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) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright 2018 Joyent, Inc.
31 * Copyright (c) 2017 Datto Inc.
32 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
33 */
34
35 /*
36 * SPA: Storage Pool Allocator
37 *
38 * This file contains all the routines used when modifying on-disk SPA state.
39 * This includes opening, importing, destroying, exporting a pool, and syncing a
40 * pool.
41 */
42
43 #include <sys/zfs_context.h>
44 #include <sys/fm/fs/zfs.h>
45 #include <sys/spa_impl.h>
46 #include <sys/zio.h>
47 #include <sys/zio_checksum.h>
48 #include <sys/dmu.h>
49 #include <sys/dmu_tx.h>
50 #include <sys/zap.h>
51 #include <sys/zil.h>
52 #include <sys/ddt.h>
53 #include <sys/vdev_impl.h>
54 #include <sys/vdev_removal.h>
55 #include <sys/vdev_indirect_mapping.h>
56 #include <sys/vdev_indirect_births.h>
57 #include <sys/vdev_initialize.h>
58 #include <sys/metaslab.h>
59 #include <sys/metaslab_impl.h>
60 #include <sys/uberblock_impl.h>
61 #include <sys/txg.h>
62 #include <sys/avl.h>
63 #include <sys/bpobj.h>
64 #include <sys/dmu_traverse.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/unique.h>
67 #include <sys/dsl_pool.h>
68 #include <sys/dsl_dataset.h>
69 #include <sys/dsl_dir.h>
70 #include <sys/dsl_prop.h>
71 #include <sys/dsl_synctask.h>
72 #include <sys/fs/zfs.h>
73 #include <sys/arc.h>
74 #include <sys/callb.h>
75 #include <sys/systeminfo.h>
76 #include <sys/spa_boot.h>
77 #include <sys/zfs_ioctl.h>
78 #include <sys/dsl_scan.h>
79 #include <sys/zfeature.h>
80 #include <sys/dsl_destroy.h>
81 #include <sys/abd.h>
82
83 #ifdef _KERNEL
84 #include <sys/bootprops.h>
85 #include <sys/callb.h>
86 #include <sys/cpupart.h>
87 #include <sys/pool.h>
88 #include <sys/sysdc.h>
89 #include <sys/zone.h>
90 #endif /* _KERNEL */
91
92 #include "zfs_prop.h"
93 #include "zfs_comutil.h"
94
95 /*
96 * The interval, in seconds, at which failed configuration cache file writes
97 * should be retried.
98 */
99 int zfs_ccw_retry_interval = 300;
100
101 typedef enum zti_modes {
102 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
103 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
104 ZTI_MODE_NULL, /* don't create a taskq */
105 ZTI_NMODES
106 } zti_modes_t;
107
108 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
109 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
110 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
111
112 #define ZTI_N(n) ZTI_P(n, 1)
113 #define ZTI_ONE ZTI_N(1)
114
115 typedef struct zio_taskq_info {
116 zti_modes_t zti_mode;
117 uint_t zti_value;
118 uint_t zti_count;
119 } zio_taskq_info_t;
120
121 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
122 "issue", "issue_high", "intr", "intr_high"
123 };
124
125 /*
126 * This table defines the taskq settings for each ZFS I/O type. When
127 * initializing a pool, we use this table to create an appropriately sized
128 * taskq. Some operations are low volume and therefore have a small, static
129 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
130 * macros. Other operations process a large amount of data; the ZTI_BATCH
131 * macro causes us to create a taskq oriented for throughput. Some operations
132 * are so high frequency and short-lived that the taskq itself can become a a
133 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
134 * additional degree of parallelism specified by the number of threads per-
135 * taskq and the number of taskqs; when dispatching an event in this case, the
136 * particular taskq is chosen at random.
137 *
138 * The different taskq priorities are to handle the different contexts (issue
139 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
140 * need to be handled with minimum delay.
141 */
142 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
143 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
145 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
146 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
147 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
149 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
150 };
151
152 static void spa_sync_version(void *arg, dmu_tx_t *tx);
153 static void spa_sync_props(void *arg, dmu_tx_t *tx);
154 static boolean_t spa_has_active_shared_spare(spa_t *spa);
155 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
156 static void spa_vdev_resilver_done(spa_t *spa);
157
158 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
159 id_t zio_taskq_psrset_bind = PS_NONE;
160 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
161 uint_t zio_taskq_basedc = 80; /* base duty cycle */
162
163 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
164 extern int zfs_sync_pass_deferred_free;
165
166 /*
167 * Report any spa_load_verify errors found, but do not fail spa_load.
168 * This is used by zdb to analyze non-idle pools.
169 */
170 boolean_t spa_load_verify_dryrun = B_FALSE;
171
172 /*
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
175 */
176 #define TRYIMPORT_NAME "$import"
177
178 /*
179 * For debugging purposes: print out vdev tree during pool import.
180 */
181 boolean_t spa_load_print_vdev_tree = B_FALSE;
182
183 /*
184 * A non-zero value for zfs_max_missing_tvds means that we allow importing
185 * pools with missing top-level vdevs. This is strictly intended for advanced
186 * pool recovery cases since missing data is almost inevitable. Pools with
187 * missing devices can only be imported read-only for safety reasons, and their
188 * fail-mode will be automatically set to "continue".
189 *
190 * With 1 missing vdev we should be able to import the pool and mount all
191 * datasets. User data that was not modified after the missing device has been
192 * added should be recoverable. This means that snapshots created prior to the
193 * addition of that device should be completely intact.
194 *
195 * With 2 missing vdevs, some datasets may fail to mount since there are
196 * dataset statistics that are stored as regular metadata. Some data might be
197 * recoverable if those vdevs were added recently.
198 *
199 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
200 * may be missing entirely. Chances of data recovery are very low. Note that
201 * there are also risks of performing an inadvertent rewind as we might be
202 * missing all the vdevs with the latest uberblocks.
203 */
204 uint64_t zfs_max_missing_tvds = 0;
205
206 /*
207 * The parameters below are similar to zfs_max_missing_tvds but are only
208 * intended for a preliminary open of the pool with an untrusted config which
209 * might be incomplete or out-dated.
210 *
211 * We are more tolerant for pools opened from a cachefile since we could have
212 * an out-dated cachefile where a device removal was not registered.
213 * We could have set the limit arbitrarily high but in the case where devices
214 * are really missing we would want to return the proper error codes; we chose
215 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
216 * and we get a chance to retrieve the trusted config.
217 */
218 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
219
220 /*
221 * In the case where config was assembled by scanning device paths (/dev/dsks
222 * by default) we are less tolerant since all the existing devices should have
223 * been detected and we want spa_load to return the right error codes.
224 */
225 uint64_t zfs_max_missing_tvds_scan = 0;
226
227 /*
228 * Debugging aid that pauses spa_sync() towards the end.
229 */
230 boolean_t zfs_pause_spa_sync = B_FALSE;
231
232 /*
233 * ==========================================================================
234 * SPA properties routines
235 * ==========================================================================
236 */
237
238 /*
239 * Add a (source=src, propname=propval) list to an nvlist.
240 */
241 static void
242 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
243 uint64_t intval, zprop_source_t src)
244 {
245 const char *propname = zpool_prop_to_name(prop);
246 nvlist_t *propval;
247
248 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
249 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
250
251 if (strval != NULL)
252 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
253 else
254 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
255
256 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
257 nvlist_free(propval);
258 }
259
260 /*
261 * Get property values from the spa configuration.
262 */
263 static void
264 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
265 {
266 vdev_t *rvd = spa->spa_root_vdev;
267 dsl_pool_t *pool = spa->spa_dsl_pool;
268 uint64_t size, alloc, cap, version;
269 zprop_source_t src = ZPROP_SRC_NONE;
270 spa_config_dirent_t *dp;
271 metaslab_class_t *mc = spa_normal_class(spa);
272
273 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
274
275 if (rvd != NULL) {
276 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
277 size = metaslab_class_get_space(spa_normal_class(spa));
278 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
279 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
280 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
281 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
282 size - alloc, src);
283 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
284 spa->spa_checkpoint_info.sci_dspace, src);
285
286 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
287 metaslab_class_fragmentation(mc), src);
288 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
289 metaslab_class_expandable_space(mc), src);
290 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
291 (spa_mode(spa) == FREAD), src);
292
293 cap = (size == 0) ? 0 : (alloc * 100 / size);
294 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
295
296 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
297 ddt_get_pool_dedup_ratio(spa), src);
298
299 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
300 rvd->vdev_state, src);
301
302 version = spa_version(spa);
303 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
304 src = ZPROP_SRC_DEFAULT;
305 else
306 src = ZPROP_SRC_LOCAL;
307 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
308 }
309
310 if (pool != NULL) {
311 /*
312 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
313 * when opening pools before this version freedir will be NULL.
314 */
315 if (pool->dp_free_dir != NULL) {
316 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
317 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
318 src);
319 } else {
320 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
321 NULL, 0, src);
322 }
323
324 if (pool->dp_leak_dir != NULL) {
325 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
326 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
327 src);
328 } else {
329 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
330 NULL, 0, src);
331 }
332 }
333
334 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
335
336 if (spa->spa_comment != NULL) {
337 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
338 0, ZPROP_SRC_LOCAL);
339 }
340
341 if (spa->spa_root != NULL)
342 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
343 0, ZPROP_SRC_LOCAL);
344
345 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
346 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
347 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
348 } else {
349 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
350 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
351 }
352
353 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
354 if (dp->scd_path == NULL) {
355 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
356 "none", 0, ZPROP_SRC_LOCAL);
357 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
358 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
359 dp->scd_path, 0, ZPROP_SRC_LOCAL);
360 }
361 }
362 }
363
364 /*
365 * Get zpool property values.
366 */
367 int
368 spa_prop_get(spa_t *spa, nvlist_t **nvp)
369 {
370 objset_t *mos = spa->spa_meta_objset;
371 zap_cursor_t zc;
372 zap_attribute_t za;
373 int err;
374
375 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
376
377 mutex_enter(&spa->spa_props_lock);
378
379 /*
380 * Get properties from the spa config.
381 */
382 spa_prop_get_config(spa, nvp);
383
384 /* If no pool property object, no more prop to get. */
385 if (mos == NULL || spa->spa_pool_props_object == 0) {
386 mutex_exit(&spa->spa_props_lock);
387 return (0);
388 }
389
390 /*
391 * Get properties from the MOS pool property object.
392 */
393 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
394 (err = zap_cursor_retrieve(&zc, &za)) == 0;
395 zap_cursor_advance(&zc)) {
396 uint64_t intval = 0;
397 char *strval = NULL;
398 zprop_source_t src = ZPROP_SRC_DEFAULT;
399 zpool_prop_t prop;
400
401 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
402 continue;
403
404 switch (za.za_integer_length) {
405 case 8:
406 /* integer property */
407 if (za.za_first_integer !=
408 zpool_prop_default_numeric(prop))
409 src = ZPROP_SRC_LOCAL;
410
411 if (prop == ZPOOL_PROP_BOOTFS) {
412 dsl_pool_t *dp;
413 dsl_dataset_t *ds = NULL;
414
415 dp = spa_get_dsl(spa);
416 dsl_pool_config_enter(dp, FTAG);
417 err = dsl_dataset_hold_obj(dp,
418 za.za_first_integer, FTAG, &ds);
419 if (err != 0) {
420 dsl_pool_config_exit(dp, FTAG);
421 break;
422 }
423
424 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
425 KM_SLEEP);
426 dsl_dataset_name(ds, strval);
427 dsl_dataset_rele(ds, FTAG);
428 dsl_pool_config_exit(dp, FTAG);
429 } else {
430 strval = NULL;
431 intval = za.za_first_integer;
432 }
433
434 spa_prop_add_list(*nvp, prop, strval, intval, src);
435
436 if (strval != NULL)
437 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
438
439 break;
440
441 case 1:
442 /* string property */
443 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
444 err = zap_lookup(mos, spa->spa_pool_props_object,
445 za.za_name, 1, za.za_num_integers, strval);
446 if (err) {
447 kmem_free(strval, za.za_num_integers);
448 break;
449 }
450 spa_prop_add_list(*nvp, prop, strval, 0, src);
451 kmem_free(strval, za.za_num_integers);
452 break;
453
454 default:
455 break;
456 }
457 }
458 zap_cursor_fini(&zc);
459 mutex_exit(&spa->spa_props_lock);
460 out:
461 if (err && err != ENOENT) {
462 nvlist_free(*nvp);
463 *nvp = NULL;
464 return (err);
465 }
466
467 return (0);
468 }
469
470 /*
471 * Validate the given pool properties nvlist and modify the list
472 * for the property values to be set.
473 */
474 static int
475 spa_prop_validate(spa_t *spa, nvlist_t *props)
476 {
477 nvpair_t *elem;
478 int error = 0, reset_bootfs = 0;
479 uint64_t objnum = 0;
480 boolean_t has_feature = B_FALSE;
481
482 elem = NULL;
483 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
484 uint64_t intval;
485 char *strval, *slash, *check, *fname;
486 const char *propname = nvpair_name(elem);
487 zpool_prop_t prop = zpool_name_to_prop(propname);
488
489 switch (prop) {
490 case ZPOOL_PROP_INVAL:
491 if (!zpool_prop_feature(propname)) {
492 error = SET_ERROR(EINVAL);
493 break;
494 }
495
496 /*
497 * Sanitize the input.
498 */
499 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
500 error = SET_ERROR(EINVAL);
501 break;
502 }
503
504 if (nvpair_value_uint64(elem, &intval) != 0) {
505 error = SET_ERROR(EINVAL);
506 break;
507 }
508
509 if (intval != 0) {
510 error = SET_ERROR(EINVAL);
511 break;
512 }
513
514 fname = strchr(propname, '@') + 1;
515 if (zfeature_lookup_name(fname, NULL) != 0) {
516 error = SET_ERROR(EINVAL);
517 break;
518 }
519
520 has_feature = B_TRUE;
521 break;
522
523 case ZPOOL_PROP_VERSION:
524 error = nvpair_value_uint64(elem, &intval);
525 if (!error &&
526 (intval < spa_version(spa) ||
527 intval > SPA_VERSION_BEFORE_FEATURES ||
528 has_feature))
529 error = SET_ERROR(EINVAL);
530 break;
531
532 case ZPOOL_PROP_DELEGATION:
533 case ZPOOL_PROP_AUTOREPLACE:
534 case ZPOOL_PROP_LISTSNAPS:
535 case ZPOOL_PROP_AUTOEXPAND:
536 error = nvpair_value_uint64(elem, &intval);
537 if (!error && intval > 1)
538 error = SET_ERROR(EINVAL);
539 break;
540
541 case ZPOOL_PROP_BOOTFS:
542 /*
543 * If the pool version is less than SPA_VERSION_BOOTFS,
544 * or the pool is still being created (version == 0),
545 * the bootfs property cannot be set.
546 */
547 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
548 error = SET_ERROR(ENOTSUP);
549 break;
550 }
551
552 /*
553 * Make sure the vdev config is bootable
554 */
555 if (!vdev_is_bootable(spa->spa_root_vdev)) {
556 error = SET_ERROR(ENOTSUP);
557 break;
558 }
559
560 reset_bootfs = 1;
561
562 error = nvpair_value_string(elem, &strval);
563
564 if (!error) {
565 objset_t *os;
566 uint64_t propval;
567
568 if (strval == NULL || strval[0] == '\0') {
569 objnum = zpool_prop_default_numeric(
570 ZPOOL_PROP_BOOTFS);
571 break;
572 }
573
574 error = dmu_objset_hold(strval, FTAG, &os);
575 if (error != 0)
576 break;
577
578 /*
579 * Must be ZPL, and its property settings
580 * must be supported by GRUB (compression
581 * is not gzip, and large blocks are not used).
582 */
583
584 if (dmu_objset_type(os) != DMU_OST_ZFS) {
585 error = SET_ERROR(ENOTSUP);
586 } else if ((error =
587 dsl_prop_get_int_ds(dmu_objset_ds(os),
588 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
589 &propval)) == 0 &&
590 !BOOTFS_COMPRESS_VALID(propval)) {
591 error = SET_ERROR(ENOTSUP);
592 } else {
593 objnum = dmu_objset_id(os);
594 }
595 dmu_objset_rele(os, FTAG);
596 }
597 break;
598
599 case ZPOOL_PROP_FAILUREMODE:
600 error = nvpair_value_uint64(elem, &intval);
601 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
602 intval > ZIO_FAILURE_MODE_PANIC))
603 error = SET_ERROR(EINVAL);
604
605 /*
606 * This is a special case which only occurs when
607 * the pool has completely failed. This allows
608 * the user to change the in-core failmode property
609 * without syncing it out to disk (I/Os might
610 * currently be blocked). We do this by returning
611 * EIO to the caller (spa_prop_set) to trick it
612 * into thinking we encountered a property validation
613 * error.
614 */
615 if (!error && spa_suspended(spa)) {
616 spa->spa_failmode = intval;
617 error = SET_ERROR(EIO);
618 }
619 break;
620
621 case ZPOOL_PROP_CACHEFILE:
622 if ((error = nvpair_value_string(elem, &strval)) != 0)
623 break;
624
625 if (strval[0] == '\0')
626 break;
627
628 if (strcmp(strval, "none") == 0)
629 break;
630
631 if (strval[0] != '/') {
632 error = SET_ERROR(EINVAL);
633 break;
634 }
635
636 slash = strrchr(strval, '/');
637 ASSERT(slash != NULL);
638
639 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
640 strcmp(slash, "/..") == 0)
641 error = SET_ERROR(EINVAL);
642 break;
643
644 case ZPOOL_PROP_COMMENT:
645 if ((error = nvpair_value_string(elem, &strval)) != 0)
646 break;
647 for (check = strval; *check != '\0'; check++) {
648 /*
649 * The kernel doesn't have an easy isprint()
650 * check. For this kernel check, we merely
651 * check ASCII apart from DEL. Fix this if
652 * there is an easy-to-use kernel isprint().
653 */
654 if (*check >= 0x7f) {
655 error = SET_ERROR(EINVAL);
656 break;
657 }
658 }
659 if (strlen(strval) > ZPROP_MAX_COMMENT)
660 error = E2BIG;
661 break;
662
663 case ZPOOL_PROP_DEDUPDITTO:
664 if (spa_version(spa) < SPA_VERSION_DEDUP)
665 error = SET_ERROR(ENOTSUP);
666 else
667 error = nvpair_value_uint64(elem, &intval);
668 if (error == 0 &&
669 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
670 error = SET_ERROR(EINVAL);
671 break;
672 }
673
674 if (error)
675 break;
676 }
677
678 if (!error && reset_bootfs) {
679 error = nvlist_remove(props,
680 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
681
682 if (!error) {
683 error = nvlist_add_uint64(props,
684 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
685 }
686 }
687
688 return (error);
689 }
690
691 void
692 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
693 {
694 char *cachefile;
695 spa_config_dirent_t *dp;
696
697 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
698 &cachefile) != 0)
699 return;
700
701 dp = kmem_alloc(sizeof (spa_config_dirent_t),
702 KM_SLEEP);
703
704 if (cachefile[0] == '\0')
705 dp->scd_path = spa_strdup(spa_config_path);
706 else if (strcmp(cachefile, "none") == 0)
707 dp->scd_path = NULL;
708 else
709 dp->scd_path = spa_strdup(cachefile);
710
711 list_insert_head(&spa->spa_config_list, dp);
712 if (need_sync)
713 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
714 }
715
716 int
717 spa_prop_set(spa_t *spa, nvlist_t *nvp)
718 {
719 int error;
720 nvpair_t *elem = NULL;
721 boolean_t need_sync = B_FALSE;
722
723 if ((error = spa_prop_validate(spa, nvp)) != 0)
724 return (error);
725
726 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
727 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
728
729 if (prop == ZPOOL_PROP_CACHEFILE ||
730 prop == ZPOOL_PROP_ALTROOT ||
731 prop == ZPOOL_PROP_READONLY)
732 continue;
733
734 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
735 uint64_t ver;
736
737 if (prop == ZPOOL_PROP_VERSION) {
738 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
739 } else {
740 ASSERT(zpool_prop_feature(nvpair_name(elem)));
741 ver = SPA_VERSION_FEATURES;
742 need_sync = B_TRUE;
743 }
744
745 /* Save time if the version is already set. */
746 if (ver == spa_version(spa))
747 continue;
748
749 /*
750 * In addition to the pool directory object, we might
751 * create the pool properties object, the features for
752 * read object, the features for write object, or the
753 * feature descriptions object.
754 */
755 error = dsl_sync_task(spa->spa_name, NULL,
756 spa_sync_version, &ver,
757 6, ZFS_SPACE_CHECK_RESERVED);
758 if (error)
759 return (error);
760 continue;
761 }
762
763 need_sync = B_TRUE;
764 break;
765 }
766
767 if (need_sync) {
768 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
769 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
770 }
771
772 return (0);
773 }
774
775 /*
776 * If the bootfs property value is dsobj, clear it.
777 */
778 void
779 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
780 {
781 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
782 VERIFY(zap_remove(spa->spa_meta_objset,
783 spa->spa_pool_props_object,
784 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
785 spa->spa_bootfs = 0;
786 }
787 }
788
789 /*ARGSUSED*/
790 static int
791 spa_change_guid_check(void *arg, dmu_tx_t *tx)
792 {
793 uint64_t *newguid = arg;
794 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
795 vdev_t *rvd = spa->spa_root_vdev;
796 uint64_t vdev_state;
797
798 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
799 int error = (spa_has_checkpoint(spa)) ?
800 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
801 return (SET_ERROR(error));
802 }
803
804 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
805 vdev_state = rvd->vdev_state;
806 spa_config_exit(spa, SCL_STATE, FTAG);
807
808 if (vdev_state != VDEV_STATE_HEALTHY)
809 return (SET_ERROR(ENXIO));
810
811 ASSERT3U(spa_guid(spa), !=, *newguid);
812
813 return (0);
814 }
815
816 static void
817 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
818 {
819 uint64_t *newguid = arg;
820 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
821 uint64_t oldguid;
822 vdev_t *rvd = spa->spa_root_vdev;
823
824 oldguid = spa_guid(spa);
825
826 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
827 rvd->vdev_guid = *newguid;
828 rvd->vdev_guid_sum += (*newguid - oldguid);
829 vdev_config_dirty(rvd);
830 spa_config_exit(spa, SCL_STATE, FTAG);
831
832 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
833 oldguid, *newguid);
834 }
835
836 /*
837 * Change the GUID for the pool. This is done so that we can later
838 * re-import a pool built from a clone of our own vdevs. We will modify
839 * the root vdev's guid, our own pool guid, and then mark all of our
840 * vdevs dirty. Note that we must make sure that all our vdevs are
841 * online when we do this, or else any vdevs that weren't present
842 * would be orphaned from our pool. We are also going to issue a
843 * sysevent to update any watchers.
844 */
845 int
846 spa_change_guid(spa_t *spa)
847 {
848 int error;
849 uint64_t guid;
850
851 mutex_enter(&spa->spa_vdev_top_lock);
852 mutex_enter(&spa_namespace_lock);
853 guid = spa_generate_guid(NULL);
854
855 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
856 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
857
858 if (error == 0) {
859 spa_write_cachefile(spa, B_FALSE, B_TRUE);
860 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
861 }
862
863 mutex_exit(&spa_namespace_lock);
864 mutex_exit(&spa->spa_vdev_top_lock);
865
866 return (error);
867 }
868
869 /*
870 * ==========================================================================
871 * SPA state manipulation (open/create/destroy/import/export)
872 * ==========================================================================
873 */
874
875 static int
876 spa_error_entry_compare(const void *a, const void *b)
877 {
878 spa_error_entry_t *sa = (spa_error_entry_t *)a;
879 spa_error_entry_t *sb = (spa_error_entry_t *)b;
880 int ret;
881
882 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
883 sizeof (zbookmark_phys_t));
884
885 if (ret < 0)
886 return (-1);
887 else if (ret > 0)
888 return (1);
889 else
890 return (0);
891 }
892
893 /*
894 * Utility function which retrieves copies of the current logs and
895 * re-initializes them in the process.
896 */
897 void
898 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
899 {
900 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
901
902 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
903 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
904
905 avl_create(&spa->spa_errlist_scrub,
906 spa_error_entry_compare, sizeof (spa_error_entry_t),
907 offsetof(spa_error_entry_t, se_avl));
908 avl_create(&spa->spa_errlist_last,
909 spa_error_entry_compare, sizeof (spa_error_entry_t),
910 offsetof(spa_error_entry_t, se_avl));
911 }
912
913 static void
914 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
915 {
916 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
917 enum zti_modes mode = ztip->zti_mode;
918 uint_t value = ztip->zti_value;
919 uint_t count = ztip->zti_count;
920 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
921 char name[32];
922 uint_t flags = 0;
923 boolean_t batch = B_FALSE;
924
925 if (mode == ZTI_MODE_NULL) {
926 tqs->stqs_count = 0;
927 tqs->stqs_taskq = NULL;
928 return;
929 }
930
931 ASSERT3U(count, >, 0);
932
933 tqs->stqs_count = count;
934 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
935
936 switch (mode) {
937 case ZTI_MODE_FIXED:
938 ASSERT3U(value, >=, 1);
939 value = MAX(value, 1);
940 break;
941
942 case ZTI_MODE_BATCH:
943 batch = B_TRUE;
944 flags |= TASKQ_THREADS_CPU_PCT;
945 value = zio_taskq_batch_pct;
946 break;
947
948 default:
949 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
950 "spa_activate()",
951 zio_type_name[t], zio_taskq_types[q], mode, value);
952 break;
953 }
954
955 for (uint_t i = 0; i < count; i++) {
956 taskq_t *tq;
957
958 if (count > 1) {
959 (void) snprintf(name, sizeof (name), "%s_%s_%u",
960 zio_type_name[t], zio_taskq_types[q], i);
961 } else {
962 (void) snprintf(name, sizeof (name), "%s_%s",
963 zio_type_name[t], zio_taskq_types[q]);
964 }
965
966 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
967 if (batch)
968 flags |= TASKQ_DC_BATCH;
969
970 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
971 spa->spa_proc, zio_taskq_basedc, flags);
972 } else {
973 pri_t pri = maxclsyspri;
974 /*
975 * The write issue taskq can be extremely CPU
976 * intensive. Run it at slightly lower priority
977 * than the other taskqs.
978 */
979 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
980 pri--;
981
982 tq = taskq_create_proc(name, value, pri, 50,
983 INT_MAX, spa->spa_proc, flags);
984 }
985
986 tqs->stqs_taskq[i] = tq;
987 }
988 }
989
990 static void
991 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
992 {
993 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
994
995 if (tqs->stqs_taskq == NULL) {
996 ASSERT0(tqs->stqs_count);
997 return;
998 }
999
1000 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1001 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1002 taskq_destroy(tqs->stqs_taskq[i]);
1003 }
1004
1005 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1006 tqs->stqs_taskq = NULL;
1007 }
1008
1009 /*
1010 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1011 * Note that a type may have multiple discrete taskqs to avoid lock contention
1012 * on the taskq itself. In that case we choose which taskq at random by using
1013 * the low bits of gethrtime().
1014 */
1015 void
1016 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1017 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1018 {
1019 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1020 taskq_t *tq;
1021
1022 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1023 ASSERT3U(tqs->stqs_count, !=, 0);
1024
1025 if (tqs->stqs_count == 1) {
1026 tq = tqs->stqs_taskq[0];
1027 } else {
1028 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1029 }
1030
1031 taskq_dispatch_ent(tq, func, arg, flags, ent);
1032 }
1033
1034 static void
1035 spa_create_zio_taskqs(spa_t *spa)
1036 {
1037 for (int t = 0; t < ZIO_TYPES; t++) {
1038 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1039 spa_taskqs_init(spa, t, q);
1040 }
1041 }
1042 }
1043
1044 #ifdef _KERNEL
1045 static void
1046 spa_thread(void *arg)
1047 {
1048 callb_cpr_t cprinfo;
1049
1050 spa_t *spa = arg;
1051 user_t *pu = PTOU(curproc);
1052
1053 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1054 spa->spa_name);
1055
1056 ASSERT(curproc != &p0);
1057 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1058 "zpool-%s", spa->spa_name);
1059 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1060
1061 /* bind this thread to the requested psrset */
1062 if (zio_taskq_psrset_bind != PS_NONE) {
1063 pool_lock();
1064 mutex_enter(&cpu_lock);
1065 mutex_enter(&pidlock);
1066 mutex_enter(&curproc->p_lock);
1067
1068 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1069 0, NULL, NULL) == 0) {
1070 curthread->t_bind_pset = zio_taskq_psrset_bind;
1071 } else {
1072 cmn_err(CE_WARN,
1073 "Couldn't bind process for zfs pool \"%s\" to "
1074 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1075 }
1076
1077 mutex_exit(&curproc->p_lock);
1078 mutex_exit(&pidlock);
1079 mutex_exit(&cpu_lock);
1080 pool_unlock();
1081 }
1082
1083 if (zio_taskq_sysdc) {
1084 sysdc_thread_enter(curthread, 100, 0);
1085 }
1086
1087 spa->spa_proc = curproc;
1088 spa->spa_did = curthread->t_did;
1089
1090 spa_create_zio_taskqs(spa);
1091
1092 mutex_enter(&spa->spa_proc_lock);
1093 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1094
1095 spa->spa_proc_state = SPA_PROC_ACTIVE;
1096 cv_broadcast(&spa->spa_proc_cv);
1097
1098 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1099 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1100 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1101 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1102
1103 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1104 spa->spa_proc_state = SPA_PROC_GONE;
1105 spa->spa_proc = &p0;
1106 cv_broadcast(&spa->spa_proc_cv);
1107 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1108
1109 mutex_enter(&curproc->p_lock);
1110 lwp_exit();
1111 }
1112 #endif
1113
1114 /*
1115 * Activate an uninitialized pool.
1116 */
1117 static void
1118 spa_activate(spa_t *spa, int mode)
1119 {
1120 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1121
1122 spa->spa_state = POOL_STATE_ACTIVE;
1123 spa->spa_mode = mode;
1124
1125 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1126 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1127
1128 /* Try to create a covering process */
1129 mutex_enter(&spa->spa_proc_lock);
1130 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1131 ASSERT(spa->spa_proc == &p0);
1132 spa->spa_did = 0;
1133
1134 /* Only create a process if we're going to be around a while. */
1135 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1136 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1137 NULL, 0) == 0) {
1138 spa->spa_proc_state = SPA_PROC_CREATED;
1139 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1140 cv_wait(&spa->spa_proc_cv,
1141 &spa->spa_proc_lock);
1142 }
1143 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1144 ASSERT(spa->spa_proc != &p0);
1145 ASSERT(spa->spa_did != 0);
1146 } else {
1147 #ifdef _KERNEL
1148 cmn_err(CE_WARN,
1149 "Couldn't create process for zfs pool \"%s\"\n",
1150 spa->spa_name);
1151 #endif
1152 }
1153 }
1154 mutex_exit(&spa->spa_proc_lock);
1155
1156 /* If we didn't create a process, we need to create our taskqs. */
1157 if (spa->spa_proc == &p0) {
1158 spa_create_zio_taskqs(spa);
1159 }
1160
1161 for (size_t i = 0; i < TXG_SIZE; i++) {
1162 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1163 ZIO_FLAG_CANFAIL);
1164 }
1165
1166 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1167 offsetof(vdev_t, vdev_config_dirty_node));
1168 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1169 offsetof(objset_t, os_evicting_node));
1170 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1171 offsetof(vdev_t, vdev_state_dirty_node));
1172
1173 txg_list_create(&spa->spa_vdev_txg_list, spa,
1174 offsetof(struct vdev, vdev_txg_node));
1175
1176 avl_create(&spa->spa_errlist_scrub,
1177 spa_error_entry_compare, sizeof (spa_error_entry_t),
1178 offsetof(spa_error_entry_t, se_avl));
1179 avl_create(&spa->spa_errlist_last,
1180 spa_error_entry_compare, sizeof (spa_error_entry_t),
1181 offsetof(spa_error_entry_t, se_avl));
1182 }
1183
1184 /*
1185 * Opposite of spa_activate().
1186 */
1187 static void
1188 spa_deactivate(spa_t *spa)
1189 {
1190 ASSERT(spa->spa_sync_on == B_FALSE);
1191 ASSERT(spa->spa_dsl_pool == NULL);
1192 ASSERT(spa->spa_root_vdev == NULL);
1193 ASSERT(spa->spa_async_zio_root == NULL);
1194 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1195
1196 spa_evicting_os_wait(spa);
1197
1198 txg_list_destroy(&spa->spa_vdev_txg_list);
1199
1200 list_destroy(&spa->spa_config_dirty_list);
1201 list_destroy(&spa->spa_evicting_os_list);
1202 list_destroy(&spa->spa_state_dirty_list);
1203
1204 for (int t = 0; t < ZIO_TYPES; t++) {
1205 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1206 spa_taskqs_fini(spa, t, q);
1207 }
1208 }
1209
1210 for (size_t i = 0; i < TXG_SIZE; i++) {
1211 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1212 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1213 spa->spa_txg_zio[i] = NULL;
1214 }
1215
1216 metaslab_class_destroy(spa->spa_normal_class);
1217 spa->spa_normal_class = NULL;
1218
1219 metaslab_class_destroy(spa->spa_log_class);
1220 spa->spa_log_class = NULL;
1221
1222 /*
1223 * If this was part of an import or the open otherwise failed, we may
1224 * still have errors left in the queues. Empty them just in case.
1225 */
1226 spa_errlog_drain(spa);
1227
1228 avl_destroy(&spa->spa_errlist_scrub);
1229 avl_destroy(&spa->spa_errlist_last);
1230
1231 spa->spa_state = POOL_STATE_UNINITIALIZED;
1232
1233 mutex_enter(&spa->spa_proc_lock);
1234 if (spa->spa_proc_state != SPA_PROC_NONE) {
1235 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1236 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1237 cv_broadcast(&spa->spa_proc_cv);
1238 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1239 ASSERT(spa->spa_proc != &p0);
1240 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1241 }
1242 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1243 spa->spa_proc_state = SPA_PROC_NONE;
1244 }
1245 ASSERT(spa->spa_proc == &p0);
1246 mutex_exit(&spa->spa_proc_lock);
1247
1248 /*
1249 * We want to make sure spa_thread() has actually exited the ZFS
1250 * module, so that the module can't be unloaded out from underneath
1251 * it.
1252 */
1253 if (spa->spa_did != 0) {
1254 thread_join(spa->spa_did);
1255 spa->spa_did = 0;
1256 }
1257 }
1258
1259 /*
1260 * Verify a pool configuration, and construct the vdev tree appropriately. This
1261 * will create all the necessary vdevs in the appropriate layout, with each vdev
1262 * in the CLOSED state. This will prep the pool before open/creation/import.
1263 * All vdev validation is done by the vdev_alloc() routine.
1264 */
1265 static int
1266 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1267 uint_t id, int atype)
1268 {
1269 nvlist_t **child;
1270 uint_t children;
1271 int error;
1272
1273 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1274 return (error);
1275
1276 if ((*vdp)->vdev_ops->vdev_op_leaf)
1277 return (0);
1278
1279 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1280 &child, &children);
1281
1282 if (error == ENOENT)
1283 return (0);
1284
1285 if (error) {
1286 vdev_free(*vdp);
1287 *vdp = NULL;
1288 return (SET_ERROR(EINVAL));
1289 }
1290
1291 for (int c = 0; c < children; c++) {
1292 vdev_t *vd;
1293 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1294 atype)) != 0) {
1295 vdev_free(*vdp);
1296 *vdp = NULL;
1297 return (error);
1298 }
1299 }
1300
1301 ASSERT(*vdp != NULL);
1302
1303 return (0);
1304 }
1305
1306 /*
1307 * Opposite of spa_load().
1308 */
1309 static void
1310 spa_unload(spa_t *spa)
1311 {
1312 int i;
1313
1314 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1315
1316 spa_load_note(spa, "UNLOADING");
1317
1318 /*
1319 * Stop async tasks.
1320 */
1321 spa_async_suspend(spa);
1322
1323 if (spa->spa_root_vdev) {
1324 vdev_initialize_stop_all(spa->spa_root_vdev,
1325 VDEV_INITIALIZE_ACTIVE);
1326 }
1327
1328 /*
1329 * Stop syncing.
1330 */
1331 if (spa->spa_sync_on) {
1332 txg_sync_stop(spa->spa_dsl_pool);
1333 spa->spa_sync_on = B_FALSE;
1334 }
1335
1336 /*
1337 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1338 * to call it earlier, before we wait for async i/o to complete.
1339 * This ensures that there is no async metaslab prefetching, by
1340 * calling taskq_wait(mg_taskq).
1341 */
1342 if (spa->spa_root_vdev != NULL) {
1343 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1344 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1345 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1346 spa_config_exit(spa, SCL_ALL, spa);
1347 }
1348
1349 /*
1350 * Wait for any outstanding async I/O to complete.
1351 */
1352 if (spa->spa_async_zio_root != NULL) {
1353 for (int i = 0; i < max_ncpus; i++)
1354 (void) zio_wait(spa->spa_async_zio_root[i]);
1355 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1356 spa->spa_async_zio_root = NULL;
1357 }
1358
1359 if (spa->spa_vdev_removal != NULL) {
1360 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1361 spa->spa_vdev_removal = NULL;
1362 }
1363
1364 if (spa->spa_condense_zthr != NULL) {
1365 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1366 zthr_destroy(spa->spa_condense_zthr);
1367 spa->spa_condense_zthr = NULL;
1368 }
1369
1370 if (spa->spa_checkpoint_discard_zthr != NULL) {
1371 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1372 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1373 spa->spa_checkpoint_discard_zthr = NULL;
1374 }
1375
1376 spa_condense_fini(spa);
1377
1378 bpobj_close(&spa->spa_deferred_bpobj);
1379
1380 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1381
1382 /*
1383 * Close all vdevs.
1384 */
1385 if (spa->spa_root_vdev)
1386 vdev_free(spa->spa_root_vdev);
1387 ASSERT(spa->spa_root_vdev == NULL);
1388
1389 /*
1390 * Close the dsl pool.
1391 */
1392 if (spa->spa_dsl_pool) {
1393 dsl_pool_close(spa->spa_dsl_pool);
1394 spa->spa_dsl_pool = NULL;
1395 spa->spa_meta_objset = NULL;
1396 }
1397
1398 ddt_unload(spa);
1399
1400 /*
1401 * Drop and purge level 2 cache
1402 */
1403 spa_l2cache_drop(spa);
1404
1405 for (i = 0; i < spa->spa_spares.sav_count; i++)
1406 vdev_free(spa->spa_spares.sav_vdevs[i]);
1407 if (spa->spa_spares.sav_vdevs) {
1408 kmem_free(spa->spa_spares.sav_vdevs,
1409 spa->spa_spares.sav_count * sizeof (void *));
1410 spa->spa_spares.sav_vdevs = NULL;
1411 }
1412 if (spa->spa_spares.sav_config) {
1413 nvlist_free(spa->spa_spares.sav_config);
1414 spa->spa_spares.sav_config = NULL;
1415 }
1416 spa->spa_spares.sav_count = 0;
1417
1418 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1419 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1420 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1421 }
1422 if (spa->spa_l2cache.sav_vdevs) {
1423 kmem_free(spa->spa_l2cache.sav_vdevs,
1424 spa->spa_l2cache.sav_count * sizeof (void *));
1425 spa->spa_l2cache.sav_vdevs = NULL;
1426 }
1427 if (spa->spa_l2cache.sav_config) {
1428 nvlist_free(spa->spa_l2cache.sav_config);
1429 spa->spa_l2cache.sav_config = NULL;
1430 }
1431 spa->spa_l2cache.sav_count = 0;
1432
1433 spa->spa_async_suspended = 0;
1434
1435 spa->spa_indirect_vdevs_loaded = B_FALSE;
1436
1437 if (spa->spa_comment != NULL) {
1438 spa_strfree(spa->spa_comment);
1439 spa->spa_comment = NULL;
1440 }
1441
1442 spa_config_exit(spa, SCL_ALL, spa);
1443 }
1444
1445 /*
1446 * Load (or re-load) the current list of vdevs describing the active spares for
1447 * this pool. When this is called, we have some form of basic information in
1448 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1449 * then re-generate a more complete list including status information.
1450 */
1451 void
1452 spa_load_spares(spa_t *spa)
1453 {
1454 nvlist_t **spares;
1455 uint_t nspares;
1456 int i;
1457 vdev_t *vd, *tvd;
1458
1459 #ifndef _KERNEL
1460 /*
1461 * zdb opens both the current state of the pool and the
1462 * checkpointed state (if present), with a different spa_t.
1463 *
1464 * As spare vdevs are shared among open pools, we skip loading
1465 * them when we load the checkpointed state of the pool.
1466 */
1467 if (!spa_writeable(spa))
1468 return;
1469 #endif
1470
1471 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1472
1473 /*
1474 * First, close and free any existing spare vdevs.
1475 */
1476 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1477 vd = spa->spa_spares.sav_vdevs[i];
1478
1479 /* Undo the call to spa_activate() below */
1480 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1481 B_FALSE)) != NULL && tvd->vdev_isspare)
1482 spa_spare_remove(tvd);
1483 vdev_close(vd);
1484 vdev_free(vd);
1485 }
1486
1487 if (spa->spa_spares.sav_vdevs)
1488 kmem_free(spa->spa_spares.sav_vdevs,
1489 spa->spa_spares.sav_count * sizeof (void *));
1490
1491 if (spa->spa_spares.sav_config == NULL)
1492 nspares = 0;
1493 else
1494 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1495 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1496
1497 spa->spa_spares.sav_count = (int)nspares;
1498 spa->spa_spares.sav_vdevs = NULL;
1499
1500 if (nspares == 0)
1501 return;
1502
1503 /*
1504 * Construct the array of vdevs, opening them to get status in the
1505 * process. For each spare, there is potentially two different vdev_t
1506 * structures associated with it: one in the list of spares (used only
1507 * for basic validation purposes) and one in the active vdev
1508 * configuration (if it's spared in). During this phase we open and
1509 * validate each vdev on the spare list. If the vdev also exists in the
1510 * active configuration, then we also mark this vdev as an active spare.
1511 */
1512 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1513 KM_SLEEP);
1514 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1515 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1516 VDEV_ALLOC_SPARE) == 0);
1517 ASSERT(vd != NULL);
1518
1519 spa->spa_spares.sav_vdevs[i] = vd;
1520
1521 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1522 B_FALSE)) != NULL) {
1523 if (!tvd->vdev_isspare)
1524 spa_spare_add(tvd);
1525
1526 /*
1527 * We only mark the spare active if we were successfully
1528 * able to load the vdev. Otherwise, importing a pool
1529 * with a bad active spare would result in strange
1530 * behavior, because multiple pool would think the spare
1531 * is actively in use.
1532 *
1533 * There is a vulnerability here to an equally bizarre
1534 * circumstance, where a dead active spare is later
1535 * brought back to life (onlined or otherwise). Given
1536 * the rarity of this scenario, and the extra complexity
1537 * it adds, we ignore the possibility.
1538 */
1539 if (!vdev_is_dead(tvd))
1540 spa_spare_activate(tvd);
1541 }
1542
1543 vd->vdev_top = vd;
1544 vd->vdev_aux = &spa->spa_spares;
1545
1546 if (vdev_open(vd) != 0)
1547 continue;
1548
1549 if (vdev_validate_aux(vd) == 0)
1550 spa_spare_add(vd);
1551 }
1552
1553 /*
1554 * Recompute the stashed list of spares, with status information
1555 * this time.
1556 */
1557 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1558 DATA_TYPE_NVLIST_ARRAY) == 0);
1559
1560 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1561 KM_SLEEP);
1562 for (i = 0; i < spa->spa_spares.sav_count; i++)
1563 spares[i] = vdev_config_generate(spa,
1564 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1565 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1566 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1567 for (i = 0; i < spa->spa_spares.sav_count; i++)
1568 nvlist_free(spares[i]);
1569 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1570 }
1571
1572 /*
1573 * Load (or re-load) the current list of vdevs describing the active l2cache for
1574 * this pool. When this is called, we have some form of basic information in
1575 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1576 * then re-generate a more complete list including status information.
1577 * Devices which are already active have their details maintained, and are
1578 * not re-opened.
1579 */
1580 void
1581 spa_load_l2cache(spa_t *spa)
1582 {
1583 nvlist_t **l2cache;
1584 uint_t nl2cache;
1585 int i, j, oldnvdevs;
1586 uint64_t guid;
1587 vdev_t *vd, **oldvdevs, **newvdevs;
1588 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1589
1590 #ifndef _KERNEL
1591 /*
1592 * zdb opens both the current state of the pool and the
1593 * checkpointed state (if present), with a different spa_t.
1594 *
1595 * As L2 caches are part of the ARC which is shared among open
1596 * pools, we skip loading them when we load the checkpointed
1597 * state of the pool.
1598 */
1599 if (!spa_writeable(spa))
1600 return;
1601 #endif
1602
1603 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1604
1605 if (sav->sav_config != NULL) {
1606 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1607 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1608 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1609 } else {
1610 nl2cache = 0;
1611 newvdevs = NULL;
1612 }
1613
1614 oldvdevs = sav->sav_vdevs;
1615 oldnvdevs = sav->sav_count;
1616 sav->sav_vdevs = NULL;
1617 sav->sav_count = 0;
1618
1619 /*
1620 * Process new nvlist of vdevs.
1621 */
1622 for (i = 0; i < nl2cache; i++) {
1623 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1624 &guid) == 0);
1625
1626 newvdevs[i] = NULL;
1627 for (j = 0; j < oldnvdevs; j++) {
1628 vd = oldvdevs[j];
1629 if (vd != NULL && guid == vd->vdev_guid) {
1630 /*
1631 * Retain previous vdev for add/remove ops.
1632 */
1633 newvdevs[i] = vd;
1634 oldvdevs[j] = NULL;
1635 break;
1636 }
1637 }
1638
1639 if (newvdevs[i] == NULL) {
1640 /*
1641 * Create new vdev
1642 */
1643 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1644 VDEV_ALLOC_L2CACHE) == 0);
1645 ASSERT(vd != NULL);
1646 newvdevs[i] = vd;
1647
1648 /*
1649 * Commit this vdev as an l2cache device,
1650 * even if it fails to open.
1651 */
1652 spa_l2cache_add(vd);
1653
1654 vd->vdev_top = vd;
1655 vd->vdev_aux = sav;
1656
1657 spa_l2cache_activate(vd);
1658
1659 if (vdev_open(vd) != 0)
1660 continue;
1661
1662 (void) vdev_validate_aux(vd);
1663
1664 if (!vdev_is_dead(vd))
1665 l2arc_add_vdev(spa, vd);
1666 }
1667 }
1668
1669 /*
1670 * Purge vdevs that were dropped
1671 */
1672 for (i = 0; i < oldnvdevs; i++) {
1673 uint64_t pool;
1674
1675 vd = oldvdevs[i];
1676 if (vd != NULL) {
1677 ASSERT(vd->vdev_isl2cache);
1678
1679 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1680 pool != 0ULL && l2arc_vdev_present(vd))
1681 l2arc_remove_vdev(vd);
1682 vdev_clear_stats(vd);
1683 vdev_free(vd);
1684 }
1685 }
1686
1687 if (oldvdevs)
1688 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1689
1690 if (sav->sav_config == NULL)
1691 goto out;
1692
1693 sav->sav_vdevs = newvdevs;
1694 sav->sav_count = (int)nl2cache;
1695
1696 /*
1697 * Recompute the stashed list of l2cache devices, with status
1698 * information this time.
1699 */
1700 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1701 DATA_TYPE_NVLIST_ARRAY) == 0);
1702
1703 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1704 for (i = 0; i < sav->sav_count; i++)
1705 l2cache[i] = vdev_config_generate(spa,
1706 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1707 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1708 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1709 out:
1710 for (i = 0; i < sav->sav_count; i++)
1711 nvlist_free(l2cache[i]);
1712 if (sav->sav_count)
1713 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1714 }
1715
1716 static int
1717 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1718 {
1719 dmu_buf_t *db;
1720 char *packed = NULL;
1721 size_t nvsize = 0;
1722 int error;
1723 *value = NULL;
1724
1725 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1726 if (error != 0)
1727 return (error);
1728
1729 nvsize = *(uint64_t *)db->db_data;
1730 dmu_buf_rele(db, FTAG);
1731
1732 packed = kmem_alloc(nvsize, KM_SLEEP);
1733 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1734 DMU_READ_PREFETCH);
1735 if (error == 0)
1736 error = nvlist_unpack(packed, nvsize, value, 0);
1737 kmem_free(packed, nvsize);
1738
1739 return (error);
1740 }
1741
1742 /*
1743 * Concrete top-level vdevs that are not missing and are not logs. At every
1744 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1745 */
1746 static uint64_t
1747 spa_healthy_core_tvds(spa_t *spa)
1748 {
1749 vdev_t *rvd = spa->spa_root_vdev;
1750 uint64_t tvds = 0;
1751
1752 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1753 vdev_t *vd = rvd->vdev_child[i];
1754 if (vd->vdev_islog)
1755 continue;
1756 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1757 tvds++;
1758 }
1759
1760 return (tvds);
1761 }
1762
1763 /*
1764 * Checks to see if the given vdev could not be opened, in which case we post a
1765 * sysevent to notify the autoreplace code that the device has been removed.
1766 */
1767 static void
1768 spa_check_removed(vdev_t *vd)
1769 {
1770 for (uint64_t c = 0; c < vd->vdev_children; c++)
1771 spa_check_removed(vd->vdev_child[c]);
1772
1773 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1774 vdev_is_concrete(vd)) {
1775 zfs_post_autoreplace(vd->vdev_spa, vd);
1776 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1777 }
1778 }
1779
1780 static int
1781 spa_check_for_missing_logs(spa_t *spa)
1782 {
1783 vdev_t *rvd = spa->spa_root_vdev;
1784
1785 /*
1786 * If we're doing a normal import, then build up any additional
1787 * diagnostic information about missing log devices.
1788 * We'll pass this up to the user for further processing.
1789 */
1790 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1791 nvlist_t **child, *nv;
1792 uint64_t idx = 0;
1793
1794 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1795 KM_SLEEP);
1796 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1797
1798 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1799 vdev_t *tvd = rvd->vdev_child[c];
1800
1801 /*
1802 * We consider a device as missing only if it failed
1803 * to open (i.e. offline or faulted is not considered
1804 * as missing).
1805 */
1806 if (tvd->vdev_islog &&
1807 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1808 child[idx++] = vdev_config_generate(spa, tvd,
1809 B_FALSE, VDEV_CONFIG_MISSING);
1810 }
1811 }
1812
1813 if (idx > 0) {
1814 fnvlist_add_nvlist_array(nv,
1815 ZPOOL_CONFIG_CHILDREN, child, idx);
1816 fnvlist_add_nvlist(spa->spa_load_info,
1817 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1818
1819 for (uint64_t i = 0; i < idx; i++)
1820 nvlist_free(child[i]);
1821 }
1822 nvlist_free(nv);
1823 kmem_free(child, rvd->vdev_children * sizeof (char **));
1824
1825 if (idx > 0) {
1826 spa_load_failed(spa, "some log devices are missing");
1827 vdev_dbgmsg_print_tree(rvd, 2);
1828 return (SET_ERROR(ENXIO));
1829 }
1830 } else {
1831 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1832 vdev_t *tvd = rvd->vdev_child[c];
1833
1834 if (tvd->vdev_islog &&
1835 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1836 spa_set_log_state(spa, SPA_LOG_CLEAR);
1837 spa_load_note(spa, "some log devices are "
1838 "missing, ZIL is dropped.");
1839 vdev_dbgmsg_print_tree(rvd, 2);
1840 break;
1841 }
1842 }
1843 }
1844
1845 return (0);
1846 }
1847
1848 /*
1849 * Check for missing log devices
1850 */
1851 static boolean_t
1852 spa_check_logs(spa_t *spa)
1853 {
1854 boolean_t rv = B_FALSE;
1855 dsl_pool_t *dp = spa_get_dsl(spa);
1856
1857 switch (spa->spa_log_state) {
1858 case SPA_LOG_MISSING:
1859 /* need to recheck in case slog has been restored */
1860 case SPA_LOG_UNKNOWN:
1861 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1862 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1863 if (rv)
1864 spa_set_log_state(spa, SPA_LOG_MISSING);
1865 break;
1866 }
1867 return (rv);
1868 }
1869
1870 static boolean_t
1871 spa_passivate_log(spa_t *spa)
1872 {
1873 vdev_t *rvd = spa->spa_root_vdev;
1874 boolean_t slog_found = B_FALSE;
1875
1876 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1877
1878 if (!spa_has_slogs(spa))
1879 return (B_FALSE);
1880
1881 for (int c = 0; c < rvd->vdev_children; c++) {
1882 vdev_t *tvd = rvd->vdev_child[c];
1883 metaslab_group_t *mg = tvd->vdev_mg;
1884
1885 if (tvd->vdev_islog) {
1886 metaslab_group_passivate(mg);
1887 slog_found = B_TRUE;
1888 }
1889 }
1890
1891 return (slog_found);
1892 }
1893
1894 static void
1895 spa_activate_log(spa_t *spa)
1896 {
1897 vdev_t *rvd = spa->spa_root_vdev;
1898
1899 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1900
1901 for (int c = 0; c < rvd->vdev_children; c++) {
1902 vdev_t *tvd = rvd->vdev_child[c];
1903 metaslab_group_t *mg = tvd->vdev_mg;
1904
1905 if (tvd->vdev_islog)
1906 metaslab_group_activate(mg);
1907 }
1908 }
1909
1910 int
1911 spa_reset_logs(spa_t *spa)
1912 {
1913 int error;
1914
1915 error = dmu_objset_find(spa_name(spa), zil_reset,
1916 NULL, DS_FIND_CHILDREN);
1917 if (error == 0) {
1918 /*
1919 * We successfully offlined the log device, sync out the
1920 * current txg so that the "stubby" block can be removed
1921 * by zil_sync().
1922 */
1923 txg_wait_synced(spa->spa_dsl_pool, 0);
1924 }
1925 return (error);
1926 }
1927
1928 static void
1929 spa_aux_check_removed(spa_aux_vdev_t *sav)
1930 {
1931 for (int i = 0; i < sav->sav_count; i++)
1932 spa_check_removed(sav->sav_vdevs[i]);
1933 }
1934
1935 void
1936 spa_claim_notify(zio_t *zio)
1937 {
1938 spa_t *spa = zio->io_spa;
1939
1940 if (zio->io_error)
1941 return;
1942
1943 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1944 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1945 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1946 mutex_exit(&spa->spa_props_lock);
1947 }
1948
1949 typedef struct spa_load_error {
1950 uint64_t sle_meta_count;
1951 uint64_t sle_data_count;
1952 } spa_load_error_t;
1953
1954 static void
1955 spa_load_verify_done(zio_t *zio)
1956 {
1957 blkptr_t *bp = zio->io_bp;
1958 spa_load_error_t *sle = zio->io_private;
1959 dmu_object_type_t type = BP_GET_TYPE(bp);
1960 int error = zio->io_error;
1961 spa_t *spa = zio->io_spa;
1962
1963 abd_free(zio->io_abd);
1964 if (error) {
1965 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1966 type != DMU_OT_INTENT_LOG)
1967 atomic_inc_64(&sle->sle_meta_count);
1968 else
1969 atomic_inc_64(&sle->sle_data_count);
1970 }
1971
1972 mutex_enter(&spa->spa_scrub_lock);
1973 spa->spa_scrub_inflight--;
1974 cv_broadcast(&spa->spa_scrub_io_cv);
1975 mutex_exit(&spa->spa_scrub_lock);
1976 }
1977
1978 /*
1979 * Maximum number of concurrent scrub i/os to create while verifying
1980 * a pool while importing it.
1981 */
1982 int spa_load_verify_maxinflight = 10000;
1983 boolean_t spa_load_verify_metadata = B_TRUE;
1984 boolean_t spa_load_verify_data = B_TRUE;
1985
1986 /*ARGSUSED*/
1987 static int
1988 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1989 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1990 {
1991 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1992 return (0);
1993 /*
1994 * Note: normally this routine will not be called if
1995 * spa_load_verify_metadata is not set. However, it may be useful
1996 * to manually set the flag after the traversal has begun.
1997 */
1998 if (!spa_load_verify_metadata)
1999 return (0);
2000 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2001 return (0);
2002
2003 zio_t *rio = arg;
2004 size_t size = BP_GET_PSIZE(bp);
2005
2006 mutex_enter(&spa->spa_scrub_lock);
2007 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2008 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2009 spa->spa_scrub_inflight++;
2010 mutex_exit(&spa->spa_scrub_lock);
2011
2012 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2013 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2014 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2015 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2016 return (0);
2017 }
2018
2019 /* ARGSUSED */
2020 int
2021 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2022 {
2023 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2024 return (SET_ERROR(ENAMETOOLONG));
2025
2026 return (0);
2027 }
2028
2029 static int
2030 spa_load_verify(spa_t *spa)
2031 {
2032 zio_t *rio;
2033 spa_load_error_t sle = { 0 };
2034 zpool_load_policy_t policy;
2035 boolean_t verify_ok = B_FALSE;
2036 int error = 0;
2037
2038 zpool_get_load_policy(spa->spa_config, &policy);
2039
2040 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2041 return (0);
2042
2043 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2044 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2045 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2046 DS_FIND_CHILDREN);
2047 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2048 if (error != 0)
2049 return (error);
2050
2051 rio = zio_root(spa, NULL, &sle,
2052 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2053
2054 if (spa_load_verify_metadata) {
2055 if (spa->spa_extreme_rewind) {
2056 spa_load_note(spa, "performing a complete scan of the "
2057 "pool since extreme rewind is on. This may take "
2058 "a very long time.\n (spa_load_verify_data=%u, "
2059 "spa_load_verify_metadata=%u)",
2060 spa_load_verify_data, spa_load_verify_metadata);
2061 }
2062 error = traverse_pool(spa, spa->spa_verify_min_txg,
2063 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2064 spa_load_verify_cb, rio);
2065 }
2066
2067 (void) zio_wait(rio);
2068
2069 spa->spa_load_meta_errors = sle.sle_meta_count;
2070 spa->spa_load_data_errors = sle.sle_data_count;
2071
2072 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2073 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2074 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2075 (u_longlong_t)sle.sle_data_count);
2076 }
2077
2078 if (spa_load_verify_dryrun ||
2079 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2080 sle.sle_data_count <= policy.zlp_maxdata)) {
2081 int64_t loss = 0;
2082
2083 verify_ok = B_TRUE;
2084 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2085 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2086
2087 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2088 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2089 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2090 VERIFY(nvlist_add_int64(spa->spa_load_info,
2091 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2092 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2093 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2094 } else {
2095 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2096 }
2097
2098 if (spa_load_verify_dryrun)
2099 return (0);
2100
2101 if (error) {
2102 if (error != ENXIO && error != EIO)
2103 error = SET_ERROR(EIO);
2104 return (error);
2105 }
2106
2107 return (verify_ok ? 0 : EIO);
2108 }
2109
2110 /*
2111 * Find a value in the pool props object.
2112 */
2113 static void
2114 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2115 {
2116 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2117 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2118 }
2119
2120 /*
2121 * Find a value in the pool directory object.
2122 */
2123 static int
2124 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2125 {
2126 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2127 name, sizeof (uint64_t), 1, val);
2128
2129 if (error != 0 && (error != ENOENT || log_enoent)) {
2130 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2131 "[error=%d]", name, error);
2132 }
2133
2134 return (error);
2135 }
2136
2137 static int
2138 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2139 {
2140 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2141 return (SET_ERROR(err));
2142 }
2143
2144 static void
2145 spa_spawn_aux_threads(spa_t *spa)
2146 {
2147 ASSERT(spa_writeable(spa));
2148
2149 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2150
2151 spa_start_indirect_condensing_thread(spa);
2152
2153 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2154 spa->spa_checkpoint_discard_zthr =
2155 zthr_create(spa_checkpoint_discard_thread_check,
2156 spa_checkpoint_discard_thread, spa);
2157 }
2158
2159 /*
2160 * Fix up config after a partly-completed split. This is done with the
2161 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2162 * pool have that entry in their config, but only the splitting one contains
2163 * a list of all the guids of the vdevs that are being split off.
2164 *
2165 * This function determines what to do with that list: either rejoin
2166 * all the disks to the pool, or complete the splitting process. To attempt
2167 * the rejoin, each disk that is offlined is marked online again, and
2168 * we do a reopen() call. If the vdev label for every disk that was
2169 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2170 * then we call vdev_split() on each disk, and complete the split.
2171 *
2172 * Otherwise we leave the config alone, with all the vdevs in place in
2173 * the original pool.
2174 */
2175 static void
2176 spa_try_repair(spa_t *spa, nvlist_t *config)
2177 {
2178 uint_t extracted;
2179 uint64_t *glist;
2180 uint_t i, gcount;
2181 nvlist_t *nvl;
2182 vdev_t **vd;
2183 boolean_t attempt_reopen;
2184
2185 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2186 return;
2187
2188 /* check that the config is complete */
2189 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2190 &glist, &gcount) != 0)
2191 return;
2192
2193 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2194
2195 /* attempt to online all the vdevs & validate */
2196 attempt_reopen = B_TRUE;
2197 for (i = 0; i < gcount; i++) {
2198 if (glist[i] == 0) /* vdev is hole */
2199 continue;
2200
2201 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2202 if (vd[i] == NULL) {
2203 /*
2204 * Don't bother attempting to reopen the disks;
2205 * just do the split.
2206 */
2207 attempt_reopen = B_FALSE;
2208 } else {
2209 /* attempt to re-online it */
2210 vd[i]->vdev_offline = B_FALSE;
2211 }
2212 }
2213
2214 if (attempt_reopen) {
2215 vdev_reopen(spa->spa_root_vdev);
2216
2217 /* check each device to see what state it's in */
2218 for (extracted = 0, i = 0; i < gcount; i++) {
2219 if (vd[i] != NULL &&
2220 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2221 break;
2222 ++extracted;
2223 }
2224 }
2225
2226 /*
2227 * If every disk has been moved to the new pool, or if we never
2228 * even attempted to look at them, then we split them off for
2229 * good.
2230 */
2231 if (!attempt_reopen || gcount == extracted) {
2232 for (i = 0; i < gcount; i++)
2233 if (vd[i] != NULL)
2234 vdev_split(vd[i]);
2235 vdev_reopen(spa->spa_root_vdev);
2236 }
2237
2238 kmem_free(vd, gcount * sizeof (vdev_t *));
2239 }
2240
2241 static int
2242 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2243 {
2244 char *ereport = FM_EREPORT_ZFS_POOL;
2245 int error;
2246
2247 spa->spa_load_state = state;
2248
2249 gethrestime(&spa->spa_loaded_ts);
2250 error = spa_load_impl(spa, type, &ereport);
2251
2252 /*
2253 * Don't count references from objsets that are already closed
2254 * and are making their way through the eviction process.
2255 */
2256 spa_evicting_os_wait(spa);
2257 spa->spa_minref = refcount_count(&spa->spa_refcount);
2258 if (error) {
2259 if (error != EEXIST) {
2260 spa->spa_loaded_ts.tv_sec = 0;
2261 spa->spa_loaded_ts.tv_nsec = 0;
2262 }
2263 if (error != EBADF) {
2264 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2265 }
2266 }
2267 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2268 spa->spa_ena = 0;
2269
2270 return (error);
2271 }
2272
2273 /*
2274 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2275 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2276 * spa's per-vdev ZAP list.
2277 */
2278 static uint64_t
2279 vdev_count_verify_zaps(vdev_t *vd)
2280 {
2281 spa_t *spa = vd->vdev_spa;
2282 uint64_t total = 0;
2283 if (vd->vdev_top_zap != 0) {
2284 total++;
2285 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2286 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2287 }
2288 if (vd->vdev_leaf_zap != 0) {
2289 total++;
2290 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2291 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2292 }
2293
2294 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2295 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2296 }
2297
2298 return (total);
2299 }
2300
2301 static int
2302 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2303 {
2304 uint64_t hostid;
2305 char *hostname;
2306 uint64_t myhostid = 0;
2307
2308 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2309 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2310 hostname = fnvlist_lookup_string(mos_config,
2311 ZPOOL_CONFIG_HOSTNAME);
2312
2313 myhostid = zone_get_hostid(NULL);
2314
2315 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2316 cmn_err(CE_WARN, "pool '%s' could not be "
2317 "loaded as it was last accessed by "
2318 "another system (host: %s hostid: 0x%llx). "
2319 "See: http://illumos.org/msg/ZFS-8000-EY",
2320 spa_name(spa), hostname, (u_longlong_t)hostid);
2321 spa_load_failed(spa, "hostid verification failed: pool "
2322 "last accessed by host: %s (hostid: 0x%llx)",
2323 hostname, (u_longlong_t)hostid);
2324 return (SET_ERROR(EBADF));
2325 }
2326 }
2327
2328 return (0);
2329 }
2330
2331 static int
2332 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2333 {
2334 int error = 0;
2335 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2336 int parse;
2337 vdev_t *rvd;
2338 uint64_t pool_guid;
2339 char *comment;
2340
2341 /*
2342 * Versioning wasn't explicitly added to the label until later, so if
2343 * it's not present treat it as the initial version.
2344 */
2345 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2346 &spa->spa_ubsync.ub_version) != 0)
2347 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2348
2349 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2350 spa_load_failed(spa, "invalid config provided: '%s' missing",
2351 ZPOOL_CONFIG_POOL_GUID);
2352 return (SET_ERROR(EINVAL));
2353 }
2354
2355 /*
2356 * If we are doing an import, ensure that the pool is not already
2357 * imported by checking if its pool guid already exists in the
2358 * spa namespace.
2359 *
2360 * The only case that we allow an already imported pool to be
2361 * imported again, is when the pool is checkpointed and we want to
2362 * look at its checkpointed state from userland tools like zdb.
2363 */
2364 #ifdef _KERNEL
2365 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2366 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2367 spa_guid_exists(pool_guid, 0)) {
2368 #else
2369 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2370 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2371 spa_guid_exists(pool_guid, 0) &&
2372 !spa_importing_readonly_checkpoint(spa)) {
2373 #endif
2374 spa_load_failed(spa, "a pool with guid %llu is already open",
2375 (u_longlong_t)pool_guid);
2376 return (SET_ERROR(EEXIST));
2377 }
2378
2379 spa->spa_config_guid = pool_guid;
2380
2381 nvlist_free(spa->spa_load_info);
2382 spa->spa_load_info = fnvlist_alloc();
2383
2384 ASSERT(spa->spa_comment == NULL);
2385 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2386 spa->spa_comment = spa_strdup(comment);
2387
2388 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2389 &spa->spa_config_txg);
2390
2391 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2392 spa->spa_config_splitting = fnvlist_dup(nvl);
2393
2394 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2395 spa_load_failed(spa, "invalid config provided: '%s' missing",
2396 ZPOOL_CONFIG_VDEV_TREE);
2397 return (SET_ERROR(EINVAL));
2398 }
2399
2400 /*
2401 * Create "The Godfather" zio to hold all async IOs
2402 */
2403 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2404 KM_SLEEP);
2405 for (int i = 0; i < max_ncpus; i++) {
2406 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2407 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2408 ZIO_FLAG_GODFATHER);
2409 }
2410
2411 /*
2412 * Parse the configuration into a vdev tree. We explicitly set the
2413 * value that will be returned by spa_version() since parsing the
2414 * configuration requires knowing the version number.
2415 */
2416 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2417 parse = (type == SPA_IMPORT_EXISTING ?
2418 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2419 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2420 spa_config_exit(spa, SCL_ALL, FTAG);
2421
2422 if (error != 0) {
2423 spa_load_failed(spa, "unable to parse config [error=%d]",
2424 error);
2425 return (error);
2426 }
2427
2428 ASSERT(spa->spa_root_vdev == rvd);
2429 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2430 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2431
2432 if (type != SPA_IMPORT_ASSEMBLE) {
2433 ASSERT(spa_guid(spa) == pool_guid);
2434 }
2435
2436 return (0);
2437 }
2438
2439 /*
2440 * Recursively open all vdevs in the vdev tree. This function is called twice:
2441 * first with the untrusted config, then with the trusted config.
2442 */
2443 static int
2444 spa_ld_open_vdevs(spa_t *spa)
2445 {
2446 int error = 0;
2447
2448 /*
2449 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2450 * missing/unopenable for the root vdev to be still considered openable.
2451 */
2452 if (spa->spa_trust_config) {
2453 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2454 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2455 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2456 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2457 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2458 } else {
2459 spa->spa_missing_tvds_allowed = 0;
2460 }
2461
2462 spa->spa_missing_tvds_allowed =
2463 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2464
2465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2466 error = vdev_open(spa->spa_root_vdev);
2467 spa_config_exit(spa, SCL_ALL, FTAG);
2468
2469 if (spa->spa_missing_tvds != 0) {
2470 spa_load_note(spa, "vdev tree has %lld missing top-level "
2471 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2472 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2473 /*
2474 * Although theoretically we could allow users to open
2475 * incomplete pools in RW mode, we'd need to add a lot
2476 * of extra logic (e.g. adjust pool space to account
2477 * for missing vdevs).
2478 * This limitation also prevents users from accidentally
2479 * opening the pool in RW mode during data recovery and
2480 * damaging it further.
2481 */
2482 spa_load_note(spa, "pools with missing top-level "
2483 "vdevs can only be opened in read-only mode.");
2484 error = SET_ERROR(ENXIO);
2485 } else {
2486 spa_load_note(spa, "current settings allow for maximum "
2487 "%lld missing top-level vdevs at this stage.",
2488 (u_longlong_t)spa->spa_missing_tvds_allowed);
2489 }
2490 }
2491 if (error != 0) {
2492 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2493 error);
2494 }
2495 if (spa->spa_missing_tvds != 0 || error != 0)
2496 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2497
2498 return (error);
2499 }
2500
2501 /*
2502 * We need to validate the vdev labels against the configuration that
2503 * we have in hand. This function is called twice: first with an untrusted
2504 * config, then with a trusted config. The validation is more strict when the
2505 * config is trusted.
2506 */
2507 static int
2508 spa_ld_validate_vdevs(spa_t *spa)
2509 {
2510 int error = 0;
2511 vdev_t *rvd = spa->spa_root_vdev;
2512
2513 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2514 error = vdev_validate(rvd);
2515 spa_config_exit(spa, SCL_ALL, FTAG);
2516
2517 if (error != 0) {
2518 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2519 return (error);
2520 }
2521
2522 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2523 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2524 "some vdevs");
2525 vdev_dbgmsg_print_tree(rvd, 2);
2526 return (SET_ERROR(ENXIO));
2527 }
2528
2529 return (0);
2530 }
2531
2532 static void
2533 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2534 {
2535 spa->spa_state = POOL_STATE_ACTIVE;
2536 spa->spa_ubsync = spa->spa_uberblock;
2537 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2538 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2539 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2540 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2541 spa->spa_claim_max_txg = spa->spa_first_txg;
2542 spa->spa_prev_software_version = ub->ub_software_version;
2543 }
2544
2545 static int
2546 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2547 {
2548 vdev_t *rvd = spa->spa_root_vdev;
2549 nvlist_t *label;
2550 uberblock_t *ub = &spa->spa_uberblock;
2551
2552 /*
2553 * If we are opening the checkpointed state of the pool by
2554 * rewinding to it, at this point we will have written the
2555 * checkpointed uberblock to the vdev labels, so searching
2556 * the labels will find the right uberblock. However, if
2557 * we are opening the checkpointed state read-only, we have
2558 * not modified the labels. Therefore, we must ignore the
2559 * labels and continue using the spa_uberblock that was set
2560 * by spa_ld_checkpoint_rewind.
2561 *
2562 * Note that it would be fine to ignore the labels when
2563 * rewinding (opening writeable) as well. However, if we
2564 * crash just after writing the labels, we will end up
2565 * searching the labels. Doing so in the common case means
2566 * that this code path gets exercised normally, rather than
2567 * just in the edge case.
2568 */
2569 if (ub->ub_checkpoint_txg != 0 &&
2570 spa_importing_readonly_checkpoint(spa)) {
2571 spa_ld_select_uberblock_done(spa, ub);
2572 return (0);
2573 }
2574
2575 /*
2576 * Find the best uberblock.
2577 */
2578 vdev_uberblock_load(rvd, ub, &label);
2579
2580 /*
2581 * If we weren't able to find a single valid uberblock, return failure.
2582 */
2583 if (ub->ub_txg == 0) {
2584 nvlist_free(label);
2585 spa_load_failed(spa, "no valid uberblock found");
2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2587 }
2588
2589 spa_load_note(spa, "using uberblock with txg=%llu",
2590 (u_longlong_t)ub->ub_txg);
2591
2592 /*
2593 * If the pool has an unsupported version we can't open it.
2594 */
2595 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2596 nvlist_free(label);
2597 spa_load_failed(spa, "version %llu is not supported",
2598 (u_longlong_t)ub->ub_version);
2599 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2600 }
2601
2602 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2603 nvlist_t *features;
2604
2605 /*
2606 * If we weren't able to find what's necessary for reading the
2607 * MOS in the label, return failure.
2608 */
2609 if (label == NULL) {
2610 spa_load_failed(spa, "label config unavailable");
2611 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2612 ENXIO));
2613 }
2614
2615 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2616 &features) != 0) {
2617 nvlist_free(label);
2618 spa_load_failed(spa, "invalid label: '%s' missing",
2619 ZPOOL_CONFIG_FEATURES_FOR_READ);
2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2621 ENXIO));
2622 }
2623
2624 /*
2625 * Update our in-core representation with the definitive values
2626 * from the label.
2627 */
2628 nvlist_free(spa->spa_label_features);
2629 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2630 }
2631
2632 nvlist_free(label);
2633
2634 /*
2635 * Look through entries in the label nvlist's features_for_read. If
2636 * there is a feature listed there which we don't understand then we
2637 * cannot open a pool.
2638 */
2639 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2640 nvlist_t *unsup_feat;
2641
2642 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2643 0);
2644
2645 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2646 NULL); nvp != NULL;
2647 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2648 if (!zfeature_is_supported(nvpair_name(nvp))) {
2649 VERIFY(nvlist_add_string(unsup_feat,
2650 nvpair_name(nvp), "") == 0);
2651 }
2652 }
2653
2654 if (!nvlist_empty(unsup_feat)) {
2655 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2656 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2657 nvlist_free(unsup_feat);
2658 spa_load_failed(spa, "some features are unsupported");
2659 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2660 ENOTSUP));
2661 }
2662
2663 nvlist_free(unsup_feat);
2664 }
2665
2666 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2667 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2668 spa_try_repair(spa, spa->spa_config);
2669 spa_config_exit(spa, SCL_ALL, FTAG);
2670 nvlist_free(spa->spa_config_splitting);
2671 spa->spa_config_splitting = NULL;
2672 }
2673
2674 /*
2675 * Initialize internal SPA structures.
2676 */
2677 spa_ld_select_uberblock_done(spa, ub);
2678
2679 return (0);
2680 }
2681
2682 static int
2683 spa_ld_open_rootbp(spa_t *spa)
2684 {
2685 int error = 0;
2686 vdev_t *rvd = spa->spa_root_vdev;
2687
2688 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2689 if (error != 0) {
2690 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2691 "[error=%d]", error);
2692 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2693 }
2694 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2695
2696 return (0);
2697 }
2698
2699 static int
2700 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2701 boolean_t reloading)
2702 {
2703 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2704 nvlist_t *nv, *mos_config, *policy;
2705 int error = 0, copy_error;
2706 uint64_t healthy_tvds, healthy_tvds_mos;
2707 uint64_t mos_config_txg;
2708
2709 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2710 != 0)
2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2712
2713 /*
2714 * If we're assembling a pool from a split, the config provided is
2715 * already trusted so there is nothing to do.
2716 */
2717 if (type == SPA_IMPORT_ASSEMBLE)
2718 return (0);
2719
2720 healthy_tvds = spa_healthy_core_tvds(spa);
2721
2722 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2723 != 0) {
2724 spa_load_failed(spa, "unable to retrieve MOS config");
2725 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2726 }
2727
2728 /*
2729 * If we are doing an open, pool owner wasn't verified yet, thus do
2730 * the verification here.
2731 */
2732 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2733 error = spa_verify_host(spa, mos_config);
2734 if (error != 0) {
2735 nvlist_free(mos_config);
2736 return (error);
2737 }
2738 }
2739
2740 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2741
2742 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2743
2744 /*
2745 * Build a new vdev tree from the trusted config
2746 */
2747 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2748
2749 /*
2750 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2751 * obtained by scanning /dev/dsk, then it will have the right vdev
2752 * paths. We update the trusted MOS config with this information.
2753 * We first try to copy the paths with vdev_copy_path_strict, which
2754 * succeeds only when both configs have exactly the same vdev tree.
2755 * If that fails, we fall back to a more flexible method that has a
2756 * best effort policy.
2757 */
2758 copy_error = vdev_copy_path_strict(rvd, mrvd);
2759 if (copy_error != 0 || spa_load_print_vdev_tree) {
2760 spa_load_note(spa, "provided vdev tree:");
2761 vdev_dbgmsg_print_tree(rvd, 2);
2762 spa_load_note(spa, "MOS vdev tree:");
2763 vdev_dbgmsg_print_tree(mrvd, 2);
2764 }
2765 if (copy_error != 0) {
2766 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2767 "back to vdev_copy_path_relaxed");
2768 vdev_copy_path_relaxed(rvd, mrvd);
2769 }
2770
2771 vdev_close(rvd);
2772 vdev_free(rvd);
2773 spa->spa_root_vdev = mrvd;
2774 rvd = mrvd;
2775 spa_config_exit(spa, SCL_ALL, FTAG);
2776
2777 /*
2778 * We will use spa_config if we decide to reload the spa or if spa_load
2779 * fails and we rewind. We must thus regenerate the config using the
2780 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
2781 * pass settings on how to load the pool and is not stored in the MOS.
2782 * We copy it over to our new, trusted config.
2783 */
2784 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2785 ZPOOL_CONFIG_POOL_TXG);
2786 nvlist_free(mos_config);
2787 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
2788 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
2789 &policy) == 0)
2790 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
2791 spa_config_set(spa, mos_config);
2792 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2793
2794 /*
2795 * Now that we got the config from the MOS, we should be more strict
2796 * in checking blkptrs and can make assumptions about the consistency
2797 * of the vdev tree. spa_trust_config must be set to true before opening
2798 * vdevs in order for them to be writeable.
2799 */
2800 spa->spa_trust_config = B_TRUE;
2801
2802 /*
2803 * Open and validate the new vdev tree
2804 */
2805 error = spa_ld_open_vdevs(spa);
2806 if (error != 0)
2807 return (error);
2808
2809 error = spa_ld_validate_vdevs(spa);
2810 if (error != 0)
2811 return (error);
2812
2813 if (copy_error != 0 || spa_load_print_vdev_tree) {
2814 spa_load_note(spa, "final vdev tree:");
2815 vdev_dbgmsg_print_tree(rvd, 2);
2816 }
2817
2818 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2819 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2820 /*
2821 * Sanity check to make sure that we are indeed loading the
2822 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2823 * in the config provided and they happened to be the only ones
2824 * to have the latest uberblock, we could involuntarily perform
2825 * an extreme rewind.
2826 */
2827 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2828 if (healthy_tvds_mos - healthy_tvds >=
2829 SPA_SYNC_MIN_VDEVS) {
2830 spa_load_note(spa, "config provided misses too many "
2831 "top-level vdevs compared to MOS (%lld vs %lld). ",
2832 (u_longlong_t)healthy_tvds,
2833 (u_longlong_t)healthy_tvds_mos);
2834 spa_load_note(spa, "vdev tree:");
2835 vdev_dbgmsg_print_tree(rvd, 2);
2836 if (reloading) {
2837 spa_load_failed(spa, "config was already "
2838 "provided from MOS. Aborting.");
2839 return (spa_vdev_err(rvd,
2840 VDEV_AUX_CORRUPT_DATA, EIO));
2841 }
2842 spa_load_note(spa, "spa must be reloaded using MOS "
2843 "config");
2844 return (SET_ERROR(EAGAIN));
2845 }
2846 }
2847
2848 error = spa_check_for_missing_logs(spa);
2849 if (error != 0)
2850 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2851
2852 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2853 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2854 "guid sum (%llu != %llu)",
2855 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2856 (u_longlong_t)rvd->vdev_guid_sum);
2857 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2858 ENXIO));
2859 }
2860
2861 return (0);
2862 }
2863
2864 static int
2865 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2866 {
2867 int error = 0;
2868 vdev_t *rvd = spa->spa_root_vdev;
2869
2870 /*
2871 * Everything that we read before spa_remove_init() must be stored
2872 * on concreted vdevs. Therefore we do this as early as possible.
2873 */
2874 error = spa_remove_init(spa);
2875 if (error != 0) {
2876 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2877 error);
2878 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2879 }
2880
2881 /*
2882 * Retrieve information needed to condense indirect vdev mappings.
2883 */
2884 error = spa_condense_init(spa);
2885 if (error != 0) {
2886 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2887 error);
2888 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2889 }
2890
2891 return (0);
2892 }
2893
2894 static int
2895 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2896 {
2897 int error = 0;
2898 vdev_t *rvd = spa->spa_root_vdev;
2899
2900 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2901 boolean_t missing_feat_read = B_FALSE;
2902 nvlist_t *unsup_feat, *enabled_feat;
2903
2904 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2905 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2906 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2907 }
2908
2909 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2910 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
2911 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2912 }
2913
2914 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2915 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
2916 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2917 }
2918
2919 enabled_feat = fnvlist_alloc();
2920 unsup_feat = fnvlist_alloc();
2921
2922 if (!spa_features_check(spa, B_FALSE,
2923 unsup_feat, enabled_feat))
2924 missing_feat_read = B_TRUE;
2925
2926 if (spa_writeable(spa) ||
2927 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
2928 if (!spa_features_check(spa, B_TRUE,
2929 unsup_feat, enabled_feat)) {
2930 *missing_feat_writep = B_TRUE;
2931 }
2932 }
2933
2934 fnvlist_add_nvlist(spa->spa_load_info,
2935 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2936
2937 if (!nvlist_empty(unsup_feat)) {
2938 fnvlist_add_nvlist(spa->spa_load_info,
2939 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2940 }
2941
2942 fnvlist_free(enabled_feat);
2943 fnvlist_free(unsup_feat);
2944
2945 if (!missing_feat_read) {
2946 fnvlist_add_boolean(spa->spa_load_info,
2947 ZPOOL_CONFIG_CAN_RDONLY);
2948 }
2949
2950 /*
2951 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2952 * twofold: to determine whether the pool is available for
2953 * import in read-write mode and (if it is not) whether the
2954 * pool is available for import in read-only mode. If the pool
2955 * is available for import in read-write mode, it is displayed
2956 * as available in userland; if it is not available for import
2957 * in read-only mode, it is displayed as unavailable in
2958 * userland. If the pool is available for import in read-only
2959 * mode but not read-write mode, it is displayed as unavailable
2960 * in userland with a special note that the pool is actually
2961 * available for open in read-only mode.
2962 *
2963 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2964 * missing a feature for write, we must first determine whether
2965 * the pool can be opened read-only before returning to
2966 * userland in order to know whether to display the
2967 * abovementioned note.
2968 */
2969 if (missing_feat_read || (*missing_feat_writep &&
2970 spa_writeable(spa))) {
2971 spa_load_failed(spa, "pool uses unsupported features");
2972 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2973 ENOTSUP));
2974 }
2975
2976 /*
2977 * Load refcounts for ZFS features from disk into an in-memory
2978 * cache during SPA initialization.
2979 */
2980 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2981 uint64_t refcount;
2982
2983 error = feature_get_refcount_from_disk(spa,
2984 &spa_feature_table[i], &refcount);
2985 if (error == 0) {
2986 spa->spa_feat_refcount_cache[i] = refcount;
2987 } else if (error == ENOTSUP) {
2988 spa->spa_feat_refcount_cache[i] =
2989 SPA_FEATURE_DISABLED;
2990 } else {
2991 spa_load_failed(spa, "error getting refcount "
2992 "for feature %s [error=%d]",
2993 spa_feature_table[i].fi_guid, error);
2994 return (spa_vdev_err(rvd,
2995 VDEV_AUX_CORRUPT_DATA, EIO));
2996 }
2997 }
2998 }
2999
3000 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3001 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3002 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3003 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3004 }
3005
3006 return (0);
3007 }
3008
3009 static int
3010 spa_ld_load_special_directories(spa_t *spa)
3011 {
3012 int error = 0;
3013 vdev_t *rvd = spa->spa_root_vdev;
3014
3015 spa->spa_is_initializing = B_TRUE;
3016 error = dsl_pool_open(spa->spa_dsl_pool);
3017 spa->spa_is_initializing = B_FALSE;
3018 if (error != 0) {
3019 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3020 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3021 }
3022
3023 return (0);
3024 }
3025
3026 static int
3027 spa_ld_get_props(spa_t *spa)
3028 {
3029 int error = 0;
3030 uint64_t obj;
3031 vdev_t *rvd = spa->spa_root_vdev;
3032
3033 /* Grab the secret checksum salt from the MOS. */
3034 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3035 DMU_POOL_CHECKSUM_SALT, 1,
3036 sizeof (spa->spa_cksum_salt.zcs_bytes),
3037 spa->spa_cksum_salt.zcs_bytes);
3038 if (error == ENOENT) {
3039 /* Generate a new salt for subsequent use */
3040 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3041 sizeof (spa->spa_cksum_salt.zcs_bytes));
3042 } else if (error != 0) {
3043 spa_load_failed(spa, "unable to retrieve checksum salt from "
3044 "MOS [error=%d]", error);
3045 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3046 }
3047
3048 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3049 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3050 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3051 if (error != 0) {
3052 spa_load_failed(spa, "error opening deferred-frees bpobj "
3053 "[error=%d]", error);
3054 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3055 }
3056
3057 /*
3058 * Load the bit that tells us to use the new accounting function
3059 * (raid-z deflation). If we have an older pool, this will not
3060 * be present.
3061 */
3062 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3063 if (error != 0 && error != ENOENT)
3064 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3065
3066 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3067 &spa->spa_creation_version, B_FALSE);
3068 if (error != 0 && error != ENOENT)
3069 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3070
3071 /*
3072 * Load the persistent error log. If we have an older pool, this will
3073 * not be present.
3074 */
3075 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3076 B_FALSE);
3077 if (error != 0 && error != ENOENT)
3078 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3079
3080 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3081 &spa->spa_errlog_scrub, B_FALSE);
3082 if (error != 0 && error != ENOENT)
3083 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3084
3085 /*
3086 * Load the history object. If we have an older pool, this
3087 * will not be present.
3088 */
3089 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3090 if (error != 0 && error != ENOENT)
3091 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3092
3093 /*
3094 * Load the per-vdev ZAP map. If we have an older pool, this will not
3095 * be present; in this case, defer its creation to a later time to
3096 * avoid dirtying the MOS this early / out of sync context. See
3097 * spa_sync_config_object.
3098 */
3099
3100 /* The sentinel is only available in the MOS config. */
3101 nvlist_t *mos_config;
3102 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3103 spa_load_failed(spa, "unable to retrieve MOS config");
3104 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3105 }
3106
3107 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3108 &spa->spa_all_vdev_zaps, B_FALSE);
3109
3110 if (error == ENOENT) {
3111 VERIFY(!nvlist_exists(mos_config,
3112 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3113 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3114 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3115 } else if (error != 0) {
3116 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3117 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3118 /*
3119 * An older version of ZFS overwrote the sentinel value, so
3120 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3121 * destruction to later; see spa_sync_config_object.
3122 */
3123 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3124 /*
3125 * We're assuming that no vdevs have had their ZAPs created
3126 * before this. Better be sure of it.
3127 */
3128 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3129 }
3130 nvlist_free(mos_config);
3131
3132 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3133
3134 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3135 B_FALSE);
3136 if (error && error != ENOENT)
3137 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3138
3139 if (error == 0) {
3140 uint64_t autoreplace;
3141
3142 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3143 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3144 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3145 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3146 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3147 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3148 &spa->spa_dedup_ditto);
3149
3150 spa->spa_autoreplace = (autoreplace != 0);
3151 }
3152
3153 /*
3154 * If we are importing a pool with missing top-level vdevs,
3155 * we enforce that the pool doesn't panic or get suspended on
3156 * error since the likelihood of missing data is extremely high.
3157 */
3158 if (spa->spa_missing_tvds > 0 &&
3159 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3160 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3161 spa_load_note(spa, "forcing failmode to 'continue' "
3162 "as some top level vdevs are missing");
3163 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3164 }
3165
3166 return (0);
3167 }
3168
3169 static int
3170 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3171 {
3172 int error = 0;
3173 vdev_t *rvd = spa->spa_root_vdev;
3174
3175 /*
3176 * If we're assembling the pool from the split-off vdevs of
3177 * an existing pool, we don't want to attach the spares & cache
3178 * devices.
3179 */
3180
3181 /*
3182 * Load any hot spares for this pool.
3183 */
3184 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3185 B_FALSE);
3186 if (error != 0 && error != ENOENT)
3187 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3188 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3189 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3190 if (load_nvlist(spa, spa->spa_spares.sav_object,
3191 &spa->spa_spares.sav_config) != 0) {
3192 spa_load_failed(spa, "error loading spares nvlist");
3193 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3194 }
3195
3196 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3197 spa_load_spares(spa);
3198 spa_config_exit(spa, SCL_ALL, FTAG);
3199 } else if (error == 0) {
3200 spa->spa_spares.sav_sync = B_TRUE;
3201 }
3202
3203 /*
3204 * Load any level 2 ARC devices for this pool.
3205 */
3206 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3207 &spa->spa_l2cache.sav_object, B_FALSE);
3208 if (error != 0 && error != ENOENT)
3209 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3210 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3211 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3212 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3213 &spa->spa_l2cache.sav_config) != 0) {
3214 spa_load_failed(spa, "error loading l2cache nvlist");
3215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3216 }
3217
3218 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3219 spa_load_l2cache(spa);
3220 spa_config_exit(spa, SCL_ALL, FTAG);
3221 } else if (error == 0) {
3222 spa->spa_l2cache.sav_sync = B_TRUE;
3223 }
3224
3225 return (0);
3226 }
3227
3228 static int
3229 spa_ld_load_vdev_metadata(spa_t *spa)
3230 {
3231 int error = 0;
3232 vdev_t *rvd = spa->spa_root_vdev;
3233
3234 /*
3235 * If the 'autoreplace' property is set, then post a resource notifying
3236 * the ZFS DE that it should not issue any faults for unopenable
3237 * devices. We also iterate over the vdevs, and post a sysevent for any
3238 * unopenable vdevs so that the normal autoreplace handler can take
3239 * over.
3240 */
3241 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3242 spa_check_removed(spa->spa_root_vdev);
3243 /*
3244 * For the import case, this is done in spa_import(), because
3245 * at this point we're using the spare definitions from
3246 * the MOS config, not necessarily from the userland config.
3247 */
3248 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3249 spa_aux_check_removed(&spa->spa_spares);
3250 spa_aux_check_removed(&spa->spa_l2cache);
3251 }
3252 }
3253
3254 /*
3255 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3256 */
3257 error = vdev_load(rvd);
3258 if (error != 0) {
3259 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3260 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3261 }
3262
3263 /*
3264 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3265 */
3266 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3267 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3268 spa_config_exit(spa, SCL_ALL, FTAG);
3269
3270 return (0);
3271 }
3272
3273 static int
3274 spa_ld_load_dedup_tables(spa_t *spa)
3275 {
3276 int error = 0;
3277 vdev_t *rvd = spa->spa_root_vdev;
3278
3279 error = ddt_load(spa);
3280 if (error != 0) {
3281 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3282 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3283 }
3284
3285 return (0);
3286 }
3287
3288 static int
3289 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3290 {
3291 vdev_t *rvd = spa->spa_root_vdev;
3292
3293 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3294 boolean_t missing = spa_check_logs(spa);
3295 if (missing) {
3296 if (spa->spa_missing_tvds != 0) {
3297 spa_load_note(spa, "spa_check_logs failed "
3298 "so dropping the logs");
3299 } else {
3300 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3301 spa_load_failed(spa, "spa_check_logs failed");
3302 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3303 ENXIO));
3304 }
3305 }
3306 }
3307
3308 return (0);
3309 }
3310
3311 static int
3312 spa_ld_verify_pool_data(spa_t *spa)
3313 {
3314 int error = 0;
3315 vdev_t *rvd = spa->spa_root_vdev;
3316
3317 /*
3318 * We've successfully opened the pool, verify that we're ready
3319 * to start pushing transactions.
3320 */
3321 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3322 error = spa_load_verify(spa);
3323 if (error != 0) {
3324 spa_load_failed(spa, "spa_load_verify failed "
3325 "[error=%d]", error);
3326 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3327 error));
3328 }
3329 }
3330
3331 return (0);
3332 }
3333
3334 static void
3335 spa_ld_claim_log_blocks(spa_t *spa)
3336 {
3337 dmu_tx_t *tx;
3338 dsl_pool_t *dp = spa_get_dsl(spa);
3339
3340 /*
3341 * Claim log blocks that haven't been committed yet.
3342 * This must all happen in a single txg.
3343 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3344 * invoked from zil_claim_log_block()'s i/o done callback.
3345 * Price of rollback is that we abandon the log.
3346 */
3347 spa->spa_claiming = B_TRUE;
3348
3349 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3350 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3351 zil_claim, tx, DS_FIND_CHILDREN);
3352 dmu_tx_commit(tx);
3353
3354 spa->spa_claiming = B_FALSE;
3355
3356 spa_set_log_state(spa, SPA_LOG_GOOD);
3357 }
3358
3359 static void
3360 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3361 boolean_t update_config_cache)
3362 {
3363 vdev_t *rvd = spa->spa_root_vdev;
3364 int need_update = B_FALSE;
3365
3366 /*
3367 * If the config cache is stale, or we have uninitialized
3368 * metaslabs (see spa_vdev_add()), then update the config.
3369 *
3370 * If this is a verbatim import, trust the current
3371 * in-core spa_config and update the disk labels.
3372 */
3373 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3374 spa->spa_load_state == SPA_LOAD_IMPORT ||
3375 spa->spa_load_state == SPA_LOAD_RECOVER ||
3376 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3377 need_update = B_TRUE;
3378
3379 for (int c = 0; c < rvd->vdev_children; c++)
3380 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3381 need_update = B_TRUE;
3382
3383 /*
3384 * Update the config cache asychronously in case we're the
3385 * root pool, in which case the config cache isn't writable yet.
3386 */
3387 if (need_update)
3388 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3389 }
3390
3391 static void
3392 spa_ld_prepare_for_reload(spa_t *spa)
3393 {
3394 int mode = spa->spa_mode;
3395 int async_suspended = spa->spa_async_suspended;
3396
3397 spa_unload(spa);
3398 spa_deactivate(spa);
3399 spa_activate(spa, mode);
3400
3401 /*
3402 * We save the value of spa_async_suspended as it gets reset to 0 by
3403 * spa_unload(). We want to restore it back to the original value before
3404 * returning as we might be calling spa_async_resume() later.
3405 */
3406 spa->spa_async_suspended = async_suspended;
3407 }
3408
3409 static int
3410 spa_ld_read_checkpoint_txg(spa_t *spa)
3411 {
3412 uberblock_t checkpoint;
3413 int error = 0;
3414
3415 ASSERT0(spa->spa_checkpoint_txg);
3416 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3417
3418 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3419 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3420 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3421
3422 if (error == ENOENT)
3423 return (0);
3424
3425 if (error != 0)
3426 return (error);
3427
3428 ASSERT3U(checkpoint.ub_txg, !=, 0);
3429 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3430 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3431 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3432 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3433
3434 return (0);
3435 }
3436
3437 static int
3438 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3439 {
3440 int error = 0;
3441
3442 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3443 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3444
3445 /*
3446 * Never trust the config that is provided unless we are assembling
3447 * a pool following a split.
3448 * This means don't trust blkptrs and the vdev tree in general. This
3449 * also effectively puts the spa in read-only mode since
3450 * spa_writeable() checks for spa_trust_config to be true.
3451 * We will later load a trusted config from the MOS.
3452 */
3453 if (type != SPA_IMPORT_ASSEMBLE)
3454 spa->spa_trust_config = B_FALSE;
3455
3456 /*
3457 * Parse the config provided to create a vdev tree.
3458 */
3459 error = spa_ld_parse_config(spa, type);
3460 if (error != 0)
3461 return (error);
3462
3463 /*
3464 * Now that we have the vdev tree, try to open each vdev. This involves
3465 * opening the underlying physical device, retrieving its geometry and
3466 * probing the vdev with a dummy I/O. The state of each vdev will be set
3467 * based on the success of those operations. After this we'll be ready
3468 * to read from the vdevs.
3469 */
3470 error = spa_ld_open_vdevs(spa);
3471 if (error != 0)
3472 return (error);
3473
3474 /*
3475 * Read the label of each vdev and make sure that the GUIDs stored
3476 * there match the GUIDs in the config provided.
3477 * If we're assembling a new pool that's been split off from an
3478 * existing pool, the labels haven't yet been updated so we skip
3479 * validation for now.
3480 */
3481 if (type != SPA_IMPORT_ASSEMBLE) {
3482 error = spa_ld_validate_vdevs(spa);
3483 if (error != 0)
3484 return (error);
3485 }
3486
3487 /*
3488 * Read all vdev labels to find the best uberblock (i.e. latest,
3489 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3490 * get the list of features required to read blkptrs in the MOS from
3491 * the vdev label with the best uberblock and verify that our version
3492 * of zfs supports them all.
3493 */
3494 error = spa_ld_select_uberblock(spa, type);
3495 if (error != 0)
3496 return (error);
3497
3498 /*
3499 * Pass that uberblock to the dsl_pool layer which will open the root
3500 * blkptr. This blkptr points to the latest version of the MOS and will
3501 * allow us to read its contents.
3502 */
3503 error = spa_ld_open_rootbp(spa);
3504 if (error != 0)
3505 return (error);
3506
3507 return (0);
3508 }
3509
3510 static int
3511 spa_ld_checkpoint_rewind(spa_t *spa)
3512 {
3513 uberblock_t checkpoint;
3514 int error = 0;
3515
3516 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3517 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3518
3519 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3520 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3521 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3522
3523 if (error != 0) {
3524 spa_load_failed(spa, "unable to retrieve checkpointed "
3525 "uberblock from the MOS config [error=%d]", error);
3526
3527 if (error == ENOENT)
3528 error = ZFS_ERR_NO_CHECKPOINT;
3529
3530 return (error);
3531 }
3532
3533 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3534 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3535
3536 /*
3537 * We need to update the txg and timestamp of the checkpointed
3538 * uberblock to be higher than the latest one. This ensures that
3539 * the checkpointed uberblock is selected if we were to close and
3540 * reopen the pool right after we've written it in the vdev labels.
3541 * (also see block comment in vdev_uberblock_compare)
3542 */
3543 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3544 checkpoint.ub_timestamp = gethrestime_sec();
3545
3546 /*
3547 * Set current uberblock to be the checkpointed uberblock.
3548 */
3549 spa->spa_uberblock = checkpoint;
3550
3551 /*
3552 * If we are doing a normal rewind, then the pool is open for
3553 * writing and we sync the "updated" checkpointed uberblock to
3554 * disk. Once this is done, we've basically rewound the whole
3555 * pool and there is no way back.
3556 *
3557 * There are cases when we don't want to attempt and sync the
3558 * checkpointed uberblock to disk because we are opening a
3559 * pool as read-only. Specifically, verifying the checkpointed
3560 * state with zdb, and importing the checkpointed state to get
3561 * a "preview" of its content.
3562 */
3563 if (spa_writeable(spa)) {
3564 vdev_t *rvd = spa->spa_root_vdev;
3565
3566 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3567 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3568 int svdcount = 0;
3569 int children = rvd->vdev_children;
3570 int c0 = spa_get_random(children);
3571
3572 for (int c = 0; c < children; c++) {
3573 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3574
3575 /* Stop when revisiting the first vdev */
3576 if (c > 0 && svd[0] == vd)
3577 break;
3578
3579 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3580 !vdev_is_concrete(vd))
3581 continue;
3582
3583 svd[svdcount++] = vd;
3584 if (svdcount == SPA_SYNC_MIN_VDEVS)
3585 break;
3586 }
3587 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3588 if (error == 0)
3589 spa->spa_last_synced_guid = rvd->vdev_guid;
3590 spa_config_exit(spa, SCL_ALL, FTAG);
3591
3592 if (error != 0) {
3593 spa_load_failed(spa, "failed to write checkpointed "
3594 "uberblock to the vdev labels [error=%d]", error);
3595 return (error);
3596 }
3597 }
3598
3599 return (0);
3600 }
3601
3602 static int
3603 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3604 boolean_t *update_config_cache)
3605 {
3606 int error;
3607
3608 /*
3609 * Parse the config for pool, open and validate vdevs,
3610 * select an uberblock, and use that uberblock to open
3611 * the MOS.
3612 */
3613 error = spa_ld_mos_init(spa, type);
3614 if (error != 0)
3615 return (error);
3616
3617 /*
3618 * Retrieve the trusted config stored in the MOS and use it to create
3619 * a new, exact version of the vdev tree, then reopen all vdevs.
3620 */
3621 error = spa_ld_trusted_config(spa, type, B_FALSE);
3622 if (error == EAGAIN) {
3623 if (update_config_cache != NULL)
3624 *update_config_cache = B_TRUE;
3625
3626 /*
3627 * Redo the loading process with the trusted config if it is
3628 * too different from the untrusted config.
3629 */
3630 spa_ld_prepare_for_reload(spa);
3631 spa_load_note(spa, "RELOADING");
3632 error = spa_ld_mos_init(spa, type);
3633 if (error != 0)
3634 return (error);
3635
3636 error = spa_ld_trusted_config(spa, type, B_TRUE);
3637 if (error != 0)
3638 return (error);
3639
3640 } else if (error != 0) {
3641 return (error);
3642 }
3643
3644 return (0);
3645 }
3646
3647 /*
3648 * Load an existing storage pool, using the config provided. This config
3649 * describes which vdevs are part of the pool and is later validated against
3650 * partial configs present in each vdev's label and an entire copy of the
3651 * config stored in the MOS.
3652 */
3653 static int
3654 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3655 {
3656 int error = 0;
3657 boolean_t missing_feat_write = B_FALSE;
3658 boolean_t checkpoint_rewind =
3659 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3660 boolean_t update_config_cache = B_FALSE;
3661
3662 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3663 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3664
3665 spa_load_note(spa, "LOADING");
3666
3667 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3668 if (error != 0)
3669 return (error);
3670
3671 /*
3672 * If we are rewinding to the checkpoint then we need to repeat
3673 * everything we've done so far in this function but this time
3674 * selecting the checkpointed uberblock and using that to open
3675 * the MOS.
3676 */
3677 if (checkpoint_rewind) {
3678 /*
3679 * If we are rewinding to the checkpoint update config cache
3680 * anyway.
3681 */
3682 update_config_cache = B_TRUE;
3683
3684 /*
3685 * Extract the checkpointed uberblock from the current MOS
3686 * and use this as the pool's uberblock from now on. If the
3687 * pool is imported as writeable we also write the checkpoint
3688 * uberblock to the labels, making the rewind permanent.
3689 */
3690 error = spa_ld_checkpoint_rewind(spa);
3691 if (error != 0)
3692 return (error);
3693
3694 /*
3695 * Redo the loading process process again with the
3696 * checkpointed uberblock.
3697 */
3698 spa_ld_prepare_for_reload(spa);
3699 spa_load_note(spa, "LOADING checkpointed uberblock");
3700 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3701 if (error != 0)
3702 return (error);
3703 }
3704
3705 /*
3706 * Retrieve the checkpoint txg if the pool has a checkpoint.
3707 */
3708 error = spa_ld_read_checkpoint_txg(spa);
3709 if (error != 0)
3710 return (error);
3711
3712 /*
3713 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3714 * from the pool and their contents were re-mapped to other vdevs. Note
3715 * that everything that we read before this step must have been
3716 * rewritten on concrete vdevs after the last device removal was
3717 * initiated. Otherwise we could be reading from indirect vdevs before
3718 * we have loaded their mappings.
3719 */
3720 error = spa_ld_open_indirect_vdev_metadata(spa);
3721 if (error != 0)
3722 return (error);
3723
3724 /*
3725 * Retrieve the full list of active features from the MOS and check if
3726 * they are all supported.
3727 */
3728 error = spa_ld_check_features(spa, &missing_feat_write);
3729 if (error != 0)
3730 return (error);
3731
3732 /*
3733 * Load several special directories from the MOS needed by the dsl_pool
3734 * layer.
3735 */
3736 error = spa_ld_load_special_directories(spa);
3737 if (error != 0)
3738 return (error);
3739
3740 /*
3741 * Retrieve pool properties from the MOS.
3742 */
3743 error = spa_ld_get_props(spa);
3744 if (error != 0)
3745 return (error);
3746
3747 /*
3748 * Retrieve the list of auxiliary devices - cache devices and spares -
3749 * and open them.
3750 */
3751 error = spa_ld_open_aux_vdevs(spa, type);
3752 if (error != 0)
3753 return (error);
3754
3755 /*
3756 * Load the metadata for all vdevs. Also check if unopenable devices
3757 * should be autoreplaced.
3758 */
3759 error = spa_ld_load_vdev_metadata(spa);
3760 if (error != 0)
3761 return (error);
3762
3763 error = spa_ld_load_dedup_tables(spa);
3764 if (error != 0)
3765 return (error);
3766
3767 /*
3768 * Verify the logs now to make sure we don't have any unexpected errors
3769 * when we claim log blocks later.
3770 */
3771 error = spa_ld_verify_logs(spa, type, ereport);
3772 if (error != 0)
3773 return (error);
3774
3775 if (missing_feat_write) {
3776 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3777
3778 /*
3779 * At this point, we know that we can open the pool in
3780 * read-only mode but not read-write mode. We now have enough
3781 * information and can return to userland.
3782 */
3783 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3784 ENOTSUP));
3785 }
3786
3787 /*
3788 * Traverse the last txgs to make sure the pool was left off in a safe
3789 * state. When performing an extreme rewind, we verify the whole pool,
3790 * which can take a very long time.
3791 */
3792 error = spa_ld_verify_pool_data(spa);
3793 if (error != 0)
3794 return (error);
3795
3796 /*
3797 * Calculate the deflated space for the pool. This must be done before
3798 * we write anything to the pool because we'd need to update the space
3799 * accounting using the deflated sizes.
3800 */
3801 spa_update_dspace(spa);
3802
3803 /*
3804 * We have now retrieved all the information we needed to open the
3805 * pool. If we are importing the pool in read-write mode, a few
3806 * additional steps must be performed to finish the import.
3807 */
3808 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3809 spa->spa_load_max_txg == UINT64_MAX)) {
3810 uint64_t config_cache_txg = spa->spa_config_txg;
3811
3812 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3813
3814 /*
3815 * In case of a checkpoint rewind, log the original txg
3816 * of the checkpointed uberblock.
3817 */
3818 if (checkpoint_rewind) {
3819 spa_history_log_internal(spa, "checkpoint rewind",
3820 NULL, "rewound state to txg=%llu",
3821 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3822 }
3823
3824 /*
3825 * Traverse the ZIL and claim all blocks.
3826 */
3827 spa_ld_claim_log_blocks(spa);
3828
3829 /*
3830 * Kick-off the syncing thread.
3831 */
3832 spa->spa_sync_on = B_TRUE;
3833 txg_sync_start(spa->spa_dsl_pool);
3834
3835 /*
3836 * Wait for all claims to sync. We sync up to the highest
3837 * claimed log block birth time so that claimed log blocks
3838 * don't appear to be from the future. spa_claim_max_txg
3839 * will have been set for us by ZIL traversal operations
3840 * performed above.
3841 */
3842 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3843
3844 /*
3845 * Check if we need to request an update of the config. On the
3846 * next sync, we would update the config stored in vdev labels
3847 * and the cachefile (by default /etc/zfs/zpool.cache).
3848 */
3849 spa_ld_check_for_config_update(spa, config_cache_txg,
3850 update_config_cache);
3851
3852 /*
3853 * Check all DTLs to see if anything needs resilvering.
3854 */
3855 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3856 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3857 spa_async_request(spa, SPA_ASYNC_RESILVER);
3858
3859 /*
3860 * Log the fact that we booted up (so that we can detect if
3861 * we rebooted in the middle of an operation).
3862 */
3863 spa_history_log_version(spa, "open");
3864
3865 spa_restart_removal(spa);
3866 spa_spawn_aux_threads(spa);
3867
3868 /*
3869 * Delete any inconsistent datasets.
3870 *
3871 * Note:
3872 * Since we may be issuing deletes for clones here,
3873 * we make sure to do so after we've spawned all the
3874 * auxiliary threads above (from which the livelist
3875 * deletion zthr is part of).
3876 */
3877 (void) dmu_objset_find(spa_name(spa),
3878 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3879
3880 /*
3881 * Clean up any stale temporary dataset userrefs.
3882 */
3883 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3884
3885 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3886 vdev_initialize_restart(spa->spa_root_vdev);
3887 spa_config_exit(spa, SCL_CONFIG, FTAG);
3888 }
3889
3890 spa_load_note(spa, "LOADED");
3891
3892 return (0);
3893 }
3894
3895 static int
3896 spa_load_retry(spa_t *spa, spa_load_state_t state)
3897 {
3898 int mode = spa->spa_mode;
3899
3900 spa_unload(spa);
3901 spa_deactivate(spa);
3902
3903 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3904
3905 spa_activate(spa, mode);
3906 spa_async_suspend(spa);
3907
3908 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3909 (u_longlong_t)spa->spa_load_max_txg);
3910
3911 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
3912 }
3913
3914 /*
3915 * If spa_load() fails this function will try loading prior txg's. If
3916 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3917 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3918 * function will not rewind the pool and will return the same error as
3919 * spa_load().
3920 */
3921 static int
3922 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
3923 int rewind_flags)
3924 {
3925 nvlist_t *loadinfo = NULL;
3926 nvlist_t *config = NULL;
3927 int load_error, rewind_error;
3928 uint64_t safe_rewind_txg;
3929 uint64_t min_txg;
3930
3931 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
3932 spa->spa_load_max_txg = spa->spa_load_txg;
3933 spa_set_log_state(spa, SPA_LOG_CLEAR);
3934 } else {
3935 spa->spa_load_max_txg = max_request;
3936 if (max_request != UINT64_MAX)
3937 spa->spa_extreme_rewind = B_TRUE;
3938 }
3939
3940 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
3941 if (load_error == 0)
3942 return (0);
3943 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
3944 /*
3945 * When attempting checkpoint-rewind on a pool with no
3946 * checkpoint, we should not attempt to load uberblocks
3947 * from previous txgs when spa_load fails.
3948 */
3949 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3950 return (load_error);
3951 }
3952
3953 if (spa->spa_root_vdev != NULL)
3954 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3955
3956 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
3957 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
3958
3959 if (rewind_flags & ZPOOL_NEVER_REWIND) {
3960 nvlist_free(config);
3961 return (load_error);
3962 }
3963
3964 if (state == SPA_LOAD_RECOVER) {
3965 /* Price of rolling back is discarding txgs, including log */
3966 spa_set_log_state(spa, SPA_LOG_CLEAR);
3967 } else {
3968 /*
3969 * If we aren't rolling back save the load info from our first
3970 * import attempt so that we can restore it after attempting
3971 * to rewind.
3972 */
3973 loadinfo = spa->spa_load_info;
3974 spa->spa_load_info = fnvlist_alloc();
3975 }
3976
3977 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3978 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3979 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3980 TXG_INITIAL : safe_rewind_txg;
3981
3982 /*
3983 * Continue as long as we're finding errors, we're still within
3984 * the acceptable rewind range, and we're still finding uberblocks
3985 */
3986 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3987 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3988 if (spa->spa_load_max_txg < safe_rewind_txg)
3989 spa->spa_extreme_rewind = B_TRUE;
3990 rewind_error = spa_load_retry(spa, state);
3991 }
3992
3993 spa->spa_extreme_rewind = B_FALSE;
3994 spa->spa_load_max_txg = UINT64_MAX;
3995
3996 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3997 spa_config_set(spa, config);
3998 else
3999 nvlist_free(config);
4000
4001 if (state == SPA_LOAD_RECOVER) {
4002 ASSERT3P(loadinfo, ==, NULL);
4003 return (rewind_error);
4004 } else {
4005 /* Store the rewind info as part of the initial load info */
4006 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4007 spa->spa_load_info);
4008
4009 /* Restore the initial load info */
4010 fnvlist_free(spa->spa_load_info);
4011 spa->spa_load_info = loadinfo;
4012
4013 return (load_error);
4014 }
4015 }
4016
4017 /*
4018 * Pool Open/Import
4019 *
4020 * The import case is identical to an open except that the configuration is sent
4021 * down from userland, instead of grabbed from the configuration cache. For the
4022 * case of an open, the pool configuration will exist in the
4023 * POOL_STATE_UNINITIALIZED state.
4024 *
4025 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4026 * the same time open the pool, without having to keep around the spa_t in some
4027 * ambiguous state.
4028 */
4029 static int
4030 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4031 nvlist_t **config)
4032 {
4033 spa_t *spa;
4034 spa_load_state_t state = SPA_LOAD_OPEN;
4035 int error;
4036 int locked = B_FALSE;
4037
4038 *spapp = NULL;
4039
4040 /*
4041 * As disgusting as this is, we need to support recursive calls to this
4042 * function because dsl_dir_open() is called during spa_load(), and ends
4043 * up calling spa_open() again. The real fix is to figure out how to
4044 * avoid dsl_dir_open() calling this in the first place.
4045 */
4046 if (mutex_owner(&spa_namespace_lock) != curthread) {
4047 mutex_enter(&spa_namespace_lock);
4048 locked = B_TRUE;
4049 }
4050
4051 if ((spa = spa_lookup(pool)) == NULL) {
4052 if (locked)
4053 mutex_exit(&spa_namespace_lock);
4054 return (SET_ERROR(ENOENT));
4055 }
4056
4057 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4058 zpool_load_policy_t policy;
4059
4060 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
4061 &policy);
4062 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
4063 state = SPA_LOAD_RECOVER;
4064
4065 spa_activate(spa, spa_mode_global);
4066
4067 if (state != SPA_LOAD_RECOVER)
4068 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4069 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4070
4071 zfs_dbgmsg("spa_open_common: opening %s", pool);
4072 error = spa_load_best(spa, state, policy.zlp_txg,
4073 policy.zlp_rewind);
4074
4075 if (error == EBADF) {
4076 /*
4077 * If vdev_validate() returns failure (indicated by
4078 * EBADF), it indicates that one of the vdevs indicates
4079 * that the pool has been exported or destroyed. If
4080 * this is the case, the config cache is out of sync and
4081 * we should remove the pool from the namespace.
4082 */
4083 spa_unload(spa);
4084 spa_deactivate(spa);
4085 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4086 spa_remove(spa);
4087 if (locked)
4088 mutex_exit(&spa_namespace_lock);
4089 return (SET_ERROR(ENOENT));
4090 }
4091
4092 if (error) {
4093 /*
4094 * We can't open the pool, but we still have useful
4095 * information: the state of each vdev after the
4096 * attempted vdev_open(). Return this to the user.
4097 */
4098 if (config != NULL && spa->spa_config) {
4099 VERIFY(nvlist_dup(spa->spa_config, config,
4100 KM_SLEEP) == 0);
4101 VERIFY(nvlist_add_nvlist(*config,
4102 ZPOOL_CONFIG_LOAD_INFO,
4103 spa->spa_load_info) == 0);
4104 }
4105 spa_unload(spa);
4106 spa_deactivate(spa);
4107 spa->spa_last_open_failed = error;
4108 if (locked)
4109 mutex_exit(&spa_namespace_lock);
4110 *spapp = NULL;
4111 return (error);
4112 }
4113 }
4114
4115 spa_open_ref(spa, tag);
4116
4117 if (config != NULL)
4118 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4119
4120 /*
4121 * If we've recovered the pool, pass back any information we
4122 * gathered while doing the load.
4123 */
4124 if (state == SPA_LOAD_RECOVER) {
4125 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4126 spa->spa_load_info) == 0);
4127 }
4128
4129 if (locked) {
4130 spa->spa_last_open_failed = 0;
4131 spa->spa_last_ubsync_txg = 0;
4132 spa->spa_load_txg = 0;
4133 mutex_exit(&spa_namespace_lock);
4134 }
4135
4136 *spapp = spa;
4137
4138 return (0);
4139 }
4140
4141 int
4142 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4143 nvlist_t **config)
4144 {
4145 return (spa_open_common(name, spapp, tag, policy, config));
4146 }
4147
4148 int
4149 spa_open(const char *name, spa_t **spapp, void *tag)
4150 {
4151 return (spa_open_common(name, spapp, tag, NULL, NULL));
4152 }
4153
4154 /*
4155 * Lookup the given spa_t, incrementing the inject count in the process,
4156 * preventing it from being exported or destroyed.
4157 */
4158 spa_t *
4159 spa_inject_addref(char *name)
4160 {
4161 spa_t *spa;
4162
4163 mutex_enter(&spa_namespace_lock);
4164 if ((spa = spa_lookup(name)) == NULL) {
4165 mutex_exit(&spa_namespace_lock);
4166 return (NULL);
4167 }
4168 spa->spa_inject_ref++;
4169 mutex_exit(&spa_namespace_lock);
4170
4171 return (spa);
4172 }
4173
4174 void
4175 spa_inject_delref(spa_t *spa)
4176 {
4177 mutex_enter(&spa_namespace_lock);
4178 spa->spa_inject_ref--;
4179 mutex_exit(&spa_namespace_lock);
4180 }
4181
4182 /*
4183 * Add spares device information to the nvlist.
4184 */
4185 static void
4186 spa_add_spares(spa_t *spa, nvlist_t *config)
4187 {
4188 nvlist_t **spares;
4189 uint_t i, nspares;
4190 nvlist_t *nvroot;
4191 uint64_t guid;
4192 vdev_stat_t *vs;
4193 uint_t vsc;
4194 uint64_t pool;
4195
4196 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4197
4198 if (spa->spa_spares.sav_count == 0)
4199 return;
4200
4201 VERIFY(nvlist_lookup_nvlist(config,
4202 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4203 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4204 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4205 if (nspares != 0) {
4206 VERIFY(nvlist_add_nvlist_array(nvroot,
4207 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4208 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4209 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4210
4211 /*
4212 * Go through and find any spares which have since been
4213 * repurposed as an active spare. If this is the case, update
4214 * their status appropriately.
4215 */
4216 for (i = 0; i < nspares; i++) {
4217 VERIFY(nvlist_lookup_uint64(spares[i],
4218 ZPOOL_CONFIG_GUID, &guid) == 0);
4219 if (spa_spare_exists(guid, &pool, NULL) &&
4220 pool != 0ULL) {
4221 VERIFY(nvlist_lookup_uint64_array(
4222 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4223 (uint64_t **)&vs, &vsc) == 0);
4224 vs->vs_state = VDEV_STATE_CANT_OPEN;
4225 vs->vs_aux = VDEV_AUX_SPARED;
4226 }
4227 }
4228 }
4229 }
4230
4231 /*
4232 * Add l2cache device information to the nvlist, including vdev stats.
4233 */
4234 static void
4235 spa_add_l2cache(spa_t *spa, nvlist_t *config)
4236 {
4237 nvlist_t **l2cache;
4238 uint_t i, j, nl2cache;
4239 nvlist_t *nvroot;
4240 uint64_t guid;
4241 vdev_t *vd;
4242 vdev_stat_t *vs;
4243 uint_t vsc;
4244
4245 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4246
4247 if (spa->spa_l2cache.sav_count == 0)
4248 return;
4249
4250 VERIFY(nvlist_lookup_nvlist(config,
4251 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4252 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4253 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4254 if (nl2cache != 0) {
4255 VERIFY(nvlist_add_nvlist_array(nvroot,
4256 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4257 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4258 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4259
4260 /*
4261 * Update level 2 cache device stats.
4262 */
4263
4264 for (i = 0; i < nl2cache; i++) {
4265 VERIFY(nvlist_lookup_uint64(l2cache[i],
4266 ZPOOL_CONFIG_GUID, &guid) == 0);
4267
4268 vd = NULL;
4269 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4270 if (guid ==
4271 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4272 vd = spa->spa_l2cache.sav_vdevs[j];
4273 break;
4274 }
4275 }
4276 ASSERT(vd != NULL);
4277
4278 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4279 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4280 == 0);
4281 vdev_get_stats(vd, vs);
4282 }
4283 }
4284 }
4285
4286 static void
4287 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4288 {
4289 nvlist_t *features;
4290 zap_cursor_t zc;
4291 zap_attribute_t za;
4292
4293 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4294 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4295
4296 if (spa->spa_feat_for_read_obj != 0) {
4297 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4298 spa->spa_feat_for_read_obj);
4299 zap_cursor_retrieve(&zc, &za) == 0;
4300 zap_cursor_advance(&zc)) {
4301 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4302 za.za_num_integers == 1);
4303 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4304 za.za_first_integer));
4305 }
4306 zap_cursor_fini(&zc);
4307 }
4308
4309 if (spa->spa_feat_for_write_obj != 0) {
4310 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4311 spa->spa_feat_for_write_obj);
4312 zap_cursor_retrieve(&zc, &za) == 0;
4313 zap_cursor_advance(&zc)) {
4314 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4315 za.za_num_integers == 1);
4316 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4317 za.za_first_integer));
4318 }
4319 zap_cursor_fini(&zc);
4320 }
4321
4322 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4323 features) == 0);
4324 nvlist_free(features);
4325 }
4326
4327 int
4328 spa_get_stats(const char *name, nvlist_t **config,
4329 char *altroot, size_t buflen)
4330 {
4331 int error;
4332 spa_t *spa;
4333
4334 *config = NULL;
4335 error = spa_open_common(name, &spa, FTAG, NULL, config);
4336
4337 if (spa != NULL) {
4338 /*
4339 * This still leaves a window of inconsistency where the spares
4340 * or l2cache devices could change and the config would be
4341 * self-inconsistent.
4342 */
4343 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4344
4345 if (*config != NULL) {
4346 uint64_t loadtimes[2];
4347
4348 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4349 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4350 VERIFY(nvlist_add_uint64_array(*config,
4351 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4352
4353 VERIFY(nvlist_add_uint64(*config,
4354 ZPOOL_CONFIG_ERRCOUNT,
4355 spa_get_errlog_size(spa)) == 0);
4356
4357 if (spa_suspended(spa))
4358 VERIFY(nvlist_add_uint64(*config,
4359 ZPOOL_CONFIG_SUSPENDED,
4360 spa->spa_failmode) == 0);
4361
4362 spa_add_spares(spa, *config);
4363 spa_add_l2cache(spa, *config);
4364 spa_add_feature_stats(spa, *config);
4365 }
4366 }
4367
4368 /*
4369 * We want to get the alternate root even for faulted pools, so we cheat
4370 * and call spa_lookup() directly.
4371 */
4372 if (altroot) {
4373 if (spa == NULL) {
4374 mutex_enter(&spa_namespace_lock);
4375 spa = spa_lookup(name);
4376 if (spa)
4377 spa_altroot(spa, altroot, buflen);
4378 else
4379 altroot[0] = '\0';
4380 spa = NULL;
4381 mutex_exit(&spa_namespace_lock);
4382 } else {
4383 spa_altroot(spa, altroot, buflen);
4384 }
4385 }
4386
4387 if (spa != NULL) {
4388 spa_config_exit(spa, SCL_CONFIG, FTAG);
4389 spa_close(spa, FTAG);
4390 }
4391
4392 return (error);
4393 }
4394
4395 /*
4396 * Validate that the auxiliary device array is well formed. We must have an
4397 * array of nvlists, each which describes a valid leaf vdev. If this is an
4398 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4399 * specified, as long as they are well-formed.
4400 */
4401 static int
4402 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4403 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4404 vdev_labeltype_t label)
4405 {
4406 nvlist_t **dev;
4407 uint_t i, ndev;
4408 vdev_t *vd;
4409 int error;
4410
4411 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4412
4413 /*
4414 * It's acceptable to have no devs specified.
4415 */
4416 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4417 return (0);
4418
4419 if (ndev == 0)
4420 return (SET_ERROR(EINVAL));
4421
4422 /*
4423 * Make sure the pool is formatted with a version that supports this
4424 * device type.
4425 */
4426 if (spa_version(spa) < version)
4427 return (SET_ERROR(ENOTSUP));
4428
4429 /*
4430 * Set the pending device list so we correctly handle device in-use
4431 * checking.
4432 */
4433 sav->sav_pending = dev;
4434 sav->sav_npending = ndev;
4435
4436 for (i = 0; i < ndev; i++) {
4437 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4438 mode)) != 0)
4439 goto out;
4440
4441 if (!vd->vdev_ops->vdev_op_leaf) {
4442 vdev_free(vd);
4443 error = SET_ERROR(EINVAL);
4444 goto out;
4445 }
4446
4447 /*
4448 * The L2ARC currently only supports disk devices in
4449 * kernel context. For user-level testing, we allow it.
4450 */
4451 #ifdef _KERNEL
4452 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4453 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4454 error = SET_ERROR(ENOTBLK);
4455 vdev_free(vd);
4456 goto out;
4457 }
4458 #endif
4459 vd->vdev_top = vd;
4460
4461 if ((error = vdev_open(vd)) == 0 &&
4462 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4463 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4464 vd->vdev_guid) == 0);
4465 }
4466
4467 vdev_free(vd);
4468
4469 if (error &&
4470 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4471 goto out;
4472 else
4473 error = 0;
4474 }
4475
4476 out:
4477 sav->sav_pending = NULL;
4478 sav->sav_npending = 0;
4479 return (error);
4480 }
4481
4482 static int
4483 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4484 {
4485 int error;
4486
4487 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4488
4489 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4490 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4491 VDEV_LABEL_SPARE)) != 0) {
4492 return (error);
4493 }
4494
4495 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4496 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4497 VDEV_LABEL_L2CACHE));
4498 }
4499
4500 static void
4501 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4502 const char *config)
4503 {
4504 int i;
4505
4506 if (sav->sav_config != NULL) {
4507 nvlist_t **olddevs;
4508 uint_t oldndevs;
4509 nvlist_t **newdevs;
4510
4511 /*
4512 * Generate new dev list by concatentating with the
4513 * current dev list.
4514 */
4515 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4516 &olddevs, &oldndevs) == 0);
4517
4518 newdevs = kmem_alloc(sizeof (void *) *
4519 (ndevs + oldndevs), KM_SLEEP);
4520 for (i = 0; i < oldndevs; i++)
4521 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4522 KM_SLEEP) == 0);
4523 for (i = 0; i < ndevs; i++)
4524 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4525 KM_SLEEP) == 0);
4526
4527 VERIFY(nvlist_remove(sav->sav_config, config,
4528 DATA_TYPE_NVLIST_ARRAY) == 0);
4529
4530 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4531 config, newdevs, ndevs + oldndevs) == 0);
4532 for (i = 0; i < oldndevs + ndevs; i++)
4533 nvlist_free(newdevs[i]);
4534 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4535 } else {
4536 /*
4537 * Generate a new dev list.
4538 */
4539 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4540 KM_SLEEP) == 0);
4541 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4542 devs, ndevs) == 0);
4543 }
4544 }
4545
4546 /*
4547 * Stop and drop level 2 ARC devices
4548 */
4549 void
4550 spa_l2cache_drop(spa_t *spa)
4551 {
4552 vdev_t *vd;
4553 int i;
4554 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4555
4556 for (i = 0; i < sav->sav_count; i++) {
4557 uint64_t pool;
4558
4559 vd = sav->sav_vdevs[i];
4560 ASSERT(vd != NULL);
4561
4562 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4563 pool != 0ULL && l2arc_vdev_present(vd))
4564 l2arc_remove_vdev(vd);
4565 }
4566 }
4567
4568 /*
4569 * Pool Creation
4570 */
4571 int
4572 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4573 nvlist_t *zplprops)
4574 {
4575 spa_t *spa;
4576 char *altroot = NULL;
4577 vdev_t *rvd;
4578 dsl_pool_t *dp;
4579 dmu_tx_t *tx;
4580 int error = 0;
4581 uint64_t txg = TXG_INITIAL;
4582 nvlist_t **spares, **l2cache;
4583 uint_t nspares, nl2cache;
4584 uint64_t version, obj;
4585 boolean_t has_features;
4586 char *poolname;
4587 nvlist_t *nvl;
4588
4589 if (nvlist_lookup_string(props,
4590 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
4591 poolname = (char *)pool;
4592
4593 /*
4594 * If this pool already exists, return failure.
4595 */
4596 mutex_enter(&spa_namespace_lock);
4597 if (spa_lookup(poolname) != NULL) {
4598 mutex_exit(&spa_namespace_lock);
4599 return (SET_ERROR(EEXIST));
4600 }
4601
4602 /*
4603 * Allocate a new spa_t structure.
4604 */
4605 nvl = fnvlist_alloc();
4606 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
4607 (void) nvlist_lookup_string(props,
4608 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4609 spa = spa_add(poolname, nvl, altroot);
4610 fnvlist_free(nvl);
4611 spa_activate(spa, spa_mode_global);
4612
4613 if (props && (error = spa_prop_validate(spa, props))) {
4614 spa_deactivate(spa);
4615 spa_remove(spa);
4616 mutex_exit(&spa_namespace_lock);
4617 return (error);
4618 }
4619
4620 /*
4621 * Temporary pool names should never be written to disk.
4622 */
4623 if (poolname != pool)
4624 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
4625
4626 has_features = B_FALSE;
4627 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4628 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4629 if (zpool_prop_feature(nvpair_name(elem)))
4630 has_features = B_TRUE;
4631 }
4632
4633 if (has_features || nvlist_lookup_uint64(props,
4634 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4635 version = SPA_VERSION;
4636 }
4637 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4638
4639 spa->spa_first_txg = txg;
4640 spa->spa_uberblock.ub_txg = txg - 1;
4641 spa->spa_uberblock.ub_version = version;
4642 spa->spa_ubsync = spa->spa_uberblock;
4643 spa->spa_load_state = SPA_LOAD_CREATE;
4644 spa->spa_removing_phys.sr_state = DSS_NONE;
4645 spa->spa_removing_phys.sr_removing_vdev = -1;
4646 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4647
4648 /*
4649 * Create "The Godfather" zio to hold all async IOs
4650 */
4651 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4652 KM_SLEEP);
4653 for (int i = 0; i < max_ncpus; i++) {
4654 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4655 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4656 ZIO_FLAG_GODFATHER);
4657 }
4658
4659 /*
4660 * Create the root vdev.
4661 */
4662 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4663
4664 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4665
4666 ASSERT(error != 0 || rvd != NULL);
4667 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4668
4669 if (error == 0 && !zfs_allocatable_devs(nvroot))
4670 error = SET_ERROR(EINVAL);
4671
4672 if (error == 0 &&
4673 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4674 (error = spa_validate_aux(spa, nvroot, txg,
4675 VDEV_ALLOC_ADD)) == 0) {
4676 for (int c = 0; c < rvd->vdev_children; c++) {
4677 vdev_metaslab_set_size(rvd->vdev_child[c]);
4678 vdev_expand(rvd->vdev_child[c], txg);
4679 }
4680 }
4681
4682 spa_config_exit(spa, SCL_ALL, FTAG);
4683
4684 if (error != 0) {
4685 spa_unload(spa);
4686 spa_deactivate(spa);
4687 spa_remove(spa);
4688 mutex_exit(&spa_namespace_lock);
4689 return (error);
4690 }
4691
4692 /*
4693 * Get the list of spares, if specified.
4694 */
4695 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4696 &spares, &nspares) == 0) {
4697 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4698 KM_SLEEP) == 0);
4699 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4700 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4701 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4702 spa_load_spares(spa);
4703 spa_config_exit(spa, SCL_ALL, FTAG);
4704 spa->spa_spares.sav_sync = B_TRUE;
4705 }
4706
4707 /*
4708 * Get the list of level 2 cache devices, if specified.
4709 */
4710 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4711 &l2cache, &nl2cache) == 0) {
4712 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4713 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4714 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4715 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4716 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4717 spa_load_l2cache(spa);
4718 spa_config_exit(spa, SCL_ALL, FTAG);
4719 spa->spa_l2cache.sav_sync = B_TRUE;
4720 }
4721
4722 spa->spa_is_initializing = B_TRUE;
4723 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4724 spa->spa_meta_objset = dp->dp_meta_objset;
4725 spa->spa_is_initializing = B_FALSE;
4726
4727 /*
4728 * Create DDTs (dedup tables).
4729 */
4730 ddt_create(spa);
4731
4732 spa_update_dspace(spa);
4733
4734 tx = dmu_tx_create_assigned(dp, txg);
4735
4736 /*
4737 * Create the pool config object.
4738 */
4739 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4740 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4741 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4742
4743 if (zap_add(spa->spa_meta_objset,
4744 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4745 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4746 cmn_err(CE_PANIC, "failed to add pool config");
4747 }
4748
4749 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4750 spa_feature_create_zap_objects(spa, tx);
4751
4752 if (zap_add(spa->spa_meta_objset,
4753 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4754 sizeof (uint64_t), 1, &version, tx) != 0) {
4755 cmn_err(CE_PANIC, "failed to add pool version");
4756 }
4757
4758 /* Newly created pools with the right version are always deflated. */
4759 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4760 spa->spa_deflate = TRUE;
4761 if (zap_add(spa->spa_meta_objset,
4762 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4763 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4764 cmn_err(CE_PANIC, "failed to add deflate");
4765 }
4766 }
4767
4768 /*
4769 * Create the deferred-free bpobj. Turn off compression
4770 * because sync-to-convergence takes longer if the blocksize
4771 * keeps changing.
4772 */
4773 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4774 dmu_object_set_compress(spa->spa_meta_objset, obj,
4775 ZIO_COMPRESS_OFF, tx);
4776 if (zap_add(spa->spa_meta_objset,
4777 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4778 sizeof (uint64_t), 1, &obj, tx) != 0) {
4779 cmn_err(CE_PANIC, "failed to add bpobj");
4780 }
4781 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4782 spa->spa_meta_objset, obj));
4783
4784 /*
4785 * Create the pool's history object.
4786 */
4787 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4788 spa_history_create_obj(spa, tx);
4789
4790 /*
4791 * Generate some random noise for salted checksums to operate on.
4792 */
4793 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4794 sizeof (spa->spa_cksum_salt.zcs_bytes));
4795
4796 /*
4797 * Set pool properties.
4798 */
4799 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4800 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4801 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4802 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4803
4804 if (props != NULL) {
4805 spa_configfile_set(spa, props, B_FALSE);
4806 spa_sync_props(props, tx);
4807 }
4808
4809 dmu_tx_commit(tx);
4810
4811 spa->spa_sync_on = B_TRUE;
4812 txg_sync_start(spa->spa_dsl_pool);
4813
4814 /*
4815 * We explicitly wait for the first transaction to complete so that our
4816 * bean counters are appropriately updated.
4817 */
4818 txg_wait_synced(spa->spa_dsl_pool, txg);
4819
4820 spa_spawn_aux_threads(spa);
4821
4822 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4823 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4824
4825 spa_history_log_version(spa, "create");
4826
4827 /*
4828 * Don't count references from objsets that are already closed
4829 * and are making their way through the eviction process.
4830 */
4831 spa_evicting_os_wait(spa);
4832 spa->spa_minref = refcount_count(&spa->spa_refcount);
4833 spa->spa_load_state = SPA_LOAD_NONE;
4834
4835 mutex_exit(&spa_namespace_lock);
4836
4837 return (0);
4838 }
4839
4840 #ifdef _KERNEL
4841 /*
4842 * Get the root pool information from the root disk, then import the root pool
4843 * during the system boot up time.
4844 */
4845 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4846
4847 static nvlist_t *
4848 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4849 {
4850 nvlist_t *config;
4851 nvlist_t *nvtop, *nvroot;
4852 uint64_t pgid;
4853
4854 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
4855 return (NULL);
4856
4857 /*
4858 * Add this top-level vdev to the child array.
4859 */
4860 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4861 &nvtop) == 0);
4862 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4863 &pgid) == 0);
4864 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
4865
4866 /*
4867 * Put this pool's top-level vdevs into a root vdev.
4868 */
4869 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4870 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4871 VDEV_TYPE_ROOT) == 0);
4872 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4873 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4874 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4875 &nvtop, 1) == 0);
4876
4877 /*
4878 * Replace the existing vdev_tree with the new root vdev in
4879 * this pool's configuration (remove the old, add the new).
4880 */
4881 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4882 nvlist_free(nvroot);
4883 return (config);
4884 }
4885
4886 /*
4887 * Walk the vdev tree and see if we can find a device with "better"
4888 * configuration. A configuration is "better" if the label on that
4889 * device has a more recent txg.
4890 */
4891 static void
4892 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
4893 {
4894 for (int c = 0; c < vd->vdev_children; c++)
4895 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
4896
4897 if (vd->vdev_ops->vdev_op_leaf) {
4898 nvlist_t *label;
4899 uint64_t label_txg;
4900
4901 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4902 &label) != 0)
4903 return;
4904
4905 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4906 &label_txg) == 0);
4907
4908 /*
4909 * Do we have a better boot device?
4910 */
4911 if (label_txg > *txg) {
4912 *txg = label_txg;
4913 *avd = vd;
4914 }
4915 nvlist_free(label);
4916 }
4917 }
4918
4919 /*
4920 * Import a root pool.
4921 *
4922 * For x86. devpath_list will consist of devid and/or physpath name of
4923 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4924 * The GRUB "findroot" command will return the vdev we should boot.
4925 *
4926 * For Sparc, devpath_list consists the physpath name of the booting device
4927 * no matter the rootpool is a single device pool or a mirrored pool.
4928 * e.g.
4929 * "/pci@1f,0/ide@d/disk@0,0:a"
4930 */
4931 int
4932 spa_import_rootpool(char *devpath, char *devid)
4933 {
4934 spa_t *spa;
4935 vdev_t *rvd, *bvd, *avd = NULL;
4936 nvlist_t *config, *nvtop;
4937 uint64_t guid, txg;
4938 char *pname;
4939 int error;
4940
4941 /*
4942 * Read the label from the boot device and generate a configuration.
4943 */
4944 config = spa_generate_rootconf(devpath, devid, &guid);
4945 #if defined(_OBP) && defined(_KERNEL)
4946 if (config == NULL) {
4947 if (strstr(devpath, "/iscsi/ssd") != NULL) {
4948 /* iscsi boot */
4949 get_iscsi_bootpath_phy(devpath);
4950 config = spa_generate_rootconf(devpath, devid, &guid);
4951 }
4952 }
4953 #endif
4954 if (config == NULL) {
4955 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
4956 devpath);
4957 return (SET_ERROR(EIO));
4958 }
4959
4960 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4961 &pname) == 0);
4962 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
4963
4964 mutex_enter(&spa_namespace_lock);
4965 if ((spa = spa_lookup(pname)) != NULL) {
4966 /*
4967 * Remove the existing root pool from the namespace so that we
4968 * can replace it with the correct config we just read in.
4969 */
4970 spa_remove(spa);
4971 }
4972
4973 spa = spa_add(pname, config, NULL);
4974 spa->spa_is_root = B_TRUE;
4975 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4976 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4977 &spa->spa_ubsync.ub_version) != 0)
4978 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4979
4980 /*
4981 * Build up a vdev tree based on the boot device's label config.
4982 */
4983 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4984 &nvtop) == 0);
4985 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4986 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4987 VDEV_ALLOC_ROOTPOOL);
4988 spa_config_exit(spa, SCL_ALL, FTAG);
4989 if (error) {
4990 mutex_exit(&spa_namespace_lock);
4991 nvlist_free(config);
4992 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4993 pname);
4994 return (error);
4995 }
4996
4997 /*
4998 * Get the boot vdev.
4999 */
5000 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5001 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5002 (u_longlong_t)guid);
5003 error = SET_ERROR(ENOENT);
5004 goto out;
5005 }
5006
5007 /*
5008 * Determine if there is a better boot device.
5009 */
5010 avd = bvd;
5011 spa_alt_rootvdev(rvd, &avd, &txg);
5012 if (avd != bvd) {
5013 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5014 "try booting from '%s'", avd->vdev_path);
5015 error = SET_ERROR(EINVAL);
5016 goto out;
5017 }
5018
5019 /*
5020 * If the boot device is part of a spare vdev then ensure that
5021 * we're booting off the active spare.
5022 */
5023 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5024 !bvd->vdev_isspare) {
5025 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5026 "try booting from '%s'",
5027 bvd->vdev_parent->
5028 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5029 error = SET_ERROR(EINVAL);
5030 goto out;
5031 }
5032
5033 error = 0;
5034 out:
5035 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5036 vdev_free(rvd);
5037 spa_config_exit(spa, SCL_ALL, FTAG);
5038 mutex_exit(&spa_namespace_lock);
5039
5040 nvlist_free(config);
5041 return (error);
5042 }
5043
5044 #endif
5045
5046 /*
5047 * Import a non-root pool into the system.
5048 */
5049 int
5050 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5051 {
5052 spa_t *spa;
5053 char *altroot = NULL;
5054 spa_load_state_t state = SPA_LOAD_IMPORT;
5055 zpool_load_policy_t policy;
5056 uint64_t mode = spa_mode_global;
5057 uint64_t readonly = B_FALSE;
5058 int error;
5059 nvlist_t *nvroot;
5060 nvlist_t **spares, **l2cache;
5061 uint_t nspares, nl2cache;
5062
5063 /*
5064 * If a pool with this name exists, return failure.
5065 */
5066 mutex_enter(&spa_namespace_lock);
5067 if (spa_lookup(pool) != NULL) {
5068 mutex_exit(&spa_namespace_lock);
5069 return (SET_ERROR(EEXIST));
5070 }
5071
5072 /*
5073 * Create and initialize the spa structure.
5074 */
5075 (void) nvlist_lookup_string(props,
5076 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5077 (void) nvlist_lookup_uint64(props,
5078 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5079 if (readonly)
5080 mode = FREAD;
5081 spa = spa_add(pool, config, altroot);
5082 spa->spa_import_flags = flags;
5083
5084 /*
5085 * Verbatim import - Take a pool and insert it into the namespace
5086 * as if it had been loaded at boot.
5087 */
5088 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5089 if (props != NULL)
5090 spa_configfile_set(spa, props, B_FALSE);
5091
5092 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5093 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5094 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5095 mutex_exit(&spa_namespace_lock);
5096 return (0);
5097 }
5098
5099 spa_activate(spa, mode);
5100
5101 /*
5102 * Don't start async tasks until we know everything is healthy.
5103 */
5104 spa_async_suspend(spa);
5105
5106 zpool_get_load_policy(config, &policy);
5107 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5108 state = SPA_LOAD_RECOVER;
5109
5110 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5111
5112 if (state != SPA_LOAD_RECOVER) {
5113 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5114 zfs_dbgmsg("spa_import: importing %s", pool);
5115 } else {
5116 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5117 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5118 }
5119 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5120
5121 /*
5122 * Propagate anything learned while loading the pool and pass it
5123 * back to caller (i.e. rewind info, missing devices, etc).
5124 */
5125 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5126 spa->spa_load_info) == 0);
5127
5128 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5129 /*
5130 * Toss any existing sparelist, as it doesn't have any validity
5131 * anymore, and conflicts with spa_has_spare().
5132 */
5133 if (spa->spa_spares.sav_config) {
5134 nvlist_free(spa->spa_spares.sav_config);
5135 spa->spa_spares.sav_config = NULL;
5136 spa_load_spares(spa);
5137 }
5138 if (spa->spa_l2cache.sav_config) {
5139 nvlist_free(spa->spa_l2cache.sav_config);
5140 spa->spa_l2cache.sav_config = NULL;
5141 spa_load_l2cache(spa);
5142 }
5143
5144 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5145 &nvroot) == 0);
5146 if (error == 0)
5147 error = spa_validate_aux(spa, nvroot, -1ULL,
5148 VDEV_ALLOC_SPARE);
5149 if (error == 0)
5150 error = spa_validate_aux(spa, nvroot, -1ULL,
5151 VDEV_ALLOC_L2CACHE);
5152 spa_config_exit(spa, SCL_ALL, FTAG);
5153
5154 if (props != NULL)
5155 spa_configfile_set(spa, props, B_FALSE);
5156
5157 if (error != 0 || (props && spa_writeable(spa) &&
5158 (error = spa_prop_set(spa, props)))) {
5159 spa_unload(spa);
5160 spa_deactivate(spa);
5161 spa_remove(spa);
5162 mutex_exit(&spa_namespace_lock);
5163 return (error);
5164 }
5165
5166 spa_async_resume(spa);
5167
5168 /*
5169 * Override any spares and level 2 cache devices as specified by
5170 * the user, as these may have correct device names/devids, etc.
5171 */
5172 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5173 &spares, &nspares) == 0) {
5174 if (spa->spa_spares.sav_config)
5175 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5176 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5177 else
5178 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5179 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5180 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5181 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5182 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5183 spa_load_spares(spa);
5184 spa_config_exit(spa, SCL_ALL, FTAG);
5185 spa->spa_spares.sav_sync = B_TRUE;
5186 }
5187 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5188 &l2cache, &nl2cache) == 0) {
5189 if (spa->spa_l2cache.sav_config)
5190 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5191 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5192 else
5193 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5194 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5195 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5196 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5197 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5198 spa_load_l2cache(spa);
5199 spa_config_exit(spa, SCL_ALL, FTAG);
5200 spa->spa_l2cache.sav_sync = B_TRUE;
5201 }
5202
5203 /*
5204 * Check for any removed devices.
5205 */
5206 if (spa->spa_autoreplace) {
5207 spa_aux_check_removed(&spa->spa_spares);
5208 spa_aux_check_removed(&spa->spa_l2cache);
5209 }
5210
5211 if (spa_writeable(spa)) {
5212 /*
5213 * Update the config cache to include the newly-imported pool.
5214 */
5215 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5216 }
5217
5218 /*
5219 * It's possible that the pool was expanded while it was exported.
5220 * We kick off an async task to handle this for us.
5221 */
5222 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5223
5224 spa_history_log_version(spa, "import");
5225
5226 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5227
5228 mutex_exit(&spa_namespace_lock);
5229
5230 return (0);
5231 }
5232
5233 nvlist_t *
5234 spa_tryimport(nvlist_t *tryconfig)
5235 {
5236 nvlist_t *config = NULL;
5237 char *poolname, *cachefile;
5238 spa_t *spa;
5239 uint64_t state;
5240 int error;
5241 zpool_load_policy_t policy;
5242
5243 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5244 return (NULL);
5245
5246 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5247 return (NULL);
5248
5249 /*
5250 * Create and initialize the spa structure.
5251 */
5252 mutex_enter(&spa_namespace_lock);
5253 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5254 spa_activate(spa, FREAD);
5255
5256 /*
5257 * Rewind pool if a max txg was provided.
5258 */
5259 zpool_get_load_policy(spa->spa_config, &policy);
5260 if (policy.zlp_txg != UINT64_MAX) {
5261 spa->spa_load_max_txg = policy.zlp_txg;
5262 spa->spa_extreme_rewind = B_TRUE;
5263 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5264 poolname, (longlong_t)policy.zlp_txg);
5265 } else {
5266 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5267 }
5268
5269 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5270 == 0) {
5271 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5272 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5273 } else {
5274 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5275 }
5276
5277 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5278
5279 /*
5280 * If 'tryconfig' was at least parsable, return the current config.
5281 */
5282 if (spa->spa_root_vdev != NULL) {
5283 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5284 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5285 poolname) == 0);
5286 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5287 state) == 0);
5288 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5289 spa->spa_uberblock.ub_timestamp) == 0);
5290 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5291 spa->spa_load_info) == 0);
5292
5293 /*
5294 * If the bootfs property exists on this pool then we
5295 * copy it out so that external consumers can tell which
5296 * pools are bootable.
5297 */
5298 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5299 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5300
5301 /*
5302 * We have to play games with the name since the
5303 * pool was opened as TRYIMPORT_NAME.
5304 */
5305 if (dsl_dsobj_to_dsname(spa_name(spa),
5306 spa->spa_bootfs, tmpname) == 0) {
5307 char *cp;
5308 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5309
5310 cp = strchr(tmpname, '/');
5311 if (cp == NULL) {
5312 (void) strlcpy(dsname, tmpname,
5313 MAXPATHLEN);
5314 } else {
5315 (void) snprintf(dsname, MAXPATHLEN,
5316 "%s/%s", poolname, ++cp);
5317 }
5318 VERIFY(nvlist_add_string(config,
5319 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5320 kmem_free(dsname, MAXPATHLEN);
5321 }
5322 kmem_free(tmpname, MAXPATHLEN);
5323 }
5324
5325 /*
5326 * Add the list of hot spares and level 2 cache devices.
5327 */
5328 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5329 spa_add_spares(spa, config);
5330 spa_add_l2cache(spa, config);
5331 spa_config_exit(spa, SCL_CONFIG, FTAG);
5332 }
5333
5334 spa_unload(spa);
5335 spa_deactivate(spa);
5336 spa_remove(spa);
5337 mutex_exit(&spa_namespace_lock);
5338
5339 return (config);
5340 }
5341
5342 /*
5343 * Pool export/destroy
5344 *
5345 * The act of destroying or exporting a pool is very simple. We make sure there
5346 * is no more pending I/O and any references to the pool are gone. Then, we
5347 * update the pool state and sync all the labels to disk, removing the
5348 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5349 * we don't sync the labels or remove the configuration cache.
5350 */
5351 static int
5352 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5353 boolean_t force, boolean_t hardforce)
5354 {
5355 spa_t *spa;
5356
5357 if (oldconfig)
5358 *oldconfig = NULL;
5359
5360 if (!(spa_mode_global & FWRITE))
5361 return (SET_ERROR(EROFS));
5362
5363 mutex_enter(&spa_namespace_lock);
5364 if ((spa = spa_lookup(pool)) == NULL) {
5365 mutex_exit(&spa_namespace_lock);
5366 return (SET_ERROR(ENOENT));
5367 }
5368
5369 /*
5370 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5371 * reacquire the namespace lock, and see if we can export.
5372 */
5373 spa_open_ref(spa, FTAG);
5374 mutex_exit(&spa_namespace_lock);
5375 spa_async_suspend(spa);
5376 mutex_enter(&spa_namespace_lock);
5377 spa_close(spa, FTAG);
5378
5379 /*
5380 * The pool will be in core if it's openable,
5381 * in which case we can modify its state.
5382 */
5383 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5384
5385 /*
5386 * Objsets may be open only because they're dirty, so we
5387 * have to force it to sync before checking spa_refcnt.
5388 */
5389 txg_wait_synced(spa->spa_dsl_pool, 0);
5390 spa_evicting_os_wait(spa);
5391
5392 /*
5393 * A pool cannot be exported or destroyed if there are active
5394 * references. If we are resetting a pool, allow references by
5395 * fault injection handlers.
5396 */
5397 if (!spa_refcount_zero(spa) ||
5398 (spa->spa_inject_ref != 0 &&
5399 new_state != POOL_STATE_UNINITIALIZED)) {
5400 spa_async_resume(spa);
5401 mutex_exit(&spa_namespace_lock);
5402 return (SET_ERROR(EBUSY));
5403 }
5404
5405 /*
5406 * A pool cannot be exported if it has an active shared spare.
5407 * This is to prevent other pools stealing the active spare
5408 * from an exported pool. At user's own will, such pool can
5409 * be forcedly exported.
5410 */
5411 if (!force && new_state == POOL_STATE_EXPORTED &&
5412 spa_has_active_shared_spare(spa)) {
5413 spa_async_resume(spa);
5414 mutex_exit(&spa_namespace_lock);
5415 return (SET_ERROR(EXDEV));
5416 }
5417
5418 /*
5419 * We're about to export or destroy this pool. Make sure
5420 * we stop all initializtion activity here before we
5421 * set the spa_final_txg. This will ensure that all
5422 * dirty data resulting from the initialization is
5423 * committed to disk before we unload the pool.
5424 */
5425 if (spa->spa_root_vdev != NULL) {
5426 vdev_initialize_stop_all(spa->spa_root_vdev,
5427 VDEV_INITIALIZE_ACTIVE);
5428 }
5429
5430 /*
5431 * We want this to be reflected on every label,
5432 * so mark them all dirty. spa_unload() will do the
5433 * final sync that pushes these changes out.
5434 */
5435 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5437 spa->spa_state = new_state;
5438 spa->spa_final_txg = spa_last_synced_txg(spa) +
5439 TXG_DEFER_SIZE + 1;
5440 vdev_config_dirty(spa->spa_root_vdev);
5441 spa_config_exit(spa, SCL_ALL, FTAG);
5442 }
5443 }
5444
5445 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5446
5447 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5448 spa_unload(spa);
5449 spa_deactivate(spa);
5450 }
5451
5452 if (oldconfig && spa->spa_config)
5453 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5454
5455 if (new_state != POOL_STATE_UNINITIALIZED) {
5456 if (!hardforce)
5457 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5458 spa_remove(spa);
5459 }
5460 mutex_exit(&spa_namespace_lock);
5461
5462 return (0);
5463 }
5464
5465 /*
5466 * Destroy a storage pool.
5467 */
5468 int
5469 spa_destroy(char *pool)
5470 {
5471 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5472 B_FALSE, B_FALSE));
5473 }
5474
5475 /*
5476 * Export a storage pool.
5477 */
5478 int
5479 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5480 boolean_t hardforce)
5481 {
5482 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5483 force, hardforce));
5484 }
5485
5486 /*
5487 * Similar to spa_export(), this unloads the spa_t without actually removing it
5488 * from the namespace in any way.
5489 */
5490 int
5491 spa_reset(char *pool)
5492 {
5493 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5494 B_FALSE, B_FALSE));
5495 }
5496
5497 /*
5498 * ==========================================================================
5499 * Device manipulation
5500 * ==========================================================================
5501 */
5502
5503 /*
5504 * Add a device to a storage pool.
5505 */
5506 int
5507 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5508 {
5509 uint64_t txg, id;
5510 int error;
5511 vdev_t *rvd = spa->spa_root_vdev;
5512 vdev_t *vd, *tvd;
5513 nvlist_t **spares, **l2cache;
5514 uint_t nspares, nl2cache;
5515
5516 ASSERT(spa_writeable(spa));
5517
5518 txg = spa_vdev_enter(spa);
5519
5520 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5521 VDEV_ALLOC_ADD)) != 0)
5522 return (spa_vdev_exit(spa, NULL, txg, error));
5523
5524 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5525
5526 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5527 &nspares) != 0)
5528 nspares = 0;
5529
5530 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5531 &nl2cache) != 0)
5532 nl2cache = 0;
5533
5534 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5535 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5536
5537 if (vd->vdev_children != 0 &&
5538 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5539 return (spa_vdev_exit(spa, vd, txg, error));
5540
5541 /*
5542 * We must validate the spares and l2cache devices after checking the
5543 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5544 */
5545 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5546 return (spa_vdev_exit(spa, vd, txg, error));
5547
5548 /*
5549 * If we are in the middle of a device removal, we can only add
5550 * devices which match the existing devices in the pool.
5551 * If we are in the middle of a removal, or have some indirect
5552 * vdevs, we can not add raidz toplevels.
5553 */
5554 if (spa->spa_vdev_removal != NULL ||
5555 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5556 for (int c = 0; c < vd->vdev_children; c++) {
5557 tvd = vd->vdev_child[c];
5558 if (spa->spa_vdev_removal != NULL &&
5559 tvd->vdev_ashift != spa->spa_max_ashift) {
5560 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5561 }
5562 /* Fail if top level vdev is raidz */
5563 if (tvd->vdev_ops == &vdev_raidz_ops) {
5564 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5565 }
5566 /*
5567 * Need the top level mirror to be
5568 * a mirror of leaf vdevs only
5569 */
5570 if (tvd->vdev_ops == &vdev_mirror_ops) {
5571 for (uint64_t cid = 0;
5572 cid < tvd->vdev_children; cid++) {
5573 vdev_t *cvd = tvd->vdev_child[cid];
5574 if (!cvd->vdev_ops->vdev_op_leaf) {
5575 return (spa_vdev_exit(spa, vd,
5576 txg, EINVAL));
5577 }
5578 }
5579 }
5580 }
5581 }
5582
5583 for (int c = 0; c < vd->vdev_children; c++) {
5584
5585 /*
5586 * Set the vdev id to the first hole, if one exists.
5587 */
5588 for (id = 0; id < rvd->vdev_children; id++) {
5589 if (rvd->vdev_child[id]->vdev_ishole) {
5590 vdev_free(rvd->vdev_child[id]);
5591 break;
5592 }
5593 }
5594 tvd = vd->vdev_child[c];
5595 vdev_remove_child(vd, tvd);
5596 tvd->vdev_id = id;
5597 vdev_add_child(rvd, tvd);
5598 vdev_config_dirty(tvd);
5599 }
5600
5601 if (nspares != 0) {
5602 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5603 ZPOOL_CONFIG_SPARES);
5604 spa_load_spares(spa);
5605 spa->spa_spares.sav_sync = B_TRUE;
5606 }
5607
5608 if (nl2cache != 0) {
5609 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5610 ZPOOL_CONFIG_L2CACHE);
5611 spa_load_l2cache(spa);
5612 spa->spa_l2cache.sav_sync = B_TRUE;
5613 }
5614
5615 /*
5616 * We have to be careful when adding new vdevs to an existing pool.
5617 * If other threads start allocating from these vdevs before we
5618 * sync the config cache, and we lose power, then upon reboot we may
5619 * fail to open the pool because there are DVAs that the config cache
5620 * can't translate. Therefore, we first add the vdevs without
5621 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5622 * and then let spa_config_update() initialize the new metaslabs.
5623 *
5624 * spa_load() checks for added-but-not-initialized vdevs, so that
5625 * if we lose power at any point in this sequence, the remaining
5626 * steps will be completed the next time we load the pool.
5627 */
5628 (void) spa_vdev_exit(spa, vd, txg, 0);
5629
5630 mutex_enter(&spa_namespace_lock);
5631 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5632 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5633 mutex_exit(&spa_namespace_lock);
5634
5635 return (0);
5636 }
5637
5638 /*
5639 * Attach a device to a mirror. The arguments are the path to any device
5640 * in the mirror, and the nvroot for the new device. If the path specifies
5641 * a device that is not mirrored, we automatically insert the mirror vdev.
5642 *
5643 * If 'replacing' is specified, the new device is intended to replace the
5644 * existing device; in this case the two devices are made into their own
5645 * mirror using the 'replacing' vdev, which is functionally identical to
5646 * the mirror vdev (it actually reuses all the same ops) but has a few
5647 * extra rules: you can't attach to it after it's been created, and upon
5648 * completion of resilvering, the first disk (the one being replaced)
5649 * is automatically detached.
5650 */
5651 int
5652 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5653 {
5654 uint64_t txg, dtl_max_txg;
5655 vdev_t *rvd = spa->spa_root_vdev;
5656 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5657 vdev_ops_t *pvops;
5658 char *oldvdpath, *newvdpath;
5659 int newvd_isspare;
5660 int error;
5661
5662 ASSERT(spa_writeable(spa));
5663
5664 txg = spa_vdev_enter(spa);
5665
5666 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5667
5668 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5669 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5670 error = (spa_has_checkpoint(spa)) ?
5671 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5672 return (spa_vdev_exit(spa, NULL, txg, error));
5673 }
5674
5675 if (spa->spa_vdev_removal != NULL)
5676 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5677
5678 if (oldvd == NULL)
5679 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5680
5681 if (!oldvd->vdev_ops->vdev_op_leaf)
5682 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5683
5684 pvd = oldvd->vdev_parent;
5685
5686 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
5687 VDEV_ALLOC_ATTACH)) != 0)
5688 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5689
5690 if (newrootvd->vdev_children != 1)
5691 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5692
5693 newvd = newrootvd->vdev_child[0];
5694
5695 if (!newvd->vdev_ops->vdev_op_leaf)
5696 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5697
5698 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
5699 return (spa_vdev_exit(spa, newrootvd, txg, error));
5700
5701 /*
5702 * Spares can't replace logs
5703 */
5704 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
5705 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5706
5707 if (!replacing) {
5708 /*
5709 * For attach, the only allowable parent is a mirror or the root
5710 * vdev.
5711 */
5712 if (pvd->vdev_ops != &vdev_mirror_ops &&
5713 pvd->vdev_ops != &vdev_root_ops)
5714 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5715
5716 pvops = &vdev_mirror_ops;
5717 } else {
5718 /*
5719 * Active hot spares can only be replaced by inactive hot
5720 * spares.
5721 */
5722 if (pvd->vdev_ops == &vdev_spare_ops &&
5723 oldvd->vdev_isspare &&
5724 !spa_has_spare(spa, newvd->vdev_guid))
5725 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5726
5727 /*
5728 * If the source is a hot spare, and the parent isn't already a
5729 * spare, then we want to create a new hot spare. Otherwise, we
5730 * want to create a replacing vdev. The user is not allowed to
5731 * attach to a spared vdev child unless the 'isspare' state is
5732 * the same (spare replaces spare, non-spare replaces
5733 * non-spare).
5734 */
5735 if (pvd->vdev_ops == &vdev_replacing_ops &&
5736 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
5737 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5738 } else if (pvd->vdev_ops == &vdev_spare_ops &&
5739 newvd->vdev_isspare != oldvd->vdev_isspare) {
5740 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5741 }
5742
5743 if (newvd->vdev_isspare)
5744 pvops = &vdev_spare_ops;
5745 else
5746 pvops = &vdev_replacing_ops;
5747 }
5748
5749 /*
5750 * Make sure the new device is big enough.
5751 */
5752 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
5753 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
5754
5755 /*
5756 * The new device cannot have a higher alignment requirement
5757 * than the top-level vdev.
5758 */
5759 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
5760 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
5761
5762 /*
5763 * If this is an in-place replacement, update oldvd's path and devid
5764 * to make it distinguishable from newvd, and unopenable from now on.
5765 */
5766 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
5767 spa_strfree(oldvd->vdev_path);
5768 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
5769 KM_SLEEP);
5770 (void) sprintf(oldvd->vdev_path, "%s/%s",
5771 newvd->vdev_path, "old");
5772 if (oldvd->vdev_devid != NULL) {
5773 spa_strfree(oldvd->vdev_devid);
5774 oldvd->vdev_devid = NULL;
5775 }
5776 }
5777
5778 /* mark the device being resilvered */
5779 newvd->vdev_resilver_txg = txg;
5780
5781 /*
5782 * If the parent is not a mirror, or if we're replacing, insert the new
5783 * mirror/replacing/spare vdev above oldvd.
5784 */
5785 if (pvd->vdev_ops != pvops)
5786 pvd = vdev_add_parent(oldvd, pvops);
5787
5788 ASSERT(pvd->vdev_top->vdev_parent == rvd);
5789 ASSERT(pvd->vdev_ops == pvops);
5790 ASSERT(oldvd->vdev_parent == pvd);
5791
5792 /*
5793 * Extract the new device from its root and add it to pvd.
5794 */
5795 vdev_remove_child(newrootvd, newvd);
5796 newvd->vdev_id = pvd->vdev_children;
5797 newvd->vdev_crtxg = oldvd->vdev_crtxg;
5798 vdev_add_child(pvd, newvd);
5799
5800 tvd = newvd->vdev_top;
5801 ASSERT(pvd->vdev_top == tvd);
5802 ASSERT(tvd->vdev_parent == rvd);
5803
5804 vdev_config_dirty(tvd);
5805
5806 /*
5807 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5808 * for any dmu_sync-ed blocks. It will propagate upward when
5809 * spa_vdev_exit() calls vdev_dtl_reassess().
5810 */
5811 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
5812
5813 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
5814 dtl_max_txg - TXG_INITIAL);
5815
5816 if (newvd->vdev_isspare) {
5817 spa_spare_activate(newvd);
5818 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
5819 }
5820
5821 oldvdpath = spa_strdup(oldvd->vdev_path);
5822 newvdpath = spa_strdup(newvd->vdev_path);
5823 newvd_isspare = newvd->vdev_isspare;
5824
5825 /*
5826 * Mark newvd's DTL dirty in this txg.
5827 */
5828 vdev_dirty(tvd, VDD_DTL, newvd, txg);
5829
5830 /*
5831 * Schedule the resilver to restart in the future. We do this to
5832 * ensure that dmu_sync-ed blocks have been stitched into the
5833 * respective datasets.
5834 */
5835 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
5836
5837 if (spa->spa_bootfs)
5838 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
5839
5840 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
5841
5842 /*
5843 * Commit the config
5844 */
5845 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
5846
5847 spa_history_log_internal(spa, "vdev attach", NULL,
5848 "%s vdev=%s %s vdev=%s",
5849 replacing && newvd_isspare ? "spare in" :
5850 replacing ? "replace" : "attach", newvdpath,
5851 replacing ? "for" : "to", oldvdpath);
5852
5853 spa_strfree(oldvdpath);
5854 spa_strfree(newvdpath);
5855
5856 return (0);
5857 }
5858
5859 /*
5860 * Detach a device from a mirror or replacing vdev.
5861 *
5862 * If 'replace_done' is specified, only detach if the parent
5863 * is a replacing vdev.
5864 */
5865 int
5866 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5867 {
5868 uint64_t txg;
5869 int error;
5870 vdev_t *rvd = spa->spa_root_vdev;
5871 vdev_t *vd, *pvd, *cvd, *tvd;
5872 boolean_t unspare = B_FALSE;
5873 uint64_t unspare_guid = 0;
5874 char *vdpath;
5875
5876 ASSERT(spa_writeable(spa));
5877
5878 txg = spa_vdev_enter(spa);
5879
5880 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5881
5882 /*
5883 * Besides being called directly from the userland through the
5884 * ioctl interface, spa_vdev_detach() can be potentially called
5885 * at the end of spa_vdev_resilver_done().
5886 *
5887 * In the regular case, when we have a checkpoint this shouldn't
5888 * happen as we never empty the DTLs of a vdev during the scrub
5889 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
5890 * should never get here when we have a checkpoint.
5891 *
5892 * That said, even in a case when we checkpoint the pool exactly
5893 * as spa_vdev_resilver_done() calls this function everything
5894 * should be fine as the resilver will return right away.
5895 */
5896 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5897 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5898 error = (spa_has_checkpoint(spa)) ?
5899 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5900 return (spa_vdev_exit(spa, NULL, txg, error));
5901 }
5902
5903 if (vd == NULL)
5904 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5905
5906 if (!vd->vdev_ops->vdev_op_leaf)
5907 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5908
5909 pvd = vd->vdev_parent;
5910
5911 /*
5912 * If the parent/child relationship is not as expected, don't do it.
5913 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5914 * vdev that's replacing B with C. The user's intent in replacing
5915 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5916 * the replace by detaching C, the expected behavior is to end up
5917 * M(A,B). But suppose that right after deciding to detach C,
5918 * the replacement of B completes. We would have M(A,C), and then
5919 * ask to detach C, which would leave us with just A -- not what
5920 * the user wanted. To prevent this, we make sure that the
5921 * parent/child relationship hasn't changed -- in this example,
5922 * that C's parent is still the replacing vdev R.
5923 */
5924 if (pvd->vdev_guid != pguid && pguid != 0)
5925 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5926
5927 /*
5928 * Only 'replacing' or 'spare' vdevs can be replaced.
5929 */
5930 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5931 pvd->vdev_ops != &vdev_spare_ops)
5932 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5933
5934 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5935 spa_version(spa) >= SPA_VERSION_SPARES);
5936
5937 /*
5938 * Only mirror, replacing, and spare vdevs support detach.
5939 */
5940 if (pvd->vdev_ops != &vdev_replacing_ops &&
5941 pvd->vdev_ops != &vdev_mirror_ops &&
5942 pvd->vdev_ops != &vdev_spare_ops)
5943 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5944
5945 /*
5946 * If this device has the only valid copy of some data,
5947 * we cannot safely detach it.
5948 */
5949 if (vdev_dtl_required(vd))
5950 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5951
5952 ASSERT(pvd->vdev_children >= 2);
5953
5954 /*
5955 * If we are detaching the second disk from a replacing vdev, then
5956 * check to see if we changed the original vdev's path to have "/old"
5957 * at the end in spa_vdev_attach(). If so, undo that change now.
5958 */
5959 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5960 vd->vdev_path != NULL) {
5961 size_t len = strlen(vd->vdev_path);
5962
5963 for (int c = 0; c < pvd->vdev_children; c++) {
5964 cvd = pvd->vdev_child[c];
5965
5966 if (cvd == vd || cvd->vdev_path == NULL)
5967 continue;
5968
5969 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5970 strcmp(cvd->vdev_path + len, "/old") == 0) {
5971 spa_strfree(cvd->vdev_path);
5972 cvd->vdev_path = spa_strdup(vd->vdev_path);
5973 break;
5974 }
5975 }
5976 }
5977
5978 /*
5979 * If we are detaching the original disk from a spare, then it implies
5980 * that the spare should become a real disk, and be removed from the
5981 * active spare list for the pool.
5982 */
5983 if (pvd->vdev_ops == &vdev_spare_ops &&
5984 vd->vdev_id == 0 &&
5985 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5986 unspare = B_TRUE;
5987
5988 /*
5989 * Erase the disk labels so the disk can be used for other things.
5990 * This must be done after all other error cases are handled,
5991 * but before we disembowel vd (so we can still do I/O to it).
5992 * But if we can't do it, don't treat the error as fatal --
5993 * it may be that the unwritability of the disk is the reason
5994 * it's being detached!
5995 */
5996 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5997
5998 /*
5999 * Remove vd from its parent and compact the parent's children.
6000 */
6001 vdev_remove_child(pvd, vd);
6002 vdev_compact_children(pvd);
6003
6004 /*
6005 * Remember one of the remaining children so we can get tvd below.
6006 */
6007 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6008
6009 /*
6010 * If we need to remove the remaining child from the list of hot spares,
6011 * do it now, marking the vdev as no longer a spare in the process.
6012 * We must do this before vdev_remove_parent(), because that can
6013 * change the GUID if it creates a new toplevel GUID. For a similar
6014 * reason, we must remove the spare now, in the same txg as the detach;
6015 * otherwise someone could attach a new sibling, change the GUID, and
6016 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6017 */
6018 if (unspare) {
6019 ASSERT(cvd->vdev_isspare);
6020 spa_spare_remove(cvd);
6021 unspare_guid = cvd->vdev_guid;
6022 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6023 cvd->vdev_unspare = B_TRUE;
6024 }
6025
6026 /*
6027 * If the parent mirror/replacing vdev only has one child,
6028 * the parent is no longer needed. Remove it from the tree.
6029 */
6030 if (pvd->vdev_children == 1) {
6031 if (pvd->vdev_ops == &vdev_spare_ops)
6032 cvd->vdev_unspare = B_FALSE;
6033 vdev_remove_parent(cvd);
6034 }
6035
6036
6037 /*
6038 * We don't set tvd until now because the parent we just removed
6039 * may have been the previous top-level vdev.
6040 */
6041 tvd = cvd->vdev_top;
6042 ASSERT(tvd->vdev_parent == rvd);
6043
6044 /*
6045 * Reevaluate the parent vdev state.
6046 */
6047 vdev_propagate_state(cvd);
6048
6049 /*
6050 * If the 'autoexpand' property is set on the pool then automatically
6051 * try to expand the size of the pool. For example if the device we
6052 * just detached was smaller than the others, it may be possible to
6053 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6054 * first so that we can obtain the updated sizes of the leaf vdevs.
6055 */
6056 if (spa->spa_autoexpand) {
6057 vdev_reopen(tvd);
6058 vdev_expand(tvd, txg);
6059 }
6060
6061 vdev_config_dirty(tvd);
6062
6063 /*
6064 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6065 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6066 * But first make sure we're not on any *other* txg's DTL list, to
6067 * prevent vd from being accessed after it's freed.
6068 */
6069 vdpath = spa_strdup(vd->vdev_path);
6070 for (int t = 0; t < TXG_SIZE; t++)
6071 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6072 vd->vdev_detached = B_TRUE;
6073 vdev_dirty(tvd, VDD_DTL, vd, txg);
6074
6075 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6076
6077 /* hang on to the spa before we release the lock */
6078 spa_open_ref(spa, FTAG);
6079
6080 error = spa_vdev_exit(spa, vd, txg, 0);
6081
6082 spa_history_log_internal(spa, "detach", NULL,
6083 "vdev=%s", vdpath);
6084 spa_strfree(vdpath);
6085
6086 /*
6087 * If this was the removal of the original device in a hot spare vdev,
6088 * then we want to go through and remove the device from the hot spare
6089 * list of every other pool.
6090 */
6091 if (unspare) {
6092 spa_t *altspa = NULL;
6093
6094 mutex_enter(&spa_namespace_lock);
6095 while ((altspa = spa_next(altspa)) != NULL) {
6096 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6097 altspa == spa)
6098 continue;
6099
6100 spa_open_ref(altspa, FTAG);
6101 mutex_exit(&spa_namespace_lock);
6102 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6103 mutex_enter(&spa_namespace_lock);
6104 spa_close(altspa, FTAG);
6105 }
6106 mutex_exit(&spa_namespace_lock);
6107
6108 /* search the rest of the vdevs for spares to remove */
6109 spa_vdev_resilver_done(spa);
6110 }
6111
6112 /* all done with the spa; OK to release */
6113 mutex_enter(&spa_namespace_lock);
6114 spa_close(spa, FTAG);
6115 mutex_exit(&spa_namespace_lock);
6116
6117 return (error);
6118 }
6119
6120 int
6121 spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type)
6122 {
6123 /*
6124 * We hold the namespace lock through the whole function
6125 * to prevent any changes to the pool while we're starting or
6126 * stopping initialization. The config and state locks are held so that
6127 * we can properly assess the vdev state before we commit to
6128 * the initializing operation.
6129 */
6130 mutex_enter(&spa_namespace_lock);
6131 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6132
6133 /* Look up vdev and ensure it's a leaf. */
6134 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6135 if (vd == NULL || vd->vdev_detached) {
6136 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6137 mutex_exit(&spa_namespace_lock);
6138 return (SET_ERROR(ENODEV));
6139 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
6140 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6141 mutex_exit(&spa_namespace_lock);
6142 return (SET_ERROR(EINVAL));
6143 } else if (!vdev_writeable(vd)) {
6144 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6145 mutex_exit(&spa_namespace_lock);
6146 return (SET_ERROR(EROFS));
6147 }
6148 mutex_enter(&vd->vdev_initialize_lock);
6149 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6150
6151 /*
6152 * When we activate an initialize action we check to see
6153 * if the vdev_initialize_thread is NULL. We do this instead
6154 * of using the vdev_initialize_state since there might be
6155 * a previous initialization process which has completed but
6156 * the thread is not exited.
6157 */
6158 if (cmd_type == POOL_INITIALIZE_DO &&
6159 (vd->vdev_initialize_thread != NULL ||
6160 vd->vdev_top->vdev_removing)) {
6161 mutex_exit(&vd->vdev_initialize_lock);
6162 mutex_exit(&spa_namespace_lock);
6163 return (SET_ERROR(EBUSY));
6164 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
6165 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
6166 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
6167 mutex_exit(&vd->vdev_initialize_lock);
6168 mutex_exit(&spa_namespace_lock);
6169 return (SET_ERROR(ESRCH));
6170 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
6171 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
6172 mutex_exit(&vd->vdev_initialize_lock);
6173 mutex_exit(&spa_namespace_lock);
6174 return (SET_ERROR(ESRCH));
6175 }
6176
6177 switch (cmd_type) {
6178 case POOL_INITIALIZE_DO:
6179 vdev_initialize(vd);
6180 break;
6181 case POOL_INITIALIZE_CANCEL:
6182 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
6183 break;
6184 case POOL_INITIALIZE_SUSPEND:
6185 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED);
6186 break;
6187 default:
6188 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
6189 }
6190 mutex_exit(&vd->vdev_initialize_lock);
6191
6192 /* Sync out the initializing state */
6193 txg_wait_synced(spa->spa_dsl_pool, 0);
6194 mutex_exit(&spa_namespace_lock);
6195
6196 return (0);
6197 }
6198
6199
6200 /*
6201 * Split a set of devices from their mirrors, and create a new pool from them.
6202 */
6203 int
6204 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6205 nvlist_t *props, boolean_t exp)
6206 {
6207 int error = 0;
6208 uint64_t txg, *glist;
6209 spa_t *newspa;
6210 uint_t c, children, lastlog;
6211 nvlist_t **child, *nvl, *tmp;
6212 dmu_tx_t *tx;
6213 char *altroot = NULL;
6214 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6215 boolean_t activate_slog;
6216
6217 ASSERT(spa_writeable(spa));
6218
6219 txg = spa_vdev_enter(spa);
6220
6221 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6222 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6223 error = (spa_has_checkpoint(spa)) ?
6224 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6225 return (spa_vdev_exit(spa, NULL, txg, error));
6226 }
6227
6228 /* clear the log and flush everything up to now */
6229 activate_slog = spa_passivate_log(spa);
6230 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6231 error = spa_reset_logs(spa);
6232 txg = spa_vdev_config_enter(spa);
6233
6234 if (activate_slog)
6235 spa_activate_log(spa);
6236
6237 if (error != 0)
6238 return (spa_vdev_exit(spa, NULL, txg, error));
6239
6240 /* check new spa name before going any further */
6241 if (spa_lookup(newname) != NULL)
6242 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6243
6244 /*
6245 * scan through all the children to ensure they're all mirrors
6246 */
6247 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6248 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6249 &children) != 0)
6250 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6251
6252 /* first, check to ensure we've got the right child count */
6253 rvd = spa->spa_root_vdev;
6254 lastlog = 0;
6255 for (c = 0; c < rvd->vdev_children; c++) {
6256 vdev_t *vd = rvd->vdev_child[c];
6257
6258 /* don't count the holes & logs as children */
6259 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6260 if (lastlog == 0)
6261 lastlog = c;
6262 continue;
6263 }
6264
6265 lastlog = 0;
6266 }
6267 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6268 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6269
6270 /* next, ensure no spare or cache devices are part of the split */
6271 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6272 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6273 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6274
6275 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6276 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6277
6278 /* then, loop over each vdev and validate it */
6279 for (c = 0; c < children; c++) {
6280 uint64_t is_hole = 0;
6281
6282 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6283 &is_hole);
6284
6285 if (is_hole != 0) {
6286 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6287 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6288 continue;
6289 } else {
6290 error = SET_ERROR(EINVAL);
6291 break;
6292 }
6293 }
6294
6295 /* which disk is going to be split? */
6296 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6297 &glist[c]) != 0) {
6298 error = SET_ERROR(EINVAL);
6299 break;
6300 }
6301
6302 /* look it up in the spa */
6303 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6304 if (vml[c] == NULL) {
6305 error = SET_ERROR(ENODEV);
6306 break;
6307 }
6308
6309 /* make sure there's nothing stopping the split */
6310 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6311 vml[c]->vdev_islog ||
6312 !vdev_is_concrete(vml[c]) ||
6313 vml[c]->vdev_isspare ||
6314 vml[c]->vdev_isl2cache ||
6315 !vdev_writeable(vml[c]) ||
6316 vml[c]->vdev_children != 0 ||
6317 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6318 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6319 error = SET_ERROR(EINVAL);
6320 break;
6321 }
6322
6323 if (vdev_dtl_required(vml[c])) {
6324 error = SET_ERROR(EBUSY);
6325 break;
6326 }
6327
6328 /* we need certain info from the top level */
6329 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6330 vml[c]->vdev_top->vdev_ms_array) == 0);
6331 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6332 vml[c]->vdev_top->vdev_ms_shift) == 0);
6333 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6334 vml[c]->vdev_top->vdev_asize) == 0);
6335 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6336 vml[c]->vdev_top->vdev_ashift) == 0);
6337
6338 /* transfer per-vdev ZAPs */
6339 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6340 VERIFY0(nvlist_add_uint64(child[c],
6341 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6342
6343 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6344 VERIFY0(nvlist_add_uint64(child[c],
6345 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6346 vml[c]->vdev_parent->vdev_top_zap));
6347 }
6348
6349 if (error != 0) {
6350 kmem_free(vml, children * sizeof (vdev_t *));
6351 kmem_free(glist, children * sizeof (uint64_t));
6352 return (spa_vdev_exit(spa, NULL, txg, error));
6353 }
6354
6355 /* stop writers from using the disks */
6356 for (c = 0; c < children; c++) {
6357 if (vml[c] != NULL)
6358 vml[c]->vdev_offline = B_TRUE;
6359 }
6360 vdev_reopen(spa->spa_root_vdev);
6361
6362 /*
6363 * Temporarily record the splitting vdevs in the spa config. This
6364 * will disappear once the config is regenerated.
6365 */
6366 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6367 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6368 glist, children) == 0);
6369 kmem_free(glist, children * sizeof (uint64_t));
6370
6371 mutex_enter(&spa->spa_props_lock);
6372 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6373 nvl) == 0);
6374 mutex_exit(&spa->spa_props_lock);
6375 spa->spa_config_splitting = nvl;
6376 vdev_config_dirty(spa->spa_root_vdev);
6377
6378 /* configure and create the new pool */
6379 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6380 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6381 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6382 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6383 spa_version(spa)) == 0);
6384 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6385 spa->spa_config_txg) == 0);
6386 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6387 spa_generate_guid(NULL)) == 0);
6388 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6389 (void) nvlist_lookup_string(props,
6390 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6391
6392 /* add the new pool to the namespace */
6393 newspa = spa_add(newname, config, altroot);
6394 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6395 newspa->spa_config_txg = spa->spa_config_txg;
6396 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6397
6398 /* release the spa config lock, retaining the namespace lock */
6399 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6400
6401 if (zio_injection_enabled)
6402 zio_handle_panic_injection(spa, FTAG, 1);
6403
6404 spa_activate(newspa, spa_mode_global);
6405 spa_async_suspend(newspa);
6406
6407 for (c = 0; c < children; c++) {
6408 if (vml[c] != NULL) {
6409 /*
6410 * Temporarily stop the initializing activity. We set
6411 * the state to ACTIVE so that we know to resume
6412 * the initializing once the split has completed.
6413 */
6414 mutex_enter(&vml[c]->vdev_initialize_lock);
6415 vdev_initialize_stop(vml[c], VDEV_INITIALIZE_ACTIVE);
6416 mutex_exit(&vml[c]->vdev_initialize_lock);
6417 }
6418 }
6419
6420 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6421
6422 /* create the new pool from the disks of the original pool */
6423 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6424 if (error)
6425 goto out;
6426
6427 /* if that worked, generate a real config for the new pool */
6428 if (newspa->spa_root_vdev != NULL) {
6429 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6430 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6431 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6432 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6433 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6434 B_TRUE));
6435 }
6436
6437 /* set the props */
6438 if (props != NULL) {
6439 spa_configfile_set(newspa, props, B_FALSE);
6440 error = spa_prop_set(newspa, props);
6441 if (error)
6442 goto out;
6443 }
6444
6445 /* flush everything */
6446 txg = spa_vdev_config_enter(newspa);
6447 vdev_config_dirty(newspa->spa_root_vdev);
6448 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6449
6450 if (zio_injection_enabled)
6451 zio_handle_panic_injection(spa, FTAG, 2);
6452
6453 spa_async_resume(newspa);
6454
6455 /* finally, update the original pool's config */
6456 txg = spa_vdev_config_enter(spa);
6457 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6458 error = dmu_tx_assign(tx, TXG_WAIT);
6459 if (error != 0)
6460 dmu_tx_abort(tx);
6461 for (c = 0; c < children; c++) {
6462 if (vml[c] != NULL) {
6463 vdev_split(vml[c]);
6464 if (error == 0)
6465 spa_history_log_internal(spa, "detach", tx,
6466 "vdev=%s", vml[c]->vdev_path);
6467
6468 vdev_free(vml[c]);
6469 }
6470 }
6471 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6472 vdev_config_dirty(spa->spa_root_vdev);
6473 spa->spa_config_splitting = NULL;
6474 nvlist_free(nvl);
6475 if (error == 0)
6476 dmu_tx_commit(tx);
6477 (void) spa_vdev_exit(spa, NULL, txg, 0);
6478
6479 if (zio_injection_enabled)
6480 zio_handle_panic_injection(spa, FTAG, 3);
6481
6482 /* split is complete; log a history record */
6483 spa_history_log_internal(newspa, "split", NULL,
6484 "from pool %s", spa_name(spa));
6485
6486 kmem_free(vml, children * sizeof (vdev_t *));
6487
6488 /* if we're not going to mount the filesystems in userland, export */
6489 if (exp)
6490 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6491 B_FALSE, B_FALSE);
6492
6493 return (error);
6494
6495 out:
6496 spa_unload(newspa);
6497 spa_deactivate(newspa);
6498 spa_remove(newspa);
6499
6500 txg = spa_vdev_config_enter(spa);
6501
6502 /* re-online all offlined disks */
6503 for (c = 0; c < children; c++) {
6504 if (vml[c] != NULL)
6505 vml[c]->vdev_offline = B_FALSE;
6506 }
6507
6508 /* restart initializing disks as necessary */
6509 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
6510
6511 vdev_reopen(spa->spa_root_vdev);
6512
6513 nvlist_free(spa->spa_config_splitting);
6514 spa->spa_config_splitting = NULL;
6515 (void) spa_vdev_exit(spa, NULL, txg, error);
6516
6517 kmem_free(vml, children * sizeof (vdev_t *));
6518 return (error);
6519 }
6520
6521 /*
6522 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6523 * currently spared, so we can detach it.
6524 */
6525 static vdev_t *
6526 spa_vdev_resilver_done_hunt(vdev_t *vd)
6527 {
6528 vdev_t *newvd, *oldvd;
6529
6530 for (int c = 0; c < vd->vdev_children; c++) {
6531 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6532 if (oldvd != NULL)
6533 return (oldvd);
6534 }
6535
6536 /*
6537 * Check for a completed replacement. We always consider the first
6538 * vdev in the list to be the oldest vdev, and the last one to be
6539 * the newest (see spa_vdev_attach() for how that works). In
6540 * the case where the newest vdev is faulted, we will not automatically
6541 * remove it after a resilver completes. This is OK as it will require
6542 * user intervention to determine which disk the admin wishes to keep.
6543 */
6544 if (vd->vdev_ops == &vdev_replacing_ops) {
6545 ASSERT(vd->vdev_children > 1);
6546
6547 newvd = vd->vdev_child[vd->vdev_children - 1];
6548 oldvd = vd->vdev_child[0];
6549
6550 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6551 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6552 !vdev_dtl_required(oldvd))
6553 return (oldvd);
6554 }
6555
6556 /*
6557 * Check for a completed resilver with the 'unspare' flag set.
6558 * Also potentially update faulted state.
6559 */
6560 if (vd->vdev_ops == &vdev_spare_ops) {
6561 vdev_t *first = vd->vdev_child[0];
6562 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6563
6564 if (last->vdev_unspare) {
6565 oldvd = first;
6566 newvd = last;
6567 } else if (first->vdev_unspare) {
6568 oldvd = last;
6569 newvd = first;
6570 } else {
6571 oldvd = NULL;
6572 }
6573
6574 if (oldvd != NULL &&
6575 vdev_dtl_empty(newvd, DTL_MISSING) &&
6576 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6577 !vdev_dtl_required(oldvd))
6578 return (oldvd);
6579
6580 vdev_propagate_state(vd);
6581
6582 /*
6583 * If there are more than two spares attached to a disk,
6584 * and those spares are not required, then we want to
6585 * attempt to free them up now so that they can be used
6586 * by other pools. Once we're back down to a single
6587 * disk+spare, we stop removing them.
6588 */
6589 if (vd->vdev_children > 2) {
6590 newvd = vd->vdev_child[1];
6591
6592 if (newvd->vdev_isspare && last->vdev_isspare &&
6593 vdev_dtl_empty(last, DTL_MISSING) &&
6594 vdev_dtl_empty(last, DTL_OUTAGE) &&
6595 !vdev_dtl_required(newvd))
6596 return (newvd);
6597 }
6598 }
6599
6600 return (NULL);
6601 }
6602
6603 static void
6604 spa_vdev_resilver_done(spa_t *spa)
6605 {
6606 vdev_t *vd, *pvd, *ppvd;
6607 uint64_t guid, sguid, pguid, ppguid;
6608
6609 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6610
6611 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6612 pvd = vd->vdev_parent;
6613 ppvd = pvd->vdev_parent;
6614 guid = vd->vdev_guid;
6615 pguid = pvd->vdev_guid;
6616 ppguid = ppvd->vdev_guid;
6617 sguid = 0;
6618 /*
6619 * If we have just finished replacing a hot spared device, then
6620 * we need to detach the parent's first child (the original hot
6621 * spare) as well.
6622 */
6623 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6624 ppvd->vdev_children == 2) {
6625 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6626 sguid = ppvd->vdev_child[1]->vdev_guid;
6627 }
6628 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6629
6630 spa_config_exit(spa, SCL_ALL, FTAG);
6631 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6632 return;
6633 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6634 return;
6635 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6636 }
6637
6638 spa_config_exit(spa, SCL_ALL, FTAG);
6639 }
6640
6641 /*
6642 * Update the stored path or FRU for this vdev.
6643 */
6644 int
6645 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
6646 boolean_t ispath)
6647 {
6648 vdev_t *vd;
6649 boolean_t sync = B_FALSE;
6650
6651 ASSERT(spa_writeable(spa));
6652
6653 spa_vdev_state_enter(spa, SCL_ALL);
6654
6655 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
6656 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6657
6658 if (!vd->vdev_ops->vdev_op_leaf)
6659 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6660
6661 if (ispath) {
6662 if (strcmp(value, vd->vdev_path) != 0) {
6663 spa_strfree(vd->vdev_path);
6664 vd->vdev_path = spa_strdup(value);
6665 sync = B_TRUE;
6666 }
6667 } else {
6668 if (vd->vdev_fru == NULL) {
6669 vd->vdev_fru = spa_strdup(value);
6670 sync = B_TRUE;
6671 } else if (strcmp(value, vd->vdev_fru) != 0) {
6672 spa_strfree(vd->vdev_fru);
6673 vd->vdev_fru = spa_strdup(value);
6674 sync = B_TRUE;
6675 }
6676 }
6677
6678 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6679 }
6680
6681 int
6682 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6683 {
6684 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6685 }
6686
6687 int
6688 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6689 {
6690 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6691 }
6692
6693 /*
6694 * ==========================================================================
6695 * SPA Scanning
6696 * ==========================================================================
6697 */
6698 int
6699 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
6700 {
6701 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6702
6703 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6704 return (SET_ERROR(EBUSY));
6705
6706 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
6707 }
6708
6709 int
6710 spa_scan_stop(spa_t *spa)
6711 {
6712 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6713 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6714 return (SET_ERROR(EBUSY));
6715 return (dsl_scan_cancel(spa->spa_dsl_pool));
6716 }
6717
6718 int
6719 spa_scan(spa_t *spa, pool_scan_func_t func)
6720 {
6721 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6722
6723 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6724 return (SET_ERROR(ENOTSUP));
6725
6726 /*
6727 * If a resilver was requested, but there is no DTL on a
6728 * writeable leaf device, we have nothing to do.
6729 */
6730 if (func == POOL_SCAN_RESILVER &&
6731 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6732 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6733 return (0);
6734 }
6735
6736 return (dsl_scan(spa->spa_dsl_pool, func));
6737 }
6738
6739 /*
6740 * ==========================================================================
6741 * SPA async task processing
6742 * ==========================================================================
6743 */
6744
6745 static void
6746 spa_async_remove(spa_t *spa, vdev_t *vd)
6747 {
6748 if (vd->vdev_remove_wanted) {
6749 vd->vdev_remove_wanted = B_FALSE;
6750 vd->vdev_delayed_close = B_FALSE;
6751 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6752
6753 /*
6754 * We want to clear the stats, but we don't want to do a full
6755 * vdev_clear() as that will cause us to throw away
6756 * degraded/faulted state as well as attempt to reopen the
6757 * device, all of which is a waste.
6758 */
6759 vd->vdev_stat.vs_read_errors = 0;
6760 vd->vdev_stat.vs_write_errors = 0;
6761 vd->vdev_stat.vs_checksum_errors = 0;
6762
6763 vdev_state_dirty(vd->vdev_top);
6764 }
6765
6766 for (int c = 0; c < vd->vdev_children; c++)
6767 spa_async_remove(spa, vd->vdev_child[c]);
6768 }
6769
6770 static void
6771 spa_async_probe(spa_t *spa, vdev_t *vd)
6772 {
6773 if (vd->vdev_probe_wanted) {
6774 vd->vdev_probe_wanted = B_FALSE;
6775 vdev_reopen(vd); /* vdev_open() does the actual probe */
6776 }
6777
6778 for (int c = 0; c < vd->vdev_children; c++)
6779 spa_async_probe(spa, vd->vdev_child[c]);
6780 }
6781
6782 static void
6783 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6784 {
6785 sysevent_id_t eid;
6786 nvlist_t *attr;
6787 char *physpath;
6788
6789 if (!spa->spa_autoexpand)
6790 return;
6791
6792 for (int c = 0; c < vd->vdev_children; c++) {
6793 vdev_t *cvd = vd->vdev_child[c];
6794 spa_async_autoexpand(spa, cvd);
6795 }
6796
6797 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6798 return;
6799
6800 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6801 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6802
6803 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6804 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6805
6806 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6807 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
6808
6809 nvlist_free(attr);
6810 kmem_free(physpath, MAXPATHLEN);
6811 }
6812
6813 static void
6814 spa_async_thread(void *arg)
6815 {
6816 spa_t *spa = (spa_t *)arg;
6817 int tasks;
6818
6819 ASSERT(spa->spa_sync_on);
6820
6821 mutex_enter(&spa->spa_async_lock);
6822 tasks = spa->spa_async_tasks;
6823 spa->spa_async_tasks = 0;
6824 mutex_exit(&spa->spa_async_lock);
6825
6826 /*
6827 * See if the config needs to be updated.
6828 */
6829 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6830 uint64_t old_space, new_space;
6831
6832 mutex_enter(&spa_namespace_lock);
6833 old_space = metaslab_class_get_space(spa_normal_class(spa));
6834 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6835 new_space = metaslab_class_get_space(spa_normal_class(spa));
6836 mutex_exit(&spa_namespace_lock);
6837
6838 /*
6839 * If the pool grew as a result of the config update,
6840 * then log an internal history event.
6841 */
6842 if (new_space != old_space) {
6843 spa_history_log_internal(spa, "vdev online", NULL,
6844 "pool '%s' size: %llu(+%llu)",
6845 spa_name(spa), new_space, new_space - old_space);
6846 }
6847 }
6848
6849 /*
6850 * See if any devices need to be marked REMOVED.
6851 */
6852 if (tasks & SPA_ASYNC_REMOVE) {
6853 spa_vdev_state_enter(spa, SCL_NONE);
6854 spa_async_remove(spa, spa->spa_root_vdev);
6855 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6856 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6857 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6858 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6859 (void) spa_vdev_state_exit(spa, NULL, 0);
6860 }
6861
6862 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6863 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6864 spa_async_autoexpand(spa, spa->spa_root_vdev);
6865 spa_config_exit(spa, SCL_CONFIG, FTAG);
6866 }
6867
6868 /*
6869 * See if any devices need to be probed.
6870 */
6871 if (tasks & SPA_ASYNC_PROBE) {
6872 spa_vdev_state_enter(spa, SCL_NONE);
6873 spa_async_probe(spa, spa->spa_root_vdev);
6874 (void) spa_vdev_state_exit(spa, NULL, 0);
6875 }
6876
6877 /*
6878 * If any devices are done replacing, detach them.
6879 */
6880 if (tasks & SPA_ASYNC_RESILVER_DONE)
6881 spa_vdev_resilver_done(spa);
6882
6883 /*
6884 * Kick off a resilver.
6885 */
6886 if (tasks & SPA_ASYNC_RESILVER)
6887 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6888
6889 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
6890 mutex_enter(&spa_namespace_lock);
6891 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6892 vdev_initialize_restart(spa->spa_root_vdev);
6893 spa_config_exit(spa, SCL_CONFIG, FTAG);
6894 mutex_exit(&spa_namespace_lock);
6895 }
6896
6897 /*
6898 * Let the world know that we're done.
6899 */
6900 mutex_enter(&spa->spa_async_lock);
6901 spa->spa_async_thread = NULL;
6902 cv_broadcast(&spa->spa_async_cv);
6903 mutex_exit(&spa->spa_async_lock);
6904 thread_exit();
6905 }
6906
6907 void
6908 spa_async_suspend(spa_t *spa)
6909 {
6910 mutex_enter(&spa->spa_async_lock);
6911 spa->spa_async_suspended++;
6912 while (spa->spa_async_thread != NULL)
6913 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6914 mutex_exit(&spa->spa_async_lock);
6915
6916 spa_vdev_remove_suspend(spa);
6917
6918 zthr_t *condense_thread = spa->spa_condense_zthr;
6919 if (condense_thread != NULL && zthr_isrunning(condense_thread))
6920 VERIFY0(zthr_cancel(condense_thread));
6921
6922 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
6923 if (discard_thread != NULL && zthr_isrunning(discard_thread))
6924 VERIFY0(zthr_cancel(discard_thread));
6925 }
6926
6927 void
6928 spa_async_resume(spa_t *spa)
6929 {
6930 mutex_enter(&spa->spa_async_lock);
6931 ASSERT(spa->spa_async_suspended != 0);
6932 spa->spa_async_suspended--;
6933 mutex_exit(&spa->spa_async_lock);
6934 spa_restart_removal(spa);
6935
6936 zthr_t *condense_thread = spa->spa_condense_zthr;
6937 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
6938 zthr_resume(condense_thread);
6939
6940 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
6941 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
6942 zthr_resume(discard_thread);
6943 }
6944
6945 static boolean_t
6946 spa_async_tasks_pending(spa_t *spa)
6947 {
6948 uint_t non_config_tasks;
6949 uint_t config_task;
6950 boolean_t config_task_suspended;
6951
6952 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
6953 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6954 if (spa->spa_ccw_fail_time == 0) {
6955 config_task_suspended = B_FALSE;
6956 } else {
6957 config_task_suspended =
6958 (gethrtime() - spa->spa_ccw_fail_time) <
6959 (zfs_ccw_retry_interval * NANOSEC);
6960 }
6961
6962 return (non_config_tasks || (config_task && !config_task_suspended));
6963 }
6964
6965 static void
6966 spa_async_dispatch(spa_t *spa)
6967 {
6968 mutex_enter(&spa->spa_async_lock);
6969 if (spa_async_tasks_pending(spa) &&
6970 !spa->spa_async_suspended &&
6971 spa->spa_async_thread == NULL &&
6972 rootdir != NULL)
6973 spa->spa_async_thread = thread_create(NULL, 0,
6974 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6975 mutex_exit(&spa->spa_async_lock);
6976 }
6977
6978 void
6979 spa_async_request(spa_t *spa, int task)
6980 {
6981 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6982 mutex_enter(&spa->spa_async_lock);
6983 spa->spa_async_tasks |= task;
6984 mutex_exit(&spa->spa_async_lock);
6985 }
6986
6987 /*
6988 * ==========================================================================
6989 * SPA syncing routines
6990 * ==========================================================================
6991 */
6992
6993 static int
6994 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6995 {
6996 bpobj_t *bpo = arg;
6997 bpobj_enqueue(bpo, bp, tx);
6998 return (0);
6999 }
7000
7001 static int
7002 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7003 {
7004 zio_t *zio = arg;
7005
7006 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7007 zio->io_flags));
7008 return (0);
7009 }
7010
7011 /*
7012 * Note: this simple function is not inlined to make it easier to dtrace the
7013 * amount of time spent syncing frees.
7014 */
7015 static void
7016 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7017 {
7018 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7019 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7020 VERIFY(zio_wait(zio) == 0);
7021 }
7022
7023 /*
7024 * Note: this simple function is not inlined to make it easier to dtrace the
7025 * amount of time spent syncing deferred frees.
7026 */
7027 static void
7028 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7029 {
7030 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7031 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7032 spa_free_sync_cb, zio, tx), ==, 0);
7033 VERIFY0(zio_wait(zio));
7034 }
7035
7036
7037 static void
7038 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7039 {
7040 char *packed = NULL;
7041 size_t bufsize;
7042 size_t nvsize = 0;
7043 dmu_buf_t *db;
7044
7045 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7046
7047 /*
7048 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7049 * information. This avoids the dmu_buf_will_dirty() path and
7050 * saves us a pre-read to get data we don't actually care about.
7051 */
7052 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7053 packed = kmem_alloc(bufsize, KM_SLEEP);
7054
7055 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7056 KM_SLEEP) == 0);
7057 bzero(packed + nvsize, bufsize - nvsize);
7058
7059 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7060
7061 kmem_free(packed, bufsize);
7062
7063 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7064 dmu_buf_will_dirty(db, tx);
7065 *(uint64_t *)db->db_data = nvsize;
7066 dmu_buf_rele(db, FTAG);
7067 }
7068
7069 static void
7070 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7071 const char *config, const char *entry)
7072 {
7073 nvlist_t *nvroot;
7074 nvlist_t **list;
7075 int i;
7076
7077 if (!sav->sav_sync)
7078 return;
7079
7080 /*
7081 * Update the MOS nvlist describing the list of available devices.
7082 * spa_validate_aux() will have already made sure this nvlist is
7083 * valid and the vdevs are labeled appropriately.
7084 */
7085 if (sav->sav_object == 0) {
7086 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7087 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7088 sizeof (uint64_t), tx);
7089 VERIFY(zap_update(spa->spa_meta_objset,
7090 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7091 &sav->sav_object, tx) == 0);
7092 }
7093
7094 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7095 if (sav->sav_count == 0) {
7096 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7097 } else {
7098 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7099 for (i = 0; i < sav->sav_count; i++)
7100 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7101 B_FALSE, VDEV_CONFIG_L2CACHE);
7102 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7103 sav->sav_count) == 0);
7104 for (i = 0; i < sav->sav_count; i++)
7105 nvlist_free(list[i]);
7106 kmem_free(list, sav->sav_count * sizeof (void *));
7107 }
7108
7109 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7110 nvlist_free(nvroot);
7111
7112 sav->sav_sync = B_FALSE;
7113 }
7114
7115 /*
7116 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7117 * The all-vdev ZAP must be empty.
7118 */
7119 static void
7120 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7121 {
7122 spa_t *spa = vd->vdev_spa;
7123 if (vd->vdev_top_zap != 0) {
7124 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7125 vd->vdev_top_zap, tx));
7126 }
7127 if (vd->vdev_leaf_zap != 0) {
7128 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7129 vd->vdev_leaf_zap, tx));
7130 }
7131 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7132 spa_avz_build(vd->vdev_child[i], avz, tx);
7133 }
7134 }
7135
7136 static void
7137 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7138 {
7139 nvlist_t *config;
7140
7141 /*
7142 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7143 * its config may not be dirty but we still need to build per-vdev ZAPs.
7144 * Similarly, if the pool is being assembled (e.g. after a split), we
7145 * need to rebuild the AVZ although the config may not be dirty.
7146 */
7147 if (list_is_empty(&spa->spa_config_dirty_list) &&
7148 spa->spa_avz_action == AVZ_ACTION_NONE)
7149 return;
7150
7151 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7152
7153 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7154 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7155 spa->spa_all_vdev_zaps != 0);
7156
7157 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7158 /* Make and build the new AVZ */
7159 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7160 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7161 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7162
7163 /* Diff old AVZ with new one */
7164 zap_cursor_t zc;
7165 zap_attribute_t za;
7166
7167 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7168 spa->spa_all_vdev_zaps);
7169 zap_cursor_retrieve(&zc, &za) == 0;
7170 zap_cursor_advance(&zc)) {
7171 uint64_t vdzap = za.za_first_integer;
7172 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7173 vdzap) == ENOENT) {
7174 /*
7175 * ZAP is listed in old AVZ but not in new one;
7176 * destroy it
7177 */
7178 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7179 tx));
7180 }
7181 }
7182
7183 zap_cursor_fini(&zc);
7184
7185 /* Destroy the old AVZ */
7186 VERIFY0(zap_destroy(spa->spa_meta_objset,
7187 spa->spa_all_vdev_zaps, tx));
7188
7189 /* Replace the old AVZ in the dir obj with the new one */
7190 VERIFY0(zap_update(spa->spa_meta_objset,
7191 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7192 sizeof (new_avz), 1, &new_avz, tx));
7193
7194 spa->spa_all_vdev_zaps = new_avz;
7195 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7196 zap_cursor_t zc;
7197 zap_attribute_t za;
7198
7199 /* Walk through the AVZ and destroy all listed ZAPs */
7200 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7201 spa->spa_all_vdev_zaps);
7202 zap_cursor_retrieve(&zc, &za) == 0;
7203 zap_cursor_advance(&zc)) {
7204 uint64_t zap = za.za_first_integer;
7205 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7206 }
7207
7208 zap_cursor_fini(&zc);
7209
7210 /* Destroy and unlink the AVZ itself */
7211 VERIFY0(zap_destroy(spa->spa_meta_objset,
7212 spa->spa_all_vdev_zaps, tx));
7213 VERIFY0(zap_remove(spa->spa_meta_objset,
7214 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7215 spa->spa_all_vdev_zaps = 0;
7216 }
7217
7218 if (spa->spa_all_vdev_zaps == 0) {
7219 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7220 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7221 DMU_POOL_VDEV_ZAP_MAP, tx);
7222 }
7223 spa->spa_avz_action = AVZ_ACTION_NONE;
7224
7225 /* Create ZAPs for vdevs that don't have them. */
7226 vdev_construct_zaps(spa->spa_root_vdev, tx);
7227
7228 config = spa_config_generate(spa, spa->spa_root_vdev,
7229 dmu_tx_get_txg(tx), B_FALSE);
7230
7231 /*
7232 * If we're upgrading the spa version then make sure that
7233 * the config object gets updated with the correct version.
7234 */
7235 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7236 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7237 spa->spa_uberblock.ub_version);
7238
7239 spa_config_exit(spa, SCL_STATE, FTAG);
7240
7241 nvlist_free(spa->spa_config_syncing);
7242 spa->spa_config_syncing = config;
7243
7244 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7245 }
7246
7247 static void
7248 spa_sync_version(void *arg, dmu_tx_t *tx)
7249 {
7250 uint64_t *versionp = arg;
7251 uint64_t version = *versionp;
7252 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7253
7254 /*
7255 * Setting the version is special cased when first creating the pool.
7256 */
7257 ASSERT(tx->tx_txg != TXG_INITIAL);
7258
7259 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7260 ASSERT(version >= spa_version(spa));
7261
7262 spa->spa_uberblock.ub_version = version;
7263 vdev_config_dirty(spa->spa_root_vdev);
7264 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7265 }
7266
7267 /*
7268 * Set zpool properties.
7269 */
7270 static void
7271 spa_sync_props(void *arg, dmu_tx_t *tx)
7272 {
7273 nvlist_t *nvp = arg;
7274 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7275 objset_t *mos = spa->spa_meta_objset;
7276 nvpair_t *elem = NULL;
7277
7278 mutex_enter(&spa->spa_props_lock);
7279
7280 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7281 uint64_t intval;
7282 char *strval, *fname;
7283 zpool_prop_t prop;
7284 const char *propname;
7285 zprop_type_t proptype;
7286 spa_feature_t fid;
7287
7288 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7289 case ZPOOL_PROP_INVAL:
7290 /*
7291 * We checked this earlier in spa_prop_validate().
7292 */
7293 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7294
7295 fname = strchr(nvpair_name(elem), '@') + 1;
7296 VERIFY0(zfeature_lookup_name(fname, &fid));
7297
7298 spa_feature_enable(spa, fid, tx);
7299 spa_history_log_internal(spa, "set", tx,
7300 "%s=enabled", nvpair_name(elem));
7301 break;
7302
7303 case ZPOOL_PROP_VERSION:
7304 intval = fnvpair_value_uint64(elem);
7305 /*
7306 * The version is synced seperatly before other
7307 * properties and should be correct by now.
7308 */
7309 ASSERT3U(spa_version(spa), >=, intval);
7310 break;
7311
7312 case ZPOOL_PROP_ALTROOT:
7313 /*
7314 * 'altroot' is a non-persistent property. It should
7315 * have been set temporarily at creation or import time.
7316 */
7317 ASSERT(spa->spa_root != NULL);
7318 break;
7319
7320 case ZPOOL_PROP_READONLY:
7321 case ZPOOL_PROP_CACHEFILE:
7322 /*
7323 * 'readonly' and 'cachefile' are also non-persisitent
7324 * properties.
7325 */
7326 break;
7327 case ZPOOL_PROP_COMMENT:
7328 strval = fnvpair_value_string(elem);
7329 if (spa->spa_comment != NULL)
7330 spa_strfree(spa->spa_comment);
7331 spa->spa_comment = spa_strdup(strval);
7332 /*
7333 * We need to dirty the configuration on all the vdevs
7334 * so that their labels get updated. It's unnecessary
7335 * to do this for pool creation since the vdev's
7336 * configuratoin has already been dirtied.
7337 */
7338 if (tx->tx_txg != TXG_INITIAL)
7339 vdev_config_dirty(spa->spa_root_vdev);
7340 spa_history_log_internal(spa, "set", tx,
7341 "%s=%s", nvpair_name(elem), strval);
7342 break;
7343 default:
7344 /*
7345 * Set pool property values in the poolprops mos object.
7346 */
7347 if (spa->spa_pool_props_object == 0) {
7348 spa->spa_pool_props_object =
7349 zap_create_link(mos, DMU_OT_POOL_PROPS,
7350 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7351 tx);
7352 }
7353
7354 /* normalize the property name */
7355 propname = zpool_prop_to_name(prop);
7356 proptype = zpool_prop_get_type(prop);
7357
7358 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7359 ASSERT(proptype == PROP_TYPE_STRING);
7360 strval = fnvpair_value_string(elem);
7361 VERIFY0(zap_update(mos,
7362 spa->spa_pool_props_object, propname,
7363 1, strlen(strval) + 1, strval, tx));
7364 spa_history_log_internal(spa, "set", tx,
7365 "%s=%s", nvpair_name(elem), strval);
7366 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7367 intval = fnvpair_value_uint64(elem);
7368
7369 if (proptype == PROP_TYPE_INDEX) {
7370 const char *unused;
7371 VERIFY0(zpool_prop_index_to_string(
7372 prop, intval, &unused));
7373 }
7374 VERIFY0(zap_update(mos,
7375 spa->spa_pool_props_object, propname,
7376 8, 1, &intval, tx));
7377 spa_history_log_internal(spa, "set", tx,
7378 "%s=%lld", nvpair_name(elem), intval);
7379 } else {
7380 ASSERT(0); /* not allowed */
7381 }
7382
7383 switch (prop) {
7384 case ZPOOL_PROP_DELEGATION:
7385 spa->spa_delegation = intval;
7386 break;
7387 case ZPOOL_PROP_BOOTFS:
7388 spa->spa_bootfs = intval;
7389 break;
7390 case ZPOOL_PROP_FAILUREMODE:
7391 spa->spa_failmode = intval;
7392 break;
7393 case ZPOOL_PROP_AUTOEXPAND:
7394 spa->spa_autoexpand = intval;
7395 if (tx->tx_txg != TXG_INITIAL)
7396 spa_async_request(spa,
7397 SPA_ASYNC_AUTOEXPAND);
7398 break;
7399 case ZPOOL_PROP_DEDUPDITTO:
7400 spa->spa_dedup_ditto = intval;
7401 break;
7402 default:
7403 break;
7404 }
7405 }
7406
7407 }
7408
7409 mutex_exit(&spa->spa_props_lock);
7410 }
7411
7412 /*
7413 * Perform one-time upgrade on-disk changes. spa_version() does not
7414 * reflect the new version this txg, so there must be no changes this
7415 * txg to anything that the upgrade code depends on after it executes.
7416 * Therefore this must be called after dsl_pool_sync() does the sync
7417 * tasks.
7418 */
7419 static void
7420 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7421 {
7422 dsl_pool_t *dp = spa->spa_dsl_pool;
7423
7424 ASSERT(spa->spa_sync_pass == 1);
7425
7426 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7427
7428 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7429 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7430 dsl_pool_create_origin(dp, tx);
7431
7432 /* Keeping the origin open increases spa_minref */
7433 spa->spa_minref += 3;
7434 }
7435
7436 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7437 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7438 dsl_pool_upgrade_clones(dp, tx);
7439 }
7440
7441 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7442 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7443 dsl_pool_upgrade_dir_clones(dp, tx);
7444
7445 /* Keeping the freedir open increases spa_minref */
7446 spa->spa_minref += 3;
7447 }
7448
7449 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7450 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7451 spa_feature_create_zap_objects(spa, tx);
7452 }
7453
7454 /*
7455 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7456 * when possibility to use lz4 compression for metadata was added
7457 * Old pools that have this feature enabled must be upgraded to have
7458 * this feature active
7459 */
7460 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7461 boolean_t lz4_en = spa_feature_is_enabled(spa,
7462 SPA_FEATURE_LZ4_COMPRESS);
7463 boolean_t lz4_ac = spa_feature_is_active(spa,
7464 SPA_FEATURE_LZ4_COMPRESS);
7465
7466 if (lz4_en && !lz4_ac)
7467 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7468 }
7469
7470 /*
7471 * If we haven't written the salt, do so now. Note that the
7472 * feature may not be activated yet, but that's fine since
7473 * the presence of this ZAP entry is backwards compatible.
7474 */
7475 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7476 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7477 VERIFY0(zap_add(spa->spa_meta_objset,
7478 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7479 sizeof (spa->spa_cksum_salt.zcs_bytes),
7480 spa->spa_cksum_salt.zcs_bytes, tx));
7481 }
7482
7483 rrw_exit(&dp->dp_config_rwlock, FTAG);
7484 }
7485
7486 static void
7487 vdev_indirect_state_sync_verify(vdev_t *vd)
7488 {
7489 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7490 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7491
7492 if (vd->vdev_ops == &vdev_indirect_ops) {
7493 ASSERT(vim != NULL);
7494 ASSERT(vib != NULL);
7495 }
7496
7497 if (vdev_obsolete_sm_object(vd) != 0) {
7498 ASSERT(vd->vdev_obsolete_sm != NULL);
7499 ASSERT(vd->vdev_removing ||
7500 vd->vdev_ops == &vdev_indirect_ops);
7501 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7502 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7503
7504 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7505 space_map_object(vd->vdev_obsolete_sm));
7506 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7507 space_map_allocated(vd->vdev_obsolete_sm));
7508 }
7509 ASSERT(vd->vdev_obsolete_segments != NULL);
7510
7511 /*
7512 * Since frees / remaps to an indirect vdev can only
7513 * happen in syncing context, the obsolete segments
7514 * tree must be empty when we start syncing.
7515 */
7516 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7517 }
7518
7519 /*
7520 * Sync the specified transaction group. New blocks may be dirtied as
7521 * part of the process, so we iterate until it converges.
7522 */
7523 void
7524 spa_sync(spa_t *spa, uint64_t txg)
7525 {
7526 dsl_pool_t *dp = spa->spa_dsl_pool;
7527 objset_t *mos = spa->spa_meta_objset;
7528 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7529 vdev_t *rvd = spa->spa_root_vdev;
7530 vdev_t *vd;
7531 dmu_tx_t *tx;
7532 int error;
7533 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7534 zfs_vdev_queue_depth_pct / 100;
7535
7536 VERIFY(spa_writeable(spa));
7537
7538 /*
7539 * Wait for i/os issued in open context that need to complete
7540 * before this txg syncs.
7541 */
7542 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
7543 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
7544 ZIO_FLAG_CANFAIL);
7545
7546 /*
7547 * Lock out configuration changes.
7548 */
7549 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7550
7551 spa->spa_syncing_txg = txg;
7552 spa->spa_sync_pass = 0;
7553
7554 for (int i = 0; i < spa->spa_alloc_count; i++) {
7555 mutex_enter(&spa->spa_alloc_locks[i]);
7556 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
7557 mutex_exit(&spa->spa_alloc_locks[i]);
7558 }
7559
7560 /*
7561 * If there are any pending vdev state changes, convert them
7562 * into config changes that go out with this transaction group.
7563 */
7564 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7565 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7566 /*
7567 * We need the write lock here because, for aux vdevs,
7568 * calling vdev_config_dirty() modifies sav_config.
7569 * This is ugly and will become unnecessary when we
7570 * eliminate the aux vdev wart by integrating all vdevs
7571 * into the root vdev tree.
7572 */
7573 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7574 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7575 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7576 vdev_state_clean(vd);
7577 vdev_config_dirty(vd);
7578 }
7579 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7580 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7581 }
7582 spa_config_exit(spa, SCL_STATE, FTAG);
7583
7584 tx = dmu_tx_create_assigned(dp, txg);
7585
7586 spa->spa_sync_starttime = gethrtime();
7587 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7588 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7589
7590 /*
7591 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7592 * set spa_deflate if we have no raid-z vdevs.
7593 */
7594 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
7595 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
7596 int i;
7597
7598 for (i = 0; i < rvd->vdev_children; i++) {
7599 vd = rvd->vdev_child[i];
7600 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
7601 break;
7602 }
7603 if (i == rvd->vdev_children) {
7604 spa->spa_deflate = TRUE;
7605 VERIFY(0 == zap_add(spa->spa_meta_objset,
7606 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
7607 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
7608 }
7609 }
7610
7611 /*
7612 * Set the top-level vdev's max queue depth. Evaluate each
7613 * top-level's async write queue depth in case it changed.
7614 * The max queue depth will not change in the middle of syncing
7615 * out this txg.
7616 */
7617 uint64_t slots_per_allocator = 0;
7618 for (int c = 0; c < rvd->vdev_children; c++) {
7619 vdev_t *tvd = rvd->vdev_child[c];
7620 metaslab_group_t *mg = tvd->vdev_mg;
7621
7622 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
7623 !metaslab_group_initialized(mg))
7624 continue;
7625
7626 /*
7627 * It is safe to do a lock-free check here because only async
7628 * allocations look at mg_max_alloc_queue_depth, and async
7629 * allocations all happen from spa_sync().
7630 */
7631 for (int i = 0; i < spa->spa_alloc_count; i++)
7632 ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i])));
7633 mg->mg_max_alloc_queue_depth = max_queue_depth;
7634
7635 for (int i = 0; i < spa->spa_alloc_count; i++) {
7636 mg->mg_cur_max_alloc_queue_depth[i] =
7637 zfs_vdev_def_queue_depth;
7638 }
7639 slots_per_allocator += zfs_vdev_def_queue_depth;
7640 }
7641 metaslab_class_t *mc = spa_normal_class(spa);
7642 for (int i = 0; i < spa->spa_alloc_count; i++) {
7643 ASSERT0(refcount_count(&mc->mc_alloc_slots[i]));
7644 mc->mc_alloc_max_slots[i] = slots_per_allocator;
7645 }
7646 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
7647
7648 for (int c = 0; c < rvd->vdev_children; c++) {
7649 vdev_t *vd = rvd->vdev_child[c];
7650 vdev_indirect_state_sync_verify(vd);
7651
7652 if (vdev_indirect_should_condense(vd)) {
7653 spa_condense_indirect_start_sync(vd, tx);
7654 break;
7655 }
7656 }
7657
7658 /*
7659 * Iterate to convergence.
7660 */
7661 do {
7662 int pass = ++spa->spa_sync_pass;
7663
7664 spa_sync_config_object(spa, tx);
7665 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
7666 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
7667 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
7668 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
7669 spa_errlog_sync(spa, txg);
7670 dsl_pool_sync(dp, txg);
7671
7672 if (pass < zfs_sync_pass_deferred_free) {
7673 spa_sync_frees(spa, free_bpl, tx);
7674 } else {
7675 /*
7676 * We can not defer frees in pass 1, because
7677 * we sync the deferred frees later in pass 1.
7678 */
7679 ASSERT3U(pass, >, 1);
7680 bplist_iterate(free_bpl, bpobj_enqueue_cb,
7681 &spa->spa_deferred_bpobj, tx);
7682 }
7683
7684 ddt_sync(spa, txg);
7685 dsl_scan_sync(dp, tx);
7686
7687 if (spa->spa_vdev_removal != NULL)
7688 svr_sync(spa, tx);
7689
7690 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
7691 != NULL)
7692 vdev_sync(vd, txg);
7693
7694 if (pass == 1) {
7695 spa_sync_upgrades(spa, tx);
7696 ASSERT3U(txg, >=,
7697 spa->spa_uberblock.ub_rootbp.blk_birth);
7698 /*
7699 * Note: We need to check if the MOS is dirty
7700 * because we could have marked the MOS dirty
7701 * without updating the uberblock (e.g. if we
7702 * have sync tasks but no dirty user data). We
7703 * need to check the uberblock's rootbp because
7704 * it is updated if we have synced out dirty
7705 * data (though in this case the MOS will most
7706 * likely also be dirty due to second order
7707 * effects, we don't want to rely on that here).
7708 */
7709 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7710 !dmu_objset_is_dirty(mos, txg)) {
7711 /*
7712 * Nothing changed on the first pass,
7713 * therefore this TXG is a no-op. Avoid
7714 * syncing deferred frees, so that we
7715 * can keep this TXG as a no-op.
7716 */
7717 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7718 txg));
7719 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7720 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7721 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
7722 txg));
7723 break;
7724 }
7725 spa_sync_deferred_frees(spa, tx);
7726 }
7727
7728 } while (dmu_objset_is_dirty(mos, txg));
7729
7730 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7731 /*
7732 * Make sure that the number of ZAPs for all the vdevs matches
7733 * the number of ZAPs in the per-vdev ZAP list. This only gets
7734 * called if the config is dirty; otherwise there may be
7735 * outstanding AVZ operations that weren't completed in
7736 * spa_sync_config_object.
7737 */
7738 uint64_t all_vdev_zap_entry_count;
7739 ASSERT0(zap_count(spa->spa_meta_objset,
7740 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7741 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7742 all_vdev_zap_entry_count);
7743 }
7744
7745 if (spa->spa_vdev_removal != NULL) {
7746 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
7747 }
7748
7749 /*
7750 * Rewrite the vdev configuration (which includes the uberblock)
7751 * to commit the transaction group.
7752 *
7753 * If there are no dirty vdevs, we sync the uberblock to a few
7754 * random top-level vdevs that are known to be visible in the
7755 * config cache (see spa_vdev_add() for a complete description).
7756 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7757 */
7758 for (;;) {
7759 /*
7760 * We hold SCL_STATE to prevent vdev open/close/etc.
7761 * while we're attempting to write the vdev labels.
7762 */
7763 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7764
7765 if (list_is_empty(&spa->spa_config_dirty_list)) {
7766 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
7767 int svdcount = 0;
7768 int children = rvd->vdev_children;
7769 int c0 = spa_get_random(children);
7770
7771 for (int c = 0; c < children; c++) {
7772 vd = rvd->vdev_child[(c0 + c) % children];
7773
7774 /* Stop when revisiting the first vdev */
7775 if (c > 0 && svd[0] == vd)
7776 break;
7777
7778 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
7779 !vdev_is_concrete(vd))
7780 continue;
7781
7782 svd[svdcount++] = vd;
7783 if (svdcount == SPA_SYNC_MIN_VDEVS)
7784 break;
7785 }
7786 error = vdev_config_sync(svd, svdcount, txg);
7787 } else {
7788 error = vdev_config_sync(rvd->vdev_child,
7789 rvd->vdev_children, txg);
7790 }
7791
7792 if (error == 0)
7793 spa->spa_last_synced_guid = rvd->vdev_guid;
7794
7795 spa_config_exit(spa, SCL_STATE, FTAG);
7796
7797 if (error == 0)
7798 break;
7799 zio_suspend(spa, NULL);
7800 zio_resume_wait(spa);
7801 }
7802 dmu_tx_commit(tx);
7803
7804 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
7805
7806 /*
7807 * Clear the dirty config list.
7808 */
7809 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
7810 vdev_config_clean(vd);
7811
7812 /*
7813 * Now that the new config has synced transactionally,
7814 * let it become visible to the config cache.
7815 */
7816 if (spa->spa_config_syncing != NULL) {
7817 spa_config_set(spa, spa->spa_config_syncing);
7818 spa->spa_config_txg = txg;
7819 spa->spa_config_syncing = NULL;
7820 }
7821
7822 dsl_pool_sync_done(dp, txg);
7823
7824 for (int i = 0; i < spa->spa_alloc_count; i++) {
7825 mutex_enter(&spa->spa_alloc_locks[i]);
7826 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
7827 mutex_exit(&spa->spa_alloc_locks[i]);
7828 }
7829
7830 /*
7831 * Update usable space statistics.
7832 */
7833 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
7834 != NULL)
7835 vdev_sync_done(vd, txg);
7836
7837 spa_update_dspace(spa);
7838
7839 /*
7840 * It had better be the case that we didn't dirty anything
7841 * since vdev_config_sync().
7842 */
7843 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
7844 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7845 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
7846
7847 while (zfs_pause_spa_sync)
7848 delay(1);
7849
7850 spa->spa_sync_pass = 0;
7851
7852 /*
7853 * Update the last synced uberblock here. We want to do this at
7854 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7855 * will be guaranteed that all the processing associated with
7856 * that txg has been completed.
7857 */
7858 spa->spa_ubsync = spa->spa_uberblock;
7859 spa_config_exit(spa, SCL_CONFIG, FTAG);
7860
7861 spa_handle_ignored_writes(spa);
7862
7863 /*
7864 * If any async tasks have been requested, kick them off.
7865 */
7866 spa_async_dispatch(spa);
7867 }
7868
7869 /*
7870 * Sync all pools. We don't want to hold the namespace lock across these
7871 * operations, so we take a reference on the spa_t and drop the lock during the
7872 * sync.
7873 */
7874 void
7875 spa_sync_allpools(void)
7876 {
7877 spa_t *spa = NULL;
7878 mutex_enter(&spa_namespace_lock);
7879 while ((spa = spa_next(spa)) != NULL) {
7880 if (spa_state(spa) != POOL_STATE_ACTIVE ||
7881 !spa_writeable(spa) || spa_suspended(spa))
7882 continue;
7883 spa_open_ref(spa, FTAG);
7884 mutex_exit(&spa_namespace_lock);
7885 txg_wait_synced(spa_get_dsl(spa), 0);
7886 mutex_enter(&spa_namespace_lock);
7887 spa_close(spa, FTAG);
7888 }
7889 mutex_exit(&spa_namespace_lock);
7890 }
7891
7892 /*
7893 * ==========================================================================
7894 * Miscellaneous routines
7895 * ==========================================================================
7896 */
7897
7898 /*
7899 * Remove all pools in the system.
7900 */
7901 void
7902 spa_evict_all(void)
7903 {
7904 spa_t *spa;
7905
7906 /*
7907 * Remove all cached state. All pools should be closed now,
7908 * so every spa in the AVL tree should be unreferenced.
7909 */
7910 mutex_enter(&spa_namespace_lock);
7911 while ((spa = spa_next(NULL)) != NULL) {
7912 /*
7913 * Stop async tasks. The async thread may need to detach
7914 * a device that's been replaced, which requires grabbing
7915 * spa_namespace_lock, so we must drop it here.
7916 */
7917 spa_open_ref(spa, FTAG);
7918 mutex_exit(&spa_namespace_lock);
7919 spa_async_suspend(spa);
7920 mutex_enter(&spa_namespace_lock);
7921 spa_close(spa, FTAG);
7922
7923 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
7924 spa_unload(spa);
7925 spa_deactivate(spa);
7926 }
7927 spa_remove(spa);
7928 }
7929 mutex_exit(&spa_namespace_lock);
7930 }
7931
7932 vdev_t *
7933 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
7934 {
7935 vdev_t *vd;
7936 int i;
7937
7938 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
7939 return (vd);
7940
7941 if (aux) {
7942 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7943 vd = spa->spa_l2cache.sav_vdevs[i];
7944 if (vd->vdev_guid == guid)
7945 return (vd);
7946 }
7947
7948 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7949 vd = spa->spa_spares.sav_vdevs[i];
7950 if (vd->vdev_guid == guid)
7951 return (vd);
7952 }
7953 }
7954
7955 return (NULL);
7956 }
7957
7958 void
7959 spa_upgrade(spa_t *spa, uint64_t version)
7960 {
7961 ASSERT(spa_writeable(spa));
7962
7963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7964
7965 /*
7966 * This should only be called for a non-faulted pool, and since a
7967 * future version would result in an unopenable pool, this shouldn't be
7968 * possible.
7969 */
7970 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7971 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7972
7973 spa->spa_uberblock.ub_version = version;
7974 vdev_config_dirty(spa->spa_root_vdev);
7975
7976 spa_config_exit(spa, SCL_ALL, FTAG);
7977
7978 txg_wait_synced(spa_get_dsl(spa), 0);
7979 }
7980
7981 boolean_t
7982 spa_has_spare(spa_t *spa, uint64_t guid)
7983 {
7984 int i;
7985 uint64_t spareguid;
7986 spa_aux_vdev_t *sav = &spa->spa_spares;
7987
7988 for (i = 0; i < sav->sav_count; i++)
7989 if (sav->sav_vdevs[i]->vdev_guid == guid)
7990 return (B_TRUE);
7991
7992 for (i = 0; i < sav->sav_npending; i++) {
7993 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7994 &spareguid) == 0 && spareguid == guid)
7995 return (B_TRUE);
7996 }
7997
7998 return (B_FALSE);
7999 }
8000
8001 /*
8002 * Check if a pool has an active shared spare device.
8003 * Note: reference count of an active spare is 2, as a spare and as a replace
8004 */
8005 static boolean_t
8006 spa_has_active_shared_spare(spa_t *spa)
8007 {
8008 int i, refcnt;
8009 uint64_t pool;
8010 spa_aux_vdev_t *sav = &spa->spa_spares;
8011
8012 for (i = 0; i < sav->sav_count; i++) {
8013 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8014 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8015 refcnt > 2)
8016 return (B_TRUE);
8017 }
8018
8019 return (B_FALSE);
8020 }
8021
8022 sysevent_t *
8023 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8024 {
8025 sysevent_t *ev = NULL;
8026 #ifdef _KERNEL
8027 sysevent_attr_list_t *attr = NULL;
8028 sysevent_value_t value;
8029
8030 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8031 SE_SLEEP);
8032 ASSERT(ev != NULL);
8033
8034 value.value_type = SE_DATA_TYPE_STRING;
8035 value.value.sv_string = spa_name(spa);
8036 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8037 goto done;
8038
8039 value.value_type = SE_DATA_TYPE_UINT64;
8040 value.value.sv_uint64 = spa_guid(spa);
8041 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8042 goto done;
8043
8044 if (vd) {
8045 value.value_type = SE_DATA_TYPE_UINT64;
8046 value.value.sv_uint64 = vd->vdev_guid;
8047 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8048 SE_SLEEP) != 0)
8049 goto done;
8050
8051 if (vd->vdev_path) {
8052 value.value_type = SE_DATA_TYPE_STRING;
8053 value.value.sv_string = vd->vdev_path;
8054 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8055 &value, SE_SLEEP) != 0)
8056 goto done;
8057 }
8058 }
8059
8060 if (hist_nvl != NULL) {
8061 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8062 }
8063
8064 if (sysevent_attach_attributes(ev, attr) != 0)
8065 goto done;
8066 attr = NULL;
8067
8068 done:
8069 if (attr)
8070 sysevent_free_attr(attr);
8071
8072 #endif
8073 return (ev);
8074 }
8075
8076 void
8077 spa_event_post(sysevent_t *ev)
8078 {
8079 #ifdef _KERNEL
8080 sysevent_id_t eid;
8081
8082 (void) log_sysevent(ev, SE_SLEEP, &eid);
8083 sysevent_free(ev);
8084 #endif
8085 }
8086
8087 void
8088 spa_event_discard(sysevent_t *ev)
8089 {
8090 #ifdef _KERNEL
8091 sysevent_free(ev);
8092 #endif
8093 }
8094
8095 /*
8096 * Post a sysevent corresponding to the given event. The 'name' must be one of
8097 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8098 * filled in from the spa and (optionally) the vdev and history nvl. This
8099 * doesn't do anything in the userland libzpool, as we don't want consumers to
8100 * misinterpret ztest or zdb as real changes.
8101 */
8102 void
8103 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8104 {
8105 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
8106 }