]>
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. |
c3520e7f | 24 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. |
153b2285 | 25 | * Copyright 2017 Nexenta Systems, Inc. |
e550644f BB |
26 | * Copyright (c) 2014 Integros [integros.com] |
27 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
12fa0466 | 28 | * Copyright 2017 Joyent, Inc. |
34dc7c2f BB |
29 | */ |
30 | ||
34dc7c2f BB |
31 | #include <sys/zfs_context.h> |
32 | #include <sys/fm/fs/zfs.h> | |
33 | #include <sys/spa.h> | |
34 | #include <sys/spa_impl.h> | |
a1d477c2 | 35 | #include <sys/bpobj.h> |
34dc7c2f BB |
36 | #include <sys/dmu.h> |
37 | #include <sys/dmu_tx.h> | |
a1d477c2 | 38 | #include <sys/dsl_dir.h> |
34dc7c2f BB |
39 | #include <sys/vdev_impl.h> |
40 | #include <sys/uberblock_impl.h> | |
41 | #include <sys/metaslab.h> | |
42 | #include <sys/metaslab_impl.h> | |
43 | #include <sys/space_map.h> | |
93cf2076 | 44 | #include <sys/space_reftree.h> |
34dc7c2f BB |
45 | #include <sys/zio.h> |
46 | #include <sys/zap.h> | |
47 | #include <sys/fs/zfs.h> | |
b128c09f | 48 | #include <sys/arc.h> |
9babb374 | 49 | #include <sys/zil.h> |
428870ff | 50 | #include <sys/dsl_scan.h> |
a6255b7f | 51 | #include <sys/abd.h> |
6c285672 | 52 | #include <sys/zvol.h> |
6078881a | 53 | #include <sys/zfs_ratelimit.h> |
34dc7c2f | 54 | |
b8bcca18 MA |
55 | /* |
56 | * When a vdev is added, it will be divided into approximately (but no | |
57 | * more than) this number of metaslabs. | |
58 | */ | |
59 | int metaslabs_per_vdev = 200; | |
60 | ||
80d52c39 TH |
61 | /* |
62 | * Rate limit delay events to this many IO delays per second. | |
63 | */ | |
64 | unsigned int zfs_delays_per_second = 20; | |
65 | ||
66 | /* | |
67 | * Rate limit checksum events after this many checksum errors per second. | |
68 | */ | |
69 | unsigned int zfs_checksums_per_second = 20; | |
70 | ||
02638a30 TC |
71 | /* |
72 | * Ignore errors during scrub/resilver. Allows to work around resilver | |
73 | * upon import when there are pool errors. | |
74 | */ | |
75 | int zfs_scan_ignore_errors = 0; | |
76 | ||
34dc7c2f BB |
77 | /* |
78 | * Virtual device management. | |
79 | */ | |
80 | ||
81 | static vdev_ops_t *vdev_ops_table[] = { | |
82 | &vdev_root_ops, | |
83 | &vdev_raidz_ops, | |
84 | &vdev_mirror_ops, | |
85 | &vdev_replacing_ops, | |
86 | &vdev_spare_ops, | |
87 | &vdev_disk_ops, | |
88 | &vdev_file_ops, | |
89 | &vdev_missing_ops, | |
428870ff | 90 | &vdev_hole_ops, |
a1d477c2 | 91 | &vdev_indirect_ops, |
34dc7c2f BB |
92 | NULL |
93 | }; | |
94 | ||
34dc7c2f BB |
95 | /* |
96 | * Given a vdev type, return the appropriate ops vector. | |
97 | */ | |
98 | static vdev_ops_t * | |
99 | vdev_getops(const char *type) | |
100 | { | |
101 | vdev_ops_t *ops, **opspp; | |
102 | ||
103 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
104 | if (strcmp(ops->vdev_op_type, type) == 0) | |
105 | break; | |
106 | ||
107 | return (ops); | |
108 | } | |
109 | ||
110 | /* | |
111 | * Default asize function: return the MAX of psize with the asize of | |
112 | * all children. This is what's used by anything other than RAID-Z. | |
113 | */ | |
114 | uint64_t | |
115 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
116 | { | |
117 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
118 | uint64_t csize; | |
34dc7c2f | 119 | |
1c27024e | 120 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
121 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
122 | asize = MAX(asize, csize); | |
123 | } | |
124 | ||
125 | return (asize); | |
126 | } | |
127 | ||
128 | /* | |
9babb374 BB |
129 | * Get the minimum allocatable size. We define the allocatable size as |
130 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
131 | * replace or attach devices which don't have the same physical size but | |
132 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
133 | */ |
134 | uint64_t | |
9babb374 | 135 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 136 | { |
9babb374 | 137 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 138 | |
9babb374 | 139 | /* |
1bd201e7 | 140 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
141 | * just return our own asize. |
142 | */ | |
143 | if (pvd == NULL) | |
144 | return (vd->vdev_asize); | |
34dc7c2f BB |
145 | |
146 | /* | |
9babb374 BB |
147 | * The top-level vdev just returns the allocatable size rounded |
148 | * to the nearest metaslab. | |
34dc7c2f | 149 | */ |
9babb374 BB |
150 | if (vd == vd->vdev_top) |
151 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 152 | |
9babb374 BB |
153 | /* |
154 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
155 | * so each child must provide at least 1/Nth of its asize. | |
156 | */ | |
157 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
2e215fec SH |
158 | return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / |
159 | pvd->vdev_children); | |
34dc7c2f | 160 | |
9babb374 BB |
161 | return (pvd->vdev_min_asize); |
162 | } | |
163 | ||
164 | void | |
165 | vdev_set_min_asize(vdev_t *vd) | |
166 | { | |
167 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 168 | |
1c27024e | 169 | for (int c = 0; c < vd->vdev_children; c++) |
9babb374 | 170 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
171 | } |
172 | ||
173 | vdev_t * | |
174 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
175 | { | |
176 | vdev_t *rvd = spa->spa_root_vdev; | |
177 | ||
b128c09f | 178 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 179 | |
b128c09f BB |
180 | if (vdev < rvd->vdev_children) { |
181 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 182 | return (rvd->vdev_child[vdev]); |
b128c09f | 183 | } |
34dc7c2f BB |
184 | |
185 | return (NULL); | |
186 | } | |
187 | ||
188 | vdev_t * | |
189 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
190 | { | |
34dc7c2f BB |
191 | vdev_t *mvd; |
192 | ||
193 | if (vd->vdev_guid == guid) | |
194 | return (vd); | |
195 | ||
1c27024e | 196 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
197 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
198 | NULL) | |
199 | return (mvd); | |
200 | ||
201 | return (NULL); | |
202 | } | |
203 | ||
9c43027b AJ |
204 | static int |
205 | vdev_count_leaves_impl(vdev_t *vd) | |
206 | { | |
207 | int n = 0; | |
9c43027b AJ |
208 | |
209 | if (vd->vdev_ops->vdev_op_leaf) | |
210 | return (1); | |
211 | ||
1c27024e | 212 | for (int c = 0; c < vd->vdev_children; c++) |
9c43027b AJ |
213 | n += vdev_count_leaves_impl(vd->vdev_child[c]); |
214 | ||
215 | return (n); | |
216 | } | |
217 | ||
218 | int | |
219 | vdev_count_leaves(spa_t *spa) | |
220 | { | |
743253df OF |
221 | int rc; |
222 | ||
223 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
224 | rc = vdev_count_leaves_impl(spa->spa_root_vdev); | |
225 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
226 | ||
227 | return (rc); | |
9c43027b AJ |
228 | } |
229 | ||
34dc7c2f BB |
230 | void |
231 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
232 | { | |
233 | size_t oldsize, newsize; | |
234 | uint64_t id = cvd->vdev_id; | |
235 | vdev_t **newchild; | |
236 | ||
44de2f02 | 237 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
238 | ASSERT(cvd->vdev_parent == NULL); |
239 | ||
240 | cvd->vdev_parent = pvd; | |
241 | ||
242 | if (pvd == NULL) | |
243 | return; | |
244 | ||
245 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
246 | ||
247 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
248 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
249 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
250 | ||
79c76d5b | 251 | newchild = kmem_alloc(newsize, KM_SLEEP); |
34dc7c2f BB |
252 | if (pvd->vdev_child != NULL) { |
253 | bcopy(pvd->vdev_child, newchild, oldsize); | |
254 | kmem_free(pvd->vdev_child, oldsize); | |
255 | } | |
256 | ||
257 | pvd->vdev_child = newchild; | |
258 | pvd->vdev_child[id] = cvd; | |
259 | ||
260 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
261 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
262 | ||
263 | /* | |
264 | * Walk up all ancestors to update guid sum. | |
265 | */ | |
266 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
267 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
268 | } |
269 | ||
270 | void | |
271 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
272 | { | |
273 | int c; | |
274 | uint_t id = cvd->vdev_id; | |
275 | ||
276 | ASSERT(cvd->vdev_parent == pvd); | |
277 | ||
278 | if (pvd == NULL) | |
279 | return; | |
280 | ||
281 | ASSERT(id < pvd->vdev_children); | |
282 | ASSERT(pvd->vdev_child[id] == cvd); | |
283 | ||
284 | pvd->vdev_child[id] = NULL; | |
285 | cvd->vdev_parent = NULL; | |
286 | ||
287 | for (c = 0; c < pvd->vdev_children; c++) | |
288 | if (pvd->vdev_child[c]) | |
289 | break; | |
290 | ||
291 | if (c == pvd->vdev_children) { | |
292 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
293 | pvd->vdev_child = NULL; | |
294 | pvd->vdev_children = 0; | |
295 | } | |
296 | ||
297 | /* | |
298 | * Walk up all ancestors to update guid sum. | |
299 | */ | |
300 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
301 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
302 | } |
303 | ||
304 | /* | |
305 | * Remove any holes in the child array. | |
306 | */ | |
307 | void | |
308 | vdev_compact_children(vdev_t *pvd) | |
309 | { | |
310 | vdev_t **newchild, *cvd; | |
311 | int oldc = pvd->vdev_children; | |
9babb374 | 312 | int newc; |
34dc7c2f | 313 | |
b128c09f | 314 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 315 | |
a1d477c2 MA |
316 | if (oldc == 0) |
317 | return; | |
318 | ||
1c27024e | 319 | for (int c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
320 | if (pvd->vdev_child[c]) |
321 | newc++; | |
322 | ||
a1d477c2 MA |
323 | if (newc > 0) { |
324 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); | |
34dc7c2f | 325 | |
a1d477c2 MA |
326 | for (int c = newc = 0; c < oldc; c++) { |
327 | if ((cvd = pvd->vdev_child[c]) != NULL) { | |
328 | newchild[newc] = cvd; | |
329 | cvd->vdev_id = newc++; | |
330 | } | |
34dc7c2f | 331 | } |
a1d477c2 MA |
332 | } else { |
333 | newchild = NULL; | |
34dc7c2f BB |
334 | } |
335 | ||
336 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
337 | pvd->vdev_child = newchild; | |
338 | pvd->vdev_children = newc; | |
339 | } | |
340 | ||
341 | /* | |
342 | * Allocate and minimally initialize a vdev_t. | |
343 | */ | |
428870ff | 344 | vdev_t * |
34dc7c2f BB |
345 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
346 | { | |
347 | vdev_t *vd; | |
a1d477c2 | 348 | vdev_indirect_config_t *vic; |
34dc7c2f | 349 | |
79c76d5b | 350 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); |
a1d477c2 | 351 | vic = &vd->vdev_indirect_config; |
34dc7c2f BB |
352 | |
353 | if (spa->spa_root_vdev == NULL) { | |
354 | ASSERT(ops == &vdev_root_ops); | |
355 | spa->spa_root_vdev = vd; | |
3541dc6d | 356 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
357 | } |
358 | ||
428870ff | 359 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
360 | if (spa->spa_root_vdev == vd) { |
361 | /* | |
362 | * The root vdev's guid will also be the pool guid, | |
363 | * which must be unique among all pools. | |
364 | */ | |
428870ff | 365 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
366 | } else { |
367 | /* | |
368 | * Any other vdev's guid must be unique within the pool. | |
369 | */ | |
428870ff | 370 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
371 | } |
372 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
373 | } | |
374 | ||
375 | vd->vdev_spa = spa; | |
376 | vd->vdev_id = id; | |
377 | vd->vdev_guid = guid; | |
378 | vd->vdev_guid_sum = guid; | |
379 | vd->vdev_ops = ops; | |
380 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 381 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
a1d477c2 MA |
382 | vic->vic_prev_indirect_vdev = UINT64_MAX; |
383 | ||
384 | rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); | |
385 | mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); | |
386 | vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); | |
34dc7c2f | 387 | |
6078881a TH |
388 | /* |
389 | * Initialize rate limit structs for events. We rate limit ZIO delay | |
390 | * and checksum events so that we don't overwhelm ZED with thousands | |
391 | * of events when a disk is acting up. | |
392 | */ | |
80d52c39 TH |
393 | zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_delays_per_second, 1); |
394 | zfs_ratelimit_init(&vd->vdev_checksum_rl, &zfs_checksums_per_second, 1); | |
6078881a | 395 | |
98f72a53 BB |
396 | list_link_init(&vd->vdev_config_dirty_node); |
397 | list_link_init(&vd->vdev_state_dirty_node); | |
448d7aaa | 398 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); |
34dc7c2f | 399 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); |
b128c09f | 400 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
3dfb57a3 | 401 | mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); |
d4a72f23 | 402 | mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); |
6078881a | 403 | |
1c27024e | 404 | for (int t = 0; t < DTL_TYPES; t++) { |
a1d477c2 | 405 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL); |
fb5f0bc8 | 406 | } |
4747a7d3 | 407 | txg_list_create(&vd->vdev_ms_list, spa, |
34dc7c2f | 408 | offsetof(struct metaslab, ms_txg_node)); |
4747a7d3 | 409 | txg_list_create(&vd->vdev_dtl_list, spa, |
34dc7c2f BB |
410 | offsetof(struct vdev, vdev_dtl_node)); |
411 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
412 | vdev_queue_init(vd); | |
413 | vdev_cache_init(vd); | |
414 | ||
415 | return (vd); | |
416 | } | |
417 | ||
418 | /* | |
419 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
420 | * creating a new vdev or loading an existing one - the behavior is slightly | |
421 | * different for each case. | |
422 | */ | |
423 | int | |
424 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
425 | int alloctype) | |
426 | { | |
427 | vdev_ops_t *ops; | |
428 | char *type; | |
429 | uint64_t guid = 0, islog, nparity; | |
430 | vdev_t *vd; | |
a1d477c2 | 431 | vdev_indirect_config_t *vic; |
4a283c7f TH |
432 | char *tmp = NULL; |
433 | int rc; | |
34dc7c2f | 434 | |
b128c09f | 435 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
436 | |
437 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
2e528b49 | 438 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
439 | |
440 | if ((ops = vdev_getops(type)) == NULL) | |
2e528b49 | 441 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
442 | |
443 | /* | |
444 | * If this is a load, get the vdev guid from the nvlist. | |
445 | * Otherwise, vdev_alloc_common() will generate one for us. | |
446 | */ | |
447 | if (alloctype == VDEV_ALLOC_LOAD) { | |
448 | uint64_t label_id; | |
449 | ||
450 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
451 | label_id != id) | |
2e528b49 | 452 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
453 | |
454 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 455 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
456 | } else if (alloctype == VDEV_ALLOC_SPARE) { |
457 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 458 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
459 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { |
460 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 461 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
462 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
463 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 464 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
465 | } |
466 | ||
467 | /* | |
468 | * The first allocated vdev must be of type 'root'. | |
469 | */ | |
470 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
2e528b49 | 471 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
472 | |
473 | /* | |
474 | * Determine whether we're a log vdev. | |
475 | */ | |
476 | islog = 0; | |
477 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
478 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
2e528b49 | 479 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 480 | |
428870ff | 481 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
2e528b49 | 482 | return (SET_ERROR(ENOTSUP)); |
428870ff | 483 | |
34dc7c2f BB |
484 | /* |
485 | * Set the nparity property for RAID-Z vdevs. | |
486 | */ | |
487 | nparity = -1ULL; | |
488 | if (ops == &vdev_raidz_ops) { | |
489 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
490 | &nparity) == 0) { | |
428870ff | 491 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
2e528b49 | 492 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 493 | /* |
45d1cae3 BB |
494 | * Previous versions could only support 1 or 2 parity |
495 | * device. | |
34dc7c2f | 496 | */ |
45d1cae3 BB |
497 | if (nparity > 1 && |
498 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
2e528b49 | 499 | return (SET_ERROR(ENOTSUP)); |
45d1cae3 BB |
500 | if (nparity > 2 && |
501 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
2e528b49 | 502 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
503 | } else { |
504 | /* | |
505 | * We require the parity to be specified for SPAs that | |
506 | * support multiple parity levels. | |
507 | */ | |
45d1cae3 | 508 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
2e528b49 | 509 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
510 | /* |
511 | * Otherwise, we default to 1 parity device for RAID-Z. | |
512 | */ | |
513 | nparity = 1; | |
514 | } | |
515 | } else { | |
516 | nparity = 0; | |
517 | } | |
518 | ASSERT(nparity != -1ULL); | |
519 | ||
520 | vd = vdev_alloc_common(spa, id, guid, ops); | |
a1d477c2 | 521 | vic = &vd->vdev_indirect_config; |
34dc7c2f BB |
522 | |
523 | vd->vdev_islog = islog; | |
524 | vd->vdev_nparity = nparity; | |
525 | ||
526 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
527 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
4a283c7f TH |
528 | |
529 | /* | |
530 | * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a | |
531 | * fault on a vdev and want it to persist across imports (like with | |
532 | * zpool offline -f). | |
533 | */ | |
534 | rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp); | |
535 | if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) { | |
536 | vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; | |
537 | vd->vdev_faulted = 1; | |
538 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
539 | } | |
540 | ||
34dc7c2f BB |
541 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) |
542 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
543 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
544 | &vd->vdev_physpath) == 0) | |
545 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
1bbd8770 TH |
546 | |
547 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, | |
548 | &vd->vdev_enc_sysfs_path) == 0) | |
549 | vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); | |
550 | ||
9babb374 BB |
551 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
552 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
553 | |
554 | /* | |
555 | * Set the whole_disk property. If it's not specified, leave the value | |
556 | * as -1. | |
557 | */ | |
558 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
559 | &vd->vdev_wholedisk) != 0) | |
560 | vd->vdev_wholedisk = -1ULL; | |
561 | ||
a1d477c2 MA |
562 | ASSERT0(vic->vic_mapping_object); |
563 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, | |
564 | &vic->vic_mapping_object); | |
565 | ASSERT0(vic->vic_births_object); | |
566 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, | |
567 | &vic->vic_births_object); | |
568 | ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); | |
569 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, | |
570 | &vic->vic_prev_indirect_vdev); | |
571 | ||
34dc7c2f BB |
572 | /* |
573 | * Look for the 'not present' flag. This will only be set if the device | |
574 | * was not present at the time of import. | |
575 | */ | |
9babb374 BB |
576 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
577 | &vd->vdev_not_present); | |
34dc7c2f BB |
578 | |
579 | /* | |
580 | * Get the alignment requirement. | |
581 | */ | |
582 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
583 | ||
428870ff BB |
584 | /* |
585 | * Retrieve the vdev creation time. | |
586 | */ | |
587 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
588 | &vd->vdev_crtxg); | |
589 | ||
34dc7c2f BB |
590 | /* |
591 | * If we're a top-level vdev, try to load the allocation parameters. | |
592 | */ | |
428870ff BB |
593 | if (parent && !parent->vdev_parent && |
594 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
595 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
596 | &vd->vdev_ms_array); | |
597 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
598 | &vd->vdev_ms_shift); | |
599 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
600 | &vd->vdev_asize); | |
428870ff BB |
601 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
602 | &vd->vdev_removing); | |
e0ab3ab5 JS |
603 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, |
604 | &vd->vdev_top_zap); | |
605 | } else { | |
606 | ASSERT0(vd->vdev_top_zap); | |
428870ff BB |
607 | } |
608 | ||
5ffb9d1d | 609 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
610 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
611 | alloctype == VDEV_ALLOC_ADD || | |
612 | alloctype == VDEV_ALLOC_SPLIT || | |
613 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
614 | vd->vdev_mg = metaslab_group_create(islog ? | |
615 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
616 | } |
617 | ||
e0ab3ab5 JS |
618 | if (vd->vdev_ops->vdev_op_leaf && |
619 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
620 | (void) nvlist_lookup_uint64(nv, | |
621 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); | |
622 | } else { | |
623 | ASSERT0(vd->vdev_leaf_zap); | |
624 | } | |
625 | ||
34dc7c2f BB |
626 | /* |
627 | * If we're a leaf vdev, try to load the DTL object and other state. | |
628 | */ | |
e0ab3ab5 | 629 | |
b128c09f | 630 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
631 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
632 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
633 | if (alloctype == VDEV_ALLOC_LOAD) { |
634 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
93cf2076 | 635 | &vd->vdev_dtl_object); |
b128c09f BB |
636 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
637 | &vd->vdev_unspare); | |
638 | } | |
9babb374 BB |
639 | |
640 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
641 | uint64_t spare = 0; | |
642 | ||
643 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
644 | &spare) == 0 && spare) | |
645 | spa_spare_add(vd); | |
646 | } | |
647 | ||
34dc7c2f BB |
648 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
649 | &vd->vdev_offline); | |
b128c09f | 650 | |
5d1f7fb6 GW |
651 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, |
652 | &vd->vdev_resilver_txg); | |
572e2857 | 653 | |
34dc7c2f | 654 | /* |
4a283c7f TH |
655 | * In general, when importing a pool we want to ignore the |
656 | * persistent fault state, as the diagnosis made on another | |
657 | * system may not be valid in the current context. The only | |
658 | * exception is if we forced a vdev to a persistently faulted | |
659 | * state with 'zpool offline -f'. The persistent fault will | |
660 | * remain across imports until cleared. | |
661 | * | |
662 | * Local vdevs will remain in the faulted state. | |
34dc7c2f | 663 | */ |
4a283c7f TH |
664 | if (spa_load_state(spa) == SPA_LOAD_OPEN || |
665 | spa_load_state(spa) == SPA_LOAD_IMPORT) { | |
34dc7c2f BB |
666 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
667 | &vd->vdev_faulted); | |
668 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
669 | &vd->vdev_degraded); | |
670 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
671 | &vd->vdev_removed); | |
428870ff BB |
672 | |
673 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
674 | char *aux; | |
675 | ||
676 | vd->vdev_label_aux = | |
677 | VDEV_AUX_ERR_EXCEEDED; | |
678 | if (nvlist_lookup_string(nv, | |
679 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
680 | strcmp(aux, "external") == 0) | |
681 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
682 | } | |
34dc7c2f BB |
683 | } |
684 | } | |
685 | ||
686 | /* | |
687 | * Add ourselves to the parent's list of children. | |
688 | */ | |
689 | vdev_add_child(parent, vd); | |
690 | ||
691 | *vdp = vd; | |
692 | ||
693 | return (0); | |
694 | } | |
695 | ||
696 | void | |
697 | vdev_free(vdev_t *vd) | |
698 | { | |
34dc7c2f BB |
699 | spa_t *spa = vd->vdev_spa; |
700 | ||
d4a72f23 TC |
701 | /* |
702 | * Scan queues are normally destroyed at the end of a scan. If the | |
703 | * queue exists here, that implies the vdev is being removed while | |
704 | * the scan is still running. | |
705 | */ | |
706 | if (vd->vdev_scan_io_queue != NULL) { | |
707 | mutex_enter(&vd->vdev_scan_io_queue_lock); | |
708 | dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); | |
709 | vd->vdev_scan_io_queue = NULL; | |
710 | mutex_exit(&vd->vdev_scan_io_queue_lock); | |
711 | } | |
712 | ||
34dc7c2f BB |
713 | /* |
714 | * vdev_free() implies closing the vdev first. This is simpler than | |
715 | * trying to ensure complicated semantics for all callers. | |
716 | */ | |
717 | vdev_close(vd); | |
718 | ||
b128c09f | 719 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 720 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
721 | |
722 | /* | |
723 | * Free all children. | |
724 | */ | |
1c27024e | 725 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
726 | vdev_free(vd->vdev_child[c]); |
727 | ||
728 | ASSERT(vd->vdev_child == NULL); | |
729 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
730 | ||
731 | /* | |
732 | * Discard allocation state. | |
733 | */ | |
428870ff | 734 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 735 | vdev_metaslab_fini(vd); |
428870ff BB |
736 | metaslab_group_destroy(vd->vdev_mg); |
737 | } | |
34dc7c2f | 738 | |
c99c9001 MS |
739 | ASSERT0(vd->vdev_stat.vs_space); |
740 | ASSERT0(vd->vdev_stat.vs_dspace); | |
741 | ASSERT0(vd->vdev_stat.vs_alloc); | |
34dc7c2f BB |
742 | |
743 | /* | |
744 | * Remove this vdev from its parent's child list. | |
745 | */ | |
746 | vdev_remove_child(vd->vdev_parent, vd); | |
747 | ||
748 | ASSERT(vd->vdev_parent == NULL); | |
749 | ||
750 | /* | |
751 | * Clean up vdev structure. | |
752 | */ | |
753 | vdev_queue_fini(vd); | |
754 | vdev_cache_fini(vd); | |
755 | ||
756 | if (vd->vdev_path) | |
757 | spa_strfree(vd->vdev_path); | |
758 | if (vd->vdev_devid) | |
759 | spa_strfree(vd->vdev_devid); | |
760 | if (vd->vdev_physpath) | |
761 | spa_strfree(vd->vdev_physpath); | |
1bbd8770 TH |
762 | |
763 | if (vd->vdev_enc_sysfs_path) | |
764 | spa_strfree(vd->vdev_enc_sysfs_path); | |
765 | ||
9babb374 BB |
766 | if (vd->vdev_fru) |
767 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
768 | |
769 | if (vd->vdev_isspare) | |
770 | spa_spare_remove(vd); | |
771 | if (vd->vdev_isl2cache) | |
772 | spa_l2cache_remove(vd); | |
773 | ||
774 | txg_list_destroy(&vd->vdev_ms_list); | |
775 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 776 | |
34dc7c2f | 777 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 | 778 | space_map_close(vd->vdev_dtl_sm); |
1c27024e | 779 | for (int t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
780 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); |
781 | range_tree_destroy(vd->vdev_dtl[t]); | |
fb5f0bc8 | 782 | } |
34dc7c2f | 783 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 784 | |
a1d477c2 MA |
785 | EQUIV(vd->vdev_indirect_births != NULL, |
786 | vd->vdev_indirect_mapping != NULL); | |
787 | if (vd->vdev_indirect_births != NULL) { | |
788 | vdev_indirect_mapping_close(vd->vdev_indirect_mapping); | |
789 | vdev_indirect_births_close(vd->vdev_indirect_births); | |
790 | } | |
791 | ||
792 | if (vd->vdev_obsolete_sm != NULL) { | |
793 | ASSERT(vd->vdev_removing || | |
794 | vd->vdev_ops == &vdev_indirect_ops); | |
795 | space_map_close(vd->vdev_obsolete_sm); | |
796 | vd->vdev_obsolete_sm = NULL; | |
797 | } | |
798 | range_tree_destroy(vd->vdev_obsolete_segments); | |
799 | rw_destroy(&vd->vdev_indirect_rwlock); | |
800 | mutex_destroy(&vd->vdev_obsolete_lock); | |
801 | ||
3dfb57a3 | 802 | mutex_destroy(&vd->vdev_queue_lock); |
34dc7c2f BB |
803 | mutex_destroy(&vd->vdev_dtl_lock); |
804 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 805 | mutex_destroy(&vd->vdev_probe_lock); |
d4a72f23 | 806 | mutex_destroy(&vd->vdev_scan_io_queue_lock); |
34dc7c2f | 807 | |
c17486b2 GN |
808 | zfs_ratelimit_fini(&vd->vdev_delay_rl); |
809 | zfs_ratelimit_fini(&vd->vdev_checksum_rl); | |
810 | ||
34dc7c2f BB |
811 | if (vd == spa->spa_root_vdev) |
812 | spa->spa_root_vdev = NULL; | |
813 | ||
814 | kmem_free(vd, sizeof (vdev_t)); | |
815 | } | |
816 | ||
817 | /* | |
818 | * Transfer top-level vdev state from svd to tvd. | |
819 | */ | |
820 | static void | |
821 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
822 | { | |
823 | spa_t *spa = svd->vdev_spa; | |
824 | metaslab_t *msp; | |
825 | vdev_t *vd; | |
826 | int t; | |
827 | ||
828 | ASSERT(tvd == tvd->vdev_top); | |
829 | ||
77943bc1 | 830 | tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; |
34dc7c2f BB |
831 | tvd->vdev_ms_array = svd->vdev_ms_array; |
832 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
833 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
e0ab3ab5 | 834 | tvd->vdev_top_zap = svd->vdev_top_zap; |
34dc7c2f BB |
835 | |
836 | svd->vdev_ms_array = 0; | |
837 | svd->vdev_ms_shift = 0; | |
838 | svd->vdev_ms_count = 0; | |
e0ab3ab5 | 839 | svd->vdev_top_zap = 0; |
34dc7c2f | 840 | |
5ffb9d1d GW |
841 | if (tvd->vdev_mg) |
842 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
843 | tvd->vdev_mg = svd->vdev_mg; |
844 | tvd->vdev_ms = svd->vdev_ms; | |
845 | ||
846 | svd->vdev_mg = NULL; | |
847 | svd->vdev_ms = NULL; | |
848 | ||
849 | if (tvd->vdev_mg != NULL) | |
850 | tvd->vdev_mg->mg_vd = tvd; | |
851 | ||
852 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
853 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
854 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
855 | ||
856 | svd->vdev_stat.vs_alloc = 0; | |
857 | svd->vdev_stat.vs_space = 0; | |
858 | svd->vdev_stat.vs_dspace = 0; | |
859 | ||
9e052db4 MA |
860 | /* |
861 | * State which may be set on a top-level vdev that's in the | |
862 | * process of being removed. | |
863 | */ | |
864 | ASSERT0(tvd->vdev_indirect_config.vic_births_object); | |
865 | ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); | |
866 | ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); | |
867 | ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); | |
868 | ASSERT3P(tvd->vdev_indirect_births, ==, NULL); | |
869 | ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); | |
870 | ASSERT0(tvd->vdev_removing); | |
871 | tvd->vdev_removing = svd->vdev_removing; | |
872 | tvd->vdev_indirect_config = svd->vdev_indirect_config; | |
873 | tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; | |
874 | tvd->vdev_indirect_births = svd->vdev_indirect_births; | |
875 | range_tree_swap(&svd->vdev_obsolete_segments, | |
876 | &tvd->vdev_obsolete_segments); | |
877 | tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; | |
878 | svd->vdev_indirect_config.vic_mapping_object = 0; | |
879 | svd->vdev_indirect_config.vic_births_object = 0; | |
880 | svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; | |
881 | svd->vdev_indirect_mapping = NULL; | |
882 | svd->vdev_indirect_births = NULL; | |
883 | svd->vdev_obsolete_sm = NULL; | |
884 | svd->vdev_removing = 0; | |
885 | ||
34dc7c2f BB |
886 | for (t = 0; t < TXG_SIZE; t++) { |
887 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
888 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
889 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
890 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
891 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
892 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
893 | } | |
894 | ||
b128c09f | 895 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
896 | vdev_config_clean(svd); |
897 | vdev_config_dirty(tvd); | |
898 | } | |
899 | ||
b128c09f BB |
900 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
901 | vdev_state_clean(svd); | |
902 | vdev_state_dirty(tvd); | |
903 | } | |
904 | ||
34dc7c2f BB |
905 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
906 | svd->vdev_deflate_ratio = 0; | |
907 | ||
908 | tvd->vdev_islog = svd->vdev_islog; | |
909 | svd->vdev_islog = 0; | |
d4a72f23 TC |
910 | |
911 | dsl_scan_io_queue_vdev_xfer(svd, tvd); | |
34dc7c2f BB |
912 | } |
913 | ||
914 | static void | |
915 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
916 | { | |
34dc7c2f BB |
917 | if (vd == NULL) |
918 | return; | |
919 | ||
920 | vd->vdev_top = tvd; | |
921 | ||
1c27024e | 922 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
923 | vdev_top_update(tvd, vd->vdev_child[c]); |
924 | } | |
925 | ||
926 | /* | |
927 | * Add a mirror/replacing vdev above an existing vdev. | |
928 | */ | |
929 | vdev_t * | |
930 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
931 | { | |
932 | spa_t *spa = cvd->vdev_spa; | |
933 | vdev_t *pvd = cvd->vdev_parent; | |
934 | vdev_t *mvd; | |
935 | ||
b128c09f | 936 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
937 | |
938 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
939 | ||
940 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 941 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 942 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
a1d477c2 | 943 | mvd->vdev_psize = cvd->vdev_psize; |
34dc7c2f BB |
944 | mvd->vdev_ashift = cvd->vdev_ashift; |
945 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 946 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
947 | |
948 | vdev_remove_child(pvd, cvd); | |
949 | vdev_add_child(pvd, mvd); | |
950 | cvd->vdev_id = mvd->vdev_children; | |
951 | vdev_add_child(mvd, cvd); | |
952 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
953 | ||
954 | if (mvd == mvd->vdev_top) | |
955 | vdev_top_transfer(cvd, mvd); | |
956 | ||
957 | return (mvd); | |
958 | } | |
959 | ||
960 | /* | |
961 | * Remove a 1-way mirror/replacing vdev from the tree. | |
962 | */ | |
963 | void | |
964 | vdev_remove_parent(vdev_t *cvd) | |
965 | { | |
966 | vdev_t *mvd = cvd->vdev_parent; | |
967 | vdev_t *pvd = mvd->vdev_parent; | |
968 | ||
b128c09f | 969 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
970 | |
971 | ASSERT(mvd->vdev_children == 1); | |
972 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
973 | mvd->vdev_ops == &vdev_replacing_ops || | |
974 | mvd->vdev_ops == &vdev_spare_ops); | |
975 | cvd->vdev_ashift = mvd->vdev_ashift; | |
976 | ||
977 | vdev_remove_child(mvd, cvd); | |
978 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 979 | |
34dc7c2f | 980 | /* |
b128c09f BB |
981 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
982 | * Otherwise, we could have detached an offline device, and when we | |
983 | * go to import the pool we'll think we have two top-level vdevs, | |
984 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 985 | */ |
fb5f0bc8 BB |
986 | if (mvd->vdev_top == mvd) { |
987 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 988 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
989 | cvd->vdev_guid += guid_delta; |
990 | cvd->vdev_guid_sum += guid_delta; | |
61e99a73 AB |
991 | |
992 | /* | |
993 | * If pool not set for autoexpand, we need to also preserve | |
994 | * mvd's asize to prevent automatic expansion of cvd. | |
995 | * Otherwise if we are adjusting the mirror by attaching and | |
996 | * detaching children of non-uniform sizes, the mirror could | |
997 | * autoexpand, unexpectedly requiring larger devices to | |
998 | * re-establish the mirror. | |
999 | */ | |
1000 | if (!cvd->vdev_spa->spa_autoexpand) | |
1001 | cvd->vdev_asize = mvd->vdev_asize; | |
fb5f0bc8 | 1002 | } |
b128c09f BB |
1003 | cvd->vdev_id = mvd->vdev_id; |
1004 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
1005 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
1006 | ||
1007 | if (cvd == cvd->vdev_top) | |
1008 | vdev_top_transfer(mvd, cvd); | |
1009 | ||
1010 | ASSERT(mvd->vdev_children == 0); | |
1011 | vdev_free(mvd); | |
1012 | } | |
1013 | ||
1014 | int | |
1015 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
1016 | { | |
1017 | spa_t *spa = vd->vdev_spa; | |
1018 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
1019 | uint64_t m; |
1020 | uint64_t oldc = vd->vdev_ms_count; | |
1021 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
1022 | metaslab_t **mspp; | |
1023 | int error; | |
1024 | ||
428870ff BB |
1025 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
1026 | ||
1027 | /* | |
1028 | * This vdev is not being allocated from yet or is a hole. | |
1029 | */ | |
1030 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
1031 | return (0); |
1032 | ||
428870ff BB |
1033 | ASSERT(!vd->vdev_ishole); |
1034 | ||
34dc7c2f BB |
1035 | ASSERT(oldc <= newc); |
1036 | ||
bffb68a2 | 1037 | mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
34dc7c2f BB |
1038 | |
1039 | if (oldc != 0) { | |
1040 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
bffb68a2 | 1041 | vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); |
34dc7c2f BB |
1042 | } |
1043 | ||
1044 | vd->vdev_ms = mspp; | |
1045 | vd->vdev_ms_count = newc; | |
1046 | ||
1047 | for (m = oldc; m < newc; m++) { | |
93cf2076 GW |
1048 | uint64_t object = 0; |
1049 | ||
a1d477c2 MA |
1050 | /* |
1051 | * vdev_ms_array may be 0 if we are creating the "fake" | |
1052 | * metaslabs for an indirect vdev for zdb's leak detection. | |
1053 | * See zdb_leak_init(). | |
1054 | */ | |
1055 | if (txg == 0 && vd->vdev_ms_array != 0) { | |
34dc7c2f | 1056 | error = dmu_read(mos, vd->vdev_ms_array, |
9babb374 BB |
1057 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
1058 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
1059 | if (error) |
1060 | return (error); | |
34dc7c2f | 1061 | } |
fb42a493 PS |
1062 | |
1063 | error = metaslab_init(vd->vdev_mg, m, object, txg, | |
1064 | &(vd->vdev_ms[m])); | |
1065 | if (error) | |
1066 | return (error); | |
34dc7c2f BB |
1067 | } |
1068 | ||
428870ff BB |
1069 | if (txg == 0) |
1070 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
1071 | ||
1072 | /* | |
1073 | * If the vdev is being removed we don't activate | |
1074 | * the metaslabs since we want to ensure that no new | |
1075 | * allocations are performed on this device. | |
1076 | */ | |
1077 | if (oldc == 0 && !vd->vdev_removing) | |
1078 | metaslab_group_activate(vd->vdev_mg); | |
1079 | ||
1080 | if (txg == 0) | |
1081 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
1082 | ||
34dc7c2f BB |
1083 | return (0); |
1084 | } | |
1085 | ||
1086 | void | |
1087 | vdev_metaslab_fini(vdev_t *vd) | |
1088 | { | |
34dc7c2f | 1089 | if (vd->vdev_ms != NULL) { |
a1d477c2 MA |
1090 | uint64_t count = vd->vdev_ms_count; |
1091 | ||
428870ff | 1092 | metaslab_group_passivate(vd->vdev_mg); |
a1d477c2 | 1093 | for (uint64_t m = 0; m < count; m++) { |
93cf2076 GW |
1094 | metaslab_t *msp = vd->vdev_ms[m]; |
1095 | ||
1096 | if (msp != NULL) | |
1097 | metaslab_fini(msp); | |
1098 | } | |
bffb68a2 | 1099 | vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); |
34dc7c2f | 1100 | vd->vdev_ms = NULL; |
920dd524 | 1101 | |
a1d477c2 MA |
1102 | vd->vdev_ms_count = 0; |
1103 | } | |
1104 | ASSERT0(vd->vdev_ms_count); | |
920dd524 | 1105 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); |
34dc7c2f BB |
1106 | } |
1107 | ||
b128c09f BB |
1108 | typedef struct vdev_probe_stats { |
1109 | boolean_t vps_readable; | |
1110 | boolean_t vps_writeable; | |
1111 | int vps_flags; | |
b128c09f BB |
1112 | } vdev_probe_stats_t; |
1113 | ||
1114 | static void | |
1115 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 1116 | { |
fb5f0bc8 | 1117 | spa_t *spa = zio->io_spa; |
d164b209 | 1118 | vdev_t *vd = zio->io_vd; |
b128c09f | 1119 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
1120 | |
1121 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
1122 | |
1123 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
1124 | if (zio->io_error == 0) |
1125 | vps->vps_readable = 1; | |
fb5f0bc8 | 1126 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 1127 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
a6255b7f | 1128 | zio->io_offset, zio->io_size, zio->io_abd, |
b128c09f BB |
1129 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1130 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
1131 | } else { | |
a6255b7f | 1132 | abd_free(zio->io_abd); |
b128c09f BB |
1133 | } |
1134 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
1135 | if (zio->io_error == 0) |
1136 | vps->vps_writeable = 1; | |
a6255b7f | 1137 | abd_free(zio->io_abd); |
b128c09f | 1138 | } else if (zio->io_type == ZIO_TYPE_NULL) { |
d164b209 | 1139 | zio_t *pio; |
3dfb57a3 | 1140 | zio_link_t *zl; |
b128c09f BB |
1141 | |
1142 | vd->vdev_cant_read |= !vps->vps_readable; | |
1143 | vd->vdev_cant_write |= !vps->vps_writeable; | |
1144 | ||
1145 | if (vdev_readable(vd) && | |
fb5f0bc8 | 1146 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
1147 | zio->io_error = 0; |
1148 | } else { | |
1149 | ASSERT(zio->io_error != 0); | |
a1d477c2 MA |
1150 | zfs_dbgmsg("failed probe on vdev %llu", |
1151 | (longlong_t)vd->vdev_id); | |
b128c09f | 1152 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, |
b5256303 | 1153 | spa, vd, NULL, NULL, 0, 0); |
2e528b49 | 1154 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f | 1155 | } |
d164b209 BB |
1156 | |
1157 | mutex_enter(&vd->vdev_probe_lock); | |
1158 | ASSERT(vd->vdev_probe_zio == zio); | |
1159 | vd->vdev_probe_zio = NULL; | |
1160 | mutex_exit(&vd->vdev_probe_lock); | |
1161 | ||
3dfb57a3 DB |
1162 | zl = NULL; |
1163 | while ((pio = zio_walk_parents(zio, &zl)) != NULL) | |
d164b209 | 1164 | if (!vdev_accessible(vd, pio)) |
2e528b49 | 1165 | pio->io_error = SET_ERROR(ENXIO); |
d164b209 | 1166 | |
b128c09f BB |
1167 | kmem_free(vps, sizeof (*vps)); |
1168 | } | |
1169 | } | |
34dc7c2f | 1170 | |
b128c09f | 1171 | /* |
d3cc8b15 WA |
1172 | * Determine whether this device is accessible. |
1173 | * | |
1174 | * Read and write to several known locations: the pad regions of each | |
1175 | * vdev label but the first, which we leave alone in case it contains | |
1176 | * a VTOC. | |
b128c09f BB |
1177 | */ |
1178 | zio_t * | |
d164b209 | 1179 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
1180 | { |
1181 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
1182 | vdev_probe_stats_t *vps = NULL; |
1183 | zio_t *pio; | |
1184 | ||
1185 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 1186 | |
d164b209 BB |
1187 | /* |
1188 | * Don't probe the probe. | |
1189 | */ | |
1190 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
1191 | return (NULL); | |
b128c09f | 1192 | |
d164b209 BB |
1193 | /* |
1194 | * To prevent 'probe storms' when a device fails, we create | |
1195 | * just one probe i/o at a time. All zios that want to probe | |
1196 | * this vdev will become parents of the probe io. | |
1197 | */ | |
1198 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 1199 | |
d164b209 | 1200 | if ((pio = vd->vdev_probe_zio) == NULL) { |
79c76d5b | 1201 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); |
d164b209 BB |
1202 | |
1203 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1204 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1205 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1206 | |
1207 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1208 | /* | |
1209 | * vdev_cant_read and vdev_cant_write can only | |
1210 | * transition from TRUE to FALSE when we have the | |
1211 | * SCL_ZIO lock as writer; otherwise they can only | |
1212 | * transition from FALSE to TRUE. This ensures that | |
1213 | * any zio looking at these values can assume that | |
1214 | * failures persist for the life of the I/O. That's | |
1215 | * important because when a device has intermittent | |
1216 | * connectivity problems, we want to ensure that | |
1217 | * they're ascribed to the device (ENXIO) and not | |
1218 | * the zio (EIO). | |
1219 | * | |
1220 | * Since we hold SCL_ZIO as writer here, clear both | |
1221 | * values so the probe can reevaluate from first | |
1222 | * principles. | |
1223 | */ | |
1224 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1225 | vd->vdev_cant_read = B_FALSE; | |
1226 | vd->vdev_cant_write = B_FALSE; | |
1227 | } | |
1228 | ||
1229 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1230 | vdev_probe_done, vps, | |
1231 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1232 | ||
428870ff BB |
1233 | /* |
1234 | * We can't change the vdev state in this context, so we | |
1235 | * kick off an async task to do it on our behalf. | |
1236 | */ | |
d164b209 BB |
1237 | if (zio != NULL) { |
1238 | vd->vdev_probe_wanted = B_TRUE; | |
1239 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1240 | } | |
b128c09f BB |
1241 | } |
1242 | ||
d164b209 BB |
1243 | if (zio != NULL) |
1244 | zio_add_child(zio, pio); | |
b128c09f | 1245 | |
d164b209 | 1246 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1247 | |
d164b209 BB |
1248 | if (vps == NULL) { |
1249 | ASSERT(zio != NULL); | |
1250 | return (NULL); | |
1251 | } | |
b128c09f | 1252 | |
1c27024e | 1253 | for (int l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1254 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1255 | vdev_label_offset(vd->vdev_psize, l, |
a6255b7f DQ |
1256 | offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, |
1257 | abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), | |
b128c09f BB |
1258 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1259 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1260 | } | |
1261 | ||
d164b209 BB |
1262 | if (zio == NULL) |
1263 | return (pio); | |
1264 | ||
1265 | zio_nowait(pio); | |
1266 | return (NULL); | |
34dc7c2f BB |
1267 | } |
1268 | ||
45d1cae3 BB |
1269 | static void |
1270 | vdev_open_child(void *arg) | |
1271 | { | |
1272 | vdev_t *vd = arg; | |
1273 | ||
1274 | vd->vdev_open_thread = curthread; | |
1275 | vd->vdev_open_error = vdev_open(vd); | |
1276 | vd->vdev_open_thread = NULL; | |
1277 | } | |
1278 | ||
6c285672 | 1279 | static boolean_t |
428870ff BB |
1280 | vdev_uses_zvols(vdev_t *vd) |
1281 | { | |
6c285672 JL |
1282 | #ifdef _KERNEL |
1283 | if (zvol_is_zvol(vd->vdev_path)) | |
428870ff | 1284 | return (B_TRUE); |
6c285672 JL |
1285 | #endif |
1286 | ||
1c27024e | 1287 | for (int c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1288 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1289 | return (B_TRUE); | |
6c285672 | 1290 | |
428870ff BB |
1291 | return (B_FALSE); |
1292 | } | |
1293 | ||
45d1cae3 BB |
1294 | void |
1295 | vdev_open_children(vdev_t *vd) | |
1296 | { | |
1297 | taskq_t *tq; | |
1298 | int children = vd->vdev_children; | |
1299 | ||
428870ff BB |
1300 | /* |
1301 | * in order to handle pools on top of zvols, do the opens | |
1302 | * in a single thread so that the same thread holds the | |
1303 | * spa_namespace_lock | |
1304 | */ | |
1305 | if (vdev_uses_zvols(vd)) { | |
13d9a004 | 1306 | retry_sync: |
1c27024e | 1307 | for (int c = 0; c < children; c++) |
428870ff BB |
1308 | vd->vdev_child[c]->vdev_open_error = |
1309 | vdev_open(vd->vdev_child[c]); | |
4770aa06 HJ |
1310 | } else { |
1311 | tq = taskq_create("vdev_open", children, minclsyspri, | |
1312 | children, children, TASKQ_PREPOPULATE); | |
13d9a004 BB |
1313 | if (tq == NULL) |
1314 | goto retry_sync; | |
45d1cae3 | 1315 | |
1c27024e | 1316 | for (int c = 0; c < children; c++) |
4770aa06 | 1317 | VERIFY(taskq_dispatch(tq, vdev_open_child, |
48d3eb40 | 1318 | vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); |
45d1cae3 | 1319 | |
4770aa06 HJ |
1320 | taskq_destroy(tq); |
1321 | } | |
1322 | ||
1323 | vd->vdev_nonrot = B_TRUE; | |
fb40095f | 1324 | |
1c27024e | 1325 | for (int c = 0; c < children; c++) |
fb40095f | 1326 | vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; |
45d1cae3 BB |
1327 | } |
1328 | ||
a1d477c2 MA |
1329 | /* |
1330 | * Compute the raidz-deflation ratio. Note, we hard-code | |
1331 | * in 128k (1 << 17) because it is the "typical" blocksize. | |
1332 | * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, | |
1333 | * otherwise it would inconsistently account for existing bp's. | |
1334 | */ | |
1335 | static void | |
1336 | vdev_set_deflate_ratio(vdev_t *vd) | |
1337 | { | |
1338 | if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { | |
1339 | vd->vdev_deflate_ratio = (1 << 17) / | |
1340 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
1341 | } | |
1342 | } | |
1343 | ||
34dc7c2f BB |
1344 | /* |
1345 | * Prepare a virtual device for access. | |
1346 | */ | |
1347 | int | |
1348 | vdev_open(vdev_t *vd) | |
1349 | { | |
fb5f0bc8 | 1350 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1351 | int error; |
34dc7c2f | 1352 | uint64_t osize = 0; |
1bd201e7 CS |
1353 | uint64_t max_osize = 0; |
1354 | uint64_t asize, max_asize, psize; | |
34dc7c2f BB |
1355 | uint64_t ashift = 0; |
1356 | ||
45d1cae3 BB |
1357 | ASSERT(vd->vdev_open_thread == curthread || |
1358 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1359 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1360 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1361 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1362 | ||
34dc7c2f | 1363 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1364 | vd->vdev_cant_read = B_FALSE; |
1365 | vd->vdev_cant_write = B_FALSE; | |
1366 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1367 | |
428870ff BB |
1368 | /* |
1369 | * If this vdev is not removed, check its fault status. If it's | |
1370 | * faulted, bail out of the open. | |
1371 | */ | |
34dc7c2f BB |
1372 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1373 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1374 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1375 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1376 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1377 | vd->vdev_label_aux); |
2e528b49 | 1378 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1379 | } else if (vd->vdev_offline) { |
1380 | ASSERT(vd->vdev_children == 0); | |
1381 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
2e528b49 | 1382 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1383 | } |
1384 | ||
1bd201e7 | 1385 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1386 | |
428870ff BB |
1387 | /* |
1388 | * Reset the vdev_reopening flag so that we actually close | |
1389 | * the vdev on error. | |
1390 | */ | |
1391 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1392 | if (zio_injection_enabled && error == 0) |
9babb374 | 1393 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1394 | |
1395 | if (error) { | |
1396 | if (vd->vdev_removed && | |
1397 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1398 | vd->vdev_removed = B_FALSE; | |
1399 | ||
1400 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1401 | vd->vdev_stat.vs_aux); | |
1402 | return (error); | |
1403 | } | |
1404 | ||
1405 | vd->vdev_removed = B_FALSE; | |
1406 | ||
428870ff BB |
1407 | /* |
1408 | * Recheck the faulted flag now that we have confirmed that | |
1409 | * the vdev is accessible. If we're faulted, bail. | |
1410 | */ | |
1411 | if (vd->vdev_faulted) { | |
1412 | ASSERT(vd->vdev_children == 0); | |
1413 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1414 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1415 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1416 | vd->vdev_label_aux); | |
2e528b49 | 1417 | return (SET_ERROR(ENXIO)); |
428870ff BB |
1418 | } |
1419 | ||
34dc7c2f BB |
1420 | if (vd->vdev_degraded) { |
1421 | ASSERT(vd->vdev_children == 0); | |
1422 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1423 | VDEV_AUX_ERR_EXCEEDED); | |
1424 | } else { | |
428870ff | 1425 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1426 | } |
1427 | ||
428870ff BB |
1428 | /* |
1429 | * For hole or missing vdevs we just return success. | |
1430 | */ | |
1431 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1432 | return (0); | |
1433 | ||
1c27024e | 1434 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1435 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1436 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1437 | VDEV_AUX_NONE); | |
1438 | break; | |
1439 | } | |
9babb374 | 1440 | } |
34dc7c2f BB |
1441 | |
1442 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1443 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1444 | |
1445 | if (vd->vdev_children == 0) { | |
1446 | if (osize < SPA_MINDEVSIZE) { | |
1447 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1448 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1449 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1450 | } |
1451 | psize = osize; | |
1452 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1453 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1454 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1455 | } else { |
1456 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1457 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1458 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1459 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1460 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1461 | } |
1462 | psize = 0; | |
1463 | asize = osize; | |
1bd201e7 | 1464 | max_asize = max_osize; |
34dc7c2f BB |
1465 | } |
1466 | ||
9d3f7b87 OF |
1467 | /* |
1468 | * If the vdev was expanded, record this so that we can re-create the | |
1469 | * uberblock rings in labels {2,3}, during the next sync. | |
1470 | */ | |
1471 | if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0)) | |
1472 | vd->vdev_copy_uberblocks = B_TRUE; | |
1473 | ||
34dc7c2f BB |
1474 | vd->vdev_psize = psize; |
1475 | ||
9babb374 | 1476 | /* |
2e215fec | 1477 | * Make sure the allocatable size hasn't shrunk too much. |
9babb374 BB |
1478 | */ |
1479 | if (asize < vd->vdev_min_asize) { | |
1480 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1481 | VDEV_AUX_BAD_LABEL); | |
2e528b49 | 1482 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
1483 | } |
1484 | ||
34dc7c2f BB |
1485 | if (vd->vdev_asize == 0) { |
1486 | /* | |
1487 | * This is the first-ever open, so use the computed values. | |
b28e57cb | 1488 | * For compatibility, a different ashift can be requested. |
34dc7c2f BB |
1489 | */ |
1490 | vd->vdev_asize = asize; | |
1bd201e7 | 1491 | vd->vdev_max_asize = max_asize; |
ff61d1a4 | 1492 | if (vd->vdev_ashift == 0) { |
1493 | vd->vdev_ashift = ashift; /* use detected value */ | |
1494 | } | |
1495 | if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || | |
1496 | vd->vdev_ashift > ASHIFT_MAX)) { | |
1497 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1498 | VDEV_AUX_BAD_ASHIFT); | |
1499 | return (SET_ERROR(EDOM)); | |
1500 | } | |
34dc7c2f BB |
1501 | } else { |
1502 | /* | |
32a9872b GW |
1503 | * Detect if the alignment requirement has increased. |
1504 | * We don't want to make the pool unavailable, just | |
1505 | * post an event instead. | |
34dc7c2f | 1506 | */ |
32a9872b GW |
1507 | if (ashift > vd->vdev_top->vdev_ashift && |
1508 | vd->vdev_ops->vdev_op_leaf) { | |
1509 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
b5256303 | 1510 | spa, vd, NULL, NULL, 0, 0); |
34dc7c2f | 1511 | } |
32a9872b | 1512 | |
1bd201e7 | 1513 | vd->vdev_max_asize = max_asize; |
9babb374 | 1514 | } |
34dc7c2f | 1515 | |
9babb374 | 1516 | /* |
2e215fec SH |
1517 | * If all children are healthy we update asize if either: |
1518 | * The asize has increased, due to a device expansion caused by dynamic | |
1519 | * LUN growth or vdev replacement, and automatic expansion is enabled; | |
1520 | * making the additional space available. | |
1521 | * | |
1522 | * The asize has decreased, due to a device shrink usually caused by a | |
1523 | * vdev replace with a smaller device. This ensures that calculations | |
1524 | * based of max_asize and asize e.g. esize are always valid. It's safe | |
1525 | * to do this as we've already validated that asize is greater than | |
1526 | * vdev_min_asize. | |
9babb374 | 1527 | */ |
2e215fec SH |
1528 | if (vd->vdev_state == VDEV_STATE_HEALTHY && |
1529 | ((asize > vd->vdev_asize && | |
1530 | (vd->vdev_expanding || spa->spa_autoexpand)) || | |
1531 | (asize < vd->vdev_asize))) | |
9babb374 | 1532 | vd->vdev_asize = asize; |
34dc7c2f | 1533 | |
9babb374 | 1534 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1535 | |
1536 | /* | |
1537 | * Ensure we can issue some IO before declaring the | |
1538 | * vdev open for business. | |
1539 | */ | |
b128c09f BB |
1540 | if (vd->vdev_ops->vdev_op_leaf && |
1541 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1542 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1543 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1544 | return (error); |
1545 | } | |
1546 | ||
c3520e7f MA |
1547 | /* |
1548 | * Track the min and max ashift values for normal data devices. | |
1549 | */ | |
1550 | if (vd->vdev_top == vd && vd->vdev_ashift != 0 && | |
1551 | !vd->vdev_islog && vd->vdev_aux == NULL) { | |
1552 | if (vd->vdev_ashift > spa->spa_max_ashift) | |
1553 | spa->spa_max_ashift = vd->vdev_ashift; | |
1554 | if (vd->vdev_ashift < spa->spa_min_ashift) | |
1555 | spa->spa_min_ashift = vd->vdev_ashift; | |
1556 | } | |
1557 | ||
34dc7c2f | 1558 | /* |
b128c09f | 1559 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1560 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1561 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1562 | */ |
fb5f0bc8 BB |
1563 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1564 | vdev_resilver_needed(vd, NULL, NULL)) | |
1565 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1566 | |
1567 | return (0); | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Called once the vdevs are all opened, this routine validates the label | |
1572 | * contents. This needs to be done before vdev_load() so that we don't | |
1573 | * inadvertently do repair I/Os to the wrong device. | |
1574 | * | |
c7f2d69d GW |
1575 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1576 | * if the machine crashed during a re-guid the new guid might have been written | |
1577 | * to all of the vdev labels, but not the cached config. The strict check | |
1578 | * will be performed when the pool is opened again using the mos config. | |
1579 | * | |
34dc7c2f BB |
1580 | * This function will only return failure if one of the vdevs indicates that it |
1581 | * has since been destroyed or exported. This is only possible if | |
1582 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1583 | * will be updated but the function will return 0. | |
1584 | */ | |
1585 | int | |
c7f2d69d | 1586 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1587 | { |
1588 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1589 | nvlist_t *label; |
428870ff | 1590 | uint64_t guid = 0, top_guid; |
34dc7c2f BB |
1591 | uint64_t state; |
1592 | ||
1c27024e | 1593 | for (int c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1594 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
2e528b49 | 1595 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
1596 | |
1597 | /* | |
1598 | * If the device has already failed, or was marked offline, don't do | |
1599 | * any further validation. Otherwise, label I/O will fail and we will | |
1600 | * overwrite the previous state. | |
1601 | */ | |
b128c09f | 1602 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1603 | uint64_t aux_guid = 0; |
1604 | nvlist_t *nvl; | |
295304be GW |
1605 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? |
1606 | spa_last_synced_txg(spa) : -1ULL; | |
34dc7c2f | 1607 | |
3bc7e0fb | 1608 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { |
e35c5a82 | 1609 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
34dc7c2f BB |
1610 | VDEV_AUX_BAD_LABEL); |
1611 | return (0); | |
1612 | } | |
1613 | ||
428870ff BB |
1614 | /* |
1615 | * Determine if this vdev has been split off into another | |
1616 | * pool. If so, then refuse to open it. | |
1617 | */ | |
1618 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1619 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1620 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1621 | VDEV_AUX_SPLIT_POOL); | |
1622 | nvlist_free(label); | |
1623 | return (0); | |
1624 | } | |
1625 | ||
c7f2d69d GW |
1626 | if (strict && (nvlist_lookup_uint64(label, |
1627 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1628 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1629 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1630 | VDEV_AUX_CORRUPT_DATA); | |
1631 | nvlist_free(label); | |
1632 | return (0); | |
1633 | } | |
1634 | ||
428870ff BB |
1635 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1636 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1637 | &aux_guid) != 0) | |
1638 | aux_guid = 0; | |
1639 | ||
b128c09f BB |
1640 | /* |
1641 | * If this vdev just became a top-level vdev because its | |
1642 | * sibling was detached, it will have adopted the parent's | |
1643 | * vdev guid -- but the label may or may not be on disk yet. | |
1644 | * Fortunately, either version of the label will have the | |
1645 | * same top guid, so if we're a top-level vdev, we can | |
1646 | * safely compare to that instead. | |
428870ff BB |
1647 | * |
1648 | * If we split this vdev off instead, then we also check the | |
1649 | * original pool's guid. We don't want to consider the vdev | |
1650 | * corrupt if it is partway through a split operation. | |
b128c09f | 1651 | */ |
34dc7c2f | 1652 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1653 | &guid) != 0 || |
1654 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1655 | &top_guid) != 0 || | |
428870ff | 1656 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1657 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1658 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1659 | VDEV_AUX_CORRUPT_DATA); | |
1660 | nvlist_free(label); | |
1661 | return (0); | |
1662 | } | |
1663 | ||
1664 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1665 | &state) != 0) { | |
1666 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1667 | VDEV_AUX_CORRUPT_DATA); | |
1668 | nvlist_free(label); | |
1669 | return (0); | |
1670 | } | |
1671 | ||
1672 | nvlist_free(label); | |
1673 | ||
45d1cae3 | 1674 | /* |
572e2857 | 1675 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1676 | * state of the pool. |
1677 | */ | |
572e2857 | 1678 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1679 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f | 1680 | state != POOL_STATE_ACTIVE) |
2e528b49 | 1681 | return (SET_ERROR(EBADF)); |
34dc7c2f | 1682 | |
b128c09f BB |
1683 | /* |
1684 | * If we were able to open and validate a vdev that was | |
1685 | * previously marked permanently unavailable, clear that state | |
1686 | * now. | |
1687 | */ | |
1688 | if (vd->vdev_not_present) | |
1689 | vd->vdev_not_present = 0; | |
1690 | } | |
34dc7c2f BB |
1691 | |
1692 | return (0); | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * Close a virtual device. | |
1697 | */ | |
1698 | void | |
1699 | vdev_close(vdev_t *vd) | |
1700 | { | |
428870ff | 1701 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1702 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1703 | |
1704 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1705 | ||
428870ff BB |
1706 | /* |
1707 | * If our parent is reopening, then we are as well, unless we are | |
1708 | * going offline. | |
1709 | */ | |
1710 | if (pvd != NULL && pvd->vdev_reopening) | |
1711 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1712 | ||
34dc7c2f BB |
1713 | vd->vdev_ops->vdev_op_close(vd); |
1714 | ||
1715 | vdev_cache_purge(vd); | |
1716 | ||
1717 | /* | |
9babb374 | 1718 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1719 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1720 | * it's still faulted. | |
1721 | */ | |
1722 | vd->vdev_prevstate = vd->vdev_state; | |
1723 | ||
1724 | if (vd->vdev_offline) | |
1725 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1726 | else | |
1727 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1728 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1729 | } | |
1730 | ||
428870ff BB |
1731 | void |
1732 | vdev_hold(vdev_t *vd) | |
1733 | { | |
1734 | spa_t *spa = vd->vdev_spa; | |
1735 | ||
1736 | ASSERT(spa_is_root(spa)); | |
1737 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1738 | return; | |
1739 | ||
1c27024e | 1740 | for (int c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1741 | vdev_hold(vd->vdev_child[c]); |
1742 | ||
1743 | if (vd->vdev_ops->vdev_op_leaf) | |
1744 | vd->vdev_ops->vdev_op_hold(vd); | |
1745 | } | |
1746 | ||
1747 | void | |
1748 | vdev_rele(vdev_t *vd) | |
1749 | { | |
d6320ddb | 1750 | ASSERT(spa_is_root(vd->vdev_spa)); |
1c27024e | 1751 | for (int c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1752 | vdev_rele(vd->vdev_child[c]); |
1753 | ||
1754 | if (vd->vdev_ops->vdev_op_leaf) | |
1755 | vd->vdev_ops->vdev_op_rele(vd); | |
1756 | } | |
1757 | ||
1758 | /* | |
1759 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1760 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1761 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1762 | * If the leaf has never been opened then open it, as usual. | |
1763 | */ | |
34dc7c2f BB |
1764 | void |
1765 | vdev_reopen(vdev_t *vd) | |
1766 | { | |
1767 | spa_t *spa = vd->vdev_spa; | |
1768 | ||
b128c09f | 1769 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1770 | |
428870ff BB |
1771 | /* set the reopening flag unless we're taking the vdev offline */ |
1772 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1773 | vdev_close(vd); |
1774 | (void) vdev_open(vd); | |
1775 | ||
1776 | /* | |
1777 | * Call vdev_validate() here to make sure we have the same device. | |
1778 | * Otherwise, a device with an invalid label could be successfully | |
1779 | * opened in response to vdev_reopen(). | |
1780 | */ | |
b128c09f BB |
1781 | if (vd->vdev_aux) { |
1782 | (void) vdev_validate_aux(vd); | |
1783 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1784 | vd->vdev_aux == &spa->spa_l2cache && |
1785 | !l2arc_vdev_present(vd)) | |
1786 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1787 | } else { |
295304be | 1788 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1789 | } |
34dc7c2f BB |
1790 | |
1791 | /* | |
1792 | * Reassess parent vdev's health. | |
1793 | */ | |
1794 | vdev_propagate_state(vd); | |
1795 | } | |
1796 | ||
1797 | int | |
1798 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1799 | { | |
1800 | int error; | |
1801 | ||
1802 | /* | |
1803 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1804 | * For a create, however, we want to fail the request if | |
1805 | * there are any components we can't open. | |
1806 | */ | |
1807 | error = vdev_open(vd); | |
1808 | ||
1809 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1810 | vdev_close(vd); | |
1811 | return (error ? error : ENXIO); | |
1812 | } | |
1813 | ||
1814 | /* | |
93cf2076 | 1815 | * Recursively load DTLs and initialize all labels. |
34dc7c2f | 1816 | */ |
93cf2076 GW |
1817 | if ((error = vdev_dtl_load(vd)) != 0 || |
1818 | (error = vdev_label_init(vd, txg, isreplacing ? | |
34dc7c2f BB |
1819 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { |
1820 | vdev_close(vd); | |
1821 | return (error); | |
1822 | } | |
1823 | ||
1824 | return (0); | |
1825 | } | |
1826 | ||
34dc7c2f | 1827 | void |
9babb374 | 1828 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1829 | { |
1830 | /* | |
b8bcca18 | 1831 | * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. |
34dc7c2f | 1832 | */ |
b8bcca18 | 1833 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); |
34dc7c2f | 1834 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); |
34dc7c2f BB |
1835 | } |
1836 | ||
1837 | void | |
1838 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1839 | { | |
1840 | ASSERT(vd == vd->vdev_top); | |
a1d477c2 MA |
1841 | /* indirect vdevs don't have metaslabs or dtls */ |
1842 | ASSERT(vdev_is_concrete(vd) || flags == 0); | |
34dc7c2f | 1843 | ASSERT(ISP2(flags)); |
572e2857 | 1844 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1845 | |
1846 | if (flags & VDD_METASLAB) | |
1847 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1848 | ||
1849 | if (flags & VDD_DTL) | |
1850 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1851 | ||
1852 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1853 | } | |
1854 | ||
93cf2076 GW |
1855 | void |
1856 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1857 | { | |
1c27024e | 1858 | for (int c = 0; c < vd->vdev_children; c++) |
93cf2076 GW |
1859 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); |
1860 | ||
1861 | if (vd->vdev_ops->vdev_op_leaf) | |
1862 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1863 | } | |
1864 | ||
fb5f0bc8 BB |
1865 | /* |
1866 | * DTLs. | |
1867 | * | |
1868 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1869 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1870 | * |
1871 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1872 | * | |
1873 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1874 | * | |
1875 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1876 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1877 | * txgs that was scrubbed. | |
1878 | * | |
1879 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1880 | * persistent errors or just some device being offline. | |
1881 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1882 | * maintained; it's only computed when needed, typically to | |
1883 | * determine whether a device can be detached. | |
1884 | * | |
1885 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1886 | * either has the data or it doesn't. | |
1887 | * | |
1888 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1889 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1890 | * if any child is less than fully replicated, then so is its parent. | |
1891 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1892 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1893 | * those are the txgs we don't have enough replication to read. For example, | |
1894 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1895 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1896 | * two child DTL_MISSING maps. | |
1897 | * | |
1898 | * It should be clear from the above that to compute the DTLs and outage maps | |
1899 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1900 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1901 | * a configuration change, we generate all other DTLs from first principles. | |
1902 | */ | |
34dc7c2f | 1903 | void |
fb5f0bc8 | 1904 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1905 | { |
93cf2076 | 1906 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1907 | |
1908 | ASSERT(t < DTL_TYPES); | |
1909 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1910 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1911 | |
a1d477c2 | 1912 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 GW |
1913 | if (!range_tree_contains(rt, txg, size)) |
1914 | range_tree_add(rt, txg, size); | |
a1d477c2 | 1915 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1916 | } |
1917 | ||
fb5f0bc8 BB |
1918 | boolean_t |
1919 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1920 | { |
93cf2076 | 1921 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 | 1922 | boolean_t dirty = B_FALSE; |
34dc7c2f | 1923 | |
fb5f0bc8 BB |
1924 | ASSERT(t < DTL_TYPES); |
1925 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f | 1926 | |
a1d477c2 MA |
1927 | /* |
1928 | * While we are loading the pool, the DTLs have not been loaded yet. | |
1929 | * Ignore the DTLs and try all devices. This avoids a recursive | |
1930 | * mutex enter on the vdev_dtl_lock, and also makes us try hard | |
1931 | * when loading the pool (relying on the checksum to ensure that | |
1932 | * we get the right data -- note that we while loading, we are | |
1933 | * only reading the MOS, which is always checksummed). | |
1934 | */ | |
1935 | if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) | |
1936 | return (B_FALSE); | |
1937 | ||
1938 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 GW |
1939 | if (range_tree_space(rt) != 0) |
1940 | dirty = range_tree_contains(rt, txg, size); | |
a1d477c2 | 1941 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1942 | |
1943 | return (dirty); | |
1944 | } | |
1945 | ||
fb5f0bc8 BB |
1946 | boolean_t |
1947 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1948 | { | |
93cf2076 | 1949 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1950 | boolean_t empty; |
1951 | ||
a1d477c2 | 1952 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 | 1953 | empty = (range_tree_space(rt) == 0); |
a1d477c2 | 1954 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 BB |
1955 | |
1956 | return (empty); | |
1957 | } | |
1958 | ||
3d6da72d IH |
1959 | /* |
1960 | * Returns B_TRUE if vdev determines offset needs to be resilvered. | |
1961 | */ | |
1962 | boolean_t | |
1963 | vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) | |
1964 | { | |
1965 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
1966 | ||
1967 | if (vd->vdev_ops->vdev_op_need_resilver == NULL || | |
1968 | vd->vdev_ops->vdev_op_leaf) | |
1969 | return (B_TRUE); | |
1970 | ||
1971 | return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); | |
1972 | } | |
1973 | ||
5d1f7fb6 GW |
1974 | /* |
1975 | * Returns the lowest txg in the DTL range. | |
1976 | */ | |
1977 | static uint64_t | |
1978 | vdev_dtl_min(vdev_t *vd) | |
1979 | { | |
93cf2076 | 1980 | range_seg_t *rs; |
5d1f7fb6 GW |
1981 | |
1982 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1983 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1984 | ASSERT0(vd->vdev_children); |
1985 | ||
93cf2076 GW |
1986 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1987 | return (rs->rs_start - 1); | |
5d1f7fb6 GW |
1988 | } |
1989 | ||
1990 | /* | |
1991 | * Returns the highest txg in the DTL. | |
1992 | */ | |
1993 | static uint64_t | |
1994 | vdev_dtl_max(vdev_t *vd) | |
1995 | { | |
93cf2076 | 1996 | range_seg_t *rs; |
5d1f7fb6 GW |
1997 | |
1998 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1999 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
2000 | ASSERT0(vd->vdev_children); |
2001 | ||
93cf2076 GW |
2002 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
2003 | return (rs->rs_end); | |
5d1f7fb6 GW |
2004 | } |
2005 | ||
2006 | /* | |
2007 | * Determine if a resilvering vdev should remove any DTL entries from | |
2008 | * its range. If the vdev was resilvering for the entire duration of the | |
2009 | * scan then it should excise that range from its DTLs. Otherwise, this | |
2010 | * vdev is considered partially resilvered and should leave its DTL | |
2011 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
2012 | * excise the DTLs. | |
2013 | */ | |
2014 | static boolean_t | |
2015 | vdev_dtl_should_excise(vdev_t *vd) | |
2016 | { | |
2017 | spa_t *spa = vd->vdev_spa; | |
2018 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
2019 | ||
2020 | ASSERT0(scn->scn_phys.scn_errors); | |
2021 | ASSERT0(vd->vdev_children); | |
2022 | ||
335b251a MA |
2023 | if (vd->vdev_state < VDEV_STATE_DEGRADED) |
2024 | return (B_FALSE); | |
2025 | ||
5d1f7fb6 | 2026 | if (vd->vdev_resilver_txg == 0 || |
93cf2076 | 2027 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) |
5d1f7fb6 GW |
2028 | return (B_TRUE); |
2029 | ||
2030 | /* | |
2031 | * When a resilver is initiated the scan will assign the scn_max_txg | |
2032 | * value to the highest txg value that exists in all DTLs. If this | |
2033 | * device's max DTL is not part of this scan (i.e. it is not in | |
2034 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
2035 | * for excision. | |
2036 | */ | |
2037 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
2038 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
2039 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
2040 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
2041 | return (B_TRUE); | |
2042 | } | |
2043 | return (B_FALSE); | |
2044 | } | |
2045 | ||
34dc7c2f BB |
2046 | /* |
2047 | * Reassess DTLs after a config change or scrub completion. | |
2048 | */ | |
2049 | void | |
2050 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
2051 | { | |
2052 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 2053 | avl_tree_t reftree; |
1c27024e | 2054 | int minref; |
34dc7c2f | 2055 | |
fb5f0bc8 | 2056 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 2057 | |
1c27024e | 2058 | for (int c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
2059 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
2060 | scrub_txg, scrub_done); | |
2061 | ||
a1d477c2 | 2062 | if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) |
fb5f0bc8 BB |
2063 | return; |
2064 | ||
2065 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
2066 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
2067 | ||
34dc7c2f | 2068 | mutex_enter(&vd->vdev_dtl_lock); |
5d1f7fb6 | 2069 | |
02638a30 TC |
2070 | /* |
2071 | * If requested, pretend the scan completed cleanly. | |
2072 | */ | |
2073 | if (zfs_scan_ignore_errors && scn) | |
2074 | scn->scn_phys.scn_errors = 0; | |
2075 | ||
5d1f7fb6 GW |
2076 | /* |
2077 | * If we've completed a scan cleanly then determine | |
2078 | * if this vdev should remove any DTLs. We only want to | |
2079 | * excise regions on vdevs that were available during | |
2080 | * the entire duration of this scan. | |
2081 | */ | |
b128c09f | 2082 | if (scrub_txg != 0 && |
428870ff | 2083 | (spa->spa_scrub_started || |
5d1f7fb6 GW |
2084 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && |
2085 | vdev_dtl_should_excise(vd)) { | |
b128c09f BB |
2086 | /* |
2087 | * We completed a scrub up to scrub_txg. If we | |
2088 | * did it without rebooting, then the scrub dtl | |
2089 | * will be valid, so excise the old region and | |
2090 | * fold in the scrub dtl. Otherwise, leave the | |
2091 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
2092 | * |
2093 | * There's little trick here: to excise the beginning | |
2094 | * of the DTL_MISSING map, we put it into a reference | |
2095 | * tree and then add a segment with refcnt -1 that | |
2096 | * covers the range [0, scrub_txg). This means | |
2097 | * that each txg in that range has refcnt -1 or 0. | |
2098 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
2099 | * entries in the range [0, scrub_txg) will have a | |
2100 | * positive refcnt -- either 1 or 2. We then convert | |
2101 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 2102 | */ |
93cf2076 GW |
2103 | space_reftree_create(&reftree); |
2104 | space_reftree_add_map(&reftree, | |
2105 | vd->vdev_dtl[DTL_MISSING], 1); | |
2106 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
2107 | space_reftree_add_map(&reftree, | |
2108 | vd->vdev_dtl[DTL_SCRUB], 2); | |
2109 | space_reftree_generate_map(&reftree, | |
2110 | vd->vdev_dtl[DTL_MISSING], 1); | |
2111 | space_reftree_destroy(&reftree); | |
34dc7c2f | 2112 | } |
93cf2076 GW |
2113 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
2114 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
2115 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 2116 | if (scrub_done) |
93cf2076 GW |
2117 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
2118 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
fb5f0bc8 | 2119 | if (!vdev_readable(vd)) |
93cf2076 | 2120 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); |
fb5f0bc8 | 2121 | else |
93cf2076 GW |
2122 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], |
2123 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
5d1f7fb6 GW |
2124 | |
2125 | /* | |
2126 | * If the vdev was resilvering and no longer has any | |
d14fa5db | 2127 | * DTLs then reset its resilvering flag and dirty |
2128 | * the top level so that we persist the change. | |
5d1f7fb6 GW |
2129 | */ |
2130 | if (vd->vdev_resilver_txg != 0 && | |
93cf2076 | 2131 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && |
d14fa5db | 2132 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) { |
5d1f7fb6 | 2133 | vd->vdev_resilver_txg = 0; |
d14fa5db | 2134 | vdev_config_dirty(vd->vdev_top); |
2135 | } | |
5d1f7fb6 | 2136 | |
34dc7c2f | 2137 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 2138 | |
34dc7c2f BB |
2139 | if (txg != 0) |
2140 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
2141 | return; | |
2142 | } | |
2143 | ||
34dc7c2f | 2144 | mutex_enter(&vd->vdev_dtl_lock); |
1c27024e | 2145 | for (int t = 0; t < DTL_TYPES; t++) { |
428870ff BB |
2146 | /* account for child's outage in parent's missing map */ |
2147 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
2148 | if (t == DTL_SCRUB) |
2149 | continue; /* leaf vdevs only */ | |
2150 | if (t == DTL_PARTIAL) | |
2151 | minref = 1; /* i.e. non-zero */ | |
2152 | else if (vd->vdev_nparity != 0) | |
2153 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
2154 | else | |
2155 | minref = vd->vdev_children; /* any kind of mirror */ | |
93cf2076 | 2156 | space_reftree_create(&reftree); |
1c27024e | 2157 | for (int c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
2158 | vdev_t *cvd = vd->vdev_child[c]; |
2159 | mutex_enter(&cvd->vdev_dtl_lock); | |
93cf2076 | 2160 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
2161 | mutex_exit(&cvd->vdev_dtl_lock); |
2162 | } | |
93cf2076 GW |
2163 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); |
2164 | space_reftree_destroy(&reftree); | |
34dc7c2f | 2165 | } |
fb5f0bc8 | 2166 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
2167 | } |
2168 | ||
93cf2076 | 2169 | int |
34dc7c2f BB |
2170 | vdev_dtl_load(vdev_t *vd) |
2171 | { | |
2172 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 2173 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 | 2174 | int error = 0; |
34dc7c2f | 2175 | |
93cf2076 | 2176 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { |
a1d477c2 | 2177 | ASSERT(vdev_is_concrete(vd)); |
34dc7c2f | 2178 | |
93cf2076 | 2179 | error = space_map_open(&vd->vdev_dtl_sm, mos, |
a1d477c2 | 2180 | vd->vdev_dtl_object, 0, -1ULL, 0); |
93cf2076 GW |
2181 | if (error) |
2182 | return (error); | |
2183 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f | 2184 | |
93cf2076 | 2185 | mutex_enter(&vd->vdev_dtl_lock); |
428870ff | 2186 | |
93cf2076 GW |
2187 | /* |
2188 | * Now that we've opened the space_map we need to update | |
2189 | * the in-core DTL. | |
2190 | */ | |
2191 | space_map_update(vd->vdev_dtl_sm); | |
34dc7c2f | 2192 | |
93cf2076 GW |
2193 | error = space_map_load(vd->vdev_dtl_sm, |
2194 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
2195 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 2196 | |
93cf2076 GW |
2197 | return (error); |
2198 | } | |
2199 | ||
1c27024e | 2200 | for (int c = 0; c < vd->vdev_children; c++) { |
93cf2076 GW |
2201 | error = vdev_dtl_load(vd->vdev_child[c]); |
2202 | if (error != 0) | |
2203 | break; | |
2204 | } | |
34dc7c2f BB |
2205 | |
2206 | return (error); | |
2207 | } | |
2208 | ||
e0ab3ab5 JS |
2209 | void |
2210 | vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) | |
2211 | { | |
2212 | spa_t *spa = vd->vdev_spa; | |
2213 | ||
2214 | VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); | |
2215 | VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2216 | zapobj, tx)); | |
2217 | } | |
2218 | ||
2219 | uint64_t | |
2220 | vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) | |
2221 | { | |
2222 | spa_t *spa = vd->vdev_spa; | |
2223 | uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, | |
2224 | DMU_OT_NONE, 0, tx); | |
2225 | ||
2226 | ASSERT(zap != 0); | |
2227 | VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2228 | zap, tx)); | |
2229 | ||
2230 | return (zap); | |
2231 | } | |
2232 | ||
2233 | void | |
2234 | vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) | |
2235 | { | |
e0ab3ab5 JS |
2236 | if (vd->vdev_ops != &vdev_hole_ops && |
2237 | vd->vdev_ops != &vdev_missing_ops && | |
2238 | vd->vdev_ops != &vdev_root_ops && | |
2239 | !vd->vdev_top->vdev_removing) { | |
2240 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { | |
2241 | vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); | |
2242 | } | |
2243 | if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { | |
2244 | vd->vdev_top_zap = vdev_create_link_zap(vd, tx); | |
2245 | } | |
2246 | } | |
1c27024e | 2247 | for (uint64_t i = 0; i < vd->vdev_children; i++) { |
e0ab3ab5 JS |
2248 | vdev_construct_zaps(vd->vdev_child[i], tx); |
2249 | } | |
2250 | } | |
2251 | ||
34dc7c2f BB |
2252 | void |
2253 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
2254 | { | |
2255 | spa_t *spa = vd->vdev_spa; | |
93cf2076 | 2256 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; |
34dc7c2f | 2257 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 | 2258 | range_tree_t *rtsync; |
34dc7c2f | 2259 | dmu_tx_t *tx; |
93cf2076 | 2260 | uint64_t object = space_map_object(vd->vdev_dtl_sm); |
34dc7c2f | 2261 | |
a1d477c2 | 2262 | ASSERT(vdev_is_concrete(vd)); |
93cf2076 | 2263 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
428870ff | 2264 | |
34dc7c2f BB |
2265 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
2266 | ||
93cf2076 GW |
2267 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { |
2268 | mutex_enter(&vd->vdev_dtl_lock); | |
2269 | space_map_free(vd->vdev_dtl_sm, tx); | |
2270 | space_map_close(vd->vdev_dtl_sm); | |
2271 | vd->vdev_dtl_sm = NULL; | |
2272 | mutex_exit(&vd->vdev_dtl_lock); | |
e0ab3ab5 JS |
2273 | |
2274 | /* | |
2275 | * We only destroy the leaf ZAP for detached leaves or for | |
2276 | * removed log devices. Removed data devices handle leaf ZAP | |
2277 | * cleanup later, once cancellation is no longer possible. | |
2278 | */ | |
2279 | if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || | |
2280 | vd->vdev_top->vdev_islog)) { | |
2281 | vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); | |
2282 | vd->vdev_leaf_zap = 0; | |
2283 | } | |
2284 | ||
34dc7c2f | 2285 | dmu_tx_commit(tx); |
34dc7c2f BB |
2286 | return; |
2287 | } | |
2288 | ||
93cf2076 GW |
2289 | if (vd->vdev_dtl_sm == NULL) { |
2290 | uint64_t new_object; | |
2291 | ||
2292 | new_object = space_map_alloc(mos, tx); | |
2293 | VERIFY3U(new_object, !=, 0); | |
2294 | ||
2295 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
a1d477c2 | 2296 | 0, -1ULL, 0)); |
93cf2076 | 2297 | ASSERT(vd->vdev_dtl_sm != NULL); |
34dc7c2f BB |
2298 | } |
2299 | ||
a1d477c2 | 2300 | rtsync = range_tree_create(NULL, NULL); |
34dc7c2f BB |
2301 | |
2302 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2303 | range_tree_walk(rt, range_tree_add, rtsync); |
34dc7c2f BB |
2304 | mutex_exit(&vd->vdev_dtl_lock); |
2305 | ||
93cf2076 GW |
2306 | space_map_truncate(vd->vdev_dtl_sm, tx); |
2307 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
2308 | range_tree_vacate(rtsync, NULL, NULL); | |
34dc7c2f | 2309 | |
93cf2076 | 2310 | range_tree_destroy(rtsync); |
34dc7c2f | 2311 | |
93cf2076 GW |
2312 | /* |
2313 | * If the object for the space map has changed then dirty | |
2314 | * the top level so that we update the config. | |
2315 | */ | |
2316 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
2317 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
2318 | "new object %llu", txg, spa_name(spa), object, | |
2319 | space_map_object(vd->vdev_dtl_sm)); | |
2320 | vdev_config_dirty(vd->vdev_top); | |
2321 | } | |
34dc7c2f BB |
2322 | |
2323 | dmu_tx_commit(tx); | |
93cf2076 GW |
2324 | |
2325 | mutex_enter(&vd->vdev_dtl_lock); | |
2326 | space_map_update(vd->vdev_dtl_sm); | |
2327 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f BB |
2328 | } |
2329 | ||
fb5f0bc8 BB |
2330 | /* |
2331 | * Determine whether the specified vdev can be offlined/detached/removed | |
2332 | * without losing data. | |
2333 | */ | |
2334 | boolean_t | |
2335 | vdev_dtl_required(vdev_t *vd) | |
2336 | { | |
2337 | spa_t *spa = vd->vdev_spa; | |
2338 | vdev_t *tvd = vd->vdev_top; | |
2339 | uint8_t cant_read = vd->vdev_cant_read; | |
2340 | boolean_t required; | |
2341 | ||
2342 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2343 | ||
2344 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2345 | return (B_TRUE); | |
2346 | ||
2347 | /* | |
2348 | * Temporarily mark the device as unreadable, and then determine | |
2349 | * whether this results in any DTL outages in the top-level vdev. | |
2350 | * If not, we can safely offline/detach/remove the device. | |
2351 | */ | |
2352 | vd->vdev_cant_read = B_TRUE; | |
2353 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2354 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2355 | vd->vdev_cant_read = cant_read; | |
2356 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2357 | ||
572e2857 BB |
2358 | if (!required && zio_injection_enabled) |
2359 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2360 | ||
fb5f0bc8 BB |
2361 | return (required); |
2362 | } | |
2363 | ||
b128c09f BB |
2364 | /* |
2365 | * Determine if resilver is needed, and if so the txg range. | |
2366 | */ | |
2367 | boolean_t | |
2368 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2369 | { | |
2370 | boolean_t needed = B_FALSE; | |
2371 | uint64_t thismin = UINT64_MAX; | |
2372 | uint64_t thismax = 0; | |
2373 | ||
2374 | if (vd->vdev_children == 0) { | |
2375 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2376 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && |
fb5f0bc8 | 2377 | vdev_writeable(vd)) { |
b128c09f | 2378 | |
5d1f7fb6 GW |
2379 | thismin = vdev_dtl_min(vd); |
2380 | thismax = vdev_dtl_max(vd); | |
b128c09f BB |
2381 | needed = B_TRUE; |
2382 | } | |
2383 | mutex_exit(&vd->vdev_dtl_lock); | |
2384 | } else { | |
1c27024e | 2385 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2386 | vdev_t *cvd = vd->vdev_child[c]; |
2387 | uint64_t cmin, cmax; | |
2388 | ||
2389 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2390 | thismin = MIN(thismin, cmin); | |
2391 | thismax = MAX(thismax, cmax); | |
2392 | needed = B_TRUE; | |
2393 | } | |
2394 | } | |
2395 | } | |
2396 | ||
2397 | if (needed && minp) { | |
2398 | *minp = thismin; | |
2399 | *maxp = thismax; | |
2400 | } | |
2401 | return (needed); | |
2402 | } | |
2403 | ||
a1d477c2 | 2404 | int |
34dc7c2f BB |
2405 | vdev_load(vdev_t *vd) |
2406 | { | |
a1d477c2 MA |
2407 | int error = 0; |
2408 | ||
34dc7c2f BB |
2409 | /* |
2410 | * Recursively load all children. | |
2411 | */ | |
a1d477c2 MA |
2412 | for (int c = 0; c < vd->vdev_children; c++) { |
2413 | error = vdev_load(vd->vdev_child[c]); | |
2414 | if (error != 0) { | |
2415 | return (error); | |
2416 | } | |
2417 | } | |
2418 | ||
2419 | vdev_set_deflate_ratio(vd); | |
34dc7c2f BB |
2420 | |
2421 | /* | |
2422 | * If this is a top-level vdev, initialize its metaslabs. | |
2423 | */ | |
a1d477c2 MA |
2424 | if (vd == vd->vdev_top && vdev_is_concrete(vd)) { |
2425 | if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { | |
2426 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2427 | VDEV_AUX_CORRUPT_DATA); | |
2428 | return (SET_ERROR(ENXIO)); | |
2429 | } else if ((error = vdev_metaslab_init(vd, 0)) != 0) { | |
2430 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2431 | VDEV_AUX_CORRUPT_DATA); | |
2432 | return (error); | |
2433 | } | |
2434 | } | |
2435 | ||
34dc7c2f BB |
2436 | /* |
2437 | * If this is a leaf vdev, load its DTL. | |
2438 | */ | |
a1d477c2 | 2439 | if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { |
34dc7c2f BB |
2440 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
2441 | VDEV_AUX_CORRUPT_DATA); | |
a1d477c2 MA |
2442 | return (error); |
2443 | } | |
2444 | ||
2445 | uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); | |
2446 | if (obsolete_sm_object != 0) { | |
2447 | objset_t *mos = vd->vdev_spa->spa_meta_objset; | |
2448 | ASSERT(vd->vdev_asize != 0); | |
2449 | ASSERT(vd->vdev_obsolete_sm == NULL); | |
2450 | ||
2451 | if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, | |
2452 | obsolete_sm_object, 0, vd->vdev_asize, 0))) { | |
2453 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2454 | VDEV_AUX_CORRUPT_DATA); | |
2455 | return (error); | |
2456 | } | |
2457 | space_map_update(vd->vdev_obsolete_sm); | |
2458 | } | |
2459 | ||
2460 | return (0); | |
34dc7c2f BB |
2461 | } |
2462 | ||
2463 | /* | |
2464 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2465 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2466 | * we make sure that we can open the underlying device, then try to read the | |
2467 | * label, and make sure that the label is sane and that it hasn't been | |
2468 | * repurposed to another pool. | |
2469 | */ | |
2470 | int | |
2471 | vdev_validate_aux(vdev_t *vd) | |
2472 | { | |
2473 | nvlist_t *label; | |
2474 | uint64_t guid, version; | |
2475 | uint64_t state; | |
2476 | ||
b128c09f BB |
2477 | if (!vdev_readable(vd)) |
2478 | return (0); | |
2479 | ||
3bc7e0fb | 2480 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { |
34dc7c2f BB |
2481 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
2482 | VDEV_AUX_CORRUPT_DATA); | |
2483 | return (-1); | |
2484 | } | |
2485 | ||
2486 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
9ae529ec | 2487 | !SPA_VERSION_IS_SUPPORTED(version) || |
34dc7c2f BB |
2488 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || |
2489 | guid != vd->vdev_guid || | |
2490 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2491 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2492 | VDEV_AUX_CORRUPT_DATA); | |
2493 | nvlist_free(label); | |
2494 | return (-1); | |
2495 | } | |
2496 | ||
2497 | /* | |
2498 | * We don't actually check the pool state here. If it's in fact in | |
2499 | * use by another pool, we update this fact on the fly when requested. | |
2500 | */ | |
2501 | nvlist_free(label); | |
2502 | return (0); | |
2503 | } | |
2504 | ||
a1d477c2 MA |
2505 | /* |
2506 | * Free the objects used to store this vdev's spacemaps, and the array | |
2507 | * that points to them. | |
2508 | */ | |
428870ff | 2509 | void |
a1d477c2 MA |
2510 | vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) |
2511 | { | |
2512 | if (vd->vdev_ms_array == 0) | |
2513 | return; | |
2514 | ||
2515 | objset_t *mos = vd->vdev_spa->spa_meta_objset; | |
2516 | uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; | |
2517 | size_t array_bytes = array_count * sizeof (uint64_t); | |
2518 | uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); | |
2519 | VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, | |
2520 | array_bytes, smobj_array, 0)); | |
2521 | ||
2522 | for (uint64_t i = 0; i < array_count; i++) { | |
2523 | uint64_t smobj = smobj_array[i]; | |
2524 | if (smobj == 0) | |
2525 | continue; | |
2526 | ||
2527 | space_map_free_obj(mos, smobj, tx); | |
2528 | } | |
2529 | ||
2530 | kmem_free(smobj_array, array_bytes); | |
2531 | VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); | |
2532 | vd->vdev_ms_array = 0; | |
2533 | } | |
2534 | ||
2535 | static void | |
2536 | vdev_remove_empty(vdev_t *vd, uint64_t txg) | |
428870ff BB |
2537 | { |
2538 | spa_t *spa = vd->vdev_spa; | |
428870ff BB |
2539 | dmu_tx_t *tx; |
2540 | ||
e0ab3ab5 JS |
2541 | ASSERT(vd == vd->vdev_top); |
2542 | ASSERT3U(txg, ==, spa_syncing_txg(spa)); | |
428870ff | 2543 | |
428870ff | 2544 | if (vd->vdev_ms != NULL) { |
f3a7f661 GW |
2545 | metaslab_group_t *mg = vd->vdev_mg; |
2546 | ||
2547 | metaslab_group_histogram_verify(mg); | |
2548 | metaslab_class_histogram_verify(mg->mg_class); | |
2549 | ||
1c27024e | 2550 | for (int m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2551 | metaslab_t *msp = vd->vdev_ms[m]; |
2552 | ||
93cf2076 | 2553 | if (msp == NULL || msp->ms_sm == NULL) |
428870ff BB |
2554 | continue; |
2555 | ||
93cf2076 | 2556 | mutex_enter(&msp->ms_lock); |
f3a7f661 GW |
2557 | /* |
2558 | * If the metaslab was not loaded when the vdev | |
2559 | * was removed then the histogram accounting may | |
2560 | * not be accurate. Update the histogram information | |
2561 | * here so that we ensure that the metaslab group | |
2562 | * and metaslab class are up-to-date. | |
2563 | */ | |
2564 | metaslab_group_histogram_remove(mg, msp); | |
2565 | ||
93cf2076 | 2566 | VERIFY0(space_map_allocated(msp->ms_sm)); |
93cf2076 GW |
2567 | space_map_close(msp->ms_sm); |
2568 | msp->ms_sm = NULL; | |
2569 | mutex_exit(&msp->ms_lock); | |
428870ff | 2570 | } |
f3a7f661 GW |
2571 | |
2572 | metaslab_group_histogram_verify(mg); | |
2573 | metaslab_class_histogram_verify(mg->mg_class); | |
1c27024e | 2574 | for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) |
f3a7f661 | 2575 | ASSERT0(mg->mg_histogram[i]); |
428870ff BB |
2576 | } |
2577 | ||
a1d477c2 MA |
2578 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); |
2579 | vdev_destroy_spacemaps(vd, tx); | |
e0ab3ab5 JS |
2580 | |
2581 | if (vd->vdev_islog && vd->vdev_top_zap != 0) { | |
2582 | vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); | |
2583 | vd->vdev_top_zap = 0; | |
2584 | } | |
428870ff BB |
2585 | dmu_tx_commit(tx); |
2586 | } | |
2587 | ||
34dc7c2f BB |
2588 | void |
2589 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2590 | { | |
2591 | metaslab_t *msp; | |
428870ff BB |
2592 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2593 | ||
a1d477c2 | 2594 | ASSERT(vdev_is_concrete(vd)); |
34dc7c2f | 2595 | |
c65aa5b2 | 2596 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2597 | metaslab_sync_done(msp, txg); |
428870ff BB |
2598 | |
2599 | if (reassess) | |
2600 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2601 | } |
2602 | ||
2603 | void | |
2604 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2605 | { | |
2606 | spa_t *spa = vd->vdev_spa; | |
2607 | vdev_t *lvd; | |
2608 | metaslab_t *msp; | |
2609 | dmu_tx_t *tx; | |
2610 | ||
a1d477c2 MA |
2611 | if (range_tree_space(vd->vdev_obsolete_segments) > 0) { |
2612 | dmu_tx_t *tx; | |
428870ff | 2613 | |
a1d477c2 MA |
2614 | ASSERT(vd->vdev_removing || |
2615 | vd->vdev_ops == &vdev_indirect_ops); | |
2616 | ||
2617 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2618 | vdev_indirect_sync_obsolete(vd, tx); | |
2619 | dmu_tx_commit(tx); | |
2620 | ||
2621 | /* | |
2622 | * If the vdev is indirect, it can't have dirty | |
2623 | * metaslabs or DTLs. | |
2624 | */ | |
2625 | if (vd->vdev_ops == &vdev_indirect_ops) { | |
2626 | ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); | |
2627 | ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); | |
2628 | return; | |
2629 | } | |
2630 | } | |
2631 | ||
2632 | ASSERT(vdev_is_concrete(vd)); | |
2633 | ||
2634 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && | |
2635 | !vd->vdev_removing) { | |
34dc7c2f | 2636 | ASSERT(vd == vd->vdev_top); |
a1d477c2 | 2637 | ASSERT0(vd->vdev_indirect_config.vic_mapping_object); |
34dc7c2f BB |
2638 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
2639 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2640 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2641 | ASSERT(vd->vdev_ms_array != 0); | |
2642 | vdev_config_dirty(vd); | |
2643 | dmu_tx_commit(tx); | |
2644 | } | |
2645 | ||
2646 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { | |
2647 | metaslab_sync(msp, txg); | |
2648 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2649 | } | |
2650 | ||
2651 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2652 | vdev_dtl_sync(lvd, txg); | |
2653 | ||
a1d477c2 MA |
2654 | /* |
2655 | * Remove the metadata associated with this vdev once it's empty. | |
2656 | * Note that this is typically used for log/cache device removal; | |
2657 | * we don't empty toplevel vdevs when removing them. But if | |
2658 | * a toplevel happens to be emptied, this is not harmful. | |
2659 | */ | |
2660 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) { | |
2661 | vdev_remove_empty(vd, txg); | |
2662 | } | |
2663 | ||
34dc7c2f BB |
2664 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); |
2665 | } | |
2666 | ||
2667 | uint64_t | |
2668 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2669 | { | |
2670 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2671 | } | |
2672 | ||
34dc7c2f BB |
2673 | /* |
2674 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2675 | * not be opened, and no I/O is attempted. | |
2676 | */ | |
2677 | int | |
428870ff | 2678 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2679 | { |
572e2857 | 2680 | vdev_t *vd, *tvd; |
34dc7c2f | 2681 | |
428870ff | 2682 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2683 | |
b128c09f BB |
2684 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2685 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2686 | |
34dc7c2f | 2687 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2688 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2689 | |
572e2857 BB |
2690 | tvd = vd->vdev_top; |
2691 | ||
4a283c7f TH |
2692 | /* |
2693 | * If user did a 'zpool offline -f' then make the fault persist across | |
2694 | * reboots. | |
2695 | */ | |
2696 | if (aux == VDEV_AUX_EXTERNAL_PERSIST) { | |
2697 | /* | |
2698 | * There are two kinds of forced faults: temporary and | |
2699 | * persistent. Temporary faults go away at pool import, while | |
2700 | * persistent faults stay set. Both types of faults can be | |
2701 | * cleared with a zpool clear. | |
2702 | * | |
2703 | * We tell if a vdev is persistently faulted by looking at the | |
2704 | * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at | |
2705 | * import then it's a persistent fault. Otherwise, it's | |
2706 | * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external" | |
2707 | * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This | |
2708 | * tells vdev_config_generate() (which gets run later) to set | |
2709 | * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist. | |
2710 | */ | |
2711 | vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; | |
2712 | vd->vdev_tmpoffline = B_FALSE; | |
2713 | aux = VDEV_AUX_EXTERNAL; | |
2714 | } else { | |
2715 | vd->vdev_tmpoffline = B_TRUE; | |
2716 | } | |
2717 | ||
428870ff BB |
2718 | /* |
2719 | * We don't directly use the aux state here, but if we do a | |
2720 | * vdev_reopen(), we need this value to be present to remember why we | |
2721 | * were faulted. | |
2722 | */ | |
2723 | vd->vdev_label_aux = aux; | |
2724 | ||
34dc7c2f BB |
2725 | /* |
2726 | * Faulted state takes precedence over degraded. | |
2727 | */ | |
428870ff | 2728 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2729 | vd->vdev_faulted = 1ULL; |
2730 | vd->vdev_degraded = 0ULL; | |
428870ff | 2731 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2732 | |
2733 | /* | |
428870ff BB |
2734 | * If this device has the only valid copy of the data, then |
2735 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2736 | */ |
572e2857 | 2737 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2738 | vd->vdev_degraded = 1ULL; |
2739 | vd->vdev_faulted = 0ULL; | |
2740 | ||
2741 | /* | |
2742 | * If we reopen the device and it's not dead, only then do we | |
2743 | * mark it degraded. | |
2744 | */ | |
572e2857 | 2745 | vdev_reopen(tvd); |
34dc7c2f | 2746 | |
428870ff BB |
2747 | if (vdev_readable(vd)) |
2748 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2749 | } |
2750 | ||
b128c09f | 2751 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2752 | } |
2753 | ||
2754 | /* | |
2755 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2756 | * user that something is wrong. The vdev continues to operate as normal as far | |
2757 | * as I/O is concerned. | |
2758 | */ | |
2759 | int | |
428870ff | 2760 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2761 | { |
b128c09f | 2762 | vdev_t *vd; |
34dc7c2f | 2763 | |
428870ff | 2764 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2765 | |
b128c09f BB |
2766 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2767 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2768 | |
34dc7c2f | 2769 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2770 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2771 | |
2772 | /* | |
2773 | * If the vdev is already faulted, then don't do anything. | |
2774 | */ | |
b128c09f BB |
2775 | if (vd->vdev_faulted || vd->vdev_degraded) |
2776 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2777 | |
2778 | vd->vdev_degraded = 1ULL; | |
2779 | if (!vdev_is_dead(vd)) | |
2780 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2781 | aux); |
34dc7c2f | 2782 | |
b128c09f | 2783 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2784 | } |
2785 | ||
2786 | /* | |
d3cc8b15 WA |
2787 | * Online the given vdev. |
2788 | * | |
2789 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2790 | * spare device should be detached when the device finishes resilvering. | |
2791 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2792 | * events are generated if the device fails to open. | |
34dc7c2f BB |
2793 | */ |
2794 | int | |
b128c09f | 2795 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2796 | { |
9babb374 | 2797 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
153b2285 YP |
2798 | boolean_t wasoffline; |
2799 | vdev_state_t oldstate; | |
34dc7c2f | 2800 | |
428870ff | 2801 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2802 | |
b128c09f BB |
2803 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2804 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2805 | |
2806 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2807 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2808 | |
153b2285 YP |
2809 | wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); |
2810 | oldstate = vd->vdev_state; | |
fb390aaf | 2811 | |
9babb374 | 2812 | tvd = vd->vdev_top; |
34dc7c2f BB |
2813 | vd->vdev_offline = B_FALSE; |
2814 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2815 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2816 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2817 | |
2818 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2819 | if (!vd->vdev_aux) { | |
2820 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2821 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2822 | } | |
2823 | ||
2824 | vdev_reopen(tvd); | |
34dc7c2f BB |
2825 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2826 | ||
9babb374 BB |
2827 | if (!vd->vdev_aux) { |
2828 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2829 | pvd->vdev_expanding = B_FALSE; | |
2830 | } | |
2831 | ||
34dc7c2f BB |
2832 | if (newstate) |
2833 | *newstate = vd->vdev_state; | |
2834 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2835 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2836 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2837 | vd->vdev_parent->vdev_child[0] == vd) | |
2838 | vd->vdev_unspare = B_TRUE; | |
2839 | ||
9babb374 BB |
2840 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2841 | ||
2842 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2843 | if (vd->vdev_aux) | |
2844 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2845 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2846 | } | |
fb390aaf | 2847 | |
153b2285 YP |
2848 | if (wasoffline || |
2849 | (oldstate < VDEV_STATE_DEGRADED && | |
2850 | vd->vdev_state >= VDEV_STATE_DEGRADED)) | |
12fa0466 | 2851 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); |
fb390aaf | 2852 | |
fb5f0bc8 | 2853 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2854 | } |
2855 | ||
428870ff BB |
2856 | static int |
2857 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2858 | { |
9babb374 | 2859 | vdev_t *vd, *tvd; |
428870ff BB |
2860 | int error = 0; |
2861 | uint64_t generation; | |
2862 | metaslab_group_t *mg; | |
34dc7c2f | 2863 | |
428870ff BB |
2864 | top: |
2865 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2866 | |
b128c09f BB |
2867 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2868 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2869 | |
2870 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2871 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2872 | |
9babb374 | 2873 | tvd = vd->vdev_top; |
428870ff BB |
2874 | mg = tvd->vdev_mg; |
2875 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2876 | |
34dc7c2f BB |
2877 | /* |
2878 | * If the device isn't already offline, try to offline it. | |
2879 | */ | |
2880 | if (!vd->vdev_offline) { | |
2881 | /* | |
fb5f0bc8 | 2882 | * If this device has the only valid copy of some data, |
9babb374 BB |
2883 | * don't allow it to be offlined. Log devices are always |
2884 | * expendable. | |
34dc7c2f | 2885 | */ |
9babb374 BB |
2886 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2887 | vdev_dtl_required(vd)) | |
b128c09f | 2888 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2889 | |
428870ff BB |
2890 | /* |
2891 | * If the top-level is a slog and it has had allocations | |
2892 | * then proceed. We check that the vdev's metaslab group | |
2893 | * is not NULL since it's possible that we may have just | |
2894 | * added this vdev but not yet initialized its metaslabs. | |
2895 | */ | |
2896 | if (tvd->vdev_islog && mg != NULL) { | |
2897 | /* | |
2898 | * Prevent any future allocations. | |
2899 | */ | |
2900 | metaslab_group_passivate(mg); | |
2901 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2902 | ||
a1d477c2 | 2903 | error = spa_reset_logs(spa); |
428870ff BB |
2904 | |
2905 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2906 | ||
2907 | /* | |
2908 | * Check to see if the config has changed. | |
2909 | */ | |
2910 | if (error || generation != spa->spa_config_generation) { | |
2911 | metaslab_group_activate(mg); | |
2912 | if (error) | |
2913 | return (spa_vdev_state_exit(spa, | |
2914 | vd, error)); | |
2915 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2916 | goto top; | |
2917 | } | |
c99c9001 | 2918 | ASSERT0(tvd->vdev_stat.vs_alloc); |
428870ff BB |
2919 | } |
2920 | ||
34dc7c2f BB |
2921 | /* |
2922 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2923 | * If the top-level vdev is a log device then just offline |
2924 | * it. Otherwise, if this action results in the top-level | |
2925 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2926 | */ |
2927 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2928 | vdev_reopen(tvd); |
2929 | ||
2930 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2931 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2932 | vd->vdev_offline = B_FALSE; |
9babb374 | 2933 | vdev_reopen(tvd); |
b128c09f | 2934 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2935 | } |
428870ff BB |
2936 | |
2937 | /* | |
2938 | * Add the device back into the metaslab rotor so that | |
2939 | * once we online the device it's open for business. | |
2940 | */ | |
2941 | if (tvd->vdev_islog && mg != NULL) | |
2942 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2943 | } |
2944 | ||
b128c09f | 2945 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2946 | |
428870ff BB |
2947 | return (spa_vdev_state_exit(spa, vd, 0)); |
2948 | } | |
9babb374 | 2949 | |
428870ff BB |
2950 | int |
2951 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2952 | { | |
2953 | int error; | |
9babb374 | 2954 | |
428870ff BB |
2955 | mutex_enter(&spa->spa_vdev_top_lock); |
2956 | error = vdev_offline_locked(spa, guid, flags); | |
2957 | mutex_exit(&spa->spa_vdev_top_lock); | |
2958 | ||
2959 | return (error); | |
34dc7c2f BB |
2960 | } |
2961 | ||
2962 | /* | |
2963 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2964 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2965 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2966 | */ |
2967 | void | |
b128c09f | 2968 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2969 | { |
b128c09f BB |
2970 | vdev_t *rvd = spa->spa_root_vdev; |
2971 | ||
2972 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2973 | |
2974 | if (vd == NULL) | |
b128c09f | 2975 | vd = rvd; |
34dc7c2f BB |
2976 | |
2977 | vd->vdev_stat.vs_read_errors = 0; | |
2978 | vd->vdev_stat.vs_write_errors = 0; | |
2979 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2980 | |
1c27024e | 2981 | for (int c = 0; c < vd->vdev_children; c++) |
b128c09f | 2982 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f | 2983 | |
a1d477c2 MA |
2984 | /* |
2985 | * It makes no sense to "clear" an indirect vdev. | |
2986 | */ | |
2987 | if (!vdev_is_concrete(vd)) | |
2988 | return; | |
2989 | ||
34dc7c2f | 2990 | /* |
b128c09f BB |
2991 | * If we're in the FAULTED state or have experienced failed I/O, then |
2992 | * clear the persistent state and attempt to reopen the device. We | |
2993 | * also mark the vdev config dirty, so that the new faulted state is | |
2994 | * written out to disk. | |
34dc7c2f | 2995 | */ |
b128c09f BB |
2996 | if (vd->vdev_faulted || vd->vdev_degraded || |
2997 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
428870ff | 2998 | /* |
4e33ba4c | 2999 | * When reopening in response to a clear event, it may be due to |
428870ff BB |
3000 | * a fmadm repair request. In this case, if the device is |
3001 | * still broken, we want to still post the ereport again. | |
3002 | */ | |
3003 | vd->vdev_forcefault = B_TRUE; | |
3004 | ||
572e2857 | 3005 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
3006 | vd->vdev_cant_read = B_FALSE; |
3007 | vd->vdev_cant_write = B_FALSE; | |
4a283c7f | 3008 | vd->vdev_stat.vs_aux = 0; |
b128c09f | 3009 | |
572e2857 | 3010 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 3011 | |
428870ff BB |
3012 | vd->vdev_forcefault = B_FALSE; |
3013 | ||
572e2857 | 3014 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
3015 | vdev_state_dirty(vd->vdev_top); |
3016 | ||
3017 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
3018 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
3019 | ||
12fa0466 | 3020 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); |
34dc7c2f | 3021 | } |
428870ff BB |
3022 | |
3023 | /* | |
3024 | * When clearing a FMA-diagnosed fault, we always want to | |
3025 | * unspare the device, as we assume that the original spare was | |
3026 | * done in response to the FMA fault. | |
3027 | */ | |
3028 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
3029 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
3030 | vd->vdev_parent->vdev_child[0] == vd) | |
3031 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
3032 | } |
3033 | ||
b128c09f BB |
3034 | boolean_t |
3035 | vdev_is_dead(vdev_t *vd) | |
3036 | { | |
428870ff BB |
3037 | /* |
3038 | * Holes and missing devices are always considered "dead". | |
3039 | * This simplifies the code since we don't have to check for | |
3040 | * these types of devices in the various code paths. | |
3041 | * Instead we rely on the fact that we skip over dead devices | |
3042 | * before issuing I/O to them. | |
3043 | */ | |
a1d477c2 MA |
3044 | return (vd->vdev_state < VDEV_STATE_DEGRADED || |
3045 | vd->vdev_ops == &vdev_hole_ops || | |
428870ff | 3046 | vd->vdev_ops == &vdev_missing_ops); |
b128c09f BB |
3047 | } |
3048 | ||
3049 | boolean_t | |
34dc7c2f BB |
3050 | vdev_readable(vdev_t *vd) |
3051 | { | |
b128c09f | 3052 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
3053 | } |
3054 | ||
b128c09f | 3055 | boolean_t |
34dc7c2f BB |
3056 | vdev_writeable(vdev_t *vd) |
3057 | { | |
a1d477c2 MA |
3058 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write && |
3059 | vdev_is_concrete(vd)); | |
34dc7c2f BB |
3060 | } |
3061 | ||
b128c09f BB |
3062 | boolean_t |
3063 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 3064 | { |
fb5f0bc8 BB |
3065 | uint64_t state = vd->vdev_state; |
3066 | ||
b128c09f | 3067 | /* |
fb5f0bc8 | 3068 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
3069 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
3070 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
3071 | * the proper locks. Note that we have to get the vdev state |
3072 | * in a local variable because although it changes atomically, | |
3073 | * we're asking two separate questions about it. | |
b128c09f | 3074 | */ |
fb5f0bc8 | 3075 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
a1d477c2 | 3076 | !vd->vdev_cant_write && vdev_is_concrete(vd) && |
3dfb57a3 | 3077 | vd->vdev_mg->mg_initialized); |
34dc7c2f BB |
3078 | } |
3079 | ||
b128c09f BB |
3080 | boolean_t |
3081 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 3082 | { |
b128c09f | 3083 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 3084 | |
b128c09f BB |
3085 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
3086 | return (B_FALSE); | |
34dc7c2f | 3087 | |
b128c09f BB |
3088 | if (zio->io_type == ZIO_TYPE_READ) |
3089 | return (!vd->vdev_cant_read); | |
34dc7c2f | 3090 | |
b128c09f BB |
3091 | if (zio->io_type == ZIO_TYPE_WRITE) |
3092 | return (!vd->vdev_cant_write); | |
34dc7c2f | 3093 | |
b128c09f | 3094 | return (B_TRUE); |
34dc7c2f BB |
3095 | } |
3096 | ||
193a37cb TH |
3097 | static void |
3098 | vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) | |
34dc7c2f | 3099 | { |
193a37cb TH |
3100 | int t; |
3101 | for (t = 0; t < ZIO_TYPES; t++) { | |
3102 | vs->vs_ops[t] += cvs->vs_ops[t]; | |
3103 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
3104 | } | |
34dc7c2f | 3105 | |
193a37cb TH |
3106 | cvs->vs_scan_removing = cvd->vdev_removing; |
3107 | } | |
f3a7f661 | 3108 | |
193a37cb TH |
3109 | /* |
3110 | * Get extended stats | |
3111 | */ | |
3112 | static void | |
3113 | vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) | |
3114 | { | |
3115 | int t, b; | |
3116 | for (t = 0; t < ZIO_TYPES; t++) { | |
7e945072 | 3117 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) |
193a37cb | 3118 | vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; |
7e945072 TH |
3119 | |
3120 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { | |
193a37cb TH |
3121 | vsx->vsx_total_histo[t][b] += |
3122 | cvsx->vsx_total_histo[t][b]; | |
3123 | } | |
f38dfec3 | 3124 | } |
34dc7c2f | 3125 | |
193a37cb | 3126 | for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { |
7e945072 | 3127 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { |
193a37cb TH |
3128 | vsx->vsx_queue_histo[t][b] += |
3129 | cvsx->vsx_queue_histo[t][b]; | |
3130 | } | |
3131 | vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; | |
3132 | vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; | |
7e945072 TH |
3133 | |
3134 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) | |
3135 | vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; | |
3136 | ||
3137 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) | |
3138 | vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; | |
193a37cb | 3139 | } |
7e945072 | 3140 | |
193a37cb TH |
3141 | } |
3142 | ||
3143 | /* | |
3144 | * Get statistics for the given vdev. | |
3145 | */ | |
3146 | static void | |
3147 | vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
3148 | { | |
1c27024e | 3149 | int t; |
34dc7c2f BB |
3150 | /* |
3151 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
3152 | * over all top-level vdevs (i.e. the direct children of the root). | |
3153 | */ | |
193a37cb TH |
3154 | if (!vd->vdev_ops->vdev_op_leaf) { |
3155 | if (vs) { | |
3156 | memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); | |
3157 | memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); | |
3158 | } | |
3159 | if (vsx) | |
3160 | memset(vsx, 0, sizeof (*vsx)); | |
3161 | ||
1c27024e | 3162 | for (int c = 0; c < vd->vdev_children; c++) { |
193a37cb | 3163 | vdev_t *cvd = vd->vdev_child[c]; |
34dc7c2f | 3164 | vdev_stat_t *cvs = &cvd->vdev_stat; |
193a37cb TH |
3165 | vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; |
3166 | ||
3167 | vdev_get_stats_ex_impl(cvd, cvs, cvsx); | |
3168 | if (vs) | |
3169 | vdev_get_child_stat(cvd, vs, cvs); | |
3170 | if (vsx) | |
3171 | vdev_get_child_stat_ex(cvd, vsx, cvsx); | |
34dc7c2f | 3172 | |
193a37cb TH |
3173 | } |
3174 | } else { | |
3175 | /* | |
3176 | * We're a leaf. Just copy our ZIO active queue stats in. The | |
3177 | * other leaf stats are updated in vdev_stat_update(). | |
3178 | */ | |
3179 | if (!vsx) | |
3180 | return; | |
3181 | ||
3182 | memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); | |
3183 | ||
3184 | for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { | |
3185 | vsx->vsx_active_queue[t] = | |
3186 | vd->vdev_queue.vq_class[t].vqc_active; | |
3187 | vsx->vsx_pend_queue[t] = avl_numnodes( | |
3188 | &vd->vdev_queue.vq_class[t].vqc_queued_tree); | |
34dc7c2f BB |
3189 | } |
3190 | } | |
193a37cb TH |
3191 | } |
3192 | ||
3193 | void | |
3194 | vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
3195 | { | |
0f676dc2 | 3196 | vdev_t *tvd = vd->vdev_top; |
193a37cb TH |
3197 | mutex_enter(&vd->vdev_stat_lock); |
3198 | if (vs) { | |
3199 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
3200 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
3201 | vs->vs_state = vd->vdev_state; | |
3202 | vs->vs_rsize = vdev_get_min_asize(vd); | |
3203 | if (vd->vdev_ops->vdev_op_leaf) | |
3204 | vs->vs_rsize += VDEV_LABEL_START_SIZE + | |
3205 | VDEV_LABEL_END_SIZE; | |
0f676dc2 GM |
3206 | /* |
3207 | * Report expandable space on top-level, non-auxillary devices | |
3208 | * only. The expandable space is reported in terms of metaslab | |
3209 | * sized units since that determines how much space the pool | |
3210 | * can expand. | |
3211 | */ | |
3212 | if (vd->vdev_aux == NULL && tvd != NULL) { | |
3213 | vs->vs_esize = P2ALIGN( | |
3214 | vd->vdev_max_asize - vd->vdev_asize, | |
3215 | 1ULL << tvd->vdev_ms_shift); | |
3216 | } | |
193a37cb TH |
3217 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
3218 | if (vd->vdev_aux == NULL && vd == vd->vdev_top && | |
a1d477c2 | 3219 | vdev_is_concrete(vd)) { |
193a37cb TH |
3220 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; |
3221 | } | |
3222 | } | |
3223 | ||
3224 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_READER) != 0); | |
3225 | vdev_get_stats_ex_impl(vd, vs, vsx); | |
f3a7f661 | 3226 | mutex_exit(&vd->vdev_stat_lock); |
34dc7c2f BB |
3227 | } |
3228 | ||
193a37cb TH |
3229 | void |
3230 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
3231 | { | |
3232 | return (vdev_get_stats_ex(vd, vs, NULL)); | |
3233 | } | |
3234 | ||
34dc7c2f BB |
3235 | void |
3236 | vdev_clear_stats(vdev_t *vd) | |
3237 | { | |
3238 | mutex_enter(&vd->vdev_stat_lock); | |
3239 | vd->vdev_stat.vs_space = 0; | |
3240 | vd->vdev_stat.vs_dspace = 0; | |
3241 | vd->vdev_stat.vs_alloc = 0; | |
3242 | mutex_exit(&vd->vdev_stat_lock); | |
3243 | } | |
3244 | ||
428870ff BB |
3245 | void |
3246 | vdev_scan_stat_init(vdev_t *vd) | |
3247 | { | |
3248 | vdev_stat_t *vs = &vd->vdev_stat; | |
3249 | ||
1c27024e | 3250 | for (int c = 0; c < vd->vdev_children; c++) |
428870ff BB |
3251 | vdev_scan_stat_init(vd->vdev_child[c]); |
3252 | ||
3253 | mutex_enter(&vd->vdev_stat_lock); | |
3254 | vs->vs_scan_processed = 0; | |
3255 | mutex_exit(&vd->vdev_stat_lock); | |
3256 | } | |
3257 | ||
34dc7c2f | 3258 | void |
b128c09f | 3259 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 3260 | { |
fb5f0bc8 BB |
3261 | spa_t *spa = zio->io_spa; |
3262 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 3263 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
3264 | vdev_t *pvd; |
3265 | uint64_t txg = zio->io_txg; | |
3266 | vdev_stat_t *vs = &vd->vdev_stat; | |
193a37cb | 3267 | vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; |
34dc7c2f BB |
3268 | zio_type_t type = zio->io_type; |
3269 | int flags = zio->io_flags; | |
3270 | ||
b128c09f BB |
3271 | /* |
3272 | * If this i/o is a gang leader, it didn't do any actual work. | |
3273 | */ | |
3274 | if (zio->io_gang_tree) | |
3275 | return; | |
3276 | ||
34dc7c2f | 3277 | if (zio->io_error == 0) { |
b128c09f BB |
3278 | /* |
3279 | * If this is a root i/o, don't count it -- we've already | |
3280 | * counted the top-level vdevs, and vdev_get_stats() will | |
3281 | * aggregate them when asked. This reduces contention on | |
3282 | * the root vdev_stat_lock and implicitly handles blocks | |
3283 | * that compress away to holes, for which there is no i/o. | |
3284 | * (Holes never create vdev children, so all the counters | |
3285 | * remain zero, which is what we want.) | |
3286 | * | |
3287 | * Note: this only applies to successful i/o (io_error == 0) | |
3288 | * because unlike i/o counts, errors are not additive. | |
3289 | * When reading a ditto block, for example, failure of | |
3290 | * one top-level vdev does not imply a root-level error. | |
3291 | */ | |
3292 | if (vd == rvd) | |
3293 | return; | |
3294 | ||
3295 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
3296 | |
3297 | if (flags & ZIO_FLAG_IO_BYPASS) | |
3298 | return; | |
3299 | ||
3300 | mutex_enter(&vd->vdev_stat_lock); | |
3301 | ||
b128c09f | 3302 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 3303 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
3304 | dsl_scan_phys_t *scn_phys = |
3305 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
3306 | uint64_t *processed = &scn_phys->scn_processed; | |
3307 | ||
3308 | /* XXX cleanup? */ | |
3309 | if (vd->vdev_ops->vdev_op_leaf) | |
3310 | atomic_add_64(processed, psize); | |
3311 | vs->vs_scan_processed += psize; | |
3312 | } | |
3313 | ||
fb5f0bc8 | 3314 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 3315 | vs->vs_self_healed += psize; |
34dc7c2f | 3316 | } |
fb5f0bc8 | 3317 | |
193a37cb TH |
3318 | /* |
3319 | * The bytes/ops/histograms are recorded at the leaf level and | |
3320 | * aggregated into the higher level vdevs in vdev_get_stats(). | |
3321 | */ | |
4eb0db42 TH |
3322 | if (vd->vdev_ops->vdev_op_leaf && |
3323 | (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { | |
193a37cb TH |
3324 | |
3325 | vs->vs_ops[type]++; | |
3326 | vs->vs_bytes[type] += psize; | |
3327 | ||
7e945072 TH |
3328 | if (flags & ZIO_FLAG_DELEGATED) { |
3329 | vsx->vsx_agg_histo[zio->io_priority] | |
3330 | [RQ_HISTO(zio->io_size)]++; | |
3331 | } else { | |
3332 | vsx->vsx_ind_histo[zio->io_priority] | |
3333 | [RQ_HISTO(zio->io_size)]++; | |
3334 | } | |
3335 | ||
193a37cb TH |
3336 | if (zio->io_delta && zio->io_delay) { |
3337 | vsx->vsx_queue_histo[zio->io_priority] | |
7e945072 | 3338 | [L_HISTO(zio->io_delta - zio->io_delay)]++; |
193a37cb | 3339 | vsx->vsx_disk_histo[type] |
7e945072 | 3340 | [L_HISTO(zio->io_delay)]++; |
193a37cb | 3341 | vsx->vsx_total_histo[type] |
7e945072 | 3342 | [L_HISTO(zio->io_delta)]++; |
193a37cb TH |
3343 | } |
3344 | } | |
fb5f0bc8 BB |
3345 | |
3346 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
3347 | return; |
3348 | } | |
3349 | ||
3350 | if (flags & ZIO_FLAG_SPECULATIVE) | |
3351 | return; | |
3352 | ||
9babb374 BB |
3353 | /* |
3354 | * If this is an I/O error that is going to be retried, then ignore the | |
3355 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
3356 | * hard errors, when in reality they can happen for any number of | |
3357 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
3358 | */ | |
3359 | if (zio->io_error == EIO && | |
3360 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
3361 | return; | |
3362 | ||
428870ff BB |
3363 | /* |
3364 | * Intent logs writes won't propagate their error to the root | |
3365 | * I/O so don't mark these types of failures as pool-level | |
3366 | * errors. | |
3367 | */ | |
3368 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
3369 | return; | |
3370 | ||
b128c09f | 3371 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 3372 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
3373 | if (zio->io_error == ECKSUM) |
3374 | vs->vs_checksum_errors++; | |
3375 | else | |
3376 | vs->vs_read_errors++; | |
34dc7c2f | 3377 | } |
9babb374 | 3378 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
3379 | vs->vs_write_errors++; |
3380 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 3381 | |
a1d477c2 MA |
3382 | if (spa->spa_load_state == SPA_LOAD_NONE && |
3383 | type == ZIO_TYPE_WRITE && txg != 0 && | |
fb5f0bc8 | 3384 | (!(flags & ZIO_FLAG_IO_REPAIR) || |
572e2857 | 3385 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 3386 | spa->spa_claiming)) { |
fb5f0bc8 | 3387 | /* |
428870ff BB |
3388 | * This is either a normal write (not a repair), or it's |
3389 | * a repair induced by the scrub thread, or it's a repair | |
3390 | * made by zil_claim() during spa_load() in the first txg. | |
3391 | * In the normal case, we commit the DTL change in the same | |
3392 | * txg as the block was born. In the scrub-induced repair | |
3393 | * case, we know that scrubs run in first-pass syncing context, | |
3394 | * so we commit the DTL change in spa_syncing_txg(spa). | |
3395 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
3396 | * |
3397 | * We currently do not make DTL entries for failed spontaneous | |
3398 | * self-healing writes triggered by normal (non-scrubbing) | |
3399 | * reads, because we have no transactional context in which to | |
3400 | * do so -- and it's not clear that it'd be desirable anyway. | |
3401 | */ | |
3402 | if (vd->vdev_ops->vdev_op_leaf) { | |
3403 | uint64_t commit_txg = txg; | |
572e2857 | 3404 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
3405 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
3406 | ASSERT(spa_sync_pass(spa) == 1); | |
3407 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
3408 | commit_txg = spa_syncing_txg(spa); |
3409 | } else if (spa->spa_claiming) { | |
3410 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
3411 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 3412 | } |
428870ff | 3413 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 3414 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 3415 | return; |
fb5f0bc8 BB |
3416 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
3417 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
3418 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 3419 | } |
fb5f0bc8 BB |
3420 | if (vd != rvd) |
3421 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
3422 | } |
3423 | } | |
3424 | ||
34dc7c2f | 3425 | /* |
428870ff BB |
3426 | * Update the in-core space usage stats for this vdev, its metaslab class, |
3427 | * and the root vdev. | |
34dc7c2f BB |
3428 | */ |
3429 | void | |
428870ff BB |
3430 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
3431 | int64_t space_delta) | |
34dc7c2f BB |
3432 | { |
3433 | int64_t dspace_delta = space_delta; | |
3434 | spa_t *spa = vd->vdev_spa; | |
3435 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
3436 | metaslab_group_t *mg = vd->vdev_mg; |
3437 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
3438 | |
3439 | ASSERT(vd == vd->vdev_top); | |
3440 | ||
3441 | /* | |
3442 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
3443 | * factor. We must calculate this here and not at the root vdev | |
3444 | * because the root vdev's psize-to-asize is simply the max of its | |
3445 | * childrens', thus not accurate enough for us. | |
3446 | */ | |
3447 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 3448 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
3449 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
3450 | vd->vdev_deflate_ratio; | |
3451 | ||
3452 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 3453 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3454 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3455 | vd->vdev_stat.vs_dspace += dspace_delta; |
3456 | mutex_exit(&vd->vdev_stat_lock); | |
3457 | ||
428870ff | 3458 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 3459 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 3460 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3461 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3462 | rvd->vdev_stat.vs_dspace += dspace_delta; |
3463 | mutex_exit(&rvd->vdev_stat_lock); | |
3464 | } | |
428870ff BB |
3465 | |
3466 | if (mc != NULL) { | |
3467 | ASSERT(rvd == vd->vdev_parent); | |
3468 | ASSERT(vd->vdev_ms_count != 0); | |
3469 | ||
3470 | metaslab_class_space_update(mc, | |
3471 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
3472 | } | |
34dc7c2f BB |
3473 | } |
3474 | ||
3475 | /* | |
3476 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
3477 | * so that it will be written out next time the vdev configuration is synced. | |
3478 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
3479 | */ | |
3480 | void | |
3481 | vdev_config_dirty(vdev_t *vd) | |
3482 | { | |
3483 | spa_t *spa = vd->vdev_spa; | |
3484 | vdev_t *rvd = spa->spa_root_vdev; | |
3485 | int c; | |
3486 | ||
572e2857 BB |
3487 | ASSERT(spa_writeable(spa)); |
3488 | ||
34dc7c2f | 3489 | /* |
9babb374 BB |
3490 | * If this is an aux vdev (as with l2cache and spare devices), then we |
3491 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
3492 | */ |
3493 | if (vd->vdev_aux != NULL) { | |
3494 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
3495 | nvlist_t **aux; | |
3496 | uint_t naux; | |
3497 | ||
3498 | for (c = 0; c < sav->sav_count; c++) { | |
3499 | if (sav->sav_vdevs[c] == vd) | |
3500 | break; | |
3501 | } | |
3502 | ||
3503 | if (c == sav->sav_count) { | |
3504 | /* | |
3505 | * We're being removed. There's nothing more to do. | |
3506 | */ | |
3507 | ASSERT(sav->sav_sync == B_TRUE); | |
3508 | return; | |
3509 | } | |
3510 | ||
3511 | sav->sav_sync = B_TRUE; | |
3512 | ||
9babb374 BB |
3513 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
3514 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
3515 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
3516 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
3517 | } | |
b128c09f BB |
3518 | |
3519 | ASSERT(c < naux); | |
3520 | ||
3521 | /* | |
3522 | * Setting the nvlist in the middle if the array is a little | |
3523 | * sketchy, but it will work. | |
3524 | */ | |
3525 | nvlist_free(aux[c]); | |
428870ff | 3526 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
3527 | |
3528 | return; | |
3529 | } | |
3530 | ||
3531 | /* | |
3532 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
3533 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
3534 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
3535 | * so this is sufficient to ensure mutual exclusion. |
3536 | */ | |
b128c09f BB |
3537 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3538 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3539 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
3540 | |
3541 | if (vd == rvd) { | |
3542 | for (c = 0; c < rvd->vdev_children; c++) | |
3543 | vdev_config_dirty(rvd->vdev_child[c]); | |
3544 | } else { | |
3545 | ASSERT(vd == vd->vdev_top); | |
3546 | ||
428870ff | 3547 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
a1d477c2 | 3548 | vdev_is_concrete(vd)) { |
b128c09f | 3549 | list_insert_head(&spa->spa_config_dirty_list, vd); |
a1d477c2 | 3550 | } |
34dc7c2f BB |
3551 | } |
3552 | } | |
3553 | ||
3554 | void | |
3555 | vdev_config_clean(vdev_t *vd) | |
3556 | { | |
3557 | spa_t *spa = vd->vdev_spa; | |
3558 | ||
b128c09f BB |
3559 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3560 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3561 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 3562 | |
b128c09f BB |
3563 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
3564 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
3565 | } |
3566 | ||
b128c09f BB |
3567 | /* |
3568 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
3569 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
3570 | * the state changes from larger config changes because they require | |
3571 | * much less locking, and are often needed for administrative actions. | |
3572 | */ | |
3573 | void | |
3574 | vdev_state_dirty(vdev_t *vd) | |
3575 | { | |
3576 | spa_t *spa = vd->vdev_spa; | |
3577 | ||
572e2857 | 3578 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
3579 | ASSERT(vd == vd->vdev_top); |
3580 | ||
3581 | /* | |
3582 | * The state list is protected by the SCL_STATE lock. The caller | |
3583 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3584 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3585 | * so this is sufficient to ensure mutual exclusion. | |
3586 | */ | |
3587 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3588 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3589 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3590 | ||
a1d477c2 MA |
3591 | if (!list_link_active(&vd->vdev_state_dirty_node) && |
3592 | vdev_is_concrete(vd)) | |
b128c09f BB |
3593 | list_insert_head(&spa->spa_state_dirty_list, vd); |
3594 | } | |
3595 | ||
3596 | void | |
3597 | vdev_state_clean(vdev_t *vd) | |
3598 | { | |
3599 | spa_t *spa = vd->vdev_spa; | |
3600 | ||
3601 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3602 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3603 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3604 | ||
3605 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3606 | list_remove(&spa->spa_state_dirty_list, vd); | |
3607 | } | |
3608 | ||
3609 | /* | |
3610 | * Propagate vdev state up from children to parent. | |
3611 | */ | |
34dc7c2f BB |
3612 | void |
3613 | vdev_propagate_state(vdev_t *vd) | |
3614 | { | |
fb5f0bc8 BB |
3615 | spa_t *spa = vd->vdev_spa; |
3616 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
3617 | int degraded = 0, faulted = 0; |
3618 | int corrupted = 0; | |
34dc7c2f BB |
3619 | vdev_t *child; |
3620 | ||
3621 | if (vd->vdev_children > 0) { | |
1c27024e | 3622 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 3623 | child = vd->vdev_child[c]; |
b128c09f | 3624 | |
428870ff | 3625 | /* |
a1d477c2 MA |
3626 | * Don't factor holes or indirect vdevs into the |
3627 | * decision. | |
428870ff | 3628 | */ |
a1d477c2 | 3629 | if (!vdev_is_concrete(child)) |
428870ff BB |
3630 | continue; |
3631 | ||
b128c09f | 3632 | if (!vdev_readable(child) || |
fb5f0bc8 | 3633 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
3634 | /* |
3635 | * Root special: if there is a top-level log | |
3636 | * device, treat the root vdev as if it were | |
3637 | * degraded. | |
3638 | */ | |
3639 | if (child->vdev_islog && vd == rvd) | |
3640 | degraded++; | |
3641 | else | |
3642 | faulted++; | |
3643 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 3644 | degraded++; |
b128c09f | 3645 | } |
34dc7c2f BB |
3646 | |
3647 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3648 | corrupted++; | |
3649 | } | |
3650 | ||
3651 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3652 | ||
3653 | /* | |
b128c09f | 3654 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
3655 | * opened due to corrupted metadata, then propagate the root |
3656 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3657 | * replicas'. | |
3658 | */ | |
3659 | if (corrupted && vd == rvd && | |
3660 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3661 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3662 | VDEV_AUX_CORRUPT_DATA); | |
3663 | } | |
3664 | ||
b128c09f | 3665 | if (vd->vdev_parent) |
34dc7c2f BB |
3666 | vdev_propagate_state(vd->vdev_parent); |
3667 | } | |
3668 | ||
3669 | /* | |
3670 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3671 | * state, because we're in the process of opening children depth-first. | |
3672 | * Otherwise, we propagate the change to the parent. | |
3673 | * | |
3674 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3675 | * generated. | |
3676 | */ | |
3677 | void | |
3678 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3679 | { | |
3680 | uint64_t save_state; | |
b128c09f | 3681 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
3682 | |
3683 | if (state == vd->vdev_state) { | |
976246fa DB |
3684 | /* |
3685 | * Since vdev_offline() code path is already in an offline | |
3686 | * state we can miss a statechange event to OFFLINE. Check | |
3687 | * the previous state to catch this condition. | |
3688 | */ | |
3689 | if (vd->vdev_ops->vdev_op_leaf && | |
3690 | (state == VDEV_STATE_OFFLINE) && | |
3691 | (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { | |
3692 | /* post an offline state change */ | |
3693 | zfs_post_state_change(spa, vd, vd->vdev_prevstate); | |
3694 | } | |
34dc7c2f BB |
3695 | vd->vdev_stat.vs_aux = aux; |
3696 | return; | |
3697 | } | |
3698 | ||
3699 | save_state = vd->vdev_state; | |
3700 | ||
3701 | vd->vdev_state = state; | |
3702 | vd->vdev_stat.vs_aux = aux; | |
3703 | ||
3704 | /* | |
3705 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
3706 | * always close the underlying device unless the device has requested |
3707 | * a delayed close (i.e. we're about to remove or fault the device). | |
3708 | * Otherwise, we keep accessible but invalid devices open forever. | |
3709 | * We don't call vdev_close() itself, because that implies some extra | |
3710 | * checks (offline, etc) that we don't want here. This is limited to | |
3711 | * leaf devices, because otherwise closing the device will affect other | |
3712 | * children. | |
34dc7c2f | 3713 | */ |
428870ff BB |
3714 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
3715 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
3716 | vd->vdev_ops->vdev_op_close(vd); |
3717 | ||
3718 | if (vd->vdev_removed && | |
3719 | state == VDEV_STATE_CANT_OPEN && | |
3720 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3721 | /* | |
3722 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3723 | * device was previously marked removed and someone attempted to | |
3724 | * reopen it. If this failed due to a nonexistent device, then | |
3725 | * keep the device in the REMOVED state. We also let this be if | |
3726 | * it is one of our special test online cases, which is only | |
3727 | * attempting to online the device and shouldn't generate an FMA | |
3728 | * fault. | |
3729 | */ | |
3730 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3731 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3732 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3733 | vd->vdev_removed = B_TRUE; |
3734 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3735 | /* | |
572e2857 BB |
3736 | * If we fail to open a vdev during an import or recovery, we |
3737 | * mark it as "not available", which signifies that it was | |
3738 | * never there to begin with. Failure to open such a device | |
3739 | * is not considered an error. | |
34dc7c2f | 3740 | */ |
572e2857 BB |
3741 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3742 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3743 | vd->vdev_ops->vdev_op_leaf) |
3744 | vd->vdev_not_present = 1; | |
3745 | ||
3746 | /* | |
3747 | * Post the appropriate ereport. If the 'prevstate' field is | |
3748 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3749 | * that this is part of a vdev_reopen(). In this case, we don't | |
3750 | * want to post the ereport if the device was already in the | |
3751 | * CANT_OPEN state beforehand. | |
3752 | * | |
3753 | * If the 'checkremove' flag is set, then this is an attempt to | |
3754 | * online the device in response to an insertion event. If we | |
3755 | * hit this case, then we have detected an insertion event for a | |
3756 | * faulted or offline device that wasn't in the removed state. | |
3757 | * In this scenario, we don't post an ereport because we are | |
3758 | * about to replace the device, or attempt an online with | |
3759 | * vdev_forcefault, which will generate the fault for us. | |
3760 | */ | |
3761 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3762 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3763 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3764 | const char *class; |
3765 | ||
3766 | switch (aux) { | |
3767 | case VDEV_AUX_OPEN_FAILED: | |
3768 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3769 | break; | |
3770 | case VDEV_AUX_CORRUPT_DATA: | |
3771 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3772 | break; | |
3773 | case VDEV_AUX_NO_REPLICAS: | |
3774 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3775 | break; | |
3776 | case VDEV_AUX_BAD_GUID_SUM: | |
3777 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3778 | break; | |
3779 | case VDEV_AUX_TOO_SMALL: | |
3780 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3781 | break; | |
3782 | case VDEV_AUX_BAD_LABEL: | |
3783 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3784 | break; | |
ff61d1a4 | 3785 | case VDEV_AUX_BAD_ASHIFT: |
3786 | class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; | |
3787 | break; | |
34dc7c2f BB |
3788 | default: |
3789 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3790 | } | |
3791 | ||
b5256303 TC |
3792 | zfs_ereport_post(class, spa, vd, NULL, NULL, |
3793 | save_state, 0); | |
34dc7c2f BB |
3794 | } |
3795 | ||
3796 | /* Erase any notion of persistent removed state */ | |
3797 | vd->vdev_removed = B_FALSE; | |
3798 | } else { | |
3799 | vd->vdev_removed = B_FALSE; | |
3800 | } | |
3801 | ||
d02ca379 DB |
3802 | /* |
3803 | * Notify ZED of any significant state-change on a leaf vdev. | |
3804 | * | |
d02ca379 | 3805 | */ |
6078881a TH |
3806 | if (vd->vdev_ops->vdev_op_leaf) { |
3807 | /* preserve original state from a vdev_reopen() */ | |
3808 | if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && | |
3809 | (vd->vdev_prevstate != vd->vdev_state) && | |
3810 | (save_state <= VDEV_STATE_CLOSED)) | |
3811 | save_state = vd->vdev_prevstate; | |
3812 | ||
3813 | /* filter out state change due to initial vdev_open */ | |
3814 | if (save_state > VDEV_STATE_CLOSED) | |
3815 | zfs_post_state_change(spa, vd, save_state); | |
d02ca379 DB |
3816 | } |
3817 | ||
9babb374 BB |
3818 | if (!isopen && vd->vdev_parent) |
3819 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3820 | } |
b128c09f BB |
3821 | |
3822 | /* | |
3823 | * Check the vdev configuration to ensure that it's capable of supporting | |
e550644f | 3824 | * a root pool. We do not support partial configuration. |
b128c09f BB |
3825 | */ |
3826 | boolean_t | |
3827 | vdev_is_bootable(vdev_t *vd) | |
3828 | { | |
b128c09f | 3829 | if (!vd->vdev_ops->vdev_op_leaf) { |
e550644f | 3830 | const char *vdev_type = vd->vdev_ops->vdev_op_type; |
b128c09f | 3831 | |
a1d477c2 MA |
3832 | if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || |
3833 | strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { | |
b128c09f | 3834 | return (B_FALSE); |
a1d477c2 | 3835 | } |
b128c09f BB |
3836 | } |
3837 | ||
e550644f | 3838 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3839 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3840 | return (B_FALSE); | |
3841 | } | |
3842 | return (B_TRUE); | |
3843 | } | |
9babb374 | 3844 | |
a1d477c2 MA |
3845 | boolean_t |
3846 | vdev_is_concrete(vdev_t *vd) | |
3847 | { | |
3848 | vdev_ops_t *ops = vd->vdev_ops; | |
3849 | if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || | |
3850 | ops == &vdev_missing_ops || ops == &vdev_root_ops) { | |
3851 | return (B_FALSE); | |
3852 | } else { | |
3853 | return (B_TRUE); | |
3854 | } | |
3855 | } | |
3856 | ||
428870ff BB |
3857 | /* |
3858 | * Load the state from the original vdev tree (ovd) which | |
3859 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3860 | * vdev was offline or faulted then we transfer that state to the |
3861 | * device in the current vdev tree (nvd). | |
428870ff | 3862 | */ |
9babb374 | 3863 | void |
428870ff | 3864 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3865 | { |
572e2857 | 3866 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3867 | ASSERT(spa_config_held(nvd->vdev_spa, |
3868 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3869 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3870 | |
1c27024e | 3871 | for (int c = 0; c < nvd->vdev_children; c++) |
428870ff | 3872 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3873 | |
572e2857 | 3874 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3875 | /* |
572e2857 | 3876 | * Restore the persistent vdev state |
9babb374 | 3877 | */ |
428870ff | 3878 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3879 | nvd->vdev_faulted = ovd->vdev_faulted; |
3880 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3881 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3882 | } |
3883 | } | |
3884 | ||
572e2857 BB |
3885 | /* |
3886 | * Determine if a log device has valid content. If the vdev was | |
3887 | * removed or faulted in the MOS config then we know that | |
3888 | * the content on the log device has already been written to the pool. | |
3889 | */ | |
3890 | boolean_t | |
3891 | vdev_log_state_valid(vdev_t *vd) | |
3892 | { | |
3893 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && | |
3894 | !vd->vdev_removed) | |
3895 | return (B_TRUE); | |
3896 | ||
1c27024e | 3897 | for (int c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3898 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3899 | return (B_TRUE); | |
3900 | ||
3901 | return (B_FALSE); | |
3902 | } | |
3903 | ||
9babb374 BB |
3904 | /* |
3905 | * Expand a vdev if possible. | |
3906 | */ | |
3907 | void | |
3908 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3909 | { | |
3910 | ASSERT(vd->vdev_top == vd); | |
3911 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3912 | ||
a1d477c2 MA |
3913 | vdev_set_deflate_ratio(vd); |
3914 | ||
3915 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && | |
3916 | vdev_is_concrete(vd)) { | |
9babb374 BB |
3917 | VERIFY(vdev_metaslab_init(vd, txg) == 0); |
3918 | vdev_config_dirty(vd); | |
3919 | } | |
3920 | } | |
428870ff BB |
3921 | |
3922 | /* | |
3923 | * Split a vdev. | |
3924 | */ | |
3925 | void | |
3926 | vdev_split(vdev_t *vd) | |
3927 | { | |
3928 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3929 | ||
3930 | vdev_remove_child(pvd, vd); | |
3931 | vdev_compact_children(pvd); | |
3932 | ||
3933 | cvd = pvd->vdev_child[0]; | |
3934 | if (pvd->vdev_children == 1) { | |
3935 | vdev_remove_parent(cvd); | |
3936 | cvd->vdev_splitting = B_TRUE; | |
3937 | } | |
3938 | vdev_propagate_state(cvd); | |
3939 | } | |
c28b2279 | 3940 | |
cc92e9d0 | 3941 | void |
8fb1ede1 | 3942 | vdev_deadman(vdev_t *vd, char *tag) |
cc92e9d0 | 3943 | { |
1c27024e | 3944 | for (int c = 0; c < vd->vdev_children; c++) { |
cc92e9d0 GW |
3945 | vdev_t *cvd = vd->vdev_child[c]; |
3946 | ||
8fb1ede1 | 3947 | vdev_deadman(cvd, tag); |
cc92e9d0 GW |
3948 | } |
3949 | ||
3950 | if (vd->vdev_ops->vdev_op_leaf) { | |
3951 | vdev_queue_t *vq = &vd->vdev_queue; | |
3952 | ||
3953 | mutex_enter(&vq->vq_lock); | |
e8b96c60 | 3954 | if (avl_numnodes(&vq->vq_active_tree) > 0) { |
cc92e9d0 GW |
3955 | spa_t *spa = vd->vdev_spa; |
3956 | zio_t *fio; | |
3957 | uint64_t delta; | |
3958 | ||
8fb1ede1 BB |
3959 | zfs_dbgmsg("slow vdev: %s has %d active IOs", |
3960 | vd->vdev_path, avl_numnodes(&vq->vq_active_tree)); | |
3961 | ||
cc92e9d0 GW |
3962 | /* |
3963 | * Look at the head of all the pending queues, | |
3964 | * if any I/O has been outstanding for longer than | |
8fb1ede1 | 3965 | * the spa_deadman_synctime invoke the deadman logic. |
cc92e9d0 | 3966 | */ |
e8b96c60 | 3967 | fio = avl_first(&vq->vq_active_tree); |
cb682a17 | 3968 | delta = gethrtime() - fio->io_timestamp; |
8fb1ede1 BB |
3969 | if (delta > spa_deadman_synctime(spa)) |
3970 | zio_deadman(fio, tag); | |
cc92e9d0 GW |
3971 | } |
3972 | mutex_exit(&vq->vq_lock); | |
3973 | } | |
3974 | } | |
3975 | ||
c28b2279 BB |
3976 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3977 | EXPORT_SYMBOL(vdev_fault); | |
3978 | EXPORT_SYMBOL(vdev_degrade); | |
3979 | EXPORT_SYMBOL(vdev_online); | |
3980 | EXPORT_SYMBOL(vdev_offline); | |
3981 | EXPORT_SYMBOL(vdev_clear); | |
4ea3f864 | 3982 | /* BEGIN CSTYLED */ |
b8bcca18 MA |
3983 | module_param(metaslabs_per_vdev, int, 0644); |
3984 | MODULE_PARM_DESC(metaslabs_per_vdev, | |
3985 | "Divide added vdev into approximately (but no more than) this number " | |
3986 | "of metaslabs"); | |
80d52c39 TH |
3987 | |
3988 | module_param(zfs_delays_per_second, uint, 0644); | |
3989 | MODULE_PARM_DESC(zfs_delays_per_second, "Rate limit delay events to this many " | |
3990 | "IO delays per second"); | |
3991 | ||
3992 | module_param(zfs_checksums_per_second, uint, 0644); | |
3993 | MODULE_PARM_DESC(zfs_checksums_per_second, "Rate limit checksum events " | |
3994 | "to this many checksum errors per second (do not set below zed" | |
3995 | "threshold)."); | |
02638a30 TC |
3996 | |
3997 | module_param(zfs_scan_ignore_errors, int, 0644); | |
3998 | MODULE_PARM_DESC(zfs_scan_ignore_errors, | |
3999 | "Ignore errors during resilver/scrub"); | |
4ea3f864 | 4000 | /* END CSTYLED */ |
c28b2279 | 4001 | #endif |