]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
3541dc6d | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
1bd201e7 | 25 | * Copyright (c) 2012 by Delphix. All rights reserved. |
34dc7c2f BB |
26 | */ |
27 | ||
34dc7c2f BB |
28 | #include <sys/zfs_context.h> |
29 | #include <sys/fm/fs/zfs.h> | |
30 | #include <sys/spa.h> | |
31 | #include <sys/spa_impl.h> | |
32 | #include <sys/dmu.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/vdev_impl.h> | |
35 | #include <sys/uberblock_impl.h> | |
36 | #include <sys/metaslab.h> | |
37 | #include <sys/metaslab_impl.h> | |
38 | #include <sys/space_map.h> | |
39 | #include <sys/zio.h> | |
40 | #include <sys/zap.h> | |
41 | #include <sys/fs/zfs.h> | |
b128c09f | 42 | #include <sys/arc.h> |
9babb374 | 43 | #include <sys/zil.h> |
428870ff | 44 | #include <sys/dsl_scan.h> |
34dc7c2f BB |
45 | |
46 | /* | |
47 | * Virtual device management. | |
48 | */ | |
49 | ||
50 | static vdev_ops_t *vdev_ops_table[] = { | |
51 | &vdev_root_ops, | |
52 | &vdev_raidz_ops, | |
53 | &vdev_mirror_ops, | |
54 | &vdev_replacing_ops, | |
55 | &vdev_spare_ops, | |
56 | &vdev_disk_ops, | |
57 | &vdev_file_ops, | |
58 | &vdev_missing_ops, | |
428870ff | 59 | &vdev_hole_ops, |
34dc7c2f BB |
60 | NULL |
61 | }; | |
62 | ||
b128c09f BB |
63 | /* maximum scrub/resilver I/O queue per leaf vdev */ |
64 | int zfs_scrub_limit = 10; | |
34dc7c2f BB |
65 | |
66 | /* | |
67 | * Given a vdev type, return the appropriate ops vector. | |
68 | */ | |
69 | static vdev_ops_t * | |
70 | vdev_getops(const char *type) | |
71 | { | |
72 | vdev_ops_t *ops, **opspp; | |
73 | ||
74 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
75 | if (strcmp(ops->vdev_op_type, type) == 0) | |
76 | break; | |
77 | ||
78 | return (ops); | |
79 | } | |
80 | ||
81 | /* | |
82 | * Default asize function: return the MAX of psize with the asize of | |
83 | * all children. This is what's used by anything other than RAID-Z. | |
84 | */ | |
85 | uint64_t | |
86 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
87 | { | |
88 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
89 | uint64_t csize; | |
d6320ddb | 90 | int c; |
34dc7c2f | 91 | |
d6320ddb | 92 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
93 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
94 | asize = MAX(asize, csize); | |
95 | } | |
96 | ||
97 | return (asize); | |
98 | } | |
99 | ||
100 | /* | |
9babb374 BB |
101 | * Get the minimum allocatable size. We define the allocatable size as |
102 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
103 | * replace or attach devices which don't have the same physical size but | |
104 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
105 | */ |
106 | uint64_t | |
9babb374 | 107 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 108 | { |
9babb374 | 109 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 110 | |
9babb374 | 111 | /* |
1bd201e7 | 112 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
113 | * just return our own asize. |
114 | */ | |
115 | if (pvd == NULL) | |
116 | return (vd->vdev_asize); | |
34dc7c2f BB |
117 | |
118 | /* | |
9babb374 BB |
119 | * The top-level vdev just returns the allocatable size rounded |
120 | * to the nearest metaslab. | |
34dc7c2f | 121 | */ |
9babb374 BB |
122 | if (vd == vd->vdev_top) |
123 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 124 | |
9babb374 BB |
125 | /* |
126 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
127 | * so each child must provide at least 1/Nth of its asize. | |
128 | */ | |
129 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
130 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 131 | |
9babb374 BB |
132 | return (pvd->vdev_min_asize); |
133 | } | |
134 | ||
135 | void | |
136 | vdev_set_min_asize(vdev_t *vd) | |
137 | { | |
d6320ddb | 138 | int c; |
9babb374 | 139 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 140 | |
d6320ddb | 141 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 142 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
143 | } |
144 | ||
145 | vdev_t * | |
146 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
147 | { | |
148 | vdev_t *rvd = spa->spa_root_vdev; | |
149 | ||
b128c09f | 150 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 151 | |
b128c09f BB |
152 | if (vdev < rvd->vdev_children) { |
153 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 154 | return (rvd->vdev_child[vdev]); |
b128c09f | 155 | } |
34dc7c2f BB |
156 | |
157 | return (NULL); | |
158 | } | |
159 | ||
160 | vdev_t * | |
161 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
162 | { | |
34dc7c2f | 163 | vdev_t *mvd; |
d6320ddb | 164 | int c; |
34dc7c2f BB |
165 | |
166 | if (vd->vdev_guid == guid) | |
167 | return (vd); | |
168 | ||
d6320ddb | 169 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
170 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
171 | NULL) | |
172 | return (mvd); | |
173 | ||
174 | return (NULL); | |
175 | } | |
176 | ||
177 | void | |
178 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
179 | { | |
180 | size_t oldsize, newsize; | |
181 | uint64_t id = cvd->vdev_id; | |
182 | vdev_t **newchild; | |
183 | ||
b128c09f | 184 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
185 | ASSERT(cvd->vdev_parent == NULL); |
186 | ||
187 | cvd->vdev_parent = pvd; | |
188 | ||
189 | if (pvd == NULL) | |
190 | return; | |
191 | ||
192 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
193 | ||
194 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
195 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
196 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
197 | ||
b8d06fca | 198 | newchild = kmem_zalloc(newsize, KM_PUSHPAGE); |
34dc7c2f BB |
199 | if (pvd->vdev_child != NULL) { |
200 | bcopy(pvd->vdev_child, newchild, oldsize); | |
201 | kmem_free(pvd->vdev_child, oldsize); | |
202 | } | |
203 | ||
204 | pvd->vdev_child = newchild; | |
205 | pvd->vdev_child[id] = cvd; | |
206 | ||
207 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
208 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
209 | ||
210 | /* | |
211 | * Walk up all ancestors to update guid sum. | |
212 | */ | |
213 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
214 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
215 | } |
216 | ||
217 | void | |
218 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
219 | { | |
220 | int c; | |
221 | uint_t id = cvd->vdev_id; | |
222 | ||
223 | ASSERT(cvd->vdev_parent == pvd); | |
224 | ||
225 | if (pvd == NULL) | |
226 | return; | |
227 | ||
228 | ASSERT(id < pvd->vdev_children); | |
229 | ASSERT(pvd->vdev_child[id] == cvd); | |
230 | ||
231 | pvd->vdev_child[id] = NULL; | |
232 | cvd->vdev_parent = NULL; | |
233 | ||
234 | for (c = 0; c < pvd->vdev_children; c++) | |
235 | if (pvd->vdev_child[c]) | |
236 | break; | |
237 | ||
238 | if (c == pvd->vdev_children) { | |
239 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
240 | pvd->vdev_child = NULL; | |
241 | pvd->vdev_children = 0; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Walk up all ancestors to update guid sum. | |
246 | */ | |
247 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
248 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
249 | } |
250 | ||
251 | /* | |
252 | * Remove any holes in the child array. | |
253 | */ | |
254 | void | |
255 | vdev_compact_children(vdev_t *pvd) | |
256 | { | |
257 | vdev_t **newchild, *cvd; | |
258 | int oldc = pvd->vdev_children; | |
9babb374 | 259 | int newc; |
d6320ddb | 260 | int c; |
34dc7c2f | 261 | |
b128c09f | 262 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 263 | |
d6320ddb | 264 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
265 | if (pvd->vdev_child[c]) |
266 | newc++; | |
267 | ||
b8d06fca | 268 | newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_PUSHPAGE); |
34dc7c2f | 269 | |
d6320ddb | 270 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
271 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
272 | newchild[newc] = cvd; | |
273 | cvd->vdev_id = newc++; | |
274 | } | |
275 | } | |
276 | ||
277 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
278 | pvd->vdev_child = newchild; | |
279 | pvd->vdev_children = newc; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Allocate and minimally initialize a vdev_t. | |
284 | */ | |
428870ff | 285 | vdev_t * |
34dc7c2f BB |
286 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
287 | { | |
288 | vdev_t *vd; | |
d6320ddb | 289 | int t; |
34dc7c2f | 290 | |
b8d06fca | 291 | vd = kmem_zalloc(sizeof (vdev_t), KM_PUSHPAGE); |
34dc7c2f BB |
292 | |
293 | if (spa->spa_root_vdev == NULL) { | |
294 | ASSERT(ops == &vdev_root_ops); | |
295 | spa->spa_root_vdev = vd; | |
3541dc6d | 296 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
297 | } |
298 | ||
428870ff | 299 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
300 | if (spa->spa_root_vdev == vd) { |
301 | /* | |
302 | * The root vdev's guid will also be the pool guid, | |
303 | * which must be unique among all pools. | |
304 | */ | |
428870ff | 305 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
306 | } else { |
307 | /* | |
308 | * Any other vdev's guid must be unique within the pool. | |
309 | */ | |
428870ff | 310 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
311 | } |
312 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
313 | } | |
314 | ||
315 | vd->vdev_spa = spa; | |
316 | vd->vdev_id = id; | |
317 | vd->vdev_guid = guid; | |
318 | vd->vdev_guid_sum = guid; | |
319 | vd->vdev_ops = ops; | |
320 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 321 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f | 322 | |
98f72a53 BB |
323 | list_link_init(&vd->vdev_config_dirty_node); |
324 | list_link_init(&vd->vdev_state_dirty_node); | |
34dc7c2f BB |
325 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); |
326 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
b128c09f | 327 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
d6320ddb | 328 | for (t = 0; t < DTL_TYPES; t++) { |
fb5f0bc8 BB |
329 | space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, |
330 | &vd->vdev_dtl_lock); | |
331 | } | |
34dc7c2f BB |
332 | txg_list_create(&vd->vdev_ms_list, |
333 | offsetof(struct metaslab, ms_txg_node)); | |
334 | txg_list_create(&vd->vdev_dtl_list, | |
335 | offsetof(struct vdev, vdev_dtl_node)); | |
336 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
337 | vdev_queue_init(vd); | |
338 | vdev_cache_init(vd); | |
339 | ||
340 | return (vd); | |
341 | } | |
342 | ||
343 | /* | |
344 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
345 | * creating a new vdev or loading an existing one - the behavior is slightly | |
346 | * different for each case. | |
347 | */ | |
348 | int | |
349 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
350 | int alloctype) | |
351 | { | |
352 | vdev_ops_t *ops; | |
353 | char *type; | |
354 | uint64_t guid = 0, islog, nparity; | |
355 | vdev_t *vd; | |
356 | ||
b128c09f | 357 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
358 | |
359 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
360 | return (EINVAL); | |
361 | ||
362 | if ((ops = vdev_getops(type)) == NULL) | |
363 | return (EINVAL); | |
364 | ||
365 | /* | |
366 | * If this is a load, get the vdev guid from the nvlist. | |
367 | * Otherwise, vdev_alloc_common() will generate one for us. | |
368 | */ | |
369 | if (alloctype == VDEV_ALLOC_LOAD) { | |
370 | uint64_t label_id; | |
371 | ||
372 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
373 | label_id != id) | |
374 | return (EINVAL); | |
375 | ||
376 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
377 | return (EINVAL); | |
378 | } else if (alloctype == VDEV_ALLOC_SPARE) { | |
379 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
380 | return (EINVAL); | |
381 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { | |
382 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
383 | return (EINVAL); | |
9babb374 BB |
384 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
385 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
386 | return (EINVAL); | |
34dc7c2f BB |
387 | } |
388 | ||
389 | /* | |
390 | * The first allocated vdev must be of type 'root'. | |
391 | */ | |
392 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
393 | return (EINVAL); | |
394 | ||
395 | /* | |
396 | * Determine whether we're a log vdev. | |
397 | */ | |
398 | islog = 0; | |
399 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
400 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
401 | return (ENOTSUP); | |
402 | ||
428870ff BB |
403 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
404 | return (ENOTSUP); | |
405 | ||
34dc7c2f BB |
406 | /* |
407 | * Set the nparity property for RAID-Z vdevs. | |
408 | */ | |
409 | nparity = -1ULL; | |
410 | if (ops == &vdev_raidz_ops) { | |
411 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
412 | &nparity) == 0) { | |
428870ff | 413 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
34dc7c2f BB |
414 | return (EINVAL); |
415 | /* | |
45d1cae3 BB |
416 | * Previous versions could only support 1 or 2 parity |
417 | * device. | |
34dc7c2f | 418 | */ |
45d1cae3 BB |
419 | if (nparity > 1 && |
420 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
421 | return (ENOTSUP); | |
422 | if (nparity > 2 && | |
423 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
34dc7c2f BB |
424 | return (ENOTSUP); |
425 | } else { | |
426 | /* | |
427 | * We require the parity to be specified for SPAs that | |
428 | * support multiple parity levels. | |
429 | */ | |
45d1cae3 | 430 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
34dc7c2f BB |
431 | return (EINVAL); |
432 | /* | |
433 | * Otherwise, we default to 1 parity device for RAID-Z. | |
434 | */ | |
435 | nparity = 1; | |
436 | } | |
437 | } else { | |
438 | nparity = 0; | |
439 | } | |
440 | ASSERT(nparity != -1ULL); | |
441 | ||
442 | vd = vdev_alloc_common(spa, id, guid, ops); | |
443 | ||
444 | vd->vdev_islog = islog; | |
445 | vd->vdev_nparity = nparity; | |
446 | ||
447 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
448 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
449 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
450 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
451 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
452 | &vd->vdev_physpath) == 0) | |
453 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
9babb374 BB |
454 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
455 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
456 | |
457 | /* | |
458 | * Set the whole_disk property. If it's not specified, leave the value | |
459 | * as -1. | |
460 | */ | |
461 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
462 | &vd->vdev_wholedisk) != 0) | |
463 | vd->vdev_wholedisk = -1ULL; | |
464 | ||
465 | /* | |
466 | * Look for the 'not present' flag. This will only be set if the device | |
467 | * was not present at the time of import. | |
468 | */ | |
9babb374 BB |
469 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
470 | &vd->vdev_not_present); | |
34dc7c2f BB |
471 | |
472 | /* | |
473 | * Get the alignment requirement. | |
474 | */ | |
475 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
476 | ||
428870ff BB |
477 | /* |
478 | * Retrieve the vdev creation time. | |
479 | */ | |
480 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
481 | &vd->vdev_crtxg); | |
482 | ||
34dc7c2f BB |
483 | /* |
484 | * If we're a top-level vdev, try to load the allocation parameters. | |
485 | */ | |
428870ff BB |
486 | if (parent && !parent->vdev_parent && |
487 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
488 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
489 | &vd->vdev_ms_array); | |
490 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
491 | &vd->vdev_ms_shift); | |
492 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
493 | &vd->vdev_asize); | |
428870ff BB |
494 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
495 | &vd->vdev_removing); | |
496 | } | |
497 | ||
5ffb9d1d | 498 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
499 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
500 | alloctype == VDEV_ALLOC_ADD || | |
501 | alloctype == VDEV_ALLOC_SPLIT || | |
502 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
503 | vd->vdev_mg = metaslab_group_create(islog ? | |
504 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
505 | } |
506 | ||
507 | /* | |
508 | * If we're a leaf vdev, try to load the DTL object and other state. | |
509 | */ | |
b128c09f | 510 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
511 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
512 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
513 | if (alloctype == VDEV_ALLOC_LOAD) { |
514 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
fb5f0bc8 | 515 | &vd->vdev_dtl_smo.smo_object); |
b128c09f BB |
516 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
517 | &vd->vdev_unspare); | |
518 | } | |
9babb374 BB |
519 | |
520 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
521 | uint64_t spare = 0; | |
522 | ||
523 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
524 | &spare) == 0 && spare) | |
525 | spa_spare_add(vd); | |
526 | } | |
527 | ||
34dc7c2f BB |
528 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
529 | &vd->vdev_offline); | |
b128c09f | 530 | |
572e2857 BB |
531 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, |
532 | &vd->vdev_resilvering); | |
533 | ||
34dc7c2f BB |
534 | /* |
535 | * When importing a pool, we want to ignore the persistent fault | |
536 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
537 | * valid in the current context. Local vdevs will |
538 | * remain in the faulted state. | |
34dc7c2f | 539 | */ |
428870ff | 540 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
541 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
542 | &vd->vdev_faulted); | |
543 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
544 | &vd->vdev_degraded); | |
545 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
546 | &vd->vdev_removed); | |
428870ff BB |
547 | |
548 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
549 | char *aux; | |
550 | ||
551 | vd->vdev_label_aux = | |
552 | VDEV_AUX_ERR_EXCEEDED; | |
553 | if (nvlist_lookup_string(nv, | |
554 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
555 | strcmp(aux, "external") == 0) | |
556 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
557 | } | |
34dc7c2f BB |
558 | } |
559 | } | |
560 | ||
561 | /* | |
562 | * Add ourselves to the parent's list of children. | |
563 | */ | |
564 | vdev_add_child(parent, vd); | |
565 | ||
566 | *vdp = vd; | |
567 | ||
568 | return (0); | |
569 | } | |
570 | ||
571 | void | |
572 | vdev_free(vdev_t *vd) | |
573 | { | |
d6320ddb | 574 | int c, t; |
34dc7c2f BB |
575 | spa_t *spa = vd->vdev_spa; |
576 | ||
577 | /* | |
578 | * vdev_free() implies closing the vdev first. This is simpler than | |
579 | * trying to ensure complicated semantics for all callers. | |
580 | */ | |
581 | vdev_close(vd); | |
582 | ||
b128c09f | 583 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 584 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
585 | |
586 | /* | |
587 | * Free all children. | |
588 | */ | |
d6320ddb | 589 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
590 | vdev_free(vd->vdev_child[c]); |
591 | ||
592 | ASSERT(vd->vdev_child == NULL); | |
593 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
594 | ||
595 | /* | |
596 | * Discard allocation state. | |
597 | */ | |
428870ff | 598 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 599 | vdev_metaslab_fini(vd); |
428870ff BB |
600 | metaslab_group_destroy(vd->vdev_mg); |
601 | } | |
34dc7c2f BB |
602 | |
603 | ASSERT3U(vd->vdev_stat.vs_space, ==, 0); | |
604 | ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); | |
605 | ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); | |
606 | ||
607 | /* | |
608 | * Remove this vdev from its parent's child list. | |
609 | */ | |
610 | vdev_remove_child(vd->vdev_parent, vd); | |
611 | ||
612 | ASSERT(vd->vdev_parent == NULL); | |
613 | ||
614 | /* | |
615 | * Clean up vdev structure. | |
616 | */ | |
617 | vdev_queue_fini(vd); | |
618 | vdev_cache_fini(vd); | |
619 | ||
620 | if (vd->vdev_path) | |
621 | spa_strfree(vd->vdev_path); | |
622 | if (vd->vdev_devid) | |
623 | spa_strfree(vd->vdev_devid); | |
624 | if (vd->vdev_physpath) | |
625 | spa_strfree(vd->vdev_physpath); | |
9babb374 BB |
626 | if (vd->vdev_fru) |
627 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
628 | |
629 | if (vd->vdev_isspare) | |
630 | spa_spare_remove(vd); | |
631 | if (vd->vdev_isl2cache) | |
632 | spa_l2cache_remove(vd); | |
633 | ||
634 | txg_list_destroy(&vd->vdev_ms_list); | |
635 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 636 | |
34dc7c2f | 637 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 638 | for (t = 0; t < DTL_TYPES; t++) { |
fb5f0bc8 BB |
639 | space_map_unload(&vd->vdev_dtl[t]); |
640 | space_map_destroy(&vd->vdev_dtl[t]); | |
641 | } | |
34dc7c2f | 642 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 643 | |
34dc7c2f BB |
644 | mutex_destroy(&vd->vdev_dtl_lock); |
645 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 646 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
647 | |
648 | if (vd == spa->spa_root_vdev) | |
649 | spa->spa_root_vdev = NULL; | |
650 | ||
651 | kmem_free(vd, sizeof (vdev_t)); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Transfer top-level vdev state from svd to tvd. | |
656 | */ | |
657 | static void | |
658 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
659 | { | |
660 | spa_t *spa = svd->vdev_spa; | |
661 | metaslab_t *msp; | |
662 | vdev_t *vd; | |
663 | int t; | |
664 | ||
665 | ASSERT(tvd == tvd->vdev_top); | |
666 | ||
667 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
668 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
669 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
670 | ||
671 | svd->vdev_ms_array = 0; | |
672 | svd->vdev_ms_shift = 0; | |
673 | svd->vdev_ms_count = 0; | |
674 | ||
5ffb9d1d GW |
675 | if (tvd->vdev_mg) |
676 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
677 | tvd->vdev_mg = svd->vdev_mg; |
678 | tvd->vdev_ms = svd->vdev_ms; | |
679 | ||
680 | svd->vdev_mg = NULL; | |
681 | svd->vdev_ms = NULL; | |
682 | ||
683 | if (tvd->vdev_mg != NULL) | |
684 | tvd->vdev_mg->mg_vd = tvd; | |
685 | ||
686 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
687 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
688 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
689 | ||
690 | svd->vdev_stat.vs_alloc = 0; | |
691 | svd->vdev_stat.vs_space = 0; | |
692 | svd->vdev_stat.vs_dspace = 0; | |
693 | ||
694 | for (t = 0; t < TXG_SIZE; t++) { | |
695 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
696 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
697 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
698 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
699 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
700 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
701 | } | |
702 | ||
b128c09f | 703 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
704 | vdev_config_clean(svd); |
705 | vdev_config_dirty(tvd); | |
706 | } | |
707 | ||
b128c09f BB |
708 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
709 | vdev_state_clean(svd); | |
710 | vdev_state_dirty(tvd); | |
711 | } | |
712 | ||
34dc7c2f BB |
713 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
714 | svd->vdev_deflate_ratio = 0; | |
715 | ||
716 | tvd->vdev_islog = svd->vdev_islog; | |
717 | svd->vdev_islog = 0; | |
718 | } | |
719 | ||
720 | static void | |
721 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
722 | { | |
d6320ddb BB |
723 | int c; |
724 | ||
34dc7c2f BB |
725 | if (vd == NULL) |
726 | return; | |
727 | ||
728 | vd->vdev_top = tvd; | |
729 | ||
d6320ddb | 730 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
731 | vdev_top_update(tvd, vd->vdev_child[c]); |
732 | } | |
733 | ||
734 | /* | |
735 | * Add a mirror/replacing vdev above an existing vdev. | |
736 | */ | |
737 | vdev_t * | |
738 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
739 | { | |
740 | spa_t *spa = cvd->vdev_spa; | |
741 | vdev_t *pvd = cvd->vdev_parent; | |
742 | vdev_t *mvd; | |
743 | ||
b128c09f | 744 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
745 | |
746 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
747 | ||
748 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 749 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 750 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
34dc7c2f BB |
751 | mvd->vdev_ashift = cvd->vdev_ashift; |
752 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 753 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
754 | |
755 | vdev_remove_child(pvd, cvd); | |
756 | vdev_add_child(pvd, mvd); | |
757 | cvd->vdev_id = mvd->vdev_children; | |
758 | vdev_add_child(mvd, cvd); | |
759 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
760 | ||
761 | if (mvd == mvd->vdev_top) | |
762 | vdev_top_transfer(cvd, mvd); | |
763 | ||
764 | return (mvd); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Remove a 1-way mirror/replacing vdev from the tree. | |
769 | */ | |
770 | void | |
771 | vdev_remove_parent(vdev_t *cvd) | |
772 | { | |
773 | vdev_t *mvd = cvd->vdev_parent; | |
774 | vdev_t *pvd = mvd->vdev_parent; | |
775 | ||
b128c09f | 776 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
777 | |
778 | ASSERT(mvd->vdev_children == 1); | |
779 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
780 | mvd->vdev_ops == &vdev_replacing_ops || | |
781 | mvd->vdev_ops == &vdev_spare_ops); | |
782 | cvd->vdev_ashift = mvd->vdev_ashift; | |
783 | ||
784 | vdev_remove_child(mvd, cvd); | |
785 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 786 | |
34dc7c2f | 787 | /* |
b128c09f BB |
788 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
789 | * Otherwise, we could have detached an offline device, and when we | |
790 | * go to import the pool we'll think we have two top-level vdevs, | |
791 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 792 | */ |
fb5f0bc8 BB |
793 | if (mvd->vdev_top == mvd) { |
794 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 795 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
796 | cvd->vdev_guid += guid_delta; |
797 | cvd->vdev_guid_sum += guid_delta; | |
798 | } | |
b128c09f BB |
799 | cvd->vdev_id = mvd->vdev_id; |
800 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
801 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
802 | ||
803 | if (cvd == cvd->vdev_top) | |
804 | vdev_top_transfer(mvd, cvd); | |
805 | ||
806 | ASSERT(mvd->vdev_children == 0); | |
807 | vdev_free(mvd); | |
808 | } | |
809 | ||
810 | int | |
811 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
812 | { | |
813 | spa_t *spa = vd->vdev_spa; | |
814 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
815 | uint64_t m; |
816 | uint64_t oldc = vd->vdev_ms_count; | |
817 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
818 | metaslab_t **mspp; | |
819 | int error; | |
820 | ||
428870ff BB |
821 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
822 | ||
823 | /* | |
824 | * This vdev is not being allocated from yet or is a hole. | |
825 | */ | |
826 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
827 | return (0); |
828 | ||
428870ff BB |
829 | ASSERT(!vd->vdev_ishole); |
830 | ||
9babb374 BB |
831 | /* |
832 | * Compute the raidz-deflation ratio. Note, we hard-code | |
833 | * in 128k (1 << 17) because it is the current "typical" blocksize. | |
834 | * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, | |
835 | * or we will inconsistently account for existing bp's. | |
836 | */ | |
837 | vd->vdev_deflate_ratio = (1 << 17) / | |
838 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
839 | ||
34dc7c2f BB |
840 | ASSERT(oldc <= newc); |
841 | ||
b8d06fca | 842 | mspp = kmem_zalloc(newc * sizeof (*mspp), KM_PUSHPAGE | KM_NODEBUG); |
34dc7c2f BB |
843 | |
844 | if (oldc != 0) { | |
845 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
846 | kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
847 | } | |
848 | ||
849 | vd->vdev_ms = mspp; | |
850 | vd->vdev_ms_count = newc; | |
851 | ||
852 | for (m = oldc; m < newc; m++) { | |
853 | space_map_obj_t smo = { 0, 0, 0 }; | |
854 | if (txg == 0) { | |
855 | uint64_t object = 0; | |
856 | error = dmu_read(mos, vd->vdev_ms_array, | |
9babb374 BB |
857 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
858 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
859 | if (error) |
860 | return (error); | |
861 | if (object != 0) { | |
862 | dmu_buf_t *db; | |
863 | error = dmu_bonus_hold(mos, object, FTAG, &db); | |
864 | if (error) | |
865 | return (error); | |
866 | ASSERT3U(db->db_size, >=, sizeof (smo)); | |
867 | bcopy(db->db_data, &smo, sizeof (smo)); | |
868 | ASSERT3U(smo.smo_object, ==, object); | |
869 | dmu_buf_rele(db, FTAG); | |
870 | } | |
871 | } | |
872 | vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, | |
873 | m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); | |
874 | } | |
875 | ||
428870ff BB |
876 | if (txg == 0) |
877 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
878 | ||
879 | /* | |
880 | * If the vdev is being removed we don't activate | |
881 | * the metaslabs since we want to ensure that no new | |
882 | * allocations are performed on this device. | |
883 | */ | |
884 | if (oldc == 0 && !vd->vdev_removing) | |
885 | metaslab_group_activate(vd->vdev_mg); | |
886 | ||
887 | if (txg == 0) | |
888 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
889 | ||
34dc7c2f BB |
890 | return (0); |
891 | } | |
892 | ||
893 | void | |
894 | vdev_metaslab_fini(vdev_t *vd) | |
895 | { | |
896 | uint64_t m; | |
897 | uint64_t count = vd->vdev_ms_count; | |
898 | ||
899 | if (vd->vdev_ms != NULL) { | |
428870ff | 900 | metaslab_group_passivate(vd->vdev_mg); |
34dc7c2f BB |
901 | for (m = 0; m < count; m++) |
902 | if (vd->vdev_ms[m] != NULL) | |
903 | metaslab_fini(vd->vdev_ms[m]); | |
904 | kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); | |
905 | vd->vdev_ms = NULL; | |
906 | } | |
920dd524 ED |
907 | |
908 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
34dc7c2f BB |
909 | } |
910 | ||
b128c09f BB |
911 | typedef struct vdev_probe_stats { |
912 | boolean_t vps_readable; | |
913 | boolean_t vps_writeable; | |
914 | int vps_flags; | |
b128c09f BB |
915 | } vdev_probe_stats_t; |
916 | ||
917 | static void | |
918 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 919 | { |
fb5f0bc8 | 920 | spa_t *spa = zio->io_spa; |
d164b209 | 921 | vdev_t *vd = zio->io_vd; |
b128c09f | 922 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
923 | |
924 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
925 | |
926 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
927 | if (zio->io_error == 0) |
928 | vps->vps_readable = 1; | |
fb5f0bc8 | 929 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 930 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
b128c09f BB |
931 | zio->io_offset, zio->io_size, zio->io_data, |
932 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
933 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
934 | } else { | |
935 | zio_buf_free(zio->io_data, zio->io_size); | |
936 | } | |
937 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
938 | if (zio->io_error == 0) |
939 | vps->vps_writeable = 1; | |
940 | zio_buf_free(zio->io_data, zio->io_size); | |
941 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
d164b209 | 942 | zio_t *pio; |
b128c09f BB |
943 | |
944 | vd->vdev_cant_read |= !vps->vps_readable; | |
945 | vd->vdev_cant_write |= !vps->vps_writeable; | |
946 | ||
947 | if (vdev_readable(vd) && | |
fb5f0bc8 | 948 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
949 | zio->io_error = 0; |
950 | } else { | |
951 | ASSERT(zio->io_error != 0); | |
952 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 953 | spa, vd, NULL, 0, 0); |
b128c09f BB |
954 | zio->io_error = ENXIO; |
955 | } | |
d164b209 BB |
956 | |
957 | mutex_enter(&vd->vdev_probe_lock); | |
958 | ASSERT(vd->vdev_probe_zio == zio); | |
959 | vd->vdev_probe_zio = NULL; | |
960 | mutex_exit(&vd->vdev_probe_lock); | |
961 | ||
962 | while ((pio = zio_walk_parents(zio)) != NULL) | |
963 | if (!vdev_accessible(vd, pio)) | |
964 | pio->io_error = ENXIO; | |
965 | ||
b128c09f BB |
966 | kmem_free(vps, sizeof (*vps)); |
967 | } | |
968 | } | |
34dc7c2f | 969 | |
b128c09f BB |
970 | /* |
971 | * Determine whether this device is accessible by reading and writing | |
972 | * to several known locations: the pad regions of each vdev label | |
973 | * but the first (which we leave alone in case it contains a VTOC). | |
974 | */ | |
975 | zio_t * | |
d164b209 | 976 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
977 | { |
978 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
979 | vdev_probe_stats_t *vps = NULL; |
980 | zio_t *pio; | |
d6320ddb | 981 | int l; |
d164b209 BB |
982 | |
983 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 984 | |
d164b209 BB |
985 | /* |
986 | * Don't probe the probe. | |
987 | */ | |
988 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
989 | return (NULL); | |
b128c09f | 990 | |
d164b209 BB |
991 | /* |
992 | * To prevent 'probe storms' when a device fails, we create | |
993 | * just one probe i/o at a time. All zios that want to probe | |
994 | * this vdev will become parents of the probe io. | |
995 | */ | |
996 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 997 | |
d164b209 | 998 | if ((pio = vd->vdev_probe_zio) == NULL) { |
b8d06fca | 999 | vps = kmem_zalloc(sizeof (*vps), KM_PUSHPAGE); |
d164b209 BB |
1000 | |
1001 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1002 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1003 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1004 | |
1005 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1006 | /* | |
1007 | * vdev_cant_read and vdev_cant_write can only | |
1008 | * transition from TRUE to FALSE when we have the | |
1009 | * SCL_ZIO lock as writer; otherwise they can only | |
1010 | * transition from FALSE to TRUE. This ensures that | |
1011 | * any zio looking at these values can assume that | |
1012 | * failures persist for the life of the I/O. That's | |
1013 | * important because when a device has intermittent | |
1014 | * connectivity problems, we want to ensure that | |
1015 | * they're ascribed to the device (ENXIO) and not | |
1016 | * the zio (EIO). | |
1017 | * | |
1018 | * Since we hold SCL_ZIO as writer here, clear both | |
1019 | * values so the probe can reevaluate from first | |
1020 | * principles. | |
1021 | */ | |
1022 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1023 | vd->vdev_cant_read = B_FALSE; | |
1024 | vd->vdev_cant_write = B_FALSE; | |
1025 | } | |
1026 | ||
1027 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1028 | vdev_probe_done, vps, | |
1029 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1030 | ||
428870ff BB |
1031 | /* |
1032 | * We can't change the vdev state in this context, so we | |
1033 | * kick off an async task to do it on our behalf. | |
1034 | */ | |
d164b209 BB |
1035 | if (zio != NULL) { |
1036 | vd->vdev_probe_wanted = B_TRUE; | |
1037 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1038 | } | |
b128c09f BB |
1039 | } |
1040 | ||
d164b209 BB |
1041 | if (zio != NULL) |
1042 | zio_add_child(zio, pio); | |
b128c09f | 1043 | |
d164b209 | 1044 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1045 | |
d164b209 BB |
1046 | if (vps == NULL) { |
1047 | ASSERT(zio != NULL); | |
1048 | return (NULL); | |
1049 | } | |
b128c09f | 1050 | |
d6320ddb | 1051 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1052 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1053 | vdev_label_offset(vd->vdev_psize, l, |
9babb374 BB |
1054 | offsetof(vdev_label_t, vl_pad2)), |
1055 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
b128c09f BB |
1056 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1057 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1058 | } | |
1059 | ||
d164b209 BB |
1060 | if (zio == NULL) |
1061 | return (pio); | |
1062 | ||
1063 | zio_nowait(pio); | |
1064 | return (NULL); | |
34dc7c2f BB |
1065 | } |
1066 | ||
45d1cae3 BB |
1067 | static void |
1068 | vdev_open_child(void *arg) | |
1069 | { | |
1070 | vdev_t *vd = arg; | |
1071 | ||
1072 | vd->vdev_open_thread = curthread; | |
1073 | vd->vdev_open_error = vdev_open(vd); | |
1074 | vd->vdev_open_thread = NULL; | |
1075 | } | |
1076 | ||
428870ff BB |
1077 | boolean_t |
1078 | vdev_uses_zvols(vdev_t *vd) | |
1079 | { | |
60101509 BB |
1080 | /* |
1081 | * Stacking zpools on top of zvols is unsupported until we implement a method | |
1082 | * for determining if an arbitrary block device is a zvol without using the | |
1083 | * path. Solaris would check the 'zvol' path component but this does not | |
1084 | * exist in the Linux port, so we really should do something like stat the | |
1085 | * file and check the major number. This is complicated by the fact that | |
1086 | * we need to do this portably in user or kernel space. | |
1087 | */ | |
1088 | #if 0 | |
d6320ddb BB |
1089 | int c; |
1090 | ||
428870ff BB |
1091 | if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, |
1092 | strlen(ZVOL_DIR)) == 0) | |
1093 | return (B_TRUE); | |
d6320ddb | 1094 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1095 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1096 | return (B_TRUE); | |
60101509 | 1097 | #endif |
428870ff BB |
1098 | return (B_FALSE); |
1099 | } | |
1100 | ||
45d1cae3 BB |
1101 | void |
1102 | vdev_open_children(vdev_t *vd) | |
1103 | { | |
1104 | taskq_t *tq; | |
1105 | int children = vd->vdev_children; | |
d6320ddb | 1106 | int c; |
45d1cae3 | 1107 | |
428870ff BB |
1108 | /* |
1109 | * in order to handle pools on top of zvols, do the opens | |
1110 | * in a single thread so that the same thread holds the | |
1111 | * spa_namespace_lock | |
1112 | */ | |
1113 | if (vdev_uses_zvols(vd)) { | |
d6320ddb | 1114 | for (c = 0; c < children; c++) |
428870ff BB |
1115 | vd->vdev_child[c]->vdev_open_error = |
1116 | vdev_open(vd->vdev_child[c]); | |
1117 | return; | |
1118 | } | |
45d1cae3 BB |
1119 | tq = taskq_create("vdev_open", children, minclsyspri, |
1120 | children, children, TASKQ_PREPOPULATE); | |
1121 | ||
d6320ddb | 1122 | for (c = 0; c < children; c++) |
45d1cae3 | 1123 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], |
d6320ddb | 1124 | TQ_SLEEP) != 0); |
45d1cae3 BB |
1125 | |
1126 | taskq_destroy(tq); | |
1127 | } | |
1128 | ||
34dc7c2f BB |
1129 | /* |
1130 | * Prepare a virtual device for access. | |
1131 | */ | |
1132 | int | |
1133 | vdev_open(vdev_t *vd) | |
1134 | { | |
fb5f0bc8 | 1135 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1136 | int error; |
34dc7c2f | 1137 | uint64_t osize = 0; |
1bd201e7 CS |
1138 | uint64_t max_osize = 0; |
1139 | uint64_t asize, max_asize, psize; | |
34dc7c2f | 1140 | uint64_t ashift = 0; |
d6320ddb | 1141 | int c; |
34dc7c2f | 1142 | |
45d1cae3 BB |
1143 | ASSERT(vd->vdev_open_thread == curthread || |
1144 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1145 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1146 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1147 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1148 | ||
34dc7c2f | 1149 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1150 | vd->vdev_cant_read = B_FALSE; |
1151 | vd->vdev_cant_write = B_FALSE; | |
1152 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1153 | |
428870ff BB |
1154 | /* |
1155 | * If this vdev is not removed, check its fault status. If it's | |
1156 | * faulted, bail out of the open. | |
1157 | */ | |
34dc7c2f BB |
1158 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1159 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1160 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1161 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1162 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1163 | vd->vdev_label_aux); |
34dc7c2f BB |
1164 | return (ENXIO); |
1165 | } else if (vd->vdev_offline) { | |
1166 | ASSERT(vd->vdev_children == 0); | |
1167 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
1168 | return (ENXIO); | |
1169 | } | |
1170 | ||
1bd201e7 | 1171 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1172 | |
428870ff BB |
1173 | /* |
1174 | * Reset the vdev_reopening flag so that we actually close | |
1175 | * the vdev on error. | |
1176 | */ | |
1177 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1178 | if (zio_injection_enabled && error == 0) |
9babb374 | 1179 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1180 | |
1181 | if (error) { | |
1182 | if (vd->vdev_removed && | |
1183 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1184 | vd->vdev_removed = B_FALSE; | |
1185 | ||
1186 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1187 | vd->vdev_stat.vs_aux); | |
1188 | return (error); | |
1189 | } | |
1190 | ||
1191 | vd->vdev_removed = B_FALSE; | |
1192 | ||
428870ff BB |
1193 | /* |
1194 | * Recheck the faulted flag now that we have confirmed that | |
1195 | * the vdev is accessible. If we're faulted, bail. | |
1196 | */ | |
1197 | if (vd->vdev_faulted) { | |
1198 | ASSERT(vd->vdev_children == 0); | |
1199 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1200 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1201 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1202 | vd->vdev_label_aux); | |
1203 | return (ENXIO); | |
1204 | } | |
1205 | ||
34dc7c2f BB |
1206 | if (vd->vdev_degraded) { |
1207 | ASSERT(vd->vdev_children == 0); | |
1208 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1209 | VDEV_AUX_ERR_EXCEEDED); | |
1210 | } else { | |
428870ff | 1211 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1212 | } |
1213 | ||
428870ff BB |
1214 | /* |
1215 | * For hole or missing vdevs we just return success. | |
1216 | */ | |
1217 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1218 | return (0); | |
1219 | ||
d6320ddb | 1220 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1221 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1222 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1223 | VDEV_AUX_NONE); | |
1224 | break; | |
1225 | } | |
9babb374 | 1226 | } |
34dc7c2f BB |
1227 | |
1228 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1229 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1230 | |
1231 | if (vd->vdev_children == 0) { | |
1232 | if (osize < SPA_MINDEVSIZE) { | |
1233 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1234 | VDEV_AUX_TOO_SMALL); | |
1235 | return (EOVERFLOW); | |
1236 | } | |
1237 | psize = osize; | |
1238 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1239 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1240 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1241 | } else { |
1242 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1243 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1244 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1245 | VDEV_AUX_TOO_SMALL); | |
1246 | return (EOVERFLOW); | |
1247 | } | |
1248 | psize = 0; | |
1249 | asize = osize; | |
1bd201e7 | 1250 | max_asize = max_osize; |
34dc7c2f BB |
1251 | } |
1252 | ||
1253 | vd->vdev_psize = psize; | |
1254 | ||
9babb374 BB |
1255 | /* |
1256 | * Make sure the allocatable size hasn't shrunk. | |
1257 | */ | |
1258 | if (asize < vd->vdev_min_asize) { | |
1259 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1260 | VDEV_AUX_BAD_LABEL); | |
1261 | return (EINVAL); | |
1262 | } | |
1263 | ||
34dc7c2f BB |
1264 | if (vd->vdev_asize == 0) { |
1265 | /* | |
1266 | * This is the first-ever open, so use the computed values. | |
1267 | * For testing purposes, a higher ashift can be requested. | |
1268 | */ | |
1269 | vd->vdev_asize = asize; | |
1bd201e7 | 1270 | vd->vdev_max_asize = max_asize; |
34dc7c2f BB |
1271 | vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); |
1272 | } else { | |
1273 | /* | |
1274 | * Make sure the alignment requirement hasn't increased. | |
1275 | */ | |
1276 | if (ashift > vd->vdev_top->vdev_ashift) { | |
1277 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1278 | VDEV_AUX_BAD_LABEL); | |
1279 | return (EINVAL); | |
1280 | } | |
1bd201e7 | 1281 | vd->vdev_max_asize = max_asize; |
9babb374 | 1282 | } |
34dc7c2f | 1283 | |
9babb374 BB |
1284 | /* |
1285 | * If all children are healthy and the asize has increased, | |
1286 | * then we've experienced dynamic LUN growth. If automatic | |
1287 | * expansion is enabled then use the additional space. | |
1288 | */ | |
1289 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1290 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1291 | vd->vdev_asize = asize; | |
34dc7c2f | 1292 | |
9babb374 | 1293 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1294 | |
1295 | /* | |
1296 | * Ensure we can issue some IO before declaring the | |
1297 | * vdev open for business. | |
1298 | */ | |
b128c09f BB |
1299 | if (vd->vdev_ops->vdev_op_leaf && |
1300 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1301 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1302 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1303 | return (error); |
1304 | } | |
1305 | ||
34dc7c2f | 1306 | /* |
b128c09f | 1307 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1308 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1309 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1310 | */ |
fb5f0bc8 BB |
1311 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1312 | vdev_resilver_needed(vd, NULL, NULL)) | |
1313 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1314 | |
1315 | return (0); | |
1316 | } | |
1317 | ||
1318 | /* | |
1319 | * Called once the vdevs are all opened, this routine validates the label | |
1320 | * contents. This needs to be done before vdev_load() so that we don't | |
1321 | * inadvertently do repair I/Os to the wrong device. | |
1322 | * | |
c7f2d69d GW |
1323 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1324 | * if the machine crashed during a re-guid the new guid might have been written | |
1325 | * to all of the vdev labels, but not the cached config. The strict check | |
1326 | * will be performed when the pool is opened again using the mos config. | |
1327 | * | |
34dc7c2f BB |
1328 | * This function will only return failure if one of the vdevs indicates that it |
1329 | * has since been destroyed or exported. This is only possible if | |
1330 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1331 | * will be updated but the function will return 0. | |
1332 | */ | |
1333 | int | |
c7f2d69d | 1334 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1335 | { |
1336 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1337 | nvlist_t *label; |
428870ff | 1338 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1339 | uint64_t state; |
d6320ddb | 1340 | int c; |
34dc7c2f | 1341 | |
d6320ddb | 1342 | for (c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1343 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
34dc7c2f BB |
1344 | return (EBADF); |
1345 | ||
1346 | /* | |
1347 | * If the device has already failed, or was marked offline, don't do | |
1348 | * any further validation. Otherwise, label I/O will fail and we will | |
1349 | * overwrite the previous state. | |
1350 | */ | |
b128c09f | 1351 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1352 | uint64_t aux_guid = 0; |
1353 | nvlist_t *nvl; | |
34dc7c2f BB |
1354 | |
1355 | if ((label = vdev_label_read_config(vd)) == NULL) { | |
1356 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1357 | VDEV_AUX_BAD_LABEL); | |
1358 | return (0); | |
1359 | } | |
1360 | ||
428870ff BB |
1361 | /* |
1362 | * Determine if this vdev has been split off into another | |
1363 | * pool. If so, then refuse to open it. | |
1364 | */ | |
1365 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1366 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1367 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1368 | VDEV_AUX_SPLIT_POOL); | |
1369 | nvlist_free(label); | |
1370 | return (0); | |
1371 | } | |
1372 | ||
c7f2d69d GW |
1373 | if (strict && (nvlist_lookup_uint64(label, |
1374 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1375 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1376 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1377 | VDEV_AUX_CORRUPT_DATA); | |
1378 | nvlist_free(label); | |
1379 | return (0); | |
1380 | } | |
1381 | ||
428870ff BB |
1382 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1383 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1384 | &aux_guid) != 0) | |
1385 | aux_guid = 0; | |
1386 | ||
b128c09f BB |
1387 | /* |
1388 | * If this vdev just became a top-level vdev because its | |
1389 | * sibling was detached, it will have adopted the parent's | |
1390 | * vdev guid -- but the label may or may not be on disk yet. | |
1391 | * Fortunately, either version of the label will have the | |
1392 | * same top guid, so if we're a top-level vdev, we can | |
1393 | * safely compare to that instead. | |
428870ff BB |
1394 | * |
1395 | * If we split this vdev off instead, then we also check the | |
1396 | * original pool's guid. We don't want to consider the vdev | |
1397 | * corrupt if it is partway through a split operation. | |
b128c09f | 1398 | */ |
34dc7c2f | 1399 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1400 | &guid) != 0 || |
1401 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1402 | &top_guid) != 0 || | |
428870ff | 1403 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1404 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1405 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1406 | VDEV_AUX_CORRUPT_DATA); | |
1407 | nvlist_free(label); | |
1408 | return (0); | |
1409 | } | |
1410 | ||
1411 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1412 | &state) != 0) { | |
1413 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1414 | VDEV_AUX_CORRUPT_DATA); | |
1415 | nvlist_free(label); | |
1416 | return (0); | |
1417 | } | |
1418 | ||
1419 | nvlist_free(label); | |
1420 | ||
45d1cae3 | 1421 | /* |
572e2857 | 1422 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1423 | * state of the pool. |
1424 | */ | |
572e2857 | 1425 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1426 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f BB |
1427 | state != POOL_STATE_ACTIVE) |
1428 | return (EBADF); | |
34dc7c2f | 1429 | |
b128c09f BB |
1430 | /* |
1431 | * If we were able to open and validate a vdev that was | |
1432 | * previously marked permanently unavailable, clear that state | |
1433 | * now. | |
1434 | */ | |
1435 | if (vd->vdev_not_present) | |
1436 | vd->vdev_not_present = 0; | |
1437 | } | |
34dc7c2f BB |
1438 | |
1439 | return (0); | |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * Close a virtual device. | |
1444 | */ | |
1445 | void | |
1446 | vdev_close(vdev_t *vd) | |
1447 | { | |
428870ff | 1448 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1449 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1450 | |
1451 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1452 | ||
428870ff BB |
1453 | /* |
1454 | * If our parent is reopening, then we are as well, unless we are | |
1455 | * going offline. | |
1456 | */ | |
1457 | if (pvd != NULL && pvd->vdev_reopening) | |
1458 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1459 | ||
34dc7c2f BB |
1460 | vd->vdev_ops->vdev_op_close(vd); |
1461 | ||
1462 | vdev_cache_purge(vd); | |
1463 | ||
1464 | /* | |
9babb374 | 1465 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1466 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1467 | * it's still faulted. | |
1468 | */ | |
1469 | vd->vdev_prevstate = vd->vdev_state; | |
1470 | ||
1471 | if (vd->vdev_offline) | |
1472 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1473 | else | |
1474 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1475 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1476 | } | |
1477 | ||
428870ff BB |
1478 | void |
1479 | vdev_hold(vdev_t *vd) | |
1480 | { | |
1481 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1482 | int c; |
428870ff BB |
1483 | |
1484 | ASSERT(spa_is_root(spa)); | |
1485 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1486 | return; | |
1487 | ||
d6320ddb | 1488 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1489 | vdev_hold(vd->vdev_child[c]); |
1490 | ||
1491 | if (vd->vdev_ops->vdev_op_leaf) | |
1492 | vd->vdev_ops->vdev_op_hold(vd); | |
1493 | } | |
1494 | ||
1495 | void | |
1496 | vdev_rele(vdev_t *vd) | |
1497 | { | |
d6320ddb | 1498 | int c; |
428870ff | 1499 | |
d6320ddb BB |
1500 | ASSERT(spa_is_root(vd->vdev_spa)); |
1501 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1502 | vdev_rele(vd->vdev_child[c]); |
1503 | ||
1504 | if (vd->vdev_ops->vdev_op_leaf) | |
1505 | vd->vdev_ops->vdev_op_rele(vd); | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1510 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1511 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1512 | * If the leaf has never been opened then open it, as usual. | |
1513 | */ | |
34dc7c2f BB |
1514 | void |
1515 | vdev_reopen(vdev_t *vd) | |
1516 | { | |
1517 | spa_t *spa = vd->vdev_spa; | |
1518 | ||
b128c09f | 1519 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1520 | |
428870ff BB |
1521 | /* set the reopening flag unless we're taking the vdev offline */ |
1522 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1523 | vdev_close(vd); |
1524 | (void) vdev_open(vd); | |
1525 | ||
1526 | /* | |
1527 | * Call vdev_validate() here to make sure we have the same device. | |
1528 | * Otherwise, a device with an invalid label could be successfully | |
1529 | * opened in response to vdev_reopen(). | |
1530 | */ | |
b128c09f BB |
1531 | if (vd->vdev_aux) { |
1532 | (void) vdev_validate_aux(vd); | |
1533 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1534 | vd->vdev_aux == &spa->spa_l2cache && |
1535 | !l2arc_vdev_present(vd)) | |
1536 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1537 | } else { |
c7f2d69d | 1538 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1539 | } |
34dc7c2f BB |
1540 | |
1541 | /* | |
1542 | * Reassess parent vdev's health. | |
1543 | */ | |
1544 | vdev_propagate_state(vd); | |
1545 | } | |
1546 | ||
1547 | int | |
1548 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1549 | { | |
1550 | int error; | |
1551 | ||
1552 | /* | |
1553 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1554 | * For a create, however, we want to fail the request if | |
1555 | * there are any components we can't open. | |
1556 | */ | |
1557 | error = vdev_open(vd); | |
1558 | ||
1559 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1560 | vdev_close(vd); | |
1561 | return (error ? error : ENXIO); | |
1562 | } | |
1563 | ||
1564 | /* | |
1565 | * Recursively initialize all labels. | |
1566 | */ | |
1567 | if ((error = vdev_label_init(vd, txg, isreplacing ? | |
1568 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { | |
1569 | vdev_close(vd); | |
1570 | return (error); | |
1571 | } | |
1572 | ||
1573 | return (0); | |
1574 | } | |
1575 | ||
34dc7c2f | 1576 | void |
9babb374 | 1577 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1578 | { |
1579 | /* | |
1580 | * Aim for roughly 200 metaslabs per vdev. | |
1581 | */ | |
1582 | vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); | |
1583 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
1584 | } |
1585 | ||
1586 | void | |
1587 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1588 | { | |
1589 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1590 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1591 | ASSERT(ISP2(flags)); |
572e2857 | 1592 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1593 | |
1594 | if (flags & VDD_METASLAB) | |
1595 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1596 | ||
1597 | if (flags & VDD_DTL) | |
1598 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1599 | ||
1600 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1601 | } | |
1602 | ||
fb5f0bc8 BB |
1603 | /* |
1604 | * DTLs. | |
1605 | * | |
1606 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1607 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1608 | * |
1609 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1610 | * | |
1611 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1612 | * | |
1613 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1614 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1615 | * txgs that was scrubbed. | |
1616 | * | |
1617 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1618 | * persistent errors or just some device being offline. | |
1619 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1620 | * maintained; it's only computed when needed, typically to | |
1621 | * determine whether a device can be detached. | |
1622 | * | |
1623 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1624 | * either has the data or it doesn't. | |
1625 | * | |
1626 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1627 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1628 | * if any child is less than fully replicated, then so is its parent. | |
1629 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1630 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1631 | * those are the txgs we don't have enough replication to read. For example, | |
1632 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1633 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1634 | * two child DTL_MISSING maps. | |
1635 | * | |
1636 | * It should be clear from the above that to compute the DTLs and outage maps | |
1637 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1638 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1639 | * a configuration change, we generate all other DTLs from first principles. | |
1640 | */ | |
34dc7c2f | 1641 | void |
fb5f0bc8 | 1642 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1643 | { |
fb5f0bc8 BB |
1644 | space_map_t *sm = &vd->vdev_dtl[t]; |
1645 | ||
1646 | ASSERT(t < DTL_TYPES); | |
1647 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1648 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1649 | |
34dc7c2f BB |
1650 | mutex_enter(sm->sm_lock); |
1651 | if (!space_map_contains(sm, txg, size)) | |
1652 | space_map_add(sm, txg, size); | |
1653 | mutex_exit(sm->sm_lock); | |
1654 | } | |
1655 | ||
fb5f0bc8 BB |
1656 | boolean_t |
1657 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1658 | { |
fb5f0bc8 BB |
1659 | space_map_t *sm = &vd->vdev_dtl[t]; |
1660 | boolean_t dirty = B_FALSE; | |
34dc7c2f | 1661 | |
fb5f0bc8 BB |
1662 | ASSERT(t < DTL_TYPES); |
1663 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f BB |
1664 | |
1665 | mutex_enter(sm->sm_lock); | |
fb5f0bc8 BB |
1666 | if (sm->sm_space != 0) |
1667 | dirty = space_map_contains(sm, txg, size); | |
34dc7c2f BB |
1668 | mutex_exit(sm->sm_lock); |
1669 | ||
1670 | return (dirty); | |
1671 | } | |
1672 | ||
fb5f0bc8 BB |
1673 | boolean_t |
1674 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1675 | { | |
1676 | space_map_t *sm = &vd->vdev_dtl[t]; | |
1677 | boolean_t empty; | |
1678 | ||
1679 | mutex_enter(sm->sm_lock); | |
1680 | empty = (sm->sm_space == 0); | |
1681 | mutex_exit(sm->sm_lock); | |
1682 | ||
1683 | return (empty); | |
1684 | } | |
1685 | ||
34dc7c2f BB |
1686 | /* |
1687 | * Reassess DTLs after a config change or scrub completion. | |
1688 | */ | |
1689 | void | |
1690 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1691 | { | |
1692 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1693 | avl_tree_t reftree; |
d6320ddb | 1694 | int c, t, minref; |
34dc7c2f | 1695 | |
fb5f0bc8 | 1696 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1697 | |
d6320ddb | 1698 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1699 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1700 | scrub_txg, scrub_done); | |
1701 | ||
428870ff | 1702 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1703 | return; |
1704 | ||
1705 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1706 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1707 | ||
34dc7c2f | 1708 | mutex_enter(&vd->vdev_dtl_lock); |
b128c09f | 1709 | if (scrub_txg != 0 && |
428870ff BB |
1710 | (spa->spa_scrub_started || |
1711 | (scn && scn->scn_phys.scn_errors == 0))) { | |
b128c09f BB |
1712 | /* |
1713 | * We completed a scrub up to scrub_txg. If we | |
1714 | * did it without rebooting, then the scrub dtl | |
1715 | * will be valid, so excise the old region and | |
1716 | * fold in the scrub dtl. Otherwise, leave the | |
1717 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1718 | * |
1719 | * There's little trick here: to excise the beginning | |
1720 | * of the DTL_MISSING map, we put it into a reference | |
1721 | * tree and then add a segment with refcnt -1 that | |
1722 | * covers the range [0, scrub_txg). This means | |
1723 | * that each txg in that range has refcnt -1 or 0. | |
1724 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1725 | * entries in the range [0, scrub_txg) will have a | |
1726 | * positive refcnt -- either 1 or 2. We then convert | |
1727 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1728 | */ |
fb5f0bc8 BB |
1729 | space_map_ref_create(&reftree); |
1730 | space_map_ref_add_map(&reftree, | |
1731 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1732 | space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); | |
1733 | space_map_ref_add_map(&reftree, | |
1734 | &vd->vdev_dtl[DTL_SCRUB], 2); | |
1735 | space_map_ref_generate_map(&reftree, | |
1736 | &vd->vdev_dtl[DTL_MISSING], 1); | |
1737 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1738 | } |
fb5f0bc8 BB |
1739 | space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1740 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1741 | space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1742 | if (scrub_done) |
fb5f0bc8 BB |
1743 | space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1744 | space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
1745 | if (!vdev_readable(vd)) | |
1746 | space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); | |
1747 | else | |
1748 | space_map_walk(&vd->vdev_dtl[DTL_MISSING], | |
1749 | space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); | |
34dc7c2f | 1750 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1751 | |
34dc7c2f BB |
1752 | if (txg != 0) |
1753 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1754 | return; | |
1755 | } | |
1756 | ||
34dc7c2f | 1757 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1758 | for (t = 0; t < DTL_TYPES; t++) { |
428870ff BB |
1759 | /* account for child's outage in parent's missing map */ |
1760 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1761 | if (t == DTL_SCRUB) |
1762 | continue; /* leaf vdevs only */ | |
1763 | if (t == DTL_PARTIAL) | |
1764 | minref = 1; /* i.e. non-zero */ | |
1765 | else if (vd->vdev_nparity != 0) | |
1766 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1767 | else | |
1768 | minref = vd->vdev_children; /* any kind of mirror */ | |
1769 | space_map_ref_create(&reftree); | |
d6320ddb | 1770 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1771 | vdev_t *cvd = vd->vdev_child[c]; |
1772 | mutex_enter(&cvd->vdev_dtl_lock); | |
428870ff | 1773 | space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1774 | mutex_exit(&cvd->vdev_dtl_lock); |
1775 | } | |
1776 | space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); | |
1777 | space_map_ref_destroy(&reftree); | |
34dc7c2f | 1778 | } |
fb5f0bc8 | 1779 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1780 | } |
1781 | ||
1782 | static int | |
1783 | vdev_dtl_load(vdev_t *vd) | |
1784 | { | |
1785 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1786 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
34dc7c2f BB |
1787 | objset_t *mos = spa->spa_meta_objset; |
1788 | dmu_buf_t *db; | |
1789 | int error; | |
1790 | ||
1791 | ASSERT(vd->vdev_children == 0); | |
1792 | ||
1793 | if (smo->smo_object == 0) | |
1794 | return (0); | |
1795 | ||
428870ff BB |
1796 | ASSERT(!vd->vdev_ishole); |
1797 | ||
34dc7c2f BB |
1798 | if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) |
1799 | return (error); | |
1800 | ||
1801 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1802 | bcopy(db->db_data, smo, sizeof (*smo)); | |
1803 | dmu_buf_rele(db, FTAG); | |
1804 | ||
1805 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1806 | error = space_map_load(&vd->vdev_dtl[DTL_MISSING], |
1807 | NULL, SM_ALLOC, smo, mos); | |
34dc7c2f BB |
1808 | mutex_exit(&vd->vdev_dtl_lock); |
1809 | ||
1810 | return (error); | |
1811 | } | |
1812 | ||
1813 | void | |
1814 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1815 | { | |
1816 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 BB |
1817 | space_map_obj_t *smo = &vd->vdev_dtl_smo; |
1818 | space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; | |
34dc7c2f BB |
1819 | objset_t *mos = spa->spa_meta_objset; |
1820 | space_map_t smsync; | |
1821 | kmutex_t smlock; | |
1822 | dmu_buf_t *db; | |
1823 | dmu_tx_t *tx; | |
1824 | ||
428870ff BB |
1825 | ASSERT(!vd->vdev_ishole); |
1826 | ||
34dc7c2f BB |
1827 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
1828 | ||
1829 | if (vd->vdev_detached) { | |
1830 | if (smo->smo_object != 0) { | |
1fde1e37 | 1831 | VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx)); |
34dc7c2f BB |
1832 | smo->smo_object = 0; |
1833 | } | |
1834 | dmu_tx_commit(tx); | |
34dc7c2f BB |
1835 | return; |
1836 | } | |
1837 | ||
1838 | if (smo->smo_object == 0) { | |
1839 | ASSERT(smo->smo_objsize == 0); | |
1840 | ASSERT(smo->smo_alloc == 0); | |
1841 | smo->smo_object = dmu_object_alloc(mos, | |
1842 | DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, | |
1843 | DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); | |
1844 | ASSERT(smo->smo_object != 0); | |
1845 | vdev_config_dirty(vd->vdev_top); | |
1846 | } | |
1847 | ||
1848 | mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); | |
1849 | ||
1850 | space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, | |
1851 | &smlock); | |
1852 | ||
1853 | mutex_enter(&smlock); | |
1854 | ||
1855 | mutex_enter(&vd->vdev_dtl_lock); | |
1856 | space_map_walk(sm, space_map_add, &smsync); | |
1857 | mutex_exit(&vd->vdev_dtl_lock); | |
1858 | ||
1859 | space_map_truncate(smo, mos, tx); | |
1860 | space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); | |
1861 | ||
1862 | space_map_destroy(&smsync); | |
1863 | ||
1864 | mutex_exit(&smlock); | |
1865 | mutex_destroy(&smlock); | |
1866 | ||
1867 | VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); | |
1868 | dmu_buf_will_dirty(db, tx); | |
1869 | ASSERT3U(db->db_size, >=, sizeof (*smo)); | |
1870 | bcopy(smo, db->db_data, sizeof (*smo)); | |
1871 | dmu_buf_rele(db, FTAG); | |
1872 | ||
1873 | dmu_tx_commit(tx); | |
1874 | } | |
1875 | ||
fb5f0bc8 BB |
1876 | /* |
1877 | * Determine whether the specified vdev can be offlined/detached/removed | |
1878 | * without losing data. | |
1879 | */ | |
1880 | boolean_t | |
1881 | vdev_dtl_required(vdev_t *vd) | |
1882 | { | |
1883 | spa_t *spa = vd->vdev_spa; | |
1884 | vdev_t *tvd = vd->vdev_top; | |
1885 | uint8_t cant_read = vd->vdev_cant_read; | |
1886 | boolean_t required; | |
1887 | ||
1888 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1889 | ||
1890 | if (vd == spa->spa_root_vdev || vd == tvd) | |
1891 | return (B_TRUE); | |
1892 | ||
1893 | /* | |
1894 | * Temporarily mark the device as unreadable, and then determine | |
1895 | * whether this results in any DTL outages in the top-level vdev. | |
1896 | * If not, we can safely offline/detach/remove the device. | |
1897 | */ | |
1898 | vd->vdev_cant_read = B_TRUE; | |
1899 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1900 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
1901 | vd->vdev_cant_read = cant_read; | |
1902 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
1903 | ||
572e2857 BB |
1904 | if (!required && zio_injection_enabled) |
1905 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
1906 | ||
fb5f0bc8 BB |
1907 | return (required); |
1908 | } | |
1909 | ||
b128c09f BB |
1910 | /* |
1911 | * Determine if resilver is needed, and if so the txg range. | |
1912 | */ | |
1913 | boolean_t | |
1914 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
1915 | { | |
1916 | boolean_t needed = B_FALSE; | |
1917 | uint64_t thismin = UINT64_MAX; | |
1918 | uint64_t thismax = 0; | |
d6320ddb | 1919 | int c; |
b128c09f BB |
1920 | |
1921 | if (vd->vdev_children == 0) { | |
1922 | mutex_enter(&vd->vdev_dtl_lock); | |
fb5f0bc8 BB |
1923 | if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && |
1924 | vdev_writeable(vd)) { | |
b128c09f BB |
1925 | space_seg_t *ss; |
1926 | ||
fb5f0bc8 | 1927 | ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f | 1928 | thismin = ss->ss_start - 1; |
fb5f0bc8 | 1929 | ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); |
b128c09f BB |
1930 | thismax = ss->ss_end; |
1931 | needed = B_TRUE; | |
1932 | } | |
1933 | mutex_exit(&vd->vdev_dtl_lock); | |
1934 | } else { | |
d6320ddb | 1935 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
1936 | vdev_t *cvd = vd->vdev_child[c]; |
1937 | uint64_t cmin, cmax; | |
1938 | ||
1939 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
1940 | thismin = MIN(thismin, cmin); | |
1941 | thismax = MAX(thismax, cmax); | |
1942 | needed = B_TRUE; | |
1943 | } | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | if (needed && minp) { | |
1948 | *minp = thismin; | |
1949 | *maxp = thismax; | |
1950 | } | |
1951 | return (needed); | |
1952 | } | |
1953 | ||
34dc7c2f BB |
1954 | void |
1955 | vdev_load(vdev_t *vd) | |
1956 | { | |
d6320ddb BB |
1957 | int c; |
1958 | ||
34dc7c2f BB |
1959 | /* |
1960 | * Recursively load all children. | |
1961 | */ | |
d6320ddb | 1962 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
1963 | vdev_load(vd->vdev_child[c]); |
1964 | ||
1965 | /* | |
1966 | * If this is a top-level vdev, initialize its metaslabs. | |
1967 | */ | |
428870ff | 1968 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
1969 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
1970 | vdev_metaslab_init(vd, 0) != 0)) | |
1971 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1972 | VDEV_AUX_CORRUPT_DATA); | |
1973 | ||
1974 | /* | |
1975 | * If this is a leaf vdev, load its DTL. | |
1976 | */ | |
1977 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
1978 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1979 | VDEV_AUX_CORRUPT_DATA); | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * The special vdev case is used for hot spares and l2cache devices. Its | |
1984 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
1985 | * we make sure that we can open the underlying device, then try to read the | |
1986 | * label, and make sure that the label is sane and that it hasn't been | |
1987 | * repurposed to another pool. | |
1988 | */ | |
1989 | int | |
1990 | vdev_validate_aux(vdev_t *vd) | |
1991 | { | |
1992 | nvlist_t *label; | |
1993 | uint64_t guid, version; | |
1994 | uint64_t state; | |
1995 | ||
b128c09f BB |
1996 | if (!vdev_readable(vd)) |
1997 | return (0); | |
1998 | ||
34dc7c2f BB |
1999 | if ((label = vdev_label_read_config(vd)) == NULL) { |
2000 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2001 | VDEV_AUX_CORRUPT_DATA); | |
2002 | return (-1); | |
2003 | } | |
2004 | ||
2005 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
2006 | version > SPA_VERSION || | |
2007 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || | |
2008 | guid != vd->vdev_guid || | |
2009 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2010 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2011 | VDEV_AUX_CORRUPT_DATA); | |
2012 | nvlist_free(label); | |
2013 | return (-1); | |
2014 | } | |
2015 | ||
2016 | /* | |
2017 | * We don't actually check the pool state here. If it's in fact in | |
2018 | * use by another pool, we update this fact on the fly when requested. | |
2019 | */ | |
2020 | nvlist_free(label); | |
2021 | return (0); | |
2022 | } | |
2023 | ||
428870ff BB |
2024 | void |
2025 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2026 | { | |
2027 | spa_t *spa = vd->vdev_spa; | |
2028 | objset_t *mos = spa->spa_meta_objset; | |
2029 | dmu_tx_t *tx; | |
d6320ddb | 2030 | int m; |
428870ff BB |
2031 | |
2032 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
2033 | ||
2034 | if (vd->vdev_dtl_smo.smo_object) { | |
2035 | ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); | |
2036 | (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); | |
2037 | vd->vdev_dtl_smo.smo_object = 0; | |
2038 | } | |
2039 | ||
2040 | if (vd->vdev_ms != NULL) { | |
d6320ddb | 2041 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2042 | metaslab_t *msp = vd->vdev_ms[m]; |
2043 | ||
2044 | if (msp == NULL || msp->ms_smo.smo_object == 0) | |
2045 | continue; | |
2046 | ||
2047 | ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); | |
2048 | (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); | |
2049 | msp->ms_smo.smo_object = 0; | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | if (vd->vdev_ms_array) { | |
2054 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2055 | vd->vdev_ms_array = 0; | |
2056 | vd->vdev_ms_shift = 0; | |
2057 | } | |
2058 | dmu_tx_commit(tx); | |
2059 | } | |
2060 | ||
34dc7c2f BB |
2061 | void |
2062 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2063 | { | |
2064 | metaslab_t *msp; | |
428870ff BB |
2065 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2066 | ||
2067 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2068 | |
c65aa5b2 | 2069 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2070 | metaslab_sync_done(msp, txg); |
428870ff BB |
2071 | |
2072 | if (reassess) | |
2073 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2074 | } |
2075 | ||
2076 | void | |
2077 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2078 | { | |
2079 | spa_t *spa = vd->vdev_spa; | |
2080 | vdev_t *lvd; | |
2081 | metaslab_t *msp; | |
2082 | dmu_tx_t *tx; | |
2083 | ||
428870ff BB |
2084 | ASSERT(!vd->vdev_ishole); |
2085 | ||
34dc7c2f BB |
2086 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2087 | ASSERT(vd == vd->vdev_top); | |
2088 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2089 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2090 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2091 | ASSERT(vd->vdev_ms_array != 0); | |
2092 | vdev_config_dirty(vd); | |
2093 | dmu_tx_commit(tx); | |
2094 | } | |
2095 | ||
428870ff BB |
2096 | /* |
2097 | * Remove the metadata associated with this vdev once it's empty. | |
2098 | */ | |
2099 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2100 | vdev_remove(vd, txg); | |
2101 | ||
34dc7c2f BB |
2102 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2103 | metaslab_sync(msp, txg); | |
2104 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2105 | } | |
2106 | ||
2107 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2108 | vdev_dtl_sync(lvd, txg); | |
2109 | ||
2110 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2111 | } | |
2112 | ||
2113 | uint64_t | |
2114 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2115 | { | |
2116 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2117 | } | |
2118 | ||
34dc7c2f BB |
2119 | /* |
2120 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2121 | * not be opened, and no I/O is attempted. | |
2122 | */ | |
2123 | int | |
428870ff | 2124 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2125 | { |
572e2857 | 2126 | vdev_t *vd, *tvd; |
34dc7c2f | 2127 | |
428870ff | 2128 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2129 | |
b128c09f BB |
2130 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2131 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2132 | |
34dc7c2f | 2133 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2134 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2135 | |
572e2857 BB |
2136 | tvd = vd->vdev_top; |
2137 | ||
428870ff BB |
2138 | /* |
2139 | * We don't directly use the aux state here, but if we do a | |
2140 | * vdev_reopen(), we need this value to be present to remember why we | |
2141 | * were faulted. | |
2142 | */ | |
2143 | vd->vdev_label_aux = aux; | |
2144 | ||
34dc7c2f BB |
2145 | /* |
2146 | * Faulted state takes precedence over degraded. | |
2147 | */ | |
428870ff | 2148 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2149 | vd->vdev_faulted = 1ULL; |
2150 | vd->vdev_degraded = 0ULL; | |
428870ff | 2151 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2152 | |
2153 | /* | |
428870ff BB |
2154 | * If this device has the only valid copy of the data, then |
2155 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2156 | */ |
572e2857 | 2157 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2158 | vd->vdev_degraded = 1ULL; |
2159 | vd->vdev_faulted = 0ULL; | |
2160 | ||
2161 | /* | |
2162 | * If we reopen the device and it's not dead, only then do we | |
2163 | * mark it degraded. | |
2164 | */ | |
572e2857 | 2165 | vdev_reopen(tvd); |
34dc7c2f | 2166 | |
428870ff BB |
2167 | if (vdev_readable(vd)) |
2168 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2169 | } |
2170 | ||
b128c09f | 2171 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2172 | } |
2173 | ||
2174 | /* | |
2175 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2176 | * user that something is wrong. The vdev continues to operate as normal as far | |
2177 | * as I/O is concerned. | |
2178 | */ | |
2179 | int | |
428870ff | 2180 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2181 | { |
b128c09f | 2182 | vdev_t *vd; |
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 | 2188 | |
34dc7c2f | 2189 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2190 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2191 | |
2192 | /* | |
2193 | * If the vdev is already faulted, then don't do anything. | |
2194 | */ | |
b128c09f BB |
2195 | if (vd->vdev_faulted || vd->vdev_degraded) |
2196 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2197 | |
2198 | vd->vdev_degraded = 1ULL; | |
2199 | if (!vdev_is_dead(vd)) | |
2200 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2201 | aux); |
34dc7c2f | 2202 | |
b128c09f | 2203 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2204 | } |
2205 | ||
2206 | /* | |
2207 | * Online the given vdev. If 'unspare' is set, it implies two things. First, | |
2208 | * any attached spare device should be detached when the device finishes | |
2209 | * resilvering. Second, the online should be treated like a 'test' online case, | |
2210 | * so no FMA events are generated if the device fails to open. | |
2211 | */ | |
2212 | int | |
b128c09f | 2213 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2214 | { |
9babb374 | 2215 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
34dc7c2f | 2216 | |
428870ff | 2217 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2218 | |
b128c09f BB |
2219 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2220 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2221 | |
2222 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2223 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2224 | |
9babb374 | 2225 | tvd = vd->vdev_top; |
34dc7c2f BB |
2226 | vd->vdev_offline = B_FALSE; |
2227 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2228 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2229 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2230 | |
2231 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2232 | if (!vd->vdev_aux) { | |
2233 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2234 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2235 | } | |
2236 | ||
2237 | vdev_reopen(tvd); | |
34dc7c2f BB |
2238 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2239 | ||
9babb374 BB |
2240 | if (!vd->vdev_aux) { |
2241 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2242 | pvd->vdev_expanding = B_FALSE; | |
2243 | } | |
2244 | ||
34dc7c2f BB |
2245 | if (newstate) |
2246 | *newstate = vd->vdev_state; | |
2247 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2248 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2249 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2250 | vd->vdev_parent->vdev_child[0] == vd) | |
2251 | vd->vdev_unspare = B_TRUE; | |
2252 | ||
9babb374 BB |
2253 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2254 | ||
2255 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2256 | if (vd->vdev_aux) | |
2257 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2258 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2259 | } | |
fb5f0bc8 | 2260 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2261 | } |
2262 | ||
428870ff BB |
2263 | static int |
2264 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2265 | { |
9babb374 | 2266 | vdev_t *vd, *tvd; |
428870ff BB |
2267 | int error = 0; |
2268 | uint64_t generation; | |
2269 | metaslab_group_t *mg; | |
34dc7c2f | 2270 | |
428870ff BB |
2271 | top: |
2272 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2273 | |
b128c09f BB |
2274 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2275 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2276 | |
2277 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2278 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2279 | |
9babb374 | 2280 | tvd = vd->vdev_top; |
428870ff BB |
2281 | mg = tvd->vdev_mg; |
2282 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2283 | |
34dc7c2f BB |
2284 | /* |
2285 | * If the device isn't already offline, try to offline it. | |
2286 | */ | |
2287 | if (!vd->vdev_offline) { | |
2288 | /* | |
fb5f0bc8 | 2289 | * If this device has the only valid copy of some data, |
9babb374 BB |
2290 | * don't allow it to be offlined. Log devices are always |
2291 | * expendable. | |
34dc7c2f | 2292 | */ |
9babb374 BB |
2293 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2294 | vdev_dtl_required(vd)) | |
b128c09f | 2295 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2296 | |
428870ff BB |
2297 | /* |
2298 | * If the top-level is a slog and it has had allocations | |
2299 | * then proceed. We check that the vdev's metaslab group | |
2300 | * is not NULL since it's possible that we may have just | |
2301 | * added this vdev but not yet initialized its metaslabs. | |
2302 | */ | |
2303 | if (tvd->vdev_islog && mg != NULL) { | |
2304 | /* | |
2305 | * Prevent any future allocations. | |
2306 | */ | |
2307 | metaslab_group_passivate(mg); | |
2308 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2309 | ||
2310 | error = spa_offline_log(spa); | |
2311 | ||
2312 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2313 | ||
2314 | /* | |
2315 | * Check to see if the config has changed. | |
2316 | */ | |
2317 | if (error || generation != spa->spa_config_generation) { | |
2318 | metaslab_group_activate(mg); | |
2319 | if (error) | |
2320 | return (spa_vdev_state_exit(spa, | |
2321 | vd, error)); | |
2322 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2323 | goto top; | |
2324 | } | |
2325 | ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); | |
2326 | } | |
2327 | ||
34dc7c2f BB |
2328 | /* |
2329 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2330 | * If the top-level vdev is a log device then just offline |
2331 | * it. Otherwise, if this action results in the top-level | |
2332 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2333 | */ |
2334 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2335 | vdev_reopen(tvd); |
2336 | ||
2337 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2338 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2339 | vd->vdev_offline = B_FALSE; |
9babb374 | 2340 | vdev_reopen(tvd); |
b128c09f | 2341 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2342 | } |
428870ff BB |
2343 | |
2344 | /* | |
2345 | * Add the device back into the metaslab rotor so that | |
2346 | * once we online the device it's open for business. | |
2347 | */ | |
2348 | if (tvd->vdev_islog && mg != NULL) | |
2349 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2350 | } |
2351 | ||
b128c09f | 2352 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2353 | |
428870ff BB |
2354 | return (spa_vdev_state_exit(spa, vd, 0)); |
2355 | } | |
9babb374 | 2356 | |
428870ff BB |
2357 | int |
2358 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2359 | { | |
2360 | int error; | |
9babb374 | 2361 | |
428870ff BB |
2362 | mutex_enter(&spa->spa_vdev_top_lock); |
2363 | error = vdev_offline_locked(spa, guid, flags); | |
2364 | mutex_exit(&spa->spa_vdev_top_lock); | |
2365 | ||
2366 | return (error); | |
34dc7c2f BB |
2367 | } |
2368 | ||
2369 | /* | |
2370 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2371 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2372 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2373 | */ |
2374 | void | |
b128c09f | 2375 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2376 | { |
b128c09f | 2377 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2378 | int c; |
b128c09f BB |
2379 | |
2380 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2381 | |
2382 | if (vd == NULL) | |
b128c09f | 2383 | vd = rvd; |
34dc7c2f BB |
2384 | |
2385 | vd->vdev_stat.vs_read_errors = 0; | |
2386 | vd->vdev_stat.vs_write_errors = 0; | |
2387 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2388 | |
d6320ddb | 2389 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2390 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2391 | |
2392 | /* | |
b128c09f BB |
2393 | * If we're in the FAULTED state or have experienced failed I/O, then |
2394 | * clear the persistent state and attempt to reopen the device. We | |
2395 | * also mark the vdev config dirty, so that the new faulted state is | |
2396 | * written out to disk. | |
34dc7c2f | 2397 | */ |
b128c09f BB |
2398 | if (vd->vdev_faulted || vd->vdev_degraded || |
2399 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2400 | ||
428870ff BB |
2401 | /* |
2402 | * When reopening in reponse to a clear event, it may be due to | |
2403 | * a fmadm repair request. In this case, if the device is | |
2404 | * still broken, we want to still post the ereport again. | |
2405 | */ | |
2406 | vd->vdev_forcefault = B_TRUE; | |
2407 | ||
572e2857 | 2408 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2409 | vd->vdev_cant_read = B_FALSE; |
2410 | vd->vdev_cant_write = B_FALSE; | |
2411 | ||
572e2857 | 2412 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2413 | |
428870ff BB |
2414 | vd->vdev_forcefault = B_FALSE; |
2415 | ||
572e2857 | 2416 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2417 | vdev_state_dirty(vd->vdev_top); |
2418 | ||
2419 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2420 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2421 | ||
26685276 | 2422 | spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR); |
34dc7c2f | 2423 | } |
428870ff BB |
2424 | |
2425 | /* | |
2426 | * When clearing a FMA-diagnosed fault, we always want to | |
2427 | * unspare the device, as we assume that the original spare was | |
2428 | * done in response to the FMA fault. | |
2429 | */ | |
2430 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2431 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2432 | vd->vdev_parent->vdev_child[0] == vd) | |
2433 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2434 | } |
2435 | ||
b128c09f BB |
2436 | boolean_t |
2437 | vdev_is_dead(vdev_t *vd) | |
2438 | { | |
428870ff BB |
2439 | /* |
2440 | * Holes and missing devices are always considered "dead". | |
2441 | * This simplifies the code since we don't have to check for | |
2442 | * these types of devices in the various code paths. | |
2443 | * Instead we rely on the fact that we skip over dead devices | |
2444 | * before issuing I/O to them. | |
2445 | */ | |
2446 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2447 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2448 | } |
2449 | ||
2450 | boolean_t | |
34dc7c2f BB |
2451 | vdev_readable(vdev_t *vd) |
2452 | { | |
b128c09f | 2453 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2454 | } |
2455 | ||
b128c09f | 2456 | boolean_t |
34dc7c2f BB |
2457 | vdev_writeable(vdev_t *vd) |
2458 | { | |
b128c09f | 2459 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2460 | } |
2461 | ||
b128c09f BB |
2462 | boolean_t |
2463 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2464 | { |
fb5f0bc8 BB |
2465 | uint64_t state = vd->vdev_state; |
2466 | ||
b128c09f | 2467 | /* |
fb5f0bc8 | 2468 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2469 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2470 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2471 | * the proper locks. Note that we have to get the vdev state |
2472 | * in a local variable because although it changes atomically, | |
2473 | * we're asking two separate questions about it. | |
b128c09f | 2474 | */ |
fb5f0bc8 | 2475 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
428870ff | 2476 | !vd->vdev_cant_write && !vd->vdev_ishole); |
34dc7c2f BB |
2477 | } |
2478 | ||
b128c09f BB |
2479 | boolean_t |
2480 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2481 | { |
b128c09f | 2482 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2483 | |
b128c09f BB |
2484 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2485 | return (B_FALSE); | |
34dc7c2f | 2486 | |
b128c09f BB |
2487 | if (zio->io_type == ZIO_TYPE_READ) |
2488 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2489 | |
b128c09f BB |
2490 | if (zio->io_type == ZIO_TYPE_WRITE) |
2491 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2492 | |
b128c09f | 2493 | return (B_TRUE); |
34dc7c2f BB |
2494 | } |
2495 | ||
2496 | /* | |
2497 | * Get statistics for the given vdev. | |
2498 | */ | |
2499 | void | |
2500 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2501 | { | |
2502 | vdev_t *rvd = vd->vdev_spa->spa_root_vdev; | |
d6320ddb | 2503 | int c, t; |
34dc7c2f BB |
2504 | |
2505 | mutex_enter(&vd->vdev_stat_lock); | |
2506 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2507 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2508 | vs->vs_state = vd->vdev_state; | |
9babb374 BB |
2509 | vs->vs_rsize = vdev_get_min_asize(vd); |
2510 | if (vd->vdev_ops->vdev_op_leaf) | |
2511 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
1bd201e7 | 2512 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
34dc7c2f BB |
2513 | mutex_exit(&vd->vdev_stat_lock); |
2514 | ||
2515 | /* | |
2516 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2517 | * over all top-level vdevs (i.e. the direct children of the root). | |
2518 | */ | |
2519 | if (vd == rvd) { | |
d6320ddb | 2520 | for (c = 0; c < rvd->vdev_children; c++) { |
34dc7c2f BB |
2521 | vdev_t *cvd = rvd->vdev_child[c]; |
2522 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2523 | ||
2524 | mutex_enter(&vd->vdev_stat_lock); | |
d6320ddb | 2525 | for (t = 0; t < ZIO_TYPES; t++) { |
34dc7c2f BB |
2526 | vs->vs_ops[t] += cvs->vs_ops[t]; |
2527 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2528 | } | |
428870ff | 2529 | cvs->vs_scan_removing = cvd->vdev_removing; |
34dc7c2f BB |
2530 | mutex_exit(&vd->vdev_stat_lock); |
2531 | } | |
2532 | } | |
2533 | } | |
2534 | ||
2535 | void | |
2536 | vdev_clear_stats(vdev_t *vd) | |
2537 | { | |
2538 | mutex_enter(&vd->vdev_stat_lock); | |
2539 | vd->vdev_stat.vs_space = 0; | |
2540 | vd->vdev_stat.vs_dspace = 0; | |
2541 | vd->vdev_stat.vs_alloc = 0; | |
2542 | mutex_exit(&vd->vdev_stat_lock); | |
2543 | } | |
2544 | ||
428870ff BB |
2545 | void |
2546 | vdev_scan_stat_init(vdev_t *vd) | |
2547 | { | |
2548 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2549 | int c; |
428870ff | 2550 | |
d6320ddb | 2551 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2552 | vdev_scan_stat_init(vd->vdev_child[c]); |
2553 | ||
2554 | mutex_enter(&vd->vdev_stat_lock); | |
2555 | vs->vs_scan_processed = 0; | |
2556 | mutex_exit(&vd->vdev_stat_lock); | |
2557 | } | |
2558 | ||
34dc7c2f | 2559 | void |
b128c09f | 2560 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2561 | { |
fb5f0bc8 BB |
2562 | spa_t *spa = zio->io_spa; |
2563 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2564 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2565 | vdev_t *pvd; |
2566 | uint64_t txg = zio->io_txg; | |
2567 | vdev_stat_t *vs = &vd->vdev_stat; | |
2568 | zio_type_t type = zio->io_type; | |
2569 | int flags = zio->io_flags; | |
2570 | ||
b128c09f BB |
2571 | /* |
2572 | * If this i/o is a gang leader, it didn't do any actual work. | |
2573 | */ | |
2574 | if (zio->io_gang_tree) | |
2575 | return; | |
2576 | ||
34dc7c2f | 2577 | if (zio->io_error == 0) { |
b128c09f BB |
2578 | /* |
2579 | * If this is a root i/o, don't count it -- we've already | |
2580 | * counted the top-level vdevs, and vdev_get_stats() will | |
2581 | * aggregate them when asked. This reduces contention on | |
2582 | * the root vdev_stat_lock and implicitly handles blocks | |
2583 | * that compress away to holes, for which there is no i/o. | |
2584 | * (Holes never create vdev children, so all the counters | |
2585 | * remain zero, which is what we want.) | |
2586 | * | |
2587 | * Note: this only applies to successful i/o (io_error == 0) | |
2588 | * because unlike i/o counts, errors are not additive. | |
2589 | * When reading a ditto block, for example, failure of | |
2590 | * one top-level vdev does not imply a root-level error. | |
2591 | */ | |
2592 | if (vd == rvd) | |
2593 | return; | |
2594 | ||
2595 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2596 | |
2597 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2598 | return; | |
2599 | ||
2600 | mutex_enter(&vd->vdev_stat_lock); | |
2601 | ||
b128c09f | 2602 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 2603 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
2604 | dsl_scan_phys_t *scn_phys = |
2605 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2606 | uint64_t *processed = &scn_phys->scn_processed; | |
2607 | ||
2608 | /* XXX cleanup? */ | |
2609 | if (vd->vdev_ops->vdev_op_leaf) | |
2610 | atomic_add_64(processed, psize); | |
2611 | vs->vs_scan_processed += psize; | |
2612 | } | |
2613 | ||
fb5f0bc8 | 2614 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 2615 | vs->vs_self_healed += psize; |
34dc7c2f | 2616 | } |
fb5f0bc8 BB |
2617 | |
2618 | vs->vs_ops[type]++; | |
2619 | vs->vs_bytes[type] += psize; | |
2620 | ||
2621 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
2622 | return; |
2623 | } | |
2624 | ||
2625 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2626 | return; | |
2627 | ||
9babb374 BB |
2628 | /* |
2629 | * If this is an I/O error that is going to be retried, then ignore the | |
2630 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2631 | * hard errors, when in reality they can happen for any number of | |
2632 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2633 | */ | |
2634 | if (zio->io_error == EIO && | |
2635 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2636 | return; | |
2637 | ||
428870ff BB |
2638 | /* |
2639 | * Intent logs writes won't propagate their error to the root | |
2640 | * I/O so don't mark these types of failures as pool-level | |
2641 | * errors. | |
2642 | */ | |
2643 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2644 | return; | |
2645 | ||
b128c09f | 2646 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 2647 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
2648 | if (zio->io_error == ECKSUM) |
2649 | vs->vs_checksum_errors++; | |
2650 | else | |
2651 | vs->vs_read_errors++; | |
34dc7c2f | 2652 | } |
9babb374 | 2653 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
2654 | vs->vs_write_errors++; |
2655 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 2656 | |
fb5f0bc8 BB |
2657 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
2658 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 2659 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 2660 | spa->spa_claiming)) { |
fb5f0bc8 | 2661 | /* |
428870ff BB |
2662 | * This is either a normal write (not a repair), or it's |
2663 | * a repair induced by the scrub thread, or it's a repair | |
2664 | * made by zil_claim() during spa_load() in the first txg. | |
2665 | * In the normal case, we commit the DTL change in the same | |
2666 | * txg as the block was born. In the scrub-induced repair | |
2667 | * case, we know that scrubs run in first-pass syncing context, | |
2668 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2669 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
2670 | * |
2671 | * We currently do not make DTL entries for failed spontaneous | |
2672 | * self-healing writes triggered by normal (non-scrubbing) | |
2673 | * reads, because we have no transactional context in which to | |
2674 | * do so -- and it's not clear that it'd be desirable anyway. | |
2675 | */ | |
2676 | if (vd->vdev_ops->vdev_op_leaf) { | |
2677 | uint64_t commit_txg = txg; | |
572e2857 | 2678 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
2679 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
2680 | ASSERT(spa_sync_pass(spa) == 1); | |
2681 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
2682 | commit_txg = spa_syncing_txg(spa); |
2683 | } else if (spa->spa_claiming) { | |
2684 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2685 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 2686 | } |
428870ff | 2687 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 2688 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 2689 | return; |
fb5f0bc8 BB |
2690 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
2691 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2692 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 2693 | } |
fb5f0bc8 BB |
2694 | if (vd != rvd) |
2695 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
2696 | } |
2697 | } | |
2698 | ||
34dc7c2f | 2699 | /* |
428870ff BB |
2700 | * Update the in-core space usage stats for this vdev, its metaslab class, |
2701 | * and the root vdev. | |
34dc7c2f BB |
2702 | */ |
2703 | void | |
428870ff BB |
2704 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
2705 | int64_t space_delta) | |
34dc7c2f BB |
2706 | { |
2707 | int64_t dspace_delta = space_delta; | |
2708 | spa_t *spa = vd->vdev_spa; | |
2709 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
2710 | metaslab_group_t *mg = vd->vdev_mg; |
2711 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
2712 | |
2713 | ASSERT(vd == vd->vdev_top); | |
2714 | ||
2715 | /* | |
2716 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2717 | * factor. We must calculate this here and not at the root vdev | |
2718 | * because the root vdev's psize-to-asize is simply the max of its | |
2719 | * childrens', thus not accurate enough for us. | |
2720 | */ | |
2721 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 2722 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
2723 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
2724 | vd->vdev_deflate_ratio; | |
2725 | ||
2726 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 2727 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2728 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2729 | vd->vdev_stat.vs_dspace += dspace_delta; |
2730 | mutex_exit(&vd->vdev_stat_lock); | |
2731 | ||
428870ff | 2732 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 2733 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 2734 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 2735 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
2736 | rvd->vdev_stat.vs_dspace += dspace_delta; |
2737 | mutex_exit(&rvd->vdev_stat_lock); | |
2738 | } | |
428870ff BB |
2739 | |
2740 | if (mc != NULL) { | |
2741 | ASSERT(rvd == vd->vdev_parent); | |
2742 | ASSERT(vd->vdev_ms_count != 0); | |
2743 | ||
2744 | metaslab_class_space_update(mc, | |
2745 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2746 | } | |
34dc7c2f BB |
2747 | } |
2748 | ||
2749 | /* | |
2750 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2751 | * so that it will be written out next time the vdev configuration is synced. | |
2752 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2753 | */ | |
2754 | void | |
2755 | vdev_config_dirty(vdev_t *vd) | |
2756 | { | |
2757 | spa_t *spa = vd->vdev_spa; | |
2758 | vdev_t *rvd = spa->spa_root_vdev; | |
2759 | int c; | |
2760 | ||
572e2857 BB |
2761 | ASSERT(spa_writeable(spa)); |
2762 | ||
34dc7c2f | 2763 | /* |
9babb374 BB |
2764 | * If this is an aux vdev (as with l2cache and spare devices), then we |
2765 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
2766 | */ |
2767 | if (vd->vdev_aux != NULL) { | |
2768 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2769 | nvlist_t **aux; | |
2770 | uint_t naux; | |
2771 | ||
2772 | for (c = 0; c < sav->sav_count; c++) { | |
2773 | if (sav->sav_vdevs[c] == vd) | |
2774 | break; | |
2775 | } | |
2776 | ||
2777 | if (c == sav->sav_count) { | |
2778 | /* | |
2779 | * We're being removed. There's nothing more to do. | |
2780 | */ | |
2781 | ASSERT(sav->sav_sync == B_TRUE); | |
2782 | return; | |
2783 | } | |
2784 | ||
2785 | sav->sav_sync = B_TRUE; | |
2786 | ||
9babb374 BB |
2787 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
2788 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2789 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2790 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2791 | } | |
b128c09f BB |
2792 | |
2793 | ASSERT(c < naux); | |
2794 | ||
2795 | /* | |
2796 | * Setting the nvlist in the middle if the array is a little | |
2797 | * sketchy, but it will work. | |
2798 | */ | |
2799 | nvlist_free(aux[c]); | |
428870ff | 2800 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
2801 | |
2802 | return; | |
2803 | } | |
2804 | ||
2805 | /* | |
2806 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
2807 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
2808 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
2809 | * so this is sufficient to ensure mutual exclusion. |
2810 | */ | |
b128c09f BB |
2811 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2812 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2813 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
2814 | |
2815 | if (vd == rvd) { | |
2816 | for (c = 0; c < rvd->vdev_children; c++) | |
2817 | vdev_config_dirty(rvd->vdev_child[c]); | |
2818 | } else { | |
2819 | ASSERT(vd == vd->vdev_top); | |
2820 | ||
428870ff BB |
2821 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
2822 | !vd->vdev_ishole) | |
b128c09f | 2823 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
2824 | } |
2825 | } | |
2826 | ||
2827 | void | |
2828 | vdev_config_clean(vdev_t *vd) | |
2829 | { | |
2830 | spa_t *spa = vd->vdev_spa; | |
2831 | ||
b128c09f BB |
2832 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
2833 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2834 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 2835 | |
b128c09f BB |
2836 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
2837 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
2838 | } |
2839 | ||
b128c09f BB |
2840 | /* |
2841 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
2842 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
2843 | * the state changes from larger config changes because they require | |
2844 | * much less locking, and are often needed for administrative actions. | |
2845 | */ | |
2846 | void | |
2847 | vdev_state_dirty(vdev_t *vd) | |
2848 | { | |
2849 | spa_t *spa = vd->vdev_spa; | |
2850 | ||
572e2857 | 2851 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
2852 | ASSERT(vd == vd->vdev_top); |
2853 | ||
2854 | /* | |
2855 | * The state list is protected by the SCL_STATE lock. The caller | |
2856 | * must either hold SCL_STATE as writer, or must be the sync thread | |
2857 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
2858 | * so this is sufficient to ensure mutual exclusion. | |
2859 | */ | |
2860 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2861 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2862 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2863 | ||
428870ff | 2864 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
2865 | list_insert_head(&spa->spa_state_dirty_list, vd); |
2866 | } | |
2867 | ||
2868 | void | |
2869 | vdev_state_clean(vdev_t *vd) | |
2870 | { | |
2871 | spa_t *spa = vd->vdev_spa; | |
2872 | ||
2873 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
2874 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
2875 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
2876 | ||
2877 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
2878 | list_remove(&spa->spa_state_dirty_list, vd); | |
2879 | } | |
2880 | ||
2881 | /* | |
2882 | * Propagate vdev state up from children to parent. | |
2883 | */ | |
34dc7c2f BB |
2884 | void |
2885 | vdev_propagate_state(vdev_t *vd) | |
2886 | { | |
fb5f0bc8 BB |
2887 | spa_t *spa = vd->vdev_spa; |
2888 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
2889 | int degraded = 0, faulted = 0; |
2890 | int corrupted = 0; | |
34dc7c2f | 2891 | vdev_t *child; |
d6320ddb | 2892 | int c; |
34dc7c2f BB |
2893 | |
2894 | if (vd->vdev_children > 0) { | |
d6320ddb | 2895 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 2896 | child = vd->vdev_child[c]; |
b128c09f | 2897 | |
428870ff BB |
2898 | /* |
2899 | * Don't factor holes into the decision. | |
2900 | */ | |
2901 | if (child->vdev_ishole) | |
2902 | continue; | |
2903 | ||
b128c09f | 2904 | if (!vdev_readable(child) || |
fb5f0bc8 | 2905 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
2906 | /* |
2907 | * Root special: if there is a top-level log | |
2908 | * device, treat the root vdev as if it were | |
2909 | * degraded. | |
2910 | */ | |
2911 | if (child->vdev_islog && vd == rvd) | |
2912 | degraded++; | |
2913 | else | |
2914 | faulted++; | |
2915 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 2916 | degraded++; |
b128c09f | 2917 | } |
34dc7c2f BB |
2918 | |
2919 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
2920 | corrupted++; | |
2921 | } | |
2922 | ||
2923 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
2924 | ||
2925 | /* | |
b128c09f | 2926 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
2927 | * opened due to corrupted metadata, then propagate the root |
2928 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
2929 | * replicas'. | |
2930 | */ | |
2931 | if (corrupted && vd == rvd && | |
2932 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
2933 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2934 | VDEV_AUX_CORRUPT_DATA); | |
2935 | } | |
2936 | ||
b128c09f | 2937 | if (vd->vdev_parent) |
34dc7c2f BB |
2938 | vdev_propagate_state(vd->vdev_parent); |
2939 | } | |
2940 | ||
2941 | /* | |
2942 | * Set a vdev's state. If this is during an open, we don't update the parent | |
2943 | * state, because we're in the process of opening children depth-first. | |
2944 | * Otherwise, we propagate the change to the parent. | |
2945 | * | |
2946 | * If this routine places a device in a faulted state, an appropriate ereport is | |
2947 | * generated. | |
2948 | */ | |
2949 | void | |
2950 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
2951 | { | |
2952 | uint64_t save_state; | |
b128c09f | 2953 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
2954 | |
2955 | if (state == vd->vdev_state) { | |
2956 | vd->vdev_stat.vs_aux = aux; | |
2957 | return; | |
2958 | } | |
2959 | ||
2960 | save_state = vd->vdev_state; | |
2961 | ||
2962 | vd->vdev_state = state; | |
2963 | vd->vdev_stat.vs_aux = aux; | |
2964 | ||
2965 | /* | |
2966 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
2967 | * always close the underlying device unless the device has requested |
2968 | * a delayed close (i.e. we're about to remove or fault the device). | |
2969 | * Otherwise, we keep accessible but invalid devices open forever. | |
2970 | * We don't call vdev_close() itself, because that implies some extra | |
2971 | * checks (offline, etc) that we don't want here. This is limited to | |
2972 | * leaf devices, because otherwise closing the device will affect other | |
2973 | * children. | |
34dc7c2f | 2974 | */ |
428870ff BB |
2975 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
2976 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
2977 | vd->vdev_ops->vdev_op_close(vd); |
2978 | ||
428870ff BB |
2979 | /* |
2980 | * If we have brought this vdev back into service, we need | |
2981 | * to notify fmd so that it can gracefully repair any outstanding | |
2982 | * cases due to a missing device. We do this in all cases, even those | |
2983 | * that probably don't correlate to a repaired fault. This is sure to | |
2984 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
2985 | * this is a transient state it's OK, as the retire agent will | |
2986 | * double-check the state of the vdev before repairing it. | |
2987 | */ | |
2988 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
2989 | vd->vdev_prevstate != state) | |
2990 | zfs_post_state_change(spa, vd); | |
2991 | ||
34dc7c2f BB |
2992 | if (vd->vdev_removed && |
2993 | state == VDEV_STATE_CANT_OPEN && | |
2994 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
2995 | /* | |
2996 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
2997 | * device was previously marked removed and someone attempted to | |
2998 | * reopen it. If this failed due to a nonexistent device, then | |
2999 | * keep the device in the REMOVED state. We also let this be if | |
3000 | * it is one of our special test online cases, which is only | |
3001 | * attempting to online the device and shouldn't generate an FMA | |
3002 | * fault. | |
3003 | */ | |
3004 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3005 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3006 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3007 | vd->vdev_removed = B_TRUE; |
3008 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3009 | /* | |
572e2857 BB |
3010 | * If we fail to open a vdev during an import or recovery, we |
3011 | * mark it as "not available", which signifies that it was | |
3012 | * never there to begin with. Failure to open such a device | |
3013 | * is not considered an error. | |
34dc7c2f | 3014 | */ |
572e2857 BB |
3015 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3016 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3017 | vd->vdev_ops->vdev_op_leaf) |
3018 | vd->vdev_not_present = 1; | |
3019 | ||
3020 | /* | |
3021 | * Post the appropriate ereport. If the 'prevstate' field is | |
3022 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3023 | * that this is part of a vdev_reopen(). In this case, we don't | |
3024 | * want to post the ereport if the device was already in the | |
3025 | * CANT_OPEN state beforehand. | |
3026 | * | |
3027 | * If the 'checkremove' flag is set, then this is an attempt to | |
3028 | * online the device in response to an insertion event. If we | |
3029 | * hit this case, then we have detected an insertion event for a | |
3030 | * faulted or offline device that wasn't in the removed state. | |
3031 | * In this scenario, we don't post an ereport because we are | |
3032 | * about to replace the device, or attempt an online with | |
3033 | * vdev_forcefault, which will generate the fault for us. | |
3034 | */ | |
3035 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3036 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3037 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3038 | const char *class; |
3039 | ||
3040 | switch (aux) { | |
3041 | case VDEV_AUX_OPEN_FAILED: | |
3042 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3043 | break; | |
3044 | case VDEV_AUX_CORRUPT_DATA: | |
3045 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3046 | break; | |
3047 | case VDEV_AUX_NO_REPLICAS: | |
3048 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3049 | break; | |
3050 | case VDEV_AUX_BAD_GUID_SUM: | |
3051 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3052 | break; | |
3053 | case VDEV_AUX_TOO_SMALL: | |
3054 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3055 | break; | |
3056 | case VDEV_AUX_BAD_LABEL: | |
3057 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3058 | break; | |
3059 | default: | |
3060 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3061 | } | |
3062 | ||
b128c09f | 3063 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3064 | } |
3065 | ||
3066 | /* Erase any notion of persistent removed state */ | |
3067 | vd->vdev_removed = B_FALSE; | |
3068 | } else { | |
3069 | vd->vdev_removed = B_FALSE; | |
3070 | } | |
3071 | ||
9babb374 BB |
3072 | if (!isopen && vd->vdev_parent) |
3073 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3074 | } |
b128c09f BB |
3075 | |
3076 | /* | |
3077 | * Check the vdev configuration to ensure that it's capable of supporting | |
57a4eddc | 3078 | * a root pool. |
b128c09f BB |
3079 | */ |
3080 | boolean_t | |
3081 | vdev_is_bootable(vdev_t *vd) | |
3082 | { | |
57a4eddc RL |
3083 | #if defined(__sun__) || defined(__sun) |
3084 | /* | |
3085 | * Currently, we do not support RAID-Z or partial configuration. | |
3086 | * In addition, only a single top-level vdev is allowed and none of the | |
3087 | * leaves can be wholedisks. | |
3088 | */ | |
d6320ddb BB |
3089 | int c; |
3090 | ||
b128c09f BB |
3091 | if (!vd->vdev_ops->vdev_op_leaf) { |
3092 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3093 | ||
3094 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3095 | vd->vdev_children > 1) { | |
3096 | return (B_FALSE); | |
3097 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3098 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3099 | return (B_FALSE); | |
3100 | } | |
3101 | } else if (vd->vdev_wholedisk == 1) { | |
3102 | return (B_FALSE); | |
3103 | } | |
3104 | ||
d6320ddb | 3105 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3106 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3107 | return (B_FALSE); | |
3108 | } | |
57a4eddc | 3109 | #endif /* __sun__ || __sun */ |
b128c09f BB |
3110 | return (B_TRUE); |
3111 | } | |
9babb374 | 3112 | |
428870ff BB |
3113 | /* |
3114 | * Load the state from the original vdev tree (ovd) which | |
3115 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3116 | * vdev was offline or faulted then we transfer that state to the |
3117 | * device in the current vdev tree (nvd). | |
428870ff | 3118 | */ |
9babb374 | 3119 | void |
428870ff | 3120 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3121 | { |
d6320ddb | 3122 | int c; |
9babb374 | 3123 | |
572e2857 | 3124 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3125 | ASSERT(spa_config_held(nvd->vdev_spa, |
3126 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3127 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3128 | |
d6320ddb | 3129 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3130 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3131 | |
572e2857 | 3132 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3133 | /* |
572e2857 | 3134 | * Restore the persistent vdev state |
9babb374 | 3135 | */ |
428870ff | 3136 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3137 | nvd->vdev_faulted = ovd->vdev_faulted; |
3138 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3139 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3140 | } |
3141 | } | |
3142 | ||
572e2857 BB |
3143 | /* |
3144 | * Determine if a log device has valid content. If the vdev was | |
3145 | * removed or faulted in the MOS config then we know that | |
3146 | * the content on the log device has already been written to the pool. | |
3147 | */ | |
3148 | boolean_t | |
3149 | vdev_log_state_valid(vdev_t *vd) | |
3150 | { | |
d6320ddb BB |
3151 | int c; |
3152 | ||
572e2857 BB |
3153 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3154 | !vd->vdev_removed) | |
3155 | return (B_TRUE); | |
3156 | ||
d6320ddb | 3157 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3158 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3159 | return (B_TRUE); | |
3160 | ||
3161 | return (B_FALSE); | |
3162 | } | |
3163 | ||
9babb374 BB |
3164 | /* |
3165 | * Expand a vdev if possible. | |
3166 | */ | |
3167 | void | |
3168 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3169 | { | |
3170 | ASSERT(vd->vdev_top == vd); | |
3171 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3172 | ||
3173 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3174 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3175 | vdev_config_dirty(vd); | |
3176 | } | |
3177 | } | |
428870ff BB |
3178 | |
3179 | /* | |
3180 | * Split a vdev. | |
3181 | */ | |
3182 | void | |
3183 | vdev_split(vdev_t *vd) | |
3184 | { | |
3185 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3186 | ||
3187 | vdev_remove_child(pvd, vd); | |
3188 | vdev_compact_children(pvd); | |
3189 | ||
3190 | cvd = pvd->vdev_child[0]; | |
3191 | if (pvd->vdev_children == 1) { | |
3192 | vdev_remove_parent(cvd); | |
3193 | cvd->vdev_splitting = B_TRUE; | |
3194 | } | |
3195 | vdev_propagate_state(cvd); | |
3196 | } | |
c28b2279 BB |
3197 | |
3198 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
3199 | EXPORT_SYMBOL(vdev_fault); | |
3200 | EXPORT_SYMBOL(vdev_degrade); | |
3201 | EXPORT_SYMBOL(vdev_online); | |
3202 | EXPORT_SYMBOL(vdev_offline); | |
3203 | EXPORT_SYMBOL(vdev_clear); | |
c409e464 BB |
3204 | |
3205 | module_param(zfs_scrub_limit, int, 0644); | |
3206 | MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev"); | |
c28b2279 | 3207 | #endif |