1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 */
32
33 /*
34 * SPA: Storage Pool Allocator
35 *
36 * This file contains all the routines used when modifying on-disk SPA state.
37 * This includes opening, importing, destroying, exporting a pool, and syncing a
38 * pool.
39 */
40
41 #include <sys/zfs_context.h>
42 #include <sys/fm/fs/zfs.h>
43 #include <sys/spa_impl.h>
44 #include <sys/zio.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/dmu.h>
47 #include <sys/dmu_tx.h>
48 #include <sys/zap.h>
49 #include <sys/zil.h>
50 #include <sys/ddt.h>
51 #include <sys/vdev_impl.h>
52 #include <sys/metaslab.h>
53 #include <sys/metaslab_impl.h>
54 #include <sys/uberblock_impl.h>
55 #include <sys/txg.h>
56 #include <sys/avl.h>
57 #include <sys/dmu_traverse.h>
58 #include <sys/dmu_objset.h>
59 #include <sys/unique.h>
60 #include <sys/dsl_pool.h>
61 #include <sys/dsl_dataset.h>
62 #include <sys/dsl_dir.h>
63 #include <sys/dsl_prop.h>
64 #include <sys/dsl_synctask.h>
65 #include <sys/fs/zfs.h>
66 #include <sys/arc.h>
67 #include <sys/callb.h>
68 #include <sys/spa_boot.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/dsl_scan.h>
71 #include <sys/dmu_send.h>
72 #include <sys/dsl_destroy.h>
73 #include <sys/dsl_userhold.h>
74 #include <sys/zfeature.h>
75 #include <sys/zvol.h>
76 #include <sys/trim_map.h>
77
78 #ifdef _KERNEL
79 #include <sys/callb.h>
80 #ifndef __NetBSD__
81 #include <sys/cpupart.h>
82 #endif
83 #include <sys/zone.h>
84 #endif /* _KERNEL */
85
86 #include "zfs_prop.h"
87 #include "zfs_comutil.h"
88
89 /* Check hostid on import? */
90 static int check_hostid = 1;
91
92 /*
93 * The interval, in seconds, at which failed configuration cache file writes
94 * should be retried.
95 */
96 static int zfs_ccw_retry_interval = 300;
97
98 SYSCTL_DECL(_vfs_zfs);
99 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
100 "Check hostid on import?");
101 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
102 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
103 &zfs_ccw_retry_interval, 0,
104 "Configuration cache file write, retry after failure, interval (seconds)");
105
106 typedef enum zti_modes {
107 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
108 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
109 ZTI_MODE_NULL, /* don't create a taskq */
110 ZTI_NMODES
111 } zti_modes_t;
112
113 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
114 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
115 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
116
117 #define ZTI_N(n) ZTI_P(n, 1)
118 #define ZTI_ONE ZTI_N(1)
119
120 typedef struct zio_taskq_info {
121 zti_modes_t zti_mode;
122 uint_t zti_value;
123 uint_t zti_count;
124 } zio_taskq_info_t;
125
126 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
127 "issue", "issue_high", "intr", "intr_high"
128 };
129
130 /*
131 * This table defines the taskq settings for each ZFS I/O type. When
132 * initializing a pool, we use this table to create an appropriately sized
133 * taskq. Some operations are low volume and therefore have a small, static
134 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
135 * macros. Other operations process a large amount of data; the ZTI_BATCH
136 * macro causes us to create a taskq oriented for throughput. Some operations
137 * are so high frequency and short-lived that the taskq itself can become a a
138 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
139 * additional degree of parallelism specified by the number of threads per-
140 * taskq and the number of taskqs; when dispatching an event in this case, the
141 * particular taskq is chosen at random.
142 *
143 * The different taskq priorities are to handle the different contexts (issue
144 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
145 * need to be handled with minimum delay.
146 */
147 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
148 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
149 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
150 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
151 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
152 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
153 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
154 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
155 };
156
157 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
158 static void spa_event_post(sysevent_t *ev);
159 static void spa_sync_version(void *arg, dmu_tx_t *tx);
160 static void spa_sync_props(void *arg, dmu_tx_t *tx);
161 static boolean_t spa_has_active_shared_spare(spa_t *spa);
162 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
163 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
164 char **ereport);
165 static void spa_vdev_resilver_done(spa_t *spa);
166
167 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
168 #ifdef PSRSET_BIND
169 id_t zio_taskq_psrset_bind = PS_NONE;
170 #endif
171 #ifdef SYSDC
172 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
173 uint_t zio_taskq_basedc = 80; /* base duty cycle */
174 #endif
175
176 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
177 extern int zfs_sync_pass_deferred_free;
178
179 /*
180 * This (illegal) pool name is used when temporarily importing a spa_t in order
181 * to get the vdev stats associated with the imported devices.
182 */
183 #define TRYIMPORT_NAME "$import"
184
185 /*
186 * ==========================================================================
187 * SPA properties routines
188 * ==========================================================================
189 */
190
191 /*
192 * Add a (source=src, propname=propval) list to an nvlist.
193 */
194 static void
spa_prop_add_list(nvlist_t * nvl,zpool_prop_t prop,char * strval,uint64_t intval,zprop_source_t src)195 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
196 uint64_t intval, zprop_source_t src)
197 {
198 const char *propname = zpool_prop_to_name(prop);
199 nvlist_t *propval;
200
201 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
202 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
203
204 if (strval != NULL)
205 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
206 else
207 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
208
209 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
210 nvlist_free(propval);
211 }
212
213 /*
214 * Get property values from the spa configuration.
215 */
216 static void
spa_prop_get_config(spa_t * spa,nvlist_t ** nvp)217 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
218 {
219 vdev_t *rvd = spa->spa_root_vdev;
220 dsl_pool_t *pool = spa->spa_dsl_pool;
221 uint64_t size, alloc, cap, version;
222 zprop_source_t src = ZPROP_SRC_NONE;
223 spa_config_dirent_t *dp;
224 metaslab_class_t *mc = spa_normal_class(spa);
225
226 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
227
228 if (rvd != NULL) {
229 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
230 size = metaslab_class_get_space(spa_normal_class(spa));
231 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
235 size - alloc, src);
236
237 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
238 metaslab_class_fragmentation(mc), src);
239 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
240 metaslab_class_expandable_space(mc), src);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
242 (spa_mode(spa) == FREAD), src);
243
244 cap = (size == 0) ? 0 : (alloc * 100 / size);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
246
247 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
248 ddt_get_pool_dedup_ratio(spa), src);
249
250 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
251 rvd->vdev_state, src);
252
253 version = spa_version(spa);
254 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
255 src = ZPROP_SRC_DEFAULT;
256 else
257 src = ZPROP_SRC_LOCAL;
258 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
259 }
260
261 if (pool != NULL) {
262 /*
263 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
264 * when opening pools before this version freedir will be NULL.
265 */
266 if (pool->dp_free_dir != NULL) {
267 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
268 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
269 src);
270 } else {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
272 NULL, 0, src);
273 }
274
275 if (pool->dp_leak_dir != NULL) {
276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
277 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
278 src);
279 } else {
280 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
281 NULL, 0, src);
282 }
283 }
284
285 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
286
287 if (spa->spa_comment != NULL) {
288 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
289 0, ZPROP_SRC_LOCAL);
290 }
291
292 if (spa->spa_root != NULL)
293 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
294 0, ZPROP_SRC_LOCAL);
295
296 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
297 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
298 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
299 } else {
300 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
301 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
302 }
303
304 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
305 if (dp->scd_path == NULL) {
306 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
307 "none", 0, ZPROP_SRC_LOCAL);
308 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
309 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
310 dp->scd_path, 0, ZPROP_SRC_LOCAL);
311 }
312 }
313 }
314
315 /*
316 * Get zpool property values.
317 */
318 int
spa_prop_get(spa_t * spa,nvlist_t ** nvp)319 spa_prop_get(spa_t *spa, nvlist_t **nvp)
320 {
321 objset_t *mos = spa->spa_meta_objset;
322 zap_cursor_t zc;
323 zap_attribute_t za;
324 int err;
325
326 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
327
328 mutex_enter(&spa->spa_props_lock);
329
330 /*
331 * Get properties from the spa config.
332 */
333 spa_prop_get_config(spa, nvp);
334
335 /* If no pool property object, no more prop to get. */
336 if (mos == NULL || spa->spa_pool_props_object == 0) {
337 mutex_exit(&spa->spa_props_lock);
338 return (0);
339 }
340
341 /*
342 * Get properties from the MOS pool property object.
343 */
344 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
345 (err = zap_cursor_retrieve(&zc, &za)) == 0;
346 zap_cursor_advance(&zc)) {
347 uint64_t intval = 0;
348 char *strval = NULL;
349 zprop_source_t src = ZPROP_SRC_DEFAULT;
350 zpool_prop_t prop;
351
352 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
353 continue;
354
355 switch (za.za_integer_length) {
356 case 8:
357 /* integer property */
358 if (za.za_first_integer !=
359 zpool_prop_default_numeric(prop))
360 src = ZPROP_SRC_LOCAL;
361
362 if (prop == ZPOOL_PROP_BOOTFS) {
363 dsl_pool_t *dp;
364 dsl_dataset_t *ds = NULL;
365
366 dp = spa_get_dsl(spa);
367 dsl_pool_config_enter(dp, FTAG);
368 if (err = dsl_dataset_hold_obj(dp,
369 za.za_first_integer, FTAG, &ds)) {
370 dsl_pool_config_exit(dp, FTAG);
371 break;
372 }
373
374 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
375 KM_SLEEP);
376 dsl_dataset_name(ds, strval);
377 dsl_dataset_rele(ds, FTAG);
378 dsl_pool_config_exit(dp, FTAG);
379 } else {
380 strval = NULL;
381 intval = za.za_first_integer;
382 }
383
384 spa_prop_add_list(*nvp, prop, strval, intval, src);
385
386 if (strval != NULL)
387 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
388
389 break;
390
391 case 1:
392 /* string property */
393 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
394 err = zap_lookup(mos, spa->spa_pool_props_object,
395 za.za_name, 1, za.za_num_integers, strval);
396 if (err) {
397 kmem_free(strval, za.za_num_integers);
398 break;
399 }
400 spa_prop_add_list(*nvp, prop, strval, 0, src);
401 kmem_free(strval, za.za_num_integers);
402 break;
403
404 default:
405 break;
406 }
407 }
408 zap_cursor_fini(&zc);
409 mutex_exit(&spa->spa_props_lock);
410 out:
411 if (err && err != ENOENT) {
412 nvlist_free(*nvp);
413 *nvp = NULL;
414 return (err);
415 }
416
417 return (0);
418 }
419
420 /*
421 * Validate the given pool properties nvlist and modify the list
422 * for the property values to be set.
423 */
424 static int
spa_prop_validate(spa_t * spa,nvlist_t * props)425 spa_prop_validate(spa_t *spa, nvlist_t *props)
426 {
427 nvpair_t *elem;
428 int error = 0, reset_bootfs = 0;
429 uint64_t objnum = 0;
430 boolean_t has_feature = B_FALSE;
431
432 elem = NULL;
433 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
434 uint64_t intval;
435 char *strval, *slash, *check, *fname;
436 const char *propname = nvpair_name(elem);
437 zpool_prop_t prop = zpool_name_to_prop(propname);
438
439 switch (prop) {
440 case ZPROP_INVAL:
441 if (!zpool_prop_feature(propname)) {
442 error = SET_ERROR(EINVAL);
443 break;
444 }
445
446 /*
447 * Sanitize the input.
448 */
449 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
450 error = SET_ERROR(EINVAL);
451 break;
452 }
453
454 if (nvpair_value_uint64(elem, &intval) != 0) {
455 error = SET_ERROR(EINVAL);
456 break;
457 }
458
459 if (intval != 0) {
460 error = SET_ERROR(EINVAL);
461 break;
462 }
463
464 fname = strchr(propname, '@') + 1;
465 if (zfeature_lookup_name(fname, NULL) != 0) {
466 error = SET_ERROR(EINVAL);
467 break;
468 }
469
470 has_feature = B_TRUE;
471 break;
472
473 case ZPOOL_PROP_VERSION:
474 error = nvpair_value_uint64(elem, &intval);
475 if (!error &&
476 (intval < spa_version(spa) ||
477 intval > SPA_VERSION_BEFORE_FEATURES ||
478 has_feature))
479 error = SET_ERROR(EINVAL);
480 break;
481
482 case ZPOOL_PROP_DELEGATION:
483 case ZPOOL_PROP_AUTOREPLACE:
484 case ZPOOL_PROP_LISTSNAPS:
485 case ZPOOL_PROP_AUTOEXPAND:
486 error = nvpair_value_uint64(elem, &intval);
487 if (!error && intval > 1)
488 error = SET_ERROR(EINVAL);
489 break;
490
491 case ZPOOL_PROP_BOOTFS:
492 /*
493 * If the pool version is less than SPA_VERSION_BOOTFS,
494 * or the pool is still being created (version == 0),
495 * the bootfs property cannot be set.
496 */
497 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
498 error = SET_ERROR(ENOTSUP);
499 break;
500 }
501
502 /*
503 * Make sure the vdev config is bootable
504 */
505 if (!vdev_is_bootable(spa->spa_root_vdev)) {
506 error = SET_ERROR(ENOTSUP);
507 break;
508 }
509
510 reset_bootfs = 1;
511
512 error = nvpair_value_string(elem, &strval);
513
514 if (!error) {
515 objset_t *os;
516 uint64_t propval;
517
518 if (strval == NULL || strval[0] == '\0') {
519 objnum = zpool_prop_default_numeric(
520 ZPOOL_PROP_BOOTFS);
521 break;
522 }
523
524 if (error = dmu_objset_hold(strval, FTAG, &os))
525 break;
526
527 /*
528 * Must be ZPL, and its property settings
529 * must be supported by GRUB (compression
530 * is not gzip, and large blocks are not used).
531 */
532
533 if (dmu_objset_type(os) != DMU_OST_ZFS) {
534 error = SET_ERROR(ENOTSUP);
535 } else if ((error =
536 dsl_prop_get_int_ds(dmu_objset_ds(os),
537 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
538 &propval)) == 0 &&
539 !BOOTFS_COMPRESS_VALID(propval)) {
540 error = SET_ERROR(ENOTSUP);
541 } else {
542 objnum = dmu_objset_id(os);
543 }
544 dmu_objset_rele(os, FTAG);
545 }
546 break;
547
548 case ZPOOL_PROP_FAILUREMODE:
549 error = nvpair_value_uint64(elem, &intval);
550 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
551 intval > ZIO_FAILURE_MODE_PANIC))
552 error = SET_ERROR(EINVAL);
553
554 /*
555 * This is a special case which only occurs when
556 * the pool has completely failed. This allows
557 * the user to change the in-core failmode property
558 * without syncing it out to disk (I/Os might
559 * currently be blocked). We do this by returning
560 * EIO to the caller (spa_prop_set) to trick it
561 * into thinking we encountered a property validation
562 * error.
563 */
564 if (!error && spa_suspended(spa)) {
565 spa->spa_failmode = intval;
566 error = SET_ERROR(EIO);
567 }
568 break;
569
570 case ZPOOL_PROP_CACHEFILE:
571 if ((error = nvpair_value_string(elem, &strval)) != 0)
572 break;
573
574 if (strval[0] == '\0')
575 break;
576
577 if (strcmp(strval, "none") == 0)
578 break;
579
580 if (strval[0] != '/') {
581 error = SET_ERROR(EINVAL);
582 break;
583 }
584
585 slash = strrchr(strval, '/');
586 ASSERT(slash != NULL);
587
588 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
589 strcmp(slash, "/..") == 0)
590 error = SET_ERROR(EINVAL);
591 break;
592
593 case ZPOOL_PROP_COMMENT:
594 if ((error = nvpair_value_string(elem, &strval)) != 0)
595 break;
596 for (check = strval; *check != '\0'; check++) {
597 /*
598 * The kernel doesn't have an easy isprint()
599 * check. For this kernel check, we merely
600 * check ASCII apart from DEL. Fix this if
601 * there is an easy-to-use kernel isprint().
602 */
603 if (*check >= 0x7f) {
604 error = SET_ERROR(EINVAL);
605 break;
606 }
607 }
608 if (strlen(strval) > ZPROP_MAX_COMMENT)
609 error = E2BIG;
610 break;
611
612 case ZPOOL_PROP_DEDUPDITTO:
613 if (spa_version(spa) < SPA_VERSION_DEDUP)
614 error = SET_ERROR(ENOTSUP);
615 else
616 error = nvpair_value_uint64(elem, &intval);
617 if (error == 0 &&
618 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
619 error = SET_ERROR(EINVAL);
620 break;
621 }
622
623 if (error)
624 break;
625 }
626
627 if (!error && reset_bootfs) {
628 error = nvlist_remove(props,
629 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
630
631 if (!error) {
632 error = nvlist_add_uint64(props,
633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
634 }
635 }
636
637 return (error);
638 }
639
640 void
spa_configfile_set(spa_t * spa,nvlist_t * nvp,boolean_t need_sync)641 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
642 {
643 char *cachefile;
644 spa_config_dirent_t *dp;
645
646 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
647 &cachefile) != 0)
648 return;
649
650 dp = kmem_alloc(sizeof (spa_config_dirent_t),
651 KM_SLEEP);
652
653 if (cachefile[0] == '\0')
654 dp->scd_path = spa_strdup(spa_config_path);
655 else if (strcmp(cachefile, "none") == 0)
656 dp->scd_path = NULL;
657 else
658 dp->scd_path = spa_strdup(cachefile);
659
660 list_insert_head(&spa->spa_config_list, dp);
661 if (need_sync)
662 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
663 }
664
665 int
spa_prop_set(spa_t * spa,nvlist_t * nvp)666 spa_prop_set(spa_t *spa, nvlist_t *nvp)
667 {
668 int error;
669 nvpair_t *elem = NULL;
670 boolean_t need_sync = B_FALSE;
671
672 if ((error = spa_prop_validate(spa, nvp)) != 0)
673 return (error);
674
675 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
676 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
677
678 if (prop == ZPOOL_PROP_CACHEFILE ||
679 prop == ZPOOL_PROP_ALTROOT ||
680 prop == ZPOOL_PROP_READONLY)
681 continue;
682
683 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
684 uint64_t ver;
685
686 if (prop == ZPOOL_PROP_VERSION) {
687 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
688 } else {
689 ASSERT(zpool_prop_feature(nvpair_name(elem)));
690 ver = SPA_VERSION_FEATURES;
691 need_sync = B_TRUE;
692 }
693
694 /* Save time if the version is already set. */
695 if (ver == spa_version(spa))
696 continue;
697
698 /*
699 * In addition to the pool directory object, we might
700 * create the pool properties object, the features for
701 * read object, the features for write object, or the
702 * feature descriptions object.
703 */
704 error = dsl_sync_task(spa->spa_name, NULL,
705 spa_sync_version, &ver,
706 6, ZFS_SPACE_CHECK_RESERVED);
707 if (error)
708 return (error);
709 continue;
710 }
711
712 need_sync = B_TRUE;
713 break;
714 }
715
716 if (need_sync) {
717 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
718 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
719 }
720
721 return (0);
722 }
723
724 /*
725 * If the bootfs property value is dsobj, clear it.
726 */
727 void
spa_prop_clear_bootfs(spa_t * spa,uint64_t dsobj,dmu_tx_t * tx)728 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
729 {
730 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
731 VERIFY(zap_remove(spa->spa_meta_objset,
732 spa->spa_pool_props_object,
733 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
734 spa->spa_bootfs = 0;
735 }
736 }
737
738 /*ARGSUSED*/
739 static int
spa_change_guid_check(void * arg,dmu_tx_t * tx)740 spa_change_guid_check(void *arg, dmu_tx_t *tx)
741 {
742 uint64_t *newguid = arg;
743 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
744 vdev_t *rvd = spa->spa_root_vdev;
745 uint64_t vdev_state;
746
747 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
748 vdev_state = rvd->vdev_state;
749 spa_config_exit(spa, SCL_STATE, FTAG);
750
751 if (vdev_state != VDEV_STATE_HEALTHY)
752 return (SET_ERROR(ENXIO));
753
754 ASSERT3U(spa_guid(spa), !=, *newguid);
755
756 return (0);
757 }
758
759 static void
spa_change_guid_sync(void * arg,dmu_tx_t * tx)760 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
761 {
762 uint64_t *newguid = arg;
763 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
764 uint64_t oldguid;
765 vdev_t *rvd = spa->spa_root_vdev;
766
767 oldguid = spa_guid(spa);
768
769 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
770 rvd->vdev_guid = *newguid;
771 rvd->vdev_guid_sum += (*newguid - oldguid);
772 vdev_config_dirty(rvd);
773 spa_config_exit(spa, SCL_STATE, FTAG);
774
775 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
776 oldguid, *newguid);
777 }
778
779 /*
780 * Change the GUID for the pool. This is done so that we can later
781 * re-import a pool built from a clone of our own vdevs. We will modify
782 * the root vdev's guid, our own pool guid, and then mark all of our
783 * vdevs dirty. Note that we must make sure that all our vdevs are
784 * online when we do this, or else any vdevs that weren't present
785 * would be orphaned from our pool. We are also going to issue a
786 * sysevent to update any watchers.
787 */
788 int
spa_change_guid(spa_t * spa)789 spa_change_guid(spa_t *spa)
790 {
791 int error;
792 uint64_t guid;
793
794 mutex_enter(&spa->spa_vdev_top_lock);
795 mutex_enter(&spa_namespace_lock);
796 guid = spa_generate_guid(NULL);
797
798 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
799 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
800
801 if (error == 0) {
802 spa_config_sync(spa, B_FALSE, B_TRUE);
803 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
804 }
805
806 mutex_exit(&spa_namespace_lock);
807 mutex_exit(&spa->spa_vdev_top_lock);
808
809 return (error);
810 }
811
812 /*
813 * ==========================================================================
814 * SPA state manipulation (open/create/destroy/import/export)
815 * ==========================================================================
816 */
817
818 static int
spa_error_entry_compare(const void * a,const void * b)819 spa_error_entry_compare(const void *a, const void *b)
820 {
821 spa_error_entry_t *sa = (spa_error_entry_t *)a;
822 spa_error_entry_t *sb = (spa_error_entry_t *)b;
823 int ret;
824
825 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
826 sizeof (zbookmark_phys_t));
827
828 if (ret < 0)
829 return (-1);
830 else if (ret > 0)
831 return (1);
832 else
833 return (0);
834 }
835
836 /*
837 * Utility function which retrieves copies of the current logs and
838 * re-initializes them in the process.
839 */
840 void
spa_get_errlists(spa_t * spa,avl_tree_t * last,avl_tree_t * scrub)841 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
842 {
843 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
844
845 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
846 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
847
848 avl_create(&spa->spa_errlist_scrub,
849 spa_error_entry_compare, sizeof (spa_error_entry_t),
850 offsetof(spa_error_entry_t, se_avl));
851 avl_create(&spa->spa_errlist_last,
852 spa_error_entry_compare, sizeof (spa_error_entry_t),
853 offsetof(spa_error_entry_t, se_avl));
854 }
855
856 static void
spa_taskqs_init(spa_t * spa,zio_type_t t,zio_taskq_type_t q)857 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
858 {
859 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
860 enum zti_modes mode = ztip->zti_mode;
861 uint_t value = ztip->zti_value;
862 uint_t count = ztip->zti_count;
863 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
864 char name[32];
865 uint_t flags = 0;
866 boolean_t batch = B_FALSE;
867
868 if (mode == ZTI_MODE_NULL) {
869 tqs->stqs_count = 0;
870 tqs->stqs_taskq = NULL;
871 return;
872 }
873
874 ASSERT3U(count, >, 0);
875
876 tqs->stqs_count = count;
877 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
878
879 switch (mode) {
880 case ZTI_MODE_FIXED:
881 ASSERT3U(value, >=, 1);
882 value = MAX(value, 1);
883 break;
884
885 case ZTI_MODE_BATCH:
886 batch = B_TRUE;
887 flags |= TASKQ_THREADS_CPU_PCT;
888 value = zio_taskq_batch_pct;
889 break;
890
891 default:
892 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
893 "spa_activate()",
894 zio_type_name[t], zio_taskq_types[q], mode, value);
895 break;
896 }
897
898 for (uint_t i = 0; i < count; i++) {
899 taskq_t *tq;
900
901 if (count > 1) {
902 (void) snprintf(name, sizeof (name), "%s_%s_%u",
903 zio_type_name[t], zio_taskq_types[q], i);
904 } else {
905 (void) snprintf(name, sizeof (name), "%s_%s",
906 zio_type_name[t], zio_taskq_types[q]);
907 }
908
909 #ifdef SYSDC
910 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
911 if (batch)
912 flags |= TASKQ_DC_BATCH;
913
914 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
915 spa->spa_proc, zio_taskq_basedc, flags);
916 } else {
917 #endif
918 pri_t pri = maxclsyspri;
919 /*
920 * The write issue taskq can be extremely CPU
921 * intensive. Run it at slightly lower priority
922 * than the other taskqs.
923 * FreeBSD notes:
924 * - numerically higher priorities are lower priorities;
925 * - if priorities divided by four (RQ_PPQ) are equal
926 * then a difference between them is insignificant.
927 */
928 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
929 #if defined(illumos) || defined(__NetBSD__)
930 pri--;
931 #else
932 pri += 4;
933 #endif
934
935 tq = taskq_create_proc(name, value, pri, 50,
936 INT_MAX, spa->spa_proc, flags);
937 #ifdef SYSDC
938 }
939 #endif
940
941 tqs->stqs_taskq[i] = tq;
942 }
943 }
944
945 static void
spa_taskqs_fini(spa_t * spa,zio_type_t t,zio_taskq_type_t q)946 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
947 {
948 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
949
950 if (tqs->stqs_taskq == NULL) {
951 ASSERT0(tqs->stqs_count);
952 return;
953 }
954
955 for (uint_t i = 0; i < tqs->stqs_count; i++) {
956 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
957 taskq_destroy(tqs->stqs_taskq[i]);
958 }
959
960 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
961 tqs->stqs_taskq = NULL;
962 }
963
964 /*
965 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
966 * Note that a type may have multiple discrete taskqs to avoid lock contention
967 * on the taskq itself. In that case we choose which taskq at random by using
968 * the low bits of gethrtime().
969 */
970 void
spa_taskq_dispatch_ent(spa_t * spa,zio_type_t t,zio_taskq_type_t q,task_func_t * func,void * arg,uint_t flags,taskq_ent_t * ent)971 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
972 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
973 {
974 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
975 taskq_t *tq;
976
977 ASSERT3P(tqs->stqs_taskq, !=, NULL);
978 ASSERT3U(tqs->stqs_count, !=, 0);
979
980 if (tqs->stqs_count == 1) {
981 tq = tqs->stqs_taskq[0];
982 } else {
983 #if defined(__FreeBSD__) && defined(_KERNEL)
984 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
985 #else
986 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
987 #endif
988 }
989
990 taskq_dispatch_ent(tq, func, arg, flags, ent);
991 }
992
993 static void
spa_create_zio_taskqs(spa_t * spa)994 spa_create_zio_taskqs(spa_t *spa)
995 {
996 for (int t = 0; t < ZIO_TYPES; t++) {
997 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
998 spa_taskqs_init(spa, t, q);
999 }
1000 }
1001 }
1002
1003 #ifdef _KERNEL
1004 #ifdef SPA_PROCESS
1005 static void
spa_thread(void * arg)1006 spa_thread(void *arg)
1007 {
1008 callb_cpr_t cprinfo;
1009
1010 spa_t *spa = arg;
1011 user_t *pu = PTOU(curproc);
1012
1013 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1014 spa->spa_name);
1015
1016 ASSERT(curproc != &p0);
1017 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1018 "zpool-%s", spa->spa_name);
1019 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1020
1021 #ifdef PSRSET_BIND
1022 /* bind this thread to the requested psrset */
1023 if (zio_taskq_psrset_bind != PS_NONE) {
1024 pool_lock();
1025 mutex_enter(&cpu_lock);
1026 mutex_enter(&pidlock);
1027 mutex_enter(&curproc->p_lock);
1028
1029 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1030 0, NULL, NULL) == 0) {
1031 curthread->t_bind_pset = zio_taskq_psrset_bind;
1032 } else {
1033 cmn_err(CE_WARN,
1034 "Couldn't bind process for zfs pool \"%s\" to "
1035 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1036 }
1037
1038 mutex_exit(&curproc->p_lock);
1039 mutex_exit(&pidlock);
1040 mutex_exit(&cpu_lock);
1041 pool_unlock();
1042 }
1043 #endif
1044
1045 #ifdef SYSDC
1046 if (zio_taskq_sysdc) {
1047 sysdc_thread_enter(curthread, 100, 0);
1048 }
1049 #endif
1050
1051 spa->spa_proc = curproc;
1052 spa->spa_did = curthread->t_did;
1053
1054 spa_create_zio_taskqs(spa);
1055
1056 mutex_enter(&spa->spa_proc_lock);
1057 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1058
1059 spa->spa_proc_state = SPA_PROC_ACTIVE;
1060 cv_broadcast(&spa->spa_proc_cv);
1061
1062 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1063 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1064 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1065 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1066
1067 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1068 spa->spa_proc_state = SPA_PROC_GONE;
1069 spa->spa_proc = &p0;
1070 cv_broadcast(&spa->spa_proc_cv);
1071 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1072
1073 mutex_enter(&curproc->p_lock);
1074 lwp_exit();
1075 }
1076 #endif /* SPA_PROCESS */
1077 #endif
1078
1079 /*
1080 * Activate an uninitialized pool.
1081 */
1082 static void
spa_activate(spa_t * spa,int mode)1083 spa_activate(spa_t *spa, int mode)
1084 {
1085 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1086
1087 spa->spa_state = POOL_STATE_ACTIVE;
1088 spa->spa_mode = mode;
1089
1090 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1091 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1092
1093 /* Try to create a covering process */
1094 mutex_enter(&spa->spa_proc_lock);
1095 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1096 ASSERT(spa->spa_proc == &p0);
1097 spa->spa_did = 0;
1098
1099 #ifdef SPA_PROCESS
1100 /* Only create a process if we're going to be around a while. */
1101 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1102 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1103 NULL, 0) == 0) {
1104 spa->spa_proc_state = SPA_PROC_CREATED;
1105 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1106 cv_wait(&spa->spa_proc_cv,
1107 &spa->spa_proc_lock);
1108 }
1109 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1110 ASSERT(spa->spa_proc != &p0);
1111 ASSERT(spa->spa_did != 0);
1112 } else {
1113 #ifdef _KERNEL
1114 cmn_err(CE_WARN,
1115 "Couldn't create process for zfs pool \"%s\"\n",
1116 spa->spa_name);
1117 #endif
1118 }
1119 }
1120 #endif /* SPA_PROCESS */
1121 mutex_exit(&spa->spa_proc_lock);
1122
1123 /* If we didn't create a process, we need to create our taskqs. */
1124 ASSERT(spa->spa_proc == &p0);
1125 if (spa->spa_proc == &p0) {
1126 spa_create_zio_taskqs(spa);
1127 }
1128
1129 /*
1130 * Start TRIM thread.
1131 */
1132 trim_thread_create(spa);
1133
1134 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1135 offsetof(vdev_t, vdev_config_dirty_node));
1136 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1137 offsetof(objset_t, os_evicting_node));
1138 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1139 offsetof(vdev_t, vdev_state_dirty_node));
1140
1141 txg_list_create(&spa->spa_vdev_txg_list,
1142 offsetof(struct vdev, vdev_txg_node));
1143
1144 avl_create(&spa->spa_errlist_scrub,
1145 spa_error_entry_compare, sizeof (spa_error_entry_t),
1146 offsetof(spa_error_entry_t, se_avl));
1147 avl_create(&spa->spa_errlist_last,
1148 spa_error_entry_compare, sizeof (spa_error_entry_t),
1149 offsetof(spa_error_entry_t, se_avl));
1150 }
1151
1152 /*
1153 * Opposite of spa_activate().
1154 */
1155 static void
spa_deactivate(spa_t * spa)1156 spa_deactivate(spa_t *spa)
1157 {
1158 ASSERT(spa->spa_sync_on == B_FALSE);
1159 ASSERT(spa->spa_dsl_pool == NULL);
1160 ASSERT(spa->spa_root_vdev == NULL);
1161 ASSERT(spa->spa_async_zio_root == NULL);
1162 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1163
1164 /*
1165 * Stop TRIM thread in case spa_unload() wasn't called directly
1166 * before spa_deactivate().
1167 */
1168 trim_thread_destroy(spa);
1169
1170 spa_evicting_os_wait(spa);
1171
1172 txg_list_destroy(&spa->spa_vdev_txg_list);
1173
1174 list_destroy(&spa->spa_config_dirty_list);
1175 list_destroy(&spa->spa_evicting_os_list);
1176 list_destroy(&spa->spa_state_dirty_list);
1177
1178 for (int t = 0; t < ZIO_TYPES; t++) {
1179 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1180 spa_taskqs_fini(spa, t, q);
1181 }
1182 }
1183
1184 metaslab_class_destroy(spa->spa_normal_class);
1185 spa->spa_normal_class = NULL;
1186
1187 metaslab_class_destroy(spa->spa_log_class);
1188 spa->spa_log_class = NULL;
1189
1190 /*
1191 * If this was part of an import or the open otherwise failed, we may
1192 * still have errors left in the queues. Empty them just in case.
1193 */
1194 spa_errlog_drain(spa);
1195
1196 avl_destroy(&spa->spa_errlist_scrub);
1197 avl_destroy(&spa->spa_errlist_last);
1198
1199 spa->spa_state = POOL_STATE_UNINITIALIZED;
1200
1201 mutex_enter(&spa->spa_proc_lock);
1202 if (spa->spa_proc_state != SPA_PROC_NONE) {
1203 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1204 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1205 cv_broadcast(&spa->spa_proc_cv);
1206 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1207 ASSERT(spa->spa_proc != &p0);
1208 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1209 }
1210 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1211 spa->spa_proc_state = SPA_PROC_NONE;
1212 }
1213 ASSERT(spa->spa_proc == &p0);
1214 mutex_exit(&spa->spa_proc_lock);
1215
1216 #ifdef SPA_PROCESS
1217 /*
1218 * We want to make sure spa_thread() has actually exited the ZFS
1219 * module, so that the module can't be unloaded out from underneath
1220 * it.
1221 */
1222 if (spa->spa_did != 0) {
1223 thread_join(spa->spa_did);
1224 spa->spa_did = 0;
1225 }
1226 #endif /* SPA_PROCESS */
1227 }
1228
1229 /*
1230 * Verify a pool configuration, and construct the vdev tree appropriately. This
1231 * will create all the necessary vdevs in the appropriate layout, with each vdev
1232 * in the CLOSED state. This will prep the pool before open/creation/import.
1233 * All vdev validation is done by the vdev_alloc() routine.
1234 */
1235 static int
spa_config_parse(spa_t * spa,vdev_t ** vdp,nvlist_t * nv,vdev_t * parent,uint_t id,int atype)1236 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1237 uint_t id, int atype)
1238 {
1239 nvlist_t **child;
1240 uint_t children;
1241 int error;
1242
1243 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1244 return (error);
1245
1246 if ((*vdp)->vdev_ops->vdev_op_leaf)
1247 return (0);
1248
1249 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1250 &child, &children);
1251
1252 if (error == ENOENT)
1253 return (0);
1254
1255 if (error) {
1256 vdev_free(*vdp);
1257 *vdp = NULL;
1258 return (SET_ERROR(EINVAL));
1259 }
1260
1261 for (int c = 0; c < children; c++) {
1262 vdev_t *vd;
1263 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1264 atype)) != 0) {
1265 vdev_free(*vdp);
1266 *vdp = NULL;
1267 return (error);
1268 }
1269 }
1270
1271 ASSERT(*vdp != NULL);
1272
1273 return (0);
1274 }
1275
1276 /*
1277 * Opposite of spa_load().
1278 */
1279 static void
spa_unload(spa_t * spa)1280 spa_unload(spa_t *spa)
1281 {
1282 int i;
1283
1284 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1285
1286 /*
1287 * Stop TRIM thread.
1288 */
1289 trim_thread_destroy(spa);
1290
1291 /*
1292 * Stop async tasks.
1293 */
1294 spa_async_suspend(spa);
1295
1296 /*
1297 * Stop syncing.
1298 */
1299 if (spa->spa_sync_on) {
1300 txg_sync_stop(spa->spa_dsl_pool);
1301 spa->spa_sync_on = B_FALSE;
1302 }
1303
1304 /*
1305 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1306 * to call it earlier, before we wait for async i/o to complete.
1307 * This ensures that there is no async metaslab prefetching, by
1308 * calling taskq_wait(mg_taskq).
1309 */
1310 if (spa->spa_root_vdev != NULL) {
1311 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1312 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1313 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1314 spa_config_exit(spa, SCL_ALL, FTAG);
1315 }
1316
1317 /*
1318 * Wait for any outstanding async I/O to complete.
1319 */
1320 if (spa->spa_async_zio_root != NULL) {
1321 for (int i = 0; i < max_ncpus; i++)
1322 (void) zio_wait(spa->spa_async_zio_root[i]);
1323 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1324 spa->spa_async_zio_root = NULL;
1325 }
1326
1327 bpobj_close(&spa->spa_deferred_bpobj);
1328
1329 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1330
1331 /*
1332 * Close all vdevs.
1333 */
1334 if (spa->spa_root_vdev)
1335 vdev_free(spa->spa_root_vdev);
1336 ASSERT(spa->spa_root_vdev == NULL);
1337
1338 /*
1339 * Close the dsl pool.
1340 */
1341 if (spa->spa_dsl_pool) {
1342 dsl_pool_close(spa->spa_dsl_pool);
1343 spa->spa_dsl_pool = NULL;
1344 spa->spa_meta_objset = NULL;
1345 }
1346
1347 ddt_unload(spa);
1348
1349 /*
1350 * Drop and purge level 2 cache
1351 */
1352 spa_l2cache_drop(spa);
1353
1354 for (i = 0; i < spa->spa_spares.sav_count; i++)
1355 vdev_free(spa->spa_spares.sav_vdevs[i]);
1356 if (spa->spa_spares.sav_vdevs) {
1357 kmem_free(spa->spa_spares.sav_vdevs,
1358 spa->spa_spares.sav_count * sizeof (void *));
1359 spa->spa_spares.sav_vdevs = NULL;
1360 }
1361 if (spa->spa_spares.sav_config) {
1362 nvlist_free(spa->spa_spares.sav_config);
1363 spa->spa_spares.sav_config = NULL;
1364 }
1365 spa->spa_spares.sav_count = 0;
1366
1367 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1368 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1369 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1370 }
1371 if (spa->spa_l2cache.sav_vdevs) {
1372 kmem_free(spa->spa_l2cache.sav_vdevs,
1373 spa->spa_l2cache.sav_count * sizeof (void *));
1374 spa->spa_l2cache.sav_vdevs = NULL;
1375 }
1376 if (spa->spa_l2cache.sav_config) {
1377 nvlist_free(spa->spa_l2cache.sav_config);
1378 spa->spa_l2cache.sav_config = NULL;
1379 }
1380 spa->spa_l2cache.sav_count = 0;
1381
1382 spa->spa_async_suspended = 0;
1383
1384 if (spa->spa_comment != NULL) {
1385 spa_strfree(spa->spa_comment);
1386 spa->spa_comment = NULL;
1387 }
1388
1389 spa_config_exit(spa, SCL_ALL, FTAG);
1390 }
1391
1392 /*
1393 * Load (or re-load) the current list of vdevs describing the active spares for
1394 * this pool. When this is called, we have some form of basic information in
1395 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1396 * then re-generate a more complete list including status information.
1397 */
1398 static void
spa_load_spares(spa_t * spa)1399 spa_load_spares(spa_t *spa)
1400 {
1401 nvlist_t **spares;
1402 uint_t nspares;
1403 int i;
1404 vdev_t *vd, *tvd;
1405
1406 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1407
1408 /*
1409 * First, close and free any existing spare vdevs.
1410 */
1411 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1412 vd = spa->spa_spares.sav_vdevs[i];
1413
1414 /* Undo the call to spa_activate() below */
1415 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1416 B_FALSE)) != NULL && tvd->vdev_isspare)
1417 spa_spare_remove(tvd);
1418 vdev_close(vd);
1419 vdev_free(vd);
1420 }
1421
1422 if (spa->spa_spares.sav_vdevs)
1423 kmem_free(spa->spa_spares.sav_vdevs,
1424 spa->spa_spares.sav_count * sizeof (void *));
1425
1426 if (spa->spa_spares.sav_config == NULL)
1427 nspares = 0;
1428 else
1429 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1430 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1431
1432 spa->spa_spares.sav_count = (int)nspares;
1433 spa->spa_spares.sav_vdevs = NULL;
1434
1435 if (nspares == 0)
1436 return;
1437
1438 /*
1439 * Construct the array of vdevs, opening them to get status in the
1440 * process. For each spare, there is potentially two different vdev_t
1441 * structures associated with it: one in the list of spares (used only
1442 * for basic validation purposes) and one in the active vdev
1443 * configuration (if it's spared in). During this phase we open and
1444 * validate each vdev on the spare list. If the vdev also exists in the
1445 * active configuration, then we also mark this vdev as an active spare.
1446 */
1447 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1448 KM_SLEEP);
1449 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1450 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1451 VDEV_ALLOC_SPARE) == 0);
1452 ASSERT(vd != NULL);
1453
1454 spa->spa_spares.sav_vdevs[i] = vd;
1455
1456 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1457 B_FALSE)) != NULL) {
1458 if (!tvd->vdev_isspare)
1459 spa_spare_add(tvd);
1460
1461 /*
1462 * We only mark the spare active if we were successfully
1463 * able to load the vdev. Otherwise, importing a pool
1464 * with a bad active spare would result in strange
1465 * behavior, because multiple pool would think the spare
1466 * is actively in use.
1467 *
1468 * There is a vulnerability here to an equally bizarre
1469 * circumstance, where a dead active spare is later
1470 * brought back to life (onlined or otherwise). Given
1471 * the rarity of this scenario, and the extra complexity
1472 * it adds, we ignore the possibility.
1473 */
1474 if (!vdev_is_dead(tvd))
1475 spa_spare_activate(tvd);
1476 }
1477
1478 vd->vdev_top = vd;
1479 vd->vdev_aux = &spa->spa_spares;
1480
1481 if (vdev_open(vd) != 0)
1482 continue;
1483
1484 if (vdev_validate_aux(vd) == 0)
1485 spa_spare_add(vd);
1486 }
1487
1488 /*
1489 * Recompute the stashed list of spares, with status information
1490 * this time.
1491 */
1492 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1493 DATA_TYPE_NVLIST_ARRAY) == 0);
1494
1495 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1496 KM_SLEEP);
1497 for (i = 0; i < spa->spa_spares.sav_count; i++)
1498 spares[i] = vdev_config_generate(spa,
1499 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1500 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1501 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1502 for (i = 0; i < spa->spa_spares.sav_count; i++)
1503 nvlist_free(spares[i]);
1504 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1505 }
1506
1507 /*
1508 * Load (or re-load) the current list of vdevs describing the active l2cache for
1509 * this pool. When this is called, we have some form of basic information in
1510 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1511 * then re-generate a more complete list including status information.
1512 * Devices which are already active have their details maintained, and are
1513 * not re-opened.
1514 */
1515 static void
spa_load_l2cache(spa_t * spa)1516 spa_load_l2cache(spa_t *spa)
1517 {
1518 nvlist_t **l2cache;
1519 uint_t nl2cache;
1520 int i, j, oldnvdevs;
1521 uint64_t guid;
1522 vdev_t *vd, **oldvdevs, **newvdevs;
1523 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1524
1525 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1526
1527 if (sav->sav_config != NULL) {
1528 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1529 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1530 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1531 } else {
1532 nl2cache = 0;
1533 newvdevs = NULL;
1534 }
1535
1536 oldvdevs = sav->sav_vdevs;
1537 oldnvdevs = sav->sav_count;
1538 sav->sav_vdevs = NULL;
1539 sav->sav_count = 0;
1540
1541 /*
1542 * Process new nvlist of vdevs.
1543 */
1544 for (i = 0; i < nl2cache; i++) {
1545 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1546 &guid) == 0);
1547
1548 newvdevs[i] = NULL;
1549 for (j = 0; j < oldnvdevs; j++) {
1550 vd = oldvdevs[j];
1551 if (vd != NULL && guid == vd->vdev_guid) {
1552 /*
1553 * Retain previous vdev for add/remove ops.
1554 */
1555 newvdevs[i] = vd;
1556 oldvdevs[j] = NULL;
1557 break;
1558 }
1559 }
1560
1561 if (newvdevs[i] == NULL) {
1562 /*
1563 * Create new vdev
1564 */
1565 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1566 VDEV_ALLOC_L2CACHE) == 0);
1567 ASSERT(vd != NULL);
1568 newvdevs[i] = vd;
1569
1570 /*
1571 * Commit this vdev as an l2cache device,
1572 * even if it fails to open.
1573 */
1574 spa_l2cache_add(vd);
1575
1576 vd->vdev_top = vd;
1577 vd->vdev_aux = sav;
1578
1579 spa_l2cache_activate(vd);
1580
1581 if (vdev_open(vd) != 0)
1582 continue;
1583
1584 (void) vdev_validate_aux(vd);
1585
1586 if (!vdev_is_dead(vd))
1587 l2arc_add_vdev(spa, vd);
1588 }
1589 }
1590
1591 /*
1592 * Purge vdevs that were dropped
1593 */
1594 for (i = 0; i < oldnvdevs; i++) {
1595 uint64_t pool;
1596
1597 vd = oldvdevs[i];
1598 if (vd != NULL) {
1599 ASSERT(vd->vdev_isl2cache);
1600
1601 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1602 pool != 0ULL && l2arc_vdev_present(vd))
1603 l2arc_remove_vdev(vd);
1604 vdev_clear_stats(vd);
1605 vdev_free(vd);
1606 }
1607 }
1608
1609 if (oldvdevs)
1610 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1611
1612 if (sav->sav_config == NULL)
1613 goto out;
1614
1615 sav->sav_vdevs = newvdevs;
1616 sav->sav_count = (int)nl2cache;
1617
1618 /*
1619 * Recompute the stashed list of l2cache devices, with status
1620 * information this time.
1621 */
1622 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1623 DATA_TYPE_NVLIST_ARRAY) == 0);
1624
1625 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1626 for (i = 0; i < sav->sav_count; i++)
1627 l2cache[i] = vdev_config_generate(spa,
1628 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1629 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1630 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1631 out:
1632 for (i = 0; i < sav->sav_count; i++)
1633 nvlist_free(l2cache[i]);
1634 if (sav->sav_count)
1635 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1636 }
1637
1638 static int
load_nvlist(spa_t * spa,uint64_t obj,nvlist_t ** value)1639 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1640 {
1641 dmu_buf_t *db;
1642 char *packed = NULL;
1643 size_t nvsize = 0;
1644 int error;
1645 *value = NULL;
1646
1647 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1648 if (error != 0)
1649 return (error);
1650
1651 nvsize = *(uint64_t *)db->db_data;
1652 dmu_buf_rele(db, FTAG);
1653
1654 packed = kmem_alloc(nvsize, KM_SLEEP);
1655 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1656 DMU_READ_PREFETCH);
1657 if (error == 0)
1658 error = nvlist_unpack(packed, nvsize, value, 0);
1659 kmem_free(packed, nvsize);
1660
1661 return (error);
1662 }
1663
1664 /*
1665 * Checks to see if the given vdev could not be opened, in which case we post a
1666 * sysevent to notify the autoreplace code that the device has been removed.
1667 */
1668 static void
spa_check_removed(vdev_t * vd)1669 spa_check_removed(vdev_t *vd)
1670 {
1671 for (int c = 0; c < vd->vdev_children; c++)
1672 spa_check_removed(vd->vdev_child[c]);
1673
1674 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1675 !vd->vdev_ishole) {
1676 zfs_post_autoreplace(vd->vdev_spa, vd);
1677 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1678 }
1679 }
1680
1681 static void
spa_config_valid_zaps(vdev_t * vd,vdev_t * mvd)1682 spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1683 {
1684 ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1685
1686 vd->vdev_top_zap = mvd->vdev_top_zap;
1687 vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1688
1689 for (uint64_t i = 0; i < vd->vdev_children; i++) {
1690 spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1691 }
1692 }
1693
1694 /*
1695 * Validate the current config against the MOS config
1696 */
1697 static boolean_t
spa_config_valid(spa_t * spa,nvlist_t * config)1698 spa_config_valid(spa_t *spa, nvlist_t *config)
1699 {
1700 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1701 nvlist_t *nv;
1702
1703 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1704
1705 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1706 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1707
1708 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1709
1710 /*
1711 * If we're doing a normal import, then build up any additional
1712 * diagnostic information about missing devices in this config.
1713 * We'll pass this up to the user for further processing.
1714 */
1715 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1716 nvlist_t **child, *nv;
1717 uint64_t idx = 0;
1718
1719 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1720 KM_SLEEP);
1721 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1722
1723 for (int c = 0; c < rvd->vdev_children; c++) {
1724 vdev_t *tvd = rvd->vdev_child[c];
1725 vdev_t *mtvd = mrvd->vdev_child[c];
1726
1727 if (tvd->vdev_ops == &vdev_missing_ops &&
1728 mtvd->vdev_ops != &vdev_missing_ops &&
1729 mtvd->vdev_islog)
1730 child[idx++] = vdev_config_generate(spa, mtvd,
1731 B_FALSE, 0);
1732 }
1733
1734 if (idx) {
1735 VERIFY(nvlist_add_nvlist_array(nv,
1736 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1737 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1738 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1739
1740 for (int i = 0; i < idx; i++)
1741 nvlist_free(child[i]);
1742 }
1743 nvlist_free(nv);
1744 kmem_free(child, rvd->vdev_children * sizeof (char **));
1745 }
1746
1747 /*
1748 * Compare the root vdev tree with the information we have
1749 * from the MOS config (mrvd). Check each top-level vdev
1750 * with the corresponding MOS config top-level (mtvd).
1751 */
1752 for (int c = 0; c < rvd->vdev_children; c++) {
1753 vdev_t *tvd = rvd->vdev_child[c];
1754 vdev_t *mtvd = mrvd->vdev_child[c];
1755
1756 /*
1757 * Resolve any "missing" vdevs in the current configuration.
1758 * If we find that the MOS config has more accurate information
1759 * about the top-level vdev then use that vdev instead.
1760 */
1761 if (tvd->vdev_ops == &vdev_missing_ops &&
1762 mtvd->vdev_ops != &vdev_missing_ops) {
1763
1764 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1765 continue;
1766
1767 /*
1768 * Device specific actions.
1769 */
1770 if (mtvd->vdev_islog) {
1771 spa_set_log_state(spa, SPA_LOG_CLEAR);
1772 } else {
1773 /*
1774 * XXX - once we have 'readonly' pool
1775 * support we should be able to handle
1776 * missing data devices by transitioning
1777 * the pool to readonly.
1778 */
1779 continue;
1780 }
1781
1782 /*
1783 * Swap the missing vdev with the data we were
1784 * able to obtain from the MOS config.
1785 */
1786 vdev_remove_child(rvd, tvd);
1787 vdev_remove_child(mrvd, mtvd);
1788
1789 vdev_add_child(rvd, mtvd);
1790 vdev_add_child(mrvd, tvd);
1791
1792 spa_config_exit(spa, SCL_ALL, FTAG);
1793 vdev_load(mtvd);
1794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1795
1796 vdev_reopen(rvd);
1797 } else {
1798 if (mtvd->vdev_islog) {
1799 /*
1800 * Load the slog device's state from the MOS
1801 * config since it's possible that the label
1802 * does not contain the most up-to-date
1803 * information.
1804 */
1805 vdev_load_log_state(tvd, mtvd);
1806 vdev_reopen(tvd);
1807 }
1808
1809 /*
1810 * Per-vdev ZAP info is stored exclusively in the MOS.
1811 */
1812 spa_config_valid_zaps(tvd, mtvd);
1813 }
1814 }
1815
1816 vdev_free(mrvd);
1817 spa_config_exit(spa, SCL_ALL, FTAG);
1818
1819 /*
1820 * Ensure we were able to validate the config.
1821 */
1822 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1823 }
1824
1825 /*
1826 * Check for missing log devices
1827 */
1828 static boolean_t
spa_check_logs(spa_t * spa)1829 spa_check_logs(spa_t *spa)
1830 {
1831 boolean_t rv = B_FALSE;
1832 dsl_pool_t *dp = spa_get_dsl(spa);
1833
1834 switch (spa->spa_log_state) {
1835 case SPA_LOG_MISSING:
1836 /* need to recheck in case slog has been restored */
1837 case SPA_LOG_UNKNOWN:
1838 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1839 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1840 if (rv)
1841 spa_set_log_state(spa, SPA_LOG_MISSING);
1842 break;
1843 }
1844 return (rv);
1845 }
1846
1847 static boolean_t
spa_passivate_log(spa_t * spa)1848 spa_passivate_log(spa_t *spa)
1849 {
1850 vdev_t *rvd = spa->spa_root_vdev;
1851 boolean_t slog_found = B_FALSE;
1852
1853 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1854
1855 if (!spa_has_slogs(spa))
1856 return (B_FALSE);
1857
1858 for (int c = 0; c < rvd->vdev_children; c++) {
1859 vdev_t *tvd = rvd->vdev_child[c];
1860 metaslab_group_t *mg = tvd->vdev_mg;
1861
1862 if (tvd->vdev_islog) {
1863 metaslab_group_passivate(mg);
1864 slog_found = B_TRUE;
1865 }
1866 }
1867
1868 return (slog_found);
1869 }
1870
1871 static void
spa_activate_log(spa_t * spa)1872 spa_activate_log(spa_t *spa)
1873 {
1874 vdev_t *rvd = spa->spa_root_vdev;
1875
1876 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1877
1878 for (int c = 0; c < rvd->vdev_children; c++) {
1879 vdev_t *tvd = rvd->vdev_child[c];
1880 metaslab_group_t *mg = tvd->vdev_mg;
1881
1882 if (tvd->vdev_islog)
1883 metaslab_group_activate(mg);
1884 }
1885 }
1886
1887 int
spa_offline_log(spa_t * spa)1888 spa_offline_log(spa_t *spa)
1889 {
1890 int error;
1891
1892 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1893 NULL, DS_FIND_CHILDREN);
1894 if (error == 0) {
1895 /*
1896 * We successfully offlined the log device, sync out the
1897 * current txg so that the "stubby" block can be removed
1898 * by zil_sync().
1899 */
1900 txg_wait_synced(spa->spa_dsl_pool, 0);
1901 }
1902 return (error);
1903 }
1904
1905 static void
spa_aux_check_removed(spa_aux_vdev_t * sav)1906 spa_aux_check_removed(spa_aux_vdev_t *sav)
1907 {
1908 int i;
1909
1910 for (i = 0; i < sav->sav_count; i++)
1911 spa_check_removed(sav->sav_vdevs[i]);
1912 }
1913
1914 void
spa_claim_notify(zio_t * zio)1915 spa_claim_notify(zio_t *zio)
1916 {
1917 spa_t *spa = zio->io_spa;
1918
1919 if (zio->io_error)
1920 return;
1921
1922 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1923 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1924 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1925 mutex_exit(&spa->spa_props_lock);
1926 }
1927
1928 typedef struct spa_load_error {
1929 uint64_t sle_meta_count;
1930 uint64_t sle_data_count;
1931 } spa_load_error_t;
1932
1933 static void
spa_load_verify_done(zio_t * zio)1934 spa_load_verify_done(zio_t *zio)
1935 {
1936 blkptr_t *bp = zio->io_bp;
1937 spa_load_error_t *sle = zio->io_private;
1938 dmu_object_type_t type = BP_GET_TYPE(bp);
1939 int error = zio->io_error;
1940 spa_t *spa = zio->io_spa;
1941
1942 if (error) {
1943 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1944 type != DMU_OT_INTENT_LOG)
1945 atomic_inc_64(&sle->sle_meta_count);
1946 else
1947 atomic_inc_64(&sle->sle_data_count);
1948 }
1949 zio_data_buf_free(zio->io_data, zio->io_size);
1950
1951 mutex_enter(&spa->spa_scrub_lock);
1952 spa->spa_scrub_inflight--;
1953 cv_broadcast(&spa->spa_scrub_io_cv);
1954 mutex_exit(&spa->spa_scrub_lock);
1955 }
1956
1957 /*
1958 * Maximum number of concurrent scrub i/os to create while verifying
1959 * a pool while importing it.
1960 */
1961 int spa_load_verify_maxinflight = 10000;
1962 boolean_t spa_load_verify_metadata = B_TRUE;
1963 boolean_t spa_load_verify_data = B_TRUE;
1964
1965 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1966 &spa_load_verify_maxinflight, 0,
1967 "Maximum number of concurrent scrub I/Os to create while verifying a "
1968 "pool while importing it");
1969
1970 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1971 &spa_load_verify_metadata, 0,
1972 "Check metadata on import?");
1973
1974 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1975 &spa_load_verify_data, 0,
1976 "Check user data on import?");
1977
1978 /*ARGSUSED*/
1979 static int
spa_load_verify_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)1980 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1981 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1982 {
1983 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1984 return (0);
1985 /*
1986 * Note: normally this routine will not be called if
1987 * spa_load_verify_metadata is not set. However, it may be useful
1988 * to manually set the flag after the traversal has begun.
1989 */
1990 if (!spa_load_verify_metadata)
1991 return (0);
1992 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1993 return (0);
1994
1995 zio_t *rio = arg;
1996 size_t size = BP_GET_PSIZE(bp);
1997 void *data = zio_data_buf_alloc(size);
1998
1999 mutex_enter(&spa->spa_scrub_lock);
2000 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2001 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2002 spa->spa_scrub_inflight++;
2003 mutex_exit(&spa->spa_scrub_lock);
2004
2005 zio_nowait(zio_read(rio, spa, bp, data, size,
2006 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2007 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2008 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2009 return (0);
2010 }
2011
2012 /* ARGSUSED */
2013 int
verify_dataset_name_len(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)2014 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2015 {
2016 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2017 return (SET_ERROR(ENAMETOOLONG));
2018
2019 return (0);
2020 }
2021
2022 static int
spa_load_verify(spa_t * spa)2023 spa_load_verify(spa_t *spa)
2024 {
2025 zio_t *rio;
2026 spa_load_error_t sle = { 0 };
2027 zpool_rewind_policy_t policy;
2028 boolean_t verify_ok = B_FALSE;
2029 int error = 0;
2030
2031 zpool_get_rewind_policy(spa->spa_config, &policy);
2032
2033 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2034 return (0);
2035
2036 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2037 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2038 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2039 DS_FIND_CHILDREN);
2040 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2041 if (error != 0)
2042 return (error);
2043
2044 rio = zio_root(spa, NULL, &sle,
2045 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2046
2047 if (spa_load_verify_metadata) {
2048 error = traverse_pool(spa, spa->spa_verify_min_txg,
2049 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2050 spa_load_verify_cb, rio);
2051 }
2052
2053 (void) zio_wait(rio);
2054
2055 spa->spa_load_meta_errors = sle.sle_meta_count;
2056 spa->spa_load_data_errors = sle.sle_data_count;
2057
2058 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2059 sle.sle_data_count <= policy.zrp_maxdata) {
2060 int64_t loss = 0;
2061
2062 verify_ok = B_TRUE;
2063 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2064 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2065
2066 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2067 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2068 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2069 VERIFY(nvlist_add_int64(spa->spa_load_info,
2070 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2071 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2072 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2073 } else {
2074 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2075 }
2076
2077 if (error) {
2078 if (error != ENXIO && error != EIO)
2079 error = SET_ERROR(EIO);
2080 return (error);
2081 }
2082
2083 return (verify_ok ? 0 : EIO);
2084 }
2085
2086 /*
2087 * Find a value in the pool props object.
2088 */
2089 static void
spa_prop_find(spa_t * spa,zpool_prop_t prop,uint64_t * val)2090 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2091 {
2092 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2093 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2094 }
2095
2096 /*
2097 * Find a value in the pool directory object.
2098 */
2099 static int
spa_dir_prop(spa_t * spa,const char * name,uint64_t * val)2100 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2101 {
2102 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2103 name, sizeof (uint64_t), 1, val));
2104 }
2105
2106 static int
spa_vdev_err(vdev_t * vdev,vdev_aux_t aux,int err)2107 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2108 {
2109 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2110 return (err);
2111 }
2112
2113 /*
2114 * Fix up config after a partly-completed split. This is done with the
2115 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2116 * pool have that entry in their config, but only the splitting one contains
2117 * a list of all the guids of the vdevs that are being split off.
2118 *
2119 * This function determines what to do with that list: either rejoin
2120 * all the disks to the pool, or complete the splitting process. To attempt
2121 * the rejoin, each disk that is offlined is marked online again, and
2122 * we do a reopen() call. If the vdev label for every disk that was
2123 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2124 * then we call vdev_split() on each disk, and complete the split.
2125 *
2126 * Otherwise we leave the config alone, with all the vdevs in place in
2127 * the original pool.
2128 */
2129 static void
spa_try_repair(spa_t * spa,nvlist_t * config)2130 spa_try_repair(spa_t *spa, nvlist_t *config)
2131 {
2132 uint_t extracted;
2133 uint64_t *glist;
2134 uint_t i, gcount;
2135 nvlist_t *nvl;
2136 vdev_t **vd;
2137 boolean_t attempt_reopen;
2138
2139 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2140 return;
2141
2142 /* check that the config is complete */
2143 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2144 &glist, &gcount) != 0)
2145 return;
2146
2147 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2148
2149 /* attempt to online all the vdevs & validate */
2150 attempt_reopen = B_TRUE;
2151 for (i = 0; i < gcount; i++) {
2152 if (glist[i] == 0) /* vdev is hole */
2153 continue;
2154
2155 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2156 if (vd[i] == NULL) {
2157 /*
2158 * Don't bother attempting to reopen the disks;
2159 * just do the split.
2160 */
2161 attempt_reopen = B_FALSE;
2162 } else {
2163 /* attempt to re-online it */
2164 vd[i]->vdev_offline = B_FALSE;
2165 }
2166 }
2167
2168 if (attempt_reopen) {
2169 vdev_reopen(spa->spa_root_vdev);
2170
2171 /* check each device to see what state it's in */
2172 for (extracted = 0, i = 0; i < gcount; i++) {
2173 if (vd[i] != NULL &&
2174 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2175 break;
2176 ++extracted;
2177 }
2178 }
2179
2180 /*
2181 * If every disk has been moved to the new pool, or if we never
2182 * even attempted to look at them, then we split them off for
2183 * good.
2184 */
2185 if (!attempt_reopen || gcount == extracted) {
2186 for (i = 0; i < gcount; i++)
2187 if (vd[i] != NULL)
2188 vdev_split(vd[i]);
2189 vdev_reopen(spa->spa_root_vdev);
2190 }
2191
2192 kmem_free(vd, gcount * sizeof (vdev_t *));
2193 }
2194
2195 static int
spa_load(spa_t * spa,spa_load_state_t state,spa_import_type_t type,boolean_t mosconfig)2196 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2197 boolean_t mosconfig)
2198 {
2199 nvlist_t *config = spa->spa_config;
2200 char *ereport = FM_EREPORT_ZFS_POOL;
2201 char *comment;
2202 int error;
2203 uint64_t pool_guid;
2204 nvlist_t *nvl;
2205
2206 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2207 return (SET_ERROR(EINVAL));
2208
2209 ASSERT(spa->spa_comment == NULL);
2210 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2211 spa->spa_comment = spa_strdup(comment);
2212
2213 /*
2214 * Versioning wasn't explicitly added to the label until later, so if
2215 * it's not present treat it as the initial version.
2216 */
2217 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2218 &spa->spa_ubsync.ub_version) != 0)
2219 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2220
2221 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2222 &spa->spa_config_txg);
2223
2224 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2225 spa_guid_exists(pool_guid, 0)) {
2226 error = SET_ERROR(EEXIST);
2227 } else {
2228 spa->spa_config_guid = pool_guid;
2229
2230 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2231 &nvl) == 0) {
2232 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2233 KM_SLEEP) == 0);
2234 }
2235
2236 nvlist_free(spa->spa_load_info);
2237 spa->spa_load_info = fnvlist_alloc();
2238
2239 gethrestime(&spa->spa_loaded_ts);
2240 error = spa_load_impl(spa, pool_guid, config, state, type,
2241 mosconfig, &ereport);
2242 }
2243
2244 /*
2245 * Don't count references from objsets that are already closed
2246 * and are making their way through the eviction process.
2247 */
2248 spa_evicting_os_wait(spa);
2249 spa->spa_minref = refcount_count(&spa->spa_refcount);
2250 if (error) {
2251 if (error != EEXIST) {
2252 spa->spa_loaded_ts.tv_sec = 0;
2253 spa->spa_loaded_ts.tv_nsec = 0;
2254 }
2255 if (error != EBADF) {
2256 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2257 }
2258 }
2259 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2260 spa->spa_ena = 0;
2261
2262 return (error);
2263 }
2264
2265 /*
2266 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2267 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2268 * spa's per-vdev ZAP list.
2269 */
2270 static uint64_t
vdev_count_verify_zaps(vdev_t * vd)2271 vdev_count_verify_zaps(vdev_t *vd)
2272 {
2273 spa_t *spa = vd->vdev_spa;
2274 uint64_t total = 0;
2275 if (vd->vdev_top_zap != 0) {
2276 total++;
2277 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2278 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2279 }
2280 if (vd->vdev_leaf_zap != 0) {
2281 total++;
2282 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2283 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2284 }
2285
2286 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2287 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2288 }
2289
2290 return (total);
2291 }
2292
2293 /*
2294 * Load an existing storage pool, using the pool's builtin spa_config as a
2295 * source of configuration information.
2296 */
2297 static int
spa_load_impl(spa_t * spa,uint64_t pool_guid,nvlist_t * config,spa_load_state_t state,spa_import_type_t type,boolean_t mosconfig,char ** ereport)2298 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2299 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2300 char **ereport)
2301 {
2302 int error = 0;
2303 nvlist_t *nvroot = NULL;
2304 nvlist_t *label;
2305 vdev_t *rvd;
2306 uberblock_t *ub = &spa->spa_uberblock;
2307 uint64_t children, config_cache_txg = spa->spa_config_txg;
2308 int orig_mode = spa->spa_mode;
2309 int parse;
2310 uint64_t obj;
2311 boolean_t missing_feat_write = B_FALSE;
2312
2313 /*
2314 * If this is an untrusted config, access the pool in read-only mode.
2315 * This prevents things like resilvering recently removed devices.
2316 */
2317 if (!mosconfig)
2318 spa->spa_mode = FREAD;
2319
2320 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2321
2322 spa->spa_load_state = state;
2323
2324 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2325 return (SET_ERROR(EINVAL));
2326
2327 parse = (type == SPA_IMPORT_EXISTING ?
2328 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2329
2330 /*
2331 * Create "The Godfather" zio to hold all async IOs
2332 */
2333 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2334 KM_SLEEP);
2335 for (int i = 0; i < max_ncpus; i++) {
2336 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2337 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2338 ZIO_FLAG_GODFATHER);
2339 }
2340
2341 /*
2342 * Parse the configuration into a vdev tree. We explicitly set the
2343 * value that will be returned by spa_version() since parsing the
2344 * configuration requires knowing the version number.
2345 */
2346 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2347 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2348 spa_config_exit(spa, SCL_ALL, FTAG);
2349
2350 if (error != 0)
2351 return (error);
2352
2353 ASSERT(spa->spa_root_vdev == rvd);
2354 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2355 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2356
2357 if (type != SPA_IMPORT_ASSEMBLE) {
2358 ASSERT(spa_guid(spa) == pool_guid);
2359 }
2360
2361 /*
2362 * Try to open all vdevs, loading each label in the process.
2363 */
2364 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2365 error = vdev_open(rvd);
2366 spa_config_exit(spa, SCL_ALL, FTAG);
2367 if (error != 0)
2368 return (error);
2369
2370 /*
2371 * We need to validate the vdev labels against the configuration that
2372 * we have in hand, which is dependent on the setting of mosconfig. If
2373 * mosconfig is true then we're validating the vdev labels based on
2374 * that config. Otherwise, we're validating against the cached config
2375 * (zpool.cache) that was read when we loaded the zfs module, and then
2376 * later we will recursively call spa_load() and validate against
2377 * the vdev config.
2378 *
2379 * If we're assembling a new pool that's been split off from an
2380 * existing pool, the labels haven't yet been updated so we skip
2381 * validation for now.
2382 */
2383 if (type != SPA_IMPORT_ASSEMBLE) {
2384 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2385 error = vdev_validate(rvd, mosconfig);
2386 spa_config_exit(spa, SCL_ALL, FTAG);
2387
2388 if (error != 0)
2389 return (error);
2390
2391 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2392 return (SET_ERROR(ENXIO));
2393 }
2394
2395 /*
2396 * Find the best uberblock.
2397 */
2398 vdev_uberblock_load(rvd, ub, &label);
2399
2400 /*
2401 * If we weren't able to find a single valid uberblock, return failure.
2402 */
2403 if (ub->ub_txg == 0) {
2404 nvlist_free(label);
2405 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2406 }
2407
2408 /*
2409 * If the pool has an unsupported version we can't open it.
2410 */
2411 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2412 nvlist_free(label);
2413 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2414 }
2415
2416 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2417 nvlist_t *features;
2418
2419 /*
2420 * If we weren't able to find what's necessary for reading the
2421 * MOS in the label, return failure.
2422 */
2423 if (label == NULL || nvlist_lookup_nvlist(label,
2424 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2425 nvlist_free(label);
2426 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2427 ENXIO));
2428 }
2429
2430 /*
2431 * Update our in-core representation with the definitive values
2432 * from the label.
2433 */
2434 nvlist_free(spa->spa_label_features);
2435 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2436 }
2437
2438 nvlist_free(label);
2439
2440 /*
2441 * Look through entries in the label nvlist's features_for_read. If
2442 * there is a feature listed there which we don't understand then we
2443 * cannot open a pool.
2444 */
2445 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2446 nvlist_t *unsup_feat;
2447
2448 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2449 0);
2450
2451 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2452 NULL); nvp != NULL;
2453 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2454 if (!zfeature_is_supported(nvpair_name(nvp))) {
2455 VERIFY(nvlist_add_string(unsup_feat,
2456 nvpair_name(nvp), "") == 0);
2457 }
2458 }
2459
2460 if (!nvlist_empty(unsup_feat)) {
2461 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2462 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2463 nvlist_free(unsup_feat);
2464 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2465 ENOTSUP));
2466 }
2467
2468 nvlist_free(unsup_feat);
2469 }
2470
2471 /*
2472 * If the vdev guid sum doesn't match the uberblock, we have an
2473 * incomplete configuration. We first check to see if the pool
2474 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2475 * If it is, defer the vdev_guid_sum check till later so we
2476 * can handle missing vdevs.
2477 */
2478 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2479 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2480 rvd->vdev_guid_sum != ub->ub_guid_sum)
2481 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2482
2483 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2484 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2485 spa_try_repair(spa, config);
2486 spa_config_exit(spa, SCL_ALL, FTAG);
2487 nvlist_free(spa->spa_config_splitting);
2488 spa->spa_config_splitting = NULL;
2489 }
2490
2491 /*
2492 * Initialize internal SPA structures.
2493 */
2494 spa->spa_state = POOL_STATE_ACTIVE;
2495 spa->spa_ubsync = spa->spa_uberblock;
2496 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2497 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2498 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2499 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2500 spa->spa_claim_max_txg = spa->spa_first_txg;
2501 spa->spa_prev_software_version = ub->ub_software_version;
2502
2503 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2504 if (error)
2505 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2506 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2507
2508 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2509 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2510
2511 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2512 boolean_t missing_feat_read = B_FALSE;
2513 nvlist_t *unsup_feat, *enabled_feat;
2514
2515 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2516 &spa->spa_feat_for_read_obj) != 0) {
2517 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2518 }
2519
2520 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2521 &spa->spa_feat_for_write_obj) != 0) {
2522 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2523 }
2524
2525 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2526 &spa->spa_feat_desc_obj) != 0) {
2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528 }
2529
2530 enabled_feat = fnvlist_alloc();
2531 unsup_feat = fnvlist_alloc();
2532
2533 if (!spa_features_check(spa, B_FALSE,
2534 unsup_feat, enabled_feat))
2535 missing_feat_read = B_TRUE;
2536
2537 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2538 if (!spa_features_check(spa, B_TRUE,
2539 unsup_feat, enabled_feat)) {
2540 missing_feat_write = B_TRUE;
2541 }
2542 }
2543
2544 fnvlist_add_nvlist(spa->spa_load_info,
2545 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2546
2547 if (!nvlist_empty(unsup_feat)) {
2548 fnvlist_add_nvlist(spa->spa_load_info,
2549 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2550 }
2551
2552 fnvlist_free(enabled_feat);
2553 fnvlist_free(unsup_feat);
2554
2555 if (!missing_feat_read) {
2556 fnvlist_add_boolean(spa->spa_load_info,
2557 ZPOOL_CONFIG_CAN_RDONLY);
2558 }
2559
2560 /*
2561 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2562 * twofold: to determine whether the pool is available for
2563 * import in read-write mode and (if it is not) whether the
2564 * pool is available for import in read-only mode. If the pool
2565 * is available for import in read-write mode, it is displayed
2566 * as available in userland; if it is not available for import
2567 * in read-only mode, it is displayed as unavailable in
2568 * userland. If the pool is available for import in read-only
2569 * mode but not read-write mode, it is displayed as unavailable
2570 * in userland with a special note that the pool is actually
2571 * available for open in read-only mode.
2572 *
2573 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2574 * missing a feature for write, we must first determine whether
2575 * the pool can be opened read-only before returning to
2576 * userland in order to know whether to display the
2577 * abovementioned note.
2578 */
2579 if (missing_feat_read || (missing_feat_write &&
2580 spa_writeable(spa))) {
2581 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2582 ENOTSUP));
2583 }
2584
2585 /*
2586 * Load refcounts for ZFS features from disk into an in-memory
2587 * cache during SPA initialization.
2588 */
2589 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2590 uint64_t refcount;
2591
2592 error = feature_get_refcount_from_disk(spa,
2593 &spa_feature_table[i], &refcount);
2594 if (error == 0) {
2595 spa->spa_feat_refcount_cache[i] = refcount;
2596 } else if (error == ENOTSUP) {
2597 spa->spa_feat_refcount_cache[i] =
2598 SPA_FEATURE_DISABLED;
2599 } else {
2600 return (spa_vdev_err(rvd,
2601 VDEV_AUX_CORRUPT_DATA, EIO));
2602 }
2603 }
2604 }
2605
2606 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2607 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2608 &spa->spa_feat_enabled_txg_obj) != 0)
2609 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2610 }
2611
2612 spa->spa_is_initializing = B_TRUE;
2613 error = dsl_pool_open(spa->spa_dsl_pool);
2614 spa->spa_is_initializing = B_FALSE;
2615 if (error != 0)
2616 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2617
2618 if (!mosconfig) {
2619 uint64_t hostid;
2620 nvlist_t *policy = NULL, *nvconfig;
2621
2622 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2623 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624
2625 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2626 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2627 char *hostname;
2628 unsigned long myhostid = 0;
2629
2630 VERIFY(nvlist_lookup_string(nvconfig,
2631 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2632
2633 #ifdef _KERNEL
2634 myhostid = zone_get_hostid(NULL);
2635 #else /* _KERNEL */
2636 /*
2637 * We're emulating the system's hostid in userland, so
2638 * we can't use zone_get_hostid().
2639 */
2640 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2641 #endif /* _KERNEL */
2642 if (check_hostid && hostid != 0 && myhostid != 0 &&
2643 hostid != myhostid) {
2644 nvlist_free(nvconfig);
2645 cmn_err(CE_WARN, "pool '%s' could not be "
2646 "loaded as it was last accessed by "
2647 "another system (host: %s hostid: 0x%lx). "
2648 "See: http://illumos.org/msg/ZFS-8000-EY",
2649 spa_name(spa), hostname,
2650 (unsigned long)hostid);
2651 return (SET_ERROR(EBADF));
2652 }
2653 }
2654 if (nvlist_lookup_nvlist(spa->spa_config,
2655 ZPOOL_REWIND_POLICY, &policy) == 0)
2656 VERIFY(nvlist_add_nvlist(nvconfig,
2657 ZPOOL_REWIND_POLICY, policy) == 0);
2658
2659 spa_config_set(spa, nvconfig);
2660 spa_unload(spa);
2661 spa_deactivate(spa);
2662 spa_activate(spa, orig_mode);
2663
2664 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2665 }
2666
2667 /* Grab the secret checksum salt from the MOS. */
2668 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2669 DMU_POOL_CHECKSUM_SALT, 1,
2670 sizeof (spa->spa_cksum_salt.zcs_bytes),
2671 spa->spa_cksum_salt.zcs_bytes);
2672 if (error == ENOENT) {
2673 /* Generate a new salt for subsequent use */
2674 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2675 sizeof (spa->spa_cksum_salt.zcs_bytes));
2676 } else if (error != 0) {
2677 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2678 }
2679
2680 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2681 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2682 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2683 if (error != 0)
2684 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2685
2686 /*
2687 * Load the bit that tells us to use the new accounting function
2688 * (raid-z deflation). If we have an older pool, this will not
2689 * be present.
2690 */
2691 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2692 if (error != 0 && error != ENOENT)
2693 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2694
2695 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2696 &spa->spa_creation_version);
2697 if (error != 0 && error != ENOENT)
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2699
2700 /*
2701 * Load the persistent error log. If we have an older pool, this will
2702 * not be present.
2703 */
2704 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2705 if (error != 0 && error != ENOENT)
2706 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2707
2708 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2709 &spa->spa_errlog_scrub);
2710 if (error != 0 && error != ENOENT)
2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2712
2713 /*
2714 * Load the history object. If we have an older pool, this
2715 * will not be present.
2716 */
2717 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2718 if (error != 0 && error != ENOENT)
2719 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2720
2721 /*
2722 * Load the per-vdev ZAP map. If we have an older pool, this will not
2723 * be present; in this case, defer its creation to a later time to
2724 * avoid dirtying the MOS this early / out of sync context. See
2725 * spa_sync_config_object.
2726 */
2727
2728 /* The sentinel is only available in the MOS config. */
2729 nvlist_t *mos_config;
2730 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2731 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2732
2733 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2734 &spa->spa_all_vdev_zaps);
2735
2736 if (error != ENOENT && error != 0) {
2737 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2738 } else if (error == 0 && !nvlist_exists(mos_config,
2739 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2740 /*
2741 * An older version of ZFS overwrote the sentinel value, so
2742 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2743 * destruction to later; see spa_sync_config_object.
2744 */
2745 spa->spa_avz_action = AVZ_ACTION_DESTROY;
2746 /*
2747 * We're assuming that no vdevs have had their ZAPs created
2748 * before this. Better be sure of it.
2749 */
2750 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2751 }
2752 nvlist_free(mos_config);
2753
2754 /*
2755 * If we're assembling the pool from the split-off vdevs of
2756 * an existing pool, we don't want to attach the spares & cache
2757 * devices.
2758 */
2759
2760 /*
2761 * Load any hot spares for this pool.
2762 */
2763 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2764 if (error != 0 && error != ENOENT)
2765 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2766 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2767 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2768 if (load_nvlist(spa, spa->spa_spares.sav_object,
2769 &spa->spa_spares.sav_config) != 0)
2770 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2771
2772 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2773 spa_load_spares(spa);
2774 spa_config_exit(spa, SCL_ALL, FTAG);
2775 } else if (error == 0) {
2776 spa->spa_spares.sav_sync = B_TRUE;
2777 }
2778
2779 /*
2780 * Load any level 2 ARC devices for this pool.
2781 */
2782 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2783 &spa->spa_l2cache.sav_object);
2784 if (error != 0 && error != ENOENT)
2785 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2786 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2787 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2788 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2789 &spa->spa_l2cache.sav_config) != 0)
2790 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2791
2792 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2793 spa_load_l2cache(spa);
2794 spa_config_exit(spa, SCL_ALL, FTAG);
2795 } else if (error == 0) {
2796 spa->spa_l2cache.sav_sync = B_TRUE;
2797 }
2798
2799 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2800
2801 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2802 if (error && error != ENOENT)
2803 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2804
2805 if (error == 0) {
2806 uint64_t autoreplace;
2807
2808 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2809 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2810 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2811 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2812 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2813 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2814 &spa->spa_dedup_ditto);
2815
2816 spa->spa_autoreplace = (autoreplace != 0);
2817 }
2818
2819 /*
2820 * If the 'autoreplace' property is set, then post a resource notifying
2821 * the ZFS DE that it should not issue any faults for unopenable
2822 * devices. We also iterate over the vdevs, and post a sysevent for any
2823 * unopenable vdevs so that the normal autoreplace handler can take
2824 * over.
2825 */
2826 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2827 spa_check_removed(spa->spa_root_vdev);
2828 /*
2829 * For the import case, this is done in spa_import(), because
2830 * at this point we're using the spare definitions from
2831 * the MOS config, not necessarily from the userland config.
2832 */
2833 if (state != SPA_LOAD_IMPORT) {
2834 spa_aux_check_removed(&spa->spa_spares);
2835 spa_aux_check_removed(&spa->spa_l2cache);
2836 }
2837 }
2838
2839 /*
2840 * Load the vdev state for all toplevel vdevs.
2841 */
2842 vdev_load(rvd);
2843
2844 /*
2845 * Propagate the leaf DTLs we just loaded all the way up the tree.
2846 */
2847 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2848 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2849 spa_config_exit(spa, SCL_ALL, FTAG);
2850
2851 /*
2852 * Load the DDTs (dedup tables).
2853 */
2854 error = ddt_load(spa);
2855 if (error != 0)
2856 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2857
2858 spa_update_dspace(spa);
2859
2860 /*
2861 * Validate the config, using the MOS config to fill in any
2862 * information which might be missing. If we fail to validate
2863 * the config then declare the pool unfit for use. If we're
2864 * assembling a pool from a split, the log is not transferred
2865 * over.
2866 */
2867 if (type != SPA_IMPORT_ASSEMBLE) {
2868 nvlist_t *nvconfig;
2869
2870 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2871 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2872
2873 if (!spa_config_valid(spa, nvconfig)) {
2874 nvlist_free(nvconfig);
2875 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2876 ENXIO));
2877 }
2878 nvlist_free(nvconfig);
2879
2880 /*
2881 * Now that we've validated the config, check the state of the
2882 * root vdev. If it can't be opened, it indicates one or
2883 * more toplevel vdevs are faulted.
2884 */
2885 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2886 return (SET_ERROR(ENXIO));
2887
2888 if (spa_writeable(spa) && spa_check_logs(spa)) {
2889 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2890 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2891 }
2892 }
2893
2894 if (missing_feat_write) {
2895 ASSERT(state == SPA_LOAD_TRYIMPORT);
2896
2897 /*
2898 * At this point, we know that we can open the pool in
2899 * read-only mode but not read-write mode. We now have enough
2900 * information and can return to userland.
2901 */
2902 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2903 }
2904
2905 /*
2906 * We've successfully opened the pool, verify that we're ready
2907 * to start pushing transactions.
2908 */
2909 if (state != SPA_LOAD_TRYIMPORT) {
2910 if (error = spa_load_verify(spa))
2911 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2912 error));
2913 }
2914
2915 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2916 spa->spa_load_max_txg == UINT64_MAX)) {
2917 dmu_tx_t *tx;
2918 int need_update = B_FALSE;
2919 dsl_pool_t *dp = spa_get_dsl(spa);
2920
2921 ASSERT(state != SPA_LOAD_TRYIMPORT);
2922
2923 /*
2924 * Claim log blocks that haven't been committed yet.
2925 * This must all happen in a single txg.
2926 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2927 * invoked from zil_claim_log_block()'s i/o done callback.
2928 * Price of rollback is that we abandon the log.
2929 */
2930 spa->spa_claiming = B_TRUE;
2931
2932 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2933 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2934 zil_claim, tx, DS_FIND_CHILDREN);
2935 dmu_tx_commit(tx);
2936
2937 spa->spa_claiming = B_FALSE;
2938
2939 spa_set_log_state(spa, SPA_LOG_GOOD);
2940 spa->spa_sync_on = B_TRUE;
2941 txg_sync_start(spa->spa_dsl_pool);
2942
2943 /*
2944 * Wait for all claims to sync. We sync up to the highest
2945 * claimed log block birth time so that claimed log blocks
2946 * don't appear to be from the future. spa_claim_max_txg
2947 * will have been set for us by either zil_check_log_chain()
2948 * (invoked from spa_check_logs()) or zil_claim() above.
2949 */
2950 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2951
2952 /*
2953 * If the config cache is stale, or we have uninitialized
2954 * metaslabs (see spa_vdev_add()), then update the config.
2955 *
2956 * If this is a verbatim import, trust the current
2957 * in-core spa_config and update the disk labels.
2958 */
2959 if (config_cache_txg != spa->spa_config_txg ||
2960 state == SPA_LOAD_IMPORT ||
2961 state == SPA_LOAD_RECOVER ||
2962 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2963 need_update = B_TRUE;
2964
2965 for (int c = 0; c < rvd->vdev_children; c++)
2966 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2967 need_update = B_TRUE;
2968
2969 /*
2970 * Update the config cache asychronously in case we're the
2971 * root pool, in which case the config cache isn't writable yet.
2972 */
2973 if (need_update)
2974 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2975
2976 /*
2977 * Check all DTLs to see if anything needs resilvering.
2978 */
2979 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2980 vdev_resilver_needed(rvd, NULL, NULL))
2981 spa_async_request(spa, SPA_ASYNC_RESILVER);
2982
2983 /*
2984 * Log the fact that we booted up (so that we can detect if
2985 * we rebooted in the middle of an operation).
2986 */
2987 spa_history_log_version(spa, "open");
2988
2989 /*
2990 * Delete any inconsistent datasets.
2991 */
2992 (void) dmu_objset_find(spa_name(spa),
2993 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2994
2995 /*
2996 * Clean up any stale temporary dataset userrefs.
2997 */
2998 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2999 }
3000
3001 return (0);
3002 }
3003
3004 static int
spa_load_retry(spa_t * spa,spa_load_state_t state,int mosconfig)3005 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
3006 {
3007 int mode = spa->spa_mode;
3008
3009 spa_unload(spa);
3010 spa_deactivate(spa);
3011
3012 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3013
3014 spa_activate(spa, mode);
3015 spa_async_suspend(spa);
3016
3017 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
3018 }
3019
3020 /*
3021 * If spa_load() fails this function will try loading prior txg's. If
3022 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3023 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3024 * function will not rewind the pool and will return the same error as
3025 * spa_load().
3026 */
3027 static int
spa_load_best(spa_t * spa,spa_load_state_t state,int mosconfig,uint64_t max_request,int rewind_flags)3028 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
3029 uint64_t max_request, int rewind_flags)
3030 {
3031 nvlist_t *loadinfo = NULL;
3032 nvlist_t *config = NULL;
3033 int load_error, rewind_error;
3034 uint64_t safe_rewind_txg;
3035 uint64_t min_txg;
3036
3037 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
3038 spa->spa_load_max_txg = spa->spa_load_txg;
3039 spa_set_log_state(spa, SPA_LOG_CLEAR);
3040 } else {
3041 spa->spa_load_max_txg = max_request;
3042 if (max_request != UINT64_MAX)
3043 spa->spa_extreme_rewind = B_TRUE;
3044 }
3045
3046 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
3047 mosconfig);
3048 if (load_error == 0)
3049 return (0);
3050
3051 if (spa->spa_root_vdev != NULL)
3052 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3053
3054 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
3055 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
3056
3057 if (rewind_flags & ZPOOL_NEVER_REWIND) {
3058 nvlist_free(config);
3059 return (load_error);
3060 }
3061
3062 if (state == SPA_LOAD_RECOVER) {
3063 /* Price of rolling back is discarding txgs, including log */
3064 spa_set_log_state(spa, SPA_LOG_CLEAR);
3065 } else {
3066 /*
3067 * If we aren't rolling back save the load info from our first
3068 * import attempt so that we can restore it after attempting
3069 * to rewind.
3070 */
3071 loadinfo = spa->spa_load_info;
3072 spa->spa_load_info = fnvlist_alloc();
3073 }
3074
3075 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3076 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3077 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3078 TXG_INITIAL : safe_rewind_txg;
3079
3080 /*
3081 * Continue as long as we're finding errors, we're still within
3082 * the acceptable rewind range, and we're still finding uberblocks
3083 */
3084 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3085 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3086 if (spa->spa_load_max_txg < safe_rewind_txg)
3087 spa->spa_extreme_rewind = B_TRUE;
3088 rewind_error = spa_load_retry(spa, state, mosconfig);
3089 }
3090
3091 spa->spa_extreme_rewind = B_FALSE;
3092 spa->spa_load_max_txg = UINT64_MAX;
3093
3094 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3095 spa_config_set(spa, config);
3096
3097 if (state == SPA_LOAD_RECOVER) {
3098 ASSERT3P(loadinfo, ==, NULL);
3099 return (rewind_error);
3100 } else {
3101 /* Store the rewind info as part of the initial load info */
3102 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3103 spa->spa_load_info);
3104
3105 /* Restore the initial load info */
3106 fnvlist_free(spa->spa_load_info);
3107 spa->spa_load_info = loadinfo;
3108
3109 return (load_error);
3110 }
3111 }
3112
3113 /*
3114 * Pool Open/Import
3115 *
3116 * The import case is identical to an open except that the configuration is sent
3117 * down from userland, instead of grabbed from the configuration cache. For the
3118 * case of an open, the pool configuration will exist in the
3119 * POOL_STATE_UNINITIALIZED state.
3120 *
3121 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3122 * the same time open the pool, without having to keep around the spa_t in some
3123 * ambiguous state.
3124 */
3125 static int
spa_open_common(const char * pool,spa_t ** spapp,void * tag,nvlist_t * nvpolicy,nvlist_t ** config)3126 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3127 nvlist_t **config)
3128 {
3129 spa_t *spa;
3130 spa_load_state_t state = SPA_LOAD_OPEN;
3131 int error;
3132 int locked = B_FALSE;
3133 int firstopen = B_FALSE;
3134
3135 *spapp = NULL;
3136
3137 /*
3138 * As disgusting as this is, we need to support recursive calls to this
3139 * function because dsl_dir_open() is called during spa_load(), and ends
3140 * up calling spa_open() again. The real fix is to figure out how to
3141 * avoid dsl_dir_open() calling this in the first place.
3142 */
3143 if (!mutex_owned(&spa_namespace_lock)) {
3144 mutex_enter(&spa_namespace_lock);
3145 locked = B_TRUE;
3146 }
3147
3148 if ((spa = spa_lookup(pool)) == NULL) {
3149 if (locked)
3150 mutex_exit(&spa_namespace_lock);
3151 return (SET_ERROR(ENOENT));
3152 }
3153
3154 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3155 zpool_rewind_policy_t policy;
3156
3157 firstopen = B_TRUE;
3158
3159 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3160 &policy);
3161 if (policy.zrp_request & ZPOOL_DO_REWIND)
3162 state = SPA_LOAD_RECOVER;
3163
3164 spa_activate(spa, spa_mode_global);
3165
3166 if (state != SPA_LOAD_RECOVER)
3167 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3168
3169 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3170 policy.zrp_request);
3171
3172 if (error == EBADF) {
3173 /*
3174 * If vdev_validate() returns failure (indicated by
3175 * EBADF), it indicates that one of the vdevs indicates
3176 * that the pool has been exported or destroyed. If
3177 * this is the case, the config cache is out of sync and
3178 * we should remove the pool from the namespace.
3179 */
3180 spa_unload(spa);
3181 spa_deactivate(spa);
3182 spa_config_sync(spa, B_TRUE, B_TRUE);
3183 spa_remove(spa);
3184 if (locked)
3185 mutex_exit(&spa_namespace_lock);
3186 return (SET_ERROR(ENOENT));
3187 }
3188
3189 if (error) {
3190 /*
3191 * We can't open the pool, but we still have useful
3192 * information: the state of each vdev after the
3193 * attempted vdev_open(). Return this to the user.
3194 */
3195 if (config != NULL && spa->spa_config) {
3196 VERIFY(nvlist_dup(spa->spa_config, config,
3197 KM_SLEEP) == 0);
3198 VERIFY(nvlist_add_nvlist(*config,
3199 ZPOOL_CONFIG_LOAD_INFO,
3200 spa->spa_load_info) == 0);
3201 }
3202 spa_unload(spa);
3203 spa_deactivate(spa);
3204 spa->spa_last_open_failed = error;
3205 if (locked)
3206 mutex_exit(&spa_namespace_lock);
3207 *spapp = NULL;
3208 return (error);
3209 }
3210 }
3211
3212 spa_open_ref(spa, tag);
3213
3214 if (config != NULL)
3215 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3216
3217 /*
3218 * If we've recovered the pool, pass back any information we
3219 * gathered while doing the load.
3220 */
3221 if (state == SPA_LOAD_RECOVER) {
3222 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3223 spa->spa_load_info) == 0);
3224 }
3225
3226 if (locked) {
3227 spa->spa_last_open_failed = 0;
3228 spa->spa_last_ubsync_txg = 0;
3229 spa->spa_load_txg = 0;
3230 mutex_exit(&spa_namespace_lock);
3231 #if defined(__FreeBSD__) || defined(__NetBSD__)
3232 #ifdef _KERNEL
3233 if (firstopen)
3234 zvol_create_minors(spa->spa_name);
3235 #endif
3236 #endif
3237 }
3238
3239 *spapp = spa;
3240
3241 return (0);
3242 }
3243
3244 int
spa_open_rewind(const char * name,spa_t ** spapp,void * tag,nvlist_t * policy,nvlist_t ** config)3245 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3246 nvlist_t **config)
3247 {
3248 return (spa_open_common(name, spapp, tag, policy, config));
3249 }
3250
3251 int
spa_open(const char * name,spa_t ** spapp,void * tag)3252 spa_open(const char *name, spa_t **spapp, void *tag)
3253 {
3254 return (spa_open_common(name, spapp, tag, NULL, NULL));
3255 }
3256
3257 /*
3258 * Lookup the given spa_t, incrementing the inject count in the process,
3259 * preventing it from being exported or destroyed.
3260 */
3261 spa_t *
spa_inject_addref(char * name)3262 spa_inject_addref(char *name)
3263 {
3264 spa_t *spa;
3265
3266 mutex_enter(&spa_namespace_lock);
3267 if ((spa = spa_lookup(name)) == NULL) {
3268 mutex_exit(&spa_namespace_lock);
3269 return (NULL);
3270 }
3271 spa->spa_inject_ref++;
3272 mutex_exit(&spa_namespace_lock);
3273
3274 return (spa);
3275 }
3276
3277 void
spa_inject_delref(spa_t * spa)3278 spa_inject_delref(spa_t *spa)
3279 {
3280 mutex_enter(&spa_namespace_lock);
3281 spa->spa_inject_ref--;
3282 mutex_exit(&spa_namespace_lock);
3283 }
3284
3285 /*
3286 * Add spares device information to the nvlist.
3287 */
3288 static void
spa_add_spares(spa_t * spa,nvlist_t * config)3289 spa_add_spares(spa_t *spa, nvlist_t *config)
3290 {
3291 nvlist_t **spares;
3292 uint_t i, nspares;
3293 nvlist_t *nvroot;
3294 uint64_t guid;
3295 vdev_stat_t *vs;
3296 uint_t vsc;
3297 uint64_t pool;
3298
3299 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3300
3301 if (spa->spa_spares.sav_count == 0)
3302 return;
3303
3304 VERIFY(nvlist_lookup_nvlist(config,
3305 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3306 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3307 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3308 if (nspares != 0) {
3309 VERIFY(nvlist_add_nvlist_array(nvroot,
3310 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3311 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3312 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3313
3314 /*
3315 * Go through and find any spares which have since been
3316 * repurposed as an active spare. If this is the case, update
3317 * their status appropriately.
3318 */
3319 for (i = 0; i < nspares; i++) {
3320 VERIFY(nvlist_lookup_uint64(spares[i],
3321 ZPOOL_CONFIG_GUID, &guid) == 0);
3322 if (spa_spare_exists(guid, &pool, NULL) &&
3323 pool != 0ULL) {
3324 VERIFY(nvlist_lookup_uint64_array(
3325 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3326 (uint64_t **)&vs, &vsc) == 0);
3327 vs->vs_state = VDEV_STATE_CANT_OPEN;
3328 vs->vs_aux = VDEV_AUX_SPARED;
3329 }
3330 }
3331 }
3332 }
3333
3334 /*
3335 * Add l2cache device information to the nvlist, including vdev stats.
3336 */
3337 static void
spa_add_l2cache(spa_t * spa,nvlist_t * config)3338 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3339 {
3340 nvlist_t **l2cache;
3341 uint_t i, j, nl2cache;
3342 nvlist_t *nvroot;
3343 uint64_t guid;
3344 vdev_t *vd;
3345 vdev_stat_t *vs;
3346 uint_t vsc;
3347
3348 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3349
3350 if (spa->spa_l2cache.sav_count == 0)
3351 return;
3352
3353 VERIFY(nvlist_lookup_nvlist(config,
3354 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3355 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3356 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3357 if (nl2cache != 0) {
3358 VERIFY(nvlist_add_nvlist_array(nvroot,
3359 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3360 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3361 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3362
3363 /*
3364 * Update level 2 cache device stats.
3365 */
3366
3367 for (i = 0; i < nl2cache; i++) {
3368 VERIFY(nvlist_lookup_uint64(l2cache[i],
3369 ZPOOL_CONFIG_GUID, &guid) == 0);
3370
3371 vd = NULL;
3372 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3373 if (guid ==
3374 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3375 vd = spa->spa_l2cache.sav_vdevs[j];
3376 break;
3377 }
3378 }
3379 ASSERT(vd != NULL);
3380
3381 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3382 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3383 == 0);
3384 vdev_get_stats(vd, vs);
3385 }
3386 }
3387 }
3388
3389 static void
spa_add_feature_stats(spa_t * spa,nvlist_t * config)3390 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3391 {
3392 nvlist_t *features;
3393 zap_cursor_t zc;
3394 zap_attribute_t za;
3395
3396 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3397 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3398
3399 /* We may be unable to read features if pool is suspended. */
3400 if (spa_suspended(spa))
3401 goto out;
3402
3403 if (spa->spa_feat_for_read_obj != 0) {
3404 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3405 spa->spa_feat_for_read_obj);
3406 zap_cursor_retrieve(&zc, &za) == 0;
3407 zap_cursor_advance(&zc)) {
3408 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3409 za.za_num_integers == 1);
3410 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3411 za.za_first_integer));
3412 }
3413 zap_cursor_fini(&zc);
3414 }
3415
3416 if (spa->spa_feat_for_write_obj != 0) {
3417 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3418 spa->spa_feat_for_write_obj);
3419 zap_cursor_retrieve(&zc, &za) == 0;
3420 zap_cursor_advance(&zc)) {
3421 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3422 za.za_num_integers == 1);
3423 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3424 za.za_first_integer));
3425 }
3426 zap_cursor_fini(&zc);
3427 }
3428
3429 out:
3430 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3431 features) == 0);
3432 nvlist_free(features);
3433 }
3434
3435 int
spa_get_stats(const char * name,nvlist_t ** config,char * altroot,size_t buflen)3436 spa_get_stats(const char *name, nvlist_t **config,
3437 char *altroot, size_t buflen)
3438 {
3439 int error;
3440 spa_t *spa;
3441
3442 *config = NULL;
3443 error = spa_open_common(name, &spa, FTAG, NULL, config);
3444
3445 if (spa != NULL) {
3446 /*
3447 * This still leaves a window of inconsistency where the spares
3448 * or l2cache devices could change and the config would be
3449 * self-inconsistent.
3450 */
3451 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3452
3453 if (*config != NULL) {
3454 uint64_t loadtimes[2];
3455
3456 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3457 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3458 VERIFY(nvlist_add_uint64_array(*config,
3459 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3460
3461 VERIFY(nvlist_add_uint64(*config,
3462 ZPOOL_CONFIG_ERRCOUNT,
3463 spa_get_errlog_size(spa)) == 0);
3464
3465 if (spa_suspended(spa))
3466 VERIFY(nvlist_add_uint64(*config,
3467 ZPOOL_CONFIG_SUSPENDED,
3468 spa->spa_failmode) == 0);
3469
3470 spa_add_spares(spa, *config);
3471 spa_add_l2cache(spa, *config);
3472 spa_add_feature_stats(spa, *config);
3473 }
3474 }
3475
3476 /*
3477 * We want to get the alternate root even for faulted pools, so we cheat
3478 * and call spa_lookup() directly.
3479 */
3480 if (altroot) {
3481 if (spa == NULL) {
3482 mutex_enter(&spa_namespace_lock);
3483 spa = spa_lookup(name);
3484 if (spa)
3485 spa_altroot(spa, altroot, buflen);
3486 else
3487 altroot[0] = '\0';
3488 spa = NULL;
3489 mutex_exit(&spa_namespace_lock);
3490 } else {
3491 spa_altroot(spa, altroot, buflen);
3492 }
3493 }
3494
3495 if (spa != NULL) {
3496 spa_config_exit(spa, SCL_CONFIG, FTAG);
3497 spa_close(spa, FTAG);
3498 }
3499
3500 return (error);
3501 }
3502
3503 /*
3504 * Validate that the auxiliary device array is well formed. We must have an
3505 * array of nvlists, each which describes a valid leaf vdev. If this is an
3506 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3507 * specified, as long as they are well-formed.
3508 */
3509 static int
spa_validate_aux_devs(spa_t * spa,nvlist_t * nvroot,uint64_t crtxg,int mode,spa_aux_vdev_t * sav,const char * config,uint64_t version,vdev_labeltype_t label)3510 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3511 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3512 vdev_labeltype_t label)
3513 {
3514 nvlist_t **dev;
3515 uint_t i, ndev;
3516 vdev_t *vd;
3517 int error;
3518
3519 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3520
3521 /*
3522 * It's acceptable to have no devs specified.
3523 */
3524 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3525 return (0);
3526
3527 if (ndev == 0)
3528 return (SET_ERROR(EINVAL));
3529
3530 /*
3531 * Make sure the pool is formatted with a version that supports this
3532 * device type.
3533 */
3534 if (spa_version(spa) < version)
3535 return (SET_ERROR(ENOTSUP));
3536
3537 /*
3538 * Set the pending device list so we correctly handle device in-use
3539 * checking.
3540 */
3541 sav->sav_pending = dev;
3542 sav->sav_npending = ndev;
3543
3544 for (i = 0; i < ndev; i++) {
3545 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3546 mode)) != 0)
3547 goto out;
3548
3549 if (!vd->vdev_ops->vdev_op_leaf) {
3550 vdev_free(vd);
3551 error = SET_ERROR(EINVAL);
3552 goto out;
3553 }
3554
3555 /*
3556 * The L2ARC currently only supports disk devices in
3557 * kernel context. For user-level testing, we allow it.
3558 */
3559 #ifdef _KERNEL
3560 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3561 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3562 error = SET_ERROR(ENOTBLK);
3563 vdev_free(vd);
3564 goto out;
3565 }
3566 #endif
3567 vd->vdev_top = vd;
3568
3569 if ((error = vdev_open(vd)) == 0 &&
3570 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3571 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3572 vd->vdev_guid) == 0);
3573 }
3574
3575 vdev_free(vd);
3576
3577 if (error &&
3578 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3579 goto out;
3580 else
3581 error = 0;
3582 }
3583
3584 out:
3585 sav->sav_pending = NULL;
3586 sav->sav_npending = 0;
3587 return (error);
3588 }
3589
3590 static int
spa_validate_aux(spa_t * spa,nvlist_t * nvroot,uint64_t crtxg,int mode)3591 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3592 {
3593 int error;
3594
3595 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3596
3597 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3598 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3599 VDEV_LABEL_SPARE)) != 0) {
3600 return (error);
3601 }
3602
3603 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3604 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3605 VDEV_LABEL_L2CACHE));
3606 }
3607
3608 static void
spa_set_aux_vdevs(spa_aux_vdev_t * sav,nvlist_t ** devs,int ndevs,const char * config)3609 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3610 const char *config)
3611 {
3612 int i;
3613
3614 if (sav->sav_config != NULL) {
3615 nvlist_t **olddevs;
3616 uint_t oldndevs;
3617 nvlist_t **newdevs;
3618
3619 /*
3620 * Generate new dev list by concatentating with the
3621 * current dev list.
3622 */
3623 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3624 &olddevs, &oldndevs) == 0);
3625
3626 newdevs = kmem_alloc(sizeof (void *) *
3627 (ndevs + oldndevs), KM_SLEEP);
3628 for (i = 0; i < oldndevs; i++)
3629 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3630 KM_SLEEP) == 0);
3631 for (i = 0; i < ndevs; i++)
3632 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3633 KM_SLEEP) == 0);
3634
3635 VERIFY(nvlist_remove(sav->sav_config, config,
3636 DATA_TYPE_NVLIST_ARRAY) == 0);
3637
3638 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3639 config, newdevs, ndevs + oldndevs) == 0);
3640 for (i = 0; i < oldndevs + ndevs; i++)
3641 nvlist_free(newdevs[i]);
3642 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3643 } else {
3644 /*
3645 * Generate a new dev list.
3646 */
3647 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3648 KM_SLEEP) == 0);
3649 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3650 devs, ndevs) == 0);
3651 }
3652 }
3653
3654 /*
3655 * Stop and drop level 2 ARC devices
3656 */
3657 void
spa_l2cache_drop(spa_t * spa)3658 spa_l2cache_drop(spa_t *spa)
3659 {
3660 vdev_t *vd;
3661 int i;
3662 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3663
3664 for (i = 0; i < sav->sav_count; i++) {
3665 uint64_t pool;
3666
3667 vd = sav->sav_vdevs[i];
3668 ASSERT(vd != NULL);
3669
3670 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3671 pool != 0ULL && l2arc_vdev_present(vd))
3672 l2arc_remove_vdev(vd);
3673 }
3674 }
3675
3676 /*
3677 * Pool Creation
3678 */
3679 int
spa_create(const char * pool,nvlist_t * nvroot,nvlist_t * props,nvlist_t * zplprops)3680 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3681 nvlist_t *zplprops)
3682 {
3683 spa_t *spa;
3684 char *altroot = NULL;
3685 vdev_t *rvd;
3686 dsl_pool_t *dp;
3687 dmu_tx_t *tx;
3688 int error = 0;
3689 uint64_t txg = TXG_INITIAL;
3690 nvlist_t **spares, **l2cache;
3691 uint_t nspares, nl2cache;
3692 uint64_t version, obj;
3693 boolean_t has_features;
3694
3695 /*
3696 * If this pool already exists, return failure.
3697 */
3698 mutex_enter(&spa_namespace_lock);
3699 if (spa_lookup(pool) != NULL) {
3700 mutex_exit(&spa_namespace_lock);
3701 return (SET_ERROR(EEXIST));
3702 }
3703
3704 /*
3705 * Allocate a new spa_t structure.
3706 */
3707 (void) nvlist_lookup_string(props,
3708 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3709 spa = spa_add(pool, NULL, altroot);
3710 spa_activate(spa, spa_mode_global);
3711
3712 if (props && (error = spa_prop_validate(spa, props))) {
3713 spa_deactivate(spa);
3714 spa_remove(spa);
3715 mutex_exit(&spa_namespace_lock);
3716 return (error);
3717 }
3718
3719 has_features = B_FALSE;
3720 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3721 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3722 if (zpool_prop_feature(nvpair_name(elem)))
3723 has_features = B_TRUE;
3724 }
3725
3726 if (has_features || nvlist_lookup_uint64(props,
3727 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3728 version = SPA_VERSION;
3729 }
3730 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3731
3732 spa->spa_first_txg = txg;
3733 spa->spa_uberblock.ub_txg = txg - 1;
3734 spa->spa_uberblock.ub_version = version;
3735 spa->spa_ubsync = spa->spa_uberblock;
3736 spa->spa_load_state = SPA_LOAD_CREATE;
3737
3738 /*
3739 * Create "The Godfather" zio to hold all async IOs
3740 */
3741 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3742 KM_SLEEP);
3743 for (int i = 0; i < max_ncpus; i++) {
3744 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3745 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3746 ZIO_FLAG_GODFATHER);
3747 }
3748
3749 /*
3750 * Create the root vdev.
3751 */
3752 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3753
3754 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3755
3756 ASSERT(error != 0 || rvd != NULL);
3757 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3758
3759 if (error == 0 && !zfs_allocatable_devs(nvroot))
3760 error = SET_ERROR(EINVAL);
3761
3762 if (error == 0 &&
3763 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3764 (error = spa_validate_aux(spa, nvroot, txg,
3765 VDEV_ALLOC_ADD)) == 0) {
3766 for (int c = 0; c < rvd->vdev_children; c++) {
3767 vdev_ashift_optimize(rvd->vdev_child[c]);
3768 vdev_metaslab_set_size(rvd->vdev_child[c]);
3769 vdev_expand(rvd->vdev_child[c], txg);
3770 }
3771 }
3772
3773 spa_config_exit(spa, SCL_ALL, FTAG);
3774
3775 if (error != 0) {
3776 spa_unload(spa);
3777 spa_deactivate(spa);
3778 spa_remove(spa);
3779 mutex_exit(&spa_namespace_lock);
3780 return (error);
3781 }
3782
3783 /*
3784 * Get the list of spares, if specified.
3785 */
3786 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3787 &spares, &nspares) == 0) {
3788 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3789 KM_SLEEP) == 0);
3790 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3791 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3792 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3793 spa_load_spares(spa);
3794 spa_config_exit(spa, SCL_ALL, FTAG);
3795 spa->spa_spares.sav_sync = B_TRUE;
3796 }
3797
3798 /*
3799 * Get the list of level 2 cache devices, if specified.
3800 */
3801 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3802 &l2cache, &nl2cache) == 0) {
3803 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3804 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3805 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3806 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3807 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3808 spa_load_l2cache(spa);
3809 spa_config_exit(spa, SCL_ALL, FTAG);
3810 spa->spa_l2cache.sav_sync = B_TRUE;
3811 }
3812
3813 spa->spa_is_initializing = B_TRUE;
3814 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3815 spa->spa_meta_objset = dp->dp_meta_objset;
3816 spa->spa_is_initializing = B_FALSE;
3817
3818 /*
3819 * Create DDTs (dedup tables).
3820 */
3821 ddt_create(spa);
3822
3823 spa_update_dspace(spa);
3824
3825 tx = dmu_tx_create_assigned(dp, txg);
3826
3827 /*
3828 * Create the pool config object.
3829 */
3830 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3831 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3832 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3833
3834 if (zap_add(spa->spa_meta_objset,
3835 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3836 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3837 cmn_err(CE_PANIC, "failed to add pool config");
3838 }
3839
3840 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3841 spa_feature_create_zap_objects(spa, tx);
3842
3843 if (zap_add(spa->spa_meta_objset,
3844 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3845 sizeof (uint64_t), 1, &version, tx) != 0) {
3846 cmn_err(CE_PANIC, "failed to add pool version");
3847 }
3848
3849 /* Newly created pools with the right version are always deflated. */
3850 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3851 spa->spa_deflate = TRUE;
3852 if (zap_add(spa->spa_meta_objset,
3853 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3854 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3855 cmn_err(CE_PANIC, "failed to add deflate");
3856 }
3857 }
3858
3859 /*
3860 * Create the deferred-free bpobj. Turn off compression
3861 * because sync-to-convergence takes longer if the blocksize
3862 * keeps changing.
3863 */
3864 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3865 dmu_object_set_compress(spa->spa_meta_objset, obj,
3866 ZIO_COMPRESS_OFF, tx);
3867 if (zap_add(spa->spa_meta_objset,
3868 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3869 sizeof (uint64_t), 1, &obj, tx) != 0) {
3870 cmn_err(CE_PANIC, "failed to add bpobj");
3871 }
3872 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3873 spa->spa_meta_objset, obj));
3874
3875 /*
3876 * Create the pool's history object.
3877 */
3878 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3879 spa_history_create_obj(spa, tx);
3880
3881 /*
3882 * Generate some random noise for salted checksums to operate on.
3883 */
3884 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3885 sizeof (spa->spa_cksum_salt.zcs_bytes));
3886
3887 /*
3888 * Set pool properties.
3889 */
3890 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3891 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3892 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3893 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3894
3895 if (props != NULL) {
3896 spa_configfile_set(spa, props, B_FALSE);
3897 spa_sync_props(props, tx);
3898 }
3899
3900 dmu_tx_commit(tx);
3901
3902 spa->spa_sync_on = B_TRUE;
3903 txg_sync_start(spa->spa_dsl_pool);
3904
3905 /*
3906 * We explicitly wait for the first transaction to complete so that our
3907 * bean counters are appropriately updated.
3908 */
3909 txg_wait_synced(spa->spa_dsl_pool, txg);
3910
3911 spa_config_sync(spa, B_FALSE, B_TRUE);
3912 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3913
3914 spa_history_log_version(spa, "create");
3915
3916 /*
3917 * Don't count references from objsets that are already closed
3918 * and are making their way through the eviction process.
3919 */
3920 spa_evicting_os_wait(spa);
3921 spa->spa_minref = refcount_count(&spa->spa_refcount);
3922 spa->spa_load_state = SPA_LOAD_NONE;
3923
3924 mutex_exit(&spa_namespace_lock);
3925
3926 return (0);
3927 }
3928
3929 #ifndef __NetBSD__
3930 #ifdef _KERNEL
3931 #ifdef illumos
3932 /*
3933 * Get the root pool information from the root disk, then import the root pool
3934 * during the system boot up time.
3935 */
3936 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3937
3938 static nvlist_t *
spa_generate_rootconf(char * devpath,char * devid,uint64_t * guid)3939 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3940 {
3941 nvlist_t *config;
3942 nvlist_t *nvtop, *nvroot;
3943 uint64_t pgid;
3944
3945 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3946 return (NULL);
3947
3948 /*
3949 * Add this top-level vdev to the child array.
3950 */
3951 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3952 &nvtop) == 0);
3953 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3954 &pgid) == 0);
3955 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3956
3957 /*
3958 * Put this pool's top-level vdevs into a root vdev.
3959 */
3960 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3961 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3962 VDEV_TYPE_ROOT) == 0);
3963 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3964 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3965 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3966 &nvtop, 1) == 0);
3967
3968 /*
3969 * Replace the existing vdev_tree with the new root vdev in
3970 * this pool's configuration (remove the old, add the new).
3971 */
3972 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3973 nvlist_free(nvroot);
3974 return (config);
3975 }
3976
3977 /*
3978 * Walk the vdev tree and see if we can find a device with "better"
3979 * configuration. A configuration is "better" if the label on that
3980 * device has a more recent txg.
3981 */
3982 static void
spa_alt_rootvdev(vdev_t * vd,vdev_t ** avd,uint64_t * txg)3983 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3984 {
3985 for (int c = 0; c < vd->vdev_children; c++)
3986 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3987
3988 if (vd->vdev_ops->vdev_op_leaf) {
3989 nvlist_t *label;
3990 uint64_t label_txg;
3991
3992 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3993 &label) != 0)
3994 return;
3995
3996 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3997 &label_txg) == 0);
3998
3999 /*
4000 * Do we have a better boot device?
4001 */
4002 if (label_txg > *txg) {
4003 *txg = label_txg;
4004 *avd = vd;
4005 }
4006 nvlist_free(label);
4007 }
4008 }
4009
4010 /*
4011 * Import a root pool.
4012 *
4013 * For x86. devpath_list will consist of devid and/or physpath name of
4014 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4015 * The GRUB "findroot" command will return the vdev we should boot.
4016 *
4017 * For Sparc, devpath_list consists the physpath name of the booting device
4018 * no matter the rootpool is a single device pool or a mirrored pool.
4019 * e.g.
4020 * "/pci@1f,0/ide@d/disk@0,0:a"
4021 */
4022 int
spa_import_rootpool(char * devpath,char * devid)4023 spa_import_rootpool(char *devpath, char *devid)
4024 {
4025 spa_t *spa;
4026 vdev_t *rvd, *bvd, *avd = NULL;
4027 nvlist_t *config, *nvtop;
4028 uint64_t guid, txg;
4029 char *pname;
4030 int error;
4031
4032 /*
4033 * Read the label from the boot device and generate a configuration.
4034 */
4035 config = spa_generate_rootconf(devpath, devid, &guid);
4036 #if defined(_OBP) && defined(_KERNEL)
4037 if (config == NULL) {
4038 if (strstr(devpath, "/iscsi/ssd") != NULL) {
4039 /* iscsi boot */
4040 get_iscsi_bootpath_phy(devpath);
4041 config = spa_generate_rootconf(devpath, devid, &guid);
4042 }
4043 }
4044 #endif
4045 if (config == NULL) {
4046 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
4047 devpath);
4048 return (SET_ERROR(EIO));
4049 }
4050
4051 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4052 &pname) == 0);
4053 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
4054
4055 mutex_enter(&spa_namespace_lock);
4056 if ((spa = spa_lookup(pname)) != NULL) {
4057 /*
4058 * Remove the existing root pool from the namespace so that we
4059 * can replace it with the correct config we just read in.
4060 */
4061 spa_remove(spa);
4062 }
4063
4064 spa = spa_add(pname, config, NULL);
4065 spa->spa_is_root = B_TRUE;
4066 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4067
4068 /*
4069 * Build up a vdev tree based on the boot device's label config.
4070 */
4071 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4072 &nvtop) == 0);
4073 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4074 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4075 VDEV_ALLOC_ROOTPOOL);
4076 spa_config_exit(spa, SCL_ALL, FTAG);
4077 if (error) {
4078 mutex_exit(&spa_namespace_lock);
4079 nvlist_free(config);
4080 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4081 pname);
4082 return (error);
4083 }
4084
4085 /*
4086 * Get the boot vdev.
4087 */
4088 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4089 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4090 (u_longlong_t)guid);
4091 error = SET_ERROR(ENOENT);
4092 goto out;
4093 }
4094
4095 /*
4096 * Determine if there is a better boot device.
4097 */
4098 avd = bvd;
4099 spa_alt_rootvdev(rvd, &avd, &txg);
4100 if (avd != bvd) {
4101 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4102 "try booting from '%s'", avd->vdev_path);
4103 error = SET_ERROR(EINVAL);
4104 goto out;
4105 }
4106
4107 /*
4108 * If the boot device is part of a spare vdev then ensure that
4109 * we're booting off the active spare.
4110 */
4111 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4112 !bvd->vdev_isspare) {
4113 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4114 "try booting from '%s'",
4115 bvd->vdev_parent->
4116 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4117 error = SET_ERROR(EINVAL);
4118 goto out;
4119 }
4120
4121 error = 0;
4122 out:
4123 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4124 vdev_free(rvd);
4125 spa_config_exit(spa, SCL_ALL, FTAG);
4126 mutex_exit(&spa_namespace_lock);
4127
4128 nvlist_free(config);
4129 return (error);
4130 }
4131
4132 #else /* !illumos */
4133
4134 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4135 uint64_t *count);
4136
4137 static nvlist_t *
spa_generate_rootconf(const char * name)4138 spa_generate_rootconf(const char *name)
4139 {
4140 nvlist_t **configs, **tops;
4141 nvlist_t *config;
4142 nvlist_t *best_cfg, *nvtop, *nvroot;
4143 uint64_t *holes;
4144 uint64_t best_txg;
4145 uint64_t nchildren;
4146 uint64_t pgid;
4147 uint64_t count;
4148 uint64_t i;
4149 uint_t nholes;
4150
4151 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4152 return (NULL);
4153
4154 ASSERT3U(count, !=, 0);
4155 best_txg = 0;
4156 for (i = 0; i < count; i++) {
4157 uint64_t txg;
4158
4159 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4160 &txg) == 0);
4161 if (txg > best_txg) {
4162 best_txg = txg;
4163 best_cfg = configs[i];
4164 }
4165 }
4166
4167 nchildren = 1;
4168 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4169 holes = NULL;
4170 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4171 &holes, &nholes);
4172
4173 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4174 for (i = 0; i < nchildren; i++) {
4175 if (i >= count)
4176 break;
4177 if (configs[i] == NULL)
4178 continue;
4179 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4180 &nvtop) == 0);
4181 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4182 }
4183 for (i = 0; holes != NULL && i < nholes; i++) {
4184 if (i >= nchildren)
4185 continue;
4186 if (tops[holes[i]] != NULL)
4187 continue;
4188 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4189 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4190 VDEV_TYPE_HOLE) == 0);
4191 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4192 holes[i]) == 0);
4193 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4194 0) == 0);
4195 }
4196 for (i = 0; i < nchildren; i++) {
4197 if (tops[i] != NULL)
4198 continue;
4199 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4200 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4201 VDEV_TYPE_MISSING) == 0);
4202 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4203 i) == 0);
4204 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4205 0) == 0);
4206 }
4207
4208 /*
4209 * Create pool config based on the best vdev config.
4210 */
4211 nvlist_dup(best_cfg, &config, KM_SLEEP);
4212
4213 /*
4214 * Put this pool's top-level vdevs into a root vdev.
4215 */
4216 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4217 &pgid) == 0);
4218 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4219 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4220 VDEV_TYPE_ROOT) == 0);
4221 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4222 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4223 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4224 tops, nchildren) == 0);
4225
4226 /*
4227 * Replace the existing vdev_tree with the new root vdev in
4228 * this pool's configuration (remove the old, add the new).
4229 */
4230 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4231
4232 /*
4233 * Drop vdev config elements that should not be present at pool level.
4234 */
4235 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4236 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4237
4238 for (i = 0; i < count; i++)
4239 nvlist_free(configs[i]);
4240 kmem_free(configs, count * sizeof(void *));
4241 for (i = 0; i < nchildren; i++)
4242 nvlist_free(tops[i]);
4243 kmem_free(tops, nchildren * sizeof(void *));
4244 nvlist_free(nvroot);
4245 return (config);
4246 }
4247
4248 int
spa_import_rootpool(const char * name)4249 spa_import_rootpool(const char *name)
4250 {
4251 spa_t *spa;
4252 vdev_t *rvd, *bvd, *avd = NULL;
4253 nvlist_t *config, *nvtop;
4254 uint64_t txg;
4255 char *pname;
4256 int error;
4257
4258 /*
4259 * Read the label from the boot device and generate a configuration.
4260 */
4261 config = spa_generate_rootconf(name);
4262
4263 mutex_enter(&spa_namespace_lock);
4264 if (config != NULL) {
4265 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4266 &pname) == 0 && strcmp(name, pname) == 0);
4267 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4268 == 0);
4269
4270 if ((spa = spa_lookup(pname)) != NULL) {
4271 /*
4272 * Remove the existing root pool from the namespace so
4273 * that we can replace it with the correct config
4274 * we just read in.
4275 */
4276 spa_remove(spa);
4277 }
4278 spa = spa_add(pname, config, NULL);
4279
4280 /*
4281 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4282 * via spa_version().
4283 */
4284 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4285 &spa->spa_ubsync.ub_version) != 0)
4286 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4287 } else if ((spa = spa_lookup(name)) == NULL) {
4288 mutex_exit(&spa_namespace_lock);
4289 nvlist_free(config);
4290 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4291 name);
4292 return (EIO);
4293 } else {
4294 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4295 }
4296 spa->spa_is_root = B_TRUE;
4297 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4298
4299 /*
4300 * Build up a vdev tree based on the boot device's label config.
4301 */
4302 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4303 &nvtop) == 0);
4304 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4305 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4306 VDEV_ALLOC_ROOTPOOL);
4307 spa_config_exit(spa, SCL_ALL, FTAG);
4308 if (error) {
4309 mutex_exit(&spa_namespace_lock);
4310 nvlist_free(config);
4311 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4312 pname);
4313 return (error);
4314 }
4315
4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4317 vdev_free(rvd);
4318 spa_config_exit(spa, SCL_ALL, FTAG);
4319 mutex_exit(&spa_namespace_lock);
4320
4321 nvlist_free(config);
4322 return (0);
4323 }
4324
4325 #endif /* illumos */
4326 #endif /* _KERNEL */
4327 #endif /* !__NetBSD__ */
4328
4329 /*
4330 * Import a non-root pool into the system.
4331 */
4332 int
spa_import(const char * pool,nvlist_t * config,nvlist_t * props,uint64_t flags)4333 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4334 {
4335 spa_t *spa;
4336 char *altroot = NULL;
4337 spa_load_state_t state = SPA_LOAD_IMPORT;
4338 zpool_rewind_policy_t policy;
4339 uint64_t mode = spa_mode_global;
4340 uint64_t readonly = B_FALSE;
4341 int error;
4342 nvlist_t *nvroot;
4343 nvlist_t **spares, **l2cache;
4344 uint_t nspares, nl2cache;
4345
4346 /*
4347 * If a pool with this name exists, return failure.
4348 */
4349 mutex_enter(&spa_namespace_lock);
4350 if (spa_lookup(pool) != NULL) {
4351 mutex_exit(&spa_namespace_lock);
4352 return (SET_ERROR(EEXIST));
4353 }
4354
4355 /*
4356 * Create and initialize the spa structure.
4357 */
4358 (void) nvlist_lookup_string(props,
4359 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4360 (void) nvlist_lookup_uint64(props,
4361 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4362 if (readonly)
4363 mode = FREAD;
4364 spa = spa_add(pool, config, altroot);
4365 spa->spa_import_flags = flags;
4366
4367 /*
4368 * Verbatim import - Take a pool and insert it into the namespace
4369 * as if it had been loaded at boot.
4370 */
4371 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4372 if (props != NULL)
4373 spa_configfile_set(spa, props, B_FALSE);
4374
4375 spa_config_sync(spa, B_FALSE, B_TRUE);
4376 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4377
4378 mutex_exit(&spa_namespace_lock);
4379 return (0);
4380 }
4381
4382 spa_activate(spa, mode);
4383
4384 /*
4385 * Don't start async tasks until we know everything is healthy.
4386 */
4387 spa_async_suspend(spa);
4388
4389 zpool_get_rewind_policy(config, &policy);
4390 if (policy.zrp_request & ZPOOL_DO_REWIND)
4391 state = SPA_LOAD_RECOVER;
4392
4393 /*
4394 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4395 * because the user-supplied config is actually the one to trust when
4396 * doing an import.
4397 */
4398 if (state != SPA_LOAD_RECOVER)
4399 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4400
4401 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4402 policy.zrp_request);
4403
4404 /*
4405 * Propagate anything learned while loading the pool and pass it
4406 * back to caller (i.e. rewind info, missing devices, etc).
4407 */
4408 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4409 spa->spa_load_info) == 0);
4410
4411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4412 /*
4413 * Toss any existing sparelist, as it doesn't have any validity
4414 * anymore, and conflicts with spa_has_spare().
4415 */
4416 if (spa->spa_spares.sav_config) {
4417 nvlist_free(spa->spa_spares.sav_config);
4418 spa->spa_spares.sav_config = NULL;
4419 spa_load_spares(spa);
4420 }
4421 if (spa->spa_l2cache.sav_config) {
4422 nvlist_free(spa->spa_l2cache.sav_config);
4423 spa->spa_l2cache.sav_config = NULL;
4424 spa_load_l2cache(spa);
4425 }
4426
4427 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4428 &nvroot) == 0);
4429 if (error == 0)
4430 error = spa_validate_aux(spa, nvroot, -1ULL,
4431 VDEV_ALLOC_SPARE);
4432 if (error == 0)
4433 error = spa_validate_aux(spa, nvroot, -1ULL,
4434 VDEV_ALLOC_L2CACHE);
4435 spa_config_exit(spa, SCL_ALL, FTAG);
4436
4437 if (props != NULL)
4438 spa_configfile_set(spa, props, B_FALSE);
4439
4440 if (error != 0 || (props && spa_writeable(spa) &&
4441 (error = spa_prop_set(spa, props)))) {
4442 spa_unload(spa);
4443 spa_deactivate(spa);
4444 spa_remove(spa);
4445 mutex_exit(&spa_namespace_lock);
4446 return (error);
4447 }
4448
4449 spa_async_resume(spa);
4450
4451 /*
4452 * Override any spares and level 2 cache devices as specified by
4453 * the user, as these may have correct device names/devids, etc.
4454 */
4455 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4456 &spares, &nspares) == 0) {
4457 if (spa->spa_spares.sav_config)
4458 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4459 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4460 else
4461 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4462 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4463 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4464 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4466 spa_load_spares(spa);
4467 spa_config_exit(spa, SCL_ALL, FTAG);
4468 spa->spa_spares.sav_sync = B_TRUE;
4469 }
4470 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4471 &l2cache, &nl2cache) == 0) {
4472 if (spa->spa_l2cache.sav_config)
4473 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4474 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4475 else
4476 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4477 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4478 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4479 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4480 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4481 spa_load_l2cache(spa);
4482 spa_config_exit(spa, SCL_ALL, FTAG);
4483 spa->spa_l2cache.sav_sync = B_TRUE;
4484 }
4485
4486 /*
4487 * Check for any removed devices.
4488 */
4489 if (spa->spa_autoreplace) {
4490 spa_aux_check_removed(&spa->spa_spares);
4491 spa_aux_check_removed(&spa->spa_l2cache);
4492 }
4493
4494 if (spa_writeable(spa)) {
4495 /*
4496 * Update the config cache to include the newly-imported pool.
4497 */
4498 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4499 }
4500
4501 /*
4502 * It's possible that the pool was expanded while it was exported.
4503 * We kick off an async task to handle this for us.
4504 */
4505 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4506
4507 spa_history_log_version(spa, "import");
4508
4509 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4510
4511 mutex_exit(&spa_namespace_lock);
4512
4513 #if defined(__FreeBSD__) || defined(__NetBSD__)
4514 #ifdef _KERNEL
4515 zvol_create_minors(pool);
4516 #endif
4517 #endif
4518 return (0);
4519 }
4520
4521 nvlist_t *
spa_tryimport(nvlist_t * tryconfig)4522 spa_tryimport(nvlist_t *tryconfig)
4523 {
4524 nvlist_t *config = NULL;
4525 char *poolname;
4526 spa_t *spa;
4527 uint64_t state;
4528 int error;
4529
4530 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4531 return (NULL);
4532
4533 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4534 return (NULL);
4535
4536 /*
4537 * Create and initialize the spa structure.
4538 */
4539 mutex_enter(&spa_namespace_lock);
4540 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4541 spa_activate(spa, FREAD);
4542
4543 /*
4544 * Pass off the heavy lifting to spa_load().
4545 * Pass TRUE for mosconfig because the user-supplied config
4546 * is actually the one to trust when doing an import.
4547 */
4548 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4549
4550 /*
4551 * If 'tryconfig' was at least parsable, return the current config.
4552 */
4553 if (spa->spa_root_vdev != NULL) {
4554 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4555 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4556 poolname) == 0);
4557 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4558 state) == 0);
4559 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4560 spa->spa_uberblock.ub_timestamp) == 0);
4561 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4562 spa->spa_load_info) == 0);
4563
4564 /*
4565 * If the bootfs property exists on this pool then we
4566 * copy it out so that external consumers can tell which
4567 * pools are bootable.
4568 */
4569 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4570 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4571
4572 /*
4573 * We have to play games with the name since the
4574 * pool was opened as TRYIMPORT_NAME.
4575 */
4576 if (dsl_dsobj_to_dsname(spa_name(spa),
4577 spa->spa_bootfs, tmpname) == 0) {
4578 char *cp;
4579 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4580
4581 cp = strchr(tmpname, '/');
4582 if (cp == NULL) {
4583 (void) strlcpy(dsname, tmpname,
4584 MAXPATHLEN);
4585 } else {
4586 (void) snprintf(dsname, MAXPATHLEN,
4587 "%s/%s", poolname, ++cp);
4588 }
4589 VERIFY(nvlist_add_string(config,
4590 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4591 kmem_free(dsname, MAXPATHLEN);
4592 }
4593 kmem_free(tmpname, MAXPATHLEN);
4594 }
4595
4596 /*
4597 * Add the list of hot spares and level 2 cache devices.
4598 */
4599 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4600 spa_add_spares(spa, config);
4601 spa_add_l2cache(spa, config);
4602 spa_config_exit(spa, SCL_CONFIG, FTAG);
4603 }
4604
4605 spa_unload(spa);
4606 spa_deactivate(spa);
4607 spa_remove(spa);
4608 mutex_exit(&spa_namespace_lock);
4609
4610 return (config);
4611 }
4612
4613 /*
4614 * Pool export/destroy
4615 *
4616 * The act of destroying or exporting a pool is very simple. We make sure there
4617 * is no more pending I/O and any references to the pool are gone. Then, we
4618 * update the pool state and sync all the labels to disk, removing the
4619 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4620 * we don't sync the labels or remove the configuration cache.
4621 */
4622 static int
spa_export_common(char * pool,int new_state,nvlist_t ** oldconfig,boolean_t force,boolean_t hardforce)4623 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4624 boolean_t force, boolean_t hardforce)
4625 {
4626 spa_t *spa;
4627
4628 if (oldconfig)
4629 *oldconfig = NULL;
4630
4631 if (!(spa_mode_global & FWRITE))
4632 return (SET_ERROR(EROFS));
4633
4634 mutex_enter(&spa_namespace_lock);
4635 if ((spa = spa_lookup(pool)) == NULL) {
4636 mutex_exit(&spa_namespace_lock);
4637 return (SET_ERROR(ENOENT));
4638 }
4639
4640 /*
4641 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4642 * reacquire the namespace lock, and see if we can export.
4643 */
4644 spa_open_ref(spa, FTAG);
4645 mutex_exit(&spa_namespace_lock);
4646 spa_async_suspend(spa);
4647 mutex_enter(&spa_namespace_lock);
4648 spa_close(spa, FTAG);
4649
4650 /*
4651 * The pool will be in core if it's openable,
4652 * in which case we can modify its state.
4653 */
4654 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4655 /*
4656 * Objsets may be open only because they're dirty, so we
4657 * have to force it to sync before checking spa_refcnt.
4658 */
4659 txg_wait_synced(spa->spa_dsl_pool, 0);
4660 spa_evicting_os_wait(spa);
4661
4662 /*
4663 * A pool cannot be exported or destroyed if there are active
4664 * references. If we are resetting a pool, allow references by
4665 * fault injection handlers.
4666 */
4667 if (!spa_refcount_zero(spa) ||
4668 (spa->spa_inject_ref != 0 &&
4669 new_state != POOL_STATE_UNINITIALIZED)) {
4670 spa_async_resume(spa);
4671 mutex_exit(&spa_namespace_lock);
4672 return (SET_ERROR(EBUSY));
4673 }
4674
4675 /*
4676 * A pool cannot be exported if it has an active shared spare.
4677 * This is to prevent other pools stealing the active spare
4678 * from an exported pool. At user's own will, such pool can
4679 * be forcedly exported.
4680 */
4681 if (!force && new_state == POOL_STATE_EXPORTED &&
4682 spa_has_active_shared_spare(spa)) {
4683 spa_async_resume(spa);
4684 mutex_exit(&spa_namespace_lock);
4685 return (SET_ERROR(EXDEV));
4686 }
4687
4688 /*
4689 * We want this to be reflected on every label,
4690 * so mark them all dirty. spa_unload() will do the
4691 * final sync that pushes these changes out.
4692 */
4693 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4694 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4695 spa->spa_state = new_state;
4696 spa->spa_final_txg = spa_last_synced_txg(spa) +
4697 TXG_DEFER_SIZE + 1;
4698 vdev_config_dirty(spa->spa_root_vdev);
4699 spa_config_exit(spa, SCL_ALL, FTAG);
4700 }
4701 }
4702
4703 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4704
4705 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4706 spa_unload(spa);
4707 spa_deactivate(spa);
4708 }
4709
4710 if (oldconfig && spa->spa_config)
4711 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4712
4713 if (new_state != POOL_STATE_UNINITIALIZED) {
4714 if (!hardforce)
4715 spa_config_sync(spa, B_TRUE, B_TRUE);
4716 spa_remove(spa);
4717 }
4718 mutex_exit(&spa_namespace_lock);
4719
4720 return (0);
4721 }
4722
4723 /*
4724 * Destroy a storage pool.
4725 */
4726 int
spa_destroy(char * pool)4727 spa_destroy(char *pool)
4728 {
4729 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4730 B_FALSE, B_FALSE));
4731 }
4732
4733 /*
4734 * Export a storage pool.
4735 */
4736 int
spa_export(char * pool,nvlist_t ** oldconfig,boolean_t force,boolean_t hardforce)4737 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4738 boolean_t hardforce)
4739 {
4740 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4741 force, hardforce));
4742 }
4743
4744 /*
4745 * Similar to spa_export(), this unloads the spa_t without actually removing it
4746 * from the namespace in any way.
4747 */
4748 int
spa_reset(char * pool)4749 spa_reset(char *pool)
4750 {
4751 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4752 B_FALSE, B_FALSE));
4753 }
4754
4755 /*
4756 * ==========================================================================
4757 * Device manipulation
4758 * ==========================================================================
4759 */
4760
4761 /*
4762 * Add a device to a storage pool.
4763 */
4764 int
spa_vdev_add(spa_t * spa,nvlist_t * nvroot)4765 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4766 {
4767 uint64_t txg, id;
4768 int error;
4769 vdev_t *rvd = spa->spa_root_vdev;
4770 vdev_t *vd, *tvd;
4771 nvlist_t **spares, **l2cache;
4772 uint_t nspares, nl2cache;
4773
4774 ASSERT(spa_writeable(spa));
4775
4776 txg = spa_vdev_enter(spa);
4777
4778 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4779 VDEV_ALLOC_ADD)) != 0)
4780 return (spa_vdev_exit(spa, NULL, txg, error));
4781
4782 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4783
4784 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4785 &nspares) != 0)
4786 nspares = 0;
4787
4788 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4789 &nl2cache) != 0)
4790 nl2cache = 0;
4791
4792 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4793 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4794
4795 if (vd->vdev_children != 0 &&
4796 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4797 return (spa_vdev_exit(spa, vd, txg, error));
4798
4799 /*
4800 * We must validate the spares and l2cache devices after checking the
4801 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4802 */
4803 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4804 return (spa_vdev_exit(spa, vd, txg, error));
4805
4806 /*
4807 * Transfer each new top-level vdev from vd to rvd.
4808 */
4809 for (int c = 0; c < vd->vdev_children; c++) {
4810
4811 /*
4812 * Set the vdev id to the first hole, if one exists.
4813 */
4814 for (id = 0; id < rvd->vdev_children; id++) {
4815 if (rvd->vdev_child[id]->vdev_ishole) {
4816 vdev_free(rvd->vdev_child[id]);
4817 break;
4818 }
4819 }
4820 tvd = vd->vdev_child[c];
4821 vdev_remove_child(vd, tvd);
4822 tvd->vdev_id = id;
4823 vdev_add_child(rvd, tvd);
4824 vdev_config_dirty(tvd);
4825 }
4826
4827 if (nspares != 0) {
4828 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4829 ZPOOL_CONFIG_SPARES);
4830 spa_load_spares(spa);
4831 spa->spa_spares.sav_sync = B_TRUE;
4832 }
4833
4834 if (nl2cache != 0) {
4835 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4836 ZPOOL_CONFIG_L2CACHE);
4837 spa_load_l2cache(spa);
4838 spa->spa_l2cache.sav_sync = B_TRUE;
4839 }
4840
4841 /*
4842 * We have to be careful when adding new vdevs to an existing pool.
4843 * If other threads start allocating from these vdevs before we
4844 * sync the config cache, and we lose power, then upon reboot we may
4845 * fail to open the pool because there are DVAs that the config cache
4846 * can't translate. Therefore, we first add the vdevs without
4847 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4848 * and then let spa_config_update() initialize the new metaslabs.
4849 *
4850 * spa_load() checks for added-but-not-initialized vdevs, so that
4851 * if we lose power at any point in this sequence, the remaining
4852 * steps will be completed the next time we load the pool.
4853 */
4854 (void) spa_vdev_exit(spa, vd, txg, 0);
4855
4856 mutex_enter(&spa_namespace_lock);
4857 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4858 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4859 mutex_exit(&spa_namespace_lock);
4860
4861 return (0);
4862 }
4863
4864 /*
4865 * Attach a device to a mirror. The arguments are the path to any device
4866 * in the mirror, and the nvroot for the new device. If the path specifies
4867 * a device that is not mirrored, we automatically insert the mirror vdev.
4868 *
4869 * If 'replacing' is specified, the new device is intended to replace the
4870 * existing device; in this case the two devices are made into their own
4871 * mirror using the 'replacing' vdev, which is functionally identical to
4872 * the mirror vdev (it actually reuses all the same ops) but has a few
4873 * extra rules: you can't attach to it after it's been created, and upon
4874 * completion of resilvering, the first disk (the one being replaced)
4875 * is automatically detached.
4876 */
4877 int
spa_vdev_attach(spa_t * spa,uint64_t guid,nvlist_t * nvroot,int replacing)4878 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4879 {
4880 uint64_t txg, dtl_max_txg;
4881 vdev_t *rvd = spa->spa_root_vdev;
4882 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4883 vdev_ops_t *pvops;
4884 char *oldvdpath, *newvdpath;
4885 int newvd_isspare;
4886 int error;
4887
4888 ASSERT(spa_writeable(spa));
4889
4890 txg = spa_vdev_enter(spa);
4891
4892 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4893
4894 if (oldvd == NULL)
4895 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4896
4897 if (!oldvd->vdev_ops->vdev_op_leaf)
4898 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4899
4900 pvd = oldvd->vdev_parent;
4901
4902 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4903 VDEV_ALLOC_ATTACH)) != 0)
4904 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4905
4906 if (newrootvd->vdev_children != 1)
4907 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4908
4909 newvd = newrootvd->vdev_child[0];
4910
4911 if (!newvd->vdev_ops->vdev_op_leaf)
4912 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4913
4914 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4915 return (spa_vdev_exit(spa, newrootvd, txg, error));
4916
4917 /*
4918 * Spares can't replace logs
4919 */
4920 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4921 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4922
4923 if (!replacing) {
4924 /*
4925 * For attach, the only allowable parent is a mirror or the root
4926 * vdev.
4927 */
4928 if (pvd->vdev_ops != &vdev_mirror_ops &&
4929 pvd->vdev_ops != &vdev_root_ops)
4930 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4931
4932 pvops = &vdev_mirror_ops;
4933 } else {
4934 /*
4935 * Active hot spares can only be replaced by inactive hot
4936 * spares.
4937 */
4938 if (pvd->vdev_ops == &vdev_spare_ops &&
4939 oldvd->vdev_isspare &&
4940 !spa_has_spare(spa, newvd->vdev_guid))
4941 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4942
4943 /*
4944 * If the source is a hot spare, and the parent isn't already a
4945 * spare, then we want to create a new hot spare. Otherwise, we
4946 * want to create a replacing vdev. The user is not allowed to
4947 * attach to a spared vdev child unless the 'isspare' state is
4948 * the same (spare replaces spare, non-spare replaces
4949 * non-spare).
4950 */
4951 if (pvd->vdev_ops == &vdev_replacing_ops &&
4952 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4953 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4954 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4955 newvd->vdev_isspare != oldvd->vdev_isspare) {
4956 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4957 }
4958
4959 if (newvd->vdev_isspare)
4960 pvops = &vdev_spare_ops;
4961 else
4962 pvops = &vdev_replacing_ops;
4963 }
4964
4965 /*
4966 * Make sure the new device is big enough.
4967 */
4968 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4969 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4970
4971 /*
4972 * The new device cannot have a higher alignment requirement
4973 * than the top-level vdev.
4974 */
4975 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4976 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4977
4978 /*
4979 * If this is an in-place replacement, update oldvd's path and devid
4980 * to make it distinguishable from newvd, and unopenable from now on.
4981 */
4982 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4983 spa_strfree(oldvd->vdev_path);
4984 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4985 KM_SLEEP);
4986 (void) sprintf(oldvd->vdev_path, "%s/%s",
4987 newvd->vdev_path, "old");
4988 if (oldvd->vdev_devid != NULL) {
4989 spa_strfree(oldvd->vdev_devid);
4990 oldvd->vdev_devid = NULL;
4991 }
4992 }
4993
4994 /* mark the device being resilvered */
4995 newvd->vdev_resilver_txg = txg;
4996
4997 /*
4998 * If the parent is not a mirror, or if we're replacing, insert the new
4999 * mirror/replacing/spare vdev above oldvd.
5000 */
5001 if (pvd->vdev_ops != pvops)
5002 pvd = vdev_add_parent(oldvd, pvops);
5003
5004 ASSERT(pvd->vdev_top->vdev_parent == rvd);
5005 ASSERT(pvd->vdev_ops == pvops);
5006 ASSERT(oldvd->vdev_parent == pvd);
5007
5008 /*
5009 * Extract the new device from its root and add it to pvd.
5010 */
5011 vdev_remove_child(newrootvd, newvd);
5012 newvd->vdev_id = pvd->vdev_children;
5013 newvd->vdev_crtxg = oldvd->vdev_crtxg;
5014 vdev_add_child(pvd, newvd);
5015
5016 tvd = newvd->vdev_top;
5017 ASSERT(pvd->vdev_top == tvd);
5018 ASSERT(tvd->vdev_parent == rvd);
5019
5020 vdev_config_dirty(tvd);
5021
5022 /*
5023 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5024 * for any dmu_sync-ed blocks. It will propagate upward when
5025 * spa_vdev_exit() calls vdev_dtl_reassess().
5026 */
5027 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
5028
5029 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
5030 dtl_max_txg - TXG_INITIAL);
5031
5032 if (newvd->vdev_isspare) {
5033 spa_spare_activate(newvd);
5034 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
5035 }
5036
5037 oldvdpath = spa_strdup(oldvd->vdev_path);
5038 newvdpath = spa_strdup(newvd->vdev_path);
5039 newvd_isspare = newvd->vdev_isspare;
5040
5041 /*
5042 * Mark newvd's DTL dirty in this txg.
5043 */
5044 vdev_dirty(tvd, VDD_DTL, newvd, txg);
5045
5046 /*
5047 * Schedule the resilver to restart in the future. We do this to
5048 * ensure that dmu_sync-ed blocks have been stitched into the
5049 * respective datasets.
5050 */
5051 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
5052
5053 if (spa->spa_bootfs)
5054 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
5055
5056 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
5057
5058 /*
5059 * Commit the config
5060 */
5061 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
5062
5063 spa_history_log_internal(spa, "vdev attach", NULL,
5064 "%s vdev=%s %s vdev=%s",
5065 replacing && newvd_isspare ? "spare in" :
5066 replacing ? "replace" : "attach", newvdpath,
5067 replacing ? "for" : "to", oldvdpath);
5068
5069 spa_strfree(oldvdpath);
5070 spa_strfree(newvdpath);
5071
5072 return (0);
5073 }
5074
5075 /*
5076 * Detach a device from a mirror or replacing vdev.
5077 *
5078 * If 'replace_done' is specified, only detach if the parent
5079 * is a replacing vdev.
5080 */
5081 int
spa_vdev_detach(spa_t * spa,uint64_t guid,uint64_t pguid,int replace_done)5082 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5083 {
5084 uint64_t txg;
5085 int error;
5086 vdev_t *rvd = spa->spa_root_vdev;
5087 vdev_t *vd, *pvd, *cvd, *tvd;
5088 boolean_t unspare = B_FALSE;
5089 uint64_t unspare_guid = 0;
5090 char *vdpath;
5091
5092 ASSERT(spa_writeable(spa));
5093
5094 txg = spa_vdev_enter(spa);
5095
5096 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5097
5098 if (vd == NULL)
5099 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5100
5101 if (!vd->vdev_ops->vdev_op_leaf)
5102 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5103
5104 pvd = vd->vdev_parent;
5105
5106 /*
5107 * If the parent/child relationship is not as expected, don't do it.
5108 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5109 * vdev that's replacing B with C. The user's intent in replacing
5110 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5111 * the replace by detaching C, the expected behavior is to end up
5112 * M(A,B). But suppose that right after deciding to detach C,
5113 * the replacement of B completes. We would have M(A,C), and then
5114 * ask to detach C, which would leave us with just A -- not what
5115 * the user wanted. To prevent this, we make sure that the
5116 * parent/child relationship hasn't changed -- in this example,
5117 * that C's parent is still the replacing vdev R.
5118 */
5119 if (pvd->vdev_guid != pguid && pguid != 0)
5120 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5121
5122 /*
5123 * Only 'replacing' or 'spare' vdevs can be replaced.
5124 */
5125 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5126 pvd->vdev_ops != &vdev_spare_ops)
5127 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5128
5129 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5130 spa_version(spa) >= SPA_VERSION_SPARES);
5131
5132 /*
5133 * Only mirror, replacing, and spare vdevs support detach.
5134 */
5135 if (pvd->vdev_ops != &vdev_replacing_ops &&
5136 pvd->vdev_ops != &vdev_mirror_ops &&
5137 pvd->vdev_ops != &vdev_spare_ops)
5138 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5139
5140 /*
5141 * If this device has the only valid copy of some data,
5142 * we cannot safely detach it.
5143 */
5144 if (vdev_dtl_required(vd))
5145 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5146
5147 ASSERT(pvd->vdev_children >= 2);
5148
5149 /*
5150 * If we are detaching the second disk from a replacing vdev, then
5151 * check to see if we changed the original vdev's path to have "/old"
5152 * at the end in spa_vdev_attach(). If so, undo that change now.
5153 */
5154 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5155 vd->vdev_path != NULL) {
5156 size_t len = strlen(vd->vdev_path);
5157
5158 for (int c = 0; c < pvd->vdev_children; c++) {
5159 cvd = pvd->vdev_child[c];
5160
5161 if (cvd == vd || cvd->vdev_path == NULL)
5162 continue;
5163
5164 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5165 strcmp(cvd->vdev_path + len, "/old") == 0) {
5166 spa_strfree(cvd->vdev_path);
5167 cvd->vdev_path = spa_strdup(vd->vdev_path);
5168 break;
5169 }
5170 }
5171 }
5172
5173 /*
5174 * If we are detaching the original disk from a spare, then it implies
5175 * that the spare should become a real disk, and be removed from the
5176 * active spare list for the pool.
5177 */
5178 if (pvd->vdev_ops == &vdev_spare_ops &&
5179 vd->vdev_id == 0 &&
5180 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5181 unspare = B_TRUE;
5182
5183 /*
5184 * Erase the disk labels so the disk can be used for other things.
5185 * This must be done after all other error cases are handled,
5186 * but before we disembowel vd (so we can still do I/O to it).
5187 * But if we can't do it, don't treat the error as fatal --
5188 * it may be that the unwritability of the disk is the reason
5189 * it's being detached!
5190 */
5191 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5192
5193 /*
5194 * Remove vd from its parent and compact the parent's children.
5195 */
5196 vdev_remove_child(pvd, vd);
5197 vdev_compact_children(pvd);
5198
5199 /*
5200 * Remember one of the remaining children so we can get tvd below.
5201 */
5202 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5203
5204 /*
5205 * If we need to remove the remaining child from the list of hot spares,
5206 * do it now, marking the vdev as no longer a spare in the process.
5207 * We must do this before vdev_remove_parent(), because that can
5208 * change the GUID if it creates a new toplevel GUID. For a similar
5209 * reason, we must remove the spare now, in the same txg as the detach;
5210 * otherwise someone could attach a new sibling, change the GUID, and
5211 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5212 */
5213 if (unspare) {
5214 ASSERT(cvd->vdev_isspare);
5215 spa_spare_remove(cvd);
5216 unspare_guid = cvd->vdev_guid;
5217 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5218 cvd->vdev_unspare = B_TRUE;
5219 }
5220
5221 /*
5222 * If the parent mirror/replacing vdev only has one child,
5223 * the parent is no longer needed. Remove it from the tree.
5224 */
5225 if (pvd->vdev_children == 1) {
5226 if (pvd->vdev_ops == &vdev_spare_ops)
5227 cvd->vdev_unspare = B_FALSE;
5228 vdev_remove_parent(cvd);
5229 }
5230
5231
5232 /*
5233 * We don't set tvd until now because the parent we just removed
5234 * may have been the previous top-level vdev.
5235 */
5236 tvd = cvd->vdev_top;
5237 ASSERT(tvd->vdev_parent == rvd);
5238
5239 /*
5240 * Reevaluate the parent vdev state.
5241 */
5242 vdev_propagate_state(cvd);
5243
5244 /*
5245 * If the 'autoexpand' property is set on the pool then automatically
5246 * try to expand the size of the pool. For example if the device we
5247 * just detached was smaller than the others, it may be possible to
5248 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5249 * first so that we can obtain the updated sizes of the leaf vdevs.
5250 */
5251 if (spa->spa_autoexpand) {
5252 vdev_reopen(tvd);
5253 vdev_expand(tvd, txg);
5254 }
5255
5256 vdev_config_dirty(tvd);
5257
5258 /*
5259 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5260 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5261 * But first make sure we're not on any *other* txg's DTL list, to
5262 * prevent vd from being accessed after it's freed.
5263 */
5264 vdpath = spa_strdup(vd->vdev_path);
5265 for (int t = 0; t < TXG_SIZE; t++)
5266 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5267 vd->vdev_detached = B_TRUE;
5268 vdev_dirty(tvd, VDD_DTL, vd, txg);
5269
5270 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5271
5272 /* hang on to the spa before we release the lock */
5273 spa_open_ref(spa, FTAG);
5274
5275 error = spa_vdev_exit(spa, vd, txg, 0);
5276
5277 spa_history_log_internal(spa, "detach", NULL,
5278 "vdev=%s", vdpath);
5279 spa_strfree(vdpath);
5280
5281 /*
5282 * If this was the removal of the original device in a hot spare vdev,
5283 * then we want to go through and remove the device from the hot spare
5284 * list of every other pool.
5285 */
5286 if (unspare) {
5287 spa_t *altspa = NULL;
5288
5289 mutex_enter(&spa_namespace_lock);
5290 while ((altspa = spa_next(altspa)) != NULL) {
5291 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5292 altspa == spa)
5293 continue;
5294
5295 spa_open_ref(altspa, FTAG);
5296 mutex_exit(&spa_namespace_lock);
5297 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5298 mutex_enter(&spa_namespace_lock);
5299 spa_close(altspa, FTAG);
5300 }
5301 mutex_exit(&spa_namespace_lock);
5302
5303 /* search the rest of the vdevs for spares to remove */
5304 spa_vdev_resilver_done(spa);
5305 }
5306
5307 /* all done with the spa; OK to release */
5308 mutex_enter(&spa_namespace_lock);
5309 spa_close(spa, FTAG);
5310 mutex_exit(&spa_namespace_lock);
5311
5312 return (error);
5313 }
5314
5315 /*
5316 * Split a set of devices from their mirrors, and create a new pool from them.
5317 */
5318 int
spa_vdev_split_mirror(spa_t * spa,char * newname,nvlist_t * config,nvlist_t * props,boolean_t exp)5319 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5320 nvlist_t *props, boolean_t exp)
5321 {
5322 int error = 0;
5323 uint64_t txg, *glist;
5324 spa_t *newspa;
5325 uint_t c, children, lastlog;
5326 nvlist_t **child, *nvl, *tmp;
5327 dmu_tx_t *tx;
5328 char *altroot = NULL;
5329 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5330 boolean_t activate_slog;
5331
5332 ASSERT(spa_writeable(spa));
5333
5334 txg = spa_vdev_enter(spa);
5335
5336 /* clear the log and flush everything up to now */
5337 activate_slog = spa_passivate_log(spa);
5338 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5339 error = spa_offline_log(spa);
5340 txg = spa_vdev_config_enter(spa);
5341
5342 if (activate_slog)
5343 spa_activate_log(spa);
5344
5345 if (error != 0)
5346 return (spa_vdev_exit(spa, NULL, txg, error));
5347
5348 /* check new spa name before going any further */
5349 if (spa_lookup(newname) != NULL)
5350 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5351
5352 /*
5353 * scan through all the children to ensure they're all mirrors
5354 */
5355 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5356 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5357 &children) != 0)
5358 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5359
5360 /* first, check to ensure we've got the right child count */
5361 rvd = spa->spa_root_vdev;
5362 lastlog = 0;
5363 for (c = 0; c < rvd->vdev_children; c++) {
5364 vdev_t *vd = rvd->vdev_child[c];
5365
5366 /* don't count the holes & logs as children */
5367 if (vd->vdev_islog || vd->vdev_ishole) {
5368 if (lastlog == 0)
5369 lastlog = c;
5370 continue;
5371 }
5372
5373 lastlog = 0;
5374 }
5375 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5376 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5377
5378 /* next, ensure no spare or cache devices are part of the split */
5379 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5380 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5381 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5382
5383 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5384 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5385
5386 /* then, loop over each vdev and validate it */
5387 for (c = 0; c < children; c++) {
5388 uint64_t is_hole = 0;
5389
5390 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5391 &is_hole);
5392
5393 if (is_hole != 0) {
5394 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5395 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5396 continue;
5397 } else {
5398 error = SET_ERROR(EINVAL);
5399 break;
5400 }
5401 }
5402
5403 /* which disk is going to be split? */
5404 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5405 &glist[c]) != 0) {
5406 error = SET_ERROR(EINVAL);
5407 break;
5408 }
5409
5410 /* look it up in the spa */
5411 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5412 if (vml[c] == NULL) {
5413 error = SET_ERROR(ENODEV);
5414 break;
5415 }
5416
5417 /* make sure there's nothing stopping the split */
5418 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5419 vml[c]->vdev_islog ||
5420 vml[c]->vdev_ishole ||
5421 vml[c]->vdev_isspare ||
5422 vml[c]->vdev_isl2cache ||
5423 !vdev_writeable(vml[c]) ||
5424 vml[c]->vdev_children != 0 ||
5425 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5426 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5427 error = SET_ERROR(EINVAL);
5428 break;
5429 }
5430
5431 if (vdev_dtl_required(vml[c])) {
5432 error = SET_ERROR(EBUSY);
5433 break;
5434 }
5435
5436 /* we need certain info from the top level */
5437 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5438 vml[c]->vdev_top->vdev_ms_array) == 0);
5439 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5440 vml[c]->vdev_top->vdev_ms_shift) == 0);
5441 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5442 vml[c]->vdev_top->vdev_asize) == 0);
5443 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5444 vml[c]->vdev_top->vdev_ashift) == 0);
5445
5446 /* transfer per-vdev ZAPs */
5447 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5448 VERIFY0(nvlist_add_uint64(child[c],
5449 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5450
5451 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5452 VERIFY0(nvlist_add_uint64(child[c],
5453 ZPOOL_CONFIG_VDEV_TOP_ZAP,
5454 vml[c]->vdev_parent->vdev_top_zap));
5455 }
5456
5457 if (error != 0) {
5458 kmem_free(vml, children * sizeof (vdev_t *));
5459 kmem_free(glist, children * sizeof (uint64_t));
5460 return (spa_vdev_exit(spa, NULL, txg, error));
5461 }
5462
5463 /* stop writers from using the disks */
5464 for (c = 0; c < children; c++) {
5465 if (vml[c] != NULL)
5466 vml[c]->vdev_offline = B_TRUE;
5467 }
5468 vdev_reopen(spa->spa_root_vdev);
5469
5470 /*
5471 * Temporarily record the splitting vdevs in the spa config. This
5472 * will disappear once the config is regenerated.
5473 */
5474 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5475 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5476 glist, children) == 0);
5477 kmem_free(glist, children * sizeof (uint64_t));
5478
5479 mutex_enter(&spa->spa_props_lock);
5480 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5481 nvl) == 0);
5482 mutex_exit(&spa->spa_props_lock);
5483 spa->spa_config_splitting = nvl;
5484 vdev_config_dirty(spa->spa_root_vdev);
5485
5486 /* configure and create the new pool */
5487 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5488 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5489 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5490 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5491 spa_version(spa)) == 0);
5492 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5493 spa->spa_config_txg) == 0);
5494 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5495 spa_generate_guid(NULL)) == 0);
5496 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5497 (void) nvlist_lookup_string(props,
5498 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5499
5500 /* add the new pool to the namespace */
5501 newspa = spa_add(newname, config, altroot);
5502 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5503 newspa->spa_config_txg = spa->spa_config_txg;
5504 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5505
5506 /* release the spa config lock, retaining the namespace lock */
5507 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5508
5509 if (zio_injection_enabled)
5510 zio_handle_panic_injection(spa, FTAG, 1);
5511
5512 spa_activate(newspa, spa_mode_global);
5513 spa_async_suspend(newspa);
5514
5515 #ifndef illumos
5516 /* mark that we are creating new spa by splitting */
5517 newspa->spa_splitting_newspa = B_TRUE;
5518 #endif
5519 /* create the new pool from the disks of the original pool */
5520 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5521 #ifndef illumos
5522 newspa->spa_splitting_newspa = B_FALSE;
5523 #endif
5524 if (error)
5525 goto out;
5526
5527 /* if that worked, generate a real config for the new pool */
5528 if (newspa->spa_root_vdev != NULL) {
5529 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5530 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5531 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5532 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5533 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5534 B_TRUE));
5535 }
5536
5537 /* set the props */
5538 if (props != NULL) {
5539 spa_configfile_set(newspa, props, B_FALSE);
5540 error = spa_prop_set(newspa, props);
5541 if (error)
5542 goto out;
5543 }
5544
5545 /* flush everything */
5546 txg = spa_vdev_config_enter(newspa);
5547 vdev_config_dirty(newspa->spa_root_vdev);
5548 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5549
5550 if (zio_injection_enabled)
5551 zio_handle_panic_injection(spa, FTAG, 2);
5552
5553 spa_async_resume(newspa);
5554
5555 /* finally, update the original pool's config */
5556 txg = spa_vdev_config_enter(spa);
5557 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5558 error = dmu_tx_assign(tx, TXG_WAIT);
5559 if (error != 0)
5560 dmu_tx_abort(tx);
5561 for (c = 0; c < children; c++) {
5562 if (vml[c] != NULL) {
5563 vdev_split(vml[c]);
5564 if (error == 0)
5565 spa_history_log_internal(spa, "detach", tx,
5566 "vdev=%s", vml[c]->vdev_path);
5567
5568 vdev_free(vml[c]);
5569 }
5570 }
5571 spa->spa_avz_action = AVZ_ACTION_REBUILD;
5572 vdev_config_dirty(spa->spa_root_vdev);
5573 spa->spa_config_splitting = NULL;
5574 nvlist_free(nvl);
5575 if (error == 0)
5576 dmu_tx_commit(tx);
5577 (void) spa_vdev_exit(spa, NULL, txg, 0);
5578
5579 if (zio_injection_enabled)
5580 zio_handle_panic_injection(spa, FTAG, 3);
5581
5582 /* split is complete; log a history record */
5583 spa_history_log_internal(newspa, "split", NULL,
5584 "from pool %s", spa_name(spa));
5585
5586 kmem_free(vml, children * sizeof (vdev_t *));
5587
5588 /* if we're not going to mount the filesystems in userland, export */
5589 if (exp)
5590 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5591 B_FALSE, B_FALSE);
5592
5593 return (error);
5594
5595 out:
5596 spa_unload(newspa);
5597 spa_deactivate(newspa);
5598 spa_remove(newspa);
5599
5600 txg = spa_vdev_config_enter(spa);
5601
5602 /* re-online all offlined disks */
5603 for (c = 0; c < children; c++) {
5604 if (vml[c] != NULL)
5605 vml[c]->vdev_offline = B_FALSE;
5606 }
5607 vdev_reopen(spa->spa_root_vdev);
5608
5609 nvlist_free(spa->spa_config_splitting);
5610 spa->spa_config_splitting = NULL;
5611 (void) spa_vdev_exit(spa, NULL, txg, error);
5612
5613 kmem_free(vml, children * sizeof (vdev_t *));
5614 return (error);
5615 }
5616
5617 static nvlist_t *
spa_nvlist_lookup_by_guid(nvlist_t ** nvpp,int count,uint64_t target_guid)5618 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5619 {
5620 for (int i = 0; i < count; i++) {
5621 uint64_t guid;
5622
5623 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5624 &guid) == 0);
5625
5626 if (guid == target_guid)
5627 return (nvpp[i]);
5628 }
5629
5630 return (NULL);
5631 }
5632
5633 static void
spa_vdev_remove_aux(nvlist_t * config,char * name,nvlist_t ** dev,int count,nvlist_t * dev_to_remove)5634 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5635 nvlist_t *dev_to_remove)
5636 {
5637 nvlist_t **newdev = NULL;
5638
5639 if (count > 1)
5640 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5641
5642 for (int i = 0, j = 0; i < count; i++) {
5643 if (dev[i] == dev_to_remove)
5644 continue;
5645 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5646 }
5647
5648 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5649 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5650
5651 for (int i = 0; i < count - 1; i++)
5652 nvlist_free(newdev[i]);
5653
5654 if (count > 1)
5655 kmem_free(newdev, (count - 1) * sizeof (void *));
5656 }
5657
5658 /*
5659 * Evacuate the device.
5660 */
5661 static int
spa_vdev_remove_evacuate(spa_t * spa,vdev_t * vd)5662 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5663 {
5664 uint64_t txg;
5665 int error = 0;
5666
5667 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5668 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5669 ASSERT(vd == vd->vdev_top);
5670
5671 /*
5672 * Evacuate the device. We don't hold the config lock as writer
5673 * since we need to do I/O but we do keep the
5674 * spa_namespace_lock held. Once this completes the device
5675 * should no longer have any blocks allocated on it.
5676 */
5677 if (vd->vdev_islog) {
5678 if (vd->vdev_stat.vs_alloc != 0)
5679 error = spa_offline_log(spa);
5680 } else {
5681 error = SET_ERROR(ENOTSUP);
5682 }
5683
5684 if (error)
5685 return (error);
5686
5687 /*
5688 * The evacuation succeeded. Remove any remaining MOS metadata
5689 * associated with this vdev, and wait for these changes to sync.
5690 */
5691 ASSERT0(vd->vdev_stat.vs_alloc);
5692 txg = spa_vdev_config_enter(spa);
5693 vd->vdev_removing = B_TRUE;
5694 vdev_dirty_leaves(vd, VDD_DTL, txg);
5695 vdev_config_dirty(vd);
5696 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5697
5698 return (0);
5699 }
5700
5701 /*
5702 * Complete the removal by cleaning up the namespace.
5703 */
5704 static void
spa_vdev_remove_from_namespace(spa_t * spa,vdev_t * vd)5705 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5706 {
5707 vdev_t *rvd = spa->spa_root_vdev;
5708 uint64_t id = vd->vdev_id;
5709 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5710
5711 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5712 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5713 ASSERT(vd == vd->vdev_top);
5714
5715 /*
5716 * Only remove any devices which are empty.
5717 */
5718 if (vd->vdev_stat.vs_alloc != 0)
5719 return;
5720
5721 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5722
5723 if (list_link_active(&vd->vdev_state_dirty_node))
5724 vdev_state_clean(vd);
5725 if (list_link_active(&vd->vdev_config_dirty_node))
5726 vdev_config_clean(vd);
5727
5728 vdev_free(vd);
5729
5730 if (last_vdev) {
5731 vdev_compact_children(rvd);
5732 } else {
5733 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5734 vdev_add_child(rvd, vd);
5735 }
5736 vdev_config_dirty(rvd);
5737
5738 /*
5739 * Reassess the health of our root vdev.
5740 */
5741 vdev_reopen(rvd);
5742 }
5743
5744 /*
5745 * Remove a device from the pool -
5746 *
5747 * Removing a device from the vdev namespace requires several steps
5748 * and can take a significant amount of time. As a result we use
5749 * the spa_vdev_config_[enter/exit] functions which allow us to
5750 * grab and release the spa_config_lock while still holding the namespace
5751 * lock. During each step the configuration is synced out.
5752 *
5753 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5754 * devices.
5755 */
5756 int
spa_vdev_remove(spa_t * spa,uint64_t guid,boolean_t unspare)5757 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5758 {
5759 vdev_t *vd;
5760 sysevent_t *ev = NULL;
5761 metaslab_group_t *mg;
5762 nvlist_t **spares, **l2cache, *nv;
5763 uint64_t txg = 0;
5764 uint_t nspares, nl2cache;
5765 int error = 0;
5766 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5767
5768 ASSERT(spa_writeable(spa));
5769
5770 if (!locked)
5771 txg = spa_vdev_enter(spa);
5772
5773 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5774
5775 if (spa->spa_spares.sav_vdevs != NULL &&
5776 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5777 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5778 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5779 /*
5780 * Only remove the hot spare if it's not currently in use
5781 * in this pool.
5782 */
5783 if (vd == NULL || unspare) {
5784 if (vd == NULL)
5785 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5786 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5787 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5788 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5789 spa_load_spares(spa);
5790 spa->spa_spares.sav_sync = B_TRUE;
5791 } else {
5792 error = SET_ERROR(EBUSY);
5793 }
5794 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5795 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5796 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5797 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5798 /*
5799 * Cache devices can always be removed.
5800 */
5801 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5802 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5803 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5804 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5805 spa_load_l2cache(spa);
5806 spa->spa_l2cache.sav_sync = B_TRUE;
5807 } else if (vd != NULL && vd->vdev_islog) {
5808 ASSERT(!locked);
5809 ASSERT(vd == vd->vdev_top);
5810
5811 mg = vd->vdev_mg;
5812
5813 /*
5814 * Stop allocating from this vdev.
5815 */
5816 metaslab_group_passivate(mg);
5817
5818 /*
5819 * Wait for the youngest allocations and frees to sync,
5820 * and then wait for the deferral of those frees to finish.
5821 */
5822 spa_vdev_config_exit(spa, NULL,
5823 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5824
5825 /*
5826 * Attempt to evacuate the vdev.
5827 */
5828 error = spa_vdev_remove_evacuate(spa, vd);
5829
5830 txg = spa_vdev_config_enter(spa);
5831
5832 /*
5833 * If we couldn't evacuate the vdev, unwind.
5834 */
5835 if (error) {
5836 metaslab_group_activate(mg);
5837 return (spa_vdev_exit(spa, NULL, txg, error));
5838 }
5839
5840 /*
5841 * Clean up the vdev namespace.
5842 */
5843 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5844 spa_vdev_remove_from_namespace(spa, vd);
5845
5846 } else if (vd != NULL) {
5847 /*
5848 * Normal vdevs cannot be removed (yet).
5849 */
5850 error = SET_ERROR(ENOTSUP);
5851 } else {
5852 /*
5853 * There is no vdev of any kind with the specified guid.
5854 */
5855 error = SET_ERROR(ENOENT);
5856 }
5857
5858 if (!locked)
5859 error = spa_vdev_exit(spa, NULL, txg, error);
5860
5861 if (ev)
5862 spa_event_post(ev);
5863
5864 return (error);
5865 }
5866
5867 /*
5868 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5869 * currently spared, so we can detach it.
5870 */
5871 static vdev_t *
spa_vdev_resilver_done_hunt(vdev_t * vd)5872 spa_vdev_resilver_done_hunt(vdev_t *vd)
5873 {
5874 vdev_t *newvd, *oldvd;
5875
5876 for (int c = 0; c < vd->vdev_children; c++) {
5877 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5878 if (oldvd != NULL)
5879 return (oldvd);
5880 }
5881
5882 /*
5883 * Check for a completed replacement. We always consider the first
5884 * vdev in the list to be the oldest vdev, and the last one to be
5885 * the newest (see spa_vdev_attach() for how that works). In
5886 * the case where the newest vdev is faulted, we will not automatically
5887 * remove it after a resilver completes. This is OK as it will require
5888 * user intervention to determine which disk the admin wishes to keep.
5889 */
5890 if (vd->vdev_ops == &vdev_replacing_ops) {
5891 ASSERT(vd->vdev_children > 1);
5892
5893 newvd = vd->vdev_child[vd->vdev_children - 1];
5894 oldvd = vd->vdev_child[0];
5895
5896 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5897 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5898 !vdev_dtl_required(oldvd))
5899 return (oldvd);
5900 }
5901
5902 /*
5903 * Check for a completed resilver with the 'unspare' flag set.
5904 */
5905 if (vd->vdev_ops == &vdev_spare_ops) {
5906 vdev_t *first = vd->vdev_child[0];
5907 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5908
5909 if (last->vdev_unspare) {
5910 oldvd = first;
5911 newvd = last;
5912 } else if (first->vdev_unspare) {
5913 oldvd = last;
5914 newvd = first;
5915 } else {
5916 oldvd = NULL;
5917 }
5918
5919 if (oldvd != NULL &&
5920 vdev_dtl_empty(newvd, DTL_MISSING) &&
5921 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5922 !vdev_dtl_required(oldvd))
5923 return (oldvd);
5924
5925 /*
5926 * If there are more than two spares attached to a disk,
5927 * and those spares are not required, then we want to
5928 * attempt to free them up now so that they can be used
5929 * by other pools. Once we're back down to a single
5930 * disk+spare, we stop removing them.
5931 */
5932 if (vd->vdev_children > 2) {
5933 newvd = vd->vdev_child[1];
5934
5935 if (newvd->vdev_isspare && last->vdev_isspare &&
5936 vdev_dtl_empty(last, DTL_MISSING) &&
5937 vdev_dtl_empty(last, DTL_OUTAGE) &&
5938 !vdev_dtl_required(newvd))
5939 return (newvd);
5940 }
5941 }
5942
5943 return (NULL);
5944 }
5945
5946 static void
spa_vdev_resilver_done(spa_t * spa)5947 spa_vdev_resilver_done(spa_t *spa)
5948 {
5949 vdev_t *vd, *pvd, *ppvd;
5950 uint64_t guid, sguid, pguid, ppguid;
5951
5952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5953
5954 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5955 pvd = vd->vdev_parent;
5956 ppvd = pvd->vdev_parent;
5957 guid = vd->vdev_guid;
5958 pguid = pvd->vdev_guid;
5959 ppguid = ppvd->vdev_guid;
5960 sguid = 0;
5961 /*
5962 * If we have just finished replacing a hot spared device, then
5963 * we need to detach the parent's first child (the original hot
5964 * spare) as well.
5965 */
5966 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5967 ppvd->vdev_children == 2) {
5968 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5969 sguid = ppvd->vdev_child[1]->vdev_guid;
5970 }
5971 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5972
5973 spa_config_exit(spa, SCL_ALL, FTAG);
5974 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5975 return;
5976 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5977 return;
5978 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5979 }
5980
5981 spa_config_exit(spa, SCL_ALL, FTAG);
5982 }
5983
5984 /*
5985 * Update the stored path or FRU for this vdev.
5986 */
5987 int
spa_vdev_set_common(spa_t * spa,uint64_t guid,const char * value,boolean_t ispath)5988 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5989 boolean_t ispath)
5990 {
5991 vdev_t *vd;
5992 boolean_t sync = B_FALSE;
5993
5994 ASSERT(spa_writeable(spa));
5995
5996 spa_vdev_state_enter(spa, SCL_ALL);
5997
5998 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5999 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6000
6001 if (!vd->vdev_ops->vdev_op_leaf)
6002 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6003
6004 if (ispath) {
6005 if (strcmp(value, vd->vdev_path) != 0) {
6006 spa_strfree(vd->vdev_path);
6007 vd->vdev_path = spa_strdup(value);
6008 sync = B_TRUE;
6009 }
6010 } else {
6011 if (vd->vdev_fru == NULL) {
6012 vd->vdev_fru = spa_strdup(value);
6013 sync = B_TRUE;
6014 } else if (strcmp(value, vd->vdev_fru) != 0) {
6015 spa_strfree(vd->vdev_fru);
6016 vd->vdev_fru = spa_strdup(value);
6017 sync = B_TRUE;
6018 }
6019 }
6020
6021 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6022 }
6023
6024 int
spa_vdev_setpath(spa_t * spa,uint64_t guid,const char * newpath)6025 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6026 {
6027 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6028 }
6029
6030 int
spa_vdev_setfru(spa_t * spa,uint64_t guid,const char * newfru)6031 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6032 {
6033 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6034 }
6035
6036 /*
6037 * ==========================================================================
6038 * SPA Scanning
6039 * ==========================================================================
6040 */
6041
6042 int
spa_scan_stop(spa_t * spa)6043 spa_scan_stop(spa_t *spa)
6044 {
6045 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6046 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6047 return (SET_ERROR(EBUSY));
6048 return (dsl_scan_cancel(spa->spa_dsl_pool));
6049 }
6050
6051 int
spa_scan(spa_t * spa,pool_scan_func_t func)6052 spa_scan(spa_t *spa, pool_scan_func_t func)
6053 {
6054 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6055
6056 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6057 return (SET_ERROR(ENOTSUP));
6058
6059 /*
6060 * If a resilver was requested, but there is no DTL on a
6061 * writeable leaf device, we have nothing to do.
6062 */
6063 if (func == POOL_SCAN_RESILVER &&
6064 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6065 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6066 return (0);
6067 }
6068
6069 return (dsl_scan(spa->spa_dsl_pool, func));
6070 }
6071
6072 /*
6073 * ==========================================================================
6074 * SPA async task processing
6075 * ==========================================================================
6076 */
6077
6078 static void
spa_async_remove(spa_t * spa,vdev_t * vd)6079 spa_async_remove(spa_t *spa, vdev_t *vd)
6080 {
6081 if (vd->vdev_remove_wanted) {
6082 vd->vdev_remove_wanted = B_FALSE;
6083 vd->vdev_delayed_close = B_FALSE;
6084 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6085
6086 /*
6087 * We want to clear the stats, but we don't want to do a full
6088 * vdev_clear() as that will cause us to throw away
6089 * degraded/faulted state as well as attempt to reopen the
6090 * device, all of which is a waste.
6091 */
6092 vd->vdev_stat.vs_read_errors = 0;
6093 vd->vdev_stat.vs_write_errors = 0;
6094 vd->vdev_stat.vs_checksum_errors = 0;
6095
6096 vdev_state_dirty(vd->vdev_top);
6097 /* Tell userspace that the vdev is gone. */
6098 zfs_post_remove(spa, vd);
6099 }
6100
6101 for (int c = 0; c < vd->vdev_children; c++)
6102 spa_async_remove(spa, vd->vdev_child[c]);
6103 }
6104
6105 static void
spa_async_probe(spa_t * spa,vdev_t * vd)6106 spa_async_probe(spa_t *spa, vdev_t *vd)
6107 {
6108 if (vd->vdev_probe_wanted) {
6109 vd->vdev_probe_wanted = B_FALSE;
6110 vdev_reopen(vd); /* vdev_open() does the actual probe */
6111 }
6112
6113 for (int c = 0; c < vd->vdev_children; c++)
6114 spa_async_probe(spa, vd->vdev_child[c]);
6115 }
6116
6117 static void
spa_async_autoexpand(spa_t * spa,vdev_t * vd)6118 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6119 {
6120 sysevent_id_t eid;
6121 nvlist_t *attr;
6122 char *physpath;
6123
6124 if (!spa->spa_autoexpand)
6125 return;
6126
6127 for (int c = 0; c < vd->vdev_children; c++) {
6128 vdev_t *cvd = vd->vdev_child[c];
6129 spa_async_autoexpand(spa, cvd);
6130 }
6131
6132 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6133 return;
6134
6135 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6136 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6137
6138 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6139 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6140
6141 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6142 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6143
6144 nvlist_free(attr);
6145 kmem_free(physpath, MAXPATHLEN);
6146 }
6147
6148 static void
spa_async_thread(void * arg)6149 spa_async_thread(void *arg)
6150 {
6151 spa_t *spa = arg;
6152 int tasks;
6153
6154 ASSERT(spa->spa_sync_on);
6155
6156 mutex_enter(&spa->spa_async_lock);
6157 tasks = spa->spa_async_tasks;
6158 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6159 mutex_exit(&spa->spa_async_lock);
6160
6161 /*
6162 * See if the config needs to be updated.
6163 */
6164 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6165 uint64_t old_space, new_space;
6166
6167 mutex_enter(&spa_namespace_lock);
6168 old_space = metaslab_class_get_space(spa_normal_class(spa));
6169 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6170 new_space = metaslab_class_get_space(spa_normal_class(spa));
6171 mutex_exit(&spa_namespace_lock);
6172
6173 /*
6174 * If the pool grew as a result of the config update,
6175 * then log an internal history event.
6176 */
6177 if (new_space != old_space) {
6178 spa_history_log_internal(spa, "vdev online", NULL,
6179 "pool '%s' size: %llu(+%llu)",
6180 spa_name(spa), new_space, new_space - old_space);
6181 }
6182 }
6183
6184 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6185 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6186 spa_async_autoexpand(spa, spa->spa_root_vdev);
6187 spa_config_exit(spa, SCL_CONFIG, FTAG);
6188 }
6189
6190 /*
6191 * See if any devices need to be probed.
6192 */
6193 if (tasks & SPA_ASYNC_PROBE) {
6194 spa_vdev_state_enter(spa, SCL_NONE);
6195 spa_async_probe(spa, spa->spa_root_vdev);
6196 (void) spa_vdev_state_exit(spa, NULL, 0);
6197 }
6198
6199 /*
6200 * If any devices are done replacing, detach them.
6201 */
6202 if (tasks & SPA_ASYNC_RESILVER_DONE)
6203 spa_vdev_resilver_done(spa);
6204
6205 /*
6206 * Kick off a resilver.
6207 */
6208 if (tasks & SPA_ASYNC_RESILVER)
6209 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6210
6211 /*
6212 * Let the world know that we're done.
6213 */
6214 mutex_enter(&spa->spa_async_lock);
6215 spa->spa_async_thread = NULL;
6216 cv_broadcast(&spa->spa_async_cv);
6217 mutex_exit(&spa->spa_async_lock);
6218 thread_exit();
6219 }
6220
6221 static void
spa_async_thread_vd(void * arg)6222 spa_async_thread_vd(void *arg)
6223 {
6224 spa_t *spa = arg;
6225 int tasks;
6226
6227 ASSERT(spa->spa_sync_on);
6228
6229 mutex_enter(&spa->spa_async_lock);
6230 tasks = spa->spa_async_tasks;
6231 retry:
6232 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6233 mutex_exit(&spa->spa_async_lock);
6234
6235 /*
6236 * See if any devices need to be marked REMOVED.
6237 */
6238 if (tasks & SPA_ASYNC_REMOVE) {
6239 spa_vdev_state_enter(spa, SCL_NONE);
6240 spa_async_remove(spa, spa->spa_root_vdev);
6241 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6242 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6243 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6244 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6245 (void) spa_vdev_state_exit(spa, NULL, 0);
6246 }
6247
6248 /*
6249 * Let the world know that we're done.
6250 */
6251 mutex_enter(&spa->spa_async_lock);
6252 tasks = spa->spa_async_tasks;
6253 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6254 goto retry;
6255 spa->spa_async_thread_vd = NULL;
6256 cv_broadcast(&spa->spa_async_cv);
6257 mutex_exit(&spa->spa_async_lock);
6258 thread_exit();
6259 }
6260
6261 void
spa_async_suspend(spa_t * spa)6262 spa_async_suspend(spa_t *spa)
6263 {
6264 mutex_enter(&spa->spa_async_lock);
6265 spa->spa_async_suspended++;
6266 while (spa->spa_async_thread != NULL &&
6267 spa->spa_async_thread_vd != NULL)
6268 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6269 mutex_exit(&spa->spa_async_lock);
6270 }
6271
6272 void
spa_async_resume(spa_t * spa)6273 spa_async_resume(spa_t *spa)
6274 {
6275 mutex_enter(&spa->spa_async_lock);
6276 ASSERT(spa->spa_async_suspended != 0);
6277 spa->spa_async_suspended--;
6278 mutex_exit(&spa->spa_async_lock);
6279 }
6280
6281 static boolean_t
spa_async_tasks_pending(spa_t * spa)6282 spa_async_tasks_pending(spa_t *spa)
6283 {
6284 uint_t non_config_tasks;
6285 uint_t config_task;
6286 boolean_t config_task_suspended;
6287
6288 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6289 SPA_ASYNC_REMOVE);
6290 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6291 if (spa->spa_ccw_fail_time == 0) {
6292 config_task_suspended = B_FALSE;
6293 } else {
6294 config_task_suspended =
6295 (gethrtime() - spa->spa_ccw_fail_time) <
6296 ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
6297 }
6298
6299 return (non_config_tasks || (config_task && !config_task_suspended));
6300 }
6301
6302 static void
spa_async_dispatch(spa_t * spa)6303 spa_async_dispatch(spa_t *spa)
6304 {
6305 mutex_enter(&spa->spa_async_lock);
6306 if (spa_async_tasks_pending(spa) &&
6307 !spa->spa_async_suspended &&
6308 spa->spa_async_thread == NULL &&
6309 rootdir != NULL)
6310 spa->spa_async_thread = thread_create(NULL, 0,
6311 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6312 mutex_exit(&spa->spa_async_lock);
6313 }
6314
6315 static void
spa_async_dispatch_vd(spa_t * spa)6316 spa_async_dispatch_vd(spa_t *spa)
6317 {
6318 mutex_enter(&spa->spa_async_lock);
6319 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6320 !spa->spa_async_suspended &&
6321 spa->spa_async_thread_vd == NULL &&
6322 rootdir != NULL)
6323 spa->spa_async_thread_vd = thread_create(NULL, 0,
6324 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6325 mutex_exit(&spa->spa_async_lock);
6326 }
6327
6328 void
spa_async_request(spa_t * spa,int task)6329 spa_async_request(spa_t *spa, int task)
6330 {
6331 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6332 mutex_enter(&spa->spa_async_lock);
6333 spa->spa_async_tasks |= task;
6334 mutex_exit(&spa->spa_async_lock);
6335 spa_async_dispatch_vd(spa);
6336 }
6337
6338 /*
6339 * ==========================================================================
6340 * SPA syncing routines
6341 * ==========================================================================
6342 */
6343
6344 static int
bpobj_enqueue_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6345 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6346 {
6347 bpobj_t *bpo = arg;
6348 bpobj_enqueue(bpo, bp, tx);
6349 return (0);
6350 }
6351
6352 static int
spa_free_sync_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6353 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6354 {
6355 zio_t *zio = arg;
6356
6357 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6358 BP_GET_PSIZE(bp), zio->io_flags));
6359 return (0);
6360 }
6361
6362 /*
6363 * Note: this simple function is not inlined to make it easier to dtrace the
6364 * amount of time spent syncing frees.
6365 */
6366 static void
spa_sync_frees(spa_t * spa,bplist_t * bpl,dmu_tx_t * tx)6367 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6368 {
6369 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6370 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6371 VERIFY(zio_wait(zio) == 0);
6372 }
6373
6374 /*
6375 * Note: this simple function is not inlined to make it easier to dtrace the
6376 * amount of time spent syncing deferred frees.
6377 */
6378 static void
spa_sync_deferred_frees(spa_t * spa,dmu_tx_t * tx)6379 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6380 {
6381 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6382 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6383 spa_free_sync_cb, zio, tx), ==, 0);
6384 VERIFY0(zio_wait(zio));
6385 }
6386
6387
6388 static void
spa_sync_nvlist(spa_t * spa,uint64_t obj,nvlist_t * nv,dmu_tx_t * tx)6389 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6390 {
6391 char *packed = NULL;
6392 size_t bufsize;
6393 size_t nvsize = 0;
6394 dmu_buf_t *db;
6395
6396 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6397
6398 /*
6399 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6400 * information. This avoids the dmu_buf_will_dirty() path and
6401 * saves us a pre-read to get data we don't actually care about.
6402 */
6403 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6404 packed = kmem_alloc(bufsize, KM_SLEEP);
6405
6406 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6407 KM_SLEEP) == 0);
6408 bzero(packed + nvsize, bufsize - nvsize);
6409
6410 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6411
6412 kmem_free(packed, bufsize);
6413
6414 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6415 dmu_buf_will_dirty(db, tx);
6416 *(uint64_t *)db->db_data = nvsize;
6417 dmu_buf_rele(db, FTAG);
6418 }
6419
6420 static void
spa_sync_aux_dev(spa_t * spa,spa_aux_vdev_t * sav,dmu_tx_t * tx,const char * config,const char * entry)6421 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6422 const char *config, const char *entry)
6423 {
6424 nvlist_t *nvroot;
6425 nvlist_t **list;
6426 int i;
6427
6428 if (!sav->sav_sync)
6429 return;
6430
6431 /*
6432 * Update the MOS nvlist describing the list of available devices.
6433 * spa_validate_aux() will have already made sure this nvlist is
6434 * valid and the vdevs are labeled appropriately.
6435 */
6436 if (sav->sav_object == 0) {
6437 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6438 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6439 sizeof (uint64_t), tx);
6440 VERIFY(zap_update(spa->spa_meta_objset,
6441 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6442 &sav->sav_object, tx) == 0);
6443 }
6444
6445 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6446 if (sav->sav_count == 0) {
6447 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6448 } else {
6449 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6450 for (i = 0; i < sav->sav_count; i++)
6451 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6452 B_FALSE, VDEV_CONFIG_L2CACHE);
6453 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6454 sav->sav_count) == 0);
6455 for (i = 0; i < sav->sav_count; i++)
6456 nvlist_free(list[i]);
6457 kmem_free(list, sav->sav_count * sizeof (void *));
6458 }
6459
6460 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6461 nvlist_free(nvroot);
6462
6463 sav->sav_sync = B_FALSE;
6464 }
6465
6466 /*
6467 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6468 * The all-vdev ZAP must be empty.
6469 */
6470 static void
spa_avz_build(vdev_t * vd,uint64_t avz,dmu_tx_t * tx)6471 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
6472 {
6473 spa_t *spa = vd->vdev_spa;
6474 if (vd->vdev_top_zap != 0) {
6475 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6476 vd->vdev_top_zap, tx));
6477 }
6478 if (vd->vdev_leaf_zap != 0) {
6479 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6480 vd->vdev_leaf_zap, tx));
6481 }
6482 for (uint64_t i = 0; i < vd->vdev_children; i++) {
6483 spa_avz_build(vd->vdev_child[i], avz, tx);
6484 }
6485 }
6486
6487 static void
spa_sync_config_object(spa_t * spa,dmu_tx_t * tx)6488 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6489 {
6490 nvlist_t *config;
6491
6492 /*
6493 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6494 * its config may not be dirty but we still need to build per-vdev ZAPs.
6495 * Similarly, if the pool is being assembled (e.g. after a split), we
6496 * need to rebuild the AVZ although the config may not be dirty.
6497 */
6498 if (list_is_empty(&spa->spa_config_dirty_list) &&
6499 spa->spa_avz_action == AVZ_ACTION_NONE)
6500 return;
6501
6502 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6503
6504 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
6505 spa->spa_all_vdev_zaps != 0);
6506
6507 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
6508 /* Make and build the new AVZ */
6509 uint64_t new_avz = zap_create(spa->spa_meta_objset,
6510 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
6511 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
6512
6513 /* Diff old AVZ with new one */
6514 zap_cursor_t zc;
6515 zap_attribute_t za;
6516
6517 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6518 spa->spa_all_vdev_zaps);
6519 zap_cursor_retrieve(&zc, &za) == 0;
6520 zap_cursor_advance(&zc)) {
6521 uint64_t vdzap = za.za_first_integer;
6522 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
6523 vdzap) == ENOENT) {
6524 /*
6525 * ZAP is listed in old AVZ but not in new one;
6526 * destroy it
6527 */
6528 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
6529 tx));
6530 }
6531 }
6532
6533 zap_cursor_fini(&zc);
6534
6535 /* Destroy the old AVZ */
6536 VERIFY0(zap_destroy(spa->spa_meta_objset,
6537 spa->spa_all_vdev_zaps, tx));
6538
6539 /* Replace the old AVZ in the dir obj with the new one */
6540 VERIFY0(zap_update(spa->spa_meta_objset,
6541 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
6542 sizeof (new_avz), 1, &new_avz, tx));
6543
6544 spa->spa_all_vdev_zaps = new_avz;
6545 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
6546 zap_cursor_t zc;
6547 zap_attribute_t za;
6548
6549 /* Walk through the AVZ and destroy all listed ZAPs */
6550 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6551 spa->spa_all_vdev_zaps);
6552 zap_cursor_retrieve(&zc, &za) == 0;
6553 zap_cursor_advance(&zc)) {
6554 uint64_t zap = za.za_first_integer;
6555 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
6556 }
6557
6558 zap_cursor_fini(&zc);
6559
6560 /* Destroy and unlink the AVZ itself */
6561 VERIFY0(zap_destroy(spa->spa_meta_objset,
6562 spa->spa_all_vdev_zaps, tx));
6563 VERIFY0(zap_remove(spa->spa_meta_objset,
6564 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
6565 spa->spa_all_vdev_zaps = 0;
6566 }
6567
6568 if (spa->spa_all_vdev_zaps == 0) {
6569 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
6570 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
6571 DMU_POOL_VDEV_ZAP_MAP, tx);
6572 }
6573 spa->spa_avz_action = AVZ_ACTION_NONE;
6574
6575 /* Create ZAPs for vdevs that don't have them. */
6576 vdev_construct_zaps(spa->spa_root_vdev, tx);
6577
6578 config = spa_config_generate(spa, spa->spa_root_vdev,
6579 dmu_tx_get_txg(tx), B_FALSE);
6580
6581 /*
6582 * If we're upgrading the spa version then make sure that
6583 * the config object gets updated with the correct version.
6584 */
6585 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6586 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6587 spa->spa_uberblock.ub_version);
6588
6589 spa_config_exit(spa, SCL_STATE, FTAG);
6590
6591 nvlist_free(spa->spa_config_syncing);
6592 spa->spa_config_syncing = config;
6593
6594 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6595 }
6596
6597 static void
spa_sync_version(void * arg,dmu_tx_t * tx)6598 spa_sync_version(void *arg, dmu_tx_t *tx)
6599 {
6600 uint64_t *versionp = arg;
6601 uint64_t version = *versionp;
6602 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6603
6604 /*
6605 * Setting the version is special cased when first creating the pool.
6606 */
6607 ASSERT(tx->tx_txg != TXG_INITIAL);
6608
6609 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6610 ASSERT(version >= spa_version(spa));
6611
6612 spa->spa_uberblock.ub_version = version;
6613 vdev_config_dirty(spa->spa_root_vdev);
6614 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6615 }
6616
6617 /*
6618 * Set zpool properties.
6619 */
6620 static void
spa_sync_props(void * arg,dmu_tx_t * tx)6621 spa_sync_props(void *arg, dmu_tx_t *tx)
6622 {
6623 nvlist_t *nvp = arg;
6624 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6625 objset_t *mos = spa->spa_meta_objset;
6626 nvpair_t *elem = NULL;
6627
6628 mutex_enter(&spa->spa_props_lock);
6629
6630 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6631 uint64_t intval;
6632 char *strval, *fname;
6633 zpool_prop_t prop;
6634 const char *propname;
6635 zprop_type_t proptype;
6636 spa_feature_t fid;
6637
6638 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6639 case ZPROP_INVAL:
6640 /*
6641 * We checked this earlier in spa_prop_validate().
6642 */
6643 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6644
6645 fname = strchr(nvpair_name(elem), '@') + 1;
6646 VERIFY0(zfeature_lookup_name(fname, &fid));
6647
6648 spa_feature_enable(spa, fid, tx);
6649 spa_history_log_internal(spa, "set", tx,
6650 "%s=enabled", nvpair_name(elem));
6651 break;
6652
6653 case ZPOOL_PROP_VERSION:
6654 intval = fnvpair_value_uint64(elem);
6655 /*
6656 * The version is synced seperatly before other
6657 * properties and should be correct by now.
6658 */
6659 ASSERT3U(spa_version(spa), >=, intval);
6660 break;
6661
6662 case ZPOOL_PROP_ALTROOT:
6663 /*
6664 * 'altroot' is a non-persistent property. It should
6665 * have been set temporarily at creation or import time.
6666 */
6667 ASSERT(spa->spa_root != NULL);
6668 break;
6669
6670 case ZPOOL_PROP_READONLY:
6671 case ZPOOL_PROP_CACHEFILE:
6672 /*
6673 * 'readonly' and 'cachefile' are also non-persisitent
6674 * properties.
6675 */
6676 break;
6677 case ZPOOL_PROP_COMMENT:
6678 strval = fnvpair_value_string(elem);
6679 if (spa->spa_comment != NULL)
6680 spa_strfree(spa->spa_comment);
6681 spa->spa_comment = spa_strdup(strval);
6682 /*
6683 * We need to dirty the configuration on all the vdevs
6684 * so that their labels get updated. It's unnecessary
6685 * to do this for pool creation since the vdev's
6686 * configuratoin has already been dirtied.
6687 */
6688 if (tx->tx_txg != TXG_INITIAL)
6689 vdev_config_dirty(spa->spa_root_vdev);
6690 spa_history_log_internal(spa, "set", tx,
6691 "%s=%s", nvpair_name(elem), strval);
6692 break;
6693 default:
6694 /*
6695 * Set pool property values in the poolprops mos object.
6696 */
6697 if (spa->spa_pool_props_object == 0) {
6698 spa->spa_pool_props_object =
6699 zap_create_link(mos, DMU_OT_POOL_PROPS,
6700 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6701 tx);
6702 }
6703
6704 /* normalize the property name */
6705 propname = zpool_prop_to_name(prop);
6706 proptype = zpool_prop_get_type(prop);
6707
6708 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6709 ASSERT(proptype == PROP_TYPE_STRING);
6710 strval = fnvpair_value_string(elem);
6711 VERIFY0(zap_update(mos,
6712 spa->spa_pool_props_object, propname,
6713 1, strlen(strval) + 1, strval, tx));
6714 spa_history_log_internal(spa, "set", tx,
6715 "%s=%s", nvpair_name(elem), strval);
6716 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6717 intval = fnvpair_value_uint64(elem);
6718
6719 if (proptype == PROP_TYPE_INDEX) {
6720 const char *unused;
6721 VERIFY0(zpool_prop_index_to_string(
6722 prop, intval, &unused));
6723 }
6724 VERIFY0(zap_update(mos,
6725 spa->spa_pool_props_object, propname,
6726 8, 1, &intval, tx));
6727 spa_history_log_internal(spa, "set", tx,
6728 "%s=%lld", nvpair_name(elem), intval);
6729 } else {
6730 ASSERT(0); /* not allowed */
6731 }
6732
6733 switch (prop) {
6734 case ZPOOL_PROP_DELEGATION:
6735 spa->spa_delegation = intval;
6736 break;
6737 case ZPOOL_PROP_BOOTFS:
6738 spa->spa_bootfs = intval;
6739 break;
6740 case ZPOOL_PROP_FAILUREMODE:
6741 spa->spa_failmode = intval;
6742 break;
6743 case ZPOOL_PROP_AUTOEXPAND:
6744 spa->spa_autoexpand = intval;
6745 if (tx->tx_txg != TXG_INITIAL)
6746 spa_async_request(spa,
6747 SPA_ASYNC_AUTOEXPAND);
6748 break;
6749 case ZPOOL_PROP_DEDUPDITTO:
6750 spa->spa_dedup_ditto = intval;
6751 break;
6752 default:
6753 break;
6754 }
6755 }
6756
6757 }
6758
6759 mutex_exit(&spa->spa_props_lock);
6760 }
6761
6762 /*
6763 * Perform one-time upgrade on-disk changes. spa_version() does not
6764 * reflect the new version this txg, so there must be no changes this
6765 * txg to anything that the upgrade code depends on after it executes.
6766 * Therefore this must be called after dsl_pool_sync() does the sync
6767 * tasks.
6768 */
6769 static void
spa_sync_upgrades(spa_t * spa,dmu_tx_t * tx)6770 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6771 {
6772 dsl_pool_t *dp = spa->spa_dsl_pool;
6773
6774 ASSERT(spa->spa_sync_pass == 1);
6775
6776 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6777
6778 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6779 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6780 dsl_pool_create_origin(dp, tx);
6781
6782 /* Keeping the origin open increases spa_minref */
6783 spa->spa_minref += 3;
6784 }
6785
6786 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6787 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6788 dsl_pool_upgrade_clones(dp, tx);
6789 }
6790
6791 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6792 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6793 dsl_pool_upgrade_dir_clones(dp, tx);
6794
6795 /* Keeping the freedir open increases spa_minref */
6796 spa->spa_minref += 3;
6797 }
6798
6799 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6800 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6801 spa_feature_create_zap_objects(spa, tx);
6802 }
6803
6804 /*
6805 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6806 * when possibility to use lz4 compression for metadata was added
6807 * Old pools that have this feature enabled must be upgraded to have
6808 * this feature active
6809 */
6810 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6811 boolean_t lz4_en = spa_feature_is_enabled(spa,
6812 SPA_FEATURE_LZ4_COMPRESS);
6813 boolean_t lz4_ac = spa_feature_is_active(spa,
6814 SPA_FEATURE_LZ4_COMPRESS);
6815
6816 if (lz4_en && !lz4_ac)
6817 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6818 }
6819
6820 /*
6821 * If we haven't written the salt, do so now. Note that the
6822 * feature may not be activated yet, but that's fine since
6823 * the presence of this ZAP entry is backwards compatible.
6824 */
6825 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6826 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6827 VERIFY0(zap_add(spa->spa_meta_objset,
6828 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6829 sizeof (spa->spa_cksum_salt.zcs_bytes),
6830 spa->spa_cksum_salt.zcs_bytes, tx));
6831 }
6832
6833 rrw_exit(&dp->dp_config_rwlock, FTAG);
6834 }
6835
6836 /*
6837 * Sync the specified transaction group. New blocks may be dirtied as
6838 * part of the process, so we iterate until it converges.
6839 */
6840
6841 void
spa_sync(spa_t * spa,uint64_t txg)6842 spa_sync(spa_t *spa, uint64_t txg)
6843 {
6844 dsl_pool_t *dp = spa->spa_dsl_pool;
6845 objset_t *mos = spa->spa_meta_objset;
6846 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6847 vdev_t *rvd = spa->spa_root_vdev;
6848 vdev_t *vd;
6849 dmu_tx_t *tx;
6850 int error;
6851 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6852 zfs_vdev_queue_depth_pct / 100;
6853
6854 VERIFY(spa_writeable(spa));
6855
6856 /*
6857 * Lock out configuration changes.
6858 */
6859 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6860
6861 spa->spa_syncing_txg = txg;
6862 spa->spa_sync_pass = 0;
6863
6864 mutex_enter(&spa->spa_alloc_lock);
6865 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6866 mutex_exit(&spa->spa_alloc_lock);
6867
6868 /*
6869 * If there are any pending vdev state changes, convert them
6870 * into config changes that go out with this transaction group.
6871 */
6872 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6873 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6874 /*
6875 * We need the write lock here because, for aux vdevs,
6876 * calling vdev_config_dirty() modifies sav_config.
6877 * This is ugly and will become unnecessary when we
6878 * eliminate the aux vdev wart by integrating all vdevs
6879 * into the root vdev tree.
6880 */
6881 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6882 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6883 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6884 vdev_state_clean(vd);
6885 vdev_config_dirty(vd);
6886 }
6887 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6888 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6889 }
6890 spa_config_exit(spa, SCL_STATE, FTAG);
6891
6892 tx = dmu_tx_create_assigned(dp, txg);
6893
6894 spa->spa_sync_starttime = gethrtime();
6895 #ifdef illumos
6896 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6897 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6898 #endif /* illumos */
6899 #ifdef __FreeBSD__
6900 #ifdef _KERNEL
6901 callout_schedule(&spa->spa_deadman_cycid,
6902 hz * spa->spa_deadman_synctime / NANOSEC);
6903 #endif
6904 #endif /* __FreeBSD__ */
6905 #ifdef __NetBSD__
6906 #ifdef _KERNEL
6907 callout_schedule(&spa->spa_deadman_cycid,
6908 hz * spa->spa_deadman_synctime / NANOSEC);
6909 #endif
6910 #endif
6911
6912 /*
6913 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6914 * set spa_deflate if we have no raid-z vdevs.
6915 */
6916 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6917 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6918 int i;
6919
6920 for (i = 0; i < rvd->vdev_children; i++) {
6921 vd = rvd->vdev_child[i];
6922 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6923 break;
6924 }
6925 if (i == rvd->vdev_children) {
6926 spa->spa_deflate = TRUE;
6927 VERIFY(0 == zap_add(spa->spa_meta_objset,
6928 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6929 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6930 }
6931 }
6932
6933 /*
6934 * Set the top-level vdev's max queue depth. Evaluate each
6935 * top-level's async write queue depth in case it changed.
6936 * The max queue depth will not change in the middle of syncing
6937 * out this txg.
6938 */
6939 uint64_t queue_depth_total = 0;
6940 for (int c = 0; c < rvd->vdev_children; c++) {
6941 vdev_t *tvd = rvd->vdev_child[c];
6942 metaslab_group_t *mg = tvd->vdev_mg;
6943
6944 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6945 !metaslab_group_initialized(mg))
6946 continue;
6947
6948 /*
6949 * It is safe to do a lock-free check here because only async
6950 * allocations look at mg_max_alloc_queue_depth, and async
6951 * allocations all happen from spa_sync().
6952 */
6953 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6954 mg->mg_max_alloc_queue_depth = max_queue_depth;
6955 queue_depth_total += mg->mg_max_alloc_queue_depth;
6956 }
6957 metaslab_class_t *mc = spa_normal_class(spa);
6958 ASSERT0(refcount_count(&mc->mc_alloc_slots));
6959 mc->mc_alloc_max_slots = queue_depth_total;
6960 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6961
6962 ASSERT3U(mc->mc_alloc_max_slots, <=,
6963 max_queue_depth * rvd->vdev_children);
6964
6965 /*
6966 * Iterate to convergence.
6967 */
6968 do {
6969 int pass = ++spa->spa_sync_pass;
6970
6971 spa_sync_config_object(spa, tx);
6972 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6973 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6974 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6975 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6976 spa_errlog_sync(spa, txg);
6977 dsl_pool_sync(dp, txg);
6978
6979 if (pass < zfs_sync_pass_deferred_free) {
6980 spa_sync_frees(spa, free_bpl, tx);
6981 } else {
6982 /*
6983 * We can not defer frees in pass 1, because
6984 * we sync the deferred frees later in pass 1.
6985 */
6986 ASSERT3U(pass, >, 1);
6987 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6988 &spa->spa_deferred_bpobj, tx);
6989 }
6990
6991 ddt_sync(spa, txg);
6992 dsl_scan_sync(dp, tx);
6993
6994 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6995 vdev_sync(vd, txg);
6996
6997 if (pass == 1) {
6998 spa_sync_upgrades(spa, tx);
6999 ASSERT3U(txg, >=,
7000 spa->spa_uberblock.ub_rootbp.blk_birth);
7001 /*
7002 * Note: We need to check if the MOS is dirty
7003 * because we could have marked the MOS dirty
7004 * without updating the uberblock (e.g. if we
7005 * have sync tasks but no dirty user data). We
7006 * need to check the uberblock's rootbp because
7007 * it is updated if we have synced out dirty
7008 * data (though in this case the MOS will most
7009 * likely also be dirty due to second order
7010 * effects, we don't want to rely on that here).
7011 */
7012 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7013 !dmu_objset_is_dirty(mos, txg)) {
7014 /*
7015 * Nothing changed on the first pass,
7016 * therefore this TXG is a no-op. Avoid
7017 * syncing deferred frees, so that we
7018 * can keep this TXG as a no-op.
7019 */
7020 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7021 txg));
7022 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7023 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7024 break;
7025 }
7026 spa_sync_deferred_frees(spa, tx);
7027 }
7028
7029 } while (dmu_objset_is_dirty(mos, txg));
7030
7031 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7032 /*
7033 * Make sure that the number of ZAPs for all the vdevs matches
7034 * the number of ZAPs in the per-vdev ZAP list. This only gets
7035 * called if the config is dirty; otherwise there may be
7036 * outstanding AVZ operations that weren't completed in
7037 * spa_sync_config_object.
7038 */
7039 uint64_t all_vdev_zap_entry_count;
7040 ASSERT0(zap_count(spa->spa_meta_objset,
7041 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7042 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7043 all_vdev_zap_entry_count);
7044 }
7045
7046 /*
7047 * Rewrite the vdev configuration (which includes the uberblock)
7048 * to commit the transaction group.
7049 *
7050 * If there are no dirty vdevs, we sync the uberblock to a few
7051 * random top-level vdevs that are known to be visible in the
7052 * config cache (see spa_vdev_add() for a complete description).
7053 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7054 */
7055 for (;;) {
7056 /*
7057 * We hold SCL_STATE to prevent vdev open/close/etc.
7058 * while we're attempting to write the vdev labels.
7059 */
7060 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7061
7062 if (list_is_empty(&spa->spa_config_dirty_list)) {
7063 vdev_t *svd[SPA_DVAS_PER_BP];
7064 int svdcount = 0;
7065 int children = rvd->vdev_children;
7066 int c0 = spa_get_random(children);
7067
7068 for (int c = 0; c < children; c++) {
7069 vd = rvd->vdev_child[(c0 + c) % children];
7070 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
7071 continue;
7072 svd[svdcount++] = vd;
7073 if (svdcount == SPA_DVAS_PER_BP)
7074 break;
7075 }
7076 error = vdev_config_sync(svd, svdcount, txg);
7077 } else {
7078 error = vdev_config_sync(rvd->vdev_child,
7079 rvd->vdev_children, txg);
7080 }
7081
7082 if (error == 0)
7083 spa->spa_last_synced_guid = rvd->vdev_guid;
7084
7085 spa_config_exit(spa, SCL_STATE, FTAG);
7086
7087 if (error == 0)
7088 break;
7089 zio_suspend(spa, NULL);
7090 zio_resume_wait(spa);
7091 }
7092 dmu_tx_commit(tx);
7093
7094 #ifdef illumos
7095 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
7096 #endif /* illumos */
7097 #ifdef __FreeBSD__
7098 #ifdef _KERNEL
7099 callout_drain(&spa->spa_deadman_cycid);
7100 #endif
7101 #endif /* __FreeBSD__ */
7102 #ifdef __NetBSD__
7103 #ifdef _KERNEL
7104 callout_drain(&spa->spa_deadman_cycid);
7105 #endif
7106 #endif /* __NetBSD__ */
7107
7108 /*
7109 * Clear the dirty config list.
7110 */
7111 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
7112 vdev_config_clean(vd);
7113
7114 /*
7115 * Now that the new config has synced transactionally,
7116 * let it become visible to the config cache.
7117 */
7118 if (spa->spa_config_syncing != NULL) {
7119 spa_config_set(spa, spa->spa_config_syncing);
7120 spa->spa_config_txg = txg;
7121 spa->spa_config_syncing = NULL;
7122 }
7123
7124 dsl_pool_sync_done(dp, txg);
7125
7126 mutex_enter(&spa->spa_alloc_lock);
7127 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7128 mutex_exit(&spa->spa_alloc_lock);
7129
7130 /*
7131 * Update usable space statistics.
7132 */
7133 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
7134 vdev_sync_done(vd, txg);
7135
7136 spa_update_dspace(spa);
7137
7138 /*
7139 * It had better be the case that we didn't dirty anything
7140 * since vdev_config_sync().
7141 */
7142 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
7143 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7144 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
7145
7146 spa->spa_sync_pass = 0;
7147
7148 /*
7149 * Update the last synced uberblock here. We want to do this at
7150 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7151 * will be guaranteed that all the processing associated with
7152 * that txg has been completed.
7153 */
7154 spa->spa_ubsync = spa->spa_uberblock;
7155 spa_config_exit(spa, SCL_CONFIG, FTAG);
7156
7157 spa_handle_ignored_writes(spa);
7158
7159 /*
7160 * If any async tasks have been requested, kick them off.
7161 */
7162 spa_async_dispatch(spa);
7163 spa_async_dispatch_vd(spa);
7164 }
7165
7166 /*
7167 * Sync all pools. We don't want to hold the namespace lock across these
7168 * operations, so we take a reference on the spa_t and drop the lock during the
7169 * sync.
7170 */
7171 void
spa_sync_allpools(void)7172 spa_sync_allpools(void)
7173 {
7174 spa_t *spa = NULL;
7175 mutex_enter(&spa_namespace_lock);
7176 while ((spa = spa_next(spa)) != NULL) {
7177 if (spa_state(spa) != POOL_STATE_ACTIVE ||
7178 !spa_writeable(spa) || spa_suspended(spa))
7179 continue;
7180 spa_open_ref(spa, FTAG);
7181 mutex_exit(&spa_namespace_lock);
7182 txg_wait_synced(spa_get_dsl(spa), 0);
7183 mutex_enter(&spa_namespace_lock);
7184 spa_close(spa, FTAG);
7185 }
7186 mutex_exit(&spa_namespace_lock);
7187 }
7188
7189 /*
7190 * ==========================================================================
7191 * Miscellaneous routines
7192 * ==========================================================================
7193 */
7194
7195 /*
7196 * Remove all pools in the system.
7197 */
7198 void
spa_evict_all(void)7199 spa_evict_all(void)
7200 {
7201 spa_t *spa;
7202
7203 /*
7204 * Remove all cached state. All pools should be closed now,
7205 * so every spa in the AVL tree should be unreferenced.
7206 */
7207 mutex_enter(&spa_namespace_lock);
7208 while ((spa = spa_next(NULL)) != NULL) {
7209 /*
7210 * Stop async tasks. The async thread may need to detach
7211 * a device that's been replaced, which requires grabbing
7212 * spa_namespace_lock, so we must drop it here.
7213 */
7214 spa_open_ref(spa, FTAG);
7215 mutex_exit(&spa_namespace_lock);
7216 spa_async_suspend(spa);
7217 mutex_enter(&spa_namespace_lock);
7218 spa_close(spa, FTAG);
7219
7220 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
7221 spa_unload(spa);
7222 spa_deactivate(spa);
7223 }
7224 spa_remove(spa);
7225 }
7226 mutex_exit(&spa_namespace_lock);
7227 }
7228
7229 vdev_t *
spa_lookup_by_guid(spa_t * spa,uint64_t guid,boolean_t aux)7230 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
7231 {
7232 vdev_t *vd;
7233 int i;
7234
7235 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
7236 return (vd);
7237
7238 if (aux) {
7239 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7240 vd = spa->spa_l2cache.sav_vdevs[i];
7241 if (vd->vdev_guid == guid)
7242 return (vd);
7243 }
7244
7245 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7246 vd = spa->spa_spares.sav_vdevs[i];
7247 if (vd->vdev_guid == guid)
7248 return (vd);
7249 }
7250 }
7251
7252 return (NULL);
7253 }
7254
7255 void
spa_upgrade(spa_t * spa,uint64_t version)7256 spa_upgrade(spa_t *spa, uint64_t version)
7257 {
7258 ASSERT(spa_writeable(spa));
7259
7260 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7261
7262 /*
7263 * This should only be called for a non-faulted pool, and since a
7264 * future version would result in an unopenable pool, this shouldn't be
7265 * possible.
7266 */
7267 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7268 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7269
7270 spa->spa_uberblock.ub_version = version;
7271 vdev_config_dirty(spa->spa_root_vdev);
7272
7273 spa_config_exit(spa, SCL_ALL, FTAG);
7274
7275 txg_wait_synced(spa_get_dsl(spa), 0);
7276 }
7277
7278 boolean_t
spa_has_spare(spa_t * spa,uint64_t guid)7279 spa_has_spare(spa_t *spa, uint64_t guid)
7280 {
7281 int i;
7282 uint64_t spareguid;
7283 spa_aux_vdev_t *sav = &spa->spa_spares;
7284
7285 for (i = 0; i < sav->sav_count; i++)
7286 if (sav->sav_vdevs[i]->vdev_guid == guid)
7287 return (B_TRUE);
7288
7289 for (i = 0; i < sav->sav_npending; i++) {
7290 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7291 &spareguid) == 0 && spareguid == guid)
7292 return (B_TRUE);
7293 }
7294
7295 return (B_FALSE);
7296 }
7297
7298 /*
7299 * Check if a pool has an active shared spare device.
7300 * Note: reference count of an active spare is 2, as a spare and as a replace
7301 */
7302 static boolean_t
spa_has_active_shared_spare(spa_t * spa)7303 spa_has_active_shared_spare(spa_t *spa)
7304 {
7305 int i, refcnt;
7306 uint64_t pool;
7307 spa_aux_vdev_t *sav = &spa->spa_spares;
7308
7309 for (i = 0; i < sav->sav_count; i++) {
7310 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
7311 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7312 refcnt > 2)
7313 return (B_TRUE);
7314 }
7315
7316 return (B_FALSE);
7317 }
7318
7319 static sysevent_t *
spa_event_create(spa_t * spa,vdev_t * vd,const char * name)7320 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
7321 {
7322 sysevent_t *ev = NULL;
7323 #ifdef _KERNEL
7324 sysevent_attr_list_t *attr = NULL;
7325 sysevent_value_t value;
7326
7327 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7328 SE_SLEEP);
7329 ASSERT(ev != NULL);
7330
7331 value.value_type = SE_DATA_TYPE_STRING;
7332 value.value.sv_string = spa_name(spa);
7333 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7334 goto done;
7335
7336 value.value_type = SE_DATA_TYPE_UINT64;
7337 value.value.sv_uint64 = spa_guid(spa);
7338 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7339 goto done;
7340
7341 if (vd) {
7342 value.value_type = SE_DATA_TYPE_UINT64;
7343 value.value.sv_uint64 = vd->vdev_guid;
7344 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7345 SE_SLEEP) != 0)
7346 goto done;
7347
7348 if (vd->vdev_path) {
7349 value.value_type = SE_DATA_TYPE_STRING;
7350 value.value.sv_string = vd->vdev_path;
7351 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7352 &value, SE_SLEEP) != 0)
7353 goto done;
7354 }
7355 }
7356
7357 if (sysevent_attach_attributes(ev, attr) != 0)
7358 goto done;
7359 attr = NULL;
7360
7361 done:
7362 if (attr)
7363 sysevent_free_attr(attr);
7364
7365 #endif
7366 return (ev);
7367 }
7368
7369 static void
spa_event_post(sysevent_t * ev)7370 spa_event_post(sysevent_t *ev)
7371 {
7372 #ifdef _KERNEL
7373 sysevent_id_t eid;
7374
7375 (void) log_sysevent(ev, SE_SLEEP, &eid);
7376 sysevent_free(ev);
7377 #endif
7378 }
7379
7380 /*
7381 * Post a sysevent corresponding to the given event. The 'name' must be one of
7382 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7383 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7384 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7385 * or zdb as real changes.
7386 */
7387 void
spa_event_notify(spa_t * spa,vdev_t * vd,const char * name)7388 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7389 {
7390 spa_event_post(spa_event_create(spa, vd, name));
7391 }
7392