]>
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> | |
34dc7c2f BB |
28 | */ |
29 | ||
34dc7c2f BB |
30 | #include <sys/zfs_context.h> |
31 | #include <sys/fm/fs/zfs.h> | |
32 | #include <sys/spa.h> | |
33 | #include <sys/spa_impl.h> | |
34 | #include <sys/dmu.h> | |
35 | #include <sys/dmu_tx.h> | |
36 | #include <sys/vdev_impl.h> | |
37 | #include <sys/uberblock_impl.h> | |
38 | #include <sys/metaslab.h> | |
39 | #include <sys/metaslab_impl.h> | |
40 | #include <sys/space_map.h> | |
93cf2076 | 41 | #include <sys/space_reftree.h> |
34dc7c2f BB |
42 | #include <sys/zio.h> |
43 | #include <sys/zap.h> | |
44 | #include <sys/fs/zfs.h> | |
b128c09f | 45 | #include <sys/arc.h> |
9babb374 | 46 | #include <sys/zil.h> |
428870ff | 47 | #include <sys/dsl_scan.h> |
a6255b7f | 48 | #include <sys/abd.h> |
6c285672 | 49 | #include <sys/zvol.h> |
6078881a | 50 | #include <sys/zfs_ratelimit.h> |
34dc7c2f | 51 | |
b8bcca18 MA |
52 | /* |
53 | * When a vdev is added, it will be divided into approximately (but no | |
54 | * more than) this number of metaslabs. | |
55 | */ | |
56 | int metaslabs_per_vdev = 200; | |
57 | ||
34dc7c2f BB |
58 | /* |
59 | * Virtual device management. | |
60 | */ | |
61 | ||
62 | static vdev_ops_t *vdev_ops_table[] = { | |
63 | &vdev_root_ops, | |
64 | &vdev_raidz_ops, | |
65 | &vdev_mirror_ops, | |
66 | &vdev_replacing_ops, | |
67 | &vdev_spare_ops, | |
68 | &vdev_disk_ops, | |
69 | &vdev_file_ops, | |
70 | &vdev_missing_ops, | |
428870ff | 71 | &vdev_hole_ops, |
34dc7c2f BB |
72 | NULL |
73 | }; | |
74 | ||
34dc7c2f BB |
75 | /* |
76 | * Given a vdev type, return the appropriate ops vector. | |
77 | */ | |
78 | static vdev_ops_t * | |
79 | vdev_getops(const char *type) | |
80 | { | |
81 | vdev_ops_t *ops, **opspp; | |
82 | ||
83 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
84 | if (strcmp(ops->vdev_op_type, type) == 0) | |
85 | break; | |
86 | ||
87 | return (ops); | |
88 | } | |
89 | ||
90 | /* | |
91 | * Default asize function: return the MAX of psize with the asize of | |
92 | * all children. This is what's used by anything other than RAID-Z. | |
93 | */ | |
94 | uint64_t | |
95 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
96 | { | |
97 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
98 | uint64_t csize; | |
d6320ddb | 99 | int c; |
34dc7c2f | 100 | |
d6320ddb | 101 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
102 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
103 | asize = MAX(asize, csize); | |
104 | } | |
105 | ||
106 | return (asize); | |
107 | } | |
108 | ||
109 | /* | |
9babb374 BB |
110 | * Get the minimum allocatable size. We define the allocatable size as |
111 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
112 | * replace or attach devices which don't have the same physical size but | |
113 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
114 | */ |
115 | uint64_t | |
9babb374 | 116 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 117 | { |
9babb374 | 118 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 119 | |
9babb374 | 120 | /* |
1bd201e7 | 121 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
122 | * just return our own asize. |
123 | */ | |
124 | if (pvd == NULL) | |
125 | return (vd->vdev_asize); | |
34dc7c2f BB |
126 | |
127 | /* | |
9babb374 BB |
128 | * The top-level vdev just returns the allocatable size rounded |
129 | * to the nearest metaslab. | |
34dc7c2f | 130 | */ |
9babb374 BB |
131 | if (vd == vd->vdev_top) |
132 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 133 | |
9babb374 BB |
134 | /* |
135 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
136 | * so each child must provide at least 1/Nth of its asize. | |
137 | */ | |
138 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
2e215fec SH |
139 | return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / |
140 | pvd->vdev_children); | |
34dc7c2f | 141 | |
9babb374 BB |
142 | return (pvd->vdev_min_asize); |
143 | } | |
144 | ||
145 | void | |
146 | vdev_set_min_asize(vdev_t *vd) | |
147 | { | |
d6320ddb | 148 | int c; |
9babb374 | 149 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 150 | |
d6320ddb | 151 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 152 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
153 | } |
154 | ||
155 | vdev_t * | |
156 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
157 | { | |
158 | vdev_t *rvd = spa->spa_root_vdev; | |
159 | ||
b128c09f | 160 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 161 | |
b128c09f BB |
162 | if (vdev < rvd->vdev_children) { |
163 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 164 | return (rvd->vdev_child[vdev]); |
b128c09f | 165 | } |
34dc7c2f BB |
166 | |
167 | return (NULL); | |
168 | } | |
169 | ||
170 | vdev_t * | |
171 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
172 | { | |
34dc7c2f | 173 | vdev_t *mvd; |
d6320ddb | 174 | int c; |
34dc7c2f BB |
175 | |
176 | if (vd->vdev_guid == guid) | |
177 | return (vd); | |
178 | ||
d6320ddb | 179 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
180 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
181 | NULL) | |
182 | return (mvd); | |
183 | ||
184 | return (NULL); | |
185 | } | |
186 | ||
9c43027b AJ |
187 | static int |
188 | vdev_count_leaves_impl(vdev_t *vd) | |
189 | { | |
190 | int n = 0; | |
191 | int c; | |
192 | ||
193 | if (vd->vdev_ops->vdev_op_leaf) | |
194 | return (1); | |
195 | ||
196 | for (c = 0; c < vd->vdev_children; c++) | |
197 | n += vdev_count_leaves_impl(vd->vdev_child[c]); | |
198 | ||
199 | return (n); | |
200 | } | |
201 | ||
202 | int | |
203 | vdev_count_leaves(spa_t *spa) | |
204 | { | |
205 | return (vdev_count_leaves_impl(spa->spa_root_vdev)); | |
206 | } | |
207 | ||
34dc7c2f BB |
208 | void |
209 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
210 | { | |
211 | size_t oldsize, newsize; | |
212 | uint64_t id = cvd->vdev_id; | |
213 | vdev_t **newchild; | |
214 | ||
44de2f02 | 215 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
216 | ASSERT(cvd->vdev_parent == NULL); |
217 | ||
218 | cvd->vdev_parent = pvd; | |
219 | ||
220 | if (pvd == NULL) | |
221 | return; | |
222 | ||
223 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
224 | ||
225 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
226 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
227 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
228 | ||
79c76d5b | 229 | newchild = kmem_alloc(newsize, KM_SLEEP); |
34dc7c2f BB |
230 | if (pvd->vdev_child != NULL) { |
231 | bcopy(pvd->vdev_child, newchild, oldsize); | |
232 | kmem_free(pvd->vdev_child, oldsize); | |
233 | } | |
234 | ||
235 | pvd->vdev_child = newchild; | |
236 | pvd->vdev_child[id] = cvd; | |
237 | ||
238 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
239 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
240 | ||
241 | /* | |
242 | * Walk up all ancestors to update guid sum. | |
243 | */ | |
244 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
245 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
246 | } |
247 | ||
248 | void | |
249 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
250 | { | |
251 | int c; | |
252 | uint_t id = cvd->vdev_id; | |
253 | ||
254 | ASSERT(cvd->vdev_parent == pvd); | |
255 | ||
256 | if (pvd == NULL) | |
257 | return; | |
258 | ||
259 | ASSERT(id < pvd->vdev_children); | |
260 | ASSERT(pvd->vdev_child[id] == cvd); | |
261 | ||
262 | pvd->vdev_child[id] = NULL; | |
263 | cvd->vdev_parent = NULL; | |
264 | ||
265 | for (c = 0; c < pvd->vdev_children; c++) | |
266 | if (pvd->vdev_child[c]) | |
267 | break; | |
268 | ||
269 | if (c == pvd->vdev_children) { | |
270 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
271 | pvd->vdev_child = NULL; | |
272 | pvd->vdev_children = 0; | |
273 | } | |
274 | ||
275 | /* | |
276 | * Walk up all ancestors to update guid sum. | |
277 | */ | |
278 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
279 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
280 | } |
281 | ||
282 | /* | |
283 | * Remove any holes in the child array. | |
284 | */ | |
285 | void | |
286 | vdev_compact_children(vdev_t *pvd) | |
287 | { | |
288 | vdev_t **newchild, *cvd; | |
289 | int oldc = pvd->vdev_children; | |
9babb374 | 290 | int newc; |
d6320ddb | 291 | int c; |
34dc7c2f | 292 | |
b128c09f | 293 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 294 | |
d6320ddb | 295 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
296 | if (pvd->vdev_child[c]) |
297 | newc++; | |
298 | ||
79c76d5b | 299 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); |
34dc7c2f | 300 | |
d6320ddb | 301 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
302 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
303 | newchild[newc] = cvd; | |
304 | cvd->vdev_id = newc++; | |
305 | } | |
306 | } | |
307 | ||
308 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
309 | pvd->vdev_child = newchild; | |
310 | pvd->vdev_children = newc; | |
311 | } | |
312 | ||
313 | /* | |
314 | * Allocate and minimally initialize a vdev_t. | |
315 | */ | |
428870ff | 316 | vdev_t * |
34dc7c2f BB |
317 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
318 | { | |
319 | vdev_t *vd; | |
d6320ddb | 320 | int t; |
34dc7c2f | 321 | |
79c76d5b | 322 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); |
34dc7c2f BB |
323 | |
324 | if (spa->spa_root_vdev == NULL) { | |
325 | ASSERT(ops == &vdev_root_ops); | |
326 | spa->spa_root_vdev = vd; | |
3541dc6d | 327 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
328 | } |
329 | ||
428870ff | 330 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
331 | if (spa->spa_root_vdev == vd) { |
332 | /* | |
333 | * The root vdev's guid will also be the pool guid, | |
334 | * which must be unique among all pools. | |
335 | */ | |
428870ff | 336 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
337 | } else { |
338 | /* | |
339 | * Any other vdev's guid must be unique within the pool. | |
340 | */ | |
428870ff | 341 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
342 | } |
343 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
344 | } | |
345 | ||
346 | vd->vdev_spa = spa; | |
347 | vd->vdev_id = id; | |
348 | vd->vdev_guid = guid; | |
349 | vd->vdev_guid_sum = guid; | |
350 | vd->vdev_ops = ops; | |
351 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 352 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f | 353 | |
6078881a TH |
354 | /* |
355 | * Initialize rate limit structs for events. We rate limit ZIO delay | |
356 | * and checksum events so that we don't overwhelm ZED with thousands | |
357 | * of events when a disk is acting up. | |
358 | */ | |
359 | zfs_ratelimit_init(&vd->vdev_delay_rl, DELAYS_PER_SECOND, 1); | |
360 | zfs_ratelimit_init(&vd->vdev_checksum_rl, CHECKSUMS_PER_SECOND, 1); | |
361 | ||
98f72a53 BB |
362 | list_link_init(&vd->vdev_config_dirty_node); |
363 | list_link_init(&vd->vdev_state_dirty_node); | |
448d7aaa | 364 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); |
34dc7c2f | 365 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); |
b128c09f | 366 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
3dfb57a3 | 367 | mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); |
6078881a | 368 | |
d6320ddb | 369 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 | 370 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL, |
fb5f0bc8 BB |
371 | &vd->vdev_dtl_lock); |
372 | } | |
34dc7c2f BB |
373 | txg_list_create(&vd->vdev_ms_list, |
374 | offsetof(struct metaslab, ms_txg_node)); | |
375 | txg_list_create(&vd->vdev_dtl_list, | |
376 | offsetof(struct vdev, vdev_dtl_node)); | |
377 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
378 | vdev_queue_init(vd); | |
379 | vdev_cache_init(vd); | |
380 | ||
381 | return (vd); | |
382 | } | |
383 | ||
384 | /* | |
385 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
386 | * creating a new vdev or loading an existing one - the behavior is slightly | |
387 | * different for each case. | |
388 | */ | |
389 | int | |
390 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
391 | int alloctype) | |
392 | { | |
393 | vdev_ops_t *ops; | |
394 | char *type; | |
395 | uint64_t guid = 0, islog, nparity; | |
396 | vdev_t *vd; | |
397 | ||
b128c09f | 398 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
399 | |
400 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
2e528b49 | 401 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
402 | |
403 | if ((ops = vdev_getops(type)) == NULL) | |
2e528b49 | 404 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
405 | |
406 | /* | |
407 | * If this is a load, get the vdev guid from the nvlist. | |
408 | * Otherwise, vdev_alloc_common() will generate one for us. | |
409 | */ | |
410 | if (alloctype == VDEV_ALLOC_LOAD) { | |
411 | uint64_t label_id; | |
412 | ||
413 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
414 | label_id != id) | |
2e528b49 | 415 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
416 | |
417 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 418 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
419 | } else if (alloctype == VDEV_ALLOC_SPARE) { |
420 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 421 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
422 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { |
423 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 424 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
425 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
426 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 427 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
428 | } |
429 | ||
430 | /* | |
431 | * The first allocated vdev must be of type 'root'. | |
432 | */ | |
433 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
2e528b49 | 434 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
435 | |
436 | /* | |
437 | * Determine whether we're a log vdev. | |
438 | */ | |
439 | islog = 0; | |
440 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
441 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
2e528b49 | 442 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 443 | |
428870ff | 444 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
2e528b49 | 445 | return (SET_ERROR(ENOTSUP)); |
428870ff | 446 | |
34dc7c2f BB |
447 | /* |
448 | * Set the nparity property for RAID-Z vdevs. | |
449 | */ | |
450 | nparity = -1ULL; | |
451 | if (ops == &vdev_raidz_ops) { | |
452 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
453 | &nparity) == 0) { | |
428870ff | 454 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
2e528b49 | 455 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 456 | /* |
45d1cae3 BB |
457 | * Previous versions could only support 1 or 2 parity |
458 | * device. | |
34dc7c2f | 459 | */ |
45d1cae3 BB |
460 | if (nparity > 1 && |
461 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
2e528b49 | 462 | return (SET_ERROR(ENOTSUP)); |
45d1cae3 BB |
463 | if (nparity > 2 && |
464 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
2e528b49 | 465 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
466 | } else { |
467 | /* | |
468 | * We require the parity to be specified for SPAs that | |
469 | * support multiple parity levels. | |
470 | */ | |
45d1cae3 | 471 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
2e528b49 | 472 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
473 | /* |
474 | * Otherwise, we default to 1 parity device for RAID-Z. | |
475 | */ | |
476 | nparity = 1; | |
477 | } | |
478 | } else { | |
479 | nparity = 0; | |
480 | } | |
481 | ASSERT(nparity != -1ULL); | |
482 | ||
483 | vd = vdev_alloc_common(spa, id, guid, ops); | |
484 | ||
485 | vd->vdev_islog = islog; | |
486 | vd->vdev_nparity = nparity; | |
487 | ||
488 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
489 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
490 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
491 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
492 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
493 | &vd->vdev_physpath) == 0) | |
494 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
1bbd8770 TH |
495 | |
496 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, | |
497 | &vd->vdev_enc_sysfs_path) == 0) | |
498 | vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); | |
499 | ||
9babb374 BB |
500 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
501 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
502 | |
503 | /* | |
504 | * Set the whole_disk property. If it's not specified, leave the value | |
505 | * as -1. | |
506 | */ | |
507 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
508 | &vd->vdev_wholedisk) != 0) | |
509 | vd->vdev_wholedisk = -1ULL; | |
510 | ||
511 | /* | |
512 | * Look for the 'not present' flag. This will only be set if the device | |
513 | * was not present at the time of import. | |
514 | */ | |
9babb374 BB |
515 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
516 | &vd->vdev_not_present); | |
34dc7c2f BB |
517 | |
518 | /* | |
519 | * Get the alignment requirement. | |
520 | */ | |
521 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
522 | ||
428870ff BB |
523 | /* |
524 | * Retrieve the vdev creation time. | |
525 | */ | |
526 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
527 | &vd->vdev_crtxg); | |
528 | ||
34dc7c2f BB |
529 | /* |
530 | * If we're a top-level vdev, try to load the allocation parameters. | |
531 | */ | |
428870ff BB |
532 | if (parent && !parent->vdev_parent && |
533 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
534 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
535 | &vd->vdev_ms_array); | |
536 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
537 | &vd->vdev_ms_shift); | |
538 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
539 | &vd->vdev_asize); | |
428870ff BB |
540 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
541 | &vd->vdev_removing); | |
e0ab3ab5 JS |
542 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, |
543 | &vd->vdev_top_zap); | |
544 | } else { | |
545 | ASSERT0(vd->vdev_top_zap); | |
428870ff BB |
546 | } |
547 | ||
5ffb9d1d | 548 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
549 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
550 | alloctype == VDEV_ALLOC_ADD || | |
551 | alloctype == VDEV_ALLOC_SPLIT || | |
552 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
553 | vd->vdev_mg = metaslab_group_create(islog ? | |
554 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
555 | } |
556 | ||
e0ab3ab5 JS |
557 | if (vd->vdev_ops->vdev_op_leaf && |
558 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
559 | (void) nvlist_lookup_uint64(nv, | |
560 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); | |
561 | } else { | |
562 | ASSERT0(vd->vdev_leaf_zap); | |
563 | } | |
564 | ||
34dc7c2f BB |
565 | /* |
566 | * If we're a leaf vdev, try to load the DTL object and other state. | |
567 | */ | |
e0ab3ab5 | 568 | |
b128c09f | 569 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
570 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
571 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
572 | if (alloctype == VDEV_ALLOC_LOAD) { |
573 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
93cf2076 | 574 | &vd->vdev_dtl_object); |
b128c09f BB |
575 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
576 | &vd->vdev_unspare); | |
577 | } | |
9babb374 BB |
578 | |
579 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
580 | uint64_t spare = 0; | |
581 | ||
582 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
583 | &spare) == 0 && spare) | |
584 | spa_spare_add(vd); | |
585 | } | |
586 | ||
34dc7c2f BB |
587 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
588 | &vd->vdev_offline); | |
b128c09f | 589 | |
5d1f7fb6 GW |
590 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, |
591 | &vd->vdev_resilver_txg); | |
572e2857 | 592 | |
34dc7c2f BB |
593 | /* |
594 | * When importing a pool, we want to ignore the persistent fault | |
595 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
596 | * valid in the current context. Local vdevs will |
597 | * remain in the faulted state. | |
34dc7c2f | 598 | */ |
428870ff | 599 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
600 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
601 | &vd->vdev_faulted); | |
602 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
603 | &vd->vdev_degraded); | |
604 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
605 | &vd->vdev_removed); | |
428870ff BB |
606 | |
607 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
608 | char *aux; | |
609 | ||
610 | vd->vdev_label_aux = | |
611 | VDEV_AUX_ERR_EXCEEDED; | |
612 | if (nvlist_lookup_string(nv, | |
613 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
614 | strcmp(aux, "external") == 0) | |
615 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
616 | } | |
34dc7c2f BB |
617 | } |
618 | } | |
619 | ||
620 | /* | |
621 | * Add ourselves to the parent's list of children. | |
622 | */ | |
623 | vdev_add_child(parent, vd); | |
624 | ||
625 | *vdp = vd; | |
626 | ||
627 | return (0); | |
628 | } | |
629 | ||
630 | void | |
631 | vdev_free(vdev_t *vd) | |
632 | { | |
d6320ddb | 633 | int c, t; |
34dc7c2f BB |
634 | spa_t *spa = vd->vdev_spa; |
635 | ||
636 | /* | |
637 | * vdev_free() implies closing the vdev first. This is simpler than | |
638 | * trying to ensure complicated semantics for all callers. | |
639 | */ | |
640 | vdev_close(vd); | |
641 | ||
b128c09f | 642 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 643 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
644 | |
645 | /* | |
646 | * Free all children. | |
647 | */ | |
d6320ddb | 648 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
649 | vdev_free(vd->vdev_child[c]); |
650 | ||
651 | ASSERT(vd->vdev_child == NULL); | |
652 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
653 | ||
654 | /* | |
655 | * Discard allocation state. | |
656 | */ | |
428870ff | 657 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 658 | vdev_metaslab_fini(vd); |
428870ff BB |
659 | metaslab_group_destroy(vd->vdev_mg); |
660 | } | |
34dc7c2f | 661 | |
c99c9001 MS |
662 | ASSERT0(vd->vdev_stat.vs_space); |
663 | ASSERT0(vd->vdev_stat.vs_dspace); | |
664 | ASSERT0(vd->vdev_stat.vs_alloc); | |
34dc7c2f BB |
665 | |
666 | /* | |
667 | * Remove this vdev from its parent's child list. | |
668 | */ | |
669 | vdev_remove_child(vd->vdev_parent, vd); | |
670 | ||
671 | ASSERT(vd->vdev_parent == NULL); | |
672 | ||
673 | /* | |
674 | * Clean up vdev structure. | |
675 | */ | |
676 | vdev_queue_fini(vd); | |
677 | vdev_cache_fini(vd); | |
678 | ||
679 | if (vd->vdev_path) | |
680 | spa_strfree(vd->vdev_path); | |
681 | if (vd->vdev_devid) | |
682 | spa_strfree(vd->vdev_devid); | |
683 | if (vd->vdev_physpath) | |
684 | spa_strfree(vd->vdev_physpath); | |
1bbd8770 TH |
685 | |
686 | if (vd->vdev_enc_sysfs_path) | |
687 | spa_strfree(vd->vdev_enc_sysfs_path); | |
688 | ||
9babb374 BB |
689 | if (vd->vdev_fru) |
690 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
691 | |
692 | if (vd->vdev_isspare) | |
693 | spa_spare_remove(vd); | |
694 | if (vd->vdev_isl2cache) | |
695 | spa_l2cache_remove(vd); | |
696 | ||
697 | txg_list_destroy(&vd->vdev_ms_list); | |
698 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 699 | |
34dc7c2f | 700 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 | 701 | space_map_close(vd->vdev_dtl_sm); |
d6320ddb | 702 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
703 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); |
704 | range_tree_destroy(vd->vdev_dtl[t]); | |
fb5f0bc8 | 705 | } |
34dc7c2f | 706 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 707 | |
3dfb57a3 | 708 | mutex_destroy(&vd->vdev_queue_lock); |
34dc7c2f BB |
709 | mutex_destroy(&vd->vdev_dtl_lock); |
710 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 711 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
712 | |
713 | if (vd == spa->spa_root_vdev) | |
714 | spa->spa_root_vdev = NULL; | |
715 | ||
716 | kmem_free(vd, sizeof (vdev_t)); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Transfer top-level vdev state from svd to tvd. | |
721 | */ | |
722 | static void | |
723 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
724 | { | |
725 | spa_t *spa = svd->vdev_spa; | |
726 | metaslab_t *msp; | |
727 | vdev_t *vd; | |
728 | int t; | |
729 | ||
730 | ASSERT(tvd == tvd->vdev_top); | |
731 | ||
77943bc1 | 732 | tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; |
34dc7c2f BB |
733 | tvd->vdev_ms_array = svd->vdev_ms_array; |
734 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
735 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
e0ab3ab5 | 736 | tvd->vdev_top_zap = svd->vdev_top_zap; |
34dc7c2f BB |
737 | |
738 | svd->vdev_ms_array = 0; | |
739 | svd->vdev_ms_shift = 0; | |
740 | svd->vdev_ms_count = 0; | |
e0ab3ab5 | 741 | svd->vdev_top_zap = 0; |
34dc7c2f | 742 | |
5ffb9d1d GW |
743 | if (tvd->vdev_mg) |
744 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
745 | tvd->vdev_mg = svd->vdev_mg; |
746 | tvd->vdev_ms = svd->vdev_ms; | |
747 | ||
748 | svd->vdev_mg = NULL; | |
749 | svd->vdev_ms = NULL; | |
750 | ||
751 | if (tvd->vdev_mg != NULL) | |
752 | tvd->vdev_mg->mg_vd = tvd; | |
753 | ||
754 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
755 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
756 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
757 | ||
758 | svd->vdev_stat.vs_alloc = 0; | |
759 | svd->vdev_stat.vs_space = 0; | |
760 | svd->vdev_stat.vs_dspace = 0; | |
761 | ||
762 | for (t = 0; t < TXG_SIZE; t++) { | |
763 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
764 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
765 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
766 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
767 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
768 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
769 | } | |
770 | ||
b128c09f | 771 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
772 | vdev_config_clean(svd); |
773 | vdev_config_dirty(tvd); | |
774 | } | |
775 | ||
b128c09f BB |
776 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
777 | vdev_state_clean(svd); | |
778 | vdev_state_dirty(tvd); | |
779 | } | |
780 | ||
34dc7c2f BB |
781 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
782 | svd->vdev_deflate_ratio = 0; | |
783 | ||
784 | tvd->vdev_islog = svd->vdev_islog; | |
785 | svd->vdev_islog = 0; | |
786 | } | |
787 | ||
788 | static void | |
789 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
790 | { | |
d6320ddb BB |
791 | int c; |
792 | ||
34dc7c2f BB |
793 | if (vd == NULL) |
794 | return; | |
795 | ||
796 | vd->vdev_top = tvd; | |
797 | ||
d6320ddb | 798 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
799 | vdev_top_update(tvd, vd->vdev_child[c]); |
800 | } | |
801 | ||
802 | /* | |
803 | * Add a mirror/replacing vdev above an existing vdev. | |
804 | */ | |
805 | vdev_t * | |
806 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
807 | { | |
808 | spa_t *spa = cvd->vdev_spa; | |
809 | vdev_t *pvd = cvd->vdev_parent; | |
810 | vdev_t *mvd; | |
811 | ||
b128c09f | 812 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
813 | |
814 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
815 | ||
816 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 817 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 818 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
34dc7c2f BB |
819 | mvd->vdev_ashift = cvd->vdev_ashift; |
820 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 821 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
822 | |
823 | vdev_remove_child(pvd, cvd); | |
824 | vdev_add_child(pvd, mvd); | |
825 | cvd->vdev_id = mvd->vdev_children; | |
826 | vdev_add_child(mvd, cvd); | |
827 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
828 | ||
829 | if (mvd == mvd->vdev_top) | |
830 | vdev_top_transfer(cvd, mvd); | |
831 | ||
832 | return (mvd); | |
833 | } | |
834 | ||
835 | /* | |
836 | * Remove a 1-way mirror/replacing vdev from the tree. | |
837 | */ | |
838 | void | |
839 | vdev_remove_parent(vdev_t *cvd) | |
840 | { | |
841 | vdev_t *mvd = cvd->vdev_parent; | |
842 | vdev_t *pvd = mvd->vdev_parent; | |
843 | ||
b128c09f | 844 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
845 | |
846 | ASSERT(mvd->vdev_children == 1); | |
847 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
848 | mvd->vdev_ops == &vdev_replacing_ops || | |
849 | mvd->vdev_ops == &vdev_spare_ops); | |
850 | cvd->vdev_ashift = mvd->vdev_ashift; | |
851 | ||
852 | vdev_remove_child(mvd, cvd); | |
853 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 854 | |
34dc7c2f | 855 | /* |
b128c09f BB |
856 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
857 | * Otherwise, we could have detached an offline device, and when we | |
858 | * go to import the pool we'll think we have two top-level vdevs, | |
859 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 860 | */ |
fb5f0bc8 BB |
861 | if (mvd->vdev_top == mvd) { |
862 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 863 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
864 | cvd->vdev_guid += guid_delta; |
865 | cvd->vdev_guid_sum += guid_delta; | |
61e99a73 AB |
866 | |
867 | /* | |
868 | * If pool not set for autoexpand, we need to also preserve | |
869 | * mvd's asize to prevent automatic expansion of cvd. | |
870 | * Otherwise if we are adjusting the mirror by attaching and | |
871 | * detaching children of non-uniform sizes, the mirror could | |
872 | * autoexpand, unexpectedly requiring larger devices to | |
873 | * re-establish the mirror. | |
874 | */ | |
875 | if (!cvd->vdev_spa->spa_autoexpand) | |
876 | cvd->vdev_asize = mvd->vdev_asize; | |
fb5f0bc8 | 877 | } |
b128c09f BB |
878 | cvd->vdev_id = mvd->vdev_id; |
879 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
880 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
881 | ||
882 | if (cvd == cvd->vdev_top) | |
883 | vdev_top_transfer(mvd, cvd); | |
884 | ||
885 | ASSERT(mvd->vdev_children == 0); | |
886 | vdev_free(mvd); | |
887 | } | |
888 | ||
889 | int | |
890 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
891 | { | |
892 | spa_t *spa = vd->vdev_spa; | |
893 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
894 | uint64_t m; |
895 | uint64_t oldc = vd->vdev_ms_count; | |
896 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
897 | metaslab_t **mspp; | |
898 | int error; | |
899 | ||
428870ff BB |
900 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
901 | ||
902 | /* | |
903 | * This vdev is not being allocated from yet or is a hole. | |
904 | */ | |
905 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
906 | return (0); |
907 | ||
428870ff BB |
908 | ASSERT(!vd->vdev_ishole); |
909 | ||
9babb374 BB |
910 | /* |
911 | * Compute the raidz-deflation ratio. Note, we hard-code | |
f1512ee6 MA |
912 | * in 128k (1 << 17) because it is the "typical" blocksize. |
913 | * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, | |
914 | * otherwise it would inconsistently account for existing bp's. | |
9babb374 BB |
915 | */ |
916 | vd->vdev_deflate_ratio = (1 << 17) / | |
917 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
918 | ||
34dc7c2f BB |
919 | ASSERT(oldc <= newc); |
920 | ||
bffb68a2 | 921 | mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
34dc7c2f BB |
922 | |
923 | if (oldc != 0) { | |
924 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
bffb68a2 | 925 | vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); |
34dc7c2f BB |
926 | } |
927 | ||
928 | vd->vdev_ms = mspp; | |
929 | vd->vdev_ms_count = newc; | |
930 | ||
931 | for (m = oldc; m < newc; m++) { | |
93cf2076 GW |
932 | uint64_t object = 0; |
933 | ||
34dc7c2f | 934 | if (txg == 0) { |
34dc7c2f | 935 | error = dmu_read(mos, vd->vdev_ms_array, |
9babb374 BB |
936 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
937 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
938 | if (error) |
939 | return (error); | |
34dc7c2f | 940 | } |
fb42a493 PS |
941 | |
942 | error = metaslab_init(vd->vdev_mg, m, object, txg, | |
943 | &(vd->vdev_ms[m])); | |
944 | if (error) | |
945 | return (error); | |
34dc7c2f BB |
946 | } |
947 | ||
428870ff BB |
948 | if (txg == 0) |
949 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
950 | ||
951 | /* | |
952 | * If the vdev is being removed we don't activate | |
953 | * the metaslabs since we want to ensure that no new | |
954 | * allocations are performed on this device. | |
955 | */ | |
956 | if (oldc == 0 && !vd->vdev_removing) | |
957 | metaslab_group_activate(vd->vdev_mg); | |
958 | ||
959 | if (txg == 0) | |
960 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
961 | ||
34dc7c2f BB |
962 | return (0); |
963 | } | |
964 | ||
965 | void | |
966 | vdev_metaslab_fini(vdev_t *vd) | |
967 | { | |
968 | uint64_t m; | |
969 | uint64_t count = vd->vdev_ms_count; | |
970 | ||
971 | if (vd->vdev_ms != NULL) { | |
428870ff | 972 | metaslab_group_passivate(vd->vdev_mg); |
93cf2076 GW |
973 | for (m = 0; m < count; m++) { |
974 | metaslab_t *msp = vd->vdev_ms[m]; | |
975 | ||
976 | if (msp != NULL) | |
977 | metaslab_fini(msp); | |
978 | } | |
bffb68a2 | 979 | vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); |
34dc7c2f BB |
980 | vd->vdev_ms = NULL; |
981 | } | |
920dd524 ED |
982 | |
983 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
34dc7c2f BB |
984 | } |
985 | ||
b128c09f BB |
986 | typedef struct vdev_probe_stats { |
987 | boolean_t vps_readable; | |
988 | boolean_t vps_writeable; | |
989 | int vps_flags; | |
b128c09f BB |
990 | } vdev_probe_stats_t; |
991 | ||
992 | static void | |
993 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 994 | { |
fb5f0bc8 | 995 | spa_t *spa = zio->io_spa; |
d164b209 | 996 | vdev_t *vd = zio->io_vd; |
b128c09f | 997 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
998 | |
999 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
1000 | |
1001 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
1002 | if (zio->io_error == 0) |
1003 | vps->vps_readable = 1; | |
fb5f0bc8 | 1004 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 1005 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
a6255b7f | 1006 | zio->io_offset, zio->io_size, zio->io_abd, |
b128c09f BB |
1007 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1008 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
1009 | } else { | |
a6255b7f | 1010 | abd_free(zio->io_abd); |
b128c09f BB |
1011 | } |
1012 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
1013 | if (zio->io_error == 0) |
1014 | vps->vps_writeable = 1; | |
a6255b7f | 1015 | abd_free(zio->io_abd); |
b128c09f | 1016 | } else if (zio->io_type == ZIO_TYPE_NULL) { |
d164b209 | 1017 | zio_t *pio; |
3dfb57a3 | 1018 | zio_link_t *zl; |
b128c09f BB |
1019 | |
1020 | vd->vdev_cant_read |= !vps->vps_readable; | |
1021 | vd->vdev_cant_write |= !vps->vps_writeable; | |
1022 | ||
1023 | if (vdev_readable(vd) && | |
fb5f0bc8 | 1024 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
1025 | zio->io_error = 0; |
1026 | } else { | |
1027 | ASSERT(zio->io_error != 0); | |
1028 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 1029 | spa, vd, NULL, 0, 0); |
2e528b49 | 1030 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f | 1031 | } |
d164b209 BB |
1032 | |
1033 | mutex_enter(&vd->vdev_probe_lock); | |
1034 | ASSERT(vd->vdev_probe_zio == zio); | |
1035 | vd->vdev_probe_zio = NULL; | |
1036 | mutex_exit(&vd->vdev_probe_lock); | |
1037 | ||
3dfb57a3 DB |
1038 | zl = NULL; |
1039 | while ((pio = zio_walk_parents(zio, &zl)) != NULL) | |
d164b209 | 1040 | if (!vdev_accessible(vd, pio)) |
2e528b49 | 1041 | pio->io_error = SET_ERROR(ENXIO); |
d164b209 | 1042 | |
b128c09f BB |
1043 | kmem_free(vps, sizeof (*vps)); |
1044 | } | |
1045 | } | |
34dc7c2f | 1046 | |
b128c09f | 1047 | /* |
d3cc8b15 WA |
1048 | * Determine whether this device is accessible. |
1049 | * | |
1050 | * Read and write to several known locations: the pad regions of each | |
1051 | * vdev label but the first, which we leave alone in case it contains | |
1052 | * a VTOC. | |
b128c09f BB |
1053 | */ |
1054 | zio_t * | |
d164b209 | 1055 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
1056 | { |
1057 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
1058 | vdev_probe_stats_t *vps = NULL; |
1059 | zio_t *pio; | |
d6320ddb | 1060 | int l; |
d164b209 BB |
1061 | |
1062 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 1063 | |
d164b209 BB |
1064 | /* |
1065 | * Don't probe the probe. | |
1066 | */ | |
1067 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
1068 | return (NULL); | |
b128c09f | 1069 | |
d164b209 BB |
1070 | /* |
1071 | * To prevent 'probe storms' when a device fails, we create | |
1072 | * just one probe i/o at a time. All zios that want to probe | |
1073 | * this vdev will become parents of the probe io. | |
1074 | */ | |
1075 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 1076 | |
d164b209 | 1077 | if ((pio = vd->vdev_probe_zio) == NULL) { |
79c76d5b | 1078 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); |
d164b209 BB |
1079 | |
1080 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1081 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1082 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1083 | |
1084 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1085 | /* | |
1086 | * vdev_cant_read and vdev_cant_write can only | |
1087 | * transition from TRUE to FALSE when we have the | |
1088 | * SCL_ZIO lock as writer; otherwise they can only | |
1089 | * transition from FALSE to TRUE. This ensures that | |
1090 | * any zio looking at these values can assume that | |
1091 | * failures persist for the life of the I/O. That's | |
1092 | * important because when a device has intermittent | |
1093 | * connectivity problems, we want to ensure that | |
1094 | * they're ascribed to the device (ENXIO) and not | |
1095 | * the zio (EIO). | |
1096 | * | |
1097 | * Since we hold SCL_ZIO as writer here, clear both | |
1098 | * values so the probe can reevaluate from first | |
1099 | * principles. | |
1100 | */ | |
1101 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1102 | vd->vdev_cant_read = B_FALSE; | |
1103 | vd->vdev_cant_write = B_FALSE; | |
1104 | } | |
1105 | ||
1106 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1107 | vdev_probe_done, vps, | |
1108 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1109 | ||
428870ff BB |
1110 | /* |
1111 | * We can't change the vdev state in this context, so we | |
1112 | * kick off an async task to do it on our behalf. | |
1113 | */ | |
d164b209 BB |
1114 | if (zio != NULL) { |
1115 | vd->vdev_probe_wanted = B_TRUE; | |
1116 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1117 | } | |
b128c09f BB |
1118 | } |
1119 | ||
d164b209 BB |
1120 | if (zio != NULL) |
1121 | zio_add_child(zio, pio); | |
b128c09f | 1122 | |
d164b209 | 1123 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1124 | |
d164b209 BB |
1125 | if (vps == NULL) { |
1126 | ASSERT(zio != NULL); | |
1127 | return (NULL); | |
1128 | } | |
b128c09f | 1129 | |
d6320ddb | 1130 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1131 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1132 | vdev_label_offset(vd->vdev_psize, l, |
a6255b7f DQ |
1133 | offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, |
1134 | abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), | |
b128c09f BB |
1135 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1136 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1137 | } | |
1138 | ||
d164b209 BB |
1139 | if (zio == NULL) |
1140 | return (pio); | |
1141 | ||
1142 | zio_nowait(pio); | |
1143 | return (NULL); | |
34dc7c2f BB |
1144 | } |
1145 | ||
45d1cae3 BB |
1146 | static void |
1147 | vdev_open_child(void *arg) | |
1148 | { | |
1149 | vdev_t *vd = arg; | |
1150 | ||
1151 | vd->vdev_open_thread = curthread; | |
1152 | vd->vdev_open_error = vdev_open(vd); | |
1153 | vd->vdev_open_thread = NULL; | |
1154 | } | |
1155 | ||
6c285672 | 1156 | static boolean_t |
428870ff BB |
1157 | vdev_uses_zvols(vdev_t *vd) |
1158 | { | |
d6320ddb BB |
1159 | int c; |
1160 | ||
6c285672 JL |
1161 | #ifdef _KERNEL |
1162 | if (zvol_is_zvol(vd->vdev_path)) | |
428870ff | 1163 | return (B_TRUE); |
6c285672 JL |
1164 | #endif |
1165 | ||
d6320ddb | 1166 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1167 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1168 | return (B_TRUE); | |
6c285672 | 1169 | |
428870ff BB |
1170 | return (B_FALSE); |
1171 | } | |
1172 | ||
45d1cae3 BB |
1173 | void |
1174 | vdev_open_children(vdev_t *vd) | |
1175 | { | |
1176 | taskq_t *tq; | |
1177 | int children = vd->vdev_children; | |
d6320ddb | 1178 | int c; |
45d1cae3 | 1179 | |
428870ff BB |
1180 | /* |
1181 | * in order to handle pools on top of zvols, do the opens | |
1182 | * in a single thread so that the same thread holds the | |
1183 | * spa_namespace_lock | |
1184 | */ | |
1185 | if (vdev_uses_zvols(vd)) { | |
13d9a004 | 1186 | retry_sync: |
4770aa06 | 1187 | for (c = 0; c < children; c++) |
428870ff BB |
1188 | vd->vdev_child[c]->vdev_open_error = |
1189 | vdev_open(vd->vdev_child[c]); | |
4770aa06 HJ |
1190 | } else { |
1191 | tq = taskq_create("vdev_open", children, minclsyspri, | |
1192 | children, children, TASKQ_PREPOPULATE); | |
13d9a004 BB |
1193 | if (tq == NULL) |
1194 | goto retry_sync; | |
45d1cae3 | 1195 | |
4770aa06 HJ |
1196 | for (c = 0; c < children; c++) |
1197 | VERIFY(taskq_dispatch(tq, vdev_open_child, | |
48d3eb40 | 1198 | vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); |
45d1cae3 | 1199 | |
4770aa06 HJ |
1200 | taskq_destroy(tq); |
1201 | } | |
1202 | ||
1203 | vd->vdev_nonrot = B_TRUE; | |
fb40095f RY |
1204 | |
1205 | for (c = 0; c < children; c++) | |
1206 | vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; | |
45d1cae3 BB |
1207 | } |
1208 | ||
34dc7c2f BB |
1209 | /* |
1210 | * Prepare a virtual device for access. | |
1211 | */ | |
1212 | int | |
1213 | vdev_open(vdev_t *vd) | |
1214 | { | |
fb5f0bc8 | 1215 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1216 | int error; |
34dc7c2f | 1217 | uint64_t osize = 0; |
1bd201e7 CS |
1218 | uint64_t max_osize = 0; |
1219 | uint64_t asize, max_asize, psize; | |
34dc7c2f | 1220 | uint64_t ashift = 0; |
d6320ddb | 1221 | int c; |
34dc7c2f | 1222 | |
45d1cae3 BB |
1223 | ASSERT(vd->vdev_open_thread == curthread || |
1224 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1225 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1226 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1227 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1228 | ||
34dc7c2f | 1229 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1230 | vd->vdev_cant_read = B_FALSE; |
1231 | vd->vdev_cant_write = B_FALSE; | |
1232 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1233 | |
428870ff BB |
1234 | /* |
1235 | * If this vdev is not removed, check its fault status. If it's | |
1236 | * faulted, bail out of the open. | |
1237 | */ | |
34dc7c2f BB |
1238 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1239 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1240 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1241 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1242 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1243 | vd->vdev_label_aux); |
2e528b49 | 1244 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1245 | } else if (vd->vdev_offline) { |
1246 | ASSERT(vd->vdev_children == 0); | |
1247 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
2e528b49 | 1248 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1249 | } |
1250 | ||
1bd201e7 | 1251 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1252 | |
428870ff BB |
1253 | /* |
1254 | * Reset the vdev_reopening flag so that we actually close | |
1255 | * the vdev on error. | |
1256 | */ | |
1257 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1258 | if (zio_injection_enabled && error == 0) |
9babb374 | 1259 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1260 | |
1261 | if (error) { | |
1262 | if (vd->vdev_removed && | |
1263 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1264 | vd->vdev_removed = B_FALSE; | |
1265 | ||
1266 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1267 | vd->vdev_stat.vs_aux); | |
1268 | return (error); | |
1269 | } | |
1270 | ||
1271 | vd->vdev_removed = B_FALSE; | |
1272 | ||
428870ff BB |
1273 | /* |
1274 | * Recheck the faulted flag now that we have confirmed that | |
1275 | * the vdev is accessible. If we're faulted, bail. | |
1276 | */ | |
1277 | if (vd->vdev_faulted) { | |
1278 | ASSERT(vd->vdev_children == 0); | |
1279 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1280 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1281 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1282 | vd->vdev_label_aux); | |
2e528b49 | 1283 | return (SET_ERROR(ENXIO)); |
428870ff BB |
1284 | } |
1285 | ||
34dc7c2f BB |
1286 | if (vd->vdev_degraded) { |
1287 | ASSERT(vd->vdev_children == 0); | |
1288 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1289 | VDEV_AUX_ERR_EXCEEDED); | |
1290 | } else { | |
428870ff | 1291 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1292 | } |
1293 | ||
428870ff BB |
1294 | /* |
1295 | * For hole or missing vdevs we just return success. | |
1296 | */ | |
1297 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1298 | return (0); | |
1299 | ||
d6320ddb | 1300 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1301 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1302 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1303 | VDEV_AUX_NONE); | |
1304 | break; | |
1305 | } | |
9babb374 | 1306 | } |
34dc7c2f BB |
1307 | |
1308 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1309 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1310 | |
1311 | if (vd->vdev_children == 0) { | |
1312 | if (osize < SPA_MINDEVSIZE) { | |
1313 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1314 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1315 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1316 | } |
1317 | psize = osize; | |
1318 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1319 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1320 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1321 | } else { |
1322 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1323 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1324 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1325 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1326 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1327 | } |
1328 | psize = 0; | |
1329 | asize = osize; | |
1bd201e7 | 1330 | max_asize = max_osize; |
34dc7c2f BB |
1331 | } |
1332 | ||
1333 | vd->vdev_psize = psize; | |
1334 | ||
9babb374 | 1335 | /* |
2e215fec | 1336 | * Make sure the allocatable size hasn't shrunk too much. |
9babb374 BB |
1337 | */ |
1338 | if (asize < vd->vdev_min_asize) { | |
1339 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1340 | VDEV_AUX_BAD_LABEL); | |
2e528b49 | 1341 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
1342 | } |
1343 | ||
34dc7c2f BB |
1344 | if (vd->vdev_asize == 0) { |
1345 | /* | |
1346 | * This is the first-ever open, so use the computed values. | |
b28e57cb | 1347 | * For compatibility, a different ashift can be requested. |
34dc7c2f BB |
1348 | */ |
1349 | vd->vdev_asize = asize; | |
1bd201e7 | 1350 | vd->vdev_max_asize = max_asize; |
ff61d1a4 | 1351 | if (vd->vdev_ashift == 0) { |
1352 | vd->vdev_ashift = ashift; /* use detected value */ | |
1353 | } | |
1354 | if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || | |
1355 | vd->vdev_ashift > ASHIFT_MAX)) { | |
1356 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1357 | VDEV_AUX_BAD_ASHIFT); | |
1358 | return (SET_ERROR(EDOM)); | |
1359 | } | |
34dc7c2f BB |
1360 | } else { |
1361 | /* | |
32a9872b GW |
1362 | * Detect if the alignment requirement has increased. |
1363 | * We don't want to make the pool unavailable, just | |
1364 | * post an event instead. | |
34dc7c2f | 1365 | */ |
32a9872b GW |
1366 | if (ashift > vd->vdev_top->vdev_ashift && |
1367 | vd->vdev_ops->vdev_op_leaf) { | |
1368 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
1369 | spa, vd, NULL, 0, 0); | |
34dc7c2f | 1370 | } |
32a9872b | 1371 | |
1bd201e7 | 1372 | vd->vdev_max_asize = max_asize; |
9babb374 | 1373 | } |
34dc7c2f | 1374 | |
9babb374 | 1375 | /* |
2e215fec SH |
1376 | * If all children are healthy we update asize if either: |
1377 | * The asize has increased, due to a device expansion caused by dynamic | |
1378 | * LUN growth or vdev replacement, and automatic expansion is enabled; | |
1379 | * making the additional space available. | |
1380 | * | |
1381 | * The asize has decreased, due to a device shrink usually caused by a | |
1382 | * vdev replace with a smaller device. This ensures that calculations | |
1383 | * based of max_asize and asize e.g. esize are always valid. It's safe | |
1384 | * to do this as we've already validated that asize is greater than | |
1385 | * vdev_min_asize. | |
9babb374 | 1386 | */ |
2e215fec SH |
1387 | if (vd->vdev_state == VDEV_STATE_HEALTHY && |
1388 | ((asize > vd->vdev_asize && | |
1389 | (vd->vdev_expanding || spa->spa_autoexpand)) || | |
1390 | (asize < vd->vdev_asize))) | |
9babb374 | 1391 | vd->vdev_asize = asize; |
34dc7c2f | 1392 | |
9babb374 | 1393 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1394 | |
1395 | /* | |
1396 | * Ensure we can issue some IO before declaring the | |
1397 | * vdev open for business. | |
1398 | */ | |
b128c09f BB |
1399 | if (vd->vdev_ops->vdev_op_leaf && |
1400 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1401 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1402 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1403 | return (error); |
1404 | } | |
1405 | ||
c3520e7f MA |
1406 | /* |
1407 | * Track the min and max ashift values for normal data devices. | |
1408 | */ | |
1409 | if (vd->vdev_top == vd && vd->vdev_ashift != 0 && | |
1410 | !vd->vdev_islog && vd->vdev_aux == NULL) { | |
1411 | if (vd->vdev_ashift > spa->spa_max_ashift) | |
1412 | spa->spa_max_ashift = vd->vdev_ashift; | |
1413 | if (vd->vdev_ashift < spa->spa_min_ashift) | |
1414 | spa->spa_min_ashift = vd->vdev_ashift; | |
1415 | } | |
1416 | ||
34dc7c2f | 1417 | /* |
b128c09f | 1418 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1419 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1420 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1421 | */ |
fb5f0bc8 BB |
1422 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1423 | vdev_resilver_needed(vd, NULL, NULL)) | |
1424 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1425 | |
1426 | return (0); | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * Called once the vdevs are all opened, this routine validates the label | |
1431 | * contents. This needs to be done before vdev_load() so that we don't | |
1432 | * inadvertently do repair I/Os to the wrong device. | |
1433 | * | |
c7f2d69d GW |
1434 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1435 | * if the machine crashed during a re-guid the new guid might have been written | |
1436 | * to all of the vdev labels, but not the cached config. The strict check | |
1437 | * will be performed when the pool is opened again using the mos config. | |
1438 | * | |
34dc7c2f BB |
1439 | * This function will only return failure if one of the vdevs indicates that it |
1440 | * has since been destroyed or exported. This is only possible if | |
1441 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1442 | * will be updated but the function will return 0. | |
1443 | */ | |
1444 | int | |
c7f2d69d | 1445 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1446 | { |
1447 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1448 | nvlist_t *label; |
428870ff | 1449 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1450 | uint64_t state; |
d6320ddb | 1451 | int c; |
34dc7c2f | 1452 | |
d6320ddb | 1453 | for (c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1454 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
2e528b49 | 1455 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
1456 | |
1457 | /* | |
1458 | * If the device has already failed, or was marked offline, don't do | |
1459 | * any further validation. Otherwise, label I/O will fail and we will | |
1460 | * overwrite the previous state. | |
1461 | */ | |
b128c09f | 1462 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1463 | uint64_t aux_guid = 0; |
1464 | nvlist_t *nvl; | |
295304be GW |
1465 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? |
1466 | spa_last_synced_txg(spa) : -1ULL; | |
34dc7c2f | 1467 | |
3bc7e0fb | 1468 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { |
e35c5a82 | 1469 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
34dc7c2f BB |
1470 | VDEV_AUX_BAD_LABEL); |
1471 | return (0); | |
1472 | } | |
1473 | ||
428870ff BB |
1474 | /* |
1475 | * Determine if this vdev has been split off into another | |
1476 | * pool. If so, then refuse to open it. | |
1477 | */ | |
1478 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1479 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1480 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1481 | VDEV_AUX_SPLIT_POOL); | |
1482 | nvlist_free(label); | |
1483 | return (0); | |
1484 | } | |
1485 | ||
c7f2d69d GW |
1486 | if (strict && (nvlist_lookup_uint64(label, |
1487 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1488 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1489 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1490 | VDEV_AUX_CORRUPT_DATA); | |
1491 | nvlist_free(label); | |
1492 | return (0); | |
1493 | } | |
1494 | ||
428870ff BB |
1495 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1496 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1497 | &aux_guid) != 0) | |
1498 | aux_guid = 0; | |
1499 | ||
b128c09f BB |
1500 | /* |
1501 | * If this vdev just became a top-level vdev because its | |
1502 | * sibling was detached, it will have adopted the parent's | |
1503 | * vdev guid -- but the label may or may not be on disk yet. | |
1504 | * Fortunately, either version of the label will have the | |
1505 | * same top guid, so if we're a top-level vdev, we can | |
1506 | * safely compare to that instead. | |
428870ff BB |
1507 | * |
1508 | * If we split this vdev off instead, then we also check the | |
1509 | * original pool's guid. We don't want to consider the vdev | |
1510 | * corrupt if it is partway through a split operation. | |
b128c09f | 1511 | */ |
34dc7c2f | 1512 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1513 | &guid) != 0 || |
1514 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1515 | &top_guid) != 0 || | |
428870ff | 1516 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1517 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1518 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1519 | VDEV_AUX_CORRUPT_DATA); | |
1520 | nvlist_free(label); | |
1521 | return (0); | |
1522 | } | |
1523 | ||
1524 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1525 | &state) != 0) { | |
1526 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1527 | VDEV_AUX_CORRUPT_DATA); | |
1528 | nvlist_free(label); | |
1529 | return (0); | |
1530 | } | |
1531 | ||
1532 | nvlist_free(label); | |
1533 | ||
45d1cae3 | 1534 | /* |
572e2857 | 1535 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1536 | * state of the pool. |
1537 | */ | |
572e2857 | 1538 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1539 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f | 1540 | state != POOL_STATE_ACTIVE) |
2e528b49 | 1541 | return (SET_ERROR(EBADF)); |
34dc7c2f | 1542 | |
b128c09f BB |
1543 | /* |
1544 | * If we were able to open and validate a vdev that was | |
1545 | * previously marked permanently unavailable, clear that state | |
1546 | * now. | |
1547 | */ | |
1548 | if (vd->vdev_not_present) | |
1549 | vd->vdev_not_present = 0; | |
1550 | } | |
34dc7c2f BB |
1551 | |
1552 | return (0); | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Close a virtual device. | |
1557 | */ | |
1558 | void | |
1559 | vdev_close(vdev_t *vd) | |
1560 | { | |
428870ff | 1561 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1562 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1563 | |
1564 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1565 | ||
428870ff BB |
1566 | /* |
1567 | * If our parent is reopening, then we are as well, unless we are | |
1568 | * going offline. | |
1569 | */ | |
1570 | if (pvd != NULL && pvd->vdev_reopening) | |
1571 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1572 | ||
34dc7c2f BB |
1573 | vd->vdev_ops->vdev_op_close(vd); |
1574 | ||
1575 | vdev_cache_purge(vd); | |
1576 | ||
1577 | /* | |
9babb374 | 1578 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1579 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1580 | * it's still faulted. | |
1581 | */ | |
1582 | vd->vdev_prevstate = vd->vdev_state; | |
1583 | ||
1584 | if (vd->vdev_offline) | |
1585 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1586 | else | |
1587 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1588 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1589 | } | |
1590 | ||
428870ff BB |
1591 | void |
1592 | vdev_hold(vdev_t *vd) | |
1593 | { | |
1594 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1595 | int c; |
428870ff BB |
1596 | |
1597 | ASSERT(spa_is_root(spa)); | |
1598 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1599 | return; | |
1600 | ||
d6320ddb | 1601 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1602 | vdev_hold(vd->vdev_child[c]); |
1603 | ||
1604 | if (vd->vdev_ops->vdev_op_leaf) | |
1605 | vd->vdev_ops->vdev_op_hold(vd); | |
1606 | } | |
1607 | ||
1608 | void | |
1609 | vdev_rele(vdev_t *vd) | |
1610 | { | |
d6320ddb | 1611 | int c; |
428870ff | 1612 | |
d6320ddb BB |
1613 | ASSERT(spa_is_root(vd->vdev_spa)); |
1614 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1615 | vdev_rele(vd->vdev_child[c]); |
1616 | ||
1617 | if (vd->vdev_ops->vdev_op_leaf) | |
1618 | vd->vdev_ops->vdev_op_rele(vd); | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1623 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1624 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1625 | * If the leaf has never been opened then open it, as usual. | |
1626 | */ | |
34dc7c2f BB |
1627 | void |
1628 | vdev_reopen(vdev_t *vd) | |
1629 | { | |
1630 | spa_t *spa = vd->vdev_spa; | |
1631 | ||
b128c09f | 1632 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1633 | |
428870ff BB |
1634 | /* set the reopening flag unless we're taking the vdev offline */ |
1635 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1636 | vdev_close(vd); |
1637 | (void) vdev_open(vd); | |
1638 | ||
1639 | /* | |
1640 | * Call vdev_validate() here to make sure we have the same device. | |
1641 | * Otherwise, a device with an invalid label could be successfully | |
1642 | * opened in response to vdev_reopen(). | |
1643 | */ | |
b128c09f BB |
1644 | if (vd->vdev_aux) { |
1645 | (void) vdev_validate_aux(vd); | |
1646 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1647 | vd->vdev_aux == &spa->spa_l2cache && |
1648 | !l2arc_vdev_present(vd)) | |
1649 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1650 | } else { |
295304be | 1651 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1652 | } |
34dc7c2f BB |
1653 | |
1654 | /* | |
1655 | * Reassess parent vdev's health. | |
1656 | */ | |
1657 | vdev_propagate_state(vd); | |
1658 | } | |
1659 | ||
1660 | int | |
1661 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1662 | { | |
1663 | int error; | |
1664 | ||
1665 | /* | |
1666 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1667 | * For a create, however, we want to fail the request if | |
1668 | * there are any components we can't open. | |
1669 | */ | |
1670 | error = vdev_open(vd); | |
1671 | ||
1672 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1673 | vdev_close(vd); | |
1674 | return (error ? error : ENXIO); | |
1675 | } | |
1676 | ||
1677 | /* | |
93cf2076 | 1678 | * Recursively load DTLs and initialize all labels. |
34dc7c2f | 1679 | */ |
93cf2076 GW |
1680 | if ((error = vdev_dtl_load(vd)) != 0 || |
1681 | (error = vdev_label_init(vd, txg, isreplacing ? | |
34dc7c2f BB |
1682 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { |
1683 | vdev_close(vd); | |
1684 | return (error); | |
1685 | } | |
1686 | ||
1687 | return (0); | |
1688 | } | |
1689 | ||
34dc7c2f | 1690 | void |
9babb374 | 1691 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1692 | { |
1693 | /* | |
b8bcca18 | 1694 | * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. |
34dc7c2f | 1695 | */ |
b8bcca18 | 1696 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); |
34dc7c2f | 1697 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); |
34dc7c2f BB |
1698 | } |
1699 | ||
1700 | void | |
1701 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1702 | { | |
1703 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1704 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1705 | ASSERT(ISP2(flags)); |
572e2857 | 1706 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1707 | |
1708 | if (flags & VDD_METASLAB) | |
1709 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1710 | ||
1711 | if (flags & VDD_DTL) | |
1712 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1713 | ||
1714 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1715 | } | |
1716 | ||
93cf2076 GW |
1717 | void |
1718 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1719 | { | |
1720 | int c; | |
1721 | ||
1722 | for (c = 0; c < vd->vdev_children; c++) | |
1723 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); | |
1724 | ||
1725 | if (vd->vdev_ops->vdev_op_leaf) | |
1726 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1727 | } | |
1728 | ||
fb5f0bc8 BB |
1729 | /* |
1730 | * DTLs. | |
1731 | * | |
1732 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1733 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1734 | * |
1735 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1736 | * | |
1737 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1738 | * | |
1739 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1740 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1741 | * txgs that was scrubbed. | |
1742 | * | |
1743 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1744 | * persistent errors or just some device being offline. | |
1745 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1746 | * maintained; it's only computed when needed, typically to | |
1747 | * determine whether a device can be detached. | |
1748 | * | |
1749 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1750 | * either has the data or it doesn't. | |
1751 | * | |
1752 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1753 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1754 | * if any child is less than fully replicated, then so is its parent. | |
1755 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1756 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1757 | * those are the txgs we don't have enough replication to read. For example, | |
1758 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1759 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1760 | * two child DTL_MISSING maps. | |
1761 | * | |
1762 | * It should be clear from the above that to compute the DTLs and outage maps | |
1763 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1764 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1765 | * a configuration change, we generate all other DTLs from first principles. | |
1766 | */ | |
34dc7c2f | 1767 | void |
fb5f0bc8 | 1768 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1769 | { |
93cf2076 | 1770 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1771 | |
1772 | ASSERT(t < DTL_TYPES); | |
1773 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1774 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1775 | |
93cf2076 GW |
1776 | mutex_enter(rt->rt_lock); |
1777 | if (!range_tree_contains(rt, txg, size)) | |
1778 | range_tree_add(rt, txg, size); | |
1779 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1780 | } |
1781 | ||
fb5f0bc8 BB |
1782 | boolean_t |
1783 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1784 | { |
93cf2076 | 1785 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 | 1786 | boolean_t dirty = B_FALSE; |
34dc7c2f | 1787 | |
fb5f0bc8 BB |
1788 | ASSERT(t < DTL_TYPES); |
1789 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f | 1790 | |
93cf2076 GW |
1791 | mutex_enter(rt->rt_lock); |
1792 | if (range_tree_space(rt) != 0) | |
1793 | dirty = range_tree_contains(rt, txg, size); | |
1794 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1795 | |
1796 | return (dirty); | |
1797 | } | |
1798 | ||
fb5f0bc8 BB |
1799 | boolean_t |
1800 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1801 | { | |
93cf2076 | 1802 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1803 | boolean_t empty; |
1804 | ||
93cf2076 GW |
1805 | mutex_enter(rt->rt_lock); |
1806 | empty = (range_tree_space(rt) == 0); | |
1807 | mutex_exit(rt->rt_lock); | |
fb5f0bc8 BB |
1808 | |
1809 | return (empty); | |
1810 | } | |
1811 | ||
5d1f7fb6 GW |
1812 | /* |
1813 | * Returns the lowest txg in the DTL range. | |
1814 | */ | |
1815 | static uint64_t | |
1816 | vdev_dtl_min(vdev_t *vd) | |
1817 | { | |
93cf2076 | 1818 | range_seg_t *rs; |
5d1f7fb6 GW |
1819 | |
1820 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1821 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1822 | ASSERT0(vd->vdev_children); |
1823 | ||
93cf2076 GW |
1824 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1825 | return (rs->rs_start - 1); | |
5d1f7fb6 GW |
1826 | } |
1827 | ||
1828 | /* | |
1829 | * Returns the highest txg in the DTL. | |
1830 | */ | |
1831 | static uint64_t | |
1832 | vdev_dtl_max(vdev_t *vd) | |
1833 | { | |
93cf2076 | 1834 | range_seg_t *rs; |
5d1f7fb6 GW |
1835 | |
1836 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1837 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1838 | ASSERT0(vd->vdev_children); |
1839 | ||
93cf2076 GW |
1840 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1841 | return (rs->rs_end); | |
5d1f7fb6 GW |
1842 | } |
1843 | ||
1844 | /* | |
1845 | * Determine if a resilvering vdev should remove any DTL entries from | |
1846 | * its range. If the vdev was resilvering for the entire duration of the | |
1847 | * scan then it should excise that range from its DTLs. Otherwise, this | |
1848 | * vdev is considered partially resilvered and should leave its DTL | |
1849 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
1850 | * excise the DTLs. | |
1851 | */ | |
1852 | static boolean_t | |
1853 | vdev_dtl_should_excise(vdev_t *vd) | |
1854 | { | |
1855 | spa_t *spa = vd->vdev_spa; | |
1856 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1857 | ||
1858 | ASSERT0(scn->scn_phys.scn_errors); | |
1859 | ASSERT0(vd->vdev_children); | |
1860 | ||
1861 | if (vd->vdev_resilver_txg == 0 || | |
93cf2076 | 1862 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) |
5d1f7fb6 GW |
1863 | return (B_TRUE); |
1864 | ||
1865 | /* | |
1866 | * When a resilver is initiated the scan will assign the scn_max_txg | |
1867 | * value to the highest txg value that exists in all DTLs. If this | |
1868 | * device's max DTL is not part of this scan (i.e. it is not in | |
1869 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
1870 | * for excision. | |
1871 | */ | |
1872 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
1873 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
1874 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
1875 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
1876 | return (B_TRUE); | |
1877 | } | |
1878 | return (B_FALSE); | |
1879 | } | |
1880 | ||
34dc7c2f BB |
1881 | /* |
1882 | * Reassess DTLs after a config change or scrub completion. | |
1883 | */ | |
1884 | void | |
1885 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1886 | { | |
1887 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1888 | avl_tree_t reftree; |
d6320ddb | 1889 | int c, t, minref; |
34dc7c2f | 1890 | |
fb5f0bc8 | 1891 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1892 | |
d6320ddb | 1893 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1894 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1895 | scrub_txg, scrub_done); | |
1896 | ||
428870ff | 1897 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1898 | return; |
1899 | ||
1900 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1901 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1902 | ||
34dc7c2f | 1903 | mutex_enter(&vd->vdev_dtl_lock); |
5d1f7fb6 GW |
1904 | |
1905 | /* | |
1906 | * If we've completed a scan cleanly then determine | |
1907 | * if this vdev should remove any DTLs. We only want to | |
1908 | * excise regions on vdevs that were available during | |
1909 | * the entire duration of this scan. | |
1910 | */ | |
b128c09f | 1911 | if (scrub_txg != 0 && |
428870ff | 1912 | (spa->spa_scrub_started || |
5d1f7fb6 GW |
1913 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && |
1914 | vdev_dtl_should_excise(vd)) { | |
b128c09f BB |
1915 | /* |
1916 | * We completed a scrub up to scrub_txg. If we | |
1917 | * did it without rebooting, then the scrub dtl | |
1918 | * will be valid, so excise the old region and | |
1919 | * fold in the scrub dtl. Otherwise, leave the | |
1920 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1921 | * |
1922 | * There's little trick here: to excise the beginning | |
1923 | * of the DTL_MISSING map, we put it into a reference | |
1924 | * tree and then add a segment with refcnt -1 that | |
1925 | * covers the range [0, scrub_txg). This means | |
1926 | * that each txg in that range has refcnt -1 or 0. | |
1927 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1928 | * entries in the range [0, scrub_txg) will have a | |
1929 | * positive refcnt -- either 1 or 2. We then convert | |
1930 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1931 | */ |
93cf2076 GW |
1932 | space_reftree_create(&reftree); |
1933 | space_reftree_add_map(&reftree, | |
1934 | vd->vdev_dtl[DTL_MISSING], 1); | |
1935 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
1936 | space_reftree_add_map(&reftree, | |
1937 | vd->vdev_dtl[DTL_SCRUB], 2); | |
1938 | space_reftree_generate_map(&reftree, | |
1939 | vd->vdev_dtl[DTL_MISSING], 1); | |
1940 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1941 | } |
93cf2076 GW |
1942 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1943 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1944 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1945 | if (scrub_done) |
93cf2076 GW |
1946 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1947 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
fb5f0bc8 | 1948 | if (!vdev_readable(vd)) |
93cf2076 | 1949 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); |
fb5f0bc8 | 1950 | else |
93cf2076 GW |
1951 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], |
1952 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
5d1f7fb6 GW |
1953 | |
1954 | /* | |
1955 | * If the vdev was resilvering and no longer has any | |
d14fa5db | 1956 | * DTLs then reset its resilvering flag and dirty |
1957 | * the top level so that we persist the change. | |
5d1f7fb6 GW |
1958 | */ |
1959 | if (vd->vdev_resilver_txg != 0 && | |
93cf2076 | 1960 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && |
d14fa5db | 1961 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) { |
5d1f7fb6 | 1962 | vd->vdev_resilver_txg = 0; |
d14fa5db | 1963 | vdev_config_dirty(vd->vdev_top); |
1964 | } | |
5d1f7fb6 | 1965 | |
34dc7c2f | 1966 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1967 | |
34dc7c2f BB |
1968 | if (txg != 0) |
1969 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1970 | return; | |
1971 | } | |
1972 | ||
34dc7c2f | 1973 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1974 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
1975 | int c; |
1976 | ||
428870ff BB |
1977 | /* account for child's outage in parent's missing map */ |
1978 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1979 | if (t == DTL_SCRUB) |
1980 | continue; /* leaf vdevs only */ | |
1981 | if (t == DTL_PARTIAL) | |
1982 | minref = 1; /* i.e. non-zero */ | |
1983 | else if (vd->vdev_nparity != 0) | |
1984 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1985 | else | |
1986 | minref = vd->vdev_children; /* any kind of mirror */ | |
93cf2076 | 1987 | space_reftree_create(&reftree); |
d6320ddb | 1988 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1989 | vdev_t *cvd = vd->vdev_child[c]; |
1990 | mutex_enter(&cvd->vdev_dtl_lock); | |
93cf2076 | 1991 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1992 | mutex_exit(&cvd->vdev_dtl_lock); |
1993 | } | |
93cf2076 GW |
1994 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); |
1995 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1996 | } |
fb5f0bc8 | 1997 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1998 | } |
1999 | ||
93cf2076 | 2000 | int |
34dc7c2f BB |
2001 | vdev_dtl_load(vdev_t *vd) |
2002 | { | |
2003 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 2004 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
2005 | int error = 0; |
2006 | int c; | |
34dc7c2f | 2007 | |
93cf2076 GW |
2008 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { |
2009 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2010 | |
93cf2076 GW |
2011 | error = space_map_open(&vd->vdev_dtl_sm, mos, |
2012 | vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); | |
2013 | if (error) | |
2014 | return (error); | |
2015 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f | 2016 | |
93cf2076 | 2017 | mutex_enter(&vd->vdev_dtl_lock); |
428870ff | 2018 | |
93cf2076 GW |
2019 | /* |
2020 | * Now that we've opened the space_map we need to update | |
2021 | * the in-core DTL. | |
2022 | */ | |
2023 | space_map_update(vd->vdev_dtl_sm); | |
34dc7c2f | 2024 | |
93cf2076 GW |
2025 | error = space_map_load(vd->vdev_dtl_sm, |
2026 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
2027 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 2028 | |
93cf2076 GW |
2029 | return (error); |
2030 | } | |
2031 | ||
2032 | for (c = 0; c < vd->vdev_children; c++) { | |
2033 | error = vdev_dtl_load(vd->vdev_child[c]); | |
2034 | if (error != 0) | |
2035 | break; | |
2036 | } | |
34dc7c2f BB |
2037 | |
2038 | return (error); | |
2039 | } | |
2040 | ||
e0ab3ab5 JS |
2041 | void |
2042 | vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) | |
2043 | { | |
2044 | spa_t *spa = vd->vdev_spa; | |
2045 | ||
2046 | VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); | |
2047 | VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2048 | zapobj, tx)); | |
2049 | } | |
2050 | ||
2051 | uint64_t | |
2052 | vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) | |
2053 | { | |
2054 | spa_t *spa = vd->vdev_spa; | |
2055 | uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, | |
2056 | DMU_OT_NONE, 0, tx); | |
2057 | ||
2058 | ASSERT(zap != 0); | |
2059 | VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2060 | zap, tx)); | |
2061 | ||
2062 | return (zap); | |
2063 | } | |
2064 | ||
2065 | void | |
2066 | vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) | |
2067 | { | |
2068 | uint64_t i; | |
2069 | ||
2070 | if (vd->vdev_ops != &vdev_hole_ops && | |
2071 | vd->vdev_ops != &vdev_missing_ops && | |
2072 | vd->vdev_ops != &vdev_root_ops && | |
2073 | !vd->vdev_top->vdev_removing) { | |
2074 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { | |
2075 | vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); | |
2076 | } | |
2077 | if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { | |
2078 | vd->vdev_top_zap = vdev_create_link_zap(vd, tx); | |
2079 | } | |
2080 | } | |
2081 | for (i = 0; i < vd->vdev_children; i++) { | |
2082 | vdev_construct_zaps(vd->vdev_child[i], tx); | |
2083 | } | |
2084 | } | |
2085 | ||
34dc7c2f BB |
2086 | void |
2087 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
2088 | { | |
2089 | spa_t *spa = vd->vdev_spa; | |
93cf2076 | 2090 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; |
34dc7c2f | 2091 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
2092 | range_tree_t *rtsync; |
2093 | kmutex_t rtlock; | |
34dc7c2f | 2094 | dmu_tx_t *tx; |
93cf2076 | 2095 | uint64_t object = space_map_object(vd->vdev_dtl_sm); |
34dc7c2f | 2096 | |
428870ff | 2097 | ASSERT(!vd->vdev_ishole); |
93cf2076 | 2098 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
428870ff | 2099 | |
34dc7c2f BB |
2100 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
2101 | ||
93cf2076 GW |
2102 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { |
2103 | mutex_enter(&vd->vdev_dtl_lock); | |
2104 | space_map_free(vd->vdev_dtl_sm, tx); | |
2105 | space_map_close(vd->vdev_dtl_sm); | |
2106 | vd->vdev_dtl_sm = NULL; | |
2107 | mutex_exit(&vd->vdev_dtl_lock); | |
e0ab3ab5 JS |
2108 | |
2109 | /* | |
2110 | * We only destroy the leaf ZAP for detached leaves or for | |
2111 | * removed log devices. Removed data devices handle leaf ZAP | |
2112 | * cleanup later, once cancellation is no longer possible. | |
2113 | */ | |
2114 | if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || | |
2115 | vd->vdev_top->vdev_islog)) { | |
2116 | vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); | |
2117 | vd->vdev_leaf_zap = 0; | |
2118 | } | |
2119 | ||
34dc7c2f | 2120 | dmu_tx_commit(tx); |
34dc7c2f BB |
2121 | return; |
2122 | } | |
2123 | ||
93cf2076 GW |
2124 | if (vd->vdev_dtl_sm == NULL) { |
2125 | uint64_t new_object; | |
2126 | ||
2127 | new_object = space_map_alloc(mos, tx); | |
2128 | VERIFY3U(new_object, !=, 0); | |
2129 | ||
2130 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
2131 | 0, -1ULL, 0, &vd->vdev_dtl_lock)); | |
2132 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f BB |
2133 | } |
2134 | ||
93cf2076 | 2135 | mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 2136 | |
93cf2076 | 2137 | rtsync = range_tree_create(NULL, NULL, &rtlock); |
34dc7c2f | 2138 | |
93cf2076 | 2139 | mutex_enter(&rtlock); |
34dc7c2f BB |
2140 | |
2141 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2142 | range_tree_walk(rt, range_tree_add, rtsync); |
34dc7c2f BB |
2143 | mutex_exit(&vd->vdev_dtl_lock); |
2144 | ||
93cf2076 GW |
2145 | space_map_truncate(vd->vdev_dtl_sm, tx); |
2146 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
2147 | range_tree_vacate(rtsync, NULL, NULL); | |
34dc7c2f | 2148 | |
93cf2076 | 2149 | range_tree_destroy(rtsync); |
34dc7c2f | 2150 | |
93cf2076 GW |
2151 | mutex_exit(&rtlock); |
2152 | mutex_destroy(&rtlock); | |
34dc7c2f | 2153 | |
93cf2076 GW |
2154 | /* |
2155 | * If the object for the space map has changed then dirty | |
2156 | * the top level so that we update the config. | |
2157 | */ | |
2158 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
2159 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
2160 | "new object %llu", txg, spa_name(spa), object, | |
2161 | space_map_object(vd->vdev_dtl_sm)); | |
2162 | vdev_config_dirty(vd->vdev_top); | |
2163 | } | |
34dc7c2f BB |
2164 | |
2165 | dmu_tx_commit(tx); | |
93cf2076 GW |
2166 | |
2167 | mutex_enter(&vd->vdev_dtl_lock); | |
2168 | space_map_update(vd->vdev_dtl_sm); | |
2169 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f BB |
2170 | } |
2171 | ||
fb5f0bc8 BB |
2172 | /* |
2173 | * Determine whether the specified vdev can be offlined/detached/removed | |
2174 | * without losing data. | |
2175 | */ | |
2176 | boolean_t | |
2177 | vdev_dtl_required(vdev_t *vd) | |
2178 | { | |
2179 | spa_t *spa = vd->vdev_spa; | |
2180 | vdev_t *tvd = vd->vdev_top; | |
2181 | uint8_t cant_read = vd->vdev_cant_read; | |
2182 | boolean_t required; | |
2183 | ||
2184 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2185 | ||
2186 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2187 | return (B_TRUE); | |
2188 | ||
2189 | /* | |
2190 | * Temporarily mark the device as unreadable, and then determine | |
2191 | * whether this results in any DTL outages in the top-level vdev. | |
2192 | * If not, we can safely offline/detach/remove the device. | |
2193 | */ | |
2194 | vd->vdev_cant_read = B_TRUE; | |
2195 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2196 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2197 | vd->vdev_cant_read = cant_read; | |
2198 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2199 | ||
572e2857 BB |
2200 | if (!required && zio_injection_enabled) |
2201 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2202 | ||
fb5f0bc8 BB |
2203 | return (required); |
2204 | } | |
2205 | ||
b128c09f BB |
2206 | /* |
2207 | * Determine if resilver is needed, and if so the txg range. | |
2208 | */ | |
2209 | boolean_t | |
2210 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2211 | { | |
2212 | boolean_t needed = B_FALSE; | |
2213 | uint64_t thismin = UINT64_MAX; | |
2214 | uint64_t thismax = 0; | |
d6320ddb | 2215 | int c; |
b128c09f BB |
2216 | |
2217 | if (vd->vdev_children == 0) { | |
2218 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2219 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && |
fb5f0bc8 | 2220 | vdev_writeable(vd)) { |
b128c09f | 2221 | |
5d1f7fb6 GW |
2222 | thismin = vdev_dtl_min(vd); |
2223 | thismax = vdev_dtl_max(vd); | |
b128c09f BB |
2224 | needed = B_TRUE; |
2225 | } | |
2226 | mutex_exit(&vd->vdev_dtl_lock); | |
2227 | } else { | |
d6320ddb | 2228 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2229 | vdev_t *cvd = vd->vdev_child[c]; |
2230 | uint64_t cmin, cmax; | |
2231 | ||
2232 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2233 | thismin = MIN(thismin, cmin); | |
2234 | thismax = MAX(thismax, cmax); | |
2235 | needed = B_TRUE; | |
2236 | } | |
2237 | } | |
2238 | } | |
2239 | ||
2240 | if (needed && minp) { | |
2241 | *minp = thismin; | |
2242 | *maxp = thismax; | |
2243 | } | |
2244 | return (needed); | |
2245 | } | |
2246 | ||
34dc7c2f BB |
2247 | void |
2248 | vdev_load(vdev_t *vd) | |
2249 | { | |
d6320ddb BB |
2250 | int c; |
2251 | ||
34dc7c2f BB |
2252 | /* |
2253 | * Recursively load all children. | |
2254 | */ | |
d6320ddb | 2255 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2256 | vdev_load(vd->vdev_child[c]); |
2257 | ||
2258 | /* | |
2259 | * If this is a top-level vdev, initialize its metaslabs. | |
2260 | */ | |
428870ff | 2261 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
2262 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
2263 | vdev_metaslab_init(vd, 0) != 0)) | |
2264 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2265 | VDEV_AUX_CORRUPT_DATA); | |
34dc7c2f BB |
2266 | /* |
2267 | * If this is a leaf vdev, load its DTL. | |
2268 | */ | |
2269 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
2270 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2271 | VDEV_AUX_CORRUPT_DATA); | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2276 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2277 | * we make sure that we can open the underlying device, then try to read the | |
2278 | * label, and make sure that the label is sane and that it hasn't been | |
2279 | * repurposed to another pool. | |
2280 | */ | |
2281 | int | |
2282 | vdev_validate_aux(vdev_t *vd) | |
2283 | { | |
2284 | nvlist_t *label; | |
2285 | uint64_t guid, version; | |
2286 | uint64_t state; | |
2287 | ||
b128c09f BB |
2288 | if (!vdev_readable(vd)) |
2289 | return (0); | |
2290 | ||
3bc7e0fb | 2291 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { |
34dc7c2f BB |
2292 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
2293 | VDEV_AUX_CORRUPT_DATA); | |
2294 | return (-1); | |
2295 | } | |
2296 | ||
2297 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
9ae529ec | 2298 | !SPA_VERSION_IS_SUPPORTED(version) || |
34dc7c2f BB |
2299 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || |
2300 | guid != vd->vdev_guid || | |
2301 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2302 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2303 | VDEV_AUX_CORRUPT_DATA); | |
2304 | nvlist_free(label); | |
2305 | return (-1); | |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * We don't actually check the pool state here. If it's in fact in | |
2310 | * use by another pool, we update this fact on the fly when requested. | |
2311 | */ | |
2312 | nvlist_free(label); | |
2313 | return (0); | |
2314 | } | |
2315 | ||
428870ff BB |
2316 | void |
2317 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2318 | { | |
2319 | spa_t *spa = vd->vdev_spa; | |
2320 | objset_t *mos = spa->spa_meta_objset; | |
2321 | dmu_tx_t *tx; | |
f3a7f661 | 2322 | int m, i; |
428870ff BB |
2323 | |
2324 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
e0ab3ab5 JS |
2325 | ASSERT(vd == vd->vdev_top); |
2326 | ASSERT3U(txg, ==, spa_syncing_txg(spa)); | |
428870ff | 2327 | |
428870ff | 2328 | if (vd->vdev_ms != NULL) { |
f3a7f661 GW |
2329 | metaslab_group_t *mg = vd->vdev_mg; |
2330 | ||
2331 | metaslab_group_histogram_verify(mg); | |
2332 | metaslab_class_histogram_verify(mg->mg_class); | |
2333 | ||
d6320ddb | 2334 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2335 | metaslab_t *msp = vd->vdev_ms[m]; |
2336 | ||
93cf2076 | 2337 | if (msp == NULL || msp->ms_sm == NULL) |
428870ff BB |
2338 | continue; |
2339 | ||
93cf2076 | 2340 | mutex_enter(&msp->ms_lock); |
f3a7f661 GW |
2341 | /* |
2342 | * If the metaslab was not loaded when the vdev | |
2343 | * was removed then the histogram accounting may | |
2344 | * not be accurate. Update the histogram information | |
2345 | * here so that we ensure that the metaslab group | |
2346 | * and metaslab class are up-to-date. | |
2347 | */ | |
2348 | metaslab_group_histogram_remove(mg, msp); | |
2349 | ||
93cf2076 GW |
2350 | VERIFY0(space_map_allocated(msp->ms_sm)); |
2351 | space_map_free(msp->ms_sm, tx); | |
2352 | space_map_close(msp->ms_sm); | |
2353 | msp->ms_sm = NULL; | |
2354 | mutex_exit(&msp->ms_lock); | |
428870ff | 2355 | } |
f3a7f661 GW |
2356 | |
2357 | metaslab_group_histogram_verify(mg); | |
2358 | metaslab_class_histogram_verify(mg->mg_class); | |
2359 | for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) | |
2360 | ASSERT0(mg->mg_histogram[i]); | |
2361 | ||
428870ff BB |
2362 | } |
2363 | ||
2364 | if (vd->vdev_ms_array) { | |
2365 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2366 | vd->vdev_ms_array = 0; | |
428870ff | 2367 | } |
e0ab3ab5 JS |
2368 | |
2369 | if (vd->vdev_islog && vd->vdev_top_zap != 0) { | |
2370 | vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); | |
2371 | vd->vdev_top_zap = 0; | |
2372 | } | |
428870ff BB |
2373 | dmu_tx_commit(tx); |
2374 | } | |
2375 | ||
34dc7c2f BB |
2376 | void |
2377 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2378 | { | |
2379 | metaslab_t *msp; | |
428870ff BB |
2380 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2381 | ||
2382 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2383 | |
c65aa5b2 | 2384 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2385 | metaslab_sync_done(msp, txg); |
428870ff BB |
2386 | |
2387 | if (reassess) | |
2388 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2389 | } |
2390 | ||
2391 | void | |
2392 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2393 | { | |
2394 | spa_t *spa = vd->vdev_spa; | |
2395 | vdev_t *lvd; | |
2396 | metaslab_t *msp; | |
2397 | dmu_tx_t *tx; | |
2398 | ||
428870ff BB |
2399 | ASSERT(!vd->vdev_ishole); |
2400 | ||
34dc7c2f BB |
2401 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2402 | ASSERT(vd == vd->vdev_top); | |
2403 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2404 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2405 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2406 | ASSERT(vd->vdev_ms_array != 0); | |
2407 | vdev_config_dirty(vd); | |
2408 | dmu_tx_commit(tx); | |
2409 | } | |
2410 | ||
428870ff BB |
2411 | /* |
2412 | * Remove the metadata associated with this vdev once it's empty. | |
2413 | */ | |
2414 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2415 | vdev_remove(vd, txg); | |
2416 | ||
34dc7c2f BB |
2417 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2418 | metaslab_sync(msp, txg); | |
2419 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2420 | } | |
2421 | ||
2422 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2423 | vdev_dtl_sync(lvd, txg); | |
2424 | ||
2425 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2426 | } | |
2427 | ||
2428 | uint64_t | |
2429 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2430 | { | |
2431 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2432 | } | |
2433 | ||
34dc7c2f BB |
2434 | /* |
2435 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2436 | * not be opened, and no I/O is attempted. | |
2437 | */ | |
2438 | int | |
428870ff | 2439 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2440 | { |
572e2857 | 2441 | vdev_t *vd, *tvd; |
34dc7c2f | 2442 | |
428870ff | 2443 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2444 | |
b128c09f BB |
2445 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2446 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2447 | |
34dc7c2f | 2448 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2449 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2450 | |
572e2857 BB |
2451 | tvd = vd->vdev_top; |
2452 | ||
428870ff BB |
2453 | /* |
2454 | * We don't directly use the aux state here, but if we do a | |
2455 | * vdev_reopen(), we need this value to be present to remember why we | |
2456 | * were faulted. | |
2457 | */ | |
2458 | vd->vdev_label_aux = aux; | |
2459 | ||
34dc7c2f BB |
2460 | /* |
2461 | * Faulted state takes precedence over degraded. | |
2462 | */ | |
428870ff | 2463 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2464 | vd->vdev_faulted = 1ULL; |
2465 | vd->vdev_degraded = 0ULL; | |
428870ff | 2466 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2467 | |
2468 | /* | |
428870ff BB |
2469 | * If this device has the only valid copy of the data, then |
2470 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2471 | */ |
572e2857 | 2472 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2473 | vd->vdev_degraded = 1ULL; |
2474 | vd->vdev_faulted = 0ULL; | |
2475 | ||
2476 | /* | |
2477 | * If we reopen the device and it's not dead, only then do we | |
2478 | * mark it degraded. | |
2479 | */ | |
572e2857 | 2480 | vdev_reopen(tvd); |
34dc7c2f | 2481 | |
428870ff BB |
2482 | if (vdev_readable(vd)) |
2483 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2484 | } |
2485 | ||
b128c09f | 2486 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2487 | } |
2488 | ||
2489 | /* | |
2490 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2491 | * user that something is wrong. The vdev continues to operate as normal as far | |
2492 | * as I/O is concerned. | |
2493 | */ | |
2494 | int | |
428870ff | 2495 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2496 | { |
b128c09f | 2497 | vdev_t *vd; |
34dc7c2f | 2498 | |
428870ff | 2499 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2500 | |
b128c09f BB |
2501 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2502 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2503 | |
34dc7c2f | 2504 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2505 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2506 | |
2507 | /* | |
2508 | * If the vdev is already faulted, then don't do anything. | |
2509 | */ | |
b128c09f BB |
2510 | if (vd->vdev_faulted || vd->vdev_degraded) |
2511 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2512 | |
2513 | vd->vdev_degraded = 1ULL; | |
2514 | if (!vdev_is_dead(vd)) | |
2515 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2516 | aux); |
34dc7c2f | 2517 | |
b128c09f | 2518 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2519 | } |
2520 | ||
2521 | /* | |
d3cc8b15 WA |
2522 | * Online the given vdev. |
2523 | * | |
2524 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2525 | * spare device should be detached when the device finishes resilvering. | |
2526 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2527 | * events are generated if the device fails to open. | |
34dc7c2f BB |
2528 | */ |
2529 | int | |
b128c09f | 2530 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2531 | { |
9babb374 | 2532 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
153b2285 YP |
2533 | boolean_t wasoffline; |
2534 | vdev_state_t oldstate; | |
34dc7c2f | 2535 | |
428870ff | 2536 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2537 | |
b128c09f BB |
2538 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2539 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2540 | |
2541 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2542 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2543 | |
153b2285 YP |
2544 | wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); |
2545 | oldstate = vd->vdev_state; | |
fb390aaf | 2546 | |
9babb374 | 2547 | tvd = vd->vdev_top; |
34dc7c2f BB |
2548 | vd->vdev_offline = B_FALSE; |
2549 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2550 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2551 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2552 | |
2553 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2554 | if (!vd->vdev_aux) { | |
2555 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2556 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2557 | } | |
2558 | ||
2559 | vdev_reopen(tvd); | |
34dc7c2f BB |
2560 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2561 | ||
9babb374 BB |
2562 | if (!vd->vdev_aux) { |
2563 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2564 | pvd->vdev_expanding = B_FALSE; | |
2565 | } | |
2566 | ||
34dc7c2f BB |
2567 | if (newstate) |
2568 | *newstate = vd->vdev_state; | |
2569 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2570 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2571 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2572 | vd->vdev_parent->vdev_child[0] == vd) | |
2573 | vd->vdev_unspare = B_TRUE; | |
2574 | ||
9babb374 BB |
2575 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2576 | ||
2577 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2578 | if (vd->vdev_aux) | |
2579 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2580 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2581 | } | |
fb390aaf | 2582 | |
153b2285 YP |
2583 | if (wasoffline || |
2584 | (oldstate < VDEV_STATE_DEGRADED && | |
2585 | vd->vdev_state >= VDEV_STATE_DEGRADED)) | |
fb390aaf HR |
2586 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE); |
2587 | ||
fb5f0bc8 | 2588 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2589 | } |
2590 | ||
428870ff BB |
2591 | static int |
2592 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2593 | { |
9babb374 | 2594 | vdev_t *vd, *tvd; |
428870ff BB |
2595 | int error = 0; |
2596 | uint64_t generation; | |
2597 | metaslab_group_t *mg; | |
34dc7c2f | 2598 | |
428870ff BB |
2599 | top: |
2600 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2601 | |
b128c09f BB |
2602 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2603 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2604 | |
2605 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2606 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2607 | |
9babb374 | 2608 | tvd = vd->vdev_top; |
428870ff BB |
2609 | mg = tvd->vdev_mg; |
2610 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2611 | |
34dc7c2f BB |
2612 | /* |
2613 | * If the device isn't already offline, try to offline it. | |
2614 | */ | |
2615 | if (!vd->vdev_offline) { | |
2616 | /* | |
fb5f0bc8 | 2617 | * If this device has the only valid copy of some data, |
9babb374 BB |
2618 | * don't allow it to be offlined. Log devices are always |
2619 | * expendable. | |
34dc7c2f | 2620 | */ |
9babb374 BB |
2621 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2622 | vdev_dtl_required(vd)) | |
b128c09f | 2623 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2624 | |
428870ff BB |
2625 | /* |
2626 | * If the top-level is a slog and it has had allocations | |
2627 | * then proceed. We check that the vdev's metaslab group | |
2628 | * is not NULL since it's possible that we may have just | |
2629 | * added this vdev but not yet initialized its metaslabs. | |
2630 | */ | |
2631 | if (tvd->vdev_islog && mg != NULL) { | |
2632 | /* | |
2633 | * Prevent any future allocations. | |
2634 | */ | |
2635 | metaslab_group_passivate(mg); | |
2636 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2637 | ||
2638 | error = spa_offline_log(spa); | |
2639 | ||
2640 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2641 | ||
2642 | /* | |
2643 | * Check to see if the config has changed. | |
2644 | */ | |
2645 | if (error || generation != spa->spa_config_generation) { | |
2646 | metaslab_group_activate(mg); | |
2647 | if (error) | |
2648 | return (spa_vdev_state_exit(spa, | |
2649 | vd, error)); | |
2650 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2651 | goto top; | |
2652 | } | |
c99c9001 | 2653 | ASSERT0(tvd->vdev_stat.vs_alloc); |
428870ff BB |
2654 | } |
2655 | ||
34dc7c2f BB |
2656 | /* |
2657 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2658 | * If the top-level vdev is a log device then just offline |
2659 | * it. Otherwise, if this action results in the top-level | |
2660 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2661 | */ |
2662 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2663 | vdev_reopen(tvd); |
2664 | ||
2665 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2666 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2667 | vd->vdev_offline = B_FALSE; |
9babb374 | 2668 | vdev_reopen(tvd); |
b128c09f | 2669 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2670 | } |
428870ff BB |
2671 | |
2672 | /* | |
2673 | * Add the device back into the metaslab rotor so that | |
2674 | * once we online the device it's open for business. | |
2675 | */ | |
2676 | if (tvd->vdev_islog && mg != NULL) | |
2677 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2678 | } |
2679 | ||
b128c09f | 2680 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2681 | |
428870ff BB |
2682 | return (spa_vdev_state_exit(spa, vd, 0)); |
2683 | } | |
9babb374 | 2684 | |
428870ff BB |
2685 | int |
2686 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2687 | { | |
2688 | int error; | |
9babb374 | 2689 | |
428870ff BB |
2690 | mutex_enter(&spa->spa_vdev_top_lock); |
2691 | error = vdev_offline_locked(spa, guid, flags); | |
2692 | mutex_exit(&spa->spa_vdev_top_lock); | |
2693 | ||
2694 | return (error); | |
34dc7c2f BB |
2695 | } |
2696 | ||
2697 | /* | |
2698 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2699 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2700 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2701 | */ |
2702 | void | |
b128c09f | 2703 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2704 | { |
b128c09f | 2705 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2706 | int c; |
b128c09f BB |
2707 | |
2708 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2709 | |
2710 | if (vd == NULL) | |
b128c09f | 2711 | vd = rvd; |
34dc7c2f BB |
2712 | |
2713 | vd->vdev_stat.vs_read_errors = 0; | |
2714 | vd->vdev_stat.vs_write_errors = 0; | |
2715 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2716 | |
d6320ddb | 2717 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2718 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2719 | |
2720 | /* | |
b128c09f BB |
2721 | * If we're in the FAULTED state or have experienced failed I/O, then |
2722 | * clear the persistent state and attempt to reopen the device. We | |
2723 | * also mark the vdev config dirty, so that the new faulted state is | |
2724 | * written out to disk. | |
34dc7c2f | 2725 | */ |
b128c09f BB |
2726 | if (vd->vdev_faulted || vd->vdev_degraded || |
2727 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2728 | ||
428870ff | 2729 | /* |
4e33ba4c | 2730 | * When reopening in response to a clear event, it may be due to |
428870ff BB |
2731 | * a fmadm repair request. In this case, if the device is |
2732 | * still broken, we want to still post the ereport again. | |
2733 | */ | |
2734 | vd->vdev_forcefault = B_TRUE; | |
2735 | ||
572e2857 | 2736 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2737 | vd->vdev_cant_read = B_FALSE; |
2738 | vd->vdev_cant_write = B_FALSE; | |
2739 | ||
572e2857 | 2740 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2741 | |
428870ff BB |
2742 | vd->vdev_forcefault = B_FALSE; |
2743 | ||
572e2857 | 2744 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2745 | vdev_state_dirty(vd->vdev_top); |
2746 | ||
2747 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2748 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2749 | ||
fb390aaf | 2750 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); |
34dc7c2f | 2751 | } |
428870ff BB |
2752 | |
2753 | /* | |
2754 | * When clearing a FMA-diagnosed fault, we always want to | |
2755 | * unspare the device, as we assume that the original spare was | |
2756 | * done in response to the FMA fault. | |
2757 | */ | |
2758 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2759 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2760 | vd->vdev_parent->vdev_child[0] == vd) | |
2761 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2762 | } |
2763 | ||
b128c09f BB |
2764 | boolean_t |
2765 | vdev_is_dead(vdev_t *vd) | |
2766 | { | |
428870ff BB |
2767 | /* |
2768 | * Holes and missing devices are always considered "dead". | |
2769 | * This simplifies the code since we don't have to check for | |
2770 | * these types of devices in the various code paths. | |
2771 | * Instead we rely on the fact that we skip over dead devices | |
2772 | * before issuing I/O to them. | |
2773 | */ | |
2774 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2775 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2776 | } |
2777 | ||
2778 | boolean_t | |
34dc7c2f BB |
2779 | vdev_readable(vdev_t *vd) |
2780 | { | |
b128c09f | 2781 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2782 | } |
2783 | ||
b128c09f | 2784 | boolean_t |
34dc7c2f BB |
2785 | vdev_writeable(vdev_t *vd) |
2786 | { | |
b128c09f | 2787 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2788 | } |
2789 | ||
b128c09f BB |
2790 | boolean_t |
2791 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2792 | { |
fb5f0bc8 BB |
2793 | uint64_t state = vd->vdev_state; |
2794 | ||
b128c09f | 2795 | /* |
fb5f0bc8 | 2796 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2797 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2798 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2799 | * the proper locks. Note that we have to get the vdev state |
2800 | * in a local variable because although it changes atomically, | |
2801 | * we're asking two separate questions about it. | |
b128c09f | 2802 | */ |
fb5f0bc8 | 2803 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
3dfb57a3 DB |
2804 | !vd->vdev_cant_write && !vd->vdev_ishole && |
2805 | vd->vdev_mg->mg_initialized); | |
34dc7c2f BB |
2806 | } |
2807 | ||
b128c09f BB |
2808 | boolean_t |
2809 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2810 | { |
b128c09f | 2811 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2812 | |
b128c09f BB |
2813 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2814 | return (B_FALSE); | |
34dc7c2f | 2815 | |
b128c09f BB |
2816 | if (zio->io_type == ZIO_TYPE_READ) |
2817 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2818 | |
b128c09f BB |
2819 | if (zio->io_type == ZIO_TYPE_WRITE) |
2820 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2821 | |
b128c09f | 2822 | return (B_TRUE); |
34dc7c2f BB |
2823 | } |
2824 | ||
193a37cb TH |
2825 | static void |
2826 | vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) | |
34dc7c2f | 2827 | { |
193a37cb TH |
2828 | int t; |
2829 | for (t = 0; t < ZIO_TYPES; t++) { | |
2830 | vs->vs_ops[t] += cvs->vs_ops[t]; | |
2831 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2832 | } | |
34dc7c2f | 2833 | |
193a37cb TH |
2834 | cvs->vs_scan_removing = cvd->vdev_removing; |
2835 | } | |
f3a7f661 | 2836 | |
193a37cb TH |
2837 | /* |
2838 | * Get extended stats | |
2839 | */ | |
2840 | static void | |
2841 | vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) | |
2842 | { | |
2843 | int t, b; | |
2844 | for (t = 0; t < ZIO_TYPES; t++) { | |
7e945072 | 2845 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) |
193a37cb | 2846 | vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; |
7e945072 TH |
2847 | |
2848 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { | |
193a37cb TH |
2849 | vsx->vsx_total_histo[t][b] += |
2850 | cvsx->vsx_total_histo[t][b]; | |
2851 | } | |
f38dfec3 | 2852 | } |
34dc7c2f | 2853 | |
193a37cb | 2854 | for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { |
7e945072 | 2855 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { |
193a37cb TH |
2856 | vsx->vsx_queue_histo[t][b] += |
2857 | cvsx->vsx_queue_histo[t][b]; | |
2858 | } | |
2859 | vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; | |
2860 | vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; | |
7e945072 TH |
2861 | |
2862 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) | |
2863 | vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; | |
2864 | ||
2865 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) | |
2866 | vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; | |
193a37cb | 2867 | } |
7e945072 | 2868 | |
193a37cb TH |
2869 | } |
2870 | ||
2871 | /* | |
2872 | * Get statistics for the given vdev. | |
2873 | */ | |
2874 | static void | |
2875 | vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
2876 | { | |
2877 | int c, t; | |
34dc7c2f BB |
2878 | /* |
2879 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2880 | * over all top-level vdevs (i.e. the direct children of the root). | |
2881 | */ | |
193a37cb TH |
2882 | if (!vd->vdev_ops->vdev_op_leaf) { |
2883 | if (vs) { | |
2884 | memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); | |
2885 | memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); | |
2886 | } | |
2887 | if (vsx) | |
2888 | memset(vsx, 0, sizeof (*vsx)); | |
2889 | ||
2890 | for (c = 0; c < vd->vdev_children; c++) { | |
2891 | vdev_t *cvd = vd->vdev_child[c]; | |
34dc7c2f | 2892 | vdev_stat_t *cvs = &cvd->vdev_stat; |
193a37cb TH |
2893 | vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; |
2894 | ||
2895 | vdev_get_stats_ex_impl(cvd, cvs, cvsx); | |
2896 | if (vs) | |
2897 | vdev_get_child_stat(cvd, vs, cvs); | |
2898 | if (vsx) | |
2899 | vdev_get_child_stat_ex(cvd, vsx, cvsx); | |
34dc7c2f | 2900 | |
193a37cb TH |
2901 | } |
2902 | } else { | |
2903 | /* | |
2904 | * We're a leaf. Just copy our ZIO active queue stats in. The | |
2905 | * other leaf stats are updated in vdev_stat_update(). | |
2906 | */ | |
2907 | if (!vsx) | |
2908 | return; | |
2909 | ||
2910 | memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); | |
2911 | ||
2912 | for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { | |
2913 | vsx->vsx_active_queue[t] = | |
2914 | vd->vdev_queue.vq_class[t].vqc_active; | |
2915 | vsx->vsx_pend_queue[t] = avl_numnodes( | |
2916 | &vd->vdev_queue.vq_class[t].vqc_queued_tree); | |
34dc7c2f BB |
2917 | } |
2918 | } | |
193a37cb TH |
2919 | } |
2920 | ||
2921 | void | |
2922 | vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
2923 | { | |
0f676dc2 | 2924 | vdev_t *tvd = vd->vdev_top; |
193a37cb TH |
2925 | mutex_enter(&vd->vdev_stat_lock); |
2926 | if (vs) { | |
2927 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2928 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2929 | vs->vs_state = vd->vdev_state; | |
2930 | vs->vs_rsize = vdev_get_min_asize(vd); | |
2931 | if (vd->vdev_ops->vdev_op_leaf) | |
2932 | vs->vs_rsize += VDEV_LABEL_START_SIZE + | |
2933 | VDEV_LABEL_END_SIZE; | |
0f676dc2 GM |
2934 | /* |
2935 | * Report expandable space on top-level, non-auxillary devices | |
2936 | * only. The expandable space is reported in terms of metaslab | |
2937 | * sized units since that determines how much space the pool | |
2938 | * can expand. | |
2939 | */ | |
2940 | if (vd->vdev_aux == NULL && tvd != NULL) { | |
2941 | vs->vs_esize = P2ALIGN( | |
2942 | vd->vdev_max_asize - vd->vdev_asize, | |
2943 | 1ULL << tvd->vdev_ms_shift); | |
2944 | } | |
193a37cb TH |
2945 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
2946 | if (vd->vdev_aux == NULL && vd == vd->vdev_top && | |
2947 | !vd->vdev_ishole) { | |
2948 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; | |
2949 | } | |
2950 | } | |
2951 | ||
2952 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_READER) != 0); | |
2953 | vdev_get_stats_ex_impl(vd, vs, vsx); | |
f3a7f661 | 2954 | mutex_exit(&vd->vdev_stat_lock); |
34dc7c2f BB |
2955 | } |
2956 | ||
193a37cb TH |
2957 | void |
2958 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2959 | { | |
2960 | return (vdev_get_stats_ex(vd, vs, NULL)); | |
2961 | } | |
2962 | ||
34dc7c2f BB |
2963 | void |
2964 | vdev_clear_stats(vdev_t *vd) | |
2965 | { | |
2966 | mutex_enter(&vd->vdev_stat_lock); | |
2967 | vd->vdev_stat.vs_space = 0; | |
2968 | vd->vdev_stat.vs_dspace = 0; | |
2969 | vd->vdev_stat.vs_alloc = 0; | |
2970 | mutex_exit(&vd->vdev_stat_lock); | |
2971 | } | |
2972 | ||
428870ff BB |
2973 | void |
2974 | vdev_scan_stat_init(vdev_t *vd) | |
2975 | { | |
2976 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2977 | int c; |
428870ff | 2978 | |
d6320ddb | 2979 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2980 | vdev_scan_stat_init(vd->vdev_child[c]); |
2981 | ||
2982 | mutex_enter(&vd->vdev_stat_lock); | |
2983 | vs->vs_scan_processed = 0; | |
2984 | mutex_exit(&vd->vdev_stat_lock); | |
2985 | } | |
2986 | ||
34dc7c2f | 2987 | void |
b128c09f | 2988 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2989 | { |
fb5f0bc8 BB |
2990 | spa_t *spa = zio->io_spa; |
2991 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2992 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2993 | vdev_t *pvd; |
2994 | uint64_t txg = zio->io_txg; | |
2995 | vdev_stat_t *vs = &vd->vdev_stat; | |
193a37cb | 2996 | vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; |
34dc7c2f BB |
2997 | zio_type_t type = zio->io_type; |
2998 | int flags = zio->io_flags; | |
2999 | ||
b128c09f BB |
3000 | /* |
3001 | * If this i/o is a gang leader, it didn't do any actual work. | |
3002 | */ | |
3003 | if (zio->io_gang_tree) | |
3004 | return; | |
3005 | ||
34dc7c2f | 3006 | if (zio->io_error == 0) { |
b128c09f BB |
3007 | /* |
3008 | * If this is a root i/o, don't count it -- we've already | |
3009 | * counted the top-level vdevs, and vdev_get_stats() will | |
3010 | * aggregate them when asked. This reduces contention on | |
3011 | * the root vdev_stat_lock and implicitly handles blocks | |
3012 | * that compress away to holes, for which there is no i/o. | |
3013 | * (Holes never create vdev children, so all the counters | |
3014 | * remain zero, which is what we want.) | |
3015 | * | |
3016 | * Note: this only applies to successful i/o (io_error == 0) | |
3017 | * because unlike i/o counts, errors are not additive. | |
3018 | * When reading a ditto block, for example, failure of | |
3019 | * one top-level vdev does not imply a root-level error. | |
3020 | */ | |
3021 | if (vd == rvd) | |
3022 | return; | |
3023 | ||
3024 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
3025 | |
3026 | if (flags & ZIO_FLAG_IO_BYPASS) | |
3027 | return; | |
3028 | ||
3029 | mutex_enter(&vd->vdev_stat_lock); | |
3030 | ||
b128c09f | 3031 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 3032 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
3033 | dsl_scan_phys_t *scn_phys = |
3034 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
3035 | uint64_t *processed = &scn_phys->scn_processed; | |
3036 | ||
3037 | /* XXX cleanup? */ | |
3038 | if (vd->vdev_ops->vdev_op_leaf) | |
3039 | atomic_add_64(processed, psize); | |
3040 | vs->vs_scan_processed += psize; | |
3041 | } | |
3042 | ||
fb5f0bc8 | 3043 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 3044 | vs->vs_self_healed += psize; |
34dc7c2f | 3045 | } |
fb5f0bc8 | 3046 | |
193a37cb TH |
3047 | /* |
3048 | * The bytes/ops/histograms are recorded at the leaf level and | |
3049 | * aggregated into the higher level vdevs in vdev_get_stats(). | |
3050 | */ | |
4eb0db42 TH |
3051 | if (vd->vdev_ops->vdev_op_leaf && |
3052 | (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { | |
193a37cb TH |
3053 | |
3054 | vs->vs_ops[type]++; | |
3055 | vs->vs_bytes[type] += psize; | |
3056 | ||
7e945072 TH |
3057 | if (flags & ZIO_FLAG_DELEGATED) { |
3058 | vsx->vsx_agg_histo[zio->io_priority] | |
3059 | [RQ_HISTO(zio->io_size)]++; | |
3060 | } else { | |
3061 | vsx->vsx_ind_histo[zio->io_priority] | |
3062 | [RQ_HISTO(zio->io_size)]++; | |
3063 | } | |
3064 | ||
193a37cb TH |
3065 | if (zio->io_delta && zio->io_delay) { |
3066 | vsx->vsx_queue_histo[zio->io_priority] | |
7e945072 | 3067 | [L_HISTO(zio->io_delta - zio->io_delay)]++; |
193a37cb | 3068 | vsx->vsx_disk_histo[type] |
7e945072 | 3069 | [L_HISTO(zio->io_delay)]++; |
193a37cb | 3070 | vsx->vsx_total_histo[type] |
7e945072 | 3071 | [L_HISTO(zio->io_delta)]++; |
193a37cb TH |
3072 | } |
3073 | } | |
fb5f0bc8 BB |
3074 | |
3075 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
3076 | return; |
3077 | } | |
3078 | ||
3079 | if (flags & ZIO_FLAG_SPECULATIVE) | |
3080 | return; | |
3081 | ||
9babb374 BB |
3082 | /* |
3083 | * If this is an I/O error that is going to be retried, then ignore the | |
3084 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
3085 | * hard errors, when in reality they can happen for any number of | |
3086 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
3087 | */ | |
3088 | if (zio->io_error == EIO && | |
3089 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
3090 | return; | |
3091 | ||
428870ff BB |
3092 | /* |
3093 | * Intent logs writes won't propagate their error to the root | |
3094 | * I/O so don't mark these types of failures as pool-level | |
3095 | * errors. | |
3096 | */ | |
3097 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
3098 | return; | |
3099 | ||
b128c09f | 3100 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 3101 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
3102 | if (zio->io_error == ECKSUM) |
3103 | vs->vs_checksum_errors++; | |
3104 | else | |
3105 | vs->vs_read_errors++; | |
34dc7c2f | 3106 | } |
9babb374 | 3107 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
3108 | vs->vs_write_errors++; |
3109 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 3110 | |
fb5f0bc8 BB |
3111 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
3112 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 3113 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 3114 | spa->spa_claiming)) { |
fb5f0bc8 | 3115 | /* |
428870ff BB |
3116 | * This is either a normal write (not a repair), or it's |
3117 | * a repair induced by the scrub thread, or it's a repair | |
3118 | * made by zil_claim() during spa_load() in the first txg. | |
3119 | * In the normal case, we commit the DTL change in the same | |
3120 | * txg as the block was born. In the scrub-induced repair | |
3121 | * case, we know that scrubs run in first-pass syncing context, | |
3122 | * so we commit the DTL change in spa_syncing_txg(spa). | |
3123 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
3124 | * |
3125 | * We currently do not make DTL entries for failed spontaneous | |
3126 | * self-healing writes triggered by normal (non-scrubbing) | |
3127 | * reads, because we have no transactional context in which to | |
3128 | * do so -- and it's not clear that it'd be desirable anyway. | |
3129 | */ | |
3130 | if (vd->vdev_ops->vdev_op_leaf) { | |
3131 | uint64_t commit_txg = txg; | |
572e2857 | 3132 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
3133 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
3134 | ASSERT(spa_sync_pass(spa) == 1); | |
3135 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
3136 | commit_txg = spa_syncing_txg(spa); |
3137 | } else if (spa->spa_claiming) { | |
3138 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
3139 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 3140 | } |
428870ff | 3141 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 3142 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 3143 | return; |
fb5f0bc8 BB |
3144 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
3145 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
3146 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 3147 | } |
fb5f0bc8 BB |
3148 | if (vd != rvd) |
3149 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
3150 | } |
3151 | } | |
3152 | ||
34dc7c2f | 3153 | /* |
428870ff BB |
3154 | * Update the in-core space usage stats for this vdev, its metaslab class, |
3155 | * and the root vdev. | |
34dc7c2f BB |
3156 | */ |
3157 | void | |
428870ff BB |
3158 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
3159 | int64_t space_delta) | |
34dc7c2f BB |
3160 | { |
3161 | int64_t dspace_delta = space_delta; | |
3162 | spa_t *spa = vd->vdev_spa; | |
3163 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
3164 | metaslab_group_t *mg = vd->vdev_mg; |
3165 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
3166 | |
3167 | ASSERT(vd == vd->vdev_top); | |
3168 | ||
3169 | /* | |
3170 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
3171 | * factor. We must calculate this here and not at the root vdev | |
3172 | * because the root vdev's psize-to-asize is simply the max of its | |
3173 | * childrens', thus not accurate enough for us. | |
3174 | */ | |
3175 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 3176 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
3177 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
3178 | vd->vdev_deflate_ratio; | |
3179 | ||
3180 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 3181 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3182 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3183 | vd->vdev_stat.vs_dspace += dspace_delta; |
3184 | mutex_exit(&vd->vdev_stat_lock); | |
3185 | ||
428870ff | 3186 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 3187 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 3188 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3189 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3190 | rvd->vdev_stat.vs_dspace += dspace_delta; |
3191 | mutex_exit(&rvd->vdev_stat_lock); | |
3192 | } | |
428870ff BB |
3193 | |
3194 | if (mc != NULL) { | |
3195 | ASSERT(rvd == vd->vdev_parent); | |
3196 | ASSERT(vd->vdev_ms_count != 0); | |
3197 | ||
3198 | metaslab_class_space_update(mc, | |
3199 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
3200 | } | |
34dc7c2f BB |
3201 | } |
3202 | ||
3203 | /* | |
3204 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
3205 | * so that it will be written out next time the vdev configuration is synced. | |
3206 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
3207 | */ | |
3208 | void | |
3209 | vdev_config_dirty(vdev_t *vd) | |
3210 | { | |
3211 | spa_t *spa = vd->vdev_spa; | |
3212 | vdev_t *rvd = spa->spa_root_vdev; | |
3213 | int c; | |
3214 | ||
572e2857 BB |
3215 | ASSERT(spa_writeable(spa)); |
3216 | ||
34dc7c2f | 3217 | /* |
9babb374 BB |
3218 | * If this is an aux vdev (as with l2cache and spare devices), then we |
3219 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
3220 | */ |
3221 | if (vd->vdev_aux != NULL) { | |
3222 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
3223 | nvlist_t **aux; | |
3224 | uint_t naux; | |
3225 | ||
3226 | for (c = 0; c < sav->sav_count; c++) { | |
3227 | if (sav->sav_vdevs[c] == vd) | |
3228 | break; | |
3229 | } | |
3230 | ||
3231 | if (c == sav->sav_count) { | |
3232 | /* | |
3233 | * We're being removed. There's nothing more to do. | |
3234 | */ | |
3235 | ASSERT(sav->sav_sync == B_TRUE); | |
3236 | return; | |
3237 | } | |
3238 | ||
3239 | sav->sav_sync = B_TRUE; | |
3240 | ||
9babb374 BB |
3241 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
3242 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
3243 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
3244 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
3245 | } | |
b128c09f BB |
3246 | |
3247 | ASSERT(c < naux); | |
3248 | ||
3249 | /* | |
3250 | * Setting the nvlist in the middle if the array is a little | |
3251 | * sketchy, but it will work. | |
3252 | */ | |
3253 | nvlist_free(aux[c]); | |
428870ff | 3254 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
3255 | |
3256 | return; | |
3257 | } | |
3258 | ||
3259 | /* | |
3260 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
3261 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
3262 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
3263 | * so this is sufficient to ensure mutual exclusion. |
3264 | */ | |
b128c09f BB |
3265 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3266 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3267 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
3268 | |
3269 | if (vd == rvd) { | |
3270 | for (c = 0; c < rvd->vdev_children; c++) | |
3271 | vdev_config_dirty(rvd->vdev_child[c]); | |
3272 | } else { | |
3273 | ASSERT(vd == vd->vdev_top); | |
3274 | ||
428870ff BB |
3275 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
3276 | !vd->vdev_ishole) | |
b128c09f | 3277 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
3278 | } |
3279 | } | |
3280 | ||
3281 | void | |
3282 | vdev_config_clean(vdev_t *vd) | |
3283 | { | |
3284 | spa_t *spa = vd->vdev_spa; | |
3285 | ||
b128c09f BB |
3286 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3287 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3288 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 3289 | |
b128c09f BB |
3290 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
3291 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
3292 | } |
3293 | ||
b128c09f BB |
3294 | /* |
3295 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
3296 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
3297 | * the state changes from larger config changes because they require | |
3298 | * much less locking, and are often needed for administrative actions. | |
3299 | */ | |
3300 | void | |
3301 | vdev_state_dirty(vdev_t *vd) | |
3302 | { | |
3303 | spa_t *spa = vd->vdev_spa; | |
3304 | ||
572e2857 | 3305 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
3306 | ASSERT(vd == vd->vdev_top); |
3307 | ||
3308 | /* | |
3309 | * The state list is protected by the SCL_STATE lock. The caller | |
3310 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3311 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3312 | * so this is sufficient to ensure mutual exclusion. | |
3313 | */ | |
3314 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3315 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3316 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3317 | ||
428870ff | 3318 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
3319 | list_insert_head(&spa->spa_state_dirty_list, vd); |
3320 | } | |
3321 | ||
3322 | void | |
3323 | vdev_state_clean(vdev_t *vd) | |
3324 | { | |
3325 | spa_t *spa = vd->vdev_spa; | |
3326 | ||
3327 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3328 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3329 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3330 | ||
3331 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3332 | list_remove(&spa->spa_state_dirty_list, vd); | |
3333 | } | |
3334 | ||
3335 | /* | |
3336 | * Propagate vdev state up from children to parent. | |
3337 | */ | |
34dc7c2f BB |
3338 | void |
3339 | vdev_propagate_state(vdev_t *vd) | |
3340 | { | |
fb5f0bc8 BB |
3341 | spa_t *spa = vd->vdev_spa; |
3342 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
3343 | int degraded = 0, faulted = 0; |
3344 | int corrupted = 0; | |
34dc7c2f | 3345 | vdev_t *child; |
d6320ddb | 3346 | int c; |
34dc7c2f BB |
3347 | |
3348 | if (vd->vdev_children > 0) { | |
d6320ddb | 3349 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 3350 | child = vd->vdev_child[c]; |
b128c09f | 3351 | |
428870ff BB |
3352 | /* |
3353 | * Don't factor holes into the decision. | |
3354 | */ | |
3355 | if (child->vdev_ishole) | |
3356 | continue; | |
3357 | ||
b128c09f | 3358 | if (!vdev_readable(child) || |
fb5f0bc8 | 3359 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
3360 | /* |
3361 | * Root special: if there is a top-level log | |
3362 | * device, treat the root vdev as if it were | |
3363 | * degraded. | |
3364 | */ | |
3365 | if (child->vdev_islog && vd == rvd) | |
3366 | degraded++; | |
3367 | else | |
3368 | faulted++; | |
3369 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 3370 | degraded++; |
b128c09f | 3371 | } |
34dc7c2f BB |
3372 | |
3373 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3374 | corrupted++; | |
3375 | } | |
3376 | ||
3377 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3378 | ||
3379 | /* | |
b128c09f | 3380 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
3381 | * opened due to corrupted metadata, then propagate the root |
3382 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3383 | * replicas'. | |
3384 | */ | |
3385 | if (corrupted && vd == rvd && | |
3386 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3387 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3388 | VDEV_AUX_CORRUPT_DATA); | |
3389 | } | |
3390 | ||
b128c09f | 3391 | if (vd->vdev_parent) |
34dc7c2f BB |
3392 | vdev_propagate_state(vd->vdev_parent); |
3393 | } | |
3394 | ||
3395 | /* | |
3396 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3397 | * state, because we're in the process of opening children depth-first. | |
3398 | * Otherwise, we propagate the change to the parent. | |
3399 | * | |
3400 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3401 | * generated. | |
3402 | */ | |
3403 | void | |
3404 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3405 | { | |
3406 | uint64_t save_state; | |
b128c09f | 3407 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
3408 | |
3409 | if (state == vd->vdev_state) { | |
976246fa DB |
3410 | /* |
3411 | * Since vdev_offline() code path is already in an offline | |
3412 | * state we can miss a statechange event to OFFLINE. Check | |
3413 | * the previous state to catch this condition. | |
3414 | */ | |
3415 | if (vd->vdev_ops->vdev_op_leaf && | |
3416 | (state == VDEV_STATE_OFFLINE) && | |
3417 | (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { | |
3418 | /* post an offline state change */ | |
3419 | zfs_post_state_change(spa, vd, vd->vdev_prevstate); | |
3420 | } | |
34dc7c2f BB |
3421 | vd->vdev_stat.vs_aux = aux; |
3422 | return; | |
3423 | } | |
3424 | ||
3425 | save_state = vd->vdev_state; | |
3426 | ||
3427 | vd->vdev_state = state; | |
3428 | vd->vdev_stat.vs_aux = aux; | |
3429 | ||
3430 | /* | |
3431 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
3432 | * always close the underlying device unless the device has requested |
3433 | * a delayed close (i.e. we're about to remove or fault the device). | |
3434 | * Otherwise, we keep accessible but invalid devices open forever. | |
3435 | * We don't call vdev_close() itself, because that implies some extra | |
3436 | * checks (offline, etc) that we don't want here. This is limited to | |
3437 | * leaf devices, because otherwise closing the device will affect other | |
3438 | * children. | |
34dc7c2f | 3439 | */ |
428870ff BB |
3440 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
3441 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
3442 | vd->vdev_ops->vdev_op_close(vd); |
3443 | ||
3444 | if (vd->vdev_removed && | |
3445 | state == VDEV_STATE_CANT_OPEN && | |
3446 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3447 | /* | |
3448 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3449 | * device was previously marked removed and someone attempted to | |
3450 | * reopen it. If this failed due to a nonexistent device, then | |
3451 | * keep the device in the REMOVED state. We also let this be if | |
3452 | * it is one of our special test online cases, which is only | |
3453 | * attempting to online the device and shouldn't generate an FMA | |
3454 | * fault. | |
3455 | */ | |
3456 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3457 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3458 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3459 | vd->vdev_removed = B_TRUE; |
3460 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3461 | /* | |
572e2857 BB |
3462 | * If we fail to open a vdev during an import or recovery, we |
3463 | * mark it as "not available", which signifies that it was | |
3464 | * never there to begin with. Failure to open such a device | |
3465 | * is not considered an error. | |
34dc7c2f | 3466 | */ |
572e2857 BB |
3467 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3468 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3469 | vd->vdev_ops->vdev_op_leaf) |
3470 | vd->vdev_not_present = 1; | |
3471 | ||
3472 | /* | |
3473 | * Post the appropriate ereport. If the 'prevstate' field is | |
3474 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3475 | * that this is part of a vdev_reopen(). In this case, we don't | |
3476 | * want to post the ereport if the device was already in the | |
3477 | * CANT_OPEN state beforehand. | |
3478 | * | |
3479 | * If the 'checkremove' flag is set, then this is an attempt to | |
3480 | * online the device in response to an insertion event. If we | |
3481 | * hit this case, then we have detected an insertion event for a | |
3482 | * faulted or offline device that wasn't in the removed state. | |
3483 | * In this scenario, we don't post an ereport because we are | |
3484 | * about to replace the device, or attempt an online with | |
3485 | * vdev_forcefault, which will generate the fault for us. | |
3486 | */ | |
3487 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3488 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3489 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3490 | const char *class; |
3491 | ||
3492 | switch (aux) { | |
3493 | case VDEV_AUX_OPEN_FAILED: | |
3494 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3495 | break; | |
3496 | case VDEV_AUX_CORRUPT_DATA: | |
3497 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3498 | break; | |
3499 | case VDEV_AUX_NO_REPLICAS: | |
3500 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3501 | break; | |
3502 | case VDEV_AUX_BAD_GUID_SUM: | |
3503 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3504 | break; | |
3505 | case VDEV_AUX_TOO_SMALL: | |
3506 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3507 | break; | |
3508 | case VDEV_AUX_BAD_LABEL: | |
3509 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3510 | break; | |
ff61d1a4 | 3511 | case VDEV_AUX_BAD_ASHIFT: |
3512 | class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; | |
3513 | break; | |
34dc7c2f BB |
3514 | default: |
3515 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3516 | } | |
3517 | ||
b128c09f | 3518 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3519 | } |
3520 | ||
3521 | /* Erase any notion of persistent removed state */ | |
3522 | vd->vdev_removed = B_FALSE; | |
3523 | } else { | |
3524 | vd->vdev_removed = B_FALSE; | |
3525 | } | |
3526 | ||
d02ca379 DB |
3527 | /* |
3528 | * Notify ZED of any significant state-change on a leaf vdev. | |
3529 | * | |
d02ca379 | 3530 | */ |
6078881a TH |
3531 | if (vd->vdev_ops->vdev_op_leaf) { |
3532 | /* preserve original state from a vdev_reopen() */ | |
3533 | if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && | |
3534 | (vd->vdev_prevstate != vd->vdev_state) && | |
3535 | (save_state <= VDEV_STATE_CLOSED)) | |
3536 | save_state = vd->vdev_prevstate; | |
3537 | ||
3538 | /* filter out state change due to initial vdev_open */ | |
3539 | if (save_state > VDEV_STATE_CLOSED) | |
3540 | zfs_post_state_change(spa, vd, save_state); | |
d02ca379 DB |
3541 | } |
3542 | ||
9babb374 BB |
3543 | if (!isopen && vd->vdev_parent) |
3544 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3545 | } |
b128c09f BB |
3546 | |
3547 | /* | |
3548 | * Check the vdev configuration to ensure that it's capable of supporting | |
e550644f | 3549 | * a root pool. We do not support partial configuration. |
b128c09f BB |
3550 | */ |
3551 | boolean_t | |
3552 | vdev_is_bootable(vdev_t *vd) | |
3553 | { | |
b128c09f | 3554 | if (!vd->vdev_ops->vdev_op_leaf) { |
e550644f | 3555 | const char *vdev_type = vd->vdev_ops->vdev_op_type; |
b128c09f | 3556 | |
e550644f | 3557 | if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) |
b128c09f | 3558 | return (B_FALSE); |
b128c09f BB |
3559 | } |
3560 | ||
e550644f | 3561 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3562 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3563 | return (B_FALSE); | |
3564 | } | |
3565 | return (B_TRUE); | |
3566 | } | |
9babb374 | 3567 | |
428870ff BB |
3568 | /* |
3569 | * Load the state from the original vdev tree (ovd) which | |
3570 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3571 | * vdev was offline or faulted then we transfer that state to the |
3572 | * device in the current vdev tree (nvd). | |
428870ff | 3573 | */ |
9babb374 | 3574 | void |
428870ff | 3575 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3576 | { |
d6320ddb | 3577 | int c; |
9babb374 | 3578 | |
572e2857 | 3579 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3580 | ASSERT(spa_config_held(nvd->vdev_spa, |
3581 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3582 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3583 | |
d6320ddb | 3584 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3585 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3586 | |
572e2857 | 3587 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3588 | /* |
572e2857 | 3589 | * Restore the persistent vdev state |
9babb374 | 3590 | */ |
428870ff | 3591 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3592 | nvd->vdev_faulted = ovd->vdev_faulted; |
3593 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3594 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3595 | } |
3596 | } | |
3597 | ||
572e2857 BB |
3598 | /* |
3599 | * Determine if a log device has valid content. If the vdev was | |
3600 | * removed or faulted in the MOS config then we know that | |
3601 | * the content on the log device has already been written to the pool. | |
3602 | */ | |
3603 | boolean_t | |
3604 | vdev_log_state_valid(vdev_t *vd) | |
3605 | { | |
d6320ddb BB |
3606 | int c; |
3607 | ||
572e2857 BB |
3608 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3609 | !vd->vdev_removed) | |
3610 | return (B_TRUE); | |
3611 | ||
d6320ddb | 3612 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3613 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3614 | return (B_TRUE); | |
3615 | ||
3616 | return (B_FALSE); | |
3617 | } | |
3618 | ||
9babb374 BB |
3619 | /* |
3620 | * Expand a vdev if possible. | |
3621 | */ | |
3622 | void | |
3623 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3624 | { | |
3625 | ASSERT(vd->vdev_top == vd); | |
3626 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3627 | ||
3628 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3629 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3630 | vdev_config_dirty(vd); | |
3631 | } | |
3632 | } | |
428870ff BB |
3633 | |
3634 | /* | |
3635 | * Split a vdev. | |
3636 | */ | |
3637 | void | |
3638 | vdev_split(vdev_t *vd) | |
3639 | { | |
3640 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3641 | ||
3642 | vdev_remove_child(pvd, vd); | |
3643 | vdev_compact_children(pvd); | |
3644 | ||
3645 | cvd = pvd->vdev_child[0]; | |
3646 | if (pvd->vdev_children == 1) { | |
3647 | vdev_remove_parent(cvd); | |
3648 | cvd->vdev_splitting = B_TRUE; | |
3649 | } | |
3650 | vdev_propagate_state(cvd); | |
3651 | } | |
c28b2279 | 3652 | |
cc92e9d0 GW |
3653 | void |
3654 | vdev_deadman(vdev_t *vd) | |
3655 | { | |
3656 | int c; | |
3657 | ||
3658 | for (c = 0; c < vd->vdev_children; c++) { | |
3659 | vdev_t *cvd = vd->vdev_child[c]; | |
3660 | ||
3661 | vdev_deadman(cvd); | |
3662 | } | |
3663 | ||
3664 | if (vd->vdev_ops->vdev_op_leaf) { | |
3665 | vdev_queue_t *vq = &vd->vdev_queue; | |
3666 | ||
3667 | mutex_enter(&vq->vq_lock); | |
e8b96c60 | 3668 | if (avl_numnodes(&vq->vq_active_tree) > 0) { |
cc92e9d0 GW |
3669 | spa_t *spa = vd->vdev_spa; |
3670 | zio_t *fio; | |
3671 | uint64_t delta; | |
3672 | ||
3673 | /* | |
3674 | * Look at the head of all the pending queues, | |
3675 | * if any I/O has been outstanding for longer than | |
3676 | * the spa_deadman_synctime we log a zevent. | |
3677 | */ | |
e8b96c60 | 3678 | fio = avl_first(&vq->vq_active_tree); |
cb682a17 MA |
3679 | delta = gethrtime() - fio->io_timestamp; |
3680 | if (delta > spa_deadman_synctime(spa)) { | |
3681 | zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " | |
3682 | "delta %lluns, last io %lluns", | |
cc92e9d0 GW |
3683 | fio->io_timestamp, delta, |
3684 | vq->vq_io_complete_ts); | |
3685 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, | |
3686 | spa, vd, fio, 0, 0); | |
3687 | } | |
3688 | } | |
3689 | mutex_exit(&vq->vq_lock); | |
3690 | } | |
3691 | } | |
3692 | ||
c28b2279 BB |
3693 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3694 | EXPORT_SYMBOL(vdev_fault); | |
3695 | EXPORT_SYMBOL(vdev_degrade); | |
3696 | EXPORT_SYMBOL(vdev_online); | |
3697 | EXPORT_SYMBOL(vdev_offline); | |
3698 | EXPORT_SYMBOL(vdev_clear); | |
4ea3f864 | 3699 | /* BEGIN CSTYLED */ |
b8bcca18 MA |
3700 | module_param(metaslabs_per_vdev, int, 0644); |
3701 | MODULE_PARM_DESC(metaslabs_per_vdev, | |
3702 | "Divide added vdev into approximately (but no more than) this number " | |
3703 | "of metaslabs"); | |
4ea3f864 | 3704 | /* END CSTYLED */ |
c28b2279 | 3705 | #endif |