pool_props_object, propname,
5969 1, strlen(strval) + 1, strval, tx) == 0);
5970 spa_history_log_internal(spa, "set", tx,
5971 "%s=%s", nvpair_name(elem), strval);
5972 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5973 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5974
5975 if (proptype == PROP_TYPE_INDEX) {
5976 const char *unused;
5977 VERIFY(zpool_prop_index_to_string(
5978 prop, intval, &unused) == 0);
5979 }
5980 VERIFY(zap_update(mos,
5981 spa->spa_pool_props_object, propname,
5982 8, 1, &intval, tx) == 0);
5983 spa_history_log_internal(spa, "set", tx,
5984 "%s=%lld", nvpair_name(elem), intval);
5985 } else {
5986 ASSERT(0); /* not allowed */
5987 }
5988
5989 switch (prop) {
5990 case ZPOOL_PROP_DELEGATION:
5991 spa->spa_delegation = intval;
5992 break;
5993 case ZPOOL_PROP_BOOTFS:
5994 spa->spa_bootfs = intval;
5995 break;
5996 case ZPOOL_PROP_FAILUREMODE:
5997 spa->spa_failmode = intval;
5998 break;
5999 case ZPOOL_PROP_AUTOEXPAND:
6000 spa->spa_autoexpand = intval;
6001 if (tx->tx_txg != TXG_INITIAL)
6002 spa_async_request(spa,
6003 SPA_ASYNC_AUTOEXPAND);
6004 break;
6005 case ZPOOL_PROP_DEDUPDITTO:
6006 spa->spa_dedup_ditto = intval;
6007 break;
6008 default:
6009 break;
6010 }
6011 }
6012
6013 }
6014
6015 mutex_exit(&spa->spa_props_lock);
6016 }
6017
6018 /*
6019 * Perform one-time upgrade on-disk changes. spa_version() does not
6020 * reflect the new version this txg, so there must be no changes this
6021 * txg to anything that the upgrade code depends on after it executes.
6022 * Therefore this must be called after dsl_pool_sync() does the sync
6023 * tasks.
6024 */
6025 static void
6026 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6027 {
6028 dsl_pool_t *dp = spa->spa_dsl_pool;
6029
6030 ASSERT(spa->spa_sync_pass == 1);
6031
6032 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6033
6034 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6035 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6036 dsl_pool_create_origin(dp, tx);
6037
6038 /* Keeping the origin open increases spa_minref */
6039 spa->spa_minref += 3;
6040 }
6041
6042 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6043 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6044 dsl_pool_upgrade_clones(dp, tx);
6045 }
6046
6047 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6048 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6049 dsl_pool_upgrade_dir_clones(dp, tx);
6050
6051 /* Keeping the freedir open increases spa_minref */
6052 spa->spa_minref += 3;
6053 }
6054
6055 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6056 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6057 spa_feature_create_zap_objects(spa, tx);
6058 }
6059 rrw_exit(&dp->dp_config_rwlock, FTAG);
6060 }
6061
6062 /*
6063 * Sync the specified transaction group. New blocks may be dirtied as
6064 * part of the process, so we iterate until it converges.
6065 */
6066 void
6067 spa_sync(spa_t *spa, uint64_t txg)
6068 {
6069 dsl_pool_t *dp = spa->spa_dsl_pool;
6070 objset_t *mos = spa->spa_meta_objset;
6071 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6072 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6073 vdev_t *rvd = spa->spa_root_vdev;
6074 vdev_t *vd;
6075 dmu_tx_t *tx;
6076 int error;
6077
6078 VERIFY(spa_writeable(spa));
6079
6080 /*
6081 * Lock out configuration changes.
6082 */
6083 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6084
6085 spa->spa_syncing_txg = txg;
6086 spa->spa_sync_pass = 0;
6087
6088 /*
6089 * If there are any pending vdev state changes, convert them
6090 * into config changes that go out with this transaction group.
6091 */
6092 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6093 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6094 /*
6095 * We need the write lock here because, for aux vdevs,
6096 * calling vdev_config_dirty() modifies sav_config.
6097 * This is ugly and will become unnecessary when we
6098 * eliminate the aux vdev wart by integrating all vdevs
6099 * into the root vdev tree.
6100 */
6101 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6102 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6103 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6104 vdev_state_clean(vd);
6105 vdev_config_dirty(vd);
6106 }
6107 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6108 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6109 }
6110 spa_config_exit(spa, SCL_STATE, FTAG);
6111
6112 tx = dmu_tx_create_assigned(dp, txg);
6113
6114 spa->spa_sync_starttime = gethrtime();
6115 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6116 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6117
6118 /*
6119 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6120 * set spa_deflate if we have no raid-z vdevs.
6121 */
6122 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6123 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6124 int i;
6125
6126 for (i = 0; i < rvd->vdev_children; i++) {
6127 vd = rvd->vdev_child[i];
6128 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6129 break;
6130 }
6131 if (i == rvd->vdev_children) {
6132 spa->spa_deflate = TRUE;
6133 VERIFY(0 == zap_add(spa->spa_meta_objset,
6134 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6135 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6136 }
6137 }
6138
6139 /*
6140 * If anything has changed in this txg, or if someone is waiting
6141 * for this txg to sync (eg, spa_vdev_remove()), push the
6142 * deferred frees from the previous txg. If not, leave them
6143 * alone so that we don't generate work on an otherwise idle
6144 * system.
6145 */
6146 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6147 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6148 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6149 ((dsl_scan_active(dp->dp_scan) ||
6150 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6151 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6152 VERIFY3U(bpobj_iterate(defer_bpo,
6153 spa_free_sync_cb, zio, tx), ==, 0);
6154 VERIFY0(zio_wait(zio));
6155 }
6156
6157 /*
6158 * Iterate to convergence.
6159 */
6160 do {
6161 int pass = ++spa->spa_sync_pass;
6162
6163 spa_sync_config_object(spa, tx);
6164 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6165 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6166 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6167 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6168 spa_errlog_sync(spa, txg);
6169 dsl_pool_sync(dp, txg);
6170
6171 if (pass < zfs_sync_pass_deferred_free) {
6172 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6173 bplist_iterate(free_bpl, spa_free_sync_cb,
6174 zio, tx);
6175 VERIFY(zio_wait(zio) == 0);
6176 } else {
6177 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6178 defer_bpo, tx);
6179 }
6180
6181 ddt_sync(spa, txg);
6182 dsl_scan_sync(dp, tx);
6183
6184 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6185 vdev_sync(vd, txg);
6186
6187 if (pass == 1)
6188 spa_sync_upgrades(spa, tx);
6189
6190 } while (dmu_objset_is_dirty(mos, txg));
6191
6192 /*
6193 * Rewrite the vdev configuration (which includes the uberblock)
6194 * to commit the transaction group.
6195 *
6196 * If there are no dirty vdevs, we sync the uberblock to a few
6197 * random top-level vdevs that are known to be visible in the
6198 * config cache (see spa_vdev_add() for a complete description).
6199 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6200 */
6201 for (;;) {
6202 /*
6203 * We hold SCL_STATE to prevent vdev open/close/etc.
6204 * while we're attempting to write the vdev labels.
6205 */
6206 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6207
6208 if (list_is_empty(&spa->spa_config_dirty_list)) {
6209 vdev_t *svd[SPA_DVAS_PER_BP];
6210 int svdcount = 0;
6211 int children = rvd->vdev_children;
6212 int c0 = spa_get_random(children);
6213
6214 for (int c = 0; c < children; c++) {
6215 vd = rvd->vdev_child[(c0 + c) % children];
6216 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6217 continue;
6218 svd[svdcount++] = vd;
6219 if (svdcount == SPA_DVAS_PER_BP)
6220 break;
6221 }
6222 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6223 if (error != 0)
6224 error = vdev_config_sync(svd, svdcount, txg,
6225 B_TRUE);
6226 } else {
6227 error = vdev_config_sync(rvd->vdev_child,
6228 rvd->vdev_children, txg, B_FALSE);
6229 if (error != 0)
6230 error = vdev_config_sync(rvd->vdev_child,
6231 rvd->vdev_children, txg, B_TRUE);
6232 }
6233
6234 if (error == 0)
6235 spa->spa_last_synced_guid = rvd->vdev_guid;
6236
6237 spa_config_exit(spa, SCL_STATE, FTAG);
6238
6239 if (error == 0)
6240 break;
6241 zio_suspend(spa, NULL);
6242 zio_resume_wait(spa);
6243 }
6244 dmu_tx_commit(tx);
6245
6246 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6247
6248 /*
6249 * Clear the dirty config list.
6250 */
6251 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6252 vdev_config_clean(vd);
6253
6254 /*
6255 * Now that the new config has synced transactionally,
6256 * let it become visible to the config cache.
6257 */
6258 if (spa->spa_config_syncing != NULL) {
6259 spa_config_set(spa, spa->spa_config_syncing);
6260 spa->spa_config_txg = txg;
6261 spa->spa_config_syncing = NULL;
6262 }
6263
6264 spa->spa_ubsync = spa->spa_uberblock;
6265
6266 dsl_pool_sync_done(dp, txg);
6267
6268 /*
6269 * Update usable space statistics.
6270 */
6271 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6272 vdev_sync_done(vd, txg);
6273
6274 spa_update_dspace(spa);
6275
6276 /*
6277 * It had better be the case that we didn't dirty anything
6278 * since vdev_config_sync().
6279 */
6280 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6281 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6282 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6283
6284 spa->spa_sync_pass = 0;
6285
6286 spa_config_exit(spa, SCL_CONFIG, FTAG);
6287
6288 spa_handle_ignored_writes(spa);
6289
6290 /*
6291 * If any async tasks have been requested, kick them off.
6292 */
6293 spa_async_dispatch(spa);
6294 }
6295
6296 /*
6297 * Sync all pools. We don't want to hold the namespace lock across these
6298 * operations, so we take a reference on the spa_t and drop the lock during the
6299 * sync.
6300 */
6301 void
6302 spa_sync_allpools(void)
6303 {
6304 spa_t *spa = NULL;
6305 mutex_enter(&spa_namespace_lock);
6306 while ((spa = spa_next(spa)) != NULL) {
6307 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6308 !spa_writeable(spa) || spa_suspended(spa))
6309 continue;
6310 spa_open_ref(spa, FTAG);
6311 mutex_exit(&spa_namespace_lock);
6312 txg_wait_synced(spa_get_dsl(spa), 0);
6313 mutex_enter(&spa_namespace_lock);
6314 spa_close(spa, FTAG);
6315 }
6316 mutex_exit(&spa_namespace_lock);
6317 }
6318
6319 /*
6320 * ==========================================================================
6321 * Miscellaneous routines
6322 * ==========================================================================
6323 */
6324
6325 /*
6326 * Remove all pools in the system.
6327 */
6328 void
6329 spa_evict_all(void)
6330 {
6331 spa_t *spa;
6332
6333 /*
6334 * Remove all cached state. All pools should be closed now,
6335 * so every spa in the AVL tree should be unreferenced.
6336 */
6337 mutex_enter(&spa_namespace_lock);
6338 while ((spa = spa_next(NULL)) != NULL) {
6339 /*
6340 * Stop async tasks. The async thread may need to detach
6341 * a device that's been replaced, which requires grabbing
6342 * spa_namespace_lock, so we must drop it here.
6343 */
6344 spa_open_ref(spa, FTAG);
6345 mutex_exit(&spa_namespace_lock);
6346 spa_async_suspend(spa);
6347 mutex_enter(&spa_namespace_lock);
6348 spa_close(spa, FTAG);
6349
6350 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6351 spa_unload(spa);
6352 spa_deactivate(spa);
6353 }
6354 spa_remove(spa);
6355 }
6356 mutex_exit(&spa_namespace_lock);
6357 }
6358
6359 vdev_t *
6360 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6361 {
6362 vdev_t *vd;
6363 int i;
6364
6365 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6366 return (vd);
6367
6368 if (aux) {
6369 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6370 vd = spa->spa_l2cache.sav_vdevs[i];
6371 if (vd->vdev_guid == guid)
6372 return (vd);
6373 }
6374
6375 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6376 vd = spa->spa_spares.sav_vdevs[i];
6377 if (vd->vdev_guid == guid)
6378 return (vd);
6379 }
6380 }
6381
6382 return (NULL);
6383 }
6384
6385 void
6386 spa_upgrade(spa_t *spa, uint64_t version)
6387 {
6388 ASSERT(spa_writeable(spa));
6389
6390 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6391
6392 /*
6393 * This should only be called for a non-faulted pool, and since a
6394 * future version would result in an unopenable pool, this shouldn't be
6395 * possible.
6396 */
6397 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6398 ASSERT(version >= spa->spa_uberblock.ub_version);
6399
6400 spa->spa_uberblock.ub_version = version;
6401 vdev_config_dirty(spa->spa_root_vdev);
6402
6403 spa_config_exit(spa, SCL_ALL, FTAG);
6404
6405 txg_wait_synced(spa_get_dsl(spa), 0);
6406 }
6407
6408 boolean_t
6409 spa_has_spare(spa_t *spa, uint64_t guid)
6410 {
6411 int i;
6412 uint64_t spareguid;
6413 spa_aux_vdev_t *sav = &spa->spa_spares;
6414
6415 for (i = 0; i < sav->sav_count; i++)
6416 if (sav->sav_vdevs[i]->vdev_guid == guid)
6417 return (B_TRUE);
6418
6419 for (i = 0; i < sav->sav_npending; i++) {
6420 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6421 &spareguid) == 0 && spareguid == guid)
6422 return (B_TRUE);
6423 }
6424
6425 return (B_FALSE);
6426 }
6427
6428 /*
6429 * Check if a pool has an active shared spare device.
6430 * Note: reference count of an active spare is 2, as a spare and as a replace
6431 */
6432 static boolean_t
6433 spa_has_active_shared_spare(spa_t *spa)
6434 {
6435 int i, refcnt;
6436 uint64_t pool;
6437 spa_aux_vdev_t *sav = &spa->spa_spares;
6438
6439 for (i = 0; i < sav->sav_count; i++) {
6440 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6441 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6442 refcnt > 2)
6443 return (B_TRUE);
6444 }
6445
6446 return (B_FALSE);
6447 }
6448
6449 /*
6450 * Post a sysevent corresponding to the given event. The 'name' must be one of
6451 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6452 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6453 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6454 * or zdb as real changes.
6455 */
6456 void
6457 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6458 {
6459 #ifdef _KERNEL
6460 sysevent_t *ev;
6461 sysevent_attr_list_t *attr = NULL;
6462 sysevent_value_t value;
6463 sysevent_id_t eid;
6464
6465 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6466 SE_SLEEP);
6467
6468 value.value_type = SE_DATA_TYPE_STRING;
6469 value.value.sv_string = spa_name(spa);
6470 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6471 goto done;
6472
6473 value.value_type = SE_DATA_TYPE_UINT64;
6474 value.value.sv_uint64 = spa_guid(spa);
6475 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6476 goto done;
6477
6478 if (vd) {
6479 value.value_type = SE_DATA_TYPE_UINT64;
6480 value.value.sv_uint64 = vd->vdev_guid;
6481 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6482 SE_SLEEP) != 0)
6483 goto done;
6484
6485 if (vd->vdev_path) {
6486 value.value_type = SE_DATA_TYPE_STRING;
6487 value.value.sv_string = vd->vdev_path;
6488 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6489 &value, SE_SLEEP) != 0)
6490 goto done;
6491 }
6492 }
6493
6494 if (sysevent_attach_attributes(ev, attr) != 0)
6495 goto done;
6496 attr = NULL;
6497
6498 (void) log_sysevent(ev, SE_SLEEP, &eid);
6499
6500 done:
6501 if (attr)
6502 sysevent_free_attr(attr);
6503 sysevent_free(ev);
6504 #endif
6505 }