]>
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 | /* | |
d164b209 | 23 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
24 | * Use is subject to license terms. |
25 | */ | |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/fm/fs/zfs.h> | |
29 | #include <sys/spa.h> | |
30 | #include <sys/spa_impl.h> | |
31 | #include <sys/dmu.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/vdev_impl.h> | |
34 | #include <sys/uberblock_impl.h> | |
35 | #include <sys/metaslab.h> | |
36 | #include <sys/metaslab_impl.h> | |
37 | #include <sys/space_map.h> | |
38 | #include <sys/zio.h> | |
39 | #include <sys/zap.h> | |
40 | #include <sys/fs/zfs.h> | |
b128c09f | 41 | #include <sys/arc.h> |
9babb374 | 42 | #include <sys/zil.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, | |
57 | NULL | |
58 | }; | |
59 | ||
b128c09f BB |
60 | /* maximum scrub/resilver I/O queue per leaf vdev */ |
61 | int zfs_scrub_limit = 10; | |
34dc7c2f BB |
62 | |
63 | /* | |
64 | * Given a vdev type, return the appropriate ops vector. | |
65 | */ | |
66 | static vdev_ops_t * | |
67 | vdev_getops(const char *type) | |
68 | { | |
69 | vdev_ops_t *ops, **opspp; | |
70 | ||
71 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
72 | if (strcmp(ops->vdev_op_type, type) == 0) | |
73 | break; | |
74 | ||
75 | return (ops); | |
76 | } | |
77 | ||
78 | /* | |
79 | * Default asize function: return the MAX of psize with the asize of | |
80 | * all children. This is what's used by anything other than RAID-Z. | |
81 | */ | |
82 | uint64_t | |
83 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
84 | { | |
85 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
86 | uint64_t csize; | |
34dc7c2f | 87 | |
9babb374 | 88 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
89 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
90 | asize = MAX(asize, csize); | |
91 | } | |
92 | ||
93 | return (asize); | |
94 | } | |
95 | ||
96 | /* | |
9babb374 BB |
97 | * Get the minimum allocatable size. We define the allocatable size as |
98 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
99 | * replace or attach devices which don't have the same physical size but | |
100 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
101 | */ |
102 | uint64_t | |
9babb374 | 103 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 104 | { |
9babb374 | 105 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 106 | |
9babb374 BB |
107 | /* |
108 | * The our parent is NULL (inactive spare or cache) or is the root, | |
109 | * just return our own asize. | |
110 | */ | |
111 | if (pvd == NULL) | |
112 | return (vd->vdev_asize); | |
34dc7c2f BB |
113 | |
114 | /* | |
9babb374 BB |
115 | * The top-level vdev just returns the allocatable size rounded |
116 | * to the nearest metaslab. | |
34dc7c2f | 117 | */ |
9babb374 BB |
118 | if (vd == vd->vdev_top) |
119 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 120 | |
9babb374 BB |
121 | /* |
122 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
123 | * so each child must provide at least 1/Nth of its asize. | |
124 | */ | |
125 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
126 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 127 | |
9babb374 BB |
128 | return (pvd->vdev_min_asize); |
129 | } | |
130 | ||
131 | void | |
132 | vdev_set_min_asize(vdev_t *vd) | |
133 | { | |
134 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 135 | |
9babb374 BB |
136 | for (int c = 0; c < vd->vdev_children; c++) |
137 | vdev_set_min_asize(vd->vdev_child[c]); | |
34dc7c2f BB |
138 | } |
139 | ||
140 | vdev_t * | |
141 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
142 | { | |
143 | vdev_t *rvd = spa->spa_root_vdev; | |
144 | ||
b128c09f | 145 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 146 | |
b128c09f BB |
147 | if (vdev < rvd->vdev_children) { |
148 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 149 | return (rvd->vdev_child[vdev]); |
b128c09f | 150 | } |
34dc7c2f BB |
151 | |
152 | return (NULL); | |
153 | } | |
154 | ||
155 | vdev_t * | |
156 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
157 | { | |
34dc7c2f BB |
158 | vdev_t *mvd; |
159 | ||
160 | if (vd->vdev_guid == guid) | |
161 | return (vd); | |
162 | ||
9babb374 | 163 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
164 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
165 | NULL) | |
166 | return (mvd); | |
167 | ||
168 | return (NULL); | |
169 | } | |
170 | ||
171 | void | |
172 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
173 | { | |
174 | size_t oldsize, newsize; | |
175 | uint64_t id = cvd->vdev_id; | |
176 | vdev_t **newchild; | |
177 | ||
b128c09f | 178 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
179 | ASSERT(cvd->vdev_parent == NULL); |
180 | ||
181 | cvd->vdev_parent = pvd; | |
182 | ||
183 | if (pvd == NULL) | |
184 | return; | |
185 | ||
186 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
187 | ||
188 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
189 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
190 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
191 | ||
192 | newchild = kmem_zalloc(newsize, KM_SLEEP); | |
193 | if (pvd->vdev_child != NULL) { | |
194 | bcopy(pvd->vdev_child, newchild, oldsize); | |
195 | kmem_free(pvd->vdev_child, oldsize); | |
196 | } | |
197 | ||
198 | pvd->vdev_child = newchild; | |
199 | pvd->vdev_child[id] = cvd; | |
200 | ||
201 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
202 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
203 | ||
204 | /* | |
205 | * Walk up all ancestors to update guid sum. | |
206 | */ | |
207 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
208 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
209 | ||
210 | if (cvd->vdev_ops->vdev_op_leaf) | |
211 | cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; | |
212 | } | |
213 | ||
214 | void | |
215 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
216 | { | |
217 | int c; | |
218 | uint_t id = cvd->vdev_id; | |
219 | ||
220 | ASSERT(cvd->vdev_parent == pvd); | |
221 | ||
222 | if (pvd == NULL) | |
223 | return; | |
224 | ||
225 | ASSERT(id < pvd->vdev_children); | |
226 | ASSERT(pvd->vdev_child[id] == cvd); | |
227 | ||
228 | pvd->vdev_child[id] = NULL; | |
229 | cvd->vdev_parent = NULL; | |
230 | ||
231 | for (c = 0; c < pvd->vdev_children; c++) | |
232 | if (pvd->vdev_child[c]) | |
233 | break; | |
234 | ||
235 | if (c == pvd->vdev_children) { | |
236 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
237 | pvd->vdev_child = NULL; | |
238 | pvd->vdev_children = 0; | |
239 | } | |
240 | ||
241 | /* | |
242 | * Walk up all ancestors to update guid sum. | |
243 | */ | |
244 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
245 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
246 | ||
247 | if (cvd->vdev_ops->vdev_op_leaf) | |
248 | cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Remove any holes in the child array. | |
253 | */ | |
254 | void | |
255 | vdev_compact_children(vdev_t *pvd) | |
256 | { | |
257 | vdev_t **newchild, *cvd; | |
258 | int oldc = pvd->vdev_children; | |
9babb374 | 259 | int newc; |
34dc7c2f | 260 | |
b128c09f | 261 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 262 | |
9babb374 | 263 | for (int c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
264 | if (pvd->vdev_child[c]) |
265 | newc++; | |
266 | ||
267 | newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); | |
268 | ||
9babb374 | 269 | for (int c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
270 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
271 | newchild[newc] = cvd; | |
272 | cvd->vdev_id = newc++; | |
273 | } | |
274 | } | |
275 | ||
276 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
277 | pvd->vdev_child = newchild; | |
278 | pvd->vdev_children = newc; | |
279 | } | |
280 | ||
281 | /* | |
282 | * Allocate and minimally initialize a vdev_t. | |
283 | */ | |
284 | static vdev_t * | |
285 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) | |
286 | { | |
287 | vdev_t *vd; | |
288 | ||
289 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); | |
290 | ||
291 | if (spa->spa_root_vdev == NULL) { | |
292 | ASSERT(ops == &vdev_root_ops); | |
293 | spa->spa_root_vdev = vd; | |
294 | } | |
295 | ||
296 | if (guid == 0) { | |
297 | if (spa->spa_root_vdev == vd) { | |
298 | /* | |
299 | * The root vdev's guid will also be the pool guid, | |
300 | * which must be unique among all pools. | |
301 | */ | |
302 | while (guid == 0 || spa_guid_exists(guid, 0)) | |
303 | guid = spa_get_random(-1ULL); | |
304 | } else { | |
305 | /* | |
306 | * Any other vdev's guid must be unique within the pool. | |
307 | */ | |
308 | while (guid == 0 || | |
309 | spa_guid_exists(spa_guid(spa), guid)) | |
310 | guid = spa_get_random(-1ULL); | |
311 | } | |
312 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
313 | } | |
314 | ||
315 | vd->vdev_spa = spa; | |
316 | vd->vdev_id = id; | |
317 | vd->vdev_guid = guid; | |
318 | vd->vdev_guid_sum = guid; | |
319 | vd->vdev_ops = ops; | |
320 | vd->vdev_state = VDEV_STATE_CLOSED; | |
321 | ||
322 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); | |
323 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 324 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
fb5f0bc8 BB |
325 | for (int t = 0; t < DTL_TYPES; t++) { |
326 | space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, | |
327 | &vd->vdev_dtl_lock); | |
328 | } | |
34dc7c2f BB |
329 | txg_list_create(&vd->vdev_ms_list, |
330 | offsetof(struct metaslab, ms_txg_node)); | |
331 | txg_list_create(&vd->vdev_dtl_list, | |
332 | offsetof(struct vdev, vdev_dtl_node)); | |
333 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
334 | vdev_queue_init(vd); | |
335 | vdev_cache_init(vd); | |
336 | ||
337 | return (vd); | |
338 | } | |
339 | ||
340 | /* | |
341 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
342 | * creating a new vdev or loading an existing one - the behavior is slightly | |
343 | * different for each case. | |
344 | */ | |
345 | int | |
346 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
347 | int alloctype) | |
348 | { | |
349 | vdev_ops_t *ops; | |
350 | char *type; | |
351 | uint64_t guid = 0, islog, nparity; | |
352 | vdev_t *vd; | |
353 | ||
b128c09f | 354 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
355 | |
356 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
357 | return (EINVAL); | |
358 | ||
359 | if ((ops = vdev_getops(type)) == NULL) | |
360 | return (EINVAL); | |
361 | ||
362 | /* | |
363 | * If this is a load, get the vdev guid from the nvlist. | |
364 | * Otherwise, vdev_alloc_common() will generate one for us. | |
365 | */ | |
366 | if (alloctype == VDEV_ALLOC_LOAD) { | |
367 | uint64_t label_id; | |
368 | ||
369 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
370 | label_id != id) | |
371 | return (EINVAL); | |
372 | ||
373 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
374 | return (EINVAL); | |
375 | } else if (alloctype == VDEV_ALLOC_SPARE) { | |
376 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
377 | return (EINVAL); | |
378 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { | |
379 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
380 | return (EINVAL); | |
9babb374 BB |
381 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
382 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
383 | return (EINVAL); | |
34dc7c2f BB |
384 | } |
385 | ||
386 | /* | |
387 | * The first allocated vdev must be of type 'root'. | |
388 | */ | |
389 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
390 | return (EINVAL); | |
391 | ||
392 | /* | |
393 | * Determine whether we're a log vdev. | |
394 | */ | |
395 | islog = 0; | |
396 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
397 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
398 | return (ENOTSUP); | |
399 | ||
400 | /* | |
401 | * Set the nparity property for RAID-Z vdevs. | |
402 | */ | |
403 | nparity = -1ULL; | |
404 | if (ops == &vdev_raidz_ops) { | |
405 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
406 | &nparity) == 0) { | |
407 | /* | |
45d1cae3 | 408 | * Currently, we can only support 3 parity devices. |
34dc7c2f | 409 | */ |
45d1cae3 | 410 | if (nparity == 0 || nparity > 3) |
34dc7c2f BB |
411 | return (EINVAL); |
412 | /* | |
45d1cae3 BB |
413 | * Previous versions could only support 1 or 2 parity |
414 | * device. | |
34dc7c2f | 415 | */ |
45d1cae3 BB |
416 | if (nparity > 1 && |
417 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
418 | return (ENOTSUP); | |
419 | if (nparity > 2 && | |
420 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
34dc7c2f BB |
421 | return (ENOTSUP); |
422 | } else { | |
423 | /* | |
424 | * We require the parity to be specified for SPAs that | |
425 | * support multiple parity levels. | |
426 | */ | |
45d1cae3 | 427 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
34dc7c2f BB |
428 | return (EINVAL); |
429 | /* | |
430 | * Otherwise, we default to 1 parity device for RAID-Z. | |
431 | */ | |
432 | nparity = 1; | |
433 | } | |
434 | } else { | |
435 | nparity = 0; | |
436 | } | |
437 | ASSERT(nparity != -1ULL); | |
438 | ||
439 | vd = vdev_alloc_common(spa, id, guid, ops); | |
440 | ||
441 | vd->vdev_islog = islog; | |
442 | vd->vdev_nparity = nparity; | |
443 | ||
444 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
445 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
446 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
447 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
448 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
449 | &vd->vdev_physpath) == 0) | |
450 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
451 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
452 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
453 | |
454 | /* | |
455 | * Set the whole_disk property. If it's not specified, leave the value | |
456 | * as -1. | |
457 | */ | |
458 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
459 | &vd->vdev_wholedisk) != 0) | |
460 | vd->vdev_wholedisk = -1ULL; | |
461 | ||
462 | /* | |
463 | * Look for the 'not present' flag. This will only be set if the device | |
464 | * was not present at the time of import. | |
465 | */ | |
9babb374 BB |
466 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
467 | &vd->vdev_not_present); | |
34dc7c2f BB |
468 | |
469 | /* | |
470 | * Get the alignment requirement. | |
471 | */ | |
472 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
473 | ||
474 | /* | |
475 | * If we're a top-level vdev, try to load the allocation parameters. | |
476 | */ | |
477 | if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) { | |
478 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, | |
479 | &vd->vdev_ms_array); | |
480 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
481 | &vd->vdev_ms_shift); | |
482 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
483 | &vd->vdev_asize); | |
484 | } | |
485 | ||
486 | /* | |
487 | * If we're a leaf vdev, try to load the DTL object and other state. | |
488 | */ | |
b128c09f | 489 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
490 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
491 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
492 | if (alloctype == VDEV_ALLOC_LOAD) { |
493 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
fb5f0bc8 | 494 | &vd->vdev_dtl_smo.smo_object); |
b128c09f BB |
495 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
496 | &vd->vdev_unspare); | |
497 | } | |
9babb374 BB |
498 | |
499 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
500 | uint64_t spare = 0; | |
501 | ||
502 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
503 | &spare) == 0 && spare) | |
504 | spa_spare_add(vd); | |
505 | } | |
506 | ||
34dc7c2f BB |
507 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
508 | &vd->vdev_offline); | |
b128c09f | 509 | |
34dc7c2f BB |
510 | /* |
511 | * When importing a pool, we want to ignore the persistent fault | |
512 | * state, as the diagnosis made on another system may not be | |
513 | * valid in the current context. | |
514 | */ | |
515 | if (spa->spa_load_state == SPA_LOAD_OPEN) { | |
516 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, | |
517 | &vd->vdev_faulted); | |
518 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
519 | &vd->vdev_degraded); | |
520 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
521 | &vd->vdev_removed); | |
522 | } | |
523 | } | |
524 | ||
525 | /* | |
526 | * Add ourselves to the parent's list of children. | |
527 | */ | |
528 | vdev_add_child(parent, vd); | |
529 | ||
530 | *vdp = vd; | |
531 | ||
532 | return (0); | |
533 | } | |
534 | ||
535 | void | |
536 | vdev_free(vdev_t *vd) | |
537 | { | |
34dc7c2f BB |
538 | spa_t *spa = vd->vdev_spa; |
539 | ||
540 | /* | |
541 | * vdev_free() implies closing the vdev first. This is simpler than | |
542 | * trying to ensure complicated semantics for all callers. | |
543 | */ | |
544 | vdev_close(vd); | |
545 | ||
b128c09f | 546 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
34dc7c2f BB |
547 | |
548 | /* | |
549 | * Free all children. | |
550 | */ | |
9babb374 | 551 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
552 | vdev_free(vd->vdev_child[c]); |
553 | ||
554 | ASSERT(vd->vdev_child == NULL); | |
555 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
556 | ||
557 | /* | |
558 | * Discard allocation state. | |
559 | */ | |
560 | if (vd == vd->vdev_top) | |
561 | vdev_metaslab_fini(vd); | |
562 | ||
563 | ASSERT3U(vd->vdev_stat.vs_space, ==, 0); | |
564 | ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); | |
565 | ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); | |
566 | ||
567 | /* | |
568 | * Remove this vdev from its parent's child list. | |
569 | */ | |
570 | vdev_remove_child(vd->vdev_parent, vd); | |
571 | ||
572 | ASSERT(vd->vdev_parent == NULL); | |
573 | ||
574 | /* | |
575 | * Clean up vdev structure. | |
576 | */ | |
577 | vdev_queue_fini(vd); | |
578 | vdev_cache_fini(vd); | |
579 | ||
580 | if (vd->vdev_path) | |
581 | spa_strfree(vd->vdev_path); | |
582 | if (vd->vdev_devid) | |
583 | spa_strfree(vd->vdev_devid); | |
584 | if (vd->vdev_physpath) | |
585 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
586 | if (vd->vdev_fru) |
587 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
588 | |
589 | if (vd->vdev_isspare) | |
590 | spa_spare_remove(vd); | |
591 | if (vd->vdev_isl2cache) | |
592 | spa_l2cache_remove(vd); | |
593 | ||
594 | txg_list_destroy(&vd->vdev_ms_list); | |
595 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 596 | |
34dc7c2f | 597 | mutex_enter(&vd->vdev_dtl_lock); |
fb5f0bc8 BB |
598 | for (int t = 0; t < DTL_TYPES; t++) { |
599 | space_map_unload(&vd->vdev_dtl[t]); | |
600 | space_map_destroy(&vd->vdev_dtl[t]); | |
601 | } | |
34dc7c2f | 602 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 603 | |
34dc7c2f BB |
604 | mutex_destroy(&vd->vdev_dtl_lock); |
605 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 606 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
607 | |
608 | if (vd == spa->spa_root_vdev) | |
609 | spa->spa_root_vdev = NULL; | |
610 | ||
611 | kmem_free(vd, sizeof (vdev_t)); | |
612 | } | |
613 | ||
614 | /* | |
615 | * Transfer top-level vdev state from svd to tvd. | |
616 | */ | |
617 | static void | |
618 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
619 | { | |
620 | spa_t *spa = svd->vdev_spa; | |
621 | metaslab_t *msp; | |
622 | vdev_t *vd; | |
623 | int t; | |
624 | ||
625 | ASSERT(tvd == tvd->vdev_top); | |
626 | ||
627 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
628 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
629 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
630 | ||
631 | svd->vdev_ms_array = 0; | |
632 | svd->vdev_ms_shift = 0; | |
633 | svd->vdev_ms_count = 0; | |
634 | ||
635 | tvd->vdev_mg = svd->vdev_mg; | |
636 | tvd->vdev_ms = svd->vdev_ms; | |
637 | ||
638 | svd->vdev_mg = NULL; | |
639 | svd->vdev_ms = NULL; | |
640 | ||
641 | if (tvd->vdev_mg != NULL) | |
642 | tvd->vdev_mg->mg_vd = tvd; | |
643 | ||
644 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
645 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
646 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
647 | ||
648 | svd->vdev_stat.vs_alloc = 0; | |
649 | svd->vdev_stat.vs_space = 0; | |
650 | svd->vdev_stat.vs_dspace = 0; | |
651 | ||
652 | for (t = 0; t < TXG_SIZE; t++) { | |
653 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
654 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
655 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
656 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
657 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
658 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
659 | } | |
660 | ||
b128c09f | 661 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
662 | vdev_config_clean(svd); |
663 | vdev_config_dirty(tvd); | |
664 | } | |
665 | ||
b128c09f BB |
666 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
667 | vdev_state_clean(svd); | |
668 | vdev_state_dirty(tvd); | |
669 | } | |
670 | ||
34dc7c2f BB |
671 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
672 | svd->vdev_deflate_ratio = 0; | |
673 | ||
674 | tvd->vdev_islog = svd->vdev_islog; | |
675 | svd->vdev_islog = 0; | |
676 | } | |
677 | ||
678 | static void | |
679 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
680 | { | |
34dc7c2f BB |
681 | if (vd == NULL) |
682 | return; | |
683 | ||
684 | vd->vdev_top = tvd; | |
685 | ||
9babb374 | 686 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
687 | vdev_top_update(tvd, vd->vdev_child[c]); |
688 | } | |
689 | ||
690 | /* | |
691 | * Add a mirror/replacing vdev above an existing vdev. | |
692 | */ | |
693 | vdev_t * | |
694 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
695 | { | |
696 | spa_t *spa = cvd->vdev_spa; | |
697 | vdev_t *pvd = cvd->vdev_parent; | |
698 | vdev_t *mvd; | |
699 | ||
b128c09f | 700 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
701 | |
702 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
703 | ||
704 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 705 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
34dc7c2f BB |
706 | mvd->vdev_ashift = cvd->vdev_ashift; |
707 | mvd->vdev_state = cvd->vdev_state; | |
708 | ||
709 | vdev_remove_child(pvd, cvd); | |
710 | vdev_add_child(pvd, mvd); | |
711 | cvd->vdev_id = mvd->vdev_children; | |
712 | vdev_add_child(mvd, cvd); | |
713 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
714 | ||
715 | if (mvd == mvd->vdev_top) | |
716 | vdev_top_transfer(cvd, mvd); | |
717 | ||
718 | return (mvd); | |
719 | } | |
720 | ||
721 | /* | |
722 | * Remove a 1-way mirror/replacing vdev from the tree. | |
723 | */ | |
724 | void | |
725 | vdev_remove_parent(vdev_t *cvd) | |
726 | { | |
727 | vdev_t *mvd = cvd->vdev_parent; | |
728 | vdev_t *pvd = mvd->vdev_parent; | |
729 | ||
b128c09f | 730 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
731 | |
732 | ASSERT(mvd->vdev_children == 1); | |
733 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
734 | mvd->vdev_ops == &vdev_replacing_ops || | |
735 | mvd->vdev_ops == &vdev_spare_ops); | |
736 | cvd->vdev_ashift = mvd->vdev_ashift; | |
737 | ||
738 | vdev_remove_child(mvd, cvd); | |
739 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 740 | |
34dc7c2f | 741 | /* |
b128c09f BB |
742 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
743 | * Otherwise, we could have detached an offline device, and when we | |
744 | * go to import the pool we'll think we have two top-level vdevs, | |
745 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 746 | */ |
fb5f0bc8 BB |
747 | if (mvd->vdev_top == mvd) { |
748 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
749 | cvd->vdev_guid += guid_delta; | |
750 | cvd->vdev_guid_sum += guid_delta; | |
751 | } | |
b128c09f BB |
752 | cvd->vdev_id = mvd->vdev_id; |
753 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
754 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
755 | ||
756 | if (cvd == cvd->vdev_top) | |
757 | vdev_top_transfer(mvd, cvd); | |
758 | ||
759 | ASSERT(mvd->vdev_children == 0); | |
760 | vdev_free(mvd); | |
761 | } | |
762 | ||
763 | int | |
764 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
765 | { | |
766 | spa_t *spa = vd->vdev_spa; | |
767 | objset_t *mos = spa->spa_meta_objset; | |
768 | metaslab_class_t *mc; | |
769 | uint64_t m; | |
770 | uint64_t oldc = vd->vdev_ms_count; | |
771 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
772 | metaslab_t **mspp; | |
773 | int error; | |
774 | ||
775 | if (vd->vdev_ms_shift == 0) /* not being allocated from yet */ | |
776 | return (0); | |
777 | ||
9babb374 BB |
778 | /* |
779 | * Compute the raidz-deflation ratio. Note, we hard-code | |
780 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
781 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
782 | * or we will inconsistently account for existing bp's. | |
783 | */ | |
784 | vd->vdev_deflate_ratio = (1 << 17) / | |
785 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
786 | ||
34dc7c2f BB |
787 | ASSERT(oldc <= newc); |
788 | ||
789 | if (vd->vdev_islog) | |
790 | mc = spa->spa_log_class; | |
791 | else | |
792 | mc = spa->spa_normal_class; | |
793 | ||
794 | if (vd->vdev_mg == NULL) | |
795 | vd->vdev_mg = metaslab_group_create(mc, vd); | |
796 | ||
797 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); | |
798 | ||
799 | if (oldc != 0) { | |
800 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
801 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
802 | } | |
803 | ||
804 | vd->vdev_ms = mspp; | |
805 | vd->vdev_ms_count = newc; | |
806 | ||
807 | for (m = oldc; m < newc; m++) { | |
808 | space_map_obj_t smo = { 0, 0, 0 }; | |
809 | if (txg == 0) { | |
810 | uint64_t object = 0; | |
811 | error = dmu_read(mos, vd->vdev_ms_array, | |
9babb374 BB |
812 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
813 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
814 | if (error) |
815 | return (error); | |
816 | if (object != 0) { | |
817 | dmu_buf_t *db; | |
818 | error = dmu_bonus_hold(mos, object, FTAG, &db); | |
819 | if (error) | |
820 | return (error); | |
821 | ASSERT3U(db->db_size, >=, sizeof (smo)); | |
822 | bcopy(db->db_data, &smo, sizeof (smo)); | |
823 | ASSERT3U(smo.smo_object, ==, object); | |
824 | dmu_buf_rele(db, FTAG); | |
825 | } | |
826 | } | |
827 | vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, | |
828 | m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); | |
829 | } | |
830 | ||
831 | return (0); | |
832 | } | |
833 | ||
834 | void | |
835 | vdev_metaslab_fini(vdev_t *vd) | |
836 | { | |
837 | uint64_t m; | |
838 | uint64_t count = vd->vdev_ms_count; | |
839 | ||
840 | if (vd->vdev_ms != NULL) { | |
841 | for (m = 0; m < count; m++) | |
842 | if (vd->vdev_ms[m] != NULL) | |
843 | metaslab_fini(vd->vdev_ms[m]); | |
844 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); | |
845 | vd->vdev_ms = NULL; | |
846 | } | |
847 | } | |
848 | ||
b128c09f BB |
849 | typedef struct vdev_probe_stats { |
850 | boolean_t vps_readable; | |
851 | boolean_t vps_writeable; | |
852 | int vps_flags; | |
b128c09f BB |
853 | } vdev_probe_stats_t; |
854 | ||
855 | static void | |
856 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 857 | { |
fb5f0bc8 | 858 | spa_t *spa = zio->io_spa; |
d164b209 | 859 | vdev_t *vd = zio->io_vd; |
b128c09f | 860 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
861 | |
862 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
863 | |
864 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
865 | if (zio->io_error == 0) |
866 | vps->vps_readable = 1; | |
fb5f0bc8 | 867 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 868 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
869 | zio->io_offset, zio->io_size, zio->io_data, |
870 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
871 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
872 | } else { | |
873 | zio_buf_free(zio->io_data, zio->io_size); | |
874 | } | |
875 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
876 | if (zio->io_error == 0) |
877 | vps->vps_writeable = 1; | |
878 | zio_buf_free(zio->io_data, zio->io_size); | |
879 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 880 | zio_t *pio; |
b128c09f BB |
881 | |
882 | vd->vdev_cant_read |= !vps->vps_readable; | |
883 | vd->vdev_cant_write |= !vps->vps_writeable; | |
884 | ||
885 | if (vdev_readable(vd) && | |
fb5f0bc8 | 886 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
887 | zio->io_error = 0; |
888 | } else { | |
889 | ASSERT(zio->io_error != 0); | |
890 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 891 | spa, vd, NULL, 0, 0); |
b128c09f BB |
892 | zio->io_error = ENXIO; |
893 | } | |
d164b209 BB |
894 | |
895 | mutex_enter(&vd->vdev_probe_lock); | |
896 | ASSERT(vd->vdev_probe_zio == zio); | |
897 | vd->vdev_probe_zio = NULL; | |
898 | mutex_exit(&vd->vdev_probe_lock); | |
899 | ||
900 | while ((pio = zio_walk_parents(zio)) != NULL) | |
901 | if (!vdev_accessible(vd, pio)) | |
902 | pio->io_error = ENXIO; | |
903 | ||
b128c09f BB |
904 | kmem_free(vps, sizeof (*vps)); |
905 | } | |
906 | } | |
34dc7c2f | 907 | |
b128c09f BB |
908 | /* |
909 | * Determine whether this device is accessible by reading and writing | |
910 | * to several known locations: the pad regions of each vdev label | |
911 | * but the first (which we leave alone in case it contains a VTOC). | |
912 | */ | |
913 | zio_t * | |
d164b209 | 914 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
915 | { |
916 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
917 | vdev_probe_stats_t *vps = NULL; |
918 | zio_t *pio; | |
919 | ||
920 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 921 | |
d164b209 BB |
922 | /* |
923 | * Don't probe the probe. | |
924 | */ | |
925 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
926 | return (NULL); | |
b128c09f | 927 | |
d164b209 BB |
928 | /* |
929 | * To prevent 'probe storms' when a device fails, we create | |
930 | * just one probe i/o at a time. All zios that want to probe | |
931 | * this vdev will become parents of the probe io. | |
932 | */ | |
933 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 934 | |
d164b209 BB |
935 | if ((pio = vd->vdev_probe_zio) == NULL) { |
936 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); | |
937 | ||
938 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
939 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 940 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
941 | |
942 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
943 | /* | |
944 | * vdev_cant_read and vdev_cant_write can only | |
945 | * transition from TRUE to FALSE when we have the | |
946 | * SCL_ZIO lock as writer; otherwise they can only | |
947 | * transition from FALSE to TRUE. This ensures that | |
948 | * any zio looking at these values can assume that | |
949 | * failures persist for the life of the I/O. That's | |
950 | * important because when a device has intermittent | |
951 | * connectivity problems, we want to ensure that | |
952 | * they're ascribed to the device (ENXIO) and not | |
953 | * the zio (EIO). | |
954 | * | |
955 | * Since we hold SCL_ZIO as writer here, clear both | |
956 | * values so the probe can reevaluate from first | |
957 | * principles. | |
958 | */ | |
959 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
960 | vd->vdev_cant_read = B_FALSE; | |
961 | vd->vdev_cant_write = B_FALSE; | |
962 | } | |
963 | ||
964 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
965 | vdev_probe_done, vps, | |
966 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
967 | ||
968 | if (zio != NULL) { | |
969 | vd->vdev_probe_wanted = B_TRUE; | |
970 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
971 | } | |
b128c09f BB |
972 | } |
973 | ||
d164b209 BB |
974 | if (zio != NULL) |
975 | zio_add_child(zio, pio); | |
b128c09f | 976 | |
d164b209 | 977 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 978 | |
d164b209 BB |
979 | if (vps == NULL) { |
980 | ASSERT(zio != NULL); | |
981 | return (NULL); | |
982 | } | |
b128c09f BB |
983 | |
984 | for (int l = 1; l < VDEV_LABELS; l++) { | |
d164b209 | 985 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 986 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
987 | offsetof(vdev_label_t, vl_pad2)), |
988 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
989 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
990 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
991 | } | |
992 | ||
d164b209 BB |
993 | if (zio == NULL) |
994 | return (pio); | |
995 | ||
996 | zio_nowait(pio); | |
997 | return (NULL); | |
34dc7c2f BB |
998 | } |
999 | ||
45d1cae3 BB |
1000 | static void |
1001 | vdev_open_child(void *arg) | |
1002 | { | |
1003 | vdev_t *vd = arg; | |
1004 | ||
1005 | vd->vdev_open_thread = curthread; | |
1006 | vd->vdev_open_error = vdev_open(vd); | |
1007 | vd->vdev_open_thread = NULL; | |
1008 | } | |
1009 | ||
1010 | void | |
1011 | vdev_open_children(vdev_t *vd) | |
1012 | { | |
1013 | taskq_t *tq; | |
1014 | int children = vd->vdev_children; | |
1015 | ||
1016 | tq = taskq_create("vdev_open", children, minclsyspri, | |
1017 | children, children, TASKQ_PREPOPULATE); | |
1018 | ||
1019 | for (int c = 0; c < children; c++) | |
1020 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], | |
1021 | TQ_SLEEP) != NULL); | |
1022 | ||
1023 | taskq_destroy(tq); | |
1024 | } | |
1025 | ||
34dc7c2f BB |
1026 | /* |
1027 | * Prepare a virtual device for access. | |
1028 | */ | |
1029 | int | |
1030 | vdev_open(vdev_t *vd) | |
1031 | { | |
fb5f0bc8 | 1032 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1033 | int error; |
34dc7c2f BB |
1034 | uint64_t osize = 0; |
1035 | uint64_t asize, psize; | |
1036 | uint64_t ashift = 0; | |
1037 | ||
45d1cae3 BB |
1038 | ASSERT(vd->vdev_open_thread == curthread || |
1039 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1040 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1041 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1042 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1043 | ||
34dc7c2f | 1044 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1045 | vd->vdev_cant_read = B_FALSE; |
1046 | vd->vdev_cant_write = B_FALSE; | |
1047 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f BB |
1048 | |
1049 | if (!vd->vdev_removed && vd->vdev_faulted) { | |
1050 | ASSERT(vd->vdev_children == 0); | |
1051 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1052 | VDEV_AUX_ERR_EXCEEDED); | |
1053 | return (ENXIO); | |
1054 | } else if (vd->vdev_offline) { | |
1055 | ASSERT(vd->vdev_children == 0); | |
1056 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
1057 | return (ENXIO); | |
1058 | } | |
1059 | ||
1060 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); | |
1061 | ||
1062 | if (zio_injection_enabled && error == 0) | |
9babb374 | 1063 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1064 | |
1065 | if (error) { | |
1066 | if (vd->vdev_removed && | |
1067 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1068 | vd->vdev_removed = B_FALSE; | |
1069 | ||
1070 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1071 | vd->vdev_stat.vs_aux); | |
1072 | return (error); | |
1073 | } | |
1074 | ||
1075 | vd->vdev_removed = B_FALSE; | |
1076 | ||
1077 | if (vd->vdev_degraded) { | |
1078 | ASSERT(vd->vdev_children == 0); | |
1079 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1080 | VDEV_AUX_ERR_EXCEEDED); | |
1081 | } else { | |
1082 | vd->vdev_state = VDEV_STATE_HEALTHY; | |
1083 | } | |
1084 | ||
9babb374 | 1085 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1086 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1087 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1088 | VDEV_AUX_NONE); | |
1089 | break; | |
1090 | } | |
9babb374 | 1091 | } |
34dc7c2f BB |
1092 | |
1093 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1094 | ||
1095 | if (vd->vdev_children == 0) { | |
1096 | if (osize < SPA_MINDEVSIZE) { | |
1097 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1098 | VDEV_AUX_TOO_SMALL); | |
1099 | return (EOVERFLOW); | |
1100 | } | |
1101 | psize = osize; | |
1102 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1103 | } else { | |
1104 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1105 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1106 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1107 | VDEV_AUX_TOO_SMALL); | |
1108 | return (EOVERFLOW); | |
1109 | } | |
1110 | psize = 0; | |
1111 | asize = osize; | |
1112 | } | |
1113 | ||
1114 | vd->vdev_psize = psize; | |
1115 | ||
9babb374 BB |
1116 | /* |
1117 | * Make sure the allocatable size hasn't shrunk. | |
1118 | */ | |
1119 | if (asize < vd->vdev_min_asize) { | |
1120 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1121 | VDEV_AUX_BAD_LABEL); | |
1122 | return (EINVAL); | |
1123 | } | |
1124 | ||
34dc7c2f BB |
1125 | if (vd->vdev_asize == 0) { |
1126 | /* | |
1127 | * This is the first-ever open, so use the computed values. | |
1128 | * For testing purposes, a higher ashift can be requested. | |
1129 | */ | |
1130 | vd->vdev_asize = asize; | |
1131 | vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); | |
1132 | } else { | |
1133 | /* | |
1134 | * Make sure the alignment requirement hasn't increased. | |
1135 | */ | |
1136 | if (ashift > vd->vdev_top->vdev_ashift) { | |
1137 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1138 | VDEV_AUX_BAD_LABEL); | |
1139 | return (EINVAL); | |
1140 | } | |
9babb374 | 1141 | } |
34dc7c2f | 1142 | |
9babb374 BB |
1143 | /* |
1144 | * If all children are healthy and the asize has increased, | |
1145 | * then we've experienced dynamic LUN growth. If automatic | |
1146 | * expansion is enabled then use the additional space. | |
1147 | */ | |
1148 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1149 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1150 | vd->vdev_asize = asize; | |
34dc7c2f | 1151 | |
9babb374 | 1152 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1153 | |
1154 | /* | |
1155 | * Ensure we can issue some IO before declaring the | |
1156 | * vdev open for business. | |
1157 | */ | |
b128c09f BB |
1158 | if (vd->vdev_ops->vdev_op_leaf && |
1159 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
34dc7c2f | 1160 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
b128c09f | 1161 | VDEV_AUX_IO_FAILURE); |
34dc7c2f BB |
1162 | return (error); |
1163 | } | |
1164 | ||
34dc7c2f | 1165 | /* |
b128c09f | 1166 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1167 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1168 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1169 | */ |
fb5f0bc8 BB |
1170 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1171 | vdev_resilver_needed(vd, NULL, NULL)) | |
1172 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1173 | |
1174 | return (0); | |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * Called once the vdevs are all opened, this routine validates the label | |
1179 | * contents. This needs to be done before vdev_load() so that we don't | |
1180 | * inadvertently do repair I/Os to the wrong device. | |
1181 | * | |
1182 | * This function will only return failure if one of the vdevs indicates that it | |
1183 | * has since been destroyed or exported. This is only possible if | |
1184 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1185 | * will be updated but the function will return 0. | |
1186 | */ | |
1187 | int | |
1188 | vdev_validate(vdev_t *vd) | |
1189 | { | |
1190 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1191 | nvlist_t *label; |
b128c09f | 1192 | uint64_t guid, top_guid; |
34dc7c2f BB |
1193 | uint64_t state; |
1194 | ||
9babb374 | 1195 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1196 | if (vdev_validate(vd->vdev_child[c]) != 0) |
1197 | return (EBADF); | |
1198 | ||
1199 | /* | |
1200 | * If the device has already failed, or was marked offline, don't do | |
1201 | * any further validation. Otherwise, label I/O will fail and we will | |
1202 | * overwrite the previous state. | |
1203 | */ | |
b128c09f | 1204 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
34dc7c2f BB |
1205 | |
1206 | if ((label = vdev_label_read_config(vd)) == NULL) { | |
1207 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1208 | VDEV_AUX_BAD_LABEL); | |
1209 | return (0); | |
1210 | } | |
1211 | ||
1212 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, | |
1213 | &guid) != 0 || guid != spa_guid(spa)) { | |
1214 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1215 | VDEV_AUX_CORRUPT_DATA); | |
1216 | nvlist_free(label); | |
1217 | return (0); | |
1218 | } | |
1219 | ||
b128c09f BB |
1220 | /* |
1221 | * If this vdev just became a top-level vdev because its | |
1222 | * sibling was detached, it will have adopted the parent's | |
1223 | * vdev guid -- but the label may or may not be on disk yet. | |
1224 | * Fortunately, either version of the label will have the | |
1225 | * same top guid, so if we're a top-level vdev, we can | |
1226 | * safely compare to that instead. | |
1227 | */ | |
34dc7c2f | 1228 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1229 | &guid) != 0 || |
1230 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1231 | &top_guid) != 0 || | |
1232 | (vd->vdev_guid != guid && | |
1233 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { | |
34dc7c2f BB |
1234 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1235 | VDEV_AUX_CORRUPT_DATA); | |
1236 | nvlist_free(label); | |
1237 | return (0); | |
1238 | } | |
1239 | ||
1240 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1241 | &state) != 0) { | |
1242 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1243 | VDEV_AUX_CORRUPT_DATA); | |
1244 | nvlist_free(label); | |
1245 | return (0); | |
1246 | } | |
1247 | ||
1248 | nvlist_free(label); | |
1249 | ||
45d1cae3 BB |
1250 | /* |
1251 | * If spa->spa_load_verbatim is true, no need to check the | |
1252 | * state of the pool. | |
1253 | */ | |
1254 | if (!spa->spa_load_verbatim && | |
1255 | spa->spa_load_state == SPA_LOAD_OPEN && | |
34dc7c2f BB |
1256 | state != POOL_STATE_ACTIVE) |
1257 | return (EBADF); | |
34dc7c2f | 1258 | |
b128c09f BB |
1259 | /* |
1260 | * If we were able to open and validate a vdev that was | |
1261 | * previously marked permanently unavailable, clear that state | |
1262 | * now. | |
1263 | */ | |
1264 | if (vd->vdev_not_present) | |
1265 | vd->vdev_not_present = 0; | |
1266 | } | |
34dc7c2f BB |
1267 | |
1268 | return (0); | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * Close a virtual device. | |
1273 | */ | |
1274 | void | |
1275 | vdev_close(vdev_t *vd) | |
1276 | { | |
fb5f0bc8 BB |
1277 | spa_t *spa = vd->vdev_spa; |
1278 | ||
1279 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1280 | ||
34dc7c2f BB |
1281 | vd->vdev_ops->vdev_op_close(vd); |
1282 | ||
1283 | vdev_cache_purge(vd); | |
1284 | ||
1285 | /* | |
9babb374 | 1286 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1287 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1288 | * it's still faulted. | |
1289 | */ | |
1290 | vd->vdev_prevstate = vd->vdev_state; | |
1291 | ||
1292 | if (vd->vdev_offline) | |
1293 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1294 | else | |
1295 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1296 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1297 | } | |
1298 | ||
1299 | void | |
1300 | vdev_reopen(vdev_t *vd) | |
1301 | { | |
1302 | spa_t *spa = vd->vdev_spa; | |
1303 | ||
b128c09f | 1304 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f BB |
1305 | |
1306 | vdev_close(vd); | |
1307 | (void) vdev_open(vd); | |
1308 | ||
1309 | /* | |
1310 | * Call vdev_validate() here to make sure we have the same device. | |
1311 | * Otherwise, a device with an invalid label could be successfully | |
1312 | * opened in response to vdev_reopen(). | |
1313 | */ | |
b128c09f BB |
1314 | if (vd->vdev_aux) { |
1315 | (void) vdev_validate_aux(vd); | |
1316 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1317 | vd->vdev_aux == &spa->spa_l2cache && |
1318 | !l2arc_vdev_present(vd)) | |
1319 | l2arc_add_vdev(spa, vd); | |
b128c09f BB |
1320 | } else { |
1321 | (void) vdev_validate(vd); | |
1322 | } | |
34dc7c2f BB |
1323 | |
1324 | /* | |
1325 | * Reassess parent vdev's health. | |
1326 | */ | |
1327 | vdev_propagate_state(vd); | |
1328 | } | |
1329 | ||
1330 | int | |
1331 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1332 | { | |
1333 | int error; | |
1334 | ||
1335 | /* | |
1336 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1337 | * For a create, however, we want to fail the request if | |
1338 | * there are any components we can't open. | |
1339 | */ | |
1340 | error = vdev_open(vd); | |
1341 | ||
1342 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1343 | vdev_close(vd); | |
1344 | return (error ? error : ENXIO); | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * Recursively initialize all labels. | |
1349 | */ | |
1350 | if ((error = vdev_label_init(vd, txg, isreplacing ? | |
1351 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { | |
1352 | vdev_close(vd); | |
1353 | return (error); | |
1354 | } | |
1355 | ||
1356 | return (0); | |
1357 | } | |
1358 | ||
34dc7c2f | 1359 | void |
9babb374 | 1360 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1361 | { |
1362 | /* | |
1363 | * Aim for roughly 200 metaslabs per vdev. | |
1364 | */ | |
1365 | vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); | |
1366 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
1367 | } |
1368 | ||
1369 | void | |
1370 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1371 | { | |
1372 | ASSERT(vd == vd->vdev_top); | |
1373 | ASSERT(ISP2(flags)); | |
1374 | ||
1375 | if (flags & VDD_METASLAB) | |
1376 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1377 | ||
1378 | if (flags & VDD_DTL) | |
1379 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1380 | ||
1381 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1382 | } | |
1383 | ||
fb5f0bc8 BB |
1384 | /* |
1385 | * DTLs. | |
1386 | * | |
1387 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
1388 | * the vdev has less than perfect replication. There are three kinds of DTL: | |
1389 | * | |
1390 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1391 | * | |
1392 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1393 | * | |
1394 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1395 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1396 | * txgs that was scrubbed. | |
1397 | * | |
1398 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1399 | * persistent errors or just some device being offline. | |
1400 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1401 | * maintained; it's only computed when needed, typically to | |
1402 | * determine whether a device can be detached. | |
1403 | * | |
1404 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1405 | * either has the data or it doesn't. | |
1406 | * | |
1407 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1408 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1409 | * if any child is less than fully replicated, then so is its parent. | |
1410 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1411 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1412 | * those are the txgs we don't have enough replication to read. For example, | |
1413 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1414 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1415 | * two child DTL_MISSING maps. | |
1416 | * | |
1417 | * It should be clear from the above that to compute the DTLs and outage maps | |
1418 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1419 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1420 | * a configuration change, we generate all other DTLs from first principles. | |
1421 | */ | |
34dc7c2f | 1422 | void |
fb5f0bc8 | 1423 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1424 | { |
fb5f0bc8 BB |
1425 | space_map_t *sm = &vd->vdev_dtl[t]; |
1426 | ||
1427 | ASSERT(t < DTL_TYPES); | |
1428 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
1429 | ||
34dc7c2f BB |
1430 | mutex_enter(sm->sm_lock); |
1431 | if (!space_map_contains(sm, txg, size)) | |
1432 | space_map_add(sm, txg, size); | |
1433 | mutex_exit(sm->sm_lock); | |
1434 | } | |
1435 | ||
fb5f0bc8 BB |
1436 | boolean_t |
1437 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1438 | { |
fb5f0bc8 BB |
1439 | space_map_t *sm = &vd->vdev_dtl[t]; |
1440 | boolean_t dirty = B_FALSE; | |
34dc7c2f | 1441 | |
fb5f0bc8 BB |
1442 | ASSERT(t < DTL_TYPES); |
1443 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f BB |
1444 | |
1445 | mutex_enter(sm->sm_lock); | |
fb5f0bc8 BB |
1446 | if (sm->sm_space != 0) |
1447 | dirty = space_map_contains(sm, txg, size); | |
34dc7c2f BB |
1448 | mutex_exit(sm->sm_lock); |
1449 | ||
1450 | return (dirty); | |
1451 | } | |
1452 | ||
fb5f0bc8 BB |
1453 | boolean_t |
1454 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1455 | { | |
1456 | space_map_t *sm = &vd->vdev_dtl[t]; | |
1457 | boolean_t empty; | |
1458 | ||
1459 | mutex_enter(sm->sm_lock); | |
1460 | empty = (sm->sm_space == 0); | |
1461 | mutex_exit(sm->sm_lock); | |
1462 | ||
1463 | return (empty); | |
1464 | } | |
1465 | ||
34dc7c2f BB |
1466 | /* |
1467 | * Reassess DTLs after a config change or scrub completion. | |
1468 | */ | |
1469 | void | |
1470 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1471 | { | |
1472 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1473 | avl_tree_t reftree; |
1474 | int minref; | |
34dc7c2f | 1475 | |
fb5f0bc8 | 1476 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1477 | |
fb5f0bc8 BB |
1478 | for (int c = 0; c < vd->vdev_children; c++) |
1479 | vdev_dtl_reassess(vd->vdev_child[c], txg, | |
1480 | scrub_txg, scrub_done); | |
1481 | ||
1482 | if (vd == spa->spa_root_vdev) | |
1483 | return; | |
1484 | ||
1485 | if (vd->vdev_ops->vdev_op_leaf) { | |
34dc7c2f | 1486 | mutex_enter(&vd->vdev_dtl_lock); |
b128c09f BB |
1487 | if (scrub_txg != 0 && |
1488 | (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { | |
1489 | /* XXX should check scrub_done? */ | |
1490 | /* | |
1491 | * We completed a scrub up to scrub_txg. If we | |
1492 | * did it without rebooting, then the scrub dtl | |
1493 | * will be valid, so excise the old region and | |
1494 | * fold in the scrub dtl. Otherwise, leave the | |
1495 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1496 | * |
1497 | * There's little trick here: to excise the beginning | |
1498 | * of the DTL_MISSING map, we put it into a reference | |
1499 | * tree and then add a segment with refcnt -1 that | |
1500 | * covers the range [0, scrub_txg). This means | |
1501 | * that each txg in that range has refcnt -1 or 0. | |
1502 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1503 | * entries in the range [0, scrub_txg) will have a | |
1504 | * positive refcnt -- either 1 or 2. We then convert | |
1505 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1506 | */ |
fb5f0bc8 BB |
1507 | space_map_ref_create(&reftree); |
1508 | space_map_ref_add_map(&reftree, | |
1509 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1510 | space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); | |
1511 | space_map_ref_add_map(&reftree, | |
1512 | &vd->vdev_dtl[DTL_SCRUB], 2); | |
1513 | space_map_ref_generate_map(&reftree, | |
1514 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1515 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1516 | } |
fb5f0bc8 BB |
1517 | space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1518 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1519 | space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1520 | if (scrub_done) |
fb5f0bc8 BB |
1521 | space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1522 | space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
1523 | if (!vdev_readable(vd)) | |
1524 | space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); | |
1525 | else | |
1526 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1527 | space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); | |
34dc7c2f | 1528 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1529 | |
34dc7c2f BB |
1530 | if (txg != 0) |
1531 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1532 | return; | |
1533 | } | |
1534 | ||
34dc7c2f | 1535 | mutex_enter(&vd->vdev_dtl_lock); |
fb5f0bc8 BB |
1536 | for (int t = 0; t < DTL_TYPES; t++) { |
1537 | if (t == DTL_SCRUB) | |
1538 | continue; /* leaf vdevs only */ | |
1539 | if (t == DTL_PARTIAL) | |
1540 | minref = 1; /* i.e. non-zero */ | |
1541 | else if (vd->vdev_nparity != 0) | |
1542 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1543 | else | |
1544 | minref = vd->vdev_children; /* any kind of mirror */ | |
1545 | space_map_ref_create(&reftree); | |
1546 | for (int c = 0; c < vd->vdev_children; c++) { | |
1547 | vdev_t *cvd = vd->vdev_child[c]; | |
1548 | mutex_enter(&cvd->vdev_dtl_lock); | |
1549 | space_map_ref_add_map(&reftree, &cvd->vdev_dtl[t], 1); | |
1550 | mutex_exit(&cvd->vdev_dtl_lock); | |
1551 | } | |
1552 | space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); | |
1553 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1554 | } |
fb5f0bc8 | 1555 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1556 | } |
1557 | ||
1558 | static int | |
1559 | vdev_dtl_load(vdev_t *vd) | |
1560 | { | |
1561 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1562 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
34dc7c2f BB |
1563 | objset_t *mos = spa->spa_meta_objset; |
1564 | dmu_buf_t *db; | |
1565 | int error; | |
1566 | ||
1567 | ASSERT(vd->vdev_children == 0); | |
1568 | ||
1569 | if (smo->smo_object == 0) | |
1570 | return (0); | |
1571 | ||
1572 | if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) | |
1573 | return (error); | |
1574 | ||
1575 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1576 | bcopy(db->db_data, smo, sizeof (*smo)); | |
1577 | dmu_buf_rele(db, FTAG); | |
1578 | ||
1579 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1580 | error = space_map_load(&vd->vdev_dtl[DTL_MISSING], |
1581 | NULL, SM_ALLOC, smo, mos); | |
34dc7c2f BB |
1582 | mutex_exit(&vd->vdev_dtl_lock); |
1583 | ||
1584 | return (error); | |
1585 | } | |
1586 | ||
1587 | void | |
1588 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1589 | { | |
1590 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1591 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
1592 | space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; | |
34dc7c2f BB |
1593 | objset_t *mos = spa->spa_meta_objset; |
1594 | space_map_t smsync; | |
1595 | kmutex_t smlock; | |
1596 | dmu_buf_t *db; | |
1597 | dmu_tx_t *tx; | |
1598 | ||
34dc7c2f BB |
1599 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1600 | ||
1601 | if (vd->vdev_detached) { | |
1602 | if (smo->smo_object != 0) { | |
1603 | int err = dmu_object_free(mos, smo->smo_object, tx); | |
1604 | ASSERT3U(err, ==, 0); | |
1605 | smo->smo_object = 0; | |
1606 | } | |
1607 | dmu_tx_commit(tx); | |
34dc7c2f BB |
1608 | return; |
1609 | } | |
1610 | ||
1611 | if (smo->smo_object == 0) { | |
1612 | ASSERT(smo->smo_objsize == 0); | |
1613 | ASSERT(smo->smo_alloc == 0); | |
1614 | smo->smo_object = dmu_object_alloc(mos, | |
1615 | DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, | |
1616 | DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); | |
1617 | ASSERT(smo->smo_object != 0); | |
1618 | vdev_config_dirty(vd->vdev_top); | |
1619 | } | |
1620 | ||
1621 | mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); | |
1622 | ||
1623 | space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, | |
1624 | &smlock); | |
1625 | ||
1626 | mutex_enter(&smlock); | |
1627 | ||
1628 | mutex_enter(&vd->vdev_dtl_lock); | |
1629 | space_map_walk(sm, space_map_add, &smsync); | |
1630 | mutex_exit(&vd->vdev_dtl_lock); | |
1631 | ||
1632 | space_map_truncate(smo, mos, tx); | |
1633 | space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); | |
1634 | ||
1635 | space_map_destroy(&smsync); | |
1636 | ||
1637 | mutex_exit(&smlock); | |
1638 | mutex_destroy(&smlock); | |
1639 | ||
1640 | VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); | |
1641 | dmu_buf_will_dirty(db, tx); | |
1642 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1643 | bcopy(smo, db->db_data, sizeof (*smo)); | |
1644 | dmu_buf_rele(db, FTAG); | |
1645 | ||
1646 | dmu_tx_commit(tx); | |
1647 | } | |
1648 | ||
fb5f0bc8 BB |
1649 | /* |
1650 | * Determine whether the specified vdev can be offlined/detached/removed | |
1651 | * without losing data. | |
1652 | */ | |
1653 | boolean_t | |
1654 | vdev_dtl_required(vdev_t *vd) | |
1655 | { | |
1656 | spa_t *spa = vd->vdev_spa; | |
1657 | vdev_t *tvd = vd->vdev_top; | |
1658 | uint8_t cant_read = vd->vdev_cant_read; | |
1659 | boolean_t required; | |
1660 | ||
1661 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1662 | ||
1663 | if (vd == spa->spa_root_vdev || vd == tvd) | |
1664 | return (B_TRUE); | |
1665 | ||
1666 | /* | |
1667 | * Temporarily mark the device as unreadable, and then determine | |
1668 | * whether this results in any DTL outages in the top-level vdev. | |
1669 | * If not, we can safely offline/detach/remove the device. | |
1670 | */ | |
1671 | vd->vdev_cant_read = B_TRUE; | |
1672 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1673 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
1674 | vd->vdev_cant_read = cant_read; | |
1675 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1676 | ||
1677 | return (required); | |
1678 | } | |
1679 | ||
b128c09f BB |
1680 | /* |
1681 | * Determine if resilver is needed, and if so the txg range. | |
1682 | */ | |
1683 | boolean_t | |
1684 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
1685 | { | |
1686 | boolean_t needed = B_FALSE; | |
1687 | uint64_t thismin = UINT64_MAX; | |
1688 | uint64_t thismax = 0; | |
1689 | ||
1690 | if (vd->vdev_children == 0) { | |
1691 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1692 | if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && |
1693 | vdev_writeable(vd)) { | |
b128c09f BB |
1694 | space_seg_t *ss; |
1695 | ||
fb5f0bc8 | 1696 | ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f | 1697 | thismin = ss->ss_start - 1; |
fb5f0bc8 | 1698 | ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f BB |
1699 | thismax = ss->ss_end; |
1700 | needed = B_TRUE; | |
1701 | } | |
1702 | mutex_exit(&vd->vdev_dtl_lock); | |
1703 | } else { | |
fb5f0bc8 | 1704 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
1705 | vdev_t *cvd = vd->vdev_child[c]; |
1706 | uint64_t cmin, cmax; | |
1707 | ||
1708 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
1709 | thismin = MIN(thismin, cmin); | |
1710 | thismax = MAX(thismax, cmax); | |
1711 | needed = B_TRUE; | |
1712 | } | |
1713 | } | |
1714 | } | |
1715 | ||
1716 | if (needed && minp) { | |
1717 | *minp = thismin; | |
1718 | *maxp = thismax; | |
1719 | } | |
1720 | return (needed); | |
1721 | } | |
1722 | ||
34dc7c2f BB |
1723 | void |
1724 | vdev_load(vdev_t *vd) | |
1725 | { | |
34dc7c2f BB |
1726 | /* |
1727 | * Recursively load all children. | |
1728 | */ | |
fb5f0bc8 | 1729 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1730 | vdev_load(vd->vdev_child[c]); |
1731 | ||
1732 | /* | |
1733 | * If this is a top-level vdev, initialize its metaslabs. | |
1734 | */ | |
1735 | if (vd == vd->vdev_top && | |
1736 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || | |
1737 | vdev_metaslab_init(vd, 0) != 0)) | |
1738 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1739 | VDEV_AUX_CORRUPT_DATA); | |
1740 | ||
1741 | /* | |
1742 | * If this is a leaf vdev, load its DTL. | |
1743 | */ | |
1744 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
1745 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1746 | VDEV_AUX_CORRUPT_DATA); | |
1747 | } | |
1748 | ||
1749 | /* | |
1750 | * The special vdev case is used for hot spares and l2cache devices. Its | |
1751 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
1752 | * we make sure that we can open the underlying device, then try to read the | |
1753 | * label, and make sure that the label is sane and that it hasn't been | |
1754 | * repurposed to another pool. | |
1755 | */ | |
1756 | int | |
1757 | vdev_validate_aux(vdev_t *vd) | |
1758 | { | |
1759 | nvlist_t *label; | |
1760 | uint64_t guid, version; | |
1761 | uint64_t state; | |
1762 | ||
b128c09f BB |
1763 | if (!vdev_readable(vd)) |
1764 | return (0); | |
1765 | ||
34dc7c2f BB |
1766 | if ((label = vdev_label_read_config(vd)) == NULL) { |
1767 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1768 | VDEV_AUX_CORRUPT_DATA); | |
1769 | return (-1); | |
1770 | } | |
1771 | ||
1772 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
1773 | version > SPA_VERSION || | |
1774 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || | |
1775 | guid != vd->vdev_guid || | |
1776 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
1777 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1778 | VDEV_AUX_CORRUPT_DATA); | |
1779 | nvlist_free(label); | |
1780 | return (-1); | |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * We don't actually check the pool state here. If it's in fact in | |
1785 | * use by another pool, we update this fact on the fly when requested. | |
1786 | */ | |
1787 | nvlist_free(label); | |
1788 | return (0); | |
1789 | } | |
1790 | ||
1791 | void | |
1792 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
1793 | { | |
1794 | metaslab_t *msp; | |
1795 | ||
34dc7c2f BB |
1796 | while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) |
1797 | metaslab_sync_done(msp, txg); | |
1798 | } | |
1799 | ||
1800 | void | |
1801 | vdev_sync(vdev_t *vd, uint64_t txg) | |
1802 | { | |
1803 | spa_t *spa = vd->vdev_spa; | |
1804 | vdev_t *lvd; | |
1805 | metaslab_t *msp; | |
1806 | dmu_tx_t *tx; | |
1807 | ||
34dc7c2f BB |
1808 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
1809 | ASSERT(vd == vd->vdev_top); | |
1810 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
1811 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
1812 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
1813 | ASSERT(vd->vdev_ms_array != 0); | |
1814 | vdev_config_dirty(vd); | |
1815 | dmu_tx_commit(tx); | |
1816 | } | |
1817 | ||
1818 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { | |
1819 | metaslab_sync(msp, txg); | |
1820 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
1821 | } | |
1822 | ||
1823 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
1824 | vdev_dtl_sync(lvd, txg); | |
1825 | ||
1826 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
1827 | } | |
1828 | ||
1829 | uint64_t | |
1830 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
1831 | { | |
1832 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
1833 | } | |
1834 | ||
34dc7c2f BB |
1835 | /* |
1836 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
1837 | * not be opened, and no I/O is attempted. | |
1838 | */ | |
1839 | int | |
1840 | vdev_fault(spa_t *spa, uint64_t guid) | |
1841 | { | |
b128c09f | 1842 | vdev_t *vd; |
34dc7c2f | 1843 | |
b128c09f | 1844 | spa_vdev_state_enter(spa); |
34dc7c2f | 1845 | |
b128c09f BB |
1846 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
1847 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 1848 | |
34dc7c2f | 1849 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 1850 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
1851 | |
1852 | /* | |
1853 | * Faulted state takes precedence over degraded. | |
1854 | */ | |
1855 | vd->vdev_faulted = 1ULL; | |
1856 | vd->vdev_degraded = 0ULL; | |
b128c09f | 1857 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED); |
34dc7c2f BB |
1858 | |
1859 | /* | |
b128c09f | 1860 | * If marking the vdev as faulted cause the top-level vdev to become |
34dc7c2f BB |
1861 | * unavailable, then back off and simply mark the vdev as degraded |
1862 | * instead. | |
1863 | */ | |
b128c09f | 1864 | if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) { |
34dc7c2f BB |
1865 | vd->vdev_degraded = 1ULL; |
1866 | vd->vdev_faulted = 0ULL; | |
1867 | ||
1868 | /* | |
1869 | * If we reopen the device and it's not dead, only then do we | |
1870 | * mark it degraded. | |
1871 | */ | |
1872 | vdev_reopen(vd); | |
1873 | ||
1874 | if (vdev_readable(vd)) { | |
1875 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
1876 | VDEV_AUX_ERR_EXCEEDED); | |
1877 | } | |
1878 | } | |
1879 | ||
b128c09f | 1880 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
1881 | } |
1882 | ||
1883 | /* | |
1884 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
1885 | * user that something is wrong. The vdev continues to operate as normal as far | |
1886 | * as I/O is concerned. | |
1887 | */ | |
1888 | int | |
1889 | vdev_degrade(spa_t *spa, uint64_t guid) | |
1890 | { | |
b128c09f | 1891 | vdev_t *vd; |
34dc7c2f | 1892 | |
b128c09f | 1893 | spa_vdev_state_enter(spa); |
34dc7c2f | 1894 | |
b128c09f BB |
1895 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
1896 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 1897 | |
34dc7c2f | 1898 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 1899 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
1900 | |
1901 | /* | |
1902 | * If the vdev is already faulted, then don't do anything. | |
1903 | */ | |
b128c09f BB |
1904 | if (vd->vdev_faulted || vd->vdev_degraded) |
1905 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
1906 | |
1907 | vd->vdev_degraded = 1ULL; | |
1908 | if (!vdev_is_dead(vd)) | |
1909 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
1910 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f | 1911 | |
b128c09f | 1912 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
1913 | } |
1914 | ||
1915 | /* | |
1916 | * Online the given vdev. If 'unspare' is set, it implies two things. First, | |
1917 | * any attached spare device should be detached when the device finishes | |
1918 | * resilvering. Second, the online should be treated like a 'test' online case, | |
1919 | * so no FMA events are generated if the device fails to open. | |
1920 | */ | |
1921 | int | |
b128c09f | 1922 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 1923 | { |
9babb374 | 1924 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 1925 | |
b128c09f | 1926 | spa_vdev_state_enter(spa); |
34dc7c2f | 1927 | |
b128c09f BB |
1928 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
1929 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
1930 | |
1931 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 1932 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 1933 | |
9babb374 | 1934 | tvd = vd->vdev_top; |
34dc7c2f BB |
1935 | vd->vdev_offline = B_FALSE; |
1936 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
1937 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
1938 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
1939 | |
1940 | /* XXX - L2ARC 1.0 does not support expansion */ | |
1941 | if (!vd->vdev_aux) { | |
1942 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
1943 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
1944 | } | |
1945 | ||
1946 | vdev_reopen(tvd); | |
34dc7c2f BB |
1947 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
1948 | ||
9babb374 BB |
1949 | if (!vd->vdev_aux) { |
1950 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
1951 | pvd->vdev_expanding = B_FALSE; | |
1952 | } | |
1953 | ||
34dc7c2f BB |
1954 | if (newstate) |
1955 | *newstate = vd->vdev_state; | |
1956 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
1957 | !vdev_is_dead(vd) && vd->vdev_parent && | |
1958 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
1959 | vd->vdev_parent->vdev_child[0] == vd) | |
1960 | vd->vdev_unspare = B_TRUE; | |
1961 | ||
9babb374 BB |
1962 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
1963 | ||
1964 | /* XXX - L2ARC 1.0 does not support expansion */ | |
1965 | if (vd->vdev_aux) | |
1966 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
1967 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
1968 | } | |
fb5f0bc8 | 1969 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
1970 | } |
1971 | ||
1972 | int | |
1973 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
1974 | { | |
9babb374 BB |
1975 | vdev_t *vd, *tvd; |
1976 | int error; | |
34dc7c2f | 1977 | |
b128c09f | 1978 | spa_vdev_state_enter(spa); |
34dc7c2f | 1979 | |
b128c09f BB |
1980 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
1981 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
1982 | |
1983 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 1984 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 1985 | |
9babb374 BB |
1986 | tvd = vd->vdev_top; |
1987 | ||
34dc7c2f BB |
1988 | /* |
1989 | * If the device isn't already offline, try to offline it. | |
1990 | */ | |
1991 | if (!vd->vdev_offline) { | |
1992 | /* | |
fb5f0bc8 | 1993 | * If this device has the only valid copy of some data, |
9babb374 BB |
1994 | * don't allow it to be offlined. Log devices are always |
1995 | * expendable. | |
34dc7c2f | 1996 | */ |
9babb374 BB |
1997 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
1998 | vdev_dtl_required(vd)) | |
b128c09f | 1999 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f BB |
2000 | |
2001 | /* | |
2002 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2003 | * If the top-level vdev is a log device then just offline |
2004 | * it. Otherwise, if this action results in the top-level | |
2005 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2006 | */ |
2007 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2008 | vdev_reopen(tvd); |
2009 | ||
2010 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2011 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2012 | vd->vdev_offline = B_FALSE; |
9babb374 | 2013 | vdev_reopen(tvd); |
b128c09f | 2014 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f BB |
2015 | } |
2016 | } | |
2017 | ||
b128c09f | 2018 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2019 | |
9babb374 BB |
2020 | if (!tvd->vdev_islog || !vdev_is_dead(tvd)) |
2021 | return (spa_vdev_state_exit(spa, vd, 0)); | |
2022 | ||
2023 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2024 | ||
2025 | error = dmu_objset_find(spa_name(spa), zil_vdev_offline, | |
2026 | NULL, DS_FIND_CHILDREN); | |
2027 | if (error) { | |
2028 | (void) vdev_online(spa, guid, 0, NULL); | |
2029 | return (error); | |
2030 | } | |
2031 | /* | |
2032 | * If we successfully offlined the log device then we need to | |
2033 | * sync out the current txg so that the "stubby" block can be | |
2034 | * removed by zil_sync(). | |
2035 | */ | |
2036 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
2037 | return (0); | |
34dc7c2f BB |
2038 | } |
2039 | ||
2040 | /* | |
2041 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2042 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2043 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2044 | */ |
2045 | void | |
b128c09f | 2046 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2047 | { |
b128c09f BB |
2048 | vdev_t *rvd = spa->spa_root_vdev; |
2049 | ||
2050 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2051 | |
2052 | if (vd == NULL) | |
b128c09f | 2053 | vd = rvd; |
34dc7c2f BB |
2054 | |
2055 | vd->vdev_stat.vs_read_errors = 0; | |
2056 | vd->vdev_stat.vs_write_errors = 0; | |
2057 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2058 | |
b128c09f BB |
2059 | for (int c = 0; c < vd->vdev_children; c++) |
2060 | vdev_clear(spa, vd->vdev_child[c]); | |
34dc7c2f BB |
2061 | |
2062 | /* | |
b128c09f BB |
2063 | * If we're in the FAULTED state or have experienced failed I/O, then |
2064 | * clear the persistent state and attempt to reopen the device. We | |
2065 | * also mark the vdev config dirty, so that the new faulted state is | |
2066 | * written out to disk. | |
34dc7c2f | 2067 | */ |
b128c09f BB |
2068 | if (vd->vdev_faulted || vd->vdev_degraded || |
2069 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2070 | ||
34dc7c2f | 2071 | vd->vdev_faulted = vd->vdev_degraded = 0; |
b128c09f BB |
2072 | vd->vdev_cant_read = B_FALSE; |
2073 | vd->vdev_cant_write = B_FALSE; | |
2074 | ||
34dc7c2f | 2075 | vdev_reopen(vd); |
34dc7c2f | 2076 | |
b128c09f BB |
2077 | if (vd != rvd) |
2078 | vdev_state_dirty(vd->vdev_top); | |
2079 | ||
2080 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2081 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2082 | ||
2083 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); | |
2084 | } | |
2085 | } | |
2086 | ||
b128c09f BB |
2087 | boolean_t |
2088 | vdev_is_dead(vdev_t *vd) | |
2089 | { | |
2090 | return (vd->vdev_state < VDEV_STATE_DEGRADED); | |
2091 | } | |
2092 | ||
2093 | boolean_t | |
34dc7c2f BB |
2094 | vdev_readable(vdev_t *vd) |
2095 | { | |
b128c09f | 2096 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2097 | } |
2098 | ||
b128c09f | 2099 | boolean_t |
34dc7c2f BB |
2100 | vdev_writeable(vdev_t *vd) |
2101 | { | |
b128c09f | 2102 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2103 | } |
2104 | ||
b128c09f BB |
2105 | boolean_t |
2106 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2107 | { |
fb5f0bc8 BB |
2108 | uint64_t state = vd->vdev_state; |
2109 | ||
b128c09f | 2110 | /* |
fb5f0bc8 | 2111 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2112 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2113 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2114 | * the proper locks. Note that we have to get the vdev state |
2115 | * in a local variable because although it changes atomically, | |
2116 | * we're asking two separate questions about it. | |
b128c09f | 2117 | */ |
fb5f0bc8 | 2118 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
b128c09f | 2119 | !vd->vdev_cant_write); |
34dc7c2f BB |
2120 | } |
2121 | ||
b128c09f BB |
2122 | boolean_t |
2123 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2124 | { |
b128c09f | 2125 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2126 | |
b128c09f BB |
2127 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2128 | return (B_FALSE); | |
34dc7c2f | 2129 | |
b128c09f BB |
2130 | if (zio->io_type == ZIO_TYPE_READ) |
2131 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2132 | |
b128c09f BB |
2133 | if (zio->io_type == ZIO_TYPE_WRITE) |
2134 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2135 | |
b128c09f | 2136 | return (B_TRUE); |
34dc7c2f BB |
2137 | } |
2138 | ||
2139 | /* | |
2140 | * Get statistics for the given vdev. | |
2141 | */ | |
2142 | void | |
2143 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2144 | { | |
2145 | vdev_t *rvd = vd->vdev_spa->spa_root_vdev; | |
34dc7c2f BB |
2146 | |
2147 | mutex_enter(&vd->vdev_stat_lock); | |
2148 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
b128c09f | 2149 | vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; |
34dc7c2f BB |
2150 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; |
2151 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2152 | vs->vs_rsize = vdev_get_min_asize(vd); |
2153 | if (vd->vdev_ops->vdev_op_leaf) | |
2154 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
34dc7c2f BB |
2155 | mutex_exit(&vd->vdev_stat_lock); |
2156 | ||
2157 | /* | |
2158 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2159 | * over all top-level vdevs (i.e. the direct children of the root). | |
2160 | */ | |
2161 | if (vd == rvd) { | |
b128c09f | 2162 | for (int c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2163 | vdev_t *cvd = rvd->vdev_child[c]; |
2164 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2165 | ||
2166 | mutex_enter(&vd->vdev_stat_lock); | |
b128c09f | 2167 | for (int t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2168 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2169 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2170 | } | |
34dc7c2f | 2171 | vs->vs_scrub_examined += cvs->vs_scrub_examined; |
34dc7c2f BB |
2172 | mutex_exit(&vd->vdev_stat_lock); |
2173 | } | |
2174 | } | |
2175 | } | |
2176 | ||
2177 | void | |
2178 | vdev_clear_stats(vdev_t *vd) | |
2179 | { | |
2180 | mutex_enter(&vd->vdev_stat_lock); | |
2181 | vd->vdev_stat.vs_space = 0; | |
2182 | vd->vdev_stat.vs_dspace = 0; | |
2183 | vd->vdev_stat.vs_alloc = 0; | |
2184 | mutex_exit(&vd->vdev_stat_lock); | |
2185 | } | |
2186 | ||
2187 | void | |
b128c09f | 2188 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2189 | { |
fb5f0bc8 BB |
2190 | spa_t *spa = zio->io_spa; |
2191 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2192 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2193 | vdev_t *pvd; |
2194 | uint64_t txg = zio->io_txg; | |
2195 | vdev_stat_t *vs = &vd->vdev_stat; | |
2196 | zio_type_t type = zio->io_type; | |
2197 | int flags = zio->io_flags; | |
2198 | ||
b128c09f BB |
2199 | /* |
2200 | * If this i/o is a gang leader, it didn't do any actual work. | |
2201 | */ | |
2202 | if (zio->io_gang_tree) | |
2203 | return; | |
2204 | ||
34dc7c2f | 2205 | if (zio->io_error == 0) { |
b128c09f BB |
2206 | /* |
2207 | * If this is a root i/o, don't count it -- we've already | |
2208 | * counted the top-level vdevs, and vdev_get_stats() will | |
2209 | * aggregate them when asked. This reduces contention on | |
2210 | * the root vdev_stat_lock and implicitly handles blocks | |
2211 | * that compress away to holes, for which there is no i/o. | |
2212 | * (Holes never create vdev children, so all the counters | |
2213 | * remain zero, which is what we want.) | |
2214 | * | |
2215 | * Note: this only applies to successful i/o (io_error == 0) | |
2216 | * because unlike i/o counts, errors are not additive. | |
2217 | * When reading a ditto block, for example, failure of | |
2218 | * one top-level vdev does not imply a root-level error. | |
2219 | */ | |
2220 | if (vd == rvd) | |
2221 | return; | |
2222 | ||
2223 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2224 | |
2225 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2226 | return; | |
2227 | ||
2228 | mutex_enter(&vd->vdev_stat_lock); | |
2229 | ||
b128c09f | 2230 | if (flags & ZIO_FLAG_IO_REPAIR) { |
34dc7c2f | 2231 | if (flags & ZIO_FLAG_SCRUB_THREAD) |
b128c09f | 2232 | vs->vs_scrub_repaired += psize; |
fb5f0bc8 | 2233 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2234 | vs->vs_self_healed += psize; |
34dc7c2f | 2235 | } |
fb5f0bc8 BB |
2236 | |
2237 | vs->vs_ops[type]++; | |
2238 | vs->vs_bytes[type] += psize; | |
2239 | ||
2240 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2241 | return; |
2242 | } | |
2243 | ||
2244 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2245 | return; | |
2246 | ||
9babb374 BB |
2247 | /* |
2248 | * If this is an I/O error that is going to be retried, then ignore the | |
2249 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2250 | * hard errors, when in reality they can happen for any number of | |
2251 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2252 | */ | |
2253 | if (zio->io_error == EIO && | |
2254 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2255 | return; | |
2256 | ||
b128c09f | 2257 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2258 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2259 | if (zio->io_error == ECKSUM) |
2260 | vs->vs_checksum_errors++; | |
2261 | else | |
2262 | vs->vs_read_errors++; | |
34dc7c2f | 2263 | } |
9babb374 | 2264 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2265 | vs->vs_write_errors++; |
2266 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2267 | |
fb5f0bc8 BB |
2268 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2269 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
2270 | (flags & ZIO_FLAG_SCRUB_THREAD))) { | |
2271 | /* | |
2272 | * This is either a normal write (not a repair), or it's a | |
2273 | * repair induced by the scrub thread. In the normal case, | |
2274 | * we commit the DTL change in the same txg as the block | |
2275 | * was born. In the scrub-induced repair case, we know that | |
2276 | * scrubs run in first-pass syncing context, so we commit | |
2277 | * the DTL change in spa->spa_syncing_txg. | |
2278 | * | |
2279 | * We currently do not make DTL entries for failed spontaneous | |
2280 | * self-healing writes triggered by normal (non-scrubbing) | |
2281 | * reads, because we have no transactional context in which to | |
2282 | * do so -- and it's not clear that it'd be desirable anyway. | |
2283 | */ | |
2284 | if (vd->vdev_ops->vdev_op_leaf) { | |
2285 | uint64_t commit_txg = txg; | |
2286 | if (flags & ZIO_FLAG_SCRUB_THREAD) { | |
2287 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2288 | ASSERT(spa_sync_pass(spa) == 1); | |
2289 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
2290 | commit_txg = spa->spa_syncing_txg; | |
2291 | } | |
2292 | ASSERT(commit_txg >= spa->spa_syncing_txg); | |
2293 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) | |
34dc7c2f | 2294 | return; |
fb5f0bc8 BB |
2295 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2296 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2297 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2298 | } |
fb5f0bc8 BB |
2299 | if (vd != rvd) |
2300 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2301 | } |
2302 | } | |
2303 | ||
2304 | void | |
2305 | vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) | |
2306 | { | |
34dc7c2f BB |
2307 | vdev_stat_t *vs = &vd->vdev_stat; |
2308 | ||
9babb374 | 2309 | for (int c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2310 | vdev_scrub_stat_update(vd->vdev_child[c], type, complete); |
2311 | ||
2312 | mutex_enter(&vd->vdev_stat_lock); | |
2313 | ||
2314 | if (type == POOL_SCRUB_NONE) { | |
2315 | /* | |
2316 | * Update completion and end time. Leave everything else alone | |
2317 | * so we can report what happened during the previous scrub. | |
2318 | */ | |
2319 | vs->vs_scrub_complete = complete; | |
2320 | vs->vs_scrub_end = gethrestime_sec(); | |
2321 | } else { | |
2322 | vs->vs_scrub_type = type; | |
2323 | vs->vs_scrub_complete = 0; | |
2324 | vs->vs_scrub_examined = 0; | |
2325 | vs->vs_scrub_repaired = 0; | |
34dc7c2f BB |
2326 | vs->vs_scrub_start = gethrestime_sec(); |
2327 | vs->vs_scrub_end = 0; | |
2328 | } | |
2329 | ||
2330 | mutex_exit(&vd->vdev_stat_lock); | |
2331 | } | |
2332 | ||
2333 | /* | |
2334 | * Update the in-core space usage stats for this vdev and the root vdev. | |
2335 | */ | |
2336 | void | |
2337 | vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta, | |
2338 | boolean_t update_root) | |
2339 | { | |
2340 | int64_t dspace_delta = space_delta; | |
2341 | spa_t *spa = vd->vdev_spa; | |
2342 | vdev_t *rvd = spa->spa_root_vdev; | |
2343 | ||
2344 | ASSERT(vd == vd->vdev_top); | |
2345 | ||
2346 | /* | |
2347 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2348 | * factor. We must calculate this here and not at the root vdev | |
2349 | * because the root vdev's psize-to-asize is simply the max of its | |
2350 | * childrens', thus not accurate enough for us. | |
2351 | */ | |
2352 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2353 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2354 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2355 | vd->vdev_deflate_ratio; | |
2356 | ||
2357 | mutex_enter(&vd->vdev_stat_lock); | |
2358 | vd->vdev_stat.vs_space += space_delta; | |
2359 | vd->vdev_stat.vs_alloc += alloc_delta; | |
2360 | vd->vdev_stat.vs_dspace += dspace_delta; | |
2361 | mutex_exit(&vd->vdev_stat_lock); | |
2362 | ||
2363 | if (update_root) { | |
2364 | ASSERT(rvd == vd->vdev_parent); | |
2365 | ASSERT(vd->vdev_ms_count != 0); | |
2366 | ||
2367 | /* | |
2368 | * Don't count non-normal (e.g. intent log) space as part of | |
2369 | * the pool's capacity. | |
2370 | */ | |
2371 | if (vd->vdev_mg->mg_class != spa->spa_normal_class) | |
2372 | return; | |
2373 | ||
2374 | mutex_enter(&rvd->vdev_stat_lock); | |
2375 | rvd->vdev_stat.vs_space += space_delta; | |
2376 | rvd->vdev_stat.vs_alloc += alloc_delta; | |
2377 | rvd->vdev_stat.vs_dspace += dspace_delta; | |
2378 | mutex_exit(&rvd->vdev_stat_lock); | |
2379 | } | |
2380 | } | |
2381 | ||
2382 | /* | |
2383 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2384 | * so that it will be written out next time the vdev configuration is synced. | |
2385 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2386 | */ | |
2387 | void | |
2388 | vdev_config_dirty(vdev_t *vd) | |
2389 | { | |
2390 | spa_t *spa = vd->vdev_spa; | |
2391 | vdev_t *rvd = spa->spa_root_vdev; | |
2392 | int c; | |
2393 | ||
2394 | /* | |
9babb374 BB |
2395 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2396 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2397 | */ |
2398 | if (vd->vdev_aux != NULL) { | |
2399 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2400 | nvlist_t **aux; | |
2401 | uint_t naux; | |
2402 | ||
2403 | for (c = 0; c < sav->sav_count; c++) { | |
2404 | if (sav->sav_vdevs[c] == vd) | |
2405 | break; | |
2406 | } | |
2407 | ||
2408 | if (c == sav->sav_count) { | |
2409 | /* | |
2410 | * We're being removed. There's nothing more to do. | |
2411 | */ | |
2412 | ASSERT(sav->sav_sync == B_TRUE); | |
2413 | return; | |
2414 | } | |
2415 | ||
2416 | sav->sav_sync = B_TRUE; | |
2417 | ||
9babb374 BB |
2418 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2419 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2420 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2421 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2422 | } | |
b128c09f BB |
2423 | |
2424 | ASSERT(c < naux); | |
2425 | ||
2426 | /* | |
2427 | * Setting the nvlist in the middle if the array is a little | |
2428 | * sketchy, but it will work. | |
2429 | */ | |
2430 | nvlist_free(aux[c]); | |
2431 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); | |
2432 | ||
2433 | return; | |
2434 | } | |
2435 | ||
2436 | /* | |
2437 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2438 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2439 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2440 | * so this is sufficient to ensure mutual exclusion. |
2441 | */ | |
b128c09f BB |
2442 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2443 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2444 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2445 | |
2446 | if (vd == rvd) { | |
2447 | for (c = 0; c < rvd->vdev_children; c++) | |
2448 | vdev_config_dirty(rvd->vdev_child[c]); | |
2449 | } else { | |
2450 | ASSERT(vd == vd->vdev_top); | |
2451 | ||
b128c09f BB |
2452 | if (!list_link_active(&vd->vdev_config_dirty_node)) |
2453 | list_insert_head(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2454 | } |
2455 | } | |
2456 | ||
2457 | void | |
2458 | vdev_config_clean(vdev_t *vd) | |
2459 | { | |
2460 | spa_t *spa = vd->vdev_spa; | |
2461 | ||
b128c09f BB |
2462 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2463 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2464 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2465 | |
b128c09f BB |
2466 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2467 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2468 | } |
2469 | ||
b128c09f BB |
2470 | /* |
2471 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2472 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2473 | * the state changes from larger config changes because they require | |
2474 | * much less locking, and are often needed for administrative actions. | |
2475 | */ | |
2476 | void | |
2477 | vdev_state_dirty(vdev_t *vd) | |
2478 | { | |
2479 | spa_t *spa = vd->vdev_spa; | |
2480 | ||
2481 | ASSERT(vd == vd->vdev_top); | |
2482 | ||
2483 | /* | |
2484 | * The state list is protected by the SCL_STATE lock. The caller | |
2485 | * must either hold SCL_STATE as writer, or must be the sync thread | |
2486 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
2487 | * so this is sufficient to ensure mutual exclusion. | |
2488 | */ | |
2489 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2490 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2491 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2492 | ||
2493 | if (!list_link_active(&vd->vdev_state_dirty_node)) | |
2494 | list_insert_head(&spa->spa_state_dirty_list, vd); | |
2495 | } | |
2496 | ||
2497 | void | |
2498 | vdev_state_clean(vdev_t *vd) | |
2499 | { | |
2500 | spa_t *spa = vd->vdev_spa; | |
2501 | ||
2502 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2503 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2504 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2505 | ||
2506 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
2507 | list_remove(&spa->spa_state_dirty_list, vd); | |
2508 | } | |
2509 | ||
2510 | /* | |
2511 | * Propagate vdev state up from children to parent. | |
2512 | */ | |
34dc7c2f BB |
2513 | void |
2514 | vdev_propagate_state(vdev_t *vd) | |
2515 | { | |
fb5f0bc8 BB |
2516 | spa_t *spa = vd->vdev_spa; |
2517 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
2518 | int degraded = 0, faulted = 0; |
2519 | int corrupted = 0; | |
34dc7c2f BB |
2520 | vdev_t *child; |
2521 | ||
2522 | if (vd->vdev_children > 0) { | |
9babb374 | 2523 | for (int c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 2524 | child = vd->vdev_child[c]; |
b128c09f BB |
2525 | |
2526 | if (!vdev_readable(child) || | |
fb5f0bc8 | 2527 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
2528 | /* |
2529 | * Root special: if there is a top-level log | |
2530 | * device, treat the root vdev as if it were | |
2531 | * degraded. | |
2532 | */ | |
2533 | if (child->vdev_islog && vd == rvd) | |
2534 | degraded++; | |
2535 | else | |
2536 | faulted++; | |
2537 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 2538 | degraded++; |
b128c09f | 2539 | } |
34dc7c2f BB |
2540 | |
2541 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
2542 | corrupted++; | |
2543 | } | |
2544 | ||
2545 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
2546 | ||
2547 | /* | |
b128c09f | 2548 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
2549 | * opened due to corrupted metadata, then propagate the root |
2550 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
2551 | * replicas'. | |
2552 | */ | |
2553 | if (corrupted && vd == rvd && | |
2554 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
2555 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2556 | VDEV_AUX_CORRUPT_DATA); | |
2557 | } | |
2558 | ||
b128c09f | 2559 | if (vd->vdev_parent) |
34dc7c2f BB |
2560 | vdev_propagate_state(vd->vdev_parent); |
2561 | } | |
2562 | ||
2563 | /* | |
2564 | * Set a vdev's state. If this is during an open, we don't update the parent | |
2565 | * state, because we're in the process of opening children depth-first. | |
2566 | * Otherwise, we propagate the change to the parent. | |
2567 | * | |
2568 | * If this routine places a device in a faulted state, an appropriate ereport is | |
2569 | * generated. | |
2570 | */ | |
2571 | void | |
2572 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
2573 | { | |
2574 | uint64_t save_state; | |
b128c09f | 2575 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
2576 | |
2577 | if (state == vd->vdev_state) { | |
2578 | vd->vdev_stat.vs_aux = aux; | |
2579 | return; | |
2580 | } | |
2581 | ||
2582 | save_state = vd->vdev_state; | |
2583 | ||
2584 | vd->vdev_state = state; | |
2585 | vd->vdev_stat.vs_aux = aux; | |
2586 | ||
2587 | /* | |
2588 | * If we are setting the vdev state to anything but an open state, then | |
2589 | * always close the underlying device. Otherwise, we keep accessible | |
2590 | * but invalid devices open forever. We don't call vdev_close() itself, | |
2591 | * because that implies some extra checks (offline, etc) that we don't | |
2592 | * want here. This is limited to leaf devices, because otherwise | |
2593 | * closing the device will affect other children. | |
2594 | */ | |
b128c09f | 2595 | if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) |
34dc7c2f BB |
2596 | vd->vdev_ops->vdev_op_close(vd); |
2597 | ||
2598 | if (vd->vdev_removed && | |
2599 | state == VDEV_STATE_CANT_OPEN && | |
2600 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
2601 | /* | |
2602 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
2603 | * device was previously marked removed and someone attempted to | |
2604 | * reopen it. If this failed due to a nonexistent device, then | |
2605 | * keep the device in the REMOVED state. We also let this be if | |
2606 | * it is one of our special test online cases, which is only | |
2607 | * attempting to online the device and shouldn't generate an FMA | |
2608 | * fault. | |
2609 | */ | |
2610 | vd->vdev_state = VDEV_STATE_REMOVED; | |
2611 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
2612 | } else if (state == VDEV_STATE_REMOVED) { | |
2613 | /* | |
2614 | * Indicate to the ZFS DE that this device has been removed, and | |
2615 | * any recent errors should be ignored. | |
2616 | */ | |
b128c09f | 2617 | zfs_post_remove(spa, vd); |
34dc7c2f BB |
2618 | vd->vdev_removed = B_TRUE; |
2619 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
2620 | /* | |
2621 | * If we fail to open a vdev during an import, we mark it as | |
2622 | * "not available", which signifies that it was never there to | |
2623 | * begin with. Failure to open such a device is not considered | |
2624 | * an error. | |
2625 | */ | |
b128c09f | 2626 | if (spa->spa_load_state == SPA_LOAD_IMPORT && |
34dc7c2f BB |
2627 | vd->vdev_ops->vdev_op_leaf) |
2628 | vd->vdev_not_present = 1; | |
2629 | ||
2630 | /* | |
2631 | * Post the appropriate ereport. If the 'prevstate' field is | |
2632 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
2633 | * that this is part of a vdev_reopen(). In this case, we don't | |
2634 | * want to post the ereport if the device was already in the | |
2635 | * CANT_OPEN state beforehand. | |
2636 | * | |
2637 | * If the 'checkremove' flag is set, then this is an attempt to | |
2638 | * online the device in response to an insertion event. If we | |
2639 | * hit this case, then we have detected an insertion event for a | |
2640 | * faulted or offline device that wasn't in the removed state. | |
2641 | * In this scenario, we don't post an ereport because we are | |
2642 | * about to replace the device, or attempt an online with | |
2643 | * vdev_forcefault, which will generate the fault for us. | |
2644 | */ | |
2645 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
2646 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 2647 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
2648 | const char *class; |
2649 | ||
2650 | switch (aux) { | |
2651 | case VDEV_AUX_OPEN_FAILED: | |
2652 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
2653 | break; | |
2654 | case VDEV_AUX_CORRUPT_DATA: | |
2655 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
2656 | break; | |
2657 | case VDEV_AUX_NO_REPLICAS: | |
2658 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
2659 | break; | |
2660 | case VDEV_AUX_BAD_GUID_SUM: | |
2661 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
2662 | break; | |
2663 | case VDEV_AUX_TOO_SMALL: | |
2664 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
2665 | break; | |
2666 | case VDEV_AUX_BAD_LABEL: | |
2667 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
2668 | break; | |
b128c09f BB |
2669 | case VDEV_AUX_IO_FAILURE: |
2670 | class = FM_EREPORT_ZFS_IO_FAILURE; | |
2671 | break; | |
34dc7c2f BB |
2672 | default: |
2673 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
2674 | } | |
2675 | ||
b128c09f | 2676 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
2677 | } |
2678 | ||
2679 | /* Erase any notion of persistent removed state */ | |
2680 | vd->vdev_removed = B_FALSE; | |
2681 | } else { | |
2682 | vd->vdev_removed = B_FALSE; | |
2683 | } | |
2684 | ||
9babb374 BB |
2685 | if (!isopen && vd->vdev_parent) |
2686 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 2687 | } |
b128c09f BB |
2688 | |
2689 | /* | |
2690 | * Check the vdev configuration to ensure that it's capable of supporting | |
2691 | * a root pool. Currently, we do not support RAID-Z or partial configuration. | |
2692 | * In addition, only a single top-level vdev is allowed and none of the leaves | |
2693 | * can be wholedisks. | |
2694 | */ | |
2695 | boolean_t | |
2696 | vdev_is_bootable(vdev_t *vd) | |
2697 | { | |
b128c09f BB |
2698 | if (!vd->vdev_ops->vdev_op_leaf) { |
2699 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
2700 | ||
2701 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
2702 | vd->vdev_children > 1) { | |
2703 | return (B_FALSE); | |
2704 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
2705 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
2706 | return (B_FALSE); | |
2707 | } | |
2708 | } else if (vd->vdev_wholedisk == 1) { | |
2709 | return (B_FALSE); | |
2710 | } | |
2711 | ||
9babb374 | 2712 | for (int c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2713 | if (!vdev_is_bootable(vd->vdev_child[c])) |
2714 | return (B_FALSE); | |
2715 | } | |
2716 | return (B_TRUE); | |
2717 | } | |
9babb374 BB |
2718 | |
2719 | void | |
2720 | vdev_load_log_state(vdev_t *vd, nvlist_t *nv) | |
2721 | { | |
2722 | uint_t children; | |
2723 | nvlist_t **child; | |
2724 | uint64_t val; | |
2725 | spa_t *spa = vd->vdev_spa; | |
2726 | ||
2727 | if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
2728 | &child, &children) == 0) { | |
2729 | for (int c = 0; c < children; c++) | |
2730 | vdev_load_log_state(vd->vdev_child[c], child[c]); | |
2731 | } | |
2732 | ||
2733 | if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv, | |
2734 | ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) { | |
2735 | ||
2736 | /* | |
2737 | * It would be nice to call vdev_offline() | |
2738 | * directly but the pool isn't fully loaded and | |
2739 | * the txg threads have not been started yet. | |
2740 | */ | |
2741 | spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER); | |
2742 | vd->vdev_offline = val; | |
2743 | vdev_reopen(vd->vdev_top); | |
2744 | spa_config_exit(spa, SCL_STATE_ALL, FTAG); | |
2745 | } | |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * Expand a vdev if possible. | |
2750 | */ | |
2751 | void | |
2752 | vdev_expand(vdev_t *vd, uint64_t txg) | |
2753 | { | |
2754 | ASSERT(vd->vdev_top == vd); | |
2755 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
2756 | ||
2757 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
2758 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
2759 | vdev_config_dirty(vd); | |
2760 | } | |
2761 | } |