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