]>
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. |
34dc7c2f BB |
24 | */ |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/fm/fs/zfs.h> | |
28 | #include <sys/spa.h> | |
29 | #include <sys/spa_impl.h> | |
30 | #include <sys/dmu.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/uberblock_impl.h> | |
34 | #include <sys/metaslab.h> | |
35 | #include <sys/metaslab_impl.h> | |
36 | #include <sys/space_map.h> | |
37 | #include <sys/zio.h> | |
38 | #include <sys/zap.h> | |
39 | #include <sys/fs/zfs.h> | |
b128c09f | 40 | #include <sys/arc.h> |
9babb374 | 41 | #include <sys/zil.h> |
428870ff | 42 | #include <sys/dsl_scan.h> |
34dc7c2f BB |
43 | |
44 | /* | |
45 | * Virtual device management. | |
46 | */ | |
47 | ||
48 | static vdev_ops_t *vdev_ops_table[] = { | |
49 | &vdev_root_ops, | |
50 | &vdev_raidz_ops, | |
51 | &vdev_mirror_ops, | |
52 | &vdev_replacing_ops, | |
53 | &vdev_spare_ops, | |
54 | &vdev_disk_ops, | |
55 | &vdev_file_ops, | |
56 | &vdev_missing_ops, | |
428870ff | 57 | &vdev_hole_ops, |
34dc7c2f BB |
58 | NULL |
59 | }; | |
60 | ||
b128c09f BB |
61 | /* maximum scrub/resilver I/O queue per leaf vdev */ |
62 | int zfs_scrub_limit = 10; | |
34dc7c2f BB |
63 | |
64 | /* | |
65 | * Given a vdev type, return the appropriate ops vector. | |
66 | */ | |
67 | static vdev_ops_t * | |
68 | vdev_getops(const char *type) | |
69 | { | |
70 | vdev_ops_t *ops, **opspp; | |
71 | ||
72 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
73 | if (strcmp(ops->vdev_op_type, type) == 0) | |
74 | break; | |
75 | ||
76 | return (ops); | |
77 | } | |
78 | ||
79 | /* | |
80 | * Default asize function: return the MAX of psize with the asize of | |
81 | * all children. This is what's used by anything other than RAID-Z. | |
82 | */ | |
83 | uint64_t | |
84 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
85 | { | |
86 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
87 | uint64_t csize; | |
34dc7c2f | 88 | |
9babb374 | 89 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
90 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
91 | asize = MAX(asize, csize); | |
92 | } | |
93 | ||
94 | return (asize); | |
95 | } | |
96 | ||
97 | /* | |
9babb374 BB |
98 | * Get the minimum allocatable size. We define the allocatable size as |
99 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
100 | * replace or attach devices which don't have the same physical size but | |
101 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
102 | */ |
103 | uint64_t | |
9babb374 | 104 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 105 | { |
9babb374 | 106 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 107 | |
9babb374 BB |
108 | /* |
109 | * The our parent is NULL (inactive spare or cache) or is the root, | |
110 | * just return our own asize. | |
111 | */ | |
112 | if (pvd == NULL) | |
113 | return (vd->vdev_asize); | |
34dc7c2f BB |
114 | |
115 | /* | |
9babb374 BB |
116 | * The top-level vdev just returns the allocatable size rounded |
117 | * to the nearest metaslab. | |
34dc7c2f | 118 | */ |
9babb374 BB |
119 | if (vd == vd->vdev_top) |
120 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 121 | |
9babb374 BB |
122 | /* |
123 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
124 | * so each child must provide at least 1/Nth of its asize. | |
125 | */ | |
126 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
127 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 128 | |
9babb374 BB |
129 | return (pvd->vdev_min_asize); |
130 | } | |
131 | ||
132 | void | |
133 | vdev_set_min_asize(vdev_t *vd) | |
134 | { | |
135 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 136 | |
9babb374 BB |
137 | for (int c = 0; c < vd->vdev_children; c++) |
138 | vdev_set_min_asize(vd->vdev_child[c]); | |
34dc7c2f BB |
139 | } |
140 | ||
141 | vdev_t * | |
142 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
143 | { | |
144 | vdev_t *rvd = spa->spa_root_vdev; | |
145 | ||
b128c09f | 146 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 147 | |
b128c09f BB |
148 | if (vdev < rvd->vdev_children) { |
149 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 150 | return (rvd->vdev_child[vdev]); |
b128c09f | 151 | } |
34dc7c2f BB |
152 | |
153 | return (NULL); | |
154 | } | |
155 | ||
156 | vdev_t * | |
157 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
158 | { | |
34dc7c2f BB |
159 | vdev_t *mvd; |
160 | ||
161 | if (vd->vdev_guid == guid) | |
162 | return (vd); | |
163 | ||
9babb374 | 164 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
165 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
166 | NULL) | |
167 | return (mvd); | |
168 | ||
169 | return (NULL); | |
170 | } | |
171 | ||
172 | void | |
173 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
174 | { | |
175 | size_t oldsize, newsize; | |
176 | uint64_t id = cvd->vdev_id; | |
177 | vdev_t **newchild; | |
178 | ||
b128c09f | 179 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
180 | ASSERT(cvd->vdev_parent == NULL); |
181 | ||
182 | cvd->vdev_parent = pvd; | |
183 | ||
184 | if (pvd == NULL) | |
185 | return; | |
186 | ||
187 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
188 | ||
189 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
190 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
191 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
192 | ||
193 | newchild = kmem_zalloc(newsize, KM_SLEEP); | |
194 | if (pvd->vdev_child != NULL) { | |
195 | bcopy(pvd->vdev_child, newchild, oldsize); | |
196 | kmem_free(pvd->vdev_child, oldsize); | |
197 | } | |
198 | ||
199 | pvd->vdev_child = newchild; | |
200 | pvd->vdev_child[id] = cvd; | |
201 | ||
202 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
203 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
204 | ||
205 | /* | |
206 | * Walk up all ancestors to update guid sum. | |
207 | */ | |
208 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
209 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
210 | ||
211 | if (cvd->vdev_ops->vdev_op_leaf) | |
212 | cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; | |
213 | } | |
214 | ||
215 | void | |
216 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
217 | { | |
218 | int c; | |
219 | uint_t id = cvd->vdev_id; | |
220 | ||
221 | ASSERT(cvd->vdev_parent == pvd); | |
222 | ||
223 | if (pvd == NULL) | |
224 | return; | |
225 | ||
226 | ASSERT(id < pvd->vdev_children); | |
227 | ASSERT(pvd->vdev_child[id] == cvd); | |
228 | ||
229 | pvd->vdev_child[id] = NULL; | |
230 | cvd->vdev_parent = NULL; | |
231 | ||
232 | for (c = 0; c < pvd->vdev_children; c++) | |
233 | if (pvd->vdev_child[c]) | |
234 | break; | |
235 | ||
236 | if (c == pvd->vdev_children) { | |
237 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
238 | pvd->vdev_child = NULL; | |
239 | pvd->vdev_children = 0; | |
240 | } | |
241 | ||
242 | /* | |
243 | * Walk up all ancestors to update guid sum. | |
244 | */ | |
245 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
246 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
247 | ||
248 | if (cvd->vdev_ops->vdev_op_leaf) | |
249 | cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; | |
250 | } | |
251 | ||
252 | /* | |
253 | * Remove any holes in the child array. | |
254 | */ | |
255 | void | |
256 | vdev_compact_children(vdev_t *pvd) | |
257 | { | |
258 | vdev_t **newchild, *cvd; | |
259 | int oldc = pvd->vdev_children; | |
9babb374 | 260 | int newc; |
34dc7c2f | 261 | |
b128c09f | 262 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 263 | |
9babb374 | 264 | for (int c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
265 | if (pvd->vdev_child[c]) |
266 | newc++; | |
267 | ||
268 | newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); | |
269 | ||
9babb374 | 270 | for (int c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
271 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
272 | newchild[newc] = cvd; | |
273 | cvd->vdev_id = newc++; | |
274 | } | |
275 | } | |
276 | ||
277 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
278 | pvd->vdev_child = newchild; | |
279 | pvd->vdev_children = newc; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Allocate and minimally initialize a vdev_t. | |
284 | */ | |
428870ff | 285 | vdev_t * |
34dc7c2f BB |
286 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
287 | { | |
288 | vdev_t *vd; | |
289 | ||
290 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); | |
291 | ||
292 | if (spa->spa_root_vdev == NULL) { | |
293 | ASSERT(ops == &vdev_root_ops); | |
294 | spa->spa_root_vdev = vd; | |
295 | } | |
296 | ||
428870ff | 297 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
298 | if (spa->spa_root_vdev == vd) { |
299 | /* | |
300 | * The root vdev's guid will also be the pool guid, | |
301 | * which must be unique among all pools. | |
302 | */ | |
428870ff | 303 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
304 | } else { |
305 | /* | |
306 | * Any other vdev's guid must be unique within the pool. | |
307 | */ | |
428870ff | 308 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
309 | } |
310 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
311 | } | |
312 | ||
313 | vd->vdev_spa = spa; | |
314 | vd->vdev_id = id; | |
315 | vd->vdev_guid = guid; | |
316 | vd->vdev_guid_sum = guid; | |
317 | vd->vdev_ops = ops; | |
318 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 319 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f BB |
320 | |
321 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); | |
322 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 323 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
fb5f0bc8 BB |
324 | for (int t = 0; t < DTL_TYPES; t++) { |
325 | space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, | |
326 | &vd->vdev_dtl_lock); | |
327 | } | |
34dc7c2f BB |
328 | txg_list_create(&vd->vdev_ms_list, |
329 | offsetof(struct metaslab, ms_txg_node)); | |
330 | txg_list_create(&vd->vdev_dtl_list, | |
331 | offsetof(struct vdev, vdev_dtl_node)); | |
332 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
333 | vdev_queue_init(vd); | |
334 | vdev_cache_init(vd); | |
335 | ||
336 | return (vd); | |
337 | } | |
338 | ||
339 | /* | |
340 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
341 | * creating a new vdev or loading an existing one - the behavior is slightly | |
342 | * different for each case. | |
343 | */ | |
344 | int | |
345 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
346 | int alloctype) | |
347 | { | |
348 | vdev_ops_t *ops; | |
349 | char *type; | |
350 | uint64_t guid = 0, islog, nparity; | |
351 | vdev_t *vd; | |
352 | ||
b128c09f | 353 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
354 | |
355 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
356 | return (EINVAL); | |
357 | ||
358 | if ((ops = vdev_getops(type)) == NULL) | |
359 | return (EINVAL); | |
360 | ||
361 | /* | |
362 | * If this is a load, get the vdev guid from the nvlist. | |
363 | * Otherwise, vdev_alloc_common() will generate one for us. | |
364 | */ | |
365 | if (alloctype == VDEV_ALLOC_LOAD) { | |
366 | uint64_t label_id; | |
367 | ||
368 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
369 | label_id != id) | |
370 | return (EINVAL); | |
371 | ||
372 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
373 | return (EINVAL); | |
374 | } else if (alloctype == VDEV_ALLOC_SPARE) { | |
375 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
376 | return (EINVAL); | |
377 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { | |
378 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
379 | return (EINVAL); | |
9babb374 BB |
380 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
381 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
382 | return (EINVAL); | |
34dc7c2f BB |
383 | } |
384 | ||
385 | /* | |
386 | * The first allocated vdev must be of type 'root'. | |
387 | */ | |
388 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
389 | return (EINVAL); | |
390 | ||
391 | /* | |
392 | * Determine whether we're a log vdev. | |
393 | */ | |
394 | islog = 0; | |
395 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
396 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
397 | return (ENOTSUP); | |
398 | ||
428870ff BB |
399 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
400 | return (ENOTSUP); | |
401 | ||
34dc7c2f BB |
402 | /* |
403 | * Set the nparity property for RAID-Z vdevs. | |
404 | */ | |
405 | nparity = -1ULL; | |
406 | if (ops == &vdev_raidz_ops) { | |
407 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
408 | &nparity) == 0) { | |
428870ff | 409 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
34dc7c2f BB |
410 | return (EINVAL); |
411 | /* | |
45d1cae3 BB |
412 | * Previous versions could only support 1 or 2 parity |
413 | * device. | |
34dc7c2f | 414 | */ |
45d1cae3 BB |
415 | if (nparity > 1 && |
416 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
417 | return (ENOTSUP); | |
418 | if (nparity > 2 && | |
419 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
34dc7c2f BB |
420 | return (ENOTSUP); |
421 | } else { | |
422 | /* | |
423 | * We require the parity to be specified for SPAs that | |
424 | * support multiple parity levels. | |
425 | */ | |
45d1cae3 | 426 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
34dc7c2f BB |
427 | return (EINVAL); |
428 | /* | |
429 | * Otherwise, we default to 1 parity device for RAID-Z. | |
430 | */ | |
431 | nparity = 1; | |
432 | } | |
433 | } else { | |
434 | nparity = 0; | |
435 | } | |
436 | ASSERT(nparity != -1ULL); | |
437 | ||
438 | vd = vdev_alloc_common(spa, id, guid, ops); | |
439 | ||
440 | vd->vdev_islog = islog; | |
441 | vd->vdev_nparity = nparity; | |
442 | ||
443 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
444 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
445 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
446 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
447 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
448 | &vd->vdev_physpath) == 0) | |
449 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
450 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
451 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
452 | |
453 | /* | |
454 | * Set the whole_disk property. If it's not specified, leave the value | |
455 | * as -1. | |
456 | */ | |
457 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
458 | &vd->vdev_wholedisk) != 0) | |
459 | vd->vdev_wholedisk = -1ULL; | |
460 | ||
461 | /* | |
462 | * Look for the 'not present' flag. This will only be set if the device | |
463 | * was not present at the time of import. | |
464 | */ | |
9babb374 BB |
465 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
466 | &vd->vdev_not_present); | |
34dc7c2f BB |
467 | |
468 | /* | |
469 | * Get the alignment requirement. | |
470 | */ | |
471 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
472 | ||
428870ff BB |
473 | /* |
474 | * Retrieve the vdev creation time. | |
475 | */ | |
476 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
477 | &vd->vdev_crtxg); | |
478 | ||
34dc7c2f BB |
479 | /* |
480 | * If we're a top-level vdev, try to load the allocation parameters. | |
481 | */ | |
428870ff BB |
482 | if (parent && !parent->vdev_parent && |
483 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
484 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
485 | &vd->vdev_ms_array); | |
486 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
487 | &vd->vdev_ms_shift); | |
488 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
489 | &vd->vdev_asize); | |
428870ff BB |
490 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
491 | &vd->vdev_removing); | |
492 | } | |
493 | ||
494 | if (parent && !parent->vdev_parent) { | |
495 | ASSERT(alloctype == VDEV_ALLOC_LOAD || | |
496 | alloctype == VDEV_ALLOC_ADD || | |
497 | alloctype == VDEV_ALLOC_SPLIT || | |
498 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
499 | vd->vdev_mg = metaslab_group_create(islog ? | |
500 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
501 | } |
502 | ||
503 | /* | |
504 | * If we're a leaf vdev, try to load the DTL object and other state. | |
505 | */ | |
b128c09f | 506 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
507 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
508 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
509 | if (alloctype == VDEV_ALLOC_LOAD) { |
510 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
fb5f0bc8 | 511 | &vd->vdev_dtl_smo.smo_object); |
b128c09f BB |
512 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
513 | &vd->vdev_unspare); | |
514 | } | |
9babb374 BB |
515 | |
516 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
517 | uint64_t spare = 0; | |
518 | ||
519 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
520 | &spare) == 0 && spare) | |
521 | spa_spare_add(vd); | |
522 | } | |
523 | ||
34dc7c2f BB |
524 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
525 | &vd->vdev_offline); | |
b128c09f | 526 | |
34dc7c2f BB |
527 | /* |
528 | * When importing a pool, we want to ignore the persistent fault | |
529 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
530 | * valid in the current context. Local vdevs will |
531 | * remain in the faulted state. | |
34dc7c2f | 532 | */ |
428870ff | 533 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
534 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
535 | &vd->vdev_faulted); | |
536 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
537 | &vd->vdev_degraded); | |
538 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
539 | &vd->vdev_removed); | |
428870ff BB |
540 | |
541 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
542 | char *aux; | |
543 | ||
544 | vd->vdev_label_aux = | |
545 | VDEV_AUX_ERR_EXCEEDED; | |
546 | if (nvlist_lookup_string(nv, | |
547 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
548 | strcmp(aux, "external") == 0) | |
549 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
550 | } | |
34dc7c2f BB |
551 | } |
552 | } | |
553 | ||
554 | /* | |
555 | * Add ourselves to the parent's list of children. | |
556 | */ | |
557 | vdev_add_child(parent, vd); | |
558 | ||
559 | *vdp = vd; | |
560 | ||
561 | return (0); | |
562 | } | |
563 | ||
564 | void | |
565 | vdev_free(vdev_t *vd) | |
566 | { | |
34dc7c2f BB |
567 | spa_t *spa = vd->vdev_spa; |
568 | ||
569 | /* | |
570 | * vdev_free() implies closing the vdev first. This is simpler than | |
571 | * trying to ensure complicated semantics for all callers. | |
572 | */ | |
573 | vdev_close(vd); | |
574 | ||
b128c09f | 575 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 576 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
577 | |
578 | /* | |
579 | * Free all children. | |
580 | */ | |
9babb374 | 581 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
582 | vdev_free(vd->vdev_child[c]); |
583 | ||
584 | ASSERT(vd->vdev_child == NULL); | |
585 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
586 | ||
587 | /* | |
588 | * Discard allocation state. | |
589 | */ | |
428870ff | 590 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 591 | vdev_metaslab_fini(vd); |
428870ff BB |
592 | metaslab_group_destroy(vd->vdev_mg); |
593 | } | |
34dc7c2f BB |
594 | |
595 | ASSERT3U(vd->vdev_stat.vs_space, ==, 0); | |
596 | ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); | |
597 | ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); | |
598 | ||
599 | /* | |
600 | * Remove this vdev from its parent's child list. | |
601 | */ | |
602 | vdev_remove_child(vd->vdev_parent, vd); | |
603 | ||
604 | ASSERT(vd->vdev_parent == NULL); | |
605 | ||
606 | /* | |
607 | * Clean up vdev structure. | |
608 | */ | |
609 | vdev_queue_fini(vd); | |
610 | vdev_cache_fini(vd); | |
611 | ||
612 | if (vd->vdev_path) | |
613 | spa_strfree(vd->vdev_path); | |
614 | if (vd->vdev_devid) | |
615 | spa_strfree(vd->vdev_devid); | |
616 | if (vd->vdev_physpath) | |
617 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
618 | if (vd->vdev_fru) |
619 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
620 | |
621 | if (vd->vdev_isspare) | |
622 | spa_spare_remove(vd); | |
623 | if (vd->vdev_isl2cache) | |
624 | spa_l2cache_remove(vd); | |
625 | ||
626 | txg_list_destroy(&vd->vdev_ms_list); | |
627 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 628 | |
34dc7c2f | 629 | mutex_enter(&vd->vdev_dtl_lock); |
fb5f0bc8 BB |
630 | for (int t = 0; t < DTL_TYPES; t++) { |
631 | space_map_unload(&vd->vdev_dtl[t]); | |
632 | space_map_destroy(&vd->vdev_dtl[t]); | |
633 | } | |
34dc7c2f | 634 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 635 | |
34dc7c2f BB |
636 | mutex_destroy(&vd->vdev_dtl_lock); |
637 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 638 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
639 | |
640 | if (vd == spa->spa_root_vdev) | |
641 | spa->spa_root_vdev = NULL; | |
642 | ||
643 | kmem_free(vd, sizeof (vdev_t)); | |
644 | } | |
645 | ||
646 | /* | |
647 | * Transfer top-level vdev state from svd to tvd. | |
648 | */ | |
649 | static void | |
650 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
651 | { | |
652 | spa_t *spa = svd->vdev_spa; | |
653 | metaslab_t *msp; | |
654 | vdev_t *vd; | |
655 | int t; | |
656 | ||
657 | ASSERT(tvd == tvd->vdev_top); | |
658 | ||
659 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
660 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
661 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
662 | ||
663 | svd->vdev_ms_array = 0; | |
664 | svd->vdev_ms_shift = 0; | |
665 | svd->vdev_ms_count = 0; | |
666 | ||
667 | tvd->vdev_mg = svd->vdev_mg; | |
668 | tvd->vdev_ms = svd->vdev_ms; | |
669 | ||
670 | svd->vdev_mg = NULL; | |
671 | svd->vdev_ms = NULL; | |
672 | ||
673 | if (tvd->vdev_mg != NULL) | |
674 | tvd->vdev_mg->mg_vd = tvd; | |
675 | ||
676 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
677 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
678 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
679 | ||
680 | svd->vdev_stat.vs_alloc = 0; | |
681 | svd->vdev_stat.vs_space = 0; | |
682 | svd->vdev_stat.vs_dspace = 0; | |
683 | ||
684 | for (t = 0; t < TXG_SIZE; t++) { | |
685 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
686 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
687 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
688 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
689 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
690 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
691 | } | |
692 | ||
b128c09f | 693 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
694 | vdev_config_clean(svd); |
695 | vdev_config_dirty(tvd); | |
696 | } | |
697 | ||
b128c09f BB |
698 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
699 | vdev_state_clean(svd); | |
700 | vdev_state_dirty(tvd); | |
701 | } | |
702 | ||
34dc7c2f BB |
703 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
704 | svd->vdev_deflate_ratio = 0; | |
705 | ||
706 | tvd->vdev_islog = svd->vdev_islog; | |
707 | svd->vdev_islog = 0; | |
708 | } | |
709 | ||
710 | static void | |
711 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
712 | { | |
34dc7c2f BB |
713 | if (vd == NULL) |
714 | return; | |
715 | ||
716 | vd->vdev_top = tvd; | |
717 | ||
9babb374 | 718 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
719 | vdev_top_update(tvd, vd->vdev_child[c]); |
720 | } | |
721 | ||
722 | /* | |
723 | * Add a mirror/replacing vdev above an existing vdev. | |
724 | */ | |
725 | vdev_t * | |
726 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
727 | { | |
728 | spa_t *spa = cvd->vdev_spa; | |
729 | vdev_t *pvd = cvd->vdev_parent; | |
730 | vdev_t *mvd; | |
731 | ||
b128c09f | 732 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
733 | |
734 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
735 | ||
736 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 737 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
34dc7c2f BB |
738 | mvd->vdev_ashift = cvd->vdev_ashift; |
739 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 740 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
741 | |
742 | vdev_remove_child(pvd, cvd); | |
743 | vdev_add_child(pvd, mvd); | |
744 | cvd->vdev_id = mvd->vdev_children; | |
745 | vdev_add_child(mvd, cvd); | |
746 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
747 | ||
748 | if (mvd == mvd->vdev_top) | |
749 | vdev_top_transfer(cvd, mvd); | |
750 | ||
751 | return (mvd); | |
752 | } | |
753 | ||
754 | /* | |
755 | * Remove a 1-way mirror/replacing vdev from the tree. | |
756 | */ | |
757 | void | |
758 | vdev_remove_parent(vdev_t *cvd) | |
759 | { | |
760 | vdev_t *mvd = cvd->vdev_parent; | |
761 | vdev_t *pvd = mvd->vdev_parent; | |
762 | ||
b128c09f | 763 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
764 | |
765 | ASSERT(mvd->vdev_children == 1); | |
766 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
767 | mvd->vdev_ops == &vdev_replacing_ops || | |
768 | mvd->vdev_ops == &vdev_spare_ops); | |
769 | cvd->vdev_ashift = mvd->vdev_ashift; | |
770 | ||
771 | vdev_remove_child(mvd, cvd); | |
772 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 773 | |
34dc7c2f | 774 | /* |
b128c09f BB |
775 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
776 | * Otherwise, we could have detached an offline device, and when we | |
777 | * go to import the pool we'll think we have two top-level vdevs, | |
778 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 779 | */ |
fb5f0bc8 BB |
780 | if (mvd->vdev_top == mvd) { |
781 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 782 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
783 | cvd->vdev_guid += guid_delta; |
784 | cvd->vdev_guid_sum += guid_delta; | |
785 | } | |
b128c09f BB |
786 | cvd->vdev_id = mvd->vdev_id; |
787 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
788 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
789 | ||
790 | if (cvd == cvd->vdev_top) | |
791 | vdev_top_transfer(mvd, cvd); | |
792 | ||
793 | ASSERT(mvd->vdev_children == 0); | |
794 | vdev_free(mvd); | |
795 | } | |
796 | ||
797 | int | |
798 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
799 | { | |
800 | spa_t *spa = vd->vdev_spa; | |
801 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
802 | uint64_t m; |
803 | uint64_t oldc = vd->vdev_ms_count; | |
804 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
805 | metaslab_t **mspp; | |
806 | int error; | |
807 | ||
428870ff BB |
808 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
809 | ||
810 | /* | |
811 | * This vdev is not being allocated from yet or is a hole. | |
812 | */ | |
813 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
814 | return (0); |
815 | ||
428870ff BB |
816 | ASSERT(!vd->vdev_ishole); |
817 | ||
9babb374 BB |
818 | /* |
819 | * Compute the raidz-deflation ratio. Note, we hard-code | |
820 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
821 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
822 | * or we will inconsistently account for existing bp's. | |
823 | */ | |
824 | vd->vdev_deflate_ratio = (1 << 17) / | |
825 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
826 | ||
34dc7c2f BB |
827 | ASSERT(oldc <= newc); |
828 | ||
34dc7c2f BB |
829 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
830 | ||
831 | if (oldc != 0) { | |
832 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
833 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
834 | } | |
835 | ||
836 | vd->vdev_ms = mspp; | |
837 | vd->vdev_ms_count = newc; | |
838 | ||
839 | for (m = oldc; m < newc; m++) { | |
840 | space_map_obj_t smo = { 0, 0, 0 }; | |
841 | if (txg == 0) { | |
842 | uint64_t object = 0; | |
843 | error = dmu_read(mos, vd->vdev_ms_array, | |
9babb374 BB |
844 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
845 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
846 | if (error) |
847 | return (error); | |
848 | if (object != 0) { | |
849 | dmu_buf_t *db; | |
850 | error = dmu_bonus_hold(mos, object, FTAG, &db); | |
851 | if (error) | |
852 | return (error); | |
853 | ASSERT3U(db->db_size, >=, sizeof (smo)); | |
854 | bcopy(db->db_data, &smo, sizeof (smo)); | |
855 | ASSERT3U(smo.smo_object, ==, object); | |
856 | dmu_buf_rele(db, FTAG); | |
857 | } | |
858 | } | |
859 | vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, | |
860 | m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); | |
861 | } | |
862 | ||
428870ff BB |
863 | if (txg == 0) |
864 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
865 | ||
866 | /* | |
867 | * If the vdev is being removed we don't activate | |
868 | * the metaslabs since we want to ensure that no new | |
869 | * allocations are performed on this device. | |
870 | */ | |
871 | if (oldc == 0 && !vd->vdev_removing) | |
872 | metaslab_group_activate(vd->vdev_mg); | |
873 | ||
874 | if (txg == 0) | |
875 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
876 | ||
34dc7c2f BB |
877 | return (0); |
878 | } | |
879 | ||
880 | void | |
881 | vdev_metaslab_fini(vdev_t *vd) | |
882 | { | |
883 | uint64_t m; | |
884 | uint64_t count = vd->vdev_ms_count; | |
885 | ||
886 | if (vd->vdev_ms != NULL) { | |
428870ff | 887 | metaslab_group_passivate(vd->vdev_mg); |
34dc7c2f BB |
888 | for (m = 0; m < count; m++) |
889 | if (vd->vdev_ms[m] != NULL) | |
890 | metaslab_fini(vd->vdev_ms[m]); | |
891 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); | |
892 | vd->vdev_ms = NULL; | |
893 | } | |
894 | } | |
895 | ||
b128c09f BB |
896 | typedef struct vdev_probe_stats { |
897 | boolean_t vps_readable; | |
898 | boolean_t vps_writeable; | |
899 | int vps_flags; | |
b128c09f BB |
900 | } vdev_probe_stats_t; |
901 | ||
902 | static void | |
903 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 904 | { |
fb5f0bc8 | 905 | spa_t *spa = zio->io_spa; |
d164b209 | 906 | vdev_t *vd = zio->io_vd; |
b128c09f | 907 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
908 | |
909 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
910 | |
911 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
912 | if (zio->io_error == 0) |
913 | vps->vps_readable = 1; | |
fb5f0bc8 | 914 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 915 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
916 | zio->io_offset, zio->io_size, zio->io_data, |
917 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
918 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
919 | } else { | |
920 | zio_buf_free(zio->io_data, zio->io_size); | |
921 | } | |
922 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
923 | if (zio->io_error == 0) |
924 | vps->vps_writeable = 1; | |
925 | zio_buf_free(zio->io_data, zio->io_size); | |
926 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 927 | zio_t *pio; |
b128c09f BB |
928 | |
929 | vd->vdev_cant_read |= !vps->vps_readable; | |
930 | vd->vdev_cant_write |= !vps->vps_writeable; | |
931 | ||
932 | if (vdev_readable(vd) && | |
fb5f0bc8 | 933 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
934 | zio->io_error = 0; |
935 | } else { | |
936 | ASSERT(zio->io_error != 0); | |
937 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 938 | spa, vd, NULL, 0, 0); |
b128c09f BB |
939 | zio->io_error = ENXIO; |
940 | } | |
d164b209 BB |
941 | |
942 | mutex_enter(&vd->vdev_probe_lock); | |
943 | ASSERT(vd->vdev_probe_zio == zio); | |
944 | vd->vdev_probe_zio = NULL; | |
945 | mutex_exit(&vd->vdev_probe_lock); | |
946 | ||
947 | while ((pio = zio_walk_parents(zio)) != NULL) | |
948 | if (!vdev_accessible(vd, pio)) | |
949 | pio->io_error = ENXIO; | |
950 | ||
b128c09f BB |
951 | kmem_free(vps, sizeof (*vps)); |
952 | } | |
953 | } | |
34dc7c2f | 954 | |
b128c09f BB |
955 | /* |
956 | * Determine whether this device is accessible by reading and writing | |
957 | * to several known locations: the pad regions of each vdev label | |
958 | * but the first (which we leave alone in case it contains a VTOC). | |
959 | */ | |
960 | zio_t * | |
d164b209 | 961 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
962 | { |
963 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
964 | vdev_probe_stats_t *vps = NULL; |
965 | zio_t *pio; | |
966 | ||
967 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 968 | |
d164b209 BB |
969 | /* |
970 | * Don't probe the probe. | |
971 | */ | |
972 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
973 | return (NULL); | |
b128c09f | 974 | |
d164b209 BB |
975 | /* |
976 | * To prevent 'probe storms' when a device fails, we create | |
977 | * just one probe i/o at a time. All zios that want to probe | |
978 | * this vdev will become parents of the probe io. | |
979 | */ | |
980 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 981 | |
d164b209 BB |
982 | if ((pio = vd->vdev_probe_zio) == NULL) { |
983 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); | |
984 | ||
985 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
986 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 987 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
988 | |
989 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
990 | /* | |
991 | * vdev_cant_read and vdev_cant_write can only | |
992 | * transition from TRUE to FALSE when we have the | |
993 | * SCL_ZIO lock as writer; otherwise they can only | |
994 | * transition from FALSE to TRUE. This ensures that | |
995 | * any zio looking at these values can assume that | |
996 | * failures persist for the life of the I/O. That's | |
997 | * important because when a device has intermittent | |
998 | * connectivity problems, we want to ensure that | |
999 | * they're ascribed to the device (ENXIO) and not | |
1000 | * the zio (EIO). | |
1001 | * | |
1002 | * Since we hold SCL_ZIO as writer here, clear both | |
1003 | * values so the probe can reevaluate from first | |
1004 | * principles. | |
1005 | */ | |
1006 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1007 | vd->vdev_cant_read = B_FALSE; | |
1008 | vd->vdev_cant_write = B_FALSE; | |
1009 | } | |
1010 | ||
1011 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1012 | vdev_probe_done, vps, | |
1013 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1014 | ||
428870ff BB |
1015 | /* |
1016 | * We can't change the vdev state in this context, so we | |
1017 | * kick off an async task to do it on our behalf. | |
1018 | */ | |
d164b209 BB |
1019 | if (zio != NULL) { |
1020 | vd->vdev_probe_wanted = B_TRUE; | |
1021 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1022 | } | |
b128c09f BB |
1023 | } |
1024 | ||
d164b209 BB |
1025 | if (zio != NULL) |
1026 | zio_add_child(zio, pio); | |
b128c09f | 1027 | |
d164b209 | 1028 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1029 | |
d164b209 BB |
1030 | if (vps == NULL) { |
1031 | ASSERT(zio != NULL); | |
1032 | return (NULL); | |
1033 | } | |
b128c09f BB |
1034 | |
1035 | for (int l = 1; l < VDEV_LABELS; l++) { | |
d164b209 | 1036 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1037 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
1038 | offsetof(vdev_label_t, vl_pad2)), |
1039 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
1040 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1041 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1042 | } | |
1043 | ||
d164b209 BB |
1044 | if (zio == NULL) |
1045 | return (pio); | |
1046 | ||
1047 | zio_nowait(pio); | |
1048 | return (NULL); | |
34dc7c2f BB |
1049 | } |
1050 | ||
45d1cae3 BB |
1051 | static void |
1052 | vdev_open_child(void *arg) | |
1053 | { | |
1054 | vdev_t *vd = arg; | |
1055 | ||
1056 | vd->vdev_open_thread = curthread; | |
1057 | vd->vdev_open_error = vdev_open(vd); | |
1058 | vd->vdev_open_thread = NULL; | |
1059 | } | |
1060 | ||
428870ff BB |
1061 | boolean_t |
1062 | vdev_uses_zvols(vdev_t *vd) | |
1063 | { | |
1064 | if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, | |
1065 | strlen(ZVOL_DIR)) == 0) | |
1066 | return (B_TRUE); | |
1067 | for (int c = 0; c < vd->vdev_children; c++) | |
1068 | if (vdev_uses_zvols(vd->vdev_child[c])) | |
1069 | return (B_TRUE); | |
1070 | return (B_FALSE); | |
1071 | } | |
1072 | ||
45d1cae3 BB |
1073 | void |
1074 | vdev_open_children(vdev_t *vd) | |
1075 | { | |
1076 | taskq_t *tq; | |
1077 | int children = vd->vdev_children; | |
1078 | ||
428870ff BB |
1079 | /* |
1080 | * in order to handle pools on top of zvols, do the opens | |
1081 | * in a single thread so that the same thread holds the | |
1082 | * spa_namespace_lock | |
1083 | */ | |
1084 | if (vdev_uses_zvols(vd)) { | |
1085 | for (int c = 0; c < children; c++) | |
1086 | vd->vdev_child[c]->vdev_open_error = | |
1087 | vdev_open(vd->vdev_child[c]); | |
1088 | return; | |
1089 | } | |
45d1cae3 BB |
1090 | tq = taskq_create("vdev_open", children, minclsyspri, |
1091 | children, children, TASKQ_PREPOPULATE); | |
1092 | ||
1093 | for (int c = 0; c < children; c++) | |
1094 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], | |
1095 | TQ_SLEEP) != NULL); | |
1096 | ||
1097 | taskq_destroy(tq); | |
1098 | } | |
1099 | ||
34dc7c2f BB |
1100 | /* |
1101 | * Prepare a virtual device for access. | |
1102 | */ | |
1103 | int | |
1104 | vdev_open(vdev_t *vd) | |
1105 | { | |
fb5f0bc8 | 1106 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1107 | int error; |
34dc7c2f BB |
1108 | uint64_t osize = 0; |
1109 | uint64_t asize, psize; | |
1110 | uint64_t ashift = 0; | |
1111 | ||
45d1cae3 BB |
1112 | ASSERT(vd->vdev_open_thread == curthread || |
1113 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1114 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1115 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1116 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1117 | ||
34dc7c2f | 1118 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1119 | vd->vdev_cant_read = B_FALSE; |
1120 | vd->vdev_cant_write = B_FALSE; | |
1121 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1122 | |
428870ff BB |
1123 | /* |
1124 | * If this vdev is not removed, check its fault status. If it's | |
1125 | * faulted, bail out of the open. | |
1126 | */ | |
34dc7c2f BB |
1127 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1128 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1129 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1130 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1131 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1132 | vd->vdev_label_aux); |
34dc7c2f BB |
1133 | return (ENXIO); |
1134 | } else if (vd->vdev_offline) { | |
1135 | ASSERT(vd->vdev_children == 0); | |
1136 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
1137 | return (ENXIO); | |
1138 | } | |
1139 | ||
1140 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); | |
1141 | ||
428870ff BB |
1142 | /* |
1143 | * Reset the vdev_reopening flag so that we actually close | |
1144 | * the vdev on error. | |
1145 | */ | |
1146 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1147 | if (zio_injection_enabled && error == 0) |
9babb374 | 1148 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1149 | |
1150 | if (error) { | |
1151 | if (vd->vdev_removed && | |
1152 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1153 | vd->vdev_removed = B_FALSE; | |
1154 | ||
1155 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1156 | vd->vdev_stat.vs_aux); | |
1157 | return (error); | |
1158 | } | |
1159 | ||
1160 | vd->vdev_removed = B_FALSE; | |
1161 | ||
428870ff BB |
1162 | /* |
1163 | * Recheck the faulted flag now that we have confirmed that | |
1164 | * the vdev is accessible. If we're faulted, bail. | |
1165 | */ | |
1166 | if (vd->vdev_faulted) { | |
1167 | ASSERT(vd->vdev_children == 0); | |
1168 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1169 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1170 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1171 | vd->vdev_label_aux); | |
1172 | return (ENXIO); | |
1173 | } | |
1174 | ||
34dc7c2f BB |
1175 | if (vd->vdev_degraded) { |
1176 | ASSERT(vd->vdev_children == 0); | |
1177 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1178 | VDEV_AUX_ERR_EXCEEDED); | |
1179 | } else { | |
428870ff | 1180 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1181 | } |
1182 | ||
428870ff BB |
1183 | /* |
1184 | * For hole or missing vdevs we just return success. | |
1185 | */ | |
1186 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1187 | return (0); | |
1188 | ||
9babb374 | 1189 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1190 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1191 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1192 | VDEV_AUX_NONE); | |
1193 | break; | |
1194 | } | |
9babb374 | 1195 | } |
34dc7c2f BB |
1196 | |
1197 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1198 | ||
1199 | if (vd->vdev_children == 0) { | |
1200 | if (osize < SPA_MINDEVSIZE) { | |
1201 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1202 | VDEV_AUX_TOO_SMALL); | |
1203 | return (EOVERFLOW); | |
1204 | } | |
1205 | psize = osize; | |
1206 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1207 | } else { | |
1208 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1209 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1210 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1211 | VDEV_AUX_TOO_SMALL); | |
1212 | return (EOVERFLOW); | |
1213 | } | |
1214 | psize = 0; | |
1215 | asize = osize; | |
1216 | } | |
1217 | ||
1218 | vd->vdev_psize = psize; | |
1219 | ||
9babb374 BB |
1220 | /* |
1221 | * Make sure the allocatable size hasn't shrunk. | |
1222 | */ | |
1223 | if (asize < vd->vdev_min_asize) { | |
1224 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1225 | VDEV_AUX_BAD_LABEL); | |
1226 | return (EINVAL); | |
1227 | } | |
1228 | ||
34dc7c2f BB |
1229 | if (vd->vdev_asize == 0) { |
1230 | /* | |
1231 | * This is the first-ever open, so use the computed values. | |
1232 | * For testing purposes, a higher ashift can be requested. | |
1233 | */ | |
1234 | vd->vdev_asize = asize; | |
1235 | vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); | |
1236 | } else { | |
1237 | /* | |
1238 | * Make sure the alignment requirement hasn't increased. | |
1239 | */ | |
1240 | if (ashift > vd->vdev_top->vdev_ashift) { | |
1241 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1242 | VDEV_AUX_BAD_LABEL); | |
1243 | return (EINVAL); | |
1244 | } | |
9babb374 | 1245 | } |
34dc7c2f | 1246 | |
9babb374 BB |
1247 | /* |
1248 | * If all children are healthy and the asize has increased, | |
1249 | * then we've experienced dynamic LUN growth. If automatic | |
1250 | * expansion is enabled then use the additional space. | |
1251 | */ | |
1252 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1253 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1254 | vd->vdev_asize = asize; | |
34dc7c2f | 1255 | |
9babb374 | 1256 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1257 | |
1258 | /* | |
1259 | * Ensure we can issue some IO before declaring the | |
1260 | * vdev open for business. | |
1261 | */ | |
b128c09f BB |
1262 | if (vd->vdev_ops->vdev_op_leaf && |
1263 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1264 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1265 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1266 | return (error); |
1267 | } | |
1268 | ||
34dc7c2f | 1269 | /* |
b128c09f | 1270 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1271 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1272 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1273 | */ |
fb5f0bc8 BB |
1274 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1275 | vdev_resilver_needed(vd, NULL, NULL)) | |
1276 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1277 | |
1278 | return (0); | |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * Called once the vdevs are all opened, this routine validates the label | |
1283 | * contents. This needs to be done before vdev_load() so that we don't | |
1284 | * inadvertently do repair I/Os to the wrong device. | |
1285 | * | |
1286 | * This function will only return failure if one of the vdevs indicates that it | |
1287 | * has since been destroyed or exported. This is only possible if | |
1288 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1289 | * will be updated but the function will return 0. | |
1290 | */ | |
1291 | int | |
1292 | vdev_validate(vdev_t *vd) | |
1293 | { | |
1294 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1295 | nvlist_t *label; |
428870ff | 1296 | uint64_t guid = 0, top_guid; |
34dc7c2f BB |
1297 | uint64_t state; |
1298 | ||
9babb374 | 1299 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1300 | if (vdev_validate(vd->vdev_child[c]) != 0) |
1301 | return (EBADF); | |
1302 | ||
1303 | /* | |
1304 | * If the device has already failed, or was marked offline, don't do | |
1305 | * any further validation. Otherwise, label I/O will fail and we will | |
1306 | * overwrite the previous state. | |
1307 | */ | |
b128c09f | 1308 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1309 | uint64_t aux_guid = 0; |
1310 | nvlist_t *nvl; | |
34dc7c2f BB |
1311 | |
1312 | if ((label = vdev_label_read_config(vd)) == NULL) { | |
1313 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1314 | VDEV_AUX_BAD_LABEL); | |
1315 | return (0); | |
1316 | } | |
1317 | ||
428870ff BB |
1318 | /* |
1319 | * Determine if this vdev has been split off into another | |
1320 | * pool. If so, then refuse to open it. | |
1321 | */ | |
1322 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1323 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1324 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1325 | VDEV_AUX_SPLIT_POOL); | |
1326 | nvlist_free(label); | |
1327 | return (0); | |
1328 | } | |
1329 | ||
34dc7c2f BB |
1330 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, |
1331 | &guid) != 0 || guid != spa_guid(spa)) { | |
1332 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1333 | VDEV_AUX_CORRUPT_DATA); | |
1334 | nvlist_free(label); | |
1335 | return (0); | |
1336 | } | |
1337 | ||
428870ff BB |
1338 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1339 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1340 | &aux_guid) != 0) | |
1341 | aux_guid = 0; | |
1342 | ||
b128c09f BB |
1343 | /* |
1344 | * If this vdev just became a top-level vdev because its | |
1345 | * sibling was detached, it will have adopted the parent's | |
1346 | * vdev guid -- but the label may or may not be on disk yet. | |
1347 | * Fortunately, either version of the label will have the | |
1348 | * same top guid, so if we're a top-level vdev, we can | |
1349 | * safely compare to that instead. | |
428870ff BB |
1350 | * |
1351 | * If we split this vdev off instead, then we also check the | |
1352 | * original pool's guid. We don't want to consider the vdev | |
1353 | * corrupt if it is partway through a split operation. | |
b128c09f | 1354 | */ |
34dc7c2f | 1355 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1356 | &guid) != 0 || |
1357 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1358 | &top_guid) != 0 || | |
428870ff | 1359 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1360 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1361 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1362 | VDEV_AUX_CORRUPT_DATA); | |
1363 | nvlist_free(label); | |
1364 | return (0); | |
1365 | } | |
1366 | ||
1367 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1368 | &state) != 0) { | |
1369 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1370 | VDEV_AUX_CORRUPT_DATA); | |
1371 | nvlist_free(label); | |
1372 | return (0); | |
1373 | } | |
1374 | ||
1375 | nvlist_free(label); | |
1376 | ||
45d1cae3 BB |
1377 | /* |
1378 | * If spa->spa_load_verbatim is true, no need to check the | |
1379 | * state of the pool. | |
1380 | */ | |
1381 | if (!spa->spa_load_verbatim && | |
428870ff | 1382 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f BB |
1383 | state != POOL_STATE_ACTIVE) |
1384 | return (EBADF); | |
34dc7c2f | 1385 | |
b128c09f BB |
1386 | /* |
1387 | * If we were able to open and validate a vdev that was | |
1388 | * previously marked permanently unavailable, clear that state | |
1389 | * now. | |
1390 | */ | |
1391 | if (vd->vdev_not_present) | |
1392 | vd->vdev_not_present = 0; | |
1393 | } | |
34dc7c2f BB |
1394 | |
1395 | return (0); | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Close a virtual device. | |
1400 | */ | |
1401 | void | |
1402 | vdev_close(vdev_t *vd) | |
1403 | { | |
fb5f0bc8 | 1404 | spa_t *spa = vd->vdev_spa; |
428870ff | 1405 | vdev_t *pvd = vd->vdev_parent; |
fb5f0bc8 BB |
1406 | |
1407 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1408 | ||
428870ff BB |
1409 | /* |
1410 | * If our parent is reopening, then we are as well, unless we are | |
1411 | * going offline. | |
1412 | */ | |
1413 | if (pvd != NULL && pvd->vdev_reopening) | |
1414 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1415 | ||
34dc7c2f BB |
1416 | vd->vdev_ops->vdev_op_close(vd); |
1417 | ||
1418 | vdev_cache_purge(vd); | |
1419 | ||
1420 | /* | |
9babb374 | 1421 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1422 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1423 | * it's still faulted. | |
1424 | */ | |
1425 | vd->vdev_prevstate = vd->vdev_state; | |
1426 | ||
1427 | if (vd->vdev_offline) | |
1428 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1429 | else | |
1430 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1431 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1432 | } | |
1433 | ||
428870ff BB |
1434 | void |
1435 | vdev_hold(vdev_t *vd) | |
1436 | { | |
1437 | spa_t *spa = vd->vdev_spa; | |
1438 | ||
1439 | ASSERT(spa_is_root(spa)); | |
1440 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1441 | return; | |
1442 | ||
1443 | for (int c = 0; c < vd->vdev_children; c++) | |
1444 | vdev_hold(vd->vdev_child[c]); | |
1445 | ||
1446 | if (vd->vdev_ops->vdev_op_leaf) | |
1447 | vd->vdev_ops->vdev_op_hold(vd); | |
1448 | } | |
1449 | ||
1450 | void | |
1451 | vdev_rele(vdev_t *vd) | |
1452 | { | |
1453 | spa_t *spa = vd->vdev_spa; | |
1454 | ||
1455 | ASSERT(spa_is_root(spa)); | |
1456 | for (int c = 0; c < vd->vdev_children; c++) | |
1457 | vdev_rele(vd->vdev_child[c]); | |
1458 | ||
1459 | if (vd->vdev_ops->vdev_op_leaf) | |
1460 | vd->vdev_ops->vdev_op_rele(vd); | |
1461 | } | |
1462 | ||
1463 | /* | |
1464 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1465 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1466 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1467 | * If the leaf has never been opened then open it, as usual. | |
1468 | */ | |
34dc7c2f BB |
1469 | void |
1470 | vdev_reopen(vdev_t *vd) | |
1471 | { | |
1472 | spa_t *spa = vd->vdev_spa; | |
1473 | ||
b128c09f | 1474 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1475 | |
428870ff BB |
1476 | /* set the reopening flag unless we're taking the vdev offline */ |
1477 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1478 | vdev_close(vd); |
1479 | (void) vdev_open(vd); | |
1480 | ||
1481 | /* | |
1482 | * Call vdev_validate() here to make sure we have the same device. | |
1483 | * Otherwise, a device with an invalid label could be successfully | |
1484 | * opened in response to vdev_reopen(). | |
1485 | */ | |
b128c09f BB |
1486 | if (vd->vdev_aux) { |
1487 | (void) vdev_validate_aux(vd); | |
1488 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1489 | vd->vdev_aux == &spa->spa_l2cache && |
1490 | !l2arc_vdev_present(vd)) | |
1491 | l2arc_add_vdev(spa, vd); | |
b128c09f BB |
1492 | } else { |
1493 | (void) vdev_validate(vd); | |
1494 | } | |
34dc7c2f BB |
1495 | |
1496 | /* | |
1497 | * Reassess parent vdev's health. | |
1498 | */ | |
1499 | vdev_propagate_state(vd); | |
1500 | } | |
1501 | ||
1502 | int | |
1503 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1504 | { | |
1505 | int error; | |
1506 | ||
1507 | /* | |
1508 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1509 | * For a create, however, we want to fail the request if | |
1510 | * there are any components we can't open. | |
1511 | */ | |
1512 | error = vdev_open(vd); | |
1513 | ||
1514 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1515 | vdev_close(vd); | |
1516 | return (error ? error : ENXIO); | |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Recursively initialize all labels. | |
1521 | */ | |
1522 | if ((error = vdev_label_init(vd, txg, isreplacing ? | |
1523 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { | |
1524 | vdev_close(vd); | |
1525 | return (error); | |
1526 | } | |
1527 | ||
1528 | return (0); | |
1529 | } | |
1530 | ||
34dc7c2f | 1531 | void |
9babb374 | 1532 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1533 | { |
1534 | /* | |
1535 | * Aim for roughly 200 metaslabs per vdev. | |
1536 | */ | |
1537 | vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); | |
1538 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
1539 | } |
1540 | ||
1541 | void | |
1542 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1543 | { | |
1544 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1545 | ASSERT(!vd->vdev_ishole); |
34dc7c2f BB |
1546 | ASSERT(ISP2(flags)); |
1547 | ||
1548 | if (flags & VDD_METASLAB) | |
1549 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1550 | ||
1551 | if (flags & VDD_DTL) | |
1552 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1553 | ||
1554 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1555 | } | |
1556 | ||
fb5f0bc8 BB |
1557 | /* |
1558 | * DTLs. | |
1559 | * | |
1560 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1561 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1562 | * |
1563 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1564 | * | |
1565 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1566 | * | |
1567 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1568 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1569 | * txgs that was scrubbed. | |
1570 | * | |
1571 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1572 | * persistent errors or just some device being offline. | |
1573 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1574 | * maintained; it's only computed when needed, typically to | |
1575 | * determine whether a device can be detached. | |
1576 | * | |
1577 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1578 | * either has the data or it doesn't. | |
1579 | * | |
1580 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1581 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1582 | * if any child is less than fully replicated, then so is its parent. | |
1583 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1584 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1585 | * those are the txgs we don't have enough replication to read. For example, | |
1586 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1587 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1588 | * two child DTL_MISSING maps. | |
1589 | * | |
1590 | * It should be clear from the above that to compute the DTLs and outage maps | |
1591 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1592 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1593 | * a configuration change, we generate all other DTLs from first principles. | |
1594 | */ | |
34dc7c2f | 1595 | void |
fb5f0bc8 | 1596 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1597 | { |
fb5f0bc8 BB |
1598 | space_map_t *sm = &vd->vdev_dtl[t]; |
1599 | ||
1600 | ASSERT(t < DTL_TYPES); | |
1601 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
1602 | ||
34dc7c2f BB |
1603 | mutex_enter(sm->sm_lock); |
1604 | if (!space_map_contains(sm, txg, size)) | |
1605 | space_map_add(sm, txg, size); | |
1606 | mutex_exit(sm->sm_lock); | |
1607 | } | |
1608 | ||
fb5f0bc8 BB |
1609 | boolean_t |
1610 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1611 | { |
fb5f0bc8 BB |
1612 | space_map_t *sm = &vd->vdev_dtl[t]; |
1613 | boolean_t dirty = B_FALSE; | |
34dc7c2f | 1614 | |
fb5f0bc8 BB |
1615 | ASSERT(t < DTL_TYPES); |
1616 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f BB |
1617 | |
1618 | mutex_enter(sm->sm_lock); | |
fb5f0bc8 BB |
1619 | if (sm->sm_space != 0) |
1620 | dirty = space_map_contains(sm, txg, size); | |
34dc7c2f BB |
1621 | mutex_exit(sm->sm_lock); |
1622 | ||
1623 | return (dirty); | |
1624 | } | |
1625 | ||
fb5f0bc8 BB |
1626 | boolean_t |
1627 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1628 | { | |
1629 | space_map_t *sm = &vd->vdev_dtl[t]; | |
1630 | boolean_t empty; | |
1631 | ||
1632 | mutex_enter(sm->sm_lock); | |
1633 | empty = (sm->sm_space == 0); | |
1634 | mutex_exit(sm->sm_lock); | |
1635 | ||
1636 | return (empty); | |
1637 | } | |
1638 | ||
34dc7c2f BB |
1639 | /* |
1640 | * Reassess DTLs after a config change or scrub completion. | |
1641 | */ | |
1642 | void | |
1643 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1644 | { | |
1645 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1646 | avl_tree_t reftree; |
1647 | int minref; | |
34dc7c2f | 1648 | |
fb5f0bc8 | 1649 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1650 | |
fb5f0bc8 BB |
1651 | for (int c = 0; c < vd->vdev_children; c++) |
1652 | vdev_dtl_reassess(vd->vdev_child[c], txg, | |
1653 | scrub_txg, scrub_done); | |
1654 | ||
428870ff | 1655 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1656 | return; |
1657 | ||
1658 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1659 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1660 | ||
34dc7c2f | 1661 | mutex_enter(&vd->vdev_dtl_lock); |
b128c09f | 1662 | if (scrub_txg != 0 && |
428870ff BB |
1663 | (spa->spa_scrub_started || |
1664 | (scn && scn->scn_phys.scn_errors == 0))) { | |
b128c09f BB |
1665 | /* |
1666 | * We completed a scrub up to scrub_txg. If we | |
1667 | * did it without rebooting, then the scrub dtl | |
1668 | * will be valid, so excise the old region and | |
1669 | * fold in the scrub dtl. Otherwise, leave the | |
1670 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1671 | * |
1672 | * There's little trick here: to excise the beginning | |
1673 | * of the DTL_MISSING map, we put it into a reference | |
1674 | * tree and then add a segment with refcnt -1 that | |
1675 | * covers the range [0, scrub_txg). This means | |
1676 | * that each txg in that range has refcnt -1 or 0. | |
1677 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1678 | * entries in the range [0, scrub_txg) will have a | |
1679 | * positive refcnt -- either 1 or 2. We then convert | |
1680 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1681 | */ |
fb5f0bc8 BB |
1682 | space_map_ref_create(&reftree); |
1683 | space_map_ref_add_map(&reftree, | |
1684 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1685 | space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); | |
1686 | space_map_ref_add_map(&reftree, | |
1687 | &vd->vdev_dtl[DTL_SCRUB], 2); | |
1688 | space_map_ref_generate_map(&reftree, | |
1689 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1690 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1691 | } |
fb5f0bc8 BB |
1692 | space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1693 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1694 | space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1695 | if (scrub_done) |
fb5f0bc8 BB |
1696 | space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1697 | space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
1698 | if (!vdev_readable(vd)) | |
1699 | space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); | |
1700 | else | |
1701 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1702 | space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); | |
34dc7c2f | 1703 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1704 | |
34dc7c2f BB |
1705 | if (txg != 0) |
1706 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1707 | return; | |
1708 | } | |
1709 | ||
34dc7c2f | 1710 | mutex_enter(&vd->vdev_dtl_lock); |
fb5f0bc8 | 1711 | for (int t = 0; t < DTL_TYPES; t++) { |
428870ff BB |
1712 | /* account for child's outage in parent's missing map */ |
1713 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1714 | if (t == DTL_SCRUB) |
1715 | continue; /* leaf vdevs only */ | |
1716 | if (t == DTL_PARTIAL) | |
1717 | minref = 1; /* i.e. non-zero */ | |
1718 | else if (vd->vdev_nparity != 0) | |
1719 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1720 | else | |
1721 | minref = vd->vdev_children; /* any kind of mirror */ | |
1722 | space_map_ref_create(&reftree); | |
1723 | for (int c = 0; c < vd->vdev_children; c++) { | |
1724 | vdev_t *cvd = vd->vdev_child[c]; | |
1725 | mutex_enter(&cvd->vdev_dtl_lock); | |
428870ff | 1726 | space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1727 | mutex_exit(&cvd->vdev_dtl_lock); |
1728 | } | |
1729 | space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); | |
1730 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1731 | } |
fb5f0bc8 | 1732 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1733 | } |
1734 | ||
1735 | static int | |
1736 | vdev_dtl_load(vdev_t *vd) | |
1737 | { | |
1738 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1739 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
34dc7c2f BB |
1740 | objset_t *mos = spa->spa_meta_objset; |
1741 | dmu_buf_t *db; | |
1742 | int error; | |
1743 | ||
1744 | ASSERT(vd->vdev_children == 0); | |
1745 | ||
1746 | if (smo->smo_object == 0) | |
1747 | return (0); | |
1748 | ||
428870ff BB |
1749 | ASSERT(!vd->vdev_ishole); |
1750 | ||
34dc7c2f BB |
1751 | if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) |
1752 | return (error); | |
1753 | ||
1754 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1755 | bcopy(db->db_data, smo, sizeof (*smo)); | |
1756 | dmu_buf_rele(db, FTAG); | |
1757 | ||
1758 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1759 | error = space_map_load(&vd->vdev_dtl[DTL_MISSING], |
1760 | NULL, SM_ALLOC, smo, mos); | |
34dc7c2f BB |
1761 | mutex_exit(&vd->vdev_dtl_lock); |
1762 | ||
1763 | return (error); | |
1764 | } | |
1765 | ||
1766 | void | |
1767 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1768 | { | |
1769 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1770 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
1771 | space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; | |
34dc7c2f BB |
1772 | objset_t *mos = spa->spa_meta_objset; |
1773 | space_map_t smsync; | |
1774 | kmutex_t smlock; | |
1775 | dmu_buf_t *db; | |
1776 | dmu_tx_t *tx; | |
1777 | ||
428870ff BB |
1778 | ASSERT(!vd->vdev_ishole); |
1779 | ||
34dc7c2f BB |
1780 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1781 | ||
1782 | if (vd->vdev_detached) { | |
1783 | if (smo->smo_object != 0) { | |
1784 | int err = dmu_object_free(mos, smo->smo_object, tx); | |
1785 | ASSERT3U(err, ==, 0); | |
1786 | smo->smo_object = 0; | |
1787 | } | |
1788 | dmu_tx_commit(tx); | |
34dc7c2f BB |
1789 | return; |
1790 | } | |
1791 | ||
1792 | if (smo->smo_object == 0) { | |
1793 | ASSERT(smo->smo_objsize == 0); | |
1794 | ASSERT(smo->smo_alloc == 0); | |
1795 | smo->smo_object = dmu_object_alloc(mos, | |
1796 | DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, | |
1797 | DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); | |
1798 | ASSERT(smo->smo_object != 0); | |
1799 | vdev_config_dirty(vd->vdev_top); | |
1800 | } | |
1801 | ||
1802 | mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); | |
1803 | ||
1804 | space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, | |
1805 | &smlock); | |
1806 | ||
1807 | mutex_enter(&smlock); | |
1808 | ||
1809 | mutex_enter(&vd->vdev_dtl_lock); | |
1810 | space_map_walk(sm, space_map_add, &smsync); | |
1811 | mutex_exit(&vd->vdev_dtl_lock); | |
1812 | ||
1813 | space_map_truncate(smo, mos, tx); | |
1814 | space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); | |
1815 | ||
1816 | space_map_destroy(&smsync); | |
1817 | ||
1818 | mutex_exit(&smlock); | |
1819 | mutex_destroy(&smlock); | |
1820 | ||
1821 | VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); | |
1822 | dmu_buf_will_dirty(db, tx); | |
1823 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1824 | bcopy(smo, db->db_data, sizeof (*smo)); | |
1825 | dmu_buf_rele(db, FTAG); | |
1826 | ||
1827 | dmu_tx_commit(tx); | |
1828 | } | |
1829 | ||
fb5f0bc8 BB |
1830 | /* |
1831 | * Determine whether the specified vdev can be offlined/detached/removed | |
1832 | * without losing data. | |
1833 | */ | |
1834 | boolean_t | |
1835 | vdev_dtl_required(vdev_t *vd) | |
1836 | { | |
1837 | spa_t *spa = vd->vdev_spa; | |
1838 | vdev_t *tvd = vd->vdev_top; | |
1839 | uint8_t cant_read = vd->vdev_cant_read; | |
1840 | boolean_t required; | |
1841 | ||
1842 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1843 | ||
1844 | if (vd == spa->spa_root_vdev || vd == tvd) | |
1845 | return (B_TRUE); | |
1846 | ||
1847 | /* | |
1848 | * Temporarily mark the device as unreadable, and then determine | |
1849 | * whether this results in any DTL outages in the top-level vdev. | |
1850 | * If not, we can safely offline/detach/remove the device. | |
1851 | */ | |
1852 | vd->vdev_cant_read = B_TRUE; | |
1853 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1854 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
1855 | vd->vdev_cant_read = cant_read; | |
1856 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1857 | ||
1858 | return (required); | |
1859 | } | |
1860 | ||
b128c09f BB |
1861 | /* |
1862 | * Determine if resilver is needed, and if so the txg range. | |
1863 | */ | |
1864 | boolean_t | |
1865 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
1866 | { | |
1867 | boolean_t needed = B_FALSE; | |
1868 | uint64_t thismin = UINT64_MAX; | |
1869 | uint64_t thismax = 0; | |
1870 | ||
1871 | if (vd->vdev_children == 0) { | |
1872 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1873 | if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && |
1874 | vdev_writeable(vd)) { | |
b128c09f BB |
1875 | space_seg_t *ss; |
1876 | ||
fb5f0bc8 | 1877 | ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f | 1878 | thismin = ss->ss_start - 1; |
fb5f0bc8 | 1879 | ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f BB |
1880 | thismax = ss->ss_end; |
1881 | needed = B_TRUE; | |
1882 | } | |
1883 | mutex_exit(&vd->vdev_dtl_lock); | |
1884 | } else { | |
fb5f0bc8 | 1885 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
1886 | vdev_t *cvd = vd->vdev_child[c]; |
1887 | uint64_t cmin, cmax; | |
1888 | ||
1889 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
1890 | thismin = MIN(thismin, cmin); | |
1891 | thismax = MAX(thismax, cmax); | |
1892 | needed = B_TRUE; | |
1893 | } | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | if (needed && minp) { | |
1898 | *minp = thismin; | |
1899 | *maxp = thismax; | |
1900 | } | |
1901 | return (needed); | |
1902 | } | |
1903 | ||
34dc7c2f BB |
1904 | void |
1905 | vdev_load(vdev_t *vd) | |
1906 | { | |
34dc7c2f BB |
1907 | /* |
1908 | * Recursively load all children. | |
1909 | */ | |
fb5f0bc8 | 1910 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1911 | vdev_load(vd->vdev_child[c]); |
1912 | ||
1913 | /* | |
1914 | * If this is a top-level vdev, initialize its metaslabs. | |
1915 | */ | |
428870ff | 1916 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
1917 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
1918 | vdev_metaslab_init(vd, 0) != 0)) | |
1919 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1920 | VDEV_AUX_CORRUPT_DATA); | |
1921 | ||
1922 | /* | |
1923 | * If this is a leaf vdev, load its DTL. | |
1924 | */ | |
1925 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
1926 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1927 | VDEV_AUX_CORRUPT_DATA); | |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * The special vdev case is used for hot spares and l2cache devices. Its | |
1932 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
1933 | * we make sure that we can open the underlying device, then try to read the | |
1934 | * label, and make sure that the label is sane and that it hasn't been | |
1935 | * repurposed to another pool. | |
1936 | */ | |
1937 | int | |
1938 | vdev_validate_aux(vdev_t *vd) | |
1939 | { | |
1940 | nvlist_t *label; | |
1941 | uint64_t guid, version; | |
1942 | uint64_t state; | |
1943 | ||
b128c09f BB |
1944 | if (!vdev_readable(vd)) |
1945 | return (0); | |
1946 | ||
34dc7c2f BB |
1947 | if ((label = vdev_label_read_config(vd)) == NULL) { |
1948 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1949 | VDEV_AUX_CORRUPT_DATA); | |
1950 | return (-1); | |
1951 | } | |
1952 | ||
1953 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
1954 | version > SPA_VERSION || | |
1955 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || | |
1956 | guid != vd->vdev_guid || | |
1957 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
1958 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1959 | VDEV_AUX_CORRUPT_DATA); | |
1960 | nvlist_free(label); | |
1961 | return (-1); | |
1962 | } | |
1963 | ||
1964 | /* | |
1965 | * We don't actually check the pool state here. If it's in fact in | |
1966 | * use by another pool, we update this fact on the fly when requested. | |
1967 | */ | |
1968 | nvlist_free(label); | |
1969 | return (0); | |
1970 | } | |
1971 | ||
428870ff BB |
1972 | void |
1973 | vdev_remove(vdev_t *vd, uint64_t txg) | |
1974 | { | |
1975 | spa_t *spa = vd->vdev_spa; | |
1976 | objset_t *mos = spa->spa_meta_objset; | |
1977 | dmu_tx_t *tx; | |
1978 | ||
1979 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
1980 | ||
1981 | if (vd->vdev_dtl_smo.smo_object) { | |
1982 | ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); | |
1983 | (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); | |
1984 | vd->vdev_dtl_smo.smo_object = 0; | |
1985 | } | |
1986 | ||
1987 | if (vd->vdev_ms != NULL) { | |
1988 | for (int m = 0; m < vd->vdev_ms_count; m++) { | |
1989 | metaslab_t *msp = vd->vdev_ms[m]; | |
1990 | ||
1991 | if (msp == NULL || msp->ms_smo.smo_object == 0) | |
1992 | continue; | |
1993 | ||
1994 | ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); | |
1995 | (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); | |
1996 | msp->ms_smo.smo_object = 0; | |
1997 | } | |
1998 | } | |
1999 | ||
2000 | if (vd->vdev_ms_array) { | |
2001 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2002 | vd->vdev_ms_array = 0; | |
2003 | vd->vdev_ms_shift = 0; | |
2004 | } | |
2005 | dmu_tx_commit(tx); | |
2006 | } | |
2007 | ||
34dc7c2f BB |
2008 | void |
2009 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2010 | { | |
2011 | metaslab_t *msp; | |
428870ff BB |
2012 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2013 | ||
2014 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2015 | |
34dc7c2f BB |
2016 | while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) |
2017 | metaslab_sync_done(msp, txg); | |
428870ff BB |
2018 | |
2019 | if (reassess) | |
2020 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2021 | } |
2022 | ||
2023 | void | |
2024 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2025 | { | |
2026 | spa_t *spa = vd->vdev_spa; | |
2027 | vdev_t *lvd; | |
2028 | metaslab_t *msp; | |
2029 | dmu_tx_t *tx; | |
2030 | ||
428870ff BB |
2031 | ASSERT(!vd->vdev_ishole); |
2032 | ||
34dc7c2f BB |
2033 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2034 | ASSERT(vd == vd->vdev_top); | |
2035 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2036 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2037 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2038 | ASSERT(vd->vdev_ms_array != 0); | |
2039 | vdev_config_dirty(vd); | |
2040 | dmu_tx_commit(tx); | |
2041 | } | |
2042 | ||
428870ff BB |
2043 | /* |
2044 | * Remove the metadata associated with this vdev once it's empty. | |
2045 | */ | |
2046 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2047 | vdev_remove(vd, txg); | |
2048 | ||
34dc7c2f BB |
2049 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2050 | metaslab_sync(msp, txg); | |
2051 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2052 | } | |
2053 | ||
2054 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2055 | vdev_dtl_sync(lvd, txg); | |
2056 | ||
2057 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2058 | } | |
2059 | ||
2060 | uint64_t | |
2061 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2062 | { | |
2063 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2064 | } | |
2065 | ||
34dc7c2f BB |
2066 | /* |
2067 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2068 | * not be opened, and no I/O is attempted. | |
2069 | */ | |
2070 | int | |
428870ff | 2071 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2072 | { |
b128c09f | 2073 | vdev_t *vd; |
34dc7c2f | 2074 | |
428870ff | 2075 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2076 | |
b128c09f BB |
2077 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2078 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2079 | |
34dc7c2f | 2080 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2081 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2082 | |
428870ff BB |
2083 | /* |
2084 | * We don't directly use the aux state here, but if we do a | |
2085 | * vdev_reopen(), we need this value to be present to remember why we | |
2086 | * were faulted. | |
2087 | */ | |
2088 | vd->vdev_label_aux = aux; | |
2089 | ||
34dc7c2f BB |
2090 | /* |
2091 | * Faulted state takes precedence over degraded. | |
2092 | */ | |
428870ff | 2093 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2094 | vd->vdev_faulted = 1ULL; |
2095 | vd->vdev_degraded = 0ULL; | |
428870ff | 2096 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2097 | |
2098 | /* | |
428870ff BB |
2099 | * If this device has the only valid copy of the data, then |
2100 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2101 | */ |
428870ff | 2102 | if (!vd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2103 | vd->vdev_degraded = 1ULL; |
2104 | vd->vdev_faulted = 0ULL; | |
2105 | ||
2106 | /* | |
2107 | * If we reopen the device and it's not dead, only then do we | |
2108 | * mark it degraded. | |
2109 | */ | |
2110 | vdev_reopen(vd); | |
2111 | ||
428870ff BB |
2112 | if (vdev_readable(vd)) |
2113 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2114 | } |
2115 | ||
b128c09f | 2116 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2117 | } |
2118 | ||
2119 | /* | |
2120 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2121 | * user that something is wrong. The vdev continues to operate as normal as far | |
2122 | * as I/O is concerned. | |
2123 | */ | |
2124 | int | |
428870ff | 2125 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2126 | { |
b128c09f | 2127 | vdev_t *vd; |
34dc7c2f | 2128 | |
428870ff | 2129 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2130 | |
b128c09f BB |
2131 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2132 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2133 | |
34dc7c2f | 2134 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2135 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2136 | |
2137 | /* | |
2138 | * If the vdev is already faulted, then don't do anything. | |
2139 | */ | |
b128c09f BB |
2140 | if (vd->vdev_faulted || vd->vdev_degraded) |
2141 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2142 | |
2143 | vd->vdev_degraded = 1ULL; | |
2144 | if (!vdev_is_dead(vd)) | |
2145 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2146 | aux); |
34dc7c2f | 2147 | |
b128c09f | 2148 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2149 | } |
2150 | ||
2151 | /* | |
2152 | * Online the given vdev. If 'unspare' is set, it implies two things. First, | |
2153 | * any attached spare device should be detached when the device finishes | |
2154 | * resilvering. Second, the online should be treated like a 'test' online case, | |
2155 | * so no FMA events are generated if the device fails to open. | |
2156 | */ | |
2157 | int | |
b128c09f | 2158 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2159 | { |
9babb374 | 2160 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 2161 | |
428870ff | 2162 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2163 | |
b128c09f BB |
2164 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2165 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2166 | |
2167 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2168 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2169 | |
9babb374 | 2170 | tvd = vd->vdev_top; |
34dc7c2f BB |
2171 | vd->vdev_offline = B_FALSE; |
2172 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2173 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2174 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2175 | |
2176 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2177 | if (!vd->vdev_aux) { | |
2178 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2179 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2180 | } | |
2181 | ||
2182 | vdev_reopen(tvd); | |
34dc7c2f BB |
2183 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2184 | ||
9babb374 BB |
2185 | if (!vd->vdev_aux) { |
2186 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2187 | pvd->vdev_expanding = B_FALSE; | |
2188 | } | |
2189 | ||
34dc7c2f BB |
2190 | if (newstate) |
2191 | *newstate = vd->vdev_state; | |
2192 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2193 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2194 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2195 | vd->vdev_parent->vdev_child[0] == vd) | |
2196 | vd->vdev_unspare = B_TRUE; | |
2197 | ||
9babb374 BB |
2198 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2199 | ||
2200 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2201 | if (vd->vdev_aux) | |
2202 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2203 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2204 | } | |
fb5f0bc8 | 2205 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2206 | } |
2207 | ||
428870ff BB |
2208 | static int |
2209 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2210 | { |
9babb374 | 2211 | vdev_t *vd, *tvd; |
428870ff BB |
2212 | int error = 0; |
2213 | uint64_t generation; | |
2214 | metaslab_group_t *mg; | |
34dc7c2f | 2215 | |
428870ff BB |
2216 | top: |
2217 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2218 | |
b128c09f BB |
2219 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2220 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2221 | |
2222 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2223 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2224 | |
9babb374 | 2225 | tvd = vd->vdev_top; |
428870ff BB |
2226 | mg = tvd->vdev_mg; |
2227 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2228 | |
34dc7c2f BB |
2229 | /* |
2230 | * If the device isn't already offline, try to offline it. | |
2231 | */ | |
2232 | if (!vd->vdev_offline) { | |
2233 | /* | |
fb5f0bc8 | 2234 | * If this device has the only valid copy of some data, |
9babb374 BB |
2235 | * don't allow it to be offlined. Log devices are always |
2236 | * expendable. | |
34dc7c2f | 2237 | */ |
9babb374 BB |
2238 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2239 | vdev_dtl_required(vd)) | |
b128c09f | 2240 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2241 | |
428870ff BB |
2242 | /* |
2243 | * If the top-level is a slog and it has had allocations | |
2244 | * then proceed. We check that the vdev's metaslab group | |
2245 | * is not NULL since it's possible that we may have just | |
2246 | * added this vdev but not yet initialized its metaslabs. | |
2247 | */ | |
2248 | if (tvd->vdev_islog && mg != NULL) { | |
2249 | /* | |
2250 | * Prevent any future allocations. | |
2251 | */ | |
2252 | metaslab_group_passivate(mg); | |
2253 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2254 | ||
2255 | error = spa_offline_log(spa); | |
2256 | ||
2257 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2258 | ||
2259 | /* | |
2260 | * Check to see if the config has changed. | |
2261 | */ | |
2262 | if (error || generation != spa->spa_config_generation) { | |
2263 | metaslab_group_activate(mg); | |
2264 | if (error) | |
2265 | return (spa_vdev_state_exit(spa, | |
2266 | vd, error)); | |
2267 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2268 | goto top; | |
2269 | } | |
2270 | ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); | |
2271 | } | |
2272 | ||
34dc7c2f BB |
2273 | /* |
2274 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2275 | * If the top-level vdev is a log device then just offline |
2276 | * it. Otherwise, if this action results in the top-level | |
2277 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2278 | */ |
2279 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2280 | vdev_reopen(tvd); |
2281 | ||
2282 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2283 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2284 | vd->vdev_offline = B_FALSE; |
9babb374 | 2285 | vdev_reopen(tvd); |
b128c09f | 2286 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2287 | } |
428870ff BB |
2288 | |
2289 | /* | |
2290 | * Add the device back into the metaslab rotor so that | |
2291 | * once we online the device it's open for business. | |
2292 | */ | |
2293 | if (tvd->vdev_islog && mg != NULL) | |
2294 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2295 | } |
2296 | ||
b128c09f | 2297 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2298 | |
428870ff BB |
2299 | return (spa_vdev_state_exit(spa, vd, 0)); |
2300 | } | |
9babb374 | 2301 | |
428870ff BB |
2302 | int |
2303 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2304 | { | |
2305 | int error; | |
9babb374 | 2306 | |
428870ff BB |
2307 | mutex_enter(&spa->spa_vdev_top_lock); |
2308 | error = vdev_offline_locked(spa, guid, flags); | |
2309 | mutex_exit(&spa->spa_vdev_top_lock); | |
2310 | ||
2311 | return (error); | |
34dc7c2f BB |
2312 | } |
2313 | ||
2314 | /* | |
2315 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2316 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2317 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2318 | */ |
2319 | void | |
b128c09f | 2320 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2321 | { |
b128c09f BB |
2322 | vdev_t *rvd = spa->spa_root_vdev; |
2323 | ||
2324 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2325 | |
2326 | if (vd == NULL) | |
b128c09f | 2327 | vd = rvd; |
34dc7c2f BB |
2328 | |
2329 | vd->vdev_stat.vs_read_errors = 0; | |
2330 | vd->vdev_stat.vs_write_errors = 0; | |
2331 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2332 | |
b128c09f BB |
2333 | for (int c = 0; c < vd->vdev_children; c++) |
2334 | vdev_clear(spa, vd->vdev_child[c]); | |
34dc7c2f BB |
2335 | |
2336 | /* | |
b128c09f BB |
2337 | * If we're in the FAULTED state or have experienced failed I/O, then |
2338 | * clear the persistent state and attempt to reopen the device. We | |
2339 | * also mark the vdev config dirty, so that the new faulted state is | |
2340 | * written out to disk. | |
34dc7c2f | 2341 | */ |
b128c09f BB |
2342 | if (vd->vdev_faulted || vd->vdev_degraded || |
2343 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2344 | ||
428870ff BB |
2345 | /* |
2346 | * When reopening in reponse to a clear event, it may be due to | |
2347 | * a fmadm repair request. In this case, if the device is | |
2348 | * still broken, we want to still post the ereport again. | |
2349 | */ | |
2350 | vd->vdev_forcefault = B_TRUE; | |
2351 | ||
34dc7c2f | 2352 | vd->vdev_faulted = vd->vdev_degraded = 0; |
b128c09f BB |
2353 | vd->vdev_cant_read = B_FALSE; |
2354 | vd->vdev_cant_write = B_FALSE; | |
2355 | ||
34dc7c2f | 2356 | vdev_reopen(vd); |
34dc7c2f | 2357 | |
428870ff BB |
2358 | vd->vdev_forcefault = B_FALSE; |
2359 | ||
b128c09f BB |
2360 | if (vd != rvd) |
2361 | vdev_state_dirty(vd->vdev_top); | |
2362 | ||
2363 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2364 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2365 | ||
2366 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); | |
2367 | } | |
428870ff BB |
2368 | |
2369 | /* | |
2370 | * When clearing a FMA-diagnosed fault, we always want to | |
2371 | * unspare the device, as we assume that the original spare was | |
2372 | * done in response to the FMA fault. | |
2373 | */ | |
2374 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2375 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2376 | vd->vdev_parent->vdev_child[0] == vd) | |
2377 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2378 | } |
2379 | ||
b128c09f BB |
2380 | boolean_t |
2381 | vdev_is_dead(vdev_t *vd) | |
2382 | { | |
428870ff BB |
2383 | /* |
2384 | * Holes and missing devices are always considered "dead". | |
2385 | * This simplifies the code since we don't have to check for | |
2386 | * these types of devices in the various code paths. | |
2387 | * Instead we rely on the fact that we skip over dead devices | |
2388 | * before issuing I/O to them. | |
2389 | */ | |
2390 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2391 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2392 | } |
2393 | ||
2394 | boolean_t | |
34dc7c2f BB |
2395 | vdev_readable(vdev_t *vd) |
2396 | { | |
b128c09f | 2397 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2398 | } |
2399 | ||
b128c09f | 2400 | boolean_t |
34dc7c2f BB |
2401 | vdev_writeable(vdev_t *vd) |
2402 | { | |
b128c09f | 2403 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2404 | } |
2405 | ||
b128c09f BB |
2406 | boolean_t |
2407 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2408 | { |
fb5f0bc8 BB |
2409 | uint64_t state = vd->vdev_state; |
2410 | ||
b128c09f | 2411 | /* |
fb5f0bc8 | 2412 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2413 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2414 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2415 | * the proper locks. Note that we have to get the vdev state |
2416 | * in a local variable because although it changes atomically, | |
2417 | * we're asking two separate questions about it. | |
b128c09f | 2418 | */ |
fb5f0bc8 | 2419 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
428870ff | 2420 | !vd->vdev_cant_write && !vd->vdev_ishole); |
34dc7c2f BB |
2421 | } |
2422 | ||
b128c09f BB |
2423 | boolean_t |
2424 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2425 | { |
b128c09f | 2426 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2427 | |
b128c09f BB |
2428 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2429 | return (B_FALSE); | |
34dc7c2f | 2430 | |
b128c09f BB |
2431 | if (zio->io_type == ZIO_TYPE_READ) |
2432 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2433 | |
b128c09f BB |
2434 | if (zio->io_type == ZIO_TYPE_WRITE) |
2435 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2436 | |
b128c09f | 2437 | return (B_TRUE); |
34dc7c2f BB |
2438 | } |
2439 | ||
2440 | /* | |
2441 | * Get statistics for the given vdev. | |
2442 | */ | |
2443 | void | |
2444 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2445 | { | |
2446 | vdev_t *rvd = vd->vdev_spa->spa_root_vdev; | |
34dc7c2f BB |
2447 | |
2448 | mutex_enter(&vd->vdev_stat_lock); | |
2449 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2450 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2451 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2452 | vs->vs_rsize = vdev_get_min_asize(vd); |
2453 | if (vd->vdev_ops->vdev_op_leaf) | |
2454 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
34dc7c2f BB |
2455 | mutex_exit(&vd->vdev_stat_lock); |
2456 | ||
2457 | /* | |
2458 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2459 | * over all top-level vdevs (i.e. the direct children of the root). | |
2460 | */ | |
2461 | if (vd == rvd) { | |
b128c09f | 2462 | for (int c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2463 | vdev_t *cvd = rvd->vdev_child[c]; |
2464 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2465 | ||
2466 | mutex_enter(&vd->vdev_stat_lock); | |
b128c09f | 2467 | for (int t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2468 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2469 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2470 | } | |
428870ff | 2471 | cvs->vs_scan_removing = cvd->vdev_removing; |
34dc7c2f BB |
2472 | mutex_exit(&vd->vdev_stat_lock); |
2473 | } | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | void | |
2478 | vdev_clear_stats(vdev_t *vd) | |
2479 | { | |
2480 | mutex_enter(&vd->vdev_stat_lock); | |
2481 | vd->vdev_stat.vs_space = 0; | |
2482 | vd->vdev_stat.vs_dspace = 0; | |
2483 | vd->vdev_stat.vs_alloc = 0; | |
2484 | mutex_exit(&vd->vdev_stat_lock); | |
2485 | } | |
2486 | ||
428870ff BB |
2487 | void |
2488 | vdev_scan_stat_init(vdev_t *vd) | |
2489 | { | |
2490 | vdev_stat_t *vs = &vd->vdev_stat; | |
2491 | ||
2492 | for (int c = 0; c < vd->vdev_children; c++) | |
2493 | vdev_scan_stat_init(vd->vdev_child[c]); | |
2494 | ||
2495 | mutex_enter(&vd->vdev_stat_lock); | |
2496 | vs->vs_scan_processed = 0; | |
2497 | mutex_exit(&vd->vdev_stat_lock); | |
2498 | } | |
2499 | ||
34dc7c2f | 2500 | void |
b128c09f | 2501 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2502 | { |
fb5f0bc8 BB |
2503 | spa_t *spa = zio->io_spa; |
2504 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2505 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2506 | vdev_t *pvd; |
2507 | uint64_t txg = zio->io_txg; | |
2508 | vdev_stat_t *vs = &vd->vdev_stat; | |
2509 | zio_type_t type = zio->io_type; | |
2510 | int flags = zio->io_flags; | |
2511 | ||
b128c09f BB |
2512 | /* |
2513 | * If this i/o is a gang leader, it didn't do any actual work. | |
2514 | */ | |
2515 | if (zio->io_gang_tree) | |
2516 | return; | |
2517 | ||
34dc7c2f | 2518 | if (zio->io_error == 0) { |
b128c09f BB |
2519 | /* |
2520 | * If this is a root i/o, don't count it -- we've already | |
2521 | * counted the top-level vdevs, and vdev_get_stats() will | |
2522 | * aggregate them when asked. This reduces contention on | |
2523 | * the root vdev_stat_lock and implicitly handles blocks | |
2524 | * that compress away to holes, for which there is no i/o. | |
2525 | * (Holes never create vdev children, so all the counters | |
2526 | * remain zero, which is what we want.) | |
2527 | * | |
2528 | * Note: this only applies to successful i/o (io_error == 0) | |
2529 | * because unlike i/o counts, errors are not additive. | |
2530 | * When reading a ditto block, for example, failure of | |
2531 | * one top-level vdev does not imply a root-level error. | |
2532 | */ | |
2533 | if (vd == rvd) | |
2534 | return; | |
2535 | ||
2536 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2537 | |
2538 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2539 | return; | |
2540 | ||
2541 | mutex_enter(&vd->vdev_stat_lock); | |
2542 | ||
b128c09f | 2543 | if (flags & ZIO_FLAG_IO_REPAIR) { |
428870ff BB |
2544 | if (flags & ZIO_FLAG_SCRUB_THREAD) { |
2545 | dsl_scan_phys_t *scn_phys = | |
2546 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2547 | uint64_t *processed = &scn_phys->scn_processed; | |
2548 | ||
2549 | /* XXX cleanup? */ | |
2550 | if (vd->vdev_ops->vdev_op_leaf) | |
2551 | atomic_add_64(processed, psize); | |
2552 | vs->vs_scan_processed += psize; | |
2553 | } | |
2554 | ||
fb5f0bc8 | 2555 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2556 | vs->vs_self_healed += psize; |
34dc7c2f | 2557 | } |
fb5f0bc8 BB |
2558 | |
2559 | vs->vs_ops[type]++; | |
2560 | vs->vs_bytes[type] += psize; | |
2561 | ||
2562 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2563 | return; |
2564 | } | |
2565 | ||
2566 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2567 | return; | |
2568 | ||
9babb374 BB |
2569 | /* |
2570 | * If this is an I/O error that is going to be retried, then ignore the | |
2571 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2572 | * hard errors, when in reality they can happen for any number of | |
2573 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2574 | */ | |
2575 | if (zio->io_error == EIO && | |
2576 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2577 | return; | |
2578 | ||
428870ff BB |
2579 | /* |
2580 | * Intent logs writes won't propagate their error to the root | |
2581 | * I/O so don't mark these types of failures as pool-level | |
2582 | * errors. | |
2583 | */ | |
2584 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2585 | return; | |
2586 | ||
b128c09f | 2587 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2588 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2589 | if (zio->io_error == ECKSUM) |
2590 | vs->vs_checksum_errors++; | |
2591 | else | |
2592 | vs->vs_read_errors++; | |
34dc7c2f | 2593 | } |
9babb374 | 2594 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2595 | vs->vs_write_errors++; |
2596 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2597 | |
fb5f0bc8 BB |
2598 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2599 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
428870ff BB |
2600 | (flags & ZIO_FLAG_SCRUB_THREAD) || |
2601 | spa->spa_claiming)) { | |
fb5f0bc8 | 2602 | /* |
428870ff BB |
2603 | * This is either a normal write (not a repair), or it's |
2604 | * a repair induced by the scrub thread, or it's a repair | |
2605 | * made by zil_claim() during spa_load() in the first txg. | |
2606 | * In the normal case, we commit the DTL change in the same | |
2607 | * txg as the block was born. In the scrub-induced repair | |
2608 | * case, we know that scrubs run in first-pass syncing context, | |
2609 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2610 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
2611 | * |
2612 | * We currently do not make DTL entries for failed spontaneous | |
2613 | * self-healing writes triggered by normal (non-scrubbing) | |
2614 | * reads, because we have no transactional context in which to | |
2615 | * do so -- and it's not clear that it'd be desirable anyway. | |
2616 | */ | |
2617 | if (vd->vdev_ops->vdev_op_leaf) { | |
2618 | uint64_t commit_txg = txg; | |
2619 | if (flags & ZIO_FLAG_SCRUB_THREAD) { | |
2620 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2621 | ASSERT(spa_sync_pass(spa) == 1); | |
2622 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
2623 | commit_txg = spa_syncing_txg(spa); |
2624 | } else if (spa->spa_claiming) { | |
2625 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2626 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 2627 | } |
428870ff | 2628 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 2629 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 2630 | return; |
fb5f0bc8 BB |
2631 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2632 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2633 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2634 | } |
fb5f0bc8 BB |
2635 | if (vd != rvd) |
2636 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2637 | } |
2638 | } | |
2639 | ||
34dc7c2f | 2640 | /* |
428870ff BB |
2641 | * Update the in-core space usage stats for this vdev, its metaslab class, |
2642 | * and the root vdev. | |
34dc7c2f BB |
2643 | */ |
2644 | void | |
428870ff BB |
2645 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
2646 | int64_t space_delta) | |
34dc7c2f BB |
2647 | { |
2648 | int64_t dspace_delta = space_delta; | |
2649 | spa_t *spa = vd->vdev_spa; | |
2650 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
2651 | metaslab_group_t *mg = vd->vdev_mg; |
2652 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
2653 | |
2654 | ASSERT(vd == vd->vdev_top); | |
2655 | ||
2656 | /* | |
2657 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2658 | * factor. We must calculate this here and not at the root vdev | |
2659 | * because the root vdev's psize-to-asize is simply the max of its | |
2660 | * childrens', thus not accurate enough for us. | |
2661 | */ | |
2662 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2663 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2664 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2665 | vd->vdev_deflate_ratio; | |
2666 | ||
2667 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 2668 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2669 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2670 | vd->vdev_stat.vs_dspace += dspace_delta; |
2671 | mutex_exit(&vd->vdev_stat_lock); | |
2672 | ||
428870ff | 2673 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 2674 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 2675 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2676 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2677 | rvd->vdev_stat.vs_dspace += dspace_delta; |
2678 | mutex_exit(&rvd->vdev_stat_lock); | |
2679 | } | |
428870ff BB |
2680 | |
2681 | if (mc != NULL) { | |
2682 | ASSERT(rvd == vd->vdev_parent); | |
2683 | ASSERT(vd->vdev_ms_count != 0); | |
2684 | ||
2685 | metaslab_class_space_update(mc, | |
2686 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2687 | } | |
34dc7c2f BB |
2688 | } |
2689 | ||
2690 | /* | |
2691 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2692 | * so that it will be written out next time the vdev configuration is synced. | |
2693 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2694 | */ | |
2695 | void | |
2696 | vdev_config_dirty(vdev_t *vd) | |
2697 | { | |
2698 | spa_t *spa = vd->vdev_spa; | |
2699 | vdev_t *rvd = spa->spa_root_vdev; | |
2700 | int c; | |
2701 | ||
2702 | /* | |
9babb374 BB |
2703 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2704 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2705 | */ |
2706 | if (vd->vdev_aux != NULL) { | |
2707 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2708 | nvlist_t **aux; | |
2709 | uint_t naux; | |
2710 | ||
2711 | for (c = 0; c < sav->sav_count; c++) { | |
2712 | if (sav->sav_vdevs[c] == vd) | |
2713 | break; | |
2714 | } | |
2715 | ||
2716 | if (c == sav->sav_count) { | |
2717 | /* | |
2718 | * We're being removed. There's nothing more to do. | |
2719 | */ | |
2720 | ASSERT(sav->sav_sync == B_TRUE); | |
2721 | return; | |
2722 | } | |
2723 | ||
2724 | sav->sav_sync = B_TRUE; | |
2725 | ||
9babb374 BB |
2726 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2727 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2728 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2729 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2730 | } | |
b128c09f BB |
2731 | |
2732 | ASSERT(c < naux); | |
2733 | ||
2734 | /* | |
2735 | * Setting the nvlist in the middle if the array is a little | |
2736 | * sketchy, but it will work. | |
2737 | */ | |
2738 | nvlist_free(aux[c]); | |
428870ff | 2739 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
2740 | |
2741 | return; | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2746 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2747 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2748 | * so this is sufficient to ensure mutual exclusion. |
2749 | */ | |
b128c09f BB |
2750 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2751 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2752 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2753 | |
2754 | if (vd == rvd) { | |
2755 | for (c = 0; c < rvd->vdev_children; c++) | |
2756 | vdev_config_dirty(rvd->vdev_child[c]); | |
2757 | } else { | |
2758 | ASSERT(vd == vd->vdev_top); | |
2759 | ||
428870ff BB |
2760 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
2761 | !vd->vdev_ishole) | |
b128c09f | 2762 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
2763 | } |
2764 | } | |
2765 | ||
2766 | void | |
2767 | vdev_config_clean(vdev_t *vd) | |
2768 | { | |
2769 | spa_t *spa = vd->vdev_spa; | |
2770 | ||
b128c09f BB |
2771 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2772 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2773 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2774 | |
b128c09f BB |
2775 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2776 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2777 | } |
2778 | ||
b128c09f BB |
2779 | /* |
2780 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2781 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2782 | * the state changes from larger config changes because they require | |
2783 | * much less locking, and are often needed for administrative actions. | |
2784 | */ | |
2785 | void | |
2786 | vdev_state_dirty(vdev_t *vd) | |
2787 | { | |
2788 | spa_t *spa = vd->vdev_spa; | |
2789 | ||
2790 | ASSERT(vd == vd->vdev_top); | |
2791 | ||
2792 | /* | |
2793 | * The state list is protected by the SCL_STATE lock. The caller | |
2794 | * must either hold SCL_STATE as writer, or must be the sync thread | |
2795 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
2796 | * so this is sufficient to ensure mutual exclusion. | |
2797 | */ | |
2798 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2799 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2800 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2801 | ||
428870ff | 2802 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
2803 | list_insert_head(&spa->spa_state_dirty_list, vd); |
2804 | } | |
2805 | ||
2806 | void | |
2807 | vdev_state_clean(vdev_t *vd) | |
2808 | { | |
2809 | spa_t *spa = vd->vdev_spa; | |
2810 | ||
2811 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2812 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2813 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2814 | ||
2815 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
2816 | list_remove(&spa->spa_state_dirty_list, vd); | |
2817 | } | |
2818 | ||
2819 | /* | |
2820 | * Propagate vdev state up from children to parent. | |
2821 | */ | |
34dc7c2f BB |
2822 | void |
2823 | vdev_propagate_state(vdev_t *vd) | |
2824 | { | |
fb5f0bc8 BB |
2825 | spa_t *spa = vd->vdev_spa; |
2826 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
2827 | int degraded = 0, faulted = 0; |
2828 | int corrupted = 0; | |
34dc7c2f BB |
2829 | vdev_t *child; |
2830 | ||
2831 | if (vd->vdev_children > 0) { | |
9babb374 | 2832 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 2833 | child = vd->vdev_child[c]; |
b128c09f | 2834 | |
428870ff BB |
2835 | /* |
2836 | * Don't factor holes into the decision. | |
2837 | */ | |
2838 | if (child->vdev_ishole) | |
2839 | continue; | |
2840 | ||
b128c09f | 2841 | if (!vdev_readable(child) || |
fb5f0bc8 | 2842 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
2843 | /* |
2844 | * Root special: if there is a top-level log | |
2845 | * device, treat the root vdev as if it were | |
2846 | * degraded. | |
2847 | */ | |
2848 | if (child->vdev_islog && vd == rvd) | |
2849 | degraded++; | |
2850 | else | |
2851 | faulted++; | |
2852 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 2853 | degraded++; |
b128c09f | 2854 | } |
34dc7c2f BB |
2855 | |
2856 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
2857 | corrupted++; | |
2858 | } | |
2859 | ||
2860 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
2861 | ||
2862 | /* | |
b128c09f | 2863 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
2864 | * opened due to corrupted metadata, then propagate the root |
2865 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
2866 | * replicas'. | |
2867 | */ | |
2868 | if (corrupted && vd == rvd && | |
2869 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
2870 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2871 | VDEV_AUX_CORRUPT_DATA); | |
2872 | } | |
2873 | ||
b128c09f | 2874 | if (vd->vdev_parent) |
34dc7c2f BB |
2875 | vdev_propagate_state(vd->vdev_parent); |
2876 | } | |
2877 | ||
2878 | /* | |
2879 | * Set a vdev's state. If this is during an open, we don't update the parent | |
2880 | * state, because we're in the process of opening children depth-first. | |
2881 | * Otherwise, we propagate the change to the parent. | |
2882 | * | |
2883 | * If this routine places a device in a faulted state, an appropriate ereport is | |
2884 | * generated. | |
2885 | */ | |
2886 | void | |
2887 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
2888 | { | |
2889 | uint64_t save_state; | |
b128c09f | 2890 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
2891 | |
2892 | if (state == vd->vdev_state) { | |
2893 | vd->vdev_stat.vs_aux = aux; | |
2894 | return; | |
2895 | } | |
2896 | ||
2897 | save_state = vd->vdev_state; | |
2898 | ||
2899 | vd->vdev_state = state; | |
2900 | vd->vdev_stat.vs_aux = aux; | |
2901 | ||
2902 | /* | |
2903 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
2904 | * always close the underlying device unless the device has requested |
2905 | * a delayed close (i.e. we're about to remove or fault the device). | |
2906 | * Otherwise, we keep accessible but invalid devices open forever. | |
2907 | * We don't call vdev_close() itself, because that implies some extra | |
2908 | * checks (offline, etc) that we don't want here. This is limited to | |
2909 | * leaf devices, because otherwise closing the device will affect other | |
2910 | * children. | |
34dc7c2f | 2911 | */ |
428870ff BB |
2912 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
2913 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
2914 | vd->vdev_ops->vdev_op_close(vd); |
2915 | ||
428870ff BB |
2916 | /* |
2917 | * If we have brought this vdev back into service, we need | |
2918 | * to notify fmd so that it can gracefully repair any outstanding | |
2919 | * cases due to a missing device. We do this in all cases, even those | |
2920 | * that probably don't correlate to a repaired fault. This is sure to | |
2921 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
2922 | * this is a transient state it's OK, as the retire agent will | |
2923 | * double-check the state of the vdev before repairing it. | |
2924 | */ | |
2925 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
2926 | vd->vdev_prevstate != state) | |
2927 | zfs_post_state_change(spa, vd); | |
2928 | ||
34dc7c2f BB |
2929 | if (vd->vdev_removed && |
2930 | state == VDEV_STATE_CANT_OPEN && | |
2931 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
2932 | /* | |
2933 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
2934 | * device was previously marked removed and someone attempted to | |
2935 | * reopen it. If this failed due to a nonexistent device, then | |
2936 | * keep the device in the REMOVED state. We also let this be if | |
2937 | * it is one of our special test online cases, which is only | |
2938 | * attempting to online the device and shouldn't generate an FMA | |
2939 | * fault. | |
2940 | */ | |
2941 | vd->vdev_state = VDEV_STATE_REMOVED; | |
2942 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
2943 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
2944 | vd->vdev_removed = B_TRUE; |
2945 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
2946 | /* | |
2947 | * If we fail to open a vdev during an import, we mark it as | |
2948 | * "not available", which signifies that it was never there to | |
2949 | * begin with. Failure to open such a device is not considered | |
2950 | * an error. | |
2951 | */ | |
428870ff | 2952 | if (spa_load_state(spa) == SPA_LOAD_IMPORT && |
34dc7c2f BB |
2953 | vd->vdev_ops->vdev_op_leaf) |
2954 | vd->vdev_not_present = 1; | |
2955 | ||
2956 | /* | |
2957 | * Post the appropriate ereport. If the 'prevstate' field is | |
2958 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
2959 | * that this is part of a vdev_reopen(). In this case, we don't | |
2960 | * want to post the ereport if the device was already in the | |
2961 | * CANT_OPEN state beforehand. | |
2962 | * | |
2963 | * If the 'checkremove' flag is set, then this is an attempt to | |
2964 | * online the device in response to an insertion event. If we | |
2965 | * hit this case, then we have detected an insertion event for a | |
2966 | * faulted or offline device that wasn't in the removed state. | |
2967 | * In this scenario, we don't post an ereport because we are | |
2968 | * about to replace the device, or attempt an online with | |
2969 | * vdev_forcefault, which will generate the fault for us. | |
2970 | */ | |
2971 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
2972 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 2973 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
2974 | const char *class; |
2975 | ||
2976 | switch (aux) { | |
2977 | case VDEV_AUX_OPEN_FAILED: | |
2978 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
2979 | break; | |
2980 | case VDEV_AUX_CORRUPT_DATA: | |
2981 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
2982 | break; | |
2983 | case VDEV_AUX_NO_REPLICAS: | |
2984 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
2985 | break; | |
2986 | case VDEV_AUX_BAD_GUID_SUM: | |
2987 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
2988 | break; | |
2989 | case VDEV_AUX_TOO_SMALL: | |
2990 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
2991 | break; | |
2992 | case VDEV_AUX_BAD_LABEL: | |
2993 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
2994 | break; | |
2995 | default: | |
2996 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
2997 | } | |
2998 | ||
b128c09f | 2999 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3000 | } |
3001 | ||
3002 | /* Erase any notion of persistent removed state */ | |
3003 | vd->vdev_removed = B_FALSE; | |
3004 | } else { | |
3005 | vd->vdev_removed = B_FALSE; | |
3006 | } | |
3007 | ||
9babb374 BB |
3008 | if (!isopen && vd->vdev_parent) |
3009 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3010 | } |
b128c09f BB |
3011 | |
3012 | /* | |
3013 | * Check the vdev configuration to ensure that it's capable of supporting | |
3014 | * a root pool. Currently, we do not support RAID-Z or partial configuration. | |
3015 | * In addition, only a single top-level vdev is allowed and none of the leaves | |
3016 | * can be wholedisks. | |
3017 | */ | |
3018 | boolean_t | |
3019 | vdev_is_bootable(vdev_t *vd) | |
3020 | { | |
b128c09f BB |
3021 | if (!vd->vdev_ops->vdev_op_leaf) { |
3022 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3023 | ||
3024 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3025 | vd->vdev_children > 1) { | |
3026 | return (B_FALSE); | |
3027 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3028 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3029 | return (B_FALSE); | |
3030 | } | |
3031 | } else if (vd->vdev_wholedisk == 1) { | |
3032 | return (B_FALSE); | |
3033 | } | |
3034 | ||
9babb374 | 3035 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3036 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3037 | return (B_FALSE); | |
3038 | } | |
3039 | return (B_TRUE); | |
3040 | } | |
9babb374 | 3041 | |
428870ff BB |
3042 | /* |
3043 | * Load the state from the original vdev tree (ovd) which | |
3044 | * we've retrieved from the MOS config object. If the original | |
3045 | * vdev was offline then we transfer that state to the device | |
3046 | * in the current vdev tree (nvd). | |
3047 | */ | |
9babb374 | 3048 | void |
428870ff | 3049 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3050 | { |
428870ff | 3051 | spa_t *spa = nvd->vdev_spa; |
9babb374 | 3052 | |
428870ff BB |
3053 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
3054 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); | |
9babb374 | 3055 | |
428870ff BB |
3056 | for (int c = 0; c < nvd->vdev_children; c++) |
3057 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); | |
9babb374 | 3058 | |
428870ff | 3059 | if (nvd->vdev_ops->vdev_op_leaf && ovd->vdev_offline) { |
9babb374 BB |
3060 | /* |
3061 | * It would be nice to call vdev_offline() | |
3062 | * directly but the pool isn't fully loaded and | |
3063 | * the txg threads have not been started yet. | |
3064 | */ | |
428870ff BB |
3065 | nvd->vdev_offline = ovd->vdev_offline; |
3066 | vdev_reopen(nvd->vdev_top); | |
9babb374 BB |
3067 | } |
3068 | } | |
3069 | ||
3070 | /* | |
3071 | * Expand a vdev if possible. | |
3072 | */ | |
3073 | void | |
3074 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3075 | { | |
3076 | ASSERT(vd->vdev_top == vd); | |
3077 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3078 | ||
3079 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3080 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3081 | vdev_config_dirty(vd); | |
3082 | } | |
3083 | } | |
428870ff BB |
3084 | |
3085 | /* | |
3086 | * Split a vdev. | |
3087 | */ | |
3088 | void | |
3089 | vdev_split(vdev_t *vd) | |
3090 | { | |
3091 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3092 | ||
3093 | vdev_remove_child(pvd, vd); | |
3094 | vdev_compact_children(pvd); | |
3095 | ||
3096 | cvd = pvd->vdev_child[0]; | |
3097 | if (pvd->vdev_children == 1) { | |
3098 | vdev_remove_parent(cvd); | |
3099 | cvd->vdev_splitting = B_TRUE; | |
3100 | } | |
3101 | vdev_propagate_state(cvd); | |
3102 | } |