]>
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. |
3541dc6d | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
ea04106b | 25 | * Copyright (c) 2011, 2014 by Delphix. All rights reserved. |
34dc7c2f BB |
26 | */ |
27 | ||
34dc7c2f BB |
28 | #include <sys/zfs_context.h> |
29 | #include <sys/fm/fs/zfs.h> | |
30 | #include <sys/spa.h> | |
31 | #include <sys/spa_impl.h> | |
32 | #include <sys/dmu.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/vdev_impl.h> | |
35 | #include <sys/uberblock_impl.h> | |
36 | #include <sys/metaslab.h> | |
37 | #include <sys/metaslab_impl.h> | |
38 | #include <sys/space_map.h> | |
ea04106b | 39 | #include <sys/space_reftree.h> |
34dc7c2f BB |
40 | #include <sys/zio.h> |
41 | #include <sys/zap.h> | |
42 | #include <sys/fs/zfs.h> | |
b128c09f | 43 | #include <sys/arc.h> |
9babb374 | 44 | #include <sys/zil.h> |
428870ff | 45 | #include <sys/dsl_scan.h> |
6c285672 | 46 | #include <sys/zvol.h> |
34dc7c2f | 47 | |
ea04106b AX |
48 | /* |
49 | * When a vdev is added, it will be divided into approximately (but no | |
50 | * more than) this number of metaslabs. | |
51 | */ | |
52 | int metaslabs_per_vdev = 200; | |
53 | ||
34dc7c2f BB |
54 | /* |
55 | * Virtual device management. | |
56 | */ | |
57 | ||
58 | static vdev_ops_t *vdev_ops_table[] = { | |
59 | &vdev_root_ops, | |
60 | &vdev_raidz_ops, | |
61 | &vdev_mirror_ops, | |
62 | &vdev_replacing_ops, | |
63 | &vdev_spare_ops, | |
64 | &vdev_disk_ops, | |
65 | &vdev_file_ops, | |
66 | &vdev_missing_ops, | |
428870ff | 67 | &vdev_hole_ops, |
34dc7c2f BB |
68 | NULL |
69 | }; | |
70 | ||
34dc7c2f BB |
71 | /* |
72 | * Given a vdev type, return the appropriate ops vector. | |
73 | */ | |
74 | static vdev_ops_t * | |
75 | vdev_getops(const char *type) | |
76 | { | |
77 | vdev_ops_t *ops, **opspp; | |
78 | ||
79 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
80 | if (strcmp(ops->vdev_op_type, type) == 0) | |
81 | break; | |
82 | ||
83 | return (ops); | |
84 | } | |
85 | ||
86 | /* | |
87 | * Default asize function: return the MAX of psize with the asize of | |
88 | * all children. This is what's used by anything other than RAID-Z. | |
89 | */ | |
90 | uint64_t | |
91 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
92 | { | |
93 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
94 | uint64_t csize; | |
d6320ddb | 95 | int c; |
34dc7c2f | 96 | |
d6320ddb | 97 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
98 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
99 | asize = MAX(asize, csize); | |
100 | } | |
101 | ||
102 | return (asize); | |
103 | } | |
104 | ||
105 | /* | |
9babb374 BB |
106 | * Get the minimum allocatable size. We define the allocatable size as |
107 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
108 | * replace or attach devices which don't have the same physical size but | |
109 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
110 | */ |
111 | uint64_t | |
9babb374 | 112 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 113 | { |
9babb374 | 114 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 115 | |
9babb374 | 116 | /* |
1bd201e7 | 117 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
118 | * just return our own asize. |
119 | */ | |
120 | if (pvd == NULL) | |
121 | return (vd->vdev_asize); | |
34dc7c2f BB |
122 | |
123 | /* | |
9babb374 BB |
124 | * The top-level vdev just returns the allocatable size rounded |
125 | * to the nearest metaslab. | |
34dc7c2f | 126 | */ |
9babb374 BB |
127 | if (vd == vd->vdev_top) |
128 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 129 | |
9babb374 BB |
130 | /* |
131 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
132 | * so each child must provide at least 1/Nth of its asize. | |
133 | */ | |
134 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
135 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 136 | |
9babb374 BB |
137 | return (pvd->vdev_min_asize); |
138 | } | |
139 | ||
140 | void | |
141 | vdev_set_min_asize(vdev_t *vd) | |
142 | { | |
d6320ddb | 143 | int c; |
9babb374 | 144 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 145 | |
d6320ddb | 146 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 147 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
148 | } |
149 | ||
150 | vdev_t * | |
151 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
152 | { | |
153 | vdev_t *rvd = spa->spa_root_vdev; | |
154 | ||
b128c09f | 155 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 156 | |
b128c09f BB |
157 | if (vdev < rvd->vdev_children) { |
158 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 159 | return (rvd->vdev_child[vdev]); |
b128c09f | 160 | } |
34dc7c2f BB |
161 | |
162 | return (NULL); | |
163 | } | |
164 | ||
165 | vdev_t * | |
166 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
167 | { | |
34dc7c2f | 168 | vdev_t *mvd; |
d6320ddb | 169 | int c; |
34dc7c2f BB |
170 | |
171 | if (vd->vdev_guid == guid) | |
172 | return (vd); | |
173 | ||
d6320ddb | 174 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
175 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
176 | NULL) | |
177 | return (mvd); | |
178 | ||
179 | return (NULL); | |
180 | } | |
181 | ||
182 | void | |
183 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
184 | { | |
185 | size_t oldsize, newsize; | |
186 | uint64_t id = cvd->vdev_id; | |
187 | vdev_t **newchild; | |
188 | ||
b128c09f | 189 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
190 | ASSERT(cvd->vdev_parent == NULL); |
191 | ||
192 | cvd->vdev_parent = pvd; | |
193 | ||
194 | if (pvd == NULL) | |
195 | return; | |
196 | ||
197 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
198 | ||
199 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
200 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
201 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
202 | ||
ea04106b | 203 | newchild = kmem_alloc(newsize, KM_SLEEP); |
34dc7c2f BB |
204 | if (pvd->vdev_child != NULL) { |
205 | bcopy(pvd->vdev_child, newchild, oldsize); | |
206 | kmem_free(pvd->vdev_child, oldsize); | |
207 | } | |
208 | ||
209 | pvd->vdev_child = newchild; | |
210 | pvd->vdev_child[id] = cvd; | |
211 | ||
212 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
213 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
214 | ||
215 | /* | |
216 | * Walk up all ancestors to update guid sum. | |
217 | */ | |
218 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
219 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
220 | } |
221 | ||
222 | void | |
223 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
224 | { | |
225 | int c; | |
226 | uint_t id = cvd->vdev_id; | |
227 | ||
228 | ASSERT(cvd->vdev_parent == pvd); | |
229 | ||
230 | if (pvd == NULL) | |
231 | return; | |
232 | ||
233 | ASSERT(id < pvd->vdev_children); | |
234 | ASSERT(pvd->vdev_child[id] == cvd); | |
235 | ||
236 | pvd->vdev_child[id] = NULL; | |
237 | cvd->vdev_parent = NULL; | |
238 | ||
239 | for (c = 0; c < pvd->vdev_children; c++) | |
240 | if (pvd->vdev_child[c]) | |
241 | break; | |
242 | ||
243 | if (c == pvd->vdev_children) { | |
244 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
245 | pvd->vdev_child = NULL; | |
246 | pvd->vdev_children = 0; | |
247 | } | |
248 | ||
249 | /* | |
250 | * Walk up all ancestors to update guid sum. | |
251 | */ | |
252 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
253 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
254 | } |
255 | ||
256 | /* | |
257 | * Remove any holes in the child array. | |
258 | */ | |
259 | void | |
260 | vdev_compact_children(vdev_t *pvd) | |
261 | { | |
262 | vdev_t **newchild, *cvd; | |
263 | int oldc = pvd->vdev_children; | |
9babb374 | 264 | int newc; |
d6320ddb | 265 | int c; |
34dc7c2f | 266 | |
b128c09f | 267 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 268 | |
d6320ddb | 269 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
270 | if (pvd->vdev_child[c]) |
271 | newc++; | |
272 | ||
ea04106b | 273 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); |
34dc7c2f | 274 | |
d6320ddb | 275 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
276 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
277 | newchild[newc] = cvd; | |
278 | cvd->vdev_id = newc++; | |
279 | } | |
280 | } | |
281 | ||
282 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
283 | pvd->vdev_child = newchild; | |
284 | pvd->vdev_children = newc; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Allocate and minimally initialize a vdev_t. | |
289 | */ | |
428870ff | 290 | vdev_t * |
34dc7c2f BB |
291 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
292 | { | |
293 | vdev_t *vd; | |
d6320ddb | 294 | int t; |
34dc7c2f | 295 | |
ea04106b | 296 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); |
34dc7c2f BB |
297 | |
298 | if (spa->spa_root_vdev == NULL) { | |
299 | ASSERT(ops == &vdev_root_ops); | |
300 | spa->spa_root_vdev = vd; | |
3541dc6d | 301 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
302 | } |
303 | ||
428870ff | 304 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
305 | if (spa->spa_root_vdev == vd) { |
306 | /* | |
307 | * The root vdev's guid will also be the pool guid, | |
308 | * which must be unique among all pools. | |
309 | */ | |
428870ff | 310 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
311 | } else { |
312 | /* | |
313 | * Any other vdev's guid must be unique within the pool. | |
314 | */ | |
428870ff | 315 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
316 | } |
317 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
318 | } | |
319 | ||
320 | vd->vdev_spa = spa; | |
321 | vd->vdev_id = id; | |
322 | vd->vdev_guid = guid; | |
323 | vd->vdev_guid_sum = guid; | |
324 | vd->vdev_ops = ops; | |
325 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 326 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f | 327 | |
98f72a53 BB |
328 | list_link_init(&vd->vdev_config_dirty_node); |
329 | list_link_init(&vd->vdev_state_dirty_node); | |
34dc7c2f BB |
330 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); |
331 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 332 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
d6320ddb | 333 | for (t = 0; t < DTL_TYPES; t++) { |
ea04106b | 334 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL, |
fb5f0bc8 BB |
335 | &vd->vdev_dtl_lock); |
336 | } | |
34dc7c2f BB |
337 | txg_list_create(&vd->vdev_ms_list, |
338 | offsetof(struct metaslab, ms_txg_node)); | |
339 | txg_list_create(&vd->vdev_dtl_list, | |
340 | offsetof(struct vdev, vdev_dtl_node)); | |
341 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
342 | vdev_queue_init(vd); | |
343 | vdev_cache_init(vd); | |
344 | ||
345 | return (vd); | |
346 | } | |
347 | ||
348 | /* | |
349 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
350 | * creating a new vdev or loading an existing one - the behavior is slightly | |
351 | * different for each case. | |
352 | */ | |
353 | int | |
354 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
355 | int alloctype) | |
356 | { | |
357 | vdev_ops_t *ops; | |
358 | char *type; | |
359 | uint64_t guid = 0, islog, nparity; | |
360 | vdev_t *vd; | |
361 | ||
b128c09f | 362 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
363 | |
364 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
a08ee875 | 365 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
366 | |
367 | if ((ops = vdev_getops(type)) == NULL) | |
a08ee875 | 368 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
369 | |
370 | /* | |
371 | * If this is a load, get the vdev guid from the nvlist. | |
372 | * Otherwise, vdev_alloc_common() will generate one for us. | |
373 | */ | |
374 | if (alloctype == VDEV_ALLOC_LOAD) { | |
375 | uint64_t label_id; | |
376 | ||
377 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
378 | label_id != id) | |
a08ee875 | 379 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
380 | |
381 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
a08ee875 | 382 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
383 | } else if (alloctype == VDEV_ALLOC_SPARE) { |
384 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
a08ee875 | 385 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
386 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { |
387 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
a08ee875 | 388 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
389 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
390 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
a08ee875 | 391 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
392 | } |
393 | ||
394 | /* | |
395 | * The first allocated vdev must be of type 'root'. | |
396 | */ | |
397 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
a08ee875 | 398 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
399 | |
400 | /* | |
401 | * Determine whether we're a log vdev. | |
402 | */ | |
403 | islog = 0; | |
404 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
405 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
a08ee875 | 406 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 407 | |
428870ff | 408 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
a08ee875 | 409 | return (SET_ERROR(ENOTSUP)); |
428870ff | 410 | |
34dc7c2f BB |
411 | /* |
412 | * Set the nparity property for RAID-Z vdevs. | |
413 | */ | |
414 | nparity = -1ULL; | |
415 | if (ops == &vdev_raidz_ops) { | |
416 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
417 | &nparity) == 0) { | |
428870ff | 418 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
a08ee875 | 419 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 420 | /* |
45d1cae3 BB |
421 | * Previous versions could only support 1 or 2 parity |
422 | * device. | |
34dc7c2f | 423 | */ |
45d1cae3 BB |
424 | if (nparity > 1 && |
425 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
a08ee875 | 426 | return (SET_ERROR(ENOTSUP)); |
45d1cae3 BB |
427 | if (nparity > 2 && |
428 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
a08ee875 | 429 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
430 | } else { |
431 | /* | |
432 | * We require the parity to be specified for SPAs that | |
433 | * support multiple parity levels. | |
434 | */ | |
45d1cae3 | 435 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
a08ee875 | 436 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
437 | /* |
438 | * Otherwise, we default to 1 parity device for RAID-Z. | |
439 | */ | |
440 | nparity = 1; | |
441 | } | |
442 | } else { | |
443 | nparity = 0; | |
444 | } | |
445 | ASSERT(nparity != -1ULL); | |
446 | ||
447 | vd = vdev_alloc_common(spa, id, guid, ops); | |
448 | ||
449 | vd->vdev_islog = islog; | |
450 | vd->vdev_nparity = nparity; | |
451 | ||
452 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
453 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
454 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
455 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
456 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
457 | &vd->vdev_physpath) == 0) | |
458 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
459 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
460 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
461 | |
462 | /* | |
463 | * Set the whole_disk property. If it's not specified, leave the value | |
464 | * as -1. | |
465 | */ | |
466 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
467 | &vd->vdev_wholedisk) != 0) | |
468 | vd->vdev_wholedisk = -1ULL; | |
469 | ||
470 | /* | |
471 | * Look for the 'not present' flag. This will only be set if the device | |
472 | * was not present at the time of import. | |
473 | */ | |
9babb374 BB |
474 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
475 | &vd->vdev_not_present); | |
34dc7c2f BB |
476 | |
477 | /* | |
478 | * Get the alignment requirement. | |
479 | */ | |
480 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
481 | ||
428870ff BB |
482 | /* |
483 | * Retrieve the vdev creation time. | |
484 | */ | |
485 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
486 | &vd->vdev_crtxg); | |
487 | ||
34dc7c2f BB |
488 | /* |
489 | * If we're a top-level vdev, try to load the allocation parameters. | |
490 | */ | |
428870ff BB |
491 | if (parent && !parent->vdev_parent && |
492 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
493 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
494 | &vd->vdev_ms_array); | |
495 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
496 | &vd->vdev_ms_shift); | |
497 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
498 | &vd->vdev_asize); | |
428870ff BB |
499 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
500 | &vd->vdev_removing); | |
501 | } | |
502 | ||
5ffb9d1d | 503 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
504 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
505 | alloctype == VDEV_ALLOC_ADD || | |
506 | alloctype == VDEV_ALLOC_SPLIT || | |
507 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
508 | vd->vdev_mg = metaslab_group_create(islog ? | |
509 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
510 | } |
511 | ||
512 | /* | |
513 | * If we're a leaf vdev, try to load the DTL object and other state. | |
514 | */ | |
b128c09f | 515 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
516 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
517 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
518 | if (alloctype == VDEV_ALLOC_LOAD) { |
519 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
ea04106b | 520 | &vd->vdev_dtl_object); |
b128c09f BB |
521 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
522 | &vd->vdev_unspare); | |
523 | } | |
9babb374 BB |
524 | |
525 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
526 | uint64_t spare = 0; | |
527 | ||
528 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
529 | &spare) == 0 && spare) | |
530 | spa_spare_add(vd); | |
531 | } | |
532 | ||
34dc7c2f BB |
533 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
534 | &vd->vdev_offline); | |
b128c09f | 535 | |
a08ee875 LG |
536 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, |
537 | &vd->vdev_resilver_txg); | |
572e2857 | 538 | |
34dc7c2f BB |
539 | /* |
540 | * When importing a pool, we want to ignore the persistent fault | |
541 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
542 | * valid in the current context. Local vdevs will |
543 | * remain in the faulted state. | |
34dc7c2f | 544 | */ |
428870ff | 545 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
546 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
547 | &vd->vdev_faulted); | |
548 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
549 | &vd->vdev_degraded); | |
550 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
551 | &vd->vdev_removed); | |
428870ff BB |
552 | |
553 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
554 | char *aux; | |
555 | ||
556 | vd->vdev_label_aux = | |
557 | VDEV_AUX_ERR_EXCEEDED; | |
558 | if (nvlist_lookup_string(nv, | |
559 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
560 | strcmp(aux, "external") == 0) | |
561 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
562 | } | |
34dc7c2f BB |
563 | } |
564 | } | |
565 | ||
566 | /* | |
567 | * Add ourselves to the parent's list of children. | |
568 | */ | |
569 | vdev_add_child(parent, vd); | |
570 | ||
571 | *vdp = vd; | |
572 | ||
573 | return (0); | |
574 | } | |
575 | ||
576 | void | |
577 | vdev_free(vdev_t *vd) | |
578 | { | |
d6320ddb | 579 | int c, t; |
34dc7c2f BB |
580 | spa_t *spa = vd->vdev_spa; |
581 | ||
582 | /* | |
583 | * vdev_free() implies closing the vdev first. This is simpler than | |
584 | * trying to ensure complicated semantics for all callers. | |
585 | */ | |
586 | vdev_close(vd); | |
587 | ||
b128c09f | 588 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 589 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
590 | |
591 | /* | |
592 | * Free all children. | |
593 | */ | |
d6320ddb | 594 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
595 | vdev_free(vd->vdev_child[c]); |
596 | ||
597 | ASSERT(vd->vdev_child == NULL); | |
598 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
599 | ||
600 | /* | |
601 | * Discard allocation state. | |
602 | */ | |
428870ff | 603 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 604 | vdev_metaslab_fini(vd); |
428870ff BB |
605 | metaslab_group_destroy(vd->vdev_mg); |
606 | } | |
34dc7c2f | 607 | |
c06d4368 AX |
608 | ASSERT0(vd->vdev_stat.vs_space); |
609 | ASSERT0(vd->vdev_stat.vs_dspace); | |
610 | ASSERT0(vd->vdev_stat.vs_alloc); | |
34dc7c2f BB |
611 | |
612 | /* | |
613 | * Remove this vdev from its parent's child list. | |
614 | */ | |
615 | vdev_remove_child(vd->vdev_parent, vd); | |
616 | ||
617 | ASSERT(vd->vdev_parent == NULL); | |
618 | ||
619 | /* | |
620 | * Clean up vdev structure. | |
621 | */ | |
622 | vdev_queue_fini(vd); | |
623 | vdev_cache_fini(vd); | |
624 | ||
625 | if (vd->vdev_path) | |
626 | spa_strfree(vd->vdev_path); | |
627 | if (vd->vdev_devid) | |
628 | spa_strfree(vd->vdev_devid); | |
629 | if (vd->vdev_physpath) | |
630 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
631 | if (vd->vdev_fru) |
632 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
633 | |
634 | if (vd->vdev_isspare) | |
635 | spa_spare_remove(vd); | |
636 | if (vd->vdev_isl2cache) | |
637 | spa_l2cache_remove(vd); | |
638 | ||
639 | txg_list_destroy(&vd->vdev_ms_list); | |
640 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 641 | |
34dc7c2f | 642 | mutex_enter(&vd->vdev_dtl_lock); |
ea04106b | 643 | space_map_close(vd->vdev_dtl_sm); |
d6320ddb | 644 | for (t = 0; t < DTL_TYPES; t++) { |
ea04106b AX |
645 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); |
646 | range_tree_destroy(vd->vdev_dtl[t]); | |
fb5f0bc8 | 647 | } |
34dc7c2f | 648 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 649 | |
34dc7c2f BB |
650 | mutex_destroy(&vd->vdev_dtl_lock); |
651 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 652 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
653 | |
654 | if (vd == spa->spa_root_vdev) | |
655 | spa->spa_root_vdev = NULL; | |
656 | ||
657 | kmem_free(vd, sizeof (vdev_t)); | |
658 | } | |
659 | ||
660 | /* | |
661 | * Transfer top-level vdev state from svd to tvd. | |
662 | */ | |
663 | static void | |
664 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
665 | { | |
666 | spa_t *spa = svd->vdev_spa; | |
667 | metaslab_t *msp; | |
668 | vdev_t *vd; | |
669 | int t; | |
670 | ||
671 | ASSERT(tvd == tvd->vdev_top); | |
672 | ||
673 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
674 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
675 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
676 | ||
677 | svd->vdev_ms_array = 0; | |
678 | svd->vdev_ms_shift = 0; | |
679 | svd->vdev_ms_count = 0; | |
680 | ||
5ffb9d1d GW |
681 | if (tvd->vdev_mg) |
682 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
683 | tvd->vdev_mg = svd->vdev_mg; |
684 | tvd->vdev_ms = svd->vdev_ms; | |
685 | ||
686 | svd->vdev_mg = NULL; | |
687 | svd->vdev_ms = NULL; | |
688 | ||
689 | if (tvd->vdev_mg != NULL) | |
690 | tvd->vdev_mg->mg_vd = tvd; | |
691 | ||
692 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
693 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
694 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
695 | ||
696 | svd->vdev_stat.vs_alloc = 0; | |
697 | svd->vdev_stat.vs_space = 0; | |
698 | svd->vdev_stat.vs_dspace = 0; | |
699 | ||
700 | for (t = 0; t < TXG_SIZE; t++) { | |
701 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
702 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
703 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
704 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
705 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
706 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
707 | } | |
708 | ||
b128c09f | 709 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
710 | vdev_config_clean(svd); |
711 | vdev_config_dirty(tvd); | |
712 | } | |
713 | ||
b128c09f BB |
714 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
715 | vdev_state_clean(svd); | |
716 | vdev_state_dirty(tvd); | |
717 | } | |
718 | ||
34dc7c2f BB |
719 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
720 | svd->vdev_deflate_ratio = 0; | |
721 | ||
722 | tvd->vdev_islog = svd->vdev_islog; | |
723 | svd->vdev_islog = 0; | |
724 | } | |
725 | ||
726 | static void | |
727 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
728 | { | |
d6320ddb BB |
729 | int c; |
730 | ||
34dc7c2f BB |
731 | if (vd == NULL) |
732 | return; | |
733 | ||
734 | vd->vdev_top = tvd; | |
735 | ||
d6320ddb | 736 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
737 | vdev_top_update(tvd, vd->vdev_child[c]); |
738 | } | |
739 | ||
740 | /* | |
741 | * Add a mirror/replacing vdev above an existing vdev. | |
742 | */ | |
743 | vdev_t * | |
744 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
745 | { | |
746 | spa_t *spa = cvd->vdev_spa; | |
747 | vdev_t *pvd = cvd->vdev_parent; | |
748 | vdev_t *mvd; | |
749 | ||
b128c09f | 750 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
751 | |
752 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
753 | ||
754 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 755 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 756 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
34dc7c2f BB |
757 | mvd->vdev_ashift = cvd->vdev_ashift; |
758 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 759 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
760 | |
761 | vdev_remove_child(pvd, cvd); | |
762 | vdev_add_child(pvd, mvd); | |
763 | cvd->vdev_id = mvd->vdev_children; | |
764 | vdev_add_child(mvd, cvd); | |
765 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
766 | ||
767 | if (mvd == mvd->vdev_top) | |
768 | vdev_top_transfer(cvd, mvd); | |
769 | ||
770 | return (mvd); | |
771 | } | |
772 | ||
773 | /* | |
774 | * Remove a 1-way mirror/replacing vdev from the tree. | |
775 | */ | |
776 | void | |
777 | vdev_remove_parent(vdev_t *cvd) | |
778 | { | |
779 | vdev_t *mvd = cvd->vdev_parent; | |
780 | vdev_t *pvd = mvd->vdev_parent; | |
781 | ||
b128c09f | 782 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
783 | |
784 | ASSERT(mvd->vdev_children == 1); | |
785 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
786 | mvd->vdev_ops == &vdev_replacing_ops || | |
787 | mvd->vdev_ops == &vdev_spare_ops); | |
788 | cvd->vdev_ashift = mvd->vdev_ashift; | |
789 | ||
790 | vdev_remove_child(mvd, cvd); | |
791 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 792 | |
34dc7c2f | 793 | /* |
b128c09f BB |
794 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
795 | * Otherwise, we could have detached an offline device, and when we | |
796 | * go to import the pool we'll think we have two top-level vdevs, | |
797 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 798 | */ |
fb5f0bc8 BB |
799 | if (mvd->vdev_top == mvd) { |
800 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 801 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
802 | cvd->vdev_guid += guid_delta; |
803 | cvd->vdev_guid_sum += guid_delta; | |
ea04106b AX |
804 | |
805 | /* | |
806 | * If pool not set for autoexpand, we need to also preserve | |
807 | * mvd's asize to prevent automatic expansion of cvd. | |
808 | * Otherwise if we are adjusting the mirror by attaching and | |
809 | * detaching children of non-uniform sizes, the mirror could | |
810 | * autoexpand, unexpectedly requiring larger devices to | |
811 | * re-establish the mirror. | |
812 | */ | |
813 | if (!cvd->vdev_spa->spa_autoexpand) | |
814 | cvd->vdev_asize = mvd->vdev_asize; | |
fb5f0bc8 | 815 | } |
b128c09f BB |
816 | cvd->vdev_id = mvd->vdev_id; |
817 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
818 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
819 | ||
820 | if (cvd == cvd->vdev_top) | |
821 | vdev_top_transfer(mvd, cvd); | |
822 | ||
823 | ASSERT(mvd->vdev_children == 0); | |
824 | vdev_free(mvd); | |
825 | } | |
826 | ||
827 | int | |
828 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
829 | { | |
830 | spa_t *spa = vd->vdev_spa; | |
831 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
832 | uint64_t m; |
833 | uint64_t oldc = vd->vdev_ms_count; | |
834 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
835 | metaslab_t **mspp; | |
836 | int error; | |
837 | ||
428870ff BB |
838 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
839 | ||
840 | /* | |
841 | * This vdev is not being allocated from yet or is a hole. | |
842 | */ | |
843 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
844 | return (0); |
845 | ||
428870ff BB |
846 | ASSERT(!vd->vdev_ishole); |
847 | ||
9babb374 BB |
848 | /* |
849 | * Compute the raidz-deflation ratio. Note, we hard-code | |
850 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
851 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
852 | * or we will inconsistently account for existing bp's. | |
853 | */ | |
854 | vd->vdev_deflate_ratio = (1 << 17) / | |
855 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
856 | ||
34dc7c2f BB |
857 | ASSERT(oldc <= newc); |
858 | ||
ea04106b | 859 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
34dc7c2f BB |
860 | |
861 | if (oldc != 0) { | |
862 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
863 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
864 | } | |
865 | ||
866 | vd->vdev_ms = mspp; | |
867 | vd->vdev_ms_count = newc; | |
868 | ||
869 | for (m = oldc; m < newc; m++) { | |
ea04106b AX |
870 | uint64_t object = 0; |
871 | ||
34dc7c2f | 872 | if (txg == 0) { |
34dc7c2f | 873 | error = dmu_read(mos, vd->vdev_ms_array, |
9babb374 BB |
874 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
875 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
876 | if (error) |
877 | return (error); | |
34dc7c2f | 878 | } |
ea04106b AX |
879 | |
880 | error = metaslab_init(vd->vdev_mg, m, object, txg, | |
881 | &(vd->vdev_ms[m])); | |
882 | if (error) | |
883 | return (error); | |
34dc7c2f BB |
884 | } |
885 | ||
428870ff BB |
886 | if (txg == 0) |
887 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
888 | ||
889 | /* | |
890 | * If the vdev is being removed we don't activate | |
891 | * the metaslabs since we want to ensure that no new | |
892 | * allocations are performed on this device. | |
893 | */ | |
894 | if (oldc == 0 && !vd->vdev_removing) | |
895 | metaslab_group_activate(vd->vdev_mg); | |
896 | ||
897 | if (txg == 0) | |
898 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
899 | ||
34dc7c2f BB |
900 | return (0); |
901 | } | |
902 | ||
903 | void | |
904 | vdev_metaslab_fini(vdev_t *vd) | |
905 | { | |
906 | uint64_t m; | |
907 | uint64_t count = vd->vdev_ms_count; | |
908 | ||
909 | if (vd->vdev_ms != NULL) { | |
428870ff | 910 | metaslab_group_passivate(vd->vdev_mg); |
ea04106b AX |
911 | for (m = 0; m < count; m++) { |
912 | metaslab_t *msp = vd->vdev_ms[m]; | |
913 | ||
914 | if (msp != NULL) | |
915 | metaslab_fini(msp); | |
916 | } | |
34dc7c2f BB |
917 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); |
918 | vd->vdev_ms = NULL; | |
919 | } | |
920dd524 ED |
920 | |
921 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
34dc7c2f BB |
922 | } |
923 | ||
b128c09f BB |
924 | typedef struct vdev_probe_stats { |
925 | boolean_t vps_readable; | |
926 | boolean_t vps_writeable; | |
927 | int vps_flags; | |
b128c09f BB |
928 | } vdev_probe_stats_t; |
929 | ||
930 | static void | |
931 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 932 | { |
fb5f0bc8 | 933 | spa_t *spa = zio->io_spa; |
d164b209 | 934 | vdev_t *vd = zio->io_vd; |
b128c09f | 935 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
936 | |
937 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
938 | |
939 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
940 | if (zio->io_error == 0) |
941 | vps->vps_readable = 1; | |
fb5f0bc8 | 942 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 943 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
944 | zio->io_offset, zio->io_size, zio->io_data, |
945 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
946 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
947 | } else { | |
948 | zio_buf_free(zio->io_data, zio->io_size); | |
949 | } | |
950 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
951 | if (zio->io_error == 0) |
952 | vps->vps_writeable = 1; | |
953 | zio_buf_free(zio->io_data, zio->io_size); | |
954 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 955 | zio_t *pio; |
b128c09f BB |
956 | |
957 | vd->vdev_cant_read |= !vps->vps_readable; | |
958 | vd->vdev_cant_write |= !vps->vps_writeable; | |
959 | ||
960 | if (vdev_readable(vd) && | |
fb5f0bc8 | 961 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
962 | zio->io_error = 0; |
963 | } else { | |
964 | ASSERT(zio->io_error != 0); | |
965 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 966 | spa, vd, NULL, 0, 0); |
a08ee875 | 967 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f | 968 | } |
d164b209 BB |
969 | |
970 | mutex_enter(&vd->vdev_probe_lock); | |
971 | ASSERT(vd->vdev_probe_zio == zio); | |
972 | vd->vdev_probe_zio = NULL; | |
973 | mutex_exit(&vd->vdev_probe_lock); | |
974 | ||
975 | while ((pio = zio_walk_parents(zio)) != NULL) | |
976 | if (!vdev_accessible(vd, pio)) | |
a08ee875 | 977 | pio->io_error = SET_ERROR(ENXIO); |
d164b209 | 978 | |
b128c09f BB |
979 | kmem_free(vps, sizeof (*vps)); |
980 | } | |
981 | } | |
34dc7c2f | 982 | |
b128c09f | 983 | /* |
a08ee875 LG |
984 | * Determine whether this device is accessible. |
985 | * | |
986 | * Read and write to several known locations: the pad regions of each | |
987 | * vdev label but the first, which we leave alone in case it contains | |
988 | * a VTOC. | |
b128c09f BB |
989 | */ |
990 | zio_t * | |
d164b209 | 991 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
992 | { |
993 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
994 | vdev_probe_stats_t *vps = NULL; |
995 | zio_t *pio; | |
d6320ddb | 996 | int l; |
d164b209 BB |
997 | |
998 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 999 | |
d164b209 BB |
1000 | /* |
1001 | * Don't probe the probe. | |
1002 | */ | |
1003 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
1004 | return (NULL); | |
b128c09f | 1005 | |
d164b209 BB |
1006 | /* |
1007 | * To prevent 'probe storms' when a device fails, we create | |
1008 | * just one probe i/o at a time. All zios that want to probe | |
1009 | * this vdev will become parents of the probe io. | |
1010 | */ | |
1011 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 1012 | |
d164b209 | 1013 | if ((pio = vd->vdev_probe_zio) == NULL) { |
ea04106b | 1014 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); |
d164b209 BB |
1015 | |
1016 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1017 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1018 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1019 | |
1020 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1021 | /* | |
1022 | * vdev_cant_read and vdev_cant_write can only | |
1023 | * transition from TRUE to FALSE when we have the | |
1024 | * SCL_ZIO lock as writer; otherwise they can only | |
1025 | * transition from FALSE to TRUE. This ensures that | |
1026 | * any zio looking at these values can assume that | |
1027 | * failures persist for the life of the I/O. That's | |
1028 | * important because when a device has intermittent | |
1029 | * connectivity problems, we want to ensure that | |
1030 | * they're ascribed to the device (ENXIO) and not | |
1031 | * the zio (EIO). | |
1032 | * | |
1033 | * Since we hold SCL_ZIO as writer here, clear both | |
1034 | * values so the probe can reevaluate from first | |
1035 | * principles. | |
1036 | */ | |
1037 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1038 | vd->vdev_cant_read = B_FALSE; | |
1039 | vd->vdev_cant_write = B_FALSE; | |
1040 | } | |
1041 | ||
1042 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1043 | vdev_probe_done, vps, | |
1044 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1045 | ||
428870ff BB |
1046 | /* |
1047 | * We can't change the vdev state in this context, so we | |
1048 | * kick off an async task to do it on our behalf. | |
1049 | */ | |
d164b209 BB |
1050 | if (zio != NULL) { |
1051 | vd->vdev_probe_wanted = B_TRUE; | |
1052 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1053 | } | |
b128c09f BB |
1054 | } |
1055 | ||
d164b209 BB |
1056 | if (zio != NULL) |
1057 | zio_add_child(zio, pio); | |
b128c09f | 1058 | |
d164b209 | 1059 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1060 | |
d164b209 BB |
1061 | if (vps == NULL) { |
1062 | ASSERT(zio != NULL); | |
1063 | return (NULL); | |
1064 | } | |
b128c09f | 1065 | |
d6320ddb | 1066 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1067 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1068 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
1069 | offsetof(vdev_label_t, vl_pad2)), |
1070 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
1071 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1072 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1073 | } | |
1074 | ||
d164b209 BB |
1075 | if (zio == NULL) |
1076 | return (pio); | |
1077 | ||
1078 | zio_nowait(pio); | |
1079 | return (NULL); | |
34dc7c2f BB |
1080 | } |
1081 | ||
45d1cae3 BB |
1082 | static void |
1083 | vdev_open_child(void *arg) | |
1084 | { | |
1085 | vdev_t *vd = arg; | |
1086 | ||
1087 | vd->vdev_open_thread = curthread; | |
1088 | vd->vdev_open_error = vdev_open(vd); | |
1089 | vd->vdev_open_thread = NULL; | |
1090 | } | |
1091 | ||
6c285672 | 1092 | static boolean_t |
428870ff BB |
1093 | vdev_uses_zvols(vdev_t *vd) |
1094 | { | |
d6320ddb BB |
1095 | int c; |
1096 | ||
6c285672 JL |
1097 | #ifdef _KERNEL |
1098 | if (zvol_is_zvol(vd->vdev_path)) | |
428870ff | 1099 | return (B_TRUE); |
6c285672 JL |
1100 | #endif |
1101 | ||
d6320ddb | 1102 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1103 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1104 | return (B_TRUE); | |
6c285672 | 1105 | |
428870ff BB |
1106 | return (B_FALSE); |
1107 | } | |
1108 | ||
45d1cae3 BB |
1109 | void |
1110 | vdev_open_children(vdev_t *vd) | |
1111 | { | |
1112 | taskq_t *tq; | |
1113 | int children = vd->vdev_children; | |
d6320ddb | 1114 | int c; |
45d1cae3 | 1115 | |
428870ff BB |
1116 | /* |
1117 | * in order to handle pools on top of zvols, do the opens | |
1118 | * in a single thread so that the same thread holds the | |
1119 | * spa_namespace_lock | |
1120 | */ | |
1121 | if (vdev_uses_zvols(vd)) { | |
d6320ddb | 1122 | for (c = 0; c < children; c++) |
428870ff BB |
1123 | vd->vdev_child[c]->vdev_open_error = |
1124 | vdev_open(vd->vdev_child[c]); | |
1125 | return; | |
1126 | } | |
45d1cae3 BB |
1127 | tq = taskq_create("vdev_open", children, minclsyspri, |
1128 | children, children, TASKQ_PREPOPULATE); | |
1129 | ||
d6320ddb | 1130 | for (c = 0; c < children; c++) |
45d1cae3 | 1131 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], |
d6320ddb | 1132 | TQ_SLEEP) != 0); |
45d1cae3 BB |
1133 | |
1134 | taskq_destroy(tq); | |
1135 | } | |
1136 | ||
34dc7c2f BB |
1137 | /* |
1138 | * Prepare a virtual device for access. | |
1139 | */ | |
1140 | int | |
1141 | vdev_open(vdev_t *vd) | |
1142 | { | |
fb5f0bc8 | 1143 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1144 | int error; |
34dc7c2f | 1145 | uint64_t osize = 0; |
1bd201e7 CS |
1146 | uint64_t max_osize = 0; |
1147 | uint64_t asize, max_asize, psize; | |
34dc7c2f | 1148 | uint64_t ashift = 0; |
d6320ddb | 1149 | int c; |
34dc7c2f | 1150 | |
45d1cae3 BB |
1151 | ASSERT(vd->vdev_open_thread == curthread || |
1152 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1153 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1154 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1155 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1156 | ||
34dc7c2f | 1157 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1158 | vd->vdev_cant_read = B_FALSE; |
1159 | vd->vdev_cant_write = B_FALSE; | |
1160 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1161 | |
428870ff BB |
1162 | /* |
1163 | * If this vdev is not removed, check its fault status. If it's | |
1164 | * faulted, bail out of the open. | |
1165 | */ | |
34dc7c2f BB |
1166 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1167 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1168 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1169 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1170 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1171 | vd->vdev_label_aux); |
a08ee875 | 1172 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1173 | } else if (vd->vdev_offline) { |
1174 | ASSERT(vd->vdev_children == 0); | |
1175 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
a08ee875 | 1176 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1177 | } |
1178 | ||
1bd201e7 | 1179 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1180 | |
428870ff BB |
1181 | /* |
1182 | * Reset the vdev_reopening flag so that we actually close | |
1183 | * the vdev on error. | |
1184 | */ | |
1185 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1186 | if (zio_injection_enabled && error == 0) |
9babb374 | 1187 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1188 | |
1189 | if (error) { | |
1190 | if (vd->vdev_removed && | |
1191 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1192 | vd->vdev_removed = B_FALSE; | |
1193 | ||
1194 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1195 | vd->vdev_stat.vs_aux); | |
1196 | return (error); | |
1197 | } | |
1198 | ||
1199 | vd->vdev_removed = B_FALSE; | |
1200 | ||
428870ff BB |
1201 | /* |
1202 | * Recheck the faulted flag now that we have confirmed that | |
1203 | * the vdev is accessible. If we're faulted, bail. | |
1204 | */ | |
1205 | if (vd->vdev_faulted) { | |
1206 | ASSERT(vd->vdev_children == 0); | |
1207 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1208 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1209 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1210 | vd->vdev_label_aux); | |
a08ee875 | 1211 | return (SET_ERROR(ENXIO)); |
428870ff BB |
1212 | } |
1213 | ||
34dc7c2f BB |
1214 | if (vd->vdev_degraded) { |
1215 | ASSERT(vd->vdev_children == 0); | |
1216 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1217 | VDEV_AUX_ERR_EXCEEDED); | |
1218 | } else { | |
428870ff | 1219 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1220 | } |
1221 | ||
428870ff BB |
1222 | /* |
1223 | * For hole or missing vdevs we just return success. | |
1224 | */ | |
1225 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1226 | return (0); | |
1227 | ||
d6320ddb | 1228 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1229 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1230 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1231 | VDEV_AUX_NONE); | |
1232 | break; | |
1233 | } | |
9babb374 | 1234 | } |
34dc7c2f BB |
1235 | |
1236 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1237 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1238 | |
1239 | if (vd->vdev_children == 0) { | |
1240 | if (osize < SPA_MINDEVSIZE) { | |
1241 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1242 | VDEV_AUX_TOO_SMALL); | |
a08ee875 | 1243 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1244 | } |
1245 | psize = osize; | |
1246 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1247 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1248 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1249 | } else { |
1250 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1251 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1252 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1253 | VDEV_AUX_TOO_SMALL); | |
a08ee875 | 1254 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1255 | } |
1256 | psize = 0; | |
1257 | asize = osize; | |
1bd201e7 | 1258 | max_asize = max_osize; |
34dc7c2f BB |
1259 | } |
1260 | ||
1261 | vd->vdev_psize = psize; | |
1262 | ||
9babb374 BB |
1263 | /* |
1264 | * Make sure the allocatable size hasn't shrunk. | |
1265 | */ | |
1266 | if (asize < vd->vdev_min_asize) { | |
1267 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1268 | VDEV_AUX_BAD_LABEL); | |
a08ee875 | 1269 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
1270 | } |
1271 | ||
34dc7c2f BB |
1272 | if (vd->vdev_asize == 0) { |
1273 | /* | |
1274 | * This is the first-ever open, so use the computed values. | |
c06d4368 | 1275 | * For compatibility, a different ashift can be requested. |
34dc7c2f BB |
1276 | */ |
1277 | vd->vdev_asize = asize; | |
1bd201e7 | 1278 | vd->vdev_max_asize = max_asize; |
c06d4368 AX |
1279 | if (vd->vdev_ashift == 0) |
1280 | vd->vdev_ashift = ashift; | |
34dc7c2f BB |
1281 | } else { |
1282 | /* | |
32a9872b GW |
1283 | * Detect if the alignment requirement has increased. |
1284 | * We don't want to make the pool unavailable, just | |
1285 | * post an event instead. | |
34dc7c2f | 1286 | */ |
32a9872b GW |
1287 | if (ashift > vd->vdev_top->vdev_ashift && |
1288 | vd->vdev_ops->vdev_op_leaf) { | |
1289 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
1290 | spa, vd, NULL, 0, 0); | |
34dc7c2f | 1291 | } |
32a9872b | 1292 | |
1bd201e7 | 1293 | vd->vdev_max_asize = max_asize; |
9babb374 | 1294 | } |
34dc7c2f | 1295 | |
9babb374 BB |
1296 | /* |
1297 | * If all children are healthy and the asize has increased, | |
1298 | * then we've experienced dynamic LUN growth. If automatic | |
1299 | * expansion is enabled then use the additional space. | |
1300 | */ | |
1301 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1302 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1303 | vd->vdev_asize = asize; | |
34dc7c2f | 1304 | |
9babb374 | 1305 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1306 | |
1307 | /* | |
1308 | * Ensure we can issue some IO before declaring the | |
1309 | * vdev open for business. | |
1310 | */ | |
b128c09f BB |
1311 | if (vd->vdev_ops->vdev_op_leaf && |
1312 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1313 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1314 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1315 | return (error); |
1316 | } | |
1317 | ||
34dc7c2f | 1318 | /* |
b128c09f | 1319 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1320 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1321 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1322 | */ |
fb5f0bc8 BB |
1323 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1324 | vdev_resilver_needed(vd, NULL, NULL)) | |
1325 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1326 | |
1327 | return (0); | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Called once the vdevs are all opened, this routine validates the label | |
1332 | * contents. This needs to be done before vdev_load() so that we don't | |
1333 | * inadvertently do repair I/Os to the wrong device. | |
1334 | * | |
c7f2d69d GW |
1335 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1336 | * if the machine crashed during a re-guid the new guid might have been written | |
1337 | * to all of the vdev labels, but not the cached config. The strict check | |
1338 | * will be performed when the pool is opened again using the mos config. | |
1339 | * | |
34dc7c2f BB |
1340 | * This function will only return failure if one of the vdevs indicates that it |
1341 | * has since been destroyed or exported. This is only possible if | |
1342 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1343 | * will be updated but the function will return 0. | |
1344 | */ | |
1345 | int | |
c7f2d69d | 1346 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1347 | { |
1348 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1349 | nvlist_t *label; |
428870ff | 1350 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1351 | uint64_t state; |
d6320ddb | 1352 | int c; |
34dc7c2f | 1353 | |
d6320ddb | 1354 | for (c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1355 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
a08ee875 | 1356 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
1357 | |
1358 | /* | |
1359 | * If the device has already failed, or was marked offline, don't do | |
1360 | * any further validation. Otherwise, label I/O will fail and we will | |
1361 | * overwrite the previous state. | |
1362 | */ | |
b128c09f | 1363 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1364 | uint64_t aux_guid = 0; |
1365 | nvlist_t *nvl; | |
c06d4368 AX |
1366 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? |
1367 | spa_last_synced_txg(spa) : -1ULL; | |
34dc7c2f | 1368 | |
3bc7e0fb | 1369 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { |
34dc7c2f BB |
1370 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
1371 | VDEV_AUX_BAD_LABEL); | |
1372 | return (0); | |
1373 | } | |
1374 | ||
428870ff BB |
1375 | /* |
1376 | * Determine if this vdev has been split off into another | |
1377 | * pool. If so, then refuse to open it. | |
1378 | */ | |
1379 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1380 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1381 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1382 | VDEV_AUX_SPLIT_POOL); | |
1383 | nvlist_free(label); | |
1384 | return (0); | |
1385 | } | |
1386 | ||
c7f2d69d GW |
1387 | if (strict && (nvlist_lookup_uint64(label, |
1388 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1389 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1390 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1391 | VDEV_AUX_CORRUPT_DATA); | |
1392 | nvlist_free(label); | |
1393 | return (0); | |
1394 | } | |
1395 | ||
428870ff BB |
1396 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1397 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1398 | &aux_guid) != 0) | |
1399 | aux_guid = 0; | |
1400 | ||
b128c09f BB |
1401 | /* |
1402 | * If this vdev just became a top-level vdev because its | |
1403 | * sibling was detached, it will have adopted the parent's | |
1404 | * vdev guid -- but the label may or may not be on disk yet. | |
1405 | * Fortunately, either version of the label will have the | |
1406 | * same top guid, so if we're a top-level vdev, we can | |
1407 | * safely compare to that instead. | |
428870ff BB |
1408 | * |
1409 | * If we split this vdev off instead, then we also check the | |
1410 | * original pool's guid. We don't want to consider the vdev | |
1411 | * corrupt if it is partway through a split operation. | |
b128c09f | 1412 | */ |
34dc7c2f | 1413 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1414 | &guid) != 0 || |
1415 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1416 | &top_guid) != 0 || | |
428870ff | 1417 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1418 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1419 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1420 | VDEV_AUX_CORRUPT_DATA); | |
1421 | nvlist_free(label); | |
1422 | return (0); | |
1423 | } | |
1424 | ||
1425 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1426 | &state) != 0) { | |
1427 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1428 | VDEV_AUX_CORRUPT_DATA); | |
1429 | nvlist_free(label); | |
1430 | return (0); | |
1431 | } | |
1432 | ||
1433 | nvlist_free(label); | |
1434 | ||
45d1cae3 | 1435 | /* |
572e2857 | 1436 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1437 | * state of the pool. |
1438 | */ | |
572e2857 | 1439 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1440 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f | 1441 | state != POOL_STATE_ACTIVE) |
a08ee875 | 1442 | return (SET_ERROR(EBADF)); |
34dc7c2f | 1443 | |
b128c09f BB |
1444 | /* |
1445 | * If we were able to open and validate a vdev that was | |
1446 | * previously marked permanently unavailable, clear that state | |
1447 | * now. | |
1448 | */ | |
1449 | if (vd->vdev_not_present) | |
1450 | vd->vdev_not_present = 0; | |
1451 | } | |
34dc7c2f BB |
1452 | |
1453 | return (0); | |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * Close a virtual device. | |
1458 | */ | |
1459 | void | |
1460 | vdev_close(vdev_t *vd) | |
1461 | { | |
428870ff | 1462 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1463 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1464 | |
1465 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1466 | ||
428870ff BB |
1467 | /* |
1468 | * If our parent is reopening, then we are as well, unless we are | |
1469 | * going offline. | |
1470 | */ | |
1471 | if (pvd != NULL && pvd->vdev_reopening) | |
1472 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1473 | ||
34dc7c2f BB |
1474 | vd->vdev_ops->vdev_op_close(vd); |
1475 | ||
1476 | vdev_cache_purge(vd); | |
1477 | ||
1478 | /* | |
9babb374 | 1479 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1480 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1481 | * it's still faulted. | |
1482 | */ | |
1483 | vd->vdev_prevstate = vd->vdev_state; | |
1484 | ||
1485 | if (vd->vdev_offline) | |
1486 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1487 | else | |
1488 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1489 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1490 | } | |
1491 | ||
428870ff BB |
1492 | void |
1493 | vdev_hold(vdev_t *vd) | |
1494 | { | |
1495 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1496 | int c; |
428870ff BB |
1497 | |
1498 | ASSERT(spa_is_root(spa)); | |
1499 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1500 | return; | |
1501 | ||
d6320ddb | 1502 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1503 | vdev_hold(vd->vdev_child[c]); |
1504 | ||
1505 | if (vd->vdev_ops->vdev_op_leaf) | |
1506 | vd->vdev_ops->vdev_op_hold(vd); | |
1507 | } | |
1508 | ||
1509 | void | |
1510 | vdev_rele(vdev_t *vd) | |
1511 | { | |
d6320ddb | 1512 | int c; |
428870ff | 1513 | |
d6320ddb BB |
1514 | ASSERT(spa_is_root(vd->vdev_spa)); |
1515 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1516 | vdev_rele(vd->vdev_child[c]); |
1517 | ||
1518 | if (vd->vdev_ops->vdev_op_leaf) | |
1519 | vd->vdev_ops->vdev_op_rele(vd); | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1524 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1525 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1526 | * If the leaf has never been opened then open it, as usual. | |
1527 | */ | |
34dc7c2f BB |
1528 | void |
1529 | vdev_reopen(vdev_t *vd) | |
1530 | { | |
1531 | spa_t *spa = vd->vdev_spa; | |
1532 | ||
b128c09f | 1533 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1534 | |
428870ff BB |
1535 | /* set the reopening flag unless we're taking the vdev offline */ |
1536 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1537 | vdev_close(vd); |
1538 | (void) vdev_open(vd); | |
1539 | ||
1540 | /* | |
1541 | * Call vdev_validate() here to make sure we have the same device. | |
1542 | * Otherwise, a device with an invalid label could be successfully | |
1543 | * opened in response to vdev_reopen(). | |
1544 | */ | |
b128c09f BB |
1545 | if (vd->vdev_aux) { |
1546 | (void) vdev_validate_aux(vd); | |
1547 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1548 | vd->vdev_aux == &spa->spa_l2cache && |
1549 | !l2arc_vdev_present(vd)) | |
1550 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1551 | } else { |
c06d4368 | 1552 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1553 | } |
34dc7c2f BB |
1554 | |
1555 | /* | |
1556 | * Reassess parent vdev's health. | |
1557 | */ | |
1558 | vdev_propagate_state(vd); | |
1559 | } | |
1560 | ||
1561 | int | |
1562 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1563 | { | |
1564 | int error; | |
1565 | ||
1566 | /* | |
1567 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1568 | * For a create, however, we want to fail the request if | |
1569 | * there are any components we can't open. | |
1570 | */ | |
1571 | error = vdev_open(vd); | |
1572 | ||
1573 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1574 | vdev_close(vd); | |
1575 | return (error ? error : ENXIO); | |
1576 | } | |
1577 | ||
1578 | /* | |
ea04106b | 1579 | * Recursively load DTLs and initialize all labels. |
34dc7c2f | 1580 | */ |
ea04106b AX |
1581 | if ((error = vdev_dtl_load(vd)) != 0 || |
1582 | (error = vdev_label_init(vd, txg, isreplacing ? | |
34dc7c2f BB |
1583 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { |
1584 | vdev_close(vd); | |
1585 | return (error); | |
1586 | } | |
1587 | ||
1588 | return (0); | |
1589 | } | |
1590 | ||
34dc7c2f | 1591 | void |
9babb374 | 1592 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1593 | { |
1594 | /* | |
ea04106b | 1595 | * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. |
34dc7c2f | 1596 | */ |
ea04106b | 1597 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); |
34dc7c2f | 1598 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); |
34dc7c2f BB |
1599 | } |
1600 | ||
1601 | void | |
1602 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1603 | { | |
1604 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1605 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1606 | ASSERT(ISP2(flags)); |
572e2857 | 1607 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1608 | |
1609 | if (flags & VDD_METASLAB) | |
1610 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1611 | ||
1612 | if (flags & VDD_DTL) | |
1613 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1614 | ||
1615 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1616 | } | |
1617 | ||
ea04106b AX |
1618 | void |
1619 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1620 | { | |
1621 | int c; | |
1622 | ||
1623 | for (c = 0; c < vd->vdev_children; c++) | |
1624 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); | |
1625 | ||
1626 | if (vd->vdev_ops->vdev_op_leaf) | |
1627 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1628 | } | |
1629 | ||
fb5f0bc8 BB |
1630 | /* |
1631 | * DTLs. | |
1632 | * | |
1633 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1634 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1635 | * |
1636 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1637 | * | |
1638 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1639 | * | |
1640 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1641 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1642 | * txgs that was scrubbed. | |
1643 | * | |
1644 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1645 | * persistent errors or just some device being offline. | |
1646 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1647 | * maintained; it's only computed when needed, typically to | |
1648 | * determine whether a device can be detached. | |
1649 | * | |
1650 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1651 | * either has the data or it doesn't. | |
1652 | * | |
1653 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1654 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1655 | * if any child is less than fully replicated, then so is its parent. | |
1656 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1657 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1658 | * those are the txgs we don't have enough replication to read. For example, | |
1659 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1660 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1661 | * two child DTL_MISSING maps. | |
1662 | * | |
1663 | * It should be clear from the above that to compute the DTLs and outage maps | |
1664 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1665 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1666 | * a configuration change, we generate all other DTLs from first principles. | |
1667 | */ | |
34dc7c2f | 1668 | void |
fb5f0bc8 | 1669 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1670 | { |
ea04106b | 1671 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1672 | |
1673 | ASSERT(t < DTL_TYPES); | |
1674 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1675 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1676 | |
ea04106b AX |
1677 | mutex_enter(rt->rt_lock); |
1678 | if (!range_tree_contains(rt, txg, size)) | |
1679 | range_tree_add(rt, txg, size); | |
1680 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1681 | } |
1682 | ||
fb5f0bc8 BB |
1683 | boolean_t |
1684 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1685 | { |
ea04106b | 1686 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 | 1687 | boolean_t dirty = B_FALSE; |
34dc7c2f | 1688 | |
fb5f0bc8 BB |
1689 | ASSERT(t < DTL_TYPES); |
1690 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f | 1691 | |
ea04106b AX |
1692 | mutex_enter(rt->rt_lock); |
1693 | if (range_tree_space(rt) != 0) | |
1694 | dirty = range_tree_contains(rt, txg, size); | |
1695 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1696 | |
1697 | return (dirty); | |
1698 | } | |
1699 | ||
fb5f0bc8 BB |
1700 | boolean_t |
1701 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1702 | { | |
ea04106b | 1703 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1704 | boolean_t empty; |
1705 | ||
ea04106b AX |
1706 | mutex_enter(rt->rt_lock); |
1707 | empty = (range_tree_space(rt) == 0); | |
1708 | mutex_exit(rt->rt_lock); | |
fb5f0bc8 BB |
1709 | |
1710 | return (empty); | |
1711 | } | |
1712 | ||
a08ee875 LG |
1713 | /* |
1714 | * Returns the lowest txg in the DTL range. | |
1715 | */ | |
1716 | static uint64_t | |
1717 | vdev_dtl_min(vdev_t *vd) | |
1718 | { | |
ea04106b | 1719 | range_seg_t *rs; |
a08ee875 LG |
1720 | |
1721 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
ea04106b | 1722 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
a08ee875 LG |
1723 | ASSERT0(vd->vdev_children); |
1724 | ||
ea04106b AX |
1725 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1726 | return (rs->rs_start - 1); | |
a08ee875 LG |
1727 | } |
1728 | ||
1729 | /* | |
1730 | * Returns the highest txg in the DTL. | |
1731 | */ | |
1732 | static uint64_t | |
1733 | vdev_dtl_max(vdev_t *vd) | |
1734 | { | |
ea04106b | 1735 | range_seg_t *rs; |
a08ee875 LG |
1736 | |
1737 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
ea04106b | 1738 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
a08ee875 LG |
1739 | ASSERT0(vd->vdev_children); |
1740 | ||
ea04106b AX |
1741 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1742 | return (rs->rs_end); | |
a08ee875 LG |
1743 | } |
1744 | ||
1745 | /* | |
1746 | * Determine if a resilvering vdev should remove any DTL entries from | |
1747 | * its range. If the vdev was resilvering for the entire duration of the | |
1748 | * scan then it should excise that range from its DTLs. Otherwise, this | |
1749 | * vdev is considered partially resilvered and should leave its DTL | |
1750 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
1751 | * excise the DTLs. | |
1752 | */ | |
1753 | static boolean_t | |
1754 | vdev_dtl_should_excise(vdev_t *vd) | |
1755 | { | |
1756 | spa_t *spa = vd->vdev_spa; | |
1757 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1758 | ||
1759 | ASSERT0(scn->scn_phys.scn_errors); | |
1760 | ASSERT0(vd->vdev_children); | |
1761 | ||
1762 | if (vd->vdev_resilver_txg == 0 || | |
ea04106b | 1763 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) |
a08ee875 LG |
1764 | return (B_TRUE); |
1765 | ||
1766 | /* | |
1767 | * When a resilver is initiated the scan will assign the scn_max_txg | |
1768 | * value to the highest txg value that exists in all DTLs. If this | |
1769 | * device's max DTL is not part of this scan (i.e. it is not in | |
1770 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
1771 | * for excision. | |
1772 | */ | |
1773 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
1774 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
1775 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
1776 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
1777 | return (B_TRUE); | |
1778 | } | |
1779 | return (B_FALSE); | |
1780 | } | |
1781 | ||
34dc7c2f BB |
1782 | /* |
1783 | * Reassess DTLs after a config change or scrub completion. | |
1784 | */ | |
1785 | void | |
1786 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1787 | { | |
1788 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1789 | avl_tree_t reftree; |
d6320ddb | 1790 | int c, t, minref; |
34dc7c2f | 1791 | |
fb5f0bc8 | 1792 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1793 | |
d6320ddb | 1794 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1795 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1796 | scrub_txg, scrub_done); | |
1797 | ||
428870ff | 1798 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1799 | return; |
1800 | ||
1801 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1802 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1803 | ||
34dc7c2f | 1804 | mutex_enter(&vd->vdev_dtl_lock); |
a08ee875 LG |
1805 | |
1806 | /* | |
1807 | * If we've completed a scan cleanly then determine | |
1808 | * if this vdev should remove any DTLs. We only want to | |
1809 | * excise regions on vdevs that were available during | |
1810 | * the entire duration of this scan. | |
1811 | */ | |
b128c09f | 1812 | if (scrub_txg != 0 && |
428870ff | 1813 | (spa->spa_scrub_started || |
a08ee875 LG |
1814 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && |
1815 | vdev_dtl_should_excise(vd)) { | |
b128c09f BB |
1816 | /* |
1817 | * We completed a scrub up to scrub_txg. If we | |
1818 | * did it without rebooting, then the scrub dtl | |
1819 | * will be valid, so excise the old region and | |
1820 | * fold in the scrub dtl. Otherwise, leave the | |
1821 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1822 | * |
1823 | * There's little trick here: to excise the beginning | |
1824 | * of the DTL_MISSING map, we put it into a reference | |
1825 | * tree and then add a segment with refcnt -1 that | |
1826 | * covers the range [0, scrub_txg). This means | |
1827 | * that each txg in that range has refcnt -1 or 0. | |
1828 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1829 | * entries in the range [0, scrub_txg) will have a | |
1830 | * positive refcnt -- either 1 or 2. We then convert | |
1831 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1832 | */ |
ea04106b AX |
1833 | space_reftree_create(&reftree); |
1834 | space_reftree_add_map(&reftree, | |
1835 | vd->vdev_dtl[DTL_MISSING], 1); | |
1836 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
1837 | space_reftree_add_map(&reftree, | |
1838 | vd->vdev_dtl[DTL_SCRUB], 2); | |
1839 | space_reftree_generate_map(&reftree, | |
1840 | vd->vdev_dtl[DTL_MISSING], 1); | |
1841 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1842 | } |
ea04106b AX |
1843 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1844 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1845 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1846 | if (scrub_done) |
ea04106b AX |
1847 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1848 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
fb5f0bc8 | 1849 | if (!vdev_readable(vd)) |
ea04106b | 1850 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); |
fb5f0bc8 | 1851 | else |
ea04106b AX |
1852 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], |
1853 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
a08ee875 LG |
1854 | |
1855 | /* | |
1856 | * If the vdev was resilvering and no longer has any | |
1857 | * DTLs then reset its resilvering flag. | |
1858 | */ | |
1859 | if (vd->vdev_resilver_txg != 0 && | |
ea04106b AX |
1860 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && |
1861 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) | |
a08ee875 LG |
1862 | vd->vdev_resilver_txg = 0; |
1863 | ||
34dc7c2f | 1864 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1865 | |
34dc7c2f BB |
1866 | if (txg != 0) |
1867 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1868 | return; | |
1869 | } | |
1870 | ||
34dc7c2f | 1871 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1872 | for (t = 0; t < DTL_TYPES; t++) { |
ea04106b AX |
1873 | int c; |
1874 | ||
428870ff BB |
1875 | /* account for child's outage in parent's missing map */ |
1876 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1877 | if (t == DTL_SCRUB) |
1878 | continue; /* leaf vdevs only */ | |
1879 | if (t == DTL_PARTIAL) | |
1880 | minref = 1; /* i.e. non-zero */ | |
1881 | else if (vd->vdev_nparity != 0) | |
1882 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1883 | else | |
1884 | minref = vd->vdev_children; /* any kind of mirror */ | |
ea04106b | 1885 | space_reftree_create(&reftree); |
d6320ddb | 1886 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1887 | vdev_t *cvd = vd->vdev_child[c]; |
1888 | mutex_enter(&cvd->vdev_dtl_lock); | |
ea04106b | 1889 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1890 | mutex_exit(&cvd->vdev_dtl_lock); |
1891 | } | |
ea04106b AX |
1892 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); |
1893 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1894 | } |
fb5f0bc8 | 1895 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1896 | } |
1897 | ||
ea04106b | 1898 | int |
34dc7c2f BB |
1899 | vdev_dtl_load(vdev_t *vd) |
1900 | { | |
1901 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1902 | objset_t *mos = spa->spa_meta_objset; |
ea04106b AX |
1903 | int error = 0; |
1904 | int c; | |
34dc7c2f | 1905 | |
ea04106b AX |
1906 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { |
1907 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 1908 | |
ea04106b AX |
1909 | error = space_map_open(&vd->vdev_dtl_sm, mos, |
1910 | vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); | |
1911 | if (error) | |
1912 | return (error); | |
1913 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f | 1914 | |
ea04106b | 1915 | mutex_enter(&vd->vdev_dtl_lock); |
428870ff | 1916 | |
ea04106b AX |
1917 | /* |
1918 | * Now that we've opened the space_map we need to update | |
1919 | * the in-core DTL. | |
1920 | */ | |
1921 | space_map_update(vd->vdev_dtl_sm); | |
34dc7c2f | 1922 | |
ea04106b AX |
1923 | error = space_map_load(vd->vdev_dtl_sm, |
1924 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
1925 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 1926 | |
ea04106b AX |
1927 | return (error); |
1928 | } | |
1929 | ||
1930 | for (c = 0; c < vd->vdev_children; c++) { | |
1931 | error = vdev_dtl_load(vd->vdev_child[c]); | |
1932 | if (error != 0) | |
1933 | break; | |
1934 | } | |
34dc7c2f BB |
1935 | |
1936 | return (error); | |
1937 | } | |
1938 | ||
1939 | void | |
1940 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1941 | { | |
1942 | spa_t *spa = vd->vdev_spa; | |
ea04106b | 1943 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; |
34dc7c2f | 1944 | objset_t *mos = spa->spa_meta_objset; |
ea04106b AX |
1945 | range_tree_t *rtsync; |
1946 | kmutex_t rtlock; | |
34dc7c2f | 1947 | dmu_tx_t *tx; |
ea04106b | 1948 | uint64_t object = space_map_object(vd->vdev_dtl_sm); |
34dc7c2f | 1949 | |
428870ff | 1950 | ASSERT(!vd->vdev_ishole); |
ea04106b | 1951 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
428870ff | 1952 | |
34dc7c2f BB |
1953 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1954 | ||
ea04106b AX |
1955 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { |
1956 | mutex_enter(&vd->vdev_dtl_lock); | |
1957 | space_map_free(vd->vdev_dtl_sm, tx); | |
1958 | space_map_close(vd->vdev_dtl_sm); | |
1959 | vd->vdev_dtl_sm = NULL; | |
1960 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 1961 | dmu_tx_commit(tx); |
34dc7c2f BB |
1962 | return; |
1963 | } | |
1964 | ||
ea04106b AX |
1965 | if (vd->vdev_dtl_sm == NULL) { |
1966 | uint64_t new_object; | |
1967 | ||
1968 | new_object = space_map_alloc(mos, tx); | |
1969 | VERIFY3U(new_object, !=, 0); | |
1970 | ||
1971 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
1972 | 0, -1ULL, 0, &vd->vdev_dtl_lock)); | |
1973 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f BB |
1974 | } |
1975 | ||
ea04106b | 1976 | mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 1977 | |
ea04106b | 1978 | rtsync = range_tree_create(NULL, NULL, &rtlock); |
34dc7c2f | 1979 | |
ea04106b | 1980 | mutex_enter(&rtlock); |
34dc7c2f BB |
1981 | |
1982 | mutex_enter(&vd->vdev_dtl_lock); | |
ea04106b | 1983 | range_tree_walk(rt, range_tree_add, rtsync); |
34dc7c2f BB |
1984 | mutex_exit(&vd->vdev_dtl_lock); |
1985 | ||
ea04106b AX |
1986 | space_map_truncate(vd->vdev_dtl_sm, tx); |
1987 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
1988 | range_tree_vacate(rtsync, NULL, NULL); | |
34dc7c2f | 1989 | |
ea04106b | 1990 | range_tree_destroy(rtsync); |
34dc7c2f | 1991 | |
ea04106b AX |
1992 | mutex_exit(&rtlock); |
1993 | mutex_destroy(&rtlock); | |
34dc7c2f | 1994 | |
ea04106b AX |
1995 | /* |
1996 | * If the object for the space map has changed then dirty | |
1997 | * the top level so that we update the config. | |
1998 | */ | |
1999 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
2000 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
2001 | "new object %llu", txg, spa_name(spa), object, | |
2002 | space_map_object(vd->vdev_dtl_sm)); | |
2003 | vdev_config_dirty(vd->vdev_top); | |
2004 | } | |
34dc7c2f BB |
2005 | |
2006 | dmu_tx_commit(tx); | |
ea04106b AX |
2007 | |
2008 | mutex_enter(&vd->vdev_dtl_lock); | |
2009 | space_map_update(vd->vdev_dtl_sm); | |
2010 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f BB |
2011 | } |
2012 | ||
fb5f0bc8 BB |
2013 | /* |
2014 | * Determine whether the specified vdev can be offlined/detached/removed | |
2015 | * without losing data. | |
2016 | */ | |
2017 | boolean_t | |
2018 | vdev_dtl_required(vdev_t *vd) | |
2019 | { | |
2020 | spa_t *spa = vd->vdev_spa; | |
2021 | vdev_t *tvd = vd->vdev_top; | |
2022 | uint8_t cant_read = vd->vdev_cant_read; | |
2023 | boolean_t required; | |
2024 | ||
2025 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2026 | ||
2027 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2028 | return (B_TRUE); | |
2029 | ||
2030 | /* | |
2031 | * Temporarily mark the device as unreadable, and then determine | |
2032 | * whether this results in any DTL outages in the top-level vdev. | |
2033 | * If not, we can safely offline/detach/remove the device. | |
2034 | */ | |
2035 | vd->vdev_cant_read = B_TRUE; | |
2036 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2037 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2038 | vd->vdev_cant_read = cant_read; | |
2039 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2040 | ||
572e2857 BB |
2041 | if (!required && zio_injection_enabled) |
2042 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2043 | ||
fb5f0bc8 BB |
2044 | return (required); |
2045 | } | |
2046 | ||
b128c09f BB |
2047 | /* |
2048 | * Determine if resilver is needed, and if so the txg range. | |
2049 | */ | |
2050 | boolean_t | |
2051 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2052 | { | |
2053 | boolean_t needed = B_FALSE; | |
2054 | uint64_t thismin = UINT64_MAX; | |
2055 | uint64_t thismax = 0; | |
d6320ddb | 2056 | int c; |
b128c09f BB |
2057 | |
2058 | if (vd->vdev_children == 0) { | |
2059 | mutex_enter(&vd->vdev_dtl_lock); | |
ea04106b | 2060 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && |
fb5f0bc8 | 2061 | vdev_writeable(vd)) { |
b128c09f | 2062 | |
a08ee875 LG |
2063 | thismin = vdev_dtl_min(vd); |
2064 | thismax = vdev_dtl_max(vd); | |
b128c09f BB |
2065 | needed = B_TRUE; |
2066 | } | |
2067 | mutex_exit(&vd->vdev_dtl_lock); | |
2068 | } else { | |
d6320ddb | 2069 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2070 | vdev_t *cvd = vd->vdev_child[c]; |
2071 | uint64_t cmin, cmax; | |
2072 | ||
2073 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2074 | thismin = MIN(thismin, cmin); | |
2075 | thismax = MAX(thismax, cmax); | |
2076 | needed = B_TRUE; | |
2077 | } | |
2078 | } | |
2079 | } | |
2080 | ||
2081 | if (needed && minp) { | |
2082 | *minp = thismin; | |
2083 | *maxp = thismax; | |
2084 | } | |
2085 | return (needed); | |
2086 | } | |
2087 | ||
34dc7c2f BB |
2088 | void |
2089 | vdev_load(vdev_t *vd) | |
2090 | { | |
d6320ddb BB |
2091 | int c; |
2092 | ||
34dc7c2f BB |
2093 | /* |
2094 | * Recursively load all children. | |
2095 | */ | |
d6320ddb | 2096 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2097 | vdev_load(vd->vdev_child[c]); |
2098 | ||
2099 | /* | |
2100 | * If this is a top-level vdev, initialize its metaslabs. | |
2101 | */ | |
428870ff | 2102 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
2103 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
2104 | vdev_metaslab_init(vd, 0) != 0)) | |
2105 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2106 | VDEV_AUX_CORRUPT_DATA); | |
2107 | ||
2108 | /* | |
2109 | * If this is a leaf vdev, load its DTL. | |
2110 | */ | |
2111 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
2112 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2113 | VDEV_AUX_CORRUPT_DATA); | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2118 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2119 | * we make sure that we can open the underlying device, then try to read the | |
2120 | * label, and make sure that the label is sane and that it hasn't been | |
2121 | * repurposed to another pool. | |
2122 | */ | |
2123 | int | |
2124 | vdev_validate_aux(vdev_t *vd) | |
2125 | { | |
2126 | nvlist_t *label; | |
2127 | uint64_t guid, version; | |
2128 | uint64_t state; | |
2129 | ||
b128c09f BB |
2130 | if (!vdev_readable(vd)) |
2131 | return (0); | |
2132 | ||
3bc7e0fb | 2133 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { |
34dc7c2f BB |
2134 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
2135 | VDEV_AUX_CORRUPT_DATA); | |
2136 | return (-1); | |
2137 | } | |
2138 | ||
2139 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
9ae529ec | 2140 | !SPA_VERSION_IS_SUPPORTED(version) || |
34dc7c2f BB |
2141 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || |
2142 | guid != vd->vdev_guid || | |
2143 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2144 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2145 | VDEV_AUX_CORRUPT_DATA); | |
2146 | nvlist_free(label); | |
2147 | return (-1); | |
2148 | } | |
2149 | ||
2150 | /* | |
2151 | * We don't actually check the pool state here. If it's in fact in | |
2152 | * use by another pool, we update this fact on the fly when requested. | |
2153 | */ | |
2154 | nvlist_free(label); | |
2155 | return (0); | |
2156 | } | |
2157 | ||
428870ff BB |
2158 | void |
2159 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2160 | { | |
2161 | spa_t *spa = vd->vdev_spa; | |
2162 | objset_t *mos = spa->spa_meta_objset; | |
2163 | dmu_tx_t *tx; | |
ea04106b | 2164 | int m, i; |
428870ff BB |
2165 | |
2166 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
2167 | ||
428870ff | 2168 | if (vd->vdev_ms != NULL) { |
ea04106b AX |
2169 | metaslab_group_t *mg = vd->vdev_mg; |
2170 | ||
2171 | metaslab_group_histogram_verify(mg); | |
2172 | metaslab_class_histogram_verify(mg->mg_class); | |
2173 | ||
d6320ddb | 2174 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2175 | metaslab_t *msp = vd->vdev_ms[m]; |
2176 | ||
ea04106b | 2177 | if (msp == NULL || msp->ms_sm == NULL) |
428870ff BB |
2178 | continue; |
2179 | ||
ea04106b AX |
2180 | mutex_enter(&msp->ms_lock); |
2181 | /* | |
2182 | * If the metaslab was not loaded when the vdev | |
2183 | * was removed then the histogram accounting may | |
2184 | * not be accurate. Update the histogram information | |
2185 | * here so that we ensure that the metaslab group | |
2186 | * and metaslab class are up-to-date. | |
2187 | */ | |
2188 | metaslab_group_histogram_remove(mg, msp); | |
2189 | ||
2190 | VERIFY0(space_map_allocated(msp->ms_sm)); | |
2191 | space_map_free(msp->ms_sm, tx); | |
2192 | space_map_close(msp->ms_sm); | |
2193 | msp->ms_sm = NULL; | |
2194 | mutex_exit(&msp->ms_lock); | |
428870ff | 2195 | } |
ea04106b AX |
2196 | |
2197 | metaslab_group_histogram_verify(mg); | |
2198 | metaslab_class_histogram_verify(mg->mg_class); | |
2199 | for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) | |
2200 | ASSERT0(mg->mg_histogram[i]); | |
2201 | ||
428870ff BB |
2202 | } |
2203 | ||
2204 | if (vd->vdev_ms_array) { | |
2205 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2206 | vd->vdev_ms_array = 0; | |
428870ff BB |
2207 | } |
2208 | dmu_tx_commit(tx); | |
2209 | } | |
2210 | ||
34dc7c2f BB |
2211 | void |
2212 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2213 | { | |
2214 | metaslab_t *msp; | |
428870ff BB |
2215 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2216 | ||
2217 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2218 | |
c65aa5b2 | 2219 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2220 | metaslab_sync_done(msp, txg); |
428870ff BB |
2221 | |
2222 | if (reassess) | |
2223 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2224 | } |
2225 | ||
2226 | void | |
2227 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2228 | { | |
2229 | spa_t *spa = vd->vdev_spa; | |
2230 | vdev_t *lvd; | |
2231 | metaslab_t *msp; | |
2232 | dmu_tx_t *tx; | |
2233 | ||
428870ff BB |
2234 | ASSERT(!vd->vdev_ishole); |
2235 | ||
34dc7c2f BB |
2236 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2237 | ASSERT(vd == vd->vdev_top); | |
2238 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2239 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2240 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2241 | ASSERT(vd->vdev_ms_array != 0); | |
2242 | vdev_config_dirty(vd); | |
2243 | dmu_tx_commit(tx); | |
2244 | } | |
2245 | ||
428870ff BB |
2246 | /* |
2247 | * Remove the metadata associated with this vdev once it's empty. | |
2248 | */ | |
2249 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2250 | vdev_remove(vd, txg); | |
2251 | ||
34dc7c2f BB |
2252 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2253 | metaslab_sync(msp, txg); | |
2254 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2255 | } | |
2256 | ||
2257 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2258 | vdev_dtl_sync(lvd, txg); | |
2259 | ||
2260 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2261 | } | |
2262 | ||
2263 | uint64_t | |
2264 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2265 | { | |
2266 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2267 | } | |
2268 | ||
34dc7c2f BB |
2269 | /* |
2270 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2271 | * not be opened, and no I/O is attempted. | |
2272 | */ | |
2273 | int | |
428870ff | 2274 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2275 | { |
572e2857 | 2276 | vdev_t *vd, *tvd; |
34dc7c2f | 2277 | |
428870ff | 2278 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2279 | |
b128c09f BB |
2280 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2281 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2282 | |
34dc7c2f | 2283 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2284 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2285 | |
572e2857 BB |
2286 | tvd = vd->vdev_top; |
2287 | ||
428870ff BB |
2288 | /* |
2289 | * We don't directly use the aux state here, but if we do a | |
2290 | * vdev_reopen(), we need this value to be present to remember why we | |
2291 | * were faulted. | |
2292 | */ | |
2293 | vd->vdev_label_aux = aux; | |
2294 | ||
34dc7c2f BB |
2295 | /* |
2296 | * Faulted state takes precedence over degraded. | |
2297 | */ | |
428870ff | 2298 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2299 | vd->vdev_faulted = 1ULL; |
2300 | vd->vdev_degraded = 0ULL; | |
428870ff | 2301 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2302 | |
2303 | /* | |
428870ff BB |
2304 | * If this device has the only valid copy of the data, then |
2305 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2306 | */ |
572e2857 | 2307 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2308 | vd->vdev_degraded = 1ULL; |
2309 | vd->vdev_faulted = 0ULL; | |
2310 | ||
2311 | /* | |
2312 | * If we reopen the device and it's not dead, only then do we | |
2313 | * mark it degraded. | |
2314 | */ | |
572e2857 | 2315 | vdev_reopen(tvd); |
34dc7c2f | 2316 | |
428870ff BB |
2317 | if (vdev_readable(vd)) |
2318 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2319 | } |
2320 | ||
b128c09f | 2321 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2322 | } |
2323 | ||
2324 | /* | |
2325 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2326 | * user that something is wrong. The vdev continues to operate as normal as far | |
2327 | * as I/O is concerned. | |
2328 | */ | |
2329 | int | |
428870ff | 2330 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2331 | { |
b128c09f | 2332 | vdev_t *vd; |
34dc7c2f | 2333 | |
428870ff | 2334 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2335 | |
b128c09f BB |
2336 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2337 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2338 | |
34dc7c2f | 2339 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2340 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2341 | |
2342 | /* | |
2343 | * If the vdev is already faulted, then don't do anything. | |
2344 | */ | |
b128c09f BB |
2345 | if (vd->vdev_faulted || vd->vdev_degraded) |
2346 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2347 | |
2348 | vd->vdev_degraded = 1ULL; | |
2349 | if (!vdev_is_dead(vd)) | |
2350 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2351 | aux); |
34dc7c2f | 2352 | |
b128c09f | 2353 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2354 | } |
2355 | ||
2356 | /* | |
a08ee875 LG |
2357 | * Online the given vdev. |
2358 | * | |
2359 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2360 | * spare device should be detached when the device finishes resilvering. | |
2361 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2362 | * events are generated if the device fails to open. | |
34dc7c2f BB |
2363 | */ |
2364 | int | |
b128c09f | 2365 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2366 | { |
9babb374 | 2367 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 2368 | |
428870ff | 2369 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2370 | |
b128c09f BB |
2371 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2372 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2373 | |
2374 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2375 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2376 | |
9babb374 | 2377 | tvd = vd->vdev_top; |
34dc7c2f BB |
2378 | vd->vdev_offline = B_FALSE; |
2379 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2380 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2381 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2382 | |
2383 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2384 | if (!vd->vdev_aux) { | |
2385 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2386 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2387 | } | |
2388 | ||
2389 | vdev_reopen(tvd); | |
34dc7c2f BB |
2390 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2391 | ||
9babb374 BB |
2392 | if (!vd->vdev_aux) { |
2393 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2394 | pvd->vdev_expanding = B_FALSE; | |
2395 | } | |
2396 | ||
34dc7c2f BB |
2397 | if (newstate) |
2398 | *newstate = vd->vdev_state; | |
2399 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2400 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2401 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2402 | vd->vdev_parent->vdev_child[0] == vd) | |
2403 | vd->vdev_unspare = B_TRUE; | |
2404 | ||
9babb374 BB |
2405 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2406 | ||
2407 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2408 | if (vd->vdev_aux) | |
2409 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2410 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2411 | } | |
fb5f0bc8 | 2412 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2413 | } |
2414 | ||
428870ff BB |
2415 | static int |
2416 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2417 | { |
9babb374 | 2418 | vdev_t *vd, *tvd; |
428870ff BB |
2419 | int error = 0; |
2420 | uint64_t generation; | |
2421 | metaslab_group_t *mg; | |
34dc7c2f | 2422 | |
428870ff BB |
2423 | top: |
2424 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2425 | |
b128c09f BB |
2426 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2427 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2428 | |
2429 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2430 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2431 | |
9babb374 | 2432 | tvd = vd->vdev_top; |
428870ff BB |
2433 | mg = tvd->vdev_mg; |
2434 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2435 | |
34dc7c2f BB |
2436 | /* |
2437 | * If the device isn't already offline, try to offline it. | |
2438 | */ | |
2439 | if (!vd->vdev_offline) { | |
2440 | /* | |
fb5f0bc8 | 2441 | * If this device has the only valid copy of some data, |
9babb374 BB |
2442 | * don't allow it to be offlined. Log devices are always |
2443 | * expendable. | |
34dc7c2f | 2444 | */ |
9babb374 BB |
2445 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2446 | vdev_dtl_required(vd)) | |
b128c09f | 2447 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2448 | |
428870ff BB |
2449 | /* |
2450 | * If the top-level is a slog and it has had allocations | |
2451 | * then proceed. We check that the vdev's metaslab group | |
2452 | * is not NULL since it's possible that we may have just | |
2453 | * added this vdev but not yet initialized its metaslabs. | |
2454 | */ | |
2455 | if (tvd->vdev_islog && mg != NULL) { | |
2456 | /* | |
2457 | * Prevent any future allocations. | |
2458 | */ | |
2459 | metaslab_group_passivate(mg); | |
2460 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2461 | ||
2462 | error = spa_offline_log(spa); | |
2463 | ||
2464 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2465 | ||
2466 | /* | |
2467 | * Check to see if the config has changed. | |
2468 | */ | |
2469 | if (error || generation != spa->spa_config_generation) { | |
2470 | metaslab_group_activate(mg); | |
2471 | if (error) | |
2472 | return (spa_vdev_state_exit(spa, | |
2473 | vd, error)); | |
2474 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2475 | goto top; | |
2476 | } | |
c06d4368 | 2477 | ASSERT0(tvd->vdev_stat.vs_alloc); |
428870ff BB |
2478 | } |
2479 | ||
34dc7c2f BB |
2480 | /* |
2481 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2482 | * If the top-level vdev is a log device then just offline |
2483 | * it. Otherwise, if this action results in the top-level | |
2484 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2485 | */ |
2486 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2487 | vdev_reopen(tvd); |
2488 | ||
2489 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2490 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2491 | vd->vdev_offline = B_FALSE; |
9babb374 | 2492 | vdev_reopen(tvd); |
b128c09f | 2493 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2494 | } |
428870ff BB |
2495 | |
2496 | /* | |
2497 | * Add the device back into the metaslab rotor so that | |
2498 | * once we online the device it's open for business. | |
2499 | */ | |
2500 | if (tvd->vdev_islog && mg != NULL) | |
2501 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2502 | } |
2503 | ||
b128c09f | 2504 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2505 | |
428870ff BB |
2506 | return (spa_vdev_state_exit(spa, vd, 0)); |
2507 | } | |
9babb374 | 2508 | |
428870ff BB |
2509 | int |
2510 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2511 | { | |
2512 | int error; | |
9babb374 | 2513 | |
428870ff BB |
2514 | mutex_enter(&spa->spa_vdev_top_lock); |
2515 | error = vdev_offline_locked(spa, guid, flags); | |
2516 | mutex_exit(&spa->spa_vdev_top_lock); | |
2517 | ||
2518 | return (error); | |
34dc7c2f BB |
2519 | } |
2520 | ||
2521 | /* | |
2522 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2523 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2524 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2525 | */ |
2526 | void | |
b128c09f | 2527 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2528 | { |
b128c09f | 2529 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2530 | int c; |
b128c09f BB |
2531 | |
2532 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2533 | |
2534 | if (vd == NULL) | |
b128c09f | 2535 | vd = rvd; |
34dc7c2f BB |
2536 | |
2537 | vd->vdev_stat.vs_read_errors = 0; | |
2538 | vd->vdev_stat.vs_write_errors = 0; | |
2539 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2540 | |
d6320ddb | 2541 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2542 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2543 | |
2544 | /* | |
b128c09f BB |
2545 | * If we're in the FAULTED state or have experienced failed I/O, then |
2546 | * clear the persistent state and attempt to reopen the device. We | |
2547 | * also mark the vdev config dirty, so that the new faulted state is | |
2548 | * written out to disk. | |
34dc7c2f | 2549 | */ |
b128c09f BB |
2550 | if (vd->vdev_faulted || vd->vdev_degraded || |
2551 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2552 | ||
428870ff BB |
2553 | /* |
2554 | * When reopening in reponse to a clear event, it may be due to | |
2555 | * a fmadm repair request. In this case, if the device is | |
2556 | * still broken, we want to still post the ereport again. | |
2557 | */ | |
2558 | vd->vdev_forcefault = B_TRUE; | |
2559 | ||
572e2857 | 2560 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2561 | vd->vdev_cant_read = B_FALSE; |
2562 | vd->vdev_cant_write = B_FALSE; | |
2563 | ||
572e2857 | 2564 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2565 | |
428870ff BB |
2566 | vd->vdev_forcefault = B_FALSE; |
2567 | ||
572e2857 | 2568 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2569 | vdev_state_dirty(vd->vdev_top); |
2570 | ||
2571 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2572 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2573 | ||
26685276 | 2574 | spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR); |
34dc7c2f | 2575 | } |
428870ff BB |
2576 | |
2577 | /* | |
2578 | * When clearing a FMA-diagnosed fault, we always want to | |
2579 | * unspare the device, as we assume that the original spare was | |
2580 | * done in response to the FMA fault. | |
2581 | */ | |
2582 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2583 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2584 | vd->vdev_parent->vdev_child[0] == vd) | |
2585 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2586 | } |
2587 | ||
b128c09f BB |
2588 | boolean_t |
2589 | vdev_is_dead(vdev_t *vd) | |
2590 | { | |
428870ff BB |
2591 | /* |
2592 | * Holes and missing devices are always considered "dead". | |
2593 | * This simplifies the code since we don't have to check for | |
2594 | * these types of devices in the various code paths. | |
2595 | * Instead we rely on the fact that we skip over dead devices | |
2596 | * before issuing I/O to them. | |
2597 | */ | |
2598 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2599 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2600 | } |
2601 | ||
2602 | boolean_t | |
34dc7c2f BB |
2603 | vdev_readable(vdev_t *vd) |
2604 | { | |
b128c09f | 2605 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2606 | } |
2607 | ||
b128c09f | 2608 | boolean_t |
34dc7c2f BB |
2609 | vdev_writeable(vdev_t *vd) |
2610 | { | |
b128c09f | 2611 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2612 | } |
2613 | ||
b128c09f BB |
2614 | boolean_t |
2615 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2616 | { |
fb5f0bc8 BB |
2617 | uint64_t state = vd->vdev_state; |
2618 | ||
b128c09f | 2619 | /* |
fb5f0bc8 | 2620 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2621 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2622 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2623 | * the proper locks. Note that we have to get the vdev state |
2624 | * in a local variable because although it changes atomically, | |
2625 | * we're asking two separate questions about it. | |
b128c09f | 2626 | */ |
fb5f0bc8 | 2627 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
428870ff | 2628 | !vd->vdev_cant_write && !vd->vdev_ishole); |
34dc7c2f BB |
2629 | } |
2630 | ||
b128c09f BB |
2631 | boolean_t |
2632 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2633 | { |
b128c09f | 2634 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2635 | |
b128c09f BB |
2636 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2637 | return (B_FALSE); | |
34dc7c2f | 2638 | |
b128c09f BB |
2639 | if (zio->io_type == ZIO_TYPE_READ) |
2640 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2641 | |
b128c09f BB |
2642 | if (zio->io_type == ZIO_TYPE_WRITE) |
2643 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2644 | |
b128c09f | 2645 | return (B_TRUE); |
34dc7c2f BB |
2646 | } |
2647 | ||
2648 | /* | |
2649 | * Get statistics for the given vdev. | |
2650 | */ | |
2651 | void | |
2652 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2653 | { | |
ea04106b AX |
2654 | spa_t *spa = vd->vdev_spa; |
2655 | vdev_t *rvd = spa->spa_root_vdev; | |
d6320ddb | 2656 | int c, t; |
34dc7c2f | 2657 | |
ea04106b AX |
2658 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
2659 | ||
34dc7c2f BB |
2660 | mutex_enter(&vd->vdev_stat_lock); |
2661 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2662 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2663 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2664 | vs->vs_rsize = vdev_get_min_asize(vd); |
2665 | if (vd->vdev_ops->vdev_op_leaf) | |
2666 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
1bd201e7 | 2667 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
ea04106b AX |
2668 | if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { |
2669 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; | |
2670 | } | |
34dc7c2f BB |
2671 | |
2672 | /* | |
2673 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2674 | * over all top-level vdevs (i.e. the direct children of the root). | |
2675 | */ | |
2676 | if (vd == rvd) { | |
d6320ddb | 2677 | for (c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2678 | vdev_t *cvd = rvd->vdev_child[c]; |
2679 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2680 | ||
d6320ddb | 2681 | for (t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2682 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2683 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2684 | } | |
428870ff | 2685 | cvs->vs_scan_removing = cvd->vdev_removing; |
34dc7c2f BB |
2686 | } |
2687 | } | |
ea04106b | 2688 | mutex_exit(&vd->vdev_stat_lock); |
34dc7c2f BB |
2689 | } |
2690 | ||
2691 | void | |
2692 | vdev_clear_stats(vdev_t *vd) | |
2693 | { | |
2694 | mutex_enter(&vd->vdev_stat_lock); | |
2695 | vd->vdev_stat.vs_space = 0; | |
2696 | vd->vdev_stat.vs_dspace = 0; | |
2697 | vd->vdev_stat.vs_alloc = 0; | |
2698 | mutex_exit(&vd->vdev_stat_lock); | |
2699 | } | |
2700 | ||
428870ff BB |
2701 | void |
2702 | vdev_scan_stat_init(vdev_t *vd) | |
2703 | { | |
2704 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2705 | int c; |
428870ff | 2706 | |
d6320ddb | 2707 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2708 | vdev_scan_stat_init(vd->vdev_child[c]); |
2709 | ||
2710 | mutex_enter(&vd->vdev_stat_lock); | |
2711 | vs->vs_scan_processed = 0; | |
2712 | mutex_exit(&vd->vdev_stat_lock); | |
2713 | } | |
2714 | ||
34dc7c2f | 2715 | void |
b128c09f | 2716 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2717 | { |
fb5f0bc8 BB |
2718 | spa_t *spa = zio->io_spa; |
2719 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2720 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2721 | vdev_t *pvd; |
2722 | uint64_t txg = zio->io_txg; | |
2723 | vdev_stat_t *vs = &vd->vdev_stat; | |
2724 | zio_type_t type = zio->io_type; | |
2725 | int flags = zio->io_flags; | |
2726 | ||
b128c09f BB |
2727 | /* |
2728 | * If this i/o is a gang leader, it didn't do any actual work. | |
2729 | */ | |
2730 | if (zio->io_gang_tree) | |
2731 | return; | |
2732 | ||
34dc7c2f | 2733 | if (zio->io_error == 0) { |
b128c09f BB |
2734 | /* |
2735 | * If this is a root i/o, don't count it -- we've already | |
2736 | * counted the top-level vdevs, and vdev_get_stats() will | |
2737 | * aggregate them when asked. This reduces contention on | |
2738 | * the root vdev_stat_lock and implicitly handles blocks | |
2739 | * that compress away to holes, for which there is no i/o. | |
2740 | * (Holes never create vdev children, so all the counters | |
2741 | * remain zero, which is what we want.) | |
2742 | * | |
2743 | * Note: this only applies to successful i/o (io_error == 0) | |
2744 | * because unlike i/o counts, errors are not additive. | |
2745 | * When reading a ditto block, for example, failure of | |
2746 | * one top-level vdev does not imply a root-level error. | |
2747 | */ | |
2748 | if (vd == rvd) | |
2749 | return; | |
2750 | ||
2751 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2752 | |
2753 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2754 | return; | |
2755 | ||
2756 | mutex_enter(&vd->vdev_stat_lock); | |
2757 | ||
b128c09f | 2758 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 2759 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
2760 | dsl_scan_phys_t *scn_phys = |
2761 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2762 | uint64_t *processed = &scn_phys->scn_processed; | |
2763 | ||
2764 | /* XXX cleanup? */ | |
2765 | if (vd->vdev_ops->vdev_op_leaf) | |
2766 | atomic_add_64(processed, psize); | |
2767 | vs->vs_scan_processed += psize; | |
2768 | } | |
2769 | ||
fb5f0bc8 | 2770 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2771 | vs->vs_self_healed += psize; |
34dc7c2f | 2772 | } |
fb5f0bc8 BB |
2773 | |
2774 | vs->vs_ops[type]++; | |
2775 | vs->vs_bytes[type] += psize; | |
2776 | ||
2777 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2778 | return; |
2779 | } | |
2780 | ||
2781 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2782 | return; | |
2783 | ||
9babb374 BB |
2784 | /* |
2785 | * If this is an I/O error that is going to be retried, then ignore the | |
2786 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2787 | * hard errors, when in reality they can happen for any number of | |
2788 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2789 | */ | |
2790 | if (zio->io_error == EIO && | |
2791 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2792 | return; | |
2793 | ||
428870ff BB |
2794 | /* |
2795 | * Intent logs writes won't propagate their error to the root | |
2796 | * I/O so don't mark these types of failures as pool-level | |
2797 | * errors. | |
2798 | */ | |
2799 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2800 | return; | |
2801 | ||
b128c09f | 2802 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2803 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2804 | if (zio->io_error == ECKSUM) |
2805 | vs->vs_checksum_errors++; | |
2806 | else | |
2807 | vs->vs_read_errors++; | |
34dc7c2f | 2808 | } |
9babb374 | 2809 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2810 | vs->vs_write_errors++; |
2811 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2812 | |
fb5f0bc8 BB |
2813 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2814 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 2815 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 2816 | spa->spa_claiming)) { |
fb5f0bc8 | 2817 | /* |
428870ff BB |
2818 | * This is either a normal write (not a repair), or it's |
2819 | * a repair induced by the scrub thread, or it's a repair | |
2820 | * made by zil_claim() during spa_load() in the first txg. | |
2821 | * In the normal case, we commit the DTL change in the same | |
2822 | * txg as the block was born. In the scrub-induced repair | |
2823 | * case, we know that scrubs run in first-pass syncing context, | |
2824 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2825 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
2826 | * |
2827 | * We currently do not make DTL entries for failed spontaneous | |
2828 | * self-healing writes triggered by normal (non-scrubbing) | |
2829 | * reads, because we have no transactional context in which to | |
2830 | * do so -- and it's not clear that it'd be desirable anyway. | |
2831 | */ | |
2832 | if (vd->vdev_ops->vdev_op_leaf) { | |
2833 | uint64_t commit_txg = txg; | |
572e2857 | 2834 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
2835 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
2836 | ASSERT(spa_sync_pass(spa) == 1); | |
2837 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
2838 | commit_txg = spa_syncing_txg(spa); |
2839 | } else if (spa->spa_claiming) { | |
2840 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2841 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 2842 | } |
428870ff | 2843 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 2844 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 2845 | return; |
fb5f0bc8 BB |
2846 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2847 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2848 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2849 | } |
fb5f0bc8 BB |
2850 | if (vd != rvd) |
2851 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2852 | } |
2853 | } | |
2854 | ||
34dc7c2f | 2855 | /* |
428870ff BB |
2856 | * Update the in-core space usage stats for this vdev, its metaslab class, |
2857 | * and the root vdev. | |
34dc7c2f BB |
2858 | */ |
2859 | void | |
428870ff BB |
2860 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
2861 | int64_t space_delta) | |
34dc7c2f BB |
2862 | { |
2863 | int64_t dspace_delta = space_delta; | |
2864 | spa_t *spa = vd->vdev_spa; | |
2865 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
2866 | metaslab_group_t *mg = vd->vdev_mg; |
2867 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
2868 | |
2869 | ASSERT(vd == vd->vdev_top); | |
2870 | ||
2871 | /* | |
2872 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2873 | * factor. We must calculate this here and not at the root vdev | |
2874 | * because the root vdev's psize-to-asize is simply the max of its | |
2875 | * childrens', thus not accurate enough for us. | |
2876 | */ | |
2877 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2878 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2879 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2880 | vd->vdev_deflate_ratio; | |
2881 | ||
2882 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 2883 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2884 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2885 | vd->vdev_stat.vs_dspace += dspace_delta; |
2886 | mutex_exit(&vd->vdev_stat_lock); | |
2887 | ||
428870ff | 2888 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 2889 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 2890 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2891 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2892 | rvd->vdev_stat.vs_dspace += dspace_delta; |
2893 | mutex_exit(&rvd->vdev_stat_lock); | |
2894 | } | |
428870ff BB |
2895 | |
2896 | if (mc != NULL) { | |
2897 | ASSERT(rvd == vd->vdev_parent); | |
2898 | ASSERT(vd->vdev_ms_count != 0); | |
2899 | ||
2900 | metaslab_class_space_update(mc, | |
2901 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2902 | } | |
34dc7c2f BB |
2903 | } |
2904 | ||
2905 | /* | |
2906 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2907 | * so that it will be written out next time the vdev configuration is synced. | |
2908 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2909 | */ | |
2910 | void | |
2911 | vdev_config_dirty(vdev_t *vd) | |
2912 | { | |
2913 | spa_t *spa = vd->vdev_spa; | |
2914 | vdev_t *rvd = spa->spa_root_vdev; | |
2915 | int c; | |
2916 | ||
572e2857 BB |
2917 | ASSERT(spa_writeable(spa)); |
2918 | ||
34dc7c2f | 2919 | /* |
9babb374 BB |
2920 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2921 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2922 | */ |
2923 | if (vd->vdev_aux != NULL) { | |
2924 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2925 | nvlist_t **aux; | |
2926 | uint_t naux; | |
2927 | ||
2928 | for (c = 0; c < sav->sav_count; c++) { | |
2929 | if (sav->sav_vdevs[c] == vd) | |
2930 | break; | |
2931 | } | |
2932 | ||
2933 | if (c == sav->sav_count) { | |
2934 | /* | |
2935 | * We're being removed. There's nothing more to do. | |
2936 | */ | |
2937 | ASSERT(sav->sav_sync == B_TRUE); | |
2938 | return; | |
2939 | } | |
2940 | ||
2941 | sav->sav_sync = B_TRUE; | |
2942 | ||
9babb374 BB |
2943 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2944 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2945 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2946 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2947 | } | |
b128c09f BB |
2948 | |
2949 | ASSERT(c < naux); | |
2950 | ||
2951 | /* | |
2952 | * Setting the nvlist in the middle if the array is a little | |
2953 | * sketchy, but it will work. | |
2954 | */ | |
2955 | nvlist_free(aux[c]); | |
428870ff | 2956 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
2957 | |
2958 | return; | |
2959 | } | |
2960 | ||
2961 | /* | |
2962 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2963 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2964 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2965 | * so this is sufficient to ensure mutual exclusion. |
2966 | */ | |
b128c09f BB |
2967 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2968 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2969 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2970 | |
2971 | if (vd == rvd) { | |
2972 | for (c = 0; c < rvd->vdev_children; c++) | |
2973 | vdev_config_dirty(rvd->vdev_child[c]); | |
2974 | } else { | |
2975 | ASSERT(vd == vd->vdev_top); | |
2976 | ||
428870ff BB |
2977 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
2978 | !vd->vdev_ishole) | |
b128c09f | 2979 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
2980 | } |
2981 | } | |
2982 | ||
2983 | void | |
2984 | vdev_config_clean(vdev_t *vd) | |
2985 | { | |
2986 | spa_t *spa = vd->vdev_spa; | |
2987 | ||
b128c09f BB |
2988 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2989 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2990 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2991 | |
b128c09f BB |
2992 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2993 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2994 | } |
2995 | ||
b128c09f BB |
2996 | /* |
2997 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2998 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2999 | * the state changes from larger config changes because they require | |
3000 | * much less locking, and are often needed for administrative actions. | |
3001 | */ | |
3002 | void | |
3003 | vdev_state_dirty(vdev_t *vd) | |
3004 | { | |
3005 | spa_t *spa = vd->vdev_spa; | |
3006 | ||
572e2857 | 3007 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
3008 | ASSERT(vd == vd->vdev_top); |
3009 | ||
3010 | /* | |
3011 | * The state list is protected by the SCL_STATE lock. The caller | |
3012 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3013 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3014 | * so this is sufficient to ensure mutual exclusion. | |
3015 | */ | |
3016 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3017 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3018 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3019 | ||
428870ff | 3020 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
3021 | list_insert_head(&spa->spa_state_dirty_list, vd); |
3022 | } | |
3023 | ||
3024 | void | |
3025 | vdev_state_clean(vdev_t *vd) | |
3026 | { | |
3027 | spa_t *spa = vd->vdev_spa; | |
3028 | ||
3029 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3030 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3031 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3032 | ||
3033 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3034 | list_remove(&spa->spa_state_dirty_list, vd); | |
3035 | } | |
3036 | ||
3037 | /* | |
3038 | * Propagate vdev state up from children to parent. | |
3039 | */ | |
34dc7c2f BB |
3040 | void |
3041 | vdev_propagate_state(vdev_t *vd) | |
3042 | { | |
fb5f0bc8 BB |
3043 | spa_t *spa = vd->vdev_spa; |
3044 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
3045 | int degraded = 0, faulted = 0; |
3046 | int corrupted = 0; | |
34dc7c2f | 3047 | vdev_t *child; |
d6320ddb | 3048 | int c; |
34dc7c2f BB |
3049 | |
3050 | if (vd->vdev_children > 0) { | |
d6320ddb | 3051 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 3052 | child = vd->vdev_child[c]; |
b128c09f | 3053 | |
428870ff BB |
3054 | /* |
3055 | * Don't factor holes into the decision. | |
3056 | */ | |
3057 | if (child->vdev_ishole) | |
3058 | continue; | |
3059 | ||
b128c09f | 3060 | if (!vdev_readable(child) || |
fb5f0bc8 | 3061 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
3062 | /* |
3063 | * Root special: if there is a top-level log | |
3064 | * device, treat the root vdev as if it were | |
3065 | * degraded. | |
3066 | */ | |
3067 | if (child->vdev_islog && vd == rvd) | |
3068 | degraded++; | |
3069 | else | |
3070 | faulted++; | |
3071 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 3072 | degraded++; |
b128c09f | 3073 | } |
34dc7c2f BB |
3074 | |
3075 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3076 | corrupted++; | |
3077 | } | |
3078 | ||
3079 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3080 | ||
3081 | /* | |
b128c09f | 3082 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
3083 | * opened due to corrupted metadata, then propagate the root |
3084 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3085 | * replicas'. | |
3086 | */ | |
3087 | if (corrupted && vd == rvd && | |
3088 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3089 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3090 | VDEV_AUX_CORRUPT_DATA); | |
3091 | } | |
3092 | ||
b128c09f | 3093 | if (vd->vdev_parent) |
34dc7c2f BB |
3094 | vdev_propagate_state(vd->vdev_parent); |
3095 | } | |
3096 | ||
3097 | /* | |
3098 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3099 | * state, because we're in the process of opening children depth-first. | |
3100 | * Otherwise, we propagate the change to the parent. | |
3101 | * | |
3102 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3103 | * generated. | |
3104 | */ | |
3105 | void | |
3106 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3107 | { | |
3108 | uint64_t save_state; | |
b128c09f | 3109 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
3110 | |
3111 | if (state == vd->vdev_state) { | |
3112 | vd->vdev_stat.vs_aux = aux; | |
3113 | return; | |
3114 | } | |
3115 | ||
3116 | save_state = vd->vdev_state; | |
3117 | ||
3118 | vd->vdev_state = state; | |
3119 | vd->vdev_stat.vs_aux = aux; | |
3120 | ||
3121 | /* | |
3122 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
3123 | * always close the underlying device unless the device has requested |
3124 | * a delayed close (i.e. we're about to remove or fault the device). | |
3125 | * Otherwise, we keep accessible but invalid devices open forever. | |
3126 | * We don't call vdev_close() itself, because that implies some extra | |
3127 | * checks (offline, etc) that we don't want here. This is limited to | |
3128 | * leaf devices, because otherwise closing the device will affect other | |
3129 | * children. | |
34dc7c2f | 3130 | */ |
428870ff BB |
3131 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
3132 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
3133 | vd->vdev_ops->vdev_op_close(vd); |
3134 | ||
428870ff BB |
3135 | /* |
3136 | * If we have brought this vdev back into service, we need | |
3137 | * to notify fmd so that it can gracefully repair any outstanding | |
3138 | * cases due to a missing device. We do this in all cases, even those | |
3139 | * that probably don't correlate to a repaired fault. This is sure to | |
3140 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
3141 | * this is a transient state it's OK, as the retire agent will | |
3142 | * double-check the state of the vdev before repairing it. | |
3143 | */ | |
3144 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
3145 | vd->vdev_prevstate != state) | |
3146 | zfs_post_state_change(spa, vd); | |
3147 | ||
34dc7c2f BB |
3148 | if (vd->vdev_removed && |
3149 | state == VDEV_STATE_CANT_OPEN && | |
3150 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3151 | /* | |
3152 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3153 | * device was previously marked removed and someone attempted to | |
3154 | * reopen it. If this failed due to a nonexistent device, then | |
3155 | * keep the device in the REMOVED state. We also let this be if | |
3156 | * it is one of our special test online cases, which is only | |
3157 | * attempting to online the device and shouldn't generate an FMA | |
3158 | * fault. | |
3159 | */ | |
3160 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3161 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3162 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3163 | vd->vdev_removed = B_TRUE; |
3164 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3165 | /* | |
572e2857 BB |
3166 | * If we fail to open a vdev during an import or recovery, we |
3167 | * mark it as "not available", which signifies that it was | |
3168 | * never there to begin with. Failure to open such a device | |
3169 | * is not considered an error. | |
34dc7c2f | 3170 | */ |
572e2857 BB |
3171 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3172 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3173 | vd->vdev_ops->vdev_op_leaf) |
3174 | vd->vdev_not_present = 1; | |
3175 | ||
3176 | /* | |
3177 | * Post the appropriate ereport. If the 'prevstate' field is | |
3178 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3179 | * that this is part of a vdev_reopen(). In this case, we don't | |
3180 | * want to post the ereport if the device was already in the | |
3181 | * CANT_OPEN state beforehand. | |
3182 | * | |
3183 | * If the 'checkremove' flag is set, then this is an attempt to | |
3184 | * online the device in response to an insertion event. If we | |
3185 | * hit this case, then we have detected an insertion event for a | |
3186 | * faulted or offline device that wasn't in the removed state. | |
3187 | * In this scenario, we don't post an ereport because we are | |
3188 | * about to replace the device, or attempt an online with | |
3189 | * vdev_forcefault, which will generate the fault for us. | |
3190 | */ | |
3191 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3192 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3193 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3194 | const char *class; |
3195 | ||
3196 | switch (aux) { | |
3197 | case VDEV_AUX_OPEN_FAILED: | |
3198 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3199 | break; | |
3200 | case VDEV_AUX_CORRUPT_DATA: | |
3201 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3202 | break; | |
3203 | case VDEV_AUX_NO_REPLICAS: | |
3204 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3205 | break; | |
3206 | case VDEV_AUX_BAD_GUID_SUM: | |
3207 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3208 | break; | |
3209 | case VDEV_AUX_TOO_SMALL: | |
3210 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3211 | break; | |
3212 | case VDEV_AUX_BAD_LABEL: | |
3213 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3214 | break; | |
3215 | default: | |
3216 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3217 | } | |
3218 | ||
b128c09f | 3219 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3220 | } |
3221 | ||
3222 | /* Erase any notion of persistent removed state */ | |
3223 | vd->vdev_removed = B_FALSE; | |
3224 | } else { | |
3225 | vd->vdev_removed = B_FALSE; | |
3226 | } | |
3227 | ||
9babb374 BB |
3228 | if (!isopen && vd->vdev_parent) |
3229 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3230 | } |
b128c09f BB |
3231 | |
3232 | /* | |
3233 | * Check the vdev configuration to ensure that it's capable of supporting | |
57a4eddc | 3234 | * a root pool. |
b128c09f BB |
3235 | */ |
3236 | boolean_t | |
3237 | vdev_is_bootable(vdev_t *vd) | |
3238 | { | |
57a4eddc RL |
3239 | #if defined(__sun__) || defined(__sun) |
3240 | /* | |
3241 | * Currently, we do not support RAID-Z or partial configuration. | |
3242 | * In addition, only a single top-level vdev is allowed and none of the | |
3243 | * leaves can be wholedisks. | |
3244 | */ | |
d6320ddb BB |
3245 | int c; |
3246 | ||
b128c09f BB |
3247 | if (!vd->vdev_ops->vdev_op_leaf) { |
3248 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3249 | ||
3250 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3251 | vd->vdev_children > 1) { | |
3252 | return (B_FALSE); | |
3253 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3254 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3255 | return (B_FALSE); | |
3256 | } | |
3257 | } else if (vd->vdev_wholedisk == 1) { | |
3258 | return (B_FALSE); | |
3259 | } | |
3260 | ||
d6320ddb | 3261 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3262 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3263 | return (B_FALSE); | |
3264 | } | |
57a4eddc | 3265 | #endif /* __sun__ || __sun */ |
b128c09f BB |
3266 | return (B_TRUE); |
3267 | } | |
9babb374 | 3268 | |
428870ff BB |
3269 | /* |
3270 | * Load the state from the original vdev tree (ovd) which | |
3271 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3272 | * vdev was offline or faulted then we transfer that state to the |
3273 | * device in the current vdev tree (nvd). | |
428870ff | 3274 | */ |
9babb374 | 3275 | void |
428870ff | 3276 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3277 | { |
d6320ddb | 3278 | int c; |
9babb374 | 3279 | |
572e2857 | 3280 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3281 | ASSERT(spa_config_held(nvd->vdev_spa, |
3282 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3283 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3284 | |
d6320ddb | 3285 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3286 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3287 | |
572e2857 | 3288 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3289 | /* |
572e2857 | 3290 | * Restore the persistent vdev state |
9babb374 | 3291 | */ |
428870ff | 3292 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3293 | nvd->vdev_faulted = ovd->vdev_faulted; |
3294 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3295 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3296 | } |
3297 | } | |
3298 | ||
572e2857 BB |
3299 | /* |
3300 | * Determine if a log device has valid content. If the vdev was | |
3301 | * removed or faulted in the MOS config then we know that | |
3302 | * the content on the log device has already been written to the pool. | |
3303 | */ | |
3304 | boolean_t | |
3305 | vdev_log_state_valid(vdev_t *vd) | |
3306 | { | |
d6320ddb BB |
3307 | int c; |
3308 | ||
572e2857 BB |
3309 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3310 | !vd->vdev_removed) | |
3311 | return (B_TRUE); | |
3312 | ||
d6320ddb | 3313 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3314 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3315 | return (B_TRUE); | |
3316 | ||
3317 | return (B_FALSE); | |
3318 | } | |
3319 | ||
9babb374 BB |
3320 | /* |
3321 | * Expand a vdev if possible. | |
3322 | */ | |
3323 | void | |
3324 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3325 | { | |
3326 | ASSERT(vd->vdev_top == vd); | |
3327 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3328 | ||
3329 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3330 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3331 | vdev_config_dirty(vd); | |
3332 | } | |
3333 | } | |
428870ff BB |
3334 | |
3335 | /* | |
3336 | * Split a vdev. | |
3337 | */ | |
3338 | void | |
3339 | vdev_split(vdev_t *vd) | |
3340 | { | |
3341 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3342 | ||
3343 | vdev_remove_child(pvd, vd); | |
3344 | vdev_compact_children(pvd); | |
3345 | ||
3346 | cvd = pvd->vdev_child[0]; | |
3347 | if (pvd->vdev_children == 1) { | |
3348 | vdev_remove_parent(cvd); | |
3349 | cvd->vdev_splitting = B_TRUE; | |
3350 | } | |
3351 | vdev_propagate_state(cvd); | |
3352 | } | |
c28b2279 | 3353 | |
c06d4368 AX |
3354 | void |
3355 | vdev_deadman(vdev_t *vd) | |
3356 | { | |
3357 | int c; | |
3358 | ||
3359 | for (c = 0; c < vd->vdev_children; c++) { | |
3360 | vdev_t *cvd = vd->vdev_child[c]; | |
3361 | ||
3362 | vdev_deadman(cvd); | |
3363 | } | |
3364 | ||
3365 | if (vd->vdev_ops->vdev_op_leaf) { | |
3366 | vdev_queue_t *vq = &vd->vdev_queue; | |
3367 | ||
3368 | mutex_enter(&vq->vq_lock); | |
a08ee875 | 3369 | if (avl_numnodes(&vq->vq_active_tree) > 0) { |
c06d4368 AX |
3370 | spa_t *spa = vd->vdev_spa; |
3371 | zio_t *fio; | |
3372 | uint64_t delta; | |
3373 | ||
3374 | /* | |
3375 | * Look at the head of all the pending queues, | |
3376 | * if any I/O has been outstanding for longer than | |
3377 | * the spa_deadman_synctime we log a zevent. | |
3378 | */ | |
a08ee875 | 3379 | fio = avl_first(&vq->vq_active_tree); |
c06d4368 AX |
3380 | delta = gethrtime() - fio->io_timestamp; |
3381 | if (delta > spa_deadman_synctime(spa)) { | |
3382 | zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " | |
3383 | "delta %lluns, last io %lluns", | |
3384 | fio->io_timestamp, delta, | |
3385 | vq->vq_io_complete_ts); | |
3386 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, | |
3387 | spa, vd, fio, 0, 0); | |
3388 | } | |
3389 | } | |
3390 | mutex_exit(&vq->vq_lock); | |
3391 | } | |
3392 | } | |
3393 | ||
c28b2279 BB |
3394 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3395 | EXPORT_SYMBOL(vdev_fault); | |
3396 | EXPORT_SYMBOL(vdev_degrade); | |
3397 | EXPORT_SYMBOL(vdev_online); | |
3398 | EXPORT_SYMBOL(vdev_offline); | |
3399 | EXPORT_SYMBOL(vdev_clear); | |
ea04106b AX |
3400 | |
3401 | module_param(metaslabs_per_vdev, int, 0644); | |
3402 | MODULE_PARM_DESC(metaslabs_per_vdev, | |
3403 | "Divide added vdev into approximately (but no more than) this number " | |
3404 | "of metaslabs"); | |
c28b2279 | 3405 | #endif |