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