]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
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 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
4f072827 | 24 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
733b5722 | 25 | * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved. |
0c66c32d | 26 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
3c67d83a | 27 | * Copyright 2013 Saso Kiselkov. All rights reserved. |
e550644f BB |
28 | * Copyright (c) 2014 Integros [integros.com] |
29 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
a0bd735a | 30 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
f65fbee1 | 31 | * Copyright 2018 Joyent, Inc. |
3c819a2c | 32 | * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. |
12fa0466 | 33 | * Copyright 2017 Joyent, Inc. |
cc99f275 | 34 | * Copyright (c) 2017, Intel Corporation. |
658fb802 | 35 | * Copyright (c) 2021, Colm Buckley <colm@tuatha.org> |
a38718a6 | 36 | */ |
34dc7c2f | 37 | |
34dc7c2f | 38 | /* |
e49f1e20 WA |
39 | * SPA: Storage Pool Allocator |
40 | * | |
34dc7c2f BB |
41 | * This file contains all the routines used when modifying on-disk SPA state. |
42 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
43 | * pool. | |
44 | */ | |
45 | ||
46 | #include <sys/zfs_context.h> | |
47 | #include <sys/fm/fs/zfs.h> | |
48 | #include <sys/spa_impl.h> | |
49 | #include <sys/zio.h> | |
50 | #include <sys/zio_checksum.h> | |
34dc7c2f BB |
51 | #include <sys/dmu.h> |
52 | #include <sys/dmu_tx.h> | |
53 | #include <sys/zap.h> | |
54 | #include <sys/zil.h> | |
67a1b037 | 55 | #include <sys/brt.h> |
428870ff | 56 | #include <sys/ddt.h> |
34dc7c2f | 57 | #include <sys/vdev_impl.h> |
a1d477c2 MA |
58 | #include <sys/vdev_removal.h> |
59 | #include <sys/vdev_indirect_mapping.h> | |
60 | #include <sys/vdev_indirect_births.h> | |
619f0976 | 61 | #include <sys/vdev_initialize.h> |
9a49d3f3 | 62 | #include <sys/vdev_rebuild.h> |
1b939560 | 63 | #include <sys/vdev_trim.h> |
c28b2279 | 64 | #include <sys/vdev_disk.h> |
b2255edc | 65 | #include <sys/vdev_draid.h> |
34dc7c2f | 66 | #include <sys/metaslab.h> |
428870ff | 67 | #include <sys/metaslab_impl.h> |
379ca9cf | 68 | #include <sys/mmp.h> |
34dc7c2f BB |
69 | #include <sys/uberblock_impl.h> |
70 | #include <sys/txg.h> | |
71 | #include <sys/avl.h> | |
a1d477c2 | 72 | #include <sys/bpobj.h> |
34dc7c2f BB |
73 | #include <sys/dmu_traverse.h> |
74 | #include <sys/dmu_objset.h> | |
75 | #include <sys/unique.h> | |
76 | #include <sys/dsl_pool.h> | |
77 | #include <sys/dsl_dataset.h> | |
78 | #include <sys/dsl_dir.h> | |
79 | #include <sys/dsl_prop.h> | |
80 | #include <sys/dsl_synctask.h> | |
81 | #include <sys/fs/zfs.h> | |
82 | #include <sys/arc.h> | |
83 | #include <sys/callb.h> | |
84 | #include <sys/systeminfo.h> | |
9babb374 | 85 | #include <sys/zfs_ioctl.h> |
428870ff | 86 | #include <sys/dsl_scan.h> |
9ae529ec | 87 | #include <sys/zfeature.h> |
13fe0198 | 88 | #include <sys/dsl_destroy.h> |
526af785 | 89 | #include <sys/zvol.h> |
34dc7c2f | 90 | |
d164b209 | 91 | #ifdef _KERNEL |
12fa0466 DE |
92 | #include <sys/fm/protocol.h> |
93 | #include <sys/fm/util.h> | |
428870ff | 94 | #include <sys/callb.h> |
d164b209 | 95 | #include <sys/zone.h> |
c8242a96 | 96 | #include <sys/vmsystm.h> |
d164b209 BB |
97 | #endif /* _KERNEL */ |
98 | ||
34dc7c2f BB |
99 | #include "zfs_prop.h" |
100 | #include "zfs_comutil.h" | |
101 | ||
e6cfd633 WA |
102 | /* |
103 | * The interval, in seconds, at which failed configuration cache file writes | |
104 | * should be retried. | |
105 | */ | |
a1d477c2 | 106 | int zfs_ccw_retry_interval = 300; |
e6cfd633 | 107 | |
428870ff | 108 | typedef enum zti_modes { |
7ef5e54e | 109 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ |
7ef5e54e | 110 | ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ |
7457b024 | 111 | ZTI_MODE_SCALE, /* Taskqs scale with CPUs. */ |
7ef5e54e AL |
112 | ZTI_MODE_NULL, /* don't create a taskq */ |
113 | ZTI_NMODES | |
428870ff | 114 | } zti_modes_t; |
34dc7c2f | 115 | |
7ef5e54e AL |
116 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } |
117 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
118 | #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } | |
7457b024 | 119 | #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 } |
7ef5e54e | 120 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } |
9babb374 | 121 | |
7ef5e54e AL |
122 | #define ZTI_N(n) ZTI_P(n, 1) |
123 | #define ZTI_ONE ZTI_N(1) | |
9babb374 BB |
124 | |
125 | typedef struct zio_taskq_info { | |
7ef5e54e | 126 | zti_modes_t zti_mode; |
428870ff | 127 | uint_t zti_value; |
7ef5e54e | 128 | uint_t zti_count; |
9babb374 BB |
129 | } zio_taskq_info_t; |
130 | ||
131 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
451041db | 132 | "iss", "iss_h", "int", "int_h" |
9babb374 BB |
133 | }; |
134 | ||
428870ff | 135 | /* |
7ef5e54e AL |
136 | * This table defines the taskq settings for each ZFS I/O type. When |
137 | * initializing a pool, we use this table to create an appropriately sized | |
138 | * taskq. Some operations are low volume and therefore have a small, static | |
139 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
140 | * macros. Other operations process a large amount of data; the ZTI_BATCH | |
141 | * macro causes us to create a taskq oriented for throughput. Some operations | |
1b939560 | 142 | * are so high frequency and short-lived that the taskq itself can become a |
7ef5e54e AL |
143 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an |
144 | * additional degree of parallelism specified by the number of threads per- | |
145 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
7457b024 AM |
146 | * particular taskq is chosen at random. ZTI_SCALE is similar to ZTI_BATCH, |
147 | * but with number of taskqs also scaling with number of CPUs. | |
7ef5e54e AL |
148 | * |
149 | * The different taskq priorities are to handle the different contexts (issue | |
150 | * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that | |
151 | * need to be handled with minimum delay. | |
428870ff | 152 | */ |
18168da7 | 153 | static const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { |
428870ff | 154 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ |
7ef5e54e | 155 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ |
7457b024 AM |
156 | { ZTI_N(8), ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* READ */ |
157 | { ZTI_BATCH, ZTI_N(5), ZTI_SCALE, ZTI_N(5) }, /* WRITE */ | |
158 | { ZTI_SCALE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ | |
7ef5e54e AL |
159 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ |
160 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ | |
1b939560 | 161 | { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */ |
9babb374 BB |
162 | }; |
163 | ||
13fe0198 MA |
164 | static void spa_sync_version(void *arg, dmu_tx_t *tx); |
165 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
b128c09f | 166 | static boolean_t spa_has_active_shared_spare(spa_t *spa); |
a926aab9 AZ |
167 | static int spa_load_impl(spa_t *spa, spa_import_type_t type, |
168 | const char **ereport); | |
572e2857 | 169 | static void spa_vdev_resilver_done(spa_t *spa); |
428870ff | 170 | |
18168da7 AZ |
171 | static uint_t zio_taskq_batch_pct = 80; /* 1 thread per cpu in pset */ |
172 | static uint_t zio_taskq_batch_tpq; /* threads per taskq */ | |
173 | static const boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ | |
174 | static const uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
428870ff | 175 | |
18168da7 | 176 | static const boolean_t spa_create_process = B_TRUE; /* no process => no sysdc */ |
428870ff | 177 | |
afd2f7b7 PZ |
178 | /* |
179 | * Report any spa_load_verify errors found, but do not fail spa_load. | |
180 | * This is used by zdb to analyze non-idle pools. | |
181 | */ | |
182 | boolean_t spa_load_verify_dryrun = B_FALSE; | |
183 | ||
e39fe05b FU |
184 | /* |
185 | * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ). | |
186 | * This is used by zdb for spacemaps verification. | |
187 | */ | |
188 | boolean_t spa_mode_readable_spacemaps = B_FALSE; | |
189 | ||
428870ff BB |
190 | /* |
191 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
192 | * to get the vdev stats associated with the imported devices. | |
193 | */ | |
194 | #define TRYIMPORT_NAME "$import" | |
34dc7c2f | 195 | |
6cb8e530 PZ |
196 | /* |
197 | * For debugging purposes: print out vdev tree during pool import. | |
198 | */ | |
18168da7 | 199 | static int spa_load_print_vdev_tree = B_FALSE; |
6cb8e530 PZ |
200 | |
201 | /* | |
202 | * A non-zero value for zfs_max_missing_tvds means that we allow importing | |
203 | * pools with missing top-level vdevs. This is strictly intended for advanced | |
204 | * pool recovery cases since missing data is almost inevitable. Pools with | |
205 | * missing devices can only be imported read-only for safety reasons, and their | |
206 | * fail-mode will be automatically set to "continue". | |
207 | * | |
208 | * With 1 missing vdev we should be able to import the pool and mount all | |
209 | * datasets. User data that was not modified after the missing device has been | |
210 | * added should be recoverable. This means that snapshots created prior to the | |
211 | * addition of that device should be completely intact. | |
212 | * | |
213 | * With 2 missing vdevs, some datasets may fail to mount since there are | |
214 | * dataset statistics that are stored as regular metadata. Some data might be | |
215 | * recoverable if those vdevs were added recently. | |
216 | * | |
217 | * With 3 or more missing vdevs, the pool is severely damaged and MOS entries | |
218 | * may be missing entirely. Chances of data recovery are very low. Note that | |
219 | * there are also risks of performing an inadvertent rewind as we might be | |
220 | * missing all the vdevs with the latest uberblocks. | |
221 | */ | |
ab8d9c17 | 222 | uint64_t zfs_max_missing_tvds = 0; |
6cb8e530 PZ |
223 | |
224 | /* | |
225 | * The parameters below are similar to zfs_max_missing_tvds but are only | |
226 | * intended for a preliminary open of the pool with an untrusted config which | |
227 | * might be incomplete or out-dated. | |
228 | * | |
229 | * We are more tolerant for pools opened from a cachefile since we could have | |
230 | * an out-dated cachefile where a device removal was not registered. | |
231 | * We could have set the limit arbitrarily high but in the case where devices | |
232 | * are really missing we would want to return the proper error codes; we chose | |
233 | * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available | |
234 | * and we get a chance to retrieve the trusted config. | |
235 | */ | |
236 | uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; | |
d2734cce | 237 | |
6cb8e530 PZ |
238 | /* |
239 | * In the case where config was assembled by scanning device paths (/dev/dsks | |
240 | * by default) we are less tolerant since all the existing devices should have | |
241 | * been detected and we want spa_load to return the right error codes. | |
242 | */ | |
243 | uint64_t zfs_max_missing_tvds_scan = 0; | |
244 | ||
d2734cce SD |
245 | /* |
246 | * Debugging aid that pauses spa_sync() towards the end. | |
247 | */ | |
18168da7 | 248 | static const boolean_t zfs_pause_spa_sync = B_FALSE; |
d2734cce | 249 | |
37f03da8 SH |
250 | /* |
251 | * Variables to indicate the livelist condense zthr func should wait at certain | |
252 | * points for the livelist to be removed - used to test condense/destroy races | |
253 | */ | |
18168da7 AZ |
254 | static int zfs_livelist_condense_zthr_pause = 0; |
255 | static int zfs_livelist_condense_sync_pause = 0; | |
37f03da8 SH |
256 | |
257 | /* | |
258 | * Variables to track whether or not condense cancellation has been | |
259 | * triggered in testing. | |
260 | */ | |
18168da7 AZ |
261 | static int zfs_livelist_condense_sync_cancel = 0; |
262 | static int zfs_livelist_condense_zthr_cancel = 0; | |
37f03da8 SH |
263 | |
264 | /* | |
265 | * Variable to track whether or not extra ALLOC blkptrs were added to a | |
266 | * livelist entry while it was being condensed (caused by the way we track | |
267 | * remapped blkptrs in dbuf_remap_impl) | |
268 | */ | |
18168da7 | 269 | static int zfs_livelist_condense_new_alloc = 0; |
37f03da8 | 270 | |
34dc7c2f BB |
271 | /* |
272 | * ========================================================================== | |
273 | * SPA properties routines | |
274 | * ========================================================================== | |
275 | */ | |
276 | ||
277 | /* | |
278 | * Add a (source=src, propname=propval) list to an nvlist. | |
279 | */ | |
280 | static void | |
a926aab9 | 281 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, const char *strval, |
34dc7c2f BB |
282 | uint64_t intval, zprop_source_t src) |
283 | { | |
284 | const char *propname = zpool_prop_to_name(prop); | |
285 | nvlist_t *propval; | |
286 | ||
65ad5d11 AJ |
287 | propval = fnvlist_alloc(); |
288 | fnvlist_add_uint64(propval, ZPROP_SOURCE, src); | |
34dc7c2f BB |
289 | |
290 | if (strval != NULL) | |
65ad5d11 | 291 | fnvlist_add_string(propval, ZPROP_VALUE, strval); |
34dc7c2f | 292 | else |
65ad5d11 | 293 | fnvlist_add_uint64(propval, ZPROP_VALUE, intval); |
34dc7c2f | 294 | |
65ad5d11 | 295 | fnvlist_add_nvlist(nvl, propname, propval); |
34dc7c2f BB |
296 | nvlist_free(propval); |
297 | } | |
298 | ||
299 | /* | |
300 | * Get property values from the spa configuration. | |
301 | */ | |
302 | static void | |
303 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
304 | { | |
1bd201e7 | 305 | vdev_t *rvd = spa->spa_root_vdev; |
9ae529ec | 306 | dsl_pool_t *pool = spa->spa_dsl_pool; |
f3a7f661 | 307 | uint64_t size, alloc, cap, version; |
82ab6848 | 308 | const zprop_source_t src = ZPROP_SRC_NONE; |
b128c09f | 309 | spa_config_dirent_t *dp; |
f3a7f661 | 310 | metaslab_class_t *mc = spa_normal_class(spa); |
b128c09f BB |
311 | |
312 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
34dc7c2f | 313 | |
1bd201e7 | 314 | if (rvd != NULL) { |
cc99f275 DB |
315 | alloc = metaslab_class_get_alloc(mc); |
316 | alloc += metaslab_class_get_alloc(spa_special_class(spa)); | |
317 | alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); | |
aa755b35 | 318 | alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa)); |
cc99f275 DB |
319 | |
320 | size = metaslab_class_get_space(mc); | |
321 | size += metaslab_class_get_space(spa_special_class(spa)); | |
322 | size += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 | 323 | size += metaslab_class_get_space(spa_embedded_log_class(spa)); |
cc99f275 | 324 | |
d164b209 BB |
325 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); |
326 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
428870ff BB |
327 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); |
328 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
329 | size - alloc, src); | |
d2734cce SD |
330 | spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, |
331 | spa->spa_checkpoint_info.sci_dspace, src); | |
1bd201e7 | 332 | |
f3a7f661 GW |
333 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, |
334 | metaslab_class_fragmentation(mc), src); | |
335 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
336 | metaslab_class_expandable_space(mc), src); | |
572e2857 | 337 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, |
da92d5cb | 338 | (spa_mode(spa) == SPA_MODE_READ), src); |
d164b209 | 339 | |
428870ff | 340 | cap = (size == 0) ? 0 : (alloc * 100 / size); |
d164b209 BB |
341 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); |
342 | ||
428870ff BB |
343 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, |
344 | ddt_get_pool_dedup_ratio(spa), src); | |
67a1b037 PJD |
345 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONEUSED, NULL, |
346 | brt_get_used(spa), src); | |
347 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONESAVED, NULL, | |
348 | brt_get_saved(spa), src); | |
349 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONERATIO, NULL, | |
350 | brt_get_ratio(spa), src); | |
428870ff | 351 | |
d164b209 | 352 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, |
1bd201e7 | 353 | rvd->vdev_state, src); |
d164b209 BB |
354 | |
355 | version = spa_version(spa); | |
82ab6848 HM |
356 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { |
357 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
358 | version, ZPROP_SRC_DEFAULT); | |
359 | } else { | |
360 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
361 | version, ZPROP_SRC_LOCAL); | |
362 | } | |
a448a255 SD |
363 | spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID, |
364 | NULL, spa_load_guid(spa), src); | |
d164b209 | 365 | } |
34dc7c2f | 366 | |
9ae529ec | 367 | if (pool != NULL) { |
9ae529ec CS |
368 | /* |
369 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
370 | * when opening pools before this version freedir will be NULL. | |
371 | */ | |
fbeddd60 | 372 | if (pool->dp_free_dir != NULL) { |
9ae529ec | 373 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, |
d683ddbb JG |
374 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, |
375 | src); | |
9ae529ec CS |
376 | } else { |
377 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
378 | NULL, 0, src); | |
379 | } | |
fbeddd60 MA |
380 | |
381 | if (pool->dp_leak_dir != NULL) { | |
382 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
d683ddbb JG |
383 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, |
384 | src); | |
fbeddd60 MA |
385 | } else { |
386 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
387 | NULL, 0, src); | |
388 | } | |
9ae529ec CS |
389 | } |
390 | ||
34dc7c2f | 391 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); |
34dc7c2f | 392 | |
d96eb2b1 DM |
393 | if (spa->spa_comment != NULL) { |
394 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
395 | 0, ZPROP_SRC_LOCAL); | |
396 | } | |
397 | ||
658fb802 CB |
398 | if (spa->spa_compatibility != NULL) { |
399 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMPATIBILITY, | |
400 | spa->spa_compatibility, 0, ZPROP_SRC_LOCAL); | |
401 | } | |
402 | ||
34dc7c2f BB |
403 | if (spa->spa_root != NULL) |
404 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
405 | 0, ZPROP_SRC_LOCAL); | |
406 | ||
f1512ee6 MA |
407 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { |
408 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
409 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
410 | } else { | |
411 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
412 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
413 | } | |
414 | ||
50c957f7 NB |
415 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { |
416 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
417 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
418 | } else { | |
419 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
420 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
421 | } | |
422 | ||
b128c09f BB |
423 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { |
424 | if (dp->scd_path == NULL) { | |
34dc7c2f | 425 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f BB |
426 | "none", 0, ZPROP_SRC_LOCAL); |
427 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
34dc7c2f | 428 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f | 429 | dp->scd_path, 0, ZPROP_SRC_LOCAL); |
34dc7c2f BB |
430 | } |
431 | } | |
432 | } | |
433 | ||
434 | /* | |
435 | * Get zpool property values. | |
436 | */ | |
437 | int | |
438 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
439 | { | |
428870ff | 440 | objset_t *mos = spa->spa_meta_objset; |
34dc7c2f BB |
441 | zap_cursor_t zc; |
442 | zap_attribute_t za; | |
1743c737 | 443 | dsl_pool_t *dp; |
34dc7c2f BB |
444 | int err; |
445 | ||
79c76d5b | 446 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); |
c28b2279 | 447 | if (err) |
d1d7e268 | 448 | return (err); |
34dc7c2f | 449 | |
1743c737 AM |
450 | dp = spa_get_dsl(spa); |
451 | dsl_pool_config_enter(dp, FTAG); | |
b128c09f BB |
452 | mutex_enter(&spa->spa_props_lock); |
453 | ||
34dc7c2f BB |
454 | /* |
455 | * Get properties from the spa config. | |
456 | */ | |
457 | spa_prop_get_config(spa, nvp); | |
458 | ||
34dc7c2f | 459 | /* If no pool property object, no more prop to get. */ |
1743c737 | 460 | if (mos == NULL || spa->spa_pool_props_object == 0) |
c28b2279 | 461 | goto out; |
34dc7c2f BB |
462 | |
463 | /* | |
464 | * Get properties from the MOS pool property object. | |
465 | */ | |
466 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
467 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
468 | zap_cursor_advance(&zc)) { | |
469 | uint64_t intval = 0; | |
470 | char *strval = NULL; | |
471 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
472 | zpool_prop_t prop; | |
473 | ||
31864e3d | 474 | if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) |
34dc7c2f BB |
475 | continue; |
476 | ||
477 | switch (za.za_integer_length) { | |
478 | case 8: | |
479 | /* integer property */ | |
480 | if (za.za_first_integer != | |
481 | zpool_prop_default_numeric(prop)) | |
482 | src = ZPROP_SRC_LOCAL; | |
483 | ||
484 | if (prop == ZPOOL_PROP_BOOTFS) { | |
34dc7c2f BB |
485 | dsl_dataset_t *ds = NULL; |
486 | ||
619f0976 GW |
487 | err = dsl_dataset_hold_obj(dp, |
488 | za.za_first_integer, FTAG, &ds); | |
1743c737 | 489 | if (err != 0) |
34dc7c2f | 490 | break; |
34dc7c2f | 491 | |
eca7b760 | 492 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, |
79c76d5b | 493 | KM_SLEEP); |
34dc7c2f | 494 | dsl_dataset_name(ds, strval); |
b128c09f | 495 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
496 | } else { |
497 | strval = NULL; | |
498 | intval = za.za_first_integer; | |
499 | } | |
500 | ||
501 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
502 | ||
503 | if (strval != NULL) | |
eca7b760 | 504 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
505 | |
506 | break; | |
507 | ||
508 | case 1: | |
509 | /* string property */ | |
79c76d5b | 510 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); |
34dc7c2f BB |
511 | err = zap_lookup(mos, spa->spa_pool_props_object, |
512 | za.za_name, 1, za.za_num_integers, strval); | |
513 | if (err) { | |
514 | kmem_free(strval, za.za_num_integers); | |
515 | break; | |
516 | } | |
517 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
518 | kmem_free(strval, za.za_num_integers); | |
519 | break; | |
520 | ||
521 | default: | |
522 | break; | |
523 | } | |
524 | } | |
525 | zap_cursor_fini(&zc); | |
34dc7c2f | 526 | out: |
1743c737 AM |
527 | mutex_exit(&spa->spa_props_lock); |
528 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
529 | if (err && err != ENOENT) { |
530 | nvlist_free(*nvp); | |
531 | *nvp = NULL; | |
532 | return (err); | |
533 | } | |
534 | ||
535 | return (0); | |
536 | } | |
537 | ||
538 | /* | |
539 | * Validate the given pool properties nvlist and modify the list | |
540 | * for the property values to be set. | |
541 | */ | |
542 | static int | |
543 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
544 | { | |
545 | nvpair_t *elem; | |
546 | int error = 0, reset_bootfs = 0; | |
d4ed6673 | 547 | uint64_t objnum = 0; |
9ae529ec | 548 | boolean_t has_feature = B_FALSE; |
34dc7c2f BB |
549 | |
550 | elem = NULL; | |
551 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
34dc7c2f | 552 | uint64_t intval; |
9ae529ec CS |
553 | char *strval, *slash, *check, *fname; |
554 | const char *propname = nvpair_name(elem); | |
555 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
556 | ||
31864e3d BB |
557 | switch (prop) { |
558 | case ZPOOL_PROP_INVAL: | |
9ae529ec | 559 | if (!zpool_prop_feature(propname)) { |
2e528b49 | 560 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
561 | break; |
562 | } | |
563 | ||
564 | /* | |
565 | * Sanitize the input. | |
566 | */ | |
567 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
2e528b49 | 568 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
569 | break; |
570 | } | |
571 | ||
572 | if (nvpair_value_uint64(elem, &intval) != 0) { | |
2e528b49 | 573 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
574 | break; |
575 | } | |
34dc7c2f | 576 | |
9ae529ec | 577 | if (intval != 0) { |
2e528b49 | 578 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
579 | break; |
580 | } | |
34dc7c2f | 581 | |
9ae529ec CS |
582 | fname = strchr(propname, '@') + 1; |
583 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
2e528b49 | 584 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
585 | break; |
586 | } | |
587 | ||
588 | has_feature = B_TRUE; | |
589 | break; | |
34dc7c2f | 590 | |
34dc7c2f BB |
591 | case ZPOOL_PROP_VERSION: |
592 | error = nvpair_value_uint64(elem, &intval); | |
593 | if (!error && | |
9ae529ec CS |
594 | (intval < spa_version(spa) || |
595 | intval > SPA_VERSION_BEFORE_FEATURES || | |
596 | has_feature)) | |
2e528b49 | 597 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
598 | break; |
599 | ||
600 | case ZPOOL_PROP_DELEGATION: | |
601 | case ZPOOL_PROP_AUTOREPLACE: | |
b128c09f | 602 | case ZPOOL_PROP_LISTSNAPS: |
9babb374 | 603 | case ZPOOL_PROP_AUTOEXPAND: |
1b939560 | 604 | case ZPOOL_PROP_AUTOTRIM: |
34dc7c2f BB |
605 | error = nvpair_value_uint64(elem, &intval); |
606 | if (!error && intval > 1) | |
2e528b49 | 607 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
608 | break; |
609 | ||
379ca9cf OF |
610 | case ZPOOL_PROP_MULTIHOST: |
611 | error = nvpair_value_uint64(elem, &intval); | |
612 | if (!error && intval > 1) | |
613 | error = SET_ERROR(EINVAL); | |
614 | ||
25f06d67 BB |
615 | if (!error) { |
616 | uint32_t hostid = zone_get_hostid(NULL); | |
617 | if (hostid) | |
618 | spa->spa_hostid = hostid; | |
619 | else | |
620 | error = SET_ERROR(ENOTSUP); | |
621 | } | |
379ca9cf OF |
622 | |
623 | break; | |
624 | ||
34dc7c2f | 625 | case ZPOOL_PROP_BOOTFS: |
9babb374 BB |
626 | /* |
627 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
628 | * or the pool is still being created (version == 0), | |
629 | * the bootfs property cannot be set. | |
630 | */ | |
34dc7c2f | 631 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { |
2e528b49 | 632 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
633 | break; |
634 | } | |
635 | ||
636 | /* | |
b128c09f | 637 | * Make sure the vdev config is bootable |
34dc7c2f | 638 | */ |
b128c09f | 639 | if (!vdev_is_bootable(spa->spa_root_vdev)) { |
2e528b49 | 640 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
641 | break; |
642 | } | |
643 | ||
644 | reset_bootfs = 1; | |
645 | ||
646 | error = nvpair_value_string(elem, &strval); | |
647 | ||
648 | if (!error) { | |
9ae529ec | 649 | objset_t *os; |
b128c09f | 650 | |
34dc7c2f BB |
651 | if (strval == NULL || strval[0] == '\0') { |
652 | objnum = zpool_prop_default_numeric( | |
653 | ZPOOL_PROP_BOOTFS); | |
654 | break; | |
655 | } | |
656 | ||
d1d7e268 | 657 | error = dmu_objset_hold(strval, FTAG, &os); |
619f0976 | 658 | if (error != 0) |
34dc7c2f | 659 | break; |
b128c09f | 660 | |
eaa25f1a | 661 | /* Must be ZPL. */ |
428870ff | 662 | if (dmu_objset_type(os) != DMU_OST_ZFS) { |
2e528b49 | 663 | error = SET_ERROR(ENOTSUP); |
b128c09f BB |
664 | } else { |
665 | objnum = dmu_objset_id(os); | |
666 | } | |
428870ff | 667 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
668 | } |
669 | break; | |
b128c09f | 670 | |
34dc7c2f BB |
671 | case ZPOOL_PROP_FAILUREMODE: |
672 | error = nvpair_value_uint64(elem, &intval); | |
3bfd95d5 | 673 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) |
2e528b49 | 674 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
675 | |
676 | /* | |
677 | * This is a special case which only occurs when | |
678 | * the pool has completely failed. This allows | |
679 | * the user to change the in-core failmode property | |
680 | * without syncing it out to disk (I/Os might | |
681 | * currently be blocked). We do this by returning | |
682 | * EIO to the caller (spa_prop_set) to trick it | |
683 | * into thinking we encountered a property validation | |
684 | * error. | |
685 | */ | |
b128c09f | 686 | if (!error && spa_suspended(spa)) { |
34dc7c2f | 687 | spa->spa_failmode = intval; |
2e528b49 | 688 | error = SET_ERROR(EIO); |
34dc7c2f BB |
689 | } |
690 | break; | |
691 | ||
692 | case ZPOOL_PROP_CACHEFILE: | |
693 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
694 | break; | |
695 | ||
696 | if (strval[0] == '\0') | |
697 | break; | |
698 | ||
699 | if (strcmp(strval, "none") == 0) | |
700 | break; | |
701 | ||
702 | if (strval[0] != '/') { | |
2e528b49 | 703 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
704 | break; |
705 | } | |
706 | ||
707 | slash = strrchr(strval, '/'); | |
708 | ASSERT(slash != NULL); | |
709 | ||
710 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
711 | strcmp(slash, "/..") == 0) | |
2e528b49 | 712 | error = SET_ERROR(EINVAL); |
34dc7c2f | 713 | break; |
428870ff | 714 | |
d96eb2b1 DM |
715 | case ZPOOL_PROP_COMMENT: |
716 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
717 | break; | |
718 | for (check = strval; *check != '\0'; check++) { | |
719 | if (!isprint(*check)) { | |
2e528b49 | 720 | error = SET_ERROR(EINVAL); |
d96eb2b1 DM |
721 | break; |
722 | } | |
d96eb2b1 DM |
723 | } |
724 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
2e528b49 | 725 | error = SET_ERROR(E2BIG); |
d96eb2b1 DM |
726 | break; |
727 | ||
e75c13c3 BB |
728 | default: |
729 | break; | |
34dc7c2f BB |
730 | } |
731 | ||
732 | if (error) | |
733 | break; | |
734 | } | |
735 | ||
050d720c MA |
736 | (void) nvlist_remove_all(props, |
737 | zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO)); | |
738 | ||
34dc7c2f BB |
739 | if (!error && reset_bootfs) { |
740 | error = nvlist_remove(props, | |
741 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
742 | ||
743 | if (!error) { | |
744 | error = nvlist_add_uint64(props, | |
745 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
746 | } | |
747 | } | |
748 | ||
749 | return (error); | |
750 | } | |
751 | ||
d164b209 BB |
752 | void |
753 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
754 | { | |
755 | char *cachefile; | |
756 | spa_config_dirent_t *dp; | |
757 | ||
758 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
759 | &cachefile) != 0) | |
760 | return; | |
761 | ||
762 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
79c76d5b | 763 | KM_SLEEP); |
d164b209 BB |
764 | |
765 | if (cachefile[0] == '\0') | |
766 | dp->scd_path = spa_strdup(spa_config_path); | |
767 | else if (strcmp(cachefile, "none") == 0) | |
768 | dp->scd_path = NULL; | |
769 | else | |
770 | dp->scd_path = spa_strdup(cachefile); | |
771 | ||
772 | list_insert_head(&spa->spa_config_list, dp); | |
773 | if (need_sync) | |
774 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
775 | } | |
776 | ||
34dc7c2f BB |
777 | int |
778 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
779 | { | |
780 | int error; | |
9ae529ec | 781 | nvpair_t *elem = NULL; |
d164b209 | 782 | boolean_t need_sync = B_FALSE; |
34dc7c2f BB |
783 | |
784 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
785 | return (error); | |
786 | ||
d164b209 | 787 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { |
9ae529ec | 788 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); |
d164b209 | 789 | |
572e2857 BB |
790 | if (prop == ZPOOL_PROP_CACHEFILE || |
791 | prop == ZPOOL_PROP_ALTROOT || | |
792 | prop == ZPOOL_PROP_READONLY) | |
d164b209 BB |
793 | continue; |
794 | ||
31864e3d | 795 | if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { |
2a673e76 | 796 | uint64_t ver = 0; |
9ae529ec CS |
797 | |
798 | if (prop == ZPOOL_PROP_VERSION) { | |
799 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
800 | } else { | |
801 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
802 | ver = SPA_VERSION_FEATURES; | |
803 | need_sync = B_TRUE; | |
804 | } | |
805 | ||
806 | /* Save time if the version is already set. */ | |
807 | if (ver == spa_version(spa)) | |
808 | continue; | |
809 | ||
810 | /* | |
811 | * In addition to the pool directory object, we might | |
812 | * create the pool properties object, the features for | |
813 | * read object, the features for write object, or the | |
814 | * feature descriptions object. | |
815 | */ | |
13fe0198 | 816 | error = dsl_sync_task(spa->spa_name, NULL, |
3d45fdd6 MA |
817 | spa_sync_version, &ver, |
818 | 6, ZFS_SPACE_CHECK_RESERVED); | |
9ae529ec CS |
819 | if (error) |
820 | return (error); | |
821 | continue; | |
822 | } | |
823 | ||
d164b209 BB |
824 | need_sync = B_TRUE; |
825 | break; | |
826 | } | |
827 | ||
9ae529ec | 828 | if (need_sync) { |
13fe0198 | 829 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, |
3d45fdd6 | 830 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); |
9ae529ec CS |
831 | } |
832 | ||
833 | return (0); | |
34dc7c2f BB |
834 | } |
835 | ||
836 | /* | |
837 | * If the bootfs property value is dsobj, clear it. | |
838 | */ | |
839 | void | |
840 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
841 | { | |
842 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
843 | VERIFY(zap_remove(spa->spa_meta_objset, | |
844 | spa->spa_pool_props_object, | |
845 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
846 | spa->spa_bootfs = 0; | |
847 | } | |
848 | } | |
849 | ||
3bc7e0fb | 850 | static int |
13fe0198 | 851 | spa_change_guid_check(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 852 | { |
2a8ba608 | 853 | uint64_t *newguid __maybe_unused = arg; |
13fe0198 | 854 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; |
3bc7e0fb GW |
855 | vdev_t *rvd = spa->spa_root_vdev; |
856 | uint64_t vdev_state; | |
3bc7e0fb | 857 | |
d2734cce SD |
858 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { |
859 | int error = (spa_has_checkpoint(spa)) ? | |
860 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
861 | return (SET_ERROR(error)); | |
862 | } | |
863 | ||
3bc7e0fb GW |
864 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
865 | vdev_state = rvd->vdev_state; | |
866 | spa_config_exit(spa, SCL_STATE, FTAG); | |
867 | ||
868 | if (vdev_state != VDEV_STATE_HEALTHY) | |
2e528b49 | 869 | return (SET_ERROR(ENXIO)); |
3bc7e0fb GW |
870 | |
871 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
872 | ||
873 | return (0); | |
874 | } | |
875 | ||
876 | static void | |
13fe0198 | 877 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 878 | { |
13fe0198 MA |
879 | uint64_t *newguid = arg; |
880 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
3bc7e0fb GW |
881 | uint64_t oldguid; |
882 | vdev_t *rvd = spa->spa_root_vdev; | |
883 | ||
884 | oldguid = spa_guid(spa); | |
885 | ||
886 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
887 | rvd->vdev_guid = *newguid; | |
888 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
889 | vdev_config_dirty(rvd); | |
890 | spa_config_exit(spa, SCL_STATE, FTAG); | |
891 | ||
6f1ffb06 | 892 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", |
74756182 | 893 | (u_longlong_t)oldguid, (u_longlong_t)*newguid); |
3bc7e0fb GW |
894 | } |
895 | ||
3541dc6d GA |
896 | /* |
897 | * Change the GUID for the pool. This is done so that we can later | |
898 | * re-import a pool built from a clone of our own vdevs. We will modify | |
899 | * the root vdev's guid, our own pool guid, and then mark all of our | |
900 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
901 | * online when we do this, or else any vdevs that weren't present | |
902 | * would be orphaned from our pool. We are also going to issue a | |
903 | * sysevent to update any watchers. | |
904 | */ | |
905 | int | |
906 | spa_change_guid(spa_t *spa) | |
907 | { | |
3bc7e0fb GW |
908 | int error; |
909 | uint64_t guid; | |
3541dc6d | 910 | |
621dd7bb | 911 | mutex_enter(&spa->spa_vdev_top_lock); |
3bc7e0fb GW |
912 | mutex_enter(&spa_namespace_lock); |
913 | guid = spa_generate_guid(NULL); | |
3541dc6d | 914 | |
13fe0198 | 915 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, |
3d45fdd6 | 916 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); |
3541dc6d | 917 | |
3bc7e0fb | 918 | if (error == 0) { |
55c12724 AH |
919 | /* |
920 | * Clear the kobj flag from all the vdevs to allow | |
921 | * vdev_cache_process_kobj_evt() to post events to all the | |
922 | * vdevs since GUID is updated. | |
923 | */ | |
924 | vdev_clear_kobj_evt(spa->spa_root_vdev); | |
925 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
926 | vdev_clear_kobj_evt(spa->spa_l2cache.sav_vdevs[i]); | |
927 | ||
928 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); | |
12fa0466 | 929 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); |
3bc7e0fb | 930 | } |
3541dc6d | 931 | |
3bc7e0fb | 932 | mutex_exit(&spa_namespace_lock); |
621dd7bb | 933 | mutex_exit(&spa->spa_vdev_top_lock); |
3541dc6d | 934 | |
3bc7e0fb | 935 | return (error); |
3541dc6d GA |
936 | } |
937 | ||
34dc7c2f BB |
938 | /* |
939 | * ========================================================================== | |
940 | * SPA state manipulation (open/create/destroy/import/export) | |
941 | * ========================================================================== | |
942 | */ | |
943 | ||
944 | static int | |
945 | spa_error_entry_compare(const void *a, const void *b) | |
946 | { | |
ee36c709 GN |
947 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; |
948 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
34dc7c2f BB |
949 | int ret; |
950 | ||
ee36c709 | 951 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, |
5dbd68a3 | 952 | sizeof (zbookmark_phys_t)); |
34dc7c2f | 953 | |
ca577779 | 954 | return (TREE_ISIGN(ret)); |
34dc7c2f BB |
955 | } |
956 | ||
957 | /* | |
958 | * Utility function which retrieves copies of the current logs and | |
959 | * re-initializes them in the process. | |
960 | */ | |
961 | void | |
962 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
963 | { | |
964 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
965 | ||
861166b0 AZ |
966 | memcpy(last, &spa->spa_errlist_last, sizeof (avl_tree_t)); |
967 | memcpy(scrub, &spa->spa_errlist_scrub, sizeof (avl_tree_t)); | |
34dc7c2f BB |
968 | |
969 | avl_create(&spa->spa_errlist_scrub, | |
970 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
971 | offsetof(spa_error_entry_t, se_avl)); | |
972 | avl_create(&spa->spa_errlist_last, | |
973 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
974 | offsetof(spa_error_entry_t, se_avl)); | |
975 | } | |
976 | ||
7ef5e54e AL |
977 | static void |
978 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
34dc7c2f | 979 | { |
7ef5e54e AL |
980 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; |
981 | enum zti_modes mode = ztip->zti_mode; | |
982 | uint_t value = ztip->zti_value; | |
983 | uint_t count = ztip->zti_count; | |
984 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7457b024 | 985 | uint_t cpus, flags = TASKQ_DYNAMIC; |
428870ff | 986 | boolean_t batch = B_FALSE; |
34dc7c2f | 987 | |
e8b96c60 MA |
988 | switch (mode) { |
989 | case ZTI_MODE_FIXED: | |
7457b024 | 990 | ASSERT3U(value, >, 0); |
e8b96c60 | 991 | break; |
7ef5e54e | 992 | |
e8b96c60 MA |
993 | case ZTI_MODE_BATCH: |
994 | batch = B_TRUE; | |
995 | flags |= TASKQ_THREADS_CPU_PCT; | |
dcb6bed1 | 996 | value = MIN(zio_taskq_batch_pct, 100); |
e8b96c60 | 997 | break; |
7ef5e54e | 998 | |
7457b024 AM |
999 | case ZTI_MODE_SCALE: |
1000 | flags |= TASKQ_THREADS_CPU_PCT; | |
1001 | /* | |
1002 | * We want more taskqs to reduce lock contention, but we want | |
1003 | * less for better request ordering and CPU utilization. | |
1004 | */ | |
1005 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); | |
1006 | if (zio_taskq_batch_tpq > 0) { | |
1007 | count = MAX(1, (cpus + zio_taskq_batch_tpq / 2) / | |
1008 | zio_taskq_batch_tpq); | |
1009 | } else { | |
1010 | /* | |
1011 | * Prefer 6 threads per taskq, but no more taskqs | |
1012 | * than threads in them on large systems. For 80%: | |
1013 | * | |
1014 | * taskq taskq total | |
1015 | * cpus taskqs percent threads threads | |
1016 | * ------- ------- ------- ------- ------- | |
1017 | * 1 1 80% 1 1 | |
1018 | * 2 1 80% 1 1 | |
1019 | * 4 1 80% 3 3 | |
1020 | * 8 2 40% 3 6 | |
1021 | * 16 3 27% 4 12 | |
1022 | * 32 5 16% 5 25 | |
1023 | * 64 7 11% 7 49 | |
1024 | * 128 10 8% 10 100 | |
1025 | * 256 14 6% 15 210 | |
1026 | */ | |
1027 | count = 1 + cpus / 6; | |
1028 | while (count * count > cpus) | |
1029 | count--; | |
1030 | } | |
1031 | /* Limit each taskq within 100% to not trigger assertion. */ | |
1032 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1033 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1034 | break; | |
1035 | ||
1036 | case ZTI_MODE_NULL: | |
1037 | tqs->stqs_count = 0; | |
1038 | tqs->stqs_taskq = NULL; | |
1039 | return; | |
1040 | ||
e8b96c60 MA |
1041 | default: |
1042 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
1043 | "spa_activate()", | |
1044 | zio_type_name[t], zio_taskq_types[q], mode, value); | |
1045 | break; | |
1046 | } | |
7ef5e54e | 1047 | |
7457b024 AM |
1048 | ASSERT3U(count, >, 0); |
1049 | tqs->stqs_count = count; | |
1050 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
1051 | ||
1c27024e | 1052 | for (uint_t i = 0; i < count; i++) { |
e8b96c60 | 1053 | taskq_t *tq; |
af430294 | 1054 | char name[32]; |
7ef5e54e | 1055 | |
7457b024 AM |
1056 | if (count > 1) |
1057 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
1058 | zio_type_name[t], zio_taskq_types[q], i); | |
1059 | else | |
1060 | (void) snprintf(name, sizeof (name), "%s_%s", | |
1061 | zio_type_name[t], zio_taskq_types[q]); | |
7ef5e54e AL |
1062 | |
1063 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { | |
1064 | if (batch) | |
1065 | flags |= TASKQ_DC_BATCH; | |
1066 | ||
18168da7 | 1067 | (void) zio_taskq_basedc; |
7ef5e54e AL |
1068 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, |
1069 | spa->spa_proc, zio_taskq_basedc, flags); | |
1070 | } else { | |
e8b96c60 MA |
1071 | pri_t pri = maxclsyspri; |
1072 | /* | |
1073 | * The write issue taskq can be extremely CPU | |
1229323d | 1074 | * intensive. Run it at slightly less important |
7432d297 MM |
1075 | * priority than the other taskqs. |
1076 | * | |
1077 | * Under Linux and FreeBSD this means incrementing | |
1078 | * the priority value as opposed to platforms like | |
1079 | * illumos where it should be decremented. | |
1080 | * | |
1081 | * On FreeBSD, if priorities divided by four (RQ_PPQ) | |
1082 | * are equal then a difference between them is | |
1083 | * insignificant. | |
e8b96c60 | 1084 | */ |
7432d297 MM |
1085 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) { |
1086 | #if defined(__linux__) | |
1229323d | 1087 | pri++; |
7432d297 MM |
1088 | #elif defined(__FreeBSD__) |
1089 | pri += 4; | |
1090 | #else | |
1091 | #error "unknown OS" | |
1092 | #endif | |
1093 | } | |
e8b96c60 | 1094 | tq = taskq_create_proc(name, value, pri, 50, |
7ef5e54e AL |
1095 | INT_MAX, spa->spa_proc, flags); |
1096 | } | |
1097 | ||
1098 | tqs->stqs_taskq[i] = tq; | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | static void | |
1103 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
1104 | { | |
1105 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7ef5e54e AL |
1106 | |
1107 | if (tqs->stqs_taskq == NULL) { | |
1108 | ASSERT3U(tqs->stqs_count, ==, 0); | |
1109 | return; | |
1110 | } | |
1111 | ||
1c27024e | 1112 | for (uint_t i = 0; i < tqs->stqs_count; i++) { |
7ef5e54e AL |
1113 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); |
1114 | taskq_destroy(tqs->stqs_taskq[i]); | |
428870ff | 1115 | } |
34dc7c2f | 1116 | |
7ef5e54e AL |
1117 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); |
1118 | tqs->stqs_taskq = NULL; | |
1119 | } | |
34dc7c2f | 1120 | |
7ef5e54e AL |
1121 | /* |
1122 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
1123 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
1124 | * on the taskq itself. In that case we choose which taskq at random by using | |
1125 | * the low bits of gethrtime(). | |
1126 | */ | |
1127 | void | |
1128 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1129 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) | |
1130 | { | |
1131 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1132 | taskq_t *tq; | |
1133 | ||
1134 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1135 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1136 | ||
1137 | if (tqs->stqs_count == 1) { | |
1138 | tq = tqs->stqs_taskq[0]; | |
1139 | } else { | |
c12936b1 | 1140 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
428870ff | 1141 | } |
7ef5e54e AL |
1142 | |
1143 | taskq_dispatch_ent(tq, func, arg, flags, ent); | |
428870ff BB |
1144 | } |
1145 | ||
044baf00 BB |
1146 | /* |
1147 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
1148 | */ | |
1149 | void | |
1150 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1151 | task_func_t *func, void *arg, uint_t flags) | |
1152 | { | |
1153 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1154 | taskq_t *tq; | |
1155 | taskqid_t id; | |
1156 | ||
1157 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1158 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1159 | ||
1160 | if (tqs->stqs_count == 1) { | |
1161 | tq = tqs->stqs_taskq[0]; | |
1162 | } else { | |
c12936b1 | 1163 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
044baf00 BB |
1164 | } |
1165 | ||
1166 | id = taskq_dispatch(tq, func, arg, flags); | |
1167 | if (id) | |
1168 | taskq_wait_id(tq, id); | |
1169 | } | |
1170 | ||
428870ff BB |
1171 | static void |
1172 | spa_create_zio_taskqs(spa_t *spa) | |
1173 | { | |
1c27024e DB |
1174 | for (int t = 0; t < ZIO_TYPES; t++) { |
1175 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1176 | spa_taskqs_init(spa, t, q); |
428870ff BB |
1177 | } |
1178 | } | |
1179 | } | |
9babb374 | 1180 | |
c25b8f99 BB |
1181 | /* |
1182 | * Disabled until spa_thread() can be adapted for Linux. | |
1183 | */ | |
1184 | #undef HAVE_SPA_THREAD | |
1185 | ||
7b89a549 | 1186 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) |
428870ff BB |
1187 | static void |
1188 | spa_thread(void *arg) | |
1189 | { | |
93ce2b4c | 1190 | psetid_t zio_taskq_psrset_bind = PS_NONE; |
428870ff | 1191 | callb_cpr_t cprinfo; |
9babb374 | 1192 | |
428870ff BB |
1193 | spa_t *spa = arg; |
1194 | user_t *pu = PTOU(curproc); | |
9babb374 | 1195 | |
428870ff BB |
1196 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, |
1197 | spa->spa_name); | |
9babb374 | 1198 | |
428870ff BB |
1199 | ASSERT(curproc != &p0); |
1200 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1201 | "zpool-%s", spa->spa_name); | |
1202 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1203 | ||
1204 | /* bind this thread to the requested psrset */ | |
1205 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1206 | pool_lock(); | |
1207 | mutex_enter(&cpu_lock); | |
1208 | mutex_enter(&pidlock); | |
1209 | mutex_enter(&curproc->p_lock); | |
1210 | ||
1211 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1212 | 0, NULL, NULL) == 0) { | |
1213 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1214 | } else { | |
1215 | cmn_err(CE_WARN, | |
1216 | "Couldn't bind process for zfs pool \"%s\" to " | |
1217 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1218 | } | |
1219 | ||
1220 | mutex_exit(&curproc->p_lock); | |
1221 | mutex_exit(&pidlock); | |
1222 | mutex_exit(&cpu_lock); | |
1223 | pool_unlock(); | |
1224 | } | |
1225 | ||
1226 | if (zio_taskq_sysdc) { | |
1227 | sysdc_thread_enter(curthread, 100, 0); | |
1228 | } | |
1229 | ||
1230 | spa->spa_proc = curproc; | |
1231 | spa->spa_did = curthread->t_did; | |
1232 | ||
1233 | spa_create_zio_taskqs(spa); | |
1234 | ||
1235 | mutex_enter(&spa->spa_proc_lock); | |
1236 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1237 | ||
1238 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1239 | cv_broadcast(&spa->spa_proc_cv); | |
1240 | ||
1241 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1242 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1243 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1244 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1245 | ||
1246 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1247 | spa->spa_proc_state = SPA_PROC_GONE; | |
1248 | spa->spa_proc = &p0; | |
1249 | cv_broadcast(&spa->spa_proc_cv); | |
1250 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1251 | ||
1252 | mutex_enter(&curproc->p_lock); | |
1253 | lwp_exit(); | |
1254 | } | |
1255 | #endif | |
1256 | ||
1257 | /* | |
1258 | * Activate an uninitialized pool. | |
1259 | */ | |
1260 | static void | |
da92d5cb | 1261 | spa_activate(spa_t *spa, spa_mode_t mode) |
428870ff BB |
1262 | { |
1263 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
1264 | ||
1265 | spa->spa_state = POOL_STATE_ACTIVE; | |
1266 | spa->spa_mode = mode; | |
e39fe05b | 1267 | spa->spa_read_spacemaps = spa_mode_readable_spacemaps; |
428870ff | 1268 | |
18168da7 AZ |
1269 | spa->spa_normal_class = metaslab_class_create(spa, &zfs_metaslab_ops); |
1270 | spa->spa_log_class = metaslab_class_create(spa, &zfs_metaslab_ops); | |
aa755b35 | 1271 | spa->spa_embedded_log_class = |
18168da7 AZ |
1272 | metaslab_class_create(spa, &zfs_metaslab_ops); |
1273 | spa->spa_special_class = metaslab_class_create(spa, &zfs_metaslab_ops); | |
1274 | spa->spa_dedup_class = metaslab_class_create(spa, &zfs_metaslab_ops); | |
428870ff BB |
1275 | |
1276 | /* Try to create a covering process */ | |
1277 | mutex_enter(&spa->spa_proc_lock); | |
1278 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1279 | ASSERT(spa->spa_proc == &p0); | |
1280 | spa->spa_did = 0; | |
1281 | ||
18168da7 | 1282 | (void) spa_create_process; |
7b89a549 | 1283 | #ifdef HAVE_SPA_THREAD |
428870ff BB |
1284 | /* Only create a process if we're going to be around a while. */ |
1285 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1286 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1287 | NULL, 0) == 0) { | |
1288 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1289 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1290 | cv_wait(&spa->spa_proc_cv, | |
1291 | &spa->spa_proc_lock); | |
9babb374 | 1292 | } |
428870ff BB |
1293 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); |
1294 | ASSERT(spa->spa_proc != &p0); | |
1295 | ASSERT(spa->spa_did != 0); | |
1296 | } else { | |
1297 | #ifdef _KERNEL | |
1298 | cmn_err(CE_WARN, | |
1299 | "Couldn't create process for zfs pool \"%s\"\n", | |
1300 | spa->spa_name); | |
1301 | #endif | |
b128c09f | 1302 | } |
34dc7c2f | 1303 | } |
7b89a549 | 1304 | #endif /* HAVE_SPA_THREAD */ |
428870ff BB |
1305 | mutex_exit(&spa->spa_proc_lock); |
1306 | ||
1307 | /* If we didn't create a process, we need to create our taskqs. */ | |
1308 | if (spa->spa_proc == &p0) { | |
1309 | spa_create_zio_taskqs(spa); | |
1310 | } | |
34dc7c2f | 1311 | |
619f0976 GW |
1312 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1313 | spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, | |
1314 | ZIO_FLAG_CANFAIL); | |
1315 | } | |
a1d477c2 | 1316 | |
b128c09f BB |
1317 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), |
1318 | offsetof(vdev_t, vdev_config_dirty_node)); | |
0c66c32d JG |
1319 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), |
1320 | offsetof(objset_t, os_evicting_node)); | |
b128c09f BB |
1321 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), |
1322 | offsetof(vdev_t, vdev_state_dirty_node)); | |
34dc7c2f | 1323 | |
4747a7d3 | 1324 | txg_list_create(&spa->spa_vdev_txg_list, spa, |
34dc7c2f BB |
1325 | offsetof(struct vdev, vdev_txg_node)); |
1326 | ||
1327 | avl_create(&spa->spa_errlist_scrub, | |
1328 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1329 | offsetof(spa_error_entry_t, se_avl)); | |
1330 | avl_create(&spa->spa_errlist_last, | |
1331 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1332 | offsetof(spa_error_entry_t, se_avl)); | |
e8cf3a4f AP |
1333 | avl_create(&spa->spa_errlist_healed, |
1334 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1335 | offsetof(spa_error_entry_t, se_avl)); | |
a0bd735a | 1336 | |
4759342a JL |
1337 | spa_activate_os(spa); |
1338 | ||
b5256303 TC |
1339 | spa_keystore_init(&spa->spa_keystore); |
1340 | ||
a0bd735a BP |
1341 | /* |
1342 | * This taskq is used to perform zvol-minor-related tasks | |
1343 | * asynchronously. This has several advantages, including easy | |
d0249a4b | 1344 | * resolution of various deadlocks. |
a0bd735a BP |
1345 | * |
1346 | * The taskq must be single threaded to ensure tasks are always | |
1347 | * processed in the order in which they were dispatched. | |
1348 | * | |
1349 | * A taskq per pool allows one to keep the pools independent. | |
1350 | * This way if one pool is suspended, it will not impact another. | |
1351 | * | |
1352 | * The preferred location to dispatch a zvol minor task is a sync | |
1353 | * task. In this context, there is easy access to the spa_t and minimal | |
1354 | * error handling is required because the sync task must succeed. | |
1355 | */ | |
1356 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1357 | 1, INT_MAX, 0); | |
1de321e6 | 1358 | |
77d8a0f1 | 1359 | /* |
1360 | * Taskq dedicated to prefetcher threads: this is used to prevent the | |
1361 | * pool traverse code from monopolizing the global (and limited) | |
1362 | * system_taskq by inappropriately scheduling long running tasks on it. | |
1363 | */ | |
60a4c7d2 PD |
1364 | spa->spa_prefetch_taskq = taskq_create("z_prefetch", 100, |
1365 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
77d8a0f1 | 1366 | |
1de321e6 JX |
1367 | /* |
1368 | * The taskq to upgrade datasets in this pool. Currently used by | |
9c5167d1 | 1369 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA. |
1de321e6 | 1370 | */ |
60a4c7d2 PD |
1371 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", 100, |
1372 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
34dc7c2f BB |
1373 | } |
1374 | ||
1375 | /* | |
1376 | * Opposite of spa_activate(). | |
1377 | */ | |
1378 | static void | |
1379 | spa_deactivate(spa_t *spa) | |
1380 | { | |
34dc7c2f BB |
1381 | ASSERT(spa->spa_sync_on == B_FALSE); |
1382 | ASSERT(spa->spa_dsl_pool == NULL); | |
1383 | ASSERT(spa->spa_root_vdev == NULL); | |
9babb374 | 1384 | ASSERT(spa->spa_async_zio_root == NULL); |
34dc7c2f BB |
1385 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); |
1386 | ||
0c66c32d JG |
1387 | spa_evicting_os_wait(spa); |
1388 | ||
a0bd735a BP |
1389 | if (spa->spa_zvol_taskq) { |
1390 | taskq_destroy(spa->spa_zvol_taskq); | |
1391 | spa->spa_zvol_taskq = NULL; | |
1392 | } | |
1393 | ||
77d8a0f1 | 1394 | if (spa->spa_prefetch_taskq) { |
1395 | taskq_destroy(spa->spa_prefetch_taskq); | |
1396 | spa->spa_prefetch_taskq = NULL; | |
1397 | } | |
1398 | ||
1de321e6 JX |
1399 | if (spa->spa_upgrade_taskq) { |
1400 | taskq_destroy(spa->spa_upgrade_taskq); | |
1401 | spa->spa_upgrade_taskq = NULL; | |
1402 | } | |
1403 | ||
34dc7c2f BB |
1404 | txg_list_destroy(&spa->spa_vdev_txg_list); |
1405 | ||
b128c09f | 1406 | list_destroy(&spa->spa_config_dirty_list); |
0c66c32d | 1407 | list_destroy(&spa->spa_evicting_os_list); |
b128c09f | 1408 | list_destroy(&spa->spa_state_dirty_list); |
34dc7c2f | 1409 | |
57ddcda1 | 1410 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 | 1411 | |
1c27024e DB |
1412 | for (int t = 0; t < ZIO_TYPES; t++) { |
1413 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1414 | spa_taskqs_fini(spa, t, q); |
b128c09f | 1415 | } |
34dc7c2f BB |
1416 | } |
1417 | ||
a1d477c2 MA |
1418 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1419 | ASSERT3P(spa->spa_txg_zio[i], !=, NULL); | |
1420 | VERIFY0(zio_wait(spa->spa_txg_zio[i])); | |
1421 | spa->spa_txg_zio[i] = NULL; | |
1422 | } | |
1423 | ||
34dc7c2f BB |
1424 | metaslab_class_destroy(spa->spa_normal_class); |
1425 | spa->spa_normal_class = NULL; | |
1426 | ||
1427 | metaslab_class_destroy(spa->spa_log_class); | |
1428 | spa->spa_log_class = NULL; | |
1429 | ||
aa755b35 MA |
1430 | metaslab_class_destroy(spa->spa_embedded_log_class); |
1431 | spa->spa_embedded_log_class = NULL; | |
1432 | ||
cc99f275 DB |
1433 | metaslab_class_destroy(spa->spa_special_class); |
1434 | spa->spa_special_class = NULL; | |
1435 | ||
1436 | metaslab_class_destroy(spa->spa_dedup_class); | |
1437 | spa->spa_dedup_class = NULL; | |
1438 | ||
34dc7c2f BB |
1439 | /* |
1440 | * If this was part of an import or the open otherwise failed, we may | |
1441 | * still have errors left in the queues. Empty them just in case. | |
1442 | */ | |
1443 | spa_errlog_drain(spa); | |
34dc7c2f BB |
1444 | avl_destroy(&spa->spa_errlist_scrub); |
1445 | avl_destroy(&spa->spa_errlist_last); | |
e8cf3a4f | 1446 | avl_destroy(&spa->spa_errlist_healed); |
34dc7c2f | 1447 | |
b5256303 TC |
1448 | spa_keystore_fini(&spa->spa_keystore); |
1449 | ||
34dc7c2f | 1450 | spa->spa_state = POOL_STATE_UNINITIALIZED; |
428870ff BB |
1451 | |
1452 | mutex_enter(&spa->spa_proc_lock); | |
1453 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1454 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1455 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1456 | cv_broadcast(&spa->spa_proc_cv); | |
1457 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1458 | ASSERT(spa->spa_proc != &p0); | |
1459 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1460 | } | |
1461 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1462 | spa->spa_proc_state = SPA_PROC_NONE; | |
1463 | } | |
1464 | ASSERT(spa->spa_proc == &p0); | |
1465 | mutex_exit(&spa->spa_proc_lock); | |
1466 | ||
1467 | /* | |
1468 | * We want to make sure spa_thread() has actually exited the ZFS | |
1469 | * module, so that the module can't be unloaded out from underneath | |
1470 | * it. | |
1471 | */ | |
1472 | if (spa->spa_did != 0) { | |
1473 | thread_join(spa->spa_did); | |
1474 | spa->spa_did = 0; | |
1475 | } | |
4759342a JL |
1476 | |
1477 | spa_deactivate_os(spa); | |
1478 | ||
34dc7c2f BB |
1479 | } |
1480 | ||
1481 | /* | |
1482 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1483 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1484 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1485 | * All vdev validation is done by the vdev_alloc() routine. | |
1486 | */ | |
4a22ba5b | 1487 | int |
34dc7c2f BB |
1488 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, |
1489 | uint_t id, int atype) | |
1490 | { | |
1491 | nvlist_t **child; | |
9babb374 | 1492 | uint_t children; |
34dc7c2f BB |
1493 | int error; |
1494 | ||
1495 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1496 | return (error); | |
1497 | ||
1498 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1499 | return (0); | |
1500 | ||
b128c09f BB |
1501 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
1502 | &child, &children); | |
1503 | ||
1504 | if (error == ENOENT) | |
1505 | return (0); | |
1506 | ||
1507 | if (error) { | |
34dc7c2f BB |
1508 | vdev_free(*vdp); |
1509 | *vdp = NULL; | |
2e528b49 | 1510 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
1511 | } |
1512 | ||
1c27024e | 1513 | for (int c = 0; c < children; c++) { |
34dc7c2f BB |
1514 | vdev_t *vd; |
1515 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1516 | atype)) != 0) { | |
1517 | vdev_free(*vdp); | |
1518 | *vdp = NULL; | |
1519 | return (error); | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | ASSERT(*vdp != NULL); | |
1524 | ||
1525 | return (0); | |
1526 | } | |
1527 | ||
93e28d66 SD |
1528 | static boolean_t |
1529 | spa_should_flush_logs_on_unload(spa_t *spa) | |
1530 | { | |
1531 | if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) | |
1532 | return (B_FALSE); | |
1533 | ||
1534 | if (!spa_writeable(spa)) | |
1535 | return (B_FALSE); | |
1536 | ||
1537 | if (!spa->spa_sync_on) | |
1538 | return (B_FALSE); | |
1539 | ||
1540 | if (spa_state(spa) != POOL_STATE_EXPORTED) | |
1541 | return (B_FALSE); | |
1542 | ||
1543 | if (zfs_keep_log_spacemaps_at_export) | |
1544 | return (B_FALSE); | |
1545 | ||
1546 | return (B_TRUE); | |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * Opens a transaction that will set the flag that will instruct | |
1551 | * spa_sync to attempt to flush all the metaslabs for that txg. | |
1552 | */ | |
1553 | static void | |
1554 | spa_unload_log_sm_flush_all(spa_t *spa) | |
1555 | { | |
1556 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
1557 | VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); | |
1558 | ||
1559 | ASSERT3U(spa->spa_log_flushall_txg, ==, 0); | |
1560 | spa->spa_log_flushall_txg = dmu_tx_get_txg(tx); | |
1561 | ||
1562 | dmu_tx_commit(tx); | |
1563 | txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg); | |
1564 | } | |
1565 | ||
1566 | static void | |
1567 | spa_unload_log_sm_metadata(spa_t *spa) | |
1568 | { | |
1569 | void *cookie = NULL; | |
1570 | spa_log_sm_t *sls; | |
1571 | while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg, | |
1572 | &cookie)) != NULL) { | |
1573 | VERIFY0(sls->sls_mscount); | |
1574 | kmem_free(sls, sizeof (spa_log_sm_t)); | |
1575 | } | |
1576 | ||
1577 | for (log_summary_entry_t *e = list_head(&spa->spa_log_summary); | |
1578 | e != NULL; e = list_head(&spa->spa_log_summary)) { | |
1579 | VERIFY0(e->lse_mscount); | |
1580 | list_remove(&spa->spa_log_summary, e); | |
1581 | kmem_free(e, sizeof (log_summary_entry_t)); | |
1582 | } | |
1583 | ||
1584 | spa->spa_unflushed_stats.sus_nblocks = 0; | |
1585 | spa->spa_unflushed_stats.sus_memused = 0; | |
1586 | spa->spa_unflushed_stats.sus_blocklimit = 0; | |
1587 | } | |
1588 | ||
37f03da8 SH |
1589 | static void |
1590 | spa_destroy_aux_threads(spa_t *spa) | |
1591 | { | |
1592 | if (spa->spa_condense_zthr != NULL) { | |
1593 | zthr_destroy(spa->spa_condense_zthr); | |
1594 | spa->spa_condense_zthr = NULL; | |
1595 | } | |
1596 | if (spa->spa_checkpoint_discard_zthr != NULL) { | |
1597 | zthr_destroy(spa->spa_checkpoint_discard_zthr); | |
1598 | spa->spa_checkpoint_discard_zthr = NULL; | |
1599 | } | |
1600 | if (spa->spa_livelist_delete_zthr != NULL) { | |
1601 | zthr_destroy(spa->spa_livelist_delete_zthr); | |
1602 | spa->spa_livelist_delete_zthr = NULL; | |
1603 | } | |
1604 | if (spa->spa_livelist_condense_zthr != NULL) { | |
1605 | zthr_destroy(spa->spa_livelist_condense_zthr); | |
1606 | spa->spa_livelist_condense_zthr = NULL; | |
1607 | } | |
1608 | } | |
1609 | ||
34dc7c2f BB |
1610 | /* |
1611 | * Opposite of spa_load(). | |
1612 | */ | |
1613 | static void | |
1614 | spa_unload(spa_t *spa) | |
1615 | { | |
b128c09f | 1616 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
93e28d66 | 1617 | ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED); |
b128c09f | 1618 | |
ca95f70d | 1619 | spa_import_progress_remove(spa_guid(spa)); |
4a0ee12a PZ |
1620 | spa_load_note(spa, "UNLOADING"); |
1621 | ||
e60e158e JG |
1622 | spa_wake_waiters(spa); |
1623 | ||
93e28d66 | 1624 | /* |
2fb52853 GA |
1625 | * If we have set the spa_final_txg, we have already performed the |
1626 | * tasks below in spa_export_common(). We should not redo it here since | |
1627 | * we delay the final TXGs beyond what spa_final_txg is set at. | |
93e28d66 | 1628 | */ |
2fb52853 GA |
1629 | if (spa->spa_final_txg == UINT64_MAX) { |
1630 | /* | |
1631 | * If the log space map feature is enabled and the pool is | |
1632 | * getting exported (but not destroyed), we want to spend some | |
1633 | * time flushing as many metaslabs as we can in an attempt to | |
1634 | * destroy log space maps and save import time. | |
1635 | */ | |
1636 | if (spa_should_flush_logs_on_unload(spa)) | |
1637 | spa_unload_log_sm_flush_all(spa); | |
93e28d66 | 1638 | |
2fb52853 GA |
1639 | /* |
1640 | * Stop async tasks. | |
1641 | */ | |
1642 | spa_async_suspend(spa); | |
34dc7c2f | 1643 | |
2fb52853 GA |
1644 | if (spa->spa_root_vdev) { |
1645 | vdev_t *root_vdev = spa->spa_root_vdev; | |
1646 | vdev_initialize_stop_all(root_vdev, | |
1647 | VDEV_INITIALIZE_ACTIVE); | |
1648 | vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE); | |
1649 | vdev_autotrim_stop_all(spa); | |
1650 | vdev_rebuild_stop_all(spa); | |
1651 | } | |
619f0976 GW |
1652 | } |
1653 | ||
34dc7c2f BB |
1654 | /* |
1655 | * Stop syncing. | |
1656 | */ | |
1657 | if (spa->spa_sync_on) { | |
1658 | txg_sync_stop(spa->spa_dsl_pool); | |
1659 | spa->spa_sync_on = B_FALSE; | |
1660 | } | |
1661 | ||
4e21fd06 | 1662 | /* |
93e28d66 SD |
1663 | * This ensures that there is no async metaslab prefetching |
1664 | * while we attempt to unload the spa. | |
4e21fd06 DB |
1665 | */ |
1666 | if (spa->spa_root_vdev != NULL) { | |
93e28d66 SD |
1667 | for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) { |
1668 | vdev_t *vc = spa->spa_root_vdev->vdev_child[c]; | |
1669 | if (vc->vdev_mg != NULL) | |
1670 | taskq_wait(vc->vdev_mg->mg_taskq); | |
1671 | } | |
4e21fd06 DB |
1672 | } |
1673 | ||
379ca9cf OF |
1674 | if (spa->spa_mmp.mmp_thread) |
1675 | mmp_thread_stop(spa); | |
1676 | ||
34dc7c2f | 1677 | /* |
b128c09f | 1678 | * Wait for any outstanding async I/O to complete. |
34dc7c2f | 1679 | */ |
9babb374 | 1680 | if (spa->spa_async_zio_root != NULL) { |
1c27024e | 1681 | for (int i = 0; i < max_ncpus; i++) |
e022864d MA |
1682 | (void) zio_wait(spa->spa_async_zio_root[i]); |
1683 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
9babb374 BB |
1684 | spa->spa_async_zio_root = NULL; |
1685 | } | |
34dc7c2f | 1686 | |
a1d477c2 MA |
1687 | if (spa->spa_vdev_removal != NULL) { |
1688 | spa_vdev_removal_destroy(spa->spa_vdev_removal); | |
1689 | spa->spa_vdev_removal = NULL; | |
1690 | } | |
1691 | ||
37f03da8 | 1692 | spa_destroy_aux_threads(spa); |
d2734cce | 1693 | |
a1d477c2 MA |
1694 | spa_condense_fini(spa); |
1695 | ||
428870ff BB |
1696 | bpobj_close(&spa->spa_deferred_bpobj); |
1697 | ||
619f0976 | 1698 | spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); |
93cf2076 GW |
1699 | |
1700 | /* | |
1701 | * Close all vdevs. | |
1702 | */ | |
1703 | if (spa->spa_root_vdev) | |
1704 | vdev_free(spa->spa_root_vdev); | |
1705 | ASSERT(spa->spa_root_vdev == NULL); | |
1706 | ||
34dc7c2f BB |
1707 | /* |
1708 | * Close the dsl pool. | |
1709 | */ | |
1710 | if (spa->spa_dsl_pool) { | |
1711 | dsl_pool_close(spa->spa_dsl_pool); | |
1712 | spa->spa_dsl_pool = NULL; | |
428870ff | 1713 | spa->spa_meta_objset = NULL; |
34dc7c2f BB |
1714 | } |
1715 | ||
428870ff | 1716 | ddt_unload(spa); |
67a1b037 | 1717 | brt_unload(spa); |
93e28d66 | 1718 | spa_unload_log_sm_metadata(spa); |
428870ff | 1719 | |
fb5f0bc8 BB |
1720 | /* |
1721 | * Drop and purge level 2 cache | |
1722 | */ | |
1723 | spa_l2cache_drop(spa); | |
1724 | ||
34dc7c2f | 1725 | if (spa->spa_spares.sav_vdevs) { |
cfb49616 RY |
1726 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
1727 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
34dc7c2f BB |
1728 | kmem_free(spa->spa_spares.sav_vdevs, |
1729 | spa->spa_spares.sav_count * sizeof (void *)); | |
1730 | spa->spa_spares.sav_vdevs = NULL; | |
1731 | } | |
1732 | if (spa->spa_spares.sav_config) { | |
1733 | nvlist_free(spa->spa_spares.sav_config); | |
1734 | spa->spa_spares.sav_config = NULL; | |
1735 | } | |
b128c09f | 1736 | spa->spa_spares.sav_count = 0; |
34dc7c2f | 1737 | |
34dc7c2f | 1738 | if (spa->spa_l2cache.sav_vdevs) { |
cfb49616 RY |
1739 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) { |
1740 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
1741 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
1742 | } | |
34dc7c2f BB |
1743 | kmem_free(spa->spa_l2cache.sav_vdevs, |
1744 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
1745 | spa->spa_l2cache.sav_vdevs = NULL; | |
1746 | } | |
1747 | if (spa->spa_l2cache.sav_config) { | |
1748 | nvlist_free(spa->spa_l2cache.sav_config); | |
1749 | spa->spa_l2cache.sav_config = NULL; | |
1750 | } | |
b128c09f | 1751 | spa->spa_l2cache.sav_count = 0; |
34dc7c2f BB |
1752 | |
1753 | spa->spa_async_suspended = 0; | |
fb5f0bc8 | 1754 | |
a1d477c2 MA |
1755 | spa->spa_indirect_vdevs_loaded = B_FALSE; |
1756 | ||
d96eb2b1 DM |
1757 | if (spa->spa_comment != NULL) { |
1758 | spa_strfree(spa->spa_comment); | |
1759 | spa->spa_comment = NULL; | |
1760 | } | |
658fb802 CB |
1761 | if (spa->spa_compatibility != NULL) { |
1762 | spa_strfree(spa->spa_compatibility); | |
1763 | spa->spa_compatibility = NULL; | |
1764 | } | |
d96eb2b1 | 1765 | |
619f0976 | 1766 | spa_config_exit(spa, SCL_ALL, spa); |
34dc7c2f BB |
1767 | } |
1768 | ||
1769 | /* | |
1770 | * Load (or re-load) the current list of vdevs describing the active spares for | |
1771 | * this pool. When this is called, we have some form of basic information in | |
1772 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
1773 | * then re-generate a more complete list including status information. | |
1774 | */ | |
a1d477c2 | 1775 | void |
34dc7c2f BB |
1776 | spa_load_spares(spa_t *spa) |
1777 | { | |
1778 | nvlist_t **spares; | |
1779 | uint_t nspares; | |
1780 | int i; | |
1781 | vdev_t *vd, *tvd; | |
1782 | ||
d2734cce SD |
1783 | #ifndef _KERNEL |
1784 | /* | |
1785 | * zdb opens both the current state of the pool and the | |
1786 | * checkpointed state (if present), with a different spa_t. | |
1787 | * | |
1788 | * As spare vdevs are shared among open pools, we skip loading | |
1789 | * them when we load the checkpointed state of the pool. | |
1790 | */ | |
1791 | if (!spa_writeable(spa)) | |
1792 | return; | |
1793 | #endif | |
1794 | ||
b128c09f BB |
1795 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
1796 | ||
34dc7c2f BB |
1797 | /* |
1798 | * First, close and free any existing spare vdevs. | |
1799 | */ | |
cfb49616 RY |
1800 | if (spa->spa_spares.sav_vdevs) { |
1801 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1802 | vd = spa->spa_spares.sav_vdevs[i]; | |
1803 | ||
1804 | /* Undo the call to spa_activate() below */ | |
1805 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1806 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
1807 | spa_spare_remove(tvd); | |
1808 | vdev_close(vd); | |
1809 | vdev_free(vd); | |
1810 | } | |
34dc7c2f | 1811 | |
34dc7c2f BB |
1812 | kmem_free(spa->spa_spares.sav_vdevs, |
1813 | spa->spa_spares.sav_count * sizeof (void *)); | |
cfb49616 | 1814 | } |
34dc7c2f BB |
1815 | |
1816 | if (spa->spa_spares.sav_config == NULL) | |
1817 | nspares = 0; | |
1818 | else | |
65ad5d11 AJ |
1819 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, |
1820 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f BB |
1821 | |
1822 | spa->spa_spares.sav_count = (int)nspares; | |
1823 | spa->spa_spares.sav_vdevs = NULL; | |
1824 | ||
1825 | if (nspares == 0) | |
1826 | return; | |
1827 | ||
1828 | /* | |
1829 | * Construct the array of vdevs, opening them to get status in the | |
1830 | * process. For each spare, there is potentially two different vdev_t | |
1831 | * structures associated with it: one in the list of spares (used only | |
1832 | * for basic validation purposes) and one in the active vdev | |
1833 | * configuration (if it's spared in). During this phase we open and | |
1834 | * validate each vdev on the spare list. If the vdev also exists in the | |
1835 | * active configuration, then we also mark this vdev as an active spare. | |
1836 | */ | |
904ea276 | 1837 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), |
79c76d5b | 1838 | KM_SLEEP); |
34dc7c2f BB |
1839 | for (i = 0; i < spa->spa_spares.sav_count; i++) { |
1840 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
1841 | VDEV_ALLOC_SPARE) == 0); | |
1842 | ASSERT(vd != NULL); | |
1843 | ||
1844 | spa->spa_spares.sav_vdevs[i] = vd; | |
1845 | ||
b128c09f BB |
1846 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, |
1847 | B_FALSE)) != NULL) { | |
34dc7c2f BB |
1848 | if (!tvd->vdev_isspare) |
1849 | spa_spare_add(tvd); | |
1850 | ||
1851 | /* | |
1852 | * We only mark the spare active if we were successfully | |
1853 | * able to load the vdev. Otherwise, importing a pool | |
1854 | * with a bad active spare would result in strange | |
1855 | * behavior, because multiple pool would think the spare | |
1856 | * is actively in use. | |
1857 | * | |
1858 | * There is a vulnerability here to an equally bizarre | |
1859 | * circumstance, where a dead active spare is later | |
1860 | * brought back to life (onlined or otherwise). Given | |
1861 | * the rarity of this scenario, and the extra complexity | |
1862 | * it adds, we ignore the possibility. | |
1863 | */ | |
1864 | if (!vdev_is_dead(tvd)) | |
1865 | spa_spare_activate(tvd); | |
1866 | } | |
1867 | ||
b128c09f | 1868 | vd->vdev_top = vd; |
9babb374 | 1869 | vd->vdev_aux = &spa->spa_spares; |
b128c09f | 1870 | |
34dc7c2f BB |
1871 | if (vdev_open(vd) != 0) |
1872 | continue; | |
1873 | ||
34dc7c2f BB |
1874 | if (vdev_validate_aux(vd) == 0) |
1875 | spa_spare_add(vd); | |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * Recompute the stashed list of spares, with status information | |
1880 | * this time. | |
1881 | */ | |
65ad5d11 | 1882 | fnvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES); |
34dc7c2f BB |
1883 | |
1884 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
79c76d5b | 1885 | KM_SLEEP); |
34dc7c2f BB |
1886 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1887 | spares[i] = vdev_config_generate(spa, | |
428870ff | 1888 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); |
65ad5d11 | 1889 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, |
795075e6 PD |
1890 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
1891 | spa->spa_spares.sav_count); | |
34dc7c2f BB |
1892 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1893 | nvlist_free(spares[i]); | |
1894 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
1899 | * this pool. When this is called, we have some form of basic information in | |
1900 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
1901 | * then re-generate a more complete list including status information. | |
1902 | * Devices which are already active have their details maintained, and are | |
1903 | * not re-opened. | |
1904 | */ | |
a1d477c2 | 1905 | void |
34dc7c2f BB |
1906 | spa_load_l2cache(spa_t *spa) |
1907 | { | |
460f239e | 1908 | nvlist_t **l2cache = NULL; |
34dc7c2f BB |
1909 | uint_t nl2cache; |
1910 | int i, j, oldnvdevs; | |
9babb374 | 1911 | uint64_t guid; |
a117a6d6 | 1912 | vdev_t *vd, **oldvdevs, **newvdevs; |
34dc7c2f BB |
1913 | spa_aux_vdev_t *sav = &spa->spa_l2cache; |
1914 | ||
d2734cce SD |
1915 | #ifndef _KERNEL |
1916 | /* | |
1917 | * zdb opens both the current state of the pool and the | |
1918 | * checkpointed state (if present), with a different spa_t. | |
1919 | * | |
1920 | * As L2 caches are part of the ARC which is shared among open | |
1921 | * pools, we skip loading them when we load the checkpointed | |
1922 | * state of the pool. | |
1923 | */ | |
1924 | if (!spa_writeable(spa)) | |
1925 | return; | |
1926 | #endif | |
1927 | ||
b128c09f BB |
1928 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
1929 | ||
34dc7c2f BB |
1930 | oldvdevs = sav->sav_vdevs; |
1931 | oldnvdevs = sav->sav_count; | |
1932 | sav->sav_vdevs = NULL; | |
1933 | sav->sav_count = 0; | |
1934 | ||
67d60824 NB |
1935 | if (sav->sav_config == NULL) { |
1936 | nl2cache = 0; | |
1937 | newvdevs = NULL; | |
1938 | goto out; | |
1939 | } | |
1940 | ||
65ad5d11 AJ |
1941 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, |
1942 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
67d60824 NB |
1943 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); |
1944 | ||
34dc7c2f BB |
1945 | /* |
1946 | * Process new nvlist of vdevs. | |
1947 | */ | |
1948 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 | 1949 | guid = fnvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID); |
34dc7c2f BB |
1950 | |
1951 | newvdevs[i] = NULL; | |
1952 | for (j = 0; j < oldnvdevs; j++) { | |
1953 | vd = oldvdevs[j]; | |
1954 | if (vd != NULL && guid == vd->vdev_guid) { | |
1955 | /* | |
1956 | * Retain previous vdev for add/remove ops. | |
1957 | */ | |
1958 | newvdevs[i] = vd; | |
1959 | oldvdevs[j] = NULL; | |
1960 | break; | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | if (newvdevs[i] == NULL) { | |
1965 | /* | |
1966 | * Create new vdev | |
1967 | */ | |
1968 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
1969 | VDEV_ALLOC_L2CACHE) == 0); | |
1970 | ASSERT(vd != NULL); | |
1971 | newvdevs[i] = vd; | |
1972 | ||
1973 | /* | |
1974 | * Commit this vdev as an l2cache device, | |
1975 | * even if it fails to open. | |
1976 | */ | |
1977 | spa_l2cache_add(vd); | |
1978 | ||
b128c09f BB |
1979 | vd->vdev_top = vd; |
1980 | vd->vdev_aux = sav; | |
1981 | ||
1982 | spa_l2cache_activate(vd); | |
1983 | ||
34dc7c2f BB |
1984 | if (vdev_open(vd) != 0) |
1985 | continue; | |
1986 | ||
34dc7c2f BB |
1987 | (void) vdev_validate_aux(vd); |
1988 | ||
9babb374 BB |
1989 | if (!vdev_is_dead(vd)) |
1990 | l2arc_add_vdev(spa, vd); | |
b7654bd7 GA |
1991 | |
1992 | /* | |
1993 | * Upon cache device addition to a pool or pool | |
1994 | * creation with a cache device or if the header | |
1995 | * of the device is invalid we issue an async | |
1996 | * TRIM command for the whole device which will | |
1997 | * execute if l2arc_trim_ahead > 0. | |
1998 | */ | |
1999 | spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); | |
34dc7c2f BB |
2000 | } |
2001 | } | |
2002 | ||
67d60824 NB |
2003 | sav->sav_vdevs = newvdevs; |
2004 | sav->sav_count = (int)nl2cache; | |
2005 | ||
2006 | /* | |
2007 | * Recompute the stashed list of l2cache devices, with status | |
2008 | * information this time. | |
2009 | */ | |
65ad5d11 | 2010 | fnvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE); |
67d60824 | 2011 | |
460f239e D |
2012 | if (sav->sav_count > 0) |
2013 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
2014 | KM_SLEEP); | |
67d60824 NB |
2015 | for (i = 0; i < sav->sav_count; i++) |
2016 | l2cache[i] = vdev_config_generate(spa, | |
2017 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
795075e6 PD |
2018 | fnvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, |
2019 | (const nvlist_t * const *)l2cache, sav->sav_count); | |
67d60824 NB |
2020 | |
2021 | out: | |
34dc7c2f BB |
2022 | /* |
2023 | * Purge vdevs that were dropped | |
2024 | */ | |
cfb49616 RY |
2025 | if (oldvdevs) { |
2026 | for (i = 0; i < oldnvdevs; i++) { | |
2027 | uint64_t pool; | |
2028 | ||
2029 | vd = oldvdevs[i]; | |
2030 | if (vd != NULL) { | |
2031 | ASSERT(vd->vdev_isl2cache); | |
2032 | ||
2033 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
2034 | pool != 0ULL && l2arc_vdev_present(vd)) | |
2035 | l2arc_remove_vdev(vd); | |
2036 | vdev_clear_stats(vd); | |
2037 | vdev_free(vd); | |
2038 | } | |
34dc7c2f | 2039 | } |
34dc7c2f | 2040 | |
34dc7c2f | 2041 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); |
cfb49616 | 2042 | } |
34dc7c2f | 2043 | |
34dc7c2f BB |
2044 | for (i = 0; i < sav->sav_count; i++) |
2045 | nvlist_free(l2cache[i]); | |
2046 | if (sav->sav_count) | |
2047 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
2048 | } | |
2049 | ||
2050 | static int | |
2051 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
2052 | { | |
2053 | dmu_buf_t *db; | |
2054 | char *packed = NULL; | |
2055 | size_t nvsize = 0; | |
2056 | int error; | |
2057 | *value = NULL; | |
2058 | ||
c3275b56 BB |
2059 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); |
2060 | if (error) | |
2061 | return (error); | |
2062 | ||
34dc7c2f BB |
2063 | nvsize = *(uint64_t *)db->db_data; |
2064 | dmu_buf_rele(db, FTAG); | |
2065 | ||
77aef6f6 | 2066 | packed = vmem_alloc(nvsize, KM_SLEEP); |
9babb374 BB |
2067 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, |
2068 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
2069 | if (error == 0) |
2070 | error = nvlist_unpack(packed, nvsize, value, 0); | |
77aef6f6 | 2071 | vmem_free(packed, nvsize); |
34dc7c2f BB |
2072 | |
2073 | return (error); | |
2074 | } | |
2075 | ||
6cb8e530 PZ |
2076 | /* |
2077 | * Concrete top-level vdevs that are not missing and are not logs. At every | |
2078 | * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. | |
2079 | */ | |
2080 | static uint64_t | |
2081 | spa_healthy_core_tvds(spa_t *spa) | |
2082 | { | |
2083 | vdev_t *rvd = spa->spa_root_vdev; | |
2084 | uint64_t tvds = 0; | |
2085 | ||
2086 | for (uint64_t i = 0; i < rvd->vdev_children; i++) { | |
2087 | vdev_t *vd = rvd->vdev_child[i]; | |
2088 | if (vd->vdev_islog) | |
2089 | continue; | |
2090 | if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) | |
2091 | tvds++; | |
2092 | } | |
2093 | ||
2094 | return (tvds); | |
2095 | } | |
2096 | ||
34dc7c2f BB |
2097 | /* |
2098 | * Checks to see if the given vdev could not be opened, in which case we post a | |
2099 | * sysevent to notify the autoreplace code that the device has been removed. | |
2100 | */ | |
2101 | static void | |
2102 | spa_check_removed(vdev_t *vd) | |
2103 | { | |
6cb8e530 | 2104 | for (uint64_t c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2105 | spa_check_removed(vd->vdev_child[c]); |
2106 | ||
7011fb60 | 2107 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && |
a1d477c2 | 2108 | vdev_is_concrete(vd)) { |
fb390aaf | 2109 | zfs_post_autoreplace(vd->vdev_spa, vd); |
12fa0466 | 2110 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); |
34dc7c2f BB |
2111 | } |
2112 | } | |
2113 | ||
6cb8e530 PZ |
2114 | static int |
2115 | spa_check_for_missing_logs(spa_t *spa) | |
9babb374 | 2116 | { |
6cb8e530 | 2117 | vdev_t *rvd = spa->spa_root_vdev; |
9babb374 | 2118 | |
428870ff | 2119 | /* |
572e2857 | 2120 | * If we're doing a normal import, then build up any additional |
6cb8e530 | 2121 | * diagnostic information about missing log devices. |
572e2857 | 2122 | * We'll pass this up to the user for further processing. |
428870ff | 2123 | */ |
572e2857 BB |
2124 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { |
2125 | nvlist_t **child, *nv; | |
2126 | uint64_t idx = 0; | |
2127 | ||
160987b5 | 2128 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), |
79c76d5b | 2129 | KM_SLEEP); |
65ad5d11 | 2130 | nv = fnvlist_alloc(); |
572e2857 | 2131 | |
6cb8e530 | 2132 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { |
572e2857 | 2133 | vdev_t *tvd = rvd->vdev_child[c]; |
572e2857 | 2134 | |
6cb8e530 PZ |
2135 | /* |
2136 | * We consider a device as missing only if it failed | |
2137 | * to open (i.e. offline or faulted is not considered | |
2138 | * as missing). | |
2139 | */ | |
2140 | if (tvd->vdev_islog && | |
2141 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
2142 | child[idx++] = vdev_config_generate(spa, tvd, | |
2143 | B_FALSE, VDEV_CONFIG_MISSING); | |
2144 | } | |
572e2857 | 2145 | } |
9babb374 | 2146 | |
6cb8e530 | 2147 | if (idx > 0) { |
795075e6 PD |
2148 | fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
2149 | (const nvlist_t * const *)child, idx); | |
6cb8e530 PZ |
2150 | fnvlist_add_nvlist(spa->spa_load_info, |
2151 | ZPOOL_CONFIG_MISSING_DEVICES, nv); | |
572e2857 | 2152 | |
6cb8e530 | 2153 | for (uint64_t i = 0; i < idx; i++) |
572e2857 BB |
2154 | nvlist_free(child[i]); |
2155 | } | |
2156 | nvlist_free(nv); | |
2157 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
572e2857 | 2158 | |
6cb8e530 PZ |
2159 | if (idx > 0) { |
2160 | spa_load_failed(spa, "some log devices are missing"); | |
db7d07e1 | 2161 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 PZ |
2162 | return (SET_ERROR(ENXIO)); |
2163 | } | |
2164 | } else { | |
2165 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
2166 | vdev_t *tvd = rvd->vdev_child[c]; | |
a1d477c2 | 2167 | |
6cb8e530 PZ |
2168 | if (tvd->vdev_islog && |
2169 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
572e2857 | 2170 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
6cb8e530 PZ |
2171 | spa_load_note(spa, "some log devices are " |
2172 | "missing, ZIL is dropped."); | |
db7d07e1 | 2173 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 | 2174 | break; |
e0ab3ab5 | 2175 | } |
572e2857 | 2176 | } |
9babb374 | 2177 | } |
e0ab3ab5 | 2178 | |
6cb8e530 | 2179 | return (0); |
9babb374 BB |
2180 | } |
2181 | ||
b128c09f BB |
2182 | /* |
2183 | * Check for missing log devices | |
2184 | */ | |
13fe0198 | 2185 | static boolean_t |
b128c09f BB |
2186 | spa_check_logs(spa_t *spa) |
2187 | { | |
13fe0198 | 2188 | boolean_t rv = B_FALSE; |
9c43027b | 2189 | dsl_pool_t *dp = spa_get_dsl(spa); |
13fe0198 | 2190 | |
b128c09f | 2191 | switch (spa->spa_log_state) { |
e75c13c3 BB |
2192 | default: |
2193 | break; | |
b128c09f BB |
2194 | case SPA_LOG_MISSING: |
2195 | /* need to recheck in case slog has been restored */ | |
2196 | case SPA_LOG_UNKNOWN: | |
9c43027b AJ |
2197 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
2198 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
13fe0198 | 2199 | if (rv) |
428870ff | 2200 | spa_set_log_state(spa, SPA_LOG_MISSING); |
b128c09f | 2201 | break; |
b128c09f | 2202 | } |
13fe0198 | 2203 | return (rv); |
b128c09f BB |
2204 | } |
2205 | ||
aa755b35 MA |
2206 | /* |
2207 | * Passivate any log vdevs (note, does not apply to embedded log metaslabs). | |
2208 | */ | |
428870ff BB |
2209 | static boolean_t |
2210 | spa_passivate_log(spa_t *spa) | |
34dc7c2f | 2211 | { |
428870ff BB |
2212 | vdev_t *rvd = spa->spa_root_vdev; |
2213 | boolean_t slog_found = B_FALSE; | |
b128c09f | 2214 | |
428870ff | 2215 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
fb5f0bc8 | 2216 | |
1c27024e | 2217 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2218 | vdev_t *tvd = rvd->vdev_child[c]; |
34dc7c2f | 2219 | |
428870ff | 2220 | if (tvd->vdev_islog) { |
aa755b35 MA |
2221 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); |
2222 | metaslab_group_passivate(tvd->vdev_mg); | |
428870ff BB |
2223 | slog_found = B_TRUE; |
2224 | } | |
34dc7c2f BB |
2225 | } |
2226 | ||
428870ff BB |
2227 | return (slog_found); |
2228 | } | |
34dc7c2f | 2229 | |
aa755b35 MA |
2230 | /* |
2231 | * Activate any log vdevs (note, does not apply to embedded log metaslabs). | |
2232 | */ | |
428870ff BB |
2233 | static void |
2234 | spa_activate_log(spa_t *spa) | |
2235 | { | |
2236 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 2237 | |
428870ff BB |
2238 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
2239 | ||
1c27024e | 2240 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2241 | vdev_t *tvd = rvd->vdev_child[c]; |
428870ff | 2242 | |
aa755b35 MA |
2243 | if (tvd->vdev_islog) { |
2244 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); | |
2245 | metaslab_group_activate(tvd->vdev_mg); | |
2246 | } | |
34dc7c2f | 2247 | } |
428870ff | 2248 | } |
34dc7c2f | 2249 | |
428870ff | 2250 | int |
a1d477c2 | 2251 | spa_reset_logs(spa_t *spa) |
428870ff | 2252 | { |
13fe0198 | 2253 | int error; |
9babb374 | 2254 | |
a1d477c2 | 2255 | error = dmu_objset_find(spa_name(spa), zil_reset, |
13fe0198 MA |
2256 | NULL, DS_FIND_CHILDREN); |
2257 | if (error == 0) { | |
428870ff BB |
2258 | /* |
2259 | * We successfully offlined the log device, sync out the | |
2260 | * current txg so that the "stubby" block can be removed | |
2261 | * by zil_sync(). | |
2262 | */ | |
2263 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
2264 | } | |
2265 | return (error); | |
2266 | } | |
34dc7c2f | 2267 | |
428870ff BB |
2268 | static void |
2269 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
2270 | { | |
1c27024e | 2271 | for (int i = 0; i < sav->sav_count; i++) |
428870ff BB |
2272 | spa_check_removed(sav->sav_vdevs[i]); |
2273 | } | |
34dc7c2f | 2274 | |
428870ff BB |
2275 | void |
2276 | spa_claim_notify(zio_t *zio) | |
2277 | { | |
2278 | spa_t *spa = zio->io_spa; | |
34dc7c2f | 2279 | |
428870ff BB |
2280 | if (zio->io_error) |
2281 | return; | |
34dc7c2f | 2282 | |
428870ff BB |
2283 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ |
2284 | if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) | |
2285 | spa->spa_claim_max_txg = zio->io_bp->blk_birth; | |
2286 | mutex_exit(&spa->spa_props_lock); | |
2287 | } | |
34dc7c2f | 2288 | |
428870ff | 2289 | typedef struct spa_load_error { |
f2c5bc15 | 2290 | boolean_t sle_verify_data; |
428870ff BB |
2291 | uint64_t sle_meta_count; |
2292 | uint64_t sle_data_count; | |
2293 | } spa_load_error_t; | |
34dc7c2f | 2294 | |
428870ff BB |
2295 | static void |
2296 | spa_load_verify_done(zio_t *zio) | |
2297 | { | |
2298 | blkptr_t *bp = zio->io_bp; | |
2299 | spa_load_error_t *sle = zio->io_private; | |
2300 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
2301 | int error = zio->io_error; | |
dea377c0 | 2302 | spa_t *spa = zio->io_spa; |
34dc7c2f | 2303 | |
a6255b7f | 2304 | abd_free(zio->io_abd); |
428870ff | 2305 | if (error) { |
9ae529ec | 2306 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && |
428870ff | 2307 | type != DMU_OT_INTENT_LOG) |
bc89ac84 | 2308 | atomic_inc_64(&sle->sle_meta_count); |
428870ff | 2309 | else |
bc89ac84 | 2310 | atomic_inc_64(&sle->sle_data_count); |
34dc7c2f | 2311 | } |
dea377c0 MA |
2312 | |
2313 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2314 | spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp); |
dea377c0 MA |
2315 | cv_broadcast(&spa->spa_scrub_io_cv); |
2316 | mutex_exit(&spa->spa_scrub_lock); | |
428870ff | 2317 | } |
34dc7c2f | 2318 | |
dea377c0 | 2319 | /* |
e1cfd73f | 2320 | * Maximum number of inflight bytes is the log2 fraction of the arc size. |
c8242a96 | 2321 | * By default, we set it to 1/16th of the arc. |
dea377c0 | 2322 | */ |
fdc2d303 | 2323 | static uint_t spa_load_verify_shift = 4; |
18168da7 AZ |
2324 | static int spa_load_verify_metadata = B_TRUE; |
2325 | static int spa_load_verify_data = B_TRUE; | |
dea377c0 | 2326 | |
428870ff BB |
2327 | static int |
2328 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
5dbd68a3 | 2329 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) |
428870ff | 2330 | { |
f2c5bc15 AM |
2331 | zio_t *rio = arg; |
2332 | spa_load_error_t *sle = rio->io_private; | |
2333 | ||
14e4e3cb AZ |
2334 | (void) zilog, (void) dnp; |
2335 | ||
dea377c0 MA |
2336 | /* |
2337 | * Note: normally this routine will not be called if | |
2338 | * spa_load_verify_metadata is not set. However, it may be useful | |
2339 | * to manually set the flag after the traversal has begun. | |
2340 | */ | |
2341 | if (!spa_load_verify_metadata) | |
2342 | return (0); | |
2cd0f98f BB |
2343 | |
2344 | /* | |
2345 | * Sanity check the block pointer in order to detect obvious damage | |
2346 | * before using the contents in subsequent checks or in zio_read(). | |
2347 | * When damaged consider it to be a metadata error since we cannot | |
2348 | * trust the BP_GET_TYPE and BP_GET_LEVEL values. | |
2349 | */ | |
2350 | if (!zfs_blkptr_verify(spa, bp, B_FALSE, BLK_VERIFY_LOG)) { | |
2351 | atomic_inc_64(&sle->sle_meta_count); | |
2352 | return (0); | |
2353 | } | |
2354 | ||
2355 | if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) || | |
2356 | BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) | |
2357 | return (0); | |
2358 | ||
f2c5bc15 AM |
2359 | if (!BP_IS_METADATA(bp) && |
2360 | (!spa_load_verify_data || !sle->sle_verify_data)) | |
dea377c0 MA |
2361 | return (0); |
2362 | ||
1e527162 GW |
2363 | uint64_t maxinflight_bytes = |
2364 | arc_target_bytes() >> spa_load_verify_shift; | |
1c27024e | 2365 | size_t size = BP_GET_PSIZE(bp); |
dea377c0 MA |
2366 | |
2367 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2368 | while (spa->spa_load_verify_bytes >= maxinflight_bytes) |
dea377c0 | 2369 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); |
c8242a96 | 2370 | spa->spa_load_verify_bytes += size; |
dea377c0 MA |
2371 | mutex_exit(&spa->spa_scrub_lock); |
2372 | ||
a6255b7f | 2373 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, |
dea377c0 MA |
2374 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, |
2375 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2376 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
428870ff BB |
2377 | return (0); |
2378 | } | |
34dc7c2f | 2379 | |
65c7cc49 | 2380 | static int |
d1d19c78 PD |
2381 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
2382 | { | |
14e4e3cb AZ |
2383 | (void) dp, (void) arg; |
2384 | ||
d1d19c78 PD |
2385 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) |
2386 | return (SET_ERROR(ENAMETOOLONG)); | |
2387 | ||
2388 | return (0); | |
2389 | } | |
2390 | ||
428870ff BB |
2391 | static int |
2392 | spa_load_verify(spa_t *spa) | |
2393 | { | |
2394 | zio_t *rio; | |
2395 | spa_load_error_t sle = { 0 }; | |
8a393be3 | 2396 | zpool_load_policy_t policy; |
428870ff | 2397 | boolean_t verify_ok = B_FALSE; |
dea377c0 | 2398 | int error = 0; |
34dc7c2f | 2399 | |
8a393be3 | 2400 | zpool_get_load_policy(spa->spa_config, &policy); |
34dc7c2f | 2401 | |
f2c5bc15 AM |
2402 | if (policy.zlp_rewind & ZPOOL_NEVER_REWIND || |
2403 | policy.zlp_maxmeta == UINT64_MAX) | |
428870ff | 2404 | return (0); |
34dc7c2f | 2405 | |
d1d19c78 PD |
2406 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); |
2407 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2408 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2409 | DS_FIND_CHILDREN); | |
2410 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2411 | if (error != 0) | |
2412 | return (error); | |
2413 | ||
f2c5bc15 AM |
2414 | /* |
2415 | * Verify data only if we are rewinding or error limit was set. | |
2416 | * Otherwise nothing except dbgmsg care about it to waste time. | |
2417 | */ | |
2418 | sle.sle_verify_data = (policy.zlp_rewind & ZPOOL_REWIND_MASK) || | |
2419 | (policy.zlp_maxdata < UINT64_MAX); | |
2420 | ||
428870ff BB |
2421 | rio = zio_root(spa, NULL, &sle, |
2422 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
34dc7c2f | 2423 | |
dea377c0 | 2424 | if (spa_load_verify_metadata) { |
4a0ee12a PZ |
2425 | if (spa->spa_extreme_rewind) { |
2426 | spa_load_note(spa, "performing a complete scan of the " | |
2427 | "pool since extreme rewind is on. This may take " | |
2428 | "a very long time.\n (spa_load_verify_data=%u, " | |
2429 | "spa_load_verify_metadata=%u)", | |
2430 | spa_load_verify_data, spa_load_verify_metadata); | |
2431 | } | |
c8242a96 | 2432 | |
dea377c0 | 2433 | error = traverse_pool(spa, spa->spa_verify_min_txg, |
b5256303 TC |
2434 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | |
2435 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
dea377c0 | 2436 | } |
428870ff BB |
2437 | |
2438 | (void) zio_wait(rio); | |
c8242a96 | 2439 | ASSERT0(spa->spa_load_verify_bytes); |
428870ff BB |
2440 | |
2441 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2442 | spa->spa_load_data_errors = sle.sle_data_count; | |
2443 | ||
afd2f7b7 PZ |
2444 | if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { |
2445 | spa_load_note(spa, "spa_load_verify found %llu metadata errors " | |
2446 | "and %llu data errors", (u_longlong_t)sle.sle_meta_count, | |
2447 | (u_longlong_t)sle.sle_data_count); | |
2448 | } | |
2449 | ||
2450 | if (spa_load_verify_dryrun || | |
8a393be3 PZ |
2451 | (!error && sle.sle_meta_count <= policy.zlp_maxmeta && |
2452 | sle.sle_data_count <= policy.zlp_maxdata)) { | |
572e2857 BB |
2453 | int64_t loss = 0; |
2454 | ||
428870ff BB |
2455 | verify_ok = B_TRUE; |
2456 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2457 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
572e2857 BB |
2458 | |
2459 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
65ad5d11 AJ |
2460 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, |
2461 | spa->spa_load_txg_ts); | |
2462 | fnvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, | |
2463 | loss); | |
f2c5bc15 AM |
2464 | fnvlist_add_uint64(spa->spa_load_info, |
2465 | ZPOOL_CONFIG_LOAD_META_ERRORS, sle.sle_meta_count); | |
65ad5d11 AJ |
2466 | fnvlist_add_uint64(spa->spa_load_info, |
2467 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count); | |
428870ff BB |
2468 | } else { |
2469 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2470 | } | |
2471 | ||
afd2f7b7 PZ |
2472 | if (spa_load_verify_dryrun) |
2473 | return (0); | |
2474 | ||
428870ff BB |
2475 | if (error) { |
2476 | if (error != ENXIO && error != EIO) | |
2e528b49 | 2477 | error = SET_ERROR(EIO); |
428870ff BB |
2478 | return (error); |
2479 | } | |
2480 | ||
2481 | return (verify_ok ? 0 : EIO); | |
2482 | } | |
2483 | ||
2484 | /* | |
2485 | * Find a value in the pool props object. | |
2486 | */ | |
2487 | static void | |
2488 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2489 | { | |
2490 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2491 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2492 | } | |
2493 | ||
2494 | /* | |
2495 | * Find a value in the pool directory object. | |
2496 | */ | |
2497 | static int | |
4a0ee12a | 2498 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) |
428870ff | 2499 | { |
4a0ee12a PZ |
2500 | int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
2501 | name, sizeof (uint64_t), 1, val); | |
2502 | ||
2503 | if (error != 0 && (error != ENOENT || log_enoent)) { | |
2504 | spa_load_failed(spa, "couldn't get '%s' value in MOS directory " | |
2505 | "[error=%d]", name, error); | |
2506 | } | |
2507 | ||
2508 | return (error); | |
428870ff BB |
2509 | } |
2510 | ||
2511 | static int | |
2512 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2513 | { | |
2514 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
a1d477c2 | 2515 | return (SET_ERROR(err)); |
428870ff BB |
2516 | } |
2517 | ||
37f03da8 SH |
2518 | boolean_t |
2519 | spa_livelist_delete_check(spa_t *spa) | |
2520 | { | |
2521 | return (spa->spa_livelists_to_delete != 0); | |
2522 | } | |
2523 | ||
37f03da8 SH |
2524 | static boolean_t |
2525 | spa_livelist_delete_cb_check(void *arg, zthr_t *z) | |
2526 | { | |
14e4e3cb | 2527 | (void) z; |
37f03da8 SH |
2528 | spa_t *spa = arg; |
2529 | return (spa_livelist_delete_check(spa)); | |
2530 | } | |
2531 | ||
2532 | static int | |
2533 | delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
2534 | { | |
2535 | spa_t *spa = arg; | |
2536 | zio_free(spa, tx->tx_txg, bp); | |
2537 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, | |
2538 | -bp_get_dsize_sync(spa, bp), | |
2539 | -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); | |
2540 | return (0); | |
2541 | } | |
2542 | ||
2543 | static int | |
2544 | dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp) | |
2545 | { | |
2546 | int err; | |
2547 | zap_cursor_t zc; | |
2548 | zap_attribute_t za; | |
2549 | zap_cursor_init(&zc, os, zap_obj); | |
2550 | err = zap_cursor_retrieve(&zc, &za); | |
2551 | zap_cursor_fini(&zc); | |
2552 | if (err == 0) | |
2553 | *llp = za.za_first_integer; | |
2554 | return (err); | |
2555 | } | |
2556 | ||
2557 | /* | |
2558 | * Components of livelist deletion that must be performed in syncing | |
2559 | * context: freeing block pointers and updating the pool-wide data | |
2560 | * structures to indicate how much work is left to do | |
2561 | */ | |
2562 | typedef struct sublist_delete_arg { | |
2563 | spa_t *spa; | |
2564 | dsl_deadlist_t *ll; | |
2565 | uint64_t key; | |
2566 | bplist_t *to_free; | |
2567 | } sublist_delete_arg_t; | |
2568 | ||
2569 | static void | |
2570 | sublist_delete_sync(void *arg, dmu_tx_t *tx) | |
2571 | { | |
2572 | sublist_delete_arg_t *sda = arg; | |
2573 | spa_t *spa = sda->spa; | |
2574 | dsl_deadlist_t *ll = sda->ll; | |
2575 | uint64_t key = sda->key; | |
2576 | bplist_t *to_free = sda->to_free; | |
2577 | ||
2578 | bplist_iterate(to_free, delete_blkptr_cb, spa, tx); | |
2579 | dsl_deadlist_remove_entry(ll, key, tx); | |
2580 | } | |
2581 | ||
2582 | typedef struct livelist_delete_arg { | |
2583 | spa_t *spa; | |
2584 | uint64_t ll_obj; | |
2585 | uint64_t zap_obj; | |
2586 | } livelist_delete_arg_t; | |
2587 | ||
2588 | static void | |
2589 | livelist_delete_sync(void *arg, dmu_tx_t *tx) | |
2590 | { | |
2591 | livelist_delete_arg_t *lda = arg; | |
2592 | spa_t *spa = lda->spa; | |
2593 | uint64_t ll_obj = lda->ll_obj; | |
2594 | uint64_t zap_obj = lda->zap_obj; | |
2595 | objset_t *mos = spa->spa_meta_objset; | |
2596 | uint64_t count; | |
2597 | ||
2598 | /* free the livelist and decrement the feature count */ | |
2599 | VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx)); | |
2600 | dsl_deadlist_free(mos, ll_obj, tx); | |
2601 | spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx); | |
2602 | VERIFY0(zap_count(mos, zap_obj, &count)); | |
2603 | if (count == 0) { | |
2604 | /* no more livelists to delete */ | |
2605 | VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT, | |
2606 | DMU_POOL_DELETED_CLONES, tx)); | |
2607 | VERIFY0(zap_destroy(mos, zap_obj, tx)); | |
2608 | spa->spa_livelists_to_delete = 0; | |
e60e158e | 2609 | spa_notify_waiters(spa); |
37f03da8 SH |
2610 | } |
2611 | } | |
2612 | ||
2613 | /* | |
2614 | * Load in the value for the livelist to be removed and open it. Then, | |
2615 | * load its first sublist and determine which block pointers should actually | |
2616 | * be freed. Then, call a synctask which performs the actual frees and updates | |
2617 | * the pool-wide livelist data. | |
2618 | */ | |
65c7cc49 | 2619 | static void |
37f03da8 SH |
2620 | spa_livelist_delete_cb(void *arg, zthr_t *z) |
2621 | { | |
2622 | spa_t *spa = arg; | |
2623 | uint64_t ll_obj = 0, count; | |
2624 | objset_t *mos = spa->spa_meta_objset; | |
2625 | uint64_t zap_obj = spa->spa_livelists_to_delete; | |
2626 | /* | |
2627 | * Determine the next livelist to delete. This function should only | |
2628 | * be called if there is at least one deleted clone. | |
2629 | */ | |
2630 | VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj)); | |
2631 | VERIFY0(zap_count(mos, ll_obj, &count)); | |
2632 | if (count > 0) { | |
c9562576 | 2633 | dsl_deadlist_t *ll; |
37f03da8 SH |
2634 | dsl_deadlist_entry_t *dle; |
2635 | bplist_t to_free; | |
c9562576 PS |
2636 | ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP); |
2637 | dsl_deadlist_open(ll, mos, ll_obj); | |
2638 | dle = dsl_deadlist_first(ll); | |
37f03da8 SH |
2639 | ASSERT3P(dle, !=, NULL); |
2640 | bplist_create(&to_free); | |
2641 | int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free, | |
2642 | z, NULL); | |
2643 | if (err == 0) { | |
2644 | sublist_delete_arg_t sync_arg = { | |
2645 | .spa = spa, | |
c9562576 | 2646 | .ll = ll, |
37f03da8 SH |
2647 | .key = dle->dle_mintxg, |
2648 | .to_free = &to_free | |
2649 | }; | |
2650 | zfs_dbgmsg("deleting sublist (id %llu) from" | |
8e739b2c RE |
2651 | " livelist %llu, %lld remaining", |
2652 | (u_longlong_t)dle->dle_bpobj.bpo_object, | |
2653 | (u_longlong_t)ll_obj, (longlong_t)count - 1); | |
37f03da8 SH |
2654 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, |
2655 | sublist_delete_sync, &sync_arg, 0, | |
2656 | ZFS_SPACE_CHECK_DESTROY)); | |
2657 | } else { | |
d87676a9 | 2658 | VERIFY3U(err, ==, EINTR); |
37f03da8 SH |
2659 | } |
2660 | bplist_clear(&to_free); | |
2661 | bplist_destroy(&to_free); | |
c9562576 PS |
2662 | dsl_deadlist_close(ll); |
2663 | kmem_free(ll, sizeof (dsl_deadlist_t)); | |
37f03da8 SH |
2664 | } else { |
2665 | livelist_delete_arg_t sync_arg = { | |
2666 | .spa = spa, | |
2667 | .ll_obj = ll_obj, | |
2668 | .zap_obj = zap_obj | |
2669 | }; | |
8e739b2c RE |
2670 | zfs_dbgmsg("deletion of livelist %llu completed", |
2671 | (u_longlong_t)ll_obj); | |
37f03da8 SH |
2672 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync, |
2673 | &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY)); | |
2674 | } | |
2675 | } | |
2676 | ||
65c7cc49 | 2677 | static void |
37f03da8 SH |
2678 | spa_start_livelist_destroy_thread(spa_t *spa) |
2679 | { | |
2680 | ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL); | |
843e9ca2 SD |
2681 | spa->spa_livelist_delete_zthr = |
2682 | zthr_create("z_livelist_destroy", | |
6bc61d22 TN |
2683 | spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa, |
2684 | minclsyspri); | |
37f03da8 SH |
2685 | } |
2686 | ||
2687 | typedef struct livelist_new_arg { | |
2688 | bplist_t *allocs; | |
2689 | bplist_t *frees; | |
2690 | } livelist_new_arg_t; | |
2691 | ||
2692 | static int | |
2693 | livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
2694 | dmu_tx_t *tx) | |
2695 | { | |
2696 | ASSERT(tx == NULL); | |
2697 | livelist_new_arg_t *lna = arg; | |
2698 | if (bp_freed) { | |
2699 | bplist_append(lna->frees, bp); | |
2700 | } else { | |
2701 | bplist_append(lna->allocs, bp); | |
2702 | zfs_livelist_condense_new_alloc++; | |
2703 | } | |
2704 | return (0); | |
2705 | } | |
2706 | ||
2707 | typedef struct livelist_condense_arg { | |
2708 | spa_t *spa; | |
2709 | bplist_t to_keep; | |
2710 | uint64_t first_size; | |
2711 | uint64_t next_size; | |
2712 | } livelist_condense_arg_t; | |
2713 | ||
2714 | static void | |
2715 | spa_livelist_condense_sync(void *arg, dmu_tx_t *tx) | |
2716 | { | |
2717 | livelist_condense_arg_t *lca = arg; | |
2718 | spa_t *spa = lca->spa; | |
2719 | bplist_t new_frees; | |
2720 | dsl_dataset_t *ds = spa->spa_to_condense.ds; | |
2721 | ||
2722 | /* Have we been cancelled? */ | |
2723 | if (spa->spa_to_condense.cancelled) { | |
2724 | zfs_livelist_condense_sync_cancel++; | |
2725 | goto out; | |
2726 | } | |
2727 | ||
2728 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
2729 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
2730 | dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; | |
2731 | ||
2732 | /* | |
2733 | * It's possible that the livelist was changed while the zthr was | |
2734 | * running. Therefore, we need to check for new blkptrs in the two | |
2735 | * entries being condensed and continue to track them in the livelist. | |
2736 | * Because of the way we handle remapped blkptrs (see dbuf_remap_impl), | |
2737 | * it's possible that the newly added blkptrs are FREEs or ALLOCs so | |
2738 | * we need to sort them into two different bplists. | |
2739 | */ | |
2740 | uint64_t first_obj = first->dle_bpobj.bpo_object; | |
2741 | uint64_t next_obj = next->dle_bpobj.bpo_object; | |
2742 | uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
2743 | uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
2744 | ||
2745 | bplist_create(&new_frees); | |
2746 | livelist_new_arg_t new_bps = { | |
2747 | .allocs = &lca->to_keep, | |
2748 | .frees = &new_frees, | |
2749 | }; | |
2750 | ||
2751 | if (cur_first_size > lca->first_size) { | |
2752 | VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj, | |
2753 | livelist_track_new_cb, &new_bps, lca->first_size)); | |
2754 | } | |
2755 | if (cur_next_size > lca->next_size) { | |
2756 | VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj, | |
2757 | livelist_track_new_cb, &new_bps, lca->next_size)); | |
2758 | } | |
2759 | ||
2760 | dsl_deadlist_clear_entry(first, ll, tx); | |
2761 | ASSERT(bpobj_is_empty(&first->dle_bpobj)); | |
2762 | dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx); | |
2763 | ||
2764 | bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx); | |
2765 | bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx); | |
2766 | bplist_destroy(&new_frees); | |
2767 | ||
2768 | char dsname[ZFS_MAX_DATASET_NAME_LEN]; | |
2769 | dsl_dataset_name(ds, dsname); | |
2770 | zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu " | |
2771 | "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu " | |
8e739b2c RE |
2772 | "(%llu blkptrs)", (u_longlong_t)tx->tx_txg, dsname, |
2773 | (u_longlong_t)ds->ds_object, (u_longlong_t)first_obj, | |
2774 | (u_longlong_t)cur_first_size, (u_longlong_t)next_obj, | |
2775 | (u_longlong_t)cur_next_size, | |
2776 | (u_longlong_t)first->dle_bpobj.bpo_object, | |
2777 | (u_longlong_t)first->dle_bpobj.bpo_phys->bpo_num_blkptrs); | |
37f03da8 SH |
2778 | out: |
2779 | dmu_buf_rele(ds->ds_dbuf, spa); | |
2780 | spa->spa_to_condense.ds = NULL; | |
2781 | bplist_clear(&lca->to_keep); | |
2782 | bplist_destroy(&lca->to_keep); | |
2783 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
2784 | spa->spa_to_condense.syncing = B_FALSE; | |
2785 | } | |
2786 | ||
65c7cc49 | 2787 | static void |
37f03da8 SH |
2788 | spa_livelist_condense_cb(void *arg, zthr_t *t) |
2789 | { | |
2790 | while (zfs_livelist_condense_zthr_pause && | |
2791 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
2792 | delay(1); | |
2793 | ||
2794 | spa_t *spa = arg; | |
2795 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
2796 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
2797 | uint64_t first_size, next_size; | |
2798 | ||
2799 | livelist_condense_arg_t *lca = | |
2800 | kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP); | |
2801 | bplist_create(&lca->to_keep); | |
2802 | ||
2803 | /* | |
2804 | * Process the livelists (matching FREEs and ALLOCs) in open context | |
2805 | * so we have minimal work in syncing context to condense. | |
2806 | * | |
2807 | * We save bpobj sizes (first_size and next_size) to use later in | |
2808 | * syncing context to determine if entries were added to these sublists | |
2809 | * while in open context. This is possible because the clone is still | |
2810 | * active and open for normal writes and we want to make sure the new, | |
2811 | * unprocessed blockpointers are inserted into the livelist normally. | |
2812 | * | |
2813 | * Note that dsl_process_sub_livelist() both stores the size number of | |
2814 | * blockpointers and iterates over them while the bpobj's lock held, so | |
2815 | * the sizes returned to us are consistent which what was actually | |
2816 | * processed. | |
2817 | */ | |
2818 | int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t, | |
2819 | &first_size); | |
2820 | if (err == 0) | |
2821 | err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep, | |
2822 | t, &next_size); | |
2823 | ||
2824 | if (err == 0) { | |
2825 | while (zfs_livelist_condense_sync_pause && | |
2826 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
2827 | delay(1); | |
2828 | ||
2829 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
2830 | dmu_tx_mark_netfree(tx); | |
2831 | dmu_tx_hold_space(tx, 1); | |
2832 | err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE); | |
2833 | if (err == 0) { | |
2834 | /* | |
2835 | * Prevent the condense zthr restarting before | |
2836 | * the synctask completes. | |
2837 | */ | |
2838 | spa->spa_to_condense.syncing = B_TRUE; | |
2839 | lca->spa = spa; | |
2840 | lca->first_size = first_size; | |
2841 | lca->next_size = next_size; | |
2842 | dsl_sync_task_nowait(spa_get_dsl(spa), | |
38080324 | 2843 | spa_livelist_condense_sync, lca, tx); |
37f03da8 SH |
2844 | dmu_tx_commit(tx); |
2845 | return; | |
2846 | } | |
2847 | } | |
2848 | /* | |
2849 | * Condensing can not continue: either it was externally stopped or | |
2850 | * we were unable to assign to a tx because the pool has run out of | |
2851 | * space. In the second case, we'll just end up trying to condense | |
2852 | * again in a later txg. | |
2853 | */ | |
2854 | ASSERT(err != 0); | |
2855 | bplist_clear(&lca->to_keep); | |
2856 | bplist_destroy(&lca->to_keep); | |
2857 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
2858 | dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa); | |
2859 | spa->spa_to_condense.ds = NULL; | |
2860 | if (err == EINTR) | |
2861 | zfs_livelist_condense_zthr_cancel++; | |
2862 | } | |
2863 | ||
37f03da8 SH |
2864 | /* |
2865 | * Check that there is something to condense but that a condense is not | |
2866 | * already in progress and that condensing has not been cancelled. | |
2867 | */ | |
2868 | static boolean_t | |
2869 | spa_livelist_condense_cb_check(void *arg, zthr_t *z) | |
2870 | { | |
14e4e3cb | 2871 | (void) z; |
37f03da8 SH |
2872 | spa_t *spa = arg; |
2873 | if ((spa->spa_to_condense.ds != NULL) && | |
2874 | (spa->spa_to_condense.syncing == B_FALSE) && | |
2875 | (spa->spa_to_condense.cancelled == B_FALSE)) { | |
2876 | return (B_TRUE); | |
2877 | } | |
2878 | return (B_FALSE); | |
2879 | } | |
2880 | ||
65c7cc49 | 2881 | static void |
37f03da8 SH |
2882 | spa_start_livelist_condensing_thread(spa_t *spa) |
2883 | { | |
2884 | spa->spa_to_condense.ds = NULL; | |
2885 | spa->spa_to_condense.first = NULL; | |
2886 | spa->spa_to_condense.next = NULL; | |
2887 | spa->spa_to_condense.syncing = B_FALSE; | |
2888 | spa->spa_to_condense.cancelled = B_FALSE; | |
2889 | ||
2890 | ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL); | |
843e9ca2 SD |
2891 | spa->spa_livelist_condense_zthr = |
2892 | zthr_create("z_livelist_condense", | |
2893 | spa_livelist_condense_cb_check, | |
6bc61d22 | 2894 | spa_livelist_condense_cb, spa, minclsyspri); |
37f03da8 SH |
2895 | } |
2896 | ||
9d5b5245 SD |
2897 | static void |
2898 | spa_spawn_aux_threads(spa_t *spa) | |
2899 | { | |
2900 | ASSERT(spa_writeable(spa)); | |
2901 | ||
2902 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
2903 | ||
2904 | spa_start_indirect_condensing_thread(spa); | |
37f03da8 SH |
2905 | spa_start_livelist_destroy_thread(spa); |
2906 | spa_start_livelist_condensing_thread(spa); | |
d2734cce SD |
2907 | |
2908 | ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); | |
2909 | spa->spa_checkpoint_discard_zthr = | |
843e9ca2 SD |
2910 | zthr_create("z_checkpoint_discard", |
2911 | spa_checkpoint_discard_thread_check, | |
6bc61d22 | 2912 | spa_checkpoint_discard_thread, spa, minclsyspri); |
9d5b5245 SD |
2913 | } |
2914 | ||
428870ff BB |
2915 | /* |
2916 | * Fix up config after a partly-completed split. This is done with the | |
2917 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
2918 | * pool have that entry in their config, but only the splitting one contains | |
2919 | * a list of all the guids of the vdevs that are being split off. | |
2920 | * | |
2921 | * This function determines what to do with that list: either rejoin | |
2922 | * all the disks to the pool, or complete the splitting process. To attempt | |
2923 | * the rejoin, each disk that is offlined is marked online again, and | |
2924 | * we do a reopen() call. If the vdev label for every disk that was | |
2925 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
2926 | * then we call vdev_split() on each disk, and complete the split. | |
2927 | * | |
2928 | * Otherwise we leave the config alone, with all the vdevs in place in | |
2929 | * the original pool. | |
2930 | */ | |
2931 | static void | |
2932 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
2933 | { | |
2934 | uint_t extracted; | |
2935 | uint64_t *glist; | |
2936 | uint_t i, gcount; | |
2937 | nvlist_t *nvl; | |
2938 | vdev_t **vd; | |
2939 | boolean_t attempt_reopen; | |
2940 | ||
2941 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
2942 | return; | |
2943 | ||
2944 | /* check that the config is complete */ | |
2945 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
2946 | &glist, &gcount) != 0) | |
2947 | return; | |
2948 | ||
79c76d5b | 2949 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); |
428870ff BB |
2950 | |
2951 | /* attempt to online all the vdevs & validate */ | |
2952 | attempt_reopen = B_TRUE; | |
2953 | for (i = 0; i < gcount; i++) { | |
2954 | if (glist[i] == 0) /* vdev is hole */ | |
2955 | continue; | |
2956 | ||
2957 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
2958 | if (vd[i] == NULL) { | |
2959 | /* | |
2960 | * Don't bother attempting to reopen the disks; | |
2961 | * just do the split. | |
2962 | */ | |
2963 | attempt_reopen = B_FALSE; | |
2964 | } else { | |
2965 | /* attempt to re-online it */ | |
2966 | vd[i]->vdev_offline = B_FALSE; | |
2967 | } | |
2968 | } | |
2969 | ||
2970 | if (attempt_reopen) { | |
2971 | vdev_reopen(spa->spa_root_vdev); | |
2972 | ||
2973 | /* check each device to see what state it's in */ | |
2974 | for (extracted = 0, i = 0; i < gcount; i++) { | |
2975 | if (vd[i] != NULL && | |
2976 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
2977 | break; | |
2978 | ++extracted; | |
2979 | } | |
2980 | } | |
2981 | ||
2982 | /* | |
2983 | * If every disk has been moved to the new pool, or if we never | |
2984 | * even attempted to look at them, then we split them off for | |
2985 | * good. | |
2986 | */ | |
2987 | if (!attempt_reopen || gcount == extracted) { | |
2988 | for (i = 0; i < gcount; i++) | |
2989 | if (vd[i] != NULL) | |
2990 | vdev_split(vd[i]); | |
2991 | vdev_reopen(spa->spa_root_vdev); | |
2992 | } | |
2993 | ||
2994 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
2995 | } | |
2996 | ||
2997 | static int | |
6cb8e530 | 2998 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) |
428870ff | 2999 | { |
a926aab9 | 3000 | const char *ereport = FM_EREPORT_ZFS_POOL; |
428870ff | 3001 | int error; |
428870ff | 3002 | |
6cb8e530 | 3003 | spa->spa_load_state = state; |
ca95f70d OF |
3004 | (void) spa_import_progress_set_state(spa_guid(spa), |
3005 | spa_load_state(spa)); | |
9ae529ec | 3006 | |
6cb8e530 | 3007 | gethrestime(&spa->spa_loaded_ts); |
d2734cce | 3008 | error = spa_load_impl(spa, type, &ereport); |
428870ff | 3009 | |
0c66c32d JG |
3010 | /* |
3011 | * Don't count references from objsets that are already closed | |
3012 | * and are making their way through the eviction process. | |
3013 | */ | |
3014 | spa_evicting_os_wait(spa); | |
424fd7c3 | 3015 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
572e2857 BB |
3016 | if (error) { |
3017 | if (error != EEXIST) { | |
3018 | spa->spa_loaded_ts.tv_sec = 0; | |
3019 | spa->spa_loaded_ts.tv_nsec = 0; | |
3020 | } | |
3021 | if (error != EBADF) { | |
1144586b | 3022 | (void) zfs_ereport_post(ereport, spa, |
4f072827 | 3023 | NULL, NULL, NULL, 0); |
572e2857 BB |
3024 | } |
3025 | } | |
428870ff BB |
3026 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; |
3027 | spa->spa_ena = 0; | |
3028 | ||
ca95f70d OF |
3029 | (void) spa_import_progress_set_state(spa_guid(spa), |
3030 | spa_load_state(spa)); | |
3031 | ||
428870ff BB |
3032 | return (error); |
3033 | } | |
3034 | ||
33cf67cd | 3035 | #ifdef ZFS_DEBUG |
e0ab3ab5 JS |
3036 | /* |
3037 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
3038 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
3039 | * spa's per-vdev ZAP list. | |
3040 | */ | |
3041 | static uint64_t | |
3042 | vdev_count_verify_zaps(vdev_t *vd) | |
3043 | { | |
3044 | spa_t *spa = vd->vdev_spa; | |
3045 | uint64_t total = 0; | |
e0ab3ab5 JS |
3046 | |
3047 | if (vd->vdev_top_zap != 0) { | |
3048 | total++; | |
3049 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3050 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
3051 | } | |
3052 | if (vd->vdev_leaf_zap != 0) { | |
3053 | total++; | |
3054 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3055 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
3056 | } | |
3057 | ||
1c27024e | 3058 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
3059 | total += vdev_count_verify_zaps(vd->vdev_child[i]); |
3060 | } | |
3061 | ||
3062 | return (total); | |
3063 | } | |
36542b06 AZ |
3064 | #else |
3065 | #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0) | |
33cf67cd | 3066 | #endif |
e0ab3ab5 | 3067 | |
379ca9cf OF |
3068 | /* |
3069 | * Determine whether the activity check is required. | |
3070 | */ | |
3071 | static boolean_t | |
bbffb59e BB |
3072 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, |
3073 | nvlist_t *config) | |
379ca9cf OF |
3074 | { |
3075 | uint64_t state = 0; | |
3076 | uint64_t hostid = 0; | |
3077 | uint64_t tryconfig_txg = 0; | |
3078 | uint64_t tryconfig_timestamp = 0; | |
060f0226 | 3079 | uint16_t tryconfig_mmp_seq = 0; |
379ca9cf OF |
3080 | nvlist_t *nvinfo; |
3081 | ||
3082 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3083 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
3084 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
3085 | &tryconfig_txg); | |
3086 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
3087 | &tryconfig_timestamp); | |
060f0226 OF |
3088 | (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ, |
3089 | &tryconfig_mmp_seq); | |
379ca9cf OF |
3090 | } |
3091 | ||
3092 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
379ca9cf OF |
3093 | |
3094 | /* | |
3095 | * Disable the MMP activity check - This is used by zdb which | |
3096 | * is intended to be used on potentially active pools. | |
3097 | */ | |
3098 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
3099 | return (B_FALSE); | |
3100 | ||
3101 | /* | |
3102 | * Skip the activity check when the MMP feature is disabled. | |
3103 | */ | |
3104 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
3105 | return (B_FALSE); | |
ca95f70d | 3106 | |
379ca9cf | 3107 | /* |
060f0226 OF |
3108 | * If the tryconfig_ values are nonzero, they are the results of an |
3109 | * earlier tryimport. If they all match the uberblock we just found, | |
3110 | * then the pool has not changed and we return false so we do not test | |
3111 | * a second time. | |
379ca9cf OF |
3112 | */ |
3113 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
060f0226 OF |
3114 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp && |
3115 | tryconfig_mmp_seq && tryconfig_mmp_seq == | |
3116 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) | |
379ca9cf OF |
3117 | return (B_FALSE); |
3118 | ||
3119 | /* | |
3120 | * Allow the activity check to be skipped when importing the pool | |
bbffb59e BB |
3121 | * on the same host which last imported it. Since the hostid from |
3122 | * configuration may be stale use the one read from the label. | |
379ca9cf | 3123 | */ |
bbffb59e BB |
3124 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) |
3125 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
3126 | ||
25f06d67 | 3127 | if (hostid == spa_get_hostid(spa)) |
379ca9cf OF |
3128 | return (B_FALSE); |
3129 | ||
3130 | /* | |
3131 | * Skip the activity test when the pool was cleanly exported. | |
3132 | */ | |
3133 | if (state != POOL_STATE_ACTIVE) | |
3134 | return (B_FALSE); | |
3135 | ||
3136 | return (B_TRUE); | |
3137 | } | |
3138 | ||
060f0226 OF |
3139 | /* |
3140 | * Nanoseconds the activity check must watch for changes on-disk. | |
3141 | */ | |
3142 | static uint64_t | |
3143 | spa_activity_check_duration(spa_t *spa, uberblock_t *ub) | |
3144 | { | |
3145 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
3146 | uint64_t multihost_interval = MSEC2NSEC( | |
3147 | MMP_INTERVAL_OK(zfs_multihost_interval)); | |
3148 | uint64_t import_delay = MAX(NANOSEC, import_intervals * | |
3149 | multihost_interval); | |
3150 | ||
3151 | /* | |
3152 | * Local tunables determine a minimum duration except for the case | |
3153 | * where we know when the remote host will suspend the pool if MMP | |
3154 | * writes do not land. | |
3155 | * | |
3156 | * See Big Theory comment at the top of mmp.c for the reasoning behind | |
3157 | * these cases and times. | |
3158 | */ | |
3159 | ||
3160 | ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100); | |
3161 | ||
3162 | if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3163 | MMP_FAIL_INT(ub) > 0) { | |
3164 | ||
3165 | /* MMP on remote host will suspend pool after failed writes */ | |
3166 | import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) * | |
3167 | MMP_IMPORT_SAFETY_FACTOR / 100; | |
3168 | ||
3169 | zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp " | |
3170 | "mmp_fails=%llu ub_mmp mmp_interval=%llu " | |
8e739b2c RE |
3171 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3172 | (u_longlong_t)MMP_FAIL_INT(ub), | |
3173 | (u_longlong_t)MMP_INTERVAL(ub), | |
3174 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3175 | |
3176 | } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3177 | MMP_FAIL_INT(ub) == 0) { | |
3178 | ||
3179 | /* MMP on remote host will never suspend pool */ | |
3180 | import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) + | |
3181 | ub->ub_mmp_delay) * import_intervals); | |
3182 | ||
3183 | zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp " | |
3184 | "mmp_interval=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3185 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3186 | (u_longlong_t)MMP_INTERVAL(ub), | |
3187 | (u_longlong_t)ub->ub_mmp_delay, | |
3188 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3189 | |
3190 | } else if (MMP_VALID(ub)) { | |
3191 | /* | |
e1cfd73f | 3192 | * zfs-0.7 compatibility case |
060f0226 OF |
3193 | */ |
3194 | ||
3195 | import_delay = MAX(import_delay, (multihost_interval + | |
3196 | ub->ub_mmp_delay) * import_intervals); | |
3197 | ||
3198 | zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3199 | "import_intervals=%llu leaves=%u", |
3200 | (u_longlong_t)import_delay, | |
3201 | (u_longlong_t)ub->ub_mmp_delay, | |
3202 | (u_longlong_t)import_intervals, | |
060f0226 OF |
3203 | vdev_count_leaves(spa)); |
3204 | } else { | |
3205 | /* Using local tunings is the only reasonable option */ | |
3206 | zfs_dbgmsg("pool last imported on non-MMP aware " | |
3207 | "host using import_delay=%llu multihost_interval=%llu " | |
8e739b2c RE |
3208 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3209 | (u_longlong_t)multihost_interval, | |
3210 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3211 | } |
3212 | ||
3213 | return (import_delay); | |
3214 | } | |
3215 | ||
379ca9cf OF |
3216 | /* |
3217 | * Perform the import activity check. If the user canceled the import or | |
3218 | * we detected activity then fail. | |
3219 | */ | |
3220 | static int | |
3221 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config) | |
3222 | { | |
379ca9cf OF |
3223 | uint64_t txg = ub->ub_txg; |
3224 | uint64_t timestamp = ub->ub_timestamp; | |
060f0226 OF |
3225 | uint64_t mmp_config = ub->ub_mmp_config; |
3226 | uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0; | |
3227 | uint64_t import_delay; | |
379ca9cf OF |
3228 | hrtime_t import_expire; |
3229 | nvlist_t *mmp_label = NULL; | |
3230 | vdev_t *rvd = spa->spa_root_vdev; | |
3231 | kcondvar_t cv; | |
3232 | kmutex_t mtx; | |
3233 | int error = 0; | |
3234 | ||
3235 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
3236 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
3237 | mutex_enter(&mtx); | |
3238 | ||
3239 | /* | |
3240 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
3241 | * during the earlier tryimport. If the txg recorded there is 0 then | |
3242 | * the pool is known to be active on another host. | |
3243 | * | |
060f0226 | 3244 | * Otherwise, the pool might be in use on another host. Check for |
379ca9cf OF |
3245 | * changes in the uberblocks on disk if necessary. |
3246 | */ | |
3247 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3248 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
3249 | ZPOOL_CONFIG_LOAD_INFO); | |
3250 | ||
3251 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
3252 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
3253 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
3254 | error = SET_ERROR(EREMOTEIO); | |
3255 | goto out; | |
3256 | } | |
3257 | } | |
3258 | ||
060f0226 | 3259 | import_delay = spa_activity_check_duration(spa, ub); |
533ea041 | 3260 | |
379ca9cf | 3261 | /* Add a small random factor in case of simultaneous imports (0-25%) */ |
29274c9f | 3262 | import_delay += import_delay * random_in_range(250) / 1000; |
ca95f70d OF |
3263 | |
3264 | import_expire = gethrtime() + import_delay; | |
379ca9cf OF |
3265 | |
3266 | while (gethrtime() < import_expire) { | |
ca95f70d OF |
3267 | (void) spa_import_progress_set_mmp_check(spa_guid(spa), |
3268 | NSEC2SEC(import_expire - gethrtime())); | |
3269 | ||
379ca9cf OF |
3270 | vdev_uberblock_load(rvd, ub, &mmp_label); |
3271 | ||
060f0226 OF |
3272 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp || |
3273 | mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) { | |
3274 | zfs_dbgmsg("multihost activity detected " | |
3275 | "txg %llu ub_txg %llu " | |
3276 | "timestamp %llu ub_timestamp %llu " | |
3277 | "mmp_config %#llx ub_mmp_config %#llx", | |
8e739b2c RE |
3278 | (u_longlong_t)txg, (u_longlong_t)ub->ub_txg, |
3279 | (u_longlong_t)timestamp, | |
3280 | (u_longlong_t)ub->ub_timestamp, | |
3281 | (u_longlong_t)mmp_config, | |
3282 | (u_longlong_t)ub->ub_mmp_config); | |
060f0226 | 3283 | |
379ca9cf OF |
3284 | error = SET_ERROR(EREMOTEIO); |
3285 | break; | |
3286 | } | |
3287 | ||
3288 | if (mmp_label) { | |
3289 | nvlist_free(mmp_label); | |
3290 | mmp_label = NULL; | |
3291 | } | |
3292 | ||
3293 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
3294 | if (error != -1) { | |
3295 | error = SET_ERROR(EINTR); | |
3296 | break; | |
3297 | } | |
3298 | error = 0; | |
3299 | } | |
3300 | ||
3301 | out: | |
3302 | mutex_exit(&mtx); | |
3303 | mutex_destroy(&mtx); | |
3304 | cv_destroy(&cv); | |
3305 | ||
3306 | /* | |
3307 | * If the pool is determined to be active store the status in the | |
3308 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
3309 | * available from configuration read from disk store them as well. | |
3310 | * This allows 'zpool import' to generate a more useful message. | |
3311 | * | |
3312 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
3313 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
3314 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
3315 | */ | |
3316 | if (error == EREMOTEIO) { | |
a926aab9 | 3317 | const char *hostname = "<unknown>"; |
379ca9cf OF |
3318 | uint64_t hostid = 0; |
3319 | ||
3320 | if (mmp_label) { | |
3321 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
3322 | hostname = fnvlist_lookup_string(mmp_label, | |
3323 | ZPOOL_CONFIG_HOSTNAME); | |
3324 | fnvlist_add_string(spa->spa_load_info, | |
3325 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
3326 | } | |
3327 | ||
3328 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
3329 | hostid = fnvlist_lookup_uint64(mmp_label, | |
3330 | ZPOOL_CONFIG_HOSTID); | |
3331 | fnvlist_add_uint64(spa->spa_load_info, | |
3332 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
3333 | } | |
3334 | } | |
3335 | ||
3336 | fnvlist_add_uint64(spa->spa_load_info, | |
3337 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
3338 | fnvlist_add_uint64(spa->spa_load_info, | |
3339 | ZPOOL_CONFIG_MMP_TXG, 0); | |
3340 | ||
3341 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
3342 | } | |
3343 | ||
3344 | if (mmp_label) | |
3345 | nvlist_free(mmp_label); | |
3346 | ||
3347 | return (error); | |
3348 | } | |
3349 | ||
9eb7b46e | 3350 | static int |
6cb8e530 PZ |
3351 | spa_verify_host(spa_t *spa, nvlist_t *mos_config) |
3352 | { | |
3353 | uint64_t hostid; | |
3354 | char *hostname; | |
3355 | uint64_t myhostid = 0; | |
3356 | ||
3357 | if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, | |
3358 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
3359 | hostname = fnvlist_lookup_string(mos_config, | |
3360 | ZPOOL_CONFIG_HOSTNAME); | |
3361 | ||
3362 | myhostid = zone_get_hostid(NULL); | |
3363 | ||
3364 | if (hostid != 0 && myhostid != 0 && hostid != myhostid) { | |
3365 | cmn_err(CE_WARN, "pool '%s' could not be " | |
3366 | "loaded as it was last accessed by " | |
3367 | "another system (host: %s hostid: 0x%llx). " | |
a2f944a1 RM |
3368 | "See: https://openzfs.github.io/openzfs-docs/msg/" |
3369 | "ZFS-8000-EY", | |
6cb8e530 PZ |
3370 | spa_name(spa), hostname, (u_longlong_t)hostid); |
3371 | spa_load_failed(spa, "hostid verification failed: pool " | |
3372 | "last accessed by host: %s (hostid: 0x%llx)", | |
3373 | hostname, (u_longlong_t)hostid); | |
3374 | return (SET_ERROR(EBADF)); | |
3375 | } | |
3376 | } | |
3377 | ||
3378 | return (0); | |
3379 | } | |
3380 | ||
3381 | static int | |
3382 | spa_ld_parse_config(spa_t *spa, spa_import_type_t type) | |
428870ff BB |
3383 | { |
3384 | int error = 0; | |
6cb8e530 | 3385 | nvlist_t *nvtree, *nvl, *config = spa->spa_config; |
1c27024e | 3386 | int parse; |
9eb7b46e | 3387 | vdev_t *rvd; |
6cb8e530 PZ |
3388 | uint64_t pool_guid; |
3389 | char *comment; | |
658fb802 | 3390 | char *compatibility; |
6cb8e530 PZ |
3391 | |
3392 | /* | |
3393 | * Versioning wasn't explicitly added to the label until later, so if | |
3394 | * it's not present treat it as the initial version. | |
3395 | */ | |
3396 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
3397 | &spa->spa_ubsync.ub_version) != 0) | |
3398 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
3399 | ||
3400 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { | |
3401 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3402 | ZPOOL_CONFIG_POOL_GUID); | |
3403 | return (SET_ERROR(EINVAL)); | |
3404 | } | |
3405 | ||
d2734cce SD |
3406 | /* |
3407 | * If we are doing an import, ensure that the pool is not already | |
3408 | * imported by checking if its pool guid already exists in the | |
3409 | * spa namespace. | |
3410 | * | |
3411 | * The only case that we allow an already imported pool to be | |
3412 | * imported again, is when the pool is checkpointed and we want to | |
3413 | * look at its checkpointed state from userland tools like zdb. | |
3414 | */ | |
3415 | #ifdef _KERNEL | |
3416 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3417 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3418 | spa_guid_exists(pool_guid, 0)) { | |
3419 | #else | |
3420 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3421 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3422 | spa_guid_exists(pool_guid, 0) && | |
3423 | !spa_importing_readonly_checkpoint(spa)) { | |
3424 | #endif | |
6cb8e530 PZ |
3425 | spa_load_failed(spa, "a pool with guid %llu is already open", |
3426 | (u_longlong_t)pool_guid); | |
3427 | return (SET_ERROR(EEXIST)); | |
3428 | } | |
3429 | ||
3430 | spa->spa_config_guid = pool_guid; | |
3431 | ||
3432 | nvlist_free(spa->spa_load_info); | |
3433 | spa->spa_load_info = fnvlist_alloc(); | |
3434 | ||
3435 | ASSERT(spa->spa_comment == NULL); | |
3436 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
3437 | spa->spa_comment = spa_strdup(comment); | |
3438 | ||
658fb802 CB |
3439 | ASSERT(spa->spa_compatibility == NULL); |
3440 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMPATIBILITY, | |
3441 | &compatibility) == 0) | |
3442 | spa->spa_compatibility = spa_strdup(compatibility); | |
3443 | ||
6cb8e530 PZ |
3444 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
3445 | &spa->spa_config_txg); | |
3446 | ||
3447 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) | |
3448 | spa->spa_config_splitting = fnvlist_dup(nvl); | |
428870ff | 3449 | |
4a0ee12a PZ |
3450 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { |
3451 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3452 | ZPOOL_CONFIG_VDEV_TREE); | |
2e528b49 | 3453 | return (SET_ERROR(EINVAL)); |
4a0ee12a | 3454 | } |
428870ff | 3455 | |
428870ff BB |
3456 | /* |
3457 | * Create "The Godfather" zio to hold all async IOs | |
3458 | */ | |
e022864d MA |
3459 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
3460 | KM_SLEEP); | |
1c27024e | 3461 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
3462 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
3463 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
3464 | ZIO_FLAG_GODFATHER); | |
3465 | } | |
428870ff BB |
3466 | |
3467 | /* | |
3468 | * Parse the configuration into a vdev tree. We explicitly set the | |
3469 | * value that will be returned by spa_version() since parsing the | |
3470 | * configuration requires knowing the version number. | |
3471 | */ | |
3472 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 PZ |
3473 | parse = (type == SPA_IMPORT_EXISTING ? |
3474 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
9eb7b46e | 3475 | error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); |
428870ff BB |
3476 | spa_config_exit(spa, SCL_ALL, FTAG); |
3477 | ||
4a0ee12a PZ |
3478 | if (error != 0) { |
3479 | spa_load_failed(spa, "unable to parse config [error=%d]", | |
3480 | error); | |
428870ff | 3481 | return (error); |
4a0ee12a | 3482 | } |
428870ff BB |
3483 | |
3484 | ASSERT(spa->spa_root_vdev == rvd); | |
c3520e7f MA |
3485 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); |
3486 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
428870ff BB |
3487 | |
3488 | if (type != SPA_IMPORT_ASSEMBLE) { | |
3489 | ASSERT(spa_guid(spa) == pool_guid); | |
3490 | } | |
3491 | ||
9eb7b46e PZ |
3492 | return (0); |
3493 | } | |
3494 | ||
6cb8e530 PZ |
3495 | /* |
3496 | * Recursively open all vdevs in the vdev tree. This function is called twice: | |
3497 | * first with the untrusted config, then with the trusted config. | |
3498 | */ | |
9eb7b46e PZ |
3499 | static int |
3500 | spa_ld_open_vdevs(spa_t *spa) | |
3501 | { | |
3502 | int error = 0; | |
3503 | ||
6cb8e530 PZ |
3504 | /* |
3505 | * spa_missing_tvds_allowed defines how many top-level vdevs can be | |
3506 | * missing/unopenable for the root vdev to be still considered openable. | |
3507 | */ | |
3508 | if (spa->spa_trust_config) { | |
3509 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; | |
3510 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { | |
3511 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; | |
3512 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { | |
3513 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; | |
3514 | } else { | |
3515 | spa->spa_missing_tvds_allowed = 0; | |
3516 | } | |
3517 | ||
3518 | spa->spa_missing_tvds_allowed = | |
3519 | MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); | |
3520 | ||
428870ff | 3521 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9eb7b46e | 3522 | error = vdev_open(spa->spa_root_vdev); |
428870ff | 3523 | spa_config_exit(spa, SCL_ALL, FTAG); |
6cb8e530 PZ |
3524 | |
3525 | if (spa->spa_missing_tvds != 0) { | |
3526 | spa_load_note(spa, "vdev tree has %lld missing top-level " | |
3527 | "vdevs.", (u_longlong_t)spa->spa_missing_tvds); | |
da92d5cb | 3528 | if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) { |
6cb8e530 PZ |
3529 | /* |
3530 | * Although theoretically we could allow users to open | |
3531 | * incomplete pools in RW mode, we'd need to add a lot | |
3532 | * of extra logic (e.g. adjust pool space to account | |
3533 | * for missing vdevs). | |
3534 | * This limitation also prevents users from accidentally | |
3535 | * opening the pool in RW mode during data recovery and | |
3536 | * damaging it further. | |
3537 | */ | |
3538 | spa_load_note(spa, "pools with missing top-level " | |
3539 | "vdevs can only be opened in read-only mode."); | |
3540 | error = SET_ERROR(ENXIO); | |
3541 | } else { | |
3542 | spa_load_note(spa, "current settings allow for maximum " | |
3543 | "%lld missing top-level vdevs at this stage.", | |
3544 | (u_longlong_t)spa->spa_missing_tvds_allowed); | |
3545 | } | |
3546 | } | |
4a0ee12a PZ |
3547 | if (error != 0) { |
3548 | spa_load_failed(spa, "unable to open vdev tree [error=%d]", | |
3549 | error); | |
3550 | } | |
6cb8e530 PZ |
3551 | if (spa->spa_missing_tvds != 0 || error != 0) |
3552 | vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); | |
9eb7b46e PZ |
3553 | |
3554 | return (error); | |
3555 | } | |
3556 | ||
6cb8e530 PZ |
3557 | /* |
3558 | * We need to validate the vdev labels against the configuration that | |
3559 | * we have in hand. This function is called twice: first with an untrusted | |
3560 | * config, then with a trusted config. The validation is more strict when the | |
3561 | * config is trusted. | |
3562 | */ | |
9eb7b46e | 3563 | static int |
6cb8e530 | 3564 | spa_ld_validate_vdevs(spa_t *spa) |
9eb7b46e PZ |
3565 | { |
3566 | int error = 0; | |
3567 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 3568 | |
6cb8e530 PZ |
3569 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
3570 | error = vdev_validate(rvd); | |
3571 | spa_config_exit(spa, SCL_ALL, FTAG); | |
428870ff | 3572 | |
6cb8e530 PZ |
3573 | if (error != 0) { |
3574 | spa_load_failed(spa, "vdev_validate failed [error=%d]", error); | |
3575 | return (error); | |
3576 | } | |
428870ff | 3577 | |
6cb8e530 PZ |
3578 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
3579 | spa_load_failed(spa, "cannot open vdev tree after invalidating " | |
3580 | "some vdevs"); | |
3581 | vdev_dbgmsg_print_tree(rvd, 2); | |
3582 | return (SET_ERROR(ENXIO)); | |
428870ff BB |
3583 | } |
3584 | ||
9eb7b46e PZ |
3585 | return (0); |
3586 | } | |
3587 | ||
d2734cce SD |
3588 | static void |
3589 | spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) | |
3590 | { | |
3591 | spa->spa_state = POOL_STATE_ACTIVE; | |
3592 | spa->spa_ubsync = spa->spa_uberblock; | |
3593 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
3594 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
3595 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
3596 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
3597 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
3598 | spa->spa_prev_software_version = ub->ub_software_version; | |
3599 | } | |
3600 | ||
9eb7b46e | 3601 | static int |
6cb8e530 | 3602 | spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) |
9eb7b46e PZ |
3603 | { |
3604 | vdev_t *rvd = spa->spa_root_vdev; | |
3605 | nvlist_t *label; | |
3606 | uberblock_t *ub = &spa->spa_uberblock; | |
9eb7b46e PZ |
3607 | boolean_t activity_check = B_FALSE; |
3608 | ||
d2734cce SD |
3609 | /* |
3610 | * If we are opening the checkpointed state of the pool by | |
3611 | * rewinding to it, at this point we will have written the | |
3612 | * checkpointed uberblock to the vdev labels, so searching | |
3613 | * the labels will find the right uberblock. However, if | |
3614 | * we are opening the checkpointed state read-only, we have | |
3615 | * not modified the labels. Therefore, we must ignore the | |
3616 | * labels and continue using the spa_uberblock that was set | |
3617 | * by spa_ld_checkpoint_rewind. | |
3618 | * | |
3619 | * Note that it would be fine to ignore the labels when | |
3620 | * rewinding (opening writeable) as well. However, if we | |
3621 | * crash just after writing the labels, we will end up | |
3622 | * searching the labels. Doing so in the common case means | |
3623 | * that this code path gets exercised normally, rather than | |
3624 | * just in the edge case. | |
3625 | */ | |
3626 | if (ub->ub_checkpoint_txg != 0 && | |
3627 | spa_importing_readonly_checkpoint(spa)) { | |
3628 | spa_ld_select_uberblock_done(spa, ub); | |
3629 | return (0); | |
3630 | } | |
3631 | ||
428870ff BB |
3632 | /* |
3633 | * Find the best uberblock. | |
3634 | */ | |
9ae529ec | 3635 | vdev_uberblock_load(rvd, ub, &label); |
428870ff BB |
3636 | |
3637 | /* | |
3638 | * If we weren't able to find a single valid uberblock, return failure. | |
3639 | */ | |
9ae529ec CS |
3640 | if (ub->ub_txg == 0) { |
3641 | nvlist_free(label); | |
4a0ee12a | 3642 | spa_load_failed(spa, "no valid uberblock found"); |
428870ff | 3643 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); |
9ae529ec | 3644 | } |
428870ff | 3645 | |
ca95f70d OF |
3646 | if (spa->spa_load_max_txg != UINT64_MAX) { |
3647 | (void) spa_import_progress_set_max_txg(spa_guid(spa), | |
3648 | (u_longlong_t)spa->spa_load_max_txg); | |
3649 | } | |
4a0ee12a PZ |
3650 | spa_load_note(spa, "using uberblock with txg=%llu", |
3651 | (u_longlong_t)ub->ub_txg); | |
3652 | ||
3653 | ||
379ca9cf OF |
3654 | /* |
3655 | * For pools which have the multihost property on determine if the | |
3656 | * pool is truly inactive and can be safely imported. Prevent | |
3657 | * hosts which don't have a hostid set from importing the pool. | |
3658 | */ | |
6cb8e530 PZ |
3659 | activity_check = spa_activity_check_required(spa, ub, label, |
3660 | spa->spa_config); | |
379ca9cf | 3661 | if (activity_check) { |
379ca9cf | 3662 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && |
25f06d67 | 3663 | spa_get_hostid(spa) == 0) { |
379ca9cf OF |
3664 | nvlist_free(label); |
3665 | fnvlist_add_uint64(spa->spa_load_info, | |
3666 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
3667 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
3668 | } | |
3669 | ||
6cb8e530 | 3670 | int error = spa_activity_check(spa, ub, spa->spa_config); |
e889f0f5 OF |
3671 | if (error) { |
3672 | nvlist_free(label); | |
3673 | return (error); | |
3674 | } | |
3675 | ||
379ca9cf OF |
3676 | fnvlist_add_uint64(spa->spa_load_info, |
3677 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
3678 | fnvlist_add_uint64(spa->spa_load_info, | |
3679 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
060f0226 OF |
3680 | fnvlist_add_uint16(spa->spa_load_info, |
3681 | ZPOOL_CONFIG_MMP_SEQ, | |
3682 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)); | |
379ca9cf OF |
3683 | } |
3684 | ||
428870ff | 3685 | /* |
9ae529ec | 3686 | * If the pool has an unsupported version we can't open it. |
428870ff | 3687 | */ |
9ae529ec CS |
3688 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { |
3689 | nvlist_free(label); | |
4a0ee12a PZ |
3690 | spa_load_failed(spa, "version %llu is not supported", |
3691 | (u_longlong_t)ub->ub_version); | |
428870ff | 3692 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); |
9ae529ec CS |
3693 | } |
3694 | ||
3695 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
3696 | nvlist_t *features; | |
3697 | ||
3698 | /* | |
3699 | * If we weren't able to find what's necessary for reading the | |
3700 | * MOS in the label, return failure. | |
3701 | */ | |
4a0ee12a PZ |
3702 | if (label == NULL) { |
3703 | spa_load_failed(spa, "label config unavailable"); | |
3704 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
3705 | ENXIO)); | |
3706 | } | |
3707 | ||
3708 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, | |
3709 | &features) != 0) { | |
9ae529ec | 3710 | nvlist_free(label); |
4a0ee12a PZ |
3711 | spa_load_failed(spa, "invalid label: '%s' missing", |
3712 | ZPOOL_CONFIG_FEATURES_FOR_READ); | |
9ae529ec CS |
3713 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
3714 | ENXIO)); | |
3715 | } | |
3716 | ||
3717 | /* | |
3718 | * Update our in-core representation with the definitive values | |
3719 | * from the label. | |
3720 | */ | |
3721 | nvlist_free(spa->spa_label_features); | |
65ad5d11 | 3722 | spa->spa_label_features = fnvlist_dup(features); |
9ae529ec CS |
3723 | } |
3724 | ||
3725 | nvlist_free(label); | |
3726 | ||
3727 | /* | |
3728 | * Look through entries in the label nvlist's features_for_read. If | |
3729 | * there is a feature listed there which we don't understand then we | |
3730 | * cannot open a pool. | |
3731 | */ | |
3732 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
3733 | nvlist_t *unsup_feat; | |
9ae529ec | 3734 | |
65ad5d11 | 3735 | unsup_feat = fnvlist_alloc(); |
9ae529ec | 3736 | |
1c27024e DB |
3737 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, |
3738 | NULL); nvp != NULL; | |
9ae529ec CS |
3739 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { |
3740 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
65ad5d11 AJ |
3741 | fnvlist_add_string(unsup_feat, |
3742 | nvpair_name(nvp), ""); | |
9ae529ec CS |
3743 | } |
3744 | } | |
3745 | ||
3746 | if (!nvlist_empty(unsup_feat)) { | |
65ad5d11 AJ |
3747 | fnvlist_add_nvlist(spa->spa_load_info, |
3748 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec | 3749 | nvlist_free(unsup_feat); |
4a0ee12a | 3750 | spa_load_failed(spa, "some features are unsupported"); |
9ae529ec CS |
3751 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
3752 | ENOTSUP)); | |
3753 | } | |
3754 | ||
3755 | nvlist_free(unsup_feat); | |
3756 | } | |
428870ff | 3757 | |
428870ff BB |
3758 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { |
3759 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 | 3760 | spa_try_repair(spa, spa->spa_config); |
428870ff BB |
3761 | spa_config_exit(spa, SCL_ALL, FTAG); |
3762 | nvlist_free(spa->spa_config_splitting); | |
3763 | spa->spa_config_splitting = NULL; | |
3764 | } | |
3765 | ||
3766 | /* | |
3767 | * Initialize internal SPA structures. | |
3768 | */ | |
d2734cce | 3769 | spa_ld_select_uberblock_done(spa, ub); |
428870ff | 3770 | |
9eb7b46e PZ |
3771 | return (0); |
3772 | } | |
3773 | ||
3774 | static int | |
3775 | spa_ld_open_rootbp(spa_t *spa) | |
3776 | { | |
3777 | int error = 0; | |
3778 | vdev_t *rvd = spa->spa_root_vdev; | |
a1d477c2 | 3779 | |
9ae529ec | 3780 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
4a0ee12a PZ |
3781 | if (error != 0) { |
3782 | spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " | |
3783 | "[error=%d]", error); | |
428870ff | 3784 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 3785 | } |
428870ff BB |
3786 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
3787 | ||
9eb7b46e PZ |
3788 | return (0); |
3789 | } | |
3790 | ||
3791 | static int | |
d2734cce | 3792 | spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, |
6cb8e530 | 3793 | boolean_t reloading) |
9eb7b46e | 3794 | { |
6cb8e530 PZ |
3795 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; |
3796 | nvlist_t *nv, *mos_config, *policy; | |
3797 | int error = 0, copy_error; | |
3798 | uint64_t healthy_tvds, healthy_tvds_mos; | |
3799 | uint64_t mos_config_txg; | |
9eb7b46e | 3800 | |
4a0ee12a PZ |
3801 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) |
3802 | != 0) | |
428870ff BB |
3803 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
3804 | ||
a1d477c2 | 3805 | /* |
6cb8e530 PZ |
3806 | * If we're assembling a pool from a split, the config provided is |
3807 | * already trusted so there is nothing to do. | |
a1d477c2 | 3808 | */ |
6cb8e530 PZ |
3809 | if (type == SPA_IMPORT_ASSEMBLE) |
3810 | return (0); | |
3811 | ||
3812 | healthy_tvds = spa_healthy_core_tvds(spa); | |
a1d477c2 | 3813 | |
6cb8e530 PZ |
3814 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) |
3815 | != 0) { | |
3816 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
3817 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3818 | } | |
3819 | ||
3820 | /* | |
3821 | * If we are doing an open, pool owner wasn't verified yet, thus do | |
3822 | * the verification here. | |
3823 | */ | |
3824 | if (spa->spa_load_state == SPA_LOAD_OPEN) { | |
3825 | error = spa_verify_host(spa, mos_config); | |
3826 | if (error != 0) { | |
a1d477c2 | 3827 | nvlist_free(mos_config); |
6cb8e530 | 3828 | return (error); |
a1d477c2 | 3829 | } |
6cb8e530 PZ |
3830 | } |
3831 | ||
3832 | nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); | |
a1d477c2 | 3833 | |
6cb8e530 PZ |
3834 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
3835 | ||
3836 | /* | |
3837 | * Build a new vdev tree from the trusted config | |
3838 | */ | |
b2255edc BB |
3839 | error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD); |
3840 | if (error != 0) { | |
3841 | nvlist_free(mos_config); | |
3842 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3843 | spa_load_failed(spa, "spa_config_parse failed [error=%d]", | |
3844 | error); | |
3845 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
3846 | } | |
6cb8e530 PZ |
3847 | |
3848 | /* | |
3849 | * Vdev paths in the MOS may be obsolete. If the untrusted config was | |
3850 | * obtained by scanning /dev/dsk, then it will have the right vdev | |
3851 | * paths. We update the trusted MOS config with this information. | |
3852 | * We first try to copy the paths with vdev_copy_path_strict, which | |
3853 | * succeeds only when both configs have exactly the same vdev tree. | |
3854 | * If that fails, we fall back to a more flexible method that has a | |
3855 | * best effort policy. | |
3856 | */ | |
3857 | copy_error = vdev_copy_path_strict(rvd, mrvd); | |
3858 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
3859 | spa_load_note(spa, "provided vdev tree:"); | |
3860 | vdev_dbgmsg_print_tree(rvd, 2); | |
3861 | spa_load_note(spa, "MOS vdev tree:"); | |
3862 | vdev_dbgmsg_print_tree(mrvd, 2); | |
3863 | } | |
3864 | if (copy_error != 0) { | |
3865 | spa_load_note(spa, "vdev_copy_path_strict failed, falling " | |
3866 | "back to vdev_copy_path_relaxed"); | |
3867 | vdev_copy_path_relaxed(rvd, mrvd); | |
3868 | } | |
3869 | ||
3870 | vdev_close(rvd); | |
3871 | vdev_free(rvd); | |
3872 | spa->spa_root_vdev = mrvd; | |
3873 | rvd = mrvd; | |
3874 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3875 | ||
3876 | /* | |
3877 | * We will use spa_config if we decide to reload the spa or if spa_load | |
3878 | * fails and we rewind. We must thus regenerate the config using the | |
8a393be3 PZ |
3879 | * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to |
3880 | * pass settings on how to load the pool and is not stored in the MOS. | |
3881 | * We copy it over to our new, trusted config. | |
6cb8e530 PZ |
3882 | */ |
3883 | mos_config_txg = fnvlist_lookup_uint64(mos_config, | |
3884 | ZPOOL_CONFIG_POOL_TXG); | |
3885 | nvlist_free(mos_config); | |
3886 | mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); | |
8a393be3 | 3887 | if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, |
6cb8e530 | 3888 | &policy) == 0) |
8a393be3 | 3889 | fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); |
6cb8e530 PZ |
3890 | spa_config_set(spa, mos_config); |
3891 | spa->spa_config_source = SPA_CONFIG_SRC_MOS; | |
3892 | ||
3893 | /* | |
3894 | * Now that we got the config from the MOS, we should be more strict | |
3895 | * in checking blkptrs and can make assumptions about the consistency | |
3896 | * of the vdev tree. spa_trust_config must be set to true before opening | |
3897 | * vdevs in order for them to be writeable. | |
3898 | */ | |
3899 | spa->spa_trust_config = B_TRUE; | |
3900 | ||
3901 | /* | |
3902 | * Open and validate the new vdev tree | |
3903 | */ | |
3904 | error = spa_ld_open_vdevs(spa); | |
3905 | if (error != 0) | |
3906 | return (error); | |
3907 | ||
3908 | error = spa_ld_validate_vdevs(spa); | |
3909 | if (error != 0) | |
3910 | return (error); | |
3911 | ||
3912 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
3913 | spa_load_note(spa, "final vdev tree:"); | |
3914 | vdev_dbgmsg_print_tree(rvd, 2); | |
3915 | } | |
3916 | ||
3917 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && | |
3918 | !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { | |
a1d477c2 | 3919 | /* |
6cb8e530 PZ |
3920 | * Sanity check to make sure that we are indeed loading the |
3921 | * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds | |
3922 | * in the config provided and they happened to be the only ones | |
3923 | * to have the latest uberblock, we could involuntarily perform | |
3924 | * an extreme rewind. | |
a1d477c2 | 3925 | */ |
6cb8e530 PZ |
3926 | healthy_tvds_mos = spa_healthy_core_tvds(spa); |
3927 | if (healthy_tvds_mos - healthy_tvds >= | |
3928 | SPA_SYNC_MIN_VDEVS) { | |
3929 | spa_load_note(spa, "config provided misses too many " | |
3930 | "top-level vdevs compared to MOS (%lld vs %lld). ", | |
3931 | (u_longlong_t)healthy_tvds, | |
3932 | (u_longlong_t)healthy_tvds_mos); | |
3933 | spa_load_note(spa, "vdev tree:"); | |
3934 | vdev_dbgmsg_print_tree(rvd, 2); | |
3935 | if (reloading) { | |
3936 | spa_load_failed(spa, "config was already " | |
3937 | "provided from MOS. Aborting."); | |
3938 | return (spa_vdev_err(rvd, | |
3939 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
3940 | } | |
3941 | spa_load_note(spa, "spa must be reloaded using MOS " | |
3942 | "config"); | |
3943 | return (SET_ERROR(EAGAIN)); | |
4a0ee12a | 3944 | } |
a1d477c2 MA |
3945 | } |
3946 | ||
6cb8e530 PZ |
3947 | error = spa_check_for_missing_logs(spa); |
3948 | if (error != 0) | |
3949 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
3950 | ||
3951 | if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { | |
3952 | spa_load_failed(spa, "uberblock guid sum doesn't match MOS " | |
3953 | "guid sum (%llu != %llu)", | |
3954 | (u_longlong_t)spa->spa_uberblock.ub_guid_sum, | |
3955 | (u_longlong_t)rvd->vdev_guid_sum); | |
3956 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
3957 | ENXIO)); | |
3958 | } | |
3959 | ||
9eb7b46e PZ |
3960 | return (0); |
3961 | } | |
3962 | ||
3963 | static int | |
3964 | spa_ld_open_indirect_vdev_metadata(spa_t *spa) | |
3965 | { | |
3966 | int error = 0; | |
3967 | vdev_t *rvd = spa->spa_root_vdev; | |
3968 | ||
a1d477c2 MA |
3969 | /* |
3970 | * Everything that we read before spa_remove_init() must be stored | |
3971 | * on concreted vdevs. Therefore we do this as early as possible. | |
3972 | */ | |
4a0ee12a PZ |
3973 | error = spa_remove_init(spa); |
3974 | if (error != 0) { | |
3975 | spa_load_failed(spa, "spa_remove_init failed [error=%d]", | |
3976 | error); | |
a1d477c2 | 3977 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 3978 | } |
a1d477c2 | 3979 | |
9eb7b46e PZ |
3980 | /* |
3981 | * Retrieve information needed to condense indirect vdev mappings. | |
3982 | */ | |
3983 | error = spa_condense_init(spa); | |
3984 | if (error != 0) { | |
4a0ee12a PZ |
3985 | spa_load_failed(spa, "spa_condense_init failed [error=%d]", |
3986 | error); | |
9eb7b46e PZ |
3987 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
3988 | } | |
3989 | ||
3990 | return (0); | |
3991 | } | |
3992 | ||
3993 | static int | |
4a0ee12a | 3994 | spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) |
9eb7b46e PZ |
3995 | { |
3996 | int error = 0; | |
3997 | vdev_t *rvd = spa->spa_root_vdev; | |
3998 | ||
9ae529ec CS |
3999 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { |
4000 | boolean_t missing_feat_read = B_FALSE; | |
b9b24bb4 | 4001 | nvlist_t *unsup_feat, *enabled_feat; |
9ae529ec CS |
4002 | |
4003 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
4a0ee12a | 4004 | &spa->spa_feat_for_read_obj, B_TRUE) != 0) { |
9ae529ec CS |
4005 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4006 | } | |
4007 | ||
4008 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
4a0ee12a | 4009 | &spa->spa_feat_for_write_obj, B_TRUE) != 0) { |
9ae529ec CS |
4010 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4011 | } | |
4012 | ||
4013 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
4a0ee12a | 4014 | &spa->spa_feat_desc_obj, B_TRUE) != 0) { |
9ae529ec CS |
4015 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4016 | } | |
4017 | ||
b9b24bb4 CS |
4018 | enabled_feat = fnvlist_alloc(); |
4019 | unsup_feat = fnvlist_alloc(); | |
9ae529ec | 4020 | |
fa86b5db | 4021 | if (!spa_features_check(spa, B_FALSE, |
b9b24bb4 | 4022 | unsup_feat, enabled_feat)) |
9ae529ec CS |
4023 | missing_feat_read = B_TRUE; |
4024 | ||
4a0ee12a PZ |
4025 | if (spa_writeable(spa) || |
4026 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) { | |
fa86b5db | 4027 | if (!spa_features_check(spa, B_TRUE, |
b9b24bb4 | 4028 | unsup_feat, enabled_feat)) { |
9eb7b46e | 4029 | *missing_feat_writep = B_TRUE; |
b9b24bb4 | 4030 | } |
9ae529ec CS |
4031 | } |
4032 | ||
b9b24bb4 CS |
4033 | fnvlist_add_nvlist(spa->spa_load_info, |
4034 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
4035 | ||
9ae529ec | 4036 | if (!nvlist_empty(unsup_feat)) { |
b9b24bb4 CS |
4037 | fnvlist_add_nvlist(spa->spa_load_info, |
4038 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec CS |
4039 | } |
4040 | ||
b9b24bb4 CS |
4041 | fnvlist_free(enabled_feat); |
4042 | fnvlist_free(unsup_feat); | |
9ae529ec CS |
4043 | |
4044 | if (!missing_feat_read) { | |
4045 | fnvlist_add_boolean(spa->spa_load_info, | |
4046 | ZPOOL_CONFIG_CAN_RDONLY); | |
4047 | } | |
4048 | ||
4049 | /* | |
4050 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
4051 | * twofold: to determine whether the pool is available for | |
4052 | * import in read-write mode and (if it is not) whether the | |
4053 | * pool is available for import in read-only mode. If the pool | |
4054 | * is available for import in read-write mode, it is displayed | |
4055 | * as available in userland; if it is not available for import | |
4056 | * in read-only mode, it is displayed as unavailable in | |
4057 | * userland. If the pool is available for import in read-only | |
4058 | * mode but not read-write mode, it is displayed as unavailable | |
4059 | * in userland with a special note that the pool is actually | |
4060 | * available for open in read-only mode. | |
4061 | * | |
4062 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
4063 | * missing a feature for write, we must first determine whether | |
4064 | * the pool can be opened read-only before returning to | |
4065 | * userland in order to know whether to display the | |
4066 | * abovementioned note. | |
4067 | */ | |
9eb7b46e | 4068 | if (missing_feat_read || (*missing_feat_writep && |
9ae529ec | 4069 | spa_writeable(spa))) { |
4a0ee12a | 4070 | spa_load_failed(spa, "pool uses unsupported features"); |
9ae529ec CS |
4071 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
4072 | ENOTSUP)); | |
4073 | } | |
b0bc7a84 MG |
4074 | |
4075 | /* | |
4076 | * Load refcounts for ZFS features from disk into an in-memory | |
4077 | * cache during SPA initialization. | |
4078 | */ | |
1c27024e | 4079 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { |
b0bc7a84 MG |
4080 | uint64_t refcount; |
4081 | ||
4082 | error = feature_get_refcount_from_disk(spa, | |
4083 | &spa_feature_table[i], &refcount); | |
4084 | if (error == 0) { | |
4085 | spa->spa_feat_refcount_cache[i] = refcount; | |
4086 | } else if (error == ENOTSUP) { | |
4087 | spa->spa_feat_refcount_cache[i] = | |
4088 | SPA_FEATURE_DISABLED; | |
4089 | } else { | |
4a0ee12a PZ |
4090 | spa_load_failed(spa, "error getting refcount " |
4091 | "for feature %s [error=%d]", | |
4092 | spa_feature_table[i].fi_guid, error); | |
b0bc7a84 MG |
4093 | return (spa_vdev_err(rvd, |
4094 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4095 | } | |
4096 | } | |
4097 | } | |
4098 | ||
4099 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
4100 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
4a0ee12a | 4101 | &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) |
b0bc7a84 | 4102 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
9ae529ec CS |
4103 | } |
4104 | ||
f00ab3f2 TC |
4105 | /* |
4106 | * Encryption was added before bookmark_v2, even though bookmark_v2 | |
4107 | * is now a dependency. If this pool has encryption enabled without | |
4108 | * bookmark_v2, trigger an errata message. | |
4109 | */ | |
4110 | if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) && | |
4111 | !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) { | |
4112 | spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; | |
4113 | } | |
4114 | ||
9eb7b46e PZ |
4115 | return (0); |
4116 | } | |
4117 | ||
4118 | static int | |
4119 | spa_ld_load_special_directories(spa_t *spa) | |
4120 | { | |
4121 | int error = 0; | |
4122 | vdev_t *rvd = spa->spa_root_vdev; | |
4123 | ||
9ae529ec CS |
4124 | spa->spa_is_initializing = B_TRUE; |
4125 | error = dsl_pool_open(spa->spa_dsl_pool); | |
4126 | spa->spa_is_initializing = B_FALSE; | |
4a0ee12a PZ |
4127 | if (error != 0) { |
4128 | spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); | |
9ae529ec | 4129 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4130 | } |
9ae529ec | 4131 | |
9eb7b46e PZ |
4132 | return (0); |
4133 | } | |
428870ff | 4134 | |
9eb7b46e PZ |
4135 | static int |
4136 | spa_ld_get_props(spa_t *spa) | |
4137 | { | |
4138 | int error = 0; | |
4139 | uint64_t obj; | |
4140 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4141 | |
3c67d83a TH |
4142 | /* Grab the checksum salt from the MOS. */ |
4143 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4144 | DMU_POOL_CHECKSUM_SALT, 1, | |
4145 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
4146 | spa->spa_cksum_salt.zcs_bytes); | |
4147 | if (error == ENOENT) { | |
4148 | /* Generate a new salt for subsequent use */ | |
4149 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4150 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4151 | } else if (error != 0) { | |
4a0ee12a PZ |
4152 | spa_load_failed(spa, "unable to retrieve checksum salt from " |
4153 | "MOS [error=%d]", error); | |
3c67d83a TH |
4154 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4155 | } | |
4156 | ||
4a0ee12a | 4157 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) |
428870ff BB |
4158 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4159 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
4a0ee12a PZ |
4160 | if (error != 0) { |
4161 | spa_load_failed(spa, "error opening deferred-frees bpobj " | |
4162 | "[error=%d]", error); | |
428870ff | 4163 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4164 | } |
34dc7c2f BB |
4165 | |
4166 | /* | |
4167 | * Load the bit that tells us to use the new accounting function | |
4168 | * (raid-z deflation). If we have an older pool, this will not | |
4169 | * be present. | |
4170 | */ | |
4a0ee12a | 4171 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); |
428870ff BB |
4172 | if (error != 0 && error != ENOENT) |
4173 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4174 | ||
4175 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
4a0ee12a | 4176 | &spa->spa_creation_version, B_FALSE); |
428870ff BB |
4177 | if (error != 0 && error != ENOENT) |
4178 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
4179 | |
4180 | /* | |
4181 | * Load the persistent error log. If we have an older pool, this will | |
4182 | * not be present. | |
4183 | */ | |
4a0ee12a PZ |
4184 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, |
4185 | B_FALSE); | |
428870ff BB |
4186 | if (error != 0 && error != ENOENT) |
4187 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4188 | |
428870ff | 4189 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, |
4a0ee12a | 4190 | &spa->spa_errlog_scrub, B_FALSE); |
428870ff BB |
4191 | if (error != 0 && error != ENOENT) |
4192 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4193 | |
37f03da8 SH |
4194 | /* |
4195 | * Load the livelist deletion field. If a livelist is queued for | |
4196 | * deletion, indicate that in the spa | |
4197 | */ | |
4198 | error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES, | |
4199 | &spa->spa_livelists_to_delete, B_FALSE); | |
4200 | if (error != 0 && error != ENOENT) | |
4201 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4202 | ||
34dc7c2f BB |
4203 | /* |
4204 | * Load the history object. If we have an older pool, this | |
4205 | * will not be present. | |
4206 | */ | |
4a0ee12a | 4207 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); |
428870ff BB |
4208 | if (error != 0 && error != ENOENT) |
4209 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4210 | ||
e0ab3ab5 JS |
4211 | /* |
4212 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
4213 | * be present; in this case, defer its creation to a later time to | |
4214 | * avoid dirtying the MOS this early / out of sync context. See | |
4215 | * spa_sync_config_object. | |
4216 | */ | |
4217 | ||
4218 | /* The sentinel is only available in the MOS config. */ | |
1c27024e | 4219 | nvlist_t *mos_config; |
4a0ee12a PZ |
4220 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { |
4221 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
e0ab3ab5 | 4222 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4223 | } |
e0ab3ab5 JS |
4224 | |
4225 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
4a0ee12a | 4226 | &spa->spa_all_vdev_zaps, B_FALSE); |
e0ab3ab5 | 4227 | |
38640550 DB |
4228 | if (error == ENOENT) { |
4229 | VERIFY(!nvlist_exists(mos_config, | |
4230 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
4231 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
4232 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4233 | } else if (error != 0) { | |
cb01da68 | 4234 | nvlist_free(mos_config); |
e0ab3ab5 | 4235 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
38640550 | 4236 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { |
e0ab3ab5 JS |
4237 | /* |
4238 | * An older version of ZFS overwrote the sentinel value, so | |
4239 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
4240 | * destruction to later; see spa_sync_config_object. | |
4241 | */ | |
4242 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
4243 | /* | |
4244 | * We're assuming that no vdevs have had their ZAPs created | |
4245 | * before this. Better be sure of it. | |
4246 | */ | |
4247 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4248 | } | |
4249 | nvlist_free(mos_config); | |
4250 | ||
9eb7b46e PZ |
4251 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); |
4252 | ||
4a0ee12a PZ |
4253 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, |
4254 | B_FALSE); | |
9eb7b46e PZ |
4255 | if (error && error != ENOENT) |
4256 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4257 | ||
4258 | if (error == 0) { | |
da27b8bc | 4259 | uint64_t autoreplace = 0; |
9eb7b46e PZ |
4260 | |
4261 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
4262 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
4263 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
4264 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
4265 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
c02c1bec | 4266 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); |
1b939560 | 4267 | spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim); |
9eb7b46e PZ |
4268 | spa->spa_autoreplace = (autoreplace != 0); |
4269 | } | |
4270 | ||
6cb8e530 PZ |
4271 | /* |
4272 | * If we are importing a pool with missing top-level vdevs, | |
4273 | * we enforce that the pool doesn't panic or get suspended on | |
4274 | * error since the likelihood of missing data is extremely high. | |
4275 | */ | |
4276 | if (spa->spa_missing_tvds > 0 && | |
4277 | spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && | |
4278 | spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4279 | spa_load_note(spa, "forcing failmode to 'continue' " | |
4280 | "as some top level vdevs are missing"); | |
4281 | spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; | |
4282 | } | |
4283 | ||
9eb7b46e PZ |
4284 | return (0); |
4285 | } | |
4286 | ||
4287 | static int | |
4288 | spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) | |
4289 | { | |
4290 | int error = 0; | |
4291 | vdev_t *rvd = spa->spa_root_vdev; | |
4292 | ||
428870ff BB |
4293 | /* |
4294 | * If we're assembling the pool from the split-off vdevs of | |
4295 | * an existing pool, we don't want to attach the spares & cache | |
4296 | * devices. | |
4297 | */ | |
34dc7c2f BB |
4298 | |
4299 | /* | |
4300 | * Load any hot spares for this pool. | |
4301 | */ | |
4a0ee12a PZ |
4302 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, |
4303 | B_FALSE); | |
428870ff BB |
4304 | if (error != 0 && error != ENOENT) |
4305 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4306 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4307 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); |
4308 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
4a0ee12a PZ |
4309 | &spa->spa_spares.sav_config) != 0) { |
4310 | spa_load_failed(spa, "error loading spares nvlist"); | |
428870ff | 4311 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4312 | } |
34dc7c2f | 4313 | |
b128c09f | 4314 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4315 | spa_load_spares(spa); |
b128c09f | 4316 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4317 | } else if (error == 0) { |
4318 | spa->spa_spares.sav_sync = B_TRUE; | |
34dc7c2f BB |
4319 | } |
4320 | ||
4321 | /* | |
4322 | * Load any level 2 ARC devices for this pool. | |
4323 | */ | |
428870ff | 4324 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, |
4a0ee12a | 4325 | &spa->spa_l2cache.sav_object, B_FALSE); |
428870ff BB |
4326 | if (error != 0 && error != ENOENT) |
4327 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4328 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4329 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); |
4330 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
4a0ee12a PZ |
4331 | &spa->spa_l2cache.sav_config) != 0) { |
4332 | spa_load_failed(spa, "error loading l2cache nvlist"); | |
428870ff | 4333 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4334 | } |
34dc7c2f | 4335 | |
b128c09f | 4336 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4337 | spa_load_l2cache(spa); |
b128c09f | 4338 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4339 | } else if (error == 0) { |
4340 | spa->spa_l2cache.sav_sync = B_TRUE; | |
b128c09f BB |
4341 | } |
4342 | ||
9eb7b46e PZ |
4343 | return (0); |
4344 | } | |
428870ff | 4345 | |
9eb7b46e | 4346 | static int |
4a0ee12a | 4347 | spa_ld_load_vdev_metadata(spa_t *spa) |
9eb7b46e PZ |
4348 | { |
4349 | int error = 0; | |
4350 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4351 | |
379ca9cf OF |
4352 | /* |
4353 | * If the 'multihost' property is set, then never allow a pool to | |
4354 | * be imported when the system hostid is zero. The exception to | |
4355 | * this rule is zdb which is always allowed to access pools. | |
4356 | */ | |
25f06d67 | 4357 | if (spa_multihost(spa) && spa_get_hostid(spa) == 0 && |
379ca9cf OF |
4358 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { |
4359 | fnvlist_add_uint64(spa->spa_load_info, | |
4360 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4361 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4362 | } | |
4363 | ||
34dc7c2f BB |
4364 | /* |
4365 | * If the 'autoreplace' property is set, then post a resource notifying | |
4366 | * the ZFS DE that it should not issue any faults for unopenable | |
4367 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
4368 | * unopenable vdevs so that the normal autoreplace handler can take | |
4369 | * over. | |
4370 | */ | |
4a0ee12a | 4371 | if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
34dc7c2f | 4372 | spa_check_removed(spa->spa_root_vdev); |
428870ff BB |
4373 | /* |
4374 | * For the import case, this is done in spa_import(), because | |
4375 | * at this point we're using the spare definitions from | |
4376 | * the MOS config, not necessarily from the userland config. | |
4377 | */ | |
4a0ee12a | 4378 | if (spa->spa_load_state != SPA_LOAD_IMPORT) { |
428870ff BB |
4379 | spa_aux_check_removed(&spa->spa_spares); |
4380 | spa_aux_check_removed(&spa->spa_l2cache); | |
4381 | } | |
4382 | } | |
34dc7c2f BB |
4383 | |
4384 | /* | |
9eb7b46e | 4385 | * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. |
34dc7c2f | 4386 | */ |
a1d477c2 MA |
4387 | error = vdev_load(rvd); |
4388 | if (error != 0) { | |
4a0ee12a | 4389 | spa_load_failed(spa, "vdev_load failed [error=%d]", error); |
a1d477c2 MA |
4390 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
4391 | } | |
4392 | ||
93e28d66 SD |
4393 | error = spa_ld_log_spacemaps(spa); |
4394 | if (error != 0) { | |
600a02b8 | 4395 | spa_load_failed(spa, "spa_ld_log_spacemaps failed [error=%d]", |
93e28d66 SD |
4396 | error); |
4397 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4398 | } | |
4399 | ||
34dc7c2f | 4400 | /* |
9eb7b46e | 4401 | * Propagate the leaf DTLs we just loaded all the way up the vdev tree. |
34dc7c2f | 4402 | */ |
b128c09f | 4403 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9a49d3f3 | 4404 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE); |
b128c09f | 4405 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 4406 | |
9eb7b46e PZ |
4407 | return (0); |
4408 | } | |
4409 | ||
4410 | static int | |
4411 | spa_ld_load_dedup_tables(spa_t *spa) | |
4412 | { | |
4413 | int error = 0; | |
4414 | vdev_t *rvd = spa->spa_root_vdev; | |
4415 | ||
428870ff | 4416 | error = ddt_load(spa); |
4a0ee12a PZ |
4417 | if (error != 0) { |
4418 | spa_load_failed(spa, "ddt_load failed [error=%d]", error); | |
428870ff | 4419 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4420 | } |
428870ff | 4421 | |
9eb7b46e PZ |
4422 | return (0); |
4423 | } | |
4424 | ||
67a1b037 PJD |
4425 | static int |
4426 | spa_ld_load_brt(spa_t *spa) | |
4427 | { | |
4428 | int error = 0; | |
4429 | vdev_t *rvd = spa->spa_root_vdev; | |
4430 | ||
4431 | error = brt_load(spa); | |
4432 | if (error != 0) { | |
4433 | spa_load_failed(spa, "brt_load failed [error=%d]", error); | |
4434 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4435 | } | |
4436 | ||
4437 | return (0); | |
4438 | } | |
4439 | ||
9eb7b46e | 4440 | static int |
a926aab9 | 4441 | spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, const char **ereport) |
9eb7b46e PZ |
4442 | { |
4443 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 4444 | |
4a0ee12a PZ |
4445 | if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { |
4446 | boolean_t missing = spa_check_logs(spa); | |
4447 | if (missing) { | |
6cb8e530 PZ |
4448 | if (spa->spa_missing_tvds != 0) { |
4449 | spa_load_note(spa, "spa_check_logs failed " | |
4450 | "so dropping the logs"); | |
4451 | } else { | |
4452 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
4453 | spa_load_failed(spa, "spa_check_logs failed"); | |
4454 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, | |
4455 | ENXIO)); | |
4456 | } | |
4a0ee12a | 4457 | } |
428870ff BB |
4458 | } |
4459 | ||
9eb7b46e PZ |
4460 | return (0); |
4461 | } | |
4462 | ||
4463 | static int | |
4a0ee12a | 4464 | spa_ld_verify_pool_data(spa_t *spa) |
9eb7b46e PZ |
4465 | { |
4466 | int error = 0; | |
4467 | vdev_t *rvd = spa->spa_root_vdev; | |
4468 | ||
4469 | /* | |
4470 | * We've successfully opened the pool, verify that we're ready | |
4471 | * to start pushing transactions. | |
4472 | */ | |
4a0ee12a | 4473 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
9eb7b46e PZ |
4474 | error = spa_load_verify(spa); |
4475 | if (error != 0) { | |
4a0ee12a PZ |
4476 | spa_load_failed(spa, "spa_load_verify failed " |
4477 | "[error=%d]", error); | |
9eb7b46e PZ |
4478 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
4479 | error)); | |
4480 | } | |
4481 | } | |
4482 | ||
4483 | return (0); | |
4484 | } | |
4485 | ||
4486 | static void | |
4487 | spa_ld_claim_log_blocks(spa_t *spa) | |
4488 | { | |
4489 | dmu_tx_t *tx; | |
4490 | dsl_pool_t *dp = spa_get_dsl(spa); | |
4491 | ||
4492 | /* | |
4493 | * Claim log blocks that haven't been committed yet. | |
4494 | * This must all happen in a single txg. | |
4495 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
4496 | * invoked from zil_claim_log_block()'s i/o done callback. | |
4497 | * Price of rollback is that we abandon the log. | |
4498 | */ | |
4499 | spa->spa_claiming = B_TRUE; | |
4500 | ||
4501 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
4502 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
4503 | zil_claim, tx, DS_FIND_CHILDREN); | |
4504 | dmu_tx_commit(tx); | |
4505 | ||
4506 | spa->spa_claiming = B_FALSE; | |
4507 | ||
4508 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
4509 | } | |
4510 | ||
4511 | static void | |
6cb8e530 | 4512 | spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, |
d2734cce | 4513 | boolean_t update_config_cache) |
9eb7b46e PZ |
4514 | { |
4515 | vdev_t *rvd = spa->spa_root_vdev; | |
4516 | int need_update = B_FALSE; | |
4517 | ||
4518 | /* | |
4519 | * If the config cache is stale, or we have uninitialized | |
4520 | * metaslabs (see spa_vdev_add()), then update the config. | |
4521 | * | |
4522 | * If this is a verbatim import, trust the current | |
4523 | * in-core spa_config and update the disk labels. | |
4524 | */ | |
d2734cce | 4525 | if (update_config_cache || config_cache_txg != spa->spa_config_txg || |
4a0ee12a PZ |
4526 | spa->spa_load_state == SPA_LOAD_IMPORT || |
4527 | spa->spa_load_state == SPA_LOAD_RECOVER || | |
9eb7b46e PZ |
4528 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) |
4529 | need_update = B_TRUE; | |
4530 | ||
4531 | for (int c = 0; c < rvd->vdev_children; c++) | |
4532 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
4533 | need_update = B_TRUE; | |
4534 | ||
4535 | /* | |
e1cfd73f | 4536 | * Update the config cache asynchronously in case we're the |
9eb7b46e PZ |
4537 | * root pool, in which case the config cache isn't writable yet. |
4538 | */ | |
4539 | if (need_update) | |
4540 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
4541 | } | |
4542 | ||
6cb8e530 PZ |
4543 | static void |
4544 | spa_ld_prepare_for_reload(spa_t *spa) | |
4545 | { | |
da92d5cb | 4546 | spa_mode_t mode = spa->spa_mode; |
6cb8e530 PZ |
4547 | int async_suspended = spa->spa_async_suspended; |
4548 | ||
4549 | spa_unload(spa); | |
4550 | spa_deactivate(spa); | |
4551 | spa_activate(spa, mode); | |
4552 | ||
4553 | /* | |
4554 | * We save the value of spa_async_suspended as it gets reset to 0 by | |
4555 | * spa_unload(). We want to restore it back to the original value before | |
4556 | * returning as we might be calling spa_async_resume() later. | |
4557 | */ | |
4558 | spa->spa_async_suspended = async_suspended; | |
4559 | } | |
4560 | ||
9eb7b46e | 4561 | static int |
d2734cce SD |
4562 | spa_ld_read_checkpoint_txg(spa_t *spa) |
4563 | { | |
4564 | uberblock_t checkpoint; | |
4565 | int error = 0; | |
4566 | ||
4567 | ASSERT0(spa->spa_checkpoint_txg); | |
4568 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4569 | ||
4570 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4571 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
4572 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
4573 | ||
4574 | if (error == ENOENT) | |
4575 | return (0); | |
4576 | ||
4577 | if (error != 0) | |
4578 | return (error); | |
4579 | ||
4580 | ASSERT3U(checkpoint.ub_txg, !=, 0); | |
4581 | ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); | |
4582 | ASSERT3U(checkpoint.ub_timestamp, !=, 0); | |
4583 | spa->spa_checkpoint_txg = checkpoint.ub_txg; | |
4584 | spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; | |
4585 | ||
4586 | return (0); | |
4587 | } | |
4588 | ||
4589 | static int | |
4590 | spa_ld_mos_init(spa_t *spa, spa_import_type_t type) | |
9eb7b46e PZ |
4591 | { |
4592 | int error = 0; | |
9eb7b46e | 4593 | |
4a0ee12a | 4594 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
6cb8e530 | 4595 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); |
4a0ee12a | 4596 | |
9eb7b46e | 4597 | /* |
6cb8e530 PZ |
4598 | * Never trust the config that is provided unless we are assembling |
4599 | * a pool following a split. | |
4600 | * This means don't trust blkptrs and the vdev tree in general. This | |
4601 | * also effectively puts the spa in read-only mode since | |
4602 | * spa_writeable() checks for spa_trust_config to be true. | |
4603 | * We will later load a trusted config from the MOS. | |
9eb7b46e | 4604 | */ |
6cb8e530 PZ |
4605 | if (type != SPA_IMPORT_ASSEMBLE) |
4606 | spa->spa_trust_config = B_FALSE; | |
4607 | ||
9eb7b46e PZ |
4608 | /* |
4609 | * Parse the config provided to create a vdev tree. | |
4610 | */ | |
6cb8e530 | 4611 | error = spa_ld_parse_config(spa, type); |
9eb7b46e PZ |
4612 | if (error != 0) |
4613 | return (error); | |
4614 | ||
ca95f70d OF |
4615 | spa_import_progress_add(spa); |
4616 | ||
9eb7b46e PZ |
4617 | /* |
4618 | * Now that we have the vdev tree, try to open each vdev. This involves | |
4619 | * opening the underlying physical device, retrieving its geometry and | |
4620 | * probing the vdev with a dummy I/O. The state of each vdev will be set | |
4621 | * based on the success of those operations. After this we'll be ready | |
4622 | * to read from the vdevs. | |
4623 | */ | |
4624 | error = spa_ld_open_vdevs(spa); | |
4625 | if (error != 0) | |
4626 | return (error); | |
4627 | ||
4628 | /* | |
4629 | * Read the label of each vdev and make sure that the GUIDs stored | |
4630 | * there match the GUIDs in the config provided. | |
6cb8e530 PZ |
4631 | * If we're assembling a new pool that's been split off from an |
4632 | * existing pool, the labels haven't yet been updated so we skip | |
4633 | * validation for now. | |
9eb7b46e | 4634 | */ |
6cb8e530 PZ |
4635 | if (type != SPA_IMPORT_ASSEMBLE) { |
4636 | error = spa_ld_validate_vdevs(spa); | |
4637 | if (error != 0) | |
4638 | return (error); | |
4639 | } | |
9eb7b46e PZ |
4640 | |
4641 | /* | |
d2734cce SD |
4642 | * Read all vdev labels to find the best uberblock (i.e. latest, |
4643 | * unless spa_load_max_txg is set) and store it in spa_uberblock. We | |
4644 | * get the list of features required to read blkptrs in the MOS from | |
4645 | * the vdev label with the best uberblock and verify that our version | |
4646 | * of zfs supports them all. | |
9eb7b46e | 4647 | */ |
6cb8e530 | 4648 | error = spa_ld_select_uberblock(spa, type); |
9eb7b46e PZ |
4649 | if (error != 0) |
4650 | return (error); | |
4651 | ||
4652 | /* | |
4653 | * Pass that uberblock to the dsl_pool layer which will open the root | |
4654 | * blkptr. This blkptr points to the latest version of the MOS and will | |
4655 | * allow us to read its contents. | |
4656 | */ | |
4657 | error = spa_ld_open_rootbp(spa); | |
4658 | if (error != 0) | |
4659 | return (error); | |
4660 | ||
d2734cce SD |
4661 | return (0); |
4662 | } | |
4663 | ||
4664 | static int | |
4665 | spa_ld_checkpoint_rewind(spa_t *spa) | |
4666 | { | |
4667 | uberblock_t checkpoint; | |
4668 | int error = 0; | |
4669 | ||
4670 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4671 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
4672 | ||
4673 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4674 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
4675 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
4676 | ||
4677 | if (error != 0) { | |
4678 | spa_load_failed(spa, "unable to retrieve checkpointed " | |
4679 | "uberblock from the MOS config [error=%d]", error); | |
4680 | ||
4681 | if (error == ENOENT) | |
4682 | error = ZFS_ERR_NO_CHECKPOINT; | |
4683 | ||
4684 | return (error); | |
4685 | } | |
4686 | ||
4687 | ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); | |
4688 | ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); | |
4689 | ||
4690 | /* | |
4691 | * We need to update the txg and timestamp of the checkpointed | |
4692 | * uberblock to be higher than the latest one. This ensures that | |
4693 | * the checkpointed uberblock is selected if we were to close and | |
4694 | * reopen the pool right after we've written it in the vdev labels. | |
4695 | * (also see block comment in vdev_uberblock_compare) | |
4696 | */ | |
4697 | checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; | |
4698 | checkpoint.ub_timestamp = gethrestime_sec(); | |
4699 | ||
4700 | /* | |
4701 | * Set current uberblock to be the checkpointed uberblock. | |
4702 | */ | |
4703 | spa->spa_uberblock = checkpoint; | |
4704 | ||
4705 | /* | |
4706 | * If we are doing a normal rewind, then the pool is open for | |
4707 | * writing and we sync the "updated" checkpointed uberblock to | |
4708 | * disk. Once this is done, we've basically rewound the whole | |
4709 | * pool and there is no way back. | |
4710 | * | |
4711 | * There are cases when we don't want to attempt and sync the | |
4712 | * checkpointed uberblock to disk because we are opening a | |
4713 | * pool as read-only. Specifically, verifying the checkpointed | |
4714 | * state with zdb, and importing the checkpointed state to get | |
4715 | * a "preview" of its content. | |
4716 | */ | |
4717 | if (spa_writeable(spa)) { | |
4718 | vdev_t *rvd = spa->spa_root_vdev; | |
4719 | ||
4720 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4721 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; | |
4722 | int svdcount = 0; | |
4723 | int children = rvd->vdev_children; | |
29274c9f | 4724 | int c0 = random_in_range(children); |
d2734cce SD |
4725 | |
4726 | for (int c = 0; c < children; c++) { | |
4727 | vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; | |
4728 | ||
4729 | /* Stop when revisiting the first vdev */ | |
4730 | if (c > 0 && svd[0] == vd) | |
4731 | break; | |
4732 | ||
4733 | if (vd->vdev_ms_array == 0 || vd->vdev_islog || | |
4734 | !vdev_is_concrete(vd)) | |
4735 | continue; | |
4736 | ||
4737 | svd[svdcount++] = vd; | |
4738 | if (svdcount == SPA_SYNC_MIN_VDEVS) | |
4739 | break; | |
4740 | } | |
4741 | error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); | |
4742 | if (error == 0) | |
4743 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
4744 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4745 | ||
4746 | if (error != 0) { | |
4747 | spa_load_failed(spa, "failed to write checkpointed " | |
4748 | "uberblock to the vdev labels [error=%d]", error); | |
4749 | return (error); | |
4750 | } | |
4751 | } | |
4752 | ||
4753 | return (0); | |
4754 | } | |
4755 | ||
4756 | static int | |
4757 | spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, | |
4758 | boolean_t *update_config_cache) | |
4759 | { | |
4760 | int error; | |
4761 | ||
4762 | /* | |
4763 | * Parse the config for pool, open and validate vdevs, | |
4764 | * select an uberblock, and use that uberblock to open | |
4765 | * the MOS. | |
4766 | */ | |
4767 | error = spa_ld_mos_init(spa, type); | |
4768 | if (error != 0) | |
4769 | return (error); | |
4770 | ||
9eb7b46e | 4771 | /* |
6cb8e530 PZ |
4772 | * Retrieve the trusted config stored in the MOS and use it to create |
4773 | * a new, exact version of the vdev tree, then reopen all vdevs. | |
9eb7b46e | 4774 | */ |
d2734cce | 4775 | error = spa_ld_trusted_config(spa, type, B_FALSE); |
6cb8e530 | 4776 | if (error == EAGAIN) { |
d2734cce SD |
4777 | if (update_config_cache != NULL) |
4778 | *update_config_cache = B_TRUE; | |
4779 | ||
6cb8e530 PZ |
4780 | /* |
4781 | * Redo the loading process with the trusted config if it is | |
4782 | * too different from the untrusted config. | |
4783 | */ | |
4784 | spa_ld_prepare_for_reload(spa); | |
d2734cce SD |
4785 | spa_load_note(spa, "RELOADING"); |
4786 | error = spa_ld_mos_init(spa, type); | |
4787 | if (error != 0) | |
4788 | return (error); | |
4789 | ||
4790 | error = spa_ld_trusted_config(spa, type, B_TRUE); | |
4791 | if (error != 0) | |
4792 | return (error); | |
4793 | ||
6cb8e530 | 4794 | } else if (error != 0) { |
9eb7b46e | 4795 | return (error); |
6cb8e530 | 4796 | } |
9eb7b46e | 4797 | |
d2734cce SD |
4798 | return (0); |
4799 | } | |
4800 | ||
4801 | /* | |
4802 | * Load an existing storage pool, using the config provided. This config | |
4803 | * describes which vdevs are part of the pool and is later validated against | |
4804 | * partial configs present in each vdev's label and an entire copy of the | |
4805 | * config stored in the MOS. | |
4806 | */ | |
4807 | static int | |
a926aab9 | 4808 | spa_load_impl(spa_t *spa, spa_import_type_t type, const char **ereport) |
d2734cce SD |
4809 | { |
4810 | int error = 0; | |
4811 | boolean_t missing_feat_write = B_FALSE; | |
4812 | boolean_t checkpoint_rewind = | |
4813 | (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
4814 | boolean_t update_config_cache = B_FALSE; | |
4815 | ||
4816 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4817 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); | |
4818 | ||
4819 | spa_load_note(spa, "LOADING"); | |
4820 | ||
4821 | error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); | |
4822 | if (error != 0) | |
4823 | return (error); | |
4824 | ||
4825 | /* | |
4826 | * If we are rewinding to the checkpoint then we need to repeat | |
4827 | * everything we've done so far in this function but this time | |
4828 | * selecting the checkpointed uberblock and using that to open | |
4829 | * the MOS. | |
4830 | */ | |
4831 | if (checkpoint_rewind) { | |
4832 | /* | |
4833 | * If we are rewinding to the checkpoint update config cache | |
4834 | * anyway. | |
4835 | */ | |
4836 | update_config_cache = B_TRUE; | |
4837 | ||
4838 | /* | |
4839 | * Extract the checkpointed uberblock from the current MOS | |
4840 | * and use this as the pool's uberblock from now on. If the | |
4841 | * pool is imported as writeable we also write the checkpoint | |
4842 | * uberblock to the labels, making the rewind permanent. | |
4843 | */ | |
4844 | error = spa_ld_checkpoint_rewind(spa); | |
4845 | if (error != 0) | |
4846 | return (error); | |
4847 | ||
4848 | /* | |
e1cfd73f | 4849 | * Redo the loading process again with the |
d2734cce SD |
4850 | * checkpointed uberblock. |
4851 | */ | |
4852 | spa_ld_prepare_for_reload(spa); | |
4853 | spa_load_note(spa, "LOADING checkpointed uberblock"); | |
4854 | error = spa_ld_mos_with_trusted_config(spa, type, NULL); | |
4855 | if (error != 0) | |
4856 | return (error); | |
4857 | } | |
4858 | ||
4859 | /* | |
4860 | * Retrieve the checkpoint txg if the pool has a checkpoint. | |
4861 | */ | |
4862 | error = spa_ld_read_checkpoint_txg(spa); | |
4863 | if (error != 0) | |
4864 | return (error); | |
4865 | ||
9eb7b46e PZ |
4866 | /* |
4867 | * Retrieve the mapping of indirect vdevs. Those vdevs were removed | |
4868 | * from the pool and their contents were re-mapped to other vdevs. Note | |
4869 | * that everything that we read before this step must have been | |
4870 | * rewritten on concrete vdevs after the last device removal was | |
4871 | * initiated. Otherwise we could be reading from indirect vdevs before | |
4872 | * we have loaded their mappings. | |
4873 | */ | |
4874 | error = spa_ld_open_indirect_vdev_metadata(spa); | |
4875 | if (error != 0) | |
4876 | return (error); | |
4877 | ||
4878 | /* | |
4879 | * Retrieve the full list of active features from the MOS and check if | |
4880 | * they are all supported. | |
4881 | */ | |
4a0ee12a | 4882 | error = spa_ld_check_features(spa, &missing_feat_write); |
9eb7b46e PZ |
4883 | if (error != 0) |
4884 | return (error); | |
4885 | ||
4886 | /* | |
4887 | * Load several special directories from the MOS needed by the dsl_pool | |
4888 | * layer. | |
4889 | */ | |
4890 | error = spa_ld_load_special_directories(spa); | |
4891 | if (error != 0) | |
4892 | return (error); | |
4893 | ||
9eb7b46e PZ |
4894 | /* |
4895 | * Retrieve pool properties from the MOS. | |
4896 | */ | |
4897 | error = spa_ld_get_props(spa); | |
4898 | if (error != 0) | |
4899 | return (error); | |
4900 | ||
4901 | /* | |
4902 | * Retrieve the list of auxiliary devices - cache devices and spares - | |
4903 | * and open them. | |
4904 | */ | |
4905 | error = spa_ld_open_aux_vdevs(spa, type); | |
4906 | if (error != 0) | |
4907 | return (error); | |
4908 | ||
4909 | /* | |
4910 | * Load the metadata for all vdevs. Also check if unopenable devices | |
4911 | * should be autoreplaced. | |
4912 | */ | |
4a0ee12a | 4913 | error = spa_ld_load_vdev_metadata(spa); |
9eb7b46e PZ |
4914 | if (error != 0) |
4915 | return (error); | |
4916 | ||
4917 | error = spa_ld_load_dedup_tables(spa); | |
4918 | if (error != 0) | |
4919 | return (error); | |
4920 | ||
67a1b037 PJD |
4921 | error = spa_ld_load_brt(spa); |
4922 | if (error != 0) | |
4923 | return (error); | |
4924 | ||
9eb7b46e PZ |
4925 | /* |
4926 | * Verify the logs now to make sure we don't have any unexpected errors | |
4927 | * when we claim log blocks later. | |
4928 | */ | |
4929 | error = spa_ld_verify_logs(spa, type, ereport); | |
4930 | if (error != 0) | |
4931 | return (error); | |
4932 | ||
9ae529ec | 4933 | if (missing_feat_write) { |
6cb8e530 | 4934 | ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); |
9ae529ec CS |
4935 | |
4936 | /* | |
4937 | * At this point, we know that we can open the pool in | |
4938 | * read-only mode but not read-write mode. We now have enough | |
4939 | * information and can return to userland. | |
4940 | */ | |
9eb7b46e PZ |
4941 | return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, |
4942 | ENOTSUP)); | |
9ae529ec CS |
4943 | } |
4944 | ||
572e2857 | 4945 | /* |
9eb7b46e PZ |
4946 | * Traverse the last txgs to make sure the pool was left off in a safe |
4947 | * state. When performing an extreme rewind, we verify the whole pool, | |
4948 | * which can take a very long time. | |
572e2857 | 4949 | */ |
4a0ee12a | 4950 | error = spa_ld_verify_pool_data(spa); |
9eb7b46e PZ |
4951 | if (error != 0) |
4952 | return (error); | |
572e2857 | 4953 | |
9eb7b46e PZ |
4954 | /* |
4955 | * Calculate the deflated space for the pool. This must be done before | |
4956 | * we write anything to the pool because we'd need to update the space | |
4957 | * accounting using the deflated sizes. | |
4958 | */ | |
4959 | spa_update_dspace(spa); | |
4960 | ||
4961 | /* | |
4962 | * We have now retrieved all the information we needed to open the | |
4963 | * pool. If we are importing the pool in read-write mode, a few | |
4964 | * additional steps must be performed to finish the import. | |
4965 | */ | |
6cb8e530 | 4966 | if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || |
428870ff | 4967 | spa->spa_load_max_txg == UINT64_MAX)) { |
6cb8e530 PZ |
4968 | uint64_t config_cache_txg = spa->spa_config_txg; |
4969 | ||
4970 | ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); | |
34dc7c2f | 4971 | |
d2734cce SD |
4972 | /* |
4973 | * In case of a checkpoint rewind, log the original txg | |
4974 | * of the checkpointed uberblock. | |
4975 | */ | |
4976 | if (checkpoint_rewind) { | |
4977 | spa_history_log_internal(spa, "checkpoint rewind", | |
4978 | NULL, "rewound state to txg=%llu", | |
4979 | (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); | |
4980 | } | |
4981 | ||
34dc7c2f | 4982 | /* |
9eb7b46e | 4983 | * Traverse the ZIL and claim all blocks. |
34dc7c2f | 4984 | */ |
9eb7b46e | 4985 | spa_ld_claim_log_blocks(spa); |
428870ff | 4986 | |
9eb7b46e PZ |
4987 | /* |
4988 | * Kick-off the syncing thread. | |
4989 | */ | |
34dc7c2f BB |
4990 | spa->spa_sync_on = B_TRUE; |
4991 | txg_sync_start(spa->spa_dsl_pool); | |
379ca9cf | 4992 | mmp_thread_start(spa); |
34dc7c2f BB |
4993 | |
4994 | /* | |
428870ff BB |
4995 | * Wait for all claims to sync. We sync up to the highest |
4996 | * claimed log block birth time so that claimed log blocks | |
4997 | * don't appear to be from the future. spa_claim_max_txg | |
9eb7b46e PZ |
4998 | * will have been set for us by ZIL traversal operations |
4999 | * performed above. | |
34dc7c2f | 5000 | */ |
428870ff | 5001 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); |
34dc7c2f BB |
5002 | |
5003 | /* | |
9eb7b46e PZ |
5004 | * Check if we need to request an update of the config. On the |
5005 | * next sync, we would update the config stored in vdev labels | |
5006 | * and the cachefile (by default /etc/zfs/zpool.cache). | |
34dc7c2f | 5007 | */ |
6cb8e530 | 5008 | spa_ld_check_for_config_update(spa, config_cache_txg, |
d2734cce | 5009 | update_config_cache); |
fb5f0bc8 BB |
5010 | |
5011 | /* | |
9a49d3f3 BB |
5012 | * Check if a rebuild was in progress and if so resume it. |
5013 | * Then check all DTLs to see if anything needs resilvering. | |
5014 | * The resilver will be deferred if a rebuild was started. | |
fb5f0bc8 | 5015 | */ |
9a49d3f3 BB |
5016 | if (vdev_rebuild_active(spa->spa_root_vdev)) { |
5017 | vdev_rebuild_restart(spa); | |
5018 | } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
5019 | vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
fb5f0bc8 | 5020 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
9a49d3f3 | 5021 | } |
428870ff | 5022 | |
6f1ffb06 MA |
5023 | /* |
5024 | * Log the fact that we booted up (so that we can detect if | |
5025 | * we rebooted in the middle of an operation). | |
5026 | */ | |
d5e024cb | 5027 | spa_history_log_version(spa, "open", NULL); |
6f1ffb06 | 5028 | |
9b2266e3 SD |
5029 | spa_restart_removal(spa); |
5030 | spa_spawn_aux_threads(spa); | |
5031 | ||
428870ff BB |
5032 | /* |
5033 | * Delete any inconsistent datasets. | |
9b2266e3 SD |
5034 | * |
5035 | * Note: | |
5036 | * Since we may be issuing deletes for clones here, | |
5037 | * we make sure to do so after we've spawned all the | |
5038 | * auxiliary threads above (from which the livelist | |
5039 | * deletion zthr is part of). | |
428870ff BB |
5040 | */ |
5041 | (void) dmu_objset_find(spa_name(spa), | |
5042 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
5043 | ||
5044 | /* | |
5045 | * Clean up any stale temporary dataset userrefs. | |
5046 | */ | |
5047 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); | |
a1d477c2 | 5048 | |
619f0976 GW |
5049 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
5050 | vdev_initialize_restart(spa->spa_root_vdev); | |
1b939560 BB |
5051 | vdev_trim_restart(spa->spa_root_vdev); |
5052 | vdev_autotrim_restart(spa); | |
619f0976 | 5053 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f BB |
5054 | } |
5055 | ||
ca95f70d | 5056 | spa_import_progress_remove(spa_guid(spa)); |
77f6826b GA |
5057 | spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); |
5058 | ||
4a0ee12a PZ |
5059 | spa_load_note(spa, "LOADED"); |
5060 | ||
428870ff BB |
5061 | return (0); |
5062 | } | |
34dc7c2f | 5063 | |
428870ff | 5064 | static int |
6cb8e530 | 5065 | spa_load_retry(spa_t *spa, spa_load_state_t state) |
428870ff | 5066 | { |
da92d5cb | 5067 | spa_mode_t mode = spa->spa_mode; |
572e2857 | 5068 | |
428870ff BB |
5069 | spa_unload(spa); |
5070 | spa_deactivate(spa); | |
5071 | ||
dea377c0 | 5072 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; |
428870ff | 5073 | |
572e2857 | 5074 | spa_activate(spa, mode); |
428870ff BB |
5075 | spa_async_suspend(spa); |
5076 | ||
4a0ee12a PZ |
5077 | spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", |
5078 | (u_longlong_t)spa->spa_load_max_txg); | |
5079 | ||
6cb8e530 | 5080 | return (spa_load(spa, state, SPA_IMPORT_EXISTING)); |
428870ff BB |
5081 | } |
5082 | ||
9ae529ec CS |
5083 | /* |
5084 | * If spa_load() fails this function will try loading prior txg's. If | |
5085 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
5086 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
5087 | * function will not rewind the pool and will return the same error as | |
5088 | * spa_load(). | |
5089 | */ | |
428870ff | 5090 | static int |
6cb8e530 PZ |
5091 | spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, |
5092 | int rewind_flags) | |
428870ff | 5093 | { |
9ae529ec | 5094 | nvlist_t *loadinfo = NULL; |
428870ff BB |
5095 | nvlist_t *config = NULL; |
5096 | int load_error, rewind_error; | |
5097 | uint64_t safe_rewind_txg; | |
5098 | uint64_t min_txg; | |
5099 | ||
5100 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
5101 | spa->spa_load_max_txg = spa->spa_load_txg; | |
5102 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
5103 | } else { | |
5104 | spa->spa_load_max_txg = max_request; | |
dea377c0 MA |
5105 | if (max_request != UINT64_MAX) |
5106 | spa->spa_extreme_rewind = B_TRUE; | |
428870ff BB |
5107 | } |
5108 | ||
6cb8e530 | 5109 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); |
428870ff BB |
5110 | if (load_error == 0) |
5111 | return (0); | |
d2734cce SD |
5112 | if (load_error == ZFS_ERR_NO_CHECKPOINT) { |
5113 | /* | |
5114 | * When attempting checkpoint-rewind on a pool with no | |
5115 | * checkpoint, we should not attempt to load uberblocks | |
5116 | * from previous txgs when spa_load fails. | |
5117 | */ | |
5118 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
ca95f70d | 5119 | spa_import_progress_remove(spa_guid(spa)); |
d2734cce SD |
5120 | return (load_error); |
5121 | } | |
428870ff BB |
5122 | |
5123 | if (spa->spa_root_vdev != NULL) | |
5124 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
5125 | ||
5126 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
5127 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
5128 | ||
5129 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
5130 | nvlist_free(config); | |
ca95f70d | 5131 | spa_import_progress_remove(spa_guid(spa)); |
428870ff BB |
5132 | return (load_error); |
5133 | } | |
5134 | ||
9ae529ec CS |
5135 | if (state == SPA_LOAD_RECOVER) { |
5136 | /* Price of rolling back is discarding txgs, including log */ | |
428870ff | 5137 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
9ae529ec CS |
5138 | } else { |
5139 | /* | |
5140 | * If we aren't rolling back save the load info from our first | |
5141 | * import attempt so that we can restore it after attempting | |
5142 | * to rewind. | |
5143 | */ | |
5144 | loadinfo = spa->spa_load_info; | |
5145 | spa->spa_load_info = fnvlist_alloc(); | |
5146 | } | |
428870ff BB |
5147 | |
5148 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
5149 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
5150 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
5151 | TXG_INITIAL : safe_rewind_txg; | |
5152 | ||
5153 | /* | |
5154 | * Continue as long as we're finding errors, we're still within | |
5155 | * the acceptable rewind range, and we're still finding uberblocks | |
5156 | */ | |
5157 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
5158 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
5159 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
5160 | spa->spa_extreme_rewind = B_TRUE; | |
6cb8e530 | 5161 | rewind_error = spa_load_retry(spa, state); |
428870ff BB |
5162 | } |
5163 | ||
428870ff BB |
5164 | spa->spa_extreme_rewind = B_FALSE; |
5165 | spa->spa_load_max_txg = UINT64_MAX; | |
5166 | ||
5167 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
5168 | spa_config_set(spa, config); | |
ee6370a7 | 5169 | else |
5170 | nvlist_free(config); | |
428870ff | 5171 | |
9ae529ec CS |
5172 | if (state == SPA_LOAD_RECOVER) { |
5173 | ASSERT3P(loadinfo, ==, NULL); | |
ca95f70d | 5174 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5175 | return (rewind_error); |
5176 | } else { | |
5177 | /* Store the rewind info as part of the initial load info */ | |
5178 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
5179 | spa->spa_load_info); | |
5180 | ||
5181 | /* Restore the initial load info */ | |
5182 | fnvlist_free(spa->spa_load_info); | |
5183 | spa->spa_load_info = loadinfo; | |
5184 | ||
ca95f70d | 5185 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5186 | return (load_error); |
5187 | } | |
34dc7c2f BB |
5188 | } |
5189 | ||
5190 | /* | |
5191 | * Pool Open/Import | |
5192 | * | |
5193 | * The import case is identical to an open except that the configuration is sent | |
5194 | * down from userland, instead of grabbed from the configuration cache. For the | |
5195 | * case of an open, the pool configuration will exist in the | |
5196 | * POOL_STATE_UNINITIALIZED state. | |
5197 | * | |
5198 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
5199 | * the same time open the pool, without having to keep around the spa_t in some | |
5200 | * ambiguous state. | |
5201 | */ | |
5202 | static int | |
a926aab9 AZ |
5203 | spa_open_common(const char *pool, spa_t **spapp, const void *tag, |
5204 | nvlist_t *nvpolicy, nvlist_t **config) | |
34dc7c2f BB |
5205 | { |
5206 | spa_t *spa; | |
572e2857 | 5207 | spa_load_state_t state = SPA_LOAD_OPEN; |
34dc7c2f | 5208 | int error; |
34dc7c2f | 5209 | int locked = B_FALSE; |
526af785 | 5210 | int firstopen = B_FALSE; |
34dc7c2f BB |
5211 | |
5212 | *spapp = NULL; | |
5213 | ||
5214 | /* | |
5215 | * As disgusting as this is, we need to support recursive calls to this | |
5216 | * function because dsl_dir_open() is called during spa_load(), and ends | |
5217 | * up calling spa_open() again. The real fix is to figure out how to | |
5218 | * avoid dsl_dir_open() calling this in the first place. | |
5219 | */ | |
c25b8f99 | 5220 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { |
34dc7c2f BB |
5221 | mutex_enter(&spa_namespace_lock); |
5222 | locked = B_TRUE; | |
5223 | } | |
5224 | ||
5225 | if ((spa = spa_lookup(pool)) == NULL) { | |
5226 | if (locked) | |
5227 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5228 | return (SET_ERROR(ENOENT)); |
34dc7c2f | 5229 | } |
428870ff | 5230 | |
34dc7c2f | 5231 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { |
8a393be3 | 5232 | zpool_load_policy_t policy; |
428870ff | 5233 | |
526af785 PJD |
5234 | firstopen = B_TRUE; |
5235 | ||
8a393be3 | 5236 | zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, |
428870ff | 5237 | &policy); |
8a393be3 | 5238 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) |
428870ff | 5239 | state = SPA_LOAD_RECOVER; |
34dc7c2f | 5240 | |
fb5f0bc8 | 5241 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 5242 | |
428870ff BB |
5243 | if (state != SPA_LOAD_RECOVER) |
5244 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6cb8e530 | 5245 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; |
428870ff | 5246 | |
4a0ee12a | 5247 | zfs_dbgmsg("spa_open_common: opening %s", pool); |
8a393be3 PZ |
5248 | error = spa_load_best(spa, state, policy.zlp_txg, |
5249 | policy.zlp_rewind); | |
34dc7c2f BB |
5250 | |
5251 | if (error == EBADF) { | |
5252 | /* | |
5253 | * If vdev_validate() returns failure (indicated by | |
5254 | * EBADF), it indicates that one of the vdevs indicates | |
5255 | * that the pool has been exported or destroyed. If | |
5256 | * this is the case, the config cache is out of sync and | |
5257 | * we should remove the pool from the namespace. | |
5258 | */ | |
34dc7c2f BB |
5259 | spa_unload(spa); |
5260 | spa_deactivate(spa); | |
55c12724 | 5261 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 5262 | spa_remove(spa); |
34dc7c2f BB |
5263 | if (locked) |
5264 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5265 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
5266 | } |
5267 | ||
5268 | if (error) { | |
5269 | /* | |
5270 | * We can't open the pool, but we still have useful | |
5271 | * information: the state of each vdev after the | |
5272 | * attempted vdev_open(). Return this to the user. | |
5273 | */ | |
572e2857 | 5274 | if (config != NULL && spa->spa_config) { |
65ad5d11 AJ |
5275 | *config = fnvlist_dup(spa->spa_config); |
5276 | fnvlist_add_nvlist(*config, | |
572e2857 | 5277 | ZPOOL_CONFIG_LOAD_INFO, |
65ad5d11 | 5278 | spa->spa_load_info); |
572e2857 | 5279 | } |
34dc7c2f BB |
5280 | spa_unload(spa); |
5281 | spa_deactivate(spa); | |
428870ff | 5282 | spa->spa_last_open_failed = error; |
34dc7c2f BB |
5283 | if (locked) |
5284 | mutex_exit(&spa_namespace_lock); | |
5285 | *spapp = NULL; | |
5286 | return (error); | |
34dc7c2f | 5287 | } |
34dc7c2f BB |
5288 | } |
5289 | ||
5290 | spa_open_ref(spa, tag); | |
5291 | ||
b128c09f | 5292 | if (config != NULL) |
34dc7c2f | 5293 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
34dc7c2f | 5294 | |
572e2857 BB |
5295 | /* |
5296 | * If we've recovered the pool, pass back any information we | |
5297 | * gathered while doing the load. | |
5298 | */ | |
1bd02680 | 5299 | if (state == SPA_LOAD_RECOVER && config != NULL) { |
65ad5d11 AJ |
5300 | fnvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, |
5301 | spa->spa_load_info); | |
572e2857 BB |
5302 | } |
5303 | ||
428870ff BB |
5304 | if (locked) { |
5305 | spa->spa_last_open_failed = 0; | |
5306 | spa->spa_last_ubsync_txg = 0; | |
5307 | spa->spa_load_txg = 0; | |
5308 | mutex_exit(&spa_namespace_lock); | |
5309 | } | |
5310 | ||
526af785 | 5311 | if (firstopen) |
ec213971 | 5312 | zvol_create_minors_recursive(spa_name(spa)); |
526af785 | 5313 | |
428870ff BB |
5314 | *spapp = spa; |
5315 | ||
34dc7c2f BB |
5316 | return (0); |
5317 | } | |
5318 | ||
428870ff | 5319 | int |
a926aab9 AZ |
5320 | spa_open_rewind(const char *name, spa_t **spapp, const void *tag, |
5321 | nvlist_t *policy, nvlist_t **config) | |
428870ff BB |
5322 | { |
5323 | return (spa_open_common(name, spapp, tag, policy, config)); | |
5324 | } | |
5325 | ||
34dc7c2f | 5326 | int |
a926aab9 | 5327 | spa_open(const char *name, spa_t **spapp, const void *tag) |
34dc7c2f | 5328 | { |
428870ff | 5329 | return (spa_open_common(name, spapp, tag, NULL, NULL)); |
34dc7c2f BB |
5330 | } |
5331 | ||
5332 | /* | |
5333 | * Lookup the given spa_t, incrementing the inject count in the process, | |
5334 | * preventing it from being exported or destroyed. | |
5335 | */ | |
5336 | spa_t * | |
5337 | spa_inject_addref(char *name) | |
5338 | { | |
5339 | spa_t *spa; | |
5340 | ||
5341 | mutex_enter(&spa_namespace_lock); | |
5342 | if ((spa = spa_lookup(name)) == NULL) { | |
5343 | mutex_exit(&spa_namespace_lock); | |
5344 | return (NULL); | |
5345 | } | |
5346 | spa->spa_inject_ref++; | |
5347 | mutex_exit(&spa_namespace_lock); | |
5348 | ||
5349 | return (spa); | |
5350 | } | |
5351 | ||
5352 | void | |
5353 | spa_inject_delref(spa_t *spa) | |
5354 | { | |
5355 | mutex_enter(&spa_namespace_lock); | |
5356 | spa->spa_inject_ref--; | |
5357 | mutex_exit(&spa_namespace_lock); | |
5358 | } | |
5359 | ||
5360 | /* | |
5361 | * Add spares device information to the nvlist. | |
5362 | */ | |
5363 | static void | |
5364 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
5365 | { | |
5366 | nvlist_t **spares; | |
5367 | uint_t i, nspares; | |
5368 | nvlist_t *nvroot; | |
5369 | uint64_t guid; | |
5370 | vdev_stat_t *vs; | |
5371 | uint_t vsc; | |
5372 | uint64_t pool; | |
5373 | ||
9babb374 BB |
5374 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5375 | ||
34dc7c2f BB |
5376 | if (spa->spa_spares.sav_count == 0) |
5377 | return; | |
5378 | ||
65ad5d11 AJ |
5379 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5380 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5381 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f | 5382 | if (nspares != 0) { |
795075e6 PD |
5383 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5384 | (const nvlist_t * const *)spares, nspares); | |
65ad5d11 AJ |
5385 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5386 | &spares, &nspares)); | |
34dc7c2f BB |
5387 | |
5388 | /* | |
5389 | * Go through and find any spares which have since been | |
5390 | * repurposed as an active spare. If this is the case, update | |
5391 | * their status appropriately. | |
5392 | */ | |
5393 | for (i = 0; i < nspares; i++) { | |
65ad5d11 AJ |
5394 | guid = fnvlist_lookup_uint64(spares[i], |
5395 | ZPOOL_CONFIG_GUID); | |
b128c09f BB |
5396 | if (spa_spare_exists(guid, &pool, NULL) && |
5397 | pool != 0ULL) { | |
65ad5d11 AJ |
5398 | VERIFY0(nvlist_lookup_uint64_array(spares[i], |
5399 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, | |
5400 | &vsc)); | |
34dc7c2f BB |
5401 | vs->vs_state = VDEV_STATE_CANT_OPEN; |
5402 | vs->vs_aux = VDEV_AUX_SPARED; | |
5403 | } | |
5404 | } | |
5405 | } | |
5406 | } | |
5407 | ||
5408 | /* | |
5409 | * Add l2cache device information to the nvlist, including vdev stats. | |
5410 | */ | |
5411 | static void | |
5412 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
5413 | { | |
5414 | nvlist_t **l2cache; | |
5415 | uint_t i, j, nl2cache; | |
5416 | nvlist_t *nvroot; | |
5417 | uint64_t guid; | |
5418 | vdev_t *vd; | |
5419 | vdev_stat_t *vs; | |
5420 | uint_t vsc; | |
5421 | ||
9babb374 BB |
5422 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5423 | ||
34dc7c2f BB |
5424 | if (spa->spa_l2cache.sav_count == 0) |
5425 | return; | |
5426 | ||
65ad5d11 AJ |
5427 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5428 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5429 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
34dc7c2f | 5430 | if (nl2cache != 0) { |
795075e6 PD |
5431 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5432 | (const nvlist_t * const *)l2cache, nl2cache); | |
65ad5d11 AJ |
5433 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5434 | &l2cache, &nl2cache)); | |
34dc7c2f BB |
5435 | |
5436 | /* | |
5437 | * Update level 2 cache device stats. | |
5438 | */ | |
5439 | ||
5440 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 AJ |
5441 | guid = fnvlist_lookup_uint64(l2cache[i], |
5442 | ZPOOL_CONFIG_GUID); | |
34dc7c2f BB |
5443 | |
5444 | vd = NULL; | |
5445 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
5446 | if (guid == | |
5447 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
5448 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
5449 | break; | |
5450 | } | |
5451 | } | |
5452 | ASSERT(vd != NULL); | |
5453 | ||
65ad5d11 AJ |
5454 | VERIFY0(nvlist_lookup_uint64_array(l2cache[i], |
5455 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
34dc7c2f | 5456 | vdev_get_stats(vd, vs); |
193a37cb TH |
5457 | vdev_config_generate_stats(vd, l2cache[i]); |
5458 | ||
34dc7c2f BB |
5459 | } |
5460 | } | |
34dc7c2f BB |
5461 | } |
5462 | ||
9ae529ec | 5463 | static void |
417104bd | 5464 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) |
9ae529ec | 5465 | { |
9ae529ec CS |
5466 | zap_cursor_t zc; |
5467 | zap_attribute_t za; | |
5468 | ||
9ae529ec CS |
5469 | if (spa->spa_feat_for_read_obj != 0) { |
5470 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
5471 | spa->spa_feat_for_read_obj); | |
5472 | zap_cursor_retrieve(&zc, &za) == 0; | |
5473 | zap_cursor_advance(&zc)) { | |
5474 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
5475 | za.za_num_integers == 1); | |
417104bd | 5476 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
5477 | za.za_first_integer)); |
5478 | } | |
5479 | zap_cursor_fini(&zc); | |
5480 | } | |
5481 | ||
5482 | if (spa->spa_feat_for_write_obj != 0) { | |
5483 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
5484 | spa->spa_feat_for_write_obj); | |
5485 | zap_cursor_retrieve(&zc, &za) == 0; | |
5486 | zap_cursor_advance(&zc)) { | |
5487 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
5488 | za.za_num_integers == 1); | |
417104bd | 5489 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
5490 | za.za_first_integer)); |
5491 | } | |
5492 | zap_cursor_fini(&zc); | |
5493 | } | |
417104bd NB |
5494 | } |
5495 | ||
5496 | static void | |
5497 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
5498 | { | |
5499 | int i; | |
5500 | ||
5501 | for (i = 0; i < SPA_FEATURES; i++) { | |
5502 | zfeature_info_t feature = spa_feature_table[i]; | |
5503 | uint64_t refcount; | |
5504 | ||
5505 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
5506 | continue; | |
5507 | ||
5508 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
5509 | } | |
5510 | } | |
5511 | ||
5512 | /* | |
5513 | * Store a list of pool features and their reference counts in the | |
5514 | * config. | |
5515 | * | |
5516 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
5517 | * the pool features and reference counts from disk, then save the list | |
5518 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
5519 | * and refresh its values from the cached reference counts. This | |
5520 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
5521 | * clear' can resume the pool. | |
5522 | */ | |
5523 | static void | |
5524 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
5525 | { | |
4eb30c68 | 5526 | nvlist_t *features; |
417104bd NB |
5527 | |
5528 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
5529 | ||
4eb30c68 NB |
5530 | mutex_enter(&spa->spa_feat_stats_lock); |
5531 | features = spa->spa_feat_stats; | |
5532 | ||
417104bd NB |
5533 | if (features != NULL) { |
5534 | spa_feature_stats_from_cache(spa, features); | |
5535 | } else { | |
5536 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
5537 | spa->spa_feat_stats = features; | |
5538 | spa_feature_stats_from_disk(spa, features); | |
5539 | } | |
9ae529ec | 5540 | |
417104bd NB |
5541 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, |
5542 | features)); | |
4eb30c68 NB |
5543 | |
5544 | mutex_exit(&spa->spa_feat_stats_lock); | |
9ae529ec CS |
5545 | } |
5546 | ||
34dc7c2f | 5547 | int |
9ae529ec CS |
5548 | spa_get_stats(const char *name, nvlist_t **config, |
5549 | char *altroot, size_t buflen) | |
34dc7c2f BB |
5550 | { |
5551 | int error; | |
5552 | spa_t *spa; | |
5553 | ||
5554 | *config = NULL; | |
428870ff | 5555 | error = spa_open_common(name, &spa, FTAG, NULL, config); |
34dc7c2f | 5556 | |
9babb374 BB |
5557 | if (spa != NULL) { |
5558 | /* | |
5559 | * This still leaves a window of inconsistency where the spares | |
5560 | * or l2cache devices could change and the config would be | |
5561 | * self-inconsistent. | |
5562 | */ | |
5563 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
34dc7c2f | 5564 | |
9babb374 | 5565 | if (*config != NULL) { |
572e2857 BB |
5566 | uint64_t loadtimes[2]; |
5567 | ||
5568 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
5569 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
65ad5d11 AJ |
5570 | fnvlist_add_uint64_array(*config, |
5571 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2); | |
572e2857 | 5572 | |
65ad5d11 | 5573 | fnvlist_add_uint64(*config, |
9babb374 | 5574 | ZPOOL_CONFIG_ERRCOUNT, |
018f2604 | 5575 | spa_approx_errlog_size(spa)); |
9babb374 | 5576 | |
cec3a0a1 | 5577 | if (spa_suspended(spa)) { |
65ad5d11 | 5578 | fnvlist_add_uint64(*config, |
9babb374 | 5579 | ZPOOL_CONFIG_SUSPENDED, |
65ad5d11 AJ |
5580 | spa->spa_failmode); |
5581 | fnvlist_add_uint64(*config, | |
cec3a0a1 | 5582 | ZPOOL_CONFIG_SUSPENDED_REASON, |
65ad5d11 | 5583 | spa->spa_suspended); |
cec3a0a1 | 5584 | } |
b128c09f | 5585 | |
9babb374 BB |
5586 | spa_add_spares(spa, *config); |
5587 | spa_add_l2cache(spa, *config); | |
9ae529ec | 5588 | spa_add_feature_stats(spa, *config); |
9babb374 | 5589 | } |
34dc7c2f BB |
5590 | } |
5591 | ||
5592 | /* | |
5593 | * We want to get the alternate root even for faulted pools, so we cheat | |
5594 | * and call spa_lookup() directly. | |
5595 | */ | |
5596 | if (altroot) { | |
5597 | if (spa == NULL) { | |
5598 | mutex_enter(&spa_namespace_lock); | |
5599 | spa = spa_lookup(name); | |
5600 | if (spa) | |
5601 | spa_altroot(spa, altroot, buflen); | |
5602 | else | |
5603 | altroot[0] = '\0'; | |
5604 | spa = NULL; | |
5605 | mutex_exit(&spa_namespace_lock); | |
5606 | } else { | |
5607 | spa_altroot(spa, altroot, buflen); | |
5608 | } | |
5609 | } | |
5610 | ||
9babb374 BB |
5611 | if (spa != NULL) { |
5612 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
34dc7c2f | 5613 | spa_close(spa, FTAG); |
9babb374 | 5614 | } |
34dc7c2f BB |
5615 | |
5616 | return (error); | |
5617 | } | |
5618 | ||
5619 | /* | |
5620 | * Validate that the auxiliary device array is well formed. We must have an | |
5621 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
5622 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
5623 | * specified, as long as they are well-formed. | |
5624 | */ | |
5625 | static int | |
5626 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
5627 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
5628 | vdev_labeltype_t label) | |
5629 | { | |
5630 | nvlist_t **dev; | |
5631 | uint_t i, ndev; | |
5632 | vdev_t *vd; | |
5633 | int error; | |
5634 | ||
b128c09f BB |
5635 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
5636 | ||
34dc7c2f BB |
5637 | /* |
5638 | * It's acceptable to have no devs specified. | |
5639 | */ | |
5640 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
5641 | return (0); | |
5642 | ||
5643 | if (ndev == 0) | |
2e528b49 | 5644 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
5645 | |
5646 | /* | |
5647 | * Make sure the pool is formatted with a version that supports this | |
5648 | * device type. | |
5649 | */ | |
5650 | if (spa_version(spa) < version) | |
2e528b49 | 5651 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
5652 | |
5653 | /* | |
5654 | * Set the pending device list so we correctly handle device in-use | |
5655 | * checking. | |
5656 | */ | |
5657 | sav->sav_pending = dev; | |
5658 | sav->sav_npending = ndev; | |
5659 | ||
5660 | for (i = 0; i < ndev; i++) { | |
5661 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
5662 | mode)) != 0) | |
5663 | goto out; | |
5664 | ||
5665 | if (!vd->vdev_ops->vdev_op_leaf) { | |
5666 | vdev_free(vd); | |
2e528b49 | 5667 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
5668 | goto out; |
5669 | } | |
5670 | ||
34dc7c2f BB |
5671 | vd->vdev_top = vd; |
5672 | ||
5673 | if ((error = vdev_open(vd)) == 0 && | |
5674 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
65ad5d11 AJ |
5675 | fnvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, |
5676 | vd->vdev_guid); | |
34dc7c2f BB |
5677 | } |
5678 | ||
5679 | vdev_free(vd); | |
5680 | ||
5681 | if (error && | |
5682 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
5683 | goto out; | |
5684 | else | |
5685 | error = 0; | |
5686 | } | |
5687 | ||
5688 | out: | |
5689 | sav->sav_pending = NULL; | |
5690 | sav->sav_npending = 0; | |
5691 | return (error); | |
5692 | } | |
5693 | ||
5694 | static int | |
5695 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
5696 | { | |
5697 | int error; | |
5698 | ||
b128c09f BB |
5699 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
5700 | ||
34dc7c2f BB |
5701 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, |
5702 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
5703 | VDEV_LABEL_SPARE)) != 0) { | |
5704 | return (error); | |
5705 | } | |
5706 | ||
5707 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
5708 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
5709 | VDEV_LABEL_L2CACHE)); | |
5710 | } | |
5711 | ||
5712 | static void | |
5713 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
5714 | const char *config) | |
5715 | { | |
5716 | int i; | |
5717 | ||
5718 | if (sav->sav_config != NULL) { | |
5719 | nvlist_t **olddevs; | |
5720 | uint_t oldndevs; | |
5721 | nvlist_t **newdevs; | |
5722 | ||
5723 | /* | |
4e33ba4c | 5724 | * Generate new dev list by concatenating with the |
34dc7c2f BB |
5725 | * current dev list. |
5726 | */ | |
65ad5d11 AJ |
5727 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, config, |
5728 | &olddevs, &oldndevs)); | |
34dc7c2f BB |
5729 | |
5730 | newdevs = kmem_alloc(sizeof (void *) * | |
79c76d5b | 5731 | (ndevs + oldndevs), KM_SLEEP); |
34dc7c2f | 5732 | for (i = 0; i < oldndevs; i++) |
65ad5d11 | 5733 | newdevs[i] = fnvlist_dup(olddevs[i]); |
34dc7c2f | 5734 | for (i = 0; i < ndevs; i++) |
65ad5d11 | 5735 | newdevs[i + oldndevs] = fnvlist_dup(devs[i]); |
34dc7c2f | 5736 | |
65ad5d11 | 5737 | fnvlist_remove(sav->sav_config, config); |
34dc7c2f | 5738 | |
795075e6 PD |
5739 | fnvlist_add_nvlist_array(sav->sav_config, config, |
5740 | (const nvlist_t * const *)newdevs, ndevs + oldndevs); | |
34dc7c2f BB |
5741 | for (i = 0; i < oldndevs + ndevs; i++) |
5742 | nvlist_free(newdevs[i]); | |
5743 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
5744 | } else { | |
5745 | /* | |
5746 | * Generate a new dev list. | |
5747 | */ | |
65ad5d11 | 5748 | sav->sav_config = fnvlist_alloc(); |
795075e6 PD |
5749 | fnvlist_add_nvlist_array(sav->sav_config, config, |
5750 | (const nvlist_t * const *)devs, ndevs); | |
34dc7c2f BB |
5751 | } |
5752 | } | |
5753 | ||
5754 | /* | |
5755 | * Stop and drop level 2 ARC devices | |
5756 | */ | |
5757 | void | |
5758 | spa_l2cache_drop(spa_t *spa) | |
5759 | { | |
5760 | vdev_t *vd; | |
5761 | int i; | |
5762 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
5763 | ||
5764 | for (i = 0; i < sav->sav_count; i++) { | |
5765 | uint64_t pool; | |
5766 | ||
5767 | vd = sav->sav_vdevs[i]; | |
5768 | ASSERT(vd != NULL); | |
5769 | ||
fb5f0bc8 BB |
5770 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && |
5771 | pool != 0ULL && l2arc_vdev_present(vd)) | |
34dc7c2f | 5772 | l2arc_remove_vdev(vd); |
34dc7c2f BB |
5773 | } |
5774 | } | |
5775 | ||
b5256303 TC |
5776 | /* |
5777 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
5778 | * have the encryption feature flag enabled. | |
5779 | */ | |
5780 | static int | |
5781 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
5782 | boolean_t has_encryption) | |
5783 | { | |
5784 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
5785 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
5786 | !has_encryption) | |
5787 | return (SET_ERROR(ENOTSUP)); | |
5788 | ||
1fff937a | 5789 | return (dmu_objset_create_crypt_check(NULL, dcp, NULL)); |
b5256303 TC |
5790 | } |
5791 | ||
34dc7c2f BB |
5792 | /* |
5793 | * Pool Creation | |
5794 | */ | |
5795 | int | |
5796 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
b5256303 | 5797 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) |
34dc7c2f BB |
5798 | { |
5799 | spa_t *spa; | |
5800 | char *altroot = NULL; | |
5801 | vdev_t *rvd; | |
5802 | dsl_pool_t *dp; | |
5803 | dmu_tx_t *tx; | |
9babb374 | 5804 | int error = 0; |
34dc7c2f BB |
5805 | uint64_t txg = TXG_INITIAL; |
5806 | nvlist_t **spares, **l2cache; | |
5807 | uint_t nspares, nl2cache; | |
b2255edc | 5808 | uint64_t version, obj, ndraid = 0; |
9ae529ec | 5809 | boolean_t has_features; |
b5256303 | 5810 | boolean_t has_encryption; |
715c996d | 5811 | boolean_t has_allocclass; |
b5256303 TC |
5812 | spa_feature_t feat; |
5813 | char *feat_name; | |
83e9986f RY |
5814 | char *poolname; |
5815 | nvlist_t *nvl; | |
5816 | ||
cc99f275 DB |
5817 | if (props == NULL || |
5818 | nvlist_lookup_string(props, "tname", &poolname) != 0) | |
83e9986f | 5819 | poolname = (char *)pool; |
34dc7c2f BB |
5820 | |
5821 | /* | |
5822 | * If this pool already exists, return failure. | |
5823 | */ | |
5824 | mutex_enter(&spa_namespace_lock); | |
83e9986f | 5825 | if (spa_lookup(poolname) != NULL) { |
34dc7c2f | 5826 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 5827 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
5828 | } |
5829 | ||
5830 | /* | |
5831 | * Allocate a new spa_t structure. | |
5832 | */ | |
83e9986f RY |
5833 | nvl = fnvlist_alloc(); |
5834 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
34dc7c2f BB |
5835 | (void) nvlist_lookup_string(props, |
5836 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
83e9986f RY |
5837 | spa = spa_add(poolname, nvl, altroot); |
5838 | fnvlist_free(nvl); | |
fb5f0bc8 | 5839 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 5840 | |
34dc7c2f | 5841 | if (props && (error = spa_prop_validate(spa, props))) { |
34dc7c2f BB |
5842 | spa_deactivate(spa); |
5843 | spa_remove(spa); | |
b128c09f | 5844 | mutex_exit(&spa_namespace_lock); |
34dc7c2f BB |
5845 | return (error); |
5846 | } | |
5847 | ||
83e9986f RY |
5848 | /* |
5849 | * Temporary pool names should never be written to disk. | |
5850 | */ | |
5851 | if (poolname != pool) | |
5852 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
5853 | ||
9ae529ec | 5854 | has_features = B_FALSE; |
b5256303 | 5855 | has_encryption = B_FALSE; |
715c996d | 5856 | has_allocclass = B_FALSE; |
1c27024e | 5857 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); |
9ae529ec | 5858 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { |
b5256303 | 5859 | if (zpool_prop_feature(nvpair_name(elem))) { |
9ae529ec | 5860 | has_features = B_TRUE; |
b5256303 TC |
5861 | |
5862 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
5863 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
5864 | if (feat == SPA_FEATURE_ENCRYPTION) | |
5865 | has_encryption = B_TRUE; | |
715c996d | 5866 | if (feat == SPA_FEATURE_ALLOCATION_CLASSES) |
5867 | has_allocclass = B_TRUE; | |
b5256303 TC |
5868 | } |
5869 | } | |
5870 | ||
5871 | /* verify encryption params, if they were provided */ | |
5872 | if (dcp != NULL) { | |
5873 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
5874 | if (error != 0) { | |
5875 | spa_deactivate(spa); | |
5876 | spa_remove(spa); | |
5877 | mutex_exit(&spa_namespace_lock); | |
5878 | return (error); | |
5879 | } | |
9ae529ec | 5880 | } |
c24fa4b1 | 5881 | if (!has_allocclass && zfs_special_devs(nvroot, NULL)) { |
715c996d | 5882 | spa_deactivate(spa); |
5883 | spa_remove(spa); | |
5884 | mutex_exit(&spa_namespace_lock); | |
5885 | return (ENOTSUP); | |
5886 | } | |
9ae529ec CS |
5887 | |
5888 | if (has_features || nvlist_lookup_uint64(props, | |
5889 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
34dc7c2f | 5890 | version = SPA_VERSION; |
9ae529ec CS |
5891 | } |
5892 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
428870ff BB |
5893 | |
5894 | spa->spa_first_txg = txg; | |
5895 | spa->spa_uberblock.ub_txg = txg - 1; | |
34dc7c2f BB |
5896 | spa->spa_uberblock.ub_version = version; |
5897 | spa->spa_ubsync = spa->spa_uberblock; | |
3dfb57a3 | 5898 | spa->spa_load_state = SPA_LOAD_CREATE; |
a1d477c2 MA |
5899 | spa->spa_removing_phys.sr_state = DSS_NONE; |
5900 | spa->spa_removing_phys.sr_removing_vdev = -1; | |
5901 | spa->spa_removing_phys.sr_prev_indirect_vdev = -1; | |
944a3724 | 5902 | spa->spa_indirect_vdevs_loaded = B_TRUE; |
34dc7c2f | 5903 | |
9babb374 BB |
5904 | /* |
5905 | * Create "The Godfather" zio to hold all async IOs | |
5906 | */ | |
e022864d MA |
5907 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
5908 | KM_SLEEP); | |
1c27024e | 5909 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
5910 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
5911 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
5912 | ZIO_FLAG_GODFATHER); | |
5913 | } | |
9babb374 | 5914 | |
34dc7c2f BB |
5915 | /* |
5916 | * Create the root vdev. | |
5917 | */ | |
b128c09f | 5918 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
5919 | |
5920 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
5921 | ||
5922 | ASSERT(error != 0 || rvd != NULL); | |
5923 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
5924 | ||
5925 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
2e528b49 | 5926 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
5927 | |
5928 | if (error == 0 && | |
5929 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
b2255edc BB |
5930 | (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 && |
5931 | (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { | |
cc99f275 DB |
5932 | /* |
5933 | * instantiate the metaslab groups (this will dirty the vdevs) | |
5934 | * we can no longer error exit past this point | |
5935 | */ | |
5936 | for (int c = 0; error == 0 && c < rvd->vdev_children; c++) { | |
5937 | vdev_t *vd = rvd->vdev_child[c]; | |
5938 | ||
5939 | vdev_metaslab_set_size(vd); | |
5940 | vdev_expand(vd, txg); | |
9babb374 | 5941 | } |
34dc7c2f BB |
5942 | } |
5943 | ||
b128c09f | 5944 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
5945 | |
5946 | if (error != 0) { | |
5947 | spa_unload(spa); | |
5948 | spa_deactivate(spa); | |
5949 | spa_remove(spa); | |
5950 | mutex_exit(&spa_namespace_lock); | |
5951 | return (error); | |
5952 | } | |
5953 | ||
5954 | /* | |
5955 | * Get the list of spares, if specified. | |
5956 | */ | |
5957 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
5958 | &spares, &nspares) == 0) { | |
65ad5d11 AJ |
5959 | spa->spa_spares.sav_config = fnvlist_alloc(); |
5960 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
5961 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
5962 | nspares); | |
b128c09f | 5963 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 5964 | spa_load_spares(spa); |
b128c09f | 5965 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
5966 | spa->spa_spares.sav_sync = B_TRUE; |
5967 | } | |
5968 | ||
5969 | /* | |
5970 | * Get the list of level 2 cache devices, if specified. | |
5971 | */ | |
5972 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
5973 | &l2cache, &nl2cache) == 0) { | |
795075e6 PD |
5974 | VERIFY0(nvlist_alloc(&spa->spa_l2cache.sav_config, |
5975 | NV_UNIQUE_NAME, KM_SLEEP)); | |
65ad5d11 | 5976 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, |
795075e6 PD |
5977 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
5978 | nl2cache); | |
b128c09f | 5979 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 5980 | spa_load_l2cache(spa); |
b128c09f | 5981 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
5982 | spa->spa_l2cache.sav_sync = B_TRUE; |
5983 | } | |
5984 | ||
9ae529ec | 5985 | spa->spa_is_initializing = B_TRUE; |
b5256303 | 5986 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); |
9ae529ec | 5987 | spa->spa_is_initializing = B_FALSE; |
34dc7c2f | 5988 | |
428870ff BB |
5989 | /* |
5990 | * Create DDTs (dedup tables). | |
5991 | */ | |
5992 | ddt_create(spa); | |
67a1b037 PJD |
5993 | /* |
5994 | * Create BRT table and BRT table object. | |
5995 | */ | |
5996 | brt_create(spa); | |
428870ff BB |
5997 | |
5998 | spa_update_dspace(spa); | |
5999 | ||
34dc7c2f BB |
6000 | tx = dmu_tx_create_assigned(dp, txg); |
6001 | ||
d5e024cb BB |
6002 | /* |
6003 | * Create the pool's history object. | |
6004 | */ | |
6005 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
6006 | spa_history_create_obj(spa, tx); | |
6007 | ||
6008 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
6009 | spa_history_log_version(spa, "create", tx); | |
6010 | ||
34dc7c2f BB |
6011 | /* |
6012 | * Create the pool config object. | |
6013 | */ | |
6014 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
b128c09f | 6015 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, |
34dc7c2f BB |
6016 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); |
6017 | ||
6018 | if (zap_add(spa->spa_meta_objset, | |
6019 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
6020 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
6021 | cmn_err(CE_PANIC, "failed to add pool config"); | |
6022 | } | |
6023 | ||
428870ff BB |
6024 | if (zap_add(spa->spa_meta_objset, |
6025 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
6026 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
6027 | cmn_err(CE_PANIC, "failed to add pool version"); | |
6028 | } | |
6029 | ||
34dc7c2f BB |
6030 | /* Newly created pools with the right version are always deflated. */ |
6031 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6032 | spa->spa_deflate = TRUE; | |
6033 | if (zap_add(spa->spa_meta_objset, | |
6034 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6035 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
6036 | cmn_err(CE_PANIC, "failed to add deflate"); | |
6037 | } | |
6038 | } | |
6039 | ||
6040 | /* | |
428870ff | 6041 | * Create the deferred-free bpobj. Turn off compression |
34dc7c2f BB |
6042 | * because sync-to-convergence takes longer if the blocksize |
6043 | * keeps changing. | |
6044 | */ | |
428870ff BB |
6045 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); |
6046 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
34dc7c2f | 6047 | ZIO_COMPRESS_OFF, tx); |
34dc7c2f | 6048 | if (zap_add(spa->spa_meta_objset, |
428870ff BB |
6049 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, |
6050 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
6051 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
34dc7c2f | 6052 | } |
428870ff BB |
6053 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, |
6054 | spa->spa_meta_objset, obj)); | |
34dc7c2f | 6055 | |
3c67d83a TH |
6056 | /* |
6057 | * Generate some random noise for salted checksums to operate on. | |
6058 | */ | |
6059 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
6060 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
6061 | ||
34dc7c2f BB |
6062 | /* |
6063 | * Set pool properties. | |
6064 | */ | |
6065 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
6066 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
6067 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
9babb374 | 6068 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); |
379ca9cf | 6069 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); |
1b939560 | 6070 | spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM); |
428870ff | 6071 | |
d164b209 BB |
6072 | if (props != NULL) { |
6073 | spa_configfile_set(spa, props, B_FALSE); | |
13fe0198 | 6074 | spa_sync_props(props, tx); |
d164b209 | 6075 | } |
34dc7c2f | 6076 | |
b2255edc BB |
6077 | for (int i = 0; i < ndraid; i++) |
6078 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6079 | ||
34dc7c2f BB |
6080 | dmu_tx_commit(tx); |
6081 | ||
6082 | spa->spa_sync_on = B_TRUE; | |
b5256303 | 6083 | txg_sync_start(dp); |
379ca9cf | 6084 | mmp_thread_start(spa); |
b5256303 | 6085 | txg_wait_synced(dp, txg); |
34dc7c2f | 6086 | |
9d5b5245 SD |
6087 | spa_spawn_aux_threads(spa); |
6088 | ||
55c12724 | 6089 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); |
34dc7c2f | 6090 | |
0c66c32d JG |
6091 | /* |
6092 | * Don't count references from objsets that are already closed | |
6093 | * and are making their way through the eviction process. | |
6094 | */ | |
6095 | spa_evicting_os_wait(spa); | |
424fd7c3 | 6096 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
3dfb57a3 | 6097 | spa->spa_load_state = SPA_LOAD_NONE; |
b128c09f | 6098 | |
4759342a JL |
6099 | spa_import_os(spa); |
6100 | ||
d164b209 BB |
6101 | mutex_exit(&spa_namespace_lock); |
6102 | ||
34dc7c2f BB |
6103 | return (0); |
6104 | } | |
6105 | ||
9babb374 BB |
6106 | /* |
6107 | * Import a non-root pool into the system. | |
6108 | */ | |
6109 | int | |
13fe0198 | 6110 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) |
34dc7c2f BB |
6111 | { |
6112 | spa_t *spa; | |
6113 | char *altroot = NULL; | |
428870ff | 6114 | spa_load_state_t state = SPA_LOAD_IMPORT; |
8a393be3 | 6115 | zpool_load_policy_t policy; |
da92d5cb | 6116 | spa_mode_t mode = spa_mode_global; |
572e2857 | 6117 | uint64_t readonly = B_FALSE; |
9babb374 | 6118 | int error; |
34dc7c2f BB |
6119 | nvlist_t *nvroot; |
6120 | nvlist_t **spares, **l2cache; | |
6121 | uint_t nspares, nl2cache; | |
34dc7c2f BB |
6122 | |
6123 | /* | |
6124 | * If a pool with this name exists, return failure. | |
6125 | */ | |
6126 | mutex_enter(&spa_namespace_lock); | |
428870ff | 6127 | if (spa_lookup(pool) != NULL) { |
9babb374 | 6128 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 6129 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
6130 | } |
6131 | ||
6132 | /* | |
6133 | * Create and initialize the spa structure. | |
6134 | */ | |
6135 | (void) nvlist_lookup_string(props, | |
6136 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
572e2857 BB |
6137 | (void) nvlist_lookup_uint64(props, |
6138 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
6139 | if (readonly) | |
da92d5cb | 6140 | mode = SPA_MODE_READ; |
428870ff | 6141 | spa = spa_add(pool, config, altroot); |
572e2857 BB |
6142 | spa->spa_import_flags = flags; |
6143 | ||
6144 | /* | |
6145 | * Verbatim import - Take a pool and insert it into the namespace | |
6146 | * as if it had been loaded at boot. | |
6147 | */ | |
6148 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
6149 | if (props != NULL) | |
6150 | spa_configfile_set(spa, props, B_FALSE); | |
6151 | ||
55c12724 | 6152 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE); |
12fa0466 | 6153 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
4a0ee12a | 6154 | zfs_dbgmsg("spa_import: verbatim import of %s", pool); |
572e2857 | 6155 | mutex_exit(&spa_namespace_lock); |
572e2857 BB |
6156 | return (0); |
6157 | } | |
6158 | ||
6159 | spa_activate(spa, mode); | |
34dc7c2f | 6160 | |
9babb374 BB |
6161 | /* |
6162 | * Don't start async tasks until we know everything is healthy. | |
6163 | */ | |
6164 | spa_async_suspend(spa); | |
b128c09f | 6165 | |
8a393be3 PZ |
6166 | zpool_get_load_policy(config, &policy); |
6167 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) | |
572e2857 BB |
6168 | state = SPA_LOAD_RECOVER; |
6169 | ||
6cb8e530 | 6170 | spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; |
572e2857 | 6171 | |
6cb8e530 PZ |
6172 | if (state != SPA_LOAD_RECOVER) { |
6173 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6174 | zfs_dbgmsg("spa_import: importing %s", pool); | |
6175 | } else { | |
6176 | zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " | |
8a393be3 | 6177 | "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); |
6cb8e530 | 6178 | } |
8a393be3 | 6179 | error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); |
428870ff BB |
6180 | |
6181 | /* | |
572e2857 BB |
6182 | * Propagate anything learned while loading the pool and pass it |
6183 | * back to caller (i.e. rewind info, missing devices, etc). | |
428870ff | 6184 | */ |
65ad5d11 | 6185 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info); |
34dc7c2f | 6186 | |
b128c09f | 6187 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6188 | /* |
9babb374 BB |
6189 | * Toss any existing sparelist, as it doesn't have any validity |
6190 | * anymore, and conflicts with spa_has_spare(). | |
34dc7c2f | 6191 | */ |
9babb374 | 6192 | if (spa->spa_spares.sav_config) { |
34dc7c2f BB |
6193 | nvlist_free(spa->spa_spares.sav_config); |
6194 | spa->spa_spares.sav_config = NULL; | |
6195 | spa_load_spares(spa); | |
6196 | } | |
9babb374 | 6197 | if (spa->spa_l2cache.sav_config) { |
34dc7c2f BB |
6198 | nvlist_free(spa->spa_l2cache.sav_config); |
6199 | spa->spa_l2cache.sav_config = NULL; | |
6200 | spa_load_l2cache(spa); | |
6201 | } | |
6202 | ||
65ad5d11 | 6203 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
b128c09f | 6204 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 6205 | |
d164b209 BB |
6206 | if (props != NULL) |
6207 | spa_configfile_set(spa, props, B_FALSE); | |
6208 | ||
fb5f0bc8 BB |
6209 | if (error != 0 || (props && spa_writeable(spa) && |
6210 | (error = spa_prop_set(spa, props)))) { | |
9babb374 BB |
6211 | spa_unload(spa); |
6212 | spa_deactivate(spa); | |
6213 | spa_remove(spa); | |
34dc7c2f BB |
6214 | mutex_exit(&spa_namespace_lock); |
6215 | return (error); | |
6216 | } | |
6217 | ||
572e2857 BB |
6218 | spa_async_resume(spa); |
6219 | ||
34dc7c2f BB |
6220 | /* |
6221 | * Override any spares and level 2 cache devices as specified by | |
6222 | * the user, as these may have correct device names/devids, etc. | |
6223 | */ | |
6224 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6225 | &spares, &nspares) == 0) { | |
6226 | if (spa->spa_spares.sav_config) | |
65ad5d11 AJ |
6227 | fnvlist_remove(spa->spa_spares.sav_config, |
6228 | ZPOOL_CONFIG_SPARES); | |
34dc7c2f | 6229 | else |
65ad5d11 AJ |
6230 | spa->spa_spares.sav_config = fnvlist_alloc(); |
6231 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
6232 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
6233 | nspares); | |
b128c09f | 6234 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6235 | spa_load_spares(spa); |
b128c09f | 6236 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6237 | spa->spa_spares.sav_sync = B_TRUE; |
6238 | } | |
6239 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6240 | &l2cache, &nl2cache) == 0) { | |
6241 | if (spa->spa_l2cache.sav_config) | |
65ad5d11 AJ |
6242 | fnvlist_remove(spa->spa_l2cache.sav_config, |
6243 | ZPOOL_CONFIG_L2CACHE); | |
34dc7c2f | 6244 | else |
65ad5d11 AJ |
6245 | spa->spa_l2cache.sav_config = fnvlist_alloc(); |
6246 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
795075e6 PD |
6247 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
6248 | nl2cache); | |
b128c09f | 6249 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6250 | spa_load_l2cache(spa); |
b128c09f | 6251 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6252 | spa->spa_l2cache.sav_sync = B_TRUE; |
6253 | } | |
6254 | ||
428870ff BB |
6255 | /* |
6256 | * Check for any removed devices. | |
6257 | */ | |
6258 | if (spa->spa_autoreplace) { | |
6259 | spa_aux_check_removed(&spa->spa_spares); | |
6260 | spa_aux_check_removed(&spa->spa_l2cache); | |
6261 | } | |
6262 | ||
fb5f0bc8 | 6263 | if (spa_writeable(spa)) { |
b128c09f BB |
6264 | /* |
6265 | * Update the config cache to include the newly-imported pool. | |
6266 | */ | |
45d1cae3 | 6267 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
b128c09f | 6268 | } |
34dc7c2f | 6269 | |
34dc7c2f | 6270 | /* |
9babb374 BB |
6271 | * It's possible that the pool was expanded while it was exported. |
6272 | * We kick off an async task to handle this for us. | |
34dc7c2f | 6273 | */ |
9babb374 | 6274 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); |
b128c09f | 6275 | |
d5e024cb | 6276 | spa_history_log_version(spa, "import", NULL); |
fb390aaf | 6277 | |
12fa0466 | 6278 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
fb390aaf | 6279 | |
fb390aaf HR |
6280 | mutex_exit(&spa_namespace_lock); |
6281 | ||
ec213971 | 6282 | zvol_create_minors_recursive(pool); |
4a22ba5b | 6283 | |
4759342a JL |
6284 | spa_import_os(spa); |
6285 | ||
b128c09f BB |
6286 | return (0); |
6287 | } | |
6288 | ||
34dc7c2f BB |
6289 | nvlist_t * |
6290 | spa_tryimport(nvlist_t *tryconfig) | |
6291 | { | |
6292 | nvlist_t *config = NULL; | |
6cb8e530 | 6293 | char *poolname, *cachefile; |
34dc7c2f BB |
6294 | spa_t *spa; |
6295 | uint64_t state; | |
d164b209 | 6296 | int error; |
8a393be3 | 6297 | zpool_load_policy_t policy; |
34dc7c2f BB |
6298 | |
6299 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
6300 | return (NULL); | |
6301 | ||
6302 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
6303 | return (NULL); | |
6304 | ||
6305 | /* | |
6306 | * Create and initialize the spa structure. | |
6307 | */ | |
6308 | mutex_enter(&spa_namespace_lock); | |
428870ff | 6309 | spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); |
da92d5cb | 6310 | spa_activate(spa, SPA_MODE_READ); |
34dc7c2f BB |
6311 | |
6312 | /* | |
8a393be3 | 6313 | * Rewind pool if a max txg was provided. |
34dc7c2f | 6314 | */ |
8a393be3 PZ |
6315 | zpool_get_load_policy(spa->spa_config, &policy); |
6316 | if (policy.zlp_txg != UINT64_MAX) { | |
6317 | spa->spa_load_max_txg = policy.zlp_txg; | |
6cb8e530 PZ |
6318 | spa->spa_extreme_rewind = B_TRUE; |
6319 | zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", | |
8a393be3 | 6320 | poolname, (longlong_t)policy.zlp_txg); |
6cb8e530 PZ |
6321 | } else { |
6322 | zfs_dbgmsg("spa_tryimport: importing %s", poolname); | |
6323 | } | |
6324 | ||
6325 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) | |
6326 | == 0) { | |
6327 | zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); | |
6328 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; | |
6329 | } else { | |
6330 | spa->spa_config_source = SPA_CONFIG_SRC_SCAN; | |
6331 | } | |
6332 | ||
6333 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); | |
34dc7c2f BB |
6334 | |
6335 | /* | |
6336 | * If 'tryconfig' was at least parsable, return the current config. | |
6337 | */ | |
6338 | if (spa->spa_root_vdev != NULL) { | |
34dc7c2f | 6339 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
65ad5d11 AJ |
6340 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname); |
6341 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state); | |
6342 | fnvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
6343 | spa->spa_uberblock.ub_timestamp); | |
6344 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
6345 | spa->spa_load_info); | |
6346 | fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
6347 | spa->spa_errata); | |
34dc7c2f BB |
6348 | |
6349 | /* | |
6350 | * If the bootfs property exists on this pool then we | |
6351 | * copy it out so that external consumers can tell which | |
6352 | * pools are bootable. | |
6353 | */ | |
d164b209 | 6354 | if ((!error || error == EEXIST) && spa->spa_bootfs) { |
79c76d5b | 6355 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6356 | |
6357 | /* | |
6358 | * We have to play games with the name since the | |
6359 | * pool was opened as TRYIMPORT_NAME. | |
6360 | */ | |
b128c09f | 6361 | if (dsl_dsobj_to_dsname(spa_name(spa), |
34dc7c2f BB |
6362 | spa->spa_bootfs, tmpname) == 0) { |
6363 | char *cp; | |
d1d7e268 MK |
6364 | char *dsname; |
6365 | ||
79c76d5b | 6366 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6367 | |
6368 | cp = strchr(tmpname, '/'); | |
6369 | if (cp == NULL) { | |
6370 | (void) strlcpy(dsname, tmpname, | |
6371 | MAXPATHLEN); | |
6372 | } else { | |
6373 | (void) snprintf(dsname, MAXPATHLEN, | |
6374 | "%s/%s", poolname, ++cp); | |
6375 | } | |
65ad5d11 AJ |
6376 | fnvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, |
6377 | dsname); | |
34dc7c2f BB |
6378 | kmem_free(dsname, MAXPATHLEN); |
6379 | } | |
6380 | kmem_free(tmpname, MAXPATHLEN); | |
6381 | } | |
6382 | ||
6383 | /* | |
6384 | * Add the list of hot spares and level 2 cache devices. | |
6385 | */ | |
9babb374 | 6386 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
34dc7c2f BB |
6387 | spa_add_spares(spa, config); |
6388 | spa_add_l2cache(spa, config); | |
9babb374 | 6389 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f BB |
6390 | } |
6391 | ||
6392 | spa_unload(spa); | |
6393 | spa_deactivate(spa); | |
6394 | spa_remove(spa); | |
6395 | mutex_exit(&spa_namespace_lock); | |
6396 | ||
6397 | return (config); | |
6398 | } | |
6399 | ||
6400 | /* | |
6401 | * Pool export/destroy | |
6402 | * | |
6403 | * The act of destroying or exporting a pool is very simple. We make sure there | |
6404 | * is no more pending I/O and any references to the pool are gone. Then, we | |
6405 | * update the pool state and sync all the labels to disk, removing the | |
fb5f0bc8 BB |
6406 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then |
6407 | * we don't sync the labels or remove the configuration cache. | |
34dc7c2f BB |
6408 | */ |
6409 | static int | |
4d55ea81 | 6410 | spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig, |
fb5f0bc8 | 6411 | boolean_t force, boolean_t hardforce) |
34dc7c2f | 6412 | { |
f4f50a70 | 6413 | int error; |
34dc7c2f BB |
6414 | spa_t *spa; |
6415 | ||
6416 | if (oldconfig) | |
6417 | *oldconfig = NULL; | |
6418 | ||
da92d5cb | 6419 | if (!(spa_mode_global & SPA_MODE_WRITE)) |
2e528b49 | 6420 | return (SET_ERROR(EROFS)); |
34dc7c2f BB |
6421 | |
6422 | mutex_enter(&spa_namespace_lock); | |
6423 | if ((spa = spa_lookup(pool)) == NULL) { | |
6424 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 6425 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
6426 | } |
6427 | ||
43a85362 SD |
6428 | if (spa->spa_is_exporting) { |
6429 | /* the pool is being exported by another thread */ | |
6430 | mutex_exit(&spa_namespace_lock); | |
6431 | return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS)); | |
6432 | } | |
6433 | spa->spa_is_exporting = B_TRUE; | |
6434 | ||
34dc7c2f BB |
6435 | /* |
6436 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
6437 | * reacquire the namespace lock, and see if we can export. | |
6438 | */ | |
6439 | spa_open_ref(spa, FTAG); | |
6440 | mutex_exit(&spa_namespace_lock); | |
6441 | spa_async_suspend(spa); | |
a0bd735a BP |
6442 | if (spa->spa_zvol_taskq) { |
6443 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
6444 | taskq_wait(spa->spa_zvol_taskq); | |
6445 | } | |
34dc7c2f BB |
6446 | mutex_enter(&spa_namespace_lock); |
6447 | spa_close(spa, FTAG); | |
6448 | ||
d14cfd83 IH |
6449 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) |
6450 | goto export_spa; | |
34dc7c2f | 6451 | /* |
d14cfd83 IH |
6452 | * The pool will be in core if it's openable, in which case we can |
6453 | * modify its state. Objsets may be open only because they're dirty, | |
6454 | * so we have to force it to sync before checking spa_refcnt. | |
34dc7c2f | 6455 | */ |
0c66c32d | 6456 | if (spa->spa_sync_on) { |
34dc7c2f | 6457 | txg_wait_synced(spa->spa_dsl_pool, 0); |
0c66c32d JG |
6458 | spa_evicting_os_wait(spa); |
6459 | } | |
34dc7c2f | 6460 | |
d14cfd83 IH |
6461 | /* |
6462 | * A pool cannot be exported or destroyed if there are active | |
6463 | * references. If we are resetting a pool, allow references by | |
6464 | * fault injection handlers. | |
6465 | */ | |
f4f50a70 WA |
6466 | if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0)) { |
6467 | error = SET_ERROR(EBUSY); | |
6468 | goto fail; | |
d14cfd83 | 6469 | } |
34dc7c2f | 6470 | |
d14cfd83 | 6471 | if (spa->spa_sync_on) { |
88b199c2 | 6472 | vdev_t *rvd = spa->spa_root_vdev; |
b128c09f BB |
6473 | /* |
6474 | * A pool cannot be exported if it has an active shared spare. | |
6475 | * This is to prevent other pools stealing the active spare | |
6476 | * from an exported pool. At user's own will, such pool can | |
6477 | * be forcedly exported. | |
6478 | */ | |
6479 | if (!force && new_state == POOL_STATE_EXPORTED && | |
6480 | spa_has_active_shared_spare(spa)) { | |
f4f50a70 WA |
6481 | error = SET_ERROR(EXDEV); |
6482 | goto fail; | |
b128c09f | 6483 | } |
34dc7c2f | 6484 | |
619f0976 GW |
6485 | /* |
6486 | * We're about to export or destroy this pool. Make sure | |
1b939560 BB |
6487 | * we stop all initialization and trim activity here before |
6488 | * we set the spa_final_txg. This will ensure that all | |
619f0976 GW |
6489 | * dirty data resulting from the initialization is |
6490 | * committed to disk before we unload the pool. | |
6491 | */ | |
88b199c2 RY |
6492 | vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE); |
6493 | vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE); | |
6494 | vdev_autotrim_stop_all(spa); | |
6495 | vdev_rebuild_stop_all(spa); | |
619f0976 | 6496 | |
34dc7c2f BB |
6497 | /* |
6498 | * We want this to be reflected on every label, | |
6499 | * so mark them all dirty. spa_unload() will do the | |
6500 | * final sync that pushes these changes out. | |
6501 | */ | |
fb5f0bc8 | 6502 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { |
b128c09f | 6503 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6504 | spa->spa_state = new_state; |
88b199c2 | 6505 | vdev_config_dirty(rvd); |
2fb52853 GA |
6506 | spa_config_exit(spa, SCL_ALL, FTAG); |
6507 | } | |
6508 | ||
6509 | /* | |
6510 | * If the log space map feature is enabled and the pool is | |
6511 | * getting exported (but not destroyed), we want to spend some | |
6512 | * time flushing as many metaslabs as we can in an attempt to | |
6513 | * destroy log space maps and save import time. This has to be | |
6514 | * done before we set the spa_final_txg, otherwise | |
6515 | * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs. | |
6516 | * spa_should_flush_logs_on_unload() should be called after | |
6517 | * spa_state has been set to the new_state. | |
6518 | */ | |
6519 | if (spa_should_flush_logs_on_unload(spa)) | |
6520 | spa_unload_log_sm_flush_all(spa); | |
6521 | ||
6522 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
6523 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
428870ff BB |
6524 | spa->spa_final_txg = spa_last_synced_txg(spa) + |
6525 | TXG_DEFER_SIZE + 1; | |
b128c09f | 6526 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6527 | } |
6528 | } | |
6529 | ||
d14cfd83 | 6530 | export_spa: |
4759342a JL |
6531 | spa_export_os(spa); |
6532 | ||
d5e024cb BB |
6533 | if (new_state == POOL_STATE_DESTROYED) |
6534 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
6535 | else if (new_state == POOL_STATE_EXPORTED) | |
6536 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
34dc7c2f BB |
6537 | |
6538 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
6539 | spa_unload(spa); | |
6540 | spa_deactivate(spa); | |
6541 | } | |
6542 | ||
6543 | if (oldconfig && spa->spa_config) | |
65ad5d11 | 6544 | *oldconfig = fnvlist_dup(spa->spa_config); |
34dc7c2f BB |
6545 | |
6546 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
fb5f0bc8 | 6547 | if (!hardforce) |
55c12724 | 6548 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 6549 | spa_remove(spa); |
43a85362 SD |
6550 | } else { |
6551 | /* | |
6552 | * If spa_remove() is not called for this spa_t and | |
6553 | * there is any possibility that it can be reused, | |
6554 | * we make sure to reset the exporting flag. | |
6555 | */ | |
6556 | spa->spa_is_exporting = B_FALSE; | |
34dc7c2f | 6557 | } |
34dc7c2f | 6558 | |
43a85362 | 6559 | mutex_exit(&spa_namespace_lock); |
34dc7c2f | 6560 | return (0); |
f4f50a70 WA |
6561 | |
6562 | fail: | |
6563 | spa->spa_is_exporting = B_FALSE; | |
6564 | spa_async_resume(spa); | |
6565 | mutex_exit(&spa_namespace_lock); | |
6566 | return (error); | |
34dc7c2f BB |
6567 | } |
6568 | ||
6569 | /* | |
6570 | * Destroy a storage pool. | |
6571 | */ | |
6572 | int | |
4d55ea81 | 6573 | spa_destroy(const char *pool) |
34dc7c2f | 6574 | { |
fb5f0bc8 BB |
6575 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, |
6576 | B_FALSE, B_FALSE)); | |
34dc7c2f BB |
6577 | } |
6578 | ||
6579 | /* | |
6580 | * Export a storage pool. | |
6581 | */ | |
6582 | int | |
4d55ea81 | 6583 | spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force, |
fb5f0bc8 | 6584 | boolean_t hardforce) |
34dc7c2f | 6585 | { |
fb5f0bc8 BB |
6586 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, |
6587 | force, hardforce)); | |
34dc7c2f BB |
6588 | } |
6589 | ||
6590 | /* | |
6591 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
6592 | * from the namespace in any way. | |
6593 | */ | |
6594 | int | |
4d55ea81 | 6595 | spa_reset(const char *pool) |
34dc7c2f | 6596 | { |
b128c09f | 6597 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, |
fb5f0bc8 | 6598 | B_FALSE, B_FALSE)); |
34dc7c2f BB |
6599 | } |
6600 | ||
34dc7c2f BB |
6601 | /* |
6602 | * ========================================================================== | |
6603 | * Device manipulation | |
6604 | * ========================================================================== | |
6605 | */ | |
6606 | ||
b2255edc BB |
6607 | /* |
6608 | * This is called as a synctask to increment the draid feature flag | |
6609 | */ | |
6610 | static void | |
6611 | spa_draid_feature_incr(void *arg, dmu_tx_t *tx) | |
6612 | { | |
6613 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6614 | int draid = (int)(uintptr_t)arg; | |
6615 | ||
6616 | for (int c = 0; c < draid; c++) | |
6617 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6618 | } | |
6619 | ||
34dc7c2f BB |
6620 | /* |
6621 | * Add a device to a storage pool. | |
6622 | */ | |
6623 | int | |
6624 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
6625 | { | |
b2255edc | 6626 | uint64_t txg, ndraid = 0; |
fb5f0bc8 | 6627 | int error; |
34dc7c2f BB |
6628 | vdev_t *rvd = spa->spa_root_vdev; |
6629 | vdev_t *vd, *tvd; | |
6630 | nvlist_t **spares, **l2cache; | |
6631 | uint_t nspares, nl2cache; | |
6632 | ||
572e2857 BB |
6633 | ASSERT(spa_writeable(spa)); |
6634 | ||
34dc7c2f BB |
6635 | txg = spa_vdev_enter(spa); |
6636 | ||
6637 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
6638 | VDEV_ALLOC_ADD)) != 0) | |
6639 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
6640 | ||
b128c09f | 6641 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ |
34dc7c2f BB |
6642 | |
6643 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
6644 | &nspares) != 0) | |
6645 | nspares = 0; | |
6646 | ||
6647 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
6648 | &nl2cache) != 0) | |
6649 | nl2cache = 0; | |
6650 | ||
b128c09f | 6651 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) |
34dc7c2f | 6652 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
34dc7c2f | 6653 | |
b128c09f | 6654 | if (vd->vdev_children != 0 && |
b2255edc | 6655 | (error = vdev_create(vd, txg, B_FALSE)) != 0) { |
b128c09f | 6656 | return (spa_vdev_exit(spa, vd, txg, error)); |
b2255edc BB |
6657 | } |
6658 | ||
6659 | /* | |
6660 | * The virtual dRAID spares must be added after vdev tree is created | |
bf169e9f | 6661 | * and the vdev guids are generated. The guid of their associated |
b2255edc BB |
6662 | * dRAID is stored in the config and used when opening the spare. |
6663 | */ | |
6664 | if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid, | |
6665 | rvd->vdev_children)) == 0) { | |
6666 | if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot, | |
6667 | ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) | |
6668 | nspares = 0; | |
6669 | } else { | |
6670 | return (spa_vdev_exit(spa, vd, txg, error)); | |
6671 | } | |
34dc7c2f BB |
6672 | |
6673 | /* | |
6674 | * We must validate the spares and l2cache devices after checking the | |
6675 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
6676 | */ | |
b128c09f | 6677 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) |
34dc7c2f | 6678 | return (spa_vdev_exit(spa, vd, txg, error)); |
34dc7c2f BB |
6679 | |
6680 | /* | |
a1d477c2 MA |
6681 | * If we are in the middle of a device removal, we can only add |
6682 | * devices which match the existing devices in the pool. | |
6683 | * If we are in the middle of a removal, or have some indirect | |
b2255edc | 6684 | * vdevs, we can not add raidz or dRAID top levels. |
34dc7c2f | 6685 | */ |
a1d477c2 MA |
6686 | if (spa->spa_vdev_removal != NULL || |
6687 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
6688 | for (int c = 0; c < vd->vdev_children; c++) { | |
6689 | tvd = vd->vdev_child[c]; | |
6690 | if (spa->spa_vdev_removal != NULL && | |
9e052db4 | 6691 | tvd->vdev_ashift != spa->spa_max_ashift) { |
a1d477c2 MA |
6692 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
6693 | } | |
b2255edc BB |
6694 | /* Fail if top level vdev is raidz or a dRAID */ |
6695 | if (vdev_get_nparity(tvd) != 0) | |
a1d477c2 | 6696 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
b2255edc | 6697 | |
a1d477c2 MA |
6698 | /* |
6699 | * Need the top level mirror to be | |
6700 | * a mirror of leaf vdevs only | |
6701 | */ | |
6702 | if (tvd->vdev_ops == &vdev_mirror_ops) { | |
6703 | for (uint64_t cid = 0; | |
6704 | cid < tvd->vdev_children; cid++) { | |
6705 | vdev_t *cvd = tvd->vdev_child[cid]; | |
6706 | if (!cvd->vdev_ops->vdev_op_leaf) { | |
6707 | return (spa_vdev_exit(spa, vd, | |
6708 | txg, EINVAL)); | |
6709 | } | |
6710 | } | |
6711 | } | |
6712 | } | |
6713 | } | |
6714 | ||
1c27024e | 6715 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
6716 | tvd = vd->vdev_child[c]; |
6717 | vdev_remove_child(vd, tvd); | |
93e28d66 | 6718 | tvd->vdev_id = rvd->vdev_children; |
34dc7c2f BB |
6719 | vdev_add_child(rvd, tvd); |
6720 | vdev_config_dirty(tvd); | |
6721 | } | |
6722 | ||
6723 | if (nspares != 0) { | |
6724 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
6725 | ZPOOL_CONFIG_SPARES); | |
6726 | spa_load_spares(spa); | |
6727 | spa->spa_spares.sav_sync = B_TRUE; | |
6728 | } | |
6729 | ||
6730 | if (nl2cache != 0) { | |
6731 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
6732 | ZPOOL_CONFIG_L2CACHE); | |
6733 | spa_load_l2cache(spa); | |
6734 | spa->spa_l2cache.sav_sync = B_TRUE; | |
6735 | } | |
6736 | ||
b2255edc BB |
6737 | /* |
6738 | * We can't increment a feature while holding spa_vdev so we | |
6739 | * have to do it in a synctask. | |
6740 | */ | |
6741 | if (ndraid != 0) { | |
6742 | dmu_tx_t *tx; | |
6743 | ||
6744 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
6745 | dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr, | |
6746 | (void *)(uintptr_t)ndraid, tx); | |
6747 | dmu_tx_commit(tx); | |
6748 | } | |
6749 | ||
34dc7c2f BB |
6750 | /* |
6751 | * We have to be careful when adding new vdevs to an existing pool. | |
6752 | * If other threads start allocating from these vdevs before we | |
6753 | * sync the config cache, and we lose power, then upon reboot we may | |
6754 | * fail to open the pool because there are DVAs that the config cache | |
6755 | * can't translate. Therefore, we first add the vdevs without | |
6756 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
6757 | * and then let spa_config_update() initialize the new metaslabs. | |
6758 | * | |
6759 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
6760 | * if we lose power at any point in this sequence, the remaining | |
6761 | * steps will be completed the next time we load the pool. | |
6762 | */ | |
6763 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
6764 | ||
6765 | mutex_enter(&spa_namespace_lock); | |
6766 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
12fa0466 | 6767 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); |
34dc7c2f BB |
6768 | mutex_exit(&spa_namespace_lock); |
6769 | ||
6770 | return (0); | |
6771 | } | |
6772 | ||
6773 | /* | |
6774 | * Attach a device to a mirror. The arguments are the path to any device | |
6775 | * in the mirror, and the nvroot for the new device. If the path specifies | |
6776 | * a device that is not mirrored, we automatically insert the mirror vdev. | |
6777 | * | |
6778 | * If 'replacing' is specified, the new device is intended to replace the | |
6779 | * existing device; in this case the two devices are made into their own | |
6780 | * mirror using the 'replacing' vdev, which is functionally identical to | |
6781 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
6782 | * extra rules: you can't attach to it after it's been created, and upon | |
6783 | * completion of resilvering, the first disk (the one being replaced) | |
6784 | * is automatically detached. | |
9a49d3f3 BB |
6785 | * |
6786 | * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild) | |
6787 | * should be performed instead of traditional healing reconstruction. From | |
6788 | * an administrators perspective these are both resilver operations. | |
34dc7c2f BB |
6789 | */ |
6790 | int | |
9a49d3f3 BB |
6791 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing, |
6792 | int rebuild) | |
34dc7c2f | 6793 | { |
428870ff | 6794 | uint64_t txg, dtl_max_txg; |
9a49d3f3 | 6795 | vdev_t *rvd = spa->spa_root_vdev; |
34dc7c2f BB |
6796 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; |
6797 | vdev_ops_t *pvops; | |
b128c09f BB |
6798 | char *oldvdpath, *newvdpath; |
6799 | int newvd_isspare; | |
6800 | int error; | |
34dc7c2f | 6801 | |
572e2857 BB |
6802 | ASSERT(spa_writeable(spa)); |
6803 | ||
34dc7c2f BB |
6804 | txg = spa_vdev_enter(spa); |
6805 | ||
b128c09f | 6806 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 6807 | |
d2734cce SD |
6808 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
6809 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
6810 | error = (spa_has_checkpoint(spa)) ? | |
6811 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
6812 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
6813 | } | |
6814 | ||
9a49d3f3 BB |
6815 | if (rebuild) { |
6816 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD)) | |
6817 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
6818 | ||
6819 | if (dsl_scan_resilvering(spa_get_dsl(spa))) | |
6820 | return (spa_vdev_exit(spa, NULL, txg, | |
6821 | ZFS_ERR_RESILVER_IN_PROGRESS)); | |
6822 | } else { | |
6823 | if (vdev_rebuild_active(rvd)) | |
6824 | return (spa_vdev_exit(spa, NULL, txg, | |
6825 | ZFS_ERR_REBUILD_IN_PROGRESS)); | |
6826 | } | |
6827 | ||
9e052db4 | 6828 | if (spa->spa_vdev_removal != NULL) |
a1d477c2 | 6829 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
a1d477c2 | 6830 | |
34dc7c2f BB |
6831 | if (oldvd == NULL) |
6832 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
6833 | ||
6834 | if (!oldvd->vdev_ops->vdev_op_leaf) | |
6835 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
6836 | ||
6837 | pvd = oldvd->vdev_parent; | |
6838 | ||
6a42939f RY |
6839 | if (spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, |
6840 | VDEV_ALLOC_ATTACH) != 0) | |
34dc7c2f BB |
6841 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); |
6842 | ||
6843 | if (newrootvd->vdev_children != 1) | |
6844 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
6845 | ||
6846 | newvd = newrootvd->vdev_child[0]; | |
6847 | ||
6848 | if (!newvd->vdev_ops->vdev_op_leaf) | |
6849 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
6850 | ||
6851 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
6852 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
6853 | ||
6854 | /* | |
c23738c7 | 6855 | * log, dedup and special vdevs should not be replaced by spares. |
34dc7c2f | 6856 | */ |
c23738c7 AH |
6857 | if ((oldvd->vdev_top->vdev_alloc_bias != VDEV_BIAS_NONE || |
6858 | oldvd->vdev_top->vdev_islog) && newvd->vdev_isspare) { | |
34dc7c2f | 6859 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
c23738c7 | 6860 | } |
34dc7c2f | 6861 | |
b2255edc BB |
6862 | /* |
6863 | * A dRAID spare can only replace a child of its parent dRAID vdev. | |
6864 | */ | |
6865 | if (newvd->vdev_ops == &vdev_draid_spare_ops && | |
6866 | oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) { | |
6867 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
6868 | } | |
6869 | ||
9a49d3f3 BB |
6870 | if (rebuild) { |
6871 | /* | |
b2255edc | 6872 | * For rebuilds, the top vdev must support reconstruction |
9a49d3f3 | 6873 | * using only space maps. This means the only allowable |
b2255edc | 6874 | * vdevs types are the root vdev, a mirror, or dRAID. |
9a49d3f3 | 6875 | */ |
b2255edc BB |
6876 | tvd = pvd; |
6877 | if (pvd->vdev_top != NULL) | |
6878 | tvd = pvd->vdev_top; | |
6879 | ||
6880 | if (tvd->vdev_ops != &vdev_mirror_ops && | |
6881 | tvd->vdev_ops != &vdev_root_ops && | |
6882 | tvd->vdev_ops != &vdev_draid_ops) { | |
9a49d3f3 BB |
6883 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
6884 | } | |
6885 | } | |
6886 | ||
34dc7c2f BB |
6887 | if (!replacing) { |
6888 | /* | |
6889 | * For attach, the only allowable parent is a mirror or the root | |
6890 | * vdev. | |
6891 | */ | |
6892 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
6893 | pvd->vdev_ops != &vdev_root_ops) | |
6894 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
6895 | ||
6896 | pvops = &vdev_mirror_ops; | |
6897 | } else { | |
6898 | /* | |
6899 | * Active hot spares can only be replaced by inactive hot | |
6900 | * spares. | |
6901 | */ | |
6902 | if (pvd->vdev_ops == &vdev_spare_ops && | |
572e2857 | 6903 | oldvd->vdev_isspare && |
34dc7c2f BB |
6904 | !spa_has_spare(spa, newvd->vdev_guid)) |
6905 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
6906 | ||
6907 | /* | |
6908 | * If the source is a hot spare, and the parent isn't already a | |
6909 | * spare, then we want to create a new hot spare. Otherwise, we | |
6910 | * want to create a replacing vdev. The user is not allowed to | |
6911 | * attach to a spared vdev child unless the 'isspare' state is | |
6912 | * the same (spare replaces spare, non-spare replaces | |
6913 | * non-spare). | |
6914 | */ | |
572e2857 BB |
6915 | if (pvd->vdev_ops == &vdev_replacing_ops && |
6916 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
34dc7c2f | 6917 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
6918 | } else if (pvd->vdev_ops == &vdev_spare_ops && |
6919 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
34dc7c2f | 6920 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
6921 | } |
6922 | ||
6923 | if (newvd->vdev_isspare) | |
34dc7c2f BB |
6924 | pvops = &vdev_spare_ops; |
6925 | else | |
6926 | pvops = &vdev_replacing_ops; | |
6927 | } | |
6928 | ||
6929 | /* | |
9babb374 | 6930 | * Make sure the new device is big enough. |
34dc7c2f | 6931 | */ |
9babb374 | 6932 | if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) |
34dc7c2f BB |
6933 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
6934 | ||
6935 | /* | |
6936 | * The new device cannot have a higher alignment requirement | |
6937 | * than the top-level vdev. | |
6938 | */ | |
6939 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
9a49d3f3 | 6940 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
34dc7c2f BB |
6941 | |
6942 | /* | |
6943 | * If this is an in-place replacement, update oldvd's path and devid | |
6944 | * to make it distinguishable from newvd, and unopenable from now on. | |
6945 | */ | |
6946 | if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
6947 | spa_strfree(oldvd->vdev_path); | |
6948 | oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
79c76d5b | 6949 | KM_SLEEP); |
c9e319fa JL |
6950 | (void) snprintf(oldvd->vdev_path, strlen(newvd->vdev_path) + 5, |
6951 | "%s/%s", newvd->vdev_path, "old"); | |
34dc7c2f BB |
6952 | if (oldvd->vdev_devid != NULL) { |
6953 | spa_strfree(oldvd->vdev_devid); | |
6954 | oldvd->vdev_devid = NULL; | |
6955 | } | |
6956 | } | |
6957 | ||
6958 | /* | |
6959 | * If the parent is not a mirror, or if we're replacing, insert the new | |
6960 | * mirror/replacing/spare vdev above oldvd. | |
6961 | */ | |
6962 | if (pvd->vdev_ops != pvops) | |
6963 | pvd = vdev_add_parent(oldvd, pvops); | |
6964 | ||
6965 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
6966 | ASSERT(pvd->vdev_ops == pvops); | |
6967 | ASSERT(oldvd->vdev_parent == pvd); | |
6968 | ||
6969 | /* | |
6970 | * Extract the new device from its root and add it to pvd. | |
6971 | */ | |
6972 | vdev_remove_child(newrootvd, newvd); | |
6973 | newvd->vdev_id = pvd->vdev_children; | |
428870ff | 6974 | newvd->vdev_crtxg = oldvd->vdev_crtxg; |
34dc7c2f BB |
6975 | vdev_add_child(pvd, newvd); |
6976 | ||
6d82f98c IH |
6977 | /* |
6978 | * Reevaluate the parent vdev state. | |
6979 | */ | |
6980 | vdev_propagate_state(pvd); | |
6981 | ||
34dc7c2f BB |
6982 | tvd = newvd->vdev_top; |
6983 | ASSERT(pvd->vdev_top == tvd); | |
6984 | ASSERT(tvd->vdev_parent == rvd); | |
6985 | ||
6986 | vdev_config_dirty(tvd); | |
6987 | ||
6988 | /* | |
428870ff BB |
6989 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account |
6990 | * for any dmu_sync-ed blocks. It will propagate upward when | |
6991 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
34dc7c2f | 6992 | */ |
428870ff | 6993 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; |
34dc7c2f | 6994 | |
9a49d3f3 BB |
6995 | vdev_dtl_dirty(newvd, DTL_MISSING, |
6996 | TXG_INITIAL, dtl_max_txg - TXG_INITIAL); | |
34dc7c2f | 6997 | |
9babb374 | 6998 | if (newvd->vdev_isspare) { |
34dc7c2f | 6999 | spa_spare_activate(newvd); |
12fa0466 | 7000 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); |
9babb374 BB |
7001 | } |
7002 | ||
b128c09f BB |
7003 | oldvdpath = spa_strdup(oldvd->vdev_path); |
7004 | newvdpath = spa_strdup(newvd->vdev_path); | |
7005 | newvd_isspare = newvd->vdev_isspare; | |
34dc7c2f BB |
7006 | |
7007 | /* | |
7008 | * Mark newvd's DTL dirty in this txg. | |
7009 | */ | |
7010 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
7011 | ||
428870ff | 7012 | /* |
9a49d3f3 BB |
7013 | * Schedule the resilver or rebuild to restart in the future. We do |
7014 | * this to ensure that dmu_sync-ed blocks have been stitched into the | |
7015 | * respective datasets. | |
428870ff | 7016 | */ |
9a49d3f3 BB |
7017 | if (rebuild) { |
7018 | newvd->vdev_rebuild_txg = txg; | |
7019 | ||
7020 | vdev_rebuild(tvd); | |
7021 | } else { | |
7022 | newvd->vdev_resilver_txg = txg; | |
7023 | ||
7024 | if (dsl_scan_resilvering(spa_get_dsl(spa)) && | |
7025 | spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) { | |
7026 | vdev_defer_resilver(newvd); | |
7027 | } else { | |
7028 | dsl_scan_restart_resilver(spa->spa_dsl_pool, | |
7029 | dtl_max_txg); | |
7030 | } | |
7031 | } | |
428870ff | 7032 | |
fb390aaf | 7033 | if (spa->spa_bootfs) |
12fa0466 | 7034 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); |
fb390aaf | 7035 | |
12fa0466 | 7036 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); |
fb390aaf | 7037 | |
428870ff BB |
7038 | /* |
7039 | * Commit the config | |
7040 | */ | |
7041 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
34dc7c2f | 7042 | |
6f1ffb06 | 7043 | spa_history_log_internal(spa, "vdev attach", NULL, |
428870ff | 7044 | "%s vdev=%s %s vdev=%s", |
45d1cae3 BB |
7045 | replacing && newvd_isspare ? "spare in" : |
7046 | replacing ? "replace" : "attach", newvdpath, | |
7047 | replacing ? "for" : "to", oldvdpath); | |
b128c09f BB |
7048 | |
7049 | spa_strfree(oldvdpath); | |
7050 | spa_strfree(newvdpath); | |
7051 | ||
34dc7c2f BB |
7052 | return (0); |
7053 | } | |
7054 | ||
7055 | /* | |
7056 | * Detach a device from a mirror or replacing vdev. | |
d3cc8b15 | 7057 | * |
34dc7c2f BB |
7058 | * If 'replace_done' is specified, only detach if the parent |
7059 | * is a replacing vdev. | |
7060 | */ | |
7061 | int | |
fb5f0bc8 | 7062 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) |
34dc7c2f BB |
7063 | { |
7064 | uint64_t txg; | |
fb5f0bc8 | 7065 | int error; |
2a8ba608 | 7066 | vdev_t *rvd __maybe_unused = spa->spa_root_vdev; |
34dc7c2f BB |
7067 | vdev_t *vd, *pvd, *cvd, *tvd; |
7068 | boolean_t unspare = B_FALSE; | |
d4ed6673 | 7069 | uint64_t unspare_guid = 0; |
428870ff | 7070 | char *vdpath; |
1c27024e | 7071 | |
572e2857 BB |
7072 | ASSERT(spa_writeable(spa)); |
7073 | ||
9a49d3f3 | 7074 | txg = spa_vdev_detach_enter(spa, guid); |
34dc7c2f | 7075 | |
b128c09f | 7076 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 7077 | |
d2734cce SD |
7078 | /* |
7079 | * Besides being called directly from the userland through the | |
7080 | * ioctl interface, spa_vdev_detach() can be potentially called | |
7081 | * at the end of spa_vdev_resilver_done(). | |
7082 | * | |
7083 | * In the regular case, when we have a checkpoint this shouldn't | |
7084 | * happen as we never empty the DTLs of a vdev during the scrub | |
7085 | * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() | |
7086 | * should never get here when we have a checkpoint. | |
7087 | * | |
7088 | * That said, even in a case when we checkpoint the pool exactly | |
7089 | * as spa_vdev_resilver_done() calls this function everything | |
7090 | * should be fine as the resilver will return right away. | |
7091 | */ | |
7092 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7093 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7094 | error = (spa_has_checkpoint(spa)) ? | |
7095 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7096 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7097 | } | |
7098 | ||
34dc7c2f BB |
7099 | if (vd == NULL) |
7100 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7101 | ||
7102 | if (!vd->vdev_ops->vdev_op_leaf) | |
7103 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7104 | ||
7105 | pvd = vd->vdev_parent; | |
7106 | ||
fb5f0bc8 BB |
7107 | /* |
7108 | * If the parent/child relationship is not as expected, don't do it. | |
7109 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
7110 | * vdev that's replacing B with C. The user's intent in replacing | |
7111 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
7112 | * the replace by detaching C, the expected behavior is to end up | |
7113 | * M(A,B). But suppose that right after deciding to detach C, | |
7114 | * the replacement of B completes. We would have M(A,C), and then | |
7115 | * ask to detach C, which would leave us with just A -- not what | |
7116 | * the user wanted. To prevent this, we make sure that the | |
7117 | * parent/child relationship hasn't changed -- in this example, | |
7118 | * that C's parent is still the replacing vdev R. | |
7119 | */ | |
7120 | if (pvd->vdev_guid != pguid && pguid != 0) | |
7121 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
7122 | ||
34dc7c2f | 7123 | /* |
572e2857 | 7124 | * Only 'replacing' or 'spare' vdevs can be replaced. |
34dc7c2f | 7125 | */ |
572e2857 BB |
7126 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && |
7127 | pvd->vdev_ops != &vdev_spare_ops) | |
7128 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
34dc7c2f BB |
7129 | |
7130 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
7131 | spa_version(spa) >= SPA_VERSION_SPARES); | |
7132 | ||
7133 | /* | |
7134 | * Only mirror, replacing, and spare vdevs support detach. | |
7135 | */ | |
7136 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
7137 | pvd->vdev_ops != &vdev_mirror_ops && | |
7138 | pvd->vdev_ops != &vdev_spare_ops) | |
7139 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7140 | ||
7141 | /* | |
fb5f0bc8 BB |
7142 | * If this device has the only valid copy of some data, |
7143 | * we cannot safely detach it. | |
34dc7c2f | 7144 | */ |
fb5f0bc8 | 7145 | if (vdev_dtl_required(vd)) |
34dc7c2f BB |
7146 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
7147 | ||
fb5f0bc8 | 7148 | ASSERT(pvd->vdev_children >= 2); |
34dc7c2f | 7149 | |
b128c09f BB |
7150 | /* |
7151 | * If we are detaching the second disk from a replacing vdev, then | |
7152 | * check to see if we changed the original vdev's path to have "/old" | |
7153 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
7154 | */ | |
572e2857 BB |
7155 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && |
7156 | vd->vdev_path != NULL) { | |
7157 | size_t len = strlen(vd->vdev_path); | |
7158 | ||
1c27024e | 7159 | for (int c = 0; c < pvd->vdev_children; c++) { |
572e2857 BB |
7160 | cvd = pvd->vdev_child[c]; |
7161 | ||
7162 | if (cvd == vd || cvd->vdev_path == NULL) | |
7163 | continue; | |
7164 | ||
7165 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
7166 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
7167 | spa_strfree(cvd->vdev_path); | |
7168 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
7169 | break; | |
7170 | } | |
b128c09f BB |
7171 | } |
7172 | } | |
7173 | ||
34dc7c2f | 7174 | /* |
b2255edc BB |
7175 | * If we are detaching the original disk from a normal spare, then it |
7176 | * implies that the spare should become a real disk, and be removed | |
7177 | * from the active spare list for the pool. dRAID spares on the | |
7178 | * other hand are coupled to the pool and thus should never be removed | |
7179 | * from the spares list. | |
34dc7c2f | 7180 | */ |
b2255edc BB |
7181 | if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) { |
7182 | vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
7183 | ||
7184 | if (last_cvd->vdev_isspare && | |
7185 | last_cvd->vdev_ops != &vdev_draid_spare_ops) { | |
7186 | unspare = B_TRUE; | |
7187 | } | |
7188 | } | |
34dc7c2f BB |
7189 | |
7190 | /* | |
7191 | * Erase the disk labels so the disk can be used for other things. | |
7192 | * This must be done after all other error cases are handled, | |
7193 | * but before we disembowel vd (so we can still do I/O to it). | |
7194 | * But if we can't do it, don't treat the error as fatal -- | |
7195 | * it may be that the unwritability of the disk is the reason | |
7196 | * it's being detached! | |
7197 | */ | |
6a42939f | 7198 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); |
34dc7c2f BB |
7199 | |
7200 | /* | |
7201 | * Remove vd from its parent and compact the parent's children. | |
7202 | */ | |
7203 | vdev_remove_child(pvd, vd); | |
7204 | vdev_compact_children(pvd); | |
7205 | ||
7206 | /* | |
7207 | * Remember one of the remaining children so we can get tvd below. | |
7208 | */ | |
572e2857 | 7209 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; |
34dc7c2f BB |
7210 | |
7211 | /* | |
7212 | * If we need to remove the remaining child from the list of hot spares, | |
fb5f0bc8 BB |
7213 | * do it now, marking the vdev as no longer a spare in the process. |
7214 | * We must do this before vdev_remove_parent(), because that can | |
7215 | * change the GUID if it creates a new toplevel GUID. For a similar | |
7216 | * reason, we must remove the spare now, in the same txg as the detach; | |
7217 | * otherwise someone could attach a new sibling, change the GUID, and | |
7218 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
34dc7c2f BB |
7219 | */ |
7220 | if (unspare) { | |
7221 | ASSERT(cvd->vdev_isspare); | |
7222 | spa_spare_remove(cvd); | |
7223 | unspare_guid = cvd->vdev_guid; | |
fb5f0bc8 | 7224 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); |
572e2857 | 7225 | cvd->vdev_unspare = B_TRUE; |
34dc7c2f BB |
7226 | } |
7227 | ||
428870ff BB |
7228 | /* |
7229 | * If the parent mirror/replacing vdev only has one child, | |
7230 | * the parent is no longer needed. Remove it from the tree. | |
7231 | */ | |
572e2857 BB |
7232 | if (pvd->vdev_children == 1) { |
7233 | if (pvd->vdev_ops == &vdev_spare_ops) | |
7234 | cvd->vdev_unspare = B_FALSE; | |
428870ff | 7235 | vdev_remove_parent(cvd); |
572e2857 BB |
7236 | } |
7237 | ||
428870ff BB |
7238 | /* |
7239 | * We don't set tvd until now because the parent we just removed | |
7240 | * may have been the previous top-level vdev. | |
7241 | */ | |
7242 | tvd = cvd->vdev_top; | |
7243 | ASSERT(tvd->vdev_parent == rvd); | |
7244 | ||
7245 | /* | |
7246 | * Reevaluate the parent vdev state. | |
7247 | */ | |
7248 | vdev_propagate_state(cvd); | |
7249 | ||
7250 | /* | |
7251 | * If the 'autoexpand' property is set on the pool then automatically | |
7252 | * try to expand the size of the pool. For example if the device we | |
7253 | * just detached was smaller than the others, it may be possible to | |
7254 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
7255 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
7256 | */ | |
7257 | if (spa->spa_autoexpand) { | |
7258 | vdev_reopen(tvd); | |
7259 | vdev_expand(tvd, txg); | |
7260 | } | |
7261 | ||
7262 | vdev_config_dirty(tvd); | |
7263 | ||
7264 | /* | |
7265 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
7266 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
7267 | * But first make sure we're not on any *other* txg's DTL list, to | |
7268 | * prevent vd from being accessed after it's freed. | |
7269 | */ | |
b6ca6193 | 7270 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); |
1c27024e | 7271 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff BB |
7272 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); |
7273 | vd->vdev_detached = B_TRUE; | |
7274 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
7275 | ||
12fa0466 | 7276 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); |
e60e158e | 7277 | spa_notify_waiters(spa); |
428870ff | 7278 | |
572e2857 BB |
7279 | /* hang on to the spa before we release the lock */ |
7280 | spa_open_ref(spa, FTAG); | |
7281 | ||
428870ff BB |
7282 | error = spa_vdev_exit(spa, vd, txg, 0); |
7283 | ||
6f1ffb06 | 7284 | spa_history_log_internal(spa, "detach", NULL, |
428870ff BB |
7285 | "vdev=%s", vdpath); |
7286 | spa_strfree(vdpath); | |
7287 | ||
7288 | /* | |
7289 | * If this was the removal of the original device in a hot spare vdev, | |
7290 | * then we want to go through and remove the device from the hot spare | |
7291 | * list of every other pool. | |
7292 | */ | |
7293 | if (unspare) { | |
572e2857 BB |
7294 | spa_t *altspa = NULL; |
7295 | ||
428870ff | 7296 | mutex_enter(&spa_namespace_lock); |
572e2857 BB |
7297 | while ((altspa = spa_next(altspa)) != NULL) { |
7298 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
7299 | altspa == spa) | |
428870ff | 7300 | continue; |
572e2857 BB |
7301 | |
7302 | spa_open_ref(altspa, FTAG); | |
428870ff | 7303 | mutex_exit(&spa_namespace_lock); |
572e2857 | 7304 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); |
428870ff | 7305 | mutex_enter(&spa_namespace_lock); |
572e2857 | 7306 | spa_close(altspa, FTAG); |
428870ff BB |
7307 | } |
7308 | mutex_exit(&spa_namespace_lock); | |
572e2857 BB |
7309 | |
7310 | /* search the rest of the vdevs for spares to remove */ | |
7311 | spa_vdev_resilver_done(spa); | |
428870ff BB |
7312 | } |
7313 | ||
572e2857 BB |
7314 | /* all done with the spa; OK to release */ |
7315 | mutex_enter(&spa_namespace_lock); | |
7316 | spa_close(spa, FTAG); | |
7317 | mutex_exit(&spa_namespace_lock); | |
7318 | ||
428870ff BB |
7319 | return (error); |
7320 | } | |
7321 | ||
c10d37dd GW |
7322 | static int |
7323 | spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7324 | list_t *vd_list) | |
619f0976 | 7325 | { |
c10d37dd GW |
7326 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7327 | ||
619f0976 GW |
7328 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); |
7329 | ||
7330 | /* Look up vdev and ensure it's a leaf. */ | |
7331 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7332 | if (vd == NULL || vd->vdev_detached) { | |
7333 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7334 | return (SET_ERROR(ENODEV)); |
7335 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
7336 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7337 | return (SET_ERROR(EINVAL)); |
7338 | } else if (!vdev_writeable(vd)) { | |
7339 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7340 | return (SET_ERROR(EROFS)); |
7341 | } | |
7342 | mutex_enter(&vd->vdev_initialize_lock); | |
7343 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7344 | ||
7345 | /* | |
7346 | * When we activate an initialize action we check to see | |
7347 | * if the vdev_initialize_thread is NULL. We do this instead | |
7348 | * of using the vdev_initialize_state since there might be | |
7349 | * a previous initialization process which has completed but | |
7350 | * the thread is not exited. | |
7351 | */ | |
1b939560 | 7352 | if (cmd_type == POOL_INITIALIZE_START && |
619f0976 GW |
7353 | (vd->vdev_initialize_thread != NULL || |
7354 | vd->vdev_top->vdev_removing)) { | |
7355 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 GW |
7356 | return (SET_ERROR(EBUSY)); |
7357 | } else if (cmd_type == POOL_INITIALIZE_CANCEL && | |
7358 | (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE && | |
7359 | vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) { | |
7360 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 GW |
7361 | return (SET_ERROR(ESRCH)); |
7362 | } else if (cmd_type == POOL_INITIALIZE_SUSPEND && | |
7363 | vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) { | |
7364 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 GW |
7365 | return (SET_ERROR(ESRCH)); |
7366 | } | |
7367 | ||
7368 | switch (cmd_type) { | |
1b939560 | 7369 | case POOL_INITIALIZE_START: |
619f0976 GW |
7370 | vdev_initialize(vd); |
7371 | break; | |
7372 | case POOL_INITIALIZE_CANCEL: | |
c10d37dd | 7373 | vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list); |
619f0976 GW |
7374 | break; |
7375 | case POOL_INITIALIZE_SUSPEND: | |
c10d37dd | 7376 | vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list); |
619f0976 GW |
7377 | break; |
7378 | default: | |
7379 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
7380 | } | |
7381 | mutex_exit(&vd->vdev_initialize_lock); | |
7382 | ||
c10d37dd GW |
7383 | return (0); |
7384 | } | |
7385 | ||
7386 | int | |
7387 | spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, | |
7388 | nvlist_t *vdev_errlist) | |
7389 | { | |
7390 | int total_errors = 0; | |
7391 | list_t vd_list; | |
7392 | ||
7393 | list_create(&vd_list, sizeof (vdev_t), | |
7394 | offsetof(vdev_t, vdev_initialize_node)); | |
7395 | ||
7396 | /* | |
7397 | * We hold the namespace lock through the whole function | |
7398 | * to prevent any changes to the pool while we're starting or | |
7399 | * stopping initialization. The config and state locks are held so that | |
7400 | * we can properly assess the vdev state before we commit to | |
7401 | * the initializing operation. | |
7402 | */ | |
7403 | mutex_enter(&spa_namespace_lock); | |
7404 | ||
7405 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
7406 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
7407 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
7408 | ||
7409 | int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type, | |
7410 | &vd_list); | |
7411 | if (error != 0) { | |
7412 | char guid_as_str[MAXNAMELEN]; | |
7413 | ||
7414 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
7415 | "%llu", (unsigned long long)vdev_guid); | |
7416 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
7417 | total_errors++; | |
7418 | } | |
7419 | } | |
7420 | ||
7421 | /* Wait for all initialize threads to stop. */ | |
7422 | vdev_initialize_stop_wait(spa, &vd_list); | |
7423 | ||
619f0976 GW |
7424 | /* Sync out the initializing state */ |
7425 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
7426 | mutex_exit(&spa_namespace_lock); | |
7427 | ||
c10d37dd | 7428 | list_destroy(&vd_list); |
619f0976 | 7429 | |
c10d37dd GW |
7430 | return (total_errors); |
7431 | } | |
619f0976 | 7432 | |
1b939560 BB |
7433 | static int |
7434 | spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7435 | uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list) | |
7436 | { | |
7437 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7438 | ||
7439 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
7440 | ||
7441 | /* Look up vdev and ensure it's a leaf. */ | |
7442 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7443 | if (vd == NULL || vd->vdev_detached) { | |
7444 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7445 | return (SET_ERROR(ENODEV)); | |
7446 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
7447 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7448 | return (SET_ERROR(EINVAL)); | |
7449 | } else if (!vdev_writeable(vd)) { | |
7450 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7451 | return (SET_ERROR(EROFS)); | |
7452 | } else if (!vd->vdev_has_trim) { | |
7453 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7454 | return (SET_ERROR(EOPNOTSUPP)); | |
7455 | } else if (secure && !vd->vdev_has_securetrim) { | |
7456 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7457 | return (SET_ERROR(EOPNOTSUPP)); | |
7458 | } | |
7459 | mutex_enter(&vd->vdev_trim_lock); | |
7460 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7461 | ||
7462 | /* | |
7463 | * When we activate a TRIM action we check to see if the | |
7464 | * vdev_trim_thread is NULL. We do this instead of using the | |
7465 | * vdev_trim_state since there might be a previous TRIM process | |
7466 | * which has completed but the thread is not exited. | |
7467 | */ | |
7468 | if (cmd_type == POOL_TRIM_START && | |
7469 | (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing)) { | |
7470 | mutex_exit(&vd->vdev_trim_lock); | |
7471 | return (SET_ERROR(EBUSY)); | |
7472 | } else if (cmd_type == POOL_TRIM_CANCEL && | |
7473 | (vd->vdev_trim_state != VDEV_TRIM_ACTIVE && | |
7474 | vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) { | |
7475 | mutex_exit(&vd->vdev_trim_lock); | |
7476 | return (SET_ERROR(ESRCH)); | |
7477 | } else if (cmd_type == POOL_TRIM_SUSPEND && | |
7478 | vd->vdev_trim_state != VDEV_TRIM_ACTIVE) { | |
7479 | mutex_exit(&vd->vdev_trim_lock); | |
7480 | return (SET_ERROR(ESRCH)); | |
7481 | } | |
7482 | ||
7483 | switch (cmd_type) { | |
7484 | case POOL_TRIM_START: | |
7485 | vdev_trim(vd, rate, partial, secure); | |
7486 | break; | |
7487 | case POOL_TRIM_CANCEL: | |
7488 | vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list); | |
7489 | break; | |
7490 | case POOL_TRIM_SUSPEND: | |
7491 | vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list); | |
7492 | break; | |
7493 | default: | |
7494 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
7495 | } | |
7496 | mutex_exit(&vd->vdev_trim_lock); | |
7497 | ||
7498 | return (0); | |
7499 | } | |
7500 | ||
7501 | /* | |
7502 | * Initiates a manual TRIM for the requested vdevs. This kicks off individual | |
7503 | * TRIM threads for each child vdev. These threads pass over all of the free | |
7504 | * space in the vdev's metaslabs and issues TRIM commands for that space. | |
7505 | */ | |
7506 | int | |
7507 | spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate, | |
7508 | boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist) | |
7509 | { | |
7510 | int total_errors = 0; | |
7511 | list_t vd_list; | |
7512 | ||
7513 | list_create(&vd_list, sizeof (vdev_t), | |
7514 | offsetof(vdev_t, vdev_trim_node)); | |
7515 | ||
7516 | /* | |
7517 | * We hold the namespace lock through the whole function | |
7518 | * to prevent any changes to the pool while we're starting or | |
7519 | * stopping TRIM. The config and state locks are held so that | |
7520 | * we can properly assess the vdev state before we commit to | |
7521 | * the TRIM operation. | |
7522 | */ | |
7523 | mutex_enter(&spa_namespace_lock); | |
7524 | ||
7525 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
7526 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
7527 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
7528 | ||
7529 | int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type, | |
7530 | rate, partial, secure, &vd_list); | |
7531 | if (error != 0) { | |
7532 | char guid_as_str[MAXNAMELEN]; | |
7533 | ||
7534 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
7535 | "%llu", (unsigned long long)vdev_guid); | |
7536 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
7537 | total_errors++; | |
7538 | } | |
7539 | } | |
7540 | ||
7541 | /* Wait for all TRIM threads to stop. */ | |
7542 | vdev_trim_stop_wait(spa, &vd_list); | |
7543 | ||
7544 | /* Sync out the TRIM state */ | |
7545 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
7546 | mutex_exit(&spa_namespace_lock); | |
7547 | ||
7548 | list_destroy(&vd_list); | |
7549 | ||
7550 | return (total_errors); | |
7551 | } | |
7552 | ||
428870ff BB |
7553 | /* |
7554 | * Split a set of devices from their mirrors, and create a new pool from them. | |
7555 | */ | |
7556 | int | |
a926aab9 | 7557 | spa_vdev_split_mirror(spa_t *spa, const char *newname, nvlist_t *config, |
428870ff BB |
7558 | nvlist_t *props, boolean_t exp) |
7559 | { | |
7560 | int error = 0; | |
7561 | uint64_t txg, *glist; | |
7562 | spa_t *newspa; | |
7563 | uint_t c, children, lastlog; | |
7564 | nvlist_t **child, *nvl, *tmp; | |
7565 | dmu_tx_t *tx; | |
7566 | char *altroot = NULL; | |
7567 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ | |
7568 | boolean_t activate_slog; | |
7569 | ||
572e2857 | 7570 | ASSERT(spa_writeable(spa)); |
428870ff BB |
7571 | |
7572 | txg = spa_vdev_enter(spa); | |
7573 | ||
d2734cce SD |
7574 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7575 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7576 | error = (spa_has_checkpoint(spa)) ? | |
7577 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7578 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7579 | } | |
7580 | ||
428870ff BB |
7581 | /* clear the log and flush everything up to now */ |
7582 | activate_slog = spa_passivate_log(spa); | |
7583 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
a1d477c2 | 7584 | error = spa_reset_logs(spa); |
428870ff BB |
7585 | txg = spa_vdev_config_enter(spa); |
7586 | ||
7587 | if (activate_slog) | |
7588 | spa_activate_log(spa); | |
7589 | ||
7590 | if (error != 0) | |
7591 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7592 | ||
7593 | /* check new spa name before going any further */ | |
7594 | if (spa_lookup(newname) != NULL) | |
7595 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
7596 | ||
7597 | /* | |
7598 | * scan through all the children to ensure they're all mirrors | |
7599 | */ | |
7600 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
7601 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
7602 | &children) != 0) | |
7603 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7604 | ||
7605 | /* first, check to ensure we've got the right child count */ | |
7606 | rvd = spa->spa_root_vdev; | |
7607 | lastlog = 0; | |
7608 | for (c = 0; c < rvd->vdev_children; c++) { | |
7609 | vdev_t *vd = rvd->vdev_child[c]; | |
7610 | ||
7611 | /* don't count the holes & logs as children */ | |
1b664952 GA |
7612 | if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops && |
7613 | !vdev_is_concrete(vd))) { | |
428870ff BB |
7614 | if (lastlog == 0) |
7615 | lastlog = c; | |
7616 | continue; | |
7617 | } | |
7618 | ||
7619 | lastlog = 0; | |
7620 | } | |
7621 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
7622 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7623 | ||
7624 | /* next, ensure no spare or cache devices are part of the split */ | |
7625 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
7626 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
7627 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7628 | ||
79c76d5b BB |
7629 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); |
7630 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
428870ff BB |
7631 | |
7632 | /* then, loop over each vdev and validate it */ | |
7633 | for (c = 0; c < children; c++) { | |
7634 | uint64_t is_hole = 0; | |
7635 | ||
7636 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
7637 | &is_hole); | |
7638 | ||
7639 | if (is_hole != 0) { | |
7640 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
7641 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
7642 | continue; | |
7643 | } else { | |
2e528b49 | 7644 | error = SET_ERROR(EINVAL); |
428870ff BB |
7645 | break; |
7646 | } | |
7647 | } | |
7648 | ||
1b664952 GA |
7649 | /* deal with indirect vdevs */ |
7650 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ops == | |
7651 | &vdev_indirect_ops) | |
7652 | continue; | |
7653 | ||
428870ff BB |
7654 | /* which disk is going to be split? */ |
7655 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
7656 | &glist[c]) != 0) { | |
2e528b49 | 7657 | error = SET_ERROR(EINVAL); |
428870ff BB |
7658 | break; |
7659 | } | |
7660 | ||
7661 | /* look it up in the spa */ | |
7662 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
7663 | if (vml[c] == NULL) { | |
2e528b49 | 7664 | error = SET_ERROR(ENODEV); |
428870ff BB |
7665 | break; |
7666 | } | |
7667 | ||
7668 | /* make sure there's nothing stopping the split */ | |
7669 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
7670 | vml[c]->vdev_islog || | |
a1d477c2 | 7671 | !vdev_is_concrete(vml[c]) || |
428870ff BB |
7672 | vml[c]->vdev_isspare || |
7673 | vml[c]->vdev_isl2cache || | |
7674 | !vdev_writeable(vml[c]) || | |
7675 | vml[c]->vdev_children != 0 || | |
7676 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
7677 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
2e528b49 | 7678 | error = SET_ERROR(EINVAL); |
428870ff BB |
7679 | break; |
7680 | } | |
7681 | ||
733b5722 RS |
7682 | if (vdev_dtl_required(vml[c]) || |
7683 | vdev_resilver_needed(vml[c], NULL, NULL)) { | |
2e528b49 | 7684 | error = SET_ERROR(EBUSY); |
428870ff BB |
7685 | break; |
7686 | } | |
7687 | ||
7688 | /* we need certain info from the top level */ | |
65ad5d11 AJ |
7689 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, |
7690 | vml[c]->vdev_top->vdev_ms_array); | |
7691 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
7692 | vml[c]->vdev_top->vdev_ms_shift); | |
7693 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
7694 | vml[c]->vdev_top->vdev_asize); | |
7695 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
7696 | vml[c]->vdev_top->vdev_ashift); | |
e0ab3ab5 JS |
7697 | |
7698 | /* transfer per-vdev ZAPs */ | |
7699 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
7700 | VERIFY0(nvlist_add_uint64(child[c], | |
7701 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
7702 | ||
7703 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
7704 | VERIFY0(nvlist_add_uint64(child[c], | |
7705 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
7706 | vml[c]->vdev_parent->vdev_top_zap)); | |
428870ff BB |
7707 | } |
7708 | ||
7709 | if (error != 0) { | |
7710 | kmem_free(vml, children * sizeof (vdev_t *)); | |
7711 | kmem_free(glist, children * sizeof (uint64_t)); | |
7712 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7713 | } | |
7714 | ||
7715 | /* stop writers from using the disks */ | |
7716 | for (c = 0; c < children; c++) { | |
7717 | if (vml[c] != NULL) | |
7718 | vml[c]->vdev_offline = B_TRUE; | |
7719 | } | |
7720 | vdev_reopen(spa->spa_root_vdev); | |
34dc7c2f BB |
7721 | |
7722 | /* | |
428870ff BB |
7723 | * Temporarily record the splitting vdevs in the spa config. This |
7724 | * will disappear once the config is regenerated. | |
34dc7c2f | 7725 | */ |
65ad5d11 AJ |
7726 | nvl = fnvlist_alloc(); |
7727 | fnvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children); | |
428870ff | 7728 | kmem_free(glist, children * sizeof (uint64_t)); |
34dc7c2f | 7729 | |
428870ff | 7730 | mutex_enter(&spa->spa_props_lock); |
65ad5d11 | 7731 | fnvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl); |
428870ff BB |
7732 | mutex_exit(&spa->spa_props_lock); |
7733 | spa->spa_config_splitting = nvl; | |
7734 | vdev_config_dirty(spa->spa_root_vdev); | |
7735 | ||
7736 | /* configure and create the new pool */ | |
65ad5d11 AJ |
7737 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname); |
7738 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
7739 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE); | |
7740 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); | |
7741 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg); | |
7742 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
7743 | spa_generate_guid(NULL)); | |
e0ab3ab5 | 7744 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); |
428870ff BB |
7745 | (void) nvlist_lookup_string(props, |
7746 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
34dc7c2f | 7747 | |
428870ff BB |
7748 | /* add the new pool to the namespace */ |
7749 | newspa = spa_add(newname, config, altroot); | |
e0ab3ab5 | 7750 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
7751 | newspa->spa_config_txg = spa->spa_config_txg; |
7752 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
7753 | ||
7754 | /* release the spa config lock, retaining the namespace lock */ | |
7755 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
7756 | ||
7757 | if (zio_injection_enabled) | |
7758 | zio_handle_panic_injection(spa, FTAG, 1); | |
7759 | ||
7760 | spa_activate(newspa, spa_mode_global); | |
7761 | spa_async_suspend(newspa); | |
7762 | ||
c10d37dd | 7763 | /* |
1b939560 BB |
7764 | * Temporarily stop the initializing and TRIM activity. We set the |
7765 | * state to ACTIVE so that we know to resume initializing or TRIM | |
7766 | * once the split has completed. | |
c10d37dd | 7767 | */ |
1b939560 BB |
7768 | list_t vd_initialize_list; |
7769 | list_create(&vd_initialize_list, sizeof (vdev_t), | |
c10d37dd GW |
7770 | offsetof(vdev_t, vdev_initialize_node)); |
7771 | ||
1b939560 BB |
7772 | list_t vd_trim_list; |
7773 | list_create(&vd_trim_list, sizeof (vdev_t), | |
7774 | offsetof(vdev_t, vdev_trim_node)); | |
7775 | ||
619f0976 | 7776 | for (c = 0; c < children; c++) { |
1b664952 | 7777 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
619f0976 | 7778 | mutex_enter(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
7779 | vdev_initialize_stop(vml[c], |
7780 | VDEV_INITIALIZE_ACTIVE, &vd_initialize_list); | |
619f0976 | 7781 | mutex_exit(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
7782 | |
7783 | mutex_enter(&vml[c]->vdev_trim_lock); | |
7784 | vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list); | |
7785 | mutex_exit(&vml[c]->vdev_trim_lock); | |
619f0976 GW |
7786 | } |
7787 | } | |
1b939560 BB |
7788 | |
7789 | vdev_initialize_stop_wait(spa, &vd_initialize_list); | |
7790 | vdev_trim_stop_wait(spa, &vd_trim_list); | |
7791 | ||
7792 | list_destroy(&vd_initialize_list); | |
7793 | list_destroy(&vd_trim_list); | |
619f0976 | 7794 | |
6cb8e530 | 7795 | newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; |
8b27e08e | 7796 | newspa->spa_is_splitting = B_TRUE; |
6cb8e530 | 7797 | |
428870ff | 7798 | /* create the new pool from the disks of the original pool */ |
6cb8e530 | 7799 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); |
428870ff BB |
7800 | if (error) |
7801 | goto out; | |
7802 | ||
7803 | /* if that worked, generate a real config for the new pool */ | |
7804 | if (newspa->spa_root_vdev != NULL) { | |
65ad5d11 AJ |
7805 | newspa->spa_config_splitting = fnvlist_alloc(); |
7806 | fnvlist_add_uint64(newspa->spa_config_splitting, | |
7807 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)); | |
428870ff BB |
7808 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, |
7809 | B_TRUE)); | |
9babb374 | 7810 | } |
34dc7c2f | 7811 | |
428870ff BB |
7812 | /* set the props */ |
7813 | if (props != NULL) { | |
7814 | spa_configfile_set(newspa, props, B_FALSE); | |
7815 | error = spa_prop_set(newspa, props); | |
7816 | if (error) | |
7817 | goto out; | |
7818 | } | |
34dc7c2f | 7819 | |
428870ff BB |
7820 | /* flush everything */ |
7821 | txg = spa_vdev_config_enter(newspa); | |
7822 | vdev_config_dirty(newspa->spa_root_vdev); | |
7823 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
34dc7c2f | 7824 | |
428870ff BB |
7825 | if (zio_injection_enabled) |
7826 | zio_handle_panic_injection(spa, FTAG, 2); | |
34dc7c2f | 7827 | |
428870ff | 7828 | spa_async_resume(newspa); |
34dc7c2f | 7829 | |
428870ff BB |
7830 | /* finally, update the original pool's config */ |
7831 | txg = spa_vdev_config_enter(spa); | |
7832 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
7833 | error = dmu_tx_assign(tx, TXG_WAIT); | |
7834 | if (error != 0) | |
7835 | dmu_tx_abort(tx); | |
7836 | for (c = 0; c < children; c++) { | |
1b664952 | 7837 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
234234ca RS |
7838 | vdev_t *tvd = vml[c]->vdev_top; |
7839 | ||
7840 | /* | |
7841 | * Need to be sure the detachable VDEV is not | |
7842 | * on any *other* txg's DTL list to prevent it | |
7843 | * from being accessed after it's freed. | |
7844 | */ | |
7845 | for (int t = 0; t < TXG_SIZE; t++) { | |
7846 | (void) txg_list_remove_this( | |
7847 | &tvd->vdev_dtl_list, vml[c], t); | |
7848 | } | |
7849 | ||
428870ff BB |
7850 | vdev_split(vml[c]); |
7851 | if (error == 0) | |
6f1ffb06 MA |
7852 | spa_history_log_internal(spa, "detach", tx, |
7853 | "vdev=%s", vml[c]->vdev_path); | |
e0ab3ab5 | 7854 | |
428870ff | 7855 | vdev_free(vml[c]); |
34dc7c2f | 7856 | } |
34dc7c2f | 7857 | } |
e0ab3ab5 | 7858 | spa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
7859 | vdev_config_dirty(spa->spa_root_vdev); |
7860 | spa->spa_config_splitting = NULL; | |
7861 | nvlist_free(nvl); | |
7862 | if (error == 0) | |
7863 | dmu_tx_commit(tx); | |
7864 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
7865 | ||
7866 | if (zio_injection_enabled) | |
7867 | zio_handle_panic_injection(spa, FTAG, 3); | |
7868 | ||
7869 | /* split is complete; log a history record */ | |
6f1ffb06 MA |
7870 | spa_history_log_internal(newspa, "split", NULL, |
7871 | "from pool %s", spa_name(spa)); | |
428870ff | 7872 | |
8b27e08e | 7873 | newspa->spa_is_splitting = B_FALSE; |
428870ff BB |
7874 | kmem_free(vml, children * sizeof (vdev_t *)); |
7875 | ||
7876 | /* if we're not going to mount the filesystems in userland, export */ | |
7877 | if (exp) | |
7878 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
7879 | B_FALSE, B_FALSE); | |
7880 | ||
7881 | return (error); | |
7882 | ||
7883 | out: | |
7884 | spa_unload(newspa); | |
7885 | spa_deactivate(newspa); | |
7886 | spa_remove(newspa); | |
7887 | ||
7888 | txg = spa_vdev_config_enter(spa); | |
7889 | ||
7890 | /* re-online all offlined disks */ | |
7891 | for (c = 0; c < children; c++) { | |
7892 | if (vml[c] != NULL) | |
7893 | vml[c]->vdev_offline = B_FALSE; | |
7894 | } | |
619f0976 | 7895 | |
1b939560 | 7896 | /* restart initializing or trimming disks as necessary */ |
619f0976 | 7897 | spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART); |
1b939560 BB |
7898 | spa_async_request(spa, SPA_ASYNC_TRIM_RESTART); |
7899 | spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART); | |
619f0976 | 7900 | |
428870ff BB |
7901 | vdev_reopen(spa->spa_root_vdev); |
7902 | ||
7903 | nvlist_free(spa->spa_config_splitting); | |
7904 | spa->spa_config_splitting = NULL; | |
7905 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
34dc7c2f | 7906 | |
428870ff | 7907 | kmem_free(vml, children * sizeof (vdev_t *)); |
34dc7c2f BB |
7908 | return (error); |
7909 | } | |
7910 | ||
34dc7c2f BB |
7911 | /* |
7912 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
d3cc8b15 | 7913 | * currently spared, so we can detach it. |
34dc7c2f BB |
7914 | */ |
7915 | static vdev_t * | |
7916 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
7917 | { | |
7918 | vdev_t *newvd, *oldvd; | |
34dc7c2f | 7919 | |
1c27024e | 7920 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
7921 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); |
7922 | if (oldvd != NULL) | |
7923 | return (oldvd); | |
7924 | } | |
7925 | ||
7926 | /* | |
572e2857 BB |
7927 | * Check for a completed replacement. We always consider the first |
7928 | * vdev in the list to be the oldest vdev, and the last one to be | |
7929 | * the newest (see spa_vdev_attach() for how that works). In | |
7930 | * the case where the newest vdev is faulted, we will not automatically | |
7931 | * remove it after a resilver completes. This is OK as it will require | |
7932 | * user intervention to determine which disk the admin wishes to keep. | |
34dc7c2f | 7933 | */ |
572e2857 BB |
7934 | if (vd->vdev_ops == &vdev_replacing_ops) { |
7935 | ASSERT(vd->vdev_children > 1); | |
7936 | ||
7937 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
34dc7c2f | 7938 | oldvd = vd->vdev_child[0]; |
34dc7c2f | 7939 | |
fb5f0bc8 | 7940 | if (vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 7941 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
fb5f0bc8 | 7942 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 7943 | return (oldvd); |
34dc7c2f BB |
7944 | } |
7945 | ||
7946 | /* | |
7947 | * Check for a completed resilver with the 'unspare' flag set. | |
f65fbee1 | 7948 | * Also potentially update faulted state. |
34dc7c2f | 7949 | */ |
572e2857 BB |
7950 | if (vd->vdev_ops == &vdev_spare_ops) { |
7951 | vdev_t *first = vd->vdev_child[0]; | |
7952 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
7953 | ||
7954 | if (last->vdev_unspare) { | |
7955 | oldvd = first; | |
7956 | newvd = last; | |
7957 | } else if (first->vdev_unspare) { | |
7958 | oldvd = last; | |
7959 | newvd = first; | |
7960 | } else { | |
7961 | oldvd = NULL; | |
7962 | } | |
34dc7c2f | 7963 | |
572e2857 | 7964 | if (oldvd != NULL && |
fb5f0bc8 | 7965 | vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 7966 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
572e2857 | 7967 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 7968 | return (oldvd); |
572e2857 | 7969 | |
f65fbee1 JJ |
7970 | vdev_propagate_state(vd); |
7971 | ||
572e2857 BB |
7972 | /* |
7973 | * If there are more than two spares attached to a disk, | |
7974 | * and those spares are not required, then we want to | |
7975 | * attempt to free them up now so that they can be used | |
7976 | * by other pools. Once we're back down to a single | |
7977 | * disk+spare, we stop removing them. | |
7978 | */ | |
7979 | if (vd->vdev_children > 2) { | |
7980 | newvd = vd->vdev_child[1]; | |
7981 | ||
7982 | if (newvd->vdev_isspare && last->vdev_isspare && | |
7983 | vdev_dtl_empty(last, DTL_MISSING) && | |
7984 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
7985 | !vdev_dtl_required(newvd)) | |
7986 | return (newvd); | |
34dc7c2f | 7987 | } |
34dc7c2f BB |
7988 | } |
7989 | ||
7990 | return (NULL); | |
7991 | } | |
7992 | ||
7993 | static void | |
7994 | spa_vdev_resilver_done(spa_t *spa) | |
7995 | { | |
fb5f0bc8 BB |
7996 | vdev_t *vd, *pvd, *ppvd; |
7997 | uint64_t guid, sguid, pguid, ppguid; | |
34dc7c2f | 7998 | |
fb5f0bc8 | 7999 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8000 | |
8001 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
fb5f0bc8 BB |
8002 | pvd = vd->vdev_parent; |
8003 | ppvd = pvd->vdev_parent; | |
34dc7c2f | 8004 | guid = vd->vdev_guid; |
fb5f0bc8 BB |
8005 | pguid = pvd->vdev_guid; |
8006 | ppguid = ppvd->vdev_guid; | |
8007 | sguid = 0; | |
34dc7c2f BB |
8008 | /* |
8009 | * If we have just finished replacing a hot spared device, then | |
8010 | * we need to detach the parent's first child (the original hot | |
8011 | * spare) as well. | |
8012 | */ | |
572e2857 BB |
8013 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && |
8014 | ppvd->vdev_children == 2) { | |
34dc7c2f | 8015 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); |
fb5f0bc8 | 8016 | sguid = ppvd->vdev_child[1]->vdev_guid; |
34dc7c2f | 8017 | } |
5d1f7fb6 GW |
8018 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); |
8019 | ||
fb5f0bc8 BB |
8020 | spa_config_exit(spa, SCL_ALL, FTAG); |
8021 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
34dc7c2f | 8022 | return; |
fb5f0bc8 | 8023 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) |
34dc7c2f | 8024 | return; |
fb5f0bc8 | 8025 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8026 | } |
8027 | ||
fb5f0bc8 | 8028 | spa_config_exit(spa, SCL_ALL, FTAG); |
9a49d3f3 BB |
8029 | |
8030 | /* | |
8031 | * If a detach was not performed above replace waiters will not have | |
8032 | * been notified. In which case we must do so now. | |
8033 | */ | |
8034 | spa_notify_waiters(spa); | |
34dc7c2f BB |
8035 | } |
8036 | ||
8037 | /* | |
428870ff | 8038 | * Update the stored path or FRU for this vdev. |
34dc7c2f | 8039 | */ |
65c7cc49 | 8040 | static int |
9babb374 BB |
8041 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, |
8042 | boolean_t ispath) | |
34dc7c2f | 8043 | { |
b128c09f | 8044 | vdev_t *vd; |
428870ff | 8045 | boolean_t sync = B_FALSE; |
34dc7c2f | 8046 | |
572e2857 BB |
8047 | ASSERT(spa_writeable(spa)); |
8048 | ||
428870ff | 8049 | spa_vdev_state_enter(spa, SCL_ALL); |
34dc7c2f | 8050 | |
9babb374 | 8051 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
428870ff | 8052 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); |
34dc7c2f BB |
8053 | |
8054 | if (!vd->vdev_ops->vdev_op_leaf) | |
428870ff | 8055 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 8056 | |
9babb374 | 8057 | if (ispath) { |
428870ff BB |
8058 | if (strcmp(value, vd->vdev_path) != 0) { |
8059 | spa_strfree(vd->vdev_path); | |
8060 | vd->vdev_path = spa_strdup(value); | |
8061 | sync = B_TRUE; | |
8062 | } | |
9babb374 | 8063 | } else { |
428870ff BB |
8064 | if (vd->vdev_fru == NULL) { |
8065 | vd->vdev_fru = spa_strdup(value); | |
8066 | sync = B_TRUE; | |
8067 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
9babb374 | 8068 | spa_strfree(vd->vdev_fru); |
428870ff BB |
8069 | vd->vdev_fru = spa_strdup(value); |
8070 | sync = B_TRUE; | |
8071 | } | |
9babb374 | 8072 | } |
34dc7c2f | 8073 | |
428870ff | 8074 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); |
34dc7c2f BB |
8075 | } |
8076 | ||
9babb374 BB |
8077 | int |
8078 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
8079 | { | |
8080 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
8081 | } | |
8082 | ||
8083 | int | |
8084 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
8085 | { | |
8086 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
8087 | } | |
8088 | ||
34dc7c2f BB |
8089 | /* |
8090 | * ========================================================================== | |
428870ff | 8091 | * SPA Scanning |
34dc7c2f BB |
8092 | * ========================================================================== |
8093 | */ | |
0ea05c64 AP |
8094 | int |
8095 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
8096 | { | |
8097 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8098 | ||
8099 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
8100 | return (SET_ERROR(EBUSY)); | |
8101 | ||
8102 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
8103 | } | |
34dc7c2f | 8104 | |
34dc7c2f | 8105 | int |
428870ff BB |
8106 | spa_scan_stop(spa_t *spa) |
8107 | { | |
8108 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8109 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
2e528b49 | 8110 | return (SET_ERROR(EBUSY)); |
428870ff BB |
8111 | return (dsl_scan_cancel(spa->spa_dsl_pool)); |
8112 | } | |
8113 | ||
8114 | int | |
8115 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
34dc7c2f | 8116 | { |
b128c09f | 8117 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); |
34dc7c2f | 8118 | |
428870ff | 8119 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) |
2e528b49 | 8120 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 8121 | |
fa241660 TC |
8122 | if (func == POOL_SCAN_RESILVER && |
8123 | !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) | |
8124 | return (SET_ERROR(ENOTSUP)); | |
8125 | ||
34dc7c2f | 8126 | /* |
b128c09f BB |
8127 | * If a resilver was requested, but there is no DTL on a |
8128 | * writeable leaf device, we have nothing to do. | |
34dc7c2f | 8129 | */ |
428870ff | 8130 | if (func == POOL_SCAN_RESILVER && |
b128c09f BB |
8131 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { |
8132 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
34dc7c2f BB |
8133 | return (0); |
8134 | } | |
8135 | ||
428870ff | 8136 | return (dsl_scan(spa->spa_dsl_pool, func)); |
34dc7c2f BB |
8137 | } |
8138 | ||
8139 | /* | |
8140 | * ========================================================================== | |
8141 | * SPA async task processing | |
8142 | * ========================================================================== | |
8143 | */ | |
8144 | ||
8145 | static void | |
8146 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
8147 | { | |
b128c09f | 8148 | if (vd->vdev_remove_wanted) { |
428870ff BB |
8149 | vd->vdev_remove_wanted = B_FALSE; |
8150 | vd->vdev_delayed_close = B_FALSE; | |
b128c09f | 8151 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); |
428870ff BB |
8152 | |
8153 | /* | |
8154 | * We want to clear the stats, but we don't want to do a full | |
8155 | * vdev_clear() as that will cause us to throw away | |
8156 | * degraded/faulted state as well as attempt to reopen the | |
8157 | * device, all of which is a waste. | |
8158 | */ | |
8159 | vd->vdev_stat.vs_read_errors = 0; | |
8160 | vd->vdev_stat.vs_write_errors = 0; | |
8161 | vd->vdev_stat.vs_checksum_errors = 0; | |
8162 | ||
b128c09f | 8163 | vdev_state_dirty(vd->vdev_top); |
0aacde2e RM |
8164 | |
8165 | /* Tell userspace that the vdev is gone. */ | |
8166 | zfs_post_remove(spa, vd); | |
b128c09f | 8167 | } |
34dc7c2f | 8168 | |
1c27024e | 8169 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f BB |
8170 | spa_async_remove(spa, vd->vdev_child[c]); |
8171 | } | |
8172 | ||
8173 | static void | |
8174 | spa_async_probe(spa_t *spa, vdev_t *vd) | |
8175 | { | |
8176 | if (vd->vdev_probe_wanted) { | |
428870ff | 8177 | vd->vdev_probe_wanted = B_FALSE; |
b128c09f | 8178 | vdev_reopen(vd); /* vdev_open() does the actual probe */ |
34dc7c2f | 8179 | } |
b128c09f | 8180 | |
1c27024e | 8181 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f | 8182 | spa_async_probe(spa, vd->vdev_child[c]); |
34dc7c2f BB |
8183 | } |
8184 | ||
9babb374 BB |
8185 | static void |
8186 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
8187 | { | |
9babb374 BB |
8188 | if (!spa->spa_autoexpand) |
8189 | return; | |
8190 | ||
1c27024e | 8191 | for (int c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
8192 | vdev_t *cvd = vd->vdev_child[c]; |
8193 | spa_async_autoexpand(spa, cvd); | |
8194 | } | |
8195 | ||
8196 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
8197 | return; | |
8198 | ||
12fa0466 | 8199 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); |
9babb374 BB |
8200 | } |
8201 | ||
460748d4 | 8202 | static __attribute__((noreturn)) void |
c25b8f99 | 8203 | spa_async_thread(void *arg) |
34dc7c2f | 8204 | { |
c25b8f99 | 8205 | spa_t *spa = (spa_t *)arg; |
80a91e74 | 8206 | dsl_pool_t *dp = spa->spa_dsl_pool; |
867959b5 | 8207 | int tasks; |
34dc7c2f BB |
8208 | |
8209 | ASSERT(spa->spa_sync_on); | |
8210 | ||
8211 | mutex_enter(&spa->spa_async_lock); | |
8212 | tasks = spa->spa_async_tasks; | |
8213 | spa->spa_async_tasks = 0; | |
8214 | mutex_exit(&spa->spa_async_lock); | |
8215 | ||
8216 | /* | |
8217 | * See if the config needs to be updated. | |
8218 | */ | |
8219 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
428870ff | 8220 | uint64_t old_space, new_space; |
9babb374 | 8221 | |
34dc7c2f | 8222 | mutex_enter(&spa_namespace_lock); |
428870ff | 8223 | old_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8224 | old_space += metaslab_class_get_space(spa_special_class(spa)); |
8225 | old_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8226 | old_space += metaslab_class_get_space( |
8227 | spa_embedded_log_class(spa)); | |
cc99f275 | 8228 | |
34dc7c2f | 8229 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
cc99f275 | 8230 | |
428870ff | 8231 | new_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8232 | new_space += metaslab_class_get_space(spa_special_class(spa)); |
8233 | new_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8234 | new_space += metaslab_class_get_space( |
8235 | spa_embedded_log_class(spa)); | |
34dc7c2f | 8236 | mutex_exit(&spa_namespace_lock); |
9babb374 BB |
8237 | |
8238 | /* | |
8239 | * If the pool grew as a result of the config update, | |
8240 | * then log an internal history event. | |
8241 | */ | |
428870ff | 8242 | if (new_space != old_space) { |
6f1ffb06 | 8243 | spa_history_log_internal(spa, "vdev online", NULL, |
45d1cae3 | 8244 | "pool '%s' size: %llu(+%llu)", |
74756182 MM |
8245 | spa_name(spa), (u_longlong_t)new_space, |
8246 | (u_longlong_t)(new_space - old_space)); | |
9babb374 | 8247 | } |
34dc7c2f BB |
8248 | } |
8249 | ||
8250 | /* | |
8251 | * See if any devices need to be marked REMOVED. | |
34dc7c2f | 8252 | */ |
b128c09f | 8253 | if (tasks & SPA_ASYNC_REMOVE) { |
428870ff | 8254 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 8255 | spa_async_remove(spa, spa->spa_root_vdev); |
867959b5 | 8256 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) |
b128c09f | 8257 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); |
867959b5 | 8258 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
b128c09f BB |
8259 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); |
8260 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
34dc7c2f BB |
8261 | } |
8262 | ||
9babb374 BB |
8263 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { |
8264 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8265 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
8266 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8267 | } | |
8268 | ||
34dc7c2f | 8269 | /* |
b128c09f | 8270 | * See if any devices need to be probed. |
34dc7c2f | 8271 | */ |
b128c09f | 8272 | if (tasks & SPA_ASYNC_PROBE) { |
428870ff | 8273 | spa_vdev_state_enter(spa, SCL_NONE); |
b128c09f BB |
8274 | spa_async_probe(spa, spa->spa_root_vdev); |
8275 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
8276 | } | |
34dc7c2f BB |
8277 | |
8278 | /* | |
b128c09f | 8279 | * If any devices are done replacing, detach them. |
34dc7c2f | 8280 | */ |
b2255edc BB |
8281 | if (tasks & SPA_ASYNC_RESILVER_DONE || |
8282 | tasks & SPA_ASYNC_REBUILD_DONE) { | |
b128c09f | 8283 | spa_vdev_resilver_done(spa); |
9a49d3f3 BB |
8284 | } |
8285 | ||
34dc7c2f BB |
8286 | /* |
8287 | * Kick off a resilver. | |
8288 | */ | |
80a91e74 | 8289 | if (tasks & SPA_ASYNC_RESILVER && |
9a49d3f3 | 8290 | !vdev_rebuild_active(spa->spa_root_vdev) && |
80a91e74 TC |
8291 | (!dsl_scan_resilvering(dp) || |
8292 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER))) | |
3c819a2c | 8293 | dsl_scan_restart_resilver(dp, 0); |
34dc7c2f | 8294 | |
619f0976 GW |
8295 | if (tasks & SPA_ASYNC_INITIALIZE_RESTART) { |
8296 | mutex_enter(&spa_namespace_lock); | |
8297 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8298 | vdev_initialize_restart(spa->spa_root_vdev); | |
8299 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8300 | mutex_exit(&spa_namespace_lock); | |
8301 | } | |
8302 | ||
1b939560 BB |
8303 | if (tasks & SPA_ASYNC_TRIM_RESTART) { |
8304 | mutex_enter(&spa_namespace_lock); | |
8305 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8306 | vdev_trim_restart(spa->spa_root_vdev); | |
8307 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8308 | mutex_exit(&spa_namespace_lock); | |
8309 | } | |
8310 | ||
8311 | if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) { | |
8312 | mutex_enter(&spa_namespace_lock); | |
8313 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8314 | vdev_autotrim_restart(spa); | |
8315 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8316 | mutex_exit(&spa_namespace_lock); | |
8317 | } | |
8318 | ||
b7654bd7 GA |
8319 | /* |
8320 | * Kick off L2 cache whole device TRIM. | |
8321 | */ | |
8322 | if (tasks & SPA_ASYNC_L2CACHE_TRIM) { | |
8323 | mutex_enter(&spa_namespace_lock); | |
8324 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8325 | vdev_trim_l2arc(spa); | |
8326 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8327 | mutex_exit(&spa_namespace_lock); | |
8328 | } | |
8329 | ||
77f6826b GA |
8330 | /* |
8331 | * Kick off L2 cache rebuilding. | |
8332 | */ | |
8333 | if (tasks & SPA_ASYNC_L2CACHE_REBUILD) { | |
8334 | mutex_enter(&spa_namespace_lock); | |
8335 | spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER); | |
8336 | l2arc_spa_rebuild_start(spa); | |
8337 | spa_config_exit(spa, SCL_L2ARC, FTAG); | |
8338 | mutex_exit(&spa_namespace_lock); | |
8339 | } | |
8340 | ||
34dc7c2f BB |
8341 | /* |
8342 | * Let the world know that we're done. | |
8343 | */ | |
8344 | mutex_enter(&spa->spa_async_lock); | |
8345 | spa->spa_async_thread = NULL; | |
8346 | cv_broadcast(&spa->spa_async_cv); | |
8347 | mutex_exit(&spa->spa_async_lock); | |
8348 | thread_exit(); | |
8349 | } | |
8350 | ||
8351 | void | |
8352 | spa_async_suspend(spa_t *spa) | |
8353 | { | |
8354 | mutex_enter(&spa->spa_async_lock); | |
8355 | spa->spa_async_suspended++; | |
9d5b5245 | 8356 | while (spa->spa_async_thread != NULL) |
34dc7c2f BB |
8357 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); |
8358 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 MA |
8359 | |
8360 | spa_vdev_remove_suspend(spa); | |
9d5b5245 SD |
8361 | |
8362 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a SD |
8363 | if (condense_thread != NULL) |
8364 | zthr_cancel(condense_thread); | |
d2734cce SD |
8365 | |
8366 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; | |
61c3391a SD |
8367 | if (discard_thread != NULL) |
8368 | zthr_cancel(discard_thread); | |
37f03da8 SH |
8369 | |
8370 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
8371 | if (ll_delete_thread != NULL) | |
8372 | zthr_cancel(ll_delete_thread); | |
8373 | ||
8374 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
8375 | if (ll_condense_thread != NULL) | |
8376 | zthr_cancel(ll_condense_thread); | |
34dc7c2f BB |
8377 | } |
8378 | ||
8379 | void | |
8380 | spa_async_resume(spa_t *spa) | |
8381 | { | |
8382 | mutex_enter(&spa->spa_async_lock); | |
8383 | ASSERT(spa->spa_async_suspended != 0); | |
8384 | spa->spa_async_suspended--; | |
8385 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 | 8386 | spa_restart_removal(spa); |
9d5b5245 SD |
8387 | |
8388 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a | 8389 | if (condense_thread != NULL) |
9d5b5245 | 8390 | zthr_resume(condense_thread); |
d2734cce SD |
8391 | |
8392 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; | |
61c3391a | 8393 | if (discard_thread != NULL) |
d2734cce | 8394 | zthr_resume(discard_thread); |
37f03da8 SH |
8395 | |
8396 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
8397 | if (ll_delete_thread != NULL) | |
8398 | zthr_resume(ll_delete_thread); | |
8399 | ||
8400 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
8401 | if (ll_condense_thread != NULL) | |
8402 | zthr_resume(ll_condense_thread); | |
34dc7c2f BB |
8403 | } |
8404 | ||
e6cfd633 WA |
8405 | static boolean_t |
8406 | spa_async_tasks_pending(spa_t *spa) | |
8407 | { | |
8408 | uint_t non_config_tasks; | |
8409 | uint_t config_task; | |
8410 | boolean_t config_task_suspended; | |
8411 | ||
8412 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
8413 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
8414 | if (spa->spa_ccw_fail_time == 0) { | |
8415 | config_task_suspended = B_FALSE; | |
8416 | } else { | |
8417 | config_task_suspended = | |
8418 | (gethrtime() - spa->spa_ccw_fail_time) < | |
05852b34 | 8419 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); |
e6cfd633 WA |
8420 | } |
8421 | ||
8422 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
8423 | } | |
8424 | ||
34dc7c2f BB |
8425 | static void |
8426 | spa_async_dispatch(spa_t *spa) | |
8427 | { | |
8428 | mutex_enter(&spa->spa_async_lock); | |
e6cfd633 WA |
8429 | if (spa_async_tasks_pending(spa) && |
8430 | !spa->spa_async_suspended && | |
da92d5cb | 8431 | spa->spa_async_thread == NULL) |
34dc7c2f BB |
8432 | spa->spa_async_thread = thread_create(NULL, 0, |
8433 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
8434 | mutex_exit(&spa->spa_async_lock); | |
8435 | } | |
8436 | ||
8437 | void | |
8438 | spa_async_request(spa_t *spa, int task) | |
8439 | { | |
428870ff | 8440 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); |
34dc7c2f BB |
8441 | mutex_enter(&spa->spa_async_lock); |
8442 | spa->spa_async_tasks |= task; | |
8443 | mutex_exit(&spa->spa_async_lock); | |
8444 | } | |
8445 | ||
3c819a2c JP |
8446 | int |
8447 | spa_async_tasks(spa_t *spa) | |
8448 | { | |
8449 | return (spa->spa_async_tasks); | |
8450 | } | |
8451 | ||
34dc7c2f BB |
8452 | /* |
8453 | * ========================================================================== | |
8454 | * SPA syncing routines | |
8455 | * ========================================================================== | |
8456 | */ | |
8457 | ||
37f03da8 | 8458 | |
428870ff | 8459 | static int |
37f03da8 SH |
8460 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, |
8461 | dmu_tx_t *tx) | |
34dc7c2f | 8462 | { |
428870ff | 8463 | bpobj_t *bpo = arg; |
37f03da8 | 8464 | bpobj_enqueue(bpo, bp, bp_freed, tx); |
428870ff BB |
8465 | return (0); |
8466 | } | |
34dc7c2f | 8467 | |
37f03da8 SH |
8468 | int |
8469 | bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8470 | { | |
8471 | return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx)); | |
8472 | } | |
8473 | ||
8474 | int | |
8475 | bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8476 | { | |
8477 | return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx)); | |
8478 | } | |
8479 | ||
428870ff BB |
8480 | static int |
8481 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8482 | { | |
9cdf7b1f | 8483 | zio_t *pio = arg; |
34dc7c2f | 8484 | |
9cdf7b1f MA |
8485 | zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp, |
8486 | pio->io_flags)); | |
428870ff | 8487 | return (0); |
34dc7c2f BB |
8488 | } |
8489 | ||
37f03da8 SH |
8490 | static int |
8491 | bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
8492 | dmu_tx_t *tx) | |
8493 | { | |
8494 | ASSERT(!bp_freed); | |
8495 | return (spa_free_sync_cb(arg, bp, tx)); | |
8496 | } | |
8497 | ||
e8b96c60 MA |
8498 | /* |
8499 | * Note: this simple function is not inlined to make it easier to dtrace the | |
8500 | * amount of time spent syncing frees. | |
8501 | */ | |
8502 | static void | |
8503 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
8504 | { | |
8505 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
8506 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
8507 | VERIFY(zio_wait(zio) == 0); | |
8508 | } | |
8509 | ||
8510 | /* | |
8511 | * Note: this simple function is not inlined to make it easier to dtrace the | |
8512 | * amount of time spent syncing deferred frees. | |
8513 | */ | |
8514 | static void | |
8515 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
8516 | { | |
8dc2197b SD |
8517 | if (spa_sync_pass(spa) != 1) |
8518 | return; | |
8519 | ||
93e28d66 SD |
8520 | /* |
8521 | * Note: | |
8522 | * If the log space map feature is active, we stop deferring | |
8523 | * frees to the next TXG and therefore running this function | |
8524 | * would be considered a no-op as spa_deferred_bpobj should | |
8525 | * not have any entries. | |
8526 | * | |
8527 | * That said we run this function anyway (instead of returning | |
8528 | * immediately) for the edge-case scenario where we just | |
8529 | * activated the log space map feature in this TXG but we have | |
8530 | * deferred frees from the previous TXG. | |
8531 | */ | |
e8b96c60 MA |
8532 | zio_t *zio = zio_root(spa, NULL, NULL, 0); |
8533 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
37f03da8 | 8534 | bpobj_spa_free_sync_cb, zio, tx), ==, 0); |
e8b96c60 MA |
8535 | VERIFY0(zio_wait(zio)); |
8536 | } | |
8537 | ||
34dc7c2f BB |
8538 | static void |
8539 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
8540 | { | |
8541 | char *packed = NULL; | |
b128c09f | 8542 | size_t bufsize; |
34dc7c2f BB |
8543 | size_t nvsize = 0; |
8544 | dmu_buf_t *db; | |
8545 | ||
8546 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
8547 | ||
b128c09f BB |
8548 | /* |
8549 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
b0bc7a84 | 8550 | * information. This avoids the dmu_buf_will_dirty() path and |
b128c09f BB |
8551 | * saves us a pre-read to get data we don't actually care about. |
8552 | */ | |
9ae529ec | 8553 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); |
79c76d5b | 8554 | packed = vmem_alloc(bufsize, KM_SLEEP); |
34dc7c2f BB |
8555 | |
8556 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
79c76d5b | 8557 | KM_SLEEP) == 0); |
861166b0 | 8558 | memset(packed + nvsize, 0, bufsize - nvsize); |
34dc7c2f | 8559 | |
b128c09f | 8560 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); |
34dc7c2f | 8561 | |
00b46022 | 8562 | vmem_free(packed, bufsize); |
34dc7c2f BB |
8563 | |
8564 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
8565 | dmu_buf_will_dirty(db, tx); | |
8566 | *(uint64_t *)db->db_data = nvsize; | |
8567 | dmu_buf_rele(db, FTAG); | |
8568 | } | |
8569 | ||
8570 | static void | |
8571 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
8572 | const char *config, const char *entry) | |
8573 | { | |
8574 | nvlist_t *nvroot; | |
8575 | nvlist_t **list; | |
8576 | int i; | |
8577 | ||
8578 | if (!sav->sav_sync) | |
8579 | return; | |
8580 | ||
8581 | /* | |
8582 | * Update the MOS nvlist describing the list of available devices. | |
8583 | * spa_validate_aux() will have already made sure this nvlist is | |
8584 | * valid and the vdevs are labeled appropriately. | |
8585 | */ | |
8586 | if (sav->sav_object == 0) { | |
8587 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
8588 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
8589 | sizeof (uint64_t), tx); | |
8590 | VERIFY(zap_update(spa->spa_meta_objset, | |
8591 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
8592 | &sav->sav_object, tx) == 0); | |
8593 | } | |
8594 | ||
65ad5d11 | 8595 | nvroot = fnvlist_alloc(); |
34dc7c2f | 8596 | if (sav->sav_count == 0) { |
795075e6 PD |
8597 | fnvlist_add_nvlist_array(nvroot, config, |
8598 | (const nvlist_t * const *)NULL, 0); | |
34dc7c2f | 8599 | } else { |
79c76d5b | 8600 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); |
34dc7c2f BB |
8601 | for (i = 0; i < sav->sav_count; i++) |
8602 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
428870ff | 8603 | B_FALSE, VDEV_CONFIG_L2CACHE); |
795075e6 PD |
8604 | fnvlist_add_nvlist_array(nvroot, config, |
8605 | (const nvlist_t * const *)list, sav->sav_count); | |
34dc7c2f BB |
8606 | for (i = 0; i < sav->sav_count; i++) |
8607 | nvlist_free(list[i]); | |
8608 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
8609 | } | |
8610 | ||
8611 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
8612 | nvlist_free(nvroot); | |
8613 | ||
8614 | sav->sav_sync = B_FALSE; | |
8615 | } | |
8616 | ||
e0ab3ab5 JS |
8617 | /* |
8618 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
8619 | * The all-vdev ZAP must be empty. | |
8620 | */ | |
8621 | static void | |
8622 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
8623 | { | |
8624 | spa_t *spa = vd->vdev_spa; | |
e0ab3ab5 JS |
8625 | |
8626 | if (vd->vdev_top_zap != 0) { | |
8627 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
8628 | vd->vdev_top_zap, tx)); | |
8629 | } | |
8630 | if (vd->vdev_leaf_zap != 0) { | |
8631 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
8632 | vd->vdev_leaf_zap, tx)); | |
8633 | } | |
1c27024e | 8634 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
8635 | spa_avz_build(vd->vdev_child[i], avz, tx); |
8636 | } | |
8637 | } | |
8638 | ||
34dc7c2f BB |
8639 | static void |
8640 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
8641 | { | |
8642 | nvlist_t *config; | |
8643 | ||
e0ab3ab5 JS |
8644 | /* |
8645 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
8646 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
8647 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
8648 | * need to rebuild the AVZ although the config may not be dirty. | |
8649 | */ | |
8650 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
8651 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
34dc7c2f BB |
8652 | return; |
8653 | ||
b128c09f BB |
8654 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
8655 | ||
e0ab3ab5 | 8656 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || |
38640550 | 8657 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || |
e0ab3ab5 JS |
8658 | spa->spa_all_vdev_zaps != 0); |
8659 | ||
8660 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
e0ab3ab5 JS |
8661 | /* Make and build the new AVZ */ |
8662 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
8663 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
8664 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
8665 | ||
8666 | /* Diff old AVZ with new one */ | |
1c27024e DB |
8667 | zap_cursor_t zc; |
8668 | zap_attribute_t za; | |
8669 | ||
e0ab3ab5 JS |
8670 | for (zap_cursor_init(&zc, spa->spa_meta_objset, |
8671 | spa->spa_all_vdev_zaps); | |
8672 | zap_cursor_retrieve(&zc, &za) == 0; | |
8673 | zap_cursor_advance(&zc)) { | |
8674 | uint64_t vdzap = za.za_first_integer; | |
8675 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
8676 | vdzap) == ENOENT) { | |
8677 | /* | |
8678 | * ZAP is listed in old AVZ but not in new one; | |
8679 | * destroy it | |
8680 | */ | |
8681 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
8682 | tx)); | |
8683 | } | |
8684 | } | |
8685 | ||
8686 | zap_cursor_fini(&zc); | |
8687 | ||
8688 | /* Destroy the old AVZ */ | |
8689 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
8690 | spa->spa_all_vdev_zaps, tx)); | |
8691 | ||
8692 | /* Replace the old AVZ in the dir obj with the new one */ | |
8693 | VERIFY0(zap_update(spa->spa_meta_objset, | |
8694 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
8695 | sizeof (new_avz), 1, &new_avz, tx)); | |
8696 | ||
8697 | spa->spa_all_vdev_zaps = new_avz; | |
8698 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
8699 | zap_cursor_t zc; | |
8700 | zap_attribute_t za; | |
8701 | ||
8702 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
8703 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
8704 | spa->spa_all_vdev_zaps); | |
8705 | zap_cursor_retrieve(&zc, &za) == 0; | |
8706 | zap_cursor_advance(&zc)) { | |
8707 | uint64_t zap = za.za_first_integer; | |
8708 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
8709 | } | |
8710 | ||
8711 | zap_cursor_fini(&zc); | |
8712 | ||
8713 | /* Destroy and unlink the AVZ itself */ | |
8714 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
8715 | spa->spa_all_vdev_zaps, tx)); | |
8716 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
8717 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
8718 | spa->spa_all_vdev_zaps = 0; | |
8719 | } | |
8720 | ||
8721 | if (spa->spa_all_vdev_zaps == 0) { | |
8722 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
8723 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
8724 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
8725 | } | |
8726 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
8727 | ||
8728 | /* Create ZAPs for vdevs that don't have them. */ | |
8729 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
8730 | ||
b128c09f BB |
8731 | config = spa_config_generate(spa, spa->spa_root_vdev, |
8732 | dmu_tx_get_txg(tx), B_FALSE); | |
8733 | ||
ea0b2538 GW |
8734 | /* |
8735 | * If we're upgrading the spa version then make sure that | |
8736 | * the config object gets updated with the correct version. | |
8737 | */ | |
8738 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
8739 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
8740 | spa->spa_uberblock.ub_version); | |
8741 | ||
b128c09f | 8742 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f | 8743 | |
8a5fc748 | 8744 | nvlist_free(spa->spa_config_syncing); |
34dc7c2f BB |
8745 | spa->spa_config_syncing = config; |
8746 | ||
8747 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
8748 | } | |
8749 | ||
9ae529ec | 8750 | static void |
13fe0198 | 8751 | spa_sync_version(void *arg, dmu_tx_t *tx) |
9ae529ec | 8752 | { |
13fe0198 MA |
8753 | uint64_t *versionp = arg; |
8754 | uint64_t version = *versionp; | |
8755 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9ae529ec CS |
8756 | |
8757 | /* | |
8758 | * Setting the version is special cased when first creating the pool. | |
8759 | */ | |
8760 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
8761 | ||
8dca0a9a | 8762 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); |
9ae529ec CS |
8763 | ASSERT(version >= spa_version(spa)); |
8764 | ||
8765 | spa->spa_uberblock.ub_version = version; | |
8766 | vdev_config_dirty(spa->spa_root_vdev); | |
74756182 MM |
8767 | spa_history_log_internal(spa, "set", tx, "version=%lld", |
8768 | (longlong_t)version); | |
9ae529ec CS |
8769 | } |
8770 | ||
34dc7c2f BB |
8771 | /* |
8772 | * Set zpool properties. | |
8773 | */ | |
8774 | static void | |
13fe0198 | 8775 | spa_sync_props(void *arg, dmu_tx_t *tx) |
34dc7c2f | 8776 | { |
13fe0198 MA |
8777 | nvlist_t *nvp = arg; |
8778 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
34dc7c2f | 8779 | objset_t *mos = spa->spa_meta_objset; |
9ae529ec | 8780 | nvpair_t *elem = NULL; |
b128c09f BB |
8781 | |
8782 | mutex_enter(&spa->spa_props_lock); | |
34dc7c2f | 8783 | |
34dc7c2f | 8784 | while ((elem = nvlist_next_nvpair(nvp, elem))) { |
9ae529ec CS |
8785 | uint64_t intval; |
8786 | char *strval, *fname; | |
8787 | zpool_prop_t prop; | |
8788 | const char *propname; | |
8789 | zprop_type_t proptype; | |
fa86b5db | 8790 | spa_feature_t fid; |
9ae529ec | 8791 | |
31864e3d BB |
8792 | switch (prop = zpool_name_to_prop(nvpair_name(elem))) { |
8793 | case ZPOOL_PROP_INVAL: | |
9ae529ec CS |
8794 | /* |
8795 | * We checked this earlier in spa_prop_validate(). | |
8796 | */ | |
8797 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
8798 | ||
8799 | fname = strchr(nvpair_name(elem), '@') + 1; | |
fa86b5db | 8800 | VERIFY0(zfeature_lookup_name(fname, &fid)); |
9ae529ec | 8801 | |
fa86b5db | 8802 | spa_feature_enable(spa, fid, tx); |
6f1ffb06 MA |
8803 | spa_history_log_internal(spa, "set", tx, |
8804 | "%s=enabled", nvpair_name(elem)); | |
9ae529ec CS |
8805 | break; |
8806 | ||
34dc7c2f | 8807 | case ZPOOL_PROP_VERSION: |
93cf2076 | 8808 | intval = fnvpair_value_uint64(elem); |
34dc7c2f | 8809 | /* |
4e33ba4c | 8810 | * The version is synced separately before other |
9ae529ec | 8811 | * properties and should be correct by now. |
34dc7c2f | 8812 | */ |
9ae529ec | 8813 | ASSERT3U(spa_version(spa), >=, intval); |
34dc7c2f BB |
8814 | break; |
8815 | ||
8816 | case ZPOOL_PROP_ALTROOT: | |
8817 | /* | |
8818 | * 'altroot' is a non-persistent property. It should | |
8819 | * have been set temporarily at creation or import time. | |
8820 | */ | |
8821 | ASSERT(spa->spa_root != NULL); | |
8822 | break; | |
8823 | ||
572e2857 | 8824 | case ZPOOL_PROP_READONLY: |
34dc7c2f BB |
8825 | case ZPOOL_PROP_CACHEFILE: |
8826 | /* | |
e1cfd73f | 8827 | * 'readonly' and 'cachefile' are also non-persistent |
572e2857 | 8828 | * properties. |
34dc7c2f | 8829 | */ |
34dc7c2f | 8830 | break; |
d96eb2b1 | 8831 | case ZPOOL_PROP_COMMENT: |
93cf2076 | 8832 | strval = fnvpair_value_string(elem); |
d96eb2b1 DM |
8833 | if (spa->spa_comment != NULL) |
8834 | spa_strfree(spa->spa_comment); | |
8835 | spa->spa_comment = spa_strdup(strval); | |
8836 | /* | |
8837 | * We need to dirty the configuration on all the vdevs | |
88a48330 BB |
8838 | * so that their labels get updated. We also need to |
8839 | * update the cache file to keep it in sync with the | |
8840 | * MOS version. It's unnecessary to do this for pool | |
8841 | * creation since the vdev's configuration has already | |
8842 | * been dirtied. | |
d96eb2b1 | 8843 | */ |
88a48330 | 8844 | if (tx->tx_txg != TXG_INITIAL) { |
d96eb2b1 | 8845 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
8846 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
8847 | } | |
6f1ffb06 MA |
8848 | spa_history_log_internal(spa, "set", tx, |
8849 | "%s=%s", nvpair_name(elem), strval); | |
d96eb2b1 | 8850 | break; |
658fb802 CB |
8851 | case ZPOOL_PROP_COMPATIBILITY: |
8852 | strval = fnvpair_value_string(elem); | |
8853 | if (spa->spa_compatibility != NULL) | |
8854 | spa_strfree(spa->spa_compatibility); | |
8855 | spa->spa_compatibility = spa_strdup(strval); | |
8856 | /* | |
8857 | * Dirty the configuration on vdevs as above. | |
8858 | */ | |
88a48330 | 8859 | if (tx->tx_txg != TXG_INITIAL) { |
658fb802 | 8860 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
8861 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
8862 | } | |
8863 | ||
658fb802 CB |
8864 | spa_history_log_internal(spa, "set", tx, |
8865 | "%s=%s", nvpair_name(elem), strval); | |
8866 | break; | |
8867 | ||
34dc7c2f BB |
8868 | default: |
8869 | /* | |
8870 | * Set pool property values in the poolprops mos object. | |
8871 | */ | |
34dc7c2f | 8872 | if (spa->spa_pool_props_object == 0) { |
9ae529ec CS |
8873 | spa->spa_pool_props_object = |
8874 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
34dc7c2f | 8875 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, |
9ae529ec | 8876 | tx); |
34dc7c2f | 8877 | } |
34dc7c2f BB |
8878 | |
8879 | /* normalize the property name */ | |
8880 | propname = zpool_prop_to_name(prop); | |
8881 | proptype = zpool_prop_get_type(prop); | |
8882 | ||
8883 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
8884 | ASSERT(proptype == PROP_TYPE_STRING); | |
93cf2076 GW |
8885 | strval = fnvpair_value_string(elem); |
8886 | VERIFY0(zap_update(mos, | |
34dc7c2f | 8887 | spa->spa_pool_props_object, propname, |
93cf2076 | 8888 | 1, strlen(strval) + 1, strval, tx)); |
6f1ffb06 MA |
8889 | spa_history_log_internal(spa, "set", tx, |
8890 | "%s=%s", nvpair_name(elem), strval); | |
34dc7c2f | 8891 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { |
93cf2076 | 8892 | intval = fnvpair_value_uint64(elem); |
34dc7c2f BB |
8893 | |
8894 | if (proptype == PROP_TYPE_INDEX) { | |
8895 | const char *unused; | |
93cf2076 GW |
8896 | VERIFY0(zpool_prop_index_to_string( |
8897 | prop, intval, &unused)); | |
34dc7c2f | 8898 | } |
93cf2076 | 8899 | VERIFY0(zap_update(mos, |
34dc7c2f | 8900 | spa->spa_pool_props_object, propname, |
93cf2076 | 8901 | 8, 1, &intval, tx)); |
6f1ffb06 | 8902 | spa_history_log_internal(spa, "set", tx, |
74756182 MM |
8903 | "%s=%lld", nvpair_name(elem), |
8904 | (longlong_t)intval); | |
34dc7c2f | 8905 | |
44f71818 RY |
8906 | switch (prop) { |
8907 | case ZPOOL_PROP_DELEGATION: | |
8908 | spa->spa_delegation = intval; | |
8909 | break; | |
8910 | case ZPOOL_PROP_BOOTFS: | |
8911 | spa->spa_bootfs = intval; | |
8912 | break; | |
8913 | case ZPOOL_PROP_FAILUREMODE: | |
8914 | spa->spa_failmode = intval; | |
8915 | break; | |
8916 | case ZPOOL_PROP_AUTOTRIM: | |
8917 | spa->spa_autotrim = intval; | |
428870ff | 8918 | spa_async_request(spa, |
44f71818 RY |
8919 | SPA_ASYNC_AUTOTRIM_RESTART); |
8920 | break; | |
8921 | case ZPOOL_PROP_AUTOEXPAND: | |
8922 | spa->spa_autoexpand = intval; | |
8923 | if (tx->tx_txg != TXG_INITIAL) | |
8924 | spa_async_request(spa, | |
8925 | SPA_ASYNC_AUTOEXPAND); | |
8926 | break; | |
8927 | case ZPOOL_PROP_MULTIHOST: | |
8928 | spa->spa_multihost = intval; | |
8929 | break; | |
8930 | default: | |
8931 | break; | |
8932 | } | |
8933 | } else { | |
8934 | ASSERT(0); /* not allowed */ | |
34dc7c2f BB |
8935 | } |
8936 | } | |
8937 | ||
34dc7c2f | 8938 | } |
b128c09f BB |
8939 | |
8940 | mutex_exit(&spa->spa_props_lock); | |
34dc7c2f BB |
8941 | } |
8942 | ||
428870ff BB |
8943 | /* |
8944 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
8945 | * reflect the new version this txg, so there must be no changes this | |
8946 | * txg to anything that the upgrade code depends on after it executes. | |
8947 | * Therefore this must be called after dsl_pool_sync() does the sync | |
8948 | * tasks. | |
8949 | */ | |
8950 | static void | |
8951 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
8952 | { | |
8dc2197b SD |
8953 | if (spa_sync_pass(spa) != 1) |
8954 | return; | |
428870ff | 8955 | |
8dc2197b | 8956 | dsl_pool_t *dp = spa->spa_dsl_pool; |
13fe0198 MA |
8957 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
8958 | ||
428870ff BB |
8959 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && |
8960 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
8961 | dsl_pool_create_origin(dp, tx); | |
8962 | ||
8963 | /* Keeping the origin open increases spa_minref */ | |
8964 | spa->spa_minref += 3; | |
8965 | } | |
8966 | ||
8967 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
8968 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
8969 | dsl_pool_upgrade_clones(dp, tx); | |
8970 | } | |
8971 | ||
8972 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
8973 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
8974 | dsl_pool_upgrade_dir_clones(dp, tx); | |
8975 | ||
8976 | /* Keeping the freedir open increases spa_minref */ | |
8977 | spa->spa_minref += 3; | |
8978 | } | |
9ae529ec CS |
8979 | |
8980 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
8981 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
8982 | spa_feature_create_zap_objects(spa, tx); | |
8983 | } | |
62bdd5eb DL |
8984 | |
8985 | /* | |
8986 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
8987 | * when possibility to use lz4 compression for metadata was added | |
8988 | * Old pools that have this feature enabled must be upgraded to have | |
8989 | * this feature active | |
8990 | */ | |
8991 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
8992 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
8993 | SPA_FEATURE_LZ4_COMPRESS); | |
8994 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
8995 | SPA_FEATURE_LZ4_COMPRESS); | |
8996 | ||
8997 | if (lz4_en && !lz4_ac) | |
8998 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
8999 | } | |
3c67d83a TH |
9000 | |
9001 | /* | |
9002 | * If we haven't written the salt, do so now. Note that the | |
9003 | * feature may not be activated yet, but that's fine since | |
9004 | * the presence of this ZAP entry is backwards compatible. | |
9005 | */ | |
9006 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
9007 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
9008 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9009 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
9010 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
9011 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
9012 | } | |
9013 | ||
13fe0198 | 9014 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
428870ff BB |
9015 | } |
9016 | ||
a1d477c2 MA |
9017 | static void |
9018 | vdev_indirect_state_sync_verify(vdev_t *vd) | |
9019 | { | |
2a8ba608 MM |
9020 | vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping; |
9021 | vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births; | |
a1d477c2 MA |
9022 | |
9023 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
9024 | ASSERT(vim != NULL); | |
9025 | ASSERT(vib != NULL); | |
9026 | } | |
9027 | ||
27f80e85 BB |
9028 | uint64_t obsolete_sm_object = 0; |
9029 | ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); | |
9030 | if (obsolete_sm_object != 0) { | |
a1d477c2 MA |
9031 | ASSERT(vd->vdev_obsolete_sm != NULL); |
9032 | ASSERT(vd->vdev_removing || | |
9033 | vd->vdev_ops == &vdev_indirect_ops); | |
9034 | ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); | |
9035 | ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); | |
27f80e85 | 9036 | ASSERT3U(obsolete_sm_object, ==, |
a1d477c2 MA |
9037 | space_map_object(vd->vdev_obsolete_sm)); |
9038 | ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, | |
9039 | space_map_allocated(vd->vdev_obsolete_sm)); | |
9040 | } | |
9041 | ASSERT(vd->vdev_obsolete_segments != NULL); | |
9042 | ||
9043 | /* | |
9044 | * Since frees / remaps to an indirect vdev can only | |
9045 | * happen in syncing context, the obsolete segments | |
9046 | * tree must be empty when we start syncing. | |
9047 | */ | |
9048 | ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); | |
9049 | } | |
9050 | ||
34dc7c2f | 9051 | /* |
8dc2197b SD |
9052 | * Set the top-level vdev's max queue depth. Evaluate each top-level's |
9053 | * async write queue depth in case it changed. The max queue depth will | |
9054 | * not change in the middle of syncing out this txg. | |
34dc7c2f | 9055 | */ |
8dc2197b SD |
9056 | static void |
9057 | spa_sync_adjust_vdev_max_queue_depth(spa_t *spa) | |
34dc7c2f | 9058 | { |
8dc2197b SD |
9059 | ASSERT(spa_writeable(spa)); |
9060 | ||
34dc7c2f | 9061 | vdev_t *rvd = spa->spa_root_vdev; |
3dfb57a3 DB |
9062 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * |
9063 | zfs_vdev_queue_depth_pct / 100; | |
8dc2197b SD |
9064 | metaslab_class_t *normal = spa_normal_class(spa); |
9065 | metaslab_class_t *special = spa_special_class(spa); | |
9066 | metaslab_class_t *dedup = spa_dedup_class(spa); | |
34dc7c2f | 9067 | |
492f64e9 | 9068 | uint64_t slots_per_allocator = 0; |
1c27024e | 9069 | for (int c = 0; c < rvd->vdev_children; c++) { |
3dfb57a3 | 9070 | vdev_t *tvd = rvd->vdev_child[c]; |
cc99f275 | 9071 | |
8dc2197b | 9072 | metaslab_group_t *mg = tvd->vdev_mg; |
cc99f275 DB |
9073 | if (mg == NULL || !metaslab_group_initialized(mg)) |
9074 | continue; | |
3dfb57a3 | 9075 | |
8dc2197b | 9076 | metaslab_class_t *mc = mg->mg_class; |
cc99f275 | 9077 | if (mc != normal && mc != special && mc != dedup) |
3dfb57a3 DB |
9078 | continue; |
9079 | ||
9080 | /* | |
9081 | * It is safe to do a lock-free check here because only async | |
9082 | * allocations look at mg_max_alloc_queue_depth, and async | |
9083 | * allocations all happen from spa_sync(). | |
9084 | */ | |
32d805c3 | 9085 | for (int i = 0; i < mg->mg_allocators; i++) { |
424fd7c3 | 9086 | ASSERT0(zfs_refcount_count( |
32d805c3 MA |
9087 | &(mg->mg_allocator[i].mga_alloc_queue_depth))); |
9088 | } | |
3dfb57a3 | 9089 | mg->mg_max_alloc_queue_depth = max_queue_depth; |
492f64e9 | 9090 | |
32d805c3 MA |
9091 | for (int i = 0; i < mg->mg_allocators; i++) { |
9092 | mg->mg_allocator[i].mga_cur_max_alloc_queue_depth = | |
492f64e9 PD |
9093 | zfs_vdev_def_queue_depth; |
9094 | } | |
9095 | slots_per_allocator += zfs_vdev_def_queue_depth; | |
3dfb57a3 | 9096 | } |
cc99f275 | 9097 | |
492f64e9 | 9098 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
f8020c93 AM |
9099 | ASSERT0(zfs_refcount_count(&normal->mc_allocator[i]. |
9100 | mca_alloc_slots)); | |
9101 | ASSERT0(zfs_refcount_count(&special->mc_allocator[i]. | |
9102 | mca_alloc_slots)); | |
9103 | ASSERT0(zfs_refcount_count(&dedup->mc_allocator[i]. | |
9104 | mca_alloc_slots)); | |
9105 | normal->mc_allocator[i].mca_alloc_max_slots = | |
9106 | slots_per_allocator; | |
9107 | special->mc_allocator[i].mca_alloc_max_slots = | |
9108 | slots_per_allocator; | |
9109 | dedup->mc_allocator[i].mca_alloc_max_slots = | |
9110 | slots_per_allocator; | |
cc99f275 DB |
9111 | } |
9112 | normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9113 | special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9114 | dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
8dc2197b SD |
9115 | } |
9116 | ||
9117 | static void | |
9118 | spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx) | |
9119 | { | |
9120 | ASSERT(spa_writeable(spa)); | |
3dfb57a3 | 9121 | |
8dc2197b | 9122 | vdev_t *rvd = spa->spa_root_vdev; |
a1d477c2 MA |
9123 | for (int c = 0; c < rvd->vdev_children; c++) { |
9124 | vdev_t *vd = rvd->vdev_child[c]; | |
9125 | vdev_indirect_state_sync_verify(vd); | |
9126 | ||
9127 | if (vdev_indirect_should_condense(vd)) { | |
9128 | spa_condense_indirect_start_sync(vd, tx); | |
9129 | break; | |
9130 | } | |
9131 | } | |
8dc2197b SD |
9132 | } |
9133 | ||
9134 | static void | |
9135 | spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx) | |
9136 | { | |
9137 | objset_t *mos = spa->spa_meta_objset; | |
9138 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9139 | uint64_t txg = tx->tx_txg; | |
9140 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
a1d477c2 | 9141 | |
34dc7c2f | 9142 | do { |
428870ff | 9143 | int pass = ++spa->spa_sync_pass; |
34dc7c2f BB |
9144 | |
9145 | spa_sync_config_object(spa, tx); | |
9146 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
9147 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
9148 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
9149 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
9150 | spa_errlog_sync(spa, txg); | |
9151 | dsl_pool_sync(dp, txg); | |
9152 | ||
93e28d66 SD |
9153 | if (pass < zfs_sync_pass_deferred_free || |
9154 | spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { | |
9155 | /* | |
9156 | * If the log space map feature is active we don't | |
9157 | * care about deferred frees and the deferred bpobj | |
9158 | * as the log space map should effectively have the | |
9159 | * same results (i.e. appending only to one object). | |
9160 | */ | |
e8b96c60 | 9161 | spa_sync_frees(spa, free_bpl, tx); |
428870ff | 9162 | } else { |
905edb40 MA |
9163 | /* |
9164 | * We can not defer frees in pass 1, because | |
9165 | * we sync the deferred frees later in pass 1. | |
9166 | */ | |
9167 | ASSERT3U(pass, >, 1); | |
37f03da8 | 9168 | bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb, |
e8b96c60 | 9169 | &spa->spa_deferred_bpobj, tx); |
34dc7c2f BB |
9170 | } |
9171 | ||
67a1b037 | 9172 | brt_sync(spa, txg); |
428870ff BB |
9173 | ddt_sync(spa, txg); |
9174 | dsl_scan_sync(dp, tx); | |
8dc2197b SD |
9175 | svr_sync(spa, tx); |
9176 | spa_sync_upgrades(spa, tx); | |
34dc7c2f | 9177 | |
93e28d66 SD |
9178 | spa_flush_metaslabs(spa, tx); |
9179 | ||
8dc2197b | 9180 | vdev_t *vd = NULL; |
a1d477c2 MA |
9181 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) |
9182 | != NULL) | |
428870ff BB |
9183 | vdev_sync(vd, txg); |
9184 | ||
8dc2197b SD |
9185 | /* |
9186 | * Note: We need to check if the MOS is dirty because we could | |
9187 | * have marked the MOS dirty without updating the uberblock | |
9188 | * (e.g. if we have sync tasks but no dirty user data). We need | |
9189 | * to check the uberblock's rootbp because it is updated if we | |
9190 | * have synced out dirty data (though in this case the MOS will | |
9191 | * most likely also be dirty due to second order effects, we | |
9192 | * don't want to rely on that here). | |
9193 | */ | |
9194 | if (pass == 1 && | |
9195 | spa->spa_uberblock.ub_rootbp.blk_birth < txg && | |
9196 | !dmu_objset_is_dirty(mos, txg)) { | |
905edb40 | 9197 | /* |
8dc2197b SD |
9198 | * Nothing changed on the first pass, therefore this |
9199 | * TXG is a no-op. Avoid syncing deferred frees, so | |
9200 | * that we can keep this TXG as a no-op. | |
905edb40 | 9201 | */ |
8dc2197b SD |
9202 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); |
9203 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9204 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
9205 | ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg)); | |
9206 | break; | |
905edb40 | 9207 | } |
34dc7c2f | 9208 | |
8dc2197b | 9209 | spa_sync_deferred_frees(spa, tx); |
428870ff | 9210 | } while (dmu_objset_is_dirty(mos, txg)); |
8dc2197b | 9211 | } |
34dc7c2f | 9212 | |
8dc2197b SD |
9213 | /* |
9214 | * Rewrite the vdev configuration (which includes the uberblock) to | |
9215 | * commit the transaction group. | |
9216 | * | |
9217 | * If there are no dirty vdevs, we sync the uberblock to a few random | |
9218 | * top-level vdevs that are known to be visible in the config cache | |
9219 | * (see spa_vdev_add() for a complete description). If there *are* dirty | |
9220 | * vdevs, sync the uberblock to all vdevs. | |
9221 | */ | |
9222 | static void | |
9223 | spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx) | |
9224 | { | |
9225 | vdev_t *rvd = spa->spa_root_vdev; | |
9226 | uint64_t txg = tx->tx_txg; | |
a1d477c2 | 9227 | |
b128c09f | 9228 | for (;;) { |
8dc2197b SD |
9229 | int error = 0; |
9230 | ||
b128c09f BB |
9231 | /* |
9232 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
9233 | * while we're attempting to write the vdev labels. | |
9234 | */ | |
9235 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9236 | ||
9237 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
d2734cce | 9238 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; |
b128c09f BB |
9239 | int svdcount = 0; |
9240 | int children = rvd->vdev_children; | |
29274c9f | 9241 | int c0 = random_in_range(children); |
b128c09f | 9242 | |
1c27024e | 9243 | for (int c = 0; c < children; c++) { |
8dc2197b SD |
9244 | vdev_t *vd = |
9245 | rvd->vdev_child[(c0 + c) % children]; | |
d2734cce SD |
9246 | |
9247 | /* Stop when revisiting the first vdev */ | |
9248 | if (c > 0 && svd[0] == vd) | |
9249 | break; | |
9250 | ||
8dc2197b SD |
9251 | if (vd->vdev_ms_array == 0 || |
9252 | vd->vdev_islog || | |
a1d477c2 | 9253 | !vdev_is_concrete(vd)) |
b128c09f | 9254 | continue; |
d2734cce | 9255 | |
b128c09f | 9256 | svd[svdcount++] = vd; |
6cb8e530 | 9257 | if (svdcount == SPA_SYNC_MIN_VDEVS) |
b128c09f BB |
9258 | break; |
9259 | } | |
b6fcb792 | 9260 | error = vdev_config_sync(svd, svdcount, txg); |
b128c09f BB |
9261 | } else { |
9262 | error = vdev_config_sync(rvd->vdev_child, | |
b6fcb792 | 9263 | rvd->vdev_children, txg); |
34dc7c2f | 9264 | } |
34dc7c2f | 9265 | |
3bc7e0fb GW |
9266 | if (error == 0) |
9267 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
9268 | ||
b128c09f BB |
9269 | spa_config_exit(spa, SCL_STATE, FTAG); |
9270 | ||
9271 | if (error == 0) | |
9272 | break; | |
cec3a0a1 | 9273 | zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR); |
b128c09f BB |
9274 | zio_resume_wait(spa); |
9275 | } | |
8dc2197b SD |
9276 | } |
9277 | ||
9278 | /* | |
9279 | * Sync the specified transaction group. New blocks may be dirtied as | |
9280 | * part of the process, so we iterate until it converges. | |
9281 | */ | |
9282 | void | |
9283 | spa_sync(spa_t *spa, uint64_t txg) | |
9284 | { | |
9285 | vdev_t *vd = NULL; | |
9286 | ||
9287 | VERIFY(spa_writeable(spa)); | |
9288 | ||
9289 | /* | |
9290 | * Wait for i/os issued in open context that need to complete | |
9291 | * before this txg syncs. | |
9292 | */ | |
9293 | (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]); | |
9294 | spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, | |
9295 | ZIO_FLAG_CANFAIL); | |
9296 | ||
67a1b037 PJD |
9297 | /* |
9298 | * Now that there can be no more cloning in this transaction group, | |
9299 | * but we are still before issuing frees, we can process pending BRT | |
9300 | * updates. | |
9301 | */ | |
9302 | brt_pending_apply(spa, txg); | |
9303 | ||
8dc2197b SD |
9304 | /* |
9305 | * Lock out configuration changes. | |
9306 | */ | |
9307 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
9308 | ||
9309 | spa->spa_syncing_txg = txg; | |
9310 | spa->spa_sync_pass = 0; | |
9311 | ||
9312 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
1b50749c AM |
9313 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
9314 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
9315 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
8dc2197b SD |
9316 | } |
9317 | ||
9318 | /* | |
9319 | * If there are any pending vdev state changes, convert them | |
9320 | * into config changes that go out with this transaction group. | |
9321 | */ | |
9322 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9f08b6e3 RY |
9323 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { |
9324 | /* Avoid holding the write lock unless actually necessary */ | |
9325 | if (vd->vdev_aux == NULL) { | |
9326 | vdev_state_clean(vd); | |
9327 | vdev_config_dirty(vd); | |
9328 | continue; | |
9329 | } | |
8dc2197b SD |
9330 | /* |
9331 | * We need the write lock here because, for aux vdevs, | |
9332 | * calling vdev_config_dirty() modifies sav_config. | |
9333 | * This is ugly and will become unnecessary when we | |
9334 | * eliminate the aux vdev wart by integrating all vdevs | |
9335 | * into the root vdev tree. | |
9336 | */ | |
9337 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9338 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
9339 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
9340 | vdev_state_clean(vd); | |
9341 | vdev_config_dirty(vd); | |
9342 | } | |
9343 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9344 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
9345 | } | |
9346 | spa_config_exit(spa, SCL_STATE, FTAG); | |
9347 | ||
9348 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9349 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
9350 | ||
9351 | spa->spa_sync_starttime = gethrtime(); | |
9352 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
9353 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
9354 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
9355 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
9356 | ||
9357 | /* | |
9358 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
9359 | * set spa_deflate if we have no raid-z vdevs. | |
9360 | */ | |
9361 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
9362 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
9363 | vdev_t *rvd = spa->spa_root_vdev; | |
9364 | ||
9365 | int i; | |
9366 | for (i = 0; i < rvd->vdev_children; i++) { | |
9367 | vd = rvd->vdev_child[i]; | |
9368 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
9369 | break; | |
9370 | } | |
9371 | if (i == rvd->vdev_children) { | |
9372 | spa->spa_deflate = TRUE; | |
9373 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9374 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
9375 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
9376 | } | |
9377 | } | |
9378 | ||
9379 | spa_sync_adjust_vdev_max_queue_depth(spa); | |
9380 | ||
9381 | spa_sync_condense_indirect(spa, tx); | |
9382 | ||
9383 | spa_sync_iterate_to_convergence(spa, tx); | |
9384 | ||
9385 | #ifdef ZFS_DEBUG | |
9386 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
9387 | /* | |
9388 | * Make sure that the number of ZAPs for all the vdevs matches | |
9389 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
9390 | * called if the config is dirty; otherwise there may be | |
9391 | * outstanding AVZ operations that weren't completed in | |
9392 | * spa_sync_config_object. | |
9393 | */ | |
9394 | uint64_t all_vdev_zap_entry_count; | |
9395 | ASSERT0(zap_count(spa->spa_meta_objset, | |
9396 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
9397 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
9398 | all_vdev_zap_entry_count); | |
9399 | } | |
9400 | #endif | |
9401 | ||
9402 | if (spa->spa_vdev_removal != NULL) { | |
9403 | ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); | |
9404 | } | |
9405 | ||
9406 | spa_sync_rewrite_vdev_config(spa, tx); | |
34dc7c2f BB |
9407 | dmu_tx_commit(tx); |
9408 | ||
57ddcda1 | 9409 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 GW |
9410 | spa->spa_deadman_tqid = 0; |
9411 | ||
34dc7c2f BB |
9412 | /* |
9413 | * Clear the dirty config list. | |
9414 | */ | |
b128c09f | 9415 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) |
34dc7c2f BB |
9416 | vdev_config_clean(vd); |
9417 | ||
9418 | /* | |
9419 | * Now that the new config has synced transactionally, | |
9420 | * let it become visible to the config cache. | |
9421 | */ | |
9422 | if (spa->spa_config_syncing != NULL) { | |
9423 | spa_config_set(spa, spa->spa_config_syncing); | |
9424 | spa->spa_config_txg = txg; | |
9425 | spa->spa_config_syncing = NULL; | |
9426 | } | |
9427 | ||
428870ff | 9428 | dsl_pool_sync_done(dp, txg); |
34dc7c2f | 9429 | |
492f64e9 | 9430 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
1b50749c AM |
9431 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
9432 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
9433 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
492f64e9 | 9434 | } |
3dfb57a3 | 9435 | |
34dc7c2f BB |
9436 | /* |
9437 | * Update usable space statistics. | |
9438 | */ | |
619f0976 GW |
9439 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) |
9440 | != NULL) | |
34dc7c2f | 9441 | vdev_sync_done(vd, txg); |
f09fda50 PD |
9442 | |
9443 | metaslab_class_evict_old(spa->spa_normal_class, txg); | |
9444 | metaslab_class_evict_old(spa->spa_log_class, txg); | |
9445 | ||
93e28d66 | 9446 | spa_sync_close_syncing_log_sm(spa); |
34dc7c2f | 9447 | |
428870ff BB |
9448 | spa_update_dspace(spa); |
9449 | ||
34dc7c2f BB |
9450 | /* |
9451 | * It had better be the case that we didn't dirty anything | |
9452 | * since vdev_config_sync(). | |
9453 | */ | |
9454 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
9455 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9456 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
428870ff | 9457 | |
d2734cce SD |
9458 | while (zfs_pause_spa_sync) |
9459 | delay(1); | |
9460 | ||
428870ff | 9461 | spa->spa_sync_pass = 0; |
34dc7c2f | 9462 | |
55922e73 GW |
9463 | /* |
9464 | * Update the last synced uberblock here. We want to do this at | |
9465 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
9466 | * will be guaranteed that all the processing associated with | |
9467 | * that txg has been completed. | |
9468 | */ | |
9469 | spa->spa_ubsync = spa->spa_uberblock; | |
b128c09f | 9470 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f | 9471 | |
428870ff BB |
9472 | spa_handle_ignored_writes(spa); |
9473 | ||
34dc7c2f BB |
9474 | /* |
9475 | * If any async tasks have been requested, kick them off. | |
9476 | */ | |
9477 | spa_async_dispatch(spa); | |
9478 | } | |
9479 | ||
9480 | /* | |
9481 | * Sync all pools. We don't want to hold the namespace lock across these | |
9482 | * operations, so we take a reference on the spa_t and drop the lock during the | |
9483 | * sync. | |
9484 | */ | |
9485 | void | |
9486 | spa_sync_allpools(void) | |
9487 | { | |
9488 | spa_t *spa = NULL; | |
9489 | mutex_enter(&spa_namespace_lock); | |
9490 | while ((spa = spa_next(spa)) != NULL) { | |
572e2857 BB |
9491 | if (spa_state(spa) != POOL_STATE_ACTIVE || |
9492 | !spa_writeable(spa) || spa_suspended(spa)) | |
34dc7c2f BB |
9493 | continue; |
9494 | spa_open_ref(spa, FTAG); | |
9495 | mutex_exit(&spa_namespace_lock); | |
9496 | txg_wait_synced(spa_get_dsl(spa), 0); | |
9497 | mutex_enter(&spa_namespace_lock); | |
9498 | spa_close(spa, FTAG); | |
9499 | } | |
9500 | mutex_exit(&spa_namespace_lock); | |
9501 | } | |
9502 | ||
9503 | /* | |
9504 | * ========================================================================== | |
9505 | * Miscellaneous routines | |
9506 | * ========================================================================== | |
9507 | */ | |
9508 | ||
9509 | /* | |
9510 | * Remove all pools in the system. | |
9511 | */ | |
9512 | void | |
9513 | spa_evict_all(void) | |
9514 | { | |
9515 | spa_t *spa; | |
9516 | ||
9517 | /* | |
9518 | * Remove all cached state. All pools should be closed now, | |
9519 | * so every spa in the AVL tree should be unreferenced. | |
9520 | */ | |
9521 | mutex_enter(&spa_namespace_lock); | |
9522 | while ((spa = spa_next(NULL)) != NULL) { | |
9523 | /* | |
9524 | * Stop async tasks. The async thread may need to detach | |
9525 | * a device that's been replaced, which requires grabbing | |
9526 | * spa_namespace_lock, so we must drop it here. | |
9527 | */ | |
9528 | spa_open_ref(spa, FTAG); | |
9529 | mutex_exit(&spa_namespace_lock); | |
9530 | spa_async_suspend(spa); | |
9531 | mutex_enter(&spa_namespace_lock); | |
34dc7c2f BB |
9532 | spa_close(spa, FTAG); |
9533 | ||
9534 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
9535 | spa_unload(spa); | |
9536 | spa_deactivate(spa); | |
9537 | } | |
9538 | spa_remove(spa); | |
9539 | } | |
9540 | mutex_exit(&spa_namespace_lock); | |
9541 | } | |
9542 | ||
9543 | vdev_t * | |
9babb374 | 9544 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) |
34dc7c2f | 9545 | { |
b128c09f BB |
9546 | vdev_t *vd; |
9547 | int i; | |
9548 | ||
9549 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
9550 | return (vd); | |
9551 | ||
9babb374 | 9552 | if (aux) { |
b128c09f BB |
9553 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { |
9554 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
9babb374 BB |
9555 | if (vd->vdev_guid == guid) |
9556 | return (vd); | |
9557 | } | |
9558 | ||
9559 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
9560 | vd = spa->spa_spares.sav_vdevs[i]; | |
b128c09f BB |
9561 | if (vd->vdev_guid == guid) |
9562 | return (vd); | |
9563 | } | |
9564 | } | |
9565 | ||
9566 | return (NULL); | |
34dc7c2f BB |
9567 | } |
9568 | ||
9569 | void | |
9570 | spa_upgrade(spa_t *spa, uint64_t version) | |
9571 | { | |
572e2857 BB |
9572 | ASSERT(spa_writeable(spa)); |
9573 | ||
b128c09f | 9574 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
9575 | |
9576 | /* | |
9577 | * This should only be called for a non-faulted pool, and since a | |
9578 | * future version would result in an unopenable pool, this shouldn't be | |
9579 | * possible. | |
9580 | */ | |
8dca0a9a | 9581 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); |
9b67f605 | 9582 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); |
34dc7c2f BB |
9583 | |
9584 | spa->spa_uberblock.ub_version = version; | |
9585 | vdev_config_dirty(spa->spa_root_vdev); | |
9586 | ||
b128c09f | 9587 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
9588 | |
9589 | txg_wait_synced(spa_get_dsl(spa), 0); | |
9590 | } | |
9591 | ||
49d42425 FU |
9592 | static boolean_t |
9593 | spa_has_aux_vdev(spa_t *spa, uint64_t guid, spa_aux_vdev_t *sav) | |
34dc7c2f | 9594 | { |
14e4e3cb | 9595 | (void) spa; |
34dc7c2f | 9596 | int i; |
49d42425 | 9597 | uint64_t vdev_guid; |
34dc7c2f BB |
9598 | |
9599 | for (i = 0; i < sav->sav_count; i++) | |
9600 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
9601 | return (B_TRUE); | |
9602 | ||
9603 | for (i = 0; i < sav->sav_npending; i++) { | |
9604 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
49d42425 | 9605 | &vdev_guid) == 0 && vdev_guid == guid) |
34dc7c2f BB |
9606 | return (B_TRUE); |
9607 | } | |
9608 | ||
9609 | return (B_FALSE); | |
9610 | } | |
9611 | ||
49d42425 FU |
9612 | boolean_t |
9613 | spa_has_l2cache(spa_t *spa, uint64_t guid) | |
9614 | { | |
9615 | return (spa_has_aux_vdev(spa, guid, &spa->spa_l2cache)); | |
9616 | } | |
9617 | ||
9618 | boolean_t | |
9619 | spa_has_spare(spa_t *spa, uint64_t guid) | |
9620 | { | |
9621 | return (spa_has_aux_vdev(spa, guid, &spa->spa_spares)); | |
9622 | } | |
9623 | ||
b128c09f BB |
9624 | /* |
9625 | * Check if a pool has an active shared spare device. | |
9626 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
9627 | */ | |
9628 | static boolean_t | |
9629 | spa_has_active_shared_spare(spa_t *spa) | |
9630 | { | |
9631 | int i, refcnt; | |
9632 | uint64_t pool; | |
9633 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
9634 | ||
9635 | for (i = 0; i < sav->sav_count; i++) { | |
9636 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
9637 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
9638 | refcnt > 2) | |
9639 | return (B_TRUE); | |
9640 | } | |
9641 | ||
9642 | return (B_FALSE); | |
9643 | } | |
9644 | ||
93e28d66 SD |
9645 | uint64_t |
9646 | spa_total_metaslabs(spa_t *spa) | |
9647 | { | |
9648 | vdev_t *rvd = spa->spa_root_vdev; | |
9649 | ||
9650 | uint64_t m = 0; | |
9651 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
9652 | vdev_t *vd = rvd->vdev_child[c]; | |
9653 | if (!vdev_is_concrete(vd)) | |
9654 | continue; | |
9655 | m += vd->vdev_ms_count; | |
9656 | } | |
9657 | return (m); | |
9658 | } | |
9659 | ||
e60e158e JG |
9660 | /* |
9661 | * Notify any waiting threads that some activity has switched from being in- | |
9662 | * progress to not-in-progress so that the thread can wake up and determine | |
9663 | * whether it is finished waiting. | |
9664 | */ | |
9665 | void | |
9666 | spa_notify_waiters(spa_t *spa) | |
9667 | { | |
9668 | /* | |
9669 | * Acquiring spa_activities_lock here prevents the cv_broadcast from | |
9670 | * happening between the waiting thread's check and cv_wait. | |
9671 | */ | |
9672 | mutex_enter(&spa->spa_activities_lock); | |
9673 | cv_broadcast(&spa->spa_activities_cv); | |
9674 | mutex_exit(&spa->spa_activities_lock); | |
9675 | } | |
9676 | ||
9677 | /* | |
9678 | * Notify any waiting threads that the pool is exporting, and then block until | |
9679 | * they are finished using the spa_t. | |
9680 | */ | |
9681 | void | |
9682 | spa_wake_waiters(spa_t *spa) | |
9683 | { | |
9684 | mutex_enter(&spa->spa_activities_lock); | |
9685 | spa->spa_waiters_cancel = B_TRUE; | |
9686 | cv_broadcast(&spa->spa_activities_cv); | |
9687 | while (spa->spa_waiters != 0) | |
9688 | cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock); | |
9689 | spa->spa_waiters_cancel = B_FALSE; | |
9690 | mutex_exit(&spa->spa_activities_lock); | |
9691 | } | |
9692 | ||
2288d419 | 9693 | /* Whether the vdev or any of its descendants are being initialized/trimmed. */ |
e60e158e | 9694 | static boolean_t |
2288d419 | 9695 | spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity) |
e60e158e JG |
9696 | { |
9697 | spa_t *spa = vd->vdev_spa; | |
e60e158e JG |
9698 | |
9699 | ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER)); | |
9700 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
2288d419 BB |
9701 | ASSERT(activity == ZPOOL_WAIT_INITIALIZE || |
9702 | activity == ZPOOL_WAIT_TRIM); | |
9703 | ||
9704 | kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ? | |
9705 | &vd->vdev_initialize_lock : &vd->vdev_trim_lock; | |
e60e158e JG |
9706 | |
9707 | mutex_exit(&spa->spa_activities_lock); | |
2288d419 | 9708 | mutex_enter(lock); |
e60e158e JG |
9709 | mutex_enter(&spa->spa_activities_lock); |
9710 | ||
2288d419 BB |
9711 | boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ? |
9712 | (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) : | |
9713 | (vd->vdev_trim_state == VDEV_TRIM_ACTIVE); | |
9714 | mutex_exit(lock); | |
e60e158e | 9715 | |
2288d419 | 9716 | if (in_progress) |
e60e158e JG |
9717 | return (B_TRUE); |
9718 | ||
9719 | for (int i = 0; i < vd->vdev_children; i++) { | |
2288d419 BB |
9720 | if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i], |
9721 | activity)) | |
e60e158e JG |
9722 | return (B_TRUE); |
9723 | } | |
9724 | ||
9725 | return (B_FALSE); | |
9726 | } | |
9727 | ||
9728 | /* | |
9729 | * If use_guid is true, this checks whether the vdev specified by guid is | |
2288d419 BB |
9730 | * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool |
9731 | * is being initialized/trimmed. The caller must hold the config lock and | |
9732 | * spa_activities_lock. | |
e60e158e JG |
9733 | */ |
9734 | static int | |
2288d419 BB |
9735 | spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid, |
9736 | zpool_wait_activity_t activity, boolean_t *in_progress) | |
e60e158e JG |
9737 | { |
9738 | mutex_exit(&spa->spa_activities_lock); | |
9739 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
9740 | mutex_enter(&spa->spa_activities_lock); | |
9741 | ||
9742 | vdev_t *vd; | |
9743 | if (use_guid) { | |
9744 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
9745 | if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) { | |
9746 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9747 | return (EINVAL); | |
9748 | } | |
9749 | } else { | |
9750 | vd = spa->spa_root_vdev; | |
9751 | } | |
9752 | ||
2288d419 | 9753 | *in_progress = spa_vdev_activity_in_progress_impl(vd, activity); |
e60e158e JG |
9754 | |
9755 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9756 | return (0); | |
9757 | } | |
9758 | ||
9759 | /* | |
9760 | * Locking for waiting threads | |
9761 | * --------------------------- | |
9762 | * | |
9763 | * Waiting threads need a way to check whether a given activity is in progress, | |
9764 | * and then, if it is, wait for it to complete. Each activity will have some | |
9765 | * in-memory representation of the relevant on-disk state which can be used to | |
9766 | * determine whether or not the activity is in progress. The in-memory state and | |
9767 | * the locking used to protect it will be different for each activity, and may | |
9768 | * not be suitable for use with a cvar (e.g., some state is protected by the | |
9769 | * config lock). To allow waiting threads to wait without any races, another | |
9770 | * lock, spa_activities_lock, is used. | |
9771 | * | |
9772 | * When the state is checked, both the activity-specific lock (if there is one) | |
9773 | * and spa_activities_lock are held. In some cases, the activity-specific lock | |
9774 | * is acquired explicitly (e.g. the config lock). In others, the locking is | |
9775 | * internal to some check (e.g. bpobj_is_empty). After checking, the waiting | |
9776 | * thread releases the activity-specific lock and, if the activity is in | |
9777 | * progress, then cv_waits using spa_activities_lock. | |
9778 | * | |
9779 | * The waiting thread is woken when another thread, one completing some | |
9780 | * activity, updates the state of the activity and then calls | |
9781 | * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only | |
9782 | * needs to hold its activity-specific lock when updating the state, and this | |
9783 | * lock can (but doesn't have to) be dropped before calling spa_notify_waiters. | |
9784 | * | |
9785 | * Because spa_notify_waiters acquires spa_activities_lock before broadcasting, | |
9786 | * and because it is held when the waiting thread checks the state of the | |
9787 | * activity, it can never be the case that the completing thread both updates | |
9788 | * the activity state and cv_broadcasts in between the waiting thread's check | |
9789 | * and cv_wait. Thus, a waiting thread can never miss a wakeup. | |
9790 | * | |
9791 | * In order to prevent deadlock, when the waiting thread does its check, in some | |
9792 | * cases it will temporarily drop spa_activities_lock in order to acquire the | |
9793 | * activity-specific lock. The order in which spa_activities_lock and the | |
9794 | * activity specific lock are acquired in the waiting thread is determined by | |
9795 | * the order in which they are acquired in the completing thread; if the | |
9796 | * completing thread calls spa_notify_waiters with the activity-specific lock | |
9797 | * held, then the waiting thread must also acquire the activity-specific lock | |
9798 | * first. | |
9799 | */ | |
9800 | ||
9801 | static int | |
9802 | spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity, | |
9803 | boolean_t use_tag, uint64_t tag, boolean_t *in_progress) | |
9804 | { | |
9805 | int error = 0; | |
9806 | ||
9807 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
9808 | ||
9809 | switch (activity) { | |
9810 | case ZPOOL_WAIT_CKPT_DISCARD: | |
9811 | *in_progress = | |
9812 | (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) && | |
9813 | zap_contains(spa_meta_objset(spa), | |
9814 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) == | |
9815 | ENOENT); | |
9816 | break; | |
9817 | case ZPOOL_WAIT_FREE: | |
9818 | *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS && | |
9819 | !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) || | |
9820 | spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) || | |
9821 | spa_livelist_delete_check(spa)); | |
9822 | break; | |
9823 | case ZPOOL_WAIT_INITIALIZE: | |
2288d419 BB |
9824 | case ZPOOL_WAIT_TRIM: |
9825 | error = spa_vdev_activity_in_progress(spa, use_tag, tag, | |
9826 | activity, in_progress); | |
e60e158e JG |
9827 | break; |
9828 | case ZPOOL_WAIT_REPLACE: | |
9829 | mutex_exit(&spa->spa_activities_lock); | |
9830 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
9831 | mutex_enter(&spa->spa_activities_lock); | |
9832 | ||
9833 | *in_progress = vdev_replace_in_progress(spa->spa_root_vdev); | |
9834 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9835 | break; | |
9836 | case ZPOOL_WAIT_REMOVE: | |
9837 | *in_progress = (spa->spa_removing_phys.sr_state == | |
9838 | DSS_SCANNING); | |
9839 | break; | |
9840 | case ZPOOL_WAIT_RESILVER: | |
9a49d3f3 BB |
9841 | if ((*in_progress = vdev_rebuild_active(spa->spa_root_vdev))) |
9842 | break; | |
9a70e97f | 9843 | zfs_fallthrough; |
e60e158e JG |
9844 | case ZPOOL_WAIT_SCRUB: |
9845 | { | |
9846 | boolean_t scanning, paused, is_scrub; | |
9847 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
9848 | ||
9849 | is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB); | |
9850 | scanning = (scn->scn_phys.scn_state == DSS_SCANNING); | |
9851 | paused = dsl_scan_is_paused_scrub(scn); | |
9852 | *in_progress = (scanning && !paused && | |
9853 | is_scrub == (activity == ZPOOL_WAIT_SCRUB)); | |
9854 | break; | |
9855 | } | |
9856 | default: | |
9857 | panic("unrecognized value for activity %d", activity); | |
9858 | } | |
9859 | ||
9860 | return (error); | |
9861 | } | |
9862 | ||
9863 | static int | |
9864 | spa_wait_common(const char *pool, zpool_wait_activity_t activity, | |
9865 | boolean_t use_tag, uint64_t tag, boolean_t *waited) | |
9866 | { | |
9867 | /* | |
9868 | * The tag is used to distinguish between instances of an activity. | |
2288d419 BB |
9869 | * 'initialize' and 'trim' are the only activities that we use this for. |
9870 | * The other activities can only have a single instance in progress in a | |
9871 | * pool at one time, making the tag unnecessary. | |
e60e158e JG |
9872 | * |
9873 | * There can be multiple devices being replaced at once, but since they | |
9874 | * all finish once resilvering finishes, we don't bother keeping track | |
9875 | * of them individually, we just wait for them all to finish. | |
9876 | */ | |
2288d419 BB |
9877 | if (use_tag && activity != ZPOOL_WAIT_INITIALIZE && |
9878 | activity != ZPOOL_WAIT_TRIM) | |
e60e158e JG |
9879 | return (EINVAL); |
9880 | ||
9881 | if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES) | |
9882 | return (EINVAL); | |
9883 | ||
9884 | spa_t *spa; | |
9885 | int error = spa_open(pool, &spa, FTAG); | |
9886 | if (error != 0) | |
9887 | return (error); | |
9888 | ||
9889 | /* | |
9890 | * Increment the spa's waiter count so that we can call spa_close and | |
9891 | * still ensure that the spa_t doesn't get freed before this thread is | |
9892 | * finished with it when the pool is exported. We want to call spa_close | |
9893 | * before we start waiting because otherwise the additional ref would | |
9894 | * prevent the pool from being exported or destroyed throughout the | |
9895 | * potentially long wait. | |
9896 | */ | |
9897 | mutex_enter(&spa->spa_activities_lock); | |
9898 | spa->spa_waiters++; | |
9899 | spa_close(spa, FTAG); | |
9900 | ||
9901 | *waited = B_FALSE; | |
9902 | for (;;) { | |
9903 | boolean_t in_progress; | |
9904 | error = spa_activity_in_progress(spa, activity, use_tag, tag, | |
9905 | &in_progress); | |
9906 | ||
b24771a8 | 9907 | if (error || !in_progress || spa->spa_waiters_cancel) |
e60e158e JG |
9908 | break; |
9909 | ||
9910 | *waited = B_TRUE; | |
9911 | ||
9912 | if (cv_wait_sig(&spa->spa_activities_cv, | |
9913 | &spa->spa_activities_lock) == 0) { | |
9914 | error = EINTR; | |
9915 | break; | |
9916 | } | |
9917 | } | |
9918 | ||
9919 | spa->spa_waiters--; | |
9920 | cv_signal(&spa->spa_waiters_cv); | |
9921 | mutex_exit(&spa->spa_activities_lock); | |
9922 | ||
9923 | return (error); | |
9924 | } | |
9925 | ||
9926 | /* | |
9927 | * Wait for a particular instance of the specified activity to complete, where | |
9928 | * the instance is identified by 'tag' | |
9929 | */ | |
9930 | int | |
9931 | spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag, | |
9932 | boolean_t *waited) | |
9933 | { | |
9934 | return (spa_wait_common(pool, activity, B_TRUE, tag, waited)); | |
9935 | } | |
9936 | ||
9937 | /* | |
9938 | * Wait for all instances of the specified activity complete | |
9939 | */ | |
9940 | int | |
9941 | spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited) | |
9942 | { | |
9943 | ||
9944 | return (spa_wait_common(pool, activity, B_FALSE, 0, waited)); | |
9945 | } | |
9946 | ||
a1d477c2 | 9947 | sysevent_t * |
12fa0466 DE |
9948 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
9949 | { | |
9950 | sysevent_t *ev = NULL; | |
9951 | #ifdef _KERNEL | |
9952 | nvlist_t *resource; | |
9953 | ||
9954 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
9955 | if (resource) { | |
9956 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
9957 | ev->resource = resource; | |
9958 | } | |
14e4e3cb AZ |
9959 | #else |
9960 | (void) spa, (void) vd, (void) hist_nvl, (void) name; | |
12fa0466 DE |
9961 | #endif |
9962 | return (ev); | |
9963 | } | |
9964 | ||
a1d477c2 | 9965 | void |
12fa0466 DE |
9966 | spa_event_post(sysevent_t *ev) |
9967 | { | |
9968 | #ifdef _KERNEL | |
9969 | if (ev) { | |
9970 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
9971 | kmem_free(ev, sizeof (*ev)); | |
9972 | } | |
14e4e3cb AZ |
9973 | #else |
9974 | (void) ev; | |
12fa0466 DE |
9975 | #endif |
9976 | } | |
9977 | ||
34dc7c2f | 9978 | /* |
fb390aaf HR |
9979 | * Post a zevent corresponding to the given sysevent. The 'name' must be one |
9980 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
34dc7c2f BB |
9981 | * filled in from the spa and (optionally) the vdev. This doesn't do anything |
9982 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
9983 | * or zdb as real changes. | |
9984 | */ | |
9985 | void | |
12fa0466 | 9986 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
34dc7c2f | 9987 | { |
12fa0466 | 9988 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); |
34dc7c2f | 9989 | } |
c28b2279 | 9990 | |
c28b2279 BB |
9991 | /* state manipulation functions */ |
9992 | EXPORT_SYMBOL(spa_open); | |
9993 | EXPORT_SYMBOL(spa_open_rewind); | |
9994 | EXPORT_SYMBOL(spa_get_stats); | |
9995 | EXPORT_SYMBOL(spa_create); | |
c28b2279 BB |
9996 | EXPORT_SYMBOL(spa_import); |
9997 | EXPORT_SYMBOL(spa_tryimport); | |
9998 | EXPORT_SYMBOL(spa_destroy); | |
9999 | EXPORT_SYMBOL(spa_export); | |
10000 | EXPORT_SYMBOL(spa_reset); | |
10001 | EXPORT_SYMBOL(spa_async_request); | |
10002 | EXPORT_SYMBOL(spa_async_suspend); | |
10003 | EXPORT_SYMBOL(spa_async_resume); | |
10004 | EXPORT_SYMBOL(spa_inject_addref); | |
10005 | EXPORT_SYMBOL(spa_inject_delref); | |
10006 | EXPORT_SYMBOL(spa_scan_stat_init); | |
10007 | EXPORT_SYMBOL(spa_scan_get_stats); | |
10008 | ||
e1cfd73f | 10009 | /* device manipulation */ |
c28b2279 BB |
10010 | EXPORT_SYMBOL(spa_vdev_add); |
10011 | EXPORT_SYMBOL(spa_vdev_attach); | |
10012 | EXPORT_SYMBOL(spa_vdev_detach); | |
c28b2279 BB |
10013 | EXPORT_SYMBOL(spa_vdev_setpath); |
10014 | EXPORT_SYMBOL(spa_vdev_setfru); | |
10015 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
10016 | ||
10017 | /* spare statech is global across all pools) */ | |
10018 | EXPORT_SYMBOL(spa_spare_add); | |
10019 | EXPORT_SYMBOL(spa_spare_remove); | |
10020 | EXPORT_SYMBOL(spa_spare_exists); | |
10021 | EXPORT_SYMBOL(spa_spare_activate); | |
10022 | ||
10023 | /* L2ARC statech is global across all pools) */ | |
10024 | EXPORT_SYMBOL(spa_l2cache_add); | |
10025 | EXPORT_SYMBOL(spa_l2cache_remove); | |
10026 | EXPORT_SYMBOL(spa_l2cache_exists); | |
10027 | EXPORT_SYMBOL(spa_l2cache_activate); | |
10028 | EXPORT_SYMBOL(spa_l2cache_drop); | |
10029 | ||
10030 | /* scanning */ | |
10031 | EXPORT_SYMBOL(spa_scan); | |
10032 | EXPORT_SYMBOL(spa_scan_stop); | |
10033 | ||
10034 | /* spa syncing */ | |
10035 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
10036 | EXPORT_SYMBOL(spa_sync_allpools); | |
10037 | ||
10038 | /* properties */ | |
10039 | EXPORT_SYMBOL(spa_prop_set); | |
10040 | EXPORT_SYMBOL(spa_prop_get); | |
10041 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
10042 | ||
10043 | /* asynchronous event notification */ | |
10044 | EXPORT_SYMBOL(spa_event_notify); | |
dea377c0 | 10045 | |
c8242a96 | 10046 | /* BEGIN CSTYLED */ |
fdc2d303 | 10047 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, UINT, ZMOD_RW, |
458f8231 | 10048 | "log2 fraction of arc that can be used by inflight I/Os when " |
03fdcb9a | 10049 | "verifying pool during import"); |
7ada752a | 10050 | /* END CSTYLED */ |
dea377c0 | 10051 | |
03fdcb9a | 10052 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW, |
dea377c0 MA |
10053 | "Set to traverse metadata on pool import"); |
10054 | ||
03fdcb9a | 10055 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW, |
dea377c0 | 10056 | "Set to traverse data on pool import"); |
dcb6bed1 | 10057 | |
03fdcb9a | 10058 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW, |
6cb8e530 PZ |
10059 | "Print vdev tree to zfs_dbgmsg during pool import"); |
10060 | ||
03fdcb9a | 10061 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD, |
dcb6bed1 D |
10062 | "Percentage of CPUs to run an IO worker thread"); |
10063 | ||
7457b024 AM |
10064 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_tpq, UINT, ZMOD_RD, |
10065 | "Number of threads per IO worker taskqueue"); | |
10066 | ||
7ada752a | 10067 | /* BEGIN CSTYLED */ |
ab8d9c17 | 10068 | ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, U64, ZMOD_RW, |
03fdcb9a MM |
10069 | "Allow importing pool with up to this number of missing top-level " |
10070 | "vdevs (in read-only mode)"); | |
7ada752a | 10071 | /* END CSTYLED */ |
6cb8e530 | 10072 | |
7ada752a AZ |
10073 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, |
10074 | ZMOD_RW, "Set the livelist condense zthr to pause"); | |
03fdcb9a | 10075 | |
7ada752a AZ |
10076 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, |
10077 | ZMOD_RW, "Set the livelist condense synctask to pause"); | |
37f03da8 | 10078 | |
7ada752a AZ |
10079 | /* BEGIN CSTYLED */ |
10080 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, | |
10081 | INT, ZMOD_RW, | |
37f03da8 | 10082 | "Whether livelist condensing was canceled in the synctask"); |
03fdcb9a | 10083 | |
7ada752a AZ |
10084 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, |
10085 | INT, ZMOD_RW, | |
37f03da8 SH |
10086 | "Whether livelist condensing was canceled in the zthr function"); |
10087 | ||
7ada752a AZ |
10088 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, |
10089 | ZMOD_RW, | |
03fdcb9a MM |
10090 | "Whether extra ALLOC blkptrs were added to a livelist entry while it " |
10091 | "was being condensed"); | |
37f03da8 | 10092 | /* END CSTYLED */ |