]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
34dc7c2f BB |
24 | */ |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/fm/fs/zfs.h> | |
28 | #include <sys/spa.h> | |
29 | #include <sys/spa_impl.h> | |
30 | #include <sys/dmu.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/uberblock_impl.h> | |
34 | #include <sys/metaslab.h> | |
35 | #include <sys/metaslab_impl.h> | |
36 | #include <sys/space_map.h> | |
37 | #include <sys/zio.h> | |
38 | #include <sys/zap.h> | |
39 | #include <sys/fs/zfs.h> | |
b128c09f | 40 | #include <sys/arc.h> |
9babb374 | 41 | #include <sys/zil.h> |
428870ff | 42 | #include <sys/dsl_scan.h> |
34dc7c2f BB |
43 | |
44 | /* | |
45 | * Virtual device management. | |
46 | */ | |
47 | ||
48 | static vdev_ops_t *vdev_ops_table[] = { | |
49 | &vdev_root_ops, | |
50 | &vdev_raidz_ops, | |
51 | &vdev_mirror_ops, | |
52 | &vdev_replacing_ops, | |
53 | &vdev_spare_ops, | |
54 | &vdev_disk_ops, | |
55 | &vdev_file_ops, | |
56 | &vdev_missing_ops, | |
428870ff | 57 | &vdev_hole_ops, |
34dc7c2f BB |
58 | NULL |
59 | }; | |
60 | ||
b128c09f BB |
61 | /* maximum scrub/resilver I/O queue per leaf vdev */ |
62 | int zfs_scrub_limit = 10; | |
34dc7c2f BB |
63 | |
64 | /* | |
65 | * Given a vdev type, return the appropriate ops vector. | |
66 | */ | |
67 | static vdev_ops_t * | |
68 | vdev_getops(const char *type) | |
69 | { | |
70 | vdev_ops_t *ops, **opspp; | |
71 | ||
72 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
73 | if (strcmp(ops->vdev_op_type, type) == 0) | |
74 | break; | |
75 | ||
76 | return (ops); | |
77 | } | |
78 | ||
79 | /* | |
80 | * Default asize function: return the MAX of psize with the asize of | |
81 | * all children. This is what's used by anything other than RAID-Z. | |
82 | */ | |
83 | uint64_t | |
84 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
85 | { | |
86 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
87 | uint64_t csize; | |
d6320ddb | 88 | int c; |
34dc7c2f | 89 | |
d6320ddb | 90 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
91 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
92 | asize = MAX(asize, csize); | |
93 | } | |
94 | ||
95 | return (asize); | |
96 | } | |
97 | ||
98 | /* | |
9babb374 BB |
99 | * Get the minimum allocatable size. We define the allocatable size as |
100 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
101 | * replace or attach devices which don't have the same physical size but | |
102 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
103 | */ |
104 | uint64_t | |
9babb374 | 105 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 106 | { |
9babb374 | 107 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 108 | |
9babb374 BB |
109 | /* |
110 | * The our parent is NULL (inactive spare or cache) or is the root, | |
111 | * just return our own asize. | |
112 | */ | |
113 | if (pvd == NULL) | |
114 | return (vd->vdev_asize); | |
34dc7c2f BB |
115 | |
116 | /* | |
9babb374 BB |
117 | * The top-level vdev just returns the allocatable size rounded |
118 | * to the nearest metaslab. | |
34dc7c2f | 119 | */ |
9babb374 BB |
120 | if (vd == vd->vdev_top) |
121 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 122 | |
9babb374 BB |
123 | /* |
124 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
125 | * so each child must provide at least 1/Nth of its asize. | |
126 | */ | |
127 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
128 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 129 | |
9babb374 BB |
130 | return (pvd->vdev_min_asize); |
131 | } | |
132 | ||
133 | void | |
134 | vdev_set_min_asize(vdev_t *vd) | |
135 | { | |
d6320ddb | 136 | int c; |
9babb374 | 137 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 138 | |
d6320ddb | 139 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 140 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
141 | } |
142 | ||
143 | vdev_t * | |
144 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
145 | { | |
146 | vdev_t *rvd = spa->spa_root_vdev; | |
147 | ||
b128c09f | 148 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 149 | |
b128c09f BB |
150 | if (vdev < rvd->vdev_children) { |
151 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 152 | return (rvd->vdev_child[vdev]); |
b128c09f | 153 | } |
34dc7c2f BB |
154 | |
155 | return (NULL); | |
156 | } | |
157 | ||
158 | vdev_t * | |
159 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
160 | { | |
34dc7c2f | 161 | vdev_t *mvd; |
d6320ddb | 162 | int c; |
34dc7c2f BB |
163 | |
164 | if (vd->vdev_guid == guid) | |
165 | return (vd); | |
166 | ||
d6320ddb | 167 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
168 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
169 | NULL) | |
170 | return (mvd); | |
171 | ||
172 | return (NULL); | |
173 | } | |
174 | ||
175 | void | |
176 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
177 | { | |
178 | size_t oldsize, newsize; | |
179 | uint64_t id = cvd->vdev_id; | |
180 | vdev_t **newchild; | |
181 | ||
b128c09f | 182 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
183 | ASSERT(cvd->vdev_parent == NULL); |
184 | ||
185 | cvd->vdev_parent = pvd; | |
186 | ||
187 | if (pvd == NULL) | |
188 | return; | |
189 | ||
190 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
191 | ||
192 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
193 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
194 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
195 | ||
196 | newchild = kmem_zalloc(newsize, KM_SLEEP); | |
197 | if (pvd->vdev_child != NULL) { | |
198 | bcopy(pvd->vdev_child, newchild, oldsize); | |
199 | kmem_free(pvd->vdev_child, oldsize); | |
200 | } | |
201 | ||
202 | pvd->vdev_child = newchild; | |
203 | pvd->vdev_child[id] = cvd; | |
204 | ||
205 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
206 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
207 | ||
208 | /* | |
209 | * Walk up all ancestors to update guid sum. | |
210 | */ | |
211 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
212 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
213 | } |
214 | ||
215 | void | |
216 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
217 | { | |
218 | int c; | |
219 | uint_t id = cvd->vdev_id; | |
220 | ||
221 | ASSERT(cvd->vdev_parent == pvd); | |
222 | ||
223 | if (pvd == NULL) | |
224 | return; | |
225 | ||
226 | ASSERT(id < pvd->vdev_children); | |
227 | ASSERT(pvd->vdev_child[id] == cvd); | |
228 | ||
229 | pvd->vdev_child[id] = NULL; | |
230 | cvd->vdev_parent = NULL; | |
231 | ||
232 | for (c = 0; c < pvd->vdev_children; c++) | |
233 | if (pvd->vdev_child[c]) | |
234 | break; | |
235 | ||
236 | if (c == pvd->vdev_children) { | |
237 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
238 | pvd->vdev_child = NULL; | |
239 | pvd->vdev_children = 0; | |
240 | } | |
241 | ||
242 | /* | |
243 | * Walk up all ancestors to update guid sum. | |
244 | */ | |
245 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
246 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
247 | } |
248 | ||
249 | /* | |
250 | * Remove any holes in the child array. | |
251 | */ | |
252 | void | |
253 | vdev_compact_children(vdev_t *pvd) | |
254 | { | |
255 | vdev_t **newchild, *cvd; | |
256 | int oldc = pvd->vdev_children; | |
9babb374 | 257 | int newc; |
d6320ddb | 258 | int c; |
34dc7c2f | 259 | |
b128c09f | 260 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 261 | |
d6320ddb | 262 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
263 | if (pvd->vdev_child[c]) |
264 | newc++; | |
265 | ||
266 | newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); | |
267 | ||
d6320ddb | 268 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
269 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
270 | newchild[newc] = cvd; | |
271 | cvd->vdev_id = newc++; | |
272 | } | |
273 | } | |
274 | ||
275 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
276 | pvd->vdev_child = newchild; | |
277 | pvd->vdev_children = newc; | |
278 | } | |
279 | ||
280 | /* | |
281 | * Allocate and minimally initialize a vdev_t. | |
282 | */ | |
428870ff | 283 | vdev_t * |
34dc7c2f BB |
284 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
285 | { | |
286 | vdev_t *vd; | |
d6320ddb | 287 | int t; |
34dc7c2f BB |
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 | ||
428870ff | 296 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
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 | */ | |
428870ff | 302 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
303 | } else { |
304 | /* | |
305 | * Any other vdev's guid must be unique within the pool. | |
306 | */ | |
428870ff | 307 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
308 | } |
309 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
310 | } | |
311 | ||
312 | vd->vdev_spa = spa; | |
313 | vd->vdev_id = id; | |
314 | vd->vdev_guid = guid; | |
315 | vd->vdev_guid_sum = guid; | |
316 | vd->vdev_ops = ops; | |
317 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 318 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f BB |
319 | |
320 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); | |
321 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 322 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
d6320ddb | 323 | for (t = 0; t < DTL_TYPES; t++) { |
fb5f0bc8 BB |
324 | space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, |
325 | &vd->vdev_dtl_lock); | |
326 | } | |
34dc7c2f BB |
327 | txg_list_create(&vd->vdev_ms_list, |
328 | offsetof(struct metaslab, ms_txg_node)); | |
329 | txg_list_create(&vd->vdev_dtl_list, | |
330 | offsetof(struct vdev, vdev_dtl_node)); | |
331 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
332 | vdev_queue_init(vd); | |
333 | vdev_cache_init(vd); | |
334 | ||
335 | return (vd); | |
336 | } | |
337 | ||
338 | /* | |
339 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
340 | * creating a new vdev or loading an existing one - the behavior is slightly | |
341 | * different for each case. | |
342 | */ | |
343 | int | |
344 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
345 | int alloctype) | |
346 | { | |
347 | vdev_ops_t *ops; | |
348 | char *type; | |
349 | uint64_t guid = 0, islog, nparity; | |
350 | vdev_t *vd; | |
351 | ||
b128c09f | 352 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
353 | |
354 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
355 | return (EINVAL); | |
356 | ||
357 | if ((ops = vdev_getops(type)) == NULL) | |
358 | return (EINVAL); | |
359 | ||
360 | /* | |
361 | * If this is a load, get the vdev guid from the nvlist. | |
362 | * Otherwise, vdev_alloc_common() will generate one for us. | |
363 | */ | |
364 | if (alloctype == VDEV_ALLOC_LOAD) { | |
365 | uint64_t label_id; | |
366 | ||
367 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
368 | label_id != id) | |
369 | return (EINVAL); | |
370 | ||
371 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
372 | return (EINVAL); | |
373 | } else if (alloctype == VDEV_ALLOC_SPARE) { | |
374 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
375 | return (EINVAL); | |
376 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { | |
377 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
378 | return (EINVAL); | |
9babb374 BB |
379 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
380 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
381 | return (EINVAL); | |
34dc7c2f BB |
382 | } |
383 | ||
384 | /* | |
385 | * The first allocated vdev must be of type 'root'. | |
386 | */ | |
387 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
388 | return (EINVAL); | |
389 | ||
390 | /* | |
391 | * Determine whether we're a log vdev. | |
392 | */ | |
393 | islog = 0; | |
394 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
395 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
396 | return (ENOTSUP); | |
397 | ||
428870ff BB |
398 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
399 | return (ENOTSUP); | |
400 | ||
34dc7c2f BB |
401 | /* |
402 | * Set the nparity property for RAID-Z vdevs. | |
403 | */ | |
404 | nparity = -1ULL; | |
405 | if (ops == &vdev_raidz_ops) { | |
406 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
407 | &nparity) == 0) { | |
428870ff | 408 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
34dc7c2f BB |
409 | return (EINVAL); |
410 | /* | |
45d1cae3 BB |
411 | * Previous versions could only support 1 or 2 parity |
412 | * device. | |
34dc7c2f | 413 | */ |
45d1cae3 BB |
414 | if (nparity > 1 && |
415 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
416 | return (ENOTSUP); | |
417 | if (nparity > 2 && | |
418 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
34dc7c2f BB |
419 | return (ENOTSUP); |
420 | } else { | |
421 | /* | |
422 | * We require the parity to be specified for SPAs that | |
423 | * support multiple parity levels. | |
424 | */ | |
45d1cae3 | 425 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
34dc7c2f BB |
426 | return (EINVAL); |
427 | /* | |
428 | * Otherwise, we default to 1 parity device for RAID-Z. | |
429 | */ | |
430 | nparity = 1; | |
431 | } | |
432 | } else { | |
433 | nparity = 0; | |
434 | } | |
435 | ASSERT(nparity != -1ULL); | |
436 | ||
437 | vd = vdev_alloc_common(spa, id, guid, ops); | |
438 | ||
439 | vd->vdev_islog = islog; | |
440 | vd->vdev_nparity = nparity; | |
441 | ||
442 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
443 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
444 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
445 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
446 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
447 | &vd->vdev_physpath) == 0) | |
448 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
449 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
450 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
451 | |
452 | /* | |
453 | * Set the whole_disk property. If it's not specified, leave the value | |
454 | * as -1. | |
455 | */ | |
456 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
457 | &vd->vdev_wholedisk) != 0) | |
458 | vd->vdev_wholedisk = -1ULL; | |
459 | ||
460 | /* | |
461 | * Look for the 'not present' flag. This will only be set if the device | |
462 | * was not present at the time of import. | |
463 | */ | |
9babb374 BB |
464 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
465 | &vd->vdev_not_present); | |
34dc7c2f BB |
466 | |
467 | /* | |
468 | * Get the alignment requirement. | |
469 | */ | |
470 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
471 | ||
428870ff BB |
472 | /* |
473 | * Retrieve the vdev creation time. | |
474 | */ | |
475 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
476 | &vd->vdev_crtxg); | |
477 | ||
34dc7c2f BB |
478 | /* |
479 | * If we're a top-level vdev, try to load the allocation parameters. | |
480 | */ | |
428870ff BB |
481 | if (parent && !parent->vdev_parent && |
482 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
483 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
484 | &vd->vdev_ms_array); | |
485 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
486 | &vd->vdev_ms_shift); | |
487 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
488 | &vd->vdev_asize); | |
428870ff BB |
489 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
490 | &vd->vdev_removing); | |
491 | } | |
492 | ||
493 | if (parent && !parent->vdev_parent) { | |
494 | ASSERT(alloctype == VDEV_ALLOC_LOAD || | |
495 | alloctype == VDEV_ALLOC_ADD || | |
496 | alloctype == VDEV_ALLOC_SPLIT || | |
497 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
498 | vd->vdev_mg = metaslab_group_create(islog ? | |
499 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
500 | } |
501 | ||
502 | /* | |
503 | * If we're a leaf vdev, try to load the DTL object and other state. | |
504 | */ | |
b128c09f | 505 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
506 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
507 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
508 | if (alloctype == VDEV_ALLOC_LOAD) { |
509 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
fb5f0bc8 | 510 | &vd->vdev_dtl_smo.smo_object); |
b128c09f BB |
511 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
512 | &vd->vdev_unspare); | |
513 | } | |
9babb374 BB |
514 | |
515 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
516 | uint64_t spare = 0; | |
517 | ||
518 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
519 | &spare) == 0 && spare) | |
520 | spa_spare_add(vd); | |
521 | } | |
522 | ||
34dc7c2f BB |
523 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
524 | &vd->vdev_offline); | |
b128c09f | 525 | |
572e2857 BB |
526 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, |
527 | &vd->vdev_resilvering); | |
528 | ||
34dc7c2f BB |
529 | /* |
530 | * When importing a pool, we want to ignore the persistent fault | |
531 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
532 | * valid in the current context. Local vdevs will |
533 | * remain in the faulted state. | |
34dc7c2f | 534 | */ |
428870ff | 535 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
536 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
537 | &vd->vdev_faulted); | |
538 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
539 | &vd->vdev_degraded); | |
540 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
541 | &vd->vdev_removed); | |
428870ff BB |
542 | |
543 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
544 | char *aux; | |
545 | ||
546 | vd->vdev_label_aux = | |
547 | VDEV_AUX_ERR_EXCEEDED; | |
548 | if (nvlist_lookup_string(nv, | |
549 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
550 | strcmp(aux, "external") == 0) | |
551 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
552 | } | |
34dc7c2f BB |
553 | } |
554 | } | |
555 | ||
556 | /* | |
557 | * Add ourselves to the parent's list of children. | |
558 | */ | |
559 | vdev_add_child(parent, vd); | |
560 | ||
561 | *vdp = vd; | |
562 | ||
563 | return (0); | |
564 | } | |
565 | ||
566 | void | |
567 | vdev_free(vdev_t *vd) | |
568 | { | |
d6320ddb | 569 | int c, t; |
34dc7c2f BB |
570 | spa_t *spa = vd->vdev_spa; |
571 | ||
572 | /* | |
573 | * vdev_free() implies closing the vdev first. This is simpler than | |
574 | * trying to ensure complicated semantics for all callers. | |
575 | */ | |
576 | vdev_close(vd); | |
577 | ||
b128c09f | 578 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 579 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
580 | |
581 | /* | |
582 | * Free all children. | |
583 | */ | |
d6320ddb | 584 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
585 | vdev_free(vd->vdev_child[c]); |
586 | ||
587 | ASSERT(vd->vdev_child == NULL); | |
588 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
589 | ||
590 | /* | |
591 | * Discard allocation state. | |
592 | */ | |
428870ff | 593 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 594 | vdev_metaslab_fini(vd); |
428870ff BB |
595 | metaslab_group_destroy(vd->vdev_mg); |
596 | } | |
34dc7c2f BB |
597 | |
598 | ASSERT3U(vd->vdev_stat.vs_space, ==, 0); | |
599 | ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); | |
600 | ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); | |
601 | ||
602 | /* | |
603 | * Remove this vdev from its parent's child list. | |
604 | */ | |
605 | vdev_remove_child(vd->vdev_parent, vd); | |
606 | ||
607 | ASSERT(vd->vdev_parent == NULL); | |
608 | ||
609 | /* | |
610 | * Clean up vdev structure. | |
611 | */ | |
612 | vdev_queue_fini(vd); | |
613 | vdev_cache_fini(vd); | |
614 | ||
615 | if (vd->vdev_path) | |
616 | spa_strfree(vd->vdev_path); | |
617 | if (vd->vdev_devid) | |
618 | spa_strfree(vd->vdev_devid); | |
619 | if (vd->vdev_physpath) | |
620 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
621 | if (vd->vdev_fru) |
622 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
623 | |
624 | if (vd->vdev_isspare) | |
625 | spa_spare_remove(vd); | |
626 | if (vd->vdev_isl2cache) | |
627 | spa_l2cache_remove(vd); | |
628 | ||
629 | txg_list_destroy(&vd->vdev_ms_list); | |
630 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 631 | |
34dc7c2f | 632 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 633 | for (t = 0; t < DTL_TYPES; t++) { |
fb5f0bc8 BB |
634 | space_map_unload(&vd->vdev_dtl[t]); |
635 | space_map_destroy(&vd->vdev_dtl[t]); | |
636 | } | |
34dc7c2f | 637 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 638 | |
34dc7c2f BB |
639 | mutex_destroy(&vd->vdev_dtl_lock); |
640 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 641 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
642 | |
643 | if (vd == spa->spa_root_vdev) | |
644 | spa->spa_root_vdev = NULL; | |
645 | ||
646 | kmem_free(vd, sizeof (vdev_t)); | |
647 | } | |
648 | ||
649 | /* | |
650 | * Transfer top-level vdev state from svd to tvd. | |
651 | */ | |
652 | static void | |
653 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
654 | { | |
655 | spa_t *spa = svd->vdev_spa; | |
656 | metaslab_t *msp; | |
657 | vdev_t *vd; | |
658 | int t; | |
659 | ||
660 | ASSERT(tvd == tvd->vdev_top); | |
661 | ||
662 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
663 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
664 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
665 | ||
666 | svd->vdev_ms_array = 0; | |
667 | svd->vdev_ms_shift = 0; | |
668 | svd->vdev_ms_count = 0; | |
669 | ||
670 | tvd->vdev_mg = svd->vdev_mg; | |
671 | tvd->vdev_ms = svd->vdev_ms; | |
672 | ||
673 | svd->vdev_mg = NULL; | |
674 | svd->vdev_ms = NULL; | |
675 | ||
676 | if (tvd->vdev_mg != NULL) | |
677 | tvd->vdev_mg->mg_vd = tvd; | |
678 | ||
679 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
680 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
681 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
682 | ||
683 | svd->vdev_stat.vs_alloc = 0; | |
684 | svd->vdev_stat.vs_space = 0; | |
685 | svd->vdev_stat.vs_dspace = 0; | |
686 | ||
687 | for (t = 0; t < TXG_SIZE; t++) { | |
688 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
689 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
690 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
691 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
692 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
693 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
694 | } | |
695 | ||
b128c09f | 696 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
697 | vdev_config_clean(svd); |
698 | vdev_config_dirty(tvd); | |
699 | } | |
700 | ||
b128c09f BB |
701 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
702 | vdev_state_clean(svd); | |
703 | vdev_state_dirty(tvd); | |
704 | } | |
705 | ||
34dc7c2f BB |
706 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
707 | svd->vdev_deflate_ratio = 0; | |
708 | ||
709 | tvd->vdev_islog = svd->vdev_islog; | |
710 | svd->vdev_islog = 0; | |
711 | } | |
712 | ||
713 | static void | |
714 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
715 | { | |
d6320ddb BB |
716 | int c; |
717 | ||
34dc7c2f BB |
718 | if (vd == NULL) |
719 | return; | |
720 | ||
721 | vd->vdev_top = tvd; | |
722 | ||
d6320ddb | 723 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
724 | vdev_top_update(tvd, vd->vdev_child[c]); |
725 | } | |
726 | ||
727 | /* | |
728 | * Add a mirror/replacing vdev above an existing vdev. | |
729 | */ | |
730 | vdev_t * | |
731 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
732 | { | |
733 | spa_t *spa = cvd->vdev_spa; | |
734 | vdev_t *pvd = cvd->vdev_parent; | |
735 | vdev_t *mvd; | |
736 | ||
b128c09f | 737 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
738 | |
739 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
740 | ||
741 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 742 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
34dc7c2f BB |
743 | mvd->vdev_ashift = cvd->vdev_ashift; |
744 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 745 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
746 | |
747 | vdev_remove_child(pvd, cvd); | |
748 | vdev_add_child(pvd, mvd); | |
749 | cvd->vdev_id = mvd->vdev_children; | |
750 | vdev_add_child(mvd, cvd); | |
751 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
752 | ||
753 | if (mvd == mvd->vdev_top) | |
754 | vdev_top_transfer(cvd, mvd); | |
755 | ||
756 | return (mvd); | |
757 | } | |
758 | ||
759 | /* | |
760 | * Remove a 1-way mirror/replacing vdev from the tree. | |
761 | */ | |
762 | void | |
763 | vdev_remove_parent(vdev_t *cvd) | |
764 | { | |
765 | vdev_t *mvd = cvd->vdev_parent; | |
766 | vdev_t *pvd = mvd->vdev_parent; | |
767 | ||
b128c09f | 768 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
769 | |
770 | ASSERT(mvd->vdev_children == 1); | |
771 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
772 | mvd->vdev_ops == &vdev_replacing_ops || | |
773 | mvd->vdev_ops == &vdev_spare_ops); | |
774 | cvd->vdev_ashift = mvd->vdev_ashift; | |
775 | ||
776 | vdev_remove_child(mvd, cvd); | |
777 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 778 | |
34dc7c2f | 779 | /* |
b128c09f BB |
780 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
781 | * Otherwise, we could have detached an offline device, and when we | |
782 | * go to import the pool we'll think we have two top-level vdevs, | |
783 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 784 | */ |
fb5f0bc8 BB |
785 | if (mvd->vdev_top == mvd) { |
786 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 787 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
788 | cvd->vdev_guid += guid_delta; |
789 | cvd->vdev_guid_sum += guid_delta; | |
790 | } | |
b128c09f BB |
791 | cvd->vdev_id = mvd->vdev_id; |
792 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
793 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
794 | ||
795 | if (cvd == cvd->vdev_top) | |
796 | vdev_top_transfer(mvd, cvd); | |
797 | ||
798 | ASSERT(mvd->vdev_children == 0); | |
799 | vdev_free(mvd); | |
800 | } | |
801 | ||
802 | int | |
803 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
804 | { | |
805 | spa_t *spa = vd->vdev_spa; | |
806 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
807 | uint64_t m; |
808 | uint64_t oldc = vd->vdev_ms_count; | |
809 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
810 | metaslab_t **mspp; | |
811 | int error; | |
812 | ||
428870ff BB |
813 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
814 | ||
815 | /* | |
816 | * This vdev is not being allocated from yet or is a hole. | |
817 | */ | |
818 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
819 | return (0); |
820 | ||
428870ff BB |
821 | ASSERT(!vd->vdev_ishole); |
822 | ||
9babb374 BB |
823 | /* |
824 | * Compute the raidz-deflation ratio. Note, we hard-code | |
825 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
826 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
827 | * or we will inconsistently account for existing bp's. | |
828 | */ | |
829 | vd->vdev_deflate_ratio = (1 << 17) / | |
830 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
831 | ||
34dc7c2f BB |
832 | ASSERT(oldc <= newc); |
833 | ||
34dc7c2f BB |
834 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
835 | ||
836 | if (oldc != 0) { | |
837 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
838 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
839 | } | |
840 | ||
841 | vd->vdev_ms = mspp; | |
842 | vd->vdev_ms_count = newc; | |
843 | ||
844 | for (m = oldc; m < newc; m++) { | |
845 | space_map_obj_t smo = { 0, 0, 0 }; | |
846 | if (txg == 0) { | |
847 | uint64_t object = 0; | |
848 | error = dmu_read(mos, vd->vdev_ms_array, | |
9babb374 BB |
849 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
850 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
851 | if (error) |
852 | return (error); | |
853 | if (object != 0) { | |
854 | dmu_buf_t *db; | |
855 | error = dmu_bonus_hold(mos, object, FTAG, &db); | |
856 | if (error) | |
857 | return (error); | |
858 | ASSERT3U(db->db_size, >=, sizeof (smo)); | |
859 | bcopy(db->db_data, &smo, sizeof (smo)); | |
860 | ASSERT3U(smo.smo_object, ==, object); | |
861 | dmu_buf_rele(db, FTAG); | |
862 | } | |
863 | } | |
864 | vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, | |
865 | m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); | |
866 | } | |
867 | ||
428870ff BB |
868 | if (txg == 0) |
869 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
870 | ||
871 | /* | |
872 | * If the vdev is being removed we don't activate | |
873 | * the metaslabs since we want to ensure that no new | |
874 | * allocations are performed on this device. | |
875 | */ | |
876 | if (oldc == 0 && !vd->vdev_removing) | |
877 | metaslab_group_activate(vd->vdev_mg); | |
878 | ||
879 | if (txg == 0) | |
880 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
881 | ||
34dc7c2f BB |
882 | return (0); |
883 | } | |
884 | ||
885 | void | |
886 | vdev_metaslab_fini(vdev_t *vd) | |
887 | { | |
888 | uint64_t m; | |
889 | uint64_t count = vd->vdev_ms_count; | |
890 | ||
891 | if (vd->vdev_ms != NULL) { | |
428870ff | 892 | metaslab_group_passivate(vd->vdev_mg); |
34dc7c2f BB |
893 | for (m = 0; m < count; m++) |
894 | if (vd->vdev_ms[m] != NULL) | |
895 | metaslab_fini(vd->vdev_ms[m]); | |
896 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); | |
897 | vd->vdev_ms = NULL; | |
898 | } | |
899 | } | |
900 | ||
b128c09f BB |
901 | typedef struct vdev_probe_stats { |
902 | boolean_t vps_readable; | |
903 | boolean_t vps_writeable; | |
904 | int vps_flags; | |
b128c09f BB |
905 | } vdev_probe_stats_t; |
906 | ||
907 | static void | |
908 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 909 | { |
fb5f0bc8 | 910 | spa_t *spa = zio->io_spa; |
d164b209 | 911 | vdev_t *vd = zio->io_vd; |
b128c09f | 912 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
913 | |
914 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
915 | |
916 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
917 | if (zio->io_error == 0) |
918 | vps->vps_readable = 1; | |
fb5f0bc8 | 919 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 920 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
921 | zio->io_offset, zio->io_size, zio->io_data, |
922 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
923 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
924 | } else { | |
925 | zio_buf_free(zio->io_data, zio->io_size); | |
926 | } | |
927 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
928 | if (zio->io_error == 0) |
929 | vps->vps_writeable = 1; | |
930 | zio_buf_free(zio->io_data, zio->io_size); | |
931 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 932 | zio_t *pio; |
b128c09f BB |
933 | |
934 | vd->vdev_cant_read |= !vps->vps_readable; | |
935 | vd->vdev_cant_write |= !vps->vps_writeable; | |
936 | ||
937 | if (vdev_readable(vd) && | |
fb5f0bc8 | 938 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
939 | zio->io_error = 0; |
940 | } else { | |
941 | ASSERT(zio->io_error != 0); | |
942 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 943 | spa, vd, NULL, 0, 0); |
b128c09f BB |
944 | zio->io_error = ENXIO; |
945 | } | |
d164b209 BB |
946 | |
947 | mutex_enter(&vd->vdev_probe_lock); | |
948 | ASSERT(vd->vdev_probe_zio == zio); | |
949 | vd->vdev_probe_zio = NULL; | |
950 | mutex_exit(&vd->vdev_probe_lock); | |
951 | ||
952 | while ((pio = zio_walk_parents(zio)) != NULL) | |
953 | if (!vdev_accessible(vd, pio)) | |
954 | pio->io_error = ENXIO; | |
955 | ||
b128c09f BB |
956 | kmem_free(vps, sizeof (*vps)); |
957 | } | |
958 | } | |
34dc7c2f | 959 | |
b128c09f BB |
960 | /* |
961 | * Determine whether this device is accessible by reading and writing | |
962 | * to several known locations: the pad regions of each vdev label | |
963 | * but the first (which we leave alone in case it contains a VTOC). | |
964 | */ | |
965 | zio_t * | |
d164b209 | 966 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
967 | { |
968 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
969 | vdev_probe_stats_t *vps = NULL; |
970 | zio_t *pio; | |
d6320ddb | 971 | int l; |
d164b209 BB |
972 | |
973 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 974 | |
d164b209 BB |
975 | /* |
976 | * Don't probe the probe. | |
977 | */ | |
978 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
979 | return (NULL); | |
b128c09f | 980 | |
d164b209 BB |
981 | /* |
982 | * To prevent 'probe storms' when a device fails, we create | |
983 | * just one probe i/o at a time. All zios that want to probe | |
984 | * this vdev will become parents of the probe io. | |
985 | */ | |
986 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 987 | |
d164b209 BB |
988 | if ((pio = vd->vdev_probe_zio) == NULL) { |
989 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); | |
990 | ||
991 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
992 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 993 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
994 | |
995 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
996 | /* | |
997 | * vdev_cant_read and vdev_cant_write can only | |
998 | * transition from TRUE to FALSE when we have the | |
999 | * SCL_ZIO lock as writer; otherwise they can only | |
1000 | * transition from FALSE to TRUE. This ensures that | |
1001 | * any zio looking at these values can assume that | |
1002 | * failures persist for the life of the I/O. That's | |
1003 | * important because when a device has intermittent | |
1004 | * connectivity problems, we want to ensure that | |
1005 | * they're ascribed to the device (ENXIO) and not | |
1006 | * the zio (EIO). | |
1007 | * | |
1008 | * Since we hold SCL_ZIO as writer here, clear both | |
1009 | * values so the probe can reevaluate from first | |
1010 | * principles. | |
1011 | */ | |
1012 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1013 | vd->vdev_cant_read = B_FALSE; | |
1014 | vd->vdev_cant_write = B_FALSE; | |
1015 | } | |
1016 | ||
1017 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1018 | vdev_probe_done, vps, | |
1019 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1020 | ||
428870ff BB |
1021 | /* |
1022 | * We can't change the vdev state in this context, so we | |
1023 | * kick off an async task to do it on our behalf. | |
1024 | */ | |
d164b209 BB |
1025 | if (zio != NULL) { |
1026 | vd->vdev_probe_wanted = B_TRUE; | |
1027 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1028 | } | |
b128c09f BB |
1029 | } |
1030 | ||
d164b209 BB |
1031 | if (zio != NULL) |
1032 | zio_add_child(zio, pio); | |
b128c09f | 1033 | |
d164b209 | 1034 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1035 | |
d164b209 BB |
1036 | if (vps == NULL) { |
1037 | ASSERT(zio != NULL); | |
1038 | return (NULL); | |
1039 | } | |
b128c09f | 1040 | |
d6320ddb | 1041 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1042 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1043 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
1044 | offsetof(vdev_label_t, vl_pad2)), |
1045 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
1046 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1047 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1048 | } | |
1049 | ||
d164b209 BB |
1050 | if (zio == NULL) |
1051 | return (pio); | |
1052 | ||
1053 | zio_nowait(pio); | |
1054 | return (NULL); | |
34dc7c2f BB |
1055 | } |
1056 | ||
45d1cae3 BB |
1057 | static void |
1058 | vdev_open_child(void *arg) | |
1059 | { | |
1060 | vdev_t *vd = arg; | |
1061 | ||
1062 | vd->vdev_open_thread = curthread; | |
1063 | vd->vdev_open_error = vdev_open(vd); | |
1064 | vd->vdev_open_thread = NULL; | |
1065 | } | |
1066 | ||
428870ff BB |
1067 | boolean_t |
1068 | vdev_uses_zvols(vdev_t *vd) | |
1069 | { | |
d6320ddb BB |
1070 | int c; |
1071 | ||
428870ff BB |
1072 | if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, |
1073 | strlen(ZVOL_DIR)) == 0) | |
1074 | return (B_TRUE); | |
d6320ddb | 1075 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1076 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1077 | return (B_TRUE); | |
1078 | return (B_FALSE); | |
1079 | } | |
1080 | ||
45d1cae3 BB |
1081 | void |
1082 | vdev_open_children(vdev_t *vd) | |
1083 | { | |
1084 | taskq_t *tq; | |
1085 | int children = vd->vdev_children; | |
d6320ddb | 1086 | int c; |
45d1cae3 | 1087 | |
428870ff BB |
1088 | /* |
1089 | * in order to handle pools on top of zvols, do the opens | |
1090 | * in a single thread so that the same thread holds the | |
1091 | * spa_namespace_lock | |
1092 | */ | |
1093 | if (vdev_uses_zvols(vd)) { | |
d6320ddb | 1094 | for (c = 0; c < children; c++) |
428870ff BB |
1095 | vd->vdev_child[c]->vdev_open_error = |
1096 | vdev_open(vd->vdev_child[c]); | |
1097 | return; | |
1098 | } | |
45d1cae3 BB |
1099 | tq = taskq_create("vdev_open", children, minclsyspri, |
1100 | children, children, TASKQ_PREPOPULATE); | |
1101 | ||
d6320ddb | 1102 | for (c = 0; c < children; c++) |
45d1cae3 | 1103 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], |
d6320ddb | 1104 | TQ_SLEEP) != 0); |
45d1cae3 BB |
1105 | |
1106 | taskq_destroy(tq); | |
1107 | } | |
1108 | ||
34dc7c2f BB |
1109 | /* |
1110 | * Prepare a virtual device for access. | |
1111 | */ | |
1112 | int | |
1113 | vdev_open(vdev_t *vd) | |
1114 | { | |
fb5f0bc8 | 1115 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1116 | int error; |
34dc7c2f BB |
1117 | uint64_t osize = 0; |
1118 | uint64_t asize, psize; | |
1119 | uint64_t ashift = 0; | |
d6320ddb | 1120 | int c; |
34dc7c2f | 1121 | |
45d1cae3 BB |
1122 | ASSERT(vd->vdev_open_thread == curthread || |
1123 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1124 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1125 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1126 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1127 | ||
34dc7c2f | 1128 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1129 | vd->vdev_cant_read = B_FALSE; |
1130 | vd->vdev_cant_write = B_FALSE; | |
1131 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1132 | |
428870ff BB |
1133 | /* |
1134 | * If this vdev is not removed, check its fault status. If it's | |
1135 | * faulted, bail out of the open. | |
1136 | */ | |
34dc7c2f BB |
1137 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1138 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1139 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1140 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1141 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1142 | vd->vdev_label_aux); |
34dc7c2f BB |
1143 | return (ENXIO); |
1144 | } else if (vd->vdev_offline) { | |
1145 | ASSERT(vd->vdev_children == 0); | |
1146 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
1147 | return (ENXIO); | |
1148 | } | |
1149 | ||
1150 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); | |
1151 | ||
428870ff BB |
1152 | /* |
1153 | * Reset the vdev_reopening flag so that we actually close | |
1154 | * the vdev on error. | |
1155 | */ | |
1156 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1157 | if (zio_injection_enabled && error == 0) |
9babb374 | 1158 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1159 | |
1160 | if (error) { | |
1161 | if (vd->vdev_removed && | |
1162 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1163 | vd->vdev_removed = B_FALSE; | |
1164 | ||
1165 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1166 | vd->vdev_stat.vs_aux); | |
1167 | return (error); | |
1168 | } | |
1169 | ||
1170 | vd->vdev_removed = B_FALSE; | |
1171 | ||
428870ff BB |
1172 | /* |
1173 | * Recheck the faulted flag now that we have confirmed that | |
1174 | * the vdev is accessible. If we're faulted, bail. | |
1175 | */ | |
1176 | if (vd->vdev_faulted) { | |
1177 | ASSERT(vd->vdev_children == 0); | |
1178 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1179 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1180 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1181 | vd->vdev_label_aux); | |
1182 | return (ENXIO); | |
1183 | } | |
1184 | ||
34dc7c2f BB |
1185 | if (vd->vdev_degraded) { |
1186 | ASSERT(vd->vdev_children == 0); | |
1187 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1188 | VDEV_AUX_ERR_EXCEEDED); | |
1189 | } else { | |
428870ff | 1190 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1191 | } |
1192 | ||
428870ff BB |
1193 | /* |
1194 | * For hole or missing vdevs we just return success. | |
1195 | */ | |
1196 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1197 | return (0); | |
1198 | ||
d6320ddb | 1199 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1200 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1201 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1202 | VDEV_AUX_NONE); | |
1203 | break; | |
1204 | } | |
9babb374 | 1205 | } |
34dc7c2f BB |
1206 | |
1207 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1208 | ||
1209 | if (vd->vdev_children == 0) { | |
1210 | if (osize < SPA_MINDEVSIZE) { | |
1211 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1212 | VDEV_AUX_TOO_SMALL); | |
1213 | return (EOVERFLOW); | |
1214 | } | |
1215 | psize = osize; | |
1216 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1217 | } else { | |
1218 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1219 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1220 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1221 | VDEV_AUX_TOO_SMALL); | |
1222 | return (EOVERFLOW); | |
1223 | } | |
1224 | psize = 0; | |
1225 | asize = osize; | |
1226 | } | |
1227 | ||
1228 | vd->vdev_psize = psize; | |
1229 | ||
9babb374 BB |
1230 | /* |
1231 | * Make sure the allocatable size hasn't shrunk. | |
1232 | */ | |
1233 | if (asize < vd->vdev_min_asize) { | |
1234 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1235 | VDEV_AUX_BAD_LABEL); | |
1236 | return (EINVAL); | |
1237 | } | |
1238 | ||
34dc7c2f BB |
1239 | if (vd->vdev_asize == 0) { |
1240 | /* | |
1241 | * This is the first-ever open, so use the computed values. | |
1242 | * For testing purposes, a higher ashift can be requested. | |
1243 | */ | |
1244 | vd->vdev_asize = asize; | |
1245 | vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); | |
1246 | } else { | |
1247 | /* | |
1248 | * Make sure the alignment requirement hasn't increased. | |
1249 | */ | |
1250 | if (ashift > vd->vdev_top->vdev_ashift) { | |
1251 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1252 | VDEV_AUX_BAD_LABEL); | |
1253 | return (EINVAL); | |
1254 | } | |
9babb374 | 1255 | } |
34dc7c2f | 1256 | |
9babb374 BB |
1257 | /* |
1258 | * If all children are healthy and the asize has increased, | |
1259 | * then we've experienced dynamic LUN growth. If automatic | |
1260 | * expansion is enabled then use the additional space. | |
1261 | */ | |
1262 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1263 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1264 | vd->vdev_asize = asize; | |
34dc7c2f | 1265 | |
9babb374 | 1266 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1267 | |
1268 | /* | |
1269 | * Ensure we can issue some IO before declaring the | |
1270 | * vdev open for business. | |
1271 | */ | |
b128c09f BB |
1272 | if (vd->vdev_ops->vdev_op_leaf && |
1273 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1274 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1275 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1276 | return (error); |
1277 | } | |
1278 | ||
34dc7c2f | 1279 | /* |
b128c09f | 1280 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1281 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1282 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1283 | */ |
fb5f0bc8 BB |
1284 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1285 | vdev_resilver_needed(vd, NULL, NULL)) | |
1286 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1287 | |
1288 | return (0); | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Called once the vdevs are all opened, this routine validates the label | |
1293 | * contents. This needs to be done before vdev_load() so that we don't | |
1294 | * inadvertently do repair I/Os to the wrong device. | |
1295 | * | |
1296 | * This function will only return failure if one of the vdevs indicates that it | |
1297 | * has since been destroyed or exported. This is only possible if | |
1298 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1299 | * will be updated but the function will return 0. | |
1300 | */ | |
1301 | int | |
1302 | vdev_validate(vdev_t *vd) | |
1303 | { | |
1304 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1305 | nvlist_t *label; |
428870ff | 1306 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1307 | uint64_t state; |
d6320ddb | 1308 | int c; |
34dc7c2f | 1309 | |
d6320ddb | 1310 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1311 | if (vdev_validate(vd->vdev_child[c]) != 0) |
1312 | return (EBADF); | |
1313 | ||
1314 | /* | |
1315 | * If the device has already failed, or was marked offline, don't do | |
1316 | * any further validation. Otherwise, label I/O will fail and we will | |
1317 | * overwrite the previous state. | |
1318 | */ | |
b128c09f | 1319 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1320 | uint64_t aux_guid = 0; |
1321 | nvlist_t *nvl; | |
34dc7c2f BB |
1322 | |
1323 | if ((label = vdev_label_read_config(vd)) == NULL) { | |
1324 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1325 | VDEV_AUX_BAD_LABEL); | |
1326 | return (0); | |
1327 | } | |
1328 | ||
428870ff BB |
1329 | /* |
1330 | * Determine if this vdev has been split off into another | |
1331 | * pool. If so, then refuse to open it. | |
1332 | */ | |
1333 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1334 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1335 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1336 | VDEV_AUX_SPLIT_POOL); | |
1337 | nvlist_free(label); | |
1338 | return (0); | |
1339 | } | |
1340 | ||
34dc7c2f BB |
1341 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, |
1342 | &guid) != 0 || guid != spa_guid(spa)) { | |
1343 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1344 | VDEV_AUX_CORRUPT_DATA); | |
1345 | nvlist_free(label); | |
1346 | return (0); | |
1347 | } | |
1348 | ||
428870ff BB |
1349 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1350 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1351 | &aux_guid) != 0) | |
1352 | aux_guid = 0; | |
1353 | ||
b128c09f BB |
1354 | /* |
1355 | * If this vdev just became a top-level vdev because its | |
1356 | * sibling was detached, it will have adopted the parent's | |
1357 | * vdev guid -- but the label may or may not be on disk yet. | |
1358 | * Fortunately, either version of the label will have the | |
1359 | * same top guid, so if we're a top-level vdev, we can | |
1360 | * safely compare to that instead. | |
428870ff BB |
1361 | * |
1362 | * If we split this vdev off instead, then we also check the | |
1363 | * original pool's guid. We don't want to consider the vdev | |
1364 | * corrupt if it is partway through a split operation. | |
b128c09f | 1365 | */ |
34dc7c2f | 1366 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1367 | &guid) != 0 || |
1368 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1369 | &top_guid) != 0 || | |
428870ff | 1370 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1371 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1372 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1373 | VDEV_AUX_CORRUPT_DATA); | |
1374 | nvlist_free(label); | |
1375 | return (0); | |
1376 | } | |
1377 | ||
1378 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1379 | &state) != 0) { | |
1380 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1381 | VDEV_AUX_CORRUPT_DATA); | |
1382 | nvlist_free(label); | |
1383 | return (0); | |
1384 | } | |
1385 | ||
1386 | nvlist_free(label); | |
1387 | ||
45d1cae3 | 1388 | /* |
572e2857 | 1389 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1390 | * state of the pool. |
1391 | */ | |
572e2857 | 1392 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1393 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f BB |
1394 | state != POOL_STATE_ACTIVE) |
1395 | return (EBADF); | |
34dc7c2f | 1396 | |
b128c09f BB |
1397 | /* |
1398 | * If we were able to open and validate a vdev that was | |
1399 | * previously marked permanently unavailable, clear that state | |
1400 | * now. | |
1401 | */ | |
1402 | if (vd->vdev_not_present) | |
1403 | vd->vdev_not_present = 0; | |
1404 | } | |
34dc7c2f BB |
1405 | |
1406 | return (0); | |
1407 | } | |
1408 | ||
1409 | /* | |
1410 | * Close a virtual device. | |
1411 | */ | |
1412 | void | |
1413 | vdev_close(vdev_t *vd) | |
1414 | { | |
428870ff | 1415 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1416 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1417 | |
1418 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1419 | ||
428870ff BB |
1420 | /* |
1421 | * If our parent is reopening, then we are as well, unless we are | |
1422 | * going offline. | |
1423 | */ | |
1424 | if (pvd != NULL && pvd->vdev_reopening) | |
1425 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1426 | ||
34dc7c2f BB |
1427 | vd->vdev_ops->vdev_op_close(vd); |
1428 | ||
1429 | vdev_cache_purge(vd); | |
1430 | ||
1431 | /* | |
9babb374 | 1432 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1433 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1434 | * it's still faulted. | |
1435 | */ | |
1436 | vd->vdev_prevstate = vd->vdev_state; | |
1437 | ||
1438 | if (vd->vdev_offline) | |
1439 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1440 | else | |
1441 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1442 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1443 | } | |
1444 | ||
428870ff BB |
1445 | void |
1446 | vdev_hold(vdev_t *vd) | |
1447 | { | |
1448 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1449 | int c; |
428870ff BB |
1450 | |
1451 | ASSERT(spa_is_root(spa)); | |
1452 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1453 | return; | |
1454 | ||
d6320ddb | 1455 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1456 | vdev_hold(vd->vdev_child[c]); |
1457 | ||
1458 | if (vd->vdev_ops->vdev_op_leaf) | |
1459 | vd->vdev_ops->vdev_op_hold(vd); | |
1460 | } | |
1461 | ||
1462 | void | |
1463 | vdev_rele(vdev_t *vd) | |
1464 | { | |
d6320ddb | 1465 | int c; |
428870ff | 1466 | |
d6320ddb BB |
1467 | ASSERT(spa_is_root(vd->vdev_spa)); |
1468 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1469 | vdev_rele(vd->vdev_child[c]); |
1470 | ||
1471 | if (vd->vdev_ops->vdev_op_leaf) | |
1472 | vd->vdev_ops->vdev_op_rele(vd); | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1477 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1478 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1479 | * If the leaf has never been opened then open it, as usual. | |
1480 | */ | |
34dc7c2f BB |
1481 | void |
1482 | vdev_reopen(vdev_t *vd) | |
1483 | { | |
1484 | spa_t *spa = vd->vdev_spa; | |
1485 | ||
b128c09f | 1486 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1487 | |
428870ff BB |
1488 | /* set the reopening flag unless we're taking the vdev offline */ |
1489 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1490 | vdev_close(vd); |
1491 | (void) vdev_open(vd); | |
1492 | ||
1493 | /* | |
1494 | * Call vdev_validate() here to make sure we have the same device. | |
1495 | * Otherwise, a device with an invalid label could be successfully | |
1496 | * opened in response to vdev_reopen(). | |
1497 | */ | |
b128c09f BB |
1498 | if (vd->vdev_aux) { |
1499 | (void) vdev_validate_aux(vd); | |
1500 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1501 | vd->vdev_aux == &spa->spa_l2cache && |
1502 | !l2arc_vdev_present(vd)) | |
1503 | l2arc_add_vdev(spa, vd); | |
b128c09f BB |
1504 | } else { |
1505 | (void) vdev_validate(vd); | |
1506 | } | |
34dc7c2f BB |
1507 | |
1508 | /* | |
1509 | * Reassess parent vdev's health. | |
1510 | */ | |
1511 | vdev_propagate_state(vd); | |
1512 | } | |
1513 | ||
1514 | int | |
1515 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1516 | { | |
1517 | int error; | |
1518 | ||
1519 | /* | |
1520 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1521 | * For a create, however, we want to fail the request if | |
1522 | * there are any components we can't open. | |
1523 | */ | |
1524 | error = vdev_open(vd); | |
1525 | ||
1526 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1527 | vdev_close(vd); | |
1528 | return (error ? error : ENXIO); | |
1529 | } | |
1530 | ||
1531 | /* | |
1532 | * Recursively initialize all labels. | |
1533 | */ | |
1534 | if ((error = vdev_label_init(vd, txg, isreplacing ? | |
1535 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { | |
1536 | vdev_close(vd); | |
1537 | return (error); | |
1538 | } | |
1539 | ||
1540 | return (0); | |
1541 | } | |
1542 | ||
34dc7c2f | 1543 | void |
9babb374 | 1544 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1545 | { |
1546 | /* | |
1547 | * Aim for roughly 200 metaslabs per vdev. | |
1548 | */ | |
1549 | vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); | |
1550 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
1551 | } |
1552 | ||
1553 | void | |
1554 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1555 | { | |
1556 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1557 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1558 | ASSERT(ISP2(flags)); |
572e2857 | 1559 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1560 | |
1561 | if (flags & VDD_METASLAB) | |
1562 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1563 | ||
1564 | if (flags & VDD_DTL) | |
1565 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1566 | ||
1567 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1568 | } | |
1569 | ||
fb5f0bc8 BB |
1570 | /* |
1571 | * DTLs. | |
1572 | * | |
1573 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1574 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1575 | * |
1576 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1577 | * | |
1578 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1579 | * | |
1580 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1581 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1582 | * txgs that was scrubbed. | |
1583 | * | |
1584 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1585 | * persistent errors or just some device being offline. | |
1586 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1587 | * maintained; it's only computed when needed, typically to | |
1588 | * determine whether a device can be detached. | |
1589 | * | |
1590 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1591 | * either has the data or it doesn't. | |
1592 | * | |
1593 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1594 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1595 | * if any child is less than fully replicated, then so is its parent. | |
1596 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1597 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1598 | * those are the txgs we don't have enough replication to read. For example, | |
1599 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1600 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1601 | * two child DTL_MISSING maps. | |
1602 | * | |
1603 | * It should be clear from the above that to compute the DTLs and outage maps | |
1604 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1605 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1606 | * a configuration change, we generate all other DTLs from first principles. | |
1607 | */ | |
34dc7c2f | 1608 | void |
fb5f0bc8 | 1609 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1610 | { |
fb5f0bc8 BB |
1611 | space_map_t *sm = &vd->vdev_dtl[t]; |
1612 | ||
1613 | ASSERT(t < DTL_TYPES); | |
1614 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1615 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1616 | |
34dc7c2f BB |
1617 | mutex_enter(sm->sm_lock); |
1618 | if (!space_map_contains(sm, txg, size)) | |
1619 | space_map_add(sm, txg, size); | |
1620 | mutex_exit(sm->sm_lock); | |
1621 | } | |
1622 | ||
fb5f0bc8 BB |
1623 | boolean_t |
1624 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1625 | { |
fb5f0bc8 BB |
1626 | space_map_t *sm = &vd->vdev_dtl[t]; |
1627 | boolean_t dirty = B_FALSE; | |
34dc7c2f | 1628 | |
fb5f0bc8 BB |
1629 | ASSERT(t < DTL_TYPES); |
1630 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f BB |
1631 | |
1632 | mutex_enter(sm->sm_lock); | |
fb5f0bc8 BB |
1633 | if (sm->sm_space != 0) |
1634 | dirty = space_map_contains(sm, txg, size); | |
34dc7c2f BB |
1635 | mutex_exit(sm->sm_lock); |
1636 | ||
1637 | return (dirty); | |
1638 | } | |
1639 | ||
fb5f0bc8 BB |
1640 | boolean_t |
1641 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1642 | { | |
1643 | space_map_t *sm = &vd->vdev_dtl[t]; | |
1644 | boolean_t empty; | |
1645 | ||
1646 | mutex_enter(sm->sm_lock); | |
1647 | empty = (sm->sm_space == 0); | |
1648 | mutex_exit(sm->sm_lock); | |
1649 | ||
1650 | return (empty); | |
1651 | } | |
1652 | ||
34dc7c2f BB |
1653 | /* |
1654 | * Reassess DTLs after a config change or scrub completion. | |
1655 | */ | |
1656 | void | |
1657 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1658 | { | |
1659 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1660 | avl_tree_t reftree; |
d6320ddb | 1661 | int c, t, minref; |
34dc7c2f | 1662 | |
fb5f0bc8 | 1663 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1664 | |
d6320ddb | 1665 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1666 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1667 | scrub_txg, scrub_done); | |
1668 | ||
428870ff | 1669 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1670 | return; |
1671 | ||
1672 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1673 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1674 | ||
34dc7c2f | 1675 | mutex_enter(&vd->vdev_dtl_lock); |
b128c09f | 1676 | if (scrub_txg != 0 && |
428870ff BB |
1677 | (spa->spa_scrub_started || |
1678 | (scn && scn->scn_phys.scn_errors == 0))) { | |
b128c09f BB |
1679 | /* |
1680 | * We completed a scrub up to scrub_txg. If we | |
1681 | * did it without rebooting, then the scrub dtl | |
1682 | * will be valid, so excise the old region and | |
1683 | * fold in the scrub dtl. Otherwise, leave the | |
1684 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1685 | * |
1686 | * There's little trick here: to excise the beginning | |
1687 | * of the DTL_MISSING map, we put it into a reference | |
1688 | * tree and then add a segment with refcnt -1 that | |
1689 | * covers the range [0, scrub_txg). This means | |
1690 | * that each txg in that range has refcnt -1 or 0. | |
1691 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1692 | * entries in the range [0, scrub_txg) will have a | |
1693 | * positive refcnt -- either 1 or 2. We then convert | |
1694 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1695 | */ |
fb5f0bc8 BB |
1696 | space_map_ref_create(&reftree); |
1697 | space_map_ref_add_map(&reftree, | |
1698 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1699 | space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); | |
1700 | space_map_ref_add_map(&reftree, | |
1701 | &vd->vdev_dtl[DTL_SCRUB], 2); | |
1702 | space_map_ref_generate_map(&reftree, | |
1703 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1704 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1705 | } |
fb5f0bc8 BB |
1706 | space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1707 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1708 | space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1709 | if (scrub_done) |
fb5f0bc8 BB |
1710 | space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1711 | space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
1712 | if (!vdev_readable(vd)) | |
1713 | space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); | |
1714 | else | |
1715 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1716 | space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); | |
34dc7c2f | 1717 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1718 | |
34dc7c2f BB |
1719 | if (txg != 0) |
1720 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1721 | return; | |
1722 | } | |
1723 | ||
34dc7c2f | 1724 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1725 | for (t = 0; t < DTL_TYPES; t++) { |
428870ff BB |
1726 | /* account for child's outage in parent's missing map */ |
1727 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1728 | if (t == DTL_SCRUB) |
1729 | continue; /* leaf vdevs only */ | |
1730 | if (t == DTL_PARTIAL) | |
1731 | minref = 1; /* i.e. non-zero */ | |
1732 | else if (vd->vdev_nparity != 0) | |
1733 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1734 | else | |
1735 | minref = vd->vdev_children; /* any kind of mirror */ | |
1736 | space_map_ref_create(&reftree); | |
d6320ddb | 1737 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1738 | vdev_t *cvd = vd->vdev_child[c]; |
1739 | mutex_enter(&cvd->vdev_dtl_lock); | |
428870ff | 1740 | space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1741 | mutex_exit(&cvd->vdev_dtl_lock); |
1742 | } | |
1743 | space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); | |
1744 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1745 | } |
fb5f0bc8 | 1746 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1747 | } |
1748 | ||
1749 | static int | |
1750 | vdev_dtl_load(vdev_t *vd) | |
1751 | { | |
1752 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1753 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
34dc7c2f BB |
1754 | objset_t *mos = spa->spa_meta_objset; |
1755 | dmu_buf_t *db; | |
1756 | int error; | |
1757 | ||
1758 | ASSERT(vd->vdev_children == 0); | |
1759 | ||
1760 | if (smo->smo_object == 0) | |
1761 | return (0); | |
1762 | ||
428870ff BB |
1763 | ASSERT(!vd->vdev_ishole); |
1764 | ||
34dc7c2f BB |
1765 | if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) |
1766 | return (error); | |
1767 | ||
1768 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1769 | bcopy(db->db_data, smo, sizeof (*smo)); | |
1770 | dmu_buf_rele(db, FTAG); | |
1771 | ||
1772 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1773 | error = space_map_load(&vd->vdev_dtl[DTL_MISSING], |
1774 | NULL, SM_ALLOC, smo, mos); | |
34dc7c2f BB |
1775 | mutex_exit(&vd->vdev_dtl_lock); |
1776 | ||
1777 | return (error); | |
1778 | } | |
1779 | ||
1780 | void | |
1781 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1782 | { | |
1783 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1784 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
1785 | space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; | |
34dc7c2f BB |
1786 | objset_t *mos = spa->spa_meta_objset; |
1787 | space_map_t smsync; | |
1788 | kmutex_t smlock; | |
1789 | dmu_buf_t *db; | |
1790 | dmu_tx_t *tx; | |
1791 | ||
428870ff BB |
1792 | ASSERT(!vd->vdev_ishole); |
1793 | ||
34dc7c2f BB |
1794 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1795 | ||
1796 | if (vd->vdev_detached) { | |
1797 | if (smo->smo_object != 0) { | |
1fde1e37 | 1798 | VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx)); |
34dc7c2f BB |
1799 | smo->smo_object = 0; |
1800 | } | |
1801 | dmu_tx_commit(tx); | |
34dc7c2f BB |
1802 | return; |
1803 | } | |
1804 | ||
1805 | if (smo->smo_object == 0) { | |
1806 | ASSERT(smo->smo_objsize == 0); | |
1807 | ASSERT(smo->smo_alloc == 0); | |
1808 | smo->smo_object = dmu_object_alloc(mos, | |
1809 | DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, | |
1810 | DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); | |
1811 | ASSERT(smo->smo_object != 0); | |
1812 | vdev_config_dirty(vd->vdev_top); | |
1813 | } | |
1814 | ||
1815 | mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); | |
1816 | ||
1817 | space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, | |
1818 | &smlock); | |
1819 | ||
1820 | mutex_enter(&smlock); | |
1821 | ||
1822 | mutex_enter(&vd->vdev_dtl_lock); | |
1823 | space_map_walk(sm, space_map_add, &smsync); | |
1824 | mutex_exit(&vd->vdev_dtl_lock); | |
1825 | ||
1826 | space_map_truncate(smo, mos, tx); | |
1827 | space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); | |
1828 | ||
1829 | space_map_destroy(&smsync); | |
1830 | ||
1831 | mutex_exit(&smlock); | |
1832 | mutex_destroy(&smlock); | |
1833 | ||
1834 | VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); | |
1835 | dmu_buf_will_dirty(db, tx); | |
1836 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1837 | bcopy(smo, db->db_data, sizeof (*smo)); | |
1838 | dmu_buf_rele(db, FTAG); | |
1839 | ||
1840 | dmu_tx_commit(tx); | |
1841 | } | |
1842 | ||
fb5f0bc8 BB |
1843 | /* |
1844 | * Determine whether the specified vdev can be offlined/detached/removed | |
1845 | * without losing data. | |
1846 | */ | |
1847 | boolean_t | |
1848 | vdev_dtl_required(vdev_t *vd) | |
1849 | { | |
1850 | spa_t *spa = vd->vdev_spa; | |
1851 | vdev_t *tvd = vd->vdev_top; | |
1852 | uint8_t cant_read = vd->vdev_cant_read; | |
1853 | boolean_t required; | |
1854 | ||
1855 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1856 | ||
1857 | if (vd == spa->spa_root_vdev || vd == tvd) | |
1858 | return (B_TRUE); | |
1859 | ||
1860 | /* | |
1861 | * Temporarily mark the device as unreadable, and then determine | |
1862 | * whether this results in any DTL outages in the top-level vdev. | |
1863 | * If not, we can safely offline/detach/remove the device. | |
1864 | */ | |
1865 | vd->vdev_cant_read = B_TRUE; | |
1866 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1867 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
1868 | vd->vdev_cant_read = cant_read; | |
1869 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1870 | ||
572e2857 BB |
1871 | if (!required && zio_injection_enabled) |
1872 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
1873 | ||
fb5f0bc8 BB |
1874 | return (required); |
1875 | } | |
1876 | ||
b128c09f BB |
1877 | /* |
1878 | * Determine if resilver is needed, and if so the txg range. | |
1879 | */ | |
1880 | boolean_t | |
1881 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
1882 | { | |
1883 | boolean_t needed = B_FALSE; | |
1884 | uint64_t thismin = UINT64_MAX; | |
1885 | uint64_t thismax = 0; | |
d6320ddb | 1886 | int c; |
b128c09f BB |
1887 | |
1888 | if (vd->vdev_children == 0) { | |
1889 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1890 | if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && |
1891 | vdev_writeable(vd)) { | |
b128c09f BB |
1892 | space_seg_t *ss; |
1893 | ||
fb5f0bc8 | 1894 | ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f | 1895 | thismin = ss->ss_start - 1; |
fb5f0bc8 | 1896 | ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f BB |
1897 | thismax = ss->ss_end; |
1898 | needed = B_TRUE; | |
1899 | } | |
1900 | mutex_exit(&vd->vdev_dtl_lock); | |
1901 | } else { | |
d6320ddb | 1902 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
1903 | vdev_t *cvd = vd->vdev_child[c]; |
1904 | uint64_t cmin, cmax; | |
1905 | ||
1906 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
1907 | thismin = MIN(thismin, cmin); | |
1908 | thismax = MAX(thismax, cmax); | |
1909 | needed = B_TRUE; | |
1910 | } | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | if (needed && minp) { | |
1915 | *minp = thismin; | |
1916 | *maxp = thismax; | |
1917 | } | |
1918 | return (needed); | |
1919 | } | |
1920 | ||
34dc7c2f BB |
1921 | void |
1922 | vdev_load(vdev_t *vd) | |
1923 | { | |
d6320ddb BB |
1924 | int c; |
1925 | ||
34dc7c2f BB |
1926 | /* |
1927 | * Recursively load all children. | |
1928 | */ | |
d6320ddb | 1929 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1930 | vdev_load(vd->vdev_child[c]); |
1931 | ||
1932 | /* | |
1933 | * If this is a top-level vdev, initialize its metaslabs. | |
1934 | */ | |
428870ff | 1935 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
1936 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
1937 | vdev_metaslab_init(vd, 0) != 0)) | |
1938 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1939 | VDEV_AUX_CORRUPT_DATA); | |
1940 | ||
1941 | /* | |
1942 | * If this is a leaf vdev, load its DTL. | |
1943 | */ | |
1944 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
1945 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1946 | VDEV_AUX_CORRUPT_DATA); | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * The special vdev case is used for hot spares and l2cache devices. Its | |
1951 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
1952 | * we make sure that we can open the underlying device, then try to read the | |
1953 | * label, and make sure that the label is sane and that it hasn't been | |
1954 | * repurposed to another pool. | |
1955 | */ | |
1956 | int | |
1957 | vdev_validate_aux(vdev_t *vd) | |
1958 | { | |
1959 | nvlist_t *label; | |
1960 | uint64_t guid, version; | |
1961 | uint64_t state; | |
1962 | ||
b128c09f BB |
1963 | if (!vdev_readable(vd)) |
1964 | return (0); | |
1965 | ||
34dc7c2f BB |
1966 | if ((label = vdev_label_read_config(vd)) == NULL) { |
1967 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1968 | VDEV_AUX_CORRUPT_DATA); | |
1969 | return (-1); | |
1970 | } | |
1971 | ||
1972 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
1973 | version > SPA_VERSION || | |
1974 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || | |
1975 | guid != vd->vdev_guid || | |
1976 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
1977 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1978 | VDEV_AUX_CORRUPT_DATA); | |
1979 | nvlist_free(label); | |
1980 | return (-1); | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * We don't actually check the pool state here. If it's in fact in | |
1985 | * use by another pool, we update this fact on the fly when requested. | |
1986 | */ | |
1987 | nvlist_free(label); | |
1988 | return (0); | |
1989 | } | |
1990 | ||
428870ff BB |
1991 | void |
1992 | vdev_remove(vdev_t *vd, uint64_t txg) | |
1993 | { | |
1994 | spa_t *spa = vd->vdev_spa; | |
1995 | objset_t *mos = spa->spa_meta_objset; | |
1996 | dmu_tx_t *tx; | |
d6320ddb | 1997 | int m; |
428870ff BB |
1998 | |
1999 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
2000 | ||
2001 | if (vd->vdev_dtl_smo.smo_object) { | |
2002 | ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); | |
2003 | (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); | |
2004 | vd->vdev_dtl_smo.smo_object = 0; | |
2005 | } | |
2006 | ||
2007 | if (vd->vdev_ms != NULL) { | |
d6320ddb | 2008 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2009 | metaslab_t *msp = vd->vdev_ms[m]; |
2010 | ||
2011 | if (msp == NULL || msp->ms_smo.smo_object == 0) | |
2012 | continue; | |
2013 | ||
2014 | ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); | |
2015 | (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); | |
2016 | msp->ms_smo.smo_object = 0; | |
2017 | } | |
2018 | } | |
2019 | ||
2020 | if (vd->vdev_ms_array) { | |
2021 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2022 | vd->vdev_ms_array = 0; | |
2023 | vd->vdev_ms_shift = 0; | |
2024 | } | |
2025 | dmu_tx_commit(tx); | |
2026 | } | |
2027 | ||
34dc7c2f BB |
2028 | void |
2029 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2030 | { | |
2031 | metaslab_t *msp; | |
428870ff BB |
2032 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2033 | ||
2034 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2035 | |
c65aa5b2 | 2036 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2037 | metaslab_sync_done(msp, txg); |
428870ff BB |
2038 | |
2039 | if (reassess) | |
2040 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2041 | } |
2042 | ||
2043 | void | |
2044 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2045 | { | |
2046 | spa_t *spa = vd->vdev_spa; | |
2047 | vdev_t *lvd; | |
2048 | metaslab_t *msp; | |
2049 | dmu_tx_t *tx; | |
2050 | ||
428870ff BB |
2051 | ASSERT(!vd->vdev_ishole); |
2052 | ||
34dc7c2f BB |
2053 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2054 | ASSERT(vd == vd->vdev_top); | |
2055 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2056 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2057 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2058 | ASSERT(vd->vdev_ms_array != 0); | |
2059 | vdev_config_dirty(vd); | |
2060 | dmu_tx_commit(tx); | |
2061 | } | |
2062 | ||
428870ff BB |
2063 | /* |
2064 | * Remove the metadata associated with this vdev once it's empty. | |
2065 | */ | |
2066 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2067 | vdev_remove(vd, txg); | |
2068 | ||
34dc7c2f BB |
2069 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2070 | metaslab_sync(msp, txg); | |
2071 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2072 | } | |
2073 | ||
2074 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2075 | vdev_dtl_sync(lvd, txg); | |
2076 | ||
2077 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2078 | } | |
2079 | ||
2080 | uint64_t | |
2081 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2082 | { | |
2083 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2084 | } | |
2085 | ||
34dc7c2f BB |
2086 | /* |
2087 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2088 | * not be opened, and no I/O is attempted. | |
2089 | */ | |
2090 | int | |
428870ff | 2091 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2092 | { |
572e2857 | 2093 | vdev_t *vd, *tvd; |
34dc7c2f | 2094 | |
428870ff | 2095 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2096 | |
b128c09f BB |
2097 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2098 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2099 | |
34dc7c2f | 2100 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2101 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2102 | |
572e2857 BB |
2103 | tvd = vd->vdev_top; |
2104 | ||
428870ff BB |
2105 | /* |
2106 | * We don't directly use the aux state here, but if we do a | |
2107 | * vdev_reopen(), we need this value to be present to remember why we | |
2108 | * were faulted. | |
2109 | */ | |
2110 | vd->vdev_label_aux = aux; | |
2111 | ||
34dc7c2f BB |
2112 | /* |
2113 | * Faulted state takes precedence over degraded. | |
2114 | */ | |
428870ff | 2115 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2116 | vd->vdev_faulted = 1ULL; |
2117 | vd->vdev_degraded = 0ULL; | |
428870ff | 2118 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2119 | |
2120 | /* | |
428870ff BB |
2121 | * If this device has the only valid copy of the data, then |
2122 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2123 | */ |
572e2857 | 2124 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2125 | vd->vdev_degraded = 1ULL; |
2126 | vd->vdev_faulted = 0ULL; | |
2127 | ||
2128 | /* | |
2129 | * If we reopen the device and it's not dead, only then do we | |
2130 | * mark it degraded. | |
2131 | */ | |
572e2857 | 2132 | vdev_reopen(tvd); |
34dc7c2f | 2133 | |
428870ff BB |
2134 | if (vdev_readable(vd)) |
2135 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2136 | } |
2137 | ||
b128c09f | 2138 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2139 | } |
2140 | ||
2141 | /* | |
2142 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2143 | * user that something is wrong. The vdev continues to operate as normal as far | |
2144 | * as I/O is concerned. | |
2145 | */ | |
2146 | int | |
428870ff | 2147 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2148 | { |
b128c09f | 2149 | vdev_t *vd; |
34dc7c2f | 2150 | |
428870ff | 2151 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2152 | |
b128c09f BB |
2153 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2154 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2155 | |
34dc7c2f | 2156 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2157 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2158 | |
2159 | /* | |
2160 | * If the vdev is already faulted, then don't do anything. | |
2161 | */ | |
b128c09f BB |
2162 | if (vd->vdev_faulted || vd->vdev_degraded) |
2163 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2164 | |
2165 | vd->vdev_degraded = 1ULL; | |
2166 | if (!vdev_is_dead(vd)) | |
2167 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2168 | aux); |
34dc7c2f | 2169 | |
b128c09f | 2170 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2171 | } |
2172 | ||
2173 | /* | |
2174 | * Online the given vdev. If 'unspare' is set, it implies two things. First, | |
2175 | * any attached spare device should be detached when the device finishes | |
2176 | * resilvering. Second, the online should be treated like a 'test' online case, | |
2177 | * so no FMA events are generated if the device fails to open. | |
2178 | */ | |
2179 | int | |
b128c09f | 2180 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2181 | { |
9babb374 | 2182 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 2183 | |
428870ff | 2184 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2185 | |
b128c09f BB |
2186 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2187 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2188 | |
2189 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2190 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2191 | |
9babb374 | 2192 | tvd = vd->vdev_top; |
34dc7c2f BB |
2193 | vd->vdev_offline = B_FALSE; |
2194 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2195 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2196 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2197 | |
2198 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2199 | if (!vd->vdev_aux) { | |
2200 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2201 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2202 | } | |
2203 | ||
2204 | vdev_reopen(tvd); | |
34dc7c2f BB |
2205 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2206 | ||
9babb374 BB |
2207 | if (!vd->vdev_aux) { |
2208 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2209 | pvd->vdev_expanding = B_FALSE; | |
2210 | } | |
2211 | ||
34dc7c2f BB |
2212 | if (newstate) |
2213 | *newstate = vd->vdev_state; | |
2214 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2215 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2216 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2217 | vd->vdev_parent->vdev_child[0] == vd) | |
2218 | vd->vdev_unspare = B_TRUE; | |
2219 | ||
9babb374 BB |
2220 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2221 | ||
2222 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2223 | if (vd->vdev_aux) | |
2224 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2225 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2226 | } | |
fb5f0bc8 | 2227 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2228 | } |
2229 | ||
428870ff BB |
2230 | static int |
2231 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2232 | { |
9babb374 | 2233 | vdev_t *vd, *tvd; |
428870ff BB |
2234 | int error = 0; |
2235 | uint64_t generation; | |
2236 | metaslab_group_t *mg; | |
34dc7c2f | 2237 | |
428870ff BB |
2238 | top: |
2239 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2240 | |
b128c09f BB |
2241 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2242 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2243 | |
2244 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2245 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2246 | |
9babb374 | 2247 | tvd = vd->vdev_top; |
428870ff BB |
2248 | mg = tvd->vdev_mg; |
2249 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2250 | |
34dc7c2f BB |
2251 | /* |
2252 | * If the device isn't already offline, try to offline it. | |
2253 | */ | |
2254 | if (!vd->vdev_offline) { | |
2255 | /* | |
fb5f0bc8 | 2256 | * If this device has the only valid copy of some data, |
9babb374 BB |
2257 | * don't allow it to be offlined. Log devices are always |
2258 | * expendable. | |
34dc7c2f | 2259 | */ |
9babb374 BB |
2260 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2261 | vdev_dtl_required(vd)) | |
b128c09f | 2262 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2263 | |
428870ff BB |
2264 | /* |
2265 | * If the top-level is a slog and it has had allocations | |
2266 | * then proceed. We check that the vdev's metaslab group | |
2267 | * is not NULL since it's possible that we may have just | |
2268 | * added this vdev but not yet initialized its metaslabs. | |
2269 | */ | |
2270 | if (tvd->vdev_islog && mg != NULL) { | |
2271 | /* | |
2272 | * Prevent any future allocations. | |
2273 | */ | |
2274 | metaslab_group_passivate(mg); | |
2275 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2276 | ||
2277 | error = spa_offline_log(spa); | |
2278 | ||
2279 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2280 | ||
2281 | /* | |
2282 | * Check to see if the config has changed. | |
2283 | */ | |
2284 | if (error || generation != spa->spa_config_generation) { | |
2285 | metaslab_group_activate(mg); | |
2286 | if (error) | |
2287 | return (spa_vdev_state_exit(spa, | |
2288 | vd, error)); | |
2289 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2290 | goto top; | |
2291 | } | |
2292 | ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); | |
2293 | } | |
2294 | ||
34dc7c2f BB |
2295 | /* |
2296 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2297 | * If the top-level vdev is a log device then just offline |
2298 | * it. Otherwise, if this action results in the top-level | |
2299 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2300 | */ |
2301 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2302 | vdev_reopen(tvd); |
2303 | ||
2304 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2305 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2306 | vd->vdev_offline = B_FALSE; |
9babb374 | 2307 | vdev_reopen(tvd); |
b128c09f | 2308 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2309 | } |
428870ff BB |
2310 | |
2311 | /* | |
2312 | * Add the device back into the metaslab rotor so that | |
2313 | * once we online the device it's open for business. | |
2314 | */ | |
2315 | if (tvd->vdev_islog && mg != NULL) | |
2316 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2317 | } |
2318 | ||
b128c09f | 2319 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2320 | |
428870ff BB |
2321 | return (spa_vdev_state_exit(spa, vd, 0)); |
2322 | } | |
9babb374 | 2323 | |
428870ff BB |
2324 | int |
2325 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2326 | { | |
2327 | int error; | |
9babb374 | 2328 | |
428870ff BB |
2329 | mutex_enter(&spa->spa_vdev_top_lock); |
2330 | error = vdev_offline_locked(spa, guid, flags); | |
2331 | mutex_exit(&spa->spa_vdev_top_lock); | |
2332 | ||
2333 | return (error); | |
34dc7c2f BB |
2334 | } |
2335 | ||
2336 | /* | |
2337 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2338 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2339 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2340 | */ |
2341 | void | |
b128c09f | 2342 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2343 | { |
b128c09f | 2344 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2345 | int c; |
b128c09f BB |
2346 | |
2347 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2348 | |
2349 | if (vd == NULL) | |
b128c09f | 2350 | vd = rvd; |
34dc7c2f BB |
2351 | |
2352 | vd->vdev_stat.vs_read_errors = 0; | |
2353 | vd->vdev_stat.vs_write_errors = 0; | |
2354 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2355 | |
d6320ddb | 2356 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2357 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2358 | |
2359 | /* | |
b128c09f BB |
2360 | * If we're in the FAULTED state or have experienced failed I/O, then |
2361 | * clear the persistent state and attempt to reopen the device. We | |
2362 | * also mark the vdev config dirty, so that the new faulted state is | |
2363 | * written out to disk. | |
34dc7c2f | 2364 | */ |
b128c09f BB |
2365 | if (vd->vdev_faulted || vd->vdev_degraded || |
2366 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2367 | ||
428870ff BB |
2368 | /* |
2369 | * When reopening in reponse to a clear event, it may be due to | |
2370 | * a fmadm repair request. In this case, if the device is | |
2371 | * still broken, we want to still post the ereport again. | |
2372 | */ | |
2373 | vd->vdev_forcefault = B_TRUE; | |
2374 | ||
572e2857 | 2375 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2376 | vd->vdev_cant_read = B_FALSE; |
2377 | vd->vdev_cant_write = B_FALSE; | |
2378 | ||
572e2857 | 2379 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2380 | |
428870ff BB |
2381 | vd->vdev_forcefault = B_FALSE; |
2382 | ||
572e2857 | 2383 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2384 | vdev_state_dirty(vd->vdev_top); |
2385 | ||
2386 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2387 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2388 | ||
2389 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); | |
2390 | } | |
428870ff BB |
2391 | |
2392 | /* | |
2393 | * When clearing a FMA-diagnosed fault, we always want to | |
2394 | * unspare the device, as we assume that the original spare was | |
2395 | * done in response to the FMA fault. | |
2396 | */ | |
2397 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2398 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2399 | vd->vdev_parent->vdev_child[0] == vd) | |
2400 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2401 | } |
2402 | ||
b128c09f BB |
2403 | boolean_t |
2404 | vdev_is_dead(vdev_t *vd) | |
2405 | { | |
428870ff BB |
2406 | /* |
2407 | * Holes and missing devices are always considered "dead". | |
2408 | * This simplifies the code since we don't have to check for | |
2409 | * these types of devices in the various code paths. | |
2410 | * Instead we rely on the fact that we skip over dead devices | |
2411 | * before issuing I/O to them. | |
2412 | */ | |
2413 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2414 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2415 | } |
2416 | ||
2417 | boolean_t | |
34dc7c2f BB |
2418 | vdev_readable(vdev_t *vd) |
2419 | { | |
b128c09f | 2420 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2421 | } |
2422 | ||
b128c09f | 2423 | boolean_t |
34dc7c2f BB |
2424 | vdev_writeable(vdev_t *vd) |
2425 | { | |
b128c09f | 2426 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2427 | } |
2428 | ||
b128c09f BB |
2429 | boolean_t |
2430 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2431 | { |
fb5f0bc8 BB |
2432 | uint64_t state = vd->vdev_state; |
2433 | ||
b128c09f | 2434 | /* |
fb5f0bc8 | 2435 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2436 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2437 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2438 | * the proper locks. Note that we have to get the vdev state |
2439 | * in a local variable because although it changes atomically, | |
2440 | * we're asking two separate questions about it. | |
b128c09f | 2441 | */ |
fb5f0bc8 | 2442 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
428870ff | 2443 | !vd->vdev_cant_write && !vd->vdev_ishole); |
34dc7c2f BB |
2444 | } |
2445 | ||
b128c09f BB |
2446 | boolean_t |
2447 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2448 | { |
b128c09f | 2449 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2450 | |
b128c09f BB |
2451 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2452 | return (B_FALSE); | |
34dc7c2f | 2453 | |
b128c09f BB |
2454 | if (zio->io_type == ZIO_TYPE_READ) |
2455 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2456 | |
b128c09f BB |
2457 | if (zio->io_type == ZIO_TYPE_WRITE) |
2458 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2459 | |
b128c09f | 2460 | return (B_TRUE); |
34dc7c2f BB |
2461 | } |
2462 | ||
2463 | /* | |
2464 | * Get statistics for the given vdev. | |
2465 | */ | |
2466 | void | |
2467 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2468 | { | |
2469 | vdev_t *rvd = vd->vdev_spa->spa_root_vdev; | |
d6320ddb | 2470 | int c, t; |
34dc7c2f BB |
2471 | |
2472 | mutex_enter(&vd->vdev_stat_lock); | |
2473 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2474 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2475 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2476 | vs->vs_rsize = vdev_get_min_asize(vd); |
2477 | if (vd->vdev_ops->vdev_op_leaf) | |
2478 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
34dc7c2f BB |
2479 | mutex_exit(&vd->vdev_stat_lock); |
2480 | ||
2481 | /* | |
2482 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2483 | * over all top-level vdevs (i.e. the direct children of the root). | |
2484 | */ | |
2485 | if (vd == rvd) { | |
d6320ddb | 2486 | for (c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2487 | vdev_t *cvd = rvd->vdev_child[c]; |
2488 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2489 | ||
2490 | mutex_enter(&vd->vdev_stat_lock); | |
d6320ddb | 2491 | for (t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2492 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2493 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2494 | } | |
428870ff | 2495 | cvs->vs_scan_removing = cvd->vdev_removing; |
34dc7c2f BB |
2496 | mutex_exit(&vd->vdev_stat_lock); |
2497 | } | |
2498 | } | |
2499 | } | |
2500 | ||
2501 | void | |
2502 | vdev_clear_stats(vdev_t *vd) | |
2503 | { | |
2504 | mutex_enter(&vd->vdev_stat_lock); | |
2505 | vd->vdev_stat.vs_space = 0; | |
2506 | vd->vdev_stat.vs_dspace = 0; | |
2507 | vd->vdev_stat.vs_alloc = 0; | |
2508 | mutex_exit(&vd->vdev_stat_lock); | |
2509 | } | |
2510 | ||
428870ff BB |
2511 | void |
2512 | vdev_scan_stat_init(vdev_t *vd) | |
2513 | { | |
2514 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2515 | int c; |
428870ff | 2516 | |
d6320ddb | 2517 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2518 | vdev_scan_stat_init(vd->vdev_child[c]); |
2519 | ||
2520 | mutex_enter(&vd->vdev_stat_lock); | |
2521 | vs->vs_scan_processed = 0; | |
2522 | mutex_exit(&vd->vdev_stat_lock); | |
2523 | } | |
2524 | ||
34dc7c2f | 2525 | void |
b128c09f | 2526 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2527 | { |
fb5f0bc8 BB |
2528 | spa_t *spa = zio->io_spa; |
2529 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2530 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2531 | vdev_t *pvd; |
2532 | uint64_t txg = zio->io_txg; | |
2533 | vdev_stat_t *vs = &vd->vdev_stat; | |
2534 | zio_type_t type = zio->io_type; | |
2535 | int flags = zio->io_flags; | |
2536 | ||
b128c09f BB |
2537 | /* |
2538 | * If this i/o is a gang leader, it didn't do any actual work. | |
2539 | */ | |
2540 | if (zio->io_gang_tree) | |
2541 | return; | |
2542 | ||
34dc7c2f | 2543 | if (zio->io_error == 0) { |
b128c09f BB |
2544 | /* |
2545 | * If this is a root i/o, don't count it -- we've already | |
2546 | * counted the top-level vdevs, and vdev_get_stats() will | |
2547 | * aggregate them when asked. This reduces contention on | |
2548 | * the root vdev_stat_lock and implicitly handles blocks | |
2549 | * that compress away to holes, for which there is no i/o. | |
2550 | * (Holes never create vdev children, so all the counters | |
2551 | * remain zero, which is what we want.) | |
2552 | * | |
2553 | * Note: this only applies to successful i/o (io_error == 0) | |
2554 | * because unlike i/o counts, errors are not additive. | |
2555 | * When reading a ditto block, for example, failure of | |
2556 | * one top-level vdev does not imply a root-level error. | |
2557 | */ | |
2558 | if (vd == rvd) | |
2559 | return; | |
2560 | ||
2561 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2562 | |
2563 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2564 | return; | |
2565 | ||
2566 | mutex_enter(&vd->vdev_stat_lock); | |
2567 | ||
b128c09f | 2568 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 2569 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
2570 | dsl_scan_phys_t *scn_phys = |
2571 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2572 | uint64_t *processed = &scn_phys->scn_processed; | |
2573 | ||
2574 | /* XXX cleanup? */ | |
2575 | if (vd->vdev_ops->vdev_op_leaf) | |
2576 | atomic_add_64(processed, psize); | |
2577 | vs->vs_scan_processed += psize; | |
2578 | } | |
2579 | ||
fb5f0bc8 | 2580 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2581 | vs->vs_self_healed += psize; |
34dc7c2f | 2582 | } |
fb5f0bc8 BB |
2583 | |
2584 | vs->vs_ops[type]++; | |
2585 | vs->vs_bytes[type] += psize; | |
2586 | ||
2587 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2588 | return; |
2589 | } | |
2590 | ||
2591 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2592 | return; | |
2593 | ||
9babb374 BB |
2594 | /* |
2595 | * If this is an I/O error that is going to be retried, then ignore the | |
2596 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2597 | * hard errors, when in reality they can happen for any number of | |
2598 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2599 | */ | |
2600 | if (zio->io_error == EIO && | |
2601 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2602 | return; | |
2603 | ||
428870ff BB |
2604 | /* |
2605 | * Intent logs writes won't propagate their error to the root | |
2606 | * I/O so don't mark these types of failures as pool-level | |
2607 | * errors. | |
2608 | */ | |
2609 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2610 | return; | |
2611 | ||
b128c09f | 2612 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2613 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2614 | if (zio->io_error == ECKSUM) |
2615 | vs->vs_checksum_errors++; | |
2616 | else | |
2617 | vs->vs_read_errors++; | |
34dc7c2f | 2618 | } |
9babb374 | 2619 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2620 | vs->vs_write_errors++; |
2621 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2622 | |
fb5f0bc8 BB |
2623 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2624 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 2625 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 2626 | spa->spa_claiming)) { |
fb5f0bc8 | 2627 | /* |
428870ff BB |
2628 | * This is either a normal write (not a repair), or it's |
2629 | * a repair induced by the scrub thread, or it's a repair | |
2630 | * made by zil_claim() during spa_load() in the first txg. | |
2631 | * In the normal case, we commit the DTL change in the same | |
2632 | * txg as the block was born. In the scrub-induced repair | |
2633 | * case, we know that scrubs run in first-pass syncing context, | |
2634 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2635 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
2636 | * |
2637 | * We currently do not make DTL entries for failed spontaneous | |
2638 | * self-healing writes triggered by normal (non-scrubbing) | |
2639 | * reads, because we have no transactional context in which to | |
2640 | * do so -- and it's not clear that it'd be desirable anyway. | |
2641 | */ | |
2642 | if (vd->vdev_ops->vdev_op_leaf) { | |
2643 | uint64_t commit_txg = txg; | |
572e2857 | 2644 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
2645 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
2646 | ASSERT(spa_sync_pass(spa) == 1); | |
2647 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
2648 | commit_txg = spa_syncing_txg(spa); |
2649 | } else if (spa->spa_claiming) { | |
2650 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2651 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 2652 | } |
428870ff | 2653 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 2654 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 2655 | return; |
fb5f0bc8 BB |
2656 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2657 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2658 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2659 | } |
fb5f0bc8 BB |
2660 | if (vd != rvd) |
2661 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2662 | } |
2663 | } | |
2664 | ||
34dc7c2f | 2665 | /* |
428870ff BB |
2666 | * Update the in-core space usage stats for this vdev, its metaslab class, |
2667 | * and the root vdev. | |
34dc7c2f BB |
2668 | */ |
2669 | void | |
428870ff BB |
2670 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
2671 | int64_t space_delta) | |
34dc7c2f BB |
2672 | { |
2673 | int64_t dspace_delta = space_delta; | |
2674 | spa_t *spa = vd->vdev_spa; | |
2675 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
2676 | metaslab_group_t *mg = vd->vdev_mg; |
2677 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
2678 | |
2679 | ASSERT(vd == vd->vdev_top); | |
2680 | ||
2681 | /* | |
2682 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2683 | * factor. We must calculate this here and not at the root vdev | |
2684 | * because the root vdev's psize-to-asize is simply the max of its | |
2685 | * childrens', thus not accurate enough for us. | |
2686 | */ | |
2687 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2688 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2689 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2690 | vd->vdev_deflate_ratio; | |
2691 | ||
2692 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 2693 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2694 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2695 | vd->vdev_stat.vs_dspace += dspace_delta; |
2696 | mutex_exit(&vd->vdev_stat_lock); | |
2697 | ||
428870ff | 2698 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 2699 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 2700 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2701 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2702 | rvd->vdev_stat.vs_dspace += dspace_delta; |
2703 | mutex_exit(&rvd->vdev_stat_lock); | |
2704 | } | |
428870ff BB |
2705 | |
2706 | if (mc != NULL) { | |
2707 | ASSERT(rvd == vd->vdev_parent); | |
2708 | ASSERT(vd->vdev_ms_count != 0); | |
2709 | ||
2710 | metaslab_class_space_update(mc, | |
2711 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2712 | } | |
34dc7c2f BB |
2713 | } |
2714 | ||
2715 | /* | |
2716 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2717 | * so that it will be written out next time the vdev configuration is synced. | |
2718 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2719 | */ | |
2720 | void | |
2721 | vdev_config_dirty(vdev_t *vd) | |
2722 | { | |
2723 | spa_t *spa = vd->vdev_spa; | |
2724 | vdev_t *rvd = spa->spa_root_vdev; | |
2725 | int c; | |
2726 | ||
572e2857 BB |
2727 | ASSERT(spa_writeable(spa)); |
2728 | ||
34dc7c2f | 2729 | /* |
9babb374 BB |
2730 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2731 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2732 | */ |
2733 | if (vd->vdev_aux != NULL) { | |
2734 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2735 | nvlist_t **aux; | |
2736 | uint_t naux; | |
2737 | ||
2738 | for (c = 0; c < sav->sav_count; c++) { | |
2739 | if (sav->sav_vdevs[c] == vd) | |
2740 | break; | |
2741 | } | |
2742 | ||
2743 | if (c == sav->sav_count) { | |
2744 | /* | |
2745 | * We're being removed. There's nothing more to do. | |
2746 | */ | |
2747 | ASSERT(sav->sav_sync == B_TRUE); | |
2748 | return; | |
2749 | } | |
2750 | ||
2751 | sav->sav_sync = B_TRUE; | |
2752 | ||
9babb374 BB |
2753 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2754 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2755 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2756 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2757 | } | |
b128c09f BB |
2758 | |
2759 | ASSERT(c < naux); | |
2760 | ||
2761 | /* | |
2762 | * Setting the nvlist in the middle if the array is a little | |
2763 | * sketchy, but it will work. | |
2764 | */ | |
2765 | nvlist_free(aux[c]); | |
428870ff | 2766 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
2767 | |
2768 | return; | |
2769 | } | |
2770 | ||
2771 | /* | |
2772 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2773 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2774 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2775 | * so this is sufficient to ensure mutual exclusion. |
2776 | */ | |
b128c09f BB |
2777 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2778 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2779 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2780 | |
2781 | if (vd == rvd) { | |
2782 | for (c = 0; c < rvd->vdev_children; c++) | |
2783 | vdev_config_dirty(rvd->vdev_child[c]); | |
2784 | } else { | |
2785 | ASSERT(vd == vd->vdev_top); | |
2786 | ||
428870ff BB |
2787 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
2788 | !vd->vdev_ishole) | |
b128c09f | 2789 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
2790 | } |
2791 | } | |
2792 | ||
2793 | void | |
2794 | vdev_config_clean(vdev_t *vd) | |
2795 | { | |
2796 | spa_t *spa = vd->vdev_spa; | |
2797 | ||
b128c09f BB |
2798 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2799 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2800 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2801 | |
b128c09f BB |
2802 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2803 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2804 | } |
2805 | ||
b128c09f BB |
2806 | /* |
2807 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2808 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2809 | * the state changes from larger config changes because they require | |
2810 | * much less locking, and are often needed for administrative actions. | |
2811 | */ | |
2812 | void | |
2813 | vdev_state_dirty(vdev_t *vd) | |
2814 | { | |
2815 | spa_t *spa = vd->vdev_spa; | |
2816 | ||
572e2857 | 2817 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
2818 | ASSERT(vd == vd->vdev_top); |
2819 | ||
2820 | /* | |
2821 | * The state list is protected by the SCL_STATE lock. The caller | |
2822 | * must either hold SCL_STATE as writer, or must be the sync thread | |
2823 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
2824 | * so this is sufficient to ensure mutual exclusion. | |
2825 | */ | |
2826 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2827 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2828 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2829 | ||
428870ff | 2830 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
2831 | list_insert_head(&spa->spa_state_dirty_list, vd); |
2832 | } | |
2833 | ||
2834 | void | |
2835 | vdev_state_clean(vdev_t *vd) | |
2836 | { | |
2837 | spa_t *spa = vd->vdev_spa; | |
2838 | ||
2839 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2840 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2841 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2842 | ||
2843 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
2844 | list_remove(&spa->spa_state_dirty_list, vd); | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Propagate vdev state up from children to parent. | |
2849 | */ | |
34dc7c2f BB |
2850 | void |
2851 | vdev_propagate_state(vdev_t *vd) | |
2852 | { | |
fb5f0bc8 BB |
2853 | spa_t *spa = vd->vdev_spa; |
2854 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
2855 | int degraded = 0, faulted = 0; |
2856 | int corrupted = 0; | |
34dc7c2f | 2857 | vdev_t *child; |
d6320ddb | 2858 | int c; |
34dc7c2f BB |
2859 | |
2860 | if (vd->vdev_children > 0) { | |
d6320ddb | 2861 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 2862 | child = vd->vdev_child[c]; |
b128c09f | 2863 | |
428870ff BB |
2864 | /* |
2865 | * Don't factor holes into the decision. | |
2866 | */ | |
2867 | if (child->vdev_ishole) | |
2868 | continue; | |
2869 | ||
b128c09f | 2870 | if (!vdev_readable(child) || |
fb5f0bc8 | 2871 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
2872 | /* |
2873 | * Root special: if there is a top-level log | |
2874 | * device, treat the root vdev as if it were | |
2875 | * degraded. | |
2876 | */ | |
2877 | if (child->vdev_islog && vd == rvd) | |
2878 | degraded++; | |
2879 | else | |
2880 | faulted++; | |
2881 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 2882 | degraded++; |
b128c09f | 2883 | } |
34dc7c2f BB |
2884 | |
2885 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
2886 | corrupted++; | |
2887 | } | |
2888 | ||
2889 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
2890 | ||
2891 | /* | |
b128c09f | 2892 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
2893 | * opened due to corrupted metadata, then propagate the root |
2894 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
2895 | * replicas'. | |
2896 | */ | |
2897 | if (corrupted && vd == rvd && | |
2898 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
2899 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2900 | VDEV_AUX_CORRUPT_DATA); | |
2901 | } | |
2902 | ||
b128c09f | 2903 | if (vd->vdev_parent) |
34dc7c2f BB |
2904 | vdev_propagate_state(vd->vdev_parent); |
2905 | } | |
2906 | ||
2907 | /* | |
2908 | * Set a vdev's state. If this is during an open, we don't update the parent | |
2909 | * state, because we're in the process of opening children depth-first. | |
2910 | * Otherwise, we propagate the change to the parent. | |
2911 | * | |
2912 | * If this routine places a device in a faulted state, an appropriate ereport is | |
2913 | * generated. | |
2914 | */ | |
2915 | void | |
2916 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
2917 | { | |
2918 | uint64_t save_state; | |
b128c09f | 2919 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
2920 | |
2921 | if (state == vd->vdev_state) { | |
2922 | vd->vdev_stat.vs_aux = aux; | |
2923 | return; | |
2924 | } | |
2925 | ||
2926 | save_state = vd->vdev_state; | |
2927 | ||
2928 | vd->vdev_state = state; | |
2929 | vd->vdev_stat.vs_aux = aux; | |
2930 | ||
2931 | /* | |
2932 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
2933 | * always close the underlying device unless the device has requested |
2934 | * a delayed close (i.e. we're about to remove or fault the device). | |
2935 | * Otherwise, we keep accessible but invalid devices open forever. | |
2936 | * We don't call vdev_close() itself, because that implies some extra | |
2937 | * checks (offline, etc) that we don't want here. This is limited to | |
2938 | * leaf devices, because otherwise closing the device will affect other | |
2939 | * children. | |
34dc7c2f | 2940 | */ |
428870ff BB |
2941 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
2942 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
2943 | vd->vdev_ops->vdev_op_close(vd); |
2944 | ||
428870ff BB |
2945 | /* |
2946 | * If we have brought this vdev back into service, we need | |
2947 | * to notify fmd so that it can gracefully repair any outstanding | |
2948 | * cases due to a missing device. We do this in all cases, even those | |
2949 | * that probably don't correlate to a repaired fault. This is sure to | |
2950 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
2951 | * this is a transient state it's OK, as the retire agent will | |
2952 | * double-check the state of the vdev before repairing it. | |
2953 | */ | |
2954 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
2955 | vd->vdev_prevstate != state) | |
2956 | zfs_post_state_change(spa, vd); | |
2957 | ||
34dc7c2f BB |
2958 | if (vd->vdev_removed && |
2959 | state == VDEV_STATE_CANT_OPEN && | |
2960 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
2961 | /* | |
2962 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
2963 | * device was previously marked removed and someone attempted to | |
2964 | * reopen it. If this failed due to a nonexistent device, then | |
2965 | * keep the device in the REMOVED state. We also let this be if | |
2966 | * it is one of our special test online cases, which is only | |
2967 | * attempting to online the device and shouldn't generate an FMA | |
2968 | * fault. | |
2969 | */ | |
2970 | vd->vdev_state = VDEV_STATE_REMOVED; | |
2971 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
2972 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
2973 | vd->vdev_removed = B_TRUE; |
2974 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
2975 | /* | |
572e2857 BB |
2976 | * If we fail to open a vdev during an import or recovery, we |
2977 | * mark it as "not available", which signifies that it was | |
2978 | * never there to begin with. Failure to open such a device | |
2979 | * is not considered an error. | |
34dc7c2f | 2980 | */ |
572e2857 BB |
2981 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
2982 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
2983 | vd->vdev_ops->vdev_op_leaf) |
2984 | vd->vdev_not_present = 1; | |
2985 | ||
2986 | /* | |
2987 | * Post the appropriate ereport. If the 'prevstate' field is | |
2988 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
2989 | * that this is part of a vdev_reopen(). In this case, we don't | |
2990 | * want to post the ereport if the device was already in the | |
2991 | * CANT_OPEN state beforehand. | |
2992 | * | |
2993 | * If the 'checkremove' flag is set, then this is an attempt to | |
2994 | * online the device in response to an insertion event. If we | |
2995 | * hit this case, then we have detected an insertion event for a | |
2996 | * faulted or offline device that wasn't in the removed state. | |
2997 | * In this scenario, we don't post an ereport because we are | |
2998 | * about to replace the device, or attempt an online with | |
2999 | * vdev_forcefault, which will generate the fault for us. | |
3000 | */ | |
3001 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3002 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3003 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3004 | const char *class; |
3005 | ||
3006 | switch (aux) { | |
3007 | case VDEV_AUX_OPEN_FAILED: | |
3008 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3009 | break; | |
3010 | case VDEV_AUX_CORRUPT_DATA: | |
3011 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3012 | break; | |
3013 | case VDEV_AUX_NO_REPLICAS: | |
3014 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3015 | break; | |
3016 | case VDEV_AUX_BAD_GUID_SUM: | |
3017 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3018 | break; | |
3019 | case VDEV_AUX_TOO_SMALL: | |
3020 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3021 | break; | |
3022 | case VDEV_AUX_BAD_LABEL: | |
3023 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3024 | break; | |
3025 | default: | |
3026 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3027 | } | |
3028 | ||
b128c09f | 3029 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3030 | } |
3031 | ||
3032 | /* Erase any notion of persistent removed state */ | |
3033 | vd->vdev_removed = B_FALSE; | |
3034 | } else { | |
3035 | vd->vdev_removed = B_FALSE; | |
3036 | } | |
3037 | ||
9babb374 BB |
3038 | if (!isopen && vd->vdev_parent) |
3039 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3040 | } |
b128c09f BB |
3041 | |
3042 | /* | |
3043 | * Check the vdev configuration to ensure that it's capable of supporting | |
3044 | * a root pool. Currently, we do not support RAID-Z or partial configuration. | |
3045 | * In addition, only a single top-level vdev is allowed and none of the leaves | |
3046 | * can be wholedisks. | |
3047 | */ | |
3048 | boolean_t | |
3049 | vdev_is_bootable(vdev_t *vd) | |
3050 | { | |
d6320ddb BB |
3051 | int c; |
3052 | ||
b128c09f BB |
3053 | if (!vd->vdev_ops->vdev_op_leaf) { |
3054 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3055 | ||
3056 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3057 | vd->vdev_children > 1) { | |
3058 | return (B_FALSE); | |
3059 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3060 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3061 | return (B_FALSE); | |
3062 | } | |
3063 | } else if (vd->vdev_wholedisk == 1) { | |
3064 | return (B_FALSE); | |
3065 | } | |
3066 | ||
d6320ddb | 3067 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3068 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3069 | return (B_FALSE); | |
3070 | } | |
3071 | return (B_TRUE); | |
3072 | } | |
9babb374 | 3073 | |
428870ff BB |
3074 | /* |
3075 | * Load the state from the original vdev tree (ovd) which | |
3076 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3077 | * vdev was offline or faulted then we transfer that state to the |
3078 | * device in the current vdev tree (nvd). | |
428870ff | 3079 | */ |
9babb374 | 3080 | void |
428870ff | 3081 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3082 | { |
d6320ddb | 3083 | int c; |
9babb374 | 3084 | |
572e2857 | 3085 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3086 | ASSERT(spa_config_held(nvd->vdev_spa, |
3087 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3088 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3089 | |
d6320ddb | 3090 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3091 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3092 | |
572e2857 | 3093 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3094 | /* |
572e2857 | 3095 | * Restore the persistent vdev state |
9babb374 | 3096 | */ |
428870ff | 3097 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3098 | nvd->vdev_faulted = ovd->vdev_faulted; |
3099 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3100 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3101 | } |
3102 | } | |
3103 | ||
572e2857 BB |
3104 | /* |
3105 | * Determine if a log device has valid content. If the vdev was | |
3106 | * removed or faulted in the MOS config then we know that | |
3107 | * the content on the log device has already been written to the pool. | |
3108 | */ | |
3109 | boolean_t | |
3110 | vdev_log_state_valid(vdev_t *vd) | |
3111 | { | |
d6320ddb BB |
3112 | int c; |
3113 | ||
572e2857 BB |
3114 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3115 | !vd->vdev_removed) | |
3116 | return (B_TRUE); | |
3117 | ||
d6320ddb | 3118 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3119 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3120 | return (B_TRUE); | |
3121 | ||
3122 | return (B_FALSE); | |
3123 | } | |
3124 | ||
9babb374 BB |
3125 | /* |
3126 | * Expand a vdev if possible. | |
3127 | */ | |
3128 | void | |
3129 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3130 | { | |
3131 | ASSERT(vd->vdev_top == vd); | |
3132 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3133 | ||
3134 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3135 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3136 | vdev_config_dirty(vd); | |
3137 | } | |
3138 | } | |
428870ff BB |
3139 | |
3140 | /* | |
3141 | * Split a vdev. | |
3142 | */ | |
3143 | void | |
3144 | vdev_split(vdev_t *vd) | |
3145 | { | |
3146 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3147 | ||
3148 | vdev_remove_child(pvd, vd); | |
3149 | vdev_compact_children(pvd); | |
3150 | ||
3151 | cvd = pvd->vdev_child[0]; | |
3152 | if (pvd->vdev_children == 1) { | |
3153 | vdev_remove_parent(cvd); | |
3154 | cvd->vdev_splitting = B_TRUE; | |
3155 | } | |
3156 | vdev_propagate_state(cvd); | |
3157 | } |