]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
10 | * See the License for the specific language governing permissions |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
4f072827 | 24 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
733b5722 | 25 | * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved. |
0c66c32d | 26 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
3c67d83a | 27 | * Copyright 2013 Saso Kiselkov. All rights reserved. |
e550644f BB |
28 | * Copyright (c) 2014 Integros [integros.com] |
29 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
a0bd735a | 30 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
f65fbee1 | 31 | * Copyright 2018 Joyent, Inc. |
3c819a2c | 32 | * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. |
12fa0466 | 33 | * Copyright 2017 Joyent, Inc. |
cc99f275 | 34 | * Copyright (c) 2017, Intel Corporation. |
658fb802 | 35 | * Copyright (c) 2021, Colm Buckley <colm@tuatha.org> |
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 | |
173 | * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that | |
174 | * need to be handled with minimum delay. | |
428870ff | 175 | */ |
18168da7 | 176 | static const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { |
428870ff | 177 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ |
7ef5e54e | 178 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ |
7457b024 | 179 | { ZTI_N(8), ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* READ */ |
3bd4df38 | 180 | { ZTI_SYNC, ZTI_N(5), ZTI_SCALE, ZTI_N(5) }, /* WRITE */ |
7457b024 | 181 | { ZTI_SCALE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ |
7ef5e54e AL |
182 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ |
183 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ | |
1b939560 | 184 | { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */ |
9babb374 BB |
185 | }; |
186 | ||
13fe0198 MA |
187 | static void spa_sync_version(void *arg, dmu_tx_t *tx); |
188 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
b128c09f | 189 | static boolean_t spa_has_active_shared_spare(spa_t *spa); |
a926aab9 AZ |
190 | static int spa_load_impl(spa_t *spa, spa_import_type_t type, |
191 | const char **ereport); | |
572e2857 | 192 | static void spa_vdev_resilver_done(spa_t *spa); |
428870ff | 193 | |
342357cd AM |
194 | /* |
195 | * Percentage of all CPUs that can be used by the metaslab preload taskq. | |
196 | */ | |
197 | static uint_t metaslab_preload_pct = 50; | |
198 | ||
18168da7 AZ |
199 | static uint_t zio_taskq_batch_pct = 80; /* 1 thread per cpu in pset */ |
200 | static uint_t zio_taskq_batch_tpq; /* threads per taskq */ | |
b5e60918 RN |
201 | |
202 | #ifdef HAVE_SYSDC | |
18168da7 AZ |
203 | static const boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ |
204 | static const uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
b5e60918 | 205 | #endif |
428870ff | 206 | |
b5e60918 | 207 | #ifdef HAVE_SPA_THREAD |
18168da7 | 208 | static const boolean_t spa_create_process = B_TRUE; /* no process => no sysdc */ |
b5e60918 | 209 | #endif |
428870ff | 210 | |
3bd4df38 EN |
211 | static uint_t zio_taskq_wr_iss_ncpus = 0; |
212 | ||
afd2f7b7 PZ |
213 | /* |
214 | * Report any spa_load_verify errors found, but do not fail spa_load. | |
215 | * This is used by zdb to analyze non-idle pools. | |
216 | */ | |
217 | boolean_t spa_load_verify_dryrun = B_FALSE; | |
218 | ||
e39fe05b FU |
219 | /* |
220 | * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ). | |
221 | * This is used by zdb for spacemaps verification. | |
222 | */ | |
223 | boolean_t spa_mode_readable_spacemaps = B_FALSE; | |
224 | ||
428870ff BB |
225 | /* |
226 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
227 | * to get the vdev stats associated with the imported devices. | |
228 | */ | |
229 | #define TRYIMPORT_NAME "$import" | |
34dc7c2f | 230 | |
6cb8e530 PZ |
231 | /* |
232 | * For debugging purposes: print out vdev tree during pool import. | |
233 | */ | |
18168da7 | 234 | static int spa_load_print_vdev_tree = B_FALSE; |
6cb8e530 PZ |
235 | |
236 | /* | |
237 | * A non-zero value for zfs_max_missing_tvds means that we allow importing | |
238 | * pools with missing top-level vdevs. This is strictly intended for advanced | |
239 | * pool recovery cases since missing data is almost inevitable. Pools with | |
240 | * missing devices can only be imported read-only for safety reasons, and their | |
241 | * fail-mode will be automatically set to "continue". | |
242 | * | |
243 | * With 1 missing vdev we should be able to import the pool and mount all | |
244 | * datasets. User data that was not modified after the missing device has been | |
245 | * added should be recoverable. This means that snapshots created prior to the | |
246 | * addition of that device should be completely intact. | |
247 | * | |
248 | * With 2 missing vdevs, some datasets may fail to mount since there are | |
249 | * dataset statistics that are stored as regular metadata. Some data might be | |
250 | * recoverable if those vdevs were added recently. | |
251 | * | |
252 | * With 3 or more missing vdevs, the pool is severely damaged and MOS entries | |
253 | * may be missing entirely. Chances of data recovery are very low. Note that | |
254 | * there are also risks of performing an inadvertent rewind as we might be | |
255 | * missing all the vdevs with the latest uberblocks. | |
256 | */ | |
ab8d9c17 | 257 | uint64_t zfs_max_missing_tvds = 0; |
6cb8e530 PZ |
258 | |
259 | /* | |
260 | * The parameters below are similar to zfs_max_missing_tvds but are only | |
261 | * intended for a preliminary open of the pool with an untrusted config which | |
262 | * might be incomplete or out-dated. | |
263 | * | |
264 | * We are more tolerant for pools opened from a cachefile since we could have | |
265 | * an out-dated cachefile where a device removal was not registered. | |
266 | * We could have set the limit arbitrarily high but in the case where devices | |
267 | * are really missing we would want to return the proper error codes; we chose | |
268 | * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available | |
269 | * and we get a chance to retrieve the trusted config. | |
270 | */ | |
271 | uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; | |
d2734cce | 272 | |
6cb8e530 PZ |
273 | /* |
274 | * In the case where config was assembled by scanning device paths (/dev/dsks | |
275 | * by default) we are less tolerant since all the existing devices should have | |
276 | * been detected and we want spa_load to return the right error codes. | |
277 | */ | |
278 | uint64_t zfs_max_missing_tvds_scan = 0; | |
279 | ||
d2734cce SD |
280 | /* |
281 | * Debugging aid that pauses spa_sync() towards the end. | |
282 | */ | |
18168da7 | 283 | static const boolean_t zfs_pause_spa_sync = B_FALSE; |
d2734cce | 284 | |
37f03da8 SH |
285 | /* |
286 | * Variables to indicate the livelist condense zthr func should wait at certain | |
287 | * points for the livelist to be removed - used to test condense/destroy races | |
288 | */ | |
18168da7 AZ |
289 | static int zfs_livelist_condense_zthr_pause = 0; |
290 | static int zfs_livelist_condense_sync_pause = 0; | |
37f03da8 SH |
291 | |
292 | /* | |
293 | * Variables to track whether or not condense cancellation has been | |
294 | * triggered in testing. | |
295 | */ | |
18168da7 AZ |
296 | static int zfs_livelist_condense_sync_cancel = 0; |
297 | static int zfs_livelist_condense_zthr_cancel = 0; | |
37f03da8 SH |
298 | |
299 | /* | |
300 | * Variable to track whether or not extra ALLOC blkptrs were added to a | |
301 | * livelist entry while it was being condensed (caused by the way we track | |
302 | * remapped blkptrs in dbuf_remap_impl) | |
303 | */ | |
18168da7 | 304 | static int zfs_livelist_condense_new_alloc = 0; |
37f03da8 | 305 | |
34dc7c2f BB |
306 | /* |
307 | * ========================================================================== | |
308 | * SPA properties routines | |
309 | * ========================================================================== | |
310 | */ | |
311 | ||
312 | /* | |
313 | * Add a (source=src, propname=propval) list to an nvlist. | |
314 | */ | |
315 | static void | |
a926aab9 | 316 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, const char *strval, |
34dc7c2f BB |
317 | uint64_t intval, zprop_source_t src) |
318 | { | |
319 | const char *propname = zpool_prop_to_name(prop); | |
320 | nvlist_t *propval; | |
321 | ||
65ad5d11 AJ |
322 | propval = fnvlist_alloc(); |
323 | fnvlist_add_uint64(propval, ZPROP_SOURCE, src); | |
34dc7c2f BB |
324 | |
325 | if (strval != NULL) | |
65ad5d11 | 326 | fnvlist_add_string(propval, ZPROP_VALUE, strval); |
34dc7c2f | 327 | else |
65ad5d11 | 328 | fnvlist_add_uint64(propval, ZPROP_VALUE, intval); |
34dc7c2f | 329 | |
65ad5d11 | 330 | fnvlist_add_nvlist(nvl, propname, propval); |
34dc7c2f BB |
331 | nvlist_free(propval); |
332 | } | |
333 | ||
8eae2d21 AJ |
334 | /* |
335 | * Add a user property (source=src, propname=propval) to an nvlist. | |
336 | */ | |
337 | static void | |
338 | spa_prop_add_user(nvlist_t *nvl, const char *propname, char *strval, | |
339 | zprop_source_t src) | |
340 | { | |
341 | nvlist_t *propval; | |
342 | ||
343 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
344 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); | |
345 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
346 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
347 | nvlist_free(propval); | |
348 | } | |
349 | ||
34dc7c2f BB |
350 | /* |
351 | * Get property values from the spa configuration. | |
352 | */ | |
353 | static void | |
354 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
355 | { | |
1bd201e7 | 356 | vdev_t *rvd = spa->spa_root_vdev; |
9ae529ec | 357 | dsl_pool_t *pool = spa->spa_dsl_pool; |
f3a7f661 | 358 | uint64_t size, alloc, cap, version; |
82ab6848 | 359 | const zprop_source_t src = ZPROP_SRC_NONE; |
b128c09f | 360 | spa_config_dirent_t *dp; |
f3a7f661 | 361 | metaslab_class_t *mc = spa_normal_class(spa); |
b128c09f BB |
362 | |
363 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
34dc7c2f | 364 | |
1bd201e7 | 365 | if (rvd != NULL) { |
cc99f275 DB |
366 | alloc = metaslab_class_get_alloc(mc); |
367 | alloc += metaslab_class_get_alloc(spa_special_class(spa)); | |
368 | alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); | |
aa755b35 | 369 | alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa)); |
cc99f275 DB |
370 | |
371 | size = metaslab_class_get_space(mc); | |
372 | size += metaslab_class_get_space(spa_special_class(spa)); | |
373 | size += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 | 374 | size += metaslab_class_get_space(spa_embedded_log_class(spa)); |
cc99f275 | 375 | |
d164b209 BB |
376 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); |
377 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
428870ff BB |
378 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); |
379 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
380 | size - alloc, src); | |
d2734cce SD |
381 | spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, |
382 | spa->spa_checkpoint_info.sci_dspace, src); | |
1bd201e7 | 383 | |
f3a7f661 GW |
384 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, |
385 | metaslab_class_fragmentation(mc), src); | |
386 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
387 | metaslab_class_expandable_space(mc), src); | |
572e2857 | 388 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, |
da92d5cb | 389 | (spa_mode(spa) == SPA_MODE_READ), src); |
d164b209 | 390 | |
428870ff | 391 | cap = (size == 0) ? 0 : (alloc * 100 / size); |
d164b209 BB |
392 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); |
393 | ||
428870ff BB |
394 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, |
395 | ddt_get_pool_dedup_ratio(spa), src); | |
67a1b037 PJD |
396 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONEUSED, NULL, |
397 | brt_get_used(spa), src); | |
398 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONESAVED, NULL, | |
399 | brt_get_saved(spa), src); | |
400 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONERATIO, NULL, | |
401 | brt_get_ratio(spa), src); | |
428870ff | 402 | |
d164b209 | 403 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, |
1bd201e7 | 404 | rvd->vdev_state, src); |
d164b209 BB |
405 | |
406 | version = spa_version(spa); | |
82ab6848 HM |
407 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { |
408 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
409 | version, ZPROP_SRC_DEFAULT); | |
410 | } else { | |
411 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
412 | version, ZPROP_SRC_LOCAL); | |
413 | } | |
a448a255 SD |
414 | spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID, |
415 | NULL, spa_load_guid(spa), src); | |
d164b209 | 416 | } |
34dc7c2f | 417 | |
9ae529ec | 418 | if (pool != NULL) { |
9ae529ec CS |
419 | /* |
420 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
421 | * when opening pools before this version freedir will be NULL. | |
422 | */ | |
fbeddd60 | 423 | if (pool->dp_free_dir != NULL) { |
9ae529ec | 424 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, |
d683ddbb JG |
425 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, |
426 | src); | |
9ae529ec CS |
427 | } else { |
428 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
429 | NULL, 0, src); | |
430 | } | |
fbeddd60 MA |
431 | |
432 | if (pool->dp_leak_dir != NULL) { | |
433 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
d683ddbb JG |
434 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, |
435 | src); | |
fbeddd60 MA |
436 | } else { |
437 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
438 | NULL, 0, src); | |
439 | } | |
9ae529ec CS |
440 | } |
441 | ||
34dc7c2f | 442 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); |
34dc7c2f | 443 | |
d96eb2b1 DM |
444 | if (spa->spa_comment != NULL) { |
445 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
446 | 0, ZPROP_SRC_LOCAL); | |
447 | } | |
448 | ||
658fb802 CB |
449 | if (spa->spa_compatibility != NULL) { |
450 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMPATIBILITY, | |
451 | spa->spa_compatibility, 0, ZPROP_SRC_LOCAL); | |
452 | } | |
453 | ||
34dc7c2f BB |
454 | if (spa->spa_root != NULL) |
455 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
456 | 0, ZPROP_SRC_LOCAL); | |
457 | ||
f1512ee6 MA |
458 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { |
459 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
460 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
461 | } else { | |
462 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
463 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
464 | } | |
465 | ||
50c957f7 NB |
466 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { |
467 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
468 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
469 | } else { | |
470 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
471 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
472 | } | |
473 | ||
b128c09f BB |
474 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { |
475 | if (dp->scd_path == NULL) { | |
34dc7c2f | 476 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f BB |
477 | "none", 0, ZPROP_SRC_LOCAL); |
478 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
34dc7c2f | 479 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f | 480 | dp->scd_path, 0, ZPROP_SRC_LOCAL); |
34dc7c2f BB |
481 | } |
482 | } | |
483 | } | |
484 | ||
485 | /* | |
486 | * Get zpool property values. | |
487 | */ | |
488 | int | |
489 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
490 | { | |
428870ff | 491 | objset_t *mos = spa->spa_meta_objset; |
34dc7c2f BB |
492 | zap_cursor_t zc; |
493 | zap_attribute_t za; | |
1743c737 | 494 | dsl_pool_t *dp; |
34dc7c2f BB |
495 | int err; |
496 | ||
79c76d5b | 497 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); |
c28b2279 | 498 | if (err) |
d1d7e268 | 499 | return (err); |
34dc7c2f | 500 | |
1743c737 AM |
501 | dp = spa_get_dsl(spa); |
502 | dsl_pool_config_enter(dp, FTAG); | |
b128c09f BB |
503 | mutex_enter(&spa->spa_props_lock); |
504 | ||
34dc7c2f BB |
505 | /* |
506 | * Get properties from the spa config. | |
507 | */ | |
508 | spa_prop_get_config(spa, nvp); | |
509 | ||
34dc7c2f | 510 | /* If no pool property object, no more prop to get. */ |
1743c737 | 511 | if (mos == NULL || spa->spa_pool_props_object == 0) |
c28b2279 | 512 | goto out; |
34dc7c2f BB |
513 | |
514 | /* | |
515 | * Get properties from the MOS pool property object. | |
516 | */ | |
517 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
518 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
519 | zap_cursor_advance(&zc)) { | |
520 | uint64_t intval = 0; | |
521 | char *strval = NULL; | |
522 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
523 | zpool_prop_t prop; | |
524 | ||
8eae2d21 AJ |
525 | if ((prop = zpool_name_to_prop(za.za_name)) == |
526 | ZPOOL_PROP_INVAL && !zfs_prop_user(za.za_name)) | |
34dc7c2f BB |
527 | continue; |
528 | ||
529 | switch (za.za_integer_length) { | |
530 | case 8: | |
531 | /* integer property */ | |
532 | if (za.za_first_integer != | |
533 | zpool_prop_default_numeric(prop)) | |
534 | src = ZPROP_SRC_LOCAL; | |
535 | ||
536 | if (prop == ZPOOL_PROP_BOOTFS) { | |
34dc7c2f BB |
537 | dsl_dataset_t *ds = NULL; |
538 | ||
619f0976 GW |
539 | err = dsl_dataset_hold_obj(dp, |
540 | za.za_first_integer, FTAG, &ds); | |
1743c737 | 541 | if (err != 0) |
34dc7c2f | 542 | break; |
34dc7c2f | 543 | |
eca7b760 | 544 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, |
79c76d5b | 545 | KM_SLEEP); |
34dc7c2f | 546 | dsl_dataset_name(ds, strval); |
b128c09f | 547 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
548 | } else { |
549 | strval = NULL; | |
550 | intval = za.za_first_integer; | |
551 | } | |
552 | ||
553 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
554 | ||
555 | if (strval != NULL) | |
eca7b760 | 556 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
557 | |
558 | break; | |
559 | ||
560 | case 1: | |
561 | /* string property */ | |
79c76d5b | 562 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); |
34dc7c2f BB |
563 | err = zap_lookup(mos, spa->spa_pool_props_object, |
564 | za.za_name, 1, za.za_num_integers, strval); | |
565 | if (err) { | |
566 | kmem_free(strval, za.za_num_integers); | |
567 | break; | |
568 | } | |
8eae2d21 AJ |
569 | if (prop != ZPOOL_PROP_INVAL) { |
570 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
571 | } else { | |
572 | src = ZPROP_SRC_LOCAL; | |
573 | spa_prop_add_user(*nvp, za.za_name, strval, | |
574 | src); | |
575 | } | |
34dc7c2f BB |
576 | kmem_free(strval, za.za_num_integers); |
577 | break; | |
578 | ||
579 | default: | |
580 | break; | |
581 | } | |
582 | } | |
583 | zap_cursor_fini(&zc); | |
34dc7c2f | 584 | out: |
1743c737 AM |
585 | mutex_exit(&spa->spa_props_lock); |
586 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
587 | if (err && err != ENOENT) { |
588 | nvlist_free(*nvp); | |
589 | *nvp = NULL; | |
590 | return (err); | |
591 | } | |
592 | ||
593 | return (0); | |
594 | } | |
595 | ||
596 | /* | |
597 | * Validate the given pool properties nvlist and modify the list | |
598 | * for the property values to be set. | |
599 | */ | |
600 | static int | |
601 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
602 | { | |
603 | nvpair_t *elem; | |
604 | int error = 0, reset_bootfs = 0; | |
d4ed6673 | 605 | uint64_t objnum = 0; |
9ae529ec | 606 | boolean_t has_feature = B_FALSE; |
34dc7c2f BB |
607 | |
608 | elem = NULL; | |
609 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
34dc7c2f | 610 | uint64_t intval; |
d1807f16 | 611 | const char *strval, *slash, *check, *fname; |
9ae529ec CS |
612 | const char *propname = nvpair_name(elem); |
613 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
614 | ||
31864e3d BB |
615 | switch (prop) { |
616 | case ZPOOL_PROP_INVAL: | |
9ae529ec CS |
617 | /* |
618 | * Sanitize the input. | |
619 | */ | |
8eae2d21 AJ |
620 | if (zfs_prop_user(propname)) { |
621 | if (strlen(propname) >= ZAP_MAXNAMELEN) { | |
622 | error = SET_ERROR(ENAMETOOLONG); | |
623 | break; | |
624 | } | |
9ae529ec | 625 | |
8eae2d21 AJ |
626 | if (strlen(fnvpair_value_string(elem)) >= |
627 | ZAP_MAXVALUELEN) { | |
628 | error = SET_ERROR(E2BIG); | |
629 | break; | |
630 | } | |
631 | } else if (zpool_prop_feature(propname)) { | |
632 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
633 | error = SET_ERROR(EINVAL); | |
634 | break; | |
635 | } | |
34dc7c2f | 636 | |
8eae2d21 AJ |
637 | if (nvpair_value_uint64(elem, &intval) != 0) { |
638 | error = SET_ERROR(EINVAL); | |
639 | break; | |
640 | } | |
641 | ||
642 | if (intval != 0) { | |
643 | error = SET_ERROR(EINVAL); | |
644 | break; | |
645 | } | |
646 | ||
647 | fname = strchr(propname, '@') + 1; | |
648 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
649 | error = SET_ERROR(EINVAL); | |
650 | break; | |
651 | } | |
34dc7c2f | 652 | |
8eae2d21 AJ |
653 | has_feature = B_TRUE; |
654 | } else { | |
2e528b49 | 655 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
656 | break; |
657 | } | |
9ae529ec | 658 | break; |
34dc7c2f | 659 | |
34dc7c2f BB |
660 | case ZPOOL_PROP_VERSION: |
661 | error = nvpair_value_uint64(elem, &intval); | |
662 | if (!error && | |
9ae529ec CS |
663 | (intval < spa_version(spa) || |
664 | intval > SPA_VERSION_BEFORE_FEATURES || | |
665 | has_feature)) | |
2e528b49 | 666 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
667 | break; |
668 | ||
669 | case ZPOOL_PROP_DELEGATION: | |
670 | case ZPOOL_PROP_AUTOREPLACE: | |
b128c09f | 671 | case ZPOOL_PROP_LISTSNAPS: |
9babb374 | 672 | case ZPOOL_PROP_AUTOEXPAND: |
1b939560 | 673 | case ZPOOL_PROP_AUTOTRIM: |
34dc7c2f BB |
674 | error = nvpair_value_uint64(elem, &intval); |
675 | if (!error && intval > 1) | |
2e528b49 | 676 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
677 | break; |
678 | ||
379ca9cf OF |
679 | case ZPOOL_PROP_MULTIHOST: |
680 | error = nvpair_value_uint64(elem, &intval); | |
681 | if (!error && intval > 1) | |
682 | error = SET_ERROR(EINVAL); | |
683 | ||
25f06d67 BB |
684 | if (!error) { |
685 | uint32_t hostid = zone_get_hostid(NULL); | |
686 | if (hostid) | |
687 | spa->spa_hostid = hostid; | |
688 | else | |
689 | error = SET_ERROR(ENOTSUP); | |
690 | } | |
379ca9cf OF |
691 | |
692 | break; | |
693 | ||
34dc7c2f | 694 | case ZPOOL_PROP_BOOTFS: |
9babb374 BB |
695 | /* |
696 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
697 | * or the pool is still being created (version == 0), | |
698 | * the bootfs property cannot be set. | |
699 | */ | |
34dc7c2f | 700 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { |
2e528b49 | 701 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
702 | break; |
703 | } | |
704 | ||
705 | /* | |
b128c09f | 706 | * Make sure the vdev config is bootable |
34dc7c2f | 707 | */ |
b128c09f | 708 | if (!vdev_is_bootable(spa->spa_root_vdev)) { |
2e528b49 | 709 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
710 | break; |
711 | } | |
712 | ||
713 | reset_bootfs = 1; | |
714 | ||
715 | error = nvpair_value_string(elem, &strval); | |
716 | ||
717 | if (!error) { | |
9ae529ec | 718 | objset_t *os; |
b128c09f | 719 | |
34dc7c2f BB |
720 | if (strval == NULL || strval[0] == '\0') { |
721 | objnum = zpool_prop_default_numeric( | |
722 | ZPOOL_PROP_BOOTFS); | |
723 | break; | |
724 | } | |
725 | ||
d1d7e268 | 726 | error = dmu_objset_hold(strval, FTAG, &os); |
619f0976 | 727 | if (error != 0) |
34dc7c2f | 728 | break; |
b128c09f | 729 | |
eaa25f1a | 730 | /* Must be ZPL. */ |
428870ff | 731 | if (dmu_objset_type(os) != DMU_OST_ZFS) { |
2e528b49 | 732 | error = SET_ERROR(ENOTSUP); |
b128c09f BB |
733 | } else { |
734 | objnum = dmu_objset_id(os); | |
735 | } | |
428870ff | 736 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
737 | } |
738 | break; | |
b128c09f | 739 | |
34dc7c2f BB |
740 | case ZPOOL_PROP_FAILUREMODE: |
741 | error = nvpair_value_uint64(elem, &intval); | |
3bfd95d5 | 742 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) |
2e528b49 | 743 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
744 | |
745 | /* | |
746 | * This is a special case which only occurs when | |
747 | * the pool has completely failed. This allows | |
748 | * the user to change the in-core failmode property | |
749 | * without syncing it out to disk (I/Os might | |
750 | * currently be blocked). We do this by returning | |
751 | * EIO to the caller (spa_prop_set) to trick it | |
752 | * into thinking we encountered a property validation | |
753 | * error. | |
754 | */ | |
b128c09f | 755 | if (!error && spa_suspended(spa)) { |
34dc7c2f | 756 | spa->spa_failmode = intval; |
2e528b49 | 757 | error = SET_ERROR(EIO); |
34dc7c2f BB |
758 | } |
759 | break; | |
760 | ||
761 | case ZPOOL_PROP_CACHEFILE: | |
762 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
763 | break; | |
764 | ||
765 | if (strval[0] == '\0') | |
766 | break; | |
767 | ||
768 | if (strcmp(strval, "none") == 0) | |
769 | break; | |
770 | ||
771 | if (strval[0] != '/') { | |
2e528b49 | 772 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
773 | break; |
774 | } | |
775 | ||
776 | slash = strrchr(strval, '/'); | |
777 | ASSERT(slash != NULL); | |
778 | ||
779 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
780 | strcmp(slash, "/..") == 0) | |
2e528b49 | 781 | error = SET_ERROR(EINVAL); |
34dc7c2f | 782 | break; |
428870ff | 783 | |
d96eb2b1 DM |
784 | case ZPOOL_PROP_COMMENT: |
785 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
786 | break; | |
787 | for (check = strval; *check != '\0'; check++) { | |
788 | if (!isprint(*check)) { | |
2e528b49 | 789 | error = SET_ERROR(EINVAL); |
d96eb2b1 DM |
790 | break; |
791 | } | |
d96eb2b1 DM |
792 | } |
793 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
2e528b49 | 794 | error = SET_ERROR(E2BIG); |
d96eb2b1 DM |
795 | break; |
796 | ||
e75c13c3 BB |
797 | default: |
798 | break; | |
34dc7c2f BB |
799 | } |
800 | ||
801 | if (error) | |
802 | break; | |
803 | } | |
804 | ||
050d720c MA |
805 | (void) nvlist_remove_all(props, |
806 | zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO)); | |
807 | ||
34dc7c2f BB |
808 | if (!error && reset_bootfs) { |
809 | error = nvlist_remove(props, | |
810 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
811 | ||
812 | if (!error) { | |
813 | error = nvlist_add_uint64(props, | |
814 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
815 | } | |
816 | } | |
817 | ||
818 | return (error); | |
819 | } | |
820 | ||
d164b209 BB |
821 | void |
822 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
823 | { | |
d1807f16 | 824 | const char *cachefile; |
d164b209 BB |
825 | spa_config_dirent_t *dp; |
826 | ||
827 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
828 | &cachefile) != 0) | |
829 | return; | |
830 | ||
831 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
79c76d5b | 832 | KM_SLEEP); |
d164b209 BB |
833 | |
834 | if (cachefile[0] == '\0') | |
835 | dp->scd_path = spa_strdup(spa_config_path); | |
836 | else if (strcmp(cachefile, "none") == 0) | |
837 | dp->scd_path = NULL; | |
838 | else | |
839 | dp->scd_path = spa_strdup(cachefile); | |
840 | ||
841 | list_insert_head(&spa->spa_config_list, dp); | |
842 | if (need_sync) | |
843 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
844 | } | |
845 | ||
34dc7c2f BB |
846 | int |
847 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
848 | { | |
849 | int error; | |
9ae529ec | 850 | nvpair_t *elem = NULL; |
d164b209 | 851 | boolean_t need_sync = B_FALSE; |
34dc7c2f BB |
852 | |
853 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
854 | return (error); | |
855 | ||
d164b209 | 856 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { |
9ae529ec | 857 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); |
d164b209 | 858 | |
572e2857 BB |
859 | if (prop == ZPOOL_PROP_CACHEFILE || |
860 | prop == ZPOOL_PROP_ALTROOT || | |
861 | prop == ZPOOL_PROP_READONLY) | |
d164b209 BB |
862 | continue; |
863 | ||
8eae2d21 AJ |
864 | if (prop == ZPOOL_PROP_INVAL && |
865 | zfs_prop_user(nvpair_name(elem))) { | |
866 | need_sync = B_TRUE; | |
867 | break; | |
868 | } | |
869 | ||
31864e3d | 870 | if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { |
2a673e76 | 871 | uint64_t ver = 0; |
9ae529ec CS |
872 | |
873 | if (prop == ZPOOL_PROP_VERSION) { | |
874 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
875 | } else { | |
876 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
877 | ver = SPA_VERSION_FEATURES; | |
878 | need_sync = B_TRUE; | |
879 | } | |
880 | ||
881 | /* Save time if the version is already set. */ | |
882 | if (ver == spa_version(spa)) | |
883 | continue; | |
884 | ||
885 | /* | |
886 | * In addition to the pool directory object, we might | |
887 | * create the pool properties object, the features for | |
888 | * read object, the features for write object, or the | |
889 | * feature descriptions object. | |
890 | */ | |
13fe0198 | 891 | error = dsl_sync_task(spa->spa_name, NULL, |
3d45fdd6 MA |
892 | spa_sync_version, &ver, |
893 | 6, ZFS_SPACE_CHECK_RESERVED); | |
9ae529ec CS |
894 | if (error) |
895 | return (error); | |
896 | continue; | |
897 | } | |
898 | ||
d164b209 BB |
899 | need_sync = B_TRUE; |
900 | break; | |
901 | } | |
902 | ||
9ae529ec | 903 | if (need_sync) { |
13fe0198 | 904 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, |
3d45fdd6 | 905 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); |
9ae529ec CS |
906 | } |
907 | ||
908 | return (0); | |
34dc7c2f BB |
909 | } |
910 | ||
911 | /* | |
912 | * If the bootfs property value is dsobj, clear it. | |
913 | */ | |
914 | void | |
915 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
916 | { | |
917 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
918 | VERIFY(zap_remove(spa->spa_meta_objset, | |
919 | spa->spa_pool_props_object, | |
920 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
921 | spa->spa_bootfs = 0; | |
922 | } | |
923 | } | |
924 | ||
3bc7e0fb | 925 | static int |
13fe0198 | 926 | spa_change_guid_check(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 927 | { |
2a8ba608 | 928 | uint64_t *newguid __maybe_unused = arg; |
13fe0198 | 929 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; |
3bc7e0fb GW |
930 | vdev_t *rvd = spa->spa_root_vdev; |
931 | uint64_t vdev_state; | |
3bc7e0fb | 932 | |
d2734cce SD |
933 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { |
934 | int error = (spa_has_checkpoint(spa)) ? | |
935 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
936 | return (SET_ERROR(error)); | |
937 | } | |
938 | ||
3bc7e0fb GW |
939 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
940 | vdev_state = rvd->vdev_state; | |
941 | spa_config_exit(spa, SCL_STATE, FTAG); | |
942 | ||
943 | if (vdev_state != VDEV_STATE_HEALTHY) | |
2e528b49 | 944 | return (SET_ERROR(ENXIO)); |
3bc7e0fb GW |
945 | |
946 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
947 | ||
948 | return (0); | |
949 | } | |
950 | ||
951 | static void | |
13fe0198 | 952 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 953 | { |
13fe0198 MA |
954 | uint64_t *newguid = arg; |
955 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
3bc7e0fb GW |
956 | uint64_t oldguid; |
957 | vdev_t *rvd = spa->spa_root_vdev; | |
958 | ||
959 | oldguid = spa_guid(spa); | |
960 | ||
961 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
962 | rvd->vdev_guid = *newguid; | |
963 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
964 | vdev_config_dirty(rvd); | |
965 | spa_config_exit(spa, SCL_STATE, FTAG); | |
966 | ||
6f1ffb06 | 967 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", |
74756182 | 968 | (u_longlong_t)oldguid, (u_longlong_t)*newguid); |
3bc7e0fb GW |
969 | } |
970 | ||
3541dc6d GA |
971 | /* |
972 | * Change the GUID for the pool. This is done so that we can later | |
973 | * re-import a pool built from a clone of our own vdevs. We will modify | |
974 | * the root vdev's guid, our own pool guid, and then mark all of our | |
975 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
976 | * online when we do this, or else any vdevs that weren't present | |
977 | * would be orphaned from our pool. We are also going to issue a | |
978 | * sysevent to update any watchers. | |
979 | */ | |
980 | int | |
981 | spa_change_guid(spa_t *spa) | |
982 | { | |
3bc7e0fb GW |
983 | int error; |
984 | uint64_t guid; | |
3541dc6d | 985 | |
621dd7bb | 986 | mutex_enter(&spa->spa_vdev_top_lock); |
3bc7e0fb GW |
987 | mutex_enter(&spa_namespace_lock); |
988 | guid = spa_generate_guid(NULL); | |
3541dc6d | 989 | |
13fe0198 | 990 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, |
3d45fdd6 | 991 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); |
3541dc6d | 992 | |
3bc7e0fb | 993 | if (error == 0) { |
55c12724 AH |
994 | /* |
995 | * Clear the kobj flag from all the vdevs to allow | |
996 | * vdev_cache_process_kobj_evt() to post events to all the | |
997 | * vdevs since GUID is updated. | |
998 | */ | |
999 | vdev_clear_kobj_evt(spa->spa_root_vdev); | |
1000 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
1001 | vdev_clear_kobj_evt(spa->spa_l2cache.sav_vdevs[i]); | |
1002 | ||
1003 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); | |
12fa0466 | 1004 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); |
3bc7e0fb | 1005 | } |
3541dc6d | 1006 | |
3bc7e0fb | 1007 | mutex_exit(&spa_namespace_lock); |
621dd7bb | 1008 | mutex_exit(&spa->spa_vdev_top_lock); |
3541dc6d | 1009 | |
3bc7e0fb | 1010 | return (error); |
3541dc6d GA |
1011 | } |
1012 | ||
34dc7c2f BB |
1013 | /* |
1014 | * ========================================================================== | |
1015 | * SPA state manipulation (open/create/destroy/import/export) | |
1016 | * ========================================================================== | |
1017 | */ | |
1018 | ||
1019 | static int | |
1020 | spa_error_entry_compare(const void *a, const void *b) | |
1021 | { | |
ee36c709 GN |
1022 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; |
1023 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
34dc7c2f BB |
1024 | int ret; |
1025 | ||
ee36c709 | 1026 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, |
5dbd68a3 | 1027 | sizeof (zbookmark_phys_t)); |
34dc7c2f | 1028 | |
ca577779 | 1029 | return (TREE_ISIGN(ret)); |
34dc7c2f BB |
1030 | } |
1031 | ||
1032 | /* | |
1033 | * Utility function which retrieves copies of the current logs and | |
1034 | * re-initializes them in the process. | |
1035 | */ | |
1036 | void | |
1037 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
1038 | { | |
1039 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
1040 | ||
861166b0 AZ |
1041 | memcpy(last, &spa->spa_errlist_last, sizeof (avl_tree_t)); |
1042 | memcpy(scrub, &spa->spa_errlist_scrub, sizeof (avl_tree_t)); | |
34dc7c2f BB |
1043 | |
1044 | avl_create(&spa->spa_errlist_scrub, | |
1045 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1046 | offsetof(spa_error_entry_t, se_avl)); | |
1047 | avl_create(&spa->spa_errlist_last, | |
1048 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1049 | offsetof(spa_error_entry_t, se_avl)); | |
1050 | } | |
1051 | ||
7ef5e54e AL |
1052 | static void |
1053 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
34dc7c2f | 1054 | { |
7ef5e54e AL |
1055 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; |
1056 | enum zti_modes mode = ztip->zti_mode; | |
1057 | uint_t value = ztip->zti_value; | |
1058 | uint_t count = ztip->zti_count; | |
1059 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7457b024 | 1060 | uint_t cpus, flags = TASKQ_DYNAMIC; |
34dc7c2f | 1061 | |
e8b96c60 MA |
1062 | switch (mode) { |
1063 | case ZTI_MODE_FIXED: | |
7457b024 | 1064 | ASSERT3U(value, >, 0); |
e8b96c60 | 1065 | break; |
7ef5e54e | 1066 | |
3bd4df38 EN |
1067 | case ZTI_MODE_SYNC: |
1068 | ||
1069 | /* | |
1070 | * Create one wr_iss taskq for every 'zio_taskq_wr_iss_ncpus', | |
1071 | * not to exceed the number of spa allocators. | |
1072 | */ | |
1073 | if (zio_taskq_wr_iss_ncpus == 0) { | |
1074 | count = MAX(boot_ncpus / spa->spa_alloc_count, 1); | |
1075 | } else { | |
1076 | count = MAX(1, | |
1077 | boot_ncpus / MAX(1, zio_taskq_wr_iss_ncpus)); | |
1078 | } | |
1079 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1080 | count = MIN(count, spa->spa_alloc_count); | |
1081 | ||
1082 | /* | |
1083 | * zio_taskq_batch_pct is unbounded and may exceed 100%, but no | |
1084 | * single taskq may have more threads than 100% of online cpus. | |
1085 | */ | |
1086 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1087 | value = MIN(value, 100); | |
e8b96c60 | 1088 | flags |= TASKQ_THREADS_CPU_PCT; |
e8b96c60 | 1089 | break; |
7ef5e54e | 1090 | |
7457b024 AM |
1091 | case ZTI_MODE_SCALE: |
1092 | flags |= TASKQ_THREADS_CPU_PCT; | |
1093 | /* | |
1094 | * We want more taskqs to reduce lock contention, but we want | |
1095 | * less for better request ordering and CPU utilization. | |
1096 | */ | |
1097 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); | |
1098 | if (zio_taskq_batch_tpq > 0) { | |
1099 | count = MAX(1, (cpus + zio_taskq_batch_tpq / 2) / | |
1100 | zio_taskq_batch_tpq); | |
1101 | } else { | |
1102 | /* | |
1103 | * Prefer 6 threads per taskq, but no more taskqs | |
1104 | * than threads in them on large systems. For 80%: | |
1105 | * | |
1106 | * taskq taskq total | |
1107 | * cpus taskqs percent threads threads | |
1108 | * ------- ------- ------- ------- ------- | |
1109 | * 1 1 80% 1 1 | |
1110 | * 2 1 80% 1 1 | |
1111 | * 4 1 80% 3 3 | |
1112 | * 8 2 40% 3 6 | |
1113 | * 16 3 27% 4 12 | |
1114 | * 32 5 16% 5 25 | |
1115 | * 64 7 11% 7 49 | |
1116 | * 128 10 8% 10 100 | |
1117 | * 256 14 6% 15 210 | |
1118 | */ | |
1119 | count = 1 + cpus / 6; | |
1120 | while (count * count > cpus) | |
1121 | count--; | |
1122 | } | |
1123 | /* Limit each taskq within 100% to not trigger assertion. */ | |
1124 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1125 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1126 | break; | |
1127 | ||
1128 | case ZTI_MODE_NULL: | |
1129 | tqs->stqs_count = 0; | |
1130 | tqs->stqs_taskq = NULL; | |
1131 | return; | |
1132 | ||
e8b96c60 MA |
1133 | default: |
1134 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
3bd4df38 | 1135 | "spa_taskqs_init()", |
e8b96c60 MA |
1136 | zio_type_name[t], zio_taskq_types[q], mode, value); |
1137 | break; | |
1138 | } | |
7ef5e54e | 1139 | |
7457b024 AM |
1140 | ASSERT3U(count, >, 0); |
1141 | tqs->stqs_count = count; | |
1142 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
1143 | ||
1c27024e | 1144 | for (uint_t i = 0; i < count; i++) { |
e8b96c60 | 1145 | taskq_t *tq; |
af430294 | 1146 | char name[32]; |
7ef5e54e | 1147 | |
7457b024 AM |
1148 | if (count > 1) |
1149 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
1150 | zio_type_name[t], zio_taskq_types[q], i); | |
1151 | else | |
1152 | (void) snprintf(name, sizeof (name), "%s_%s", | |
1153 | zio_type_name[t], zio_taskq_types[q]); | |
7ef5e54e | 1154 | |
b5e60918 | 1155 | #ifdef HAVE_SYSDC |
7ef5e54e | 1156 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { |
18168da7 | 1157 | (void) zio_taskq_basedc; |
7ef5e54e AL |
1158 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, |
1159 | spa->spa_proc, zio_taskq_basedc, flags); | |
1160 | } else { | |
b5e60918 | 1161 | #endif |
e8b96c60 MA |
1162 | pri_t pri = maxclsyspri; |
1163 | /* | |
1164 | * The write issue taskq can be extremely CPU | |
1229323d | 1165 | * intensive. Run it at slightly less important |
7432d297 MM |
1166 | * priority than the other taskqs. |
1167 | * | |
1168 | * Under Linux and FreeBSD this means incrementing | |
1169 | * the priority value as opposed to platforms like | |
1170 | * illumos where it should be decremented. | |
1171 | * | |
1172 | * On FreeBSD, if priorities divided by four (RQ_PPQ) | |
1173 | * are equal then a difference between them is | |
1174 | * insignificant. | |
e8b96c60 | 1175 | */ |
7432d297 MM |
1176 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) { |
1177 | #if defined(__linux__) | |
1229323d | 1178 | pri++; |
7432d297 MM |
1179 | #elif defined(__FreeBSD__) |
1180 | pri += 4; | |
1181 | #else | |
1182 | #error "unknown OS" | |
1183 | #endif | |
1184 | } | |
e8b96c60 | 1185 | tq = taskq_create_proc(name, value, pri, 50, |
7ef5e54e | 1186 | INT_MAX, spa->spa_proc, flags); |
b5e60918 | 1187 | #ifdef HAVE_SYSDC |
7ef5e54e | 1188 | } |
b5e60918 | 1189 | #endif |
7ef5e54e AL |
1190 | |
1191 | tqs->stqs_taskq[i] = tq; | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | static void | |
1196 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
1197 | { | |
1198 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7ef5e54e AL |
1199 | |
1200 | if (tqs->stqs_taskq == NULL) { | |
1201 | ASSERT3U(tqs->stqs_count, ==, 0); | |
1202 | return; | |
1203 | } | |
1204 | ||
1c27024e | 1205 | for (uint_t i = 0; i < tqs->stqs_count; i++) { |
7ef5e54e AL |
1206 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); |
1207 | taskq_destroy(tqs->stqs_taskq[i]); | |
428870ff | 1208 | } |
34dc7c2f | 1209 | |
7ef5e54e AL |
1210 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); |
1211 | tqs->stqs_taskq = NULL; | |
1212 | } | |
34dc7c2f | 1213 | |
7ef5e54e AL |
1214 | /* |
1215 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
1216 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
3bd4df38 | 1217 | * on the taskq itself. |
7ef5e54e | 1218 | */ |
3bd4df38 EN |
1219 | static taskq_t * |
1220 | spa_taskq_dispatch_select(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1221 | zio_t *zio) | |
7ef5e54e AL |
1222 | { |
1223 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1224 | taskq_t *tq; | |
1225 | ||
1226 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1227 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1228 | ||
3bd4df38 EN |
1229 | if ((t == ZIO_TYPE_WRITE) && (q == ZIO_TASKQ_ISSUE) && |
1230 | (zio != NULL) && (zio->io_wr_iss_tq != NULL)) { | |
1231 | /* dispatch to assigned write issue taskq */ | |
1232 | tq = zio->io_wr_iss_tq; | |
1233 | return (tq); | |
1234 | } | |
1235 | ||
7ef5e54e AL |
1236 | if (tqs->stqs_count == 1) { |
1237 | tq = tqs->stqs_taskq[0]; | |
1238 | } else { | |
c12936b1 | 1239 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
428870ff | 1240 | } |
3bd4df38 EN |
1241 | return (tq); |
1242 | } | |
7ef5e54e | 1243 | |
3bd4df38 EN |
1244 | void |
1245 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1246 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent, | |
1247 | zio_t *zio) | |
1248 | { | |
1249 | taskq_t *tq = spa_taskq_dispatch_select(spa, t, q, zio); | |
7ef5e54e | 1250 | taskq_dispatch_ent(tq, func, arg, flags, ent); |
428870ff BB |
1251 | } |
1252 | ||
044baf00 BB |
1253 | /* |
1254 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
1255 | */ | |
1256 | void | |
1257 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1258 | task_func_t *func, void *arg, uint_t flags) | |
1259 | { | |
3bd4df38 EN |
1260 | taskq_t *tq = spa_taskq_dispatch_select(spa, t, q, NULL); |
1261 | taskqid_t id = taskq_dispatch(tq, func, arg, flags); | |
044baf00 BB |
1262 | if (id) |
1263 | taskq_wait_id(tq, id); | |
1264 | } | |
1265 | ||
428870ff BB |
1266 | static void |
1267 | spa_create_zio_taskqs(spa_t *spa) | |
1268 | { | |
1c27024e DB |
1269 | for (int t = 0; t < ZIO_TYPES; t++) { |
1270 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1271 | spa_taskqs_init(spa, t, q); |
428870ff BB |
1272 | } |
1273 | } | |
1274 | } | |
9babb374 | 1275 | |
7b89a549 | 1276 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) |
428870ff BB |
1277 | static void |
1278 | spa_thread(void *arg) | |
1279 | { | |
93ce2b4c | 1280 | psetid_t zio_taskq_psrset_bind = PS_NONE; |
428870ff | 1281 | callb_cpr_t cprinfo; |
9babb374 | 1282 | |
428870ff BB |
1283 | spa_t *spa = arg; |
1284 | user_t *pu = PTOU(curproc); | |
9babb374 | 1285 | |
428870ff BB |
1286 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, |
1287 | spa->spa_name); | |
9babb374 | 1288 | |
428870ff BB |
1289 | ASSERT(curproc != &p0); |
1290 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1291 | "zpool-%s", spa->spa_name); | |
1292 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1293 | ||
1294 | /* bind this thread to the requested psrset */ | |
1295 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1296 | pool_lock(); | |
1297 | mutex_enter(&cpu_lock); | |
1298 | mutex_enter(&pidlock); | |
1299 | mutex_enter(&curproc->p_lock); | |
1300 | ||
1301 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1302 | 0, NULL, NULL) == 0) { | |
1303 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1304 | } else { | |
1305 | cmn_err(CE_WARN, | |
1306 | "Couldn't bind process for zfs pool \"%s\" to " | |
1307 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1308 | } | |
1309 | ||
1310 | mutex_exit(&curproc->p_lock); | |
1311 | mutex_exit(&pidlock); | |
1312 | mutex_exit(&cpu_lock); | |
1313 | pool_unlock(); | |
1314 | } | |
1315 | ||
b5e60918 | 1316 | #ifdef HAVE_SYSDC |
428870ff BB |
1317 | if (zio_taskq_sysdc) { |
1318 | sysdc_thread_enter(curthread, 100, 0); | |
1319 | } | |
b5e60918 | 1320 | #endif |
428870ff BB |
1321 | |
1322 | spa->spa_proc = curproc; | |
1323 | spa->spa_did = curthread->t_did; | |
1324 | ||
1325 | spa_create_zio_taskqs(spa); | |
1326 | ||
1327 | mutex_enter(&spa->spa_proc_lock); | |
1328 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1329 | ||
1330 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1331 | cv_broadcast(&spa->spa_proc_cv); | |
1332 | ||
1333 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1334 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1335 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1336 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1337 | ||
1338 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1339 | spa->spa_proc_state = SPA_PROC_GONE; | |
1340 | spa->spa_proc = &p0; | |
1341 | cv_broadcast(&spa->spa_proc_cv); | |
1342 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1343 | ||
1344 | mutex_enter(&curproc->p_lock); | |
1345 | lwp_exit(); | |
1346 | } | |
1347 | #endif | |
1348 | ||
95f71c01 EN |
1349 | extern metaslab_ops_t *metaslab_allocator(spa_t *spa); |
1350 | ||
428870ff BB |
1351 | /* |
1352 | * Activate an uninitialized pool. | |
1353 | */ | |
1354 | static void | |
da92d5cb | 1355 | spa_activate(spa_t *spa, spa_mode_t mode) |
428870ff | 1356 | { |
95f71c01 | 1357 | metaslab_ops_t *msp = metaslab_allocator(spa); |
428870ff BB |
1358 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); |
1359 | ||
1360 | spa->spa_state = POOL_STATE_ACTIVE; | |
1361 | spa->spa_mode = mode; | |
e39fe05b | 1362 | spa->spa_read_spacemaps = spa_mode_readable_spacemaps; |
428870ff | 1363 | |
95f71c01 EN |
1364 | spa->spa_normal_class = metaslab_class_create(spa, msp); |
1365 | spa->spa_log_class = metaslab_class_create(spa, msp); | |
1366 | spa->spa_embedded_log_class = metaslab_class_create(spa, msp); | |
1367 | spa->spa_special_class = metaslab_class_create(spa, msp); | |
1368 | spa->spa_dedup_class = metaslab_class_create(spa, msp); | |
428870ff BB |
1369 | |
1370 | /* Try to create a covering process */ | |
1371 | mutex_enter(&spa->spa_proc_lock); | |
1372 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1373 | ASSERT(spa->spa_proc == &p0); | |
1374 | spa->spa_did = 0; | |
1375 | ||
7b89a549 | 1376 | #ifdef HAVE_SPA_THREAD |
428870ff BB |
1377 | /* Only create a process if we're going to be around a while. */ |
1378 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1379 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1380 | NULL, 0) == 0) { | |
1381 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1382 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1383 | cv_wait(&spa->spa_proc_cv, | |
1384 | &spa->spa_proc_lock); | |
9babb374 | 1385 | } |
428870ff BB |
1386 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); |
1387 | ASSERT(spa->spa_proc != &p0); | |
1388 | ASSERT(spa->spa_did != 0); | |
1389 | } else { | |
1390 | #ifdef _KERNEL | |
1391 | cmn_err(CE_WARN, | |
1392 | "Couldn't create process for zfs pool \"%s\"\n", | |
1393 | spa->spa_name); | |
1394 | #endif | |
b128c09f | 1395 | } |
34dc7c2f | 1396 | } |
7b89a549 | 1397 | #endif /* HAVE_SPA_THREAD */ |
428870ff BB |
1398 | mutex_exit(&spa->spa_proc_lock); |
1399 | ||
1400 | /* If we didn't create a process, we need to create our taskqs. */ | |
1401 | if (spa->spa_proc == &p0) { | |
1402 | spa_create_zio_taskqs(spa); | |
1403 | } | |
34dc7c2f | 1404 | |
619f0976 GW |
1405 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1406 | spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, | |
1407 | ZIO_FLAG_CANFAIL); | |
1408 | } | |
a1d477c2 | 1409 | |
b128c09f BB |
1410 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), |
1411 | offsetof(vdev_t, vdev_config_dirty_node)); | |
0c66c32d JG |
1412 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), |
1413 | offsetof(objset_t, os_evicting_node)); | |
b128c09f BB |
1414 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), |
1415 | offsetof(vdev_t, vdev_state_dirty_node)); | |
34dc7c2f | 1416 | |
4747a7d3 | 1417 | txg_list_create(&spa->spa_vdev_txg_list, spa, |
34dc7c2f BB |
1418 | offsetof(struct vdev, vdev_txg_node)); |
1419 | ||
1420 | avl_create(&spa->spa_errlist_scrub, | |
1421 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1422 | offsetof(spa_error_entry_t, se_avl)); | |
1423 | avl_create(&spa->spa_errlist_last, | |
1424 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1425 | offsetof(spa_error_entry_t, se_avl)); | |
e8cf3a4f AP |
1426 | avl_create(&spa->spa_errlist_healed, |
1427 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1428 | offsetof(spa_error_entry_t, se_avl)); | |
a0bd735a | 1429 | |
4759342a JL |
1430 | spa_activate_os(spa); |
1431 | ||
b5256303 TC |
1432 | spa_keystore_init(&spa->spa_keystore); |
1433 | ||
a0bd735a BP |
1434 | /* |
1435 | * This taskq is used to perform zvol-minor-related tasks | |
1436 | * asynchronously. This has several advantages, including easy | |
d0249a4b | 1437 | * resolution of various deadlocks. |
a0bd735a BP |
1438 | * |
1439 | * The taskq must be single threaded to ensure tasks are always | |
1440 | * processed in the order in which they were dispatched. | |
1441 | * | |
1442 | * A taskq per pool allows one to keep the pools independent. | |
1443 | * This way if one pool is suspended, it will not impact another. | |
1444 | * | |
1445 | * The preferred location to dispatch a zvol minor task is a sync | |
1446 | * task. In this context, there is easy access to the spa_t and minimal | |
1447 | * error handling is required because the sync task must succeed. | |
1448 | */ | |
1449 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1450 | 1, INT_MAX, 0); | |
1de321e6 | 1451 | |
342357cd AM |
1452 | /* |
1453 | * The taskq to preload metaslabs. | |
1454 | */ | |
1455 | spa->spa_metaslab_taskq = taskq_create("z_metaslab", | |
1456 | metaslab_preload_pct, maxclsyspri, 1, INT_MAX, | |
1457 | TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
1458 | ||
77d8a0f1 | 1459 | /* |
1460 | * Taskq dedicated to prefetcher threads: this is used to prevent the | |
1461 | * pool traverse code from monopolizing the global (and limited) | |
1462 | * system_taskq by inappropriately scheduling long running tasks on it. | |
1463 | */ | |
60a4c7d2 PD |
1464 | spa->spa_prefetch_taskq = taskq_create("z_prefetch", 100, |
1465 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
77d8a0f1 | 1466 | |
1de321e6 JX |
1467 | /* |
1468 | * The taskq to upgrade datasets in this pool. Currently used by | |
9c5167d1 | 1469 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA. |
1de321e6 | 1470 | */ |
60a4c7d2 PD |
1471 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", 100, |
1472 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
34dc7c2f BB |
1473 | } |
1474 | ||
1475 | /* | |
1476 | * Opposite of spa_activate(). | |
1477 | */ | |
1478 | static void | |
1479 | spa_deactivate(spa_t *spa) | |
1480 | { | |
34dc7c2f BB |
1481 | ASSERT(spa->spa_sync_on == B_FALSE); |
1482 | ASSERT(spa->spa_dsl_pool == NULL); | |
1483 | ASSERT(spa->spa_root_vdev == NULL); | |
9babb374 | 1484 | ASSERT(spa->spa_async_zio_root == NULL); |
34dc7c2f BB |
1485 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); |
1486 | ||
0c66c32d JG |
1487 | spa_evicting_os_wait(spa); |
1488 | ||
a0bd735a BP |
1489 | if (spa->spa_zvol_taskq) { |
1490 | taskq_destroy(spa->spa_zvol_taskq); | |
1491 | spa->spa_zvol_taskq = NULL; | |
1492 | } | |
1493 | ||
342357cd AM |
1494 | if (spa->spa_metaslab_taskq) { |
1495 | taskq_destroy(spa->spa_metaslab_taskq); | |
1496 | spa->spa_metaslab_taskq = NULL; | |
1497 | } | |
1498 | ||
77d8a0f1 | 1499 | if (spa->spa_prefetch_taskq) { |
1500 | taskq_destroy(spa->spa_prefetch_taskq); | |
1501 | spa->spa_prefetch_taskq = NULL; | |
1502 | } | |
1503 | ||
1de321e6 JX |
1504 | if (spa->spa_upgrade_taskq) { |
1505 | taskq_destroy(spa->spa_upgrade_taskq); | |
1506 | spa->spa_upgrade_taskq = NULL; | |
1507 | } | |
1508 | ||
34dc7c2f BB |
1509 | txg_list_destroy(&spa->spa_vdev_txg_list); |
1510 | ||
b128c09f | 1511 | list_destroy(&spa->spa_config_dirty_list); |
0c66c32d | 1512 | list_destroy(&spa->spa_evicting_os_list); |
b128c09f | 1513 | list_destroy(&spa->spa_state_dirty_list); |
34dc7c2f | 1514 | |
57ddcda1 | 1515 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 | 1516 | |
1c27024e DB |
1517 | for (int t = 0; t < ZIO_TYPES; t++) { |
1518 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1519 | spa_taskqs_fini(spa, t, q); |
b128c09f | 1520 | } |
34dc7c2f BB |
1521 | } |
1522 | ||
a1d477c2 MA |
1523 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1524 | ASSERT3P(spa->spa_txg_zio[i], !=, NULL); | |
1525 | VERIFY0(zio_wait(spa->spa_txg_zio[i])); | |
1526 | spa->spa_txg_zio[i] = NULL; | |
1527 | } | |
1528 | ||
34dc7c2f BB |
1529 | metaslab_class_destroy(spa->spa_normal_class); |
1530 | spa->spa_normal_class = NULL; | |
1531 | ||
1532 | metaslab_class_destroy(spa->spa_log_class); | |
1533 | spa->spa_log_class = NULL; | |
1534 | ||
aa755b35 MA |
1535 | metaslab_class_destroy(spa->spa_embedded_log_class); |
1536 | spa->spa_embedded_log_class = NULL; | |
1537 | ||
cc99f275 DB |
1538 | metaslab_class_destroy(spa->spa_special_class); |
1539 | spa->spa_special_class = NULL; | |
1540 | ||
1541 | metaslab_class_destroy(spa->spa_dedup_class); | |
1542 | spa->spa_dedup_class = NULL; | |
1543 | ||
34dc7c2f BB |
1544 | /* |
1545 | * If this was part of an import or the open otherwise failed, we may | |
1546 | * still have errors left in the queues. Empty them just in case. | |
1547 | */ | |
1548 | spa_errlog_drain(spa); | |
34dc7c2f BB |
1549 | avl_destroy(&spa->spa_errlist_scrub); |
1550 | avl_destroy(&spa->spa_errlist_last); | |
e8cf3a4f | 1551 | avl_destroy(&spa->spa_errlist_healed); |
34dc7c2f | 1552 | |
b5256303 TC |
1553 | spa_keystore_fini(&spa->spa_keystore); |
1554 | ||
34dc7c2f | 1555 | spa->spa_state = POOL_STATE_UNINITIALIZED; |
428870ff BB |
1556 | |
1557 | mutex_enter(&spa->spa_proc_lock); | |
1558 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1559 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1560 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1561 | cv_broadcast(&spa->spa_proc_cv); | |
1562 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1563 | ASSERT(spa->spa_proc != &p0); | |
1564 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1565 | } | |
1566 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1567 | spa->spa_proc_state = SPA_PROC_NONE; | |
1568 | } | |
1569 | ASSERT(spa->spa_proc == &p0); | |
1570 | mutex_exit(&spa->spa_proc_lock); | |
1571 | ||
1572 | /* | |
1573 | * We want to make sure spa_thread() has actually exited the ZFS | |
1574 | * module, so that the module can't be unloaded out from underneath | |
1575 | * it. | |
1576 | */ | |
1577 | if (spa->spa_did != 0) { | |
1578 | thread_join(spa->spa_did); | |
1579 | spa->spa_did = 0; | |
1580 | } | |
4759342a JL |
1581 | |
1582 | spa_deactivate_os(spa); | |
1583 | ||
34dc7c2f BB |
1584 | } |
1585 | ||
1586 | /* | |
1587 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1588 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1589 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1590 | * All vdev validation is done by the vdev_alloc() routine. | |
1591 | */ | |
4a22ba5b | 1592 | int |
34dc7c2f BB |
1593 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, |
1594 | uint_t id, int atype) | |
1595 | { | |
1596 | nvlist_t **child; | |
9babb374 | 1597 | uint_t children; |
34dc7c2f BB |
1598 | int error; |
1599 | ||
1600 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1601 | return (error); | |
1602 | ||
1603 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1604 | return (0); | |
1605 | ||
b128c09f BB |
1606 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
1607 | &child, &children); | |
1608 | ||
1609 | if (error == ENOENT) | |
1610 | return (0); | |
1611 | ||
1612 | if (error) { | |
34dc7c2f BB |
1613 | vdev_free(*vdp); |
1614 | *vdp = NULL; | |
2e528b49 | 1615 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
1616 | } |
1617 | ||
1c27024e | 1618 | for (int c = 0; c < children; c++) { |
34dc7c2f BB |
1619 | vdev_t *vd; |
1620 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1621 | atype)) != 0) { | |
1622 | vdev_free(*vdp); | |
1623 | *vdp = NULL; | |
1624 | return (error); | |
1625 | } | |
1626 | } | |
1627 | ||
1628 | ASSERT(*vdp != NULL); | |
1629 | ||
1630 | return (0); | |
1631 | } | |
1632 | ||
93e28d66 SD |
1633 | static boolean_t |
1634 | spa_should_flush_logs_on_unload(spa_t *spa) | |
1635 | { | |
1636 | if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) | |
1637 | return (B_FALSE); | |
1638 | ||
1639 | if (!spa_writeable(spa)) | |
1640 | return (B_FALSE); | |
1641 | ||
1642 | if (!spa->spa_sync_on) | |
1643 | return (B_FALSE); | |
1644 | ||
1645 | if (spa_state(spa) != POOL_STATE_EXPORTED) | |
1646 | return (B_FALSE); | |
1647 | ||
1648 | if (zfs_keep_log_spacemaps_at_export) | |
1649 | return (B_FALSE); | |
1650 | ||
1651 | return (B_TRUE); | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * Opens a transaction that will set the flag that will instruct | |
1656 | * spa_sync to attempt to flush all the metaslabs for that txg. | |
1657 | */ | |
1658 | static void | |
1659 | spa_unload_log_sm_flush_all(spa_t *spa) | |
1660 | { | |
1661 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
1662 | VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); | |
1663 | ||
1664 | ASSERT3U(spa->spa_log_flushall_txg, ==, 0); | |
1665 | spa->spa_log_flushall_txg = dmu_tx_get_txg(tx); | |
1666 | ||
1667 | dmu_tx_commit(tx); | |
1668 | txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg); | |
1669 | } | |
1670 | ||
1671 | static void | |
1672 | spa_unload_log_sm_metadata(spa_t *spa) | |
1673 | { | |
1674 | void *cookie = NULL; | |
1675 | spa_log_sm_t *sls; | |
b3ad3f48 AM |
1676 | log_summary_entry_t *e; |
1677 | ||
93e28d66 SD |
1678 | while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg, |
1679 | &cookie)) != NULL) { | |
1680 | VERIFY0(sls->sls_mscount); | |
1681 | kmem_free(sls, sizeof (spa_log_sm_t)); | |
1682 | } | |
1683 | ||
b3ad3f48 | 1684 | while ((e = list_remove_head(&spa->spa_log_summary)) != NULL) { |
93e28d66 | 1685 | VERIFY0(e->lse_mscount); |
93e28d66 SD |
1686 | kmem_free(e, sizeof (log_summary_entry_t)); |
1687 | } | |
1688 | ||
1689 | spa->spa_unflushed_stats.sus_nblocks = 0; | |
1690 | spa->spa_unflushed_stats.sus_memused = 0; | |
1691 | spa->spa_unflushed_stats.sus_blocklimit = 0; | |
1692 | } | |
1693 | ||
37f03da8 SH |
1694 | static void |
1695 | spa_destroy_aux_threads(spa_t *spa) | |
1696 | { | |
1697 | if (spa->spa_condense_zthr != NULL) { | |
1698 | zthr_destroy(spa->spa_condense_zthr); | |
1699 | spa->spa_condense_zthr = NULL; | |
1700 | } | |
1701 | if (spa->spa_checkpoint_discard_zthr != NULL) { | |
1702 | zthr_destroy(spa->spa_checkpoint_discard_zthr); | |
1703 | spa->spa_checkpoint_discard_zthr = NULL; | |
1704 | } | |
1705 | if (spa->spa_livelist_delete_zthr != NULL) { | |
1706 | zthr_destroy(spa->spa_livelist_delete_zthr); | |
1707 | spa->spa_livelist_delete_zthr = NULL; | |
1708 | } | |
1709 | if (spa->spa_livelist_condense_zthr != NULL) { | |
1710 | zthr_destroy(spa->spa_livelist_condense_zthr); | |
1711 | spa->spa_livelist_condense_zthr = NULL; | |
1712 | } | |
5caeef02 DB |
1713 | if (spa->spa_raidz_expand_zthr != NULL) { |
1714 | zthr_destroy(spa->spa_raidz_expand_zthr); | |
1715 | spa->spa_raidz_expand_zthr = NULL; | |
1716 | } | |
37f03da8 SH |
1717 | } |
1718 | ||
34dc7c2f BB |
1719 | /* |
1720 | * Opposite of spa_load(). | |
1721 | */ | |
1722 | static void | |
1723 | spa_unload(spa_t *spa) | |
1724 | { | |
b128c09f | 1725 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
93e28d66 | 1726 | ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED); |
b128c09f | 1727 | |
ca95f70d | 1728 | spa_import_progress_remove(spa_guid(spa)); |
4a0ee12a PZ |
1729 | spa_load_note(spa, "UNLOADING"); |
1730 | ||
e60e158e JG |
1731 | spa_wake_waiters(spa); |
1732 | ||
93e28d66 | 1733 | /* |
2fb52853 GA |
1734 | * If we have set the spa_final_txg, we have already performed the |
1735 | * tasks below in spa_export_common(). We should not redo it here since | |
1736 | * we delay the final TXGs beyond what spa_final_txg is set at. | |
93e28d66 | 1737 | */ |
2fb52853 GA |
1738 | if (spa->spa_final_txg == UINT64_MAX) { |
1739 | /* | |
1740 | * If the log space map feature is enabled and the pool is | |
1741 | * getting exported (but not destroyed), we want to spend some | |
1742 | * time flushing as many metaslabs as we can in an attempt to | |
1743 | * destroy log space maps and save import time. | |
1744 | */ | |
1745 | if (spa_should_flush_logs_on_unload(spa)) | |
1746 | spa_unload_log_sm_flush_all(spa); | |
93e28d66 | 1747 | |
2fb52853 GA |
1748 | /* |
1749 | * Stop async tasks. | |
1750 | */ | |
1751 | spa_async_suspend(spa); | |
34dc7c2f | 1752 | |
2fb52853 GA |
1753 | if (spa->spa_root_vdev) { |
1754 | vdev_t *root_vdev = spa->spa_root_vdev; | |
1755 | vdev_initialize_stop_all(root_vdev, | |
1756 | VDEV_INITIALIZE_ACTIVE); | |
1757 | vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE); | |
1758 | vdev_autotrim_stop_all(spa); | |
1759 | vdev_rebuild_stop_all(spa); | |
1760 | } | |
619f0976 GW |
1761 | } |
1762 | ||
34dc7c2f BB |
1763 | /* |
1764 | * Stop syncing. | |
1765 | */ | |
1766 | if (spa->spa_sync_on) { | |
1767 | txg_sync_stop(spa->spa_dsl_pool); | |
1768 | spa->spa_sync_on = B_FALSE; | |
1769 | } | |
1770 | ||
4e21fd06 | 1771 | /* |
93e28d66 SD |
1772 | * This ensures that there is no async metaslab prefetching |
1773 | * while we attempt to unload the spa. | |
4e21fd06 | 1774 | */ |
342357cd | 1775 | taskq_wait(spa->spa_metaslab_taskq); |
4e21fd06 | 1776 | |
379ca9cf OF |
1777 | if (spa->spa_mmp.mmp_thread) |
1778 | mmp_thread_stop(spa); | |
1779 | ||
34dc7c2f | 1780 | /* |
b128c09f | 1781 | * Wait for any outstanding async I/O to complete. |
34dc7c2f | 1782 | */ |
9babb374 | 1783 | if (spa->spa_async_zio_root != NULL) { |
1c27024e | 1784 | for (int i = 0; i < max_ncpus; i++) |
e022864d MA |
1785 | (void) zio_wait(spa->spa_async_zio_root[i]); |
1786 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
9babb374 BB |
1787 | spa->spa_async_zio_root = NULL; |
1788 | } | |
34dc7c2f | 1789 | |
a1d477c2 MA |
1790 | if (spa->spa_vdev_removal != NULL) { |
1791 | spa_vdev_removal_destroy(spa->spa_vdev_removal); | |
1792 | spa->spa_vdev_removal = NULL; | |
1793 | } | |
1794 | ||
37f03da8 | 1795 | spa_destroy_aux_threads(spa); |
d2734cce | 1796 | |
a1d477c2 MA |
1797 | spa_condense_fini(spa); |
1798 | ||
428870ff BB |
1799 | bpobj_close(&spa->spa_deferred_bpobj); |
1800 | ||
619f0976 | 1801 | spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); |
93cf2076 GW |
1802 | |
1803 | /* | |
1804 | * Close all vdevs. | |
1805 | */ | |
1806 | if (spa->spa_root_vdev) | |
1807 | vdev_free(spa->spa_root_vdev); | |
1808 | ASSERT(spa->spa_root_vdev == NULL); | |
1809 | ||
34dc7c2f BB |
1810 | /* |
1811 | * Close the dsl pool. | |
1812 | */ | |
1813 | if (spa->spa_dsl_pool) { | |
1814 | dsl_pool_close(spa->spa_dsl_pool); | |
1815 | spa->spa_dsl_pool = NULL; | |
428870ff | 1816 | spa->spa_meta_objset = NULL; |
34dc7c2f BB |
1817 | } |
1818 | ||
428870ff | 1819 | ddt_unload(spa); |
67a1b037 | 1820 | brt_unload(spa); |
93e28d66 | 1821 | spa_unload_log_sm_metadata(spa); |
428870ff | 1822 | |
fb5f0bc8 BB |
1823 | /* |
1824 | * Drop and purge level 2 cache | |
1825 | */ | |
1826 | spa_l2cache_drop(spa); | |
1827 | ||
34dc7c2f | 1828 | if (spa->spa_spares.sav_vdevs) { |
cfb49616 RY |
1829 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
1830 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
34dc7c2f BB |
1831 | kmem_free(spa->spa_spares.sav_vdevs, |
1832 | spa->spa_spares.sav_count * sizeof (void *)); | |
1833 | spa->spa_spares.sav_vdevs = NULL; | |
1834 | } | |
1835 | if (spa->spa_spares.sav_config) { | |
1836 | nvlist_free(spa->spa_spares.sav_config); | |
1837 | spa->spa_spares.sav_config = NULL; | |
1838 | } | |
b128c09f | 1839 | spa->spa_spares.sav_count = 0; |
34dc7c2f | 1840 | |
34dc7c2f | 1841 | if (spa->spa_l2cache.sav_vdevs) { |
cfb49616 RY |
1842 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) { |
1843 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
1844 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
1845 | } | |
34dc7c2f BB |
1846 | kmem_free(spa->spa_l2cache.sav_vdevs, |
1847 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
1848 | spa->spa_l2cache.sav_vdevs = NULL; | |
1849 | } | |
1850 | if (spa->spa_l2cache.sav_config) { | |
1851 | nvlist_free(spa->spa_l2cache.sav_config); | |
1852 | spa->spa_l2cache.sav_config = NULL; | |
1853 | } | |
b128c09f | 1854 | spa->spa_l2cache.sav_count = 0; |
34dc7c2f BB |
1855 | |
1856 | spa->spa_async_suspended = 0; | |
fb5f0bc8 | 1857 | |
a1d477c2 MA |
1858 | spa->spa_indirect_vdevs_loaded = B_FALSE; |
1859 | ||
d96eb2b1 DM |
1860 | if (spa->spa_comment != NULL) { |
1861 | spa_strfree(spa->spa_comment); | |
1862 | spa->spa_comment = NULL; | |
1863 | } | |
658fb802 CB |
1864 | if (spa->spa_compatibility != NULL) { |
1865 | spa_strfree(spa->spa_compatibility); | |
1866 | spa->spa_compatibility = NULL; | |
1867 | } | |
d96eb2b1 | 1868 | |
5caeef02 DB |
1869 | spa->spa_raidz_expand = NULL; |
1870 | ||
619f0976 | 1871 | spa_config_exit(spa, SCL_ALL, spa); |
34dc7c2f BB |
1872 | } |
1873 | ||
1874 | /* | |
1875 | * Load (or re-load) the current list of vdevs describing the active spares for | |
1876 | * this pool. When this is called, we have some form of basic information in | |
1877 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
1878 | * then re-generate a more complete list including status information. | |
1879 | */ | |
a1d477c2 | 1880 | void |
34dc7c2f BB |
1881 | spa_load_spares(spa_t *spa) |
1882 | { | |
1883 | nvlist_t **spares; | |
1884 | uint_t nspares; | |
1885 | int i; | |
1886 | vdev_t *vd, *tvd; | |
1887 | ||
d2734cce SD |
1888 | #ifndef _KERNEL |
1889 | /* | |
1890 | * zdb opens both the current state of the pool and the | |
1891 | * checkpointed state (if present), with a different spa_t. | |
1892 | * | |
1893 | * As spare vdevs are shared among open pools, we skip loading | |
1894 | * them when we load the checkpointed state of the pool. | |
1895 | */ | |
1896 | if (!spa_writeable(spa)) | |
1897 | return; | |
1898 | #endif | |
1899 | ||
b128c09f BB |
1900 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
1901 | ||
34dc7c2f BB |
1902 | /* |
1903 | * First, close and free any existing spare vdevs. | |
1904 | */ | |
cfb49616 RY |
1905 | if (spa->spa_spares.sav_vdevs) { |
1906 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1907 | vd = spa->spa_spares.sav_vdevs[i]; | |
1908 | ||
1909 | /* Undo the call to spa_activate() below */ | |
1910 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1911 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
1912 | spa_spare_remove(tvd); | |
1913 | vdev_close(vd); | |
1914 | vdev_free(vd); | |
1915 | } | |
34dc7c2f | 1916 | |
34dc7c2f BB |
1917 | kmem_free(spa->spa_spares.sav_vdevs, |
1918 | spa->spa_spares.sav_count * sizeof (void *)); | |
cfb49616 | 1919 | } |
34dc7c2f BB |
1920 | |
1921 | if (spa->spa_spares.sav_config == NULL) | |
1922 | nspares = 0; | |
1923 | else | |
65ad5d11 AJ |
1924 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, |
1925 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f BB |
1926 | |
1927 | spa->spa_spares.sav_count = (int)nspares; | |
1928 | spa->spa_spares.sav_vdevs = NULL; | |
1929 | ||
1930 | if (nspares == 0) | |
1931 | return; | |
1932 | ||
1933 | /* | |
1934 | * Construct the array of vdevs, opening them to get status in the | |
1935 | * process. For each spare, there is potentially two different vdev_t | |
1936 | * structures associated with it: one in the list of spares (used only | |
1937 | * for basic validation purposes) and one in the active vdev | |
1938 | * configuration (if it's spared in). During this phase we open and | |
1939 | * validate each vdev on the spare list. If the vdev also exists in the | |
1940 | * active configuration, then we also mark this vdev as an active spare. | |
1941 | */ | |
904ea276 | 1942 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), |
79c76d5b | 1943 | KM_SLEEP); |
34dc7c2f BB |
1944 | for (i = 0; i < spa->spa_spares.sav_count; i++) { |
1945 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
1946 | VDEV_ALLOC_SPARE) == 0); | |
1947 | ASSERT(vd != NULL); | |
1948 | ||
1949 | spa->spa_spares.sav_vdevs[i] = vd; | |
1950 | ||
b128c09f BB |
1951 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, |
1952 | B_FALSE)) != NULL) { | |
34dc7c2f BB |
1953 | if (!tvd->vdev_isspare) |
1954 | spa_spare_add(tvd); | |
1955 | ||
1956 | /* | |
1957 | * We only mark the spare active if we were successfully | |
1958 | * able to load the vdev. Otherwise, importing a pool | |
1959 | * with a bad active spare would result in strange | |
1960 | * behavior, because multiple pool would think the spare | |
1961 | * is actively in use. | |
1962 | * | |
1963 | * There is a vulnerability here to an equally bizarre | |
1964 | * circumstance, where a dead active spare is later | |
1965 | * brought back to life (onlined or otherwise). Given | |
1966 | * the rarity of this scenario, and the extra complexity | |
1967 | * it adds, we ignore the possibility. | |
1968 | */ | |
1969 | if (!vdev_is_dead(tvd)) | |
1970 | spa_spare_activate(tvd); | |
1971 | } | |
1972 | ||
b128c09f | 1973 | vd->vdev_top = vd; |
9babb374 | 1974 | vd->vdev_aux = &spa->spa_spares; |
b128c09f | 1975 | |
34dc7c2f BB |
1976 | if (vdev_open(vd) != 0) |
1977 | continue; | |
1978 | ||
34dc7c2f BB |
1979 | if (vdev_validate_aux(vd) == 0) |
1980 | spa_spare_add(vd); | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * Recompute the stashed list of spares, with status information | |
1985 | * this time. | |
1986 | */ | |
65ad5d11 | 1987 | fnvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES); |
34dc7c2f BB |
1988 | |
1989 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
79c76d5b | 1990 | KM_SLEEP); |
34dc7c2f BB |
1991 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1992 | spares[i] = vdev_config_generate(spa, | |
428870ff | 1993 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); |
65ad5d11 | 1994 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, |
795075e6 PD |
1995 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
1996 | spa->spa_spares.sav_count); | |
34dc7c2f BB |
1997 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1998 | nvlist_free(spares[i]); | |
1999 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
2000 | } | |
2001 | ||
2002 | /* | |
2003 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
2004 | * this pool. When this is called, we have some form of basic information in | |
2005 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
2006 | * then re-generate a more complete list including status information. | |
2007 | * Devices which are already active have their details maintained, and are | |
2008 | * not re-opened. | |
2009 | */ | |
a1d477c2 | 2010 | void |
34dc7c2f BB |
2011 | spa_load_l2cache(spa_t *spa) |
2012 | { | |
460f239e | 2013 | nvlist_t **l2cache = NULL; |
34dc7c2f BB |
2014 | uint_t nl2cache; |
2015 | int i, j, oldnvdevs; | |
9babb374 | 2016 | uint64_t guid; |
a117a6d6 | 2017 | vdev_t *vd, **oldvdevs, **newvdevs; |
34dc7c2f BB |
2018 | spa_aux_vdev_t *sav = &spa->spa_l2cache; |
2019 | ||
d2734cce SD |
2020 | #ifndef _KERNEL |
2021 | /* | |
2022 | * zdb opens both the current state of the pool and the | |
2023 | * checkpointed state (if present), with a different spa_t. | |
2024 | * | |
2025 | * As L2 caches are part of the ARC which is shared among open | |
2026 | * pools, we skip loading them when we load the checkpointed | |
2027 | * state of the pool. | |
2028 | */ | |
2029 | if (!spa_writeable(spa)) | |
2030 | return; | |
2031 | #endif | |
2032 | ||
b128c09f BB |
2033 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
2034 | ||
34dc7c2f BB |
2035 | oldvdevs = sav->sav_vdevs; |
2036 | oldnvdevs = sav->sav_count; | |
2037 | sav->sav_vdevs = NULL; | |
2038 | sav->sav_count = 0; | |
2039 | ||
67d60824 NB |
2040 | if (sav->sav_config == NULL) { |
2041 | nl2cache = 0; | |
2042 | newvdevs = NULL; | |
2043 | goto out; | |
2044 | } | |
2045 | ||
65ad5d11 AJ |
2046 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, |
2047 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
67d60824 NB |
2048 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); |
2049 | ||
34dc7c2f BB |
2050 | /* |
2051 | * Process new nvlist of vdevs. | |
2052 | */ | |
2053 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 | 2054 | guid = fnvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID); |
34dc7c2f BB |
2055 | |
2056 | newvdevs[i] = NULL; | |
2057 | for (j = 0; j < oldnvdevs; j++) { | |
2058 | vd = oldvdevs[j]; | |
2059 | if (vd != NULL && guid == vd->vdev_guid) { | |
2060 | /* | |
2061 | * Retain previous vdev for add/remove ops. | |
2062 | */ | |
2063 | newvdevs[i] = vd; | |
2064 | oldvdevs[j] = NULL; | |
2065 | break; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | if (newvdevs[i] == NULL) { | |
2070 | /* | |
2071 | * Create new vdev | |
2072 | */ | |
2073 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
2074 | VDEV_ALLOC_L2CACHE) == 0); | |
2075 | ASSERT(vd != NULL); | |
2076 | newvdevs[i] = vd; | |
2077 | ||
2078 | /* | |
2079 | * Commit this vdev as an l2cache device, | |
2080 | * even if it fails to open. | |
2081 | */ | |
2082 | spa_l2cache_add(vd); | |
2083 | ||
b128c09f BB |
2084 | vd->vdev_top = vd; |
2085 | vd->vdev_aux = sav; | |
2086 | ||
2087 | spa_l2cache_activate(vd); | |
2088 | ||
34dc7c2f BB |
2089 | if (vdev_open(vd) != 0) |
2090 | continue; | |
2091 | ||
34dc7c2f BB |
2092 | (void) vdev_validate_aux(vd); |
2093 | ||
9babb374 BB |
2094 | if (!vdev_is_dead(vd)) |
2095 | l2arc_add_vdev(spa, vd); | |
b7654bd7 GA |
2096 | |
2097 | /* | |
2098 | * Upon cache device addition to a pool or pool | |
2099 | * creation with a cache device or if the header | |
2100 | * of the device is invalid we issue an async | |
2101 | * TRIM command for the whole device which will | |
2102 | * execute if l2arc_trim_ahead > 0. | |
2103 | */ | |
2104 | spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); | |
34dc7c2f BB |
2105 | } |
2106 | } | |
2107 | ||
67d60824 NB |
2108 | sav->sav_vdevs = newvdevs; |
2109 | sav->sav_count = (int)nl2cache; | |
2110 | ||
2111 | /* | |
2112 | * Recompute the stashed list of l2cache devices, with status | |
2113 | * information this time. | |
2114 | */ | |
65ad5d11 | 2115 | fnvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE); |
67d60824 | 2116 | |
460f239e D |
2117 | if (sav->sav_count > 0) |
2118 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
2119 | KM_SLEEP); | |
67d60824 NB |
2120 | for (i = 0; i < sav->sav_count; i++) |
2121 | l2cache[i] = vdev_config_generate(spa, | |
2122 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
795075e6 PD |
2123 | fnvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, |
2124 | (const nvlist_t * const *)l2cache, sav->sav_count); | |
67d60824 NB |
2125 | |
2126 | out: | |
34dc7c2f BB |
2127 | /* |
2128 | * Purge vdevs that were dropped | |
2129 | */ | |
cfb49616 RY |
2130 | if (oldvdevs) { |
2131 | for (i = 0; i < oldnvdevs; i++) { | |
2132 | uint64_t pool; | |
2133 | ||
2134 | vd = oldvdevs[i]; | |
2135 | if (vd != NULL) { | |
2136 | ASSERT(vd->vdev_isl2cache); | |
2137 | ||
2138 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
2139 | pool != 0ULL && l2arc_vdev_present(vd)) | |
2140 | l2arc_remove_vdev(vd); | |
2141 | vdev_clear_stats(vd); | |
2142 | vdev_free(vd); | |
2143 | } | |
34dc7c2f | 2144 | } |
34dc7c2f | 2145 | |
34dc7c2f | 2146 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); |
cfb49616 | 2147 | } |
34dc7c2f | 2148 | |
34dc7c2f BB |
2149 | for (i = 0; i < sav->sav_count; i++) |
2150 | nvlist_free(l2cache[i]); | |
2151 | if (sav->sav_count) | |
2152 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
2153 | } | |
2154 | ||
2155 | static int | |
2156 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
2157 | { | |
2158 | dmu_buf_t *db; | |
2159 | char *packed = NULL; | |
2160 | size_t nvsize = 0; | |
2161 | int error; | |
2162 | *value = NULL; | |
2163 | ||
c3275b56 BB |
2164 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); |
2165 | if (error) | |
2166 | return (error); | |
2167 | ||
34dc7c2f BB |
2168 | nvsize = *(uint64_t *)db->db_data; |
2169 | dmu_buf_rele(db, FTAG); | |
2170 | ||
77aef6f6 | 2171 | packed = vmem_alloc(nvsize, KM_SLEEP); |
9babb374 BB |
2172 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, |
2173 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
2174 | if (error == 0) |
2175 | error = nvlist_unpack(packed, nvsize, value, 0); | |
77aef6f6 | 2176 | vmem_free(packed, nvsize); |
34dc7c2f BB |
2177 | |
2178 | return (error); | |
2179 | } | |
2180 | ||
6cb8e530 PZ |
2181 | /* |
2182 | * Concrete top-level vdevs that are not missing and are not logs. At every | |
2183 | * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. | |
2184 | */ | |
2185 | static uint64_t | |
2186 | spa_healthy_core_tvds(spa_t *spa) | |
2187 | { | |
2188 | vdev_t *rvd = spa->spa_root_vdev; | |
2189 | uint64_t tvds = 0; | |
2190 | ||
2191 | for (uint64_t i = 0; i < rvd->vdev_children; i++) { | |
2192 | vdev_t *vd = rvd->vdev_child[i]; | |
2193 | if (vd->vdev_islog) | |
2194 | continue; | |
2195 | if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) | |
2196 | tvds++; | |
2197 | } | |
2198 | ||
2199 | return (tvds); | |
2200 | } | |
2201 | ||
34dc7c2f BB |
2202 | /* |
2203 | * Checks to see if the given vdev could not be opened, in which case we post a | |
2204 | * sysevent to notify the autoreplace code that the device has been removed. | |
2205 | */ | |
2206 | static void | |
2207 | spa_check_removed(vdev_t *vd) | |
2208 | { | |
6cb8e530 | 2209 | for (uint64_t c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2210 | spa_check_removed(vd->vdev_child[c]); |
2211 | ||
7011fb60 | 2212 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && |
a1d477c2 | 2213 | vdev_is_concrete(vd)) { |
fb390aaf | 2214 | zfs_post_autoreplace(vd->vdev_spa, vd); |
12fa0466 | 2215 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); |
34dc7c2f BB |
2216 | } |
2217 | } | |
2218 | ||
6cb8e530 PZ |
2219 | static int |
2220 | spa_check_for_missing_logs(spa_t *spa) | |
9babb374 | 2221 | { |
6cb8e530 | 2222 | vdev_t *rvd = spa->spa_root_vdev; |
9babb374 | 2223 | |
428870ff | 2224 | /* |
572e2857 | 2225 | * If we're doing a normal import, then build up any additional |
6cb8e530 | 2226 | * diagnostic information about missing log devices. |
572e2857 | 2227 | * We'll pass this up to the user for further processing. |
428870ff | 2228 | */ |
572e2857 BB |
2229 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { |
2230 | nvlist_t **child, *nv; | |
2231 | uint64_t idx = 0; | |
2232 | ||
160987b5 | 2233 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), |
79c76d5b | 2234 | KM_SLEEP); |
65ad5d11 | 2235 | nv = fnvlist_alloc(); |
572e2857 | 2236 | |
6cb8e530 | 2237 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { |
572e2857 | 2238 | vdev_t *tvd = rvd->vdev_child[c]; |
572e2857 | 2239 | |
6cb8e530 PZ |
2240 | /* |
2241 | * We consider a device as missing only if it failed | |
2242 | * to open (i.e. offline or faulted is not considered | |
2243 | * as missing). | |
2244 | */ | |
2245 | if (tvd->vdev_islog && | |
2246 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
2247 | child[idx++] = vdev_config_generate(spa, tvd, | |
2248 | B_FALSE, VDEV_CONFIG_MISSING); | |
2249 | } | |
572e2857 | 2250 | } |
9babb374 | 2251 | |
6cb8e530 | 2252 | if (idx > 0) { |
795075e6 PD |
2253 | fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
2254 | (const nvlist_t * const *)child, idx); | |
6cb8e530 PZ |
2255 | fnvlist_add_nvlist(spa->spa_load_info, |
2256 | ZPOOL_CONFIG_MISSING_DEVICES, nv); | |
572e2857 | 2257 | |
6cb8e530 | 2258 | for (uint64_t i = 0; i < idx; i++) |
572e2857 BB |
2259 | nvlist_free(child[i]); |
2260 | } | |
2261 | nvlist_free(nv); | |
2262 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
572e2857 | 2263 | |
6cb8e530 PZ |
2264 | if (idx > 0) { |
2265 | spa_load_failed(spa, "some log devices are missing"); | |
db7d07e1 | 2266 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 PZ |
2267 | return (SET_ERROR(ENXIO)); |
2268 | } | |
2269 | } else { | |
2270 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
2271 | vdev_t *tvd = rvd->vdev_child[c]; | |
a1d477c2 | 2272 | |
6cb8e530 PZ |
2273 | if (tvd->vdev_islog && |
2274 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
572e2857 | 2275 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
6cb8e530 PZ |
2276 | spa_load_note(spa, "some log devices are " |
2277 | "missing, ZIL is dropped."); | |
db7d07e1 | 2278 | vdev_dbgmsg_print_tree(rvd, 2); |
6cb8e530 | 2279 | break; |
e0ab3ab5 | 2280 | } |
572e2857 | 2281 | } |
9babb374 | 2282 | } |
e0ab3ab5 | 2283 | |
6cb8e530 | 2284 | return (0); |
9babb374 BB |
2285 | } |
2286 | ||
b128c09f BB |
2287 | /* |
2288 | * Check for missing log devices | |
2289 | */ | |
13fe0198 | 2290 | static boolean_t |
b128c09f BB |
2291 | spa_check_logs(spa_t *spa) |
2292 | { | |
13fe0198 | 2293 | boolean_t rv = B_FALSE; |
9c43027b | 2294 | dsl_pool_t *dp = spa_get_dsl(spa); |
13fe0198 | 2295 | |
b128c09f | 2296 | switch (spa->spa_log_state) { |
e75c13c3 BB |
2297 | default: |
2298 | break; | |
b128c09f BB |
2299 | case SPA_LOG_MISSING: |
2300 | /* need to recheck in case slog has been restored */ | |
2301 | case SPA_LOG_UNKNOWN: | |
9c43027b AJ |
2302 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
2303 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
13fe0198 | 2304 | if (rv) |
428870ff | 2305 | spa_set_log_state(spa, SPA_LOG_MISSING); |
b128c09f | 2306 | break; |
b128c09f | 2307 | } |
13fe0198 | 2308 | return (rv); |
b128c09f BB |
2309 | } |
2310 | ||
aa755b35 MA |
2311 | /* |
2312 | * Passivate any log vdevs (note, does not apply to embedded log metaslabs). | |
2313 | */ | |
428870ff BB |
2314 | static boolean_t |
2315 | spa_passivate_log(spa_t *spa) | |
34dc7c2f | 2316 | { |
428870ff BB |
2317 | vdev_t *rvd = spa->spa_root_vdev; |
2318 | boolean_t slog_found = B_FALSE; | |
b128c09f | 2319 | |
428870ff | 2320 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
fb5f0bc8 | 2321 | |
1c27024e | 2322 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2323 | vdev_t *tvd = rvd->vdev_child[c]; |
34dc7c2f | 2324 | |
428870ff | 2325 | if (tvd->vdev_islog) { |
aa755b35 MA |
2326 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); |
2327 | metaslab_group_passivate(tvd->vdev_mg); | |
428870ff BB |
2328 | slog_found = B_TRUE; |
2329 | } | |
34dc7c2f BB |
2330 | } |
2331 | ||
428870ff BB |
2332 | return (slog_found); |
2333 | } | |
34dc7c2f | 2334 | |
aa755b35 MA |
2335 | /* |
2336 | * Activate any log vdevs (note, does not apply to embedded log metaslabs). | |
2337 | */ | |
428870ff BB |
2338 | static void |
2339 | spa_activate_log(spa_t *spa) | |
2340 | { | |
2341 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 2342 | |
428870ff BB |
2343 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
2344 | ||
1c27024e | 2345 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff | 2346 | vdev_t *tvd = rvd->vdev_child[c]; |
428870ff | 2347 | |
aa755b35 MA |
2348 | if (tvd->vdev_islog) { |
2349 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); | |
2350 | metaslab_group_activate(tvd->vdev_mg); | |
2351 | } | |
34dc7c2f | 2352 | } |
428870ff | 2353 | } |
34dc7c2f | 2354 | |
428870ff | 2355 | int |
a1d477c2 | 2356 | spa_reset_logs(spa_t *spa) |
428870ff | 2357 | { |
13fe0198 | 2358 | int error; |
9babb374 | 2359 | |
a1d477c2 | 2360 | error = dmu_objset_find(spa_name(spa), zil_reset, |
13fe0198 MA |
2361 | NULL, DS_FIND_CHILDREN); |
2362 | if (error == 0) { | |
428870ff BB |
2363 | /* |
2364 | * We successfully offlined the log device, sync out the | |
2365 | * current txg so that the "stubby" block can be removed | |
2366 | * by zil_sync(). | |
2367 | */ | |
2368 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
2369 | } | |
2370 | return (error); | |
2371 | } | |
34dc7c2f | 2372 | |
428870ff BB |
2373 | static void |
2374 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
2375 | { | |
1c27024e | 2376 | for (int i = 0; i < sav->sav_count; i++) |
428870ff BB |
2377 | spa_check_removed(sav->sav_vdevs[i]); |
2378 | } | |
34dc7c2f | 2379 | |
428870ff BB |
2380 | void |
2381 | spa_claim_notify(zio_t *zio) | |
2382 | { | |
2383 | spa_t *spa = zio->io_spa; | |
34dc7c2f | 2384 | |
428870ff BB |
2385 | if (zio->io_error) |
2386 | return; | |
34dc7c2f | 2387 | |
428870ff BB |
2388 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ |
2389 | if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) | |
2390 | spa->spa_claim_max_txg = zio->io_bp->blk_birth; | |
2391 | mutex_exit(&spa->spa_props_lock); | |
2392 | } | |
34dc7c2f | 2393 | |
428870ff | 2394 | typedef struct spa_load_error { |
f2c5bc15 | 2395 | boolean_t sle_verify_data; |
428870ff BB |
2396 | uint64_t sle_meta_count; |
2397 | uint64_t sle_data_count; | |
2398 | } spa_load_error_t; | |
34dc7c2f | 2399 | |
428870ff BB |
2400 | static void |
2401 | spa_load_verify_done(zio_t *zio) | |
2402 | { | |
2403 | blkptr_t *bp = zio->io_bp; | |
2404 | spa_load_error_t *sle = zio->io_private; | |
2405 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
2406 | int error = zio->io_error; | |
dea377c0 | 2407 | spa_t *spa = zio->io_spa; |
34dc7c2f | 2408 | |
a6255b7f | 2409 | abd_free(zio->io_abd); |
428870ff | 2410 | if (error) { |
9ae529ec | 2411 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && |
428870ff | 2412 | type != DMU_OT_INTENT_LOG) |
bc89ac84 | 2413 | atomic_inc_64(&sle->sle_meta_count); |
428870ff | 2414 | else |
bc89ac84 | 2415 | atomic_inc_64(&sle->sle_data_count); |
34dc7c2f | 2416 | } |
dea377c0 MA |
2417 | |
2418 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2419 | spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp); |
dea377c0 MA |
2420 | cv_broadcast(&spa->spa_scrub_io_cv); |
2421 | mutex_exit(&spa->spa_scrub_lock); | |
428870ff | 2422 | } |
34dc7c2f | 2423 | |
dea377c0 | 2424 | /* |
e1cfd73f | 2425 | * Maximum number of inflight bytes is the log2 fraction of the arc size. |
c8242a96 | 2426 | * By default, we set it to 1/16th of the arc. |
dea377c0 | 2427 | */ |
fdc2d303 | 2428 | static uint_t spa_load_verify_shift = 4; |
18168da7 AZ |
2429 | static int spa_load_verify_metadata = B_TRUE; |
2430 | static int spa_load_verify_data = B_TRUE; | |
dea377c0 | 2431 | |
428870ff BB |
2432 | static int |
2433 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
5dbd68a3 | 2434 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) |
428870ff | 2435 | { |
f2c5bc15 AM |
2436 | zio_t *rio = arg; |
2437 | spa_load_error_t *sle = rio->io_private; | |
2438 | ||
14e4e3cb AZ |
2439 | (void) zilog, (void) dnp; |
2440 | ||
dea377c0 MA |
2441 | /* |
2442 | * Note: normally this routine will not be called if | |
2443 | * spa_load_verify_metadata is not set. However, it may be useful | |
2444 | * to manually set the flag after the traversal has begun. | |
2445 | */ | |
2446 | if (!spa_load_verify_metadata) | |
2447 | return (0); | |
2cd0f98f BB |
2448 | |
2449 | /* | |
2450 | * Sanity check the block pointer in order to detect obvious damage | |
2451 | * before using the contents in subsequent checks or in zio_read(). | |
2452 | * When damaged consider it to be a metadata error since we cannot | |
2453 | * trust the BP_GET_TYPE and BP_GET_LEVEL values. | |
2454 | */ | |
3095ca91 | 2455 | if (!zfs_blkptr_verify(spa, bp, BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) { |
2cd0f98f BB |
2456 | atomic_inc_64(&sle->sle_meta_count); |
2457 | return (0); | |
2458 | } | |
2459 | ||
2460 | if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) || | |
2461 | BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) | |
2462 | return (0); | |
2463 | ||
f2c5bc15 AM |
2464 | if (!BP_IS_METADATA(bp) && |
2465 | (!spa_load_verify_data || !sle->sle_verify_data)) | |
dea377c0 MA |
2466 | return (0); |
2467 | ||
1e527162 GW |
2468 | uint64_t maxinflight_bytes = |
2469 | arc_target_bytes() >> spa_load_verify_shift; | |
1c27024e | 2470 | size_t size = BP_GET_PSIZE(bp); |
dea377c0 MA |
2471 | |
2472 | mutex_enter(&spa->spa_scrub_lock); | |
c8242a96 | 2473 | while (spa->spa_load_verify_bytes >= maxinflight_bytes) |
dea377c0 | 2474 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); |
c8242a96 | 2475 | spa->spa_load_verify_bytes += size; |
dea377c0 MA |
2476 | mutex_exit(&spa->spa_scrub_lock); |
2477 | ||
a6255b7f | 2478 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, |
dea377c0 MA |
2479 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, |
2480 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2481 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
428870ff BB |
2482 | return (0); |
2483 | } | |
34dc7c2f | 2484 | |
65c7cc49 | 2485 | static int |
d1d19c78 PD |
2486 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
2487 | { | |
14e4e3cb AZ |
2488 | (void) dp, (void) arg; |
2489 | ||
d1d19c78 PD |
2490 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) |
2491 | return (SET_ERROR(ENAMETOOLONG)); | |
2492 | ||
2493 | return (0); | |
2494 | } | |
2495 | ||
428870ff BB |
2496 | static int |
2497 | spa_load_verify(spa_t *spa) | |
2498 | { | |
2499 | zio_t *rio; | |
2500 | spa_load_error_t sle = { 0 }; | |
8a393be3 | 2501 | zpool_load_policy_t policy; |
428870ff | 2502 | boolean_t verify_ok = B_FALSE; |
dea377c0 | 2503 | int error = 0; |
34dc7c2f | 2504 | |
8a393be3 | 2505 | zpool_get_load_policy(spa->spa_config, &policy); |
34dc7c2f | 2506 | |
f2c5bc15 AM |
2507 | if (policy.zlp_rewind & ZPOOL_NEVER_REWIND || |
2508 | policy.zlp_maxmeta == UINT64_MAX) | |
428870ff | 2509 | return (0); |
34dc7c2f | 2510 | |
d1d19c78 PD |
2511 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); |
2512 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2513 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2514 | DS_FIND_CHILDREN); | |
2515 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2516 | if (error != 0) | |
2517 | return (error); | |
2518 | ||
f2c5bc15 AM |
2519 | /* |
2520 | * Verify data only if we are rewinding or error limit was set. | |
2521 | * Otherwise nothing except dbgmsg care about it to waste time. | |
2522 | */ | |
2523 | sle.sle_verify_data = (policy.zlp_rewind & ZPOOL_REWIND_MASK) || | |
2524 | (policy.zlp_maxdata < UINT64_MAX); | |
2525 | ||
428870ff BB |
2526 | rio = zio_root(spa, NULL, &sle, |
2527 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
34dc7c2f | 2528 | |
dea377c0 | 2529 | if (spa_load_verify_metadata) { |
4a0ee12a PZ |
2530 | if (spa->spa_extreme_rewind) { |
2531 | spa_load_note(spa, "performing a complete scan of the " | |
2532 | "pool since extreme rewind is on. This may take " | |
2533 | "a very long time.\n (spa_load_verify_data=%u, " | |
2534 | "spa_load_verify_metadata=%u)", | |
2535 | spa_load_verify_data, spa_load_verify_metadata); | |
2536 | } | |
c8242a96 | 2537 | |
dea377c0 | 2538 | error = traverse_pool(spa, spa->spa_verify_min_txg, |
b5256303 TC |
2539 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | |
2540 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
dea377c0 | 2541 | } |
428870ff BB |
2542 | |
2543 | (void) zio_wait(rio); | |
c8242a96 | 2544 | ASSERT0(spa->spa_load_verify_bytes); |
428870ff BB |
2545 | |
2546 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2547 | spa->spa_load_data_errors = sle.sle_data_count; | |
2548 | ||
afd2f7b7 PZ |
2549 | if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { |
2550 | spa_load_note(spa, "spa_load_verify found %llu metadata errors " | |
2551 | "and %llu data errors", (u_longlong_t)sle.sle_meta_count, | |
2552 | (u_longlong_t)sle.sle_data_count); | |
2553 | } | |
2554 | ||
2555 | if (spa_load_verify_dryrun || | |
8a393be3 PZ |
2556 | (!error && sle.sle_meta_count <= policy.zlp_maxmeta && |
2557 | sle.sle_data_count <= policy.zlp_maxdata)) { | |
572e2857 BB |
2558 | int64_t loss = 0; |
2559 | ||
428870ff BB |
2560 | verify_ok = B_TRUE; |
2561 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2562 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
572e2857 BB |
2563 | |
2564 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
65ad5d11 AJ |
2565 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, |
2566 | spa->spa_load_txg_ts); | |
2567 | fnvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, | |
2568 | loss); | |
f2c5bc15 AM |
2569 | fnvlist_add_uint64(spa->spa_load_info, |
2570 | ZPOOL_CONFIG_LOAD_META_ERRORS, sle.sle_meta_count); | |
65ad5d11 AJ |
2571 | fnvlist_add_uint64(spa->spa_load_info, |
2572 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count); | |
428870ff BB |
2573 | } else { |
2574 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2575 | } | |
2576 | ||
afd2f7b7 PZ |
2577 | if (spa_load_verify_dryrun) |
2578 | return (0); | |
2579 | ||
428870ff BB |
2580 | if (error) { |
2581 | if (error != ENXIO && error != EIO) | |
2e528b49 | 2582 | error = SET_ERROR(EIO); |
428870ff BB |
2583 | return (error); |
2584 | } | |
2585 | ||
2586 | return (verify_ok ? 0 : EIO); | |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * Find a value in the pool props object. | |
2591 | */ | |
2592 | static void | |
2593 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2594 | { | |
2595 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2596 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2597 | } | |
2598 | ||
2599 | /* | |
2600 | * Find a value in the pool directory object. | |
2601 | */ | |
2602 | static int | |
4a0ee12a | 2603 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) |
428870ff | 2604 | { |
4a0ee12a PZ |
2605 | int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
2606 | name, sizeof (uint64_t), 1, val); | |
2607 | ||
2608 | if (error != 0 && (error != ENOENT || log_enoent)) { | |
2609 | spa_load_failed(spa, "couldn't get '%s' value in MOS directory " | |
2610 | "[error=%d]", name, error); | |
2611 | } | |
2612 | ||
2613 | return (error); | |
428870ff BB |
2614 | } |
2615 | ||
2616 | static int | |
2617 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2618 | { | |
2619 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
a1d477c2 | 2620 | return (SET_ERROR(err)); |
428870ff BB |
2621 | } |
2622 | ||
37f03da8 SH |
2623 | boolean_t |
2624 | spa_livelist_delete_check(spa_t *spa) | |
2625 | { | |
2626 | return (spa->spa_livelists_to_delete != 0); | |
2627 | } | |
2628 | ||
37f03da8 SH |
2629 | static boolean_t |
2630 | spa_livelist_delete_cb_check(void *arg, zthr_t *z) | |
2631 | { | |
14e4e3cb | 2632 | (void) z; |
37f03da8 SH |
2633 | spa_t *spa = arg; |
2634 | return (spa_livelist_delete_check(spa)); | |
2635 | } | |
2636 | ||
2637 | static int | |
2638 | delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
2639 | { | |
2640 | spa_t *spa = arg; | |
2641 | zio_free(spa, tx->tx_txg, bp); | |
2642 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, | |
2643 | -bp_get_dsize_sync(spa, bp), | |
2644 | -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); | |
2645 | return (0); | |
2646 | } | |
2647 | ||
2648 | static int | |
2649 | dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp) | |
2650 | { | |
2651 | int err; | |
2652 | zap_cursor_t zc; | |
2653 | zap_attribute_t za; | |
2654 | zap_cursor_init(&zc, os, zap_obj); | |
2655 | err = zap_cursor_retrieve(&zc, &za); | |
2656 | zap_cursor_fini(&zc); | |
2657 | if (err == 0) | |
2658 | *llp = za.za_first_integer; | |
2659 | return (err); | |
2660 | } | |
2661 | ||
2662 | /* | |
2663 | * Components of livelist deletion that must be performed in syncing | |
2664 | * context: freeing block pointers and updating the pool-wide data | |
2665 | * structures to indicate how much work is left to do | |
2666 | */ | |
2667 | typedef struct sublist_delete_arg { | |
2668 | spa_t *spa; | |
2669 | dsl_deadlist_t *ll; | |
2670 | uint64_t key; | |
2671 | bplist_t *to_free; | |
2672 | } sublist_delete_arg_t; | |
2673 | ||
2674 | static void | |
2675 | sublist_delete_sync(void *arg, dmu_tx_t *tx) | |
2676 | { | |
2677 | sublist_delete_arg_t *sda = arg; | |
2678 | spa_t *spa = sda->spa; | |
2679 | dsl_deadlist_t *ll = sda->ll; | |
2680 | uint64_t key = sda->key; | |
2681 | bplist_t *to_free = sda->to_free; | |
2682 | ||
2683 | bplist_iterate(to_free, delete_blkptr_cb, spa, tx); | |
2684 | dsl_deadlist_remove_entry(ll, key, tx); | |
2685 | } | |
2686 | ||
2687 | typedef struct livelist_delete_arg { | |
2688 | spa_t *spa; | |
2689 | uint64_t ll_obj; | |
2690 | uint64_t zap_obj; | |
2691 | } livelist_delete_arg_t; | |
2692 | ||
2693 | static void | |
2694 | livelist_delete_sync(void *arg, dmu_tx_t *tx) | |
2695 | { | |
2696 | livelist_delete_arg_t *lda = arg; | |
2697 | spa_t *spa = lda->spa; | |
2698 | uint64_t ll_obj = lda->ll_obj; | |
2699 | uint64_t zap_obj = lda->zap_obj; | |
2700 | objset_t *mos = spa->spa_meta_objset; | |
2701 | uint64_t count; | |
2702 | ||
2703 | /* free the livelist and decrement the feature count */ | |
2704 | VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx)); | |
2705 | dsl_deadlist_free(mos, ll_obj, tx); | |
2706 | spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx); | |
2707 | VERIFY0(zap_count(mos, zap_obj, &count)); | |
2708 | if (count == 0) { | |
2709 | /* no more livelists to delete */ | |
2710 | VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT, | |
2711 | DMU_POOL_DELETED_CLONES, tx)); | |
2712 | VERIFY0(zap_destroy(mos, zap_obj, tx)); | |
2713 | spa->spa_livelists_to_delete = 0; | |
e60e158e | 2714 | spa_notify_waiters(spa); |
37f03da8 SH |
2715 | } |
2716 | } | |
2717 | ||
2718 | /* | |
2719 | * Load in the value for the livelist to be removed and open it. Then, | |
2720 | * load its first sublist and determine which block pointers should actually | |
2721 | * be freed. Then, call a synctask which performs the actual frees and updates | |
2722 | * the pool-wide livelist data. | |
2723 | */ | |
65c7cc49 | 2724 | static void |
37f03da8 SH |
2725 | spa_livelist_delete_cb(void *arg, zthr_t *z) |
2726 | { | |
2727 | spa_t *spa = arg; | |
2728 | uint64_t ll_obj = 0, count; | |
2729 | objset_t *mos = spa->spa_meta_objset; | |
2730 | uint64_t zap_obj = spa->spa_livelists_to_delete; | |
2731 | /* | |
2732 | * Determine the next livelist to delete. This function should only | |
2733 | * be called if there is at least one deleted clone. | |
2734 | */ | |
2735 | VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj)); | |
2736 | VERIFY0(zap_count(mos, ll_obj, &count)); | |
2737 | if (count > 0) { | |
c9562576 | 2738 | dsl_deadlist_t *ll; |
37f03da8 SH |
2739 | dsl_deadlist_entry_t *dle; |
2740 | bplist_t to_free; | |
c9562576 PS |
2741 | ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP); |
2742 | dsl_deadlist_open(ll, mos, ll_obj); | |
2743 | dle = dsl_deadlist_first(ll); | |
37f03da8 SH |
2744 | ASSERT3P(dle, !=, NULL); |
2745 | bplist_create(&to_free); | |
2746 | int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free, | |
2747 | z, NULL); | |
2748 | if (err == 0) { | |
2749 | sublist_delete_arg_t sync_arg = { | |
2750 | .spa = spa, | |
c9562576 | 2751 | .ll = ll, |
37f03da8 SH |
2752 | .key = dle->dle_mintxg, |
2753 | .to_free = &to_free | |
2754 | }; | |
2755 | zfs_dbgmsg("deleting sublist (id %llu) from" | |
8e739b2c RE |
2756 | " livelist %llu, %lld remaining", |
2757 | (u_longlong_t)dle->dle_bpobj.bpo_object, | |
2758 | (u_longlong_t)ll_obj, (longlong_t)count - 1); | |
37f03da8 SH |
2759 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, |
2760 | sublist_delete_sync, &sync_arg, 0, | |
2761 | ZFS_SPACE_CHECK_DESTROY)); | |
2762 | } else { | |
d87676a9 | 2763 | VERIFY3U(err, ==, EINTR); |
37f03da8 SH |
2764 | } |
2765 | bplist_clear(&to_free); | |
2766 | bplist_destroy(&to_free); | |
c9562576 PS |
2767 | dsl_deadlist_close(ll); |
2768 | kmem_free(ll, sizeof (dsl_deadlist_t)); | |
37f03da8 SH |
2769 | } else { |
2770 | livelist_delete_arg_t sync_arg = { | |
2771 | .spa = spa, | |
2772 | .ll_obj = ll_obj, | |
2773 | .zap_obj = zap_obj | |
2774 | }; | |
8e739b2c RE |
2775 | zfs_dbgmsg("deletion of livelist %llu completed", |
2776 | (u_longlong_t)ll_obj); | |
37f03da8 SH |
2777 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync, |
2778 | &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY)); | |
2779 | } | |
2780 | } | |
2781 | ||
65c7cc49 | 2782 | static void |
37f03da8 SH |
2783 | spa_start_livelist_destroy_thread(spa_t *spa) |
2784 | { | |
2785 | ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL); | |
843e9ca2 SD |
2786 | spa->spa_livelist_delete_zthr = |
2787 | zthr_create("z_livelist_destroy", | |
6bc61d22 TN |
2788 | spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa, |
2789 | minclsyspri); | |
37f03da8 SH |
2790 | } |
2791 | ||
2792 | typedef struct livelist_new_arg { | |
2793 | bplist_t *allocs; | |
2794 | bplist_t *frees; | |
2795 | } livelist_new_arg_t; | |
2796 | ||
2797 | static int | |
2798 | livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
2799 | dmu_tx_t *tx) | |
2800 | { | |
2801 | ASSERT(tx == NULL); | |
2802 | livelist_new_arg_t *lna = arg; | |
2803 | if (bp_freed) { | |
2804 | bplist_append(lna->frees, bp); | |
2805 | } else { | |
2806 | bplist_append(lna->allocs, bp); | |
2807 | zfs_livelist_condense_new_alloc++; | |
2808 | } | |
2809 | return (0); | |
2810 | } | |
2811 | ||
2812 | typedef struct livelist_condense_arg { | |
2813 | spa_t *spa; | |
2814 | bplist_t to_keep; | |
2815 | uint64_t first_size; | |
2816 | uint64_t next_size; | |
2817 | } livelist_condense_arg_t; | |
2818 | ||
2819 | static void | |
2820 | spa_livelist_condense_sync(void *arg, dmu_tx_t *tx) | |
2821 | { | |
2822 | livelist_condense_arg_t *lca = arg; | |
2823 | spa_t *spa = lca->spa; | |
2824 | bplist_t new_frees; | |
2825 | dsl_dataset_t *ds = spa->spa_to_condense.ds; | |
2826 | ||
2827 | /* Have we been cancelled? */ | |
2828 | if (spa->spa_to_condense.cancelled) { | |
2829 | zfs_livelist_condense_sync_cancel++; | |
2830 | goto out; | |
2831 | } | |
2832 | ||
2833 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
2834 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
2835 | dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; | |
2836 | ||
2837 | /* | |
2838 | * It's possible that the livelist was changed while the zthr was | |
2839 | * running. Therefore, we need to check for new blkptrs in the two | |
2840 | * entries being condensed and continue to track them in the livelist. | |
2841 | * Because of the way we handle remapped blkptrs (see dbuf_remap_impl), | |
2842 | * it's possible that the newly added blkptrs are FREEs or ALLOCs so | |
2843 | * we need to sort them into two different bplists. | |
2844 | */ | |
2845 | uint64_t first_obj = first->dle_bpobj.bpo_object; | |
2846 | uint64_t next_obj = next->dle_bpobj.bpo_object; | |
2847 | uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
2848 | uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
2849 | ||
2850 | bplist_create(&new_frees); | |
2851 | livelist_new_arg_t new_bps = { | |
2852 | .allocs = &lca->to_keep, | |
2853 | .frees = &new_frees, | |
2854 | }; | |
2855 | ||
2856 | if (cur_first_size > lca->first_size) { | |
2857 | VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj, | |
2858 | livelist_track_new_cb, &new_bps, lca->first_size)); | |
2859 | } | |
2860 | if (cur_next_size > lca->next_size) { | |
2861 | VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj, | |
2862 | livelist_track_new_cb, &new_bps, lca->next_size)); | |
2863 | } | |
2864 | ||
2865 | dsl_deadlist_clear_entry(first, ll, tx); | |
2866 | ASSERT(bpobj_is_empty(&first->dle_bpobj)); | |
2867 | dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx); | |
2868 | ||
2869 | bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx); | |
2870 | bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx); | |
2871 | bplist_destroy(&new_frees); | |
2872 | ||
2873 | char dsname[ZFS_MAX_DATASET_NAME_LEN]; | |
2874 | dsl_dataset_name(ds, dsname); | |
2875 | zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu " | |
2876 | "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu " | |
8e739b2c RE |
2877 | "(%llu blkptrs)", (u_longlong_t)tx->tx_txg, dsname, |
2878 | (u_longlong_t)ds->ds_object, (u_longlong_t)first_obj, | |
2879 | (u_longlong_t)cur_first_size, (u_longlong_t)next_obj, | |
2880 | (u_longlong_t)cur_next_size, | |
2881 | (u_longlong_t)first->dle_bpobj.bpo_object, | |
2882 | (u_longlong_t)first->dle_bpobj.bpo_phys->bpo_num_blkptrs); | |
37f03da8 SH |
2883 | out: |
2884 | dmu_buf_rele(ds->ds_dbuf, spa); | |
2885 | spa->spa_to_condense.ds = NULL; | |
2886 | bplist_clear(&lca->to_keep); | |
2887 | bplist_destroy(&lca->to_keep); | |
2888 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
2889 | spa->spa_to_condense.syncing = B_FALSE; | |
2890 | } | |
2891 | ||
65c7cc49 | 2892 | static void |
37f03da8 SH |
2893 | spa_livelist_condense_cb(void *arg, zthr_t *t) |
2894 | { | |
2895 | while (zfs_livelist_condense_zthr_pause && | |
2896 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
2897 | delay(1); | |
2898 | ||
2899 | spa_t *spa = arg; | |
2900 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
2901 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
2902 | uint64_t first_size, next_size; | |
2903 | ||
2904 | livelist_condense_arg_t *lca = | |
2905 | kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP); | |
2906 | bplist_create(&lca->to_keep); | |
2907 | ||
2908 | /* | |
2909 | * Process the livelists (matching FREEs and ALLOCs) in open context | |
2910 | * so we have minimal work in syncing context to condense. | |
2911 | * | |
2912 | * We save bpobj sizes (first_size and next_size) to use later in | |
2913 | * syncing context to determine if entries were added to these sublists | |
2914 | * while in open context. This is possible because the clone is still | |
2915 | * active and open for normal writes and we want to make sure the new, | |
2916 | * unprocessed blockpointers are inserted into the livelist normally. | |
2917 | * | |
2918 | * Note that dsl_process_sub_livelist() both stores the size number of | |
2919 | * blockpointers and iterates over them while the bpobj's lock held, so | |
2920 | * the sizes returned to us are consistent which what was actually | |
2921 | * processed. | |
2922 | */ | |
2923 | int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t, | |
2924 | &first_size); | |
2925 | if (err == 0) | |
2926 | err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep, | |
2927 | t, &next_size); | |
2928 | ||
2929 | if (err == 0) { | |
2930 | while (zfs_livelist_condense_sync_pause && | |
2931 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
2932 | delay(1); | |
2933 | ||
2934 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
2935 | dmu_tx_mark_netfree(tx); | |
2936 | dmu_tx_hold_space(tx, 1); | |
2937 | err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE); | |
2938 | if (err == 0) { | |
2939 | /* | |
2940 | * Prevent the condense zthr restarting before | |
2941 | * the synctask completes. | |
2942 | */ | |
2943 | spa->spa_to_condense.syncing = B_TRUE; | |
2944 | lca->spa = spa; | |
2945 | lca->first_size = first_size; | |
2946 | lca->next_size = next_size; | |
2947 | dsl_sync_task_nowait(spa_get_dsl(spa), | |
38080324 | 2948 | spa_livelist_condense_sync, lca, tx); |
37f03da8 SH |
2949 | dmu_tx_commit(tx); |
2950 | return; | |
2951 | } | |
2952 | } | |
2953 | /* | |
2954 | * Condensing can not continue: either it was externally stopped or | |
2955 | * we were unable to assign to a tx because the pool has run out of | |
2956 | * space. In the second case, we'll just end up trying to condense | |
2957 | * again in a later txg. | |
2958 | */ | |
2959 | ASSERT(err != 0); | |
2960 | bplist_clear(&lca->to_keep); | |
2961 | bplist_destroy(&lca->to_keep); | |
2962 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
2963 | dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa); | |
2964 | spa->spa_to_condense.ds = NULL; | |
2965 | if (err == EINTR) | |
2966 | zfs_livelist_condense_zthr_cancel++; | |
2967 | } | |
2968 | ||
37f03da8 SH |
2969 | /* |
2970 | * Check that there is something to condense but that a condense is not | |
2971 | * already in progress and that condensing has not been cancelled. | |
2972 | */ | |
2973 | static boolean_t | |
2974 | spa_livelist_condense_cb_check(void *arg, zthr_t *z) | |
2975 | { | |
14e4e3cb | 2976 | (void) z; |
37f03da8 SH |
2977 | spa_t *spa = arg; |
2978 | if ((spa->spa_to_condense.ds != NULL) && | |
2979 | (spa->spa_to_condense.syncing == B_FALSE) && | |
2980 | (spa->spa_to_condense.cancelled == B_FALSE)) { | |
2981 | return (B_TRUE); | |
2982 | } | |
2983 | return (B_FALSE); | |
2984 | } | |
2985 | ||
65c7cc49 | 2986 | static void |
37f03da8 SH |
2987 | spa_start_livelist_condensing_thread(spa_t *spa) |
2988 | { | |
2989 | spa->spa_to_condense.ds = NULL; | |
2990 | spa->spa_to_condense.first = NULL; | |
2991 | spa->spa_to_condense.next = NULL; | |
2992 | spa->spa_to_condense.syncing = B_FALSE; | |
2993 | spa->spa_to_condense.cancelled = B_FALSE; | |
2994 | ||
2995 | ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL); | |
843e9ca2 SD |
2996 | spa->spa_livelist_condense_zthr = |
2997 | zthr_create("z_livelist_condense", | |
2998 | spa_livelist_condense_cb_check, | |
6bc61d22 | 2999 | spa_livelist_condense_cb, spa, minclsyspri); |
37f03da8 SH |
3000 | } |
3001 | ||
9d5b5245 SD |
3002 | static void |
3003 | spa_spawn_aux_threads(spa_t *spa) | |
3004 | { | |
3005 | ASSERT(spa_writeable(spa)); | |
3006 | ||
3007 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
3008 | ||
5caeef02 | 3009 | spa_start_raidz_expansion_thread(spa); |
9d5b5245 | 3010 | spa_start_indirect_condensing_thread(spa); |
37f03da8 SH |
3011 | spa_start_livelist_destroy_thread(spa); |
3012 | spa_start_livelist_condensing_thread(spa); | |
d2734cce SD |
3013 | |
3014 | ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); | |
3015 | spa->spa_checkpoint_discard_zthr = | |
843e9ca2 SD |
3016 | zthr_create("z_checkpoint_discard", |
3017 | spa_checkpoint_discard_thread_check, | |
6bc61d22 | 3018 | spa_checkpoint_discard_thread, spa, minclsyspri); |
9d5b5245 SD |
3019 | } |
3020 | ||
428870ff BB |
3021 | /* |
3022 | * Fix up config after a partly-completed split. This is done with the | |
3023 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
3024 | * pool have that entry in their config, but only the splitting one contains | |
3025 | * a list of all the guids of the vdevs that are being split off. | |
3026 | * | |
3027 | * This function determines what to do with that list: either rejoin | |
3028 | * all the disks to the pool, or complete the splitting process. To attempt | |
3029 | * the rejoin, each disk that is offlined is marked online again, and | |
3030 | * we do a reopen() call. If the vdev label for every disk that was | |
3031 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
3032 | * then we call vdev_split() on each disk, and complete the split. | |
3033 | * | |
3034 | * Otherwise we leave the config alone, with all the vdevs in place in | |
3035 | * the original pool. | |
3036 | */ | |
3037 | static void | |
3038 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
3039 | { | |
3040 | uint_t extracted; | |
3041 | uint64_t *glist; | |
3042 | uint_t i, gcount; | |
3043 | nvlist_t *nvl; | |
3044 | vdev_t **vd; | |
3045 | boolean_t attempt_reopen; | |
3046 | ||
3047 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
3048 | return; | |
3049 | ||
3050 | /* check that the config is complete */ | |
3051 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
3052 | &glist, &gcount) != 0) | |
3053 | return; | |
3054 | ||
79c76d5b | 3055 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); |
428870ff BB |
3056 | |
3057 | /* attempt to online all the vdevs & validate */ | |
3058 | attempt_reopen = B_TRUE; | |
3059 | for (i = 0; i < gcount; i++) { | |
3060 | if (glist[i] == 0) /* vdev is hole */ | |
3061 | continue; | |
3062 | ||
3063 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
3064 | if (vd[i] == NULL) { | |
3065 | /* | |
3066 | * Don't bother attempting to reopen the disks; | |
3067 | * just do the split. | |
3068 | */ | |
3069 | attempt_reopen = B_FALSE; | |
3070 | } else { | |
3071 | /* attempt to re-online it */ | |
3072 | vd[i]->vdev_offline = B_FALSE; | |
3073 | } | |
3074 | } | |
3075 | ||
3076 | if (attempt_reopen) { | |
3077 | vdev_reopen(spa->spa_root_vdev); | |
3078 | ||
3079 | /* check each device to see what state it's in */ | |
3080 | for (extracted = 0, i = 0; i < gcount; i++) { | |
3081 | if (vd[i] != NULL && | |
3082 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
3083 | break; | |
3084 | ++extracted; | |
3085 | } | |
3086 | } | |
3087 | ||
3088 | /* | |
3089 | * If every disk has been moved to the new pool, or if we never | |
3090 | * even attempted to look at them, then we split them off for | |
3091 | * good. | |
3092 | */ | |
3093 | if (!attempt_reopen || gcount == extracted) { | |
3094 | for (i = 0; i < gcount; i++) | |
3095 | if (vd[i] != NULL) | |
3096 | vdev_split(vd[i]); | |
3097 | vdev_reopen(spa->spa_root_vdev); | |
3098 | } | |
3099 | ||
3100 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
3101 | } | |
3102 | ||
3103 | static int | |
6cb8e530 | 3104 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) |
428870ff | 3105 | { |
a926aab9 | 3106 | const char *ereport = FM_EREPORT_ZFS_POOL; |
428870ff | 3107 | int error; |
428870ff | 3108 | |
6cb8e530 | 3109 | spa->spa_load_state = state; |
ca95f70d OF |
3110 | (void) spa_import_progress_set_state(spa_guid(spa), |
3111 | spa_load_state(spa)); | |
687e4d7f | 3112 | spa_import_progress_set_notes(spa, "spa_load()"); |
9ae529ec | 3113 | |
6cb8e530 | 3114 | gethrestime(&spa->spa_loaded_ts); |
d2734cce | 3115 | error = spa_load_impl(spa, type, &ereport); |
428870ff | 3116 | |
0c66c32d JG |
3117 | /* |
3118 | * Don't count references from objsets that are already closed | |
3119 | * and are making their way through the eviction process. | |
3120 | */ | |
3121 | spa_evicting_os_wait(spa); | |
424fd7c3 | 3122 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
572e2857 BB |
3123 | if (error) { |
3124 | if (error != EEXIST) { | |
3125 | spa->spa_loaded_ts.tv_sec = 0; | |
3126 | spa->spa_loaded_ts.tv_nsec = 0; | |
3127 | } | |
3128 | if (error != EBADF) { | |
1144586b | 3129 | (void) zfs_ereport_post(ereport, spa, |
4f072827 | 3130 | NULL, NULL, NULL, 0); |
572e2857 BB |
3131 | } |
3132 | } | |
428870ff BB |
3133 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; |
3134 | spa->spa_ena = 0; | |
3135 | ||
ca95f70d OF |
3136 | (void) spa_import_progress_set_state(spa_guid(spa), |
3137 | spa_load_state(spa)); | |
3138 | ||
428870ff BB |
3139 | return (error); |
3140 | } | |
3141 | ||
33cf67cd | 3142 | #ifdef ZFS_DEBUG |
e0ab3ab5 JS |
3143 | /* |
3144 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
3145 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
3146 | * spa's per-vdev ZAP list. | |
3147 | */ | |
3148 | static uint64_t | |
3149 | vdev_count_verify_zaps(vdev_t *vd) | |
3150 | { | |
3151 | spa_t *spa = vd->vdev_spa; | |
3152 | uint64_t total = 0; | |
e0ab3ab5 | 3153 | |
3e4ed421 RW |
3154 | if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2) && |
3155 | vd->vdev_root_zap != 0) { | |
3156 | total++; | |
3157 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3158 | spa->spa_all_vdev_zaps, vd->vdev_root_zap)); | |
3159 | } | |
e0ab3ab5 JS |
3160 | if (vd->vdev_top_zap != 0) { |
3161 | total++; | |
3162 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3163 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
3164 | } | |
3165 | if (vd->vdev_leaf_zap != 0) { | |
3166 | total++; | |
3167 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3168 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
3169 | } | |
3170 | ||
1c27024e | 3171 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
3172 | total += vdev_count_verify_zaps(vd->vdev_child[i]); |
3173 | } | |
3174 | ||
3175 | return (total); | |
3176 | } | |
36542b06 AZ |
3177 | #else |
3178 | #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0) | |
33cf67cd | 3179 | #endif |
e0ab3ab5 | 3180 | |
379ca9cf OF |
3181 | /* |
3182 | * Determine whether the activity check is required. | |
3183 | */ | |
3184 | static boolean_t | |
bbffb59e BB |
3185 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, |
3186 | nvlist_t *config) | |
379ca9cf OF |
3187 | { |
3188 | uint64_t state = 0; | |
3189 | uint64_t hostid = 0; | |
3190 | uint64_t tryconfig_txg = 0; | |
3191 | uint64_t tryconfig_timestamp = 0; | |
060f0226 | 3192 | uint16_t tryconfig_mmp_seq = 0; |
379ca9cf OF |
3193 | nvlist_t *nvinfo; |
3194 | ||
3195 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3196 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
3197 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
3198 | &tryconfig_txg); | |
3199 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
3200 | &tryconfig_timestamp); | |
060f0226 OF |
3201 | (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ, |
3202 | &tryconfig_mmp_seq); | |
379ca9cf OF |
3203 | } |
3204 | ||
3205 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
379ca9cf OF |
3206 | |
3207 | /* | |
3208 | * Disable the MMP activity check - This is used by zdb which | |
3209 | * is intended to be used on potentially active pools. | |
3210 | */ | |
3211 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
3212 | return (B_FALSE); | |
3213 | ||
3214 | /* | |
3215 | * Skip the activity check when the MMP feature is disabled. | |
3216 | */ | |
3217 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
3218 | return (B_FALSE); | |
ca95f70d | 3219 | |
379ca9cf | 3220 | /* |
060f0226 OF |
3221 | * If the tryconfig_ values are nonzero, they are the results of an |
3222 | * earlier tryimport. If they all match the uberblock we just found, | |
3223 | * then the pool has not changed and we return false so we do not test | |
3224 | * a second time. | |
379ca9cf OF |
3225 | */ |
3226 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
060f0226 OF |
3227 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp && |
3228 | tryconfig_mmp_seq && tryconfig_mmp_seq == | |
3229 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) | |
379ca9cf OF |
3230 | return (B_FALSE); |
3231 | ||
3232 | /* | |
3233 | * Allow the activity check to be skipped when importing the pool | |
bbffb59e BB |
3234 | * on the same host which last imported it. Since the hostid from |
3235 | * configuration may be stale use the one read from the label. | |
379ca9cf | 3236 | */ |
bbffb59e BB |
3237 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) |
3238 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
3239 | ||
25f06d67 | 3240 | if (hostid == spa_get_hostid(spa)) |
379ca9cf OF |
3241 | return (B_FALSE); |
3242 | ||
3243 | /* | |
3244 | * Skip the activity test when the pool was cleanly exported. | |
3245 | */ | |
3246 | if (state != POOL_STATE_ACTIVE) | |
3247 | return (B_FALSE); | |
3248 | ||
3249 | return (B_TRUE); | |
3250 | } | |
3251 | ||
060f0226 OF |
3252 | /* |
3253 | * Nanoseconds the activity check must watch for changes on-disk. | |
3254 | */ | |
3255 | static uint64_t | |
3256 | spa_activity_check_duration(spa_t *spa, uberblock_t *ub) | |
3257 | { | |
3258 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
3259 | uint64_t multihost_interval = MSEC2NSEC( | |
3260 | MMP_INTERVAL_OK(zfs_multihost_interval)); | |
3261 | uint64_t import_delay = MAX(NANOSEC, import_intervals * | |
3262 | multihost_interval); | |
3263 | ||
3264 | /* | |
3265 | * Local tunables determine a minimum duration except for the case | |
3266 | * where we know when the remote host will suspend the pool if MMP | |
3267 | * writes do not land. | |
3268 | * | |
3269 | * See Big Theory comment at the top of mmp.c for the reasoning behind | |
3270 | * these cases and times. | |
3271 | */ | |
3272 | ||
3273 | ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100); | |
3274 | ||
3275 | if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3276 | MMP_FAIL_INT(ub) > 0) { | |
3277 | ||
3278 | /* MMP on remote host will suspend pool after failed writes */ | |
3279 | import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) * | |
3280 | MMP_IMPORT_SAFETY_FACTOR / 100; | |
3281 | ||
3282 | zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp " | |
3283 | "mmp_fails=%llu ub_mmp mmp_interval=%llu " | |
8e739b2c RE |
3284 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3285 | (u_longlong_t)MMP_FAIL_INT(ub), | |
3286 | (u_longlong_t)MMP_INTERVAL(ub), | |
3287 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3288 | |
3289 | } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3290 | MMP_FAIL_INT(ub) == 0) { | |
3291 | ||
3292 | /* MMP on remote host will never suspend pool */ | |
3293 | import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) + | |
3294 | ub->ub_mmp_delay) * import_intervals); | |
3295 | ||
3296 | zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp " | |
3297 | "mmp_interval=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3298 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3299 | (u_longlong_t)MMP_INTERVAL(ub), | |
3300 | (u_longlong_t)ub->ub_mmp_delay, | |
3301 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3302 | |
3303 | } else if (MMP_VALID(ub)) { | |
3304 | /* | |
e1cfd73f | 3305 | * zfs-0.7 compatibility case |
060f0226 OF |
3306 | */ |
3307 | ||
3308 | import_delay = MAX(import_delay, (multihost_interval + | |
3309 | ub->ub_mmp_delay) * import_intervals); | |
3310 | ||
3311 | zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu " | |
8e739b2c RE |
3312 | "import_intervals=%llu leaves=%u", |
3313 | (u_longlong_t)import_delay, | |
3314 | (u_longlong_t)ub->ub_mmp_delay, | |
3315 | (u_longlong_t)import_intervals, | |
060f0226 OF |
3316 | vdev_count_leaves(spa)); |
3317 | } else { | |
3318 | /* Using local tunings is the only reasonable option */ | |
3319 | zfs_dbgmsg("pool last imported on non-MMP aware " | |
3320 | "host using import_delay=%llu multihost_interval=%llu " | |
8e739b2c RE |
3321 | "import_intervals=%llu", (u_longlong_t)import_delay, |
3322 | (u_longlong_t)multihost_interval, | |
3323 | (u_longlong_t)import_intervals); | |
060f0226 OF |
3324 | } |
3325 | ||
3326 | return (import_delay); | |
3327 | } | |
3328 | ||
379ca9cf OF |
3329 | /* |
3330 | * Perform the import activity check. If the user canceled the import or | |
3331 | * we detected activity then fail. | |
3332 | */ | |
3333 | static int | |
3334 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config) | |
3335 | { | |
379ca9cf OF |
3336 | uint64_t txg = ub->ub_txg; |
3337 | uint64_t timestamp = ub->ub_timestamp; | |
060f0226 OF |
3338 | uint64_t mmp_config = ub->ub_mmp_config; |
3339 | uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0; | |
3340 | uint64_t import_delay; | |
687e4d7f | 3341 | hrtime_t import_expire, now; |
379ca9cf OF |
3342 | nvlist_t *mmp_label = NULL; |
3343 | vdev_t *rvd = spa->spa_root_vdev; | |
3344 | kcondvar_t cv; | |
3345 | kmutex_t mtx; | |
3346 | int error = 0; | |
3347 | ||
3348 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
3349 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
3350 | mutex_enter(&mtx); | |
3351 | ||
3352 | /* | |
3353 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
3354 | * during the earlier tryimport. If the txg recorded there is 0 then | |
3355 | * the pool is known to be active on another host. | |
3356 | * | |
060f0226 | 3357 | * Otherwise, the pool might be in use on another host. Check for |
379ca9cf OF |
3358 | * changes in the uberblocks on disk if necessary. |
3359 | */ | |
3360 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3361 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
3362 | ZPOOL_CONFIG_LOAD_INFO); | |
3363 | ||
3364 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
3365 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
3366 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
3367 | error = SET_ERROR(EREMOTEIO); | |
3368 | goto out; | |
3369 | } | |
3370 | } | |
3371 | ||
060f0226 | 3372 | import_delay = spa_activity_check_duration(spa, ub); |
533ea041 | 3373 | |
379ca9cf | 3374 | /* Add a small random factor in case of simultaneous imports (0-25%) */ |
29274c9f | 3375 | import_delay += import_delay * random_in_range(250) / 1000; |
ca95f70d OF |
3376 | |
3377 | import_expire = gethrtime() + import_delay; | |
379ca9cf | 3378 | |
687e4d7f DB |
3379 | spa_import_progress_set_notes(spa, "Checking MMP activity, waiting " |
3380 | "%llu ms", (u_longlong_t)NSEC2MSEC(import_delay)); | |
3381 | ||
3382 | int interations = 0; | |
3383 | while ((now = gethrtime()) < import_expire) { | |
3384 | if (interations++ % 30 == 0) { | |
3385 | spa_import_progress_set_notes(spa, "Checking MMP " | |
3386 | "activity, %llu ms remaining", | |
3387 | (u_longlong_t)NSEC2MSEC(import_expire - now)); | |
3388 | } | |
3389 | ||
ca95f70d OF |
3390 | (void) spa_import_progress_set_mmp_check(spa_guid(spa), |
3391 | NSEC2SEC(import_expire - gethrtime())); | |
3392 | ||
379ca9cf OF |
3393 | vdev_uberblock_load(rvd, ub, &mmp_label); |
3394 | ||
060f0226 OF |
3395 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp || |
3396 | mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) { | |
3397 | zfs_dbgmsg("multihost activity detected " | |
3398 | "txg %llu ub_txg %llu " | |
3399 | "timestamp %llu ub_timestamp %llu " | |
3400 | "mmp_config %#llx ub_mmp_config %#llx", | |
8e739b2c RE |
3401 | (u_longlong_t)txg, (u_longlong_t)ub->ub_txg, |
3402 | (u_longlong_t)timestamp, | |
3403 | (u_longlong_t)ub->ub_timestamp, | |
3404 | (u_longlong_t)mmp_config, | |
3405 | (u_longlong_t)ub->ub_mmp_config); | |
060f0226 | 3406 | |
379ca9cf OF |
3407 | error = SET_ERROR(EREMOTEIO); |
3408 | break; | |
3409 | } | |
3410 | ||
3411 | if (mmp_label) { | |
3412 | nvlist_free(mmp_label); | |
3413 | mmp_label = NULL; | |
3414 | } | |
3415 | ||
3416 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
3417 | if (error != -1) { | |
3418 | error = SET_ERROR(EINTR); | |
3419 | break; | |
3420 | } | |
3421 | error = 0; | |
3422 | } | |
3423 | ||
3424 | out: | |
3425 | mutex_exit(&mtx); | |
3426 | mutex_destroy(&mtx); | |
3427 | cv_destroy(&cv); | |
3428 | ||
3429 | /* | |
3430 | * If the pool is determined to be active store the status in the | |
3431 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
3432 | * available from configuration read from disk store them as well. | |
3433 | * This allows 'zpool import' to generate a more useful message. | |
3434 | * | |
3435 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
3436 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
3437 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
3438 | */ | |
3439 | if (error == EREMOTEIO) { | |
a926aab9 | 3440 | const char *hostname = "<unknown>"; |
379ca9cf OF |
3441 | uint64_t hostid = 0; |
3442 | ||
3443 | if (mmp_label) { | |
3444 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
3445 | hostname = fnvlist_lookup_string(mmp_label, | |
3446 | ZPOOL_CONFIG_HOSTNAME); | |
3447 | fnvlist_add_string(spa->spa_load_info, | |
3448 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
3449 | } | |
3450 | ||
3451 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
3452 | hostid = fnvlist_lookup_uint64(mmp_label, | |
3453 | ZPOOL_CONFIG_HOSTID); | |
3454 | fnvlist_add_uint64(spa->spa_load_info, | |
3455 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
3456 | } | |
3457 | } | |
3458 | ||
3459 | fnvlist_add_uint64(spa->spa_load_info, | |
3460 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
3461 | fnvlist_add_uint64(spa->spa_load_info, | |
3462 | ZPOOL_CONFIG_MMP_TXG, 0); | |
3463 | ||
3464 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
3465 | } | |
3466 | ||
3467 | if (mmp_label) | |
3468 | nvlist_free(mmp_label); | |
3469 | ||
3470 | return (error); | |
3471 | } | |
3472 | ||
9eb7b46e | 3473 | static int |
6cb8e530 PZ |
3474 | spa_verify_host(spa_t *spa, nvlist_t *mos_config) |
3475 | { | |
3476 | uint64_t hostid; | |
d1807f16 | 3477 | const char *hostname; |
6cb8e530 PZ |
3478 | uint64_t myhostid = 0; |
3479 | ||
3480 | if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, | |
3481 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
3482 | hostname = fnvlist_lookup_string(mos_config, | |
3483 | ZPOOL_CONFIG_HOSTNAME); | |
3484 | ||
3485 | myhostid = zone_get_hostid(NULL); | |
3486 | ||
3487 | if (hostid != 0 && myhostid != 0 && hostid != myhostid) { | |
3488 | cmn_err(CE_WARN, "pool '%s' could not be " | |
3489 | "loaded as it was last accessed by " | |
3490 | "another system (host: %s hostid: 0x%llx). " | |
a2f944a1 RM |
3491 | "See: https://openzfs.github.io/openzfs-docs/msg/" |
3492 | "ZFS-8000-EY", | |
6cb8e530 PZ |
3493 | spa_name(spa), hostname, (u_longlong_t)hostid); |
3494 | spa_load_failed(spa, "hostid verification failed: pool " | |
3495 | "last accessed by host: %s (hostid: 0x%llx)", | |
3496 | hostname, (u_longlong_t)hostid); | |
3497 | return (SET_ERROR(EBADF)); | |
3498 | } | |
3499 | } | |
3500 | ||
3501 | return (0); | |
3502 | } | |
3503 | ||
3504 | static int | |
3505 | spa_ld_parse_config(spa_t *spa, spa_import_type_t type) | |
428870ff BB |
3506 | { |
3507 | int error = 0; | |
6cb8e530 | 3508 | nvlist_t *nvtree, *nvl, *config = spa->spa_config; |
1c27024e | 3509 | int parse; |
9eb7b46e | 3510 | vdev_t *rvd; |
6cb8e530 | 3511 | uint64_t pool_guid; |
d1807f16 RY |
3512 | const char *comment; |
3513 | const char *compatibility; | |
6cb8e530 PZ |
3514 | |
3515 | /* | |
3516 | * Versioning wasn't explicitly added to the label until later, so if | |
3517 | * it's not present treat it as the initial version. | |
3518 | */ | |
3519 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
3520 | &spa->spa_ubsync.ub_version) != 0) | |
3521 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
3522 | ||
3523 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { | |
3524 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3525 | ZPOOL_CONFIG_POOL_GUID); | |
3526 | return (SET_ERROR(EINVAL)); | |
3527 | } | |
3528 | ||
d2734cce SD |
3529 | /* |
3530 | * If we are doing an import, ensure that the pool is not already | |
3531 | * imported by checking if its pool guid already exists in the | |
3532 | * spa namespace. | |
3533 | * | |
3534 | * The only case that we allow an already imported pool to be | |
3535 | * imported again, is when the pool is checkpointed and we want to | |
3536 | * look at its checkpointed state from userland tools like zdb. | |
3537 | */ | |
3538 | #ifdef _KERNEL | |
3539 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3540 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3541 | spa_guid_exists(pool_guid, 0)) { | |
3542 | #else | |
3543 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3544 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3545 | spa_guid_exists(pool_guid, 0) && | |
3546 | !spa_importing_readonly_checkpoint(spa)) { | |
3547 | #endif | |
6cb8e530 PZ |
3548 | spa_load_failed(spa, "a pool with guid %llu is already open", |
3549 | (u_longlong_t)pool_guid); | |
3550 | return (SET_ERROR(EEXIST)); | |
3551 | } | |
3552 | ||
3553 | spa->spa_config_guid = pool_guid; | |
3554 | ||
3555 | nvlist_free(spa->spa_load_info); | |
3556 | spa->spa_load_info = fnvlist_alloc(); | |
3557 | ||
3558 | ASSERT(spa->spa_comment == NULL); | |
3559 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
3560 | spa->spa_comment = spa_strdup(comment); | |
3561 | ||
658fb802 CB |
3562 | ASSERT(spa->spa_compatibility == NULL); |
3563 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMPATIBILITY, | |
3564 | &compatibility) == 0) | |
3565 | spa->spa_compatibility = spa_strdup(compatibility); | |
3566 | ||
6cb8e530 PZ |
3567 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
3568 | &spa->spa_config_txg); | |
3569 | ||
3570 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) | |
3571 | spa->spa_config_splitting = fnvlist_dup(nvl); | |
428870ff | 3572 | |
4a0ee12a PZ |
3573 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { |
3574 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3575 | ZPOOL_CONFIG_VDEV_TREE); | |
2e528b49 | 3576 | return (SET_ERROR(EINVAL)); |
4a0ee12a | 3577 | } |
428870ff | 3578 | |
428870ff BB |
3579 | /* |
3580 | * Create "The Godfather" zio to hold all async IOs | |
3581 | */ | |
e022864d MA |
3582 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
3583 | KM_SLEEP); | |
1c27024e | 3584 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
3585 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
3586 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
3587 | ZIO_FLAG_GODFATHER); | |
3588 | } | |
428870ff BB |
3589 | |
3590 | /* | |
3591 | * Parse the configuration into a vdev tree. We explicitly set the | |
3592 | * value that will be returned by spa_version() since parsing the | |
3593 | * configuration requires knowing the version number. | |
3594 | */ | |
3595 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 PZ |
3596 | parse = (type == SPA_IMPORT_EXISTING ? |
3597 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
9eb7b46e | 3598 | error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); |
428870ff BB |
3599 | spa_config_exit(spa, SCL_ALL, FTAG); |
3600 | ||
4a0ee12a PZ |
3601 | if (error != 0) { |
3602 | spa_load_failed(spa, "unable to parse config [error=%d]", | |
3603 | error); | |
428870ff | 3604 | return (error); |
4a0ee12a | 3605 | } |
428870ff BB |
3606 | |
3607 | ASSERT(spa->spa_root_vdev == rvd); | |
c3520e7f MA |
3608 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); |
3609 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
428870ff BB |
3610 | |
3611 | if (type != SPA_IMPORT_ASSEMBLE) { | |
3612 | ASSERT(spa_guid(spa) == pool_guid); | |
3613 | } | |
3614 | ||
9eb7b46e PZ |
3615 | return (0); |
3616 | } | |
3617 | ||
6cb8e530 PZ |
3618 | /* |
3619 | * Recursively open all vdevs in the vdev tree. This function is called twice: | |
3620 | * first with the untrusted config, then with the trusted config. | |
3621 | */ | |
9eb7b46e PZ |
3622 | static int |
3623 | spa_ld_open_vdevs(spa_t *spa) | |
3624 | { | |
3625 | int error = 0; | |
3626 | ||
6cb8e530 PZ |
3627 | /* |
3628 | * spa_missing_tvds_allowed defines how many top-level vdevs can be | |
3629 | * missing/unopenable for the root vdev to be still considered openable. | |
3630 | */ | |
3631 | if (spa->spa_trust_config) { | |
3632 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; | |
3633 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { | |
3634 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; | |
3635 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { | |
3636 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; | |
3637 | } else { | |
3638 | spa->spa_missing_tvds_allowed = 0; | |
3639 | } | |
3640 | ||
3641 | spa->spa_missing_tvds_allowed = | |
3642 | MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); | |
3643 | ||
428870ff | 3644 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9eb7b46e | 3645 | error = vdev_open(spa->spa_root_vdev); |
428870ff | 3646 | spa_config_exit(spa, SCL_ALL, FTAG); |
6cb8e530 PZ |
3647 | |
3648 | if (spa->spa_missing_tvds != 0) { | |
3649 | spa_load_note(spa, "vdev tree has %lld missing top-level " | |
3650 | "vdevs.", (u_longlong_t)spa->spa_missing_tvds); | |
da92d5cb | 3651 | if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) { |
6cb8e530 PZ |
3652 | /* |
3653 | * Although theoretically we could allow users to open | |
3654 | * incomplete pools in RW mode, we'd need to add a lot | |
3655 | * of extra logic (e.g. adjust pool space to account | |
3656 | * for missing vdevs). | |
3657 | * This limitation also prevents users from accidentally | |
3658 | * opening the pool in RW mode during data recovery and | |
3659 | * damaging it further. | |
3660 | */ | |
3661 | spa_load_note(spa, "pools with missing top-level " | |
3662 | "vdevs can only be opened in read-only mode."); | |
3663 | error = SET_ERROR(ENXIO); | |
3664 | } else { | |
3665 | spa_load_note(spa, "current settings allow for maximum " | |
3666 | "%lld missing top-level vdevs at this stage.", | |
3667 | (u_longlong_t)spa->spa_missing_tvds_allowed); | |
3668 | } | |
3669 | } | |
4a0ee12a PZ |
3670 | if (error != 0) { |
3671 | spa_load_failed(spa, "unable to open vdev tree [error=%d]", | |
3672 | error); | |
3673 | } | |
6cb8e530 PZ |
3674 | if (spa->spa_missing_tvds != 0 || error != 0) |
3675 | vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); | |
9eb7b46e PZ |
3676 | |
3677 | return (error); | |
3678 | } | |
3679 | ||
6cb8e530 PZ |
3680 | /* |
3681 | * We need to validate the vdev labels against the configuration that | |
3682 | * we have in hand. This function is called twice: first with an untrusted | |
3683 | * config, then with a trusted config. The validation is more strict when the | |
3684 | * config is trusted. | |
3685 | */ | |
9eb7b46e | 3686 | static int |
6cb8e530 | 3687 | spa_ld_validate_vdevs(spa_t *spa) |
9eb7b46e PZ |
3688 | { |
3689 | int error = 0; | |
3690 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 3691 | |
6cb8e530 PZ |
3692 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
3693 | error = vdev_validate(rvd); | |
3694 | spa_config_exit(spa, SCL_ALL, FTAG); | |
428870ff | 3695 | |
6cb8e530 PZ |
3696 | if (error != 0) { |
3697 | spa_load_failed(spa, "vdev_validate failed [error=%d]", error); | |
3698 | return (error); | |
3699 | } | |
428870ff | 3700 | |
6cb8e530 PZ |
3701 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
3702 | spa_load_failed(spa, "cannot open vdev tree after invalidating " | |
3703 | "some vdevs"); | |
3704 | vdev_dbgmsg_print_tree(rvd, 2); | |
3705 | return (SET_ERROR(ENXIO)); | |
428870ff BB |
3706 | } |
3707 | ||
9eb7b46e PZ |
3708 | return (0); |
3709 | } | |
3710 | ||
d2734cce SD |
3711 | static void |
3712 | spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) | |
3713 | { | |
3714 | spa->spa_state = POOL_STATE_ACTIVE; | |
3715 | spa->spa_ubsync = spa->spa_uberblock; | |
3716 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
3717 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
3718 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
3719 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
3720 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
3721 | spa->spa_prev_software_version = ub->ub_software_version; | |
3722 | } | |
3723 | ||
9eb7b46e | 3724 | static int |
6cb8e530 | 3725 | spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) |
9eb7b46e PZ |
3726 | { |
3727 | vdev_t *rvd = spa->spa_root_vdev; | |
3728 | nvlist_t *label; | |
3729 | uberblock_t *ub = &spa->spa_uberblock; | |
9eb7b46e PZ |
3730 | boolean_t activity_check = B_FALSE; |
3731 | ||
d2734cce SD |
3732 | /* |
3733 | * If we are opening the checkpointed state of the pool by | |
3734 | * rewinding to it, at this point we will have written the | |
3735 | * checkpointed uberblock to the vdev labels, so searching | |
3736 | * the labels will find the right uberblock. However, if | |
3737 | * we are opening the checkpointed state read-only, we have | |
3738 | * not modified the labels. Therefore, we must ignore the | |
3739 | * labels and continue using the spa_uberblock that was set | |
3740 | * by spa_ld_checkpoint_rewind. | |
3741 | * | |
3742 | * Note that it would be fine to ignore the labels when | |
3743 | * rewinding (opening writeable) as well. However, if we | |
3744 | * crash just after writing the labels, we will end up | |
3745 | * searching the labels. Doing so in the common case means | |
3746 | * that this code path gets exercised normally, rather than | |
3747 | * just in the edge case. | |
3748 | */ | |
3749 | if (ub->ub_checkpoint_txg != 0 && | |
3750 | spa_importing_readonly_checkpoint(spa)) { | |
3751 | spa_ld_select_uberblock_done(spa, ub); | |
3752 | return (0); | |
3753 | } | |
3754 | ||
428870ff BB |
3755 | /* |
3756 | * Find the best uberblock. | |
3757 | */ | |
9ae529ec | 3758 | vdev_uberblock_load(rvd, ub, &label); |
428870ff BB |
3759 | |
3760 | /* | |
3761 | * If we weren't able to find a single valid uberblock, return failure. | |
3762 | */ | |
9ae529ec CS |
3763 | if (ub->ub_txg == 0) { |
3764 | nvlist_free(label); | |
4a0ee12a | 3765 | spa_load_failed(spa, "no valid uberblock found"); |
428870ff | 3766 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); |
9ae529ec | 3767 | } |
428870ff | 3768 | |
ca95f70d OF |
3769 | if (spa->spa_load_max_txg != UINT64_MAX) { |
3770 | (void) spa_import_progress_set_max_txg(spa_guid(spa), | |
3771 | (u_longlong_t)spa->spa_load_max_txg); | |
3772 | } | |
4a0ee12a PZ |
3773 | spa_load_note(spa, "using uberblock with txg=%llu", |
3774 | (u_longlong_t)ub->ub_txg); | |
5caeef02 DB |
3775 | if (ub->ub_raidz_reflow_info != 0) { |
3776 | spa_load_note(spa, "uberblock raidz_reflow_info: " | |
3777 | "state=%u offset=%llu", | |
3778 | (int)RRSS_GET_STATE(ub), | |
3779 | (u_longlong_t)RRSS_GET_OFFSET(ub)); | |
3780 | } | |
4a0ee12a PZ |
3781 | |
3782 | ||
379ca9cf OF |
3783 | /* |
3784 | * For pools which have the multihost property on determine if the | |
3785 | * pool is truly inactive and can be safely imported. Prevent | |
3786 | * hosts which don't have a hostid set from importing the pool. | |
3787 | */ | |
6cb8e530 PZ |
3788 | activity_check = spa_activity_check_required(spa, ub, label, |
3789 | spa->spa_config); | |
379ca9cf | 3790 | if (activity_check) { |
379ca9cf | 3791 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && |
25f06d67 | 3792 | spa_get_hostid(spa) == 0) { |
379ca9cf OF |
3793 | nvlist_free(label); |
3794 | fnvlist_add_uint64(spa->spa_load_info, | |
3795 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
3796 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
3797 | } | |
3798 | ||
6cb8e530 | 3799 | int error = spa_activity_check(spa, ub, spa->spa_config); |
e889f0f5 OF |
3800 | if (error) { |
3801 | nvlist_free(label); | |
3802 | return (error); | |
3803 | } | |
3804 | ||
379ca9cf OF |
3805 | fnvlist_add_uint64(spa->spa_load_info, |
3806 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
3807 | fnvlist_add_uint64(spa->spa_load_info, | |
3808 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
060f0226 OF |
3809 | fnvlist_add_uint16(spa->spa_load_info, |
3810 | ZPOOL_CONFIG_MMP_SEQ, | |
3811 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)); | |
379ca9cf OF |
3812 | } |
3813 | ||
428870ff | 3814 | /* |
9ae529ec | 3815 | * If the pool has an unsupported version we can't open it. |
428870ff | 3816 | */ |
9ae529ec CS |
3817 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { |
3818 | nvlist_free(label); | |
4a0ee12a PZ |
3819 | spa_load_failed(spa, "version %llu is not supported", |
3820 | (u_longlong_t)ub->ub_version); | |
428870ff | 3821 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); |
9ae529ec CS |
3822 | } |
3823 | ||
3824 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
3825 | nvlist_t *features; | |
3826 | ||
3827 | /* | |
3828 | * If we weren't able to find what's necessary for reading the | |
3829 | * MOS in the label, return failure. | |
3830 | */ | |
4a0ee12a PZ |
3831 | if (label == NULL) { |
3832 | spa_load_failed(spa, "label config unavailable"); | |
3833 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
3834 | ENXIO)); | |
3835 | } | |
3836 | ||
3837 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, | |
3838 | &features) != 0) { | |
9ae529ec | 3839 | nvlist_free(label); |
4a0ee12a PZ |
3840 | spa_load_failed(spa, "invalid label: '%s' missing", |
3841 | ZPOOL_CONFIG_FEATURES_FOR_READ); | |
9ae529ec CS |
3842 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
3843 | ENXIO)); | |
3844 | } | |
3845 | ||
3846 | /* | |
3847 | * Update our in-core representation with the definitive values | |
3848 | * from the label. | |
3849 | */ | |
3850 | nvlist_free(spa->spa_label_features); | |
65ad5d11 | 3851 | spa->spa_label_features = fnvlist_dup(features); |
9ae529ec CS |
3852 | } |
3853 | ||
3854 | nvlist_free(label); | |
3855 | ||
3856 | /* | |
3857 | * Look through entries in the label nvlist's features_for_read. If | |
3858 | * there is a feature listed there which we don't understand then we | |
3859 | * cannot open a pool. | |
3860 | */ | |
3861 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
3862 | nvlist_t *unsup_feat; | |
9ae529ec | 3863 | |
65ad5d11 | 3864 | unsup_feat = fnvlist_alloc(); |
9ae529ec | 3865 | |
1c27024e DB |
3866 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, |
3867 | NULL); nvp != NULL; | |
9ae529ec CS |
3868 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { |
3869 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
65ad5d11 AJ |
3870 | fnvlist_add_string(unsup_feat, |
3871 | nvpair_name(nvp), ""); | |
9ae529ec CS |
3872 | } |
3873 | } | |
3874 | ||
3875 | if (!nvlist_empty(unsup_feat)) { | |
65ad5d11 AJ |
3876 | fnvlist_add_nvlist(spa->spa_load_info, |
3877 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec | 3878 | nvlist_free(unsup_feat); |
4a0ee12a | 3879 | spa_load_failed(spa, "some features are unsupported"); |
9ae529ec CS |
3880 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
3881 | ENOTSUP)); | |
3882 | } | |
3883 | ||
3884 | nvlist_free(unsup_feat); | |
3885 | } | |
428870ff | 3886 | |
428870ff BB |
3887 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { |
3888 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6cb8e530 | 3889 | spa_try_repair(spa, spa->spa_config); |
428870ff BB |
3890 | spa_config_exit(spa, SCL_ALL, FTAG); |
3891 | nvlist_free(spa->spa_config_splitting); | |
3892 | spa->spa_config_splitting = NULL; | |
3893 | } | |
3894 | ||
3895 | /* | |
3896 | * Initialize internal SPA structures. | |
3897 | */ | |
d2734cce | 3898 | spa_ld_select_uberblock_done(spa, ub); |
428870ff | 3899 | |
9eb7b46e PZ |
3900 | return (0); |
3901 | } | |
3902 | ||
3903 | static int | |
3904 | spa_ld_open_rootbp(spa_t *spa) | |
3905 | { | |
3906 | int error = 0; | |
3907 | vdev_t *rvd = spa->spa_root_vdev; | |
a1d477c2 | 3908 | |
9ae529ec | 3909 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
4a0ee12a PZ |
3910 | if (error != 0) { |
3911 | spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " | |
3912 | "[error=%d]", error); | |
428870ff | 3913 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 3914 | } |
428870ff BB |
3915 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
3916 | ||
9eb7b46e PZ |
3917 | return (0); |
3918 | } | |
3919 | ||
3920 | static int | |
d2734cce | 3921 | spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, |
6cb8e530 | 3922 | boolean_t reloading) |
9eb7b46e | 3923 | { |
6cb8e530 PZ |
3924 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; |
3925 | nvlist_t *nv, *mos_config, *policy; | |
3926 | int error = 0, copy_error; | |
3927 | uint64_t healthy_tvds, healthy_tvds_mos; | |
3928 | uint64_t mos_config_txg; | |
9eb7b46e | 3929 | |
4a0ee12a PZ |
3930 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) |
3931 | != 0) | |
428870ff BB |
3932 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
3933 | ||
a1d477c2 | 3934 | /* |
6cb8e530 PZ |
3935 | * If we're assembling a pool from a split, the config provided is |
3936 | * already trusted so there is nothing to do. | |
a1d477c2 | 3937 | */ |
6cb8e530 PZ |
3938 | if (type == SPA_IMPORT_ASSEMBLE) |
3939 | return (0); | |
3940 | ||
3941 | healthy_tvds = spa_healthy_core_tvds(spa); | |
a1d477c2 | 3942 | |
6cb8e530 PZ |
3943 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) |
3944 | != 0) { | |
3945 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
3946 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3947 | } | |
3948 | ||
3949 | /* | |
3950 | * If we are doing an open, pool owner wasn't verified yet, thus do | |
3951 | * the verification here. | |
3952 | */ | |
3953 | if (spa->spa_load_state == SPA_LOAD_OPEN) { | |
3954 | error = spa_verify_host(spa, mos_config); | |
3955 | if (error != 0) { | |
a1d477c2 | 3956 | nvlist_free(mos_config); |
6cb8e530 | 3957 | return (error); |
a1d477c2 | 3958 | } |
6cb8e530 PZ |
3959 | } |
3960 | ||
3961 | nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); | |
a1d477c2 | 3962 | |
6cb8e530 PZ |
3963 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
3964 | ||
3965 | /* | |
3966 | * Build a new vdev tree from the trusted config | |
3967 | */ | |
b2255edc BB |
3968 | error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD); |
3969 | if (error != 0) { | |
3970 | nvlist_free(mos_config); | |
3971 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3972 | spa_load_failed(spa, "spa_config_parse failed [error=%d]", | |
3973 | error); | |
3974 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
3975 | } | |
6cb8e530 PZ |
3976 | |
3977 | /* | |
3978 | * Vdev paths in the MOS may be obsolete. If the untrusted config was | |
3979 | * obtained by scanning /dev/dsk, then it will have the right vdev | |
3980 | * paths. We update the trusted MOS config with this information. | |
3981 | * We first try to copy the paths with vdev_copy_path_strict, which | |
3982 | * succeeds only when both configs have exactly the same vdev tree. | |
3983 | * If that fails, we fall back to a more flexible method that has a | |
3984 | * best effort policy. | |
3985 | */ | |
3986 | copy_error = vdev_copy_path_strict(rvd, mrvd); | |
3987 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
3988 | spa_load_note(spa, "provided vdev tree:"); | |
3989 | vdev_dbgmsg_print_tree(rvd, 2); | |
3990 | spa_load_note(spa, "MOS vdev tree:"); | |
3991 | vdev_dbgmsg_print_tree(mrvd, 2); | |
3992 | } | |
3993 | if (copy_error != 0) { | |
3994 | spa_load_note(spa, "vdev_copy_path_strict failed, falling " | |
3995 | "back to vdev_copy_path_relaxed"); | |
3996 | vdev_copy_path_relaxed(rvd, mrvd); | |
3997 | } | |
3998 | ||
3999 | vdev_close(rvd); | |
4000 | vdev_free(rvd); | |
4001 | spa->spa_root_vdev = mrvd; | |
4002 | rvd = mrvd; | |
4003 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4004 | ||
54b1b1d8 RN |
4005 | /* |
4006 | * If 'zpool import' used a cached config, then the on-disk hostid and | |
4007 | * hostname may be different to the cached config in ways that should | |
4008 | * prevent import. Userspace can't discover this without a scan, but | |
4009 | * we know, so we add these values to LOAD_INFO so the caller can know | |
4010 | * the difference. | |
4011 | * | |
4012 | * Note that we have to do this before the config is regenerated, | |
4013 | * because the new config will have the hostid and hostname for this | |
4014 | * host, in readiness for import. | |
4015 | */ | |
4016 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTID)) | |
4017 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_HOSTID, | |
4018 | fnvlist_lookup_uint64(mos_config, ZPOOL_CONFIG_HOSTID)); | |
4019 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTNAME)) | |
4020 | fnvlist_add_string(spa->spa_load_info, ZPOOL_CONFIG_HOSTNAME, | |
4021 | fnvlist_lookup_string(mos_config, ZPOOL_CONFIG_HOSTNAME)); | |
4022 | ||
6cb8e530 PZ |
4023 | /* |
4024 | * We will use spa_config if we decide to reload the spa or if spa_load | |
4025 | * fails and we rewind. We must thus regenerate the config using the | |
8a393be3 PZ |
4026 | * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to |
4027 | * pass settings on how to load the pool and is not stored in the MOS. | |
4028 | * We copy it over to our new, trusted config. | |
6cb8e530 PZ |
4029 | */ |
4030 | mos_config_txg = fnvlist_lookup_uint64(mos_config, | |
4031 | ZPOOL_CONFIG_POOL_TXG); | |
4032 | nvlist_free(mos_config); | |
4033 | mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); | |
8a393be3 | 4034 | if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, |
6cb8e530 | 4035 | &policy) == 0) |
8a393be3 | 4036 | fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); |
6cb8e530 PZ |
4037 | spa_config_set(spa, mos_config); |
4038 | spa->spa_config_source = SPA_CONFIG_SRC_MOS; | |
4039 | ||
4040 | /* | |
4041 | * Now that we got the config from the MOS, we should be more strict | |
4042 | * in checking blkptrs and can make assumptions about the consistency | |
4043 | * of the vdev tree. spa_trust_config must be set to true before opening | |
4044 | * vdevs in order for them to be writeable. | |
4045 | */ | |
4046 | spa->spa_trust_config = B_TRUE; | |
4047 | ||
4048 | /* | |
4049 | * Open and validate the new vdev tree | |
4050 | */ | |
4051 | error = spa_ld_open_vdevs(spa); | |
4052 | if (error != 0) | |
4053 | return (error); | |
4054 | ||
4055 | error = spa_ld_validate_vdevs(spa); | |
4056 | if (error != 0) | |
4057 | return (error); | |
4058 | ||
4059 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
4060 | spa_load_note(spa, "final vdev tree:"); | |
4061 | vdev_dbgmsg_print_tree(rvd, 2); | |
4062 | } | |
4063 | ||
4064 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && | |
4065 | !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { | |
a1d477c2 | 4066 | /* |
6cb8e530 PZ |
4067 | * Sanity check to make sure that we are indeed loading the |
4068 | * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds | |
4069 | * in the config provided and they happened to be the only ones | |
4070 | * to have the latest uberblock, we could involuntarily perform | |
4071 | * an extreme rewind. | |
a1d477c2 | 4072 | */ |
6cb8e530 PZ |
4073 | healthy_tvds_mos = spa_healthy_core_tvds(spa); |
4074 | if (healthy_tvds_mos - healthy_tvds >= | |
4075 | SPA_SYNC_MIN_VDEVS) { | |
4076 | spa_load_note(spa, "config provided misses too many " | |
4077 | "top-level vdevs compared to MOS (%lld vs %lld). ", | |
4078 | (u_longlong_t)healthy_tvds, | |
4079 | (u_longlong_t)healthy_tvds_mos); | |
4080 | spa_load_note(spa, "vdev tree:"); | |
4081 | vdev_dbgmsg_print_tree(rvd, 2); | |
4082 | if (reloading) { | |
4083 | spa_load_failed(spa, "config was already " | |
4084 | "provided from MOS. Aborting."); | |
4085 | return (spa_vdev_err(rvd, | |
4086 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4087 | } | |
4088 | spa_load_note(spa, "spa must be reloaded using MOS " | |
4089 | "config"); | |
4090 | return (SET_ERROR(EAGAIN)); | |
4a0ee12a | 4091 | } |
a1d477c2 MA |
4092 | } |
4093 | ||
6cb8e530 PZ |
4094 | error = spa_check_for_missing_logs(spa); |
4095 | if (error != 0) | |
4096 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
4097 | ||
4098 | if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { | |
4099 | spa_load_failed(spa, "uberblock guid sum doesn't match MOS " | |
4100 | "guid sum (%llu != %llu)", | |
4101 | (u_longlong_t)spa->spa_uberblock.ub_guid_sum, | |
4102 | (u_longlong_t)rvd->vdev_guid_sum); | |
4103 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
4104 | ENXIO)); | |
4105 | } | |
4106 | ||
9eb7b46e PZ |
4107 | return (0); |
4108 | } | |
4109 | ||
4110 | static int | |
4111 | spa_ld_open_indirect_vdev_metadata(spa_t *spa) | |
4112 | { | |
4113 | int error = 0; | |
4114 | vdev_t *rvd = spa->spa_root_vdev; | |
4115 | ||
a1d477c2 MA |
4116 | /* |
4117 | * Everything that we read before spa_remove_init() must be stored | |
4118 | * on concreted vdevs. Therefore we do this as early as possible. | |
4119 | */ | |
4a0ee12a PZ |
4120 | error = spa_remove_init(spa); |
4121 | if (error != 0) { | |
4122 | spa_load_failed(spa, "spa_remove_init failed [error=%d]", | |
4123 | error); | |
a1d477c2 | 4124 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4125 | } |
a1d477c2 | 4126 | |
9eb7b46e PZ |
4127 | /* |
4128 | * Retrieve information needed to condense indirect vdev mappings. | |
4129 | */ | |
4130 | error = spa_condense_init(spa); | |
4131 | if (error != 0) { | |
4a0ee12a PZ |
4132 | spa_load_failed(spa, "spa_condense_init failed [error=%d]", |
4133 | error); | |
9eb7b46e PZ |
4134 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
4135 | } | |
4136 | ||
4137 | return (0); | |
4138 | } | |
4139 | ||
4140 | static int | |
4a0ee12a | 4141 | spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) |
9eb7b46e PZ |
4142 | { |
4143 | int error = 0; | |
4144 | vdev_t *rvd = spa->spa_root_vdev; | |
4145 | ||
9ae529ec CS |
4146 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { |
4147 | boolean_t missing_feat_read = B_FALSE; | |
b9b24bb4 | 4148 | nvlist_t *unsup_feat, *enabled_feat; |
9ae529ec CS |
4149 | |
4150 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
4a0ee12a | 4151 | &spa->spa_feat_for_read_obj, B_TRUE) != 0) { |
9ae529ec CS |
4152 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4153 | } | |
4154 | ||
4155 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
4a0ee12a | 4156 | &spa->spa_feat_for_write_obj, B_TRUE) != 0) { |
9ae529ec CS |
4157 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4158 | } | |
4159 | ||
4160 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
4a0ee12a | 4161 | &spa->spa_feat_desc_obj, B_TRUE) != 0) { |
9ae529ec CS |
4162 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4163 | } | |
4164 | ||
b9b24bb4 CS |
4165 | enabled_feat = fnvlist_alloc(); |
4166 | unsup_feat = fnvlist_alloc(); | |
9ae529ec | 4167 | |
fa86b5db | 4168 | if (!spa_features_check(spa, B_FALSE, |
b9b24bb4 | 4169 | unsup_feat, enabled_feat)) |
9ae529ec CS |
4170 | missing_feat_read = B_TRUE; |
4171 | ||
4a0ee12a PZ |
4172 | if (spa_writeable(spa) || |
4173 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) { | |
fa86b5db | 4174 | if (!spa_features_check(spa, B_TRUE, |
b9b24bb4 | 4175 | unsup_feat, enabled_feat)) { |
9eb7b46e | 4176 | *missing_feat_writep = B_TRUE; |
b9b24bb4 | 4177 | } |
9ae529ec CS |
4178 | } |
4179 | ||
b9b24bb4 CS |
4180 | fnvlist_add_nvlist(spa->spa_load_info, |
4181 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
4182 | ||
9ae529ec | 4183 | if (!nvlist_empty(unsup_feat)) { |
b9b24bb4 CS |
4184 | fnvlist_add_nvlist(spa->spa_load_info, |
4185 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec CS |
4186 | } |
4187 | ||
b9b24bb4 CS |
4188 | fnvlist_free(enabled_feat); |
4189 | fnvlist_free(unsup_feat); | |
9ae529ec CS |
4190 | |
4191 | if (!missing_feat_read) { | |
4192 | fnvlist_add_boolean(spa->spa_load_info, | |
4193 | ZPOOL_CONFIG_CAN_RDONLY); | |
4194 | } | |
4195 | ||
4196 | /* | |
4197 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
4198 | * twofold: to determine whether the pool is available for | |
4199 | * import in read-write mode and (if it is not) whether the | |
4200 | * pool is available for import in read-only mode. If the pool | |
4201 | * is available for import in read-write mode, it is displayed | |
4202 | * as available in userland; if it is not available for import | |
4203 | * in read-only mode, it is displayed as unavailable in | |
4204 | * userland. If the pool is available for import in read-only | |
4205 | * mode but not read-write mode, it is displayed as unavailable | |
4206 | * in userland with a special note that the pool is actually | |
4207 | * available for open in read-only mode. | |
4208 | * | |
4209 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
4210 | * missing a feature for write, we must first determine whether | |
4211 | * the pool can be opened read-only before returning to | |
4212 | * userland in order to know whether to display the | |
4213 | * abovementioned note. | |
4214 | */ | |
9eb7b46e | 4215 | if (missing_feat_read || (*missing_feat_writep && |
9ae529ec | 4216 | spa_writeable(spa))) { |
4a0ee12a | 4217 | spa_load_failed(spa, "pool uses unsupported features"); |
9ae529ec CS |
4218 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, |
4219 | ENOTSUP)); | |
4220 | } | |
b0bc7a84 MG |
4221 | |
4222 | /* | |
4223 | * Load refcounts for ZFS features from disk into an in-memory | |
4224 | * cache during SPA initialization. | |
4225 | */ | |
1c27024e | 4226 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { |
b0bc7a84 MG |
4227 | uint64_t refcount; |
4228 | ||
4229 | error = feature_get_refcount_from_disk(spa, | |
4230 | &spa_feature_table[i], &refcount); | |
4231 | if (error == 0) { | |
4232 | spa->spa_feat_refcount_cache[i] = refcount; | |
4233 | } else if (error == ENOTSUP) { | |
4234 | spa->spa_feat_refcount_cache[i] = | |
4235 | SPA_FEATURE_DISABLED; | |
4236 | } else { | |
4a0ee12a PZ |
4237 | spa_load_failed(spa, "error getting refcount " |
4238 | "for feature %s [error=%d]", | |
4239 | spa_feature_table[i].fi_guid, error); | |
b0bc7a84 MG |
4240 | return (spa_vdev_err(rvd, |
4241 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4242 | } | |
4243 | } | |
4244 | } | |
4245 | ||
4246 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
4247 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
4a0ee12a | 4248 | &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) |
b0bc7a84 | 4249 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
9ae529ec CS |
4250 | } |
4251 | ||
f00ab3f2 TC |
4252 | /* |
4253 | * Encryption was added before bookmark_v2, even though bookmark_v2 | |
4254 | * is now a dependency. If this pool has encryption enabled without | |
4255 | * bookmark_v2, trigger an errata message. | |
4256 | */ | |
4257 | if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) && | |
4258 | !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) { | |
4259 | spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; | |
4260 | } | |
4261 | ||
9eb7b46e PZ |
4262 | return (0); |
4263 | } | |
4264 | ||
4265 | static int | |
4266 | spa_ld_load_special_directories(spa_t *spa) | |
4267 | { | |
4268 | int error = 0; | |
4269 | vdev_t *rvd = spa->spa_root_vdev; | |
4270 | ||
9ae529ec CS |
4271 | spa->spa_is_initializing = B_TRUE; |
4272 | error = dsl_pool_open(spa->spa_dsl_pool); | |
4273 | spa->spa_is_initializing = B_FALSE; | |
4a0ee12a PZ |
4274 | if (error != 0) { |
4275 | spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); | |
9ae529ec | 4276 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4277 | } |
9ae529ec | 4278 | |
9eb7b46e PZ |
4279 | return (0); |
4280 | } | |
428870ff | 4281 | |
9eb7b46e PZ |
4282 | static int |
4283 | spa_ld_get_props(spa_t *spa) | |
4284 | { | |
4285 | int error = 0; | |
4286 | uint64_t obj; | |
4287 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4288 | |
3c67d83a TH |
4289 | /* Grab the checksum salt from the MOS. */ |
4290 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4291 | DMU_POOL_CHECKSUM_SALT, 1, | |
4292 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
4293 | spa->spa_cksum_salt.zcs_bytes); | |
4294 | if (error == ENOENT) { | |
4295 | /* Generate a new salt for subsequent use */ | |
4296 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4297 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4298 | } else if (error != 0) { | |
4a0ee12a PZ |
4299 | spa_load_failed(spa, "unable to retrieve checksum salt from " |
4300 | "MOS [error=%d]", error); | |
3c67d83a TH |
4301 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4302 | } | |
4303 | ||
4a0ee12a | 4304 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) |
428870ff BB |
4305 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4306 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
4a0ee12a PZ |
4307 | if (error != 0) { |
4308 | spa_load_failed(spa, "error opening deferred-frees bpobj " | |
4309 | "[error=%d]", error); | |
428870ff | 4310 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4311 | } |
34dc7c2f BB |
4312 | |
4313 | /* | |
4314 | * Load the bit that tells us to use the new accounting function | |
4315 | * (raid-z deflation). If we have an older pool, this will not | |
4316 | * be present. | |
4317 | */ | |
4a0ee12a | 4318 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); |
428870ff BB |
4319 | if (error != 0 && error != ENOENT) |
4320 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4321 | ||
4322 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
4a0ee12a | 4323 | &spa->spa_creation_version, B_FALSE); |
428870ff BB |
4324 | if (error != 0 && error != ENOENT) |
4325 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
4326 | |
4327 | /* | |
4328 | * Load the persistent error log. If we have an older pool, this will | |
4329 | * not be present. | |
4330 | */ | |
4a0ee12a PZ |
4331 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, |
4332 | B_FALSE); | |
428870ff BB |
4333 | if (error != 0 && error != ENOENT) |
4334 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4335 | |
428870ff | 4336 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, |
4a0ee12a | 4337 | &spa->spa_errlog_scrub, B_FALSE); |
428870ff BB |
4338 | if (error != 0 && error != ENOENT) |
4339 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 4340 | |
37f03da8 SH |
4341 | /* |
4342 | * Load the livelist deletion field. If a livelist is queued for | |
4343 | * deletion, indicate that in the spa | |
4344 | */ | |
4345 | error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES, | |
4346 | &spa->spa_livelists_to_delete, B_FALSE); | |
4347 | if (error != 0 && error != ENOENT) | |
4348 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4349 | ||
34dc7c2f BB |
4350 | /* |
4351 | * Load the history object. If we have an older pool, this | |
4352 | * will not be present. | |
4353 | */ | |
4a0ee12a | 4354 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); |
428870ff BB |
4355 | if (error != 0 && error != ENOENT) |
4356 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4357 | ||
e0ab3ab5 JS |
4358 | /* |
4359 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
4360 | * be present; in this case, defer its creation to a later time to | |
4361 | * avoid dirtying the MOS this early / out of sync context. See | |
4362 | * spa_sync_config_object. | |
4363 | */ | |
4364 | ||
4365 | /* The sentinel is only available in the MOS config. */ | |
1c27024e | 4366 | nvlist_t *mos_config; |
4a0ee12a PZ |
4367 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { |
4368 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
e0ab3ab5 | 4369 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4370 | } |
e0ab3ab5 JS |
4371 | |
4372 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
4a0ee12a | 4373 | &spa->spa_all_vdev_zaps, B_FALSE); |
e0ab3ab5 | 4374 | |
38640550 DB |
4375 | if (error == ENOENT) { |
4376 | VERIFY(!nvlist_exists(mos_config, | |
4377 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
4378 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
4379 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4380 | } else if (error != 0) { | |
cb01da68 | 4381 | nvlist_free(mos_config); |
e0ab3ab5 | 4382 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
38640550 | 4383 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { |
e0ab3ab5 JS |
4384 | /* |
4385 | * An older version of ZFS overwrote the sentinel value, so | |
4386 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
4387 | * destruction to later; see spa_sync_config_object. | |
4388 | */ | |
4389 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
4390 | /* | |
4391 | * We're assuming that no vdevs have had their ZAPs created | |
4392 | * before this. Better be sure of it. | |
4393 | */ | |
4394 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4395 | } | |
4396 | nvlist_free(mos_config); | |
4397 | ||
9eb7b46e PZ |
4398 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); |
4399 | ||
4a0ee12a PZ |
4400 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, |
4401 | B_FALSE); | |
9eb7b46e PZ |
4402 | if (error && error != ENOENT) |
4403 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4404 | ||
4405 | if (error == 0) { | |
da27b8bc | 4406 | uint64_t autoreplace = 0; |
9eb7b46e PZ |
4407 | |
4408 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
4409 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
4410 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
4411 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
4412 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
c02c1bec | 4413 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); |
1b939560 | 4414 | spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim); |
9eb7b46e PZ |
4415 | spa->spa_autoreplace = (autoreplace != 0); |
4416 | } | |
4417 | ||
6cb8e530 PZ |
4418 | /* |
4419 | * If we are importing a pool with missing top-level vdevs, | |
4420 | * we enforce that the pool doesn't panic or get suspended on | |
4421 | * error since the likelihood of missing data is extremely high. | |
4422 | */ | |
4423 | if (spa->spa_missing_tvds > 0 && | |
4424 | spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && | |
4425 | spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4426 | spa_load_note(spa, "forcing failmode to 'continue' " | |
4427 | "as some top level vdevs are missing"); | |
4428 | spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; | |
4429 | } | |
4430 | ||
9eb7b46e PZ |
4431 | return (0); |
4432 | } | |
4433 | ||
4434 | static int | |
4435 | spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) | |
4436 | { | |
4437 | int error = 0; | |
4438 | vdev_t *rvd = spa->spa_root_vdev; | |
4439 | ||
428870ff BB |
4440 | /* |
4441 | * If we're assembling the pool from the split-off vdevs of | |
4442 | * an existing pool, we don't want to attach the spares & cache | |
4443 | * devices. | |
4444 | */ | |
34dc7c2f BB |
4445 | |
4446 | /* | |
4447 | * Load any hot spares for this pool. | |
4448 | */ | |
4a0ee12a PZ |
4449 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, |
4450 | B_FALSE); | |
428870ff BB |
4451 | if (error != 0 && error != ENOENT) |
4452 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4453 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4454 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); |
4455 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
4a0ee12a PZ |
4456 | &spa->spa_spares.sav_config) != 0) { |
4457 | spa_load_failed(spa, "error loading spares nvlist"); | |
428870ff | 4458 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4459 | } |
34dc7c2f | 4460 | |
b128c09f | 4461 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4462 | spa_load_spares(spa); |
b128c09f | 4463 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4464 | } else if (error == 0) { |
4465 | spa->spa_spares.sav_sync = B_TRUE; | |
34dc7c2f BB |
4466 | } |
4467 | ||
4468 | /* | |
4469 | * Load any level 2 ARC devices for this pool. | |
4470 | */ | |
428870ff | 4471 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, |
4a0ee12a | 4472 | &spa->spa_l2cache.sav_object, B_FALSE); |
428870ff BB |
4473 | if (error != 0 && error != ENOENT) |
4474 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4475 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
4476 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); |
4477 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
4a0ee12a PZ |
4478 | &spa->spa_l2cache.sav_config) != 0) { |
4479 | spa_load_failed(spa, "error loading l2cache nvlist"); | |
428870ff | 4480 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4481 | } |
34dc7c2f | 4482 | |
b128c09f | 4483 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4484 | spa_load_l2cache(spa); |
b128c09f | 4485 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
4486 | } else if (error == 0) { |
4487 | spa->spa_l2cache.sav_sync = B_TRUE; | |
b128c09f BB |
4488 | } |
4489 | ||
9eb7b46e PZ |
4490 | return (0); |
4491 | } | |
428870ff | 4492 | |
9eb7b46e | 4493 | static int |
4a0ee12a | 4494 | spa_ld_load_vdev_metadata(spa_t *spa) |
9eb7b46e PZ |
4495 | { |
4496 | int error = 0; | |
4497 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 4498 | |
379ca9cf OF |
4499 | /* |
4500 | * If the 'multihost' property is set, then never allow a pool to | |
4501 | * be imported when the system hostid is zero. The exception to | |
4502 | * this rule is zdb which is always allowed to access pools. | |
4503 | */ | |
25f06d67 | 4504 | if (spa_multihost(spa) && spa_get_hostid(spa) == 0 && |
379ca9cf OF |
4505 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { |
4506 | fnvlist_add_uint64(spa->spa_load_info, | |
4507 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4508 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4509 | } | |
4510 | ||
34dc7c2f BB |
4511 | /* |
4512 | * If the 'autoreplace' property is set, then post a resource notifying | |
4513 | * the ZFS DE that it should not issue any faults for unopenable | |
4514 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
4515 | * unopenable vdevs so that the normal autoreplace handler can take | |
4516 | * over. | |
4517 | */ | |
4a0ee12a | 4518 | if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
34dc7c2f | 4519 | spa_check_removed(spa->spa_root_vdev); |
428870ff BB |
4520 | /* |
4521 | * For the import case, this is done in spa_import(), because | |
4522 | * at this point we're using the spare definitions from | |
4523 | * the MOS config, not necessarily from the userland config. | |
4524 | */ | |
4a0ee12a | 4525 | if (spa->spa_load_state != SPA_LOAD_IMPORT) { |
428870ff BB |
4526 | spa_aux_check_removed(&spa->spa_spares); |
4527 | spa_aux_check_removed(&spa->spa_l2cache); | |
4528 | } | |
4529 | } | |
34dc7c2f BB |
4530 | |
4531 | /* | |
9eb7b46e | 4532 | * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. |
34dc7c2f | 4533 | */ |
a1d477c2 MA |
4534 | error = vdev_load(rvd); |
4535 | if (error != 0) { | |
4a0ee12a | 4536 | spa_load_failed(spa, "vdev_load failed [error=%d]", error); |
a1d477c2 MA |
4537 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); |
4538 | } | |
4539 | ||
93e28d66 SD |
4540 | error = spa_ld_log_spacemaps(spa); |
4541 | if (error != 0) { | |
600a02b8 | 4542 | spa_load_failed(spa, "spa_ld_log_spacemaps failed [error=%d]", |
93e28d66 SD |
4543 | error); |
4544 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4545 | } | |
4546 | ||
34dc7c2f | 4547 | /* |
9eb7b46e | 4548 | * Propagate the leaf DTLs we just loaded all the way up the vdev tree. |
34dc7c2f | 4549 | */ |
b128c09f | 4550 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
9a49d3f3 | 4551 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE); |
b128c09f | 4552 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 4553 | |
9eb7b46e PZ |
4554 | return (0); |
4555 | } | |
4556 | ||
4557 | static int | |
4558 | spa_ld_load_dedup_tables(spa_t *spa) | |
4559 | { | |
4560 | int error = 0; | |
4561 | vdev_t *rvd = spa->spa_root_vdev; | |
4562 | ||
428870ff | 4563 | error = ddt_load(spa); |
4a0ee12a PZ |
4564 | if (error != 0) { |
4565 | spa_load_failed(spa, "ddt_load failed [error=%d]", error); | |
428870ff | 4566 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
4a0ee12a | 4567 | } |
428870ff | 4568 | |
9eb7b46e PZ |
4569 | return (0); |
4570 | } | |
4571 | ||
67a1b037 PJD |
4572 | static int |
4573 | spa_ld_load_brt(spa_t *spa) | |
4574 | { | |
4575 | int error = 0; | |
4576 | vdev_t *rvd = spa->spa_root_vdev; | |
4577 | ||
4578 | error = brt_load(spa); | |
4579 | if (error != 0) { | |
4580 | spa_load_failed(spa, "brt_load failed [error=%d]", error); | |
4581 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4582 | } | |
4583 | ||
4584 | return (0); | |
4585 | } | |
4586 | ||
9eb7b46e | 4587 | static int |
a926aab9 | 4588 | spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, const char **ereport) |
9eb7b46e PZ |
4589 | { |
4590 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff | 4591 | |
4a0ee12a PZ |
4592 | if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { |
4593 | boolean_t missing = spa_check_logs(spa); | |
4594 | if (missing) { | |
6cb8e530 PZ |
4595 | if (spa->spa_missing_tvds != 0) { |
4596 | spa_load_note(spa, "spa_check_logs failed " | |
4597 | "so dropping the logs"); | |
4598 | } else { | |
4599 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
4600 | spa_load_failed(spa, "spa_check_logs failed"); | |
4601 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, | |
4602 | ENXIO)); | |
4603 | } | |
4a0ee12a | 4604 | } |
428870ff BB |
4605 | } |
4606 | ||
9eb7b46e PZ |
4607 | return (0); |
4608 | } | |
4609 | ||
4610 | static int | |
4a0ee12a | 4611 | spa_ld_verify_pool_data(spa_t *spa) |
9eb7b46e PZ |
4612 | { |
4613 | int error = 0; | |
4614 | vdev_t *rvd = spa->spa_root_vdev; | |
4615 | ||
4616 | /* | |
4617 | * We've successfully opened the pool, verify that we're ready | |
4618 | * to start pushing transactions. | |
4619 | */ | |
4a0ee12a | 4620 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { |
9eb7b46e PZ |
4621 | error = spa_load_verify(spa); |
4622 | if (error != 0) { | |
4a0ee12a PZ |
4623 | spa_load_failed(spa, "spa_load_verify failed " |
4624 | "[error=%d]", error); | |
9eb7b46e PZ |
4625 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
4626 | error)); | |
4627 | } | |
4628 | } | |
4629 | ||
4630 | return (0); | |
4631 | } | |
4632 | ||
4633 | static void | |
4634 | spa_ld_claim_log_blocks(spa_t *spa) | |
4635 | { | |
4636 | dmu_tx_t *tx; | |
4637 | dsl_pool_t *dp = spa_get_dsl(spa); | |
4638 | ||
4639 | /* | |
4640 | * Claim log blocks that haven't been committed yet. | |
4641 | * This must all happen in a single txg. | |
4642 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
4643 | * invoked from zil_claim_log_block()'s i/o done callback. | |
4644 | * Price of rollback is that we abandon the log. | |
4645 | */ | |
4646 | spa->spa_claiming = B_TRUE; | |
4647 | ||
4648 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
4649 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
4650 | zil_claim, tx, DS_FIND_CHILDREN); | |
4651 | dmu_tx_commit(tx); | |
4652 | ||
4653 | spa->spa_claiming = B_FALSE; | |
4654 | ||
4655 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
4656 | } | |
4657 | ||
4658 | static void | |
6cb8e530 | 4659 | spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, |
d2734cce | 4660 | boolean_t update_config_cache) |
9eb7b46e PZ |
4661 | { |
4662 | vdev_t *rvd = spa->spa_root_vdev; | |
4663 | int need_update = B_FALSE; | |
4664 | ||
4665 | /* | |
4666 | * If the config cache is stale, or we have uninitialized | |
4667 | * metaslabs (see spa_vdev_add()), then update the config. | |
4668 | * | |
4669 | * If this is a verbatim import, trust the current | |
4670 | * in-core spa_config and update the disk labels. | |
4671 | */ | |
d2734cce | 4672 | if (update_config_cache || config_cache_txg != spa->spa_config_txg || |
4a0ee12a PZ |
4673 | spa->spa_load_state == SPA_LOAD_IMPORT || |
4674 | spa->spa_load_state == SPA_LOAD_RECOVER || | |
9eb7b46e PZ |
4675 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) |
4676 | need_update = B_TRUE; | |
4677 | ||
4678 | for (int c = 0; c < rvd->vdev_children; c++) | |
4679 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
4680 | need_update = B_TRUE; | |
4681 | ||
4682 | /* | |
e1cfd73f | 4683 | * Update the config cache asynchronously in case we're the |
9eb7b46e PZ |
4684 | * root pool, in which case the config cache isn't writable yet. |
4685 | */ | |
4686 | if (need_update) | |
4687 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
4688 | } | |
4689 | ||
6cb8e530 PZ |
4690 | static void |
4691 | spa_ld_prepare_for_reload(spa_t *spa) | |
4692 | { | |
da92d5cb | 4693 | spa_mode_t mode = spa->spa_mode; |
6cb8e530 PZ |
4694 | int async_suspended = spa->spa_async_suspended; |
4695 | ||
4696 | spa_unload(spa); | |
4697 | spa_deactivate(spa); | |
4698 | spa_activate(spa, mode); | |
4699 | ||
4700 | /* | |
4701 | * We save the value of spa_async_suspended as it gets reset to 0 by | |
4702 | * spa_unload(). We want to restore it back to the original value before | |
4703 | * returning as we might be calling spa_async_resume() later. | |
4704 | */ | |
4705 | spa->spa_async_suspended = async_suspended; | |
4706 | } | |
4707 | ||
9eb7b46e | 4708 | static int |
d2734cce SD |
4709 | spa_ld_read_checkpoint_txg(spa_t *spa) |
4710 | { | |
4711 | uberblock_t checkpoint; | |
4712 | int error = 0; | |
4713 | ||
4714 | ASSERT0(spa->spa_checkpoint_txg); | |
4715 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4716 | ||
4717 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4718 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
4719 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
4720 | ||
4721 | if (error == ENOENT) | |
4722 | return (0); | |
4723 | ||
4724 | if (error != 0) | |
4725 | return (error); | |
4726 | ||
4727 | ASSERT3U(checkpoint.ub_txg, !=, 0); | |
4728 | ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); | |
4729 | ASSERT3U(checkpoint.ub_timestamp, !=, 0); | |
4730 | spa->spa_checkpoint_txg = checkpoint.ub_txg; | |
4731 | spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; | |
4732 | ||
4733 | return (0); | |
4734 | } | |
4735 | ||
4736 | static int | |
4737 | spa_ld_mos_init(spa_t *spa, spa_import_type_t type) | |
9eb7b46e PZ |
4738 | { |
4739 | int error = 0; | |
9eb7b46e | 4740 | |
4a0ee12a | 4741 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
6cb8e530 | 4742 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); |
4a0ee12a | 4743 | |
9eb7b46e | 4744 | /* |
6cb8e530 PZ |
4745 | * Never trust the config that is provided unless we are assembling |
4746 | * a pool following a split. | |
4747 | * This means don't trust blkptrs and the vdev tree in general. This | |
4748 | * also effectively puts the spa in read-only mode since | |
4749 | * spa_writeable() checks for spa_trust_config to be true. | |
4750 | * We will later load a trusted config from the MOS. | |
9eb7b46e | 4751 | */ |
6cb8e530 PZ |
4752 | if (type != SPA_IMPORT_ASSEMBLE) |
4753 | spa->spa_trust_config = B_FALSE; | |
4754 | ||
9eb7b46e PZ |
4755 | /* |
4756 | * Parse the config provided to create a vdev tree. | |
4757 | */ | |
6cb8e530 | 4758 | error = spa_ld_parse_config(spa, type); |
9eb7b46e PZ |
4759 | if (error != 0) |
4760 | return (error); | |
4761 | ||
ca95f70d OF |
4762 | spa_import_progress_add(spa); |
4763 | ||
9eb7b46e PZ |
4764 | /* |
4765 | * Now that we have the vdev tree, try to open each vdev. This involves | |
4766 | * opening the underlying physical device, retrieving its geometry and | |
4767 | * probing the vdev with a dummy I/O. The state of each vdev will be set | |
4768 | * based on the success of those operations. After this we'll be ready | |
4769 | * to read from the vdevs. | |
4770 | */ | |
4771 | error = spa_ld_open_vdevs(spa); | |
4772 | if (error != 0) | |
4773 | return (error); | |
4774 | ||
4775 | /* | |
4776 | * Read the label of each vdev and make sure that the GUIDs stored | |
4777 | * there match the GUIDs in the config provided. | |
6cb8e530 PZ |
4778 | * If we're assembling a new pool that's been split off from an |
4779 | * existing pool, the labels haven't yet been updated so we skip | |
4780 | * validation for now. | |
9eb7b46e | 4781 | */ |
6cb8e530 PZ |
4782 | if (type != SPA_IMPORT_ASSEMBLE) { |
4783 | error = spa_ld_validate_vdevs(spa); | |
4784 | if (error != 0) | |
4785 | return (error); | |
4786 | } | |
9eb7b46e PZ |
4787 | |
4788 | /* | |
d2734cce SD |
4789 | * Read all vdev labels to find the best uberblock (i.e. latest, |
4790 | * unless spa_load_max_txg is set) and store it in spa_uberblock. We | |
4791 | * get the list of features required to read blkptrs in the MOS from | |
4792 | * the vdev label with the best uberblock and verify that our version | |
4793 | * of zfs supports them all. | |
9eb7b46e | 4794 | */ |
6cb8e530 | 4795 | error = spa_ld_select_uberblock(spa, type); |
9eb7b46e PZ |
4796 | if (error != 0) |
4797 | return (error); | |
4798 | ||
4799 | /* | |
4800 | * Pass that uberblock to the dsl_pool layer which will open the root | |
4801 | * blkptr. This blkptr points to the latest version of the MOS and will | |
4802 | * allow us to read its contents. | |
4803 | */ | |
4804 | error = spa_ld_open_rootbp(spa); | |
4805 | if (error != 0) | |
4806 | return (error); | |
4807 | ||
d2734cce SD |
4808 | return (0); |
4809 | } | |
4810 | ||
4811 | static int | |
4812 | spa_ld_checkpoint_rewind(spa_t *spa) | |
4813 | { | |
4814 | uberblock_t checkpoint; | |
4815 | int error = 0; | |
4816 | ||
4817 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4818 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
4819 | ||
4820 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4821 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
4822 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
4823 | ||
4824 | if (error != 0) { | |
4825 | spa_load_failed(spa, "unable to retrieve checkpointed " | |
4826 | "uberblock from the MOS config [error=%d]", error); | |
4827 | ||
4828 | if (error == ENOENT) | |
4829 | error = ZFS_ERR_NO_CHECKPOINT; | |
4830 | ||
4831 | return (error); | |
4832 | } | |
4833 | ||
4834 | ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); | |
4835 | ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); | |
4836 | ||
4837 | /* | |
4838 | * We need to update the txg and timestamp of the checkpointed | |
4839 | * uberblock to be higher than the latest one. This ensures that | |
4840 | * the checkpointed uberblock is selected if we were to close and | |
4841 | * reopen the pool right after we've written it in the vdev labels. | |
4842 | * (also see block comment in vdev_uberblock_compare) | |
4843 | */ | |
4844 | checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; | |
4845 | checkpoint.ub_timestamp = gethrestime_sec(); | |
4846 | ||
4847 | /* | |
4848 | * Set current uberblock to be the checkpointed uberblock. | |
4849 | */ | |
4850 | spa->spa_uberblock = checkpoint; | |
4851 | ||
4852 | /* | |
4853 | * If we are doing a normal rewind, then the pool is open for | |
4854 | * writing and we sync the "updated" checkpointed uberblock to | |
4855 | * disk. Once this is done, we've basically rewound the whole | |
4856 | * pool and there is no way back. | |
4857 | * | |
4858 | * There are cases when we don't want to attempt and sync the | |
4859 | * checkpointed uberblock to disk because we are opening a | |
4860 | * pool as read-only. Specifically, verifying the checkpointed | |
4861 | * state with zdb, and importing the checkpointed state to get | |
4862 | * a "preview" of its content. | |
4863 | */ | |
4864 | if (spa_writeable(spa)) { | |
4865 | vdev_t *rvd = spa->spa_root_vdev; | |
4866 | ||
4867 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4868 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; | |
4869 | int svdcount = 0; | |
4870 | int children = rvd->vdev_children; | |
29274c9f | 4871 | int c0 = random_in_range(children); |
d2734cce SD |
4872 | |
4873 | for (int c = 0; c < children; c++) { | |
4874 | vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; | |
4875 | ||
4876 | /* Stop when revisiting the first vdev */ | |
4877 | if (c > 0 && svd[0] == vd) | |
4878 | break; | |
4879 | ||
4880 | if (vd->vdev_ms_array == 0 || vd->vdev_islog || | |
4881 | !vdev_is_concrete(vd)) | |
4882 | continue; | |
4883 | ||
4884 | svd[svdcount++] = vd; | |
4885 | if (svdcount == SPA_SYNC_MIN_VDEVS) | |
4886 | break; | |
4887 | } | |
4888 | error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); | |
4889 | if (error == 0) | |
4890 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
4891 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4892 | ||
4893 | if (error != 0) { | |
4894 | spa_load_failed(spa, "failed to write checkpointed " | |
4895 | "uberblock to the vdev labels [error=%d]", error); | |
4896 | return (error); | |
4897 | } | |
4898 | } | |
4899 | ||
4900 | return (0); | |
4901 | } | |
4902 | ||
4903 | static int | |
4904 | spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, | |
4905 | boolean_t *update_config_cache) | |
4906 | { | |
4907 | int error; | |
4908 | ||
4909 | /* | |
4910 | * Parse the config for pool, open and validate vdevs, | |
4911 | * select an uberblock, and use that uberblock to open | |
4912 | * the MOS. | |
4913 | */ | |
4914 | error = spa_ld_mos_init(spa, type); | |
4915 | if (error != 0) | |
4916 | return (error); | |
4917 | ||
9eb7b46e | 4918 | /* |
6cb8e530 PZ |
4919 | * Retrieve the trusted config stored in the MOS and use it to create |
4920 | * a new, exact version of the vdev tree, then reopen all vdevs. | |
9eb7b46e | 4921 | */ |
d2734cce | 4922 | error = spa_ld_trusted_config(spa, type, B_FALSE); |
6cb8e530 | 4923 | if (error == EAGAIN) { |
d2734cce SD |
4924 | if (update_config_cache != NULL) |
4925 | *update_config_cache = B_TRUE; | |
4926 | ||
6cb8e530 PZ |
4927 | /* |
4928 | * Redo the loading process with the trusted config if it is | |
4929 | * too different from the untrusted config. | |
4930 | */ | |
4931 | spa_ld_prepare_for_reload(spa); | |
d2734cce SD |
4932 | spa_load_note(spa, "RELOADING"); |
4933 | error = spa_ld_mos_init(spa, type); | |
4934 | if (error != 0) | |
4935 | return (error); | |
4936 | ||
4937 | error = spa_ld_trusted_config(spa, type, B_TRUE); | |
4938 | if (error != 0) | |
4939 | return (error); | |
4940 | ||
6cb8e530 | 4941 | } else if (error != 0) { |
9eb7b46e | 4942 | return (error); |
6cb8e530 | 4943 | } |
9eb7b46e | 4944 | |
d2734cce SD |
4945 | return (0); |
4946 | } | |
4947 | ||
4948 | /* | |
4949 | * Load an existing storage pool, using the config provided. This config | |
4950 | * describes which vdevs are part of the pool and is later validated against | |
4951 | * partial configs present in each vdev's label and an entire copy of the | |
4952 | * config stored in the MOS. | |
4953 | */ | |
4954 | static int | |
a926aab9 | 4955 | spa_load_impl(spa_t *spa, spa_import_type_t type, const char **ereport) |
d2734cce SD |
4956 | { |
4957 | int error = 0; | |
4958 | boolean_t missing_feat_write = B_FALSE; | |
4959 | boolean_t checkpoint_rewind = | |
4960 | (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
4961 | boolean_t update_config_cache = B_FALSE; | |
4962 | ||
4963 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
4964 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); | |
4965 | ||
4966 | spa_load_note(spa, "LOADING"); | |
4967 | ||
4968 | error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); | |
4969 | if (error != 0) | |
4970 | return (error); | |
4971 | ||
4972 | /* | |
4973 | * If we are rewinding to the checkpoint then we need to repeat | |
4974 | * everything we've done so far in this function but this time | |
4975 | * selecting the checkpointed uberblock and using that to open | |
4976 | * the MOS. | |
4977 | */ | |
4978 | if (checkpoint_rewind) { | |
4979 | /* | |
4980 | * If we are rewinding to the checkpoint update config cache | |
4981 | * anyway. | |
4982 | */ | |
4983 | update_config_cache = B_TRUE; | |
4984 | ||
4985 | /* | |
4986 | * Extract the checkpointed uberblock from the current MOS | |
4987 | * and use this as the pool's uberblock from now on. If the | |
4988 | * pool is imported as writeable we also write the checkpoint | |
4989 | * uberblock to the labels, making the rewind permanent. | |
4990 | */ | |
4991 | error = spa_ld_checkpoint_rewind(spa); | |
4992 | if (error != 0) | |
4993 | return (error); | |
4994 | ||
4995 | /* | |
e1cfd73f | 4996 | * Redo the loading process again with the |
d2734cce SD |
4997 | * checkpointed uberblock. |
4998 | */ | |
4999 | spa_ld_prepare_for_reload(spa); | |
5000 | spa_load_note(spa, "LOADING checkpointed uberblock"); | |
5001 | error = spa_ld_mos_with_trusted_config(spa, type, NULL); | |
5002 | if (error != 0) | |
5003 | return (error); | |
5004 | } | |
5005 | ||
5006 | /* | |
5007 | * Retrieve the checkpoint txg if the pool has a checkpoint. | |
5008 | */ | |
687e4d7f | 5009 | spa_import_progress_set_notes(spa, "Loading checkpoint txg"); |
d2734cce SD |
5010 | error = spa_ld_read_checkpoint_txg(spa); |
5011 | if (error != 0) | |
5012 | return (error); | |
5013 | ||
9eb7b46e PZ |
5014 | /* |
5015 | * Retrieve the mapping of indirect vdevs. Those vdevs were removed | |
5016 | * from the pool and their contents were re-mapped to other vdevs. Note | |
5017 | * that everything that we read before this step must have been | |
5018 | * rewritten on concrete vdevs after the last device removal was | |
5019 | * initiated. Otherwise we could be reading from indirect vdevs before | |
5020 | * we have loaded their mappings. | |
5021 | */ | |
687e4d7f | 5022 | spa_import_progress_set_notes(spa, "Loading indirect vdev metadata"); |
9eb7b46e PZ |
5023 | error = spa_ld_open_indirect_vdev_metadata(spa); |
5024 | if (error != 0) | |
5025 | return (error); | |
5026 | ||
5027 | /* | |
5028 | * Retrieve the full list of active features from the MOS and check if | |
5029 | * they are all supported. | |
5030 | */ | |
687e4d7f | 5031 | spa_import_progress_set_notes(spa, "Checking feature flags"); |
4a0ee12a | 5032 | error = spa_ld_check_features(spa, &missing_feat_write); |
9eb7b46e PZ |
5033 | if (error != 0) |
5034 | return (error); | |
5035 | ||
5036 | /* | |
5037 | * Load several special directories from the MOS needed by the dsl_pool | |
5038 | * layer. | |
5039 | */ | |
687e4d7f | 5040 | spa_import_progress_set_notes(spa, "Loading special MOS directories"); |
9eb7b46e PZ |
5041 | error = spa_ld_load_special_directories(spa); |
5042 | if (error != 0) | |
5043 | return (error); | |
5044 | ||
9eb7b46e PZ |
5045 | /* |
5046 | * Retrieve pool properties from the MOS. | |
5047 | */ | |
687e4d7f | 5048 | spa_import_progress_set_notes(spa, "Loading properties"); |
9eb7b46e PZ |
5049 | error = spa_ld_get_props(spa); |
5050 | if (error != 0) | |
5051 | return (error); | |
5052 | ||
5053 | /* | |
5054 | * Retrieve the list of auxiliary devices - cache devices and spares - | |
5055 | * and open them. | |
5056 | */ | |
687e4d7f | 5057 | spa_import_progress_set_notes(spa, "Loading AUX vdevs"); |
9eb7b46e PZ |
5058 | error = spa_ld_open_aux_vdevs(spa, type); |
5059 | if (error != 0) | |
5060 | return (error); | |
5061 | ||
5062 | /* | |
5063 | * Load the metadata for all vdevs. Also check if unopenable devices | |
5064 | * should be autoreplaced. | |
5065 | */ | |
687e4d7f | 5066 | spa_import_progress_set_notes(spa, "Loading vdev metadata"); |
4a0ee12a | 5067 | error = spa_ld_load_vdev_metadata(spa); |
9eb7b46e PZ |
5068 | if (error != 0) |
5069 | return (error); | |
5070 | ||
687e4d7f | 5071 | spa_import_progress_set_notes(spa, "Loading dedup tables"); |
9eb7b46e PZ |
5072 | error = spa_ld_load_dedup_tables(spa); |
5073 | if (error != 0) | |
5074 | return (error); | |
5075 | ||
687e4d7f | 5076 | spa_import_progress_set_notes(spa, "Loading BRT"); |
67a1b037 PJD |
5077 | error = spa_ld_load_brt(spa); |
5078 | if (error != 0) | |
5079 | return (error); | |
5080 | ||
9eb7b46e PZ |
5081 | /* |
5082 | * Verify the logs now to make sure we don't have any unexpected errors | |
5083 | * when we claim log blocks later. | |
5084 | */ | |
687e4d7f | 5085 | spa_import_progress_set_notes(spa, "Verifying Log Devices"); |
9eb7b46e PZ |
5086 | error = spa_ld_verify_logs(spa, type, ereport); |
5087 | if (error != 0) | |
5088 | return (error); | |
5089 | ||
9ae529ec | 5090 | if (missing_feat_write) { |
6cb8e530 | 5091 | ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); |
9ae529ec CS |
5092 | |
5093 | /* | |
5094 | * At this point, we know that we can open the pool in | |
5095 | * read-only mode but not read-write mode. We now have enough | |
5096 | * information and can return to userland. | |
5097 | */ | |
9eb7b46e PZ |
5098 | return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, |
5099 | ENOTSUP)); | |
9ae529ec CS |
5100 | } |
5101 | ||
572e2857 | 5102 | /* |
9eb7b46e PZ |
5103 | * Traverse the last txgs to make sure the pool was left off in a safe |
5104 | * state. When performing an extreme rewind, we verify the whole pool, | |
5105 | * which can take a very long time. | |
572e2857 | 5106 | */ |
687e4d7f | 5107 | spa_import_progress_set_notes(spa, "Verifying pool data"); |
4a0ee12a | 5108 | error = spa_ld_verify_pool_data(spa); |
9eb7b46e PZ |
5109 | if (error != 0) |
5110 | return (error); | |
572e2857 | 5111 | |
9eb7b46e PZ |
5112 | /* |
5113 | * Calculate the deflated space for the pool. This must be done before | |
5114 | * we write anything to the pool because we'd need to update the space | |
5115 | * accounting using the deflated sizes. | |
5116 | */ | |
687e4d7f | 5117 | spa_import_progress_set_notes(spa, "Calculating deflated space"); |
9eb7b46e PZ |
5118 | spa_update_dspace(spa); |
5119 | ||
5120 | /* | |
5121 | * We have now retrieved all the information we needed to open the | |
5122 | * pool. If we are importing the pool in read-write mode, a few | |
5123 | * additional steps must be performed to finish the import. | |
5124 | */ | |
687e4d7f | 5125 | spa_import_progress_set_notes(spa, "Starting import"); |
6cb8e530 | 5126 | if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || |
428870ff | 5127 | spa->spa_load_max_txg == UINT64_MAX)) { |
6cb8e530 PZ |
5128 | uint64_t config_cache_txg = spa->spa_config_txg; |
5129 | ||
5130 | ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); | |
34dc7c2f | 5131 | |
5caeef02 DB |
5132 | /* |
5133 | * Before we do any zio_write's, complete the raidz expansion | |
5134 | * scratch space copying, if necessary. | |
5135 | */ | |
5136 | if (RRSS_GET_STATE(&spa->spa_uberblock) == RRSS_SCRATCH_VALID) | |
5137 | vdev_raidz_reflow_copy_scratch(spa); | |
5138 | ||
d2734cce SD |
5139 | /* |
5140 | * In case of a checkpoint rewind, log the original txg | |
5141 | * of the checkpointed uberblock. | |
5142 | */ | |
5143 | if (checkpoint_rewind) { | |
5144 | spa_history_log_internal(spa, "checkpoint rewind", | |
5145 | NULL, "rewound state to txg=%llu", | |
5146 | (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); | |
5147 | } | |
5148 | ||
687e4d7f | 5149 | spa_import_progress_set_notes(spa, "Claiming ZIL blocks"); |
34dc7c2f | 5150 | /* |
9eb7b46e | 5151 | * Traverse the ZIL and claim all blocks. |
34dc7c2f | 5152 | */ |
9eb7b46e | 5153 | spa_ld_claim_log_blocks(spa); |
428870ff | 5154 | |
9eb7b46e PZ |
5155 | /* |
5156 | * Kick-off the syncing thread. | |
5157 | */ | |
34dc7c2f BB |
5158 | spa->spa_sync_on = B_TRUE; |
5159 | txg_sync_start(spa->spa_dsl_pool); | |
379ca9cf | 5160 | mmp_thread_start(spa); |
34dc7c2f BB |
5161 | |
5162 | /* | |
428870ff BB |
5163 | * Wait for all claims to sync. We sync up to the highest |
5164 | * claimed log block birth time so that claimed log blocks | |
5165 | * don't appear to be from the future. spa_claim_max_txg | |
9eb7b46e PZ |
5166 | * will have been set for us by ZIL traversal operations |
5167 | * performed above. | |
34dc7c2f | 5168 | */ |
687e4d7f | 5169 | spa_import_progress_set_notes(spa, "Syncing ZIL claims"); |
428870ff | 5170 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); |
34dc7c2f BB |
5171 | |
5172 | /* | |
9eb7b46e PZ |
5173 | * Check if we need to request an update of the config. On the |
5174 | * next sync, we would update the config stored in vdev labels | |
5175 | * and the cachefile (by default /etc/zfs/zpool.cache). | |
34dc7c2f | 5176 | */ |
687e4d7f | 5177 | spa_import_progress_set_notes(spa, "Updating configs"); |
6cb8e530 | 5178 | spa_ld_check_for_config_update(spa, config_cache_txg, |
d2734cce | 5179 | update_config_cache); |
fb5f0bc8 BB |
5180 | |
5181 | /* | |
9a49d3f3 BB |
5182 | * Check if a rebuild was in progress and if so resume it. |
5183 | * Then check all DTLs to see if anything needs resilvering. | |
5184 | * The resilver will be deferred if a rebuild was started. | |
fb5f0bc8 | 5185 | */ |
687e4d7f | 5186 | spa_import_progress_set_notes(spa, "Starting resilvers"); |
9a49d3f3 BB |
5187 | if (vdev_rebuild_active(spa->spa_root_vdev)) { |
5188 | vdev_rebuild_restart(spa); | |
5189 | } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
5190 | vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
fb5f0bc8 | 5191 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
9a49d3f3 | 5192 | } |
428870ff | 5193 | |
6f1ffb06 MA |
5194 | /* |
5195 | * Log the fact that we booted up (so that we can detect if | |
5196 | * we rebooted in the middle of an operation). | |
5197 | */ | |
d5e024cb | 5198 | spa_history_log_version(spa, "open", NULL); |
6f1ffb06 | 5199 | |
687e4d7f DB |
5200 | spa_import_progress_set_notes(spa, |
5201 | "Restarting device removals"); | |
9b2266e3 SD |
5202 | spa_restart_removal(spa); |
5203 | spa_spawn_aux_threads(spa); | |
5204 | ||
428870ff BB |
5205 | /* |
5206 | * Delete any inconsistent datasets. | |
9b2266e3 SD |
5207 | * |
5208 | * Note: | |
5209 | * Since we may be issuing deletes for clones here, | |
5210 | * we make sure to do so after we've spawned all the | |
5211 | * auxiliary threads above (from which the livelist | |
5212 | * deletion zthr is part of). | |
428870ff | 5213 | */ |
687e4d7f DB |
5214 | spa_import_progress_set_notes(spa, |
5215 | "Cleaning up inconsistent objsets"); | |
428870ff BB |
5216 | (void) dmu_objset_find(spa_name(spa), |
5217 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
5218 | ||
5219 | /* | |
5220 | * Clean up any stale temporary dataset userrefs. | |
5221 | */ | |
687e4d7f DB |
5222 | spa_import_progress_set_notes(spa, |
5223 | "Cleaning up temporary userrefs"); | |
428870ff | 5224 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); |
a1d477c2 | 5225 | |
619f0976 | 5226 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
687e4d7f | 5227 | spa_import_progress_set_notes(spa, "Restarting initialize"); |
619f0976 | 5228 | vdev_initialize_restart(spa->spa_root_vdev); |
687e4d7f | 5229 | spa_import_progress_set_notes(spa, "Restarting TRIM"); |
1b939560 BB |
5230 | vdev_trim_restart(spa->spa_root_vdev); |
5231 | vdev_autotrim_restart(spa); | |
619f0976 | 5232 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
687e4d7f | 5233 | spa_import_progress_set_notes(spa, "Finished importing"); |
34dc7c2f BB |
5234 | } |
5235 | ||
ca95f70d | 5236 | spa_import_progress_remove(spa_guid(spa)); |
77f6826b GA |
5237 | spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); |
5238 | ||
4a0ee12a PZ |
5239 | spa_load_note(spa, "LOADED"); |
5240 | ||
428870ff BB |
5241 | return (0); |
5242 | } | |
34dc7c2f | 5243 | |
428870ff | 5244 | static int |
6cb8e530 | 5245 | spa_load_retry(spa_t *spa, spa_load_state_t state) |
428870ff | 5246 | { |
da92d5cb | 5247 | spa_mode_t mode = spa->spa_mode; |
572e2857 | 5248 | |
428870ff BB |
5249 | spa_unload(spa); |
5250 | spa_deactivate(spa); | |
5251 | ||
dea377c0 | 5252 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; |
428870ff | 5253 | |
572e2857 | 5254 | spa_activate(spa, mode); |
428870ff BB |
5255 | spa_async_suspend(spa); |
5256 | ||
4a0ee12a PZ |
5257 | spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", |
5258 | (u_longlong_t)spa->spa_load_max_txg); | |
5259 | ||
6cb8e530 | 5260 | return (spa_load(spa, state, SPA_IMPORT_EXISTING)); |
428870ff BB |
5261 | } |
5262 | ||
9ae529ec CS |
5263 | /* |
5264 | * If spa_load() fails this function will try loading prior txg's. If | |
5265 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
5266 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
5267 | * function will not rewind the pool and will return the same error as | |
5268 | * spa_load(). | |
5269 | */ | |
428870ff | 5270 | static int |
6cb8e530 PZ |
5271 | spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, |
5272 | int rewind_flags) | |
428870ff | 5273 | { |
9ae529ec | 5274 | nvlist_t *loadinfo = NULL; |
428870ff BB |
5275 | nvlist_t *config = NULL; |
5276 | int load_error, rewind_error; | |
5277 | uint64_t safe_rewind_txg; | |
5278 | uint64_t min_txg; | |
5279 | ||
5280 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
5281 | spa->spa_load_max_txg = spa->spa_load_txg; | |
5282 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
5283 | } else { | |
5284 | spa->spa_load_max_txg = max_request; | |
dea377c0 MA |
5285 | if (max_request != UINT64_MAX) |
5286 | spa->spa_extreme_rewind = B_TRUE; | |
428870ff BB |
5287 | } |
5288 | ||
6cb8e530 | 5289 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); |
428870ff BB |
5290 | if (load_error == 0) |
5291 | return (0); | |
d2734cce SD |
5292 | if (load_error == ZFS_ERR_NO_CHECKPOINT) { |
5293 | /* | |
5294 | * When attempting checkpoint-rewind on a pool with no | |
5295 | * checkpoint, we should not attempt to load uberblocks | |
5296 | * from previous txgs when spa_load fails. | |
5297 | */ | |
5298 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
ca95f70d | 5299 | spa_import_progress_remove(spa_guid(spa)); |
d2734cce SD |
5300 | return (load_error); |
5301 | } | |
428870ff BB |
5302 | |
5303 | if (spa->spa_root_vdev != NULL) | |
5304 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
5305 | ||
5306 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
5307 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
5308 | ||
5309 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
5310 | nvlist_free(config); | |
ca95f70d | 5311 | spa_import_progress_remove(spa_guid(spa)); |
428870ff BB |
5312 | return (load_error); |
5313 | } | |
5314 | ||
9ae529ec CS |
5315 | if (state == SPA_LOAD_RECOVER) { |
5316 | /* Price of rolling back is discarding txgs, including log */ | |
428870ff | 5317 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
9ae529ec CS |
5318 | } else { |
5319 | /* | |
5320 | * If we aren't rolling back save the load info from our first | |
5321 | * import attempt so that we can restore it after attempting | |
5322 | * to rewind. | |
5323 | */ | |
5324 | loadinfo = spa->spa_load_info; | |
5325 | spa->spa_load_info = fnvlist_alloc(); | |
5326 | } | |
428870ff BB |
5327 | |
5328 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
5329 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
5330 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
5331 | TXG_INITIAL : safe_rewind_txg; | |
5332 | ||
5333 | /* | |
5334 | * Continue as long as we're finding errors, we're still within | |
5335 | * the acceptable rewind range, and we're still finding uberblocks | |
5336 | */ | |
5337 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
5338 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
5339 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
5340 | spa->spa_extreme_rewind = B_TRUE; | |
6cb8e530 | 5341 | rewind_error = spa_load_retry(spa, state); |
428870ff BB |
5342 | } |
5343 | ||
428870ff BB |
5344 | spa->spa_extreme_rewind = B_FALSE; |
5345 | spa->spa_load_max_txg = UINT64_MAX; | |
5346 | ||
5347 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
5348 | spa_config_set(spa, config); | |
ee6370a7 | 5349 | else |
5350 | nvlist_free(config); | |
428870ff | 5351 | |
9ae529ec CS |
5352 | if (state == SPA_LOAD_RECOVER) { |
5353 | ASSERT3P(loadinfo, ==, NULL); | |
ca95f70d | 5354 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5355 | return (rewind_error); |
5356 | } else { | |
5357 | /* Store the rewind info as part of the initial load info */ | |
5358 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
5359 | spa->spa_load_info); | |
5360 | ||
5361 | /* Restore the initial load info */ | |
5362 | fnvlist_free(spa->spa_load_info); | |
5363 | spa->spa_load_info = loadinfo; | |
5364 | ||
ca95f70d | 5365 | spa_import_progress_remove(spa_guid(spa)); |
9ae529ec CS |
5366 | return (load_error); |
5367 | } | |
34dc7c2f BB |
5368 | } |
5369 | ||
5370 | /* | |
5371 | * Pool Open/Import | |
5372 | * | |
5373 | * The import case is identical to an open except that the configuration is sent | |
5374 | * down from userland, instead of grabbed from the configuration cache. For the | |
5375 | * case of an open, the pool configuration will exist in the | |
5376 | * POOL_STATE_UNINITIALIZED state. | |
5377 | * | |
5378 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
5379 | * the same time open the pool, without having to keep around the spa_t in some | |
5380 | * ambiguous state. | |
5381 | */ | |
5382 | static int | |
a926aab9 AZ |
5383 | spa_open_common(const char *pool, spa_t **spapp, const void *tag, |
5384 | nvlist_t *nvpolicy, nvlist_t **config) | |
34dc7c2f BB |
5385 | { |
5386 | spa_t *spa; | |
572e2857 | 5387 | spa_load_state_t state = SPA_LOAD_OPEN; |
34dc7c2f | 5388 | int error; |
34dc7c2f | 5389 | int locked = B_FALSE; |
526af785 | 5390 | int firstopen = B_FALSE; |
34dc7c2f BB |
5391 | |
5392 | *spapp = NULL; | |
5393 | ||
5394 | /* | |
5395 | * As disgusting as this is, we need to support recursive calls to this | |
5396 | * function because dsl_dir_open() is called during spa_load(), and ends | |
5397 | * up calling spa_open() again. The real fix is to figure out how to | |
5398 | * avoid dsl_dir_open() calling this in the first place. | |
5399 | */ | |
c25b8f99 | 5400 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { |
34dc7c2f BB |
5401 | mutex_enter(&spa_namespace_lock); |
5402 | locked = B_TRUE; | |
5403 | } | |
5404 | ||
5405 | if ((spa = spa_lookup(pool)) == NULL) { | |
5406 | if (locked) | |
5407 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5408 | return (SET_ERROR(ENOENT)); |
34dc7c2f | 5409 | } |
428870ff | 5410 | |
34dc7c2f | 5411 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { |
8a393be3 | 5412 | zpool_load_policy_t policy; |
428870ff | 5413 | |
526af785 PJD |
5414 | firstopen = B_TRUE; |
5415 | ||
8a393be3 | 5416 | zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, |
428870ff | 5417 | &policy); |
8a393be3 | 5418 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) |
428870ff | 5419 | state = SPA_LOAD_RECOVER; |
34dc7c2f | 5420 | |
fb5f0bc8 | 5421 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 5422 | |
428870ff BB |
5423 | if (state != SPA_LOAD_RECOVER) |
5424 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6cb8e530 | 5425 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; |
428870ff | 5426 | |
4a0ee12a | 5427 | zfs_dbgmsg("spa_open_common: opening %s", pool); |
8a393be3 PZ |
5428 | error = spa_load_best(spa, state, policy.zlp_txg, |
5429 | policy.zlp_rewind); | |
34dc7c2f BB |
5430 | |
5431 | if (error == EBADF) { | |
5432 | /* | |
5433 | * If vdev_validate() returns failure (indicated by | |
5434 | * EBADF), it indicates that one of the vdevs indicates | |
5435 | * that the pool has been exported or destroyed. If | |
5436 | * this is the case, the config cache is out of sync and | |
5437 | * we should remove the pool from the namespace. | |
5438 | */ | |
34dc7c2f BB |
5439 | spa_unload(spa); |
5440 | spa_deactivate(spa); | |
55c12724 | 5441 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 5442 | spa_remove(spa); |
34dc7c2f BB |
5443 | if (locked) |
5444 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 5445 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
5446 | } |
5447 | ||
5448 | if (error) { | |
5449 | /* | |
5450 | * We can't open the pool, but we still have useful | |
5451 | * information: the state of each vdev after the | |
5452 | * attempted vdev_open(). Return this to the user. | |
5453 | */ | |
572e2857 | 5454 | if (config != NULL && spa->spa_config) { |
65ad5d11 AJ |
5455 | *config = fnvlist_dup(spa->spa_config); |
5456 | fnvlist_add_nvlist(*config, | |
572e2857 | 5457 | ZPOOL_CONFIG_LOAD_INFO, |
65ad5d11 | 5458 | spa->spa_load_info); |
572e2857 | 5459 | } |
34dc7c2f BB |
5460 | spa_unload(spa); |
5461 | spa_deactivate(spa); | |
428870ff | 5462 | spa->spa_last_open_failed = error; |
34dc7c2f BB |
5463 | if (locked) |
5464 | mutex_exit(&spa_namespace_lock); | |
5465 | *spapp = NULL; | |
5466 | return (error); | |
34dc7c2f | 5467 | } |
34dc7c2f BB |
5468 | } |
5469 | ||
5470 | spa_open_ref(spa, tag); | |
5471 | ||
b128c09f | 5472 | if (config != NULL) |
34dc7c2f | 5473 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
34dc7c2f | 5474 | |
572e2857 BB |
5475 | /* |
5476 | * If we've recovered the pool, pass back any information we | |
5477 | * gathered while doing the load. | |
5478 | */ | |
1bd02680 | 5479 | if (state == SPA_LOAD_RECOVER && config != NULL) { |
65ad5d11 AJ |
5480 | fnvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, |
5481 | spa->spa_load_info); | |
572e2857 BB |
5482 | } |
5483 | ||
428870ff BB |
5484 | if (locked) { |
5485 | spa->spa_last_open_failed = 0; | |
5486 | spa->spa_last_ubsync_txg = 0; | |
5487 | spa->spa_load_txg = 0; | |
5488 | mutex_exit(&spa_namespace_lock); | |
5489 | } | |
5490 | ||
526af785 | 5491 | if (firstopen) |
ec213971 | 5492 | zvol_create_minors_recursive(spa_name(spa)); |
526af785 | 5493 | |
428870ff BB |
5494 | *spapp = spa; |
5495 | ||
34dc7c2f BB |
5496 | return (0); |
5497 | } | |
5498 | ||
428870ff | 5499 | int |
a926aab9 AZ |
5500 | spa_open_rewind(const char *name, spa_t **spapp, const void *tag, |
5501 | nvlist_t *policy, nvlist_t **config) | |
428870ff BB |
5502 | { |
5503 | return (spa_open_common(name, spapp, tag, policy, config)); | |
5504 | } | |
5505 | ||
34dc7c2f | 5506 | int |
a926aab9 | 5507 | spa_open(const char *name, spa_t **spapp, const void *tag) |
34dc7c2f | 5508 | { |
428870ff | 5509 | return (spa_open_common(name, spapp, tag, NULL, NULL)); |
34dc7c2f BB |
5510 | } |
5511 | ||
5512 | /* | |
5513 | * Lookup the given spa_t, incrementing the inject count in the process, | |
5514 | * preventing it from being exported or destroyed. | |
5515 | */ | |
5516 | spa_t * | |
5517 | spa_inject_addref(char *name) | |
5518 | { | |
5519 | spa_t *spa; | |
5520 | ||
5521 | mutex_enter(&spa_namespace_lock); | |
5522 | if ((spa = spa_lookup(name)) == NULL) { | |
5523 | mutex_exit(&spa_namespace_lock); | |
5524 | return (NULL); | |
5525 | } | |
5526 | spa->spa_inject_ref++; | |
5527 | mutex_exit(&spa_namespace_lock); | |
5528 | ||
5529 | return (spa); | |
5530 | } | |
5531 | ||
5532 | void | |
5533 | spa_inject_delref(spa_t *spa) | |
5534 | { | |
5535 | mutex_enter(&spa_namespace_lock); | |
5536 | spa->spa_inject_ref--; | |
5537 | mutex_exit(&spa_namespace_lock); | |
5538 | } | |
5539 | ||
5540 | /* | |
5541 | * Add spares device information to the nvlist. | |
5542 | */ | |
5543 | static void | |
5544 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
5545 | { | |
5546 | nvlist_t **spares; | |
5547 | uint_t i, nspares; | |
5548 | nvlist_t *nvroot; | |
5549 | uint64_t guid; | |
5550 | vdev_stat_t *vs; | |
5551 | uint_t vsc; | |
5552 | uint64_t pool; | |
5553 | ||
9babb374 BB |
5554 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5555 | ||
34dc7c2f BB |
5556 | if (spa->spa_spares.sav_count == 0) |
5557 | return; | |
5558 | ||
65ad5d11 AJ |
5559 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5560 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5561 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
34dc7c2f | 5562 | if (nspares != 0) { |
795075e6 PD |
5563 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5564 | (const nvlist_t * const *)spares, nspares); | |
65ad5d11 AJ |
5565 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
5566 | &spares, &nspares)); | |
34dc7c2f BB |
5567 | |
5568 | /* | |
5569 | * Go through and find any spares which have since been | |
5570 | * repurposed as an active spare. If this is the case, update | |
5571 | * their status appropriately. | |
5572 | */ | |
5573 | for (i = 0; i < nspares; i++) { | |
65ad5d11 AJ |
5574 | guid = fnvlist_lookup_uint64(spares[i], |
5575 | ZPOOL_CONFIG_GUID); | |
a05263b7 AH |
5576 | VERIFY0(nvlist_lookup_uint64_array(spares[i], |
5577 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
b128c09f BB |
5578 | if (spa_spare_exists(guid, &pool, NULL) && |
5579 | pool != 0ULL) { | |
34dc7c2f BB |
5580 | vs->vs_state = VDEV_STATE_CANT_OPEN; |
5581 | vs->vs_aux = VDEV_AUX_SPARED; | |
a05263b7 AH |
5582 | } else { |
5583 | vs->vs_state = | |
5584 | spa->spa_spares.sav_vdevs[i]->vdev_state; | |
34dc7c2f BB |
5585 | } |
5586 | } | |
5587 | } | |
5588 | } | |
5589 | ||
5590 | /* | |
5591 | * Add l2cache device information to the nvlist, including vdev stats. | |
5592 | */ | |
5593 | static void | |
5594 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
5595 | { | |
5596 | nvlist_t **l2cache; | |
5597 | uint_t i, j, nl2cache; | |
5598 | nvlist_t *nvroot; | |
5599 | uint64_t guid; | |
5600 | vdev_t *vd; | |
5601 | vdev_stat_t *vs; | |
5602 | uint_t vsc; | |
5603 | ||
9babb374 BB |
5604 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
5605 | ||
34dc7c2f BB |
5606 | if (spa->spa_l2cache.sav_count == 0) |
5607 | return; | |
5608 | ||
65ad5d11 AJ |
5609 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
5610 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5611 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
34dc7c2f | 5612 | if (nl2cache != 0) { |
795075e6 PD |
5613 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5614 | (const nvlist_t * const *)l2cache, nl2cache); | |
65ad5d11 AJ |
5615 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, |
5616 | &l2cache, &nl2cache)); | |
34dc7c2f BB |
5617 | |
5618 | /* | |
5619 | * Update level 2 cache device stats. | |
5620 | */ | |
5621 | ||
5622 | for (i = 0; i < nl2cache; i++) { | |
65ad5d11 AJ |
5623 | guid = fnvlist_lookup_uint64(l2cache[i], |
5624 | ZPOOL_CONFIG_GUID); | |
34dc7c2f BB |
5625 | |
5626 | vd = NULL; | |
5627 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
5628 | if (guid == | |
5629 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
5630 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
5631 | break; | |
5632 | } | |
5633 | } | |
5634 | ASSERT(vd != NULL); | |
5635 | ||
65ad5d11 AJ |
5636 | VERIFY0(nvlist_lookup_uint64_array(l2cache[i], |
5637 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
34dc7c2f | 5638 | vdev_get_stats(vd, vs); |
193a37cb TH |
5639 | vdev_config_generate_stats(vd, l2cache[i]); |
5640 | ||
34dc7c2f BB |
5641 | } |
5642 | } | |
34dc7c2f BB |
5643 | } |
5644 | ||
9ae529ec | 5645 | static void |
417104bd | 5646 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) |
9ae529ec | 5647 | { |
9ae529ec CS |
5648 | zap_cursor_t zc; |
5649 | zap_attribute_t za; | |
5650 | ||
9ae529ec CS |
5651 | if (spa->spa_feat_for_read_obj != 0) { |
5652 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
5653 | spa->spa_feat_for_read_obj); | |
5654 | zap_cursor_retrieve(&zc, &za) == 0; | |
5655 | zap_cursor_advance(&zc)) { | |
5656 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
5657 | za.za_num_integers == 1); | |
417104bd | 5658 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
5659 | za.za_first_integer)); |
5660 | } | |
5661 | zap_cursor_fini(&zc); | |
5662 | } | |
5663 | ||
5664 | if (spa->spa_feat_for_write_obj != 0) { | |
5665 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
5666 | spa->spa_feat_for_write_obj); | |
5667 | zap_cursor_retrieve(&zc, &za) == 0; | |
5668 | zap_cursor_advance(&zc)) { | |
5669 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
5670 | za.za_num_integers == 1); | |
417104bd | 5671 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
5672 | za.za_first_integer)); |
5673 | } | |
5674 | zap_cursor_fini(&zc); | |
5675 | } | |
417104bd NB |
5676 | } |
5677 | ||
5678 | static void | |
5679 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
5680 | { | |
5681 | int i; | |
5682 | ||
5683 | for (i = 0; i < SPA_FEATURES; i++) { | |
5684 | zfeature_info_t feature = spa_feature_table[i]; | |
5685 | uint64_t refcount; | |
5686 | ||
5687 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
5688 | continue; | |
5689 | ||
5690 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
5691 | } | |
5692 | } | |
5693 | ||
5694 | /* | |
5695 | * Store a list of pool features and their reference counts in the | |
5696 | * config. | |
5697 | * | |
5698 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
5699 | * the pool features and reference counts from disk, then save the list | |
5700 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
5701 | * and refresh its values from the cached reference counts. This | |
5702 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
5703 | * clear' can resume the pool. | |
5704 | */ | |
5705 | static void | |
5706 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
5707 | { | |
4eb30c68 | 5708 | nvlist_t *features; |
417104bd NB |
5709 | |
5710 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
5711 | ||
4eb30c68 NB |
5712 | mutex_enter(&spa->spa_feat_stats_lock); |
5713 | features = spa->spa_feat_stats; | |
5714 | ||
417104bd NB |
5715 | if (features != NULL) { |
5716 | spa_feature_stats_from_cache(spa, features); | |
5717 | } else { | |
5718 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
5719 | spa->spa_feat_stats = features; | |
5720 | spa_feature_stats_from_disk(spa, features); | |
5721 | } | |
9ae529ec | 5722 | |
417104bd NB |
5723 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, |
5724 | features)); | |
4eb30c68 NB |
5725 | |
5726 | mutex_exit(&spa->spa_feat_stats_lock); | |
9ae529ec CS |
5727 | } |
5728 | ||
34dc7c2f | 5729 | int |
9ae529ec CS |
5730 | spa_get_stats(const char *name, nvlist_t **config, |
5731 | char *altroot, size_t buflen) | |
34dc7c2f BB |
5732 | { |
5733 | int error; | |
5734 | spa_t *spa; | |
5735 | ||
5736 | *config = NULL; | |
428870ff | 5737 | error = spa_open_common(name, &spa, FTAG, NULL, config); |
34dc7c2f | 5738 | |
9babb374 BB |
5739 | if (spa != NULL) { |
5740 | /* | |
5741 | * This still leaves a window of inconsistency where the spares | |
5742 | * or l2cache devices could change and the config would be | |
5743 | * self-inconsistent. | |
5744 | */ | |
5745 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
34dc7c2f | 5746 | |
9babb374 | 5747 | if (*config != NULL) { |
572e2857 BB |
5748 | uint64_t loadtimes[2]; |
5749 | ||
5750 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
5751 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
65ad5d11 AJ |
5752 | fnvlist_add_uint64_array(*config, |
5753 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2); | |
572e2857 | 5754 | |
65ad5d11 | 5755 | fnvlist_add_uint64(*config, |
9babb374 | 5756 | ZPOOL_CONFIG_ERRCOUNT, |
018f2604 | 5757 | spa_approx_errlog_size(spa)); |
9babb374 | 5758 | |
cec3a0a1 | 5759 | if (spa_suspended(spa)) { |
65ad5d11 | 5760 | fnvlist_add_uint64(*config, |
9babb374 | 5761 | ZPOOL_CONFIG_SUSPENDED, |
65ad5d11 AJ |
5762 | spa->spa_failmode); |
5763 | fnvlist_add_uint64(*config, | |
cec3a0a1 | 5764 | ZPOOL_CONFIG_SUSPENDED_REASON, |
65ad5d11 | 5765 | spa->spa_suspended); |
cec3a0a1 | 5766 | } |
b128c09f | 5767 | |
9babb374 BB |
5768 | spa_add_spares(spa, *config); |
5769 | spa_add_l2cache(spa, *config); | |
9ae529ec | 5770 | spa_add_feature_stats(spa, *config); |
9babb374 | 5771 | } |
34dc7c2f BB |
5772 | } |
5773 | ||
5774 | /* | |
5775 | * We want to get the alternate root even for faulted pools, so we cheat | |
5776 | * and call spa_lookup() directly. | |
5777 | */ | |
5778 | if (altroot) { | |
5779 | if (spa == NULL) { | |
5780 | mutex_enter(&spa_namespace_lock); | |
5781 | spa = spa_lookup(name); | |
5782 | if (spa) | |
5783 | spa_altroot(spa, altroot, buflen); | |
5784 | else | |
5785 | altroot[0] = '\0'; | |
5786 | spa = NULL; | |
5787 | mutex_exit(&spa_namespace_lock); | |
5788 | } else { | |
5789 | spa_altroot(spa, altroot, buflen); | |
5790 | } | |
5791 | } | |
5792 | ||
9babb374 BB |
5793 | if (spa != NULL) { |
5794 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
34dc7c2f | 5795 | spa_close(spa, FTAG); |
9babb374 | 5796 | } |
34dc7c2f BB |
5797 | |
5798 | return (error); | |
5799 | } | |
5800 | ||
5801 | /* | |
5802 | * Validate that the auxiliary device array is well formed. We must have an | |
5803 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
5804 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
5805 | * specified, as long as they are well-formed. | |
5806 | */ | |
5807 | static int | |
5808 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
5809 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
5810 | vdev_labeltype_t label) | |
5811 | { | |
5812 | nvlist_t **dev; | |
5813 | uint_t i, ndev; | |
5814 | vdev_t *vd; | |
5815 | int error; | |
5816 | ||
b128c09f BB |
5817 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
5818 | ||
34dc7c2f BB |
5819 | /* |
5820 | * It's acceptable to have no devs specified. | |
5821 | */ | |
5822 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
5823 | return (0); | |
5824 | ||
5825 | if (ndev == 0) | |
2e528b49 | 5826 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
5827 | |
5828 | /* | |
5829 | * Make sure the pool is formatted with a version that supports this | |
5830 | * device type. | |
5831 | */ | |
5832 | if (spa_version(spa) < version) | |
2e528b49 | 5833 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
5834 | |
5835 | /* | |
5836 | * Set the pending device list so we correctly handle device in-use | |
5837 | * checking. | |
5838 | */ | |
5839 | sav->sav_pending = dev; | |
5840 | sav->sav_npending = ndev; | |
5841 | ||
5842 | for (i = 0; i < ndev; i++) { | |
5843 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
5844 | mode)) != 0) | |
5845 | goto out; | |
5846 | ||
5847 | if (!vd->vdev_ops->vdev_op_leaf) { | |
5848 | vdev_free(vd); | |
2e528b49 | 5849 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
5850 | goto out; |
5851 | } | |
5852 | ||
34dc7c2f BB |
5853 | vd->vdev_top = vd; |
5854 | ||
5855 | if ((error = vdev_open(vd)) == 0 && | |
5856 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
65ad5d11 AJ |
5857 | fnvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, |
5858 | vd->vdev_guid); | |
34dc7c2f BB |
5859 | } |
5860 | ||
5861 | vdev_free(vd); | |
5862 | ||
5863 | if (error && | |
5864 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
5865 | goto out; | |
5866 | else | |
5867 | error = 0; | |
5868 | } | |
5869 | ||
5870 | out: | |
5871 | sav->sav_pending = NULL; | |
5872 | sav->sav_npending = 0; | |
5873 | return (error); | |
5874 | } | |
5875 | ||
5876 | static int | |
5877 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
5878 | { | |
5879 | int error; | |
5880 | ||
b128c09f BB |
5881 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
5882 | ||
34dc7c2f BB |
5883 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, |
5884 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
5885 | VDEV_LABEL_SPARE)) != 0) { | |
5886 | return (error); | |
5887 | } | |
5888 | ||
5889 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
5890 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
5891 | VDEV_LABEL_L2CACHE)); | |
5892 | } | |
5893 | ||
5894 | static void | |
5895 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
5896 | const char *config) | |
5897 | { | |
5898 | int i; | |
5899 | ||
5900 | if (sav->sav_config != NULL) { | |
5901 | nvlist_t **olddevs; | |
5902 | uint_t oldndevs; | |
5903 | nvlist_t **newdevs; | |
5904 | ||
5905 | /* | |
4e33ba4c | 5906 | * Generate new dev list by concatenating with the |
34dc7c2f BB |
5907 | * current dev list. |
5908 | */ | |
65ad5d11 AJ |
5909 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, config, |
5910 | &olddevs, &oldndevs)); | |
34dc7c2f BB |
5911 | |
5912 | newdevs = kmem_alloc(sizeof (void *) * | |
79c76d5b | 5913 | (ndevs + oldndevs), KM_SLEEP); |
34dc7c2f | 5914 | for (i = 0; i < oldndevs; i++) |
65ad5d11 | 5915 | newdevs[i] = fnvlist_dup(olddevs[i]); |
34dc7c2f | 5916 | for (i = 0; i < ndevs; i++) |
65ad5d11 | 5917 | newdevs[i + oldndevs] = fnvlist_dup(devs[i]); |
34dc7c2f | 5918 | |
65ad5d11 | 5919 | fnvlist_remove(sav->sav_config, config); |
34dc7c2f | 5920 | |
795075e6 PD |
5921 | fnvlist_add_nvlist_array(sav->sav_config, config, |
5922 | (const nvlist_t * const *)newdevs, ndevs + oldndevs); | |
34dc7c2f BB |
5923 | for (i = 0; i < oldndevs + ndevs; i++) |
5924 | nvlist_free(newdevs[i]); | |
5925 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
5926 | } else { | |
5927 | /* | |
5928 | * Generate a new dev list. | |
5929 | */ | |
65ad5d11 | 5930 | sav->sav_config = fnvlist_alloc(); |
795075e6 PD |
5931 | fnvlist_add_nvlist_array(sav->sav_config, config, |
5932 | (const nvlist_t * const *)devs, ndevs); | |
34dc7c2f BB |
5933 | } |
5934 | } | |
5935 | ||
5936 | /* | |
5937 | * Stop and drop level 2 ARC devices | |
5938 | */ | |
5939 | void | |
5940 | spa_l2cache_drop(spa_t *spa) | |
5941 | { | |
5942 | vdev_t *vd; | |
5943 | int i; | |
5944 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
5945 | ||
5946 | for (i = 0; i < sav->sav_count; i++) { | |
5947 | uint64_t pool; | |
5948 | ||
5949 | vd = sav->sav_vdevs[i]; | |
5950 | ASSERT(vd != NULL); | |
5951 | ||
fb5f0bc8 BB |
5952 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && |
5953 | pool != 0ULL && l2arc_vdev_present(vd)) | |
34dc7c2f | 5954 | l2arc_remove_vdev(vd); |
34dc7c2f BB |
5955 | } |
5956 | } | |
5957 | ||
b5256303 TC |
5958 | /* |
5959 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
5960 | * have the encryption feature flag enabled. | |
5961 | */ | |
5962 | static int | |
5963 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
5964 | boolean_t has_encryption) | |
5965 | { | |
5966 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
5967 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
5968 | !has_encryption) | |
5969 | return (SET_ERROR(ENOTSUP)); | |
5970 | ||
1fff937a | 5971 | return (dmu_objset_create_crypt_check(NULL, dcp, NULL)); |
b5256303 TC |
5972 | } |
5973 | ||
34dc7c2f BB |
5974 | /* |
5975 | * Pool Creation | |
5976 | */ | |
5977 | int | |
5978 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
b5256303 | 5979 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) |
34dc7c2f BB |
5980 | { |
5981 | spa_t *spa; | |
d1807f16 | 5982 | const char *altroot = NULL; |
34dc7c2f BB |
5983 | vdev_t *rvd; |
5984 | dsl_pool_t *dp; | |
5985 | dmu_tx_t *tx; | |
9babb374 | 5986 | int error = 0; |
34dc7c2f BB |
5987 | uint64_t txg = TXG_INITIAL; |
5988 | nvlist_t **spares, **l2cache; | |
5989 | uint_t nspares, nl2cache; | |
b2255edc | 5990 | uint64_t version, obj, ndraid = 0; |
9ae529ec | 5991 | boolean_t has_features; |
b5256303 | 5992 | boolean_t has_encryption; |
715c996d | 5993 | boolean_t has_allocclass; |
b5256303 | 5994 | spa_feature_t feat; |
d1807f16 RY |
5995 | const char *feat_name; |
5996 | const char *poolname; | |
83e9986f RY |
5997 | nvlist_t *nvl; |
5998 | ||
cc99f275 DB |
5999 | if (props == NULL || |
6000 | nvlist_lookup_string(props, "tname", &poolname) != 0) | |
83e9986f | 6001 | poolname = (char *)pool; |
34dc7c2f BB |
6002 | |
6003 | /* | |
6004 | * If this pool already exists, return failure. | |
6005 | */ | |
6006 | mutex_enter(&spa_namespace_lock); | |
83e9986f | 6007 | if (spa_lookup(poolname) != NULL) { |
34dc7c2f | 6008 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 6009 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
6010 | } |
6011 | ||
6012 | /* | |
6013 | * Allocate a new spa_t structure. | |
6014 | */ | |
83e9986f RY |
6015 | nvl = fnvlist_alloc(); |
6016 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
34dc7c2f BB |
6017 | (void) nvlist_lookup_string(props, |
6018 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
83e9986f RY |
6019 | spa = spa_add(poolname, nvl, altroot); |
6020 | fnvlist_free(nvl); | |
fb5f0bc8 | 6021 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 6022 | |
34dc7c2f | 6023 | if (props && (error = spa_prop_validate(spa, props))) { |
34dc7c2f BB |
6024 | spa_deactivate(spa); |
6025 | spa_remove(spa); | |
b128c09f | 6026 | mutex_exit(&spa_namespace_lock); |
34dc7c2f BB |
6027 | return (error); |
6028 | } | |
6029 | ||
83e9986f RY |
6030 | /* |
6031 | * Temporary pool names should never be written to disk. | |
6032 | */ | |
6033 | if (poolname != pool) | |
6034 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
6035 | ||
9ae529ec | 6036 | has_features = B_FALSE; |
b5256303 | 6037 | has_encryption = B_FALSE; |
715c996d | 6038 | has_allocclass = B_FALSE; |
1c27024e | 6039 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); |
9ae529ec | 6040 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { |
b5256303 | 6041 | if (zpool_prop_feature(nvpair_name(elem))) { |
9ae529ec | 6042 | has_features = B_TRUE; |
b5256303 TC |
6043 | |
6044 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
6045 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
6046 | if (feat == SPA_FEATURE_ENCRYPTION) | |
6047 | has_encryption = B_TRUE; | |
715c996d | 6048 | if (feat == SPA_FEATURE_ALLOCATION_CLASSES) |
6049 | has_allocclass = B_TRUE; | |
b5256303 TC |
6050 | } |
6051 | } | |
6052 | ||
6053 | /* verify encryption params, if they were provided */ | |
6054 | if (dcp != NULL) { | |
6055 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
6056 | if (error != 0) { | |
6057 | spa_deactivate(spa); | |
6058 | spa_remove(spa); | |
6059 | mutex_exit(&spa_namespace_lock); | |
6060 | return (error); | |
6061 | } | |
9ae529ec | 6062 | } |
c24fa4b1 | 6063 | if (!has_allocclass && zfs_special_devs(nvroot, NULL)) { |
715c996d | 6064 | spa_deactivate(spa); |
6065 | spa_remove(spa); | |
6066 | mutex_exit(&spa_namespace_lock); | |
6067 | return (ENOTSUP); | |
6068 | } | |
9ae529ec CS |
6069 | |
6070 | if (has_features || nvlist_lookup_uint64(props, | |
6071 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
34dc7c2f | 6072 | version = SPA_VERSION; |
9ae529ec CS |
6073 | } |
6074 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
428870ff BB |
6075 | |
6076 | spa->spa_first_txg = txg; | |
6077 | spa->spa_uberblock.ub_txg = txg - 1; | |
34dc7c2f BB |
6078 | spa->spa_uberblock.ub_version = version; |
6079 | spa->spa_ubsync = spa->spa_uberblock; | |
3dfb57a3 | 6080 | spa->spa_load_state = SPA_LOAD_CREATE; |
a1d477c2 MA |
6081 | spa->spa_removing_phys.sr_state = DSS_NONE; |
6082 | spa->spa_removing_phys.sr_removing_vdev = -1; | |
6083 | spa->spa_removing_phys.sr_prev_indirect_vdev = -1; | |
944a3724 | 6084 | spa->spa_indirect_vdevs_loaded = B_TRUE; |
34dc7c2f | 6085 | |
9babb374 BB |
6086 | /* |
6087 | * Create "The Godfather" zio to hold all async IOs | |
6088 | */ | |
e022864d MA |
6089 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
6090 | KM_SLEEP); | |
1c27024e | 6091 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
6092 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
6093 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
6094 | ZIO_FLAG_GODFATHER); | |
6095 | } | |
9babb374 | 6096 | |
34dc7c2f BB |
6097 | /* |
6098 | * Create the root vdev. | |
6099 | */ | |
b128c09f | 6100 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
6101 | |
6102 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
6103 | ||
6104 | ASSERT(error != 0 || rvd != NULL); | |
6105 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
6106 | ||
6107 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
2e528b49 | 6108 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
6109 | |
6110 | if (error == 0 && | |
6111 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
b2255edc BB |
6112 | (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 && |
6113 | (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { | |
cc99f275 DB |
6114 | /* |
6115 | * instantiate the metaslab groups (this will dirty the vdevs) | |
6116 | * we can no longer error exit past this point | |
6117 | */ | |
6118 | for (int c = 0; error == 0 && c < rvd->vdev_children; c++) { | |
6119 | vdev_t *vd = rvd->vdev_child[c]; | |
6120 | ||
6121 | vdev_metaslab_set_size(vd); | |
6122 | vdev_expand(vd, txg); | |
9babb374 | 6123 | } |
34dc7c2f BB |
6124 | } |
6125 | ||
b128c09f | 6126 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6127 | |
6128 | if (error != 0) { | |
6129 | spa_unload(spa); | |
6130 | spa_deactivate(spa); | |
6131 | spa_remove(spa); | |
6132 | mutex_exit(&spa_namespace_lock); | |
6133 | return (error); | |
6134 | } | |
6135 | ||
6136 | /* | |
6137 | * Get the list of spares, if specified. | |
6138 | */ | |
6139 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6140 | &spares, &nspares) == 0) { | |
65ad5d11 AJ |
6141 | spa->spa_spares.sav_config = fnvlist_alloc(); |
6142 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
6143 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
6144 | nspares); | |
b128c09f | 6145 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6146 | spa_load_spares(spa); |
b128c09f | 6147 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6148 | spa->spa_spares.sav_sync = B_TRUE; |
6149 | } | |
6150 | ||
6151 | /* | |
6152 | * Get the list of level 2 cache devices, if specified. | |
6153 | */ | |
6154 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6155 | &l2cache, &nl2cache) == 0) { | |
795075e6 PD |
6156 | VERIFY0(nvlist_alloc(&spa->spa_l2cache.sav_config, |
6157 | NV_UNIQUE_NAME, KM_SLEEP)); | |
65ad5d11 | 6158 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, |
795075e6 PD |
6159 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
6160 | nl2cache); | |
b128c09f | 6161 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6162 | spa_load_l2cache(spa); |
b128c09f | 6163 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6164 | spa->spa_l2cache.sav_sync = B_TRUE; |
6165 | } | |
6166 | ||
9ae529ec | 6167 | spa->spa_is_initializing = B_TRUE; |
b5256303 | 6168 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); |
9ae529ec | 6169 | spa->spa_is_initializing = B_FALSE; |
34dc7c2f | 6170 | |
428870ff BB |
6171 | /* |
6172 | * Create DDTs (dedup tables). | |
6173 | */ | |
6174 | ddt_create(spa); | |
67a1b037 PJD |
6175 | /* |
6176 | * Create BRT table and BRT table object. | |
6177 | */ | |
6178 | brt_create(spa); | |
428870ff BB |
6179 | |
6180 | spa_update_dspace(spa); | |
6181 | ||
34dc7c2f BB |
6182 | tx = dmu_tx_create_assigned(dp, txg); |
6183 | ||
d5e024cb BB |
6184 | /* |
6185 | * Create the pool's history object. | |
6186 | */ | |
6187 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
6188 | spa_history_create_obj(spa, tx); | |
6189 | ||
6190 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
6191 | spa_history_log_version(spa, "create", tx); | |
6192 | ||
34dc7c2f BB |
6193 | /* |
6194 | * Create the pool config object. | |
6195 | */ | |
6196 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
b128c09f | 6197 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, |
34dc7c2f BB |
6198 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); |
6199 | ||
6200 | if (zap_add(spa->spa_meta_objset, | |
6201 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
6202 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
6203 | cmn_err(CE_PANIC, "failed to add pool config"); | |
6204 | } | |
6205 | ||
428870ff BB |
6206 | if (zap_add(spa->spa_meta_objset, |
6207 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
6208 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
6209 | cmn_err(CE_PANIC, "failed to add pool version"); | |
6210 | } | |
6211 | ||
34dc7c2f BB |
6212 | /* Newly created pools with the right version are always deflated. */ |
6213 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6214 | spa->spa_deflate = TRUE; | |
6215 | if (zap_add(spa->spa_meta_objset, | |
6216 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6217 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
6218 | cmn_err(CE_PANIC, "failed to add deflate"); | |
6219 | } | |
6220 | } | |
6221 | ||
6222 | /* | |
428870ff | 6223 | * Create the deferred-free bpobj. Turn off compression |
34dc7c2f BB |
6224 | * because sync-to-convergence takes longer if the blocksize |
6225 | * keeps changing. | |
6226 | */ | |
428870ff BB |
6227 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); |
6228 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
34dc7c2f | 6229 | ZIO_COMPRESS_OFF, tx); |
34dc7c2f | 6230 | if (zap_add(spa->spa_meta_objset, |
428870ff BB |
6231 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, |
6232 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
6233 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
34dc7c2f | 6234 | } |
428870ff BB |
6235 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, |
6236 | spa->spa_meta_objset, obj)); | |
34dc7c2f | 6237 | |
3c67d83a TH |
6238 | /* |
6239 | * Generate some random noise for salted checksums to operate on. | |
6240 | */ | |
6241 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
6242 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
6243 | ||
34dc7c2f BB |
6244 | /* |
6245 | * Set pool properties. | |
6246 | */ | |
6247 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
6248 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
6249 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
9babb374 | 6250 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); |
379ca9cf | 6251 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); |
1b939560 | 6252 | spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM); |
428870ff | 6253 | |
d164b209 BB |
6254 | if (props != NULL) { |
6255 | spa_configfile_set(spa, props, B_FALSE); | |
13fe0198 | 6256 | spa_sync_props(props, tx); |
d164b209 | 6257 | } |
34dc7c2f | 6258 | |
b2255edc BB |
6259 | for (int i = 0; i < ndraid; i++) |
6260 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6261 | ||
34dc7c2f BB |
6262 | dmu_tx_commit(tx); |
6263 | ||
6264 | spa->spa_sync_on = B_TRUE; | |
b5256303 | 6265 | txg_sync_start(dp); |
379ca9cf | 6266 | mmp_thread_start(spa); |
b5256303 | 6267 | txg_wait_synced(dp, txg); |
34dc7c2f | 6268 | |
9d5b5245 SD |
6269 | spa_spawn_aux_threads(spa); |
6270 | ||
55c12724 | 6271 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); |
34dc7c2f | 6272 | |
0c66c32d JG |
6273 | /* |
6274 | * Don't count references from objsets that are already closed | |
6275 | * and are making their way through the eviction process. | |
6276 | */ | |
6277 | spa_evicting_os_wait(spa); | |
424fd7c3 | 6278 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); |
3dfb57a3 | 6279 | spa->spa_load_state = SPA_LOAD_NONE; |
b128c09f | 6280 | |
4759342a JL |
6281 | spa_import_os(spa); |
6282 | ||
d164b209 BB |
6283 | mutex_exit(&spa_namespace_lock); |
6284 | ||
34dc7c2f BB |
6285 | return (0); |
6286 | } | |
6287 | ||
9babb374 BB |
6288 | /* |
6289 | * Import a non-root pool into the system. | |
6290 | */ | |
6291 | int | |
13fe0198 | 6292 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) |
34dc7c2f BB |
6293 | { |
6294 | spa_t *spa; | |
d1807f16 | 6295 | const char *altroot = NULL; |
428870ff | 6296 | spa_load_state_t state = SPA_LOAD_IMPORT; |
8a393be3 | 6297 | zpool_load_policy_t policy; |
da92d5cb | 6298 | spa_mode_t mode = spa_mode_global; |
572e2857 | 6299 | uint64_t readonly = B_FALSE; |
9babb374 | 6300 | int error; |
34dc7c2f BB |
6301 | nvlist_t *nvroot; |
6302 | nvlist_t **spares, **l2cache; | |
6303 | uint_t nspares, nl2cache; | |
34dc7c2f BB |
6304 | |
6305 | /* | |
6306 | * If a pool with this name exists, return failure. | |
6307 | */ | |
6308 | mutex_enter(&spa_namespace_lock); | |
428870ff | 6309 | if (spa_lookup(pool) != NULL) { |
9babb374 | 6310 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 6311 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
6312 | } |
6313 | ||
6314 | /* | |
6315 | * Create and initialize the spa structure. | |
6316 | */ | |
6317 | (void) nvlist_lookup_string(props, | |
6318 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
572e2857 BB |
6319 | (void) nvlist_lookup_uint64(props, |
6320 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
6321 | if (readonly) | |
da92d5cb | 6322 | mode = SPA_MODE_READ; |
428870ff | 6323 | spa = spa_add(pool, config, altroot); |
572e2857 BB |
6324 | spa->spa_import_flags = flags; |
6325 | ||
6326 | /* | |
6327 | * Verbatim import - Take a pool and insert it into the namespace | |
6328 | * as if it had been loaded at boot. | |
6329 | */ | |
6330 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
6331 | if (props != NULL) | |
6332 | spa_configfile_set(spa, props, B_FALSE); | |
6333 | ||
55c12724 | 6334 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE); |
12fa0466 | 6335 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
4a0ee12a | 6336 | zfs_dbgmsg("spa_import: verbatim import of %s", pool); |
572e2857 | 6337 | mutex_exit(&spa_namespace_lock); |
572e2857 BB |
6338 | return (0); |
6339 | } | |
6340 | ||
6341 | spa_activate(spa, mode); | |
34dc7c2f | 6342 | |
9babb374 BB |
6343 | /* |
6344 | * Don't start async tasks until we know everything is healthy. | |
6345 | */ | |
6346 | spa_async_suspend(spa); | |
b128c09f | 6347 | |
8a393be3 PZ |
6348 | zpool_get_load_policy(config, &policy); |
6349 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) | |
572e2857 BB |
6350 | state = SPA_LOAD_RECOVER; |
6351 | ||
6cb8e530 | 6352 | spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; |
572e2857 | 6353 | |
6cb8e530 PZ |
6354 | if (state != SPA_LOAD_RECOVER) { |
6355 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6356 | zfs_dbgmsg("spa_import: importing %s", pool); | |
6357 | } else { | |
6358 | zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " | |
8a393be3 | 6359 | "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); |
6cb8e530 | 6360 | } |
8a393be3 | 6361 | error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); |
428870ff BB |
6362 | |
6363 | /* | |
572e2857 BB |
6364 | * Propagate anything learned while loading the pool and pass it |
6365 | * back to caller (i.e. rewind info, missing devices, etc). | |
428870ff | 6366 | */ |
65ad5d11 | 6367 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info); |
34dc7c2f | 6368 | |
b128c09f | 6369 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6370 | /* |
9babb374 BB |
6371 | * Toss any existing sparelist, as it doesn't have any validity |
6372 | * anymore, and conflicts with spa_has_spare(). | |
34dc7c2f | 6373 | */ |
9babb374 | 6374 | if (spa->spa_spares.sav_config) { |
34dc7c2f BB |
6375 | nvlist_free(spa->spa_spares.sav_config); |
6376 | spa->spa_spares.sav_config = NULL; | |
6377 | spa_load_spares(spa); | |
6378 | } | |
9babb374 | 6379 | if (spa->spa_l2cache.sav_config) { |
34dc7c2f BB |
6380 | nvlist_free(spa->spa_l2cache.sav_config); |
6381 | spa->spa_l2cache.sav_config = NULL; | |
6382 | spa_load_l2cache(spa); | |
6383 | } | |
6384 | ||
65ad5d11 | 6385 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); |
b128c09f | 6386 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 6387 | |
d164b209 BB |
6388 | if (props != NULL) |
6389 | spa_configfile_set(spa, props, B_FALSE); | |
6390 | ||
fb5f0bc8 BB |
6391 | if (error != 0 || (props && spa_writeable(spa) && |
6392 | (error = spa_prop_set(spa, props)))) { | |
9babb374 BB |
6393 | spa_unload(spa); |
6394 | spa_deactivate(spa); | |
6395 | spa_remove(spa); | |
34dc7c2f BB |
6396 | mutex_exit(&spa_namespace_lock); |
6397 | return (error); | |
6398 | } | |
6399 | ||
572e2857 BB |
6400 | spa_async_resume(spa); |
6401 | ||
34dc7c2f BB |
6402 | /* |
6403 | * Override any spares and level 2 cache devices as specified by | |
6404 | * the user, as these may have correct device names/devids, etc. | |
6405 | */ | |
6406 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6407 | &spares, &nspares) == 0) { | |
6408 | if (spa->spa_spares.sav_config) | |
65ad5d11 AJ |
6409 | fnvlist_remove(spa->spa_spares.sav_config, |
6410 | ZPOOL_CONFIG_SPARES); | |
34dc7c2f | 6411 | else |
65ad5d11 AJ |
6412 | spa->spa_spares.sav_config = fnvlist_alloc(); |
6413 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
795075e6 PD |
6414 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, |
6415 | nspares); | |
b128c09f | 6416 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6417 | spa_load_spares(spa); |
b128c09f | 6418 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6419 | spa->spa_spares.sav_sync = B_TRUE; |
6420 | } | |
6421 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6422 | &l2cache, &nl2cache) == 0) { | |
6423 | if (spa->spa_l2cache.sav_config) | |
65ad5d11 AJ |
6424 | fnvlist_remove(spa->spa_l2cache.sav_config, |
6425 | ZPOOL_CONFIG_L2CACHE); | |
34dc7c2f | 6426 | else |
65ad5d11 AJ |
6427 | spa->spa_l2cache.sav_config = fnvlist_alloc(); |
6428 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
795075e6 PD |
6429 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, |
6430 | nl2cache); | |
b128c09f | 6431 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6432 | spa_load_l2cache(spa); |
b128c09f | 6433 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6434 | spa->spa_l2cache.sav_sync = B_TRUE; |
6435 | } | |
6436 | ||
428870ff BB |
6437 | /* |
6438 | * Check for any removed devices. | |
6439 | */ | |
6440 | if (spa->spa_autoreplace) { | |
6441 | spa_aux_check_removed(&spa->spa_spares); | |
6442 | spa_aux_check_removed(&spa->spa_l2cache); | |
6443 | } | |
6444 | ||
fb5f0bc8 | 6445 | if (spa_writeable(spa)) { |
b128c09f BB |
6446 | /* |
6447 | * Update the config cache to include the newly-imported pool. | |
6448 | */ | |
45d1cae3 | 6449 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
b128c09f | 6450 | } |
34dc7c2f | 6451 | |
34dc7c2f | 6452 | /* |
9babb374 BB |
6453 | * It's possible that the pool was expanded while it was exported. |
6454 | * We kick off an async task to handle this for us. | |
34dc7c2f | 6455 | */ |
9babb374 | 6456 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); |
b128c09f | 6457 | |
d5e024cb | 6458 | spa_history_log_version(spa, "import", NULL); |
fb390aaf | 6459 | |
12fa0466 | 6460 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
fb390aaf | 6461 | |
fb390aaf HR |
6462 | mutex_exit(&spa_namespace_lock); |
6463 | ||
ec213971 | 6464 | zvol_create_minors_recursive(pool); |
4a22ba5b | 6465 | |
4759342a JL |
6466 | spa_import_os(spa); |
6467 | ||
b128c09f BB |
6468 | return (0); |
6469 | } | |
6470 | ||
34dc7c2f BB |
6471 | nvlist_t * |
6472 | spa_tryimport(nvlist_t *tryconfig) | |
6473 | { | |
6474 | nvlist_t *config = NULL; | |
d1807f16 | 6475 | const char *poolname, *cachefile; |
34dc7c2f BB |
6476 | spa_t *spa; |
6477 | uint64_t state; | |
d164b209 | 6478 | int error; |
8a393be3 | 6479 | zpool_load_policy_t policy; |
34dc7c2f BB |
6480 | |
6481 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
6482 | return (NULL); | |
6483 | ||
6484 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
6485 | return (NULL); | |
6486 | ||
6487 | /* | |
6488 | * Create and initialize the spa structure. | |
6489 | */ | |
6490 | mutex_enter(&spa_namespace_lock); | |
428870ff | 6491 | spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); |
da92d5cb | 6492 | spa_activate(spa, SPA_MODE_READ); |
34dc7c2f BB |
6493 | |
6494 | /* | |
8a393be3 | 6495 | * Rewind pool if a max txg was provided. |
34dc7c2f | 6496 | */ |
8a393be3 PZ |
6497 | zpool_get_load_policy(spa->spa_config, &policy); |
6498 | if (policy.zlp_txg != UINT64_MAX) { | |
6499 | spa->spa_load_max_txg = policy.zlp_txg; | |
6cb8e530 PZ |
6500 | spa->spa_extreme_rewind = B_TRUE; |
6501 | zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", | |
8a393be3 | 6502 | poolname, (longlong_t)policy.zlp_txg); |
6cb8e530 PZ |
6503 | } else { |
6504 | zfs_dbgmsg("spa_tryimport: importing %s", poolname); | |
6505 | } | |
6506 | ||
6507 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) | |
6508 | == 0) { | |
6509 | zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); | |
6510 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; | |
6511 | } else { | |
6512 | spa->spa_config_source = SPA_CONFIG_SRC_SCAN; | |
6513 | } | |
6514 | ||
82ac409a AH |
6515 | /* |
6516 | * spa_import() relies on a pool config fetched by spa_try_import() | |
6517 | * for spare/cache devices. Import flags are not passed to | |
6518 | * spa_tryimport(), which makes it return early due to a missing log | |
6519 | * device and missing retrieving the cache device and spare eventually. | |
6520 | * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch | |
6521 | * the correct configuration regardless of the missing log device. | |
6522 | */ | |
6523 | spa->spa_import_flags |= ZFS_IMPORT_MISSING_LOG; | |
6524 | ||
6cb8e530 | 6525 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); |
34dc7c2f BB |
6526 | |
6527 | /* | |
6528 | * If 'tryconfig' was at least parsable, return the current config. | |
6529 | */ | |
6530 | if (spa->spa_root_vdev != NULL) { | |
34dc7c2f | 6531 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
65ad5d11 AJ |
6532 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname); |
6533 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state); | |
6534 | fnvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
6535 | spa->spa_uberblock.ub_timestamp); | |
6536 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
6537 | spa->spa_load_info); | |
6538 | fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
6539 | spa->spa_errata); | |
34dc7c2f BB |
6540 | |
6541 | /* | |
6542 | * If the bootfs property exists on this pool then we | |
6543 | * copy it out so that external consumers can tell which | |
6544 | * pools are bootable. | |
6545 | */ | |
d164b209 | 6546 | if ((!error || error == EEXIST) && spa->spa_bootfs) { |
79c76d5b | 6547 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6548 | |
6549 | /* | |
6550 | * We have to play games with the name since the | |
6551 | * pool was opened as TRYIMPORT_NAME. | |
6552 | */ | |
b128c09f | 6553 | if (dsl_dsobj_to_dsname(spa_name(spa), |
34dc7c2f BB |
6554 | spa->spa_bootfs, tmpname) == 0) { |
6555 | char *cp; | |
d1d7e268 MK |
6556 | char *dsname; |
6557 | ||
79c76d5b | 6558 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
6559 | |
6560 | cp = strchr(tmpname, '/'); | |
6561 | if (cp == NULL) { | |
6562 | (void) strlcpy(dsname, tmpname, | |
6563 | MAXPATHLEN); | |
6564 | } else { | |
6565 | (void) snprintf(dsname, MAXPATHLEN, | |
6566 | "%s/%s", poolname, ++cp); | |
6567 | } | |
65ad5d11 AJ |
6568 | fnvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, |
6569 | dsname); | |
34dc7c2f BB |
6570 | kmem_free(dsname, MAXPATHLEN); |
6571 | } | |
6572 | kmem_free(tmpname, MAXPATHLEN); | |
6573 | } | |
6574 | ||
6575 | /* | |
6576 | * Add the list of hot spares and level 2 cache devices. | |
6577 | */ | |
9babb374 | 6578 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
34dc7c2f BB |
6579 | spa_add_spares(spa, config); |
6580 | spa_add_l2cache(spa, config); | |
9babb374 | 6581 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f BB |
6582 | } |
6583 | ||
6584 | spa_unload(spa); | |
6585 | spa_deactivate(spa); | |
6586 | spa_remove(spa); | |
6587 | mutex_exit(&spa_namespace_lock); | |
6588 | ||
6589 | return (config); | |
6590 | } | |
6591 | ||
6592 | /* | |
6593 | * Pool export/destroy | |
6594 | * | |
6595 | * The act of destroying or exporting a pool is very simple. We make sure there | |
6596 | * is no more pending I/O and any references to the pool are gone. Then, we | |
6597 | * update the pool state and sync all the labels to disk, removing the | |
fb5f0bc8 BB |
6598 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then |
6599 | * we don't sync the labels or remove the configuration cache. | |
34dc7c2f BB |
6600 | */ |
6601 | static int | |
4d55ea81 | 6602 | spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig, |
fb5f0bc8 | 6603 | boolean_t force, boolean_t hardforce) |
34dc7c2f | 6604 | { |
f4f50a70 | 6605 | int error; |
34dc7c2f BB |
6606 | spa_t *spa; |
6607 | ||
6608 | if (oldconfig) | |
6609 | *oldconfig = NULL; | |
6610 | ||
da92d5cb | 6611 | if (!(spa_mode_global & SPA_MODE_WRITE)) |
2e528b49 | 6612 | return (SET_ERROR(EROFS)); |
34dc7c2f BB |
6613 | |
6614 | mutex_enter(&spa_namespace_lock); | |
6615 | if ((spa = spa_lookup(pool)) == NULL) { | |
6616 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 6617 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
6618 | } |
6619 | ||
43a85362 SD |
6620 | if (spa->spa_is_exporting) { |
6621 | /* the pool is being exported by another thread */ | |
6622 | mutex_exit(&spa_namespace_lock); | |
6623 | return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS)); | |
6624 | } | |
6625 | spa->spa_is_exporting = B_TRUE; | |
6626 | ||
34dc7c2f BB |
6627 | /* |
6628 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
6629 | * reacquire the namespace lock, and see if we can export. | |
6630 | */ | |
6631 | spa_open_ref(spa, FTAG); | |
6632 | mutex_exit(&spa_namespace_lock); | |
6633 | spa_async_suspend(spa); | |
a0bd735a BP |
6634 | if (spa->spa_zvol_taskq) { |
6635 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
6636 | taskq_wait(spa->spa_zvol_taskq); | |
6637 | } | |
34dc7c2f BB |
6638 | mutex_enter(&spa_namespace_lock); |
6639 | spa_close(spa, FTAG); | |
6640 | ||
d14cfd83 IH |
6641 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) |
6642 | goto export_spa; | |
34dc7c2f | 6643 | /* |
d14cfd83 IH |
6644 | * The pool will be in core if it's openable, in which case we can |
6645 | * modify its state. Objsets may be open only because they're dirty, | |
6646 | * so we have to force it to sync before checking spa_refcnt. | |
34dc7c2f | 6647 | */ |
0c66c32d | 6648 | if (spa->spa_sync_on) { |
34dc7c2f | 6649 | txg_wait_synced(spa->spa_dsl_pool, 0); |
0c66c32d JG |
6650 | spa_evicting_os_wait(spa); |
6651 | } | |
34dc7c2f | 6652 | |
d14cfd83 IH |
6653 | /* |
6654 | * A pool cannot be exported or destroyed if there are active | |
6655 | * references. If we are resetting a pool, allow references by | |
6656 | * fault injection handlers. | |
6657 | */ | |
f4f50a70 WA |
6658 | if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0)) { |
6659 | error = SET_ERROR(EBUSY); | |
6660 | goto fail; | |
d14cfd83 | 6661 | } |
34dc7c2f | 6662 | |
d14cfd83 | 6663 | if (spa->spa_sync_on) { |
88b199c2 | 6664 | vdev_t *rvd = spa->spa_root_vdev; |
b128c09f BB |
6665 | /* |
6666 | * A pool cannot be exported if it has an active shared spare. | |
6667 | * This is to prevent other pools stealing the active spare | |
6668 | * from an exported pool. At user's own will, such pool can | |
6669 | * be forcedly exported. | |
6670 | */ | |
6671 | if (!force && new_state == POOL_STATE_EXPORTED && | |
6672 | spa_has_active_shared_spare(spa)) { | |
f4f50a70 WA |
6673 | error = SET_ERROR(EXDEV); |
6674 | goto fail; | |
b128c09f | 6675 | } |
34dc7c2f | 6676 | |
619f0976 GW |
6677 | /* |
6678 | * We're about to export or destroy this pool. Make sure | |
1b939560 BB |
6679 | * we stop all initialization and trim activity here before |
6680 | * we set the spa_final_txg. This will ensure that all | |
619f0976 GW |
6681 | * dirty data resulting from the initialization is |
6682 | * committed to disk before we unload the pool. | |
6683 | */ | |
88b199c2 RY |
6684 | vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE); |
6685 | vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE); | |
6686 | vdev_autotrim_stop_all(spa); | |
6687 | vdev_rebuild_stop_all(spa); | |
619f0976 | 6688 | |
34dc7c2f BB |
6689 | /* |
6690 | * We want this to be reflected on every label, | |
6691 | * so mark them all dirty. spa_unload() will do the | |
6692 | * final sync that pushes these changes out. | |
6693 | */ | |
fb5f0bc8 | 6694 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { |
b128c09f | 6695 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 6696 | spa->spa_state = new_state; |
88b199c2 | 6697 | vdev_config_dirty(rvd); |
2fb52853 GA |
6698 | spa_config_exit(spa, SCL_ALL, FTAG); |
6699 | } | |
6700 | ||
6701 | /* | |
6702 | * If the log space map feature is enabled and the pool is | |
6703 | * getting exported (but not destroyed), we want to spend some | |
6704 | * time flushing as many metaslabs as we can in an attempt to | |
6705 | * destroy log space maps and save import time. This has to be | |
6706 | * done before we set the spa_final_txg, otherwise | |
6707 | * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs. | |
6708 | * spa_should_flush_logs_on_unload() should be called after | |
6709 | * spa_state has been set to the new_state. | |
6710 | */ | |
6711 | if (spa_should_flush_logs_on_unload(spa)) | |
6712 | spa_unload_log_sm_flush_all(spa); | |
6713 | ||
6714 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
6715 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
428870ff BB |
6716 | spa->spa_final_txg = spa_last_synced_txg(spa) + |
6717 | TXG_DEFER_SIZE + 1; | |
b128c09f | 6718 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
6719 | } |
6720 | } | |
6721 | ||
d14cfd83 | 6722 | export_spa: |
4759342a JL |
6723 | spa_export_os(spa); |
6724 | ||
d5e024cb BB |
6725 | if (new_state == POOL_STATE_DESTROYED) |
6726 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
6727 | else if (new_state == POOL_STATE_EXPORTED) | |
6728 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
34dc7c2f BB |
6729 | |
6730 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
6731 | spa_unload(spa); | |
6732 | spa_deactivate(spa); | |
6733 | } | |
6734 | ||
6735 | if (oldconfig && spa->spa_config) | |
65ad5d11 | 6736 | *oldconfig = fnvlist_dup(spa->spa_config); |
34dc7c2f BB |
6737 | |
6738 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
fb5f0bc8 | 6739 | if (!hardforce) |
55c12724 | 6740 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); |
34dc7c2f | 6741 | spa_remove(spa); |
43a85362 SD |
6742 | } else { |
6743 | /* | |
6744 | * If spa_remove() is not called for this spa_t and | |
6745 | * there is any possibility that it can be reused, | |
6746 | * we make sure to reset the exporting flag. | |
6747 | */ | |
6748 | spa->spa_is_exporting = B_FALSE; | |
34dc7c2f | 6749 | } |
34dc7c2f | 6750 | |
43a85362 | 6751 | mutex_exit(&spa_namespace_lock); |
34dc7c2f | 6752 | return (0); |
f4f50a70 WA |
6753 | |
6754 | fail: | |
6755 | spa->spa_is_exporting = B_FALSE; | |
6756 | spa_async_resume(spa); | |
6757 | mutex_exit(&spa_namespace_lock); | |
6758 | return (error); | |
34dc7c2f BB |
6759 | } |
6760 | ||
6761 | /* | |
6762 | * Destroy a storage pool. | |
6763 | */ | |
6764 | int | |
4d55ea81 | 6765 | spa_destroy(const char *pool) |
34dc7c2f | 6766 | { |
fb5f0bc8 BB |
6767 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, |
6768 | B_FALSE, B_FALSE)); | |
34dc7c2f BB |
6769 | } |
6770 | ||
6771 | /* | |
6772 | * Export a storage pool. | |
6773 | */ | |
6774 | int | |
4d55ea81 | 6775 | spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force, |
fb5f0bc8 | 6776 | boolean_t hardforce) |
34dc7c2f | 6777 | { |
fb5f0bc8 BB |
6778 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, |
6779 | force, hardforce)); | |
34dc7c2f BB |
6780 | } |
6781 | ||
6782 | /* | |
6783 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
6784 | * from the namespace in any way. | |
6785 | */ | |
6786 | int | |
4d55ea81 | 6787 | spa_reset(const char *pool) |
34dc7c2f | 6788 | { |
b128c09f | 6789 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, |
fb5f0bc8 | 6790 | B_FALSE, B_FALSE)); |
34dc7c2f BB |
6791 | } |
6792 | ||
34dc7c2f BB |
6793 | /* |
6794 | * ========================================================================== | |
6795 | * Device manipulation | |
6796 | * ========================================================================== | |
6797 | */ | |
6798 | ||
b2255edc BB |
6799 | /* |
6800 | * This is called as a synctask to increment the draid feature flag | |
6801 | */ | |
6802 | static void | |
6803 | spa_draid_feature_incr(void *arg, dmu_tx_t *tx) | |
6804 | { | |
6805 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6806 | int draid = (int)(uintptr_t)arg; | |
6807 | ||
6808 | for (int c = 0; c < draid; c++) | |
6809 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6810 | } | |
6811 | ||
34dc7c2f BB |
6812 | /* |
6813 | * Add a device to a storage pool. | |
6814 | */ | |
6815 | int | |
6816 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
6817 | { | |
b2255edc | 6818 | uint64_t txg, ndraid = 0; |
fb5f0bc8 | 6819 | int error; |
34dc7c2f BB |
6820 | vdev_t *rvd = spa->spa_root_vdev; |
6821 | vdev_t *vd, *tvd; | |
6822 | nvlist_t **spares, **l2cache; | |
6823 | uint_t nspares, nl2cache; | |
6824 | ||
572e2857 BB |
6825 | ASSERT(spa_writeable(spa)); |
6826 | ||
34dc7c2f BB |
6827 | txg = spa_vdev_enter(spa); |
6828 | ||
6829 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
6830 | VDEV_ALLOC_ADD)) != 0) | |
6831 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
6832 | ||
b128c09f | 6833 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ |
34dc7c2f BB |
6834 | |
6835 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
6836 | &nspares) != 0) | |
6837 | nspares = 0; | |
6838 | ||
6839 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
6840 | &nl2cache) != 0) | |
6841 | nl2cache = 0; | |
6842 | ||
b128c09f | 6843 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) |
34dc7c2f | 6844 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
34dc7c2f | 6845 | |
b128c09f | 6846 | if (vd->vdev_children != 0 && |
b2255edc | 6847 | (error = vdev_create(vd, txg, B_FALSE)) != 0) { |
b128c09f | 6848 | return (spa_vdev_exit(spa, vd, txg, error)); |
b2255edc BB |
6849 | } |
6850 | ||
6851 | /* | |
6852 | * The virtual dRAID spares must be added after vdev tree is created | |
bf169e9f | 6853 | * and the vdev guids are generated. The guid of their associated |
b2255edc BB |
6854 | * dRAID is stored in the config and used when opening the spare. |
6855 | */ | |
6856 | if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid, | |
6857 | rvd->vdev_children)) == 0) { | |
6858 | if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot, | |
6859 | ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) | |
6860 | nspares = 0; | |
6861 | } else { | |
6862 | return (spa_vdev_exit(spa, vd, txg, error)); | |
6863 | } | |
34dc7c2f BB |
6864 | |
6865 | /* | |
6866 | * We must validate the spares and l2cache devices after checking the | |
6867 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
6868 | */ | |
b128c09f | 6869 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) |
34dc7c2f | 6870 | return (spa_vdev_exit(spa, vd, txg, error)); |
34dc7c2f BB |
6871 | |
6872 | /* | |
a1d477c2 MA |
6873 | * If we are in the middle of a device removal, we can only add |
6874 | * devices which match the existing devices in the pool. | |
6875 | * If we are in the middle of a removal, or have some indirect | |
b2255edc | 6876 | * vdevs, we can not add raidz or dRAID top levels. |
34dc7c2f | 6877 | */ |
a1d477c2 MA |
6878 | if (spa->spa_vdev_removal != NULL || |
6879 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
6880 | for (int c = 0; c < vd->vdev_children; c++) { | |
6881 | tvd = vd->vdev_child[c]; | |
6882 | if (spa->spa_vdev_removal != NULL && | |
9e052db4 | 6883 | tvd->vdev_ashift != spa->spa_max_ashift) { |
a1d477c2 MA |
6884 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
6885 | } | |
b2255edc BB |
6886 | /* Fail if top level vdev is raidz or a dRAID */ |
6887 | if (vdev_get_nparity(tvd) != 0) | |
a1d477c2 | 6888 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
b2255edc | 6889 | |
a1d477c2 MA |
6890 | /* |
6891 | * Need the top level mirror to be | |
6892 | * a mirror of leaf vdevs only | |
6893 | */ | |
6894 | if (tvd->vdev_ops == &vdev_mirror_ops) { | |
6895 | for (uint64_t cid = 0; | |
6896 | cid < tvd->vdev_children; cid++) { | |
6897 | vdev_t *cvd = tvd->vdev_child[cid]; | |
6898 | if (!cvd->vdev_ops->vdev_op_leaf) { | |
6899 | return (spa_vdev_exit(spa, vd, | |
6900 | txg, EINVAL)); | |
6901 | } | |
6902 | } | |
6903 | } | |
6904 | } | |
6905 | } | |
6906 | ||
1c27024e | 6907 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
6908 | tvd = vd->vdev_child[c]; |
6909 | vdev_remove_child(vd, tvd); | |
93e28d66 | 6910 | tvd->vdev_id = rvd->vdev_children; |
34dc7c2f BB |
6911 | vdev_add_child(rvd, tvd); |
6912 | vdev_config_dirty(tvd); | |
6913 | } | |
6914 | ||
6915 | if (nspares != 0) { | |
6916 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
6917 | ZPOOL_CONFIG_SPARES); | |
6918 | spa_load_spares(spa); | |
6919 | spa->spa_spares.sav_sync = B_TRUE; | |
6920 | } | |
6921 | ||
6922 | if (nl2cache != 0) { | |
6923 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
6924 | ZPOOL_CONFIG_L2CACHE); | |
6925 | spa_load_l2cache(spa); | |
6926 | spa->spa_l2cache.sav_sync = B_TRUE; | |
6927 | } | |
6928 | ||
b2255edc BB |
6929 | /* |
6930 | * We can't increment a feature while holding spa_vdev so we | |
6931 | * have to do it in a synctask. | |
6932 | */ | |
6933 | if (ndraid != 0) { | |
6934 | dmu_tx_t *tx; | |
6935 | ||
6936 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
6937 | dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr, | |
6938 | (void *)(uintptr_t)ndraid, tx); | |
6939 | dmu_tx_commit(tx); | |
6940 | } | |
6941 | ||
34dc7c2f BB |
6942 | /* |
6943 | * We have to be careful when adding new vdevs to an existing pool. | |
6944 | * If other threads start allocating from these vdevs before we | |
6945 | * sync the config cache, and we lose power, then upon reboot we may | |
6946 | * fail to open the pool because there are DVAs that the config cache | |
6947 | * can't translate. Therefore, we first add the vdevs without | |
6948 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
6949 | * and then let spa_config_update() initialize the new metaslabs. | |
6950 | * | |
6951 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
6952 | * if we lose power at any point in this sequence, the remaining | |
6953 | * steps will be completed the next time we load the pool. | |
6954 | */ | |
6955 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
6956 | ||
6957 | mutex_enter(&spa_namespace_lock); | |
6958 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
12fa0466 | 6959 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); |
34dc7c2f BB |
6960 | mutex_exit(&spa_namespace_lock); |
6961 | ||
6962 | return (0); | |
6963 | } | |
6964 | ||
6965 | /* | |
5caeef02 DB |
6966 | * Attach a device to a vdev specified by its guid. The vdev type can be |
6967 | * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a | |
6968 | * single device). When the vdev is a single device, a mirror vdev will be | |
6969 | * automatically inserted. | |
34dc7c2f BB |
6970 | * |
6971 | * If 'replacing' is specified, the new device is intended to replace the | |
6972 | * existing device; in this case the two devices are made into their own | |
6973 | * mirror using the 'replacing' vdev, which is functionally identical to | |
6974 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
6975 | * extra rules: you can't attach to it after it's been created, and upon | |
6976 | * completion of resilvering, the first disk (the one being replaced) | |
6977 | * is automatically detached. | |
9a49d3f3 BB |
6978 | * |
6979 | * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild) | |
6980 | * should be performed instead of traditional healing reconstruction. From | |
6981 | * an administrators perspective these are both resilver operations. | |
34dc7c2f BB |
6982 | */ |
6983 | int | |
9a49d3f3 BB |
6984 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing, |
6985 | int rebuild) | |
34dc7c2f | 6986 | { |
428870ff | 6987 | uint64_t txg, dtl_max_txg; |
9a49d3f3 | 6988 | vdev_t *rvd = spa->spa_root_vdev; |
34dc7c2f BB |
6989 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; |
6990 | vdev_ops_t *pvops; | |
b128c09f | 6991 | char *oldvdpath, *newvdpath; |
5caeef02 | 6992 | int newvd_isspare = B_FALSE; |
b128c09f | 6993 | int error; |
34dc7c2f | 6994 | |
572e2857 BB |
6995 | ASSERT(spa_writeable(spa)); |
6996 | ||
34dc7c2f BB |
6997 | txg = spa_vdev_enter(spa); |
6998 | ||
b128c09f | 6999 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 7000 | |
d2734cce SD |
7001 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7002 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7003 | error = (spa_has_checkpoint(spa)) ? | |
7004 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7005 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7006 | } | |
7007 | ||
9a49d3f3 BB |
7008 | if (rebuild) { |
7009 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD)) | |
7010 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7011 | ||
9d618615 A |
7012 | if (dsl_scan_resilvering(spa_get_dsl(spa)) || |
7013 | dsl_scan_resilver_scheduled(spa_get_dsl(spa))) { | |
9a49d3f3 BB |
7014 | return (spa_vdev_exit(spa, NULL, txg, |
7015 | ZFS_ERR_RESILVER_IN_PROGRESS)); | |
9d618615 | 7016 | } |
9a49d3f3 BB |
7017 | } else { |
7018 | if (vdev_rebuild_active(rvd)) | |
7019 | return (spa_vdev_exit(spa, NULL, txg, | |
7020 | ZFS_ERR_REBUILD_IN_PROGRESS)); | |
7021 | } | |
7022 | ||
5caeef02 DB |
7023 | if (spa->spa_vdev_removal != NULL) { |
7024 | return (spa_vdev_exit(spa, NULL, txg, | |
7025 | ZFS_ERR_DEVRM_IN_PROGRESS)); | |
7026 | } | |
a1d477c2 | 7027 | |
34dc7c2f BB |
7028 | if (oldvd == NULL) |
7029 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7030 | ||
5caeef02 DB |
7031 | boolean_t raidz = oldvd->vdev_ops == &vdev_raidz_ops; |
7032 | ||
7033 | if (raidz) { | |
7034 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_RAIDZ_EXPANSION)) | |
7035 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7036 | ||
7037 | /* | |
7038 | * Can't expand a raidz while prior expand is in progress. | |
7039 | */ | |
7040 | if (spa->spa_raidz_expand != NULL) { | |
7041 | return (spa_vdev_exit(spa, NULL, txg, | |
7042 | ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS)); | |
7043 | } | |
7044 | } else if (!oldvd->vdev_ops->vdev_op_leaf) { | |
34dc7c2f | 7045 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
5caeef02 | 7046 | } |
34dc7c2f | 7047 | |
5caeef02 DB |
7048 | if (raidz) |
7049 | pvd = oldvd; | |
7050 | else | |
7051 | pvd = oldvd->vdev_parent; | |
34dc7c2f | 7052 | |
6a42939f RY |
7053 | if (spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, |
7054 | VDEV_ALLOC_ATTACH) != 0) | |
34dc7c2f BB |
7055 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); |
7056 | ||
7057 | if (newrootvd->vdev_children != 1) | |
7058 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7059 | ||
7060 | newvd = newrootvd->vdev_child[0]; | |
7061 | ||
7062 | if (!newvd->vdev_ops->vdev_op_leaf) | |
7063 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7064 | ||
7065 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
7066 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
7067 | ||
7068 | /* | |
c23738c7 | 7069 | * log, dedup and special vdevs should not be replaced by spares. |
34dc7c2f | 7070 | */ |
c23738c7 AH |
7071 | if ((oldvd->vdev_top->vdev_alloc_bias != VDEV_BIAS_NONE || |
7072 | oldvd->vdev_top->vdev_islog) && newvd->vdev_isspare) { | |
34dc7c2f | 7073 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
c23738c7 | 7074 | } |
34dc7c2f | 7075 | |
b2255edc BB |
7076 | /* |
7077 | * A dRAID spare can only replace a child of its parent dRAID vdev. | |
7078 | */ | |
7079 | if (newvd->vdev_ops == &vdev_draid_spare_ops && | |
7080 | oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) { | |
7081 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7082 | } | |
7083 | ||
9a49d3f3 BB |
7084 | if (rebuild) { |
7085 | /* | |
b2255edc | 7086 | * For rebuilds, the top vdev must support reconstruction |
9a49d3f3 | 7087 | * using only space maps. This means the only allowable |
b2255edc | 7088 | * vdevs types are the root vdev, a mirror, or dRAID. |
9a49d3f3 | 7089 | */ |
b2255edc BB |
7090 | tvd = pvd; |
7091 | if (pvd->vdev_top != NULL) | |
7092 | tvd = pvd->vdev_top; | |
7093 | ||
7094 | if (tvd->vdev_ops != &vdev_mirror_ops && | |
7095 | tvd->vdev_ops != &vdev_root_ops && | |
7096 | tvd->vdev_ops != &vdev_draid_ops) { | |
9a49d3f3 BB |
7097 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
7098 | } | |
7099 | } | |
7100 | ||
34dc7c2f BB |
7101 | if (!replacing) { |
7102 | /* | |
7bbd42ef DB |
7103 | * For attach, the only allowable parent is a mirror or |
7104 | * the root vdev. A raidz vdev can be attached to, but | |
7105 | * you cannot attach to a raidz child. | |
34dc7c2f BB |
7106 | */ |
7107 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
7bbd42ef DB |
7108 | pvd->vdev_ops != &vdev_root_ops && |
7109 | !raidz) | |
34dc7c2f BB |
7110 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
7111 | ||
7112 | pvops = &vdev_mirror_ops; | |
7113 | } else { | |
7114 | /* | |
7115 | * Active hot spares can only be replaced by inactive hot | |
7116 | * spares. | |
7117 | */ | |
7118 | if (pvd->vdev_ops == &vdev_spare_ops && | |
572e2857 | 7119 | oldvd->vdev_isspare && |
34dc7c2f BB |
7120 | !spa_has_spare(spa, newvd->vdev_guid)) |
7121 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7122 | ||
7123 | /* | |
7124 | * If the source is a hot spare, and the parent isn't already a | |
7125 | * spare, then we want to create a new hot spare. Otherwise, we | |
7126 | * want to create a replacing vdev. The user is not allowed to | |
7127 | * attach to a spared vdev child unless the 'isspare' state is | |
7128 | * the same (spare replaces spare, non-spare replaces | |
7129 | * non-spare). | |
7130 | */ | |
572e2857 BB |
7131 | if (pvd->vdev_ops == &vdev_replacing_ops && |
7132 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
34dc7c2f | 7133 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
7134 | } else if (pvd->vdev_ops == &vdev_spare_ops && |
7135 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
34dc7c2f | 7136 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
7137 | } |
7138 | ||
7139 | if (newvd->vdev_isspare) | |
34dc7c2f BB |
7140 | pvops = &vdev_spare_ops; |
7141 | else | |
7142 | pvops = &vdev_replacing_ops; | |
7143 | } | |
7144 | ||
7145 | /* | |
9babb374 | 7146 | * Make sure the new device is big enough. |
34dc7c2f | 7147 | */ |
5caeef02 DB |
7148 | vdev_t *min_vdev = raidz ? oldvd->vdev_child[0] : oldvd; |
7149 | if (newvd->vdev_asize < vdev_get_min_asize(min_vdev)) | |
34dc7c2f BB |
7150 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
7151 | ||
7152 | /* | |
7153 | * The new device cannot have a higher alignment requirement | |
7154 | * than the top-level vdev. | |
7155 | */ | |
7156 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
9a49d3f3 | 7157 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
34dc7c2f | 7158 | |
5caeef02 DB |
7159 | /* |
7160 | * RAIDZ-expansion-specific checks. | |
7161 | */ | |
7162 | if (raidz) { | |
7163 | if (vdev_raidz_attach_check(newvd) != 0) | |
7164 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7165 | ||
7166 | /* | |
7167 | * Fail early if a child is not healthy or being replaced | |
7168 | */ | |
7169 | for (int i = 0; i < oldvd->vdev_children; i++) { | |
7170 | if (vdev_is_dead(oldvd->vdev_child[i]) || | |
7171 | !oldvd->vdev_child[i]->vdev_ops->vdev_op_leaf) { | |
7172 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7173 | ENXIO)); | |
7174 | } | |
7175 | /* Also fail if reserved boot area is in-use */ | |
7176 | if (vdev_check_boot_reserve(spa, oldvd->vdev_child[i]) | |
7177 | != 0) { | |
7178 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7179 | EADDRINUSE)); | |
7180 | } | |
7181 | } | |
7182 | } | |
7183 | ||
7184 | if (raidz) { | |
7185 | /* | |
7186 | * Note: oldvdpath is freed by spa_strfree(), but | |
7187 | * kmem_asprintf() is freed by kmem_strfree(), so we have to | |
7188 | * move it to a spa_strdup-ed string. | |
7189 | */ | |
7190 | char *tmp = kmem_asprintf("raidz%u-%u", | |
7191 | (uint_t)vdev_get_nparity(oldvd), (uint_t)oldvd->vdev_id); | |
7192 | oldvdpath = spa_strdup(tmp); | |
7193 | kmem_strfree(tmp); | |
7194 | } else { | |
7195 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
7196 | } | |
7197 | newvdpath = spa_strdup(newvd->vdev_path); | |
7198 | ||
34dc7c2f BB |
7199 | /* |
7200 | * If this is an in-place replacement, update oldvd's path and devid | |
7201 | * to make it distinguishable from newvd, and unopenable from now on. | |
7202 | */ | |
5caeef02 | 7203 | if (strcmp(oldvdpath, newvdpath) == 0) { |
34dc7c2f | 7204 | spa_strfree(oldvd->vdev_path); |
5caeef02 | 7205 | oldvd->vdev_path = kmem_alloc(strlen(newvdpath) + 5, |
79c76d5b | 7206 | KM_SLEEP); |
5caeef02 DB |
7207 | (void) sprintf(oldvd->vdev_path, "%s/old", |
7208 | newvdpath); | |
34dc7c2f BB |
7209 | if (oldvd->vdev_devid != NULL) { |
7210 | spa_strfree(oldvd->vdev_devid); | |
7211 | oldvd->vdev_devid = NULL; | |
7212 | } | |
5caeef02 DB |
7213 | spa_strfree(oldvdpath); |
7214 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
34dc7c2f BB |
7215 | } |
7216 | ||
7217 | /* | |
7218 | * If the parent is not a mirror, or if we're replacing, insert the new | |
7219 | * mirror/replacing/spare vdev above oldvd. | |
7220 | */ | |
5caeef02 | 7221 | if (!raidz && pvd->vdev_ops != pvops) { |
34dc7c2f | 7222 | pvd = vdev_add_parent(oldvd, pvops); |
5caeef02 DB |
7223 | ASSERT(pvd->vdev_ops == pvops); |
7224 | ASSERT(oldvd->vdev_parent == pvd); | |
7225 | } | |
34dc7c2f BB |
7226 | |
7227 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
34dc7c2f BB |
7228 | |
7229 | /* | |
7230 | * Extract the new device from its root and add it to pvd. | |
7231 | */ | |
7232 | vdev_remove_child(newrootvd, newvd); | |
7233 | newvd->vdev_id = pvd->vdev_children; | |
428870ff | 7234 | newvd->vdev_crtxg = oldvd->vdev_crtxg; |
34dc7c2f BB |
7235 | vdev_add_child(pvd, newvd); |
7236 | ||
6d82f98c IH |
7237 | /* |
7238 | * Reevaluate the parent vdev state. | |
7239 | */ | |
7240 | vdev_propagate_state(pvd); | |
7241 | ||
34dc7c2f BB |
7242 | tvd = newvd->vdev_top; |
7243 | ASSERT(pvd->vdev_top == tvd); | |
7244 | ASSERT(tvd->vdev_parent == rvd); | |
7245 | ||
7246 | vdev_config_dirty(tvd); | |
7247 | ||
7248 | /* | |
428870ff BB |
7249 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account |
7250 | * for any dmu_sync-ed blocks. It will propagate upward when | |
7251 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
34dc7c2f | 7252 | */ |
428870ff | 7253 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; |
34dc7c2f | 7254 | |
5caeef02 DB |
7255 | if (raidz) { |
7256 | /* | |
7257 | * Wait for the youngest allocations and frees to sync, | |
7258 | * and then wait for the deferral of those frees to finish. | |
7259 | */ | |
7260 | spa_vdev_config_exit(spa, NULL, | |
7261 | txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); | |
34dc7c2f | 7262 | |
5caeef02 DB |
7263 | vdev_initialize_stop_all(tvd, VDEV_INITIALIZE_ACTIVE); |
7264 | vdev_trim_stop_all(tvd, VDEV_TRIM_ACTIVE); | |
7265 | vdev_autotrim_stop_wait(tvd); | |
9babb374 | 7266 | |
5caeef02 | 7267 | dtl_max_txg = spa_vdev_config_enter(spa); |
34dc7c2f | 7268 | |
5caeef02 | 7269 | tvd->vdev_rz_expanding = B_TRUE; |
34dc7c2f | 7270 | |
5caeef02 DB |
7271 | vdev_dirty_leaves(tvd, VDD_DTL, dtl_max_txg); |
7272 | vdev_config_dirty(tvd); | |
9a49d3f3 | 7273 | |
5caeef02 DB |
7274 | dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, |
7275 | dtl_max_txg); | |
7276 | dsl_sync_task_nowait(spa->spa_dsl_pool, vdev_raidz_attach_sync, | |
7277 | newvd, tx); | |
7278 | dmu_tx_commit(tx); | |
9a49d3f3 | 7279 | } else { |
5caeef02 DB |
7280 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, |
7281 | dtl_max_txg - TXG_INITIAL); | |
7282 | ||
7283 | if (newvd->vdev_isspare) { | |
7284 | spa_spare_activate(newvd); | |
7285 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); | |
7286 | } | |
7287 | ||
7288 | newvd_isspare = newvd->vdev_isspare; | |
7289 | ||
7290 | /* | |
7291 | * Mark newvd's DTL dirty in this txg. | |
7292 | */ | |
7293 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
9a49d3f3 | 7294 | |
5caeef02 DB |
7295 | /* |
7296 | * Schedule the resilver or rebuild to restart in the future. | |
7297 | * We do this to ensure that dmu_sync-ed blocks have been | |
7298 | * stitched into the respective datasets. | |
7299 | */ | |
7300 | if (rebuild) { | |
7301 | newvd->vdev_rebuild_txg = txg; | |
7302 | ||
7303 | vdev_rebuild(tvd); | |
9a49d3f3 | 7304 | } else { |
5caeef02 DB |
7305 | newvd->vdev_resilver_txg = txg; |
7306 | ||
7307 | if (dsl_scan_resilvering(spa_get_dsl(spa)) && | |
7308 | spa_feature_is_enabled(spa, | |
7309 | SPA_FEATURE_RESILVER_DEFER)) { | |
7310 | vdev_defer_resilver(newvd); | |
7311 | } else { | |
7312 | dsl_scan_restart_resilver(spa->spa_dsl_pool, | |
7313 | dtl_max_txg); | |
7314 | } | |
9a49d3f3 BB |
7315 | } |
7316 | } | |
428870ff | 7317 | |
fb390aaf | 7318 | if (spa->spa_bootfs) |
12fa0466 | 7319 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); |
fb390aaf | 7320 | |
12fa0466 | 7321 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); |
fb390aaf | 7322 | |
428870ff BB |
7323 | /* |
7324 | * Commit the config | |
7325 | */ | |
7326 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
34dc7c2f | 7327 | |
6f1ffb06 | 7328 | spa_history_log_internal(spa, "vdev attach", NULL, |
428870ff | 7329 | "%s vdev=%s %s vdev=%s", |
45d1cae3 BB |
7330 | replacing && newvd_isspare ? "spare in" : |
7331 | replacing ? "replace" : "attach", newvdpath, | |
7332 | replacing ? "for" : "to", oldvdpath); | |
b128c09f BB |
7333 | |
7334 | spa_strfree(oldvdpath); | |
7335 | spa_strfree(newvdpath); | |
7336 | ||
34dc7c2f BB |
7337 | return (0); |
7338 | } | |
7339 | ||
7340 | /* | |
7341 | * Detach a device from a mirror or replacing vdev. | |
d3cc8b15 | 7342 | * |
34dc7c2f | 7343 | * If 'replace_done' is specified, only detach if the parent |
719534ca | 7344 | * is a replacing or a spare vdev. |
34dc7c2f BB |
7345 | */ |
7346 | int | |
fb5f0bc8 | 7347 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) |
34dc7c2f BB |
7348 | { |
7349 | uint64_t txg; | |
fb5f0bc8 | 7350 | int error; |
2a8ba608 | 7351 | vdev_t *rvd __maybe_unused = spa->spa_root_vdev; |
34dc7c2f BB |
7352 | vdev_t *vd, *pvd, *cvd, *tvd; |
7353 | boolean_t unspare = B_FALSE; | |
d4ed6673 | 7354 | uint64_t unspare_guid = 0; |
428870ff | 7355 | char *vdpath; |
1c27024e | 7356 | |
572e2857 BB |
7357 | ASSERT(spa_writeable(spa)); |
7358 | ||
9a49d3f3 | 7359 | txg = spa_vdev_detach_enter(spa, guid); |
34dc7c2f | 7360 | |
b128c09f | 7361 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 7362 | |
d2734cce SD |
7363 | /* |
7364 | * Besides being called directly from the userland through the | |
7365 | * ioctl interface, spa_vdev_detach() can be potentially called | |
7366 | * at the end of spa_vdev_resilver_done(). | |
7367 | * | |
7368 | * In the regular case, when we have a checkpoint this shouldn't | |
7369 | * happen as we never empty the DTLs of a vdev during the scrub | |
7370 | * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() | |
7371 | * should never get here when we have a checkpoint. | |
7372 | * | |
7373 | * That said, even in a case when we checkpoint the pool exactly | |
7374 | * as spa_vdev_resilver_done() calls this function everything | |
7375 | * should be fine as the resilver will return right away. | |
7376 | */ | |
7377 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7378 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7379 | error = (spa_has_checkpoint(spa)) ? | |
7380 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7381 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7382 | } | |
7383 | ||
34dc7c2f BB |
7384 | if (vd == NULL) |
7385 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7386 | ||
7387 | if (!vd->vdev_ops->vdev_op_leaf) | |
7388 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7389 | ||
7390 | pvd = vd->vdev_parent; | |
7391 | ||
fb5f0bc8 BB |
7392 | /* |
7393 | * If the parent/child relationship is not as expected, don't do it. | |
7394 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
7395 | * vdev that's replacing B with C. The user's intent in replacing | |
7396 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
7397 | * the replace by detaching C, the expected behavior is to end up | |
7398 | * M(A,B). But suppose that right after deciding to detach C, | |
7399 | * the replacement of B completes. We would have M(A,C), and then | |
7400 | * ask to detach C, which would leave us with just A -- not what | |
7401 | * the user wanted. To prevent this, we make sure that the | |
7402 | * parent/child relationship hasn't changed -- in this example, | |
7403 | * that C's parent is still the replacing vdev R. | |
7404 | */ | |
7405 | if (pvd->vdev_guid != pguid && pguid != 0) | |
7406 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
7407 | ||
34dc7c2f | 7408 | /* |
572e2857 | 7409 | * Only 'replacing' or 'spare' vdevs can be replaced. |
34dc7c2f | 7410 | */ |
572e2857 BB |
7411 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && |
7412 | pvd->vdev_ops != &vdev_spare_ops) | |
7413 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
34dc7c2f BB |
7414 | |
7415 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
7416 | spa_version(spa) >= SPA_VERSION_SPARES); | |
7417 | ||
7418 | /* | |
7419 | * Only mirror, replacing, and spare vdevs support detach. | |
7420 | */ | |
7421 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
7422 | pvd->vdev_ops != &vdev_mirror_ops && | |
7423 | pvd->vdev_ops != &vdev_spare_ops) | |
7424 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7425 | ||
7426 | /* | |
fb5f0bc8 BB |
7427 | * If this device has the only valid copy of some data, |
7428 | * we cannot safely detach it. | |
34dc7c2f | 7429 | */ |
fb5f0bc8 | 7430 | if (vdev_dtl_required(vd)) |
34dc7c2f BB |
7431 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
7432 | ||
fb5f0bc8 | 7433 | ASSERT(pvd->vdev_children >= 2); |
34dc7c2f | 7434 | |
b128c09f BB |
7435 | /* |
7436 | * If we are detaching the second disk from a replacing vdev, then | |
7437 | * check to see if we changed the original vdev's path to have "/old" | |
7438 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
7439 | */ | |
572e2857 BB |
7440 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && |
7441 | vd->vdev_path != NULL) { | |
7442 | size_t len = strlen(vd->vdev_path); | |
7443 | ||
1c27024e | 7444 | for (int c = 0; c < pvd->vdev_children; c++) { |
572e2857 BB |
7445 | cvd = pvd->vdev_child[c]; |
7446 | ||
7447 | if (cvd == vd || cvd->vdev_path == NULL) | |
7448 | continue; | |
7449 | ||
7450 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
7451 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
7452 | spa_strfree(cvd->vdev_path); | |
7453 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
7454 | break; | |
7455 | } | |
b128c09f BB |
7456 | } |
7457 | } | |
7458 | ||
34dc7c2f | 7459 | /* |
b2255edc BB |
7460 | * If we are detaching the original disk from a normal spare, then it |
7461 | * implies that the spare should become a real disk, and be removed | |
7462 | * from the active spare list for the pool. dRAID spares on the | |
7463 | * other hand are coupled to the pool and thus should never be removed | |
7464 | * from the spares list. | |
34dc7c2f | 7465 | */ |
b2255edc BB |
7466 | if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) { |
7467 | vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
7468 | ||
7469 | if (last_cvd->vdev_isspare && | |
7470 | last_cvd->vdev_ops != &vdev_draid_spare_ops) { | |
7471 | unspare = B_TRUE; | |
7472 | } | |
7473 | } | |
34dc7c2f BB |
7474 | |
7475 | /* | |
7476 | * Erase the disk labels so the disk can be used for other things. | |
7477 | * This must be done after all other error cases are handled, | |
7478 | * but before we disembowel vd (so we can still do I/O to it). | |
7479 | * But if we can't do it, don't treat the error as fatal -- | |
7480 | * it may be that the unwritability of the disk is the reason | |
7481 | * it's being detached! | |
7482 | */ | |
6a42939f | 7483 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); |
34dc7c2f BB |
7484 | |
7485 | /* | |
7486 | * Remove vd from its parent and compact the parent's children. | |
7487 | */ | |
7488 | vdev_remove_child(pvd, vd); | |
7489 | vdev_compact_children(pvd); | |
7490 | ||
7491 | /* | |
7492 | * Remember one of the remaining children so we can get tvd below. | |
7493 | */ | |
572e2857 | 7494 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; |
34dc7c2f BB |
7495 | |
7496 | /* | |
7497 | * If we need to remove the remaining child from the list of hot spares, | |
fb5f0bc8 BB |
7498 | * do it now, marking the vdev as no longer a spare in the process. |
7499 | * We must do this before vdev_remove_parent(), because that can | |
7500 | * change the GUID if it creates a new toplevel GUID. For a similar | |
7501 | * reason, we must remove the spare now, in the same txg as the detach; | |
7502 | * otherwise someone could attach a new sibling, change the GUID, and | |
7503 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
34dc7c2f BB |
7504 | */ |
7505 | if (unspare) { | |
7506 | ASSERT(cvd->vdev_isspare); | |
7507 | spa_spare_remove(cvd); | |
7508 | unspare_guid = cvd->vdev_guid; | |
fb5f0bc8 | 7509 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); |
572e2857 | 7510 | cvd->vdev_unspare = B_TRUE; |
34dc7c2f BB |
7511 | } |
7512 | ||
428870ff BB |
7513 | /* |
7514 | * If the parent mirror/replacing vdev only has one child, | |
7515 | * the parent is no longer needed. Remove it from the tree. | |
7516 | */ | |
572e2857 BB |
7517 | if (pvd->vdev_children == 1) { |
7518 | if (pvd->vdev_ops == &vdev_spare_ops) | |
7519 | cvd->vdev_unspare = B_FALSE; | |
428870ff | 7520 | vdev_remove_parent(cvd); |
572e2857 BB |
7521 | } |
7522 | ||
428870ff BB |
7523 | /* |
7524 | * We don't set tvd until now because the parent we just removed | |
7525 | * may have been the previous top-level vdev. | |
7526 | */ | |
7527 | tvd = cvd->vdev_top; | |
7528 | ASSERT(tvd->vdev_parent == rvd); | |
7529 | ||
7530 | /* | |
7531 | * Reevaluate the parent vdev state. | |
7532 | */ | |
7533 | vdev_propagate_state(cvd); | |
7534 | ||
7535 | /* | |
7536 | * If the 'autoexpand' property is set on the pool then automatically | |
7537 | * try to expand the size of the pool. For example if the device we | |
7538 | * just detached was smaller than the others, it may be possible to | |
7539 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
7540 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
7541 | */ | |
7542 | if (spa->spa_autoexpand) { | |
7543 | vdev_reopen(tvd); | |
7544 | vdev_expand(tvd, txg); | |
7545 | } | |
7546 | ||
7547 | vdev_config_dirty(tvd); | |
7548 | ||
7549 | /* | |
7550 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
7551 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
7552 | * But first make sure we're not on any *other* txg's DTL list, to | |
7553 | * prevent vd from being accessed after it's freed. | |
7554 | */ | |
b6ca6193 | 7555 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); |
1c27024e | 7556 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff BB |
7557 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); |
7558 | vd->vdev_detached = B_TRUE; | |
7559 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
7560 | ||
12fa0466 | 7561 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); |
e60e158e | 7562 | spa_notify_waiters(spa); |
428870ff | 7563 | |
572e2857 BB |
7564 | /* hang on to the spa before we release the lock */ |
7565 | spa_open_ref(spa, FTAG); | |
7566 | ||
428870ff BB |
7567 | error = spa_vdev_exit(spa, vd, txg, 0); |
7568 | ||
6f1ffb06 | 7569 | spa_history_log_internal(spa, "detach", NULL, |
428870ff BB |
7570 | "vdev=%s", vdpath); |
7571 | spa_strfree(vdpath); | |
7572 | ||
7573 | /* | |
7574 | * If this was the removal of the original device in a hot spare vdev, | |
7575 | * then we want to go through and remove the device from the hot spare | |
7576 | * list of every other pool. | |
7577 | */ | |
7578 | if (unspare) { | |
572e2857 BB |
7579 | spa_t *altspa = NULL; |
7580 | ||
428870ff | 7581 | mutex_enter(&spa_namespace_lock); |
572e2857 BB |
7582 | while ((altspa = spa_next(altspa)) != NULL) { |
7583 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
7584 | altspa == spa) | |
428870ff | 7585 | continue; |
572e2857 BB |
7586 | |
7587 | spa_open_ref(altspa, FTAG); | |
428870ff | 7588 | mutex_exit(&spa_namespace_lock); |
572e2857 | 7589 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); |
428870ff | 7590 | mutex_enter(&spa_namespace_lock); |
572e2857 | 7591 | spa_close(altspa, FTAG); |
428870ff BB |
7592 | } |
7593 | mutex_exit(&spa_namespace_lock); | |
572e2857 BB |
7594 | |
7595 | /* search the rest of the vdevs for spares to remove */ | |
7596 | spa_vdev_resilver_done(spa); | |
428870ff BB |
7597 | } |
7598 | ||
572e2857 BB |
7599 | /* all done with the spa; OK to release */ |
7600 | mutex_enter(&spa_namespace_lock); | |
7601 | spa_close(spa, FTAG); | |
7602 | mutex_exit(&spa_namespace_lock); | |
7603 | ||
428870ff BB |
7604 | return (error); |
7605 | } | |
7606 | ||
c10d37dd GW |
7607 | static int |
7608 | spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7609 | list_t *vd_list) | |
619f0976 | 7610 | { |
c10d37dd GW |
7611 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7612 | ||
619f0976 GW |
7613 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); |
7614 | ||
7615 | /* Look up vdev and ensure it's a leaf. */ | |
7616 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7617 | if (vd == NULL || vd->vdev_detached) { | |
7618 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7619 | return (SET_ERROR(ENODEV)); |
7620 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
7621 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7622 | return (SET_ERROR(EINVAL)); |
7623 | } else if (!vdev_writeable(vd)) { | |
7624 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
619f0976 GW |
7625 | return (SET_ERROR(EROFS)); |
7626 | } | |
7627 | mutex_enter(&vd->vdev_initialize_lock); | |
7628 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7629 | ||
7630 | /* | |
7631 | * When we activate an initialize action we check to see | |
7632 | * if the vdev_initialize_thread is NULL. We do this instead | |
7633 | * of using the vdev_initialize_state since there might be | |
7634 | * a previous initialization process which has completed but | |
7635 | * the thread is not exited. | |
7636 | */ | |
1b939560 | 7637 | if (cmd_type == POOL_INITIALIZE_START && |
619f0976 | 7638 | (vd->vdev_initialize_thread != NULL || |
5caeef02 | 7639 | vd->vdev_top->vdev_removing || vd->vdev_top->vdev_rz_expanding)) { |
619f0976 | 7640 | mutex_exit(&vd->vdev_initialize_lock); |
619f0976 GW |
7641 | return (SET_ERROR(EBUSY)); |
7642 | } else if (cmd_type == POOL_INITIALIZE_CANCEL && | |
7643 | (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE && | |
7644 | vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) { | |
7645 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 GW |
7646 | return (SET_ERROR(ESRCH)); |
7647 | } else if (cmd_type == POOL_INITIALIZE_SUSPEND && | |
7648 | vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) { | |
7649 | mutex_exit(&vd->vdev_initialize_lock); | |
619f0976 | 7650 | return (SET_ERROR(ESRCH)); |
e34e15ed BB |
7651 | } else if (cmd_type == POOL_INITIALIZE_UNINIT && |
7652 | vd->vdev_initialize_thread != NULL) { | |
7653 | mutex_exit(&vd->vdev_initialize_lock); | |
7654 | return (SET_ERROR(EBUSY)); | |
619f0976 GW |
7655 | } |
7656 | ||
7657 | switch (cmd_type) { | |
1b939560 | 7658 | case POOL_INITIALIZE_START: |
619f0976 GW |
7659 | vdev_initialize(vd); |
7660 | break; | |
7661 | case POOL_INITIALIZE_CANCEL: | |
c10d37dd | 7662 | vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list); |
619f0976 GW |
7663 | break; |
7664 | case POOL_INITIALIZE_SUSPEND: | |
c10d37dd | 7665 | vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list); |
619f0976 | 7666 | break; |
e34e15ed BB |
7667 | case POOL_INITIALIZE_UNINIT: |
7668 | vdev_uninitialize(vd); | |
7669 | break; | |
619f0976 GW |
7670 | default: |
7671 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
7672 | } | |
7673 | mutex_exit(&vd->vdev_initialize_lock); | |
7674 | ||
c10d37dd GW |
7675 | return (0); |
7676 | } | |
7677 | ||
7678 | int | |
7679 | spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, | |
7680 | nvlist_t *vdev_errlist) | |
7681 | { | |
7682 | int total_errors = 0; | |
7683 | list_t vd_list; | |
7684 | ||
7685 | list_create(&vd_list, sizeof (vdev_t), | |
7686 | offsetof(vdev_t, vdev_initialize_node)); | |
7687 | ||
7688 | /* | |
7689 | * We hold the namespace lock through the whole function | |
7690 | * to prevent any changes to the pool while we're starting or | |
7691 | * stopping initialization. The config and state locks are held so that | |
7692 | * we can properly assess the vdev state before we commit to | |
7693 | * the initializing operation. | |
7694 | */ | |
7695 | mutex_enter(&spa_namespace_lock); | |
7696 | ||
7697 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
7698 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
7699 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
7700 | ||
7701 | int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type, | |
7702 | &vd_list); | |
7703 | if (error != 0) { | |
7704 | char guid_as_str[MAXNAMELEN]; | |
7705 | ||
7706 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
7707 | "%llu", (unsigned long long)vdev_guid); | |
7708 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
7709 | total_errors++; | |
7710 | } | |
7711 | } | |
7712 | ||
7713 | /* Wait for all initialize threads to stop. */ | |
7714 | vdev_initialize_stop_wait(spa, &vd_list); | |
7715 | ||
619f0976 GW |
7716 | /* Sync out the initializing state */ |
7717 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
7718 | mutex_exit(&spa_namespace_lock); | |
7719 | ||
c10d37dd | 7720 | list_destroy(&vd_list); |
619f0976 | 7721 | |
c10d37dd GW |
7722 | return (total_errors); |
7723 | } | |
619f0976 | 7724 | |
1b939560 BB |
7725 | static int |
7726 | spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7727 | uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list) | |
7728 | { | |
7729 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7730 | ||
7731 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
7732 | ||
7733 | /* Look up vdev and ensure it's a leaf. */ | |
7734 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7735 | if (vd == NULL || vd->vdev_detached) { | |
7736 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7737 | return (SET_ERROR(ENODEV)); | |
7738 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
7739 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7740 | return (SET_ERROR(EINVAL)); | |
7741 | } else if (!vdev_writeable(vd)) { | |
7742 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7743 | return (SET_ERROR(EROFS)); | |
7744 | } else if (!vd->vdev_has_trim) { | |
7745 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7746 | return (SET_ERROR(EOPNOTSUPP)); | |
7747 | } else if (secure && !vd->vdev_has_securetrim) { | |
7748 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7749 | return (SET_ERROR(EOPNOTSUPP)); | |
7750 | } | |
7751 | mutex_enter(&vd->vdev_trim_lock); | |
7752 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
7753 | ||
7754 | /* | |
7755 | * When we activate a TRIM action we check to see if the | |
7756 | * vdev_trim_thread is NULL. We do this instead of using the | |
7757 | * vdev_trim_state since there might be a previous TRIM process | |
7758 | * which has completed but the thread is not exited. | |
7759 | */ | |
7760 | if (cmd_type == POOL_TRIM_START && | |
5caeef02 DB |
7761 | (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing || |
7762 | vd->vdev_top->vdev_rz_expanding)) { | |
1b939560 BB |
7763 | mutex_exit(&vd->vdev_trim_lock); |
7764 | return (SET_ERROR(EBUSY)); | |
7765 | } else if (cmd_type == POOL_TRIM_CANCEL && | |
7766 | (vd->vdev_trim_state != VDEV_TRIM_ACTIVE && | |
7767 | vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) { | |
7768 | mutex_exit(&vd->vdev_trim_lock); | |
7769 | return (SET_ERROR(ESRCH)); | |
7770 | } else if (cmd_type == POOL_TRIM_SUSPEND && | |
7771 | vd->vdev_trim_state != VDEV_TRIM_ACTIVE) { | |
7772 | mutex_exit(&vd->vdev_trim_lock); | |
7773 | return (SET_ERROR(ESRCH)); | |
7774 | } | |
7775 | ||
7776 | switch (cmd_type) { | |
7777 | case POOL_TRIM_START: | |
7778 | vdev_trim(vd, rate, partial, secure); | |
7779 | break; | |
7780 | case POOL_TRIM_CANCEL: | |
7781 | vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list); | |
7782 | break; | |
7783 | case POOL_TRIM_SUSPEND: | |
7784 | vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list); | |
7785 | break; | |
7786 | default: | |
7787 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
7788 | } | |
7789 | mutex_exit(&vd->vdev_trim_lock); | |
7790 | ||
7791 | return (0); | |
7792 | } | |
7793 | ||
7794 | /* | |
7795 | * Initiates a manual TRIM for the requested vdevs. This kicks off individual | |
7796 | * TRIM threads for each child vdev. These threads pass over all of the free | |
7797 | * space in the vdev's metaslabs and issues TRIM commands for that space. | |
7798 | */ | |
7799 | int | |
7800 | spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate, | |
7801 | boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist) | |
7802 | { | |
7803 | int total_errors = 0; | |
7804 | list_t vd_list; | |
7805 | ||
7806 | list_create(&vd_list, sizeof (vdev_t), | |
7807 | offsetof(vdev_t, vdev_trim_node)); | |
7808 | ||
7809 | /* | |
7810 | * We hold the namespace lock through the whole function | |
7811 | * to prevent any changes to the pool while we're starting or | |
7812 | * stopping TRIM. The config and state locks are held so that | |
7813 | * we can properly assess the vdev state before we commit to | |
7814 | * the TRIM operation. | |
7815 | */ | |
7816 | mutex_enter(&spa_namespace_lock); | |
7817 | ||
7818 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
7819 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
7820 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
7821 | ||
7822 | int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type, | |
7823 | rate, partial, secure, &vd_list); | |
7824 | if (error != 0) { | |
7825 | char guid_as_str[MAXNAMELEN]; | |
7826 | ||
7827 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
7828 | "%llu", (unsigned long long)vdev_guid); | |
7829 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
7830 | total_errors++; | |
7831 | } | |
7832 | } | |
7833 | ||
7834 | /* Wait for all TRIM threads to stop. */ | |
7835 | vdev_trim_stop_wait(spa, &vd_list); | |
7836 | ||
7837 | /* Sync out the TRIM state */ | |
7838 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
7839 | mutex_exit(&spa_namespace_lock); | |
7840 | ||
7841 | list_destroy(&vd_list); | |
7842 | ||
7843 | return (total_errors); | |
7844 | } | |
7845 | ||
428870ff BB |
7846 | /* |
7847 | * Split a set of devices from their mirrors, and create a new pool from them. | |
7848 | */ | |
7849 | int | |
a926aab9 | 7850 | spa_vdev_split_mirror(spa_t *spa, const char *newname, nvlist_t *config, |
428870ff BB |
7851 | nvlist_t *props, boolean_t exp) |
7852 | { | |
7853 | int error = 0; | |
7854 | uint64_t txg, *glist; | |
7855 | spa_t *newspa; | |
7856 | uint_t c, children, lastlog; | |
7857 | nvlist_t **child, *nvl, *tmp; | |
7858 | dmu_tx_t *tx; | |
d1807f16 | 7859 | const char *altroot = NULL; |
428870ff BB |
7860 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ |
7861 | boolean_t activate_slog; | |
7862 | ||
572e2857 | 7863 | ASSERT(spa_writeable(spa)); |
428870ff BB |
7864 | |
7865 | txg = spa_vdev_enter(spa); | |
7866 | ||
d2734cce SD |
7867 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
7868 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7869 | error = (spa_has_checkpoint(spa)) ? | |
7870 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7871 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7872 | } | |
7873 | ||
428870ff BB |
7874 | /* clear the log and flush everything up to now */ |
7875 | activate_slog = spa_passivate_log(spa); | |
7876 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
a1d477c2 | 7877 | error = spa_reset_logs(spa); |
428870ff BB |
7878 | txg = spa_vdev_config_enter(spa); |
7879 | ||
7880 | if (activate_slog) | |
7881 | spa_activate_log(spa); | |
7882 | ||
7883 | if (error != 0) | |
7884 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7885 | ||
7886 | /* check new spa name before going any further */ | |
7887 | if (spa_lookup(newname) != NULL) | |
7888 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
7889 | ||
7890 | /* | |
7891 | * scan through all the children to ensure they're all mirrors | |
7892 | */ | |
7893 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
7894 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
7895 | &children) != 0) | |
7896 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7897 | ||
7898 | /* first, check to ensure we've got the right child count */ | |
7899 | rvd = spa->spa_root_vdev; | |
7900 | lastlog = 0; | |
7901 | for (c = 0; c < rvd->vdev_children; c++) { | |
7902 | vdev_t *vd = rvd->vdev_child[c]; | |
7903 | ||
7904 | /* don't count the holes & logs as children */ | |
1b664952 GA |
7905 | if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops && |
7906 | !vdev_is_concrete(vd))) { | |
428870ff BB |
7907 | if (lastlog == 0) |
7908 | lastlog = c; | |
7909 | continue; | |
7910 | } | |
7911 | ||
7912 | lastlog = 0; | |
7913 | } | |
7914 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
7915 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7916 | ||
7917 | /* next, ensure no spare or cache devices are part of the split */ | |
7918 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
7919 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
7920 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7921 | ||
79c76d5b BB |
7922 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); |
7923 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
428870ff BB |
7924 | |
7925 | /* then, loop over each vdev and validate it */ | |
7926 | for (c = 0; c < children; c++) { | |
7927 | uint64_t is_hole = 0; | |
7928 | ||
7929 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
7930 | &is_hole); | |
7931 | ||
7932 | if (is_hole != 0) { | |
7933 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
7934 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
7935 | continue; | |
7936 | } else { | |
2e528b49 | 7937 | error = SET_ERROR(EINVAL); |
428870ff BB |
7938 | break; |
7939 | } | |
7940 | } | |
7941 | ||
1b664952 GA |
7942 | /* deal with indirect vdevs */ |
7943 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ops == | |
7944 | &vdev_indirect_ops) | |
7945 | continue; | |
7946 | ||
428870ff BB |
7947 | /* which disk is going to be split? */ |
7948 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
7949 | &glist[c]) != 0) { | |
2e528b49 | 7950 | error = SET_ERROR(EINVAL); |
428870ff BB |
7951 | break; |
7952 | } | |
7953 | ||
7954 | /* look it up in the spa */ | |
7955 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
7956 | if (vml[c] == NULL) { | |
2e528b49 | 7957 | error = SET_ERROR(ENODEV); |
428870ff BB |
7958 | break; |
7959 | } | |
7960 | ||
7961 | /* make sure there's nothing stopping the split */ | |
7962 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
7963 | vml[c]->vdev_islog || | |
a1d477c2 | 7964 | !vdev_is_concrete(vml[c]) || |
428870ff BB |
7965 | vml[c]->vdev_isspare || |
7966 | vml[c]->vdev_isl2cache || | |
7967 | !vdev_writeable(vml[c]) || | |
7968 | vml[c]->vdev_children != 0 || | |
7969 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
7970 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
2e528b49 | 7971 | error = SET_ERROR(EINVAL); |
428870ff BB |
7972 | break; |
7973 | } | |
7974 | ||
733b5722 RS |
7975 | if (vdev_dtl_required(vml[c]) || |
7976 | vdev_resilver_needed(vml[c], NULL, NULL)) { | |
2e528b49 | 7977 | error = SET_ERROR(EBUSY); |
428870ff BB |
7978 | break; |
7979 | } | |
7980 | ||
7981 | /* we need certain info from the top level */ | |
65ad5d11 AJ |
7982 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, |
7983 | vml[c]->vdev_top->vdev_ms_array); | |
7984 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
7985 | vml[c]->vdev_top->vdev_ms_shift); | |
7986 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
7987 | vml[c]->vdev_top->vdev_asize); | |
7988 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
7989 | vml[c]->vdev_top->vdev_ashift); | |
e0ab3ab5 JS |
7990 | |
7991 | /* transfer per-vdev ZAPs */ | |
7992 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
7993 | VERIFY0(nvlist_add_uint64(child[c], | |
7994 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
7995 | ||
7996 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
7997 | VERIFY0(nvlist_add_uint64(child[c], | |
7998 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
7999 | vml[c]->vdev_parent->vdev_top_zap)); | |
428870ff BB |
8000 | } |
8001 | ||
8002 | if (error != 0) { | |
8003 | kmem_free(vml, children * sizeof (vdev_t *)); | |
8004 | kmem_free(glist, children * sizeof (uint64_t)); | |
8005 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8006 | } | |
8007 | ||
8008 | /* stop writers from using the disks */ | |
8009 | for (c = 0; c < children; c++) { | |
8010 | if (vml[c] != NULL) | |
8011 | vml[c]->vdev_offline = B_TRUE; | |
8012 | } | |
8013 | vdev_reopen(spa->spa_root_vdev); | |
34dc7c2f BB |
8014 | |
8015 | /* | |
428870ff BB |
8016 | * Temporarily record the splitting vdevs in the spa config. This |
8017 | * will disappear once the config is regenerated. | |
34dc7c2f | 8018 | */ |
65ad5d11 AJ |
8019 | nvl = fnvlist_alloc(); |
8020 | fnvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children); | |
428870ff | 8021 | kmem_free(glist, children * sizeof (uint64_t)); |
34dc7c2f | 8022 | |
428870ff | 8023 | mutex_enter(&spa->spa_props_lock); |
65ad5d11 | 8024 | fnvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl); |
428870ff BB |
8025 | mutex_exit(&spa->spa_props_lock); |
8026 | spa->spa_config_splitting = nvl; | |
8027 | vdev_config_dirty(spa->spa_root_vdev); | |
8028 | ||
8029 | /* configure and create the new pool */ | |
65ad5d11 AJ |
8030 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname); |
8031 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
8032 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE); | |
8033 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); | |
8034 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg); | |
8035 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
8036 | spa_generate_guid(NULL)); | |
e0ab3ab5 | 8037 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); |
428870ff BB |
8038 | (void) nvlist_lookup_string(props, |
8039 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
34dc7c2f | 8040 | |
428870ff BB |
8041 | /* add the new pool to the namespace */ |
8042 | newspa = spa_add(newname, config, altroot); | |
e0ab3ab5 | 8043 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
8044 | newspa->spa_config_txg = spa->spa_config_txg; |
8045 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
8046 | ||
8047 | /* release the spa config lock, retaining the namespace lock */ | |
8048 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
8049 | ||
8050 | if (zio_injection_enabled) | |
8051 | zio_handle_panic_injection(spa, FTAG, 1); | |
8052 | ||
8053 | spa_activate(newspa, spa_mode_global); | |
8054 | spa_async_suspend(newspa); | |
8055 | ||
c10d37dd | 8056 | /* |
1b939560 BB |
8057 | * Temporarily stop the initializing and TRIM activity. We set the |
8058 | * state to ACTIVE so that we know to resume initializing or TRIM | |
8059 | * once the split has completed. | |
c10d37dd | 8060 | */ |
1b939560 BB |
8061 | list_t vd_initialize_list; |
8062 | list_create(&vd_initialize_list, sizeof (vdev_t), | |
c10d37dd GW |
8063 | offsetof(vdev_t, vdev_initialize_node)); |
8064 | ||
1b939560 BB |
8065 | list_t vd_trim_list; |
8066 | list_create(&vd_trim_list, sizeof (vdev_t), | |
8067 | offsetof(vdev_t, vdev_trim_node)); | |
8068 | ||
619f0976 | 8069 | for (c = 0; c < children; c++) { |
1b664952 | 8070 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
619f0976 | 8071 | mutex_enter(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
8072 | vdev_initialize_stop(vml[c], |
8073 | VDEV_INITIALIZE_ACTIVE, &vd_initialize_list); | |
619f0976 | 8074 | mutex_exit(&vml[c]->vdev_initialize_lock); |
1b939560 BB |
8075 | |
8076 | mutex_enter(&vml[c]->vdev_trim_lock); | |
8077 | vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list); | |
8078 | mutex_exit(&vml[c]->vdev_trim_lock); | |
619f0976 GW |
8079 | } |
8080 | } | |
1b939560 BB |
8081 | |
8082 | vdev_initialize_stop_wait(spa, &vd_initialize_list); | |
8083 | vdev_trim_stop_wait(spa, &vd_trim_list); | |
8084 | ||
8085 | list_destroy(&vd_initialize_list); | |
8086 | list_destroy(&vd_trim_list); | |
619f0976 | 8087 | |
6cb8e530 | 8088 | newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; |
8b27e08e | 8089 | newspa->spa_is_splitting = B_TRUE; |
6cb8e530 | 8090 | |
428870ff | 8091 | /* create the new pool from the disks of the original pool */ |
6cb8e530 | 8092 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); |
428870ff BB |
8093 | if (error) |
8094 | goto out; | |
8095 | ||
8096 | /* if that worked, generate a real config for the new pool */ | |
8097 | if (newspa->spa_root_vdev != NULL) { | |
65ad5d11 AJ |
8098 | newspa->spa_config_splitting = fnvlist_alloc(); |
8099 | fnvlist_add_uint64(newspa->spa_config_splitting, | |
8100 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)); | |
428870ff BB |
8101 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, |
8102 | B_TRUE)); | |
9babb374 | 8103 | } |
34dc7c2f | 8104 | |
428870ff BB |
8105 | /* set the props */ |
8106 | if (props != NULL) { | |
8107 | spa_configfile_set(newspa, props, B_FALSE); | |
8108 | error = spa_prop_set(newspa, props); | |
8109 | if (error) | |
8110 | goto out; | |
8111 | } | |
34dc7c2f | 8112 | |
428870ff BB |
8113 | /* flush everything */ |
8114 | txg = spa_vdev_config_enter(newspa); | |
8115 | vdev_config_dirty(newspa->spa_root_vdev); | |
8116 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
34dc7c2f | 8117 | |
428870ff BB |
8118 | if (zio_injection_enabled) |
8119 | zio_handle_panic_injection(spa, FTAG, 2); | |
34dc7c2f | 8120 | |
428870ff | 8121 | spa_async_resume(newspa); |
34dc7c2f | 8122 | |
428870ff BB |
8123 | /* finally, update the original pool's config */ |
8124 | txg = spa_vdev_config_enter(spa); | |
8125 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
8126 | error = dmu_tx_assign(tx, TXG_WAIT); | |
8127 | if (error != 0) | |
8128 | dmu_tx_abort(tx); | |
8129 | for (c = 0; c < children; c++) { | |
1b664952 | 8130 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { |
234234ca RS |
8131 | vdev_t *tvd = vml[c]->vdev_top; |
8132 | ||
8133 | /* | |
8134 | * Need to be sure the detachable VDEV is not | |
8135 | * on any *other* txg's DTL list to prevent it | |
8136 | * from being accessed after it's freed. | |
8137 | */ | |
8138 | for (int t = 0; t < TXG_SIZE; t++) { | |
8139 | (void) txg_list_remove_this( | |
8140 | &tvd->vdev_dtl_list, vml[c], t); | |
8141 | } | |
8142 | ||
428870ff BB |
8143 | vdev_split(vml[c]); |
8144 | if (error == 0) | |
6f1ffb06 MA |
8145 | spa_history_log_internal(spa, "detach", tx, |
8146 | "vdev=%s", vml[c]->vdev_path); | |
e0ab3ab5 | 8147 | |
428870ff | 8148 | vdev_free(vml[c]); |
34dc7c2f | 8149 | } |
34dc7c2f | 8150 | } |
e0ab3ab5 | 8151 | spa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
8152 | vdev_config_dirty(spa->spa_root_vdev); |
8153 | spa->spa_config_splitting = NULL; | |
8154 | nvlist_free(nvl); | |
8155 | if (error == 0) | |
8156 | dmu_tx_commit(tx); | |
8157 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
8158 | ||
8159 | if (zio_injection_enabled) | |
8160 | zio_handle_panic_injection(spa, FTAG, 3); | |
8161 | ||
8162 | /* split is complete; log a history record */ | |
6f1ffb06 MA |
8163 | spa_history_log_internal(newspa, "split", NULL, |
8164 | "from pool %s", spa_name(spa)); | |
428870ff | 8165 | |
8b27e08e | 8166 | newspa->spa_is_splitting = B_FALSE; |
428870ff BB |
8167 | kmem_free(vml, children * sizeof (vdev_t *)); |
8168 | ||
8169 | /* if we're not going to mount the filesystems in userland, export */ | |
8170 | if (exp) | |
8171 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
8172 | B_FALSE, B_FALSE); | |
8173 | ||
8174 | return (error); | |
8175 | ||
8176 | out: | |
8177 | spa_unload(newspa); | |
8178 | spa_deactivate(newspa); | |
8179 | spa_remove(newspa); | |
8180 | ||
8181 | txg = spa_vdev_config_enter(spa); | |
8182 | ||
8183 | /* re-online all offlined disks */ | |
8184 | for (c = 0; c < children; c++) { | |
8185 | if (vml[c] != NULL) | |
8186 | vml[c]->vdev_offline = B_FALSE; | |
8187 | } | |
619f0976 | 8188 | |
1b939560 | 8189 | /* restart initializing or trimming disks as necessary */ |
619f0976 | 8190 | spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART); |
1b939560 BB |
8191 | spa_async_request(spa, SPA_ASYNC_TRIM_RESTART); |
8192 | spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART); | |
619f0976 | 8193 | |
428870ff BB |
8194 | vdev_reopen(spa->spa_root_vdev); |
8195 | ||
8196 | nvlist_free(spa->spa_config_splitting); | |
8197 | spa->spa_config_splitting = NULL; | |
8198 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
34dc7c2f | 8199 | |
428870ff | 8200 | kmem_free(vml, children * sizeof (vdev_t *)); |
34dc7c2f BB |
8201 | return (error); |
8202 | } | |
8203 | ||
34dc7c2f BB |
8204 | /* |
8205 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
d3cc8b15 | 8206 | * currently spared, so we can detach it. |
34dc7c2f BB |
8207 | */ |
8208 | static vdev_t * | |
8209 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
8210 | { | |
8211 | vdev_t *newvd, *oldvd; | |
34dc7c2f | 8212 | |
1c27024e | 8213 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
8214 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); |
8215 | if (oldvd != NULL) | |
8216 | return (oldvd); | |
8217 | } | |
8218 | ||
8219 | /* | |
572e2857 BB |
8220 | * Check for a completed replacement. We always consider the first |
8221 | * vdev in the list to be the oldest vdev, and the last one to be | |
8222 | * the newest (see spa_vdev_attach() for how that works). In | |
8223 | * the case where the newest vdev is faulted, we will not automatically | |
8224 | * remove it after a resilver completes. This is OK as it will require | |
8225 | * user intervention to determine which disk the admin wishes to keep. | |
34dc7c2f | 8226 | */ |
572e2857 BB |
8227 | if (vd->vdev_ops == &vdev_replacing_ops) { |
8228 | ASSERT(vd->vdev_children > 1); | |
8229 | ||
8230 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
34dc7c2f | 8231 | oldvd = vd->vdev_child[0]; |
34dc7c2f | 8232 | |
fb5f0bc8 | 8233 | if (vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 8234 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
fb5f0bc8 | 8235 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 8236 | return (oldvd); |
34dc7c2f BB |
8237 | } |
8238 | ||
8239 | /* | |
8240 | * Check for a completed resilver with the 'unspare' flag set. | |
f65fbee1 | 8241 | * Also potentially update faulted state. |
34dc7c2f | 8242 | */ |
572e2857 BB |
8243 | if (vd->vdev_ops == &vdev_spare_ops) { |
8244 | vdev_t *first = vd->vdev_child[0]; | |
8245 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
8246 | ||
8247 | if (last->vdev_unspare) { | |
8248 | oldvd = first; | |
8249 | newvd = last; | |
8250 | } else if (first->vdev_unspare) { | |
8251 | oldvd = last; | |
8252 | newvd = first; | |
8253 | } else { | |
8254 | oldvd = NULL; | |
8255 | } | |
34dc7c2f | 8256 | |
572e2857 | 8257 | if (oldvd != NULL && |
fb5f0bc8 | 8258 | vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 8259 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
572e2857 | 8260 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 8261 | return (oldvd); |
572e2857 | 8262 | |
f65fbee1 JJ |
8263 | vdev_propagate_state(vd); |
8264 | ||
572e2857 BB |
8265 | /* |
8266 | * If there are more than two spares attached to a disk, | |
8267 | * and those spares are not required, then we want to | |
8268 | * attempt to free them up now so that they can be used | |
8269 | * by other pools. Once we're back down to a single | |
8270 | * disk+spare, we stop removing them. | |
8271 | */ | |
8272 | if (vd->vdev_children > 2) { | |
8273 | newvd = vd->vdev_child[1]; | |
8274 | ||
8275 | if (newvd->vdev_isspare && last->vdev_isspare && | |
8276 | vdev_dtl_empty(last, DTL_MISSING) && | |
8277 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
8278 | !vdev_dtl_required(newvd)) | |
8279 | return (newvd); | |
34dc7c2f | 8280 | } |
34dc7c2f BB |
8281 | } |
8282 | ||
8283 | return (NULL); | |
8284 | } | |
8285 | ||
8286 | static void | |
8287 | spa_vdev_resilver_done(spa_t *spa) | |
8288 | { | |
fb5f0bc8 BB |
8289 | vdev_t *vd, *pvd, *ppvd; |
8290 | uint64_t guid, sguid, pguid, ppguid; | |
34dc7c2f | 8291 | |
fb5f0bc8 | 8292 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8293 | |
8294 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
fb5f0bc8 BB |
8295 | pvd = vd->vdev_parent; |
8296 | ppvd = pvd->vdev_parent; | |
34dc7c2f | 8297 | guid = vd->vdev_guid; |
fb5f0bc8 BB |
8298 | pguid = pvd->vdev_guid; |
8299 | ppguid = ppvd->vdev_guid; | |
8300 | sguid = 0; | |
34dc7c2f BB |
8301 | /* |
8302 | * If we have just finished replacing a hot spared device, then | |
8303 | * we need to detach the parent's first child (the original hot | |
8304 | * spare) as well. | |
8305 | */ | |
572e2857 BB |
8306 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && |
8307 | ppvd->vdev_children == 2) { | |
34dc7c2f | 8308 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); |
fb5f0bc8 | 8309 | sguid = ppvd->vdev_child[1]->vdev_guid; |
34dc7c2f | 8310 | } |
5d1f7fb6 GW |
8311 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); |
8312 | ||
fb5f0bc8 BB |
8313 | spa_config_exit(spa, SCL_ALL, FTAG); |
8314 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
34dc7c2f | 8315 | return; |
fb5f0bc8 | 8316 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) |
34dc7c2f | 8317 | return; |
fb5f0bc8 | 8318 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
8319 | } |
8320 | ||
fb5f0bc8 | 8321 | spa_config_exit(spa, SCL_ALL, FTAG); |
9a49d3f3 BB |
8322 | |
8323 | /* | |
8324 | * If a detach was not performed above replace waiters will not have | |
8325 | * been notified. In which case we must do so now. | |
8326 | */ | |
8327 | spa_notify_waiters(spa); | |
34dc7c2f BB |
8328 | } |
8329 | ||
8330 | /* | |
428870ff | 8331 | * Update the stored path or FRU for this vdev. |
34dc7c2f | 8332 | */ |
65c7cc49 | 8333 | static int |
9babb374 BB |
8334 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, |
8335 | boolean_t ispath) | |
34dc7c2f | 8336 | { |
b128c09f | 8337 | vdev_t *vd; |
428870ff | 8338 | boolean_t sync = B_FALSE; |
34dc7c2f | 8339 | |
572e2857 BB |
8340 | ASSERT(spa_writeable(spa)); |
8341 | ||
428870ff | 8342 | spa_vdev_state_enter(spa, SCL_ALL); |
34dc7c2f | 8343 | |
9babb374 | 8344 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
428870ff | 8345 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); |
34dc7c2f BB |
8346 | |
8347 | if (!vd->vdev_ops->vdev_op_leaf) | |
428870ff | 8348 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 8349 | |
9babb374 | 8350 | if (ispath) { |
428870ff BB |
8351 | if (strcmp(value, vd->vdev_path) != 0) { |
8352 | spa_strfree(vd->vdev_path); | |
8353 | vd->vdev_path = spa_strdup(value); | |
8354 | sync = B_TRUE; | |
8355 | } | |
9babb374 | 8356 | } else { |
428870ff BB |
8357 | if (vd->vdev_fru == NULL) { |
8358 | vd->vdev_fru = spa_strdup(value); | |
8359 | sync = B_TRUE; | |
8360 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
9babb374 | 8361 | spa_strfree(vd->vdev_fru); |
428870ff BB |
8362 | vd->vdev_fru = spa_strdup(value); |
8363 | sync = B_TRUE; | |
8364 | } | |
9babb374 | 8365 | } |
34dc7c2f | 8366 | |
428870ff | 8367 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); |
34dc7c2f BB |
8368 | } |
8369 | ||
9babb374 BB |
8370 | int |
8371 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
8372 | { | |
8373 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
8374 | } | |
8375 | ||
8376 | int | |
8377 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
8378 | { | |
8379 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
8380 | } | |
8381 | ||
34dc7c2f BB |
8382 | /* |
8383 | * ========================================================================== | |
428870ff | 8384 | * SPA Scanning |
34dc7c2f BB |
8385 | * ========================================================================== |
8386 | */ | |
0ea05c64 AP |
8387 | int |
8388 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
8389 | { | |
8390 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8391 | ||
8392 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
8393 | return (SET_ERROR(EBUSY)); | |
8394 | ||
8395 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
8396 | } | |
34dc7c2f | 8397 | |
34dc7c2f | 8398 | int |
428870ff BB |
8399 | spa_scan_stop(spa_t *spa) |
8400 | { | |
8401 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8402 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
2e528b49 | 8403 | return (SET_ERROR(EBUSY)); |
482eeef8 | 8404 | |
428870ff BB |
8405 | return (dsl_scan_cancel(spa->spa_dsl_pool)); |
8406 | } | |
8407 | ||
8408 | int | |
8409 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
34dc7c2f | 8410 | { |
b128c09f | 8411 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); |
34dc7c2f | 8412 | |
428870ff | 8413 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) |
2e528b49 | 8414 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 8415 | |
fa241660 TC |
8416 | if (func == POOL_SCAN_RESILVER && |
8417 | !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) | |
8418 | return (SET_ERROR(ENOTSUP)); | |
8419 | ||
34dc7c2f | 8420 | /* |
b128c09f BB |
8421 | * If a resilver was requested, but there is no DTL on a |
8422 | * writeable leaf device, we have nothing to do. | |
34dc7c2f | 8423 | */ |
428870ff | 8424 | if (func == POOL_SCAN_RESILVER && |
b128c09f BB |
8425 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { |
8426 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
34dc7c2f BB |
8427 | return (0); |
8428 | } | |
8429 | ||
482eeef8 GA |
8430 | if (func == POOL_SCAN_ERRORSCRUB && |
8431 | !spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) | |
8432 | return (SET_ERROR(ENOTSUP)); | |
8433 | ||
428870ff | 8434 | return (dsl_scan(spa->spa_dsl_pool, func)); |
34dc7c2f BB |
8435 | } |
8436 | ||
8437 | /* | |
8438 | * ========================================================================== | |
8439 | * SPA async task processing | |
8440 | * ========================================================================== | |
8441 | */ | |
8442 | ||
8443 | static void | |
8444 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
8445 | { | |
b128c09f | 8446 | if (vd->vdev_remove_wanted) { |
428870ff BB |
8447 | vd->vdev_remove_wanted = B_FALSE; |
8448 | vd->vdev_delayed_close = B_FALSE; | |
b128c09f | 8449 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); |
428870ff BB |
8450 | |
8451 | /* | |
8452 | * We want to clear the stats, but we don't want to do a full | |
8453 | * vdev_clear() as that will cause us to throw away | |
8454 | * degraded/faulted state as well as attempt to reopen the | |
8455 | * device, all of which is a waste. | |
8456 | */ | |
8457 | vd->vdev_stat.vs_read_errors = 0; | |
8458 | vd->vdev_stat.vs_write_errors = 0; | |
8459 | vd->vdev_stat.vs_checksum_errors = 0; | |
8460 | ||
b128c09f | 8461 | vdev_state_dirty(vd->vdev_top); |
0aacde2e RM |
8462 | |
8463 | /* Tell userspace that the vdev is gone. */ | |
8464 | zfs_post_remove(spa, vd); | |
b128c09f | 8465 | } |
34dc7c2f | 8466 | |
1c27024e | 8467 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f BB |
8468 | spa_async_remove(spa, vd->vdev_child[c]); |
8469 | } | |
8470 | ||
8471 | static void | |
8472 | spa_async_probe(spa_t *spa, vdev_t *vd) | |
8473 | { | |
8474 | if (vd->vdev_probe_wanted) { | |
428870ff | 8475 | vd->vdev_probe_wanted = B_FALSE; |
b128c09f | 8476 | vdev_reopen(vd); /* vdev_open() does the actual probe */ |
34dc7c2f | 8477 | } |
b128c09f | 8478 | |
1c27024e | 8479 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f | 8480 | spa_async_probe(spa, vd->vdev_child[c]); |
34dc7c2f BB |
8481 | } |
8482 | ||
9babb374 BB |
8483 | static void |
8484 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
8485 | { | |
9babb374 BB |
8486 | if (!spa->spa_autoexpand) |
8487 | return; | |
8488 | ||
1c27024e | 8489 | for (int c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
8490 | vdev_t *cvd = vd->vdev_child[c]; |
8491 | spa_async_autoexpand(spa, cvd); | |
8492 | } | |
8493 | ||
8494 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
8495 | return; | |
8496 | ||
12fa0466 | 8497 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); |
9babb374 BB |
8498 | } |
8499 | ||
460748d4 | 8500 | static __attribute__((noreturn)) void |
c25b8f99 | 8501 | spa_async_thread(void *arg) |
34dc7c2f | 8502 | { |
c25b8f99 | 8503 | spa_t *spa = (spa_t *)arg; |
80a91e74 | 8504 | dsl_pool_t *dp = spa->spa_dsl_pool; |
867959b5 | 8505 | int tasks; |
34dc7c2f BB |
8506 | |
8507 | ASSERT(spa->spa_sync_on); | |
8508 | ||
8509 | mutex_enter(&spa->spa_async_lock); | |
8510 | tasks = spa->spa_async_tasks; | |
8511 | spa->spa_async_tasks = 0; | |
8512 | mutex_exit(&spa->spa_async_lock); | |
8513 | ||
8514 | /* | |
8515 | * See if the config needs to be updated. | |
8516 | */ | |
8517 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
428870ff | 8518 | uint64_t old_space, new_space; |
9babb374 | 8519 | |
34dc7c2f | 8520 | mutex_enter(&spa_namespace_lock); |
428870ff | 8521 | old_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8522 | old_space += metaslab_class_get_space(spa_special_class(spa)); |
8523 | old_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8524 | old_space += metaslab_class_get_space( |
8525 | spa_embedded_log_class(spa)); | |
cc99f275 | 8526 | |
34dc7c2f | 8527 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
cc99f275 | 8528 | |
428870ff | 8529 | new_space = metaslab_class_get_space(spa_normal_class(spa)); |
cc99f275 DB |
8530 | new_space += metaslab_class_get_space(spa_special_class(spa)); |
8531 | new_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
aa755b35 MA |
8532 | new_space += metaslab_class_get_space( |
8533 | spa_embedded_log_class(spa)); | |
34dc7c2f | 8534 | mutex_exit(&spa_namespace_lock); |
9babb374 BB |
8535 | |
8536 | /* | |
8537 | * If the pool grew as a result of the config update, | |
8538 | * then log an internal history event. | |
8539 | */ | |
428870ff | 8540 | if (new_space != old_space) { |
6f1ffb06 | 8541 | spa_history_log_internal(spa, "vdev online", NULL, |
45d1cae3 | 8542 | "pool '%s' size: %llu(+%llu)", |
74756182 MM |
8543 | spa_name(spa), (u_longlong_t)new_space, |
8544 | (u_longlong_t)(new_space - old_space)); | |
9babb374 | 8545 | } |
34dc7c2f BB |
8546 | } |
8547 | ||
8548 | /* | |
8549 | * See if any devices need to be marked REMOVED. | |
34dc7c2f | 8550 | */ |
b128c09f | 8551 | if (tasks & SPA_ASYNC_REMOVE) { |
428870ff | 8552 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 8553 | spa_async_remove(spa, spa->spa_root_vdev); |
867959b5 | 8554 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) |
b128c09f | 8555 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); |
867959b5 | 8556 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
b128c09f BB |
8557 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); |
8558 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
34dc7c2f BB |
8559 | } |
8560 | ||
9babb374 BB |
8561 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { |
8562 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8563 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
8564 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8565 | } | |
8566 | ||
34dc7c2f | 8567 | /* |
b128c09f | 8568 | * See if any devices need to be probed. |
34dc7c2f | 8569 | */ |
b128c09f | 8570 | if (tasks & SPA_ASYNC_PROBE) { |
428870ff | 8571 | spa_vdev_state_enter(spa, SCL_NONE); |
b128c09f BB |
8572 | spa_async_probe(spa, spa->spa_root_vdev); |
8573 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
8574 | } | |
34dc7c2f BB |
8575 | |
8576 | /* | |
b128c09f | 8577 | * If any devices are done replacing, detach them. |
34dc7c2f | 8578 | */ |
b2255edc | 8579 | if (tasks & SPA_ASYNC_RESILVER_DONE || |
719534ca AH |
8580 | tasks & SPA_ASYNC_REBUILD_DONE || |
8581 | tasks & SPA_ASYNC_DETACH_SPARE) { | |
b128c09f | 8582 | spa_vdev_resilver_done(spa); |
9a49d3f3 BB |
8583 | } |
8584 | ||
34dc7c2f BB |
8585 | /* |
8586 | * Kick off a resilver. | |
8587 | */ | |
80a91e74 | 8588 | if (tasks & SPA_ASYNC_RESILVER && |
9a49d3f3 | 8589 | !vdev_rebuild_active(spa->spa_root_vdev) && |
80a91e74 TC |
8590 | (!dsl_scan_resilvering(dp) || |
8591 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER))) | |
3c819a2c | 8592 | dsl_scan_restart_resilver(dp, 0); |
34dc7c2f | 8593 | |
619f0976 GW |
8594 | if (tasks & SPA_ASYNC_INITIALIZE_RESTART) { |
8595 | mutex_enter(&spa_namespace_lock); | |
8596 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8597 | vdev_initialize_restart(spa->spa_root_vdev); | |
8598 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8599 | mutex_exit(&spa_namespace_lock); | |
8600 | } | |
8601 | ||
1b939560 BB |
8602 | if (tasks & SPA_ASYNC_TRIM_RESTART) { |
8603 | mutex_enter(&spa_namespace_lock); | |
8604 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8605 | vdev_trim_restart(spa->spa_root_vdev); | |
8606 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8607 | mutex_exit(&spa_namespace_lock); | |
8608 | } | |
8609 | ||
8610 | if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) { | |
8611 | mutex_enter(&spa_namespace_lock); | |
8612 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8613 | vdev_autotrim_restart(spa); | |
8614 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8615 | mutex_exit(&spa_namespace_lock); | |
8616 | } | |
8617 | ||
b7654bd7 GA |
8618 | /* |
8619 | * Kick off L2 cache whole device TRIM. | |
8620 | */ | |
8621 | if (tasks & SPA_ASYNC_L2CACHE_TRIM) { | |
8622 | mutex_enter(&spa_namespace_lock); | |
8623 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8624 | vdev_trim_l2arc(spa); | |
8625 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8626 | mutex_exit(&spa_namespace_lock); | |
8627 | } | |
8628 | ||
77f6826b GA |
8629 | /* |
8630 | * Kick off L2 cache rebuilding. | |
8631 | */ | |
8632 | if (tasks & SPA_ASYNC_L2CACHE_REBUILD) { | |
8633 | mutex_enter(&spa_namespace_lock); | |
8634 | spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER); | |
8635 | l2arc_spa_rebuild_start(spa); | |
8636 | spa_config_exit(spa, SCL_L2ARC, FTAG); | |
8637 | mutex_exit(&spa_namespace_lock); | |
8638 | } | |
8639 | ||
34dc7c2f BB |
8640 | /* |
8641 | * Let the world know that we're done. | |
8642 | */ | |
8643 | mutex_enter(&spa->spa_async_lock); | |
8644 | spa->spa_async_thread = NULL; | |
8645 | cv_broadcast(&spa->spa_async_cv); | |
8646 | mutex_exit(&spa->spa_async_lock); | |
8647 | thread_exit(); | |
8648 | } | |
8649 | ||
8650 | void | |
8651 | spa_async_suspend(spa_t *spa) | |
8652 | { | |
8653 | mutex_enter(&spa->spa_async_lock); | |
8654 | spa->spa_async_suspended++; | |
9d5b5245 | 8655 | while (spa->spa_async_thread != NULL) |
34dc7c2f BB |
8656 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); |
8657 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 MA |
8658 | |
8659 | spa_vdev_remove_suspend(spa); | |
9d5b5245 SD |
8660 | |
8661 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a SD |
8662 | if (condense_thread != NULL) |
8663 | zthr_cancel(condense_thread); | |
d2734cce | 8664 | |
5caeef02 DB |
8665 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; |
8666 | if (raidz_expand_thread != NULL) | |
8667 | zthr_cancel(raidz_expand_thread); | |
8668 | ||
d2734cce | 8669 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; |
61c3391a SD |
8670 | if (discard_thread != NULL) |
8671 | zthr_cancel(discard_thread); | |
37f03da8 SH |
8672 | |
8673 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
8674 | if (ll_delete_thread != NULL) | |
8675 | zthr_cancel(ll_delete_thread); | |
8676 | ||
8677 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
8678 | if (ll_condense_thread != NULL) | |
8679 | zthr_cancel(ll_condense_thread); | |
34dc7c2f BB |
8680 | } |
8681 | ||
8682 | void | |
8683 | spa_async_resume(spa_t *spa) | |
8684 | { | |
8685 | mutex_enter(&spa->spa_async_lock); | |
8686 | ASSERT(spa->spa_async_suspended != 0); | |
8687 | spa->spa_async_suspended--; | |
8688 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 | 8689 | spa_restart_removal(spa); |
9d5b5245 SD |
8690 | |
8691 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
61c3391a | 8692 | if (condense_thread != NULL) |
9d5b5245 | 8693 | zthr_resume(condense_thread); |
d2734cce | 8694 | |
5caeef02 DB |
8695 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; |
8696 | if (raidz_expand_thread != NULL) | |
8697 | zthr_resume(raidz_expand_thread); | |
8698 | ||
d2734cce | 8699 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; |
61c3391a | 8700 | if (discard_thread != NULL) |
d2734cce | 8701 | zthr_resume(discard_thread); |
37f03da8 SH |
8702 | |
8703 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
8704 | if (ll_delete_thread != NULL) | |
8705 | zthr_resume(ll_delete_thread); | |
8706 | ||
8707 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
8708 | if (ll_condense_thread != NULL) | |
8709 | zthr_resume(ll_condense_thread); | |
34dc7c2f BB |
8710 | } |
8711 | ||
e6cfd633 WA |
8712 | static boolean_t |
8713 | spa_async_tasks_pending(spa_t *spa) | |
8714 | { | |
8715 | uint_t non_config_tasks; | |
8716 | uint_t config_task; | |
8717 | boolean_t config_task_suspended; | |
8718 | ||
8719 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
8720 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
8721 | if (spa->spa_ccw_fail_time == 0) { | |
8722 | config_task_suspended = B_FALSE; | |
8723 | } else { | |
8724 | config_task_suspended = | |
8725 | (gethrtime() - spa->spa_ccw_fail_time) < | |
05852b34 | 8726 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); |
e6cfd633 WA |
8727 | } |
8728 | ||
8729 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
8730 | } | |
8731 | ||
34dc7c2f BB |
8732 | static void |
8733 | spa_async_dispatch(spa_t *spa) | |
8734 | { | |
8735 | mutex_enter(&spa->spa_async_lock); | |
e6cfd633 WA |
8736 | if (spa_async_tasks_pending(spa) && |
8737 | !spa->spa_async_suspended && | |
da92d5cb | 8738 | spa->spa_async_thread == NULL) |
34dc7c2f BB |
8739 | spa->spa_async_thread = thread_create(NULL, 0, |
8740 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
8741 | mutex_exit(&spa->spa_async_lock); | |
8742 | } | |
8743 | ||
8744 | void | |
8745 | spa_async_request(spa_t *spa, int task) | |
8746 | { | |
428870ff | 8747 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); |
34dc7c2f BB |
8748 | mutex_enter(&spa->spa_async_lock); |
8749 | spa->spa_async_tasks |= task; | |
8750 | mutex_exit(&spa->spa_async_lock); | |
8751 | } | |
8752 | ||
3c819a2c JP |
8753 | int |
8754 | spa_async_tasks(spa_t *spa) | |
8755 | { | |
8756 | return (spa->spa_async_tasks); | |
8757 | } | |
8758 | ||
34dc7c2f BB |
8759 | /* |
8760 | * ========================================================================== | |
8761 | * SPA syncing routines | |
8762 | * ========================================================================== | |
8763 | */ | |
8764 | ||
37f03da8 | 8765 | |
428870ff | 8766 | static int |
37f03da8 SH |
8767 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, |
8768 | dmu_tx_t *tx) | |
34dc7c2f | 8769 | { |
428870ff | 8770 | bpobj_t *bpo = arg; |
37f03da8 | 8771 | bpobj_enqueue(bpo, bp, bp_freed, tx); |
428870ff BB |
8772 | return (0); |
8773 | } | |
34dc7c2f | 8774 | |
37f03da8 SH |
8775 | int |
8776 | bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8777 | { | |
8778 | return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx)); | |
8779 | } | |
8780 | ||
8781 | int | |
8782 | bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8783 | { | |
8784 | return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx)); | |
8785 | } | |
8786 | ||
428870ff BB |
8787 | static int |
8788 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
8789 | { | |
9cdf7b1f | 8790 | zio_t *pio = arg; |
34dc7c2f | 8791 | |
9cdf7b1f MA |
8792 | zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp, |
8793 | pio->io_flags)); | |
428870ff | 8794 | return (0); |
34dc7c2f BB |
8795 | } |
8796 | ||
37f03da8 SH |
8797 | static int |
8798 | bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
8799 | dmu_tx_t *tx) | |
8800 | { | |
8801 | ASSERT(!bp_freed); | |
8802 | return (spa_free_sync_cb(arg, bp, tx)); | |
8803 | } | |
8804 | ||
e8b96c60 MA |
8805 | /* |
8806 | * Note: this simple function is not inlined to make it easier to dtrace the | |
8807 | * amount of time spent syncing frees. | |
8808 | */ | |
8809 | static void | |
8810 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
8811 | { | |
8812 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
8813 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
8814 | VERIFY(zio_wait(zio) == 0); | |
8815 | } | |
8816 | ||
8817 | /* | |
8818 | * Note: this simple function is not inlined to make it easier to dtrace the | |
8819 | * amount of time spent syncing deferred frees. | |
8820 | */ | |
8821 | static void | |
8822 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
8823 | { | |
8dc2197b SD |
8824 | if (spa_sync_pass(spa) != 1) |
8825 | return; | |
8826 | ||
93e28d66 SD |
8827 | /* |
8828 | * Note: | |
8829 | * If the log space map feature is active, we stop deferring | |
8830 | * frees to the next TXG and therefore running this function | |
8831 | * would be considered a no-op as spa_deferred_bpobj should | |
8832 | * not have any entries. | |
8833 | * | |
8834 | * That said we run this function anyway (instead of returning | |
8835 | * immediately) for the edge-case scenario where we just | |
8836 | * activated the log space map feature in this TXG but we have | |
8837 | * deferred frees from the previous TXG. | |
8838 | */ | |
e8b96c60 MA |
8839 | zio_t *zio = zio_root(spa, NULL, NULL, 0); |
8840 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
37f03da8 | 8841 | bpobj_spa_free_sync_cb, zio, tx), ==, 0); |
e8b96c60 MA |
8842 | VERIFY0(zio_wait(zio)); |
8843 | } | |
8844 | ||
34dc7c2f BB |
8845 | static void |
8846 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
8847 | { | |
8848 | char *packed = NULL; | |
b128c09f | 8849 | size_t bufsize; |
34dc7c2f BB |
8850 | size_t nvsize = 0; |
8851 | dmu_buf_t *db; | |
8852 | ||
8853 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
8854 | ||
b128c09f BB |
8855 | /* |
8856 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
b0bc7a84 | 8857 | * information. This avoids the dmu_buf_will_dirty() path and |
b128c09f BB |
8858 | * saves us a pre-read to get data we don't actually care about. |
8859 | */ | |
9ae529ec | 8860 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); |
79c76d5b | 8861 | packed = vmem_alloc(bufsize, KM_SLEEP); |
34dc7c2f BB |
8862 | |
8863 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
79c76d5b | 8864 | KM_SLEEP) == 0); |
861166b0 | 8865 | memset(packed + nvsize, 0, bufsize - nvsize); |
34dc7c2f | 8866 | |
b128c09f | 8867 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); |
34dc7c2f | 8868 | |
00b46022 | 8869 | vmem_free(packed, bufsize); |
34dc7c2f BB |
8870 | |
8871 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
8872 | dmu_buf_will_dirty(db, tx); | |
8873 | *(uint64_t *)db->db_data = nvsize; | |
8874 | dmu_buf_rele(db, FTAG); | |
8875 | } | |
8876 | ||
8877 | static void | |
8878 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
8879 | const char *config, const char *entry) | |
8880 | { | |
8881 | nvlist_t *nvroot; | |
8882 | nvlist_t **list; | |
8883 | int i; | |
8884 | ||
8885 | if (!sav->sav_sync) | |
8886 | return; | |
8887 | ||
8888 | /* | |
8889 | * Update the MOS nvlist describing the list of available devices. | |
8890 | * spa_validate_aux() will have already made sure this nvlist is | |
8891 | * valid and the vdevs are labeled appropriately. | |
8892 | */ | |
8893 | if (sav->sav_object == 0) { | |
8894 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
8895 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
8896 | sizeof (uint64_t), tx); | |
8897 | VERIFY(zap_update(spa->spa_meta_objset, | |
8898 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
8899 | &sav->sav_object, tx) == 0); | |
8900 | } | |
8901 | ||
65ad5d11 | 8902 | nvroot = fnvlist_alloc(); |
34dc7c2f | 8903 | if (sav->sav_count == 0) { |
795075e6 PD |
8904 | fnvlist_add_nvlist_array(nvroot, config, |
8905 | (const nvlist_t * const *)NULL, 0); | |
34dc7c2f | 8906 | } else { |
79c76d5b | 8907 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); |
34dc7c2f BB |
8908 | for (i = 0; i < sav->sav_count; i++) |
8909 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
428870ff | 8910 | B_FALSE, VDEV_CONFIG_L2CACHE); |
795075e6 PD |
8911 | fnvlist_add_nvlist_array(nvroot, config, |
8912 | (const nvlist_t * const *)list, sav->sav_count); | |
34dc7c2f BB |
8913 | for (i = 0; i < sav->sav_count; i++) |
8914 | nvlist_free(list[i]); | |
8915 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
8916 | } | |
8917 | ||
8918 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
8919 | nvlist_free(nvroot); | |
8920 | ||
8921 | sav->sav_sync = B_FALSE; | |
8922 | } | |
8923 | ||
e0ab3ab5 JS |
8924 | /* |
8925 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
8926 | * The all-vdev ZAP must be empty. | |
8927 | */ | |
8928 | static void | |
8929 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
8930 | { | |
8931 | spa_t *spa = vd->vdev_spa; | |
e0ab3ab5 | 8932 | |
3e4ed421 RW |
8933 | if (vd->vdev_root_zap != 0 && |
8934 | spa_feature_is_active(spa, SPA_FEATURE_AVZ_V2)) { | |
8935 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
8936 | vd->vdev_root_zap, tx)); | |
8937 | } | |
e0ab3ab5 JS |
8938 | if (vd->vdev_top_zap != 0) { |
8939 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
8940 | vd->vdev_top_zap, tx)); | |
8941 | } | |
8942 | if (vd->vdev_leaf_zap != 0) { | |
8943 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
8944 | vd->vdev_leaf_zap, tx)); | |
8945 | } | |
1c27024e | 8946 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
8947 | spa_avz_build(vd->vdev_child[i], avz, tx); |
8948 | } | |
8949 | } | |
8950 | ||
34dc7c2f BB |
8951 | static void |
8952 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
8953 | { | |
8954 | nvlist_t *config; | |
8955 | ||
e0ab3ab5 JS |
8956 | /* |
8957 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
8958 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
8959 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
8960 | * need to rebuild the AVZ although the config may not be dirty. | |
8961 | */ | |
8962 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
8963 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
34dc7c2f BB |
8964 | return; |
8965 | ||
b128c09f BB |
8966 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
8967 | ||
e0ab3ab5 | 8968 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || |
38640550 | 8969 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || |
e0ab3ab5 JS |
8970 | spa->spa_all_vdev_zaps != 0); |
8971 | ||
8972 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
e0ab3ab5 JS |
8973 | /* Make and build the new AVZ */ |
8974 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
8975 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
8976 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
8977 | ||
8978 | /* Diff old AVZ with new one */ | |
1c27024e DB |
8979 | zap_cursor_t zc; |
8980 | zap_attribute_t za; | |
8981 | ||
e0ab3ab5 JS |
8982 | for (zap_cursor_init(&zc, spa->spa_meta_objset, |
8983 | spa->spa_all_vdev_zaps); | |
8984 | zap_cursor_retrieve(&zc, &za) == 0; | |
8985 | zap_cursor_advance(&zc)) { | |
8986 | uint64_t vdzap = za.za_first_integer; | |
8987 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
8988 | vdzap) == ENOENT) { | |
8989 | /* | |
8990 | * ZAP is listed in old AVZ but not in new one; | |
8991 | * destroy it | |
8992 | */ | |
8993 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
8994 | tx)); | |
8995 | } | |
8996 | } | |
8997 | ||
8998 | zap_cursor_fini(&zc); | |
8999 | ||
9000 | /* Destroy the old AVZ */ | |
9001 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9002 | spa->spa_all_vdev_zaps, tx)); | |
9003 | ||
9004 | /* Replace the old AVZ in the dir obj with the new one */ | |
9005 | VERIFY0(zap_update(spa->spa_meta_objset, | |
9006 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
9007 | sizeof (new_avz), 1, &new_avz, tx)); | |
9008 | ||
9009 | spa->spa_all_vdev_zaps = new_avz; | |
9010 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
9011 | zap_cursor_t zc; | |
9012 | zap_attribute_t za; | |
9013 | ||
9014 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
9015 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
9016 | spa->spa_all_vdev_zaps); | |
9017 | zap_cursor_retrieve(&zc, &za) == 0; | |
9018 | zap_cursor_advance(&zc)) { | |
9019 | uint64_t zap = za.za_first_integer; | |
9020 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
9021 | } | |
9022 | ||
9023 | zap_cursor_fini(&zc); | |
9024 | ||
9025 | /* Destroy and unlink the AVZ itself */ | |
9026 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9027 | spa->spa_all_vdev_zaps, tx)); | |
9028 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
9029 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
9030 | spa->spa_all_vdev_zaps = 0; | |
9031 | } | |
9032 | ||
9033 | if (spa->spa_all_vdev_zaps == 0) { | |
9034 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
9035 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
9036 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
9037 | } | |
9038 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
9039 | ||
9040 | /* Create ZAPs for vdevs that don't have them. */ | |
9041 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
9042 | ||
b128c09f BB |
9043 | config = spa_config_generate(spa, spa->spa_root_vdev, |
9044 | dmu_tx_get_txg(tx), B_FALSE); | |
9045 | ||
ea0b2538 GW |
9046 | /* |
9047 | * If we're upgrading the spa version then make sure that | |
9048 | * the config object gets updated with the correct version. | |
9049 | */ | |
9050 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
9051 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
9052 | spa->spa_uberblock.ub_version); | |
9053 | ||
b128c09f | 9054 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f | 9055 | |
8a5fc748 | 9056 | nvlist_free(spa->spa_config_syncing); |
34dc7c2f BB |
9057 | spa->spa_config_syncing = config; |
9058 | ||
9059 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
9060 | } | |
9061 | ||
9ae529ec | 9062 | static void |
13fe0198 | 9063 | spa_sync_version(void *arg, dmu_tx_t *tx) |
9ae529ec | 9064 | { |
13fe0198 MA |
9065 | uint64_t *versionp = arg; |
9066 | uint64_t version = *versionp; | |
9067 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9ae529ec CS |
9068 | |
9069 | /* | |
9070 | * Setting the version is special cased when first creating the pool. | |
9071 | */ | |
9072 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
9073 | ||
8dca0a9a | 9074 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); |
9ae529ec CS |
9075 | ASSERT(version >= spa_version(spa)); |
9076 | ||
9077 | spa->spa_uberblock.ub_version = version; | |
9078 | vdev_config_dirty(spa->spa_root_vdev); | |
74756182 MM |
9079 | spa_history_log_internal(spa, "set", tx, "version=%lld", |
9080 | (longlong_t)version); | |
9ae529ec CS |
9081 | } |
9082 | ||
34dc7c2f BB |
9083 | /* |
9084 | * Set zpool properties. | |
9085 | */ | |
9086 | static void | |
13fe0198 | 9087 | spa_sync_props(void *arg, dmu_tx_t *tx) |
34dc7c2f | 9088 | { |
13fe0198 MA |
9089 | nvlist_t *nvp = arg; |
9090 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
34dc7c2f | 9091 | objset_t *mos = spa->spa_meta_objset; |
9ae529ec | 9092 | nvpair_t *elem = NULL; |
b128c09f BB |
9093 | |
9094 | mutex_enter(&spa->spa_props_lock); | |
34dc7c2f | 9095 | |
34dc7c2f | 9096 | while ((elem = nvlist_next_nvpair(nvp, elem))) { |
9ae529ec | 9097 | uint64_t intval; |
d1807f16 | 9098 | const char *strval, *fname; |
9ae529ec CS |
9099 | zpool_prop_t prop; |
9100 | const char *propname; | |
8eae2d21 | 9101 | const char *elemname = nvpair_name(elem); |
9ae529ec | 9102 | zprop_type_t proptype; |
fa86b5db | 9103 | spa_feature_t fid; |
9ae529ec | 9104 | |
8eae2d21 | 9105 | switch (prop = zpool_name_to_prop(elemname)) { |
34dc7c2f | 9106 | case ZPOOL_PROP_VERSION: |
93cf2076 | 9107 | intval = fnvpair_value_uint64(elem); |
34dc7c2f | 9108 | /* |
4e33ba4c | 9109 | * The version is synced separately before other |
9ae529ec | 9110 | * properties and should be correct by now. |
34dc7c2f | 9111 | */ |
9ae529ec | 9112 | ASSERT3U(spa_version(spa), >=, intval); |
34dc7c2f BB |
9113 | break; |
9114 | ||
9115 | case ZPOOL_PROP_ALTROOT: | |
9116 | /* | |
9117 | * 'altroot' is a non-persistent property. It should | |
9118 | * have been set temporarily at creation or import time. | |
9119 | */ | |
9120 | ASSERT(spa->spa_root != NULL); | |
9121 | break; | |
9122 | ||
572e2857 | 9123 | case ZPOOL_PROP_READONLY: |
34dc7c2f BB |
9124 | case ZPOOL_PROP_CACHEFILE: |
9125 | /* | |
e1cfd73f | 9126 | * 'readonly' and 'cachefile' are also non-persistent |
572e2857 | 9127 | * properties. |
34dc7c2f | 9128 | */ |
34dc7c2f | 9129 | break; |
d96eb2b1 | 9130 | case ZPOOL_PROP_COMMENT: |
93cf2076 | 9131 | strval = fnvpair_value_string(elem); |
d96eb2b1 DM |
9132 | if (spa->spa_comment != NULL) |
9133 | spa_strfree(spa->spa_comment); | |
9134 | spa->spa_comment = spa_strdup(strval); | |
9135 | /* | |
9136 | * We need to dirty the configuration on all the vdevs | |
88a48330 BB |
9137 | * so that their labels get updated. We also need to |
9138 | * update the cache file to keep it in sync with the | |
9139 | * MOS version. It's unnecessary to do this for pool | |
9140 | * creation since the vdev's configuration has already | |
9141 | * been dirtied. | |
d96eb2b1 | 9142 | */ |
88a48330 | 9143 | if (tx->tx_txg != TXG_INITIAL) { |
d96eb2b1 | 9144 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
9145 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
9146 | } | |
6f1ffb06 | 9147 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9148 | "%s=%s", elemname, strval); |
d96eb2b1 | 9149 | break; |
658fb802 CB |
9150 | case ZPOOL_PROP_COMPATIBILITY: |
9151 | strval = fnvpair_value_string(elem); | |
9152 | if (spa->spa_compatibility != NULL) | |
9153 | spa_strfree(spa->spa_compatibility); | |
9154 | spa->spa_compatibility = spa_strdup(strval); | |
9155 | /* | |
9156 | * Dirty the configuration on vdevs as above. | |
9157 | */ | |
88a48330 | 9158 | if (tx->tx_txg != TXG_INITIAL) { |
658fb802 | 9159 | vdev_config_dirty(spa->spa_root_vdev); |
88a48330 BB |
9160 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
9161 | } | |
9162 | ||
658fb802 CB |
9163 | spa_history_log_internal(spa, "set", tx, |
9164 | "%s=%s", nvpair_name(elem), strval); | |
9165 | break; | |
9166 | ||
8eae2d21 AJ |
9167 | case ZPOOL_PROP_INVAL: |
9168 | if (zpool_prop_feature(elemname)) { | |
9169 | fname = strchr(elemname, '@') + 1; | |
9170 | VERIFY0(zfeature_lookup_name(fname, &fid)); | |
9171 | ||
9172 | spa_feature_enable(spa, fid, tx); | |
9173 | spa_history_log_internal(spa, "set", tx, | |
9174 | "%s=enabled", elemname); | |
9175 | break; | |
9176 | } else if (!zfs_prop_user(elemname)) { | |
9177 | ASSERT(zpool_prop_feature(elemname)); | |
9178 | break; | |
9179 | } | |
9180 | zfs_fallthrough; | |
34dc7c2f BB |
9181 | default: |
9182 | /* | |
9183 | * Set pool property values in the poolprops mos object. | |
9184 | */ | |
34dc7c2f | 9185 | if (spa->spa_pool_props_object == 0) { |
9ae529ec CS |
9186 | spa->spa_pool_props_object = |
9187 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
34dc7c2f | 9188 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, |
9ae529ec | 9189 | tx); |
34dc7c2f | 9190 | } |
34dc7c2f BB |
9191 | |
9192 | /* normalize the property name */ | |
ee7b71db | 9193 | if (prop == ZPOOL_PROP_INVAL) { |
8eae2d21 AJ |
9194 | propname = elemname; |
9195 | proptype = PROP_TYPE_STRING; | |
ee7b71db RY |
9196 | } else { |
9197 | propname = zpool_prop_to_name(prop); | |
9198 | proptype = zpool_prop_get_type(prop); | |
8eae2d21 | 9199 | } |
34dc7c2f BB |
9200 | |
9201 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
9202 | ASSERT(proptype == PROP_TYPE_STRING); | |
93cf2076 GW |
9203 | strval = fnvpair_value_string(elem); |
9204 | VERIFY0(zap_update(mos, | |
34dc7c2f | 9205 | spa->spa_pool_props_object, propname, |
93cf2076 | 9206 | 1, strlen(strval) + 1, strval, tx)); |
6f1ffb06 | 9207 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9208 | "%s=%s", elemname, strval); |
34dc7c2f | 9209 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { |
93cf2076 | 9210 | intval = fnvpair_value_uint64(elem); |
34dc7c2f BB |
9211 | |
9212 | if (proptype == PROP_TYPE_INDEX) { | |
9213 | const char *unused; | |
93cf2076 GW |
9214 | VERIFY0(zpool_prop_index_to_string( |
9215 | prop, intval, &unused)); | |
34dc7c2f | 9216 | } |
93cf2076 | 9217 | VERIFY0(zap_update(mos, |
34dc7c2f | 9218 | spa->spa_pool_props_object, propname, |
93cf2076 | 9219 | 8, 1, &intval, tx)); |
6f1ffb06 | 9220 | spa_history_log_internal(spa, "set", tx, |
8eae2d21 | 9221 | "%s=%lld", elemname, |
74756182 | 9222 | (longlong_t)intval); |
34dc7c2f | 9223 | |
44f71818 RY |
9224 | switch (prop) { |
9225 | case ZPOOL_PROP_DELEGATION: | |
9226 | spa->spa_delegation = intval; | |
9227 | break; | |
9228 | case ZPOOL_PROP_BOOTFS: | |
9229 | spa->spa_bootfs = intval; | |
9230 | break; | |
9231 | case ZPOOL_PROP_FAILUREMODE: | |
9232 | spa->spa_failmode = intval; | |
9233 | break; | |
9234 | case ZPOOL_PROP_AUTOTRIM: | |
9235 | spa->spa_autotrim = intval; | |
428870ff | 9236 | spa_async_request(spa, |
44f71818 RY |
9237 | SPA_ASYNC_AUTOTRIM_RESTART); |
9238 | break; | |
9239 | case ZPOOL_PROP_AUTOEXPAND: | |
9240 | spa->spa_autoexpand = intval; | |
9241 | if (tx->tx_txg != TXG_INITIAL) | |
9242 | spa_async_request(spa, | |
9243 | SPA_ASYNC_AUTOEXPAND); | |
9244 | break; | |
9245 | case ZPOOL_PROP_MULTIHOST: | |
9246 | spa->spa_multihost = intval; | |
9247 | break; | |
9248 | default: | |
9249 | break; | |
9250 | } | |
9251 | } else { | |
9252 | ASSERT(0); /* not allowed */ | |
34dc7c2f BB |
9253 | } |
9254 | } | |
9255 | ||
34dc7c2f | 9256 | } |
b128c09f BB |
9257 | |
9258 | mutex_exit(&spa->spa_props_lock); | |
34dc7c2f BB |
9259 | } |
9260 | ||
428870ff BB |
9261 | /* |
9262 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
9263 | * reflect the new version this txg, so there must be no changes this | |
9264 | * txg to anything that the upgrade code depends on after it executes. | |
9265 | * Therefore this must be called after dsl_pool_sync() does the sync | |
9266 | * tasks. | |
9267 | */ | |
9268 | static void | |
9269 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
9270 | { | |
8dc2197b SD |
9271 | if (spa_sync_pass(spa) != 1) |
9272 | return; | |
428870ff | 9273 | |
8dc2197b | 9274 | dsl_pool_t *dp = spa->spa_dsl_pool; |
13fe0198 MA |
9275 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
9276 | ||
428870ff BB |
9277 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && |
9278 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
9279 | dsl_pool_create_origin(dp, tx); | |
9280 | ||
9281 | /* Keeping the origin open increases spa_minref */ | |
9282 | spa->spa_minref += 3; | |
9283 | } | |
9284 | ||
9285 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
9286 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
9287 | dsl_pool_upgrade_clones(dp, tx); | |
9288 | } | |
9289 | ||
9290 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
9291 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
9292 | dsl_pool_upgrade_dir_clones(dp, tx); | |
9293 | ||
9294 | /* Keeping the freedir open increases spa_minref */ | |
9295 | spa->spa_minref += 3; | |
9296 | } | |
9ae529ec CS |
9297 | |
9298 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
9299 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9300 | spa_feature_create_zap_objects(spa, tx); | |
9301 | } | |
62bdd5eb DL |
9302 | |
9303 | /* | |
9304 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
9305 | * when possibility to use lz4 compression for metadata was added | |
9306 | * Old pools that have this feature enabled must be upgraded to have | |
9307 | * this feature active | |
9308 | */ | |
9309 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9310 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
9311 | SPA_FEATURE_LZ4_COMPRESS); | |
9312 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
9313 | SPA_FEATURE_LZ4_COMPRESS); | |
9314 | ||
9315 | if (lz4_en && !lz4_ac) | |
9316 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
9317 | } | |
3c67d83a TH |
9318 | |
9319 | /* | |
9320 | * If we haven't written the salt, do so now. Note that the | |
9321 | * feature may not be activated yet, but that's fine since | |
9322 | * the presence of this ZAP entry is backwards compatible. | |
9323 | */ | |
9324 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
9325 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
9326 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9327 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
9328 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
9329 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
9330 | } | |
9331 | ||
13fe0198 | 9332 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
428870ff BB |
9333 | } |
9334 | ||
a1d477c2 MA |
9335 | static void |
9336 | vdev_indirect_state_sync_verify(vdev_t *vd) | |
9337 | { | |
2a8ba608 MM |
9338 | vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping; |
9339 | vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births; | |
a1d477c2 MA |
9340 | |
9341 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
9342 | ASSERT(vim != NULL); | |
9343 | ASSERT(vib != NULL); | |
9344 | } | |
9345 | ||
27f80e85 BB |
9346 | uint64_t obsolete_sm_object = 0; |
9347 | ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); | |
9348 | if (obsolete_sm_object != 0) { | |
a1d477c2 MA |
9349 | ASSERT(vd->vdev_obsolete_sm != NULL); |
9350 | ASSERT(vd->vdev_removing || | |
9351 | vd->vdev_ops == &vdev_indirect_ops); | |
9352 | ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); | |
9353 | ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); | |
27f80e85 | 9354 | ASSERT3U(obsolete_sm_object, ==, |
a1d477c2 MA |
9355 | space_map_object(vd->vdev_obsolete_sm)); |
9356 | ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, | |
9357 | space_map_allocated(vd->vdev_obsolete_sm)); | |
9358 | } | |
9359 | ASSERT(vd->vdev_obsolete_segments != NULL); | |
9360 | ||
9361 | /* | |
9362 | * Since frees / remaps to an indirect vdev can only | |
9363 | * happen in syncing context, the obsolete segments | |
9364 | * tree must be empty when we start syncing. | |
9365 | */ | |
9366 | ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); | |
9367 | } | |
9368 | ||
34dc7c2f | 9369 | /* |
8dc2197b SD |
9370 | * Set the top-level vdev's max queue depth. Evaluate each top-level's |
9371 | * async write queue depth in case it changed. The max queue depth will | |
9372 | * not change in the middle of syncing out this txg. | |
34dc7c2f | 9373 | */ |
8dc2197b SD |
9374 | static void |
9375 | spa_sync_adjust_vdev_max_queue_depth(spa_t *spa) | |
34dc7c2f | 9376 | { |
8dc2197b SD |
9377 | ASSERT(spa_writeable(spa)); |
9378 | ||
34dc7c2f | 9379 | vdev_t *rvd = spa->spa_root_vdev; |
3dfb57a3 DB |
9380 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * |
9381 | zfs_vdev_queue_depth_pct / 100; | |
8dc2197b SD |
9382 | metaslab_class_t *normal = spa_normal_class(spa); |
9383 | metaslab_class_t *special = spa_special_class(spa); | |
9384 | metaslab_class_t *dedup = spa_dedup_class(spa); | |
34dc7c2f | 9385 | |
492f64e9 | 9386 | uint64_t slots_per_allocator = 0; |
1c27024e | 9387 | for (int c = 0; c < rvd->vdev_children; c++) { |
3dfb57a3 | 9388 | vdev_t *tvd = rvd->vdev_child[c]; |
cc99f275 | 9389 | |
8dc2197b | 9390 | metaslab_group_t *mg = tvd->vdev_mg; |
cc99f275 DB |
9391 | if (mg == NULL || !metaslab_group_initialized(mg)) |
9392 | continue; | |
3dfb57a3 | 9393 | |
8dc2197b | 9394 | metaslab_class_t *mc = mg->mg_class; |
cc99f275 | 9395 | if (mc != normal && mc != special && mc != dedup) |
3dfb57a3 DB |
9396 | continue; |
9397 | ||
9398 | /* | |
9399 | * It is safe to do a lock-free check here because only async | |
9400 | * allocations look at mg_max_alloc_queue_depth, and async | |
9401 | * allocations all happen from spa_sync(). | |
9402 | */ | |
32d805c3 | 9403 | for (int i = 0; i < mg->mg_allocators; i++) { |
424fd7c3 | 9404 | ASSERT0(zfs_refcount_count( |
32d805c3 MA |
9405 | &(mg->mg_allocator[i].mga_alloc_queue_depth))); |
9406 | } | |
3dfb57a3 | 9407 | mg->mg_max_alloc_queue_depth = max_queue_depth; |
492f64e9 | 9408 | |
32d805c3 MA |
9409 | for (int i = 0; i < mg->mg_allocators; i++) { |
9410 | mg->mg_allocator[i].mga_cur_max_alloc_queue_depth = | |
492f64e9 PD |
9411 | zfs_vdev_def_queue_depth; |
9412 | } | |
9413 | slots_per_allocator += zfs_vdev_def_queue_depth; | |
3dfb57a3 | 9414 | } |
cc99f275 | 9415 | |
492f64e9 | 9416 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
f8020c93 AM |
9417 | ASSERT0(zfs_refcount_count(&normal->mc_allocator[i]. |
9418 | mca_alloc_slots)); | |
9419 | ASSERT0(zfs_refcount_count(&special->mc_allocator[i]. | |
9420 | mca_alloc_slots)); | |
9421 | ASSERT0(zfs_refcount_count(&dedup->mc_allocator[i]. | |
9422 | mca_alloc_slots)); | |
9423 | normal->mc_allocator[i].mca_alloc_max_slots = | |
9424 | slots_per_allocator; | |
9425 | special->mc_allocator[i].mca_alloc_max_slots = | |
9426 | slots_per_allocator; | |
9427 | dedup->mc_allocator[i].mca_alloc_max_slots = | |
9428 | slots_per_allocator; | |
cc99f275 DB |
9429 | } |
9430 | normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9431 | special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9432 | dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
8dc2197b SD |
9433 | } |
9434 | ||
9435 | static void | |
9436 | spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx) | |
9437 | { | |
9438 | ASSERT(spa_writeable(spa)); | |
3dfb57a3 | 9439 | |
8dc2197b | 9440 | vdev_t *rvd = spa->spa_root_vdev; |
a1d477c2 MA |
9441 | for (int c = 0; c < rvd->vdev_children; c++) { |
9442 | vdev_t *vd = rvd->vdev_child[c]; | |
9443 | vdev_indirect_state_sync_verify(vd); | |
9444 | ||
9445 | if (vdev_indirect_should_condense(vd)) { | |
9446 | spa_condense_indirect_start_sync(vd, tx); | |
9447 | break; | |
9448 | } | |
9449 | } | |
8dc2197b SD |
9450 | } |
9451 | ||
9452 | static void | |
9453 | spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx) | |
9454 | { | |
9455 | objset_t *mos = spa->spa_meta_objset; | |
9456 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9457 | uint64_t txg = tx->tx_txg; | |
9458 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
a1d477c2 | 9459 | |
34dc7c2f | 9460 | do { |
428870ff | 9461 | int pass = ++spa->spa_sync_pass; |
34dc7c2f BB |
9462 | |
9463 | spa_sync_config_object(spa, tx); | |
9464 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
9465 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
9466 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
9467 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
9468 | spa_errlog_sync(spa, txg); | |
9469 | dsl_pool_sync(dp, txg); | |
9470 | ||
93e28d66 SD |
9471 | if (pass < zfs_sync_pass_deferred_free || |
9472 | spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { | |
9473 | /* | |
9474 | * If the log space map feature is active we don't | |
9475 | * care about deferred frees and the deferred bpobj | |
9476 | * as the log space map should effectively have the | |
9477 | * same results (i.e. appending only to one object). | |
9478 | */ | |
e8b96c60 | 9479 | spa_sync_frees(spa, free_bpl, tx); |
428870ff | 9480 | } else { |
905edb40 MA |
9481 | /* |
9482 | * We can not defer frees in pass 1, because | |
9483 | * we sync the deferred frees later in pass 1. | |
9484 | */ | |
9485 | ASSERT3U(pass, >, 1); | |
37f03da8 | 9486 | bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb, |
e8b96c60 | 9487 | &spa->spa_deferred_bpobj, tx); |
34dc7c2f BB |
9488 | } |
9489 | ||
67a1b037 | 9490 | brt_sync(spa, txg); |
428870ff BB |
9491 | ddt_sync(spa, txg); |
9492 | dsl_scan_sync(dp, tx); | |
482eeef8 | 9493 | dsl_errorscrub_sync(dp, tx); |
8dc2197b SD |
9494 | svr_sync(spa, tx); |
9495 | spa_sync_upgrades(spa, tx); | |
34dc7c2f | 9496 | |
93e28d66 SD |
9497 | spa_flush_metaslabs(spa, tx); |
9498 | ||
8dc2197b | 9499 | vdev_t *vd = NULL; |
a1d477c2 MA |
9500 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) |
9501 | != NULL) | |
428870ff BB |
9502 | vdev_sync(vd, txg); |
9503 | ||
5caeef02 DB |
9504 | if (pass == 1) { |
9505 | /* | |
9506 | * dsl_pool_sync() -> dp_sync_tasks may have dirtied | |
9507 | * the config. If that happens, this txg should not | |
9508 | * be a no-op. So we must sync the config to the MOS | |
9509 | * before checking for no-op. | |
9510 | * | |
9511 | * Note that when the config is dirty, it will | |
9512 | * be written to the MOS (i.e. the MOS will be | |
9513 | * dirtied) every time we call spa_sync_config_object() | |
9514 | * in this txg. Therefore we can't call this after | |
9515 | * dsl_pool_sync() every pass, because it would | |
9516 | * prevent us from converging, since we'd dirty | |
9517 | * the MOS every pass. | |
9518 | * | |
9519 | * Sync tasks can only be processed in pass 1, so | |
9520 | * there's no need to do this in later passes. | |
9521 | */ | |
9522 | spa_sync_config_object(spa, tx); | |
9523 | } | |
9524 | ||
8dc2197b SD |
9525 | /* |
9526 | * Note: We need to check if the MOS is dirty because we could | |
9527 | * have marked the MOS dirty without updating the uberblock | |
9528 | * (e.g. if we have sync tasks but no dirty user data). We need | |
9529 | * to check the uberblock's rootbp because it is updated if we | |
9530 | * have synced out dirty data (though in this case the MOS will | |
9531 | * most likely also be dirty due to second order effects, we | |
9532 | * don't want to rely on that here). | |
9533 | */ | |
9534 | if (pass == 1 && | |
9535 | spa->spa_uberblock.ub_rootbp.blk_birth < txg && | |
9536 | !dmu_objset_is_dirty(mos, txg)) { | |
905edb40 | 9537 | /* |
8dc2197b SD |
9538 | * Nothing changed on the first pass, therefore this |
9539 | * TXG is a no-op. Avoid syncing deferred frees, so | |
9540 | * that we can keep this TXG as a no-op. | |
905edb40 | 9541 | */ |
8dc2197b SD |
9542 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); |
9543 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9544 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
9545 | ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg)); | |
9546 | break; | |
905edb40 | 9547 | } |
34dc7c2f | 9548 | |
8dc2197b | 9549 | spa_sync_deferred_frees(spa, tx); |
428870ff | 9550 | } while (dmu_objset_is_dirty(mos, txg)); |
8dc2197b | 9551 | } |
34dc7c2f | 9552 | |
8dc2197b SD |
9553 | /* |
9554 | * Rewrite the vdev configuration (which includes the uberblock) to | |
9555 | * commit the transaction group. | |
9556 | * | |
9557 | * If there are no dirty vdevs, we sync the uberblock to a few random | |
9558 | * top-level vdevs that are known to be visible in the config cache | |
9559 | * (see spa_vdev_add() for a complete description). If there *are* dirty | |
9560 | * vdevs, sync the uberblock to all vdevs. | |
9561 | */ | |
9562 | static void | |
9563 | spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx) | |
9564 | { | |
9565 | vdev_t *rvd = spa->spa_root_vdev; | |
9566 | uint64_t txg = tx->tx_txg; | |
a1d477c2 | 9567 | |
b128c09f | 9568 | for (;;) { |
8dc2197b SD |
9569 | int error = 0; |
9570 | ||
b128c09f BB |
9571 | /* |
9572 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
9573 | * while we're attempting to write the vdev labels. | |
9574 | */ | |
9575 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9576 | ||
9577 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
d2734cce | 9578 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; |
b128c09f BB |
9579 | int svdcount = 0; |
9580 | int children = rvd->vdev_children; | |
29274c9f | 9581 | int c0 = random_in_range(children); |
b128c09f | 9582 | |
1c27024e | 9583 | for (int c = 0; c < children; c++) { |
8dc2197b SD |
9584 | vdev_t *vd = |
9585 | rvd->vdev_child[(c0 + c) % children]; | |
d2734cce SD |
9586 | |
9587 | /* Stop when revisiting the first vdev */ | |
9588 | if (c > 0 && svd[0] == vd) | |
9589 | break; | |
9590 | ||
8dc2197b SD |
9591 | if (vd->vdev_ms_array == 0 || |
9592 | vd->vdev_islog || | |
a1d477c2 | 9593 | !vdev_is_concrete(vd)) |
b128c09f | 9594 | continue; |
d2734cce | 9595 | |
b128c09f | 9596 | svd[svdcount++] = vd; |
6cb8e530 | 9597 | if (svdcount == SPA_SYNC_MIN_VDEVS) |
b128c09f BB |
9598 | break; |
9599 | } | |
b6fcb792 | 9600 | error = vdev_config_sync(svd, svdcount, txg); |
b128c09f BB |
9601 | } else { |
9602 | error = vdev_config_sync(rvd->vdev_child, | |
b6fcb792 | 9603 | rvd->vdev_children, txg); |
34dc7c2f | 9604 | } |
34dc7c2f | 9605 | |
3bc7e0fb GW |
9606 | if (error == 0) |
9607 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
9608 | ||
b128c09f BB |
9609 | spa_config_exit(spa, SCL_STATE, FTAG); |
9610 | ||
9611 | if (error == 0) | |
9612 | break; | |
cec3a0a1 | 9613 | zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR); |
b128c09f BB |
9614 | zio_resume_wait(spa); |
9615 | } | |
8dc2197b SD |
9616 | } |
9617 | ||
9618 | /* | |
9619 | * Sync the specified transaction group. New blocks may be dirtied as | |
9620 | * part of the process, so we iterate until it converges. | |
9621 | */ | |
9622 | void | |
9623 | spa_sync(spa_t *spa, uint64_t txg) | |
9624 | { | |
9625 | vdev_t *vd = NULL; | |
9626 | ||
9627 | VERIFY(spa_writeable(spa)); | |
9628 | ||
9629 | /* | |
9630 | * Wait for i/os issued in open context that need to complete | |
9631 | * before this txg syncs. | |
9632 | */ | |
9633 | (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]); | |
9634 | spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, | |
9635 | ZIO_FLAG_CANFAIL); | |
9636 | ||
67a1b037 PJD |
9637 | /* |
9638 | * Now that there can be no more cloning in this transaction group, | |
9639 | * but we are still before issuing frees, we can process pending BRT | |
9640 | * updates. | |
9641 | */ | |
9642 | brt_pending_apply(spa, txg); | |
9643 | ||
8dc2197b SD |
9644 | /* |
9645 | * Lock out configuration changes. | |
9646 | */ | |
9647 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
9648 | ||
9649 | spa->spa_syncing_txg = txg; | |
9650 | spa->spa_sync_pass = 0; | |
9651 | ||
9652 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
1b50749c AM |
9653 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
9654 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
9655 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
8dc2197b SD |
9656 | } |
9657 | ||
9658 | /* | |
9659 | * If there are any pending vdev state changes, convert them | |
9660 | * into config changes that go out with this transaction group. | |
9661 | */ | |
9662 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9f08b6e3 RY |
9663 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { |
9664 | /* Avoid holding the write lock unless actually necessary */ | |
9665 | if (vd->vdev_aux == NULL) { | |
9666 | vdev_state_clean(vd); | |
9667 | vdev_config_dirty(vd); | |
9668 | continue; | |
9669 | } | |
8dc2197b SD |
9670 | /* |
9671 | * We need the write lock here because, for aux vdevs, | |
9672 | * calling vdev_config_dirty() modifies sav_config. | |
9673 | * This is ugly and will become unnecessary when we | |
9674 | * eliminate the aux vdev wart by integrating all vdevs | |
9675 | * into the root vdev tree. | |
9676 | */ | |
9677 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9678 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
9679 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
9680 | vdev_state_clean(vd); | |
9681 | vdev_config_dirty(vd); | |
9682 | } | |
9683 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
9684 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
9685 | } | |
9686 | spa_config_exit(spa, SCL_STATE, FTAG); | |
9687 | ||
9688 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9689 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
9690 | ||
9691 | spa->spa_sync_starttime = gethrtime(); | |
9692 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
9693 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
9694 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
9695 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
9696 | ||
9697 | /* | |
9698 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
9699 | * set spa_deflate if we have no raid-z vdevs. | |
9700 | */ | |
9701 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
9702 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
9703 | vdev_t *rvd = spa->spa_root_vdev; | |
9704 | ||
9705 | int i; | |
9706 | for (i = 0; i < rvd->vdev_children; i++) { | |
9707 | vd = rvd->vdev_child[i]; | |
9708 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
9709 | break; | |
9710 | } | |
9711 | if (i == rvd->vdev_children) { | |
9712 | spa->spa_deflate = TRUE; | |
9713 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9714 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
9715 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
9716 | } | |
9717 | } | |
9718 | ||
9719 | spa_sync_adjust_vdev_max_queue_depth(spa); | |
9720 | ||
9721 | spa_sync_condense_indirect(spa, tx); | |
9722 | ||
9723 | spa_sync_iterate_to_convergence(spa, tx); | |
9724 | ||
9725 | #ifdef ZFS_DEBUG | |
9726 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
9727 | /* | |
9728 | * Make sure that the number of ZAPs for all the vdevs matches | |
9729 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
9730 | * called if the config is dirty; otherwise there may be | |
9731 | * outstanding AVZ operations that weren't completed in | |
9732 | * spa_sync_config_object. | |
9733 | */ | |
9734 | uint64_t all_vdev_zap_entry_count; | |
9735 | ASSERT0(zap_count(spa->spa_meta_objset, | |
9736 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
9737 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
9738 | all_vdev_zap_entry_count); | |
9739 | } | |
9740 | #endif | |
9741 | ||
9742 | if (spa->spa_vdev_removal != NULL) { | |
9743 | ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); | |
9744 | } | |
9745 | ||
9746 | spa_sync_rewrite_vdev_config(spa, tx); | |
34dc7c2f BB |
9747 | dmu_tx_commit(tx); |
9748 | ||
57ddcda1 | 9749 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 GW |
9750 | spa->spa_deadman_tqid = 0; |
9751 | ||
34dc7c2f BB |
9752 | /* |
9753 | * Clear the dirty config list. | |
9754 | */ | |
b128c09f | 9755 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) |
34dc7c2f BB |
9756 | vdev_config_clean(vd); |
9757 | ||
9758 | /* | |
9759 | * Now that the new config has synced transactionally, | |
9760 | * let it become visible to the config cache. | |
9761 | */ | |
9762 | if (spa->spa_config_syncing != NULL) { | |
9763 | spa_config_set(spa, spa->spa_config_syncing); | |
9764 | spa->spa_config_txg = txg; | |
9765 | spa->spa_config_syncing = NULL; | |
9766 | } | |
9767 | ||
428870ff | 9768 | dsl_pool_sync_done(dp, txg); |
34dc7c2f | 9769 | |
492f64e9 | 9770 | for (int i = 0; i < spa->spa_alloc_count; i++) { |
1b50749c AM |
9771 | mutex_enter(&spa->spa_allocs[i].spaa_lock); |
9772 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
9773 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
492f64e9 | 9774 | } |
3dfb57a3 | 9775 | |
34dc7c2f BB |
9776 | /* |
9777 | * Update usable space statistics. | |
9778 | */ | |
619f0976 GW |
9779 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) |
9780 | != NULL) | |
34dc7c2f | 9781 | vdev_sync_done(vd, txg); |
f09fda50 PD |
9782 | |
9783 | metaslab_class_evict_old(spa->spa_normal_class, txg); | |
9784 | metaslab_class_evict_old(spa->spa_log_class, txg); | |
9785 | ||
93e28d66 | 9786 | spa_sync_close_syncing_log_sm(spa); |
34dc7c2f | 9787 | |
428870ff BB |
9788 | spa_update_dspace(spa); |
9789 | ||
65d10bd8 KJ |
9790 | if (spa_get_autotrim(spa) == SPA_AUTOTRIM_ON) |
9791 | vdev_autotrim_kick(spa); | |
9792 | ||
34dc7c2f BB |
9793 | /* |
9794 | * It had better be the case that we didn't dirty anything | |
9795 | * since vdev_config_sync(). | |
9796 | */ | |
9797 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
9798 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9799 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
428870ff | 9800 | |
d2734cce SD |
9801 | while (zfs_pause_spa_sync) |
9802 | delay(1); | |
9803 | ||
428870ff | 9804 | spa->spa_sync_pass = 0; |
34dc7c2f | 9805 | |
55922e73 GW |
9806 | /* |
9807 | * Update the last synced uberblock here. We want to do this at | |
9808 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
9809 | * will be guaranteed that all the processing associated with | |
9810 | * that txg has been completed. | |
9811 | */ | |
9812 | spa->spa_ubsync = spa->spa_uberblock; | |
b128c09f | 9813 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f | 9814 | |
428870ff BB |
9815 | spa_handle_ignored_writes(spa); |
9816 | ||
34dc7c2f BB |
9817 | /* |
9818 | * If any async tasks have been requested, kick them off. | |
9819 | */ | |
9820 | spa_async_dispatch(spa); | |
9821 | } | |
9822 | ||
9823 | /* | |
9824 | * Sync all pools. We don't want to hold the namespace lock across these | |
9825 | * operations, so we take a reference on the spa_t and drop the lock during the | |
9826 | * sync. | |
9827 | */ | |
9828 | void | |
9829 | spa_sync_allpools(void) | |
9830 | { | |
9831 | spa_t *spa = NULL; | |
9832 | mutex_enter(&spa_namespace_lock); | |
9833 | while ((spa = spa_next(spa)) != NULL) { | |
572e2857 BB |
9834 | if (spa_state(spa) != POOL_STATE_ACTIVE || |
9835 | !spa_writeable(spa) || spa_suspended(spa)) | |
34dc7c2f BB |
9836 | continue; |
9837 | spa_open_ref(spa, FTAG); | |
9838 | mutex_exit(&spa_namespace_lock); | |
9839 | txg_wait_synced(spa_get_dsl(spa), 0); | |
9840 | mutex_enter(&spa_namespace_lock); | |
9841 | spa_close(spa, FTAG); | |
9842 | } | |
9843 | mutex_exit(&spa_namespace_lock); | |
9844 | } | |
9845 | ||
3bd4df38 EN |
9846 | taskq_t * |
9847 | spa_sync_tq_create(spa_t *spa, const char *name) | |
9848 | { | |
9849 | kthread_t **kthreads; | |
9850 | ||
9851 | ASSERT(spa->spa_sync_tq == NULL); | |
9852 | ASSERT3S(spa->spa_alloc_count, <=, boot_ncpus); | |
9853 | ||
9854 | /* | |
9855 | * - do not allow more allocators than cpus. | |
9856 | * - there may be more cpus than allocators. | |
9857 | * - do not allow more sync taskq threads than allocators or cpus. | |
9858 | */ | |
9859 | int nthreads = spa->spa_alloc_count; | |
9860 | spa->spa_syncthreads = kmem_zalloc(sizeof (spa_syncthread_info_t) * | |
9861 | nthreads, KM_SLEEP); | |
9862 | ||
9863 | spa->spa_sync_tq = taskq_create_synced(name, nthreads, minclsyspri, | |
9864 | nthreads, INT_MAX, TASKQ_PREPOPULATE, &kthreads); | |
9865 | VERIFY(spa->spa_sync_tq != NULL); | |
9866 | VERIFY(kthreads != NULL); | |
9867 | ||
9868 | spa_taskqs_t *tqs = | |
9869 | &spa->spa_zio_taskq[ZIO_TYPE_WRITE][ZIO_TASKQ_ISSUE]; | |
9870 | ||
9871 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
9872 | for (int i = 0, w = 0; i < nthreads; i++, w++, ti++) { | |
9873 | ti->sti_thread = kthreads[i]; | |
9874 | if (w == tqs->stqs_count) { | |
9875 | w = 0; | |
9876 | } | |
9877 | ti->sti_wr_iss_tq = tqs->stqs_taskq[w]; | |
9878 | } | |
9879 | ||
9880 | kmem_free(kthreads, sizeof (*kthreads) * nthreads); | |
9881 | return (spa->spa_sync_tq); | |
9882 | } | |
9883 | ||
9884 | void | |
9885 | spa_sync_tq_destroy(spa_t *spa) | |
9886 | { | |
9887 | ASSERT(spa->spa_sync_tq != NULL); | |
9888 | ||
9889 | taskq_wait(spa->spa_sync_tq); | |
9890 | taskq_destroy(spa->spa_sync_tq); | |
9891 | kmem_free(spa->spa_syncthreads, | |
9892 | sizeof (spa_syncthread_info_t) * spa->spa_alloc_count); | |
9893 | spa->spa_sync_tq = NULL; | |
9894 | } | |
9895 | ||
9896 | void | |
9897 | spa_select_allocator(zio_t *zio) | |
9898 | { | |
9899 | zbookmark_phys_t *bm = &zio->io_bookmark; | |
9900 | spa_t *spa = zio->io_spa; | |
9901 | ||
9902 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
9903 | ||
9904 | /* | |
9905 | * A gang block (for example) may have inherited its parent's | |
9906 | * allocator, in which case there is nothing further to do here. | |
9907 | */ | |
9908 | if (ZIO_HAS_ALLOCATOR(zio)) | |
9909 | return; | |
9910 | ||
9911 | ASSERT(spa != NULL); | |
9912 | ASSERT(bm != NULL); | |
9913 | ||
9914 | /* | |
9915 | * First try to use an allocator assigned to the syncthread, and set | |
9916 | * the corresponding write issue taskq for the allocator. | |
9917 | * Note, we must have an open pool to do this. | |
9918 | */ | |
9919 | if (spa->spa_sync_tq != NULL) { | |
9920 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
9921 | for (int i = 0; i < spa->spa_alloc_count; i++, ti++) { | |
9922 | if (ti->sti_thread == curthread) { | |
9923 | zio->io_allocator = i; | |
9924 | zio->io_wr_iss_tq = ti->sti_wr_iss_tq; | |
9925 | return; | |
9926 | } | |
9927 | } | |
9928 | } | |
9929 | ||
9930 | /* | |
9931 | * We want to try to use as many allocators as possible to help improve | |
9932 | * performance, but we also want logically adjacent IOs to be physically | |
9933 | * adjacent to improve sequential read performance. We chunk each object | |
9934 | * into 2^20 block regions, and then hash based on the objset, object, | |
9935 | * level, and region to accomplish both of these goals. | |
9936 | */ | |
9937 | uint64_t hv = cityhash4(bm->zb_objset, bm->zb_object, bm->zb_level, | |
9938 | bm->zb_blkid >> 20); | |
9939 | ||
9940 | zio->io_allocator = (uint_t)hv % spa->spa_alloc_count; | |
9941 | zio->io_wr_iss_tq = NULL; | |
9942 | } | |
9943 | ||
34dc7c2f BB |
9944 | /* |
9945 | * ========================================================================== | |
9946 | * Miscellaneous routines | |
9947 | * ========================================================================== | |
9948 | */ | |
9949 | ||
9950 | /* | |
9951 | * Remove all pools in the system. | |
9952 | */ | |
9953 | void | |
9954 | spa_evict_all(void) | |
9955 | { | |
9956 | spa_t *spa; | |
9957 | ||
9958 | /* | |
9959 | * Remove all cached state. All pools should be closed now, | |
9960 | * so every spa in the AVL tree should be unreferenced. | |
9961 | */ | |
9962 | mutex_enter(&spa_namespace_lock); | |
9963 | while ((spa = spa_next(NULL)) != NULL) { | |
9964 | /* | |
9965 | * Stop async tasks. The async thread may need to detach | |
9966 | * a device that's been replaced, which requires grabbing | |
9967 | * spa_namespace_lock, so we must drop it here. | |
9968 | */ | |
9969 | spa_open_ref(spa, FTAG); | |
9970 | mutex_exit(&spa_namespace_lock); | |
9971 | spa_async_suspend(spa); | |
9972 | mutex_enter(&spa_namespace_lock); | |
34dc7c2f BB |
9973 | spa_close(spa, FTAG); |
9974 | ||
9975 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
9976 | spa_unload(spa); | |
9977 | spa_deactivate(spa); | |
9978 | } | |
9979 | spa_remove(spa); | |
9980 | } | |
9981 | mutex_exit(&spa_namespace_lock); | |
9982 | } | |
9983 | ||
9984 | vdev_t * | |
9babb374 | 9985 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) |
34dc7c2f | 9986 | { |
b128c09f BB |
9987 | vdev_t *vd; |
9988 | int i; | |
9989 | ||
9990 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
9991 | return (vd); | |
9992 | ||
9babb374 | 9993 | if (aux) { |
b128c09f BB |
9994 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { |
9995 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
9babb374 BB |
9996 | if (vd->vdev_guid == guid) |
9997 | return (vd); | |
9998 | } | |
9999 | ||
10000 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
10001 | vd = spa->spa_spares.sav_vdevs[i]; | |
b128c09f BB |
10002 | if (vd->vdev_guid == guid) |
10003 | return (vd); | |
10004 | } | |
10005 | } | |
10006 | ||
10007 | return (NULL); | |
34dc7c2f BB |
10008 | } |
10009 | ||
10010 | void | |
10011 | spa_upgrade(spa_t *spa, uint64_t version) | |
10012 | { | |
572e2857 BB |
10013 | ASSERT(spa_writeable(spa)); |
10014 | ||
b128c09f | 10015 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
10016 | |
10017 | /* | |
10018 | * This should only be called for a non-faulted pool, and since a | |
10019 | * future version would result in an unopenable pool, this shouldn't be | |
10020 | * possible. | |
10021 | */ | |
8dca0a9a | 10022 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); |
9b67f605 | 10023 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); |
34dc7c2f BB |
10024 | |
10025 | spa->spa_uberblock.ub_version = version; | |
10026 | vdev_config_dirty(spa->spa_root_vdev); | |
10027 | ||
b128c09f | 10028 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
10029 | |
10030 | txg_wait_synced(spa_get_dsl(spa), 0); | |
10031 | } | |
10032 | ||
49d42425 FU |
10033 | static boolean_t |
10034 | spa_has_aux_vdev(spa_t *spa, uint64_t guid, spa_aux_vdev_t *sav) | |
34dc7c2f | 10035 | { |
14e4e3cb | 10036 | (void) spa; |
34dc7c2f | 10037 | int i; |
49d42425 | 10038 | uint64_t vdev_guid; |
34dc7c2f BB |
10039 | |
10040 | for (i = 0; i < sav->sav_count; i++) | |
10041 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
10042 | return (B_TRUE); | |
10043 | ||
10044 | for (i = 0; i < sav->sav_npending; i++) { | |
10045 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
49d42425 | 10046 | &vdev_guid) == 0 && vdev_guid == guid) |
34dc7c2f BB |
10047 | return (B_TRUE); |
10048 | } | |
10049 | ||
10050 | return (B_FALSE); | |
10051 | } | |
10052 | ||
49d42425 FU |
10053 | boolean_t |
10054 | spa_has_l2cache(spa_t *spa, uint64_t guid) | |
10055 | { | |
10056 | return (spa_has_aux_vdev(spa, guid, &spa->spa_l2cache)); | |
10057 | } | |
10058 | ||
10059 | boolean_t | |
10060 | spa_has_spare(spa_t *spa, uint64_t guid) | |
10061 | { | |
10062 | return (spa_has_aux_vdev(spa, guid, &spa->spa_spares)); | |
10063 | } | |
10064 | ||
b128c09f BB |
10065 | /* |
10066 | * Check if a pool has an active shared spare device. | |
10067 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
10068 | */ | |
10069 | static boolean_t | |
10070 | spa_has_active_shared_spare(spa_t *spa) | |
10071 | { | |
10072 | int i, refcnt; | |
10073 | uint64_t pool; | |
10074 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
10075 | ||
10076 | for (i = 0; i < sav->sav_count; i++) { | |
10077 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
10078 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
10079 | refcnt > 2) | |
10080 | return (B_TRUE); | |
10081 | } | |
10082 | ||
10083 | return (B_FALSE); | |
10084 | } | |
10085 | ||
93e28d66 SD |
10086 | uint64_t |
10087 | spa_total_metaslabs(spa_t *spa) | |
10088 | { | |
10089 | vdev_t *rvd = spa->spa_root_vdev; | |
10090 | ||
10091 | uint64_t m = 0; | |
10092 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
10093 | vdev_t *vd = rvd->vdev_child[c]; | |
10094 | if (!vdev_is_concrete(vd)) | |
10095 | continue; | |
10096 | m += vd->vdev_ms_count; | |
10097 | } | |
10098 | return (m); | |
10099 | } | |
10100 | ||
e60e158e JG |
10101 | /* |
10102 | * Notify any waiting threads that some activity has switched from being in- | |
10103 | * progress to not-in-progress so that the thread can wake up and determine | |
10104 | * whether it is finished waiting. | |
10105 | */ | |
10106 | void | |
10107 | spa_notify_waiters(spa_t *spa) | |
10108 | { | |
10109 | /* | |
10110 | * Acquiring spa_activities_lock here prevents the cv_broadcast from | |
10111 | * happening between the waiting thread's check and cv_wait. | |
10112 | */ | |
10113 | mutex_enter(&spa->spa_activities_lock); | |
10114 | cv_broadcast(&spa->spa_activities_cv); | |
10115 | mutex_exit(&spa->spa_activities_lock); | |
10116 | } | |
10117 | ||
10118 | /* | |
10119 | * Notify any waiting threads that the pool is exporting, and then block until | |
10120 | * they are finished using the spa_t. | |
10121 | */ | |
10122 | void | |
10123 | spa_wake_waiters(spa_t *spa) | |
10124 | { | |
10125 | mutex_enter(&spa->spa_activities_lock); | |
10126 | spa->spa_waiters_cancel = B_TRUE; | |
10127 | cv_broadcast(&spa->spa_activities_cv); | |
10128 | while (spa->spa_waiters != 0) | |
10129 | cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock); | |
10130 | spa->spa_waiters_cancel = B_FALSE; | |
10131 | mutex_exit(&spa->spa_activities_lock); | |
10132 | } | |
10133 | ||
2288d419 | 10134 | /* Whether the vdev or any of its descendants are being initialized/trimmed. */ |
e60e158e | 10135 | static boolean_t |
2288d419 | 10136 | spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity) |
e60e158e JG |
10137 | { |
10138 | spa_t *spa = vd->vdev_spa; | |
e60e158e JG |
10139 | |
10140 | ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER)); | |
10141 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
2288d419 BB |
10142 | ASSERT(activity == ZPOOL_WAIT_INITIALIZE || |
10143 | activity == ZPOOL_WAIT_TRIM); | |
10144 | ||
10145 | kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ? | |
10146 | &vd->vdev_initialize_lock : &vd->vdev_trim_lock; | |
e60e158e JG |
10147 | |
10148 | mutex_exit(&spa->spa_activities_lock); | |
2288d419 | 10149 | mutex_enter(lock); |
e60e158e JG |
10150 | mutex_enter(&spa->spa_activities_lock); |
10151 | ||
2288d419 BB |
10152 | boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ? |
10153 | (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) : | |
10154 | (vd->vdev_trim_state == VDEV_TRIM_ACTIVE); | |
10155 | mutex_exit(lock); | |
e60e158e | 10156 | |
2288d419 | 10157 | if (in_progress) |
e60e158e JG |
10158 | return (B_TRUE); |
10159 | ||
10160 | for (int i = 0; i < vd->vdev_children; i++) { | |
2288d419 BB |
10161 | if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i], |
10162 | activity)) | |
e60e158e JG |
10163 | return (B_TRUE); |
10164 | } | |
10165 | ||
10166 | return (B_FALSE); | |
10167 | } | |
10168 | ||
10169 | /* | |
10170 | * If use_guid is true, this checks whether the vdev specified by guid is | |
2288d419 BB |
10171 | * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool |
10172 | * is being initialized/trimmed. The caller must hold the config lock and | |
10173 | * spa_activities_lock. | |
e60e158e JG |
10174 | */ |
10175 | static int | |
2288d419 BB |
10176 | spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid, |
10177 | zpool_wait_activity_t activity, boolean_t *in_progress) | |
e60e158e JG |
10178 | { |
10179 | mutex_exit(&spa->spa_activities_lock); | |
10180 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10181 | mutex_enter(&spa->spa_activities_lock); | |
10182 | ||
10183 | vdev_t *vd; | |
10184 | if (use_guid) { | |
10185 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
10186 | if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) { | |
10187 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10188 | return (EINVAL); | |
10189 | } | |
10190 | } else { | |
10191 | vd = spa->spa_root_vdev; | |
10192 | } | |
10193 | ||
2288d419 | 10194 | *in_progress = spa_vdev_activity_in_progress_impl(vd, activity); |
e60e158e JG |
10195 | |
10196 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10197 | return (0); | |
10198 | } | |
10199 | ||
10200 | /* | |
10201 | * Locking for waiting threads | |
10202 | * --------------------------- | |
10203 | * | |
10204 | * Waiting threads need a way to check whether a given activity is in progress, | |
10205 | * and then, if it is, wait for it to complete. Each activity will have some | |
10206 | * in-memory representation of the relevant on-disk state which can be used to | |
10207 | * determine whether or not the activity is in progress. The in-memory state and | |
10208 | * the locking used to protect it will be different for each activity, and may | |
10209 | * not be suitable for use with a cvar (e.g., some state is protected by the | |
10210 | * config lock). To allow waiting threads to wait without any races, another | |
10211 | * lock, spa_activities_lock, is used. | |
10212 | * | |
10213 | * When the state is checked, both the activity-specific lock (if there is one) | |
10214 | * and spa_activities_lock are held. In some cases, the activity-specific lock | |
10215 | * is acquired explicitly (e.g. the config lock). In others, the locking is | |
10216 | * internal to some check (e.g. bpobj_is_empty). After checking, the waiting | |
10217 | * thread releases the activity-specific lock and, if the activity is in | |
10218 | * progress, then cv_waits using spa_activities_lock. | |
10219 | * | |
10220 | * The waiting thread is woken when another thread, one completing some | |
10221 | * activity, updates the state of the activity and then calls | |
10222 | * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only | |
10223 | * needs to hold its activity-specific lock when updating the state, and this | |
10224 | * lock can (but doesn't have to) be dropped before calling spa_notify_waiters. | |
10225 | * | |
10226 | * Because spa_notify_waiters acquires spa_activities_lock before broadcasting, | |
10227 | * and because it is held when the waiting thread checks the state of the | |
10228 | * activity, it can never be the case that the completing thread both updates | |
10229 | * the activity state and cv_broadcasts in between the waiting thread's check | |
10230 | * and cv_wait. Thus, a waiting thread can never miss a wakeup. | |
10231 | * | |
10232 | * In order to prevent deadlock, when the waiting thread does its check, in some | |
10233 | * cases it will temporarily drop spa_activities_lock in order to acquire the | |
10234 | * activity-specific lock. The order in which spa_activities_lock and the | |
10235 | * activity specific lock are acquired in the waiting thread is determined by | |
10236 | * the order in which they are acquired in the completing thread; if the | |
10237 | * completing thread calls spa_notify_waiters with the activity-specific lock | |
10238 | * held, then the waiting thread must also acquire the activity-specific lock | |
10239 | * first. | |
10240 | */ | |
10241 | ||
10242 | static int | |
10243 | spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity, | |
10244 | boolean_t use_tag, uint64_t tag, boolean_t *in_progress) | |
10245 | { | |
10246 | int error = 0; | |
10247 | ||
10248 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
10249 | ||
10250 | switch (activity) { | |
10251 | case ZPOOL_WAIT_CKPT_DISCARD: | |
10252 | *in_progress = | |
10253 | (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) && | |
10254 | zap_contains(spa_meta_objset(spa), | |
10255 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) == | |
10256 | ENOENT); | |
10257 | break; | |
10258 | case ZPOOL_WAIT_FREE: | |
10259 | *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS && | |
10260 | !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) || | |
10261 | spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) || | |
10262 | spa_livelist_delete_check(spa)); | |
10263 | break; | |
10264 | case ZPOOL_WAIT_INITIALIZE: | |
2288d419 BB |
10265 | case ZPOOL_WAIT_TRIM: |
10266 | error = spa_vdev_activity_in_progress(spa, use_tag, tag, | |
10267 | activity, in_progress); | |
e60e158e JG |
10268 | break; |
10269 | case ZPOOL_WAIT_REPLACE: | |
10270 | mutex_exit(&spa->spa_activities_lock); | |
10271 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10272 | mutex_enter(&spa->spa_activities_lock); | |
10273 | ||
10274 | *in_progress = vdev_replace_in_progress(spa->spa_root_vdev); | |
10275 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10276 | break; | |
10277 | case ZPOOL_WAIT_REMOVE: | |
10278 | *in_progress = (spa->spa_removing_phys.sr_state == | |
10279 | DSS_SCANNING); | |
10280 | break; | |
10281 | case ZPOOL_WAIT_RESILVER: | |
5caeef02 DB |
10282 | *in_progress = vdev_rebuild_active(spa->spa_root_vdev); |
10283 | if (*in_progress) | |
9a49d3f3 | 10284 | break; |
9a70e97f | 10285 | zfs_fallthrough; |
e60e158e JG |
10286 | case ZPOOL_WAIT_SCRUB: |
10287 | { | |
10288 | boolean_t scanning, paused, is_scrub; | |
10289 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
10290 | ||
10291 | is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB); | |
10292 | scanning = (scn->scn_phys.scn_state == DSS_SCANNING); | |
10293 | paused = dsl_scan_is_paused_scrub(scn); | |
10294 | *in_progress = (scanning && !paused && | |
10295 | is_scrub == (activity == ZPOOL_WAIT_SCRUB)); | |
10296 | break; | |
10297 | } | |
5caeef02 DB |
10298 | case ZPOOL_WAIT_RAIDZ_EXPAND: |
10299 | { | |
10300 | vdev_raidz_expand_t *vre = spa->spa_raidz_expand; | |
10301 | *in_progress = (vre != NULL && vre->vre_state == DSS_SCANNING); | |
10302 | break; | |
10303 | } | |
e60e158e JG |
10304 | default: |
10305 | panic("unrecognized value for activity %d", activity); | |
10306 | } | |
10307 | ||
10308 | return (error); | |
10309 | } | |
10310 | ||
10311 | static int | |
10312 | spa_wait_common(const char *pool, zpool_wait_activity_t activity, | |
10313 | boolean_t use_tag, uint64_t tag, boolean_t *waited) | |
10314 | { | |
10315 | /* | |
10316 | * The tag is used to distinguish between instances of an activity. | |
2288d419 BB |
10317 | * 'initialize' and 'trim' are the only activities that we use this for. |
10318 | * The other activities can only have a single instance in progress in a | |
10319 | * pool at one time, making the tag unnecessary. | |
e60e158e JG |
10320 | * |
10321 | * There can be multiple devices being replaced at once, but since they | |
10322 | * all finish once resilvering finishes, we don't bother keeping track | |
10323 | * of them individually, we just wait for them all to finish. | |
10324 | */ | |
2288d419 BB |
10325 | if (use_tag && activity != ZPOOL_WAIT_INITIALIZE && |
10326 | activity != ZPOOL_WAIT_TRIM) | |
e60e158e JG |
10327 | return (EINVAL); |
10328 | ||
10329 | if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES) | |
10330 | return (EINVAL); | |
10331 | ||
10332 | spa_t *spa; | |
10333 | int error = spa_open(pool, &spa, FTAG); | |
10334 | if (error != 0) | |
10335 | return (error); | |
10336 | ||
10337 | /* | |
10338 | * Increment the spa's waiter count so that we can call spa_close and | |
10339 | * still ensure that the spa_t doesn't get freed before this thread is | |
10340 | * finished with it when the pool is exported. We want to call spa_close | |
10341 | * before we start waiting because otherwise the additional ref would | |
10342 | * prevent the pool from being exported or destroyed throughout the | |
10343 | * potentially long wait. | |
10344 | */ | |
10345 | mutex_enter(&spa->spa_activities_lock); | |
10346 | spa->spa_waiters++; | |
10347 | spa_close(spa, FTAG); | |
10348 | ||
10349 | *waited = B_FALSE; | |
10350 | for (;;) { | |
10351 | boolean_t in_progress; | |
10352 | error = spa_activity_in_progress(spa, activity, use_tag, tag, | |
10353 | &in_progress); | |
10354 | ||
b24771a8 | 10355 | if (error || !in_progress || spa->spa_waiters_cancel) |
e60e158e JG |
10356 | break; |
10357 | ||
10358 | *waited = B_TRUE; | |
10359 | ||
10360 | if (cv_wait_sig(&spa->spa_activities_cv, | |
10361 | &spa->spa_activities_lock) == 0) { | |
10362 | error = EINTR; | |
10363 | break; | |
10364 | } | |
10365 | } | |
10366 | ||
10367 | spa->spa_waiters--; | |
10368 | cv_signal(&spa->spa_waiters_cv); | |
10369 | mutex_exit(&spa->spa_activities_lock); | |
10370 | ||
10371 | return (error); | |
10372 | } | |
10373 | ||
10374 | /* | |
10375 | * Wait for a particular instance of the specified activity to complete, where | |
10376 | * the instance is identified by 'tag' | |
10377 | */ | |
10378 | int | |
10379 | spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag, | |
10380 | boolean_t *waited) | |
10381 | { | |
10382 | return (spa_wait_common(pool, activity, B_TRUE, tag, waited)); | |
10383 | } | |
10384 | ||
10385 | /* | |
10386 | * Wait for all instances of the specified activity complete | |
10387 | */ | |
10388 | int | |
10389 | spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited) | |
10390 | { | |
10391 | ||
10392 | return (spa_wait_common(pool, activity, B_FALSE, 0, waited)); | |
10393 | } | |
10394 | ||
a1d477c2 | 10395 | sysevent_t * |
12fa0466 DE |
10396 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
10397 | { | |
10398 | sysevent_t *ev = NULL; | |
10399 | #ifdef _KERNEL | |
10400 | nvlist_t *resource; | |
10401 | ||
10402 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
10403 | if (resource) { | |
10404 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
10405 | ev->resource = resource; | |
10406 | } | |
14e4e3cb AZ |
10407 | #else |
10408 | (void) spa, (void) vd, (void) hist_nvl, (void) name; | |
12fa0466 DE |
10409 | #endif |
10410 | return (ev); | |
10411 | } | |
10412 | ||
a1d477c2 | 10413 | void |
12fa0466 DE |
10414 | spa_event_post(sysevent_t *ev) |
10415 | { | |
10416 | #ifdef _KERNEL | |
10417 | if (ev) { | |
10418 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
10419 | kmem_free(ev, sizeof (*ev)); | |
10420 | } | |
14e4e3cb AZ |
10421 | #else |
10422 | (void) ev; | |
12fa0466 DE |
10423 | #endif |
10424 | } | |
10425 | ||
34dc7c2f | 10426 | /* |
fb390aaf HR |
10427 | * Post a zevent corresponding to the given sysevent. The 'name' must be one |
10428 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
34dc7c2f BB |
10429 | * filled in from the spa and (optionally) the vdev. This doesn't do anything |
10430 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
10431 | * or zdb as real changes. | |
10432 | */ | |
10433 | void | |
12fa0466 | 10434 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
34dc7c2f | 10435 | { |
12fa0466 | 10436 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); |
34dc7c2f | 10437 | } |
c28b2279 | 10438 | |
c28b2279 BB |
10439 | /* state manipulation functions */ |
10440 | EXPORT_SYMBOL(spa_open); | |
10441 | EXPORT_SYMBOL(spa_open_rewind); | |
10442 | EXPORT_SYMBOL(spa_get_stats); | |
10443 | EXPORT_SYMBOL(spa_create); | |
c28b2279 BB |
10444 | EXPORT_SYMBOL(spa_import); |
10445 | EXPORT_SYMBOL(spa_tryimport); | |
10446 | EXPORT_SYMBOL(spa_destroy); | |
10447 | EXPORT_SYMBOL(spa_export); | |
10448 | EXPORT_SYMBOL(spa_reset); | |
10449 | EXPORT_SYMBOL(spa_async_request); | |
10450 | EXPORT_SYMBOL(spa_async_suspend); | |
10451 | EXPORT_SYMBOL(spa_async_resume); | |
10452 | EXPORT_SYMBOL(spa_inject_addref); | |
10453 | EXPORT_SYMBOL(spa_inject_delref); | |
10454 | EXPORT_SYMBOL(spa_scan_stat_init); | |
10455 | EXPORT_SYMBOL(spa_scan_get_stats); | |
10456 | ||
e1cfd73f | 10457 | /* device manipulation */ |
c28b2279 BB |
10458 | EXPORT_SYMBOL(spa_vdev_add); |
10459 | EXPORT_SYMBOL(spa_vdev_attach); | |
10460 | EXPORT_SYMBOL(spa_vdev_detach); | |
c28b2279 BB |
10461 | EXPORT_SYMBOL(spa_vdev_setpath); |
10462 | EXPORT_SYMBOL(spa_vdev_setfru); | |
10463 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
10464 | ||
10465 | /* spare statech is global across all pools) */ | |
10466 | EXPORT_SYMBOL(spa_spare_add); | |
10467 | EXPORT_SYMBOL(spa_spare_remove); | |
10468 | EXPORT_SYMBOL(spa_spare_exists); | |
10469 | EXPORT_SYMBOL(spa_spare_activate); | |
10470 | ||
10471 | /* L2ARC statech is global across all pools) */ | |
10472 | EXPORT_SYMBOL(spa_l2cache_add); | |
10473 | EXPORT_SYMBOL(spa_l2cache_remove); | |
10474 | EXPORT_SYMBOL(spa_l2cache_exists); | |
10475 | EXPORT_SYMBOL(spa_l2cache_activate); | |
10476 | EXPORT_SYMBOL(spa_l2cache_drop); | |
10477 | ||
10478 | /* scanning */ | |
10479 | EXPORT_SYMBOL(spa_scan); | |
10480 | EXPORT_SYMBOL(spa_scan_stop); | |
10481 | ||
10482 | /* spa syncing */ | |
10483 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
10484 | EXPORT_SYMBOL(spa_sync_allpools); | |
10485 | ||
10486 | /* properties */ | |
10487 | EXPORT_SYMBOL(spa_prop_set); | |
10488 | EXPORT_SYMBOL(spa_prop_get); | |
10489 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
10490 | ||
10491 | /* asynchronous event notification */ | |
10492 | EXPORT_SYMBOL(spa_event_notify); | |
dea377c0 | 10493 | |
342357cd AM |
10494 | ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, preload_pct, UINT, ZMOD_RW, |
10495 | "Percentage of CPUs to run a metaslab preload taskq"); | |
10496 | ||
c8242a96 | 10497 | /* BEGIN CSTYLED */ |
fdc2d303 | 10498 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, UINT, ZMOD_RW, |
458f8231 | 10499 | "log2 fraction of arc that can be used by inflight I/Os when " |
03fdcb9a | 10500 | "verifying pool during import"); |
7ada752a | 10501 | /* END CSTYLED */ |
dea377c0 | 10502 | |
03fdcb9a | 10503 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW, |
dea377c0 MA |
10504 | "Set to traverse metadata on pool import"); |
10505 | ||
03fdcb9a | 10506 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW, |
dea377c0 | 10507 | "Set to traverse data on pool import"); |
dcb6bed1 | 10508 | |
03fdcb9a | 10509 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW, |
6cb8e530 PZ |
10510 | "Print vdev tree to zfs_dbgmsg during pool import"); |
10511 | ||
03fdcb9a | 10512 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD, |
dcb6bed1 D |
10513 | "Percentage of CPUs to run an IO worker thread"); |
10514 | ||
7457b024 AM |
10515 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_tpq, UINT, ZMOD_RD, |
10516 | "Number of threads per IO worker taskqueue"); | |
10517 | ||
7ada752a | 10518 | /* BEGIN CSTYLED */ |
ab8d9c17 | 10519 | ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, U64, ZMOD_RW, |
03fdcb9a MM |
10520 | "Allow importing pool with up to this number of missing top-level " |
10521 | "vdevs (in read-only mode)"); | |
7ada752a | 10522 | /* END CSTYLED */ |
6cb8e530 | 10523 | |
7ada752a AZ |
10524 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, |
10525 | ZMOD_RW, "Set the livelist condense zthr to pause"); | |
03fdcb9a | 10526 | |
7ada752a AZ |
10527 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, |
10528 | ZMOD_RW, "Set the livelist condense synctask to pause"); | |
37f03da8 | 10529 | |
7ada752a AZ |
10530 | /* BEGIN CSTYLED */ |
10531 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, | |
10532 | INT, ZMOD_RW, | |
37f03da8 | 10533 | "Whether livelist condensing was canceled in the synctask"); |
03fdcb9a | 10534 | |
7ada752a AZ |
10535 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, |
10536 | INT, ZMOD_RW, | |
37f03da8 SH |
10537 | "Whether livelist condensing was canceled in the zthr function"); |
10538 | ||
7ada752a AZ |
10539 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, |
10540 | ZMOD_RW, | |
03fdcb9a MM |
10541 | "Whether extra ALLOC blkptrs were added to a livelist entry while it " |
10542 | "was being condensed"); | |
37f03da8 | 10543 | /* END CSTYLED */ |
3bd4df38 EN |
10544 | |
10545 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_wr_iss_ncpus, UINT, ZMOD_RW, | |
10546 | "Number of CPUs to run write issue taskqs"); |