]>
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. |
b1e46f86 | 24 | * Copyright (c) 2011, 2024 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> |
9d618615 | 36 | * Copyright (c) 2023 Hewlett Packard Enterprise Development LP. |
a38718a6 | 37 | */ |
34dc7c2f | 38 | |
34dc7c2f | 39 | /* |
e49f1e20 WA |
40 | * SPA: Storage Pool Allocator |
41 | * | |
34dc7c2f BB |
42 | * This file contains all the routines used when modifying on-disk SPA state. |
43 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
44 | * pool. | |
45 | */ | |
46 | ||
47 | #include <sys/zfs_context.h> | |
48 | #include <sys/fm/fs/zfs.h> | |
49 | #include <sys/spa_impl.h> | |
50 | #include <sys/zio.h> | |
51 | #include <sys/zio_checksum.h> | |
34dc7c2f BB |
52 | #include <sys/dmu.h> |
53 | #include <sys/dmu_tx.h> | |
54 | #include <sys/zap.h> | |
55 | #include <sys/zil.h> | |
67a1b037 | 56 | #include <sys/brt.h> |
428870ff | 57 | #include <sys/ddt.h> |
34dc7c2f | 58 | #include <sys/vdev_impl.h> |
a1d477c2 MA |
59 | #include <sys/vdev_removal.h> |
60 | #include <sys/vdev_indirect_mapping.h> | |
61 | #include <sys/vdev_indirect_births.h> | |
619f0976 | 62 | #include <sys/vdev_initialize.h> |
9a49d3f3 | 63 | #include <sys/vdev_rebuild.h> |
1b939560 | 64 | #include <sys/vdev_trim.h> |
c28b2279 | 65 | #include <sys/vdev_disk.h> |
5caeef02 | 66 | #include <sys/vdev_raidz.h> |
b2255edc | 67 | #include <sys/vdev_draid.h> |
34dc7c2f | 68 | #include <sys/metaslab.h> |
428870ff | 69 | #include <sys/metaslab_impl.h> |
379ca9cf | 70 | #include <sys/mmp.h> |
34dc7c2f BB |
71 | #include <sys/uberblock_impl.h> |
72 | #include <sys/txg.h> | |
73 | #include <sys/avl.h> | |
a1d477c2 | 74 | #include <sys/bpobj.h> |
34dc7c2f BB |
75 | #include <sys/dmu_traverse.h> |
76 | #include <sys/dmu_objset.h> | |
77 | #include <sys/unique.h> | |
78 | #include <sys/dsl_pool.h> | |
79 | #include <sys/dsl_dataset.h> | |
80 | #include <sys/dsl_dir.h> | |
81 | #include <sys/dsl_prop.h> | |
82 | #include <sys/dsl_synctask.h> | |
83 | #include <sys/fs/zfs.h> | |
84 | #include <sys/arc.h> | |
85 | #include <sys/callb.h> | |
86 | #include <sys/systeminfo.h> | |
9babb374 | 87 | #include <sys/zfs_ioctl.h> |
428870ff | 88 | #include <sys/dsl_scan.h> |
9ae529ec | 89 | #include <sys/zfeature.h> |
13fe0198 | 90 | #include <sys/dsl_destroy.h> |
526af785 | 91 | #include <sys/zvol.h> |
34dc7c2f | 92 | |
d164b209 | 93 | #ifdef _KERNEL |
12fa0466 DE |
94 | #include <sys/fm/protocol.h> |
95 | #include <sys/fm/util.h> | |
428870ff | 96 | #include <sys/callb.h> |
d164b209 | 97 | #include <sys/zone.h> |
c8242a96 | 98 | #include <sys/vmsystm.h> |
d164b209 BB |
99 | #endif /* _KERNEL */ |
100 | ||
34dc7c2f BB |
101 | #include "zfs_prop.h" |
102 | #include "zfs_comutil.h" | |
3bd4df38 | 103 | #include <cityhash.h> |
34dc7c2f | 104 | |
b5e60918 RN |
105 | /* |
106 | * spa_thread() existed on Illumos as a parent thread for the various worker | |
107 | * threads that actually run the pool, as a way to both reference the entire | |
108 | * pool work as a single object, and to share properties like scheduling | |
109 | * options. It has not yet been adapted to Linux or FreeBSD. This define is | |
110 | * used to mark related parts of the code to make things easier for the reader, | |
111 | * and to compile this code out. It can be removed when someone implements it, | |
112 | * moves it to some Illumos-specific place, or removes it entirely. | |
113 | */ | |
114 | #undef HAVE_SPA_THREAD | |
115 | ||
116 | /* | |
117 | * The "System Duty Cycle" scheduling class is an Illumos feature to help | |
118 | * prevent CPU-intensive kernel threads from affecting latency on interactive | |
119 | * threads. It doesn't exist on Linux or FreeBSD, so the supporting code is | |
120 | * gated behind a define. On Illumos SDC depends on spa_thread(), but | |
121 | * spa_thread() also has other uses, so this is a separate define. | |
122 | */ | |
123 | #undef HAVE_SYSDC | |
124 | ||
e6cfd633 WA |
125 | /* |
126 | * The interval, in seconds, at which failed configuration cache file writes | |
127 | * should be retried. | |
128 | */ | |
a1d477c2 | 129 | int zfs_ccw_retry_interval = 300; |
e6cfd633 | 130 | |
428870ff | 131 | typedef enum zti_modes { |
7ef5e54e | 132 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ |
7457b024 | 133 | ZTI_MODE_SCALE, /* Taskqs scale with CPUs. */ |
3bd4df38 | 134 | ZTI_MODE_SYNC, /* sync thread assigned */ |
7ef5e54e AL |
135 | ZTI_MODE_NULL, /* don't create a taskq */ |
136 | ZTI_NMODES | |
428870ff | 137 | } zti_modes_t; |
34dc7c2f | 138 | |
7ef5e54e AL |
139 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } |
140 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
7457b024 | 141 | #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 } |
3bd4df38 | 142 | #define ZTI_SYNC { ZTI_MODE_SYNC, 0, 1 } |
7ef5e54e | 143 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } |
9babb374 | 144 | |
7ef5e54e AL |
145 | #define ZTI_N(n) ZTI_P(n, 1) |
146 | #define ZTI_ONE ZTI_N(1) | |
9babb374 BB |
147 | |
148 | typedef struct zio_taskq_info { | |
7ef5e54e | 149 | zti_modes_t zti_mode; |
428870ff | 150 | uint_t zti_value; |
7ef5e54e | 151 | uint_t zti_count; |
9babb374 BB |
152 | } zio_taskq_info_t; |
153 | ||
154 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
451041db | 155 | "iss", "iss_h", "int", "int_h" |
9babb374 BB |
156 | }; |
157 | ||
428870ff | 158 | /* |
7ef5e54e AL |
159 | * This table defines the taskq settings for each ZFS I/O type. When |
160 | * initializing a pool, we use this table to create an appropriately sized | |
161 | * taskq. Some operations are low volume and therefore have a small, static | |
162 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
3bd4df38 | 163 | * macros. Other operations process a large amount of data; the ZTI_SCALE |
7ef5e54e | 164 | * macro causes us to create a taskq oriented for throughput. Some operations |
1b939560 | 165 | * are so high frequency and short-lived that the taskq itself can become a |
7ef5e54e AL |
166 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an |
167 | * additional degree of parallelism specified by the number of threads per- | |
168 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
3bd4df38 EN |
169 | * particular taskq is chosen at random. ZTI_SCALE uses a number of taskqs |
170 | * that scales with the number of CPUs. | |
7ef5e54e AL |
171 | * |
172 | * The different taskq priorities are to handle the different contexts (issue | |
04bae5ec AM |
173 | * and interrupt) and then to reserve threads for high priority I/Os that |
174 | * need to be handled with minimum delay. Illumos taskq has unfair TQ_FRONT | |
175 | * implementation, so separate high priority threads are used there. | |
428870ff | 176 | */ |
6930ecbb | 177 | static zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { |
428870ff | 178 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ |
7ef5e54e | 179 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ |
7457b024 | 180 | { ZTI_N(8), ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* READ */ |
04bae5ec | 181 | #ifdef illumos |
3bd4df38 | 182 | { ZTI_SYNC, ZTI_N(5), ZTI_SCALE, ZTI_N(5) }, /* WRITE */ |
04bae5ec AM |
183 | #else |
184 | { ZTI_SYNC, ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* WRITE */ | |
185 | #endif | |
7457b024 | 186 | { ZTI_SCALE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ |
7ef5e54e | 187 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ |
d7605ae7 | 188 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FLUSH */ |
1b939560 | 189 | { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */ |
9babb374 BB |
190 | }; |
191 | ||
13fe0198 MA |
192 | static void spa_sync_version(void *arg, dmu_tx_t *tx); |
193 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
b128c09f | 194 | static boolean_t spa_has_active_shared_spare(spa_t *spa); |
a926aab9 AZ |
195 | static int spa_load_impl(spa_t *spa, spa_import_type_t type, |
196 | const char **ereport); | |
572e2857 | 197 | static void spa_vdev_resilver_done(spa_t *spa); |
428870ff | 198 | |
342357cd AM |
199 | /* |
200 | * Percentage of all CPUs that can be used by the metaslab preload taskq. | |
201 | */ | |
202 | static uint_t metaslab_preload_pct = 50; | |
203 | ||
18168da7 AZ |
204 | static uint_t zio_taskq_batch_pct = 80; /* 1 thread per cpu in pset */ |
205 | static uint_t zio_taskq_batch_tpq; /* threads per taskq */ | |
b5e60918 RN |
206 | |
207 | #ifdef HAVE_SYSDC | |
18168da7 AZ |
208 | static const boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ |
209 | static const uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
b5e60918 | 210 | #endif |
428870ff | 211 | |
b5e60918 | 212 | #ifdef HAVE_SPA_THREAD |
18168da7 | 213 | static const boolean_t spa_create_process = B_TRUE; /* no process => no sysdc */ |
b5e60918 | 214 | #endif |
428870ff | 215 | |
645b8330 | 216 | static uint_t zio_taskq_write_tpq = 16; |
3bd4df38 | 217 | |
afd2f7b7 PZ |
218 | /* |
219 | * Report any spa_load_verify errors found, but do not fail spa_load. | |
220 | * This is used by zdb to analyze non-idle pools. | |
221 | */ | |
222 | boolean_t spa_load_verify_dryrun = B_FALSE; | |
223 | ||
e39fe05b FU |
224 | /* |
225 | * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ). | |
226 | * This is used by zdb for spacemaps verification. | |
227 | */ | |
228 | boolean_t spa_mode_readable_spacemaps = B_FALSE; | |
229 | ||
428870ff BB |
230 | /* |
231 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
232 | * to get the vdev stats associated with the imported devices. | |
233 | */ | |
234 | #define TRYIMPORT_NAME "$import" | |
34dc7c2f | 235 | |
6cb8e530 PZ |
236 | /* |
237 | * For debugging purposes: print out vdev tree during pool import. | |
238 | */ | |
18168da7 | 239 | static int spa_load_print_vdev_tree = B_FALSE; |
6cb8e530 PZ |
240 | |
241 | /* | |
242 | * A non-zero value for zfs_max_missing_tvds means that we allow importing | |
243 | * pools with missing top-level vdevs. This is strictly intended for advanced | |
244 | * pool recovery cases since missing data is almost inevitable. Pools with | |
245 | * missing devices can only be imported read-only for safety reasons, and their | |
246 | * fail-mode will be automatically set to "continue". | |
247 | * | |
248 | * With 1 missing vdev we should be able to import the pool and mount all | |
249 | * datasets. User data that was not modified after the missing device has been | |
250 | * added should be recoverable. This means that snapshots created prior to the | |
251 | * addition of that device should be completely intact. | |
252 | * | |
253 | * With 2 missing vdevs, some datasets may fail to mount since there are | |
254 | * dataset statistics that are stored as regular metadata. Some data might be | |
255 | * recoverable if those vdevs were added recently. | |
256 | * | |
257 | * With 3 or more missing vdevs, the pool is severely damaged and MOS entries | |
258 | * may be missing entirely. Chances of data recovery are very low. Note that | |
259 | * there are also risks of performing an inadvertent rewind as we might be | |
260 | * missing all the vdevs with the latest uberblocks. | |
261 | */ | |
ab8d9c17 | 262 | uint64_t zfs_max_missing_tvds = 0; |
6cb8e530 PZ |
263 | |
264 | /* | |
265 | * The parameters below are similar to zfs_max_missing_tvds but are only | |
266 | * intended for a preliminary open of the pool with an untrusted config which | |
267 | * might be incomplete or out-dated. | |
268 | * | |
269 | * We are more tolerant for pools opened from a cachefile since we could have | |
270 | * an out-dated cachefile where a device removal was not registered. | |
271 | * We could have set the limit arbitrarily high but in the case where devices | |
272 | * are really missing we would want to return the proper error codes; we chose | |
273 | * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available | |
274 | * and we get a chance to retrieve the trusted config. | |
275 | */ | |
276 | uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; | |
d2734cce | 277 | |
6cb8e530 PZ |
278 | /* |
279 | * In the case where config was assembled by scanning device paths (/dev/dsks | |
280 | * by default) we are less tolerant since all the existing devices should have | |
281 | * been detected and we want spa_load to return the right error codes. | |
282 | */ | |
283 | uint64_t zfs_max_missing_tvds_scan = 0; | |
284 | ||
d2734cce SD |
285 | /* |
286 | * Debugging aid that pauses spa_sync() towards the end. | |
287 | */ | |
18168da7 | 288 | static const boolean_t zfs_pause_spa_sync = B_FALSE; |
d2734cce | 289 | |
37f03da8 SH |
290 | /* |
291 | * Variables to indicate the livelist condense zthr func should wait at certain | |
292 | * points for the livelist to be removed - used to test condense/destroy races | |
293 | */ | |
18168da7 AZ |
294 | static int zfs_livelist_condense_zthr_pause = 0; |
295 | static int zfs_livelist_condense_sync_pause = 0; | |
37f03da8 SH |
296 | |
297 | /* | |
298 | * Variables to track whether or not condense cancellation has been | |
299 | * triggered in testing. | |
300 | */ | |
18168da7 AZ |
301 | static int zfs_livelist_condense_sync_cancel = 0; |
302 | static int zfs_livelist_condense_zthr_cancel = 0; | |
37f03da8 SH |
303 | |
304 | /* | |
305 | * Variable to track whether or not extra ALLOC blkptrs were added to a | |
306 | * livelist entry while it was being condensed (caused by the way we track | |
307 | * remapped blkptrs in dbuf_remap_impl) | |
308 | */ | |
18168da7 | 309 | static int zfs_livelist_condense_new_alloc = 0; |
37f03da8 | 310 | |
34dc7c2f BB |
311 | /* |
312 | * ========================================================================== | |
313 | * SPA properties routines | |
314 | * ========================================================================== | |
315 | */ | |
316 | ||
317 | /* | |
318 | * Add a (source=src, propname=propval) list to an nvlist. | |
319 | */ | |
320 | static void | |
a926aab9 | 321 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, const char *strval, |
34dc7c2f BB |
322 | uint64_t intval, zprop_source_t src) |
323 | { | |
324 | const char *propname = zpool_prop_to_name(prop); | |
325 | nvlist_t *propval; | |
326 | ||
65ad5d11 AJ |
327 | propval = fnvlist_alloc(); |
328 | fnvlist_add_uint64(propval, ZPROP_SOURCE, src); | |
34dc7c2f BB |
329 | |
330 | if (strval != NULL) | |
65ad5d11 | 331 | fnvlist_add_string(propval, ZPROP_VALUE, strval); |
34dc7c2f | 332 | else |
65ad5d11 | 333 | fnvlist_add_uint64(propval, ZPROP_VALUE, intval); |
34dc7c2f | 334 | |
65ad5d11 | 335 | fnvlist_add_nvlist(nvl, propname, propval); |
34dc7c2f BB |
336 | nvlist_free(propval); |
337 | } | |
338 | ||
8eae2d21 AJ |
339 | /* |
340 | * Add a user property (source=src, propname=propval) to an nvlist. | |
341 | */ | |
342 | static void | |
343 | spa_prop_add_user(nvlist_t *nvl, const char *propname, char *strval, | |
344 | zprop_source_t src) | |
345 | { | |
346 | nvlist_t *propval; | |
347 | ||
348 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
349 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); | |
350 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
351 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
352 | nvlist_free(propval); | |
353 | } | |
354 | ||
34dc7c2f BB |
355 | /* |
356 | * Get property values from the spa configuration. | |
357 | */ | |
358 | static void | |
359 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
360 | { | |
1bd201e7 | 361 | vdev_t *rvd = spa->spa_root_vdev; |
9ae529ec | 362 | dsl_pool_t *pool = spa->spa_dsl_pool; |
f3a7f661 | 363 | uint64_t size, alloc, cap, version; |
82ab6848 | 364 | const zprop_source_t src = ZPROP_SRC_NONE; |
b128c09f | 365 | spa_config_dirent_t *dp; |
f3a7f661 | 366 | metaslab_class_t *mc = spa_normal_class(spa); |
b128c09f BB |
367 | |
368 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
34dc7c2f | 369 | |
1bd201e7 | 370 | if (rvd != NULL) { |
cc99f275 DB |
371 | alloc = metaslab_class_get_alloc(mc); |
372 | alloc += metaslab_class_get_alloc(spa_special_class(spa)); | |
373 | alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); | |
aa755b35 | 374 | alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa)); |
cc99f275 DB |
375 | |
376 | size = metaslab_class_get_space(mc); | |
377 | size += metaslab_class_get_space(spa_special_class(spa)); | |
378 | size += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 | 379 | size += metaslab_class_get_space(spa_embedded_log_class(spa)); |
cc99f275 | 380 | |
d164b209 BB |
381 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); |
382 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
428870ff BB |
383 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); |
384 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
385 | size - alloc, src); | |
d2734cce SD |
386 | spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, |
387 | spa->spa_checkpoint_info.sci_dspace, src); | |
1bd201e7 | 388 | |
f3a7f661 GW |
389 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, |
390 | metaslab_class_fragmentation(mc), src); | |
391 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
392 | metaslab_class_expandable_space(mc), src); | |
572e2857 | 393 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, |
da92d5cb | 394 | (spa_mode(spa) == SPA_MODE_READ), src); |
d164b209 | 395 | |
428870ff | 396 | cap = (size == 0) ? 0 : (alloc * 100 / size); |
d164b209 BB |
397 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); |
398 | ||
428870ff BB |
399 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, |
400 | ddt_get_pool_dedup_ratio(spa), src); | |
67a1b037 PJD |
401 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONEUSED, NULL, |
402 | brt_get_used(spa), src); | |
403 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONESAVED, NULL, | |
404 | brt_get_saved(spa), src); | |
405 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONERATIO, NULL, | |
406 | brt_get_ratio(spa), src); | |
428870ff | 407 | |
d164b209 | 408 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, |
1bd201e7 | 409 | rvd->vdev_state, src); |
d164b209 BB |
410 | |
411 | version = spa_version(spa); | |
82ab6848 HM |
412 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { |
413 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
414 | version, ZPROP_SRC_DEFAULT); | |
415 | } else { | |
416 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
417 | version, ZPROP_SRC_LOCAL); | |
418 | } | |
a448a255 SD |
419 | spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID, |
420 | NULL, spa_load_guid(spa), src); | |
d164b209 | 421 | } |
34dc7c2f | 422 | |
9ae529ec | 423 | if (pool != NULL) { |
9ae529ec CS |
424 | /* |
425 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
426 | * when opening pools before this version freedir will be NULL. | |
427 | */ | |
fbeddd60 | 428 | if (pool->dp_free_dir != NULL) { |
9ae529ec | 429 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, |
d683ddbb JG |
430 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, |
431 | src); | |
9ae529ec CS |
432 | } else { |
433 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
434 | NULL, 0, src); | |
435 | } | |
fbeddd60 MA |
436 | |
437 | if (pool->dp_leak_dir != NULL) { | |
438 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
d683ddbb JG |
439 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, |
440 | src); | |
fbeddd60 MA |
441 | } else { |
442 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
443 | NULL, 0, src); | |
444 | } | |
9ae529ec CS |
445 | } |
446 | ||
34dc7c2f | 447 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); |
34dc7c2f | 448 | |
d96eb2b1 DM |
449 | if (spa->spa_comment != NULL) { |
450 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
451 | 0, ZPROP_SRC_LOCAL); | |
452 | } | |
453 | ||
658fb802 CB |
454 | if (spa->spa_compatibility != NULL) { |
455 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMPATIBILITY, | |
456 | spa->spa_compatibility, 0, ZPROP_SRC_LOCAL); | |
457 | } | |
458 | ||
34dc7c2f BB |
459 | if (spa->spa_root != NULL) |
460 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
461 | 0, ZPROP_SRC_LOCAL); | |
462 | ||
f1512ee6 MA |
463 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { |
464 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
465 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
466 | } else { | |
467 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
468 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
469 | } | |
470 | ||
50c957f7 NB |
471 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { |
472 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
473 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
474 | } else { | |
475 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
476 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
477 | } | |
478 | ||
b128c09f BB |
479 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { |
480 | if (dp->scd_path == NULL) { | |
34dc7c2f | 481 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f BB |
482 | "none", 0, ZPROP_SRC_LOCAL); |
483 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
34dc7c2f | 484 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f | 485 | dp->scd_path, 0, ZPROP_SRC_LOCAL); |
34dc7c2f BB |
486 | } |
487 | } | |
488 | } | |
489 | ||
490 | /* | |
491 | * Get zpool property values. | |
492 | */ | |
493 | int | |
494 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
495 | { | |
428870ff | 496 | objset_t *mos = spa->spa_meta_objset; |
34dc7c2f BB |
497 | zap_cursor_t zc; |
498 | zap_attribute_t za; | |
1743c737 | 499 | dsl_pool_t *dp; |
34dc7c2f BB |
500 | int err; |
501 | ||
79c76d5b | 502 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); |
c28b2279 | 503 | if (err) |
d1d7e268 | 504 | return (err); |
34dc7c2f | 505 | |
1743c737 AM |
506 | dp = spa_get_dsl(spa); |
507 | dsl_pool_config_enter(dp, FTAG); | |
b128c09f BB |
508 | mutex_enter(&spa->spa_props_lock); |
509 | ||
34dc7c2f BB |
510 | /* |
511 | * Get properties from the spa config. | |
512 | */ | |
513 | spa_prop_get_config(spa, nvp); | |
514 | ||
34dc7c2f | 515 | /* If no pool property object, no more prop to get. */ |
1743c737 | 516 | if (mos == NULL || spa->spa_pool_props_object == 0) |
c28b2279 | 517 | goto out; |
34dc7c2f BB |
518 | |
519 | /* | |
520 | * Get properties from the MOS pool property object. | |
521 | */ | |
522 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
523 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
524 | zap_cursor_advance(&zc)) { | |
525 | uint64_t intval = 0; | |
526 | char *strval = NULL; | |
527 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
528 | zpool_prop_t prop; | |
529 | ||
8eae2d21 AJ |
530 | if ((prop = zpool_name_to_prop(za.za_name)) == |
531 | ZPOOL_PROP_INVAL && !zfs_prop_user(za.za_name)) | |
34dc7c2f BB |
532 | continue; |
533 | ||
534 | switch (za.za_integer_length) { | |
535 | case 8: | |
536 | /* integer property */ | |
537 | if (za.za_first_integer != | |
538 | zpool_prop_default_numeric(prop)) | |
539 | src = ZPROP_SRC_LOCAL; | |
540 | ||
541 | if (prop == ZPOOL_PROP_BOOTFS) { | |
34dc7c2f BB |
542 | dsl_dataset_t *ds = NULL; |
543 | ||
619f0976 GW |
544 | err = dsl_dataset_hold_obj(dp, |
545 | za.za_first_integer, FTAG, &ds); | |
1743c737 | 546 | if (err != 0) |
34dc7c2f | 547 | break; |
34dc7c2f | 548 | |
eca7b760 | 549 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, |
79c76d5b | 550 | KM_SLEEP); |
34dc7c2f | 551 | dsl_dataset_name(ds, strval); |
b128c09f | 552 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
553 | } else { |
554 | strval = NULL; | |
555 | intval = za.za_first_integer; | |
556 | } | |
557 | ||
558 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
559 | ||
560 | if (strval != NULL) | |
eca7b760 | 561 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
562 | |
563 | break; | |
564 | ||
565 | case 1: | |
566 | /* string property */ | |
79c76d5b | 567 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); |
34dc7c2f BB |
568 | err = zap_lookup(mos, spa->spa_pool_props_object, |
569 | za.za_name, 1, za.za_num_integers, strval); | |
570 | if (err) { | |
571 | kmem_free(strval, za.za_num_integers); | |
572 | break; | |
573 | } | |
8eae2d21 AJ |
574 | if (prop != ZPOOL_PROP_INVAL) { |
575 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
576 | } else { | |
577 | src = ZPROP_SRC_LOCAL; | |
578 | spa_prop_add_user(*nvp, za.za_name, strval, | |
579 | src); | |
580 | } | |
34dc7c2f BB |
581 | kmem_free(strval, za.za_num_integers); |
582 | break; | |
583 | ||
584 | default: | |
585 | break; | |
586 | } | |
587 | } | |
588 | zap_cursor_fini(&zc); | |
34dc7c2f | 589 | out: |
1743c737 AM |
590 | mutex_exit(&spa->spa_props_lock); |
591 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
592 | if (err && err != ENOENT) { |
593 | nvlist_free(*nvp); | |
594 | *nvp = NULL; | |
595 | return (err); | |
596 | } | |
597 | ||
598 | return (0); | |
599 | } | |
600 | ||
601 | /* | |
602 | * Validate the given pool properties nvlist and modify the list | |
603 | * for the property values to be set. | |
604 | */ | |
605 | static int | |
606 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
607 | { | |
608 | nvpair_t *elem; | |
609 | int error = 0, reset_bootfs = 0; | |
d4ed6673 | 610 | uint64_t objnum = 0; |
9ae529ec | 611 | boolean_t has_feature = B_FALSE; |
34dc7c2f BB |
612 | |
613 | elem = NULL; | |
614 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
34dc7c2f | 615 | uint64_t intval; |
d1807f16 | 616 | const char *strval, *slash, *check, *fname; |
9ae529ec CS |
617 | const char *propname = nvpair_name(elem); |
618 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
619 | ||
31864e3d BB |
620 | switch (prop) { |
621 | case ZPOOL_PROP_INVAL: | |
9ae529ec CS |
622 | /* |
623 | * Sanitize the input. | |
624 | */ | |
8eae2d21 AJ |
625 | if (zfs_prop_user(propname)) { |
626 | if (strlen(propname) >= ZAP_MAXNAMELEN) { | |
627 | error = SET_ERROR(ENAMETOOLONG); | |
628 | break; | |
629 | } | |
9ae529ec | 630 | |
8eae2d21 AJ |
631 | if (strlen(fnvpair_value_string(elem)) >= |
632 | ZAP_MAXVALUELEN) { | |
633 | error = SET_ERROR(E2BIG); | |
634 | break; | |
635 | } | |
636 | } else if (zpool_prop_feature(propname)) { | |
637 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
638 | error = SET_ERROR(EINVAL); | |
639 | break; | |
640 | } | |
34dc7c2f | 641 | |
8eae2d21 AJ |
642 | if (nvpair_value_uint64(elem, &intval) != 0) { |
643 | error = SET_ERROR(EINVAL); | |
644 | break; | |
645 | } | |
646 | ||
647 | if (intval != 0) { | |
648 | error = SET_ERROR(EINVAL); | |
649 | break; | |
650 | } | |
651 | ||
652 | fname = strchr(propname, '@') + 1; | |
653 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
654 | error = SET_ERROR(EINVAL); | |
655 | break; | |
656 | } | |
34dc7c2f | 657 | |
8eae2d21 AJ |
658 | has_feature = B_TRUE; |
659 | } else { | |
2e528b49 | 660 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
661 | break; |
662 | } | |
9ae529ec | 663 | break; |
34dc7c2f | 664 | |
34dc7c2f BB |
665 | case ZPOOL_PROP_VERSION: |
666 | error = nvpair_value_uint64(elem, &intval); | |
667 | if (!error && | |
9ae529ec CS |
668 | (intval < spa_version(spa) || |
669 | intval > SPA_VERSION_BEFORE_FEATURES || | |
670 | has_feature)) | |
2e528b49 | 671 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
672 | break; |
673 | ||
674 | case ZPOOL_PROP_DELEGATION: | |
675 | case ZPOOL_PROP_AUTOREPLACE: | |
b128c09f | 676 | case ZPOOL_PROP_LISTSNAPS: |
9babb374 | 677 | case ZPOOL_PROP_AUTOEXPAND: |
1b939560 | 678 | case ZPOOL_PROP_AUTOTRIM: |
34dc7c2f BB |
679 | error = nvpair_value_uint64(elem, &intval); |
680 | if (!error && intval > 1) | |
2e528b49 | 681 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
682 | break; |
683 | ||
379ca9cf OF |
684 | case ZPOOL_PROP_MULTIHOST: |
685 | error = nvpair_value_uint64(elem, &intval); | |
686 | if (!error && intval > 1) | |
687 | error = SET_ERROR(EINVAL); | |
688 | ||
25f06d67 BB |
689 | if (!error) { |
690 | uint32_t hostid = zone_get_hostid(NULL); | |
691 | if (hostid) | |
692 | spa->spa_hostid = hostid; | |
693 | else | |
694 | error = SET_ERROR(ENOTSUP); | |
695 | } | |
379ca9cf OF |
696 | |
697 | break; | |
698 | ||
34dc7c2f | 699 | case ZPOOL_PROP_BOOTFS: |
9babb374 BB |
700 | /* |
701 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
702 | * or the pool is still being created (version == 0), | |
703 | * the bootfs property cannot be set. | |
704 | */ | |
34dc7c2f | 705 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { |
2e528b49 | 706 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
707 | break; |
708 | } | |
709 | ||
710 | /* | |
b128c09f | 711 | * Make sure the vdev config is bootable |
34dc7c2f | 712 | */ |
b128c09f | 713 | if (!vdev_is_bootable(spa->spa_root_vdev)) { |
2e528b49 | 714 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
715 | break; |
716 | } | |
717 | ||
718 | reset_bootfs = 1; | |
719 | ||
720 | error = nvpair_value_string(elem, &strval); | |
721 | ||
722 | if (!error) { | |
9ae529ec | 723 | objset_t *os; |
b128c09f | 724 | |
34dc7c2f BB |
725 | if (strval == NULL || strval[0] == '\0') { |
726 | objnum = zpool_prop_default_numeric( | |
727 | ZPOOL_PROP_BOOTFS); | |
728 | break; | |
729 | } | |
730 | ||
d1d7e268 | 731 | error = dmu_objset_hold(strval, FTAG, &os); |
619f0976 | 732 | if (error != 0) |
34dc7c2f | 733 | break; |
b128c09f | 734 | |
eaa25f1a | 735 | /* Must be ZPL. */ |
428870ff | 736 | if (dmu_objset_type(os) != DMU_OST_ZFS) { |
2e528b49 | 737 | error = SET_ERROR(ENOTSUP); |
b128c09f BB |
738 | } else { |
739 | objnum = dmu_objset_id(os); | |
740 | } | |
428870ff | 741 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
742 | } |
743 | break; | |
b128c09f | 744 | |
34dc7c2f BB |
745 | case ZPOOL_PROP_FAILUREMODE: |
746 | error = nvpair_value_uint64(elem, &intval); | |
3bfd95d5 | 747 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) |
2e528b49 | 748 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
749 | |
750 | /* | |
751 | * This is a special case which only occurs when | |
752 | * the pool has completely failed. This allows | |
753 | * the user to change the in-core failmode property | |
754 | * without syncing it out to disk (I/Os might | |
755 | * currently be blocked). We do this by returning | |
756 | * EIO to the caller (spa_prop_set) to trick it | |
757 | * into thinking we encountered a property validation | |
758 | * error. | |
759 | */ | |
b128c09f | 760 | if (!error && spa_suspended(spa)) { |
34dc7c2f | 761 | spa->spa_failmode = intval; |
2e528b49 | 762 | error = SET_ERROR(EIO); |
34dc7c2f BB |
763 | } |
764 | break; | |
765 | ||
766 | case ZPOOL_PROP_CACHEFILE: | |
767 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
768 | break; | |
769 | ||
770 | if (strval[0] == '\0') | |
771 | break; | |
772 | ||
773 | if (strcmp(strval, "none") == 0) | |
774 | break; | |
775 | ||
776 | if (strval[0] != '/') { | |
2e528b49 | 777 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
778 | break; |
779 | } | |
780 | ||
781 | slash = strrchr(strval, '/'); | |
782 | ASSERT(slash != NULL); | |
783 | ||
784 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
785 | strcmp(slash, "/..") == 0) | |
2e528b49 | 786 | error = SET_ERROR(EINVAL); |
34dc7c2f | 787 | break; |
428870ff | 788 | |
d96eb2b1 DM |
789 | case ZPOOL_PROP_COMMENT: |
790 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
791 | break; | |
792 | for (check = strval; *check != '\0'; check++) { | |
793 | if (!isprint(*check)) { | |
2e528b49 | 794 | error = SET_ERROR(EINVAL); |
d96eb2b1 DM |
795 | break; |
796 | } | |
d96eb2b1 DM |
797 | } |
798 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
2e528b49 | 799 | error = SET_ERROR(E2BIG); |
d96eb2b1 DM |
800 | break; |
801 | ||
e75c13c3 BB |
802 | default: |
803 | break; | |
34dc7c2f BB |
804 | } |
805 | ||
806 | if (error) | |
807 | break; | |
808 | } | |
809 | ||
050d720c MA |
810 | (void) nvlist_remove_all(props, |
811 | zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO)); | |
812 | ||
34dc7c2f BB |
813 | if (!error && reset_bootfs) { |
814 | error = nvlist_remove(props, | |
815 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
816 | ||
817 | if (!error) { | |
818 | error = nvlist_add_uint64(props, | |
819 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
820 | } | |
821 | } | |
822 | ||
823 | return (error); | |
824 | } | |
825 | ||
d164b209 BB |
826 | void |
827 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
828 | { | |
d1807f16 | 829 | const char *cachefile; |
d164b209 BB |
830 | spa_config_dirent_t *dp; |
831 | ||
832 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
833 | &cachefile) != 0) | |
834 | return; | |
835 | ||
836 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
79c76d5b | 837 | KM_SLEEP); |
d164b209 BB |
838 | |
839 | if (cachefile[0] == '\0') | |
840 | dp->scd_path = spa_strdup(spa_config_path); | |
841 | else if (strcmp(cachefile, "none") == 0) | |
842 | dp->scd_path = NULL; | |
843 | else | |
844 | dp->scd_path = spa_strdup(cachefile); | |
845 | ||
846 | list_insert_head(&spa->spa_config_list, dp); | |
847 | if (need_sync) | |
848 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
849 | } | |
850 | ||
34dc7c2f BB |
851 | int |
852 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
853 | { | |
854 | int error; | |
9ae529ec | 855 | nvpair_t *elem = NULL; |
d164b209 | 856 | boolean_t need_sync = B_FALSE; |
34dc7c2f BB |
857 | |
858 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
859 | return (error); | |
860 | ||
d164b209 | 861 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { |
9ae529ec | 862 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); |
d164b209 | 863 | |
572e2857 BB |
864 | if (prop == ZPOOL_PROP_CACHEFILE || |
865 | prop == ZPOOL_PROP_ALTROOT || | |
866 | prop == ZPOOL_PROP_READONLY) | |
d164b209 BB |
867 | continue; |
868 | ||
8eae2d21 AJ |
869 | if (prop == ZPOOL_PROP_INVAL && |
870 | zfs_prop_user(nvpair_name(elem))) { | |
871 | need_sync = B_TRUE; | |
872 | break; | |
873 | } | |
874 | ||
31864e3d | 875 | if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { |
2a673e76 | 876 | uint64_t ver = 0; |
9ae529ec CS |
877 | |
878 | if (prop == ZPOOL_PROP_VERSION) { | |
879 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
880 | } else { | |
881 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
882 | ver = SPA_VERSION_FEATURES; | |
883 | need_sync = B_TRUE; | |
884 | } | |
885 | ||
886 | /* Save time if the version is already set. */ | |
887 | if (ver == spa_version(spa)) | |
888 | continue; | |
889 | ||
890 | /* | |
891 | * In addition to the pool directory object, we might | |
892 | * create the pool properties object, the features for | |
893 | * read object, the features for write object, or the | |
894 | * feature descriptions object. | |
895 | */ | |
13fe0198 | 896 | error = dsl_sync_task(spa->spa_name, NULL, |
3d45fdd6 MA |
897 | spa_sync_version, &ver, |
898 | 6, ZFS_SPACE_CHECK_RESERVED); | |
9ae529ec CS |
899 | if (error) |
900 | return (error); | |
901 | continue; | |
902 | } | |
903 | ||
d164b209 BB |
904 | need_sync = B_TRUE; |
905 | break; | |
906 | } | |
907 | ||
9ae529ec | 908 | if (need_sync) { |
13fe0198 | 909 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, |
3d45fdd6 | 910 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); |
9ae529ec CS |
911 | } |
912 | ||
913 | return (0); | |
34dc7c2f BB |
914 | } |
915 | ||
916 | /* | |
917 | * If the bootfs property value is dsobj, clear it. | |
918 | */ | |
919 | void | |
920 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
921 | { | |
922 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
923 | VERIFY(zap_remove(spa->spa_meta_objset, | |
924 | spa->spa_pool_props_object, | |
925 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
926 | spa->spa_bootfs = 0; | |
927 | } | |
928 | } | |
929 | ||
3bc7e0fb | 930 | static int |
13fe0198 | 931 | spa_change_guid_check(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 932 | { |
2a8ba608 | 933 | uint64_t *newguid __maybe_unused = arg; |
13fe0198 | 934 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; |
3bc7e0fb GW |
935 | vdev_t *rvd = spa->spa_root_vdev; |
936 | uint64_t vdev_state; | |
3bc7e0fb | 937 | |
d2734cce SD |
938 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { |
939 | int error = (spa_has_checkpoint(spa)) ? | |
940 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
941 | return (SET_ERROR(error)); | |
942 | } | |
943 | ||
3bc7e0fb GW |
944 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
945 | vdev_state = rvd->vdev_state; | |
946 | spa_config_exit(spa, SCL_STATE, FTAG); | |
947 | ||
948 | if (vdev_state != VDEV_STATE_HEALTHY) | |
2e528b49 | 949 | return (SET_ERROR(ENXIO)); |
3bc7e0fb GW |
950 | |
951 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
952 | ||
953 | return (0); | |
954 | } | |
955 | ||
956 | static void | |
13fe0198 | 957 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 958 | { |
13fe0198 MA |
959 | uint64_t *newguid = arg; |
960 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
3bc7e0fb GW |
961 | uint64_t oldguid; |
962 | vdev_t *rvd = spa->spa_root_vdev; | |
963 | ||
964 | oldguid = spa_guid(spa); | |
965 | ||
966 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
967 | rvd->vdev_guid = *newguid; | |
968 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
969 | vdev_config_dirty(rvd); | |
970 | spa_config_exit(spa, SCL_STATE, FTAG); | |
971 | ||
6f1ffb06 | 972 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", |
74756182 | 973 | (u_longlong_t)oldguid, (u_longlong_t)*newguid); |
3bc7e0fb GW |
974 | } |
975 | ||
3541dc6d GA |
976 | /* |
977 | * Change the GUID for the pool. This is done so that we can later | |
978 | * re-import a pool built from a clone of our own vdevs. We will modify | |
979 | * the root vdev's guid, our own pool guid, and then mark all of our | |
980 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
981 | * online when we do this, or else any vdevs that weren't present | |
982 | * would be orphaned from our pool. We are also going to issue a | |
983 | * sysevent to update any watchers. | |
984 | */ | |
985 | int | |
986 | spa_change_guid(spa_t *spa) | |
987 | { | |
3bc7e0fb GW |
988 | int error; |
989 | uint64_t guid; | |
3541dc6d | 990 | |
621dd7bb | 991 | mutex_enter(&spa->spa_vdev_top_lock); |
3bc7e0fb GW |
992 | mutex_enter(&spa_namespace_lock); |
993 | guid = spa_generate_guid(NULL); | |
3541dc6d | 994 | |
13fe0198 | 995 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, |
3d45fdd6 | 996 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); |
3541dc6d | 997 | |
3bc7e0fb | 998 | if (error == 0) { |
55c12724 AH |
999 | /* |
1000 | * Clear the kobj flag from all the vdevs to allow | |
1001 | * vdev_cache_process_kobj_evt() to post events to all the | |
1002 | * vdevs since GUID is updated. | |
1003 | */ | |
1004 | vdev_clear_kobj_evt(spa->spa_root_vdev); | |
1005 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
1006 | vdev_clear_kobj_evt(spa->spa_l2cache.sav_vdevs[i]); | |
1007 | ||
1008 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); | |
12fa0466 | 1009 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); |
3bc7e0fb | 1010 | } |
3541dc6d | 1011 | |
3bc7e0fb | 1012 | mutex_exit(&spa_namespace_lock); |
621dd7bb | 1013 | mutex_exit(&spa->spa_vdev_top_lock); |
3541dc6d | 1014 | |
3bc7e0fb | 1015 | return (error); |
3541dc6d GA |
1016 | } |
1017 | ||
34dc7c2f BB |
1018 | /* |
1019 | * ========================================================================== | |
1020 | * SPA state manipulation (open/create/destroy/import/export) | |
1021 | * ========================================================================== | |
1022 | */ | |
1023 | ||
1024 | static int | |
1025 | spa_error_entry_compare(const void *a, const void *b) | |
1026 | { | |
ee36c709 GN |
1027 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; |
1028 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
34dc7c2f BB |
1029 | int ret; |
1030 | ||
ee36c709 | 1031 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, |
5dbd68a3 | 1032 | sizeof (zbookmark_phys_t)); |
34dc7c2f | 1033 | |
ca577779 | 1034 | return (TREE_ISIGN(ret)); |
34dc7c2f BB |
1035 | } |
1036 | ||
1037 | /* | |
1038 | * Utility function which retrieves copies of the current logs and | |
1039 | * re-initializes them in the process. | |
1040 | */ | |
1041 | void | |
1042 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
1043 | { | |
1044 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
1045 | ||
861166b0 AZ |
1046 | memcpy(last, &spa->spa_errlist_last, sizeof (avl_tree_t)); |
1047 | memcpy(scrub, &spa->spa_errlist_scrub, sizeof (avl_tree_t)); | |
34dc7c2f BB |
1048 | |
1049 | avl_create(&spa->spa_errlist_scrub, | |
1050 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1051 | offsetof(spa_error_entry_t, se_avl)); | |
1052 | avl_create(&spa->spa_errlist_last, | |
1053 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1054 | offsetof(spa_error_entry_t, se_avl)); | |
1055 | } | |
1056 | ||
7ef5e54e AL |
1057 | static void |
1058 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
34dc7c2f | 1059 | { |
7ef5e54e AL |
1060 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; |
1061 | enum zti_modes mode = ztip->zti_mode; | |
1062 | uint_t value = ztip->zti_value; | |
1063 | uint_t count = ztip->zti_count; | |
1064 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7457b024 | 1065 | uint_t cpus, flags = TASKQ_DYNAMIC; |
34dc7c2f | 1066 | |
e8b96c60 MA |
1067 | switch (mode) { |
1068 | case ZTI_MODE_FIXED: | |
7457b024 | 1069 | ASSERT3U(value, >, 0); |
e8b96c60 | 1070 | break; |
7ef5e54e | 1071 | |
3bd4df38 EN |
1072 | case ZTI_MODE_SYNC: |
1073 | ||
1074 | /* | |
645b8330 AM |
1075 | * Create one wr_iss taskq for every 'zio_taskq_write_tpq' CPUs, |
1076 | * not to exceed the number of spa allocators, and align to it. | |
3bd4df38 | 1077 | */ |
645b8330 AM |
1078 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); |
1079 | count = MAX(1, cpus / MAX(1, zio_taskq_write_tpq)); | |
3bd4df38 EN |
1080 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); |
1081 | count = MIN(count, spa->spa_alloc_count); | |
645b8330 AM |
1082 | while (spa->spa_alloc_count % count != 0 && |
1083 | spa->spa_alloc_count < count * 2) | |
1084 | count--; | |
3bd4df38 EN |
1085 | |
1086 | /* | |
1087 | * zio_taskq_batch_pct is unbounded and may exceed 100%, but no | |
1088 | * single taskq may have more threads than 100% of online cpus. | |
1089 | */ | |
1090 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1091 | value = MIN(value, 100); | |
e8b96c60 | 1092 | flags |= TASKQ_THREADS_CPU_PCT; |
e8b96c60 | 1093 | break; |
7ef5e54e | 1094 | |
7457b024 AM |
1095 | case ZTI_MODE_SCALE: |
1096 | flags |= TASKQ_THREADS_CPU_PCT; | |
1097 | /* | |
1098 | * We want more taskqs to reduce lock contention, but we want | |
1099 | * less for better request ordering and CPU utilization. | |
1100 | */ | |
1101 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); | |
1102 | if (zio_taskq_batch_tpq > 0) { | |
1103 | count = MAX(1, (cpus + zio_taskq_batch_tpq / 2) / | |
1104 | zio_taskq_batch_tpq); | |
1105 | } else { | |
1106 | /* | |
1107 | * Prefer 6 threads per taskq, but no more taskqs | |
1108 | * than threads in them on large systems. For 80%: | |
1109 | * | |
1110 | * taskq taskq total | |
1111 | * cpus taskqs percent threads threads | |
1112 | * ------- ------- ------- ------- ------- | |
1113 | * 1 1 80% 1 1 | |
1114 | * 2 1 80% 1 1 | |
1115 | * 4 1 80% 3 3 | |
1116 | * 8 2 40% 3 6 | |
1117 | * 16 3 27% 4 12 | |
1118 | * 32 5 16% 5 25 | |
1119 | * 64 7 11% 7 49 | |
1120 | * 128 10 8% 10 100 | |
1121 | * 256 14 6% 15 210 | |
1122 | */ | |
1123 | count = 1 + cpus / 6; | |
1124 | while (count * count > cpus) | |
1125 | count--; | |
1126 | } | |
1127 | /* Limit each taskq within 100% to not trigger assertion. */ | |
1128 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1129 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1130 | break; | |
1131 | ||
1132 | case ZTI_MODE_NULL: | |
1133 | tqs->stqs_count = 0; | |
1134 | tqs->stqs_taskq = NULL; | |
1135 | return; | |
1136 | ||
e8b96c60 MA |
1137 | default: |
1138 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
3bd4df38 | 1139 | "spa_taskqs_init()", |
e8b96c60 MA |
1140 | zio_type_name[t], zio_taskq_types[q], mode, value); |
1141 | break; | |
1142 | } | |
7ef5e54e | 1143 | |
7457b024 AM |
1144 | ASSERT3U(count, >, 0); |
1145 | tqs->stqs_count = count; | |
1146 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
1147 | ||
1c27024e | 1148 | for (uint_t i = 0; i < count; i++) { |
e8b96c60 | 1149 | taskq_t *tq; |
af430294 | 1150 | char name[32]; |
7ef5e54e | 1151 | |
7457b024 AM |
1152 | if (count > 1) |
1153 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
1154 | zio_type_name[t], zio_taskq_types[q], i); | |
1155 | else | |
1156 | (void) snprintf(name, sizeof (name), "%s_%s", | |
1157 | zio_type_name[t], zio_taskq_types[q]); | |
7ef5e54e | 1158 | |
b5e60918 | 1159 | #ifdef HAVE_SYSDC |
7ef5e54e | 1160 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { |
18168da7 | 1161 | (void) zio_taskq_basedc; |
7ef5e54e AL |
1162 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, |
1163 | spa->spa_proc, zio_taskq_basedc, flags); | |
1164 | } else { | |
b5e60918 | 1165 | #endif |
e8b96c60 MA |
1166 | pri_t pri = maxclsyspri; |
1167 | /* | |
1168 | * The write issue taskq can be extremely CPU | |
1229323d | 1169 | * intensive. Run it at slightly less important |
7432d297 MM |
1170 | * priority than the other taskqs. |
1171 | * | |
1172 | * Under Linux and FreeBSD this means incrementing | |
1173 | * the priority value as opposed to platforms like | |
1174 | * illumos where it should be decremented. | |
1175 | * | |
1176 | * On FreeBSD, if priorities divided by four (RQ_PPQ) | |
1177 | * are equal then a difference between them is | |
1178 | * insignificant. | |
e8b96c60 | 1179 | */ |
7432d297 MM |
1180 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) { |
1181 | #if defined(__linux__) | |
1229323d | 1182 | pri++; |
7432d297 MM |
1183 | #elif defined(__FreeBSD__) |
1184 | pri += 4; | |
1185 | #else | |
1186 | #error "unknown OS" | |
1187 | #endif | |
1188 | } | |
e8b96c60 | 1189 | tq = taskq_create_proc(name, value, pri, 50, |
7ef5e54e | 1190 | INT_MAX, spa->spa_proc, flags); |
b5e60918 | 1191 | #ifdef HAVE_SYSDC |
7ef5e54e | 1192 | } |
b5e60918 | 1193 | #endif |
7ef5e54e AL |
1194 | |
1195 | tqs->stqs_taskq[i] = tq; | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | static void | |
1200 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
1201 | { | |
1202 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7ef5e54e AL |
1203 | |
1204 | if (tqs->stqs_taskq == NULL) { | |
1205 | ASSERT3U(tqs->stqs_count, ==, 0); | |
1206 | return; | |
1207 | } | |
1208 | ||
1c27024e | 1209 | for (uint_t i = 0; i < tqs->stqs_count; i++) { |
7ef5e54e AL |
1210 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); |
1211 | taskq_destroy(tqs->stqs_taskq[i]); | |
428870ff | 1212 | } |
34dc7c2f | 1213 | |
7ef5e54e AL |
1214 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); |
1215 | tqs->stqs_taskq = NULL; | |
1216 | } | |
34dc7c2f | 1217 | |
6930ecbb RN |
1218 | #ifdef _KERNEL |
1219 | /* | |
1220 | * The READ and WRITE rows of zio_taskqs are configurable at module load time | |
1221 | * by setting zio_taskq_read or zio_taskq_write. | |
1222 | * | |
1223 | * Example (the defaults for READ and WRITE) | |
1224 | * zio_taskq_read='fixed,1,8 null scale null' | |
04bae5ec | 1225 | * zio_taskq_write='sync null scale null' |
6930ecbb RN |
1226 | * |
1227 | * Each sets the entire row at a time. | |
1228 | * | |
1229 | * 'fixed' is parameterised: fixed,Q,T where Q is number of taskqs, T is number | |
1230 | * of threads per taskq. | |
1231 | * | |
1232 | * 'null' can only be set on the high-priority queues (queue selection for | |
1233 | * high-priority queues will fall back to the regular queue if the high-pri | |
1234 | * is NULL. | |
1235 | */ | |
1236 | static const char *const modes[ZTI_NMODES] = { | |
1237 | "fixed", "scale", "sync", "null" | |
1238 | }; | |
1239 | ||
1240 | /* Parse the incoming config string. Modifies cfg */ | |
1241 | static int | |
1242 | spa_taskq_param_set(zio_type_t t, char *cfg) | |
1243 | { | |
1244 | int err = 0; | |
1245 | ||
1246 | zio_taskq_info_t row[ZIO_TASKQ_TYPES] = {{0}}; | |
1247 | ||
1248 | char *next = cfg, *tok, *c; | |
1249 | ||
1250 | /* | |
1251 | * Parse out each element from the string and fill `row`. The entire | |
1252 | * row has to be set at once, so any errors are flagged by just | |
1253 | * breaking out of this loop early. | |
1254 | */ | |
1255 | uint_t q; | |
1256 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1257 | /* `next` is the start of the config */ | |
1258 | if (next == NULL) | |
1259 | break; | |
1260 | ||
1261 | /* Eat up leading space */ | |
1262 | while (isspace(*next)) | |
1263 | next++; | |
1264 | if (*next == '\0') | |
1265 | break; | |
1266 | ||
1267 | /* Mode ends at space or end of string */ | |
1268 | tok = next; | |
1269 | next = strchr(tok, ' '); | |
1270 | if (next != NULL) *next++ = '\0'; | |
1271 | ||
1272 | /* Parameters start after a comma */ | |
1273 | c = strchr(tok, ','); | |
1274 | if (c != NULL) *c++ = '\0'; | |
1275 | ||
1276 | /* Match mode string */ | |
1277 | uint_t mode; | |
1278 | for (mode = 0; mode < ZTI_NMODES; mode++) | |
1279 | if (strcmp(tok, modes[mode]) == 0) | |
1280 | break; | |
1281 | if (mode == ZTI_NMODES) | |
1282 | break; | |
1283 | ||
1284 | /* Invalid canary */ | |
1285 | row[q].zti_mode = ZTI_NMODES; | |
1286 | ||
1287 | /* Per-mode setup */ | |
1288 | switch (mode) { | |
1289 | ||
1290 | /* | |
1291 | * FIXED is parameterised: number of queues, and number of | |
1292 | * threads per queue. | |
1293 | */ | |
1294 | case ZTI_MODE_FIXED: { | |
1295 | /* No parameters? */ | |
1296 | if (c == NULL || *c == '\0') | |
1297 | break; | |
1298 | ||
1299 | /* Find next parameter */ | |
1300 | tok = c; | |
1301 | c = strchr(tok, ','); | |
1302 | if (c == NULL) | |
1303 | break; | |
1304 | ||
1305 | /* Take digits and convert */ | |
1306 | unsigned long long nq; | |
1307 | if (!(isdigit(*tok))) | |
1308 | break; | |
1309 | err = ddi_strtoull(tok, &tok, 10, &nq); | |
1310 | /* Must succeed and also end at the next param sep */ | |
1311 | if (err != 0 || tok != c) | |
1312 | break; | |
1313 | ||
1314 | /* Move past the comma */ | |
1315 | tok++; | |
1316 | /* Need another number */ | |
1317 | if (!(isdigit(*tok))) | |
1318 | break; | |
1319 | /* Remember start to make sure we moved */ | |
1320 | c = tok; | |
1321 | ||
1322 | /* Take digits */ | |
1323 | unsigned long long ntpq; | |
1324 | err = ddi_strtoull(tok, &tok, 10, &ntpq); | |
1325 | /* Must succeed, and moved forward */ | |
1326 | if (err != 0 || tok == c || *tok != '\0') | |
1327 | break; | |
1328 | ||
1329 | /* | |
1330 | * sanity; zero queues/threads make no sense, and | |
1331 | * 16K is almost certainly more than anyone will ever | |
1332 | * need and avoids silly numbers like UINT32_MAX | |
1333 | */ | |
1334 | if (nq == 0 || nq >= 16384 || | |
1335 | ntpq == 0 || ntpq >= 16384) | |
1336 | break; | |
1337 | ||
1338 | const zio_taskq_info_t zti = ZTI_P(ntpq, nq); | |
1339 | row[q] = zti; | |
1340 | break; | |
1341 | } | |
1342 | ||
1343 | case ZTI_MODE_SCALE: { | |
1344 | const zio_taskq_info_t zti = ZTI_SCALE; | |
1345 | row[q] = zti; | |
1346 | break; | |
1347 | } | |
1348 | ||
1349 | case ZTI_MODE_SYNC: { | |
1350 | const zio_taskq_info_t zti = ZTI_SYNC; | |
1351 | row[q] = zti; | |
1352 | break; | |
1353 | } | |
1354 | ||
1355 | case ZTI_MODE_NULL: { | |
1356 | /* | |
1357 | * Can only null the high-priority queues; the general- | |
1358 | * purpose ones have to exist. | |
1359 | */ | |
1360 | if (q != ZIO_TASKQ_ISSUE_HIGH && | |
1361 | q != ZIO_TASKQ_INTERRUPT_HIGH) | |
1362 | break; | |
1363 | ||
1364 | const zio_taskq_info_t zti = ZTI_NULL; | |
1365 | row[q] = zti; | |
1366 | break; | |
1367 | } | |
1368 | ||
1369 | default: | |
1370 | break; | |
1371 | } | |
1372 | ||
1373 | /* Ensure we set a mode */ | |
1374 | if (row[q].zti_mode == ZTI_NMODES) | |
1375 | break; | |
1376 | } | |
1377 | ||
1378 | /* Didn't get a full row, fail */ | |
1379 | if (q < ZIO_TASKQ_TYPES) | |
1380 | return (SET_ERROR(EINVAL)); | |
1381 | ||
1382 | /* Eat trailing space */ | |
1383 | if (next != NULL) | |
1384 | while (isspace(*next)) | |
1385 | next++; | |
1386 | ||
1387 | /* If there's anything left over then fail */ | |
1388 | if (next != NULL && *next != '\0') | |
1389 | return (SET_ERROR(EINVAL)); | |
1390 | ||
1391 | /* Success! Copy it into the real config */ | |
1392 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) | |
1393 | zio_taskqs[t][q] = row[q]; | |
1394 | ||
1395 | return (0); | |
1396 | } | |
1397 | ||
1398 | static int | |
5a703d13 | 1399 | spa_taskq_param_get(zio_type_t t, char *buf, boolean_t add_newline) |
6930ecbb RN |
1400 | { |
1401 | int pos = 0; | |
1402 | ||
1403 | /* Build paramater string from live config */ | |
1404 | const char *sep = ""; | |
1405 | for (uint_t q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1406 | const zio_taskq_info_t *zti = &zio_taskqs[t][q]; | |
1407 | if (zti->zti_mode == ZTI_MODE_FIXED) | |
1408 | pos += sprintf(&buf[pos], "%s%s,%u,%u", sep, | |
1409 | modes[zti->zti_mode], zti->zti_count, | |
1410 | zti->zti_value); | |
1411 | else | |
1412 | pos += sprintf(&buf[pos], "%s%s", sep, | |
1413 | modes[zti->zti_mode]); | |
1414 | sep = " "; | |
1415 | } | |
1416 | ||
5a703d13 MJ |
1417 | if (add_newline) |
1418 | buf[pos++] = '\n'; | |
6930ecbb RN |
1419 | buf[pos] = '\0'; |
1420 | ||
1421 | return (pos); | |
1422 | } | |
1423 | ||
1424 | #ifdef __linux__ | |
1425 | static int | |
1426 | spa_taskq_read_param_set(const char *val, zfs_kernel_param_t *kp) | |
1427 | { | |
1428 | char *cfg = kmem_strdup(val); | |
1429 | int err = spa_taskq_param_set(ZIO_TYPE_READ, cfg); | |
1430 | kmem_free(cfg, strlen(val)+1); | |
1431 | return (-err); | |
1432 | } | |
1433 | static int | |
1434 | spa_taskq_read_param_get(char *buf, zfs_kernel_param_t *kp) | |
1435 | { | |
5a703d13 | 1436 | return (spa_taskq_param_get(ZIO_TYPE_READ, buf, TRUE)); |
6930ecbb RN |
1437 | } |
1438 | ||
1439 | static int | |
1440 | spa_taskq_write_param_set(const char *val, zfs_kernel_param_t *kp) | |
1441 | { | |
1442 | char *cfg = kmem_strdup(val); | |
1443 | int err = spa_taskq_param_set(ZIO_TYPE_WRITE, cfg); | |
1444 | kmem_free(cfg, strlen(val)+1); | |
1445 | return (-err); | |
1446 | } | |
1447 | static int | |
1448 | spa_taskq_write_param_get(char *buf, zfs_kernel_param_t *kp) | |
1449 | { | |
5a703d13 | 1450 | return (spa_taskq_param_get(ZIO_TYPE_WRITE, buf, TRUE)); |
6930ecbb RN |
1451 | } |
1452 | #else | |
6930ecbb RN |
1453 | /* |
1454 | * On FreeBSD load-time parameters can be set up before malloc() is available, | |
1455 | * so we have to do all the parsing work on the stack. | |
1456 | */ | |
1457 | #define SPA_TASKQ_PARAM_MAX (128) | |
1458 | ||
1459 | static int | |
1460 | spa_taskq_read_param(ZFS_MODULE_PARAM_ARGS) | |
1461 | { | |
1462 | char buf[SPA_TASKQ_PARAM_MAX]; | |
3bddc4da | 1463 | int err; |
6930ecbb | 1464 | |
5a703d13 | 1465 | (void) spa_taskq_param_get(ZIO_TYPE_READ, buf, FALSE); |
6930ecbb | 1466 | err = sysctl_handle_string(oidp, buf, sizeof (buf), req); |
3bddc4da | 1467 | if (err || req->newptr == NULL) |
6930ecbb RN |
1468 | return (err); |
1469 | return (spa_taskq_param_set(ZIO_TYPE_READ, buf)); | |
1470 | } | |
1471 | ||
1472 | static int | |
1473 | spa_taskq_write_param(ZFS_MODULE_PARAM_ARGS) | |
1474 | { | |
1475 | char buf[SPA_TASKQ_PARAM_MAX]; | |
3bddc4da | 1476 | int err; |
6930ecbb | 1477 | |
5a703d13 | 1478 | (void) spa_taskq_param_get(ZIO_TYPE_WRITE, buf, FALSE); |
6930ecbb | 1479 | err = sysctl_handle_string(oidp, buf, sizeof (buf), req); |
3bddc4da | 1480 | if (err || req->newptr == NULL) |
6930ecbb RN |
1481 | return (err); |
1482 | return (spa_taskq_param_set(ZIO_TYPE_WRITE, buf)); | |
1483 | } | |
1484 | #endif | |
1485 | #endif /* _KERNEL */ | |
1486 | ||
7ef5e54e AL |
1487 | /* |
1488 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
1489 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
3bd4df38 | 1490 | * on the taskq itself. |
7ef5e54e | 1491 | */ |
3bd4df38 EN |
1492 | static taskq_t * |
1493 | spa_taskq_dispatch_select(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1494 | zio_t *zio) | |
7ef5e54e AL |
1495 | { |
1496 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1497 | taskq_t *tq; | |
1498 | ||
1499 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1500 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1501 | ||
1502 | if (tqs->stqs_count == 1) { | |
1503 | tq = tqs->stqs_taskq[0]; | |
645b8330 AM |
1504 | } else if ((t == ZIO_TYPE_WRITE) && (q == ZIO_TASKQ_ISSUE) && |
1505 | (zio != NULL) && ZIO_HAS_ALLOCATOR(zio)) { | |
1506 | tq = tqs->stqs_taskq[zio->io_allocator % tqs->stqs_count]; | |
7ef5e54e | 1507 | } else { |
c12936b1 | 1508 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
428870ff | 1509 | } |
3bd4df38 EN |
1510 | return (tq); |
1511 | } | |
7ef5e54e | 1512 | |
3bd4df38 EN |
1513 | void |
1514 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1515 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent, | |
1516 | zio_t *zio) | |
1517 | { | |
1518 | taskq_t *tq = spa_taskq_dispatch_select(spa, t, q, zio); | |
7ef5e54e | 1519 | taskq_dispatch_ent(tq, func, arg, flags, ent); |
428870ff BB |
1520 | } |
1521 | ||
044baf00 BB |
1522 | /* |
1523 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
1524 | */ | |
1525 | void | |
1526 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1527 | task_func_t *func, void *arg, uint_t flags) | |
1528 | { | |
3bd4df38 EN |
1529 | taskq_t *tq = spa_taskq_dispatch_select(spa, t, q, NULL); |
1530 | taskqid_t id = taskq_dispatch(tq, func, arg, flags); | |
044baf00 BB |
1531 | if (id) |
1532 | taskq_wait_id(tq, id); | |
1533 | } | |
1534 | ||
428870ff BB |
1535 | static void |
1536 | spa_create_zio_taskqs(spa_t *spa) | |
1537 | { | |
1c27024e DB |
1538 | for (int t = 0; t < ZIO_TYPES; t++) { |
1539 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1540 | spa_taskqs_init(spa, t, q); |
428870ff BB |
1541 | } |
1542 | } | |
1543 | } | |
9babb374 | 1544 | |
7b89a549 | 1545 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) |
428870ff BB |
1546 | static void |
1547 | spa_thread(void *arg) | |
1548 | { | |
93ce2b4c | 1549 | psetid_t zio_taskq_psrset_bind = PS_NONE; |
428870ff | 1550 | callb_cpr_t cprinfo; |
9babb374 | 1551 | |
428870ff BB |
1552 | spa_t *spa = arg; |
1553 | user_t *pu = PTOU(curproc); | |
9babb374 | 1554 | |
428870ff BB |
1555 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, |
1556 | spa->spa_name); | |
9babb374 | 1557 | |
428870ff BB |
1558 | ASSERT(curproc != &p0); |
1559 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1560 | "zpool-%s", spa->spa_name); | |
1561 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1562 | ||
1563 | /* bind this thread to the requested psrset */ | |
1564 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1565 | pool_lock(); | |
1566 | mutex_enter(&cpu_lock); | |
1567 | mutex_enter(&pidlock); | |
1568 | mutex_enter(&curproc->p_lock); | |
1569 | ||
1570 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1571 | 0, NULL, NULL) == 0) { | |
1572 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1573 | } else { | |
1574 | cmn_err(CE_WARN, | |
1575 | "Couldn't bind process for zfs pool \"%s\" to " | |
1576 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1577 | } | |
1578 | ||
1579 | mutex_exit(&curproc->p_lock); | |
1580 | mutex_exit(&pidlock); | |
1581 | mutex_exit(&cpu_lock); | |
1582 | pool_unlock(); | |
1583 | } | |
1584 | ||
b5e60918 | 1585 | #ifdef HAVE_SYSDC |
428870ff BB |
1586 | if (zio_taskq_sysdc) { |
1587 | sysdc_thread_enter(curthread, 100, 0); | |
1588 | } | |
b5e60918 | 1589 | #endif |
428870ff BB |
1590 | |
1591 | spa->spa_proc = curproc; | |
1592 | spa->spa_did = curthread->t_did; | |
1593 | ||
1594 | spa_create_zio_taskqs(spa); | |
1595 | ||
1596 | mutex_enter(&spa->spa_proc_lock); | |
1597 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1598 | ||
1599 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1600 | cv_broadcast(&spa->spa_proc_cv); | |
1601 | ||
1602 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1603 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1604 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1605 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1606 | ||
1607 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1608 | spa->spa_proc_state = SPA_PROC_GONE; | |
1609 | spa->spa_proc = &p0; | |
1610 | cv_broadcast(&spa->spa_proc_cv); | |
1611 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1612 | ||
1613 | mutex_enter(&curproc->p_lock); | |
1614 | lwp_exit(); | |
1615 | } | |
1616 | #endif | |
1617 | ||
95f71c01 EN |
1618 | extern metaslab_ops_t *metaslab_allocator(spa_t *spa); |
1619 | ||
428870ff BB |
1620 | /* |
1621 | * Activate an uninitialized pool. | |
1622 | */ | |
1623 | static void | |
da92d5cb | 1624 | spa_activate(spa_t *spa, spa_mode_t mode) |
428870ff | 1625 | { |
95f71c01 | 1626 | metaslab_ops_t *msp = metaslab_allocator(spa); |
428870ff BB |
1627 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); |
1628 | ||
1629 | spa->spa_state = POOL_STATE_ACTIVE; | |
1630 | spa->spa_mode = mode; | |
e39fe05b | 1631 | spa->spa_read_spacemaps = spa_mode_readable_spacemaps; |
428870ff | 1632 | |
95f71c01 EN |
1633 | spa->spa_normal_class = metaslab_class_create(spa, msp); |
1634 | spa->spa_log_class = metaslab_class_create(spa, msp); | |
1635 | spa->spa_embedded_log_class = metaslab_class_create(spa, msp); | |
1636 | spa->spa_special_class = metaslab_class_create(spa, msp); | |
1637 | spa->spa_dedup_class = metaslab_class_create(spa, msp); | |
428870ff BB |
1638 | |
1639 | /* Try to create a covering process */ | |
1640 | mutex_enter(&spa->spa_proc_lock); | |
1641 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1642 | ASSERT(spa->spa_proc == &p0); | |
1643 | spa->spa_did = 0; | |
1644 | ||
7b89a549 | 1645 | #ifdef HAVE_SPA_THREAD |
428870ff BB |
1646 | /* Only create a process if we're going to be around a while. */ |
1647 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1648 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1649 | NULL, 0) == 0) { | |
1650 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1651 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1652 | cv_wait(&spa->spa_proc_cv, | |
1653 | &spa->spa_proc_lock); | |
9babb374 | 1654 | } |
428870ff BB |
1655 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); |
1656 | ASSERT(spa->spa_proc != &p0); | |
1657 | ASSERT(spa->spa_did != 0); | |
1658 | } else { | |
1659 | #ifdef _KERNEL | |
1660 | cmn_err(CE_WARN, | |
1661 | "Couldn't create process for zfs pool \"%s\"\n", | |
1662 | spa->spa_name); | |
1663 | #endif | |
b128c09f | 1664 | } |
34dc7c2f | 1665 | } |
7b89a549 | 1666 | #endif /* HAVE_SPA_THREAD */ |
428870ff BB |
1667 | mutex_exit(&spa->spa_proc_lock); |
1668 | ||
1669 | /* If we didn't create a process, we need to create our taskqs. */ | |
1670 | if (spa->spa_proc == &p0) { | |
1671 | spa_create_zio_taskqs(spa); | |
1672 | } | |
34dc7c2f | 1673 | |
619f0976 GW |
1674 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1675 | spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, | |
1676 | ZIO_FLAG_CANFAIL); | |
1677 | } | |
a1d477c2 | 1678 | |
b128c09f BB |
1679 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), |
1680 | offsetof(vdev_t, vdev_config_dirty_node)); | |
0c66c32d JG |
1681 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), |
1682 | offsetof(objset_t, os_evicting_node)); | |
b128c09f BB |
1683 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), |
1684 | offsetof(vdev_t, vdev_state_dirty_node)); | |
34dc7c2f | 1685 | |
4747a7d3 | 1686 | txg_list_create(&spa->spa_vdev_txg_list, spa, |
34dc7c2f BB |
1687 | offsetof(struct vdev, vdev_txg_node)); |
1688 | ||
1689 | avl_create(&spa->spa_errlist_scrub, | |
1690 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1691 | offsetof(spa_error_entry_t, se_avl)); | |
1692 | avl_create(&spa->spa_errlist_last, | |
1693 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1694 | offsetof(spa_error_entry_t, se_avl)); | |
e8cf3a4f AP |
1695 | avl_create(&spa->spa_errlist_healed, |
1696 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1697 | offsetof(spa_error_entry_t, se_avl)); | |
a0bd735a | 1698 | |
4759342a JL |
1699 | spa_activate_os(spa); |
1700 | ||
b5256303 TC |
1701 | spa_keystore_init(&spa->spa_keystore); |
1702 | ||
a0bd735a BP |
1703 | /* |
1704 | * This taskq is used to perform zvol-minor-related tasks | |
1705 | * asynchronously. This has several advantages, including easy | |
d0249a4b | 1706 | * resolution of various deadlocks. |
a0bd735a BP |
1707 | * |
1708 | * The taskq must be single threaded to ensure tasks are always | |
1709 | * processed in the order in which they were dispatched. | |
1710 | * | |
1711 | * A taskq per pool allows one to keep the pools independent. | |
1712 | * This way if one pool is suspended, it will not impact another. | |
1713 | * | |
1714 | * The preferred location to dispatch a zvol minor task is a sync | |
1715 | * task. In this context, there is easy access to the spa_t and minimal | |
1716 | * error handling is required because the sync task must succeed. | |
1717 | */ | |
1718 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1719 | 1, INT_MAX, 0); | |
1de321e6 | 1720 | |
342357cd AM |
1721 | /* |
1722 | * The taskq to preload metaslabs. | |
1723 | */ | |
1724 | spa->spa_metaslab_taskq = taskq_create("z_metaslab", | |
1725 | metaslab_preload_pct, maxclsyspri, 1, INT_MAX, | |
1726 | TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
1727 | ||
77d8a0f1 | 1728 | /* |
1729 | * Taskq dedicated to prefetcher threads: this is used to prevent the | |
1730 | * pool traverse code from monopolizing the global (and limited) | |
1731 | * system_taskq by inappropriately scheduling long running tasks on it. | |
1732 | */ | |
60a4c7d2 PD |
1733 | spa->spa_prefetch_taskq = taskq_create("z_prefetch", 100, |
1734 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
77d8a0f1 | 1735 | |
1de321e6 JX |
1736 | /* |
1737 | * The taskq to upgrade datasets in this pool. Currently used by | |
9c5167d1 | 1738 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA. |
1de321e6 | 1739 | */ |
60a4c7d2 PD |
1740 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", 100, |
1741 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
34dc7c2f BB |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * Opposite of spa_activate(). | |
1746 | */ | |
1747 | static void | |
1748 | spa_deactivate(spa_t *spa) | |
1749 | { | |
34dc7c2f BB |
1750 | ASSERT(spa->spa_sync_on == B_FALSE); |
1751 | ASSERT(spa->spa_dsl_pool == NULL); | |
1752 | ASSERT(spa->spa_root_vdev == NULL); | |
9babb374 | 1753 | ASSERT(spa->spa_async_zio_root == NULL); |
34dc7c2f BB |
1754 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); |
1755 | ||
0c66c32d JG |
1756 | spa_evicting_os_wait(spa); |
1757 | ||
a0bd735a BP |
1758 | if (spa->spa_zvol_taskq) { |
1759 | taskq_destroy(spa->spa_zvol_taskq); | |
1760 | spa->spa_zvol_taskq = NULL; | |
1761 | } | |
1762 | ||
342357cd AM |
1763 | if (spa->spa_metaslab_taskq) { |
1764 | taskq_destroy(spa->spa_metaslab_taskq); | |
1765 | spa->spa_metaslab_taskq = NULL; | |
1766 | } | |
1767 | ||
77d8a0f1 | 1768 | if (spa->spa_prefetch_taskq) { |
1769 | taskq_destroy(spa->spa_prefetch_taskq); | |
1770 | spa->spa_prefetch_taskq = NULL; | |
1771 | } | |
1772 | ||
1de321e6 JX |
1773 | if (spa->spa_upgrade_taskq) { |
1774 | taskq_destroy(spa->spa_upgrade_taskq); | |
1775 | spa->spa_upgrade_taskq = NULL; | |
1776 | } | |
1777 | ||
34dc7c2f BB |
1778 | txg_list_destroy(&spa->spa_vdev_txg_list); |
1779 | ||
b128c09f | 1780 | list_destroy(&spa->spa_config_dirty_list); |
0c66c32d | 1781 | list_destroy(&spa->spa_evicting_os_list); |
b128c09f | 1782 | list_destroy(&spa->spa_state_dirty_list); |
34dc7c2f | 1783 | |
57ddcda1 | 1784 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 | 1785 | |
1c27024e DB |
1786 | for (int t = 0; t < ZIO_TYPES; t++) { |
1787 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1788 | spa_taskqs_fini(spa, t, q); |
b128c09f | 1789 | } |
34dc7c2f BB |
1790 | } |
1791 | ||
a1d477c2 MA |
1792 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1793 | ASSERT3P(spa->spa_txg_zio[i], !=, NULL); | |
1794 | VERIFY0(zio_wait(spa->spa_txg_zio[i])); | |
1795 | spa->spa_txg_zio[i] = NULL; | |
1796 | } | |
1797 | ||
34dc7c2f BB |
1798 | metaslab_class_destroy(spa->spa_normal_class); |
1799 | spa->spa_normal_class = NULL; | |
1800 | ||
1801 | metaslab_class_destroy(spa->spa_log_class); | |
1802 | spa->spa_log_class = NULL; | |
1803 | ||
aa755b35 MA |
1804 | metaslab_class_destroy(spa->spa_embedded_log_class); |
1805 | spa->spa_embedded_log_class = NULL; | |
1806 | ||
cc99f275 DB |
1807 | metaslab_class_destroy(spa->spa_special_class); |
1808 | spa->spa_special_class = NULL; | |
1809 | ||
1810 | metaslab_class_destroy(spa->spa_dedup_class); | |
1811 | spa->spa_dedup_class = NULL; | |
1812 | ||
34dc7c2f BB |
1813 | /* |
1814 | * If this was part of an import or the open otherwise failed, we may | |
1815 | * still have errors left in the queues. Empty them just in case. | |
1816 | */ | |
1817 | spa_errlog_drain(spa); | |
34dc7c2f BB |
1818 | avl_destroy(&spa->spa_errlist_scrub); |
1819 | avl_destroy(&spa->spa_errlist_last); | |
e8cf3a4f | 1820 | avl_destroy(&spa->spa_errlist_healed); |
34dc7c2f | 1821 | |
b5256303 TC |
1822 | spa_keystore_fini(&spa->spa_keystore); |
1823 | ||
34dc7c2f | 1824 | spa->spa_state = POOL_STATE_UNINITIALIZED; |
428870ff BB |
1825 | |
1826 | mutex_enter(&spa->spa_proc_lock); | |
1827 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1828 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1829 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1830 | cv_broadcast(&spa->spa_proc_cv); | |
1831 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1832 | ASSERT(spa->spa_proc != &p0); | |
1833 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1834 | } | |
1835 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1836 | spa->spa_proc_state = SPA_PROC_NONE; | |
1837 | } | |
1838 | ASSERT(spa->spa_proc == &p0); | |
1839 | mutex_exit(&spa->spa_proc_lock); | |
1840 | ||
1841 | /* | |
1842 | * We want to make sure spa_thread() has actually exited the ZFS | |
1843 | * module, so that the module can't be unloaded out from underneath | |
1844 | * it. | |
1845 | */ | |
1846 | if (spa->spa_did != 0) { | |
1847 | thread_join(spa->spa_did); | |
1848 | spa->spa_did = 0; | |
1849 | } | |
4759342a JL |
1850 | |
1851 | spa_deactivate_os(spa); | |
1852 | ||
34dc7c2f BB |
1853 | } |
1854 | ||
1855 | /* | |
1856 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1857 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1858 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1859 | * All vdev validation is done by the vdev_alloc() routine. | |
1860 | */ | |
4a22ba5b | 1861 | int |
34dc7c2f BB |
1862 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, |
1863 | uint_t id, int atype) | |
1864 | { | |
1865 | nvlist_t **child; | |
9babb374 | 1866 | uint_t children; |
34dc7c2f BB |
1867 | int error; |
1868 | ||
1869 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1870 | return (error); | |
1871 | ||
1872 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1873 | return (0); | |
1874 | ||
b128c09f BB |
1875 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
1876 | &child, &children); | |
1877 | ||
1878 | if (error == ENOENT) | |
1879 | return (0); | |
1880 | ||
1881 | if (error) { | |
34dc7c2f BB |
1882 | vdev_free(*vdp); |
1883 | *vdp = NULL; | |
2e528b49 | 1884 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
1885 | } |
1886 | ||
1c27024e | 1887 | for (int c = 0; c < children; c++) { |
34dc7c2f BB |
1888 | vdev_t *vd; |
1889 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1890 | atype)) != 0) { | |
1891 | vdev_free(*vdp); | |
1892 | *vdp = NULL; | |
1893 | return (error); | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | ASSERT(*vdp != NULL); | |
1898 | ||
1899 | return (0); | |
1900 | } | |
1901 | ||
93e28d66 SD |
1902 | static boolean_t |
1903 | spa_should_flush_logs_on_unload(spa_t *spa) | |
1904 | { | |
1905 | if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) | |
1906 | return (B_FALSE); | |
1907 | ||
1908 | if (!spa_writeable(spa)) | |
1909 | return (B_FALSE); | |
1910 | ||
1911 | if (!spa->spa_sync_on) | |
1912 | return (B_FALSE); | |
1913 | ||
1914 | if (spa_state(spa) != POOL_STATE_EXPORTED) | |
1915 | return (B_FALSE); | |
1916 | ||
1917 | if (zfs_keep_log_spacemaps_at_export) | |
1918 | return (B_FALSE); | |
1919 | ||
1920 | return (B_TRUE); | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * Opens a transaction that will set the flag that will instruct | |
1925 | * spa_sync to attempt to flush all the metaslabs for that txg. | |
1926 | */ | |
1927 | static void | |
1928 | spa_unload_log_sm_flush_all(spa_t *spa) | |
1929 | { | |
1930 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
1931 | VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); | |
1932 | ||
1933 | ASSERT3U(spa->spa_log_flushall_txg, ==, 0); | |
1934 | spa->spa_log_flushall_txg = dmu_tx_get_txg(tx); | |
1935 | ||
1936 | dmu_tx_commit(tx); | |
1937 | txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg); | |
1938 | } | |
1939 | ||
1940 | static void | |
1941 | spa_unload_log_sm_metadata(spa_t *spa) | |
1942 | { | |
1943 | void *cookie = NULL; | |
1944 | spa_log_sm_t *sls; | |
b3ad3f48 AM |
1945 | log_summary_entry_t *e; |
1946 | ||
93e28d66 SD |
1947 | while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg, |
1948 | &cookie)) != NULL) { | |
1949 | VERIFY0(sls->sls_mscount); | |
1950 | kmem_free(sls, sizeof (spa_log_sm_t)); | |
1951 | } | |
1952 | ||
b3ad3f48 | 1953 | while ((e = list_remove_head(&spa->spa_log_summary)) != NULL) { |
93e28d66 | 1954 | VERIFY0(e->lse_mscount); |
93e28d66 SD |
1955 | kmem_free(e, sizeof (log_summary_entry_t)); |
1956 | } | |
1957 | ||
1958 | spa->spa_unflushed_stats.sus_nblocks = 0; | |
1959 | spa->spa_unflushed_stats.sus_memused = 0; | |
1960 | spa->spa_unflushed_stats.sus_blocklimit = 0; | |
1961 | } | |
1962 | ||
37f03da8 SH |
1963 | static void |
1964 | spa_destroy_aux_threads(spa_t *spa) | |
1965 | { | |
1966 | if (spa->spa_condense_zthr != NULL) { | |
1967 | zthr_destroy(spa->spa_condense_zthr); | |
1968 | spa->spa_condense_zthr = NULL; | |
1969 | } | |
1970 | if (spa->spa_checkpoint_discard_zthr != NULL) { | |
1971 | zthr_destroy(spa->spa_checkpoint_discard_zthr); | |
1972 | spa->spa_checkpoint_discard_zthr = NULL; | |
1973 | } | |
1974 | if (spa->spa_livelist_delete_zthr != NULL) { | |
1975 | zthr_destroy(spa->spa_livelist_delete_zthr); | |
1976 | spa->spa_livelist_delete_zthr = NULL; | |
1977 | } | |
1978 | if (spa->spa_livelist_condense_zthr != NULL) { | |
1979 | zthr_destroy(spa->spa_livelist_condense_zthr); | |
1980 | spa->spa_livelist_condense_zthr = NULL; | |
1981 | } | |
5caeef02 DB |
1982 | if (spa->spa_raidz_expand_zthr != NULL) { |
1983 | zthr_destroy(spa->spa_raidz_expand_zthr); | |
1984 | spa->spa_raidz_expand_zthr = NULL; | |
1985 | } | |
37f03da8 SH |
1986 | } |
1987 | ||
34dc7c2f BB |
1988 | /* |
1989 | * Opposite of spa_load(). | |
1990 | */ | |
1991 | static void | |
1992 | spa_unload(spa_t *spa) | |
1993 | { | |
b128c09f | 1994 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
93e28d66 | 1995 | ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED); |
b128c09f | 1996 | |
ca95f70d | 1997 | spa_import_progress_remove(spa_guid(spa)); |
4a0ee12a PZ |
1998 | spa_load_note(spa, "UNLOADING"); |
1999 | ||
e60e158e JG |
2000 | spa_wake_waiters(spa); |
2001 | ||
93e28d66 | 2002 | /* |
2fb52853 GA |
2003 | * If we have set the spa_final_txg, we have already performed the |
2004 | * tasks below in spa_export_common(). We should not redo it here since | |
2005 | * we delay the final TXGs beyond what spa_final_txg is set at. | |
93e28d66 | 2006 | */ |
2fb52853 GA |
2007 | if (spa->spa_final_txg == UINT64_MAX) { |
2008 | /* | |
2009 | * If the log space map feature is enabled and the pool is | |
2010 | * getting exported (but not destroyed), we want to spend some | |
2011 | * time flushing as many metaslabs as we can in an attempt to | |
2012 | * destroy log space maps and save import time. | |
2013 | */ | |
2014 | if (spa_should_flush_logs_on_unload(spa)) | |
2015 | spa_unload_log_sm_flush_all(spa); | |
93e28d66 | 2016 | |
2fb52853 GA |
2017 | /* |
2018 | * Stop async tasks. | |
2019 | */ | |
2020 | spa_async_suspend(spa); | |
34dc7c2f | 2021 | |
2fb52853 GA |
2022 | if (spa->spa_root_vdev) { |
2023 | vdev_t *root_vdev = spa->spa_root_vdev; | |
2024 | vdev_initialize_stop_all(root_vdev, | |
2025 | VDEV_INITIALIZE_ACTIVE); | |
2026 | vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE); | |
2027 | vdev_autotrim_stop_all(spa); | |
2028 | vdev_rebuild_stop_all(spa); | |
2029 | } | |
619f0976 GW |
2030 | } |
2031 | ||
34dc7c2f BB |
2032 | /* |
2033 | * Stop syncing. | |
2034 | */ | |
2035 | if (spa->spa_sync_on) { | |
2036 | txg_sync_stop(spa->spa_dsl_pool); | |
2037 | spa->spa_sync_on = B_FALSE; | |
2038 | } | |
2039 | ||
4e21fd06 | 2040 | /* |
93e28d66 SD |
2041 | * This ensures that there is no async metaslab prefetching |
2042 | * while we attempt to unload the spa. | |
4e21fd06 | 2043 | */ |
342357cd | 2044 | taskq_wait(spa->spa_metaslab_taskq); |
4e21fd06 | 2045 | |
379ca9cf OF |
2046 | if (spa->spa_mmp.mmp_thread) |
2047 | mmp_thread_stop(spa); | |
2048 | ||
34dc7c2f | 2049 | /* |
b128c09f | 2050 | * Wait for any outstanding async I/O to complete. |
34dc7c2f | 2051 | */ |
9babb374 | 2052 | if (spa->spa_async_zio_root != NULL) { |
1c27024e | 2053 | for (int i = 0; i < max_ncpus; i++) |
e022864d MA |
2054 | (void) zio_wait(spa->spa_async_zio_root[i]); |
2055 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
9babb374 BB |
2056 | spa->spa_async_zio_root = NULL; |
2057 | } | |
34dc7c2f | 2058 | |
a1d477c2 MA |
2059 | if (spa->spa_vdev_removal != NULL) { |
2060 | spa_vdev_removal_destroy(spa->spa_vdev_removal); | |
2061 | spa->spa_vdev_removal = NULL; | |
2062 | } | |
2063 | ||
37f03da8 | 2064 | spa_destroy_aux_threads(spa); |
d2734cce | 2065 | |
a1d477c2 MA |
2066 | spa_condense_fini(spa); |
2067 | ||
428870ff BB |
2068 | bpobj_close(&spa->spa_deferred_bpobj); |
2069 | ||
619f0976 | 2070 | spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); |
93cf2076 GW |
2071 | |
2072 | /* | |
2073 | * Close all vdevs. | |
2074 | */ | |
2075 | if (spa->spa_root_vdev) | |
2076 | vdev_free(spa->spa_root_vdev); | |
2077 | ASSERT(spa->spa_root_vdev == NULL); | |
2078 | ||
34dc7c2f BB |
2079 | /* |
2080 | * Close the dsl pool. | |
2081 | */ | |
2082 | if (spa->spa_dsl_pool) { | |
2083 | dsl_pool_close(spa->spa_dsl_pool); | |
2084 | spa->spa_dsl_pool = NULL; | |
428870ff | 2085 | spa->spa_meta_objset = NULL; |
34dc7c2f BB |
2086 | } |
2087 | ||
428870ff | 2088 | ddt_unload(spa); |
67a1b037 | 2089 | brt_unload(spa); |
93e28d66 | 2090 | spa_unload_log_sm_metadata(spa); |
428870ff | 2091 | |
fb5f0bc8 BB |
2092 | /* |
2093 | * Drop and purge level 2 cache | |
2094 | */ | |
2095 | spa_l2cache_drop(spa); | |
2096 | ||
34dc7c2f | 2097 | if (spa->spa_spares.sav_vdevs) { |
cfb49616 RY |
2098 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
2099 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
34dc7c2f BB |
2100 | kmem_free(spa->spa_spares.sav_vdevs, |
2101 | spa->spa_spares.sav_count * sizeof (void *)); | |
2102 | spa->spa_spares.sav_vdevs = NULL; | |
2103 | } | |
2104 | if (spa->spa_spares.sav_config) { | |
2105 | nvlist_free(spa->spa_spares.sav_config); | |
2106 | spa->spa_spares.sav_config = NULL; | |
2107 | } | |
b128c09f | 2108 | spa->spa_spares.sav_count = 0; |
34dc7c2f | 2109 | |
34dc7c2f | 2110 | if (spa->spa_l2cache.sav_vdevs) { |
cfb49616 RY |
2111 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) { |
2112 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
2113 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
2114 | } | |
34dc7c2f BB |
2115 | kmem_free(spa->spa_l2cache.sav_vdevs, |
2116 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
2117 | spa->spa_l2cache.sav_vdevs = NULL; | |
2118 | } | |
2119 | if (spa->spa_l2cache.sav_config) { | |
2120 | nvlist_free(spa->spa_l2cache.sav_config); | |
2121 | spa->spa_l2cache.sav_config = NULL; | |
2122 | } | |
b128c09f | 2123 | spa->spa_l2cache.sav_count = 0; |
34dc7c2f BB |
2124 | |
2125 | spa->spa_async_suspended = 0; | |
fb5f0bc8 | 2126 | |
a1d477c2 MA |
2127 | spa->spa_indirect_vdevs_loaded = B_FALSE; |
2128 | ||
d96eb2b1 DM |
2129 | if (spa->spa_comment != NULL) { |
2130 | spa_strfree(spa->spa_comment); | |
2131 | spa->spa_comment = NULL; | |
2132 | } | |
658fb802 CB |
2133 | if (spa->spa_compatibility != NULL) { |
2134 | spa_strfree(spa->spa_compatibility); | |
2135 | spa->spa_compatibility = NULL; | |
2136 | } | |
d96eb2b1 | 2137 | |
5caeef02 DB |
2138 | spa->spa_raidz_expand = NULL; |
2139 | ||
619f0976 | 2140 | spa_config_exit(spa, SCL_ALL, spa); |
34dc7c2f BB |
2141 | } |
2142 | ||
2143 | /* | |
2144 | * Load (or re-load) the current list of vdevs describing the active spares for | |
2145 | * this pool. When this is called, we have some form of basic information in | |
2146 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
2147 | * then re-generate a more complete list including status information. | |
2148 | */ | |
a1d477c2 | 2149 | void |
34dc7c2f BB |
2150 | spa_load_spares(spa_t *spa) |
2151 | { | |
2152 | nvlist_t **spares; | |
2153 | uint_t nspares; | |
2154 | int i; | |
2155 | vdev_t *vd, *tvd; | |
2156 | ||
d2734cce SD |
2157 | #ifndef _KERNEL |
2158 | /* | |
2159 | * zdb opens both the current state of the pool and the | |
2160 | * checkpointed state (if present), with a different spa_t. | |
2161 | * | |
2162 | * As spare vdevs are shared among open pools, we skip loading | |
2163 | * them when we load the checkpointed state of the pool. | |
2164 | */ | |
2165 | if (!spa_writeable(spa)) | |
2166 | return; | |
2167 | #endif | |
2168 | ||
b128c09f BB |
2169 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
2170 | ||
34dc7c2f BB |
2171 | /* |
2172 | * First, close and free any existing spare vdevs. | |
2173 | */ | |
cfb49616 RY |
2174 | if (spa->spa_spares.sav_vdevs) { |
2175 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
2176 | vd = spa->spa_spares.sav_vdevs[i]; | |
2177 | ||
2178 | /* Undo the call to spa_activate() below */ | |
2179 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
2180 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
2181 | spa_spare_remove(tvd); | |
2182 | vdev_close(vd); | |
2183 | vdev_free(vd); | |
2184 | } | |
34dc7c2f | 2185 | |
34dc7c2f BB |
2186 | kmem_free(spa->spa_spares.sav_vdevs, |
2187 | spa->spa_spares.sav_count * sizeof (void *)); | |
cfb49616 | 2188 | } |
34dc7c2f BB |
2189 | |
2190 | if (spa->spa_spares.sav_config == NULL) | |
2191 | nspares = 0; | |
2192 | else | |
65ad5d11 AJ |
2193 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, |
2194 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f BB |
2195 | |
2196 | spa->spa_spares.sav_count = (int)nspares; | |
2197 | spa->spa_spares.sav_vdevs = NULL; | |
2198 | ||
2199 | if (nspares == 0) | |
2200 | return; | |
2201 | ||
2202 | /* | |
2203 | * Construct the array of vdevs, opening them to get status in the | |
2204 | * process. For each spare, there is potentially two different vdev_t | |
2205 | * structures associated with it: one in the list of spares (used only | |
2206 | * for basic validation purposes) and one in the active vdev | |
2207 | * configuration (if it's spared in). During this phase we open and | |
2208 | * validate each vdev on the spare list. If the vdev also exists in the | |
2209 | * active configuration, then we also mark this vdev as an active spare. | |
2210 | */ | |
904ea276 | 2211 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), |
79c76d5b | 2212 | KM_SLEEP); |
34dc7c2f BB |
2213 | for (i = 0; i < spa->spa_spares.sav_count; i++) { |
2214 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
2215 | VDEV_ALLOC_SPARE) == 0); | |
2216 | ASSERT(vd != NULL); | |
2217 | ||
2218 | spa->spa_spares.sav_vdevs[i] = vd; | |
2219 | ||
b128c09f BB |
2220 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, |
2221 | B_FALSE)) != NULL) { | |
34dc7c2f BB |
2222 | if (!tvd->vdev_isspare) |
2223 | spa_spare_add(tvd); | |
2224 | ||
2225 | /* | |
2226 | * We only mark the spare active if we were successfully | |
2227 | * able to load the vdev. Otherwise, importing a pool | |
2228 | * with a bad active spare would result in strange | |
2229 | * behavior, because multiple pool would think the spare | |
2230 | * is actively in use. | |
2231 | * | |
2232 | * There is a vulnerability here to an equally bizarre | |
2233 | * circumstance, where a dead active spare is later | |
2234 | * brought back to life (onlined or otherwise). Given | |
2235 | * the rarity of this scenario, and the extra complexity | |
2236 | * it adds, we ignore the possibility. | |
2237 | */ | |
2238 | if (!vdev_is_dead(tvd)) | |
2239 | spa_spare_activate(tvd); | |
2240 | } | |
2241 | ||
b128c09f | 2242 | vd->vdev_top = vd; |
9babb374 | 2243 | vd->vdev_aux = &spa->spa_spares; |
b128c09f | 2244 | |
34dc7c2f BB |
2245 | if (vdev_open(vd) != 0) |
2246 | continue; | |
2247 | ||
34dc7c2f BB |
2248 | if (vdev_validate_aux(vd) == 0) |
2249 | spa_spare_add(vd); | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * Recompute the stashed list of spares, with status information | |
2254 | * this time. | |
2255 | */ | |
65ad5d11 | 2256 | fnvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES); |
34dc7c2f BB |
2257 | |
2258 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
79c76d5b | 2259 | KM_SLEEP); |
34dc7c2f BB |
2260 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
2261 | spares[i] = vdev_config_generate(spa, | |
428870ff | 2262 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); |
65ad5d11 | 2263 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, |
795075e6 PD |
2264 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
2265 | spa->spa_spares.sav_count); | |
34dc7c2f BB |
2266 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
2267 | nvlist_free(spares[i]); | |
2268 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
2269 | } | |
2270 | ||
2271 | /* | |
2272 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
2273 | * this pool. When this is called, we have some form of basic information in | |
2274 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
2275 | * then re-generate a more complete list including status information. | |
2276 | * Devices which are already active have their details maintained, and are | |
2277 | * not re-opened. | |
2278 | */ | |
a1d477c2 | 2279 | void |
34dc7c2f BB |
2280 | spa_load_l2cache(spa_t *spa) |
2281 | { | |
460f239e | 2282 | nvlist_t **l2cache = NULL; |
34dc7c2f BB |
2283 | uint_t nl2cache; |
2284 | int i, j, oldnvdevs; | |
9babb374 | 2285 | uint64_t guid; |
a117a6d6 | 2286 | vdev_t *vd, **oldvdevs, **newvdevs; |
34dc7c2f BB |
2287 | spa_aux_vdev_t *sav = &spa->spa_l2cache; |
2288 | ||
d2734cce SD |
2289 | #ifndef _KERNEL |
2290 | /* | |
2291 | * zdb opens both the current state of the pool and the | |
2292 | * checkpointed state (if present), with a different spa_t. | |
2293 | * | |
2294 | * As L2 caches are part of the ARC which is shared among open | |
2295 | * pools, we skip loading them when we load the checkpointed | |
2296 | * state of the pool. | |
2297 | */ | |
2298 | if (!spa_writeable(spa)) | |
2299 | return; | |
2300 | #endif | |
2301 | ||
b128c09f BB |
2302 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
2303 | ||
34dc7c2f BB |
2304 | oldvdevs = sav->sav_vdevs; |
2305 | oldnvdevs = sav->sav_count; | |
2306 | sav->sav_vdevs = NULL; | |
2307 | sav->sav_count = 0; | |
2308 | ||
67d60824 NB |
2309 | if (sav->sav_config == NULL) { |
2310 | nl2cache = 0; | |
2311 | newvdevs = NULL; | |
2312 | goto out; | |
2313 | } | |
2314 | ||
65ad5d11 AJ |
2315 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, |
2316 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
67d60824 NB |
2317 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); |
2318 | ||
34dc7c2f BB |
2319 | /* |
2320 | * Process new nvlist of vdevs. | |
2321 | */ | |
2322 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 | 2323 | guid = fnvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID); |
34dc7c2f BB |
2324 | |
2325 | newvdevs[i] = NULL; | |
2326 | for (j = 0; j < oldnvdevs; j++) { | |
2327 | vd = oldvdevs[j]; | |
2328 | if (vd != NULL && guid == vd->vdev_guid) { | |
2329 | /* | |
2330 | * Retain previous vdev for add/remove ops. | |
2331 | */ | |
2332 | newvdevs[i] = vd; | |
2333 | oldvdevs[j] = NULL; | |
2334 | break; | |
2335 | } | |
2336 | } | |
2337 | ||
2338 | if (newvdevs[i] == NULL) { | |
2339 | /* | |
2340 | * Create new vdev | |
2341 | */ | |
2342 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
2343 | VDEV_ALLOC_L2CACHE) == 0); | |
2344 | ASSERT(vd != NULL); | |
2345 | newvdevs[i] = vd; | |
2346 | ||
2347 | /* | |
2348 | * Commit this vdev as an l2cache device, | |
2349 | * even if it fails to open. | |
2350 | */ | |
2351 | spa_l2cache_add(vd); | |
2352 | ||
b128c09f BB |
2353 | vd->vdev_top = vd; |
2354 | vd->vdev_aux = sav; | |
2355 | ||
2356 | spa_l2cache_activate(vd); | |
2357 | ||
34dc7c2f BB |
2358 | if (vdev_open(vd) != 0) |
2359 | continue; | |
2360 | ||
34dc7c2f BB |
2361 | (void) vdev_validate_aux(vd); |
2362 | ||
9babb374 BB |
2363 | if (!vdev_is_dead(vd)) |
2364 | l2arc_add_vdev(spa, vd); | |
b7654bd7 GA |
2365 | |
2366 | /* | |
2367 | * Upon cache device addition to a pool or pool | |
2368 | * creation with a cache device or if the header | |
2369 | * of the device is invalid we issue an async | |
2370 | * TRIM command for the whole device which will | |
2371 | * execute if l2arc_trim_ahead > 0. | |
2372 | */ | |
2373 | spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); | |
34dc7c2f BB |
2374 | } |
2375 | } | |
2376 | ||
67d60824 NB |
2377 | sav->sav_vdevs = newvdevs; |
2378 | sav->sav_count = (int)nl2cache; | |
2379 | ||
2380 | /* | |
2381 | * Recompute the stashed list of l2cache devices, with status | |
2382 | * information this time. | |
2383 | */ | |
65ad5d11 | 2384 | fnvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE); |
67d60824 | 2385 | |
460f239e D |
2386 | if (sav->sav_count > 0) |
2387 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
2388 | KM_SLEEP); | |
67d60824 NB |
2389 | for (i = 0; i < sav->sav_count; i++) |
2390 | l2cache[i] = vdev_config_generate(spa, | |
2391 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
795075e6 PD |
2392 | fnvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, |
2393 | (const nvlist_t * const *)l2cache, sav->sav_count); | |
67d60824 NB |
2394 | |
2395 | out: | |
34dc7c2f BB |
2396 | /* |
2397 | * Purge vdevs that were dropped | |
2398 | */ | |
cfb49616 RY |
2399 | if (oldvdevs) { |
2400 | for (i = 0; i < oldnvdevs; i++) { | |
2401 | uint64_t pool; | |
2402 | ||
2403 | vd = oldvdevs[i]; | |
2404 | if (vd != NULL) { | |
2405 | ASSERT(vd->vdev_isl2cache); | |
2406 | ||
2407 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
2408 | pool != 0ULL && l2arc_vdev_present(vd)) | |
2409 | l2arc_remove_vdev(vd); | |
2410 | vdev_clear_stats(vd); | |
2411 | vdev_free(vd); | |
2412 | } | |
34dc7c2f | 2413 | } |
34dc7c2f | 2414 | |
34dc7c2f | 2415 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); |
cfb49616 | 2416 | } |
34dc7c2f | 2417 | |
34dc7c2f BB |
2418 | for (i = 0; i < sav->sav_count; i++) |
2419 | nvlist_free(l2cache[i]); | |
2420 | if (sav->sav_count) | |
2421 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
2422 | } | |
2423 | ||
2424 | static int | |
2425 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
2426 | { | |
2427 | dmu_buf_t *db; | |
2428 | char *packed = NULL; | |
2429 | size_t nvsize = 0; | |
2430 | int error; | |
2431 | *value = NULL; | |
2432 | ||
c3275b56 BB |
2433 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); |
2434 | if (error) | |
2435 | return (error); | |
2436 | ||
34dc7c2f BB |
2437 | nvsize = *(uint64_t *)db->db_data; |
2438 | dmu_buf_rele(db, FTAG); | |
2439 | ||
77aef6f6 | 2440 | packed = vmem_alloc(nvsize, KM_SLEEP); |
9babb374 BB |
2441 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, |
2442 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
2443 | if (error == 0) |
2444 | error = nvlist_unpack(packed, nvsize, value, 0); | |
77aef6f6 | 2445 | vmem_free(packed, nvsize); |
34dc7c2f BB |
2446 | |
2447 | return (error); | |
2448 | } | |
2449 | ||
6cb8e530 PZ |
2450 | /* |
2451 | * Concrete top-level vdevs that are not missing and are not logs. At every | |
2452 | * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. | |
2453 | */ | |
2454 | static uint64_t | |
2455 | spa_healthy_core_tvds(spa_t *spa) | |
2456 | { | |
2457 | vdev_t *rvd = spa->spa_root_vdev; | |
2458 | uint64_t tvds = 0; | |
2459 | ||
2460 | for (uint64_t i = 0; i < rvd->vdev_children; i++) { | |
2461 | vdev_t *vd = rvd->vdev_child[i]; | |
2462 | if (vd->vdev_islog) | |
2463 | continue; | |
2464 | if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) | |
2465 | tvds++; | |
2466 | } | |
2467 | ||
2468 | return (tvds); | |
2469 | } | |
2470 | ||
34dc7c2f BB |
2471 | /* |
2472 | * Checks to see if the given vdev could not be opened, in which case we post a | |
2473 | * sysevent to notify the autoreplace code that the device has been removed. | |
2474 | */ | |
2475 | static void | |
2476 | spa_check_removed(vdev_t *vd) | |
2477 | { | |
6cb8e530 | 2478 | for (uint64_t c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2479 | spa_check_removed(vd->vdev_child[c]); |
2480 | ||
7011fb60 | 2481 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && |
a1d477c2 | 2482 | vdev_is_concrete(vd)) { |
fb390aaf | 2483 | zfs_post_autoreplace(vd->vdev_spa, vd); |
12fa0466 | 2484 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); |
34dc7c2f BB |
2485 | } |
2486 | } | |
2487 | ||
6cb8e530 PZ |
2488 | static int |
2489 | spa_check_for_missing_logs(spa_t *spa) | |
9babb374 | 2490 | { |
6cb8e530 | 2491 | vdev_t *rvd = spa->spa_root_vdev; |
9babb374 | 2492 | |
428870ff | 2493 | /* |
572e2857 | 2494 | * If we're doing a normal import, then build up any additional |
6cb8e530 | 2495 | * diagnostic information about missing log devices. |
572e2857 | 2496 | * We'll pass this up to the user for further processing. |
428870ff | 2497 | */ |
572e2857 BB |
2498 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { |
2499 | nvlist_t **child, *nv; | |
2500 | uint64_t idx = 0; | |
2501 | ||
160987b5 | 2502 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), |
79c76d5b | 2503 | KM_SLEEP); |
65ad5d11 | 2504 | nv = fnvlist_alloc(); |
572e2857 | 2505 | |
6cb8e530 | 2506 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { |
572e2857 | 2507 | vdev_t *tvd = rvd->vdev_child[c]; |
572e2857 | 2508 | |
6cb8e530 PZ |
2509 | /* |
2510 | * We consider a device as missing only if it failed | |
2511 | * to open (i.e. offline or faulted is not considered | |
2512 | * as missing). | |
2513 | */ | |
2514 | if (tvd->vdev_islog && | |
2515 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
2516 | child[idx++] = vdev_config_generate(spa, tvd, | |
2517 | B_FALSE, VDEV_CONFIG_MISSING); | |
2518 | } | |
572e2857 | 2519 | } |
9babb374 | 2520 | |
6cb8e530 | 2521 | if (idx > 0) { |
795075e6 PD |
2522 | fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
2523 | (const nvlist_t * const *)child, idx); | |
6cb8e530 PZ |
2524 | fnvlist_add_nvlist(spa->spa_load_info, |
2525 | ZPOOL_CONFIG_MISSING_DEVICES, nv); | |
572e2857 | 2526 | |
6cb8e530 | 2527 | for (uint64_t i = 0; i < idx; i++) |
572e2857 BB |
2528 | nvlist_free(child[i]); |
2529 | } | |
2530 | nvlist_free(nv); | |
2531 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
572e2857 | 2532 | |
6cb8e530 PZ |
2533 | if (idx > 0) { |
2534 | spa_load_failed(spa, "some log devices are missing"); | |
db7d07e1 | 2535 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 PZ |
2536 | return (SET_ERROR(ENXIO)); |
2537 | } | |
2538 | } else { | |
2539 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
2540 | vdev_t *tvd = rvd->vdev_child[c]; | |
a1d477c2 | 2541 | |
6cb8e530 PZ |
2542 | if (tvd->vdev_islog && |
2543 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
572e2857 | 2544 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
6cb8e530 PZ |
2545 | spa_load_note(spa, "some log devices are " |
2546 | "missing, ZIL is dropped."); | |
db7d07e1 | 2547 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 | 2548 | break; |
e0ab3ab5 | 2549 | } |
572e2857 | 2550 | } |
9babb374 | 2551 | } |
e0ab3ab5 | 2552 | |
6cb8e530 | 2553 | return (0); |
9babb374 BB |
2554 | } |
2555 | ||
b128c09f BB |
2556 | /* |
2557 | * Check for missing log devices | |
2558 | */ | |
13fe0198 | 2559 | static boolean_t |
b128c09f BB |
2560 | spa_check_logs(spa_t *spa) |
2561 | { | |
13fe0198 | 2562 | boolean_t rv = B_FALSE; |
9c43027b | 2563 | dsl_pool_t *dp = spa_get_dsl(spa); |
13fe0198 | 2564 | |
b128c09f | 2565 | switch (spa->spa_log_state) { |
e75c13c3 BB |
2566 | default: |
2567 | break; | |
b128c09f BB |
2568 | case SPA_LOG_MISSING: |
2569 | /* need to recheck in case slog has been restored */ | |
2570 | case SPA_LOG_UNKNOWN: | |
9c43027b AJ |
2571 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
2572 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
13fe0198 | 2573 | if (rv) |
428870ff | 2574 | spa_set_log_state(spa, SPA_LOG_MISSING); |
b128c09f | 2575 | break; |
b128c09f | 2576 | } |
13fe0198 | 2577 | return (rv); |
b128c09f BB |
2578 | } |
2579 | ||
aa755b35 MA |
2580 | /* |
2581 | * Passivate any log vdevs (note, does not apply to embedded log metaslabs). | |
2582 | */ | |
428870ff BB |
2583 | static boolean_t |
2584 | spa_passivate_log(spa_t *spa) | |
34dc7c2f | 2585 | { |
428870ff BB |
2586 | vdev_t *rvd = spa->spa_root_vdev; |
2587 | boolean_t slog_found = B_FALSE; | |
b128c09f | 2588 | |
428870ff | 2589 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
fb5f0bc8 | 2590 | |
1c27024e | 2591 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2592 | vdev_t *tvd = rvd->vdev_child[c]; |
34dc7c2f | 2593 | |
428870ff | 2594 | if (tvd->vdev_islog) { |
aa755b35 MA |
2595 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); |
2596 | metaslab_group_passivate(tvd->vdev_mg); | |
428870ff BB |
2597 | slog_found = B_TRUE; |
2598 | } | |
34dc7c2f BB |
2599 | } |
2600 | ||
428870ff BB |
2601 | return (slog_found); |
2602 | } | |
34dc7c2f | 2603 | |
aa755b35 MA |
2604 | /* |
2605 | * Activate any log vdevs (note, does not apply to embedded log metaslabs). | |
2606 | */ | |
428870ff BB |
2607 | static void |
2608 | spa_activate_log(spa_t *spa) | |
2609 | { | |
2610 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 2611 | |
428870ff BB |
2612 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
2613 | ||
1c27024e | 2614 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2615 | vdev_t *tvd = rvd->vdev_child[c]; |
428870ff | 2616 | |
aa755b35 MA |
2617 | if (tvd->vdev_islog) { |
2618 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); | |
2619 | metaslab_group_activate(tvd->vdev_mg); | |
2620 | } | |
34dc7c2f | 2621 | } |
428870ff | 2622 | } |
34dc7c2f | 2623 | |
428870ff | 2624 | int |
a1d477c2 | 2625 | spa_reset_logs(spa_t *spa) |
428870ff | 2626 | { |
13fe0198 | 2627 | int error; |
9babb374 | 2628 | |
a1d477c2 | 2629 | error = dmu_objset_find(spa_name(spa), zil_reset, |
13fe0198 MA |
2630 | NULL, DS_FIND_CHILDREN); |
2631 | if (error == 0) { | |
428870ff BB |
2632 | /* |
2633 | * We successfully offlined the log device, sync out the | |
2634 | * current txg so that the "stubby" block can be removed | |
2635 | * by zil_sync(). | |
2636 | */ | |
2637 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
2638 | } | |
2639 | return (error); | |
2640 | } | |
34dc7c2f | 2641 | |
428870ff BB |
2642 | static void |
2643 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
2644 | { | |
1c27024e | 2645 | for (int i = 0; i < sav->sav_count; i++) |
428870ff BB |
2646 | spa_check_removed(sav->sav_vdevs[i]); |
2647 | } | |
34dc7c2f | 2648 | |
428870ff BB |
2649 | void |
2650 | spa_claim_notify(zio_t *zio) | |
2651 | { | |
2652 | spa_t *spa = zio->io_spa; | |
34dc7c2f | 2653 | |
428870ff BB |
2654 | if (zio->io_error) |
2655 | return; | |
34dc7c2f | 2656 | |
428870ff | 2657 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ |
493fcce9 GW |
2658 | if (spa->spa_claim_max_txg < BP_GET_LOGICAL_BIRTH(zio->io_bp)) |
2659 | spa->spa_claim_max_txg = BP_GET_LOGICAL_BIRTH(zio->io_bp); | |
428870ff BB |
2660 | mutex_exit(&spa->spa_props_lock); |
2661 | } | |
34dc7c2f | 2662 | |
428870ff | 2663 | typedef struct spa_load_error { |
f2c5bc15 | 2664 | boolean_t sle_verify_data; |
428870ff BB |
2665 | uint64_t sle_meta_count; |
2666 | uint64_t sle_data_count; | |
2667 | } spa_load_error_t; | |
34dc7c2f | 2668 | |
428870ff BB |
2669 | static void |
2670 | spa_load_verify_done(zio_t *zio) | |
2671 | { | |
2672 | blkptr_t *bp = zio->io_bp; | |
2673 | spa_load_error_t *sle = zio->io_private; | |
2674 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
2675 | int error = zio->io_error; | |
dea377c0 | 2676 | spa_t *spa = zio->io_spa; |
34dc7c2f | 2677 | |
a6255b7f | 2678 | abd_free(zio->io_abd); |
428870ff | 2679 | if (error) { |
9ae529ec | 2680 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && |
428870ff | 2681 | type != DMU_OT_INTENT_LOG) |
bc89ac84 | 2682 | atomic_inc_64(&sle->sle_meta_count); |
428870ff | 2683 | else |
bc89ac84 | 2684 | atomic_inc_64(&sle->sle_data_count); |
34dc7c2f | 2685 | } |
dea377c0 MA |
2686 | |
2687 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2688 | spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp); |
dea377c0 MA |
2689 | cv_broadcast(&spa->spa_scrub_io_cv); |
2690 | mutex_exit(&spa->spa_scrub_lock); | |
428870ff | 2691 | } |
34dc7c2f | 2692 | |
dea377c0 | 2693 | /* |
e1cfd73f | 2694 | * Maximum number of inflight bytes is the log2 fraction of the arc size. |
c8242a96 | 2695 | * By default, we set it to 1/16th of the arc. |
dea377c0 | 2696 | */ |
fdc2d303 | 2697 | static uint_t spa_load_verify_shift = 4; |
18168da7 AZ |
2698 | static int spa_load_verify_metadata = B_TRUE; |
2699 | static int spa_load_verify_data = B_TRUE; | |
dea377c0 | 2700 | |
428870ff BB |
2701 | static int |
2702 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
5dbd68a3 | 2703 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) |
428870ff | 2704 | { |
f2c5bc15 AM |
2705 | zio_t *rio = arg; |
2706 | spa_load_error_t *sle = rio->io_private; | |
2707 | ||
14e4e3cb AZ |
2708 | (void) zilog, (void) dnp; |
2709 | ||
dea377c0 MA |
2710 | /* |
2711 | * Note: normally this routine will not be called if | |
2712 | * spa_load_verify_metadata is not set. However, it may be useful | |
2713 | * to manually set the flag after the traversal has begun. | |
2714 | */ | |
2715 | if (!spa_load_verify_metadata) | |
2716 | return (0); | |
2cd0f98f BB |
2717 | |
2718 | /* | |
2719 | * Sanity check the block pointer in order to detect obvious damage | |
2720 | * before using the contents in subsequent checks or in zio_read(). | |
2721 | * When damaged consider it to be a metadata error since we cannot | |
2722 | * trust the BP_GET_TYPE and BP_GET_LEVEL values. | |
2723 | */ | |
3095ca91 | 2724 | if (!zfs_blkptr_verify(spa, bp, BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) { |
2cd0f98f BB |
2725 | atomic_inc_64(&sle->sle_meta_count); |
2726 | return (0); | |
2727 | } | |
2728 | ||
2729 | if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) || | |
2730 | BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) | |
2731 | return (0); | |
2732 | ||
f2c5bc15 AM |
2733 | if (!BP_IS_METADATA(bp) && |
2734 | (!spa_load_verify_data || !sle->sle_verify_data)) | |
dea377c0 MA |
2735 | return (0); |
2736 | ||
1e527162 GW |
2737 | uint64_t maxinflight_bytes = |
2738 | arc_target_bytes() >> spa_load_verify_shift; | |
1c27024e | 2739 | size_t size = BP_GET_PSIZE(bp); |
dea377c0 MA |
2740 | |
2741 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2742 | while (spa->spa_load_verify_bytes >= maxinflight_bytes) |
dea377c0 | 2743 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); |
c8242a96 | 2744 | spa->spa_load_verify_bytes += size; |
dea377c0 MA |
2745 | mutex_exit(&spa->spa_scrub_lock); |
2746 | ||
a6255b7f | 2747 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, |
dea377c0 MA |
2748 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, |
2749 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2750 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
428870ff BB |
2751 | return (0); |
2752 | } | |
34dc7c2f | 2753 | |
65c7cc49 | 2754 | static int |
d1d19c78 PD |
2755 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
2756 | { | |
14e4e3cb AZ |
2757 | (void) dp, (void) arg; |
2758 | ||
d1d19c78 PD |
2759 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) |
2760 | return (SET_ERROR(ENAMETOOLONG)); | |
2761 | ||
2762 | return (0); | |
2763 | } | |
2764 | ||
428870ff BB |
2765 | static int |
2766 | spa_load_verify(spa_t *spa) | |
2767 | { | |
2768 | zio_t *rio; | |
2769 | spa_load_error_t sle = { 0 }; | |
8a393be3 | 2770 | zpool_load_policy_t policy; |
428870ff | 2771 | boolean_t verify_ok = B_FALSE; |
dea377c0 | 2772 | int error = 0; |
34dc7c2f | 2773 | |
8a393be3 | 2774 | zpool_get_load_policy(spa->spa_config, &policy); |
34dc7c2f | 2775 | |
f2c5bc15 AM |
2776 | if (policy.zlp_rewind & ZPOOL_NEVER_REWIND || |
2777 | policy.zlp_maxmeta == UINT64_MAX) | |
428870ff | 2778 | return (0); |
34dc7c2f | 2779 | |
d1d19c78 PD |
2780 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); |
2781 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2782 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2783 | DS_FIND_CHILDREN); | |
2784 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2785 | if (error != 0) | |
2786 | return (error); | |
2787 | ||
f2c5bc15 AM |
2788 | /* |
2789 | * Verify data only if we are rewinding or error limit was set. | |
2790 | * Otherwise nothing except dbgmsg care about it to waste time. | |
2791 | */ | |
2792 | sle.sle_verify_data = (policy.zlp_rewind & ZPOOL_REWIND_MASK) || | |
2793 | (policy.zlp_maxdata < UINT64_MAX); | |
2794 | ||
428870ff BB |
2795 | rio = zio_root(spa, NULL, &sle, |
2796 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
34dc7c2f | 2797 | |
dea377c0 | 2798 | if (spa_load_verify_metadata) { |
4a0ee12a PZ |
2799 | if (spa->spa_extreme_rewind) { |
2800 | spa_load_note(spa, "performing a complete scan of the " | |
2801 | "pool since extreme rewind is on. This may take " | |
2802 | "a very long time.\n (spa_load_verify_data=%u, " | |
2803 | "spa_load_verify_metadata=%u)", | |
2804 | spa_load_verify_data, spa_load_verify_metadata); | |
2805 | } | |
c8242a96 | 2806 | |
dea377c0 | 2807 | error = traverse_pool(spa, spa->spa_verify_min_txg, |
b5256303 TC |
2808 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | |
2809 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
dea377c0 | 2810 | } |
428870ff BB |
2811 | |
2812 | (void) zio_wait(rio); | |
c8242a96 | 2813 | ASSERT0(spa->spa_load_verify_bytes); |
428870ff BB |
2814 | |
2815 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2816 | spa->spa_load_data_errors = sle.sle_data_count; | |
2817 | ||
afd2f7b7 PZ |
2818 | if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { |
2819 | spa_load_note(spa, "spa_load_verify found %llu metadata errors " | |
2820 | "and %llu data errors", (u_longlong_t)sle.sle_meta_count, | |
2821 | (u_longlong_t)sle.sle_data_count); | |
2822 | } | |
2823 | ||
2824 | if (spa_load_verify_dryrun || | |
8a393be3 PZ |
2825 | (!error && sle.sle_meta_count <= policy.zlp_maxmeta && |
2826 | sle.sle_data_count <= policy.zlp_maxdata)) { | |
572e2857 BB |
2827 | int64_t loss = 0; |
2828 | ||
428870ff BB |
2829 | verify_ok = B_TRUE; |
2830 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2831 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
572e2857 BB |
2832 | |
2833 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
65ad5d11 AJ |
2834 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, |
2835 | spa->spa_load_txg_ts); | |
2836 | fnvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, | |
2837 | loss); | |
f2c5bc15 AM |
2838 | fnvlist_add_uint64(spa->spa_load_info, |
2839 | ZPOOL_CONFIG_LOAD_META_ERRORS, sle.sle_meta_count); | |
65ad5d11 AJ |
2840 | fnvlist_add_uint64(spa->spa_load_info, |
2841 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count); | |
428870ff BB |
2842 | } else { |
2843 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2844 | } | |
2845 | ||
afd2f7b7 PZ |
2846 | if (spa_load_verify_dryrun) |
2847 | return (0); | |
2848 | ||
428870ff BB |
2849 | if (error) { |
2850 | if (error != ENXIO && error != EIO) | |
2e528b49 | 2851 | error = SET_ERROR(EIO); |
428870ff BB |
2852 | return (error); |
2853 | } | |
2854 | ||
2855 | return (verify_ok ? 0 : EIO); | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * Find a value in the pool props object. | |
2860 | */ | |
2861 | static void | |
2862 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2863 | { | |
2864 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2865 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2866 | } | |
2867 | ||
2868 | /* | |
2869 | * Find a value in the pool directory object. | |
2870 | */ | |
2871 | static int | |
4a0ee12a | 2872 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) |
428870ff | 2873 | { |
4a0ee12a PZ |
2874 | int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
2875 | name, sizeof (uint64_t), 1, val); | |
2876 | ||
2877 | if (error != 0 && (error != ENOENT || log_enoent)) { | |
2878 | spa_load_failed(spa, "couldn't get '%s' value in MOS directory " | |
2879 | "[error=%d]", name, error); | |
2880 | } | |
2881 | ||
2882 | return (error); | |
428870ff BB |
2883 | } |
2884 | ||
2885 | static int | |
2886 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2887 | { | |
2888 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
a1d477c2 | 2889 | return (SET_ERROR(err)); |
428870ff BB |
2890 | } |
2891 | ||
37f03da8 SH |
2892 | boolean_t |
2893 | spa_livelist_delete_check(spa_t *spa) | |
2894 | { | |
2895 | return (spa->spa_livelists_to_delete != 0); | |
2896 | } | |
2897 | ||
37f03da8 SH |
2898 | static boolean_t |
2899 | spa_livelist_delete_cb_check(void *arg, zthr_t *z) | |
2900 | { | |
14e4e3cb | 2901 | (void) z; |
37f03da8 SH |
2902 | spa_t *spa = arg; |
2903 | return (spa_livelist_delete_check(spa)); | |
2904 | } | |
2905 | ||
2906 | static int | |
2907 | delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
2908 | { | |
2909 | spa_t *spa = arg; | |
2910 | zio_free(spa, tx->tx_txg, bp); | |
2911 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, | |
2912 | -bp_get_dsize_sync(spa, bp), | |
2913 | -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); | |
2914 | return (0); | |
2915 | } | |
2916 | ||
2917 | static int | |
2918 | dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp) | |
2919 | { | |
2920 | int err; | |
2921 | zap_cursor_t zc; | |
2922 | zap_attribute_t za; | |
2923 | zap_cursor_init(&zc, os, zap_obj); | |
2924 | err = zap_cursor_retrieve(&zc, &za); | |
2925 | zap_cursor_fini(&zc); | |
2926 | if (err == 0) | |
2927 | *llp = za.za_first_integer; | |
2928 | return (err); | |
2929 | } | |
2930 | ||
2931 | /* | |
2932 | * Components of livelist deletion that must be performed in syncing | |
2933 | * context: freeing block pointers and updating the pool-wide data | |
2934 | * structures to indicate how much work is left to do | |
2935 | */ | |
2936 | typedef struct sublist_delete_arg { | |
2937 | spa_t *spa; | |
2938 | dsl_deadlist_t *ll; | |
2939 | uint64_t key; | |
2940 | bplist_t *to_free; | |
2941 | } sublist_delete_arg_t; | |
2942 | ||
2943 | static void | |
2944 | sublist_delete_sync(void *arg, dmu_tx_t *tx) | |
2945 | { | |
2946 | sublist_delete_arg_t *sda = arg; | |
2947 | spa_t *spa = sda->spa; | |
2948 | dsl_deadlist_t *ll = sda->ll; | |
2949 | uint64_t key = sda->key; | |
2950 | bplist_t *to_free = sda->to_free; | |
2951 | ||
2952 | bplist_iterate(to_free, delete_blkptr_cb, spa, tx); | |
2953 | dsl_deadlist_remove_entry(ll, key, tx); | |
2954 | } | |
2955 | ||
2956 | typedef struct livelist_delete_arg { | |
2957 | spa_t *spa; | |
2958 | uint64_t ll_obj; | |
2959 | uint64_t zap_obj; | |
2960 | } livelist_delete_arg_t; | |
2961 | ||
2962 | static void | |
2963 | livelist_delete_sync(void *arg, dmu_tx_t *tx) | |
2964 | { | |
2965 | livelist_delete_arg_t *lda = arg; | |
2966 | spa_t *spa = lda->spa; | |
2967 | uint64_t ll_obj = lda->ll_obj; | |
2968 | uint64_t zap_obj = lda->zap_obj; | |
2969 | objset_t *mos = spa->spa_meta_objset; | |
2970 | uint64_t count; | |
2971 | ||
2972 | /* free the livelist and decrement the feature count */ | |
2973 | VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx)); | |
2974 | dsl_deadlist_free(mos, ll_obj, tx); | |
2975 | spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx); | |
2976 | VERIFY0(zap_count(mos, zap_obj, &count)); | |
2977 | if (count == 0) { | |
2978 | /* no more livelists to delete */ | |
2979 | VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT, | |
2980 | DMU_POOL_DELETED_CLONES, tx)); | |
2981 | VERIFY0(zap_destroy(mos, zap_obj, tx)); | |
2982 | spa->spa_livelists_to_delete = 0; | |
e60e158e | 2983 | spa_notify_waiters(spa); |
37f03da8 SH |
2984 | } |
2985 | } | |
2986 | ||
2987 | /* | |
2988 | * Load in the value for the livelist to be removed and open it. Then, | |
2989 | * load its first sublist and determine which block pointers should actually | |
2990 | * be freed. Then, call a synctask which performs the actual frees and updates | |
2991 | * the pool-wide livelist data. | |
2992 | */ | |
65c7cc49 | 2993 | static void |
37f03da8 SH |
2994 | spa_livelist_delete_cb(void *arg, zthr_t *z) |
2995 | { | |
2996 | spa_t *spa = arg; | |
2997 | uint64_t ll_obj = 0, count; | |
2998 | objset_t *mos = spa->spa_meta_objset; | |
2999 | uint64_t zap_obj = spa->spa_livelists_to_delete; | |
3000 | /* | |
3001 | * Determine the next livelist to delete. This function should only | |
3002 | * be called if there is at least one deleted clone. | |
3003 | */ | |
3004 | VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj)); | |
3005 | VERIFY0(zap_count(mos, ll_obj, &count)); | |
3006 | if (count > 0) { | |
c9562576 | 3007 | dsl_deadlist_t *ll; |
37f03da8 SH |
3008 | dsl_deadlist_entry_t *dle; |
3009 | bplist_t to_free; | |
c9562576 PS |
3010 | ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP); |
3011 | dsl_deadlist_open(ll, mos, ll_obj); | |
3012 | dle = dsl_deadlist_first(ll); | |
37f03da8 SH |
3013 | ASSERT3P(dle, !=, NULL); |
3014 | bplist_create(&to_free); | |
3015 | int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free, | |
3016 | z, NULL); | |
3017 | if (err == 0) { | |
3018 | sublist_delete_arg_t sync_arg = { | |
3019 | .spa = spa, | |
c9562576 | 3020 | .ll = ll, |
37f03da8 SH |
3021 | .key = dle->dle_mintxg, |
3022 | .to_free = &to_free | |
3023 | }; | |
3024 | zfs_dbgmsg("deleting sublist (id %llu) from" | |
8e739b2c RE |
3025 | " livelist %llu, %lld remaining", |
3026 | (u_longlong_t)dle->dle_bpobj.bpo_object, | |
3027 | (u_longlong_t)ll_obj, (longlong_t)count - 1); | |
37f03da8 SH |
3028 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, |
3029 | sublist_delete_sync, &sync_arg, 0, | |
3030 | ZFS_SPACE_CHECK_DESTROY)); | |
3031 | } else { | |
d87676a9 | 3032 | VERIFY3U(err, ==, EINTR); |
37f03da8 SH |
3033 | } |
3034 | bplist_clear(&to_free); | |
3035 | bplist_destroy(&to_free); | |
c9562576 PS |
3036 | dsl_deadlist_close(ll); |
3037 | kmem_free(ll, sizeof (dsl_deadlist_t)); | |
37f03da8 SH |
3038 | } else { |
3039 | livelist_delete_arg_t sync_arg = { | |
3040 | .spa = spa, | |
3041 | .ll_obj = ll_obj, | |
3042 | .zap_obj = zap_obj | |
3043 | }; | |
8e739b2c RE |
3044 | zfs_dbgmsg("deletion of livelist %llu completed", |
3045 | (u_longlong_t)ll_obj); | |
37f03da8 SH |
3046 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync, |
3047 | &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY)); | |
3048 | } | |
3049 | } | |
3050 | ||
65c7cc49 | 3051 | static void |
37f03da8 SH |
3052 | spa_start_livelist_destroy_thread(spa_t *spa) |
3053 | { | |
3054 | ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL); | |
843e9ca2 SD |
3055 | spa->spa_livelist_delete_zthr = |
3056 | zthr_create("z_livelist_destroy", | |
6bc61d22 TN |
3057 | spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa, |
3058 | minclsyspri); | |
37f03da8 SH |
3059 | } |
3060 | ||
3061 | typedef struct livelist_new_arg { | |
3062 | bplist_t *allocs; | |
3063 | bplist_t *frees; | |
3064 | } livelist_new_arg_t; | |
3065 | ||
3066 | static int | |
3067 | livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
3068 | dmu_tx_t *tx) | |
3069 | { | |
3070 | ASSERT(tx == NULL); | |
3071 | livelist_new_arg_t *lna = arg; | |
3072 | if (bp_freed) { | |
3073 | bplist_append(lna->frees, bp); | |
3074 | } else { | |
3075 | bplist_append(lna->allocs, bp); | |
3076 | zfs_livelist_condense_new_alloc++; | |
3077 | } | |
3078 | return (0); | |
3079 | } | |
3080 | ||
3081 | typedef struct livelist_condense_arg { | |
3082 | spa_t *spa; | |
3083 | bplist_t to_keep; | |
3084 | uint64_t first_size; | |
3085 | uint64_t next_size; | |
3086 | } livelist_condense_arg_t; | |
3087 | ||
3088 | static void | |
3089 | spa_livelist_condense_sync(void *arg, dmu_tx_t *tx) | |
3090 | { | |
3091 | livelist_condense_arg_t *lca = arg; | |
3092 | spa_t *spa = lca->spa; | |
3093 | bplist_t new_frees; | |
3094 | dsl_dataset_t *ds = spa->spa_to_condense.ds; | |
3095 | ||
3096 | /* Have we been cancelled? */ | |
3097 | if (spa->spa_to_condense.cancelled) { | |
3098 | zfs_livelist_condense_sync_cancel++; | |
3099 | goto out; | |
3100 | } | |
3101 | ||
3102 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
3103 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
3104 | dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; | |
3105 | ||
3106 | /* | |
3107 | * It's possible that the livelist was changed while the zthr was | |
3108 | * running. Therefore, we need to check for new blkptrs in the two | |
3109 | * entries being condensed and continue to track them in the livelist. | |
3110 | * Because of the way we handle remapped blkptrs (see dbuf_remap_impl), | |
3111 | * it's possible that the newly added blkptrs are FREEs or ALLOCs so | |
3112 | * we need to sort them into two different bplists. | |
3113 | */ | |
3114 | uint64_t first_obj = first->dle_bpobj.bpo_object; | |
3115 | uint64_t next_obj = next->dle_bpobj.bpo_object; | |
3116 | uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
3117 | uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
3118 | ||
3119 | bplist_create(&new_frees); | |
3120 | livelist_new_arg_t new_bps = { | |
3121 | .allocs = &lca->to_keep, | |
3122 | .frees = &new_frees, | |
3123 | }; | |
3124 | ||
3125 | if (cur_first_size > lca->first_size) { | |
3126 | VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj, | |
3127 | livelist_track_new_cb, &new_bps, lca->first_size)); | |
3128 | } | |
3129 | if (cur_next_size > lca->next_size) { | |
3130 | VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj, | |
3131 | livelist_track_new_cb, &new_bps, lca->next_size)); | |
3132 | } | |
3133 | ||
3134 | dsl_deadlist_clear_entry(first, ll, tx); | |
3135 | ASSERT(bpobj_is_empty(&first->dle_bpobj)); | |
3136 | dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx); | |
3137 | ||
3138 | bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx); | |
3139 | bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx); | |
3140 | bplist_destroy(&new_frees); | |
3141 | ||
3142 | char dsname[ZFS_MAX_DATASET_NAME_LEN]; | |
3143 | dsl_dataset_name(ds, dsname); | |
3144 | zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu " | |
3145 | "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu " | |
8e739b2c RE |
3146 | "(%llu blkptrs)", (u_longlong_t)tx->tx_txg, dsname, |
3147 | (u_longlong_t)ds->ds_object, (u_longlong_t)first_obj, | |
3148 | (u_longlong_t)cur_first_size, (u_longlong_t)next_obj, | |
3149 | (u_longlong_t)cur_next_size, | |
3150 | (u_longlong_t)first->dle_bpobj.bpo_object, | |
3151 | (u_longlong_t)first->dle_bpobj.bpo_phys->bpo_num_blkptrs); | |
37f03da8 SH |
3152 | out: |
3153 | dmu_buf_rele(ds->ds_dbuf, spa); | |
3154 | spa->spa_to_condense.ds = NULL; | |
3155 | bplist_clear(&lca->to_keep); | |
3156 | bplist_destroy(&lca->to_keep); | |
3157 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
3158 | spa->spa_to_condense.syncing = B_FALSE; | |
3159 | } | |
3160 | ||
65c7cc49 | 3161 | static void |
37f03da8 SH |
3162 | spa_livelist_condense_cb(void *arg, zthr_t *t) |
3163 | { | |
3164 | while (zfs_livelist_condense_zthr_pause && | |
3165 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
3166 | delay(1); | |
3167 | ||
3168 | spa_t *spa = arg; | |
3169 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
3170 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
3171 | uint64_t first_size, next_size; | |
3172 | ||
3173 | livelist_condense_arg_t *lca = | |
3174 | kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP); | |
3175 | bplist_create(&lca->to_keep); | |
3176 | ||
3177 | /* | |
3178 | * Process the livelists (matching FREEs and ALLOCs) in open context | |
3179 | * so we have minimal work in syncing context to condense. | |
3180 | * | |
3181 | * We save bpobj sizes (first_size and next_size) to use later in | |
3182 | * syncing context to determine if entries were added to these sublists | |
3183 | * while in open context. This is possible because the clone is still | |
3184 | * active and open for normal writes and we want to make sure the new, | |
3185 | * unprocessed blockpointers are inserted into the livelist normally. | |
3186 | * | |
3187 | * Note that dsl_process_sub_livelist() both stores the size number of | |
3188 | * blockpointers and iterates over them while the bpobj's lock held, so | |
3189 | * the sizes returned to us are consistent which what was actually | |
3190 | * processed. | |
3191 | */ | |
3192 | int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t, | |
3193 | &first_size); | |
3194 | if (err == 0) | |
3195 | err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep, | |
3196 | t, &next_size); | |
3197 | ||
3198 | if (err == 0) { | |
3199 | while (zfs_livelist_condense_sync_pause && | |
3200 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
3201 | delay(1); | |
3202 | ||
3203 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
3204 | dmu_tx_mark_netfree(tx); | |
3205 | dmu_tx_hold_space(tx, 1); | |
3206 | err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE); | |
3207 | if (err == 0) { | |
3208 | /* | |
3209 | * Prevent the condense zthr restarting before | |
3210 | * the synctask completes. | |
3211 | */ | |
3212 | spa->spa_to_condense.syncing = B_TRUE; | |
3213 | lca->spa = spa; | |
3214 | lca->first_size = first_size; | |
3215 | lca->next_size = next_size; | |
3216 | dsl_sync_task_nowait(spa_get_dsl(spa), | |
38080324 | 3217 | spa_livelist_condense_sync, lca, tx); |
37f03da8 SH |
3218 | dmu_tx_commit(tx); |
3219 | return; | |
3220 | } | |
3221 | } | |
3222 | /* | |
3223 | * Condensing can not continue: either it was externally stopped or | |
3224 | * we were unable to assign to a tx because the pool has run out of | |
3225 | * space. In the second case, we'll just end up trying to condense | |
3226 | * again in a later txg. | |
3227 | */ | |
3228 | ASSERT(err != 0); | |
3229 | bplist_clear(&lca->to_keep); | |
3230 | bplist_destroy(&lca->to_keep); | |
3231 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
3232 | dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa); | |
3233 | spa->spa_to_condense.ds = NULL; | |
3234 | if (err == EINTR) | |
3235 | zfs_livelist_condense_zthr_cancel++; | |
3236 | } | |
3237 | ||
37f03da8 SH |
3238 | /* |
3239 | * Check that there is something to condense but that a condense is not | |
3240 | * already in progress and that condensing has not been cancelled. | |
3241 | */ | |
3242 | static boolean_t | |
3243 | spa_livelist_condense_cb_check(void *arg, zthr_t *z) | |
3244 | { | |
14e4e3cb | 3245 | (void) z; |
37f03da8 SH |
3246 | spa_t *spa = arg; |
3247 | if ((spa->spa_to_condense.ds != NULL) && | |
3248 | (spa->spa_to_condense.syncing == B_FALSE) && | |
3249 | (spa->spa_to_condense.cancelled == B_FALSE)) { | |
3250 | return (B_TRUE); | |
3251 | } | |
3252 | return (B_FALSE); | |
3253 | } | |
3254 | ||
65c7cc49 | 3255 | static void |
37f03da8 SH |
3256 | spa_start_livelist_condensing_thread(spa_t *spa) |
3257 | { | |
3258 | spa->spa_to_condense.ds = NULL; | |
3259 | spa->spa_to_condense.first = NULL; | |
3260 | spa->spa_to_condense.next = NULL; | |
3261 | spa->spa_to_condense.syncing = B_FALSE; | |
3262 | spa->spa_to_condense.cancelled = B_FALSE; | |
3263 | ||
3264 | ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL); | |
843e9ca2 SD |
3265 | spa->spa_livelist_condense_zthr = |
3266 | zthr_create("z_livelist_condense", | |
3267 | spa_livelist_condense_cb_check, | |
6bc61d22 | 3268 | spa_livelist_condense_cb, spa, minclsyspri); |
37f03da8 SH |
3269 | } |
3270 | ||
9d5b5245 SD |
3271 | static void |
3272 | spa_spawn_aux_threads(spa_t *spa) | |
3273 | { | |
3274 | ASSERT(spa_writeable(spa)); | |
3275 | ||
5caeef02 | 3276 | spa_start_raidz_expansion_thread(spa); |
9d5b5245 | 3277 | spa_start_indirect_condensing_thread(spa); |
37f03da8 SH |
3278 | spa_start_livelist_destroy_thread(spa); |
3279 | spa_start_livelist_condensing_thread(spa); | |
d2734cce SD |
3280 | |
3281 | ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); | |
3282 | spa->spa_checkpoint_discard_zthr = | |
843e9ca2 SD |
3283 | zthr_create("z_checkpoint_discard", |
3284 | spa_checkpoint_discard_thread_check, | |
6bc61d22 | 3285 | spa_checkpoint_discard_thread, spa, minclsyspri); |
9d5b5245 SD |
3286 | } |
3287 | ||
428870ff BB |
3288 | /* |
3289 | * Fix up config after a partly-completed split. This is done with the | |
3290 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
3291 | * pool have that entry in their config, but only the splitting one contains | |
3292 | * a list of all the guids of the vdevs that are being split off. | |
3293 | * | |
3294 | * This function determines what to do with that list: either rejoin | |
3295 | * all the disks to the pool, or complete the splitting process. To attempt | |
3296 | * the rejoin, each disk that is offlined is marked online again, and | |
3297 | * we do a reopen() call. If the vdev label for every disk that was | |
3298 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
3299 | * then we call vdev_split() on each disk, and complete the split. | |
3300 | * | |
3301 | * Otherwise we leave the config alone, with all the vdevs in place in | |
3302 | * the original pool. | |
3303 | */ | |
3304 | static void | |
3305 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
3306 | { | |
3307 | uint_t extracted; | |
3308 | uint64_t *glist; | |
3309 | uint_t i, gcount; | |
3310 | nvlist_t *nvl; | |
3311 | vdev_t **vd; | |
3312 | boolean_t attempt_reopen; | |
3313 | ||
3314 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
3315 | return; | |
3316 | ||
3317 | /* check that the config is complete */ | |
3318 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
3319 | &glist, &gcount) != 0) | |
3320 | return; | |
3321 | ||
79c76d5b | 3322 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); |
428870ff BB |
3323 | |
3324 | /* attempt to online all the vdevs & validate */ | |
3325 | attempt_reopen = B_TRUE; | |
3326 | for (i = 0; i < gcount; i++) { | |
3327 | if (glist[i] == 0) /* vdev is hole */ | |
3328 | continue; | |
3329 | ||
3330 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
3331 | if (vd[i] == NULL) { | |
3332 | /* | |
3333 | * Don't bother attempting to reopen the disks; | |
3334 | * just do the split. | |
3335 | */ | |
3336 | attempt_reopen = B_FALSE; | |
3337 | } else { | |
3338 | /* attempt to re-online it */ | |
3339 | vd[i]->vdev_offline = B_FALSE; | |
3340 | } | |
3341 | } | |
3342 | ||
3343 | if (attempt_reopen) { | |
3344 | vdev_reopen(spa->spa_root_vdev); | |
3345 | ||
3346 | /* check each device to see what state it's in */ | |
3347 | for (extracted = 0, i = 0; i < gcount; i++) { | |
3348 | if (vd[i] != NULL && | |
3349 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
3350 | break; | |
3351 | ++extracted; | |
3352 | } | |
3353 | } | |
3354 | ||
3355 | /* | |
3356 | * If every disk has been moved to the new pool, or if we never | |
3357 | * even attempted to look at them, then we split them off for | |
3358 | * good. | |
3359 | */ | |
3360 | if (!attempt_reopen || gcount == extracted) { | |
3361 | for (i = 0; i < gcount; i++) | |
3362 | if (vd[i] != NULL) | |
3363 | vdev_split(vd[i]); | |
3364 | vdev_reopen(spa->spa_root_vdev); | |
3365 | } | |
3366 | ||
3367 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
3368 | } | |
3369 | ||
3370 | static int | |
6cb8e530 | 3371 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) |
428870ff | 3372 | { |
a926aab9 | 3373 | const char *ereport = FM_EREPORT_ZFS_POOL; |
428870ff | 3374 | int error; |
428870ff | 3375 | |
6cb8e530 | 3376 | spa->spa_load_state = state; |
ca95f70d OF |
3377 | (void) spa_import_progress_set_state(spa_guid(spa), |
3378 | spa_load_state(spa)); | |
687e4d7f | 3379 | spa_import_progress_set_notes(spa, "spa_load()"); |
9ae529ec | 3380 | |
6cb8e530 | 3381 | gethrestime(&spa->spa_loaded_ts); |
d2734cce | 3382 | error = spa_load_impl(spa, type, &ereport); |
428870ff | 3383 | |
0c66c32d JG |
3384 | /* |
3385 | * Don't count references from objsets that are already closed | |
3386 | * and are making their way through the eviction process. | |
3387 | */ | |
3388 | spa_evicting_os_wait(spa); | |
424fd7c3 | 3389 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
572e2857 BB |
3390 | if (error) { |
3391 | if (error != EEXIST) { | |
3392 | spa->spa_loaded_ts.tv_sec = 0; | |
3393 | spa->spa_loaded_ts.tv_nsec = 0; | |
3394 | } | |
3395 | if (error != EBADF) { | |
1144586b | 3396 | (void) zfs_ereport_post(ereport, spa, |
4f072827 | 3397 | NULL, NULL, NULL, 0); |
572e2857 BB |
3398 | } |
3399 | } | |
428870ff BB |
3400 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; |
3401 | spa->spa_ena = 0; | |
3402 | ||
ca95f70d OF |
3403 | (void) spa_import_progress_set_state(spa_guid(spa), |
3404 | spa_load_state(spa)); | |
3405 | ||
428870ff BB |
3406 | return (error); |
3407 | } | |
3408 | ||
33cf67cd | 3409 | #ifdef ZFS_DEBUG |
e0ab3ab5 JS |
3410 | /* |
3411 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
3412 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
3413 | * spa's per-vdev ZAP list. | |
3414 | */ | |
3415 | static uint64_t | |
3416 | vdev_count_verify_zaps(vdev_t *vd) | |
3417 | { | |
3418 | spa_t *spa = vd->vdev_spa; | |
3419 | uint64_t total = 0; | |
e0ab3ab5 | 3420 | |
3e4ed421 RW |
3421 | if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2) && |
3422 | vd->vdev_root_zap != 0) { | |
3423 | total++; | |
3424 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3425 | spa->spa_all_vdev_zaps, vd->vdev_root_zap)); | |
3426 | } | |
e0ab3ab5 JS |
3427 | if (vd->vdev_top_zap != 0) { |
3428 | total++; | |
3429 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3430 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
3431 | } | |
3432 | if (vd->vdev_leaf_zap != 0) { | |
3433 | total++; | |
3434 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3435 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
3436 | } | |
3437 | ||
1c27024e | 3438 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
3439 | total += vdev_count_verify_zaps(vd->vdev_child[i]); |
3440 | } | |
3441 | ||
3442 | return (total); | |
3443 | } | |
36542b06 AZ |
3444 | #else |
3445 | #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0) | |
33cf67cd | 3446 | #endif |
e0ab3ab5 | 3447 | |
379ca9cf OF |
3448 | /* |
3449 | * Determine whether the activity check is required. | |
3450 | */ | |
3451 | static boolean_t | |
bbffb59e BB |
3452 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, |
3453 | nvlist_t *config) | |
379ca9cf OF |
3454 | { |
3455 | uint64_t state = 0; | |
3456 | uint64_t hostid = 0; | |
3457 | uint64_t tryconfig_txg = 0; | |
3458 | uint64_t tryconfig_timestamp = 0; | |
060f0226 | 3459 | uint16_t tryconfig_mmp_seq = 0; |
379ca9cf OF |
3460 | nvlist_t *nvinfo; |
3461 | ||
3462 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3463 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
3464 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
3465 | &tryconfig_txg); | |
3466 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
3467 | &tryconfig_timestamp); | |
060f0226 OF |
3468 | (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ, |
3469 | &tryconfig_mmp_seq); | |
379ca9cf OF |
3470 | } |
3471 | ||
3472 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
379ca9cf OF |
3473 | |
3474 | /* | |
3475 | * Disable the MMP activity check - This is used by zdb which | |
3476 | * is intended to be used on potentially active pools. | |
3477 | */ | |
3478 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
3479 | return (B_FALSE); | |
3480 | ||
3481 | /* | |
3482 | * Skip the activity check when the MMP feature is disabled. | |
3483 | */ | |
3484 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
3485 | return (B_FALSE); | |
ca95f70d | 3486 | |
379ca9cf | 3487 | /* |
060f0226 OF |
3488 | * If the tryconfig_ values are nonzero, they are the results of an |
3489 | * earlier tryimport. If they all match the uberblock we just found, | |
3490 | * then the pool has not changed and we return false so we do not test | |
3491 | * a second time. | |
379ca9cf OF |
3492 | */ |
3493 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
060f0226 OF |
3494 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp && |
3495 | tryconfig_mmp_seq && tryconfig_mmp_seq == | |
3496 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) | |
379ca9cf OF |
3497 | return (B_FALSE); |
3498 | ||
3499 | /* | |
3500 | * Allow the activity check to be skipped when importing the pool | |
bbffb59e BB |
3501 | * on the same host which last imported it. Since the hostid from |
3502 | * configuration may be stale use the one read from the label. | |
379ca9cf | 3503 | */ |
bbffb59e BB |
3504 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) |
3505 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
3506 | ||
25f06d67 | 3507 | if (hostid == spa_get_hostid(spa)) |
379ca9cf OF |
3508 | return (B_FALSE); |
3509 | ||
3510 | /* | |
3511 | * Skip the activity test when the pool was cleanly exported. | |
3512 | */ | |
3513 | if (state != POOL_STATE_ACTIVE) | |
3514 | return (B_FALSE); | |
3515 | ||
3516 | return (B_TRUE); | |
3517 | } | |
3518 | ||
060f0226 OF |
3519 | /* |
3520 | * Nanoseconds the activity check must watch for changes on-disk. | |
3521 | */ | |
3522 | static uint64_t | |
3523 | spa_activity_check_duration(spa_t *spa, uberblock_t *ub) | |
3524 | { | |
3525 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
3526 | uint64_t multihost_interval = MSEC2NSEC( | |
3527 | MMP_INTERVAL_OK(zfs_multihost_interval)); | |
3528 | uint64_t import_delay = MAX(NANOSEC, import_intervals * | |
3529 | multihost_interval); | |
3530 | ||
3531 | /* | |
3532 | * Local tunables determine a minimum duration except for the case | |
3533 | * where we know when the remote host will suspend the pool if MMP | |
3534 | * writes do not land. | |
3535 | * | |
3536 | * See Big Theory comment at the top of mmp.c for the reasoning behind | |
3537 | * these cases and times. | |
3538 | */ | |
3539 | ||
3540 | ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100); | |
3541 | ||
3542 | if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3543 | MMP_FAIL_INT(ub) > 0) { | |
3544 | ||
3545 | /* MMP on remote host will suspend pool after failed writes */ | |
3546 | import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) * | |
3547 | MMP_IMPORT_SAFETY_FACTOR / 100; | |
3548 | ||
3549 | zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp " | |
3550 | "mmp_fails=%llu ub_mmp mmp_interval=%llu " | |
8e739b2c RE |
3551 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3552 | (u_longlong_t)MMP_FAIL_INT(ub), | |
3553 | (u_longlong_t)MMP_INTERVAL(ub), | |
3554 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3555 | |
3556 | } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3557 | MMP_FAIL_INT(ub) == 0) { | |
3558 | ||
3559 | /* MMP on remote host will never suspend pool */ | |
3560 | import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) + | |
3561 | ub->ub_mmp_delay) * import_intervals); | |
3562 | ||
3563 | zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp " | |
3564 | "mmp_interval=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3565 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3566 | (u_longlong_t)MMP_INTERVAL(ub), | |
3567 | (u_longlong_t)ub->ub_mmp_delay, | |
3568 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3569 | |
3570 | } else if (MMP_VALID(ub)) { | |
3571 | /* | |
e1cfd73f | 3572 | * zfs-0.7 compatibility case |
060f0226 OF |
3573 | */ |
3574 | ||
3575 | import_delay = MAX(import_delay, (multihost_interval + | |
3576 | ub->ub_mmp_delay) * import_intervals); | |
3577 | ||
3578 | zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3579 | "import_intervals=%llu leaves=%u", |
3580 | (u_longlong_t)import_delay, | |
3581 | (u_longlong_t)ub->ub_mmp_delay, | |
3582 | (u_longlong_t)import_intervals, | |
060f0226 OF |
3583 | vdev_count_leaves(spa)); |
3584 | } else { | |
3585 | /* Using local tunings is the only reasonable option */ | |
3586 | zfs_dbgmsg("pool last imported on non-MMP aware " | |
3587 | "host using import_delay=%llu multihost_interval=%llu " | |
8e739b2c RE |
3588 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3589 | (u_longlong_t)multihost_interval, | |
3590 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3591 | } |
3592 | ||
3593 | return (import_delay); | |
3594 | } | |
3595 | ||
379ca9cf | 3596 | /* |
c3f2f1aa DB |
3597 | * Remote host activity check. |
3598 | * | |
3599 | * error results: | |
3600 | * 0 - no activity detected | |
3601 | * EREMOTEIO - remote activity detected | |
3602 | * EINTR - user canceled the operation | |
379ca9cf OF |
3603 | */ |
3604 | static int | |
c3f2f1aa DB |
3605 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config, |
3606 | boolean_t importing) | |
379ca9cf | 3607 | { |
379ca9cf OF |
3608 | uint64_t txg = ub->ub_txg; |
3609 | uint64_t timestamp = ub->ub_timestamp; | |
060f0226 OF |
3610 | uint64_t mmp_config = ub->ub_mmp_config; |
3611 | uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0; | |
3612 | uint64_t import_delay; | |
687e4d7f | 3613 | hrtime_t import_expire, now; |
379ca9cf OF |
3614 | nvlist_t *mmp_label = NULL; |
3615 | vdev_t *rvd = spa->spa_root_vdev; | |
3616 | kcondvar_t cv; | |
3617 | kmutex_t mtx; | |
3618 | int error = 0; | |
3619 | ||
3620 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
3621 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
3622 | mutex_enter(&mtx); | |
3623 | ||
3624 | /* | |
3625 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
3626 | * during the earlier tryimport. If the txg recorded there is 0 then | |
3627 | * the pool is known to be active on another host. | |
3628 | * | |
060f0226 | 3629 | * Otherwise, the pool might be in use on another host. Check for |
379ca9cf OF |
3630 | * changes in the uberblocks on disk if necessary. |
3631 | */ | |
3632 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3633 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
3634 | ZPOOL_CONFIG_LOAD_INFO); | |
3635 | ||
3636 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
3637 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
3638 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
3639 | error = SET_ERROR(EREMOTEIO); | |
3640 | goto out; | |
3641 | } | |
3642 | } | |
3643 | ||
060f0226 | 3644 | import_delay = spa_activity_check_duration(spa, ub); |
533ea041 | 3645 | |
379ca9cf | 3646 | /* Add a small random factor in case of simultaneous imports (0-25%) */ |
29274c9f | 3647 | import_delay += import_delay * random_in_range(250) / 1000; |
ca95f70d OF |
3648 | |
3649 | import_expire = gethrtime() + import_delay; | |
379ca9cf | 3650 | |
c3f2f1aa DB |
3651 | if (importing) { |
3652 | spa_import_progress_set_notes(spa, "Checking MMP activity, " | |
3653 | "waiting %llu ms", (u_longlong_t)NSEC2MSEC(import_delay)); | |
3654 | } | |
687e4d7f | 3655 | |
c3f2f1aa | 3656 | int iterations = 0; |
687e4d7f | 3657 | while ((now = gethrtime()) < import_expire) { |
c3f2f1aa | 3658 | if (importing && iterations++ % 30 == 0) { |
687e4d7f DB |
3659 | spa_import_progress_set_notes(spa, "Checking MMP " |
3660 | "activity, %llu ms remaining", | |
3661 | (u_longlong_t)NSEC2MSEC(import_expire - now)); | |
3662 | } | |
3663 | ||
c3f2f1aa DB |
3664 | if (importing) { |
3665 | (void) spa_import_progress_set_mmp_check(spa_guid(spa), | |
3666 | NSEC2SEC(import_expire - gethrtime())); | |
3667 | } | |
ca95f70d | 3668 | |
379ca9cf OF |
3669 | vdev_uberblock_load(rvd, ub, &mmp_label); |
3670 | ||
060f0226 OF |
3671 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp || |
3672 | mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) { | |
3673 | zfs_dbgmsg("multihost activity detected " | |
3674 | "txg %llu ub_txg %llu " | |
3675 | "timestamp %llu ub_timestamp %llu " | |
3676 | "mmp_config %#llx ub_mmp_config %#llx", | |
8e739b2c RE |
3677 | (u_longlong_t)txg, (u_longlong_t)ub->ub_txg, |
3678 | (u_longlong_t)timestamp, | |
3679 | (u_longlong_t)ub->ub_timestamp, | |
3680 | (u_longlong_t)mmp_config, | |
3681 | (u_longlong_t)ub->ub_mmp_config); | |
060f0226 | 3682 | |
379ca9cf OF |
3683 | error = SET_ERROR(EREMOTEIO); |
3684 | break; | |
3685 | } | |
3686 | ||
3687 | if (mmp_label) { | |
3688 | nvlist_free(mmp_label); | |
3689 | mmp_label = NULL; | |
3690 | } | |
3691 | ||
3692 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
3693 | if (error != -1) { | |
3694 | error = SET_ERROR(EINTR); | |
3695 | break; | |
3696 | } | |
3697 | error = 0; | |
3698 | } | |
3699 | ||
3700 | out: | |
3701 | mutex_exit(&mtx); | |
3702 | mutex_destroy(&mtx); | |
3703 | cv_destroy(&cv); | |
3704 | ||
3705 | /* | |
3706 | * If the pool is determined to be active store the status in the | |
3707 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
3708 | * available from configuration read from disk store them as well. | |
3709 | * This allows 'zpool import' to generate a more useful message. | |
3710 | * | |
3711 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
3712 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
3713 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
3714 | */ | |
3715 | if (error == EREMOTEIO) { | |
a926aab9 | 3716 | const char *hostname = "<unknown>"; |
379ca9cf OF |
3717 | uint64_t hostid = 0; |
3718 | ||
3719 | if (mmp_label) { | |
3720 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
3721 | hostname = fnvlist_lookup_string(mmp_label, | |
3722 | ZPOOL_CONFIG_HOSTNAME); | |
3723 | fnvlist_add_string(spa->spa_load_info, | |
3724 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
3725 | } | |
3726 | ||
3727 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
3728 | hostid = fnvlist_lookup_uint64(mmp_label, | |
3729 | ZPOOL_CONFIG_HOSTID); | |
3730 | fnvlist_add_uint64(spa->spa_load_info, | |
3731 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
3732 | } | |
3733 | } | |
3734 | ||
3735 | fnvlist_add_uint64(spa->spa_load_info, | |
3736 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
3737 | fnvlist_add_uint64(spa->spa_load_info, | |
3738 | ZPOOL_CONFIG_MMP_TXG, 0); | |
3739 | ||
3740 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
3741 | } | |
3742 | ||
3743 | if (mmp_label) | |
3744 | nvlist_free(mmp_label); | |
3745 | ||
3746 | return (error); | |
3747 | } | |
3748 | ||
c3f2f1aa DB |
3749 | /* |
3750 | * Called from zfs_ioc_clear for a pool that was suspended | |
3751 | * after failing mmp write checks. | |
3752 | */ | |
3753 | boolean_t | |
3754 | spa_mmp_remote_host_activity(spa_t *spa) | |
3755 | { | |
3756 | ASSERT(spa_multihost(spa) && spa_suspended(spa)); | |
3757 | ||
3758 | nvlist_t *best_label; | |
3759 | uberblock_t best_ub; | |
3760 | ||
3761 | /* | |
3762 | * Locate the best uberblock on disk | |
3763 | */ | |
3764 | vdev_uberblock_load(spa->spa_root_vdev, &best_ub, &best_label); | |
3765 | if (best_label) { | |
3766 | /* | |
3767 | * confirm that the best hostid matches our hostid | |
3768 | */ | |
3769 | if (nvlist_exists(best_label, ZPOOL_CONFIG_HOSTID) && | |
3770 | spa_get_hostid(spa) != | |
3771 | fnvlist_lookup_uint64(best_label, ZPOOL_CONFIG_HOSTID)) { | |
3772 | nvlist_free(best_label); | |
3773 | return (B_TRUE); | |
3774 | } | |
3775 | nvlist_free(best_label); | |
3776 | } else { | |
3777 | return (B_TRUE); | |
3778 | } | |
3779 | ||
3780 | if (!MMP_VALID(&best_ub) || | |
3781 | !MMP_FAIL_INT_VALID(&best_ub) || | |
3782 | MMP_FAIL_INT(&best_ub) == 0) { | |
3783 | return (B_TRUE); | |
3784 | } | |
3785 | ||
3786 | if (best_ub.ub_txg != spa->spa_uberblock.ub_txg || | |
3787 | best_ub.ub_timestamp != spa->spa_uberblock.ub_timestamp) { | |
3788 | zfs_dbgmsg("txg mismatch detected during pool clear " | |
3789 | "txg %llu ub_txg %llu timestamp %llu ub_timestamp %llu", | |
3790 | (u_longlong_t)spa->spa_uberblock.ub_txg, | |
3791 | (u_longlong_t)best_ub.ub_txg, | |
3792 | (u_longlong_t)spa->spa_uberblock.ub_timestamp, | |
3793 | (u_longlong_t)best_ub.ub_timestamp); | |
3794 | return (B_TRUE); | |
3795 | } | |
3796 | ||
3797 | /* | |
3798 | * Perform an activity check looking for any remote writer | |
3799 | */ | |
3800 | return (spa_activity_check(spa, &spa->spa_uberblock, spa->spa_config, | |
3801 | B_FALSE) != 0); | |
3802 | } | |
3803 | ||
9eb7b46e | 3804 | static int |
6cb8e530 PZ |
3805 | spa_verify_host(spa_t *spa, nvlist_t *mos_config) |
3806 | { | |
3807 | uint64_t hostid; | |
d1807f16 | 3808 | const char *hostname; |
6cb8e530 PZ |
3809 | uint64_t myhostid = 0; |
3810 | ||
3811 | if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, | |
3812 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
3813 | hostname = fnvlist_lookup_string(mos_config, | |
3814 | ZPOOL_CONFIG_HOSTNAME); | |
3815 | ||
3816 | myhostid = zone_get_hostid(NULL); | |
3817 | ||
3818 | if (hostid != 0 && myhostid != 0 && hostid != myhostid) { | |
3819 | cmn_err(CE_WARN, "pool '%s' could not be " | |
3820 | "loaded as it was last accessed by " | |
3821 | "another system (host: %s hostid: 0x%llx). " | |
a2f944a1 RM |
3822 | "See: https://openzfs.github.io/openzfs-docs/msg/" |
3823 | "ZFS-8000-EY", | |
6cb8e530 PZ |
3824 | spa_name(spa), hostname, (u_longlong_t)hostid); |
3825 | spa_load_failed(spa, "hostid verification failed: pool " | |
3826 | "last accessed by host: %s (hostid: 0x%llx)", | |
3827 | hostname, (u_longlong_t)hostid); | |
3828 | return (SET_ERROR(EBADF)); | |
3829 | } | |
3830 | } | |
3831 | ||
3832 | return (0); | |
3833 | } | |
3834 | ||
3835 | static int | |
3836 | spa_ld_parse_config(spa_t *spa, spa_import_type_t type) | |
428870ff BB |
3837 | { |
3838 | int error = 0; | |
6cb8e530 | 3839 | nvlist_t *nvtree, *nvl, *config = spa->spa_config; |
1c27024e | 3840 | int parse; |
9eb7b46e | 3841 | vdev_t *rvd; |
6cb8e530 | 3842 | uint64_t pool_guid; |
d1807f16 RY |
3843 | const char *comment; |
3844 | const char *compatibility; | |
6cb8e530 PZ |
3845 | |
3846 | /* | |
3847 | * Versioning wasn't explicitly added to the label until later, so if | |
3848 | * it's not present treat it as the initial version. | |
3849 | */ | |
3850 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
3851 | &spa->spa_ubsync.ub_version) != 0) | |
3852 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
3853 | ||
3854 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { | |
3855 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3856 | ZPOOL_CONFIG_POOL_GUID); | |
3857 | return (SET_ERROR(EINVAL)); | |
3858 | } | |
3859 | ||
d2734cce SD |
3860 | /* |
3861 | * If we are doing an import, ensure that the pool is not already | |
3862 | * imported by checking if its pool guid already exists in the | |
3863 | * spa namespace. | |
3864 | * | |
3865 | * The only case that we allow an already imported pool to be | |
3866 | * imported again, is when the pool is checkpointed and we want to | |
3867 | * look at its checkpointed state from userland tools like zdb. | |
3868 | */ | |
3869 | #ifdef _KERNEL | |
3870 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3871 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3872 | spa_guid_exists(pool_guid, 0)) { | |
3873 | #else | |
3874 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3875 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3876 | spa_guid_exists(pool_guid, 0) && | |
3877 | !spa_importing_readonly_checkpoint(spa)) { | |
3878 | #endif | |
6cb8e530 PZ |
3879 | spa_load_failed(spa, "a pool with guid %llu is already open", |
3880 | (u_longlong_t)pool_guid); | |
3881 | return (SET_ERROR(EEXIST)); | |
3882 | } | |
3883 | ||
3884 | spa->spa_config_guid = pool_guid; | |
3885 | ||
3886 | nvlist_free(spa->spa_load_info); | |
3887 | spa->spa_load_info = fnvlist_alloc(); | |
3888 | ||
3889 | ASSERT(spa->spa_comment == NULL); | |
3890 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
3891 | spa->spa_comment = spa_strdup(comment); | |
3892 | ||
658fb802 CB |
3893 | ASSERT(spa->spa_compatibility == NULL); |
3894 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMPATIBILITY, | |
3895 | &compatibility) == 0) | |
3896 | spa->spa_compatibility = spa_strdup(compatibility); | |
3897 | ||
6cb8e530 PZ |
3898 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
3899 | &spa->spa_config_txg); | |
3900 | ||
3901 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) | |
3902 | spa->spa_config_splitting = fnvlist_dup(nvl); | |
428870ff | 3903 | |
4a0ee12a PZ |
3904 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { |
3905 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3906 | ZPOOL_CONFIG_VDEV_TREE); | |
2e528b49 | 3907 | return (SET_ERROR(EINVAL)); |
4a0ee12a | 3908 | } |
428870ff | 3909 | |
428870ff BB |
3910 | /* |
3911 | * Create "The Godfather" zio to hold all async IOs | |
3912 | */ | |
e022864d MA |
3913 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
3914 | KM_SLEEP); | |
1c27024e | 3915 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
3916 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
3917 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
3918 | ZIO_FLAG_GODFATHER); | |
3919 | } | |
428870ff BB |
3920 | |
3921 | /* | |
3922 | * Parse the configuration into a vdev tree. We explicitly set the | |
3923 | * value that will be returned by spa_version() since parsing the | |
3924 | * configuration requires knowing the version number. | |
3925 | */ | |
3926 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 PZ |
3927 | parse = (type == SPA_IMPORT_EXISTING ? |
3928 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
9eb7b46e | 3929 | error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); |
428870ff BB |
3930 | spa_config_exit(spa, SCL_ALL, FTAG); |
3931 | ||
4a0ee12a PZ |
3932 | if (error != 0) { |
3933 | spa_load_failed(spa, "unable to parse config [error=%d]", | |
3934 | error); | |
428870ff | 3935 | return (error); |
4a0ee12a | 3936 | } |
428870ff BB |
3937 | |
3938 | ASSERT(spa->spa_root_vdev == rvd); | |
c3520e7f MA |
3939 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); |
3940 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
428870ff BB |
3941 | |
3942 | if (type != SPA_IMPORT_ASSEMBLE) { | |
3943 | ASSERT(spa_guid(spa) == pool_guid); | |
3944 | } | |
3945 | ||
9eb7b46e PZ |
3946 | return (0); |
3947 | } | |
3948 | ||
6cb8e530 PZ |
3949 | /* |
3950 | * Recursively open all vdevs in the vdev tree. This function is called twice: | |
3951 | * first with the untrusted config, then with the trusted config. | |
3952 | */ | |
9eb7b46e PZ |
3953 | static int |
3954 | spa_ld_open_vdevs(spa_t *spa) | |
3955 | { | |
3956 | int error = 0; | |
3957 | ||
6cb8e530 PZ |
3958 | /* |
3959 | * spa_missing_tvds_allowed defines how many top-level vdevs can be | |
3960 | * missing/unopenable for the root vdev to be still considered openable. | |
3961 | */ | |
3962 | if (spa->spa_trust_config) { | |
3963 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; | |
3964 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { | |
3965 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; | |
3966 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { | |
3967 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; | |
3968 | } else { | |
3969 | spa->spa_missing_tvds_allowed = 0; | |
3970 | } | |
3971 | ||
3972 | spa->spa_missing_tvds_allowed = | |
3973 | MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); | |
3974 | ||
428870ff | 3975 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9eb7b46e | 3976 | error = vdev_open(spa->spa_root_vdev); |
428870ff | 3977 | spa_config_exit(spa, SCL_ALL, FTAG); |
6cb8e530 PZ |
3978 | |
3979 | if (spa->spa_missing_tvds != 0) { | |
3980 | spa_load_note(spa, "vdev tree has %lld missing top-level " | |
3981 | "vdevs.", (u_longlong_t)spa->spa_missing_tvds); | |
da92d5cb | 3982 | if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) { |
6cb8e530 PZ |
3983 | /* |
3984 | * Although theoretically we could allow users to open | |
3985 | * incomplete pools in RW mode, we'd need to add a lot | |
3986 | * of extra logic (e.g. adjust pool space to account | |
3987 | * for missing vdevs). | |
3988 | * This limitation also prevents users from accidentally | |
3989 | * opening the pool in RW mode during data recovery and | |
3990 | * damaging it further. | |
3991 | */ | |
3992 | spa_load_note(spa, "pools with missing top-level " | |
3993 | "vdevs can only be opened in read-only mode."); | |
3994 | error = SET_ERROR(ENXIO); | |
3995 | } else { | |
3996 | spa_load_note(spa, "current settings allow for maximum " | |
3997 | "%lld missing top-level vdevs at this stage.", | |
3998 | (u_longlong_t)spa->spa_missing_tvds_allowed); | |
3999 | } | |
4000 | } | |
4a0ee12a PZ |
4001 | if (error != 0) { |
4002 | spa_load_failed(spa, "unable to open vdev tree [error=%d]", | |
4003 | error); | |
4004 | } | |
6cb8e530 PZ |
4005 | if (spa->spa_missing_tvds != 0 || error != 0) |
4006 | vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); | |
9eb7b46e PZ |
4007 | |
4008 | return (error); | |
4009 | } | |
4010 | ||
6cb8e530 PZ |
4011 | /* |
4012 | * We need to validate the vdev labels against the configuration that | |
4013 | * we have in hand. This function is called twice: first with an untrusted | |
4014 | * config, then with a trusted config. The validation is more strict when the | |
4015 | * config is trusted. | |
4016 | */ | |
9eb7b46e | 4017 | static int |
6cb8e530 | 4018 | spa_ld_validate_vdevs(spa_t *spa) |
9eb7b46e PZ |
4019 | { |
4020 | int error = 0; | |
4021 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 4022 | |
6cb8e530 PZ |
4023 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
4024 | error = vdev_validate(rvd); | |
4025 | spa_config_exit(spa, SCL_ALL, FTAG); | |
428870ff | 4026 | |
6cb8e530 PZ |
4027 | if (error != 0) { |
4028 | spa_load_failed(spa, "vdev_validate failed [error=%d]", error); | |
4029 | return (error); | |
4030 | } | |
428870ff | 4031 | |
6cb8e530 PZ |
4032 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
4033 | spa_load_failed(spa, "cannot open vdev tree after invalidating " | |
4034 | "some vdevs"); | |
4035 | vdev_dbgmsg_print_tree(rvd, 2); | |
4036 | return (SET_ERROR(ENXIO)); | |
428870ff BB |
4037 | } |
4038 | ||
9eb7b46e PZ |
4039 | return (0); |
4040 | } | |
4041 | ||
d2734cce SD |
4042 | static void |
4043 | spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) | |
4044 | { | |
4045 | spa->spa_state = POOL_STATE_ACTIVE; | |
4046 | spa->spa_ubsync = spa->spa_uberblock; | |
4047 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
4048 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
4049 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
4050 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
4051 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
4052 | spa->spa_prev_software_version = ub->ub_software_version; | |
4053 | } | |
4054 | ||
9eb7b46e | 4055 | static int |
6cb8e530 | 4056 | spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) |
9eb7b46e PZ |
4057 | { |
4058 | vdev_t *rvd = spa->spa_root_vdev; | |
4059 | nvlist_t *label; | |
4060 | uberblock_t *ub = &spa->spa_uberblock; | |
9eb7b46e PZ |
4061 | boolean_t activity_check = B_FALSE; |
4062 | ||
d2734cce SD |
4063 | /* |
4064 | * If we are opening the checkpointed state of the pool by | |
4065 | * rewinding to it, at this point we will have written the | |
4066 | * checkpointed uberblock to the vdev labels, so searching | |
4067 | * the labels will find the right uberblock. However, if | |
4068 | * we are opening the checkpointed state read-only, we have | |
4069 | * not modified the labels. Therefore, we must ignore the | |
4070 | * labels and continue using the spa_uberblock that was set | |
4071 | * by spa_ld_checkpoint_rewind. | |
4072 | * | |
4073 | * Note that it would be fine to ignore the labels when | |
4074 | * rewinding (opening writeable) as well. However, if we | |
4075 | * crash just after writing the labels, we will end up | |
4076 | * searching the labels. Doing so in the common case means | |
4077 | * that this code path gets exercised normally, rather than | |
4078 | * just in the edge case. | |
4079 | */ | |
4080 | if (ub->ub_checkpoint_txg != 0 && | |
4081 | spa_importing_readonly_checkpoint(spa)) { | |
4082 | spa_ld_select_uberblock_done(spa, ub); | |
4083 | return (0); | |
4084 | } | |
4085 | ||
428870ff BB |
4086 | /* |
4087 | * Find the best uberblock. | |
4088 | */ | |
9ae529ec | 4089 | vdev_uberblock_load(rvd, ub, &label); |
428870ff BB |
4090 | |
4091 | /* | |
4092 | * If we weren't able to find a single valid uberblock, return failure. | |
4093 | */ | |
9ae529ec CS |
4094 | if (ub->ub_txg == 0) { |
4095 | nvlist_free(label); | |
4a0ee12a | 4096 | spa_load_failed(spa, "no valid uberblock found"); |
428870ff | 4097 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); |
9ae529ec | 4098 | } |
428870ff | 4099 | |
ca95f70d OF |
4100 | if (spa->spa_load_max_txg != UINT64_MAX) { |
4101 | (void) spa_import_progress_set_max_txg(spa_guid(spa), | |
4102 | (u_longlong_t)spa->spa_load_max_txg); | |
4103 | } | |
4a0ee12a PZ |
4104 | spa_load_note(spa, "using uberblock with txg=%llu", |
4105 | (u_longlong_t)ub->ub_txg); | |
5caeef02 DB |
4106 | if (ub->ub_raidz_reflow_info != 0) { |
4107 | spa_load_note(spa, "uberblock raidz_reflow_info: " | |
4108 | "state=%u offset=%llu", | |
4109 | (int)RRSS_GET_STATE(ub), | |
4110 | (u_longlong_t)RRSS_GET_OFFSET(ub)); | |
4111 | } | |
4a0ee12a PZ |
4112 | |
4113 | ||
379ca9cf OF |
4114 | /* |
4115 | * For pools which have the multihost property on determine if the | |
4116 | * pool is truly inactive and can be safely imported. Prevent | |
4117 | * hosts which don't have a hostid set from importing the pool. | |
4118 | */ | |
6cb8e530 PZ |
4119 | activity_check = spa_activity_check_required(spa, ub, label, |
4120 | spa->spa_config); | |
379ca9cf | 4121 | if (activity_check) { |
379ca9cf | 4122 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && |
25f06d67 | 4123 | spa_get_hostid(spa) == 0) { |
379ca9cf OF |
4124 | nvlist_free(label); |
4125 | fnvlist_add_uint64(spa->spa_load_info, | |
4126 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4127 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4128 | } | |
4129 | ||
c3f2f1aa DB |
4130 | int error = |
4131 | spa_activity_check(spa, ub, spa->spa_config, B_TRUE); | |
e889f0f5 OF |
4132 | if (error) { |
4133 | nvlist_free(label); | |
4134 | return (error); | |
4135 | } | |
4136 | ||
379ca9cf OF |
4137 | fnvlist_add_uint64(spa->spa_load_info, |
4138 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
4139 | fnvlist_add_uint64(spa->spa_load_info, | |
4140 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
060f0226 OF |
4141 | fnvlist_add_uint16(spa->spa_load_info, |
4142 | ZPOOL_CONFIG_MMP_SEQ, | |
4143 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)); | |
379ca9cf OF |
4144 | } |
4145 | ||
428870ff | 4146 | /* |
9ae529ec | 4147 | * If the pool has an unsupported version we can't open it. |
428870ff | 4148 | */ |
9ae529ec CS |
4149 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { |
4150 | nvlist_free(label); | |
4a0ee12a PZ |
4151 | spa_load_failed(spa, "version %llu is not supported", |
4152 | (u_longlong_t)ub->ub_version); | |
428870ff | 4153 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); |
9ae529ec CS |
4154 | } |
4155 | ||
4156 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
4157 | nvlist_t *features; | |
4158 | ||
4159 | /* | |
4160 | * If we weren't able to find what's necessary for reading the | |
4161 | * MOS in the label, return failure. | |
4162 | */ | |
4a0ee12a PZ |
4163 | if (label == NULL) { |
4164 | spa_load_failed(spa, "label config unavailable"); | |
4165 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
4166 | ENXIO)); | |
4167 | } | |
4168 | ||
4169 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, | |
4170 | &features) != 0) { | |
9ae529ec | 4171 | nvlist_free(label); |
4a0ee12a PZ |
4172 | spa_load_failed(spa, "invalid label: '%s' missing", |
4173 | ZPOOL_CONFIG_FEATURES_FOR_READ); | |
9ae529ec CS |
4174 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
4175 | ENXIO)); | |
4176 | } | |
4177 | ||
4178 | /* | |
4179 | * Update our in-core representation with the definitive values | |
4180 | * from the label. | |
4181 | */ | |
4182 | nvlist_free(spa->spa_label_features); | |
65ad5d11 | 4183 | spa->spa_label_features = fnvlist_dup(features); |
9ae529ec CS |
4184 | } |
4185 | ||
4186 | nvlist_free(label); | |
4187 | ||
4188 | /* | |
4189 | * Look through entries in the label nvlist's features_for_read. If | |
4190 | * there is a feature listed there which we don't understand then we | |
4191 | * cannot open a pool. | |
4192 | */ | |
4193 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
4194 | nvlist_t *unsup_feat; | |
9ae529ec | 4195 | |
65ad5d11 | 4196 | unsup_feat = fnvlist_alloc(); |
9ae529ec | 4197 | |
1c27024e DB |
4198 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, |
4199 | NULL); nvp != NULL; | |
9ae529ec CS |
4200 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { |
4201 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
65ad5d11 AJ |
4202 | fnvlist_add_string(unsup_feat, |
4203 | nvpair_name(nvp), ""); | |
9ae529ec CS |
4204 | } |
4205 | } | |
4206 | ||
4207 | if (!nvlist_empty(unsup_feat)) { | |
65ad5d11 AJ |
4208 | fnvlist_add_nvlist(spa->spa_load_info, |
4209 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec | 4210 | nvlist_free(unsup_feat); |
4a0ee12a | 4211 | spa_load_failed(spa, "some features are unsupported"); |
9ae529ec CS |
4212 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
4213 | ENOTSUP)); | |
4214 | } | |
4215 | ||
4216 | nvlist_free(unsup_feat); | |
4217 | } | |
428870ff | 4218 | |
428870ff BB |
4219 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { |
4220 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 | 4221 | spa_try_repair(spa, spa->spa_config); |
428870ff BB |
4222 | spa_config_exit(spa, SCL_ALL, FTAG); |
4223 | nvlist_free(spa->spa_config_splitting); | |
4224 | spa->spa_config_splitting = NULL; | |
4225 | } | |
4226 | ||
4227 | /* | |
4228 | * Initialize internal SPA structures. | |
4229 | */ | |
d2734cce | 4230 | spa_ld_select_uberblock_done(spa, ub); |
428870ff | 4231 | |
9eb7b46e PZ |
4232 | return (0); |
4233 | } | |
4234 | ||
4235 | static int | |
4236 | spa_ld_open_rootbp(spa_t *spa) | |
4237 | { | |
4238 | int error = 0; | |
4239 | vdev_t *rvd = spa->spa_root_vdev; | |
a1d477c2 | 4240 | |
9ae529ec | 4241 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
4a0ee12a PZ |
4242 | if (error != 0) { |
4243 | spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " | |
4244 | "[error=%d]", error); | |
428870ff | 4245 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4246 | } |
428870ff BB |
4247 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
4248 | ||
9eb7b46e PZ |
4249 | return (0); |
4250 | } | |
4251 | ||
4252 | static int | |
d2734cce | 4253 | spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, |
6cb8e530 | 4254 | boolean_t reloading) |
9eb7b46e | 4255 | { |
6cb8e530 PZ |
4256 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; |
4257 | nvlist_t *nv, *mos_config, *policy; | |
4258 | int error = 0, copy_error; | |
4259 | uint64_t healthy_tvds, healthy_tvds_mos; | |
4260 | uint64_t mos_config_txg; | |
9eb7b46e | 4261 | |
4a0ee12a PZ |
4262 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) |
4263 | != 0) | |
428870ff BB |
4264 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4265 | ||
a1d477c2 | 4266 | /* |
6cb8e530 PZ |
4267 | * If we're assembling a pool from a split, the config provided is |
4268 | * already trusted so there is nothing to do. | |
a1d477c2 | 4269 | */ |
6cb8e530 PZ |
4270 | if (type == SPA_IMPORT_ASSEMBLE) |
4271 | return (0); | |
4272 | ||
4273 | healthy_tvds = spa_healthy_core_tvds(spa); | |
a1d477c2 | 4274 | |
6cb8e530 PZ |
4275 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) |
4276 | != 0) { | |
4277 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
4278 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4279 | } | |
4280 | ||
4281 | /* | |
4282 | * If we are doing an open, pool owner wasn't verified yet, thus do | |
4283 | * the verification here. | |
4284 | */ | |
4285 | if (spa->spa_load_state == SPA_LOAD_OPEN) { | |
4286 | error = spa_verify_host(spa, mos_config); | |
4287 | if (error != 0) { | |
a1d477c2 | 4288 | nvlist_free(mos_config); |
6cb8e530 | 4289 | return (error); |
a1d477c2 | 4290 | } |
6cb8e530 PZ |
4291 | } |
4292 | ||
4293 | nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); | |
a1d477c2 | 4294 | |
6cb8e530 PZ |
4295 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
4296 | ||
4297 | /* | |
4298 | * Build a new vdev tree from the trusted config | |
4299 | */ | |
b2255edc BB |
4300 | error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD); |
4301 | if (error != 0) { | |
4302 | nvlist_free(mos_config); | |
4303 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4304 | spa_load_failed(spa, "spa_config_parse failed [error=%d]", | |
4305 | error); | |
4306 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4307 | } | |
6cb8e530 PZ |
4308 | |
4309 | /* | |
4310 | * Vdev paths in the MOS may be obsolete. If the untrusted config was | |
4311 | * obtained by scanning /dev/dsk, then it will have the right vdev | |
4312 | * paths. We update the trusted MOS config with this information. | |
4313 | * We first try to copy the paths with vdev_copy_path_strict, which | |
4314 | * succeeds only when both configs have exactly the same vdev tree. | |
4315 | * If that fails, we fall back to a more flexible method that has a | |
4316 | * best effort policy. | |
4317 | */ | |
4318 | copy_error = vdev_copy_path_strict(rvd, mrvd); | |
4319 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
4320 | spa_load_note(spa, "provided vdev tree:"); | |
4321 | vdev_dbgmsg_print_tree(rvd, 2); | |
4322 | spa_load_note(spa, "MOS vdev tree:"); | |
4323 | vdev_dbgmsg_print_tree(mrvd, 2); | |
4324 | } | |
4325 | if (copy_error != 0) { | |
4326 | spa_load_note(spa, "vdev_copy_path_strict failed, falling " | |
4327 | "back to vdev_copy_path_relaxed"); | |
4328 | vdev_copy_path_relaxed(rvd, mrvd); | |
4329 | } | |
4330 | ||
4331 | vdev_close(rvd); | |
4332 | vdev_free(rvd); | |
4333 | spa->spa_root_vdev = mrvd; | |
4334 | rvd = mrvd; | |
4335 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4336 | ||
54b1b1d8 RN |
4337 | /* |
4338 | * If 'zpool import' used a cached config, then the on-disk hostid and | |
4339 | * hostname may be different to the cached config in ways that should | |
4340 | * prevent import. Userspace can't discover this without a scan, but | |
4341 | * we know, so we add these values to LOAD_INFO so the caller can know | |
4342 | * the difference. | |
4343 | * | |
4344 | * Note that we have to do this before the config is regenerated, | |
4345 | * because the new config will have the hostid and hostname for this | |
4346 | * host, in readiness for import. | |
4347 | */ | |
4348 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTID)) | |
4349 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_HOSTID, | |
4350 | fnvlist_lookup_uint64(mos_config, ZPOOL_CONFIG_HOSTID)); | |
4351 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTNAME)) | |
4352 | fnvlist_add_string(spa->spa_load_info, ZPOOL_CONFIG_HOSTNAME, | |
4353 | fnvlist_lookup_string(mos_config, ZPOOL_CONFIG_HOSTNAME)); | |
4354 | ||
6cb8e530 PZ |
4355 | /* |
4356 | * We will use spa_config if we decide to reload the spa or if spa_load | |
4357 | * fails and we rewind. We must thus regenerate the config using the | |
8a393be3 PZ |
4358 | * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to |
4359 | * pass settings on how to load the pool and is not stored in the MOS. | |
4360 | * We copy it over to our new, trusted config. | |
6cb8e530 PZ |
4361 | */ |
4362 | mos_config_txg = fnvlist_lookup_uint64(mos_config, | |
4363 | ZPOOL_CONFIG_POOL_TXG); | |
4364 | nvlist_free(mos_config); | |
4365 | mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); | |
8a393be3 | 4366 | if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, |
6cb8e530 | 4367 | &policy) == 0) |
8a393be3 | 4368 | fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); |
6cb8e530 PZ |
4369 | spa_config_set(spa, mos_config); |
4370 | spa->spa_config_source = SPA_CONFIG_SRC_MOS; | |
4371 | ||
4372 | /* | |
4373 | * Now that we got the config from the MOS, we should be more strict | |
4374 | * in checking blkptrs and can make assumptions about the consistency | |
4375 | * of the vdev tree. spa_trust_config must be set to true before opening | |
4376 | * vdevs in order for them to be writeable. | |
4377 | */ | |
4378 | spa->spa_trust_config = B_TRUE; | |
4379 | ||
4380 | /* | |
4381 | * Open and validate the new vdev tree | |
4382 | */ | |
4383 | error = spa_ld_open_vdevs(spa); | |
4384 | if (error != 0) | |
4385 | return (error); | |
4386 | ||
4387 | error = spa_ld_validate_vdevs(spa); | |
4388 | if (error != 0) | |
4389 | return (error); | |
4390 | ||
4391 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
4392 | spa_load_note(spa, "final vdev tree:"); | |
4393 | vdev_dbgmsg_print_tree(rvd, 2); | |
4394 | } | |
4395 | ||
4396 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && | |
4397 | !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { | |
a1d477c2 | 4398 | /* |
6cb8e530 PZ |
4399 | * Sanity check to make sure that we are indeed loading the |
4400 | * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds | |
4401 | * in the config provided and they happened to be the only ones | |
4402 | * to have the latest uberblock, we could involuntarily perform | |
4403 | * an extreme rewind. | |
a1d477c2 | 4404 | */ |
6cb8e530 PZ |
4405 | healthy_tvds_mos = spa_healthy_core_tvds(spa); |
4406 | if (healthy_tvds_mos - healthy_tvds >= | |
4407 | SPA_SYNC_MIN_VDEVS) { | |
4408 | spa_load_note(spa, "config provided misses too many " | |
4409 | "top-level vdevs compared to MOS (%lld vs %lld). ", | |
4410 | (u_longlong_t)healthy_tvds, | |
4411 | (u_longlong_t)healthy_tvds_mos); | |
4412 | spa_load_note(spa, "vdev tree:"); | |
4413 | vdev_dbgmsg_print_tree(rvd, 2); | |
4414 | if (reloading) { | |
4415 | spa_load_failed(spa, "config was already " | |
4416 | "provided from MOS. Aborting."); | |
4417 | return (spa_vdev_err(rvd, | |
4418 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4419 | } | |
4420 | spa_load_note(spa, "spa must be reloaded using MOS " | |
4421 | "config"); | |
4422 | return (SET_ERROR(EAGAIN)); | |
4a0ee12a | 4423 | } |
a1d477c2 MA |
4424 | } |
4425 | ||
6cb8e530 PZ |
4426 | error = spa_check_for_missing_logs(spa); |
4427 | if (error != 0) | |
4428 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
4429 | ||
4430 | if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { | |
4431 | spa_load_failed(spa, "uberblock guid sum doesn't match MOS " | |
4432 | "guid sum (%llu != %llu)", | |
4433 | (u_longlong_t)spa->spa_uberblock.ub_guid_sum, | |
4434 | (u_longlong_t)rvd->vdev_guid_sum); | |
4435 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
4436 | ENXIO)); | |
4437 | } | |
4438 | ||
9eb7b46e PZ |
4439 | return (0); |
4440 | } | |
4441 | ||
4442 | static int | |
4443 | spa_ld_open_indirect_vdev_metadata(spa_t *spa) | |
4444 | { | |
4445 | int error = 0; | |
4446 | vdev_t *rvd = spa->spa_root_vdev; | |
4447 | ||
a1d477c2 MA |
4448 | /* |
4449 | * Everything that we read before spa_remove_init() must be stored | |
4450 | * on concreted vdevs. Therefore we do this as early as possible. | |
4451 | */ | |
4a0ee12a PZ |
4452 | error = spa_remove_init(spa); |
4453 | if (error != 0) { | |
4454 | spa_load_failed(spa, "spa_remove_init failed [error=%d]", | |
4455 | error); | |
a1d477c2 | 4456 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4457 | } |
a1d477c2 | 4458 | |
9eb7b46e PZ |
4459 | /* |
4460 | * Retrieve information needed to condense indirect vdev mappings. | |
4461 | */ | |
4462 | error = spa_condense_init(spa); | |
4463 | if (error != 0) { | |
4a0ee12a PZ |
4464 | spa_load_failed(spa, "spa_condense_init failed [error=%d]", |
4465 | error); | |
9eb7b46e PZ |
4466 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
4467 | } | |
4468 | ||
4469 | return (0); | |
4470 | } | |
4471 | ||
4472 | static int | |
4a0ee12a | 4473 | spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) |
9eb7b46e PZ |
4474 | { |
4475 | int error = 0; | |
4476 | vdev_t *rvd = spa->spa_root_vdev; | |
4477 | ||
9ae529ec CS |
4478 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { |
4479 | boolean_t missing_feat_read = B_FALSE; | |
b9b24bb4 | 4480 | nvlist_t *unsup_feat, *enabled_feat; |
9ae529ec CS |
4481 | |
4482 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
4a0ee12a | 4483 | &spa->spa_feat_for_read_obj, B_TRUE) != 0) { |
9ae529ec CS |
4484 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4485 | } | |
4486 | ||
4487 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
4a0ee12a | 4488 | &spa->spa_feat_for_write_obj, B_TRUE) != 0) { |
9ae529ec CS |
4489 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4490 | } | |
4491 | ||
4492 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
4a0ee12a | 4493 | &spa->spa_feat_desc_obj, B_TRUE) != 0) { |
9ae529ec CS |
4494 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4495 | } | |
4496 | ||
b9b24bb4 CS |
4497 | enabled_feat = fnvlist_alloc(); |
4498 | unsup_feat = fnvlist_alloc(); | |
9ae529ec | 4499 | |
fa86b5db | 4500 | if (!spa_features_check(spa, B_FALSE, |
b9b24bb4 | 4501 | unsup_feat, enabled_feat)) |
9ae529ec CS |
4502 | missing_feat_read = B_TRUE; |
4503 | ||
4a0ee12a PZ |
4504 | if (spa_writeable(spa) || |
4505 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) { | |
fa86b5db | 4506 | if (!spa_features_check(spa, B_TRUE, |
b9b24bb4 | 4507 | unsup_feat, enabled_feat)) { |
9eb7b46e | 4508 | *missing_feat_writep = B_TRUE; |
b9b24bb4 | 4509 | } |
9ae529ec CS |
4510 | } |
4511 | ||
b9b24bb4 CS |
4512 | fnvlist_add_nvlist(spa->spa_load_info, |
4513 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
4514 | ||
9ae529ec | 4515 | if (!nvlist_empty(unsup_feat)) { |
b9b24bb4 CS |
4516 | fnvlist_add_nvlist(spa->spa_load_info, |
4517 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec CS |
4518 | } |
4519 | ||
b9b24bb4 CS |
4520 | fnvlist_free(enabled_feat); |
4521 | fnvlist_free(unsup_feat); | |
9ae529ec CS |
4522 | |
4523 | if (!missing_feat_read) { | |
4524 | fnvlist_add_boolean(spa->spa_load_info, | |
4525 | ZPOOL_CONFIG_CAN_RDONLY); | |
4526 | } | |
4527 | ||
4528 | /* | |
4529 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
4530 | * twofold: to determine whether the pool is available for | |
4531 | * import in read-write mode and (if it is not) whether the | |
4532 | * pool is available for import in read-only mode. If the pool | |
4533 | * is available for import in read-write mode, it is displayed | |
4534 | * as available in userland; if it is not available for import | |
4535 | * in read-only mode, it is displayed as unavailable in | |
4536 | * userland. If the pool is available for import in read-only | |
4537 | * mode but not read-write mode, it is displayed as unavailable | |
4538 | * in userland with a special note that the pool is actually | |
4539 | * available for open in read-only mode. | |
4540 | * | |
4541 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
4542 | * missing a feature for write, we must first determine whether | |
4543 | * the pool can be opened read-only before returning to | |
4544 | * userland in order to know whether to display the | |
4545 | * abovementioned note. | |
4546 | */ | |
9eb7b46e | 4547 | if (missing_feat_read || (*missing_feat_writep && |
9ae529ec | 4548 | spa_writeable(spa))) { |
4a0ee12a | 4549 | spa_load_failed(spa, "pool uses unsupported features"); |
9ae529ec CS |
4550 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
4551 | ENOTSUP)); | |
4552 | } | |
b0bc7a84 MG |
4553 | |
4554 | /* | |
4555 | * Load refcounts for ZFS features from disk into an in-memory | |
4556 | * cache during SPA initialization. | |
4557 | */ | |
1c27024e | 4558 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { |
b0bc7a84 MG |
4559 | uint64_t refcount; |
4560 | ||
4561 | error = feature_get_refcount_from_disk(spa, | |
4562 | &spa_feature_table[i], &refcount); | |
4563 | if (error == 0) { | |
4564 | spa->spa_feat_refcount_cache[i] = refcount; | |
4565 | } else if (error == ENOTSUP) { | |
4566 | spa->spa_feat_refcount_cache[i] = | |
4567 | SPA_FEATURE_DISABLED; | |
4568 | } else { | |
4a0ee12a PZ |
4569 | spa_load_failed(spa, "error getting refcount " |
4570 | "for feature %s [error=%d]", | |
4571 | spa_feature_table[i].fi_guid, error); | |
b0bc7a84 MG |
4572 | return (spa_vdev_err(rvd, |
4573 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4574 | } | |
4575 | } | |
4576 | } | |
4577 | ||
4578 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
4579 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
4a0ee12a | 4580 | &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) |
b0bc7a84 | 4581 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
9ae529ec CS |
4582 | } |
4583 | ||
f00ab3f2 TC |
4584 | /* |
4585 | * Encryption was added before bookmark_v2, even though bookmark_v2 | |
4586 | * is now a dependency. If this pool has encryption enabled without | |
4587 | * bookmark_v2, trigger an errata message. | |
4588 | */ | |
4589 | if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) && | |
4590 | !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) { | |
4591 | spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; | |
4592 | } | |
4593 | ||
9eb7b46e PZ |
4594 | return (0); |
4595 | } | |
4596 | ||
4597 | static int | |
4598 | spa_ld_load_special_directories(spa_t *spa) | |
4599 | { | |
4600 | int error = 0; | |
4601 | vdev_t *rvd = spa->spa_root_vdev; | |
4602 | ||
9ae529ec CS |
4603 | spa->spa_is_initializing = B_TRUE; |
4604 | error = dsl_pool_open(spa->spa_dsl_pool); | |
4605 | spa->spa_is_initializing = B_FALSE; | |
4a0ee12a PZ |
4606 | if (error != 0) { |
4607 | spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); | |
9ae529ec | 4608 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4609 | } |
9ae529ec | 4610 | |
9eb7b46e PZ |
4611 | return (0); |
4612 | } | |
428870ff | 4613 | |
9eb7b46e PZ |
4614 | static int |
4615 | spa_ld_get_props(spa_t *spa) | |
4616 | { | |
4617 | int error = 0; | |
4618 | uint64_t obj; | |
4619 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4620 | |
3c67d83a TH |
4621 | /* Grab the checksum salt from the MOS. */ |
4622 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4623 | DMU_POOL_CHECKSUM_SALT, 1, | |
4624 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
4625 | spa->spa_cksum_salt.zcs_bytes); | |
4626 | if (error == ENOENT) { | |
4627 | /* Generate a new salt for subsequent use */ | |
4628 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4629 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4630 | } else if (error != 0) { | |
4a0ee12a PZ |
4631 | spa_load_failed(spa, "unable to retrieve checksum salt from " |
4632 | "MOS [error=%d]", error); | |
3c67d83a TH |
4633 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4634 | } | |
4635 | ||
4a0ee12a | 4636 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) |
428870ff BB |
4637 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4638 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
4a0ee12a PZ |
4639 | if (error != 0) { |
4640 | spa_load_failed(spa, "error opening deferred-frees bpobj " | |
4641 | "[error=%d]", error); | |
428870ff | 4642 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4643 | } |
34dc7c2f BB |
4644 | |
4645 | /* | |
4646 | * Load the bit that tells us to use the new accounting function | |
4647 | * (raid-z deflation). If we have an older pool, this will not | |
4648 | * be present. | |
4649 | */ | |
4a0ee12a | 4650 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); |
428870ff BB |
4651 | if (error != 0 && error != ENOENT) |
4652 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4653 | ||
4654 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
4a0ee12a | 4655 | &spa->spa_creation_version, B_FALSE); |
428870ff BB |
4656 | if (error != 0 && error != ENOENT) |
4657 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
4658 | |
4659 | /* | |
4660 | * Load the persistent error log. If we have an older pool, this will | |
4661 | * not be present. | |
4662 | */ | |
4a0ee12a PZ |
4663 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, |
4664 | B_FALSE); | |
428870ff BB |
4665 | if (error != 0 && error != ENOENT) |
4666 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4667 | |
428870ff | 4668 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, |
4a0ee12a | 4669 | &spa->spa_errlog_scrub, B_FALSE); |
428870ff BB |
4670 | if (error != 0 && error != ENOENT) |
4671 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4672 | |
37f03da8 SH |
4673 | /* |
4674 | * Load the livelist deletion field. If a livelist is queued for | |
4675 | * deletion, indicate that in the spa | |
4676 | */ | |
4677 | error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES, | |
4678 | &spa->spa_livelists_to_delete, B_FALSE); | |
4679 | if (error != 0 && error != ENOENT) | |
4680 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4681 | ||
34dc7c2f BB |
4682 | /* |
4683 | * Load the history object. If we have an older pool, this | |
4684 | * will not be present. | |
4685 | */ | |
4a0ee12a | 4686 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); |
428870ff BB |
4687 | if (error != 0 && error != ENOENT) |
4688 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4689 | ||
e0ab3ab5 JS |
4690 | /* |
4691 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
4692 | * be present; in this case, defer its creation to a later time to | |
4693 | * avoid dirtying the MOS this early / out of sync context. See | |
4694 | * spa_sync_config_object. | |
4695 | */ | |
4696 | ||
4697 | /* The sentinel is only available in the MOS config. */ | |
1c27024e | 4698 | nvlist_t *mos_config; |
4a0ee12a PZ |
4699 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { |
4700 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
e0ab3ab5 | 4701 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4702 | } |
e0ab3ab5 JS |
4703 | |
4704 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
4a0ee12a | 4705 | &spa->spa_all_vdev_zaps, B_FALSE); |
e0ab3ab5 | 4706 | |
38640550 DB |
4707 | if (error == ENOENT) { |
4708 | VERIFY(!nvlist_exists(mos_config, | |
4709 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
4710 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
4711 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4712 | } else if (error != 0) { | |
cb01da68 | 4713 | nvlist_free(mos_config); |
e0ab3ab5 | 4714 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
38640550 | 4715 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { |
e0ab3ab5 JS |
4716 | /* |
4717 | * An older version of ZFS overwrote the sentinel value, so | |
4718 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
4719 | * destruction to later; see spa_sync_config_object. | |
4720 | */ | |
4721 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
4722 | /* | |
4723 | * We're assuming that no vdevs have had their ZAPs created | |
4724 | * before this. Better be sure of it. | |
4725 | */ | |
4726 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4727 | } | |
4728 | nvlist_free(mos_config); | |
4729 | ||
9eb7b46e PZ |
4730 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); |
4731 | ||
4a0ee12a PZ |
4732 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, |
4733 | B_FALSE); | |
9eb7b46e PZ |
4734 | if (error && error != ENOENT) |
4735 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4736 | ||
4737 | if (error == 0) { | |
da27b8bc | 4738 | uint64_t autoreplace = 0; |
9eb7b46e PZ |
4739 | |
4740 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
4741 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
4742 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
4743 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
4744 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
c02c1bec | 4745 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); |
1b939560 | 4746 | spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim); |
9eb7b46e PZ |
4747 | spa->spa_autoreplace = (autoreplace != 0); |
4748 | } | |
4749 | ||
6cb8e530 PZ |
4750 | /* |
4751 | * If we are importing a pool with missing top-level vdevs, | |
4752 | * we enforce that the pool doesn't panic or get suspended on | |
4753 | * error since the likelihood of missing data is extremely high. | |
4754 | */ | |
4755 | if (spa->spa_missing_tvds > 0 && | |
4756 | spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && | |
4757 | spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4758 | spa_load_note(spa, "forcing failmode to 'continue' " | |
4759 | "as some top level vdevs are missing"); | |
4760 | spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; | |
4761 | } | |
4762 | ||
9eb7b46e PZ |
4763 | return (0); |
4764 | } | |
4765 | ||
4766 | static int | |
4767 | spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) | |
4768 | { | |
4769 | int error = 0; | |
4770 | vdev_t *rvd = spa->spa_root_vdev; | |
4771 | ||
428870ff BB |
4772 | /* |
4773 | * If we're assembling the pool from the split-off vdevs of | |
4774 | * an existing pool, we don't want to attach the spares & cache | |
4775 | * devices. | |
4776 | */ | |
34dc7c2f BB |
4777 | |
4778 | /* | |
4779 | * Load any hot spares for this pool. | |
4780 | */ | |
4a0ee12a PZ |
4781 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, |
4782 | B_FALSE); | |
428870ff BB |
4783 | if (error != 0 && error != ENOENT) |
4784 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4785 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4786 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); |
4787 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
4a0ee12a PZ |
4788 | &spa->spa_spares.sav_config) != 0) { |
4789 | spa_load_failed(spa, "error loading spares nvlist"); | |
428870ff | 4790 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4791 | } |
34dc7c2f | 4792 | |
b128c09f | 4793 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4794 | spa_load_spares(spa); |
b128c09f | 4795 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4796 | } else if (error == 0) { |
4797 | spa->spa_spares.sav_sync = B_TRUE; | |
34dc7c2f BB |
4798 | } |
4799 | ||
4800 | /* | |
4801 | * Load any level 2 ARC devices for this pool. | |
4802 | */ | |
428870ff | 4803 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, |
4a0ee12a | 4804 | &spa->spa_l2cache.sav_object, B_FALSE); |
428870ff BB |
4805 | if (error != 0 && error != ENOENT) |
4806 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4807 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4808 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); |
4809 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
4a0ee12a PZ |
4810 | &spa->spa_l2cache.sav_config) != 0) { |
4811 | spa_load_failed(spa, "error loading l2cache nvlist"); | |
428870ff | 4812 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4813 | } |
34dc7c2f | 4814 | |
b128c09f | 4815 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4816 | spa_load_l2cache(spa); |
b128c09f | 4817 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4818 | } else if (error == 0) { |
4819 | spa->spa_l2cache.sav_sync = B_TRUE; | |
b128c09f BB |
4820 | } |
4821 | ||
9eb7b46e PZ |
4822 | return (0); |
4823 | } | |
428870ff | 4824 | |
9eb7b46e | 4825 | static int |
4a0ee12a | 4826 | spa_ld_load_vdev_metadata(spa_t *spa) |
9eb7b46e PZ |
4827 | { |
4828 | int error = 0; | |
4829 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4830 | |
379ca9cf OF |
4831 | /* |
4832 | * If the 'multihost' property is set, then never allow a pool to | |
4833 | * be imported when the system hostid is zero. The exception to | |
4834 | * this rule is zdb which is always allowed to access pools. | |
4835 | */ | |
25f06d67 | 4836 | if (spa_multihost(spa) && spa_get_hostid(spa) == 0 && |
379ca9cf OF |
4837 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { |
4838 | fnvlist_add_uint64(spa->spa_load_info, | |
4839 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4840 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4841 | } | |
4842 | ||
34dc7c2f BB |
4843 | /* |
4844 | * If the 'autoreplace' property is set, then post a resource notifying | |
4845 | * the ZFS DE that it should not issue any faults for unopenable | |
4846 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
4847 | * unopenable vdevs so that the normal autoreplace handler can take | |
4848 | * over. | |
4849 | */ | |
4a0ee12a | 4850 | if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
34dc7c2f | 4851 | spa_check_removed(spa->spa_root_vdev); |
428870ff BB |
4852 | /* |
4853 | * For the import case, this is done in spa_import(), because | |
4854 | * at this point we're using the spare definitions from | |
4855 | * the MOS config, not necessarily from the userland config. | |
4856 | */ | |
4a0ee12a | 4857 | if (spa->spa_load_state != SPA_LOAD_IMPORT) { |
428870ff BB |
4858 | spa_aux_check_removed(&spa->spa_spares); |
4859 | spa_aux_check_removed(&spa->spa_l2cache); | |
4860 | } | |
4861 | } | |
34dc7c2f BB |
4862 | |
4863 | /* | |
9eb7b46e | 4864 | * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. |
34dc7c2f | 4865 | */ |
a1d477c2 MA |
4866 | error = vdev_load(rvd); |
4867 | if (error != 0) { | |
4a0ee12a | 4868 | spa_load_failed(spa, "vdev_load failed [error=%d]", error); |
a1d477c2 MA |
4869 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
4870 | } | |
4871 | ||
93e28d66 SD |
4872 | error = spa_ld_log_spacemaps(spa); |
4873 | if (error != 0) { | |
600a02b8 | 4874 | spa_load_failed(spa, "spa_ld_log_spacemaps failed [error=%d]", |
93e28d66 SD |
4875 | error); |
4876 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4877 | } | |
4878 | ||
34dc7c2f | 4879 | /* |
9eb7b46e | 4880 | * Propagate the leaf DTLs we just loaded all the way up the vdev tree. |
34dc7c2f | 4881 | */ |
b128c09f | 4882 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9a49d3f3 | 4883 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE); |
b128c09f | 4884 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 4885 | |
9eb7b46e PZ |
4886 | return (0); |
4887 | } | |
4888 | ||
4889 | static int | |
4890 | spa_ld_load_dedup_tables(spa_t *spa) | |
4891 | { | |
4892 | int error = 0; | |
4893 | vdev_t *rvd = spa->spa_root_vdev; | |
4894 | ||
428870ff | 4895 | error = ddt_load(spa); |
4a0ee12a PZ |
4896 | if (error != 0) { |
4897 | spa_load_failed(spa, "ddt_load failed [error=%d]", error); | |
428870ff | 4898 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4899 | } |
428870ff | 4900 | |
9eb7b46e PZ |
4901 | return (0); |
4902 | } | |
4903 | ||
67a1b037 PJD |
4904 | static int |
4905 | spa_ld_load_brt(spa_t *spa) | |
4906 | { | |
4907 | int error = 0; | |
4908 | vdev_t *rvd = spa->spa_root_vdev; | |
4909 | ||
4910 | error = brt_load(spa); | |
4911 | if (error != 0) { | |
4912 | spa_load_failed(spa, "brt_load failed [error=%d]", error); | |
4913 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4914 | } | |
4915 | ||
4916 | return (0); | |
4917 | } | |
4918 | ||
9eb7b46e | 4919 | static int |
a926aab9 | 4920 | spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, const char **ereport) |
9eb7b46e PZ |
4921 | { |
4922 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 4923 | |
4a0ee12a PZ |
4924 | if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { |
4925 | boolean_t missing = spa_check_logs(spa); | |
4926 | if (missing) { | |
6cb8e530 PZ |
4927 | if (spa->spa_missing_tvds != 0) { |
4928 | spa_load_note(spa, "spa_check_logs failed " | |
4929 | "so dropping the logs"); | |
4930 | } else { | |
4931 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
4932 | spa_load_failed(spa, "spa_check_logs failed"); | |
4933 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, | |
4934 | ENXIO)); | |
4935 | } | |
4a0ee12a | 4936 | } |
428870ff BB |
4937 | } |
4938 | ||
9eb7b46e PZ |
4939 | return (0); |
4940 | } | |
4941 | ||
4942 | static int | |
4a0ee12a | 4943 | spa_ld_verify_pool_data(spa_t *spa) |
9eb7b46e PZ |
4944 | { |
4945 | int error = 0; | |
4946 | vdev_t *rvd = spa->spa_root_vdev; | |
4947 | ||
4948 | /* | |
4949 | * We've successfully opened the pool, verify that we're ready | |
4950 | * to start pushing transactions. | |
4951 | */ | |
4a0ee12a | 4952 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
9eb7b46e PZ |
4953 | error = spa_load_verify(spa); |
4954 | if (error != 0) { | |
4a0ee12a PZ |
4955 | spa_load_failed(spa, "spa_load_verify failed " |
4956 | "[error=%d]", error); | |
9eb7b46e PZ |
4957 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
4958 | error)); | |
4959 | } | |
4960 | } | |
4961 | ||
4962 | return (0); | |
4963 | } | |
4964 | ||
4965 | static void | |
4966 | spa_ld_claim_log_blocks(spa_t *spa) | |
4967 | { | |
4968 | dmu_tx_t *tx; | |
4969 | dsl_pool_t *dp = spa_get_dsl(spa); | |
4970 | ||
4971 | /* | |
4972 | * Claim log blocks that haven't been committed yet. | |
4973 | * This must all happen in a single txg. | |
4974 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
4975 | * invoked from zil_claim_log_block()'s i/o done callback. | |
4976 | * Price of rollback is that we abandon the log. | |
4977 | */ | |
4978 | spa->spa_claiming = B_TRUE; | |
4979 | ||
4980 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
4981 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
4982 | zil_claim, tx, DS_FIND_CHILDREN); | |
4983 | dmu_tx_commit(tx); | |
4984 | ||
4985 | spa->spa_claiming = B_FALSE; | |
4986 | ||
4987 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
4988 | } | |
4989 | ||
4990 | static void | |
6cb8e530 | 4991 | spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, |
d2734cce | 4992 | boolean_t update_config_cache) |
9eb7b46e PZ |
4993 | { |
4994 | vdev_t *rvd = spa->spa_root_vdev; | |
4995 | int need_update = B_FALSE; | |
4996 | ||
4997 | /* | |
4998 | * If the config cache is stale, or we have uninitialized | |
4999 | * metaslabs (see spa_vdev_add()), then update the config. | |
5000 | * | |
5001 | * If this is a verbatim import, trust the current | |
5002 | * in-core spa_config and update the disk labels. | |
5003 | */ | |
d2734cce | 5004 | if (update_config_cache || config_cache_txg != spa->spa_config_txg || |
4a0ee12a PZ |
5005 | spa->spa_load_state == SPA_LOAD_IMPORT || |
5006 | spa->spa_load_state == SPA_LOAD_RECOVER || | |
9eb7b46e PZ |
5007 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) |
5008 | need_update = B_TRUE; | |
5009 | ||
5010 | for (int c = 0; c < rvd->vdev_children; c++) | |
5011 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
5012 | need_update = B_TRUE; | |
5013 | ||
5014 | /* | |
e1cfd73f | 5015 | * Update the config cache asynchronously in case we're the |
9eb7b46e PZ |
5016 | * root pool, in which case the config cache isn't writable yet. |
5017 | */ | |
5018 | if (need_update) | |
5019 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
5020 | } | |
5021 | ||
6cb8e530 PZ |
5022 | static void |
5023 | spa_ld_prepare_for_reload(spa_t *spa) | |
5024 | { | |
da92d5cb | 5025 | spa_mode_t mode = spa->spa_mode; |
6cb8e530 PZ |
5026 | int async_suspended = spa->spa_async_suspended; |
5027 | ||
5028 | spa_unload(spa); | |
5029 | spa_deactivate(spa); | |
5030 | spa_activate(spa, mode); | |
5031 | ||
5032 | /* | |
5033 | * We save the value of spa_async_suspended as it gets reset to 0 by | |
5034 | * spa_unload(). We want to restore it back to the original value before | |
5035 | * returning as we might be calling spa_async_resume() later. | |
5036 | */ | |
5037 | spa->spa_async_suspended = async_suspended; | |
5038 | } | |
5039 | ||
9eb7b46e | 5040 | static int |
d2734cce SD |
5041 | spa_ld_read_checkpoint_txg(spa_t *spa) |
5042 | { | |
5043 | uberblock_t checkpoint; | |
5044 | int error = 0; | |
5045 | ||
5046 | ASSERT0(spa->spa_checkpoint_txg); | |
c183d164 GW |
5047 | ASSERT(MUTEX_HELD(&spa_namespace_lock) || |
5048 | spa->spa_load_thread == curthread); | |
d2734cce SD |
5049 | |
5050 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
5051 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
5052 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
5053 | ||
5054 | if (error == ENOENT) | |
5055 | return (0); | |
5056 | ||
5057 | if (error != 0) | |
5058 | return (error); | |
5059 | ||
5060 | ASSERT3U(checkpoint.ub_txg, !=, 0); | |
5061 | ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); | |
5062 | ASSERT3U(checkpoint.ub_timestamp, !=, 0); | |
5063 | spa->spa_checkpoint_txg = checkpoint.ub_txg; | |
5064 | spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; | |
5065 | ||
5066 | return (0); | |
5067 | } | |
5068 | ||
5069 | static int | |
5070 | spa_ld_mos_init(spa_t *spa, spa_import_type_t type) | |
9eb7b46e PZ |
5071 | { |
5072 | int error = 0; | |
9eb7b46e | 5073 | |
4a0ee12a | 5074 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
6cb8e530 | 5075 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); |
4a0ee12a | 5076 | |
9eb7b46e | 5077 | /* |
6cb8e530 PZ |
5078 | * Never trust the config that is provided unless we are assembling |
5079 | * a pool following a split. | |
5080 | * This means don't trust blkptrs and the vdev tree in general. This | |
5081 | * also effectively puts the spa in read-only mode since | |
5082 | * spa_writeable() checks for spa_trust_config to be true. | |
5083 | * We will later load a trusted config from the MOS. | |
9eb7b46e | 5084 | */ |
6cb8e530 PZ |
5085 | if (type != SPA_IMPORT_ASSEMBLE) |
5086 | spa->spa_trust_config = B_FALSE; | |
5087 | ||
9eb7b46e PZ |
5088 | /* |
5089 | * Parse the config provided to create a vdev tree. | |
5090 | */ | |
6cb8e530 | 5091 | error = spa_ld_parse_config(spa, type); |
9eb7b46e PZ |
5092 | if (error != 0) |
5093 | return (error); | |
5094 | ||
ca95f70d OF |
5095 | spa_import_progress_add(spa); |
5096 | ||
9eb7b46e PZ |
5097 | /* |
5098 | * Now that we have the vdev tree, try to open each vdev. This involves | |
5099 | * opening the underlying physical device, retrieving its geometry and | |
5100 | * probing the vdev with a dummy I/O. The state of each vdev will be set | |
5101 | * based on the success of those operations. After this we'll be ready | |
5102 | * to read from the vdevs. | |
5103 | */ | |
5104 | error = spa_ld_open_vdevs(spa); | |
5105 | if (error != 0) | |
5106 | return (error); | |
5107 | ||
5108 | /* | |
5109 | * Read the label of each vdev and make sure that the GUIDs stored | |
5110 | * there match the GUIDs in the config provided. | |
6cb8e530 PZ |
5111 | * If we're assembling a new pool that's been split off from an |
5112 | * existing pool, the labels haven't yet been updated so we skip | |
5113 | * validation for now. | |
9eb7b46e | 5114 | */ |
6cb8e530 PZ |
5115 | if (type != SPA_IMPORT_ASSEMBLE) { |
5116 | error = spa_ld_validate_vdevs(spa); | |
5117 | if (error != 0) | |
5118 | return (error); | |
5119 | } | |
9eb7b46e PZ |
5120 | |
5121 | /* | |
d2734cce SD |
5122 | * Read all vdev labels to find the best uberblock (i.e. latest, |
5123 | * unless spa_load_max_txg is set) and store it in spa_uberblock. We | |
5124 | * get the list of features required to read blkptrs in the MOS from | |
5125 | * the vdev label with the best uberblock and verify that our version | |
5126 | * of zfs supports them all. | |
9eb7b46e | 5127 | */ |
6cb8e530 | 5128 | error = spa_ld_select_uberblock(spa, type); |
9eb7b46e PZ |
5129 | if (error != 0) |
5130 | return (error); | |
5131 | ||
5132 | /* | |
5133 | * Pass that uberblock to the dsl_pool layer which will open the root | |
5134 | * blkptr. This blkptr points to the latest version of the MOS and will | |
5135 | * allow us to read its contents. | |
5136 | */ | |
5137 | error = spa_ld_open_rootbp(spa); | |
5138 | if (error != 0) | |
5139 | return (error); | |
5140 | ||
d2734cce SD |
5141 | return (0); |
5142 | } | |
5143 | ||
5144 | static int | |
5145 | spa_ld_checkpoint_rewind(spa_t *spa) | |
5146 | { | |
5147 | uberblock_t checkpoint; | |
5148 | int error = 0; | |
5149 | ||
5150 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5151 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
5152 | ||
5153 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
5154 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
5155 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
5156 | ||
5157 | if (error != 0) { | |
5158 | spa_load_failed(spa, "unable to retrieve checkpointed " | |
5159 | "uberblock from the MOS config [error=%d]", error); | |
5160 | ||
5161 | if (error == ENOENT) | |
5162 | error = ZFS_ERR_NO_CHECKPOINT; | |
5163 | ||
5164 | return (error); | |
5165 | } | |
5166 | ||
5167 | ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); | |
5168 | ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); | |
5169 | ||
5170 | /* | |
5171 | * We need to update the txg and timestamp of the checkpointed | |
5172 | * uberblock to be higher than the latest one. This ensures that | |
5173 | * the checkpointed uberblock is selected if we were to close and | |
5174 | * reopen the pool right after we've written it in the vdev labels. | |
5175 | * (also see block comment in vdev_uberblock_compare) | |
5176 | */ | |
5177 | checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; | |
5178 | checkpoint.ub_timestamp = gethrestime_sec(); | |
5179 | ||
5180 | /* | |
5181 | * Set current uberblock to be the checkpointed uberblock. | |
5182 | */ | |
5183 | spa->spa_uberblock = checkpoint; | |
5184 | ||
5185 | /* | |
5186 | * If we are doing a normal rewind, then the pool is open for | |
5187 | * writing and we sync the "updated" checkpointed uberblock to | |
5188 | * disk. Once this is done, we've basically rewound the whole | |
5189 | * pool and there is no way back. | |
5190 | * | |
5191 | * There are cases when we don't want to attempt and sync the | |
5192 | * checkpointed uberblock to disk because we are opening a | |
5193 | * pool as read-only. Specifically, verifying the checkpointed | |
5194 | * state with zdb, and importing the checkpointed state to get | |
5195 | * a "preview" of its content. | |
5196 | */ | |
5197 | if (spa_writeable(spa)) { | |
5198 | vdev_t *rvd = spa->spa_root_vdev; | |
5199 | ||
5200 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5201 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; | |
5202 | int svdcount = 0; | |
5203 | int children = rvd->vdev_children; | |
29274c9f | 5204 | int c0 = random_in_range(children); |
d2734cce SD |
5205 | |
5206 | for (int c = 0; c < children; c++) { | |
5207 | vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; | |
5208 | ||
5209 | /* Stop when revisiting the first vdev */ | |
5210 | if (c > 0 && svd[0] == vd) | |
5211 | break; | |
5212 | ||
5213 | if (vd->vdev_ms_array == 0 || vd->vdev_islog || | |
5214 | !vdev_is_concrete(vd)) | |
5215 | continue; | |
5216 | ||
5217 | svd[svdcount++] = vd; | |
5218 | if (svdcount == SPA_SYNC_MIN_VDEVS) | |
5219 | break; | |
5220 | } | |
5221 | error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); | |
5222 | if (error == 0) | |
5223 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
5224 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5225 | ||
5226 | if (error != 0) { | |
5227 | spa_load_failed(spa, "failed to write checkpointed " | |
5228 | "uberblock to the vdev labels [error=%d]", error); | |
5229 | return (error); | |
5230 | } | |
5231 | } | |
5232 | ||
5233 | return (0); | |
5234 | } | |
5235 | ||
5236 | static int | |
5237 | spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, | |
5238 | boolean_t *update_config_cache) | |
5239 | { | |
5240 | int error; | |
5241 | ||
5242 | /* | |
5243 | * Parse the config for pool, open and validate vdevs, | |
5244 | * select an uberblock, and use that uberblock to open | |
5245 | * the MOS. | |
5246 | */ | |
5247 | error = spa_ld_mos_init(spa, type); | |
5248 | if (error != 0) | |
5249 | return (error); | |
5250 | ||
9eb7b46e | 5251 | /* |
6cb8e530 PZ |
5252 | * Retrieve the trusted config stored in the MOS and use it to create |
5253 | * a new, exact version of the vdev tree, then reopen all vdevs. | |
9eb7b46e | 5254 | */ |
d2734cce | 5255 | error = spa_ld_trusted_config(spa, type, B_FALSE); |
6cb8e530 | 5256 | if (error == EAGAIN) { |
d2734cce SD |
5257 | if (update_config_cache != NULL) |
5258 | *update_config_cache = B_TRUE; | |
5259 | ||
6cb8e530 PZ |
5260 | /* |
5261 | * Redo the loading process with the trusted config if it is | |
5262 | * too different from the untrusted config. | |
5263 | */ | |
5264 | spa_ld_prepare_for_reload(spa); | |
d2734cce SD |
5265 | spa_load_note(spa, "RELOADING"); |
5266 | error = spa_ld_mos_init(spa, type); | |
5267 | if (error != 0) | |
5268 | return (error); | |
5269 | ||
5270 | error = spa_ld_trusted_config(spa, type, B_TRUE); | |
5271 | if (error != 0) | |
5272 | return (error); | |
5273 | ||
6cb8e530 | 5274 | } else if (error != 0) { |
9eb7b46e | 5275 | return (error); |
6cb8e530 | 5276 | } |
9eb7b46e | 5277 | |
d2734cce SD |
5278 | return (0); |
5279 | } | |
5280 | ||
5281 | /* | |
5282 | * Load an existing storage pool, using the config provided. This config | |
5283 | * describes which vdevs are part of the pool and is later validated against | |
5284 | * partial configs present in each vdev's label and an entire copy of the | |
5285 | * config stored in the MOS. | |
5286 | */ | |
5287 | static int | |
a926aab9 | 5288 | spa_load_impl(spa_t *spa, spa_import_type_t type, const char **ereport) |
d2734cce SD |
5289 | { |
5290 | int error = 0; | |
5291 | boolean_t missing_feat_write = B_FALSE; | |
5292 | boolean_t checkpoint_rewind = | |
5293 | (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
5294 | boolean_t update_config_cache = B_FALSE; | |
c183d164 | 5295 | hrtime_t load_start = gethrtime(); |
d2734cce SD |
5296 | |
5297 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5298 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); | |
5299 | ||
5300 | spa_load_note(spa, "LOADING"); | |
5301 | ||
5302 | error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); | |
5303 | if (error != 0) | |
5304 | return (error); | |
5305 | ||
5306 | /* | |
5307 | * If we are rewinding to the checkpoint then we need to repeat | |
5308 | * everything we've done so far in this function but this time | |
5309 | * selecting the checkpointed uberblock and using that to open | |
5310 | * the MOS. | |
5311 | */ | |
5312 | if (checkpoint_rewind) { | |
5313 | /* | |
5314 | * If we are rewinding to the checkpoint update config cache | |
5315 | * anyway. | |
5316 | */ | |
5317 | update_config_cache = B_TRUE; | |
5318 | ||
5319 | /* | |
5320 | * Extract the checkpointed uberblock from the current MOS | |
5321 | * and use this as the pool's uberblock from now on. If the | |
5322 | * pool is imported as writeable we also write the checkpoint | |
5323 | * uberblock to the labels, making the rewind permanent. | |
5324 | */ | |
5325 | error = spa_ld_checkpoint_rewind(spa); | |
5326 | if (error != 0) | |
5327 | return (error); | |
5328 | ||
5329 | /* | |
e1cfd73f | 5330 | * Redo the loading process again with the |
d2734cce SD |
5331 | * checkpointed uberblock. |
5332 | */ | |
5333 | spa_ld_prepare_for_reload(spa); | |
5334 | spa_load_note(spa, "LOADING checkpointed uberblock"); | |
5335 | error = spa_ld_mos_with_trusted_config(spa, type, NULL); | |
5336 | if (error != 0) | |
5337 | return (error); | |
5338 | } | |
5339 | ||
c183d164 GW |
5340 | /* |
5341 | * Drop the namespace lock for the rest of the function. | |
5342 | */ | |
5343 | spa->spa_load_thread = curthread; | |
5344 | mutex_exit(&spa_namespace_lock); | |
5345 | ||
d2734cce SD |
5346 | /* |
5347 | * Retrieve the checkpoint txg if the pool has a checkpoint. | |
5348 | */ | |
687e4d7f | 5349 | spa_import_progress_set_notes(spa, "Loading checkpoint txg"); |
d2734cce SD |
5350 | error = spa_ld_read_checkpoint_txg(spa); |
5351 | if (error != 0) | |
c183d164 | 5352 | goto fail; |
d2734cce | 5353 | |
9eb7b46e PZ |
5354 | /* |
5355 | * Retrieve the mapping of indirect vdevs. Those vdevs were removed | |
5356 | * from the pool and their contents were re-mapped to other vdevs. Note | |
5357 | * that everything that we read before this step must have been | |
5358 | * rewritten on concrete vdevs after the last device removal was | |
5359 | * initiated. Otherwise we could be reading from indirect vdevs before | |
5360 | * we have loaded their mappings. | |
5361 | */ | |
687e4d7f | 5362 | spa_import_progress_set_notes(spa, "Loading indirect vdev metadata"); |
9eb7b46e PZ |
5363 | error = spa_ld_open_indirect_vdev_metadata(spa); |
5364 | if (error != 0) | |
c183d164 | 5365 | goto fail; |
9eb7b46e PZ |
5366 | |
5367 | /* | |
5368 | * Retrieve the full list of active features from the MOS and check if | |
5369 | * they are all supported. | |
5370 | */ | |
687e4d7f | 5371 | spa_import_progress_set_notes(spa, "Checking feature flags"); |
4a0ee12a | 5372 | error = spa_ld_check_features(spa, &missing_feat_write); |
9eb7b46e | 5373 | if (error != 0) |
c183d164 | 5374 | goto fail; |
9eb7b46e PZ |
5375 | |
5376 | /* | |
5377 | * Load several special directories from the MOS needed by the dsl_pool | |
5378 | * layer. | |
5379 | */ | |
687e4d7f | 5380 | spa_import_progress_set_notes(spa, "Loading special MOS directories"); |
9eb7b46e PZ |
5381 | error = spa_ld_load_special_directories(spa); |
5382 | if (error != 0) | |
c183d164 | 5383 | goto fail; |
9eb7b46e | 5384 | |
9eb7b46e PZ |
5385 | /* |
5386 | * Retrieve pool properties from the MOS. | |
5387 | */ | |
687e4d7f | 5388 | spa_import_progress_set_notes(spa, "Loading properties"); |
9eb7b46e PZ |
5389 | error = spa_ld_get_props(spa); |
5390 | if (error != 0) | |
c183d164 | 5391 | goto fail; |
9eb7b46e PZ |
5392 | |
5393 | /* | |
5394 | * Retrieve the list of auxiliary devices - cache devices and spares - | |
5395 | * and open them. | |
5396 | */ | |
687e4d7f | 5397 | spa_import_progress_set_notes(spa, "Loading AUX vdevs"); |
9eb7b46e PZ |
5398 | error = spa_ld_open_aux_vdevs(spa, type); |
5399 | if (error != 0) | |
c183d164 | 5400 | goto fail; |
9eb7b46e PZ |
5401 | |
5402 | /* | |
5403 | * Load the metadata for all vdevs. Also check if unopenable devices | |
5404 | * should be autoreplaced. | |
5405 | */ | |
687e4d7f | 5406 | spa_import_progress_set_notes(spa, "Loading vdev metadata"); |
4a0ee12a | 5407 | error = spa_ld_load_vdev_metadata(spa); |
9eb7b46e | 5408 | if (error != 0) |
c183d164 | 5409 | goto fail; |
9eb7b46e | 5410 | |
687e4d7f | 5411 | spa_import_progress_set_notes(spa, "Loading dedup tables"); |
9eb7b46e PZ |
5412 | error = spa_ld_load_dedup_tables(spa); |
5413 | if (error != 0) | |
c183d164 | 5414 | goto fail; |
9eb7b46e | 5415 | |
687e4d7f | 5416 | spa_import_progress_set_notes(spa, "Loading BRT"); |
67a1b037 PJD |
5417 | error = spa_ld_load_brt(spa); |
5418 | if (error != 0) | |
c183d164 | 5419 | goto fail; |
67a1b037 | 5420 | |
9eb7b46e PZ |
5421 | /* |
5422 | * Verify the logs now to make sure we don't have any unexpected errors | |
5423 | * when we claim log blocks later. | |
5424 | */ | |
687e4d7f | 5425 | spa_import_progress_set_notes(spa, "Verifying Log Devices"); |
9eb7b46e PZ |
5426 | error = spa_ld_verify_logs(spa, type, ereport); |
5427 | if (error != 0) | |
c183d164 | 5428 | goto fail; |
9eb7b46e | 5429 | |
9ae529ec | 5430 | if (missing_feat_write) { |
6cb8e530 | 5431 | ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); |
9ae529ec CS |
5432 | |
5433 | /* | |
5434 | * At this point, we know that we can open the pool in | |
5435 | * read-only mode but not read-write mode. We now have enough | |
5436 | * information and can return to userland. | |
5437 | */ | |
c183d164 GW |
5438 | error = spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, |
5439 | ENOTSUP); | |
5440 | goto fail; | |
9ae529ec CS |
5441 | } |
5442 | ||
572e2857 | 5443 | /* |
9eb7b46e PZ |
5444 | * Traverse the last txgs to make sure the pool was left off in a safe |
5445 | * state. When performing an extreme rewind, we verify the whole pool, | |
5446 | * which can take a very long time. | |
572e2857 | 5447 | */ |
687e4d7f | 5448 | spa_import_progress_set_notes(spa, "Verifying pool data"); |
4a0ee12a | 5449 | error = spa_ld_verify_pool_data(spa); |
9eb7b46e | 5450 | if (error != 0) |
c183d164 | 5451 | goto fail; |
572e2857 | 5452 | |
9eb7b46e PZ |
5453 | /* |
5454 | * Calculate the deflated space for the pool. This must be done before | |
5455 | * we write anything to the pool because we'd need to update the space | |
5456 | * accounting using the deflated sizes. | |
5457 | */ | |
687e4d7f | 5458 | spa_import_progress_set_notes(spa, "Calculating deflated space"); |
9eb7b46e PZ |
5459 | spa_update_dspace(spa); |
5460 | ||
5461 | /* | |
5462 | * We have now retrieved all the information we needed to open the | |
5463 | * pool. If we are importing the pool in read-write mode, a few | |
5464 | * additional steps must be performed to finish the import. | |
5465 | */ | |
687e4d7f | 5466 | spa_import_progress_set_notes(spa, "Starting import"); |
6cb8e530 | 5467 | if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || |
428870ff | 5468 | spa->spa_load_max_txg == UINT64_MAX)) { |
6cb8e530 PZ |
5469 | uint64_t config_cache_txg = spa->spa_config_txg; |
5470 | ||
5471 | ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); | |
34dc7c2f | 5472 | |
5caeef02 DB |
5473 | /* |
5474 | * Before we do any zio_write's, complete the raidz expansion | |
5475 | * scratch space copying, if necessary. | |
5476 | */ | |
5477 | if (RRSS_GET_STATE(&spa->spa_uberblock) == RRSS_SCRATCH_VALID) | |
5478 | vdev_raidz_reflow_copy_scratch(spa); | |
5479 | ||
d2734cce SD |
5480 | /* |
5481 | * In case of a checkpoint rewind, log the original txg | |
5482 | * of the checkpointed uberblock. | |
5483 | */ | |
5484 | if (checkpoint_rewind) { | |
5485 | spa_history_log_internal(spa, "checkpoint rewind", | |
5486 | NULL, "rewound state to txg=%llu", | |
5487 | (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); | |
5488 | } | |
5489 | ||
687e4d7f | 5490 | spa_import_progress_set_notes(spa, "Claiming ZIL blocks"); |
34dc7c2f | 5491 | /* |
9eb7b46e | 5492 | * Traverse the ZIL and claim all blocks. |
34dc7c2f | 5493 | */ |
9eb7b46e | 5494 | spa_ld_claim_log_blocks(spa); |
428870ff | 5495 | |
9eb7b46e PZ |
5496 | /* |
5497 | * Kick-off the syncing thread. | |
5498 | */ | |
34dc7c2f BB |
5499 | spa->spa_sync_on = B_TRUE; |
5500 | txg_sync_start(spa->spa_dsl_pool); | |
379ca9cf | 5501 | mmp_thread_start(spa); |
34dc7c2f BB |
5502 | |
5503 | /* | |
428870ff BB |
5504 | * Wait for all claims to sync. We sync up to the highest |
5505 | * claimed log block birth time so that claimed log blocks | |
5506 | * don't appear to be from the future. spa_claim_max_txg | |
9eb7b46e PZ |
5507 | * will have been set for us by ZIL traversal operations |
5508 | * performed above. | |
34dc7c2f | 5509 | */ |
687e4d7f | 5510 | spa_import_progress_set_notes(spa, "Syncing ZIL claims"); |
428870ff | 5511 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); |
34dc7c2f BB |
5512 | |
5513 | /* | |
9eb7b46e PZ |
5514 | * Check if we need to request an update of the config. On the |
5515 | * next sync, we would update the config stored in vdev labels | |
5516 | * and the cachefile (by default /etc/zfs/zpool.cache). | |
34dc7c2f | 5517 | */ |
687e4d7f | 5518 | spa_import_progress_set_notes(spa, "Updating configs"); |
6cb8e530 | 5519 | spa_ld_check_for_config_update(spa, config_cache_txg, |
d2734cce | 5520 | update_config_cache); |
fb5f0bc8 BB |
5521 | |
5522 | /* | |
9a49d3f3 BB |
5523 | * Check if a rebuild was in progress and if so resume it. |
5524 | * Then check all DTLs to see if anything needs resilvering. | |
5525 | * The resilver will be deferred if a rebuild was started. | |
fb5f0bc8 | 5526 | */ |
687e4d7f | 5527 | spa_import_progress_set_notes(spa, "Starting resilvers"); |
9a49d3f3 BB |
5528 | if (vdev_rebuild_active(spa->spa_root_vdev)) { |
5529 | vdev_rebuild_restart(spa); | |
5530 | } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
5531 | vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
fb5f0bc8 | 5532 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
9a49d3f3 | 5533 | } |
428870ff | 5534 | |
6f1ffb06 MA |
5535 | /* |
5536 | * Log the fact that we booted up (so that we can detect if | |
5537 | * we rebooted in the middle of an operation). | |
5538 | */ | |
d5e024cb | 5539 | spa_history_log_version(spa, "open", NULL); |
6f1ffb06 | 5540 | |
687e4d7f DB |
5541 | spa_import_progress_set_notes(spa, |
5542 | "Restarting device removals"); | |
9b2266e3 SD |
5543 | spa_restart_removal(spa); |
5544 | spa_spawn_aux_threads(spa); | |
5545 | ||
428870ff BB |
5546 | /* |
5547 | * Delete any inconsistent datasets. | |
9b2266e3 SD |
5548 | * |
5549 | * Note: | |
5550 | * Since we may be issuing deletes for clones here, | |
5551 | * we make sure to do so after we've spawned all the | |
5552 | * auxiliary threads above (from which the livelist | |
5553 | * deletion zthr is part of). | |
428870ff | 5554 | */ |
687e4d7f DB |
5555 | spa_import_progress_set_notes(spa, |
5556 | "Cleaning up inconsistent objsets"); | |
428870ff BB |
5557 | (void) dmu_objset_find(spa_name(spa), |
5558 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
5559 | ||
5560 | /* | |
5561 | * Clean up any stale temporary dataset userrefs. | |
5562 | */ | |
687e4d7f DB |
5563 | spa_import_progress_set_notes(spa, |
5564 | "Cleaning up temporary userrefs"); | |
428870ff | 5565 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); |
a1d477c2 | 5566 | |
619f0976 | 5567 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
687e4d7f | 5568 | spa_import_progress_set_notes(spa, "Restarting initialize"); |
619f0976 | 5569 | vdev_initialize_restart(spa->spa_root_vdev); |
687e4d7f | 5570 | spa_import_progress_set_notes(spa, "Restarting TRIM"); |
1b939560 BB |
5571 | vdev_trim_restart(spa->spa_root_vdev); |
5572 | vdev_autotrim_restart(spa); | |
619f0976 | 5573 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
687e4d7f | 5574 | spa_import_progress_set_notes(spa, "Finished importing"); |
34dc7c2f | 5575 | } |
c183d164 | 5576 | zio_handle_import_delay(spa, gethrtime() - load_start); |
34dc7c2f | 5577 | |
ca95f70d | 5578 | spa_import_progress_remove(spa_guid(spa)); |
77f6826b GA |
5579 | spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); |
5580 | ||
4a0ee12a | 5581 | spa_load_note(spa, "LOADED"); |
c183d164 GW |
5582 | fail: |
5583 | mutex_enter(&spa_namespace_lock); | |
5584 | spa->spa_load_thread = NULL; | |
5585 | cv_broadcast(&spa_namespace_cv); | |
5586 | ||
5587 | return (error); | |
4a0ee12a | 5588 | |
428870ff | 5589 | } |
34dc7c2f | 5590 | |
428870ff | 5591 | static int |
6cb8e530 | 5592 | spa_load_retry(spa_t *spa, spa_load_state_t state) |
428870ff | 5593 | { |
da92d5cb | 5594 | spa_mode_t mode = spa->spa_mode; |
572e2857 | 5595 | |
428870ff BB |
5596 | spa_unload(spa); |
5597 | spa_deactivate(spa); | |
5598 | ||
dea377c0 | 5599 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; |
428870ff | 5600 | |
572e2857 | 5601 | spa_activate(spa, mode); |
428870ff BB |
5602 | spa_async_suspend(spa); |
5603 | ||
4a0ee12a PZ |
5604 | spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", |
5605 | (u_longlong_t)spa->spa_load_max_txg); | |
5606 | ||
6cb8e530 | 5607 | return (spa_load(spa, state, SPA_IMPORT_EXISTING)); |
428870ff BB |
5608 | } |
5609 | ||
9ae529ec CS |
5610 | /* |
5611 | * If spa_load() fails this function will try loading prior txg's. If | |
5612 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
5613 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
5614 | * function will not rewind the pool and will return the same error as | |
5615 | * spa_load(). | |
5616 | */ | |
428870ff | 5617 | static int |
6cb8e530 PZ |
5618 | spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, |
5619 | int rewind_flags) | |
428870ff | 5620 | { |
9ae529ec | 5621 | nvlist_t *loadinfo = NULL; |
428870ff BB |
5622 | nvlist_t *config = NULL; |
5623 | int load_error, rewind_error; | |
5624 | uint64_t safe_rewind_txg; | |
5625 | uint64_t min_txg; | |
5626 | ||
5627 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
5628 | spa->spa_load_max_txg = spa->spa_load_txg; | |
5629 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
5630 | } else { | |
5631 | spa->spa_load_max_txg = max_request; | |
dea377c0 MA |
5632 | if (max_request != UINT64_MAX) |
5633 | spa->spa_extreme_rewind = B_TRUE; | |
428870ff BB |
5634 | } |
5635 | ||
6cb8e530 | 5636 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); |
428870ff BB |
5637 | if (load_error == 0) |
5638 | return (0); | |
d2734cce SD |
5639 | if (load_error == ZFS_ERR_NO_CHECKPOINT) { |
5640 | /* | |
5641 | * When attempting checkpoint-rewind on a pool with no | |
5642 | * checkpoint, we should not attempt to load uberblocks | |
5643 | * from previous txgs when spa_load fails. | |
5644 | */ | |
5645 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
ca95f70d | 5646 | spa_import_progress_remove(spa_guid(spa)); |
d2734cce SD |
5647 | return (load_error); |
5648 | } | |
428870ff BB |
5649 | |
5650 | if (spa->spa_root_vdev != NULL) | |
5651 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
5652 | ||
5653 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
5654 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
5655 | ||
5656 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
5657 | nvlist_free(config); | |
ca95f70d | 5658 | spa_import_progress_remove(spa_guid(spa)); |
428870ff BB |
5659 | return (load_error); |
5660 | } | |
5661 | ||
9ae529ec CS |
5662 | if (state == SPA_LOAD_RECOVER) { |
5663 | /* Price of rolling back is discarding txgs, including log */ | |
428870ff | 5664 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
9ae529ec CS |
5665 | } else { |
5666 | /* | |
5667 | * If we aren't rolling back save the load info from our first | |
5668 | * import attempt so that we can restore it after attempting | |
5669 | * to rewind. | |
5670 | */ | |
5671 | loadinfo = spa->spa_load_info; | |
5672 | spa->spa_load_info = fnvlist_alloc(); | |
5673 | } | |
428870ff BB |
5674 | |
5675 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
5676 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
5677 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
5678 | TXG_INITIAL : safe_rewind_txg; | |
5679 | ||
5680 | /* | |
5681 | * Continue as long as we're finding errors, we're still within | |
5682 | * the acceptable rewind range, and we're still finding uberblocks | |
5683 | */ | |
5684 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
5685 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
5686 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
5687 | spa->spa_extreme_rewind = B_TRUE; | |
6cb8e530 | 5688 | rewind_error = spa_load_retry(spa, state); |
428870ff BB |
5689 | } |
5690 | ||
428870ff BB |
5691 | spa->spa_extreme_rewind = B_FALSE; |
5692 | spa->spa_load_max_txg = UINT64_MAX; | |
5693 | ||
5694 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
5695 | spa_config_set(spa, config); | |
ee6370a7 | 5696 | else |
5697 | nvlist_free(config); | |
428870ff | 5698 | |
9ae529ec CS |
5699 | if (state == SPA_LOAD_RECOVER) { |
5700 | ASSERT3P(loadinfo, ==, NULL); | |
ca95f70d | 5701 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5702 | return (rewind_error); |
5703 | } else { | |
5704 | /* Store the rewind info as part of the initial load info */ | |
5705 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
5706 | spa->spa_load_info); | |
5707 | ||
5708 | /* Restore the initial load info */ | |
5709 | fnvlist_free(spa->spa_load_info); | |
5710 | spa->spa_load_info = loadinfo; | |
5711 | ||
ca95f70d | 5712 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5713 | return (load_error); |
5714 | } | |
34dc7c2f BB |
5715 | } |
5716 | ||
5717 | /* | |
5718 | * Pool Open/Import | |
5719 | * | |
5720 | * The import case is identical to an open except that the configuration is sent | |
5721 | * down from userland, instead of grabbed from the configuration cache. For the | |
5722 | * case of an open, the pool configuration will exist in the | |
5723 | * POOL_STATE_UNINITIALIZED state. | |
5724 | * | |
5725 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
5726 | * the same time open the pool, without having to keep around the spa_t in some | |
5727 | * ambiguous state. | |
5728 | */ | |
5729 | static int | |
a926aab9 AZ |
5730 | spa_open_common(const char *pool, spa_t **spapp, const void *tag, |
5731 | nvlist_t *nvpolicy, nvlist_t **config) | |
34dc7c2f BB |
5732 | { |
5733 | spa_t *spa; | |
572e2857 | 5734 | spa_load_state_t state = SPA_LOAD_OPEN; |
34dc7c2f | 5735 | int error; |
34dc7c2f | 5736 | int locked = B_FALSE; |
526af785 | 5737 | int firstopen = B_FALSE; |
34dc7c2f BB |
5738 | |
5739 | *spapp = NULL; | |
5740 | ||
5741 | /* | |
5742 | * As disgusting as this is, we need to support recursive calls to this | |
5743 | * function because dsl_dir_open() is called during spa_load(), and ends | |
5744 | * up calling spa_open() again. The real fix is to figure out how to | |
5745 | * avoid dsl_dir_open() calling this in the first place. | |
5746 | */ | |
c25b8f99 | 5747 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { |
34dc7c2f BB |
5748 | mutex_enter(&spa_namespace_lock); |
5749 | locked = B_TRUE; | |
5750 | } | |
5751 | ||
5752 | if ((spa = spa_lookup(pool)) == NULL) { | |
5753 | if (locked) | |
5754 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5755 | return (SET_ERROR(ENOENT)); |
34dc7c2f | 5756 | } |
428870ff | 5757 | |
34dc7c2f | 5758 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { |
8a393be3 | 5759 | zpool_load_policy_t policy; |
428870ff | 5760 | |
526af785 PJD |
5761 | firstopen = B_TRUE; |
5762 | ||
8a393be3 | 5763 | zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, |
428870ff | 5764 | &policy); |
8a393be3 | 5765 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) |
428870ff | 5766 | state = SPA_LOAD_RECOVER; |
34dc7c2f | 5767 | |
fb5f0bc8 | 5768 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 5769 | |
428870ff BB |
5770 | if (state != SPA_LOAD_RECOVER) |
5771 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6cb8e530 | 5772 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; |
428870ff | 5773 | |
4a0ee12a | 5774 | zfs_dbgmsg("spa_open_common: opening %s", pool); |
8a393be3 PZ |
5775 | error = spa_load_best(spa, state, policy.zlp_txg, |
5776 | policy.zlp_rewind); | |
34dc7c2f BB |
5777 | |
5778 | if (error == EBADF) { | |
5779 | /* | |
5780 | * If vdev_validate() returns failure (indicated by | |
5781 | * EBADF), it indicates that one of the vdevs indicates | |
5782 | * that the pool has been exported or destroyed. If | |
5783 | * this is the case, the config cache is out of sync and | |
5784 | * we should remove the pool from the namespace. | |
5785 | */ | |
34dc7c2f BB |
5786 | spa_unload(spa); |
5787 | spa_deactivate(spa); | |
55c12724 | 5788 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 5789 | spa_remove(spa); |
34dc7c2f BB |
5790 | if (locked) |
5791 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5792 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
5793 | } |
5794 | ||
5795 | if (error) { | |
5796 | /* | |
5797 | * We can't open the pool, but we still have useful | |
5798 | * information: the state of each vdev after the | |
5799 | * attempted vdev_open(). Return this to the user. | |
5800 | */ | |
572e2857 | 5801 | if (config != NULL && spa->spa_config) { |
65ad5d11 AJ |
5802 | *config = fnvlist_dup(spa->spa_config); |
5803 | fnvlist_add_nvlist(*config, | |
572e2857 | 5804 | ZPOOL_CONFIG_LOAD_INFO, |
65ad5d11 | 5805 | spa->spa_load_info); |
572e2857 | 5806 | } |
34dc7c2f BB |
5807 | spa_unload(spa); |
5808 | spa_deactivate(spa); | |
428870ff | 5809 | spa->spa_last_open_failed = error; |
34dc7c2f BB |
5810 | if (locked) |
5811 | mutex_exit(&spa_namespace_lock); | |
5812 | *spapp = NULL; | |
5813 | return (error); | |
34dc7c2f | 5814 | } |
34dc7c2f BB |
5815 | } |
5816 | ||
5817 | spa_open_ref(spa, tag); | |
5818 | ||
b128c09f | 5819 | if (config != NULL) |
34dc7c2f | 5820 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
34dc7c2f | 5821 | |
572e2857 BB |
5822 | /* |
5823 | * If we've recovered the pool, pass back any information we | |
5824 | * gathered while doing the load. | |
5825 | */ | |
1bd02680 | 5826 | if (state == SPA_LOAD_RECOVER && config != NULL) { |
65ad5d11 AJ |
5827 | fnvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, |
5828 | spa->spa_load_info); | |
572e2857 BB |
5829 | } |
5830 | ||
428870ff BB |
5831 | if (locked) { |
5832 | spa->spa_last_open_failed = 0; | |
5833 | spa->spa_last_ubsync_txg = 0; | |
5834 | spa->spa_load_txg = 0; | |
5835 | mutex_exit(&spa_namespace_lock); | |
5836 | } | |
5837 | ||
526af785 | 5838 | if (firstopen) |
ec213971 | 5839 | zvol_create_minors_recursive(spa_name(spa)); |
526af785 | 5840 | |
428870ff BB |
5841 | *spapp = spa; |
5842 | ||
34dc7c2f BB |
5843 | return (0); |
5844 | } | |
5845 | ||
428870ff | 5846 | int |
a926aab9 AZ |
5847 | spa_open_rewind(const char *name, spa_t **spapp, const void *tag, |
5848 | nvlist_t *policy, nvlist_t **config) | |
428870ff BB |
5849 | { |
5850 | return (spa_open_common(name, spapp, tag, policy, config)); | |
5851 | } | |
5852 | ||
34dc7c2f | 5853 | int |
a926aab9 | 5854 | spa_open(const char *name, spa_t **spapp, const void *tag) |
34dc7c2f | 5855 | { |
428870ff | 5856 | return (spa_open_common(name, spapp, tag, NULL, NULL)); |
34dc7c2f BB |
5857 | } |
5858 | ||
5859 | /* | |
5860 | * Lookup the given spa_t, incrementing the inject count in the process, | |
5861 | * preventing it from being exported or destroyed. | |
5862 | */ | |
5863 | spa_t * | |
5864 | spa_inject_addref(char *name) | |
5865 | { | |
5866 | spa_t *spa; | |
5867 | ||
5868 | mutex_enter(&spa_namespace_lock); | |
5869 | if ((spa = spa_lookup(name)) == NULL) { | |
5870 | mutex_exit(&spa_namespace_lock); | |
5871 | return (NULL); | |
5872 | } | |
5873 | spa->spa_inject_ref++; | |
5874 | mutex_exit(&spa_namespace_lock); | |
5875 | ||
5876 | return (spa); | |
5877 | } | |
5878 | ||
5879 | void | |
5880 | spa_inject_delref(spa_t *spa) | |
5881 | { | |
5882 | mutex_enter(&spa_namespace_lock); | |
5883 | spa->spa_inject_ref--; | |
5884 | mutex_exit(&spa_namespace_lock); | |
5885 | } | |
5886 | ||
5887 | /* | |
5888 | * Add spares device information to the nvlist. | |
5889 | */ | |
5890 | static void | |
5891 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
5892 | { | |
5893 | nvlist_t **spares; | |
5894 | uint_t i, nspares; | |
5895 | nvlist_t *nvroot; | |
5896 | uint64_t guid; | |
5897 | vdev_stat_t *vs; | |
5898 | uint_t vsc; | |
5899 | uint64_t pool; | |
5900 | ||
9babb374 BB |
5901 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5902 | ||
34dc7c2f BB |
5903 | if (spa->spa_spares.sav_count == 0) |
5904 | return; | |
5905 | ||
65ad5d11 AJ |
5906 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5907 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5908 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f | 5909 | if (nspares != 0) { |
795075e6 PD |
5910 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5911 | (const nvlist_t * const *)spares, nspares); | |
65ad5d11 AJ |
5912 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5913 | &spares, &nspares)); | |
34dc7c2f BB |
5914 | |
5915 | /* | |
5916 | * Go through and find any spares which have since been | |
5917 | * repurposed as an active spare. If this is the case, update | |
5918 | * their status appropriately. | |
5919 | */ | |
5920 | for (i = 0; i < nspares; i++) { | |
65ad5d11 AJ |
5921 | guid = fnvlist_lookup_uint64(spares[i], |
5922 | ZPOOL_CONFIG_GUID); | |
a05263b7 AH |
5923 | VERIFY0(nvlist_lookup_uint64_array(spares[i], |
5924 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
b128c09f BB |
5925 | if (spa_spare_exists(guid, &pool, NULL) && |
5926 | pool != 0ULL) { | |
34dc7c2f BB |
5927 | vs->vs_state = VDEV_STATE_CANT_OPEN; |
5928 | vs->vs_aux = VDEV_AUX_SPARED; | |
a05263b7 AH |
5929 | } else { |
5930 | vs->vs_state = | |
5931 | spa->spa_spares.sav_vdevs[i]->vdev_state; | |
34dc7c2f BB |
5932 | } |
5933 | } | |
5934 | } | |
5935 | } | |
5936 | ||
5937 | /* | |
5938 | * Add l2cache device information to the nvlist, including vdev stats. | |
5939 | */ | |
5940 | static void | |
5941 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
5942 | { | |
5943 | nvlist_t **l2cache; | |
5944 | uint_t i, j, nl2cache; | |
5945 | nvlist_t *nvroot; | |
5946 | uint64_t guid; | |
5947 | vdev_t *vd; | |
5948 | vdev_stat_t *vs; | |
5949 | uint_t vsc; | |
5950 | ||
9babb374 BB |
5951 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5952 | ||
34dc7c2f BB |
5953 | if (spa->spa_l2cache.sav_count == 0) |
5954 | return; | |
5955 | ||
65ad5d11 AJ |
5956 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5957 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5958 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
34dc7c2f | 5959 | if (nl2cache != 0) { |
795075e6 PD |
5960 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5961 | (const nvlist_t * const *)l2cache, nl2cache); | |
65ad5d11 AJ |
5962 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5963 | &l2cache, &nl2cache)); | |
34dc7c2f BB |
5964 | |
5965 | /* | |
5966 | * Update level 2 cache device stats. | |
5967 | */ | |
5968 | ||
5969 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 AJ |
5970 | guid = fnvlist_lookup_uint64(l2cache[i], |
5971 | ZPOOL_CONFIG_GUID); | |
34dc7c2f BB |
5972 | |
5973 | vd = NULL; | |
5974 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
5975 | if (guid == | |
5976 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
5977 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
5978 | break; | |
5979 | } | |
5980 | } | |
5981 | ASSERT(vd != NULL); | |
5982 | ||
65ad5d11 AJ |
5983 | VERIFY0(nvlist_lookup_uint64_array(l2cache[i], |
5984 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
34dc7c2f | 5985 | vdev_get_stats(vd, vs); |
193a37cb TH |
5986 | vdev_config_generate_stats(vd, l2cache[i]); |
5987 | ||
34dc7c2f BB |
5988 | } |
5989 | } | |
34dc7c2f BB |
5990 | } |
5991 | ||
9ae529ec | 5992 | static void |
417104bd | 5993 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) |
9ae529ec | 5994 | { |
9ae529ec CS |
5995 | zap_cursor_t zc; |
5996 | zap_attribute_t za; | |
5997 | ||
9ae529ec CS |
5998 | if (spa->spa_feat_for_read_obj != 0) { |
5999 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6000 | spa->spa_feat_for_read_obj); | |
6001 | zap_cursor_retrieve(&zc, &za) == 0; | |
6002 | zap_cursor_advance(&zc)) { | |
6003 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
6004 | za.za_num_integers == 1); | |
417104bd | 6005 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
6006 | za.za_first_integer)); |
6007 | } | |
6008 | zap_cursor_fini(&zc); | |
6009 | } | |
6010 | ||
6011 | if (spa->spa_feat_for_write_obj != 0) { | |
6012 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6013 | spa->spa_feat_for_write_obj); | |
6014 | zap_cursor_retrieve(&zc, &za) == 0; | |
6015 | zap_cursor_advance(&zc)) { | |
6016 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
6017 | za.za_num_integers == 1); | |
417104bd | 6018 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
6019 | za.za_first_integer)); |
6020 | } | |
6021 | zap_cursor_fini(&zc); | |
6022 | } | |
417104bd NB |
6023 | } |
6024 | ||
6025 | static void | |
6026 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
6027 | { | |
6028 | int i; | |
6029 | ||
6030 | for (i = 0; i < SPA_FEATURES; i++) { | |
6031 | zfeature_info_t feature = spa_feature_table[i]; | |
6032 | uint64_t refcount; | |
6033 | ||
6034 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
6035 | continue; | |
6036 | ||
6037 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
6038 | } | |
6039 | } | |
6040 | ||
6041 | /* | |
6042 | * Store a list of pool features and their reference counts in the | |
6043 | * config. | |
6044 | * | |
6045 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
6046 | * the pool features and reference counts from disk, then save the list | |
6047 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
6048 | * and refresh its values from the cached reference counts. This | |
6049 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
6050 | * clear' can resume the pool. | |
6051 | */ | |
6052 | static void | |
6053 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
6054 | { | |
4eb30c68 | 6055 | nvlist_t *features; |
417104bd NB |
6056 | |
6057 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
6058 | ||
4eb30c68 NB |
6059 | mutex_enter(&spa->spa_feat_stats_lock); |
6060 | features = spa->spa_feat_stats; | |
6061 | ||
417104bd NB |
6062 | if (features != NULL) { |
6063 | spa_feature_stats_from_cache(spa, features); | |
6064 | } else { | |
6065 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
6066 | spa->spa_feat_stats = features; | |
6067 | spa_feature_stats_from_disk(spa, features); | |
6068 | } | |
9ae529ec | 6069 | |
417104bd NB |
6070 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, |
6071 | features)); | |
4eb30c68 NB |
6072 | |
6073 | mutex_exit(&spa->spa_feat_stats_lock); | |
9ae529ec CS |
6074 | } |
6075 | ||
34dc7c2f | 6076 | int |
9ae529ec CS |
6077 | spa_get_stats(const char *name, nvlist_t **config, |
6078 | char *altroot, size_t buflen) | |
34dc7c2f BB |
6079 | { |
6080 | int error; | |
6081 | spa_t *spa; | |
6082 | ||
6083 | *config = NULL; | |
428870ff | 6084 | error = spa_open_common(name, &spa, FTAG, NULL, config); |
34dc7c2f | 6085 | |
9babb374 BB |
6086 | if (spa != NULL) { |
6087 | /* | |
6088 | * This still leaves a window of inconsistency where the spares | |
6089 | * or l2cache devices could change and the config would be | |
6090 | * self-inconsistent. | |
6091 | */ | |
6092 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
34dc7c2f | 6093 | |
9babb374 | 6094 | if (*config != NULL) { |
572e2857 BB |
6095 | uint64_t loadtimes[2]; |
6096 | ||
6097 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
6098 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
65ad5d11 AJ |
6099 | fnvlist_add_uint64_array(*config, |
6100 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2); | |
572e2857 | 6101 | |
65ad5d11 | 6102 | fnvlist_add_uint64(*config, |
9babb374 | 6103 | ZPOOL_CONFIG_ERRCOUNT, |
018f2604 | 6104 | spa_approx_errlog_size(spa)); |
9babb374 | 6105 | |
cec3a0a1 | 6106 | if (spa_suspended(spa)) { |
65ad5d11 | 6107 | fnvlist_add_uint64(*config, |
9babb374 | 6108 | ZPOOL_CONFIG_SUSPENDED, |
65ad5d11 AJ |
6109 | spa->spa_failmode); |
6110 | fnvlist_add_uint64(*config, | |
cec3a0a1 | 6111 | ZPOOL_CONFIG_SUSPENDED_REASON, |
65ad5d11 | 6112 | spa->spa_suspended); |
cec3a0a1 | 6113 | } |
b128c09f | 6114 | |
9babb374 BB |
6115 | spa_add_spares(spa, *config); |
6116 | spa_add_l2cache(spa, *config); | |
9ae529ec | 6117 | spa_add_feature_stats(spa, *config); |
9babb374 | 6118 | } |
34dc7c2f BB |
6119 | } |
6120 | ||
6121 | /* | |
6122 | * We want to get the alternate root even for faulted pools, so we cheat | |
6123 | * and call spa_lookup() directly. | |
6124 | */ | |
6125 | if (altroot) { | |
6126 | if (spa == NULL) { | |
6127 | mutex_enter(&spa_namespace_lock); | |
6128 | spa = spa_lookup(name); | |
6129 | if (spa) | |
6130 | spa_altroot(spa, altroot, buflen); | |
6131 | else | |
6132 | altroot[0] = '\0'; | |
6133 | spa = NULL; | |
6134 | mutex_exit(&spa_namespace_lock); | |
6135 | } else { | |
6136 | spa_altroot(spa, altroot, buflen); | |
6137 | } | |
6138 | } | |
6139 | ||
9babb374 BB |
6140 | if (spa != NULL) { |
6141 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
34dc7c2f | 6142 | spa_close(spa, FTAG); |
9babb374 | 6143 | } |
34dc7c2f BB |
6144 | |
6145 | return (error); | |
6146 | } | |
6147 | ||
6148 | /* | |
6149 | * Validate that the auxiliary device array is well formed. We must have an | |
6150 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
6151 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
6152 | * specified, as long as they are well-formed. | |
6153 | */ | |
6154 | static int | |
6155 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
6156 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
6157 | vdev_labeltype_t label) | |
6158 | { | |
6159 | nvlist_t **dev; | |
6160 | uint_t i, ndev; | |
6161 | vdev_t *vd; | |
6162 | int error; | |
6163 | ||
b128c09f BB |
6164 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
6165 | ||
34dc7c2f BB |
6166 | /* |
6167 | * It's acceptable to have no devs specified. | |
6168 | */ | |
6169 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
6170 | return (0); | |
6171 | ||
6172 | if (ndev == 0) | |
2e528b49 | 6173 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
6174 | |
6175 | /* | |
6176 | * Make sure the pool is formatted with a version that supports this | |
6177 | * device type. | |
6178 | */ | |
6179 | if (spa_version(spa) < version) | |
2e528b49 | 6180 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
6181 | |
6182 | /* | |
6183 | * Set the pending device list so we correctly handle device in-use | |
6184 | * checking. | |
6185 | */ | |
6186 | sav->sav_pending = dev; | |
6187 | sav->sav_npending = ndev; | |
6188 | ||
6189 | for (i = 0; i < ndev; i++) { | |
6190 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
6191 | mode)) != 0) | |
6192 | goto out; | |
6193 | ||
6194 | if (!vd->vdev_ops->vdev_op_leaf) { | |
6195 | vdev_free(vd); | |
2e528b49 | 6196 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
6197 | goto out; |
6198 | } | |
6199 | ||
34dc7c2f BB |
6200 | vd->vdev_top = vd; |
6201 | ||
6202 | if ((error = vdev_open(vd)) == 0 && | |
6203 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
65ad5d11 AJ |
6204 | fnvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, |
6205 | vd->vdev_guid); | |
34dc7c2f BB |
6206 | } |
6207 | ||
6208 | vdev_free(vd); | |
6209 | ||
6210 | if (error && | |
6211 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
6212 | goto out; | |
6213 | else | |
6214 | error = 0; | |
6215 | } | |
6216 | ||
6217 | out: | |
6218 | sav->sav_pending = NULL; | |
6219 | sav->sav_npending = 0; | |
6220 | return (error); | |
6221 | } | |
6222 | ||
6223 | static int | |
6224 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
6225 | { | |
6226 | int error; | |
6227 | ||
b128c09f BB |
6228 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
6229 | ||
34dc7c2f BB |
6230 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, |
6231 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
6232 | VDEV_LABEL_SPARE)) != 0) { | |
6233 | return (error); | |
6234 | } | |
6235 | ||
6236 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
6237 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
6238 | VDEV_LABEL_L2CACHE)); | |
6239 | } | |
6240 | ||
6241 | static void | |
6242 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
6243 | const char *config) | |
6244 | { | |
6245 | int i; | |
6246 | ||
6247 | if (sav->sav_config != NULL) { | |
6248 | nvlist_t **olddevs; | |
6249 | uint_t oldndevs; | |
6250 | nvlist_t **newdevs; | |
6251 | ||
6252 | /* | |
4e33ba4c | 6253 | * Generate new dev list by concatenating with the |
34dc7c2f BB |
6254 | * current dev list. |
6255 | */ | |
65ad5d11 AJ |
6256 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, config, |
6257 | &olddevs, &oldndevs)); | |
34dc7c2f BB |
6258 | |
6259 | newdevs = kmem_alloc(sizeof (void *) * | |
79c76d5b | 6260 | (ndevs + oldndevs), KM_SLEEP); |
34dc7c2f | 6261 | for (i = 0; i < oldndevs; i++) |
65ad5d11 | 6262 | newdevs[i] = fnvlist_dup(olddevs[i]); |
34dc7c2f | 6263 | for (i = 0; i < ndevs; i++) |
65ad5d11 | 6264 | newdevs[i + oldndevs] = fnvlist_dup(devs[i]); |
34dc7c2f | 6265 | |
65ad5d11 | 6266 | fnvlist_remove(sav->sav_config, config); |
34dc7c2f | 6267 | |
795075e6 PD |
6268 | fnvlist_add_nvlist_array(sav->sav_config, config, |
6269 | (const nvlist_t * const *)newdevs, ndevs + oldndevs); | |
34dc7c2f BB |
6270 | for (i = 0; i < oldndevs + ndevs; i++) |
6271 | nvlist_free(newdevs[i]); | |
6272 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
6273 | } else { | |
6274 | /* | |
6275 | * Generate a new dev list. | |
6276 | */ | |
65ad5d11 | 6277 | sav->sav_config = fnvlist_alloc(); |
795075e6 PD |
6278 | fnvlist_add_nvlist_array(sav->sav_config, config, |
6279 | (const nvlist_t * const *)devs, ndevs); | |
34dc7c2f BB |
6280 | } |
6281 | } | |
6282 | ||
6283 | /* | |
6284 | * Stop and drop level 2 ARC devices | |
6285 | */ | |
6286 | void | |
6287 | spa_l2cache_drop(spa_t *spa) | |
6288 | { | |
6289 | vdev_t *vd; | |
6290 | int i; | |
6291 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
6292 | ||
6293 | for (i = 0; i < sav->sav_count; i++) { | |
6294 | uint64_t pool; | |
6295 | ||
6296 | vd = sav->sav_vdevs[i]; | |
6297 | ASSERT(vd != NULL); | |
6298 | ||
fb5f0bc8 BB |
6299 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && |
6300 | pool != 0ULL && l2arc_vdev_present(vd)) | |
34dc7c2f | 6301 | l2arc_remove_vdev(vd); |
34dc7c2f BB |
6302 | } |
6303 | } | |
6304 | ||
b5256303 TC |
6305 | /* |
6306 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
6307 | * have the encryption feature flag enabled. | |
6308 | */ | |
6309 | static int | |
6310 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
6311 | boolean_t has_encryption) | |
6312 | { | |
6313 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
6314 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
6315 | !has_encryption) | |
6316 | return (SET_ERROR(ENOTSUP)); | |
6317 | ||
1fff937a | 6318 | return (dmu_objset_create_crypt_check(NULL, dcp, NULL)); |
b5256303 TC |
6319 | } |
6320 | ||
34dc7c2f BB |
6321 | /* |
6322 | * Pool Creation | |
6323 | */ | |
6324 | int | |
6325 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
b5256303 | 6326 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) |
34dc7c2f BB |
6327 | { |
6328 | spa_t *spa; | |
d1807f16 | 6329 | const char *altroot = NULL; |
34dc7c2f BB |
6330 | vdev_t *rvd; |
6331 | dsl_pool_t *dp; | |
6332 | dmu_tx_t *tx; | |
9babb374 | 6333 | int error = 0; |
34dc7c2f BB |
6334 | uint64_t txg = TXG_INITIAL; |
6335 | nvlist_t **spares, **l2cache; | |
6336 | uint_t nspares, nl2cache; | |
b2255edc | 6337 | uint64_t version, obj, ndraid = 0; |
9ae529ec | 6338 | boolean_t has_features; |
b5256303 | 6339 | boolean_t has_encryption; |
715c996d | 6340 | boolean_t has_allocclass; |
b5256303 | 6341 | spa_feature_t feat; |
d1807f16 RY |
6342 | const char *feat_name; |
6343 | const char *poolname; | |
83e9986f RY |
6344 | nvlist_t *nvl; |
6345 | ||
cc99f275 | 6346 | if (props == NULL || |
493fcce9 GW |
6347 | nvlist_lookup_string(props, |
6348 | zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) | |
83e9986f | 6349 | poolname = (char *)pool; |
34dc7c2f BB |
6350 | |
6351 | /* | |
6352 | * If this pool already exists, return failure. | |
6353 | */ | |
6354 | mutex_enter(&spa_namespace_lock); | |
83e9986f | 6355 | if (spa_lookup(poolname) != NULL) { |
34dc7c2f | 6356 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 6357 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
6358 | } |
6359 | ||
6360 | /* | |
6361 | * Allocate a new spa_t structure. | |
6362 | */ | |
83e9986f RY |
6363 | nvl = fnvlist_alloc(); |
6364 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
34dc7c2f BB |
6365 | (void) nvlist_lookup_string(props, |
6366 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
83e9986f RY |
6367 | spa = spa_add(poolname, nvl, altroot); |
6368 | fnvlist_free(nvl); | |
fb5f0bc8 | 6369 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 6370 | |
34dc7c2f | 6371 | if (props && (error = spa_prop_validate(spa, props))) { |
34dc7c2f BB |
6372 | spa_deactivate(spa); |
6373 | spa_remove(spa); | |
b128c09f | 6374 | mutex_exit(&spa_namespace_lock); |
34dc7c2f BB |
6375 | return (error); |
6376 | } | |
6377 | ||
83e9986f RY |
6378 | /* |
6379 | * Temporary pool names should never be written to disk. | |
6380 | */ | |
6381 | if (poolname != pool) | |
6382 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
6383 | ||
9ae529ec | 6384 | has_features = B_FALSE; |
b5256303 | 6385 | has_encryption = B_FALSE; |
715c996d | 6386 | has_allocclass = B_FALSE; |
1c27024e | 6387 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); |
9ae529ec | 6388 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { |
b5256303 | 6389 | if (zpool_prop_feature(nvpair_name(elem))) { |
9ae529ec | 6390 | has_features = B_TRUE; |
b5256303 TC |
6391 | |
6392 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
6393 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
6394 | if (feat == SPA_FEATURE_ENCRYPTION) | |
6395 | has_encryption = B_TRUE; | |
715c996d | 6396 | if (feat == SPA_FEATURE_ALLOCATION_CLASSES) |
6397 | has_allocclass = B_TRUE; | |
b5256303 TC |
6398 | } |
6399 | } | |
6400 | ||
6401 | /* verify encryption params, if they were provided */ | |
6402 | if (dcp != NULL) { | |
6403 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
6404 | if (error != 0) { | |
6405 | spa_deactivate(spa); | |
6406 | spa_remove(spa); | |
6407 | mutex_exit(&spa_namespace_lock); | |
6408 | return (error); | |
6409 | } | |
9ae529ec | 6410 | } |
c24fa4b1 | 6411 | if (!has_allocclass && zfs_special_devs(nvroot, NULL)) { |
715c996d | 6412 | spa_deactivate(spa); |
6413 | spa_remove(spa); | |
6414 | mutex_exit(&spa_namespace_lock); | |
6415 | return (ENOTSUP); | |
6416 | } | |
9ae529ec CS |
6417 | |
6418 | if (has_features || nvlist_lookup_uint64(props, | |
6419 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
34dc7c2f | 6420 | version = SPA_VERSION; |
9ae529ec CS |
6421 | } |
6422 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
428870ff BB |
6423 | |
6424 | spa->spa_first_txg = txg; | |
6425 | spa->spa_uberblock.ub_txg = txg - 1; | |
34dc7c2f BB |
6426 | spa->spa_uberblock.ub_version = version; |
6427 | spa->spa_ubsync = spa->spa_uberblock; | |
3dfb57a3 | 6428 | spa->spa_load_state = SPA_LOAD_CREATE; |
a1d477c2 MA |
6429 | spa->spa_removing_phys.sr_state = DSS_NONE; |
6430 | spa->spa_removing_phys.sr_removing_vdev = -1; | |
6431 | spa->spa_removing_phys.sr_prev_indirect_vdev = -1; | |
944a3724 | 6432 | spa->spa_indirect_vdevs_loaded = B_TRUE; |
34dc7c2f | 6433 | |
9babb374 BB |
6434 | /* |
6435 | * Create "The Godfather" zio to hold all async IOs | |
6436 | */ | |
e022864d MA |
6437 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
6438 | KM_SLEEP); | |
1c27024e | 6439 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
6440 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
6441 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
6442 | ZIO_FLAG_GODFATHER); | |
6443 | } | |
9babb374 | 6444 | |
34dc7c2f BB |
6445 | /* |
6446 | * Create the root vdev. | |
6447 | */ | |
b128c09f | 6448 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
6449 | |
6450 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
6451 | ||
6452 | ASSERT(error != 0 || rvd != NULL); | |
6453 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
6454 | ||
6455 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
2e528b49 | 6456 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
6457 | |
6458 | if (error == 0 && | |
6459 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
b2255edc BB |
6460 | (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 && |
6461 | (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { | |
cc99f275 DB |
6462 | /* |
6463 | * instantiate the metaslab groups (this will dirty the vdevs) | |
6464 | * we can no longer error exit past this point | |
6465 | */ | |
6466 | for (int c = 0; error == 0 && c < rvd->vdev_children; c++) { | |
6467 | vdev_t *vd = rvd->vdev_child[c]; | |
6468 | ||
6469 | vdev_metaslab_set_size(vd); | |
6470 | vdev_expand(vd, txg); | |
9babb374 | 6471 | } |
34dc7c2f BB |
6472 | } |
6473 | ||
b128c09f | 6474 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6475 | |
6476 | if (error != 0) { | |
6477 | spa_unload(spa); | |
6478 | spa_deactivate(spa); | |
6479 | spa_remove(spa); | |
6480 | mutex_exit(&spa_namespace_lock); | |
6481 | return (error); | |
6482 | } | |
6483 | ||
6484 | /* | |
6485 | * Get the list of spares, if specified. | |
6486 | */ | |
6487 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6488 | &spares, &nspares) == 0) { | |
65ad5d11 AJ |
6489 | spa->spa_spares.sav_config = fnvlist_alloc(); |
6490 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
6491 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
6492 | nspares); | |
b128c09f | 6493 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6494 | spa_load_spares(spa); |
b128c09f | 6495 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6496 | spa->spa_spares.sav_sync = B_TRUE; |
6497 | } | |
6498 | ||
6499 | /* | |
6500 | * Get the list of level 2 cache devices, if specified. | |
6501 | */ | |
6502 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6503 | &l2cache, &nl2cache) == 0) { | |
795075e6 PD |
6504 | VERIFY0(nvlist_alloc(&spa->spa_l2cache.sav_config, |
6505 | NV_UNIQUE_NAME, KM_SLEEP)); | |
65ad5d11 | 6506 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, |
795075e6 PD |
6507 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
6508 | nl2cache); | |
b128c09f | 6509 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6510 | spa_load_l2cache(spa); |
b128c09f | 6511 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6512 | spa->spa_l2cache.sav_sync = B_TRUE; |
6513 | } | |
6514 | ||
9ae529ec | 6515 | spa->spa_is_initializing = B_TRUE; |
b5256303 | 6516 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); |
9ae529ec | 6517 | spa->spa_is_initializing = B_FALSE; |
34dc7c2f | 6518 | |
428870ff BB |
6519 | /* |
6520 | * Create DDTs (dedup tables). | |
6521 | */ | |
6522 | ddt_create(spa); | |
67a1b037 PJD |
6523 | /* |
6524 | * Create BRT table and BRT table object. | |
6525 | */ | |
6526 | brt_create(spa); | |
428870ff BB |
6527 | |
6528 | spa_update_dspace(spa); | |
6529 | ||
34dc7c2f BB |
6530 | tx = dmu_tx_create_assigned(dp, txg); |
6531 | ||
d5e024cb BB |
6532 | /* |
6533 | * Create the pool's history object. | |
6534 | */ | |
6535 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
6536 | spa_history_create_obj(spa, tx); | |
6537 | ||
6538 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
6539 | spa_history_log_version(spa, "create", tx); | |
6540 | ||
34dc7c2f BB |
6541 | /* |
6542 | * Create the pool config object. | |
6543 | */ | |
6544 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
b128c09f | 6545 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, |
34dc7c2f BB |
6546 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); |
6547 | ||
6548 | if (zap_add(spa->spa_meta_objset, | |
6549 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
6550 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
6551 | cmn_err(CE_PANIC, "failed to add pool config"); | |
6552 | } | |
6553 | ||
428870ff BB |
6554 | if (zap_add(spa->spa_meta_objset, |
6555 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
6556 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
6557 | cmn_err(CE_PANIC, "failed to add pool version"); | |
6558 | } | |
6559 | ||
34dc7c2f BB |
6560 | /* Newly created pools with the right version are always deflated. */ |
6561 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6562 | spa->spa_deflate = TRUE; | |
6563 | if (zap_add(spa->spa_meta_objset, | |
6564 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6565 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
6566 | cmn_err(CE_PANIC, "failed to add deflate"); | |
6567 | } | |
6568 | } | |
6569 | ||
6570 | /* | |
428870ff | 6571 | * Create the deferred-free bpobj. Turn off compression |
34dc7c2f BB |
6572 | * because sync-to-convergence takes longer if the blocksize |
6573 | * keeps changing. | |
6574 | */ | |
428870ff BB |
6575 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); |
6576 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
34dc7c2f | 6577 | ZIO_COMPRESS_OFF, tx); |
34dc7c2f | 6578 | if (zap_add(spa->spa_meta_objset, |
428870ff BB |
6579 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, |
6580 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
6581 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
34dc7c2f | 6582 | } |
428870ff BB |
6583 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, |
6584 | spa->spa_meta_objset, obj)); | |
34dc7c2f | 6585 | |
3c67d83a TH |
6586 | /* |
6587 | * Generate some random noise for salted checksums to operate on. | |
6588 | */ | |
6589 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
6590 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
6591 | ||
34dc7c2f BB |
6592 | /* |
6593 | * Set pool properties. | |
6594 | */ | |
6595 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
6596 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
6597 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
9babb374 | 6598 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); |
379ca9cf | 6599 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); |
1b939560 | 6600 | spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM); |
428870ff | 6601 | |
d164b209 BB |
6602 | if (props != NULL) { |
6603 | spa_configfile_set(spa, props, B_FALSE); | |
13fe0198 | 6604 | spa_sync_props(props, tx); |
d164b209 | 6605 | } |
34dc7c2f | 6606 | |
b2255edc BB |
6607 | for (int i = 0; i < ndraid; i++) |
6608 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6609 | ||
34dc7c2f BB |
6610 | dmu_tx_commit(tx); |
6611 | ||
6612 | spa->spa_sync_on = B_TRUE; | |
b5256303 | 6613 | txg_sync_start(dp); |
379ca9cf | 6614 | mmp_thread_start(spa); |
b5256303 | 6615 | txg_wait_synced(dp, txg); |
34dc7c2f | 6616 | |
9d5b5245 SD |
6617 | spa_spawn_aux_threads(spa); |
6618 | ||
55c12724 | 6619 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); |
34dc7c2f | 6620 | |
0c66c32d JG |
6621 | /* |
6622 | * Don't count references from objsets that are already closed | |
6623 | * and are making their way through the eviction process. | |
6624 | */ | |
6625 | spa_evicting_os_wait(spa); | |
424fd7c3 | 6626 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
3dfb57a3 | 6627 | spa->spa_load_state = SPA_LOAD_NONE; |
b128c09f | 6628 | |
4759342a JL |
6629 | spa_import_os(spa); |
6630 | ||
d164b209 BB |
6631 | mutex_exit(&spa_namespace_lock); |
6632 | ||
34dc7c2f BB |
6633 | return (0); |
6634 | } | |
6635 | ||
9babb374 BB |
6636 | /* |
6637 | * Import a non-root pool into the system. | |
6638 | */ | |
6639 | int | |
13fe0198 | 6640 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) |
34dc7c2f BB |
6641 | { |
6642 | spa_t *spa; | |
d1807f16 | 6643 | const char *altroot = NULL; |
428870ff | 6644 | spa_load_state_t state = SPA_LOAD_IMPORT; |
8a393be3 | 6645 | zpool_load_policy_t policy; |
da92d5cb | 6646 | spa_mode_t mode = spa_mode_global; |
572e2857 | 6647 | uint64_t readonly = B_FALSE; |
9babb374 | 6648 | int error; |
34dc7c2f BB |
6649 | nvlist_t *nvroot; |
6650 | nvlist_t **spares, **l2cache; | |
6651 | uint_t nspares, nl2cache; | |
34dc7c2f BB |
6652 | |
6653 | /* | |
6654 | * If a pool with this name exists, return failure. | |
6655 | */ | |
6656 | mutex_enter(&spa_namespace_lock); | |
428870ff | 6657 | if (spa_lookup(pool) != NULL) { |
9babb374 | 6658 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 6659 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
6660 | } |
6661 | ||
6662 | /* | |
6663 | * Create and initialize the spa structure. | |
6664 | */ | |
6665 | (void) nvlist_lookup_string(props, | |
6666 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
572e2857 BB |
6667 | (void) nvlist_lookup_uint64(props, |
6668 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
6669 | if (readonly) | |
da92d5cb | 6670 | mode = SPA_MODE_READ; |
428870ff | 6671 | spa = spa_add(pool, config, altroot); |
572e2857 BB |
6672 | spa->spa_import_flags = flags; |
6673 | ||
6674 | /* | |
6675 | * Verbatim import - Take a pool and insert it into the namespace | |
6676 | * as if it had been loaded at boot. | |
6677 | */ | |
6678 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
6679 | if (props != NULL) | |
6680 | spa_configfile_set(spa, props, B_FALSE); | |
6681 | ||
55c12724 | 6682 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE); |
12fa0466 | 6683 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
4a0ee12a | 6684 | zfs_dbgmsg("spa_import: verbatim import of %s", pool); |
572e2857 | 6685 | mutex_exit(&spa_namespace_lock); |
572e2857 BB |
6686 | return (0); |
6687 | } | |
6688 | ||
6689 | spa_activate(spa, mode); | |
34dc7c2f | 6690 | |
9babb374 BB |
6691 | /* |
6692 | * Don't start async tasks until we know everything is healthy. | |
6693 | */ | |
6694 | spa_async_suspend(spa); | |
b128c09f | 6695 | |
8a393be3 PZ |
6696 | zpool_get_load_policy(config, &policy); |
6697 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) | |
572e2857 BB |
6698 | state = SPA_LOAD_RECOVER; |
6699 | ||
6cb8e530 | 6700 | spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; |
572e2857 | 6701 | |
6cb8e530 PZ |
6702 | if (state != SPA_LOAD_RECOVER) { |
6703 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6704 | zfs_dbgmsg("spa_import: importing %s", pool); | |
6705 | } else { | |
6706 | zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " | |
8a393be3 | 6707 | "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); |
6cb8e530 | 6708 | } |
8a393be3 | 6709 | error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); |
428870ff BB |
6710 | |
6711 | /* | |
572e2857 BB |
6712 | * Propagate anything learned while loading the pool and pass it |
6713 | * back to caller (i.e. rewind info, missing devices, etc). | |
428870ff | 6714 | */ |
65ad5d11 | 6715 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info); |
34dc7c2f | 6716 | |
b128c09f | 6717 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6718 | /* |
9babb374 BB |
6719 | * Toss any existing sparelist, as it doesn't have any validity |
6720 | * anymore, and conflicts with spa_has_spare(). | |
34dc7c2f | 6721 | */ |
9babb374 | 6722 | if (spa->spa_spares.sav_config) { |
34dc7c2f BB |
6723 | nvlist_free(spa->spa_spares.sav_config); |
6724 | spa->spa_spares.sav_config = NULL; | |
6725 | spa_load_spares(spa); | |
6726 | } | |
9babb374 | 6727 | if (spa->spa_l2cache.sav_config) { |
34dc7c2f BB |
6728 | nvlist_free(spa->spa_l2cache.sav_config); |
6729 | spa->spa_l2cache.sav_config = NULL; | |
6730 | spa_load_l2cache(spa); | |
6731 | } | |
6732 | ||
65ad5d11 | 6733 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
b128c09f | 6734 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 6735 | |
d164b209 BB |
6736 | if (props != NULL) |
6737 | spa_configfile_set(spa, props, B_FALSE); | |
6738 | ||
fb5f0bc8 BB |
6739 | if (error != 0 || (props && spa_writeable(spa) && |
6740 | (error = spa_prop_set(spa, props)))) { | |
9babb374 BB |
6741 | spa_unload(spa); |
6742 | spa_deactivate(spa); | |
6743 | spa_remove(spa); | |
34dc7c2f BB |
6744 | mutex_exit(&spa_namespace_lock); |
6745 | return (error); | |
6746 | } | |
6747 | ||
572e2857 BB |
6748 | spa_async_resume(spa); |
6749 | ||
34dc7c2f BB |
6750 | /* |
6751 | * Override any spares and level 2 cache devices as specified by | |
6752 | * the user, as these may have correct device names/devids, etc. | |
6753 | */ | |
6754 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6755 | &spares, &nspares) == 0) { | |
6756 | if (spa->spa_spares.sav_config) | |
65ad5d11 AJ |
6757 | fnvlist_remove(spa->spa_spares.sav_config, |
6758 | ZPOOL_CONFIG_SPARES); | |
34dc7c2f | 6759 | else |
65ad5d11 AJ |
6760 | spa->spa_spares.sav_config = fnvlist_alloc(); |
6761 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
6762 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
6763 | nspares); | |
b128c09f | 6764 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6765 | spa_load_spares(spa); |
b128c09f | 6766 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6767 | spa->spa_spares.sav_sync = B_TRUE; |
6768 | } | |
6769 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6770 | &l2cache, &nl2cache) == 0) { | |
6771 | if (spa->spa_l2cache.sav_config) | |
65ad5d11 AJ |
6772 | fnvlist_remove(spa->spa_l2cache.sav_config, |
6773 | ZPOOL_CONFIG_L2CACHE); | |
34dc7c2f | 6774 | else |
65ad5d11 AJ |
6775 | spa->spa_l2cache.sav_config = fnvlist_alloc(); |
6776 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
795075e6 PD |
6777 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
6778 | nl2cache); | |
b128c09f | 6779 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6780 | spa_load_l2cache(spa); |
b128c09f | 6781 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6782 | spa->spa_l2cache.sav_sync = B_TRUE; |
6783 | } | |
6784 | ||
428870ff BB |
6785 | /* |
6786 | * Check for any removed devices. | |
6787 | */ | |
6788 | if (spa->spa_autoreplace) { | |
6789 | spa_aux_check_removed(&spa->spa_spares); | |
6790 | spa_aux_check_removed(&spa->spa_l2cache); | |
6791 | } | |
6792 | ||
fb5f0bc8 | 6793 | if (spa_writeable(spa)) { |
b128c09f BB |
6794 | /* |
6795 | * Update the config cache to include the newly-imported pool. | |
6796 | */ | |
45d1cae3 | 6797 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
b128c09f | 6798 | } |
34dc7c2f | 6799 | |
34dc7c2f | 6800 | /* |
9babb374 BB |
6801 | * It's possible that the pool was expanded while it was exported. |
6802 | * We kick off an async task to handle this for us. | |
34dc7c2f | 6803 | */ |
9babb374 | 6804 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); |
b128c09f | 6805 | |
d5e024cb | 6806 | spa_history_log_version(spa, "import", NULL); |
fb390aaf | 6807 | |
12fa0466 | 6808 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
fb390aaf | 6809 | |
fb390aaf HR |
6810 | mutex_exit(&spa_namespace_lock); |
6811 | ||
ec213971 | 6812 | zvol_create_minors_recursive(pool); |
4a22ba5b | 6813 | |
4759342a JL |
6814 | spa_import_os(spa); |
6815 | ||
b128c09f BB |
6816 | return (0); |
6817 | } | |
6818 | ||
34dc7c2f BB |
6819 | nvlist_t * |
6820 | spa_tryimport(nvlist_t *tryconfig) | |
6821 | { | |
6822 | nvlist_t *config = NULL; | |
d1807f16 | 6823 | const char *poolname, *cachefile; |
34dc7c2f BB |
6824 | spa_t *spa; |
6825 | uint64_t state; | |
d164b209 | 6826 | int error; |
8a393be3 | 6827 | zpool_load_policy_t policy; |
34dc7c2f BB |
6828 | |
6829 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
6830 | return (NULL); | |
6831 | ||
6832 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
6833 | return (NULL); | |
6834 | ||
6835 | /* | |
6836 | * Create and initialize the spa structure. | |
6837 | */ | |
c183d164 GW |
6838 | char *name = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
6839 | (void) snprintf(name, MAXPATHLEN, "%s-%llx-%s", | |
6840 | TRYIMPORT_NAME, (u_longlong_t)curthread, poolname); | |
6841 | ||
34dc7c2f | 6842 | mutex_enter(&spa_namespace_lock); |
c183d164 | 6843 | spa = spa_add(name, tryconfig, NULL); |
da92d5cb | 6844 | spa_activate(spa, SPA_MODE_READ); |
c183d164 | 6845 | kmem_free(name, MAXPATHLEN); |
34dc7c2f BB |
6846 | |
6847 | /* | |
8a393be3 | 6848 | * Rewind pool if a max txg was provided. |
34dc7c2f | 6849 | */ |
8a393be3 PZ |
6850 | zpool_get_load_policy(spa->spa_config, &policy); |
6851 | if (policy.zlp_txg != UINT64_MAX) { | |
6852 | spa->spa_load_max_txg = policy.zlp_txg; | |
6cb8e530 PZ |
6853 | spa->spa_extreme_rewind = B_TRUE; |
6854 | zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", | |
8a393be3 | 6855 | poolname, (longlong_t)policy.zlp_txg); |
6cb8e530 PZ |
6856 | } else { |
6857 | zfs_dbgmsg("spa_tryimport: importing %s", poolname); | |
6858 | } | |
6859 | ||
6860 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) | |
6861 | == 0) { | |
6862 | zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); | |
6863 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; | |
6864 | } else { | |
6865 | spa->spa_config_source = SPA_CONFIG_SRC_SCAN; | |
6866 | } | |
6867 | ||
82ac409a AH |
6868 | /* |
6869 | * spa_import() relies on a pool config fetched by spa_try_import() | |
6870 | * for spare/cache devices. Import flags are not passed to | |
6871 | * spa_tryimport(), which makes it return early due to a missing log | |
6872 | * device and missing retrieving the cache device and spare eventually. | |
6873 | * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch | |
6874 | * the correct configuration regardless of the missing log device. | |
6875 | */ | |
6876 | spa->spa_import_flags |= ZFS_IMPORT_MISSING_LOG; | |
6877 | ||
6cb8e530 | 6878 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); |
34dc7c2f BB |
6879 | |
6880 | /* | |
6881 | * If 'tryconfig' was at least parsable, return the current config. | |
6882 | */ | |
6883 | if (spa->spa_root_vdev != NULL) { | |
34dc7c2f | 6884 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
65ad5d11 AJ |
6885 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname); |
6886 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state); | |
6887 | fnvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
6888 | spa->spa_uberblock.ub_timestamp); | |
6889 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
6890 | spa->spa_load_info); | |
6891 | fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
6892 | spa->spa_errata); | |
34dc7c2f BB |
6893 | |
6894 | /* | |
6895 | * If the bootfs property exists on this pool then we | |
6896 | * copy it out so that external consumers can tell which | |
6897 | * pools are bootable. | |
6898 | */ | |
d164b209 | 6899 | if ((!error || error == EEXIST) && spa->spa_bootfs) { |
79c76d5b | 6900 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6901 | |
6902 | /* | |
6903 | * We have to play games with the name since the | |
6904 | * pool was opened as TRYIMPORT_NAME. | |
6905 | */ | |
b128c09f | 6906 | if (dsl_dsobj_to_dsname(spa_name(spa), |
34dc7c2f BB |
6907 | spa->spa_bootfs, tmpname) == 0) { |
6908 | char *cp; | |
d1d7e268 MK |
6909 | char *dsname; |
6910 | ||
79c76d5b | 6911 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6912 | |
6913 | cp = strchr(tmpname, '/'); | |
6914 | if (cp == NULL) { | |
6915 | (void) strlcpy(dsname, tmpname, | |
6916 | MAXPATHLEN); | |
6917 | } else { | |
6918 | (void) snprintf(dsname, MAXPATHLEN, | |
6919 | "%s/%s", poolname, ++cp); | |
6920 | } | |
65ad5d11 AJ |
6921 | fnvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, |
6922 | dsname); | |
34dc7c2f BB |
6923 | kmem_free(dsname, MAXPATHLEN); |
6924 | } | |
6925 | kmem_free(tmpname, MAXPATHLEN); | |
6926 | } | |
6927 | ||
6928 | /* | |
6929 | * Add the list of hot spares and level 2 cache devices. | |
6930 | */ | |
9babb374 | 6931 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
34dc7c2f BB |
6932 | spa_add_spares(spa, config); |
6933 | spa_add_l2cache(spa, config); | |
9babb374 | 6934 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f BB |
6935 | } |
6936 | ||
6937 | spa_unload(spa); | |
6938 | spa_deactivate(spa); | |
6939 | spa_remove(spa); | |
6940 | mutex_exit(&spa_namespace_lock); | |
6941 | ||
6942 | return (config); | |
6943 | } | |
6944 | ||
6945 | /* | |
6946 | * Pool export/destroy | |
6947 | * | |
6948 | * The act of destroying or exporting a pool is very simple. We make sure there | |
6949 | * is no more pending I/O and any references to the pool are gone. Then, we | |
6950 | * update the pool state and sync all the labels to disk, removing the | |
fb5f0bc8 BB |
6951 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then |
6952 | * we don't sync the labels or remove the configuration cache. | |
34dc7c2f BB |
6953 | */ |
6954 | static int | |
4d55ea81 | 6955 | spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig, |
fb5f0bc8 | 6956 | boolean_t force, boolean_t hardforce) |
34dc7c2f | 6957 | { |
f4f50a70 | 6958 | int error; |
34dc7c2f | 6959 | spa_t *spa; |
c183d164 | 6960 | hrtime_t export_start = gethrtime(); |
34dc7c2f BB |
6961 | |
6962 | if (oldconfig) | |
6963 | *oldconfig = NULL; | |
6964 | ||
da92d5cb | 6965 | if (!(spa_mode_global & SPA_MODE_WRITE)) |
2e528b49 | 6966 | return (SET_ERROR(EROFS)); |
34dc7c2f BB |
6967 | |
6968 | mutex_enter(&spa_namespace_lock); | |
6969 | if ((spa = spa_lookup(pool)) == NULL) { | |
6970 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 6971 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
6972 | } |
6973 | ||
43a85362 SD |
6974 | if (spa->spa_is_exporting) { |
6975 | /* the pool is being exported by another thread */ | |
6976 | mutex_exit(&spa_namespace_lock); | |
6977 | return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS)); | |
6978 | } | |
6979 | spa->spa_is_exporting = B_TRUE; | |
6980 | ||
34dc7c2f BB |
6981 | /* |
6982 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
6983 | * reacquire the namespace lock, and see if we can export. | |
6984 | */ | |
6985 | spa_open_ref(spa, FTAG); | |
6986 | mutex_exit(&spa_namespace_lock); | |
6987 | spa_async_suspend(spa); | |
a0bd735a BP |
6988 | if (spa->spa_zvol_taskq) { |
6989 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
6990 | taskq_wait(spa->spa_zvol_taskq); | |
6991 | } | |
34dc7c2f BB |
6992 | mutex_enter(&spa_namespace_lock); |
6993 | spa_close(spa, FTAG); | |
6994 | ||
d14cfd83 IH |
6995 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) |
6996 | goto export_spa; | |
34dc7c2f | 6997 | /* |
d14cfd83 IH |
6998 | * The pool will be in core if it's openable, in which case we can |
6999 | * modify its state. Objsets may be open only because they're dirty, | |
7000 | * so we have to force it to sync before checking spa_refcnt. | |
34dc7c2f | 7001 | */ |
0c66c32d | 7002 | if (spa->spa_sync_on) { |
34dc7c2f | 7003 | txg_wait_synced(spa->spa_dsl_pool, 0); |
0c66c32d JG |
7004 | spa_evicting_os_wait(spa); |
7005 | } | |
34dc7c2f | 7006 | |
d14cfd83 IH |
7007 | /* |
7008 | * A pool cannot be exported or destroyed if there are active | |
7009 | * references. If we are resetting a pool, allow references by | |
7010 | * fault injection handlers. | |
7011 | */ | |
f4f50a70 WA |
7012 | if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0)) { |
7013 | error = SET_ERROR(EBUSY); | |
7014 | goto fail; | |
d14cfd83 | 7015 | } |
34dc7c2f | 7016 | |
d14cfd83 | 7017 | if (spa->spa_sync_on) { |
88b199c2 | 7018 | vdev_t *rvd = spa->spa_root_vdev; |
b128c09f BB |
7019 | /* |
7020 | * A pool cannot be exported if it has an active shared spare. | |
7021 | * This is to prevent other pools stealing the active spare | |
7022 | * from an exported pool. At user's own will, such pool can | |
7023 | * be forcedly exported. | |
7024 | */ | |
7025 | if (!force && new_state == POOL_STATE_EXPORTED && | |
7026 | spa_has_active_shared_spare(spa)) { | |
f4f50a70 WA |
7027 | error = SET_ERROR(EXDEV); |
7028 | goto fail; | |
b128c09f | 7029 | } |
34dc7c2f | 7030 | |
619f0976 GW |
7031 | /* |
7032 | * We're about to export or destroy this pool. Make sure | |
1b939560 BB |
7033 | * we stop all initialization and trim activity here before |
7034 | * we set the spa_final_txg. This will ensure that all | |
619f0976 GW |
7035 | * dirty data resulting from the initialization is |
7036 | * committed to disk before we unload the pool. | |
7037 | */ | |
88b199c2 RY |
7038 | vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE); |
7039 | vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE); | |
7040 | vdev_autotrim_stop_all(spa); | |
7041 | vdev_rebuild_stop_all(spa); | |
619f0976 | 7042 | |
34dc7c2f BB |
7043 | /* |
7044 | * We want this to be reflected on every label, | |
7045 | * so mark them all dirty. spa_unload() will do the | |
7046 | * final sync that pushes these changes out. | |
7047 | */ | |
fb5f0bc8 | 7048 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { |
b128c09f | 7049 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 7050 | spa->spa_state = new_state; |
88b199c2 | 7051 | vdev_config_dirty(rvd); |
2fb52853 GA |
7052 | spa_config_exit(spa, SCL_ALL, FTAG); |
7053 | } | |
7054 | ||
7055 | /* | |
7056 | * If the log space map feature is enabled and the pool is | |
7057 | * getting exported (but not destroyed), we want to spend some | |
7058 | * time flushing as many metaslabs as we can in an attempt to | |
7059 | * destroy log space maps and save import time. This has to be | |
7060 | * done before we set the spa_final_txg, otherwise | |
7061 | * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs. | |
7062 | * spa_should_flush_logs_on_unload() should be called after | |
7063 | * spa_state has been set to the new_state. | |
7064 | */ | |
7065 | if (spa_should_flush_logs_on_unload(spa)) | |
7066 | spa_unload_log_sm_flush_all(spa); | |
7067 | ||
7068 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
7069 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
428870ff BB |
7070 | spa->spa_final_txg = spa_last_synced_txg(spa) + |
7071 | TXG_DEFER_SIZE + 1; | |
b128c09f | 7072 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
7073 | } |
7074 | } | |
7075 | ||
d14cfd83 | 7076 | export_spa: |
4759342a JL |
7077 | spa_export_os(spa); |
7078 | ||
d5e024cb BB |
7079 | if (new_state == POOL_STATE_DESTROYED) |
7080 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
7081 | else if (new_state == POOL_STATE_EXPORTED) | |
7082 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
34dc7c2f BB |
7083 | |
7084 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
7085 | spa_unload(spa); | |
7086 | spa_deactivate(spa); | |
7087 | } | |
7088 | ||
7089 | if (oldconfig && spa->spa_config) | |
65ad5d11 | 7090 | *oldconfig = fnvlist_dup(spa->spa_config); |
34dc7c2f BB |
7091 | |
7092 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
fb5f0bc8 | 7093 | if (!hardforce) |
55c12724 | 7094 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 7095 | spa_remove(spa); |
43a85362 SD |
7096 | } else { |
7097 | /* | |
7098 | * If spa_remove() is not called for this spa_t and | |
7099 | * there is any possibility that it can be reused, | |
7100 | * we make sure to reset the exporting flag. | |
7101 | */ | |
7102 | spa->spa_is_exporting = B_FALSE; | |
34dc7c2f | 7103 | } |
34dc7c2f | 7104 | |
c183d164 GW |
7105 | if (new_state == POOL_STATE_EXPORTED) |
7106 | zio_handle_export_delay(spa, gethrtime() - export_start); | |
7107 | ||
43a85362 | 7108 | mutex_exit(&spa_namespace_lock); |
34dc7c2f | 7109 | return (0); |
f4f50a70 WA |
7110 | |
7111 | fail: | |
7112 | spa->spa_is_exporting = B_FALSE; | |
7113 | spa_async_resume(spa); | |
7114 | mutex_exit(&spa_namespace_lock); | |
7115 | return (error); | |
34dc7c2f BB |
7116 | } |
7117 | ||
7118 | /* | |
7119 | * Destroy a storage pool. | |
7120 | */ | |
7121 | int | |
4d55ea81 | 7122 | spa_destroy(const char *pool) |
34dc7c2f | 7123 | { |
fb5f0bc8 BB |
7124 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, |
7125 | B_FALSE, B_FALSE)); | |
34dc7c2f BB |
7126 | } |
7127 | ||
7128 | /* | |
7129 | * Export a storage pool. | |
7130 | */ | |
7131 | int | |
4d55ea81 | 7132 | spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force, |
fb5f0bc8 | 7133 | boolean_t hardforce) |
34dc7c2f | 7134 | { |
fb5f0bc8 BB |
7135 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, |
7136 | force, hardforce)); | |
34dc7c2f BB |
7137 | } |
7138 | ||
7139 | /* | |
7140 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
7141 | * from the namespace in any way. | |
7142 | */ | |
7143 | int | |
4d55ea81 | 7144 | spa_reset(const char *pool) |
34dc7c2f | 7145 | { |
b128c09f | 7146 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, |
fb5f0bc8 | 7147 | B_FALSE, B_FALSE)); |
34dc7c2f BB |
7148 | } |
7149 | ||
34dc7c2f BB |
7150 | /* |
7151 | * ========================================================================== | |
7152 | * Device manipulation | |
7153 | * ========================================================================== | |
7154 | */ | |
7155 | ||
b2255edc BB |
7156 | /* |
7157 | * This is called as a synctask to increment the draid feature flag | |
7158 | */ | |
7159 | static void | |
7160 | spa_draid_feature_incr(void *arg, dmu_tx_t *tx) | |
7161 | { | |
7162 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
7163 | int draid = (int)(uintptr_t)arg; | |
7164 | ||
7165 | for (int c = 0; c < draid; c++) | |
7166 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
7167 | } | |
7168 | ||
34dc7c2f BB |
7169 | /* |
7170 | * Add a device to a storage pool. | |
7171 | */ | |
7172 | int | |
b1e46f86 | 7173 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot, boolean_t check_ashift) |
34dc7c2f | 7174 | { |
b2255edc | 7175 | uint64_t txg, ndraid = 0; |
fb5f0bc8 | 7176 | int error; |
34dc7c2f BB |
7177 | vdev_t *rvd = spa->spa_root_vdev; |
7178 | vdev_t *vd, *tvd; | |
7179 | nvlist_t **spares, **l2cache; | |
7180 | uint_t nspares, nl2cache; | |
7181 | ||
572e2857 BB |
7182 | ASSERT(spa_writeable(spa)); |
7183 | ||
34dc7c2f BB |
7184 | txg = spa_vdev_enter(spa); |
7185 | ||
7186 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
7187 | VDEV_ALLOC_ADD)) != 0) | |
7188 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7189 | ||
b128c09f | 7190 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ |
34dc7c2f BB |
7191 | |
7192 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
7193 | &nspares) != 0) | |
7194 | nspares = 0; | |
7195 | ||
7196 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
7197 | &nl2cache) != 0) | |
7198 | nl2cache = 0; | |
7199 | ||
b128c09f | 7200 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) |
34dc7c2f | 7201 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
34dc7c2f | 7202 | |
b128c09f | 7203 | if (vd->vdev_children != 0 && |
b2255edc | 7204 | (error = vdev_create(vd, txg, B_FALSE)) != 0) { |
b128c09f | 7205 | return (spa_vdev_exit(spa, vd, txg, error)); |
b2255edc BB |
7206 | } |
7207 | ||
7208 | /* | |
7209 | * The virtual dRAID spares must be added after vdev tree is created | |
bf169e9f | 7210 | * and the vdev guids are generated. The guid of their associated |
b2255edc BB |
7211 | * dRAID is stored in the config and used when opening the spare. |
7212 | */ | |
7213 | if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid, | |
7214 | rvd->vdev_children)) == 0) { | |
7215 | if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot, | |
7216 | ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) | |
7217 | nspares = 0; | |
7218 | } else { | |
7219 | return (spa_vdev_exit(spa, vd, txg, error)); | |
7220 | } | |
34dc7c2f BB |
7221 | |
7222 | /* | |
7223 | * We must validate the spares and l2cache devices after checking the | |
7224 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
7225 | */ | |
b128c09f | 7226 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) |
34dc7c2f | 7227 | return (spa_vdev_exit(spa, vd, txg, error)); |
34dc7c2f BB |
7228 | |
7229 | /* | |
a1d477c2 MA |
7230 | * If we are in the middle of a device removal, we can only add |
7231 | * devices which match the existing devices in the pool. | |
7232 | * If we are in the middle of a removal, or have some indirect | |
b2255edc | 7233 | * vdevs, we can not add raidz or dRAID top levels. |
34dc7c2f | 7234 | */ |
a1d477c2 MA |
7235 | if (spa->spa_vdev_removal != NULL || |
7236 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
7237 | for (int c = 0; c < vd->vdev_children; c++) { | |
7238 | tvd = vd->vdev_child[c]; | |
7239 | if (spa->spa_vdev_removal != NULL && | |
9e052db4 | 7240 | tvd->vdev_ashift != spa->spa_max_ashift) { |
a1d477c2 MA |
7241 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
7242 | } | |
b2255edc BB |
7243 | /* Fail if top level vdev is raidz or a dRAID */ |
7244 | if (vdev_get_nparity(tvd) != 0) | |
a1d477c2 | 7245 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
b2255edc | 7246 | |
a1d477c2 MA |
7247 | /* |
7248 | * Need the top level mirror to be | |
7249 | * a mirror of leaf vdevs only | |
7250 | */ | |
7251 | if (tvd->vdev_ops == &vdev_mirror_ops) { | |
7252 | for (uint64_t cid = 0; | |
7253 | cid < tvd->vdev_children; cid++) { | |
7254 | vdev_t *cvd = tvd->vdev_child[cid]; | |
7255 | if (!cvd->vdev_ops->vdev_op_leaf) { | |
7256 | return (spa_vdev_exit(spa, vd, | |
7257 | txg, EINVAL)); | |
7258 | } | |
7259 | } | |
7260 | } | |
7261 | } | |
7262 | } | |
7263 | ||
b1e46f86 GW |
7264 | if (check_ashift && spa->spa_max_ashift == spa->spa_min_ashift) { |
7265 | for (int c = 0; c < vd->vdev_children; c++) { | |
7266 | tvd = vd->vdev_child[c]; | |
7267 | if (tvd->vdev_ashift != spa->spa_max_ashift) { | |
7268 | return (spa_vdev_exit(spa, vd, txg, | |
7269 | ZFS_ERR_ASHIFT_MISMATCH)); | |
7270 | } | |
7271 | } | |
7272 | } | |
7273 | ||
1c27024e | 7274 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
7275 | tvd = vd->vdev_child[c]; |
7276 | vdev_remove_child(vd, tvd); | |
93e28d66 | 7277 | tvd->vdev_id = rvd->vdev_children; |
34dc7c2f BB |
7278 | vdev_add_child(rvd, tvd); |
7279 | vdev_config_dirty(tvd); | |
7280 | } | |
7281 | ||
7282 | if (nspares != 0) { | |
7283 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
7284 | ZPOOL_CONFIG_SPARES); | |
7285 | spa_load_spares(spa); | |
7286 | spa->spa_spares.sav_sync = B_TRUE; | |
7287 | } | |
7288 | ||
7289 | if (nl2cache != 0) { | |
7290 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
7291 | ZPOOL_CONFIG_L2CACHE); | |
7292 | spa_load_l2cache(spa); | |
7293 | spa->spa_l2cache.sav_sync = B_TRUE; | |
7294 | } | |
7295 | ||
b2255edc BB |
7296 | /* |
7297 | * We can't increment a feature while holding spa_vdev so we | |
7298 | * have to do it in a synctask. | |
7299 | */ | |
7300 | if (ndraid != 0) { | |
7301 | dmu_tx_t *tx; | |
7302 | ||
7303 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
7304 | dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr, | |
7305 | (void *)(uintptr_t)ndraid, tx); | |
7306 | dmu_tx_commit(tx); | |
7307 | } | |
7308 | ||
34dc7c2f BB |
7309 | /* |
7310 | * We have to be careful when adding new vdevs to an existing pool. | |
7311 | * If other threads start allocating from these vdevs before we | |
7312 | * sync the config cache, and we lose power, then upon reboot we may | |
7313 | * fail to open the pool because there are DVAs that the config cache | |
7314 | * can't translate. Therefore, we first add the vdevs without | |
7315 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
7316 | * and then let spa_config_update() initialize the new metaslabs. | |
7317 | * | |
7318 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
7319 | * if we lose power at any point in this sequence, the remaining | |
7320 | * steps will be completed the next time we load the pool. | |
7321 | */ | |
7322 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
7323 | ||
7324 | mutex_enter(&spa_namespace_lock); | |
7325 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
12fa0466 | 7326 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); |
34dc7c2f BB |
7327 | mutex_exit(&spa_namespace_lock); |
7328 | ||
7329 | return (0); | |
7330 | } | |
7331 | ||
7332 | /* | |
5caeef02 DB |
7333 | * Attach a device to a vdev specified by its guid. The vdev type can be |
7334 | * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a | |
7335 | * single device). When the vdev is a single device, a mirror vdev will be | |
7336 | * automatically inserted. | |
34dc7c2f BB |
7337 | * |
7338 | * If 'replacing' is specified, the new device is intended to replace the | |
7339 | * existing device; in this case the two devices are made into their own | |
7340 | * mirror using the 'replacing' vdev, which is functionally identical to | |
7341 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
7342 | * extra rules: you can't attach to it after it's been created, and upon | |
7343 | * completion of resilvering, the first disk (the one being replaced) | |
7344 | * is automatically detached. | |
9a49d3f3 BB |
7345 | * |
7346 | * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild) | |
7347 | * should be performed instead of traditional healing reconstruction. From | |
7348 | * an administrators perspective these are both resilver operations. | |
34dc7c2f BB |
7349 | */ |
7350 | int | |
9a49d3f3 BB |
7351 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing, |
7352 | int rebuild) | |
34dc7c2f | 7353 | { |
428870ff | 7354 | uint64_t txg, dtl_max_txg; |
9a49d3f3 | 7355 | vdev_t *rvd = spa->spa_root_vdev; |
34dc7c2f BB |
7356 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; |
7357 | vdev_ops_t *pvops; | |
b128c09f | 7358 | char *oldvdpath, *newvdpath; |
5caeef02 | 7359 | int newvd_isspare = B_FALSE; |
b128c09f | 7360 | int error; |
34dc7c2f | 7361 | |
572e2857 BB |
7362 | ASSERT(spa_writeable(spa)); |
7363 | ||
34dc7c2f BB |
7364 | txg = spa_vdev_enter(spa); |
7365 | ||
b128c09f | 7366 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 7367 | |
d2734cce SD |
7368 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7369 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7370 | error = (spa_has_checkpoint(spa)) ? | |
7371 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7372 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7373 | } | |
7374 | ||
9a49d3f3 BB |
7375 | if (rebuild) { |
7376 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD)) | |
7377 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7378 | ||
9d618615 A |
7379 | if (dsl_scan_resilvering(spa_get_dsl(spa)) || |
7380 | dsl_scan_resilver_scheduled(spa_get_dsl(spa))) { | |
9a49d3f3 BB |
7381 | return (spa_vdev_exit(spa, NULL, txg, |
7382 | ZFS_ERR_RESILVER_IN_PROGRESS)); | |
9d618615 | 7383 | } |
9a49d3f3 BB |
7384 | } else { |
7385 | if (vdev_rebuild_active(rvd)) | |
7386 | return (spa_vdev_exit(spa, NULL, txg, | |
7387 | ZFS_ERR_REBUILD_IN_PROGRESS)); | |
7388 | } | |
7389 | ||
5caeef02 DB |
7390 | if (spa->spa_vdev_removal != NULL) { |
7391 | return (spa_vdev_exit(spa, NULL, txg, | |
7392 | ZFS_ERR_DEVRM_IN_PROGRESS)); | |
7393 | } | |
a1d477c2 | 7394 | |
34dc7c2f BB |
7395 | if (oldvd == NULL) |
7396 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7397 | ||
5caeef02 DB |
7398 | boolean_t raidz = oldvd->vdev_ops == &vdev_raidz_ops; |
7399 | ||
7400 | if (raidz) { | |
7401 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_RAIDZ_EXPANSION)) | |
7402 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7403 | ||
7404 | /* | |
7405 | * Can't expand a raidz while prior expand is in progress. | |
7406 | */ | |
7407 | if (spa->spa_raidz_expand != NULL) { | |
7408 | return (spa_vdev_exit(spa, NULL, txg, | |
7409 | ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS)); | |
7410 | } | |
7411 | } else if (!oldvd->vdev_ops->vdev_op_leaf) { | |
34dc7c2f | 7412 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
5caeef02 | 7413 | } |
34dc7c2f | 7414 | |
5caeef02 DB |
7415 | if (raidz) |
7416 | pvd = oldvd; | |
7417 | else | |
7418 | pvd = oldvd->vdev_parent; | |
34dc7c2f | 7419 | |
6a42939f RY |
7420 | if (spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, |
7421 | VDEV_ALLOC_ATTACH) != 0) | |
34dc7c2f BB |
7422 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); |
7423 | ||
7424 | if (newrootvd->vdev_children != 1) | |
7425 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7426 | ||
7427 | newvd = newrootvd->vdev_child[0]; | |
7428 | ||
7429 | if (!newvd->vdev_ops->vdev_op_leaf) | |
7430 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7431 | ||
7432 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
7433 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
7434 | ||
7435 | /* | |
c23738c7 | 7436 | * log, dedup and special vdevs should not be replaced by spares. |
34dc7c2f | 7437 | */ |
c23738c7 AH |
7438 | if ((oldvd->vdev_top->vdev_alloc_bias != VDEV_BIAS_NONE || |
7439 | oldvd->vdev_top->vdev_islog) && newvd->vdev_isspare) { | |
34dc7c2f | 7440 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
c23738c7 | 7441 | } |
34dc7c2f | 7442 | |
b2255edc BB |
7443 | /* |
7444 | * A dRAID spare can only replace a child of its parent dRAID vdev. | |
7445 | */ | |
7446 | if (newvd->vdev_ops == &vdev_draid_spare_ops && | |
7447 | oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) { | |
7448 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7449 | } | |
7450 | ||
9a49d3f3 BB |
7451 | if (rebuild) { |
7452 | /* | |
b2255edc | 7453 | * For rebuilds, the top vdev must support reconstruction |
9a49d3f3 | 7454 | * using only space maps. This means the only allowable |
b2255edc | 7455 | * vdevs types are the root vdev, a mirror, or dRAID. |
9a49d3f3 | 7456 | */ |
b2255edc BB |
7457 | tvd = pvd; |
7458 | if (pvd->vdev_top != NULL) | |
7459 | tvd = pvd->vdev_top; | |
7460 | ||
7461 | if (tvd->vdev_ops != &vdev_mirror_ops && | |
7462 | tvd->vdev_ops != &vdev_root_ops && | |
7463 | tvd->vdev_ops != &vdev_draid_ops) { | |
9a49d3f3 BB |
7464 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
7465 | } | |
7466 | } | |
7467 | ||
34dc7c2f BB |
7468 | if (!replacing) { |
7469 | /* | |
7bbd42ef DB |
7470 | * For attach, the only allowable parent is a mirror or |
7471 | * the root vdev. A raidz vdev can be attached to, but | |
7472 | * you cannot attach to a raidz child. | |
34dc7c2f BB |
7473 | */ |
7474 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
7bbd42ef DB |
7475 | pvd->vdev_ops != &vdev_root_ops && |
7476 | !raidz) | |
34dc7c2f BB |
7477 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
7478 | ||
7479 | pvops = &vdev_mirror_ops; | |
7480 | } else { | |
7481 | /* | |
7482 | * Active hot spares can only be replaced by inactive hot | |
7483 | * spares. | |
7484 | */ | |
7485 | if (pvd->vdev_ops == &vdev_spare_ops && | |
572e2857 | 7486 | oldvd->vdev_isspare && |
34dc7c2f BB |
7487 | !spa_has_spare(spa, newvd->vdev_guid)) |
7488 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7489 | ||
7490 | /* | |
7491 | * If the source is a hot spare, and the parent isn't already a | |
7492 | * spare, then we want to create a new hot spare. Otherwise, we | |
7493 | * want to create a replacing vdev. The user is not allowed to | |
7494 | * attach to a spared vdev child unless the 'isspare' state is | |
7495 | * the same (spare replaces spare, non-spare replaces | |
7496 | * non-spare). | |
7497 | */ | |
572e2857 BB |
7498 | if (pvd->vdev_ops == &vdev_replacing_ops && |
7499 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
34dc7c2f | 7500 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
7501 | } else if (pvd->vdev_ops == &vdev_spare_ops && |
7502 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
34dc7c2f | 7503 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
7504 | } |
7505 | ||
7506 | if (newvd->vdev_isspare) | |
34dc7c2f BB |
7507 | pvops = &vdev_spare_ops; |
7508 | else | |
7509 | pvops = &vdev_replacing_ops; | |
7510 | } | |
7511 | ||
7512 | /* | |
9babb374 | 7513 | * Make sure the new device is big enough. |
34dc7c2f | 7514 | */ |
5caeef02 DB |
7515 | vdev_t *min_vdev = raidz ? oldvd->vdev_child[0] : oldvd; |
7516 | if (newvd->vdev_asize < vdev_get_min_asize(min_vdev)) | |
34dc7c2f BB |
7517 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
7518 | ||
7519 | /* | |
7520 | * The new device cannot have a higher alignment requirement | |
7521 | * than the top-level vdev. | |
7522 | */ | |
7523 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
9a49d3f3 | 7524 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
34dc7c2f | 7525 | |
5caeef02 DB |
7526 | /* |
7527 | * RAIDZ-expansion-specific checks. | |
7528 | */ | |
7529 | if (raidz) { | |
7530 | if (vdev_raidz_attach_check(newvd) != 0) | |
7531 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7532 | ||
7533 | /* | |
7534 | * Fail early if a child is not healthy or being replaced | |
7535 | */ | |
7536 | for (int i = 0; i < oldvd->vdev_children; i++) { | |
7537 | if (vdev_is_dead(oldvd->vdev_child[i]) || | |
7538 | !oldvd->vdev_child[i]->vdev_ops->vdev_op_leaf) { | |
7539 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7540 | ENXIO)); | |
7541 | } | |
7542 | /* Also fail if reserved boot area is in-use */ | |
7543 | if (vdev_check_boot_reserve(spa, oldvd->vdev_child[i]) | |
7544 | != 0) { | |
7545 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7546 | EADDRINUSE)); | |
7547 | } | |
7548 | } | |
7549 | } | |
7550 | ||
7551 | if (raidz) { | |
7552 | /* | |
7553 | * Note: oldvdpath is freed by spa_strfree(), but | |
7554 | * kmem_asprintf() is freed by kmem_strfree(), so we have to | |
7555 | * move it to a spa_strdup-ed string. | |
7556 | */ | |
7557 | char *tmp = kmem_asprintf("raidz%u-%u", | |
7558 | (uint_t)vdev_get_nparity(oldvd), (uint_t)oldvd->vdev_id); | |
7559 | oldvdpath = spa_strdup(tmp); | |
7560 | kmem_strfree(tmp); | |
7561 | } else { | |
7562 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
7563 | } | |
7564 | newvdpath = spa_strdup(newvd->vdev_path); | |
7565 | ||
34dc7c2f BB |
7566 | /* |
7567 | * If this is an in-place replacement, update oldvd's path and devid | |
7568 | * to make it distinguishable from newvd, and unopenable from now on. | |
7569 | */ | |
5caeef02 | 7570 | if (strcmp(oldvdpath, newvdpath) == 0) { |
34dc7c2f | 7571 | spa_strfree(oldvd->vdev_path); |
5caeef02 | 7572 | oldvd->vdev_path = kmem_alloc(strlen(newvdpath) + 5, |
79c76d5b | 7573 | KM_SLEEP); |
5caeef02 DB |
7574 | (void) sprintf(oldvd->vdev_path, "%s/old", |
7575 | newvdpath); | |
34dc7c2f BB |
7576 | if (oldvd->vdev_devid != NULL) { |
7577 | spa_strfree(oldvd->vdev_devid); | |
7578 | oldvd->vdev_devid = NULL; | |
7579 | } | |
5caeef02 DB |
7580 | spa_strfree(oldvdpath); |
7581 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
34dc7c2f BB |
7582 | } |
7583 | ||
7584 | /* | |
7585 | * If the parent is not a mirror, or if we're replacing, insert the new | |
7586 | * mirror/replacing/spare vdev above oldvd. | |
7587 | */ | |
5caeef02 | 7588 | if (!raidz && pvd->vdev_ops != pvops) { |
34dc7c2f | 7589 | pvd = vdev_add_parent(oldvd, pvops); |
5caeef02 DB |
7590 | ASSERT(pvd->vdev_ops == pvops); |
7591 | ASSERT(oldvd->vdev_parent == pvd); | |
7592 | } | |
34dc7c2f BB |
7593 | |
7594 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
34dc7c2f BB |
7595 | |
7596 | /* | |
7597 | * Extract the new device from its root and add it to pvd. | |
7598 | */ | |
7599 | vdev_remove_child(newrootvd, newvd); | |
7600 | newvd->vdev_id = pvd->vdev_children; | |
428870ff | 7601 | newvd->vdev_crtxg = oldvd->vdev_crtxg; |
34dc7c2f BB |
7602 | vdev_add_child(pvd, newvd); |
7603 | ||
6d82f98c IH |
7604 | /* |
7605 | * Reevaluate the parent vdev state. | |
7606 | */ | |
7607 | vdev_propagate_state(pvd); | |
7608 | ||
34dc7c2f BB |
7609 | tvd = newvd->vdev_top; |
7610 | ASSERT(pvd->vdev_top == tvd); | |
7611 | ASSERT(tvd->vdev_parent == rvd); | |
7612 | ||
7613 | vdev_config_dirty(tvd); | |
7614 | ||
7615 | /* | |
428870ff BB |
7616 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account |
7617 | * for any dmu_sync-ed blocks. It will propagate upward when | |
7618 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
34dc7c2f | 7619 | */ |
428870ff | 7620 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; |
34dc7c2f | 7621 | |
5caeef02 DB |
7622 | if (raidz) { |
7623 | /* | |
7624 | * Wait for the youngest allocations and frees to sync, | |
7625 | * and then wait for the deferral of those frees to finish. | |
7626 | */ | |
7627 | spa_vdev_config_exit(spa, NULL, | |
7628 | txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); | |
34dc7c2f | 7629 | |
5caeef02 DB |
7630 | vdev_initialize_stop_all(tvd, VDEV_INITIALIZE_ACTIVE); |
7631 | vdev_trim_stop_all(tvd, VDEV_TRIM_ACTIVE); | |
7632 | vdev_autotrim_stop_wait(tvd); | |
9babb374 | 7633 | |
5caeef02 | 7634 | dtl_max_txg = spa_vdev_config_enter(spa); |
34dc7c2f | 7635 | |
5caeef02 | 7636 | tvd->vdev_rz_expanding = B_TRUE; |
34dc7c2f | 7637 | |
5caeef02 DB |
7638 | vdev_dirty_leaves(tvd, VDD_DTL, dtl_max_txg); |
7639 | vdev_config_dirty(tvd); | |
9a49d3f3 | 7640 | |
5caeef02 DB |
7641 | dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, |
7642 | dtl_max_txg); | |
7643 | dsl_sync_task_nowait(spa->spa_dsl_pool, vdev_raidz_attach_sync, | |
7644 | newvd, tx); | |
7645 | dmu_tx_commit(tx); | |
9a49d3f3 | 7646 | } else { |
5caeef02 DB |
7647 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, |
7648 | dtl_max_txg - TXG_INITIAL); | |
7649 | ||
7650 | if (newvd->vdev_isspare) { | |
7651 | spa_spare_activate(newvd); | |
7652 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); | |
7653 | } | |
7654 | ||
7655 | newvd_isspare = newvd->vdev_isspare; | |
7656 | ||
7657 | /* | |
7658 | * Mark newvd's DTL dirty in this txg. | |
7659 | */ | |
7660 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
9a49d3f3 | 7661 | |
5caeef02 DB |
7662 | /* |
7663 | * Schedule the resilver or rebuild to restart in the future. | |
7664 | * We do this to ensure that dmu_sync-ed blocks have been | |
7665 | * stitched into the respective datasets. | |
7666 | */ | |
7667 | if (rebuild) { | |
7668 | newvd->vdev_rebuild_txg = txg; | |
7669 | ||
7670 | vdev_rebuild(tvd); | |
9a49d3f3 | 7671 | } else { |
5caeef02 DB |
7672 | newvd->vdev_resilver_txg = txg; |
7673 | ||
7674 | if (dsl_scan_resilvering(spa_get_dsl(spa)) && | |
7675 | spa_feature_is_enabled(spa, | |
7676 | SPA_FEATURE_RESILVER_DEFER)) { | |
7677 | vdev_defer_resilver(newvd); | |
7678 | } else { | |
7679 | dsl_scan_restart_resilver(spa->spa_dsl_pool, | |
7680 | dtl_max_txg); | |
7681 | } | |
9a49d3f3 BB |
7682 | } |
7683 | } | |
428870ff | 7684 | |
fb390aaf | 7685 | if (spa->spa_bootfs) |
12fa0466 | 7686 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); |
fb390aaf | 7687 | |
12fa0466 | 7688 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); |
fb390aaf | 7689 | |
428870ff BB |
7690 | /* |
7691 | * Commit the config | |
7692 | */ | |
7693 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
34dc7c2f | 7694 | |
6f1ffb06 | 7695 | spa_history_log_internal(spa, "vdev attach", NULL, |
428870ff | 7696 | "%s vdev=%s %s vdev=%s", |
45d1cae3 BB |
7697 | replacing && newvd_isspare ? "spare in" : |
7698 | replacing ? "replace" : "attach", newvdpath, | |
7699 | replacing ? "for" : "to", oldvdpath); | |
b128c09f BB |
7700 | |
7701 | spa_strfree(oldvdpath); | |
7702 | spa_strfree(newvdpath); | |
7703 | ||
34dc7c2f BB |
7704 | return (0); |
7705 | } | |
7706 | ||
7707 | /* | |
7708 | * Detach a device from a mirror or replacing vdev. | |
d3cc8b15 | 7709 | * |
34dc7c2f | 7710 | * If 'replace_done' is specified, only detach if the parent |
719534ca | 7711 | * is a replacing or a spare vdev. |
34dc7c2f BB |
7712 | */ |
7713 | int | |
fb5f0bc8 | 7714 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) |
34dc7c2f BB |
7715 | { |
7716 | uint64_t txg; | |
fb5f0bc8 | 7717 | int error; |
2a8ba608 | 7718 | vdev_t *rvd __maybe_unused = spa->spa_root_vdev; |
34dc7c2f BB |
7719 | vdev_t *vd, *pvd, *cvd, *tvd; |
7720 | boolean_t unspare = B_FALSE; | |
d4ed6673 | 7721 | uint64_t unspare_guid = 0; |
428870ff | 7722 | char *vdpath; |
1c27024e | 7723 | |
572e2857 BB |
7724 | ASSERT(spa_writeable(spa)); |
7725 | ||
9a49d3f3 | 7726 | txg = spa_vdev_detach_enter(spa, guid); |
34dc7c2f | 7727 | |
b128c09f | 7728 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 7729 | |
d2734cce SD |
7730 | /* |
7731 | * Besides being called directly from the userland through the | |
7732 | * ioctl interface, spa_vdev_detach() can be potentially called | |
7733 | * at the end of spa_vdev_resilver_done(). | |
7734 | * | |
7735 | * In the regular case, when we have a checkpoint this shouldn't | |
7736 | * happen as we never empty the DTLs of a vdev during the scrub | |
7737 | * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() | |
7738 | * should never get here when we have a checkpoint. | |
7739 | * | |
7740 | * That said, even in a case when we checkpoint the pool exactly | |
7741 | * as spa_vdev_resilver_done() calls this function everything | |
7742 | * should be fine as the resilver will return right away. | |
7743 | */ | |
7744 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7745 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7746 | error = (spa_has_checkpoint(spa)) ? | |
7747 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7748 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7749 | } | |
7750 | ||
34dc7c2f BB |
7751 | if (vd == NULL) |
7752 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7753 | ||
7754 | if (!vd->vdev_ops->vdev_op_leaf) | |
7755 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7756 | ||
7757 | pvd = vd->vdev_parent; | |
7758 | ||
fb5f0bc8 BB |
7759 | /* |
7760 | * If the parent/child relationship is not as expected, don't do it. | |
7761 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
7762 | * vdev that's replacing B with C. The user's intent in replacing | |
7763 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
7764 | * the replace by detaching C, the expected behavior is to end up | |
7765 | * M(A,B). But suppose that right after deciding to detach C, | |
7766 | * the replacement of B completes. We would have M(A,C), and then | |
7767 | * ask to detach C, which would leave us with just A -- not what | |
7768 | * the user wanted. To prevent this, we make sure that the | |
7769 | * parent/child relationship hasn't changed -- in this example, | |
7770 | * that C's parent is still the replacing vdev R. | |
7771 | */ | |
7772 | if (pvd->vdev_guid != pguid && pguid != 0) | |
7773 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
7774 | ||
34dc7c2f | 7775 | /* |
572e2857 | 7776 | * Only 'replacing' or 'spare' vdevs can be replaced. |
34dc7c2f | 7777 | */ |
572e2857 BB |
7778 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && |
7779 | pvd->vdev_ops != &vdev_spare_ops) | |
7780 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
34dc7c2f BB |
7781 | |
7782 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
7783 | spa_version(spa) >= SPA_VERSION_SPARES); | |
7784 | ||
7785 | /* | |
7786 | * Only mirror, replacing, and spare vdevs support detach. | |
7787 | */ | |
7788 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
7789 | pvd->vdev_ops != &vdev_mirror_ops && | |
7790 | pvd->vdev_ops != &vdev_spare_ops) | |
7791 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7792 | ||
7793 | /* | |
fb5f0bc8 BB |
7794 | * If this device has the only valid copy of some data, |
7795 | * we cannot safely detach it. | |
34dc7c2f | 7796 | */ |
fb5f0bc8 | 7797 | if (vdev_dtl_required(vd)) |
34dc7c2f BB |
7798 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
7799 | ||
fb5f0bc8 | 7800 | ASSERT(pvd->vdev_children >= 2); |
34dc7c2f | 7801 | |
b128c09f BB |
7802 | /* |
7803 | * If we are detaching the second disk from a replacing vdev, then | |
7804 | * check to see if we changed the original vdev's path to have "/old" | |
7805 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
7806 | */ | |
572e2857 BB |
7807 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && |
7808 | vd->vdev_path != NULL) { | |
7809 | size_t len = strlen(vd->vdev_path); | |
7810 | ||
1c27024e | 7811 | for (int c = 0; c < pvd->vdev_children; c++) { |
572e2857 BB |
7812 | cvd = pvd->vdev_child[c]; |
7813 | ||
7814 | if (cvd == vd || cvd->vdev_path == NULL) | |
7815 | continue; | |
7816 | ||
7817 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
7818 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
7819 | spa_strfree(cvd->vdev_path); | |
7820 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
7821 | break; | |
7822 | } | |
b128c09f BB |
7823 | } |
7824 | } | |
7825 | ||
34dc7c2f | 7826 | /* |
b2255edc BB |
7827 | * If we are detaching the original disk from a normal spare, then it |
7828 | * implies that the spare should become a real disk, and be removed | |
7829 | * from the active spare list for the pool. dRAID spares on the | |
7830 | * other hand are coupled to the pool and thus should never be removed | |
7831 | * from the spares list. | |
34dc7c2f | 7832 | */ |
b2255edc BB |
7833 | if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) { |
7834 | vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
7835 | ||
7836 | if (last_cvd->vdev_isspare && | |
7837 | last_cvd->vdev_ops != &vdev_draid_spare_ops) { | |
7838 | unspare = B_TRUE; | |
7839 | } | |
7840 | } | |
34dc7c2f BB |
7841 | |
7842 | /* | |
7843 | * Erase the disk labels so the disk can be used for other things. | |
7844 | * This must be done after all other error cases are handled, | |
7845 | * but before we disembowel vd (so we can still do I/O to it). | |
7846 | * But if we can't do it, don't treat the error as fatal -- | |
7847 | * it may be that the unwritability of the disk is the reason | |
7848 | * it's being detached! | |
7849 | */ | |
6a42939f | 7850 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); |
34dc7c2f BB |
7851 | |
7852 | /* | |
7853 | * Remove vd from its parent and compact the parent's children. | |
7854 | */ | |
7855 | vdev_remove_child(pvd, vd); | |
7856 | vdev_compact_children(pvd); | |
7857 | ||
7858 | /* | |
7859 | * Remember one of the remaining children so we can get tvd below. | |
7860 | */ | |
572e2857 | 7861 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; |
34dc7c2f BB |
7862 | |
7863 | /* | |
7864 | * If we need to remove the remaining child from the list of hot spares, | |
fb5f0bc8 BB |
7865 | * do it now, marking the vdev as no longer a spare in the process. |
7866 | * We must do this before vdev_remove_parent(), because that can | |
7867 | * change the GUID if it creates a new toplevel GUID. For a similar | |
7868 | * reason, we must remove the spare now, in the same txg as the detach; | |
7869 | * otherwise someone could attach a new sibling, change the GUID, and | |
7870 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
34dc7c2f BB |
7871 | */ |
7872 | if (unspare) { | |
7873 | ASSERT(cvd->vdev_isspare); | |
7874 | spa_spare_remove(cvd); | |
7875 | unspare_guid = cvd->vdev_guid; | |
fb5f0bc8 | 7876 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); |
572e2857 | 7877 | cvd->vdev_unspare = B_TRUE; |
34dc7c2f BB |
7878 | } |
7879 | ||
428870ff BB |
7880 | /* |
7881 | * If the parent mirror/replacing vdev only has one child, | |
7882 | * the parent is no longer needed. Remove it from the tree. | |
7883 | */ | |
572e2857 BB |
7884 | if (pvd->vdev_children == 1) { |
7885 | if (pvd->vdev_ops == &vdev_spare_ops) | |
7886 | cvd->vdev_unspare = B_FALSE; | |
428870ff | 7887 | vdev_remove_parent(cvd); |
572e2857 BB |
7888 | } |
7889 | ||
428870ff BB |
7890 | /* |
7891 | * We don't set tvd until now because the parent we just removed | |
7892 | * may have been the previous top-level vdev. | |
7893 | */ | |
7894 | tvd = cvd->vdev_top; | |
7895 | ASSERT(tvd->vdev_parent == rvd); | |
7896 | ||
7897 | /* | |
7898 | * Reevaluate the parent vdev state. | |
7899 | */ | |
7900 | vdev_propagate_state(cvd); | |
7901 | ||
7902 | /* | |
7903 | * If the 'autoexpand' property is set on the pool then automatically | |
7904 | * try to expand the size of the pool. For example if the device we | |
7905 | * just detached was smaller than the others, it may be possible to | |
7906 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
7907 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
7908 | */ | |
7909 | if (spa->spa_autoexpand) { | |
7910 | vdev_reopen(tvd); | |
7911 | vdev_expand(tvd, txg); | |
7912 | } | |
7913 | ||
7914 | vdev_config_dirty(tvd); | |
7915 | ||
7916 | /* | |
7917 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
7918 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
7919 | * But first make sure we're not on any *other* txg's DTL list, to | |
7920 | * prevent vd from being accessed after it's freed. | |
7921 | */ | |
b6ca6193 | 7922 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); |
1c27024e | 7923 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff BB |
7924 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); |
7925 | vd->vdev_detached = B_TRUE; | |
7926 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
7927 | ||
12fa0466 | 7928 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); |
e60e158e | 7929 | spa_notify_waiters(spa); |
428870ff | 7930 | |
572e2857 BB |
7931 | /* hang on to the spa before we release the lock */ |
7932 | spa_open_ref(spa, FTAG); | |
7933 | ||
428870ff BB |
7934 | error = spa_vdev_exit(spa, vd, txg, 0); |
7935 | ||
6f1ffb06 | 7936 | spa_history_log_internal(spa, "detach", NULL, |
428870ff BB |
7937 | "vdev=%s", vdpath); |
7938 | spa_strfree(vdpath); | |
7939 | ||
7940 | /* | |
7941 | * If this was the removal of the original device in a hot spare vdev, | |
7942 | * then we want to go through and remove the device from the hot spare | |
7943 | * list of every other pool. | |
7944 | */ | |
7945 | if (unspare) { | |
572e2857 BB |
7946 | spa_t *altspa = NULL; |
7947 | ||
428870ff | 7948 | mutex_enter(&spa_namespace_lock); |
572e2857 BB |
7949 | while ((altspa = spa_next(altspa)) != NULL) { |
7950 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
7951 | altspa == spa) | |
428870ff | 7952 | continue; |
572e2857 BB |
7953 | |
7954 | spa_open_ref(altspa, FTAG); | |
428870ff | 7955 | mutex_exit(&spa_namespace_lock); |
572e2857 | 7956 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); |
428870ff | 7957 | mutex_enter(&spa_namespace_lock); |
572e2857 | 7958 | spa_close(altspa, FTAG); |
428870ff BB |
7959 | } |
7960 | mutex_exit(&spa_namespace_lock); | |
572e2857 BB |
7961 | |
7962 | /* search the rest of the vdevs for spares to remove */ | |
7963 | spa_vdev_resilver_done(spa); | |
428870ff BB |
7964 | } |
7965 | ||
572e2857 BB |
7966 | /* all done with the spa; OK to release */ |
7967 | mutex_enter(&spa_namespace_lock); | |
7968 | spa_close(spa, FTAG); | |
7969 | mutex_exit(&spa_namespace_lock); | |
7970 | ||
428870ff BB |
7971 | return (error); |
7972 | } | |
7973 | ||
c10d37dd GW |
7974 | static int |
7975 | spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7976 | list_t *vd_list) | |
619f0976 | 7977 | { |
c10d37dd GW |
7978 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7979 | ||
619f0976 GW |
7980 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); |
7981 | ||
7982 | /* Look up vdev and ensure it's a leaf. */ | |
7983 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7984 | if (vd == NULL || vd->vdev_detached) { | |
7985 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7986 | return (SET_ERROR(ENODEV)); |
7987 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
7988 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7989 | return (SET_ERROR(EINVAL)); |
7990 | } else if (!vdev_writeable(vd)) { | |
7991 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7992 | return (SET_ERROR(EROFS)); |
7993 | } | |
7994 | mutex_enter(&vd->vdev_initialize_lock); | |
7995 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7996 | ||
7997 | /* | |
7998 | * When we activate an initialize action we check to see | |
7999 | * if the vdev_initialize_thread is NULL. We do this instead | |
8000 | * of using the vdev_initialize_state since there might be | |
8001 | * a previous initialization process which has completed but | |
8002 | * the thread is not exited. | |
8003 | */ | |
1b939560 | 8004 | if (cmd_type == POOL_INITIALIZE_START && |
619f0976 | 8005 | (vd->vdev_initialize_thread != NULL || |
5caeef02 | 8006 | vd->vdev_top->vdev_removing || vd->vdev_top->vdev_rz_expanding)) { |
619f0976 | 8007 | mutex_exit(&vd->vdev_initialize_lock); |
619f0976 GW |
8008 | return (SET_ERROR(EBUSY)); |
8009 | } else if (cmd_type == POOL_INITIALIZE_CANCEL && | |
8010 | (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE && | |
8011 | vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) { | |
8012 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 GW |
8013 | return (SET_ERROR(ESRCH)); |
8014 | } else if (cmd_type == POOL_INITIALIZE_SUSPEND && | |
8015 | vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) { | |
8016 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 | 8017 | return (SET_ERROR(ESRCH)); |
e34e15ed BB |
8018 | } else if (cmd_type == POOL_INITIALIZE_UNINIT && |
8019 | vd->vdev_initialize_thread != NULL) { | |
8020 | mutex_exit(&vd->vdev_initialize_lock); | |
8021 | return (SET_ERROR(EBUSY)); | |
619f0976 GW |
8022 | } |
8023 | ||
8024 | switch (cmd_type) { | |
1b939560 | 8025 | case POOL_INITIALIZE_START: |
619f0976 GW |
8026 | vdev_initialize(vd); |
8027 | break; | |
8028 | case POOL_INITIALIZE_CANCEL: | |
c10d37dd | 8029 | vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list); |
619f0976 GW |
8030 | break; |
8031 | case POOL_INITIALIZE_SUSPEND: | |
c10d37dd | 8032 | vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list); |
619f0976 | 8033 | break; |
e34e15ed BB |
8034 | case POOL_INITIALIZE_UNINIT: |
8035 | vdev_uninitialize(vd); | |
8036 | break; | |
619f0976 GW |
8037 | default: |
8038 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
8039 | } | |
8040 | mutex_exit(&vd->vdev_initialize_lock); | |
8041 | ||
c10d37dd GW |
8042 | return (0); |
8043 | } | |
8044 | ||
8045 | int | |
8046 | spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, | |
8047 | nvlist_t *vdev_errlist) | |
8048 | { | |
8049 | int total_errors = 0; | |
8050 | list_t vd_list; | |
8051 | ||
8052 | list_create(&vd_list, sizeof (vdev_t), | |
8053 | offsetof(vdev_t, vdev_initialize_node)); | |
8054 | ||
8055 | /* | |
8056 | * We hold the namespace lock through the whole function | |
8057 | * to prevent any changes to the pool while we're starting or | |
8058 | * stopping initialization. The config and state locks are held so that | |
8059 | * we can properly assess the vdev state before we commit to | |
8060 | * the initializing operation. | |
8061 | */ | |
8062 | mutex_enter(&spa_namespace_lock); | |
8063 | ||
8064 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
8065 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
8066 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
8067 | ||
8068 | int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type, | |
8069 | &vd_list); | |
8070 | if (error != 0) { | |
8071 | char guid_as_str[MAXNAMELEN]; | |
8072 | ||
8073 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
8074 | "%llu", (unsigned long long)vdev_guid); | |
8075 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
8076 | total_errors++; | |
8077 | } | |
8078 | } | |
8079 | ||
8080 | /* Wait for all initialize threads to stop. */ | |
8081 | vdev_initialize_stop_wait(spa, &vd_list); | |
8082 | ||
619f0976 GW |
8083 | /* Sync out the initializing state */ |
8084 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
8085 | mutex_exit(&spa_namespace_lock); | |
8086 | ||
c10d37dd | 8087 | list_destroy(&vd_list); |
619f0976 | 8088 | |
c10d37dd GW |
8089 | return (total_errors); |
8090 | } | |
619f0976 | 8091 | |
1b939560 BB |
8092 | static int |
8093 | spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
8094 | uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list) | |
8095 | { | |
8096 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
8097 | ||
8098 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
8099 | ||
8100 | /* Look up vdev and ensure it's a leaf. */ | |
8101 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
8102 | if (vd == NULL || vd->vdev_detached) { | |
8103 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8104 | return (SET_ERROR(ENODEV)); | |
8105 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
8106 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8107 | return (SET_ERROR(EINVAL)); | |
8108 | } else if (!vdev_writeable(vd)) { | |
8109 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8110 | return (SET_ERROR(EROFS)); | |
8111 | } else if (!vd->vdev_has_trim) { | |
8112 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8113 | return (SET_ERROR(EOPNOTSUPP)); | |
8114 | } else if (secure && !vd->vdev_has_securetrim) { | |
8115 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8116 | return (SET_ERROR(EOPNOTSUPP)); | |
8117 | } | |
8118 | mutex_enter(&vd->vdev_trim_lock); | |
8119 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8120 | ||
8121 | /* | |
8122 | * When we activate a TRIM action we check to see if the | |
8123 | * vdev_trim_thread is NULL. We do this instead of using the | |
8124 | * vdev_trim_state since there might be a previous TRIM process | |
8125 | * which has completed but the thread is not exited. | |
8126 | */ | |
8127 | if (cmd_type == POOL_TRIM_START && | |
5caeef02 DB |
8128 | (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing || |
8129 | vd->vdev_top->vdev_rz_expanding)) { | |
1b939560 BB |
8130 | mutex_exit(&vd->vdev_trim_lock); |
8131 | return (SET_ERROR(EBUSY)); | |
8132 | } else if (cmd_type == POOL_TRIM_CANCEL && | |
8133 | (vd->vdev_trim_state != VDEV_TRIM_ACTIVE && | |
8134 | vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) { | |
8135 | mutex_exit(&vd->vdev_trim_lock); | |
8136 | return (SET_ERROR(ESRCH)); | |
8137 | } else if (cmd_type == POOL_TRIM_SUSPEND && | |
8138 | vd->vdev_trim_state != VDEV_TRIM_ACTIVE) { | |
8139 | mutex_exit(&vd->vdev_trim_lock); | |
8140 | return (SET_ERROR(ESRCH)); | |
8141 | } | |
8142 | ||
8143 | switch (cmd_type) { | |
8144 | case POOL_TRIM_START: | |
8145 | vdev_trim(vd, rate, partial, secure); | |
8146 | break; | |
8147 | case POOL_TRIM_CANCEL: | |
8148 | vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list); | |
8149 | break; | |
8150 | case POOL_TRIM_SUSPEND: | |
8151 | vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list); | |
8152 | break; | |
8153 | default: | |
8154 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
8155 | } | |
8156 | mutex_exit(&vd->vdev_trim_lock); | |
8157 | ||
8158 | return (0); | |
8159 | } | |
8160 | ||
8161 | /* | |
8162 | * Initiates a manual TRIM for the requested vdevs. This kicks off individual | |
8163 | * TRIM threads for each child vdev. These threads pass over all of the free | |
8164 | * space in the vdev's metaslabs and issues TRIM commands for that space. | |
8165 | */ | |
8166 | int | |
8167 | spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate, | |
8168 | boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist) | |
8169 | { | |
8170 | int total_errors = 0; | |
8171 | list_t vd_list; | |
8172 | ||
8173 | list_create(&vd_list, sizeof (vdev_t), | |
8174 | offsetof(vdev_t, vdev_trim_node)); | |
8175 | ||
8176 | /* | |
8177 | * We hold the namespace lock through the whole function | |
8178 | * to prevent any changes to the pool while we're starting or | |
8179 | * stopping TRIM. The config and state locks are held so that | |
8180 | * we can properly assess the vdev state before we commit to | |
8181 | * the TRIM operation. | |
8182 | */ | |
8183 | mutex_enter(&spa_namespace_lock); | |
8184 | ||
8185 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
8186 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
8187 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
8188 | ||
8189 | int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type, | |
8190 | rate, partial, secure, &vd_list); | |
8191 | if (error != 0) { | |
8192 | char guid_as_str[MAXNAMELEN]; | |
8193 | ||
8194 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
8195 | "%llu", (unsigned long long)vdev_guid); | |
8196 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
8197 | total_errors++; | |
8198 | } | |
8199 | } | |
8200 | ||
8201 | /* Wait for all TRIM threads to stop. */ | |
8202 | vdev_trim_stop_wait(spa, &vd_list); | |
8203 | ||
8204 | /* Sync out the TRIM state */ | |
8205 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
8206 | mutex_exit(&spa_namespace_lock); | |
8207 | ||
8208 | list_destroy(&vd_list); | |
8209 | ||
8210 | return (total_errors); | |
8211 | } | |
8212 | ||
428870ff BB |
8213 | /* |
8214 | * Split a set of devices from their mirrors, and create a new pool from them. | |
8215 | */ | |
8216 | int | |
a926aab9 | 8217 | spa_vdev_split_mirror(spa_t *spa, const char *newname, nvlist_t *config, |
428870ff BB |
8218 | nvlist_t *props, boolean_t exp) |
8219 | { | |
8220 | int error = 0; | |
8221 | uint64_t txg, *glist; | |
8222 | spa_t *newspa; | |
8223 | uint_t c, children, lastlog; | |
8224 | nvlist_t **child, *nvl, *tmp; | |
8225 | dmu_tx_t *tx; | |
d1807f16 | 8226 | const char *altroot = NULL; |
428870ff BB |
8227 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ |
8228 | boolean_t activate_slog; | |
8229 | ||
572e2857 | 8230 | ASSERT(spa_writeable(spa)); |
428870ff BB |
8231 | |
8232 | txg = spa_vdev_enter(spa); | |
8233 | ||
d2734cce SD |
8234 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
8235 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
8236 | error = (spa_has_checkpoint(spa)) ? | |
8237 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
8238 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8239 | } | |
8240 | ||
428870ff BB |
8241 | /* clear the log and flush everything up to now */ |
8242 | activate_slog = spa_passivate_log(spa); | |
8243 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
a1d477c2 | 8244 | error = spa_reset_logs(spa); |
428870ff BB |
8245 | txg = spa_vdev_config_enter(spa); |
8246 | ||
8247 | if (activate_slog) | |
8248 | spa_activate_log(spa); | |
8249 | ||
8250 | if (error != 0) | |
8251 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8252 | ||
8253 | /* check new spa name before going any further */ | |
8254 | if (spa_lookup(newname) != NULL) | |
8255 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
8256 | ||
8257 | /* | |
8258 | * scan through all the children to ensure they're all mirrors | |
8259 | */ | |
8260 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
8261 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
8262 | &children) != 0) | |
8263 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8264 | ||
8265 | /* first, check to ensure we've got the right child count */ | |
8266 | rvd = spa->spa_root_vdev; | |
8267 | lastlog = 0; | |
8268 | for (c = 0; c < rvd->vdev_children; c++) { | |
8269 | vdev_t *vd = rvd->vdev_child[c]; | |
8270 | ||
8271 | /* don't count the holes & logs as children */ | |
1b664952 GA |
8272 | if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops && |
8273 | !vdev_is_concrete(vd))) { | |
428870ff BB |
8274 | if (lastlog == 0) |
8275 | lastlog = c; | |
8276 | continue; | |
8277 | } | |
8278 | ||
8279 | lastlog = 0; | |
8280 | } | |
8281 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
8282 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8283 | ||
8284 | /* next, ensure no spare or cache devices are part of the split */ | |
8285 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
8286 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
8287 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8288 | ||
79c76d5b BB |
8289 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); |
8290 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
428870ff BB |
8291 | |
8292 | /* then, loop over each vdev and validate it */ | |
8293 | for (c = 0; c < children; c++) { | |
8294 | uint64_t is_hole = 0; | |
8295 | ||
8296 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
8297 | &is_hole); | |
8298 | ||
8299 | if (is_hole != 0) { | |
8300 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
8301 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
8302 | continue; | |
8303 | } else { | |
2e528b49 | 8304 | error = SET_ERROR(EINVAL); |
428870ff BB |
8305 | break; |
8306 | } | |
8307 | } | |
8308 | ||
1b664952 GA |
8309 | /* deal with indirect vdevs */ |
8310 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ops == | |
8311 | &vdev_indirect_ops) | |
8312 | continue; | |
8313 | ||
428870ff BB |
8314 | /* which disk is going to be split? */ |
8315 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
8316 | &glist[c]) != 0) { | |
2e528b49 | 8317 | error = SET_ERROR(EINVAL); |
428870ff BB |
8318 | break; |
8319 | } | |
8320 | ||
8321 | /* look it up in the spa */ | |
8322 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
8323 | if (vml[c] == NULL) { | |
2e528b49 | 8324 | error = SET_ERROR(ENODEV); |
428870ff BB |
8325 | break; |
8326 | } | |
8327 | ||
8328 | /* make sure there's nothing stopping the split */ | |
8329 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
8330 | vml[c]->vdev_islog || | |
a1d477c2 | 8331 | !vdev_is_concrete(vml[c]) || |
428870ff BB |
8332 | vml[c]->vdev_isspare || |
8333 | vml[c]->vdev_isl2cache || | |
8334 | !vdev_writeable(vml[c]) || | |
8335 | vml[c]->vdev_children != 0 || | |
8336 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
8337 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
2e528b49 | 8338 | error = SET_ERROR(EINVAL); |
428870ff BB |
8339 | break; |
8340 | } | |
8341 | ||
733b5722 RS |
8342 | if (vdev_dtl_required(vml[c]) || |
8343 | vdev_resilver_needed(vml[c], NULL, NULL)) { | |
2e528b49 | 8344 | error = SET_ERROR(EBUSY); |
428870ff BB |
8345 | break; |
8346 | } | |
8347 | ||
8348 | /* we need certain info from the top level */ | |
65ad5d11 AJ |
8349 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, |
8350 | vml[c]->vdev_top->vdev_ms_array); | |
8351 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
8352 | vml[c]->vdev_top->vdev_ms_shift); | |
8353 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
8354 | vml[c]->vdev_top->vdev_asize); | |
8355 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
8356 | vml[c]->vdev_top->vdev_ashift); | |
e0ab3ab5 JS |
8357 | |
8358 | /* transfer per-vdev ZAPs */ | |
8359 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
8360 | VERIFY0(nvlist_add_uint64(child[c], | |
8361 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
8362 | ||
8363 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
8364 | VERIFY0(nvlist_add_uint64(child[c], | |
8365 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
8366 | vml[c]->vdev_parent->vdev_top_zap)); | |
428870ff BB |
8367 | } |
8368 | ||
8369 | if (error != 0) { | |
8370 | kmem_free(vml, children * sizeof (vdev_t *)); | |
8371 | kmem_free(glist, children * sizeof (uint64_t)); | |
8372 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8373 | } | |
8374 | ||
8375 | /* stop writers from using the disks */ | |
8376 | for (c = 0; c < children; c++) { | |
8377 | if (vml[c] != NULL) | |
8378 | vml[c]->vdev_offline = B_TRUE; | |
8379 | } | |
8380 | vdev_reopen(spa->spa_root_vdev); | |
34dc7c2f BB |
8381 | |
8382 | /* | |
428870ff BB |
8383 | * Temporarily record the splitting vdevs in the spa config. This |
8384 | * will disappear once the config is regenerated. | |
34dc7c2f | 8385 | */ |
65ad5d11 AJ |
8386 | nvl = fnvlist_alloc(); |
8387 | fnvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children); | |
428870ff | 8388 | kmem_free(glist, children * sizeof (uint64_t)); |
34dc7c2f | 8389 | |
428870ff | 8390 | mutex_enter(&spa->spa_props_lock); |
65ad5d11 | 8391 | fnvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl); |
428870ff BB |
8392 | mutex_exit(&spa->spa_props_lock); |
8393 | spa->spa_config_splitting = nvl; | |
8394 | vdev_config_dirty(spa->spa_root_vdev); | |
8395 | ||
8396 | /* configure and create the new pool */ | |
65ad5d11 AJ |
8397 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname); |
8398 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
8399 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE); | |
8400 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); | |
8401 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg); | |
8402 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
8403 | spa_generate_guid(NULL)); | |
e0ab3ab5 | 8404 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); |
428870ff BB |
8405 | (void) nvlist_lookup_string(props, |
8406 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
34dc7c2f | 8407 | |
428870ff BB |
8408 | /* add the new pool to the namespace */ |
8409 | newspa = spa_add(newname, config, altroot); | |
e0ab3ab5 | 8410 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
8411 | newspa->spa_config_txg = spa->spa_config_txg; |
8412 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
8413 | ||
8414 | /* release the spa config lock, retaining the namespace lock */ | |
8415 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
8416 | ||
8417 | if (zio_injection_enabled) | |
8418 | zio_handle_panic_injection(spa, FTAG, 1); | |
8419 | ||
8420 | spa_activate(newspa, spa_mode_global); | |
8421 | spa_async_suspend(newspa); | |
8422 | ||
c10d37dd | 8423 | /* |
1b939560 BB |
8424 | * Temporarily stop the initializing and TRIM activity. We set the |
8425 | * state to ACTIVE so that we know to resume initializing or TRIM | |
8426 | * once the split has completed. | |
c10d37dd | 8427 | */ |
1b939560 BB |
8428 | list_t vd_initialize_list; |
8429 | list_create(&vd_initialize_list, sizeof (vdev_t), | |
c10d37dd GW |
8430 | offsetof(vdev_t, vdev_initialize_node)); |
8431 | ||
1b939560 BB |
8432 | list_t vd_trim_list; |
8433 | list_create(&vd_trim_list, sizeof (vdev_t), | |
8434 | offsetof(vdev_t, vdev_trim_node)); | |
8435 | ||
619f0976 | 8436 | for (c = 0; c < children; c++) { |
1b664952 | 8437 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
619f0976 | 8438 | mutex_enter(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
8439 | vdev_initialize_stop(vml[c], |
8440 | VDEV_INITIALIZE_ACTIVE, &vd_initialize_list); | |
619f0976 | 8441 | mutex_exit(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
8442 | |
8443 | mutex_enter(&vml[c]->vdev_trim_lock); | |
8444 | vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list); | |
8445 | mutex_exit(&vml[c]->vdev_trim_lock); | |
619f0976 GW |
8446 | } |
8447 | } | |
1b939560 BB |
8448 | |
8449 | vdev_initialize_stop_wait(spa, &vd_initialize_list); | |
8450 | vdev_trim_stop_wait(spa, &vd_trim_list); | |
8451 | ||
8452 | list_destroy(&vd_initialize_list); | |
8453 | list_destroy(&vd_trim_list); | |
619f0976 | 8454 | |
6cb8e530 | 8455 | newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; |
8b27e08e | 8456 | newspa->spa_is_splitting = B_TRUE; |
6cb8e530 | 8457 | |
428870ff | 8458 | /* create the new pool from the disks of the original pool */ |
6cb8e530 | 8459 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); |
428870ff BB |
8460 | if (error) |
8461 | goto out; | |
8462 | ||
8463 | /* if that worked, generate a real config for the new pool */ | |
8464 | if (newspa->spa_root_vdev != NULL) { | |
65ad5d11 AJ |
8465 | newspa->spa_config_splitting = fnvlist_alloc(); |
8466 | fnvlist_add_uint64(newspa->spa_config_splitting, | |
8467 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)); | |
428870ff BB |
8468 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, |
8469 | B_TRUE)); | |
9babb374 | 8470 | } |
34dc7c2f | 8471 | |
428870ff BB |
8472 | /* set the props */ |
8473 | if (props != NULL) { | |
8474 | spa_configfile_set(newspa, props, B_FALSE); | |
8475 | error = spa_prop_set(newspa, props); | |
8476 | if (error) | |
8477 | goto out; | |
8478 | } | |
34dc7c2f | 8479 | |
428870ff BB |
8480 | /* flush everything */ |
8481 | txg = spa_vdev_config_enter(newspa); | |
8482 | vdev_config_dirty(newspa->spa_root_vdev); | |
8483 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
34dc7c2f | 8484 | |
428870ff BB |
8485 | if (zio_injection_enabled) |
8486 | zio_handle_panic_injection(spa, FTAG, 2); | |
34dc7c2f | 8487 | |
428870ff | 8488 | spa_async_resume(newspa); |
34dc7c2f | 8489 | |
428870ff BB |
8490 | /* finally, update the original pool's config */ |
8491 | txg = spa_vdev_config_enter(spa); | |
8492 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
8493 | error = dmu_tx_assign(tx, TXG_WAIT); | |
8494 | if (error != 0) | |
8495 | dmu_tx_abort(tx); | |
8496 | for (c = 0; c < children; c++) { | |
1b664952 | 8497 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
234234ca RS |
8498 | vdev_t *tvd = vml[c]->vdev_top; |
8499 | ||
8500 | /* | |
8501 | * Need to be sure the detachable VDEV is not | |
8502 | * on any *other* txg's DTL list to prevent it | |
8503 | * from being accessed after it's freed. | |
8504 | */ | |
8505 | for (int t = 0; t < TXG_SIZE; t++) { | |
8506 | (void) txg_list_remove_this( | |
8507 | &tvd->vdev_dtl_list, vml[c], t); | |
8508 | } | |
8509 | ||
428870ff BB |
8510 | vdev_split(vml[c]); |
8511 | if (error == 0) | |
6f1ffb06 MA |
8512 | spa_history_log_internal(spa, "detach", tx, |
8513 | "vdev=%s", vml[c]->vdev_path); | |
e0ab3ab5 | 8514 | |
428870ff | 8515 | vdev_free(vml[c]); |
34dc7c2f | 8516 | } |
34dc7c2f | 8517 | } |
e0ab3ab5 | 8518 | spa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
8519 | vdev_config_dirty(spa->spa_root_vdev); |
8520 | spa->spa_config_splitting = NULL; | |
8521 | nvlist_free(nvl); | |
8522 | if (error == 0) | |
8523 | dmu_tx_commit(tx); | |
8524 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
8525 | ||
8526 | if (zio_injection_enabled) | |
8527 | zio_handle_panic_injection(spa, FTAG, 3); | |
8528 | ||
8529 | /* split is complete; log a history record */ | |
6f1ffb06 MA |
8530 | spa_history_log_internal(newspa, "split", NULL, |
8531 | "from pool %s", spa_name(spa)); | |
428870ff | 8532 | |
8b27e08e | 8533 | newspa->spa_is_splitting = B_FALSE; |
428870ff BB |
8534 | kmem_free(vml, children * sizeof (vdev_t *)); |
8535 | ||
8536 | /* if we're not going to mount the filesystems in userland, export */ | |
8537 | if (exp) | |
8538 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
8539 | B_FALSE, B_FALSE); | |
8540 | ||
8541 | return (error); | |
8542 | ||
8543 | out: | |
8544 | spa_unload(newspa); | |
8545 | spa_deactivate(newspa); | |
8546 | spa_remove(newspa); | |
8547 | ||
8548 | txg = spa_vdev_config_enter(spa); | |
8549 | ||
8550 | /* re-online all offlined disks */ | |
8551 | for (c = 0; c < children; c++) { | |
8552 | if (vml[c] != NULL) | |
8553 | vml[c]->vdev_offline = B_FALSE; | |
8554 | } | |
619f0976 | 8555 | |
1b939560 | 8556 | /* restart initializing or trimming disks as necessary */ |
619f0976 | 8557 | spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART); |
1b939560 BB |
8558 | spa_async_request(spa, SPA_ASYNC_TRIM_RESTART); |
8559 | spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART); | |
619f0976 | 8560 | |
428870ff BB |
8561 | vdev_reopen(spa->spa_root_vdev); |
8562 | ||
8563 | nvlist_free(spa->spa_config_splitting); | |
8564 | spa->spa_config_splitting = NULL; | |
8565 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
34dc7c2f | 8566 | |
428870ff | 8567 | kmem_free(vml, children * sizeof (vdev_t *)); |
34dc7c2f BB |
8568 | return (error); |
8569 | } | |
8570 | ||
34dc7c2f BB |
8571 | /* |
8572 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
d3cc8b15 | 8573 | * currently spared, so we can detach it. |
34dc7c2f BB |
8574 | */ |
8575 | static vdev_t * | |
8576 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
8577 | { | |
8578 | vdev_t *newvd, *oldvd; | |
34dc7c2f | 8579 | |
1c27024e | 8580 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
8581 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); |
8582 | if (oldvd != NULL) | |
8583 | return (oldvd); | |
8584 | } | |
8585 | ||
8586 | /* | |
572e2857 BB |
8587 | * Check for a completed replacement. We always consider the first |
8588 | * vdev in the list to be the oldest vdev, and the last one to be | |
8589 | * the newest (see spa_vdev_attach() for how that works). In | |
8590 | * the case where the newest vdev is faulted, we will not automatically | |
8591 | * remove it after a resilver completes. This is OK as it will require | |
8592 | * user intervention to determine which disk the admin wishes to keep. | |
34dc7c2f | 8593 | */ |
572e2857 BB |
8594 | if (vd->vdev_ops == &vdev_replacing_ops) { |
8595 | ASSERT(vd->vdev_children > 1); | |
8596 | ||
8597 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
34dc7c2f | 8598 | oldvd = vd->vdev_child[0]; |
34dc7c2f | 8599 | |
fb5f0bc8 | 8600 | if (vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 8601 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
fb5f0bc8 | 8602 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 8603 | return (oldvd); |
34dc7c2f BB |
8604 | } |
8605 | ||
8606 | /* | |
8607 | * Check for a completed resilver with the 'unspare' flag set. | |
f65fbee1 | 8608 | * Also potentially update faulted state. |
34dc7c2f | 8609 | */ |
572e2857 BB |
8610 | if (vd->vdev_ops == &vdev_spare_ops) { |
8611 | vdev_t *first = vd->vdev_child[0]; | |
8612 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
8613 | ||
8614 | if (last->vdev_unspare) { | |
8615 | oldvd = first; | |
8616 | newvd = last; | |
8617 | } else if (first->vdev_unspare) { | |
8618 | oldvd = last; | |
8619 | newvd = first; | |
8620 | } else { | |
8621 | oldvd = NULL; | |
8622 | } | |
34dc7c2f | 8623 | |
572e2857 | 8624 | if (oldvd != NULL && |
fb5f0bc8 | 8625 | vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 8626 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
572e2857 | 8627 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 8628 | return (oldvd); |
572e2857 | 8629 | |
f65fbee1 JJ |
8630 | vdev_propagate_state(vd); |
8631 | ||
572e2857 BB |
8632 | /* |
8633 | * If there are more than two spares attached to a disk, | |
8634 | * and those spares are not required, then we want to | |
8635 | * attempt to free them up now so that they can be used | |
8636 | * by other pools. Once we're back down to a single | |
8637 | * disk+spare, we stop removing them. | |
8638 | */ | |
8639 | if (vd->vdev_children > 2) { | |
8640 | newvd = vd->vdev_child[1]; | |
8641 | ||
8642 | if (newvd->vdev_isspare && last->vdev_isspare && | |
8643 | vdev_dtl_empty(last, DTL_MISSING) && | |
8644 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
8645 | !vdev_dtl_required(newvd)) | |
8646 | return (newvd); | |
34dc7c2f | 8647 | } |
34dc7c2f BB |
8648 | } |
8649 | ||
8650 | return (NULL); | |
8651 | } | |
8652 | ||
8653 | static void | |
8654 | spa_vdev_resilver_done(spa_t *spa) | |
8655 | { | |
fb5f0bc8 BB |
8656 | vdev_t *vd, *pvd, *ppvd; |
8657 | uint64_t guid, sguid, pguid, ppguid; | |
34dc7c2f | 8658 | |
fb5f0bc8 | 8659 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8660 | |
8661 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
fb5f0bc8 BB |
8662 | pvd = vd->vdev_parent; |
8663 | ppvd = pvd->vdev_parent; | |
34dc7c2f | 8664 | guid = vd->vdev_guid; |
fb5f0bc8 BB |
8665 | pguid = pvd->vdev_guid; |
8666 | ppguid = ppvd->vdev_guid; | |
8667 | sguid = 0; | |
34dc7c2f BB |
8668 | /* |
8669 | * If we have just finished replacing a hot spared device, then | |
8670 | * we need to detach the parent's first child (the original hot | |
8671 | * spare) as well. | |
8672 | */ | |
572e2857 BB |
8673 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && |
8674 | ppvd->vdev_children == 2) { | |
34dc7c2f | 8675 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); |
fb5f0bc8 | 8676 | sguid = ppvd->vdev_child[1]->vdev_guid; |
34dc7c2f | 8677 | } |
5d1f7fb6 GW |
8678 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); |
8679 | ||
fb5f0bc8 BB |
8680 | spa_config_exit(spa, SCL_ALL, FTAG); |
8681 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
34dc7c2f | 8682 | return; |
fb5f0bc8 | 8683 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) |
34dc7c2f | 8684 | return; |
fb5f0bc8 | 8685 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8686 | } |
8687 | ||
fb5f0bc8 | 8688 | spa_config_exit(spa, SCL_ALL, FTAG); |
9a49d3f3 BB |
8689 | |
8690 | /* | |
8691 | * If a detach was not performed above replace waiters will not have | |
8692 | * been notified. In which case we must do so now. | |
8693 | */ | |
8694 | spa_notify_waiters(spa); | |
34dc7c2f BB |
8695 | } |
8696 | ||
8697 | /* | |
428870ff | 8698 | * Update the stored path or FRU for this vdev. |
34dc7c2f | 8699 | */ |
65c7cc49 | 8700 | static int |
9babb374 BB |
8701 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, |
8702 | boolean_t ispath) | |
34dc7c2f | 8703 | { |
b128c09f | 8704 | vdev_t *vd; |
428870ff | 8705 | boolean_t sync = B_FALSE; |
34dc7c2f | 8706 | |
572e2857 BB |
8707 | ASSERT(spa_writeable(spa)); |
8708 | ||
428870ff | 8709 | spa_vdev_state_enter(spa, SCL_ALL); |
34dc7c2f | 8710 | |
9babb374 | 8711 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
428870ff | 8712 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); |
34dc7c2f BB |
8713 | |
8714 | if (!vd->vdev_ops->vdev_op_leaf) | |
428870ff | 8715 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 8716 | |
9babb374 | 8717 | if (ispath) { |
428870ff BB |
8718 | if (strcmp(value, vd->vdev_path) != 0) { |
8719 | spa_strfree(vd->vdev_path); | |
8720 | vd->vdev_path = spa_strdup(value); | |
8721 | sync = B_TRUE; | |
8722 | } | |
9babb374 | 8723 | } else { |
428870ff BB |
8724 | if (vd->vdev_fru == NULL) { |
8725 | vd->vdev_fru = spa_strdup(value); | |
8726 | sync = B_TRUE; | |
8727 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
9babb374 | 8728 | spa_strfree(vd->vdev_fru); |
428870ff BB |
8729 | vd->vdev_fru = spa_strdup(value); |
8730 | sync = B_TRUE; | |
8731 | } | |
9babb374 | 8732 | } |
34dc7c2f | 8733 | |
428870ff | 8734 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); |
34dc7c2f BB |
8735 | } |
8736 | ||
9babb374 BB |
8737 | int |
8738 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
8739 | { | |
8740 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
8741 | } | |
8742 | ||
8743 | int | |
8744 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
8745 | { | |
8746 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
8747 | } | |
8748 | ||
34dc7c2f BB |
8749 | /* |
8750 | * ========================================================================== | |
428870ff | 8751 | * SPA Scanning |
34dc7c2f BB |
8752 | * ========================================================================== |
8753 | */ | |
0ea05c64 AP |
8754 | int |
8755 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
8756 | { | |
8757 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8758 | ||
8759 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
8760 | return (SET_ERROR(EBUSY)); | |
8761 | ||
8762 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
8763 | } | |
34dc7c2f | 8764 | |
34dc7c2f | 8765 | int |
428870ff BB |
8766 | spa_scan_stop(spa_t *spa) |
8767 | { | |
8768 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8769 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
2e528b49 | 8770 | return (SET_ERROR(EBUSY)); |
482eeef8 | 8771 | |
428870ff BB |
8772 | return (dsl_scan_cancel(spa->spa_dsl_pool)); |
8773 | } | |
8774 | ||
8775 | int | |
8776 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
34dc7c2f | 8777 | { |
b128c09f | 8778 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); |
34dc7c2f | 8779 | |
428870ff | 8780 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) |
2e528b49 | 8781 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 8782 | |
fa241660 TC |
8783 | if (func == POOL_SCAN_RESILVER && |
8784 | !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) | |
8785 | return (SET_ERROR(ENOTSUP)); | |
8786 | ||
34dc7c2f | 8787 | /* |
b128c09f BB |
8788 | * If a resilver was requested, but there is no DTL on a |
8789 | * writeable leaf device, we have nothing to do. | |
34dc7c2f | 8790 | */ |
428870ff | 8791 | if (func == POOL_SCAN_RESILVER && |
b128c09f BB |
8792 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { |
8793 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
34dc7c2f BB |
8794 | return (0); |
8795 | } | |
8796 | ||
482eeef8 GA |
8797 | if (func == POOL_SCAN_ERRORSCRUB && |
8798 | !spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) | |
8799 | return (SET_ERROR(ENOTSUP)); | |
8800 | ||
428870ff | 8801 | return (dsl_scan(spa->spa_dsl_pool, func)); |
34dc7c2f BB |
8802 | } |
8803 | ||
8804 | /* | |
8805 | * ========================================================================== | |
8806 | * SPA async task processing | |
8807 | * ========================================================================== | |
8808 | */ | |
8809 | ||
8810 | static void | |
8811 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
8812 | { | |
b128c09f | 8813 | if (vd->vdev_remove_wanted) { |
428870ff BB |
8814 | vd->vdev_remove_wanted = B_FALSE; |
8815 | vd->vdev_delayed_close = B_FALSE; | |
b128c09f | 8816 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); |
428870ff BB |
8817 | |
8818 | /* | |
8819 | * We want to clear the stats, but we don't want to do a full | |
8820 | * vdev_clear() as that will cause us to throw away | |
8821 | * degraded/faulted state as well as attempt to reopen the | |
8822 | * device, all of which is a waste. | |
8823 | */ | |
8824 | vd->vdev_stat.vs_read_errors = 0; | |
8825 | vd->vdev_stat.vs_write_errors = 0; | |
8826 | vd->vdev_stat.vs_checksum_errors = 0; | |
8827 | ||
b128c09f | 8828 | vdev_state_dirty(vd->vdev_top); |
0aacde2e RM |
8829 | |
8830 | /* Tell userspace that the vdev is gone. */ | |
8831 | zfs_post_remove(spa, vd); | |
b128c09f | 8832 | } |
34dc7c2f | 8833 | |
1c27024e | 8834 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f BB |
8835 | spa_async_remove(spa, vd->vdev_child[c]); |
8836 | } | |
8837 | ||
8838 | static void | |
c3f2f1aa | 8839 | spa_async_fault_vdev(spa_t *spa, vdev_t *vd) |
b128c09f | 8840 | { |
c3f2f1aa DB |
8841 | if (vd->vdev_fault_wanted) { |
8842 | vd->vdev_fault_wanted = B_FALSE; | |
8843 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
8844 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f | 8845 | } |
b128c09f | 8846 | |
1c27024e | 8847 | for (int c = 0; c < vd->vdev_children; c++) |
c3f2f1aa | 8848 | spa_async_fault_vdev(spa, vd->vdev_child[c]); |
34dc7c2f BB |
8849 | } |
8850 | ||
9babb374 BB |
8851 | static void |
8852 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
8853 | { | |
9babb374 BB |
8854 | if (!spa->spa_autoexpand) |
8855 | return; | |
8856 | ||
1c27024e | 8857 | for (int c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
8858 | vdev_t *cvd = vd->vdev_child[c]; |
8859 | spa_async_autoexpand(spa, cvd); | |
8860 | } | |
8861 | ||
8862 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
8863 | return; | |
8864 | ||
12fa0466 | 8865 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); |
9babb374 BB |
8866 | } |
8867 | ||
460748d4 | 8868 | static __attribute__((noreturn)) void |
c25b8f99 | 8869 | spa_async_thread(void *arg) |
34dc7c2f | 8870 | { |
c25b8f99 | 8871 | spa_t *spa = (spa_t *)arg; |
80a91e74 | 8872 | dsl_pool_t *dp = spa->spa_dsl_pool; |
867959b5 | 8873 | int tasks; |
34dc7c2f BB |
8874 | |
8875 | ASSERT(spa->spa_sync_on); | |
8876 | ||
8877 | mutex_enter(&spa->spa_async_lock); | |
8878 | tasks = spa->spa_async_tasks; | |
8879 | spa->spa_async_tasks = 0; | |
8880 | mutex_exit(&spa->spa_async_lock); | |
8881 | ||
8882 | /* | |
8883 | * See if the config needs to be updated. | |
8884 | */ | |
8885 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
428870ff | 8886 | uint64_t old_space, new_space; |
9babb374 | 8887 | |
34dc7c2f | 8888 | mutex_enter(&spa_namespace_lock); |
428870ff | 8889 | old_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8890 | old_space += metaslab_class_get_space(spa_special_class(spa)); |
8891 | old_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8892 | old_space += metaslab_class_get_space( |
8893 | spa_embedded_log_class(spa)); | |
cc99f275 | 8894 | |
34dc7c2f | 8895 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
cc99f275 | 8896 | |
428870ff | 8897 | new_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8898 | new_space += metaslab_class_get_space(spa_special_class(spa)); |
8899 | new_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8900 | new_space += metaslab_class_get_space( |
8901 | spa_embedded_log_class(spa)); | |
34dc7c2f | 8902 | mutex_exit(&spa_namespace_lock); |
9babb374 BB |
8903 | |
8904 | /* | |
8905 | * If the pool grew as a result of the config update, | |
8906 | * then log an internal history event. | |
8907 | */ | |
428870ff | 8908 | if (new_space != old_space) { |
6f1ffb06 | 8909 | spa_history_log_internal(spa, "vdev online", NULL, |
45d1cae3 | 8910 | "pool '%s' size: %llu(+%llu)", |
74756182 MM |
8911 | spa_name(spa), (u_longlong_t)new_space, |
8912 | (u_longlong_t)(new_space - old_space)); | |
9babb374 | 8913 | } |
34dc7c2f BB |
8914 | } |
8915 | ||
8916 | /* | |
8917 | * See if any devices need to be marked REMOVED. | |
34dc7c2f | 8918 | */ |
b128c09f | 8919 | if (tasks & SPA_ASYNC_REMOVE) { |
428870ff | 8920 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 8921 | spa_async_remove(spa, spa->spa_root_vdev); |
867959b5 | 8922 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) |
b128c09f | 8923 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); |
867959b5 | 8924 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
b128c09f BB |
8925 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); |
8926 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
34dc7c2f BB |
8927 | } |
8928 | ||
9babb374 BB |
8929 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { |
8930 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8931 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
8932 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8933 | } | |
8934 | ||
34dc7c2f | 8935 | /* |
c3f2f1aa | 8936 | * See if any devices need to be marked faulted. |
34dc7c2f | 8937 | */ |
c3f2f1aa | 8938 | if (tasks & SPA_ASYNC_FAULT_VDEV) { |
428870ff | 8939 | spa_vdev_state_enter(spa, SCL_NONE); |
c3f2f1aa | 8940 | spa_async_fault_vdev(spa, spa->spa_root_vdev); |
b128c09f BB |
8941 | (void) spa_vdev_state_exit(spa, NULL, 0); |
8942 | } | |
34dc7c2f BB |
8943 | |
8944 | /* | |
b128c09f | 8945 | * If any devices are done replacing, detach them. |
34dc7c2f | 8946 | */ |
b2255edc | 8947 | if (tasks & SPA_ASYNC_RESILVER_DONE || |
719534ca AH |
8948 | tasks & SPA_ASYNC_REBUILD_DONE || |
8949 | tasks & SPA_ASYNC_DETACH_SPARE) { | |
b128c09f | 8950 | spa_vdev_resilver_done(spa); |
9a49d3f3 BB |
8951 | } |
8952 | ||
34dc7c2f BB |
8953 | /* |
8954 | * Kick off a resilver. | |
8955 | */ | |
80a91e74 | 8956 | if (tasks & SPA_ASYNC_RESILVER && |
9a49d3f3 | 8957 | !vdev_rebuild_active(spa->spa_root_vdev) && |
80a91e74 TC |
8958 | (!dsl_scan_resilvering(dp) || |
8959 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER))) | |
3c819a2c | 8960 | dsl_scan_restart_resilver(dp, 0); |
34dc7c2f | 8961 | |
619f0976 GW |
8962 | if (tasks & SPA_ASYNC_INITIALIZE_RESTART) { |
8963 | mutex_enter(&spa_namespace_lock); | |
8964 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8965 | vdev_initialize_restart(spa->spa_root_vdev); | |
8966 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8967 | mutex_exit(&spa_namespace_lock); | |
8968 | } | |
8969 | ||
1b939560 BB |
8970 | if (tasks & SPA_ASYNC_TRIM_RESTART) { |
8971 | mutex_enter(&spa_namespace_lock); | |
8972 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8973 | vdev_trim_restart(spa->spa_root_vdev); | |
8974 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8975 | mutex_exit(&spa_namespace_lock); | |
8976 | } | |
8977 | ||
8978 | if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) { | |
8979 | mutex_enter(&spa_namespace_lock); | |
8980 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8981 | vdev_autotrim_restart(spa); | |
8982 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8983 | mutex_exit(&spa_namespace_lock); | |
8984 | } | |
8985 | ||
b7654bd7 GA |
8986 | /* |
8987 | * Kick off L2 cache whole device TRIM. | |
8988 | */ | |
8989 | if (tasks & SPA_ASYNC_L2CACHE_TRIM) { | |
8990 | mutex_enter(&spa_namespace_lock); | |
8991 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8992 | vdev_trim_l2arc(spa); | |
8993 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8994 | mutex_exit(&spa_namespace_lock); | |
8995 | } | |
8996 | ||
77f6826b GA |
8997 | /* |
8998 | * Kick off L2 cache rebuilding. | |
8999 | */ | |
9000 | if (tasks & SPA_ASYNC_L2CACHE_REBUILD) { | |
9001 | mutex_enter(&spa_namespace_lock); | |
9002 | spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER); | |
9003 | l2arc_spa_rebuild_start(spa); | |
9004 | spa_config_exit(spa, SCL_L2ARC, FTAG); | |
9005 | mutex_exit(&spa_namespace_lock); | |
9006 | } | |
9007 | ||
34dc7c2f BB |
9008 | /* |
9009 | * Let the world know that we're done. | |
9010 | */ | |
9011 | mutex_enter(&spa->spa_async_lock); | |
9012 | spa->spa_async_thread = NULL; | |
9013 | cv_broadcast(&spa->spa_async_cv); | |
9014 | mutex_exit(&spa->spa_async_lock); | |
9015 | thread_exit(); | |
9016 | } | |
9017 | ||
9018 | void | |
9019 | spa_async_suspend(spa_t *spa) | |
9020 | { | |
9021 | mutex_enter(&spa->spa_async_lock); | |
9022 | spa->spa_async_suspended++; | |
9d5b5245 | 9023 | while (spa->spa_async_thread != NULL) |
34dc7c2f BB |
9024 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); |
9025 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 MA |
9026 | |
9027 | spa_vdev_remove_suspend(spa); | |
9d5b5245 SD |
9028 | |
9029 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a SD |
9030 | if (condense_thread != NULL) |
9031 | zthr_cancel(condense_thread); | |
d2734cce | 9032 | |
5caeef02 DB |
9033 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; |
9034 | if (raidz_expand_thread != NULL) | |
9035 | zthr_cancel(raidz_expand_thread); | |
9036 | ||
d2734cce | 9037 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; |
61c3391a SD |
9038 | if (discard_thread != NULL) |
9039 | zthr_cancel(discard_thread); | |
37f03da8 SH |
9040 | |
9041 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
9042 | if (ll_delete_thread != NULL) | |
9043 | zthr_cancel(ll_delete_thread); | |
9044 | ||
9045 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
9046 | if (ll_condense_thread != NULL) | |
9047 | zthr_cancel(ll_condense_thread); | |
34dc7c2f BB |
9048 | } |
9049 | ||
9050 | void | |
9051 | spa_async_resume(spa_t *spa) | |
9052 | { | |
9053 | mutex_enter(&spa->spa_async_lock); | |
9054 | ASSERT(spa->spa_async_suspended != 0); | |
9055 | spa->spa_async_suspended--; | |
9056 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 | 9057 | spa_restart_removal(spa); |
9d5b5245 SD |
9058 | |
9059 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a | 9060 | if (condense_thread != NULL) |
9d5b5245 | 9061 | zthr_resume(condense_thread); |
d2734cce | 9062 | |
5caeef02 DB |
9063 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; |
9064 | if (raidz_expand_thread != NULL) | |
9065 | zthr_resume(raidz_expand_thread); | |
9066 | ||
d2734cce | 9067 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; |
61c3391a | 9068 | if (discard_thread != NULL) |
d2734cce | 9069 | zthr_resume(discard_thread); |
37f03da8 SH |
9070 | |
9071 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
9072 | if (ll_delete_thread != NULL) | |
9073 | zthr_resume(ll_delete_thread); | |
9074 | ||
9075 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
9076 | if (ll_condense_thread != NULL) | |
9077 | zthr_resume(ll_condense_thread); | |
34dc7c2f BB |
9078 | } |
9079 | ||
e6cfd633 WA |
9080 | static boolean_t |
9081 | spa_async_tasks_pending(spa_t *spa) | |
9082 | { | |
9083 | uint_t non_config_tasks; | |
9084 | uint_t config_task; | |
9085 | boolean_t config_task_suspended; | |
9086 | ||
9087 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
9088 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
9089 | if (spa->spa_ccw_fail_time == 0) { | |
9090 | config_task_suspended = B_FALSE; | |
9091 | } else { | |
9092 | config_task_suspended = | |
9093 | (gethrtime() - spa->spa_ccw_fail_time) < | |
05852b34 | 9094 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); |
e6cfd633 WA |
9095 | } |
9096 | ||
9097 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
9098 | } | |
9099 | ||
34dc7c2f BB |
9100 | static void |
9101 | spa_async_dispatch(spa_t *spa) | |
9102 | { | |
9103 | mutex_enter(&spa->spa_async_lock); | |
e6cfd633 WA |
9104 | if (spa_async_tasks_pending(spa) && |
9105 | !spa->spa_async_suspended && | |
da92d5cb | 9106 | spa->spa_async_thread == NULL) |
34dc7c2f BB |
9107 | spa->spa_async_thread = thread_create(NULL, 0, |
9108 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
9109 | mutex_exit(&spa->spa_async_lock); | |
9110 | } | |
9111 | ||
9112 | void | |
9113 | spa_async_request(spa_t *spa, int task) | |
9114 | { | |
428870ff | 9115 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); |
34dc7c2f BB |
9116 | mutex_enter(&spa->spa_async_lock); |
9117 | spa->spa_async_tasks |= task; | |
9118 | mutex_exit(&spa->spa_async_lock); | |
9119 | } | |
9120 | ||
3c819a2c JP |
9121 | int |
9122 | spa_async_tasks(spa_t *spa) | |
9123 | { | |
9124 | return (spa->spa_async_tasks); | |
9125 | } | |
9126 | ||
34dc7c2f BB |
9127 | /* |
9128 | * ========================================================================== | |
9129 | * SPA syncing routines | |
9130 | * ========================================================================== | |
9131 | */ | |
9132 | ||
37f03da8 | 9133 | |
428870ff | 9134 | static int |
37f03da8 SH |
9135 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, |
9136 | dmu_tx_t *tx) | |
34dc7c2f | 9137 | { |
428870ff | 9138 | bpobj_t *bpo = arg; |
37f03da8 | 9139 | bpobj_enqueue(bpo, bp, bp_freed, tx); |
428870ff BB |
9140 | return (0); |
9141 | } | |
34dc7c2f | 9142 | |
37f03da8 SH |
9143 | int |
9144 | bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9145 | { | |
9146 | return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx)); | |
9147 | } | |
9148 | ||
9149 | int | |
9150 | bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9151 | { | |
9152 | return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx)); | |
9153 | } | |
9154 | ||
428870ff BB |
9155 | static int |
9156 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9157 | { | |
9cdf7b1f | 9158 | zio_t *pio = arg; |
34dc7c2f | 9159 | |
9cdf7b1f MA |
9160 | zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp, |
9161 | pio->io_flags)); | |
428870ff | 9162 | return (0); |
34dc7c2f BB |
9163 | } |
9164 | ||
37f03da8 SH |
9165 | static int |
9166 | bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
9167 | dmu_tx_t *tx) | |
9168 | { | |
9169 | ASSERT(!bp_freed); | |
9170 | return (spa_free_sync_cb(arg, bp, tx)); | |
9171 | } | |
9172 | ||
e8b96c60 MA |
9173 | /* |
9174 | * Note: this simple function is not inlined to make it easier to dtrace the | |
9175 | * amount of time spent syncing frees. | |
9176 | */ | |
9177 | static void | |
9178 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
9179 | { | |
9180 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
9181 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
9182 | VERIFY(zio_wait(zio) == 0); | |
9183 | } | |
9184 | ||
9185 | /* | |
9186 | * Note: this simple function is not inlined to make it easier to dtrace the | |
9187 | * amount of time spent syncing deferred frees. | |
9188 | */ | |
9189 | static void | |
9190 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
9191 | { | |
8dc2197b SD |
9192 | if (spa_sync_pass(spa) != 1) |
9193 | return; | |
9194 | ||
93e28d66 SD |
9195 | /* |
9196 | * Note: | |
9197 | * If the log space map feature is active, we stop deferring | |
9198 | * frees to the next TXG and therefore running this function | |
9199 | * would be considered a no-op as spa_deferred_bpobj should | |
9200 | * not have any entries. | |
9201 | * | |
9202 | * That said we run this function anyway (instead of returning | |
9203 | * immediately) for the edge-case scenario where we just | |
9204 | * activated the log space map feature in this TXG but we have | |
9205 | * deferred frees from the previous TXG. | |
9206 | */ | |
e8b96c60 MA |
9207 | zio_t *zio = zio_root(spa, NULL, NULL, 0); |
9208 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
37f03da8 | 9209 | bpobj_spa_free_sync_cb, zio, tx), ==, 0); |
e8b96c60 MA |
9210 | VERIFY0(zio_wait(zio)); |
9211 | } | |
9212 | ||
34dc7c2f BB |
9213 | static void |
9214 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
9215 | { | |
9216 | char *packed = NULL; | |
b128c09f | 9217 | size_t bufsize; |
34dc7c2f BB |
9218 | size_t nvsize = 0; |
9219 | dmu_buf_t *db; | |
9220 | ||
9221 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
9222 | ||
b128c09f BB |
9223 | /* |
9224 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
b0bc7a84 | 9225 | * information. This avoids the dmu_buf_will_dirty() path and |
b128c09f BB |
9226 | * saves us a pre-read to get data we don't actually care about. |
9227 | */ | |
9ae529ec | 9228 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); |
79c76d5b | 9229 | packed = vmem_alloc(bufsize, KM_SLEEP); |
34dc7c2f BB |
9230 | |
9231 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
79c76d5b | 9232 | KM_SLEEP) == 0); |
861166b0 | 9233 | memset(packed + nvsize, 0, bufsize - nvsize); |
34dc7c2f | 9234 | |
b128c09f | 9235 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); |
34dc7c2f | 9236 | |
00b46022 | 9237 | vmem_free(packed, bufsize); |
34dc7c2f BB |
9238 | |
9239 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
9240 | dmu_buf_will_dirty(db, tx); | |
9241 | *(uint64_t *)db->db_data = nvsize; | |
9242 | dmu_buf_rele(db, FTAG); | |
9243 | } | |
9244 | ||
9245 | static void | |
9246 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
9247 | const char *config, const char *entry) | |
9248 | { | |
9249 | nvlist_t *nvroot; | |
9250 | nvlist_t **list; | |
9251 | int i; | |
9252 | ||
9253 | if (!sav->sav_sync) | |
9254 | return; | |
9255 | ||
9256 | /* | |
9257 | * Update the MOS nvlist describing the list of available devices. | |
9258 | * spa_validate_aux() will have already made sure this nvlist is | |
9259 | * valid and the vdevs are labeled appropriately. | |
9260 | */ | |
9261 | if (sav->sav_object == 0) { | |
9262 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
9263 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
9264 | sizeof (uint64_t), tx); | |
9265 | VERIFY(zap_update(spa->spa_meta_objset, | |
9266 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
9267 | &sav->sav_object, tx) == 0); | |
9268 | } | |
9269 | ||
65ad5d11 | 9270 | nvroot = fnvlist_alloc(); |
34dc7c2f | 9271 | if (sav->sav_count == 0) { |
795075e6 PD |
9272 | fnvlist_add_nvlist_array(nvroot, config, |
9273 | (const nvlist_t * const *)NULL, 0); | |
34dc7c2f | 9274 | } else { |
79c76d5b | 9275 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); |
34dc7c2f BB |
9276 | for (i = 0; i < sav->sav_count; i++) |
9277 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
428870ff | 9278 | B_FALSE, VDEV_CONFIG_L2CACHE); |
795075e6 PD |
9279 | fnvlist_add_nvlist_array(nvroot, config, |
9280 | (const nvlist_t * const *)list, sav->sav_count); | |
34dc7c2f BB |
9281 | for (i = 0; i < sav->sav_count; i++) |
9282 | nvlist_free(list[i]); | |
9283 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
9284 | } | |
9285 | ||
9286 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
9287 | nvlist_free(nvroot); | |
9288 | ||
9289 | sav->sav_sync = B_FALSE; | |
9290 | } | |
9291 | ||
e0ab3ab5 JS |
9292 | /* |
9293 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
9294 | * The all-vdev ZAP must be empty. | |
9295 | */ | |
9296 | static void | |
9297 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
9298 | { | |
9299 | spa_t *spa = vd->vdev_spa; | |
e0ab3ab5 | 9300 | |
3e4ed421 RW |
9301 | if (vd->vdev_root_zap != 0 && |
9302 | spa_feature_is_active(spa, SPA_FEATURE_AVZ_V2)) { | |
9303 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9304 | vd->vdev_root_zap, tx)); | |
9305 | } | |
e0ab3ab5 JS |
9306 | if (vd->vdev_top_zap != 0) { |
9307 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9308 | vd->vdev_top_zap, tx)); | |
9309 | } | |
9310 | if (vd->vdev_leaf_zap != 0) { | |
9311 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9312 | vd->vdev_leaf_zap, tx)); | |
9313 | } | |
1c27024e | 9314 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
9315 | spa_avz_build(vd->vdev_child[i], avz, tx); |
9316 | } | |
9317 | } | |
9318 | ||
34dc7c2f BB |
9319 | static void |
9320 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
9321 | { | |
9322 | nvlist_t *config; | |
9323 | ||
e0ab3ab5 JS |
9324 | /* |
9325 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
9326 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
9327 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
9328 | * need to rebuild the AVZ although the config may not be dirty. | |
9329 | */ | |
9330 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
9331 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
34dc7c2f BB |
9332 | return; |
9333 | ||
b128c09f BB |
9334 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
9335 | ||
e0ab3ab5 | 9336 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || |
38640550 | 9337 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || |
e0ab3ab5 JS |
9338 | spa->spa_all_vdev_zaps != 0); |
9339 | ||
9340 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
e0ab3ab5 JS |
9341 | /* Make and build the new AVZ */ |
9342 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
9343 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
9344 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
9345 | ||
9346 | /* Diff old AVZ with new one */ | |
1c27024e DB |
9347 | zap_cursor_t zc; |
9348 | zap_attribute_t za; | |
9349 | ||
e0ab3ab5 JS |
9350 | for (zap_cursor_init(&zc, spa->spa_meta_objset, |
9351 | spa->spa_all_vdev_zaps); | |
9352 | zap_cursor_retrieve(&zc, &za) == 0; | |
9353 | zap_cursor_advance(&zc)) { | |
9354 | uint64_t vdzap = za.za_first_integer; | |
9355 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
9356 | vdzap) == ENOENT) { | |
9357 | /* | |
9358 | * ZAP is listed in old AVZ but not in new one; | |
9359 | * destroy it | |
9360 | */ | |
9361 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
9362 | tx)); | |
9363 | } | |
9364 | } | |
9365 | ||
9366 | zap_cursor_fini(&zc); | |
9367 | ||
9368 | /* Destroy the old AVZ */ | |
9369 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9370 | spa->spa_all_vdev_zaps, tx)); | |
9371 | ||
9372 | /* Replace the old AVZ in the dir obj with the new one */ | |
9373 | VERIFY0(zap_update(spa->spa_meta_objset, | |
9374 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
9375 | sizeof (new_avz), 1, &new_avz, tx)); | |
9376 | ||
9377 | spa->spa_all_vdev_zaps = new_avz; | |
9378 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
9379 | zap_cursor_t zc; | |
9380 | zap_attribute_t za; | |
9381 | ||
9382 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
9383 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
9384 | spa->spa_all_vdev_zaps); | |
9385 | zap_cursor_retrieve(&zc, &za) == 0; | |
9386 | zap_cursor_advance(&zc)) { | |
9387 | uint64_t zap = za.za_first_integer; | |
9388 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
9389 | } | |
9390 | ||
9391 | zap_cursor_fini(&zc); | |
9392 | ||
9393 | /* Destroy and unlink the AVZ itself */ | |
9394 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9395 | spa->spa_all_vdev_zaps, tx)); | |
9396 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
9397 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
9398 | spa->spa_all_vdev_zaps = 0; | |
9399 | } | |
9400 | ||
9401 | if (spa->spa_all_vdev_zaps == 0) { | |
9402 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
9403 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
9404 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
9405 | } | |
9406 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
9407 | ||
9408 | /* Create ZAPs for vdevs that don't have them. */ | |
9409 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
9410 | ||
b128c09f BB |
9411 | config = spa_config_generate(spa, spa->spa_root_vdev, |
9412 | dmu_tx_get_txg(tx), B_FALSE); | |
9413 | ||
ea0b2538 GW |
9414 | /* |
9415 | * If we're upgrading the spa version then make sure that | |
9416 | * the config object gets updated with the correct version. | |
9417 | */ | |
9418 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
9419 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
9420 | spa->spa_uberblock.ub_version); | |
9421 | ||
b128c09f | 9422 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f | 9423 | |
8a5fc748 | 9424 | nvlist_free(spa->spa_config_syncing); |
34dc7c2f BB |
9425 | spa->spa_config_syncing = config; |
9426 | ||
9427 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
9428 | } | |
9429 | ||
9ae529ec | 9430 | static void |
13fe0198 | 9431 | spa_sync_version(void *arg, dmu_tx_t *tx) |
9ae529ec | 9432 | { |
13fe0198 MA |
9433 | uint64_t *versionp = arg; |
9434 | uint64_t version = *versionp; | |
9435 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9ae529ec CS |
9436 | |
9437 | /* | |
9438 | * Setting the version is special cased when first creating the pool. | |
9439 | */ | |
9440 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
9441 | ||
8dca0a9a | 9442 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); |
9ae529ec CS |
9443 | ASSERT(version >= spa_version(spa)); |
9444 | ||
9445 | spa->spa_uberblock.ub_version = version; | |
9446 | vdev_config_dirty(spa->spa_root_vdev); | |
74756182 MM |
9447 | spa_history_log_internal(spa, "set", tx, "version=%lld", |
9448 | (longlong_t)version); | |
9ae529ec CS |
9449 | } |
9450 | ||
34dc7c2f BB |
9451 | /* |
9452 | * Set zpool properties. | |
9453 | */ | |
9454 | static void | |
13fe0198 | 9455 | spa_sync_props(void *arg, dmu_tx_t *tx) |
34dc7c2f | 9456 | { |
13fe0198 MA |
9457 | nvlist_t *nvp = arg; |
9458 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
34dc7c2f | 9459 | objset_t *mos = spa->spa_meta_objset; |
9ae529ec | 9460 | nvpair_t *elem = NULL; |
b128c09f BB |
9461 | |
9462 | mutex_enter(&spa->spa_props_lock); | |
34dc7c2f | 9463 | |
34dc7c2f | 9464 | while ((elem = nvlist_next_nvpair(nvp, elem))) { |
9ae529ec | 9465 | uint64_t intval; |
d1807f16 | 9466 | const char *strval, *fname; |
9ae529ec CS |
9467 | zpool_prop_t prop; |
9468 | const char *propname; | |
8eae2d21 | 9469 | const char *elemname = nvpair_name(elem); |
9ae529ec | 9470 | zprop_type_t proptype; |
fa86b5db | 9471 | spa_feature_t fid; |
9ae529ec | 9472 | |
8eae2d21 | 9473 | switch (prop = zpool_name_to_prop(elemname)) { |
34dc7c2f | 9474 | case ZPOOL_PROP_VERSION: |
93cf2076 | 9475 | intval = fnvpair_value_uint64(elem); |
34dc7c2f | 9476 | /* |
4e33ba4c | 9477 | * The version is synced separately before other |
9ae529ec | 9478 | * properties and should be correct by now. |
34dc7c2f | 9479 | */ |
9ae529ec | 9480 | ASSERT3U(spa_version(spa), >=, intval); |
34dc7c2f BB |
9481 | break; |
9482 | ||
9483 | case ZPOOL_PROP_ALTROOT: | |
9484 | /* | |
9485 | * 'altroot' is a non-persistent property. It should | |
9486 | * have been set temporarily at creation or import time. | |
9487 | */ | |
9488 | ASSERT(spa->spa_root != NULL); | |
9489 | break; | |
9490 | ||
572e2857 | 9491 | case ZPOOL_PROP_READONLY: |
34dc7c2f BB |
9492 | case ZPOOL_PROP_CACHEFILE: |
9493 | /* | |
e1cfd73f | 9494 | * 'readonly' and 'cachefile' are also non-persistent |
572e2857 | 9495 | * properties. |
34dc7c2f | 9496 | */ |
34dc7c2f | 9497 | break; |
d96eb2b1 | 9498 | case ZPOOL_PROP_COMMENT: |
93cf2076 | 9499 | strval = fnvpair_value_string(elem); |
d96eb2b1 DM |
9500 | if (spa->spa_comment != NULL) |
9501 | spa_strfree(spa->spa_comment); | |
9502 | spa->spa_comment = spa_strdup(strval); | |
9503 | /* | |
9504 | * We need to dirty the configuration on all the vdevs | |
88a48330 BB |
9505 | * so that their labels get updated. We also need to |
9506 | * update the cache file to keep it in sync with the | |
9507 | * MOS version. It's unnecessary to do this for pool | |
9508 | * creation since the vdev's configuration has already | |
9509 | * been dirtied. | |
d96eb2b1 | 9510 | */ |
88a48330 | 9511 | if (tx->tx_txg != TXG_INITIAL) { |
d96eb2b1 | 9512 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
9513 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
9514 | } | |
6f1ffb06 | 9515 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9516 | "%s=%s", elemname, strval); |
d96eb2b1 | 9517 | break; |
658fb802 CB |
9518 | case ZPOOL_PROP_COMPATIBILITY: |
9519 | strval = fnvpair_value_string(elem); | |
9520 | if (spa->spa_compatibility != NULL) | |
9521 | spa_strfree(spa->spa_compatibility); | |
9522 | spa->spa_compatibility = spa_strdup(strval); | |
9523 | /* | |
9524 | * Dirty the configuration on vdevs as above. | |
9525 | */ | |
88a48330 | 9526 | if (tx->tx_txg != TXG_INITIAL) { |
658fb802 | 9527 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
9528 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
9529 | } | |
9530 | ||
658fb802 CB |
9531 | spa_history_log_internal(spa, "set", tx, |
9532 | "%s=%s", nvpair_name(elem), strval); | |
9533 | break; | |
9534 | ||
8eae2d21 AJ |
9535 | case ZPOOL_PROP_INVAL: |
9536 | if (zpool_prop_feature(elemname)) { | |
9537 | fname = strchr(elemname, '@') + 1; | |
9538 | VERIFY0(zfeature_lookup_name(fname, &fid)); | |
9539 | ||
9540 | spa_feature_enable(spa, fid, tx); | |
9541 | spa_history_log_internal(spa, "set", tx, | |
9542 | "%s=enabled", elemname); | |
9543 | break; | |
9544 | } else if (!zfs_prop_user(elemname)) { | |
9545 | ASSERT(zpool_prop_feature(elemname)); | |
9546 | break; | |
9547 | } | |
9548 | zfs_fallthrough; | |
34dc7c2f BB |
9549 | default: |
9550 | /* | |
9551 | * Set pool property values in the poolprops mos object. | |
9552 | */ | |
34dc7c2f | 9553 | if (spa->spa_pool_props_object == 0) { |
9ae529ec CS |
9554 | spa->spa_pool_props_object = |
9555 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
34dc7c2f | 9556 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, |
9ae529ec | 9557 | tx); |
34dc7c2f | 9558 | } |
34dc7c2f BB |
9559 | |
9560 | /* normalize the property name */ | |
ee7b71db | 9561 | if (prop == ZPOOL_PROP_INVAL) { |
8eae2d21 AJ |
9562 | propname = elemname; |
9563 | proptype = PROP_TYPE_STRING; | |
ee7b71db RY |
9564 | } else { |
9565 | propname = zpool_prop_to_name(prop); | |
9566 | proptype = zpool_prop_get_type(prop); | |
8eae2d21 | 9567 | } |
34dc7c2f BB |
9568 | |
9569 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
9570 | ASSERT(proptype == PROP_TYPE_STRING); | |
93cf2076 GW |
9571 | strval = fnvpair_value_string(elem); |
9572 | VERIFY0(zap_update(mos, | |
34dc7c2f | 9573 | spa->spa_pool_props_object, propname, |
93cf2076 | 9574 | 1, strlen(strval) + 1, strval, tx)); |
6f1ffb06 | 9575 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9576 | "%s=%s", elemname, strval); |
34dc7c2f | 9577 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { |
93cf2076 | 9578 | intval = fnvpair_value_uint64(elem); |
34dc7c2f BB |
9579 | |
9580 | if (proptype == PROP_TYPE_INDEX) { | |
9581 | const char *unused; | |
93cf2076 GW |
9582 | VERIFY0(zpool_prop_index_to_string( |
9583 | prop, intval, &unused)); | |
34dc7c2f | 9584 | } |
93cf2076 | 9585 | VERIFY0(zap_update(mos, |
34dc7c2f | 9586 | spa->spa_pool_props_object, propname, |
93cf2076 | 9587 | 8, 1, &intval, tx)); |
6f1ffb06 | 9588 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9589 | "%s=%lld", elemname, |
74756182 | 9590 | (longlong_t)intval); |
34dc7c2f | 9591 | |
44f71818 RY |
9592 | switch (prop) { |
9593 | case ZPOOL_PROP_DELEGATION: | |
9594 | spa->spa_delegation = intval; | |
9595 | break; | |
9596 | case ZPOOL_PROP_BOOTFS: | |
9597 | spa->spa_bootfs = intval; | |
9598 | break; | |
9599 | case ZPOOL_PROP_FAILUREMODE: | |
9600 | spa->spa_failmode = intval; | |
9601 | break; | |
9602 | case ZPOOL_PROP_AUTOTRIM: | |
9603 | spa->spa_autotrim = intval; | |
428870ff | 9604 | spa_async_request(spa, |
44f71818 RY |
9605 | SPA_ASYNC_AUTOTRIM_RESTART); |
9606 | break; | |
9607 | case ZPOOL_PROP_AUTOEXPAND: | |
9608 | spa->spa_autoexpand = intval; | |
9609 | if (tx->tx_txg != TXG_INITIAL) | |
9610 | spa_async_request(spa, | |
9611 | SPA_ASYNC_AUTOEXPAND); | |
9612 | break; | |
9613 | case ZPOOL_PROP_MULTIHOST: | |
9614 | spa->spa_multihost = intval; | |
9615 | break; | |
9616 | default: | |
9617 | break; | |
9618 | } | |
9619 | } else { | |
9620 | ASSERT(0); /* not allowed */ | |
34dc7c2f BB |
9621 | } |
9622 | } | |
9623 | ||
34dc7c2f | 9624 | } |
b128c09f BB |
9625 | |
9626 | mutex_exit(&spa->spa_props_lock); | |
34dc7c2f BB |
9627 | } |
9628 | ||
428870ff BB |
9629 | /* |
9630 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
9631 | * reflect the new version this txg, so there must be no changes this | |
9632 | * txg to anything that the upgrade code depends on after it executes. | |
9633 | * Therefore this must be called after dsl_pool_sync() does the sync | |
9634 | * tasks. | |
9635 | */ | |
9636 | static void | |
9637 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
9638 | { | |
8dc2197b SD |
9639 | if (spa_sync_pass(spa) != 1) |
9640 | return; | |
428870ff | 9641 | |
8dc2197b | 9642 | dsl_pool_t *dp = spa->spa_dsl_pool; |
13fe0198 MA |
9643 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
9644 | ||
428870ff BB |
9645 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && |
9646 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
9647 | dsl_pool_create_origin(dp, tx); | |
9648 | ||
9649 | /* Keeping the origin open increases spa_minref */ | |
9650 | spa->spa_minref += 3; | |
9651 | } | |
9652 | ||
9653 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
9654 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
9655 | dsl_pool_upgrade_clones(dp, tx); | |
9656 | } | |
9657 | ||
9658 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
9659 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
9660 | dsl_pool_upgrade_dir_clones(dp, tx); | |
9661 | ||
9662 | /* Keeping the freedir open increases spa_minref */ | |
9663 | spa->spa_minref += 3; | |
9664 | } | |
9ae529ec CS |
9665 | |
9666 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
9667 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9668 | spa_feature_create_zap_objects(spa, tx); | |
9669 | } | |
62bdd5eb DL |
9670 | |
9671 | /* | |
9672 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
9673 | * when possibility to use lz4 compression for metadata was added | |
9674 | * Old pools that have this feature enabled must be upgraded to have | |
9675 | * this feature active | |
9676 | */ | |
9677 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9678 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
9679 | SPA_FEATURE_LZ4_COMPRESS); | |
9680 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
9681 | SPA_FEATURE_LZ4_COMPRESS); | |
9682 | ||
9683 | if (lz4_en && !lz4_ac) | |
9684 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
9685 | } | |
3c67d83a TH |
9686 | |
9687 | /* | |
9688 | * If we haven't written the salt, do so now. Note that the | |
9689 | * feature may not be activated yet, but that's fine since | |
9690 | * the presence of this ZAP entry is backwards compatible. | |
9691 | */ | |
9692 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
9693 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
9694 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9695 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
9696 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
9697 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
9698 | } | |
9699 | ||
13fe0198 | 9700 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
428870ff BB |
9701 | } |
9702 | ||
a1d477c2 MA |
9703 | static void |
9704 | vdev_indirect_state_sync_verify(vdev_t *vd) | |
9705 | { | |
2a8ba608 MM |
9706 | vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping; |
9707 | vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births; | |
a1d477c2 MA |
9708 | |
9709 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
9710 | ASSERT(vim != NULL); | |
9711 | ASSERT(vib != NULL); | |
9712 | } | |
9713 | ||
27f80e85 BB |
9714 | uint64_t obsolete_sm_object = 0; |
9715 | ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); | |
9716 | if (obsolete_sm_object != 0) { | |
a1d477c2 MA |
9717 | ASSERT(vd->vdev_obsolete_sm != NULL); |
9718 | ASSERT(vd->vdev_removing || | |
9719 | vd->vdev_ops == &vdev_indirect_ops); | |
9720 | ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); | |
9721 | ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); | |
27f80e85 | 9722 | ASSERT3U(obsolete_sm_object, ==, |
a1d477c2 MA |
9723 | space_map_object(vd->vdev_obsolete_sm)); |
9724 | ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, | |
9725 | space_map_allocated(vd->vdev_obsolete_sm)); | |
9726 | } | |
9727 | ASSERT(vd->vdev_obsolete_segments != NULL); | |
9728 | ||
9729 | /* | |
9730 | * Since frees / remaps to an indirect vdev can only | |
9731 | * happen in syncing context, the obsolete segments | |
9732 | * tree must be empty when we start syncing. | |
9733 | */ | |
9734 | ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); | |
9735 | } | |
9736 | ||
34dc7c2f | 9737 | /* |
8dc2197b SD |
9738 | * Set the top-level vdev's max queue depth. Evaluate each top-level's |
9739 | * async write queue depth in case it changed. The max queue depth will | |
9740 | * not change in the middle of syncing out this txg. | |
34dc7c2f | 9741 | */ |
8dc2197b SD |
9742 | static void |
9743 | spa_sync_adjust_vdev_max_queue_depth(spa_t *spa) | |
34dc7c2f | 9744 | { |
8dc2197b SD |
9745 | ASSERT(spa_writeable(spa)); |
9746 | ||
34dc7c2f | 9747 | vdev_t *rvd = spa->spa_root_vdev; |
3dfb57a3 DB |
9748 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * |
9749 | zfs_vdev_queue_depth_pct / 100; | |
8dc2197b SD |
9750 | metaslab_class_t *normal = spa_normal_class(spa); |
9751 | metaslab_class_t *special = spa_special_class(spa); | |
9752 | metaslab_class_t *dedup = spa_dedup_class(spa); | |
34dc7c2f | 9753 | |
492f64e9 | 9754 | uint64_t slots_per_allocator = 0; |
1c27024e | 9755 | for (int c = 0; c < rvd->vdev_children; c++) { |
3dfb57a3 | 9756 | vdev_t *tvd = rvd->vdev_child[c]; |
cc99f275 | 9757 | |
8dc2197b | 9758 | metaslab_group_t *mg = tvd->vdev_mg; |
cc99f275 DB |
9759 | if (mg == NULL || !metaslab_group_initialized(mg)) |
9760 | continue; | |
3dfb57a3 | 9761 | |
8dc2197b | 9762 | metaslab_class_t *mc = mg->mg_class; |
cc99f275 | 9763 | if (mc != normal && mc != special && mc != dedup) |
3dfb57a3 DB |
9764 | continue; |
9765 | ||
9766 | /* | |
9767 | * It is safe to do a lock-free check here because only async | |
9768 | * allocations look at mg_max_alloc_queue_depth, and async | |
9769 | * allocations all happen from spa_sync(). | |
9770 | */ | |
32d805c3 | 9771 | for (int i = 0; i < mg->mg_allocators; i++) { |
424fd7c3 | 9772 | ASSERT0(zfs_refcount_count( |
32d805c3 MA |
9773 | &(mg->mg_allocator[i].mga_alloc_queue_depth))); |
9774 | } | |
3dfb57a3 | 9775 | mg->mg_max_alloc_queue_depth = max_queue_depth; |
492f64e9 | 9776 | |
32d805c3 MA |
9777 | for (int i = 0; i < mg->mg_allocators; i++) { |
9778 | mg->mg_allocator[i].mga_cur_max_alloc_queue_depth = | |
492f64e9 PD |
9779 | zfs_vdev_def_queue_depth; |
9780 | } | |
9781 | slots_per_allocator += zfs_vdev_def_queue_depth; | |
3dfb57a3 | 9782 | } |
cc99f275 | 9783 | |
492f64e9 | 9784 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
f8020c93 AM |
9785 | ASSERT0(zfs_refcount_count(&normal->mc_allocator[i]. |
9786 | mca_alloc_slots)); | |
9787 | ASSERT0(zfs_refcount_count(&special->mc_allocator[i]. | |
9788 | mca_alloc_slots)); | |
9789 | ASSERT0(zfs_refcount_count(&dedup->mc_allocator[i]. | |
9790 | mca_alloc_slots)); | |
9791 | normal->mc_allocator[i].mca_alloc_max_slots = | |
9792 | slots_per_allocator; | |
9793 | special->mc_allocator[i].mca_alloc_max_slots = | |
9794 | slots_per_allocator; | |
9795 | dedup->mc_allocator[i].mca_alloc_max_slots = | |
9796 | slots_per_allocator; | |
cc99f275 DB |
9797 | } |
9798 | normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9799 | special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9800 | dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
8dc2197b SD |
9801 | } |
9802 | ||
9803 | static void | |
9804 | spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx) | |
9805 | { | |
9806 | ASSERT(spa_writeable(spa)); | |
3dfb57a3 | 9807 | |
8dc2197b | 9808 | vdev_t *rvd = spa->spa_root_vdev; |
a1d477c2 MA |
9809 | for (int c = 0; c < rvd->vdev_children; c++) { |
9810 | vdev_t *vd = rvd->vdev_child[c]; | |
9811 | vdev_indirect_state_sync_verify(vd); | |
9812 | ||
9813 | if (vdev_indirect_should_condense(vd)) { | |
9814 | spa_condense_indirect_start_sync(vd, tx); | |
9815 | break; | |
9816 | } | |
9817 | } | |
8dc2197b SD |
9818 | } |
9819 | ||
9820 | static void | |
9821 | spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx) | |
9822 | { | |
9823 | objset_t *mos = spa->spa_meta_objset; | |
9824 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9825 | uint64_t txg = tx->tx_txg; | |
9826 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
a1d477c2 | 9827 | |
34dc7c2f | 9828 | do { |
428870ff | 9829 | int pass = ++spa->spa_sync_pass; |
34dc7c2f BB |
9830 | |
9831 | spa_sync_config_object(spa, tx); | |
9832 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
9833 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
9834 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
9835 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
9836 | spa_errlog_sync(spa, txg); | |
9837 | dsl_pool_sync(dp, txg); | |
9838 | ||
93e28d66 SD |
9839 | if (pass < zfs_sync_pass_deferred_free || |
9840 | spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { | |
9841 | /* | |
9842 | * If the log space map feature is active we don't | |
9843 | * care about deferred frees and the deferred bpobj | |
9844 | * as the log space map should effectively have the | |
9845 | * same results (i.e. appending only to one object). | |
9846 | */ | |
e8b96c60 | 9847 | spa_sync_frees(spa, free_bpl, tx); |
428870ff | 9848 | } else { |
905edb40 MA |
9849 | /* |
9850 | * We can not defer frees in pass 1, because | |
9851 | * we sync the deferred frees later in pass 1. | |
9852 | */ | |
9853 | ASSERT3U(pass, >, 1); | |
37f03da8 | 9854 | bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb, |
e8b96c60 | 9855 | &spa->spa_deferred_bpobj, tx); |
34dc7c2f BB |
9856 | } |
9857 | ||
67a1b037 | 9858 | brt_sync(spa, txg); |
428870ff BB |
9859 | ddt_sync(spa, txg); |
9860 | dsl_scan_sync(dp, tx); | |
482eeef8 | 9861 | dsl_errorscrub_sync(dp, tx); |
8dc2197b SD |
9862 | svr_sync(spa, tx); |
9863 | spa_sync_upgrades(spa, tx); | |
34dc7c2f | 9864 | |
93e28d66 SD |
9865 | spa_flush_metaslabs(spa, tx); |
9866 | ||
8dc2197b | 9867 | vdev_t *vd = NULL; |
a1d477c2 MA |
9868 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) |
9869 | != NULL) | |
428870ff BB |
9870 | vdev_sync(vd, txg); |
9871 | ||
5caeef02 DB |
9872 | if (pass == 1) { |
9873 | /* | |
9874 | * dsl_pool_sync() -> dp_sync_tasks may have dirtied | |
9875 | * the config. If that happens, this txg should not | |
9876 | * be a no-op. So we must sync the config to the MOS | |
9877 | * before checking for no-op. | |
9878 | * | |
9879 | * Note that when the config is dirty, it will | |
9880 | * be written to the MOS (i.e. the MOS will be | |
9881 | * dirtied) every time we call spa_sync_config_object() | |
9882 | * in this txg. Therefore we can't call this after | |
9883 | * dsl_pool_sync() every pass, because it would | |
9884 | * prevent us from converging, since we'd dirty | |
9885 | * the MOS every pass. | |
9886 | * | |
9887 | * Sync tasks can only be processed in pass 1, so | |
9888 | * there's no need to do this in later passes. | |
9889 | */ | |
9890 | spa_sync_config_object(spa, tx); | |
9891 | } | |
9892 | ||
8dc2197b SD |
9893 | /* |
9894 | * Note: We need to check if the MOS is dirty because we could | |
9895 | * have marked the MOS dirty without updating the uberblock | |
9896 | * (e.g. if we have sync tasks but no dirty user data). We need | |
9897 | * to check the uberblock's rootbp because it is updated if we | |
9898 | * have synced out dirty data (though in this case the MOS will | |
9899 | * most likely also be dirty due to second order effects, we | |
9900 | * don't want to rely on that here). | |
9901 | */ | |
9902 | if (pass == 1 && | |
493fcce9 | 9903 | BP_GET_LOGICAL_BIRTH(&spa->spa_uberblock.ub_rootbp) < txg && |
8dc2197b | 9904 | !dmu_objset_is_dirty(mos, txg)) { |
905edb40 | 9905 | /* |
8dc2197b SD |
9906 | * Nothing changed on the first pass, therefore this |
9907 | * TXG is a no-op. Avoid syncing deferred frees, so | |
9908 | * that we can keep this TXG as a no-op. | |
905edb40 | 9909 | */ |
8dc2197b SD |
9910 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); |
9911 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9912 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
9913 | ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg)); | |
9914 | break; | |
905edb40 | 9915 | } |
34dc7c2f | 9916 | |
8dc2197b | 9917 | spa_sync_deferred_frees(spa, tx); |
428870ff | 9918 | } while (dmu_objset_is_dirty(mos, txg)); |
8dc2197b | 9919 | } |
34dc7c2f | 9920 | |
8dc2197b SD |
9921 | /* |
9922 | * Rewrite the vdev configuration (which includes the uberblock) to | |
9923 | * commit the transaction group. | |
9924 | * | |
9925 | * If there are no dirty vdevs, we sync the uberblock to a few random | |
9926 | * top-level vdevs that are known to be visible in the config cache | |
9927 | * (see spa_vdev_add() for a complete description). If there *are* dirty | |
9928 | * vdevs, sync the uberblock to all vdevs. | |
9929 | */ | |
9930 | static void | |
9931 | spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx) | |
9932 | { | |
9933 | vdev_t *rvd = spa->spa_root_vdev; | |
9934 | uint64_t txg = tx->tx_txg; | |
a1d477c2 | 9935 | |
b128c09f | 9936 | for (;;) { |
8dc2197b SD |
9937 | int error = 0; |
9938 | ||
b128c09f BB |
9939 | /* |
9940 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
9941 | * while we're attempting to write the vdev labels. | |
9942 | */ | |
9943 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9944 | ||
9945 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
d2734cce | 9946 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; |
b128c09f BB |
9947 | int svdcount = 0; |
9948 | int children = rvd->vdev_children; | |
29274c9f | 9949 | int c0 = random_in_range(children); |
b128c09f | 9950 | |
1c27024e | 9951 | for (int c = 0; c < children; c++) { |
8dc2197b SD |
9952 | vdev_t *vd = |
9953 | rvd->vdev_child[(c0 + c) % children]; | |
d2734cce SD |
9954 | |
9955 | /* Stop when revisiting the first vdev */ | |
9956 | if (c > 0 && svd[0] == vd) | |
9957 | break; | |
9958 | ||
8dc2197b SD |
9959 | if (vd->vdev_ms_array == 0 || |
9960 | vd->vdev_islog || | |
a1d477c2 | 9961 | !vdev_is_concrete(vd)) |
b128c09f | 9962 | continue; |
d2734cce | 9963 | |
b128c09f | 9964 | svd[svdcount++] = vd; |
6cb8e530 | 9965 | if (svdcount == SPA_SYNC_MIN_VDEVS) |
b128c09f BB |
9966 | break; |
9967 | } | |
b6fcb792 | 9968 | error = vdev_config_sync(svd, svdcount, txg); |
b128c09f BB |
9969 | } else { |
9970 | error = vdev_config_sync(rvd->vdev_child, | |
b6fcb792 | 9971 | rvd->vdev_children, txg); |
34dc7c2f | 9972 | } |
34dc7c2f | 9973 | |
3bc7e0fb GW |
9974 | if (error == 0) |
9975 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
9976 | ||
b128c09f BB |
9977 | spa_config_exit(spa, SCL_STATE, FTAG); |
9978 | ||
9979 | if (error == 0) | |
9980 | break; | |
cec3a0a1 | 9981 | zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR); |
b128c09f BB |
9982 | zio_resume_wait(spa); |
9983 | } | |
8dc2197b SD |
9984 | } |
9985 | ||
9986 | /* | |
9987 | * Sync the specified transaction group. New blocks may be dirtied as | |
9988 | * part of the process, so we iterate until it converges. | |
9989 | */ | |
9990 | void | |
9991 | spa_sync(spa_t *spa, uint64_t txg) | |
9992 | { | |
9993 | vdev_t *vd = NULL; | |
9994 | ||
9995 | VERIFY(spa_writeable(spa)); | |
9996 | ||
9997 | /* | |
9998 | * Wait for i/os issued in open context that need to complete | |
9999 | * before this txg syncs. | |
10000 | */ | |
10001 | (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]); | |
10002 | spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, | |
10003 | ZIO_FLAG_CANFAIL); | |
10004 | ||
67a1b037 PJD |
10005 | /* |
10006 | * Now that there can be no more cloning in this transaction group, | |
10007 | * but we are still before issuing frees, we can process pending BRT | |
10008 | * updates. | |
10009 | */ | |
10010 | brt_pending_apply(spa, txg); | |
10011 | ||
8dc2197b SD |
10012 | /* |
10013 | * Lock out configuration changes. | |
10014 | */ | |
10015 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
10016 | ||
10017 | spa->spa_syncing_txg = txg; | |
10018 | spa->spa_sync_pass = 0; | |
10019 | ||
10020 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
1b50749c AM |
10021 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
10022 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
10023 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
8dc2197b SD |
10024 | } |
10025 | ||
10026 | /* | |
10027 | * If there are any pending vdev state changes, convert them | |
10028 | * into config changes that go out with this transaction group. | |
10029 | */ | |
10030 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9f08b6e3 RY |
10031 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { |
10032 | /* Avoid holding the write lock unless actually necessary */ | |
10033 | if (vd->vdev_aux == NULL) { | |
10034 | vdev_state_clean(vd); | |
10035 | vdev_config_dirty(vd); | |
10036 | continue; | |
10037 | } | |
8dc2197b SD |
10038 | /* |
10039 | * We need the write lock here because, for aux vdevs, | |
10040 | * calling vdev_config_dirty() modifies sav_config. | |
10041 | * This is ugly and will become unnecessary when we | |
10042 | * eliminate the aux vdev wart by integrating all vdevs | |
10043 | * into the root vdev tree. | |
10044 | */ | |
10045 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10046 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
10047 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
10048 | vdev_state_clean(vd); | |
10049 | vdev_config_dirty(vd); | |
10050 | } | |
10051 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10052 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10053 | } | |
10054 | spa_config_exit(spa, SCL_STATE, FTAG); | |
10055 | ||
10056 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
10057 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
10058 | ||
10059 | spa->spa_sync_starttime = gethrtime(); | |
10060 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
10061 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
10062 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
10063 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
10064 | ||
10065 | /* | |
10066 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
10067 | * set spa_deflate if we have no raid-z vdevs. | |
10068 | */ | |
10069 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
10070 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
10071 | vdev_t *rvd = spa->spa_root_vdev; | |
10072 | ||
10073 | int i; | |
10074 | for (i = 0; i < rvd->vdev_children; i++) { | |
10075 | vd = rvd->vdev_child[i]; | |
10076 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
10077 | break; | |
10078 | } | |
10079 | if (i == rvd->vdev_children) { | |
10080 | spa->spa_deflate = TRUE; | |
10081 | VERIFY0(zap_add(spa->spa_meta_objset, | |
10082 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
10083 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
10084 | } | |
10085 | } | |
10086 | ||
10087 | spa_sync_adjust_vdev_max_queue_depth(spa); | |
10088 | ||
10089 | spa_sync_condense_indirect(spa, tx); | |
10090 | ||
10091 | spa_sync_iterate_to_convergence(spa, tx); | |
10092 | ||
10093 | #ifdef ZFS_DEBUG | |
10094 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
10095 | /* | |
10096 | * Make sure that the number of ZAPs for all the vdevs matches | |
10097 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
10098 | * called if the config is dirty; otherwise there may be | |
10099 | * outstanding AVZ operations that weren't completed in | |
10100 | * spa_sync_config_object. | |
10101 | */ | |
10102 | uint64_t all_vdev_zap_entry_count; | |
10103 | ASSERT0(zap_count(spa->spa_meta_objset, | |
10104 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
10105 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
10106 | all_vdev_zap_entry_count); | |
10107 | } | |
10108 | #endif | |
10109 | ||
10110 | if (spa->spa_vdev_removal != NULL) { | |
10111 | ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); | |
10112 | } | |
10113 | ||
10114 | spa_sync_rewrite_vdev_config(spa, tx); | |
34dc7c2f BB |
10115 | dmu_tx_commit(tx); |
10116 | ||
57ddcda1 | 10117 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 GW |
10118 | spa->spa_deadman_tqid = 0; |
10119 | ||
34dc7c2f BB |
10120 | /* |
10121 | * Clear the dirty config list. | |
10122 | */ | |
b128c09f | 10123 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) |
34dc7c2f BB |
10124 | vdev_config_clean(vd); |
10125 | ||
10126 | /* | |
10127 | * Now that the new config has synced transactionally, | |
10128 | * let it become visible to the config cache. | |
10129 | */ | |
10130 | if (spa->spa_config_syncing != NULL) { | |
10131 | spa_config_set(spa, spa->spa_config_syncing); | |
10132 | spa->spa_config_txg = txg; | |
10133 | spa->spa_config_syncing = NULL; | |
10134 | } | |
10135 | ||
428870ff | 10136 | dsl_pool_sync_done(dp, txg); |
34dc7c2f | 10137 | |
492f64e9 | 10138 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
1b50749c AM |
10139 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
10140 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
10141 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
492f64e9 | 10142 | } |
3dfb57a3 | 10143 | |
34dc7c2f BB |
10144 | /* |
10145 | * Update usable space statistics. | |
10146 | */ | |
619f0976 GW |
10147 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) |
10148 | != NULL) | |
34dc7c2f | 10149 | vdev_sync_done(vd, txg); |
f09fda50 PD |
10150 | |
10151 | metaslab_class_evict_old(spa->spa_normal_class, txg); | |
10152 | metaslab_class_evict_old(spa->spa_log_class, txg); | |
10153 | ||
93e28d66 | 10154 | spa_sync_close_syncing_log_sm(spa); |
34dc7c2f | 10155 | |
428870ff BB |
10156 | spa_update_dspace(spa); |
10157 | ||
65d10bd8 KJ |
10158 | if (spa_get_autotrim(spa) == SPA_AUTOTRIM_ON) |
10159 | vdev_autotrim_kick(spa); | |
10160 | ||
34dc7c2f BB |
10161 | /* |
10162 | * It had better be the case that we didn't dirty anything | |
10163 | * since vdev_config_sync(). | |
10164 | */ | |
10165 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
10166 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
10167 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
428870ff | 10168 | |
d2734cce SD |
10169 | while (zfs_pause_spa_sync) |
10170 | delay(1); | |
10171 | ||
428870ff | 10172 | spa->spa_sync_pass = 0; |
34dc7c2f | 10173 | |
55922e73 GW |
10174 | /* |
10175 | * Update the last synced uberblock here. We want to do this at | |
10176 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
10177 | * will be guaranteed that all the processing associated with | |
10178 | * that txg has been completed. | |
10179 | */ | |
10180 | spa->spa_ubsync = spa->spa_uberblock; | |
b128c09f | 10181 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f | 10182 | |
428870ff BB |
10183 | spa_handle_ignored_writes(spa); |
10184 | ||
34dc7c2f BB |
10185 | /* |
10186 | * If any async tasks have been requested, kick them off. | |
10187 | */ | |
10188 | spa_async_dispatch(spa); | |
10189 | } | |
10190 | ||
10191 | /* | |
10192 | * Sync all pools. We don't want to hold the namespace lock across these | |
10193 | * operations, so we take a reference on the spa_t and drop the lock during the | |
10194 | * sync. | |
10195 | */ | |
10196 | void | |
10197 | spa_sync_allpools(void) | |
10198 | { | |
10199 | spa_t *spa = NULL; | |
10200 | mutex_enter(&spa_namespace_lock); | |
10201 | while ((spa = spa_next(spa)) != NULL) { | |
572e2857 BB |
10202 | if (spa_state(spa) != POOL_STATE_ACTIVE || |
10203 | !spa_writeable(spa) || spa_suspended(spa)) | |
34dc7c2f BB |
10204 | continue; |
10205 | spa_open_ref(spa, FTAG); | |
10206 | mutex_exit(&spa_namespace_lock); | |
10207 | txg_wait_synced(spa_get_dsl(spa), 0); | |
10208 | mutex_enter(&spa_namespace_lock); | |
10209 | spa_close(spa, FTAG); | |
10210 | } | |
10211 | mutex_exit(&spa_namespace_lock); | |
10212 | } | |
10213 | ||
3bd4df38 EN |
10214 | taskq_t * |
10215 | spa_sync_tq_create(spa_t *spa, const char *name) | |
10216 | { | |
10217 | kthread_t **kthreads; | |
10218 | ||
10219 | ASSERT(spa->spa_sync_tq == NULL); | |
10220 | ASSERT3S(spa->spa_alloc_count, <=, boot_ncpus); | |
10221 | ||
10222 | /* | |
10223 | * - do not allow more allocators than cpus. | |
10224 | * - there may be more cpus than allocators. | |
10225 | * - do not allow more sync taskq threads than allocators or cpus. | |
10226 | */ | |
10227 | int nthreads = spa->spa_alloc_count; | |
10228 | spa->spa_syncthreads = kmem_zalloc(sizeof (spa_syncthread_info_t) * | |
10229 | nthreads, KM_SLEEP); | |
10230 | ||
10231 | spa->spa_sync_tq = taskq_create_synced(name, nthreads, minclsyspri, | |
10232 | nthreads, INT_MAX, TASKQ_PREPOPULATE, &kthreads); | |
10233 | VERIFY(spa->spa_sync_tq != NULL); | |
10234 | VERIFY(kthreads != NULL); | |
10235 | ||
3bd4df38 | 10236 | spa_syncthread_info_t *ti = spa->spa_syncthreads; |
645b8330 | 10237 | for (int i = 0; i < nthreads; i++, ti++) { |
3bd4df38 | 10238 | ti->sti_thread = kthreads[i]; |
645b8330 | 10239 | ti->sti_allocator = i; |
3bd4df38 EN |
10240 | } |
10241 | ||
10242 | kmem_free(kthreads, sizeof (*kthreads) * nthreads); | |
10243 | return (spa->spa_sync_tq); | |
10244 | } | |
10245 | ||
10246 | void | |
10247 | spa_sync_tq_destroy(spa_t *spa) | |
10248 | { | |
10249 | ASSERT(spa->spa_sync_tq != NULL); | |
10250 | ||
10251 | taskq_wait(spa->spa_sync_tq); | |
10252 | taskq_destroy(spa->spa_sync_tq); | |
10253 | kmem_free(spa->spa_syncthreads, | |
10254 | sizeof (spa_syncthread_info_t) * spa->spa_alloc_count); | |
10255 | spa->spa_sync_tq = NULL; | |
10256 | } | |
10257 | ||
645b8330 AM |
10258 | uint_t |
10259 | spa_acq_allocator(spa_t *spa) | |
10260 | { | |
10261 | int i; | |
10262 | ||
10263 | if (spa->spa_alloc_count == 1) | |
10264 | return (0); | |
10265 | ||
10266 | mutex_enter(&spa->spa_allocs_use->sau_lock); | |
10267 | uint_t r = spa->spa_allocs_use->sau_rotor; | |
10268 | do { | |
10269 | if (++r == spa->spa_alloc_count) | |
10270 | r = 0; | |
10271 | } while (spa->spa_allocs_use->sau_inuse[r]); | |
10272 | spa->spa_allocs_use->sau_inuse[r] = B_TRUE; | |
10273 | spa->spa_allocs_use->sau_rotor = r; | |
10274 | mutex_exit(&spa->spa_allocs_use->sau_lock); | |
10275 | ||
10276 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
10277 | for (i = 0; i < spa->spa_alloc_count; i++, ti++) { | |
10278 | if (ti->sti_thread == curthread) { | |
10279 | ti->sti_allocator = r; | |
10280 | break; | |
10281 | } | |
10282 | } | |
10283 | ASSERT3S(i, <, spa->spa_alloc_count); | |
10284 | return (r); | |
10285 | } | |
10286 | ||
10287 | void | |
10288 | spa_rel_allocator(spa_t *spa, uint_t allocator) | |
10289 | { | |
10290 | if (spa->spa_alloc_count > 1) | |
10291 | spa->spa_allocs_use->sau_inuse[allocator] = B_FALSE; | |
10292 | } | |
10293 | ||
3bd4df38 EN |
10294 | void |
10295 | spa_select_allocator(zio_t *zio) | |
10296 | { | |
10297 | zbookmark_phys_t *bm = &zio->io_bookmark; | |
10298 | spa_t *spa = zio->io_spa; | |
10299 | ||
10300 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
10301 | ||
10302 | /* | |
10303 | * A gang block (for example) may have inherited its parent's | |
10304 | * allocator, in which case there is nothing further to do here. | |
10305 | */ | |
10306 | if (ZIO_HAS_ALLOCATOR(zio)) | |
10307 | return; | |
10308 | ||
10309 | ASSERT(spa != NULL); | |
10310 | ASSERT(bm != NULL); | |
10311 | ||
10312 | /* | |
10313 | * First try to use an allocator assigned to the syncthread, and set | |
10314 | * the corresponding write issue taskq for the allocator. | |
10315 | * Note, we must have an open pool to do this. | |
10316 | */ | |
10317 | if (spa->spa_sync_tq != NULL) { | |
10318 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
10319 | for (int i = 0; i < spa->spa_alloc_count; i++, ti++) { | |
10320 | if (ti->sti_thread == curthread) { | |
645b8330 | 10321 | zio->io_allocator = ti->sti_allocator; |
3bd4df38 EN |
10322 | return; |
10323 | } | |
10324 | } | |
10325 | } | |
10326 | ||
10327 | /* | |
10328 | * We want to try to use as many allocators as possible to help improve | |
10329 | * performance, but we also want logically adjacent IOs to be physically | |
10330 | * adjacent to improve sequential read performance. We chunk each object | |
10331 | * into 2^20 block regions, and then hash based on the objset, object, | |
10332 | * level, and region to accomplish both of these goals. | |
10333 | */ | |
10334 | uint64_t hv = cityhash4(bm->zb_objset, bm->zb_object, bm->zb_level, | |
10335 | bm->zb_blkid >> 20); | |
10336 | ||
10337 | zio->io_allocator = (uint_t)hv % spa->spa_alloc_count; | |
3bd4df38 EN |
10338 | } |
10339 | ||
34dc7c2f BB |
10340 | /* |
10341 | * ========================================================================== | |
10342 | * Miscellaneous routines | |
10343 | * ========================================================================== | |
10344 | */ | |
10345 | ||
10346 | /* | |
10347 | * Remove all pools in the system. | |
10348 | */ | |
10349 | void | |
10350 | spa_evict_all(void) | |
10351 | { | |
10352 | spa_t *spa; | |
10353 | ||
10354 | /* | |
10355 | * Remove all cached state. All pools should be closed now, | |
10356 | * so every spa in the AVL tree should be unreferenced. | |
10357 | */ | |
10358 | mutex_enter(&spa_namespace_lock); | |
10359 | while ((spa = spa_next(NULL)) != NULL) { | |
10360 | /* | |
10361 | * Stop async tasks. The async thread may need to detach | |
10362 | * a device that's been replaced, which requires grabbing | |
10363 | * spa_namespace_lock, so we must drop it here. | |
10364 | */ | |
10365 | spa_open_ref(spa, FTAG); | |
10366 | mutex_exit(&spa_namespace_lock); | |
10367 | spa_async_suspend(spa); | |
10368 | mutex_enter(&spa_namespace_lock); | |
34dc7c2f BB |
10369 | spa_close(spa, FTAG); |
10370 | ||
10371 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
10372 | spa_unload(spa); | |
10373 | spa_deactivate(spa); | |
10374 | } | |
10375 | spa_remove(spa); | |
10376 | } | |
10377 | mutex_exit(&spa_namespace_lock); | |
10378 | } | |
10379 | ||
10380 | vdev_t * | |
9babb374 | 10381 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) |
34dc7c2f | 10382 | { |
b128c09f BB |
10383 | vdev_t *vd; |
10384 | int i; | |
10385 | ||
10386 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
10387 | return (vd); | |
10388 | ||
9babb374 | 10389 | if (aux) { |
b128c09f BB |
10390 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { |
10391 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
9babb374 BB |
10392 | if (vd->vdev_guid == guid) |
10393 | return (vd); | |
10394 | } | |
10395 | ||
10396 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
10397 | vd = spa->spa_spares.sav_vdevs[i]; | |
b128c09f BB |
10398 | if (vd->vdev_guid == guid) |
10399 | return (vd); | |
10400 | } | |
10401 | } | |
10402 | ||
10403 | return (NULL); | |
34dc7c2f BB |
10404 | } |
10405 | ||
10406 | void | |
10407 | spa_upgrade(spa_t *spa, uint64_t version) | |
10408 | { | |
572e2857 BB |
10409 | ASSERT(spa_writeable(spa)); |
10410 | ||
b128c09f | 10411 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
10412 | |
10413 | /* | |
10414 | * This should only be called for a non-faulted pool, and since a | |
10415 | * future version would result in an unopenable pool, this shouldn't be | |
10416 | * possible. | |
10417 | */ | |
8dca0a9a | 10418 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); |
9b67f605 | 10419 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); |
34dc7c2f BB |
10420 | |
10421 | spa->spa_uberblock.ub_version = version; | |
10422 | vdev_config_dirty(spa->spa_root_vdev); | |
10423 | ||
b128c09f | 10424 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
10425 | |
10426 | txg_wait_synced(spa_get_dsl(spa), 0); | |
10427 | } | |
10428 | ||
49d42425 FU |
10429 | static boolean_t |
10430 | spa_has_aux_vdev(spa_t *spa, uint64_t guid, spa_aux_vdev_t *sav) | |
34dc7c2f | 10431 | { |
14e4e3cb | 10432 | (void) spa; |
34dc7c2f | 10433 | int i; |
49d42425 | 10434 | uint64_t vdev_guid; |
34dc7c2f BB |
10435 | |
10436 | for (i = 0; i < sav->sav_count; i++) | |
10437 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
10438 | return (B_TRUE); | |
10439 | ||
10440 | for (i = 0; i < sav->sav_npending; i++) { | |
10441 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
49d42425 | 10442 | &vdev_guid) == 0 && vdev_guid == guid) |
34dc7c2f BB |
10443 | return (B_TRUE); |
10444 | } | |
10445 | ||
10446 | return (B_FALSE); | |
10447 | } | |
10448 | ||
49d42425 FU |
10449 | boolean_t |
10450 | spa_has_l2cache(spa_t *spa, uint64_t guid) | |
10451 | { | |
10452 | return (spa_has_aux_vdev(spa, guid, &spa->spa_l2cache)); | |
10453 | } | |
10454 | ||
10455 | boolean_t | |
10456 | spa_has_spare(spa_t *spa, uint64_t guid) | |
10457 | { | |
10458 | return (spa_has_aux_vdev(spa, guid, &spa->spa_spares)); | |
10459 | } | |
10460 | ||
b128c09f BB |
10461 | /* |
10462 | * Check if a pool has an active shared spare device. | |
10463 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
10464 | */ | |
10465 | static boolean_t | |
10466 | spa_has_active_shared_spare(spa_t *spa) | |
10467 | { | |
10468 | int i, refcnt; | |
10469 | uint64_t pool; | |
10470 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
10471 | ||
10472 | for (i = 0; i < sav->sav_count; i++) { | |
10473 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
10474 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
10475 | refcnt > 2) | |
10476 | return (B_TRUE); | |
10477 | } | |
10478 | ||
10479 | return (B_FALSE); | |
10480 | } | |
10481 | ||
93e28d66 SD |
10482 | uint64_t |
10483 | spa_total_metaslabs(spa_t *spa) | |
10484 | { | |
10485 | vdev_t *rvd = spa->spa_root_vdev; | |
10486 | ||
10487 | uint64_t m = 0; | |
10488 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
10489 | vdev_t *vd = rvd->vdev_child[c]; | |
10490 | if (!vdev_is_concrete(vd)) | |
10491 | continue; | |
10492 | m += vd->vdev_ms_count; | |
10493 | } | |
10494 | return (m); | |
10495 | } | |
10496 | ||
e60e158e JG |
10497 | /* |
10498 | * Notify any waiting threads that some activity has switched from being in- | |
10499 | * progress to not-in-progress so that the thread can wake up and determine | |
10500 | * whether it is finished waiting. | |
10501 | */ | |
10502 | void | |
10503 | spa_notify_waiters(spa_t *spa) | |
10504 | { | |
10505 | /* | |
10506 | * Acquiring spa_activities_lock here prevents the cv_broadcast from | |
10507 | * happening between the waiting thread's check and cv_wait. | |
10508 | */ | |
10509 | mutex_enter(&spa->spa_activities_lock); | |
10510 | cv_broadcast(&spa->spa_activities_cv); | |
10511 | mutex_exit(&spa->spa_activities_lock); | |
10512 | } | |
10513 | ||
10514 | /* | |
10515 | * Notify any waiting threads that the pool is exporting, and then block until | |
10516 | * they are finished using the spa_t. | |
10517 | */ | |
10518 | void | |
10519 | spa_wake_waiters(spa_t *spa) | |
10520 | { | |
10521 | mutex_enter(&spa->spa_activities_lock); | |
10522 | spa->spa_waiters_cancel = B_TRUE; | |
10523 | cv_broadcast(&spa->spa_activities_cv); | |
10524 | while (spa->spa_waiters != 0) | |
10525 | cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock); | |
10526 | spa->spa_waiters_cancel = B_FALSE; | |
10527 | mutex_exit(&spa->spa_activities_lock); | |
10528 | } | |
10529 | ||
2288d419 | 10530 | /* Whether the vdev or any of its descendants are being initialized/trimmed. */ |
e60e158e | 10531 | static boolean_t |
2288d419 | 10532 | spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity) |
e60e158e JG |
10533 | { |
10534 | spa_t *spa = vd->vdev_spa; | |
e60e158e JG |
10535 | |
10536 | ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER)); | |
10537 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
2288d419 BB |
10538 | ASSERT(activity == ZPOOL_WAIT_INITIALIZE || |
10539 | activity == ZPOOL_WAIT_TRIM); | |
10540 | ||
10541 | kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ? | |
10542 | &vd->vdev_initialize_lock : &vd->vdev_trim_lock; | |
e60e158e JG |
10543 | |
10544 | mutex_exit(&spa->spa_activities_lock); | |
2288d419 | 10545 | mutex_enter(lock); |
e60e158e JG |
10546 | mutex_enter(&spa->spa_activities_lock); |
10547 | ||
2288d419 BB |
10548 | boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ? |
10549 | (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) : | |
10550 | (vd->vdev_trim_state == VDEV_TRIM_ACTIVE); | |
10551 | mutex_exit(lock); | |
e60e158e | 10552 | |
2288d419 | 10553 | if (in_progress) |
e60e158e JG |
10554 | return (B_TRUE); |
10555 | ||
10556 | for (int i = 0; i < vd->vdev_children; i++) { | |
2288d419 BB |
10557 | if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i], |
10558 | activity)) | |
e60e158e JG |
10559 | return (B_TRUE); |
10560 | } | |
10561 | ||
10562 | return (B_FALSE); | |
10563 | } | |
10564 | ||
10565 | /* | |
10566 | * If use_guid is true, this checks whether the vdev specified by guid is | |
2288d419 BB |
10567 | * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool |
10568 | * is being initialized/trimmed. The caller must hold the config lock and | |
10569 | * spa_activities_lock. | |
e60e158e JG |
10570 | */ |
10571 | static int | |
2288d419 BB |
10572 | spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid, |
10573 | zpool_wait_activity_t activity, boolean_t *in_progress) | |
e60e158e JG |
10574 | { |
10575 | mutex_exit(&spa->spa_activities_lock); | |
10576 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10577 | mutex_enter(&spa->spa_activities_lock); | |
10578 | ||
10579 | vdev_t *vd; | |
10580 | if (use_guid) { | |
10581 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
10582 | if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) { | |
10583 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10584 | return (EINVAL); | |
10585 | } | |
10586 | } else { | |
10587 | vd = spa->spa_root_vdev; | |
10588 | } | |
10589 | ||
2288d419 | 10590 | *in_progress = spa_vdev_activity_in_progress_impl(vd, activity); |
e60e158e JG |
10591 | |
10592 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10593 | return (0); | |
10594 | } | |
10595 | ||
10596 | /* | |
10597 | * Locking for waiting threads | |
10598 | * --------------------------- | |
10599 | * | |
10600 | * Waiting threads need a way to check whether a given activity is in progress, | |
10601 | * and then, if it is, wait for it to complete. Each activity will have some | |
10602 | * in-memory representation of the relevant on-disk state which can be used to | |
10603 | * determine whether or not the activity is in progress. The in-memory state and | |
10604 | * the locking used to protect it will be different for each activity, and may | |
10605 | * not be suitable for use with a cvar (e.g., some state is protected by the | |
10606 | * config lock). To allow waiting threads to wait without any races, another | |
10607 | * lock, spa_activities_lock, is used. | |
10608 | * | |
10609 | * When the state is checked, both the activity-specific lock (if there is one) | |
10610 | * and spa_activities_lock are held. In some cases, the activity-specific lock | |
10611 | * is acquired explicitly (e.g. the config lock). In others, the locking is | |
10612 | * internal to some check (e.g. bpobj_is_empty). After checking, the waiting | |
10613 | * thread releases the activity-specific lock and, if the activity is in | |
10614 | * progress, then cv_waits using spa_activities_lock. | |
10615 | * | |
10616 | * The waiting thread is woken when another thread, one completing some | |
10617 | * activity, updates the state of the activity and then calls | |
10618 | * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only | |
10619 | * needs to hold its activity-specific lock when updating the state, and this | |
10620 | * lock can (but doesn't have to) be dropped before calling spa_notify_waiters. | |
10621 | * | |
10622 | * Because spa_notify_waiters acquires spa_activities_lock before broadcasting, | |
10623 | * and because it is held when the waiting thread checks the state of the | |
10624 | * activity, it can never be the case that the completing thread both updates | |
10625 | * the activity state and cv_broadcasts in between the waiting thread's check | |
10626 | * and cv_wait. Thus, a waiting thread can never miss a wakeup. | |
10627 | * | |
10628 | * In order to prevent deadlock, when the waiting thread does its check, in some | |
10629 | * cases it will temporarily drop spa_activities_lock in order to acquire the | |
10630 | * activity-specific lock. The order in which spa_activities_lock and the | |
10631 | * activity specific lock are acquired in the waiting thread is determined by | |
10632 | * the order in which they are acquired in the completing thread; if the | |
10633 | * completing thread calls spa_notify_waiters with the activity-specific lock | |
10634 | * held, then the waiting thread must also acquire the activity-specific lock | |
10635 | * first. | |
10636 | */ | |
10637 | ||
10638 | static int | |
10639 | spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity, | |
10640 | boolean_t use_tag, uint64_t tag, boolean_t *in_progress) | |
10641 | { | |
10642 | int error = 0; | |
10643 | ||
10644 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
10645 | ||
10646 | switch (activity) { | |
10647 | case ZPOOL_WAIT_CKPT_DISCARD: | |
10648 | *in_progress = | |
10649 | (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) && | |
10650 | zap_contains(spa_meta_objset(spa), | |
10651 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) == | |
10652 | ENOENT); | |
10653 | break; | |
10654 | case ZPOOL_WAIT_FREE: | |
10655 | *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS && | |
10656 | !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) || | |
10657 | spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) || | |
10658 | spa_livelist_delete_check(spa)); | |
10659 | break; | |
10660 | case ZPOOL_WAIT_INITIALIZE: | |
2288d419 BB |
10661 | case ZPOOL_WAIT_TRIM: |
10662 | error = spa_vdev_activity_in_progress(spa, use_tag, tag, | |
10663 | activity, in_progress); | |
e60e158e JG |
10664 | break; |
10665 | case ZPOOL_WAIT_REPLACE: | |
10666 | mutex_exit(&spa->spa_activities_lock); | |
10667 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10668 | mutex_enter(&spa->spa_activities_lock); | |
10669 | ||
10670 | *in_progress = vdev_replace_in_progress(spa->spa_root_vdev); | |
10671 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10672 | break; | |
10673 | case ZPOOL_WAIT_REMOVE: | |
10674 | *in_progress = (spa->spa_removing_phys.sr_state == | |
10675 | DSS_SCANNING); | |
10676 | break; | |
10677 | case ZPOOL_WAIT_RESILVER: | |
5caeef02 DB |
10678 | *in_progress = vdev_rebuild_active(spa->spa_root_vdev); |
10679 | if (*in_progress) | |
9a49d3f3 | 10680 | break; |
9a70e97f | 10681 | zfs_fallthrough; |
e60e158e JG |
10682 | case ZPOOL_WAIT_SCRUB: |
10683 | { | |
10684 | boolean_t scanning, paused, is_scrub; | |
10685 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
10686 | ||
10687 | is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB); | |
10688 | scanning = (scn->scn_phys.scn_state == DSS_SCANNING); | |
10689 | paused = dsl_scan_is_paused_scrub(scn); | |
10690 | *in_progress = (scanning && !paused && | |
10691 | is_scrub == (activity == ZPOOL_WAIT_SCRUB)); | |
10692 | break; | |
10693 | } | |
5caeef02 DB |
10694 | case ZPOOL_WAIT_RAIDZ_EXPAND: |
10695 | { | |
10696 | vdev_raidz_expand_t *vre = spa->spa_raidz_expand; | |
10697 | *in_progress = (vre != NULL && vre->vre_state == DSS_SCANNING); | |
10698 | break; | |
10699 | } | |
e60e158e JG |
10700 | default: |
10701 | panic("unrecognized value for activity %d", activity); | |
10702 | } | |
10703 | ||
10704 | return (error); | |
10705 | } | |
10706 | ||
10707 | static int | |
10708 | spa_wait_common(const char *pool, zpool_wait_activity_t activity, | |
10709 | boolean_t use_tag, uint64_t tag, boolean_t *waited) | |
10710 | { | |
10711 | /* | |
10712 | * The tag is used to distinguish between instances of an activity. | |
2288d419 BB |
10713 | * 'initialize' and 'trim' are the only activities that we use this for. |
10714 | * The other activities can only have a single instance in progress in a | |
10715 | * pool at one time, making the tag unnecessary. | |
e60e158e JG |
10716 | * |
10717 | * There can be multiple devices being replaced at once, but since they | |
10718 | * all finish once resilvering finishes, we don't bother keeping track | |
10719 | * of them individually, we just wait for them all to finish. | |
10720 | */ | |
2288d419 BB |
10721 | if (use_tag && activity != ZPOOL_WAIT_INITIALIZE && |
10722 | activity != ZPOOL_WAIT_TRIM) | |
e60e158e JG |
10723 | return (EINVAL); |
10724 | ||
10725 | if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES) | |
10726 | return (EINVAL); | |
10727 | ||
10728 | spa_t *spa; | |
10729 | int error = spa_open(pool, &spa, FTAG); | |
10730 | if (error != 0) | |
10731 | return (error); | |
10732 | ||
10733 | /* | |
10734 | * Increment the spa's waiter count so that we can call spa_close and | |
10735 | * still ensure that the spa_t doesn't get freed before this thread is | |
10736 | * finished with it when the pool is exported. We want to call spa_close | |
10737 | * before we start waiting because otherwise the additional ref would | |
10738 | * prevent the pool from being exported or destroyed throughout the | |
10739 | * potentially long wait. | |
10740 | */ | |
10741 | mutex_enter(&spa->spa_activities_lock); | |
10742 | spa->spa_waiters++; | |
10743 | spa_close(spa, FTAG); | |
10744 | ||
10745 | *waited = B_FALSE; | |
10746 | for (;;) { | |
10747 | boolean_t in_progress; | |
10748 | error = spa_activity_in_progress(spa, activity, use_tag, tag, | |
10749 | &in_progress); | |
10750 | ||
b24771a8 | 10751 | if (error || !in_progress || spa->spa_waiters_cancel) |
e60e158e JG |
10752 | break; |
10753 | ||
10754 | *waited = B_TRUE; | |
10755 | ||
10756 | if (cv_wait_sig(&spa->spa_activities_cv, | |
10757 | &spa->spa_activities_lock) == 0) { | |
10758 | error = EINTR; | |
10759 | break; | |
10760 | } | |
10761 | } | |
10762 | ||
10763 | spa->spa_waiters--; | |
10764 | cv_signal(&spa->spa_waiters_cv); | |
10765 | mutex_exit(&spa->spa_activities_lock); | |
10766 | ||
10767 | return (error); | |
10768 | } | |
10769 | ||
10770 | /* | |
10771 | * Wait for a particular instance of the specified activity to complete, where | |
10772 | * the instance is identified by 'tag' | |
10773 | */ | |
10774 | int | |
10775 | spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag, | |
10776 | boolean_t *waited) | |
10777 | { | |
10778 | return (spa_wait_common(pool, activity, B_TRUE, tag, waited)); | |
10779 | } | |
10780 | ||
10781 | /* | |
10782 | * Wait for all instances of the specified activity complete | |
10783 | */ | |
10784 | int | |
10785 | spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited) | |
10786 | { | |
10787 | ||
10788 | return (spa_wait_common(pool, activity, B_FALSE, 0, waited)); | |
10789 | } | |
10790 | ||
a1d477c2 | 10791 | sysevent_t * |
12fa0466 DE |
10792 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
10793 | { | |
10794 | sysevent_t *ev = NULL; | |
10795 | #ifdef _KERNEL | |
10796 | nvlist_t *resource; | |
10797 | ||
10798 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
10799 | if (resource) { | |
10800 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
10801 | ev->resource = resource; | |
10802 | } | |
14e4e3cb AZ |
10803 | #else |
10804 | (void) spa, (void) vd, (void) hist_nvl, (void) name; | |
12fa0466 DE |
10805 | #endif |
10806 | return (ev); | |
10807 | } | |
10808 | ||
a1d477c2 | 10809 | void |
12fa0466 DE |
10810 | spa_event_post(sysevent_t *ev) |
10811 | { | |
10812 | #ifdef _KERNEL | |
10813 | if (ev) { | |
10814 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
10815 | kmem_free(ev, sizeof (*ev)); | |
10816 | } | |
14e4e3cb AZ |
10817 | #else |
10818 | (void) ev; | |
12fa0466 DE |
10819 | #endif |
10820 | } | |
10821 | ||
34dc7c2f | 10822 | /* |
fb390aaf HR |
10823 | * Post a zevent corresponding to the given sysevent. The 'name' must be one |
10824 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
34dc7c2f BB |
10825 | * filled in from the spa and (optionally) the vdev. This doesn't do anything |
10826 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
10827 | * or zdb as real changes. | |
10828 | */ | |
10829 | void | |
12fa0466 | 10830 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
34dc7c2f | 10831 | { |
12fa0466 | 10832 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); |
34dc7c2f | 10833 | } |
c28b2279 | 10834 | |
c28b2279 BB |
10835 | /* state manipulation functions */ |
10836 | EXPORT_SYMBOL(spa_open); | |
10837 | EXPORT_SYMBOL(spa_open_rewind); | |
10838 | EXPORT_SYMBOL(spa_get_stats); | |
10839 | EXPORT_SYMBOL(spa_create); | |
c28b2279 BB |
10840 | EXPORT_SYMBOL(spa_import); |
10841 | EXPORT_SYMBOL(spa_tryimport); | |
10842 | EXPORT_SYMBOL(spa_destroy); | |
10843 | EXPORT_SYMBOL(spa_export); | |
10844 | EXPORT_SYMBOL(spa_reset); | |
10845 | EXPORT_SYMBOL(spa_async_request); | |
10846 | EXPORT_SYMBOL(spa_async_suspend); | |
10847 | EXPORT_SYMBOL(spa_async_resume); | |
10848 | EXPORT_SYMBOL(spa_inject_addref); | |
10849 | EXPORT_SYMBOL(spa_inject_delref); | |
10850 | EXPORT_SYMBOL(spa_scan_stat_init); | |
10851 | EXPORT_SYMBOL(spa_scan_get_stats); | |
10852 | ||
e1cfd73f | 10853 | /* device manipulation */ |
c28b2279 BB |
10854 | EXPORT_SYMBOL(spa_vdev_add); |
10855 | EXPORT_SYMBOL(spa_vdev_attach); | |
10856 | EXPORT_SYMBOL(spa_vdev_detach); | |
c28b2279 BB |
10857 | EXPORT_SYMBOL(spa_vdev_setpath); |
10858 | EXPORT_SYMBOL(spa_vdev_setfru); | |
10859 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
10860 | ||
10861 | /* spare statech is global across all pools) */ | |
10862 | EXPORT_SYMBOL(spa_spare_add); | |
10863 | EXPORT_SYMBOL(spa_spare_remove); | |
10864 | EXPORT_SYMBOL(spa_spare_exists); | |
10865 | EXPORT_SYMBOL(spa_spare_activate); | |
10866 | ||
10867 | /* L2ARC statech is global across all pools) */ | |
10868 | EXPORT_SYMBOL(spa_l2cache_add); | |
10869 | EXPORT_SYMBOL(spa_l2cache_remove); | |
10870 | EXPORT_SYMBOL(spa_l2cache_exists); | |
10871 | EXPORT_SYMBOL(spa_l2cache_activate); | |
10872 | EXPORT_SYMBOL(spa_l2cache_drop); | |
10873 | ||
10874 | /* scanning */ | |
10875 | EXPORT_SYMBOL(spa_scan); | |
10876 | EXPORT_SYMBOL(spa_scan_stop); | |
10877 | ||
10878 | /* spa syncing */ | |
10879 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
10880 | EXPORT_SYMBOL(spa_sync_allpools); | |
10881 | ||
10882 | /* properties */ | |
10883 | EXPORT_SYMBOL(spa_prop_set); | |
10884 | EXPORT_SYMBOL(spa_prop_get); | |
10885 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
10886 | ||
10887 | /* asynchronous event notification */ | |
10888 | EXPORT_SYMBOL(spa_event_notify); | |
dea377c0 | 10889 | |
342357cd AM |
10890 | ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, preload_pct, UINT, ZMOD_RW, |
10891 | "Percentage of CPUs to run a metaslab preload taskq"); | |
10892 | ||
c8242a96 | 10893 | /* BEGIN CSTYLED */ |
fdc2d303 | 10894 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, UINT, ZMOD_RW, |
458f8231 | 10895 | "log2 fraction of arc that can be used by inflight I/Os when " |
03fdcb9a | 10896 | "verifying pool during import"); |
7ada752a | 10897 | /* END CSTYLED */ |
dea377c0 | 10898 | |
03fdcb9a | 10899 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW, |
dea377c0 MA |
10900 | "Set to traverse metadata on pool import"); |
10901 | ||
03fdcb9a | 10902 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW, |
dea377c0 | 10903 | "Set to traverse data on pool import"); |
dcb6bed1 | 10904 | |
03fdcb9a | 10905 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW, |
6cb8e530 PZ |
10906 | "Print vdev tree to zfs_dbgmsg during pool import"); |
10907 | ||
67d13998 | 10908 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RW, |
dcb6bed1 D |
10909 | "Percentage of CPUs to run an IO worker thread"); |
10910 | ||
67d13998 | 10911 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_tpq, UINT, ZMOD_RW, |
7457b024 AM |
10912 | "Number of threads per IO worker taskqueue"); |
10913 | ||
7ada752a | 10914 | /* BEGIN CSTYLED */ |
ab8d9c17 | 10915 | ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, U64, ZMOD_RW, |
03fdcb9a MM |
10916 | "Allow importing pool with up to this number of missing top-level " |
10917 | "vdevs (in read-only mode)"); | |
7ada752a | 10918 | /* END CSTYLED */ |
6cb8e530 | 10919 | |
7ada752a AZ |
10920 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, |
10921 | ZMOD_RW, "Set the livelist condense zthr to pause"); | |
03fdcb9a | 10922 | |
7ada752a AZ |
10923 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, |
10924 | ZMOD_RW, "Set the livelist condense synctask to pause"); | |
37f03da8 | 10925 | |
7ada752a AZ |
10926 | /* BEGIN CSTYLED */ |
10927 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, | |
10928 | INT, ZMOD_RW, | |
37f03da8 | 10929 | "Whether livelist condensing was canceled in the synctask"); |
03fdcb9a | 10930 | |
7ada752a AZ |
10931 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, |
10932 | INT, ZMOD_RW, | |
37f03da8 SH |
10933 | "Whether livelist condensing was canceled in the zthr function"); |
10934 | ||
7ada752a AZ |
10935 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, |
10936 | ZMOD_RW, | |
03fdcb9a MM |
10937 | "Whether extra ALLOC blkptrs were added to a livelist entry while it " |
10938 | "was being condensed"); | |
6930ecbb RN |
10939 | |
10940 | #ifdef _KERNEL | |
10941 | ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio, zio_, taskq_read, | |
67d13998 | 10942 | spa_taskq_read_param_set, spa_taskq_read_param_get, ZMOD_RW, |
6930ecbb RN |
10943 | "Configure IO queues for read IO"); |
10944 | ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio, zio_, taskq_write, | |
67d13998 | 10945 | spa_taskq_write_param_set, spa_taskq_write_param_get, ZMOD_RW, |
6930ecbb RN |
10946 | "Configure IO queues for write IO"); |
10947 | #endif | |
37f03da8 | 10948 | /* END CSTYLED */ |
3bd4df38 | 10949 | |
645b8330 AM |
10950 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_write_tpq, UINT, ZMOD_RW, |
10951 | "Number of CPUs per write issue taskq"); |