]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
4747a7d3 | 24 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
fb390aaf | 25 | * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. |
62bdd5eb | 26 | * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. |
0c66c32d | 27 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
3c67d83a | 28 | * Copyright 2013 Saso Kiselkov. All rights reserved. |
e550644f BB |
29 | * Copyright (c) 2014 Integros [integros.com] |
30 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
a0bd735a | 31 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
0ea05c64 | 32 | * Copyright (c) 2017 Datto Inc. |
12fa0466 | 33 | * Copyright 2017 Joyent, Inc. |
a38718a6 | 34 | */ |
34dc7c2f | 35 | |
34dc7c2f | 36 | /* |
e49f1e20 WA |
37 | * SPA: Storage Pool Allocator |
38 | * | |
34dc7c2f BB |
39 | * This file contains all the routines used when modifying on-disk SPA state. |
40 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
41 | * pool. | |
42 | */ | |
43 | ||
44 | #include <sys/zfs_context.h> | |
45 | #include <sys/fm/fs/zfs.h> | |
46 | #include <sys/spa_impl.h> | |
47 | #include <sys/zio.h> | |
48 | #include <sys/zio_checksum.h> | |
34dc7c2f BB |
49 | #include <sys/dmu.h> |
50 | #include <sys/dmu_tx.h> | |
51 | #include <sys/zap.h> | |
52 | #include <sys/zil.h> | |
428870ff | 53 | #include <sys/ddt.h> |
34dc7c2f | 54 | #include <sys/vdev_impl.h> |
a1d477c2 MA |
55 | #include <sys/vdev_removal.h> |
56 | #include <sys/vdev_indirect_mapping.h> | |
57 | #include <sys/vdev_indirect_births.h> | |
c28b2279 | 58 | #include <sys/vdev_disk.h> |
34dc7c2f | 59 | #include <sys/metaslab.h> |
428870ff | 60 | #include <sys/metaslab_impl.h> |
379ca9cf | 61 | #include <sys/mmp.h> |
34dc7c2f BB |
62 | #include <sys/uberblock_impl.h> |
63 | #include <sys/txg.h> | |
64 | #include <sys/avl.h> | |
a1d477c2 | 65 | #include <sys/bpobj.h> |
34dc7c2f BB |
66 | #include <sys/dmu_traverse.h> |
67 | #include <sys/dmu_objset.h> | |
68 | #include <sys/unique.h> | |
69 | #include <sys/dsl_pool.h> | |
70 | #include <sys/dsl_dataset.h> | |
71 | #include <sys/dsl_dir.h> | |
72 | #include <sys/dsl_prop.h> | |
73 | #include <sys/dsl_synctask.h> | |
74 | #include <sys/fs/zfs.h> | |
75 | #include <sys/arc.h> | |
76 | #include <sys/callb.h> | |
77 | #include <sys/systeminfo.h> | |
34dc7c2f | 78 | #include <sys/spa_boot.h> |
9babb374 | 79 | #include <sys/zfs_ioctl.h> |
428870ff | 80 | #include <sys/dsl_scan.h> |
9ae529ec | 81 | #include <sys/zfeature.h> |
13fe0198 | 82 | #include <sys/dsl_destroy.h> |
526af785 | 83 | #include <sys/zvol.h> |
34dc7c2f | 84 | |
d164b209 | 85 | #ifdef _KERNEL |
12fa0466 DE |
86 | #include <sys/fm/protocol.h> |
87 | #include <sys/fm/util.h> | |
428870ff BB |
88 | #include <sys/bootprops.h> |
89 | #include <sys/callb.h> | |
90 | #include <sys/cpupart.h> | |
91 | #include <sys/pool.h> | |
92 | #include <sys/sysdc.h> | |
d164b209 BB |
93 | #include <sys/zone.h> |
94 | #endif /* _KERNEL */ | |
95 | ||
34dc7c2f BB |
96 | #include "zfs_prop.h" |
97 | #include "zfs_comutil.h" | |
98 | ||
e6cfd633 WA |
99 | /* |
100 | * The interval, in seconds, at which failed configuration cache file writes | |
101 | * should be retried. | |
102 | */ | |
a1d477c2 | 103 | int zfs_ccw_retry_interval = 300; |
e6cfd633 | 104 | |
428870ff | 105 | typedef enum zti_modes { |
7ef5e54e | 106 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ |
7ef5e54e AL |
107 | ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ |
108 | ZTI_MODE_NULL, /* don't create a taskq */ | |
109 | ZTI_NMODES | |
428870ff | 110 | } zti_modes_t; |
34dc7c2f | 111 | |
7ef5e54e AL |
112 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } |
113 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
114 | #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } | |
115 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } | |
9babb374 | 116 | |
7ef5e54e AL |
117 | #define ZTI_N(n) ZTI_P(n, 1) |
118 | #define ZTI_ONE ZTI_N(1) | |
9babb374 BB |
119 | |
120 | typedef struct zio_taskq_info { | |
7ef5e54e | 121 | zti_modes_t zti_mode; |
428870ff | 122 | uint_t zti_value; |
7ef5e54e | 123 | uint_t zti_count; |
9babb374 BB |
124 | } zio_taskq_info_t; |
125 | ||
126 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
451041db | 127 | "iss", "iss_h", "int", "int_h" |
9babb374 BB |
128 | }; |
129 | ||
428870ff | 130 | /* |
7ef5e54e AL |
131 | * This table defines the taskq settings for each ZFS I/O type. When |
132 | * initializing a pool, we use this table to create an appropriately sized | |
133 | * taskq. Some operations are low volume and therefore have a small, static | |
134 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
135 | * macros. Other operations process a large amount of data; the ZTI_BATCH | |
136 | * macro causes us to create a taskq oriented for throughput. Some operations | |
137 | * are so high frequency and short-lived that the taskq itself can become a a | |
138 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an | |
139 | * additional degree of parallelism specified by the number of threads per- | |
140 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
141 | * particular taskq is chosen at random. | |
142 | * | |
143 | * The different taskq priorities are to handle the different contexts (issue | |
144 | * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that | |
145 | * need to be handled with minimum delay. | |
428870ff BB |
146 | */ |
147 | const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { | |
148 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ | |
7ef5e54e | 149 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ |
aa9af22c BB |
150 | { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ |
151 | { ZTI_BATCH, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */ | |
152 | { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ | |
7ef5e54e AL |
153 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ |
154 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ | |
9babb374 BB |
155 | }; |
156 | ||
13fe0198 MA |
157 | static void spa_sync_version(void *arg, dmu_tx_t *tx); |
158 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
b128c09f | 159 | static boolean_t spa_has_active_shared_spare(spa_t *spa); |
bf701a83 | 160 | static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, |
a1d477c2 | 161 | spa_load_state_t state, spa_import_type_t type, boolean_t trust_config, |
428870ff | 162 | char **ereport); |
572e2857 | 163 | static void spa_vdev_resilver_done(spa_t *spa); |
428870ff | 164 | |
e8b96c60 | 165 | uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ |
428870ff BB |
166 | id_t zio_taskq_psrset_bind = PS_NONE; |
167 | boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ | |
168 | uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
169 | ||
170 | boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ | |
171 | ||
172 | /* | |
173 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
174 | * to get the vdev stats associated with the imported devices. | |
175 | */ | |
176 | #define TRYIMPORT_NAME "$import" | |
34dc7c2f BB |
177 | |
178 | /* | |
179 | * ========================================================================== | |
180 | * SPA properties routines | |
181 | * ========================================================================== | |
182 | */ | |
183 | ||
184 | /* | |
185 | * Add a (source=src, propname=propval) list to an nvlist. | |
186 | */ | |
187 | static void | |
188 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, | |
189 | uint64_t intval, zprop_source_t src) | |
190 | { | |
191 | const char *propname = zpool_prop_to_name(prop); | |
192 | nvlist_t *propval; | |
193 | ||
79c76d5b | 194 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); |
34dc7c2f BB |
195 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); |
196 | ||
197 | if (strval != NULL) | |
198 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
199 | else | |
200 | VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); | |
201 | ||
202 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
203 | nvlist_free(propval); | |
204 | } | |
205 | ||
206 | /* | |
207 | * Get property values from the spa configuration. | |
208 | */ | |
209 | static void | |
210 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
211 | { | |
1bd201e7 | 212 | vdev_t *rvd = spa->spa_root_vdev; |
9ae529ec | 213 | dsl_pool_t *pool = spa->spa_dsl_pool; |
f3a7f661 | 214 | uint64_t size, alloc, cap, version; |
82ab6848 | 215 | const zprop_source_t src = ZPROP_SRC_NONE; |
b128c09f | 216 | spa_config_dirent_t *dp; |
f3a7f661 | 217 | metaslab_class_t *mc = spa_normal_class(spa); |
b128c09f BB |
218 | |
219 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
34dc7c2f | 220 | |
1bd201e7 | 221 | if (rvd != NULL) { |
428870ff BB |
222 | alloc = metaslab_class_get_alloc(spa_normal_class(spa)); |
223 | size = metaslab_class_get_space(spa_normal_class(spa)); | |
d164b209 BB |
224 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); |
225 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
428870ff BB |
226 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); |
227 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
228 | size - alloc, src); | |
1bd201e7 | 229 | |
f3a7f661 GW |
230 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, |
231 | metaslab_class_fragmentation(mc), src); | |
232 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
233 | metaslab_class_expandable_space(mc), src); | |
572e2857 BB |
234 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, |
235 | (spa_mode(spa) == FREAD), src); | |
d164b209 | 236 | |
428870ff | 237 | cap = (size == 0) ? 0 : (alloc * 100 / size); |
d164b209 BB |
238 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); |
239 | ||
428870ff BB |
240 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, |
241 | ddt_get_pool_dedup_ratio(spa), src); | |
242 | ||
d164b209 | 243 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, |
1bd201e7 | 244 | rvd->vdev_state, src); |
d164b209 BB |
245 | |
246 | version = spa_version(spa); | |
82ab6848 HM |
247 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { |
248 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
249 | version, ZPROP_SRC_DEFAULT); | |
250 | } else { | |
251 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
252 | version, ZPROP_SRC_LOCAL); | |
253 | } | |
d164b209 | 254 | } |
34dc7c2f | 255 | |
9ae529ec | 256 | if (pool != NULL) { |
9ae529ec CS |
257 | /* |
258 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
259 | * when opening pools before this version freedir will be NULL. | |
260 | */ | |
fbeddd60 | 261 | if (pool->dp_free_dir != NULL) { |
9ae529ec | 262 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, |
d683ddbb JG |
263 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, |
264 | src); | |
9ae529ec CS |
265 | } else { |
266 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
267 | NULL, 0, src); | |
268 | } | |
fbeddd60 MA |
269 | |
270 | if (pool->dp_leak_dir != NULL) { | |
271 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
d683ddbb JG |
272 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, |
273 | src); | |
fbeddd60 MA |
274 | } else { |
275 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
276 | NULL, 0, src); | |
277 | } | |
9ae529ec CS |
278 | } |
279 | ||
34dc7c2f | 280 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); |
34dc7c2f | 281 | |
d96eb2b1 DM |
282 | if (spa->spa_comment != NULL) { |
283 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
284 | 0, ZPROP_SRC_LOCAL); | |
285 | } | |
286 | ||
34dc7c2f BB |
287 | if (spa->spa_root != NULL) |
288 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
289 | 0, ZPROP_SRC_LOCAL); | |
290 | ||
f1512ee6 MA |
291 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { |
292 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
293 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
294 | } else { | |
295 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
296 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
297 | } | |
298 | ||
50c957f7 NB |
299 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { |
300 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
301 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
302 | } else { | |
303 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
304 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
305 | } | |
306 | ||
b128c09f BB |
307 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { |
308 | if (dp->scd_path == NULL) { | |
34dc7c2f | 309 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f BB |
310 | "none", 0, ZPROP_SRC_LOCAL); |
311 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
34dc7c2f | 312 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, |
b128c09f | 313 | dp->scd_path, 0, ZPROP_SRC_LOCAL); |
34dc7c2f BB |
314 | } |
315 | } | |
316 | } | |
317 | ||
318 | /* | |
319 | * Get zpool property values. | |
320 | */ | |
321 | int | |
322 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
323 | { | |
428870ff | 324 | objset_t *mos = spa->spa_meta_objset; |
34dc7c2f BB |
325 | zap_cursor_t zc; |
326 | zap_attribute_t za; | |
34dc7c2f BB |
327 | int err; |
328 | ||
79c76d5b | 329 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); |
c28b2279 | 330 | if (err) |
d1d7e268 | 331 | return (err); |
34dc7c2f | 332 | |
b128c09f BB |
333 | mutex_enter(&spa->spa_props_lock); |
334 | ||
34dc7c2f BB |
335 | /* |
336 | * Get properties from the spa config. | |
337 | */ | |
338 | spa_prop_get_config(spa, nvp); | |
339 | ||
34dc7c2f | 340 | /* If no pool property object, no more prop to get. */ |
428870ff | 341 | if (mos == NULL || spa->spa_pool_props_object == 0) { |
34dc7c2f | 342 | mutex_exit(&spa->spa_props_lock); |
c28b2279 | 343 | goto out; |
34dc7c2f BB |
344 | } |
345 | ||
346 | /* | |
347 | * Get properties from the MOS pool property object. | |
348 | */ | |
349 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
350 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
351 | zap_cursor_advance(&zc)) { | |
352 | uint64_t intval = 0; | |
353 | char *strval = NULL; | |
354 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
355 | zpool_prop_t prop; | |
356 | ||
31864e3d | 357 | if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) |
34dc7c2f BB |
358 | continue; |
359 | ||
360 | switch (za.za_integer_length) { | |
361 | case 8: | |
362 | /* integer property */ | |
363 | if (za.za_first_integer != | |
364 | zpool_prop_default_numeric(prop)) | |
365 | src = ZPROP_SRC_LOCAL; | |
366 | ||
367 | if (prop == ZPOOL_PROP_BOOTFS) { | |
368 | dsl_pool_t *dp; | |
369 | dsl_dataset_t *ds = NULL; | |
370 | ||
371 | dp = spa_get_dsl(spa); | |
13fe0198 | 372 | dsl_pool_config_enter(dp, FTAG); |
c65aa5b2 BB |
373 | if ((err = dsl_dataset_hold_obj(dp, |
374 | za.za_first_integer, FTAG, &ds))) { | |
13fe0198 | 375 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f BB |
376 | break; |
377 | } | |
378 | ||
eca7b760 | 379 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, |
79c76d5b | 380 | KM_SLEEP); |
34dc7c2f | 381 | dsl_dataset_name(ds, strval); |
b128c09f | 382 | dsl_dataset_rele(ds, FTAG); |
13fe0198 | 383 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f BB |
384 | } else { |
385 | strval = NULL; | |
386 | intval = za.za_first_integer; | |
387 | } | |
388 | ||
389 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
390 | ||
391 | if (strval != NULL) | |
eca7b760 | 392 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
393 | |
394 | break; | |
395 | ||
396 | case 1: | |
397 | /* string property */ | |
79c76d5b | 398 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); |
34dc7c2f BB |
399 | err = zap_lookup(mos, spa->spa_pool_props_object, |
400 | za.za_name, 1, za.za_num_integers, strval); | |
401 | if (err) { | |
402 | kmem_free(strval, za.za_num_integers); | |
403 | break; | |
404 | } | |
405 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
406 | kmem_free(strval, za.za_num_integers); | |
407 | break; | |
408 | ||
409 | default: | |
410 | break; | |
411 | } | |
412 | } | |
413 | zap_cursor_fini(&zc); | |
414 | mutex_exit(&spa->spa_props_lock); | |
415 | out: | |
416 | if (err && err != ENOENT) { | |
417 | nvlist_free(*nvp); | |
418 | *nvp = NULL; | |
419 | return (err); | |
420 | } | |
421 | ||
422 | return (0); | |
423 | } | |
424 | ||
425 | /* | |
426 | * Validate the given pool properties nvlist and modify the list | |
427 | * for the property values to be set. | |
428 | */ | |
429 | static int | |
430 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
431 | { | |
432 | nvpair_t *elem; | |
433 | int error = 0, reset_bootfs = 0; | |
d4ed6673 | 434 | uint64_t objnum = 0; |
9ae529ec | 435 | boolean_t has_feature = B_FALSE; |
34dc7c2f BB |
436 | |
437 | elem = NULL; | |
438 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
34dc7c2f | 439 | uint64_t intval; |
9ae529ec CS |
440 | char *strval, *slash, *check, *fname; |
441 | const char *propname = nvpair_name(elem); | |
442 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
443 | ||
31864e3d BB |
444 | switch (prop) { |
445 | case ZPOOL_PROP_INVAL: | |
9ae529ec | 446 | if (!zpool_prop_feature(propname)) { |
2e528b49 | 447 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
448 | break; |
449 | } | |
450 | ||
451 | /* | |
452 | * Sanitize the input. | |
453 | */ | |
454 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
2e528b49 | 455 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
456 | break; |
457 | } | |
458 | ||
459 | if (nvpair_value_uint64(elem, &intval) != 0) { | |
2e528b49 | 460 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
461 | break; |
462 | } | |
34dc7c2f | 463 | |
9ae529ec | 464 | if (intval != 0) { |
2e528b49 | 465 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
466 | break; |
467 | } | |
34dc7c2f | 468 | |
9ae529ec CS |
469 | fname = strchr(propname, '@') + 1; |
470 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
2e528b49 | 471 | error = SET_ERROR(EINVAL); |
9ae529ec CS |
472 | break; |
473 | } | |
474 | ||
475 | has_feature = B_TRUE; | |
476 | break; | |
34dc7c2f | 477 | |
34dc7c2f BB |
478 | case ZPOOL_PROP_VERSION: |
479 | error = nvpair_value_uint64(elem, &intval); | |
480 | if (!error && | |
9ae529ec CS |
481 | (intval < spa_version(spa) || |
482 | intval > SPA_VERSION_BEFORE_FEATURES || | |
483 | has_feature)) | |
2e528b49 | 484 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
485 | break; |
486 | ||
487 | case ZPOOL_PROP_DELEGATION: | |
488 | case ZPOOL_PROP_AUTOREPLACE: | |
b128c09f | 489 | case ZPOOL_PROP_LISTSNAPS: |
9babb374 | 490 | case ZPOOL_PROP_AUTOEXPAND: |
34dc7c2f BB |
491 | error = nvpair_value_uint64(elem, &intval); |
492 | if (!error && intval > 1) | |
2e528b49 | 493 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
494 | break; |
495 | ||
379ca9cf OF |
496 | case ZPOOL_PROP_MULTIHOST: |
497 | error = nvpair_value_uint64(elem, &intval); | |
498 | if (!error && intval > 1) | |
499 | error = SET_ERROR(EINVAL); | |
500 | ||
501 | if (!error && !spa_get_hostid()) | |
502 | error = SET_ERROR(ENOTSUP); | |
503 | ||
504 | break; | |
505 | ||
34dc7c2f | 506 | case ZPOOL_PROP_BOOTFS: |
9babb374 BB |
507 | /* |
508 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
509 | * or the pool is still being created (version == 0), | |
510 | * the bootfs property cannot be set. | |
511 | */ | |
34dc7c2f | 512 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { |
2e528b49 | 513 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
514 | break; |
515 | } | |
516 | ||
517 | /* | |
b128c09f | 518 | * Make sure the vdev config is bootable |
34dc7c2f | 519 | */ |
b128c09f | 520 | if (!vdev_is_bootable(spa->spa_root_vdev)) { |
2e528b49 | 521 | error = SET_ERROR(ENOTSUP); |
34dc7c2f BB |
522 | break; |
523 | } | |
524 | ||
525 | reset_bootfs = 1; | |
526 | ||
527 | error = nvpair_value_string(elem, &strval); | |
528 | ||
529 | if (!error) { | |
9ae529ec | 530 | objset_t *os; |
f1512ee6 | 531 | uint64_t propval; |
b128c09f | 532 | |
34dc7c2f BB |
533 | if (strval == NULL || strval[0] == '\0') { |
534 | objnum = zpool_prop_default_numeric( | |
535 | ZPOOL_PROP_BOOTFS); | |
536 | break; | |
537 | } | |
538 | ||
d1d7e268 MK |
539 | error = dmu_objset_hold(strval, FTAG, &os); |
540 | if (error) | |
34dc7c2f | 541 | break; |
b128c09f | 542 | |
f1512ee6 MA |
543 | /* |
544 | * Must be ZPL, and its property settings | |
545 | * must be supported by GRUB (compression | |
50c957f7 NB |
546 | * is not gzip, and large blocks or large |
547 | * dnodes are not used). | |
f1512ee6 | 548 | */ |
428870ff BB |
549 | |
550 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
2e528b49 | 551 | error = SET_ERROR(ENOTSUP); |
13fe0198 MA |
552 | } else if ((error = |
553 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
b128c09f | 554 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), |
f1512ee6 MA |
555 | &propval)) == 0 && |
556 | !BOOTFS_COMPRESS_VALID(propval)) { | |
557 | error = SET_ERROR(ENOTSUP); | |
50c957f7 NB |
558 | } else if ((error = |
559 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
560 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
561 | &propval)) == 0 && | |
562 | propval != ZFS_DNSIZE_LEGACY) { | |
563 | error = SET_ERROR(ENOTSUP); | |
b128c09f BB |
564 | } else { |
565 | objnum = dmu_objset_id(os); | |
566 | } | |
428870ff | 567 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
568 | } |
569 | break; | |
b128c09f | 570 | |
34dc7c2f BB |
571 | case ZPOOL_PROP_FAILUREMODE: |
572 | error = nvpair_value_uint64(elem, &intval); | |
3bfd95d5 | 573 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) |
2e528b49 | 574 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
575 | |
576 | /* | |
577 | * This is a special case which only occurs when | |
578 | * the pool has completely failed. This allows | |
579 | * the user to change the in-core failmode property | |
580 | * without syncing it out to disk (I/Os might | |
581 | * currently be blocked). We do this by returning | |
582 | * EIO to the caller (spa_prop_set) to trick it | |
583 | * into thinking we encountered a property validation | |
584 | * error. | |
585 | */ | |
b128c09f | 586 | if (!error && spa_suspended(spa)) { |
34dc7c2f | 587 | spa->spa_failmode = intval; |
2e528b49 | 588 | error = SET_ERROR(EIO); |
34dc7c2f BB |
589 | } |
590 | break; | |
591 | ||
592 | case ZPOOL_PROP_CACHEFILE: | |
593 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
594 | break; | |
595 | ||
596 | if (strval[0] == '\0') | |
597 | break; | |
598 | ||
599 | if (strcmp(strval, "none") == 0) | |
600 | break; | |
601 | ||
602 | if (strval[0] != '/') { | |
2e528b49 | 603 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
604 | break; |
605 | } | |
606 | ||
607 | slash = strrchr(strval, '/'); | |
608 | ASSERT(slash != NULL); | |
609 | ||
610 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
611 | strcmp(slash, "/..") == 0) | |
2e528b49 | 612 | error = SET_ERROR(EINVAL); |
34dc7c2f | 613 | break; |
428870ff | 614 | |
d96eb2b1 DM |
615 | case ZPOOL_PROP_COMMENT: |
616 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
617 | break; | |
618 | for (check = strval; *check != '\0'; check++) { | |
619 | if (!isprint(*check)) { | |
2e528b49 | 620 | error = SET_ERROR(EINVAL); |
d96eb2b1 DM |
621 | break; |
622 | } | |
d96eb2b1 DM |
623 | } |
624 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
2e528b49 | 625 | error = SET_ERROR(E2BIG); |
d96eb2b1 DM |
626 | break; |
627 | ||
428870ff BB |
628 | case ZPOOL_PROP_DEDUPDITTO: |
629 | if (spa_version(spa) < SPA_VERSION_DEDUP) | |
2e528b49 | 630 | error = SET_ERROR(ENOTSUP); |
428870ff BB |
631 | else |
632 | error = nvpair_value_uint64(elem, &intval); | |
633 | if (error == 0 && | |
634 | intval != 0 && intval < ZIO_DEDUPDITTO_MIN) | |
2e528b49 | 635 | error = SET_ERROR(EINVAL); |
428870ff | 636 | break; |
e75c13c3 BB |
637 | |
638 | default: | |
639 | break; | |
34dc7c2f BB |
640 | } |
641 | ||
642 | if (error) | |
643 | break; | |
644 | } | |
645 | ||
646 | if (!error && reset_bootfs) { | |
647 | error = nvlist_remove(props, | |
648 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
649 | ||
650 | if (!error) { | |
651 | error = nvlist_add_uint64(props, | |
652 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
653 | } | |
654 | } | |
655 | ||
656 | return (error); | |
657 | } | |
658 | ||
d164b209 BB |
659 | void |
660 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
661 | { | |
662 | char *cachefile; | |
663 | spa_config_dirent_t *dp; | |
664 | ||
665 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
666 | &cachefile) != 0) | |
667 | return; | |
668 | ||
669 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
79c76d5b | 670 | KM_SLEEP); |
d164b209 BB |
671 | |
672 | if (cachefile[0] == '\0') | |
673 | dp->scd_path = spa_strdup(spa_config_path); | |
674 | else if (strcmp(cachefile, "none") == 0) | |
675 | dp->scd_path = NULL; | |
676 | else | |
677 | dp->scd_path = spa_strdup(cachefile); | |
678 | ||
679 | list_insert_head(&spa->spa_config_list, dp); | |
680 | if (need_sync) | |
681 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
682 | } | |
683 | ||
34dc7c2f BB |
684 | int |
685 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
686 | { | |
687 | int error; | |
9ae529ec | 688 | nvpair_t *elem = NULL; |
d164b209 | 689 | boolean_t need_sync = B_FALSE; |
34dc7c2f BB |
690 | |
691 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
692 | return (error); | |
693 | ||
d164b209 | 694 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { |
9ae529ec | 695 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); |
d164b209 | 696 | |
572e2857 BB |
697 | if (prop == ZPOOL_PROP_CACHEFILE || |
698 | prop == ZPOOL_PROP_ALTROOT || | |
699 | prop == ZPOOL_PROP_READONLY) | |
d164b209 BB |
700 | continue; |
701 | ||
31864e3d | 702 | if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { |
9ae529ec CS |
703 | uint64_t ver; |
704 | ||
705 | if (prop == ZPOOL_PROP_VERSION) { | |
706 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
707 | } else { | |
708 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
709 | ver = SPA_VERSION_FEATURES; | |
710 | need_sync = B_TRUE; | |
711 | } | |
712 | ||
713 | /* Save time if the version is already set. */ | |
714 | if (ver == spa_version(spa)) | |
715 | continue; | |
716 | ||
717 | /* | |
718 | * In addition to the pool directory object, we might | |
719 | * create the pool properties object, the features for | |
720 | * read object, the features for write object, or the | |
721 | * feature descriptions object. | |
722 | */ | |
13fe0198 | 723 | error = dsl_sync_task(spa->spa_name, NULL, |
3d45fdd6 MA |
724 | spa_sync_version, &ver, |
725 | 6, ZFS_SPACE_CHECK_RESERVED); | |
9ae529ec CS |
726 | if (error) |
727 | return (error); | |
728 | continue; | |
729 | } | |
730 | ||
d164b209 BB |
731 | need_sync = B_TRUE; |
732 | break; | |
733 | } | |
734 | ||
9ae529ec | 735 | if (need_sync) { |
13fe0198 | 736 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, |
3d45fdd6 | 737 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); |
9ae529ec CS |
738 | } |
739 | ||
740 | return (0); | |
34dc7c2f BB |
741 | } |
742 | ||
743 | /* | |
744 | * If the bootfs property value is dsobj, clear it. | |
745 | */ | |
746 | void | |
747 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
748 | { | |
749 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
750 | VERIFY(zap_remove(spa->spa_meta_objset, | |
751 | spa->spa_pool_props_object, | |
752 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
753 | spa->spa_bootfs = 0; | |
754 | } | |
755 | } | |
756 | ||
3bc7e0fb GW |
757 | /*ARGSUSED*/ |
758 | static int | |
13fe0198 | 759 | spa_change_guid_check(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 760 | { |
1c27024e | 761 | ASSERTV(uint64_t *newguid = arg); |
13fe0198 | 762 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; |
3bc7e0fb GW |
763 | vdev_t *rvd = spa->spa_root_vdev; |
764 | uint64_t vdev_state; | |
3bc7e0fb GW |
765 | |
766 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
767 | vdev_state = rvd->vdev_state; | |
768 | spa_config_exit(spa, SCL_STATE, FTAG); | |
769 | ||
770 | if (vdev_state != VDEV_STATE_HEALTHY) | |
2e528b49 | 771 | return (SET_ERROR(ENXIO)); |
3bc7e0fb GW |
772 | |
773 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
774 | ||
775 | return (0); | |
776 | } | |
777 | ||
778 | static void | |
13fe0198 | 779 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) |
3bc7e0fb | 780 | { |
13fe0198 MA |
781 | uint64_t *newguid = arg; |
782 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
3bc7e0fb GW |
783 | uint64_t oldguid; |
784 | vdev_t *rvd = spa->spa_root_vdev; | |
785 | ||
786 | oldguid = spa_guid(spa); | |
787 | ||
788 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
789 | rvd->vdev_guid = *newguid; | |
790 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
791 | vdev_config_dirty(rvd); | |
792 | spa_config_exit(spa, SCL_STATE, FTAG); | |
793 | ||
6f1ffb06 MA |
794 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", |
795 | oldguid, *newguid); | |
3bc7e0fb GW |
796 | } |
797 | ||
3541dc6d GA |
798 | /* |
799 | * Change the GUID for the pool. This is done so that we can later | |
800 | * re-import a pool built from a clone of our own vdevs. We will modify | |
801 | * the root vdev's guid, our own pool guid, and then mark all of our | |
802 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
803 | * online when we do this, or else any vdevs that weren't present | |
804 | * would be orphaned from our pool. We are also going to issue a | |
805 | * sysevent to update any watchers. | |
806 | */ | |
807 | int | |
808 | spa_change_guid(spa_t *spa) | |
809 | { | |
3bc7e0fb GW |
810 | int error; |
811 | uint64_t guid; | |
3541dc6d | 812 | |
621dd7bb | 813 | mutex_enter(&spa->spa_vdev_top_lock); |
3bc7e0fb GW |
814 | mutex_enter(&spa_namespace_lock); |
815 | guid = spa_generate_guid(NULL); | |
3541dc6d | 816 | |
13fe0198 | 817 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, |
3d45fdd6 | 818 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); |
3541dc6d | 819 | |
3bc7e0fb | 820 | if (error == 0) { |
a1d477c2 | 821 | spa_write_cachefile(spa, B_FALSE, B_TRUE); |
12fa0466 | 822 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); |
3bc7e0fb | 823 | } |
3541dc6d | 824 | |
3bc7e0fb | 825 | mutex_exit(&spa_namespace_lock); |
621dd7bb | 826 | mutex_exit(&spa->spa_vdev_top_lock); |
3541dc6d | 827 | |
3bc7e0fb | 828 | return (error); |
3541dc6d GA |
829 | } |
830 | ||
34dc7c2f BB |
831 | /* |
832 | * ========================================================================== | |
833 | * SPA state manipulation (open/create/destroy/import/export) | |
834 | * ========================================================================== | |
835 | */ | |
836 | ||
837 | static int | |
838 | spa_error_entry_compare(const void *a, const void *b) | |
839 | { | |
ee36c709 GN |
840 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; |
841 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
34dc7c2f BB |
842 | int ret; |
843 | ||
ee36c709 | 844 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, |
5dbd68a3 | 845 | sizeof (zbookmark_phys_t)); |
34dc7c2f | 846 | |
ee36c709 | 847 | return (AVL_ISIGN(ret)); |
34dc7c2f BB |
848 | } |
849 | ||
850 | /* | |
851 | * Utility function which retrieves copies of the current logs and | |
852 | * re-initializes them in the process. | |
853 | */ | |
854 | void | |
855 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
856 | { | |
857 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
858 | ||
859 | bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
860 | bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
861 | ||
862 | avl_create(&spa->spa_errlist_scrub, | |
863 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
864 | offsetof(spa_error_entry_t, se_avl)); | |
865 | avl_create(&spa->spa_errlist_last, | |
866 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
867 | offsetof(spa_error_entry_t, se_avl)); | |
868 | } | |
869 | ||
7ef5e54e AL |
870 | static void |
871 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
34dc7c2f | 872 | { |
7ef5e54e AL |
873 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; |
874 | enum zti_modes mode = ztip->zti_mode; | |
875 | uint_t value = ztip->zti_value; | |
876 | uint_t count = ztip->zti_count; | |
877 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
878 | char name[32]; | |
1c27024e | 879 | uint_t flags = 0; |
428870ff | 880 | boolean_t batch = B_FALSE; |
34dc7c2f | 881 | |
7ef5e54e AL |
882 | if (mode == ZTI_MODE_NULL) { |
883 | tqs->stqs_count = 0; | |
884 | tqs->stqs_taskq = NULL; | |
885 | return; | |
886 | } | |
428870ff | 887 | |
7ef5e54e | 888 | ASSERT3U(count, >, 0); |
428870ff | 889 | |
7ef5e54e AL |
890 | tqs->stqs_count = count; |
891 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
428870ff | 892 | |
e8b96c60 MA |
893 | switch (mode) { |
894 | case ZTI_MODE_FIXED: | |
895 | ASSERT3U(value, >=, 1); | |
896 | value = MAX(value, 1); | |
d33931a8 | 897 | flags |= TASKQ_DYNAMIC; |
e8b96c60 | 898 | break; |
7ef5e54e | 899 | |
e8b96c60 MA |
900 | case ZTI_MODE_BATCH: |
901 | batch = B_TRUE; | |
902 | flags |= TASKQ_THREADS_CPU_PCT; | |
dcb6bed1 | 903 | value = MIN(zio_taskq_batch_pct, 100); |
e8b96c60 | 904 | break; |
7ef5e54e | 905 | |
e8b96c60 MA |
906 | default: |
907 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
908 | "spa_activate()", | |
909 | zio_type_name[t], zio_taskq_types[q], mode, value); | |
910 | break; | |
911 | } | |
7ef5e54e | 912 | |
1c27024e | 913 | for (uint_t i = 0; i < count; i++) { |
e8b96c60 | 914 | taskq_t *tq; |
7ef5e54e AL |
915 | |
916 | if (count > 1) { | |
917 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
918 | zio_type_name[t], zio_taskq_types[q], i); | |
919 | } else { | |
920 | (void) snprintf(name, sizeof (name), "%s_%s", | |
921 | zio_type_name[t], zio_taskq_types[q]); | |
922 | } | |
923 | ||
924 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { | |
925 | if (batch) | |
926 | flags |= TASKQ_DC_BATCH; | |
927 | ||
928 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, | |
929 | spa->spa_proc, zio_taskq_basedc, flags); | |
930 | } else { | |
e8b96c60 MA |
931 | pri_t pri = maxclsyspri; |
932 | /* | |
933 | * The write issue taskq can be extremely CPU | |
1229323d BB |
934 | * intensive. Run it at slightly less important |
935 | * priority than the other taskqs. Under Linux this | |
936 | * means incrementing the priority value on platforms | |
937 | * like illumos it should be decremented. | |
e8b96c60 MA |
938 | */ |
939 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) | |
1229323d | 940 | pri++; |
e8b96c60 MA |
941 | |
942 | tq = taskq_create_proc(name, value, pri, 50, | |
7ef5e54e AL |
943 | INT_MAX, spa->spa_proc, flags); |
944 | } | |
945 | ||
946 | tqs->stqs_taskq[i] = tq; | |
947 | } | |
948 | } | |
949 | ||
950 | static void | |
951 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
952 | { | |
953 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
7ef5e54e AL |
954 | |
955 | if (tqs->stqs_taskq == NULL) { | |
956 | ASSERT3U(tqs->stqs_count, ==, 0); | |
957 | return; | |
958 | } | |
959 | ||
1c27024e | 960 | for (uint_t i = 0; i < tqs->stqs_count; i++) { |
7ef5e54e AL |
961 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); |
962 | taskq_destroy(tqs->stqs_taskq[i]); | |
428870ff | 963 | } |
34dc7c2f | 964 | |
7ef5e54e AL |
965 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); |
966 | tqs->stqs_taskq = NULL; | |
967 | } | |
34dc7c2f | 968 | |
7ef5e54e AL |
969 | /* |
970 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
971 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
972 | * on the taskq itself. In that case we choose which taskq at random by using | |
973 | * the low bits of gethrtime(). | |
974 | */ | |
975 | void | |
976 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
977 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) | |
978 | { | |
979 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
980 | taskq_t *tq; | |
981 | ||
982 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
983 | ASSERT3U(tqs->stqs_count, !=, 0); | |
984 | ||
985 | if (tqs->stqs_count == 1) { | |
986 | tq = tqs->stqs_taskq[0]; | |
987 | } else { | |
c12936b1 | 988 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
428870ff | 989 | } |
7ef5e54e AL |
990 | |
991 | taskq_dispatch_ent(tq, func, arg, flags, ent); | |
428870ff BB |
992 | } |
993 | ||
044baf00 BB |
994 | /* |
995 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
996 | */ | |
997 | void | |
998 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
999 | task_func_t *func, void *arg, uint_t flags) | |
1000 | { | |
1001 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1002 | taskq_t *tq; | |
1003 | taskqid_t id; | |
1004 | ||
1005 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1006 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1007 | ||
1008 | if (tqs->stqs_count == 1) { | |
1009 | tq = tqs->stqs_taskq[0]; | |
1010 | } else { | |
c12936b1 | 1011 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; |
044baf00 BB |
1012 | } |
1013 | ||
1014 | id = taskq_dispatch(tq, func, arg, flags); | |
1015 | if (id) | |
1016 | taskq_wait_id(tq, id); | |
1017 | } | |
1018 | ||
428870ff BB |
1019 | static void |
1020 | spa_create_zio_taskqs(spa_t *spa) | |
1021 | { | |
1c27024e DB |
1022 | for (int t = 0; t < ZIO_TYPES; t++) { |
1023 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1024 | spa_taskqs_init(spa, t, q); |
428870ff BB |
1025 | } |
1026 | } | |
1027 | } | |
9babb374 | 1028 | |
c25b8f99 BB |
1029 | /* |
1030 | * Disabled until spa_thread() can be adapted for Linux. | |
1031 | */ | |
1032 | #undef HAVE_SPA_THREAD | |
1033 | ||
7b89a549 | 1034 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) |
428870ff BB |
1035 | static void |
1036 | spa_thread(void *arg) | |
1037 | { | |
1038 | callb_cpr_t cprinfo; | |
9babb374 | 1039 | |
428870ff BB |
1040 | spa_t *spa = arg; |
1041 | user_t *pu = PTOU(curproc); | |
9babb374 | 1042 | |
428870ff BB |
1043 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, |
1044 | spa->spa_name); | |
9babb374 | 1045 | |
428870ff BB |
1046 | ASSERT(curproc != &p0); |
1047 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1048 | "zpool-%s", spa->spa_name); | |
1049 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1050 | ||
1051 | /* bind this thread to the requested psrset */ | |
1052 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1053 | pool_lock(); | |
1054 | mutex_enter(&cpu_lock); | |
1055 | mutex_enter(&pidlock); | |
1056 | mutex_enter(&curproc->p_lock); | |
1057 | ||
1058 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1059 | 0, NULL, NULL) == 0) { | |
1060 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1061 | } else { | |
1062 | cmn_err(CE_WARN, | |
1063 | "Couldn't bind process for zfs pool \"%s\" to " | |
1064 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1065 | } | |
1066 | ||
1067 | mutex_exit(&curproc->p_lock); | |
1068 | mutex_exit(&pidlock); | |
1069 | mutex_exit(&cpu_lock); | |
1070 | pool_unlock(); | |
1071 | } | |
1072 | ||
1073 | if (zio_taskq_sysdc) { | |
1074 | sysdc_thread_enter(curthread, 100, 0); | |
1075 | } | |
1076 | ||
1077 | spa->spa_proc = curproc; | |
1078 | spa->spa_did = curthread->t_did; | |
1079 | ||
1080 | spa_create_zio_taskqs(spa); | |
1081 | ||
1082 | mutex_enter(&spa->spa_proc_lock); | |
1083 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1084 | ||
1085 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1086 | cv_broadcast(&spa->spa_proc_cv); | |
1087 | ||
1088 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1089 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1090 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1091 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1092 | ||
1093 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1094 | spa->spa_proc_state = SPA_PROC_GONE; | |
1095 | spa->spa_proc = &p0; | |
1096 | cv_broadcast(&spa->spa_proc_cv); | |
1097 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1098 | ||
1099 | mutex_enter(&curproc->p_lock); | |
1100 | lwp_exit(); | |
1101 | } | |
1102 | #endif | |
1103 | ||
1104 | /* | |
1105 | * Activate an uninitialized pool. | |
1106 | */ | |
1107 | static void | |
1108 | spa_activate(spa_t *spa, int mode) | |
1109 | { | |
1110 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
1111 | ||
1112 | spa->spa_state = POOL_STATE_ACTIVE; | |
1113 | spa->spa_mode = mode; | |
1114 | ||
1115 | spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1116 | spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1117 | ||
1118 | /* Try to create a covering process */ | |
1119 | mutex_enter(&spa->spa_proc_lock); | |
1120 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1121 | ASSERT(spa->spa_proc == &p0); | |
1122 | spa->spa_did = 0; | |
1123 | ||
7b89a549 | 1124 | #ifdef HAVE_SPA_THREAD |
428870ff BB |
1125 | /* Only create a process if we're going to be around a while. */ |
1126 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1127 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1128 | NULL, 0) == 0) { | |
1129 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1130 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1131 | cv_wait(&spa->spa_proc_cv, | |
1132 | &spa->spa_proc_lock); | |
9babb374 | 1133 | } |
428870ff BB |
1134 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); |
1135 | ASSERT(spa->spa_proc != &p0); | |
1136 | ASSERT(spa->spa_did != 0); | |
1137 | } else { | |
1138 | #ifdef _KERNEL | |
1139 | cmn_err(CE_WARN, | |
1140 | "Couldn't create process for zfs pool \"%s\"\n", | |
1141 | spa->spa_name); | |
1142 | #endif | |
b128c09f | 1143 | } |
34dc7c2f | 1144 | } |
7b89a549 | 1145 | #endif /* HAVE_SPA_THREAD */ |
428870ff BB |
1146 | mutex_exit(&spa->spa_proc_lock); |
1147 | ||
1148 | /* If we didn't create a process, we need to create our taskqs. */ | |
1149 | if (spa->spa_proc == &p0) { | |
1150 | spa_create_zio_taskqs(spa); | |
1151 | } | |
34dc7c2f | 1152 | |
a1d477c2 MA |
1153 | for (size_t i = 0; i < TXG_SIZE; i++) |
1154 | spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0); | |
1155 | ||
b128c09f BB |
1156 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), |
1157 | offsetof(vdev_t, vdev_config_dirty_node)); | |
0c66c32d JG |
1158 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), |
1159 | offsetof(objset_t, os_evicting_node)); | |
b128c09f BB |
1160 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), |
1161 | offsetof(vdev_t, vdev_state_dirty_node)); | |
34dc7c2f | 1162 | |
4747a7d3 | 1163 | txg_list_create(&spa->spa_vdev_txg_list, spa, |
34dc7c2f BB |
1164 | offsetof(struct vdev, vdev_txg_node)); |
1165 | ||
1166 | avl_create(&spa->spa_errlist_scrub, | |
1167 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1168 | offsetof(spa_error_entry_t, se_avl)); | |
1169 | avl_create(&spa->spa_errlist_last, | |
1170 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1171 | offsetof(spa_error_entry_t, se_avl)); | |
a0bd735a | 1172 | |
b5256303 TC |
1173 | spa_keystore_init(&spa->spa_keystore); |
1174 | ||
a0bd735a BP |
1175 | /* |
1176 | * This taskq is used to perform zvol-minor-related tasks | |
1177 | * asynchronously. This has several advantages, including easy | |
1178 | * resolution of various deadlocks (zfsonlinux bug #3681). | |
1179 | * | |
1180 | * The taskq must be single threaded to ensure tasks are always | |
1181 | * processed in the order in which they were dispatched. | |
1182 | * | |
1183 | * A taskq per pool allows one to keep the pools independent. | |
1184 | * This way if one pool is suspended, it will not impact another. | |
1185 | * | |
1186 | * The preferred location to dispatch a zvol minor task is a sync | |
1187 | * task. In this context, there is easy access to the spa_t and minimal | |
1188 | * error handling is required because the sync task must succeed. | |
1189 | */ | |
1190 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1191 | 1, INT_MAX, 0); | |
1de321e6 | 1192 | |
77d8a0f1 | 1193 | /* |
1194 | * Taskq dedicated to prefetcher threads: this is used to prevent the | |
1195 | * pool traverse code from monopolizing the global (and limited) | |
1196 | * system_taskq by inappropriately scheduling long running tasks on it. | |
1197 | */ | |
1198 | spa->spa_prefetch_taskq = taskq_create("z_prefetch", boot_ncpus, | |
1199 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC); | |
1200 | ||
1de321e6 JX |
1201 | /* |
1202 | * The taskq to upgrade datasets in this pool. Currently used by | |
9c5167d1 | 1203 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA. |
1de321e6 JX |
1204 | */ |
1205 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus, | |
1206 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC); | |
34dc7c2f BB |
1207 | } |
1208 | ||
1209 | /* | |
1210 | * Opposite of spa_activate(). | |
1211 | */ | |
1212 | static void | |
1213 | spa_deactivate(spa_t *spa) | |
1214 | { | |
34dc7c2f BB |
1215 | ASSERT(spa->spa_sync_on == B_FALSE); |
1216 | ASSERT(spa->spa_dsl_pool == NULL); | |
1217 | ASSERT(spa->spa_root_vdev == NULL); | |
9babb374 | 1218 | ASSERT(spa->spa_async_zio_root == NULL); |
34dc7c2f BB |
1219 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); |
1220 | ||
0c66c32d JG |
1221 | spa_evicting_os_wait(spa); |
1222 | ||
a0bd735a BP |
1223 | if (spa->spa_zvol_taskq) { |
1224 | taskq_destroy(spa->spa_zvol_taskq); | |
1225 | spa->spa_zvol_taskq = NULL; | |
1226 | } | |
1227 | ||
77d8a0f1 | 1228 | if (spa->spa_prefetch_taskq) { |
1229 | taskq_destroy(spa->spa_prefetch_taskq); | |
1230 | spa->spa_prefetch_taskq = NULL; | |
1231 | } | |
1232 | ||
1de321e6 JX |
1233 | if (spa->spa_upgrade_taskq) { |
1234 | taskq_destroy(spa->spa_upgrade_taskq); | |
1235 | spa->spa_upgrade_taskq = NULL; | |
1236 | } | |
1237 | ||
34dc7c2f BB |
1238 | txg_list_destroy(&spa->spa_vdev_txg_list); |
1239 | ||
b128c09f | 1240 | list_destroy(&spa->spa_config_dirty_list); |
0c66c32d | 1241 | list_destroy(&spa->spa_evicting_os_list); |
b128c09f | 1242 | list_destroy(&spa->spa_state_dirty_list); |
34dc7c2f | 1243 | |
57ddcda1 | 1244 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 | 1245 | |
1c27024e DB |
1246 | for (int t = 0; t < ZIO_TYPES; t++) { |
1247 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
7ef5e54e | 1248 | spa_taskqs_fini(spa, t, q); |
b128c09f | 1249 | } |
34dc7c2f BB |
1250 | } |
1251 | ||
a1d477c2 MA |
1252 | for (size_t i = 0; i < TXG_SIZE; i++) { |
1253 | ASSERT3P(spa->spa_txg_zio[i], !=, NULL); | |
1254 | VERIFY0(zio_wait(spa->spa_txg_zio[i])); | |
1255 | spa->spa_txg_zio[i] = NULL; | |
1256 | } | |
1257 | ||
34dc7c2f BB |
1258 | metaslab_class_destroy(spa->spa_normal_class); |
1259 | spa->spa_normal_class = NULL; | |
1260 | ||
1261 | metaslab_class_destroy(spa->spa_log_class); | |
1262 | spa->spa_log_class = NULL; | |
1263 | ||
1264 | /* | |
1265 | * If this was part of an import or the open otherwise failed, we may | |
1266 | * still have errors left in the queues. Empty them just in case. | |
1267 | */ | |
1268 | spa_errlog_drain(spa); | |
34dc7c2f BB |
1269 | avl_destroy(&spa->spa_errlist_scrub); |
1270 | avl_destroy(&spa->spa_errlist_last); | |
1271 | ||
b5256303 TC |
1272 | spa_keystore_fini(&spa->spa_keystore); |
1273 | ||
34dc7c2f | 1274 | spa->spa_state = POOL_STATE_UNINITIALIZED; |
428870ff BB |
1275 | |
1276 | mutex_enter(&spa->spa_proc_lock); | |
1277 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1278 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1279 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1280 | cv_broadcast(&spa->spa_proc_cv); | |
1281 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1282 | ASSERT(spa->spa_proc != &p0); | |
1283 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1284 | } | |
1285 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1286 | spa->spa_proc_state = SPA_PROC_NONE; | |
1287 | } | |
1288 | ASSERT(spa->spa_proc == &p0); | |
1289 | mutex_exit(&spa->spa_proc_lock); | |
1290 | ||
1291 | /* | |
1292 | * We want to make sure spa_thread() has actually exited the ZFS | |
1293 | * module, so that the module can't be unloaded out from underneath | |
1294 | * it. | |
1295 | */ | |
1296 | if (spa->spa_did != 0) { | |
1297 | thread_join(spa->spa_did); | |
1298 | spa->spa_did = 0; | |
1299 | } | |
34dc7c2f BB |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1304 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1305 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1306 | * All vdev validation is done by the vdev_alloc() routine. | |
1307 | */ | |
1308 | static int | |
1309 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
1310 | uint_t id, int atype) | |
1311 | { | |
1312 | nvlist_t **child; | |
9babb374 | 1313 | uint_t children; |
34dc7c2f BB |
1314 | int error; |
1315 | ||
1316 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1317 | return (error); | |
1318 | ||
1319 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1320 | return (0); | |
1321 | ||
b128c09f BB |
1322 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, |
1323 | &child, &children); | |
1324 | ||
1325 | if (error == ENOENT) | |
1326 | return (0); | |
1327 | ||
1328 | if (error) { | |
34dc7c2f BB |
1329 | vdev_free(*vdp); |
1330 | *vdp = NULL; | |
2e528b49 | 1331 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
1332 | } |
1333 | ||
1c27024e | 1334 | for (int c = 0; c < children; c++) { |
34dc7c2f BB |
1335 | vdev_t *vd; |
1336 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1337 | atype)) != 0) { | |
1338 | vdev_free(*vdp); | |
1339 | *vdp = NULL; | |
1340 | return (error); | |
1341 | } | |
1342 | } | |
1343 | ||
1344 | ASSERT(*vdp != NULL); | |
1345 | ||
1346 | return (0); | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Opposite of spa_load(). | |
1351 | */ | |
1352 | static void | |
1353 | spa_unload(spa_t *spa) | |
1354 | { | |
1c27024e | 1355 | int i; |
34dc7c2f | 1356 | |
b128c09f BB |
1357 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); |
1358 | ||
34dc7c2f BB |
1359 | /* |
1360 | * Stop async tasks. | |
1361 | */ | |
1362 | spa_async_suspend(spa); | |
1363 | ||
1364 | /* | |
1365 | * Stop syncing. | |
1366 | */ | |
1367 | if (spa->spa_sync_on) { | |
1368 | txg_sync_stop(spa->spa_dsl_pool); | |
1369 | spa->spa_sync_on = B_FALSE; | |
1370 | } | |
1371 | ||
4e21fd06 DB |
1372 | /* |
1373 | * Even though vdev_free() also calls vdev_metaslab_fini, we need | |
1374 | * to call it earlier, before we wait for async i/o to complete. | |
1375 | * This ensures that there is no async metaslab prefetching, by | |
1376 | * calling taskq_wait(mg_taskq). | |
1377 | */ | |
1378 | if (spa->spa_root_vdev != NULL) { | |
1379 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1c27024e | 1380 | for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) |
4e21fd06 DB |
1381 | vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); |
1382 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1383 | } | |
1384 | ||
379ca9cf OF |
1385 | if (spa->spa_mmp.mmp_thread) |
1386 | mmp_thread_stop(spa); | |
1387 | ||
34dc7c2f | 1388 | /* |
b128c09f | 1389 | * Wait for any outstanding async I/O to complete. |
34dc7c2f | 1390 | */ |
9babb374 | 1391 | if (spa->spa_async_zio_root != NULL) { |
1c27024e | 1392 | for (int i = 0; i < max_ncpus; i++) |
e022864d MA |
1393 | (void) zio_wait(spa->spa_async_zio_root[i]); |
1394 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
9babb374 BB |
1395 | spa->spa_async_zio_root = NULL; |
1396 | } | |
34dc7c2f | 1397 | |
a1d477c2 MA |
1398 | if (spa->spa_vdev_removal != NULL) { |
1399 | spa_vdev_removal_destroy(spa->spa_vdev_removal); | |
1400 | spa->spa_vdev_removal = NULL; | |
1401 | } | |
1402 | ||
1403 | spa_condense_fini(spa); | |
1404 | ||
428870ff BB |
1405 | bpobj_close(&spa->spa_deferred_bpobj); |
1406 | ||
93cf2076 GW |
1407 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
1408 | ||
1409 | /* | |
1410 | * Close all vdevs. | |
1411 | */ | |
1412 | if (spa->spa_root_vdev) | |
1413 | vdev_free(spa->spa_root_vdev); | |
1414 | ASSERT(spa->spa_root_vdev == NULL); | |
1415 | ||
34dc7c2f BB |
1416 | /* |
1417 | * Close the dsl pool. | |
1418 | */ | |
1419 | if (spa->spa_dsl_pool) { | |
1420 | dsl_pool_close(spa->spa_dsl_pool); | |
1421 | spa->spa_dsl_pool = NULL; | |
428870ff | 1422 | spa->spa_meta_objset = NULL; |
34dc7c2f BB |
1423 | } |
1424 | ||
428870ff BB |
1425 | ddt_unload(spa); |
1426 | ||
fb5f0bc8 BB |
1427 | /* |
1428 | * Drop and purge level 2 cache | |
1429 | */ | |
1430 | spa_l2cache_drop(spa); | |
1431 | ||
34dc7c2f BB |
1432 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1433 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
1434 | if (spa->spa_spares.sav_vdevs) { | |
1435 | kmem_free(spa->spa_spares.sav_vdevs, | |
1436 | spa->spa_spares.sav_count * sizeof (void *)); | |
1437 | spa->spa_spares.sav_vdevs = NULL; | |
1438 | } | |
1439 | if (spa->spa_spares.sav_config) { | |
1440 | nvlist_free(spa->spa_spares.sav_config); | |
1441 | spa->spa_spares.sav_config = NULL; | |
1442 | } | |
b128c09f | 1443 | spa->spa_spares.sav_count = 0; |
34dc7c2f | 1444 | |
5ffb9d1d GW |
1445 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { |
1446 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
34dc7c2f | 1447 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); |
5ffb9d1d | 1448 | } |
34dc7c2f BB |
1449 | if (spa->spa_l2cache.sav_vdevs) { |
1450 | kmem_free(spa->spa_l2cache.sav_vdevs, | |
1451 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
1452 | spa->spa_l2cache.sav_vdevs = NULL; | |
1453 | } | |
1454 | if (spa->spa_l2cache.sav_config) { | |
1455 | nvlist_free(spa->spa_l2cache.sav_config); | |
1456 | spa->spa_l2cache.sav_config = NULL; | |
1457 | } | |
b128c09f | 1458 | spa->spa_l2cache.sav_count = 0; |
34dc7c2f BB |
1459 | |
1460 | spa->spa_async_suspended = 0; | |
fb5f0bc8 | 1461 | |
a1d477c2 MA |
1462 | spa->spa_indirect_vdevs_loaded = B_FALSE; |
1463 | ||
d96eb2b1 DM |
1464 | if (spa->spa_comment != NULL) { |
1465 | spa_strfree(spa->spa_comment); | |
1466 | spa->spa_comment = NULL; | |
1467 | } | |
1468 | ||
fb5f0bc8 | 1469 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
1470 | } |
1471 | ||
1472 | /* | |
1473 | * Load (or re-load) the current list of vdevs describing the active spares for | |
1474 | * this pool. When this is called, we have some form of basic information in | |
1475 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
1476 | * then re-generate a more complete list including status information. | |
1477 | */ | |
a1d477c2 | 1478 | void |
34dc7c2f BB |
1479 | spa_load_spares(spa_t *spa) |
1480 | { | |
1481 | nvlist_t **spares; | |
1482 | uint_t nspares; | |
1483 | int i; | |
1484 | vdev_t *vd, *tvd; | |
1485 | ||
b128c09f BB |
1486 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
1487 | ||
34dc7c2f BB |
1488 | /* |
1489 | * First, close and free any existing spare vdevs. | |
1490 | */ | |
1491 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1492 | vd = spa->spa_spares.sav_vdevs[i]; | |
1493 | ||
1494 | /* Undo the call to spa_activate() below */ | |
b128c09f BB |
1495 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, |
1496 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
34dc7c2f BB |
1497 | spa_spare_remove(tvd); |
1498 | vdev_close(vd); | |
1499 | vdev_free(vd); | |
1500 | } | |
1501 | ||
1502 | if (spa->spa_spares.sav_vdevs) | |
1503 | kmem_free(spa->spa_spares.sav_vdevs, | |
1504 | spa->spa_spares.sav_count * sizeof (void *)); | |
1505 | ||
1506 | if (spa->spa_spares.sav_config == NULL) | |
1507 | nspares = 0; | |
1508 | else | |
1509 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
1510 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
1511 | ||
1512 | spa->spa_spares.sav_count = (int)nspares; | |
1513 | spa->spa_spares.sav_vdevs = NULL; | |
1514 | ||
1515 | if (nspares == 0) | |
1516 | return; | |
1517 | ||
1518 | /* | |
1519 | * Construct the array of vdevs, opening them to get status in the | |
1520 | * process. For each spare, there is potentially two different vdev_t | |
1521 | * structures associated with it: one in the list of spares (used only | |
1522 | * for basic validation purposes) and one in the active vdev | |
1523 | * configuration (if it's spared in). During this phase we open and | |
1524 | * validate each vdev on the spare list. If the vdev also exists in the | |
1525 | * active configuration, then we also mark this vdev as an active spare. | |
1526 | */ | |
904ea276 | 1527 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), |
79c76d5b | 1528 | KM_SLEEP); |
34dc7c2f BB |
1529 | for (i = 0; i < spa->spa_spares.sav_count; i++) { |
1530 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
1531 | VDEV_ALLOC_SPARE) == 0); | |
1532 | ASSERT(vd != NULL); | |
1533 | ||
1534 | spa->spa_spares.sav_vdevs[i] = vd; | |
1535 | ||
b128c09f BB |
1536 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, |
1537 | B_FALSE)) != NULL) { | |
34dc7c2f BB |
1538 | if (!tvd->vdev_isspare) |
1539 | spa_spare_add(tvd); | |
1540 | ||
1541 | /* | |
1542 | * We only mark the spare active if we were successfully | |
1543 | * able to load the vdev. Otherwise, importing a pool | |
1544 | * with a bad active spare would result in strange | |
1545 | * behavior, because multiple pool would think the spare | |
1546 | * is actively in use. | |
1547 | * | |
1548 | * There is a vulnerability here to an equally bizarre | |
1549 | * circumstance, where a dead active spare is later | |
1550 | * brought back to life (onlined or otherwise). Given | |
1551 | * the rarity of this scenario, and the extra complexity | |
1552 | * it adds, we ignore the possibility. | |
1553 | */ | |
1554 | if (!vdev_is_dead(tvd)) | |
1555 | spa_spare_activate(tvd); | |
1556 | } | |
1557 | ||
b128c09f | 1558 | vd->vdev_top = vd; |
9babb374 | 1559 | vd->vdev_aux = &spa->spa_spares; |
b128c09f | 1560 | |
34dc7c2f BB |
1561 | if (vdev_open(vd) != 0) |
1562 | continue; | |
1563 | ||
34dc7c2f BB |
1564 | if (vdev_validate_aux(vd) == 0) |
1565 | spa_spare_add(vd); | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * Recompute the stashed list of spares, with status information | |
1570 | * this time. | |
1571 | */ | |
1572 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, | |
1573 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1574 | ||
1575 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
79c76d5b | 1576 | KM_SLEEP); |
34dc7c2f BB |
1577 | for (i = 0; i < spa->spa_spares.sav_count; i++) |
1578 | spares[i] = vdev_config_generate(spa, | |
428870ff | 1579 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); |
34dc7c2f BB |
1580 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, |
1581 | ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); | |
1582 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1583 | nvlist_free(spares[i]); | |
1584 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
1589 | * this pool. When this is called, we have some form of basic information in | |
1590 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
1591 | * then re-generate a more complete list including status information. | |
1592 | * Devices which are already active have their details maintained, and are | |
1593 | * not re-opened. | |
1594 | */ | |
a1d477c2 | 1595 | void |
34dc7c2f BB |
1596 | spa_load_l2cache(spa_t *spa) |
1597 | { | |
460f239e | 1598 | nvlist_t **l2cache = NULL; |
34dc7c2f BB |
1599 | uint_t nl2cache; |
1600 | int i, j, oldnvdevs; | |
9babb374 | 1601 | uint64_t guid; |
a117a6d6 | 1602 | vdev_t *vd, **oldvdevs, **newvdevs; |
34dc7c2f BB |
1603 | spa_aux_vdev_t *sav = &spa->spa_l2cache; |
1604 | ||
b128c09f BB |
1605 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
1606 | ||
34dc7c2f BB |
1607 | oldvdevs = sav->sav_vdevs; |
1608 | oldnvdevs = sav->sav_count; | |
1609 | sav->sav_vdevs = NULL; | |
1610 | sav->sav_count = 0; | |
1611 | ||
67d60824 NB |
1612 | if (sav->sav_config == NULL) { |
1613 | nl2cache = 0; | |
1614 | newvdevs = NULL; | |
1615 | goto out; | |
1616 | } | |
1617 | ||
1618 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
1619 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
1620 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); | |
1621 | ||
34dc7c2f BB |
1622 | /* |
1623 | * Process new nvlist of vdevs. | |
1624 | */ | |
1625 | for (i = 0; i < nl2cache; i++) { | |
1626 | VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, | |
1627 | &guid) == 0); | |
1628 | ||
1629 | newvdevs[i] = NULL; | |
1630 | for (j = 0; j < oldnvdevs; j++) { | |
1631 | vd = oldvdevs[j]; | |
1632 | if (vd != NULL && guid == vd->vdev_guid) { | |
1633 | /* | |
1634 | * Retain previous vdev for add/remove ops. | |
1635 | */ | |
1636 | newvdevs[i] = vd; | |
1637 | oldvdevs[j] = NULL; | |
1638 | break; | |
1639 | } | |
1640 | } | |
1641 | ||
1642 | if (newvdevs[i] == NULL) { | |
1643 | /* | |
1644 | * Create new vdev | |
1645 | */ | |
1646 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
1647 | VDEV_ALLOC_L2CACHE) == 0); | |
1648 | ASSERT(vd != NULL); | |
1649 | newvdevs[i] = vd; | |
1650 | ||
1651 | /* | |
1652 | * Commit this vdev as an l2cache device, | |
1653 | * even if it fails to open. | |
1654 | */ | |
1655 | spa_l2cache_add(vd); | |
1656 | ||
b128c09f BB |
1657 | vd->vdev_top = vd; |
1658 | vd->vdev_aux = sav; | |
1659 | ||
1660 | spa_l2cache_activate(vd); | |
1661 | ||
34dc7c2f BB |
1662 | if (vdev_open(vd) != 0) |
1663 | continue; | |
1664 | ||
34dc7c2f BB |
1665 | (void) vdev_validate_aux(vd); |
1666 | ||
9babb374 BB |
1667 | if (!vdev_is_dead(vd)) |
1668 | l2arc_add_vdev(spa, vd); | |
34dc7c2f BB |
1669 | } |
1670 | } | |
1671 | ||
67d60824 NB |
1672 | sav->sav_vdevs = newvdevs; |
1673 | sav->sav_count = (int)nl2cache; | |
1674 | ||
1675 | /* | |
1676 | * Recompute the stashed list of l2cache devices, with status | |
1677 | * information this time. | |
1678 | */ | |
1679 | VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, | |
1680 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1681 | ||
460f239e D |
1682 | if (sav->sav_count > 0) |
1683 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
1684 | KM_SLEEP); | |
67d60824 NB |
1685 | for (i = 0; i < sav->sav_count; i++) |
1686 | l2cache[i] = vdev_config_generate(spa, | |
1687 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
1688 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
1689 | ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); | |
1690 | ||
1691 | out: | |
34dc7c2f BB |
1692 | /* |
1693 | * Purge vdevs that were dropped | |
1694 | */ | |
1695 | for (i = 0; i < oldnvdevs; i++) { | |
1696 | uint64_t pool; | |
1697 | ||
1698 | vd = oldvdevs[i]; | |
1699 | if (vd != NULL) { | |
5ffb9d1d GW |
1700 | ASSERT(vd->vdev_isl2cache); |
1701 | ||
fb5f0bc8 BB |
1702 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && |
1703 | pool != 0ULL && l2arc_vdev_present(vd)) | |
34dc7c2f | 1704 | l2arc_remove_vdev(vd); |
5ffb9d1d GW |
1705 | vdev_clear_stats(vd); |
1706 | vdev_free(vd); | |
34dc7c2f BB |
1707 | } |
1708 | } | |
1709 | ||
1710 | if (oldvdevs) | |
1711 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); | |
1712 | ||
34dc7c2f BB |
1713 | for (i = 0; i < sav->sav_count; i++) |
1714 | nvlist_free(l2cache[i]); | |
1715 | if (sav->sav_count) | |
1716 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
1717 | } | |
1718 | ||
1719 | static int | |
1720 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
1721 | { | |
1722 | dmu_buf_t *db; | |
1723 | char *packed = NULL; | |
1724 | size_t nvsize = 0; | |
1725 | int error; | |
1726 | *value = NULL; | |
1727 | ||
c3275b56 BB |
1728 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); |
1729 | if (error) | |
1730 | return (error); | |
1731 | ||
34dc7c2f BB |
1732 | nvsize = *(uint64_t *)db->db_data; |
1733 | dmu_buf_rele(db, FTAG); | |
1734 | ||
77aef6f6 | 1735 | packed = vmem_alloc(nvsize, KM_SLEEP); |
9babb374 BB |
1736 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, |
1737 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
1738 | if (error == 0) |
1739 | error = nvlist_unpack(packed, nvsize, value, 0); | |
77aef6f6 | 1740 | vmem_free(packed, nvsize); |
34dc7c2f BB |
1741 | |
1742 | return (error); | |
1743 | } | |
1744 | ||
1745 | /* | |
1746 | * Checks to see if the given vdev could not be opened, in which case we post a | |
1747 | * sysevent to notify the autoreplace code that the device has been removed. | |
1748 | */ | |
1749 | static void | |
1750 | spa_check_removed(vdev_t *vd) | |
1751 | { | |
1c27024e | 1752 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1753 | spa_check_removed(vd->vdev_child[c]); |
1754 | ||
7011fb60 | 1755 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && |
a1d477c2 | 1756 | vdev_is_concrete(vd)) { |
fb390aaf | 1757 | zfs_post_autoreplace(vd->vdev_spa, vd); |
12fa0466 | 1758 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); |
34dc7c2f BB |
1759 | } |
1760 | } | |
1761 | ||
e0ab3ab5 JS |
1762 | static void |
1763 | spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) | |
1764 | { | |
e0ab3ab5 JS |
1765 | ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); |
1766 | ||
1767 | vd->vdev_top_zap = mvd->vdev_top_zap; | |
1768 | vd->vdev_leaf_zap = mvd->vdev_leaf_zap; | |
1769 | ||
1c27024e | 1770 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
1771 | spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); |
1772 | } | |
1773 | } | |
1774 | ||
9babb374 | 1775 | /* |
572e2857 | 1776 | * Validate the current config against the MOS config |
9babb374 | 1777 | */ |
572e2857 BB |
1778 | static boolean_t |
1779 | spa_config_valid(spa_t *spa, nvlist_t *config) | |
9babb374 | 1780 | { |
572e2857 BB |
1781 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; |
1782 | nvlist_t *nv; | |
1783 | ||
1784 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); | |
1785 | ||
1786 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1787 | VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); | |
1788 | ||
1789 | ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); | |
9babb374 | 1790 | |
428870ff | 1791 | /* |
572e2857 BB |
1792 | * If we're doing a normal import, then build up any additional |
1793 | * diagnostic information about missing devices in this config. | |
1794 | * We'll pass this up to the user for further processing. | |
428870ff | 1795 | */ |
572e2857 BB |
1796 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { |
1797 | nvlist_t **child, *nv; | |
1798 | uint64_t idx = 0; | |
1799 | ||
160987b5 | 1800 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), |
79c76d5b BB |
1801 | KM_SLEEP); |
1802 | VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
572e2857 | 1803 | |
1c27024e | 1804 | for (int c = 0; c < rvd->vdev_children; c++) { |
572e2857 BB |
1805 | vdev_t *tvd = rvd->vdev_child[c]; |
1806 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1807 | ||
1808 | if (tvd->vdev_ops == &vdev_missing_ops && | |
1809 | mtvd->vdev_ops != &vdev_missing_ops && | |
1810 | mtvd->vdev_islog) | |
1811 | child[idx++] = vdev_config_generate(spa, mtvd, | |
1812 | B_FALSE, 0); | |
1813 | } | |
9babb374 | 1814 | |
572e2857 BB |
1815 | if (idx) { |
1816 | VERIFY(nvlist_add_nvlist_array(nv, | |
1817 | ZPOOL_CONFIG_CHILDREN, child, idx) == 0); | |
1818 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
1819 | ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); | |
1820 | ||
1c27024e | 1821 | for (int i = 0; i < idx; i++) |
572e2857 BB |
1822 | nvlist_free(child[i]); |
1823 | } | |
1824 | nvlist_free(nv); | |
1825 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * Compare the root vdev tree with the information we have | |
1830 | * from the MOS config (mrvd). Check each top-level vdev | |
1831 | * with the corresponding MOS config top-level (mtvd). | |
1832 | */ | |
1c27024e | 1833 | for (int c = 0; c < rvd->vdev_children; c++) { |
572e2857 BB |
1834 | vdev_t *tvd = rvd->vdev_child[c]; |
1835 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1836 | ||
1837 | /* | |
1838 | * Resolve any "missing" vdevs in the current configuration. | |
a1d477c2 | 1839 | * Also trust the MOS config about any "indirect" vdevs. |
572e2857 BB |
1840 | * If we find that the MOS config has more accurate information |
1841 | * about the top-level vdev then use that vdev instead. | |
1842 | */ | |
a1d477c2 MA |
1843 | if ((tvd->vdev_ops == &vdev_missing_ops && |
1844 | mtvd->vdev_ops != &vdev_missing_ops) || | |
1845 | (mtvd->vdev_ops == &vdev_indirect_ops && | |
1846 | tvd->vdev_ops != &vdev_indirect_ops)) { | |
572e2857 BB |
1847 | |
1848 | /* | |
1849 | * Device specific actions. | |
1850 | */ | |
1851 | if (mtvd->vdev_islog) { | |
a1d477c2 MA |
1852 | if (!(spa->spa_import_flags & |
1853 | ZFS_IMPORT_MISSING_LOG)) { | |
1854 | continue; | |
1855 | } | |
1856 | ||
572e2857 | 1857 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
a1d477c2 | 1858 | } else if (mtvd->vdev_ops != &vdev_indirect_ops) { |
572e2857 BB |
1859 | continue; |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * Swap the missing vdev with the data we were | |
1864 | * able to obtain from the MOS config. | |
1865 | */ | |
1866 | vdev_remove_child(rvd, tvd); | |
1867 | vdev_remove_child(mrvd, mtvd); | |
1868 | ||
1869 | vdev_add_child(rvd, mtvd); | |
1870 | vdev_add_child(mrvd, tvd); | |
1871 | ||
572e2857 | 1872 | vdev_reopen(rvd); |
e0ab3ab5 JS |
1873 | } else { |
1874 | if (mtvd->vdev_islog) { | |
1875 | /* | |
1876 | * Load the slog device's state from the MOS | |
1877 | * config since it's possible that the label | |
1878 | * does not contain the most up-to-date | |
1879 | * information. | |
1880 | */ | |
1881 | vdev_load_log_state(tvd, mtvd); | |
1882 | vdev_reopen(tvd); | |
1883 | } | |
1884 | ||
572e2857 | 1885 | /* |
e0ab3ab5 | 1886 | * Per-vdev ZAP info is stored exclusively in the MOS. |
572e2857 | 1887 | */ |
e0ab3ab5 | 1888 | spa_config_valid_zaps(tvd, mtvd); |
572e2857 | 1889 | } |
a1d477c2 MA |
1890 | |
1891 | /* | |
1892 | * Never trust this info from userland; always use what's | |
1893 | * in the MOS. This prevents it from getting out of sync | |
1894 | * with the rest of the info in the MOS. | |
1895 | */ | |
1896 | tvd->vdev_removing = mtvd->vdev_removing; | |
1897 | tvd->vdev_indirect_config = mtvd->vdev_indirect_config; | |
9babb374 | 1898 | } |
e0ab3ab5 | 1899 | |
572e2857 | 1900 | vdev_free(mrvd); |
428870ff | 1901 | spa_config_exit(spa, SCL_ALL, FTAG); |
572e2857 BB |
1902 | |
1903 | /* | |
1904 | * Ensure we were able to validate the config. | |
1905 | */ | |
1906 | return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); | |
9babb374 BB |
1907 | } |
1908 | ||
b128c09f BB |
1909 | /* |
1910 | * Check for missing log devices | |
1911 | */ | |
13fe0198 | 1912 | static boolean_t |
b128c09f BB |
1913 | spa_check_logs(spa_t *spa) |
1914 | { | |
13fe0198 | 1915 | boolean_t rv = B_FALSE; |
9c43027b | 1916 | dsl_pool_t *dp = spa_get_dsl(spa); |
13fe0198 | 1917 | |
b128c09f | 1918 | switch (spa->spa_log_state) { |
e75c13c3 BB |
1919 | default: |
1920 | break; | |
b128c09f BB |
1921 | case SPA_LOG_MISSING: |
1922 | /* need to recheck in case slog has been restored */ | |
1923 | case SPA_LOG_UNKNOWN: | |
9c43027b AJ |
1924 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
1925 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
13fe0198 | 1926 | if (rv) |
428870ff | 1927 | spa_set_log_state(spa, SPA_LOG_MISSING); |
b128c09f | 1928 | break; |
b128c09f | 1929 | } |
13fe0198 | 1930 | return (rv); |
b128c09f BB |
1931 | } |
1932 | ||
428870ff BB |
1933 | static boolean_t |
1934 | spa_passivate_log(spa_t *spa) | |
34dc7c2f | 1935 | { |
428870ff BB |
1936 | vdev_t *rvd = spa->spa_root_vdev; |
1937 | boolean_t slog_found = B_FALSE; | |
b128c09f | 1938 | |
428870ff | 1939 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
fb5f0bc8 | 1940 | |
428870ff BB |
1941 | if (!spa_has_slogs(spa)) |
1942 | return (B_FALSE); | |
34dc7c2f | 1943 | |
1c27024e | 1944 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff BB |
1945 | vdev_t *tvd = rvd->vdev_child[c]; |
1946 | metaslab_group_t *mg = tvd->vdev_mg; | |
34dc7c2f | 1947 | |
428870ff BB |
1948 | if (tvd->vdev_islog) { |
1949 | metaslab_group_passivate(mg); | |
1950 | slog_found = B_TRUE; | |
1951 | } | |
34dc7c2f BB |
1952 | } |
1953 | ||
428870ff BB |
1954 | return (slog_found); |
1955 | } | |
34dc7c2f | 1956 | |
428870ff BB |
1957 | static void |
1958 | spa_activate_log(spa_t *spa) | |
1959 | { | |
1960 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f | 1961 | |
428870ff BB |
1962 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
1963 | ||
1c27024e | 1964 | for (int c = 0; c < rvd->vdev_children; c++) { |
428870ff BB |
1965 | vdev_t *tvd = rvd->vdev_child[c]; |
1966 | metaslab_group_t *mg = tvd->vdev_mg; | |
1967 | ||
1968 | if (tvd->vdev_islog) | |
1969 | metaslab_group_activate(mg); | |
34dc7c2f | 1970 | } |
428870ff | 1971 | } |
34dc7c2f | 1972 | |
428870ff | 1973 | int |
a1d477c2 | 1974 | spa_reset_logs(spa_t *spa) |
428870ff | 1975 | { |
13fe0198 | 1976 | int error; |
9babb374 | 1977 | |
a1d477c2 | 1978 | error = dmu_objset_find(spa_name(spa), zil_reset, |
13fe0198 MA |
1979 | NULL, DS_FIND_CHILDREN); |
1980 | if (error == 0) { | |
428870ff BB |
1981 | /* |
1982 | * We successfully offlined the log device, sync out the | |
1983 | * current txg so that the "stubby" block can be removed | |
1984 | * by zil_sync(). | |
1985 | */ | |
1986 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
1987 | } | |
1988 | return (error); | |
1989 | } | |
34dc7c2f | 1990 | |
428870ff BB |
1991 | static void |
1992 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
1993 | { | |
1c27024e | 1994 | for (int i = 0; i < sav->sav_count; i++) |
428870ff BB |
1995 | spa_check_removed(sav->sav_vdevs[i]); |
1996 | } | |
34dc7c2f | 1997 | |
428870ff BB |
1998 | void |
1999 | spa_claim_notify(zio_t *zio) | |
2000 | { | |
2001 | spa_t *spa = zio->io_spa; | |
34dc7c2f | 2002 | |
428870ff BB |
2003 | if (zio->io_error) |
2004 | return; | |
34dc7c2f | 2005 | |
428870ff BB |
2006 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ |
2007 | if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) | |
2008 | spa->spa_claim_max_txg = zio->io_bp->blk_birth; | |
2009 | mutex_exit(&spa->spa_props_lock); | |
2010 | } | |
34dc7c2f | 2011 | |
428870ff BB |
2012 | typedef struct spa_load_error { |
2013 | uint64_t sle_meta_count; | |
2014 | uint64_t sle_data_count; | |
2015 | } spa_load_error_t; | |
34dc7c2f | 2016 | |
428870ff BB |
2017 | static void |
2018 | spa_load_verify_done(zio_t *zio) | |
2019 | { | |
2020 | blkptr_t *bp = zio->io_bp; | |
2021 | spa_load_error_t *sle = zio->io_private; | |
2022 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
2023 | int error = zio->io_error; | |
dea377c0 | 2024 | spa_t *spa = zio->io_spa; |
34dc7c2f | 2025 | |
a6255b7f | 2026 | abd_free(zio->io_abd); |
428870ff | 2027 | if (error) { |
9ae529ec | 2028 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && |
428870ff | 2029 | type != DMU_OT_INTENT_LOG) |
bc89ac84 | 2030 | atomic_inc_64(&sle->sle_meta_count); |
428870ff | 2031 | else |
bc89ac84 | 2032 | atomic_inc_64(&sle->sle_data_count); |
34dc7c2f | 2033 | } |
dea377c0 MA |
2034 | |
2035 | mutex_enter(&spa->spa_scrub_lock); | |
d4a72f23 | 2036 | spa->spa_load_verify_ios--; |
dea377c0 MA |
2037 | cv_broadcast(&spa->spa_scrub_io_cv); |
2038 | mutex_exit(&spa->spa_scrub_lock); | |
428870ff | 2039 | } |
34dc7c2f | 2040 | |
dea377c0 MA |
2041 | /* |
2042 | * Maximum number of concurrent scrub i/os to create while verifying | |
2043 | * a pool while importing it. | |
2044 | */ | |
2045 | int spa_load_verify_maxinflight = 10000; | |
2046 | int spa_load_verify_metadata = B_TRUE; | |
2047 | int spa_load_verify_data = B_TRUE; | |
2048 | ||
428870ff BB |
2049 | /*ARGSUSED*/ |
2050 | static int | |
2051 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
5dbd68a3 | 2052 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) |
428870ff | 2053 | { |
fcff0f35 | 2054 | if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) |
dea377c0 MA |
2055 | return (0); |
2056 | /* | |
2057 | * Note: normally this routine will not be called if | |
2058 | * spa_load_verify_metadata is not set. However, it may be useful | |
2059 | * to manually set the flag after the traversal has begun. | |
2060 | */ | |
2061 | if (!spa_load_verify_metadata) | |
2062 | return (0); | |
a6255b7f | 2063 | if (!BP_IS_METADATA(bp) && !spa_load_verify_data) |
dea377c0 MA |
2064 | return (0); |
2065 | ||
1c27024e DB |
2066 | zio_t *rio = arg; |
2067 | size_t size = BP_GET_PSIZE(bp); | |
dea377c0 MA |
2068 | |
2069 | mutex_enter(&spa->spa_scrub_lock); | |
d4a72f23 | 2070 | while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight) |
dea377c0 | 2071 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); |
d4a72f23 | 2072 | spa->spa_load_verify_ios++; |
dea377c0 MA |
2073 | mutex_exit(&spa->spa_scrub_lock); |
2074 | ||
a6255b7f | 2075 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, |
dea377c0 MA |
2076 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, |
2077 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2078 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
428870ff BB |
2079 | return (0); |
2080 | } | |
34dc7c2f | 2081 | |
d1d19c78 PD |
2082 | /* ARGSUSED */ |
2083 | int | |
2084 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) | |
2085 | { | |
2086 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) | |
2087 | return (SET_ERROR(ENAMETOOLONG)); | |
2088 | ||
2089 | return (0); | |
2090 | } | |
2091 | ||
428870ff BB |
2092 | static int |
2093 | spa_load_verify(spa_t *spa) | |
2094 | { | |
2095 | zio_t *rio; | |
2096 | spa_load_error_t sle = { 0 }; | |
2097 | zpool_rewind_policy_t policy; | |
2098 | boolean_t verify_ok = B_FALSE; | |
dea377c0 | 2099 | int error = 0; |
34dc7c2f | 2100 | |
428870ff | 2101 | zpool_get_rewind_policy(spa->spa_config, &policy); |
34dc7c2f | 2102 | |
428870ff BB |
2103 | if (policy.zrp_request & ZPOOL_NEVER_REWIND) |
2104 | return (0); | |
34dc7c2f | 2105 | |
d1d19c78 PD |
2106 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); |
2107 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2108 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2109 | DS_FIND_CHILDREN); | |
2110 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2111 | if (error != 0) | |
2112 | return (error); | |
2113 | ||
428870ff BB |
2114 | rio = zio_root(spa, NULL, &sle, |
2115 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
34dc7c2f | 2116 | |
dea377c0 MA |
2117 | if (spa_load_verify_metadata) { |
2118 | error = traverse_pool(spa, spa->spa_verify_min_txg, | |
b5256303 TC |
2119 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | |
2120 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
dea377c0 | 2121 | } |
428870ff BB |
2122 | |
2123 | (void) zio_wait(rio); | |
2124 | ||
2125 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2126 | spa->spa_load_data_errors = sle.sle_data_count; | |
2127 | ||
2128 | if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && | |
2129 | sle.sle_data_count <= policy.zrp_maxdata) { | |
572e2857 BB |
2130 | int64_t loss = 0; |
2131 | ||
428870ff BB |
2132 | verify_ok = B_TRUE; |
2133 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2134 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
572e2857 BB |
2135 | |
2136 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
2137 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2138 | ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); | |
2139 | VERIFY(nvlist_add_int64(spa->spa_load_info, | |
2140 | ZPOOL_CONFIG_REWIND_TIME, loss) == 0); | |
2141 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2142 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); | |
428870ff BB |
2143 | } else { |
2144 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2145 | } | |
2146 | ||
2147 | if (error) { | |
2148 | if (error != ENXIO && error != EIO) | |
2e528b49 | 2149 | error = SET_ERROR(EIO); |
428870ff BB |
2150 | return (error); |
2151 | } | |
2152 | ||
2153 | return (verify_ok ? 0 : EIO); | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Find a value in the pool props object. | |
2158 | */ | |
2159 | static void | |
2160 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2161 | { | |
2162 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2163 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * Find a value in the pool directory object. | |
2168 | */ | |
2169 | static int | |
2170 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) | |
2171 | { | |
2172 | return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2173 | name, sizeof (uint64_t), 1, val)); | |
2174 | } | |
2175 | ||
2176 | static int | |
2177 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2178 | { | |
2179 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
a1d477c2 | 2180 | return (SET_ERROR(err)); |
428870ff BB |
2181 | } |
2182 | ||
2183 | /* | |
2184 | * Fix up config after a partly-completed split. This is done with the | |
2185 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
2186 | * pool have that entry in their config, but only the splitting one contains | |
2187 | * a list of all the guids of the vdevs that are being split off. | |
2188 | * | |
2189 | * This function determines what to do with that list: either rejoin | |
2190 | * all the disks to the pool, or complete the splitting process. To attempt | |
2191 | * the rejoin, each disk that is offlined is marked online again, and | |
2192 | * we do a reopen() call. If the vdev label for every disk that was | |
2193 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
2194 | * then we call vdev_split() on each disk, and complete the split. | |
2195 | * | |
2196 | * Otherwise we leave the config alone, with all the vdevs in place in | |
2197 | * the original pool. | |
2198 | */ | |
2199 | static void | |
2200 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
2201 | { | |
2202 | uint_t extracted; | |
2203 | uint64_t *glist; | |
2204 | uint_t i, gcount; | |
2205 | nvlist_t *nvl; | |
2206 | vdev_t **vd; | |
2207 | boolean_t attempt_reopen; | |
2208 | ||
2209 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
2210 | return; | |
2211 | ||
2212 | /* check that the config is complete */ | |
2213 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
2214 | &glist, &gcount) != 0) | |
2215 | return; | |
2216 | ||
79c76d5b | 2217 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); |
428870ff BB |
2218 | |
2219 | /* attempt to online all the vdevs & validate */ | |
2220 | attempt_reopen = B_TRUE; | |
2221 | for (i = 0; i < gcount; i++) { | |
2222 | if (glist[i] == 0) /* vdev is hole */ | |
2223 | continue; | |
2224 | ||
2225 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
2226 | if (vd[i] == NULL) { | |
2227 | /* | |
2228 | * Don't bother attempting to reopen the disks; | |
2229 | * just do the split. | |
2230 | */ | |
2231 | attempt_reopen = B_FALSE; | |
2232 | } else { | |
2233 | /* attempt to re-online it */ | |
2234 | vd[i]->vdev_offline = B_FALSE; | |
2235 | } | |
2236 | } | |
2237 | ||
2238 | if (attempt_reopen) { | |
2239 | vdev_reopen(spa->spa_root_vdev); | |
2240 | ||
2241 | /* check each device to see what state it's in */ | |
2242 | for (extracted = 0, i = 0; i < gcount; i++) { | |
2243 | if (vd[i] != NULL && | |
2244 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
2245 | break; | |
2246 | ++extracted; | |
2247 | } | |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * If every disk has been moved to the new pool, or if we never | |
2252 | * even attempted to look at them, then we split them off for | |
2253 | * good. | |
2254 | */ | |
2255 | if (!attempt_reopen || gcount == extracted) { | |
2256 | for (i = 0; i < gcount; i++) | |
2257 | if (vd[i] != NULL) | |
2258 | vdev_split(vd[i]); | |
2259 | vdev_reopen(spa->spa_root_vdev); | |
2260 | } | |
2261 | ||
2262 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
2263 | } | |
2264 | ||
2265 | static int | |
2266 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, | |
2267 | boolean_t mosconfig) | |
2268 | { | |
2269 | nvlist_t *config = spa->spa_config; | |
2270 | char *ereport = FM_EREPORT_ZFS_POOL; | |
d96eb2b1 | 2271 | char *comment; |
428870ff BB |
2272 | int error; |
2273 | uint64_t pool_guid; | |
2274 | nvlist_t *nvl; | |
2275 | ||
2276 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) | |
2e528b49 | 2277 | return (SET_ERROR(EINVAL)); |
428870ff | 2278 | |
d96eb2b1 DM |
2279 | ASSERT(spa->spa_comment == NULL); |
2280 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
2281 | spa->spa_comment = spa_strdup(comment); | |
2282 | ||
428870ff BB |
2283 | /* |
2284 | * Versioning wasn't explicitly added to the label until later, so if | |
2285 | * it's not present treat it as the initial version. | |
2286 | */ | |
2287 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
2288 | &spa->spa_ubsync.ub_version) != 0) | |
2289 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
2290 | ||
2291 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
2292 | &spa->spa_config_txg); | |
2293 | ||
2294 | if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && | |
2295 | spa_guid_exists(pool_guid, 0)) { | |
2e528b49 | 2296 | error = SET_ERROR(EEXIST); |
428870ff | 2297 | } else { |
3541dc6d | 2298 | spa->spa_config_guid = pool_guid; |
428870ff BB |
2299 | |
2300 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, | |
2301 | &nvl) == 0) { | |
2302 | VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, | |
79c76d5b | 2303 | KM_SLEEP) == 0); |
428870ff BB |
2304 | } |
2305 | ||
9ae529ec CS |
2306 | nvlist_free(spa->spa_load_info); |
2307 | spa->spa_load_info = fnvlist_alloc(); | |
2308 | ||
572e2857 | 2309 | gethrestime(&spa->spa_loaded_ts); |
428870ff BB |
2310 | error = spa_load_impl(spa, pool_guid, config, state, type, |
2311 | mosconfig, &ereport); | |
2312 | } | |
2313 | ||
0c66c32d JG |
2314 | /* |
2315 | * Don't count references from objsets that are already closed | |
2316 | * and are making their way through the eviction process. | |
2317 | */ | |
2318 | spa_evicting_os_wait(spa); | |
428870ff | 2319 | spa->spa_minref = refcount_count(&spa->spa_refcount); |
572e2857 BB |
2320 | if (error) { |
2321 | if (error != EEXIST) { | |
2322 | spa->spa_loaded_ts.tv_sec = 0; | |
2323 | spa->spa_loaded_ts.tv_nsec = 0; | |
2324 | } | |
2325 | if (error != EBADF) { | |
b5256303 | 2326 | zfs_ereport_post(ereport, spa, NULL, NULL, NULL, 0, 0); |
572e2857 BB |
2327 | } |
2328 | } | |
428870ff BB |
2329 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; |
2330 | spa->spa_ena = 0; | |
2331 | ||
2332 | return (error); | |
2333 | } | |
2334 | ||
33cf67cd | 2335 | #ifdef ZFS_DEBUG |
e0ab3ab5 JS |
2336 | /* |
2337 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
2338 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
2339 | * spa's per-vdev ZAP list. | |
2340 | */ | |
2341 | static uint64_t | |
2342 | vdev_count_verify_zaps(vdev_t *vd) | |
2343 | { | |
2344 | spa_t *spa = vd->vdev_spa; | |
2345 | uint64_t total = 0; | |
e0ab3ab5 JS |
2346 | |
2347 | if (vd->vdev_top_zap != 0) { | |
2348 | total++; | |
2349 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2350 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
2351 | } | |
2352 | if (vd->vdev_leaf_zap != 0) { | |
2353 | total++; | |
2354 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2355 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
2356 | } | |
2357 | ||
1c27024e | 2358 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
2359 | total += vdev_count_verify_zaps(vd->vdev_child[i]); |
2360 | } | |
2361 | ||
2362 | return (total); | |
2363 | } | |
33cf67cd | 2364 | #endif |
e0ab3ab5 | 2365 | |
379ca9cf OF |
2366 | /* |
2367 | * Determine whether the activity check is required. | |
2368 | */ | |
2369 | static boolean_t | |
bbffb59e BB |
2370 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, |
2371 | nvlist_t *config) | |
379ca9cf OF |
2372 | { |
2373 | uint64_t state = 0; | |
2374 | uint64_t hostid = 0; | |
2375 | uint64_t tryconfig_txg = 0; | |
2376 | uint64_t tryconfig_timestamp = 0; | |
2377 | nvlist_t *nvinfo; | |
2378 | ||
2379 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
2380 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
2381 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
2382 | &tryconfig_txg); | |
2383 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
2384 | &tryconfig_timestamp); | |
2385 | } | |
2386 | ||
2387 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
379ca9cf OF |
2388 | |
2389 | /* | |
2390 | * Disable the MMP activity check - This is used by zdb which | |
2391 | * is intended to be used on potentially active pools. | |
2392 | */ | |
2393 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
2394 | return (B_FALSE); | |
2395 | ||
2396 | /* | |
2397 | * Skip the activity check when the MMP feature is disabled. | |
2398 | */ | |
2399 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
2400 | return (B_FALSE); | |
2401 | /* | |
2402 | * If the tryconfig_* values are nonzero, they are the results of an | |
2403 | * earlier tryimport. If they match the uberblock we just found, then | |
2404 | * the pool has not changed and we return false so we do not test a | |
2405 | * second time. | |
2406 | */ | |
2407 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
2408 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp) | |
2409 | return (B_FALSE); | |
2410 | ||
2411 | /* | |
2412 | * Allow the activity check to be skipped when importing the pool | |
bbffb59e BB |
2413 | * on the same host which last imported it. Since the hostid from |
2414 | * configuration may be stale use the one read from the label. | |
379ca9cf | 2415 | */ |
bbffb59e BB |
2416 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) |
2417 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
2418 | ||
379ca9cf OF |
2419 | if (hostid == spa_get_hostid()) |
2420 | return (B_FALSE); | |
2421 | ||
2422 | /* | |
2423 | * Skip the activity test when the pool was cleanly exported. | |
2424 | */ | |
2425 | if (state != POOL_STATE_ACTIVE) | |
2426 | return (B_FALSE); | |
2427 | ||
2428 | return (B_TRUE); | |
2429 | } | |
2430 | ||
2431 | /* | |
2432 | * Perform the import activity check. If the user canceled the import or | |
2433 | * we detected activity then fail. | |
2434 | */ | |
2435 | static int | |
2436 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config) | |
2437 | { | |
2438 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
2439 | uint64_t txg = ub->ub_txg; | |
2440 | uint64_t timestamp = ub->ub_timestamp; | |
2441 | uint64_t import_delay = NANOSEC; | |
2442 | hrtime_t import_expire; | |
2443 | nvlist_t *mmp_label = NULL; | |
2444 | vdev_t *rvd = spa->spa_root_vdev; | |
2445 | kcondvar_t cv; | |
2446 | kmutex_t mtx; | |
2447 | int error = 0; | |
2448 | ||
2449 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
2450 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
2451 | mutex_enter(&mtx); | |
2452 | ||
2453 | /* | |
2454 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
2455 | * during the earlier tryimport. If the txg recorded there is 0 then | |
2456 | * the pool is known to be active on another host. | |
2457 | * | |
2458 | * Otherwise, the pool might be in use on another node. Check for | |
2459 | * changes in the uberblocks on disk if necessary. | |
2460 | */ | |
2461 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
2462 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
2463 | ZPOOL_CONFIG_LOAD_INFO); | |
2464 | ||
2465 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
2466 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
2467 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
2468 | error = SET_ERROR(EREMOTEIO); | |
2469 | goto out; | |
2470 | } | |
2471 | } | |
2472 | ||
2473 | /* | |
2474 | * Preferentially use the zfs_multihost_interval from the node which | |
2475 | * last imported the pool. This value is stored in an MMP uberblock as. | |
2476 | * | |
2477 | * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval | |
2478 | */ | |
2479 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay) | |
2480 | import_delay = MAX(import_delay, import_intervals * | |
802ae562 | 2481 | ub->ub_mmp_delay * MAX(vdev_count_leaves(spa), 1)); |
379ca9cf OF |
2482 | |
2483 | /* Apply a floor using the local default values. */ | |
2484 | import_delay = MAX(import_delay, import_intervals * | |
2485 | MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL))); | |
2486 | ||
533ea041 OF |
2487 | zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u " |
2488 | "leaves=%u", import_delay, ub->ub_mmp_delay, import_intervals, | |
2489 | vdev_count_leaves(spa)); | |
2490 | ||
379ca9cf OF |
2491 | /* Add a small random factor in case of simultaneous imports (0-25%) */ |
2492 | import_expire = gethrtime() + import_delay + | |
2493 | (import_delay * spa_get_random(250) / 1000); | |
2494 | ||
2495 | while (gethrtime() < import_expire) { | |
2496 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
2497 | ||
2498 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp) { | |
2499 | error = SET_ERROR(EREMOTEIO); | |
2500 | break; | |
2501 | } | |
2502 | ||
2503 | if (mmp_label) { | |
2504 | nvlist_free(mmp_label); | |
2505 | mmp_label = NULL; | |
2506 | } | |
2507 | ||
2508 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
2509 | if (error != -1) { | |
2510 | error = SET_ERROR(EINTR); | |
2511 | break; | |
2512 | } | |
2513 | error = 0; | |
2514 | } | |
2515 | ||
2516 | out: | |
2517 | mutex_exit(&mtx); | |
2518 | mutex_destroy(&mtx); | |
2519 | cv_destroy(&cv); | |
2520 | ||
2521 | /* | |
2522 | * If the pool is determined to be active store the status in the | |
2523 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
2524 | * available from configuration read from disk store them as well. | |
2525 | * This allows 'zpool import' to generate a more useful message. | |
2526 | * | |
2527 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
2528 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
2529 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
2530 | */ | |
2531 | if (error == EREMOTEIO) { | |
2532 | char *hostname = "<unknown>"; | |
2533 | uint64_t hostid = 0; | |
2534 | ||
2535 | if (mmp_label) { | |
2536 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
2537 | hostname = fnvlist_lookup_string(mmp_label, | |
2538 | ZPOOL_CONFIG_HOSTNAME); | |
2539 | fnvlist_add_string(spa->spa_load_info, | |
2540 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
2541 | } | |
2542 | ||
2543 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
2544 | hostid = fnvlist_lookup_uint64(mmp_label, | |
2545 | ZPOOL_CONFIG_HOSTID); | |
2546 | fnvlist_add_uint64(spa->spa_load_info, | |
2547 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
2548 | } | |
2549 | } | |
2550 | ||
2551 | fnvlist_add_uint64(spa->spa_load_info, | |
2552 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
2553 | fnvlist_add_uint64(spa->spa_load_info, | |
2554 | ZPOOL_CONFIG_MMP_TXG, 0); | |
2555 | ||
2556 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
2557 | } | |
2558 | ||
2559 | if (mmp_label) | |
2560 | nvlist_free(mmp_label); | |
2561 | ||
2562 | return (error); | |
2563 | } | |
2564 | ||
428870ff BB |
2565 | /* |
2566 | * Load an existing storage pool, using the pool's builtin spa_config as a | |
2567 | * source of configuration information. | |
2568 | */ | |
bf701a83 BB |
2569 | __attribute__((always_inline)) |
2570 | static inline int | |
428870ff | 2571 | spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, |
a1d477c2 | 2572 | spa_load_state_t state, spa_import_type_t type, boolean_t trust_config, |
428870ff BB |
2573 | char **ereport) |
2574 | { | |
2575 | int error = 0; | |
2576 | nvlist_t *nvroot = NULL; | |
9ae529ec | 2577 | nvlist_t *label; |
428870ff BB |
2578 | vdev_t *rvd; |
2579 | uberblock_t *ub = &spa->spa_uberblock; | |
572e2857 | 2580 | uint64_t children, config_cache_txg = spa->spa_config_txg; |
428870ff | 2581 | int orig_mode = spa->spa_mode; |
1c27024e | 2582 | int parse; |
428870ff | 2583 | uint64_t obj; |
9ae529ec | 2584 | boolean_t missing_feat_write = B_FALSE; |
379ca9cf | 2585 | boolean_t activity_check = B_FALSE; |
428870ff BB |
2586 | |
2587 | /* | |
2588 | * If this is an untrusted config, access the pool in read-only mode. | |
2589 | * This prevents things like resilvering recently removed devices. | |
2590 | */ | |
a1d477c2 | 2591 | if (!trust_config) |
428870ff BB |
2592 | spa->spa_mode = FREAD; |
2593 | ||
2594 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
2595 | ||
2596 | spa->spa_load_state = state; | |
2597 | ||
2598 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) | |
2e528b49 | 2599 | return (SET_ERROR(EINVAL)); |
428870ff BB |
2600 | |
2601 | parse = (type == SPA_IMPORT_EXISTING ? | |
2602 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
2603 | ||
2604 | /* | |
2605 | * Create "The Godfather" zio to hold all async IOs | |
2606 | */ | |
e022864d MA |
2607 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
2608 | KM_SLEEP); | |
1c27024e | 2609 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
2610 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
2611 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
2612 | ZIO_FLAG_GODFATHER); | |
2613 | } | |
428870ff BB |
2614 | |
2615 | /* | |
2616 | * Parse the configuration into a vdev tree. We explicitly set the | |
2617 | * value that will be returned by spa_version() since parsing the | |
2618 | * configuration requires knowing the version number. | |
2619 | */ | |
2620 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2621 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); | |
2622 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2623 | ||
2624 | if (error != 0) | |
2625 | return (error); | |
2626 | ||
2627 | ASSERT(spa->spa_root_vdev == rvd); | |
c3520e7f MA |
2628 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); |
2629 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
428870ff BB |
2630 | |
2631 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2632 | ASSERT(spa_guid(spa) == pool_guid); | |
2633 | } | |
2634 | ||
2635 | /* | |
2636 | * Try to open all vdevs, loading each label in the process. | |
2637 | */ | |
2638 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2639 | error = vdev_open(rvd); | |
2640 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2641 | if (error != 0) | |
2642 | return (error); | |
2643 | ||
2644 | /* | |
2645 | * We need to validate the vdev labels against the configuration that | |
2646 | * we have in hand, which is dependent on the setting of mosconfig. If | |
2647 | * mosconfig is true then we're validating the vdev labels based on | |
2648 | * that config. Otherwise, we're validating against the cached config | |
2649 | * (zpool.cache) that was read when we loaded the zfs module, and then | |
2650 | * later we will recursively call spa_load() and validate against | |
2651 | * the vdev config. | |
2652 | * | |
2653 | * If we're assembling a new pool that's been split off from an | |
2654 | * existing pool, the labels haven't yet been updated so we skip | |
2655 | * validation for now. | |
2656 | */ | |
2657 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2658 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
a1d477c2 | 2659 | error = vdev_validate(rvd, trust_config); |
428870ff BB |
2660 | spa_config_exit(spa, SCL_ALL, FTAG); |
2661 | ||
2662 | if (error != 0) | |
2663 | return (error); | |
2664 | ||
2665 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
2e528b49 | 2666 | return (SET_ERROR(ENXIO)); |
428870ff BB |
2667 | } |
2668 | ||
2669 | /* | |
2670 | * Find the best uberblock. | |
2671 | */ | |
9ae529ec | 2672 | vdev_uberblock_load(rvd, ub, &label); |
428870ff BB |
2673 | |
2674 | /* | |
2675 | * If we weren't able to find a single valid uberblock, return failure. | |
2676 | */ | |
9ae529ec CS |
2677 | if (ub->ub_txg == 0) { |
2678 | nvlist_free(label); | |
428870ff | 2679 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); |
9ae529ec | 2680 | } |
428870ff | 2681 | |
379ca9cf OF |
2682 | /* |
2683 | * For pools which have the multihost property on determine if the | |
2684 | * pool is truly inactive and can be safely imported. Prevent | |
2685 | * hosts which don't have a hostid set from importing the pool. | |
2686 | */ | |
bbffb59e | 2687 | activity_check = spa_activity_check_required(spa, ub, label, config); |
379ca9cf | 2688 | if (activity_check) { |
379ca9cf OF |
2689 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && |
2690 | spa_get_hostid() == 0) { | |
2691 | nvlist_free(label); | |
2692 | fnvlist_add_uint64(spa->spa_load_info, | |
2693 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
2694 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
2695 | } | |
2696 | ||
e889f0f5 OF |
2697 | error = spa_activity_check(spa, ub, config); |
2698 | if (error) { | |
2699 | nvlist_free(label); | |
2700 | return (error); | |
2701 | } | |
2702 | ||
379ca9cf OF |
2703 | fnvlist_add_uint64(spa->spa_load_info, |
2704 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
2705 | fnvlist_add_uint64(spa->spa_load_info, | |
2706 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
2707 | } | |
2708 | ||
428870ff | 2709 | /* |
9ae529ec | 2710 | * If the pool has an unsupported version we can't open it. |
428870ff | 2711 | */ |
9ae529ec CS |
2712 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { |
2713 | nvlist_free(label); | |
428870ff | 2714 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); |
9ae529ec CS |
2715 | } |
2716 | ||
2717 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2718 | nvlist_t *features; | |
2719 | ||
2720 | /* | |
2721 | * If we weren't able to find what's necessary for reading the | |
2722 | * MOS in the label, return failure. | |
2723 | */ | |
2724 | if (label == NULL || nvlist_lookup_nvlist(label, | |
2725 | ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { | |
2726 | nvlist_free(label); | |
2727 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
2728 | ENXIO)); | |
2729 | } | |
2730 | ||
2731 | /* | |
2732 | * Update our in-core representation with the definitive values | |
2733 | * from the label. | |
2734 | */ | |
2735 | nvlist_free(spa->spa_label_features); | |
2736 | VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); | |
2737 | } | |
2738 | ||
2739 | nvlist_free(label); | |
2740 | ||
2741 | /* | |
2742 | * Look through entries in the label nvlist's features_for_read. If | |
2743 | * there is a feature listed there which we don't understand then we | |
2744 | * cannot open a pool. | |
2745 | */ | |
2746 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2747 | nvlist_t *unsup_feat; | |
9ae529ec CS |
2748 | |
2749 | VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == | |
2750 | 0); | |
2751 | ||
1c27024e DB |
2752 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, |
2753 | NULL); nvp != NULL; | |
9ae529ec CS |
2754 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { |
2755 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
2756 | VERIFY(nvlist_add_string(unsup_feat, | |
2757 | nvpair_name(nvp), "") == 0); | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | if (!nvlist_empty(unsup_feat)) { | |
2762 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
2763 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); | |
2764 | nvlist_free(unsup_feat); | |
2765 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2766 | ENOTSUP)); | |
2767 | } | |
2768 | ||
2769 | nvlist_free(unsup_feat); | |
2770 | } | |
428870ff BB |
2771 | |
2772 | /* | |
2773 | * If the vdev guid sum doesn't match the uberblock, we have an | |
572e2857 BB |
2774 | * incomplete configuration. We first check to see if the pool |
2775 | * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). | |
2776 | * If it is, defer the vdev_guid_sum check till later so we | |
2777 | * can handle missing vdevs. | |
428870ff | 2778 | */ |
572e2857 | 2779 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, |
a1d477c2 | 2780 | &children) != 0 && trust_config && type != SPA_IMPORT_ASSEMBLE && |
428870ff BB |
2781 | rvd->vdev_guid_sum != ub->ub_guid_sum) |
2782 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
2783 | ||
2784 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { | |
2785 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2786 | spa_try_repair(spa, config); | |
2787 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2788 | nvlist_free(spa->spa_config_splitting); | |
2789 | spa->spa_config_splitting = NULL; | |
2790 | } | |
2791 | ||
2792 | /* | |
2793 | * Initialize internal SPA structures. | |
2794 | */ | |
2795 | spa->spa_state = POOL_STATE_ACTIVE; | |
2796 | spa->spa_ubsync = spa->spa_uberblock; | |
2797 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
2798 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
2799 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
2800 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
2801 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
2802 | spa->spa_prev_software_version = ub->ub_software_version; | |
2803 | ||
a1d477c2 MA |
2804 | /* |
2805 | * Everything that we read before we do spa_remove_init() must | |
2806 | * have been rewritten after the last device removal was initiated. | |
2807 | * Otherwise we could be reading from indirect vdevs before | |
2808 | * we have loaded their mappings. | |
2809 | */ | |
2810 | ||
9ae529ec | 2811 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
428870ff BB |
2812 | if (error) |
2813 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2814 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; | |
2815 | ||
2816 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) | |
2817 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2818 | ||
a1d477c2 MA |
2819 | /* |
2820 | * Validate the config, using the MOS config to fill in any | |
2821 | * information which might be missing. If we fail to validate | |
2822 | * the config then declare the pool unfit for use. If we're | |
2823 | * assembling a pool from a split, the log is not transferred | |
2824 | * over. | |
2825 | */ | |
2826 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2827 | nvlist_t *mos_config; | |
2828 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) | |
2829 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2830 | ||
2831 | if (!spa_config_valid(spa, mos_config)) { | |
2832 | nvlist_free(mos_config); | |
2833 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
2834 | ENXIO)); | |
2835 | } | |
2836 | nvlist_free(mos_config); | |
2837 | ||
2838 | /* | |
2839 | * Now that we've validated the config, check the state of the | |
2840 | * root vdev. If it can't be opened, it indicates one or | |
2841 | * more toplevel vdevs are faulted. | |
2842 | */ | |
2843 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
2844 | return (SET_ERROR(ENXIO)); | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Everything that we read before spa_remove_init() must be stored | |
2849 | * on concreted vdevs. Therefore we do this as early as possible. | |
2850 | */ | |
2851 | if (spa_remove_init(spa) != 0) | |
2852 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2853 | ||
9ae529ec CS |
2854 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { |
2855 | boolean_t missing_feat_read = B_FALSE; | |
b9b24bb4 | 2856 | nvlist_t *unsup_feat, *enabled_feat; |
9ae529ec CS |
2857 | |
2858 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
2859 | &spa->spa_feat_for_read_obj) != 0) { | |
2860 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2861 | } | |
2862 | ||
2863 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
2864 | &spa->spa_feat_for_write_obj) != 0) { | |
2865 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2866 | } | |
2867 | ||
2868 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
2869 | &spa->spa_feat_desc_obj) != 0) { | |
2870 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2871 | } | |
2872 | ||
b9b24bb4 CS |
2873 | enabled_feat = fnvlist_alloc(); |
2874 | unsup_feat = fnvlist_alloc(); | |
9ae529ec | 2875 | |
fa86b5db | 2876 | if (!spa_features_check(spa, B_FALSE, |
b9b24bb4 | 2877 | unsup_feat, enabled_feat)) |
9ae529ec CS |
2878 | missing_feat_read = B_TRUE; |
2879 | ||
2880 | if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { | |
fa86b5db | 2881 | if (!spa_features_check(spa, B_TRUE, |
b9b24bb4 | 2882 | unsup_feat, enabled_feat)) { |
9ae529ec | 2883 | missing_feat_write = B_TRUE; |
b9b24bb4 | 2884 | } |
9ae529ec CS |
2885 | } |
2886 | ||
b9b24bb4 CS |
2887 | fnvlist_add_nvlist(spa->spa_load_info, |
2888 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
2889 | ||
9ae529ec | 2890 | if (!nvlist_empty(unsup_feat)) { |
b9b24bb4 CS |
2891 | fnvlist_add_nvlist(spa->spa_load_info, |
2892 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
9ae529ec CS |
2893 | } |
2894 | ||
b9b24bb4 CS |
2895 | fnvlist_free(enabled_feat); |
2896 | fnvlist_free(unsup_feat); | |
9ae529ec CS |
2897 | |
2898 | if (!missing_feat_read) { | |
2899 | fnvlist_add_boolean(spa->spa_load_info, | |
2900 | ZPOOL_CONFIG_CAN_RDONLY); | |
2901 | } | |
2902 | ||
2903 | /* | |
2904 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
2905 | * twofold: to determine whether the pool is available for | |
2906 | * import in read-write mode and (if it is not) whether the | |
2907 | * pool is available for import in read-only mode. If the pool | |
2908 | * is available for import in read-write mode, it is displayed | |
2909 | * as available in userland; if it is not available for import | |
2910 | * in read-only mode, it is displayed as unavailable in | |
2911 | * userland. If the pool is available for import in read-only | |
2912 | * mode but not read-write mode, it is displayed as unavailable | |
2913 | * in userland with a special note that the pool is actually | |
2914 | * available for open in read-only mode. | |
2915 | * | |
2916 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
2917 | * missing a feature for write, we must first determine whether | |
2918 | * the pool can be opened read-only before returning to | |
2919 | * userland in order to know whether to display the | |
2920 | * abovementioned note. | |
2921 | */ | |
2922 | if (missing_feat_read || (missing_feat_write && | |
2923 | spa_writeable(spa))) { | |
2924 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2925 | ENOTSUP)); | |
2926 | } | |
b0bc7a84 MG |
2927 | |
2928 | /* | |
2929 | * Load refcounts for ZFS features from disk into an in-memory | |
2930 | * cache during SPA initialization. | |
2931 | */ | |
1c27024e | 2932 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { |
b0bc7a84 MG |
2933 | uint64_t refcount; |
2934 | ||
2935 | error = feature_get_refcount_from_disk(spa, | |
2936 | &spa_feature_table[i], &refcount); | |
2937 | if (error == 0) { | |
2938 | spa->spa_feat_refcount_cache[i] = refcount; | |
2939 | } else if (error == ENOTSUP) { | |
2940 | spa->spa_feat_refcount_cache[i] = | |
2941 | SPA_FEATURE_DISABLED; | |
2942 | } else { | |
2943 | return (spa_vdev_err(rvd, | |
2944 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
2945 | } | |
2946 | } | |
2947 | } | |
2948 | ||
2949 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
2950 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
9b67f605 | 2951 | &spa->spa_feat_enabled_txg_obj) != 0) |
b0bc7a84 | 2952 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
9ae529ec CS |
2953 | } |
2954 | ||
2955 | spa->spa_is_initializing = B_TRUE; | |
2956 | error = dsl_pool_open(spa->spa_dsl_pool); | |
2957 | spa->spa_is_initializing = B_FALSE; | |
2958 | if (error != 0) | |
2959 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2960 | ||
a1d477c2 | 2961 | if (!trust_config) { |
428870ff | 2962 | uint64_t hostid; |
a1d477c2 MA |
2963 | nvlist_t *policy = NULL; |
2964 | nvlist_t *mos_config; | |
428870ff | 2965 | |
a1d477c2 | 2966 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) |
428870ff BB |
2967 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
2968 | ||
a1d477c2 | 2969 | if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, |
b128c09f | 2970 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { |
34dc7c2f BB |
2971 | char *hostname; |
2972 | unsigned long myhostid = 0; | |
2973 | ||
a1d477c2 | 2974 | VERIFY(nvlist_lookup_string(mos_config, |
34dc7c2f BB |
2975 | ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); |
2976 | ||
379ca9cf OF |
2977 | myhostid = spa_get_hostid(); |
2978 | if (hostid && myhostid && hostid != myhostid) { | |
a1d477c2 | 2979 | nvlist_free(mos_config); |
2e528b49 | 2980 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
2981 | } |
2982 | } | |
428870ff BB |
2983 | if (nvlist_lookup_nvlist(spa->spa_config, |
2984 | ZPOOL_REWIND_POLICY, &policy) == 0) | |
a1d477c2 | 2985 | VERIFY(nvlist_add_nvlist(mos_config, |
428870ff | 2986 | ZPOOL_REWIND_POLICY, policy) == 0); |
34dc7c2f | 2987 | |
a1d477c2 | 2988 | spa_config_set(spa, mos_config); |
34dc7c2f BB |
2989 | spa_unload(spa); |
2990 | spa_deactivate(spa); | |
fb5f0bc8 | 2991 | spa_activate(spa, orig_mode); |
34dc7c2f | 2992 | |
428870ff | 2993 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); |
34dc7c2f BB |
2994 | } |
2995 | ||
3c67d83a TH |
2996 | /* Grab the checksum salt from the MOS. */ |
2997 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2998 | DMU_POOL_CHECKSUM_SALT, 1, | |
2999 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
3000 | spa->spa_cksum_salt.zcs_bytes); | |
3001 | if (error == ENOENT) { | |
3002 | /* Generate a new salt for subsequent use */ | |
3003 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
3004 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
3005 | } else if (error != 0) { | |
3006 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3007 | } | |
3008 | ||
428870ff BB |
3009 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) |
3010 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3011 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
3012 | if (error != 0) | |
3013 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
3014 | |
3015 | /* | |
3016 | * Load the bit that tells us to use the new accounting function | |
3017 | * (raid-z deflation). If we have an older pool, this will not | |
3018 | * be present. | |
3019 | */ | |
428870ff BB |
3020 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); |
3021 | if (error != 0 && error != ENOENT) | |
3022 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3023 | ||
3024 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
3025 | &spa->spa_creation_version); | |
3026 | if (error != 0 && error != ENOENT) | |
3027 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
3028 | |
3029 | /* | |
3030 | * Load the persistent error log. If we have an older pool, this will | |
3031 | * not be present. | |
3032 | */ | |
428870ff BB |
3033 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); |
3034 | if (error != 0 && error != ENOENT) | |
3035 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 3036 | |
428870ff BB |
3037 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, |
3038 | &spa->spa_errlog_scrub); | |
3039 | if (error != 0 && error != ENOENT) | |
3040 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
3041 | |
3042 | /* | |
3043 | * Load the history object. If we have an older pool, this | |
3044 | * will not be present. | |
3045 | */ | |
428870ff BB |
3046 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); |
3047 | if (error != 0 && error != ENOENT) | |
3048 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3049 | ||
e0ab3ab5 JS |
3050 | /* |
3051 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
3052 | * be present; in this case, defer its creation to a later time to | |
3053 | * avoid dirtying the MOS this early / out of sync context. See | |
3054 | * spa_sync_config_object. | |
3055 | */ | |
3056 | ||
3057 | /* The sentinel is only available in the MOS config. */ | |
1c27024e | 3058 | nvlist_t *mos_config; |
e0ab3ab5 JS |
3059 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) |
3060 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3061 | ||
3062 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
3063 | &spa->spa_all_vdev_zaps); | |
3064 | ||
38640550 DB |
3065 | if (error == ENOENT) { |
3066 | VERIFY(!nvlist_exists(mos_config, | |
3067 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
3068 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
3069 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
3070 | } else if (error != 0) { | |
e0ab3ab5 | 3071 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); |
38640550 | 3072 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { |
e0ab3ab5 JS |
3073 | /* |
3074 | * An older version of ZFS overwrote the sentinel value, so | |
3075 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
3076 | * destruction to later; see spa_sync_config_object. | |
3077 | */ | |
3078 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
3079 | /* | |
3080 | * We're assuming that no vdevs have had their ZAPs created | |
3081 | * before this. Better be sure of it. | |
3082 | */ | |
3083 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
3084 | } | |
3085 | nvlist_free(mos_config); | |
3086 | ||
428870ff BB |
3087 | /* |
3088 | * If we're assembling the pool from the split-off vdevs of | |
3089 | * an existing pool, we don't want to attach the spares & cache | |
3090 | * devices. | |
3091 | */ | |
34dc7c2f BB |
3092 | |
3093 | /* | |
3094 | * Load any hot spares for this pool. | |
3095 | */ | |
428870ff BB |
3096 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); |
3097 | if (error != 0 && error != ENOENT) | |
3098 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3099 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
3100 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); |
3101 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
428870ff BB |
3102 | &spa->spa_spares.sav_config) != 0) |
3103 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 3104 | |
b128c09f | 3105 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 3106 | spa_load_spares(spa); |
b128c09f | 3107 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
3108 | } else if (error == 0) { |
3109 | spa->spa_spares.sav_sync = B_TRUE; | |
34dc7c2f BB |
3110 | } |
3111 | ||
3112 | /* | |
3113 | * Load any level 2 ARC devices for this pool. | |
3114 | */ | |
428870ff | 3115 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, |
34dc7c2f | 3116 | &spa->spa_l2cache.sav_object); |
428870ff BB |
3117 | if (error != 0 && error != ENOENT) |
3118 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3119 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
34dc7c2f BB |
3120 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); |
3121 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
428870ff BB |
3122 | &spa->spa_l2cache.sav_config) != 0) |
3123 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f | 3124 | |
b128c09f | 3125 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 3126 | spa_load_l2cache(spa); |
b128c09f | 3127 | spa_config_exit(spa, SCL_ALL, FTAG); |
428870ff BB |
3128 | } else if (error == 0) { |
3129 | spa->spa_l2cache.sav_sync = B_TRUE; | |
b128c09f BB |
3130 | } |
3131 | ||
34dc7c2f BB |
3132 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); |
3133 | ||
428870ff BB |
3134 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); |
3135 | if (error && error != ENOENT) | |
3136 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
34dc7c2f BB |
3137 | |
3138 | if (error == 0) { | |
2dbedf54 | 3139 | uint64_t autoreplace = 0; |
428870ff BB |
3140 | |
3141 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
3142 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
3143 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
3144 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
3145 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
379ca9cf | 3146 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); |
428870ff BB |
3147 | spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, |
3148 | &spa->spa_dedup_ditto); | |
3149 | ||
3150 | spa->spa_autoreplace = (autoreplace != 0); | |
34dc7c2f BB |
3151 | } |
3152 | ||
379ca9cf OF |
3153 | /* |
3154 | * If the 'multihost' property is set, then never allow a pool to | |
3155 | * be imported when the system hostid is zero. The exception to | |
3156 | * this rule is zdb which is always allowed to access pools. | |
3157 | */ | |
3158 | if (spa_multihost(spa) && spa_get_hostid() == 0 && | |
3159 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { | |
3160 | fnvlist_add_uint64(spa->spa_load_info, | |
3161 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
3162 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
3163 | } | |
3164 | ||
34dc7c2f BB |
3165 | /* |
3166 | * If the 'autoreplace' property is set, then post a resource notifying | |
3167 | * the ZFS DE that it should not issue any faults for unopenable | |
3168 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
3169 | * unopenable vdevs so that the normal autoreplace handler can take | |
3170 | * over. | |
3171 | */ | |
428870ff | 3172 | if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { |
34dc7c2f | 3173 | spa_check_removed(spa->spa_root_vdev); |
428870ff BB |
3174 | /* |
3175 | * For the import case, this is done in spa_import(), because | |
3176 | * at this point we're using the spare definitions from | |
3177 | * the MOS config, not necessarily from the userland config. | |
3178 | */ | |
3179 | if (state != SPA_LOAD_IMPORT) { | |
3180 | spa_aux_check_removed(&spa->spa_spares); | |
3181 | spa_aux_check_removed(&spa->spa_l2cache); | |
3182 | } | |
3183 | } | |
34dc7c2f BB |
3184 | |
3185 | /* | |
3186 | * Load the vdev state for all toplevel vdevs. | |
3187 | */ | |
a1d477c2 MA |
3188 | error = vdev_load(rvd); |
3189 | if (error != 0) { | |
3190 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
3191 | } | |
3192 | ||
3193 | error = spa_condense_init(spa); | |
3194 | if (error != 0) { | |
3195 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); | |
3196 | } | |
34dc7c2f BB |
3197 | |
3198 | /* | |
3199 | * Propagate the leaf DTLs we just loaded all the way up the tree. | |
3200 | */ | |
b128c09f | 3201 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 3202 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE); |
b128c09f | 3203 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 3204 | |
428870ff BB |
3205 | /* |
3206 | * Load the DDTs (dedup tables). | |
3207 | */ | |
3208 | error = ddt_load(spa); | |
3209 | if (error != 0) | |
3210 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3211 | ||
3212 | spa_update_dspace(spa); | |
3213 | ||
a1d477c2 MA |
3214 | if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa) && |
3215 | spa_check_logs(spa)) { | |
3216 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
3217 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); | |
428870ff BB |
3218 | } |
3219 | ||
9ae529ec CS |
3220 | if (missing_feat_write) { |
3221 | ASSERT(state == SPA_LOAD_TRYIMPORT); | |
3222 | ||
3223 | /* | |
3224 | * At this point, we know that we can open the pool in | |
3225 | * read-only mode but not read-write mode. We now have enough | |
3226 | * information and can return to userland. | |
3227 | */ | |
3228 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); | |
3229 | } | |
3230 | ||
572e2857 BB |
3231 | /* |
3232 | * We've successfully opened the pool, verify that we're ready | |
3233 | * to start pushing transactions. | |
3234 | */ | |
3235 | if (state != SPA_LOAD_TRYIMPORT) { | |
c65aa5b2 | 3236 | if ((error = spa_load_verify(spa))) |
572e2857 BB |
3237 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, |
3238 | error)); | |
3239 | } | |
3240 | ||
428870ff BB |
3241 | if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || |
3242 | spa->spa_load_max_txg == UINT64_MAX)) { | |
34dc7c2f BB |
3243 | dmu_tx_t *tx; |
3244 | int need_update = B_FALSE; | |
9c43027b | 3245 | dsl_pool_t *dp = spa_get_dsl(spa); |
fb5f0bc8 | 3246 | |
a1d477c2 MA |
3247 | /* |
3248 | * We must check this before we start the sync thread, because | |
3249 | * we only want to start a condense thread for condense | |
3250 | * operations that were in progress when the pool was | |
3251 | * imported. Once we start syncing, spa_sync() could | |
3252 | * initiate a condense (and start a thread for it). In | |
3253 | * that case it would be wrong to start a second | |
3254 | * condense thread. | |
3255 | */ | |
3256 | boolean_t condense_in_progress = | |
3257 | (spa->spa_condensing_indirect != NULL); | |
3258 | ||
fb5f0bc8 | 3259 | ASSERT(state != SPA_LOAD_TRYIMPORT); |
34dc7c2f BB |
3260 | |
3261 | /* | |
3262 | * Claim log blocks that haven't been committed yet. | |
3263 | * This must all happen in a single txg. | |
428870ff BB |
3264 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), |
3265 | * invoked from zil_claim_log_block()'s i/o done callback. | |
3266 | * Price of rollback is that we abandon the log. | |
34dc7c2f | 3267 | */ |
428870ff BB |
3268 | spa->spa_claiming = B_TRUE; |
3269 | ||
9c43027b AJ |
3270 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); |
3271 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
34dc7c2f BB |
3272 | zil_claim, tx, DS_FIND_CHILDREN); |
3273 | dmu_tx_commit(tx); | |
3274 | ||
428870ff BB |
3275 | spa->spa_claiming = B_FALSE; |
3276 | ||
3277 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
34dc7c2f BB |
3278 | spa->spa_sync_on = B_TRUE; |
3279 | txg_sync_start(spa->spa_dsl_pool); | |
379ca9cf | 3280 | mmp_thread_start(spa); |
34dc7c2f BB |
3281 | |
3282 | /* | |
428870ff BB |
3283 | * Wait for all claims to sync. We sync up to the highest |
3284 | * claimed log block birth time so that claimed log blocks | |
3285 | * don't appear to be from the future. spa_claim_max_txg | |
3286 | * will have been set for us by either zil_check_log_chain() | |
3287 | * (invoked from spa_check_logs()) or zil_claim() above. | |
34dc7c2f | 3288 | */ |
428870ff | 3289 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); |
34dc7c2f BB |
3290 | |
3291 | /* | |
3292 | * If the config cache is stale, or we have uninitialized | |
3293 | * metaslabs (see spa_vdev_add()), then update the config. | |
45d1cae3 | 3294 | * |
572e2857 | 3295 | * If this is a verbatim import, trust the current |
45d1cae3 | 3296 | * in-core spa_config and update the disk labels. |
34dc7c2f BB |
3297 | */ |
3298 | if (config_cache_txg != spa->spa_config_txg || | |
572e2857 BB |
3299 | state == SPA_LOAD_IMPORT || |
3300 | state == SPA_LOAD_RECOVER || | |
3301 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) | |
34dc7c2f BB |
3302 | need_update = B_TRUE; |
3303 | ||
1c27024e | 3304 | for (int c = 0; c < rvd->vdev_children; c++) |
34dc7c2f BB |
3305 | if (rvd->vdev_child[c]->vdev_ms_array == 0) |
3306 | need_update = B_TRUE; | |
3307 | ||
3308 | /* | |
3309 | * Update the config cache asychronously in case we're the | |
3310 | * root pool, in which case the config cache isn't writable yet. | |
3311 | */ | |
3312 | if (need_update) | |
3313 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
fb5f0bc8 BB |
3314 | |
3315 | /* | |
3316 | * Check all DTLs to see if anything needs resilvering. | |
3317 | */ | |
428870ff BB |
3318 | if (!dsl_scan_resilvering(spa->spa_dsl_pool) && |
3319 | vdev_resilver_needed(rvd, NULL, NULL)) | |
fb5f0bc8 | 3320 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
428870ff | 3321 | |
6f1ffb06 MA |
3322 | /* |
3323 | * Log the fact that we booted up (so that we can detect if | |
3324 | * we rebooted in the middle of an operation). | |
3325 | */ | |
d5e024cb | 3326 | spa_history_log_version(spa, "open", NULL); |
6f1ffb06 | 3327 | |
428870ff BB |
3328 | /* |
3329 | * Delete any inconsistent datasets. | |
3330 | */ | |
3331 | (void) dmu_objset_find(spa_name(spa), | |
3332 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
3333 | ||
3334 | /* | |
3335 | * Clean up any stale temporary dataset userrefs. | |
3336 | */ | |
3337 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); | |
a1d477c2 MA |
3338 | |
3339 | /* | |
3340 | * Note: unlike condensing, we don't need an analogous | |
3341 | * "removal_in_progress" dance because no other thread | |
3342 | * can start a removal while we hold the spa_namespace_lock. | |
3343 | */ | |
3344 | spa_restart_removal(spa); | |
3345 | ||
3346 | if (condense_in_progress) | |
3347 | spa_condense_indirect_restart(spa); | |
34dc7c2f BB |
3348 | } |
3349 | ||
428870ff BB |
3350 | return (0); |
3351 | } | |
34dc7c2f | 3352 | |
428870ff BB |
3353 | static int |
3354 | spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) | |
3355 | { | |
572e2857 BB |
3356 | int mode = spa->spa_mode; |
3357 | ||
428870ff BB |
3358 | spa_unload(spa); |
3359 | spa_deactivate(spa); | |
3360 | ||
dea377c0 | 3361 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; |
428870ff | 3362 | |
572e2857 | 3363 | spa_activate(spa, mode); |
428870ff BB |
3364 | spa_async_suspend(spa); |
3365 | ||
3366 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); | |
3367 | } | |
3368 | ||
9ae529ec CS |
3369 | /* |
3370 | * If spa_load() fails this function will try loading prior txg's. If | |
3371 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
3372 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
3373 | * function will not rewind the pool and will return the same error as | |
3374 | * spa_load(). | |
3375 | */ | |
428870ff BB |
3376 | static int |
3377 | spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, | |
3378 | uint64_t max_request, int rewind_flags) | |
3379 | { | |
9ae529ec | 3380 | nvlist_t *loadinfo = NULL; |
428870ff BB |
3381 | nvlist_t *config = NULL; |
3382 | int load_error, rewind_error; | |
3383 | uint64_t safe_rewind_txg; | |
3384 | uint64_t min_txg; | |
3385 | ||
3386 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
3387 | spa->spa_load_max_txg = spa->spa_load_txg; | |
3388 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
3389 | } else { | |
3390 | spa->spa_load_max_txg = max_request; | |
dea377c0 MA |
3391 | if (max_request != UINT64_MAX) |
3392 | spa->spa_extreme_rewind = B_TRUE; | |
428870ff BB |
3393 | } |
3394 | ||
3395 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, | |
3396 | mosconfig); | |
3397 | if (load_error == 0) | |
3398 | return (0); | |
3399 | ||
3400 | if (spa->spa_root_vdev != NULL) | |
3401 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
3402 | ||
3403 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
3404 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
3405 | ||
3406 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
3407 | nvlist_free(config); | |
3408 | return (load_error); | |
3409 | } | |
3410 | ||
9ae529ec CS |
3411 | if (state == SPA_LOAD_RECOVER) { |
3412 | /* Price of rolling back is discarding txgs, including log */ | |
428870ff | 3413 | spa_set_log_state(spa, SPA_LOG_CLEAR); |
9ae529ec CS |
3414 | } else { |
3415 | /* | |
3416 | * If we aren't rolling back save the load info from our first | |
3417 | * import attempt so that we can restore it after attempting | |
3418 | * to rewind. | |
3419 | */ | |
3420 | loadinfo = spa->spa_load_info; | |
3421 | spa->spa_load_info = fnvlist_alloc(); | |
3422 | } | |
428870ff BB |
3423 | |
3424 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
3425 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
3426 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
3427 | TXG_INITIAL : safe_rewind_txg; | |
3428 | ||
3429 | /* | |
3430 | * Continue as long as we're finding errors, we're still within | |
3431 | * the acceptable rewind range, and we're still finding uberblocks | |
3432 | */ | |
3433 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
3434 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
3435 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
3436 | spa->spa_extreme_rewind = B_TRUE; | |
3437 | rewind_error = spa_load_retry(spa, state, mosconfig); | |
3438 | } | |
3439 | ||
428870ff BB |
3440 | spa->spa_extreme_rewind = B_FALSE; |
3441 | spa->spa_load_max_txg = UINT64_MAX; | |
3442 | ||
3443 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
3444 | spa_config_set(spa, config); | |
ee6370a7 | 3445 | else |
3446 | nvlist_free(config); | |
428870ff | 3447 | |
9ae529ec CS |
3448 | if (state == SPA_LOAD_RECOVER) { |
3449 | ASSERT3P(loadinfo, ==, NULL); | |
3450 | return (rewind_error); | |
3451 | } else { | |
3452 | /* Store the rewind info as part of the initial load info */ | |
3453 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
3454 | spa->spa_load_info); | |
3455 | ||
3456 | /* Restore the initial load info */ | |
3457 | fnvlist_free(spa->spa_load_info); | |
3458 | spa->spa_load_info = loadinfo; | |
3459 | ||
3460 | return (load_error); | |
3461 | } | |
34dc7c2f BB |
3462 | } |
3463 | ||
3464 | /* | |
3465 | * Pool Open/Import | |
3466 | * | |
3467 | * The import case is identical to an open except that the configuration is sent | |
3468 | * down from userland, instead of grabbed from the configuration cache. For the | |
3469 | * case of an open, the pool configuration will exist in the | |
3470 | * POOL_STATE_UNINITIALIZED state. | |
3471 | * | |
3472 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
3473 | * the same time open the pool, without having to keep around the spa_t in some | |
3474 | * ambiguous state. | |
3475 | */ | |
3476 | static int | |
428870ff BB |
3477 | spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, |
3478 | nvlist_t **config) | |
34dc7c2f BB |
3479 | { |
3480 | spa_t *spa; | |
572e2857 | 3481 | spa_load_state_t state = SPA_LOAD_OPEN; |
34dc7c2f | 3482 | int error; |
34dc7c2f | 3483 | int locked = B_FALSE; |
526af785 | 3484 | int firstopen = B_FALSE; |
34dc7c2f BB |
3485 | |
3486 | *spapp = NULL; | |
3487 | ||
3488 | /* | |
3489 | * As disgusting as this is, we need to support recursive calls to this | |
3490 | * function because dsl_dir_open() is called during spa_load(), and ends | |
3491 | * up calling spa_open() again. The real fix is to figure out how to | |
3492 | * avoid dsl_dir_open() calling this in the first place. | |
3493 | */ | |
c25b8f99 | 3494 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { |
34dc7c2f BB |
3495 | mutex_enter(&spa_namespace_lock); |
3496 | locked = B_TRUE; | |
3497 | } | |
3498 | ||
3499 | if ((spa = spa_lookup(pool)) == NULL) { | |
3500 | if (locked) | |
3501 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 3502 | return (SET_ERROR(ENOENT)); |
34dc7c2f | 3503 | } |
428870ff | 3504 | |
34dc7c2f | 3505 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { |
428870ff BB |
3506 | zpool_rewind_policy_t policy; |
3507 | ||
526af785 PJD |
3508 | firstopen = B_TRUE; |
3509 | ||
428870ff BB |
3510 | zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, |
3511 | &policy); | |
3512 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
3513 | state = SPA_LOAD_RECOVER; | |
34dc7c2f | 3514 | |
fb5f0bc8 | 3515 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 3516 | |
428870ff BB |
3517 | if (state != SPA_LOAD_RECOVER) |
3518 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
3519 | ||
3520 | error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, | |
3521 | policy.zrp_request); | |
34dc7c2f BB |
3522 | |
3523 | if (error == EBADF) { | |
3524 | /* | |
3525 | * If vdev_validate() returns failure (indicated by | |
3526 | * EBADF), it indicates that one of the vdevs indicates | |
3527 | * that the pool has been exported or destroyed. If | |
3528 | * this is the case, the config cache is out of sync and | |
3529 | * we should remove the pool from the namespace. | |
3530 | */ | |
34dc7c2f BB |
3531 | spa_unload(spa); |
3532 | spa_deactivate(spa); | |
a1d477c2 | 3533 | spa_write_cachefile(spa, B_TRUE, B_TRUE); |
34dc7c2f | 3534 | spa_remove(spa); |
34dc7c2f BB |
3535 | if (locked) |
3536 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 3537 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
3538 | } |
3539 | ||
3540 | if (error) { | |
3541 | /* | |
3542 | * We can't open the pool, but we still have useful | |
3543 | * information: the state of each vdev after the | |
3544 | * attempted vdev_open(). Return this to the user. | |
3545 | */ | |
572e2857 | 3546 | if (config != NULL && spa->spa_config) { |
428870ff | 3547 | VERIFY(nvlist_dup(spa->spa_config, config, |
79c76d5b | 3548 | KM_SLEEP) == 0); |
572e2857 BB |
3549 | VERIFY(nvlist_add_nvlist(*config, |
3550 | ZPOOL_CONFIG_LOAD_INFO, | |
3551 | spa->spa_load_info) == 0); | |
3552 | } | |
34dc7c2f BB |
3553 | spa_unload(spa); |
3554 | spa_deactivate(spa); | |
428870ff | 3555 | spa->spa_last_open_failed = error; |
34dc7c2f BB |
3556 | if (locked) |
3557 | mutex_exit(&spa_namespace_lock); | |
3558 | *spapp = NULL; | |
3559 | return (error); | |
34dc7c2f | 3560 | } |
34dc7c2f BB |
3561 | } |
3562 | ||
3563 | spa_open_ref(spa, tag); | |
3564 | ||
b128c09f | 3565 | if (config != NULL) |
34dc7c2f | 3566 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
34dc7c2f | 3567 | |
572e2857 BB |
3568 | /* |
3569 | * If we've recovered the pool, pass back any information we | |
3570 | * gathered while doing the load. | |
3571 | */ | |
3572 | if (state == SPA_LOAD_RECOVER) { | |
3573 | VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, | |
3574 | spa->spa_load_info) == 0); | |
3575 | } | |
3576 | ||
428870ff BB |
3577 | if (locked) { |
3578 | spa->spa_last_open_failed = 0; | |
3579 | spa->spa_last_ubsync_txg = 0; | |
3580 | spa->spa_load_txg = 0; | |
3581 | mutex_exit(&spa_namespace_lock); | |
3582 | } | |
3583 | ||
526af785 | 3584 | if (firstopen) |
a0bd735a | 3585 | zvol_create_minors(spa, spa_name(spa), B_TRUE); |
526af785 | 3586 | |
428870ff BB |
3587 | *spapp = spa; |
3588 | ||
34dc7c2f BB |
3589 | return (0); |
3590 | } | |
3591 | ||
428870ff BB |
3592 | int |
3593 | spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, | |
3594 | nvlist_t **config) | |
3595 | { | |
3596 | return (spa_open_common(name, spapp, tag, policy, config)); | |
3597 | } | |
3598 | ||
34dc7c2f BB |
3599 | int |
3600 | spa_open(const char *name, spa_t **spapp, void *tag) | |
3601 | { | |
428870ff | 3602 | return (spa_open_common(name, spapp, tag, NULL, NULL)); |
34dc7c2f BB |
3603 | } |
3604 | ||
3605 | /* | |
3606 | * Lookup the given spa_t, incrementing the inject count in the process, | |
3607 | * preventing it from being exported or destroyed. | |
3608 | */ | |
3609 | spa_t * | |
3610 | spa_inject_addref(char *name) | |
3611 | { | |
3612 | spa_t *spa; | |
3613 | ||
3614 | mutex_enter(&spa_namespace_lock); | |
3615 | if ((spa = spa_lookup(name)) == NULL) { | |
3616 | mutex_exit(&spa_namespace_lock); | |
3617 | return (NULL); | |
3618 | } | |
3619 | spa->spa_inject_ref++; | |
3620 | mutex_exit(&spa_namespace_lock); | |
3621 | ||
3622 | return (spa); | |
3623 | } | |
3624 | ||
3625 | void | |
3626 | spa_inject_delref(spa_t *spa) | |
3627 | { | |
3628 | mutex_enter(&spa_namespace_lock); | |
3629 | spa->spa_inject_ref--; | |
3630 | mutex_exit(&spa_namespace_lock); | |
3631 | } | |
3632 | ||
3633 | /* | |
3634 | * Add spares device information to the nvlist. | |
3635 | */ | |
3636 | static void | |
3637 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
3638 | { | |
3639 | nvlist_t **spares; | |
3640 | uint_t i, nspares; | |
3641 | nvlist_t *nvroot; | |
3642 | uint64_t guid; | |
3643 | vdev_stat_t *vs; | |
3644 | uint_t vsc; | |
3645 | uint64_t pool; | |
3646 | ||
9babb374 BB |
3647 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
3648 | ||
34dc7c2f BB |
3649 | if (spa->spa_spares.sav_count == 0) |
3650 | return; | |
3651 | ||
3652 | VERIFY(nvlist_lookup_nvlist(config, | |
3653 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3654 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
3655 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3656 | if (nspares != 0) { | |
3657 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3658 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
3659 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3660 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3661 | ||
3662 | /* | |
3663 | * Go through and find any spares which have since been | |
3664 | * repurposed as an active spare. If this is the case, update | |
3665 | * their status appropriately. | |
3666 | */ | |
3667 | for (i = 0; i < nspares; i++) { | |
3668 | VERIFY(nvlist_lookup_uint64(spares[i], | |
3669 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
b128c09f BB |
3670 | if (spa_spare_exists(guid, &pool, NULL) && |
3671 | pool != 0ULL) { | |
34dc7c2f | 3672 | VERIFY(nvlist_lookup_uint64_array( |
428870ff | 3673 | spares[i], ZPOOL_CONFIG_VDEV_STATS, |
34dc7c2f BB |
3674 | (uint64_t **)&vs, &vsc) == 0); |
3675 | vs->vs_state = VDEV_STATE_CANT_OPEN; | |
3676 | vs->vs_aux = VDEV_AUX_SPARED; | |
3677 | } | |
3678 | } | |
3679 | } | |
3680 | } | |
3681 | ||
3682 | /* | |
3683 | * Add l2cache device information to the nvlist, including vdev stats. | |
3684 | */ | |
3685 | static void | |
3686 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
3687 | { | |
3688 | nvlist_t **l2cache; | |
3689 | uint_t i, j, nl2cache; | |
3690 | nvlist_t *nvroot; | |
3691 | uint64_t guid; | |
3692 | vdev_t *vd; | |
3693 | vdev_stat_t *vs; | |
3694 | uint_t vsc; | |
3695 | ||
9babb374 BB |
3696 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); |
3697 | ||
34dc7c2f BB |
3698 | if (spa->spa_l2cache.sav_count == 0) |
3699 | return; | |
3700 | ||
34dc7c2f BB |
3701 | VERIFY(nvlist_lookup_nvlist(config, |
3702 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3703 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
3704 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3705 | if (nl2cache != 0) { | |
3706 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3707 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
3708 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3709 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3710 | ||
3711 | /* | |
3712 | * Update level 2 cache device stats. | |
3713 | */ | |
3714 | ||
3715 | for (i = 0; i < nl2cache; i++) { | |
3716 | VERIFY(nvlist_lookup_uint64(l2cache[i], | |
3717 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
3718 | ||
3719 | vd = NULL; | |
3720 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
3721 | if (guid == | |
3722 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
3723 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
3724 | break; | |
3725 | } | |
3726 | } | |
3727 | ASSERT(vd != NULL); | |
3728 | ||
3729 | VERIFY(nvlist_lookup_uint64_array(l2cache[i], | |
428870ff BB |
3730 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) |
3731 | == 0); | |
34dc7c2f | 3732 | vdev_get_stats(vd, vs); |
193a37cb TH |
3733 | vdev_config_generate_stats(vd, l2cache[i]); |
3734 | ||
34dc7c2f BB |
3735 | } |
3736 | } | |
34dc7c2f BB |
3737 | } |
3738 | ||
9ae529ec | 3739 | static void |
417104bd | 3740 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) |
9ae529ec | 3741 | { |
9ae529ec CS |
3742 | zap_cursor_t zc; |
3743 | zap_attribute_t za; | |
3744 | ||
9ae529ec CS |
3745 | if (spa->spa_feat_for_read_obj != 0) { |
3746 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3747 | spa->spa_feat_for_read_obj); | |
3748 | zap_cursor_retrieve(&zc, &za) == 0; | |
3749 | zap_cursor_advance(&zc)) { | |
3750 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3751 | za.za_num_integers == 1); | |
417104bd | 3752 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
3753 | za.za_first_integer)); |
3754 | } | |
3755 | zap_cursor_fini(&zc); | |
3756 | } | |
3757 | ||
3758 | if (spa->spa_feat_for_write_obj != 0) { | |
3759 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3760 | spa->spa_feat_for_write_obj); | |
3761 | zap_cursor_retrieve(&zc, &za) == 0; | |
3762 | zap_cursor_advance(&zc)) { | |
3763 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3764 | za.za_num_integers == 1); | |
417104bd | 3765 | VERIFY0(nvlist_add_uint64(features, za.za_name, |
9ae529ec CS |
3766 | za.za_first_integer)); |
3767 | } | |
3768 | zap_cursor_fini(&zc); | |
3769 | } | |
417104bd NB |
3770 | } |
3771 | ||
3772 | static void | |
3773 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
3774 | { | |
3775 | int i; | |
3776 | ||
3777 | for (i = 0; i < SPA_FEATURES; i++) { | |
3778 | zfeature_info_t feature = spa_feature_table[i]; | |
3779 | uint64_t refcount; | |
3780 | ||
3781 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
3782 | continue; | |
3783 | ||
3784 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
3785 | } | |
3786 | } | |
3787 | ||
3788 | /* | |
3789 | * Store a list of pool features and their reference counts in the | |
3790 | * config. | |
3791 | * | |
3792 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
3793 | * the pool features and reference counts from disk, then save the list | |
3794 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
3795 | * and refresh its values from the cached reference counts. This | |
3796 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
3797 | * clear' can resume the pool. | |
3798 | */ | |
3799 | static void | |
3800 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
3801 | { | |
4eb30c68 | 3802 | nvlist_t *features; |
417104bd NB |
3803 | |
3804 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3805 | ||
4eb30c68 NB |
3806 | mutex_enter(&spa->spa_feat_stats_lock); |
3807 | features = spa->spa_feat_stats; | |
3808 | ||
417104bd NB |
3809 | if (features != NULL) { |
3810 | spa_feature_stats_from_cache(spa, features); | |
3811 | } else { | |
3812 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
3813 | spa->spa_feat_stats = features; | |
3814 | spa_feature_stats_from_disk(spa, features); | |
3815 | } | |
9ae529ec | 3816 | |
417104bd NB |
3817 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, |
3818 | features)); | |
4eb30c68 NB |
3819 | |
3820 | mutex_exit(&spa->spa_feat_stats_lock); | |
9ae529ec CS |
3821 | } |
3822 | ||
34dc7c2f | 3823 | int |
9ae529ec CS |
3824 | spa_get_stats(const char *name, nvlist_t **config, |
3825 | char *altroot, size_t buflen) | |
34dc7c2f BB |
3826 | { |
3827 | int error; | |
3828 | spa_t *spa; | |
3829 | ||
3830 | *config = NULL; | |
428870ff | 3831 | error = spa_open_common(name, &spa, FTAG, NULL, config); |
34dc7c2f | 3832 | |
9babb374 BB |
3833 | if (spa != NULL) { |
3834 | /* | |
3835 | * This still leaves a window of inconsistency where the spares | |
3836 | * or l2cache devices could change and the config would be | |
3837 | * self-inconsistent. | |
3838 | */ | |
3839 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
34dc7c2f | 3840 | |
9babb374 | 3841 | if (*config != NULL) { |
572e2857 BB |
3842 | uint64_t loadtimes[2]; |
3843 | ||
3844 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
3845 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
3846 | VERIFY(nvlist_add_uint64_array(*config, | |
3847 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); | |
3848 | ||
b128c09f | 3849 | VERIFY(nvlist_add_uint64(*config, |
9babb374 BB |
3850 | ZPOOL_CONFIG_ERRCOUNT, |
3851 | spa_get_errlog_size(spa)) == 0); | |
3852 | ||
cec3a0a1 | 3853 | if (spa_suspended(spa)) { |
9babb374 BB |
3854 | VERIFY(nvlist_add_uint64(*config, |
3855 | ZPOOL_CONFIG_SUSPENDED, | |
3856 | spa->spa_failmode) == 0); | |
cec3a0a1 OF |
3857 | VERIFY(nvlist_add_uint64(*config, |
3858 | ZPOOL_CONFIG_SUSPENDED_REASON, | |
3859 | spa->spa_suspended) == 0); | |
3860 | } | |
b128c09f | 3861 | |
9babb374 BB |
3862 | spa_add_spares(spa, *config); |
3863 | spa_add_l2cache(spa, *config); | |
9ae529ec | 3864 | spa_add_feature_stats(spa, *config); |
9babb374 | 3865 | } |
34dc7c2f BB |
3866 | } |
3867 | ||
3868 | /* | |
3869 | * We want to get the alternate root even for faulted pools, so we cheat | |
3870 | * and call spa_lookup() directly. | |
3871 | */ | |
3872 | if (altroot) { | |
3873 | if (spa == NULL) { | |
3874 | mutex_enter(&spa_namespace_lock); | |
3875 | spa = spa_lookup(name); | |
3876 | if (spa) | |
3877 | spa_altroot(spa, altroot, buflen); | |
3878 | else | |
3879 | altroot[0] = '\0'; | |
3880 | spa = NULL; | |
3881 | mutex_exit(&spa_namespace_lock); | |
3882 | } else { | |
3883 | spa_altroot(spa, altroot, buflen); | |
3884 | } | |
3885 | } | |
3886 | ||
9babb374 BB |
3887 | if (spa != NULL) { |
3888 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
34dc7c2f | 3889 | spa_close(spa, FTAG); |
9babb374 | 3890 | } |
34dc7c2f BB |
3891 | |
3892 | return (error); | |
3893 | } | |
3894 | ||
3895 | /* | |
3896 | * Validate that the auxiliary device array is well formed. We must have an | |
3897 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
3898 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
3899 | * specified, as long as they are well-formed. | |
3900 | */ | |
3901 | static int | |
3902 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
3903 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
3904 | vdev_labeltype_t label) | |
3905 | { | |
3906 | nvlist_t **dev; | |
3907 | uint_t i, ndev; | |
3908 | vdev_t *vd; | |
3909 | int error; | |
3910 | ||
b128c09f BB |
3911 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
3912 | ||
34dc7c2f BB |
3913 | /* |
3914 | * It's acceptable to have no devs specified. | |
3915 | */ | |
3916 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
3917 | return (0); | |
3918 | ||
3919 | if (ndev == 0) | |
2e528b49 | 3920 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
3921 | |
3922 | /* | |
3923 | * Make sure the pool is formatted with a version that supports this | |
3924 | * device type. | |
3925 | */ | |
3926 | if (spa_version(spa) < version) | |
2e528b49 | 3927 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
3928 | |
3929 | /* | |
3930 | * Set the pending device list so we correctly handle device in-use | |
3931 | * checking. | |
3932 | */ | |
3933 | sav->sav_pending = dev; | |
3934 | sav->sav_npending = ndev; | |
3935 | ||
3936 | for (i = 0; i < ndev; i++) { | |
3937 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
3938 | mode)) != 0) | |
3939 | goto out; | |
3940 | ||
3941 | if (!vd->vdev_ops->vdev_op_leaf) { | |
3942 | vdev_free(vd); | |
2e528b49 | 3943 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
3944 | goto out; |
3945 | } | |
3946 | ||
34dc7c2f BB |
3947 | vd->vdev_top = vd; |
3948 | ||
3949 | if ((error = vdev_open(vd)) == 0 && | |
3950 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
3951 | VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, | |
3952 | vd->vdev_guid) == 0); | |
3953 | } | |
3954 | ||
3955 | vdev_free(vd); | |
3956 | ||
3957 | if (error && | |
3958 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
3959 | goto out; | |
3960 | else | |
3961 | error = 0; | |
3962 | } | |
3963 | ||
3964 | out: | |
3965 | sav->sav_pending = NULL; | |
3966 | sav->sav_npending = 0; | |
3967 | return (error); | |
3968 | } | |
3969 | ||
3970 | static int | |
3971 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
3972 | { | |
3973 | int error; | |
3974 | ||
b128c09f BB |
3975 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
3976 | ||
34dc7c2f BB |
3977 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, |
3978 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
3979 | VDEV_LABEL_SPARE)) != 0) { | |
3980 | return (error); | |
3981 | } | |
3982 | ||
3983 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
3984 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
3985 | VDEV_LABEL_L2CACHE)); | |
3986 | } | |
3987 | ||
3988 | static void | |
3989 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
3990 | const char *config) | |
3991 | { | |
3992 | int i; | |
3993 | ||
3994 | if (sav->sav_config != NULL) { | |
3995 | nvlist_t **olddevs; | |
3996 | uint_t oldndevs; | |
3997 | nvlist_t **newdevs; | |
3998 | ||
3999 | /* | |
4e33ba4c | 4000 | * Generate new dev list by concatenating with the |
34dc7c2f BB |
4001 | * current dev list. |
4002 | */ | |
4003 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, | |
4004 | &olddevs, &oldndevs) == 0); | |
4005 | ||
4006 | newdevs = kmem_alloc(sizeof (void *) * | |
79c76d5b | 4007 | (ndevs + oldndevs), KM_SLEEP); |
34dc7c2f BB |
4008 | for (i = 0; i < oldndevs; i++) |
4009 | VERIFY(nvlist_dup(olddevs[i], &newdevs[i], | |
79c76d5b | 4010 | KM_SLEEP) == 0); |
34dc7c2f BB |
4011 | for (i = 0; i < ndevs; i++) |
4012 | VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], | |
79c76d5b | 4013 | KM_SLEEP) == 0); |
34dc7c2f BB |
4014 | |
4015 | VERIFY(nvlist_remove(sav->sav_config, config, | |
4016 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
4017 | ||
4018 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
4019 | config, newdevs, ndevs + oldndevs) == 0); | |
4020 | for (i = 0; i < oldndevs + ndevs; i++) | |
4021 | nvlist_free(newdevs[i]); | |
4022 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
4023 | } else { | |
4024 | /* | |
4025 | * Generate a new dev list. | |
4026 | */ | |
4027 | VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, | |
79c76d5b | 4028 | KM_SLEEP) == 0); |
34dc7c2f BB |
4029 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, |
4030 | devs, ndevs) == 0); | |
4031 | } | |
4032 | } | |
4033 | ||
4034 | /* | |
4035 | * Stop and drop level 2 ARC devices | |
4036 | */ | |
4037 | void | |
4038 | spa_l2cache_drop(spa_t *spa) | |
4039 | { | |
4040 | vdev_t *vd; | |
4041 | int i; | |
4042 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
4043 | ||
4044 | for (i = 0; i < sav->sav_count; i++) { | |
4045 | uint64_t pool; | |
4046 | ||
4047 | vd = sav->sav_vdevs[i]; | |
4048 | ASSERT(vd != NULL); | |
4049 | ||
fb5f0bc8 BB |
4050 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && |
4051 | pool != 0ULL && l2arc_vdev_present(vd)) | |
34dc7c2f | 4052 | l2arc_remove_vdev(vd); |
34dc7c2f BB |
4053 | } |
4054 | } | |
4055 | ||
b5256303 TC |
4056 | /* |
4057 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
4058 | * have the encryption feature flag enabled. | |
4059 | */ | |
4060 | static int | |
4061 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
4062 | boolean_t has_encryption) | |
4063 | { | |
4064 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
4065 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
4066 | !has_encryption) | |
4067 | return (SET_ERROR(ENOTSUP)); | |
4068 | ||
4069 | return (dmu_objset_create_crypt_check(NULL, dcp)); | |
4070 | } | |
4071 | ||
34dc7c2f BB |
4072 | /* |
4073 | * Pool Creation | |
4074 | */ | |
4075 | int | |
4076 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
b5256303 | 4077 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) |
34dc7c2f BB |
4078 | { |
4079 | spa_t *spa; | |
4080 | char *altroot = NULL; | |
4081 | vdev_t *rvd; | |
4082 | dsl_pool_t *dp; | |
4083 | dmu_tx_t *tx; | |
9babb374 | 4084 | int error = 0; |
34dc7c2f BB |
4085 | uint64_t txg = TXG_INITIAL; |
4086 | nvlist_t **spares, **l2cache; | |
4087 | uint_t nspares, nl2cache; | |
b5256303 | 4088 | uint64_t version, obj, root_dsobj = 0; |
9ae529ec | 4089 | boolean_t has_features; |
b5256303 TC |
4090 | boolean_t has_encryption; |
4091 | spa_feature_t feat; | |
4092 | char *feat_name; | |
83e9986f RY |
4093 | char *poolname; |
4094 | nvlist_t *nvl; | |
4095 | ||
4096 | if (nvlist_lookup_string(props, "tname", &poolname) != 0) | |
4097 | poolname = (char *)pool; | |
34dc7c2f BB |
4098 | |
4099 | /* | |
4100 | * If this pool already exists, return failure. | |
4101 | */ | |
4102 | mutex_enter(&spa_namespace_lock); | |
83e9986f | 4103 | if (spa_lookup(poolname) != NULL) { |
34dc7c2f | 4104 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 4105 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
4106 | } |
4107 | ||
4108 | /* | |
4109 | * Allocate a new spa_t structure. | |
4110 | */ | |
83e9986f RY |
4111 | nvl = fnvlist_alloc(); |
4112 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
34dc7c2f BB |
4113 | (void) nvlist_lookup_string(props, |
4114 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
83e9986f RY |
4115 | spa = spa_add(poolname, nvl, altroot); |
4116 | fnvlist_free(nvl); | |
fb5f0bc8 | 4117 | spa_activate(spa, spa_mode_global); |
34dc7c2f | 4118 | |
34dc7c2f | 4119 | if (props && (error = spa_prop_validate(spa, props))) { |
34dc7c2f BB |
4120 | spa_deactivate(spa); |
4121 | spa_remove(spa); | |
b128c09f | 4122 | mutex_exit(&spa_namespace_lock); |
34dc7c2f BB |
4123 | return (error); |
4124 | } | |
4125 | ||
83e9986f RY |
4126 | /* |
4127 | * Temporary pool names should never be written to disk. | |
4128 | */ | |
4129 | if (poolname != pool) | |
4130 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
4131 | ||
9ae529ec | 4132 | has_features = B_FALSE; |
b5256303 | 4133 | has_encryption = B_FALSE; |
1c27024e | 4134 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); |
9ae529ec | 4135 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { |
b5256303 | 4136 | if (zpool_prop_feature(nvpair_name(elem))) { |
9ae529ec | 4137 | has_features = B_TRUE; |
b5256303 TC |
4138 | |
4139 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
4140 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
4141 | if (feat == SPA_FEATURE_ENCRYPTION) | |
4142 | has_encryption = B_TRUE; | |
4143 | } | |
4144 | } | |
4145 | ||
4146 | /* verify encryption params, if they were provided */ | |
4147 | if (dcp != NULL) { | |
4148 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
4149 | if (error != 0) { | |
4150 | spa_deactivate(spa); | |
4151 | spa_remove(spa); | |
4152 | mutex_exit(&spa_namespace_lock); | |
4153 | return (error); | |
4154 | } | |
9ae529ec CS |
4155 | } |
4156 | ||
4157 | if (has_features || nvlist_lookup_uint64(props, | |
4158 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
34dc7c2f | 4159 | version = SPA_VERSION; |
9ae529ec CS |
4160 | } |
4161 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
428870ff BB |
4162 | |
4163 | spa->spa_first_txg = txg; | |
4164 | spa->spa_uberblock.ub_txg = txg - 1; | |
34dc7c2f BB |
4165 | spa->spa_uberblock.ub_version = version; |
4166 | spa->spa_ubsync = spa->spa_uberblock; | |
3dfb57a3 | 4167 | spa->spa_load_state = SPA_LOAD_CREATE; |
a1d477c2 MA |
4168 | spa->spa_removing_phys.sr_state = DSS_NONE; |
4169 | spa->spa_removing_phys.sr_removing_vdev = -1; | |
4170 | spa->spa_removing_phys.sr_prev_indirect_vdev = -1; | |
34dc7c2f | 4171 | |
9babb374 BB |
4172 | /* |
4173 | * Create "The Godfather" zio to hold all async IOs | |
4174 | */ | |
e022864d MA |
4175 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), |
4176 | KM_SLEEP); | |
1c27024e | 4177 | for (int i = 0; i < max_ncpus; i++) { |
e022864d MA |
4178 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, |
4179 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
4180 | ZIO_FLAG_GODFATHER); | |
4181 | } | |
9babb374 | 4182 | |
34dc7c2f BB |
4183 | /* |
4184 | * Create the root vdev. | |
4185 | */ | |
b128c09f | 4186 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
4187 | |
4188 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
4189 | ||
4190 | ASSERT(error != 0 || rvd != NULL); | |
4191 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
4192 | ||
4193 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
2e528b49 | 4194 | error = SET_ERROR(EINVAL); |
34dc7c2f BB |
4195 | |
4196 | if (error == 0 && | |
4197 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
4198 | (error = spa_validate_aux(spa, nvroot, txg, | |
4199 | VDEV_ALLOC_ADD)) == 0) { | |
1c27024e | 4200 | for (int c = 0; c < rvd->vdev_children; c++) { |
9babb374 BB |
4201 | vdev_metaslab_set_size(rvd->vdev_child[c]); |
4202 | vdev_expand(rvd->vdev_child[c], txg); | |
4203 | } | |
34dc7c2f BB |
4204 | } |
4205 | ||
b128c09f | 4206 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4207 | |
4208 | if (error != 0) { | |
4209 | spa_unload(spa); | |
4210 | spa_deactivate(spa); | |
4211 | spa_remove(spa); | |
4212 | mutex_exit(&spa_namespace_lock); | |
4213 | return (error); | |
4214 | } | |
4215 | ||
4216 | /* | |
4217 | * Get the list of spares, if specified. | |
4218 | */ | |
4219 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
4220 | &spares, &nspares) == 0) { | |
4221 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, | |
79c76d5b | 4222 | KM_SLEEP) == 0); |
34dc7c2f BB |
4223 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, |
4224 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
b128c09f | 4225 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4226 | spa_load_spares(spa); |
b128c09f | 4227 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4228 | spa->spa_spares.sav_sync = B_TRUE; |
4229 | } | |
4230 | ||
4231 | /* | |
4232 | * Get the list of level 2 cache devices, if specified. | |
4233 | */ | |
4234 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
4235 | &l2cache, &nl2cache) == 0) { | |
4236 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
79c76d5b | 4237 | NV_UNIQUE_NAME, KM_SLEEP) == 0); |
34dc7c2f BB |
4238 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, |
4239 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
b128c09f | 4240 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4241 | spa_load_l2cache(spa); |
b128c09f | 4242 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4243 | spa->spa_l2cache.sav_sync = B_TRUE; |
4244 | } | |
4245 | ||
9ae529ec | 4246 | spa->spa_is_initializing = B_TRUE; |
b5256303 | 4247 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); |
9ae529ec | 4248 | spa->spa_is_initializing = B_FALSE; |
34dc7c2f | 4249 | |
428870ff BB |
4250 | /* |
4251 | * Create DDTs (dedup tables). | |
4252 | */ | |
4253 | ddt_create(spa); | |
4254 | ||
4255 | spa_update_dspace(spa); | |
4256 | ||
34dc7c2f BB |
4257 | tx = dmu_tx_create_assigned(dp, txg); |
4258 | ||
d5e024cb BB |
4259 | /* |
4260 | * Create the pool's history object. | |
4261 | */ | |
4262 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
4263 | spa_history_create_obj(spa, tx); | |
4264 | ||
4265 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
4266 | spa_history_log_version(spa, "create", tx); | |
4267 | ||
34dc7c2f BB |
4268 | /* |
4269 | * Create the pool config object. | |
4270 | */ | |
4271 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
b128c09f | 4272 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, |
34dc7c2f BB |
4273 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); |
4274 | ||
4275 | if (zap_add(spa->spa_meta_objset, | |
4276 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
4277 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
4278 | cmn_err(CE_PANIC, "failed to add pool config"); | |
4279 | } | |
4280 | ||
428870ff BB |
4281 | if (zap_add(spa->spa_meta_objset, |
4282 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
4283 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
4284 | cmn_err(CE_PANIC, "failed to add pool version"); | |
4285 | } | |
4286 | ||
34dc7c2f BB |
4287 | /* Newly created pools with the right version are always deflated. */ |
4288 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
4289 | spa->spa_deflate = TRUE; | |
4290 | if (zap_add(spa->spa_meta_objset, | |
4291 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
4292 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
4293 | cmn_err(CE_PANIC, "failed to add deflate"); | |
4294 | } | |
4295 | } | |
4296 | ||
4297 | /* | |
428870ff | 4298 | * Create the deferred-free bpobj. Turn off compression |
34dc7c2f BB |
4299 | * because sync-to-convergence takes longer if the blocksize |
4300 | * keeps changing. | |
4301 | */ | |
428870ff BB |
4302 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); |
4303 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
34dc7c2f | 4304 | ZIO_COMPRESS_OFF, tx); |
34dc7c2f | 4305 | if (zap_add(spa->spa_meta_objset, |
428870ff BB |
4306 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, |
4307 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
4308 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
34dc7c2f | 4309 | } |
428870ff BB |
4310 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, |
4311 | spa->spa_meta_objset, obj)); | |
34dc7c2f | 4312 | |
3c67d83a TH |
4313 | /* |
4314 | * Generate some random noise for salted checksums to operate on. | |
4315 | */ | |
4316 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4317 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4318 | ||
34dc7c2f BB |
4319 | /* |
4320 | * Set pool properties. | |
4321 | */ | |
4322 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
4323 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
4324 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
9babb374 | 4325 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); |
379ca9cf | 4326 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); |
428870ff | 4327 | |
d164b209 BB |
4328 | if (props != NULL) { |
4329 | spa_configfile_set(spa, props, B_FALSE); | |
13fe0198 | 4330 | spa_sync_props(props, tx); |
d164b209 | 4331 | } |
34dc7c2f BB |
4332 | |
4333 | dmu_tx_commit(tx); | |
4334 | ||
b5256303 TC |
4335 | /* |
4336 | * If the root dataset is encrypted we will need to create key mappings | |
4337 | * for the zio layer before we start to write any data to disk and hold | |
4338 | * them until after the first txg has been synced. Waiting for the first | |
4339 | * transaction to complete also ensures that our bean counters are | |
4340 | * appropriately updated. | |
4341 | */ | |
4342 | if (dp->dp_root_dir->dd_crypto_obj != 0) { | |
4343 | root_dsobj = dsl_dir_phys(dp->dp_root_dir)->dd_head_dataset_obj; | |
4344 | VERIFY0(spa_keystore_create_mapping_impl(spa, root_dsobj, | |
4345 | dp->dp_root_dir, FTAG)); | |
4346 | } | |
4347 | ||
34dc7c2f | 4348 | spa->spa_sync_on = B_TRUE; |
b5256303 | 4349 | txg_sync_start(dp); |
379ca9cf | 4350 | mmp_thread_start(spa); |
b5256303 | 4351 | txg_wait_synced(dp, txg); |
34dc7c2f | 4352 | |
b5256303 TC |
4353 | if (dp->dp_root_dir->dd_crypto_obj != 0) |
4354 | VERIFY0(spa_keystore_remove_mapping(spa, root_dsobj, FTAG)); | |
34dc7c2f | 4355 | |
a1d477c2 | 4356 | spa_write_cachefile(spa, B_FALSE, B_TRUE); |
34dc7c2f | 4357 | |
0c66c32d JG |
4358 | /* |
4359 | * Don't count references from objsets that are already closed | |
4360 | * and are making their way through the eviction process. | |
4361 | */ | |
4362 | spa_evicting_os_wait(spa); | |
b128c09f | 4363 | spa->spa_minref = refcount_count(&spa->spa_refcount); |
3dfb57a3 | 4364 | spa->spa_load_state = SPA_LOAD_NONE; |
b128c09f | 4365 | |
d164b209 BB |
4366 | mutex_exit(&spa_namespace_lock); |
4367 | ||
34dc7c2f BB |
4368 | return (0); |
4369 | } | |
4370 | ||
9babb374 BB |
4371 | /* |
4372 | * Import a non-root pool into the system. | |
4373 | */ | |
4374 | int | |
13fe0198 | 4375 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) |
34dc7c2f BB |
4376 | { |
4377 | spa_t *spa; | |
4378 | char *altroot = NULL; | |
428870ff BB |
4379 | spa_load_state_t state = SPA_LOAD_IMPORT; |
4380 | zpool_rewind_policy_t policy; | |
572e2857 BB |
4381 | uint64_t mode = spa_mode_global; |
4382 | uint64_t readonly = B_FALSE; | |
9babb374 | 4383 | int error; |
34dc7c2f BB |
4384 | nvlist_t *nvroot; |
4385 | nvlist_t **spares, **l2cache; | |
4386 | uint_t nspares, nl2cache; | |
34dc7c2f BB |
4387 | |
4388 | /* | |
4389 | * If a pool with this name exists, return failure. | |
4390 | */ | |
4391 | mutex_enter(&spa_namespace_lock); | |
428870ff | 4392 | if (spa_lookup(pool) != NULL) { |
9babb374 | 4393 | mutex_exit(&spa_namespace_lock); |
2e528b49 | 4394 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
4395 | } |
4396 | ||
4397 | /* | |
4398 | * Create and initialize the spa structure. | |
4399 | */ | |
4400 | (void) nvlist_lookup_string(props, | |
4401 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
572e2857 BB |
4402 | (void) nvlist_lookup_uint64(props, |
4403 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
4404 | if (readonly) | |
4405 | mode = FREAD; | |
428870ff | 4406 | spa = spa_add(pool, config, altroot); |
572e2857 BB |
4407 | spa->spa_import_flags = flags; |
4408 | ||
4409 | /* | |
4410 | * Verbatim import - Take a pool and insert it into the namespace | |
4411 | * as if it had been loaded at boot. | |
4412 | */ | |
4413 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
4414 | if (props != NULL) | |
4415 | spa_configfile_set(spa, props, B_FALSE); | |
4416 | ||
a1d477c2 | 4417 | spa_write_cachefile(spa, B_FALSE, B_TRUE); |
12fa0466 | 4418 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
572e2857 BB |
4419 | |
4420 | mutex_exit(&spa_namespace_lock); | |
572e2857 BB |
4421 | return (0); |
4422 | } | |
4423 | ||
4424 | spa_activate(spa, mode); | |
34dc7c2f | 4425 | |
9babb374 BB |
4426 | /* |
4427 | * Don't start async tasks until we know everything is healthy. | |
4428 | */ | |
4429 | spa_async_suspend(spa); | |
b128c09f | 4430 | |
572e2857 BB |
4431 | zpool_get_rewind_policy(config, &policy); |
4432 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
4433 | state = SPA_LOAD_RECOVER; | |
4434 | ||
34dc7c2f | 4435 | /* |
9babb374 BB |
4436 | * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig |
4437 | * because the user-supplied config is actually the one to trust when | |
b128c09f | 4438 | * doing an import. |
34dc7c2f | 4439 | */ |
428870ff BB |
4440 | if (state != SPA_LOAD_RECOVER) |
4441 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
572e2857 | 4442 | |
428870ff BB |
4443 | error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, |
4444 | policy.zrp_request); | |
4445 | ||
4446 | /* | |
572e2857 BB |
4447 | * Propagate anything learned while loading the pool and pass it |
4448 | * back to caller (i.e. rewind info, missing devices, etc). | |
428870ff | 4449 | */ |
572e2857 BB |
4450 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, |
4451 | spa->spa_load_info) == 0); | |
34dc7c2f | 4452 | |
b128c09f | 4453 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4454 | /* |
9babb374 BB |
4455 | * Toss any existing sparelist, as it doesn't have any validity |
4456 | * anymore, and conflicts with spa_has_spare(). | |
34dc7c2f | 4457 | */ |
9babb374 | 4458 | if (spa->spa_spares.sav_config) { |
34dc7c2f BB |
4459 | nvlist_free(spa->spa_spares.sav_config); |
4460 | spa->spa_spares.sav_config = NULL; | |
4461 | spa_load_spares(spa); | |
4462 | } | |
9babb374 | 4463 | if (spa->spa_l2cache.sav_config) { |
34dc7c2f BB |
4464 | nvlist_free(spa->spa_l2cache.sav_config); |
4465 | spa->spa_l2cache.sav_config = NULL; | |
4466 | spa_load_l2cache(spa); | |
4467 | } | |
4468 | ||
4469 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, | |
4470 | &nvroot) == 0); | |
b128c09f | 4471 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f | 4472 | |
d164b209 BB |
4473 | if (props != NULL) |
4474 | spa_configfile_set(spa, props, B_FALSE); | |
4475 | ||
fb5f0bc8 BB |
4476 | if (error != 0 || (props && spa_writeable(spa) && |
4477 | (error = spa_prop_set(spa, props)))) { | |
9babb374 BB |
4478 | spa_unload(spa); |
4479 | spa_deactivate(spa); | |
4480 | spa_remove(spa); | |
34dc7c2f BB |
4481 | mutex_exit(&spa_namespace_lock); |
4482 | return (error); | |
4483 | } | |
4484 | ||
572e2857 BB |
4485 | spa_async_resume(spa); |
4486 | ||
34dc7c2f BB |
4487 | /* |
4488 | * Override any spares and level 2 cache devices as specified by | |
4489 | * the user, as these may have correct device names/devids, etc. | |
4490 | */ | |
4491 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
4492 | &spares, &nspares) == 0) { | |
4493 | if (spa->spa_spares.sav_config) | |
4494 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, | |
4495 | ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4496 | else | |
4497 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, | |
79c76d5b | 4498 | NV_UNIQUE_NAME, KM_SLEEP) == 0); |
34dc7c2f BB |
4499 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, |
4500 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
b128c09f | 4501 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4502 | spa_load_spares(spa); |
b128c09f | 4503 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4504 | spa->spa_spares.sav_sync = B_TRUE; |
4505 | } | |
4506 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
4507 | &l2cache, &nl2cache) == 0) { | |
4508 | if (spa->spa_l2cache.sav_config) | |
4509 | VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, | |
4510 | ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4511 | else | |
4512 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
79c76d5b | 4513 | NV_UNIQUE_NAME, KM_SLEEP) == 0); |
34dc7c2f BB |
4514 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, |
4515 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
b128c09f | 4516 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4517 | spa_load_l2cache(spa); |
b128c09f | 4518 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4519 | spa->spa_l2cache.sav_sync = B_TRUE; |
4520 | } | |
4521 | ||
428870ff BB |
4522 | /* |
4523 | * Check for any removed devices. | |
4524 | */ | |
4525 | if (spa->spa_autoreplace) { | |
4526 | spa_aux_check_removed(&spa->spa_spares); | |
4527 | spa_aux_check_removed(&spa->spa_l2cache); | |
4528 | } | |
4529 | ||
fb5f0bc8 | 4530 | if (spa_writeable(spa)) { |
b128c09f BB |
4531 | /* |
4532 | * Update the config cache to include the newly-imported pool. | |
4533 | */ | |
45d1cae3 | 4534 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
b128c09f | 4535 | } |
34dc7c2f | 4536 | |
34dc7c2f | 4537 | /* |
9babb374 BB |
4538 | * It's possible that the pool was expanded while it was exported. |
4539 | * We kick off an async task to handle this for us. | |
34dc7c2f | 4540 | */ |
9babb374 | 4541 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); |
b128c09f | 4542 | |
d5e024cb | 4543 | spa_history_log_version(spa, "import", NULL); |
fb390aaf | 4544 | |
12fa0466 | 4545 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); |
fb390aaf | 4546 | |
a0bd735a | 4547 | zvol_create_minors(spa, pool, B_TRUE); |
526af785 | 4548 | |
fb390aaf HR |
4549 | mutex_exit(&spa_namespace_lock); |
4550 | ||
b128c09f BB |
4551 | return (0); |
4552 | } | |
4553 | ||
34dc7c2f BB |
4554 | nvlist_t * |
4555 | spa_tryimport(nvlist_t *tryconfig) | |
4556 | { | |
4557 | nvlist_t *config = NULL; | |
4558 | char *poolname; | |
4559 | spa_t *spa; | |
4560 | uint64_t state; | |
d164b209 | 4561 | int error; |
34dc7c2f BB |
4562 | |
4563 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
4564 | return (NULL); | |
4565 | ||
4566 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
4567 | return (NULL); | |
4568 | ||
4569 | /* | |
4570 | * Create and initialize the spa structure. | |
4571 | */ | |
4572 | mutex_enter(&spa_namespace_lock); | |
428870ff | 4573 | spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); |
fb5f0bc8 | 4574 | spa_activate(spa, FREAD); |
34dc7c2f BB |
4575 | |
4576 | /* | |
4577 | * Pass off the heavy lifting to spa_load(). | |
4578 | * Pass TRUE for mosconfig because the user-supplied config | |
4579 | * is actually the one to trust when doing an import. | |
4580 | */ | |
428870ff | 4581 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); |
34dc7c2f BB |
4582 | |
4583 | /* | |
4584 | * If 'tryconfig' was at least parsable, return the current config. | |
4585 | */ | |
4586 | if (spa->spa_root_vdev != NULL) { | |
34dc7c2f | 4587 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
34dc7c2f BB |
4588 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, |
4589 | poolname) == 0); | |
4590 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
4591 | state) == 0); | |
4592 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
4593 | spa->spa_uberblock.ub_timestamp) == 0); | |
9ae529ec CS |
4594 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, |
4595 | spa->spa_load_info) == 0); | |
ffe9d382 BB |
4596 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, |
4597 | spa->spa_errata) == 0); | |
34dc7c2f BB |
4598 | |
4599 | /* | |
4600 | * If the bootfs property exists on this pool then we | |
4601 | * copy it out so that external consumers can tell which | |
4602 | * pools are bootable. | |
4603 | */ | |
d164b209 | 4604 | if ((!error || error == EEXIST) && spa->spa_bootfs) { |
79c76d5b | 4605 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
4606 | |
4607 | /* | |
4608 | * We have to play games with the name since the | |
4609 | * pool was opened as TRYIMPORT_NAME. | |
4610 | */ | |
b128c09f | 4611 | if (dsl_dsobj_to_dsname(spa_name(spa), |
34dc7c2f BB |
4612 | spa->spa_bootfs, tmpname) == 0) { |
4613 | char *cp; | |
d1d7e268 MK |
4614 | char *dsname; |
4615 | ||
79c76d5b | 4616 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
34dc7c2f BB |
4617 | |
4618 | cp = strchr(tmpname, '/'); | |
4619 | if (cp == NULL) { | |
4620 | (void) strlcpy(dsname, tmpname, | |
4621 | MAXPATHLEN); | |
4622 | } else { | |
4623 | (void) snprintf(dsname, MAXPATHLEN, | |
4624 | "%s/%s", poolname, ++cp); | |
4625 | } | |
4626 | VERIFY(nvlist_add_string(config, | |
4627 | ZPOOL_CONFIG_BOOTFS, dsname) == 0); | |
4628 | kmem_free(dsname, MAXPATHLEN); | |
4629 | } | |
4630 | kmem_free(tmpname, MAXPATHLEN); | |
4631 | } | |
4632 | ||
4633 | /* | |
4634 | * Add the list of hot spares and level 2 cache devices. | |
4635 | */ | |
9babb374 | 4636 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
34dc7c2f BB |
4637 | spa_add_spares(spa, config); |
4638 | spa_add_l2cache(spa, config); | |
9babb374 | 4639 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f BB |
4640 | } |
4641 | ||
4642 | spa_unload(spa); | |
4643 | spa_deactivate(spa); | |
4644 | spa_remove(spa); | |
4645 | mutex_exit(&spa_namespace_lock); | |
4646 | ||
4647 | return (config); | |
4648 | } | |
4649 | ||
4650 | /* | |
4651 | * Pool export/destroy | |
4652 | * | |
4653 | * The act of destroying or exporting a pool is very simple. We make sure there | |
4654 | * is no more pending I/O and any references to the pool are gone. Then, we | |
4655 | * update the pool state and sync all the labels to disk, removing the | |
fb5f0bc8 BB |
4656 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then |
4657 | * we don't sync the labels or remove the configuration cache. | |
34dc7c2f BB |
4658 | */ |
4659 | static int | |
b128c09f | 4660 | spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, |
fb5f0bc8 | 4661 | boolean_t force, boolean_t hardforce) |
34dc7c2f BB |
4662 | { |
4663 | spa_t *spa; | |
4664 | ||
4665 | if (oldconfig) | |
4666 | *oldconfig = NULL; | |
4667 | ||
fb5f0bc8 | 4668 | if (!(spa_mode_global & FWRITE)) |
2e528b49 | 4669 | return (SET_ERROR(EROFS)); |
34dc7c2f BB |
4670 | |
4671 | mutex_enter(&spa_namespace_lock); | |
4672 | if ((spa = spa_lookup(pool)) == NULL) { | |
4673 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 4674 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
4675 | } |
4676 | ||
4677 | /* | |
4678 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
4679 | * reacquire the namespace lock, and see if we can export. | |
4680 | */ | |
4681 | spa_open_ref(spa, FTAG); | |
4682 | mutex_exit(&spa_namespace_lock); | |
4683 | spa_async_suspend(spa); | |
a0bd735a BP |
4684 | if (spa->spa_zvol_taskq) { |
4685 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
4686 | taskq_wait(spa->spa_zvol_taskq); | |
4687 | } | |
34dc7c2f BB |
4688 | mutex_enter(&spa_namespace_lock); |
4689 | spa_close(spa, FTAG); | |
4690 | ||
d14cfd83 IH |
4691 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) |
4692 | goto export_spa; | |
34dc7c2f | 4693 | /* |
d14cfd83 IH |
4694 | * The pool will be in core if it's openable, in which case we can |
4695 | * modify its state. Objsets may be open only because they're dirty, | |
4696 | * so we have to force it to sync before checking spa_refcnt. | |
34dc7c2f | 4697 | */ |
0c66c32d | 4698 | if (spa->spa_sync_on) { |
34dc7c2f | 4699 | txg_wait_synced(spa->spa_dsl_pool, 0); |
0c66c32d JG |
4700 | spa_evicting_os_wait(spa); |
4701 | } | |
34dc7c2f | 4702 | |
d14cfd83 IH |
4703 | /* |
4704 | * A pool cannot be exported or destroyed if there are active | |
4705 | * references. If we are resetting a pool, allow references by | |
4706 | * fault injection handlers. | |
4707 | */ | |
4708 | if (!spa_refcount_zero(spa) || | |
4709 | (spa->spa_inject_ref != 0 && | |
4710 | new_state != POOL_STATE_UNINITIALIZED)) { | |
4711 | spa_async_resume(spa); | |
4712 | mutex_exit(&spa_namespace_lock); | |
4713 | return (SET_ERROR(EBUSY)); | |
4714 | } | |
34dc7c2f | 4715 | |
d14cfd83 | 4716 | if (spa->spa_sync_on) { |
b128c09f BB |
4717 | /* |
4718 | * A pool cannot be exported if it has an active shared spare. | |
4719 | * This is to prevent other pools stealing the active spare | |
4720 | * from an exported pool. At user's own will, such pool can | |
4721 | * be forcedly exported. | |
4722 | */ | |
4723 | if (!force && new_state == POOL_STATE_EXPORTED && | |
4724 | spa_has_active_shared_spare(spa)) { | |
4725 | spa_async_resume(spa); | |
4726 | mutex_exit(&spa_namespace_lock); | |
2e528b49 | 4727 | return (SET_ERROR(EXDEV)); |
b128c09f | 4728 | } |
34dc7c2f BB |
4729 | |
4730 | /* | |
4731 | * We want this to be reflected on every label, | |
4732 | * so mark them all dirty. spa_unload() will do the | |
4733 | * final sync that pushes these changes out. | |
4734 | */ | |
fb5f0bc8 | 4735 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { |
b128c09f | 4736 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f | 4737 | spa->spa_state = new_state; |
428870ff BB |
4738 | spa->spa_final_txg = spa_last_synced_txg(spa) + |
4739 | TXG_DEFER_SIZE + 1; | |
34dc7c2f | 4740 | vdev_config_dirty(spa->spa_root_vdev); |
b128c09f | 4741 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
4742 | } |
4743 | } | |
4744 | ||
d14cfd83 | 4745 | export_spa: |
d5e024cb BB |
4746 | if (new_state == POOL_STATE_DESTROYED) |
4747 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
4748 | else if (new_state == POOL_STATE_EXPORTED) | |
4749 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
34dc7c2f BB |
4750 | |
4751 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
4752 | spa_unload(spa); | |
4753 | spa_deactivate(spa); | |
4754 | } | |
4755 | ||
4756 | if (oldconfig && spa->spa_config) | |
4757 | VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); | |
4758 | ||
4759 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
fb5f0bc8 | 4760 | if (!hardforce) |
a1d477c2 | 4761 | spa_write_cachefile(spa, B_TRUE, B_TRUE); |
34dc7c2f | 4762 | spa_remove(spa); |
34dc7c2f BB |
4763 | } |
4764 | mutex_exit(&spa_namespace_lock); | |
4765 | ||
4766 | return (0); | |
4767 | } | |
4768 | ||
4769 | /* | |
4770 | * Destroy a storage pool. | |
4771 | */ | |
4772 | int | |
4773 | spa_destroy(char *pool) | |
4774 | { | |
fb5f0bc8 BB |
4775 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, |
4776 | B_FALSE, B_FALSE)); | |
34dc7c2f BB |
4777 | } |
4778 | ||
4779 | /* | |
4780 | * Export a storage pool. | |
4781 | */ | |
4782 | int | |
fb5f0bc8 BB |
4783 | spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, |
4784 | boolean_t hardforce) | |
34dc7c2f | 4785 | { |
fb5f0bc8 BB |
4786 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, |
4787 | force, hardforce)); | |
34dc7c2f BB |
4788 | } |
4789 | ||
4790 | /* | |
4791 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
4792 | * from the namespace in any way. | |
4793 | */ | |
4794 | int | |
4795 | spa_reset(char *pool) | |
4796 | { | |
b128c09f | 4797 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, |
fb5f0bc8 | 4798 | B_FALSE, B_FALSE)); |
34dc7c2f BB |
4799 | } |
4800 | ||
34dc7c2f BB |
4801 | /* |
4802 | * ========================================================================== | |
4803 | * Device manipulation | |
4804 | * ========================================================================== | |
4805 | */ | |
4806 | ||
4807 | /* | |
4808 | * Add a device to a storage pool. | |
4809 | */ | |
4810 | int | |
4811 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
4812 | { | |
428870ff | 4813 | uint64_t txg, id; |
fb5f0bc8 | 4814 | int error; |
34dc7c2f BB |
4815 | vdev_t *rvd = spa->spa_root_vdev; |
4816 | vdev_t *vd, *tvd; | |
4817 | nvlist_t **spares, **l2cache; | |
4818 | uint_t nspares, nl2cache; | |
4819 | ||
572e2857 BB |
4820 | ASSERT(spa_writeable(spa)); |
4821 | ||
34dc7c2f BB |
4822 | txg = spa_vdev_enter(spa); |
4823 | ||
4824 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
4825 | VDEV_ALLOC_ADD)) != 0) | |
4826 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
4827 | ||
b128c09f | 4828 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ |
34dc7c2f BB |
4829 | |
4830 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
4831 | &nspares) != 0) | |
4832 | nspares = 0; | |
4833 | ||
4834 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
4835 | &nl2cache) != 0) | |
4836 | nl2cache = 0; | |
4837 | ||
b128c09f | 4838 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) |
34dc7c2f | 4839 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); |
34dc7c2f | 4840 | |
b128c09f BB |
4841 | if (vd->vdev_children != 0 && |
4842 | (error = vdev_create(vd, txg, B_FALSE)) != 0) | |
4843 | return (spa_vdev_exit(spa, vd, txg, error)); | |
34dc7c2f BB |
4844 | |
4845 | /* | |
4846 | * We must validate the spares and l2cache devices after checking the | |
4847 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
4848 | */ | |
b128c09f | 4849 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) |
34dc7c2f | 4850 | return (spa_vdev_exit(spa, vd, txg, error)); |
34dc7c2f BB |
4851 | |
4852 | /* | |
a1d477c2 MA |
4853 | * If we are in the middle of a device removal, we can only add |
4854 | * devices which match the existing devices in the pool. | |
4855 | * If we are in the middle of a removal, or have some indirect | |
4856 | * vdevs, we can not add raidz toplevels. | |
34dc7c2f | 4857 | */ |
a1d477c2 MA |
4858 | if (spa->spa_vdev_removal != NULL || |
4859 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
4860 | for (int c = 0; c < vd->vdev_children; c++) { | |
4861 | tvd = vd->vdev_child[c]; | |
4862 | if (spa->spa_vdev_removal != NULL && | |
4863 | tvd->vdev_ashift != | |
4864 | spa->spa_vdev_removal->svr_vdev->vdev_ashift) { | |
4865 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
4866 | } | |
4867 | /* Fail if top level vdev is raidz */ | |
4868 | if (tvd->vdev_ops == &vdev_raidz_ops) { | |
4869 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
4870 | } | |
4871 | /* | |
4872 | * Need the top level mirror to be | |
4873 | * a mirror of leaf vdevs only | |
4874 | */ | |
4875 | if (tvd->vdev_ops == &vdev_mirror_ops) { | |
4876 | for (uint64_t cid = 0; | |
4877 | cid < tvd->vdev_children; cid++) { | |
4878 | vdev_t *cvd = tvd->vdev_child[cid]; | |
4879 | if (!cvd->vdev_ops->vdev_op_leaf) { | |
4880 | return (spa_vdev_exit(spa, vd, | |
4881 | txg, EINVAL)); | |
4882 | } | |
4883 | } | |
4884 | } | |
4885 | } | |
4886 | } | |
4887 | ||
1c27024e | 4888 | for (int c = 0; c < vd->vdev_children; c++) { |
428870ff BB |
4889 | |
4890 | /* | |
4891 | * Set the vdev id to the first hole, if one exists. | |
4892 | */ | |
4893 | for (id = 0; id < rvd->vdev_children; id++) { | |
4894 | if (rvd->vdev_child[id]->vdev_ishole) { | |
4895 | vdev_free(rvd->vdev_child[id]); | |
4896 | break; | |
4897 | } | |
4898 | } | |
34dc7c2f BB |
4899 | tvd = vd->vdev_child[c]; |
4900 | vdev_remove_child(vd, tvd); | |
428870ff | 4901 | tvd->vdev_id = id; |
34dc7c2f BB |
4902 | vdev_add_child(rvd, tvd); |
4903 | vdev_config_dirty(tvd); | |
4904 | } | |
4905 | ||
4906 | if (nspares != 0) { | |
4907 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
4908 | ZPOOL_CONFIG_SPARES); | |
4909 | spa_load_spares(spa); | |
4910 | spa->spa_spares.sav_sync = B_TRUE; | |
4911 | } | |
4912 | ||
4913 | if (nl2cache != 0) { | |
4914 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
4915 | ZPOOL_CONFIG_L2CACHE); | |
4916 | spa_load_l2cache(spa); | |
4917 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4918 | } | |
4919 | ||
4920 | /* | |
4921 | * We have to be careful when adding new vdevs to an existing pool. | |
4922 | * If other threads start allocating from these vdevs before we | |
4923 | * sync the config cache, and we lose power, then upon reboot we may | |
4924 | * fail to open the pool because there are DVAs that the config cache | |
4925 | * can't translate. Therefore, we first add the vdevs without | |
4926 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
4927 | * and then let spa_config_update() initialize the new metaslabs. | |
4928 | * | |
4929 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
4930 | * if we lose power at any point in this sequence, the remaining | |
4931 | * steps will be completed the next time we load the pool. | |
4932 | */ | |
4933 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
4934 | ||
4935 | mutex_enter(&spa_namespace_lock); | |
4936 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
12fa0466 | 4937 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); |
34dc7c2f BB |
4938 | mutex_exit(&spa_namespace_lock); |
4939 | ||
4940 | return (0); | |
4941 | } | |
4942 | ||
4943 | /* | |
4944 | * Attach a device to a mirror. The arguments are the path to any device | |
4945 | * in the mirror, and the nvroot for the new device. If the path specifies | |
4946 | * a device that is not mirrored, we automatically insert the mirror vdev. | |
4947 | * | |
4948 | * If 'replacing' is specified, the new device is intended to replace the | |
4949 | * existing device; in this case the two devices are made into their own | |
4950 | * mirror using the 'replacing' vdev, which is functionally identical to | |
4951 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
4952 | * extra rules: you can't attach to it after it's been created, and upon | |
4953 | * completion of resilvering, the first disk (the one being replaced) | |
4954 | * is automatically detached. | |
4955 | */ | |
4956 | int | |
4957 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) | |
4958 | { | |
428870ff | 4959 | uint64_t txg, dtl_max_txg; |
1c27024e | 4960 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); |
34dc7c2f BB |
4961 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; |
4962 | vdev_ops_t *pvops; | |
b128c09f BB |
4963 | char *oldvdpath, *newvdpath; |
4964 | int newvd_isspare; | |
4965 | int error; | |
34dc7c2f | 4966 | |
572e2857 BB |
4967 | ASSERT(spa_writeable(spa)); |
4968 | ||
34dc7c2f BB |
4969 | txg = spa_vdev_enter(spa); |
4970 | ||
b128c09f | 4971 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f | 4972 | |
a1d477c2 MA |
4973 | if (spa->spa_vdev_removal != NULL || |
4974 | spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { | |
4975 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
4976 | } | |
4977 | ||
34dc7c2f BB |
4978 | if (oldvd == NULL) |
4979 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
4980 | ||
4981 | if (!oldvd->vdev_ops->vdev_op_leaf) | |
4982 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4983 | ||
4984 | pvd = oldvd->vdev_parent; | |
4985 | ||
4986 | if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, | |
5ffb9d1d | 4987 | VDEV_ALLOC_ATTACH)) != 0) |
34dc7c2f BB |
4988 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); |
4989 | ||
4990 | if (newrootvd->vdev_children != 1) | |
4991 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4992 | ||
4993 | newvd = newrootvd->vdev_child[0]; | |
4994 | ||
4995 | if (!newvd->vdev_ops->vdev_op_leaf) | |
4996 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4997 | ||
4998 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
4999 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
5000 | ||
5001 | /* | |
5002 | * Spares can't replace logs | |
5003 | */ | |
b128c09f | 5004 | if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) |
34dc7c2f BB |
5005 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
5006 | ||
5007 | if (!replacing) { | |
5008 | /* | |
5009 | * For attach, the only allowable parent is a mirror or the root | |
5010 | * vdev. | |
5011 | */ | |
5012 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
5013 | pvd->vdev_ops != &vdev_root_ops) | |
5014 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
5015 | ||
5016 | pvops = &vdev_mirror_ops; | |
5017 | } else { | |
5018 | /* | |
5019 | * Active hot spares can only be replaced by inactive hot | |
5020 | * spares. | |
5021 | */ | |
5022 | if (pvd->vdev_ops == &vdev_spare_ops && | |
572e2857 | 5023 | oldvd->vdev_isspare && |
34dc7c2f BB |
5024 | !spa_has_spare(spa, newvd->vdev_guid)) |
5025 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
5026 | ||
5027 | /* | |
5028 | * If the source is a hot spare, and the parent isn't already a | |
5029 | * spare, then we want to create a new hot spare. Otherwise, we | |
5030 | * want to create a replacing vdev. The user is not allowed to | |
5031 | * attach to a spared vdev child unless the 'isspare' state is | |
5032 | * the same (spare replaces spare, non-spare replaces | |
5033 | * non-spare). | |
5034 | */ | |
572e2857 BB |
5035 | if (pvd->vdev_ops == &vdev_replacing_ops && |
5036 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
34dc7c2f | 5037 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
5038 | } else if (pvd->vdev_ops == &vdev_spare_ops && |
5039 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
34dc7c2f | 5040 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); |
572e2857 BB |
5041 | } |
5042 | ||
5043 | if (newvd->vdev_isspare) | |
34dc7c2f BB |
5044 | pvops = &vdev_spare_ops; |
5045 | else | |
5046 | pvops = &vdev_replacing_ops; | |
5047 | } | |
5048 | ||
5049 | /* | |
9babb374 | 5050 | * Make sure the new device is big enough. |
34dc7c2f | 5051 | */ |
9babb374 | 5052 | if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) |
34dc7c2f BB |
5053 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
5054 | ||
5055 | /* | |
5056 | * The new device cannot have a higher alignment requirement | |
5057 | * than the top-level vdev. | |
5058 | */ | |
5059 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
5060 | return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); | |
5061 | ||
5062 | /* | |
5063 | * If this is an in-place replacement, update oldvd's path and devid | |
5064 | * to make it distinguishable from newvd, and unopenable from now on. | |
5065 | */ | |
5066 | if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
5067 | spa_strfree(oldvd->vdev_path); | |
5068 | oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
79c76d5b | 5069 | KM_SLEEP); |
34dc7c2f BB |
5070 | (void) sprintf(oldvd->vdev_path, "%s/%s", |
5071 | newvd->vdev_path, "old"); | |
5072 | if (oldvd->vdev_devid != NULL) { | |
5073 | spa_strfree(oldvd->vdev_devid); | |
5074 | oldvd->vdev_devid = NULL; | |
5075 | } | |
5076 | } | |
5077 | ||
572e2857 | 5078 | /* mark the device being resilvered */ |
5d1f7fb6 | 5079 | newvd->vdev_resilver_txg = txg; |
572e2857 | 5080 | |
34dc7c2f BB |
5081 | /* |
5082 | * If the parent is not a mirror, or if we're replacing, insert the new | |
5083 | * mirror/replacing/spare vdev above oldvd. | |
5084 | */ | |
5085 | if (pvd->vdev_ops != pvops) | |
5086 | pvd = vdev_add_parent(oldvd, pvops); | |
5087 | ||
5088 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
5089 | ASSERT(pvd->vdev_ops == pvops); | |
5090 | ASSERT(oldvd->vdev_parent == pvd); | |
5091 | ||
5092 | /* | |
5093 | * Extract the new device from its root and add it to pvd. | |
5094 | */ | |
5095 | vdev_remove_child(newrootvd, newvd); | |
5096 | newvd->vdev_id = pvd->vdev_children; | |
428870ff | 5097 | newvd->vdev_crtxg = oldvd->vdev_crtxg; |
34dc7c2f BB |
5098 | vdev_add_child(pvd, newvd); |
5099 | ||
6d82f98c IH |
5100 | /* |
5101 | * Reevaluate the parent vdev state. | |
5102 | */ | |
5103 | vdev_propagate_state(pvd); | |
5104 | ||
34dc7c2f BB |
5105 | tvd = newvd->vdev_top; |
5106 | ASSERT(pvd->vdev_top == tvd); | |
5107 | ASSERT(tvd->vdev_parent == rvd); | |
5108 | ||
5109 | vdev_config_dirty(tvd); | |
5110 | ||
5111 | /* | |
428870ff BB |
5112 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account |
5113 | * for any dmu_sync-ed blocks. It will propagate upward when | |
5114 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
34dc7c2f | 5115 | */ |
428870ff | 5116 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; |
34dc7c2f | 5117 | |
428870ff BB |
5118 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, |
5119 | dtl_max_txg - TXG_INITIAL); | |
34dc7c2f | 5120 | |
9babb374 | 5121 | if (newvd->vdev_isspare) { |
34dc7c2f | 5122 | spa_spare_activate(newvd); |
12fa0466 | 5123 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); |
9babb374 BB |
5124 | } |
5125 | ||
b128c09f BB |
5126 | oldvdpath = spa_strdup(oldvd->vdev_path); |
5127 | newvdpath = spa_strdup(newvd->vdev_path); | |
5128 | newvd_isspare = newvd->vdev_isspare; | |
34dc7c2f BB |
5129 | |
5130 | /* | |
5131 | * Mark newvd's DTL dirty in this txg. | |
5132 | */ | |
5133 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
5134 | ||
428870ff | 5135 | /* |
93cf2076 GW |
5136 | * Schedule the resilver to restart in the future. We do this to |
5137 | * ensure that dmu_sync-ed blocks have been stitched into the | |
5138 | * respective datasets. | |
428870ff BB |
5139 | */ |
5140 | dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); | |
5141 | ||
fb390aaf | 5142 | if (spa->spa_bootfs) |
12fa0466 | 5143 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); |
fb390aaf | 5144 | |
12fa0466 | 5145 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); |
fb390aaf | 5146 | |
428870ff BB |
5147 | /* |
5148 | * Commit the config | |
5149 | */ | |
5150 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
34dc7c2f | 5151 | |
6f1ffb06 | 5152 | spa_history_log_internal(spa, "vdev attach", NULL, |
428870ff | 5153 | "%s vdev=%s %s vdev=%s", |
45d1cae3 BB |
5154 | replacing && newvd_isspare ? "spare in" : |
5155 | replacing ? "replace" : "attach", newvdpath, | |
5156 | replacing ? "for" : "to", oldvdpath); | |
b128c09f BB |
5157 | |
5158 | spa_strfree(oldvdpath); | |
5159 | spa_strfree(newvdpath); | |
5160 | ||
34dc7c2f BB |
5161 | return (0); |
5162 | } | |
5163 | ||
5164 | /* | |
5165 | * Detach a device from a mirror or replacing vdev. | |
d3cc8b15 | 5166 | * |
34dc7c2f BB |
5167 | * If 'replace_done' is specified, only detach if the parent |
5168 | * is a replacing vdev. | |
5169 | */ | |
5170 | int | |
fb5f0bc8 | 5171 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) |
34dc7c2f BB |
5172 | { |
5173 | uint64_t txg; | |
fb5f0bc8 | 5174 | int error; |
1c27024e | 5175 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); |
34dc7c2f BB |
5176 | vdev_t *vd, *pvd, *cvd, *tvd; |
5177 | boolean_t unspare = B_FALSE; | |
d4ed6673 | 5178 | uint64_t unspare_guid = 0; |
428870ff | 5179 | char *vdpath; |
1c27024e | 5180 | |
572e2857 BB |
5181 | ASSERT(spa_writeable(spa)); |
5182 | ||
34dc7c2f BB |
5183 | txg = spa_vdev_enter(spa); |
5184 | ||
b128c09f | 5185 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); |
34dc7c2f BB |
5186 | |
5187 | if (vd == NULL) | |
5188 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
5189 | ||
5190 | if (!vd->vdev_ops->vdev_op_leaf) | |
5191 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
5192 | ||
5193 | pvd = vd->vdev_parent; | |
5194 | ||
fb5f0bc8 BB |
5195 | /* |
5196 | * If the parent/child relationship is not as expected, don't do it. | |
5197 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
5198 | * vdev that's replacing B with C. The user's intent in replacing | |
5199 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
5200 | * the replace by detaching C, the expected behavior is to end up | |
5201 | * M(A,B). But suppose that right after deciding to detach C, | |
5202 | * the replacement of B completes. We would have M(A,C), and then | |
5203 | * ask to detach C, which would leave us with just A -- not what | |
5204 | * the user wanted. To prevent this, we make sure that the | |
5205 | * parent/child relationship hasn't changed -- in this example, | |
5206 | * that C's parent is still the replacing vdev R. | |
5207 | */ | |
5208 | if (pvd->vdev_guid != pguid && pguid != 0) | |
5209 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
5210 | ||
34dc7c2f | 5211 | /* |
572e2857 | 5212 | * Only 'replacing' or 'spare' vdevs can be replaced. |
34dc7c2f | 5213 | */ |
572e2857 BB |
5214 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && |
5215 | pvd->vdev_ops != &vdev_spare_ops) | |
5216 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
34dc7c2f BB |
5217 | |
5218 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
5219 | spa_version(spa) >= SPA_VERSION_SPARES); | |
5220 | ||
5221 | /* | |
5222 | * Only mirror, replacing, and spare vdevs support detach. | |
5223 | */ | |
5224 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
5225 | pvd->vdev_ops != &vdev_mirror_ops && | |
5226 | pvd->vdev_ops != &vdev_spare_ops) | |
5227 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
5228 | ||
5229 | /* | |
fb5f0bc8 BB |
5230 | * If this device has the only valid copy of some data, |
5231 | * we cannot safely detach it. | |
34dc7c2f | 5232 | */ |
fb5f0bc8 | 5233 | if (vdev_dtl_required(vd)) |
34dc7c2f BB |
5234 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
5235 | ||
fb5f0bc8 | 5236 | ASSERT(pvd->vdev_children >= 2); |
34dc7c2f | 5237 | |
b128c09f BB |
5238 | /* |
5239 | * If we are detaching the second disk from a replacing vdev, then | |
5240 | * check to see if we changed the original vdev's path to have "/old" | |
5241 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
5242 | */ | |
572e2857 BB |
5243 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && |
5244 | vd->vdev_path != NULL) { | |
5245 | size_t len = strlen(vd->vdev_path); | |
5246 | ||
1c27024e | 5247 | for (int c = 0; c < pvd->vdev_children; c++) { |
572e2857 BB |
5248 | cvd = pvd->vdev_child[c]; |
5249 | ||
5250 | if (cvd == vd || cvd->vdev_path == NULL) | |
5251 | continue; | |
5252 | ||
5253 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
5254 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
5255 | spa_strfree(cvd->vdev_path); | |
5256 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
5257 | break; | |
5258 | } | |
b128c09f BB |
5259 | } |
5260 | } | |
5261 | ||
34dc7c2f BB |
5262 | /* |
5263 | * If we are detaching the original disk from a spare, then it implies | |
5264 | * that the spare should become a real disk, and be removed from the | |
5265 | * active spare list for the pool. | |
5266 | */ | |
5267 | if (pvd->vdev_ops == &vdev_spare_ops && | |
572e2857 BB |
5268 | vd->vdev_id == 0 && |
5269 | pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) | |
34dc7c2f BB |
5270 | unspare = B_TRUE; |
5271 | ||
5272 | /* | |
5273 | * Erase the disk labels so the disk can be used for other things. | |
5274 | * This must be done after all other error cases are handled, | |
5275 | * but before we disembowel vd (so we can still do I/O to it). | |
5276 | * But if we can't do it, don't treat the error as fatal -- | |
5277 | * it may be that the unwritability of the disk is the reason | |
5278 | * it's being detached! | |
5279 | */ | |
5280 | error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
5281 | ||
5282 | /* | |
5283 | * Remove vd from its parent and compact the parent's children. | |
5284 | */ | |
5285 | vdev_remove_child(pvd, vd); | |
5286 | vdev_compact_children(pvd); | |
5287 | ||
5288 | /* | |
5289 | * Remember one of the remaining children so we can get tvd below. | |
5290 | */ | |
572e2857 | 5291 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; |
34dc7c2f BB |
5292 | |
5293 | /* | |
5294 | * If we need to remove the remaining child from the list of hot spares, | |
fb5f0bc8 BB |
5295 | * do it now, marking the vdev as no longer a spare in the process. |
5296 | * We must do this before vdev_remove_parent(), because that can | |
5297 | * change the GUID if it creates a new toplevel GUID. For a similar | |
5298 | * reason, we must remove the spare now, in the same txg as the detach; | |
5299 | * otherwise someone could attach a new sibling, change the GUID, and | |
5300 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
34dc7c2f BB |
5301 | */ |
5302 | if (unspare) { | |
5303 | ASSERT(cvd->vdev_isspare); | |
5304 | spa_spare_remove(cvd); | |
5305 | unspare_guid = cvd->vdev_guid; | |
fb5f0bc8 | 5306 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); |
572e2857 | 5307 | cvd->vdev_unspare = B_TRUE; |
34dc7c2f BB |
5308 | } |
5309 | ||
428870ff BB |
5310 | /* |
5311 | * If the parent mirror/replacing vdev only has one child, | |
5312 | * the parent is no longer needed. Remove it from the tree. | |
5313 | */ | |
572e2857 BB |
5314 | if (pvd->vdev_children == 1) { |
5315 | if (pvd->vdev_ops == &vdev_spare_ops) | |
5316 | cvd->vdev_unspare = B_FALSE; | |
428870ff | 5317 | vdev_remove_parent(cvd); |
572e2857 BB |
5318 | } |
5319 | ||
428870ff BB |
5320 | |
5321 | /* | |
5322 | * We don't set tvd until now because the parent we just removed | |
5323 | * may have been the previous top-level vdev. | |
5324 | */ | |
5325 | tvd = cvd->vdev_top; | |
5326 | ASSERT(tvd->vdev_parent == rvd); | |
5327 | ||
5328 | /* | |
5329 | * Reevaluate the parent vdev state. | |
5330 | */ | |
5331 | vdev_propagate_state(cvd); | |
5332 | ||
5333 | /* | |
5334 | * If the 'autoexpand' property is set on the pool then automatically | |
5335 | * try to expand the size of the pool. For example if the device we | |
5336 | * just detached was smaller than the others, it may be possible to | |
5337 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
5338 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
5339 | */ | |
5340 | if (spa->spa_autoexpand) { | |
5341 | vdev_reopen(tvd); | |
5342 | vdev_expand(tvd, txg); | |
5343 | } | |
5344 | ||
5345 | vdev_config_dirty(tvd); | |
5346 | ||
5347 | /* | |
5348 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
5349 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
5350 | * But first make sure we're not on any *other* txg's DTL list, to | |
5351 | * prevent vd from being accessed after it's freed. | |
5352 | */ | |
b6ca6193 | 5353 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); |
1c27024e | 5354 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff BB |
5355 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); |
5356 | vd->vdev_detached = B_TRUE; | |
5357 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
5358 | ||
12fa0466 | 5359 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); |
428870ff | 5360 | |
572e2857 BB |
5361 | /* hang on to the spa before we release the lock */ |
5362 | spa_open_ref(spa, FTAG); | |
5363 | ||
428870ff BB |
5364 | error = spa_vdev_exit(spa, vd, txg, 0); |
5365 | ||
6f1ffb06 | 5366 | spa_history_log_internal(spa, "detach", NULL, |
428870ff BB |
5367 | "vdev=%s", vdpath); |
5368 | spa_strfree(vdpath); | |
5369 | ||
5370 | /* | |
5371 | * If this was the removal of the original device in a hot spare vdev, | |
5372 | * then we want to go through and remove the device from the hot spare | |
5373 | * list of every other pool. | |
5374 | */ | |
5375 | if (unspare) { | |
572e2857 BB |
5376 | spa_t *altspa = NULL; |
5377 | ||
428870ff | 5378 | mutex_enter(&spa_namespace_lock); |
572e2857 BB |
5379 | while ((altspa = spa_next(altspa)) != NULL) { |
5380 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
5381 | altspa == spa) | |
428870ff | 5382 | continue; |
572e2857 BB |
5383 | |
5384 | spa_open_ref(altspa, FTAG); | |
428870ff | 5385 | mutex_exit(&spa_namespace_lock); |
572e2857 | 5386 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); |
428870ff | 5387 | mutex_enter(&spa_namespace_lock); |
572e2857 | 5388 | spa_close(altspa, FTAG); |
428870ff BB |
5389 | } |
5390 | mutex_exit(&spa_namespace_lock); | |
572e2857 BB |
5391 | |
5392 | /* search the rest of the vdevs for spares to remove */ | |
5393 | spa_vdev_resilver_done(spa); | |
428870ff BB |
5394 | } |
5395 | ||
572e2857 BB |
5396 | /* all done with the spa; OK to release */ |
5397 | mutex_enter(&spa_namespace_lock); | |
5398 | spa_close(spa, FTAG); | |
5399 | mutex_exit(&spa_namespace_lock); | |
5400 | ||
428870ff BB |
5401 | return (error); |
5402 | } | |
5403 | ||
5404 | /* | |
5405 | * Split a set of devices from their mirrors, and create a new pool from them. | |
5406 | */ | |
5407 | int | |
5408 | spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, | |
5409 | nvlist_t *props, boolean_t exp) | |
5410 | { | |
5411 | int error = 0; | |
5412 | uint64_t txg, *glist; | |
5413 | spa_t *newspa; | |
5414 | uint_t c, children, lastlog; | |
5415 | nvlist_t **child, *nvl, *tmp; | |
5416 | dmu_tx_t *tx; | |
5417 | char *altroot = NULL; | |
5418 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ | |
5419 | boolean_t activate_slog; | |
5420 | ||
572e2857 | 5421 | ASSERT(spa_writeable(spa)); |
428870ff BB |
5422 | |
5423 | txg = spa_vdev_enter(spa); | |
5424 | ||
5425 | /* clear the log and flush everything up to now */ | |
5426 | activate_slog = spa_passivate_log(spa); | |
5427 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
a1d477c2 | 5428 | error = spa_reset_logs(spa); |
428870ff BB |
5429 | txg = spa_vdev_config_enter(spa); |
5430 | ||
5431 | if (activate_slog) | |
5432 | spa_activate_log(spa); | |
5433 | ||
5434 | if (error != 0) | |
5435 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5436 | ||
5437 | /* check new spa name before going any further */ | |
5438 | if (spa_lookup(newname) != NULL) | |
5439 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
5440 | ||
5441 | /* | |
5442 | * scan through all the children to ensure they're all mirrors | |
5443 | */ | |
5444 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
5445 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
5446 | &children) != 0) | |
5447 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5448 | ||
5449 | /* first, check to ensure we've got the right child count */ | |
5450 | rvd = spa->spa_root_vdev; | |
5451 | lastlog = 0; | |
5452 | for (c = 0; c < rvd->vdev_children; c++) { | |
5453 | vdev_t *vd = rvd->vdev_child[c]; | |
5454 | ||
5455 | /* don't count the holes & logs as children */ | |
a1d477c2 | 5456 | if (vd->vdev_islog || !vdev_is_concrete(vd)) { |
428870ff BB |
5457 | if (lastlog == 0) |
5458 | lastlog = c; | |
5459 | continue; | |
5460 | } | |
5461 | ||
5462 | lastlog = 0; | |
5463 | } | |
5464 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
5465 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5466 | ||
5467 | /* next, ensure no spare or cache devices are part of the split */ | |
5468 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
5469 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
5470 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5471 | ||
79c76d5b BB |
5472 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); |
5473 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
428870ff BB |
5474 | |
5475 | /* then, loop over each vdev and validate it */ | |
5476 | for (c = 0; c < children; c++) { | |
5477 | uint64_t is_hole = 0; | |
5478 | ||
5479 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
5480 | &is_hole); | |
5481 | ||
5482 | if (is_hole != 0) { | |
5483 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
5484 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
5485 | continue; | |
5486 | } else { | |
2e528b49 | 5487 | error = SET_ERROR(EINVAL); |
428870ff BB |
5488 | break; |
5489 | } | |
5490 | } | |
5491 | ||
5492 | /* which disk is going to be split? */ | |
5493 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
5494 | &glist[c]) != 0) { | |
2e528b49 | 5495 | error = SET_ERROR(EINVAL); |
428870ff BB |
5496 | break; |
5497 | } | |
5498 | ||
5499 | /* look it up in the spa */ | |
5500 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
5501 | if (vml[c] == NULL) { | |
2e528b49 | 5502 | error = SET_ERROR(ENODEV); |
428870ff BB |
5503 | break; |
5504 | } | |
5505 | ||
5506 | /* make sure there's nothing stopping the split */ | |
5507 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
5508 | vml[c]->vdev_islog || | |
a1d477c2 | 5509 | !vdev_is_concrete(vml[c]) || |
428870ff BB |
5510 | vml[c]->vdev_isspare || |
5511 | vml[c]->vdev_isl2cache || | |
5512 | !vdev_writeable(vml[c]) || | |
5513 | vml[c]->vdev_children != 0 || | |
5514 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
5515 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
2e528b49 | 5516 | error = SET_ERROR(EINVAL); |
428870ff BB |
5517 | break; |
5518 | } | |
5519 | ||
5520 | if (vdev_dtl_required(vml[c])) { | |
2e528b49 | 5521 | error = SET_ERROR(EBUSY); |
428870ff BB |
5522 | break; |
5523 | } | |
5524 | ||
5525 | /* we need certain info from the top level */ | |
5526 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, | |
5527 | vml[c]->vdev_top->vdev_ms_array) == 0); | |
5528 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
5529 | vml[c]->vdev_top->vdev_ms_shift) == 0); | |
5530 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
5531 | vml[c]->vdev_top->vdev_asize) == 0); | |
5532 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
5533 | vml[c]->vdev_top->vdev_ashift) == 0); | |
e0ab3ab5 JS |
5534 | |
5535 | /* transfer per-vdev ZAPs */ | |
5536 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
5537 | VERIFY0(nvlist_add_uint64(child[c], | |
5538 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
5539 | ||
5540 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
5541 | VERIFY0(nvlist_add_uint64(child[c], | |
5542 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
5543 | vml[c]->vdev_parent->vdev_top_zap)); | |
428870ff BB |
5544 | } |
5545 | ||
5546 | if (error != 0) { | |
5547 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5548 | kmem_free(glist, children * sizeof (uint64_t)); | |
5549 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5550 | } | |
5551 | ||
5552 | /* stop writers from using the disks */ | |
5553 | for (c = 0; c < children; c++) { | |
5554 | if (vml[c] != NULL) | |
5555 | vml[c]->vdev_offline = B_TRUE; | |
5556 | } | |
5557 | vdev_reopen(spa->spa_root_vdev); | |
34dc7c2f BB |
5558 | |
5559 | /* | |
428870ff BB |
5560 | * Temporarily record the splitting vdevs in the spa config. This |
5561 | * will disappear once the config is regenerated. | |
34dc7c2f | 5562 | */ |
79c76d5b | 5563 | VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); |
428870ff BB |
5564 | VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, |
5565 | glist, children) == 0); | |
5566 | kmem_free(glist, children * sizeof (uint64_t)); | |
34dc7c2f | 5567 | |
428870ff BB |
5568 | mutex_enter(&spa->spa_props_lock); |
5569 | VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, | |
5570 | nvl) == 0); | |
5571 | mutex_exit(&spa->spa_props_lock); | |
5572 | spa->spa_config_splitting = nvl; | |
5573 | vdev_config_dirty(spa->spa_root_vdev); | |
5574 | ||
5575 | /* configure and create the new pool */ | |
5576 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); | |
5577 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
5578 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); | |
5579 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
5580 | spa_version(spa)) == 0); | |
5581 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
5582 | spa->spa_config_txg) == 0); | |
5583 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
5584 | spa_generate_guid(NULL)) == 0); | |
e0ab3ab5 | 5585 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); |
428870ff BB |
5586 | (void) nvlist_lookup_string(props, |
5587 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
34dc7c2f | 5588 | |
428870ff BB |
5589 | /* add the new pool to the namespace */ |
5590 | newspa = spa_add(newname, config, altroot); | |
e0ab3ab5 | 5591 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
5592 | newspa->spa_config_txg = spa->spa_config_txg; |
5593 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
5594 | ||
5595 | /* release the spa config lock, retaining the namespace lock */ | |
5596 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5597 | ||
5598 | if (zio_injection_enabled) | |
5599 | zio_handle_panic_injection(spa, FTAG, 1); | |
5600 | ||
5601 | spa_activate(newspa, spa_mode_global); | |
5602 | spa_async_suspend(newspa); | |
5603 | ||
5604 | /* create the new pool from the disks of the original pool */ | |
5605 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); | |
5606 | if (error) | |
5607 | goto out; | |
5608 | ||
5609 | /* if that worked, generate a real config for the new pool */ | |
5610 | if (newspa->spa_root_vdev != NULL) { | |
5611 | VERIFY(nvlist_alloc(&newspa->spa_config_splitting, | |
79c76d5b | 5612 | NV_UNIQUE_NAME, KM_SLEEP) == 0); |
428870ff BB |
5613 | VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, |
5614 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); | |
5615 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, | |
5616 | B_TRUE)); | |
9babb374 | 5617 | } |
34dc7c2f | 5618 | |
428870ff BB |
5619 | /* set the props */ |
5620 | if (props != NULL) { | |
5621 | spa_configfile_set(newspa, props, B_FALSE); | |
5622 | error = spa_prop_set(newspa, props); | |
5623 | if (error) | |
5624 | goto out; | |
5625 | } | |
34dc7c2f | 5626 | |
428870ff BB |
5627 | /* flush everything */ |
5628 | txg = spa_vdev_config_enter(newspa); | |
5629 | vdev_config_dirty(newspa->spa_root_vdev); | |
5630 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
34dc7c2f | 5631 | |
428870ff BB |
5632 | if (zio_injection_enabled) |
5633 | zio_handle_panic_injection(spa, FTAG, 2); | |
34dc7c2f | 5634 | |
428870ff | 5635 | spa_async_resume(newspa); |
34dc7c2f | 5636 | |
428870ff BB |
5637 | /* finally, update the original pool's config */ |
5638 | txg = spa_vdev_config_enter(spa); | |
5639 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
5640 | error = dmu_tx_assign(tx, TXG_WAIT); | |
5641 | if (error != 0) | |
5642 | dmu_tx_abort(tx); | |
5643 | for (c = 0; c < children; c++) { | |
5644 | if (vml[c] != NULL) { | |
5645 | vdev_split(vml[c]); | |
5646 | if (error == 0) | |
6f1ffb06 MA |
5647 | spa_history_log_internal(spa, "detach", tx, |
5648 | "vdev=%s", vml[c]->vdev_path); | |
e0ab3ab5 | 5649 | |
428870ff | 5650 | vdev_free(vml[c]); |
34dc7c2f | 5651 | } |
34dc7c2f | 5652 | } |
e0ab3ab5 | 5653 | spa->spa_avz_action = AVZ_ACTION_REBUILD; |
428870ff BB |
5654 | vdev_config_dirty(spa->spa_root_vdev); |
5655 | spa->spa_config_splitting = NULL; | |
5656 | nvlist_free(nvl); | |
5657 | if (error == 0) | |
5658 | dmu_tx_commit(tx); | |
5659 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
5660 | ||
5661 | if (zio_injection_enabled) | |
5662 | zio_handle_panic_injection(spa, FTAG, 3); | |
5663 | ||
5664 | /* split is complete; log a history record */ | |
6f1ffb06 MA |
5665 | spa_history_log_internal(newspa, "split", NULL, |
5666 | "from pool %s", spa_name(spa)); | |
428870ff BB |
5667 | |
5668 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5669 | ||
5670 | /* if we're not going to mount the filesystems in userland, export */ | |
5671 | if (exp) | |
5672 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
5673 | B_FALSE, B_FALSE); | |
5674 | ||
5675 | return (error); | |
5676 | ||
5677 | out: | |
5678 | spa_unload(newspa); | |
5679 | spa_deactivate(newspa); | |
5680 | spa_remove(newspa); | |
5681 | ||
5682 | txg = spa_vdev_config_enter(spa); | |
5683 | ||
5684 | /* re-online all offlined disks */ | |
5685 | for (c = 0; c < children; c++) { | |
5686 | if (vml[c] != NULL) | |
5687 | vml[c]->vdev_offline = B_FALSE; | |
5688 | } | |
5689 | vdev_reopen(spa->spa_root_vdev); | |
5690 | ||
5691 | nvlist_free(spa->spa_config_splitting); | |
5692 | spa->spa_config_splitting = NULL; | |
5693 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
34dc7c2f | 5694 | |
428870ff | 5695 | kmem_free(vml, children * sizeof (vdev_t *)); |
34dc7c2f BB |
5696 | return (error); |
5697 | } | |
5698 | ||
34dc7c2f BB |
5699 | /* |
5700 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
d3cc8b15 | 5701 | * currently spared, so we can detach it. |
34dc7c2f BB |
5702 | */ |
5703 | static vdev_t * | |
5704 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
5705 | { | |
5706 | vdev_t *newvd, *oldvd; | |
34dc7c2f | 5707 | |
1c27024e | 5708 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
5709 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); |
5710 | if (oldvd != NULL) | |
5711 | return (oldvd); | |
5712 | } | |
5713 | ||
5714 | /* | |
572e2857 BB |
5715 | * Check for a completed replacement. We always consider the first |
5716 | * vdev in the list to be the oldest vdev, and the last one to be | |
5717 | * the newest (see spa_vdev_attach() for how that works). In | |
5718 | * the case where the newest vdev is faulted, we will not automatically | |
5719 | * remove it after a resilver completes. This is OK as it will require | |
5720 | * user intervention to determine which disk the admin wishes to keep. | |
34dc7c2f | 5721 | */ |
572e2857 BB |
5722 | if (vd->vdev_ops == &vdev_replacing_ops) { |
5723 | ASSERT(vd->vdev_children > 1); | |
5724 | ||
5725 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
34dc7c2f | 5726 | oldvd = vd->vdev_child[0]; |
34dc7c2f | 5727 | |
fb5f0bc8 | 5728 | if (vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 5729 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
fb5f0bc8 | 5730 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 5731 | return (oldvd); |
34dc7c2f BB |
5732 | } |
5733 | ||
5734 | /* | |
5735 | * Check for a completed resilver with the 'unspare' flag set. | |
5736 | */ | |
572e2857 BB |
5737 | if (vd->vdev_ops == &vdev_spare_ops) { |
5738 | vdev_t *first = vd->vdev_child[0]; | |
5739 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
5740 | ||
5741 | if (last->vdev_unspare) { | |
5742 | oldvd = first; | |
5743 | newvd = last; | |
5744 | } else if (first->vdev_unspare) { | |
5745 | oldvd = last; | |
5746 | newvd = first; | |
5747 | } else { | |
5748 | oldvd = NULL; | |
5749 | } | |
34dc7c2f | 5750 | |
572e2857 | 5751 | if (oldvd != NULL && |
fb5f0bc8 | 5752 | vdev_dtl_empty(newvd, DTL_MISSING) && |
428870ff | 5753 | vdev_dtl_empty(newvd, DTL_OUTAGE) && |
572e2857 | 5754 | !vdev_dtl_required(oldvd)) |
34dc7c2f | 5755 | return (oldvd); |
572e2857 BB |
5756 | |
5757 | /* | |
5758 | * If there are more than two spares attached to a disk, | |
5759 | * and those spares are not required, then we want to | |
5760 | * attempt to free them up now so that they can be used | |
5761 | * by other pools. Once we're back down to a single | |
5762 | * disk+spare, we stop removing them. | |
5763 | */ | |
5764 | if (vd->vdev_children > 2) { | |
5765 | newvd = vd->vdev_child[1]; | |
5766 | ||
5767 | if (newvd->vdev_isspare && last->vdev_isspare && | |
5768 | vdev_dtl_empty(last, DTL_MISSING) && | |
5769 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
5770 | !vdev_dtl_required(newvd)) | |
5771 | return (newvd); | |
34dc7c2f | 5772 | } |
34dc7c2f BB |
5773 | } |
5774 | ||
5775 | return (NULL); | |
5776 | } | |
5777 | ||
5778 | static void | |
5779 | spa_vdev_resilver_done(spa_t *spa) | |
5780 | { | |
fb5f0bc8 BB |
5781 | vdev_t *vd, *pvd, *ppvd; |
5782 | uint64_t guid, sguid, pguid, ppguid; | |
34dc7c2f | 5783 | |
fb5f0bc8 | 5784 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
5785 | |
5786 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
fb5f0bc8 BB |
5787 | pvd = vd->vdev_parent; |
5788 | ppvd = pvd->vdev_parent; | |
34dc7c2f | 5789 | guid = vd->vdev_guid; |
fb5f0bc8 BB |
5790 | pguid = pvd->vdev_guid; |
5791 | ppguid = ppvd->vdev_guid; | |
5792 | sguid = 0; | |
34dc7c2f BB |
5793 | /* |
5794 | * If we have just finished replacing a hot spared device, then | |
5795 | * we need to detach the parent's first child (the original hot | |
5796 | * spare) as well. | |
5797 | */ | |
572e2857 BB |
5798 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && |
5799 | ppvd->vdev_children == 2) { | |
34dc7c2f | 5800 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); |
fb5f0bc8 | 5801 | sguid = ppvd->vdev_child[1]->vdev_guid; |
34dc7c2f | 5802 | } |
5d1f7fb6 GW |
5803 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); |
5804 | ||
fb5f0bc8 BB |
5805 | spa_config_exit(spa, SCL_ALL, FTAG); |
5806 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
34dc7c2f | 5807 | return; |
fb5f0bc8 | 5808 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) |
34dc7c2f | 5809 | return; |
fb5f0bc8 | 5810 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
5811 | } |
5812 | ||
fb5f0bc8 | 5813 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
5814 | } |
5815 | ||
5816 | /* | |
428870ff | 5817 | * Update the stored path or FRU for this vdev. |
34dc7c2f BB |
5818 | */ |
5819 | int | |
9babb374 BB |
5820 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, |
5821 | boolean_t ispath) | |
34dc7c2f | 5822 | { |
b128c09f | 5823 | vdev_t *vd; |
428870ff | 5824 | boolean_t sync = B_FALSE; |
34dc7c2f | 5825 | |
572e2857 BB |
5826 | ASSERT(spa_writeable(spa)); |
5827 | ||
428870ff | 5828 | spa_vdev_state_enter(spa, SCL_ALL); |
34dc7c2f | 5829 | |
9babb374 | 5830 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
428870ff | 5831 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); |
34dc7c2f BB |
5832 | |
5833 | if (!vd->vdev_ops->vdev_op_leaf) | |
428870ff | 5834 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 5835 | |
9babb374 | 5836 | if (ispath) { |
428870ff BB |
5837 | if (strcmp(value, vd->vdev_path) != 0) { |
5838 | spa_strfree(vd->vdev_path); | |
5839 | vd->vdev_path = spa_strdup(value); | |
5840 | sync = B_TRUE; | |
5841 | } | |
9babb374 | 5842 | } else { |
428870ff BB |
5843 | if (vd->vdev_fru == NULL) { |
5844 | vd->vdev_fru = spa_strdup(value); | |
5845 | sync = B_TRUE; | |
5846 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
9babb374 | 5847 | spa_strfree(vd->vdev_fru); |
428870ff BB |
5848 | vd->vdev_fru = spa_strdup(value); |
5849 | sync = B_TRUE; | |
5850 | } | |
9babb374 | 5851 | } |
34dc7c2f | 5852 | |
428870ff | 5853 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); |
34dc7c2f BB |
5854 | } |
5855 | ||
9babb374 BB |
5856 | int |
5857 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
5858 | { | |
5859 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
5860 | } | |
5861 | ||
5862 | int | |
5863 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
5864 | { | |
5865 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
5866 | } | |
5867 | ||
34dc7c2f BB |
5868 | /* |
5869 | * ========================================================================== | |
428870ff | 5870 | * SPA Scanning |
34dc7c2f BB |
5871 | * ========================================================================== |
5872 | */ | |
0ea05c64 AP |
5873 | int |
5874 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
5875 | { | |
5876 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5877 | ||
5878 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
5879 | return (SET_ERROR(EBUSY)); | |
5880 | ||
5881 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
5882 | } | |
34dc7c2f | 5883 | |
34dc7c2f | 5884 | int |
428870ff BB |
5885 | spa_scan_stop(spa_t *spa) |
5886 | { | |
5887 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5888 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
2e528b49 | 5889 | return (SET_ERROR(EBUSY)); |
428870ff BB |
5890 | return (dsl_scan_cancel(spa->spa_dsl_pool)); |
5891 | } | |
5892 | ||
5893 | int | |
5894 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
34dc7c2f | 5895 | { |
b128c09f | 5896 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); |
34dc7c2f | 5897 | |
428870ff | 5898 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) |
2e528b49 | 5899 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 5900 | |
34dc7c2f | 5901 | /* |
b128c09f BB |
5902 | * If a resilver was requested, but there is no DTL on a |
5903 | * writeable leaf device, we have nothing to do. | |
34dc7c2f | 5904 | */ |
428870ff | 5905 | if (func == POOL_SCAN_RESILVER && |
b128c09f BB |
5906 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { |
5907 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
34dc7c2f BB |
5908 | return (0); |
5909 | } | |
5910 | ||
428870ff | 5911 | return (dsl_scan(spa->spa_dsl_pool, func)); |
34dc7c2f BB |
5912 | } |
5913 | ||
5914 | /* | |
5915 | * ========================================================================== | |
5916 | * SPA async task processing | |
5917 | * ========================================================================== | |
5918 | */ | |
5919 | ||
5920 | static void | |
5921 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
5922 | { | |
b128c09f | 5923 | if (vd->vdev_remove_wanted) { |
428870ff BB |
5924 | vd->vdev_remove_wanted = B_FALSE; |
5925 | vd->vdev_delayed_close = B_FALSE; | |
b128c09f | 5926 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); |
428870ff BB |
5927 | |
5928 | /* | |
5929 | * We want to clear the stats, but we don't want to do a full | |
5930 | * vdev_clear() as that will cause us to throw away | |
5931 | * degraded/faulted state as well as attempt to reopen the | |
5932 | * device, all of which is a waste. | |
5933 | */ | |
5934 | vd->vdev_stat.vs_read_errors = 0; | |
5935 | vd->vdev_stat.vs_write_errors = 0; | |
5936 | vd->vdev_stat.vs_checksum_errors = 0; | |
5937 | ||
b128c09f BB |
5938 | vdev_state_dirty(vd->vdev_top); |
5939 | } | |
34dc7c2f | 5940 | |
1c27024e | 5941 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f BB |
5942 | spa_async_remove(spa, vd->vdev_child[c]); |
5943 | } | |
5944 | ||
5945 | static void | |
5946 | spa_async_probe(spa_t *spa, vdev_t *vd) | |
5947 | { | |
5948 | if (vd->vdev_probe_wanted) { | |
428870ff | 5949 | vd->vdev_probe_wanted = B_FALSE; |
b128c09f | 5950 | vdev_reopen(vd); /* vdev_open() does the actual probe */ |
34dc7c2f | 5951 | } |
b128c09f | 5952 | |
1c27024e | 5953 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f | 5954 | spa_async_probe(spa, vd->vdev_child[c]); |
34dc7c2f BB |
5955 | } |
5956 | ||
9babb374 BB |
5957 | static void |
5958 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
5959 | { | |
9babb374 BB |
5960 | if (!spa->spa_autoexpand) |
5961 | return; | |
5962 | ||
1c27024e | 5963 | for (int c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
5964 | vdev_t *cvd = vd->vdev_child[c]; |
5965 | spa_async_autoexpand(spa, cvd); | |
5966 | } | |
5967 | ||
5968 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
5969 | return; | |
5970 | ||
12fa0466 | 5971 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); |
9babb374 BB |
5972 | } |
5973 | ||
34dc7c2f | 5974 | static void |
c25b8f99 | 5975 | spa_async_thread(void *arg) |
34dc7c2f | 5976 | { |
c25b8f99 | 5977 | spa_t *spa = (spa_t *)arg; |
867959b5 | 5978 | int tasks; |
34dc7c2f BB |
5979 | |
5980 | ASSERT(spa->spa_sync_on); | |
5981 | ||
5982 | mutex_enter(&spa->spa_async_lock); | |
5983 | tasks = spa->spa_async_tasks; | |
5984 | spa->spa_async_tasks = 0; | |
5985 | mutex_exit(&spa->spa_async_lock); | |
5986 | ||
5987 | /* | |
5988 | * See if the config needs to be updated. | |
5989 | */ | |
5990 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
428870ff | 5991 | uint64_t old_space, new_space; |
9babb374 | 5992 | |
34dc7c2f | 5993 | mutex_enter(&spa_namespace_lock); |
428870ff | 5994 | old_space = metaslab_class_get_space(spa_normal_class(spa)); |
34dc7c2f | 5995 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
428870ff | 5996 | new_space = metaslab_class_get_space(spa_normal_class(spa)); |
34dc7c2f | 5997 | mutex_exit(&spa_namespace_lock); |
9babb374 BB |
5998 | |
5999 | /* | |
6000 | * If the pool grew as a result of the config update, | |
6001 | * then log an internal history event. | |
6002 | */ | |
428870ff | 6003 | if (new_space != old_space) { |
6f1ffb06 | 6004 | spa_history_log_internal(spa, "vdev online", NULL, |
45d1cae3 | 6005 | "pool '%s' size: %llu(+%llu)", |
428870ff | 6006 | spa_name(spa), new_space, new_space - old_space); |
9babb374 | 6007 | } |
34dc7c2f BB |
6008 | } |
6009 | ||
6010 | /* | |
6011 | * See if any devices need to be marked REMOVED. | |
34dc7c2f | 6012 | */ |
b128c09f | 6013 | if (tasks & SPA_ASYNC_REMOVE) { |
428870ff | 6014 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 6015 | spa_async_remove(spa, spa->spa_root_vdev); |
867959b5 | 6016 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) |
b128c09f | 6017 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); |
867959b5 | 6018 | for (int i = 0; i < spa->spa_spares.sav_count; i++) |
b128c09f BB |
6019 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); |
6020 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
34dc7c2f BB |
6021 | } |
6022 | ||
9babb374 BB |
6023 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { |
6024 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6025 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
6026 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
6027 | } | |
6028 | ||
34dc7c2f | 6029 | /* |
b128c09f | 6030 | * See if any devices need to be probed. |
34dc7c2f | 6031 | */ |
b128c09f | 6032 | if (tasks & SPA_ASYNC_PROBE) { |
428870ff | 6033 | spa_vdev_state_enter(spa, SCL_NONE); |
b128c09f BB |
6034 | spa_async_probe(spa, spa->spa_root_vdev); |
6035 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
6036 | } | |
34dc7c2f BB |
6037 | |
6038 | /* | |
b128c09f | 6039 | * If any devices are done replacing, detach them. |
34dc7c2f | 6040 | */ |
b128c09f BB |
6041 | if (tasks & SPA_ASYNC_RESILVER_DONE) |
6042 | spa_vdev_resilver_done(spa); | |
34dc7c2f BB |
6043 | |
6044 | /* | |
6045 | * Kick off a resilver. | |
6046 | */ | |
b128c09f | 6047 | if (tasks & SPA_ASYNC_RESILVER) |
428870ff | 6048 | dsl_resilver_restart(spa->spa_dsl_pool, 0); |
34dc7c2f BB |
6049 | |
6050 | /* | |
6051 | * Let the world know that we're done. | |
6052 | */ | |
6053 | mutex_enter(&spa->spa_async_lock); | |
6054 | spa->spa_async_thread = NULL; | |
6055 | cv_broadcast(&spa->spa_async_cv); | |
6056 | mutex_exit(&spa->spa_async_lock); | |
6057 | thread_exit(); | |
6058 | } | |
6059 | ||
6060 | void | |
6061 | spa_async_suspend(spa_t *spa) | |
6062 | { | |
6063 | mutex_enter(&spa->spa_async_lock); | |
6064 | spa->spa_async_suspended++; | |
a1d477c2 MA |
6065 | while (spa->spa_async_thread != NULL || |
6066 | spa->spa_condense_thread != NULL) | |
34dc7c2f BB |
6067 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); |
6068 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 MA |
6069 | |
6070 | spa_vdev_remove_suspend(spa); | |
34dc7c2f BB |
6071 | } |
6072 | ||
6073 | void | |
6074 | spa_async_resume(spa_t *spa) | |
6075 | { | |
6076 | mutex_enter(&spa->spa_async_lock); | |
6077 | ASSERT(spa->spa_async_suspended != 0); | |
6078 | spa->spa_async_suspended--; | |
6079 | mutex_exit(&spa->spa_async_lock); | |
a1d477c2 | 6080 | spa_restart_removal(spa); |
34dc7c2f BB |
6081 | } |
6082 | ||
e6cfd633 WA |
6083 | static boolean_t |
6084 | spa_async_tasks_pending(spa_t *spa) | |
6085 | { | |
6086 | uint_t non_config_tasks; | |
6087 | uint_t config_task; | |
6088 | boolean_t config_task_suspended; | |
6089 | ||
6090 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
6091 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
6092 | if (spa->spa_ccw_fail_time == 0) { | |
6093 | config_task_suspended = B_FALSE; | |
6094 | } else { | |
6095 | config_task_suspended = | |
6096 | (gethrtime() - spa->spa_ccw_fail_time) < | |
05852b34 | 6097 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); |
e6cfd633 WA |
6098 | } |
6099 | ||
6100 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
6101 | } | |
6102 | ||
34dc7c2f BB |
6103 | static void |
6104 | spa_async_dispatch(spa_t *spa) | |
6105 | { | |
6106 | mutex_enter(&spa->spa_async_lock); | |
e6cfd633 WA |
6107 | if (spa_async_tasks_pending(spa) && |
6108 | !spa->spa_async_suspended && | |
34dc7c2f | 6109 | spa->spa_async_thread == NULL && |
e6cfd633 | 6110 | rootdir != NULL) |
34dc7c2f BB |
6111 | spa->spa_async_thread = thread_create(NULL, 0, |
6112 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
6113 | mutex_exit(&spa->spa_async_lock); | |
6114 | } | |
6115 | ||
6116 | void | |
6117 | spa_async_request(spa_t *spa, int task) | |
6118 | { | |
428870ff | 6119 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); |
34dc7c2f BB |
6120 | mutex_enter(&spa->spa_async_lock); |
6121 | spa->spa_async_tasks |= task; | |
6122 | mutex_exit(&spa->spa_async_lock); | |
6123 | } | |
6124 | ||
6125 | /* | |
6126 | * ========================================================================== | |
6127 | * SPA syncing routines | |
6128 | * ========================================================================== | |
6129 | */ | |
6130 | ||
428870ff BB |
6131 | static int |
6132 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
34dc7c2f | 6133 | { |
428870ff BB |
6134 | bpobj_t *bpo = arg; |
6135 | bpobj_enqueue(bpo, bp, tx); | |
6136 | return (0); | |
6137 | } | |
34dc7c2f | 6138 | |
428870ff BB |
6139 | static int |
6140 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
6141 | { | |
6142 | zio_t *zio = arg; | |
34dc7c2f | 6143 | |
428870ff BB |
6144 | zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, |
6145 | zio->io_flags)); | |
6146 | return (0); | |
34dc7c2f BB |
6147 | } |
6148 | ||
e8b96c60 MA |
6149 | /* |
6150 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6151 | * amount of time spent syncing frees. | |
6152 | */ | |
6153 | static void | |
6154 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
6155 | { | |
6156 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6157 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
6158 | VERIFY(zio_wait(zio) == 0); | |
6159 | } | |
6160 | ||
6161 | /* | |
6162 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6163 | * amount of time spent syncing deferred frees. | |
6164 | */ | |
6165 | static void | |
6166 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
6167 | { | |
6168 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6169 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
6170 | spa_free_sync_cb, zio, tx), ==, 0); | |
6171 | VERIFY0(zio_wait(zio)); | |
6172 | } | |
6173 | ||
34dc7c2f BB |
6174 | static void |
6175 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
6176 | { | |
6177 | char *packed = NULL; | |
b128c09f | 6178 | size_t bufsize; |
34dc7c2f BB |
6179 | size_t nvsize = 0; |
6180 | dmu_buf_t *db; | |
6181 | ||
6182 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
6183 | ||
b128c09f BB |
6184 | /* |
6185 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
b0bc7a84 | 6186 | * information. This avoids the dmu_buf_will_dirty() path and |
b128c09f BB |
6187 | * saves us a pre-read to get data we don't actually care about. |
6188 | */ | |
9ae529ec | 6189 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); |
79c76d5b | 6190 | packed = vmem_alloc(bufsize, KM_SLEEP); |
34dc7c2f BB |
6191 | |
6192 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
79c76d5b | 6193 | KM_SLEEP) == 0); |
b128c09f | 6194 | bzero(packed + nvsize, bufsize - nvsize); |
34dc7c2f | 6195 | |
b128c09f | 6196 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); |
34dc7c2f | 6197 | |
00b46022 | 6198 | vmem_free(packed, bufsize); |
34dc7c2f BB |
6199 | |
6200 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
6201 | dmu_buf_will_dirty(db, tx); | |
6202 | *(uint64_t *)db->db_data = nvsize; | |
6203 | dmu_buf_rele(db, FTAG); | |
6204 | } | |
6205 | ||
6206 | static void | |
6207 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
6208 | const char *config, const char *entry) | |
6209 | { | |
6210 | nvlist_t *nvroot; | |
6211 | nvlist_t **list; | |
6212 | int i; | |
6213 | ||
6214 | if (!sav->sav_sync) | |
6215 | return; | |
6216 | ||
6217 | /* | |
6218 | * Update the MOS nvlist describing the list of available devices. | |
6219 | * spa_validate_aux() will have already made sure this nvlist is | |
6220 | * valid and the vdevs are labeled appropriately. | |
6221 | */ | |
6222 | if (sav->sav_object == 0) { | |
6223 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
6224 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
6225 | sizeof (uint64_t), tx); | |
6226 | VERIFY(zap_update(spa->spa_meta_objset, | |
6227 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
6228 | &sav->sav_object, tx) == 0); | |
6229 | } | |
6230 | ||
79c76d5b | 6231 | VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); |
34dc7c2f BB |
6232 | if (sav->sav_count == 0) { |
6233 | VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); | |
6234 | } else { | |
79c76d5b | 6235 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); |
34dc7c2f BB |
6236 | for (i = 0; i < sav->sav_count; i++) |
6237 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
428870ff | 6238 | B_FALSE, VDEV_CONFIG_L2CACHE); |
34dc7c2f BB |
6239 | VERIFY(nvlist_add_nvlist_array(nvroot, config, list, |
6240 | sav->sav_count) == 0); | |
6241 | for (i = 0; i < sav->sav_count; i++) | |
6242 | nvlist_free(list[i]); | |
6243 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
6244 | } | |
6245 | ||
6246 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
6247 | nvlist_free(nvroot); | |
6248 | ||
6249 | sav->sav_sync = B_FALSE; | |
6250 | } | |
6251 | ||
e0ab3ab5 JS |
6252 | /* |
6253 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
6254 | * The all-vdev ZAP must be empty. | |
6255 | */ | |
6256 | static void | |
6257 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
6258 | { | |
6259 | spa_t *spa = vd->vdev_spa; | |
e0ab3ab5 JS |
6260 | |
6261 | if (vd->vdev_top_zap != 0) { | |
6262 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6263 | vd->vdev_top_zap, tx)); | |
6264 | } | |
6265 | if (vd->vdev_leaf_zap != 0) { | |
6266 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6267 | vd->vdev_leaf_zap, tx)); | |
6268 | } | |
1c27024e | 6269 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
6270 | spa_avz_build(vd->vdev_child[i], avz, tx); |
6271 | } | |
6272 | } | |
6273 | ||
34dc7c2f BB |
6274 | static void |
6275 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
6276 | { | |
6277 | nvlist_t *config; | |
6278 | ||
e0ab3ab5 JS |
6279 | /* |
6280 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
6281 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
6282 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
6283 | * need to rebuild the AVZ although the config may not be dirty. | |
6284 | */ | |
6285 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
6286 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
34dc7c2f BB |
6287 | return; |
6288 | ||
b128c09f BB |
6289 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
6290 | ||
e0ab3ab5 | 6291 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || |
38640550 | 6292 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || |
e0ab3ab5 JS |
6293 | spa->spa_all_vdev_zaps != 0); |
6294 | ||
6295 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
e0ab3ab5 JS |
6296 | /* Make and build the new AVZ */ |
6297 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
6298 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
6299 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
6300 | ||
6301 | /* Diff old AVZ with new one */ | |
1c27024e DB |
6302 | zap_cursor_t zc; |
6303 | zap_attribute_t za; | |
6304 | ||
e0ab3ab5 JS |
6305 | for (zap_cursor_init(&zc, spa->spa_meta_objset, |
6306 | spa->spa_all_vdev_zaps); | |
6307 | zap_cursor_retrieve(&zc, &za) == 0; | |
6308 | zap_cursor_advance(&zc)) { | |
6309 | uint64_t vdzap = za.za_first_integer; | |
6310 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
6311 | vdzap) == ENOENT) { | |
6312 | /* | |
6313 | * ZAP is listed in old AVZ but not in new one; | |
6314 | * destroy it | |
6315 | */ | |
6316 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
6317 | tx)); | |
6318 | } | |
6319 | } | |
6320 | ||
6321 | zap_cursor_fini(&zc); | |
6322 | ||
6323 | /* Destroy the old AVZ */ | |
6324 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6325 | spa->spa_all_vdev_zaps, tx)); | |
6326 | ||
6327 | /* Replace the old AVZ in the dir obj with the new one */ | |
6328 | VERIFY0(zap_update(spa->spa_meta_objset, | |
6329 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
6330 | sizeof (new_avz), 1, &new_avz, tx)); | |
6331 | ||
6332 | spa->spa_all_vdev_zaps = new_avz; | |
6333 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
6334 | zap_cursor_t zc; | |
6335 | zap_attribute_t za; | |
6336 | ||
6337 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
6338 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6339 | spa->spa_all_vdev_zaps); | |
6340 | zap_cursor_retrieve(&zc, &za) == 0; | |
6341 | zap_cursor_advance(&zc)) { | |
6342 | uint64_t zap = za.za_first_integer; | |
6343 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
6344 | } | |
6345 | ||
6346 | zap_cursor_fini(&zc); | |
6347 | ||
6348 | /* Destroy and unlink the AVZ itself */ | |
6349 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6350 | spa->spa_all_vdev_zaps, tx)); | |
6351 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
6352 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
6353 | spa->spa_all_vdev_zaps = 0; | |
6354 | } | |
6355 | ||
6356 | if (spa->spa_all_vdev_zaps == 0) { | |
6357 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
6358 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
6359 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
6360 | } | |
6361 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
6362 | ||
6363 | /* Create ZAPs for vdevs that don't have them. */ | |
6364 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
6365 | ||
b128c09f BB |
6366 | config = spa_config_generate(spa, spa->spa_root_vdev, |
6367 | dmu_tx_get_txg(tx), B_FALSE); | |
6368 | ||
ea0b2538 GW |
6369 | /* |
6370 | * If we're upgrading the spa version then make sure that | |
6371 | * the config object gets updated with the correct version. | |
6372 | */ | |
6373 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
6374 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
6375 | spa->spa_uberblock.ub_version); | |
6376 | ||
b128c09f | 6377 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f | 6378 | |
8a5fc748 | 6379 | nvlist_free(spa->spa_config_syncing); |
34dc7c2f BB |
6380 | spa->spa_config_syncing = config; |
6381 | ||
6382 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
6383 | } | |
6384 | ||
9ae529ec | 6385 | static void |
13fe0198 | 6386 | spa_sync_version(void *arg, dmu_tx_t *tx) |
9ae529ec | 6387 | { |
13fe0198 MA |
6388 | uint64_t *versionp = arg; |
6389 | uint64_t version = *versionp; | |
6390 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
9ae529ec CS |
6391 | |
6392 | /* | |
6393 | * Setting the version is special cased when first creating the pool. | |
6394 | */ | |
6395 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
6396 | ||
8dca0a9a | 6397 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); |
9ae529ec CS |
6398 | ASSERT(version >= spa_version(spa)); |
6399 | ||
6400 | spa->spa_uberblock.ub_version = version; | |
6401 | vdev_config_dirty(spa->spa_root_vdev); | |
6f1ffb06 | 6402 | spa_history_log_internal(spa, "set", tx, "version=%lld", version); |
9ae529ec CS |
6403 | } |
6404 | ||
34dc7c2f BB |
6405 | /* |
6406 | * Set zpool properties. | |
6407 | */ | |
6408 | static void | |
13fe0198 | 6409 | spa_sync_props(void *arg, dmu_tx_t *tx) |
34dc7c2f | 6410 | { |
13fe0198 MA |
6411 | nvlist_t *nvp = arg; |
6412 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
34dc7c2f | 6413 | objset_t *mos = spa->spa_meta_objset; |
9ae529ec | 6414 | nvpair_t *elem = NULL; |
b128c09f BB |
6415 | |
6416 | mutex_enter(&spa->spa_props_lock); | |
34dc7c2f | 6417 | |
34dc7c2f | 6418 | while ((elem = nvlist_next_nvpair(nvp, elem))) { |
9ae529ec CS |
6419 | uint64_t intval; |
6420 | char *strval, *fname; | |
6421 | zpool_prop_t prop; | |
6422 | const char *propname; | |
6423 | zprop_type_t proptype; | |
fa86b5db | 6424 | spa_feature_t fid; |
9ae529ec | 6425 | |
31864e3d BB |
6426 | switch (prop = zpool_name_to_prop(nvpair_name(elem))) { |
6427 | case ZPOOL_PROP_INVAL: | |
9ae529ec CS |
6428 | /* |
6429 | * We checked this earlier in spa_prop_validate(). | |
6430 | */ | |
6431 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
6432 | ||
6433 | fname = strchr(nvpair_name(elem), '@') + 1; | |
fa86b5db | 6434 | VERIFY0(zfeature_lookup_name(fname, &fid)); |
9ae529ec | 6435 | |
fa86b5db | 6436 | spa_feature_enable(spa, fid, tx); |
6f1ffb06 MA |
6437 | spa_history_log_internal(spa, "set", tx, |
6438 | "%s=enabled", nvpair_name(elem)); | |
9ae529ec CS |
6439 | break; |
6440 | ||
34dc7c2f | 6441 | case ZPOOL_PROP_VERSION: |
93cf2076 | 6442 | intval = fnvpair_value_uint64(elem); |
34dc7c2f | 6443 | /* |
4e33ba4c | 6444 | * The version is synced separately before other |
9ae529ec | 6445 | * properties and should be correct by now. |
34dc7c2f | 6446 | */ |
9ae529ec | 6447 | ASSERT3U(spa_version(spa), >=, intval); |
34dc7c2f BB |
6448 | break; |
6449 | ||
6450 | case ZPOOL_PROP_ALTROOT: | |
6451 | /* | |
6452 | * 'altroot' is a non-persistent property. It should | |
6453 | * have been set temporarily at creation or import time. | |
6454 | */ | |
6455 | ASSERT(spa->spa_root != NULL); | |
6456 | break; | |
6457 | ||
572e2857 | 6458 | case ZPOOL_PROP_READONLY: |
34dc7c2f BB |
6459 | case ZPOOL_PROP_CACHEFILE: |
6460 | /* | |
572e2857 BB |
6461 | * 'readonly' and 'cachefile' are also non-persisitent |
6462 | * properties. | |
34dc7c2f | 6463 | */ |
34dc7c2f | 6464 | break; |
d96eb2b1 | 6465 | case ZPOOL_PROP_COMMENT: |
93cf2076 | 6466 | strval = fnvpair_value_string(elem); |
d96eb2b1 DM |
6467 | if (spa->spa_comment != NULL) |
6468 | spa_strfree(spa->spa_comment); | |
6469 | spa->spa_comment = spa_strdup(strval); | |
6470 | /* | |
6471 | * We need to dirty the configuration on all the vdevs | |
6472 | * so that their labels get updated. It's unnecessary | |
6473 | * to do this for pool creation since the vdev's | |
4e33ba4c | 6474 | * configuration has already been dirtied. |
d96eb2b1 DM |
6475 | */ |
6476 | if (tx->tx_txg != TXG_INITIAL) | |
6477 | vdev_config_dirty(spa->spa_root_vdev); | |
6f1ffb06 MA |
6478 | spa_history_log_internal(spa, "set", tx, |
6479 | "%s=%s", nvpair_name(elem), strval); | |
d96eb2b1 | 6480 | break; |
34dc7c2f BB |
6481 | default: |
6482 | /* | |
6483 | * Set pool property values in the poolprops mos object. | |
6484 | */ | |
34dc7c2f | 6485 | if (spa->spa_pool_props_object == 0) { |
9ae529ec CS |
6486 | spa->spa_pool_props_object = |
6487 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
34dc7c2f | 6488 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, |
9ae529ec | 6489 | tx); |
34dc7c2f | 6490 | } |
34dc7c2f BB |
6491 | |
6492 | /* normalize the property name */ | |
6493 | propname = zpool_prop_to_name(prop); | |
6494 | proptype = zpool_prop_get_type(prop); | |
6495 | ||
6496 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
6497 | ASSERT(proptype == PROP_TYPE_STRING); | |
93cf2076 GW |
6498 | strval = fnvpair_value_string(elem); |
6499 | VERIFY0(zap_update(mos, | |
34dc7c2f | 6500 | spa->spa_pool_props_object, propname, |
93cf2076 | 6501 | 1, strlen(strval) + 1, strval, tx)); |
6f1ffb06 MA |
6502 | spa_history_log_internal(spa, "set", tx, |
6503 | "%s=%s", nvpair_name(elem), strval); | |
34dc7c2f | 6504 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { |
93cf2076 | 6505 | intval = fnvpair_value_uint64(elem); |
34dc7c2f BB |
6506 | |
6507 | if (proptype == PROP_TYPE_INDEX) { | |
6508 | const char *unused; | |
93cf2076 GW |
6509 | VERIFY0(zpool_prop_index_to_string( |
6510 | prop, intval, &unused)); | |
34dc7c2f | 6511 | } |
93cf2076 | 6512 | VERIFY0(zap_update(mos, |
34dc7c2f | 6513 | spa->spa_pool_props_object, propname, |
93cf2076 | 6514 | 8, 1, &intval, tx)); |
6f1ffb06 MA |
6515 | spa_history_log_internal(spa, "set", tx, |
6516 | "%s=%lld", nvpair_name(elem), intval); | |
34dc7c2f BB |
6517 | } else { |
6518 | ASSERT(0); /* not allowed */ | |
6519 | } | |
6520 | ||
6521 | switch (prop) { | |
6522 | case ZPOOL_PROP_DELEGATION: | |
6523 | spa->spa_delegation = intval; | |
6524 | break; | |
6525 | case ZPOOL_PROP_BOOTFS: | |
6526 | spa->spa_bootfs = intval; | |
6527 | break; | |
6528 | case ZPOOL_PROP_FAILUREMODE: | |
6529 | spa->spa_failmode = intval; | |
6530 | break; | |
9babb374 BB |
6531 | case ZPOOL_PROP_AUTOEXPAND: |
6532 | spa->spa_autoexpand = intval; | |
428870ff BB |
6533 | if (tx->tx_txg != TXG_INITIAL) |
6534 | spa_async_request(spa, | |
6535 | SPA_ASYNC_AUTOEXPAND); | |
6536 | break; | |
379ca9cf OF |
6537 | case ZPOOL_PROP_MULTIHOST: |
6538 | spa->spa_multihost = intval; | |
6539 | break; | |
428870ff BB |
6540 | case ZPOOL_PROP_DEDUPDITTO: |
6541 | spa->spa_dedup_ditto = intval; | |
9babb374 | 6542 | break; |
34dc7c2f BB |
6543 | default: |
6544 | break; | |
6545 | } | |
6546 | } | |
6547 | ||
34dc7c2f | 6548 | } |
b128c09f BB |
6549 | |
6550 | mutex_exit(&spa->spa_props_lock); | |
34dc7c2f BB |
6551 | } |
6552 | ||
428870ff BB |
6553 | /* |
6554 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
6555 | * reflect the new version this txg, so there must be no changes this | |
6556 | * txg to anything that the upgrade code depends on after it executes. | |
6557 | * Therefore this must be called after dsl_pool_sync() does the sync | |
6558 | * tasks. | |
6559 | */ | |
6560 | static void | |
6561 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
6562 | { | |
6563 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6564 | ||
6565 | ASSERT(spa->spa_sync_pass == 1); | |
6566 | ||
13fe0198 MA |
6567 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
6568 | ||
428870ff BB |
6569 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && |
6570 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
6571 | dsl_pool_create_origin(dp, tx); | |
6572 | ||
6573 | /* Keeping the origin open increases spa_minref */ | |
6574 | spa->spa_minref += 3; | |
6575 | } | |
6576 | ||
6577 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
6578 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
6579 | dsl_pool_upgrade_clones(dp, tx); | |
6580 | } | |
6581 | ||
6582 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
6583 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
6584 | dsl_pool_upgrade_dir_clones(dp, tx); | |
6585 | ||
6586 | /* Keeping the freedir open increases spa_minref */ | |
6587 | spa->spa_minref += 3; | |
6588 | } | |
9ae529ec CS |
6589 | |
6590 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
6591 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6592 | spa_feature_create_zap_objects(spa, tx); | |
6593 | } | |
62bdd5eb DL |
6594 | |
6595 | /* | |
6596 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
6597 | * when possibility to use lz4 compression for metadata was added | |
6598 | * Old pools that have this feature enabled must be upgraded to have | |
6599 | * this feature active | |
6600 | */ | |
6601 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6602 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
6603 | SPA_FEATURE_LZ4_COMPRESS); | |
6604 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
6605 | SPA_FEATURE_LZ4_COMPRESS); | |
6606 | ||
6607 | if (lz4_en && !lz4_ac) | |
6608 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
6609 | } | |
3c67d83a TH |
6610 | |
6611 | /* | |
6612 | * If we haven't written the salt, do so now. Note that the | |
6613 | * feature may not be activated yet, but that's fine since | |
6614 | * the presence of this ZAP entry is backwards compatible. | |
6615 | */ | |
6616 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
6617 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
6618 | VERIFY0(zap_add(spa->spa_meta_objset, | |
6619 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
6620 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
6621 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
6622 | } | |
6623 | ||
13fe0198 | 6624 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
428870ff BB |
6625 | } |
6626 | ||
a1d477c2 MA |
6627 | static void |
6628 | vdev_indirect_state_sync_verify(vdev_t *vd) | |
6629 | { | |
6630 | ASSERTV(vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping); | |
6631 | ASSERTV(vdev_indirect_births_t *vib = vd->vdev_indirect_births); | |
6632 | ||
6633 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
6634 | ASSERT(vim != NULL); | |
6635 | ASSERT(vib != NULL); | |
6636 | } | |
6637 | ||
6638 | if (vdev_obsolete_sm_object(vd) != 0) { | |
6639 | ASSERT(vd->vdev_obsolete_sm != NULL); | |
6640 | ASSERT(vd->vdev_removing || | |
6641 | vd->vdev_ops == &vdev_indirect_ops); | |
6642 | ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); | |
6643 | ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); | |
6644 | ||
6645 | ASSERT3U(vdev_obsolete_sm_object(vd), ==, | |
6646 | space_map_object(vd->vdev_obsolete_sm)); | |
6647 | ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, | |
6648 | space_map_allocated(vd->vdev_obsolete_sm)); | |
6649 | } | |
6650 | ASSERT(vd->vdev_obsolete_segments != NULL); | |
6651 | ||
6652 | /* | |
6653 | * Since frees / remaps to an indirect vdev can only | |
6654 | * happen in syncing context, the obsolete segments | |
6655 | * tree must be empty when we start syncing. | |
6656 | */ | |
6657 | ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); | |
6658 | } | |
6659 | ||
34dc7c2f BB |
6660 | /* |
6661 | * Sync the specified transaction group. New blocks may be dirtied as | |
6662 | * part of the process, so we iterate until it converges. | |
6663 | */ | |
6664 | void | |
6665 | spa_sync(spa_t *spa, uint64_t txg) | |
6666 | { | |
6667 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6668 | objset_t *mos = spa->spa_meta_objset; | |
428870ff | 6669 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; |
34dc7c2f BB |
6670 | vdev_t *rvd = spa->spa_root_vdev; |
6671 | vdev_t *vd; | |
34dc7c2f | 6672 | dmu_tx_t *tx; |
b128c09f | 6673 | int error; |
3dfb57a3 DB |
6674 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * |
6675 | zfs_vdev_queue_depth_pct / 100; | |
34dc7c2f | 6676 | |
572e2857 BB |
6677 | VERIFY(spa_writeable(spa)); |
6678 | ||
a1d477c2 MA |
6679 | /* |
6680 | * Wait for i/os issued in open context that need to complete | |
6681 | * before this txg syncs. | |
6682 | */ | |
6683 | VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK])); | |
6684 | spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0); | |
6685 | ||
34dc7c2f BB |
6686 | /* |
6687 | * Lock out configuration changes. | |
6688 | */ | |
b128c09f | 6689 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); |
34dc7c2f BB |
6690 | |
6691 | spa->spa_syncing_txg = txg; | |
6692 | spa->spa_sync_pass = 0; | |
6693 | ||
3dfb57a3 DB |
6694 | mutex_enter(&spa->spa_alloc_lock); |
6695 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
6696 | mutex_exit(&spa->spa_alloc_lock); | |
6697 | ||
b128c09f BB |
6698 | /* |
6699 | * If there are any pending vdev state changes, convert them | |
6700 | * into config changes that go out with this transaction group. | |
6701 | */ | |
6702 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
fb5f0bc8 BB |
6703 | while (list_head(&spa->spa_state_dirty_list) != NULL) { |
6704 | /* | |
6705 | * We need the write lock here because, for aux vdevs, | |
6706 | * calling vdev_config_dirty() modifies sav_config. | |
6707 | * This is ugly and will become unnecessary when we | |
6708 | * eliminate the aux vdev wart by integrating all vdevs | |
6709 | * into the root vdev tree. | |
6710 | */ | |
6711 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6712 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
6713 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
6714 | vdev_state_clean(vd); | |
6715 | vdev_config_dirty(vd); | |
6716 | } | |
6717 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6718 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
b128c09f BB |
6719 | } |
6720 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6721 | ||
34dc7c2f BB |
6722 | tx = dmu_tx_create_assigned(dp, txg); |
6723 | ||
cc92e9d0 | 6724 | spa->spa_sync_starttime = gethrtime(); |
57ddcda1 CC |
6725 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
6726 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
79c76d5b | 6727 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + |
cc92e9d0 GW |
6728 | NSEC_TO_TICK(spa->spa_deadman_synctime)); |
6729 | ||
34dc7c2f BB |
6730 | /* |
6731 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
6732 | * set spa_deflate if we have no raid-z vdevs. | |
6733 | */ | |
6734 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
6735 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6736 | int i; | |
6737 | ||
6738 | for (i = 0; i < rvd->vdev_children; i++) { | |
6739 | vd = rvd->vdev_child[i]; | |
6740 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
6741 | break; | |
6742 | } | |
6743 | if (i == rvd->vdev_children) { | |
6744 | spa->spa_deflate = TRUE; | |
6745 | VERIFY(0 == zap_add(spa->spa_meta_objset, | |
6746 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6747 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
6748 | } | |
6749 | } | |
6750 | ||
3dfb57a3 DB |
6751 | /* |
6752 | * Set the top-level vdev's max queue depth. Evaluate each | |
6753 | * top-level's async write queue depth in case it changed. | |
6754 | * The max queue depth will not change in the middle of syncing | |
6755 | * out this txg. | |
6756 | */ | |
1c27024e DB |
6757 | uint64_t queue_depth_total = 0; |
6758 | for (int c = 0; c < rvd->vdev_children; c++) { | |
3dfb57a3 DB |
6759 | vdev_t *tvd = rvd->vdev_child[c]; |
6760 | metaslab_group_t *mg = tvd->vdev_mg; | |
6761 | ||
6762 | if (mg == NULL || mg->mg_class != spa_normal_class(spa) || | |
6763 | !metaslab_group_initialized(mg)) | |
6764 | continue; | |
6765 | ||
6766 | /* | |
6767 | * It is safe to do a lock-free check here because only async | |
6768 | * allocations look at mg_max_alloc_queue_depth, and async | |
6769 | * allocations all happen from spa_sync(). | |
6770 | */ | |
6771 | ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); | |
6772 | mg->mg_max_alloc_queue_depth = max_queue_depth; | |
6773 | queue_depth_total += mg->mg_max_alloc_queue_depth; | |
6774 | } | |
1c27024e | 6775 | metaslab_class_t *mc = spa_normal_class(spa); |
3dfb57a3 DB |
6776 | ASSERT0(refcount_count(&mc->mc_alloc_slots)); |
6777 | mc->mc_alloc_max_slots = queue_depth_total; | |
6778 | mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
6779 | ||
6780 | ASSERT3U(mc->mc_alloc_max_slots, <=, | |
6781 | max_queue_depth * rvd->vdev_children); | |
6782 | ||
a1d477c2 MA |
6783 | for (int c = 0; c < rvd->vdev_children; c++) { |
6784 | vdev_t *vd = rvd->vdev_child[c]; | |
6785 | vdev_indirect_state_sync_verify(vd); | |
6786 | ||
6787 | if (vdev_indirect_should_condense(vd)) { | |
6788 | spa_condense_indirect_start_sync(vd, tx); | |
6789 | break; | |
6790 | } | |
6791 | } | |
6792 | ||
34dc7c2f BB |
6793 | /* |
6794 | * Iterate to convergence. | |
6795 | */ | |
6796 | do { | |
428870ff | 6797 | int pass = ++spa->spa_sync_pass; |
34dc7c2f BB |
6798 | |
6799 | spa_sync_config_object(spa, tx); | |
6800 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
6801 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
6802 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
6803 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
6804 | spa_errlog_sync(spa, txg); | |
6805 | dsl_pool_sync(dp, txg); | |
6806 | ||
55d85d5a | 6807 | if (pass < zfs_sync_pass_deferred_free) { |
e8b96c60 | 6808 | spa_sync_frees(spa, free_bpl, tx); |
428870ff | 6809 | } else { |
905edb40 MA |
6810 | /* |
6811 | * We can not defer frees in pass 1, because | |
6812 | * we sync the deferred frees later in pass 1. | |
6813 | */ | |
6814 | ASSERT3U(pass, >, 1); | |
428870ff | 6815 | bplist_iterate(free_bpl, bpobj_enqueue_cb, |
e8b96c60 | 6816 | &spa->spa_deferred_bpobj, tx); |
34dc7c2f BB |
6817 | } |
6818 | ||
428870ff BB |
6819 | ddt_sync(spa, txg); |
6820 | dsl_scan_sync(dp, tx); | |
34dc7c2f | 6821 | |
a1d477c2 MA |
6822 | if (spa->spa_vdev_removal != NULL) |
6823 | svr_sync(spa, tx); | |
6824 | ||
6825 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) | |
6826 | != NULL) | |
428870ff BB |
6827 | vdev_sync(vd, txg); |
6828 | ||
905edb40 | 6829 | if (pass == 1) { |
428870ff | 6830 | spa_sync_upgrades(spa, tx); |
905edb40 MA |
6831 | ASSERT3U(txg, >=, |
6832 | spa->spa_uberblock.ub_rootbp.blk_birth); | |
6833 | /* | |
6834 | * Note: We need to check if the MOS is dirty | |
6835 | * because we could have marked the MOS dirty | |
6836 | * without updating the uberblock (e.g. if we | |
6837 | * have sync tasks but no dirty user data). We | |
6838 | * need to check the uberblock's rootbp because | |
6839 | * it is updated if we have synced out dirty | |
6840 | * data (though in this case the MOS will most | |
6841 | * likely also be dirty due to second order | |
6842 | * effects, we don't want to rely on that here). | |
6843 | */ | |
6844 | if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && | |
6845 | !dmu_objset_is_dirty(mos, txg)) { | |
6846 | /* | |
6847 | * Nothing changed on the first pass, | |
6848 | * therefore this TXG is a no-op. Avoid | |
6849 | * syncing deferred frees, so that we | |
6850 | * can keep this TXG as a no-op. | |
6851 | */ | |
6852 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, | |
6853 | txg)); | |
6854 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
6855 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
6856 | break; | |
6857 | } | |
6858 | spa_sync_deferred_frees(spa, tx); | |
6859 | } | |
34dc7c2f | 6860 | |
428870ff | 6861 | } while (dmu_objset_is_dirty(mos, txg)); |
34dc7c2f | 6862 | |
33cf67cd | 6863 | #ifdef ZFS_DEBUG |
e0ab3ab5 JS |
6864 | if (!list_is_empty(&spa->spa_config_dirty_list)) { |
6865 | /* | |
6866 | * Make sure that the number of ZAPs for all the vdevs matches | |
6867 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
6868 | * called if the config is dirty; otherwise there may be | |
6869 | * outstanding AVZ operations that weren't completed in | |
6870 | * spa_sync_config_object. | |
6871 | */ | |
6872 | uint64_t all_vdev_zap_entry_count; | |
6873 | ASSERT0(zap_count(spa->spa_meta_objset, | |
6874 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
6875 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
6876 | all_vdev_zap_entry_count); | |
6877 | } | |
33cf67cd | 6878 | #endif |
e0ab3ab5 | 6879 | |
a1d477c2 MA |
6880 | if (spa->spa_vdev_removal != NULL) { |
6881 | ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); | |
6882 | } | |
6883 | ||
34dc7c2f BB |
6884 | /* |
6885 | * Rewrite the vdev configuration (which includes the uberblock) | |
6886 | * to commit the transaction group. | |
6887 | * | |
6888 | * If there are no dirty vdevs, we sync the uberblock to a few | |
6889 | * random top-level vdevs that are known to be visible in the | |
b128c09f BB |
6890 | * config cache (see spa_vdev_add() for a complete description). |
6891 | * If there *are* dirty vdevs, sync the uberblock to all vdevs. | |
34dc7c2f | 6892 | */ |
b128c09f BB |
6893 | for (;;) { |
6894 | /* | |
6895 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
6896 | * while we're attempting to write the vdev labels. | |
6897 | */ | |
6898 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6899 | ||
6900 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
6901 | vdev_t *svd[SPA_DVAS_PER_BP]; | |
6902 | int svdcount = 0; | |
6903 | int children = rvd->vdev_children; | |
6904 | int c0 = spa_get_random(children); | |
b128c09f | 6905 | |
1c27024e | 6906 | for (int c = 0; c < children; c++) { |
b128c09f | 6907 | vd = rvd->vdev_child[(c0 + c) % children]; |
a1d477c2 MA |
6908 | if (vd->vdev_ms_array == 0 || vd->vdev_islog || |
6909 | !vdev_is_concrete(vd)) | |
b128c09f BB |
6910 | continue; |
6911 | svd[svdcount++] = vd; | |
6912 | if (svdcount == SPA_DVAS_PER_BP) | |
6913 | break; | |
6914 | } | |
b6fcb792 | 6915 | error = vdev_config_sync(svd, svdcount, txg); |
b128c09f BB |
6916 | } else { |
6917 | error = vdev_config_sync(rvd->vdev_child, | |
b6fcb792 | 6918 | rvd->vdev_children, txg); |
34dc7c2f | 6919 | } |
34dc7c2f | 6920 | |
3bc7e0fb GW |
6921 | if (error == 0) |
6922 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
6923 | ||
b128c09f BB |
6924 | spa_config_exit(spa, SCL_STATE, FTAG); |
6925 | ||
6926 | if (error == 0) | |
6927 | break; | |
cec3a0a1 | 6928 | zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR); |
b128c09f BB |
6929 | zio_resume_wait(spa); |
6930 | } | |
34dc7c2f BB |
6931 | dmu_tx_commit(tx); |
6932 | ||
57ddcda1 | 6933 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); |
cc92e9d0 GW |
6934 | spa->spa_deadman_tqid = 0; |
6935 | ||
34dc7c2f BB |
6936 | /* |
6937 | * Clear the dirty config list. | |
6938 | */ | |
b128c09f | 6939 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) |
34dc7c2f BB |
6940 | vdev_config_clean(vd); |
6941 | ||
6942 | /* | |
6943 | * Now that the new config has synced transactionally, | |
6944 | * let it become visible to the config cache. | |
6945 | */ | |
6946 | if (spa->spa_config_syncing != NULL) { | |
6947 | spa_config_set(spa, spa->spa_config_syncing); | |
6948 | spa->spa_config_txg = txg; | |
6949 | spa->spa_config_syncing = NULL; | |
6950 | } | |
6951 | ||
428870ff | 6952 | dsl_pool_sync_done(dp, txg); |
34dc7c2f | 6953 | |
3dfb57a3 DB |
6954 | mutex_enter(&spa->spa_alloc_lock); |
6955 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
6956 | mutex_exit(&spa->spa_alloc_lock); | |
6957 | ||
34dc7c2f BB |
6958 | /* |
6959 | * Update usable space statistics. | |
6960 | */ | |
c65aa5b2 | 6961 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))) |
34dc7c2f BB |
6962 | vdev_sync_done(vd, txg); |
6963 | ||
428870ff BB |
6964 | spa_update_dspace(spa); |
6965 | ||
34dc7c2f BB |
6966 | /* |
6967 | * It had better be the case that we didn't dirty anything | |
6968 | * since vdev_config_sync(). | |
6969 | */ | |
6970 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
6971 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
6972 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
428870ff BB |
6973 | |
6974 | spa->spa_sync_pass = 0; | |
34dc7c2f | 6975 | |
55922e73 GW |
6976 | /* |
6977 | * Update the last synced uberblock here. We want to do this at | |
6978 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
6979 | * will be guaranteed that all the processing associated with | |
6980 | * that txg has been completed. | |
6981 | */ | |
6982 | spa->spa_ubsync = spa->spa_uberblock; | |
b128c09f | 6983 | spa_config_exit(spa, SCL_CONFIG, FTAG); |
34dc7c2f | 6984 | |
428870ff BB |
6985 | spa_handle_ignored_writes(spa); |
6986 | ||
34dc7c2f BB |
6987 | /* |
6988 | * If any async tasks have been requested, kick them off. | |
6989 | */ | |
6990 | spa_async_dispatch(spa); | |
6991 | } | |
6992 | ||
6993 | /* | |
6994 | * Sync all pools. We don't want to hold the namespace lock across these | |
6995 | * operations, so we take a reference on the spa_t and drop the lock during the | |
6996 | * sync. | |
6997 | */ | |
6998 | void | |
6999 | spa_sync_allpools(void) | |
7000 | { | |
7001 | spa_t *spa = NULL; | |
7002 | mutex_enter(&spa_namespace_lock); | |
7003 | while ((spa = spa_next(spa)) != NULL) { | |
572e2857 BB |
7004 | if (spa_state(spa) != POOL_STATE_ACTIVE || |
7005 | !spa_writeable(spa) || spa_suspended(spa)) | |
34dc7c2f BB |
7006 | continue; |
7007 | spa_open_ref(spa, FTAG); | |
7008 | mutex_exit(&spa_namespace_lock); | |
7009 | txg_wait_synced(spa_get_dsl(spa), 0); | |
7010 | mutex_enter(&spa_namespace_lock); | |
7011 | spa_close(spa, FTAG); | |
7012 | } | |
7013 | mutex_exit(&spa_namespace_lock); | |
7014 | } | |
7015 | ||
7016 | /* | |
7017 | * ========================================================================== | |
7018 | * Miscellaneous routines | |
7019 | * ========================================================================== | |
7020 | */ | |
7021 | ||
7022 | /* | |
7023 | * Remove all pools in the system. | |
7024 | */ | |
7025 | void | |
7026 | spa_evict_all(void) | |
7027 | { | |
7028 | spa_t *spa; | |
7029 | ||
7030 | /* | |
7031 | * Remove all cached state. All pools should be closed now, | |
7032 | * so every spa in the AVL tree should be unreferenced. | |
7033 | */ | |
7034 | mutex_enter(&spa_namespace_lock); | |
7035 | while ((spa = spa_next(NULL)) != NULL) { | |
7036 | /* | |
7037 | * Stop async tasks. The async thread may need to detach | |
7038 | * a device that's been replaced, which requires grabbing | |
7039 | * spa_namespace_lock, so we must drop it here. | |
7040 | */ | |
7041 | spa_open_ref(spa, FTAG); | |
7042 | mutex_exit(&spa_namespace_lock); | |
7043 | spa_async_suspend(spa); | |
7044 | mutex_enter(&spa_namespace_lock); | |
34dc7c2f BB |
7045 | spa_close(spa, FTAG); |
7046 | ||
7047 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
7048 | spa_unload(spa); | |
7049 | spa_deactivate(spa); | |
7050 | } | |
7051 | spa_remove(spa); | |
7052 | } | |
7053 | mutex_exit(&spa_namespace_lock); | |
7054 | } | |
7055 | ||
7056 | vdev_t * | |
9babb374 | 7057 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) |
34dc7c2f | 7058 | { |
b128c09f BB |
7059 | vdev_t *vd; |
7060 | int i; | |
7061 | ||
7062 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
7063 | return (vd); | |
7064 | ||
9babb374 | 7065 | if (aux) { |
b128c09f BB |
7066 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { |
7067 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
9babb374 BB |
7068 | if (vd->vdev_guid == guid) |
7069 | return (vd); | |
7070 | } | |
7071 | ||
7072 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
7073 | vd = spa->spa_spares.sav_vdevs[i]; | |
b128c09f BB |
7074 | if (vd->vdev_guid == guid) |
7075 | return (vd); | |
7076 | } | |
7077 | } | |
7078 | ||
7079 | return (NULL); | |
34dc7c2f BB |
7080 | } |
7081 | ||
7082 | void | |
7083 | spa_upgrade(spa_t *spa, uint64_t version) | |
7084 | { | |
572e2857 BB |
7085 | ASSERT(spa_writeable(spa)); |
7086 | ||
b128c09f | 7087 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); |
34dc7c2f BB |
7088 | |
7089 | /* | |
7090 | * This should only be called for a non-faulted pool, and since a | |
7091 | * future version would result in an unopenable pool, this shouldn't be | |
7092 | * possible. | |
7093 | */ | |
8dca0a9a | 7094 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); |
9b67f605 | 7095 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); |
34dc7c2f BB |
7096 | |
7097 | spa->spa_uberblock.ub_version = version; | |
7098 | vdev_config_dirty(spa->spa_root_vdev); | |
7099 | ||
b128c09f | 7100 | spa_config_exit(spa, SCL_ALL, FTAG); |
34dc7c2f BB |
7101 | |
7102 | txg_wait_synced(spa_get_dsl(spa), 0); | |
7103 | } | |
7104 | ||
7105 | boolean_t | |
7106 | spa_has_spare(spa_t *spa, uint64_t guid) | |
7107 | { | |
7108 | int i; | |
7109 | uint64_t spareguid; | |
7110 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
7111 | ||
7112 | for (i = 0; i < sav->sav_count; i++) | |
7113 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
7114 | return (B_TRUE); | |
7115 | ||
7116 | for (i = 0; i < sav->sav_npending; i++) { | |
7117 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
7118 | &spareguid) == 0 && spareguid == guid) | |
7119 | return (B_TRUE); | |
7120 | } | |
7121 | ||
7122 | return (B_FALSE); | |
7123 | } | |
7124 | ||
b128c09f BB |
7125 | /* |
7126 | * Check if a pool has an active shared spare device. | |
7127 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
7128 | */ | |
7129 | static boolean_t | |
7130 | spa_has_active_shared_spare(spa_t *spa) | |
7131 | { | |
7132 | int i, refcnt; | |
7133 | uint64_t pool; | |
7134 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
7135 | ||
7136 | for (i = 0; i < sav->sav_count; i++) { | |
7137 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
7138 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
7139 | refcnt > 2) | |
7140 | return (B_TRUE); | |
7141 | } | |
7142 | ||
7143 | return (B_FALSE); | |
7144 | } | |
7145 | ||
a1d477c2 | 7146 | sysevent_t * |
12fa0466 DE |
7147 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
7148 | { | |
7149 | sysevent_t *ev = NULL; | |
7150 | #ifdef _KERNEL | |
7151 | nvlist_t *resource; | |
7152 | ||
7153 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
7154 | if (resource) { | |
7155 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
7156 | ev->resource = resource; | |
7157 | } | |
7158 | #endif | |
7159 | return (ev); | |
7160 | } | |
7161 | ||
a1d477c2 | 7162 | void |
12fa0466 DE |
7163 | spa_event_post(sysevent_t *ev) |
7164 | { | |
7165 | #ifdef _KERNEL | |
7166 | if (ev) { | |
7167 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
7168 | kmem_free(ev, sizeof (*ev)); | |
7169 | } | |
7170 | #endif | |
7171 | } | |
7172 | ||
34dc7c2f | 7173 | /* |
fb390aaf HR |
7174 | * Post a zevent corresponding to the given sysevent. The 'name' must be one |
7175 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
34dc7c2f BB |
7176 | * filled in from the spa and (optionally) the vdev. This doesn't do anything |
7177 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
7178 | * or zdb as real changes. | |
7179 | */ | |
7180 | void | |
12fa0466 | 7181 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) |
34dc7c2f | 7182 | { |
12fa0466 | 7183 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); |
34dc7c2f | 7184 | } |
c28b2279 BB |
7185 | |
7186 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
7187 | /* state manipulation functions */ | |
7188 | EXPORT_SYMBOL(spa_open); | |
7189 | EXPORT_SYMBOL(spa_open_rewind); | |
7190 | EXPORT_SYMBOL(spa_get_stats); | |
7191 | EXPORT_SYMBOL(spa_create); | |
c28b2279 BB |
7192 | EXPORT_SYMBOL(spa_import); |
7193 | EXPORT_SYMBOL(spa_tryimport); | |
7194 | EXPORT_SYMBOL(spa_destroy); | |
7195 | EXPORT_SYMBOL(spa_export); | |
7196 | EXPORT_SYMBOL(spa_reset); | |
7197 | EXPORT_SYMBOL(spa_async_request); | |
7198 | EXPORT_SYMBOL(spa_async_suspend); | |
7199 | EXPORT_SYMBOL(spa_async_resume); | |
7200 | EXPORT_SYMBOL(spa_inject_addref); | |
7201 | EXPORT_SYMBOL(spa_inject_delref); | |
7202 | EXPORT_SYMBOL(spa_scan_stat_init); | |
7203 | EXPORT_SYMBOL(spa_scan_get_stats); | |
7204 | ||
7205 | /* device maniion */ | |
7206 | EXPORT_SYMBOL(spa_vdev_add); | |
7207 | EXPORT_SYMBOL(spa_vdev_attach); | |
7208 | EXPORT_SYMBOL(spa_vdev_detach); | |
c28b2279 BB |
7209 | EXPORT_SYMBOL(spa_vdev_setpath); |
7210 | EXPORT_SYMBOL(spa_vdev_setfru); | |
7211 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
7212 | ||
7213 | /* spare statech is global across all pools) */ | |
7214 | EXPORT_SYMBOL(spa_spare_add); | |
7215 | EXPORT_SYMBOL(spa_spare_remove); | |
7216 | EXPORT_SYMBOL(spa_spare_exists); | |
7217 | EXPORT_SYMBOL(spa_spare_activate); | |
7218 | ||
7219 | /* L2ARC statech is global across all pools) */ | |
7220 | EXPORT_SYMBOL(spa_l2cache_add); | |
7221 | EXPORT_SYMBOL(spa_l2cache_remove); | |
7222 | EXPORT_SYMBOL(spa_l2cache_exists); | |
7223 | EXPORT_SYMBOL(spa_l2cache_activate); | |
7224 | EXPORT_SYMBOL(spa_l2cache_drop); | |
7225 | ||
7226 | /* scanning */ | |
7227 | EXPORT_SYMBOL(spa_scan); | |
7228 | EXPORT_SYMBOL(spa_scan_stop); | |
7229 | ||
7230 | /* spa syncing */ | |
7231 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
7232 | EXPORT_SYMBOL(spa_sync_allpools); | |
7233 | ||
7234 | /* properties */ | |
7235 | EXPORT_SYMBOL(spa_prop_set); | |
7236 | EXPORT_SYMBOL(spa_prop_get); | |
7237 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
7238 | ||
7239 | /* asynchronous event notification */ | |
7240 | EXPORT_SYMBOL(spa_event_notify); | |
7241 | #endif | |
dea377c0 MA |
7242 | |
7243 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
7244 | module_param(spa_load_verify_maxinflight, int, 0644); | |
7245 | MODULE_PARM_DESC(spa_load_verify_maxinflight, | |
7246 | "Max concurrent traversal I/Os while verifying pool during import -X"); | |
7247 | ||
7248 | module_param(spa_load_verify_metadata, int, 0644); | |
7249 | MODULE_PARM_DESC(spa_load_verify_metadata, | |
7250 | "Set to traverse metadata on pool import"); | |
7251 | ||
7252 | module_param(spa_load_verify_data, int, 0644); | |
7253 | MODULE_PARM_DESC(spa_load_verify_data, | |
7254 | "Set to traverse data on pool import"); | |
dcb6bed1 | 7255 | |
02730c33 | 7256 | /* CSTYLED */ |
dcb6bed1 D |
7257 | module_param(zio_taskq_batch_pct, uint, 0444); |
7258 | MODULE_PARM_DESC(zio_taskq_batch_pct, | |
7259 | "Percentage of CPUs to run an IO worker thread"); | |
7260 | ||
dea377c0 | 7261 | #endif |