]>
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. |
9bd274dd | 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> | |
93cf2076 | 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 BB |
47 | |
48 | /* | |
49 | * Virtual device management. | |
50 | */ | |
51 | ||
52 | static vdev_ops_t *vdev_ops_table[] = { | |
53 | &vdev_root_ops, | |
54 | &vdev_raidz_ops, | |
55 | &vdev_mirror_ops, | |
56 | &vdev_replacing_ops, | |
57 | &vdev_spare_ops, | |
58 | &vdev_disk_ops, | |
59 | &vdev_file_ops, | |
60 | &vdev_missing_ops, | |
428870ff | 61 | &vdev_hole_ops, |
34dc7c2f BB |
62 | NULL |
63 | }; | |
64 | ||
34dc7c2f BB |
65 | /* |
66 | * Given a vdev type, return the appropriate ops vector. | |
67 | */ | |
68 | static vdev_ops_t * | |
69 | vdev_getops(const char *type) | |
70 | { | |
71 | vdev_ops_t *ops, **opspp; | |
72 | ||
73 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
74 | if (strcmp(ops->vdev_op_type, type) == 0) | |
75 | break; | |
76 | ||
77 | return (ops); | |
78 | } | |
79 | ||
80 | /* | |
81 | * Default asize function: return the MAX of psize with the asize of | |
82 | * all children. This is what's used by anything other than RAID-Z. | |
83 | */ | |
84 | uint64_t | |
85 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
86 | { | |
87 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
88 | uint64_t csize; | |
d6320ddb | 89 | int c; |
34dc7c2f | 90 | |
d6320ddb | 91 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
92 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
93 | asize = MAX(asize, csize); | |
94 | } | |
95 | ||
96 | return (asize); | |
97 | } | |
98 | ||
99 | /* | |
9babb374 BB |
100 | * Get the minimum allocatable size. We define the allocatable size as |
101 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
102 | * replace or attach devices which don't have the same physical size but | |
103 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
104 | */ |
105 | uint64_t | |
9babb374 | 106 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 107 | { |
9babb374 | 108 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 109 | |
9babb374 | 110 | /* |
1bd201e7 | 111 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
112 | * just return our own asize. |
113 | */ | |
114 | if (pvd == NULL) | |
115 | return (vd->vdev_asize); | |
34dc7c2f BB |
116 | |
117 | /* | |
9babb374 BB |
118 | * The top-level vdev just returns the allocatable size rounded |
119 | * to the nearest metaslab. | |
34dc7c2f | 120 | */ |
9babb374 BB |
121 | if (vd == vd->vdev_top) |
122 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 123 | |
9babb374 BB |
124 | /* |
125 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
126 | * so each child must provide at least 1/Nth of its asize. | |
127 | */ | |
128 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
129 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 130 | |
9babb374 BB |
131 | return (pvd->vdev_min_asize); |
132 | } | |
133 | ||
134 | void | |
135 | vdev_set_min_asize(vdev_t *vd) | |
136 | { | |
d6320ddb | 137 | int c; |
9babb374 | 138 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 139 | |
d6320ddb | 140 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 141 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
142 | } |
143 | ||
144 | vdev_t * | |
145 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
146 | { | |
147 | vdev_t *rvd = spa->spa_root_vdev; | |
148 | ||
b128c09f | 149 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 150 | |
b128c09f BB |
151 | if (vdev < rvd->vdev_children) { |
152 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 153 | return (rvd->vdev_child[vdev]); |
b128c09f | 154 | } |
34dc7c2f BB |
155 | |
156 | return (NULL); | |
157 | } | |
158 | ||
159 | vdev_t * | |
160 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
161 | { | |
34dc7c2f | 162 | vdev_t *mvd; |
d6320ddb | 163 | int c; |
34dc7c2f BB |
164 | |
165 | if (vd->vdev_guid == guid) | |
166 | return (vd); | |
167 | ||
d6320ddb | 168 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
169 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
170 | NULL) | |
171 | return (mvd); | |
172 | ||
173 | return (NULL); | |
174 | } | |
175 | ||
176 | void | |
177 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
178 | { | |
179 | size_t oldsize, newsize; | |
180 | uint64_t id = cvd->vdev_id; | |
181 | vdev_t **newchild; | |
182 | ||
b128c09f | 183 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
184 | ASSERT(cvd->vdev_parent == NULL); |
185 | ||
186 | cvd->vdev_parent = pvd; | |
187 | ||
188 | if (pvd == NULL) | |
189 | return; | |
190 | ||
191 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
192 | ||
193 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
194 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
195 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
196 | ||
93cf2076 | 197 | newchild = kmem_alloc(newsize, KM_PUSHPAGE); |
34dc7c2f BB |
198 | if (pvd->vdev_child != NULL) { |
199 | bcopy(pvd->vdev_child, newchild, oldsize); | |
200 | kmem_free(pvd->vdev_child, oldsize); | |
201 | } | |
202 | ||
203 | pvd->vdev_child = newchild; | |
204 | pvd->vdev_child[id] = cvd; | |
205 | ||
206 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
207 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
208 | ||
209 | /* | |
210 | * Walk up all ancestors to update guid sum. | |
211 | */ | |
212 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
213 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
214 | } |
215 | ||
216 | void | |
217 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
218 | { | |
219 | int c; | |
220 | uint_t id = cvd->vdev_id; | |
221 | ||
222 | ASSERT(cvd->vdev_parent == pvd); | |
223 | ||
224 | if (pvd == NULL) | |
225 | return; | |
226 | ||
227 | ASSERT(id < pvd->vdev_children); | |
228 | ASSERT(pvd->vdev_child[id] == cvd); | |
229 | ||
230 | pvd->vdev_child[id] = NULL; | |
231 | cvd->vdev_parent = NULL; | |
232 | ||
233 | for (c = 0; c < pvd->vdev_children; c++) | |
234 | if (pvd->vdev_child[c]) | |
235 | break; | |
236 | ||
237 | if (c == pvd->vdev_children) { | |
238 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
239 | pvd->vdev_child = NULL; | |
240 | pvd->vdev_children = 0; | |
241 | } | |
242 | ||
243 | /* | |
244 | * Walk up all ancestors to update guid sum. | |
245 | */ | |
246 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
247 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
248 | } |
249 | ||
250 | /* | |
251 | * Remove any holes in the child array. | |
252 | */ | |
253 | void | |
254 | vdev_compact_children(vdev_t *pvd) | |
255 | { | |
256 | vdev_t **newchild, *cvd; | |
257 | int oldc = pvd->vdev_children; | |
9babb374 | 258 | int newc; |
d6320ddb | 259 | int c; |
34dc7c2f | 260 | |
b128c09f | 261 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 262 | |
d6320ddb | 263 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
264 | if (pvd->vdev_child[c]) |
265 | newc++; | |
266 | ||
93cf2076 | 267 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_PUSHPAGE); |
34dc7c2f | 268 | |
d6320ddb | 269 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
270 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
271 | newchild[newc] = cvd; | |
272 | cvd->vdev_id = newc++; | |
273 | } | |
274 | } | |
275 | ||
276 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
277 | pvd->vdev_child = newchild; | |
278 | pvd->vdev_children = newc; | |
279 | } | |
280 | ||
281 | /* | |
282 | * Allocate and minimally initialize a vdev_t. | |
283 | */ | |
428870ff | 284 | vdev_t * |
34dc7c2f BB |
285 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
286 | { | |
287 | vdev_t *vd; | |
d6320ddb | 288 | int t; |
34dc7c2f | 289 | |
b8d06fca | 290 | vd = kmem_zalloc(sizeof (vdev_t), KM_PUSHPAGE); |
34dc7c2f BB |
291 | |
292 | if (spa->spa_root_vdev == NULL) { | |
293 | ASSERT(ops == &vdev_root_ops); | |
294 | spa->spa_root_vdev = vd; | |
3541dc6d | 295 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
296 | } |
297 | ||
428870ff | 298 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
299 | if (spa->spa_root_vdev == vd) { |
300 | /* | |
301 | * The root vdev's guid will also be the pool guid, | |
302 | * which must be unique among all pools. | |
303 | */ | |
428870ff | 304 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
305 | } else { |
306 | /* | |
307 | * Any other vdev's guid must be unique within the pool. | |
308 | */ | |
428870ff | 309 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
310 | } |
311 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
312 | } | |
313 | ||
314 | vd->vdev_spa = spa; | |
315 | vd->vdev_id = id; | |
316 | vd->vdev_guid = guid; | |
317 | vd->vdev_guid_sum = guid; | |
318 | vd->vdev_ops = ops; | |
319 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 320 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f | 321 | |
98f72a53 BB |
322 | list_link_init(&vd->vdev_config_dirty_node); |
323 | list_link_init(&vd->vdev_state_dirty_node); | |
34dc7c2f BB |
324 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); |
325 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 326 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
d6320ddb | 327 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 | 328 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL, |
fb5f0bc8 BB |
329 | &vd->vdev_dtl_lock); |
330 | } | |
34dc7c2f BB |
331 | txg_list_create(&vd->vdev_ms_list, |
332 | offsetof(struct metaslab, ms_txg_node)); | |
333 | txg_list_create(&vd->vdev_dtl_list, | |
334 | offsetof(struct vdev, vdev_dtl_node)); | |
335 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
336 | vdev_queue_init(vd); | |
337 | vdev_cache_init(vd); | |
338 | ||
339 | return (vd); | |
340 | } | |
341 | ||
342 | /* | |
343 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
344 | * creating a new vdev or loading an existing one - the behavior is slightly | |
345 | * different for each case. | |
346 | */ | |
347 | int | |
348 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
349 | int alloctype) | |
350 | { | |
351 | vdev_ops_t *ops; | |
352 | char *type; | |
353 | uint64_t guid = 0, islog, nparity; | |
354 | vdev_t *vd; | |
355 | ||
b128c09f | 356 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
357 | |
358 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
2e528b49 | 359 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
360 | |
361 | if ((ops = vdev_getops(type)) == NULL) | |
2e528b49 | 362 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
363 | |
364 | /* | |
365 | * If this is a load, get the vdev guid from the nvlist. | |
366 | * Otherwise, vdev_alloc_common() will generate one for us. | |
367 | */ | |
368 | if (alloctype == VDEV_ALLOC_LOAD) { | |
369 | uint64_t label_id; | |
370 | ||
371 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
372 | label_id != id) | |
2e528b49 | 373 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
374 | |
375 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 376 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
377 | } else if (alloctype == VDEV_ALLOC_SPARE) { |
378 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 379 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
380 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { |
381 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 382 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
383 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
384 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 385 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
386 | } |
387 | ||
388 | /* | |
389 | * The first allocated vdev must be of type 'root'. | |
390 | */ | |
391 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
2e528b49 | 392 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
393 | |
394 | /* | |
395 | * Determine whether we're a log vdev. | |
396 | */ | |
397 | islog = 0; | |
398 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
399 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
2e528b49 | 400 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 401 | |
428870ff | 402 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
2e528b49 | 403 | return (SET_ERROR(ENOTSUP)); |
428870ff | 404 | |
34dc7c2f BB |
405 | /* |
406 | * Set the nparity property for RAID-Z vdevs. | |
407 | */ | |
408 | nparity = -1ULL; | |
409 | if (ops == &vdev_raidz_ops) { | |
410 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
411 | &nparity) == 0) { | |
428870ff | 412 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
2e528b49 | 413 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 414 | /* |
45d1cae3 BB |
415 | * Previous versions could only support 1 or 2 parity |
416 | * device. | |
34dc7c2f | 417 | */ |
45d1cae3 BB |
418 | if (nparity > 1 && |
419 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
2e528b49 | 420 | return (SET_ERROR(ENOTSUP)); |
45d1cae3 BB |
421 | if (nparity > 2 && |
422 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
2e528b49 | 423 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
424 | } else { |
425 | /* | |
426 | * We require the parity to be specified for SPAs that | |
427 | * support multiple parity levels. | |
428 | */ | |
45d1cae3 | 429 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
2e528b49 | 430 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
431 | /* |
432 | * Otherwise, we default to 1 parity device for RAID-Z. | |
433 | */ | |
434 | nparity = 1; | |
435 | } | |
436 | } else { | |
437 | nparity = 0; | |
438 | } | |
439 | ASSERT(nparity != -1ULL); | |
440 | ||
441 | vd = vdev_alloc_common(spa, id, guid, ops); | |
442 | ||
443 | vd->vdev_islog = islog; | |
444 | vd->vdev_nparity = nparity; | |
445 | ||
446 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
447 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
448 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
449 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
450 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
451 | &vd->vdev_physpath) == 0) | |
452 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
453 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
454 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
455 | |
456 | /* | |
457 | * Set the whole_disk property. If it's not specified, leave the value | |
458 | * as -1. | |
459 | */ | |
460 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
461 | &vd->vdev_wholedisk) != 0) | |
462 | vd->vdev_wholedisk = -1ULL; | |
463 | ||
464 | /* | |
465 | * Look for the 'not present' flag. This will only be set if the device | |
466 | * was not present at the time of import. | |
467 | */ | |
9babb374 BB |
468 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
469 | &vd->vdev_not_present); | |
34dc7c2f BB |
470 | |
471 | /* | |
472 | * Get the alignment requirement. | |
473 | */ | |
474 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
475 | ||
428870ff BB |
476 | /* |
477 | * Retrieve the vdev creation time. | |
478 | */ | |
479 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
480 | &vd->vdev_crtxg); | |
481 | ||
34dc7c2f BB |
482 | /* |
483 | * If we're a top-level vdev, try to load the allocation parameters. | |
484 | */ | |
428870ff BB |
485 | if (parent && !parent->vdev_parent && |
486 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
487 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
488 | &vd->vdev_ms_array); | |
489 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
490 | &vd->vdev_ms_shift); | |
491 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
492 | &vd->vdev_asize); | |
428870ff BB |
493 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
494 | &vd->vdev_removing); | |
495 | } | |
496 | ||
5ffb9d1d | 497 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
498 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
499 | alloctype == VDEV_ALLOC_ADD || | |
500 | alloctype == VDEV_ALLOC_SPLIT || | |
501 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
502 | vd->vdev_mg = metaslab_group_create(islog ? | |
503 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
504 | } |
505 | ||
506 | /* | |
507 | * If we're a leaf vdev, try to load the DTL object and other state. | |
508 | */ | |
b128c09f | 509 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
510 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
511 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
512 | if (alloctype == VDEV_ALLOC_LOAD) { |
513 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
93cf2076 | 514 | &vd->vdev_dtl_object); |
b128c09f BB |
515 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
516 | &vd->vdev_unspare); | |
517 | } | |
9babb374 BB |
518 | |
519 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
520 | uint64_t spare = 0; | |
521 | ||
522 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
523 | &spare) == 0 && spare) | |
524 | spa_spare_add(vd); | |
525 | } | |
526 | ||
34dc7c2f BB |
527 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
528 | &vd->vdev_offline); | |
b128c09f | 529 | |
5d1f7fb6 GW |
530 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, |
531 | &vd->vdev_resilver_txg); | |
572e2857 | 532 | |
34dc7c2f BB |
533 | /* |
534 | * When importing a pool, we want to ignore the persistent fault | |
535 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
536 | * valid in the current context. Local vdevs will |
537 | * remain in the faulted state. | |
34dc7c2f | 538 | */ |
428870ff | 539 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
540 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
541 | &vd->vdev_faulted); | |
542 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
543 | &vd->vdev_degraded); | |
544 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
545 | &vd->vdev_removed); | |
428870ff BB |
546 | |
547 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
548 | char *aux; | |
549 | ||
550 | vd->vdev_label_aux = | |
551 | VDEV_AUX_ERR_EXCEEDED; | |
552 | if (nvlist_lookup_string(nv, | |
553 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
554 | strcmp(aux, "external") == 0) | |
555 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
556 | } | |
34dc7c2f BB |
557 | } |
558 | } | |
559 | ||
560 | /* | |
561 | * Add ourselves to the parent's list of children. | |
562 | */ | |
563 | vdev_add_child(parent, vd); | |
564 | ||
565 | *vdp = vd; | |
566 | ||
567 | return (0); | |
568 | } | |
569 | ||
570 | void | |
571 | vdev_free(vdev_t *vd) | |
572 | { | |
d6320ddb | 573 | int c, t; |
34dc7c2f BB |
574 | spa_t *spa = vd->vdev_spa; |
575 | ||
576 | /* | |
577 | * vdev_free() implies closing the vdev first. This is simpler than | |
578 | * trying to ensure complicated semantics for all callers. | |
579 | */ | |
580 | vdev_close(vd); | |
581 | ||
b128c09f | 582 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 583 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
584 | |
585 | /* | |
586 | * Free all children. | |
587 | */ | |
d6320ddb | 588 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
589 | vdev_free(vd->vdev_child[c]); |
590 | ||
591 | ASSERT(vd->vdev_child == NULL); | |
592 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
593 | ||
594 | /* | |
595 | * Discard allocation state. | |
596 | */ | |
428870ff | 597 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 598 | vdev_metaslab_fini(vd); |
428870ff BB |
599 | metaslab_group_destroy(vd->vdev_mg); |
600 | } | |
34dc7c2f | 601 | |
c99c9001 MS |
602 | ASSERT0(vd->vdev_stat.vs_space); |
603 | ASSERT0(vd->vdev_stat.vs_dspace); | |
604 | ASSERT0(vd->vdev_stat.vs_alloc); | |
34dc7c2f BB |
605 | |
606 | /* | |
607 | * Remove this vdev from its parent's child list. | |
608 | */ | |
609 | vdev_remove_child(vd->vdev_parent, vd); | |
610 | ||
611 | ASSERT(vd->vdev_parent == NULL); | |
612 | ||
613 | /* | |
614 | * Clean up vdev structure. | |
615 | */ | |
616 | vdev_queue_fini(vd); | |
617 | vdev_cache_fini(vd); | |
618 | ||
619 | if (vd->vdev_path) | |
620 | spa_strfree(vd->vdev_path); | |
621 | if (vd->vdev_devid) | |
622 | spa_strfree(vd->vdev_devid); | |
623 | if (vd->vdev_physpath) | |
624 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
625 | if (vd->vdev_fru) |
626 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
627 | |
628 | if (vd->vdev_isspare) | |
629 | spa_spare_remove(vd); | |
630 | if (vd->vdev_isl2cache) | |
631 | spa_l2cache_remove(vd); | |
632 | ||
633 | txg_list_destroy(&vd->vdev_ms_list); | |
634 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 635 | |
34dc7c2f | 636 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 | 637 | space_map_close(vd->vdev_dtl_sm); |
d6320ddb | 638 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
639 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); |
640 | range_tree_destroy(vd->vdev_dtl[t]); | |
fb5f0bc8 | 641 | } |
34dc7c2f | 642 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 643 | |
34dc7c2f BB |
644 | mutex_destroy(&vd->vdev_dtl_lock); |
645 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 646 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
647 | |
648 | if (vd == spa->spa_root_vdev) | |
649 | spa->spa_root_vdev = NULL; | |
650 | ||
651 | kmem_free(vd, sizeof (vdev_t)); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Transfer top-level vdev state from svd to tvd. | |
656 | */ | |
657 | static void | |
658 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
659 | { | |
660 | spa_t *spa = svd->vdev_spa; | |
661 | metaslab_t *msp; | |
662 | vdev_t *vd; | |
663 | int t; | |
664 | ||
665 | ASSERT(tvd == tvd->vdev_top); | |
666 | ||
667 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
668 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
669 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
670 | ||
671 | svd->vdev_ms_array = 0; | |
672 | svd->vdev_ms_shift = 0; | |
673 | svd->vdev_ms_count = 0; | |
674 | ||
5ffb9d1d GW |
675 | if (tvd->vdev_mg) |
676 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
677 | tvd->vdev_mg = svd->vdev_mg; |
678 | tvd->vdev_ms = svd->vdev_ms; | |
679 | ||
680 | svd->vdev_mg = NULL; | |
681 | svd->vdev_ms = NULL; | |
682 | ||
683 | if (tvd->vdev_mg != NULL) | |
684 | tvd->vdev_mg->mg_vd = tvd; | |
685 | ||
686 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
687 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
688 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
689 | ||
690 | svd->vdev_stat.vs_alloc = 0; | |
691 | svd->vdev_stat.vs_space = 0; | |
692 | svd->vdev_stat.vs_dspace = 0; | |
693 | ||
694 | for (t = 0; t < TXG_SIZE; t++) { | |
695 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
696 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
697 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
698 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
699 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
700 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
701 | } | |
702 | ||
b128c09f | 703 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
704 | vdev_config_clean(svd); |
705 | vdev_config_dirty(tvd); | |
706 | } | |
707 | ||
b128c09f BB |
708 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
709 | vdev_state_clean(svd); | |
710 | vdev_state_dirty(tvd); | |
711 | } | |
712 | ||
34dc7c2f BB |
713 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
714 | svd->vdev_deflate_ratio = 0; | |
715 | ||
716 | tvd->vdev_islog = svd->vdev_islog; | |
717 | svd->vdev_islog = 0; | |
718 | } | |
719 | ||
720 | static void | |
721 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
722 | { | |
d6320ddb BB |
723 | int c; |
724 | ||
34dc7c2f BB |
725 | if (vd == NULL) |
726 | return; | |
727 | ||
728 | vd->vdev_top = tvd; | |
729 | ||
d6320ddb | 730 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
731 | vdev_top_update(tvd, vd->vdev_child[c]); |
732 | } | |
733 | ||
734 | /* | |
735 | * Add a mirror/replacing vdev above an existing vdev. | |
736 | */ | |
737 | vdev_t * | |
738 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
739 | { | |
740 | spa_t *spa = cvd->vdev_spa; | |
741 | vdev_t *pvd = cvd->vdev_parent; | |
742 | vdev_t *mvd; | |
743 | ||
b128c09f | 744 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
745 | |
746 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
747 | ||
748 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 749 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 750 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
34dc7c2f BB |
751 | mvd->vdev_ashift = cvd->vdev_ashift; |
752 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 753 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
754 | |
755 | vdev_remove_child(pvd, cvd); | |
756 | vdev_add_child(pvd, mvd); | |
757 | cvd->vdev_id = mvd->vdev_children; | |
758 | vdev_add_child(mvd, cvd); | |
759 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
760 | ||
761 | if (mvd == mvd->vdev_top) | |
762 | vdev_top_transfer(cvd, mvd); | |
763 | ||
764 | return (mvd); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Remove a 1-way mirror/replacing vdev from the tree. | |
769 | */ | |
770 | void | |
771 | vdev_remove_parent(vdev_t *cvd) | |
772 | { | |
773 | vdev_t *mvd = cvd->vdev_parent; | |
774 | vdev_t *pvd = mvd->vdev_parent; | |
775 | ||
b128c09f | 776 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
777 | |
778 | ASSERT(mvd->vdev_children == 1); | |
779 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
780 | mvd->vdev_ops == &vdev_replacing_ops || | |
781 | mvd->vdev_ops == &vdev_spare_ops); | |
782 | cvd->vdev_ashift = mvd->vdev_ashift; | |
783 | ||
784 | vdev_remove_child(mvd, cvd); | |
785 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 786 | |
34dc7c2f | 787 | /* |
b128c09f BB |
788 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
789 | * Otherwise, we could have detached an offline device, and when we | |
790 | * go to import the pool we'll think we have two top-level vdevs, | |
791 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 792 | */ |
fb5f0bc8 BB |
793 | if (mvd->vdev_top == mvd) { |
794 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 795 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
796 | cvd->vdev_guid += guid_delta; |
797 | cvd->vdev_guid_sum += guid_delta; | |
61e99a73 AB |
798 | |
799 | /* | |
800 | * If pool not set for autoexpand, we need to also preserve | |
801 | * mvd's asize to prevent automatic expansion of cvd. | |
802 | * Otherwise if we are adjusting the mirror by attaching and | |
803 | * detaching children of non-uniform sizes, the mirror could | |
804 | * autoexpand, unexpectedly requiring larger devices to | |
805 | * re-establish the mirror. | |
806 | */ | |
807 | if (!cvd->vdev_spa->spa_autoexpand) | |
808 | cvd->vdev_asize = mvd->vdev_asize; | |
fb5f0bc8 | 809 | } |
b128c09f BB |
810 | cvd->vdev_id = mvd->vdev_id; |
811 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
812 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
813 | ||
814 | if (cvd == cvd->vdev_top) | |
815 | vdev_top_transfer(mvd, cvd); | |
816 | ||
817 | ASSERT(mvd->vdev_children == 0); | |
818 | vdev_free(mvd); | |
819 | } | |
820 | ||
821 | int | |
822 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
823 | { | |
824 | spa_t *spa = vd->vdev_spa; | |
825 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
826 | uint64_t m; |
827 | uint64_t oldc = vd->vdev_ms_count; | |
828 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
829 | metaslab_t **mspp; | |
830 | int error; | |
831 | ||
428870ff BB |
832 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
833 | ||
834 | /* | |
835 | * This vdev is not being allocated from yet or is a hole. | |
836 | */ | |
837 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
838 | return (0); |
839 | ||
428870ff BB |
840 | ASSERT(!vd->vdev_ishole); |
841 | ||
9babb374 BB |
842 | /* |
843 | * Compute the raidz-deflation ratio. Note, we hard-code | |
844 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
845 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
846 | * or we will inconsistently account for existing bp's. | |
847 | */ | |
848 | vd->vdev_deflate_ratio = (1 << 17) / | |
849 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
850 | ||
34dc7c2f BB |
851 | ASSERT(oldc <= newc); |
852 | ||
b8d06fca | 853 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_PUSHPAGE | KM_NODEBUG); |
34dc7c2f BB |
854 | |
855 | if (oldc != 0) { | |
856 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
857 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
858 | } | |
859 | ||
860 | vd->vdev_ms = mspp; | |
861 | vd->vdev_ms_count = newc; | |
862 | ||
863 | for (m = oldc; m < newc; m++) { | |
93cf2076 GW |
864 | uint64_t object = 0; |
865 | ||
34dc7c2f | 866 | if (txg == 0) { |
34dc7c2f | 867 | error = dmu_read(mos, vd->vdev_ms_array, |
9babb374 BB |
868 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
869 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
870 | if (error) |
871 | return (error); | |
34dc7c2f | 872 | } |
93cf2076 | 873 | vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg); |
34dc7c2f BB |
874 | } |
875 | ||
428870ff BB |
876 | if (txg == 0) |
877 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
878 | ||
879 | /* | |
880 | * If the vdev is being removed we don't activate | |
881 | * the metaslabs since we want to ensure that no new | |
882 | * allocations are performed on this device. | |
883 | */ | |
884 | if (oldc == 0 && !vd->vdev_removing) | |
885 | metaslab_group_activate(vd->vdev_mg); | |
886 | ||
887 | if (txg == 0) | |
888 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
889 | ||
34dc7c2f BB |
890 | return (0); |
891 | } | |
892 | ||
893 | void | |
894 | vdev_metaslab_fini(vdev_t *vd) | |
895 | { | |
896 | uint64_t m; | |
897 | uint64_t count = vd->vdev_ms_count; | |
898 | ||
899 | if (vd->vdev_ms != NULL) { | |
428870ff | 900 | metaslab_group_passivate(vd->vdev_mg); |
93cf2076 GW |
901 | for (m = 0; m < count; m++) { |
902 | metaslab_t *msp = vd->vdev_ms[m]; | |
903 | ||
904 | if (msp != NULL) | |
905 | metaslab_fini(msp); | |
906 | } | |
34dc7c2f BB |
907 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); |
908 | vd->vdev_ms = NULL; | |
909 | } | |
920dd524 ED |
910 | |
911 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
34dc7c2f BB |
912 | } |
913 | ||
b128c09f BB |
914 | typedef struct vdev_probe_stats { |
915 | boolean_t vps_readable; | |
916 | boolean_t vps_writeable; | |
917 | int vps_flags; | |
b128c09f BB |
918 | } vdev_probe_stats_t; |
919 | ||
920 | static void | |
921 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 922 | { |
fb5f0bc8 | 923 | spa_t *spa = zio->io_spa; |
d164b209 | 924 | vdev_t *vd = zio->io_vd; |
b128c09f | 925 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
926 | |
927 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
928 | |
929 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
930 | if (zio->io_error == 0) |
931 | vps->vps_readable = 1; | |
fb5f0bc8 | 932 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 933 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
934 | zio->io_offset, zio->io_size, zio->io_data, |
935 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
936 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
937 | } else { | |
938 | zio_buf_free(zio->io_data, zio->io_size); | |
939 | } | |
940 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
941 | if (zio->io_error == 0) |
942 | vps->vps_writeable = 1; | |
943 | zio_buf_free(zio->io_data, zio->io_size); | |
944 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 945 | zio_t *pio; |
b128c09f BB |
946 | |
947 | vd->vdev_cant_read |= !vps->vps_readable; | |
948 | vd->vdev_cant_write |= !vps->vps_writeable; | |
949 | ||
950 | if (vdev_readable(vd) && | |
fb5f0bc8 | 951 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
952 | zio->io_error = 0; |
953 | } else { | |
954 | ASSERT(zio->io_error != 0); | |
955 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 956 | spa, vd, NULL, 0, 0); |
2e528b49 | 957 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f | 958 | } |
d164b209 BB |
959 | |
960 | mutex_enter(&vd->vdev_probe_lock); | |
961 | ASSERT(vd->vdev_probe_zio == zio); | |
962 | vd->vdev_probe_zio = NULL; | |
963 | mutex_exit(&vd->vdev_probe_lock); | |
964 | ||
965 | while ((pio = zio_walk_parents(zio)) != NULL) | |
966 | if (!vdev_accessible(vd, pio)) | |
2e528b49 | 967 | pio->io_error = SET_ERROR(ENXIO); |
d164b209 | 968 | |
b128c09f BB |
969 | kmem_free(vps, sizeof (*vps)); |
970 | } | |
971 | } | |
34dc7c2f | 972 | |
b128c09f | 973 | /* |
d3cc8b15 WA |
974 | * Determine whether this device is accessible. |
975 | * | |
976 | * Read and write to several known locations: the pad regions of each | |
977 | * vdev label but the first, which we leave alone in case it contains | |
978 | * a VTOC. | |
b128c09f BB |
979 | */ |
980 | zio_t * | |
d164b209 | 981 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
982 | { |
983 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
984 | vdev_probe_stats_t *vps = NULL; |
985 | zio_t *pio; | |
d6320ddb | 986 | int l; |
d164b209 BB |
987 | |
988 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 989 | |
d164b209 BB |
990 | /* |
991 | * Don't probe the probe. | |
992 | */ | |
993 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
994 | return (NULL); | |
b128c09f | 995 | |
d164b209 BB |
996 | /* |
997 | * To prevent 'probe storms' when a device fails, we create | |
998 | * just one probe i/o at a time. All zios that want to probe | |
999 | * this vdev will become parents of the probe io. | |
1000 | */ | |
1001 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 1002 | |
d164b209 | 1003 | if ((pio = vd->vdev_probe_zio) == NULL) { |
b8d06fca | 1004 | vps = kmem_zalloc(sizeof (*vps), KM_PUSHPAGE); |
d164b209 BB |
1005 | |
1006 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1007 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1008 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1009 | |
1010 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1011 | /* | |
1012 | * vdev_cant_read and vdev_cant_write can only | |
1013 | * transition from TRUE to FALSE when we have the | |
1014 | * SCL_ZIO lock as writer; otherwise they can only | |
1015 | * transition from FALSE to TRUE. This ensures that | |
1016 | * any zio looking at these values can assume that | |
1017 | * failures persist for the life of the I/O. That's | |
1018 | * important because when a device has intermittent | |
1019 | * connectivity problems, we want to ensure that | |
1020 | * they're ascribed to the device (ENXIO) and not | |
1021 | * the zio (EIO). | |
1022 | * | |
1023 | * Since we hold SCL_ZIO as writer here, clear both | |
1024 | * values so the probe can reevaluate from first | |
1025 | * principles. | |
1026 | */ | |
1027 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1028 | vd->vdev_cant_read = B_FALSE; | |
1029 | vd->vdev_cant_write = B_FALSE; | |
1030 | } | |
1031 | ||
1032 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1033 | vdev_probe_done, vps, | |
1034 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1035 | ||
428870ff BB |
1036 | /* |
1037 | * We can't change the vdev state in this context, so we | |
1038 | * kick off an async task to do it on our behalf. | |
1039 | */ | |
d164b209 BB |
1040 | if (zio != NULL) { |
1041 | vd->vdev_probe_wanted = B_TRUE; | |
1042 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1043 | } | |
b128c09f BB |
1044 | } |
1045 | ||
d164b209 BB |
1046 | if (zio != NULL) |
1047 | zio_add_child(zio, pio); | |
b128c09f | 1048 | |
d164b209 | 1049 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1050 | |
d164b209 BB |
1051 | if (vps == NULL) { |
1052 | ASSERT(zio != NULL); | |
1053 | return (NULL); | |
1054 | } | |
b128c09f | 1055 | |
d6320ddb | 1056 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1057 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1058 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
1059 | offsetof(vdev_label_t, vl_pad2)), |
1060 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
1061 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1062 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1063 | } | |
1064 | ||
d164b209 BB |
1065 | if (zio == NULL) |
1066 | return (pio); | |
1067 | ||
1068 | zio_nowait(pio); | |
1069 | return (NULL); | |
34dc7c2f BB |
1070 | } |
1071 | ||
45d1cae3 BB |
1072 | static void |
1073 | vdev_open_child(void *arg) | |
1074 | { | |
1075 | vdev_t *vd = arg; | |
1076 | ||
1077 | vd->vdev_open_thread = curthread; | |
1078 | vd->vdev_open_error = vdev_open(vd); | |
1079 | vd->vdev_open_thread = NULL; | |
1080 | } | |
1081 | ||
6c285672 | 1082 | static boolean_t |
428870ff BB |
1083 | vdev_uses_zvols(vdev_t *vd) |
1084 | { | |
d6320ddb BB |
1085 | int c; |
1086 | ||
6c285672 JL |
1087 | #ifdef _KERNEL |
1088 | if (zvol_is_zvol(vd->vdev_path)) | |
428870ff | 1089 | return (B_TRUE); |
6c285672 JL |
1090 | #endif |
1091 | ||
d6320ddb | 1092 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1093 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1094 | return (B_TRUE); | |
6c285672 | 1095 | |
428870ff BB |
1096 | return (B_FALSE); |
1097 | } | |
1098 | ||
45d1cae3 BB |
1099 | void |
1100 | vdev_open_children(vdev_t *vd) | |
1101 | { | |
1102 | taskq_t *tq; | |
1103 | int children = vd->vdev_children; | |
d6320ddb | 1104 | int c; |
45d1cae3 | 1105 | |
428870ff BB |
1106 | /* |
1107 | * in order to handle pools on top of zvols, do the opens | |
1108 | * in a single thread so that the same thread holds the | |
1109 | * spa_namespace_lock | |
1110 | */ | |
1111 | if (vdev_uses_zvols(vd)) { | |
d6320ddb | 1112 | for (c = 0; c < children; c++) |
428870ff BB |
1113 | vd->vdev_child[c]->vdev_open_error = |
1114 | vdev_open(vd->vdev_child[c]); | |
1115 | return; | |
1116 | } | |
45d1cae3 BB |
1117 | tq = taskq_create("vdev_open", children, minclsyspri, |
1118 | children, children, TASKQ_PREPOPULATE); | |
1119 | ||
d6320ddb | 1120 | for (c = 0; c < children; c++) |
45d1cae3 | 1121 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], |
d6320ddb | 1122 | TQ_SLEEP) != 0); |
45d1cae3 BB |
1123 | |
1124 | taskq_destroy(tq); | |
1125 | } | |
1126 | ||
34dc7c2f BB |
1127 | /* |
1128 | * Prepare a virtual device for access. | |
1129 | */ | |
1130 | int | |
1131 | vdev_open(vdev_t *vd) | |
1132 | { | |
fb5f0bc8 | 1133 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1134 | int error; |
34dc7c2f | 1135 | uint64_t osize = 0; |
1bd201e7 CS |
1136 | uint64_t max_osize = 0; |
1137 | uint64_t asize, max_asize, psize; | |
34dc7c2f | 1138 | uint64_t ashift = 0; |
d6320ddb | 1139 | int c; |
34dc7c2f | 1140 | |
45d1cae3 BB |
1141 | ASSERT(vd->vdev_open_thread == curthread || |
1142 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1143 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1144 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1145 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1146 | ||
34dc7c2f | 1147 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1148 | vd->vdev_cant_read = B_FALSE; |
1149 | vd->vdev_cant_write = B_FALSE; | |
1150 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1151 | |
428870ff BB |
1152 | /* |
1153 | * If this vdev is not removed, check its fault status. If it's | |
1154 | * faulted, bail out of the open. | |
1155 | */ | |
34dc7c2f BB |
1156 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1157 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1158 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1159 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1160 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1161 | vd->vdev_label_aux); |
2e528b49 | 1162 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1163 | } else if (vd->vdev_offline) { |
1164 | ASSERT(vd->vdev_children == 0); | |
1165 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
2e528b49 | 1166 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1167 | } |
1168 | ||
1bd201e7 | 1169 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1170 | |
428870ff BB |
1171 | /* |
1172 | * Reset the vdev_reopening flag so that we actually close | |
1173 | * the vdev on error. | |
1174 | */ | |
1175 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1176 | if (zio_injection_enabled && error == 0) |
9babb374 | 1177 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1178 | |
1179 | if (error) { | |
1180 | if (vd->vdev_removed && | |
1181 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1182 | vd->vdev_removed = B_FALSE; | |
1183 | ||
1184 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1185 | vd->vdev_stat.vs_aux); | |
1186 | return (error); | |
1187 | } | |
1188 | ||
1189 | vd->vdev_removed = B_FALSE; | |
1190 | ||
428870ff BB |
1191 | /* |
1192 | * Recheck the faulted flag now that we have confirmed that | |
1193 | * the vdev is accessible. If we're faulted, bail. | |
1194 | */ | |
1195 | if (vd->vdev_faulted) { | |
1196 | ASSERT(vd->vdev_children == 0); | |
1197 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1198 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1199 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1200 | vd->vdev_label_aux); | |
2e528b49 | 1201 | return (SET_ERROR(ENXIO)); |
428870ff BB |
1202 | } |
1203 | ||
34dc7c2f BB |
1204 | if (vd->vdev_degraded) { |
1205 | ASSERT(vd->vdev_children == 0); | |
1206 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1207 | VDEV_AUX_ERR_EXCEEDED); | |
1208 | } else { | |
428870ff | 1209 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1210 | } |
1211 | ||
428870ff BB |
1212 | /* |
1213 | * For hole or missing vdevs we just return success. | |
1214 | */ | |
1215 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1216 | return (0); | |
1217 | ||
d6320ddb | 1218 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1219 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1220 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1221 | VDEV_AUX_NONE); | |
1222 | break; | |
1223 | } | |
9babb374 | 1224 | } |
34dc7c2f BB |
1225 | |
1226 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1227 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1228 | |
1229 | if (vd->vdev_children == 0) { | |
1230 | if (osize < SPA_MINDEVSIZE) { | |
1231 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1232 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1233 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1234 | } |
1235 | psize = osize; | |
1236 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1237 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1238 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1239 | } else { |
1240 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1241 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1242 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1243 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1244 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1245 | } |
1246 | psize = 0; | |
1247 | asize = osize; | |
1bd201e7 | 1248 | max_asize = max_osize; |
34dc7c2f BB |
1249 | } |
1250 | ||
1251 | vd->vdev_psize = psize; | |
1252 | ||
9babb374 BB |
1253 | /* |
1254 | * Make sure the allocatable size hasn't shrunk. | |
1255 | */ | |
1256 | if (asize < vd->vdev_min_asize) { | |
1257 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1258 | VDEV_AUX_BAD_LABEL); | |
2e528b49 | 1259 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
1260 | } |
1261 | ||
34dc7c2f BB |
1262 | if (vd->vdev_asize == 0) { |
1263 | /* | |
1264 | * This is the first-ever open, so use the computed values. | |
b28e57cb | 1265 | * For compatibility, a different ashift can be requested. |
34dc7c2f BB |
1266 | */ |
1267 | vd->vdev_asize = asize; | |
1bd201e7 | 1268 | vd->vdev_max_asize = max_asize; |
b28e57cb MM |
1269 | if (vd->vdev_ashift == 0) |
1270 | vd->vdev_ashift = ashift; | |
34dc7c2f BB |
1271 | } else { |
1272 | /* | |
32a9872b GW |
1273 | * Detect if the alignment requirement has increased. |
1274 | * We don't want to make the pool unavailable, just | |
1275 | * post an event instead. | |
34dc7c2f | 1276 | */ |
32a9872b GW |
1277 | if (ashift > vd->vdev_top->vdev_ashift && |
1278 | vd->vdev_ops->vdev_op_leaf) { | |
1279 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
1280 | spa, vd, NULL, 0, 0); | |
34dc7c2f | 1281 | } |
32a9872b | 1282 | |
1bd201e7 | 1283 | vd->vdev_max_asize = max_asize; |
9babb374 | 1284 | } |
34dc7c2f | 1285 | |
9babb374 BB |
1286 | /* |
1287 | * If all children are healthy and the asize has increased, | |
1288 | * then we've experienced dynamic LUN growth. If automatic | |
1289 | * expansion is enabled then use the additional space. | |
1290 | */ | |
1291 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1292 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1293 | vd->vdev_asize = asize; | |
34dc7c2f | 1294 | |
9babb374 | 1295 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1296 | |
1297 | /* | |
1298 | * Ensure we can issue some IO before declaring the | |
1299 | * vdev open for business. | |
1300 | */ | |
b128c09f BB |
1301 | if (vd->vdev_ops->vdev_op_leaf && |
1302 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1303 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1304 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1305 | return (error); |
1306 | } | |
1307 | ||
34dc7c2f | 1308 | /* |
b128c09f | 1309 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1310 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1311 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1312 | */ |
fb5f0bc8 BB |
1313 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1314 | vdev_resilver_needed(vd, NULL, NULL)) | |
1315 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1316 | |
1317 | return (0); | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Called once the vdevs are all opened, this routine validates the label | |
1322 | * contents. This needs to be done before vdev_load() so that we don't | |
1323 | * inadvertently do repair I/Os to the wrong device. | |
1324 | * | |
c7f2d69d GW |
1325 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1326 | * if the machine crashed during a re-guid the new guid might have been written | |
1327 | * to all of the vdev labels, but not the cached config. The strict check | |
1328 | * will be performed when the pool is opened again using the mos config. | |
1329 | * | |
34dc7c2f BB |
1330 | * This function will only return failure if one of the vdevs indicates that it |
1331 | * has since been destroyed or exported. This is only possible if | |
1332 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1333 | * will be updated but the function will return 0. | |
1334 | */ | |
1335 | int | |
c7f2d69d | 1336 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1337 | { |
1338 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1339 | nvlist_t *label; |
428870ff | 1340 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1341 | uint64_t state; |
d6320ddb | 1342 | int c; |
34dc7c2f | 1343 | |
d6320ddb | 1344 | for (c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1345 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
2e528b49 | 1346 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
1347 | |
1348 | /* | |
1349 | * If the device has already failed, or was marked offline, don't do | |
1350 | * any further validation. Otherwise, label I/O will fail and we will | |
1351 | * overwrite the previous state. | |
1352 | */ | |
b128c09f | 1353 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1354 | uint64_t aux_guid = 0; |
1355 | nvlist_t *nvl; | |
295304be GW |
1356 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? |
1357 | spa_last_synced_txg(spa) : -1ULL; | |
34dc7c2f | 1358 | |
3bc7e0fb | 1359 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { |
34dc7c2f BB |
1360 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
1361 | VDEV_AUX_BAD_LABEL); | |
1362 | return (0); | |
1363 | } | |
1364 | ||
428870ff BB |
1365 | /* |
1366 | * Determine if this vdev has been split off into another | |
1367 | * pool. If so, then refuse to open it. | |
1368 | */ | |
1369 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1370 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1371 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1372 | VDEV_AUX_SPLIT_POOL); | |
1373 | nvlist_free(label); | |
1374 | return (0); | |
1375 | } | |
1376 | ||
c7f2d69d GW |
1377 | if (strict && (nvlist_lookup_uint64(label, |
1378 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1379 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1380 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1381 | VDEV_AUX_CORRUPT_DATA); | |
1382 | nvlist_free(label); | |
1383 | return (0); | |
1384 | } | |
1385 | ||
428870ff BB |
1386 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1387 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1388 | &aux_guid) != 0) | |
1389 | aux_guid = 0; | |
1390 | ||
b128c09f BB |
1391 | /* |
1392 | * If this vdev just became a top-level vdev because its | |
1393 | * sibling was detached, it will have adopted the parent's | |
1394 | * vdev guid -- but the label may or may not be on disk yet. | |
1395 | * Fortunately, either version of the label will have the | |
1396 | * same top guid, so if we're a top-level vdev, we can | |
1397 | * safely compare to that instead. | |
428870ff BB |
1398 | * |
1399 | * If we split this vdev off instead, then we also check the | |
1400 | * original pool's guid. We don't want to consider the vdev | |
1401 | * corrupt if it is partway through a split operation. | |
b128c09f | 1402 | */ |
34dc7c2f | 1403 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1404 | &guid) != 0 || |
1405 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1406 | &top_guid) != 0 || | |
428870ff | 1407 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1408 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1409 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1410 | VDEV_AUX_CORRUPT_DATA); | |
1411 | nvlist_free(label); | |
1412 | return (0); | |
1413 | } | |
1414 | ||
1415 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1416 | &state) != 0) { | |
1417 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1418 | VDEV_AUX_CORRUPT_DATA); | |
1419 | nvlist_free(label); | |
1420 | return (0); | |
1421 | } | |
1422 | ||
1423 | nvlist_free(label); | |
1424 | ||
45d1cae3 | 1425 | /* |
572e2857 | 1426 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1427 | * state of the pool. |
1428 | */ | |
572e2857 | 1429 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1430 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f | 1431 | state != POOL_STATE_ACTIVE) |
2e528b49 | 1432 | return (SET_ERROR(EBADF)); |
34dc7c2f | 1433 | |
b128c09f BB |
1434 | /* |
1435 | * If we were able to open and validate a vdev that was | |
1436 | * previously marked permanently unavailable, clear that state | |
1437 | * now. | |
1438 | */ | |
1439 | if (vd->vdev_not_present) | |
1440 | vd->vdev_not_present = 0; | |
1441 | } | |
34dc7c2f BB |
1442 | |
1443 | return (0); | |
1444 | } | |
1445 | ||
1446 | /* | |
1447 | * Close a virtual device. | |
1448 | */ | |
1449 | void | |
1450 | vdev_close(vdev_t *vd) | |
1451 | { | |
428870ff | 1452 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1453 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1454 | |
1455 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1456 | ||
428870ff BB |
1457 | /* |
1458 | * If our parent is reopening, then we are as well, unless we are | |
1459 | * going offline. | |
1460 | */ | |
1461 | if (pvd != NULL && pvd->vdev_reopening) | |
1462 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1463 | ||
34dc7c2f BB |
1464 | vd->vdev_ops->vdev_op_close(vd); |
1465 | ||
1466 | vdev_cache_purge(vd); | |
1467 | ||
1468 | /* | |
9babb374 | 1469 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1470 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1471 | * it's still faulted. | |
1472 | */ | |
1473 | vd->vdev_prevstate = vd->vdev_state; | |
1474 | ||
1475 | if (vd->vdev_offline) | |
1476 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1477 | else | |
1478 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1479 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1480 | } | |
1481 | ||
428870ff BB |
1482 | void |
1483 | vdev_hold(vdev_t *vd) | |
1484 | { | |
1485 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1486 | int c; |
428870ff BB |
1487 | |
1488 | ASSERT(spa_is_root(spa)); | |
1489 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1490 | return; | |
1491 | ||
d6320ddb | 1492 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1493 | vdev_hold(vd->vdev_child[c]); |
1494 | ||
1495 | if (vd->vdev_ops->vdev_op_leaf) | |
1496 | vd->vdev_ops->vdev_op_hold(vd); | |
1497 | } | |
1498 | ||
1499 | void | |
1500 | vdev_rele(vdev_t *vd) | |
1501 | { | |
d6320ddb | 1502 | int c; |
428870ff | 1503 | |
d6320ddb BB |
1504 | ASSERT(spa_is_root(vd->vdev_spa)); |
1505 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1506 | vdev_rele(vd->vdev_child[c]); |
1507 | ||
1508 | if (vd->vdev_ops->vdev_op_leaf) | |
1509 | vd->vdev_ops->vdev_op_rele(vd); | |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1514 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1515 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1516 | * If the leaf has never been opened then open it, as usual. | |
1517 | */ | |
34dc7c2f BB |
1518 | void |
1519 | vdev_reopen(vdev_t *vd) | |
1520 | { | |
1521 | spa_t *spa = vd->vdev_spa; | |
1522 | ||
b128c09f | 1523 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1524 | |
428870ff BB |
1525 | /* set the reopening flag unless we're taking the vdev offline */ |
1526 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1527 | vdev_close(vd); |
1528 | (void) vdev_open(vd); | |
1529 | ||
1530 | /* | |
1531 | * Call vdev_validate() here to make sure we have the same device. | |
1532 | * Otherwise, a device with an invalid label could be successfully | |
1533 | * opened in response to vdev_reopen(). | |
1534 | */ | |
b128c09f BB |
1535 | if (vd->vdev_aux) { |
1536 | (void) vdev_validate_aux(vd); | |
1537 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1538 | vd->vdev_aux == &spa->spa_l2cache && |
1539 | !l2arc_vdev_present(vd)) | |
1540 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1541 | } else { |
295304be | 1542 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1543 | } |
34dc7c2f BB |
1544 | |
1545 | /* | |
1546 | * Reassess parent vdev's health. | |
1547 | */ | |
1548 | vdev_propagate_state(vd); | |
1549 | } | |
1550 | ||
1551 | int | |
1552 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1553 | { | |
1554 | int error; | |
1555 | ||
1556 | /* | |
1557 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1558 | * For a create, however, we want to fail the request if | |
1559 | * there are any components we can't open. | |
1560 | */ | |
1561 | error = vdev_open(vd); | |
1562 | ||
1563 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1564 | vdev_close(vd); | |
1565 | return (error ? error : ENXIO); | |
1566 | } | |
1567 | ||
1568 | /* | |
93cf2076 | 1569 | * Recursively load DTLs and initialize all labels. |
34dc7c2f | 1570 | */ |
93cf2076 GW |
1571 | if ((error = vdev_dtl_load(vd)) != 0 || |
1572 | (error = vdev_label_init(vd, txg, isreplacing ? | |
34dc7c2f BB |
1573 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { |
1574 | vdev_close(vd); | |
1575 | return (error); | |
1576 | } | |
1577 | ||
1578 | return (0); | |
1579 | } | |
1580 | ||
34dc7c2f | 1581 | void |
9babb374 | 1582 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1583 | { |
1584 | /* | |
1585 | * Aim for roughly 200 metaslabs per vdev. | |
1586 | */ | |
9bd274dd | 1587 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / 200); |
34dc7c2f | 1588 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); |
34dc7c2f BB |
1589 | } |
1590 | ||
1591 | void | |
1592 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1593 | { | |
1594 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1595 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1596 | ASSERT(ISP2(flags)); |
572e2857 | 1597 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1598 | |
1599 | if (flags & VDD_METASLAB) | |
1600 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1601 | ||
1602 | if (flags & VDD_DTL) | |
1603 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1604 | ||
1605 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1606 | } | |
1607 | ||
93cf2076 GW |
1608 | void |
1609 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1610 | { | |
1611 | int c; | |
1612 | ||
1613 | for (c = 0; c < vd->vdev_children; c++) | |
1614 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); | |
1615 | ||
1616 | if (vd->vdev_ops->vdev_op_leaf) | |
1617 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1618 | } | |
1619 | ||
fb5f0bc8 BB |
1620 | /* |
1621 | * DTLs. | |
1622 | * | |
1623 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1624 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1625 | * |
1626 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1627 | * | |
1628 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1629 | * | |
1630 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1631 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1632 | * txgs that was scrubbed. | |
1633 | * | |
1634 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1635 | * persistent errors or just some device being offline. | |
1636 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1637 | * maintained; it's only computed when needed, typically to | |
1638 | * determine whether a device can be detached. | |
1639 | * | |
1640 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1641 | * either has the data or it doesn't. | |
1642 | * | |
1643 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1644 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1645 | * if any child is less than fully replicated, then so is its parent. | |
1646 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1647 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1648 | * those are the txgs we don't have enough replication to read. For example, | |
1649 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1650 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1651 | * two child DTL_MISSING maps. | |
1652 | * | |
1653 | * It should be clear from the above that to compute the DTLs and outage maps | |
1654 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1655 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1656 | * a configuration change, we generate all other DTLs from first principles. | |
1657 | */ | |
34dc7c2f | 1658 | void |
fb5f0bc8 | 1659 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1660 | { |
93cf2076 | 1661 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1662 | |
1663 | ASSERT(t < DTL_TYPES); | |
1664 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1665 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1666 | |
93cf2076 GW |
1667 | mutex_enter(rt->rt_lock); |
1668 | if (!range_tree_contains(rt, txg, size)) | |
1669 | range_tree_add(rt, txg, size); | |
1670 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1671 | } |
1672 | ||
fb5f0bc8 BB |
1673 | boolean_t |
1674 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1675 | { |
93cf2076 | 1676 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 | 1677 | boolean_t dirty = B_FALSE; |
34dc7c2f | 1678 | |
fb5f0bc8 BB |
1679 | ASSERT(t < DTL_TYPES); |
1680 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f | 1681 | |
93cf2076 GW |
1682 | mutex_enter(rt->rt_lock); |
1683 | if (range_tree_space(rt) != 0) | |
1684 | dirty = range_tree_contains(rt, txg, size); | |
1685 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1686 | |
1687 | return (dirty); | |
1688 | } | |
1689 | ||
fb5f0bc8 BB |
1690 | boolean_t |
1691 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1692 | { | |
93cf2076 | 1693 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1694 | boolean_t empty; |
1695 | ||
93cf2076 GW |
1696 | mutex_enter(rt->rt_lock); |
1697 | empty = (range_tree_space(rt) == 0); | |
1698 | mutex_exit(rt->rt_lock); | |
fb5f0bc8 BB |
1699 | |
1700 | return (empty); | |
1701 | } | |
1702 | ||
5d1f7fb6 GW |
1703 | /* |
1704 | * Returns the lowest txg in the DTL range. | |
1705 | */ | |
1706 | static uint64_t | |
1707 | vdev_dtl_min(vdev_t *vd) | |
1708 | { | |
93cf2076 | 1709 | range_seg_t *rs; |
5d1f7fb6 GW |
1710 | |
1711 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1712 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1713 | ASSERT0(vd->vdev_children); |
1714 | ||
93cf2076 GW |
1715 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1716 | return (rs->rs_start - 1); | |
5d1f7fb6 GW |
1717 | } |
1718 | ||
1719 | /* | |
1720 | * Returns the highest txg in the DTL. | |
1721 | */ | |
1722 | static uint64_t | |
1723 | vdev_dtl_max(vdev_t *vd) | |
1724 | { | |
93cf2076 | 1725 | range_seg_t *rs; |
5d1f7fb6 GW |
1726 | |
1727 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1728 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1729 | ASSERT0(vd->vdev_children); |
1730 | ||
93cf2076 GW |
1731 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1732 | return (rs->rs_end); | |
5d1f7fb6 GW |
1733 | } |
1734 | ||
1735 | /* | |
1736 | * Determine if a resilvering vdev should remove any DTL entries from | |
1737 | * its range. If the vdev was resilvering for the entire duration of the | |
1738 | * scan then it should excise that range from its DTLs. Otherwise, this | |
1739 | * vdev is considered partially resilvered and should leave its DTL | |
1740 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
1741 | * excise the DTLs. | |
1742 | */ | |
1743 | static boolean_t | |
1744 | vdev_dtl_should_excise(vdev_t *vd) | |
1745 | { | |
1746 | spa_t *spa = vd->vdev_spa; | |
1747 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1748 | ||
1749 | ASSERT0(scn->scn_phys.scn_errors); | |
1750 | ASSERT0(vd->vdev_children); | |
1751 | ||
1752 | if (vd->vdev_resilver_txg == 0 || | |
93cf2076 | 1753 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) |
5d1f7fb6 GW |
1754 | return (B_TRUE); |
1755 | ||
1756 | /* | |
1757 | * When a resilver is initiated the scan will assign the scn_max_txg | |
1758 | * value to the highest txg value that exists in all DTLs. If this | |
1759 | * device's max DTL is not part of this scan (i.e. it is not in | |
1760 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
1761 | * for excision. | |
1762 | */ | |
1763 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
1764 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
1765 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
1766 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
1767 | return (B_TRUE); | |
1768 | } | |
1769 | return (B_FALSE); | |
1770 | } | |
1771 | ||
34dc7c2f BB |
1772 | /* |
1773 | * Reassess DTLs after a config change or scrub completion. | |
1774 | */ | |
1775 | void | |
1776 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1777 | { | |
1778 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1779 | avl_tree_t reftree; |
d6320ddb | 1780 | int c, t, minref; |
34dc7c2f | 1781 | |
fb5f0bc8 | 1782 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1783 | |
d6320ddb | 1784 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1785 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1786 | scrub_txg, scrub_done); | |
1787 | ||
428870ff | 1788 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1789 | return; |
1790 | ||
1791 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1792 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1793 | ||
34dc7c2f | 1794 | mutex_enter(&vd->vdev_dtl_lock); |
5d1f7fb6 GW |
1795 | |
1796 | /* | |
1797 | * If we've completed a scan cleanly then determine | |
1798 | * if this vdev should remove any DTLs. We only want to | |
1799 | * excise regions on vdevs that were available during | |
1800 | * the entire duration of this scan. | |
1801 | */ | |
b128c09f | 1802 | if (scrub_txg != 0 && |
428870ff | 1803 | (spa->spa_scrub_started || |
5d1f7fb6 GW |
1804 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && |
1805 | vdev_dtl_should_excise(vd)) { | |
b128c09f BB |
1806 | /* |
1807 | * We completed a scrub up to scrub_txg. If we | |
1808 | * did it without rebooting, then the scrub dtl | |
1809 | * will be valid, so excise the old region and | |
1810 | * fold in the scrub dtl. Otherwise, leave the | |
1811 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1812 | * |
1813 | * There's little trick here: to excise the beginning | |
1814 | * of the DTL_MISSING map, we put it into a reference | |
1815 | * tree and then add a segment with refcnt -1 that | |
1816 | * covers the range [0, scrub_txg). This means | |
1817 | * that each txg in that range has refcnt -1 or 0. | |
1818 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1819 | * entries in the range [0, scrub_txg) will have a | |
1820 | * positive refcnt -- either 1 or 2. We then convert | |
1821 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1822 | */ |
93cf2076 GW |
1823 | space_reftree_create(&reftree); |
1824 | space_reftree_add_map(&reftree, | |
1825 | vd->vdev_dtl[DTL_MISSING], 1); | |
1826 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
1827 | space_reftree_add_map(&reftree, | |
1828 | vd->vdev_dtl[DTL_SCRUB], 2); | |
1829 | space_reftree_generate_map(&reftree, | |
1830 | vd->vdev_dtl[DTL_MISSING], 1); | |
1831 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1832 | } |
93cf2076 GW |
1833 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1834 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1835 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1836 | if (scrub_done) |
93cf2076 GW |
1837 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1838 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
fb5f0bc8 | 1839 | if (!vdev_readable(vd)) |
93cf2076 | 1840 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); |
fb5f0bc8 | 1841 | else |
93cf2076 GW |
1842 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], |
1843 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
5d1f7fb6 GW |
1844 | |
1845 | /* | |
1846 | * If the vdev was resilvering and no longer has any | |
1847 | * DTLs then reset its resilvering flag. | |
1848 | */ | |
1849 | if (vd->vdev_resilver_txg != 0 && | |
93cf2076 GW |
1850 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && |
1851 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) | |
5d1f7fb6 GW |
1852 | vd->vdev_resilver_txg = 0; |
1853 | ||
34dc7c2f | 1854 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1855 | |
34dc7c2f BB |
1856 | if (txg != 0) |
1857 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1858 | return; | |
1859 | } | |
1860 | ||
34dc7c2f | 1861 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1862 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
1863 | int c; |
1864 | ||
428870ff BB |
1865 | /* account for child's outage in parent's missing map */ |
1866 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1867 | if (t == DTL_SCRUB) |
1868 | continue; /* leaf vdevs only */ | |
1869 | if (t == DTL_PARTIAL) | |
1870 | minref = 1; /* i.e. non-zero */ | |
1871 | else if (vd->vdev_nparity != 0) | |
1872 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1873 | else | |
1874 | minref = vd->vdev_children; /* any kind of mirror */ | |
93cf2076 | 1875 | space_reftree_create(&reftree); |
d6320ddb | 1876 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1877 | vdev_t *cvd = vd->vdev_child[c]; |
1878 | mutex_enter(&cvd->vdev_dtl_lock); | |
93cf2076 | 1879 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1880 | mutex_exit(&cvd->vdev_dtl_lock); |
1881 | } | |
93cf2076 GW |
1882 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); |
1883 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1884 | } |
fb5f0bc8 | 1885 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1886 | } |
1887 | ||
93cf2076 | 1888 | int |
34dc7c2f BB |
1889 | vdev_dtl_load(vdev_t *vd) |
1890 | { | |
1891 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1892 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
1893 | int error = 0; |
1894 | int c; | |
34dc7c2f | 1895 | |
93cf2076 GW |
1896 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { |
1897 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 1898 | |
93cf2076 GW |
1899 | error = space_map_open(&vd->vdev_dtl_sm, mos, |
1900 | vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); | |
1901 | if (error) | |
1902 | return (error); | |
1903 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f | 1904 | |
93cf2076 | 1905 | mutex_enter(&vd->vdev_dtl_lock); |
428870ff | 1906 | |
93cf2076 GW |
1907 | /* |
1908 | * Now that we've opened the space_map we need to update | |
1909 | * the in-core DTL. | |
1910 | */ | |
1911 | space_map_update(vd->vdev_dtl_sm); | |
34dc7c2f | 1912 | |
93cf2076 GW |
1913 | error = space_map_load(vd->vdev_dtl_sm, |
1914 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
1915 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 1916 | |
93cf2076 GW |
1917 | return (error); |
1918 | } | |
1919 | ||
1920 | for (c = 0; c < vd->vdev_children; c++) { | |
1921 | error = vdev_dtl_load(vd->vdev_child[c]); | |
1922 | if (error != 0) | |
1923 | break; | |
1924 | } | |
34dc7c2f BB |
1925 | |
1926 | return (error); | |
1927 | } | |
1928 | ||
1929 | void | |
1930 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1931 | { | |
1932 | spa_t *spa = vd->vdev_spa; | |
93cf2076 | 1933 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; |
34dc7c2f | 1934 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
1935 | range_tree_t *rtsync; |
1936 | kmutex_t rtlock; | |
34dc7c2f | 1937 | dmu_tx_t *tx; |
93cf2076 | 1938 | uint64_t object = space_map_object(vd->vdev_dtl_sm); |
34dc7c2f | 1939 | |
428870ff | 1940 | ASSERT(!vd->vdev_ishole); |
93cf2076 | 1941 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
428870ff | 1942 | |
34dc7c2f BB |
1943 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1944 | ||
93cf2076 GW |
1945 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { |
1946 | mutex_enter(&vd->vdev_dtl_lock); | |
1947 | space_map_free(vd->vdev_dtl_sm, tx); | |
1948 | space_map_close(vd->vdev_dtl_sm); | |
1949 | vd->vdev_dtl_sm = NULL; | |
1950 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 1951 | dmu_tx_commit(tx); |
34dc7c2f BB |
1952 | return; |
1953 | } | |
1954 | ||
93cf2076 GW |
1955 | if (vd->vdev_dtl_sm == NULL) { |
1956 | uint64_t new_object; | |
1957 | ||
1958 | new_object = space_map_alloc(mos, tx); | |
1959 | VERIFY3U(new_object, !=, 0); | |
1960 | ||
1961 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
1962 | 0, -1ULL, 0, &vd->vdev_dtl_lock)); | |
1963 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f BB |
1964 | } |
1965 | ||
93cf2076 | 1966 | mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 1967 | |
93cf2076 | 1968 | rtsync = range_tree_create(NULL, NULL, &rtlock); |
34dc7c2f | 1969 | |
93cf2076 | 1970 | mutex_enter(&rtlock); |
34dc7c2f BB |
1971 | |
1972 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 1973 | range_tree_walk(rt, range_tree_add, rtsync); |
34dc7c2f BB |
1974 | mutex_exit(&vd->vdev_dtl_lock); |
1975 | ||
93cf2076 GW |
1976 | space_map_truncate(vd->vdev_dtl_sm, tx); |
1977 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
1978 | range_tree_vacate(rtsync, NULL, NULL); | |
34dc7c2f | 1979 | |
93cf2076 | 1980 | range_tree_destroy(rtsync); |
34dc7c2f | 1981 | |
93cf2076 GW |
1982 | mutex_exit(&rtlock); |
1983 | mutex_destroy(&rtlock); | |
34dc7c2f | 1984 | |
93cf2076 GW |
1985 | /* |
1986 | * If the object for the space map has changed then dirty | |
1987 | * the top level so that we update the config. | |
1988 | */ | |
1989 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
1990 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
1991 | "new object %llu", txg, spa_name(spa), object, | |
1992 | space_map_object(vd->vdev_dtl_sm)); | |
1993 | vdev_config_dirty(vd->vdev_top); | |
1994 | } | |
34dc7c2f BB |
1995 | |
1996 | dmu_tx_commit(tx); | |
93cf2076 GW |
1997 | |
1998 | mutex_enter(&vd->vdev_dtl_lock); | |
1999 | space_map_update(vd->vdev_dtl_sm); | |
2000 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f BB |
2001 | } |
2002 | ||
fb5f0bc8 BB |
2003 | /* |
2004 | * Determine whether the specified vdev can be offlined/detached/removed | |
2005 | * without losing data. | |
2006 | */ | |
2007 | boolean_t | |
2008 | vdev_dtl_required(vdev_t *vd) | |
2009 | { | |
2010 | spa_t *spa = vd->vdev_spa; | |
2011 | vdev_t *tvd = vd->vdev_top; | |
2012 | uint8_t cant_read = vd->vdev_cant_read; | |
2013 | boolean_t required; | |
2014 | ||
2015 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2016 | ||
2017 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2018 | return (B_TRUE); | |
2019 | ||
2020 | /* | |
2021 | * Temporarily mark the device as unreadable, and then determine | |
2022 | * whether this results in any DTL outages in the top-level vdev. | |
2023 | * If not, we can safely offline/detach/remove the device. | |
2024 | */ | |
2025 | vd->vdev_cant_read = B_TRUE; | |
2026 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2027 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2028 | vd->vdev_cant_read = cant_read; | |
2029 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2030 | ||
572e2857 BB |
2031 | if (!required && zio_injection_enabled) |
2032 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2033 | ||
fb5f0bc8 BB |
2034 | return (required); |
2035 | } | |
2036 | ||
b128c09f BB |
2037 | /* |
2038 | * Determine if resilver is needed, and if so the txg range. | |
2039 | */ | |
2040 | boolean_t | |
2041 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2042 | { | |
2043 | boolean_t needed = B_FALSE; | |
2044 | uint64_t thismin = UINT64_MAX; | |
2045 | uint64_t thismax = 0; | |
d6320ddb | 2046 | int c; |
b128c09f BB |
2047 | |
2048 | if (vd->vdev_children == 0) { | |
2049 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2050 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && |
fb5f0bc8 | 2051 | vdev_writeable(vd)) { |
b128c09f | 2052 | |
5d1f7fb6 GW |
2053 | thismin = vdev_dtl_min(vd); |
2054 | thismax = vdev_dtl_max(vd); | |
b128c09f BB |
2055 | needed = B_TRUE; |
2056 | } | |
2057 | mutex_exit(&vd->vdev_dtl_lock); | |
2058 | } else { | |
d6320ddb | 2059 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2060 | vdev_t *cvd = vd->vdev_child[c]; |
2061 | uint64_t cmin, cmax; | |
2062 | ||
2063 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2064 | thismin = MIN(thismin, cmin); | |
2065 | thismax = MAX(thismax, cmax); | |
2066 | needed = B_TRUE; | |
2067 | } | |
2068 | } | |
2069 | } | |
2070 | ||
2071 | if (needed && minp) { | |
2072 | *minp = thismin; | |
2073 | *maxp = thismax; | |
2074 | } | |
2075 | return (needed); | |
2076 | } | |
2077 | ||
34dc7c2f BB |
2078 | void |
2079 | vdev_load(vdev_t *vd) | |
2080 | { | |
d6320ddb BB |
2081 | int c; |
2082 | ||
34dc7c2f BB |
2083 | /* |
2084 | * Recursively load all children. | |
2085 | */ | |
d6320ddb | 2086 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2087 | vdev_load(vd->vdev_child[c]); |
2088 | ||
2089 | /* | |
2090 | * If this is a top-level vdev, initialize its metaslabs. | |
2091 | */ | |
428870ff | 2092 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
2093 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
2094 | vdev_metaslab_init(vd, 0) != 0)) | |
2095 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2096 | VDEV_AUX_CORRUPT_DATA); | |
2097 | ||
2098 | /* | |
2099 | * If this is a leaf vdev, load its DTL. | |
2100 | */ | |
2101 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
2102 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2103 | VDEV_AUX_CORRUPT_DATA); | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2108 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2109 | * we make sure that we can open the underlying device, then try to read the | |
2110 | * label, and make sure that the label is sane and that it hasn't been | |
2111 | * repurposed to another pool. | |
2112 | */ | |
2113 | int | |
2114 | vdev_validate_aux(vdev_t *vd) | |
2115 | { | |
2116 | nvlist_t *label; | |
2117 | uint64_t guid, version; | |
2118 | uint64_t state; | |
2119 | ||
b128c09f BB |
2120 | if (!vdev_readable(vd)) |
2121 | return (0); | |
2122 | ||
3bc7e0fb | 2123 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { |
34dc7c2f BB |
2124 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
2125 | VDEV_AUX_CORRUPT_DATA); | |
2126 | return (-1); | |
2127 | } | |
2128 | ||
2129 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
9ae529ec | 2130 | !SPA_VERSION_IS_SUPPORTED(version) || |
34dc7c2f BB |
2131 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || |
2132 | guid != vd->vdev_guid || | |
2133 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2134 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2135 | VDEV_AUX_CORRUPT_DATA); | |
2136 | nvlist_free(label); | |
2137 | return (-1); | |
2138 | } | |
2139 | ||
2140 | /* | |
2141 | * We don't actually check the pool state here. If it's in fact in | |
2142 | * use by another pool, we update this fact on the fly when requested. | |
2143 | */ | |
2144 | nvlist_free(label); | |
2145 | return (0); | |
2146 | } | |
2147 | ||
428870ff BB |
2148 | void |
2149 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2150 | { | |
2151 | spa_t *spa = vd->vdev_spa; | |
2152 | objset_t *mos = spa->spa_meta_objset; | |
2153 | dmu_tx_t *tx; | |
f3a7f661 | 2154 | int m, i; |
428870ff BB |
2155 | |
2156 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
2157 | ||
428870ff | 2158 | if (vd->vdev_ms != NULL) { |
f3a7f661 GW |
2159 | metaslab_group_t *mg = vd->vdev_mg; |
2160 | ||
2161 | metaslab_group_histogram_verify(mg); | |
2162 | metaslab_class_histogram_verify(mg->mg_class); | |
2163 | ||
d6320ddb | 2164 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2165 | metaslab_t *msp = vd->vdev_ms[m]; |
2166 | ||
93cf2076 | 2167 | if (msp == NULL || msp->ms_sm == NULL) |
428870ff BB |
2168 | continue; |
2169 | ||
93cf2076 | 2170 | mutex_enter(&msp->ms_lock); |
f3a7f661 GW |
2171 | /* |
2172 | * If the metaslab was not loaded when the vdev | |
2173 | * was removed then the histogram accounting may | |
2174 | * not be accurate. Update the histogram information | |
2175 | * here so that we ensure that the metaslab group | |
2176 | * and metaslab class are up-to-date. | |
2177 | */ | |
2178 | metaslab_group_histogram_remove(mg, msp); | |
2179 | ||
93cf2076 GW |
2180 | VERIFY0(space_map_allocated(msp->ms_sm)); |
2181 | space_map_free(msp->ms_sm, tx); | |
2182 | space_map_close(msp->ms_sm); | |
2183 | msp->ms_sm = NULL; | |
2184 | mutex_exit(&msp->ms_lock); | |
428870ff | 2185 | } |
f3a7f661 GW |
2186 | |
2187 | metaslab_group_histogram_verify(mg); | |
2188 | metaslab_class_histogram_verify(mg->mg_class); | |
2189 | for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) | |
2190 | ASSERT0(mg->mg_histogram[i]); | |
2191 | ||
428870ff BB |
2192 | } |
2193 | ||
2194 | if (vd->vdev_ms_array) { | |
2195 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2196 | vd->vdev_ms_array = 0; | |
428870ff BB |
2197 | } |
2198 | dmu_tx_commit(tx); | |
2199 | } | |
2200 | ||
34dc7c2f BB |
2201 | void |
2202 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2203 | { | |
2204 | metaslab_t *msp; | |
428870ff BB |
2205 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2206 | ||
2207 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2208 | |
c65aa5b2 | 2209 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2210 | metaslab_sync_done(msp, txg); |
428870ff BB |
2211 | |
2212 | if (reassess) | |
2213 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2214 | } |
2215 | ||
2216 | void | |
2217 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2218 | { | |
2219 | spa_t *spa = vd->vdev_spa; | |
2220 | vdev_t *lvd; | |
2221 | metaslab_t *msp; | |
2222 | dmu_tx_t *tx; | |
2223 | ||
428870ff BB |
2224 | ASSERT(!vd->vdev_ishole); |
2225 | ||
34dc7c2f BB |
2226 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2227 | ASSERT(vd == vd->vdev_top); | |
2228 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2229 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2230 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2231 | ASSERT(vd->vdev_ms_array != 0); | |
2232 | vdev_config_dirty(vd); | |
2233 | dmu_tx_commit(tx); | |
2234 | } | |
2235 | ||
428870ff BB |
2236 | /* |
2237 | * Remove the metadata associated with this vdev once it's empty. | |
2238 | */ | |
2239 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2240 | vdev_remove(vd, txg); | |
2241 | ||
34dc7c2f BB |
2242 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2243 | metaslab_sync(msp, txg); | |
2244 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2245 | } | |
2246 | ||
2247 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2248 | vdev_dtl_sync(lvd, txg); | |
2249 | ||
2250 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2251 | } | |
2252 | ||
2253 | uint64_t | |
2254 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2255 | { | |
2256 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2257 | } | |
2258 | ||
34dc7c2f BB |
2259 | /* |
2260 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2261 | * not be opened, and no I/O is attempted. | |
2262 | */ | |
2263 | int | |
428870ff | 2264 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2265 | { |
572e2857 | 2266 | vdev_t *vd, *tvd; |
34dc7c2f | 2267 | |
428870ff | 2268 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2269 | |
b128c09f BB |
2270 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2271 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2272 | |
34dc7c2f | 2273 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2274 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2275 | |
572e2857 BB |
2276 | tvd = vd->vdev_top; |
2277 | ||
428870ff BB |
2278 | /* |
2279 | * We don't directly use the aux state here, but if we do a | |
2280 | * vdev_reopen(), we need this value to be present to remember why we | |
2281 | * were faulted. | |
2282 | */ | |
2283 | vd->vdev_label_aux = aux; | |
2284 | ||
34dc7c2f BB |
2285 | /* |
2286 | * Faulted state takes precedence over degraded. | |
2287 | */ | |
428870ff | 2288 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2289 | vd->vdev_faulted = 1ULL; |
2290 | vd->vdev_degraded = 0ULL; | |
428870ff | 2291 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2292 | |
2293 | /* | |
428870ff BB |
2294 | * If this device has the only valid copy of the data, then |
2295 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2296 | */ |
572e2857 | 2297 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2298 | vd->vdev_degraded = 1ULL; |
2299 | vd->vdev_faulted = 0ULL; | |
2300 | ||
2301 | /* | |
2302 | * If we reopen the device and it's not dead, only then do we | |
2303 | * mark it degraded. | |
2304 | */ | |
572e2857 | 2305 | vdev_reopen(tvd); |
34dc7c2f | 2306 | |
428870ff BB |
2307 | if (vdev_readable(vd)) |
2308 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2309 | } |
2310 | ||
b128c09f | 2311 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2312 | } |
2313 | ||
2314 | /* | |
2315 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2316 | * user that something is wrong. The vdev continues to operate as normal as far | |
2317 | * as I/O is concerned. | |
2318 | */ | |
2319 | int | |
428870ff | 2320 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2321 | { |
b128c09f | 2322 | vdev_t *vd; |
34dc7c2f | 2323 | |
428870ff | 2324 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2325 | |
b128c09f BB |
2326 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2327 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2328 | |
34dc7c2f | 2329 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2330 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2331 | |
2332 | /* | |
2333 | * If the vdev is already faulted, then don't do anything. | |
2334 | */ | |
b128c09f BB |
2335 | if (vd->vdev_faulted || vd->vdev_degraded) |
2336 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2337 | |
2338 | vd->vdev_degraded = 1ULL; | |
2339 | if (!vdev_is_dead(vd)) | |
2340 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2341 | aux); |
34dc7c2f | 2342 | |
b128c09f | 2343 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2344 | } |
2345 | ||
2346 | /* | |
d3cc8b15 WA |
2347 | * Online the given vdev. |
2348 | * | |
2349 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2350 | * spare device should be detached when the device finishes resilvering. | |
2351 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2352 | * events are generated if the device fails to open. | |
34dc7c2f BB |
2353 | */ |
2354 | int | |
b128c09f | 2355 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2356 | { |
9babb374 | 2357 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 2358 | |
428870ff | 2359 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2360 | |
b128c09f BB |
2361 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2362 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2363 | |
2364 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2365 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2366 | |
9babb374 | 2367 | tvd = vd->vdev_top; |
34dc7c2f BB |
2368 | vd->vdev_offline = B_FALSE; |
2369 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2370 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2371 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2372 | |
2373 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2374 | if (!vd->vdev_aux) { | |
2375 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2376 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2377 | } | |
2378 | ||
2379 | vdev_reopen(tvd); | |
34dc7c2f BB |
2380 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2381 | ||
9babb374 BB |
2382 | if (!vd->vdev_aux) { |
2383 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2384 | pvd->vdev_expanding = B_FALSE; | |
2385 | } | |
2386 | ||
34dc7c2f BB |
2387 | if (newstate) |
2388 | *newstate = vd->vdev_state; | |
2389 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2390 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2391 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2392 | vd->vdev_parent->vdev_child[0] == vd) | |
2393 | vd->vdev_unspare = B_TRUE; | |
2394 | ||
9babb374 BB |
2395 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2396 | ||
2397 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2398 | if (vd->vdev_aux) | |
2399 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2400 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2401 | } | |
fb5f0bc8 | 2402 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2403 | } |
2404 | ||
428870ff BB |
2405 | static int |
2406 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2407 | { |
9babb374 | 2408 | vdev_t *vd, *tvd; |
428870ff BB |
2409 | int error = 0; |
2410 | uint64_t generation; | |
2411 | metaslab_group_t *mg; | |
34dc7c2f | 2412 | |
428870ff BB |
2413 | top: |
2414 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2415 | |
b128c09f BB |
2416 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2417 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2418 | |
2419 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2420 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2421 | |
9babb374 | 2422 | tvd = vd->vdev_top; |
428870ff BB |
2423 | mg = tvd->vdev_mg; |
2424 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2425 | |
34dc7c2f BB |
2426 | /* |
2427 | * If the device isn't already offline, try to offline it. | |
2428 | */ | |
2429 | if (!vd->vdev_offline) { | |
2430 | /* | |
fb5f0bc8 | 2431 | * If this device has the only valid copy of some data, |
9babb374 BB |
2432 | * don't allow it to be offlined. Log devices are always |
2433 | * expendable. | |
34dc7c2f | 2434 | */ |
9babb374 BB |
2435 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2436 | vdev_dtl_required(vd)) | |
b128c09f | 2437 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2438 | |
428870ff BB |
2439 | /* |
2440 | * If the top-level is a slog and it has had allocations | |
2441 | * then proceed. We check that the vdev's metaslab group | |
2442 | * is not NULL since it's possible that we may have just | |
2443 | * added this vdev but not yet initialized its metaslabs. | |
2444 | */ | |
2445 | if (tvd->vdev_islog && mg != NULL) { | |
2446 | /* | |
2447 | * Prevent any future allocations. | |
2448 | */ | |
2449 | metaslab_group_passivate(mg); | |
2450 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2451 | ||
2452 | error = spa_offline_log(spa); | |
2453 | ||
2454 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2455 | ||
2456 | /* | |
2457 | * Check to see if the config has changed. | |
2458 | */ | |
2459 | if (error || generation != spa->spa_config_generation) { | |
2460 | metaslab_group_activate(mg); | |
2461 | if (error) | |
2462 | return (spa_vdev_state_exit(spa, | |
2463 | vd, error)); | |
2464 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2465 | goto top; | |
2466 | } | |
c99c9001 | 2467 | ASSERT0(tvd->vdev_stat.vs_alloc); |
428870ff BB |
2468 | } |
2469 | ||
34dc7c2f BB |
2470 | /* |
2471 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2472 | * If the top-level vdev is a log device then just offline |
2473 | * it. Otherwise, if this action results in the top-level | |
2474 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2475 | */ |
2476 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2477 | vdev_reopen(tvd); |
2478 | ||
2479 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2480 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2481 | vd->vdev_offline = B_FALSE; |
9babb374 | 2482 | vdev_reopen(tvd); |
b128c09f | 2483 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2484 | } |
428870ff BB |
2485 | |
2486 | /* | |
2487 | * Add the device back into the metaslab rotor so that | |
2488 | * once we online the device it's open for business. | |
2489 | */ | |
2490 | if (tvd->vdev_islog && mg != NULL) | |
2491 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2492 | } |
2493 | ||
b128c09f | 2494 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2495 | |
428870ff BB |
2496 | return (spa_vdev_state_exit(spa, vd, 0)); |
2497 | } | |
9babb374 | 2498 | |
428870ff BB |
2499 | int |
2500 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2501 | { | |
2502 | int error; | |
9babb374 | 2503 | |
428870ff BB |
2504 | mutex_enter(&spa->spa_vdev_top_lock); |
2505 | error = vdev_offline_locked(spa, guid, flags); | |
2506 | mutex_exit(&spa->spa_vdev_top_lock); | |
2507 | ||
2508 | return (error); | |
34dc7c2f BB |
2509 | } |
2510 | ||
2511 | /* | |
2512 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2513 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2514 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2515 | */ |
2516 | void | |
b128c09f | 2517 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2518 | { |
b128c09f | 2519 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2520 | int c; |
b128c09f BB |
2521 | |
2522 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2523 | |
2524 | if (vd == NULL) | |
b128c09f | 2525 | vd = rvd; |
34dc7c2f BB |
2526 | |
2527 | vd->vdev_stat.vs_read_errors = 0; | |
2528 | vd->vdev_stat.vs_write_errors = 0; | |
2529 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2530 | |
d6320ddb | 2531 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2532 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2533 | |
2534 | /* | |
b128c09f BB |
2535 | * If we're in the FAULTED state or have experienced failed I/O, then |
2536 | * clear the persistent state and attempt to reopen the device. We | |
2537 | * also mark the vdev config dirty, so that the new faulted state is | |
2538 | * written out to disk. | |
34dc7c2f | 2539 | */ |
b128c09f BB |
2540 | if (vd->vdev_faulted || vd->vdev_degraded || |
2541 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2542 | ||
428870ff BB |
2543 | /* |
2544 | * When reopening in reponse to a clear event, it may be due to | |
2545 | * a fmadm repair request. In this case, if the device is | |
2546 | * still broken, we want to still post the ereport again. | |
2547 | */ | |
2548 | vd->vdev_forcefault = B_TRUE; | |
2549 | ||
572e2857 | 2550 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2551 | vd->vdev_cant_read = B_FALSE; |
2552 | vd->vdev_cant_write = B_FALSE; | |
2553 | ||
572e2857 | 2554 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2555 | |
428870ff BB |
2556 | vd->vdev_forcefault = B_FALSE; |
2557 | ||
572e2857 | 2558 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2559 | vdev_state_dirty(vd->vdev_top); |
2560 | ||
2561 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2562 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2563 | ||
26685276 | 2564 | spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR); |
34dc7c2f | 2565 | } |
428870ff BB |
2566 | |
2567 | /* | |
2568 | * When clearing a FMA-diagnosed fault, we always want to | |
2569 | * unspare the device, as we assume that the original spare was | |
2570 | * done in response to the FMA fault. | |
2571 | */ | |
2572 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2573 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2574 | vd->vdev_parent->vdev_child[0] == vd) | |
2575 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2576 | } |
2577 | ||
b128c09f BB |
2578 | boolean_t |
2579 | vdev_is_dead(vdev_t *vd) | |
2580 | { | |
428870ff BB |
2581 | /* |
2582 | * Holes and missing devices are always considered "dead". | |
2583 | * This simplifies the code since we don't have to check for | |
2584 | * these types of devices in the various code paths. | |
2585 | * Instead we rely on the fact that we skip over dead devices | |
2586 | * before issuing I/O to them. | |
2587 | */ | |
2588 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2589 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2590 | } |
2591 | ||
2592 | boolean_t | |
34dc7c2f BB |
2593 | vdev_readable(vdev_t *vd) |
2594 | { | |
b128c09f | 2595 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2596 | } |
2597 | ||
b128c09f | 2598 | boolean_t |
34dc7c2f BB |
2599 | vdev_writeable(vdev_t *vd) |
2600 | { | |
b128c09f | 2601 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2602 | } |
2603 | ||
b128c09f BB |
2604 | boolean_t |
2605 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2606 | { |
fb5f0bc8 BB |
2607 | uint64_t state = vd->vdev_state; |
2608 | ||
b128c09f | 2609 | /* |
fb5f0bc8 | 2610 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2611 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2612 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2613 | * the proper locks. Note that we have to get the vdev state |
2614 | * in a local variable because although it changes atomically, | |
2615 | * we're asking two separate questions about it. | |
b128c09f | 2616 | */ |
fb5f0bc8 | 2617 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
428870ff | 2618 | !vd->vdev_cant_write && !vd->vdev_ishole); |
34dc7c2f BB |
2619 | } |
2620 | ||
b128c09f BB |
2621 | boolean_t |
2622 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2623 | { |
b128c09f | 2624 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2625 | |
b128c09f BB |
2626 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2627 | return (B_FALSE); | |
34dc7c2f | 2628 | |
b128c09f BB |
2629 | if (zio->io_type == ZIO_TYPE_READ) |
2630 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2631 | |
b128c09f BB |
2632 | if (zio->io_type == ZIO_TYPE_WRITE) |
2633 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2634 | |
b128c09f | 2635 | return (B_TRUE); |
34dc7c2f BB |
2636 | } |
2637 | ||
2638 | /* | |
2639 | * Get statistics for the given vdev. | |
2640 | */ | |
2641 | void | |
2642 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2643 | { | |
f3a7f661 GW |
2644 | spa_t *spa = vd->vdev_spa; |
2645 | vdev_t *rvd = spa->spa_root_vdev; | |
d6320ddb | 2646 | int c, t; |
34dc7c2f | 2647 | |
f3a7f661 GW |
2648 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
2649 | ||
34dc7c2f BB |
2650 | mutex_enter(&vd->vdev_stat_lock); |
2651 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2652 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2653 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2654 | vs->vs_rsize = vdev_get_min_asize(vd); |
2655 | if (vd->vdev_ops->vdev_op_leaf) | |
2656 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
1bd201e7 | 2657 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
f3a7f661 GW |
2658 | if (vd->vdev_aux == NULL && vd == vd->vdev_top) |
2659 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; | |
34dc7c2f BB |
2660 | |
2661 | /* | |
2662 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2663 | * over all top-level vdevs (i.e. the direct children of the root). | |
2664 | */ | |
2665 | if (vd == rvd) { | |
d6320ddb | 2666 | for (c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2667 | vdev_t *cvd = rvd->vdev_child[c]; |
2668 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2669 | ||
d6320ddb | 2670 | for (t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2671 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2672 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2673 | } | |
428870ff | 2674 | cvs->vs_scan_removing = cvd->vdev_removing; |
34dc7c2f BB |
2675 | } |
2676 | } | |
f3a7f661 | 2677 | mutex_exit(&vd->vdev_stat_lock); |
34dc7c2f BB |
2678 | } |
2679 | ||
2680 | void | |
2681 | vdev_clear_stats(vdev_t *vd) | |
2682 | { | |
2683 | mutex_enter(&vd->vdev_stat_lock); | |
2684 | vd->vdev_stat.vs_space = 0; | |
2685 | vd->vdev_stat.vs_dspace = 0; | |
2686 | vd->vdev_stat.vs_alloc = 0; | |
2687 | mutex_exit(&vd->vdev_stat_lock); | |
2688 | } | |
2689 | ||
428870ff BB |
2690 | void |
2691 | vdev_scan_stat_init(vdev_t *vd) | |
2692 | { | |
2693 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2694 | int c; |
428870ff | 2695 | |
d6320ddb | 2696 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2697 | vdev_scan_stat_init(vd->vdev_child[c]); |
2698 | ||
2699 | mutex_enter(&vd->vdev_stat_lock); | |
2700 | vs->vs_scan_processed = 0; | |
2701 | mutex_exit(&vd->vdev_stat_lock); | |
2702 | } | |
2703 | ||
34dc7c2f | 2704 | void |
b128c09f | 2705 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2706 | { |
fb5f0bc8 BB |
2707 | spa_t *spa = zio->io_spa; |
2708 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2709 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2710 | vdev_t *pvd; |
2711 | uint64_t txg = zio->io_txg; | |
2712 | vdev_stat_t *vs = &vd->vdev_stat; | |
2713 | zio_type_t type = zio->io_type; | |
2714 | int flags = zio->io_flags; | |
2715 | ||
b128c09f BB |
2716 | /* |
2717 | * If this i/o is a gang leader, it didn't do any actual work. | |
2718 | */ | |
2719 | if (zio->io_gang_tree) | |
2720 | return; | |
2721 | ||
34dc7c2f | 2722 | if (zio->io_error == 0) { |
b128c09f BB |
2723 | /* |
2724 | * If this is a root i/o, don't count it -- we've already | |
2725 | * counted the top-level vdevs, and vdev_get_stats() will | |
2726 | * aggregate them when asked. This reduces contention on | |
2727 | * the root vdev_stat_lock and implicitly handles blocks | |
2728 | * that compress away to holes, for which there is no i/o. | |
2729 | * (Holes never create vdev children, so all the counters | |
2730 | * remain zero, which is what we want.) | |
2731 | * | |
2732 | * Note: this only applies to successful i/o (io_error == 0) | |
2733 | * because unlike i/o counts, errors are not additive. | |
2734 | * When reading a ditto block, for example, failure of | |
2735 | * one top-level vdev does not imply a root-level error. | |
2736 | */ | |
2737 | if (vd == rvd) | |
2738 | return; | |
2739 | ||
2740 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2741 | |
2742 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2743 | return; | |
2744 | ||
2745 | mutex_enter(&vd->vdev_stat_lock); | |
2746 | ||
b128c09f | 2747 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 2748 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
2749 | dsl_scan_phys_t *scn_phys = |
2750 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2751 | uint64_t *processed = &scn_phys->scn_processed; | |
2752 | ||
2753 | /* XXX cleanup? */ | |
2754 | if (vd->vdev_ops->vdev_op_leaf) | |
2755 | atomic_add_64(processed, psize); | |
2756 | vs->vs_scan_processed += psize; | |
2757 | } | |
2758 | ||
fb5f0bc8 | 2759 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2760 | vs->vs_self_healed += psize; |
34dc7c2f | 2761 | } |
fb5f0bc8 BB |
2762 | |
2763 | vs->vs_ops[type]++; | |
2764 | vs->vs_bytes[type] += psize; | |
2765 | ||
2766 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2767 | return; |
2768 | } | |
2769 | ||
2770 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2771 | return; | |
2772 | ||
9babb374 BB |
2773 | /* |
2774 | * If this is an I/O error that is going to be retried, then ignore the | |
2775 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2776 | * hard errors, when in reality they can happen for any number of | |
2777 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2778 | */ | |
2779 | if (zio->io_error == EIO && | |
2780 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2781 | return; | |
2782 | ||
428870ff BB |
2783 | /* |
2784 | * Intent logs writes won't propagate their error to the root | |
2785 | * I/O so don't mark these types of failures as pool-level | |
2786 | * errors. | |
2787 | */ | |
2788 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2789 | return; | |
2790 | ||
b128c09f | 2791 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2792 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2793 | if (zio->io_error == ECKSUM) |
2794 | vs->vs_checksum_errors++; | |
2795 | else | |
2796 | vs->vs_read_errors++; | |
34dc7c2f | 2797 | } |
9babb374 | 2798 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2799 | vs->vs_write_errors++; |
2800 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2801 | |
fb5f0bc8 BB |
2802 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2803 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 2804 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 2805 | spa->spa_claiming)) { |
fb5f0bc8 | 2806 | /* |
428870ff BB |
2807 | * This is either a normal write (not a repair), or it's |
2808 | * a repair induced by the scrub thread, or it's a repair | |
2809 | * made by zil_claim() during spa_load() in the first txg. | |
2810 | * In the normal case, we commit the DTL change in the same | |
2811 | * txg as the block was born. In the scrub-induced repair | |
2812 | * case, we know that scrubs run in first-pass syncing context, | |
2813 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2814 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
2815 | * |
2816 | * We currently do not make DTL entries for failed spontaneous | |
2817 | * self-healing writes triggered by normal (non-scrubbing) | |
2818 | * reads, because we have no transactional context in which to | |
2819 | * do so -- and it's not clear that it'd be desirable anyway. | |
2820 | */ | |
2821 | if (vd->vdev_ops->vdev_op_leaf) { | |
2822 | uint64_t commit_txg = txg; | |
572e2857 | 2823 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
2824 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
2825 | ASSERT(spa_sync_pass(spa) == 1); | |
2826 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
2827 | commit_txg = spa_syncing_txg(spa); |
2828 | } else if (spa->spa_claiming) { | |
2829 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2830 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 2831 | } |
428870ff | 2832 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 2833 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 2834 | return; |
fb5f0bc8 BB |
2835 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2836 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2837 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2838 | } |
fb5f0bc8 BB |
2839 | if (vd != rvd) |
2840 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2841 | } |
2842 | } | |
2843 | ||
34dc7c2f | 2844 | /* |
428870ff BB |
2845 | * Update the in-core space usage stats for this vdev, its metaslab class, |
2846 | * and the root vdev. | |
34dc7c2f BB |
2847 | */ |
2848 | void | |
428870ff BB |
2849 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
2850 | int64_t space_delta) | |
34dc7c2f BB |
2851 | { |
2852 | int64_t dspace_delta = space_delta; | |
2853 | spa_t *spa = vd->vdev_spa; | |
2854 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
2855 | metaslab_group_t *mg = vd->vdev_mg; |
2856 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
2857 | |
2858 | ASSERT(vd == vd->vdev_top); | |
2859 | ||
2860 | /* | |
2861 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2862 | * factor. We must calculate this here and not at the root vdev | |
2863 | * because the root vdev's psize-to-asize is simply the max of its | |
2864 | * childrens', thus not accurate enough for us. | |
2865 | */ | |
2866 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2867 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2868 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2869 | vd->vdev_deflate_ratio; | |
2870 | ||
2871 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 2872 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2873 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2874 | vd->vdev_stat.vs_dspace += dspace_delta; |
2875 | mutex_exit(&vd->vdev_stat_lock); | |
2876 | ||
428870ff | 2877 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 2878 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 2879 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2880 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2881 | rvd->vdev_stat.vs_dspace += dspace_delta; |
2882 | mutex_exit(&rvd->vdev_stat_lock); | |
2883 | } | |
428870ff BB |
2884 | |
2885 | if (mc != NULL) { | |
2886 | ASSERT(rvd == vd->vdev_parent); | |
2887 | ASSERT(vd->vdev_ms_count != 0); | |
2888 | ||
2889 | metaslab_class_space_update(mc, | |
2890 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2891 | } | |
34dc7c2f BB |
2892 | } |
2893 | ||
2894 | /* | |
2895 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2896 | * so that it will be written out next time the vdev configuration is synced. | |
2897 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2898 | */ | |
2899 | void | |
2900 | vdev_config_dirty(vdev_t *vd) | |
2901 | { | |
2902 | spa_t *spa = vd->vdev_spa; | |
2903 | vdev_t *rvd = spa->spa_root_vdev; | |
2904 | int c; | |
2905 | ||
572e2857 BB |
2906 | ASSERT(spa_writeable(spa)); |
2907 | ||
34dc7c2f | 2908 | /* |
9babb374 BB |
2909 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2910 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2911 | */ |
2912 | if (vd->vdev_aux != NULL) { | |
2913 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2914 | nvlist_t **aux; | |
2915 | uint_t naux; | |
2916 | ||
2917 | for (c = 0; c < sav->sav_count; c++) { | |
2918 | if (sav->sav_vdevs[c] == vd) | |
2919 | break; | |
2920 | } | |
2921 | ||
2922 | if (c == sav->sav_count) { | |
2923 | /* | |
2924 | * We're being removed. There's nothing more to do. | |
2925 | */ | |
2926 | ASSERT(sav->sav_sync == B_TRUE); | |
2927 | return; | |
2928 | } | |
2929 | ||
2930 | sav->sav_sync = B_TRUE; | |
2931 | ||
9babb374 BB |
2932 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2933 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2934 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2935 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2936 | } | |
b128c09f BB |
2937 | |
2938 | ASSERT(c < naux); | |
2939 | ||
2940 | /* | |
2941 | * Setting the nvlist in the middle if the array is a little | |
2942 | * sketchy, but it will work. | |
2943 | */ | |
2944 | nvlist_free(aux[c]); | |
428870ff | 2945 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
2946 | |
2947 | return; | |
2948 | } | |
2949 | ||
2950 | /* | |
2951 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2952 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2953 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2954 | * so this is sufficient to ensure mutual exclusion. |
2955 | */ | |
b128c09f BB |
2956 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2957 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2958 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2959 | |
2960 | if (vd == rvd) { | |
2961 | for (c = 0; c < rvd->vdev_children; c++) | |
2962 | vdev_config_dirty(rvd->vdev_child[c]); | |
2963 | } else { | |
2964 | ASSERT(vd == vd->vdev_top); | |
2965 | ||
428870ff BB |
2966 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
2967 | !vd->vdev_ishole) | |
b128c09f | 2968 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
2969 | } |
2970 | } | |
2971 | ||
2972 | void | |
2973 | vdev_config_clean(vdev_t *vd) | |
2974 | { | |
2975 | spa_t *spa = vd->vdev_spa; | |
2976 | ||
b128c09f BB |
2977 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2978 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2979 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2980 | |
b128c09f BB |
2981 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2982 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2983 | } |
2984 | ||
b128c09f BB |
2985 | /* |
2986 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2987 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2988 | * the state changes from larger config changes because they require | |
2989 | * much less locking, and are often needed for administrative actions. | |
2990 | */ | |
2991 | void | |
2992 | vdev_state_dirty(vdev_t *vd) | |
2993 | { | |
2994 | spa_t *spa = vd->vdev_spa; | |
2995 | ||
572e2857 | 2996 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
2997 | ASSERT(vd == vd->vdev_top); |
2998 | ||
2999 | /* | |
3000 | * The state list is protected by the SCL_STATE lock. The caller | |
3001 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3002 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3003 | * so this is sufficient to ensure mutual exclusion. | |
3004 | */ | |
3005 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3006 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3007 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3008 | ||
428870ff | 3009 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
3010 | list_insert_head(&spa->spa_state_dirty_list, vd); |
3011 | } | |
3012 | ||
3013 | void | |
3014 | vdev_state_clean(vdev_t *vd) | |
3015 | { | |
3016 | spa_t *spa = vd->vdev_spa; | |
3017 | ||
3018 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3019 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3020 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3021 | ||
3022 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3023 | list_remove(&spa->spa_state_dirty_list, vd); | |
3024 | } | |
3025 | ||
3026 | /* | |
3027 | * Propagate vdev state up from children to parent. | |
3028 | */ | |
34dc7c2f BB |
3029 | void |
3030 | vdev_propagate_state(vdev_t *vd) | |
3031 | { | |
fb5f0bc8 BB |
3032 | spa_t *spa = vd->vdev_spa; |
3033 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
3034 | int degraded = 0, faulted = 0; |
3035 | int corrupted = 0; | |
34dc7c2f | 3036 | vdev_t *child; |
d6320ddb | 3037 | int c; |
34dc7c2f BB |
3038 | |
3039 | if (vd->vdev_children > 0) { | |
d6320ddb | 3040 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 3041 | child = vd->vdev_child[c]; |
b128c09f | 3042 | |
428870ff BB |
3043 | /* |
3044 | * Don't factor holes into the decision. | |
3045 | */ | |
3046 | if (child->vdev_ishole) | |
3047 | continue; | |
3048 | ||
b128c09f | 3049 | if (!vdev_readable(child) || |
fb5f0bc8 | 3050 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
3051 | /* |
3052 | * Root special: if there is a top-level log | |
3053 | * device, treat the root vdev as if it were | |
3054 | * degraded. | |
3055 | */ | |
3056 | if (child->vdev_islog && vd == rvd) | |
3057 | degraded++; | |
3058 | else | |
3059 | faulted++; | |
3060 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 3061 | degraded++; |
b128c09f | 3062 | } |
34dc7c2f BB |
3063 | |
3064 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3065 | corrupted++; | |
3066 | } | |
3067 | ||
3068 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3069 | ||
3070 | /* | |
b128c09f | 3071 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
3072 | * opened due to corrupted metadata, then propagate the root |
3073 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3074 | * replicas'. | |
3075 | */ | |
3076 | if (corrupted && vd == rvd && | |
3077 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3078 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3079 | VDEV_AUX_CORRUPT_DATA); | |
3080 | } | |
3081 | ||
b128c09f | 3082 | if (vd->vdev_parent) |
34dc7c2f BB |
3083 | vdev_propagate_state(vd->vdev_parent); |
3084 | } | |
3085 | ||
3086 | /* | |
3087 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3088 | * state, because we're in the process of opening children depth-first. | |
3089 | * Otherwise, we propagate the change to the parent. | |
3090 | * | |
3091 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3092 | * generated. | |
3093 | */ | |
3094 | void | |
3095 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3096 | { | |
3097 | uint64_t save_state; | |
b128c09f | 3098 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
3099 | |
3100 | if (state == vd->vdev_state) { | |
3101 | vd->vdev_stat.vs_aux = aux; | |
3102 | return; | |
3103 | } | |
3104 | ||
3105 | save_state = vd->vdev_state; | |
3106 | ||
3107 | vd->vdev_state = state; | |
3108 | vd->vdev_stat.vs_aux = aux; | |
3109 | ||
3110 | /* | |
3111 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
3112 | * always close the underlying device unless the device has requested |
3113 | * a delayed close (i.e. we're about to remove or fault the device). | |
3114 | * Otherwise, we keep accessible but invalid devices open forever. | |
3115 | * We don't call vdev_close() itself, because that implies some extra | |
3116 | * checks (offline, etc) that we don't want here. This is limited to | |
3117 | * leaf devices, because otherwise closing the device will affect other | |
3118 | * children. | |
34dc7c2f | 3119 | */ |
428870ff BB |
3120 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
3121 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
3122 | vd->vdev_ops->vdev_op_close(vd); |
3123 | ||
428870ff BB |
3124 | /* |
3125 | * If we have brought this vdev back into service, we need | |
3126 | * to notify fmd so that it can gracefully repair any outstanding | |
3127 | * cases due to a missing device. We do this in all cases, even those | |
3128 | * that probably don't correlate to a repaired fault. This is sure to | |
3129 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
3130 | * this is a transient state it's OK, as the retire agent will | |
3131 | * double-check the state of the vdev before repairing it. | |
3132 | */ | |
3133 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
3134 | vd->vdev_prevstate != state) | |
3135 | zfs_post_state_change(spa, vd); | |
3136 | ||
34dc7c2f BB |
3137 | if (vd->vdev_removed && |
3138 | state == VDEV_STATE_CANT_OPEN && | |
3139 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3140 | /* | |
3141 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3142 | * device was previously marked removed and someone attempted to | |
3143 | * reopen it. If this failed due to a nonexistent device, then | |
3144 | * keep the device in the REMOVED state. We also let this be if | |
3145 | * it is one of our special test online cases, which is only | |
3146 | * attempting to online the device and shouldn't generate an FMA | |
3147 | * fault. | |
3148 | */ | |
3149 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3150 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3151 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3152 | vd->vdev_removed = B_TRUE; |
3153 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3154 | /* | |
572e2857 BB |
3155 | * If we fail to open a vdev during an import or recovery, we |
3156 | * mark it as "not available", which signifies that it was | |
3157 | * never there to begin with. Failure to open such a device | |
3158 | * is not considered an error. | |
34dc7c2f | 3159 | */ |
572e2857 BB |
3160 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3161 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3162 | vd->vdev_ops->vdev_op_leaf) |
3163 | vd->vdev_not_present = 1; | |
3164 | ||
3165 | /* | |
3166 | * Post the appropriate ereport. If the 'prevstate' field is | |
3167 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3168 | * that this is part of a vdev_reopen(). In this case, we don't | |
3169 | * want to post the ereport if the device was already in the | |
3170 | * CANT_OPEN state beforehand. | |
3171 | * | |
3172 | * If the 'checkremove' flag is set, then this is an attempt to | |
3173 | * online the device in response to an insertion event. If we | |
3174 | * hit this case, then we have detected an insertion event for a | |
3175 | * faulted or offline device that wasn't in the removed state. | |
3176 | * In this scenario, we don't post an ereport because we are | |
3177 | * about to replace the device, or attempt an online with | |
3178 | * vdev_forcefault, which will generate the fault for us. | |
3179 | */ | |
3180 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3181 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3182 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3183 | const char *class; |
3184 | ||
3185 | switch (aux) { | |
3186 | case VDEV_AUX_OPEN_FAILED: | |
3187 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3188 | break; | |
3189 | case VDEV_AUX_CORRUPT_DATA: | |
3190 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3191 | break; | |
3192 | case VDEV_AUX_NO_REPLICAS: | |
3193 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3194 | break; | |
3195 | case VDEV_AUX_BAD_GUID_SUM: | |
3196 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3197 | break; | |
3198 | case VDEV_AUX_TOO_SMALL: | |
3199 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3200 | break; | |
3201 | case VDEV_AUX_BAD_LABEL: | |
3202 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3203 | break; | |
3204 | default: | |
3205 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3206 | } | |
3207 | ||
b128c09f | 3208 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3209 | } |
3210 | ||
3211 | /* Erase any notion of persistent removed state */ | |
3212 | vd->vdev_removed = B_FALSE; | |
3213 | } else { | |
3214 | vd->vdev_removed = B_FALSE; | |
3215 | } | |
3216 | ||
9babb374 BB |
3217 | if (!isopen && vd->vdev_parent) |
3218 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3219 | } |
b128c09f BB |
3220 | |
3221 | /* | |
3222 | * Check the vdev configuration to ensure that it's capable of supporting | |
57a4eddc | 3223 | * a root pool. |
b128c09f BB |
3224 | */ |
3225 | boolean_t | |
3226 | vdev_is_bootable(vdev_t *vd) | |
3227 | { | |
57a4eddc RL |
3228 | #if defined(__sun__) || defined(__sun) |
3229 | /* | |
3230 | * Currently, we do not support RAID-Z or partial configuration. | |
3231 | * In addition, only a single top-level vdev is allowed and none of the | |
3232 | * leaves can be wholedisks. | |
3233 | */ | |
d6320ddb BB |
3234 | int c; |
3235 | ||
b128c09f BB |
3236 | if (!vd->vdev_ops->vdev_op_leaf) { |
3237 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3238 | ||
3239 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3240 | vd->vdev_children > 1) { | |
3241 | return (B_FALSE); | |
3242 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3243 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3244 | return (B_FALSE); | |
3245 | } | |
3246 | } else if (vd->vdev_wholedisk == 1) { | |
3247 | return (B_FALSE); | |
3248 | } | |
3249 | ||
d6320ddb | 3250 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3251 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3252 | return (B_FALSE); | |
3253 | } | |
57a4eddc | 3254 | #endif /* __sun__ || __sun */ |
b128c09f BB |
3255 | return (B_TRUE); |
3256 | } | |
9babb374 | 3257 | |
428870ff BB |
3258 | /* |
3259 | * Load the state from the original vdev tree (ovd) which | |
3260 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3261 | * vdev was offline or faulted then we transfer that state to the |
3262 | * device in the current vdev tree (nvd). | |
428870ff | 3263 | */ |
9babb374 | 3264 | void |
428870ff | 3265 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3266 | { |
d6320ddb | 3267 | int c; |
9babb374 | 3268 | |
572e2857 | 3269 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3270 | ASSERT(spa_config_held(nvd->vdev_spa, |
3271 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3272 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3273 | |
d6320ddb | 3274 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3275 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3276 | |
572e2857 | 3277 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3278 | /* |
572e2857 | 3279 | * Restore the persistent vdev state |
9babb374 | 3280 | */ |
428870ff | 3281 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3282 | nvd->vdev_faulted = ovd->vdev_faulted; |
3283 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3284 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3285 | } |
3286 | } | |
3287 | ||
572e2857 BB |
3288 | /* |
3289 | * Determine if a log device has valid content. If the vdev was | |
3290 | * removed or faulted in the MOS config then we know that | |
3291 | * the content on the log device has already been written to the pool. | |
3292 | */ | |
3293 | boolean_t | |
3294 | vdev_log_state_valid(vdev_t *vd) | |
3295 | { | |
d6320ddb BB |
3296 | int c; |
3297 | ||
572e2857 BB |
3298 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3299 | !vd->vdev_removed) | |
3300 | return (B_TRUE); | |
3301 | ||
d6320ddb | 3302 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3303 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3304 | return (B_TRUE); | |
3305 | ||
3306 | return (B_FALSE); | |
3307 | } | |
3308 | ||
9babb374 BB |
3309 | /* |
3310 | * Expand a vdev if possible. | |
3311 | */ | |
3312 | void | |
3313 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3314 | { | |
3315 | ASSERT(vd->vdev_top == vd); | |
3316 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3317 | ||
3318 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3319 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3320 | vdev_config_dirty(vd); | |
3321 | } | |
3322 | } | |
428870ff BB |
3323 | |
3324 | /* | |
3325 | * Split a vdev. | |
3326 | */ | |
3327 | void | |
3328 | vdev_split(vdev_t *vd) | |
3329 | { | |
3330 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3331 | ||
3332 | vdev_remove_child(pvd, vd); | |
3333 | vdev_compact_children(pvd); | |
3334 | ||
3335 | cvd = pvd->vdev_child[0]; | |
3336 | if (pvd->vdev_children == 1) { | |
3337 | vdev_remove_parent(cvd); | |
3338 | cvd->vdev_splitting = B_TRUE; | |
3339 | } | |
3340 | vdev_propagate_state(cvd); | |
3341 | } | |
c28b2279 | 3342 | |
cc92e9d0 GW |
3343 | void |
3344 | vdev_deadman(vdev_t *vd) | |
3345 | { | |
3346 | int c; | |
3347 | ||
3348 | for (c = 0; c < vd->vdev_children; c++) { | |
3349 | vdev_t *cvd = vd->vdev_child[c]; | |
3350 | ||
3351 | vdev_deadman(cvd); | |
3352 | } | |
3353 | ||
3354 | if (vd->vdev_ops->vdev_op_leaf) { | |
3355 | vdev_queue_t *vq = &vd->vdev_queue; | |
3356 | ||
3357 | mutex_enter(&vq->vq_lock); | |
e8b96c60 | 3358 | if (avl_numnodes(&vq->vq_active_tree) > 0) { |
cc92e9d0 GW |
3359 | spa_t *spa = vd->vdev_spa; |
3360 | zio_t *fio; | |
3361 | uint64_t delta; | |
3362 | ||
3363 | /* | |
3364 | * Look at the head of all the pending queues, | |
3365 | * if any I/O has been outstanding for longer than | |
3366 | * the spa_deadman_synctime we log a zevent. | |
3367 | */ | |
e8b96c60 | 3368 | fio = avl_first(&vq->vq_active_tree); |
cb682a17 MA |
3369 | delta = gethrtime() - fio->io_timestamp; |
3370 | if (delta > spa_deadman_synctime(spa)) { | |
3371 | zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " | |
3372 | "delta %lluns, last io %lluns", | |
cc92e9d0 GW |
3373 | fio->io_timestamp, delta, |
3374 | vq->vq_io_complete_ts); | |
3375 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, | |
3376 | spa, vd, fio, 0, 0); | |
3377 | } | |
3378 | } | |
3379 | mutex_exit(&vq->vq_lock); | |
3380 | } | |
3381 | } | |
3382 | ||
c28b2279 BB |
3383 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3384 | EXPORT_SYMBOL(vdev_fault); | |
3385 | EXPORT_SYMBOL(vdev_degrade); | |
3386 | EXPORT_SYMBOL(vdev_online); | |
3387 | EXPORT_SYMBOL(vdev_offline); | |
3388 | EXPORT_SYMBOL(vdev_clear); | |
3389 | #endif |