]>
Commit | Line | Data |
---|---|---|
70e083d2 TG |
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 | /* | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
25 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. | |
26 | */ | |
27 | ||
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/space_reftree.h> | |
40 | #include <sys/zio.h> | |
41 | #include <sys/zap.h> | |
42 | #include <sys/fs/zfs.h> | |
43 | #include <sys/arc.h> | |
44 | #include <sys/zil.h> | |
45 | #include <sys/dsl_scan.h> | |
46 | #include <sys/zvol.h> | |
47 | ||
48 | /* | |
49 | * When a vdev is added, it will be divided into approximately (but no | |
50 | * more than) this number of metaslabs. | |
51 | */ | |
52 | int metaslabs_per_vdev = 200; | |
53 | ||
54 | /* | |
55 | * Virtual device management. | |
56 | */ | |
57 | ||
58 | static vdev_ops_t *vdev_ops_table[] = { | |
59 | &vdev_root_ops, | |
60 | &vdev_raidz_ops, | |
61 | &vdev_mirror_ops, | |
62 | &vdev_replacing_ops, | |
63 | &vdev_spare_ops, | |
64 | &vdev_disk_ops, | |
65 | &vdev_file_ops, | |
66 | &vdev_missing_ops, | |
67 | &vdev_hole_ops, | |
68 | NULL | |
69 | }; | |
70 | ||
71 | /* | |
72 | * Given a vdev type, return the appropriate ops vector. | |
73 | */ | |
74 | static vdev_ops_t * | |
75 | vdev_getops(const char *type) | |
76 | { | |
77 | vdev_ops_t *ops, **opspp; | |
78 | ||
79 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
80 | if (strcmp(ops->vdev_op_type, type) == 0) | |
81 | break; | |
82 | ||
83 | return (ops); | |
84 | } | |
85 | ||
86 | /* | |
87 | * Default asize function: return the MAX of psize with the asize of | |
88 | * all children. This is what's used by anything other than RAID-Z. | |
89 | */ | |
90 | uint64_t | |
91 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
92 | { | |
93 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
94 | uint64_t csize; | |
95 | int c; | |
96 | ||
97 | for (c = 0; c < vd->vdev_children; c++) { | |
98 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); | |
99 | asize = MAX(asize, csize); | |
100 | } | |
101 | ||
102 | return (asize); | |
103 | } | |
104 | ||
105 | /* | |
106 | * Get the minimum allocatable size. We define the allocatable size as | |
107 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
108 | * replace or attach devices which don't have the same physical size but | |
109 | * can still satisfy the same number of allocations. | |
110 | */ | |
111 | uint64_t | |
112 | vdev_get_min_asize(vdev_t *vd) | |
113 | { | |
114 | vdev_t *pvd = vd->vdev_parent; | |
115 | ||
116 | /* | |
117 | * If our parent is NULL (inactive spare or cache) or is the root, | |
118 | * just return our own asize. | |
119 | */ | |
120 | if (pvd == NULL) | |
121 | return (vd->vdev_asize); | |
122 | ||
123 | /* | |
124 | * The top-level vdev just returns the allocatable size rounded | |
125 | * to the nearest metaslab. | |
126 | */ | |
127 | if (vd == vd->vdev_top) | |
128 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
129 | ||
130 | /* | |
131 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
132 | * so each child must provide at least 1/Nth of its asize. | |
133 | */ | |
134 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
135 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
136 | ||
137 | return (pvd->vdev_min_asize); | |
138 | } | |
139 | ||
140 | void | |
141 | vdev_set_min_asize(vdev_t *vd) | |
142 | { | |
143 | int c; | |
144 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
145 | ||
146 | for (c = 0; c < vd->vdev_children; c++) | |
147 | vdev_set_min_asize(vd->vdev_child[c]); | |
148 | } | |
149 | ||
150 | vdev_t * | |
151 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
152 | { | |
153 | vdev_t *rvd = spa->spa_root_vdev; | |
154 | ||
155 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); | |
156 | ||
157 | if (vdev < rvd->vdev_children) { | |
158 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
159 | return (rvd->vdev_child[vdev]); | |
160 | } | |
161 | ||
162 | return (NULL); | |
163 | } | |
164 | ||
165 | vdev_t * | |
166 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
167 | { | |
168 | vdev_t *mvd; | |
169 | int c; | |
170 | ||
171 | if (vd->vdev_guid == guid) | |
172 | return (vd); | |
173 | ||
174 | for (c = 0; c < vd->vdev_children; c++) | |
175 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != | |
176 | NULL) | |
177 | return (mvd); | |
178 | ||
179 | return (NULL); | |
180 | } | |
181 | ||
182 | static int | |
183 | vdev_count_leaves_impl(vdev_t *vd) | |
184 | { | |
185 | int n = 0; | |
186 | int c; | |
187 | ||
188 | if (vd->vdev_ops->vdev_op_leaf) | |
189 | return (1); | |
190 | ||
191 | for (c = 0; c < vd->vdev_children; c++) | |
192 | n += vdev_count_leaves_impl(vd->vdev_child[c]); | |
193 | ||
194 | return (n); | |
195 | } | |
196 | ||
197 | int | |
198 | vdev_count_leaves(spa_t *spa) | |
199 | { | |
200 | return (vdev_count_leaves_impl(spa->spa_root_vdev)); | |
201 | } | |
202 | ||
203 | void | |
204 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
205 | { | |
206 | size_t oldsize, newsize; | |
207 | uint64_t id = cvd->vdev_id; | |
208 | vdev_t **newchild; | |
209 | ||
210 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
211 | ASSERT(cvd->vdev_parent == NULL); | |
212 | ||
213 | cvd->vdev_parent = pvd; | |
214 | ||
215 | if (pvd == NULL) | |
216 | return; | |
217 | ||
218 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
219 | ||
220 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
221 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
222 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
223 | ||
224 | newchild = kmem_alloc(newsize, KM_SLEEP); | |
225 | if (pvd->vdev_child != NULL) { | |
226 | bcopy(pvd->vdev_child, newchild, oldsize); | |
227 | kmem_free(pvd->vdev_child, oldsize); | |
228 | } | |
229 | ||
230 | pvd->vdev_child = newchild; | |
231 | pvd->vdev_child[id] = cvd; | |
232 | ||
233 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
234 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
235 | ||
236 | /* | |
237 | * Walk up all ancestors to update guid sum. | |
238 | */ | |
239 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
240 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
241 | } | |
242 | ||
243 | void | |
244 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
245 | { | |
246 | int c; | |
247 | uint_t id = cvd->vdev_id; | |
248 | ||
249 | ASSERT(cvd->vdev_parent == pvd); | |
250 | ||
251 | if (pvd == NULL) | |
252 | return; | |
253 | ||
254 | ASSERT(id < pvd->vdev_children); | |
255 | ASSERT(pvd->vdev_child[id] == cvd); | |
256 | ||
257 | pvd->vdev_child[id] = NULL; | |
258 | cvd->vdev_parent = NULL; | |
259 | ||
260 | for (c = 0; c < pvd->vdev_children; c++) | |
261 | if (pvd->vdev_child[c]) | |
262 | break; | |
263 | ||
264 | if (c == pvd->vdev_children) { | |
265 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
266 | pvd->vdev_child = NULL; | |
267 | pvd->vdev_children = 0; | |
268 | } | |
269 | ||
270 | /* | |
271 | * Walk up all ancestors to update guid sum. | |
272 | */ | |
273 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
274 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
275 | } | |
276 | ||
277 | /* | |
278 | * Remove any holes in the child array. | |
279 | */ | |
280 | void | |
281 | vdev_compact_children(vdev_t *pvd) | |
282 | { | |
283 | vdev_t **newchild, *cvd; | |
284 | int oldc = pvd->vdev_children; | |
285 | int newc; | |
286 | int c; | |
287 | ||
288 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
289 | ||
290 | for (c = newc = 0; c < oldc; c++) | |
291 | if (pvd->vdev_child[c]) | |
292 | newc++; | |
293 | ||
294 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); | |
295 | ||
296 | for (c = newc = 0; c < oldc; c++) { | |
297 | if ((cvd = pvd->vdev_child[c]) != NULL) { | |
298 | newchild[newc] = cvd; | |
299 | cvd->vdev_id = newc++; | |
300 | } | |
301 | } | |
302 | ||
303 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
304 | pvd->vdev_child = newchild; | |
305 | pvd->vdev_children = newc; | |
306 | } | |
307 | ||
308 | /* | |
309 | * Allocate and minimally initialize a vdev_t. | |
310 | */ | |
311 | vdev_t * | |
312 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) | |
313 | { | |
314 | vdev_t *vd; | |
315 | int t; | |
316 | ||
317 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); | |
318 | ||
319 | if (spa->spa_root_vdev == NULL) { | |
320 | ASSERT(ops == &vdev_root_ops); | |
321 | spa->spa_root_vdev = vd; | |
322 | spa->spa_load_guid = spa_generate_guid(NULL); | |
323 | } | |
324 | ||
325 | if (guid == 0 && ops != &vdev_hole_ops) { | |
326 | if (spa->spa_root_vdev == vd) { | |
327 | /* | |
328 | * The root vdev's guid will also be the pool guid, | |
329 | * which must be unique among all pools. | |
330 | */ | |
331 | guid = spa_generate_guid(NULL); | |
332 | } else { | |
333 | /* | |
334 | * Any other vdev's guid must be unique within the pool. | |
335 | */ | |
336 | guid = spa_generate_guid(spa); | |
337 | } | |
338 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
339 | } | |
340 | ||
341 | vd->vdev_spa = spa; | |
342 | vd->vdev_id = id; | |
343 | vd->vdev_guid = guid; | |
344 | vd->vdev_guid_sum = guid; | |
345 | vd->vdev_ops = ops; | |
346 | vd->vdev_state = VDEV_STATE_CLOSED; | |
347 | vd->vdev_ishole = (ops == &vdev_hole_ops); | |
348 | ||
349 | list_link_init(&vd->vdev_config_dirty_node); | |
350 | list_link_init(&vd->vdev_state_dirty_node); | |
351 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); | |
352 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); | |
353 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); | |
354 | for (t = 0; t < DTL_TYPES; t++) { | |
355 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL, | |
356 | &vd->vdev_dtl_lock); | |
357 | } | |
358 | txg_list_create(&vd->vdev_ms_list, | |
359 | offsetof(struct metaslab, ms_txg_node)); | |
360 | txg_list_create(&vd->vdev_dtl_list, | |
361 | offsetof(struct vdev, vdev_dtl_node)); | |
362 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
363 | vdev_queue_init(vd); | |
364 | vdev_cache_init(vd); | |
365 | ||
366 | return (vd); | |
367 | } | |
368 | ||
369 | /* | |
370 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
371 | * creating a new vdev or loading an existing one - the behavior is slightly | |
372 | * different for each case. | |
373 | */ | |
374 | int | |
375 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
376 | int alloctype) | |
377 | { | |
378 | vdev_ops_t *ops; | |
379 | char *type; | |
380 | uint64_t guid = 0, islog, nparity; | |
381 | vdev_t *vd; | |
382 | ||
383 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
384 | ||
385 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
386 | return (SET_ERROR(EINVAL)); | |
387 | ||
388 | if ((ops = vdev_getops(type)) == NULL) | |
389 | return (SET_ERROR(EINVAL)); | |
390 | ||
391 | /* | |
392 | * If this is a load, get the vdev guid from the nvlist. | |
393 | * Otherwise, vdev_alloc_common() will generate one for us. | |
394 | */ | |
395 | if (alloctype == VDEV_ALLOC_LOAD) { | |
396 | uint64_t label_id; | |
397 | ||
398 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
399 | label_id != id) | |
400 | return (SET_ERROR(EINVAL)); | |
401 | ||
402 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
403 | return (SET_ERROR(EINVAL)); | |
404 | } else if (alloctype == VDEV_ALLOC_SPARE) { | |
405 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
406 | return (SET_ERROR(EINVAL)); | |
407 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { | |
408 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
409 | return (SET_ERROR(EINVAL)); | |
410 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
411 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
412 | return (SET_ERROR(EINVAL)); | |
413 | } | |
414 | ||
415 | /* | |
416 | * The first allocated vdev must be of type 'root'. | |
417 | */ | |
418 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
419 | return (SET_ERROR(EINVAL)); | |
420 | ||
421 | /* | |
422 | * Determine whether we're a log vdev. | |
423 | */ | |
424 | islog = 0; | |
425 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
426 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
427 | return (SET_ERROR(ENOTSUP)); | |
428 | ||
429 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) | |
430 | return (SET_ERROR(ENOTSUP)); | |
431 | ||
432 | /* | |
433 | * Set the nparity property for RAID-Z vdevs. | |
434 | */ | |
435 | nparity = -1ULL; | |
436 | if (ops == &vdev_raidz_ops) { | |
437 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
438 | &nparity) == 0) { | |
439 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) | |
440 | return (SET_ERROR(EINVAL)); | |
441 | /* | |
442 | * Previous versions could only support 1 or 2 parity | |
443 | * device. | |
444 | */ | |
445 | if (nparity > 1 && | |
446 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
447 | return (SET_ERROR(ENOTSUP)); | |
448 | if (nparity > 2 && | |
449 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
450 | return (SET_ERROR(ENOTSUP)); | |
451 | } else { | |
452 | /* | |
453 | * We require the parity to be specified for SPAs that | |
454 | * support multiple parity levels. | |
455 | */ | |
456 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) | |
457 | return (SET_ERROR(EINVAL)); | |
458 | /* | |
459 | * Otherwise, we default to 1 parity device for RAID-Z. | |
460 | */ | |
461 | nparity = 1; | |
462 | } | |
463 | } else { | |
464 | nparity = 0; | |
465 | } | |
466 | ASSERT(nparity != -1ULL); | |
467 | ||
468 | vd = vdev_alloc_common(spa, id, guid, ops); | |
469 | ||
470 | vd->vdev_islog = islog; | |
471 | vd->vdev_nparity = nparity; | |
472 | ||
473 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
474 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
475 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
476 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
477 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
478 | &vd->vdev_physpath) == 0) | |
479 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
480 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) | |
481 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
482 | ||
483 | /* | |
484 | * Set the whole_disk property. If it's not specified, leave the value | |
485 | * as -1. | |
486 | */ | |
487 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
488 | &vd->vdev_wholedisk) != 0) | |
489 | vd->vdev_wholedisk = -1ULL; | |
490 | ||
491 | /* | |
492 | * Look for the 'not present' flag. This will only be set if the device | |
493 | * was not present at the time of import. | |
494 | */ | |
495 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, | |
496 | &vd->vdev_not_present); | |
497 | ||
498 | /* | |
499 | * Get the alignment requirement. | |
500 | */ | |
501 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
502 | ||
503 | /* | |
504 | * Retrieve the vdev creation time. | |
505 | */ | |
506 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
507 | &vd->vdev_crtxg); | |
508 | ||
509 | /* | |
510 | * If we're a top-level vdev, try to load the allocation parameters. | |
511 | */ | |
512 | if (parent && !parent->vdev_parent && | |
513 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
514 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, | |
515 | &vd->vdev_ms_array); | |
516 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
517 | &vd->vdev_ms_shift); | |
518 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
519 | &vd->vdev_asize); | |
520 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, | |
521 | &vd->vdev_removing); | |
522 | } | |
523 | ||
524 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { | |
525 | ASSERT(alloctype == VDEV_ALLOC_LOAD || | |
526 | alloctype == VDEV_ALLOC_ADD || | |
527 | alloctype == VDEV_ALLOC_SPLIT || | |
528 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
529 | vd->vdev_mg = metaslab_group_create(islog ? | |
530 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
531 | } | |
532 | ||
533 | /* | |
534 | * If we're a leaf vdev, try to load the DTL object and other state. | |
535 | */ | |
536 | if (vd->vdev_ops->vdev_op_leaf && | |
537 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || | |
538 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
539 | if (alloctype == VDEV_ALLOC_LOAD) { | |
540 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
541 | &vd->vdev_dtl_object); | |
542 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, | |
543 | &vd->vdev_unspare); | |
544 | } | |
545 | ||
546 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
547 | uint64_t spare = 0; | |
548 | ||
549 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
550 | &spare) == 0 && spare) | |
551 | spa_spare_add(vd); | |
552 | } | |
553 | ||
554 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, | |
555 | &vd->vdev_offline); | |
556 | ||
557 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, | |
558 | &vd->vdev_resilver_txg); | |
559 | ||
560 | /* | |
561 | * When importing a pool, we want to ignore the persistent fault | |
562 | * state, as the diagnosis made on another system may not be | |
563 | * valid in the current context. Local vdevs will | |
564 | * remain in the faulted state. | |
565 | */ | |
566 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { | |
567 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, | |
568 | &vd->vdev_faulted); | |
569 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
570 | &vd->vdev_degraded); | |
571 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
572 | &vd->vdev_removed); | |
573 | ||
574 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
575 | char *aux; | |
576 | ||
577 | vd->vdev_label_aux = | |
578 | VDEV_AUX_ERR_EXCEEDED; | |
579 | if (nvlist_lookup_string(nv, | |
580 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
581 | strcmp(aux, "external") == 0) | |
582 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
583 | } | |
584 | } | |
585 | } | |
586 | ||
587 | /* | |
588 | * Add ourselves to the parent's list of children. | |
589 | */ | |
590 | vdev_add_child(parent, vd); | |
591 | ||
592 | *vdp = vd; | |
593 | ||
594 | return (0); | |
595 | } | |
596 | ||
597 | void | |
598 | vdev_free(vdev_t *vd) | |
599 | { | |
600 | int c, t; | |
601 | spa_t *spa = vd->vdev_spa; | |
602 | ||
603 | /* | |
604 | * vdev_free() implies closing the vdev first. This is simpler than | |
605 | * trying to ensure complicated semantics for all callers. | |
606 | */ | |
607 | vdev_close(vd); | |
608 | ||
609 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); | |
610 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); | |
611 | ||
612 | /* | |
613 | * Free all children. | |
614 | */ | |
615 | for (c = 0; c < vd->vdev_children; c++) | |
616 | vdev_free(vd->vdev_child[c]); | |
617 | ||
618 | ASSERT(vd->vdev_child == NULL); | |
619 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
620 | ||
621 | /* | |
622 | * Discard allocation state. | |
623 | */ | |
624 | if (vd->vdev_mg != NULL) { | |
625 | vdev_metaslab_fini(vd); | |
626 | metaslab_group_destroy(vd->vdev_mg); | |
627 | } | |
628 | ||
629 | ASSERT0(vd->vdev_stat.vs_space); | |
630 | ASSERT0(vd->vdev_stat.vs_dspace); | |
631 | ASSERT0(vd->vdev_stat.vs_alloc); | |
632 | ||
633 | /* | |
634 | * Remove this vdev from its parent's child list. | |
635 | */ | |
636 | vdev_remove_child(vd->vdev_parent, vd); | |
637 | ||
638 | ASSERT(vd->vdev_parent == NULL); | |
639 | ||
640 | /* | |
641 | * Clean up vdev structure. | |
642 | */ | |
643 | vdev_queue_fini(vd); | |
644 | vdev_cache_fini(vd); | |
645 | ||
646 | if (vd->vdev_path) | |
647 | spa_strfree(vd->vdev_path); | |
648 | if (vd->vdev_devid) | |
649 | spa_strfree(vd->vdev_devid); | |
650 | if (vd->vdev_physpath) | |
651 | spa_strfree(vd->vdev_physpath); | |
652 | if (vd->vdev_fru) | |
653 | spa_strfree(vd->vdev_fru); | |
654 | ||
655 | if (vd->vdev_isspare) | |
656 | spa_spare_remove(vd); | |
657 | if (vd->vdev_isl2cache) | |
658 | spa_l2cache_remove(vd); | |
659 | ||
660 | txg_list_destroy(&vd->vdev_ms_list); | |
661 | txg_list_destroy(&vd->vdev_dtl_list); | |
662 | ||
663 | mutex_enter(&vd->vdev_dtl_lock); | |
664 | space_map_close(vd->vdev_dtl_sm); | |
665 | for (t = 0; t < DTL_TYPES; t++) { | |
666 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); | |
667 | range_tree_destroy(vd->vdev_dtl[t]); | |
668 | } | |
669 | mutex_exit(&vd->vdev_dtl_lock); | |
670 | ||
671 | mutex_destroy(&vd->vdev_dtl_lock); | |
672 | mutex_destroy(&vd->vdev_stat_lock); | |
673 | mutex_destroy(&vd->vdev_probe_lock); | |
674 | ||
675 | if (vd == spa->spa_root_vdev) | |
676 | spa->spa_root_vdev = NULL; | |
677 | ||
678 | kmem_free(vd, sizeof (vdev_t)); | |
679 | } | |
680 | ||
681 | /* | |
682 | * Transfer top-level vdev state from svd to tvd. | |
683 | */ | |
684 | static void | |
685 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
686 | { | |
687 | spa_t *spa = svd->vdev_spa; | |
688 | metaslab_t *msp; | |
689 | vdev_t *vd; | |
690 | int t; | |
691 | ||
692 | ASSERT(tvd == tvd->vdev_top); | |
693 | ||
694 | tvd->vdev_ms_array = svd->vdev_ms_array; | |
695 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
696 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
697 | ||
698 | svd->vdev_ms_array = 0; | |
699 | svd->vdev_ms_shift = 0; | |
700 | svd->vdev_ms_count = 0; | |
701 | ||
702 | if (tvd->vdev_mg) | |
703 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
704 | tvd->vdev_mg = svd->vdev_mg; | |
705 | tvd->vdev_ms = svd->vdev_ms; | |
706 | ||
707 | svd->vdev_mg = NULL; | |
708 | svd->vdev_ms = NULL; | |
709 | ||
710 | if (tvd->vdev_mg != NULL) | |
711 | tvd->vdev_mg->mg_vd = tvd; | |
712 | ||
713 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
714 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
715 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
716 | ||
717 | svd->vdev_stat.vs_alloc = 0; | |
718 | svd->vdev_stat.vs_space = 0; | |
719 | svd->vdev_stat.vs_dspace = 0; | |
720 | ||
721 | for (t = 0; t < TXG_SIZE; t++) { | |
722 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
723 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
724 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
725 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
726 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
727 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
728 | } | |
729 | ||
730 | if (list_link_active(&svd->vdev_config_dirty_node)) { | |
731 | vdev_config_clean(svd); | |
732 | vdev_config_dirty(tvd); | |
733 | } | |
734 | ||
735 | if (list_link_active(&svd->vdev_state_dirty_node)) { | |
736 | vdev_state_clean(svd); | |
737 | vdev_state_dirty(tvd); | |
738 | } | |
739 | ||
740 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; | |
741 | svd->vdev_deflate_ratio = 0; | |
742 | ||
743 | tvd->vdev_islog = svd->vdev_islog; | |
744 | svd->vdev_islog = 0; | |
745 | } | |
746 | ||
747 | static void | |
748 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
749 | { | |
750 | int c; | |
751 | ||
752 | if (vd == NULL) | |
753 | return; | |
754 | ||
755 | vd->vdev_top = tvd; | |
756 | ||
757 | for (c = 0; c < vd->vdev_children; c++) | |
758 | vdev_top_update(tvd, vd->vdev_child[c]); | |
759 | } | |
760 | ||
761 | /* | |
762 | * Add a mirror/replacing vdev above an existing vdev. | |
763 | */ | |
764 | vdev_t * | |
765 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
766 | { | |
767 | spa_t *spa = cvd->vdev_spa; | |
768 | vdev_t *pvd = cvd->vdev_parent; | |
769 | vdev_t *mvd; | |
770 | ||
771 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
772 | ||
773 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
774 | ||
775 | mvd->vdev_asize = cvd->vdev_asize; | |
776 | mvd->vdev_min_asize = cvd->vdev_min_asize; | |
777 | mvd->vdev_max_asize = cvd->vdev_max_asize; | |
778 | mvd->vdev_ashift = cvd->vdev_ashift; | |
779 | mvd->vdev_state = cvd->vdev_state; | |
780 | mvd->vdev_crtxg = cvd->vdev_crtxg; | |
781 | ||
782 | vdev_remove_child(pvd, cvd); | |
783 | vdev_add_child(pvd, mvd); | |
784 | cvd->vdev_id = mvd->vdev_children; | |
785 | vdev_add_child(mvd, cvd); | |
786 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
787 | ||
788 | if (mvd == mvd->vdev_top) | |
789 | vdev_top_transfer(cvd, mvd); | |
790 | ||
791 | return (mvd); | |
792 | } | |
793 | ||
794 | /* | |
795 | * Remove a 1-way mirror/replacing vdev from the tree. | |
796 | */ | |
797 | void | |
798 | vdev_remove_parent(vdev_t *cvd) | |
799 | { | |
800 | vdev_t *mvd = cvd->vdev_parent; | |
801 | vdev_t *pvd = mvd->vdev_parent; | |
802 | ||
803 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
804 | ||
805 | ASSERT(mvd->vdev_children == 1); | |
806 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
807 | mvd->vdev_ops == &vdev_replacing_ops || | |
808 | mvd->vdev_ops == &vdev_spare_ops); | |
809 | cvd->vdev_ashift = mvd->vdev_ashift; | |
810 | ||
811 | vdev_remove_child(mvd, cvd); | |
812 | vdev_remove_child(pvd, mvd); | |
813 | ||
814 | /* | |
815 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. | |
816 | * Otherwise, we could have detached an offline device, and when we | |
817 | * go to import the pool we'll think we have two top-level vdevs, | |
818 | * instead of a different version of the same top-level vdev. | |
819 | */ | |
820 | if (mvd->vdev_top == mvd) { | |
821 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
822 | cvd->vdev_orig_guid = cvd->vdev_guid; | |
823 | cvd->vdev_guid += guid_delta; | |
824 | cvd->vdev_guid_sum += guid_delta; | |
825 | ||
826 | /* | |
827 | * If pool not set for autoexpand, we need to also preserve | |
828 | * mvd's asize to prevent automatic expansion of cvd. | |
829 | * Otherwise if we are adjusting the mirror by attaching and | |
830 | * detaching children of non-uniform sizes, the mirror could | |
831 | * autoexpand, unexpectedly requiring larger devices to | |
832 | * re-establish the mirror. | |
833 | */ | |
834 | if (!cvd->vdev_spa->spa_autoexpand) | |
835 | cvd->vdev_asize = mvd->vdev_asize; | |
836 | } | |
837 | cvd->vdev_id = mvd->vdev_id; | |
838 | vdev_add_child(pvd, cvd); | |
839 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
840 | ||
841 | if (cvd == cvd->vdev_top) | |
842 | vdev_top_transfer(mvd, cvd); | |
843 | ||
844 | ASSERT(mvd->vdev_children == 0); | |
845 | vdev_free(mvd); | |
846 | } | |
847 | ||
848 | int | |
849 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
850 | { | |
851 | spa_t *spa = vd->vdev_spa; | |
852 | objset_t *mos = spa->spa_meta_objset; | |
853 | uint64_t m; | |
854 | uint64_t oldc = vd->vdev_ms_count; | |
855 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
856 | metaslab_t **mspp; | |
857 | int error; | |
858 | ||
859 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
860 | ||
861 | /* | |
862 | * This vdev is not being allocated from yet or is a hole. | |
863 | */ | |
864 | if (vd->vdev_ms_shift == 0) | |
865 | return (0); | |
866 | ||
867 | ASSERT(!vd->vdev_ishole); | |
868 | ||
869 | /* | |
870 | * Compute the raidz-deflation ratio. Note, we hard-code | |
871 | * in 128k (1 << 17) because it is the "typical" blocksize. | |
872 | * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, | |
873 | * otherwise it would inconsistently account for existing bp's. | |
874 | */ | |
875 | vd->vdev_deflate_ratio = (1 << 17) / | |
876 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
877 | ||
878 | ASSERT(oldc <= newc); | |
879 | ||
880 | mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); | |
881 | ||
882 | if (oldc != 0) { | |
883 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
884 | vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); | |
885 | } | |
886 | ||
887 | vd->vdev_ms = mspp; | |
888 | vd->vdev_ms_count = newc; | |
889 | ||
890 | for (m = oldc; m < newc; m++) { | |
891 | uint64_t object = 0; | |
892 | ||
893 | if (txg == 0) { | |
894 | error = dmu_read(mos, vd->vdev_ms_array, | |
895 | m * sizeof (uint64_t), sizeof (uint64_t), &object, | |
896 | DMU_READ_PREFETCH); | |
897 | if (error) | |
898 | return (error); | |
899 | } | |
900 | ||
901 | error = metaslab_init(vd->vdev_mg, m, object, txg, | |
902 | &(vd->vdev_ms[m])); | |
903 | if (error) | |
904 | return (error); | |
905 | } | |
906 | ||
907 | if (txg == 0) | |
908 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
909 | ||
910 | /* | |
911 | * If the vdev is being removed we don't activate | |
912 | * the metaslabs since we want to ensure that no new | |
913 | * allocations are performed on this device. | |
914 | */ | |
915 | if (oldc == 0 && !vd->vdev_removing) | |
916 | metaslab_group_activate(vd->vdev_mg); | |
917 | ||
918 | if (txg == 0) | |
919 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
920 | ||
921 | return (0); | |
922 | } | |
923 | ||
924 | void | |
925 | vdev_metaslab_fini(vdev_t *vd) | |
926 | { | |
927 | uint64_t m; | |
928 | uint64_t count = vd->vdev_ms_count; | |
929 | ||
930 | if (vd->vdev_ms != NULL) { | |
931 | metaslab_group_passivate(vd->vdev_mg); | |
932 | for (m = 0; m < count; m++) { | |
933 | metaslab_t *msp = vd->vdev_ms[m]; | |
934 | ||
935 | if (msp != NULL) | |
936 | metaslab_fini(msp); | |
937 | } | |
938 | vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); | |
939 | vd->vdev_ms = NULL; | |
940 | } | |
941 | ||
942 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
943 | } | |
944 | ||
945 | typedef struct vdev_probe_stats { | |
946 | boolean_t vps_readable; | |
947 | boolean_t vps_writeable; | |
948 | int vps_flags; | |
949 | } vdev_probe_stats_t; | |
950 | ||
951 | static void | |
952 | vdev_probe_done(zio_t *zio) | |
953 | { | |
954 | spa_t *spa = zio->io_spa; | |
955 | vdev_t *vd = zio->io_vd; | |
956 | vdev_probe_stats_t *vps = zio->io_private; | |
957 | ||
958 | ASSERT(vd->vdev_probe_zio != NULL); | |
959 | ||
960 | if (zio->io_type == ZIO_TYPE_READ) { | |
961 | if (zio->io_error == 0) | |
962 | vps->vps_readable = 1; | |
963 | if (zio->io_error == 0 && spa_writeable(spa)) { | |
964 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, | |
965 | zio->io_offset, zio->io_size, zio->io_data, | |
966 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
967 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
968 | } else { | |
969 | zio_buf_free(zio->io_data, zio->io_size); | |
970 | } | |
971 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
972 | if (zio->io_error == 0) | |
973 | vps->vps_writeable = 1; | |
974 | zio_buf_free(zio->io_data, zio->io_size); | |
975 | } else if (zio->io_type == ZIO_TYPE_NULL) { | |
976 | zio_t *pio; | |
977 | ||
978 | vd->vdev_cant_read |= !vps->vps_readable; | |
979 | vd->vdev_cant_write |= !vps->vps_writeable; | |
980 | ||
981 | if (vdev_readable(vd) && | |
982 | (vdev_writeable(vd) || !spa_writeable(spa))) { | |
983 | zio->io_error = 0; | |
984 | } else { | |
985 | ASSERT(zio->io_error != 0); | |
986 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
987 | spa, vd, NULL, 0, 0); | |
988 | zio->io_error = SET_ERROR(ENXIO); | |
989 | } | |
990 | ||
991 | mutex_enter(&vd->vdev_probe_lock); | |
992 | ASSERT(vd->vdev_probe_zio == zio); | |
993 | vd->vdev_probe_zio = NULL; | |
994 | mutex_exit(&vd->vdev_probe_lock); | |
995 | ||
996 | while ((pio = zio_walk_parents(zio)) != NULL) | |
997 | if (!vdev_accessible(vd, pio)) | |
998 | pio->io_error = SET_ERROR(ENXIO); | |
999 | ||
1000 | kmem_free(vps, sizeof (*vps)); | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * Determine whether this device is accessible. | |
1006 | * | |
1007 | * Read and write to several known locations: the pad regions of each | |
1008 | * vdev label but the first, which we leave alone in case it contains | |
1009 | * a VTOC. | |
1010 | */ | |
1011 | zio_t * | |
1012 | vdev_probe(vdev_t *vd, zio_t *zio) | |
1013 | { | |
1014 | spa_t *spa = vd->vdev_spa; | |
1015 | vdev_probe_stats_t *vps = NULL; | |
1016 | zio_t *pio; | |
1017 | int l; | |
1018 | ||
1019 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
1020 | ||
1021 | /* | |
1022 | * Don't probe the probe. | |
1023 | */ | |
1024 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
1025 | return (NULL); | |
1026 | ||
1027 | /* | |
1028 | * To prevent 'probe storms' when a device fails, we create | |
1029 | * just one probe i/o at a time. All zios that want to probe | |
1030 | * this vdev will become parents of the probe io. | |
1031 | */ | |
1032 | mutex_enter(&vd->vdev_probe_lock); | |
1033 | ||
1034 | if ((pio = vd->vdev_probe_zio) == NULL) { | |
1035 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); | |
1036 | ||
1037 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1038 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
1039 | ZIO_FLAG_TRYHARD; | |
1040 | ||
1041 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1042 | /* | |
1043 | * vdev_cant_read and vdev_cant_write can only | |
1044 | * transition from TRUE to FALSE when we have the | |
1045 | * SCL_ZIO lock as writer; otherwise they can only | |
1046 | * transition from FALSE to TRUE. This ensures that | |
1047 | * any zio looking at these values can assume that | |
1048 | * failures persist for the life of the I/O. That's | |
1049 | * important because when a device has intermittent | |
1050 | * connectivity problems, we want to ensure that | |
1051 | * they're ascribed to the device (ENXIO) and not | |
1052 | * the zio (EIO). | |
1053 | * | |
1054 | * Since we hold SCL_ZIO as writer here, clear both | |
1055 | * values so the probe can reevaluate from first | |
1056 | * principles. | |
1057 | */ | |
1058 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1059 | vd->vdev_cant_read = B_FALSE; | |
1060 | vd->vdev_cant_write = B_FALSE; | |
1061 | } | |
1062 | ||
1063 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1064 | vdev_probe_done, vps, | |
1065 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1066 | ||
1067 | /* | |
1068 | * We can't change the vdev state in this context, so we | |
1069 | * kick off an async task to do it on our behalf. | |
1070 | */ | |
1071 | if (zio != NULL) { | |
1072 | vd->vdev_probe_wanted = B_TRUE; | |
1073 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1074 | } | |
1075 | } | |
1076 | ||
1077 | if (zio != NULL) | |
1078 | zio_add_child(zio, pio); | |
1079 | ||
1080 | mutex_exit(&vd->vdev_probe_lock); | |
1081 | ||
1082 | if (vps == NULL) { | |
1083 | ASSERT(zio != NULL); | |
1084 | return (NULL); | |
1085 | } | |
1086 | ||
1087 | for (l = 1; l < VDEV_LABELS; l++) { | |
1088 | zio_nowait(zio_read_phys(pio, vd, | |
1089 | vdev_label_offset(vd->vdev_psize, l, | |
1090 | offsetof(vdev_label_t, vl_pad2)), | |
1091 | VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), | |
1092 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, | |
1093 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1094 | } | |
1095 | ||
1096 | if (zio == NULL) | |
1097 | return (pio); | |
1098 | ||
1099 | zio_nowait(pio); | |
1100 | return (NULL); | |
1101 | } | |
1102 | ||
1103 | static void | |
1104 | vdev_open_child(void *arg) | |
1105 | { | |
1106 | vdev_t *vd = arg; | |
1107 | ||
1108 | vd->vdev_open_thread = curthread; | |
1109 | vd->vdev_open_error = vdev_open(vd); | |
1110 | vd->vdev_open_thread = NULL; | |
1111 | vd->vdev_parent->vdev_nonrot &= vd->vdev_nonrot; | |
1112 | } | |
1113 | ||
1114 | static boolean_t | |
1115 | vdev_uses_zvols(vdev_t *vd) | |
1116 | { | |
1117 | int c; | |
1118 | ||
1119 | #ifdef _KERNEL | |
1120 | if (zvol_is_zvol(vd->vdev_path)) | |
1121 | return (B_TRUE); | |
1122 | #endif | |
1123 | ||
1124 | for (c = 0; c < vd->vdev_children; c++) | |
1125 | if (vdev_uses_zvols(vd->vdev_child[c])) | |
1126 | return (B_TRUE); | |
1127 | ||
1128 | return (B_FALSE); | |
1129 | } | |
1130 | ||
1131 | void | |
1132 | vdev_open_children(vdev_t *vd) | |
1133 | { | |
1134 | taskq_t *tq; | |
1135 | int children = vd->vdev_children; | |
1136 | int c; | |
1137 | ||
1138 | vd->vdev_nonrot = B_TRUE; | |
1139 | ||
1140 | /* | |
1141 | * in order to handle pools on top of zvols, do the opens | |
1142 | * in a single thread so that the same thread holds the | |
1143 | * spa_namespace_lock | |
1144 | */ | |
1145 | if (vdev_uses_zvols(vd)) { | |
1146 | for (c = 0; c < children; c++) { | |
1147 | vd->vdev_child[c]->vdev_open_error = | |
1148 | vdev_open(vd->vdev_child[c]); | |
1149 | vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; | |
1150 | } | |
1151 | return; | |
1152 | } | |
1153 | tq = taskq_create("vdev_open", children, minclsyspri, | |
1154 | children, children, TASKQ_PREPOPULATE); | |
1155 | ||
1156 | for (c = 0; c < children; c++) | |
1157 | VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], | |
1158 | TQ_SLEEP) != 0); | |
1159 | ||
1160 | taskq_destroy(tq); | |
1161 | ||
1162 | for (c = 0; c < children; c++) | |
1163 | vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * Prepare a virtual device for access. | |
1168 | */ | |
1169 | int | |
1170 | vdev_open(vdev_t *vd) | |
1171 | { | |
1172 | spa_t *spa = vd->vdev_spa; | |
1173 | int error; | |
1174 | uint64_t osize = 0; | |
1175 | uint64_t max_osize = 0; | |
1176 | uint64_t asize, max_asize, psize; | |
1177 | uint64_t ashift = 0; | |
1178 | int c; | |
1179 | ||
1180 | ASSERT(vd->vdev_open_thread == curthread || | |
1181 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1182 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || | |
1183 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1184 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1185 | ||
1186 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1187 | vd->vdev_cant_read = B_FALSE; | |
1188 | vd->vdev_cant_write = B_FALSE; | |
1189 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
1190 | ||
1191 | /* | |
1192 | * If this vdev is not removed, check its fault status. If it's | |
1193 | * faulted, bail out of the open. | |
1194 | */ | |
1195 | if (!vd->vdev_removed && vd->vdev_faulted) { | |
1196 | ASSERT(vd->vdev_children == 0); | |
1197 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1198 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1199 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1200 | vd->vdev_label_aux); | |
1201 | return (SET_ERROR(ENXIO)); | |
1202 | } else if (vd->vdev_offline) { | |
1203 | ASSERT(vd->vdev_children == 0); | |
1204 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
1205 | return (SET_ERROR(ENXIO)); | |
1206 | } | |
1207 | ||
1208 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); | |
1209 | ||
1210 | /* | |
1211 | * Reset the vdev_reopening flag so that we actually close | |
1212 | * the vdev on error. | |
1213 | */ | |
1214 | vd->vdev_reopening = B_FALSE; | |
1215 | if (zio_injection_enabled && error == 0) | |
1216 | error = zio_handle_device_injection(vd, NULL, ENXIO); | |
1217 | ||
1218 | if (error) { | |
1219 | if (vd->vdev_removed && | |
1220 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1221 | vd->vdev_removed = B_FALSE; | |
1222 | ||
1223 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1224 | vd->vdev_stat.vs_aux); | |
1225 | return (error); | |
1226 | } | |
1227 | ||
1228 | vd->vdev_removed = B_FALSE; | |
1229 | ||
1230 | /* | |
1231 | * Recheck the faulted flag now that we have confirmed that | |
1232 | * the vdev is accessible. If we're faulted, bail. | |
1233 | */ | |
1234 | if (vd->vdev_faulted) { | |
1235 | ASSERT(vd->vdev_children == 0); | |
1236 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1237 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1238 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1239 | vd->vdev_label_aux); | |
1240 | return (SET_ERROR(ENXIO)); | |
1241 | } | |
1242 | ||
1243 | if (vd->vdev_degraded) { | |
1244 | ASSERT(vd->vdev_children == 0); | |
1245 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1246 | VDEV_AUX_ERR_EXCEEDED); | |
1247 | } else { | |
1248 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * For hole or missing vdevs we just return success. | |
1253 | */ | |
1254 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1255 | return (0); | |
1256 | ||
1257 | for (c = 0; c < vd->vdev_children; c++) { | |
1258 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { | |
1259 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1260 | VDEV_AUX_NONE); | |
1261 | break; | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1266 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); | |
1267 | ||
1268 | if (vd->vdev_children == 0) { | |
1269 | if (osize < SPA_MINDEVSIZE) { | |
1270 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1271 | VDEV_AUX_TOO_SMALL); | |
1272 | return (SET_ERROR(EOVERFLOW)); | |
1273 | } | |
1274 | psize = osize; | |
1275 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1276 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + | |
1277 | VDEV_LABEL_END_SIZE); | |
1278 | } else { | |
1279 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1280 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1281 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1282 | VDEV_AUX_TOO_SMALL); | |
1283 | return (SET_ERROR(EOVERFLOW)); | |
1284 | } | |
1285 | psize = 0; | |
1286 | asize = osize; | |
1287 | max_asize = max_osize; | |
1288 | } | |
1289 | ||
1290 | vd->vdev_psize = psize; | |
1291 | ||
1292 | /* | |
1293 | * Make sure the allocatable size hasn't shrunk. | |
1294 | */ | |
1295 | if (asize < vd->vdev_min_asize) { | |
1296 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1297 | VDEV_AUX_BAD_LABEL); | |
1298 | return (SET_ERROR(EINVAL)); | |
1299 | } | |
1300 | ||
1301 | if (vd->vdev_asize == 0) { | |
1302 | /* | |
1303 | * This is the first-ever open, so use the computed values. | |
1304 | * For compatibility, a different ashift can be requested. | |
1305 | */ | |
1306 | vd->vdev_asize = asize; | |
1307 | vd->vdev_max_asize = max_asize; | |
1308 | if (vd->vdev_ashift == 0) | |
1309 | vd->vdev_ashift = ashift; | |
1310 | } else { | |
1311 | /* | |
1312 | * Detect if the alignment requirement has increased. | |
1313 | * We don't want to make the pool unavailable, just | |
1314 | * post an event instead. | |
1315 | */ | |
1316 | if (ashift > vd->vdev_top->vdev_ashift && | |
1317 | vd->vdev_ops->vdev_op_leaf) { | |
1318 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
1319 | spa, vd, NULL, 0, 0); | |
1320 | } | |
1321 | ||
1322 | vd->vdev_max_asize = max_asize; | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * If all children are healthy and the asize has increased, | |
1327 | * then we've experienced dynamic LUN growth. If automatic | |
1328 | * expansion is enabled then use the additional space. | |
1329 | */ | |
1330 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1331 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1332 | vd->vdev_asize = asize; | |
1333 | ||
1334 | vdev_set_min_asize(vd); | |
1335 | ||
1336 | /* | |
1337 | * Ensure we can issue some IO before declaring the | |
1338 | * vdev open for business. | |
1339 | */ | |
1340 | if (vd->vdev_ops->vdev_op_leaf && | |
1341 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
1342 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1343 | VDEV_AUX_ERR_EXCEEDED); | |
1344 | return (error); | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * Track the min and max ashift values for normal data devices. | |
1349 | */ | |
1350 | if (vd->vdev_top == vd && vd->vdev_ashift != 0 && | |
1351 | !vd->vdev_islog && vd->vdev_aux == NULL) { | |
1352 | if (vd->vdev_ashift > spa->spa_max_ashift) | |
1353 | spa->spa_max_ashift = vd->vdev_ashift; | |
1354 | if (vd->vdev_ashift < spa->spa_min_ashift) | |
1355 | spa->spa_min_ashift = vd->vdev_ashift; | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * If a leaf vdev has a DTL, and seems healthy, then kick off a | |
1360 | * resilver. But don't do this if we are doing a reopen for a scrub, | |
1361 | * since this would just restart the scrub we are already doing. | |
1362 | */ | |
1363 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && | |
1364 | vdev_resilver_needed(vd, NULL, NULL)) | |
1365 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
1366 | ||
1367 | return (0); | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Called once the vdevs are all opened, this routine validates the label | |
1372 | * contents. This needs to be done before vdev_load() so that we don't | |
1373 | * inadvertently do repair I/Os to the wrong device. | |
1374 | * | |
1375 | * If 'strict' is false ignore the spa guid check. This is necessary because | |
1376 | * if the machine crashed during a re-guid the new guid might have been written | |
1377 | * to all of the vdev labels, but not the cached config. The strict check | |
1378 | * will be performed when the pool is opened again using the mos config. | |
1379 | * | |
1380 | * This function will only return failure if one of the vdevs indicates that it | |
1381 | * has since been destroyed or exported. This is only possible if | |
1382 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1383 | * will be updated but the function will return 0. | |
1384 | */ | |
1385 | int | |
1386 | vdev_validate(vdev_t *vd, boolean_t strict) | |
1387 | { | |
1388 | spa_t *spa = vd->vdev_spa; | |
1389 | nvlist_t *label; | |
1390 | uint64_t guid = 0, top_guid; | |
1391 | uint64_t state; | |
1392 | int c; | |
1393 | ||
1394 | for (c = 0; c < vd->vdev_children; c++) | |
1395 | if (vdev_validate(vd->vdev_child[c], strict) != 0) | |
1396 | return (SET_ERROR(EBADF)); | |
1397 | ||
1398 | /* | |
1399 | * If the device has already failed, or was marked offline, don't do | |
1400 | * any further validation. Otherwise, label I/O will fail and we will | |
1401 | * overwrite the previous state. | |
1402 | */ | |
1403 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { | |
1404 | uint64_t aux_guid = 0; | |
1405 | nvlist_t *nvl; | |
1406 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? | |
1407 | spa_last_synced_txg(spa) : -1ULL; | |
1408 | ||
1409 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { | |
1410 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1411 | VDEV_AUX_BAD_LABEL); | |
1412 | return (0); | |
1413 | } | |
1414 | ||
1415 | /* | |
1416 | * Determine if this vdev has been split off into another | |
1417 | * pool. If so, then refuse to open it. | |
1418 | */ | |
1419 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1420 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1421 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1422 | VDEV_AUX_SPLIT_POOL); | |
1423 | nvlist_free(label); | |
1424 | return (0); | |
1425 | } | |
1426 | ||
1427 | if (strict && (nvlist_lookup_uint64(label, | |
1428 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1429 | guid != spa_guid(spa))) { | |
1430 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1431 | VDEV_AUX_CORRUPT_DATA); | |
1432 | nvlist_free(label); | |
1433 | return (0); | |
1434 | } | |
1435 | ||
1436 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) | |
1437 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1438 | &aux_guid) != 0) | |
1439 | aux_guid = 0; | |
1440 | ||
1441 | /* | |
1442 | * If this vdev just became a top-level vdev because its | |
1443 | * sibling was detached, it will have adopted the parent's | |
1444 | * vdev guid -- but the label may or may not be on disk yet. | |
1445 | * Fortunately, either version of the label will have the | |
1446 | * same top guid, so if we're a top-level vdev, we can | |
1447 | * safely compare to that instead. | |
1448 | * | |
1449 | * If we split this vdev off instead, then we also check the | |
1450 | * original pool's guid. We don't want to consider the vdev | |
1451 | * corrupt if it is partway through a split operation. | |
1452 | */ | |
1453 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, | |
1454 | &guid) != 0 || | |
1455 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1456 | &top_guid) != 0 || | |
1457 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && | |
1458 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { | |
1459 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1460 | VDEV_AUX_CORRUPT_DATA); | |
1461 | nvlist_free(label); | |
1462 | return (0); | |
1463 | } | |
1464 | ||
1465 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1466 | &state) != 0) { | |
1467 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1468 | VDEV_AUX_CORRUPT_DATA); | |
1469 | nvlist_free(label); | |
1470 | return (0); | |
1471 | } | |
1472 | ||
1473 | nvlist_free(label); | |
1474 | ||
1475 | /* | |
1476 | * If this is a verbatim import, no need to check the | |
1477 | * state of the pool. | |
1478 | */ | |
1479 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && | |
1480 | spa_load_state(spa) == SPA_LOAD_OPEN && | |
1481 | state != POOL_STATE_ACTIVE) | |
1482 | return (SET_ERROR(EBADF)); | |
1483 | ||
1484 | /* | |
1485 | * If we were able to open and validate a vdev that was | |
1486 | * previously marked permanently unavailable, clear that state | |
1487 | * now. | |
1488 | */ | |
1489 | if (vd->vdev_not_present) | |
1490 | vd->vdev_not_present = 0; | |
1491 | } | |
1492 | ||
1493 | return (0); | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * Close a virtual device. | |
1498 | */ | |
1499 | void | |
1500 | vdev_close(vdev_t *vd) | |
1501 | { | |
1502 | vdev_t *pvd = vd->vdev_parent; | |
1503 | ASSERTV(spa_t *spa = vd->vdev_spa); | |
1504 | ||
1505 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1506 | ||
1507 | /* | |
1508 | * If our parent is reopening, then we are as well, unless we are | |
1509 | * going offline. | |
1510 | */ | |
1511 | if (pvd != NULL && pvd->vdev_reopening) | |
1512 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1513 | ||
1514 | vd->vdev_ops->vdev_op_close(vd); | |
1515 | ||
1516 | vdev_cache_purge(vd); | |
1517 | ||
1518 | /* | |
1519 | * We record the previous state before we close it, so that if we are | |
1520 | * doing a reopen(), we don't generate FMA ereports if we notice that | |
1521 | * it's still faulted. | |
1522 | */ | |
1523 | vd->vdev_prevstate = vd->vdev_state; | |
1524 | ||
1525 | if (vd->vdev_offline) | |
1526 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1527 | else | |
1528 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1529 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1530 | } | |
1531 | ||
1532 | void | |
1533 | vdev_hold(vdev_t *vd) | |
1534 | { | |
1535 | spa_t *spa = vd->vdev_spa; | |
1536 | int c; | |
1537 | ||
1538 | ASSERT(spa_is_root(spa)); | |
1539 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1540 | return; | |
1541 | ||
1542 | for (c = 0; c < vd->vdev_children; c++) | |
1543 | vdev_hold(vd->vdev_child[c]); | |
1544 | ||
1545 | if (vd->vdev_ops->vdev_op_leaf) | |
1546 | vd->vdev_ops->vdev_op_hold(vd); | |
1547 | } | |
1548 | ||
1549 | void | |
1550 | vdev_rele(vdev_t *vd) | |
1551 | { | |
1552 | int c; | |
1553 | ||
1554 | ASSERT(spa_is_root(vd->vdev_spa)); | |
1555 | for (c = 0; c < vd->vdev_children; c++) | |
1556 | vdev_rele(vd->vdev_child[c]); | |
1557 | ||
1558 | if (vd->vdev_ops->vdev_op_leaf) | |
1559 | vd->vdev_ops->vdev_op_rele(vd); | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1564 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1565 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1566 | * If the leaf has never been opened then open it, as usual. | |
1567 | */ | |
1568 | void | |
1569 | vdev_reopen(vdev_t *vd) | |
1570 | { | |
1571 | spa_t *spa = vd->vdev_spa; | |
1572 | ||
1573 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1574 | ||
1575 | /* set the reopening flag unless we're taking the vdev offline */ | |
1576 | vd->vdev_reopening = !vd->vdev_offline; | |
1577 | vdev_close(vd); | |
1578 | (void) vdev_open(vd); | |
1579 | ||
1580 | /* | |
1581 | * Call vdev_validate() here to make sure we have the same device. | |
1582 | * Otherwise, a device with an invalid label could be successfully | |
1583 | * opened in response to vdev_reopen(). | |
1584 | */ | |
1585 | if (vd->vdev_aux) { | |
1586 | (void) vdev_validate_aux(vd); | |
1587 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
1588 | vd->vdev_aux == &spa->spa_l2cache && | |
1589 | !l2arc_vdev_present(vd)) | |
1590 | l2arc_add_vdev(spa, vd); | |
1591 | } else { | |
1592 | (void) vdev_validate(vd, B_TRUE); | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Reassess parent vdev's health. | |
1597 | */ | |
1598 | vdev_propagate_state(vd); | |
1599 | } | |
1600 | ||
1601 | int | |
1602 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1603 | { | |
1604 | int error; | |
1605 | ||
1606 | /* | |
1607 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1608 | * For a create, however, we want to fail the request if | |
1609 | * there are any components we can't open. | |
1610 | */ | |
1611 | error = vdev_open(vd); | |
1612 | ||
1613 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1614 | vdev_close(vd); | |
1615 | return (error ? error : ENXIO); | |
1616 | } | |
1617 | ||
1618 | /* | |
1619 | * Recursively load DTLs and initialize all labels. | |
1620 | */ | |
1621 | if ((error = vdev_dtl_load(vd)) != 0 || | |
1622 | (error = vdev_label_init(vd, txg, isreplacing ? | |
1623 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { | |
1624 | vdev_close(vd); | |
1625 | return (error); | |
1626 | } | |
1627 | ||
1628 | return (0); | |
1629 | } | |
1630 | ||
1631 | void | |
1632 | vdev_metaslab_set_size(vdev_t *vd) | |
1633 | { | |
1634 | /* | |
1635 | * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. | |
1636 | */ | |
1637 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); | |
1638 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); | |
1639 | } | |
1640 | ||
1641 | void | |
1642 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1643 | { | |
1644 | ASSERT(vd == vd->vdev_top); | |
1645 | ASSERT(!vd->vdev_ishole); | |
1646 | ASSERT(ISP2(flags)); | |
1647 | ASSERT(spa_writeable(vd->vdev_spa)); | |
1648 | ||
1649 | if (flags & VDD_METASLAB) | |
1650 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1651 | ||
1652 | if (flags & VDD_DTL) | |
1653 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1654 | ||
1655 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1656 | } | |
1657 | ||
1658 | void | |
1659 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1660 | { | |
1661 | int c; | |
1662 | ||
1663 | for (c = 0; c < vd->vdev_children; c++) | |
1664 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); | |
1665 | ||
1666 | if (vd->vdev_ops->vdev_op_leaf) | |
1667 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * DTLs. | |
1672 | * | |
1673 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
1674 | * the vdev has less than perfect replication. There are four kinds of DTL: | |
1675 | * | |
1676 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1677 | * | |
1678 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1679 | * | |
1680 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1681 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1682 | * txgs that was scrubbed. | |
1683 | * | |
1684 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1685 | * persistent errors or just some device being offline. | |
1686 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1687 | * maintained; it's only computed when needed, typically to | |
1688 | * determine whether a device can be detached. | |
1689 | * | |
1690 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1691 | * either has the data or it doesn't. | |
1692 | * | |
1693 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1694 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1695 | * if any child is less than fully replicated, then so is its parent. | |
1696 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1697 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1698 | * those are the txgs we don't have enough replication to read. For example, | |
1699 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1700 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1701 | * two child DTL_MISSING maps. | |
1702 | * | |
1703 | * It should be clear from the above that to compute the DTLs and outage maps | |
1704 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1705 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1706 | * a configuration change, we generate all other DTLs from first principles. | |
1707 | */ | |
1708 | void | |
1709 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
1710 | { | |
1711 | range_tree_t *rt = vd->vdev_dtl[t]; | |
1712 | ||
1713 | ASSERT(t < DTL_TYPES); | |
1714 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
1715 | ASSERT(spa_writeable(vd->vdev_spa)); | |
1716 | ||
1717 | mutex_enter(rt->rt_lock); | |
1718 | if (!range_tree_contains(rt, txg, size)) | |
1719 | range_tree_add(rt, txg, size); | |
1720 | mutex_exit(rt->rt_lock); | |
1721 | } | |
1722 | ||
1723 | boolean_t | |
1724 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
1725 | { | |
1726 | range_tree_t *rt = vd->vdev_dtl[t]; | |
1727 | boolean_t dirty = B_FALSE; | |
1728 | ||
1729 | ASSERT(t < DTL_TYPES); | |
1730 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
1731 | ||
1732 | mutex_enter(rt->rt_lock); | |
1733 | if (range_tree_space(rt) != 0) | |
1734 | dirty = range_tree_contains(rt, txg, size); | |
1735 | mutex_exit(rt->rt_lock); | |
1736 | ||
1737 | return (dirty); | |
1738 | } | |
1739 | ||
1740 | boolean_t | |
1741 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1742 | { | |
1743 | range_tree_t *rt = vd->vdev_dtl[t]; | |
1744 | boolean_t empty; | |
1745 | ||
1746 | mutex_enter(rt->rt_lock); | |
1747 | empty = (range_tree_space(rt) == 0); | |
1748 | mutex_exit(rt->rt_lock); | |
1749 | ||
1750 | return (empty); | |
1751 | } | |
1752 | ||
1753 | /* | |
1754 | * Returns the lowest txg in the DTL range. | |
1755 | */ | |
1756 | static uint64_t | |
1757 | vdev_dtl_min(vdev_t *vd) | |
1758 | { | |
1759 | range_seg_t *rs; | |
1760 | ||
1761 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
1762 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); | |
1763 | ASSERT0(vd->vdev_children); | |
1764 | ||
1765 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); | |
1766 | return (rs->rs_start - 1); | |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * Returns the highest txg in the DTL. | |
1771 | */ | |
1772 | static uint64_t | |
1773 | vdev_dtl_max(vdev_t *vd) | |
1774 | { | |
1775 | range_seg_t *rs; | |
1776 | ||
1777 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
1778 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); | |
1779 | ASSERT0(vd->vdev_children); | |
1780 | ||
1781 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); | |
1782 | return (rs->rs_end); | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * Determine if a resilvering vdev should remove any DTL entries from | |
1787 | * its range. If the vdev was resilvering for the entire duration of the | |
1788 | * scan then it should excise that range from its DTLs. Otherwise, this | |
1789 | * vdev is considered partially resilvered and should leave its DTL | |
1790 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
1791 | * excise the DTLs. | |
1792 | */ | |
1793 | static boolean_t | |
1794 | vdev_dtl_should_excise(vdev_t *vd) | |
1795 | { | |
1796 | spa_t *spa = vd->vdev_spa; | |
1797 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1798 | ||
1799 | ASSERT0(scn->scn_phys.scn_errors); | |
1800 | ASSERT0(vd->vdev_children); | |
1801 | ||
9784fa9e CIK |
1802 | if (vd->vdev_state < VDEV_STATE_DEGRADED) |
1803 | return (B_FALSE); | |
1804 | ||
70e083d2 TG |
1805 | if (vd->vdev_resilver_txg == 0 || |
1806 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) | |
1807 | return (B_TRUE); | |
1808 | ||
1809 | /* | |
1810 | * When a resilver is initiated the scan will assign the scn_max_txg | |
1811 | * value to the highest txg value that exists in all DTLs. If this | |
1812 | * device's max DTL is not part of this scan (i.e. it is not in | |
1813 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
1814 | * for excision. | |
1815 | */ | |
1816 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
1817 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
1818 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
1819 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
1820 | return (B_TRUE); | |
1821 | } | |
1822 | return (B_FALSE); | |
1823 | } | |
1824 | ||
1825 | /* | |
1826 | * Reassess DTLs after a config change or scrub completion. | |
1827 | */ | |
1828 | void | |
1829 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1830 | { | |
1831 | spa_t *spa = vd->vdev_spa; | |
1832 | avl_tree_t reftree; | |
1833 | int c, t, minref; | |
1834 | ||
1835 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); | |
1836 | ||
1837 | for (c = 0; c < vd->vdev_children; c++) | |
1838 | vdev_dtl_reassess(vd->vdev_child[c], txg, | |
1839 | scrub_txg, scrub_done); | |
1840 | ||
1841 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) | |
1842 | return; | |
1843 | ||
1844 | if (vd->vdev_ops->vdev_op_leaf) { | |
1845 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1846 | ||
1847 | mutex_enter(&vd->vdev_dtl_lock); | |
1848 | ||
1849 | /* | |
1850 | * If we've completed a scan cleanly then determine | |
1851 | * if this vdev should remove any DTLs. We only want to | |
1852 | * excise regions on vdevs that were available during | |
1853 | * the entire duration of this scan. | |
1854 | */ | |
1855 | if (scrub_txg != 0 && | |
1856 | (spa->spa_scrub_started || | |
1857 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && | |
1858 | vdev_dtl_should_excise(vd)) { | |
1859 | /* | |
1860 | * We completed a scrub up to scrub_txg. If we | |
1861 | * did it without rebooting, then the scrub dtl | |
1862 | * will be valid, so excise the old region and | |
1863 | * fold in the scrub dtl. Otherwise, leave the | |
1864 | * dtl as-is if there was an error. | |
1865 | * | |
1866 | * There's little trick here: to excise the beginning | |
1867 | * of the DTL_MISSING map, we put it into a reference | |
1868 | * tree and then add a segment with refcnt -1 that | |
1869 | * covers the range [0, scrub_txg). This means | |
1870 | * that each txg in that range has refcnt -1 or 0. | |
1871 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1872 | * entries in the range [0, scrub_txg) will have a | |
1873 | * positive refcnt -- either 1 or 2. We then convert | |
1874 | * the reference tree into the new DTL_MISSING map. | |
1875 | */ | |
1876 | space_reftree_create(&reftree); | |
1877 | space_reftree_add_map(&reftree, | |
1878 | vd->vdev_dtl[DTL_MISSING], 1); | |
1879 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
1880 | space_reftree_add_map(&reftree, | |
1881 | vd->vdev_dtl[DTL_SCRUB], 2); | |
1882 | space_reftree_generate_map(&reftree, | |
1883 | vd->vdev_dtl[DTL_MISSING], 1); | |
1884 | space_reftree_destroy(&reftree); | |
1885 | } | |
1886 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); | |
1887 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1888 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
1889 | if (scrub_done) | |
1890 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); | |
1891 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
1892 | if (!vdev_readable(vd)) | |
1893 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); | |
1894 | else | |
1895 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1896 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
1897 | ||
1898 | /* | |
1899 | * If the vdev was resilvering and no longer has any | |
1900 | * DTLs then reset its resilvering flag and dirty | |
1901 | * the top level so that we persist the change. | |
1902 | */ | |
1903 | if (vd->vdev_resilver_txg != 0 && | |
1904 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && | |
1905 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) { | |
1906 | vd->vdev_resilver_txg = 0; | |
1907 | vdev_config_dirty(vd->vdev_top); | |
1908 | } | |
1909 | ||
1910 | mutex_exit(&vd->vdev_dtl_lock); | |
1911 | ||
1912 | if (txg != 0) | |
1913 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1914 | return; | |
1915 | } | |
1916 | ||
1917 | mutex_enter(&vd->vdev_dtl_lock); | |
1918 | for (t = 0; t < DTL_TYPES; t++) { | |
1919 | int c; | |
1920 | ||
1921 | /* account for child's outage in parent's missing map */ | |
1922 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
1923 | if (t == DTL_SCRUB) | |
1924 | continue; /* leaf vdevs only */ | |
1925 | if (t == DTL_PARTIAL) | |
1926 | minref = 1; /* i.e. non-zero */ | |
1927 | else if (vd->vdev_nparity != 0) | |
1928 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1929 | else | |
1930 | minref = vd->vdev_children; /* any kind of mirror */ | |
1931 | space_reftree_create(&reftree); | |
1932 | for (c = 0; c < vd->vdev_children; c++) { | |
1933 | vdev_t *cvd = vd->vdev_child[c]; | |
1934 | mutex_enter(&cvd->vdev_dtl_lock); | |
1935 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); | |
1936 | mutex_exit(&cvd->vdev_dtl_lock); | |
1937 | } | |
1938 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); | |
1939 | space_reftree_destroy(&reftree); | |
1940 | } | |
1941 | mutex_exit(&vd->vdev_dtl_lock); | |
1942 | } | |
1943 | ||
1944 | int | |
1945 | vdev_dtl_load(vdev_t *vd) | |
1946 | { | |
1947 | spa_t *spa = vd->vdev_spa; | |
1948 | objset_t *mos = spa->spa_meta_objset; | |
1949 | int error = 0; | |
1950 | int c; | |
1951 | ||
1952 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { | |
1953 | ASSERT(!vd->vdev_ishole); | |
1954 | ||
1955 | error = space_map_open(&vd->vdev_dtl_sm, mos, | |
1956 | vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); | |
1957 | if (error) | |
1958 | return (error); | |
1959 | ASSERT(vd->vdev_dtl_sm != NULL); | |
1960 | ||
1961 | mutex_enter(&vd->vdev_dtl_lock); | |
1962 | ||
1963 | /* | |
1964 | * Now that we've opened the space_map we need to update | |
1965 | * the in-core DTL. | |
1966 | */ | |
1967 | space_map_update(vd->vdev_dtl_sm); | |
1968 | ||
1969 | error = space_map_load(vd->vdev_dtl_sm, | |
1970 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
1971 | mutex_exit(&vd->vdev_dtl_lock); | |
1972 | ||
1973 | return (error); | |
1974 | } | |
1975 | ||
1976 | for (c = 0; c < vd->vdev_children; c++) { | |
1977 | error = vdev_dtl_load(vd->vdev_child[c]); | |
1978 | if (error != 0) | |
1979 | break; | |
1980 | } | |
1981 | ||
1982 | return (error); | |
1983 | } | |
1984 | ||
1985 | void | |
1986 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
1987 | { | |
1988 | spa_t *spa = vd->vdev_spa; | |
1989 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; | |
1990 | objset_t *mos = spa->spa_meta_objset; | |
1991 | range_tree_t *rtsync; | |
1992 | kmutex_t rtlock; | |
1993 | dmu_tx_t *tx; | |
1994 | uint64_t object = space_map_object(vd->vdev_dtl_sm); | |
1995 | ||
1996 | ASSERT(!vd->vdev_ishole); | |
1997 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
1998 | ||
1999 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2000 | ||
2001 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { | |
2002 | mutex_enter(&vd->vdev_dtl_lock); | |
2003 | space_map_free(vd->vdev_dtl_sm, tx); | |
2004 | space_map_close(vd->vdev_dtl_sm); | |
2005 | vd->vdev_dtl_sm = NULL; | |
2006 | mutex_exit(&vd->vdev_dtl_lock); | |
2007 | dmu_tx_commit(tx); | |
2008 | return; | |
2009 | } | |
2010 | ||
2011 | if (vd->vdev_dtl_sm == NULL) { | |
2012 | uint64_t new_object; | |
2013 | ||
2014 | new_object = space_map_alloc(mos, tx); | |
2015 | VERIFY3U(new_object, !=, 0); | |
2016 | ||
2017 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
2018 | 0, -1ULL, 0, &vd->vdev_dtl_lock)); | |
2019 | ASSERT(vd->vdev_dtl_sm != NULL); | |
2020 | } | |
2021 | ||
2022 | mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); | |
2023 | ||
2024 | rtsync = range_tree_create(NULL, NULL, &rtlock); | |
2025 | ||
2026 | mutex_enter(&rtlock); | |
2027 | ||
2028 | mutex_enter(&vd->vdev_dtl_lock); | |
2029 | range_tree_walk(rt, range_tree_add, rtsync); | |
2030 | mutex_exit(&vd->vdev_dtl_lock); | |
2031 | ||
2032 | space_map_truncate(vd->vdev_dtl_sm, tx); | |
2033 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
2034 | range_tree_vacate(rtsync, NULL, NULL); | |
2035 | ||
2036 | range_tree_destroy(rtsync); | |
2037 | ||
2038 | mutex_exit(&rtlock); | |
2039 | mutex_destroy(&rtlock); | |
2040 | ||
2041 | /* | |
2042 | * If the object for the space map has changed then dirty | |
2043 | * the top level so that we update the config. | |
2044 | */ | |
2045 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
2046 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
2047 | "new object %llu", txg, spa_name(spa), object, | |
2048 | space_map_object(vd->vdev_dtl_sm)); | |
2049 | vdev_config_dirty(vd->vdev_top); | |
2050 | } | |
2051 | ||
2052 | dmu_tx_commit(tx); | |
2053 | ||
2054 | mutex_enter(&vd->vdev_dtl_lock); | |
2055 | space_map_update(vd->vdev_dtl_sm); | |
2056 | mutex_exit(&vd->vdev_dtl_lock); | |
2057 | } | |
2058 | ||
2059 | /* | |
2060 | * Determine whether the specified vdev can be offlined/detached/removed | |
2061 | * without losing data. | |
2062 | */ | |
2063 | boolean_t | |
2064 | vdev_dtl_required(vdev_t *vd) | |
2065 | { | |
2066 | spa_t *spa = vd->vdev_spa; | |
2067 | vdev_t *tvd = vd->vdev_top; | |
2068 | uint8_t cant_read = vd->vdev_cant_read; | |
2069 | boolean_t required; | |
2070 | ||
2071 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2072 | ||
2073 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2074 | return (B_TRUE); | |
2075 | ||
2076 | /* | |
2077 | * Temporarily mark the device as unreadable, and then determine | |
2078 | * whether this results in any DTL outages in the top-level vdev. | |
2079 | * If not, we can safely offline/detach/remove the device. | |
2080 | */ | |
2081 | vd->vdev_cant_read = B_TRUE; | |
2082 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2083 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2084 | vd->vdev_cant_read = cant_read; | |
2085 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2086 | ||
2087 | if (!required && zio_injection_enabled) | |
2088 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2089 | ||
2090 | return (required); | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * Determine if resilver is needed, and if so the txg range. | |
2095 | */ | |
2096 | boolean_t | |
2097 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2098 | { | |
2099 | boolean_t needed = B_FALSE; | |
2100 | uint64_t thismin = UINT64_MAX; | |
2101 | uint64_t thismax = 0; | |
2102 | int c; | |
2103 | ||
2104 | if (vd->vdev_children == 0) { | |
2105 | mutex_enter(&vd->vdev_dtl_lock); | |
2106 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && | |
2107 | vdev_writeable(vd)) { | |
2108 | ||
2109 | thismin = vdev_dtl_min(vd); | |
2110 | thismax = vdev_dtl_max(vd); | |
2111 | needed = B_TRUE; | |
2112 | } | |
2113 | mutex_exit(&vd->vdev_dtl_lock); | |
2114 | } else { | |
2115 | for (c = 0; c < vd->vdev_children; c++) { | |
2116 | vdev_t *cvd = vd->vdev_child[c]; | |
2117 | uint64_t cmin, cmax; | |
2118 | ||
2119 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2120 | thismin = MIN(thismin, cmin); | |
2121 | thismax = MAX(thismax, cmax); | |
2122 | needed = B_TRUE; | |
2123 | } | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | if (needed && minp) { | |
2128 | *minp = thismin; | |
2129 | *maxp = thismax; | |
2130 | } | |
2131 | return (needed); | |
2132 | } | |
2133 | ||
2134 | void | |
2135 | vdev_load(vdev_t *vd) | |
2136 | { | |
2137 | int c; | |
2138 | ||
2139 | /* | |
2140 | * Recursively load all children. | |
2141 | */ | |
2142 | for (c = 0; c < vd->vdev_children; c++) | |
2143 | vdev_load(vd->vdev_child[c]); | |
2144 | ||
2145 | /* | |
2146 | * If this is a top-level vdev, initialize its metaslabs. | |
2147 | */ | |
2148 | if (vd == vd->vdev_top && !vd->vdev_ishole && | |
2149 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || | |
2150 | vdev_metaslab_init(vd, 0) != 0)) | |
2151 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2152 | VDEV_AUX_CORRUPT_DATA); | |
2153 | ||
2154 | /* | |
2155 | * If this is a leaf vdev, load its DTL. | |
2156 | */ | |
2157 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
2158 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2159 | VDEV_AUX_CORRUPT_DATA); | |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2164 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2165 | * we make sure that we can open the underlying device, then try to read the | |
2166 | * label, and make sure that the label is sane and that it hasn't been | |
2167 | * repurposed to another pool. | |
2168 | */ | |
2169 | int | |
2170 | vdev_validate_aux(vdev_t *vd) | |
2171 | { | |
2172 | nvlist_t *label; | |
2173 | uint64_t guid, version; | |
2174 | uint64_t state; | |
2175 | ||
2176 | if (!vdev_readable(vd)) | |
2177 | return (0); | |
2178 | ||
2179 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { | |
2180 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2181 | VDEV_AUX_CORRUPT_DATA); | |
2182 | return (-1); | |
2183 | } | |
2184 | ||
2185 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
2186 | !SPA_VERSION_IS_SUPPORTED(version) || | |
2187 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || | |
2188 | guid != vd->vdev_guid || | |
2189 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2190 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2191 | VDEV_AUX_CORRUPT_DATA); | |
2192 | nvlist_free(label); | |
2193 | return (-1); | |
2194 | } | |
2195 | ||
2196 | /* | |
2197 | * We don't actually check the pool state here. If it's in fact in | |
2198 | * use by another pool, we update this fact on the fly when requested. | |
2199 | */ | |
2200 | nvlist_free(label); | |
2201 | return (0); | |
2202 | } | |
2203 | ||
2204 | void | |
2205 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2206 | { | |
2207 | spa_t *spa = vd->vdev_spa; | |
2208 | objset_t *mos = spa->spa_meta_objset; | |
2209 | dmu_tx_t *tx; | |
2210 | int m, i; | |
2211 | ||
2212 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
2213 | ||
2214 | if (vd->vdev_ms != NULL) { | |
2215 | metaslab_group_t *mg = vd->vdev_mg; | |
2216 | ||
2217 | metaslab_group_histogram_verify(mg); | |
2218 | metaslab_class_histogram_verify(mg->mg_class); | |
2219 | ||
2220 | for (m = 0; m < vd->vdev_ms_count; m++) { | |
2221 | metaslab_t *msp = vd->vdev_ms[m]; | |
2222 | ||
2223 | if (msp == NULL || msp->ms_sm == NULL) | |
2224 | continue; | |
2225 | ||
2226 | mutex_enter(&msp->ms_lock); | |
2227 | /* | |
2228 | * If the metaslab was not loaded when the vdev | |
2229 | * was removed then the histogram accounting may | |
2230 | * not be accurate. Update the histogram information | |
2231 | * here so that we ensure that the metaslab group | |
2232 | * and metaslab class are up-to-date. | |
2233 | */ | |
2234 | metaslab_group_histogram_remove(mg, msp); | |
2235 | ||
2236 | VERIFY0(space_map_allocated(msp->ms_sm)); | |
2237 | space_map_free(msp->ms_sm, tx); | |
2238 | space_map_close(msp->ms_sm); | |
2239 | msp->ms_sm = NULL; | |
2240 | mutex_exit(&msp->ms_lock); | |
2241 | } | |
2242 | ||
2243 | metaslab_group_histogram_verify(mg); | |
2244 | metaslab_class_histogram_verify(mg->mg_class); | |
2245 | for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) | |
2246 | ASSERT0(mg->mg_histogram[i]); | |
2247 | ||
2248 | } | |
2249 | ||
2250 | if (vd->vdev_ms_array) { | |
2251 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2252 | vd->vdev_ms_array = 0; | |
2253 | } | |
2254 | dmu_tx_commit(tx); | |
2255 | } | |
2256 | ||
2257 | void | |
2258 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2259 | { | |
2260 | metaslab_t *msp; | |
2261 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); | |
2262 | ||
2263 | ASSERT(!vd->vdev_ishole); | |
2264 | ||
2265 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) | |
2266 | metaslab_sync_done(msp, txg); | |
2267 | ||
2268 | if (reassess) | |
2269 | metaslab_sync_reassess(vd->vdev_mg); | |
2270 | } | |
2271 | ||
2272 | void | |
2273 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2274 | { | |
2275 | spa_t *spa = vd->vdev_spa; | |
2276 | vdev_t *lvd; | |
2277 | metaslab_t *msp; | |
2278 | dmu_tx_t *tx; | |
2279 | ||
2280 | ASSERT(!vd->vdev_ishole); | |
2281 | ||
2282 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { | |
2283 | ASSERT(vd == vd->vdev_top); | |
2284 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2285 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2286 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2287 | ASSERT(vd->vdev_ms_array != 0); | |
2288 | vdev_config_dirty(vd); | |
2289 | dmu_tx_commit(tx); | |
2290 | } | |
2291 | ||
2292 | /* | |
2293 | * Remove the metadata associated with this vdev once it's empty. | |
2294 | */ | |
2295 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2296 | vdev_remove(vd, txg); | |
2297 | ||
2298 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { | |
2299 | metaslab_sync(msp, txg); | |
2300 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2301 | } | |
2302 | ||
2303 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2304 | vdev_dtl_sync(lvd, txg); | |
2305 | ||
2306 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2307 | } | |
2308 | ||
2309 | uint64_t | |
2310 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2311 | { | |
2312 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2313 | } | |
2314 | ||
2315 | /* | |
2316 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2317 | * not be opened, and no I/O is attempted. | |
2318 | */ | |
2319 | int | |
2320 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) | |
2321 | { | |
2322 | vdev_t *vd, *tvd; | |
2323 | ||
2324 | spa_vdev_state_enter(spa, SCL_NONE); | |
2325 | ||
2326 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
2327 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
2328 | ||
2329 | if (!vd->vdev_ops->vdev_op_leaf) | |
2330 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
2331 | ||
2332 | tvd = vd->vdev_top; | |
2333 | ||
2334 | /* | |
2335 | * We don't directly use the aux state here, but if we do a | |
2336 | * vdev_reopen(), we need this value to be present to remember why we | |
2337 | * were faulted. | |
2338 | */ | |
2339 | vd->vdev_label_aux = aux; | |
2340 | ||
2341 | /* | |
2342 | * Faulted state takes precedence over degraded. | |
2343 | */ | |
2344 | vd->vdev_delayed_close = B_FALSE; | |
2345 | vd->vdev_faulted = 1ULL; | |
2346 | vd->vdev_degraded = 0ULL; | |
2347 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); | |
2348 | ||
2349 | /* | |
2350 | * If this device has the only valid copy of the data, then | |
2351 | * back off and simply mark the vdev as degraded instead. | |
2352 | */ | |
2353 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { | |
2354 | vd->vdev_degraded = 1ULL; | |
2355 | vd->vdev_faulted = 0ULL; | |
2356 | ||
2357 | /* | |
2358 | * If we reopen the device and it's not dead, only then do we | |
2359 | * mark it degraded. | |
2360 | */ | |
2361 | vdev_reopen(tvd); | |
2362 | ||
2363 | if (vdev_readable(vd)) | |
2364 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
2365 | } | |
2366 | ||
2367 | return (spa_vdev_state_exit(spa, vd, 0)); | |
2368 | } | |
2369 | ||
2370 | /* | |
2371 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2372 | * user that something is wrong. The vdev continues to operate as normal as far | |
2373 | * as I/O is concerned. | |
2374 | */ | |
2375 | int | |
2376 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) | |
2377 | { | |
2378 | vdev_t *vd; | |
2379 | ||
2380 | spa_vdev_state_enter(spa, SCL_NONE); | |
2381 | ||
2382 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
2383 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
2384 | ||
2385 | if (!vd->vdev_ops->vdev_op_leaf) | |
2386 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
2387 | ||
2388 | /* | |
2389 | * If the vdev is already faulted, then don't do anything. | |
2390 | */ | |
2391 | if (vd->vdev_faulted || vd->vdev_degraded) | |
2392 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
2393 | ||
2394 | vd->vdev_degraded = 1ULL; | |
2395 | if (!vdev_is_dead(vd)) | |
2396 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
2397 | aux); | |
2398 | ||
2399 | return (spa_vdev_state_exit(spa, vd, 0)); | |
2400 | } | |
2401 | ||
2402 | /* | |
2403 | * Online the given vdev. | |
2404 | * | |
2405 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2406 | * spare device should be detached when the device finishes resilvering. | |
2407 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2408 | * events are generated if the device fails to open. | |
2409 | */ | |
2410 | int | |
2411 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) | |
2412 | { | |
2413 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; | |
2414 | ||
2415 | spa_vdev_state_enter(spa, SCL_NONE); | |
2416 | ||
2417 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
2418 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
2419 | ||
2420 | if (!vd->vdev_ops->vdev_op_leaf) | |
2421 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
2422 | ||
2423 | tvd = vd->vdev_top; | |
2424 | vd->vdev_offline = B_FALSE; | |
2425 | vd->vdev_tmpoffline = B_FALSE; | |
2426 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); | |
2427 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
2428 | ||
2429 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2430 | if (!vd->vdev_aux) { | |
2431 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2432 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2433 | } | |
2434 | ||
2435 | vdev_reopen(tvd); | |
2436 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; | |
2437 | ||
2438 | if (!vd->vdev_aux) { | |
2439 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2440 | pvd->vdev_expanding = B_FALSE; | |
2441 | } | |
2442 | ||
2443 | if (newstate) | |
2444 | *newstate = vd->vdev_state; | |
2445 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2446 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2447 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2448 | vd->vdev_parent->vdev_child[0] == vd) | |
2449 | vd->vdev_unspare = B_TRUE; | |
2450 | ||
2451 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { | |
2452 | ||
2453 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2454 | if (vd->vdev_aux) | |
2455 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2456 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2457 | } | |
2458 | return (spa_vdev_state_exit(spa, vd, 0)); | |
2459 | } | |
2460 | ||
2461 | static int | |
2462 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
2463 | { | |
2464 | vdev_t *vd, *tvd; | |
2465 | int error = 0; | |
2466 | uint64_t generation; | |
2467 | metaslab_group_t *mg; | |
2468 | ||
2469 | top: | |
2470 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2471 | ||
2472 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
2473 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
2474 | ||
2475 | if (!vd->vdev_ops->vdev_op_leaf) | |
2476 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
2477 | ||
2478 | tvd = vd->vdev_top; | |
2479 | mg = tvd->vdev_mg; | |
2480 | generation = spa->spa_config_generation + 1; | |
2481 | ||
2482 | /* | |
2483 | * If the device isn't already offline, try to offline it. | |
2484 | */ | |
2485 | if (!vd->vdev_offline) { | |
2486 | /* | |
2487 | * If this device has the only valid copy of some data, | |
2488 | * don't allow it to be offlined. Log devices are always | |
2489 | * expendable. | |
2490 | */ | |
2491 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2492 | vdev_dtl_required(vd)) | |
2493 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); | |
2494 | ||
2495 | /* | |
2496 | * If the top-level is a slog and it has had allocations | |
2497 | * then proceed. We check that the vdev's metaslab group | |
2498 | * is not NULL since it's possible that we may have just | |
2499 | * added this vdev but not yet initialized its metaslabs. | |
2500 | */ | |
2501 | if (tvd->vdev_islog && mg != NULL) { | |
2502 | /* | |
2503 | * Prevent any future allocations. | |
2504 | */ | |
2505 | metaslab_group_passivate(mg); | |
2506 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2507 | ||
2508 | error = spa_offline_log(spa); | |
2509 | ||
2510 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2511 | ||
2512 | /* | |
2513 | * Check to see if the config has changed. | |
2514 | */ | |
2515 | if (error || generation != spa->spa_config_generation) { | |
2516 | metaslab_group_activate(mg); | |
2517 | if (error) | |
2518 | return (spa_vdev_state_exit(spa, | |
2519 | vd, error)); | |
2520 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2521 | goto top; | |
2522 | } | |
2523 | ASSERT0(tvd->vdev_stat.vs_alloc); | |
2524 | } | |
2525 | ||
2526 | /* | |
2527 | * Offline this device and reopen its top-level vdev. | |
2528 | * If the top-level vdev is a log device then just offline | |
2529 | * it. Otherwise, if this action results in the top-level | |
2530 | * vdev becoming unusable, undo it and fail the request. | |
2531 | */ | |
2532 | vd->vdev_offline = B_TRUE; | |
2533 | vdev_reopen(tvd); | |
2534 | ||
2535 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2536 | vdev_is_dead(tvd)) { | |
2537 | vd->vdev_offline = B_FALSE; | |
2538 | vdev_reopen(tvd); | |
2539 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); | |
2540 | } | |
2541 | ||
2542 | /* | |
2543 | * Add the device back into the metaslab rotor so that | |
2544 | * once we online the device it's open for business. | |
2545 | */ | |
2546 | if (tvd->vdev_islog && mg != NULL) | |
2547 | metaslab_group_activate(mg); | |
2548 | } | |
2549 | ||
2550 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); | |
2551 | ||
2552 | return (spa_vdev_state_exit(spa, vd, 0)); | |
2553 | } | |
2554 | ||
2555 | int | |
2556 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2557 | { | |
2558 | int error; | |
2559 | ||
2560 | mutex_enter(&spa->spa_vdev_top_lock); | |
2561 | error = vdev_offline_locked(spa, guid, flags); | |
2562 | mutex_exit(&spa->spa_vdev_top_lock); | |
2563 | ||
2564 | return (error); | |
2565 | } | |
2566 | ||
2567 | /* | |
2568 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2569 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2570 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
2571 | */ | |
2572 | void | |
2573 | vdev_clear(spa_t *spa, vdev_t *vd) | |
2574 | { | |
2575 | vdev_t *rvd = spa->spa_root_vdev; | |
2576 | int c; | |
2577 | ||
2578 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2579 | ||
2580 | if (vd == NULL) | |
2581 | vd = rvd; | |
2582 | ||
2583 | vd->vdev_stat.vs_read_errors = 0; | |
2584 | vd->vdev_stat.vs_write_errors = 0; | |
2585 | vd->vdev_stat.vs_checksum_errors = 0; | |
2586 | ||
2587 | for (c = 0; c < vd->vdev_children; c++) | |
2588 | vdev_clear(spa, vd->vdev_child[c]); | |
2589 | ||
2590 | /* | |
2591 | * If we're in the FAULTED state or have experienced failed I/O, then | |
2592 | * clear the persistent state and attempt to reopen the device. We | |
2593 | * also mark the vdev config dirty, so that the new faulted state is | |
2594 | * written out to disk. | |
2595 | */ | |
2596 | if (vd->vdev_faulted || vd->vdev_degraded || | |
2597 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2598 | ||
2599 | /* | |
2600 | * When reopening in reponse to a clear event, it may be due to | |
2601 | * a fmadm repair request. In this case, if the device is | |
2602 | * still broken, we want to still post the ereport again. | |
2603 | */ | |
2604 | vd->vdev_forcefault = B_TRUE; | |
2605 | ||
2606 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; | |
2607 | vd->vdev_cant_read = B_FALSE; | |
2608 | vd->vdev_cant_write = B_FALSE; | |
2609 | ||
2610 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); | |
2611 | ||
2612 | vd->vdev_forcefault = B_FALSE; | |
2613 | ||
2614 | if (vd != rvd && vdev_writeable(vd->vdev_top)) | |
2615 | vdev_state_dirty(vd->vdev_top); | |
2616 | ||
2617 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
2618 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
2619 | ||
2620 | spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR); | |
2621 | } | |
2622 | ||
2623 | /* | |
2624 | * When clearing a FMA-diagnosed fault, we always want to | |
2625 | * unspare the device, as we assume that the original spare was | |
2626 | * done in response to the FMA fault. | |
2627 | */ | |
2628 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2629 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2630 | vd->vdev_parent->vdev_child[0] == vd) | |
2631 | vd->vdev_unspare = B_TRUE; | |
2632 | } | |
2633 | ||
2634 | boolean_t | |
2635 | vdev_is_dead(vdev_t *vd) | |
2636 | { | |
2637 | /* | |
2638 | * Holes and missing devices are always considered "dead". | |
2639 | * This simplifies the code since we don't have to check for | |
2640 | * these types of devices in the various code paths. | |
2641 | * Instead we rely on the fact that we skip over dead devices | |
2642 | * before issuing I/O to them. | |
2643 | */ | |
2644 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2645 | vd->vdev_ops == &vdev_missing_ops); | |
2646 | } | |
2647 | ||
2648 | boolean_t | |
2649 | vdev_readable(vdev_t *vd) | |
2650 | { | |
2651 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); | |
2652 | } | |
2653 | ||
2654 | boolean_t | |
2655 | vdev_writeable(vdev_t *vd) | |
2656 | { | |
2657 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); | |
2658 | } | |
2659 | ||
2660 | boolean_t | |
2661 | vdev_allocatable(vdev_t *vd) | |
2662 | { | |
2663 | uint64_t state = vd->vdev_state; | |
2664 | ||
2665 | /* | |
2666 | * We currently allow allocations from vdevs which may be in the | |
2667 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device | |
2668 | * fails to reopen then we'll catch it later when we're holding | |
2669 | * the proper locks. Note that we have to get the vdev state | |
2670 | * in a local variable because although it changes atomically, | |
2671 | * we're asking two separate questions about it. | |
2672 | */ | |
2673 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && | |
2674 | !vd->vdev_cant_write && !vd->vdev_ishole); | |
2675 | } | |
2676 | ||
2677 | boolean_t | |
2678 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
2679 | { | |
2680 | ASSERT(zio->io_vd == vd); | |
2681 | ||
2682 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) | |
2683 | return (B_FALSE); | |
2684 | ||
2685 | if (zio->io_type == ZIO_TYPE_READ) | |
2686 | return (!vd->vdev_cant_read); | |
2687 | ||
2688 | if (zio->io_type == ZIO_TYPE_WRITE) | |
2689 | return (!vd->vdev_cant_write); | |
2690 | ||
2691 | return (B_TRUE); | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * Get statistics for the given vdev. | |
2696 | */ | |
2697 | void | |
2698 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2699 | { | |
2700 | spa_t *spa = vd->vdev_spa; | |
2701 | vdev_t *rvd = spa->spa_root_vdev; | |
2702 | int c, t; | |
2703 | ||
2704 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); | |
2705 | ||
2706 | mutex_enter(&vd->vdev_stat_lock); | |
2707 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2708 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2709 | vs->vs_state = vd->vdev_state; | |
2710 | vs->vs_rsize = vdev_get_min_asize(vd); | |
2711 | if (vd->vdev_ops->vdev_op_leaf) | |
2712 | vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; | |
2713 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; | |
2714 | if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { | |
2715 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; | |
2716 | } | |
2717 | ||
2718 | /* | |
2719 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2720 | * over all top-level vdevs (i.e. the direct children of the root). | |
2721 | */ | |
2722 | if (vd == rvd) { | |
2723 | for (c = 0; c < rvd->vdev_children; c++) { | |
2724 | vdev_t *cvd = rvd->vdev_child[c]; | |
2725 | vdev_stat_t *cvs = &cvd->vdev_stat; | |
2726 | ||
2727 | for (t = 0; t < ZIO_TYPES; t++) { | |
2728 | vs->vs_ops[t] += cvs->vs_ops[t]; | |
2729 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2730 | } | |
2731 | cvs->vs_scan_removing = cvd->vdev_removing; | |
2732 | } | |
2733 | } | |
2734 | mutex_exit(&vd->vdev_stat_lock); | |
2735 | } | |
2736 | ||
2737 | void | |
2738 | vdev_clear_stats(vdev_t *vd) | |
2739 | { | |
2740 | mutex_enter(&vd->vdev_stat_lock); | |
2741 | vd->vdev_stat.vs_space = 0; | |
2742 | vd->vdev_stat.vs_dspace = 0; | |
2743 | vd->vdev_stat.vs_alloc = 0; | |
2744 | mutex_exit(&vd->vdev_stat_lock); | |
2745 | } | |
2746 | ||
2747 | void | |
2748 | vdev_scan_stat_init(vdev_t *vd) | |
2749 | { | |
2750 | vdev_stat_t *vs = &vd->vdev_stat; | |
2751 | int c; | |
2752 | ||
2753 | for (c = 0; c < vd->vdev_children; c++) | |
2754 | vdev_scan_stat_init(vd->vdev_child[c]); | |
2755 | ||
2756 | mutex_enter(&vd->vdev_stat_lock); | |
2757 | vs->vs_scan_processed = 0; | |
2758 | mutex_exit(&vd->vdev_stat_lock); | |
2759 | } | |
2760 | ||
2761 | void | |
2762 | vdev_stat_update(zio_t *zio, uint64_t psize) | |
2763 | { | |
2764 | spa_t *spa = zio->io_spa; | |
2765 | vdev_t *rvd = spa->spa_root_vdev; | |
2766 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; | |
2767 | vdev_t *pvd; | |
2768 | uint64_t txg = zio->io_txg; | |
2769 | vdev_stat_t *vs = &vd->vdev_stat; | |
2770 | zio_type_t type = zio->io_type; | |
2771 | int flags = zio->io_flags; | |
2772 | ||
2773 | /* | |
2774 | * If this i/o is a gang leader, it didn't do any actual work. | |
2775 | */ | |
2776 | if (zio->io_gang_tree) | |
2777 | return; | |
2778 | ||
2779 | if (zio->io_error == 0) { | |
2780 | /* | |
2781 | * If this is a root i/o, don't count it -- we've already | |
2782 | * counted the top-level vdevs, and vdev_get_stats() will | |
2783 | * aggregate them when asked. This reduces contention on | |
2784 | * the root vdev_stat_lock and implicitly handles blocks | |
2785 | * that compress away to holes, for which there is no i/o. | |
2786 | * (Holes never create vdev children, so all the counters | |
2787 | * remain zero, which is what we want.) | |
2788 | * | |
2789 | * Note: this only applies to successful i/o (io_error == 0) | |
2790 | * because unlike i/o counts, errors are not additive. | |
2791 | * When reading a ditto block, for example, failure of | |
2792 | * one top-level vdev does not imply a root-level error. | |
2793 | */ | |
2794 | if (vd == rvd) | |
2795 | return; | |
2796 | ||
2797 | ASSERT(vd == zio->io_vd); | |
2798 | ||
2799 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2800 | return; | |
2801 | ||
2802 | mutex_enter(&vd->vdev_stat_lock); | |
2803 | ||
2804 | if (flags & ZIO_FLAG_IO_REPAIR) { | |
2805 | if (flags & ZIO_FLAG_SCAN_THREAD) { | |
2806 | dsl_scan_phys_t *scn_phys = | |
2807 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
2808 | uint64_t *processed = &scn_phys->scn_processed; | |
2809 | ||
2810 | /* XXX cleanup? */ | |
2811 | if (vd->vdev_ops->vdev_op_leaf) | |
2812 | atomic_add_64(processed, psize); | |
2813 | vs->vs_scan_processed += psize; | |
2814 | } | |
2815 | ||
2816 | if (flags & ZIO_FLAG_SELF_HEAL) | |
2817 | vs->vs_self_healed += psize; | |
2818 | } | |
2819 | ||
2820 | vs->vs_ops[type]++; | |
2821 | vs->vs_bytes[type] += psize; | |
2822 | ||
2823 | mutex_exit(&vd->vdev_stat_lock); | |
2824 | return; | |
2825 | } | |
2826 | ||
2827 | if (flags & ZIO_FLAG_SPECULATIVE) | |
2828 | return; | |
2829 | ||
2830 | /* | |
2831 | * If this is an I/O error that is going to be retried, then ignore the | |
2832 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
2833 | * hard errors, when in reality they can happen for any number of | |
2834 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
2835 | */ | |
2836 | if (zio->io_error == EIO && | |
2837 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
2838 | return; | |
2839 | ||
2840 | /* | |
2841 | * Intent logs writes won't propagate their error to the root | |
2842 | * I/O so don't mark these types of failures as pool-level | |
2843 | * errors. | |
2844 | */ | |
2845 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
2846 | return; | |
2847 | ||
2848 | mutex_enter(&vd->vdev_stat_lock); | |
2849 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { | |
2850 | if (zio->io_error == ECKSUM) | |
2851 | vs->vs_checksum_errors++; | |
2852 | else | |
2853 | vs->vs_read_errors++; | |
2854 | } | |
2855 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) | |
2856 | vs->vs_write_errors++; | |
2857 | mutex_exit(&vd->vdev_stat_lock); | |
2858 | ||
2859 | if (type == ZIO_TYPE_WRITE && txg != 0 && | |
2860 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
2861 | (flags & ZIO_FLAG_SCAN_THREAD) || | |
2862 | spa->spa_claiming)) { | |
2863 | /* | |
2864 | * This is either a normal write (not a repair), or it's | |
2865 | * a repair induced by the scrub thread, or it's a repair | |
2866 | * made by zil_claim() during spa_load() in the first txg. | |
2867 | * In the normal case, we commit the DTL change in the same | |
2868 | * txg as the block was born. In the scrub-induced repair | |
2869 | * case, we know that scrubs run in first-pass syncing context, | |
2870 | * so we commit the DTL change in spa_syncing_txg(spa). | |
2871 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
2872 | * | |
2873 | * We currently do not make DTL entries for failed spontaneous | |
2874 | * self-healing writes triggered by normal (non-scrubbing) | |
2875 | * reads, because we have no transactional context in which to | |
2876 | * do so -- and it's not clear that it'd be desirable anyway. | |
2877 | */ | |
2878 | if (vd->vdev_ops->vdev_op_leaf) { | |
2879 | uint64_t commit_txg = txg; | |
2880 | if (flags & ZIO_FLAG_SCAN_THREAD) { | |
2881 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2882 | ASSERT(spa_sync_pass(spa) == 1); | |
2883 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
2884 | commit_txg = spa_syncing_txg(spa); | |
2885 | } else if (spa->spa_claiming) { | |
2886 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
2887 | commit_txg = spa_first_txg(spa); | |
2888 | } | |
2889 | ASSERT(commit_txg >= spa_syncing_txg(spa)); | |
2890 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) | |
2891 | return; | |
2892 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2893 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
2894 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
2895 | } | |
2896 | if (vd != rvd) | |
2897 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
2898 | } | |
2899 | } | |
2900 | ||
2901 | /* | |
2902 | * Update the in-core space usage stats for this vdev, its metaslab class, | |
2903 | * and the root vdev. | |
2904 | */ | |
2905 | void | |
2906 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, | |
2907 | int64_t space_delta) | |
2908 | { | |
2909 | int64_t dspace_delta = space_delta; | |
2910 | spa_t *spa = vd->vdev_spa; | |
2911 | vdev_t *rvd = spa->spa_root_vdev; | |
2912 | metaslab_group_t *mg = vd->vdev_mg; | |
2913 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
2914 | ||
2915 | ASSERT(vd == vd->vdev_top); | |
2916 | ||
2917 | /* | |
2918 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
2919 | * factor. We must calculate this here and not at the root vdev | |
2920 | * because the root vdev's psize-to-asize is simply the max of its | |
2921 | * childrens', thus not accurate enough for us. | |
2922 | */ | |
2923 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
2924 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); | |
2925 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * | |
2926 | vd->vdev_deflate_ratio; | |
2927 | ||
2928 | mutex_enter(&vd->vdev_stat_lock); | |
2929 | vd->vdev_stat.vs_alloc += alloc_delta; | |
2930 | vd->vdev_stat.vs_space += space_delta; | |
2931 | vd->vdev_stat.vs_dspace += dspace_delta; | |
2932 | mutex_exit(&vd->vdev_stat_lock); | |
2933 | ||
2934 | if (mc == spa_normal_class(spa)) { | |
2935 | mutex_enter(&rvd->vdev_stat_lock); | |
2936 | rvd->vdev_stat.vs_alloc += alloc_delta; | |
2937 | rvd->vdev_stat.vs_space += space_delta; | |
2938 | rvd->vdev_stat.vs_dspace += dspace_delta; | |
2939 | mutex_exit(&rvd->vdev_stat_lock); | |
2940 | } | |
2941 | ||
2942 | if (mc != NULL) { | |
2943 | ASSERT(rvd == vd->vdev_parent); | |
2944 | ASSERT(vd->vdev_ms_count != 0); | |
2945 | ||
2946 | metaslab_class_space_update(mc, | |
2947 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
2948 | } | |
2949 | } | |
2950 | ||
2951 | /* | |
2952 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
2953 | * so that it will be written out next time the vdev configuration is synced. | |
2954 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
2955 | */ | |
2956 | void | |
2957 | vdev_config_dirty(vdev_t *vd) | |
2958 | { | |
2959 | spa_t *spa = vd->vdev_spa; | |
2960 | vdev_t *rvd = spa->spa_root_vdev; | |
2961 | int c; | |
2962 | ||
2963 | ASSERT(spa_writeable(spa)); | |
2964 | ||
2965 | /* | |
2966 | * If this is an aux vdev (as with l2cache and spare devices), then we | |
2967 | * update the vdev config manually and set the sync flag. | |
2968 | */ | |
2969 | if (vd->vdev_aux != NULL) { | |
2970 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
2971 | nvlist_t **aux; | |
2972 | uint_t naux; | |
2973 | ||
2974 | for (c = 0; c < sav->sav_count; c++) { | |
2975 | if (sav->sav_vdevs[c] == vd) | |
2976 | break; | |
2977 | } | |
2978 | ||
2979 | if (c == sav->sav_count) { | |
2980 | /* | |
2981 | * We're being removed. There's nothing more to do. | |
2982 | */ | |
2983 | ASSERT(sav->sav_sync == B_TRUE); | |
2984 | return; | |
2985 | } | |
2986 | ||
2987 | sav->sav_sync = B_TRUE; | |
2988 | ||
2989 | if (nvlist_lookup_nvlist_array(sav->sav_config, | |
2990 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
2991 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
2992 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
2993 | } | |
2994 | ||
2995 | ASSERT(c < naux); | |
2996 | ||
2997 | /* | |
2998 | * Setting the nvlist in the middle if the array is a little | |
2999 | * sketchy, but it will work. | |
3000 | */ | |
3001 | nvlist_free(aux[c]); | |
3002 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); | |
3003 | ||
3004 | return; | |
3005 | } | |
3006 | ||
3007 | /* | |
3008 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
3009 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
3010 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
3011 | * so this is sufficient to ensure mutual exclusion. | |
3012 | */ | |
3013 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || | |
3014 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3015 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
3016 | ||
3017 | if (vd == rvd) { | |
3018 | for (c = 0; c < rvd->vdev_children; c++) | |
3019 | vdev_config_dirty(rvd->vdev_child[c]); | |
3020 | } else { | |
3021 | ASSERT(vd == vd->vdev_top); | |
3022 | ||
3023 | if (!list_link_active(&vd->vdev_config_dirty_node) && | |
3024 | !vd->vdev_ishole) | |
3025 | list_insert_head(&spa->spa_config_dirty_list, vd); | |
3026 | } | |
3027 | } | |
3028 | ||
3029 | void | |
3030 | vdev_config_clean(vdev_t *vd) | |
3031 | { | |
3032 | spa_t *spa = vd->vdev_spa; | |
3033 | ||
3034 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || | |
3035 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3036 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
3037 | ||
3038 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); | |
3039 | list_remove(&spa->spa_config_dirty_list, vd); | |
3040 | } | |
3041 | ||
3042 | /* | |
3043 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
3044 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
3045 | * the state changes from larger config changes because they require | |
3046 | * much less locking, and are often needed for administrative actions. | |
3047 | */ | |
3048 | void | |
3049 | vdev_state_dirty(vdev_t *vd) | |
3050 | { | |
3051 | spa_t *spa = vd->vdev_spa; | |
3052 | ||
3053 | ASSERT(spa_writeable(spa)); | |
3054 | ASSERT(vd == vd->vdev_top); | |
3055 | ||
3056 | /* | |
3057 | * The state list is protected by the SCL_STATE lock. The caller | |
3058 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3059 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3060 | * so this is sufficient to ensure mutual exclusion. | |
3061 | */ | |
3062 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3063 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3064 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3065 | ||
3066 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) | |
3067 | list_insert_head(&spa->spa_state_dirty_list, vd); | |
3068 | } | |
3069 | ||
3070 | void | |
3071 | vdev_state_clean(vdev_t *vd) | |
3072 | { | |
3073 | spa_t *spa = vd->vdev_spa; | |
3074 | ||
3075 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3076 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3077 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3078 | ||
3079 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3080 | list_remove(&spa->spa_state_dirty_list, vd); | |
3081 | } | |
3082 | ||
3083 | /* | |
3084 | * Propagate vdev state up from children to parent. | |
3085 | */ | |
3086 | void | |
3087 | vdev_propagate_state(vdev_t *vd) | |
3088 | { | |
3089 | spa_t *spa = vd->vdev_spa; | |
3090 | vdev_t *rvd = spa->spa_root_vdev; | |
3091 | int degraded = 0, faulted = 0; | |
3092 | int corrupted = 0; | |
3093 | vdev_t *child; | |
3094 | int c; | |
3095 | ||
3096 | if (vd->vdev_children > 0) { | |
3097 | for (c = 0; c < vd->vdev_children; c++) { | |
3098 | child = vd->vdev_child[c]; | |
3099 | ||
3100 | /* | |
3101 | * Don't factor holes into the decision. | |
3102 | */ | |
3103 | if (child->vdev_ishole) | |
3104 | continue; | |
3105 | ||
3106 | if (!vdev_readable(child) || | |
3107 | (!vdev_writeable(child) && spa_writeable(spa))) { | |
3108 | /* | |
3109 | * Root special: if there is a top-level log | |
3110 | * device, treat the root vdev as if it were | |
3111 | * degraded. | |
3112 | */ | |
3113 | if (child->vdev_islog && vd == rvd) | |
3114 | degraded++; | |
3115 | else | |
3116 | faulted++; | |
3117 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
3118 | degraded++; | |
3119 | } | |
3120 | ||
3121 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3122 | corrupted++; | |
3123 | } | |
3124 | ||
3125 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3126 | ||
3127 | /* | |
3128 | * Root special: if there is a top-level vdev that cannot be | |
3129 | * opened due to corrupted metadata, then propagate the root | |
3130 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3131 | * replicas'. | |
3132 | */ | |
3133 | if (corrupted && vd == rvd && | |
3134 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3135 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3136 | VDEV_AUX_CORRUPT_DATA); | |
3137 | } | |
3138 | ||
3139 | if (vd->vdev_parent) | |
3140 | vdev_propagate_state(vd->vdev_parent); | |
3141 | } | |
3142 | ||
3143 | /* | |
3144 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3145 | * state, because we're in the process of opening children depth-first. | |
3146 | * Otherwise, we propagate the change to the parent. | |
3147 | * | |
3148 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3149 | * generated. | |
3150 | */ | |
3151 | void | |
3152 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3153 | { | |
3154 | uint64_t save_state; | |
3155 | spa_t *spa = vd->vdev_spa; | |
3156 | ||
3157 | if (state == vd->vdev_state) { | |
3158 | vd->vdev_stat.vs_aux = aux; | |
3159 | return; | |
3160 | } | |
3161 | ||
3162 | save_state = vd->vdev_state; | |
3163 | ||
3164 | vd->vdev_state = state; | |
3165 | vd->vdev_stat.vs_aux = aux; | |
3166 | ||
3167 | /* | |
3168 | * If we are setting the vdev state to anything but an open state, then | |
3169 | * always close the underlying device unless the device has requested | |
3170 | * a delayed close (i.e. we're about to remove or fault the device). | |
3171 | * Otherwise, we keep accessible but invalid devices open forever. | |
3172 | * We don't call vdev_close() itself, because that implies some extra | |
3173 | * checks (offline, etc) that we don't want here. This is limited to | |
3174 | * leaf devices, because otherwise closing the device will affect other | |
3175 | * children. | |
3176 | */ | |
3177 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && | |
3178 | vd->vdev_ops->vdev_op_leaf) | |
3179 | vd->vdev_ops->vdev_op_close(vd); | |
3180 | ||
3181 | /* | |
3182 | * If we have brought this vdev back into service, we need | |
3183 | * to notify fmd so that it can gracefully repair any outstanding | |
3184 | * cases due to a missing device. We do this in all cases, even those | |
3185 | * that probably don't correlate to a repaired fault. This is sure to | |
3186 | * catch all cases, and we let the zfs-retire agent sort it out. If | |
3187 | * this is a transient state it's OK, as the retire agent will | |
3188 | * double-check the state of the vdev before repairing it. | |
3189 | */ | |
3190 | if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && | |
3191 | vd->vdev_prevstate != state) | |
3192 | zfs_post_state_change(spa, vd); | |
3193 | ||
3194 | if (vd->vdev_removed && | |
3195 | state == VDEV_STATE_CANT_OPEN && | |
3196 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3197 | /* | |
3198 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3199 | * device was previously marked removed and someone attempted to | |
3200 | * reopen it. If this failed due to a nonexistent device, then | |
3201 | * keep the device in the REMOVED state. We also let this be if | |
3202 | * it is one of our special test online cases, which is only | |
3203 | * attempting to online the device and shouldn't generate an FMA | |
3204 | * fault. | |
3205 | */ | |
3206 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3207 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3208 | } else if (state == VDEV_STATE_REMOVED) { | |
3209 | vd->vdev_removed = B_TRUE; | |
3210 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3211 | /* | |
3212 | * If we fail to open a vdev during an import or recovery, we | |
3213 | * mark it as "not available", which signifies that it was | |
3214 | * never there to begin with. Failure to open such a device | |
3215 | * is not considered an error. | |
3216 | */ | |
3217 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || | |
3218 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
3219 | vd->vdev_ops->vdev_op_leaf) | |
3220 | vd->vdev_not_present = 1; | |
3221 | ||
3222 | /* | |
3223 | * Post the appropriate ereport. If the 'prevstate' field is | |
3224 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3225 | * that this is part of a vdev_reopen(). In this case, we don't | |
3226 | * want to post the ereport if the device was already in the | |
3227 | * CANT_OPEN state beforehand. | |
3228 | * | |
3229 | * If the 'checkremove' flag is set, then this is an attempt to | |
3230 | * online the device in response to an insertion event. If we | |
3231 | * hit this case, then we have detected an insertion event for a | |
3232 | * faulted or offline device that wasn't in the removed state. | |
3233 | * In this scenario, we don't post an ereport because we are | |
3234 | * about to replace the device, or attempt an online with | |
3235 | * vdev_forcefault, which will generate the fault for us. | |
3236 | */ | |
3237 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3238 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
3239 | vd != spa->spa_root_vdev) { | |
3240 | const char *class; | |
3241 | ||
3242 | switch (aux) { | |
3243 | case VDEV_AUX_OPEN_FAILED: | |
3244 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3245 | break; | |
3246 | case VDEV_AUX_CORRUPT_DATA: | |
3247 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3248 | break; | |
3249 | case VDEV_AUX_NO_REPLICAS: | |
3250 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3251 | break; | |
3252 | case VDEV_AUX_BAD_GUID_SUM: | |
3253 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3254 | break; | |
3255 | case VDEV_AUX_TOO_SMALL: | |
3256 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3257 | break; | |
3258 | case VDEV_AUX_BAD_LABEL: | |
3259 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3260 | break; | |
3261 | default: | |
3262 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3263 | } | |
3264 | ||
3265 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); | |
3266 | } | |
3267 | ||
3268 | /* Erase any notion of persistent removed state */ | |
3269 | vd->vdev_removed = B_FALSE; | |
3270 | } else { | |
3271 | vd->vdev_removed = B_FALSE; | |
3272 | } | |
3273 | ||
3274 | if (!isopen && vd->vdev_parent) | |
3275 | vdev_propagate_state(vd->vdev_parent); | |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * Check the vdev configuration to ensure that it's capable of supporting | |
3280 | * a root pool. | |
3281 | */ | |
3282 | boolean_t | |
3283 | vdev_is_bootable(vdev_t *vd) | |
3284 | { | |
3285 | #if defined(__sun__) || defined(__sun) | |
3286 | /* | |
3287 | * Currently, we do not support RAID-Z or partial configuration. | |
3288 | * In addition, only a single top-level vdev is allowed and none of the | |
3289 | * leaves can be wholedisks. | |
3290 | */ | |
3291 | int c; | |
3292 | ||
3293 | if (!vd->vdev_ops->vdev_op_leaf) { | |
3294 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3295 | ||
3296 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3297 | vd->vdev_children > 1) { | |
3298 | return (B_FALSE); | |
3299 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3300 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3301 | return (B_FALSE); | |
3302 | } | |
3303 | } else if (vd->vdev_wholedisk == 1) { | |
3304 | return (B_FALSE); | |
3305 | } | |
3306 | ||
3307 | for (c = 0; c < vd->vdev_children; c++) { | |
3308 | if (!vdev_is_bootable(vd->vdev_child[c])) | |
3309 | return (B_FALSE); | |
3310 | } | |
3311 | #endif /* __sun__ || __sun */ | |
3312 | return (B_TRUE); | |
3313 | } | |
3314 | ||
3315 | /* | |
3316 | * Load the state from the original vdev tree (ovd) which | |
3317 | * we've retrieved from the MOS config object. If the original | |
3318 | * vdev was offline or faulted then we transfer that state to the | |
3319 | * device in the current vdev tree (nvd). | |
3320 | */ | |
3321 | void | |
3322 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) | |
3323 | { | |
3324 | int c; | |
3325 | ||
3326 | ASSERT(nvd->vdev_top->vdev_islog); | |
3327 | ASSERT(spa_config_held(nvd->vdev_spa, | |
3328 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
3329 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); | |
3330 | ||
3331 | for (c = 0; c < nvd->vdev_children; c++) | |
3332 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); | |
3333 | ||
3334 | if (nvd->vdev_ops->vdev_op_leaf) { | |
3335 | /* | |
3336 | * Restore the persistent vdev state | |
3337 | */ | |
3338 | nvd->vdev_offline = ovd->vdev_offline; | |
3339 | nvd->vdev_faulted = ovd->vdev_faulted; | |
3340 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3341 | nvd->vdev_removed = ovd->vdev_removed; | |
3342 | } | |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * Determine if a log device has valid content. If the vdev was | |
3347 | * removed or faulted in the MOS config then we know that | |
3348 | * the content on the log device has already been written to the pool. | |
3349 | */ | |
3350 | boolean_t | |
3351 | vdev_log_state_valid(vdev_t *vd) | |
3352 | { | |
3353 | int c; | |
3354 | ||
3355 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && | |
3356 | !vd->vdev_removed) | |
3357 | return (B_TRUE); | |
3358 | ||
3359 | for (c = 0; c < vd->vdev_children; c++) | |
3360 | if (vdev_log_state_valid(vd->vdev_child[c])) | |
3361 | return (B_TRUE); | |
3362 | ||
3363 | return (B_FALSE); | |
3364 | } | |
3365 | ||
3366 | /* | |
3367 | * Expand a vdev if possible. | |
3368 | */ | |
3369 | void | |
3370 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3371 | { | |
3372 | ASSERT(vd->vdev_top == vd); | |
3373 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3374 | ||
3375 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3376 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3377 | vdev_config_dirty(vd); | |
3378 | } | |
3379 | } | |
3380 | ||
3381 | /* | |
3382 | * Split a vdev. | |
3383 | */ | |
3384 | void | |
3385 | vdev_split(vdev_t *vd) | |
3386 | { | |
3387 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3388 | ||
3389 | vdev_remove_child(pvd, vd); | |
3390 | vdev_compact_children(pvd); | |
3391 | ||
3392 | cvd = pvd->vdev_child[0]; | |
3393 | if (pvd->vdev_children == 1) { | |
3394 | vdev_remove_parent(cvd); | |
3395 | cvd->vdev_splitting = B_TRUE; | |
3396 | } | |
3397 | vdev_propagate_state(cvd); | |
3398 | } | |
3399 | ||
3400 | void | |
3401 | vdev_deadman(vdev_t *vd) | |
3402 | { | |
3403 | int c; | |
3404 | ||
3405 | for (c = 0; c < vd->vdev_children; c++) { | |
3406 | vdev_t *cvd = vd->vdev_child[c]; | |
3407 | ||
3408 | vdev_deadman(cvd); | |
3409 | } | |
3410 | ||
3411 | if (vd->vdev_ops->vdev_op_leaf) { | |
3412 | vdev_queue_t *vq = &vd->vdev_queue; | |
3413 | ||
3414 | mutex_enter(&vq->vq_lock); | |
3415 | if (avl_numnodes(&vq->vq_active_tree) > 0) { | |
3416 | spa_t *spa = vd->vdev_spa; | |
3417 | zio_t *fio; | |
3418 | uint64_t delta; | |
3419 | ||
3420 | /* | |
3421 | * Look at the head of all the pending queues, | |
3422 | * if any I/O has been outstanding for longer than | |
3423 | * the spa_deadman_synctime we log a zevent. | |
3424 | */ | |
3425 | fio = avl_first(&vq->vq_active_tree); | |
3426 | delta = gethrtime() - fio->io_timestamp; | |
3427 | if (delta > spa_deadman_synctime(spa)) { | |
3428 | zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " | |
3429 | "delta %lluns, last io %lluns", | |
3430 | fio->io_timestamp, delta, | |
3431 | vq->vq_io_complete_ts); | |
3432 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, | |
3433 | spa, vd, fio, 0, 0); | |
3434 | } | |
3435 | } | |
3436 | mutex_exit(&vq->vq_lock); | |
3437 | } | |
3438 | } | |
3439 | ||
3440 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
3441 | EXPORT_SYMBOL(vdev_fault); | |
3442 | EXPORT_SYMBOL(vdev_degrade); | |
3443 | EXPORT_SYMBOL(vdev_online); | |
3444 | EXPORT_SYMBOL(vdev_offline); | |
3445 | EXPORT_SYMBOL(vdev_clear); | |
3446 | ||
3447 | module_param(metaslabs_per_vdev, int, 0644); | |
3448 | MODULE_PARM_DESC(metaslabs_per_vdev, | |
3449 | "Divide added vdev into approximately (but no more than) this number " | |
3450 | "of metaslabs"); | |
3451 | #endif |