]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or https://opensource.org/licenses/CDDL-1.0. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright (c) 2011, 2024 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved. | |
26 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. | |
27 | * Copyright 2013 Saso Kiselkov. All rights reserved. | |
28 | * Copyright (c) 2014 Integros [integros.com] | |
29 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
30 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
31 | * Copyright 2018 Joyent, Inc. | |
32 | * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. | |
33 | * Copyright 2017 Joyent, Inc. | |
34 | * Copyright (c) 2017, Intel Corporation. | |
35 | * Copyright (c) 2021, Colm Buckley <colm@tuatha.org> | |
36 | * Copyright (c) 2023 Hewlett Packard Enterprise Development LP. | |
37 | * Copyright (c) 2024, Klara Inc. | |
38 | */ | |
39 | ||
40 | /* | |
41 | * SPA: Storage Pool Allocator | |
42 | * | |
43 | * This file contains all the routines used when modifying on-disk SPA state. | |
44 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
45 | * pool. | |
46 | */ | |
47 | ||
48 | #include <sys/zfs_context.h> | |
49 | #include <sys/fm/fs/zfs.h> | |
50 | #include <sys/spa_impl.h> | |
51 | #include <sys/zio.h> | |
52 | #include <sys/zio_checksum.h> | |
53 | #include <sys/dmu.h> | |
54 | #include <sys/dmu_tx.h> | |
55 | #include <sys/zap.h> | |
56 | #include <sys/zil.h> | |
57 | #include <sys/brt.h> | |
58 | #include <sys/ddt.h> | |
59 | #include <sys/vdev_impl.h> | |
60 | #include <sys/vdev_removal.h> | |
61 | #include <sys/vdev_indirect_mapping.h> | |
62 | #include <sys/vdev_indirect_births.h> | |
63 | #include <sys/vdev_initialize.h> | |
64 | #include <sys/vdev_rebuild.h> | |
65 | #include <sys/vdev_trim.h> | |
66 | #include <sys/vdev_disk.h> | |
67 | #include <sys/vdev_raidz.h> | |
68 | #include <sys/vdev_draid.h> | |
69 | #include <sys/metaslab.h> | |
70 | #include <sys/metaslab_impl.h> | |
71 | #include <sys/mmp.h> | |
72 | #include <sys/uberblock_impl.h> | |
73 | #include <sys/txg.h> | |
74 | #include <sys/avl.h> | |
75 | #include <sys/bpobj.h> | |
76 | #include <sys/dmu_traverse.h> | |
77 | #include <sys/dmu_objset.h> | |
78 | #include <sys/unique.h> | |
79 | #include <sys/dsl_pool.h> | |
80 | #include <sys/dsl_dataset.h> | |
81 | #include <sys/dsl_dir.h> | |
82 | #include <sys/dsl_prop.h> | |
83 | #include <sys/dsl_synctask.h> | |
84 | #include <sys/fs/zfs.h> | |
85 | #include <sys/arc.h> | |
86 | #include <sys/callb.h> | |
87 | #include <sys/systeminfo.h> | |
88 | #include <sys/zfs_ioctl.h> | |
89 | #include <sys/dsl_scan.h> | |
90 | #include <sys/zfeature.h> | |
91 | #include <sys/dsl_destroy.h> | |
92 | #include <sys/zvol.h> | |
93 | ||
94 | #ifdef _KERNEL | |
95 | #include <sys/fm/protocol.h> | |
96 | #include <sys/fm/util.h> | |
97 | #include <sys/callb.h> | |
98 | #include <sys/zone.h> | |
99 | #include <sys/vmsystm.h> | |
100 | #endif /* _KERNEL */ | |
101 | ||
102 | #include "zfs_prop.h" | |
103 | #include "zfs_comutil.h" | |
104 | #include <cityhash.h> | |
105 | ||
106 | /* | |
107 | * spa_thread() existed on Illumos as a parent thread for the various worker | |
108 | * threads that actually run the pool, as a way to both reference the entire | |
109 | * pool work as a single object, and to share properties like scheduling | |
110 | * options. It has not yet been adapted to Linux or FreeBSD. This define is | |
111 | * used to mark related parts of the code to make things easier for the reader, | |
112 | * and to compile this code out. It can be removed when someone implements it, | |
113 | * moves it to some Illumos-specific place, or removes it entirely. | |
114 | */ | |
115 | #undef HAVE_SPA_THREAD | |
116 | ||
117 | /* | |
118 | * The "System Duty Cycle" scheduling class is an Illumos feature to help | |
119 | * prevent CPU-intensive kernel threads from affecting latency on interactive | |
120 | * threads. It doesn't exist on Linux or FreeBSD, so the supporting code is | |
121 | * gated behind a define. On Illumos SDC depends on spa_thread(), but | |
122 | * spa_thread() also has other uses, so this is a separate define. | |
123 | */ | |
124 | #undef HAVE_SYSDC | |
125 | ||
126 | /* | |
127 | * The interval, in seconds, at which failed configuration cache file writes | |
128 | * should be retried. | |
129 | */ | |
130 | int zfs_ccw_retry_interval = 300; | |
131 | ||
132 | typedef enum zti_modes { | |
133 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ | |
134 | ZTI_MODE_SCALE, /* Taskqs scale with CPUs. */ | |
135 | ZTI_MODE_SYNC, /* sync thread assigned */ | |
136 | ZTI_MODE_NULL, /* don't create a taskq */ | |
137 | ZTI_NMODES | |
138 | } zti_modes_t; | |
139 | ||
140 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } | |
141 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
142 | #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 } | |
143 | #define ZTI_SYNC { ZTI_MODE_SYNC, 0, 1 } | |
144 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } | |
145 | ||
146 | #define ZTI_N(n) ZTI_P(n, 1) | |
147 | #define ZTI_ONE ZTI_N(1) | |
148 | ||
149 | typedef struct zio_taskq_info { | |
150 | zti_modes_t zti_mode; | |
151 | uint_t zti_value; | |
152 | uint_t zti_count; | |
153 | } zio_taskq_info_t; | |
154 | ||
155 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
156 | "iss", "iss_h", "int", "int_h" | |
157 | }; | |
158 | ||
159 | /* | |
160 | * This table defines the taskq settings for each ZFS I/O type. When | |
161 | * initializing a pool, we use this table to create an appropriately sized | |
162 | * taskq. Some operations are low volume and therefore have a small, static | |
163 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
164 | * macros. Other operations process a large amount of data; the ZTI_SCALE | |
165 | * macro causes us to create a taskq oriented for throughput. Some operations | |
166 | * are so high frequency and short-lived that the taskq itself can become a | |
167 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an | |
168 | * additional degree of parallelism specified by the number of threads per- | |
169 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
170 | * particular taskq is chosen at random. ZTI_SCALE uses a number of taskqs | |
171 | * that scales with the number of CPUs. | |
172 | * | |
173 | * The different taskq priorities are to handle the different contexts (issue | |
174 | * and interrupt) and then to reserve threads for high priority I/Os that | |
175 | * need to be handled with minimum delay. Illumos taskq has unfair TQ_FRONT | |
176 | * implementation, so separate high priority threads are used there. | |
177 | */ | |
178 | static zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { | |
179 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ | |
180 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ | |
181 | { ZTI_N(8), ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* READ */ | |
182 | #ifdef illumos | |
183 | { ZTI_SYNC, ZTI_N(5), ZTI_SCALE, ZTI_N(5) }, /* WRITE */ | |
184 | #else | |
185 | { ZTI_SYNC, ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* WRITE */ | |
186 | #endif | |
187 | { ZTI_SCALE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ | |
188 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ | |
189 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FLUSH */ | |
190 | { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */ | |
191 | }; | |
192 | ||
193 | static void spa_sync_version(void *arg, dmu_tx_t *tx); | |
194 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
195 | static boolean_t spa_has_active_shared_spare(spa_t *spa); | |
196 | static int spa_load_impl(spa_t *spa, spa_import_type_t type, | |
197 | const char **ereport); | |
198 | static void spa_vdev_resilver_done(spa_t *spa); | |
199 | ||
200 | /* | |
201 | * Percentage of all CPUs that can be used by the metaslab preload taskq. | |
202 | */ | |
203 | static uint_t metaslab_preload_pct = 50; | |
204 | ||
205 | static uint_t zio_taskq_batch_pct = 80; /* 1 thread per cpu in pset */ | |
206 | static uint_t zio_taskq_batch_tpq; /* threads per taskq */ | |
207 | ||
208 | #ifdef HAVE_SYSDC | |
209 | static const boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ | |
210 | static const uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
211 | #endif | |
212 | ||
213 | #ifdef HAVE_SPA_THREAD | |
214 | static const boolean_t spa_create_process = B_TRUE; /* no process => no sysdc */ | |
215 | #endif | |
216 | ||
217 | static uint_t zio_taskq_write_tpq = 16; | |
218 | ||
219 | /* | |
220 | * Report any spa_load_verify errors found, but do not fail spa_load. | |
221 | * This is used by zdb to analyze non-idle pools. | |
222 | */ | |
223 | boolean_t spa_load_verify_dryrun = B_FALSE; | |
224 | ||
225 | /* | |
226 | * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ). | |
227 | * This is used by zdb for spacemaps verification. | |
228 | */ | |
229 | boolean_t spa_mode_readable_spacemaps = B_FALSE; | |
230 | ||
231 | /* | |
232 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
233 | * to get the vdev stats associated with the imported devices. | |
234 | */ | |
235 | #define TRYIMPORT_NAME "$import" | |
236 | ||
237 | /* | |
238 | * For debugging purposes: print out vdev tree during pool import. | |
239 | */ | |
240 | static int spa_load_print_vdev_tree = B_FALSE; | |
241 | ||
242 | /* | |
243 | * A non-zero value for zfs_max_missing_tvds means that we allow importing | |
244 | * pools with missing top-level vdevs. This is strictly intended for advanced | |
245 | * pool recovery cases since missing data is almost inevitable. Pools with | |
246 | * missing devices can only be imported read-only for safety reasons, and their | |
247 | * fail-mode will be automatically set to "continue". | |
248 | * | |
249 | * With 1 missing vdev we should be able to import the pool and mount all | |
250 | * datasets. User data that was not modified after the missing device has been | |
251 | * added should be recoverable. This means that snapshots created prior to the | |
252 | * addition of that device should be completely intact. | |
253 | * | |
254 | * With 2 missing vdevs, some datasets may fail to mount since there are | |
255 | * dataset statistics that are stored as regular metadata. Some data might be | |
256 | * recoverable if those vdevs were added recently. | |
257 | * | |
258 | * With 3 or more missing vdevs, the pool is severely damaged and MOS entries | |
259 | * may be missing entirely. Chances of data recovery are very low. Note that | |
260 | * there are also risks of performing an inadvertent rewind as we might be | |
261 | * missing all the vdevs with the latest uberblocks. | |
262 | */ | |
263 | uint64_t zfs_max_missing_tvds = 0; | |
264 | ||
265 | /* | |
266 | * The parameters below are similar to zfs_max_missing_tvds but are only | |
267 | * intended for a preliminary open of the pool with an untrusted config which | |
268 | * might be incomplete or out-dated. | |
269 | * | |
270 | * We are more tolerant for pools opened from a cachefile since we could have | |
271 | * an out-dated cachefile where a device removal was not registered. | |
272 | * We could have set the limit arbitrarily high but in the case where devices | |
273 | * are really missing we would want to return the proper error codes; we chose | |
274 | * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available | |
275 | * and we get a chance to retrieve the trusted config. | |
276 | */ | |
277 | uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; | |
278 | ||
279 | /* | |
280 | * In the case where config was assembled by scanning device paths (/dev/dsks | |
281 | * by default) we are less tolerant since all the existing devices should have | |
282 | * been detected and we want spa_load to return the right error codes. | |
283 | */ | |
284 | uint64_t zfs_max_missing_tvds_scan = 0; | |
285 | ||
286 | /* | |
287 | * Debugging aid that pauses spa_sync() towards the end. | |
288 | */ | |
289 | static const boolean_t zfs_pause_spa_sync = B_FALSE; | |
290 | ||
291 | /* | |
292 | * Variables to indicate the livelist condense zthr func should wait at certain | |
293 | * points for the livelist to be removed - used to test condense/destroy races | |
294 | */ | |
295 | static int zfs_livelist_condense_zthr_pause = 0; | |
296 | static int zfs_livelist_condense_sync_pause = 0; | |
297 | ||
298 | /* | |
299 | * Variables to track whether or not condense cancellation has been | |
300 | * triggered in testing. | |
301 | */ | |
302 | static int zfs_livelist_condense_sync_cancel = 0; | |
303 | static int zfs_livelist_condense_zthr_cancel = 0; | |
304 | ||
305 | /* | |
306 | * Variable to track whether or not extra ALLOC blkptrs were added to a | |
307 | * livelist entry while it was being condensed (caused by the way we track | |
308 | * remapped blkptrs in dbuf_remap_impl) | |
309 | */ | |
310 | static int zfs_livelist_condense_new_alloc = 0; | |
311 | ||
312 | /* | |
313 | * ========================================================================== | |
314 | * SPA properties routines | |
315 | * ========================================================================== | |
316 | */ | |
317 | ||
318 | /* | |
319 | * Add a (source=src, propname=propval) list to an nvlist. | |
320 | */ | |
321 | static void | |
322 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, const char *strval, | |
323 | uint64_t intval, zprop_source_t src) | |
324 | { | |
325 | const char *propname = zpool_prop_to_name(prop); | |
326 | nvlist_t *propval; | |
327 | ||
328 | propval = fnvlist_alloc(); | |
329 | fnvlist_add_uint64(propval, ZPROP_SOURCE, src); | |
330 | ||
331 | if (strval != NULL) | |
332 | fnvlist_add_string(propval, ZPROP_VALUE, strval); | |
333 | else | |
334 | fnvlist_add_uint64(propval, ZPROP_VALUE, intval); | |
335 | ||
336 | fnvlist_add_nvlist(nvl, propname, propval); | |
337 | nvlist_free(propval); | |
338 | } | |
339 | ||
340 | /* | |
341 | * Add a user property (source=src, propname=propval) to an nvlist. | |
342 | */ | |
343 | static void | |
344 | spa_prop_add_user(nvlist_t *nvl, const char *propname, char *strval, | |
345 | zprop_source_t src) | |
346 | { | |
347 | nvlist_t *propval; | |
348 | ||
349 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
350 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); | |
351 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
352 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
353 | nvlist_free(propval); | |
354 | } | |
355 | ||
356 | /* | |
357 | * Get property values from the spa configuration. | |
358 | */ | |
359 | static void | |
360 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
361 | { | |
362 | vdev_t *rvd = spa->spa_root_vdev; | |
363 | dsl_pool_t *pool = spa->spa_dsl_pool; | |
364 | uint64_t size, alloc, cap, version; | |
365 | const zprop_source_t src = ZPROP_SRC_NONE; | |
366 | spa_config_dirent_t *dp; | |
367 | metaslab_class_t *mc = spa_normal_class(spa); | |
368 | ||
369 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
370 | ||
371 | if (rvd != NULL) { | |
372 | alloc = metaslab_class_get_alloc(mc); | |
373 | alloc += metaslab_class_get_alloc(spa_special_class(spa)); | |
374 | alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); | |
375 | alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa)); | |
376 | ||
377 | size = metaslab_class_get_space(mc); | |
378 | size += metaslab_class_get_space(spa_special_class(spa)); | |
379 | size += metaslab_class_get_space(spa_dedup_class(spa)); | |
380 | size += metaslab_class_get_space(spa_embedded_log_class(spa)); | |
381 | ||
382 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); | |
383 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
384 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); | |
385 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
386 | size - alloc, src); | |
387 | spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, | |
388 | spa->spa_checkpoint_info.sci_dspace, src); | |
389 | ||
390 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, | |
391 | metaslab_class_fragmentation(mc), src); | |
392 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
393 | metaslab_class_expandable_space(mc), src); | |
394 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, | |
395 | (spa_mode(spa) == SPA_MODE_READ), src); | |
396 | ||
397 | cap = (size == 0) ? 0 : (alloc * 100 / size); | |
398 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); | |
399 | ||
400 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, | |
401 | ddt_get_pool_dedup_ratio(spa), src); | |
402 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONEUSED, NULL, | |
403 | brt_get_used(spa), src); | |
404 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONESAVED, NULL, | |
405 | brt_get_saved(spa), src); | |
406 | spa_prop_add_list(*nvp, ZPOOL_PROP_BCLONERATIO, NULL, | |
407 | brt_get_ratio(spa), src); | |
408 | ||
409 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, | |
410 | rvd->vdev_state, src); | |
411 | ||
412 | version = spa_version(spa); | |
413 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { | |
414 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
415 | version, ZPROP_SRC_DEFAULT); | |
416 | } else { | |
417 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
418 | version, ZPROP_SRC_LOCAL); | |
419 | } | |
420 | spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID, | |
421 | NULL, spa_load_guid(spa), src); | |
422 | } | |
423 | ||
424 | if (pool != NULL) { | |
425 | /* | |
426 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
427 | * when opening pools before this version freedir will be NULL. | |
428 | */ | |
429 | if (pool->dp_free_dir != NULL) { | |
430 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, | |
431 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, | |
432 | src); | |
433 | } else { | |
434 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
435 | NULL, 0, src); | |
436 | } | |
437 | ||
438 | if (pool->dp_leak_dir != NULL) { | |
439 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
440 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, | |
441 | src); | |
442 | } else { | |
443 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
444 | NULL, 0, src); | |
445 | } | |
446 | } | |
447 | ||
448 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); | |
449 | ||
450 | if (spa->spa_comment != NULL) { | |
451 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
452 | 0, ZPROP_SRC_LOCAL); | |
453 | } | |
454 | ||
455 | if (spa->spa_compatibility != NULL) { | |
456 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMPATIBILITY, | |
457 | spa->spa_compatibility, 0, ZPROP_SRC_LOCAL); | |
458 | } | |
459 | ||
460 | if (spa->spa_root != NULL) | |
461 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
462 | 0, ZPROP_SRC_LOCAL); | |
463 | ||
464 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { | |
465 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
466 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
467 | } else { | |
468 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
469 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
470 | } | |
471 | ||
472 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { | |
473 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
474 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
475 | } else { | |
476 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
477 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
478 | } | |
479 | ||
480 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { | |
481 | if (dp->scd_path == NULL) { | |
482 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
483 | "none", 0, ZPROP_SRC_LOCAL); | |
484 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
485 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
486 | dp->scd_path, 0, ZPROP_SRC_LOCAL); | |
487 | } | |
488 | } | |
489 | } | |
490 | ||
491 | /* | |
492 | * Get zpool property values. | |
493 | */ | |
494 | int | |
495 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
496 | { | |
497 | objset_t *mos = spa->spa_meta_objset; | |
498 | zap_cursor_t zc; | |
499 | zap_attribute_t za; | |
500 | dsl_pool_t *dp; | |
501 | int err; | |
502 | ||
503 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); | |
504 | if (err) | |
505 | return (err); | |
506 | ||
507 | dp = spa_get_dsl(spa); | |
508 | dsl_pool_config_enter(dp, FTAG); | |
509 | mutex_enter(&spa->spa_props_lock); | |
510 | ||
511 | /* | |
512 | * Get properties from the spa config. | |
513 | */ | |
514 | spa_prop_get_config(spa, nvp); | |
515 | ||
516 | /* If no pool property object, no more prop to get. */ | |
517 | if (mos == NULL || spa->spa_pool_props_object == 0) | |
518 | goto out; | |
519 | ||
520 | /* | |
521 | * Get properties from the MOS pool property object. | |
522 | */ | |
523 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
524 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
525 | zap_cursor_advance(&zc)) { | |
526 | uint64_t intval = 0; | |
527 | char *strval = NULL; | |
528 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
529 | zpool_prop_t prop; | |
530 | ||
531 | if ((prop = zpool_name_to_prop(za.za_name)) == | |
532 | ZPOOL_PROP_INVAL && !zfs_prop_user(za.za_name)) | |
533 | continue; | |
534 | ||
535 | switch (za.za_integer_length) { | |
536 | case 8: | |
537 | /* integer property */ | |
538 | if (za.za_first_integer != | |
539 | zpool_prop_default_numeric(prop)) | |
540 | src = ZPROP_SRC_LOCAL; | |
541 | ||
542 | if (prop == ZPOOL_PROP_BOOTFS) { | |
543 | dsl_dataset_t *ds = NULL; | |
544 | ||
545 | err = dsl_dataset_hold_obj(dp, | |
546 | za.za_first_integer, FTAG, &ds); | |
547 | if (err != 0) | |
548 | break; | |
549 | ||
550 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, | |
551 | KM_SLEEP); | |
552 | dsl_dataset_name(ds, strval); | |
553 | dsl_dataset_rele(ds, FTAG); | |
554 | } else { | |
555 | strval = NULL; | |
556 | intval = za.za_first_integer; | |
557 | } | |
558 | ||
559 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
560 | ||
561 | if (strval != NULL) | |
562 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); | |
563 | ||
564 | break; | |
565 | ||
566 | case 1: | |
567 | /* string property */ | |
568 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); | |
569 | err = zap_lookup(mos, spa->spa_pool_props_object, | |
570 | za.za_name, 1, za.za_num_integers, strval); | |
571 | if (err) { | |
572 | kmem_free(strval, za.za_num_integers); | |
573 | break; | |
574 | } | |
575 | if (prop != ZPOOL_PROP_INVAL) { | |
576 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
577 | } else { | |
578 | src = ZPROP_SRC_LOCAL; | |
579 | spa_prop_add_user(*nvp, za.za_name, strval, | |
580 | src); | |
581 | } | |
582 | kmem_free(strval, za.za_num_integers); | |
583 | break; | |
584 | ||
585 | default: | |
586 | break; | |
587 | } | |
588 | } | |
589 | zap_cursor_fini(&zc); | |
590 | out: | |
591 | mutex_exit(&spa->spa_props_lock); | |
592 | dsl_pool_config_exit(dp, FTAG); | |
593 | if (err && err != ENOENT) { | |
594 | nvlist_free(*nvp); | |
595 | *nvp = NULL; | |
596 | return (err); | |
597 | } | |
598 | ||
599 | return (0); | |
600 | } | |
601 | ||
602 | /* | |
603 | * Validate the given pool properties nvlist and modify the list | |
604 | * for the property values to be set. | |
605 | */ | |
606 | static int | |
607 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
608 | { | |
609 | nvpair_t *elem; | |
610 | int error = 0, reset_bootfs = 0; | |
611 | uint64_t objnum = 0; | |
612 | boolean_t has_feature = B_FALSE; | |
613 | ||
614 | elem = NULL; | |
615 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
616 | uint64_t intval; | |
617 | const char *strval, *slash, *check, *fname; | |
618 | const char *propname = nvpair_name(elem); | |
619 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
620 | ||
621 | switch (prop) { | |
622 | case ZPOOL_PROP_INVAL: | |
623 | /* | |
624 | * Sanitize the input. | |
625 | */ | |
626 | if (zfs_prop_user(propname)) { | |
627 | if (strlen(propname) >= ZAP_MAXNAMELEN) { | |
628 | error = SET_ERROR(ENAMETOOLONG); | |
629 | break; | |
630 | } | |
631 | ||
632 | if (strlen(fnvpair_value_string(elem)) >= | |
633 | ZAP_MAXVALUELEN) { | |
634 | error = SET_ERROR(E2BIG); | |
635 | break; | |
636 | } | |
637 | } else if (zpool_prop_feature(propname)) { | |
638 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
639 | error = SET_ERROR(EINVAL); | |
640 | break; | |
641 | } | |
642 | ||
643 | if (nvpair_value_uint64(elem, &intval) != 0) { | |
644 | error = SET_ERROR(EINVAL); | |
645 | break; | |
646 | } | |
647 | ||
648 | if (intval != 0) { | |
649 | error = SET_ERROR(EINVAL); | |
650 | break; | |
651 | } | |
652 | ||
653 | fname = strchr(propname, '@') + 1; | |
654 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
655 | error = SET_ERROR(EINVAL); | |
656 | break; | |
657 | } | |
658 | ||
659 | has_feature = B_TRUE; | |
660 | } else { | |
661 | error = SET_ERROR(EINVAL); | |
662 | break; | |
663 | } | |
664 | break; | |
665 | ||
666 | case ZPOOL_PROP_VERSION: | |
667 | error = nvpair_value_uint64(elem, &intval); | |
668 | if (!error && | |
669 | (intval < spa_version(spa) || | |
670 | intval > SPA_VERSION_BEFORE_FEATURES || | |
671 | has_feature)) | |
672 | error = SET_ERROR(EINVAL); | |
673 | break; | |
674 | ||
675 | case ZPOOL_PROP_DELEGATION: | |
676 | case ZPOOL_PROP_AUTOREPLACE: | |
677 | case ZPOOL_PROP_LISTSNAPS: | |
678 | case ZPOOL_PROP_AUTOEXPAND: | |
679 | case ZPOOL_PROP_AUTOTRIM: | |
680 | error = nvpair_value_uint64(elem, &intval); | |
681 | if (!error && intval > 1) | |
682 | error = SET_ERROR(EINVAL); | |
683 | break; | |
684 | ||
685 | case ZPOOL_PROP_MULTIHOST: | |
686 | error = nvpair_value_uint64(elem, &intval); | |
687 | if (!error && intval > 1) | |
688 | error = SET_ERROR(EINVAL); | |
689 | ||
690 | if (!error) { | |
691 | uint32_t hostid = zone_get_hostid(NULL); | |
692 | if (hostid) | |
693 | spa->spa_hostid = hostid; | |
694 | else | |
695 | error = SET_ERROR(ENOTSUP); | |
696 | } | |
697 | ||
698 | break; | |
699 | ||
700 | case ZPOOL_PROP_BOOTFS: | |
701 | /* | |
702 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
703 | * or the pool is still being created (version == 0), | |
704 | * the bootfs property cannot be set. | |
705 | */ | |
706 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { | |
707 | error = SET_ERROR(ENOTSUP); | |
708 | break; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Make sure the vdev config is bootable | |
713 | */ | |
714 | if (!vdev_is_bootable(spa->spa_root_vdev)) { | |
715 | error = SET_ERROR(ENOTSUP); | |
716 | break; | |
717 | } | |
718 | ||
719 | reset_bootfs = 1; | |
720 | ||
721 | error = nvpair_value_string(elem, &strval); | |
722 | ||
723 | if (!error) { | |
724 | objset_t *os; | |
725 | ||
726 | if (strval == NULL || strval[0] == '\0') { | |
727 | objnum = zpool_prop_default_numeric( | |
728 | ZPOOL_PROP_BOOTFS); | |
729 | break; | |
730 | } | |
731 | ||
732 | error = dmu_objset_hold(strval, FTAG, &os); | |
733 | if (error != 0) | |
734 | break; | |
735 | ||
736 | /* Must be ZPL. */ | |
737 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
738 | error = SET_ERROR(ENOTSUP); | |
739 | } else { | |
740 | objnum = dmu_objset_id(os); | |
741 | } | |
742 | dmu_objset_rele(os, FTAG); | |
743 | } | |
744 | break; | |
745 | ||
746 | case ZPOOL_PROP_FAILUREMODE: | |
747 | error = nvpair_value_uint64(elem, &intval); | |
748 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) | |
749 | error = SET_ERROR(EINVAL); | |
750 | ||
751 | /* | |
752 | * This is a special case which only occurs when | |
753 | * the pool has completely failed. This allows | |
754 | * the user to change the in-core failmode property | |
755 | * without syncing it out to disk (I/Os might | |
756 | * currently be blocked). We do this by returning | |
757 | * EIO to the caller (spa_prop_set) to trick it | |
758 | * into thinking we encountered a property validation | |
759 | * error. | |
760 | */ | |
761 | if (!error && spa_suspended(spa)) { | |
762 | spa->spa_failmode = intval; | |
763 | error = SET_ERROR(EIO); | |
764 | } | |
765 | break; | |
766 | ||
767 | case ZPOOL_PROP_CACHEFILE: | |
768 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
769 | break; | |
770 | ||
771 | if (strval[0] == '\0') | |
772 | break; | |
773 | ||
774 | if (strcmp(strval, "none") == 0) | |
775 | break; | |
776 | ||
777 | if (strval[0] != '/') { | |
778 | error = SET_ERROR(EINVAL); | |
779 | break; | |
780 | } | |
781 | ||
782 | slash = strrchr(strval, '/'); | |
783 | ASSERT(slash != NULL); | |
784 | ||
785 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
786 | strcmp(slash, "/..") == 0) | |
787 | error = SET_ERROR(EINVAL); | |
788 | break; | |
789 | ||
790 | case ZPOOL_PROP_COMMENT: | |
791 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
792 | break; | |
793 | for (check = strval; *check != '\0'; check++) { | |
794 | if (!isprint(*check)) { | |
795 | error = SET_ERROR(EINVAL); | |
796 | break; | |
797 | } | |
798 | } | |
799 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
800 | error = SET_ERROR(E2BIG); | |
801 | break; | |
802 | ||
803 | default: | |
804 | break; | |
805 | } | |
806 | ||
807 | if (error) | |
808 | break; | |
809 | } | |
810 | ||
811 | (void) nvlist_remove_all(props, | |
812 | zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO)); | |
813 | ||
814 | if (!error && reset_bootfs) { | |
815 | error = nvlist_remove(props, | |
816 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
817 | ||
818 | if (!error) { | |
819 | error = nvlist_add_uint64(props, | |
820 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
821 | } | |
822 | } | |
823 | ||
824 | return (error); | |
825 | } | |
826 | ||
827 | void | |
828 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
829 | { | |
830 | const char *cachefile; | |
831 | spa_config_dirent_t *dp; | |
832 | ||
833 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
834 | &cachefile) != 0) | |
835 | return; | |
836 | ||
837 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
838 | KM_SLEEP); | |
839 | ||
840 | if (cachefile[0] == '\0') | |
841 | dp->scd_path = spa_strdup(spa_config_path); | |
842 | else if (strcmp(cachefile, "none") == 0) | |
843 | dp->scd_path = NULL; | |
844 | else | |
845 | dp->scd_path = spa_strdup(cachefile); | |
846 | ||
847 | list_insert_head(&spa->spa_config_list, dp); | |
848 | if (need_sync) | |
849 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
850 | } | |
851 | ||
852 | int | |
853 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
854 | { | |
855 | int error; | |
856 | nvpair_t *elem = NULL; | |
857 | boolean_t need_sync = B_FALSE; | |
858 | ||
859 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
860 | return (error); | |
861 | ||
862 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { | |
863 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); | |
864 | ||
865 | if (prop == ZPOOL_PROP_CACHEFILE || | |
866 | prop == ZPOOL_PROP_ALTROOT || | |
867 | prop == ZPOOL_PROP_READONLY) | |
868 | continue; | |
869 | ||
870 | if (prop == ZPOOL_PROP_INVAL && | |
871 | zfs_prop_user(nvpair_name(elem))) { | |
872 | need_sync = B_TRUE; | |
873 | break; | |
874 | } | |
875 | ||
876 | if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { | |
877 | uint64_t ver = 0; | |
878 | ||
879 | if (prop == ZPOOL_PROP_VERSION) { | |
880 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
881 | } else { | |
882 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
883 | ver = SPA_VERSION_FEATURES; | |
884 | need_sync = B_TRUE; | |
885 | } | |
886 | ||
887 | /* Save time if the version is already set. */ | |
888 | if (ver == spa_version(spa)) | |
889 | continue; | |
890 | ||
891 | /* | |
892 | * In addition to the pool directory object, we might | |
893 | * create the pool properties object, the features for | |
894 | * read object, the features for write object, or the | |
895 | * feature descriptions object. | |
896 | */ | |
897 | error = dsl_sync_task(spa->spa_name, NULL, | |
898 | spa_sync_version, &ver, | |
899 | 6, ZFS_SPACE_CHECK_RESERVED); | |
900 | if (error) | |
901 | return (error); | |
902 | continue; | |
903 | } | |
904 | ||
905 | need_sync = B_TRUE; | |
906 | break; | |
907 | } | |
908 | ||
909 | if (need_sync) { | |
910 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, | |
911 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); | |
912 | } | |
913 | ||
914 | return (0); | |
915 | } | |
916 | ||
917 | /* | |
918 | * If the bootfs property value is dsobj, clear it. | |
919 | */ | |
920 | void | |
921 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
922 | { | |
923 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
924 | VERIFY(zap_remove(spa->spa_meta_objset, | |
925 | spa->spa_pool_props_object, | |
926 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
927 | spa->spa_bootfs = 0; | |
928 | } | |
929 | } | |
930 | ||
931 | static int | |
932 | spa_change_guid_check(void *arg, dmu_tx_t *tx) | |
933 | { | |
934 | uint64_t *newguid __maybe_unused = arg; | |
935 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
936 | vdev_t *rvd = spa->spa_root_vdev; | |
937 | uint64_t vdev_state; | |
938 | ||
939 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
940 | int error = (spa_has_checkpoint(spa)) ? | |
941 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
942 | return (SET_ERROR(error)); | |
943 | } | |
944 | ||
945 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
946 | vdev_state = rvd->vdev_state; | |
947 | spa_config_exit(spa, SCL_STATE, FTAG); | |
948 | ||
949 | if (vdev_state != VDEV_STATE_HEALTHY) | |
950 | return (SET_ERROR(ENXIO)); | |
951 | ||
952 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
953 | ||
954 | return (0); | |
955 | } | |
956 | ||
957 | static void | |
958 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) | |
959 | { | |
960 | uint64_t *newguid = arg; | |
961 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
962 | uint64_t oldguid; | |
963 | vdev_t *rvd = spa->spa_root_vdev; | |
964 | ||
965 | oldguid = spa_guid(spa); | |
966 | ||
967 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
968 | rvd->vdev_guid = *newguid; | |
969 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
970 | vdev_config_dirty(rvd); | |
971 | spa_config_exit(spa, SCL_STATE, FTAG); | |
972 | ||
973 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", | |
974 | (u_longlong_t)oldguid, (u_longlong_t)*newguid); | |
975 | } | |
976 | ||
977 | /* | |
978 | * Change the GUID for the pool. This is done so that we can later | |
979 | * re-import a pool built from a clone of our own vdevs. We will modify | |
980 | * the root vdev's guid, our own pool guid, and then mark all of our | |
981 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
982 | * online when we do this, or else any vdevs that weren't present | |
983 | * would be orphaned from our pool. We are also going to issue a | |
984 | * sysevent to update any watchers. | |
985 | */ | |
986 | int | |
987 | spa_change_guid(spa_t *spa) | |
988 | { | |
989 | int error; | |
990 | uint64_t guid; | |
991 | ||
992 | mutex_enter(&spa->spa_vdev_top_lock); | |
993 | mutex_enter(&spa_namespace_lock); | |
994 | guid = spa_generate_guid(NULL); | |
995 | ||
996 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, | |
997 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); | |
998 | ||
999 | if (error == 0) { | |
1000 | /* | |
1001 | * Clear the kobj flag from all the vdevs to allow | |
1002 | * vdev_cache_process_kobj_evt() to post events to all the | |
1003 | * vdevs since GUID is updated. | |
1004 | */ | |
1005 | vdev_clear_kobj_evt(spa->spa_root_vdev); | |
1006 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
1007 | vdev_clear_kobj_evt(spa->spa_l2cache.sav_vdevs[i]); | |
1008 | ||
1009 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); | |
1010 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); | |
1011 | } | |
1012 | ||
1013 | mutex_exit(&spa_namespace_lock); | |
1014 | mutex_exit(&spa->spa_vdev_top_lock); | |
1015 | ||
1016 | return (error); | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * ========================================================================== | |
1021 | * SPA state manipulation (open/create/destroy/import/export) | |
1022 | * ========================================================================== | |
1023 | */ | |
1024 | ||
1025 | static int | |
1026 | spa_error_entry_compare(const void *a, const void *b) | |
1027 | { | |
1028 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; | |
1029 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
1030 | int ret; | |
1031 | ||
1032 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, | |
1033 | sizeof (zbookmark_phys_t)); | |
1034 | ||
1035 | return (TREE_ISIGN(ret)); | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * Utility function which retrieves copies of the current logs and | |
1040 | * re-initializes them in the process. | |
1041 | */ | |
1042 | void | |
1043 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
1044 | { | |
1045 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
1046 | ||
1047 | memcpy(last, &spa->spa_errlist_last, sizeof (avl_tree_t)); | |
1048 | memcpy(scrub, &spa->spa_errlist_scrub, sizeof (avl_tree_t)); | |
1049 | ||
1050 | avl_create(&spa->spa_errlist_scrub, | |
1051 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1052 | offsetof(spa_error_entry_t, se_avl)); | |
1053 | avl_create(&spa->spa_errlist_last, | |
1054 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1055 | offsetof(spa_error_entry_t, se_avl)); | |
1056 | } | |
1057 | ||
1058 | static void | |
1059 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
1060 | { | |
1061 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; | |
1062 | enum zti_modes mode = ztip->zti_mode; | |
1063 | uint_t value = ztip->zti_value; | |
1064 | uint_t count = ztip->zti_count; | |
1065 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1066 | uint_t cpus, flags = TASKQ_DYNAMIC; | |
1067 | ||
1068 | switch (mode) { | |
1069 | case ZTI_MODE_FIXED: | |
1070 | ASSERT3U(value, >, 0); | |
1071 | break; | |
1072 | ||
1073 | case ZTI_MODE_SYNC: | |
1074 | ||
1075 | /* | |
1076 | * Create one wr_iss taskq for every 'zio_taskq_write_tpq' CPUs, | |
1077 | * not to exceed the number of spa allocators, and align to it. | |
1078 | */ | |
1079 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); | |
1080 | count = MAX(1, cpus / MAX(1, zio_taskq_write_tpq)); | |
1081 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1082 | count = MIN(count, spa->spa_alloc_count); | |
1083 | while (spa->spa_alloc_count % count != 0 && | |
1084 | spa->spa_alloc_count < count * 2) | |
1085 | count--; | |
1086 | ||
1087 | /* | |
1088 | * zio_taskq_batch_pct is unbounded and may exceed 100%, but no | |
1089 | * single taskq may have more threads than 100% of online cpus. | |
1090 | */ | |
1091 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1092 | value = MIN(value, 100); | |
1093 | flags |= TASKQ_THREADS_CPU_PCT; | |
1094 | break; | |
1095 | ||
1096 | case ZTI_MODE_SCALE: | |
1097 | flags |= TASKQ_THREADS_CPU_PCT; | |
1098 | /* | |
1099 | * We want more taskqs to reduce lock contention, but we want | |
1100 | * less for better request ordering and CPU utilization. | |
1101 | */ | |
1102 | cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100); | |
1103 | if (zio_taskq_batch_tpq > 0) { | |
1104 | count = MAX(1, (cpus + zio_taskq_batch_tpq / 2) / | |
1105 | zio_taskq_batch_tpq); | |
1106 | } else { | |
1107 | /* | |
1108 | * Prefer 6 threads per taskq, but no more taskqs | |
1109 | * than threads in them on large systems. For 80%: | |
1110 | * | |
1111 | * taskq taskq total | |
1112 | * cpus taskqs percent threads threads | |
1113 | * ------- ------- ------- ------- ------- | |
1114 | * 1 1 80% 1 1 | |
1115 | * 2 1 80% 1 1 | |
1116 | * 4 1 80% 3 3 | |
1117 | * 8 2 40% 3 6 | |
1118 | * 16 3 27% 4 12 | |
1119 | * 32 5 16% 5 25 | |
1120 | * 64 7 11% 7 49 | |
1121 | * 128 10 8% 10 100 | |
1122 | * 256 14 6% 15 210 | |
1123 | */ | |
1124 | count = 1 + cpus / 6; | |
1125 | while (count * count > cpus) | |
1126 | count--; | |
1127 | } | |
1128 | /* Limit each taskq within 100% to not trigger assertion. */ | |
1129 | count = MAX(count, (zio_taskq_batch_pct + 99) / 100); | |
1130 | value = (zio_taskq_batch_pct + count / 2) / count; | |
1131 | break; | |
1132 | ||
1133 | case ZTI_MODE_NULL: | |
1134 | tqs->stqs_count = 0; | |
1135 | tqs->stqs_taskq = NULL; | |
1136 | return; | |
1137 | ||
1138 | default: | |
1139 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
1140 | "spa_taskqs_init()", | |
1141 | zio_type_name[t], zio_taskq_types[q], mode, value); | |
1142 | break; | |
1143 | } | |
1144 | ||
1145 | ASSERT3U(count, >, 0); | |
1146 | tqs->stqs_count = count; | |
1147 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
1148 | ||
1149 | for (uint_t i = 0; i < count; i++) { | |
1150 | taskq_t *tq; | |
1151 | char name[32]; | |
1152 | ||
1153 | if (count > 1) | |
1154 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
1155 | zio_type_name[t], zio_taskq_types[q], i); | |
1156 | else | |
1157 | (void) snprintf(name, sizeof (name), "%s_%s", | |
1158 | zio_type_name[t], zio_taskq_types[q]); | |
1159 | ||
1160 | #ifdef HAVE_SYSDC | |
1161 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { | |
1162 | (void) zio_taskq_basedc; | |
1163 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, | |
1164 | spa->spa_proc, zio_taskq_basedc, flags); | |
1165 | } else { | |
1166 | #endif | |
1167 | pri_t pri = maxclsyspri; | |
1168 | /* | |
1169 | * The write issue taskq can be extremely CPU | |
1170 | * intensive. Run it at slightly less important | |
1171 | * priority than the other taskqs. | |
1172 | * | |
1173 | * Under Linux and FreeBSD this means incrementing | |
1174 | * the priority value as opposed to platforms like | |
1175 | * illumos where it should be decremented. | |
1176 | * | |
1177 | * On FreeBSD, if priorities divided by four (RQ_PPQ) | |
1178 | * are equal then a difference between them is | |
1179 | * insignificant. | |
1180 | */ | |
1181 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) { | |
1182 | #if defined(__linux__) | |
1183 | pri++; | |
1184 | #elif defined(__FreeBSD__) | |
1185 | pri += 4; | |
1186 | #else | |
1187 | #error "unknown OS" | |
1188 | #endif | |
1189 | } | |
1190 | tq = taskq_create_proc(name, value, pri, 50, | |
1191 | INT_MAX, spa->spa_proc, flags); | |
1192 | #ifdef HAVE_SYSDC | |
1193 | } | |
1194 | #endif | |
1195 | ||
1196 | tqs->stqs_taskq[i] = tq; | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | static void | |
1201 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
1202 | { | |
1203 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1204 | ||
1205 | if (tqs->stqs_taskq == NULL) { | |
1206 | ASSERT3U(tqs->stqs_count, ==, 0); | |
1207 | return; | |
1208 | } | |
1209 | ||
1210 | for (uint_t i = 0; i < tqs->stqs_count; i++) { | |
1211 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); | |
1212 | taskq_destroy(tqs->stqs_taskq[i]); | |
1213 | } | |
1214 | ||
1215 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); | |
1216 | tqs->stqs_taskq = NULL; | |
1217 | } | |
1218 | ||
1219 | #ifdef _KERNEL | |
1220 | /* | |
1221 | * The READ and WRITE rows of zio_taskqs are configurable at module load time | |
1222 | * by setting zio_taskq_read or zio_taskq_write. | |
1223 | * | |
1224 | * Example (the defaults for READ and WRITE) | |
1225 | * zio_taskq_read='fixed,1,8 null scale null' | |
1226 | * zio_taskq_write='sync null scale null' | |
1227 | * | |
1228 | * Each sets the entire row at a time. | |
1229 | * | |
1230 | * 'fixed' is parameterised: fixed,Q,T where Q is number of taskqs, T is number | |
1231 | * of threads per taskq. | |
1232 | * | |
1233 | * 'null' can only be set on the high-priority queues (queue selection for | |
1234 | * high-priority queues will fall back to the regular queue if the high-pri | |
1235 | * is NULL. | |
1236 | */ | |
1237 | static const char *const modes[ZTI_NMODES] = { | |
1238 | "fixed", "scale", "sync", "null" | |
1239 | }; | |
1240 | ||
1241 | /* Parse the incoming config string. Modifies cfg */ | |
1242 | static int | |
1243 | spa_taskq_param_set(zio_type_t t, char *cfg) | |
1244 | { | |
1245 | int err = 0; | |
1246 | ||
1247 | zio_taskq_info_t row[ZIO_TASKQ_TYPES] = {{0}}; | |
1248 | ||
1249 | char *next = cfg, *tok, *c; | |
1250 | ||
1251 | /* | |
1252 | * Parse out each element from the string and fill `row`. The entire | |
1253 | * row has to be set at once, so any errors are flagged by just | |
1254 | * breaking out of this loop early. | |
1255 | */ | |
1256 | uint_t q; | |
1257 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1258 | /* `next` is the start of the config */ | |
1259 | if (next == NULL) | |
1260 | break; | |
1261 | ||
1262 | /* Eat up leading space */ | |
1263 | while (isspace(*next)) | |
1264 | next++; | |
1265 | if (*next == '\0') | |
1266 | break; | |
1267 | ||
1268 | /* Mode ends at space or end of string */ | |
1269 | tok = next; | |
1270 | next = strchr(tok, ' '); | |
1271 | if (next != NULL) *next++ = '\0'; | |
1272 | ||
1273 | /* Parameters start after a comma */ | |
1274 | c = strchr(tok, ','); | |
1275 | if (c != NULL) *c++ = '\0'; | |
1276 | ||
1277 | /* Match mode string */ | |
1278 | uint_t mode; | |
1279 | for (mode = 0; mode < ZTI_NMODES; mode++) | |
1280 | if (strcmp(tok, modes[mode]) == 0) | |
1281 | break; | |
1282 | if (mode == ZTI_NMODES) | |
1283 | break; | |
1284 | ||
1285 | /* Invalid canary */ | |
1286 | row[q].zti_mode = ZTI_NMODES; | |
1287 | ||
1288 | /* Per-mode setup */ | |
1289 | switch (mode) { | |
1290 | ||
1291 | /* | |
1292 | * FIXED is parameterised: number of queues, and number of | |
1293 | * threads per queue. | |
1294 | */ | |
1295 | case ZTI_MODE_FIXED: { | |
1296 | /* No parameters? */ | |
1297 | if (c == NULL || *c == '\0') | |
1298 | break; | |
1299 | ||
1300 | /* Find next parameter */ | |
1301 | tok = c; | |
1302 | c = strchr(tok, ','); | |
1303 | if (c == NULL) | |
1304 | break; | |
1305 | ||
1306 | /* Take digits and convert */ | |
1307 | unsigned long long nq; | |
1308 | if (!(isdigit(*tok))) | |
1309 | break; | |
1310 | err = ddi_strtoull(tok, &tok, 10, &nq); | |
1311 | /* Must succeed and also end at the next param sep */ | |
1312 | if (err != 0 || tok != c) | |
1313 | break; | |
1314 | ||
1315 | /* Move past the comma */ | |
1316 | tok++; | |
1317 | /* Need another number */ | |
1318 | if (!(isdigit(*tok))) | |
1319 | break; | |
1320 | /* Remember start to make sure we moved */ | |
1321 | c = tok; | |
1322 | ||
1323 | /* Take digits */ | |
1324 | unsigned long long ntpq; | |
1325 | err = ddi_strtoull(tok, &tok, 10, &ntpq); | |
1326 | /* Must succeed, and moved forward */ | |
1327 | if (err != 0 || tok == c || *tok != '\0') | |
1328 | break; | |
1329 | ||
1330 | /* | |
1331 | * sanity; zero queues/threads make no sense, and | |
1332 | * 16K is almost certainly more than anyone will ever | |
1333 | * need and avoids silly numbers like UINT32_MAX | |
1334 | */ | |
1335 | if (nq == 0 || nq >= 16384 || | |
1336 | ntpq == 0 || ntpq >= 16384) | |
1337 | break; | |
1338 | ||
1339 | const zio_taskq_info_t zti = ZTI_P(ntpq, nq); | |
1340 | row[q] = zti; | |
1341 | break; | |
1342 | } | |
1343 | ||
1344 | case ZTI_MODE_SCALE: { | |
1345 | const zio_taskq_info_t zti = ZTI_SCALE; | |
1346 | row[q] = zti; | |
1347 | break; | |
1348 | } | |
1349 | ||
1350 | case ZTI_MODE_SYNC: { | |
1351 | const zio_taskq_info_t zti = ZTI_SYNC; | |
1352 | row[q] = zti; | |
1353 | break; | |
1354 | } | |
1355 | ||
1356 | case ZTI_MODE_NULL: { | |
1357 | /* | |
1358 | * Can only null the high-priority queues; the general- | |
1359 | * purpose ones have to exist. | |
1360 | */ | |
1361 | if (q != ZIO_TASKQ_ISSUE_HIGH && | |
1362 | q != ZIO_TASKQ_INTERRUPT_HIGH) | |
1363 | break; | |
1364 | ||
1365 | const zio_taskq_info_t zti = ZTI_NULL; | |
1366 | row[q] = zti; | |
1367 | break; | |
1368 | } | |
1369 | ||
1370 | default: | |
1371 | break; | |
1372 | } | |
1373 | ||
1374 | /* Ensure we set a mode */ | |
1375 | if (row[q].zti_mode == ZTI_NMODES) | |
1376 | break; | |
1377 | } | |
1378 | ||
1379 | /* Didn't get a full row, fail */ | |
1380 | if (q < ZIO_TASKQ_TYPES) | |
1381 | return (SET_ERROR(EINVAL)); | |
1382 | ||
1383 | /* Eat trailing space */ | |
1384 | if (next != NULL) | |
1385 | while (isspace(*next)) | |
1386 | next++; | |
1387 | ||
1388 | /* If there's anything left over then fail */ | |
1389 | if (next != NULL && *next != '\0') | |
1390 | return (SET_ERROR(EINVAL)); | |
1391 | ||
1392 | /* Success! Copy it into the real config */ | |
1393 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) | |
1394 | zio_taskqs[t][q] = row[q]; | |
1395 | ||
1396 | return (0); | |
1397 | } | |
1398 | ||
1399 | static int | |
1400 | spa_taskq_param_get(zio_type_t t, char *buf, boolean_t add_newline) | |
1401 | { | |
1402 | int pos = 0; | |
1403 | ||
1404 | /* Build paramater string from live config */ | |
1405 | const char *sep = ""; | |
1406 | for (uint_t q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1407 | const zio_taskq_info_t *zti = &zio_taskqs[t][q]; | |
1408 | if (zti->zti_mode == ZTI_MODE_FIXED) | |
1409 | pos += sprintf(&buf[pos], "%s%s,%u,%u", sep, | |
1410 | modes[zti->zti_mode], zti->zti_count, | |
1411 | zti->zti_value); | |
1412 | else | |
1413 | pos += sprintf(&buf[pos], "%s%s", sep, | |
1414 | modes[zti->zti_mode]); | |
1415 | sep = " "; | |
1416 | } | |
1417 | ||
1418 | if (add_newline) | |
1419 | buf[pos++] = '\n'; | |
1420 | buf[pos] = '\0'; | |
1421 | ||
1422 | return (pos); | |
1423 | } | |
1424 | ||
1425 | #ifdef __linux__ | |
1426 | static int | |
1427 | spa_taskq_read_param_set(const char *val, zfs_kernel_param_t *kp) | |
1428 | { | |
1429 | char *cfg = kmem_strdup(val); | |
1430 | int err = spa_taskq_param_set(ZIO_TYPE_READ, cfg); | |
1431 | kmem_free(cfg, strlen(val)+1); | |
1432 | return (-err); | |
1433 | } | |
1434 | static int | |
1435 | spa_taskq_read_param_get(char *buf, zfs_kernel_param_t *kp) | |
1436 | { | |
1437 | return (spa_taskq_param_get(ZIO_TYPE_READ, buf, TRUE)); | |
1438 | } | |
1439 | ||
1440 | static int | |
1441 | spa_taskq_write_param_set(const char *val, zfs_kernel_param_t *kp) | |
1442 | { | |
1443 | char *cfg = kmem_strdup(val); | |
1444 | int err = spa_taskq_param_set(ZIO_TYPE_WRITE, cfg); | |
1445 | kmem_free(cfg, strlen(val)+1); | |
1446 | return (-err); | |
1447 | } | |
1448 | static int | |
1449 | spa_taskq_write_param_get(char *buf, zfs_kernel_param_t *kp) | |
1450 | { | |
1451 | return (spa_taskq_param_get(ZIO_TYPE_WRITE, buf, TRUE)); | |
1452 | } | |
1453 | #else | |
1454 | /* | |
1455 | * On FreeBSD load-time parameters can be set up before malloc() is available, | |
1456 | * so we have to do all the parsing work on the stack. | |
1457 | */ | |
1458 | #define SPA_TASKQ_PARAM_MAX (128) | |
1459 | ||
1460 | static int | |
1461 | spa_taskq_read_param(ZFS_MODULE_PARAM_ARGS) | |
1462 | { | |
1463 | char buf[SPA_TASKQ_PARAM_MAX]; | |
1464 | int err; | |
1465 | ||
1466 | (void) spa_taskq_param_get(ZIO_TYPE_READ, buf, FALSE); | |
1467 | err = sysctl_handle_string(oidp, buf, sizeof (buf), req); | |
1468 | if (err || req->newptr == NULL) | |
1469 | return (err); | |
1470 | return (spa_taskq_param_set(ZIO_TYPE_READ, buf)); | |
1471 | } | |
1472 | ||
1473 | static int | |
1474 | spa_taskq_write_param(ZFS_MODULE_PARAM_ARGS) | |
1475 | { | |
1476 | char buf[SPA_TASKQ_PARAM_MAX]; | |
1477 | int err; | |
1478 | ||
1479 | (void) spa_taskq_param_get(ZIO_TYPE_WRITE, buf, FALSE); | |
1480 | err = sysctl_handle_string(oidp, buf, sizeof (buf), req); | |
1481 | if (err || req->newptr == NULL) | |
1482 | return (err); | |
1483 | return (spa_taskq_param_set(ZIO_TYPE_WRITE, buf)); | |
1484 | } | |
1485 | #endif | |
1486 | #endif /* _KERNEL */ | |
1487 | ||
1488 | /* | |
1489 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
1490 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
1491 | * on the taskq itself. | |
1492 | */ | |
1493 | void | |
1494 | spa_taskq_dispatch(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
1495 | task_func_t *func, zio_t *zio, boolean_t cutinline) | |
1496 | { | |
1497 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1498 | taskq_t *tq; | |
1499 | ||
1500 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1501 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1502 | ||
1503 | /* | |
1504 | * NB: We are assuming that the zio can only be dispatched | |
1505 | * to a single taskq at a time. It would be a grievous error | |
1506 | * to dispatch the zio to another taskq at the same time. | |
1507 | */ | |
1508 | ASSERT(zio); | |
1509 | ASSERT(taskq_empty_ent(&zio->io_tqent)); | |
1510 | ||
1511 | if (tqs->stqs_count == 1) { | |
1512 | tq = tqs->stqs_taskq[0]; | |
1513 | } else if ((t == ZIO_TYPE_WRITE) && (q == ZIO_TASKQ_ISSUE) && | |
1514 | ZIO_HAS_ALLOCATOR(zio)) { | |
1515 | tq = tqs->stqs_taskq[zio->io_allocator % tqs->stqs_count]; | |
1516 | } else { | |
1517 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; | |
1518 | } | |
1519 | ||
1520 | taskq_dispatch_ent(tq, func, zio, cutinline ? TQ_FRONT : 0, | |
1521 | &zio->io_tqent); | |
1522 | } | |
1523 | ||
1524 | static void | |
1525 | spa_create_zio_taskqs(spa_t *spa) | |
1526 | { | |
1527 | for (int t = 0; t < ZIO_TYPES; t++) { | |
1528 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1529 | spa_taskqs_init(spa, t, q); | |
1530 | } | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) | |
1535 | static void | |
1536 | spa_thread(void *arg) | |
1537 | { | |
1538 | psetid_t zio_taskq_psrset_bind = PS_NONE; | |
1539 | callb_cpr_t cprinfo; | |
1540 | ||
1541 | spa_t *spa = arg; | |
1542 | user_t *pu = PTOU(curproc); | |
1543 | ||
1544 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, | |
1545 | spa->spa_name); | |
1546 | ||
1547 | ASSERT(curproc != &p0); | |
1548 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1549 | "zpool-%s", spa->spa_name); | |
1550 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1551 | ||
1552 | /* bind this thread to the requested psrset */ | |
1553 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1554 | pool_lock(); | |
1555 | mutex_enter(&cpu_lock); | |
1556 | mutex_enter(&pidlock); | |
1557 | mutex_enter(&curproc->p_lock); | |
1558 | ||
1559 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1560 | 0, NULL, NULL) == 0) { | |
1561 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1562 | } else { | |
1563 | cmn_err(CE_WARN, | |
1564 | "Couldn't bind process for zfs pool \"%s\" to " | |
1565 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1566 | } | |
1567 | ||
1568 | mutex_exit(&curproc->p_lock); | |
1569 | mutex_exit(&pidlock); | |
1570 | mutex_exit(&cpu_lock); | |
1571 | pool_unlock(); | |
1572 | } | |
1573 | ||
1574 | #ifdef HAVE_SYSDC | |
1575 | if (zio_taskq_sysdc) { | |
1576 | sysdc_thread_enter(curthread, 100, 0); | |
1577 | } | |
1578 | #endif | |
1579 | ||
1580 | spa->spa_proc = curproc; | |
1581 | spa->spa_did = curthread->t_did; | |
1582 | ||
1583 | spa_create_zio_taskqs(spa); | |
1584 | ||
1585 | mutex_enter(&spa->spa_proc_lock); | |
1586 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1587 | ||
1588 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1589 | cv_broadcast(&spa->spa_proc_cv); | |
1590 | ||
1591 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1592 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1593 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1594 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1595 | ||
1596 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1597 | spa->spa_proc_state = SPA_PROC_GONE; | |
1598 | spa->spa_proc = &p0; | |
1599 | cv_broadcast(&spa->spa_proc_cv); | |
1600 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1601 | ||
1602 | mutex_enter(&curproc->p_lock); | |
1603 | lwp_exit(); | |
1604 | } | |
1605 | #endif | |
1606 | ||
1607 | extern metaslab_ops_t *metaslab_allocator(spa_t *spa); | |
1608 | ||
1609 | /* | |
1610 | * Activate an uninitialized pool. | |
1611 | */ | |
1612 | static void | |
1613 | spa_activate(spa_t *spa, spa_mode_t mode) | |
1614 | { | |
1615 | metaslab_ops_t *msp = metaslab_allocator(spa); | |
1616 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
1617 | ||
1618 | spa->spa_state = POOL_STATE_ACTIVE; | |
1619 | spa->spa_mode = mode; | |
1620 | spa->spa_read_spacemaps = spa_mode_readable_spacemaps; | |
1621 | ||
1622 | spa->spa_normal_class = metaslab_class_create(spa, msp); | |
1623 | spa->spa_log_class = metaslab_class_create(spa, msp); | |
1624 | spa->spa_embedded_log_class = metaslab_class_create(spa, msp); | |
1625 | spa->spa_special_class = metaslab_class_create(spa, msp); | |
1626 | spa->spa_dedup_class = metaslab_class_create(spa, msp); | |
1627 | ||
1628 | /* Try to create a covering process */ | |
1629 | mutex_enter(&spa->spa_proc_lock); | |
1630 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1631 | ASSERT(spa->spa_proc == &p0); | |
1632 | spa->spa_did = 0; | |
1633 | ||
1634 | #ifdef HAVE_SPA_THREAD | |
1635 | /* Only create a process if we're going to be around a while. */ | |
1636 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1637 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1638 | NULL, 0) == 0) { | |
1639 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1640 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1641 | cv_wait(&spa->spa_proc_cv, | |
1642 | &spa->spa_proc_lock); | |
1643 | } | |
1644 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1645 | ASSERT(spa->spa_proc != &p0); | |
1646 | ASSERT(spa->spa_did != 0); | |
1647 | } else { | |
1648 | #ifdef _KERNEL | |
1649 | cmn_err(CE_WARN, | |
1650 | "Couldn't create process for zfs pool \"%s\"\n", | |
1651 | spa->spa_name); | |
1652 | #endif | |
1653 | } | |
1654 | } | |
1655 | #endif /* HAVE_SPA_THREAD */ | |
1656 | mutex_exit(&spa->spa_proc_lock); | |
1657 | ||
1658 | /* If we didn't create a process, we need to create our taskqs. */ | |
1659 | if (spa->spa_proc == &p0) { | |
1660 | spa_create_zio_taskqs(spa); | |
1661 | } | |
1662 | ||
1663 | for (size_t i = 0; i < TXG_SIZE; i++) { | |
1664 | spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, | |
1665 | ZIO_FLAG_CANFAIL); | |
1666 | } | |
1667 | ||
1668 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), | |
1669 | offsetof(vdev_t, vdev_config_dirty_node)); | |
1670 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), | |
1671 | offsetof(objset_t, os_evicting_node)); | |
1672 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), | |
1673 | offsetof(vdev_t, vdev_state_dirty_node)); | |
1674 | ||
1675 | txg_list_create(&spa->spa_vdev_txg_list, spa, | |
1676 | offsetof(struct vdev, vdev_txg_node)); | |
1677 | ||
1678 | avl_create(&spa->spa_errlist_scrub, | |
1679 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1680 | offsetof(spa_error_entry_t, se_avl)); | |
1681 | avl_create(&spa->spa_errlist_last, | |
1682 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1683 | offsetof(spa_error_entry_t, se_avl)); | |
1684 | avl_create(&spa->spa_errlist_healed, | |
1685 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1686 | offsetof(spa_error_entry_t, se_avl)); | |
1687 | ||
1688 | spa_activate_os(spa); | |
1689 | ||
1690 | spa_keystore_init(&spa->spa_keystore); | |
1691 | ||
1692 | /* | |
1693 | * This taskq is used to perform zvol-minor-related tasks | |
1694 | * asynchronously. This has several advantages, including easy | |
1695 | * resolution of various deadlocks. | |
1696 | * | |
1697 | * The taskq must be single threaded to ensure tasks are always | |
1698 | * processed in the order in which they were dispatched. | |
1699 | * | |
1700 | * A taskq per pool allows one to keep the pools independent. | |
1701 | * This way if one pool is suspended, it will not impact another. | |
1702 | * | |
1703 | * The preferred location to dispatch a zvol minor task is a sync | |
1704 | * task. In this context, there is easy access to the spa_t and minimal | |
1705 | * error handling is required because the sync task must succeed. | |
1706 | */ | |
1707 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1708 | 1, INT_MAX, 0); | |
1709 | ||
1710 | /* | |
1711 | * The taskq to preload metaslabs. | |
1712 | */ | |
1713 | spa->spa_metaslab_taskq = taskq_create("z_metaslab", | |
1714 | metaslab_preload_pct, maxclsyspri, 1, INT_MAX, | |
1715 | TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
1716 | ||
1717 | /* | |
1718 | * Taskq dedicated to prefetcher threads: this is used to prevent the | |
1719 | * pool traverse code from monopolizing the global (and limited) | |
1720 | * system_taskq by inappropriately scheduling long running tasks on it. | |
1721 | */ | |
1722 | spa->spa_prefetch_taskq = taskq_create("z_prefetch", 100, | |
1723 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
1724 | ||
1725 | /* | |
1726 | * The taskq to upgrade datasets in this pool. Currently used by | |
1727 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA. | |
1728 | */ | |
1729 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", 100, | |
1730 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Opposite of spa_activate(). | |
1735 | */ | |
1736 | static void | |
1737 | spa_deactivate(spa_t *spa) | |
1738 | { | |
1739 | ASSERT(spa->spa_sync_on == B_FALSE); | |
1740 | ASSERT(spa->spa_dsl_pool == NULL); | |
1741 | ASSERT(spa->spa_root_vdev == NULL); | |
1742 | ASSERT(spa->spa_async_zio_root == NULL); | |
1743 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
1744 | ||
1745 | spa_evicting_os_wait(spa); | |
1746 | ||
1747 | if (spa->spa_zvol_taskq) { | |
1748 | taskq_destroy(spa->spa_zvol_taskq); | |
1749 | spa->spa_zvol_taskq = NULL; | |
1750 | } | |
1751 | ||
1752 | if (spa->spa_metaslab_taskq) { | |
1753 | taskq_destroy(spa->spa_metaslab_taskq); | |
1754 | spa->spa_metaslab_taskq = NULL; | |
1755 | } | |
1756 | ||
1757 | if (spa->spa_prefetch_taskq) { | |
1758 | taskq_destroy(spa->spa_prefetch_taskq); | |
1759 | spa->spa_prefetch_taskq = NULL; | |
1760 | } | |
1761 | ||
1762 | if (spa->spa_upgrade_taskq) { | |
1763 | taskq_destroy(spa->spa_upgrade_taskq); | |
1764 | spa->spa_upgrade_taskq = NULL; | |
1765 | } | |
1766 | ||
1767 | txg_list_destroy(&spa->spa_vdev_txg_list); | |
1768 | ||
1769 | list_destroy(&spa->spa_config_dirty_list); | |
1770 | list_destroy(&spa->spa_evicting_os_list); | |
1771 | list_destroy(&spa->spa_state_dirty_list); | |
1772 | ||
1773 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
1774 | ||
1775 | for (int t = 0; t < ZIO_TYPES; t++) { | |
1776 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1777 | spa_taskqs_fini(spa, t, q); | |
1778 | } | |
1779 | } | |
1780 | ||
1781 | for (size_t i = 0; i < TXG_SIZE; i++) { | |
1782 | ASSERT3P(spa->spa_txg_zio[i], !=, NULL); | |
1783 | VERIFY0(zio_wait(spa->spa_txg_zio[i])); | |
1784 | spa->spa_txg_zio[i] = NULL; | |
1785 | } | |
1786 | ||
1787 | metaslab_class_destroy(spa->spa_normal_class); | |
1788 | spa->spa_normal_class = NULL; | |
1789 | ||
1790 | metaslab_class_destroy(spa->spa_log_class); | |
1791 | spa->spa_log_class = NULL; | |
1792 | ||
1793 | metaslab_class_destroy(spa->spa_embedded_log_class); | |
1794 | spa->spa_embedded_log_class = NULL; | |
1795 | ||
1796 | metaslab_class_destroy(spa->spa_special_class); | |
1797 | spa->spa_special_class = NULL; | |
1798 | ||
1799 | metaslab_class_destroy(spa->spa_dedup_class); | |
1800 | spa->spa_dedup_class = NULL; | |
1801 | ||
1802 | /* | |
1803 | * If this was part of an import or the open otherwise failed, we may | |
1804 | * still have errors left in the queues. Empty them just in case. | |
1805 | */ | |
1806 | spa_errlog_drain(spa); | |
1807 | avl_destroy(&spa->spa_errlist_scrub); | |
1808 | avl_destroy(&spa->spa_errlist_last); | |
1809 | avl_destroy(&spa->spa_errlist_healed); | |
1810 | ||
1811 | spa_keystore_fini(&spa->spa_keystore); | |
1812 | ||
1813 | spa->spa_state = POOL_STATE_UNINITIALIZED; | |
1814 | ||
1815 | mutex_enter(&spa->spa_proc_lock); | |
1816 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1817 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1818 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1819 | cv_broadcast(&spa->spa_proc_cv); | |
1820 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1821 | ASSERT(spa->spa_proc != &p0); | |
1822 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1823 | } | |
1824 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1825 | spa->spa_proc_state = SPA_PROC_NONE; | |
1826 | } | |
1827 | ASSERT(spa->spa_proc == &p0); | |
1828 | mutex_exit(&spa->spa_proc_lock); | |
1829 | ||
1830 | /* | |
1831 | * We want to make sure spa_thread() has actually exited the ZFS | |
1832 | * module, so that the module can't be unloaded out from underneath | |
1833 | * it. | |
1834 | */ | |
1835 | if (spa->spa_did != 0) { | |
1836 | thread_join(spa->spa_did); | |
1837 | spa->spa_did = 0; | |
1838 | } | |
1839 | ||
1840 | spa_deactivate_os(spa); | |
1841 | ||
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1846 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1847 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1848 | * All vdev validation is done by the vdev_alloc() routine. | |
1849 | */ | |
1850 | int | |
1851 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
1852 | uint_t id, int atype) | |
1853 | { | |
1854 | nvlist_t **child; | |
1855 | uint_t children; | |
1856 | int error; | |
1857 | ||
1858 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1859 | return (error); | |
1860 | ||
1861 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1862 | return (0); | |
1863 | ||
1864 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
1865 | &child, &children); | |
1866 | ||
1867 | if (error == ENOENT) | |
1868 | return (0); | |
1869 | ||
1870 | if (error) { | |
1871 | vdev_free(*vdp); | |
1872 | *vdp = NULL; | |
1873 | return (SET_ERROR(EINVAL)); | |
1874 | } | |
1875 | ||
1876 | for (int c = 0; c < children; c++) { | |
1877 | vdev_t *vd; | |
1878 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1879 | atype)) != 0) { | |
1880 | vdev_free(*vdp); | |
1881 | *vdp = NULL; | |
1882 | return (error); | |
1883 | } | |
1884 | } | |
1885 | ||
1886 | ASSERT(*vdp != NULL); | |
1887 | ||
1888 | return (0); | |
1889 | } | |
1890 | ||
1891 | static boolean_t | |
1892 | spa_should_flush_logs_on_unload(spa_t *spa) | |
1893 | { | |
1894 | if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) | |
1895 | return (B_FALSE); | |
1896 | ||
1897 | if (!spa_writeable(spa)) | |
1898 | return (B_FALSE); | |
1899 | ||
1900 | if (!spa->spa_sync_on) | |
1901 | return (B_FALSE); | |
1902 | ||
1903 | if (spa_state(spa) != POOL_STATE_EXPORTED) | |
1904 | return (B_FALSE); | |
1905 | ||
1906 | if (zfs_keep_log_spacemaps_at_export) | |
1907 | return (B_FALSE); | |
1908 | ||
1909 | return (B_TRUE); | |
1910 | } | |
1911 | ||
1912 | /* | |
1913 | * Opens a transaction that will set the flag that will instruct | |
1914 | * spa_sync to attempt to flush all the metaslabs for that txg. | |
1915 | */ | |
1916 | static void | |
1917 | spa_unload_log_sm_flush_all(spa_t *spa) | |
1918 | { | |
1919 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
1920 | VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); | |
1921 | ||
1922 | ASSERT3U(spa->spa_log_flushall_txg, ==, 0); | |
1923 | spa->spa_log_flushall_txg = dmu_tx_get_txg(tx); | |
1924 | ||
1925 | dmu_tx_commit(tx); | |
1926 | txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg); | |
1927 | } | |
1928 | ||
1929 | static void | |
1930 | spa_unload_log_sm_metadata(spa_t *spa) | |
1931 | { | |
1932 | void *cookie = NULL; | |
1933 | spa_log_sm_t *sls; | |
1934 | log_summary_entry_t *e; | |
1935 | ||
1936 | while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg, | |
1937 | &cookie)) != NULL) { | |
1938 | VERIFY0(sls->sls_mscount); | |
1939 | kmem_free(sls, sizeof (spa_log_sm_t)); | |
1940 | } | |
1941 | ||
1942 | while ((e = list_remove_head(&spa->spa_log_summary)) != NULL) { | |
1943 | VERIFY0(e->lse_mscount); | |
1944 | kmem_free(e, sizeof (log_summary_entry_t)); | |
1945 | } | |
1946 | ||
1947 | spa->spa_unflushed_stats.sus_nblocks = 0; | |
1948 | spa->spa_unflushed_stats.sus_memused = 0; | |
1949 | spa->spa_unflushed_stats.sus_blocklimit = 0; | |
1950 | } | |
1951 | ||
1952 | static void | |
1953 | spa_destroy_aux_threads(spa_t *spa) | |
1954 | { | |
1955 | if (spa->spa_condense_zthr != NULL) { | |
1956 | zthr_destroy(spa->spa_condense_zthr); | |
1957 | spa->spa_condense_zthr = NULL; | |
1958 | } | |
1959 | if (spa->spa_checkpoint_discard_zthr != NULL) { | |
1960 | zthr_destroy(spa->spa_checkpoint_discard_zthr); | |
1961 | spa->spa_checkpoint_discard_zthr = NULL; | |
1962 | } | |
1963 | if (spa->spa_livelist_delete_zthr != NULL) { | |
1964 | zthr_destroy(spa->spa_livelist_delete_zthr); | |
1965 | spa->spa_livelist_delete_zthr = NULL; | |
1966 | } | |
1967 | if (spa->spa_livelist_condense_zthr != NULL) { | |
1968 | zthr_destroy(spa->spa_livelist_condense_zthr); | |
1969 | spa->spa_livelist_condense_zthr = NULL; | |
1970 | } | |
1971 | if (spa->spa_raidz_expand_zthr != NULL) { | |
1972 | zthr_destroy(spa->spa_raidz_expand_zthr); | |
1973 | spa->spa_raidz_expand_zthr = NULL; | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | /* | |
1978 | * Opposite of spa_load(). | |
1979 | */ | |
1980 | static void | |
1981 | spa_unload(spa_t *spa) | |
1982 | { | |
1983 | ASSERT(MUTEX_HELD(&spa_namespace_lock) || | |
1984 | spa->spa_export_thread == curthread); | |
1985 | ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED); | |
1986 | ||
1987 | spa_import_progress_remove(spa_guid(spa)); | |
1988 | spa_load_note(spa, "UNLOADING"); | |
1989 | ||
1990 | spa_wake_waiters(spa); | |
1991 | ||
1992 | /* | |
1993 | * If we have set the spa_final_txg, we have already performed the | |
1994 | * tasks below in spa_export_common(). We should not redo it here since | |
1995 | * we delay the final TXGs beyond what spa_final_txg is set at. | |
1996 | */ | |
1997 | if (spa->spa_final_txg == UINT64_MAX) { | |
1998 | /* | |
1999 | * If the log space map feature is enabled and the pool is | |
2000 | * getting exported (but not destroyed), we want to spend some | |
2001 | * time flushing as many metaslabs as we can in an attempt to | |
2002 | * destroy log space maps and save import time. | |
2003 | */ | |
2004 | if (spa_should_flush_logs_on_unload(spa)) | |
2005 | spa_unload_log_sm_flush_all(spa); | |
2006 | ||
2007 | /* | |
2008 | * Stop async tasks. | |
2009 | */ | |
2010 | spa_async_suspend(spa); | |
2011 | ||
2012 | if (spa->spa_root_vdev) { | |
2013 | vdev_t *root_vdev = spa->spa_root_vdev; | |
2014 | vdev_initialize_stop_all(root_vdev, | |
2015 | VDEV_INITIALIZE_ACTIVE); | |
2016 | vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE); | |
2017 | vdev_autotrim_stop_all(spa); | |
2018 | vdev_rebuild_stop_all(spa); | |
2019 | } | |
2020 | } | |
2021 | ||
2022 | /* | |
2023 | * Stop syncing. | |
2024 | */ | |
2025 | if (spa->spa_sync_on) { | |
2026 | txg_sync_stop(spa->spa_dsl_pool); | |
2027 | spa->spa_sync_on = B_FALSE; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * This ensures that there is no async metaslab prefetching | |
2032 | * while we attempt to unload the spa. | |
2033 | */ | |
2034 | taskq_wait(spa->spa_metaslab_taskq); | |
2035 | ||
2036 | if (spa->spa_mmp.mmp_thread) | |
2037 | mmp_thread_stop(spa); | |
2038 | ||
2039 | /* | |
2040 | * Wait for any outstanding async I/O to complete. | |
2041 | */ | |
2042 | if (spa->spa_async_zio_root != NULL) { | |
2043 | for (int i = 0; i < max_ncpus; i++) | |
2044 | (void) zio_wait(spa->spa_async_zio_root[i]); | |
2045 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
2046 | spa->spa_async_zio_root = NULL; | |
2047 | } | |
2048 | ||
2049 | if (spa->spa_vdev_removal != NULL) { | |
2050 | spa_vdev_removal_destroy(spa->spa_vdev_removal); | |
2051 | spa->spa_vdev_removal = NULL; | |
2052 | } | |
2053 | ||
2054 | spa_destroy_aux_threads(spa); | |
2055 | ||
2056 | spa_condense_fini(spa); | |
2057 | ||
2058 | bpobj_close(&spa->spa_deferred_bpobj); | |
2059 | ||
2060 | spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); | |
2061 | ||
2062 | /* | |
2063 | * Close all vdevs. | |
2064 | */ | |
2065 | if (spa->spa_root_vdev) | |
2066 | vdev_free(spa->spa_root_vdev); | |
2067 | ASSERT(spa->spa_root_vdev == NULL); | |
2068 | ||
2069 | /* | |
2070 | * Close the dsl pool. | |
2071 | */ | |
2072 | if (spa->spa_dsl_pool) { | |
2073 | dsl_pool_close(spa->spa_dsl_pool); | |
2074 | spa->spa_dsl_pool = NULL; | |
2075 | spa->spa_meta_objset = NULL; | |
2076 | } | |
2077 | ||
2078 | ddt_unload(spa); | |
2079 | brt_unload(spa); | |
2080 | spa_unload_log_sm_metadata(spa); | |
2081 | ||
2082 | /* | |
2083 | * Drop and purge level 2 cache | |
2084 | */ | |
2085 | spa_l2cache_drop(spa); | |
2086 | ||
2087 | if (spa->spa_spares.sav_vdevs) { | |
2088 | for (int i = 0; i < spa->spa_spares.sav_count; i++) | |
2089 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
2090 | kmem_free(spa->spa_spares.sav_vdevs, | |
2091 | spa->spa_spares.sav_count * sizeof (void *)); | |
2092 | spa->spa_spares.sav_vdevs = NULL; | |
2093 | } | |
2094 | if (spa->spa_spares.sav_config) { | |
2095 | nvlist_free(spa->spa_spares.sav_config); | |
2096 | spa->spa_spares.sav_config = NULL; | |
2097 | } | |
2098 | spa->spa_spares.sav_count = 0; | |
2099 | ||
2100 | if (spa->spa_l2cache.sav_vdevs) { | |
2101 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
2102 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
2103 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
2104 | } | |
2105 | kmem_free(spa->spa_l2cache.sav_vdevs, | |
2106 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
2107 | spa->spa_l2cache.sav_vdevs = NULL; | |
2108 | } | |
2109 | if (spa->spa_l2cache.sav_config) { | |
2110 | nvlist_free(spa->spa_l2cache.sav_config); | |
2111 | spa->spa_l2cache.sav_config = NULL; | |
2112 | } | |
2113 | spa->spa_l2cache.sav_count = 0; | |
2114 | ||
2115 | spa->spa_async_suspended = 0; | |
2116 | ||
2117 | spa->spa_indirect_vdevs_loaded = B_FALSE; | |
2118 | ||
2119 | if (spa->spa_comment != NULL) { | |
2120 | spa_strfree(spa->spa_comment); | |
2121 | spa->spa_comment = NULL; | |
2122 | } | |
2123 | if (spa->spa_compatibility != NULL) { | |
2124 | spa_strfree(spa->spa_compatibility); | |
2125 | spa->spa_compatibility = NULL; | |
2126 | } | |
2127 | ||
2128 | spa->spa_raidz_expand = NULL; | |
2129 | ||
2130 | spa_config_exit(spa, SCL_ALL, spa); | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * Load (or re-load) the current list of vdevs describing the active spares for | |
2135 | * this pool. When this is called, we have some form of basic information in | |
2136 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
2137 | * then re-generate a more complete list including status information. | |
2138 | */ | |
2139 | void | |
2140 | spa_load_spares(spa_t *spa) | |
2141 | { | |
2142 | nvlist_t **spares; | |
2143 | uint_t nspares; | |
2144 | int i; | |
2145 | vdev_t *vd, *tvd; | |
2146 | ||
2147 | #ifndef _KERNEL | |
2148 | /* | |
2149 | * zdb opens both the current state of the pool and the | |
2150 | * checkpointed state (if present), with a different spa_t. | |
2151 | * | |
2152 | * As spare vdevs are shared among open pools, we skip loading | |
2153 | * them when we load the checkpointed state of the pool. | |
2154 | */ | |
2155 | if (!spa_writeable(spa)) | |
2156 | return; | |
2157 | #endif | |
2158 | ||
2159 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
2160 | ||
2161 | /* | |
2162 | * First, close and free any existing spare vdevs. | |
2163 | */ | |
2164 | if (spa->spa_spares.sav_vdevs) { | |
2165 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
2166 | vd = spa->spa_spares.sav_vdevs[i]; | |
2167 | ||
2168 | /* Undo the call to spa_activate() below */ | |
2169 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
2170 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
2171 | spa_spare_remove(tvd); | |
2172 | vdev_close(vd); | |
2173 | vdev_free(vd); | |
2174 | } | |
2175 | ||
2176 | kmem_free(spa->spa_spares.sav_vdevs, | |
2177 | spa->spa_spares.sav_count * sizeof (void *)); | |
2178 | } | |
2179 | ||
2180 | if (spa->spa_spares.sav_config == NULL) | |
2181 | nspares = 0; | |
2182 | else | |
2183 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
2184 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
2185 | ||
2186 | spa->spa_spares.sav_count = (int)nspares; | |
2187 | spa->spa_spares.sav_vdevs = NULL; | |
2188 | ||
2189 | if (nspares == 0) | |
2190 | return; | |
2191 | ||
2192 | /* | |
2193 | * Construct the array of vdevs, opening them to get status in the | |
2194 | * process. For each spare, there is potentially two different vdev_t | |
2195 | * structures associated with it: one in the list of spares (used only | |
2196 | * for basic validation purposes) and one in the active vdev | |
2197 | * configuration (if it's spared in). During this phase we open and | |
2198 | * validate each vdev on the spare list. If the vdev also exists in the | |
2199 | * active configuration, then we also mark this vdev as an active spare. | |
2200 | */ | |
2201 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), | |
2202 | KM_SLEEP); | |
2203 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
2204 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
2205 | VDEV_ALLOC_SPARE) == 0); | |
2206 | ASSERT(vd != NULL); | |
2207 | ||
2208 | spa->spa_spares.sav_vdevs[i] = vd; | |
2209 | ||
2210 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
2211 | B_FALSE)) != NULL) { | |
2212 | if (!tvd->vdev_isspare) | |
2213 | spa_spare_add(tvd); | |
2214 | ||
2215 | /* | |
2216 | * We only mark the spare active if we were successfully | |
2217 | * able to load the vdev. Otherwise, importing a pool | |
2218 | * with a bad active spare would result in strange | |
2219 | * behavior, because multiple pool would think the spare | |
2220 | * is actively in use. | |
2221 | * | |
2222 | * There is a vulnerability here to an equally bizarre | |
2223 | * circumstance, where a dead active spare is later | |
2224 | * brought back to life (onlined or otherwise). Given | |
2225 | * the rarity of this scenario, and the extra complexity | |
2226 | * it adds, we ignore the possibility. | |
2227 | */ | |
2228 | if (!vdev_is_dead(tvd)) | |
2229 | spa_spare_activate(tvd); | |
2230 | } | |
2231 | ||
2232 | vd->vdev_top = vd; | |
2233 | vd->vdev_aux = &spa->spa_spares; | |
2234 | ||
2235 | if (vdev_open(vd) != 0) | |
2236 | continue; | |
2237 | ||
2238 | if (vdev_validate_aux(vd) == 0) | |
2239 | spa_spare_add(vd); | |
2240 | } | |
2241 | ||
2242 | /* | |
2243 | * Recompute the stashed list of spares, with status information | |
2244 | * this time. | |
2245 | */ | |
2246 | fnvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES); | |
2247 | ||
2248 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
2249 | KM_SLEEP); | |
2250 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
2251 | spares[i] = vdev_config_generate(spa, | |
2252 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); | |
2253 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
2254 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, | |
2255 | spa->spa_spares.sav_count); | |
2256 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
2257 | nvlist_free(spares[i]); | |
2258 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
2259 | } | |
2260 | ||
2261 | /* | |
2262 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
2263 | * this pool. When this is called, we have some form of basic information in | |
2264 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
2265 | * then re-generate a more complete list including status information. | |
2266 | * Devices which are already active have their details maintained, and are | |
2267 | * not re-opened. | |
2268 | */ | |
2269 | void | |
2270 | spa_load_l2cache(spa_t *spa) | |
2271 | { | |
2272 | nvlist_t **l2cache = NULL; | |
2273 | uint_t nl2cache; | |
2274 | int i, j, oldnvdevs; | |
2275 | uint64_t guid; | |
2276 | vdev_t *vd, **oldvdevs, **newvdevs; | |
2277 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
2278 | ||
2279 | #ifndef _KERNEL | |
2280 | /* | |
2281 | * zdb opens both the current state of the pool and the | |
2282 | * checkpointed state (if present), with a different spa_t. | |
2283 | * | |
2284 | * As L2 caches are part of the ARC which is shared among open | |
2285 | * pools, we skip loading them when we load the checkpointed | |
2286 | * state of the pool. | |
2287 | */ | |
2288 | if (!spa_writeable(spa)) | |
2289 | return; | |
2290 | #endif | |
2291 | ||
2292 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
2293 | ||
2294 | oldvdevs = sav->sav_vdevs; | |
2295 | oldnvdevs = sav->sav_count; | |
2296 | sav->sav_vdevs = NULL; | |
2297 | sav->sav_count = 0; | |
2298 | ||
2299 | if (sav->sav_config == NULL) { | |
2300 | nl2cache = 0; | |
2301 | newvdevs = NULL; | |
2302 | goto out; | |
2303 | } | |
2304 | ||
2305 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, | |
2306 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
2307 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); | |
2308 | ||
2309 | /* | |
2310 | * Process new nvlist of vdevs. | |
2311 | */ | |
2312 | for (i = 0; i < nl2cache; i++) { | |
2313 | guid = fnvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID); | |
2314 | ||
2315 | newvdevs[i] = NULL; | |
2316 | for (j = 0; j < oldnvdevs; j++) { | |
2317 | vd = oldvdevs[j]; | |
2318 | if (vd != NULL && guid == vd->vdev_guid) { | |
2319 | /* | |
2320 | * Retain previous vdev for add/remove ops. | |
2321 | */ | |
2322 | newvdevs[i] = vd; | |
2323 | oldvdevs[j] = NULL; | |
2324 | break; | |
2325 | } | |
2326 | } | |
2327 | ||
2328 | if (newvdevs[i] == NULL) { | |
2329 | /* | |
2330 | * Create new vdev | |
2331 | */ | |
2332 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
2333 | VDEV_ALLOC_L2CACHE) == 0); | |
2334 | ASSERT(vd != NULL); | |
2335 | newvdevs[i] = vd; | |
2336 | ||
2337 | /* | |
2338 | * Commit this vdev as an l2cache device, | |
2339 | * even if it fails to open. | |
2340 | */ | |
2341 | spa_l2cache_add(vd); | |
2342 | ||
2343 | vd->vdev_top = vd; | |
2344 | vd->vdev_aux = sav; | |
2345 | ||
2346 | spa_l2cache_activate(vd); | |
2347 | ||
2348 | if (vdev_open(vd) != 0) | |
2349 | continue; | |
2350 | ||
2351 | (void) vdev_validate_aux(vd); | |
2352 | ||
2353 | if (!vdev_is_dead(vd)) | |
2354 | l2arc_add_vdev(spa, vd); | |
2355 | ||
2356 | /* | |
2357 | * Upon cache device addition to a pool or pool | |
2358 | * creation with a cache device or if the header | |
2359 | * of the device is invalid we issue an async | |
2360 | * TRIM command for the whole device which will | |
2361 | * execute if l2arc_trim_ahead > 0. | |
2362 | */ | |
2363 | spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); | |
2364 | } | |
2365 | } | |
2366 | ||
2367 | sav->sav_vdevs = newvdevs; | |
2368 | sav->sav_count = (int)nl2cache; | |
2369 | ||
2370 | /* | |
2371 | * Recompute the stashed list of l2cache devices, with status | |
2372 | * information this time. | |
2373 | */ | |
2374 | fnvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE); | |
2375 | ||
2376 | if (sav->sav_count > 0) | |
2377 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
2378 | KM_SLEEP); | |
2379 | for (i = 0; i < sav->sav_count; i++) | |
2380 | l2cache[i] = vdev_config_generate(spa, | |
2381 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
2382 | fnvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, | |
2383 | (const nvlist_t * const *)l2cache, sav->sav_count); | |
2384 | ||
2385 | out: | |
2386 | /* | |
2387 | * Purge vdevs that were dropped | |
2388 | */ | |
2389 | if (oldvdevs) { | |
2390 | for (i = 0; i < oldnvdevs; i++) { | |
2391 | uint64_t pool; | |
2392 | ||
2393 | vd = oldvdevs[i]; | |
2394 | if (vd != NULL) { | |
2395 | ASSERT(vd->vdev_isl2cache); | |
2396 | ||
2397 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
2398 | pool != 0ULL && l2arc_vdev_present(vd)) | |
2399 | l2arc_remove_vdev(vd); | |
2400 | vdev_clear_stats(vd); | |
2401 | vdev_free(vd); | |
2402 | } | |
2403 | } | |
2404 | ||
2405 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); | |
2406 | } | |
2407 | ||
2408 | for (i = 0; i < sav->sav_count; i++) | |
2409 | nvlist_free(l2cache[i]); | |
2410 | if (sav->sav_count) | |
2411 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
2412 | } | |
2413 | ||
2414 | static int | |
2415 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
2416 | { | |
2417 | dmu_buf_t *db; | |
2418 | char *packed = NULL; | |
2419 | size_t nvsize = 0; | |
2420 | int error; | |
2421 | *value = NULL; | |
2422 | ||
2423 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); | |
2424 | if (error) | |
2425 | return (error); | |
2426 | ||
2427 | nvsize = *(uint64_t *)db->db_data; | |
2428 | dmu_buf_rele(db, FTAG); | |
2429 | ||
2430 | packed = vmem_alloc(nvsize, KM_SLEEP); | |
2431 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, | |
2432 | DMU_READ_PREFETCH); | |
2433 | if (error == 0) | |
2434 | error = nvlist_unpack(packed, nvsize, value, 0); | |
2435 | vmem_free(packed, nvsize); | |
2436 | ||
2437 | return (error); | |
2438 | } | |
2439 | ||
2440 | /* | |
2441 | * Concrete top-level vdevs that are not missing and are not logs. At every | |
2442 | * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. | |
2443 | */ | |
2444 | static uint64_t | |
2445 | spa_healthy_core_tvds(spa_t *spa) | |
2446 | { | |
2447 | vdev_t *rvd = spa->spa_root_vdev; | |
2448 | uint64_t tvds = 0; | |
2449 | ||
2450 | for (uint64_t i = 0; i < rvd->vdev_children; i++) { | |
2451 | vdev_t *vd = rvd->vdev_child[i]; | |
2452 | if (vd->vdev_islog) | |
2453 | continue; | |
2454 | if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) | |
2455 | tvds++; | |
2456 | } | |
2457 | ||
2458 | return (tvds); | |
2459 | } | |
2460 | ||
2461 | /* | |
2462 | * Checks to see if the given vdev could not be opened, in which case we post a | |
2463 | * sysevent to notify the autoreplace code that the device has been removed. | |
2464 | */ | |
2465 | static void | |
2466 | spa_check_removed(vdev_t *vd) | |
2467 | { | |
2468 | for (uint64_t c = 0; c < vd->vdev_children; c++) | |
2469 | spa_check_removed(vd->vdev_child[c]); | |
2470 | ||
2471 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && | |
2472 | vdev_is_concrete(vd)) { | |
2473 | zfs_post_autoreplace(vd->vdev_spa, vd); | |
2474 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); | |
2475 | } | |
2476 | } | |
2477 | ||
2478 | static int | |
2479 | spa_check_for_missing_logs(spa_t *spa) | |
2480 | { | |
2481 | vdev_t *rvd = spa->spa_root_vdev; | |
2482 | ||
2483 | /* | |
2484 | * If we're doing a normal import, then build up any additional | |
2485 | * diagnostic information about missing log devices. | |
2486 | * We'll pass this up to the user for further processing. | |
2487 | */ | |
2488 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { | |
2489 | nvlist_t **child, *nv; | |
2490 | uint64_t idx = 0; | |
2491 | ||
2492 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), | |
2493 | KM_SLEEP); | |
2494 | nv = fnvlist_alloc(); | |
2495 | ||
2496 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
2497 | vdev_t *tvd = rvd->vdev_child[c]; | |
2498 | ||
2499 | /* | |
2500 | * We consider a device as missing only if it failed | |
2501 | * to open (i.e. offline or faulted is not considered | |
2502 | * as missing). | |
2503 | */ | |
2504 | if (tvd->vdev_islog && | |
2505 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
2506 | child[idx++] = vdev_config_generate(spa, tvd, | |
2507 | B_FALSE, VDEV_CONFIG_MISSING); | |
2508 | } | |
2509 | } | |
2510 | ||
2511 | if (idx > 0) { | |
2512 | fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
2513 | (const nvlist_t * const *)child, idx); | |
2514 | fnvlist_add_nvlist(spa->spa_load_info, | |
2515 | ZPOOL_CONFIG_MISSING_DEVICES, nv); | |
2516 | ||
2517 | for (uint64_t i = 0; i < idx; i++) | |
2518 | nvlist_free(child[i]); | |
2519 | } | |
2520 | nvlist_free(nv); | |
2521 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
2522 | ||
2523 | if (idx > 0) { | |
2524 | spa_load_failed(spa, "some log devices are missing"); | |
2525 | vdev_dbgmsg_print_tree(rvd, 2); | |
2526 | return (SET_ERROR(ENXIO)); | |
2527 | } | |
2528 | } else { | |
2529 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
2530 | vdev_t *tvd = rvd->vdev_child[c]; | |
2531 | ||
2532 | if (tvd->vdev_islog && | |
2533 | tvd->vdev_state == VDEV_STATE_CANT_OPEN) { | |
2534 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
2535 | spa_load_note(spa, "some log devices are " | |
2536 | "missing, ZIL is dropped."); | |
2537 | vdev_dbgmsg_print_tree(rvd, 2); | |
2538 | break; | |
2539 | } | |
2540 | } | |
2541 | } | |
2542 | ||
2543 | return (0); | |
2544 | } | |
2545 | ||
2546 | /* | |
2547 | * Check for missing log devices | |
2548 | */ | |
2549 | static boolean_t | |
2550 | spa_check_logs(spa_t *spa) | |
2551 | { | |
2552 | boolean_t rv = B_FALSE; | |
2553 | dsl_pool_t *dp = spa_get_dsl(spa); | |
2554 | ||
2555 | switch (spa->spa_log_state) { | |
2556 | default: | |
2557 | break; | |
2558 | case SPA_LOG_MISSING: | |
2559 | /* need to recheck in case slog has been restored */ | |
2560 | case SPA_LOG_UNKNOWN: | |
2561 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
2562 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
2563 | if (rv) | |
2564 | spa_set_log_state(spa, SPA_LOG_MISSING); | |
2565 | break; | |
2566 | } | |
2567 | return (rv); | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * Passivate any log vdevs (note, does not apply to embedded log metaslabs). | |
2572 | */ | |
2573 | static boolean_t | |
2574 | spa_passivate_log(spa_t *spa) | |
2575 | { | |
2576 | vdev_t *rvd = spa->spa_root_vdev; | |
2577 | boolean_t slog_found = B_FALSE; | |
2578 | ||
2579 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
2580 | ||
2581 | for (int c = 0; c < rvd->vdev_children; c++) { | |
2582 | vdev_t *tvd = rvd->vdev_child[c]; | |
2583 | ||
2584 | if (tvd->vdev_islog) { | |
2585 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); | |
2586 | metaslab_group_passivate(tvd->vdev_mg); | |
2587 | slog_found = B_TRUE; | |
2588 | } | |
2589 | } | |
2590 | ||
2591 | return (slog_found); | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * Activate any log vdevs (note, does not apply to embedded log metaslabs). | |
2596 | */ | |
2597 | static void | |
2598 | spa_activate_log(spa_t *spa) | |
2599 | { | |
2600 | vdev_t *rvd = spa->spa_root_vdev; | |
2601 | ||
2602 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
2603 | ||
2604 | for (int c = 0; c < rvd->vdev_children; c++) { | |
2605 | vdev_t *tvd = rvd->vdev_child[c]; | |
2606 | ||
2607 | if (tvd->vdev_islog) { | |
2608 | ASSERT3P(tvd->vdev_log_mg, ==, NULL); | |
2609 | metaslab_group_activate(tvd->vdev_mg); | |
2610 | } | |
2611 | } | |
2612 | } | |
2613 | ||
2614 | int | |
2615 | spa_reset_logs(spa_t *spa) | |
2616 | { | |
2617 | int error; | |
2618 | ||
2619 | error = dmu_objset_find(spa_name(spa), zil_reset, | |
2620 | NULL, DS_FIND_CHILDREN); | |
2621 | if (error == 0) { | |
2622 | /* | |
2623 | * We successfully offlined the log device, sync out the | |
2624 | * current txg so that the "stubby" block can be removed | |
2625 | * by zil_sync(). | |
2626 | */ | |
2627 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
2628 | } | |
2629 | return (error); | |
2630 | } | |
2631 | ||
2632 | static void | |
2633 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
2634 | { | |
2635 | for (int i = 0; i < sav->sav_count; i++) | |
2636 | spa_check_removed(sav->sav_vdevs[i]); | |
2637 | } | |
2638 | ||
2639 | void | |
2640 | spa_claim_notify(zio_t *zio) | |
2641 | { | |
2642 | spa_t *spa = zio->io_spa; | |
2643 | ||
2644 | if (zio->io_error) | |
2645 | return; | |
2646 | ||
2647 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ | |
2648 | if (spa->spa_claim_max_txg < BP_GET_LOGICAL_BIRTH(zio->io_bp)) | |
2649 | spa->spa_claim_max_txg = BP_GET_LOGICAL_BIRTH(zio->io_bp); | |
2650 | mutex_exit(&spa->spa_props_lock); | |
2651 | } | |
2652 | ||
2653 | typedef struct spa_load_error { | |
2654 | boolean_t sle_verify_data; | |
2655 | uint64_t sle_meta_count; | |
2656 | uint64_t sle_data_count; | |
2657 | } spa_load_error_t; | |
2658 | ||
2659 | static void | |
2660 | spa_load_verify_done(zio_t *zio) | |
2661 | { | |
2662 | blkptr_t *bp = zio->io_bp; | |
2663 | spa_load_error_t *sle = zio->io_private; | |
2664 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
2665 | int error = zio->io_error; | |
2666 | spa_t *spa = zio->io_spa; | |
2667 | ||
2668 | abd_free(zio->io_abd); | |
2669 | if (error) { | |
2670 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && | |
2671 | type != DMU_OT_INTENT_LOG) | |
2672 | atomic_inc_64(&sle->sle_meta_count); | |
2673 | else | |
2674 | atomic_inc_64(&sle->sle_data_count); | |
2675 | } | |
2676 | ||
2677 | mutex_enter(&spa->spa_scrub_lock); | |
2678 | spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp); | |
2679 | cv_broadcast(&spa->spa_scrub_io_cv); | |
2680 | mutex_exit(&spa->spa_scrub_lock); | |
2681 | } | |
2682 | ||
2683 | /* | |
2684 | * Maximum number of inflight bytes is the log2 fraction of the arc size. | |
2685 | * By default, we set it to 1/16th of the arc. | |
2686 | */ | |
2687 | static uint_t spa_load_verify_shift = 4; | |
2688 | static int spa_load_verify_metadata = B_TRUE; | |
2689 | static int spa_load_verify_data = B_TRUE; | |
2690 | ||
2691 | static int | |
2692 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
2693 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
2694 | { | |
2695 | zio_t *rio = arg; | |
2696 | spa_load_error_t *sle = rio->io_private; | |
2697 | ||
2698 | (void) zilog, (void) dnp; | |
2699 | ||
2700 | /* | |
2701 | * Note: normally this routine will not be called if | |
2702 | * spa_load_verify_metadata is not set. However, it may be useful | |
2703 | * to manually set the flag after the traversal has begun. | |
2704 | */ | |
2705 | if (!spa_load_verify_metadata) | |
2706 | return (0); | |
2707 | ||
2708 | /* | |
2709 | * Sanity check the block pointer in order to detect obvious damage | |
2710 | * before using the contents in subsequent checks or in zio_read(). | |
2711 | * When damaged consider it to be a metadata error since we cannot | |
2712 | * trust the BP_GET_TYPE and BP_GET_LEVEL values. | |
2713 | */ | |
2714 | if (!zfs_blkptr_verify(spa, bp, BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) { | |
2715 | atomic_inc_64(&sle->sle_meta_count); | |
2716 | return (0); | |
2717 | } | |
2718 | ||
2719 | if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) || | |
2720 | BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) | |
2721 | return (0); | |
2722 | ||
2723 | if (!BP_IS_METADATA(bp) && | |
2724 | (!spa_load_verify_data || !sle->sle_verify_data)) | |
2725 | return (0); | |
2726 | ||
2727 | uint64_t maxinflight_bytes = | |
2728 | arc_target_bytes() >> spa_load_verify_shift; | |
2729 | size_t size = BP_GET_PSIZE(bp); | |
2730 | ||
2731 | mutex_enter(&spa->spa_scrub_lock); | |
2732 | while (spa->spa_load_verify_bytes >= maxinflight_bytes) | |
2733 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2734 | spa->spa_load_verify_bytes += size; | |
2735 | mutex_exit(&spa->spa_scrub_lock); | |
2736 | ||
2737 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, | |
2738 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, | |
2739 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2740 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
2741 | return (0); | |
2742 | } | |
2743 | ||
2744 | static int | |
2745 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) | |
2746 | { | |
2747 | (void) dp, (void) arg; | |
2748 | ||
2749 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) | |
2750 | return (SET_ERROR(ENAMETOOLONG)); | |
2751 | ||
2752 | return (0); | |
2753 | } | |
2754 | ||
2755 | static int | |
2756 | spa_load_verify(spa_t *spa) | |
2757 | { | |
2758 | zio_t *rio; | |
2759 | spa_load_error_t sle = { 0 }; | |
2760 | zpool_load_policy_t policy; | |
2761 | boolean_t verify_ok = B_FALSE; | |
2762 | int error = 0; | |
2763 | ||
2764 | zpool_get_load_policy(spa->spa_config, &policy); | |
2765 | ||
2766 | if (policy.zlp_rewind & ZPOOL_NEVER_REWIND || | |
2767 | policy.zlp_maxmeta == UINT64_MAX) | |
2768 | return (0); | |
2769 | ||
2770 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); | |
2771 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2772 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2773 | DS_FIND_CHILDREN); | |
2774 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2775 | if (error != 0) | |
2776 | return (error); | |
2777 | ||
2778 | /* | |
2779 | * Verify data only if we are rewinding or error limit was set. | |
2780 | * Otherwise nothing except dbgmsg care about it to waste time. | |
2781 | */ | |
2782 | sle.sle_verify_data = (policy.zlp_rewind & ZPOOL_REWIND_MASK) || | |
2783 | (policy.zlp_maxdata < UINT64_MAX); | |
2784 | ||
2785 | rio = zio_root(spa, NULL, &sle, | |
2786 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
2787 | ||
2788 | if (spa_load_verify_metadata) { | |
2789 | if (spa->spa_extreme_rewind) { | |
2790 | spa_load_note(spa, "performing a complete scan of the " | |
2791 | "pool since extreme rewind is on. This may take " | |
2792 | "a very long time.\n (spa_load_verify_data=%u, " | |
2793 | "spa_load_verify_metadata=%u)", | |
2794 | spa_load_verify_data, spa_load_verify_metadata); | |
2795 | } | |
2796 | ||
2797 | error = traverse_pool(spa, spa->spa_verify_min_txg, | |
2798 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | | |
2799 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
2800 | } | |
2801 | ||
2802 | (void) zio_wait(rio); | |
2803 | ASSERT0(spa->spa_load_verify_bytes); | |
2804 | ||
2805 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2806 | spa->spa_load_data_errors = sle.sle_data_count; | |
2807 | ||
2808 | if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { | |
2809 | spa_load_note(spa, "spa_load_verify found %llu metadata errors " | |
2810 | "and %llu data errors", (u_longlong_t)sle.sle_meta_count, | |
2811 | (u_longlong_t)sle.sle_data_count); | |
2812 | } | |
2813 | ||
2814 | if (spa_load_verify_dryrun || | |
2815 | (!error && sle.sle_meta_count <= policy.zlp_maxmeta && | |
2816 | sle.sle_data_count <= policy.zlp_maxdata)) { | |
2817 | int64_t loss = 0; | |
2818 | ||
2819 | verify_ok = B_TRUE; | |
2820 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2821 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
2822 | ||
2823 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
2824 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, | |
2825 | spa->spa_load_txg_ts); | |
2826 | fnvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, | |
2827 | loss); | |
2828 | fnvlist_add_uint64(spa->spa_load_info, | |
2829 | ZPOOL_CONFIG_LOAD_META_ERRORS, sle.sle_meta_count); | |
2830 | fnvlist_add_uint64(spa->spa_load_info, | |
2831 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count); | |
2832 | } else { | |
2833 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2834 | } | |
2835 | ||
2836 | if (spa_load_verify_dryrun) | |
2837 | return (0); | |
2838 | ||
2839 | if (error) { | |
2840 | if (error != ENXIO && error != EIO) | |
2841 | error = SET_ERROR(EIO); | |
2842 | return (error); | |
2843 | } | |
2844 | ||
2845 | return (verify_ok ? 0 : EIO); | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * Find a value in the pool props object. | |
2850 | */ | |
2851 | static void | |
2852 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2853 | { | |
2854 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2855 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * Find a value in the pool directory object. | |
2860 | */ | |
2861 | static int | |
2862 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) | |
2863 | { | |
2864 | int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2865 | name, sizeof (uint64_t), 1, val); | |
2866 | ||
2867 | if (error != 0 && (error != ENOENT || log_enoent)) { | |
2868 | spa_load_failed(spa, "couldn't get '%s' value in MOS directory " | |
2869 | "[error=%d]", name, error); | |
2870 | } | |
2871 | ||
2872 | return (error); | |
2873 | } | |
2874 | ||
2875 | static int | |
2876 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2877 | { | |
2878 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
2879 | return (SET_ERROR(err)); | |
2880 | } | |
2881 | ||
2882 | boolean_t | |
2883 | spa_livelist_delete_check(spa_t *spa) | |
2884 | { | |
2885 | return (spa->spa_livelists_to_delete != 0); | |
2886 | } | |
2887 | ||
2888 | static boolean_t | |
2889 | spa_livelist_delete_cb_check(void *arg, zthr_t *z) | |
2890 | { | |
2891 | (void) z; | |
2892 | spa_t *spa = arg; | |
2893 | return (spa_livelist_delete_check(spa)); | |
2894 | } | |
2895 | ||
2896 | static int | |
2897 | delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
2898 | { | |
2899 | spa_t *spa = arg; | |
2900 | zio_free(spa, tx->tx_txg, bp); | |
2901 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, | |
2902 | -bp_get_dsize_sync(spa, bp), | |
2903 | -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); | |
2904 | return (0); | |
2905 | } | |
2906 | ||
2907 | static int | |
2908 | dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp) | |
2909 | { | |
2910 | int err; | |
2911 | zap_cursor_t zc; | |
2912 | zap_attribute_t za; | |
2913 | zap_cursor_init(&zc, os, zap_obj); | |
2914 | err = zap_cursor_retrieve(&zc, &za); | |
2915 | zap_cursor_fini(&zc); | |
2916 | if (err == 0) | |
2917 | *llp = za.za_first_integer; | |
2918 | return (err); | |
2919 | } | |
2920 | ||
2921 | /* | |
2922 | * Components of livelist deletion that must be performed in syncing | |
2923 | * context: freeing block pointers and updating the pool-wide data | |
2924 | * structures to indicate how much work is left to do | |
2925 | */ | |
2926 | typedef struct sublist_delete_arg { | |
2927 | spa_t *spa; | |
2928 | dsl_deadlist_t *ll; | |
2929 | uint64_t key; | |
2930 | bplist_t *to_free; | |
2931 | } sublist_delete_arg_t; | |
2932 | ||
2933 | static void | |
2934 | sublist_delete_sync(void *arg, dmu_tx_t *tx) | |
2935 | { | |
2936 | sublist_delete_arg_t *sda = arg; | |
2937 | spa_t *spa = sda->spa; | |
2938 | dsl_deadlist_t *ll = sda->ll; | |
2939 | uint64_t key = sda->key; | |
2940 | bplist_t *to_free = sda->to_free; | |
2941 | ||
2942 | bplist_iterate(to_free, delete_blkptr_cb, spa, tx); | |
2943 | dsl_deadlist_remove_entry(ll, key, tx); | |
2944 | } | |
2945 | ||
2946 | typedef struct livelist_delete_arg { | |
2947 | spa_t *spa; | |
2948 | uint64_t ll_obj; | |
2949 | uint64_t zap_obj; | |
2950 | } livelist_delete_arg_t; | |
2951 | ||
2952 | static void | |
2953 | livelist_delete_sync(void *arg, dmu_tx_t *tx) | |
2954 | { | |
2955 | livelist_delete_arg_t *lda = arg; | |
2956 | spa_t *spa = lda->spa; | |
2957 | uint64_t ll_obj = lda->ll_obj; | |
2958 | uint64_t zap_obj = lda->zap_obj; | |
2959 | objset_t *mos = spa->spa_meta_objset; | |
2960 | uint64_t count; | |
2961 | ||
2962 | /* free the livelist and decrement the feature count */ | |
2963 | VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx)); | |
2964 | dsl_deadlist_free(mos, ll_obj, tx); | |
2965 | spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx); | |
2966 | VERIFY0(zap_count(mos, zap_obj, &count)); | |
2967 | if (count == 0) { | |
2968 | /* no more livelists to delete */ | |
2969 | VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT, | |
2970 | DMU_POOL_DELETED_CLONES, tx)); | |
2971 | VERIFY0(zap_destroy(mos, zap_obj, tx)); | |
2972 | spa->spa_livelists_to_delete = 0; | |
2973 | spa_notify_waiters(spa); | |
2974 | } | |
2975 | } | |
2976 | ||
2977 | /* | |
2978 | * Load in the value for the livelist to be removed and open it. Then, | |
2979 | * load its first sublist and determine which block pointers should actually | |
2980 | * be freed. Then, call a synctask which performs the actual frees and updates | |
2981 | * the pool-wide livelist data. | |
2982 | */ | |
2983 | static void | |
2984 | spa_livelist_delete_cb(void *arg, zthr_t *z) | |
2985 | { | |
2986 | spa_t *spa = arg; | |
2987 | uint64_t ll_obj = 0, count; | |
2988 | objset_t *mos = spa->spa_meta_objset; | |
2989 | uint64_t zap_obj = spa->spa_livelists_to_delete; | |
2990 | /* | |
2991 | * Determine the next livelist to delete. This function should only | |
2992 | * be called if there is at least one deleted clone. | |
2993 | */ | |
2994 | VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj)); | |
2995 | VERIFY0(zap_count(mos, ll_obj, &count)); | |
2996 | if (count > 0) { | |
2997 | dsl_deadlist_t *ll; | |
2998 | dsl_deadlist_entry_t *dle; | |
2999 | bplist_t to_free; | |
3000 | ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP); | |
3001 | dsl_deadlist_open(ll, mos, ll_obj); | |
3002 | dle = dsl_deadlist_first(ll); | |
3003 | ASSERT3P(dle, !=, NULL); | |
3004 | bplist_create(&to_free); | |
3005 | int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free, | |
3006 | z, NULL); | |
3007 | if (err == 0) { | |
3008 | sublist_delete_arg_t sync_arg = { | |
3009 | .spa = spa, | |
3010 | .ll = ll, | |
3011 | .key = dle->dle_mintxg, | |
3012 | .to_free = &to_free | |
3013 | }; | |
3014 | zfs_dbgmsg("deleting sublist (id %llu) from" | |
3015 | " livelist %llu, %lld remaining", | |
3016 | (u_longlong_t)dle->dle_bpobj.bpo_object, | |
3017 | (u_longlong_t)ll_obj, (longlong_t)count - 1); | |
3018 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, | |
3019 | sublist_delete_sync, &sync_arg, 0, | |
3020 | ZFS_SPACE_CHECK_DESTROY)); | |
3021 | } else { | |
3022 | VERIFY3U(err, ==, EINTR); | |
3023 | } | |
3024 | bplist_clear(&to_free); | |
3025 | bplist_destroy(&to_free); | |
3026 | dsl_deadlist_close(ll); | |
3027 | kmem_free(ll, sizeof (dsl_deadlist_t)); | |
3028 | } else { | |
3029 | livelist_delete_arg_t sync_arg = { | |
3030 | .spa = spa, | |
3031 | .ll_obj = ll_obj, | |
3032 | .zap_obj = zap_obj | |
3033 | }; | |
3034 | zfs_dbgmsg("deletion of livelist %llu completed", | |
3035 | (u_longlong_t)ll_obj); | |
3036 | VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync, | |
3037 | &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY)); | |
3038 | } | |
3039 | } | |
3040 | ||
3041 | static void | |
3042 | spa_start_livelist_destroy_thread(spa_t *spa) | |
3043 | { | |
3044 | ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL); | |
3045 | spa->spa_livelist_delete_zthr = | |
3046 | zthr_create("z_livelist_destroy", | |
3047 | spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa, | |
3048 | minclsyspri); | |
3049 | } | |
3050 | ||
3051 | typedef struct livelist_new_arg { | |
3052 | bplist_t *allocs; | |
3053 | bplist_t *frees; | |
3054 | } livelist_new_arg_t; | |
3055 | ||
3056 | static int | |
3057 | livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
3058 | dmu_tx_t *tx) | |
3059 | { | |
3060 | ASSERT(tx == NULL); | |
3061 | livelist_new_arg_t *lna = arg; | |
3062 | if (bp_freed) { | |
3063 | bplist_append(lna->frees, bp); | |
3064 | } else { | |
3065 | bplist_append(lna->allocs, bp); | |
3066 | zfs_livelist_condense_new_alloc++; | |
3067 | } | |
3068 | return (0); | |
3069 | } | |
3070 | ||
3071 | typedef struct livelist_condense_arg { | |
3072 | spa_t *spa; | |
3073 | bplist_t to_keep; | |
3074 | uint64_t first_size; | |
3075 | uint64_t next_size; | |
3076 | } livelist_condense_arg_t; | |
3077 | ||
3078 | static void | |
3079 | spa_livelist_condense_sync(void *arg, dmu_tx_t *tx) | |
3080 | { | |
3081 | livelist_condense_arg_t *lca = arg; | |
3082 | spa_t *spa = lca->spa; | |
3083 | bplist_t new_frees; | |
3084 | dsl_dataset_t *ds = spa->spa_to_condense.ds; | |
3085 | ||
3086 | /* Have we been cancelled? */ | |
3087 | if (spa->spa_to_condense.cancelled) { | |
3088 | zfs_livelist_condense_sync_cancel++; | |
3089 | goto out; | |
3090 | } | |
3091 | ||
3092 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
3093 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
3094 | dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; | |
3095 | ||
3096 | /* | |
3097 | * It's possible that the livelist was changed while the zthr was | |
3098 | * running. Therefore, we need to check for new blkptrs in the two | |
3099 | * entries being condensed and continue to track them in the livelist. | |
3100 | * Because of the way we handle remapped blkptrs (see dbuf_remap_impl), | |
3101 | * it's possible that the newly added blkptrs are FREEs or ALLOCs so | |
3102 | * we need to sort them into two different bplists. | |
3103 | */ | |
3104 | uint64_t first_obj = first->dle_bpobj.bpo_object; | |
3105 | uint64_t next_obj = next->dle_bpobj.bpo_object; | |
3106 | uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
3107 | uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs; | |
3108 | ||
3109 | bplist_create(&new_frees); | |
3110 | livelist_new_arg_t new_bps = { | |
3111 | .allocs = &lca->to_keep, | |
3112 | .frees = &new_frees, | |
3113 | }; | |
3114 | ||
3115 | if (cur_first_size > lca->first_size) { | |
3116 | VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj, | |
3117 | livelist_track_new_cb, &new_bps, lca->first_size)); | |
3118 | } | |
3119 | if (cur_next_size > lca->next_size) { | |
3120 | VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj, | |
3121 | livelist_track_new_cb, &new_bps, lca->next_size)); | |
3122 | } | |
3123 | ||
3124 | dsl_deadlist_clear_entry(first, ll, tx); | |
3125 | ASSERT(bpobj_is_empty(&first->dle_bpobj)); | |
3126 | dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx); | |
3127 | ||
3128 | bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx); | |
3129 | bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx); | |
3130 | bplist_destroy(&new_frees); | |
3131 | ||
3132 | char dsname[ZFS_MAX_DATASET_NAME_LEN]; | |
3133 | dsl_dataset_name(ds, dsname); | |
3134 | zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu " | |
3135 | "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu " | |
3136 | "(%llu blkptrs)", (u_longlong_t)tx->tx_txg, dsname, | |
3137 | (u_longlong_t)ds->ds_object, (u_longlong_t)first_obj, | |
3138 | (u_longlong_t)cur_first_size, (u_longlong_t)next_obj, | |
3139 | (u_longlong_t)cur_next_size, | |
3140 | (u_longlong_t)first->dle_bpobj.bpo_object, | |
3141 | (u_longlong_t)first->dle_bpobj.bpo_phys->bpo_num_blkptrs); | |
3142 | out: | |
3143 | dmu_buf_rele(ds->ds_dbuf, spa); | |
3144 | spa->spa_to_condense.ds = NULL; | |
3145 | bplist_clear(&lca->to_keep); | |
3146 | bplist_destroy(&lca->to_keep); | |
3147 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
3148 | spa->spa_to_condense.syncing = B_FALSE; | |
3149 | } | |
3150 | ||
3151 | static void | |
3152 | spa_livelist_condense_cb(void *arg, zthr_t *t) | |
3153 | { | |
3154 | while (zfs_livelist_condense_zthr_pause && | |
3155 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
3156 | delay(1); | |
3157 | ||
3158 | spa_t *spa = arg; | |
3159 | dsl_deadlist_entry_t *first = spa->spa_to_condense.first; | |
3160 | dsl_deadlist_entry_t *next = spa->spa_to_condense.next; | |
3161 | uint64_t first_size, next_size; | |
3162 | ||
3163 | livelist_condense_arg_t *lca = | |
3164 | kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP); | |
3165 | bplist_create(&lca->to_keep); | |
3166 | ||
3167 | /* | |
3168 | * Process the livelists (matching FREEs and ALLOCs) in open context | |
3169 | * so we have minimal work in syncing context to condense. | |
3170 | * | |
3171 | * We save bpobj sizes (first_size and next_size) to use later in | |
3172 | * syncing context to determine if entries were added to these sublists | |
3173 | * while in open context. This is possible because the clone is still | |
3174 | * active and open for normal writes and we want to make sure the new, | |
3175 | * unprocessed blockpointers are inserted into the livelist normally. | |
3176 | * | |
3177 | * Note that dsl_process_sub_livelist() both stores the size number of | |
3178 | * blockpointers and iterates over them while the bpobj's lock held, so | |
3179 | * the sizes returned to us are consistent which what was actually | |
3180 | * processed. | |
3181 | */ | |
3182 | int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t, | |
3183 | &first_size); | |
3184 | if (err == 0) | |
3185 | err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep, | |
3186 | t, &next_size); | |
3187 | ||
3188 | if (err == 0) { | |
3189 | while (zfs_livelist_condense_sync_pause && | |
3190 | !(zthr_has_waiters(t) || zthr_iscancelled(t))) | |
3191 | delay(1); | |
3192 | ||
3193 | dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
3194 | dmu_tx_mark_netfree(tx); | |
3195 | dmu_tx_hold_space(tx, 1); | |
3196 | err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE); | |
3197 | if (err == 0) { | |
3198 | /* | |
3199 | * Prevent the condense zthr restarting before | |
3200 | * the synctask completes. | |
3201 | */ | |
3202 | spa->spa_to_condense.syncing = B_TRUE; | |
3203 | lca->spa = spa; | |
3204 | lca->first_size = first_size; | |
3205 | lca->next_size = next_size; | |
3206 | dsl_sync_task_nowait(spa_get_dsl(spa), | |
3207 | spa_livelist_condense_sync, lca, tx); | |
3208 | dmu_tx_commit(tx); | |
3209 | return; | |
3210 | } | |
3211 | } | |
3212 | /* | |
3213 | * Condensing can not continue: either it was externally stopped or | |
3214 | * we were unable to assign to a tx because the pool has run out of | |
3215 | * space. In the second case, we'll just end up trying to condense | |
3216 | * again in a later txg. | |
3217 | */ | |
3218 | ASSERT(err != 0); | |
3219 | bplist_clear(&lca->to_keep); | |
3220 | bplist_destroy(&lca->to_keep); | |
3221 | kmem_free(lca, sizeof (livelist_condense_arg_t)); | |
3222 | dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa); | |
3223 | spa->spa_to_condense.ds = NULL; | |
3224 | if (err == EINTR) | |
3225 | zfs_livelist_condense_zthr_cancel++; | |
3226 | } | |
3227 | ||
3228 | /* | |
3229 | * Check that there is something to condense but that a condense is not | |
3230 | * already in progress and that condensing has not been cancelled. | |
3231 | */ | |
3232 | static boolean_t | |
3233 | spa_livelist_condense_cb_check(void *arg, zthr_t *z) | |
3234 | { | |
3235 | (void) z; | |
3236 | spa_t *spa = arg; | |
3237 | if ((spa->spa_to_condense.ds != NULL) && | |
3238 | (spa->spa_to_condense.syncing == B_FALSE) && | |
3239 | (spa->spa_to_condense.cancelled == B_FALSE)) { | |
3240 | return (B_TRUE); | |
3241 | } | |
3242 | return (B_FALSE); | |
3243 | } | |
3244 | ||
3245 | static void | |
3246 | spa_start_livelist_condensing_thread(spa_t *spa) | |
3247 | { | |
3248 | spa->spa_to_condense.ds = NULL; | |
3249 | spa->spa_to_condense.first = NULL; | |
3250 | spa->spa_to_condense.next = NULL; | |
3251 | spa->spa_to_condense.syncing = B_FALSE; | |
3252 | spa->spa_to_condense.cancelled = B_FALSE; | |
3253 | ||
3254 | ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL); | |
3255 | spa->spa_livelist_condense_zthr = | |
3256 | zthr_create("z_livelist_condense", | |
3257 | spa_livelist_condense_cb_check, | |
3258 | spa_livelist_condense_cb, spa, minclsyspri); | |
3259 | } | |
3260 | ||
3261 | static void | |
3262 | spa_spawn_aux_threads(spa_t *spa) | |
3263 | { | |
3264 | ASSERT(spa_writeable(spa)); | |
3265 | ||
3266 | spa_start_raidz_expansion_thread(spa); | |
3267 | spa_start_indirect_condensing_thread(spa); | |
3268 | spa_start_livelist_destroy_thread(spa); | |
3269 | spa_start_livelist_condensing_thread(spa); | |
3270 | ||
3271 | ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); | |
3272 | spa->spa_checkpoint_discard_zthr = | |
3273 | zthr_create("z_checkpoint_discard", | |
3274 | spa_checkpoint_discard_thread_check, | |
3275 | spa_checkpoint_discard_thread, spa, minclsyspri); | |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * Fix up config after a partly-completed split. This is done with the | |
3280 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
3281 | * pool have that entry in their config, but only the splitting one contains | |
3282 | * a list of all the guids of the vdevs that are being split off. | |
3283 | * | |
3284 | * This function determines what to do with that list: either rejoin | |
3285 | * all the disks to the pool, or complete the splitting process. To attempt | |
3286 | * the rejoin, each disk that is offlined is marked online again, and | |
3287 | * we do a reopen() call. If the vdev label for every disk that was | |
3288 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
3289 | * then we call vdev_split() on each disk, and complete the split. | |
3290 | * | |
3291 | * Otherwise we leave the config alone, with all the vdevs in place in | |
3292 | * the original pool. | |
3293 | */ | |
3294 | static void | |
3295 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
3296 | { | |
3297 | uint_t extracted; | |
3298 | uint64_t *glist; | |
3299 | uint_t i, gcount; | |
3300 | nvlist_t *nvl; | |
3301 | vdev_t **vd; | |
3302 | boolean_t attempt_reopen; | |
3303 | ||
3304 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
3305 | return; | |
3306 | ||
3307 | /* check that the config is complete */ | |
3308 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
3309 | &glist, &gcount) != 0) | |
3310 | return; | |
3311 | ||
3312 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); | |
3313 | ||
3314 | /* attempt to online all the vdevs & validate */ | |
3315 | attempt_reopen = B_TRUE; | |
3316 | for (i = 0; i < gcount; i++) { | |
3317 | if (glist[i] == 0) /* vdev is hole */ | |
3318 | continue; | |
3319 | ||
3320 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
3321 | if (vd[i] == NULL) { | |
3322 | /* | |
3323 | * Don't bother attempting to reopen the disks; | |
3324 | * just do the split. | |
3325 | */ | |
3326 | attempt_reopen = B_FALSE; | |
3327 | } else { | |
3328 | /* attempt to re-online it */ | |
3329 | vd[i]->vdev_offline = B_FALSE; | |
3330 | } | |
3331 | } | |
3332 | ||
3333 | if (attempt_reopen) { | |
3334 | vdev_reopen(spa->spa_root_vdev); | |
3335 | ||
3336 | /* check each device to see what state it's in */ | |
3337 | for (extracted = 0, i = 0; i < gcount; i++) { | |
3338 | if (vd[i] != NULL && | |
3339 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
3340 | break; | |
3341 | ++extracted; | |
3342 | } | |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * If every disk has been moved to the new pool, or if we never | |
3347 | * even attempted to look at them, then we split them off for | |
3348 | * good. | |
3349 | */ | |
3350 | if (!attempt_reopen || gcount == extracted) { | |
3351 | for (i = 0; i < gcount; i++) | |
3352 | if (vd[i] != NULL) | |
3353 | vdev_split(vd[i]); | |
3354 | vdev_reopen(spa->spa_root_vdev); | |
3355 | } | |
3356 | ||
3357 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
3358 | } | |
3359 | ||
3360 | static int | |
3361 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) | |
3362 | { | |
3363 | const char *ereport = FM_EREPORT_ZFS_POOL; | |
3364 | int error; | |
3365 | ||
3366 | spa->spa_load_state = state; | |
3367 | (void) spa_import_progress_set_state(spa_guid(spa), | |
3368 | spa_load_state(spa)); | |
3369 | spa_import_progress_set_notes(spa, "spa_load()"); | |
3370 | ||
3371 | gethrestime(&spa->spa_loaded_ts); | |
3372 | error = spa_load_impl(spa, type, &ereport); | |
3373 | ||
3374 | /* | |
3375 | * Don't count references from objsets that are already closed | |
3376 | * and are making their way through the eviction process. | |
3377 | */ | |
3378 | spa_evicting_os_wait(spa); | |
3379 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); | |
3380 | if (error) { | |
3381 | if (error != EEXIST) { | |
3382 | spa->spa_loaded_ts.tv_sec = 0; | |
3383 | spa->spa_loaded_ts.tv_nsec = 0; | |
3384 | } | |
3385 | if (error != EBADF) { | |
3386 | (void) zfs_ereport_post(ereport, spa, | |
3387 | NULL, NULL, NULL, 0); | |
3388 | } | |
3389 | } | |
3390 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; | |
3391 | spa->spa_ena = 0; | |
3392 | ||
3393 | (void) spa_import_progress_set_state(spa_guid(spa), | |
3394 | spa_load_state(spa)); | |
3395 | ||
3396 | return (error); | |
3397 | } | |
3398 | ||
3399 | #ifdef ZFS_DEBUG | |
3400 | /* | |
3401 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
3402 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
3403 | * spa's per-vdev ZAP list. | |
3404 | */ | |
3405 | static uint64_t | |
3406 | vdev_count_verify_zaps(vdev_t *vd) | |
3407 | { | |
3408 | spa_t *spa = vd->vdev_spa; | |
3409 | uint64_t total = 0; | |
3410 | ||
3411 | if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2) && | |
3412 | vd->vdev_root_zap != 0) { | |
3413 | total++; | |
3414 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3415 | spa->spa_all_vdev_zaps, vd->vdev_root_zap)); | |
3416 | } | |
3417 | if (vd->vdev_top_zap != 0) { | |
3418 | total++; | |
3419 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3420 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
3421 | } | |
3422 | if (vd->vdev_leaf_zap != 0) { | |
3423 | total++; | |
3424 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
3425 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
3426 | } | |
3427 | ||
3428 | for (uint64_t i = 0; i < vd->vdev_children; i++) { | |
3429 | total += vdev_count_verify_zaps(vd->vdev_child[i]); | |
3430 | } | |
3431 | ||
3432 | return (total); | |
3433 | } | |
3434 | #else | |
3435 | #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0) | |
3436 | #endif | |
3437 | ||
3438 | /* | |
3439 | * Determine whether the activity check is required. | |
3440 | */ | |
3441 | static boolean_t | |
3442 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, | |
3443 | nvlist_t *config) | |
3444 | { | |
3445 | uint64_t state = 0; | |
3446 | uint64_t hostid = 0; | |
3447 | uint64_t tryconfig_txg = 0; | |
3448 | uint64_t tryconfig_timestamp = 0; | |
3449 | uint16_t tryconfig_mmp_seq = 0; | |
3450 | nvlist_t *nvinfo; | |
3451 | ||
3452 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3453 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
3454 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
3455 | &tryconfig_txg); | |
3456 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
3457 | &tryconfig_timestamp); | |
3458 | (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ, | |
3459 | &tryconfig_mmp_seq); | |
3460 | } | |
3461 | ||
3462 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
3463 | ||
3464 | /* | |
3465 | * Disable the MMP activity check - This is used by zdb which | |
3466 | * is intended to be used on potentially active pools. | |
3467 | */ | |
3468 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
3469 | return (B_FALSE); | |
3470 | ||
3471 | /* | |
3472 | * Skip the activity check when the MMP feature is disabled. | |
3473 | */ | |
3474 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
3475 | return (B_FALSE); | |
3476 | ||
3477 | /* | |
3478 | * If the tryconfig_ values are nonzero, they are the results of an | |
3479 | * earlier tryimport. If they all match the uberblock we just found, | |
3480 | * then the pool has not changed and we return false so we do not test | |
3481 | * a second time. | |
3482 | */ | |
3483 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
3484 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp && | |
3485 | tryconfig_mmp_seq && tryconfig_mmp_seq == | |
3486 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) | |
3487 | return (B_FALSE); | |
3488 | ||
3489 | /* | |
3490 | * Allow the activity check to be skipped when importing the pool | |
3491 | * on the same host which last imported it. Since the hostid from | |
3492 | * configuration may be stale use the one read from the label. | |
3493 | */ | |
3494 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) | |
3495 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
3496 | ||
3497 | if (hostid == spa_get_hostid(spa)) | |
3498 | return (B_FALSE); | |
3499 | ||
3500 | /* | |
3501 | * Skip the activity test when the pool was cleanly exported. | |
3502 | */ | |
3503 | if (state != POOL_STATE_ACTIVE) | |
3504 | return (B_FALSE); | |
3505 | ||
3506 | return (B_TRUE); | |
3507 | } | |
3508 | ||
3509 | /* | |
3510 | * Nanoseconds the activity check must watch for changes on-disk. | |
3511 | */ | |
3512 | static uint64_t | |
3513 | spa_activity_check_duration(spa_t *spa, uberblock_t *ub) | |
3514 | { | |
3515 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
3516 | uint64_t multihost_interval = MSEC2NSEC( | |
3517 | MMP_INTERVAL_OK(zfs_multihost_interval)); | |
3518 | uint64_t import_delay = MAX(NANOSEC, import_intervals * | |
3519 | multihost_interval); | |
3520 | ||
3521 | /* | |
3522 | * Local tunables determine a minimum duration except for the case | |
3523 | * where we know when the remote host will suspend the pool if MMP | |
3524 | * writes do not land. | |
3525 | * | |
3526 | * See Big Theory comment at the top of mmp.c for the reasoning behind | |
3527 | * these cases and times. | |
3528 | */ | |
3529 | ||
3530 | ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100); | |
3531 | ||
3532 | if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3533 | MMP_FAIL_INT(ub) > 0) { | |
3534 | ||
3535 | /* MMP on remote host will suspend pool after failed writes */ | |
3536 | import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) * | |
3537 | MMP_IMPORT_SAFETY_FACTOR / 100; | |
3538 | ||
3539 | zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp " | |
3540 | "mmp_fails=%llu ub_mmp mmp_interval=%llu " | |
3541 | "import_intervals=%llu", (u_longlong_t)import_delay, | |
3542 | (u_longlong_t)MMP_FAIL_INT(ub), | |
3543 | (u_longlong_t)MMP_INTERVAL(ub), | |
3544 | (u_longlong_t)import_intervals); | |
3545 | ||
3546 | } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) && | |
3547 | MMP_FAIL_INT(ub) == 0) { | |
3548 | ||
3549 | /* MMP on remote host will never suspend pool */ | |
3550 | import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) + | |
3551 | ub->ub_mmp_delay) * import_intervals); | |
3552 | ||
3553 | zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp " | |
3554 | "mmp_interval=%llu ub_mmp_delay=%llu " | |
3555 | "import_intervals=%llu", (u_longlong_t)import_delay, | |
3556 | (u_longlong_t)MMP_INTERVAL(ub), | |
3557 | (u_longlong_t)ub->ub_mmp_delay, | |
3558 | (u_longlong_t)import_intervals); | |
3559 | ||
3560 | } else if (MMP_VALID(ub)) { | |
3561 | /* | |
3562 | * zfs-0.7 compatibility case | |
3563 | */ | |
3564 | ||
3565 | import_delay = MAX(import_delay, (multihost_interval + | |
3566 | ub->ub_mmp_delay) * import_intervals); | |
3567 | ||
3568 | zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu " | |
3569 | "import_intervals=%llu leaves=%u", | |
3570 | (u_longlong_t)import_delay, | |
3571 | (u_longlong_t)ub->ub_mmp_delay, | |
3572 | (u_longlong_t)import_intervals, | |
3573 | vdev_count_leaves(spa)); | |
3574 | } else { | |
3575 | /* Using local tunings is the only reasonable option */ | |
3576 | zfs_dbgmsg("pool last imported on non-MMP aware " | |
3577 | "host using import_delay=%llu multihost_interval=%llu " | |
3578 | "import_intervals=%llu", (u_longlong_t)import_delay, | |
3579 | (u_longlong_t)multihost_interval, | |
3580 | (u_longlong_t)import_intervals); | |
3581 | } | |
3582 | ||
3583 | return (import_delay); | |
3584 | } | |
3585 | ||
3586 | /* | |
3587 | * Remote host activity check. | |
3588 | * | |
3589 | * error results: | |
3590 | * 0 - no activity detected | |
3591 | * EREMOTEIO - remote activity detected | |
3592 | * EINTR - user canceled the operation | |
3593 | */ | |
3594 | static int | |
3595 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config, | |
3596 | boolean_t importing) | |
3597 | { | |
3598 | uint64_t txg = ub->ub_txg; | |
3599 | uint64_t timestamp = ub->ub_timestamp; | |
3600 | uint64_t mmp_config = ub->ub_mmp_config; | |
3601 | uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0; | |
3602 | uint64_t import_delay; | |
3603 | hrtime_t import_expire, now; | |
3604 | nvlist_t *mmp_label = NULL; | |
3605 | vdev_t *rvd = spa->spa_root_vdev; | |
3606 | kcondvar_t cv; | |
3607 | kmutex_t mtx; | |
3608 | int error = 0; | |
3609 | ||
3610 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
3611 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
3612 | mutex_enter(&mtx); | |
3613 | ||
3614 | /* | |
3615 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
3616 | * during the earlier tryimport. If the txg recorded there is 0 then | |
3617 | * the pool is known to be active on another host. | |
3618 | * | |
3619 | * Otherwise, the pool might be in use on another host. Check for | |
3620 | * changes in the uberblocks on disk if necessary. | |
3621 | */ | |
3622 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
3623 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
3624 | ZPOOL_CONFIG_LOAD_INFO); | |
3625 | ||
3626 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
3627 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
3628 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
3629 | error = SET_ERROR(EREMOTEIO); | |
3630 | goto out; | |
3631 | } | |
3632 | } | |
3633 | ||
3634 | import_delay = spa_activity_check_duration(spa, ub); | |
3635 | ||
3636 | /* Add a small random factor in case of simultaneous imports (0-25%) */ | |
3637 | import_delay += import_delay * random_in_range(250) / 1000; | |
3638 | ||
3639 | import_expire = gethrtime() + import_delay; | |
3640 | ||
3641 | if (importing) { | |
3642 | spa_import_progress_set_notes(spa, "Checking MMP activity, " | |
3643 | "waiting %llu ms", (u_longlong_t)NSEC2MSEC(import_delay)); | |
3644 | } | |
3645 | ||
3646 | int iterations = 0; | |
3647 | while ((now = gethrtime()) < import_expire) { | |
3648 | if (importing && iterations++ % 30 == 0) { | |
3649 | spa_import_progress_set_notes(spa, "Checking MMP " | |
3650 | "activity, %llu ms remaining", | |
3651 | (u_longlong_t)NSEC2MSEC(import_expire - now)); | |
3652 | } | |
3653 | ||
3654 | if (importing) { | |
3655 | (void) spa_import_progress_set_mmp_check(spa_guid(spa), | |
3656 | NSEC2SEC(import_expire - gethrtime())); | |
3657 | } | |
3658 | ||
3659 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
3660 | ||
3661 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp || | |
3662 | mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) { | |
3663 | zfs_dbgmsg("multihost activity detected " | |
3664 | "txg %llu ub_txg %llu " | |
3665 | "timestamp %llu ub_timestamp %llu " | |
3666 | "mmp_config %#llx ub_mmp_config %#llx", | |
3667 | (u_longlong_t)txg, (u_longlong_t)ub->ub_txg, | |
3668 | (u_longlong_t)timestamp, | |
3669 | (u_longlong_t)ub->ub_timestamp, | |
3670 | (u_longlong_t)mmp_config, | |
3671 | (u_longlong_t)ub->ub_mmp_config); | |
3672 | ||
3673 | error = SET_ERROR(EREMOTEIO); | |
3674 | break; | |
3675 | } | |
3676 | ||
3677 | if (mmp_label) { | |
3678 | nvlist_free(mmp_label); | |
3679 | mmp_label = NULL; | |
3680 | } | |
3681 | ||
3682 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
3683 | if (error != -1) { | |
3684 | error = SET_ERROR(EINTR); | |
3685 | break; | |
3686 | } | |
3687 | error = 0; | |
3688 | } | |
3689 | ||
3690 | out: | |
3691 | mutex_exit(&mtx); | |
3692 | mutex_destroy(&mtx); | |
3693 | cv_destroy(&cv); | |
3694 | ||
3695 | /* | |
3696 | * If the pool is determined to be active store the status in the | |
3697 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
3698 | * available from configuration read from disk store them as well. | |
3699 | * This allows 'zpool import' to generate a more useful message. | |
3700 | * | |
3701 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
3702 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
3703 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
3704 | */ | |
3705 | if (error == EREMOTEIO) { | |
3706 | const char *hostname = "<unknown>"; | |
3707 | uint64_t hostid = 0; | |
3708 | ||
3709 | if (mmp_label) { | |
3710 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
3711 | hostname = fnvlist_lookup_string(mmp_label, | |
3712 | ZPOOL_CONFIG_HOSTNAME); | |
3713 | fnvlist_add_string(spa->spa_load_info, | |
3714 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
3715 | } | |
3716 | ||
3717 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
3718 | hostid = fnvlist_lookup_uint64(mmp_label, | |
3719 | ZPOOL_CONFIG_HOSTID); | |
3720 | fnvlist_add_uint64(spa->spa_load_info, | |
3721 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
3722 | } | |
3723 | } | |
3724 | ||
3725 | fnvlist_add_uint64(spa->spa_load_info, | |
3726 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
3727 | fnvlist_add_uint64(spa->spa_load_info, | |
3728 | ZPOOL_CONFIG_MMP_TXG, 0); | |
3729 | ||
3730 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
3731 | } | |
3732 | ||
3733 | if (mmp_label) | |
3734 | nvlist_free(mmp_label); | |
3735 | ||
3736 | return (error); | |
3737 | } | |
3738 | ||
3739 | /* | |
3740 | * Called from zfs_ioc_clear for a pool that was suspended | |
3741 | * after failing mmp write checks. | |
3742 | */ | |
3743 | boolean_t | |
3744 | spa_mmp_remote_host_activity(spa_t *spa) | |
3745 | { | |
3746 | ASSERT(spa_multihost(spa) && spa_suspended(spa)); | |
3747 | ||
3748 | nvlist_t *best_label; | |
3749 | uberblock_t best_ub; | |
3750 | ||
3751 | /* | |
3752 | * Locate the best uberblock on disk | |
3753 | */ | |
3754 | vdev_uberblock_load(spa->spa_root_vdev, &best_ub, &best_label); | |
3755 | if (best_label) { | |
3756 | /* | |
3757 | * confirm that the best hostid matches our hostid | |
3758 | */ | |
3759 | if (nvlist_exists(best_label, ZPOOL_CONFIG_HOSTID) && | |
3760 | spa_get_hostid(spa) != | |
3761 | fnvlist_lookup_uint64(best_label, ZPOOL_CONFIG_HOSTID)) { | |
3762 | nvlist_free(best_label); | |
3763 | return (B_TRUE); | |
3764 | } | |
3765 | nvlist_free(best_label); | |
3766 | } else { | |
3767 | return (B_TRUE); | |
3768 | } | |
3769 | ||
3770 | if (!MMP_VALID(&best_ub) || | |
3771 | !MMP_FAIL_INT_VALID(&best_ub) || | |
3772 | MMP_FAIL_INT(&best_ub) == 0) { | |
3773 | return (B_TRUE); | |
3774 | } | |
3775 | ||
3776 | if (best_ub.ub_txg != spa->spa_uberblock.ub_txg || | |
3777 | best_ub.ub_timestamp != spa->spa_uberblock.ub_timestamp) { | |
3778 | zfs_dbgmsg("txg mismatch detected during pool clear " | |
3779 | "txg %llu ub_txg %llu timestamp %llu ub_timestamp %llu", | |
3780 | (u_longlong_t)spa->spa_uberblock.ub_txg, | |
3781 | (u_longlong_t)best_ub.ub_txg, | |
3782 | (u_longlong_t)spa->spa_uberblock.ub_timestamp, | |
3783 | (u_longlong_t)best_ub.ub_timestamp); | |
3784 | return (B_TRUE); | |
3785 | } | |
3786 | ||
3787 | /* | |
3788 | * Perform an activity check looking for any remote writer | |
3789 | */ | |
3790 | return (spa_activity_check(spa, &spa->spa_uberblock, spa->spa_config, | |
3791 | B_FALSE) != 0); | |
3792 | } | |
3793 | ||
3794 | static int | |
3795 | spa_verify_host(spa_t *spa, nvlist_t *mos_config) | |
3796 | { | |
3797 | uint64_t hostid; | |
3798 | const char *hostname; | |
3799 | uint64_t myhostid = 0; | |
3800 | ||
3801 | if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, | |
3802 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
3803 | hostname = fnvlist_lookup_string(mos_config, | |
3804 | ZPOOL_CONFIG_HOSTNAME); | |
3805 | ||
3806 | myhostid = zone_get_hostid(NULL); | |
3807 | ||
3808 | if (hostid != 0 && myhostid != 0 && hostid != myhostid) { | |
3809 | cmn_err(CE_WARN, "pool '%s' could not be " | |
3810 | "loaded as it was last accessed by " | |
3811 | "another system (host: %s hostid: 0x%llx). " | |
3812 | "See: https://openzfs.github.io/openzfs-docs/msg/" | |
3813 | "ZFS-8000-EY", | |
3814 | spa_name(spa), hostname, (u_longlong_t)hostid); | |
3815 | spa_load_failed(spa, "hostid verification failed: pool " | |
3816 | "last accessed by host: %s (hostid: 0x%llx)", | |
3817 | hostname, (u_longlong_t)hostid); | |
3818 | return (SET_ERROR(EBADF)); | |
3819 | } | |
3820 | } | |
3821 | ||
3822 | return (0); | |
3823 | } | |
3824 | ||
3825 | static int | |
3826 | spa_ld_parse_config(spa_t *spa, spa_import_type_t type) | |
3827 | { | |
3828 | int error = 0; | |
3829 | nvlist_t *nvtree, *nvl, *config = spa->spa_config; | |
3830 | int parse; | |
3831 | vdev_t *rvd; | |
3832 | uint64_t pool_guid; | |
3833 | const char *comment; | |
3834 | const char *compatibility; | |
3835 | ||
3836 | /* | |
3837 | * Versioning wasn't explicitly added to the label until later, so if | |
3838 | * it's not present treat it as the initial version. | |
3839 | */ | |
3840 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
3841 | &spa->spa_ubsync.ub_version) != 0) | |
3842 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
3843 | ||
3844 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { | |
3845 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3846 | ZPOOL_CONFIG_POOL_GUID); | |
3847 | return (SET_ERROR(EINVAL)); | |
3848 | } | |
3849 | ||
3850 | /* | |
3851 | * If we are doing an import, ensure that the pool is not already | |
3852 | * imported by checking if its pool guid already exists in the | |
3853 | * spa namespace. | |
3854 | * | |
3855 | * The only case that we allow an already imported pool to be | |
3856 | * imported again, is when the pool is checkpointed and we want to | |
3857 | * look at its checkpointed state from userland tools like zdb. | |
3858 | */ | |
3859 | #ifdef _KERNEL | |
3860 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3861 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3862 | spa_guid_exists(pool_guid, 0)) { | |
3863 | #else | |
3864 | if ((spa->spa_load_state == SPA_LOAD_IMPORT || | |
3865 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) && | |
3866 | spa_guid_exists(pool_guid, 0) && | |
3867 | !spa_importing_readonly_checkpoint(spa)) { | |
3868 | #endif | |
3869 | spa_load_failed(spa, "a pool with guid %llu is already open", | |
3870 | (u_longlong_t)pool_guid); | |
3871 | return (SET_ERROR(EEXIST)); | |
3872 | } | |
3873 | ||
3874 | spa->spa_config_guid = pool_guid; | |
3875 | ||
3876 | nvlist_free(spa->spa_load_info); | |
3877 | spa->spa_load_info = fnvlist_alloc(); | |
3878 | ||
3879 | ASSERT(spa->spa_comment == NULL); | |
3880 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
3881 | spa->spa_comment = spa_strdup(comment); | |
3882 | ||
3883 | ASSERT(spa->spa_compatibility == NULL); | |
3884 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMPATIBILITY, | |
3885 | &compatibility) == 0) | |
3886 | spa->spa_compatibility = spa_strdup(compatibility); | |
3887 | ||
3888 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
3889 | &spa->spa_config_txg); | |
3890 | ||
3891 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) | |
3892 | spa->spa_config_splitting = fnvlist_dup(nvl); | |
3893 | ||
3894 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { | |
3895 | spa_load_failed(spa, "invalid config provided: '%s' missing", | |
3896 | ZPOOL_CONFIG_VDEV_TREE); | |
3897 | return (SET_ERROR(EINVAL)); | |
3898 | } | |
3899 | ||
3900 | /* | |
3901 | * Create "The Godfather" zio to hold all async IOs | |
3902 | */ | |
3903 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
3904 | KM_SLEEP); | |
3905 | for (int i = 0; i < max_ncpus; i++) { | |
3906 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
3907 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
3908 | ZIO_FLAG_GODFATHER); | |
3909 | } | |
3910 | ||
3911 | /* | |
3912 | * Parse the configuration into a vdev tree. We explicitly set the | |
3913 | * value that will be returned by spa_version() since parsing the | |
3914 | * configuration requires knowing the version number. | |
3915 | */ | |
3916 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3917 | parse = (type == SPA_IMPORT_EXISTING ? | |
3918 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
3919 | error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); | |
3920 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3921 | ||
3922 | if (error != 0) { | |
3923 | spa_load_failed(spa, "unable to parse config [error=%d]", | |
3924 | error); | |
3925 | return (error); | |
3926 | } | |
3927 | ||
3928 | ASSERT(spa->spa_root_vdev == rvd); | |
3929 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); | |
3930 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
3931 | ||
3932 | if (type != SPA_IMPORT_ASSEMBLE) { | |
3933 | ASSERT(spa_guid(spa) == pool_guid); | |
3934 | } | |
3935 | ||
3936 | return (0); | |
3937 | } | |
3938 | ||
3939 | /* | |
3940 | * Recursively open all vdevs in the vdev tree. This function is called twice: | |
3941 | * first with the untrusted config, then with the trusted config. | |
3942 | */ | |
3943 | static int | |
3944 | spa_ld_open_vdevs(spa_t *spa) | |
3945 | { | |
3946 | int error = 0; | |
3947 | ||
3948 | /* | |
3949 | * spa_missing_tvds_allowed defines how many top-level vdevs can be | |
3950 | * missing/unopenable for the root vdev to be still considered openable. | |
3951 | */ | |
3952 | if (spa->spa_trust_config) { | |
3953 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; | |
3954 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { | |
3955 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; | |
3956 | } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { | |
3957 | spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; | |
3958 | } else { | |
3959 | spa->spa_missing_tvds_allowed = 0; | |
3960 | } | |
3961 | ||
3962 | spa->spa_missing_tvds_allowed = | |
3963 | MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); | |
3964 | ||
3965 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3966 | error = vdev_open(spa->spa_root_vdev); | |
3967 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3968 | ||
3969 | if (spa->spa_missing_tvds != 0) { | |
3970 | spa_load_note(spa, "vdev tree has %lld missing top-level " | |
3971 | "vdevs.", (u_longlong_t)spa->spa_missing_tvds); | |
3972 | if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) { | |
3973 | /* | |
3974 | * Although theoretically we could allow users to open | |
3975 | * incomplete pools in RW mode, we'd need to add a lot | |
3976 | * of extra logic (e.g. adjust pool space to account | |
3977 | * for missing vdevs). | |
3978 | * This limitation also prevents users from accidentally | |
3979 | * opening the pool in RW mode during data recovery and | |
3980 | * damaging it further. | |
3981 | */ | |
3982 | spa_load_note(spa, "pools with missing top-level " | |
3983 | "vdevs can only be opened in read-only mode."); | |
3984 | error = SET_ERROR(ENXIO); | |
3985 | } else { | |
3986 | spa_load_note(spa, "current settings allow for maximum " | |
3987 | "%lld missing top-level vdevs at this stage.", | |
3988 | (u_longlong_t)spa->spa_missing_tvds_allowed); | |
3989 | } | |
3990 | } | |
3991 | if (error != 0) { | |
3992 | spa_load_failed(spa, "unable to open vdev tree [error=%d]", | |
3993 | error); | |
3994 | } | |
3995 | if (spa->spa_missing_tvds != 0 || error != 0) | |
3996 | vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); | |
3997 | ||
3998 | return (error); | |
3999 | } | |
4000 | ||
4001 | /* | |
4002 | * We need to validate the vdev labels against the configuration that | |
4003 | * we have in hand. This function is called twice: first with an untrusted | |
4004 | * config, then with a trusted config. The validation is more strict when the | |
4005 | * config is trusted. | |
4006 | */ | |
4007 | static int | |
4008 | spa_ld_validate_vdevs(spa_t *spa) | |
4009 | { | |
4010 | int error = 0; | |
4011 | vdev_t *rvd = spa->spa_root_vdev; | |
4012 | ||
4013 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4014 | error = vdev_validate(rvd); | |
4015 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4016 | ||
4017 | if (error != 0) { | |
4018 | spa_load_failed(spa, "vdev_validate failed [error=%d]", error); | |
4019 | return (error); | |
4020 | } | |
4021 | ||
4022 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { | |
4023 | spa_load_failed(spa, "cannot open vdev tree after invalidating " | |
4024 | "some vdevs"); | |
4025 | vdev_dbgmsg_print_tree(rvd, 2); | |
4026 | return (SET_ERROR(ENXIO)); | |
4027 | } | |
4028 | ||
4029 | return (0); | |
4030 | } | |
4031 | ||
4032 | static void | |
4033 | spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) | |
4034 | { | |
4035 | spa->spa_state = POOL_STATE_ACTIVE; | |
4036 | spa->spa_ubsync = spa->spa_uberblock; | |
4037 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
4038 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
4039 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
4040 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
4041 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
4042 | spa->spa_prev_software_version = ub->ub_software_version; | |
4043 | } | |
4044 | ||
4045 | static int | |
4046 | spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) | |
4047 | { | |
4048 | vdev_t *rvd = spa->spa_root_vdev; | |
4049 | nvlist_t *label; | |
4050 | uberblock_t *ub = &spa->spa_uberblock; | |
4051 | boolean_t activity_check = B_FALSE; | |
4052 | ||
4053 | /* | |
4054 | * If we are opening the checkpointed state of the pool by | |
4055 | * rewinding to it, at this point we will have written the | |
4056 | * checkpointed uberblock to the vdev labels, so searching | |
4057 | * the labels will find the right uberblock. However, if | |
4058 | * we are opening the checkpointed state read-only, we have | |
4059 | * not modified the labels. Therefore, we must ignore the | |
4060 | * labels and continue using the spa_uberblock that was set | |
4061 | * by spa_ld_checkpoint_rewind. | |
4062 | * | |
4063 | * Note that it would be fine to ignore the labels when | |
4064 | * rewinding (opening writeable) as well. However, if we | |
4065 | * crash just after writing the labels, we will end up | |
4066 | * searching the labels. Doing so in the common case means | |
4067 | * that this code path gets exercised normally, rather than | |
4068 | * just in the edge case. | |
4069 | */ | |
4070 | if (ub->ub_checkpoint_txg != 0 && | |
4071 | spa_importing_readonly_checkpoint(spa)) { | |
4072 | spa_ld_select_uberblock_done(spa, ub); | |
4073 | return (0); | |
4074 | } | |
4075 | ||
4076 | /* | |
4077 | * Find the best uberblock. | |
4078 | */ | |
4079 | vdev_uberblock_load(rvd, ub, &label); | |
4080 | ||
4081 | /* | |
4082 | * If we weren't able to find a single valid uberblock, return failure. | |
4083 | */ | |
4084 | if (ub->ub_txg == 0) { | |
4085 | nvlist_free(label); | |
4086 | spa_load_failed(spa, "no valid uberblock found"); | |
4087 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); | |
4088 | } | |
4089 | ||
4090 | if (spa->spa_load_max_txg != UINT64_MAX) { | |
4091 | (void) spa_import_progress_set_max_txg(spa_guid(spa), | |
4092 | (u_longlong_t)spa->spa_load_max_txg); | |
4093 | } | |
4094 | spa_load_note(spa, "using uberblock with txg=%llu", | |
4095 | (u_longlong_t)ub->ub_txg); | |
4096 | if (ub->ub_raidz_reflow_info != 0) { | |
4097 | spa_load_note(spa, "uberblock raidz_reflow_info: " | |
4098 | "state=%u offset=%llu", | |
4099 | (int)RRSS_GET_STATE(ub), | |
4100 | (u_longlong_t)RRSS_GET_OFFSET(ub)); | |
4101 | } | |
4102 | ||
4103 | ||
4104 | /* | |
4105 | * For pools which have the multihost property on determine if the | |
4106 | * pool is truly inactive and can be safely imported. Prevent | |
4107 | * hosts which don't have a hostid set from importing the pool. | |
4108 | */ | |
4109 | activity_check = spa_activity_check_required(spa, ub, label, | |
4110 | spa->spa_config); | |
4111 | if (activity_check) { | |
4112 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && | |
4113 | spa_get_hostid(spa) == 0) { | |
4114 | nvlist_free(label); | |
4115 | fnvlist_add_uint64(spa->spa_load_info, | |
4116 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4117 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4118 | } | |
4119 | ||
4120 | int error = | |
4121 | spa_activity_check(spa, ub, spa->spa_config, B_TRUE); | |
4122 | if (error) { | |
4123 | nvlist_free(label); | |
4124 | return (error); | |
4125 | } | |
4126 | ||
4127 | fnvlist_add_uint64(spa->spa_load_info, | |
4128 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
4129 | fnvlist_add_uint64(spa->spa_load_info, | |
4130 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
4131 | fnvlist_add_uint16(spa->spa_load_info, | |
4132 | ZPOOL_CONFIG_MMP_SEQ, | |
4133 | (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)); | |
4134 | } | |
4135 | ||
4136 | /* | |
4137 | * If the pool has an unsupported version we can't open it. | |
4138 | */ | |
4139 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { | |
4140 | nvlist_free(label); | |
4141 | spa_load_failed(spa, "version %llu is not supported", | |
4142 | (u_longlong_t)ub->ub_version); | |
4143 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); | |
4144 | } | |
4145 | ||
4146 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
4147 | nvlist_t *features; | |
4148 | ||
4149 | /* | |
4150 | * If we weren't able to find what's necessary for reading the | |
4151 | * MOS in the label, return failure. | |
4152 | */ | |
4153 | if (label == NULL) { | |
4154 | spa_load_failed(spa, "label config unavailable"); | |
4155 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
4156 | ENXIO)); | |
4157 | } | |
4158 | ||
4159 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, | |
4160 | &features) != 0) { | |
4161 | nvlist_free(label); | |
4162 | spa_load_failed(spa, "invalid label: '%s' missing", | |
4163 | ZPOOL_CONFIG_FEATURES_FOR_READ); | |
4164 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
4165 | ENXIO)); | |
4166 | } | |
4167 | ||
4168 | /* | |
4169 | * Update our in-core representation with the definitive values | |
4170 | * from the label. | |
4171 | */ | |
4172 | nvlist_free(spa->spa_label_features); | |
4173 | spa->spa_label_features = fnvlist_dup(features); | |
4174 | } | |
4175 | ||
4176 | nvlist_free(label); | |
4177 | ||
4178 | /* | |
4179 | * Look through entries in the label nvlist's features_for_read. If | |
4180 | * there is a feature listed there which we don't understand then we | |
4181 | * cannot open a pool. | |
4182 | */ | |
4183 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
4184 | nvlist_t *unsup_feat; | |
4185 | ||
4186 | unsup_feat = fnvlist_alloc(); | |
4187 | ||
4188 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, | |
4189 | NULL); nvp != NULL; | |
4190 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { | |
4191 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
4192 | fnvlist_add_string(unsup_feat, | |
4193 | nvpair_name(nvp), ""); | |
4194 | } | |
4195 | } | |
4196 | ||
4197 | if (!nvlist_empty(unsup_feat)) { | |
4198 | fnvlist_add_nvlist(spa->spa_load_info, | |
4199 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
4200 | nvlist_free(unsup_feat); | |
4201 | spa_load_failed(spa, "some features are unsupported"); | |
4202 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
4203 | ENOTSUP)); | |
4204 | } | |
4205 | ||
4206 | nvlist_free(unsup_feat); | |
4207 | } | |
4208 | ||
4209 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { | |
4210 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4211 | spa_try_repair(spa, spa->spa_config); | |
4212 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4213 | nvlist_free(spa->spa_config_splitting); | |
4214 | spa->spa_config_splitting = NULL; | |
4215 | } | |
4216 | ||
4217 | /* | |
4218 | * Initialize internal SPA structures. | |
4219 | */ | |
4220 | spa_ld_select_uberblock_done(spa, ub); | |
4221 | ||
4222 | return (0); | |
4223 | } | |
4224 | ||
4225 | static int | |
4226 | spa_ld_open_rootbp(spa_t *spa) | |
4227 | { | |
4228 | int error = 0; | |
4229 | vdev_t *rvd = spa->spa_root_vdev; | |
4230 | ||
4231 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); | |
4232 | if (error != 0) { | |
4233 | spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " | |
4234 | "[error=%d]", error); | |
4235 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4236 | } | |
4237 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; | |
4238 | ||
4239 | return (0); | |
4240 | } | |
4241 | ||
4242 | static int | |
4243 | spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, | |
4244 | boolean_t reloading) | |
4245 | { | |
4246 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; | |
4247 | nvlist_t *nv, *mos_config, *policy; | |
4248 | int error = 0, copy_error; | |
4249 | uint64_t healthy_tvds, healthy_tvds_mos; | |
4250 | uint64_t mos_config_txg; | |
4251 | ||
4252 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) | |
4253 | != 0) | |
4254 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4255 | ||
4256 | /* | |
4257 | * If we're assembling a pool from a split, the config provided is | |
4258 | * already trusted so there is nothing to do. | |
4259 | */ | |
4260 | if (type == SPA_IMPORT_ASSEMBLE) | |
4261 | return (0); | |
4262 | ||
4263 | healthy_tvds = spa_healthy_core_tvds(spa); | |
4264 | ||
4265 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) | |
4266 | != 0) { | |
4267 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
4268 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4269 | } | |
4270 | ||
4271 | /* | |
4272 | * If we are doing an open, pool owner wasn't verified yet, thus do | |
4273 | * the verification here. | |
4274 | */ | |
4275 | if (spa->spa_load_state == SPA_LOAD_OPEN) { | |
4276 | error = spa_verify_host(spa, mos_config); | |
4277 | if (error != 0) { | |
4278 | nvlist_free(mos_config); | |
4279 | return (error); | |
4280 | } | |
4281 | } | |
4282 | ||
4283 | nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); | |
4284 | ||
4285 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4286 | ||
4287 | /* | |
4288 | * Build a new vdev tree from the trusted config | |
4289 | */ | |
4290 | error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD); | |
4291 | if (error != 0) { | |
4292 | nvlist_free(mos_config); | |
4293 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4294 | spa_load_failed(spa, "spa_config_parse failed [error=%d]", | |
4295 | error); | |
4296 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4297 | } | |
4298 | ||
4299 | /* | |
4300 | * Vdev paths in the MOS may be obsolete. If the untrusted config was | |
4301 | * obtained by scanning /dev/dsk, then it will have the right vdev | |
4302 | * paths. We update the trusted MOS config with this information. | |
4303 | * We first try to copy the paths with vdev_copy_path_strict, which | |
4304 | * succeeds only when both configs have exactly the same vdev tree. | |
4305 | * If that fails, we fall back to a more flexible method that has a | |
4306 | * best effort policy. | |
4307 | */ | |
4308 | copy_error = vdev_copy_path_strict(rvd, mrvd); | |
4309 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
4310 | spa_load_note(spa, "provided vdev tree:"); | |
4311 | vdev_dbgmsg_print_tree(rvd, 2); | |
4312 | spa_load_note(spa, "MOS vdev tree:"); | |
4313 | vdev_dbgmsg_print_tree(mrvd, 2); | |
4314 | } | |
4315 | if (copy_error != 0) { | |
4316 | spa_load_note(spa, "vdev_copy_path_strict failed, falling " | |
4317 | "back to vdev_copy_path_relaxed"); | |
4318 | vdev_copy_path_relaxed(rvd, mrvd); | |
4319 | } | |
4320 | ||
4321 | vdev_close(rvd); | |
4322 | vdev_free(rvd); | |
4323 | spa->spa_root_vdev = mrvd; | |
4324 | rvd = mrvd; | |
4325 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4326 | ||
4327 | /* | |
4328 | * If 'zpool import' used a cached config, then the on-disk hostid and | |
4329 | * hostname may be different to the cached config in ways that should | |
4330 | * prevent import. Userspace can't discover this without a scan, but | |
4331 | * we know, so we add these values to LOAD_INFO so the caller can know | |
4332 | * the difference. | |
4333 | * | |
4334 | * Note that we have to do this before the config is regenerated, | |
4335 | * because the new config will have the hostid and hostname for this | |
4336 | * host, in readiness for import. | |
4337 | */ | |
4338 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTID)) | |
4339 | fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_HOSTID, | |
4340 | fnvlist_lookup_uint64(mos_config, ZPOOL_CONFIG_HOSTID)); | |
4341 | if (nvlist_exists(mos_config, ZPOOL_CONFIG_HOSTNAME)) | |
4342 | fnvlist_add_string(spa->spa_load_info, ZPOOL_CONFIG_HOSTNAME, | |
4343 | fnvlist_lookup_string(mos_config, ZPOOL_CONFIG_HOSTNAME)); | |
4344 | ||
4345 | /* | |
4346 | * We will use spa_config if we decide to reload the spa or if spa_load | |
4347 | * fails and we rewind. We must thus regenerate the config using the | |
4348 | * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to | |
4349 | * pass settings on how to load the pool and is not stored in the MOS. | |
4350 | * We copy it over to our new, trusted config. | |
4351 | */ | |
4352 | mos_config_txg = fnvlist_lookup_uint64(mos_config, | |
4353 | ZPOOL_CONFIG_POOL_TXG); | |
4354 | nvlist_free(mos_config); | |
4355 | mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); | |
4356 | if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, | |
4357 | &policy) == 0) | |
4358 | fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); | |
4359 | spa_config_set(spa, mos_config); | |
4360 | spa->spa_config_source = SPA_CONFIG_SRC_MOS; | |
4361 | ||
4362 | /* | |
4363 | * Now that we got the config from the MOS, we should be more strict | |
4364 | * in checking blkptrs and can make assumptions about the consistency | |
4365 | * of the vdev tree. spa_trust_config must be set to true before opening | |
4366 | * vdevs in order for them to be writeable. | |
4367 | */ | |
4368 | spa->spa_trust_config = B_TRUE; | |
4369 | ||
4370 | /* | |
4371 | * Open and validate the new vdev tree | |
4372 | */ | |
4373 | error = spa_ld_open_vdevs(spa); | |
4374 | if (error != 0) | |
4375 | return (error); | |
4376 | ||
4377 | error = spa_ld_validate_vdevs(spa); | |
4378 | if (error != 0) | |
4379 | return (error); | |
4380 | ||
4381 | if (copy_error != 0 || spa_load_print_vdev_tree) { | |
4382 | spa_load_note(spa, "final vdev tree:"); | |
4383 | vdev_dbgmsg_print_tree(rvd, 2); | |
4384 | } | |
4385 | ||
4386 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && | |
4387 | !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { | |
4388 | /* | |
4389 | * Sanity check to make sure that we are indeed loading the | |
4390 | * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds | |
4391 | * in the config provided and they happened to be the only ones | |
4392 | * to have the latest uberblock, we could involuntarily perform | |
4393 | * an extreme rewind. | |
4394 | */ | |
4395 | healthy_tvds_mos = spa_healthy_core_tvds(spa); | |
4396 | if (healthy_tvds_mos - healthy_tvds >= | |
4397 | SPA_SYNC_MIN_VDEVS) { | |
4398 | spa_load_note(spa, "config provided misses too many " | |
4399 | "top-level vdevs compared to MOS (%lld vs %lld). ", | |
4400 | (u_longlong_t)healthy_tvds, | |
4401 | (u_longlong_t)healthy_tvds_mos); | |
4402 | spa_load_note(spa, "vdev tree:"); | |
4403 | vdev_dbgmsg_print_tree(rvd, 2); | |
4404 | if (reloading) { | |
4405 | spa_load_failed(spa, "config was already " | |
4406 | "provided from MOS. Aborting."); | |
4407 | return (spa_vdev_err(rvd, | |
4408 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4409 | } | |
4410 | spa_load_note(spa, "spa must be reloaded using MOS " | |
4411 | "config"); | |
4412 | return (SET_ERROR(EAGAIN)); | |
4413 | } | |
4414 | } | |
4415 | ||
4416 | error = spa_check_for_missing_logs(spa); | |
4417 | if (error != 0) | |
4418 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
4419 | ||
4420 | if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { | |
4421 | spa_load_failed(spa, "uberblock guid sum doesn't match MOS " | |
4422 | "guid sum (%llu != %llu)", | |
4423 | (u_longlong_t)spa->spa_uberblock.ub_guid_sum, | |
4424 | (u_longlong_t)rvd->vdev_guid_sum); | |
4425 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
4426 | ENXIO)); | |
4427 | } | |
4428 | ||
4429 | return (0); | |
4430 | } | |
4431 | ||
4432 | static int | |
4433 | spa_ld_open_indirect_vdev_metadata(spa_t *spa) | |
4434 | { | |
4435 | int error = 0; | |
4436 | vdev_t *rvd = spa->spa_root_vdev; | |
4437 | ||
4438 | /* | |
4439 | * Everything that we read before spa_remove_init() must be stored | |
4440 | * on concreted vdevs. Therefore we do this as early as possible. | |
4441 | */ | |
4442 | error = spa_remove_init(spa); | |
4443 | if (error != 0) { | |
4444 | spa_load_failed(spa, "spa_remove_init failed [error=%d]", | |
4445 | error); | |
4446 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4447 | } | |
4448 | ||
4449 | /* | |
4450 | * Retrieve information needed to condense indirect vdev mappings. | |
4451 | */ | |
4452 | error = spa_condense_init(spa); | |
4453 | if (error != 0) { | |
4454 | spa_load_failed(spa, "spa_condense_init failed [error=%d]", | |
4455 | error); | |
4456 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4457 | } | |
4458 | ||
4459 | return (0); | |
4460 | } | |
4461 | ||
4462 | static int | |
4463 | spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) | |
4464 | { | |
4465 | int error = 0; | |
4466 | vdev_t *rvd = spa->spa_root_vdev; | |
4467 | ||
4468 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { | |
4469 | boolean_t missing_feat_read = B_FALSE; | |
4470 | nvlist_t *unsup_feat, *enabled_feat; | |
4471 | ||
4472 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
4473 | &spa->spa_feat_for_read_obj, B_TRUE) != 0) { | |
4474 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4475 | } | |
4476 | ||
4477 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
4478 | &spa->spa_feat_for_write_obj, B_TRUE) != 0) { | |
4479 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4480 | } | |
4481 | ||
4482 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
4483 | &spa->spa_feat_desc_obj, B_TRUE) != 0) { | |
4484 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4485 | } | |
4486 | ||
4487 | enabled_feat = fnvlist_alloc(); | |
4488 | unsup_feat = fnvlist_alloc(); | |
4489 | ||
4490 | if (!spa_features_check(spa, B_FALSE, | |
4491 | unsup_feat, enabled_feat)) | |
4492 | missing_feat_read = B_TRUE; | |
4493 | ||
4494 | if (spa_writeable(spa) || | |
4495 | spa->spa_load_state == SPA_LOAD_TRYIMPORT) { | |
4496 | if (!spa_features_check(spa, B_TRUE, | |
4497 | unsup_feat, enabled_feat)) { | |
4498 | *missing_feat_writep = B_TRUE; | |
4499 | } | |
4500 | } | |
4501 | ||
4502 | fnvlist_add_nvlist(spa->spa_load_info, | |
4503 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
4504 | ||
4505 | if (!nvlist_empty(unsup_feat)) { | |
4506 | fnvlist_add_nvlist(spa->spa_load_info, | |
4507 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
4508 | } | |
4509 | ||
4510 | fnvlist_free(enabled_feat); | |
4511 | fnvlist_free(unsup_feat); | |
4512 | ||
4513 | if (!missing_feat_read) { | |
4514 | fnvlist_add_boolean(spa->spa_load_info, | |
4515 | ZPOOL_CONFIG_CAN_RDONLY); | |
4516 | } | |
4517 | ||
4518 | /* | |
4519 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
4520 | * twofold: to determine whether the pool is available for | |
4521 | * import in read-write mode and (if it is not) whether the | |
4522 | * pool is available for import in read-only mode. If the pool | |
4523 | * is available for import in read-write mode, it is displayed | |
4524 | * as available in userland; if it is not available for import | |
4525 | * in read-only mode, it is displayed as unavailable in | |
4526 | * userland. If the pool is available for import in read-only | |
4527 | * mode but not read-write mode, it is displayed as unavailable | |
4528 | * in userland with a special note that the pool is actually | |
4529 | * available for open in read-only mode. | |
4530 | * | |
4531 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
4532 | * missing a feature for write, we must first determine whether | |
4533 | * the pool can be opened read-only before returning to | |
4534 | * userland in order to know whether to display the | |
4535 | * abovementioned note. | |
4536 | */ | |
4537 | if (missing_feat_read || (*missing_feat_writep && | |
4538 | spa_writeable(spa))) { | |
4539 | spa_load_failed(spa, "pool uses unsupported features"); | |
4540 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
4541 | ENOTSUP)); | |
4542 | } | |
4543 | ||
4544 | /* | |
4545 | * Load refcounts for ZFS features from disk into an in-memory | |
4546 | * cache during SPA initialization. | |
4547 | */ | |
4548 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { | |
4549 | uint64_t refcount; | |
4550 | ||
4551 | error = feature_get_refcount_from_disk(spa, | |
4552 | &spa_feature_table[i], &refcount); | |
4553 | if (error == 0) { | |
4554 | spa->spa_feat_refcount_cache[i] = refcount; | |
4555 | } else if (error == ENOTSUP) { | |
4556 | spa->spa_feat_refcount_cache[i] = | |
4557 | SPA_FEATURE_DISABLED; | |
4558 | } else { | |
4559 | spa_load_failed(spa, "error getting refcount " | |
4560 | "for feature %s [error=%d]", | |
4561 | spa_feature_table[i].fi_guid, error); | |
4562 | return (spa_vdev_err(rvd, | |
4563 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
4564 | } | |
4565 | } | |
4566 | } | |
4567 | ||
4568 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
4569 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
4570 | &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) | |
4571 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4572 | } | |
4573 | ||
4574 | /* | |
4575 | * Encryption was added before bookmark_v2, even though bookmark_v2 | |
4576 | * is now a dependency. If this pool has encryption enabled without | |
4577 | * bookmark_v2, trigger an errata message. | |
4578 | */ | |
4579 | if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) && | |
4580 | !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) { | |
4581 | spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; | |
4582 | } | |
4583 | ||
4584 | return (0); | |
4585 | } | |
4586 | ||
4587 | static int | |
4588 | spa_ld_load_special_directories(spa_t *spa) | |
4589 | { | |
4590 | int error = 0; | |
4591 | vdev_t *rvd = spa->spa_root_vdev; | |
4592 | ||
4593 | spa->spa_is_initializing = B_TRUE; | |
4594 | error = dsl_pool_open(spa->spa_dsl_pool); | |
4595 | spa->spa_is_initializing = B_FALSE; | |
4596 | if (error != 0) { | |
4597 | spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); | |
4598 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4599 | } | |
4600 | ||
4601 | return (0); | |
4602 | } | |
4603 | ||
4604 | static int | |
4605 | spa_ld_get_props(spa_t *spa) | |
4606 | { | |
4607 | int error = 0; | |
4608 | uint64_t obj; | |
4609 | vdev_t *rvd = spa->spa_root_vdev; | |
4610 | ||
4611 | /* Grab the checksum salt from the MOS. */ | |
4612 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
4613 | DMU_POOL_CHECKSUM_SALT, 1, | |
4614 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
4615 | spa->spa_cksum_salt.zcs_bytes); | |
4616 | if (error == ENOENT) { | |
4617 | /* Generate a new salt for subsequent use */ | |
4618 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4619 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4620 | } else if (error != 0) { | |
4621 | spa_load_failed(spa, "unable to retrieve checksum salt from " | |
4622 | "MOS [error=%d]", error); | |
4623 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4624 | } | |
4625 | ||
4626 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) | |
4627 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4628 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
4629 | if (error != 0) { | |
4630 | spa_load_failed(spa, "error opening deferred-frees bpobj " | |
4631 | "[error=%d]", error); | |
4632 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4633 | } | |
4634 | ||
4635 | /* | |
4636 | * Load the bit that tells us to use the new accounting function | |
4637 | * (raid-z deflation). If we have an older pool, this will not | |
4638 | * be present. | |
4639 | */ | |
4640 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); | |
4641 | if (error != 0 && error != ENOENT) | |
4642 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4643 | ||
4644 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
4645 | &spa->spa_creation_version, B_FALSE); | |
4646 | if (error != 0 && error != ENOENT) | |
4647 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4648 | ||
4649 | /* | |
4650 | * Load the persistent error log. If we have an older pool, this will | |
4651 | * not be present. | |
4652 | */ | |
4653 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, | |
4654 | B_FALSE); | |
4655 | if (error != 0 && error != ENOENT) | |
4656 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4657 | ||
4658 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, | |
4659 | &spa->spa_errlog_scrub, B_FALSE); | |
4660 | if (error != 0 && error != ENOENT) | |
4661 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4662 | ||
4663 | /* | |
4664 | * Load the livelist deletion field. If a livelist is queued for | |
4665 | * deletion, indicate that in the spa | |
4666 | */ | |
4667 | error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES, | |
4668 | &spa->spa_livelists_to_delete, B_FALSE); | |
4669 | if (error != 0 && error != ENOENT) | |
4670 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4671 | ||
4672 | /* | |
4673 | * Load the history object. If we have an older pool, this | |
4674 | * will not be present. | |
4675 | */ | |
4676 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); | |
4677 | if (error != 0 && error != ENOENT) | |
4678 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4679 | ||
4680 | /* | |
4681 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
4682 | * be present; in this case, defer its creation to a later time to | |
4683 | * avoid dirtying the MOS this early / out of sync context. See | |
4684 | * spa_sync_config_object. | |
4685 | */ | |
4686 | ||
4687 | /* The sentinel is only available in the MOS config. */ | |
4688 | nvlist_t *mos_config; | |
4689 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { | |
4690 | spa_load_failed(spa, "unable to retrieve MOS config"); | |
4691 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4692 | } | |
4693 | ||
4694 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
4695 | &spa->spa_all_vdev_zaps, B_FALSE); | |
4696 | ||
4697 | if (error == ENOENT) { | |
4698 | VERIFY(!nvlist_exists(mos_config, | |
4699 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
4700 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
4701 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4702 | } else if (error != 0) { | |
4703 | nvlist_free(mos_config); | |
4704 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4705 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { | |
4706 | /* | |
4707 | * An older version of ZFS overwrote the sentinel value, so | |
4708 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
4709 | * destruction to later; see spa_sync_config_object. | |
4710 | */ | |
4711 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
4712 | /* | |
4713 | * We're assuming that no vdevs have had their ZAPs created | |
4714 | * before this. Better be sure of it. | |
4715 | */ | |
4716 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
4717 | } | |
4718 | nvlist_free(mos_config); | |
4719 | ||
4720 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
4721 | ||
4722 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, | |
4723 | B_FALSE); | |
4724 | if (error && error != ENOENT) | |
4725 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4726 | ||
4727 | if (error == 0) { | |
4728 | uint64_t autoreplace = 0; | |
4729 | ||
4730 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
4731 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
4732 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
4733 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
4734 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
4735 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); | |
4736 | spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim); | |
4737 | spa->spa_autoreplace = (autoreplace != 0); | |
4738 | } | |
4739 | ||
4740 | /* | |
4741 | * If we are importing a pool with missing top-level vdevs, | |
4742 | * we enforce that the pool doesn't panic or get suspended on | |
4743 | * error since the likelihood of missing data is extremely high. | |
4744 | */ | |
4745 | if (spa->spa_missing_tvds > 0 && | |
4746 | spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && | |
4747 | spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4748 | spa_load_note(spa, "forcing failmode to 'continue' " | |
4749 | "as some top level vdevs are missing"); | |
4750 | spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; | |
4751 | } | |
4752 | ||
4753 | return (0); | |
4754 | } | |
4755 | ||
4756 | static int | |
4757 | spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) | |
4758 | { | |
4759 | int error = 0; | |
4760 | vdev_t *rvd = spa->spa_root_vdev; | |
4761 | ||
4762 | /* | |
4763 | * If we're assembling the pool from the split-off vdevs of | |
4764 | * an existing pool, we don't want to attach the spares & cache | |
4765 | * devices. | |
4766 | */ | |
4767 | ||
4768 | /* | |
4769 | * Load any hot spares for this pool. | |
4770 | */ | |
4771 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, | |
4772 | B_FALSE); | |
4773 | if (error != 0 && error != ENOENT) | |
4774 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4775 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
4776 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); | |
4777 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
4778 | &spa->spa_spares.sav_config) != 0) { | |
4779 | spa_load_failed(spa, "error loading spares nvlist"); | |
4780 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4781 | } | |
4782 | ||
4783 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4784 | spa_load_spares(spa); | |
4785 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4786 | } else if (error == 0) { | |
4787 | spa->spa_spares.sav_sync = B_TRUE; | |
4788 | } | |
4789 | ||
4790 | /* | |
4791 | * Load any level 2 ARC devices for this pool. | |
4792 | */ | |
4793 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, | |
4794 | &spa->spa_l2cache.sav_object, B_FALSE); | |
4795 | if (error != 0 && error != ENOENT) | |
4796 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4797 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
4798 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); | |
4799 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
4800 | &spa->spa_l2cache.sav_config) != 0) { | |
4801 | spa_load_failed(spa, "error loading l2cache nvlist"); | |
4802 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4803 | } | |
4804 | ||
4805 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4806 | spa_load_l2cache(spa); | |
4807 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4808 | } else if (error == 0) { | |
4809 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4810 | } | |
4811 | ||
4812 | return (0); | |
4813 | } | |
4814 | ||
4815 | static int | |
4816 | spa_ld_load_vdev_metadata(spa_t *spa) | |
4817 | { | |
4818 | int error = 0; | |
4819 | vdev_t *rvd = spa->spa_root_vdev; | |
4820 | ||
4821 | /* | |
4822 | * If the 'multihost' property is set, then never allow a pool to | |
4823 | * be imported when the system hostid is zero. The exception to | |
4824 | * this rule is zdb which is always allowed to access pools. | |
4825 | */ | |
4826 | if (spa_multihost(spa) && spa_get_hostid(spa) == 0 && | |
4827 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { | |
4828 | fnvlist_add_uint64(spa->spa_load_info, | |
4829 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
4830 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
4831 | } | |
4832 | ||
4833 | /* | |
4834 | * If the 'autoreplace' property is set, then post a resource notifying | |
4835 | * the ZFS DE that it should not issue any faults for unopenable | |
4836 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
4837 | * unopenable vdevs so that the normal autoreplace handler can take | |
4838 | * over. | |
4839 | */ | |
4840 | if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4841 | spa_check_removed(spa->spa_root_vdev); | |
4842 | /* | |
4843 | * For the import case, this is done in spa_import(), because | |
4844 | * at this point we're using the spare definitions from | |
4845 | * the MOS config, not necessarily from the userland config. | |
4846 | */ | |
4847 | if (spa->spa_load_state != SPA_LOAD_IMPORT) { | |
4848 | spa_aux_check_removed(&spa->spa_spares); | |
4849 | spa_aux_check_removed(&spa->spa_l2cache); | |
4850 | } | |
4851 | } | |
4852 | ||
4853 | /* | |
4854 | * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. | |
4855 | */ | |
4856 | error = vdev_load(rvd); | |
4857 | if (error != 0) { | |
4858 | spa_load_failed(spa, "vdev_load failed [error=%d]", error); | |
4859 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4860 | } | |
4861 | ||
4862 | error = spa_ld_log_spacemaps(spa); | |
4863 | if (error != 0) { | |
4864 | spa_load_failed(spa, "spa_ld_log_spacemaps failed [error=%d]", | |
4865 | error); | |
4866 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
4867 | } | |
4868 | ||
4869 | /* | |
4870 | * Propagate the leaf DTLs we just loaded all the way up the vdev tree. | |
4871 | */ | |
4872 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4873 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE); | |
4874 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4875 | ||
4876 | return (0); | |
4877 | } | |
4878 | ||
4879 | static int | |
4880 | spa_ld_load_dedup_tables(spa_t *spa) | |
4881 | { | |
4882 | int error = 0; | |
4883 | vdev_t *rvd = spa->spa_root_vdev; | |
4884 | ||
4885 | error = ddt_load(spa); | |
4886 | if (error != 0) { | |
4887 | spa_load_failed(spa, "ddt_load failed [error=%d]", error); | |
4888 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4889 | } | |
4890 | ||
4891 | return (0); | |
4892 | } | |
4893 | ||
4894 | static int | |
4895 | spa_ld_load_brt(spa_t *spa) | |
4896 | { | |
4897 | int error = 0; | |
4898 | vdev_t *rvd = spa->spa_root_vdev; | |
4899 | ||
4900 | error = brt_load(spa); | |
4901 | if (error != 0) { | |
4902 | spa_load_failed(spa, "brt_load failed [error=%d]", error); | |
4903 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
4904 | } | |
4905 | ||
4906 | return (0); | |
4907 | } | |
4908 | ||
4909 | static int | |
4910 | spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, const char **ereport) | |
4911 | { | |
4912 | vdev_t *rvd = spa->spa_root_vdev; | |
4913 | ||
4914 | if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { | |
4915 | boolean_t missing = spa_check_logs(spa); | |
4916 | if (missing) { | |
4917 | if (spa->spa_missing_tvds != 0) { | |
4918 | spa_load_note(spa, "spa_check_logs failed " | |
4919 | "so dropping the logs"); | |
4920 | } else { | |
4921 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
4922 | spa_load_failed(spa, "spa_check_logs failed"); | |
4923 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, | |
4924 | ENXIO)); | |
4925 | } | |
4926 | } | |
4927 | } | |
4928 | ||
4929 | return (0); | |
4930 | } | |
4931 | ||
4932 | static int | |
4933 | spa_ld_verify_pool_data(spa_t *spa) | |
4934 | { | |
4935 | int error = 0; | |
4936 | vdev_t *rvd = spa->spa_root_vdev; | |
4937 | ||
4938 | /* | |
4939 | * We've successfully opened the pool, verify that we're ready | |
4940 | * to start pushing transactions. | |
4941 | */ | |
4942 | if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { | |
4943 | error = spa_load_verify(spa); | |
4944 | if (error != 0) { | |
4945 | spa_load_failed(spa, "spa_load_verify failed " | |
4946 | "[error=%d]", error); | |
4947 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
4948 | error)); | |
4949 | } | |
4950 | } | |
4951 | ||
4952 | return (0); | |
4953 | } | |
4954 | ||
4955 | static void | |
4956 | spa_ld_claim_log_blocks(spa_t *spa) | |
4957 | { | |
4958 | dmu_tx_t *tx; | |
4959 | dsl_pool_t *dp = spa_get_dsl(spa); | |
4960 | ||
4961 | /* | |
4962 | * Claim log blocks that haven't been committed yet. | |
4963 | * This must all happen in a single txg. | |
4964 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
4965 | * invoked from zil_claim_log_block()'s i/o done callback. | |
4966 | * Price of rollback is that we abandon the log. | |
4967 | */ | |
4968 | spa->spa_claiming = B_TRUE; | |
4969 | ||
4970 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
4971 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
4972 | zil_claim, tx, DS_FIND_CHILDREN); | |
4973 | dmu_tx_commit(tx); | |
4974 | ||
4975 | spa->spa_claiming = B_FALSE; | |
4976 | ||
4977 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
4978 | } | |
4979 | ||
4980 | static void | |
4981 | spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, | |
4982 | boolean_t update_config_cache) | |
4983 | { | |
4984 | vdev_t *rvd = spa->spa_root_vdev; | |
4985 | int need_update = B_FALSE; | |
4986 | ||
4987 | /* | |
4988 | * If the config cache is stale, or we have uninitialized | |
4989 | * metaslabs (see spa_vdev_add()), then update the config. | |
4990 | * | |
4991 | * If this is a verbatim import, trust the current | |
4992 | * in-core spa_config and update the disk labels. | |
4993 | */ | |
4994 | if (update_config_cache || config_cache_txg != spa->spa_config_txg || | |
4995 | spa->spa_load_state == SPA_LOAD_IMPORT || | |
4996 | spa->spa_load_state == SPA_LOAD_RECOVER || | |
4997 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) | |
4998 | need_update = B_TRUE; | |
4999 | ||
5000 | for (int c = 0; c < rvd->vdev_children; c++) | |
5001 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
5002 | need_update = B_TRUE; | |
5003 | ||
5004 | /* | |
5005 | * Update the config cache asynchronously in case we're the | |
5006 | * root pool, in which case the config cache isn't writable yet. | |
5007 | */ | |
5008 | if (need_update) | |
5009 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
5010 | } | |
5011 | ||
5012 | static void | |
5013 | spa_ld_prepare_for_reload(spa_t *spa) | |
5014 | { | |
5015 | spa_mode_t mode = spa->spa_mode; | |
5016 | int async_suspended = spa->spa_async_suspended; | |
5017 | ||
5018 | spa_unload(spa); | |
5019 | spa_deactivate(spa); | |
5020 | spa_activate(spa, mode); | |
5021 | ||
5022 | /* | |
5023 | * We save the value of spa_async_suspended as it gets reset to 0 by | |
5024 | * spa_unload(). We want to restore it back to the original value before | |
5025 | * returning as we might be calling spa_async_resume() later. | |
5026 | */ | |
5027 | spa->spa_async_suspended = async_suspended; | |
5028 | } | |
5029 | ||
5030 | static int | |
5031 | spa_ld_read_checkpoint_txg(spa_t *spa) | |
5032 | { | |
5033 | uberblock_t checkpoint; | |
5034 | int error = 0; | |
5035 | ||
5036 | ASSERT0(spa->spa_checkpoint_txg); | |
5037 | ASSERT(MUTEX_HELD(&spa_namespace_lock) || | |
5038 | spa->spa_load_thread == curthread); | |
5039 | ||
5040 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
5041 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
5042 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
5043 | ||
5044 | if (error == ENOENT) | |
5045 | return (0); | |
5046 | ||
5047 | if (error != 0) | |
5048 | return (error); | |
5049 | ||
5050 | ASSERT3U(checkpoint.ub_txg, !=, 0); | |
5051 | ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); | |
5052 | ASSERT3U(checkpoint.ub_timestamp, !=, 0); | |
5053 | spa->spa_checkpoint_txg = checkpoint.ub_txg; | |
5054 | spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; | |
5055 | ||
5056 | return (0); | |
5057 | } | |
5058 | ||
5059 | static int | |
5060 | spa_ld_mos_init(spa_t *spa, spa_import_type_t type) | |
5061 | { | |
5062 | int error = 0; | |
5063 | ||
5064 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5065 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); | |
5066 | ||
5067 | /* | |
5068 | * Never trust the config that is provided unless we are assembling | |
5069 | * a pool following a split. | |
5070 | * This means don't trust blkptrs and the vdev tree in general. This | |
5071 | * also effectively puts the spa in read-only mode since | |
5072 | * spa_writeable() checks for spa_trust_config to be true. | |
5073 | * We will later load a trusted config from the MOS. | |
5074 | */ | |
5075 | if (type != SPA_IMPORT_ASSEMBLE) | |
5076 | spa->spa_trust_config = B_FALSE; | |
5077 | ||
5078 | /* | |
5079 | * Parse the config provided to create a vdev tree. | |
5080 | */ | |
5081 | error = spa_ld_parse_config(spa, type); | |
5082 | if (error != 0) | |
5083 | return (error); | |
5084 | ||
5085 | spa_import_progress_add(spa); | |
5086 | ||
5087 | /* | |
5088 | * Now that we have the vdev tree, try to open each vdev. This involves | |
5089 | * opening the underlying physical device, retrieving its geometry and | |
5090 | * probing the vdev with a dummy I/O. The state of each vdev will be set | |
5091 | * based on the success of those operations. After this we'll be ready | |
5092 | * to read from the vdevs. | |
5093 | */ | |
5094 | error = spa_ld_open_vdevs(spa); | |
5095 | if (error != 0) | |
5096 | return (error); | |
5097 | ||
5098 | /* | |
5099 | * Read the label of each vdev and make sure that the GUIDs stored | |
5100 | * there match the GUIDs in the config provided. | |
5101 | * If we're assembling a new pool that's been split off from an | |
5102 | * existing pool, the labels haven't yet been updated so we skip | |
5103 | * validation for now. | |
5104 | */ | |
5105 | if (type != SPA_IMPORT_ASSEMBLE) { | |
5106 | error = spa_ld_validate_vdevs(spa); | |
5107 | if (error != 0) | |
5108 | return (error); | |
5109 | } | |
5110 | ||
5111 | /* | |
5112 | * Read all vdev labels to find the best uberblock (i.e. latest, | |
5113 | * unless spa_load_max_txg is set) and store it in spa_uberblock. We | |
5114 | * get the list of features required to read blkptrs in the MOS from | |
5115 | * the vdev label with the best uberblock and verify that our version | |
5116 | * of zfs supports them all. | |
5117 | */ | |
5118 | error = spa_ld_select_uberblock(spa, type); | |
5119 | if (error != 0) | |
5120 | return (error); | |
5121 | ||
5122 | /* | |
5123 | * Pass that uberblock to the dsl_pool layer which will open the root | |
5124 | * blkptr. This blkptr points to the latest version of the MOS and will | |
5125 | * allow us to read its contents. | |
5126 | */ | |
5127 | error = spa_ld_open_rootbp(spa); | |
5128 | if (error != 0) | |
5129 | return (error); | |
5130 | ||
5131 | return (0); | |
5132 | } | |
5133 | ||
5134 | static int | |
5135 | spa_ld_checkpoint_rewind(spa_t *spa) | |
5136 | { | |
5137 | uberblock_t checkpoint; | |
5138 | int error = 0; | |
5139 | ||
5140 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5141 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
5142 | ||
5143 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
5144 | DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), | |
5145 | sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); | |
5146 | ||
5147 | if (error != 0) { | |
5148 | spa_load_failed(spa, "unable to retrieve checkpointed " | |
5149 | "uberblock from the MOS config [error=%d]", error); | |
5150 | ||
5151 | if (error == ENOENT) | |
5152 | error = ZFS_ERR_NO_CHECKPOINT; | |
5153 | ||
5154 | return (error); | |
5155 | } | |
5156 | ||
5157 | ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); | |
5158 | ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); | |
5159 | ||
5160 | /* | |
5161 | * We need to update the txg and timestamp of the checkpointed | |
5162 | * uberblock to be higher than the latest one. This ensures that | |
5163 | * the checkpointed uberblock is selected if we were to close and | |
5164 | * reopen the pool right after we've written it in the vdev labels. | |
5165 | * (also see block comment in vdev_uberblock_compare) | |
5166 | */ | |
5167 | checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; | |
5168 | checkpoint.ub_timestamp = gethrestime_sec(); | |
5169 | ||
5170 | /* | |
5171 | * Set current uberblock to be the checkpointed uberblock. | |
5172 | */ | |
5173 | spa->spa_uberblock = checkpoint; | |
5174 | ||
5175 | /* | |
5176 | * If we are doing a normal rewind, then the pool is open for | |
5177 | * writing and we sync the "updated" checkpointed uberblock to | |
5178 | * disk. Once this is done, we've basically rewound the whole | |
5179 | * pool and there is no way back. | |
5180 | * | |
5181 | * There are cases when we don't want to attempt and sync the | |
5182 | * checkpointed uberblock to disk because we are opening a | |
5183 | * pool as read-only. Specifically, verifying the checkpointed | |
5184 | * state with zdb, and importing the checkpointed state to get | |
5185 | * a "preview" of its content. | |
5186 | */ | |
5187 | if (spa_writeable(spa)) { | |
5188 | vdev_t *rvd = spa->spa_root_vdev; | |
5189 | ||
5190 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5191 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; | |
5192 | int svdcount = 0; | |
5193 | int children = rvd->vdev_children; | |
5194 | int c0 = random_in_range(children); | |
5195 | ||
5196 | for (int c = 0; c < children; c++) { | |
5197 | vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; | |
5198 | ||
5199 | /* Stop when revisiting the first vdev */ | |
5200 | if (c > 0 && svd[0] == vd) | |
5201 | break; | |
5202 | ||
5203 | if (vd->vdev_ms_array == 0 || vd->vdev_islog || | |
5204 | !vdev_is_concrete(vd)) | |
5205 | continue; | |
5206 | ||
5207 | svd[svdcount++] = vd; | |
5208 | if (svdcount == SPA_SYNC_MIN_VDEVS) | |
5209 | break; | |
5210 | } | |
5211 | error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); | |
5212 | if (error == 0) | |
5213 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
5214 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5215 | ||
5216 | if (error != 0) { | |
5217 | spa_load_failed(spa, "failed to write checkpointed " | |
5218 | "uberblock to the vdev labels [error=%d]", error); | |
5219 | return (error); | |
5220 | } | |
5221 | } | |
5222 | ||
5223 | return (0); | |
5224 | } | |
5225 | ||
5226 | static int | |
5227 | spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, | |
5228 | boolean_t *update_config_cache) | |
5229 | { | |
5230 | int error; | |
5231 | ||
5232 | /* | |
5233 | * Parse the config for pool, open and validate vdevs, | |
5234 | * select an uberblock, and use that uberblock to open | |
5235 | * the MOS. | |
5236 | */ | |
5237 | error = spa_ld_mos_init(spa, type); | |
5238 | if (error != 0) | |
5239 | return (error); | |
5240 | ||
5241 | /* | |
5242 | * Retrieve the trusted config stored in the MOS and use it to create | |
5243 | * a new, exact version of the vdev tree, then reopen all vdevs. | |
5244 | */ | |
5245 | error = spa_ld_trusted_config(spa, type, B_FALSE); | |
5246 | if (error == EAGAIN) { | |
5247 | if (update_config_cache != NULL) | |
5248 | *update_config_cache = B_TRUE; | |
5249 | ||
5250 | /* | |
5251 | * Redo the loading process with the trusted config if it is | |
5252 | * too different from the untrusted config. | |
5253 | */ | |
5254 | spa_ld_prepare_for_reload(spa); | |
5255 | spa_load_note(spa, "RELOADING"); | |
5256 | error = spa_ld_mos_init(spa, type); | |
5257 | if (error != 0) | |
5258 | return (error); | |
5259 | ||
5260 | error = spa_ld_trusted_config(spa, type, B_TRUE); | |
5261 | if (error != 0) | |
5262 | return (error); | |
5263 | ||
5264 | } else if (error != 0) { | |
5265 | return (error); | |
5266 | } | |
5267 | ||
5268 | return (0); | |
5269 | } | |
5270 | ||
5271 | /* | |
5272 | * Load an existing storage pool, using the config provided. This config | |
5273 | * describes which vdevs are part of the pool and is later validated against | |
5274 | * partial configs present in each vdev's label and an entire copy of the | |
5275 | * config stored in the MOS. | |
5276 | */ | |
5277 | static int | |
5278 | spa_load_impl(spa_t *spa, spa_import_type_t type, const char **ereport) | |
5279 | { | |
5280 | int error = 0; | |
5281 | boolean_t missing_feat_write = B_FALSE; | |
5282 | boolean_t checkpoint_rewind = | |
5283 | (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
5284 | boolean_t update_config_cache = B_FALSE; | |
5285 | hrtime_t load_start = gethrtime(); | |
5286 | ||
5287 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5288 | ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); | |
5289 | ||
5290 | spa_load_note(spa, "LOADING"); | |
5291 | ||
5292 | error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); | |
5293 | if (error != 0) | |
5294 | return (error); | |
5295 | ||
5296 | /* | |
5297 | * If we are rewinding to the checkpoint then we need to repeat | |
5298 | * everything we've done so far in this function but this time | |
5299 | * selecting the checkpointed uberblock and using that to open | |
5300 | * the MOS. | |
5301 | */ | |
5302 | if (checkpoint_rewind) { | |
5303 | /* | |
5304 | * If we are rewinding to the checkpoint update config cache | |
5305 | * anyway. | |
5306 | */ | |
5307 | update_config_cache = B_TRUE; | |
5308 | ||
5309 | /* | |
5310 | * Extract the checkpointed uberblock from the current MOS | |
5311 | * and use this as the pool's uberblock from now on. If the | |
5312 | * pool is imported as writeable we also write the checkpoint | |
5313 | * uberblock to the labels, making the rewind permanent. | |
5314 | */ | |
5315 | error = spa_ld_checkpoint_rewind(spa); | |
5316 | if (error != 0) | |
5317 | return (error); | |
5318 | ||
5319 | /* | |
5320 | * Redo the loading process again with the | |
5321 | * checkpointed uberblock. | |
5322 | */ | |
5323 | spa_ld_prepare_for_reload(spa); | |
5324 | spa_load_note(spa, "LOADING checkpointed uberblock"); | |
5325 | error = spa_ld_mos_with_trusted_config(spa, type, NULL); | |
5326 | if (error != 0) | |
5327 | return (error); | |
5328 | } | |
5329 | ||
5330 | /* | |
5331 | * Drop the namespace lock for the rest of the function. | |
5332 | */ | |
5333 | spa->spa_load_thread = curthread; | |
5334 | mutex_exit(&spa_namespace_lock); | |
5335 | ||
5336 | /* | |
5337 | * Retrieve the checkpoint txg if the pool has a checkpoint. | |
5338 | */ | |
5339 | spa_import_progress_set_notes(spa, "Loading checkpoint txg"); | |
5340 | error = spa_ld_read_checkpoint_txg(spa); | |
5341 | if (error != 0) | |
5342 | goto fail; | |
5343 | ||
5344 | /* | |
5345 | * Retrieve the mapping of indirect vdevs. Those vdevs were removed | |
5346 | * from the pool and their contents were re-mapped to other vdevs. Note | |
5347 | * that everything that we read before this step must have been | |
5348 | * rewritten on concrete vdevs after the last device removal was | |
5349 | * initiated. Otherwise we could be reading from indirect vdevs before | |
5350 | * we have loaded their mappings. | |
5351 | */ | |
5352 | spa_import_progress_set_notes(spa, "Loading indirect vdev metadata"); | |
5353 | error = spa_ld_open_indirect_vdev_metadata(spa); | |
5354 | if (error != 0) | |
5355 | goto fail; | |
5356 | ||
5357 | /* | |
5358 | * Retrieve the full list of active features from the MOS and check if | |
5359 | * they are all supported. | |
5360 | */ | |
5361 | spa_import_progress_set_notes(spa, "Checking feature flags"); | |
5362 | error = spa_ld_check_features(spa, &missing_feat_write); | |
5363 | if (error != 0) | |
5364 | goto fail; | |
5365 | ||
5366 | /* | |
5367 | * Load several special directories from the MOS needed by the dsl_pool | |
5368 | * layer. | |
5369 | */ | |
5370 | spa_import_progress_set_notes(spa, "Loading special MOS directories"); | |
5371 | error = spa_ld_load_special_directories(spa); | |
5372 | if (error != 0) | |
5373 | goto fail; | |
5374 | ||
5375 | /* | |
5376 | * Retrieve pool properties from the MOS. | |
5377 | */ | |
5378 | spa_import_progress_set_notes(spa, "Loading properties"); | |
5379 | error = spa_ld_get_props(spa); | |
5380 | if (error != 0) | |
5381 | goto fail; | |
5382 | ||
5383 | /* | |
5384 | * Retrieve the list of auxiliary devices - cache devices and spares - | |
5385 | * and open them. | |
5386 | */ | |
5387 | spa_import_progress_set_notes(spa, "Loading AUX vdevs"); | |
5388 | error = spa_ld_open_aux_vdevs(spa, type); | |
5389 | if (error != 0) | |
5390 | goto fail; | |
5391 | ||
5392 | /* | |
5393 | * Load the metadata for all vdevs. Also check if unopenable devices | |
5394 | * should be autoreplaced. | |
5395 | */ | |
5396 | spa_import_progress_set_notes(spa, "Loading vdev metadata"); | |
5397 | error = spa_ld_load_vdev_metadata(spa); | |
5398 | if (error != 0) | |
5399 | goto fail; | |
5400 | ||
5401 | spa_import_progress_set_notes(spa, "Loading dedup tables"); | |
5402 | error = spa_ld_load_dedup_tables(spa); | |
5403 | if (error != 0) | |
5404 | goto fail; | |
5405 | ||
5406 | spa_import_progress_set_notes(spa, "Loading BRT"); | |
5407 | error = spa_ld_load_brt(spa); | |
5408 | if (error != 0) | |
5409 | goto fail; | |
5410 | ||
5411 | /* | |
5412 | * Verify the logs now to make sure we don't have any unexpected errors | |
5413 | * when we claim log blocks later. | |
5414 | */ | |
5415 | spa_import_progress_set_notes(spa, "Verifying Log Devices"); | |
5416 | error = spa_ld_verify_logs(spa, type, ereport); | |
5417 | if (error != 0) | |
5418 | goto fail; | |
5419 | ||
5420 | if (missing_feat_write) { | |
5421 | ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); | |
5422 | ||
5423 | /* | |
5424 | * At this point, we know that we can open the pool in | |
5425 | * read-only mode but not read-write mode. We now have enough | |
5426 | * information and can return to userland. | |
5427 | */ | |
5428 | error = spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, | |
5429 | ENOTSUP); | |
5430 | goto fail; | |
5431 | } | |
5432 | ||
5433 | /* | |
5434 | * Traverse the last txgs to make sure the pool was left off in a safe | |
5435 | * state. When performing an extreme rewind, we verify the whole pool, | |
5436 | * which can take a very long time. | |
5437 | */ | |
5438 | spa_import_progress_set_notes(spa, "Verifying pool data"); | |
5439 | error = spa_ld_verify_pool_data(spa); | |
5440 | if (error != 0) | |
5441 | goto fail; | |
5442 | ||
5443 | /* | |
5444 | * Calculate the deflated space for the pool. This must be done before | |
5445 | * we write anything to the pool because we'd need to update the space | |
5446 | * accounting using the deflated sizes. | |
5447 | */ | |
5448 | spa_import_progress_set_notes(spa, "Calculating deflated space"); | |
5449 | spa_update_dspace(spa); | |
5450 | ||
5451 | /* | |
5452 | * We have now retrieved all the information we needed to open the | |
5453 | * pool. If we are importing the pool in read-write mode, a few | |
5454 | * additional steps must be performed to finish the import. | |
5455 | */ | |
5456 | spa_import_progress_set_notes(spa, "Starting import"); | |
5457 | if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || | |
5458 | spa->spa_load_max_txg == UINT64_MAX)) { | |
5459 | uint64_t config_cache_txg = spa->spa_config_txg; | |
5460 | ||
5461 | ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); | |
5462 | ||
5463 | /* | |
5464 | * Before we do any zio_write's, complete the raidz expansion | |
5465 | * scratch space copying, if necessary. | |
5466 | */ | |
5467 | if (RRSS_GET_STATE(&spa->spa_uberblock) == RRSS_SCRATCH_VALID) | |
5468 | vdev_raidz_reflow_copy_scratch(spa); | |
5469 | ||
5470 | /* | |
5471 | * In case of a checkpoint rewind, log the original txg | |
5472 | * of the checkpointed uberblock. | |
5473 | */ | |
5474 | if (checkpoint_rewind) { | |
5475 | spa_history_log_internal(spa, "checkpoint rewind", | |
5476 | NULL, "rewound state to txg=%llu", | |
5477 | (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); | |
5478 | } | |
5479 | ||
5480 | spa_import_progress_set_notes(spa, "Claiming ZIL blocks"); | |
5481 | /* | |
5482 | * Traverse the ZIL and claim all blocks. | |
5483 | */ | |
5484 | spa_ld_claim_log_blocks(spa); | |
5485 | ||
5486 | /* | |
5487 | * Kick-off the syncing thread. | |
5488 | */ | |
5489 | spa->spa_sync_on = B_TRUE; | |
5490 | txg_sync_start(spa->spa_dsl_pool); | |
5491 | mmp_thread_start(spa); | |
5492 | ||
5493 | /* | |
5494 | * Wait for all claims to sync. We sync up to the highest | |
5495 | * claimed log block birth time so that claimed log blocks | |
5496 | * don't appear to be from the future. spa_claim_max_txg | |
5497 | * will have been set for us by ZIL traversal operations | |
5498 | * performed above. | |
5499 | */ | |
5500 | spa_import_progress_set_notes(spa, "Syncing ZIL claims"); | |
5501 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); | |
5502 | ||
5503 | /* | |
5504 | * Check if we need to request an update of the config. On the | |
5505 | * next sync, we would update the config stored in vdev labels | |
5506 | * and the cachefile (by default /etc/zfs/zpool.cache). | |
5507 | */ | |
5508 | spa_import_progress_set_notes(spa, "Updating configs"); | |
5509 | spa_ld_check_for_config_update(spa, config_cache_txg, | |
5510 | update_config_cache); | |
5511 | ||
5512 | /* | |
5513 | * Check if a rebuild was in progress and if so resume it. | |
5514 | * Then check all DTLs to see if anything needs resilvering. | |
5515 | * The resilver will be deferred if a rebuild was started. | |
5516 | */ | |
5517 | spa_import_progress_set_notes(spa, "Starting resilvers"); | |
5518 | if (vdev_rebuild_active(spa->spa_root_vdev)) { | |
5519 | vdev_rebuild_restart(spa); | |
5520 | } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
5521 | vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
5522 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
5523 | } | |
5524 | ||
5525 | /* | |
5526 | * Log the fact that we booted up (so that we can detect if | |
5527 | * we rebooted in the middle of an operation). | |
5528 | */ | |
5529 | spa_history_log_version(spa, "open", NULL); | |
5530 | ||
5531 | spa_import_progress_set_notes(spa, | |
5532 | "Restarting device removals"); | |
5533 | spa_restart_removal(spa); | |
5534 | spa_spawn_aux_threads(spa); | |
5535 | ||
5536 | /* | |
5537 | * Delete any inconsistent datasets. | |
5538 | * | |
5539 | * Note: | |
5540 | * Since we may be issuing deletes for clones here, | |
5541 | * we make sure to do so after we've spawned all the | |
5542 | * auxiliary threads above (from which the livelist | |
5543 | * deletion zthr is part of). | |
5544 | */ | |
5545 | spa_import_progress_set_notes(spa, | |
5546 | "Cleaning up inconsistent objsets"); | |
5547 | (void) dmu_objset_find(spa_name(spa), | |
5548 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
5549 | ||
5550 | /* | |
5551 | * Clean up any stale temporary dataset userrefs. | |
5552 | */ | |
5553 | spa_import_progress_set_notes(spa, | |
5554 | "Cleaning up temporary userrefs"); | |
5555 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); | |
5556 | ||
5557 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
5558 | spa_import_progress_set_notes(spa, "Restarting initialize"); | |
5559 | vdev_initialize_restart(spa->spa_root_vdev); | |
5560 | spa_import_progress_set_notes(spa, "Restarting TRIM"); | |
5561 | vdev_trim_restart(spa->spa_root_vdev); | |
5562 | vdev_autotrim_restart(spa); | |
5563 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
5564 | spa_import_progress_set_notes(spa, "Finished importing"); | |
5565 | } | |
5566 | zio_handle_import_delay(spa, gethrtime() - load_start); | |
5567 | ||
5568 | spa_import_progress_remove(spa_guid(spa)); | |
5569 | spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); | |
5570 | ||
5571 | spa_load_note(spa, "LOADED"); | |
5572 | fail: | |
5573 | mutex_enter(&spa_namespace_lock); | |
5574 | spa->spa_load_thread = NULL; | |
5575 | cv_broadcast(&spa_namespace_cv); | |
5576 | ||
5577 | return (error); | |
5578 | ||
5579 | } | |
5580 | ||
5581 | static int | |
5582 | spa_load_retry(spa_t *spa, spa_load_state_t state) | |
5583 | { | |
5584 | spa_mode_t mode = spa->spa_mode; | |
5585 | ||
5586 | spa_unload(spa); | |
5587 | spa_deactivate(spa); | |
5588 | ||
5589 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; | |
5590 | ||
5591 | spa_activate(spa, mode); | |
5592 | spa_async_suspend(spa); | |
5593 | ||
5594 | spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", | |
5595 | (u_longlong_t)spa->spa_load_max_txg); | |
5596 | ||
5597 | return (spa_load(spa, state, SPA_IMPORT_EXISTING)); | |
5598 | } | |
5599 | ||
5600 | /* | |
5601 | * If spa_load() fails this function will try loading prior txg's. If | |
5602 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
5603 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
5604 | * function will not rewind the pool and will return the same error as | |
5605 | * spa_load(). | |
5606 | */ | |
5607 | static int | |
5608 | spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, | |
5609 | int rewind_flags) | |
5610 | { | |
5611 | nvlist_t *loadinfo = NULL; | |
5612 | nvlist_t *config = NULL; | |
5613 | int load_error, rewind_error; | |
5614 | uint64_t safe_rewind_txg; | |
5615 | uint64_t min_txg; | |
5616 | ||
5617 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
5618 | spa->spa_load_max_txg = spa->spa_load_txg; | |
5619 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
5620 | } else { | |
5621 | spa->spa_load_max_txg = max_request; | |
5622 | if (max_request != UINT64_MAX) | |
5623 | spa->spa_extreme_rewind = B_TRUE; | |
5624 | } | |
5625 | ||
5626 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); | |
5627 | if (load_error == 0) | |
5628 | return (0); | |
5629 | if (load_error == ZFS_ERR_NO_CHECKPOINT) { | |
5630 | /* | |
5631 | * When attempting checkpoint-rewind on a pool with no | |
5632 | * checkpoint, we should not attempt to load uberblocks | |
5633 | * from previous txgs when spa_load fails. | |
5634 | */ | |
5635 | ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); | |
5636 | spa_import_progress_remove(spa_guid(spa)); | |
5637 | return (load_error); | |
5638 | } | |
5639 | ||
5640 | if (spa->spa_root_vdev != NULL) | |
5641 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
5642 | ||
5643 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
5644 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
5645 | ||
5646 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
5647 | nvlist_free(config); | |
5648 | spa_import_progress_remove(spa_guid(spa)); | |
5649 | return (load_error); | |
5650 | } | |
5651 | ||
5652 | if (state == SPA_LOAD_RECOVER) { | |
5653 | /* Price of rolling back is discarding txgs, including log */ | |
5654 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
5655 | } else { | |
5656 | /* | |
5657 | * If we aren't rolling back save the load info from our first | |
5658 | * import attempt so that we can restore it after attempting | |
5659 | * to rewind. | |
5660 | */ | |
5661 | loadinfo = spa->spa_load_info; | |
5662 | spa->spa_load_info = fnvlist_alloc(); | |
5663 | } | |
5664 | ||
5665 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
5666 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
5667 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
5668 | TXG_INITIAL : safe_rewind_txg; | |
5669 | ||
5670 | /* | |
5671 | * Continue as long as we're finding errors, we're still within | |
5672 | * the acceptable rewind range, and we're still finding uberblocks | |
5673 | */ | |
5674 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
5675 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
5676 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
5677 | spa->spa_extreme_rewind = B_TRUE; | |
5678 | rewind_error = spa_load_retry(spa, state); | |
5679 | } | |
5680 | ||
5681 | spa->spa_extreme_rewind = B_FALSE; | |
5682 | spa->spa_load_max_txg = UINT64_MAX; | |
5683 | ||
5684 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
5685 | spa_config_set(spa, config); | |
5686 | else | |
5687 | nvlist_free(config); | |
5688 | ||
5689 | if (state == SPA_LOAD_RECOVER) { | |
5690 | ASSERT3P(loadinfo, ==, NULL); | |
5691 | spa_import_progress_remove(spa_guid(spa)); | |
5692 | return (rewind_error); | |
5693 | } else { | |
5694 | /* Store the rewind info as part of the initial load info */ | |
5695 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
5696 | spa->spa_load_info); | |
5697 | ||
5698 | /* Restore the initial load info */ | |
5699 | fnvlist_free(spa->spa_load_info); | |
5700 | spa->spa_load_info = loadinfo; | |
5701 | ||
5702 | spa_import_progress_remove(spa_guid(spa)); | |
5703 | return (load_error); | |
5704 | } | |
5705 | } | |
5706 | ||
5707 | /* | |
5708 | * Pool Open/Import | |
5709 | * | |
5710 | * The import case is identical to an open except that the configuration is sent | |
5711 | * down from userland, instead of grabbed from the configuration cache. For the | |
5712 | * case of an open, the pool configuration will exist in the | |
5713 | * POOL_STATE_UNINITIALIZED state. | |
5714 | * | |
5715 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
5716 | * the same time open the pool, without having to keep around the spa_t in some | |
5717 | * ambiguous state. | |
5718 | */ | |
5719 | static int | |
5720 | spa_open_common(const char *pool, spa_t **spapp, const void *tag, | |
5721 | nvlist_t *nvpolicy, nvlist_t **config) | |
5722 | { | |
5723 | spa_t *spa; | |
5724 | spa_load_state_t state = SPA_LOAD_OPEN; | |
5725 | int error; | |
5726 | int locked = B_FALSE; | |
5727 | int firstopen = B_FALSE; | |
5728 | ||
5729 | *spapp = NULL; | |
5730 | ||
5731 | /* | |
5732 | * As disgusting as this is, we need to support recursive calls to this | |
5733 | * function because dsl_dir_open() is called during spa_load(), and ends | |
5734 | * up calling spa_open() again. The real fix is to figure out how to | |
5735 | * avoid dsl_dir_open() calling this in the first place. | |
5736 | */ | |
5737 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { | |
5738 | mutex_enter(&spa_namespace_lock); | |
5739 | locked = B_TRUE; | |
5740 | } | |
5741 | ||
5742 | if ((spa = spa_lookup(pool)) == NULL) { | |
5743 | if (locked) | |
5744 | mutex_exit(&spa_namespace_lock); | |
5745 | return (SET_ERROR(ENOENT)); | |
5746 | } | |
5747 | ||
5748 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
5749 | zpool_load_policy_t policy; | |
5750 | ||
5751 | firstopen = B_TRUE; | |
5752 | ||
5753 | zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, | |
5754 | &policy); | |
5755 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) | |
5756 | state = SPA_LOAD_RECOVER; | |
5757 | ||
5758 | spa_activate(spa, spa_mode_global); | |
5759 | ||
5760 | if (state != SPA_LOAD_RECOVER) | |
5761 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
5762 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; | |
5763 | ||
5764 | zfs_dbgmsg("spa_open_common: opening %s", pool); | |
5765 | error = spa_load_best(spa, state, policy.zlp_txg, | |
5766 | policy.zlp_rewind); | |
5767 | ||
5768 | if (error == EBADF) { | |
5769 | /* | |
5770 | * If vdev_validate() returns failure (indicated by | |
5771 | * EBADF), it indicates that one of the vdevs indicates | |
5772 | * that the pool has been exported or destroyed. If | |
5773 | * this is the case, the config cache is out of sync and | |
5774 | * we should remove the pool from the namespace. | |
5775 | */ | |
5776 | spa_unload(spa); | |
5777 | spa_deactivate(spa); | |
5778 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); | |
5779 | spa_remove(spa); | |
5780 | if (locked) | |
5781 | mutex_exit(&spa_namespace_lock); | |
5782 | return (SET_ERROR(ENOENT)); | |
5783 | } | |
5784 | ||
5785 | if (error) { | |
5786 | /* | |
5787 | * We can't open the pool, but we still have useful | |
5788 | * information: the state of each vdev after the | |
5789 | * attempted vdev_open(). Return this to the user. | |
5790 | */ | |
5791 | if (config != NULL && spa->spa_config) { | |
5792 | *config = fnvlist_dup(spa->spa_config); | |
5793 | fnvlist_add_nvlist(*config, | |
5794 | ZPOOL_CONFIG_LOAD_INFO, | |
5795 | spa->spa_load_info); | |
5796 | } | |
5797 | spa_unload(spa); | |
5798 | spa_deactivate(spa); | |
5799 | spa->spa_last_open_failed = error; | |
5800 | if (locked) | |
5801 | mutex_exit(&spa_namespace_lock); | |
5802 | *spapp = NULL; | |
5803 | return (error); | |
5804 | } | |
5805 | } | |
5806 | ||
5807 | spa_open_ref(spa, tag); | |
5808 | ||
5809 | if (config != NULL) | |
5810 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
5811 | ||
5812 | /* | |
5813 | * If we've recovered the pool, pass back any information we | |
5814 | * gathered while doing the load. | |
5815 | */ | |
5816 | if (state == SPA_LOAD_RECOVER && config != NULL) { | |
5817 | fnvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, | |
5818 | spa->spa_load_info); | |
5819 | } | |
5820 | ||
5821 | if (locked) { | |
5822 | spa->spa_last_open_failed = 0; | |
5823 | spa->spa_last_ubsync_txg = 0; | |
5824 | spa->spa_load_txg = 0; | |
5825 | mutex_exit(&spa_namespace_lock); | |
5826 | } | |
5827 | ||
5828 | if (firstopen) | |
5829 | zvol_create_minors_recursive(spa_name(spa)); | |
5830 | ||
5831 | *spapp = spa; | |
5832 | ||
5833 | return (0); | |
5834 | } | |
5835 | ||
5836 | int | |
5837 | spa_open_rewind(const char *name, spa_t **spapp, const void *tag, | |
5838 | nvlist_t *policy, nvlist_t **config) | |
5839 | { | |
5840 | return (spa_open_common(name, spapp, tag, policy, config)); | |
5841 | } | |
5842 | ||
5843 | int | |
5844 | spa_open(const char *name, spa_t **spapp, const void *tag) | |
5845 | { | |
5846 | return (spa_open_common(name, spapp, tag, NULL, NULL)); | |
5847 | } | |
5848 | ||
5849 | /* | |
5850 | * Lookup the given spa_t, incrementing the inject count in the process, | |
5851 | * preventing it from being exported or destroyed. | |
5852 | */ | |
5853 | spa_t * | |
5854 | spa_inject_addref(char *name) | |
5855 | { | |
5856 | spa_t *spa; | |
5857 | ||
5858 | mutex_enter(&spa_namespace_lock); | |
5859 | if ((spa = spa_lookup(name)) == NULL) { | |
5860 | mutex_exit(&spa_namespace_lock); | |
5861 | return (NULL); | |
5862 | } | |
5863 | spa->spa_inject_ref++; | |
5864 | mutex_exit(&spa_namespace_lock); | |
5865 | ||
5866 | return (spa); | |
5867 | } | |
5868 | ||
5869 | void | |
5870 | spa_inject_delref(spa_t *spa) | |
5871 | { | |
5872 | mutex_enter(&spa_namespace_lock); | |
5873 | spa->spa_inject_ref--; | |
5874 | mutex_exit(&spa_namespace_lock); | |
5875 | } | |
5876 | ||
5877 | /* | |
5878 | * Add spares device information to the nvlist. | |
5879 | */ | |
5880 | static void | |
5881 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
5882 | { | |
5883 | nvlist_t **spares; | |
5884 | uint_t i, nspares; | |
5885 | nvlist_t *nvroot; | |
5886 | uint64_t guid; | |
5887 | vdev_stat_t *vs; | |
5888 | uint_t vsc; | |
5889 | uint64_t pool; | |
5890 | ||
5891 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
5892 | ||
5893 | if (spa->spa_spares.sav_count == 0) | |
5894 | return; | |
5895 | ||
5896 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); | |
5897 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5898 | ZPOOL_CONFIG_SPARES, &spares, &nspares)); | |
5899 | if (nspares != 0) { | |
5900 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
5901 | (const nvlist_t * const *)spares, nspares); | |
5902 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
5903 | &spares, &nspares)); | |
5904 | ||
5905 | /* | |
5906 | * Go through and find any spares which have since been | |
5907 | * repurposed as an active spare. If this is the case, update | |
5908 | * their status appropriately. | |
5909 | */ | |
5910 | for (i = 0; i < nspares; i++) { | |
5911 | guid = fnvlist_lookup_uint64(spares[i], | |
5912 | ZPOOL_CONFIG_GUID); | |
5913 | VERIFY0(nvlist_lookup_uint64_array(spares[i], | |
5914 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
5915 | if (spa_spare_exists(guid, &pool, NULL) && | |
5916 | pool != 0ULL) { | |
5917 | vs->vs_state = VDEV_STATE_CANT_OPEN; | |
5918 | vs->vs_aux = VDEV_AUX_SPARED; | |
5919 | } else { | |
5920 | vs->vs_state = | |
5921 | spa->spa_spares.sav_vdevs[i]->vdev_state; | |
5922 | } | |
5923 | } | |
5924 | } | |
5925 | } | |
5926 | ||
5927 | /* | |
5928 | * Add l2cache device information to the nvlist, including vdev stats. | |
5929 | */ | |
5930 | static void | |
5931 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
5932 | { | |
5933 | nvlist_t **l2cache; | |
5934 | uint_t i, j, nl2cache; | |
5935 | nvlist_t *nvroot; | |
5936 | uint64_t guid; | |
5937 | vdev_t *vd; | |
5938 | vdev_stat_t *vs; | |
5939 | uint_t vsc; | |
5940 | ||
5941 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
5942 | ||
5943 | if (spa->spa_l2cache.sav_count == 0) | |
5944 | return; | |
5945 | ||
5946 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); | |
5947 | VERIFY0(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5948 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache)); | |
5949 | if (nl2cache != 0) { | |
5950 | fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
5951 | (const nvlist_t * const *)l2cache, nl2cache); | |
5952 | VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
5953 | &l2cache, &nl2cache)); | |
5954 | ||
5955 | /* | |
5956 | * Update level 2 cache device stats. | |
5957 | */ | |
5958 | ||
5959 | for (i = 0; i < nl2cache; i++) { | |
5960 | guid = fnvlist_lookup_uint64(l2cache[i], | |
5961 | ZPOOL_CONFIG_GUID); | |
5962 | ||
5963 | vd = NULL; | |
5964 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
5965 | if (guid == | |
5966 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
5967 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
5968 | break; | |
5969 | } | |
5970 | } | |
5971 | ASSERT(vd != NULL); | |
5972 | ||
5973 | VERIFY0(nvlist_lookup_uint64_array(l2cache[i], | |
5974 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)); | |
5975 | vdev_get_stats(vd, vs); | |
5976 | vdev_config_generate_stats(vd, l2cache[i]); | |
5977 | ||
5978 | } | |
5979 | } | |
5980 | } | |
5981 | ||
5982 | static void | |
5983 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) | |
5984 | { | |
5985 | zap_cursor_t zc; | |
5986 | zap_attribute_t za; | |
5987 | ||
5988 | if (spa->spa_feat_for_read_obj != 0) { | |
5989 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
5990 | spa->spa_feat_for_read_obj); | |
5991 | zap_cursor_retrieve(&zc, &za) == 0; | |
5992 | zap_cursor_advance(&zc)) { | |
5993 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
5994 | za.za_num_integers == 1); | |
5995 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
5996 | za.za_first_integer)); | |
5997 | } | |
5998 | zap_cursor_fini(&zc); | |
5999 | } | |
6000 | ||
6001 | if (spa->spa_feat_for_write_obj != 0) { | |
6002 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6003 | spa->spa_feat_for_write_obj); | |
6004 | zap_cursor_retrieve(&zc, &za) == 0; | |
6005 | zap_cursor_advance(&zc)) { | |
6006 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
6007 | za.za_num_integers == 1); | |
6008 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
6009 | za.za_first_integer)); | |
6010 | } | |
6011 | zap_cursor_fini(&zc); | |
6012 | } | |
6013 | } | |
6014 | ||
6015 | static void | |
6016 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
6017 | { | |
6018 | int i; | |
6019 | ||
6020 | for (i = 0; i < SPA_FEATURES; i++) { | |
6021 | zfeature_info_t feature = spa_feature_table[i]; | |
6022 | uint64_t refcount; | |
6023 | ||
6024 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
6025 | continue; | |
6026 | ||
6027 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
6028 | } | |
6029 | } | |
6030 | ||
6031 | /* | |
6032 | * Store a list of pool features and their reference counts in the | |
6033 | * config. | |
6034 | * | |
6035 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
6036 | * the pool features and reference counts from disk, then save the list | |
6037 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
6038 | * and refresh its values from the cached reference counts. This | |
6039 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
6040 | * clear' can resume the pool. | |
6041 | */ | |
6042 | static void | |
6043 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
6044 | { | |
6045 | nvlist_t *features; | |
6046 | ||
6047 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
6048 | ||
6049 | mutex_enter(&spa->spa_feat_stats_lock); | |
6050 | features = spa->spa_feat_stats; | |
6051 | ||
6052 | if (features != NULL) { | |
6053 | spa_feature_stats_from_cache(spa, features); | |
6054 | } else { | |
6055 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
6056 | spa->spa_feat_stats = features; | |
6057 | spa_feature_stats_from_disk(spa, features); | |
6058 | } | |
6059 | ||
6060 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, | |
6061 | features)); | |
6062 | ||
6063 | mutex_exit(&spa->spa_feat_stats_lock); | |
6064 | } | |
6065 | ||
6066 | int | |
6067 | spa_get_stats(const char *name, nvlist_t **config, | |
6068 | char *altroot, size_t buflen) | |
6069 | { | |
6070 | int error; | |
6071 | spa_t *spa; | |
6072 | ||
6073 | *config = NULL; | |
6074 | error = spa_open_common(name, &spa, FTAG, NULL, config); | |
6075 | ||
6076 | if (spa != NULL) { | |
6077 | /* | |
6078 | * This still leaves a window of inconsistency where the spares | |
6079 | * or l2cache devices could change and the config would be | |
6080 | * self-inconsistent. | |
6081 | */ | |
6082 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6083 | ||
6084 | if (*config != NULL) { | |
6085 | uint64_t loadtimes[2]; | |
6086 | ||
6087 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
6088 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
6089 | fnvlist_add_uint64_array(*config, | |
6090 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2); | |
6091 | ||
6092 | fnvlist_add_uint64(*config, | |
6093 | ZPOOL_CONFIG_ERRCOUNT, | |
6094 | spa_approx_errlog_size(spa)); | |
6095 | ||
6096 | if (spa_suspended(spa)) { | |
6097 | fnvlist_add_uint64(*config, | |
6098 | ZPOOL_CONFIG_SUSPENDED, | |
6099 | spa->spa_failmode); | |
6100 | fnvlist_add_uint64(*config, | |
6101 | ZPOOL_CONFIG_SUSPENDED_REASON, | |
6102 | spa->spa_suspended); | |
6103 | } | |
6104 | ||
6105 | spa_add_spares(spa, *config); | |
6106 | spa_add_l2cache(spa, *config); | |
6107 | spa_add_feature_stats(spa, *config); | |
6108 | } | |
6109 | } | |
6110 | ||
6111 | /* | |
6112 | * We want to get the alternate root even for faulted pools, so we cheat | |
6113 | * and call spa_lookup() directly. | |
6114 | */ | |
6115 | if (altroot) { | |
6116 | if (spa == NULL) { | |
6117 | mutex_enter(&spa_namespace_lock); | |
6118 | spa = spa_lookup(name); | |
6119 | if (spa) | |
6120 | spa_altroot(spa, altroot, buflen); | |
6121 | else | |
6122 | altroot[0] = '\0'; | |
6123 | spa = NULL; | |
6124 | mutex_exit(&spa_namespace_lock); | |
6125 | } else { | |
6126 | spa_altroot(spa, altroot, buflen); | |
6127 | } | |
6128 | } | |
6129 | ||
6130 | if (spa != NULL) { | |
6131 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
6132 | spa_close(spa, FTAG); | |
6133 | } | |
6134 | ||
6135 | return (error); | |
6136 | } | |
6137 | ||
6138 | /* | |
6139 | * Validate that the auxiliary device array is well formed. We must have an | |
6140 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
6141 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
6142 | * specified, as long as they are well-formed. | |
6143 | */ | |
6144 | static int | |
6145 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
6146 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
6147 | vdev_labeltype_t label) | |
6148 | { | |
6149 | nvlist_t **dev; | |
6150 | uint_t i, ndev; | |
6151 | vdev_t *vd; | |
6152 | int error; | |
6153 | ||
6154 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
6155 | ||
6156 | /* | |
6157 | * It's acceptable to have no devs specified. | |
6158 | */ | |
6159 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
6160 | return (0); | |
6161 | ||
6162 | if (ndev == 0) | |
6163 | return (SET_ERROR(EINVAL)); | |
6164 | ||
6165 | /* | |
6166 | * Make sure the pool is formatted with a version that supports this | |
6167 | * device type. | |
6168 | */ | |
6169 | if (spa_version(spa) < version) | |
6170 | return (SET_ERROR(ENOTSUP)); | |
6171 | ||
6172 | /* | |
6173 | * Set the pending device list so we correctly handle device in-use | |
6174 | * checking. | |
6175 | */ | |
6176 | sav->sav_pending = dev; | |
6177 | sav->sav_npending = ndev; | |
6178 | ||
6179 | for (i = 0; i < ndev; i++) { | |
6180 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
6181 | mode)) != 0) | |
6182 | goto out; | |
6183 | ||
6184 | if (!vd->vdev_ops->vdev_op_leaf) { | |
6185 | vdev_free(vd); | |
6186 | error = SET_ERROR(EINVAL); | |
6187 | goto out; | |
6188 | } | |
6189 | ||
6190 | vd->vdev_top = vd; | |
6191 | ||
6192 | if ((error = vdev_open(vd)) == 0 && | |
6193 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
6194 | fnvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, | |
6195 | vd->vdev_guid); | |
6196 | } | |
6197 | ||
6198 | vdev_free(vd); | |
6199 | ||
6200 | if (error && | |
6201 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
6202 | goto out; | |
6203 | else | |
6204 | error = 0; | |
6205 | } | |
6206 | ||
6207 | out: | |
6208 | sav->sav_pending = NULL; | |
6209 | sav->sav_npending = 0; | |
6210 | return (error); | |
6211 | } | |
6212 | ||
6213 | static int | |
6214 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
6215 | { | |
6216 | int error; | |
6217 | ||
6218 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
6219 | ||
6220 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
6221 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
6222 | VDEV_LABEL_SPARE)) != 0) { | |
6223 | return (error); | |
6224 | } | |
6225 | ||
6226 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
6227 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
6228 | VDEV_LABEL_L2CACHE)); | |
6229 | } | |
6230 | ||
6231 | static void | |
6232 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
6233 | const char *config) | |
6234 | { | |
6235 | int i; | |
6236 | ||
6237 | if (sav->sav_config != NULL) { | |
6238 | nvlist_t **olddevs; | |
6239 | uint_t oldndevs; | |
6240 | nvlist_t **newdevs; | |
6241 | ||
6242 | /* | |
6243 | * Generate new dev list by concatenating with the | |
6244 | * current dev list. | |
6245 | */ | |
6246 | VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, config, | |
6247 | &olddevs, &oldndevs)); | |
6248 | ||
6249 | newdevs = kmem_alloc(sizeof (void *) * | |
6250 | (ndevs + oldndevs), KM_SLEEP); | |
6251 | for (i = 0; i < oldndevs; i++) | |
6252 | newdevs[i] = fnvlist_dup(olddevs[i]); | |
6253 | for (i = 0; i < ndevs; i++) | |
6254 | newdevs[i + oldndevs] = fnvlist_dup(devs[i]); | |
6255 | ||
6256 | fnvlist_remove(sav->sav_config, config); | |
6257 | ||
6258 | fnvlist_add_nvlist_array(sav->sav_config, config, | |
6259 | (const nvlist_t * const *)newdevs, ndevs + oldndevs); | |
6260 | for (i = 0; i < oldndevs + ndevs; i++) | |
6261 | nvlist_free(newdevs[i]); | |
6262 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
6263 | } else { | |
6264 | /* | |
6265 | * Generate a new dev list. | |
6266 | */ | |
6267 | sav->sav_config = fnvlist_alloc(); | |
6268 | fnvlist_add_nvlist_array(sav->sav_config, config, | |
6269 | (const nvlist_t * const *)devs, ndevs); | |
6270 | } | |
6271 | } | |
6272 | ||
6273 | /* | |
6274 | * Stop and drop level 2 ARC devices | |
6275 | */ | |
6276 | void | |
6277 | spa_l2cache_drop(spa_t *spa) | |
6278 | { | |
6279 | vdev_t *vd; | |
6280 | int i; | |
6281 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
6282 | ||
6283 | for (i = 0; i < sav->sav_count; i++) { | |
6284 | uint64_t pool; | |
6285 | ||
6286 | vd = sav->sav_vdevs[i]; | |
6287 | ASSERT(vd != NULL); | |
6288 | ||
6289 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
6290 | pool != 0ULL && l2arc_vdev_present(vd)) | |
6291 | l2arc_remove_vdev(vd); | |
6292 | } | |
6293 | } | |
6294 | ||
6295 | /* | |
6296 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
6297 | * have the encryption feature flag enabled. | |
6298 | */ | |
6299 | static int | |
6300 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
6301 | boolean_t has_encryption) | |
6302 | { | |
6303 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
6304 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
6305 | !has_encryption) | |
6306 | return (SET_ERROR(ENOTSUP)); | |
6307 | ||
6308 | return (dmu_objset_create_crypt_check(NULL, dcp, NULL)); | |
6309 | } | |
6310 | ||
6311 | /* | |
6312 | * Pool Creation | |
6313 | */ | |
6314 | int | |
6315 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
6316 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) | |
6317 | { | |
6318 | spa_t *spa; | |
6319 | const char *altroot = NULL; | |
6320 | vdev_t *rvd; | |
6321 | dsl_pool_t *dp; | |
6322 | dmu_tx_t *tx; | |
6323 | int error = 0; | |
6324 | uint64_t txg = TXG_INITIAL; | |
6325 | nvlist_t **spares, **l2cache; | |
6326 | uint_t nspares, nl2cache; | |
6327 | uint64_t version, obj, ndraid = 0; | |
6328 | boolean_t has_features; | |
6329 | boolean_t has_encryption; | |
6330 | boolean_t has_allocclass; | |
6331 | spa_feature_t feat; | |
6332 | const char *feat_name; | |
6333 | const char *poolname; | |
6334 | nvlist_t *nvl; | |
6335 | ||
6336 | if (props == NULL || | |
6337 | nvlist_lookup_string(props, | |
6338 | zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) | |
6339 | poolname = (char *)pool; | |
6340 | ||
6341 | /* | |
6342 | * If this pool already exists, return failure. | |
6343 | */ | |
6344 | mutex_enter(&spa_namespace_lock); | |
6345 | if (spa_lookup(poolname) != NULL) { | |
6346 | mutex_exit(&spa_namespace_lock); | |
6347 | return (SET_ERROR(EEXIST)); | |
6348 | } | |
6349 | ||
6350 | /* | |
6351 | * Allocate a new spa_t structure. | |
6352 | */ | |
6353 | nvl = fnvlist_alloc(); | |
6354 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
6355 | (void) nvlist_lookup_string(props, | |
6356 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
6357 | spa = spa_add(poolname, nvl, altroot); | |
6358 | fnvlist_free(nvl); | |
6359 | spa_activate(spa, spa_mode_global); | |
6360 | ||
6361 | if (props && (error = spa_prop_validate(spa, props))) { | |
6362 | spa_deactivate(spa); | |
6363 | spa_remove(spa); | |
6364 | mutex_exit(&spa_namespace_lock); | |
6365 | return (error); | |
6366 | } | |
6367 | ||
6368 | /* | |
6369 | * Temporary pool names should never be written to disk. | |
6370 | */ | |
6371 | if (poolname != pool) | |
6372 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
6373 | ||
6374 | has_features = B_FALSE; | |
6375 | has_encryption = B_FALSE; | |
6376 | has_allocclass = B_FALSE; | |
6377 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); | |
6378 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { | |
6379 | if (zpool_prop_feature(nvpair_name(elem))) { | |
6380 | has_features = B_TRUE; | |
6381 | ||
6382 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
6383 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
6384 | if (feat == SPA_FEATURE_ENCRYPTION) | |
6385 | has_encryption = B_TRUE; | |
6386 | if (feat == SPA_FEATURE_ALLOCATION_CLASSES) | |
6387 | has_allocclass = B_TRUE; | |
6388 | } | |
6389 | } | |
6390 | ||
6391 | /* verify encryption params, if they were provided */ | |
6392 | if (dcp != NULL) { | |
6393 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
6394 | if (error != 0) { | |
6395 | spa_deactivate(spa); | |
6396 | spa_remove(spa); | |
6397 | mutex_exit(&spa_namespace_lock); | |
6398 | return (error); | |
6399 | } | |
6400 | } | |
6401 | if (!has_allocclass && zfs_special_devs(nvroot, NULL)) { | |
6402 | spa_deactivate(spa); | |
6403 | spa_remove(spa); | |
6404 | mutex_exit(&spa_namespace_lock); | |
6405 | return (ENOTSUP); | |
6406 | } | |
6407 | ||
6408 | if (has_features || nvlist_lookup_uint64(props, | |
6409 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
6410 | version = SPA_VERSION; | |
6411 | } | |
6412 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
6413 | ||
6414 | spa->spa_first_txg = txg; | |
6415 | spa->spa_uberblock.ub_txg = txg - 1; | |
6416 | spa->spa_uberblock.ub_version = version; | |
6417 | spa->spa_ubsync = spa->spa_uberblock; | |
6418 | spa->spa_load_state = SPA_LOAD_CREATE; | |
6419 | spa->spa_removing_phys.sr_state = DSS_NONE; | |
6420 | spa->spa_removing_phys.sr_removing_vdev = -1; | |
6421 | spa->spa_removing_phys.sr_prev_indirect_vdev = -1; | |
6422 | spa->spa_indirect_vdevs_loaded = B_TRUE; | |
6423 | ||
6424 | /* | |
6425 | * Create "The Godfather" zio to hold all async IOs | |
6426 | */ | |
6427 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
6428 | KM_SLEEP); | |
6429 | for (int i = 0; i < max_ncpus; i++) { | |
6430 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
6431 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
6432 | ZIO_FLAG_GODFATHER); | |
6433 | } | |
6434 | ||
6435 | /* | |
6436 | * Create the root vdev. | |
6437 | */ | |
6438 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6439 | ||
6440 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
6441 | ||
6442 | ASSERT(error != 0 || rvd != NULL); | |
6443 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
6444 | ||
6445 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
6446 | error = SET_ERROR(EINVAL); | |
6447 | ||
6448 | if (error == 0 && | |
6449 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
6450 | (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 && | |
6451 | (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { | |
6452 | /* | |
6453 | * instantiate the metaslab groups (this will dirty the vdevs) | |
6454 | * we can no longer error exit past this point | |
6455 | */ | |
6456 | for (int c = 0; error == 0 && c < rvd->vdev_children; c++) { | |
6457 | vdev_t *vd = rvd->vdev_child[c]; | |
6458 | ||
6459 | vdev_metaslab_set_size(vd); | |
6460 | vdev_expand(vd, txg); | |
6461 | } | |
6462 | } | |
6463 | ||
6464 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6465 | ||
6466 | if (error != 0) { | |
6467 | spa_unload(spa); | |
6468 | spa_deactivate(spa); | |
6469 | spa_remove(spa); | |
6470 | mutex_exit(&spa_namespace_lock); | |
6471 | return (error); | |
6472 | } | |
6473 | ||
6474 | /* | |
6475 | * Get the list of spares, if specified. | |
6476 | */ | |
6477 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6478 | &spares, &nspares) == 0) { | |
6479 | spa->spa_spares.sav_config = fnvlist_alloc(); | |
6480 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
6481 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, | |
6482 | nspares); | |
6483 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6484 | spa_load_spares(spa); | |
6485 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6486 | spa->spa_spares.sav_sync = B_TRUE; | |
6487 | } | |
6488 | ||
6489 | /* | |
6490 | * Get the list of level 2 cache devices, if specified. | |
6491 | */ | |
6492 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6493 | &l2cache, &nl2cache) == 0) { | |
6494 | VERIFY0(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
6495 | NV_UNIQUE_NAME, KM_SLEEP)); | |
6496 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
6497 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, | |
6498 | nl2cache); | |
6499 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6500 | spa_load_l2cache(spa); | |
6501 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6502 | spa->spa_l2cache.sav_sync = B_TRUE; | |
6503 | } | |
6504 | ||
6505 | spa->spa_is_initializing = B_TRUE; | |
6506 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); | |
6507 | spa->spa_is_initializing = B_FALSE; | |
6508 | ||
6509 | /* | |
6510 | * Create DDTs (dedup tables). | |
6511 | */ | |
6512 | ddt_create(spa); | |
6513 | /* | |
6514 | * Create BRT table and BRT table object. | |
6515 | */ | |
6516 | brt_create(spa); | |
6517 | ||
6518 | spa_update_dspace(spa); | |
6519 | ||
6520 | tx = dmu_tx_create_assigned(dp, txg); | |
6521 | ||
6522 | /* | |
6523 | * Create the pool's history object. | |
6524 | */ | |
6525 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
6526 | spa_history_create_obj(spa, tx); | |
6527 | ||
6528 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
6529 | spa_history_log_version(spa, "create", tx); | |
6530 | ||
6531 | /* | |
6532 | * Create the pool config object. | |
6533 | */ | |
6534 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
6535 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, | |
6536 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
6537 | ||
6538 | if (zap_add(spa->spa_meta_objset, | |
6539 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
6540 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
6541 | cmn_err(CE_PANIC, "failed to add pool config"); | |
6542 | } | |
6543 | ||
6544 | if (zap_add(spa->spa_meta_objset, | |
6545 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
6546 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
6547 | cmn_err(CE_PANIC, "failed to add pool version"); | |
6548 | } | |
6549 | ||
6550 | /* Newly created pools with the right version are always deflated. */ | |
6551 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6552 | spa->spa_deflate = TRUE; | |
6553 | if (zap_add(spa->spa_meta_objset, | |
6554 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6555 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
6556 | cmn_err(CE_PANIC, "failed to add deflate"); | |
6557 | } | |
6558 | } | |
6559 | ||
6560 | /* | |
6561 | * Create the deferred-free bpobj. Turn off compression | |
6562 | * because sync-to-convergence takes longer if the blocksize | |
6563 | * keeps changing. | |
6564 | */ | |
6565 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); | |
6566 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
6567 | ZIO_COMPRESS_OFF, tx); | |
6568 | if (zap_add(spa->spa_meta_objset, | |
6569 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, | |
6570 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
6571 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
6572 | } | |
6573 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, | |
6574 | spa->spa_meta_objset, obj)); | |
6575 | ||
6576 | /* | |
6577 | * Generate some random noise for salted checksums to operate on. | |
6578 | */ | |
6579 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
6580 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
6581 | ||
6582 | /* | |
6583 | * Set pool properties. | |
6584 | */ | |
6585 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
6586 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
6587 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
6588 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); | |
6589 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); | |
6590 | spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM); | |
6591 | ||
6592 | if (props != NULL) { | |
6593 | spa_configfile_set(spa, props, B_FALSE); | |
6594 | spa_sync_props(props, tx); | |
6595 | } | |
6596 | ||
6597 | for (int i = 0; i < ndraid; i++) | |
6598 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
6599 | ||
6600 | dmu_tx_commit(tx); | |
6601 | ||
6602 | spa->spa_sync_on = B_TRUE; | |
6603 | txg_sync_start(dp); | |
6604 | mmp_thread_start(spa); | |
6605 | txg_wait_synced(dp, txg); | |
6606 | ||
6607 | spa_spawn_aux_threads(spa); | |
6608 | ||
6609 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE); | |
6610 | ||
6611 | /* | |
6612 | * Don't count references from objsets that are already closed | |
6613 | * and are making their way through the eviction process. | |
6614 | */ | |
6615 | spa_evicting_os_wait(spa); | |
6616 | spa->spa_minref = zfs_refcount_count(&spa->spa_refcount); | |
6617 | spa->spa_load_state = SPA_LOAD_NONE; | |
6618 | ||
6619 | spa_import_os(spa); | |
6620 | ||
6621 | mutex_exit(&spa_namespace_lock); | |
6622 | ||
6623 | return (0); | |
6624 | } | |
6625 | ||
6626 | /* | |
6627 | * Import a non-root pool into the system. | |
6628 | */ | |
6629 | int | |
6630 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) | |
6631 | { | |
6632 | spa_t *spa; | |
6633 | const char *altroot = NULL; | |
6634 | spa_load_state_t state = SPA_LOAD_IMPORT; | |
6635 | zpool_load_policy_t policy; | |
6636 | spa_mode_t mode = spa_mode_global; | |
6637 | uint64_t readonly = B_FALSE; | |
6638 | int error; | |
6639 | nvlist_t *nvroot; | |
6640 | nvlist_t **spares, **l2cache; | |
6641 | uint_t nspares, nl2cache; | |
6642 | ||
6643 | /* | |
6644 | * If a pool with this name exists, return failure. | |
6645 | */ | |
6646 | mutex_enter(&spa_namespace_lock); | |
6647 | if (spa_lookup(pool) != NULL) { | |
6648 | mutex_exit(&spa_namespace_lock); | |
6649 | return (SET_ERROR(EEXIST)); | |
6650 | } | |
6651 | ||
6652 | /* | |
6653 | * Create and initialize the spa structure. | |
6654 | */ | |
6655 | (void) nvlist_lookup_string(props, | |
6656 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
6657 | (void) nvlist_lookup_uint64(props, | |
6658 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
6659 | if (readonly) | |
6660 | mode = SPA_MODE_READ; | |
6661 | spa = spa_add(pool, config, altroot); | |
6662 | spa->spa_import_flags = flags; | |
6663 | ||
6664 | /* | |
6665 | * Verbatim import - Take a pool and insert it into the namespace | |
6666 | * as if it had been loaded at boot. | |
6667 | */ | |
6668 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
6669 | if (props != NULL) | |
6670 | spa_configfile_set(spa, props, B_FALSE); | |
6671 | ||
6672 | spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE); | |
6673 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); | |
6674 | zfs_dbgmsg("spa_import: verbatim import of %s", pool); | |
6675 | mutex_exit(&spa_namespace_lock); | |
6676 | return (0); | |
6677 | } | |
6678 | ||
6679 | spa_activate(spa, mode); | |
6680 | ||
6681 | /* | |
6682 | * Don't start async tasks until we know everything is healthy. | |
6683 | */ | |
6684 | spa_async_suspend(spa); | |
6685 | ||
6686 | zpool_get_load_policy(config, &policy); | |
6687 | if (policy.zlp_rewind & ZPOOL_DO_REWIND) | |
6688 | state = SPA_LOAD_RECOVER; | |
6689 | ||
6690 | spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; | |
6691 | ||
6692 | if (state != SPA_LOAD_RECOVER) { | |
6693 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
6694 | zfs_dbgmsg("spa_import: importing %s", pool); | |
6695 | } else { | |
6696 | zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " | |
6697 | "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); | |
6698 | } | |
6699 | error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); | |
6700 | ||
6701 | /* | |
6702 | * Propagate anything learned while loading the pool and pass it | |
6703 | * back to caller (i.e. rewind info, missing devices, etc). | |
6704 | */ | |
6705 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info); | |
6706 | ||
6707 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6708 | /* | |
6709 | * Toss any existing sparelist, as it doesn't have any validity | |
6710 | * anymore, and conflicts with spa_has_spare(). | |
6711 | */ | |
6712 | if (spa->spa_spares.sav_config) { | |
6713 | nvlist_free(spa->spa_spares.sav_config); | |
6714 | spa->spa_spares.sav_config = NULL; | |
6715 | spa_load_spares(spa); | |
6716 | } | |
6717 | if (spa->spa_l2cache.sav_config) { | |
6718 | nvlist_free(spa->spa_l2cache.sav_config); | |
6719 | spa->spa_l2cache.sav_config = NULL; | |
6720 | spa_load_l2cache(spa); | |
6721 | } | |
6722 | ||
6723 | nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE); | |
6724 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6725 | ||
6726 | if (props != NULL) | |
6727 | spa_configfile_set(spa, props, B_FALSE); | |
6728 | ||
6729 | if (error != 0 || (props && spa_writeable(spa) && | |
6730 | (error = spa_prop_set(spa, props)))) { | |
6731 | spa_unload(spa); | |
6732 | spa_deactivate(spa); | |
6733 | spa_remove(spa); | |
6734 | mutex_exit(&spa_namespace_lock); | |
6735 | return (error); | |
6736 | } | |
6737 | ||
6738 | spa_async_resume(spa); | |
6739 | ||
6740 | /* | |
6741 | * Override any spares and level 2 cache devices as specified by | |
6742 | * the user, as these may have correct device names/devids, etc. | |
6743 | */ | |
6744 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
6745 | &spares, &nspares) == 0) { | |
6746 | if (spa->spa_spares.sav_config) | |
6747 | fnvlist_remove(spa->spa_spares.sav_config, | |
6748 | ZPOOL_CONFIG_SPARES); | |
6749 | else | |
6750 | spa->spa_spares.sav_config = fnvlist_alloc(); | |
6751 | fnvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
6752 | ZPOOL_CONFIG_SPARES, (const nvlist_t * const *)spares, | |
6753 | nspares); | |
6754 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6755 | spa_load_spares(spa); | |
6756 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6757 | spa->spa_spares.sav_sync = B_TRUE; | |
6758 | } | |
6759 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
6760 | &l2cache, &nl2cache) == 0) { | |
6761 | if (spa->spa_l2cache.sav_config) | |
6762 | fnvlist_remove(spa->spa_l2cache.sav_config, | |
6763 | ZPOOL_CONFIG_L2CACHE); | |
6764 | else | |
6765 | spa->spa_l2cache.sav_config = fnvlist_alloc(); | |
6766 | fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
6767 | ZPOOL_CONFIG_L2CACHE, (const nvlist_t * const *)l2cache, | |
6768 | nl2cache); | |
6769 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6770 | spa_load_l2cache(spa); | |
6771 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6772 | spa->spa_l2cache.sav_sync = B_TRUE; | |
6773 | } | |
6774 | ||
6775 | /* | |
6776 | * Check for any removed devices. | |
6777 | */ | |
6778 | if (spa->spa_autoreplace) { | |
6779 | spa_aux_check_removed(&spa->spa_spares); | |
6780 | spa_aux_check_removed(&spa->spa_l2cache); | |
6781 | } | |
6782 | ||
6783 | if (spa_writeable(spa)) { | |
6784 | /* | |
6785 | * Update the config cache to include the newly-imported pool. | |
6786 | */ | |
6787 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
6788 | } | |
6789 | ||
6790 | /* | |
6791 | * It's possible that the pool was expanded while it was exported. | |
6792 | * We kick off an async task to handle this for us. | |
6793 | */ | |
6794 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); | |
6795 | ||
6796 | spa_history_log_version(spa, "import", NULL); | |
6797 | ||
6798 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); | |
6799 | ||
6800 | mutex_exit(&spa_namespace_lock); | |
6801 | ||
6802 | zvol_create_minors_recursive(pool); | |
6803 | ||
6804 | spa_import_os(spa); | |
6805 | ||
6806 | return (0); | |
6807 | } | |
6808 | ||
6809 | nvlist_t * | |
6810 | spa_tryimport(nvlist_t *tryconfig) | |
6811 | { | |
6812 | nvlist_t *config = NULL; | |
6813 | const char *poolname, *cachefile; | |
6814 | spa_t *spa; | |
6815 | uint64_t state; | |
6816 | int error; | |
6817 | zpool_load_policy_t policy; | |
6818 | ||
6819 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
6820 | return (NULL); | |
6821 | ||
6822 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
6823 | return (NULL); | |
6824 | ||
6825 | /* | |
6826 | * Create and initialize the spa structure. | |
6827 | */ | |
6828 | char *name = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
6829 | (void) snprintf(name, MAXPATHLEN, "%s-%llx-%s", | |
6830 | TRYIMPORT_NAME, (u_longlong_t)curthread, poolname); | |
6831 | ||
6832 | mutex_enter(&spa_namespace_lock); | |
6833 | spa = spa_add(name, tryconfig, NULL); | |
6834 | spa_activate(spa, SPA_MODE_READ); | |
6835 | kmem_free(name, MAXPATHLEN); | |
6836 | ||
6837 | /* | |
6838 | * Rewind pool if a max txg was provided. | |
6839 | */ | |
6840 | zpool_get_load_policy(spa->spa_config, &policy); | |
6841 | if (policy.zlp_txg != UINT64_MAX) { | |
6842 | spa->spa_load_max_txg = policy.zlp_txg; | |
6843 | spa->spa_extreme_rewind = B_TRUE; | |
6844 | zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", | |
6845 | poolname, (longlong_t)policy.zlp_txg); | |
6846 | } else { | |
6847 | zfs_dbgmsg("spa_tryimport: importing %s", poolname); | |
6848 | } | |
6849 | ||
6850 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) | |
6851 | == 0) { | |
6852 | zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); | |
6853 | spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; | |
6854 | } else { | |
6855 | spa->spa_config_source = SPA_CONFIG_SRC_SCAN; | |
6856 | } | |
6857 | ||
6858 | /* | |
6859 | * spa_import() relies on a pool config fetched by spa_try_import() | |
6860 | * for spare/cache devices. Import flags are not passed to | |
6861 | * spa_tryimport(), which makes it return early due to a missing log | |
6862 | * device and missing retrieving the cache device and spare eventually. | |
6863 | * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch | |
6864 | * the correct configuration regardless of the missing log device. | |
6865 | */ | |
6866 | spa->spa_import_flags |= ZFS_IMPORT_MISSING_LOG; | |
6867 | ||
6868 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); | |
6869 | ||
6870 | /* | |
6871 | * If 'tryconfig' was at least parsable, return the current config. | |
6872 | */ | |
6873 | if (spa->spa_root_vdev != NULL) { | |
6874 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
6875 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname); | |
6876 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state); | |
6877 | fnvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
6878 | spa->spa_uberblock.ub_timestamp); | |
6879 | fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
6880 | spa->spa_load_info); | |
6881 | fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
6882 | spa->spa_errata); | |
6883 | ||
6884 | /* | |
6885 | * If the bootfs property exists on this pool then we | |
6886 | * copy it out so that external consumers can tell which | |
6887 | * pools are bootable. | |
6888 | */ | |
6889 | if ((!error || error == EEXIST) && spa->spa_bootfs) { | |
6890 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
6891 | ||
6892 | /* | |
6893 | * We have to play games with the name since the | |
6894 | * pool was opened as TRYIMPORT_NAME. | |
6895 | */ | |
6896 | if (dsl_dsobj_to_dsname(spa_name(spa), | |
6897 | spa->spa_bootfs, tmpname) == 0) { | |
6898 | char *cp; | |
6899 | char *dsname; | |
6900 | ||
6901 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
6902 | ||
6903 | cp = strchr(tmpname, '/'); | |
6904 | if (cp == NULL) { | |
6905 | (void) strlcpy(dsname, tmpname, | |
6906 | MAXPATHLEN); | |
6907 | } else { | |
6908 | (void) snprintf(dsname, MAXPATHLEN, | |
6909 | "%s/%s", poolname, ++cp); | |
6910 | } | |
6911 | fnvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, | |
6912 | dsname); | |
6913 | kmem_free(dsname, MAXPATHLEN); | |
6914 | } | |
6915 | kmem_free(tmpname, MAXPATHLEN); | |
6916 | } | |
6917 | ||
6918 | /* | |
6919 | * Add the list of hot spares and level 2 cache devices. | |
6920 | */ | |
6921 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6922 | spa_add_spares(spa, config); | |
6923 | spa_add_l2cache(spa, config); | |
6924 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
6925 | } | |
6926 | ||
6927 | spa_unload(spa); | |
6928 | spa_deactivate(spa); | |
6929 | spa_remove(spa); | |
6930 | mutex_exit(&spa_namespace_lock); | |
6931 | ||
6932 | return (config); | |
6933 | } | |
6934 | ||
6935 | /* | |
6936 | * Pool export/destroy | |
6937 | * | |
6938 | * The act of destroying or exporting a pool is very simple. We make sure there | |
6939 | * is no more pending I/O and any references to the pool are gone. Then, we | |
6940 | * update the pool state and sync all the labels to disk, removing the | |
6941 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then | |
6942 | * we don't sync the labels or remove the configuration cache. | |
6943 | */ | |
6944 | static int | |
6945 | spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig, | |
6946 | boolean_t force, boolean_t hardforce) | |
6947 | { | |
6948 | int error = 0; | |
6949 | spa_t *spa; | |
6950 | hrtime_t export_start = gethrtime(); | |
6951 | ||
6952 | if (oldconfig) | |
6953 | *oldconfig = NULL; | |
6954 | ||
6955 | if (!(spa_mode_global & SPA_MODE_WRITE)) | |
6956 | return (SET_ERROR(EROFS)); | |
6957 | ||
6958 | mutex_enter(&spa_namespace_lock); | |
6959 | if ((spa = spa_lookup(pool)) == NULL) { | |
6960 | mutex_exit(&spa_namespace_lock); | |
6961 | return (SET_ERROR(ENOENT)); | |
6962 | } | |
6963 | ||
6964 | if (spa->spa_is_exporting) { | |
6965 | /* the pool is being exported by another thread */ | |
6966 | mutex_exit(&spa_namespace_lock); | |
6967 | return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS)); | |
6968 | } | |
6969 | spa->spa_is_exporting = B_TRUE; | |
6970 | ||
6971 | /* | |
6972 | * Put a hold on the pool, drop the namespace lock, stop async tasks | |
6973 | * and see if we can export. | |
6974 | */ | |
6975 | spa_open_ref(spa, FTAG); | |
6976 | mutex_exit(&spa_namespace_lock); | |
6977 | spa_async_suspend(spa); | |
6978 | if (spa->spa_zvol_taskq) { | |
6979 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
6980 | taskq_wait(spa->spa_zvol_taskq); | |
6981 | } | |
6982 | mutex_enter(&spa_namespace_lock); | |
6983 | spa->spa_export_thread = curthread; | |
6984 | spa_close(spa, FTAG); | |
6985 | ||
6986 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
6987 | mutex_exit(&spa_namespace_lock); | |
6988 | goto export_spa; | |
6989 | } | |
6990 | ||
6991 | /* | |
6992 | * The pool will be in core if it's openable, in which case we can | |
6993 | * modify its state. Objsets may be open only because they're dirty, | |
6994 | * so we have to force it to sync before checking spa_refcnt. | |
6995 | */ | |
6996 | if (spa->spa_sync_on) { | |
6997 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
6998 | spa_evicting_os_wait(spa); | |
6999 | } | |
7000 | ||
7001 | /* | |
7002 | * A pool cannot be exported or destroyed if there are active | |
7003 | * references. If we are resetting a pool, allow references by | |
7004 | * fault injection handlers. | |
7005 | */ | |
7006 | if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0)) { | |
7007 | error = SET_ERROR(EBUSY); | |
7008 | goto fail; | |
7009 | } | |
7010 | ||
7011 | mutex_exit(&spa_namespace_lock); | |
7012 | /* | |
7013 | * At this point we no longer hold the spa_namespace_lock and | |
7014 | * there were no references on the spa. Future spa_lookups will | |
7015 | * notice the spa->spa_export_thread and wait until we signal | |
7016 | * that we are finshed. | |
7017 | */ | |
7018 | ||
7019 | if (spa->spa_sync_on) { | |
7020 | vdev_t *rvd = spa->spa_root_vdev; | |
7021 | /* | |
7022 | * A pool cannot be exported if it has an active shared spare. | |
7023 | * This is to prevent other pools stealing the active spare | |
7024 | * from an exported pool. At user's own will, such pool can | |
7025 | * be forcedly exported. | |
7026 | */ | |
7027 | if (!force && new_state == POOL_STATE_EXPORTED && | |
7028 | spa_has_active_shared_spare(spa)) { | |
7029 | error = SET_ERROR(EXDEV); | |
7030 | mutex_enter(&spa_namespace_lock); | |
7031 | goto fail; | |
7032 | } | |
7033 | ||
7034 | /* | |
7035 | * We're about to export or destroy this pool. Make sure | |
7036 | * we stop all initialization and trim activity here before | |
7037 | * we set the spa_final_txg. This will ensure that all | |
7038 | * dirty data resulting from the initialization is | |
7039 | * committed to disk before we unload the pool. | |
7040 | */ | |
7041 | vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE); | |
7042 | vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE); | |
7043 | vdev_autotrim_stop_all(spa); | |
7044 | vdev_rebuild_stop_all(spa); | |
7045 | ||
7046 | /* | |
7047 | * We want this to be reflected on every label, | |
7048 | * so mark them all dirty. spa_unload() will do the | |
7049 | * final sync that pushes these changes out. | |
7050 | */ | |
7051 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
7052 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
7053 | spa->spa_state = new_state; | |
7054 | vdev_config_dirty(rvd); | |
7055 | spa_config_exit(spa, SCL_ALL, FTAG); | |
7056 | } | |
7057 | ||
7058 | /* | |
7059 | * If the log space map feature is enabled and the pool is | |
7060 | * getting exported (but not destroyed), we want to spend some | |
7061 | * time flushing as many metaslabs as we can in an attempt to | |
7062 | * destroy log space maps and save import time. This has to be | |
7063 | * done before we set the spa_final_txg, otherwise | |
7064 | * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs. | |
7065 | * spa_should_flush_logs_on_unload() should be called after | |
7066 | * spa_state has been set to the new_state. | |
7067 | */ | |
7068 | if (spa_should_flush_logs_on_unload(spa)) | |
7069 | spa_unload_log_sm_flush_all(spa); | |
7070 | ||
7071 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
7072 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
7073 | spa->spa_final_txg = spa_last_synced_txg(spa) + | |
7074 | TXG_DEFER_SIZE + 1; | |
7075 | spa_config_exit(spa, SCL_ALL, FTAG); | |
7076 | } | |
7077 | } | |
7078 | ||
7079 | export_spa: | |
7080 | spa_export_os(spa); | |
7081 | ||
7082 | if (new_state == POOL_STATE_DESTROYED) | |
7083 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
7084 | else if (new_state == POOL_STATE_EXPORTED) | |
7085 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
7086 | ||
7087 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
7088 | spa_unload(spa); | |
7089 | spa_deactivate(spa); | |
7090 | } | |
7091 | ||
7092 | if (oldconfig && spa->spa_config) | |
7093 | *oldconfig = fnvlist_dup(spa->spa_config); | |
7094 | ||
7095 | if (new_state == POOL_STATE_EXPORTED) | |
7096 | zio_handle_export_delay(spa, gethrtime() - export_start); | |
7097 | ||
7098 | /* | |
7099 | * Take the namespace lock for the actual spa_t removal | |
7100 | */ | |
7101 | mutex_enter(&spa_namespace_lock); | |
7102 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
7103 | if (!hardforce) | |
7104 | spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE); | |
7105 | spa_remove(spa); | |
7106 | } else { | |
7107 | /* | |
7108 | * If spa_remove() is not called for this spa_t and | |
7109 | * there is any possibility that it can be reused, | |
7110 | * we make sure to reset the exporting flag. | |
7111 | */ | |
7112 | spa->spa_is_exporting = B_FALSE; | |
7113 | spa->spa_export_thread = NULL; | |
7114 | } | |
7115 | ||
7116 | /* | |
7117 | * Wake up any waiters in spa_lookup() | |
7118 | */ | |
7119 | cv_broadcast(&spa_namespace_cv); | |
7120 | mutex_exit(&spa_namespace_lock); | |
7121 | return (0); | |
7122 | ||
7123 | fail: | |
7124 | spa->spa_is_exporting = B_FALSE; | |
7125 | spa->spa_export_thread = NULL; | |
7126 | ||
7127 | spa_async_resume(spa); | |
7128 | /* | |
7129 | * Wake up any waiters in spa_lookup() | |
7130 | */ | |
7131 | cv_broadcast(&spa_namespace_cv); | |
7132 | mutex_exit(&spa_namespace_lock); | |
7133 | return (error); | |
7134 | } | |
7135 | ||
7136 | /* | |
7137 | * Destroy a storage pool. | |
7138 | */ | |
7139 | int | |
7140 | spa_destroy(const char *pool) | |
7141 | { | |
7142 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, | |
7143 | B_FALSE, B_FALSE)); | |
7144 | } | |
7145 | ||
7146 | /* | |
7147 | * Export a storage pool. | |
7148 | */ | |
7149 | int | |
7150 | spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force, | |
7151 | boolean_t hardforce) | |
7152 | { | |
7153 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, | |
7154 | force, hardforce)); | |
7155 | } | |
7156 | ||
7157 | /* | |
7158 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
7159 | * from the namespace in any way. | |
7160 | */ | |
7161 | int | |
7162 | spa_reset(const char *pool) | |
7163 | { | |
7164 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, | |
7165 | B_FALSE, B_FALSE)); | |
7166 | } | |
7167 | ||
7168 | /* | |
7169 | * ========================================================================== | |
7170 | * Device manipulation | |
7171 | * ========================================================================== | |
7172 | */ | |
7173 | ||
7174 | /* | |
7175 | * This is called as a synctask to increment the draid feature flag | |
7176 | */ | |
7177 | static void | |
7178 | spa_draid_feature_incr(void *arg, dmu_tx_t *tx) | |
7179 | { | |
7180 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
7181 | int draid = (int)(uintptr_t)arg; | |
7182 | ||
7183 | for (int c = 0; c < draid; c++) | |
7184 | spa_feature_incr(spa, SPA_FEATURE_DRAID, tx); | |
7185 | } | |
7186 | ||
7187 | /* | |
7188 | * Add a device to a storage pool. | |
7189 | */ | |
7190 | int | |
7191 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot, boolean_t check_ashift) | |
7192 | { | |
7193 | uint64_t txg, ndraid = 0; | |
7194 | int error; | |
7195 | vdev_t *rvd = spa->spa_root_vdev; | |
7196 | vdev_t *vd, *tvd; | |
7197 | nvlist_t **spares, **l2cache; | |
7198 | uint_t nspares, nl2cache; | |
7199 | ||
7200 | ASSERT(spa_writeable(spa)); | |
7201 | ||
7202 | txg = spa_vdev_enter(spa); | |
7203 | ||
7204 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
7205 | VDEV_ALLOC_ADD)) != 0) | |
7206 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7207 | ||
7208 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ | |
7209 | ||
7210 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
7211 | &nspares) != 0) | |
7212 | nspares = 0; | |
7213 | ||
7214 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
7215 | &nl2cache) != 0) | |
7216 | nl2cache = 0; | |
7217 | ||
7218 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) | |
7219 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
7220 | ||
7221 | if (vd->vdev_children != 0 && | |
7222 | (error = vdev_create(vd, txg, B_FALSE)) != 0) { | |
7223 | return (spa_vdev_exit(spa, vd, txg, error)); | |
7224 | } | |
7225 | ||
7226 | /* | |
7227 | * The virtual dRAID spares must be added after vdev tree is created | |
7228 | * and the vdev guids are generated. The guid of their associated | |
7229 | * dRAID is stored in the config and used when opening the spare. | |
7230 | */ | |
7231 | if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid, | |
7232 | rvd->vdev_children)) == 0) { | |
7233 | if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot, | |
7234 | ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) | |
7235 | nspares = 0; | |
7236 | } else { | |
7237 | return (spa_vdev_exit(spa, vd, txg, error)); | |
7238 | } | |
7239 | ||
7240 | /* | |
7241 | * We must validate the spares and l2cache devices after checking the | |
7242 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
7243 | */ | |
7244 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) | |
7245 | return (spa_vdev_exit(spa, vd, txg, error)); | |
7246 | ||
7247 | /* | |
7248 | * If we are in the middle of a device removal, we can only add | |
7249 | * devices which match the existing devices in the pool. | |
7250 | * If we are in the middle of a removal, or have some indirect | |
7251 | * vdevs, we can not add raidz or dRAID top levels. | |
7252 | */ | |
7253 | if (spa->spa_vdev_removal != NULL || | |
7254 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
7255 | for (int c = 0; c < vd->vdev_children; c++) { | |
7256 | tvd = vd->vdev_child[c]; | |
7257 | if (spa->spa_vdev_removal != NULL && | |
7258 | tvd->vdev_ashift != spa->spa_max_ashift) { | |
7259 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
7260 | } | |
7261 | /* Fail if top level vdev is raidz or a dRAID */ | |
7262 | if (vdev_get_nparity(tvd) != 0) | |
7263 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
7264 | ||
7265 | /* | |
7266 | * Need the top level mirror to be | |
7267 | * a mirror of leaf vdevs only | |
7268 | */ | |
7269 | if (tvd->vdev_ops == &vdev_mirror_ops) { | |
7270 | for (uint64_t cid = 0; | |
7271 | cid < tvd->vdev_children; cid++) { | |
7272 | vdev_t *cvd = tvd->vdev_child[cid]; | |
7273 | if (!cvd->vdev_ops->vdev_op_leaf) { | |
7274 | return (spa_vdev_exit(spa, vd, | |
7275 | txg, EINVAL)); | |
7276 | } | |
7277 | } | |
7278 | } | |
7279 | } | |
7280 | } | |
7281 | ||
7282 | if (check_ashift && spa->spa_max_ashift == spa->spa_min_ashift) { | |
7283 | for (int c = 0; c < vd->vdev_children; c++) { | |
7284 | tvd = vd->vdev_child[c]; | |
7285 | if (tvd->vdev_ashift != spa->spa_max_ashift) { | |
7286 | return (spa_vdev_exit(spa, vd, txg, | |
7287 | ZFS_ERR_ASHIFT_MISMATCH)); | |
7288 | } | |
7289 | } | |
7290 | } | |
7291 | ||
7292 | for (int c = 0; c < vd->vdev_children; c++) { | |
7293 | tvd = vd->vdev_child[c]; | |
7294 | vdev_remove_child(vd, tvd); | |
7295 | tvd->vdev_id = rvd->vdev_children; | |
7296 | vdev_add_child(rvd, tvd); | |
7297 | vdev_config_dirty(tvd); | |
7298 | } | |
7299 | ||
7300 | if (nspares != 0) { | |
7301 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
7302 | ZPOOL_CONFIG_SPARES); | |
7303 | spa_load_spares(spa); | |
7304 | spa->spa_spares.sav_sync = B_TRUE; | |
7305 | } | |
7306 | ||
7307 | if (nl2cache != 0) { | |
7308 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
7309 | ZPOOL_CONFIG_L2CACHE); | |
7310 | spa_load_l2cache(spa); | |
7311 | spa->spa_l2cache.sav_sync = B_TRUE; | |
7312 | } | |
7313 | ||
7314 | /* | |
7315 | * We can't increment a feature while holding spa_vdev so we | |
7316 | * have to do it in a synctask. | |
7317 | */ | |
7318 | if (ndraid != 0) { | |
7319 | dmu_tx_t *tx; | |
7320 | ||
7321 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
7322 | dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr, | |
7323 | (void *)(uintptr_t)ndraid, tx); | |
7324 | dmu_tx_commit(tx); | |
7325 | } | |
7326 | ||
7327 | /* | |
7328 | * We have to be careful when adding new vdevs to an existing pool. | |
7329 | * If other threads start allocating from these vdevs before we | |
7330 | * sync the config cache, and we lose power, then upon reboot we may | |
7331 | * fail to open the pool because there are DVAs that the config cache | |
7332 | * can't translate. Therefore, we first add the vdevs without | |
7333 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
7334 | * and then let spa_config_update() initialize the new metaslabs. | |
7335 | * | |
7336 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
7337 | * if we lose power at any point in this sequence, the remaining | |
7338 | * steps will be completed the next time we load the pool. | |
7339 | */ | |
7340 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
7341 | ||
7342 | mutex_enter(&spa_namespace_lock); | |
7343 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
7344 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); | |
7345 | mutex_exit(&spa_namespace_lock); | |
7346 | ||
7347 | return (0); | |
7348 | } | |
7349 | ||
7350 | /* | |
7351 | * Attach a device to a vdev specified by its guid. The vdev type can be | |
7352 | * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a | |
7353 | * single device). When the vdev is a single device, a mirror vdev will be | |
7354 | * automatically inserted. | |
7355 | * | |
7356 | * If 'replacing' is specified, the new device is intended to replace the | |
7357 | * existing device; in this case the two devices are made into their own | |
7358 | * mirror using the 'replacing' vdev, which is functionally identical to | |
7359 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
7360 | * extra rules: you can't attach to it after it's been created, and upon | |
7361 | * completion of resilvering, the first disk (the one being replaced) | |
7362 | * is automatically detached. | |
7363 | * | |
7364 | * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild) | |
7365 | * should be performed instead of traditional healing reconstruction. From | |
7366 | * an administrators perspective these are both resilver operations. | |
7367 | */ | |
7368 | int | |
7369 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing, | |
7370 | int rebuild) | |
7371 | { | |
7372 | uint64_t txg, dtl_max_txg; | |
7373 | vdev_t *rvd = spa->spa_root_vdev; | |
7374 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
7375 | vdev_ops_t *pvops; | |
7376 | char *oldvdpath, *newvdpath; | |
7377 | int newvd_isspare = B_FALSE; | |
7378 | int error; | |
7379 | ||
7380 | ASSERT(spa_writeable(spa)); | |
7381 | ||
7382 | txg = spa_vdev_enter(spa); | |
7383 | ||
7384 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7385 | ||
7386 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7387 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7388 | error = (spa_has_checkpoint(spa)) ? | |
7389 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7390 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7391 | } | |
7392 | ||
7393 | if (rebuild) { | |
7394 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD)) | |
7395 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7396 | ||
7397 | if (dsl_scan_resilvering(spa_get_dsl(spa)) || | |
7398 | dsl_scan_resilver_scheduled(spa_get_dsl(spa))) { | |
7399 | return (spa_vdev_exit(spa, NULL, txg, | |
7400 | ZFS_ERR_RESILVER_IN_PROGRESS)); | |
7401 | } | |
7402 | } else { | |
7403 | if (vdev_rebuild_active(rvd)) | |
7404 | return (spa_vdev_exit(spa, NULL, txg, | |
7405 | ZFS_ERR_REBUILD_IN_PROGRESS)); | |
7406 | } | |
7407 | ||
7408 | if (spa->spa_vdev_removal != NULL) { | |
7409 | return (spa_vdev_exit(spa, NULL, txg, | |
7410 | ZFS_ERR_DEVRM_IN_PROGRESS)); | |
7411 | } | |
7412 | ||
7413 | if (oldvd == NULL) | |
7414 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7415 | ||
7416 | boolean_t raidz = oldvd->vdev_ops == &vdev_raidz_ops; | |
7417 | ||
7418 | if (raidz) { | |
7419 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_RAIDZ_EXPANSION)) | |
7420 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7421 | ||
7422 | /* | |
7423 | * Can't expand a raidz while prior expand is in progress. | |
7424 | */ | |
7425 | if (spa->spa_raidz_expand != NULL) { | |
7426 | return (spa_vdev_exit(spa, NULL, txg, | |
7427 | ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS)); | |
7428 | } | |
7429 | } else if (!oldvd->vdev_ops->vdev_op_leaf) { | |
7430 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7431 | } | |
7432 | ||
7433 | if (raidz) | |
7434 | pvd = oldvd; | |
7435 | else | |
7436 | pvd = oldvd->vdev_parent; | |
7437 | ||
7438 | if (spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, | |
7439 | VDEV_ALLOC_ATTACH) != 0) | |
7440 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
7441 | ||
7442 | if (newrootvd->vdev_children != 1) | |
7443 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7444 | ||
7445 | newvd = newrootvd->vdev_child[0]; | |
7446 | ||
7447 | if (!newvd->vdev_ops->vdev_op_leaf) | |
7448 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
7449 | ||
7450 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
7451 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
7452 | ||
7453 | /* | |
7454 | * log, dedup and special vdevs should not be replaced by spares. | |
7455 | */ | |
7456 | if ((oldvd->vdev_top->vdev_alloc_bias != VDEV_BIAS_NONE || | |
7457 | oldvd->vdev_top->vdev_islog) && newvd->vdev_isspare) { | |
7458 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7459 | } | |
7460 | ||
7461 | /* | |
7462 | * A dRAID spare can only replace a child of its parent dRAID vdev. | |
7463 | */ | |
7464 | if (newvd->vdev_ops == &vdev_draid_spare_ops && | |
7465 | oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) { | |
7466 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7467 | } | |
7468 | ||
7469 | if (rebuild) { | |
7470 | /* | |
7471 | * For rebuilds, the top vdev must support reconstruction | |
7472 | * using only space maps. This means the only allowable | |
7473 | * vdevs types are the root vdev, a mirror, or dRAID. | |
7474 | */ | |
7475 | tvd = pvd; | |
7476 | if (pvd->vdev_top != NULL) | |
7477 | tvd = pvd->vdev_top; | |
7478 | ||
7479 | if (tvd->vdev_ops != &vdev_mirror_ops && | |
7480 | tvd->vdev_ops != &vdev_root_ops && | |
7481 | tvd->vdev_ops != &vdev_draid_ops) { | |
7482 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7483 | } | |
7484 | } | |
7485 | ||
7486 | if (!replacing) { | |
7487 | /* | |
7488 | * For attach, the only allowable parent is a mirror or | |
7489 | * the root vdev. A raidz vdev can be attached to, but | |
7490 | * you cannot attach to a raidz child. | |
7491 | */ | |
7492 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
7493 | pvd->vdev_ops != &vdev_root_ops && | |
7494 | !raidz) | |
7495 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7496 | ||
7497 | pvops = &vdev_mirror_ops; | |
7498 | } else { | |
7499 | /* | |
7500 | * Active hot spares can only be replaced by inactive hot | |
7501 | * spares. | |
7502 | */ | |
7503 | if (pvd->vdev_ops == &vdev_spare_ops && | |
7504 | oldvd->vdev_isspare && | |
7505 | !spa_has_spare(spa, newvd->vdev_guid)) | |
7506 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7507 | ||
7508 | /* | |
7509 | * If the source is a hot spare, and the parent isn't already a | |
7510 | * spare, then we want to create a new hot spare. Otherwise, we | |
7511 | * want to create a replacing vdev. The user is not allowed to | |
7512 | * attach to a spared vdev child unless the 'isspare' state is | |
7513 | * the same (spare replaces spare, non-spare replaces | |
7514 | * non-spare). | |
7515 | */ | |
7516 | if (pvd->vdev_ops == &vdev_replacing_ops && | |
7517 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
7518 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7519 | } else if (pvd->vdev_ops == &vdev_spare_ops && | |
7520 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
7521 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7522 | } | |
7523 | ||
7524 | if (newvd->vdev_isspare) | |
7525 | pvops = &vdev_spare_ops; | |
7526 | else | |
7527 | pvops = &vdev_replacing_ops; | |
7528 | } | |
7529 | ||
7530 | /* | |
7531 | * Make sure the new device is big enough. | |
7532 | */ | |
7533 | vdev_t *min_vdev = raidz ? oldvd->vdev_child[0] : oldvd; | |
7534 | if (newvd->vdev_asize < vdev_get_min_asize(min_vdev)) | |
7535 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); | |
7536 | ||
7537 | /* | |
7538 | * The new device cannot have a higher alignment requirement | |
7539 | * than the top-level vdev. | |
7540 | */ | |
7541 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
7542 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7543 | ||
7544 | /* | |
7545 | * RAIDZ-expansion-specific checks. | |
7546 | */ | |
7547 | if (raidz) { | |
7548 | if (vdev_raidz_attach_check(newvd) != 0) | |
7549 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
7550 | ||
7551 | /* | |
7552 | * Fail early if a child is not healthy or being replaced | |
7553 | */ | |
7554 | for (int i = 0; i < oldvd->vdev_children; i++) { | |
7555 | if (vdev_is_dead(oldvd->vdev_child[i]) || | |
7556 | !oldvd->vdev_child[i]->vdev_ops->vdev_op_leaf) { | |
7557 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7558 | ENXIO)); | |
7559 | } | |
7560 | /* Also fail if reserved boot area is in-use */ | |
7561 | if (vdev_check_boot_reserve(spa, oldvd->vdev_child[i]) | |
7562 | != 0) { | |
7563 | return (spa_vdev_exit(spa, newrootvd, txg, | |
7564 | EADDRINUSE)); | |
7565 | } | |
7566 | } | |
7567 | } | |
7568 | ||
7569 | if (raidz) { | |
7570 | /* | |
7571 | * Note: oldvdpath is freed by spa_strfree(), but | |
7572 | * kmem_asprintf() is freed by kmem_strfree(), so we have to | |
7573 | * move it to a spa_strdup-ed string. | |
7574 | */ | |
7575 | char *tmp = kmem_asprintf("raidz%u-%u", | |
7576 | (uint_t)vdev_get_nparity(oldvd), (uint_t)oldvd->vdev_id); | |
7577 | oldvdpath = spa_strdup(tmp); | |
7578 | kmem_strfree(tmp); | |
7579 | } else { | |
7580 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
7581 | } | |
7582 | newvdpath = spa_strdup(newvd->vdev_path); | |
7583 | ||
7584 | /* | |
7585 | * If this is an in-place replacement, update oldvd's path and devid | |
7586 | * to make it distinguishable from newvd, and unopenable from now on. | |
7587 | */ | |
7588 | if (strcmp(oldvdpath, newvdpath) == 0) { | |
7589 | spa_strfree(oldvd->vdev_path); | |
7590 | oldvd->vdev_path = kmem_alloc(strlen(newvdpath) + 5, | |
7591 | KM_SLEEP); | |
7592 | (void) sprintf(oldvd->vdev_path, "%s/old", | |
7593 | newvdpath); | |
7594 | if (oldvd->vdev_devid != NULL) { | |
7595 | spa_strfree(oldvd->vdev_devid); | |
7596 | oldvd->vdev_devid = NULL; | |
7597 | } | |
7598 | spa_strfree(oldvdpath); | |
7599 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
7600 | } | |
7601 | ||
7602 | /* | |
7603 | * If the parent is not a mirror, or if we're replacing, insert the new | |
7604 | * mirror/replacing/spare vdev above oldvd. | |
7605 | */ | |
7606 | if (!raidz && pvd->vdev_ops != pvops) { | |
7607 | pvd = vdev_add_parent(oldvd, pvops); | |
7608 | ASSERT(pvd->vdev_ops == pvops); | |
7609 | ASSERT(oldvd->vdev_parent == pvd); | |
7610 | } | |
7611 | ||
7612 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
7613 | ||
7614 | /* | |
7615 | * Extract the new device from its root and add it to pvd. | |
7616 | */ | |
7617 | vdev_remove_child(newrootvd, newvd); | |
7618 | newvd->vdev_id = pvd->vdev_children; | |
7619 | newvd->vdev_crtxg = oldvd->vdev_crtxg; | |
7620 | vdev_add_child(pvd, newvd); | |
7621 | ||
7622 | /* | |
7623 | * Reevaluate the parent vdev state. | |
7624 | */ | |
7625 | vdev_propagate_state(pvd); | |
7626 | ||
7627 | tvd = newvd->vdev_top; | |
7628 | ASSERT(pvd->vdev_top == tvd); | |
7629 | ASSERT(tvd->vdev_parent == rvd); | |
7630 | ||
7631 | vdev_config_dirty(tvd); | |
7632 | ||
7633 | /* | |
7634 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account | |
7635 | * for any dmu_sync-ed blocks. It will propagate upward when | |
7636 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
7637 | */ | |
7638 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; | |
7639 | ||
7640 | if (raidz) { | |
7641 | /* | |
7642 | * Wait for the youngest allocations and frees to sync, | |
7643 | * and then wait for the deferral of those frees to finish. | |
7644 | */ | |
7645 | spa_vdev_config_exit(spa, NULL, | |
7646 | txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); | |
7647 | ||
7648 | vdev_initialize_stop_all(tvd, VDEV_INITIALIZE_ACTIVE); | |
7649 | vdev_trim_stop_all(tvd, VDEV_TRIM_ACTIVE); | |
7650 | vdev_autotrim_stop_wait(tvd); | |
7651 | ||
7652 | dtl_max_txg = spa_vdev_config_enter(spa); | |
7653 | ||
7654 | tvd->vdev_rz_expanding = B_TRUE; | |
7655 | ||
7656 | vdev_dirty_leaves(tvd, VDD_DTL, dtl_max_txg); | |
7657 | vdev_config_dirty(tvd); | |
7658 | ||
7659 | dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, | |
7660 | dtl_max_txg); | |
7661 | dsl_sync_task_nowait(spa->spa_dsl_pool, vdev_raidz_attach_sync, | |
7662 | newvd, tx); | |
7663 | dmu_tx_commit(tx); | |
7664 | } else { | |
7665 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, | |
7666 | dtl_max_txg - TXG_INITIAL); | |
7667 | ||
7668 | if (newvd->vdev_isspare) { | |
7669 | spa_spare_activate(newvd); | |
7670 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); | |
7671 | } | |
7672 | ||
7673 | newvd_isspare = newvd->vdev_isspare; | |
7674 | ||
7675 | /* | |
7676 | * Mark newvd's DTL dirty in this txg. | |
7677 | */ | |
7678 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
7679 | ||
7680 | /* | |
7681 | * Schedule the resilver or rebuild to restart in the future. | |
7682 | * We do this to ensure that dmu_sync-ed blocks have been | |
7683 | * stitched into the respective datasets. | |
7684 | */ | |
7685 | if (rebuild) { | |
7686 | newvd->vdev_rebuild_txg = txg; | |
7687 | ||
7688 | vdev_rebuild(tvd); | |
7689 | } else { | |
7690 | newvd->vdev_resilver_txg = txg; | |
7691 | ||
7692 | if (dsl_scan_resilvering(spa_get_dsl(spa)) && | |
7693 | spa_feature_is_enabled(spa, | |
7694 | SPA_FEATURE_RESILVER_DEFER)) { | |
7695 | vdev_defer_resilver(newvd); | |
7696 | } else { | |
7697 | dsl_scan_restart_resilver(spa->spa_dsl_pool, | |
7698 | dtl_max_txg); | |
7699 | } | |
7700 | } | |
7701 | } | |
7702 | ||
7703 | if (spa->spa_bootfs) | |
7704 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); | |
7705 | ||
7706 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); | |
7707 | ||
7708 | /* | |
7709 | * Commit the config | |
7710 | */ | |
7711 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
7712 | ||
7713 | spa_history_log_internal(spa, "vdev attach", NULL, | |
7714 | "%s vdev=%s %s vdev=%s", | |
7715 | replacing && newvd_isspare ? "spare in" : | |
7716 | replacing ? "replace" : "attach", newvdpath, | |
7717 | replacing ? "for" : "to", oldvdpath); | |
7718 | ||
7719 | spa_strfree(oldvdpath); | |
7720 | spa_strfree(newvdpath); | |
7721 | ||
7722 | return (0); | |
7723 | } | |
7724 | ||
7725 | /* | |
7726 | * Detach a device from a mirror or replacing vdev. | |
7727 | * | |
7728 | * If 'replace_done' is specified, only detach if the parent | |
7729 | * is a replacing or a spare vdev. | |
7730 | */ | |
7731 | int | |
7732 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) | |
7733 | { | |
7734 | uint64_t txg; | |
7735 | int error; | |
7736 | vdev_t *rvd __maybe_unused = spa->spa_root_vdev; | |
7737 | vdev_t *vd, *pvd, *cvd, *tvd; | |
7738 | boolean_t unspare = B_FALSE; | |
7739 | uint64_t unspare_guid = 0; | |
7740 | char *vdpath; | |
7741 | ||
7742 | ASSERT(spa_writeable(spa)); | |
7743 | ||
7744 | txg = spa_vdev_detach_enter(spa, guid); | |
7745 | ||
7746 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
7747 | ||
7748 | /* | |
7749 | * Besides being called directly from the userland through the | |
7750 | * ioctl interface, spa_vdev_detach() can be potentially called | |
7751 | * at the end of spa_vdev_resilver_done(). | |
7752 | * | |
7753 | * In the regular case, when we have a checkpoint this shouldn't | |
7754 | * happen as we never empty the DTLs of a vdev during the scrub | |
7755 | * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() | |
7756 | * should never get here when we have a checkpoint. | |
7757 | * | |
7758 | * That said, even in a case when we checkpoint the pool exactly | |
7759 | * as spa_vdev_resilver_done() calls this function everything | |
7760 | * should be fine as the resilver will return right away. | |
7761 | */ | |
7762 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7763 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
7764 | error = (spa_has_checkpoint(spa)) ? | |
7765 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
7766 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
7767 | } | |
7768 | ||
7769 | if (vd == NULL) | |
7770 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
7771 | ||
7772 | if (!vd->vdev_ops->vdev_op_leaf) | |
7773 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7774 | ||
7775 | pvd = vd->vdev_parent; | |
7776 | ||
7777 | /* | |
7778 | * If the parent/child relationship is not as expected, don't do it. | |
7779 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
7780 | * vdev that's replacing B with C. The user's intent in replacing | |
7781 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
7782 | * the replace by detaching C, the expected behavior is to end up | |
7783 | * M(A,B). But suppose that right after deciding to detach C, | |
7784 | * the replacement of B completes. We would have M(A,C), and then | |
7785 | * ask to detach C, which would leave us with just A -- not what | |
7786 | * the user wanted. To prevent this, we make sure that the | |
7787 | * parent/child relationship hasn't changed -- in this example, | |
7788 | * that C's parent is still the replacing vdev R. | |
7789 | */ | |
7790 | if (pvd->vdev_guid != pguid && pguid != 0) | |
7791 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
7792 | ||
7793 | /* | |
7794 | * Only 'replacing' or 'spare' vdevs can be replaced. | |
7795 | */ | |
7796 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && | |
7797 | pvd->vdev_ops != &vdev_spare_ops) | |
7798 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7799 | ||
7800 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
7801 | spa_version(spa) >= SPA_VERSION_SPARES); | |
7802 | ||
7803 | /* | |
7804 | * Only mirror, replacing, and spare vdevs support detach. | |
7805 | */ | |
7806 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
7807 | pvd->vdev_ops != &vdev_mirror_ops && | |
7808 | pvd->vdev_ops != &vdev_spare_ops) | |
7809 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
7810 | ||
7811 | /* | |
7812 | * If this device has the only valid copy of some data, | |
7813 | * we cannot safely detach it. | |
7814 | */ | |
7815 | if (vdev_dtl_required(vd)) | |
7816 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
7817 | ||
7818 | ASSERT(pvd->vdev_children >= 2); | |
7819 | ||
7820 | /* | |
7821 | * If we are detaching the second disk from a replacing vdev, then | |
7822 | * check to see if we changed the original vdev's path to have "/old" | |
7823 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
7824 | */ | |
7825 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && | |
7826 | vd->vdev_path != NULL) { | |
7827 | size_t len = strlen(vd->vdev_path); | |
7828 | ||
7829 | for (int c = 0; c < pvd->vdev_children; c++) { | |
7830 | cvd = pvd->vdev_child[c]; | |
7831 | ||
7832 | if (cvd == vd || cvd->vdev_path == NULL) | |
7833 | continue; | |
7834 | ||
7835 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
7836 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
7837 | spa_strfree(cvd->vdev_path); | |
7838 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
7839 | break; | |
7840 | } | |
7841 | } | |
7842 | } | |
7843 | ||
7844 | /* | |
7845 | * If we are detaching the original disk from a normal spare, then it | |
7846 | * implies that the spare should become a real disk, and be removed | |
7847 | * from the active spare list for the pool. dRAID spares on the | |
7848 | * other hand are coupled to the pool and thus should never be removed | |
7849 | * from the spares list. | |
7850 | */ | |
7851 | if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) { | |
7852 | vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
7853 | ||
7854 | if (last_cvd->vdev_isspare && | |
7855 | last_cvd->vdev_ops != &vdev_draid_spare_ops) { | |
7856 | unspare = B_TRUE; | |
7857 | } | |
7858 | } | |
7859 | ||
7860 | /* | |
7861 | * Erase the disk labels so the disk can be used for other things. | |
7862 | * This must be done after all other error cases are handled, | |
7863 | * but before we disembowel vd (so we can still do I/O to it). | |
7864 | * But if we can't do it, don't treat the error as fatal -- | |
7865 | * it may be that the unwritability of the disk is the reason | |
7866 | * it's being detached! | |
7867 | */ | |
7868 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
7869 | ||
7870 | /* | |
7871 | * Remove vd from its parent and compact the parent's children. | |
7872 | */ | |
7873 | vdev_remove_child(pvd, vd); | |
7874 | vdev_compact_children(pvd); | |
7875 | ||
7876 | /* | |
7877 | * Remember one of the remaining children so we can get tvd below. | |
7878 | */ | |
7879 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
7880 | ||
7881 | /* | |
7882 | * If we need to remove the remaining child from the list of hot spares, | |
7883 | * do it now, marking the vdev as no longer a spare in the process. | |
7884 | * We must do this before vdev_remove_parent(), because that can | |
7885 | * change the GUID if it creates a new toplevel GUID. For a similar | |
7886 | * reason, we must remove the spare now, in the same txg as the detach; | |
7887 | * otherwise someone could attach a new sibling, change the GUID, and | |
7888 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
7889 | */ | |
7890 | if (unspare) { | |
7891 | ASSERT(cvd->vdev_isspare); | |
7892 | spa_spare_remove(cvd); | |
7893 | unspare_guid = cvd->vdev_guid; | |
7894 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); | |
7895 | cvd->vdev_unspare = B_TRUE; | |
7896 | } | |
7897 | ||
7898 | /* | |
7899 | * If the parent mirror/replacing vdev only has one child, | |
7900 | * the parent is no longer needed. Remove it from the tree. | |
7901 | */ | |
7902 | if (pvd->vdev_children == 1) { | |
7903 | if (pvd->vdev_ops == &vdev_spare_ops) | |
7904 | cvd->vdev_unspare = B_FALSE; | |
7905 | vdev_remove_parent(cvd); | |
7906 | } | |
7907 | ||
7908 | /* | |
7909 | * We don't set tvd until now because the parent we just removed | |
7910 | * may have been the previous top-level vdev. | |
7911 | */ | |
7912 | tvd = cvd->vdev_top; | |
7913 | ASSERT(tvd->vdev_parent == rvd); | |
7914 | ||
7915 | /* | |
7916 | * Reevaluate the parent vdev state. | |
7917 | */ | |
7918 | vdev_propagate_state(cvd); | |
7919 | ||
7920 | /* | |
7921 | * If the 'autoexpand' property is set on the pool then automatically | |
7922 | * try to expand the size of the pool. For example if the device we | |
7923 | * just detached was smaller than the others, it may be possible to | |
7924 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
7925 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
7926 | */ | |
7927 | if (spa->spa_autoexpand) { | |
7928 | vdev_reopen(tvd); | |
7929 | vdev_expand(tvd, txg); | |
7930 | } | |
7931 | ||
7932 | vdev_config_dirty(tvd); | |
7933 | ||
7934 | /* | |
7935 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
7936 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
7937 | * But first make sure we're not on any *other* txg's DTL list, to | |
7938 | * prevent vd from being accessed after it's freed. | |
7939 | */ | |
7940 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); | |
7941 | for (int t = 0; t < TXG_SIZE; t++) | |
7942 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
7943 | vd->vdev_detached = B_TRUE; | |
7944 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
7945 | ||
7946 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); | |
7947 | spa_notify_waiters(spa); | |
7948 | ||
7949 | /* hang on to the spa before we release the lock */ | |
7950 | spa_open_ref(spa, FTAG); | |
7951 | ||
7952 | error = spa_vdev_exit(spa, vd, txg, 0); | |
7953 | ||
7954 | spa_history_log_internal(spa, "detach", NULL, | |
7955 | "vdev=%s", vdpath); | |
7956 | spa_strfree(vdpath); | |
7957 | ||
7958 | /* | |
7959 | * If this was the removal of the original device in a hot spare vdev, | |
7960 | * then we want to go through and remove the device from the hot spare | |
7961 | * list of every other pool. | |
7962 | */ | |
7963 | if (unspare) { | |
7964 | spa_t *altspa = NULL; | |
7965 | ||
7966 | mutex_enter(&spa_namespace_lock); | |
7967 | while ((altspa = spa_next(altspa)) != NULL) { | |
7968 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
7969 | altspa == spa) | |
7970 | continue; | |
7971 | ||
7972 | spa_open_ref(altspa, FTAG); | |
7973 | mutex_exit(&spa_namespace_lock); | |
7974 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); | |
7975 | mutex_enter(&spa_namespace_lock); | |
7976 | spa_close(altspa, FTAG); | |
7977 | } | |
7978 | mutex_exit(&spa_namespace_lock); | |
7979 | ||
7980 | /* search the rest of the vdevs for spares to remove */ | |
7981 | spa_vdev_resilver_done(spa); | |
7982 | } | |
7983 | ||
7984 | /* all done with the spa; OK to release */ | |
7985 | mutex_enter(&spa_namespace_lock); | |
7986 | spa_close(spa, FTAG); | |
7987 | mutex_exit(&spa_namespace_lock); | |
7988 | ||
7989 | return (error); | |
7990 | } | |
7991 | ||
7992 | static int | |
7993 | spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
7994 | list_t *vd_list) | |
7995 | { | |
7996 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
7997 | ||
7998 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
7999 | ||
8000 | /* Look up vdev and ensure it's a leaf. */ | |
8001 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
8002 | if (vd == NULL || vd->vdev_detached) { | |
8003 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8004 | return (SET_ERROR(ENODEV)); | |
8005 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
8006 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8007 | return (SET_ERROR(EINVAL)); | |
8008 | } else if (!vdev_writeable(vd)) { | |
8009 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8010 | return (SET_ERROR(EROFS)); | |
8011 | } | |
8012 | mutex_enter(&vd->vdev_initialize_lock); | |
8013 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8014 | ||
8015 | /* | |
8016 | * When we activate an initialize action we check to see | |
8017 | * if the vdev_initialize_thread is NULL. We do this instead | |
8018 | * of using the vdev_initialize_state since there might be | |
8019 | * a previous initialization process which has completed but | |
8020 | * the thread is not exited. | |
8021 | */ | |
8022 | if (cmd_type == POOL_INITIALIZE_START && | |
8023 | (vd->vdev_initialize_thread != NULL || | |
8024 | vd->vdev_top->vdev_removing || vd->vdev_top->vdev_rz_expanding)) { | |
8025 | mutex_exit(&vd->vdev_initialize_lock); | |
8026 | return (SET_ERROR(EBUSY)); | |
8027 | } else if (cmd_type == POOL_INITIALIZE_CANCEL && | |
8028 | (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE && | |
8029 | vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) { | |
8030 | mutex_exit(&vd->vdev_initialize_lock); | |
8031 | return (SET_ERROR(ESRCH)); | |
8032 | } else if (cmd_type == POOL_INITIALIZE_SUSPEND && | |
8033 | vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) { | |
8034 | mutex_exit(&vd->vdev_initialize_lock); | |
8035 | return (SET_ERROR(ESRCH)); | |
8036 | } else if (cmd_type == POOL_INITIALIZE_UNINIT && | |
8037 | vd->vdev_initialize_thread != NULL) { | |
8038 | mutex_exit(&vd->vdev_initialize_lock); | |
8039 | return (SET_ERROR(EBUSY)); | |
8040 | } | |
8041 | ||
8042 | switch (cmd_type) { | |
8043 | case POOL_INITIALIZE_START: | |
8044 | vdev_initialize(vd); | |
8045 | break; | |
8046 | case POOL_INITIALIZE_CANCEL: | |
8047 | vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list); | |
8048 | break; | |
8049 | case POOL_INITIALIZE_SUSPEND: | |
8050 | vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list); | |
8051 | break; | |
8052 | case POOL_INITIALIZE_UNINIT: | |
8053 | vdev_uninitialize(vd); | |
8054 | break; | |
8055 | default: | |
8056 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
8057 | } | |
8058 | mutex_exit(&vd->vdev_initialize_lock); | |
8059 | ||
8060 | return (0); | |
8061 | } | |
8062 | ||
8063 | int | |
8064 | spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, | |
8065 | nvlist_t *vdev_errlist) | |
8066 | { | |
8067 | int total_errors = 0; | |
8068 | list_t vd_list; | |
8069 | ||
8070 | list_create(&vd_list, sizeof (vdev_t), | |
8071 | offsetof(vdev_t, vdev_initialize_node)); | |
8072 | ||
8073 | /* | |
8074 | * We hold the namespace lock through the whole function | |
8075 | * to prevent any changes to the pool while we're starting or | |
8076 | * stopping initialization. The config and state locks are held so that | |
8077 | * we can properly assess the vdev state before we commit to | |
8078 | * the initializing operation. | |
8079 | */ | |
8080 | mutex_enter(&spa_namespace_lock); | |
8081 | ||
8082 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
8083 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
8084 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
8085 | ||
8086 | int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type, | |
8087 | &vd_list); | |
8088 | if (error != 0) { | |
8089 | char guid_as_str[MAXNAMELEN]; | |
8090 | ||
8091 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
8092 | "%llu", (unsigned long long)vdev_guid); | |
8093 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
8094 | total_errors++; | |
8095 | } | |
8096 | } | |
8097 | ||
8098 | /* Wait for all initialize threads to stop. */ | |
8099 | vdev_initialize_stop_wait(spa, &vd_list); | |
8100 | ||
8101 | /* Sync out the initializing state */ | |
8102 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
8103 | mutex_exit(&spa_namespace_lock); | |
8104 | ||
8105 | list_destroy(&vd_list); | |
8106 | ||
8107 | return (total_errors); | |
8108 | } | |
8109 | ||
8110 | static int | |
8111 | spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type, | |
8112 | uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list) | |
8113 | { | |
8114 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
8115 | ||
8116 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
8117 | ||
8118 | /* Look up vdev and ensure it's a leaf. */ | |
8119 | vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
8120 | if (vd == NULL || vd->vdev_detached) { | |
8121 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8122 | return (SET_ERROR(ENODEV)); | |
8123 | } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { | |
8124 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8125 | return (SET_ERROR(EINVAL)); | |
8126 | } else if (!vdev_writeable(vd)) { | |
8127 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8128 | return (SET_ERROR(EROFS)); | |
8129 | } else if (!vd->vdev_has_trim) { | |
8130 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8131 | return (SET_ERROR(EOPNOTSUPP)); | |
8132 | } else if (secure && !vd->vdev_has_securetrim) { | |
8133 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8134 | return (SET_ERROR(EOPNOTSUPP)); | |
8135 | } | |
8136 | mutex_enter(&vd->vdev_trim_lock); | |
8137 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
8138 | ||
8139 | /* | |
8140 | * When we activate a TRIM action we check to see if the | |
8141 | * vdev_trim_thread is NULL. We do this instead of using the | |
8142 | * vdev_trim_state since there might be a previous TRIM process | |
8143 | * which has completed but the thread is not exited. | |
8144 | */ | |
8145 | if (cmd_type == POOL_TRIM_START && | |
8146 | (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing || | |
8147 | vd->vdev_top->vdev_rz_expanding)) { | |
8148 | mutex_exit(&vd->vdev_trim_lock); | |
8149 | return (SET_ERROR(EBUSY)); | |
8150 | } else if (cmd_type == POOL_TRIM_CANCEL && | |
8151 | (vd->vdev_trim_state != VDEV_TRIM_ACTIVE && | |
8152 | vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) { | |
8153 | mutex_exit(&vd->vdev_trim_lock); | |
8154 | return (SET_ERROR(ESRCH)); | |
8155 | } else if (cmd_type == POOL_TRIM_SUSPEND && | |
8156 | vd->vdev_trim_state != VDEV_TRIM_ACTIVE) { | |
8157 | mutex_exit(&vd->vdev_trim_lock); | |
8158 | return (SET_ERROR(ESRCH)); | |
8159 | } | |
8160 | ||
8161 | switch (cmd_type) { | |
8162 | case POOL_TRIM_START: | |
8163 | vdev_trim(vd, rate, partial, secure); | |
8164 | break; | |
8165 | case POOL_TRIM_CANCEL: | |
8166 | vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list); | |
8167 | break; | |
8168 | case POOL_TRIM_SUSPEND: | |
8169 | vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list); | |
8170 | break; | |
8171 | default: | |
8172 | panic("invalid cmd_type %llu", (unsigned long long)cmd_type); | |
8173 | } | |
8174 | mutex_exit(&vd->vdev_trim_lock); | |
8175 | ||
8176 | return (0); | |
8177 | } | |
8178 | ||
8179 | /* | |
8180 | * Initiates a manual TRIM for the requested vdevs. This kicks off individual | |
8181 | * TRIM threads for each child vdev. These threads pass over all of the free | |
8182 | * space in the vdev's metaslabs and issues TRIM commands for that space. | |
8183 | */ | |
8184 | int | |
8185 | spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate, | |
8186 | boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist) | |
8187 | { | |
8188 | int total_errors = 0; | |
8189 | list_t vd_list; | |
8190 | ||
8191 | list_create(&vd_list, sizeof (vdev_t), | |
8192 | offsetof(vdev_t, vdev_trim_node)); | |
8193 | ||
8194 | /* | |
8195 | * We hold the namespace lock through the whole function | |
8196 | * to prevent any changes to the pool while we're starting or | |
8197 | * stopping TRIM. The config and state locks are held so that | |
8198 | * we can properly assess the vdev state before we commit to | |
8199 | * the TRIM operation. | |
8200 | */ | |
8201 | mutex_enter(&spa_namespace_lock); | |
8202 | ||
8203 | for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL); | |
8204 | pair != NULL; pair = nvlist_next_nvpair(nv, pair)) { | |
8205 | uint64_t vdev_guid = fnvpair_value_uint64(pair); | |
8206 | ||
8207 | int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type, | |
8208 | rate, partial, secure, &vd_list); | |
8209 | if (error != 0) { | |
8210 | char guid_as_str[MAXNAMELEN]; | |
8211 | ||
8212 | (void) snprintf(guid_as_str, sizeof (guid_as_str), | |
8213 | "%llu", (unsigned long long)vdev_guid); | |
8214 | fnvlist_add_int64(vdev_errlist, guid_as_str, error); | |
8215 | total_errors++; | |
8216 | } | |
8217 | } | |
8218 | ||
8219 | /* Wait for all TRIM threads to stop. */ | |
8220 | vdev_trim_stop_wait(spa, &vd_list); | |
8221 | ||
8222 | /* Sync out the TRIM state */ | |
8223 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
8224 | mutex_exit(&spa_namespace_lock); | |
8225 | ||
8226 | list_destroy(&vd_list); | |
8227 | ||
8228 | return (total_errors); | |
8229 | } | |
8230 | ||
8231 | /* | |
8232 | * Split a set of devices from their mirrors, and create a new pool from them. | |
8233 | */ | |
8234 | int | |
8235 | spa_vdev_split_mirror(spa_t *spa, const char *newname, nvlist_t *config, | |
8236 | nvlist_t *props, boolean_t exp) | |
8237 | { | |
8238 | int error = 0; | |
8239 | uint64_t txg, *glist; | |
8240 | spa_t *newspa; | |
8241 | uint_t c, children, lastlog; | |
8242 | nvlist_t **child, *nvl, *tmp; | |
8243 | dmu_tx_t *tx; | |
8244 | const char *altroot = NULL; | |
8245 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ | |
8246 | boolean_t activate_slog; | |
8247 | ||
8248 | ASSERT(spa_writeable(spa)); | |
8249 | ||
8250 | txg = spa_vdev_enter(spa); | |
8251 | ||
8252 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
8253 | if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { | |
8254 | error = (spa_has_checkpoint(spa)) ? | |
8255 | ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; | |
8256 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8257 | } | |
8258 | ||
8259 | /* clear the log and flush everything up to now */ | |
8260 | activate_slog = spa_passivate_log(spa); | |
8261 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
8262 | error = spa_reset_logs(spa); | |
8263 | txg = spa_vdev_config_enter(spa); | |
8264 | ||
8265 | if (activate_slog) | |
8266 | spa_activate_log(spa); | |
8267 | ||
8268 | if (error != 0) | |
8269 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8270 | ||
8271 | /* check new spa name before going any further */ | |
8272 | if (spa_lookup(newname) != NULL) | |
8273 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
8274 | ||
8275 | /* | |
8276 | * scan through all the children to ensure they're all mirrors | |
8277 | */ | |
8278 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
8279 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
8280 | &children) != 0) | |
8281 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8282 | ||
8283 | /* first, check to ensure we've got the right child count */ | |
8284 | rvd = spa->spa_root_vdev; | |
8285 | lastlog = 0; | |
8286 | for (c = 0; c < rvd->vdev_children; c++) { | |
8287 | vdev_t *vd = rvd->vdev_child[c]; | |
8288 | ||
8289 | /* don't count the holes & logs as children */ | |
8290 | if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops && | |
8291 | !vdev_is_concrete(vd))) { | |
8292 | if (lastlog == 0) | |
8293 | lastlog = c; | |
8294 | continue; | |
8295 | } | |
8296 | ||
8297 | lastlog = 0; | |
8298 | } | |
8299 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
8300 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8301 | ||
8302 | /* next, ensure no spare or cache devices are part of the split */ | |
8303 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
8304 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
8305 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
8306 | ||
8307 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); | |
8308 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
8309 | ||
8310 | /* then, loop over each vdev and validate it */ | |
8311 | for (c = 0; c < children; c++) { | |
8312 | uint64_t is_hole = 0; | |
8313 | ||
8314 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
8315 | &is_hole); | |
8316 | ||
8317 | if (is_hole != 0) { | |
8318 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
8319 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
8320 | continue; | |
8321 | } else { | |
8322 | error = SET_ERROR(EINVAL); | |
8323 | break; | |
8324 | } | |
8325 | } | |
8326 | ||
8327 | /* deal with indirect vdevs */ | |
8328 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ops == | |
8329 | &vdev_indirect_ops) | |
8330 | continue; | |
8331 | ||
8332 | /* which disk is going to be split? */ | |
8333 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
8334 | &glist[c]) != 0) { | |
8335 | error = SET_ERROR(EINVAL); | |
8336 | break; | |
8337 | } | |
8338 | ||
8339 | /* look it up in the spa */ | |
8340 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
8341 | if (vml[c] == NULL) { | |
8342 | error = SET_ERROR(ENODEV); | |
8343 | break; | |
8344 | } | |
8345 | ||
8346 | /* make sure there's nothing stopping the split */ | |
8347 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
8348 | vml[c]->vdev_islog || | |
8349 | !vdev_is_concrete(vml[c]) || | |
8350 | vml[c]->vdev_isspare || | |
8351 | vml[c]->vdev_isl2cache || | |
8352 | !vdev_writeable(vml[c]) || | |
8353 | vml[c]->vdev_children != 0 || | |
8354 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
8355 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
8356 | error = SET_ERROR(EINVAL); | |
8357 | break; | |
8358 | } | |
8359 | ||
8360 | if (vdev_dtl_required(vml[c]) || | |
8361 | vdev_resilver_needed(vml[c], NULL, NULL)) { | |
8362 | error = SET_ERROR(EBUSY); | |
8363 | break; | |
8364 | } | |
8365 | ||
8366 | /* we need certain info from the top level */ | |
8367 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, | |
8368 | vml[c]->vdev_top->vdev_ms_array); | |
8369 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
8370 | vml[c]->vdev_top->vdev_ms_shift); | |
8371 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
8372 | vml[c]->vdev_top->vdev_asize); | |
8373 | fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
8374 | vml[c]->vdev_top->vdev_ashift); | |
8375 | ||
8376 | /* transfer per-vdev ZAPs */ | |
8377 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
8378 | VERIFY0(nvlist_add_uint64(child[c], | |
8379 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
8380 | ||
8381 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
8382 | VERIFY0(nvlist_add_uint64(child[c], | |
8383 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
8384 | vml[c]->vdev_parent->vdev_top_zap)); | |
8385 | } | |
8386 | ||
8387 | if (error != 0) { | |
8388 | kmem_free(vml, children * sizeof (vdev_t *)); | |
8389 | kmem_free(glist, children * sizeof (uint64_t)); | |
8390 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
8391 | } | |
8392 | ||
8393 | /* stop writers from using the disks */ | |
8394 | for (c = 0; c < children; c++) { | |
8395 | if (vml[c] != NULL) | |
8396 | vml[c]->vdev_offline = B_TRUE; | |
8397 | } | |
8398 | vdev_reopen(spa->spa_root_vdev); | |
8399 | ||
8400 | /* | |
8401 | * Temporarily record the splitting vdevs in the spa config. This | |
8402 | * will disappear once the config is regenerated. | |
8403 | */ | |
8404 | nvl = fnvlist_alloc(); | |
8405 | fnvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children); | |
8406 | kmem_free(glist, children * sizeof (uint64_t)); | |
8407 | ||
8408 | mutex_enter(&spa->spa_props_lock); | |
8409 | fnvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl); | |
8410 | mutex_exit(&spa->spa_props_lock); | |
8411 | spa->spa_config_splitting = nvl; | |
8412 | vdev_config_dirty(spa->spa_root_vdev); | |
8413 | ||
8414 | /* configure and create the new pool */ | |
8415 | fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname); | |
8416 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
8417 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE); | |
8418 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); | |
8419 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg); | |
8420 | fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
8421 | spa_generate_guid(NULL)); | |
8422 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
8423 | (void) nvlist_lookup_string(props, | |
8424 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
8425 | ||
8426 | /* add the new pool to the namespace */ | |
8427 | newspa = spa_add(newname, config, altroot); | |
8428 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; | |
8429 | newspa->spa_config_txg = spa->spa_config_txg; | |
8430 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
8431 | ||
8432 | /* release the spa config lock, retaining the namespace lock */ | |
8433 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
8434 | ||
8435 | if (zio_injection_enabled) | |
8436 | zio_handle_panic_injection(spa, FTAG, 1); | |
8437 | ||
8438 | spa_activate(newspa, spa_mode_global); | |
8439 | spa_async_suspend(newspa); | |
8440 | ||
8441 | /* | |
8442 | * Temporarily stop the initializing and TRIM activity. We set the | |
8443 | * state to ACTIVE so that we know to resume initializing or TRIM | |
8444 | * once the split has completed. | |
8445 | */ | |
8446 | list_t vd_initialize_list; | |
8447 | list_create(&vd_initialize_list, sizeof (vdev_t), | |
8448 | offsetof(vdev_t, vdev_initialize_node)); | |
8449 | ||
8450 | list_t vd_trim_list; | |
8451 | list_create(&vd_trim_list, sizeof (vdev_t), | |
8452 | offsetof(vdev_t, vdev_trim_node)); | |
8453 | ||
8454 | for (c = 0; c < children; c++) { | |
8455 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { | |
8456 | mutex_enter(&vml[c]->vdev_initialize_lock); | |
8457 | vdev_initialize_stop(vml[c], | |
8458 | VDEV_INITIALIZE_ACTIVE, &vd_initialize_list); | |
8459 | mutex_exit(&vml[c]->vdev_initialize_lock); | |
8460 | ||
8461 | mutex_enter(&vml[c]->vdev_trim_lock); | |
8462 | vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list); | |
8463 | mutex_exit(&vml[c]->vdev_trim_lock); | |
8464 | } | |
8465 | } | |
8466 | ||
8467 | vdev_initialize_stop_wait(spa, &vd_initialize_list); | |
8468 | vdev_trim_stop_wait(spa, &vd_trim_list); | |
8469 | ||
8470 | list_destroy(&vd_initialize_list); | |
8471 | list_destroy(&vd_trim_list); | |
8472 | ||
8473 | newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; | |
8474 | newspa->spa_is_splitting = B_TRUE; | |
8475 | ||
8476 | /* create the new pool from the disks of the original pool */ | |
8477 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); | |
8478 | if (error) | |
8479 | goto out; | |
8480 | ||
8481 | /* if that worked, generate a real config for the new pool */ | |
8482 | if (newspa->spa_root_vdev != NULL) { | |
8483 | newspa->spa_config_splitting = fnvlist_alloc(); | |
8484 | fnvlist_add_uint64(newspa->spa_config_splitting, | |
8485 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)); | |
8486 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, | |
8487 | B_TRUE)); | |
8488 | } | |
8489 | ||
8490 | /* set the props */ | |
8491 | if (props != NULL) { | |
8492 | spa_configfile_set(newspa, props, B_FALSE); | |
8493 | error = spa_prop_set(newspa, props); | |
8494 | if (error) | |
8495 | goto out; | |
8496 | } | |
8497 | ||
8498 | /* flush everything */ | |
8499 | txg = spa_vdev_config_enter(newspa); | |
8500 | vdev_config_dirty(newspa->spa_root_vdev); | |
8501 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
8502 | ||
8503 | if (zio_injection_enabled) | |
8504 | zio_handle_panic_injection(spa, FTAG, 2); | |
8505 | ||
8506 | spa_async_resume(newspa); | |
8507 | ||
8508 | /* finally, update the original pool's config */ | |
8509 | txg = spa_vdev_config_enter(spa); | |
8510 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
8511 | error = dmu_tx_assign(tx, TXG_WAIT); | |
8512 | if (error != 0) | |
8513 | dmu_tx_abort(tx); | |
8514 | for (c = 0; c < children; c++) { | |
8515 | if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) { | |
8516 | vdev_t *tvd = vml[c]->vdev_top; | |
8517 | ||
8518 | /* | |
8519 | * Need to be sure the detachable VDEV is not | |
8520 | * on any *other* txg's DTL list to prevent it | |
8521 | * from being accessed after it's freed. | |
8522 | */ | |
8523 | for (int t = 0; t < TXG_SIZE; t++) { | |
8524 | (void) txg_list_remove_this( | |
8525 | &tvd->vdev_dtl_list, vml[c], t); | |
8526 | } | |
8527 | ||
8528 | vdev_split(vml[c]); | |
8529 | if (error == 0) | |
8530 | spa_history_log_internal(spa, "detach", tx, | |
8531 | "vdev=%s", vml[c]->vdev_path); | |
8532 | ||
8533 | vdev_free(vml[c]); | |
8534 | } | |
8535 | } | |
8536 | spa->spa_avz_action = AVZ_ACTION_REBUILD; | |
8537 | vdev_config_dirty(spa->spa_root_vdev); | |
8538 | spa->spa_config_splitting = NULL; | |
8539 | nvlist_free(nvl); | |
8540 | if (error == 0) | |
8541 | dmu_tx_commit(tx); | |
8542 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
8543 | ||
8544 | if (zio_injection_enabled) | |
8545 | zio_handle_panic_injection(spa, FTAG, 3); | |
8546 | ||
8547 | /* split is complete; log a history record */ | |
8548 | spa_history_log_internal(newspa, "split", NULL, | |
8549 | "from pool %s", spa_name(spa)); | |
8550 | ||
8551 | newspa->spa_is_splitting = B_FALSE; | |
8552 | kmem_free(vml, children * sizeof (vdev_t *)); | |
8553 | ||
8554 | /* if we're not going to mount the filesystems in userland, export */ | |
8555 | if (exp) | |
8556 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
8557 | B_FALSE, B_FALSE); | |
8558 | ||
8559 | return (error); | |
8560 | ||
8561 | out: | |
8562 | spa_unload(newspa); | |
8563 | spa_deactivate(newspa); | |
8564 | spa_remove(newspa); | |
8565 | ||
8566 | txg = spa_vdev_config_enter(spa); | |
8567 | ||
8568 | /* re-online all offlined disks */ | |
8569 | for (c = 0; c < children; c++) { | |
8570 | if (vml[c] != NULL) | |
8571 | vml[c]->vdev_offline = B_FALSE; | |
8572 | } | |
8573 | ||
8574 | /* restart initializing or trimming disks as necessary */ | |
8575 | spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART); | |
8576 | spa_async_request(spa, SPA_ASYNC_TRIM_RESTART); | |
8577 | spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART); | |
8578 | ||
8579 | vdev_reopen(spa->spa_root_vdev); | |
8580 | ||
8581 | nvlist_free(spa->spa_config_splitting); | |
8582 | spa->spa_config_splitting = NULL; | |
8583 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
8584 | ||
8585 | kmem_free(vml, children * sizeof (vdev_t *)); | |
8586 | return (error); | |
8587 | } | |
8588 | ||
8589 | /* | |
8590 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
8591 | * currently spared, so we can detach it. | |
8592 | */ | |
8593 | static vdev_t * | |
8594 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
8595 | { | |
8596 | vdev_t *newvd, *oldvd; | |
8597 | ||
8598 | for (int c = 0; c < vd->vdev_children; c++) { | |
8599 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); | |
8600 | if (oldvd != NULL) | |
8601 | return (oldvd); | |
8602 | } | |
8603 | ||
8604 | /* | |
8605 | * Check for a completed replacement. We always consider the first | |
8606 | * vdev in the list to be the oldest vdev, and the last one to be | |
8607 | * the newest (see spa_vdev_attach() for how that works). In | |
8608 | * the case where the newest vdev is faulted, we will not automatically | |
8609 | * remove it after a resilver completes. This is OK as it will require | |
8610 | * user intervention to determine which disk the admin wishes to keep. | |
8611 | */ | |
8612 | if (vd->vdev_ops == &vdev_replacing_ops) { | |
8613 | ASSERT(vd->vdev_children > 1); | |
8614 | ||
8615 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
8616 | oldvd = vd->vdev_child[0]; | |
8617 | ||
8618 | if (vdev_dtl_empty(newvd, DTL_MISSING) && | |
8619 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
8620 | !vdev_dtl_required(oldvd)) | |
8621 | return (oldvd); | |
8622 | } | |
8623 | ||
8624 | /* | |
8625 | * Check for a completed resilver with the 'unspare' flag set. | |
8626 | * Also potentially update faulted state. | |
8627 | */ | |
8628 | if (vd->vdev_ops == &vdev_spare_ops) { | |
8629 | vdev_t *first = vd->vdev_child[0]; | |
8630 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
8631 | ||
8632 | if (last->vdev_unspare) { | |
8633 | oldvd = first; | |
8634 | newvd = last; | |
8635 | } else if (first->vdev_unspare) { | |
8636 | oldvd = last; | |
8637 | newvd = first; | |
8638 | } else { | |
8639 | oldvd = NULL; | |
8640 | } | |
8641 | ||
8642 | if (oldvd != NULL && | |
8643 | vdev_dtl_empty(newvd, DTL_MISSING) && | |
8644 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
8645 | !vdev_dtl_required(oldvd)) | |
8646 | return (oldvd); | |
8647 | ||
8648 | vdev_propagate_state(vd); | |
8649 | ||
8650 | /* | |
8651 | * If there are more than two spares attached to a disk, | |
8652 | * and those spares are not required, then we want to | |
8653 | * attempt to free them up now so that they can be used | |
8654 | * by other pools. Once we're back down to a single | |
8655 | * disk+spare, we stop removing them. | |
8656 | */ | |
8657 | if (vd->vdev_children > 2) { | |
8658 | newvd = vd->vdev_child[1]; | |
8659 | ||
8660 | if (newvd->vdev_isspare && last->vdev_isspare && | |
8661 | vdev_dtl_empty(last, DTL_MISSING) && | |
8662 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
8663 | !vdev_dtl_required(newvd)) | |
8664 | return (newvd); | |
8665 | } | |
8666 | } | |
8667 | ||
8668 | return (NULL); | |
8669 | } | |
8670 | ||
8671 | static void | |
8672 | spa_vdev_resilver_done(spa_t *spa) | |
8673 | { | |
8674 | vdev_t *vd, *pvd, *ppvd; | |
8675 | uint64_t guid, sguid, pguid, ppguid; | |
8676 | ||
8677 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
8678 | ||
8679 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
8680 | pvd = vd->vdev_parent; | |
8681 | ppvd = pvd->vdev_parent; | |
8682 | guid = vd->vdev_guid; | |
8683 | pguid = pvd->vdev_guid; | |
8684 | ppguid = ppvd->vdev_guid; | |
8685 | sguid = 0; | |
8686 | /* | |
8687 | * If we have just finished replacing a hot spared device, then | |
8688 | * we need to detach the parent's first child (the original hot | |
8689 | * spare) as well. | |
8690 | */ | |
8691 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && | |
8692 | ppvd->vdev_children == 2) { | |
8693 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); | |
8694 | sguid = ppvd->vdev_child[1]->vdev_guid; | |
8695 | } | |
8696 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); | |
8697 | ||
8698 | spa_config_exit(spa, SCL_ALL, FTAG); | |
8699 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
8700 | return; | |
8701 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) | |
8702 | return; | |
8703 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
8704 | } | |
8705 | ||
8706 | spa_config_exit(spa, SCL_ALL, FTAG); | |
8707 | ||
8708 | /* | |
8709 | * If a detach was not performed above replace waiters will not have | |
8710 | * been notified. In which case we must do so now. | |
8711 | */ | |
8712 | spa_notify_waiters(spa); | |
8713 | } | |
8714 | ||
8715 | /* | |
8716 | * Update the stored path or FRU for this vdev. | |
8717 | */ | |
8718 | static int | |
8719 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, | |
8720 | boolean_t ispath) | |
8721 | { | |
8722 | vdev_t *vd; | |
8723 | boolean_t sync = B_FALSE; | |
8724 | ||
8725 | ASSERT(spa_writeable(spa)); | |
8726 | ||
8727 | spa_vdev_state_enter(spa, SCL_ALL); | |
8728 | ||
8729 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
8730 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); | |
8731 | ||
8732 | if (!vd->vdev_ops->vdev_op_leaf) | |
8733 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
8734 | ||
8735 | if (ispath) { | |
8736 | if (strcmp(value, vd->vdev_path) != 0) { | |
8737 | spa_strfree(vd->vdev_path); | |
8738 | vd->vdev_path = spa_strdup(value); | |
8739 | sync = B_TRUE; | |
8740 | } | |
8741 | } else { | |
8742 | if (vd->vdev_fru == NULL) { | |
8743 | vd->vdev_fru = spa_strdup(value); | |
8744 | sync = B_TRUE; | |
8745 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
8746 | spa_strfree(vd->vdev_fru); | |
8747 | vd->vdev_fru = spa_strdup(value); | |
8748 | sync = B_TRUE; | |
8749 | } | |
8750 | } | |
8751 | ||
8752 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); | |
8753 | } | |
8754 | ||
8755 | int | |
8756 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
8757 | { | |
8758 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
8759 | } | |
8760 | ||
8761 | int | |
8762 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
8763 | { | |
8764 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
8765 | } | |
8766 | ||
8767 | /* | |
8768 | * ========================================================================== | |
8769 | * SPA Scanning | |
8770 | * ========================================================================== | |
8771 | */ | |
8772 | int | |
8773 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
8774 | { | |
8775 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8776 | ||
8777 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
8778 | return (SET_ERROR(EBUSY)); | |
8779 | ||
8780 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
8781 | } | |
8782 | ||
8783 | int | |
8784 | spa_scan_stop(spa_t *spa) | |
8785 | { | |
8786 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8787 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
8788 | return (SET_ERROR(EBUSY)); | |
8789 | ||
8790 | return (dsl_scan_cancel(spa->spa_dsl_pool)); | |
8791 | } | |
8792 | ||
8793 | int | |
8794 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
8795 | { | |
8796 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
8797 | ||
8798 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) | |
8799 | return (SET_ERROR(ENOTSUP)); | |
8800 | ||
8801 | if (func == POOL_SCAN_RESILVER && | |
8802 | !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) | |
8803 | return (SET_ERROR(ENOTSUP)); | |
8804 | ||
8805 | /* | |
8806 | * If a resilver was requested, but there is no DTL on a | |
8807 | * writeable leaf device, we have nothing to do. | |
8808 | */ | |
8809 | if (func == POOL_SCAN_RESILVER && | |
8810 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
8811 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
8812 | return (0); | |
8813 | } | |
8814 | ||
8815 | if (func == POOL_SCAN_ERRORSCRUB && | |
8816 | !spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) | |
8817 | return (SET_ERROR(ENOTSUP)); | |
8818 | ||
8819 | return (dsl_scan(spa->spa_dsl_pool, func)); | |
8820 | } | |
8821 | ||
8822 | /* | |
8823 | * ========================================================================== | |
8824 | * SPA async task processing | |
8825 | * ========================================================================== | |
8826 | */ | |
8827 | ||
8828 | static void | |
8829 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
8830 | { | |
8831 | if (vd->vdev_remove_wanted) { | |
8832 | vd->vdev_remove_wanted = B_FALSE; | |
8833 | vd->vdev_delayed_close = B_FALSE; | |
8834 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); | |
8835 | ||
8836 | /* | |
8837 | * We want to clear the stats, but we don't want to do a full | |
8838 | * vdev_clear() as that will cause us to throw away | |
8839 | * degraded/faulted state as well as attempt to reopen the | |
8840 | * device, all of which is a waste. | |
8841 | */ | |
8842 | vd->vdev_stat.vs_read_errors = 0; | |
8843 | vd->vdev_stat.vs_write_errors = 0; | |
8844 | vd->vdev_stat.vs_checksum_errors = 0; | |
8845 | ||
8846 | vdev_state_dirty(vd->vdev_top); | |
8847 | ||
8848 | /* Tell userspace that the vdev is gone. */ | |
8849 | zfs_post_remove(spa, vd); | |
8850 | } | |
8851 | ||
8852 | for (int c = 0; c < vd->vdev_children; c++) | |
8853 | spa_async_remove(spa, vd->vdev_child[c]); | |
8854 | } | |
8855 | ||
8856 | static void | |
8857 | spa_async_fault_vdev(spa_t *spa, vdev_t *vd) | |
8858 | { | |
8859 | if (vd->vdev_fault_wanted) { | |
8860 | vd->vdev_fault_wanted = B_FALSE; | |
8861 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
8862 | VDEV_AUX_ERR_EXCEEDED); | |
8863 | } | |
8864 | ||
8865 | for (int c = 0; c < vd->vdev_children; c++) | |
8866 | spa_async_fault_vdev(spa, vd->vdev_child[c]); | |
8867 | } | |
8868 | ||
8869 | static void | |
8870 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
8871 | { | |
8872 | if (!spa->spa_autoexpand) | |
8873 | return; | |
8874 | ||
8875 | for (int c = 0; c < vd->vdev_children; c++) { | |
8876 | vdev_t *cvd = vd->vdev_child[c]; | |
8877 | spa_async_autoexpand(spa, cvd); | |
8878 | } | |
8879 | ||
8880 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
8881 | return; | |
8882 | ||
8883 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); | |
8884 | } | |
8885 | ||
8886 | static __attribute__((noreturn)) void | |
8887 | spa_async_thread(void *arg) | |
8888 | { | |
8889 | spa_t *spa = (spa_t *)arg; | |
8890 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
8891 | int tasks; | |
8892 | ||
8893 | ASSERT(spa->spa_sync_on); | |
8894 | ||
8895 | mutex_enter(&spa->spa_async_lock); | |
8896 | tasks = spa->spa_async_tasks; | |
8897 | spa->spa_async_tasks = 0; | |
8898 | mutex_exit(&spa->spa_async_lock); | |
8899 | ||
8900 | /* | |
8901 | * See if the config needs to be updated. | |
8902 | */ | |
8903 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
8904 | uint64_t old_space, new_space; | |
8905 | ||
8906 | mutex_enter(&spa_namespace_lock); | |
8907 | old_space = metaslab_class_get_space(spa_normal_class(spa)); | |
8908 | old_space += metaslab_class_get_space(spa_special_class(spa)); | |
8909 | old_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
8910 | old_space += metaslab_class_get_space( | |
8911 | spa_embedded_log_class(spa)); | |
8912 | ||
8913 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
8914 | ||
8915 | new_space = metaslab_class_get_space(spa_normal_class(spa)); | |
8916 | new_space += metaslab_class_get_space(spa_special_class(spa)); | |
8917 | new_space += metaslab_class_get_space(spa_dedup_class(spa)); | |
8918 | new_space += metaslab_class_get_space( | |
8919 | spa_embedded_log_class(spa)); | |
8920 | mutex_exit(&spa_namespace_lock); | |
8921 | ||
8922 | /* | |
8923 | * If the pool grew as a result of the config update, | |
8924 | * then log an internal history event. | |
8925 | */ | |
8926 | if (new_space != old_space) { | |
8927 | spa_history_log_internal(spa, "vdev online", NULL, | |
8928 | "pool '%s' size: %llu(+%llu)", | |
8929 | spa_name(spa), (u_longlong_t)new_space, | |
8930 | (u_longlong_t)(new_space - old_space)); | |
8931 | } | |
8932 | } | |
8933 | ||
8934 | /* | |
8935 | * See if any devices need to be marked REMOVED. | |
8936 | */ | |
8937 | if (tasks & SPA_ASYNC_REMOVE) { | |
8938 | spa_vdev_state_enter(spa, SCL_NONE); | |
8939 | spa_async_remove(spa, spa->spa_root_vdev); | |
8940 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
8941 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); | |
8942 | for (int i = 0; i < spa->spa_spares.sav_count; i++) | |
8943 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); | |
8944 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
8945 | } | |
8946 | ||
8947 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { | |
8948 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8949 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
8950 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8951 | } | |
8952 | ||
8953 | /* | |
8954 | * See if any devices need to be marked faulted. | |
8955 | */ | |
8956 | if (tasks & SPA_ASYNC_FAULT_VDEV) { | |
8957 | spa_vdev_state_enter(spa, SCL_NONE); | |
8958 | spa_async_fault_vdev(spa, spa->spa_root_vdev); | |
8959 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
8960 | } | |
8961 | ||
8962 | /* | |
8963 | * If any devices are done replacing, detach them. | |
8964 | */ | |
8965 | if (tasks & SPA_ASYNC_RESILVER_DONE || | |
8966 | tasks & SPA_ASYNC_REBUILD_DONE || | |
8967 | tasks & SPA_ASYNC_DETACH_SPARE) { | |
8968 | spa_vdev_resilver_done(spa); | |
8969 | } | |
8970 | ||
8971 | /* | |
8972 | * Kick off a resilver. | |
8973 | */ | |
8974 | if (tasks & SPA_ASYNC_RESILVER && | |
8975 | !vdev_rebuild_active(spa->spa_root_vdev) && | |
8976 | (!dsl_scan_resilvering(dp) || | |
8977 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER))) | |
8978 | dsl_scan_restart_resilver(dp, 0); | |
8979 | ||
8980 | if (tasks & SPA_ASYNC_INITIALIZE_RESTART) { | |
8981 | mutex_enter(&spa_namespace_lock); | |
8982 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8983 | vdev_initialize_restart(spa->spa_root_vdev); | |
8984 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8985 | mutex_exit(&spa_namespace_lock); | |
8986 | } | |
8987 | ||
8988 | if (tasks & SPA_ASYNC_TRIM_RESTART) { | |
8989 | mutex_enter(&spa_namespace_lock); | |
8990 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8991 | vdev_trim_restart(spa->spa_root_vdev); | |
8992 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
8993 | mutex_exit(&spa_namespace_lock); | |
8994 | } | |
8995 | ||
8996 | if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) { | |
8997 | mutex_enter(&spa_namespace_lock); | |
8998 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
8999 | vdev_autotrim_restart(spa); | |
9000 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
9001 | mutex_exit(&spa_namespace_lock); | |
9002 | } | |
9003 | ||
9004 | /* | |
9005 | * Kick off L2 cache whole device TRIM. | |
9006 | */ | |
9007 | if (tasks & SPA_ASYNC_L2CACHE_TRIM) { | |
9008 | mutex_enter(&spa_namespace_lock); | |
9009 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
9010 | vdev_trim_l2arc(spa); | |
9011 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
9012 | mutex_exit(&spa_namespace_lock); | |
9013 | } | |
9014 | ||
9015 | /* | |
9016 | * Kick off L2 cache rebuilding. | |
9017 | */ | |
9018 | if (tasks & SPA_ASYNC_L2CACHE_REBUILD) { | |
9019 | mutex_enter(&spa_namespace_lock); | |
9020 | spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER); | |
9021 | l2arc_spa_rebuild_start(spa); | |
9022 | spa_config_exit(spa, SCL_L2ARC, FTAG); | |
9023 | mutex_exit(&spa_namespace_lock); | |
9024 | } | |
9025 | ||
9026 | /* | |
9027 | * Let the world know that we're done. | |
9028 | */ | |
9029 | mutex_enter(&spa->spa_async_lock); | |
9030 | spa->spa_async_thread = NULL; | |
9031 | cv_broadcast(&spa->spa_async_cv); | |
9032 | mutex_exit(&spa->spa_async_lock); | |
9033 | thread_exit(); | |
9034 | } | |
9035 | ||
9036 | void | |
9037 | spa_async_suspend(spa_t *spa) | |
9038 | { | |
9039 | mutex_enter(&spa->spa_async_lock); | |
9040 | spa->spa_async_suspended++; | |
9041 | while (spa->spa_async_thread != NULL) | |
9042 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
9043 | mutex_exit(&spa->spa_async_lock); | |
9044 | ||
9045 | spa_vdev_remove_suspend(spa); | |
9046 | ||
9047 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
9048 | if (condense_thread != NULL) | |
9049 | zthr_cancel(condense_thread); | |
9050 | ||
9051 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; | |
9052 | if (raidz_expand_thread != NULL) | |
9053 | zthr_cancel(raidz_expand_thread); | |
9054 | ||
9055 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; | |
9056 | if (discard_thread != NULL) | |
9057 | zthr_cancel(discard_thread); | |
9058 | ||
9059 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
9060 | if (ll_delete_thread != NULL) | |
9061 | zthr_cancel(ll_delete_thread); | |
9062 | ||
9063 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
9064 | if (ll_condense_thread != NULL) | |
9065 | zthr_cancel(ll_condense_thread); | |
9066 | } | |
9067 | ||
9068 | void | |
9069 | spa_async_resume(spa_t *spa) | |
9070 | { | |
9071 | mutex_enter(&spa->spa_async_lock); | |
9072 | ASSERT(spa->spa_async_suspended != 0); | |
9073 | spa->spa_async_suspended--; | |
9074 | mutex_exit(&spa->spa_async_lock); | |
9075 | spa_restart_removal(spa); | |
9076 | ||
9077 | zthr_t *condense_thread = spa->spa_condense_zthr; | |
9078 | if (condense_thread != NULL) | |
9079 | zthr_resume(condense_thread); | |
9080 | ||
9081 | zthr_t *raidz_expand_thread = spa->spa_raidz_expand_zthr; | |
9082 | if (raidz_expand_thread != NULL) | |
9083 | zthr_resume(raidz_expand_thread); | |
9084 | ||
9085 | zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; | |
9086 | if (discard_thread != NULL) | |
9087 | zthr_resume(discard_thread); | |
9088 | ||
9089 | zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr; | |
9090 | if (ll_delete_thread != NULL) | |
9091 | zthr_resume(ll_delete_thread); | |
9092 | ||
9093 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
9094 | if (ll_condense_thread != NULL) | |
9095 | zthr_resume(ll_condense_thread); | |
9096 | } | |
9097 | ||
9098 | static boolean_t | |
9099 | spa_async_tasks_pending(spa_t *spa) | |
9100 | { | |
9101 | uint_t non_config_tasks; | |
9102 | uint_t config_task; | |
9103 | boolean_t config_task_suspended; | |
9104 | ||
9105 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
9106 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
9107 | if (spa->spa_ccw_fail_time == 0) { | |
9108 | config_task_suspended = B_FALSE; | |
9109 | } else { | |
9110 | config_task_suspended = | |
9111 | (gethrtime() - spa->spa_ccw_fail_time) < | |
9112 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); | |
9113 | } | |
9114 | ||
9115 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
9116 | } | |
9117 | ||
9118 | static void | |
9119 | spa_async_dispatch(spa_t *spa) | |
9120 | { | |
9121 | mutex_enter(&spa->spa_async_lock); | |
9122 | if (spa_async_tasks_pending(spa) && | |
9123 | !spa->spa_async_suspended && | |
9124 | spa->spa_async_thread == NULL) | |
9125 | spa->spa_async_thread = thread_create(NULL, 0, | |
9126 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
9127 | mutex_exit(&spa->spa_async_lock); | |
9128 | } | |
9129 | ||
9130 | void | |
9131 | spa_async_request(spa_t *spa, int task) | |
9132 | { | |
9133 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); | |
9134 | mutex_enter(&spa->spa_async_lock); | |
9135 | spa->spa_async_tasks |= task; | |
9136 | mutex_exit(&spa->spa_async_lock); | |
9137 | } | |
9138 | ||
9139 | int | |
9140 | spa_async_tasks(spa_t *spa) | |
9141 | { | |
9142 | return (spa->spa_async_tasks); | |
9143 | } | |
9144 | ||
9145 | /* | |
9146 | * ========================================================================== | |
9147 | * SPA syncing routines | |
9148 | * ========================================================================== | |
9149 | */ | |
9150 | ||
9151 | ||
9152 | static int | |
9153 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
9154 | dmu_tx_t *tx) | |
9155 | { | |
9156 | bpobj_t *bpo = arg; | |
9157 | bpobj_enqueue(bpo, bp, bp_freed, tx); | |
9158 | return (0); | |
9159 | } | |
9160 | ||
9161 | int | |
9162 | bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9163 | { | |
9164 | return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx)); | |
9165 | } | |
9166 | ||
9167 | int | |
9168 | bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9169 | { | |
9170 | return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx)); | |
9171 | } | |
9172 | ||
9173 | static int | |
9174 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
9175 | { | |
9176 | zio_t *pio = arg; | |
9177 | ||
9178 | zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp, | |
9179 | pio->io_flags)); | |
9180 | return (0); | |
9181 | } | |
9182 | ||
9183 | static int | |
9184 | bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, | |
9185 | dmu_tx_t *tx) | |
9186 | { | |
9187 | ASSERT(!bp_freed); | |
9188 | return (spa_free_sync_cb(arg, bp, tx)); | |
9189 | } | |
9190 | ||
9191 | /* | |
9192 | * Note: this simple function is not inlined to make it easier to dtrace the | |
9193 | * amount of time spent syncing frees. | |
9194 | */ | |
9195 | static void | |
9196 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
9197 | { | |
9198 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
9199 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
9200 | VERIFY(zio_wait(zio) == 0); | |
9201 | } | |
9202 | ||
9203 | /* | |
9204 | * Note: this simple function is not inlined to make it easier to dtrace the | |
9205 | * amount of time spent syncing deferred frees. | |
9206 | */ | |
9207 | static void | |
9208 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
9209 | { | |
9210 | if (spa_sync_pass(spa) != 1) | |
9211 | return; | |
9212 | ||
9213 | /* | |
9214 | * Note: | |
9215 | * If the log space map feature is active, we stop deferring | |
9216 | * frees to the next TXG and therefore running this function | |
9217 | * would be considered a no-op as spa_deferred_bpobj should | |
9218 | * not have any entries. | |
9219 | * | |
9220 | * That said we run this function anyway (instead of returning | |
9221 | * immediately) for the edge-case scenario where we just | |
9222 | * activated the log space map feature in this TXG but we have | |
9223 | * deferred frees from the previous TXG. | |
9224 | */ | |
9225 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
9226 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
9227 | bpobj_spa_free_sync_cb, zio, tx), ==, 0); | |
9228 | VERIFY0(zio_wait(zio)); | |
9229 | } | |
9230 | ||
9231 | static void | |
9232 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
9233 | { | |
9234 | char *packed = NULL; | |
9235 | size_t bufsize; | |
9236 | size_t nvsize = 0; | |
9237 | dmu_buf_t *db; | |
9238 | ||
9239 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
9240 | ||
9241 | /* | |
9242 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
9243 | * information. This avoids the dmu_buf_will_dirty() path and | |
9244 | * saves us a pre-read to get data we don't actually care about. | |
9245 | */ | |
9246 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); | |
9247 | packed = vmem_alloc(bufsize, KM_SLEEP); | |
9248 | ||
9249 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
9250 | KM_SLEEP) == 0); | |
9251 | memset(packed + nvsize, 0, bufsize - nvsize); | |
9252 | ||
9253 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); | |
9254 | ||
9255 | vmem_free(packed, bufsize); | |
9256 | ||
9257 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
9258 | dmu_buf_will_dirty(db, tx); | |
9259 | *(uint64_t *)db->db_data = nvsize; | |
9260 | dmu_buf_rele(db, FTAG); | |
9261 | } | |
9262 | ||
9263 | static void | |
9264 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
9265 | const char *config, const char *entry) | |
9266 | { | |
9267 | nvlist_t *nvroot; | |
9268 | nvlist_t **list; | |
9269 | int i; | |
9270 | ||
9271 | if (!sav->sav_sync) | |
9272 | return; | |
9273 | ||
9274 | /* | |
9275 | * Update the MOS nvlist describing the list of available devices. | |
9276 | * spa_validate_aux() will have already made sure this nvlist is | |
9277 | * valid and the vdevs are labeled appropriately. | |
9278 | */ | |
9279 | if (sav->sav_object == 0) { | |
9280 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
9281 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
9282 | sizeof (uint64_t), tx); | |
9283 | VERIFY(zap_update(spa->spa_meta_objset, | |
9284 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
9285 | &sav->sav_object, tx) == 0); | |
9286 | } | |
9287 | ||
9288 | nvroot = fnvlist_alloc(); | |
9289 | if (sav->sav_count == 0) { | |
9290 | fnvlist_add_nvlist_array(nvroot, config, | |
9291 | (const nvlist_t * const *)NULL, 0); | |
9292 | } else { | |
9293 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); | |
9294 | for (i = 0; i < sav->sav_count; i++) | |
9295 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
9296 | B_FALSE, VDEV_CONFIG_L2CACHE); | |
9297 | fnvlist_add_nvlist_array(nvroot, config, | |
9298 | (const nvlist_t * const *)list, sav->sav_count); | |
9299 | for (i = 0; i < sav->sav_count; i++) | |
9300 | nvlist_free(list[i]); | |
9301 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
9302 | } | |
9303 | ||
9304 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
9305 | nvlist_free(nvroot); | |
9306 | ||
9307 | sav->sav_sync = B_FALSE; | |
9308 | } | |
9309 | ||
9310 | /* | |
9311 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
9312 | * The all-vdev ZAP must be empty. | |
9313 | */ | |
9314 | static void | |
9315 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
9316 | { | |
9317 | spa_t *spa = vd->vdev_spa; | |
9318 | ||
9319 | if (vd->vdev_root_zap != 0 && | |
9320 | spa_feature_is_active(spa, SPA_FEATURE_AVZ_V2)) { | |
9321 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9322 | vd->vdev_root_zap, tx)); | |
9323 | } | |
9324 | if (vd->vdev_top_zap != 0) { | |
9325 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9326 | vd->vdev_top_zap, tx)); | |
9327 | } | |
9328 | if (vd->vdev_leaf_zap != 0) { | |
9329 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
9330 | vd->vdev_leaf_zap, tx)); | |
9331 | } | |
9332 | for (uint64_t i = 0; i < vd->vdev_children; i++) { | |
9333 | spa_avz_build(vd->vdev_child[i], avz, tx); | |
9334 | } | |
9335 | } | |
9336 | ||
9337 | static void | |
9338 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
9339 | { | |
9340 | nvlist_t *config; | |
9341 | ||
9342 | /* | |
9343 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
9344 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
9345 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
9346 | * need to rebuild the AVZ although the config may not be dirty. | |
9347 | */ | |
9348 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
9349 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
9350 | return; | |
9351 | ||
9352 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9353 | ||
9354 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || | |
9355 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || | |
9356 | spa->spa_all_vdev_zaps != 0); | |
9357 | ||
9358 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
9359 | /* Make and build the new AVZ */ | |
9360 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
9361 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
9362 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
9363 | ||
9364 | /* Diff old AVZ with new one */ | |
9365 | zap_cursor_t zc; | |
9366 | zap_attribute_t za; | |
9367 | ||
9368 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
9369 | spa->spa_all_vdev_zaps); | |
9370 | zap_cursor_retrieve(&zc, &za) == 0; | |
9371 | zap_cursor_advance(&zc)) { | |
9372 | uint64_t vdzap = za.za_first_integer; | |
9373 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
9374 | vdzap) == ENOENT) { | |
9375 | /* | |
9376 | * ZAP is listed in old AVZ but not in new one; | |
9377 | * destroy it | |
9378 | */ | |
9379 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
9380 | tx)); | |
9381 | } | |
9382 | } | |
9383 | ||
9384 | zap_cursor_fini(&zc); | |
9385 | ||
9386 | /* Destroy the old AVZ */ | |
9387 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9388 | spa->spa_all_vdev_zaps, tx)); | |
9389 | ||
9390 | /* Replace the old AVZ in the dir obj with the new one */ | |
9391 | VERIFY0(zap_update(spa->spa_meta_objset, | |
9392 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
9393 | sizeof (new_avz), 1, &new_avz, tx)); | |
9394 | ||
9395 | spa->spa_all_vdev_zaps = new_avz; | |
9396 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
9397 | zap_cursor_t zc; | |
9398 | zap_attribute_t za; | |
9399 | ||
9400 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
9401 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
9402 | spa->spa_all_vdev_zaps); | |
9403 | zap_cursor_retrieve(&zc, &za) == 0; | |
9404 | zap_cursor_advance(&zc)) { | |
9405 | uint64_t zap = za.za_first_integer; | |
9406 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
9407 | } | |
9408 | ||
9409 | zap_cursor_fini(&zc); | |
9410 | ||
9411 | /* Destroy and unlink the AVZ itself */ | |
9412 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
9413 | spa->spa_all_vdev_zaps, tx)); | |
9414 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
9415 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
9416 | spa->spa_all_vdev_zaps = 0; | |
9417 | } | |
9418 | ||
9419 | if (spa->spa_all_vdev_zaps == 0) { | |
9420 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
9421 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
9422 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
9423 | } | |
9424 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
9425 | ||
9426 | /* Create ZAPs for vdevs that don't have them. */ | |
9427 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
9428 | ||
9429 | config = spa_config_generate(spa, spa->spa_root_vdev, | |
9430 | dmu_tx_get_txg(tx), B_FALSE); | |
9431 | ||
9432 | /* | |
9433 | * If we're upgrading the spa version then make sure that | |
9434 | * the config object gets updated with the correct version. | |
9435 | */ | |
9436 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
9437 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
9438 | spa->spa_uberblock.ub_version); | |
9439 | ||
9440 | spa_config_exit(spa, SCL_STATE, FTAG); | |
9441 | ||
9442 | nvlist_free(spa->spa_config_syncing); | |
9443 | spa->spa_config_syncing = config; | |
9444 | ||
9445 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
9446 | } | |
9447 | ||
9448 | static void | |
9449 | spa_sync_version(void *arg, dmu_tx_t *tx) | |
9450 | { | |
9451 | uint64_t *versionp = arg; | |
9452 | uint64_t version = *versionp; | |
9453 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9454 | ||
9455 | /* | |
9456 | * Setting the version is special cased when first creating the pool. | |
9457 | */ | |
9458 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
9459 | ||
9460 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
9461 | ASSERT(version >= spa_version(spa)); | |
9462 | ||
9463 | spa->spa_uberblock.ub_version = version; | |
9464 | vdev_config_dirty(spa->spa_root_vdev); | |
9465 | spa_history_log_internal(spa, "set", tx, "version=%lld", | |
9466 | (longlong_t)version); | |
9467 | } | |
9468 | ||
9469 | /* | |
9470 | * Set zpool properties. | |
9471 | */ | |
9472 | static void | |
9473 | spa_sync_props(void *arg, dmu_tx_t *tx) | |
9474 | { | |
9475 | nvlist_t *nvp = arg; | |
9476 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9477 | objset_t *mos = spa->spa_meta_objset; | |
9478 | nvpair_t *elem = NULL; | |
9479 | ||
9480 | mutex_enter(&spa->spa_props_lock); | |
9481 | ||
9482 | while ((elem = nvlist_next_nvpair(nvp, elem))) { | |
9483 | uint64_t intval; | |
9484 | const char *strval, *fname; | |
9485 | zpool_prop_t prop; | |
9486 | const char *propname; | |
9487 | const char *elemname = nvpair_name(elem); | |
9488 | zprop_type_t proptype; | |
9489 | spa_feature_t fid; | |
9490 | ||
9491 | switch (prop = zpool_name_to_prop(elemname)) { | |
9492 | case ZPOOL_PROP_VERSION: | |
9493 | intval = fnvpair_value_uint64(elem); | |
9494 | /* | |
9495 | * The version is synced separately before other | |
9496 | * properties and should be correct by now. | |
9497 | */ | |
9498 | ASSERT3U(spa_version(spa), >=, intval); | |
9499 | break; | |
9500 | ||
9501 | case ZPOOL_PROP_ALTROOT: | |
9502 | /* | |
9503 | * 'altroot' is a non-persistent property. It should | |
9504 | * have been set temporarily at creation or import time. | |
9505 | */ | |
9506 | ASSERT(spa->spa_root != NULL); | |
9507 | break; | |
9508 | ||
9509 | case ZPOOL_PROP_READONLY: | |
9510 | case ZPOOL_PROP_CACHEFILE: | |
9511 | /* | |
9512 | * 'readonly' and 'cachefile' are also non-persistent | |
9513 | * properties. | |
9514 | */ | |
9515 | break; | |
9516 | case ZPOOL_PROP_COMMENT: | |
9517 | strval = fnvpair_value_string(elem); | |
9518 | if (spa->spa_comment != NULL) | |
9519 | spa_strfree(spa->spa_comment); | |
9520 | spa->spa_comment = spa_strdup(strval); | |
9521 | /* | |
9522 | * We need to dirty the configuration on all the vdevs | |
9523 | * so that their labels get updated. We also need to | |
9524 | * update the cache file to keep it in sync with the | |
9525 | * MOS version. It's unnecessary to do this for pool | |
9526 | * creation since the vdev's configuration has already | |
9527 | * been dirtied. | |
9528 | */ | |
9529 | if (tx->tx_txg != TXG_INITIAL) { | |
9530 | vdev_config_dirty(spa->spa_root_vdev); | |
9531 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
9532 | } | |
9533 | spa_history_log_internal(spa, "set", tx, | |
9534 | "%s=%s", elemname, strval); | |
9535 | break; | |
9536 | case ZPOOL_PROP_COMPATIBILITY: | |
9537 | strval = fnvpair_value_string(elem); | |
9538 | if (spa->spa_compatibility != NULL) | |
9539 | spa_strfree(spa->spa_compatibility); | |
9540 | spa->spa_compatibility = spa_strdup(strval); | |
9541 | /* | |
9542 | * Dirty the configuration on vdevs as above. | |
9543 | */ | |
9544 | if (tx->tx_txg != TXG_INITIAL) { | |
9545 | vdev_config_dirty(spa->spa_root_vdev); | |
9546 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
9547 | } | |
9548 | ||
9549 | spa_history_log_internal(spa, "set", tx, | |
9550 | "%s=%s", nvpair_name(elem), strval); | |
9551 | break; | |
9552 | ||
9553 | case ZPOOL_PROP_INVAL: | |
9554 | if (zpool_prop_feature(elemname)) { | |
9555 | fname = strchr(elemname, '@') + 1; | |
9556 | VERIFY0(zfeature_lookup_name(fname, &fid)); | |
9557 | ||
9558 | spa_feature_enable(spa, fid, tx); | |
9559 | spa_history_log_internal(spa, "set", tx, | |
9560 | "%s=enabled", elemname); | |
9561 | break; | |
9562 | } else if (!zfs_prop_user(elemname)) { | |
9563 | ASSERT(zpool_prop_feature(elemname)); | |
9564 | break; | |
9565 | } | |
9566 | zfs_fallthrough; | |
9567 | default: | |
9568 | /* | |
9569 | * Set pool property values in the poolprops mos object. | |
9570 | */ | |
9571 | if (spa->spa_pool_props_object == 0) { | |
9572 | spa->spa_pool_props_object = | |
9573 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
9574 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, | |
9575 | tx); | |
9576 | } | |
9577 | ||
9578 | /* normalize the property name */ | |
9579 | if (prop == ZPOOL_PROP_INVAL) { | |
9580 | propname = elemname; | |
9581 | proptype = PROP_TYPE_STRING; | |
9582 | } else { | |
9583 | propname = zpool_prop_to_name(prop); | |
9584 | proptype = zpool_prop_get_type(prop); | |
9585 | } | |
9586 | ||
9587 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
9588 | ASSERT(proptype == PROP_TYPE_STRING); | |
9589 | strval = fnvpair_value_string(elem); | |
9590 | VERIFY0(zap_update(mos, | |
9591 | spa->spa_pool_props_object, propname, | |
9592 | 1, strlen(strval) + 1, strval, tx)); | |
9593 | spa_history_log_internal(spa, "set", tx, | |
9594 | "%s=%s", elemname, strval); | |
9595 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { | |
9596 | intval = fnvpair_value_uint64(elem); | |
9597 | ||
9598 | if (proptype == PROP_TYPE_INDEX) { | |
9599 | const char *unused; | |
9600 | VERIFY0(zpool_prop_index_to_string( | |
9601 | prop, intval, &unused)); | |
9602 | } | |
9603 | VERIFY0(zap_update(mos, | |
9604 | spa->spa_pool_props_object, propname, | |
9605 | 8, 1, &intval, tx)); | |
9606 | spa_history_log_internal(spa, "set", tx, | |
9607 | "%s=%lld", elemname, | |
9608 | (longlong_t)intval); | |
9609 | ||
9610 | switch (prop) { | |
9611 | case ZPOOL_PROP_DELEGATION: | |
9612 | spa->spa_delegation = intval; | |
9613 | break; | |
9614 | case ZPOOL_PROP_BOOTFS: | |
9615 | spa->spa_bootfs = intval; | |
9616 | break; | |
9617 | case ZPOOL_PROP_FAILUREMODE: | |
9618 | spa->spa_failmode = intval; | |
9619 | break; | |
9620 | case ZPOOL_PROP_AUTOTRIM: | |
9621 | spa->spa_autotrim = intval; | |
9622 | spa_async_request(spa, | |
9623 | SPA_ASYNC_AUTOTRIM_RESTART); | |
9624 | break; | |
9625 | case ZPOOL_PROP_AUTOEXPAND: | |
9626 | spa->spa_autoexpand = intval; | |
9627 | if (tx->tx_txg != TXG_INITIAL) | |
9628 | spa_async_request(spa, | |
9629 | SPA_ASYNC_AUTOEXPAND); | |
9630 | break; | |
9631 | case ZPOOL_PROP_MULTIHOST: | |
9632 | spa->spa_multihost = intval; | |
9633 | break; | |
9634 | default: | |
9635 | break; | |
9636 | } | |
9637 | } else { | |
9638 | ASSERT(0); /* not allowed */ | |
9639 | } | |
9640 | } | |
9641 | ||
9642 | } | |
9643 | ||
9644 | mutex_exit(&spa->spa_props_lock); | |
9645 | } | |
9646 | ||
9647 | /* | |
9648 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
9649 | * reflect the new version this txg, so there must be no changes this | |
9650 | * txg to anything that the upgrade code depends on after it executes. | |
9651 | * Therefore this must be called after dsl_pool_sync() does the sync | |
9652 | * tasks. | |
9653 | */ | |
9654 | static void | |
9655 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
9656 | { | |
9657 | if (spa_sync_pass(spa) != 1) | |
9658 | return; | |
9659 | ||
9660 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9661 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); | |
9662 | ||
9663 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && | |
9664 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
9665 | dsl_pool_create_origin(dp, tx); | |
9666 | ||
9667 | /* Keeping the origin open increases spa_minref */ | |
9668 | spa->spa_minref += 3; | |
9669 | } | |
9670 | ||
9671 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
9672 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
9673 | dsl_pool_upgrade_clones(dp, tx); | |
9674 | } | |
9675 | ||
9676 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
9677 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
9678 | dsl_pool_upgrade_dir_clones(dp, tx); | |
9679 | ||
9680 | /* Keeping the freedir open increases spa_minref */ | |
9681 | spa->spa_minref += 3; | |
9682 | } | |
9683 | ||
9684 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
9685 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9686 | spa_feature_create_zap_objects(spa, tx); | |
9687 | } | |
9688 | ||
9689 | /* | |
9690 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
9691 | * when possibility to use lz4 compression for metadata was added | |
9692 | * Old pools that have this feature enabled must be upgraded to have | |
9693 | * this feature active | |
9694 | */ | |
9695 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
9696 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
9697 | SPA_FEATURE_LZ4_COMPRESS); | |
9698 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
9699 | SPA_FEATURE_LZ4_COMPRESS); | |
9700 | ||
9701 | if (lz4_en && !lz4_ac) | |
9702 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
9703 | } | |
9704 | ||
9705 | /* | |
9706 | * If we haven't written the salt, do so now. Note that the | |
9707 | * feature may not be activated yet, but that's fine since | |
9708 | * the presence of this ZAP entry is backwards compatible. | |
9709 | */ | |
9710 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
9711 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
9712 | VERIFY0(zap_add(spa->spa_meta_objset, | |
9713 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
9714 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
9715 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
9716 | } | |
9717 | ||
9718 | rrw_exit(&dp->dp_config_rwlock, FTAG); | |
9719 | } | |
9720 | ||
9721 | static void | |
9722 | vdev_indirect_state_sync_verify(vdev_t *vd) | |
9723 | { | |
9724 | vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping; | |
9725 | vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births; | |
9726 | ||
9727 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
9728 | ASSERT(vim != NULL); | |
9729 | ASSERT(vib != NULL); | |
9730 | } | |
9731 | ||
9732 | uint64_t obsolete_sm_object = 0; | |
9733 | ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); | |
9734 | if (obsolete_sm_object != 0) { | |
9735 | ASSERT(vd->vdev_obsolete_sm != NULL); | |
9736 | ASSERT(vd->vdev_removing || | |
9737 | vd->vdev_ops == &vdev_indirect_ops); | |
9738 | ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); | |
9739 | ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); | |
9740 | ASSERT3U(obsolete_sm_object, ==, | |
9741 | space_map_object(vd->vdev_obsolete_sm)); | |
9742 | ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, | |
9743 | space_map_allocated(vd->vdev_obsolete_sm)); | |
9744 | } | |
9745 | ASSERT(vd->vdev_obsolete_segments != NULL); | |
9746 | ||
9747 | /* | |
9748 | * Since frees / remaps to an indirect vdev can only | |
9749 | * happen in syncing context, the obsolete segments | |
9750 | * tree must be empty when we start syncing. | |
9751 | */ | |
9752 | ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); | |
9753 | } | |
9754 | ||
9755 | /* | |
9756 | * Set the top-level vdev's max queue depth. Evaluate each top-level's | |
9757 | * async write queue depth in case it changed. The max queue depth will | |
9758 | * not change in the middle of syncing out this txg. | |
9759 | */ | |
9760 | static void | |
9761 | spa_sync_adjust_vdev_max_queue_depth(spa_t *spa) | |
9762 | { | |
9763 | ASSERT(spa_writeable(spa)); | |
9764 | ||
9765 | vdev_t *rvd = spa->spa_root_vdev; | |
9766 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * | |
9767 | zfs_vdev_queue_depth_pct / 100; | |
9768 | metaslab_class_t *normal = spa_normal_class(spa); | |
9769 | metaslab_class_t *special = spa_special_class(spa); | |
9770 | metaslab_class_t *dedup = spa_dedup_class(spa); | |
9771 | ||
9772 | uint64_t slots_per_allocator = 0; | |
9773 | for (int c = 0; c < rvd->vdev_children; c++) { | |
9774 | vdev_t *tvd = rvd->vdev_child[c]; | |
9775 | ||
9776 | metaslab_group_t *mg = tvd->vdev_mg; | |
9777 | if (mg == NULL || !metaslab_group_initialized(mg)) | |
9778 | continue; | |
9779 | ||
9780 | metaslab_class_t *mc = mg->mg_class; | |
9781 | if (mc != normal && mc != special && mc != dedup) | |
9782 | continue; | |
9783 | ||
9784 | /* | |
9785 | * It is safe to do a lock-free check here because only async | |
9786 | * allocations look at mg_max_alloc_queue_depth, and async | |
9787 | * allocations all happen from spa_sync(). | |
9788 | */ | |
9789 | for (int i = 0; i < mg->mg_allocators; i++) { | |
9790 | ASSERT0(zfs_refcount_count( | |
9791 | &(mg->mg_allocator[i].mga_alloc_queue_depth))); | |
9792 | } | |
9793 | mg->mg_max_alloc_queue_depth = max_queue_depth; | |
9794 | ||
9795 | for (int i = 0; i < mg->mg_allocators; i++) { | |
9796 | mg->mg_allocator[i].mga_cur_max_alloc_queue_depth = | |
9797 | zfs_vdev_def_queue_depth; | |
9798 | } | |
9799 | slots_per_allocator += zfs_vdev_def_queue_depth; | |
9800 | } | |
9801 | ||
9802 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
9803 | ASSERT0(zfs_refcount_count(&normal->mc_allocator[i]. | |
9804 | mca_alloc_slots)); | |
9805 | ASSERT0(zfs_refcount_count(&special->mc_allocator[i]. | |
9806 | mca_alloc_slots)); | |
9807 | ASSERT0(zfs_refcount_count(&dedup->mc_allocator[i]. | |
9808 | mca_alloc_slots)); | |
9809 | normal->mc_allocator[i].mca_alloc_max_slots = | |
9810 | slots_per_allocator; | |
9811 | special->mc_allocator[i].mca_alloc_max_slots = | |
9812 | slots_per_allocator; | |
9813 | dedup->mc_allocator[i].mca_alloc_max_slots = | |
9814 | slots_per_allocator; | |
9815 | } | |
9816 | normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9817 | special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9818 | dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
9819 | } | |
9820 | ||
9821 | static void | |
9822 | spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx) | |
9823 | { | |
9824 | ASSERT(spa_writeable(spa)); | |
9825 | ||
9826 | vdev_t *rvd = spa->spa_root_vdev; | |
9827 | for (int c = 0; c < rvd->vdev_children; c++) { | |
9828 | vdev_t *vd = rvd->vdev_child[c]; | |
9829 | vdev_indirect_state_sync_verify(vd); | |
9830 | ||
9831 | if (vdev_indirect_should_condense(vd)) { | |
9832 | spa_condense_indirect_start_sync(vd, tx); | |
9833 | break; | |
9834 | } | |
9835 | } | |
9836 | } | |
9837 | ||
9838 | static void | |
9839 | spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx) | |
9840 | { | |
9841 | objset_t *mos = spa->spa_meta_objset; | |
9842 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
9843 | uint64_t txg = tx->tx_txg; | |
9844 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
9845 | ||
9846 | do { | |
9847 | int pass = ++spa->spa_sync_pass; | |
9848 | ||
9849 | spa_sync_config_object(spa, tx); | |
9850 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
9851 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
9852 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
9853 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
9854 | spa_errlog_sync(spa, txg); | |
9855 | dsl_pool_sync(dp, txg); | |
9856 | ||
9857 | if (pass < zfs_sync_pass_deferred_free || | |
9858 | spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { | |
9859 | /* | |
9860 | * If the log space map feature is active we don't | |
9861 | * care about deferred frees and the deferred bpobj | |
9862 | * as the log space map should effectively have the | |
9863 | * same results (i.e. appending only to one object). | |
9864 | */ | |
9865 | spa_sync_frees(spa, free_bpl, tx); | |
9866 | } else { | |
9867 | /* | |
9868 | * We can not defer frees in pass 1, because | |
9869 | * we sync the deferred frees later in pass 1. | |
9870 | */ | |
9871 | ASSERT3U(pass, >, 1); | |
9872 | bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb, | |
9873 | &spa->spa_deferred_bpobj, tx); | |
9874 | } | |
9875 | ||
9876 | brt_sync(spa, txg); | |
9877 | ddt_sync(spa, txg); | |
9878 | dsl_scan_sync(dp, tx); | |
9879 | dsl_errorscrub_sync(dp, tx); | |
9880 | svr_sync(spa, tx); | |
9881 | spa_sync_upgrades(spa, tx); | |
9882 | ||
9883 | spa_flush_metaslabs(spa, tx); | |
9884 | ||
9885 | vdev_t *vd = NULL; | |
9886 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) | |
9887 | != NULL) | |
9888 | vdev_sync(vd, txg); | |
9889 | ||
9890 | if (pass == 1) { | |
9891 | /* | |
9892 | * dsl_pool_sync() -> dp_sync_tasks may have dirtied | |
9893 | * the config. If that happens, this txg should not | |
9894 | * be a no-op. So we must sync the config to the MOS | |
9895 | * before checking for no-op. | |
9896 | * | |
9897 | * Note that when the config is dirty, it will | |
9898 | * be written to the MOS (i.e. the MOS will be | |
9899 | * dirtied) every time we call spa_sync_config_object() | |
9900 | * in this txg. Therefore we can't call this after | |
9901 | * dsl_pool_sync() every pass, because it would | |
9902 | * prevent us from converging, since we'd dirty | |
9903 | * the MOS every pass. | |
9904 | * | |
9905 | * Sync tasks can only be processed in pass 1, so | |
9906 | * there's no need to do this in later passes. | |
9907 | */ | |
9908 | spa_sync_config_object(spa, tx); | |
9909 | } | |
9910 | ||
9911 | /* | |
9912 | * Note: We need to check if the MOS is dirty because we could | |
9913 | * have marked the MOS dirty without updating the uberblock | |
9914 | * (e.g. if we have sync tasks but no dirty user data). We need | |
9915 | * to check the uberblock's rootbp because it is updated if we | |
9916 | * have synced out dirty data (though in this case the MOS will | |
9917 | * most likely also be dirty due to second order effects, we | |
9918 | * don't want to rely on that here). | |
9919 | */ | |
9920 | if (pass == 1 && | |
9921 | BP_GET_LOGICAL_BIRTH(&spa->spa_uberblock.ub_rootbp) < txg && | |
9922 | !dmu_objset_is_dirty(mos, txg)) { | |
9923 | /* | |
9924 | * Nothing changed on the first pass, therefore this | |
9925 | * TXG is a no-op. Avoid syncing deferred frees, so | |
9926 | * that we can keep this TXG as a no-op. | |
9927 | */ | |
9928 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
9929 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
9930 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
9931 | ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg)); | |
9932 | break; | |
9933 | } | |
9934 | ||
9935 | spa_sync_deferred_frees(spa, tx); | |
9936 | } while (dmu_objset_is_dirty(mos, txg)); | |
9937 | } | |
9938 | ||
9939 | /* | |
9940 | * Rewrite the vdev configuration (which includes the uberblock) to | |
9941 | * commit the transaction group. | |
9942 | * | |
9943 | * If there are no dirty vdevs, we sync the uberblock to a few random | |
9944 | * top-level vdevs that are known to be visible in the config cache | |
9945 | * (see spa_vdev_add() for a complete description). If there *are* dirty | |
9946 | * vdevs, sync the uberblock to all vdevs. | |
9947 | */ | |
9948 | static void | |
9949 | spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx) | |
9950 | { | |
9951 | vdev_t *rvd = spa->spa_root_vdev; | |
9952 | uint64_t txg = tx->tx_txg; | |
9953 | ||
9954 | for (;;) { | |
9955 | int error = 0; | |
9956 | ||
9957 | /* | |
9958 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
9959 | * while we're attempting to write the vdev labels. | |
9960 | */ | |
9961 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
9962 | ||
9963 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
9964 | vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; | |
9965 | int svdcount = 0; | |
9966 | int children = rvd->vdev_children; | |
9967 | int c0 = random_in_range(children); | |
9968 | ||
9969 | for (int c = 0; c < children; c++) { | |
9970 | vdev_t *vd = | |
9971 | rvd->vdev_child[(c0 + c) % children]; | |
9972 | ||
9973 | /* Stop when revisiting the first vdev */ | |
9974 | if (c > 0 && svd[0] == vd) | |
9975 | break; | |
9976 | ||
9977 | if (vd->vdev_ms_array == 0 || | |
9978 | vd->vdev_islog || | |
9979 | !vdev_is_concrete(vd)) | |
9980 | continue; | |
9981 | ||
9982 | svd[svdcount++] = vd; | |
9983 | if (svdcount == SPA_SYNC_MIN_VDEVS) | |
9984 | break; | |
9985 | } | |
9986 | error = vdev_config_sync(svd, svdcount, txg); | |
9987 | } else { | |
9988 | error = vdev_config_sync(rvd->vdev_child, | |
9989 | rvd->vdev_children, txg); | |
9990 | } | |
9991 | ||
9992 | if (error == 0) | |
9993 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
9994 | ||
9995 | spa_config_exit(spa, SCL_STATE, FTAG); | |
9996 | ||
9997 | if (error == 0) | |
9998 | break; | |
9999 | zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR); | |
10000 | zio_resume_wait(spa); | |
10001 | } | |
10002 | } | |
10003 | ||
10004 | /* | |
10005 | * Sync the specified transaction group. New blocks may be dirtied as | |
10006 | * part of the process, so we iterate until it converges. | |
10007 | */ | |
10008 | void | |
10009 | spa_sync(spa_t *spa, uint64_t txg) | |
10010 | { | |
10011 | vdev_t *vd = NULL; | |
10012 | ||
10013 | VERIFY(spa_writeable(spa)); | |
10014 | ||
10015 | /* | |
10016 | * Wait for i/os issued in open context that need to complete | |
10017 | * before this txg syncs. | |
10018 | */ | |
10019 | (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]); | |
10020 | spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, | |
10021 | ZIO_FLAG_CANFAIL); | |
10022 | ||
10023 | /* | |
10024 | * Now that there can be no more cloning in this transaction group, | |
10025 | * but we are still before issuing frees, we can process pending BRT | |
10026 | * updates. | |
10027 | */ | |
10028 | brt_pending_apply(spa, txg); | |
10029 | ||
10030 | /* | |
10031 | * Lock out configuration changes. | |
10032 | */ | |
10033 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
10034 | ||
10035 | spa->spa_syncing_txg = txg; | |
10036 | spa->spa_sync_pass = 0; | |
10037 | ||
10038 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
10039 | mutex_enter(&spa->spa_allocs[i].spaa_lock); | |
10040 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
10041 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
10042 | } | |
10043 | ||
10044 | /* | |
10045 | * If there are any pending vdev state changes, convert them | |
10046 | * into config changes that go out with this transaction group. | |
10047 | */ | |
10048 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
10049 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
10050 | /* Avoid holding the write lock unless actually necessary */ | |
10051 | if (vd->vdev_aux == NULL) { | |
10052 | vdev_state_clean(vd); | |
10053 | vdev_config_dirty(vd); | |
10054 | continue; | |
10055 | } | |
10056 | /* | |
10057 | * We need the write lock here because, for aux vdevs, | |
10058 | * calling vdev_config_dirty() modifies sav_config. | |
10059 | * This is ugly and will become unnecessary when we | |
10060 | * eliminate the aux vdev wart by integrating all vdevs | |
10061 | * into the root vdev tree. | |
10062 | */ | |
10063 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10064 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
10065 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
10066 | vdev_state_clean(vd); | |
10067 | vdev_config_dirty(vd); | |
10068 | } | |
10069 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10070 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10071 | } | |
10072 | spa_config_exit(spa, SCL_STATE, FTAG); | |
10073 | ||
10074 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
10075 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
10076 | ||
10077 | spa->spa_sync_starttime = gethrtime(); | |
10078 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
10079 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
10080 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
10081 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
10082 | ||
10083 | /* | |
10084 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
10085 | * set spa_deflate if we have no raid-z vdevs. | |
10086 | */ | |
10087 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
10088 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
10089 | vdev_t *rvd = spa->spa_root_vdev; | |
10090 | ||
10091 | int i; | |
10092 | for (i = 0; i < rvd->vdev_children; i++) { | |
10093 | vd = rvd->vdev_child[i]; | |
10094 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
10095 | break; | |
10096 | } | |
10097 | if (i == rvd->vdev_children) { | |
10098 | spa->spa_deflate = TRUE; | |
10099 | VERIFY0(zap_add(spa->spa_meta_objset, | |
10100 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
10101 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
10102 | } | |
10103 | } | |
10104 | ||
10105 | spa_sync_adjust_vdev_max_queue_depth(spa); | |
10106 | ||
10107 | spa_sync_condense_indirect(spa, tx); | |
10108 | ||
10109 | spa_sync_iterate_to_convergence(spa, tx); | |
10110 | ||
10111 | #ifdef ZFS_DEBUG | |
10112 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
10113 | /* | |
10114 | * Make sure that the number of ZAPs for all the vdevs matches | |
10115 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
10116 | * called if the config is dirty; otherwise there may be | |
10117 | * outstanding AVZ operations that weren't completed in | |
10118 | * spa_sync_config_object. | |
10119 | */ | |
10120 | uint64_t all_vdev_zap_entry_count; | |
10121 | ASSERT0(zap_count(spa->spa_meta_objset, | |
10122 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
10123 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
10124 | all_vdev_zap_entry_count); | |
10125 | } | |
10126 | #endif | |
10127 | ||
10128 | if (spa->spa_vdev_removal != NULL) { | |
10129 | ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); | |
10130 | } | |
10131 | ||
10132 | spa_sync_rewrite_vdev_config(spa, tx); | |
10133 | dmu_tx_commit(tx); | |
10134 | ||
10135 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
10136 | spa->spa_deadman_tqid = 0; | |
10137 | ||
10138 | /* | |
10139 | * Clear the dirty config list. | |
10140 | */ | |
10141 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) | |
10142 | vdev_config_clean(vd); | |
10143 | ||
10144 | /* | |
10145 | * Now that the new config has synced transactionally, | |
10146 | * let it become visible to the config cache. | |
10147 | */ | |
10148 | if (spa->spa_config_syncing != NULL) { | |
10149 | spa_config_set(spa, spa->spa_config_syncing); | |
10150 | spa->spa_config_txg = txg; | |
10151 | spa->spa_config_syncing = NULL; | |
10152 | } | |
10153 | ||
10154 | dsl_pool_sync_done(dp, txg); | |
10155 | ||
10156 | for (int i = 0; i < spa->spa_alloc_count; i++) { | |
10157 | mutex_enter(&spa->spa_allocs[i].spaa_lock); | |
10158 | VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree)); | |
10159 | mutex_exit(&spa->spa_allocs[i].spaa_lock); | |
10160 | } | |
10161 | ||
10162 | /* | |
10163 | * Update usable space statistics. | |
10164 | */ | |
10165 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) | |
10166 | != NULL) | |
10167 | vdev_sync_done(vd, txg); | |
10168 | ||
10169 | metaslab_class_evict_old(spa->spa_normal_class, txg); | |
10170 | metaslab_class_evict_old(spa->spa_log_class, txg); | |
10171 | ||
10172 | spa_sync_close_syncing_log_sm(spa); | |
10173 | ||
10174 | spa_update_dspace(spa); | |
10175 | ||
10176 | if (spa_get_autotrim(spa) == SPA_AUTOTRIM_ON) | |
10177 | vdev_autotrim_kick(spa); | |
10178 | ||
10179 | /* | |
10180 | * It had better be the case that we didn't dirty anything | |
10181 | * since vdev_config_sync(). | |
10182 | */ | |
10183 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
10184 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
10185 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
10186 | ||
10187 | while (zfs_pause_spa_sync) | |
10188 | delay(1); | |
10189 | ||
10190 | spa->spa_sync_pass = 0; | |
10191 | ||
10192 | /* | |
10193 | * Update the last synced uberblock here. We want to do this at | |
10194 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
10195 | * will be guaranteed that all the processing associated with | |
10196 | * that txg has been completed. | |
10197 | */ | |
10198 | spa->spa_ubsync = spa->spa_uberblock; | |
10199 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
10200 | ||
10201 | spa_handle_ignored_writes(spa); | |
10202 | ||
10203 | /* | |
10204 | * If any async tasks have been requested, kick them off. | |
10205 | */ | |
10206 | spa_async_dispatch(spa); | |
10207 | } | |
10208 | ||
10209 | /* | |
10210 | * Sync all pools. We don't want to hold the namespace lock across these | |
10211 | * operations, so we take a reference on the spa_t and drop the lock during the | |
10212 | * sync. | |
10213 | */ | |
10214 | void | |
10215 | spa_sync_allpools(void) | |
10216 | { | |
10217 | spa_t *spa = NULL; | |
10218 | mutex_enter(&spa_namespace_lock); | |
10219 | while ((spa = spa_next(spa)) != NULL) { | |
10220 | if (spa_state(spa) != POOL_STATE_ACTIVE || | |
10221 | !spa_writeable(spa) || spa_suspended(spa)) | |
10222 | continue; | |
10223 | spa_open_ref(spa, FTAG); | |
10224 | mutex_exit(&spa_namespace_lock); | |
10225 | txg_wait_synced(spa_get_dsl(spa), 0); | |
10226 | mutex_enter(&spa_namespace_lock); | |
10227 | spa_close(spa, FTAG); | |
10228 | } | |
10229 | mutex_exit(&spa_namespace_lock); | |
10230 | } | |
10231 | ||
10232 | taskq_t * | |
10233 | spa_sync_tq_create(spa_t *spa, const char *name) | |
10234 | { | |
10235 | kthread_t **kthreads; | |
10236 | ||
10237 | ASSERT(spa->spa_sync_tq == NULL); | |
10238 | ASSERT3S(spa->spa_alloc_count, <=, boot_ncpus); | |
10239 | ||
10240 | /* | |
10241 | * - do not allow more allocators than cpus. | |
10242 | * - there may be more cpus than allocators. | |
10243 | * - do not allow more sync taskq threads than allocators or cpus. | |
10244 | */ | |
10245 | int nthreads = spa->spa_alloc_count; | |
10246 | spa->spa_syncthreads = kmem_zalloc(sizeof (spa_syncthread_info_t) * | |
10247 | nthreads, KM_SLEEP); | |
10248 | ||
10249 | spa->spa_sync_tq = taskq_create_synced(name, nthreads, minclsyspri, | |
10250 | nthreads, INT_MAX, TASKQ_PREPOPULATE, &kthreads); | |
10251 | VERIFY(spa->spa_sync_tq != NULL); | |
10252 | VERIFY(kthreads != NULL); | |
10253 | ||
10254 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
10255 | for (int i = 0; i < nthreads; i++, ti++) { | |
10256 | ti->sti_thread = kthreads[i]; | |
10257 | ti->sti_allocator = i; | |
10258 | } | |
10259 | ||
10260 | kmem_free(kthreads, sizeof (*kthreads) * nthreads); | |
10261 | return (spa->spa_sync_tq); | |
10262 | } | |
10263 | ||
10264 | void | |
10265 | spa_sync_tq_destroy(spa_t *spa) | |
10266 | { | |
10267 | ASSERT(spa->spa_sync_tq != NULL); | |
10268 | ||
10269 | taskq_wait(spa->spa_sync_tq); | |
10270 | taskq_destroy(spa->spa_sync_tq); | |
10271 | kmem_free(spa->spa_syncthreads, | |
10272 | sizeof (spa_syncthread_info_t) * spa->spa_alloc_count); | |
10273 | spa->spa_sync_tq = NULL; | |
10274 | } | |
10275 | ||
10276 | uint_t | |
10277 | spa_acq_allocator(spa_t *spa) | |
10278 | { | |
10279 | int i; | |
10280 | ||
10281 | if (spa->spa_alloc_count == 1) | |
10282 | return (0); | |
10283 | ||
10284 | mutex_enter(&spa->spa_allocs_use->sau_lock); | |
10285 | uint_t r = spa->spa_allocs_use->sau_rotor; | |
10286 | do { | |
10287 | if (++r == spa->spa_alloc_count) | |
10288 | r = 0; | |
10289 | } while (spa->spa_allocs_use->sau_inuse[r]); | |
10290 | spa->spa_allocs_use->sau_inuse[r] = B_TRUE; | |
10291 | spa->spa_allocs_use->sau_rotor = r; | |
10292 | mutex_exit(&spa->spa_allocs_use->sau_lock); | |
10293 | ||
10294 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
10295 | for (i = 0; i < spa->spa_alloc_count; i++, ti++) { | |
10296 | if (ti->sti_thread == curthread) { | |
10297 | ti->sti_allocator = r; | |
10298 | break; | |
10299 | } | |
10300 | } | |
10301 | ASSERT3S(i, <, spa->spa_alloc_count); | |
10302 | return (r); | |
10303 | } | |
10304 | ||
10305 | void | |
10306 | spa_rel_allocator(spa_t *spa, uint_t allocator) | |
10307 | { | |
10308 | if (spa->spa_alloc_count > 1) | |
10309 | spa->spa_allocs_use->sau_inuse[allocator] = B_FALSE; | |
10310 | } | |
10311 | ||
10312 | void | |
10313 | spa_select_allocator(zio_t *zio) | |
10314 | { | |
10315 | zbookmark_phys_t *bm = &zio->io_bookmark; | |
10316 | spa_t *spa = zio->io_spa; | |
10317 | ||
10318 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
10319 | ||
10320 | /* | |
10321 | * A gang block (for example) may have inherited its parent's | |
10322 | * allocator, in which case there is nothing further to do here. | |
10323 | */ | |
10324 | if (ZIO_HAS_ALLOCATOR(zio)) | |
10325 | return; | |
10326 | ||
10327 | ASSERT(spa != NULL); | |
10328 | ASSERT(bm != NULL); | |
10329 | ||
10330 | /* | |
10331 | * First try to use an allocator assigned to the syncthread, and set | |
10332 | * the corresponding write issue taskq for the allocator. | |
10333 | * Note, we must have an open pool to do this. | |
10334 | */ | |
10335 | if (spa->spa_sync_tq != NULL) { | |
10336 | spa_syncthread_info_t *ti = spa->spa_syncthreads; | |
10337 | for (int i = 0; i < spa->spa_alloc_count; i++, ti++) { | |
10338 | if (ti->sti_thread == curthread) { | |
10339 | zio->io_allocator = ti->sti_allocator; | |
10340 | return; | |
10341 | } | |
10342 | } | |
10343 | } | |
10344 | ||
10345 | /* | |
10346 | * We want to try to use as many allocators as possible to help improve | |
10347 | * performance, but we also want logically adjacent IOs to be physically | |
10348 | * adjacent to improve sequential read performance. We chunk each object | |
10349 | * into 2^20 block regions, and then hash based on the objset, object, | |
10350 | * level, and region to accomplish both of these goals. | |
10351 | */ | |
10352 | uint64_t hv = cityhash4(bm->zb_objset, bm->zb_object, bm->zb_level, | |
10353 | bm->zb_blkid >> 20); | |
10354 | ||
10355 | zio->io_allocator = (uint_t)hv % spa->spa_alloc_count; | |
10356 | } | |
10357 | ||
10358 | /* | |
10359 | * ========================================================================== | |
10360 | * Miscellaneous routines | |
10361 | * ========================================================================== | |
10362 | */ | |
10363 | ||
10364 | /* | |
10365 | * Remove all pools in the system. | |
10366 | */ | |
10367 | void | |
10368 | spa_evict_all(void) | |
10369 | { | |
10370 | spa_t *spa; | |
10371 | ||
10372 | /* | |
10373 | * Remove all cached state. All pools should be closed now, | |
10374 | * so every spa in the AVL tree should be unreferenced. | |
10375 | */ | |
10376 | mutex_enter(&spa_namespace_lock); | |
10377 | while ((spa = spa_next(NULL)) != NULL) { | |
10378 | /* | |
10379 | * Stop async tasks. The async thread may need to detach | |
10380 | * a device that's been replaced, which requires grabbing | |
10381 | * spa_namespace_lock, so we must drop it here. | |
10382 | */ | |
10383 | spa_open_ref(spa, FTAG); | |
10384 | mutex_exit(&spa_namespace_lock); | |
10385 | spa_async_suspend(spa); | |
10386 | mutex_enter(&spa_namespace_lock); | |
10387 | spa_close(spa, FTAG); | |
10388 | ||
10389 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
10390 | spa_unload(spa); | |
10391 | spa_deactivate(spa); | |
10392 | } | |
10393 | spa_remove(spa); | |
10394 | } | |
10395 | mutex_exit(&spa_namespace_lock); | |
10396 | } | |
10397 | ||
10398 | vdev_t * | |
10399 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) | |
10400 | { | |
10401 | vdev_t *vd; | |
10402 | int i; | |
10403 | ||
10404 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
10405 | return (vd); | |
10406 | ||
10407 | if (aux) { | |
10408 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
10409 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
10410 | if (vd->vdev_guid == guid) | |
10411 | return (vd); | |
10412 | } | |
10413 | ||
10414 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
10415 | vd = spa->spa_spares.sav_vdevs[i]; | |
10416 | if (vd->vdev_guid == guid) | |
10417 | return (vd); | |
10418 | } | |
10419 | } | |
10420 | ||
10421 | return (NULL); | |
10422 | } | |
10423 | ||
10424 | void | |
10425 | spa_upgrade(spa_t *spa, uint64_t version) | |
10426 | { | |
10427 | ASSERT(spa_writeable(spa)); | |
10428 | ||
10429 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
10430 | ||
10431 | /* | |
10432 | * This should only be called for a non-faulted pool, and since a | |
10433 | * future version would result in an unopenable pool, this shouldn't be | |
10434 | * possible. | |
10435 | */ | |
10436 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); | |
10437 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); | |
10438 | ||
10439 | spa->spa_uberblock.ub_version = version; | |
10440 | vdev_config_dirty(spa->spa_root_vdev); | |
10441 | ||
10442 | spa_config_exit(spa, SCL_ALL, FTAG); | |
10443 | ||
10444 | txg_wait_synced(spa_get_dsl(spa), 0); | |
10445 | } | |
10446 | ||
10447 | static boolean_t | |
10448 | spa_has_aux_vdev(spa_t *spa, uint64_t guid, spa_aux_vdev_t *sav) | |
10449 | { | |
10450 | (void) spa; | |
10451 | int i; | |
10452 | uint64_t vdev_guid; | |
10453 | ||
10454 | for (i = 0; i < sav->sav_count; i++) | |
10455 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
10456 | return (B_TRUE); | |
10457 | ||
10458 | for (i = 0; i < sav->sav_npending; i++) { | |
10459 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
10460 | &vdev_guid) == 0 && vdev_guid == guid) | |
10461 | return (B_TRUE); | |
10462 | } | |
10463 | ||
10464 | return (B_FALSE); | |
10465 | } | |
10466 | ||
10467 | boolean_t | |
10468 | spa_has_l2cache(spa_t *spa, uint64_t guid) | |
10469 | { | |
10470 | return (spa_has_aux_vdev(spa, guid, &spa->spa_l2cache)); | |
10471 | } | |
10472 | ||
10473 | boolean_t | |
10474 | spa_has_spare(spa_t *spa, uint64_t guid) | |
10475 | { | |
10476 | return (spa_has_aux_vdev(spa, guid, &spa->spa_spares)); | |
10477 | } | |
10478 | ||
10479 | /* | |
10480 | * Check if a pool has an active shared spare device. | |
10481 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
10482 | */ | |
10483 | static boolean_t | |
10484 | spa_has_active_shared_spare(spa_t *spa) | |
10485 | { | |
10486 | int i, refcnt; | |
10487 | uint64_t pool; | |
10488 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
10489 | ||
10490 | for (i = 0; i < sav->sav_count; i++) { | |
10491 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
10492 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
10493 | refcnt > 2) | |
10494 | return (B_TRUE); | |
10495 | } | |
10496 | ||
10497 | return (B_FALSE); | |
10498 | } | |
10499 | ||
10500 | uint64_t | |
10501 | spa_total_metaslabs(spa_t *spa) | |
10502 | { | |
10503 | vdev_t *rvd = spa->spa_root_vdev; | |
10504 | ||
10505 | uint64_t m = 0; | |
10506 | for (uint64_t c = 0; c < rvd->vdev_children; c++) { | |
10507 | vdev_t *vd = rvd->vdev_child[c]; | |
10508 | if (!vdev_is_concrete(vd)) | |
10509 | continue; | |
10510 | m += vd->vdev_ms_count; | |
10511 | } | |
10512 | return (m); | |
10513 | } | |
10514 | ||
10515 | /* | |
10516 | * Notify any waiting threads that some activity has switched from being in- | |
10517 | * progress to not-in-progress so that the thread can wake up and determine | |
10518 | * whether it is finished waiting. | |
10519 | */ | |
10520 | void | |
10521 | spa_notify_waiters(spa_t *spa) | |
10522 | { | |
10523 | /* | |
10524 | * Acquiring spa_activities_lock here prevents the cv_broadcast from | |
10525 | * happening between the waiting thread's check and cv_wait. | |
10526 | */ | |
10527 | mutex_enter(&spa->spa_activities_lock); | |
10528 | cv_broadcast(&spa->spa_activities_cv); | |
10529 | mutex_exit(&spa->spa_activities_lock); | |
10530 | } | |
10531 | ||
10532 | /* | |
10533 | * Notify any waiting threads that the pool is exporting, and then block until | |
10534 | * they are finished using the spa_t. | |
10535 | */ | |
10536 | void | |
10537 | spa_wake_waiters(spa_t *spa) | |
10538 | { | |
10539 | mutex_enter(&spa->spa_activities_lock); | |
10540 | spa->spa_waiters_cancel = B_TRUE; | |
10541 | cv_broadcast(&spa->spa_activities_cv); | |
10542 | while (spa->spa_waiters != 0) | |
10543 | cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock); | |
10544 | spa->spa_waiters_cancel = B_FALSE; | |
10545 | mutex_exit(&spa->spa_activities_lock); | |
10546 | } | |
10547 | ||
10548 | /* Whether the vdev or any of its descendants are being initialized/trimmed. */ | |
10549 | static boolean_t | |
10550 | spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity) | |
10551 | { | |
10552 | spa_t *spa = vd->vdev_spa; | |
10553 | ||
10554 | ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER)); | |
10555 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
10556 | ASSERT(activity == ZPOOL_WAIT_INITIALIZE || | |
10557 | activity == ZPOOL_WAIT_TRIM); | |
10558 | ||
10559 | kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ? | |
10560 | &vd->vdev_initialize_lock : &vd->vdev_trim_lock; | |
10561 | ||
10562 | mutex_exit(&spa->spa_activities_lock); | |
10563 | mutex_enter(lock); | |
10564 | mutex_enter(&spa->spa_activities_lock); | |
10565 | ||
10566 | boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ? | |
10567 | (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) : | |
10568 | (vd->vdev_trim_state == VDEV_TRIM_ACTIVE); | |
10569 | mutex_exit(lock); | |
10570 | ||
10571 | if (in_progress) | |
10572 | return (B_TRUE); | |
10573 | ||
10574 | for (int i = 0; i < vd->vdev_children; i++) { | |
10575 | if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i], | |
10576 | activity)) | |
10577 | return (B_TRUE); | |
10578 | } | |
10579 | ||
10580 | return (B_FALSE); | |
10581 | } | |
10582 | ||
10583 | /* | |
10584 | * If use_guid is true, this checks whether the vdev specified by guid is | |
10585 | * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool | |
10586 | * is being initialized/trimmed. The caller must hold the config lock and | |
10587 | * spa_activities_lock. | |
10588 | */ | |
10589 | static int | |
10590 | spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid, | |
10591 | zpool_wait_activity_t activity, boolean_t *in_progress) | |
10592 | { | |
10593 | mutex_exit(&spa->spa_activities_lock); | |
10594 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10595 | mutex_enter(&spa->spa_activities_lock); | |
10596 | ||
10597 | vdev_t *vd; | |
10598 | if (use_guid) { | |
10599 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
10600 | if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) { | |
10601 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10602 | return (EINVAL); | |
10603 | } | |
10604 | } else { | |
10605 | vd = spa->spa_root_vdev; | |
10606 | } | |
10607 | ||
10608 | *in_progress = spa_vdev_activity_in_progress_impl(vd, activity); | |
10609 | ||
10610 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10611 | return (0); | |
10612 | } | |
10613 | ||
10614 | /* | |
10615 | * Locking for waiting threads | |
10616 | * --------------------------- | |
10617 | * | |
10618 | * Waiting threads need a way to check whether a given activity is in progress, | |
10619 | * and then, if it is, wait for it to complete. Each activity will have some | |
10620 | * in-memory representation of the relevant on-disk state which can be used to | |
10621 | * determine whether or not the activity is in progress. The in-memory state and | |
10622 | * the locking used to protect it will be different for each activity, and may | |
10623 | * not be suitable for use with a cvar (e.g., some state is protected by the | |
10624 | * config lock). To allow waiting threads to wait without any races, another | |
10625 | * lock, spa_activities_lock, is used. | |
10626 | * | |
10627 | * When the state is checked, both the activity-specific lock (if there is one) | |
10628 | * and spa_activities_lock are held. In some cases, the activity-specific lock | |
10629 | * is acquired explicitly (e.g. the config lock). In others, the locking is | |
10630 | * internal to some check (e.g. bpobj_is_empty). After checking, the waiting | |
10631 | * thread releases the activity-specific lock and, if the activity is in | |
10632 | * progress, then cv_waits using spa_activities_lock. | |
10633 | * | |
10634 | * The waiting thread is woken when another thread, one completing some | |
10635 | * activity, updates the state of the activity and then calls | |
10636 | * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only | |
10637 | * needs to hold its activity-specific lock when updating the state, and this | |
10638 | * lock can (but doesn't have to) be dropped before calling spa_notify_waiters. | |
10639 | * | |
10640 | * Because spa_notify_waiters acquires spa_activities_lock before broadcasting, | |
10641 | * and because it is held when the waiting thread checks the state of the | |
10642 | * activity, it can never be the case that the completing thread both updates | |
10643 | * the activity state and cv_broadcasts in between the waiting thread's check | |
10644 | * and cv_wait. Thus, a waiting thread can never miss a wakeup. | |
10645 | * | |
10646 | * In order to prevent deadlock, when the waiting thread does its check, in some | |
10647 | * cases it will temporarily drop spa_activities_lock in order to acquire the | |
10648 | * activity-specific lock. The order in which spa_activities_lock and the | |
10649 | * activity specific lock are acquired in the waiting thread is determined by | |
10650 | * the order in which they are acquired in the completing thread; if the | |
10651 | * completing thread calls spa_notify_waiters with the activity-specific lock | |
10652 | * held, then the waiting thread must also acquire the activity-specific lock | |
10653 | * first. | |
10654 | */ | |
10655 | ||
10656 | static int | |
10657 | spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity, | |
10658 | boolean_t use_tag, uint64_t tag, boolean_t *in_progress) | |
10659 | { | |
10660 | int error = 0; | |
10661 | ||
10662 | ASSERT(MUTEX_HELD(&spa->spa_activities_lock)); | |
10663 | ||
10664 | switch (activity) { | |
10665 | case ZPOOL_WAIT_CKPT_DISCARD: | |
10666 | *in_progress = | |
10667 | (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) && | |
10668 | zap_contains(spa_meta_objset(spa), | |
10669 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) == | |
10670 | ENOENT); | |
10671 | break; | |
10672 | case ZPOOL_WAIT_FREE: | |
10673 | *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS && | |
10674 | !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) || | |
10675 | spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) || | |
10676 | spa_livelist_delete_check(spa)); | |
10677 | break; | |
10678 | case ZPOOL_WAIT_INITIALIZE: | |
10679 | case ZPOOL_WAIT_TRIM: | |
10680 | error = spa_vdev_activity_in_progress(spa, use_tag, tag, | |
10681 | activity, in_progress); | |
10682 | break; | |
10683 | case ZPOOL_WAIT_REPLACE: | |
10684 | mutex_exit(&spa->spa_activities_lock); | |
10685 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
10686 | mutex_enter(&spa->spa_activities_lock); | |
10687 | ||
10688 | *in_progress = vdev_replace_in_progress(spa->spa_root_vdev); | |
10689 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
10690 | break; | |
10691 | case ZPOOL_WAIT_REMOVE: | |
10692 | *in_progress = (spa->spa_removing_phys.sr_state == | |
10693 | DSS_SCANNING); | |
10694 | break; | |
10695 | case ZPOOL_WAIT_RESILVER: | |
10696 | *in_progress = vdev_rebuild_active(spa->spa_root_vdev); | |
10697 | if (*in_progress) | |
10698 | break; | |
10699 | zfs_fallthrough; | |
10700 | case ZPOOL_WAIT_SCRUB: | |
10701 | { | |
10702 | boolean_t scanning, paused, is_scrub; | |
10703 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
10704 | ||
10705 | is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB); | |
10706 | scanning = (scn->scn_phys.scn_state == DSS_SCANNING); | |
10707 | paused = dsl_scan_is_paused_scrub(scn); | |
10708 | *in_progress = (scanning && !paused && | |
10709 | is_scrub == (activity == ZPOOL_WAIT_SCRUB)); | |
10710 | break; | |
10711 | } | |
10712 | case ZPOOL_WAIT_RAIDZ_EXPAND: | |
10713 | { | |
10714 | vdev_raidz_expand_t *vre = spa->spa_raidz_expand; | |
10715 | *in_progress = (vre != NULL && vre->vre_state == DSS_SCANNING); | |
10716 | break; | |
10717 | } | |
10718 | default: | |
10719 | panic("unrecognized value for activity %d", activity); | |
10720 | } | |
10721 | ||
10722 | return (error); | |
10723 | } | |
10724 | ||
10725 | static int | |
10726 | spa_wait_common(const char *pool, zpool_wait_activity_t activity, | |
10727 | boolean_t use_tag, uint64_t tag, boolean_t *waited) | |
10728 | { | |
10729 | /* | |
10730 | * The tag is used to distinguish between instances of an activity. | |
10731 | * 'initialize' and 'trim' are the only activities that we use this for. | |
10732 | * The other activities can only have a single instance in progress in a | |
10733 | * pool at one time, making the tag unnecessary. | |
10734 | * | |
10735 | * There can be multiple devices being replaced at once, but since they | |
10736 | * all finish once resilvering finishes, we don't bother keeping track | |
10737 | * of them individually, we just wait for them all to finish. | |
10738 | */ | |
10739 | if (use_tag && activity != ZPOOL_WAIT_INITIALIZE && | |
10740 | activity != ZPOOL_WAIT_TRIM) | |
10741 | return (EINVAL); | |
10742 | ||
10743 | if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES) | |
10744 | return (EINVAL); | |
10745 | ||
10746 | spa_t *spa; | |
10747 | int error = spa_open(pool, &spa, FTAG); | |
10748 | if (error != 0) | |
10749 | return (error); | |
10750 | ||
10751 | /* | |
10752 | * Increment the spa's waiter count so that we can call spa_close and | |
10753 | * still ensure that the spa_t doesn't get freed before this thread is | |
10754 | * finished with it when the pool is exported. We want to call spa_close | |
10755 | * before we start waiting because otherwise the additional ref would | |
10756 | * prevent the pool from being exported or destroyed throughout the | |
10757 | * potentially long wait. | |
10758 | */ | |
10759 | mutex_enter(&spa->spa_activities_lock); | |
10760 | spa->spa_waiters++; | |
10761 | spa_close(spa, FTAG); | |
10762 | ||
10763 | *waited = B_FALSE; | |
10764 | for (;;) { | |
10765 | boolean_t in_progress; | |
10766 | error = spa_activity_in_progress(spa, activity, use_tag, tag, | |
10767 | &in_progress); | |
10768 | ||
10769 | if (error || !in_progress || spa->spa_waiters_cancel) | |
10770 | break; | |
10771 | ||
10772 | *waited = B_TRUE; | |
10773 | ||
10774 | if (cv_wait_sig(&spa->spa_activities_cv, | |
10775 | &spa->spa_activities_lock) == 0) { | |
10776 | error = EINTR; | |
10777 | break; | |
10778 | } | |
10779 | } | |
10780 | ||
10781 | spa->spa_waiters--; | |
10782 | cv_signal(&spa->spa_waiters_cv); | |
10783 | mutex_exit(&spa->spa_activities_lock); | |
10784 | ||
10785 | return (error); | |
10786 | } | |
10787 | ||
10788 | /* | |
10789 | * Wait for a particular instance of the specified activity to complete, where | |
10790 | * the instance is identified by 'tag' | |
10791 | */ | |
10792 | int | |
10793 | spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag, | |
10794 | boolean_t *waited) | |
10795 | { | |
10796 | return (spa_wait_common(pool, activity, B_TRUE, tag, waited)); | |
10797 | } | |
10798 | ||
10799 | /* | |
10800 | * Wait for all instances of the specified activity complete | |
10801 | */ | |
10802 | int | |
10803 | spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited) | |
10804 | { | |
10805 | ||
10806 | return (spa_wait_common(pool, activity, B_FALSE, 0, waited)); | |
10807 | } | |
10808 | ||
10809 | sysevent_t * | |
10810 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) | |
10811 | { | |
10812 | sysevent_t *ev = NULL; | |
10813 | #ifdef _KERNEL | |
10814 | nvlist_t *resource; | |
10815 | ||
10816 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
10817 | if (resource) { | |
10818 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
10819 | ev->resource = resource; | |
10820 | } | |
10821 | #else | |
10822 | (void) spa, (void) vd, (void) hist_nvl, (void) name; | |
10823 | #endif | |
10824 | return (ev); | |
10825 | } | |
10826 | ||
10827 | void | |
10828 | spa_event_post(sysevent_t *ev) | |
10829 | { | |
10830 | #ifdef _KERNEL | |
10831 | if (ev) { | |
10832 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
10833 | kmem_free(ev, sizeof (*ev)); | |
10834 | } | |
10835 | #else | |
10836 | (void) ev; | |
10837 | #endif | |
10838 | } | |
10839 | ||
10840 | /* | |
10841 | * Post a zevent corresponding to the given sysevent. The 'name' must be one | |
10842 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
10843 | * filled in from the spa and (optionally) the vdev. This doesn't do anything | |
10844 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
10845 | * or zdb as real changes. | |
10846 | */ | |
10847 | void | |
10848 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) | |
10849 | { | |
10850 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); | |
10851 | } | |
10852 | ||
10853 | /* state manipulation functions */ | |
10854 | EXPORT_SYMBOL(spa_open); | |
10855 | EXPORT_SYMBOL(spa_open_rewind); | |
10856 | EXPORT_SYMBOL(spa_get_stats); | |
10857 | EXPORT_SYMBOL(spa_create); | |
10858 | EXPORT_SYMBOL(spa_import); | |
10859 | EXPORT_SYMBOL(spa_tryimport); | |
10860 | EXPORT_SYMBOL(spa_destroy); | |
10861 | EXPORT_SYMBOL(spa_export); | |
10862 | EXPORT_SYMBOL(spa_reset); | |
10863 | EXPORT_SYMBOL(spa_async_request); | |
10864 | EXPORT_SYMBOL(spa_async_suspend); | |
10865 | EXPORT_SYMBOL(spa_async_resume); | |
10866 | EXPORT_SYMBOL(spa_inject_addref); | |
10867 | EXPORT_SYMBOL(spa_inject_delref); | |
10868 | EXPORT_SYMBOL(spa_scan_stat_init); | |
10869 | EXPORT_SYMBOL(spa_scan_get_stats); | |
10870 | ||
10871 | /* device manipulation */ | |
10872 | EXPORT_SYMBOL(spa_vdev_add); | |
10873 | EXPORT_SYMBOL(spa_vdev_attach); | |
10874 | EXPORT_SYMBOL(spa_vdev_detach); | |
10875 | EXPORT_SYMBOL(spa_vdev_setpath); | |
10876 | EXPORT_SYMBOL(spa_vdev_setfru); | |
10877 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
10878 | ||
10879 | /* spare statech is global across all pools) */ | |
10880 | EXPORT_SYMBOL(spa_spare_add); | |
10881 | EXPORT_SYMBOL(spa_spare_remove); | |
10882 | EXPORT_SYMBOL(spa_spare_exists); | |
10883 | EXPORT_SYMBOL(spa_spare_activate); | |
10884 | ||
10885 | /* L2ARC statech is global across all pools) */ | |
10886 | EXPORT_SYMBOL(spa_l2cache_add); | |
10887 | EXPORT_SYMBOL(spa_l2cache_remove); | |
10888 | EXPORT_SYMBOL(spa_l2cache_exists); | |
10889 | EXPORT_SYMBOL(spa_l2cache_activate); | |
10890 | EXPORT_SYMBOL(spa_l2cache_drop); | |
10891 | ||
10892 | /* scanning */ | |
10893 | EXPORT_SYMBOL(spa_scan); | |
10894 | EXPORT_SYMBOL(spa_scan_stop); | |
10895 | ||
10896 | /* spa syncing */ | |
10897 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
10898 | EXPORT_SYMBOL(spa_sync_allpools); | |
10899 | ||
10900 | /* properties */ | |
10901 | EXPORT_SYMBOL(spa_prop_set); | |
10902 | EXPORT_SYMBOL(spa_prop_get); | |
10903 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
10904 | ||
10905 | /* asynchronous event notification */ | |
10906 | EXPORT_SYMBOL(spa_event_notify); | |
10907 | ||
10908 | ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, preload_pct, UINT, ZMOD_RW, | |
10909 | "Percentage of CPUs to run a metaslab preload taskq"); | |
10910 | ||
10911 | /* BEGIN CSTYLED */ | |
10912 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, UINT, ZMOD_RW, | |
10913 | "log2 fraction of arc that can be used by inflight I/Os when " | |
10914 | "verifying pool during import"); | |
10915 | /* END CSTYLED */ | |
10916 | ||
10917 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW, | |
10918 | "Set to traverse metadata on pool import"); | |
10919 | ||
10920 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW, | |
10921 | "Set to traverse data on pool import"); | |
10922 | ||
10923 | ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW, | |
10924 | "Print vdev tree to zfs_dbgmsg during pool import"); | |
10925 | ||
10926 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RW, | |
10927 | "Percentage of CPUs to run an IO worker thread"); | |
10928 | ||
10929 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_tpq, UINT, ZMOD_RW, | |
10930 | "Number of threads per IO worker taskqueue"); | |
10931 | ||
10932 | /* BEGIN CSTYLED */ | |
10933 | ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, U64, ZMOD_RW, | |
10934 | "Allow importing pool with up to this number of missing top-level " | |
10935 | "vdevs (in read-only mode)"); | |
10936 | /* END CSTYLED */ | |
10937 | ||
10938 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, | |
10939 | ZMOD_RW, "Set the livelist condense zthr to pause"); | |
10940 | ||
10941 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, | |
10942 | ZMOD_RW, "Set the livelist condense synctask to pause"); | |
10943 | ||
10944 | /* BEGIN CSTYLED */ | |
10945 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, | |
10946 | INT, ZMOD_RW, | |
10947 | "Whether livelist condensing was canceled in the synctask"); | |
10948 | ||
10949 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, | |
10950 | INT, ZMOD_RW, | |
10951 | "Whether livelist condensing was canceled in the zthr function"); | |
10952 | ||
10953 | ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, | |
10954 | ZMOD_RW, | |
10955 | "Whether extra ALLOC blkptrs were added to a livelist entry while it " | |
10956 | "was being condensed"); | |
10957 | ||
10958 | #ifdef _KERNEL | |
10959 | ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio, zio_, taskq_read, | |
10960 | spa_taskq_read_param_set, spa_taskq_read_param_get, ZMOD_RW, | |
10961 | "Configure IO queues for read IO"); | |
10962 | ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio, zio_, taskq_write, | |
10963 | spa_taskq_write_param_set, spa_taskq_write_param_get, ZMOD_RW, | |
10964 | "Configure IO queues for write IO"); | |
10965 | #endif | |
10966 | /* END CSTYLED */ | |
10967 | ||
10968 | ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_write_tpq, UINT, ZMOD_RW, | |
10969 | "Number of CPUs per write issue taskq"); |