1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 */
27
28 /*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36 #include <sys/zfs_context.h>
37 #include <sys/fm/fs/zfs.h>
38 #include <sys/spa_impl.h>
39 #include <sys/zio.h>
40 #include <sys/zio_checksum.h>
41 #include <sys/dmu.h>
42 #include <sys/dmu_tx.h>
43 #include <sys/zap.h>
44 #include <sys/zil.h>
45 #include <sys/ddt.h>
46 #include <sys/vdev_impl.h>
47 #include <sys/metaslab.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/uberblock_impl.h>
50 #include <sys/txg.h>
51 #include <sys/avl.h>
52 #include <sys/dmu_traverse.h>
53 #include <sys/dmu_objset.h>
54 #include <sys/unique.h>
55 #include <sys/dsl_pool.h>
56 #include <sys/dsl_dataset.h>
57 #include <sys/dsl_dir.h>
58 #include <sys/dsl_prop.h>
59 #include <sys/dsl_synctask.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/arc.h>
62 #include <sys/callb.h>
63 #include <sys/systeminfo.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/zfeature.h>
68 #include <sys/dsl_destroy.h>
69
70 #ifdef _KERNEL
71 #include <sys/bootprops.h>
72 #include <sys/callb.h>
73 #include <sys/cpupart.h>
74 #include <sys/pool.h>
75 #include <sys/sysdc.h>
76 #include <sys/zone.h>
77 #endif /* _KERNEL */
78
79 #include "zfs_prop.h"
80 #include "zfs_comutil.h"
81
82 /*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86 static int zfs_ccw_retry_interval = 300;
87
88 typedef enum zti_modes {
89 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
90 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
91 ZTI_MODE_NULL, /* don't create a taskq */
92 ZTI_NMODES
93 } zti_modes_t;
94
95 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
96 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
97 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
98
99 #define ZTI_N(n) ZTI_P(n, 1)
100 #define ZTI_ONE ZTI_N(1)
101
102 typedef struct zio_taskq_info {
103 zti_modes_t zti_mode;
104 uint_t zti_value;
105 uint_t zti_count;
106 } zio_taskq_info_t;
107
108 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109 "issue", "issue_high", "intr", "intr_high"
110 };
111
112 /*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
132 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
133 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
134 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
135 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
137 };
138
139 static void spa_sync_version(void *arg, dmu_tx_t *tx);
140 static void spa_sync_props(void *arg, dmu_tx_t *tx);
141 static boolean_t spa_has_active_shared_spare(spa_t *spa);
142 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144 char **ereport);
145 static void spa_vdev_resilver_done(spa_t *spa);
146
147 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
148 id_t zio_taskq_psrset_bind = PS_NONE;
149 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 uint_t zio_taskq_basedc = 80; /* base duty cycle */
151
152 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 extern int zfs_sync_pass_deferred_free;
154
155 /*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159 #define TRYIMPORT_NAME "$import"
160
161 /*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167 /*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170 static void
171 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172 uint64_t intval, zprop_source_t src)
173 {
174 const char *propname = zpool_prop_to_name(prop);
175 nvlist_t *propval;
176
177 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180 if (strval != NULL)
181 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182 else
183 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186 nvlist_free(propval);
187 }
188
189 /*
190 * Get property values from the spa configuration.
191 */
192 static void
193 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194 {
195 vdev_t *rvd = spa->spa_root_vdev;
196 dsl_pool_t *pool = spa->spa_dsl_pool;
197 uint64_t size;
198 uint64_t alloc;
199 uint64_t space;
200 uint64_t cap, version;
201 zprop_source_t src = ZPROP_SRC_NONE;
202 spa_config_dirent_t *dp;
203
204 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206 if (rvd != NULL) {
207 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208 size = metaslab_class_get_space(spa_normal_class(spa));
209 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 size - alloc, src);
214
215 space = 0;
216 for (int c = 0; c < rvd->vdev_children; c++) {
217 vdev_t *tvd = rvd->vdev_child[c];
218 space += tvd->vdev_max_asize - tvd->vdev_asize;
219 }
220 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221 src);
222
223 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 (spa_mode(spa) == FREAD), src);
225
226 cap = (size == 0) ? 0 : (alloc * 100 / size);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228
229 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 ddt_get_pool_dedup_ratio(spa), src);
231
232 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 rvd->vdev_state, src);
234
235 version = spa_version(spa);
236 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 src = ZPROP_SRC_DEFAULT;
238 else
239 src = ZPROP_SRC_LOCAL;
240 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 }
242
243 if (pool != NULL) {
244 dsl_dir_t *freedir = pool->dp_free_dir;
245
246 /*
247 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248 * when opening pools before this version freedir will be NULL.
249 */
250 if (freedir != NULL) {
251 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252 freedir->dd_phys->dd_used_bytes, src);
253 } else {
254 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 NULL, 0, src);
256 }
257 }
258
259 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260
261 if (spa->spa_comment != NULL) {
262 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 0, ZPROP_SRC_LOCAL);
264 }
265
266 if (spa->spa_root != NULL)
267 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268 0, ZPROP_SRC_LOCAL);
269
270 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271 if (dp->scd_path == NULL) {
272 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 "none", 0, ZPROP_SRC_LOCAL);
274 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276 dp->scd_path, 0, ZPROP_SRC_LOCAL);
277 }
278 }
279 }
280
281 /*
282 * Get zpool property values.
283 */
284 int
285 spa_prop_get(spa_t *spa, nvlist_t **nvp)
286 {
287 objset_t *mos = spa->spa_meta_objset;
288 zap_cursor_t zc;
289 zap_attribute_t za;
290 int err;
291
292 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293
294 mutex_enter(&spa->spa_props_lock);
295
296 /*
297 * Get properties from the spa config.
298 */
299 spa_prop_get_config(spa, nvp);
300
301 /* If no pool property object, no more prop to get. */
302 if (mos == NULL || spa->spa_pool_props_object == 0) {
303 mutex_exit(&spa->spa_props_lock);
304 return (0);
305 }
306
307 /*
308 * Get properties from the MOS pool property object.
309 */
310 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311 (err = zap_cursor_retrieve(&zc, &za)) == 0;
312 zap_cursor_advance(&zc)) {
313 uint64_t intval = 0;
314 char *strval = NULL;
315 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 zpool_prop_t prop;
317
318 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 continue;
320
321 switch (za.za_integer_length) {
322 case 8:
323 /* integer property */
324 if (za.za_first_integer !=
325 zpool_prop_default_numeric(prop))
326 src = ZPROP_SRC_LOCAL;
327
328 if (prop == ZPOOL_PROP_BOOTFS) {
329 dsl_pool_t *dp;
330 dsl_dataset_t *ds = NULL;
331
332 dp = spa_get_dsl(spa);
333 dsl_pool_config_enter(dp, FTAG);
334 if (err = dsl_dataset_hold_obj(dp,
335 za.za_first_integer, FTAG, &ds)) {
336 dsl_pool_config_exit(dp, FTAG);
337 break;
338 }
339
340 strval = kmem_alloc(
341 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 KM_SLEEP);
343 dsl_dataset_name(ds, strval);
344 dsl_dataset_rele(ds, FTAG);
345 dsl_pool_config_exit(dp, FTAG);
346 } else {
347 strval = NULL;
348 intval = za.za_first_integer;
349 }
350
351 spa_prop_add_list(*nvp, prop, strval, intval, src);
352
353 if (strval != NULL)
354 kmem_free(strval,
355 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356
357 break;
358
359 case 1:
360 /* string property */
361 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362 err = zap_lookup(mos, spa->spa_pool_props_object,
363 za.za_name, 1, za.za_num_integers, strval);
364 if (err) {
365 kmem_free(strval, za.za_num_integers);
366 break;
367 }
368 spa_prop_add_list(*nvp, prop, strval, 0, src);
369 kmem_free(strval, za.za_num_integers);
370 break;
371
372 default:
373 break;
374 }
375 }
376 zap_cursor_fini(&zc);
377 mutex_exit(&spa->spa_props_lock);
378 out:
379 if (err && err != ENOENT) {
380 nvlist_free(*nvp);
381 *nvp = NULL;
382 return (err);
383 }
384
385 return (0);
386 }
387
388 /*
389 * Validate the given pool properties nvlist and modify the list
390 * for the property values to be set.
391 */
392 static int
393 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 {
395 nvpair_t *elem;
396 int error = 0, reset_bootfs = 0;
397 uint64_t objnum = 0;
398 boolean_t has_feature = B_FALSE;
399
400 elem = NULL;
401 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 uint64_t intval;
403 char *strval, *slash, *check, *fname;
404 const char *propname = nvpair_name(elem);
405 zpool_prop_t prop = zpool_name_to_prop(propname);
406
407 switch (prop) {
408 case ZPROP_INVAL:
409 if (!zpool_prop_feature(propname)) {
410 error = SET_ERROR(EINVAL);
411 break;
412 }
413
414 /*
415 * Sanitize the input.
416 */
417 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418 error = SET_ERROR(EINVAL);
419 break;
420 }
421
422 if (nvpair_value_uint64(elem, &intval) != 0) {
423 error = SET_ERROR(EINVAL);
424 break;
425 }
426
427 if (intval != 0) {
428 error = SET_ERROR(EINVAL);
429 break;
430 }
431
432 fname = strchr(propname, '@') + 1;
433 if (zfeature_lookup_name(fname, NULL) != 0) {
434 error = SET_ERROR(EINVAL);
435 break;
436 }
437
438 has_feature = B_TRUE;
439 break;
440
441 case ZPOOL_PROP_VERSION:
442 error = nvpair_value_uint64(elem, &intval);
443 if (!error &&
444 (intval < spa_version(spa) ||
445 intval > SPA_VERSION_BEFORE_FEATURES ||
446 has_feature))
447 error = SET_ERROR(EINVAL);
448 break;
449
450 case ZPOOL_PROP_DELEGATION:
451 case ZPOOL_PROP_AUTOREPLACE:
452 case ZPOOL_PROP_LISTSNAPS:
453 case ZPOOL_PROP_AUTOEXPAND:
454 error = nvpair_value_uint64(elem, &intval);
455 if (!error && intval > 1)
456 error = SET_ERROR(EINVAL);
457 break;
458
459 case ZPOOL_PROP_BOOTFS:
460 /*
461 * If the pool version is less than SPA_VERSION_BOOTFS,
462 * or the pool is still being created (version == 0),
463 * the bootfs property cannot be set.
464 */
465 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466 error = SET_ERROR(ENOTSUP);
467 break;
468 }
469
470 /*
471 * Make sure the vdev config is bootable
472 */
473 if (!vdev_is_bootable(spa->spa_root_vdev)) {
474 error = SET_ERROR(ENOTSUP);
475 break;
476 }
477
478 reset_bootfs = 1;
479
480 error = nvpair_value_string(elem, &strval);
481
482 if (!error) {
483 objset_t *os;
484 uint64_t compress;
485
486 if (strval == NULL || strval[0] == '\0') {
487 objnum = zpool_prop_default_numeric(
488 ZPOOL_PROP_BOOTFS);
489 break;
490 }
491
492 if (error = dmu_objset_hold(strval, FTAG, &os))
493 break;
494
495 /* Must be ZPL and not gzip compressed. */
496
497 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 error = SET_ERROR(ENOTSUP);
499 } else if ((error =
500 dsl_prop_get_int_ds(dmu_objset_ds(os),
501 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502 &compress)) == 0 &&
503 !BOOTFS_COMPRESS_VALID(compress)) {
504 error = SET_ERROR(ENOTSUP);
505 } else {
506 objnum = dmu_objset_id(os);
507 }
508 dmu_objset_rele(os, FTAG);
509 }
510 break;
511
512 case ZPOOL_PROP_FAILUREMODE:
513 error = nvpair_value_uint64(elem, &intval);
514 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515 intval > ZIO_FAILURE_MODE_PANIC))
516 error = SET_ERROR(EINVAL);
517
518 /*
519 * This is a special case which only occurs when
520 * the pool has completely failed. This allows
521 * the user to change the in-core failmode property
522 * without syncing it out to disk (I/Os might
523 * currently be blocked). We do this by returning
524 * EIO to the caller (spa_prop_set) to trick it
525 * into thinking we encountered a property validation
526 * error.
527 */
528 if (!error && spa_suspended(spa)) {
529 spa->spa_failmode = intval;
530 error = SET_ERROR(EIO);
531 }
532 break;
533
534 case ZPOOL_PROP_CACHEFILE:
535 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 break;
537
538 if (strval[0] == '\0')
539 break;
540
541 if (strcmp(strval, "none") == 0)
542 break;
543
544 if (strval[0] != '/') {
545 error = SET_ERROR(EINVAL);
546 break;
547 }
548
549 slash = strrchr(strval, '/');
550 ASSERT(slash != NULL);
551
552 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553 strcmp(slash, "/..") == 0)
554 error = SET_ERROR(EINVAL);
555 break;
556
557 case ZPOOL_PROP_COMMENT:
558 if ((error = nvpair_value_string(elem, &strval)) != 0)
559 break;
560 for (check = strval; *check != '\0'; check++) {
561 /*
562 * The kernel doesn't have an easy isprint()
563 * check. For this kernel check, we merely
564 * check ASCII apart from DEL. Fix this if
565 * there is an easy-to-use kernel isprint().
566 */
567 if (*check >= 0x7f) {
568 error = SET_ERROR(EINVAL);
569 break;
570 }
571 check++;
572 }
573 if (strlen(strval) > ZPROP_MAX_COMMENT)
574 error = E2BIG;
575 break;
576
577 case ZPOOL_PROP_DEDUPDITTO:
578 if (spa_version(spa) < SPA_VERSION_DEDUP)
579 error = SET_ERROR(ENOTSUP);
580 else
581 error = nvpair_value_uint64(elem, &intval);
582 if (error == 0 &&
583 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584 error = SET_ERROR(EINVAL);
585 break;
586 }
587
588 if (error)
589 break;
590 }
591
592 if (!error && reset_bootfs) {
593 error = nvlist_remove(props,
594 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595
596 if (!error) {
597 error = nvlist_add_uint64(props,
598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599 }
600 }
601
602 return (error);
603 }
604
605 void
606 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607 {
608 char *cachefile;
609 spa_config_dirent_t *dp;
610
611 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612 &cachefile) != 0)
613 return;
614
615 dp = kmem_alloc(sizeof (spa_config_dirent_t),
616 KM_SLEEP);
617
618 if (cachefile[0] == '\0')
619 dp->scd_path = spa_strdup(spa_config_path);
620 else if (strcmp(cachefile, "none") == 0)
621 dp->scd_path = NULL;
622 else
623 dp->scd_path = spa_strdup(cachefile);
624
625 list_insert_head(&spa->spa_config_list, dp);
626 if (need_sync)
627 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628 }
629
630 int
631 spa_prop_set(spa_t *spa, nvlist_t *nvp)
632 {
633 int error;
634 nvpair_t *elem = NULL;
635 boolean_t need_sync = B_FALSE;
636
637 if ((error = spa_prop_validate(spa, nvp)) != 0)
638 return (error);
639
640 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642
643 if (prop == ZPOOL_PROP_CACHEFILE ||
644 prop == ZPOOL_PROP_ALTROOT ||
645 prop == ZPOOL_PROP_READONLY)
646 continue;
647
648 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649 uint64_t ver;
650
651 if (prop == ZPOOL_PROP_VERSION) {
652 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653 } else {
654 ASSERT(zpool_prop_feature(nvpair_name(elem)));
655 ver = SPA_VERSION_FEATURES;
656 need_sync = B_TRUE;
657 }
658
659 /* Save time if the version is already set. */
660 if (ver == spa_version(spa))
661 continue;
662
663 /*
664 * In addition to the pool directory object, we might
665 * create the pool properties object, the features for
666 * read object, the features for write object, or the
667 * feature descriptions object.
668 */
669 error = dsl_sync_task(spa->spa_name, NULL,
670 spa_sync_version, &ver, 6);
671 if (error)
672 return (error);
673 continue;
674 }
675
676 need_sync = B_TRUE;
677 break;
678 }
679
680 if (need_sync) {
681 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682 nvp, 6));
683 }
684
685 return (0);
686 }
687
688 /*
689 * If the bootfs property value is dsobj, clear it.
690 */
691 void
692 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693 {
694 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695 VERIFY(zap_remove(spa->spa_meta_objset,
696 spa->spa_pool_props_object,
697 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 spa->spa_bootfs = 0;
699 }
700 }
701
702 /*ARGSUSED*/
703 static int
704 spa_change_guid_check(void *arg, dmu_tx_t *tx)
705 {
706 uint64_t *newguid = arg;
707 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708 vdev_t *rvd = spa->spa_root_vdev;
709 uint64_t vdev_state;
710
711 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712 vdev_state = rvd->vdev_state;
713 spa_config_exit(spa, SCL_STATE, FTAG);
714
715 if (vdev_state != VDEV_STATE_HEALTHY)
716 return (SET_ERROR(ENXIO));
717
718 ASSERT3U(spa_guid(spa), !=, *newguid);
719
720 return (0);
721 }
722
723 static void
724 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725 {
726 uint64_t *newguid = arg;
727 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728 uint64_t oldguid;
729 vdev_t *rvd = spa->spa_root_vdev;
730
731 oldguid = spa_guid(spa);
732
733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 rvd->vdev_guid = *newguid;
735 rvd->vdev_guid_sum += (*newguid - oldguid);
736 vdev_config_dirty(rvd);
737 spa_config_exit(spa, SCL_STATE, FTAG);
738
739 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740 oldguid, *newguid);
741 }
742
743 /*
744 * Change the GUID for the pool. This is done so that we can later
745 * re-import a pool built from a clone of our own vdevs. We will modify
746 * the root vdev's guid, our own pool guid, and then mark all of our
747 * vdevs dirty. Note that we must make sure that all our vdevs are
748 * online when we do this, or else any vdevs that weren't present
749 * would be orphaned from our pool. We are also going to issue a
750 * sysevent to update any watchers.
751 */
752 int
753 spa_change_guid(spa_t *spa)
754 {
755 int error;
756 uint64_t guid;
757
758 mutex_enter(&spa->spa_vdev_top_lock);
759 mutex_enter(&spa_namespace_lock);
760 guid = spa_generate_guid(NULL);
761
762 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763 spa_change_guid_sync, &guid, 5);
764
765 if (error == 0) {
766 spa_config_sync(spa, B_FALSE, B_TRUE);
767 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768 }
769
770 mutex_exit(&spa_namespace_lock);
771 mutex_exit(&spa->spa_vdev_top_lock);
772
773 return (error);
774 }
775
776 /*
777 * ==========================================================================
778 * SPA state manipulation (open/create/destroy/import/export)
779 * ==========================================================================
780 */
781
782 static int
783 spa_error_entry_compare(const void *a, const void *b)
784 {
785 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 int ret;
788
789 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 sizeof (zbookmark_t));
791
792 if (ret < 0)
793 return (-1);
794 else if (ret > 0)
795 return (1);
796 else
797 return (0);
798 }
799
800 /*
801 * Utility function which retrieves copies of the current logs and
802 * re-initializes them in the process.
803 */
804 void
805 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 {
807 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808
809 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811
812 avl_create(&spa->spa_errlist_scrub,
813 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 offsetof(spa_error_entry_t, se_avl));
815 avl_create(&spa->spa_errlist_last,
816 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 offsetof(spa_error_entry_t, se_avl));
818 }
819
820 static void
821 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 {
823 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 enum zti_modes mode = ztip->zti_mode;
825 uint_t value = ztip->zti_value;
826 uint_t count = ztip->zti_count;
827 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 char name[32];
829 uint_t flags = 0;
830 boolean_t batch = B_FALSE;
831
832 if (mode == ZTI_MODE_NULL) {
833 tqs->stqs_count = 0;
834 tqs->stqs_taskq = NULL;
835 return;
836 }
837
838 ASSERT3U(count, >, 0);
839
840 tqs->stqs_count = count;
841 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842
843 switch (mode) {
844 case ZTI_MODE_FIXED:
845 ASSERT3U(value, >=, 1);
846 value = MAX(value, 1);
847 break;
848
849 case ZTI_MODE_BATCH:
850 batch = B_TRUE;
851 flags |= TASKQ_THREADS_CPU_PCT;
852 value = zio_taskq_batch_pct;
853 break;
854
855 default:
856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 "spa_activate()",
858 zio_type_name[t], zio_taskq_types[q], mode, value);
859 break;
860 }
861
862 for (uint_t i = 0; i < count; i++) {
863 taskq_t *tq;
864
865 if (count > 1) {
866 (void) snprintf(name, sizeof (name), "%s_%s_%u",
867 zio_type_name[t], zio_taskq_types[q], i);
868 } else {
869 (void) snprintf(name, sizeof (name), "%s_%s",
870 zio_type_name[t], zio_taskq_types[q]);
871 }
872
873 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874 if (batch)
875 flags |= TASKQ_DC_BATCH;
876
877 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878 spa->spa_proc, zio_taskq_basedc, flags);
879 } else {
880 pri_t pri = maxclsyspri;
881 /*
882 * The write issue taskq can be extremely CPU
883 * intensive. Run it at slightly lower priority
884 * than the other taskqs.
885 */
886 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887 pri--;
888
889 tq = taskq_create_proc(name, value, pri, 50,
890 INT_MAX, spa->spa_proc, flags);
891 }
892
893 tqs->stqs_taskq[i] = tq;
894 }
895 }
896
897 static void
898 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899 {
900 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901
902 if (tqs->stqs_taskq == NULL) {
903 ASSERT0(tqs->stqs_count);
904 return;
905 }
906
907 for (uint_t i = 0; i < tqs->stqs_count; i++) {
908 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909 taskq_destroy(tqs->stqs_taskq[i]);
910 }
911
912 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913 tqs->stqs_taskq = NULL;
914 }
915
916 /*
917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 * on the taskq itself. In that case we choose which taskq at random by using
920 * the low bits of gethrtime().
921 */
922 void
923 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925 {
926 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927 taskq_t *tq;
928
929 ASSERT3P(tqs->stqs_taskq, !=, NULL);
930 ASSERT3U(tqs->stqs_count, !=, 0);
931
932 if (tqs->stqs_count == 1) {
933 tq = tqs->stqs_taskq[0];
934 } else {
935 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936 }
937
938 taskq_dispatch_ent(tq, func, arg, flags, ent);
939 }
940
941 static void
942 spa_create_zio_taskqs(spa_t *spa)
943 {
944 for (int t = 0; t < ZIO_TYPES; t++) {
945 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946 spa_taskqs_init(spa, t, q);
947 }
948 }
949 }
950
951 #ifdef _KERNEL
952 static void
953 spa_thread(void *arg)
954 {
955 callb_cpr_t cprinfo;
956
957 spa_t *spa = arg;
958 user_t *pu = PTOU(curproc);
959
960 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961 spa->spa_name);
962
963 ASSERT(curproc != &p0);
964 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965 "zpool-%s", spa->spa_name);
966 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967
968 /* bind this thread to the requested psrset */
969 if (zio_taskq_psrset_bind != PS_NONE) {
970 pool_lock();
971 mutex_enter(&cpu_lock);
972 mutex_enter(&pidlock);
973 mutex_enter(&curproc->p_lock);
974
975 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976 0, NULL, NULL) == 0) {
977 curthread->t_bind_pset = zio_taskq_psrset_bind;
978 } else {
979 cmn_err(CE_WARN,
980 "Couldn't bind process for zfs pool \"%s\" to "
981 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982 }
983
984 mutex_exit(&curproc->p_lock);
985 mutex_exit(&pidlock);
986 mutex_exit(&cpu_lock);
987 pool_unlock();
988 }
989
990 if (zio_taskq_sysdc) {
991 sysdc_thread_enter(curthread, 100, 0);
992 }
993
994 spa->spa_proc = curproc;
995 spa->spa_did = curthread->t_did;
996
997 spa_create_zio_taskqs(spa);
998
999 mutex_enter(&spa->spa_proc_lock);
1000 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001
1002 spa->spa_proc_state = SPA_PROC_ACTIVE;
1003 cv_broadcast(&spa->spa_proc_cv);
1004
1005 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009
1010 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011 spa->spa_proc_state = SPA_PROC_GONE;
1012 spa->spa_proc = &p0;
1013 cv_broadcast(&spa->spa_proc_cv);
1014 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1015
1016 mutex_enter(&curproc->p_lock);
1017 lwp_exit();
1018 }
1019 #endif
1020
1021 /*
1022 * Activate an uninitialized pool.
1023 */
1024 static void
1025 spa_activate(spa_t *spa, int mode)
1026 {
1027 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028
1029 spa->spa_state = POOL_STATE_ACTIVE;
1030 spa->spa_mode = mode;
1031
1032 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034
1035 /* Try to create a covering process */
1036 mutex_enter(&spa->spa_proc_lock);
1037 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038 ASSERT(spa->spa_proc == &p0);
1039 spa->spa_did = 0;
1040
1041 /* Only create a process if we're going to be around a while. */
1042 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044 NULL, 0) == 0) {
1045 spa->spa_proc_state = SPA_PROC_CREATED;
1046 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047 cv_wait(&spa->spa_proc_cv,
1048 &spa->spa_proc_lock);
1049 }
1050 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051 ASSERT(spa->spa_proc != &p0);
1052 ASSERT(spa->spa_did != 0);
1053 } else {
1054 #ifdef _KERNEL
1055 cmn_err(CE_WARN,
1056 "Couldn't create process for zfs pool \"%s\"\n",
1057 spa->spa_name);
1058 #endif
1059 }
1060 }
1061 mutex_exit(&spa->spa_proc_lock);
1062
1063 /* If we didn't create a process, we need to create our taskqs. */
1064 if (spa->spa_proc == &p0) {
1065 spa_create_zio_taskqs(spa);
1066 }
1067
1068 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069 offsetof(vdev_t, vdev_config_dirty_node));
1070 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071 offsetof(vdev_t, vdev_state_dirty_node));
1072
1073 txg_list_create(&spa->spa_vdev_txg_list,
1074 offsetof(struct vdev, vdev_txg_node));
1075
1076 avl_create(&spa->spa_errlist_scrub,
1077 spa_error_entry_compare, sizeof (spa_error_entry_t),
1078 offsetof(spa_error_entry_t, se_avl));
1079 avl_create(&spa->spa_errlist_last,
1080 spa_error_entry_compare, sizeof (spa_error_entry_t),
1081 offsetof(spa_error_entry_t, se_avl));
1082 }
1083
1084 /*
1085 * Opposite of spa_activate().
1086 */
1087 static void
1088 spa_deactivate(spa_t *spa)
1089 {
1090 ASSERT(spa->spa_sync_on == B_FALSE);
1091 ASSERT(spa->spa_dsl_pool == NULL);
1092 ASSERT(spa->spa_root_vdev == NULL);
1093 ASSERT(spa->spa_async_zio_root == NULL);
1094 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095
1096 txg_list_destroy(&spa->spa_vdev_txg_list);
1097
1098 list_destroy(&spa->spa_config_dirty_list);
1099 list_destroy(&spa->spa_state_dirty_list);
1100
1101 for (int t = 0; t < ZIO_TYPES; t++) {
1102 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103 spa_taskqs_fini(spa, t, q);
1104 }
1105 }
1106
1107 metaslab_class_destroy(spa->spa_normal_class);
1108 spa->spa_normal_class = NULL;
1109
1110 metaslab_class_destroy(spa->spa_log_class);
1111 spa->spa_log_class = NULL;
1112
1113 /*
1114 * If this was part of an import or the open otherwise failed, we may
1115 * still have errors left in the queues. Empty them just in case.
1116 */
1117 spa_errlog_drain(spa);
1118
1119 avl_destroy(&spa->spa_errlist_scrub);
1120 avl_destroy(&spa->spa_errlist_last);
1121
1122 spa->spa_state = POOL_STATE_UNINITIALIZED;
1123
1124 mutex_enter(&spa->spa_proc_lock);
1125 if (spa->spa_proc_state != SPA_PROC_NONE) {
1126 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128 cv_broadcast(&spa->spa_proc_cv);
1129 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130 ASSERT(spa->spa_proc != &p0);
1131 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132 }
1133 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134 spa->spa_proc_state = SPA_PROC_NONE;
1135 }
1136 ASSERT(spa->spa_proc == &p0);
1137 mutex_exit(&spa->spa_proc_lock);
1138
1139 /*
1140 * We want to make sure spa_thread() has actually exited the ZFS
1141 * module, so that the module can't be unloaded out from underneath
1142 * it.
1143 */
1144 if (spa->spa_did != 0) {
1145 thread_join(spa->spa_did);
1146 spa->spa_did = 0;
1147 }
1148 }
1149
1150 /*
1151 * Verify a pool configuration, and construct the vdev tree appropriately. This
1152 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 * in the CLOSED state. This will prep the pool before open/creation/import.
1154 * All vdev validation is done by the vdev_alloc() routine.
1155 */
1156 static int
1157 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158 uint_t id, int atype)
1159 {
1160 nvlist_t **child;
1161 uint_t children;
1162 int error;
1163
1164 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165 return (error);
1166
1167 if ((*vdp)->vdev_ops->vdev_op_leaf)
1168 return (0);
1169
1170 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171 &child, &children);
1172
1173 if (error == ENOENT)
1174 return (0);
1175
1176 if (error) {
1177 vdev_free(*vdp);
1178 *vdp = NULL;
1179 return (SET_ERROR(EINVAL));
1180 }
1181
1182 for (int c = 0; c < children; c++) {
1183 vdev_t *vd;
1184 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185 atype)) != 0) {
1186 vdev_free(*vdp);
1187 *vdp = NULL;
1188 return (error);
1189 }
1190 }
1191
1192 ASSERT(*vdp != NULL);
1193
1194 return (0);
1195 }
1196
1197 /*
1198 * Opposite of spa_load().
1199 */
1200 static void
1201 spa_unload(spa_t *spa)
1202 {
1203 int i;
1204
1205 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206
1207 /*
1208 * Stop async tasks.
1209 */
1210 spa_async_suspend(spa);
1211
1212 /*
1213 * Stop syncing.
1214 */
1215 if (spa->spa_sync_on) {
1216 txg_sync_stop(spa->spa_dsl_pool);
1217 spa->spa_sync_on = B_FALSE;
1218 }
1219
1220 /*
1221 * Wait for any outstanding async I/O to complete.
1222 */
1223 if (spa->spa_async_zio_root != NULL) {
1224 (void) zio_wait(spa->spa_async_zio_root);
1225 spa->spa_async_zio_root = NULL;
1226 }
1227
1228 bpobj_close(&spa->spa_deferred_bpobj);
1229
1230 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231
1232 /*
1233 * Close all vdevs.
1234 */
1235 if (spa->spa_root_vdev)
1236 vdev_free(spa->spa_root_vdev);
1237 ASSERT(spa->spa_root_vdev == NULL);
1238
1239 /*
1240 * Close the dsl pool.
1241 */
1242 if (spa->spa_dsl_pool) {
1243 dsl_pool_close(spa->spa_dsl_pool);
1244 spa->spa_dsl_pool = NULL;
1245 spa->spa_meta_objset = NULL;
1246 }
1247
1248 ddt_unload(spa);
1249
1250
1251 /*
1252 * Drop and purge level 2 cache
1253 */
1254 spa_l2cache_drop(spa);
1255
1256 for (i = 0; i < spa->spa_spares.sav_count; i++)
1257 vdev_free(spa->spa_spares.sav_vdevs[i]);
1258 if (spa->spa_spares.sav_vdevs) {
1259 kmem_free(spa->spa_spares.sav_vdevs,
1260 spa->spa_spares.sav_count * sizeof (void *));
1261 spa->spa_spares.sav_vdevs = NULL;
1262 }
1263 if (spa->spa_spares.sav_config) {
1264 nvlist_free(spa->spa_spares.sav_config);
1265 spa->spa_spares.sav_config = NULL;
1266 }
1267 spa->spa_spares.sav_count = 0;
1268
1269 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272 }
1273 if (spa->spa_l2cache.sav_vdevs) {
1274 kmem_free(spa->spa_l2cache.sav_vdevs,
1275 spa->spa_l2cache.sav_count * sizeof (void *));
1276 spa->spa_l2cache.sav_vdevs = NULL;
1277 }
1278 if (spa->spa_l2cache.sav_config) {
1279 nvlist_free(spa->spa_l2cache.sav_config);
1280 spa->spa_l2cache.sav_config = NULL;
1281 }
1282 spa->spa_l2cache.sav_count = 0;
1283
1284 spa->spa_async_suspended = 0;
1285
1286 if (spa->spa_comment != NULL) {
1287 spa_strfree(spa->spa_comment);
1288 spa->spa_comment = NULL;
1289 }
1290
1291 spa_config_exit(spa, SCL_ALL, FTAG);
1292 }
1293
1294 /*
1295 * Load (or re-load) the current list of vdevs describing the active spares for
1296 * this pool. When this is called, we have some form of basic information in
1297 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1298 * then re-generate a more complete list including status information.
1299 */
1300 static void
1301 spa_load_spares(spa_t *spa)
1302 {
1303 nvlist_t **spares;
1304 uint_t nspares;
1305 int i;
1306 vdev_t *vd, *tvd;
1307
1308 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309
1310 /*
1311 * First, close and free any existing spare vdevs.
1312 */
1313 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314 vd = spa->spa_spares.sav_vdevs[i];
1315
1316 /* Undo the call to spa_activate() below */
1317 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318 B_FALSE)) != NULL && tvd->vdev_isspare)
1319 spa_spare_remove(tvd);
1320 vdev_close(vd);
1321 vdev_free(vd);
1322 }
1323
1324 if (spa->spa_spares.sav_vdevs)
1325 kmem_free(spa->spa_spares.sav_vdevs,
1326 spa->spa_spares.sav_count * sizeof (void *));
1327
1328 if (spa->spa_spares.sav_config == NULL)
1329 nspares = 0;
1330 else
1331 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333
1334 spa->spa_spares.sav_count = (int)nspares;
1335 spa->spa_spares.sav_vdevs = NULL;
1336
1337 if (nspares == 0)
1338 return;
1339
1340 /*
1341 * Construct the array of vdevs, opening them to get status in the
1342 * process. For each spare, there is potentially two different vdev_t
1343 * structures associated with it: one in the list of spares (used only
1344 * for basic validation purposes) and one in the active vdev
1345 * configuration (if it's spared in). During this phase we open and
1346 * validate each vdev on the spare list. If the vdev also exists in the
1347 * active configuration, then we also mark this vdev as an active spare.
1348 */
1349 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350 KM_SLEEP);
1351 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353 VDEV_ALLOC_SPARE) == 0);
1354 ASSERT(vd != NULL);
1355
1356 spa->spa_spares.sav_vdevs[i] = vd;
1357
1358 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359 B_FALSE)) != NULL) {
1360 if (!tvd->vdev_isspare)
1361 spa_spare_add(tvd);
1362
1363 /*
1364 * We only mark the spare active if we were successfully
1365 * able to load the vdev. Otherwise, importing a pool
1366 * with a bad active spare would result in strange
1367 * behavior, because multiple pool would think the spare
1368 * is actively in use.
1369 *
1370 * There is a vulnerability here to an equally bizarre
1371 * circumstance, where a dead active spare is later
1372 * brought back to life (onlined or otherwise). Given
1373 * the rarity of this scenario, and the extra complexity
1374 * it adds, we ignore the possibility.
1375 */
1376 if (!vdev_is_dead(tvd))
1377 spa_spare_activate(tvd);
1378 }
1379
1380 vd->vdev_top = vd;
1381 vd->vdev_aux = &spa->spa_spares;
1382
1383 if (vdev_open(vd) != 0)
1384 continue;
1385
1386 if (vdev_validate_aux(vd) == 0)
1387 spa_spare_add(vd);
1388 }
1389
1390 /*
1391 * Recompute the stashed list of spares, with status information
1392 * this time.
1393 */
1394 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395 DATA_TYPE_NVLIST_ARRAY) == 0);
1396
1397 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398 KM_SLEEP);
1399 for (i = 0; i < spa->spa_spares.sav_count; i++)
1400 spares[i] = vdev_config_generate(spa,
1401 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404 for (i = 0; i < spa->spa_spares.sav_count; i++)
1405 nvlist_free(spares[i]);
1406 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407 }
1408
1409 /*
1410 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 * this pool. When this is called, we have some form of basic information in
1412 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1413 * then re-generate a more complete list including status information.
1414 * Devices which are already active have their details maintained, and are
1415 * not re-opened.
1416 */
1417 static void
1418 spa_load_l2cache(spa_t *spa)
1419 {
1420 nvlist_t **l2cache;
1421 uint_t nl2cache;
1422 int i, j, oldnvdevs;
1423 uint64_t guid;
1424 vdev_t *vd, **oldvdevs, **newvdevs;
1425 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426
1427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428
1429 if (sav->sav_config != NULL) {
1430 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433 } else {
1434 nl2cache = 0;
1435 newvdevs = NULL;
1436 }
1437
1438 oldvdevs = sav->sav_vdevs;
1439 oldnvdevs = sav->sav_count;
1440 sav->sav_vdevs = NULL;
1441 sav->sav_count = 0;
1442
1443 /*
1444 * Process new nvlist of vdevs.
1445 */
1446 for (i = 0; i < nl2cache; i++) {
1447 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448 &guid) == 0);
1449
1450 newvdevs[i] = NULL;
1451 for (j = 0; j < oldnvdevs; j++) {
1452 vd = oldvdevs[j];
1453 if (vd != NULL && guid == vd->vdev_guid) {
1454 /*
1455 * Retain previous vdev for add/remove ops.
1456 */
1457 newvdevs[i] = vd;
1458 oldvdevs[j] = NULL;
1459 break;
1460 }
1461 }
1462
1463 if (newvdevs[i] == NULL) {
1464 /*
1465 * Create new vdev
1466 */
1467 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468 VDEV_ALLOC_L2CACHE) == 0);
1469 ASSERT(vd != NULL);
1470 newvdevs[i] = vd;
1471
1472 /*
1473 * Commit this vdev as an l2cache device,
1474 * even if it fails to open.
1475 */
1476 spa_l2cache_add(vd);
1477
1478 vd->vdev_top = vd;
1479 vd->vdev_aux = sav;
1480
1481 spa_l2cache_activate(vd);
1482
1483 if (vdev_open(vd) != 0)
1484 continue;
1485
1486 (void) vdev_validate_aux(vd);
1487
1488 if (!vdev_is_dead(vd)) {
1489 boolean_t do_rebuild = B_FALSE;
1490
1491 (void) nvlist_lookup_boolean_value(l2cache[i],
1492 ZPOOL_CONFIG_L2CACHE_PERSISTENT,
1493 &do_rebuild);
1494 l2arc_add_vdev(spa, vd, do_rebuild);
1495 }
1496 }
1497 }
1498
1499 /*
1500 * Purge vdevs that were dropped
1501 */
1502 for (i = 0; i < oldnvdevs; i++) {
1503 uint64_t pool;
1504
1505 vd = oldvdevs[i];
1506 if (vd != NULL) {
1507 ASSERT(vd->vdev_isl2cache);
1508
1509 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1510 pool != 0ULL && l2arc_vdev_present(vd))
1511 l2arc_remove_vdev(vd);
1512 vdev_clear_stats(vd);
1513 vdev_free(vd);
1514 }
1515 }
1516
1517 if (oldvdevs)
1518 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1519
1520 if (sav->sav_config == NULL)
1521 goto out;
1522
1523 sav->sav_vdevs = newvdevs;
1524 sav->sav_count = (int)nl2cache;
1525
1526 /*
1527 * Recompute the stashed list of l2cache devices, with status
1528 * information this time.
1529 */
1530 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1531 DATA_TYPE_NVLIST_ARRAY) == 0);
1532
1533 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1534 for (i = 0; i < sav->sav_count; i++)
1535 l2cache[i] = vdev_config_generate(spa,
1536 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1537 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1538 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1539 out:
1540 for (i = 0; i < sav->sav_count; i++)
1541 nvlist_free(l2cache[i]);
1542 if (sav->sav_count)
1543 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1544 }
1545
1546 static int
1547 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1548 {
1549 dmu_buf_t *db;
1550 char *packed = NULL;
1551 size_t nvsize = 0;
1552 int error;
1553 *value = NULL;
1554
1555 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1556 nvsize = *(uint64_t *)db->db_data;
1557 dmu_buf_rele(db, FTAG);
1558
1559 packed = kmem_alloc(nvsize, KM_SLEEP);
1560 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1561 DMU_READ_PREFETCH);
1562 if (error == 0)
1563 error = nvlist_unpack(packed, nvsize, value, 0);
1564 kmem_free(packed, nvsize);
1565
1566 return (error);
1567 }
1568
1569 /*
1570 * Checks to see if the given vdev could not be opened, in which case we post a
1571 * sysevent to notify the autoreplace code that the device has been removed.
1572 */
1573 static void
1574 spa_check_removed(vdev_t *vd)
1575 {
1576 for (int c = 0; c < vd->vdev_children; c++)
1577 spa_check_removed(vd->vdev_child[c]);
1578
1579 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1580 !vd->vdev_ishole) {
1581 zfs_post_autoreplace(vd->vdev_spa, vd);
1582 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1583 }
1584 }
1585
1586 /*
1587 * Validate the current config against the MOS config
1588 */
1589 static boolean_t
1590 spa_config_valid(spa_t *spa, nvlist_t *config)
1591 {
1592 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1593 nvlist_t *nv;
1594
1595 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1596
1597 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1598 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1599
1600 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1601
1602 /*
1603 * If we're doing a normal import, then build up any additional
1604 * diagnostic information about missing devices in this config.
1605 * We'll pass this up to the user for further processing.
1606 */
1607 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1608 nvlist_t **child, *nv;
1609 uint64_t idx = 0;
1610
1611 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1612 KM_SLEEP);
1613 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1614
1615 for (int c = 0; c < rvd->vdev_children; c++) {
1616 vdev_t *tvd = rvd->vdev_child[c];
1617 vdev_t *mtvd = mrvd->vdev_child[c];
1618
1619 if (tvd->vdev_ops == &vdev_missing_ops &&
1620 mtvd->vdev_ops != &vdev_missing_ops &&
1621 mtvd->vdev_islog)
1622 child[idx++] = vdev_config_generate(spa, mtvd,
1623 B_FALSE, 0);
1624 }
1625
1626 if (idx) {
1627 VERIFY(nvlist_add_nvlist_array(nv,
1628 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1629 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1630 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1631
1632 for (int i = 0; i < idx; i++)
1633 nvlist_free(child[i]);
1634 }
1635 nvlist_free(nv);
1636 kmem_free(child, rvd->vdev_children * sizeof (char **));
1637 }
1638
1639 /*
1640 * Compare the root vdev tree with the information we have
1641 * from the MOS config (mrvd). Check each top-level vdev
1642 * with the corresponding MOS config top-level (mtvd).
1643 */
1644 for (int c = 0; c < rvd->vdev_children; c++) {
1645 vdev_t *tvd = rvd->vdev_child[c];
1646 vdev_t *mtvd = mrvd->vdev_child[c];
1647
1648 /*
1649 * Resolve any "missing" vdevs in the current configuration.
1650 * If we find that the MOS config has more accurate information
1651 * about the top-level vdev then use that vdev instead.
1652 */
1653 if (tvd->vdev_ops == &vdev_missing_ops &&
1654 mtvd->vdev_ops != &vdev_missing_ops) {
1655
1656 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1657 continue;
1658
1659 /*
1660 * Device specific actions.
1661 */
1662 if (mtvd->vdev_islog) {
1663 spa_set_log_state(spa, SPA_LOG_CLEAR);
1664 } else {
1665 /*
1666 * XXX - once we have 'readonly' pool
1667 * support we should be able to handle
1668 * missing data devices by transitioning
1669 * the pool to readonly.
1670 */
1671 continue;
1672 }
1673
1674 /*
1675 * Swap the missing vdev with the data we were
1676 * able to obtain from the MOS config.
1677 */
1678 vdev_remove_child(rvd, tvd);
1679 vdev_remove_child(mrvd, mtvd);
1680
1681 vdev_add_child(rvd, mtvd);
1682 vdev_add_child(mrvd, tvd);
1683
1684 spa_config_exit(spa, SCL_ALL, FTAG);
1685 vdev_load(mtvd);
1686 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1687
1688 vdev_reopen(rvd);
1689 } else if (mtvd->vdev_islog) {
1690 /*
1691 * Load the slog device's state from the MOS config
1692 * since it's possible that the label does not
1693 * contain the most up-to-date information.
1694 */
1695 vdev_load_log_state(tvd, mtvd);
1696 vdev_reopen(tvd);
1697 }
1698 }
1699 vdev_free(mrvd);
1700 spa_config_exit(spa, SCL_ALL, FTAG);
1701
1702 /*
1703 * Ensure we were able to validate the config.
1704 */
1705 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1706 }
1707
1708 /*
1709 * Check for missing log devices
1710 */
1711 static boolean_t
1712 spa_check_logs(spa_t *spa)
1713 {
1714 boolean_t rv = B_FALSE;
1715
1716 switch (spa->spa_log_state) {
1717 case SPA_LOG_MISSING:
1718 /* need to recheck in case slog has been restored */
1719 case SPA_LOG_UNKNOWN:
1720 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1721 NULL, DS_FIND_CHILDREN) != 0);
1722 if (rv)
1723 spa_set_log_state(spa, SPA_LOG_MISSING);
1724 break;
1725 }
1726 return (rv);
1727 }
1728
1729 static boolean_t
1730 spa_passivate_log(spa_t *spa)
1731 {
1732 vdev_t *rvd = spa->spa_root_vdev;
1733 boolean_t slog_found = B_FALSE;
1734
1735 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1736
1737 if (!spa_has_slogs(spa))
1738 return (B_FALSE);
1739
1740 for (int c = 0; c < rvd->vdev_children; c++) {
1741 vdev_t *tvd = rvd->vdev_child[c];
1742 metaslab_group_t *mg = tvd->vdev_mg;
1743
1744 if (tvd->vdev_islog) {
1745 metaslab_group_passivate(mg);
1746 slog_found = B_TRUE;
1747 }
1748 }
1749
1750 return (slog_found);
1751 }
1752
1753 static void
1754 spa_activate_log(spa_t *spa)
1755 {
1756 vdev_t *rvd = spa->spa_root_vdev;
1757
1758 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1759
1760 for (int c = 0; c < rvd->vdev_children; c++) {
1761 vdev_t *tvd = rvd->vdev_child[c];
1762 metaslab_group_t *mg = tvd->vdev_mg;
1763
1764 if (tvd->vdev_islog)
1765 metaslab_group_activate(mg);
1766 }
1767 }
1768
1769 int
1770 spa_offline_log(spa_t *spa)
1771 {
1772 int error;
1773
1774 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1775 NULL, DS_FIND_CHILDREN);
1776 if (error == 0) {
1777 /*
1778 * We successfully offlined the log device, sync out the
1779 * current txg so that the "stubby" block can be removed
1780 * by zil_sync().
1781 */
1782 txg_wait_synced(spa->spa_dsl_pool, 0);
1783 }
1784 return (error);
1785 }
1786
1787 static void
1788 spa_aux_check_removed(spa_aux_vdev_t *sav)
1789 {
1790 for (int i = 0; i < sav->sav_count; i++)
1791 spa_check_removed(sav->sav_vdevs[i]);
1792 }
1793
1794 void
1795 spa_claim_notify(zio_t *zio)
1796 {
1797 spa_t *spa = zio->io_spa;
1798
1799 if (zio->io_error)
1800 return;
1801
1802 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1803 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1804 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1805 mutex_exit(&spa->spa_props_lock);
1806 }
1807
1808 typedef struct spa_load_error {
1809 uint64_t sle_meta_count;
1810 uint64_t sle_data_count;
1811 } spa_load_error_t;
1812
1813 static void
1814 spa_load_verify_done(zio_t *zio)
1815 {
1816 blkptr_t *bp = zio->io_bp;
1817 spa_load_error_t *sle = zio->io_private;
1818 dmu_object_type_t type = BP_GET_TYPE(bp);
1819 int error = zio->io_error;
1820
1821 if (error) {
1822 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1823 type != DMU_OT_INTENT_LOG)
1824 atomic_add_64(&sle->sle_meta_count, 1);
1825 else
1826 atomic_add_64(&sle->sle_data_count, 1);
1827 }
1828 zio_data_buf_free(zio->io_data, zio->io_size);
1829 }
1830
1831 /*ARGSUSED*/
1832 static int
1833 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1834 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1835 {
1836 if (bp != NULL) {
1837 zio_t *rio = arg;
1838 size_t size = BP_GET_PSIZE(bp);
1839 void *data = zio_data_buf_alloc(size);
1840
1841 zio_nowait(zio_read(rio, spa, bp, data, size,
1842 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1843 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1844 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1845 }
1846 return (0);
1847 }
1848
1849 static int
1850 spa_load_verify(spa_t *spa)
1851 {
1852 zio_t *rio;
1853 spa_load_error_t sle = { 0 };
1854 zpool_rewind_policy_t policy;
1855 boolean_t verify_ok = B_FALSE;
1856 int error;
1857
1858 zpool_get_rewind_policy(spa->spa_config, &policy);
1859
1860 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1861 return (0);
1862
1863 rio = zio_root(spa, NULL, &sle,
1864 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1865
1866 error = traverse_pool(spa, spa->spa_verify_min_txg,
1867 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1868
1869 (void) zio_wait(rio);
1870
1871 spa->spa_load_meta_errors = sle.sle_meta_count;
1872 spa->spa_load_data_errors = sle.sle_data_count;
1873
1874 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1875 sle.sle_data_count <= policy.zrp_maxdata) {
1876 int64_t loss = 0;
1877
1878 verify_ok = B_TRUE;
1879 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1880 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1881
1882 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1883 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1884 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1885 VERIFY(nvlist_add_int64(spa->spa_load_info,
1886 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1887 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1888 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1889 } else {
1890 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1891 }
1892
1893 if (error) {
1894 if (error != ENXIO && error != EIO)
1895 error = SET_ERROR(EIO);
1896 return (error);
1897 }
1898
1899 return (verify_ok ? 0 : EIO);
1900 }
1901
1902 /*
1903 * Find a value in the pool props object.
1904 */
1905 static void
1906 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1907 {
1908 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1909 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1910 }
1911
1912 /*
1913 * Find a value in the pool directory object.
1914 */
1915 static int
1916 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1917 {
1918 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1919 name, sizeof (uint64_t), 1, val));
1920 }
1921
1922 static int
1923 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1924 {
1925 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1926 return (err);
1927 }
1928
1929 /*
1930 * Fix up config after a partly-completed split. This is done with the
1931 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1932 * pool have that entry in their config, but only the splitting one contains
1933 * a list of all the guids of the vdevs that are being split off.
1934 *
1935 * This function determines what to do with that list: either rejoin
1936 * all the disks to the pool, or complete the splitting process. To attempt
1937 * the rejoin, each disk that is offlined is marked online again, and
1938 * we do a reopen() call. If the vdev label for every disk that was
1939 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1940 * then we call vdev_split() on each disk, and complete the split.
1941 *
1942 * Otherwise we leave the config alone, with all the vdevs in place in
1943 * the original pool.
1944 */
1945 static void
1946 spa_try_repair(spa_t *spa, nvlist_t *config)
1947 {
1948 uint_t extracted;
1949 uint64_t *glist;
1950 uint_t i, gcount;
1951 nvlist_t *nvl;
1952 vdev_t **vd;
1953 boolean_t attempt_reopen;
1954
1955 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1956 return;
1957
1958 /* check that the config is complete */
1959 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1960 &glist, &gcount) != 0)
1961 return;
1962
1963 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1964
1965 /* attempt to online all the vdevs & validate */
1966 attempt_reopen = B_TRUE;
1967 for (i = 0; i < gcount; i++) {
1968 if (glist[i] == 0) /* vdev is hole */
1969 continue;
1970
1971 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1972 if (vd[i] == NULL) {
1973 /*
1974 * Don't bother attempting to reopen the disks;
1975 * just do the split.
1976 */
1977 attempt_reopen = B_FALSE;
1978 } else {
1979 /* attempt to re-online it */
1980 vd[i]->vdev_offline = B_FALSE;
1981 }
1982 }
1983
1984 if (attempt_reopen) {
1985 vdev_reopen(spa->spa_root_vdev);
1986
1987 /* check each device to see what state it's in */
1988 for (extracted = 0, i = 0; i < gcount; i++) {
1989 if (vd[i] != NULL &&
1990 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1991 break;
1992 ++extracted;
1993 }
1994 }
1995
1996 /*
1997 * If every disk has been moved to the new pool, or if we never
1998 * even attempted to look at them, then we split them off for
1999 * good.
2000 */
2001 if (!attempt_reopen || gcount == extracted) {
2002 for (i = 0; i < gcount; i++)
2003 if (vd[i] != NULL)
2004 vdev_split(vd[i]);
2005 vdev_reopen(spa->spa_root_vdev);
2006 }
2007
2008 kmem_free(vd, gcount * sizeof (vdev_t *));
2009 }
2010
2011 static int
2012 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2013 boolean_t mosconfig)
2014 {
2015 nvlist_t *config = spa->spa_config;
2016 char *ereport = FM_EREPORT_ZFS_POOL;
2017 char *comment;
2018 int error;
2019 uint64_t pool_guid;
2020 nvlist_t *nvl;
2021
2022 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2023 return (SET_ERROR(EINVAL));
2024
2025 ASSERT(spa->spa_comment == NULL);
2026 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2027 spa->spa_comment = spa_strdup(comment);
2028
2029 /*
2030 * Versioning wasn't explicitly added to the label until later, so if
2031 * it's not present treat it as the initial version.
2032 */
2033 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2034 &spa->spa_ubsync.ub_version) != 0)
2035 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2036
2037 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2038 &spa->spa_config_txg);
2039
2040 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2041 spa_guid_exists(pool_guid, 0)) {
2042 error = SET_ERROR(EEXIST);
2043 } else {
2044 spa->spa_config_guid = pool_guid;
2045
2046 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2047 &nvl) == 0) {
2048 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2049 KM_SLEEP) == 0);
2050 }
2051
2052 nvlist_free(spa->spa_load_info);
2053 spa->spa_load_info = fnvlist_alloc();
2054
2055 gethrestime(&spa->spa_loaded_ts);
2056 error = spa_load_impl(spa, pool_guid, config, state, type,
2057 mosconfig, &ereport);
2058 }
2059
2060 spa->spa_minref = refcount_count(&spa->spa_refcount);
2061 if (error) {
2062 if (error != EEXIST) {
2063 spa->spa_loaded_ts.tv_sec = 0;
2064 spa->spa_loaded_ts.tv_nsec = 0;
2065 }
2066 if (error != EBADF) {
2067 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2068 }
2069 }
2070 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2071 spa->spa_ena = 0;
2072
2073 return (error);
2074 }
2075
2076 /*
2077 * Load an existing storage pool, using the pool's builtin spa_config as a
2078 * source of configuration information.
2079 */
2080 static int
2081 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2082 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2083 char **ereport)
2084 {
2085 int error = 0;
2086 nvlist_t *nvroot = NULL;
2087 nvlist_t *label;
2088 vdev_t *rvd;
2089 uberblock_t *ub = &spa->spa_uberblock;
2090 uint64_t children, config_cache_txg = spa->spa_config_txg;
2091 int orig_mode = spa->spa_mode;
2092 int parse;
2093 uint64_t obj;
2094 boolean_t missing_feat_write = B_FALSE;
2095
2096 /*
2097 * If this is an untrusted config, access the pool in read-only mode.
2098 * This prevents things like resilvering recently removed devices.
2099 */
2100 if (!mosconfig)
2101 spa->spa_mode = FREAD;
2102
2103 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2104
2105 spa->spa_load_state = state;
2106
2107 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2108 return (SET_ERROR(EINVAL));
2109
2110 parse = (type == SPA_IMPORT_EXISTING ?
2111 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2112
2113 /*
2114 * Create "The Godfather" zio to hold all async IOs
2115 */
2116 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2117 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2118
2119 /*
2120 * Parse the configuration into a vdev tree. We explicitly set the
2121 * value that will be returned by spa_version() since parsing the
2122 * configuration requires knowing the version number.
2123 */
2124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2125 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2126 spa_config_exit(spa, SCL_ALL, FTAG);
2127
2128 if (error != 0)
2129 return (error);
2130
2131 ASSERT(spa->spa_root_vdev == rvd);
2132
2133 if (type != SPA_IMPORT_ASSEMBLE) {
2134 ASSERT(spa_guid(spa) == pool_guid);
2135 }
2136
2137 /*
2138 * Try to open all vdevs, loading each label in the process.
2139 */
2140 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2141 error = vdev_open(rvd);
2142 spa_config_exit(spa, SCL_ALL, FTAG);
2143 if (error != 0)
2144 return (error);
2145
2146 /*
2147 * We need to validate the vdev labels against the configuration that
2148 * we have in hand, which is dependent on the setting of mosconfig. If
2149 * mosconfig is true then we're validating the vdev labels based on
2150 * that config. Otherwise, we're validating against the cached config
2151 * (zpool.cache) that was read when we loaded the zfs module, and then
2152 * later we will recursively call spa_load() and validate against
2153 * the vdev config.
2154 *
2155 * If we're assembling a new pool that's been split off from an
2156 * existing pool, the labels haven't yet been updated so we skip
2157 * validation for now.
2158 */
2159 if (type != SPA_IMPORT_ASSEMBLE) {
2160 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2161 error = vdev_validate(rvd, mosconfig);
2162 spa_config_exit(spa, SCL_ALL, FTAG);
2163
2164 if (error != 0)
2165 return (error);
2166
2167 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2168 return (SET_ERROR(ENXIO));
2169 }
2170
2171 /*
2172 * Find the best uberblock.
2173 */
2174 vdev_uberblock_load(rvd, ub, &label);
2175
2176 /*
2177 * If we weren't able to find a single valid uberblock, return failure.
2178 */
2179 if (ub->ub_txg == 0) {
2180 nvlist_free(label);
2181 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2182 }
2183
2184 /*
2185 * If the pool has an unsupported version we can't open it.
2186 */
2187 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2188 nvlist_free(label);
2189 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2190 }
2191
2192 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2193 nvlist_t *features;
2194
2195 /*
2196 * If we weren't able to find what's necessary for reading the
2197 * MOS in the label, return failure.
2198 */
2199 if (label == NULL || nvlist_lookup_nvlist(label,
2200 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2201 nvlist_free(label);
2202 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2203 ENXIO));
2204 }
2205
2206 /*
2207 * Update our in-core representation with the definitive values
2208 * from the label.
2209 */
2210 nvlist_free(spa->spa_label_features);
2211 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2212 }
2213
2214 nvlist_free(label);
2215
2216 /*
2217 * Look through entries in the label nvlist's features_for_read. If
2218 * there is a feature listed there which we don't understand then we
2219 * cannot open a pool.
2220 */
2221 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2222 nvlist_t *unsup_feat;
2223
2224 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2225 0);
2226
2227 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2228 NULL); nvp != NULL;
2229 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2230 if (!zfeature_is_supported(nvpair_name(nvp))) {
2231 VERIFY(nvlist_add_string(unsup_feat,
2232 nvpair_name(nvp), "") == 0);
2233 }
2234 }
2235
2236 if (!nvlist_empty(unsup_feat)) {
2237 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2238 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2239 nvlist_free(unsup_feat);
2240 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2241 ENOTSUP));
2242 }
2243
2244 nvlist_free(unsup_feat);
2245 }
2246
2247 /*
2248 * If the vdev guid sum doesn't match the uberblock, we have an
2249 * incomplete configuration. We first check to see if the pool
2250 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2251 * If it is, defer the vdev_guid_sum check till later so we
2252 * can handle missing vdevs.
2253 */
2254 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2255 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2256 rvd->vdev_guid_sum != ub->ub_guid_sum)
2257 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2258
2259 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2260 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2261 spa_try_repair(spa, config);
2262 spa_config_exit(spa, SCL_ALL, FTAG);
2263 nvlist_free(spa->spa_config_splitting);
2264 spa->spa_config_splitting = NULL;
2265 }
2266
2267 /*
2268 * Initialize internal SPA structures.
2269 */
2270 spa->spa_state = POOL_STATE_ACTIVE;
2271 spa->spa_ubsync = spa->spa_uberblock;
2272 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2273 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2274 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2275 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2276 spa->spa_claim_max_txg = spa->spa_first_txg;
2277 spa->spa_prev_software_version = ub->ub_software_version;
2278
2279 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2280 if (error)
2281 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2282 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2283
2284 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2285 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2286
2287 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2288 boolean_t missing_feat_read = B_FALSE;
2289 nvlist_t *unsup_feat, *enabled_feat;
2290
2291 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2292 &spa->spa_feat_for_read_obj) != 0) {
2293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2294 }
2295
2296 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2297 &spa->spa_feat_for_write_obj) != 0) {
2298 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2299 }
2300
2301 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2302 &spa->spa_feat_desc_obj) != 0) {
2303 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2304 }
2305
2306 enabled_feat = fnvlist_alloc();
2307 unsup_feat = fnvlist_alloc();
2308
2309 if (!spa_features_check(spa, B_FALSE,
2310 unsup_feat, enabled_feat))
2311 missing_feat_read = B_TRUE;
2312
2313 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2314 if (!spa_features_check(spa, B_TRUE,
2315 unsup_feat, enabled_feat)) {
2316 missing_feat_write = B_TRUE;
2317 }
2318 }
2319
2320 fnvlist_add_nvlist(spa->spa_load_info,
2321 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2322
2323 if (!nvlist_empty(unsup_feat)) {
2324 fnvlist_add_nvlist(spa->spa_load_info,
2325 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2326 }
2327
2328 fnvlist_free(enabled_feat);
2329 fnvlist_free(unsup_feat);
2330
2331 if (!missing_feat_read) {
2332 fnvlist_add_boolean(spa->spa_load_info,
2333 ZPOOL_CONFIG_CAN_RDONLY);
2334 }
2335
2336 /*
2337 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2338 * twofold: to determine whether the pool is available for
2339 * import in read-write mode and (if it is not) whether the
2340 * pool is available for import in read-only mode. If the pool
2341 * is available for import in read-write mode, it is displayed
2342 * as available in userland; if it is not available for import
2343 * in read-only mode, it is displayed as unavailable in
2344 * userland. If the pool is available for import in read-only
2345 * mode but not read-write mode, it is displayed as unavailable
2346 * in userland with a special note that the pool is actually
2347 * available for open in read-only mode.
2348 *
2349 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2350 * missing a feature for write, we must first determine whether
2351 * the pool can be opened read-only before returning to
2352 * userland in order to know whether to display the
2353 * abovementioned note.
2354 */
2355 if (missing_feat_read || (missing_feat_write &&
2356 spa_writeable(spa))) {
2357 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2358 ENOTSUP));
2359 }
2360 }
2361
2362 spa->spa_is_initializing = B_TRUE;
2363 error = dsl_pool_open(spa->spa_dsl_pool);
2364 spa->spa_is_initializing = B_FALSE;
2365 if (error != 0)
2366 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2367
2368 if (!mosconfig) {
2369 uint64_t hostid;
2370 nvlist_t *policy = NULL, *nvconfig;
2371
2372 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2374
2375 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2376 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2377 char *hostname;
2378 unsigned long myhostid = 0;
2379
2380 VERIFY(nvlist_lookup_string(nvconfig,
2381 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2382
2383 #ifdef _KERNEL
2384 myhostid = zone_get_hostid(NULL);
2385 #else /* _KERNEL */
2386 /*
2387 * We're emulating the system's hostid in userland, so
2388 * we can't use zone_get_hostid().
2389 */
2390 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2391 #endif /* _KERNEL */
2392 if (hostid != 0 && myhostid != 0 &&
2393 hostid != myhostid) {
2394 nvlist_free(nvconfig);
2395 cmn_err(CE_WARN, "pool '%s' could not be "
2396 "loaded as it was last accessed by "
2397 "another system (host: %s hostid: 0x%lx). "
2398 "See: http://illumos.org/msg/ZFS-8000-EY",
2399 spa_name(spa), hostname,
2400 (unsigned long)hostid);
2401 return (SET_ERROR(EBADF));
2402 }
2403 }
2404 if (nvlist_lookup_nvlist(spa->spa_config,
2405 ZPOOL_REWIND_POLICY, &policy) == 0)
2406 VERIFY(nvlist_add_nvlist(nvconfig,
2407 ZPOOL_REWIND_POLICY, policy) == 0);
2408
2409 spa_config_set(spa, nvconfig);
2410 spa_unload(spa);
2411 spa_deactivate(spa);
2412 spa_activate(spa, orig_mode);
2413
2414 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2415 }
2416
2417 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2419 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2420 if (error != 0)
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422
2423 /*
2424 * Load the bit that tells us to use the new accounting function
2425 * (raid-z deflation). If we have an older pool, this will not
2426 * be present.
2427 */
2428 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2429 if (error != 0 && error != ENOENT)
2430 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431
2432 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2433 &spa->spa_creation_version);
2434 if (error != 0 && error != ENOENT)
2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2436
2437 /*
2438 * Load the persistent error log. If we have an older pool, this will
2439 * not be present.
2440 */
2441 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2442 if (error != 0 && error != ENOENT)
2443 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2444
2445 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2446 &spa->spa_errlog_scrub);
2447 if (error != 0 && error != ENOENT)
2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2449
2450 /*
2451 * Load the history object. If we have an older pool, this
2452 * will not be present.
2453 */
2454 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2455 if (error != 0 && error != ENOENT)
2456 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457
2458 /*
2459 * If we're assembling the pool from the split-off vdevs of
2460 * an existing pool, we don't want to attach the spares & cache
2461 * devices.
2462 */
2463
2464 /*
2465 * Load any hot spares for this pool.
2466 */
2467 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2468 if (error != 0 && error != ENOENT)
2469 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2471 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2472 if (load_nvlist(spa, spa->spa_spares.sav_object,
2473 &spa->spa_spares.sav_config) != 0)
2474 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2475
2476 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2477 spa_load_spares(spa);
2478 spa_config_exit(spa, SCL_ALL, FTAG);
2479 } else if (error == 0) {
2480 spa->spa_spares.sav_sync = B_TRUE;
2481 }
2482
2483 /*
2484 * Load any level 2 ARC devices for this pool.
2485 */
2486 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2487 &spa->spa_l2cache.sav_object);
2488 if (error != 0 && error != ENOENT)
2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2490 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2491 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2492 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2493 &spa->spa_l2cache.sav_config) != 0)
2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2495
2496 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2497 spa_load_l2cache(spa);
2498 spa_config_exit(spa, SCL_ALL, FTAG);
2499 } else if (error == 0) {
2500 spa->spa_l2cache.sav_sync = B_TRUE;
2501 }
2502
2503 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2504
2505 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2506 if (error && error != ENOENT)
2507 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2508
2509 if (error == 0) {
2510 uint64_t autoreplace;
2511
2512 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2513 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2514 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2515 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2516 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2517 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2518 &spa->spa_dedup_ditto);
2519
2520 spa->spa_autoreplace = (autoreplace != 0);
2521 }
2522
2523 /*
2524 * If the 'autoreplace' property is set, then post a resource notifying
2525 * the ZFS DE that it should not issue any faults for unopenable
2526 * devices. We also iterate over the vdevs, and post a sysevent for any
2527 * unopenable vdevs so that the normal autoreplace handler can take
2528 * over.
2529 */
2530 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2531 spa_check_removed(spa->spa_root_vdev);
2532 /*
2533 * For the import case, this is done in spa_import(), because
2534 * at this point we're using the spare definitions from
2535 * the MOS config, not necessarily from the userland config.
2536 */
2537 if (state != SPA_LOAD_IMPORT) {
2538 spa_aux_check_removed(&spa->spa_spares);
2539 spa_aux_check_removed(&spa->spa_l2cache);
2540 }
2541 }
2542
2543 /*
2544 * Load the vdev state for all toplevel vdevs.
2545 */
2546 vdev_load(rvd);
2547
2548 /*
2549 * Propagate the leaf DTLs we just loaded all the way up the tree.
2550 */
2551 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2552 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2553 spa_config_exit(spa, SCL_ALL, FTAG);
2554
2555 /*
2556 * Load the DDTs (dedup tables).
2557 */
2558 error = ddt_load(spa);
2559 if (error != 0)
2560 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2561
2562 spa_update_dspace(spa);
2563
2564 /*
2565 * Validate the config, using the MOS config to fill in any
2566 * information which might be missing. If we fail to validate
2567 * the config then declare the pool unfit for use. If we're
2568 * assembling a pool from a split, the log is not transferred
2569 * over.
2570 */
2571 if (type != SPA_IMPORT_ASSEMBLE) {
2572 nvlist_t *nvconfig;
2573
2574 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2575 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2576
2577 if (!spa_config_valid(spa, nvconfig)) {
2578 nvlist_free(nvconfig);
2579 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2580 ENXIO));
2581 }
2582 nvlist_free(nvconfig);
2583
2584 /*
2585 * Now that we've validated the config, check the state of the
2586 * root vdev. If it can't be opened, it indicates one or
2587 * more toplevel vdevs are faulted.
2588 */
2589 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2590 return (SET_ERROR(ENXIO));
2591
2592 if (spa_check_logs(spa)) {
2593 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2594 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2595 }
2596 }
2597
2598 if (missing_feat_write) {
2599 ASSERT(state == SPA_LOAD_TRYIMPORT);
2600
2601 /*
2602 * At this point, we know that we can open the pool in
2603 * read-only mode but not read-write mode. We now have enough
2604 * information and can return to userland.
2605 */
2606 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2607 }
2608
2609 /*
2610 * We've successfully opened the pool, verify that we're ready
2611 * to start pushing transactions.
2612 */
2613 if (state != SPA_LOAD_TRYIMPORT) {
2614 if (error = spa_load_verify(spa))
2615 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2616 error));
2617 }
2618
2619 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2620 spa->spa_load_max_txg == UINT64_MAX)) {
2621 dmu_tx_t *tx;
2622 int need_update = B_FALSE;
2623
2624 ASSERT(state != SPA_LOAD_TRYIMPORT);
2625
2626 /*
2627 * Claim log blocks that haven't been committed yet.
2628 * This must all happen in a single txg.
2629 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2630 * invoked from zil_claim_log_block()'s i/o done callback.
2631 * Price of rollback is that we abandon the log.
2632 */
2633 spa->spa_claiming = B_TRUE;
2634
2635 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2636 spa_first_txg(spa));
2637 (void) dmu_objset_find(spa_name(spa),
2638 zil_claim, tx, DS_FIND_CHILDREN);
2639 dmu_tx_commit(tx);
2640
2641 spa->spa_claiming = B_FALSE;
2642
2643 spa_set_log_state(spa, SPA_LOG_GOOD);
2644 spa->spa_sync_on = B_TRUE;
2645 txg_sync_start(spa->spa_dsl_pool);
2646
2647 /*
2648 * Wait for all claims to sync. We sync up to the highest
2649 * claimed log block birth time so that claimed log blocks
2650 * don't appear to be from the future. spa_claim_max_txg
2651 * will have been set for us by either zil_check_log_chain()
2652 * (invoked from spa_check_logs()) or zil_claim() above.
2653 */
2654 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2655
2656 /*
2657 * If the config cache is stale, or we have uninitialized
2658 * metaslabs (see spa_vdev_add()), then update the config.
2659 *
2660 * If this is a verbatim import, trust the current
2661 * in-core spa_config and update the disk labels.
2662 */
2663 if (config_cache_txg != spa->spa_config_txg ||
2664 state == SPA_LOAD_IMPORT ||
2665 state == SPA_LOAD_RECOVER ||
2666 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2667 need_update = B_TRUE;
2668
2669 for (int c = 0; c < rvd->vdev_children; c++)
2670 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2671 need_update = B_TRUE;
2672
2673 /*
2674 * Update the config cache asychronously in case we're the
2675 * root pool, in which case the config cache isn't writable yet.
2676 */
2677 if (need_update)
2678 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2679
2680 /*
2681 * Check all DTLs to see if anything needs resilvering.
2682 */
2683 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2684 vdev_resilver_needed(rvd, NULL, NULL))
2685 spa_async_request(spa, SPA_ASYNC_RESILVER);
2686
2687 /*
2688 * Log the fact that we booted up (so that we can detect if
2689 * we rebooted in the middle of an operation).
2690 */
2691 spa_history_log_version(spa, "open");
2692
2693 /*
2694 * Delete any inconsistent datasets.
2695 */
2696 (void) dmu_objset_find(spa_name(spa),
2697 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2698
2699 /*
2700 * Clean up any stale temporary dataset userrefs.
2701 */
2702 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2703 }
2704
2705 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2706
2707 return (0);
2708 }
2709
2710 static int
2711 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2712 {
2713 int mode = spa->spa_mode;
2714
2715 spa_unload(spa);
2716 spa_deactivate(spa);
2717
2718 spa->spa_load_max_txg--;
2719
2720 spa_activate(spa, mode);
2721 spa_async_suspend(spa);
2722
2723 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2724 }
2725
2726 /*
2727 * If spa_load() fails this function will try loading prior txg's. If
2728 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2729 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2730 * function will not rewind the pool and will return the same error as
2731 * spa_load().
2732 */
2733 static int
2734 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2735 uint64_t max_request, int rewind_flags)
2736 {
2737 nvlist_t *loadinfo = NULL;
2738 nvlist_t *config = NULL;
2739 int load_error, rewind_error;
2740 uint64_t safe_rewind_txg;
2741 uint64_t min_txg;
2742
2743 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2744 spa->spa_load_max_txg = spa->spa_load_txg;
2745 spa_set_log_state(spa, SPA_LOG_CLEAR);
2746 } else {
2747 spa->spa_load_max_txg = max_request;
2748 }
2749
2750 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2751 mosconfig);
2752 if (load_error == 0)
2753 return (0);
2754
2755 if (spa->spa_root_vdev != NULL)
2756 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2757
2758 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2759 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2760
2761 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2762 nvlist_free(config);
2763 return (load_error);
2764 }
2765
2766 if (state == SPA_LOAD_RECOVER) {
2767 /* Price of rolling back is discarding txgs, including log */
2768 spa_set_log_state(spa, SPA_LOG_CLEAR);
2769 } else {
2770 /*
2771 * If we aren't rolling back save the load info from our first
2772 * import attempt so that we can restore it after attempting
2773 * to rewind.
2774 */
2775 loadinfo = spa->spa_load_info;
2776 spa->spa_load_info = fnvlist_alloc();
2777 }
2778
2779 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2780 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2781 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2782 TXG_INITIAL : safe_rewind_txg;
2783
2784 /*
2785 * Continue as long as we're finding errors, we're still within
2786 * the acceptable rewind range, and we're still finding uberblocks
2787 */
2788 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2789 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2790 if (spa->spa_load_max_txg < safe_rewind_txg)
2791 spa->spa_extreme_rewind = B_TRUE;
2792 rewind_error = spa_load_retry(spa, state, mosconfig);
2793 }
2794
2795 spa->spa_extreme_rewind = B_FALSE;
2796 spa->spa_load_max_txg = UINT64_MAX;
2797
2798 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2799 spa_config_set(spa, config);
2800
2801 if (state == SPA_LOAD_RECOVER) {
2802 ASSERT3P(loadinfo, ==, NULL);
2803 return (rewind_error);
2804 } else {
2805 /* Store the rewind info as part of the initial load info */
2806 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2807 spa->spa_load_info);
2808
2809 /* Restore the initial load info */
2810 fnvlist_free(spa->spa_load_info);
2811 spa->spa_load_info = loadinfo;
2812
2813 return (load_error);
2814 }
2815 }
2816
2817 /*
2818 * Pool Open/Import
2819 *
2820 * The import case is identical to an open except that the configuration is sent
2821 * down from userland, instead of grabbed from the configuration cache. For the
2822 * case of an open, the pool configuration will exist in the
2823 * POOL_STATE_UNINITIALIZED state.
2824 *
2825 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2826 * the same time open the pool, without having to keep around the spa_t in some
2827 * ambiguous state.
2828 */
2829 static int
2830 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2831 nvlist_t **config)
2832 {
2833 spa_t *spa;
2834 spa_load_state_t state = SPA_LOAD_OPEN;
2835 int error;
2836 int locked = B_FALSE;
2837
2838 *spapp = NULL;
2839
2840 /*
2841 * As disgusting as this is, we need to support recursive calls to this
2842 * function because dsl_dir_open() is called during spa_load(), and ends
2843 * up calling spa_open() again. The real fix is to figure out how to
2844 * avoid dsl_dir_open() calling this in the first place.
2845 */
2846 if (mutex_owner(&spa_namespace_lock) != curthread) {
2847 mutex_enter(&spa_namespace_lock);
2848 locked = B_TRUE;
2849 }
2850
2851 if ((spa = spa_lookup(pool)) == NULL) {
2852 if (locked)
2853 mutex_exit(&spa_namespace_lock);
2854 return (SET_ERROR(ENOENT));
2855 }
2856
2857 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2858 zpool_rewind_policy_t policy;
2859
2860 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2861 &policy);
2862 if (policy.zrp_request & ZPOOL_DO_REWIND)
2863 state = SPA_LOAD_RECOVER;
2864
2865 spa_activate(spa, spa_mode_global);
2866
2867 if (state != SPA_LOAD_RECOVER)
2868 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2869
2870 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2871 policy.zrp_request);
2872
2873 if (error == EBADF) {
2874 /*
2875 * If vdev_validate() returns failure (indicated by
2876 * EBADF), it indicates that one of the vdevs indicates
2877 * that the pool has been exported or destroyed. If
2878 * this is the case, the config cache is out of sync and
2879 * we should remove the pool from the namespace.
2880 */
2881 spa_unload(spa);
2882 spa_deactivate(spa);
2883 spa_config_sync(spa, B_TRUE, B_TRUE);
2884 spa_remove(spa);
2885 if (locked)
2886 mutex_exit(&spa_namespace_lock);
2887 return (SET_ERROR(ENOENT));
2888 }
2889
2890 if (error) {
2891 /*
2892 * We can't open the pool, but we still have useful
2893 * information: the state of each vdev after the
2894 * attempted vdev_open(). Return this to the user.
2895 */
2896 if (config != NULL && spa->spa_config) {
2897 VERIFY(nvlist_dup(spa->spa_config, config,
2898 KM_SLEEP) == 0);
2899 VERIFY(nvlist_add_nvlist(*config,
2900 ZPOOL_CONFIG_LOAD_INFO,
2901 spa->spa_load_info) == 0);
2902 }
2903 spa_unload(spa);
2904 spa_deactivate(spa);
2905 spa->spa_last_open_failed = error;
2906 if (locked)
2907 mutex_exit(&spa_namespace_lock);
2908 *spapp = NULL;
2909 return (error);
2910 }
2911 }
2912
2913 spa_open_ref(spa, tag);
2914
2915 if (config != NULL)
2916 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2917
2918 /*
2919 * If we've recovered the pool, pass back any information we
2920 * gathered while doing the load.
2921 */
2922 if (state == SPA_LOAD_RECOVER) {
2923 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2924 spa->spa_load_info) == 0);
2925 }
2926
2927 if (locked) {
2928 spa->spa_last_open_failed = 0;
2929 spa->spa_last_ubsync_txg = 0;
2930 spa->spa_load_txg = 0;
2931 mutex_exit(&spa_namespace_lock);
2932 }
2933
2934 *spapp = spa;
2935
2936 return (0);
2937 }
2938
2939 int
2940 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2941 nvlist_t **config)
2942 {
2943 return (spa_open_common(name, spapp, tag, policy, config));
2944 }
2945
2946 int
2947 spa_open(const char *name, spa_t **spapp, void *tag)
2948 {
2949 return (spa_open_common(name, spapp, tag, NULL, NULL));
2950 }
2951
2952 /*
2953 * Lookup the given spa_t, incrementing the inject count in the process,
2954 * preventing it from being exported or destroyed.
2955 */
2956 spa_t *
2957 spa_inject_addref(char *name)
2958 {
2959 spa_t *spa;
2960
2961 mutex_enter(&spa_namespace_lock);
2962 if ((spa = spa_lookup(name)) == NULL) {
2963 mutex_exit(&spa_namespace_lock);
2964 return (NULL);
2965 }
2966 spa->spa_inject_ref++;
2967 mutex_exit(&spa_namespace_lock);
2968
2969 return (spa);
2970 }
2971
2972 void
2973 spa_inject_delref(spa_t *spa)
2974 {
2975 mutex_enter(&spa_namespace_lock);
2976 spa->spa_inject_ref--;
2977 mutex_exit(&spa_namespace_lock);
2978 }
2979
2980 /*
2981 * Add spares device information to the nvlist.
2982 */
2983 static void
2984 spa_add_spares(spa_t *spa, nvlist_t *config)
2985 {
2986 nvlist_t **spares;
2987 uint_t i, nspares;
2988 nvlist_t *nvroot;
2989 uint64_t guid;
2990 vdev_stat_t *vs;
2991 uint_t vsc;
2992 uint64_t pool;
2993
2994 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2995
2996 if (spa->spa_spares.sav_count == 0)
2997 return;
2998
2999 VERIFY(nvlist_lookup_nvlist(config,
3000 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3001 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3002 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3003 if (nspares != 0) {
3004 VERIFY(nvlist_add_nvlist_array(nvroot,
3005 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3006 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3007 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3008
3009 /*
3010 * Go through and find any spares which have since been
3011 * repurposed as an active spare. If this is the case, update
3012 * their status appropriately.
3013 */
3014 for (i = 0; i < nspares; i++) {
3015 VERIFY(nvlist_lookup_uint64(spares[i],
3016 ZPOOL_CONFIG_GUID, &guid) == 0);
3017 if (spa_spare_exists(guid, &pool, NULL) &&
3018 pool != 0ULL) {
3019 VERIFY(nvlist_lookup_uint64_array(
3020 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3021 (uint64_t **)&vs, &vsc) == 0);
3022 vs->vs_state = VDEV_STATE_CANT_OPEN;
3023 vs->vs_aux = VDEV_AUX_SPARED;
3024 }
3025 }
3026 }
3027 }
3028
3029 /*
3030 * Add l2cache device information to the nvlist, including vdev stats.
3031 */
3032 static void
3033 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3034 {
3035 nvlist_t **l2cache;
3036 uint_t i, j, nl2cache;
3037 nvlist_t *nvroot;
3038 uint64_t guid;
3039 vdev_t *vd;
3040 vdev_stat_t *vs;
3041 uint_t vsc;
3042
3043 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3044
3045 if (spa->spa_l2cache.sav_count == 0)
3046 return;
3047
3048 VERIFY(nvlist_lookup_nvlist(config,
3049 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3050 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3051 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3052 if (nl2cache != 0) {
3053 VERIFY(nvlist_add_nvlist_array(nvroot,
3054 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3055 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3056 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3057
3058 /*
3059 * Update level 2 cache device stats.
3060 */
3061
3062 for (i = 0; i < nl2cache; i++) {
3063 VERIFY(nvlist_lookup_uint64(l2cache[i],
3064 ZPOOL_CONFIG_GUID, &guid) == 0);
3065
3066 vd = NULL;
3067 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3068 if (guid ==
3069 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3070 vd = spa->spa_l2cache.sav_vdevs[j];
3071 break;
3072 }
3073 }
3074 ASSERT(vd != NULL);
3075
3076 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3077 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3078 == 0);
3079 vdev_get_stats(vd, vs);
3080 }
3081 }
3082 }
3083
3084 static void
3085 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3086 {
3087 nvlist_t *features;
3088 zap_cursor_t zc;
3089 zap_attribute_t za;
3090
3091 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3092 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3093
3094 if (spa->spa_feat_for_read_obj != 0) {
3095 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3096 spa->spa_feat_for_read_obj);
3097 zap_cursor_retrieve(&zc, &za) == 0;
3098 zap_cursor_advance(&zc)) {
3099 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3100 za.za_num_integers == 1);
3101 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3102 za.za_first_integer));
3103 }
3104 zap_cursor_fini(&zc);
3105 }
3106
3107 if (spa->spa_feat_for_write_obj != 0) {
3108 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3109 spa->spa_feat_for_write_obj);
3110 zap_cursor_retrieve(&zc, &za) == 0;
3111 zap_cursor_advance(&zc)) {
3112 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3113 za.za_num_integers == 1);
3114 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3115 za.za_first_integer));
3116 }
3117 zap_cursor_fini(&zc);
3118 }
3119
3120 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3121 features) == 0);
3122 nvlist_free(features);
3123 }
3124
3125 int
3126 spa_get_stats(const char *name, nvlist_t **config,
3127 char *altroot, size_t buflen)
3128 {
3129 int error;
3130 spa_t *spa;
3131
3132 *config = NULL;
3133 error = spa_open_common(name, &spa, FTAG, NULL, config);
3134
3135 if (spa != NULL) {
3136 /*
3137 * This still leaves a window of inconsistency where the spares
3138 * or l2cache devices could change and the config would be
3139 * self-inconsistent.
3140 */
3141 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3142
3143 if (*config != NULL) {
3144 uint64_t loadtimes[2];
3145
3146 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3147 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3148 VERIFY(nvlist_add_uint64_array(*config,
3149 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3150
3151 VERIFY(nvlist_add_uint64(*config,
3152 ZPOOL_CONFIG_ERRCOUNT,
3153 spa_get_errlog_size(spa)) == 0);
3154
3155 if (spa_suspended(spa))
3156 VERIFY(nvlist_add_uint64(*config,
3157 ZPOOL_CONFIG_SUSPENDED,
3158 spa->spa_failmode) == 0);
3159
3160 spa_add_spares(spa, *config);
3161 spa_add_l2cache(spa, *config);
3162 spa_add_feature_stats(spa, *config);
3163 }
3164 }
3165
3166 /*
3167 * We want to get the alternate root even for faulted pools, so we cheat
3168 * and call spa_lookup() directly.
3169 */
3170 if (altroot) {
3171 if (spa == NULL) {
3172 mutex_enter(&spa_namespace_lock);
3173 spa = spa_lookup(name);
3174 if (spa)
3175 spa_altroot(spa, altroot, buflen);
3176 else
3177 altroot[0] = '\0';
3178 spa = NULL;
3179 mutex_exit(&spa_namespace_lock);
3180 } else {
3181 spa_altroot(spa, altroot, buflen);
3182 }
3183 }
3184
3185 if (spa != NULL) {
3186 spa_config_exit(spa, SCL_CONFIG, FTAG);
3187 spa_close(spa, FTAG);
3188 }
3189
3190 return (error);
3191 }
3192
3193 /*
3194 * Validate that the auxiliary device array is well formed. We must have an
3195 * array of nvlists, each which describes a valid leaf vdev. If this is an
3196 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3197 * specified, as long as they are well-formed.
3198 */
3199 static int
3200 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3201 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3202 vdev_labeltype_t label)
3203 {
3204 nvlist_t **dev;
3205 uint_t i, ndev;
3206 vdev_t *vd;
3207 int error;
3208
3209 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3210
3211 /*
3212 * It's acceptable to have no devs specified.
3213 */
3214 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3215 return (0);
3216
3217 if (ndev == 0)
3218 return (SET_ERROR(EINVAL));
3219
3220 /*
3221 * Make sure the pool is formatted with a version that supports this
3222 * device type.
3223 */
3224 if (spa_version(spa) < version)
3225 return (SET_ERROR(ENOTSUP));
3226
3227 /*
3228 * Set the pending device list so we correctly handle device in-use
3229 * checking.
3230 */
3231 sav->sav_pending = dev;
3232 sav->sav_npending = ndev;
3233
3234 for (i = 0; i < ndev; i++) {
3235 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3236 mode)) != 0)
3237 goto out;
3238
3239 if (!vd->vdev_ops->vdev_op_leaf) {
3240 vdev_free(vd);
3241 error = SET_ERROR(EINVAL);
3242 goto out;
3243 }
3244
3245 /*
3246 * The L2ARC currently only supports disk devices in
3247 * kernel context. For user-level testing, we allow it.
3248 */
3249 #ifdef _KERNEL
3250 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3251 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3252 error = SET_ERROR(ENOTBLK);
3253 vdev_free(vd);
3254 goto out;
3255 }
3256 #endif
3257 vd->vdev_top = vd;
3258
3259 if ((error = vdev_open(vd)) == 0 &&
3260 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3261 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3262 vd->vdev_guid) == 0);
3263 }
3264
3265 vdev_free(vd);
3266
3267 if (error &&
3268 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3269 goto out;
3270 else
3271 error = 0;
3272 }
3273
3274 out:
3275 sav->sav_pending = NULL;
3276 sav->sav_npending = 0;
3277 return (error);
3278 }
3279
3280 static int
3281 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3282 {
3283 int error;
3284
3285 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3286
3287 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3288 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3289 VDEV_LABEL_SPARE)) != 0) {
3290 return (error);
3291 }
3292
3293 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3294 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3295 VDEV_LABEL_L2CACHE));
3296 }
3297
3298 static void
3299 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3300 const char *config)
3301 {
3302 int i;
3303
3304 if (sav->sav_config != NULL) {
3305 nvlist_t **olddevs;
3306 uint_t oldndevs;
3307 nvlist_t **newdevs;
3308
3309 /*
3310 * Generate new dev list by concatentating with the
3311 * current dev list.
3312 */
3313 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3314 &olddevs, &oldndevs) == 0);
3315
3316 newdevs = kmem_alloc(sizeof (void *) *
3317 (ndevs + oldndevs), KM_SLEEP);
3318 for (i = 0; i < oldndevs; i++)
3319 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3320 KM_SLEEP) == 0);
3321 for (i = 0; i < ndevs; i++)
3322 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3323 KM_SLEEP) == 0);
3324
3325 VERIFY(nvlist_remove(sav->sav_config, config,
3326 DATA_TYPE_NVLIST_ARRAY) == 0);
3327
3328 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3329 config, newdevs, ndevs + oldndevs) == 0);
3330 for (i = 0; i < oldndevs + ndevs; i++)
3331 nvlist_free(newdevs[i]);
3332 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3333 } else {
3334 /*
3335 * Generate a new dev list.
3336 */
3337 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3338 KM_SLEEP) == 0);
3339 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3340 devs, ndevs) == 0);
3341 }
3342 }
3343
3344 /*
3345 * Stop and drop level 2 ARC devices
3346 */
3347 void
3348 spa_l2cache_drop(spa_t *spa)
3349 {
3350 vdev_t *vd;
3351 int i;
3352 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3353
3354 for (i = 0; i < sav->sav_count; i++) {
3355 uint64_t pool;
3356
3357 vd = sav->sav_vdevs[i];
3358 ASSERT(vd != NULL);
3359
3360 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3361 pool != 0ULL && l2arc_vdev_present(vd))
3362 l2arc_remove_vdev(vd);
3363 }
3364 }
3365
3366 /*
3367 * Pool Creation
3368 */
3369 int
3370 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3371 nvlist_t *zplprops)
3372 {
3373 spa_t *spa;
3374 char *altroot = NULL;
3375 vdev_t *rvd;
3376 dsl_pool_t *dp;
3377 dmu_tx_t *tx;
3378 int error = 0;
3379 uint64_t txg = TXG_INITIAL;
3380 nvlist_t **spares, **l2cache;
3381 uint_t nspares, nl2cache;
3382 uint64_t version, obj;
3383 boolean_t has_features;
3384
3385 /*
3386 * If this pool already exists, return failure.
3387 */
3388 mutex_enter(&spa_namespace_lock);
3389 if (spa_lookup(pool) != NULL) {
3390 mutex_exit(&spa_namespace_lock);
3391 return (SET_ERROR(EEXIST));
3392 }
3393
3394 /*
3395 * Allocate a new spa_t structure.
3396 */
3397 (void) nvlist_lookup_string(props,
3398 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3399 spa = spa_add(pool, NULL, altroot);
3400 spa_activate(spa, spa_mode_global);
3401
3402 if (props && (error = spa_prop_validate(spa, props))) {
3403 spa_deactivate(spa);
3404 spa_remove(spa);
3405 mutex_exit(&spa_namespace_lock);
3406 return (error);
3407 }
3408
3409 has_features = B_FALSE;
3410 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3411 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3412 if (zpool_prop_feature(nvpair_name(elem)))
3413 has_features = B_TRUE;
3414 }
3415
3416 if (has_features || nvlist_lookup_uint64(props,
3417 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3418 version = SPA_VERSION;
3419 }
3420 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3421
3422 spa->spa_first_txg = txg;
3423 spa->spa_uberblock.ub_txg = txg - 1;
3424 spa->spa_uberblock.ub_version = version;
3425 spa->spa_ubsync = spa->spa_uberblock;
3426
3427 /*
3428 * Create "The Godfather" zio to hold all async IOs
3429 */
3430 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3431 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3432
3433 /*
3434 * Create the root vdev.
3435 */
3436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3437
3438 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3439
3440 ASSERT(error != 0 || rvd != NULL);
3441 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3442
3443 if (error == 0 && !zfs_allocatable_devs(nvroot))
3444 error = SET_ERROR(EINVAL);
3445
3446 if (error == 0 &&
3447 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3448 (error = spa_validate_aux(spa, nvroot, txg,
3449 VDEV_ALLOC_ADD)) == 0) {
3450 for (int c = 0; c < rvd->vdev_children; c++) {
3451 vdev_metaslab_set_size(rvd->vdev_child[c]);
3452 vdev_expand(rvd->vdev_child[c], txg);
3453 }
3454 }
3455
3456 spa_config_exit(spa, SCL_ALL, FTAG);
3457
3458 if (error != 0) {
3459 spa_unload(spa);
3460 spa_deactivate(spa);
3461 spa_remove(spa);
3462 mutex_exit(&spa_namespace_lock);
3463 return (error);
3464 }
3465
3466 /*
3467 * Get the list of spares, if specified.
3468 */
3469 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3470 &spares, &nspares) == 0) {
3471 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3472 KM_SLEEP) == 0);
3473 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3474 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3475 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3476 spa_load_spares(spa);
3477 spa_config_exit(spa, SCL_ALL, FTAG);
3478 spa->spa_spares.sav_sync = B_TRUE;
3479 }
3480
3481 /*
3482 * Get the list of level 2 cache devices, if specified.
3483 */
3484 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3485 &l2cache, &nl2cache) == 0) {
3486 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3487 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3488 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3489 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3491 spa_load_l2cache(spa);
3492 spa_config_exit(spa, SCL_ALL, FTAG);
3493 spa->spa_l2cache.sav_sync = B_TRUE;
3494 }
3495
3496 spa->spa_is_initializing = B_TRUE;
3497 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3498 spa->spa_meta_objset = dp->dp_meta_objset;
3499 spa->spa_is_initializing = B_FALSE;
3500
3501 /*
3502 * Create DDTs (dedup tables).
3503 */
3504 ddt_create(spa);
3505
3506 spa_update_dspace(spa);
3507
3508 tx = dmu_tx_create_assigned(dp, txg);
3509
3510 /*
3511 * Create the pool config object.
3512 */
3513 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3514 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3515 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3516
3517 if (zap_add(spa->spa_meta_objset,
3518 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3519 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3520 cmn_err(CE_PANIC, "failed to add pool config");
3521 }
3522
3523 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3524 spa_feature_create_zap_objects(spa, tx);
3525
3526 if (zap_add(spa->spa_meta_objset,
3527 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3528 sizeof (uint64_t), 1, &version, tx) != 0) {
3529 cmn_err(CE_PANIC, "failed to add pool version");
3530 }
3531
3532 /* Newly created pools with the right version are always deflated. */
3533 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3534 spa->spa_deflate = TRUE;
3535 if (zap_add(spa->spa_meta_objset,
3536 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3537 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3538 cmn_err(CE_PANIC, "failed to add deflate");
3539 }
3540 }
3541
3542 /*
3543 * Create the deferred-free bpobj. Turn off compression
3544 * because sync-to-convergence takes longer if the blocksize
3545 * keeps changing.
3546 */
3547 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3548 dmu_object_set_compress(spa->spa_meta_objset, obj,
3549 ZIO_COMPRESS_OFF, tx);
3550 if (zap_add(spa->spa_meta_objset,
3551 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3552 sizeof (uint64_t), 1, &obj, tx) != 0) {
3553 cmn_err(CE_PANIC, "failed to add bpobj");
3554 }
3555 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3556 spa->spa_meta_objset, obj));
3557
3558 /*
3559 * Create the pool's history object.
3560 */
3561 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3562 spa_history_create_obj(spa, tx);
3563
3564 /*
3565 * Set pool properties.
3566 */
3567 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3568 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3569 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3570 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3571
3572 if (props != NULL) {
3573 spa_configfile_set(spa, props, B_FALSE);
3574 spa_sync_props(props, tx);
3575 }
3576
3577 dmu_tx_commit(tx);
3578
3579 spa->spa_sync_on = B_TRUE;
3580 txg_sync_start(spa->spa_dsl_pool);
3581
3582 /*
3583 * We explicitly wait for the first transaction to complete so that our
3584 * bean counters are appropriately updated.
3585 */
3586 txg_wait_synced(spa->spa_dsl_pool, txg);
3587
3588 spa_config_sync(spa, B_FALSE, B_TRUE);
3589
3590 spa_history_log_version(spa, "create");
3591
3592 spa->spa_minref = refcount_count(&spa->spa_refcount);
3593
3594 mutex_exit(&spa_namespace_lock);
3595
3596 return (0);
3597 }
3598
3599 #ifdef _KERNEL
3600 /*
3601 * Get the root pool information from the root disk, then import the root pool
3602 * during the system boot up time.
3603 */
3604 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3605
3606 static nvlist_t *
3607 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3608 {
3609 nvlist_t *config;
3610 nvlist_t *nvtop, *nvroot;
3611 uint64_t pgid;
3612
3613 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3614 return (NULL);
3615
3616 /*
3617 * Add this top-level vdev to the child array.
3618 */
3619 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3620 &nvtop) == 0);
3621 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3622 &pgid) == 0);
3623 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3624
3625 /*
3626 * Put this pool's top-level vdevs into a root vdev.
3627 */
3628 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3629 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3630 VDEV_TYPE_ROOT) == 0);
3631 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3632 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3633 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3634 &nvtop, 1) == 0);
3635
3636 /*
3637 * Replace the existing vdev_tree with the new root vdev in
3638 * this pool's configuration (remove the old, add the new).
3639 */
3640 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3641 nvlist_free(nvroot);
3642 return (config);
3643 }
3644
3645 /*
3646 * Walk the vdev tree and see if we can find a device with "better"
3647 * configuration. A configuration is "better" if the label on that
3648 * device has a more recent txg.
3649 */
3650 static void
3651 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3652 {
3653 for (int c = 0; c < vd->vdev_children; c++)
3654 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3655
3656 if (vd->vdev_ops->vdev_op_leaf) {
3657 nvlist_t *label;
3658 uint64_t label_txg;
3659
3660 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3661 &label) != 0)
3662 return;
3663
3664 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3665 &label_txg) == 0);
3666
3667 /*
3668 * Do we have a better boot device?
3669 */
3670 if (label_txg > *txg) {
3671 *txg = label_txg;
3672 *avd = vd;
3673 }
3674 nvlist_free(label);
3675 }
3676 }
3677
3678 /*
3679 * Import a root pool.
3680 *
3681 * For x86. devpath_list will consist of devid and/or physpath name of
3682 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3683 * The GRUB "findroot" command will return the vdev we should boot.
3684 *
3685 * For Sparc, devpath_list consists the physpath name of the booting device
3686 * no matter the rootpool is a single device pool or a mirrored pool.
3687 * e.g.
3688 * "/pci@1f,0/ide@d/disk@0,0:a"
3689 */
3690 int
3691 spa_import_rootpool(char *devpath, char *devid)
3692 {
3693 spa_t *spa;
3694 vdev_t *rvd, *bvd, *avd = NULL;
3695 nvlist_t *config, *nvtop;
3696 uint64_t guid, txg;
3697 char *pname;
3698 int error;
3699
3700 /*
3701 * Read the label from the boot device and generate a configuration.
3702 */
3703 config = spa_generate_rootconf(devpath, devid, &guid);
3704 #if defined(_OBP) && defined(_KERNEL)
3705 if (config == NULL) {
3706 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3707 /* iscsi boot */
3708 get_iscsi_bootpath_phy(devpath);
3709 config = spa_generate_rootconf(devpath, devid, &guid);
3710 }
3711 }
3712 #endif
3713 if (config == NULL) {
3714 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3715 devpath);
3716 return (SET_ERROR(EIO));
3717 }
3718
3719 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3720 &pname) == 0);
3721 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3722
3723 mutex_enter(&spa_namespace_lock);
3724 if ((spa = spa_lookup(pname)) != NULL) {
3725 /*
3726 * Remove the existing root pool from the namespace so that we
3727 * can replace it with the correct config we just read in.
3728 */
3729 spa_remove(spa);
3730 }
3731
3732 spa = spa_add(pname, config, NULL);
3733 spa->spa_is_root = B_TRUE;
3734 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3735
3736 /*
3737 * Build up a vdev tree based on the boot device's label config.
3738 */
3739 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3740 &nvtop) == 0);
3741 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3742 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3743 VDEV_ALLOC_ROOTPOOL);
3744 spa_config_exit(spa, SCL_ALL, FTAG);
3745 if (error) {
3746 mutex_exit(&spa_namespace_lock);
3747 nvlist_free(config);
3748 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3749 pname);
3750 return (error);
3751 }
3752
3753 /*
3754 * Get the boot vdev.
3755 */
3756 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3757 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3758 (u_longlong_t)guid);
3759 error = SET_ERROR(ENOENT);
3760 goto out;
3761 }
3762
3763 /*
3764 * Determine if there is a better boot device.
3765 */
3766 avd = bvd;
3767 spa_alt_rootvdev(rvd, &avd, &txg);
3768 if (avd != bvd) {
3769 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3770 "try booting from '%s'", avd->vdev_path);
3771 error = SET_ERROR(EINVAL);
3772 goto out;
3773 }
3774
3775 /*
3776 * If the boot device is part of a spare vdev then ensure that
3777 * we're booting off the active spare.
3778 */
3779 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3780 !bvd->vdev_isspare) {
3781 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3782 "try booting from '%s'",
3783 bvd->vdev_parent->
3784 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3785 error = SET_ERROR(EINVAL);
3786 goto out;
3787 }
3788
3789 error = 0;
3790 out:
3791 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3792 vdev_free(rvd);
3793 spa_config_exit(spa, SCL_ALL, FTAG);
3794 mutex_exit(&spa_namespace_lock);
3795
3796 nvlist_free(config);
3797 return (error);
3798 }
3799
3800 #endif
3801
3802 /*
3803 * Import a non-root pool into the system.
3804 */
3805 int
3806 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3807 {
3808 spa_t *spa;
3809 char *altroot = NULL;
3810 spa_load_state_t state = SPA_LOAD_IMPORT;
3811 zpool_rewind_policy_t policy;
3812 uint64_t mode = spa_mode_global;
3813 uint64_t readonly = B_FALSE;
3814 int error;
3815 nvlist_t *nvroot;
3816 nvlist_t **spares, **l2cache;
3817 uint_t nspares, nl2cache;
3818
3819 /*
3820 * If a pool with this name exists, return failure.
3821 */
3822 mutex_enter(&spa_namespace_lock);
3823 if (spa_lookup(pool) != NULL) {
3824 mutex_exit(&spa_namespace_lock);
3825 return (SET_ERROR(EEXIST));
3826 }
3827
3828 /*
3829 * Create and initialize the spa structure.
3830 */
3831 (void) nvlist_lookup_string(props,
3832 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3833 (void) nvlist_lookup_uint64(props,
3834 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3835 if (readonly)
3836 mode = FREAD;
3837 spa = spa_add(pool, config, altroot);
3838 spa->spa_import_flags = flags;
3839
3840 /*
3841 * Verbatim import - Take a pool and insert it into the namespace
3842 * as if it had been loaded at boot.
3843 */
3844 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3845 if (props != NULL)
3846 spa_configfile_set(spa, props, B_FALSE);
3847
3848 spa_config_sync(spa, B_FALSE, B_TRUE);
3849
3850 mutex_exit(&spa_namespace_lock);
3851 return (0);
3852 }
3853
3854 spa_activate(spa, mode);
3855
3856 /*
3857 * Don't start async tasks until we know everything is healthy.
3858 */
3859 spa_async_suspend(spa);
3860
3861 zpool_get_rewind_policy(config, &policy);
3862 if (policy.zrp_request & ZPOOL_DO_REWIND)
3863 state = SPA_LOAD_RECOVER;
3864
3865 /*
3866 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3867 * because the user-supplied config is actually the one to trust when
3868 * doing an import.
3869 */
3870 if (state != SPA_LOAD_RECOVER)
3871 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3872
3873 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3874 policy.zrp_request);
3875
3876 /*
3877 * Propagate anything learned while loading the pool and pass it
3878 * back to caller (i.e. rewind info, missing devices, etc).
3879 */
3880 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3881 spa->spa_load_info) == 0);
3882
3883 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3884 /*
3885 * Toss any existing sparelist, as it doesn't have any validity
3886 * anymore, and conflicts with spa_has_spare().
3887 */
3888 if (spa->spa_spares.sav_config) {
3889 nvlist_free(spa->spa_spares.sav_config);
3890 spa->spa_spares.sav_config = NULL;
3891 spa_load_spares(spa);
3892 }
3893 if (spa->spa_l2cache.sav_config) {
3894 nvlist_free(spa->spa_l2cache.sav_config);
3895 spa->spa_l2cache.sav_config = NULL;
3896 spa_load_l2cache(spa);
3897 }
3898
3899 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3900 &nvroot) == 0);
3901 if (error == 0)
3902 error = spa_validate_aux(spa, nvroot, -1ULL,
3903 VDEV_ALLOC_SPARE);
3904 if (error == 0)
3905 error = spa_validate_aux(spa, nvroot, -1ULL,
3906 VDEV_ALLOC_L2CACHE);
3907 spa_config_exit(spa, SCL_ALL, FTAG);
3908
3909 if (props != NULL)
3910 spa_configfile_set(spa, props, B_FALSE);
3911
3912 if (error != 0 || (props && spa_writeable(spa) &&
3913 (error = spa_prop_set(spa, props)))) {
3914 spa_unload(spa);
3915 spa_deactivate(spa);
3916 spa_remove(spa);
3917 mutex_exit(&spa_namespace_lock);
3918 return (error);
3919 }
3920
3921 spa_async_resume(spa);
3922
3923 /*
3924 * Override any spares and level 2 cache devices as specified by
3925 * the user, as these may have correct device names/devids, etc.
3926 */
3927 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3928 &spares, &nspares) == 0) {
3929 if (spa->spa_spares.sav_config)
3930 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3931 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3932 else
3933 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3934 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3935 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3936 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3937 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3938 spa_load_spares(spa);
3939 spa_config_exit(spa, SCL_ALL, FTAG);
3940 spa->spa_spares.sav_sync = B_TRUE;
3941 }
3942 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3943 &l2cache, &nl2cache) == 0) {
3944 if (spa->spa_l2cache.sav_config)
3945 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3946 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3947 else
3948 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3949 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3950 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3951 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3953 spa_load_l2cache(spa);
3954 spa_config_exit(spa, SCL_ALL, FTAG);
3955 spa->spa_l2cache.sav_sync = B_TRUE;
3956 }
3957
3958 /*
3959 * Check for any removed devices.
3960 */
3961 if (spa->spa_autoreplace) {
3962 spa_aux_check_removed(&spa->spa_spares);
3963 spa_aux_check_removed(&spa->spa_l2cache);
3964 }
3965
3966 if (spa_writeable(spa)) {
3967 /*
3968 * Update the config cache to include the newly-imported pool.
3969 */
3970 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3971 }
3972
3973 /*
3974 * It's possible that the pool was expanded while it was exported.
3975 * We kick off an async task to handle this for us.
3976 */
3977 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3978
3979 mutex_exit(&spa_namespace_lock);
3980 spa_history_log_version(spa, "import");
3981
3982 return (0);
3983 }
3984
3985 nvlist_t *
3986 spa_tryimport(nvlist_t *tryconfig)
3987 {
3988 nvlist_t *config = NULL;
3989 char *poolname;
3990 spa_t *spa;
3991 uint64_t state;
3992 int error;
3993
3994 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3995 return (NULL);
3996
3997 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3998 return (NULL);
3999
4000 /*
4001 * Create and initialize the spa structure.
4002 */
4003 mutex_enter(&spa_namespace_lock);
4004 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4005 spa_activate(spa, FREAD);
4006
4007 /*
4008 * Pass off the heavy lifting to spa_load().
4009 * Pass TRUE for mosconfig because the user-supplied config
4010 * is actually the one to trust when doing an import.
4011 */
4012 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4013
4014 /*
4015 * If 'tryconfig' was at least parsable, return the current config.
4016 */
4017 if (spa->spa_root_vdev != NULL) {
4018 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4019 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4020 poolname) == 0);
4021 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4022 state) == 0);
4023 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4024 spa->spa_uberblock.ub_timestamp) == 0);
4025 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4026 spa->spa_load_info) == 0);
4027
4028 /*
4029 * If the bootfs property exists on this pool then we
4030 * copy it out so that external consumers can tell which
4031 * pools are bootable.
4032 */
4033 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4034 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4035
4036 /*
4037 * We have to play games with the name since the
4038 * pool was opened as TRYIMPORT_NAME.
4039 */
4040 if (dsl_dsobj_to_dsname(spa_name(spa),
4041 spa->spa_bootfs, tmpname) == 0) {
4042 char *cp;
4043 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4044
4045 cp = strchr(tmpname, '/');
4046 if (cp == NULL) {
4047 (void) strlcpy(dsname, tmpname,
4048 MAXPATHLEN);
4049 } else {
4050 (void) snprintf(dsname, MAXPATHLEN,
4051 "%s/%s", poolname, ++cp);
4052 }
4053 VERIFY(nvlist_add_string(config,
4054 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4055 kmem_free(dsname, MAXPATHLEN);
4056 }
4057 kmem_free(tmpname, MAXPATHLEN);
4058 }
4059
4060 /*
4061 * Add the list of hot spares and level 2 cache devices.
4062 */
4063 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4064 spa_add_spares(spa, config);
4065 spa_add_l2cache(spa, config);
4066 spa_config_exit(spa, SCL_CONFIG, FTAG);
4067 }
4068
4069 spa_unload(spa);
4070 spa_deactivate(spa);
4071 spa_remove(spa);
4072 mutex_exit(&spa_namespace_lock);
4073
4074 return (config);
4075 }
4076
4077 /*
4078 * Pool export/destroy
4079 *
4080 * The act of destroying or exporting a pool is very simple. We make sure there
4081 * is no more pending I/O and any references to the pool are gone. Then, we
4082 * update the pool state and sync all the labels to disk, removing the
4083 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4084 * we don't sync the labels or remove the configuration cache.
4085 */
4086 static int
4087 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4088 boolean_t force, boolean_t hardforce)
4089 {
4090 spa_t *spa;
4091
4092 if (oldconfig)
4093 *oldconfig = NULL;
4094
4095 if (!(spa_mode_global & FWRITE))
4096 return (SET_ERROR(EROFS));
4097
4098 mutex_enter(&spa_namespace_lock);
4099 if ((spa = spa_lookup(pool)) == NULL) {
4100 mutex_exit(&spa_namespace_lock);
4101 return (SET_ERROR(ENOENT));
4102 }
4103
4104 /*
4105 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4106 * reacquire the namespace lock, and see if we can export.
4107 */
4108 spa_open_ref(spa, FTAG);
4109 mutex_exit(&spa_namespace_lock);
4110 spa_async_suspend(spa);
4111 mutex_enter(&spa_namespace_lock);
4112 spa_close(spa, FTAG);
4113
4114 /*
4115 * The pool will be in core if it's openable,
4116 * in which case we can modify its state.
4117 */
4118 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4119 /*
4120 * Objsets may be open only because they're dirty, so we
4121 * have to force it to sync before checking spa_refcnt.
4122 */
4123 txg_wait_synced(spa->spa_dsl_pool, 0);
4124
4125 /*
4126 * A pool cannot be exported or destroyed if there are active
4127 * references. If we are resetting a pool, allow references by
4128 * fault injection handlers.
4129 */
4130 if (!spa_refcount_zero(spa) ||
4131 (spa->spa_inject_ref != 0 &&
4132 new_state != POOL_STATE_UNINITIALIZED)) {
4133 spa_async_resume(spa);
4134 mutex_exit(&spa_namespace_lock);
4135 return (SET_ERROR(EBUSY));
4136 }
4137
4138 /*
4139 * A pool cannot be exported if it has an active shared spare.
4140 * This is to prevent other pools stealing the active spare
4141 * from an exported pool. At user's own will, such pool can
4142 * be forcedly exported.
4143 */
4144 if (!force && new_state == POOL_STATE_EXPORTED &&
4145 spa_has_active_shared_spare(spa)) {
4146 spa_async_resume(spa);
4147 mutex_exit(&spa_namespace_lock);
4148 return (SET_ERROR(EXDEV));
4149 }
4150
4151 /*
4152 * We want this to be reflected on every label,
4153 * so mark them all dirty. spa_unload() will do the
4154 * final sync that pushes these changes out.
4155 */
4156 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4157 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4158 spa->spa_state = new_state;
4159 spa->spa_final_txg = spa_last_synced_txg(spa) +
4160 TXG_DEFER_SIZE + 1;
4161 vdev_config_dirty(spa->spa_root_vdev);
4162 spa_config_exit(spa, SCL_ALL, FTAG);
4163 }
4164 }
4165
4166 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4167
4168 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4169 spa_unload(spa);
4170 spa_deactivate(spa);
4171 }
4172
4173 if (oldconfig && spa->spa_config)
4174 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4175
4176 if (new_state != POOL_STATE_UNINITIALIZED) {
4177 if (!hardforce)
4178 spa_config_sync(spa, B_TRUE, B_TRUE);
4179 spa_remove(spa);
4180 }
4181 mutex_exit(&spa_namespace_lock);
4182
4183 return (0);
4184 }
4185
4186 /*
4187 * Destroy a storage pool.
4188 */
4189 int
4190 spa_destroy(char *pool)
4191 {
4192 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4193 B_FALSE, B_FALSE));
4194 }
4195
4196 /*
4197 * Export a storage pool.
4198 */
4199 int
4200 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4201 boolean_t hardforce)
4202 {
4203 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4204 force, hardforce));
4205 }
4206
4207 /*
4208 * Similar to spa_export(), this unloads the spa_t without actually removing it
4209 * from the namespace in any way.
4210 */
4211 int
4212 spa_reset(char *pool)
4213 {
4214 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4215 B_FALSE, B_FALSE));
4216 }
4217
4218 /*
4219 * ==========================================================================
4220 * Device manipulation
4221 * ==========================================================================
4222 */
4223
4224 /*
4225 * Add a device to a storage pool.
4226 */
4227 int
4228 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4229 {
4230 uint64_t txg, id;
4231 int error;
4232 vdev_t *rvd = spa->spa_root_vdev;
4233 vdev_t *vd, *tvd;
4234 nvlist_t **spares, **l2cache;
4235 uint_t nspares, nl2cache;
4236
4237 ASSERT(spa_writeable(spa));
4238
4239 txg = spa_vdev_enter(spa);
4240
4241 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4242 VDEV_ALLOC_ADD)) != 0)
4243 return (spa_vdev_exit(spa, NULL, txg, error));
4244
4245 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4246
4247 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4248 &nspares) != 0)
4249 nspares = 0;
4250
4251 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4252 &nl2cache) != 0)
4253 nl2cache = 0;
4254
4255 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4256 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4257
4258 if (vd->vdev_children != 0 &&
4259 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4260 return (spa_vdev_exit(spa, vd, txg, error));
4261
4262 /*
4263 * We must validate the spares and l2cache devices after checking the
4264 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4265 */
4266 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4267 return (spa_vdev_exit(spa, vd, txg, error));
4268
4269 /*
4270 * Transfer each new top-level vdev from vd to rvd.
4271 */
4272 for (int c = 0; c < vd->vdev_children; c++) {
4273
4274 /*
4275 * Set the vdev id to the first hole, if one exists.
4276 */
4277 for (id = 0; id < rvd->vdev_children; id++) {
4278 if (rvd->vdev_child[id]->vdev_ishole) {
4279 vdev_free(rvd->vdev_child[id]);
4280 break;
4281 }
4282 }
4283 tvd = vd->vdev_child[c];
4284 vdev_remove_child(vd, tvd);
4285 tvd->vdev_id = id;
4286 vdev_add_child(rvd, tvd);
4287 vdev_config_dirty(tvd);
4288 }
4289
4290 if (nspares != 0) {
4291 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4292 ZPOOL_CONFIG_SPARES);
4293 spa_load_spares(spa);
4294 spa->spa_spares.sav_sync = B_TRUE;
4295 }
4296
4297 if (nl2cache != 0) {
4298 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4299 ZPOOL_CONFIG_L2CACHE);
4300 spa_load_l2cache(spa);
4301 spa->spa_l2cache.sav_sync = B_TRUE;
4302 }
4303
4304 /*
4305 * We have to be careful when adding new vdevs to an existing pool.
4306 * If other threads start allocating from these vdevs before we
4307 * sync the config cache, and we lose power, then upon reboot we may
4308 * fail to open the pool because there are DVAs that the config cache
4309 * can't translate. Therefore, we first add the vdevs without
4310 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4311 * and then let spa_config_update() initialize the new metaslabs.
4312 *
4313 * spa_load() checks for added-but-not-initialized vdevs, so that
4314 * if we lose power at any point in this sequence, the remaining
4315 * steps will be completed the next time we load the pool.
4316 */
4317 (void) spa_vdev_exit(spa, vd, txg, 0);
4318
4319 mutex_enter(&spa_namespace_lock);
4320 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4321 mutex_exit(&spa_namespace_lock);
4322
4323 return (0);
4324 }
4325
4326 /*
4327 * Attach a device to a mirror. The arguments are the path to any device
4328 * in the mirror, and the nvroot for the new device. If the path specifies
4329 * a device that is not mirrored, we automatically insert the mirror vdev.
4330 *
4331 * If 'replacing' is specified, the new device is intended to replace the
4332 * existing device; in this case the two devices are made into their own
4333 * mirror using the 'replacing' vdev, which is functionally identical to
4334 * the mirror vdev (it actually reuses all the same ops) but has a few
4335 * extra rules: you can't attach to it after it's been created, and upon
4336 * completion of resilvering, the first disk (the one being replaced)
4337 * is automatically detached.
4338 */
4339 int
4340 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4341 {
4342 uint64_t txg, dtl_max_txg;
4343 vdev_t *rvd = spa->spa_root_vdev;
4344 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4345 vdev_ops_t *pvops;
4346 char *oldvdpath, *newvdpath;
4347 int newvd_isspare;
4348 int error;
4349
4350 ASSERT(spa_writeable(spa));
4351
4352 txg = spa_vdev_enter(spa);
4353
4354 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4355
4356 if (oldvd == NULL)
4357 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4358
4359 if (!oldvd->vdev_ops->vdev_op_leaf)
4360 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4361
4362 pvd = oldvd->vdev_parent;
4363
4364 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4365 VDEV_ALLOC_ATTACH)) != 0)
4366 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4367
4368 if (newrootvd->vdev_children != 1)
4369 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4370
4371 newvd = newrootvd->vdev_child[0];
4372
4373 if (!newvd->vdev_ops->vdev_op_leaf)
4374 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4375
4376 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4377 return (spa_vdev_exit(spa, newrootvd, txg, error));
4378
4379 /*
4380 * Spares can't replace logs
4381 */
4382 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4383 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4384
4385 if (!replacing) {
4386 /*
4387 * For attach, the only allowable parent is a mirror or the root
4388 * vdev.
4389 */
4390 if (pvd->vdev_ops != &vdev_mirror_ops &&
4391 pvd->vdev_ops != &vdev_root_ops)
4392 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4393
4394 pvops = &vdev_mirror_ops;
4395 } else {
4396 /*
4397 * Active hot spares can only be replaced by inactive hot
4398 * spares.
4399 */
4400 if (pvd->vdev_ops == &vdev_spare_ops &&
4401 oldvd->vdev_isspare &&
4402 !spa_has_spare(spa, newvd->vdev_guid))
4403 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4404
4405 /*
4406 * If the source is a hot spare, and the parent isn't already a
4407 * spare, then we want to create a new hot spare. Otherwise, we
4408 * want to create a replacing vdev. The user is not allowed to
4409 * attach to a spared vdev child unless the 'isspare' state is
4410 * the same (spare replaces spare, non-spare replaces
4411 * non-spare).
4412 */
4413 if (pvd->vdev_ops == &vdev_replacing_ops &&
4414 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4415 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4416 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4417 newvd->vdev_isspare != oldvd->vdev_isspare) {
4418 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4419 }
4420
4421 if (newvd->vdev_isspare)
4422 pvops = &vdev_spare_ops;
4423 else
4424 pvops = &vdev_replacing_ops;
4425 }
4426
4427 /*
4428 * Make sure the new device is big enough.
4429 */
4430 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4431 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4432
4433 /*
4434 * The new device cannot have a higher alignment requirement
4435 * than the top-level vdev.
4436 */
4437 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4438 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4439
4440 /*
4441 * If this is an in-place replacement, update oldvd's path and devid
4442 * to make it distinguishable from newvd, and unopenable from now on.
4443 */
4444 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4445 spa_strfree(oldvd->vdev_path);
4446 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4447 KM_SLEEP);
4448 (void) sprintf(oldvd->vdev_path, "%s/%s",
4449 newvd->vdev_path, "old");
4450 if (oldvd->vdev_devid != NULL) {
4451 spa_strfree(oldvd->vdev_devid);
4452 oldvd->vdev_devid = NULL;
4453 }
4454 }
4455
4456 /* mark the device being resilvered */
4457 newvd->vdev_resilver_txg = txg;
4458
4459 /*
4460 * If the parent is not a mirror, or if we're replacing, insert the new
4461 * mirror/replacing/spare vdev above oldvd.
4462 */
4463 if (pvd->vdev_ops != pvops)
4464 pvd = vdev_add_parent(oldvd, pvops);
4465
4466 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4467 ASSERT(pvd->vdev_ops == pvops);
4468 ASSERT(oldvd->vdev_parent == pvd);
4469
4470 /*
4471 * Extract the new device from its root and add it to pvd.
4472 */
4473 vdev_remove_child(newrootvd, newvd);
4474 newvd->vdev_id = pvd->vdev_children;
4475 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4476 vdev_add_child(pvd, newvd);
4477
4478 tvd = newvd->vdev_top;
4479 ASSERT(pvd->vdev_top == tvd);
4480 ASSERT(tvd->vdev_parent == rvd);
4481
4482 vdev_config_dirty(tvd);
4483
4484 /*
4485 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4486 * for any dmu_sync-ed blocks. It will propagate upward when
4487 * spa_vdev_exit() calls vdev_dtl_reassess().
4488 */
4489 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4490
4491 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4492 dtl_max_txg - TXG_INITIAL);
4493
4494 if (newvd->vdev_isspare) {
4495 spa_spare_activate(newvd);
4496 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4497 }
4498
4499 oldvdpath = spa_strdup(oldvd->vdev_path);
4500 newvdpath = spa_strdup(newvd->vdev_path);
4501 newvd_isspare = newvd->vdev_isspare;
4502
4503 /*
4504 * Mark newvd's DTL dirty in this txg.
4505 */
4506 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4507
4508 /*
4509 * Schedule the resilver to restart in the future. We do this to
4510 * ensure that dmu_sync-ed blocks have been stitched into the
4511 * respective datasets.
4512 */
4513 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4514
4515 /*
4516 * Commit the config
4517 */
4518 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4519
4520 spa_history_log_internal(spa, "vdev attach", NULL,
4521 "%s vdev=%s %s vdev=%s",
4522 replacing && newvd_isspare ? "spare in" :
4523 replacing ? "replace" : "attach", newvdpath,
4524 replacing ? "for" : "to", oldvdpath);
4525
4526 spa_strfree(oldvdpath);
4527 spa_strfree(newvdpath);
4528
4529 if (spa->spa_bootfs)
4530 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4531
4532 return (0);
4533 }
4534
4535 /*
4536 * Detach a device from a mirror or replacing vdev.
4537 *
4538 * If 'replace_done' is specified, only detach if the parent
4539 * is a replacing vdev.
4540 */
4541 int
4542 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4543 {
4544 uint64_t txg;
4545 int error;
4546 vdev_t *rvd = spa->spa_root_vdev;
4547 vdev_t *vd, *pvd, *cvd, *tvd;
4548 boolean_t unspare = B_FALSE;
4549 uint64_t unspare_guid = 0;
4550 char *vdpath;
4551
4552 ASSERT(spa_writeable(spa));
4553
4554 txg = spa_vdev_enter(spa);
4555
4556 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4557
4558 if (vd == NULL)
4559 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4560
4561 if (!vd->vdev_ops->vdev_op_leaf)
4562 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4563
4564 pvd = vd->vdev_parent;
4565
4566 /*
4567 * If the parent/child relationship is not as expected, don't do it.
4568 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4569 * vdev that's replacing B with C. The user's intent in replacing
4570 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4571 * the replace by detaching C, the expected behavior is to end up
4572 * M(A,B). But suppose that right after deciding to detach C,
4573 * the replacement of B completes. We would have M(A,C), and then
4574 * ask to detach C, which would leave us with just A -- not what
4575 * the user wanted. To prevent this, we make sure that the
4576 * parent/child relationship hasn't changed -- in this example,
4577 * that C's parent is still the replacing vdev R.
4578 */
4579 if (pvd->vdev_guid != pguid && pguid != 0)
4580 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4581
4582 /*
4583 * Only 'replacing' or 'spare' vdevs can be replaced.
4584 */
4585 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4586 pvd->vdev_ops != &vdev_spare_ops)
4587 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4588
4589 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4590 spa_version(spa) >= SPA_VERSION_SPARES);
4591
4592 /*
4593 * Only mirror, replacing, and spare vdevs support detach.
4594 */
4595 if (pvd->vdev_ops != &vdev_replacing_ops &&
4596 pvd->vdev_ops != &vdev_mirror_ops &&
4597 pvd->vdev_ops != &vdev_spare_ops)
4598 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4599
4600 /*
4601 * If this device has the only valid copy of some data,
4602 * we cannot safely detach it.
4603 */
4604 if (vdev_dtl_required(vd))
4605 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4606
4607 ASSERT(pvd->vdev_children >= 2);
4608
4609 /*
4610 * If we are detaching the second disk from a replacing vdev, then
4611 * check to see if we changed the original vdev's path to have "/old"
4612 * at the end in spa_vdev_attach(). If so, undo that change now.
4613 */
4614 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4615 vd->vdev_path != NULL) {
4616 size_t len = strlen(vd->vdev_path);
4617
4618 for (int c = 0; c < pvd->vdev_children; c++) {
4619 cvd = pvd->vdev_child[c];
4620
4621 if (cvd == vd || cvd->vdev_path == NULL)
4622 continue;
4623
4624 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4625 strcmp(cvd->vdev_path + len, "/old") == 0) {
4626 spa_strfree(cvd->vdev_path);
4627 cvd->vdev_path = spa_strdup(vd->vdev_path);
4628 break;
4629 }
4630 }
4631 }
4632
4633 /*
4634 * If we are detaching the original disk from a spare, then it implies
4635 * that the spare should become a real disk, and be removed from the
4636 * active spare list for the pool.
4637 */
4638 if (pvd->vdev_ops == &vdev_spare_ops &&
4639 vd->vdev_id == 0 &&
4640 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4641 unspare = B_TRUE;
4642
4643 /*
4644 * Erase the disk labels so the disk can be used for other things.
4645 * This must be done after all other error cases are handled,
4646 * but before we disembowel vd (so we can still do I/O to it).
4647 * But if we can't do it, don't treat the error as fatal --
4648 * it may be that the unwritability of the disk is the reason
4649 * it's being detached!
4650 */
4651 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4652
4653 /*
4654 * Remove vd from its parent and compact the parent's children.
4655 */
4656 vdev_remove_child(pvd, vd);
4657 vdev_compact_children(pvd);
4658
4659 /*
4660 * Remember one of the remaining children so we can get tvd below.
4661 */
4662 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4663
4664 /*
4665 * If we need to remove the remaining child from the list of hot spares,
4666 * do it now, marking the vdev as no longer a spare in the process.
4667 * We must do this before vdev_remove_parent(), because that can
4668 * change the GUID if it creates a new toplevel GUID. For a similar
4669 * reason, we must remove the spare now, in the same txg as the detach;
4670 * otherwise someone could attach a new sibling, change the GUID, and
4671 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4672 */
4673 if (unspare) {
4674 ASSERT(cvd->vdev_isspare);
4675 spa_spare_remove(cvd);
4676 unspare_guid = cvd->vdev_guid;
4677 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4678 cvd->vdev_unspare = B_TRUE;
4679 }
4680
4681 /*
4682 * If the parent mirror/replacing vdev only has one child,
4683 * the parent is no longer needed. Remove it from the tree.
4684 */
4685 if (pvd->vdev_children == 1) {
4686 if (pvd->vdev_ops == &vdev_spare_ops)
4687 cvd->vdev_unspare = B_FALSE;
4688 vdev_remove_parent(cvd);
4689 }
4690
4691
4692 /*
4693 * We don't set tvd until now because the parent we just removed
4694 * may have been the previous top-level vdev.
4695 */
4696 tvd = cvd->vdev_top;
4697 ASSERT(tvd->vdev_parent == rvd);
4698
4699 /*
4700 * Reevaluate the parent vdev state.
4701 */
4702 vdev_propagate_state(cvd);
4703
4704 /*
4705 * If the 'autoexpand' property is set on the pool then automatically
4706 * try to expand the size of the pool. For example if the device we
4707 * just detached was smaller than the others, it may be possible to
4708 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4709 * first so that we can obtain the updated sizes of the leaf vdevs.
4710 */
4711 if (spa->spa_autoexpand) {
4712 vdev_reopen(tvd);
4713 vdev_expand(tvd, txg);
4714 }
4715
4716 vdev_config_dirty(tvd);
4717
4718 /*
4719 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4720 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4721 * But first make sure we're not on any *other* txg's DTL list, to
4722 * prevent vd from being accessed after it's freed.
4723 */
4724 vdpath = spa_strdup(vd->vdev_path);
4725 for (int t = 0; t < TXG_SIZE; t++)
4726 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4727 vd->vdev_detached = B_TRUE;
4728 vdev_dirty(tvd, VDD_DTL, vd, txg);
4729
4730 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4731
4732 /* hang on to the spa before we release the lock */
4733 spa_open_ref(spa, FTAG);
4734
4735 error = spa_vdev_exit(spa, vd, txg, 0);
4736
4737 spa_history_log_internal(spa, "detach", NULL,
4738 "vdev=%s", vdpath);
4739 spa_strfree(vdpath);
4740
4741 /*
4742 * If this was the removal of the original device in a hot spare vdev,
4743 * then we want to go through and remove the device from the hot spare
4744 * list of every other pool.
4745 */
4746 if (unspare) {
4747 spa_t *altspa = NULL;
4748
4749 mutex_enter(&spa_namespace_lock);
4750 while ((altspa = spa_next(altspa)) != NULL) {
4751 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4752 altspa == spa)
4753 continue;
4754
4755 spa_open_ref(altspa, FTAG);
4756 mutex_exit(&spa_namespace_lock);
4757 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4758 mutex_enter(&spa_namespace_lock);
4759 spa_close(altspa, FTAG);
4760 }
4761 mutex_exit(&spa_namespace_lock);
4762
4763 /* search the rest of the vdevs for spares to remove */
4764 spa_vdev_resilver_done(spa);
4765 }
4766
4767 /* all done with the spa; OK to release */
4768 mutex_enter(&spa_namespace_lock);
4769 spa_close(spa, FTAG);
4770 mutex_exit(&spa_namespace_lock);
4771
4772 return (error);
4773 }
4774
4775 /*
4776 * Split a set of devices from their mirrors, and create a new pool from them.
4777 */
4778 int
4779 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4780 nvlist_t *props, boolean_t exp)
4781 {
4782 int error = 0;
4783 uint64_t txg, *glist;
4784 spa_t *newspa;
4785 uint_t c, children, lastlog;
4786 nvlist_t **child, *nvl, *tmp;
4787 dmu_tx_t *tx;
4788 char *altroot = NULL;
4789 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4790 boolean_t activate_slog;
4791
4792 ASSERT(spa_writeable(spa));
4793
4794 txg = spa_vdev_enter(spa);
4795
4796 /* clear the log and flush everything up to now */
4797 activate_slog = spa_passivate_log(spa);
4798 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4799 error = spa_offline_log(spa);
4800 txg = spa_vdev_config_enter(spa);
4801
4802 if (activate_slog)
4803 spa_activate_log(spa);
4804
4805 if (error != 0)
4806 return (spa_vdev_exit(spa, NULL, txg, error));
4807
4808 /* check new spa name before going any further */
4809 if (spa_lookup(newname) != NULL)
4810 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4811
4812 /*
4813 * scan through all the children to ensure they're all mirrors
4814 */
4815 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4816 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4817 &children) != 0)
4818 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4819
4820 /* first, check to ensure we've got the right child count */
4821 rvd = spa->spa_root_vdev;
4822 lastlog = 0;
4823 for (c = 0; c < rvd->vdev_children; c++) {
4824 vdev_t *vd = rvd->vdev_child[c];
4825
4826 /* don't count the holes & logs as children */
4827 if (vd->vdev_islog || vd->vdev_ishole) {
4828 if (lastlog == 0)
4829 lastlog = c;
4830 continue;
4831 }
4832
4833 lastlog = 0;
4834 }
4835 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4836 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4837
4838 /* next, ensure no spare or cache devices are part of the split */
4839 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4840 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4841 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4842
4843 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4844 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4845
4846 /* then, loop over each vdev and validate it */
4847 for (c = 0; c < children; c++) {
4848 uint64_t is_hole = 0;
4849
4850 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4851 &is_hole);
4852
4853 if (is_hole != 0) {
4854 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4855 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4856 continue;
4857 } else {
4858 error = SET_ERROR(EINVAL);
4859 break;
4860 }
4861 }
4862
4863 /* which disk is going to be split? */
4864 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4865 &glist[c]) != 0) {
4866 error = SET_ERROR(EINVAL);
4867 break;
4868 }
4869
4870 /* look it up in the spa */
4871 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4872 if (vml[c] == NULL) {
4873 error = SET_ERROR(ENODEV);
4874 break;
4875 }
4876
4877 /* make sure there's nothing stopping the split */
4878 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4879 vml[c]->vdev_islog ||
4880 vml[c]->vdev_ishole ||
4881 vml[c]->vdev_isspare ||
4882 vml[c]->vdev_isl2cache ||
4883 !vdev_writeable(vml[c]) ||
4884 vml[c]->vdev_children != 0 ||
4885 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4886 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4887 error = SET_ERROR(EINVAL);
4888 break;
4889 }
4890
4891 if (vdev_dtl_required(vml[c])) {
4892 error = SET_ERROR(EBUSY);
4893 break;
4894 }
4895
4896 /* we need certain info from the top level */
4897 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4898 vml[c]->vdev_top->vdev_ms_array) == 0);
4899 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4900 vml[c]->vdev_top->vdev_ms_shift) == 0);
4901 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4902 vml[c]->vdev_top->vdev_asize) == 0);
4903 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4904 vml[c]->vdev_top->vdev_ashift) == 0);
4905 }
4906
4907 if (error != 0) {
4908 kmem_free(vml, children * sizeof (vdev_t *));
4909 kmem_free(glist, children * sizeof (uint64_t));
4910 return (spa_vdev_exit(spa, NULL, txg, error));
4911 }
4912
4913 /* stop writers from using the disks */
4914 for (c = 0; c < children; c++) {
4915 if (vml[c] != NULL)
4916 vml[c]->vdev_offline = B_TRUE;
4917 }
4918 vdev_reopen(spa->spa_root_vdev);
4919
4920 /*
4921 * Temporarily record the splitting vdevs in the spa config. This
4922 * will disappear once the config is regenerated.
4923 */
4924 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4925 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4926 glist, children) == 0);
4927 kmem_free(glist, children * sizeof (uint64_t));
4928
4929 mutex_enter(&spa->spa_props_lock);
4930 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4931 nvl) == 0);
4932 mutex_exit(&spa->spa_props_lock);
4933 spa->spa_config_splitting = nvl;
4934 vdev_config_dirty(spa->spa_root_vdev);
4935
4936 /* configure and create the new pool */
4937 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4938 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4939 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4940 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4941 spa_version(spa)) == 0);
4942 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4943 spa->spa_config_txg) == 0);
4944 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4945 spa_generate_guid(NULL)) == 0);
4946 (void) nvlist_lookup_string(props,
4947 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4948
4949 /* add the new pool to the namespace */
4950 newspa = spa_add(newname, config, altroot);
4951 newspa->spa_config_txg = spa->spa_config_txg;
4952 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4953
4954 /* release the spa config lock, retaining the namespace lock */
4955 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4956
4957 if (zio_injection_enabled)
4958 zio_handle_panic_injection(spa, FTAG, 1);
4959
4960 spa_activate(newspa, spa_mode_global);
4961 spa_async_suspend(newspa);
4962
4963 /* create the new pool from the disks of the original pool */
4964 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4965 if (error)
4966 goto out;
4967
4968 /* if that worked, generate a real config for the new pool */
4969 if (newspa->spa_root_vdev != NULL) {
4970 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4971 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4972 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4973 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4974 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4975 B_TRUE));
4976 }
4977
4978 /* set the props */
4979 if (props != NULL) {
4980 spa_configfile_set(newspa, props, B_FALSE);
4981 error = spa_prop_set(newspa, props);
4982 if (error)
4983 goto out;
4984 }
4985
4986 /* flush everything */
4987 txg = spa_vdev_config_enter(newspa);
4988 vdev_config_dirty(newspa->spa_root_vdev);
4989 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4990
4991 if (zio_injection_enabled)
4992 zio_handle_panic_injection(spa, FTAG, 2);
4993
4994 spa_async_resume(newspa);
4995
4996 /* finally, update the original pool's config */
4997 txg = spa_vdev_config_enter(spa);
4998 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4999 error = dmu_tx_assign(tx, TXG_WAIT);
5000 if (error != 0)
5001 dmu_tx_abort(tx);
5002 for (c = 0; c < children; c++) {
5003 if (vml[c] != NULL) {
5004 vdev_split(vml[c]);
5005 if (error == 0)
5006 spa_history_log_internal(spa, "detach", tx,
5007 "vdev=%s", vml[c]->vdev_path);
5008 vdev_free(vml[c]);
5009 }
5010 }
5011 vdev_config_dirty(spa->spa_root_vdev);
5012 spa->spa_config_splitting = NULL;
5013 nvlist_free(nvl);
5014 if (error == 0)
5015 dmu_tx_commit(tx);
5016 (void) spa_vdev_exit(spa, NULL, txg, 0);
5017
5018 if (zio_injection_enabled)
5019 zio_handle_panic_injection(spa, FTAG, 3);
5020
5021 /* split is complete; log a history record */
5022 spa_history_log_internal(newspa, "split", NULL,
5023 "from pool %s", spa_name(spa));
5024
5025 kmem_free(vml, children * sizeof (vdev_t *));
5026
5027 /* if we're not going to mount the filesystems in userland, export */
5028 if (exp)
5029 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5030 B_FALSE, B_FALSE);
5031
5032 return (error);
5033
5034 out:
5035 spa_unload(newspa);
5036 spa_deactivate(newspa);
5037 spa_remove(newspa);
5038
5039 txg = spa_vdev_config_enter(spa);
5040
5041 /* re-online all offlined disks */
5042 for (c = 0; c < children; c++) {
5043 if (vml[c] != NULL)
5044 vml[c]->vdev_offline = B_FALSE;
5045 }
5046 vdev_reopen(spa->spa_root_vdev);
5047
5048 nvlist_free(spa->spa_config_splitting);
5049 spa->spa_config_splitting = NULL;
5050 (void) spa_vdev_exit(spa, NULL, txg, error);
5051
5052 kmem_free(vml, children * sizeof (vdev_t *));
5053 return (error);
5054 }
5055
5056 static nvlist_t *
5057 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5058 {
5059 for (int i = 0; i < count; i++) {
5060 uint64_t guid;
5061
5062 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5063 &guid) == 0);
5064
5065 if (guid == target_guid)
5066 return (nvpp[i]);
5067 }
5068
5069 return (NULL);
5070 }
5071
5072 static void
5073 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5074 nvlist_t *dev_to_remove)
5075 {
5076 nvlist_t **newdev = NULL;
5077
5078 if (count > 1)
5079 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5080
5081 for (int i = 0, j = 0; i < count; i++) {
5082 if (dev[i] == dev_to_remove)
5083 continue;
5084 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5085 }
5086
5087 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5088 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5089
5090 for (int i = 0; i < count - 1; i++)
5091 nvlist_free(newdev[i]);
5092
5093 if (count > 1)
5094 kmem_free(newdev, (count - 1) * sizeof (void *));
5095 }
5096
5097 /*
5098 * Evacuate the device.
5099 */
5100 static int
5101 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5102 {
5103 uint64_t txg;
5104 int error = 0;
5105
5106 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5107 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5108 ASSERT(vd == vd->vdev_top);
5109
5110 /*
5111 * Evacuate the device. We don't hold the config lock as writer
5112 * since we need to do I/O but we do keep the
5113 * spa_namespace_lock held. Once this completes the device
5114 * should no longer have any blocks allocated on it.
5115 */
5116 if (vd->vdev_islog) {
5117 if (vd->vdev_stat.vs_alloc != 0)
5118 error = spa_offline_log(spa);
5119 } else {
5120 error = SET_ERROR(ENOTSUP);
5121 }
5122
5123 if (error)
5124 return (error);
5125
5126 /*
5127 * The evacuation succeeded. Remove any remaining MOS metadata
5128 * associated with this vdev, and wait for these changes to sync.
5129 */
5130 ASSERT0(vd->vdev_stat.vs_alloc);
5131 txg = spa_vdev_config_enter(spa);
5132 vd->vdev_removing = B_TRUE;
5133 vdev_dirty_leaves(vd, VDD_DTL, txg);
5134 vdev_config_dirty(vd);
5135 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5136
5137 return (0);
5138 }
5139
5140 /*
5141 * Complete the removal by cleaning up the namespace.
5142 */
5143 static void
5144 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5145 {
5146 vdev_t *rvd = spa->spa_root_vdev;
5147 uint64_t id = vd->vdev_id;
5148 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5149
5150 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5151 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5152 ASSERT(vd == vd->vdev_top);
5153
5154 /*
5155 * Only remove any devices which are empty.
5156 */
5157 if (vd->vdev_stat.vs_alloc != 0)
5158 return;
5159
5160 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5161
5162 if (list_link_active(&vd->vdev_state_dirty_node))
5163 vdev_state_clean(vd);
5164 if (list_link_active(&vd->vdev_config_dirty_node))
5165 vdev_config_clean(vd);
5166
5167 vdev_free(vd);
5168
5169 if (last_vdev) {
5170 vdev_compact_children(rvd);
5171 } else {
5172 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5173 vdev_add_child(rvd, vd);
5174 }
5175 vdev_config_dirty(rvd);
5176
5177 /*
5178 * Reassess the health of our root vdev.
5179 */
5180 vdev_reopen(rvd);
5181 }
5182
5183 /*
5184 * Remove a device from the pool -
5185 *
5186 * Removing a device from the vdev namespace requires several steps
5187 * and can take a significant amount of time. As a result we use
5188 * the spa_vdev_config_[enter/exit] functions which allow us to
5189 * grab and release the spa_config_lock while still holding the namespace
5190 * lock. During each step the configuration is synced out.
5191 *
5192 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5193 * devices.
5194 */
5195 int
5196 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5197 {
5198 vdev_t *vd;
5199 metaslab_group_t *mg;
5200 nvlist_t **spares, **l2cache, *nv;
5201 uint64_t txg = 0;
5202 uint_t nspares, nl2cache;
5203 int error = 0;
5204 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5205
5206 ASSERT(spa_writeable(spa));
5207
5208 if (!locked)
5209 txg = spa_vdev_enter(spa);
5210
5211 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5212
5213 if (spa->spa_spares.sav_vdevs != NULL &&
5214 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5215 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5216 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5217 /*
5218 * Only remove the hot spare if it's not currently in use
5219 * in this pool.
5220 */
5221 if (vd == NULL || unspare) {
5222 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5223 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5224 spa_load_spares(spa);
5225 spa->spa_spares.sav_sync = B_TRUE;
5226 } else {
5227 error = SET_ERROR(EBUSY);
5228 }
5229 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5230 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5231 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5232 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5233 /*
5234 * Cache devices can always be removed.
5235 */
5236 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5237 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5238 spa_load_l2cache(spa);
5239 spa->spa_l2cache.sav_sync = B_TRUE;
5240 } else if (vd != NULL && vd->vdev_islog) {
5241 ASSERT(!locked);
5242 ASSERT(vd == vd->vdev_top);
5243
5244 /*
5245 * XXX - Once we have bp-rewrite this should
5246 * become the common case.
5247 */
5248
5249 mg = vd->vdev_mg;
5250
5251 /*
5252 * Stop allocating from this vdev.
5253 */
5254 metaslab_group_passivate(mg);
5255
5256 /*
5257 * Wait for the youngest allocations and frees to sync,
5258 * and then wait for the deferral of those frees to finish.
5259 */
5260 spa_vdev_config_exit(spa, NULL,
5261 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5262
5263 /*
5264 * Attempt to evacuate the vdev.
5265 */
5266 error = spa_vdev_remove_evacuate(spa, vd);
5267
5268 txg = spa_vdev_config_enter(spa);
5269
5270 /*
5271 * If we couldn't evacuate the vdev, unwind.
5272 */
5273 if (error) {
5274 metaslab_group_activate(mg);
5275 return (spa_vdev_exit(spa, NULL, txg, error));
5276 }
5277
5278 /*
5279 * Clean up the vdev namespace.
5280 */
5281 spa_vdev_remove_from_namespace(spa, vd);
5282
5283 } else if (vd != NULL) {
5284 /*
5285 * Normal vdevs cannot be removed (yet).
5286 */
5287 error = SET_ERROR(ENOTSUP);
5288 } else {
5289 /*
5290 * There is no vdev of any kind with the specified guid.
5291 */
5292 error = SET_ERROR(ENOENT);
5293 }
5294
5295 if (!locked)
5296 return (spa_vdev_exit(spa, NULL, txg, error));
5297
5298 return (error);
5299 }
5300
5301 /*
5302 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5303 * currently spared, so we can detach it.
5304 */
5305 static vdev_t *
5306 spa_vdev_resilver_done_hunt(vdev_t *vd)
5307 {
5308 vdev_t *newvd, *oldvd;
5309
5310 for (int c = 0; c < vd->vdev_children; c++) {
5311 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5312 if (oldvd != NULL)
5313 return (oldvd);
5314 }
5315
5316 /*
5317 * Check for a completed replacement. We always consider the first
5318 * vdev in the list to be the oldest vdev, and the last one to be
5319 * the newest (see spa_vdev_attach() for how that works). In
5320 * the case where the newest vdev is faulted, we will not automatically
5321 * remove it after a resilver completes. This is OK as it will require
5322 * user intervention to determine which disk the admin wishes to keep.
5323 */
5324 if (vd->vdev_ops == &vdev_replacing_ops) {
5325 ASSERT(vd->vdev_children > 1);
5326
5327 newvd = vd->vdev_child[vd->vdev_children - 1];
5328 oldvd = vd->vdev_child[0];
5329
5330 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5331 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5332 !vdev_dtl_required(oldvd))
5333 return (oldvd);
5334 }
5335
5336 /*
5337 * Check for a completed resilver with the 'unspare' flag set.
5338 */
5339 if (vd->vdev_ops == &vdev_spare_ops) {
5340 vdev_t *first = vd->vdev_child[0];
5341 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5342
5343 if (last->vdev_unspare) {
5344 oldvd = first;
5345 newvd = last;
5346 } else if (first->vdev_unspare) {
5347 oldvd = last;
5348 newvd = first;
5349 } else {
5350 oldvd = NULL;
5351 }
5352
5353 if (oldvd != NULL &&
5354 vdev_dtl_empty(newvd, DTL_MISSING) &&
5355 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5356 !vdev_dtl_required(oldvd))
5357 return (oldvd);
5358
5359 /*
5360 * If there are more than two spares attached to a disk,
5361 * and those spares are not required, then we want to
5362 * attempt to free them up now so that they can be used
5363 * by other pools. Once we're back down to a single
5364 * disk+spare, we stop removing them.
5365 */
5366 if (vd->vdev_children > 2) {
5367 newvd = vd->vdev_child[1];
5368
5369 if (newvd->vdev_isspare && last->vdev_isspare &&
5370 vdev_dtl_empty(last, DTL_MISSING) &&
5371 vdev_dtl_empty(last, DTL_OUTAGE) &&
5372 !vdev_dtl_required(newvd))
5373 return (newvd);
5374 }
5375 }
5376
5377 return (NULL);
5378 }
5379
5380 static void
5381 spa_vdev_resilver_done(spa_t *spa)
5382 {
5383 vdev_t *vd, *pvd, *ppvd;
5384 uint64_t guid, sguid, pguid, ppguid;
5385
5386 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5387
5388 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5389 pvd = vd->vdev_parent;
5390 ppvd = pvd->vdev_parent;
5391 guid = vd->vdev_guid;
5392 pguid = pvd->vdev_guid;
5393 ppguid = ppvd->vdev_guid;
5394 sguid = 0;
5395 /*
5396 * If we have just finished replacing a hot spared device, then
5397 * we need to detach the parent's first child (the original hot
5398 * spare) as well.
5399 */
5400 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5401 ppvd->vdev_children == 2) {
5402 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5403 sguid = ppvd->vdev_child[1]->vdev_guid;
5404 }
5405 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5406
5407 spa_config_exit(spa, SCL_ALL, FTAG);
5408 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5409 return;
5410 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5411 return;
5412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5413 }
5414
5415 spa_config_exit(spa, SCL_ALL, FTAG);
5416 }
5417
5418 /*
5419 * Update the stored path or FRU for this vdev.
5420 */
5421 int
5422 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5423 boolean_t ispath)
5424 {
5425 vdev_t *vd;
5426 boolean_t sync = B_FALSE;
5427
5428 ASSERT(spa_writeable(spa));
5429
5430 spa_vdev_state_enter(spa, SCL_ALL);
5431
5432 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5433 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5434
5435 if (!vd->vdev_ops->vdev_op_leaf)
5436 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5437
5438 if (ispath) {
5439 if (strcmp(value, vd->vdev_path) != 0) {
5440 spa_strfree(vd->vdev_path);
5441 vd->vdev_path = spa_strdup(value);
5442 sync = B_TRUE;
5443 }
5444 } else {
5445 if (vd->vdev_fru == NULL) {
5446 vd->vdev_fru = spa_strdup(value);
5447 sync = B_TRUE;
5448 } else if (strcmp(value, vd->vdev_fru) != 0) {
5449 spa_strfree(vd->vdev_fru);
5450 vd->vdev_fru = spa_strdup(value);
5451 sync = B_TRUE;
5452 }
5453 }
5454
5455 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5456 }
5457
5458 int
5459 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5460 {
5461 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5462 }
5463
5464 int
5465 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5466 {
5467 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5468 }
5469
5470 /*
5471 * ==========================================================================
5472 * SPA Scanning
5473 * ==========================================================================
5474 */
5475
5476 int
5477 spa_scan_stop(spa_t *spa)
5478 {
5479 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5480 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5481 return (SET_ERROR(EBUSY));
5482 return (dsl_scan_cancel(spa->spa_dsl_pool));
5483 }
5484
5485 int
5486 spa_scan(spa_t *spa, pool_scan_func_t func)
5487 {
5488 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5489
5490 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5491 return (SET_ERROR(ENOTSUP));
5492
5493 /*
5494 * If a resilver was requested, but there is no DTL on a
5495 * writeable leaf device, we have nothing to do.
5496 */
5497 if (func == POOL_SCAN_RESILVER &&
5498 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5499 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5500 return (0);
5501 }
5502
5503 return (dsl_scan(spa->spa_dsl_pool, func));
5504 }
5505
5506 /*
5507 * ==========================================================================
5508 * SPA async task processing
5509 * ==========================================================================
5510 */
5511
5512 static void
5513 spa_async_remove(spa_t *spa, vdev_t *vd)
5514 {
5515 if (vd->vdev_remove_wanted) {
5516 vd->vdev_remove_wanted = B_FALSE;
5517 vd->vdev_delayed_close = B_FALSE;
5518 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5519
5520 /*
5521 * We want to clear the stats, but we don't want to do a full
5522 * vdev_clear() as that will cause us to throw away
5523 * degraded/faulted state as well as attempt to reopen the
5524 * device, all of which is a waste.
5525 */
5526 vd->vdev_stat.vs_read_errors = 0;
5527 vd->vdev_stat.vs_write_errors = 0;
5528 vd->vdev_stat.vs_checksum_errors = 0;
5529
5530 vdev_state_dirty(vd->vdev_top);
5531 }
5532
5533 for (int c = 0; c < vd->vdev_children; c++)
5534 spa_async_remove(spa, vd->vdev_child[c]);
5535 }
5536
5537 static void
5538 spa_async_probe(spa_t *spa, vdev_t *vd)
5539 {
5540 if (vd->vdev_probe_wanted) {
5541 vd->vdev_probe_wanted = B_FALSE;
5542 vdev_reopen(vd); /* vdev_open() does the actual probe */
5543 }
5544
5545 for (int c = 0; c < vd->vdev_children; c++)
5546 spa_async_probe(spa, vd->vdev_child[c]);
5547 }
5548
5549 static void
5550 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5551 {
5552 sysevent_id_t eid;
5553 nvlist_t *attr;
5554 char *physpath;
5555
5556 if (!spa->spa_autoexpand)
5557 return;
5558
5559 for (int c = 0; c < vd->vdev_children; c++) {
5560 vdev_t *cvd = vd->vdev_child[c];
5561 spa_async_autoexpand(spa, cvd);
5562 }
5563
5564 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5565 return;
5566
5567 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5568 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5569
5570 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5571 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5572
5573 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5574 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5575
5576 nvlist_free(attr);
5577 kmem_free(physpath, MAXPATHLEN);
5578 }
5579
5580 static void
5581 spa_async_thread(spa_t *spa)
5582 {
5583 int tasks;
5584
5585 ASSERT(spa->spa_sync_on);
5586
5587 mutex_enter(&spa->spa_async_lock);
5588 tasks = spa->spa_async_tasks;
5589 spa->spa_async_tasks = 0;
5590 mutex_exit(&spa->spa_async_lock);
5591
5592 /*
5593 * See if the config needs to be updated.
5594 */
5595 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5596 uint64_t old_space, new_space;
5597
5598 mutex_enter(&spa_namespace_lock);
5599 old_space = metaslab_class_get_space(spa_normal_class(spa));
5600 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5601 new_space = metaslab_class_get_space(spa_normal_class(spa));
5602 mutex_exit(&spa_namespace_lock);
5603
5604 /*
5605 * If the pool grew as a result of the config update,
5606 * then log an internal history event.
5607 */
5608 if (new_space != old_space) {
5609 spa_history_log_internal(spa, "vdev online", NULL,
5610 "pool '%s' size: %llu(+%llu)",
5611 spa_name(spa), new_space, new_space - old_space);
5612 }
5613 }
5614
5615 /*
5616 * See if any devices need to be marked REMOVED.
5617 */
5618 if (tasks & SPA_ASYNC_REMOVE) {
5619 spa_vdev_state_enter(spa, SCL_NONE);
5620 spa_async_remove(spa, spa->spa_root_vdev);
5621 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5622 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5623 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5624 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5625 (void) spa_vdev_state_exit(spa, NULL, 0);
5626 }
5627
5628 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5629 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5630 spa_async_autoexpand(spa, spa->spa_root_vdev);
5631 spa_config_exit(spa, SCL_CONFIG, FTAG);
5632 }
5633
5634 /*
5635 * See if any devices need to be probed.
5636 */
5637 if (tasks & SPA_ASYNC_PROBE) {
5638 spa_vdev_state_enter(spa, SCL_NONE);
5639 spa_async_probe(spa, spa->spa_root_vdev);
5640 (void) spa_vdev_state_exit(spa, NULL, 0);
5641 }
5642
5643 /*
5644 * If any devices are done replacing, detach them.
5645 */
5646 if (tasks & SPA_ASYNC_RESILVER_DONE)
5647 spa_vdev_resilver_done(spa);
5648
5649 /*
5650 * Kick off a resilver.
5651 */
5652 if (tasks & SPA_ASYNC_RESILVER)
5653 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5654
5655 /*
5656 * Kick off L2 cache rebuilding.
5657 */
5658 if (tasks & SPA_ASYNC_L2CACHE_REBUILD)
5659 l2arc_spa_rebuild_start(spa);
5660
5661 /*
5662 * Let the world know that we're done.
5663 */
5664 mutex_enter(&spa->spa_async_lock);
5665 spa->spa_async_thread = NULL;
5666 cv_broadcast(&spa->spa_async_cv);
5667 mutex_exit(&spa->spa_async_lock);
5668 thread_exit();
5669 }
5670
5671 void
5672 spa_async_suspend(spa_t *spa)
5673 {
5674 mutex_enter(&spa->spa_async_lock);
5675 spa->spa_async_suspended++;
5676 while (spa->spa_async_thread != NULL)
5677 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5678 mutex_exit(&spa->spa_async_lock);
5679 }
5680
5681 void
5682 spa_async_resume(spa_t *spa)
5683 {
5684 mutex_enter(&spa->spa_async_lock);
5685 ASSERT(spa->spa_async_suspended != 0);
5686 spa->spa_async_suspended--;
5687 mutex_exit(&spa->spa_async_lock);
5688 }
5689
5690 static boolean_t
5691 spa_async_tasks_pending(spa_t *spa)
5692 {
5693 uint_t non_config_tasks;
5694 uint_t config_task;
5695 boolean_t config_task_suspended;
5696
5697 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5698 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5699 if (spa->spa_ccw_fail_time == 0) {
5700 config_task_suspended = B_FALSE;
5701 } else {
5702 config_task_suspended =
5703 (gethrtime() - spa->spa_ccw_fail_time) <
5704 (zfs_ccw_retry_interval * NANOSEC);
5705 }
5706
5707 return (non_config_tasks || (config_task && !config_task_suspended));
5708 }
5709
5710 static void
5711 spa_async_dispatch(spa_t *spa)
5712 {
5713 mutex_enter(&spa->spa_async_lock);
5714 if (spa_async_tasks_pending(spa) &&
5715 !spa->spa_async_suspended &&
5716 spa->spa_async_thread == NULL &&
5717 rootdir != NULL)
5718 spa->spa_async_thread = thread_create(NULL, 0,
5719 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5720 mutex_exit(&spa->spa_async_lock);
5721 }
5722
5723 void
5724 spa_async_request(spa_t *spa, int task)
5725 {
5726 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5727 mutex_enter(&spa->spa_async_lock);
5728 spa->spa_async_tasks |= task;
5729 mutex_exit(&spa->spa_async_lock);
5730 }
5731
5732 /*
5733 * ==========================================================================
5734 * SPA syncing routines
5735 * ==========================================================================
5736 */
5737
5738 static int
5739 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5740 {
5741 bpobj_t *bpo = arg;
5742 bpobj_enqueue(bpo, bp, tx);
5743 return (0);
5744 }
5745
5746 static int
5747 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5748 {
5749 zio_t *zio = arg;
5750
5751 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5752 zio->io_flags));
5753 return (0);
5754 }
5755
5756 /*
5757 * Note: this simple function is not inlined to make it easier to dtrace the
5758 * amount of time spent syncing frees.
5759 */
5760 static void
5761 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5762 {
5763 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5764 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5765 VERIFY(zio_wait(zio) == 0);
5766 }
5767
5768 /*
5769 * Note: this simple function is not inlined to make it easier to dtrace the
5770 * amount of time spent syncing deferred frees.
5771 */
5772 static void
5773 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5774 {
5775 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5776 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5777 spa_free_sync_cb, zio, tx), ==, 0);
5778 VERIFY0(zio_wait(zio));
5779 }
5780
5781
5782 static void
5783 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5784 {
5785 char *packed = NULL;
5786 size_t bufsize;
5787 size_t nvsize = 0;
5788 dmu_buf_t *db;
5789
5790 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5791
5792 /*
5793 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5794 * information. This avoids the dbuf_will_dirty() path and
5795 * saves us a pre-read to get data we don't actually care about.
5796 */
5797 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5798 packed = kmem_alloc(bufsize, KM_SLEEP);
5799
5800 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5801 KM_SLEEP) == 0);
5802 bzero(packed + nvsize, bufsize - nvsize);
5803
5804 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5805
5806 kmem_free(packed, bufsize);
5807
5808 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5809 dmu_buf_will_dirty(db, tx);
5810 *(uint64_t *)db->db_data = nvsize;
5811 dmu_buf_rele(db, FTAG);
5812 }
5813
5814 static void
5815 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5816 const char *config, const char *entry)
5817 {
5818 nvlist_t *nvroot;
5819 nvlist_t **list;
5820 int i;
5821
5822 if (!sav->sav_sync)
5823 return;
5824
5825 /*
5826 * Update the MOS nvlist describing the list of available devices.
5827 * spa_validate_aux() will have already made sure this nvlist is
5828 * valid and the vdevs are labeled appropriately.
5829 */
5830 if (sav->sav_object == 0) {
5831 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5832 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5833 sizeof (uint64_t), tx);
5834 VERIFY(zap_update(spa->spa_meta_objset,
5835 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5836 &sav->sav_object, tx) == 0);
5837 }
5838
5839 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5840 if (sav->sav_count == 0) {
5841 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5842 } else {
5843 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5844 for (i = 0; i < sav->sav_count; i++)
5845 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5846 B_FALSE, VDEV_CONFIG_L2CACHE);
5847 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5848 sav->sav_count) == 0);
5849 for (i = 0; i < sav->sav_count; i++)
5850 nvlist_free(list[i]);
5851 kmem_free(list, sav->sav_count * sizeof (void *));
5852 }
5853
5854 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5855 nvlist_free(nvroot);
5856
5857 sav->sav_sync = B_FALSE;
5858 }
5859
5860 static void
5861 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5862 {
5863 nvlist_t *config;
5864
5865 if (list_is_empty(&spa->spa_config_dirty_list))
5866 return;
5867
5868 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5869
5870 config = spa_config_generate(spa, spa->spa_root_vdev,
5871 dmu_tx_get_txg(tx), B_FALSE);
5872
5873 /*
5874 * If we're upgrading the spa version then make sure that
5875 * the config object gets updated with the correct version.
5876 */
5877 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5878 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5879 spa->spa_uberblock.ub_version);
5880
5881 spa_config_exit(spa, SCL_STATE, FTAG);
5882
5883 if (spa->spa_config_syncing)
5884 nvlist_free(spa->spa_config_syncing);
5885 spa->spa_config_syncing = config;
5886
5887 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5888 }
5889
5890 static void
5891 spa_sync_version(void *arg, dmu_tx_t *tx)
5892 {
5893 uint64_t *versionp = arg;
5894 uint64_t version = *versionp;
5895 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5896
5897 /*
5898 * Setting the version is special cased when first creating the pool.
5899 */
5900 ASSERT(tx->tx_txg != TXG_INITIAL);
5901
5902 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5903 ASSERT(version >= spa_version(spa));
5904
5905 spa->spa_uberblock.ub_version = version;
5906 vdev_config_dirty(spa->spa_root_vdev);
5907 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5908 }
5909
5910 /*
5911 * Set zpool properties.
5912 */
5913 static void
5914 spa_sync_props(void *arg, dmu_tx_t *tx)
5915 {
5916 nvlist_t *nvp = arg;
5917 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5918 objset_t *mos = spa->spa_meta_objset;
5919 nvpair_t *elem = NULL;
5920
5921 mutex_enter(&spa->spa_props_lock);
5922
5923 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5924 uint64_t intval;
5925 char *strval, *fname;
5926 zpool_prop_t prop;
5927 const char *propname;
5928 zprop_type_t proptype;
5929 spa_feature_t fid;
5930
5931 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5932 case ZPROP_INVAL:
5933 /*
5934 * We checked this earlier in spa_prop_validate().
5935 */
5936 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5937
5938 fname = strchr(nvpair_name(elem), '@') + 1;
5939 VERIFY0(zfeature_lookup_name(fname, &fid));
5940
5941 spa_feature_enable(spa, fid, tx);
5942 spa_history_log_internal(spa, "set", tx,
5943 "%s=enabled", nvpair_name(elem));
5944 break;
5945
5946 case ZPOOL_PROP_VERSION:
5947 intval = fnvpair_value_uint64(elem);
5948 /*
5949 * The version is synced seperatly before other
5950 * properties and should be correct by now.
5951 */
5952 ASSERT3U(spa_version(spa), >=, intval);
5953 break;
5954
5955 case ZPOOL_PROP_ALTROOT:
5956 /*
5957 * 'altroot' is a non-persistent property. It should
5958 * have been set temporarily at creation or import time.
5959 */
5960 ASSERT(spa->spa_root != NULL);
5961 break;
5962
5963 case ZPOOL_PROP_READONLY:
5964 case ZPOOL_PROP_CACHEFILE:
5965 /*
5966 * 'readonly' and 'cachefile' are also non-persisitent
5967 * properties.
5968 */
5969 break;
5970 case ZPOOL_PROP_COMMENT:
5971 strval = fnvpair_value_string(elem);
5972 if (spa->spa_comment != NULL)
5973 spa_strfree(spa->spa_comment);
5974 spa->spa_comment = spa_strdup(strval);
5975 /*
5976 * We need to dirty the configuration on all the vdevs
5977 * so that their labels get updated. It's unnecessary
5978 * to do this for pool creation since the vdev's
5979 * configuratoin has already been dirtied.
5980 */
5981 if (tx->tx_txg != TXG_INITIAL)
5982 vdev_config_dirty(spa->spa_root_vdev);
5983 spa_history_log_internal(spa, "set", tx,
5984 "%s=%s", nvpair_name(elem), strval);
5985 break;
5986 default:
5987 /*
5988 * Set pool property values in the poolprops mos object.
5989 */
5990 if (spa->spa_pool_props_object == 0) {
5991 spa->spa_pool_props_object =
5992 zap_create_link(mos, DMU_OT_POOL_PROPS,
5993 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5994 tx);
5995 }
5996
5997 /* normalize the property name */
5998 propname = zpool_prop_to_name(prop);
5999 proptype = zpool_prop_get_type(prop);
6000
6001 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6002 ASSERT(proptype == PROP_TYPE_STRING);
6003 strval = fnvpair_value_string(elem);
6004 VERIFY0(zap_update(mos,
6005 spa->spa_pool_props_object, propname,
6006 1, strlen(strval) + 1, strval, tx));
6007 spa_history_log_internal(spa, "set", tx,
6008 "%s=%s", nvpair_name(elem), strval);
6009 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6010 intval = fnvpair_value_uint64(elem);
6011
6012 if (proptype == PROP_TYPE_INDEX) {
6013 const char *unused;
6014 VERIFY0(zpool_prop_index_to_string(
6015 prop, intval, &unused));
6016 }
6017 VERIFY0(zap_update(mos,
6018 spa->spa_pool_props_object, propname,
6019 8, 1, &intval, tx));
6020 spa_history_log_internal(spa, "set", tx,
6021 "%s=%lld", nvpair_name(elem), intval);
6022 } else {
6023 ASSERT(0); /* not allowed */
6024 }
6025
6026 switch (prop) {
6027 case ZPOOL_PROP_DELEGATION:
6028 spa->spa_delegation = intval;
6029 break;
6030 case ZPOOL_PROP_BOOTFS:
6031 spa->spa_bootfs = intval;
6032 break;
6033 case ZPOOL_PROP_FAILUREMODE:
6034 spa->spa_failmode = intval;
6035 break;
6036 case ZPOOL_PROP_AUTOEXPAND:
6037 spa->spa_autoexpand = intval;
6038 if (tx->tx_txg != TXG_INITIAL)
6039 spa_async_request(spa,
6040 SPA_ASYNC_AUTOEXPAND);
6041 break;
6042 case ZPOOL_PROP_DEDUPDITTO:
6043 spa->spa_dedup_ditto = intval;
6044 break;
6045 default:
6046 break;
6047 }
6048 }
6049
6050 }
6051
6052 mutex_exit(&spa->spa_props_lock);
6053 }
6054
6055 /*
6056 * Perform one-time upgrade on-disk changes. spa_version() does not
6057 * reflect the new version this txg, so there must be no changes this
6058 * txg to anything that the upgrade code depends on after it executes.
6059 * Therefore this must be called after dsl_pool_sync() does the sync
6060 * tasks.
6061 */
6062 static void
6063 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6064 {
6065 dsl_pool_t *dp = spa->spa_dsl_pool;
6066
6067 ASSERT(spa->spa_sync_pass == 1);
6068
6069 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6070
6071 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6072 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6073 dsl_pool_create_origin(dp, tx);
6074
6075 /* Keeping the origin open increases spa_minref */
6076 spa->spa_minref += 3;
6077 }
6078
6079 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6080 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6081 dsl_pool_upgrade_clones(dp, tx);
6082 }
6083
6084 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6085 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6086 dsl_pool_upgrade_dir_clones(dp, tx);
6087
6088 /* Keeping the freedir open increases spa_minref */
6089 spa->spa_minref += 3;
6090 }
6091
6092 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6093 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6094 spa_feature_create_zap_objects(spa, tx);
6095 }
6096 rrw_exit(&dp->dp_config_rwlock, FTAG);
6097 }
6098
6099 /*
6100 * Sync the specified transaction group. New blocks may be dirtied as
6101 * part of the process, so we iterate until it converges.
6102 */
6103 void
6104 spa_sync(spa_t *spa, uint64_t txg)
6105 {
6106 dsl_pool_t *dp = spa->spa_dsl_pool;
6107 objset_t *mos = spa->spa_meta_objset;
6108 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6109 vdev_t *rvd = spa->spa_root_vdev;
6110 vdev_t *vd;
6111 dmu_tx_t *tx;
6112 int error;
6113
6114 VERIFY(spa_writeable(spa));
6115
6116 /*
6117 * Lock out configuration changes.
6118 */
6119 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6120
6121 spa->spa_syncing_txg = txg;
6122 spa->spa_sync_pass = 0;
6123
6124 /*
6125 * If there are any pending vdev state changes, convert them
6126 * into config changes that go out with this transaction group.
6127 */
6128 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6129 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6130 /*
6131 * We need the write lock here because, for aux vdevs,
6132 * calling vdev_config_dirty() modifies sav_config.
6133 * This is ugly and will become unnecessary when we
6134 * eliminate the aux vdev wart by integrating all vdevs
6135 * into the root vdev tree.
6136 */
6137 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6138 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6139 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6140 vdev_state_clean(vd);
6141 vdev_config_dirty(vd);
6142 }
6143 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6144 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6145 }
6146 spa_config_exit(spa, SCL_STATE, FTAG);
6147
6148 tx = dmu_tx_create_assigned(dp, txg);
6149
6150 spa->spa_sync_starttime = gethrtime();
6151 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6152 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6153
6154 /*
6155 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6156 * set spa_deflate if we have no raid-z vdevs.
6157 */
6158 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6159 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6160 int i;
6161
6162 for (i = 0; i < rvd->vdev_children; i++) {
6163 vd = rvd->vdev_child[i];
6164 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6165 break;
6166 }
6167 if (i == rvd->vdev_children) {
6168 spa->spa_deflate = TRUE;
6169 VERIFY(0 == zap_add(spa->spa_meta_objset,
6170 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6171 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6172 }
6173 }
6174
6175 /*
6176 * If anything has changed in this txg, or if someone is waiting
6177 * for this txg to sync (eg, spa_vdev_remove()), push the
6178 * deferred frees from the previous txg. If not, leave them
6179 * alone so that we don't generate work on an otherwise idle
6180 * system.
6181 */
6182 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6183 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6184 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6185 ((dsl_scan_active(dp->dp_scan) ||
6186 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6187 spa_sync_deferred_frees(spa, tx);
6188 }
6189
6190 /*
6191 * Iterate to convergence.
6192 */
6193 do {
6194 int pass = ++spa->spa_sync_pass;
6195
6196 spa_sync_config_object(spa, tx);
6197 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6198 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6199 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6200 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6201 spa_errlog_sync(spa, txg);
6202 dsl_pool_sync(dp, txg);
6203
6204 if (pass < zfs_sync_pass_deferred_free) {
6205 spa_sync_frees(spa, free_bpl, tx);
6206 } else {
6207 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6208 &spa->spa_deferred_bpobj, tx);
6209 }
6210
6211 ddt_sync(spa, txg);
6212 dsl_scan_sync(dp, tx);
6213
6214 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6215 vdev_sync(vd, txg);
6216
6217 if (pass == 1)
6218 spa_sync_upgrades(spa, tx);
6219
6220 } while (dmu_objset_is_dirty(mos, txg));
6221
6222 /*
6223 * Rewrite the vdev configuration (which includes the uberblock)
6224 * to commit the transaction group.
6225 *
6226 * If there are no dirty vdevs, we sync the uberblock to a few
6227 * random top-level vdevs that are known to be visible in the
6228 * config cache (see spa_vdev_add() for a complete description).
6229 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6230 */
6231 for (;;) {
6232 /*
6233 * We hold SCL_STATE to prevent vdev open/close/etc.
6234 * while we're attempting to write the vdev labels.
6235 */
6236 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6237
6238 if (list_is_empty(&spa->spa_config_dirty_list)) {
6239 vdev_t *svd[SPA_DVAS_PER_BP];
6240 int svdcount = 0;
6241 int children = rvd->vdev_children;
6242 int c0 = spa_get_random(children);
6243
6244 for (int c = 0; c < children; c++) {
6245 vd = rvd->vdev_child[(c0 + c) % children];
6246 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6247 continue;
6248 svd[svdcount++] = vd;
6249 if (svdcount == SPA_DVAS_PER_BP)
6250 break;
6251 }
6252 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6253 if (error != 0)
6254 error = vdev_config_sync(svd, svdcount, txg,
6255 B_TRUE);
6256 } else {
6257 error = vdev_config_sync(rvd->vdev_child,
6258 rvd->vdev_children, txg, B_FALSE);
6259 if (error != 0)
6260 error = vdev_config_sync(rvd->vdev_child,
6261 rvd->vdev_children, txg, B_TRUE);
6262 }
6263
6264 if (error == 0)
6265 spa->spa_last_synced_guid = rvd->vdev_guid;
6266
6267 spa_config_exit(spa, SCL_STATE, FTAG);
6268
6269 if (error == 0)
6270 break;
6271 zio_suspend(spa, NULL);
6272 zio_resume_wait(spa);
6273 }
6274 dmu_tx_commit(tx);
6275
6276 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6277
6278 /*
6279 * Clear the dirty config list.
6280 */
6281 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6282 vdev_config_clean(vd);
6283
6284 /*
6285 * Now that the new config has synced transactionally,
6286 * let it become visible to the config cache.
6287 */
6288 if (spa->spa_config_syncing != NULL) {
6289 spa_config_set(spa, spa->spa_config_syncing);
6290 spa->spa_config_txg = txg;
6291 spa->spa_config_syncing = NULL;
6292 }
6293
6294 spa->spa_ubsync = spa->spa_uberblock;
6295
6296 dsl_pool_sync_done(dp, txg);
6297
6298 /*
6299 * Update usable space statistics.
6300 */
6301 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6302 vdev_sync_done(vd, txg);
6303
6304 spa_update_dspace(spa);
6305
6306 /*
6307 * It had better be the case that we didn't dirty anything
6308 * since vdev_config_sync().
6309 */
6310 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6311 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6312 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6313
6314 spa->spa_sync_pass = 0;
6315
6316 spa_config_exit(spa, SCL_CONFIG, FTAG);
6317
6318 spa_handle_ignored_writes(spa);
6319
6320 /*
6321 * If any async tasks have been requested, kick them off.
6322 */
6323 spa_async_dispatch(spa);
6324 }
6325
6326 /*
6327 * Sync all pools. We don't want to hold the namespace lock across these
6328 * operations, so we take a reference on the spa_t and drop the lock during the
6329 * sync.
6330 */
6331 void
6332 spa_sync_allpools(void)
6333 {
6334 spa_t *spa = NULL;
6335 mutex_enter(&spa_namespace_lock);
6336 while ((spa = spa_next(spa)) != NULL) {
6337 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6338 !spa_writeable(spa) || spa_suspended(spa))
6339 continue;
6340 spa_open_ref(spa, FTAG);
6341 mutex_exit(&spa_namespace_lock);
6342 txg_wait_synced(spa_get_dsl(spa), 0);
6343 mutex_enter(&spa_namespace_lock);
6344 spa_close(spa, FTAG);
6345 }
6346 mutex_exit(&spa_namespace_lock);
6347 }
6348
6349 /*
6350 * ==========================================================================
6351 * Miscellaneous routines
6352 * ==========================================================================
6353 */
6354
6355 /*
6356 * Remove all pools in the system.
6357 */
6358 void
6359 spa_evict_all(void)
6360 {
6361 spa_t *spa;
6362
6363 /*
6364 * Remove all cached state. All pools should be closed now,
6365 * so every spa in the AVL tree should be unreferenced.
6366 */
6367 mutex_enter(&spa_namespace_lock);
6368 while ((spa = spa_next(NULL)) != NULL) {
6369 /*
6370 * Stop async tasks. The async thread may need to detach
6371 * a device that's been replaced, which requires grabbing
6372 * spa_namespace_lock, so we must drop it here.
6373 */
6374 spa_open_ref(spa, FTAG);
6375 mutex_exit(&spa_namespace_lock);
6376 spa_async_suspend(spa);
6377 mutex_enter(&spa_namespace_lock);
6378 spa_close(spa, FTAG);
6379
6380 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6381 spa_unload(spa);
6382 spa_deactivate(spa);
6383 }
6384 spa_remove(spa);
6385 }
6386 mutex_exit(&spa_namespace_lock);
6387 }
6388
6389 vdev_t *
6390 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6391 {
6392 vdev_t *vd;
6393 int i;
6394
6395 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6396 return (vd);
6397
6398 if (aux) {
6399 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6400 vd = spa->spa_l2cache.sav_vdevs[i];
6401 if (vd->vdev_guid == guid)
6402 return (vd);
6403 }
6404
6405 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6406 vd = spa->spa_spares.sav_vdevs[i];
6407 if (vd->vdev_guid == guid)
6408 return (vd);
6409 }
6410 }
6411
6412 return (NULL);
6413 }
6414
6415 void
6416 spa_upgrade(spa_t *spa, uint64_t version)
6417 {
6418 ASSERT(spa_writeable(spa));
6419
6420 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6421
6422 /*
6423 * This should only be called for a non-faulted pool, and since a
6424 * future version would result in an unopenable pool, this shouldn't be
6425 * possible.
6426 */
6427 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6428 ASSERT(version >= spa->spa_uberblock.ub_version);
6429
6430 spa->spa_uberblock.ub_version = version;
6431 vdev_config_dirty(spa->spa_root_vdev);
6432
6433 spa_config_exit(spa, SCL_ALL, FTAG);
6434
6435 txg_wait_synced(spa_get_dsl(spa), 0);
6436 }
6437
6438 boolean_t
6439 spa_has_spare(spa_t *spa, uint64_t guid)
6440 {
6441 int i;
6442 uint64_t spareguid;
6443 spa_aux_vdev_t *sav = &spa->spa_spares;
6444
6445 for (i = 0; i < sav->sav_count; i++)
6446 if (sav->sav_vdevs[i]->vdev_guid == guid)
6447 return (B_TRUE);
6448
6449 for (i = 0; i < sav->sav_npending; i++) {
6450 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6451 &spareguid) == 0 && spareguid == guid)
6452 return (B_TRUE);
6453 }
6454
6455 return (B_FALSE);
6456 }
6457
6458 /*
6459 * Check if a pool has an active shared spare device.
6460 * Note: reference count of an active spare is 2, as a spare and as a replace
6461 */
6462 static boolean_t
6463 spa_has_active_shared_spare(spa_t *spa)
6464 {
6465 int i, refcnt;
6466 uint64_t pool;
6467 spa_aux_vdev_t *sav = &spa->spa_spares;
6468
6469 for (i = 0; i < sav->sav_count; i++) {
6470 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6471 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6472 refcnt > 2)
6473 return (B_TRUE);
6474 }
6475
6476 return (B_FALSE);
6477 }
6478
6479 /*
6480 * Post a sysevent corresponding to the given event. The 'name' must be one of
6481 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6482 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6483 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6484 * or zdb as real changes.
6485 */
6486 void
6487 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6488 {
6489 #ifdef _KERNEL
6490 sysevent_t *ev;
6491 sysevent_attr_list_t *attr = NULL;
6492 sysevent_value_t value;
6493 sysevent_id_t eid;
6494
6495 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6496 SE_SLEEP);
6497
6498 value.value_type = SE_DATA_TYPE_STRING;
6499 value.value.sv_string = spa_name(spa);
6500 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6501 goto done;
6502
6503 value.value_type = SE_DATA_TYPE_UINT64;
6504 value.value.sv_uint64 = spa_guid(spa);
6505 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6506 goto done;
6507
6508 if (vd) {
6509 value.value_type = SE_DATA_TYPE_UINT64;
6510 value.value.sv_uint64 = vd->vdev_guid;
6511 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6512 SE_SLEEP) != 0)
6513 goto done;
6514
6515 if (vd->vdev_path) {
6516 value.value_type = SE_DATA_TYPE_STRING;
6517 value.value.sv_string = vd->vdev_path;
6518 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6519 &value, SE_SLEEP) != 0)
6520 goto done;
6521 }
6522 }
6523
6524 if (sysevent_attach_attributes(ev, attr) != 0)
6525 goto done;
6526 attr = NULL;
6527
6528 (void) log_sysevent(ev, SE_SLEEP, &eid);
6529
6530 done:
6531 if (attr)
6532 sysevent_free_attr(attr);
6533 sysevent_free(ev);
6534 #endif
6535 }