Print this page
4101 metaslab_debug should allow for fine-grained control
4102 space_maps should store more information about themselves
4103 space map object blocksize should be increased
4104 ::spa_space no longer works
4105 removing a mirrored log device results in a leaked object
4106 asynchronously load metaslab
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Sebastien Roy <seb@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/vdev.c
+++ new/usr/src/uts/common/fs/zfs/vdev.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 26 */
27 27
28 28 #include <sys/zfs_context.h>
|
↓ open down ↓ |
28 lines elided |
↑ open up ↑ |
29 29 #include <sys/fm/fs/zfs.h>
30 30 #include <sys/spa.h>
31 31 #include <sys/spa_impl.h>
32 32 #include <sys/dmu.h>
33 33 #include <sys/dmu_tx.h>
34 34 #include <sys/vdev_impl.h>
35 35 #include <sys/uberblock_impl.h>
36 36 #include <sys/metaslab.h>
37 37 #include <sys/metaslab_impl.h>
38 38 #include <sys/space_map.h>
39 +#include <sys/space_reftree.h>
39 40 #include <sys/zio.h>
40 41 #include <sys/zap.h>
41 42 #include <sys/fs/zfs.h>
42 43 #include <sys/arc.h>
43 44 #include <sys/zil.h>
44 45 #include <sys/dsl_scan.h>
45 46
46 47 /*
47 48 * Virtual device management.
48 49 */
49 50
50 51 static vdev_ops_t *vdev_ops_table[] = {
51 52 &vdev_root_ops,
52 53 &vdev_raidz_ops,
53 54 &vdev_mirror_ops,
54 55 &vdev_replacing_ops,
55 56 &vdev_spare_ops,
56 57 &vdev_disk_ops,
57 58 &vdev_file_ops,
58 59 &vdev_missing_ops,
59 60 &vdev_hole_ops,
60 61 NULL
61 62 };
62 63
63 64 /* maximum scrub/resilver I/O queue per leaf vdev */
64 65 int zfs_scrub_limit = 10;
65 66
66 67 /*
67 68 * Given a vdev type, return the appropriate ops vector.
68 69 */
69 70 static vdev_ops_t *
70 71 vdev_getops(const char *type)
71 72 {
72 73 vdev_ops_t *ops, **opspp;
73 74
74 75 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
75 76 if (strcmp(ops->vdev_op_type, type) == 0)
76 77 break;
77 78
78 79 return (ops);
79 80 }
80 81
81 82 /*
82 83 * Default asize function: return the MAX of psize with the asize of
83 84 * all children. This is what's used by anything other than RAID-Z.
84 85 */
85 86 uint64_t
86 87 vdev_default_asize(vdev_t *vd, uint64_t psize)
87 88 {
88 89 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
89 90 uint64_t csize;
90 91
91 92 for (int c = 0; c < vd->vdev_children; c++) {
92 93 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
93 94 asize = MAX(asize, csize);
94 95 }
95 96
96 97 return (asize);
97 98 }
98 99
99 100 /*
100 101 * Get the minimum allocatable size. We define the allocatable size as
101 102 * the vdev's asize rounded to the nearest metaslab. This allows us to
102 103 * replace or attach devices which don't have the same physical size but
103 104 * can still satisfy the same number of allocations.
104 105 */
105 106 uint64_t
106 107 vdev_get_min_asize(vdev_t *vd)
107 108 {
108 109 vdev_t *pvd = vd->vdev_parent;
109 110
110 111 /*
111 112 * If our parent is NULL (inactive spare or cache) or is the root,
112 113 * just return our own asize.
113 114 */
114 115 if (pvd == NULL)
115 116 return (vd->vdev_asize);
116 117
117 118 /*
118 119 * The top-level vdev just returns the allocatable size rounded
119 120 * to the nearest metaslab.
120 121 */
121 122 if (vd == vd->vdev_top)
122 123 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
123 124
124 125 /*
125 126 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
126 127 * so each child must provide at least 1/Nth of its asize.
127 128 */
128 129 if (pvd->vdev_ops == &vdev_raidz_ops)
129 130 return (pvd->vdev_min_asize / pvd->vdev_children);
130 131
131 132 return (pvd->vdev_min_asize);
132 133 }
133 134
134 135 void
135 136 vdev_set_min_asize(vdev_t *vd)
136 137 {
137 138 vd->vdev_min_asize = vdev_get_min_asize(vd);
138 139
139 140 for (int c = 0; c < vd->vdev_children; c++)
140 141 vdev_set_min_asize(vd->vdev_child[c]);
141 142 }
142 143
143 144 vdev_t *
144 145 vdev_lookup_top(spa_t *spa, uint64_t vdev)
145 146 {
146 147 vdev_t *rvd = spa->spa_root_vdev;
147 148
148 149 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
149 150
150 151 if (vdev < rvd->vdev_children) {
151 152 ASSERT(rvd->vdev_child[vdev] != NULL);
152 153 return (rvd->vdev_child[vdev]);
153 154 }
154 155
155 156 return (NULL);
156 157 }
157 158
158 159 vdev_t *
159 160 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
160 161 {
161 162 vdev_t *mvd;
162 163
163 164 if (vd->vdev_guid == guid)
164 165 return (vd);
165 166
166 167 for (int c = 0; c < vd->vdev_children; c++)
167 168 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
168 169 NULL)
169 170 return (mvd);
170 171
171 172 return (NULL);
172 173 }
173 174
174 175 void
175 176 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
176 177 {
177 178 size_t oldsize, newsize;
178 179 uint64_t id = cvd->vdev_id;
179 180 vdev_t **newchild;
180 181
181 182 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
182 183 ASSERT(cvd->vdev_parent == NULL);
183 184
184 185 cvd->vdev_parent = pvd;
185 186
186 187 if (pvd == NULL)
187 188 return;
188 189
189 190 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
190 191
191 192 oldsize = pvd->vdev_children * sizeof (vdev_t *);
192 193 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
193 194 newsize = pvd->vdev_children * sizeof (vdev_t *);
194 195
195 196 newchild = kmem_zalloc(newsize, KM_SLEEP);
196 197 if (pvd->vdev_child != NULL) {
197 198 bcopy(pvd->vdev_child, newchild, oldsize);
198 199 kmem_free(pvd->vdev_child, oldsize);
199 200 }
200 201
201 202 pvd->vdev_child = newchild;
202 203 pvd->vdev_child[id] = cvd;
203 204
204 205 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
205 206 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
206 207
207 208 /*
208 209 * Walk up all ancestors to update guid sum.
209 210 */
210 211 for (; pvd != NULL; pvd = pvd->vdev_parent)
211 212 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
212 213 }
213 214
214 215 void
215 216 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
216 217 {
217 218 int c;
218 219 uint_t id = cvd->vdev_id;
219 220
220 221 ASSERT(cvd->vdev_parent == pvd);
221 222
222 223 if (pvd == NULL)
223 224 return;
224 225
225 226 ASSERT(id < pvd->vdev_children);
226 227 ASSERT(pvd->vdev_child[id] == cvd);
227 228
228 229 pvd->vdev_child[id] = NULL;
229 230 cvd->vdev_parent = NULL;
230 231
231 232 for (c = 0; c < pvd->vdev_children; c++)
232 233 if (pvd->vdev_child[c])
233 234 break;
234 235
235 236 if (c == pvd->vdev_children) {
236 237 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
237 238 pvd->vdev_child = NULL;
238 239 pvd->vdev_children = 0;
239 240 }
240 241
241 242 /*
242 243 * Walk up all ancestors to update guid sum.
243 244 */
244 245 for (; pvd != NULL; pvd = pvd->vdev_parent)
245 246 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
246 247 }
247 248
248 249 /*
249 250 * Remove any holes in the child array.
250 251 */
251 252 void
252 253 vdev_compact_children(vdev_t *pvd)
253 254 {
254 255 vdev_t **newchild, *cvd;
255 256 int oldc = pvd->vdev_children;
256 257 int newc;
257 258
258 259 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
259 260
260 261 for (int c = newc = 0; c < oldc; c++)
261 262 if (pvd->vdev_child[c])
262 263 newc++;
263 264
264 265 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
265 266
266 267 for (int c = newc = 0; c < oldc; c++) {
267 268 if ((cvd = pvd->vdev_child[c]) != NULL) {
268 269 newchild[newc] = cvd;
269 270 cvd->vdev_id = newc++;
270 271 }
271 272 }
272 273
273 274 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
274 275 pvd->vdev_child = newchild;
275 276 pvd->vdev_children = newc;
276 277 }
277 278
278 279 /*
279 280 * Allocate and minimally initialize a vdev_t.
280 281 */
281 282 vdev_t *
282 283 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
283 284 {
284 285 vdev_t *vd;
285 286
286 287 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
287 288
288 289 if (spa->spa_root_vdev == NULL) {
289 290 ASSERT(ops == &vdev_root_ops);
290 291 spa->spa_root_vdev = vd;
291 292 spa->spa_load_guid = spa_generate_guid(NULL);
292 293 }
293 294
294 295 if (guid == 0 && ops != &vdev_hole_ops) {
295 296 if (spa->spa_root_vdev == vd) {
296 297 /*
297 298 * The root vdev's guid will also be the pool guid,
298 299 * which must be unique among all pools.
299 300 */
300 301 guid = spa_generate_guid(NULL);
301 302 } else {
302 303 /*
303 304 * Any other vdev's guid must be unique within the pool.
304 305 */
305 306 guid = spa_generate_guid(spa);
306 307 }
307 308 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
308 309 }
309 310
310 311 vd->vdev_spa = spa;
311 312 vd->vdev_id = id;
|
↓ open down ↓ |
263 lines elided |
↑ open up ↑ |
312 313 vd->vdev_guid = guid;
313 314 vd->vdev_guid_sum = guid;
314 315 vd->vdev_ops = ops;
315 316 vd->vdev_state = VDEV_STATE_CLOSED;
316 317 vd->vdev_ishole = (ops == &vdev_hole_ops);
317 318
318 319 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
319 320 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
320 321 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
321 322 for (int t = 0; t < DTL_TYPES; t++) {
322 - space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
323 + vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
323 324 &vd->vdev_dtl_lock);
324 325 }
325 326 txg_list_create(&vd->vdev_ms_list,
326 327 offsetof(struct metaslab, ms_txg_node));
327 328 txg_list_create(&vd->vdev_dtl_list,
328 329 offsetof(struct vdev, vdev_dtl_node));
329 330 vd->vdev_stat.vs_timestamp = gethrtime();
330 331 vdev_queue_init(vd);
331 332 vdev_cache_init(vd);
332 333
333 334 return (vd);
334 335 }
335 336
336 337 /*
337 338 * Allocate a new vdev. The 'alloctype' is used to control whether we are
338 339 * creating a new vdev or loading an existing one - the behavior is slightly
339 340 * different for each case.
340 341 */
341 342 int
342 343 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
343 344 int alloctype)
344 345 {
345 346 vdev_ops_t *ops;
346 347 char *type;
347 348 uint64_t guid = 0, islog, nparity;
348 349 vdev_t *vd;
349 350
350 351 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
351 352
352 353 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
353 354 return (SET_ERROR(EINVAL));
354 355
355 356 if ((ops = vdev_getops(type)) == NULL)
356 357 return (SET_ERROR(EINVAL));
357 358
358 359 /*
359 360 * If this is a load, get the vdev guid from the nvlist.
360 361 * Otherwise, vdev_alloc_common() will generate one for us.
361 362 */
362 363 if (alloctype == VDEV_ALLOC_LOAD) {
363 364 uint64_t label_id;
364 365
365 366 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
366 367 label_id != id)
367 368 return (SET_ERROR(EINVAL));
368 369
369 370 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
370 371 return (SET_ERROR(EINVAL));
371 372 } else if (alloctype == VDEV_ALLOC_SPARE) {
372 373 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
373 374 return (SET_ERROR(EINVAL));
374 375 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
375 376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
376 377 return (SET_ERROR(EINVAL));
377 378 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
378 379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 380 return (SET_ERROR(EINVAL));
380 381 }
381 382
382 383 /*
383 384 * The first allocated vdev must be of type 'root'.
384 385 */
385 386 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
386 387 return (SET_ERROR(EINVAL));
387 388
388 389 /*
389 390 * Determine whether we're a log vdev.
390 391 */
391 392 islog = 0;
392 393 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
393 394 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
394 395 return (SET_ERROR(ENOTSUP));
395 396
396 397 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
397 398 return (SET_ERROR(ENOTSUP));
398 399
399 400 /*
400 401 * Set the nparity property for RAID-Z vdevs.
401 402 */
402 403 nparity = -1ULL;
403 404 if (ops == &vdev_raidz_ops) {
404 405 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
405 406 &nparity) == 0) {
406 407 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
407 408 return (SET_ERROR(EINVAL));
408 409 /*
409 410 * Previous versions could only support 1 or 2 parity
410 411 * device.
411 412 */
412 413 if (nparity > 1 &&
413 414 spa_version(spa) < SPA_VERSION_RAIDZ2)
414 415 return (SET_ERROR(ENOTSUP));
415 416 if (nparity > 2 &&
416 417 spa_version(spa) < SPA_VERSION_RAIDZ3)
417 418 return (SET_ERROR(ENOTSUP));
418 419 } else {
419 420 /*
420 421 * We require the parity to be specified for SPAs that
421 422 * support multiple parity levels.
422 423 */
423 424 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
424 425 return (SET_ERROR(EINVAL));
425 426 /*
426 427 * Otherwise, we default to 1 parity device for RAID-Z.
427 428 */
428 429 nparity = 1;
429 430 }
430 431 } else {
431 432 nparity = 0;
432 433 }
433 434 ASSERT(nparity != -1ULL);
434 435
435 436 vd = vdev_alloc_common(spa, id, guid, ops);
436 437
437 438 vd->vdev_islog = islog;
438 439 vd->vdev_nparity = nparity;
439 440
440 441 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
441 442 vd->vdev_path = spa_strdup(vd->vdev_path);
442 443 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
443 444 vd->vdev_devid = spa_strdup(vd->vdev_devid);
444 445 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
445 446 &vd->vdev_physpath) == 0)
446 447 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
447 448 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
448 449 vd->vdev_fru = spa_strdup(vd->vdev_fru);
449 450
450 451 /*
451 452 * Set the whole_disk property. If it's not specified, leave the value
452 453 * as -1.
453 454 */
454 455 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
455 456 &vd->vdev_wholedisk) != 0)
456 457 vd->vdev_wholedisk = -1ULL;
457 458
458 459 /*
459 460 * Look for the 'not present' flag. This will only be set if the device
460 461 * was not present at the time of import.
461 462 */
462 463 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
463 464 &vd->vdev_not_present);
464 465
465 466 /*
466 467 * Get the alignment requirement.
467 468 */
468 469 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
469 470
470 471 /*
471 472 * Retrieve the vdev creation time.
472 473 */
473 474 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
474 475 &vd->vdev_crtxg);
475 476
476 477 /*
477 478 * If we're a top-level vdev, try to load the allocation parameters.
478 479 */
479 480 if (parent && !parent->vdev_parent &&
480 481 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
481 482 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
482 483 &vd->vdev_ms_array);
483 484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
484 485 &vd->vdev_ms_shift);
485 486 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
486 487 &vd->vdev_asize);
487 488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
488 489 &vd->vdev_removing);
489 490 }
490 491
491 492 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
492 493 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
493 494 alloctype == VDEV_ALLOC_ADD ||
494 495 alloctype == VDEV_ALLOC_SPLIT ||
495 496 alloctype == VDEV_ALLOC_ROOTPOOL);
496 497 vd->vdev_mg = metaslab_group_create(islog ?
497 498 spa_log_class(spa) : spa_normal_class(spa), vd);
|
↓ open down ↓ |
165 lines elided |
↑ open up ↑ |
498 499 }
499 500
500 501 /*
501 502 * If we're a leaf vdev, try to load the DTL object and other state.
502 503 */
503 504 if (vd->vdev_ops->vdev_op_leaf &&
504 505 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
505 506 alloctype == VDEV_ALLOC_ROOTPOOL)) {
506 507 if (alloctype == VDEV_ALLOC_LOAD) {
507 508 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
508 - &vd->vdev_dtl_smo.smo_object);
509 + &vd->vdev_dtl_object);
509 510 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
510 511 &vd->vdev_unspare);
511 512 }
512 513
513 514 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
514 515 uint64_t spare = 0;
515 516
516 517 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
517 518 &spare) == 0 && spare)
518 519 spa_spare_add(vd);
519 520 }
520 521
521 522 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
522 523 &vd->vdev_offline);
523 524
524 525 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
525 526 &vd->vdev_resilver_txg);
526 527
527 528 /*
528 529 * When importing a pool, we want to ignore the persistent fault
529 530 * state, as the diagnosis made on another system may not be
530 531 * valid in the current context. Local vdevs will
531 532 * remain in the faulted state.
532 533 */
533 534 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
534 535 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
535 536 &vd->vdev_faulted);
536 537 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
537 538 &vd->vdev_degraded);
538 539 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
539 540 &vd->vdev_removed);
540 541
541 542 if (vd->vdev_faulted || vd->vdev_degraded) {
542 543 char *aux;
543 544
544 545 vd->vdev_label_aux =
545 546 VDEV_AUX_ERR_EXCEEDED;
546 547 if (nvlist_lookup_string(nv,
547 548 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
548 549 strcmp(aux, "external") == 0)
549 550 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
550 551 }
551 552 }
552 553 }
553 554
554 555 /*
555 556 * Add ourselves to the parent's list of children.
556 557 */
557 558 vdev_add_child(parent, vd);
558 559
559 560 *vdp = vd;
560 561
561 562 return (0);
562 563 }
563 564
564 565 void
565 566 vdev_free(vdev_t *vd)
566 567 {
567 568 spa_t *spa = vd->vdev_spa;
568 569
569 570 /*
570 571 * vdev_free() implies closing the vdev first. This is simpler than
571 572 * trying to ensure complicated semantics for all callers.
572 573 */
573 574 vdev_close(vd);
574 575
575 576 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
576 577 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
577 578
578 579 /*
579 580 * Free all children.
580 581 */
581 582 for (int c = 0; c < vd->vdev_children; c++)
582 583 vdev_free(vd->vdev_child[c]);
583 584
584 585 ASSERT(vd->vdev_child == NULL);
585 586 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
586 587
587 588 /*
588 589 * Discard allocation state.
589 590 */
590 591 if (vd->vdev_mg != NULL) {
591 592 vdev_metaslab_fini(vd);
592 593 metaslab_group_destroy(vd->vdev_mg);
593 594 }
594 595
595 596 ASSERT0(vd->vdev_stat.vs_space);
596 597 ASSERT0(vd->vdev_stat.vs_dspace);
597 598 ASSERT0(vd->vdev_stat.vs_alloc);
598 599
599 600 /*
600 601 * Remove this vdev from its parent's child list.
601 602 */
602 603 vdev_remove_child(vd->vdev_parent, vd);
603 604
604 605 ASSERT(vd->vdev_parent == NULL);
605 606
606 607 /*
607 608 * Clean up vdev structure.
608 609 */
609 610 vdev_queue_fini(vd);
610 611 vdev_cache_fini(vd);
611 612
612 613 if (vd->vdev_path)
613 614 spa_strfree(vd->vdev_path);
614 615 if (vd->vdev_devid)
615 616 spa_strfree(vd->vdev_devid);
616 617 if (vd->vdev_physpath)
617 618 spa_strfree(vd->vdev_physpath);
618 619 if (vd->vdev_fru)
619 620 spa_strfree(vd->vdev_fru);
|
↓ open down ↓ |
101 lines elided |
↑ open up ↑ |
620 621
621 622 if (vd->vdev_isspare)
622 623 spa_spare_remove(vd);
623 624 if (vd->vdev_isl2cache)
624 625 spa_l2cache_remove(vd);
625 626
626 627 txg_list_destroy(&vd->vdev_ms_list);
627 628 txg_list_destroy(&vd->vdev_dtl_list);
628 629
629 630 mutex_enter(&vd->vdev_dtl_lock);
631 + space_map_close(vd->vdev_dtl_sm);
630 632 for (int t = 0; t < DTL_TYPES; t++) {
631 - space_map_unload(&vd->vdev_dtl[t]);
632 - space_map_destroy(&vd->vdev_dtl[t]);
633 + range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
634 + range_tree_destroy(vd->vdev_dtl[t]);
633 635 }
634 636 mutex_exit(&vd->vdev_dtl_lock);
635 637
636 638 mutex_destroy(&vd->vdev_dtl_lock);
637 639 mutex_destroy(&vd->vdev_stat_lock);
638 640 mutex_destroy(&vd->vdev_probe_lock);
639 641
640 642 if (vd == spa->spa_root_vdev)
641 643 spa->spa_root_vdev = NULL;
642 644
643 645 kmem_free(vd, sizeof (vdev_t));
644 646 }
645 647
646 648 /*
647 649 * Transfer top-level vdev state from svd to tvd.
648 650 */
649 651 static void
650 652 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
651 653 {
652 654 spa_t *spa = svd->vdev_spa;
653 655 metaslab_t *msp;
654 656 vdev_t *vd;
655 657 int t;
656 658
657 659 ASSERT(tvd == tvd->vdev_top);
658 660
659 661 tvd->vdev_ms_array = svd->vdev_ms_array;
660 662 tvd->vdev_ms_shift = svd->vdev_ms_shift;
661 663 tvd->vdev_ms_count = svd->vdev_ms_count;
662 664
663 665 svd->vdev_ms_array = 0;
664 666 svd->vdev_ms_shift = 0;
665 667 svd->vdev_ms_count = 0;
666 668
667 669 if (tvd->vdev_mg)
668 670 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
669 671 tvd->vdev_mg = svd->vdev_mg;
670 672 tvd->vdev_ms = svd->vdev_ms;
671 673
672 674 svd->vdev_mg = NULL;
673 675 svd->vdev_ms = NULL;
674 676
675 677 if (tvd->vdev_mg != NULL)
676 678 tvd->vdev_mg->mg_vd = tvd;
677 679
678 680 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
679 681 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
680 682 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
681 683
682 684 svd->vdev_stat.vs_alloc = 0;
683 685 svd->vdev_stat.vs_space = 0;
684 686 svd->vdev_stat.vs_dspace = 0;
685 687
686 688 for (t = 0; t < TXG_SIZE; t++) {
687 689 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
688 690 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
689 691 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
690 692 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
691 693 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
692 694 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
693 695 }
694 696
695 697 if (list_link_active(&svd->vdev_config_dirty_node)) {
696 698 vdev_config_clean(svd);
697 699 vdev_config_dirty(tvd);
698 700 }
699 701
700 702 if (list_link_active(&svd->vdev_state_dirty_node)) {
701 703 vdev_state_clean(svd);
702 704 vdev_state_dirty(tvd);
703 705 }
704 706
705 707 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
706 708 svd->vdev_deflate_ratio = 0;
707 709
708 710 tvd->vdev_islog = svd->vdev_islog;
709 711 svd->vdev_islog = 0;
710 712 }
711 713
712 714 static void
713 715 vdev_top_update(vdev_t *tvd, vdev_t *vd)
714 716 {
715 717 if (vd == NULL)
716 718 return;
717 719
718 720 vd->vdev_top = tvd;
719 721
720 722 for (int c = 0; c < vd->vdev_children; c++)
721 723 vdev_top_update(tvd, vd->vdev_child[c]);
722 724 }
723 725
724 726 /*
725 727 * Add a mirror/replacing vdev above an existing vdev.
726 728 */
727 729 vdev_t *
728 730 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
729 731 {
730 732 spa_t *spa = cvd->vdev_spa;
731 733 vdev_t *pvd = cvd->vdev_parent;
732 734 vdev_t *mvd;
733 735
734 736 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
735 737
736 738 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
737 739
738 740 mvd->vdev_asize = cvd->vdev_asize;
739 741 mvd->vdev_min_asize = cvd->vdev_min_asize;
740 742 mvd->vdev_max_asize = cvd->vdev_max_asize;
741 743 mvd->vdev_ashift = cvd->vdev_ashift;
742 744 mvd->vdev_state = cvd->vdev_state;
743 745 mvd->vdev_crtxg = cvd->vdev_crtxg;
744 746
745 747 vdev_remove_child(pvd, cvd);
746 748 vdev_add_child(pvd, mvd);
747 749 cvd->vdev_id = mvd->vdev_children;
748 750 vdev_add_child(mvd, cvd);
749 751 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
750 752
751 753 if (mvd == mvd->vdev_top)
752 754 vdev_top_transfer(cvd, mvd);
753 755
754 756 return (mvd);
755 757 }
756 758
757 759 /*
758 760 * Remove a 1-way mirror/replacing vdev from the tree.
759 761 */
760 762 void
761 763 vdev_remove_parent(vdev_t *cvd)
762 764 {
763 765 vdev_t *mvd = cvd->vdev_parent;
764 766 vdev_t *pvd = mvd->vdev_parent;
765 767
766 768 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
767 769
768 770 ASSERT(mvd->vdev_children == 1);
769 771 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
770 772 mvd->vdev_ops == &vdev_replacing_ops ||
771 773 mvd->vdev_ops == &vdev_spare_ops);
772 774 cvd->vdev_ashift = mvd->vdev_ashift;
773 775
774 776 vdev_remove_child(mvd, cvd);
775 777 vdev_remove_child(pvd, mvd);
776 778
777 779 /*
778 780 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
779 781 * Otherwise, we could have detached an offline device, and when we
780 782 * go to import the pool we'll think we have two top-level vdevs,
781 783 * instead of a different version of the same top-level vdev.
782 784 */
783 785 if (mvd->vdev_top == mvd) {
784 786 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
785 787 cvd->vdev_orig_guid = cvd->vdev_guid;
786 788 cvd->vdev_guid += guid_delta;
787 789 cvd->vdev_guid_sum += guid_delta;
788 790 }
789 791 cvd->vdev_id = mvd->vdev_id;
790 792 vdev_add_child(pvd, cvd);
791 793 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
792 794
793 795 if (cvd == cvd->vdev_top)
794 796 vdev_top_transfer(mvd, cvd);
795 797
796 798 ASSERT(mvd->vdev_children == 0);
797 799 vdev_free(mvd);
798 800 }
799 801
800 802 int
801 803 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
802 804 {
803 805 spa_t *spa = vd->vdev_spa;
804 806 objset_t *mos = spa->spa_meta_objset;
805 807 uint64_t m;
806 808 uint64_t oldc = vd->vdev_ms_count;
807 809 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
808 810 metaslab_t **mspp;
809 811 int error;
810 812
811 813 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
812 814
813 815 /*
814 816 * This vdev is not being allocated from yet or is a hole.
815 817 */
816 818 if (vd->vdev_ms_shift == 0)
817 819 return (0);
818 820
819 821 ASSERT(!vd->vdev_ishole);
820 822
821 823 /*
822 824 * Compute the raidz-deflation ratio. Note, we hard-code
823 825 * in 128k (1 << 17) because it is the current "typical" blocksize.
824 826 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
825 827 * or we will inconsistently account for existing bp's.
826 828 */
827 829 vd->vdev_deflate_ratio = (1 << 17) /
828 830 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
829 831
830 832 ASSERT(oldc <= newc);
831 833
832 834 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
|
↓ open down ↓ |
190 lines elided |
↑ open up ↑ |
833 835
834 836 if (oldc != 0) {
835 837 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
836 838 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
837 839 }
838 840
839 841 vd->vdev_ms = mspp;
840 842 vd->vdev_ms_count = newc;
841 843
842 844 for (m = oldc; m < newc; m++) {
843 - space_map_obj_t smo = { 0, 0, 0 };
845 + uint64_t object = 0;
846 +
844 847 if (txg == 0) {
845 - uint64_t object = 0;
846 848 error = dmu_read(mos, vd->vdev_ms_array,
847 849 m * sizeof (uint64_t), sizeof (uint64_t), &object,
848 850 DMU_READ_PREFETCH);
849 851 if (error)
850 852 return (error);
851 - if (object != 0) {
852 - dmu_buf_t *db;
853 - error = dmu_bonus_hold(mos, object, FTAG, &db);
854 - if (error)
855 - return (error);
856 - ASSERT3U(db->db_size, >=, sizeof (smo));
857 - bcopy(db->db_data, &smo, sizeof (smo));
858 - ASSERT3U(smo.smo_object, ==, object);
859 - dmu_buf_rele(db, FTAG);
860 - }
861 853 }
862 - vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
863 - m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
854 + vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg);
864 855 }
865 856
866 857 if (txg == 0)
867 858 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
868 859
869 860 /*
870 861 * If the vdev is being removed we don't activate
871 862 * the metaslabs since we want to ensure that no new
872 863 * allocations are performed on this device.
873 864 */
874 865 if (oldc == 0 && !vd->vdev_removing)
875 866 metaslab_group_activate(vd->vdev_mg);
876 867
877 868 if (txg == 0)
878 869 spa_config_exit(spa, SCL_ALLOC, FTAG);
879 870
880 871 return (0);
|
↓ open down ↓ |
7 lines elided |
↑ open up ↑ |
881 872 }
882 873
883 874 void
884 875 vdev_metaslab_fini(vdev_t *vd)
885 876 {
886 877 uint64_t m;
887 878 uint64_t count = vd->vdev_ms_count;
888 879
889 880 if (vd->vdev_ms != NULL) {
890 881 metaslab_group_passivate(vd->vdev_mg);
891 - for (m = 0; m < count; m++)
892 - if (vd->vdev_ms[m] != NULL)
893 - metaslab_fini(vd->vdev_ms[m]);
882 + for (m = 0; m < count; m++) {
883 + metaslab_t *msp = vd->vdev_ms[m];
884 +
885 + if (msp != NULL)
886 + metaslab_fini(msp);
887 + }
894 888 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
895 889 vd->vdev_ms = NULL;
896 890 }
897 891 }
898 892
899 893 typedef struct vdev_probe_stats {
900 894 boolean_t vps_readable;
901 895 boolean_t vps_writeable;
902 896 int vps_flags;
903 897 } vdev_probe_stats_t;
904 898
905 899 static void
906 900 vdev_probe_done(zio_t *zio)
907 901 {
908 902 spa_t *spa = zio->io_spa;
909 903 vdev_t *vd = zio->io_vd;
910 904 vdev_probe_stats_t *vps = zio->io_private;
911 905
912 906 ASSERT(vd->vdev_probe_zio != NULL);
913 907
914 908 if (zio->io_type == ZIO_TYPE_READ) {
915 909 if (zio->io_error == 0)
916 910 vps->vps_readable = 1;
917 911 if (zio->io_error == 0 && spa_writeable(spa)) {
918 912 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
919 913 zio->io_offset, zio->io_size, zio->io_data,
920 914 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
921 915 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
922 916 } else {
923 917 zio_buf_free(zio->io_data, zio->io_size);
924 918 }
925 919 } else if (zio->io_type == ZIO_TYPE_WRITE) {
926 920 if (zio->io_error == 0)
927 921 vps->vps_writeable = 1;
928 922 zio_buf_free(zio->io_data, zio->io_size);
929 923 } else if (zio->io_type == ZIO_TYPE_NULL) {
930 924 zio_t *pio;
931 925
932 926 vd->vdev_cant_read |= !vps->vps_readable;
933 927 vd->vdev_cant_write |= !vps->vps_writeable;
934 928
935 929 if (vdev_readable(vd) &&
936 930 (vdev_writeable(vd) || !spa_writeable(spa))) {
937 931 zio->io_error = 0;
938 932 } else {
939 933 ASSERT(zio->io_error != 0);
940 934 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
941 935 spa, vd, NULL, 0, 0);
942 936 zio->io_error = SET_ERROR(ENXIO);
943 937 }
944 938
945 939 mutex_enter(&vd->vdev_probe_lock);
946 940 ASSERT(vd->vdev_probe_zio == zio);
947 941 vd->vdev_probe_zio = NULL;
948 942 mutex_exit(&vd->vdev_probe_lock);
949 943
950 944 while ((pio = zio_walk_parents(zio)) != NULL)
951 945 if (!vdev_accessible(vd, pio))
952 946 pio->io_error = SET_ERROR(ENXIO);
953 947
954 948 kmem_free(vps, sizeof (*vps));
955 949 }
956 950 }
957 951
958 952 /*
959 953 * Determine whether this device is accessible.
960 954 *
961 955 * Read and write to several known locations: the pad regions of each
962 956 * vdev label but the first, which we leave alone in case it contains
963 957 * a VTOC.
964 958 */
965 959 zio_t *
966 960 vdev_probe(vdev_t *vd, zio_t *zio)
967 961 {
968 962 spa_t *spa = vd->vdev_spa;
969 963 vdev_probe_stats_t *vps = NULL;
970 964 zio_t *pio;
971 965
972 966 ASSERT(vd->vdev_ops->vdev_op_leaf);
973 967
974 968 /*
975 969 * Don't probe the probe.
976 970 */
977 971 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
978 972 return (NULL);
979 973
980 974 /*
981 975 * To prevent 'probe storms' when a device fails, we create
982 976 * just one probe i/o at a time. All zios that want to probe
983 977 * this vdev will become parents of the probe io.
984 978 */
985 979 mutex_enter(&vd->vdev_probe_lock);
986 980
987 981 if ((pio = vd->vdev_probe_zio) == NULL) {
988 982 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
989 983
990 984 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
991 985 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
992 986 ZIO_FLAG_TRYHARD;
993 987
994 988 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
995 989 /*
996 990 * vdev_cant_read and vdev_cant_write can only
997 991 * transition from TRUE to FALSE when we have the
998 992 * SCL_ZIO lock as writer; otherwise they can only
999 993 * transition from FALSE to TRUE. This ensures that
1000 994 * any zio looking at these values can assume that
1001 995 * failures persist for the life of the I/O. That's
1002 996 * important because when a device has intermittent
1003 997 * connectivity problems, we want to ensure that
1004 998 * they're ascribed to the device (ENXIO) and not
1005 999 * the zio (EIO).
1006 1000 *
1007 1001 * Since we hold SCL_ZIO as writer here, clear both
1008 1002 * values so the probe can reevaluate from first
1009 1003 * principles.
1010 1004 */
1011 1005 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1012 1006 vd->vdev_cant_read = B_FALSE;
1013 1007 vd->vdev_cant_write = B_FALSE;
1014 1008 }
1015 1009
1016 1010 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1017 1011 vdev_probe_done, vps,
1018 1012 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1019 1013
1020 1014 /*
1021 1015 * We can't change the vdev state in this context, so we
1022 1016 * kick off an async task to do it on our behalf.
1023 1017 */
1024 1018 if (zio != NULL) {
1025 1019 vd->vdev_probe_wanted = B_TRUE;
1026 1020 spa_async_request(spa, SPA_ASYNC_PROBE);
1027 1021 }
1028 1022 }
1029 1023
1030 1024 if (zio != NULL)
1031 1025 zio_add_child(zio, pio);
1032 1026
1033 1027 mutex_exit(&vd->vdev_probe_lock);
1034 1028
1035 1029 if (vps == NULL) {
1036 1030 ASSERT(zio != NULL);
1037 1031 return (NULL);
1038 1032 }
1039 1033
1040 1034 for (int l = 1; l < VDEV_LABELS; l++) {
1041 1035 zio_nowait(zio_read_phys(pio, vd,
1042 1036 vdev_label_offset(vd->vdev_psize, l,
1043 1037 offsetof(vdev_label_t, vl_pad2)),
1044 1038 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1045 1039 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1046 1040 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1047 1041 }
1048 1042
1049 1043 if (zio == NULL)
1050 1044 return (pio);
1051 1045
1052 1046 zio_nowait(pio);
1053 1047 return (NULL);
1054 1048 }
1055 1049
1056 1050 static void
1057 1051 vdev_open_child(void *arg)
1058 1052 {
1059 1053 vdev_t *vd = arg;
1060 1054
1061 1055 vd->vdev_open_thread = curthread;
1062 1056 vd->vdev_open_error = vdev_open(vd);
1063 1057 vd->vdev_open_thread = NULL;
1064 1058 }
1065 1059
1066 1060 boolean_t
1067 1061 vdev_uses_zvols(vdev_t *vd)
1068 1062 {
1069 1063 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1070 1064 strlen(ZVOL_DIR)) == 0)
1071 1065 return (B_TRUE);
1072 1066 for (int c = 0; c < vd->vdev_children; c++)
1073 1067 if (vdev_uses_zvols(vd->vdev_child[c]))
1074 1068 return (B_TRUE);
1075 1069 return (B_FALSE);
1076 1070 }
1077 1071
1078 1072 void
1079 1073 vdev_open_children(vdev_t *vd)
1080 1074 {
1081 1075 taskq_t *tq;
1082 1076 int children = vd->vdev_children;
1083 1077
1084 1078 /*
1085 1079 * in order to handle pools on top of zvols, do the opens
1086 1080 * in a single thread so that the same thread holds the
1087 1081 * spa_namespace_lock
1088 1082 */
1089 1083 if (vdev_uses_zvols(vd)) {
1090 1084 for (int c = 0; c < children; c++)
1091 1085 vd->vdev_child[c]->vdev_open_error =
1092 1086 vdev_open(vd->vdev_child[c]);
1093 1087 return;
1094 1088 }
1095 1089 tq = taskq_create("vdev_open", children, minclsyspri,
1096 1090 children, children, TASKQ_PREPOPULATE);
1097 1091
1098 1092 for (int c = 0; c < children; c++)
1099 1093 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1100 1094 TQ_SLEEP) != NULL);
1101 1095
1102 1096 taskq_destroy(tq);
1103 1097 }
1104 1098
1105 1099 /*
1106 1100 * Prepare a virtual device for access.
1107 1101 */
1108 1102 int
1109 1103 vdev_open(vdev_t *vd)
1110 1104 {
1111 1105 spa_t *spa = vd->vdev_spa;
1112 1106 int error;
1113 1107 uint64_t osize = 0;
1114 1108 uint64_t max_osize = 0;
1115 1109 uint64_t asize, max_asize, psize;
1116 1110 uint64_t ashift = 0;
1117 1111
1118 1112 ASSERT(vd->vdev_open_thread == curthread ||
1119 1113 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1120 1114 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1121 1115 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1122 1116 vd->vdev_state == VDEV_STATE_OFFLINE);
1123 1117
1124 1118 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1125 1119 vd->vdev_cant_read = B_FALSE;
1126 1120 vd->vdev_cant_write = B_FALSE;
1127 1121 vd->vdev_min_asize = vdev_get_min_asize(vd);
1128 1122
1129 1123 /*
1130 1124 * If this vdev is not removed, check its fault status. If it's
1131 1125 * faulted, bail out of the open.
1132 1126 */
1133 1127 if (!vd->vdev_removed && vd->vdev_faulted) {
1134 1128 ASSERT(vd->vdev_children == 0);
1135 1129 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1136 1130 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1137 1131 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1138 1132 vd->vdev_label_aux);
1139 1133 return (SET_ERROR(ENXIO));
1140 1134 } else if (vd->vdev_offline) {
1141 1135 ASSERT(vd->vdev_children == 0);
1142 1136 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1143 1137 return (SET_ERROR(ENXIO));
1144 1138 }
1145 1139
1146 1140 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1147 1141
1148 1142 /*
1149 1143 * Reset the vdev_reopening flag so that we actually close
1150 1144 * the vdev on error.
1151 1145 */
1152 1146 vd->vdev_reopening = B_FALSE;
1153 1147 if (zio_injection_enabled && error == 0)
1154 1148 error = zio_handle_device_injection(vd, NULL, ENXIO);
1155 1149
1156 1150 if (error) {
1157 1151 if (vd->vdev_removed &&
1158 1152 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1159 1153 vd->vdev_removed = B_FALSE;
1160 1154
1161 1155 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1162 1156 vd->vdev_stat.vs_aux);
1163 1157 return (error);
1164 1158 }
1165 1159
1166 1160 vd->vdev_removed = B_FALSE;
1167 1161
1168 1162 /*
1169 1163 * Recheck the faulted flag now that we have confirmed that
1170 1164 * the vdev is accessible. If we're faulted, bail.
1171 1165 */
1172 1166 if (vd->vdev_faulted) {
1173 1167 ASSERT(vd->vdev_children == 0);
1174 1168 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1175 1169 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1176 1170 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1177 1171 vd->vdev_label_aux);
1178 1172 return (SET_ERROR(ENXIO));
1179 1173 }
1180 1174
1181 1175 if (vd->vdev_degraded) {
1182 1176 ASSERT(vd->vdev_children == 0);
1183 1177 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1184 1178 VDEV_AUX_ERR_EXCEEDED);
1185 1179 } else {
1186 1180 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1187 1181 }
1188 1182
1189 1183 /*
1190 1184 * For hole or missing vdevs we just return success.
1191 1185 */
1192 1186 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1193 1187 return (0);
1194 1188
1195 1189 for (int c = 0; c < vd->vdev_children; c++) {
1196 1190 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1197 1191 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1198 1192 VDEV_AUX_NONE);
1199 1193 break;
1200 1194 }
1201 1195 }
1202 1196
1203 1197 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1204 1198 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1205 1199
1206 1200 if (vd->vdev_children == 0) {
1207 1201 if (osize < SPA_MINDEVSIZE) {
1208 1202 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1209 1203 VDEV_AUX_TOO_SMALL);
1210 1204 return (SET_ERROR(EOVERFLOW));
1211 1205 }
1212 1206 psize = osize;
1213 1207 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1214 1208 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1215 1209 VDEV_LABEL_END_SIZE);
1216 1210 } else {
1217 1211 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1218 1212 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1219 1213 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1220 1214 VDEV_AUX_TOO_SMALL);
1221 1215 return (SET_ERROR(EOVERFLOW));
1222 1216 }
1223 1217 psize = 0;
1224 1218 asize = osize;
1225 1219 max_asize = max_osize;
1226 1220 }
1227 1221
1228 1222 vd->vdev_psize = psize;
1229 1223
1230 1224 /*
1231 1225 * Make sure the allocatable size hasn't shrunk.
1232 1226 */
1233 1227 if (asize < vd->vdev_min_asize) {
1234 1228 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1235 1229 VDEV_AUX_BAD_LABEL);
1236 1230 return (SET_ERROR(EINVAL));
1237 1231 }
1238 1232
1239 1233 if (vd->vdev_asize == 0) {
1240 1234 /*
1241 1235 * This is the first-ever open, so use the computed values.
1242 1236 * For testing purposes, a higher ashift can be requested.
1243 1237 */
1244 1238 vd->vdev_asize = asize;
1245 1239 vd->vdev_max_asize = max_asize;
1246 1240 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1247 1241 } else {
1248 1242 /*
1249 1243 * Detect if the alignment requirement has increased.
1250 1244 * We don't want to make the pool unavailable, just
1251 1245 * issue a warning instead.
1252 1246 */
1253 1247 if (ashift > vd->vdev_top->vdev_ashift &&
1254 1248 vd->vdev_ops->vdev_op_leaf) {
1255 1249 cmn_err(CE_WARN,
1256 1250 "Disk, '%s', has a block alignment that is "
1257 1251 "larger than the pool's alignment\n",
1258 1252 vd->vdev_path);
1259 1253 }
1260 1254 vd->vdev_max_asize = max_asize;
1261 1255 }
1262 1256
1263 1257 /*
1264 1258 * If all children are healthy and the asize has increased,
1265 1259 * then we've experienced dynamic LUN growth. If automatic
1266 1260 * expansion is enabled then use the additional space.
1267 1261 */
1268 1262 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1269 1263 (vd->vdev_expanding || spa->spa_autoexpand))
1270 1264 vd->vdev_asize = asize;
1271 1265
1272 1266 vdev_set_min_asize(vd);
1273 1267
1274 1268 /*
1275 1269 * Ensure we can issue some IO before declaring the
1276 1270 * vdev open for business.
1277 1271 */
1278 1272 if (vd->vdev_ops->vdev_op_leaf &&
1279 1273 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1280 1274 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1281 1275 VDEV_AUX_ERR_EXCEEDED);
1282 1276 return (error);
1283 1277 }
1284 1278
1285 1279 /*
1286 1280 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1287 1281 * resilver. But don't do this if we are doing a reopen for a scrub,
1288 1282 * since this would just restart the scrub we are already doing.
1289 1283 */
1290 1284 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1291 1285 vdev_resilver_needed(vd, NULL, NULL))
1292 1286 spa_async_request(spa, SPA_ASYNC_RESILVER);
1293 1287
1294 1288 return (0);
1295 1289 }
1296 1290
1297 1291 /*
1298 1292 * Called once the vdevs are all opened, this routine validates the label
1299 1293 * contents. This needs to be done before vdev_load() so that we don't
1300 1294 * inadvertently do repair I/Os to the wrong device.
1301 1295 *
1302 1296 * If 'strict' is false ignore the spa guid check. This is necessary because
1303 1297 * if the machine crashed during a re-guid the new guid might have been written
1304 1298 * to all of the vdev labels, but not the cached config. The strict check
1305 1299 * will be performed when the pool is opened again using the mos config.
1306 1300 *
1307 1301 * This function will only return failure if one of the vdevs indicates that it
1308 1302 * has since been destroyed or exported. This is only possible if
1309 1303 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1310 1304 * will be updated but the function will return 0.
1311 1305 */
1312 1306 int
1313 1307 vdev_validate(vdev_t *vd, boolean_t strict)
1314 1308 {
1315 1309 spa_t *spa = vd->vdev_spa;
1316 1310 nvlist_t *label;
1317 1311 uint64_t guid = 0, top_guid;
1318 1312 uint64_t state;
1319 1313
1320 1314 for (int c = 0; c < vd->vdev_children; c++)
1321 1315 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1322 1316 return (SET_ERROR(EBADF));
1323 1317
1324 1318 /*
1325 1319 * If the device has already failed, or was marked offline, don't do
1326 1320 * any further validation. Otherwise, label I/O will fail and we will
1327 1321 * overwrite the previous state.
1328 1322 */
1329 1323 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1330 1324 uint64_t aux_guid = 0;
1331 1325 nvlist_t *nvl;
1332 1326 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1333 1327 spa_last_synced_txg(spa) : -1ULL;
1334 1328
1335 1329 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1336 1330 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1337 1331 VDEV_AUX_BAD_LABEL);
1338 1332 return (0);
1339 1333 }
1340 1334
1341 1335 /*
1342 1336 * Determine if this vdev has been split off into another
1343 1337 * pool. If so, then refuse to open it.
1344 1338 */
1345 1339 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1346 1340 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1347 1341 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1348 1342 VDEV_AUX_SPLIT_POOL);
1349 1343 nvlist_free(label);
1350 1344 return (0);
1351 1345 }
1352 1346
1353 1347 if (strict && (nvlist_lookup_uint64(label,
1354 1348 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1355 1349 guid != spa_guid(spa))) {
1356 1350 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1357 1351 VDEV_AUX_CORRUPT_DATA);
1358 1352 nvlist_free(label);
1359 1353 return (0);
1360 1354 }
1361 1355
1362 1356 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1363 1357 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1364 1358 &aux_guid) != 0)
1365 1359 aux_guid = 0;
1366 1360
1367 1361 /*
1368 1362 * If this vdev just became a top-level vdev because its
1369 1363 * sibling was detached, it will have adopted the parent's
1370 1364 * vdev guid -- but the label may or may not be on disk yet.
1371 1365 * Fortunately, either version of the label will have the
1372 1366 * same top guid, so if we're a top-level vdev, we can
1373 1367 * safely compare to that instead.
1374 1368 *
1375 1369 * If we split this vdev off instead, then we also check the
1376 1370 * original pool's guid. We don't want to consider the vdev
1377 1371 * corrupt if it is partway through a split operation.
1378 1372 */
1379 1373 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1380 1374 &guid) != 0 ||
1381 1375 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1382 1376 &top_guid) != 0 ||
1383 1377 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1384 1378 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1385 1379 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1386 1380 VDEV_AUX_CORRUPT_DATA);
1387 1381 nvlist_free(label);
1388 1382 return (0);
1389 1383 }
1390 1384
1391 1385 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1392 1386 &state) != 0) {
1393 1387 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1394 1388 VDEV_AUX_CORRUPT_DATA);
1395 1389 nvlist_free(label);
1396 1390 return (0);
1397 1391 }
1398 1392
1399 1393 nvlist_free(label);
1400 1394
1401 1395 /*
1402 1396 * If this is a verbatim import, no need to check the
1403 1397 * state of the pool.
1404 1398 */
1405 1399 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1406 1400 spa_load_state(spa) == SPA_LOAD_OPEN &&
1407 1401 state != POOL_STATE_ACTIVE)
1408 1402 return (SET_ERROR(EBADF));
1409 1403
1410 1404 /*
1411 1405 * If we were able to open and validate a vdev that was
1412 1406 * previously marked permanently unavailable, clear that state
1413 1407 * now.
1414 1408 */
1415 1409 if (vd->vdev_not_present)
1416 1410 vd->vdev_not_present = 0;
1417 1411 }
1418 1412
1419 1413 return (0);
1420 1414 }
1421 1415
1422 1416 /*
1423 1417 * Close a virtual device.
1424 1418 */
1425 1419 void
1426 1420 vdev_close(vdev_t *vd)
1427 1421 {
1428 1422 spa_t *spa = vd->vdev_spa;
1429 1423 vdev_t *pvd = vd->vdev_parent;
1430 1424
1431 1425 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1432 1426
1433 1427 /*
1434 1428 * If our parent is reopening, then we are as well, unless we are
1435 1429 * going offline.
1436 1430 */
1437 1431 if (pvd != NULL && pvd->vdev_reopening)
1438 1432 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1439 1433
1440 1434 vd->vdev_ops->vdev_op_close(vd);
1441 1435
1442 1436 vdev_cache_purge(vd);
1443 1437
1444 1438 /*
1445 1439 * We record the previous state before we close it, so that if we are
1446 1440 * doing a reopen(), we don't generate FMA ereports if we notice that
1447 1441 * it's still faulted.
1448 1442 */
1449 1443 vd->vdev_prevstate = vd->vdev_state;
1450 1444
1451 1445 if (vd->vdev_offline)
1452 1446 vd->vdev_state = VDEV_STATE_OFFLINE;
1453 1447 else
1454 1448 vd->vdev_state = VDEV_STATE_CLOSED;
1455 1449 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1456 1450 }
1457 1451
1458 1452 void
1459 1453 vdev_hold(vdev_t *vd)
1460 1454 {
1461 1455 spa_t *spa = vd->vdev_spa;
1462 1456
1463 1457 ASSERT(spa_is_root(spa));
1464 1458 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1465 1459 return;
1466 1460
1467 1461 for (int c = 0; c < vd->vdev_children; c++)
1468 1462 vdev_hold(vd->vdev_child[c]);
1469 1463
1470 1464 if (vd->vdev_ops->vdev_op_leaf)
1471 1465 vd->vdev_ops->vdev_op_hold(vd);
1472 1466 }
1473 1467
1474 1468 void
1475 1469 vdev_rele(vdev_t *vd)
1476 1470 {
1477 1471 spa_t *spa = vd->vdev_spa;
1478 1472
1479 1473 ASSERT(spa_is_root(spa));
1480 1474 for (int c = 0; c < vd->vdev_children; c++)
1481 1475 vdev_rele(vd->vdev_child[c]);
1482 1476
1483 1477 if (vd->vdev_ops->vdev_op_leaf)
1484 1478 vd->vdev_ops->vdev_op_rele(vd);
1485 1479 }
1486 1480
1487 1481 /*
1488 1482 * Reopen all interior vdevs and any unopened leaves. We don't actually
1489 1483 * reopen leaf vdevs which had previously been opened as they might deadlock
1490 1484 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1491 1485 * If the leaf has never been opened then open it, as usual.
1492 1486 */
1493 1487 void
1494 1488 vdev_reopen(vdev_t *vd)
1495 1489 {
1496 1490 spa_t *spa = vd->vdev_spa;
1497 1491
1498 1492 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1499 1493
1500 1494 /* set the reopening flag unless we're taking the vdev offline */
1501 1495 vd->vdev_reopening = !vd->vdev_offline;
1502 1496 vdev_close(vd);
1503 1497 (void) vdev_open(vd);
1504 1498
1505 1499 /*
1506 1500 * Call vdev_validate() here to make sure we have the same device.
1507 1501 * Otherwise, a device with an invalid label could be successfully
1508 1502 * opened in response to vdev_reopen().
1509 1503 */
1510 1504 if (vd->vdev_aux) {
1511 1505 (void) vdev_validate_aux(vd);
1512 1506 if (vdev_readable(vd) && vdev_writeable(vd) &&
1513 1507 vd->vdev_aux == &spa->spa_l2cache &&
1514 1508 !l2arc_vdev_present(vd))
1515 1509 l2arc_add_vdev(spa, vd);
1516 1510 } else {
1517 1511 (void) vdev_validate(vd, B_TRUE);
1518 1512 }
1519 1513
1520 1514 /*
1521 1515 * Reassess parent vdev's health.
1522 1516 */
1523 1517 vdev_propagate_state(vd);
1524 1518 }
1525 1519
1526 1520 int
1527 1521 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1528 1522 {
1529 1523 int error;
1530 1524
1531 1525 /*
1532 1526 * Normally, partial opens (e.g. of a mirror) are allowed.
1533 1527 * For a create, however, we want to fail the request if
|
↓ open down ↓ |
630 lines elided |
↑ open up ↑ |
1534 1528 * there are any components we can't open.
1535 1529 */
1536 1530 error = vdev_open(vd);
1537 1531
1538 1532 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1539 1533 vdev_close(vd);
1540 1534 return (error ? error : ENXIO);
1541 1535 }
1542 1536
1543 1537 /*
1544 - * Recursively initialize all labels.
1538 + * Recursively load DTLs and initialize all labels.
1545 1539 */
1546 - if ((error = vdev_label_init(vd, txg, isreplacing ?
1540 + if ((error = vdev_dtl_load(vd)) != 0 ||
1541 + (error = vdev_label_init(vd, txg, isreplacing ?
1547 1542 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1548 1543 vdev_close(vd);
1549 1544 return (error);
1550 1545 }
1551 1546
1552 1547 return (0);
1553 1548 }
1554 1549
1555 1550 void
1556 1551 vdev_metaslab_set_size(vdev_t *vd)
1557 1552 {
1558 1553 /*
1559 1554 * Aim for roughly 200 metaslabs per vdev.
1560 1555 */
1561 1556 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1562 1557 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1563 1558 }
1564 1559
1565 1560 void
1566 1561 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1567 1562 {
1568 1563 ASSERT(vd == vd->vdev_top);
1569 1564 ASSERT(!vd->vdev_ishole);
1570 1565 ASSERT(ISP2(flags));
1571 1566 ASSERT(spa_writeable(vd->vdev_spa));
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
1572 1567
1573 1568 if (flags & VDD_METASLAB)
1574 1569 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1575 1570
1576 1571 if (flags & VDD_DTL)
1577 1572 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1578 1573
1579 1574 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1580 1575 }
1581 1576
1577 +void
1578 +vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1579 +{
1580 + for (int c = 0; c < vd->vdev_children; c++)
1581 + vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1582 +
1583 + if (vd->vdev_ops->vdev_op_leaf)
1584 + vdev_dirty(vd->vdev_top, flags, vd, txg);
1585 +}
1586 +
1582 1587 /*
1583 1588 * DTLs.
1584 1589 *
1585 1590 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1586 1591 * the vdev has less than perfect replication. There are four kinds of DTL:
1587 1592 *
1588 1593 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1589 1594 *
1590 1595 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1591 1596 *
1592 1597 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1593 1598 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1594 1599 * txgs that was scrubbed.
1595 1600 *
1596 1601 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1597 1602 * persistent errors or just some device being offline.
1598 1603 * Unlike the other three, the DTL_OUTAGE map is not generally
1599 1604 * maintained; it's only computed when needed, typically to
1600 1605 * determine whether a device can be detached.
1601 1606 *
1602 1607 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1603 1608 * either has the data or it doesn't.
1604 1609 *
1605 1610 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1606 1611 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1607 1612 * if any child is less than fully replicated, then so is its parent.
1608 1613 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1609 1614 * comprising only those txgs which appear in 'maxfaults' or more children;
1610 1615 * those are the txgs we don't have enough replication to read. For example,
1611 1616 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1612 1617 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
|
↓ open down ↓ |
21 lines elided |
↑ open up ↑ |
1613 1618 * two child DTL_MISSING maps.
1614 1619 *
1615 1620 * It should be clear from the above that to compute the DTLs and outage maps
1616 1621 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1617 1622 * Therefore, that is all we keep on disk. When loading the pool, or after
1618 1623 * a configuration change, we generate all other DTLs from first principles.
1619 1624 */
1620 1625 void
1621 1626 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1622 1627 {
1623 - space_map_t *sm = &vd->vdev_dtl[t];
1628 + range_tree_t *rt = vd->vdev_dtl[t];
1624 1629
1625 1630 ASSERT(t < DTL_TYPES);
1626 1631 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1627 1632 ASSERT(spa_writeable(vd->vdev_spa));
1628 1633
1629 - mutex_enter(sm->sm_lock);
1630 - if (!space_map_contains(sm, txg, size))
1631 - space_map_add(sm, txg, size);
1632 - mutex_exit(sm->sm_lock);
1634 + mutex_enter(rt->rt_lock);
1635 + if (!range_tree_contains(rt, txg, size))
1636 + range_tree_add(rt, txg, size);
1637 + mutex_exit(rt->rt_lock);
1633 1638 }
1634 1639
1635 1640 boolean_t
1636 1641 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1637 1642 {
1638 - space_map_t *sm = &vd->vdev_dtl[t];
1643 + range_tree_t *rt = vd->vdev_dtl[t];
1639 1644 boolean_t dirty = B_FALSE;
1640 1645
1641 1646 ASSERT(t < DTL_TYPES);
1642 1647 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1643 1648
1644 - mutex_enter(sm->sm_lock);
1645 - if (sm->sm_space != 0)
1646 - dirty = space_map_contains(sm, txg, size);
1647 - mutex_exit(sm->sm_lock);
1649 + mutex_enter(rt->rt_lock);
1650 + if (range_tree_space(rt) != 0)
1651 + dirty = range_tree_contains(rt, txg, size);
1652 + mutex_exit(rt->rt_lock);
1648 1653
1649 1654 return (dirty);
1650 1655 }
1651 1656
1652 1657 boolean_t
1653 1658 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1654 1659 {
1655 - space_map_t *sm = &vd->vdev_dtl[t];
1660 + range_tree_t *rt = vd->vdev_dtl[t];
1656 1661 boolean_t empty;
1657 1662
1658 - mutex_enter(sm->sm_lock);
1659 - empty = (sm->sm_space == 0);
1660 - mutex_exit(sm->sm_lock);
1663 + mutex_enter(rt->rt_lock);
1664 + empty = (range_tree_space(rt) == 0);
1665 + mutex_exit(rt->rt_lock);
1661 1666
1662 1667 return (empty);
1663 1668 }
1664 1669
1665 1670 /*
1666 1671 * Returns the lowest txg in the DTL range.
1667 1672 */
1668 1673 static uint64_t
1669 1674 vdev_dtl_min(vdev_t *vd)
1670 1675 {
1671 - space_seg_t *ss;
1676 + range_seg_t *rs;
1672 1677
1673 1678 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1674 - ASSERT3U(vd->vdev_dtl[DTL_MISSING].sm_space, !=, 0);
1679 + ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1675 1680 ASSERT0(vd->vdev_children);
1676 1681
1677 - ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1678 - return (ss->ss_start - 1);
1682 + rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1683 + return (rs->rs_start - 1);
1679 1684 }
1680 1685
1681 1686 /*
1682 1687 * Returns the highest txg in the DTL.
1683 1688 */
1684 1689 static uint64_t
1685 1690 vdev_dtl_max(vdev_t *vd)
1686 1691 {
1687 - space_seg_t *ss;
1692 + range_seg_t *rs;
1688 1693
1689 1694 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1690 - ASSERT3U(vd->vdev_dtl[DTL_MISSING].sm_space, !=, 0);
1695 + ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1691 1696 ASSERT0(vd->vdev_children);
1692 1697
1693 - ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1694 - return (ss->ss_end);
1698 + rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1699 + return (rs->rs_end);
1695 1700 }
1696 1701
1697 1702 /*
1698 1703 * Determine if a resilvering vdev should remove any DTL entries from
1699 1704 * its range. If the vdev was resilvering for the entire duration of the
1700 1705 * scan then it should excise that range from its DTLs. Otherwise, this
1701 1706 * vdev is considered partially resilvered and should leave its DTL
1702 1707 * entries intact. The comment in vdev_dtl_reassess() describes how we
1703 1708 * excise the DTLs.
1704 1709 */
1705 1710 static boolean_t
1706 1711 vdev_dtl_should_excise(vdev_t *vd)
1707 1712 {
1708 1713 spa_t *spa = vd->vdev_spa;
1709 1714 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1710 1715
1711 1716 ASSERT0(scn->scn_phys.scn_errors);
1712 1717 ASSERT0(vd->vdev_children);
1713 1718
1714 1719 if (vd->vdev_resilver_txg == 0 ||
1715 - vd->vdev_dtl[DTL_MISSING].sm_space == 0)
1720 + range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1716 1721 return (B_TRUE);
1717 1722
1718 1723 /*
1719 1724 * When a resilver is initiated the scan will assign the scn_max_txg
1720 1725 * value to the highest txg value that exists in all DTLs. If this
1721 1726 * device's max DTL is not part of this scan (i.e. it is not in
1722 1727 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1723 1728 * for excision.
1724 1729 */
1725 1730 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1726 1731 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1727 1732 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1728 1733 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1729 1734 return (B_TRUE);
1730 1735 }
1731 1736 return (B_FALSE);
1732 1737 }
1733 1738
1734 1739 /*
1735 1740 * Reassess DTLs after a config change or scrub completion.
1736 1741 */
1737 1742 void
1738 1743 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1739 1744 {
1740 1745 spa_t *spa = vd->vdev_spa;
1741 1746 avl_tree_t reftree;
1742 1747 int minref;
1743 1748
1744 1749 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1745 1750
1746 1751 for (int c = 0; c < vd->vdev_children; c++)
1747 1752 vdev_dtl_reassess(vd->vdev_child[c], txg,
1748 1753 scrub_txg, scrub_done);
1749 1754
1750 1755 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1751 1756 return;
1752 1757
1753 1758 if (vd->vdev_ops->vdev_op_leaf) {
1754 1759 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1755 1760
1756 1761 mutex_enter(&vd->vdev_dtl_lock);
1757 1762
1758 1763 /*
1759 1764 * If we've completed a scan cleanly then determine
1760 1765 * if this vdev should remove any DTLs. We only want to
1761 1766 * excise regions on vdevs that were available during
1762 1767 * the entire duration of this scan.
1763 1768 */
1764 1769 if (scrub_txg != 0 &&
1765 1770 (spa->spa_scrub_started ||
1766 1771 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1767 1772 vdev_dtl_should_excise(vd)) {
1768 1773 /*
1769 1774 * We completed a scrub up to scrub_txg. If we
1770 1775 * did it without rebooting, then the scrub dtl
1771 1776 * will be valid, so excise the old region and
1772 1777 * fold in the scrub dtl. Otherwise, leave the
1773 1778 * dtl as-is if there was an error.
1774 1779 *
|
↓ open down ↓ |
49 lines elided |
↑ open up ↑ |
1775 1780 * There's little trick here: to excise the beginning
1776 1781 * of the DTL_MISSING map, we put it into a reference
1777 1782 * tree and then add a segment with refcnt -1 that
1778 1783 * covers the range [0, scrub_txg). This means
1779 1784 * that each txg in that range has refcnt -1 or 0.
1780 1785 * We then add DTL_SCRUB with a refcnt of 2, so that
1781 1786 * entries in the range [0, scrub_txg) will have a
1782 1787 * positive refcnt -- either 1 or 2. We then convert
1783 1788 * the reference tree into the new DTL_MISSING map.
1784 1789 */
1785 - space_map_ref_create(&reftree);
1786 - space_map_ref_add_map(&reftree,
1787 - &vd->vdev_dtl[DTL_MISSING], 1);
1788 - space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1789 - space_map_ref_add_map(&reftree,
1790 - &vd->vdev_dtl[DTL_SCRUB], 2);
1791 - space_map_ref_generate_map(&reftree,
1792 - &vd->vdev_dtl[DTL_MISSING], 1);
1793 - space_map_ref_destroy(&reftree);
1790 + space_reftree_create(&reftree);
1791 + space_reftree_add_map(&reftree,
1792 + vd->vdev_dtl[DTL_MISSING], 1);
1793 + space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1794 + space_reftree_add_map(&reftree,
1795 + vd->vdev_dtl[DTL_SCRUB], 2);
1796 + space_reftree_generate_map(&reftree,
1797 + vd->vdev_dtl[DTL_MISSING], 1);
1798 + space_reftree_destroy(&reftree);
1794 1799 }
1795 - space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1796 - space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1797 - space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1800 + range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1801 + range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1802 + range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1798 1803 if (scrub_done)
1799 - space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1800 - space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1804 + range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1805 + range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1801 1806 if (!vdev_readable(vd))
1802 - space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1807 + range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1803 1808 else
1804 - space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1805 - space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1809 + range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1810 + range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1806 1811
1807 1812 /*
1808 1813 * If the vdev was resilvering and no longer has any
1809 1814 * DTLs then reset its resilvering flag.
1810 1815 */
1811 1816 if (vd->vdev_resilver_txg != 0 &&
1812 - vd->vdev_dtl[DTL_MISSING].sm_space == 0 &&
1813 - vd->vdev_dtl[DTL_OUTAGE].sm_space == 0)
1817 + range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1818 + range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
1814 1819 vd->vdev_resilver_txg = 0;
1815 1820
1816 1821 mutex_exit(&vd->vdev_dtl_lock);
1817 1822
1818 1823 if (txg != 0)
1819 1824 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1820 1825 return;
1821 1826 }
1822 1827
1823 1828 mutex_enter(&vd->vdev_dtl_lock);
1824 1829 for (int t = 0; t < DTL_TYPES; t++) {
|
↓ open down ↓ |
1 lines elided |
↑ open up ↑ |
1825 1830 /* account for child's outage in parent's missing map */
1826 1831 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1827 1832 if (t == DTL_SCRUB)
1828 1833 continue; /* leaf vdevs only */
1829 1834 if (t == DTL_PARTIAL)
1830 1835 minref = 1; /* i.e. non-zero */
1831 1836 else if (vd->vdev_nparity != 0)
1832 1837 minref = vd->vdev_nparity + 1; /* RAID-Z */
1833 1838 else
1834 1839 minref = vd->vdev_children; /* any kind of mirror */
1835 - space_map_ref_create(&reftree);
1840 + space_reftree_create(&reftree);
1836 1841 for (int c = 0; c < vd->vdev_children; c++) {
1837 1842 vdev_t *cvd = vd->vdev_child[c];
1838 1843 mutex_enter(&cvd->vdev_dtl_lock);
1839 - space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1844 + space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1840 1845 mutex_exit(&cvd->vdev_dtl_lock);
1841 1846 }
1842 - space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1843 - space_map_ref_destroy(&reftree);
1847 + space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1848 + space_reftree_destroy(&reftree);
1844 1849 }
1845 1850 mutex_exit(&vd->vdev_dtl_lock);
1846 1851 }
1847 1852
1848 -static int
1853 +int
1849 1854 vdev_dtl_load(vdev_t *vd)
1850 1855 {
1851 1856 spa_t *spa = vd->vdev_spa;
1852 - space_map_obj_t *smo = &vd->vdev_dtl_smo;
1853 1857 objset_t *mos = spa->spa_meta_objset;
1854 - dmu_buf_t *db;
1855 - int error;
1858 + int error = 0;
1856 1859
1857 - ASSERT(vd->vdev_children == 0);
1860 + if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
1861 + ASSERT(!vd->vdev_ishole);
1858 1862
1859 - if (smo->smo_object == 0)
1860 - return (0);
1863 + error = space_map_open(&vd->vdev_dtl_sm, mos,
1864 + vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
1865 + if (error)
1866 + return (error);
1867 + ASSERT(vd->vdev_dtl_sm != NULL);
1861 1868
1862 - ASSERT(!vd->vdev_ishole);
1869 + mutex_enter(&vd->vdev_dtl_lock);
1863 1870
1864 - if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1865 - return (error);
1871 + /*
1872 + * Now that we've opened the space_map we need to update
1873 + * the in-core DTL.
1874 + */
1875 + space_map_update(vd->vdev_dtl_sm);
1866 1876
1867 - ASSERT3U(db->db_size, >=, sizeof (*smo));
1868 - bcopy(db->db_data, smo, sizeof (*smo));
1869 - dmu_buf_rele(db, FTAG);
1877 + error = space_map_load(vd->vdev_dtl_sm,
1878 + vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
1879 + mutex_exit(&vd->vdev_dtl_lock);
1870 1880
1871 - mutex_enter(&vd->vdev_dtl_lock);
1872 - error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1873 - NULL, SM_ALLOC, smo, mos);
1874 - mutex_exit(&vd->vdev_dtl_lock);
1881 + return (error);
1882 + }
1875 1883
1884 + for (int c = 0; c < vd->vdev_children; c++) {
1885 + error = vdev_dtl_load(vd->vdev_child[c]);
1886 + if (error != 0)
1887 + break;
1888 + }
1889 +
1876 1890 return (error);
1877 1891 }
1878 1892
1879 1893 void
1880 1894 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1881 1895 {
1882 1896 spa_t *spa = vd->vdev_spa;
1883 - space_map_obj_t *smo = &vd->vdev_dtl_smo;
1884 - space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1897 + range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
1885 1898 objset_t *mos = spa->spa_meta_objset;
1886 - space_map_t smsync;
1887 - kmutex_t smlock;
1888 - dmu_buf_t *db;
1899 + range_tree_t *rtsync;
1900 + kmutex_t rtlock;
1889 1901 dmu_tx_t *tx;
1902 + uint64_t object = space_map_object(vd->vdev_dtl_sm);
1890 1903
1891 1904 ASSERT(!vd->vdev_ishole);
1905 + ASSERT(vd->vdev_ops->vdev_op_leaf);
1892 1906
1893 1907 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1894 1908
1895 - if (vd->vdev_detached) {
1896 - if (smo->smo_object != 0) {
1897 - int err = dmu_object_free(mos, smo->smo_object, tx);
1898 - ASSERT0(err);
1899 - smo->smo_object = 0;
1900 - }
1909 + if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
1910 + mutex_enter(&vd->vdev_dtl_lock);
1911 + space_map_free(vd->vdev_dtl_sm, tx);
1912 + space_map_close(vd->vdev_dtl_sm);
1913 + vd->vdev_dtl_sm = NULL;
1914 + mutex_exit(&vd->vdev_dtl_lock);
1901 1915 dmu_tx_commit(tx);
1902 1916 return;
1903 1917 }
1904 1918
1905 - if (smo->smo_object == 0) {
1906 - ASSERT(smo->smo_objsize == 0);
1907 - ASSERT(smo->smo_alloc == 0);
1908 - smo->smo_object = dmu_object_alloc(mos,
1909 - DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1910 - DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1911 - ASSERT(smo->smo_object != 0);
1912 - vdev_config_dirty(vd->vdev_top);
1919 + if (vd->vdev_dtl_sm == NULL) {
1920 + uint64_t new_object;
1921 +
1922 + new_object = space_map_alloc(mos, tx);
1923 + VERIFY3U(new_object, !=, 0);
1924 +
1925 + VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
1926 + 0, -1ULL, 0, &vd->vdev_dtl_lock));
1927 + ASSERT(vd->vdev_dtl_sm != NULL);
1913 1928 }
1914 1929
1915 - mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1930 + mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
1916 1931
1917 - space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1918 - &smlock);
1932 + rtsync = range_tree_create(NULL, NULL, &rtlock);
1919 1933
1920 - mutex_enter(&smlock);
1934 + mutex_enter(&rtlock);
1921 1935
1922 1936 mutex_enter(&vd->vdev_dtl_lock);
1923 - space_map_walk(sm, space_map_add, &smsync);
1937 + range_tree_walk(rt, range_tree_add, rtsync);
1924 1938 mutex_exit(&vd->vdev_dtl_lock);
1925 1939
1926 - space_map_truncate(smo, mos, tx);
1927 - space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1928 - space_map_vacate(&smsync, NULL, NULL);
1940 + space_map_truncate(vd->vdev_dtl_sm, tx);
1941 + space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
1942 + range_tree_vacate(rtsync, NULL, NULL);
1929 1943
1930 - space_map_destroy(&smsync);
1944 + range_tree_destroy(rtsync);
1931 1945
1932 - mutex_exit(&smlock);
1933 - mutex_destroy(&smlock);
1946 + mutex_exit(&rtlock);
1947 + mutex_destroy(&rtlock);
1934 1948
1935 - VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1936 - dmu_buf_will_dirty(db, tx);
1937 - ASSERT3U(db->db_size, >=, sizeof (*smo));
1938 - bcopy(smo, db->db_data, sizeof (*smo));
1939 - dmu_buf_rele(db, FTAG);
1949 + /*
1950 + * If the object for the space map has changed then dirty
1951 + * the top level so that we update the config.
1952 + */
1953 + if (object != space_map_object(vd->vdev_dtl_sm)) {
1954 + zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
1955 + "new object %llu", txg, spa_name(spa), object,
1956 + space_map_object(vd->vdev_dtl_sm));
1957 + vdev_config_dirty(vd->vdev_top);
1958 + }
1940 1959
1941 1960 dmu_tx_commit(tx);
1961 +
1962 + mutex_enter(&vd->vdev_dtl_lock);
1963 + space_map_update(vd->vdev_dtl_sm);
1964 + mutex_exit(&vd->vdev_dtl_lock);
1942 1965 }
1943 1966
1944 1967 /*
1945 1968 * Determine whether the specified vdev can be offlined/detached/removed
1946 1969 * without losing data.
1947 1970 */
1948 1971 boolean_t
1949 1972 vdev_dtl_required(vdev_t *vd)
1950 1973 {
1951 1974 spa_t *spa = vd->vdev_spa;
1952 1975 vdev_t *tvd = vd->vdev_top;
1953 1976 uint8_t cant_read = vd->vdev_cant_read;
1954 1977 boolean_t required;
1955 1978
1956 1979 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1957 1980
1958 1981 if (vd == spa->spa_root_vdev || vd == tvd)
1959 1982 return (B_TRUE);
1960 1983
1961 1984 /*
1962 1985 * Temporarily mark the device as unreadable, and then determine
1963 1986 * whether this results in any DTL outages in the top-level vdev.
1964 1987 * If not, we can safely offline/detach/remove the device.
1965 1988 */
1966 1989 vd->vdev_cant_read = B_TRUE;
1967 1990 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1968 1991 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1969 1992 vd->vdev_cant_read = cant_read;
1970 1993 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1971 1994
1972 1995 if (!required && zio_injection_enabled)
1973 1996 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1974 1997
1975 1998 return (required);
1976 1999 }
1977 2000
1978 2001 /*
1979 2002 * Determine if resilver is needed, and if so the txg range.
|
↓ open down ↓ |
28 lines elided |
↑ open up ↑ |
1980 2003 */
1981 2004 boolean_t
1982 2005 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1983 2006 {
1984 2007 boolean_t needed = B_FALSE;
1985 2008 uint64_t thismin = UINT64_MAX;
1986 2009 uint64_t thismax = 0;
1987 2010
1988 2011 if (vd->vdev_children == 0) {
1989 2012 mutex_enter(&vd->vdev_dtl_lock);
1990 - if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
2013 + if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
1991 2014 vdev_writeable(vd)) {
1992 2015
1993 2016 thismin = vdev_dtl_min(vd);
1994 2017 thismax = vdev_dtl_max(vd);
1995 2018 needed = B_TRUE;
1996 2019 }
1997 2020 mutex_exit(&vd->vdev_dtl_lock);
1998 2021 } else {
1999 2022 for (int c = 0; c < vd->vdev_children; c++) {
2000 2023 vdev_t *cvd = vd->vdev_child[c];
2001 2024 uint64_t cmin, cmax;
2002 2025
2003 2026 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2004 2027 thismin = MIN(thismin, cmin);
2005 2028 thismax = MAX(thismax, cmax);
2006 2029 needed = B_TRUE;
2007 2030 }
2008 2031 }
2009 2032 }
2010 2033
2011 2034 if (needed && minp) {
2012 2035 *minp = thismin;
2013 2036 *maxp = thismax;
2014 2037 }
2015 2038 return (needed);
2016 2039 }
2017 2040
2018 2041 void
2019 2042 vdev_load(vdev_t *vd)
2020 2043 {
2021 2044 /*
2022 2045 * Recursively load all children.
2023 2046 */
2024 2047 for (int c = 0; c < vd->vdev_children; c++)
2025 2048 vdev_load(vd->vdev_child[c]);
2026 2049
2027 2050 /*
2028 2051 * If this is a top-level vdev, initialize its metaslabs.
2029 2052 */
2030 2053 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2031 2054 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2032 2055 vdev_metaslab_init(vd, 0) != 0))
2033 2056 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2034 2057 VDEV_AUX_CORRUPT_DATA);
2035 2058
2036 2059 /*
2037 2060 * If this is a leaf vdev, load its DTL.
2038 2061 */
2039 2062 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2040 2063 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2041 2064 VDEV_AUX_CORRUPT_DATA);
2042 2065 }
2043 2066
2044 2067 /*
2045 2068 * The special vdev case is used for hot spares and l2cache devices. Its
2046 2069 * sole purpose it to set the vdev state for the associated vdev. To do this,
2047 2070 * we make sure that we can open the underlying device, then try to read the
2048 2071 * label, and make sure that the label is sane and that it hasn't been
2049 2072 * repurposed to another pool.
2050 2073 */
2051 2074 int
2052 2075 vdev_validate_aux(vdev_t *vd)
2053 2076 {
2054 2077 nvlist_t *label;
2055 2078 uint64_t guid, version;
2056 2079 uint64_t state;
2057 2080
2058 2081 if (!vdev_readable(vd))
2059 2082 return (0);
2060 2083
2061 2084 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2062 2085 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2063 2086 VDEV_AUX_CORRUPT_DATA);
2064 2087 return (-1);
2065 2088 }
2066 2089
2067 2090 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2068 2091 !SPA_VERSION_IS_SUPPORTED(version) ||
2069 2092 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2070 2093 guid != vd->vdev_guid ||
2071 2094 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2072 2095 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2073 2096 VDEV_AUX_CORRUPT_DATA);
2074 2097 nvlist_free(label);
2075 2098 return (-1);
2076 2099 }
2077 2100
2078 2101 /*
2079 2102 * We don't actually check the pool state here. If it's in fact in
2080 2103 * use by another pool, we update this fact on the fly when requested.
2081 2104 */
2082 2105 nvlist_free(label);
2083 2106 return (0);
2084 2107 }
|
↓ open down ↓ |
84 lines elided |
↑ open up ↑ |
2085 2108
2086 2109 void
2087 2110 vdev_remove(vdev_t *vd, uint64_t txg)
2088 2111 {
2089 2112 spa_t *spa = vd->vdev_spa;
2090 2113 objset_t *mos = spa->spa_meta_objset;
2091 2114 dmu_tx_t *tx;
2092 2115
2093 2116 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2094 2117
2095 - if (vd->vdev_dtl_smo.smo_object) {
2096 - ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2097 - (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2098 - vd->vdev_dtl_smo.smo_object = 0;
2099 - }
2100 -
2101 2118 if (vd->vdev_ms != NULL) {
2102 2119 for (int m = 0; m < vd->vdev_ms_count; m++) {
2103 2120 metaslab_t *msp = vd->vdev_ms[m];
2104 2121
2105 - if (msp == NULL || msp->ms_smo.smo_object == 0)
2122 + if (msp == NULL || msp->ms_sm == NULL)
2106 2123 continue;
2107 2124
2108 - ASSERT0(msp->ms_smo.smo_alloc);
2109 - (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2110 - msp->ms_smo.smo_object = 0;
2125 + mutex_enter(&msp->ms_lock);
2126 + VERIFY0(space_map_allocated(msp->ms_sm));
2127 + space_map_free(msp->ms_sm, tx);
2128 + space_map_close(msp->ms_sm);
2129 + msp->ms_sm = NULL;
2130 + mutex_exit(&msp->ms_lock);
2111 2131 }
2112 2132 }
2113 2133
2114 2134 if (vd->vdev_ms_array) {
2115 2135 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2116 2136 vd->vdev_ms_array = 0;
2117 - vd->vdev_ms_shift = 0;
2118 2137 }
2119 2138 dmu_tx_commit(tx);
2120 2139 }
2121 2140
2122 2141 void
2123 2142 vdev_sync_done(vdev_t *vd, uint64_t txg)
2124 2143 {
2125 2144 metaslab_t *msp;
2126 2145 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2127 2146
2128 2147 ASSERT(!vd->vdev_ishole);
2129 2148
2130 2149 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2131 2150 metaslab_sync_done(msp, txg);
2132 2151
2133 2152 if (reassess)
2134 2153 metaslab_sync_reassess(vd->vdev_mg);
2135 2154 }
2136 2155
2137 2156 void
2138 2157 vdev_sync(vdev_t *vd, uint64_t txg)
2139 2158 {
2140 2159 spa_t *spa = vd->vdev_spa;
2141 2160 vdev_t *lvd;
2142 2161 metaslab_t *msp;
2143 2162 dmu_tx_t *tx;
2144 2163
2145 2164 ASSERT(!vd->vdev_ishole);
2146 2165
2147 2166 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2148 2167 ASSERT(vd == vd->vdev_top);
2149 2168 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2150 2169 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2151 2170 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2152 2171 ASSERT(vd->vdev_ms_array != 0);
2153 2172 vdev_config_dirty(vd);
2154 2173 dmu_tx_commit(tx);
2155 2174 }
2156 2175
2157 2176 /*
2158 2177 * Remove the metadata associated with this vdev once it's empty.
2159 2178 */
2160 2179 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2161 2180 vdev_remove(vd, txg);
2162 2181
2163 2182 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2164 2183 metaslab_sync(msp, txg);
2165 2184 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2166 2185 }
2167 2186
2168 2187 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2169 2188 vdev_dtl_sync(lvd, txg);
2170 2189
2171 2190 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2172 2191 }
2173 2192
2174 2193 uint64_t
2175 2194 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2176 2195 {
2177 2196 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2178 2197 }
2179 2198
2180 2199 /*
2181 2200 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2182 2201 * not be opened, and no I/O is attempted.
2183 2202 */
2184 2203 int
2185 2204 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2186 2205 {
2187 2206 vdev_t *vd, *tvd;
2188 2207
2189 2208 spa_vdev_state_enter(spa, SCL_NONE);
2190 2209
2191 2210 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2192 2211 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2193 2212
2194 2213 if (!vd->vdev_ops->vdev_op_leaf)
2195 2214 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2196 2215
2197 2216 tvd = vd->vdev_top;
2198 2217
2199 2218 /*
2200 2219 * We don't directly use the aux state here, but if we do a
2201 2220 * vdev_reopen(), we need this value to be present to remember why we
2202 2221 * were faulted.
2203 2222 */
2204 2223 vd->vdev_label_aux = aux;
2205 2224
2206 2225 /*
2207 2226 * Faulted state takes precedence over degraded.
2208 2227 */
2209 2228 vd->vdev_delayed_close = B_FALSE;
2210 2229 vd->vdev_faulted = 1ULL;
2211 2230 vd->vdev_degraded = 0ULL;
2212 2231 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2213 2232
2214 2233 /*
2215 2234 * If this device has the only valid copy of the data, then
2216 2235 * back off and simply mark the vdev as degraded instead.
2217 2236 */
2218 2237 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2219 2238 vd->vdev_degraded = 1ULL;
2220 2239 vd->vdev_faulted = 0ULL;
2221 2240
2222 2241 /*
2223 2242 * If we reopen the device and it's not dead, only then do we
2224 2243 * mark it degraded.
2225 2244 */
2226 2245 vdev_reopen(tvd);
2227 2246
2228 2247 if (vdev_readable(vd))
2229 2248 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2230 2249 }
2231 2250
2232 2251 return (spa_vdev_state_exit(spa, vd, 0));
2233 2252 }
2234 2253
2235 2254 /*
2236 2255 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2237 2256 * user that something is wrong. The vdev continues to operate as normal as far
2238 2257 * as I/O is concerned.
2239 2258 */
2240 2259 int
2241 2260 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2242 2261 {
2243 2262 vdev_t *vd;
2244 2263
2245 2264 spa_vdev_state_enter(spa, SCL_NONE);
2246 2265
2247 2266 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2248 2267 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2249 2268
2250 2269 if (!vd->vdev_ops->vdev_op_leaf)
2251 2270 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2252 2271
2253 2272 /*
2254 2273 * If the vdev is already faulted, then don't do anything.
2255 2274 */
2256 2275 if (vd->vdev_faulted || vd->vdev_degraded)
2257 2276 return (spa_vdev_state_exit(spa, NULL, 0));
2258 2277
2259 2278 vd->vdev_degraded = 1ULL;
2260 2279 if (!vdev_is_dead(vd))
2261 2280 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2262 2281 aux);
2263 2282
2264 2283 return (spa_vdev_state_exit(spa, vd, 0));
2265 2284 }
2266 2285
2267 2286 /*
2268 2287 * Online the given vdev.
2269 2288 *
2270 2289 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2271 2290 * spare device should be detached when the device finishes resilvering.
2272 2291 * Second, the online should be treated like a 'test' online case, so no FMA
2273 2292 * events are generated if the device fails to open.
2274 2293 */
2275 2294 int
2276 2295 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2277 2296 {
2278 2297 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2279 2298
2280 2299 spa_vdev_state_enter(spa, SCL_NONE);
2281 2300
2282 2301 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2283 2302 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2284 2303
2285 2304 if (!vd->vdev_ops->vdev_op_leaf)
2286 2305 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2287 2306
2288 2307 tvd = vd->vdev_top;
2289 2308 vd->vdev_offline = B_FALSE;
2290 2309 vd->vdev_tmpoffline = B_FALSE;
2291 2310 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2292 2311 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2293 2312
2294 2313 /* XXX - L2ARC 1.0 does not support expansion */
2295 2314 if (!vd->vdev_aux) {
2296 2315 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2297 2316 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2298 2317 }
2299 2318
2300 2319 vdev_reopen(tvd);
2301 2320 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2302 2321
2303 2322 if (!vd->vdev_aux) {
2304 2323 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2305 2324 pvd->vdev_expanding = B_FALSE;
2306 2325 }
2307 2326
2308 2327 if (newstate)
2309 2328 *newstate = vd->vdev_state;
2310 2329 if ((flags & ZFS_ONLINE_UNSPARE) &&
2311 2330 !vdev_is_dead(vd) && vd->vdev_parent &&
2312 2331 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2313 2332 vd->vdev_parent->vdev_child[0] == vd)
2314 2333 vd->vdev_unspare = B_TRUE;
2315 2334
2316 2335 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2317 2336
2318 2337 /* XXX - L2ARC 1.0 does not support expansion */
2319 2338 if (vd->vdev_aux)
2320 2339 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2321 2340 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2322 2341 }
2323 2342 return (spa_vdev_state_exit(spa, vd, 0));
2324 2343 }
2325 2344
2326 2345 static int
2327 2346 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2328 2347 {
2329 2348 vdev_t *vd, *tvd;
2330 2349 int error = 0;
2331 2350 uint64_t generation;
2332 2351 metaslab_group_t *mg;
2333 2352
2334 2353 top:
2335 2354 spa_vdev_state_enter(spa, SCL_ALLOC);
2336 2355
2337 2356 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2338 2357 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2339 2358
2340 2359 if (!vd->vdev_ops->vdev_op_leaf)
2341 2360 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2342 2361
2343 2362 tvd = vd->vdev_top;
2344 2363 mg = tvd->vdev_mg;
2345 2364 generation = spa->spa_config_generation + 1;
2346 2365
2347 2366 /*
2348 2367 * If the device isn't already offline, try to offline it.
2349 2368 */
2350 2369 if (!vd->vdev_offline) {
2351 2370 /*
2352 2371 * If this device has the only valid copy of some data,
2353 2372 * don't allow it to be offlined. Log devices are always
2354 2373 * expendable.
2355 2374 */
2356 2375 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2357 2376 vdev_dtl_required(vd))
2358 2377 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2359 2378
2360 2379 /*
2361 2380 * If the top-level is a slog and it has had allocations
2362 2381 * then proceed. We check that the vdev's metaslab group
2363 2382 * is not NULL since it's possible that we may have just
2364 2383 * added this vdev but not yet initialized its metaslabs.
2365 2384 */
2366 2385 if (tvd->vdev_islog && mg != NULL) {
2367 2386 /*
2368 2387 * Prevent any future allocations.
2369 2388 */
2370 2389 metaslab_group_passivate(mg);
2371 2390 (void) spa_vdev_state_exit(spa, vd, 0);
2372 2391
2373 2392 error = spa_offline_log(spa);
2374 2393
2375 2394 spa_vdev_state_enter(spa, SCL_ALLOC);
2376 2395
2377 2396 /*
2378 2397 * Check to see if the config has changed.
2379 2398 */
2380 2399 if (error || generation != spa->spa_config_generation) {
2381 2400 metaslab_group_activate(mg);
2382 2401 if (error)
2383 2402 return (spa_vdev_state_exit(spa,
2384 2403 vd, error));
2385 2404 (void) spa_vdev_state_exit(spa, vd, 0);
2386 2405 goto top;
2387 2406 }
2388 2407 ASSERT0(tvd->vdev_stat.vs_alloc);
2389 2408 }
2390 2409
2391 2410 /*
2392 2411 * Offline this device and reopen its top-level vdev.
2393 2412 * If the top-level vdev is a log device then just offline
2394 2413 * it. Otherwise, if this action results in the top-level
2395 2414 * vdev becoming unusable, undo it and fail the request.
2396 2415 */
2397 2416 vd->vdev_offline = B_TRUE;
2398 2417 vdev_reopen(tvd);
2399 2418
2400 2419 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2401 2420 vdev_is_dead(tvd)) {
2402 2421 vd->vdev_offline = B_FALSE;
2403 2422 vdev_reopen(tvd);
2404 2423 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2405 2424 }
2406 2425
2407 2426 /*
2408 2427 * Add the device back into the metaslab rotor so that
2409 2428 * once we online the device it's open for business.
2410 2429 */
2411 2430 if (tvd->vdev_islog && mg != NULL)
2412 2431 metaslab_group_activate(mg);
2413 2432 }
2414 2433
2415 2434 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2416 2435
2417 2436 return (spa_vdev_state_exit(spa, vd, 0));
2418 2437 }
2419 2438
2420 2439 int
2421 2440 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2422 2441 {
2423 2442 int error;
2424 2443
2425 2444 mutex_enter(&spa->spa_vdev_top_lock);
2426 2445 error = vdev_offline_locked(spa, guid, flags);
2427 2446 mutex_exit(&spa->spa_vdev_top_lock);
2428 2447
2429 2448 return (error);
2430 2449 }
2431 2450
2432 2451 /*
2433 2452 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2434 2453 * vdev_offline(), we assume the spa config is locked. We also clear all
2435 2454 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2436 2455 */
2437 2456 void
2438 2457 vdev_clear(spa_t *spa, vdev_t *vd)
2439 2458 {
2440 2459 vdev_t *rvd = spa->spa_root_vdev;
2441 2460
2442 2461 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2443 2462
2444 2463 if (vd == NULL)
2445 2464 vd = rvd;
2446 2465
2447 2466 vd->vdev_stat.vs_read_errors = 0;
2448 2467 vd->vdev_stat.vs_write_errors = 0;
2449 2468 vd->vdev_stat.vs_checksum_errors = 0;
2450 2469
2451 2470 for (int c = 0; c < vd->vdev_children; c++)
2452 2471 vdev_clear(spa, vd->vdev_child[c]);
2453 2472
2454 2473 /*
2455 2474 * If we're in the FAULTED state or have experienced failed I/O, then
2456 2475 * clear the persistent state and attempt to reopen the device. We
2457 2476 * also mark the vdev config dirty, so that the new faulted state is
2458 2477 * written out to disk.
2459 2478 */
2460 2479 if (vd->vdev_faulted || vd->vdev_degraded ||
2461 2480 !vdev_readable(vd) || !vdev_writeable(vd)) {
2462 2481
2463 2482 /*
2464 2483 * When reopening in reponse to a clear event, it may be due to
2465 2484 * a fmadm repair request. In this case, if the device is
2466 2485 * still broken, we want to still post the ereport again.
2467 2486 */
2468 2487 vd->vdev_forcefault = B_TRUE;
2469 2488
2470 2489 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2471 2490 vd->vdev_cant_read = B_FALSE;
2472 2491 vd->vdev_cant_write = B_FALSE;
2473 2492
2474 2493 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2475 2494
2476 2495 vd->vdev_forcefault = B_FALSE;
2477 2496
2478 2497 if (vd != rvd && vdev_writeable(vd->vdev_top))
2479 2498 vdev_state_dirty(vd->vdev_top);
2480 2499
2481 2500 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2482 2501 spa_async_request(spa, SPA_ASYNC_RESILVER);
2483 2502
2484 2503 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2485 2504 }
2486 2505
2487 2506 /*
2488 2507 * When clearing a FMA-diagnosed fault, we always want to
2489 2508 * unspare the device, as we assume that the original spare was
2490 2509 * done in response to the FMA fault.
2491 2510 */
2492 2511 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2493 2512 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2494 2513 vd->vdev_parent->vdev_child[0] == vd)
2495 2514 vd->vdev_unspare = B_TRUE;
2496 2515 }
2497 2516
2498 2517 boolean_t
2499 2518 vdev_is_dead(vdev_t *vd)
2500 2519 {
2501 2520 /*
2502 2521 * Holes and missing devices are always considered "dead".
2503 2522 * This simplifies the code since we don't have to check for
2504 2523 * these types of devices in the various code paths.
2505 2524 * Instead we rely on the fact that we skip over dead devices
2506 2525 * before issuing I/O to them.
2507 2526 */
2508 2527 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2509 2528 vd->vdev_ops == &vdev_missing_ops);
2510 2529 }
2511 2530
2512 2531 boolean_t
2513 2532 vdev_readable(vdev_t *vd)
2514 2533 {
2515 2534 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2516 2535 }
2517 2536
2518 2537 boolean_t
2519 2538 vdev_writeable(vdev_t *vd)
2520 2539 {
2521 2540 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2522 2541 }
2523 2542
2524 2543 boolean_t
2525 2544 vdev_allocatable(vdev_t *vd)
2526 2545 {
2527 2546 uint64_t state = vd->vdev_state;
2528 2547
2529 2548 /*
2530 2549 * We currently allow allocations from vdevs which may be in the
2531 2550 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2532 2551 * fails to reopen then we'll catch it later when we're holding
2533 2552 * the proper locks. Note that we have to get the vdev state
2534 2553 * in a local variable because although it changes atomically,
2535 2554 * we're asking two separate questions about it.
2536 2555 */
2537 2556 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2538 2557 !vd->vdev_cant_write && !vd->vdev_ishole);
2539 2558 }
2540 2559
2541 2560 boolean_t
2542 2561 vdev_accessible(vdev_t *vd, zio_t *zio)
2543 2562 {
2544 2563 ASSERT(zio->io_vd == vd);
2545 2564
2546 2565 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2547 2566 return (B_FALSE);
2548 2567
2549 2568 if (zio->io_type == ZIO_TYPE_READ)
2550 2569 return (!vd->vdev_cant_read);
2551 2570
2552 2571 if (zio->io_type == ZIO_TYPE_WRITE)
2553 2572 return (!vd->vdev_cant_write);
2554 2573
2555 2574 return (B_TRUE);
2556 2575 }
2557 2576
2558 2577 /*
2559 2578 * Get statistics for the given vdev.
2560 2579 */
2561 2580 void
2562 2581 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2563 2582 {
2564 2583 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2565 2584
2566 2585 mutex_enter(&vd->vdev_stat_lock);
2567 2586 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2568 2587 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2569 2588 vs->vs_state = vd->vdev_state;
2570 2589 vs->vs_rsize = vdev_get_min_asize(vd);
2571 2590 if (vd->vdev_ops->vdev_op_leaf)
2572 2591 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2573 2592 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2574 2593 mutex_exit(&vd->vdev_stat_lock);
2575 2594
2576 2595 /*
2577 2596 * If we're getting stats on the root vdev, aggregate the I/O counts
2578 2597 * over all top-level vdevs (i.e. the direct children of the root).
2579 2598 */
2580 2599 if (vd == rvd) {
2581 2600 for (int c = 0; c < rvd->vdev_children; c++) {
2582 2601 vdev_t *cvd = rvd->vdev_child[c];
2583 2602 vdev_stat_t *cvs = &cvd->vdev_stat;
2584 2603
2585 2604 mutex_enter(&vd->vdev_stat_lock);
2586 2605 for (int t = 0; t < ZIO_TYPES; t++) {
2587 2606 vs->vs_ops[t] += cvs->vs_ops[t];
2588 2607 vs->vs_bytes[t] += cvs->vs_bytes[t];
2589 2608 }
2590 2609 cvs->vs_scan_removing = cvd->vdev_removing;
2591 2610 mutex_exit(&vd->vdev_stat_lock);
2592 2611 }
2593 2612 }
2594 2613 }
2595 2614
2596 2615 void
2597 2616 vdev_clear_stats(vdev_t *vd)
2598 2617 {
2599 2618 mutex_enter(&vd->vdev_stat_lock);
2600 2619 vd->vdev_stat.vs_space = 0;
2601 2620 vd->vdev_stat.vs_dspace = 0;
2602 2621 vd->vdev_stat.vs_alloc = 0;
2603 2622 mutex_exit(&vd->vdev_stat_lock);
2604 2623 }
2605 2624
2606 2625 void
2607 2626 vdev_scan_stat_init(vdev_t *vd)
2608 2627 {
2609 2628 vdev_stat_t *vs = &vd->vdev_stat;
2610 2629
2611 2630 for (int c = 0; c < vd->vdev_children; c++)
2612 2631 vdev_scan_stat_init(vd->vdev_child[c]);
2613 2632
2614 2633 mutex_enter(&vd->vdev_stat_lock);
2615 2634 vs->vs_scan_processed = 0;
2616 2635 mutex_exit(&vd->vdev_stat_lock);
2617 2636 }
2618 2637
2619 2638 void
2620 2639 vdev_stat_update(zio_t *zio, uint64_t psize)
2621 2640 {
2622 2641 spa_t *spa = zio->io_spa;
2623 2642 vdev_t *rvd = spa->spa_root_vdev;
2624 2643 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2625 2644 vdev_t *pvd;
2626 2645 uint64_t txg = zio->io_txg;
2627 2646 vdev_stat_t *vs = &vd->vdev_stat;
2628 2647 zio_type_t type = zio->io_type;
2629 2648 int flags = zio->io_flags;
2630 2649
2631 2650 /*
2632 2651 * If this i/o is a gang leader, it didn't do any actual work.
2633 2652 */
2634 2653 if (zio->io_gang_tree)
2635 2654 return;
2636 2655
2637 2656 if (zio->io_error == 0) {
2638 2657 /*
2639 2658 * If this is a root i/o, don't count it -- we've already
2640 2659 * counted the top-level vdevs, and vdev_get_stats() will
2641 2660 * aggregate them when asked. This reduces contention on
2642 2661 * the root vdev_stat_lock and implicitly handles blocks
2643 2662 * that compress away to holes, for which there is no i/o.
2644 2663 * (Holes never create vdev children, so all the counters
2645 2664 * remain zero, which is what we want.)
2646 2665 *
2647 2666 * Note: this only applies to successful i/o (io_error == 0)
2648 2667 * because unlike i/o counts, errors are not additive.
2649 2668 * When reading a ditto block, for example, failure of
2650 2669 * one top-level vdev does not imply a root-level error.
2651 2670 */
2652 2671 if (vd == rvd)
2653 2672 return;
2654 2673
2655 2674 ASSERT(vd == zio->io_vd);
2656 2675
2657 2676 if (flags & ZIO_FLAG_IO_BYPASS)
2658 2677 return;
2659 2678
2660 2679 mutex_enter(&vd->vdev_stat_lock);
2661 2680
2662 2681 if (flags & ZIO_FLAG_IO_REPAIR) {
2663 2682 if (flags & ZIO_FLAG_SCAN_THREAD) {
2664 2683 dsl_scan_phys_t *scn_phys =
2665 2684 &spa->spa_dsl_pool->dp_scan->scn_phys;
2666 2685 uint64_t *processed = &scn_phys->scn_processed;
2667 2686
2668 2687 /* XXX cleanup? */
2669 2688 if (vd->vdev_ops->vdev_op_leaf)
2670 2689 atomic_add_64(processed, psize);
2671 2690 vs->vs_scan_processed += psize;
2672 2691 }
2673 2692
2674 2693 if (flags & ZIO_FLAG_SELF_HEAL)
2675 2694 vs->vs_self_healed += psize;
2676 2695 }
2677 2696
2678 2697 vs->vs_ops[type]++;
2679 2698 vs->vs_bytes[type] += psize;
2680 2699
2681 2700 mutex_exit(&vd->vdev_stat_lock);
2682 2701 return;
2683 2702 }
2684 2703
2685 2704 if (flags & ZIO_FLAG_SPECULATIVE)
2686 2705 return;
2687 2706
2688 2707 /*
2689 2708 * If this is an I/O error that is going to be retried, then ignore the
2690 2709 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2691 2710 * hard errors, when in reality they can happen for any number of
2692 2711 * innocuous reasons (bus resets, MPxIO link failure, etc).
2693 2712 */
2694 2713 if (zio->io_error == EIO &&
2695 2714 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2696 2715 return;
2697 2716
2698 2717 /*
2699 2718 * Intent logs writes won't propagate their error to the root
2700 2719 * I/O so don't mark these types of failures as pool-level
2701 2720 * errors.
2702 2721 */
2703 2722 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2704 2723 return;
2705 2724
2706 2725 mutex_enter(&vd->vdev_stat_lock);
2707 2726 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2708 2727 if (zio->io_error == ECKSUM)
2709 2728 vs->vs_checksum_errors++;
2710 2729 else
2711 2730 vs->vs_read_errors++;
2712 2731 }
2713 2732 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2714 2733 vs->vs_write_errors++;
2715 2734 mutex_exit(&vd->vdev_stat_lock);
2716 2735
2717 2736 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2718 2737 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2719 2738 (flags & ZIO_FLAG_SCAN_THREAD) ||
2720 2739 spa->spa_claiming)) {
2721 2740 /*
2722 2741 * This is either a normal write (not a repair), or it's
2723 2742 * a repair induced by the scrub thread, or it's a repair
2724 2743 * made by zil_claim() during spa_load() in the first txg.
2725 2744 * In the normal case, we commit the DTL change in the same
2726 2745 * txg as the block was born. In the scrub-induced repair
2727 2746 * case, we know that scrubs run in first-pass syncing context,
2728 2747 * so we commit the DTL change in spa_syncing_txg(spa).
2729 2748 * In the zil_claim() case, we commit in spa_first_txg(spa).
2730 2749 *
2731 2750 * We currently do not make DTL entries for failed spontaneous
2732 2751 * self-healing writes triggered by normal (non-scrubbing)
2733 2752 * reads, because we have no transactional context in which to
2734 2753 * do so -- and it's not clear that it'd be desirable anyway.
2735 2754 */
2736 2755 if (vd->vdev_ops->vdev_op_leaf) {
2737 2756 uint64_t commit_txg = txg;
2738 2757 if (flags & ZIO_FLAG_SCAN_THREAD) {
2739 2758 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2740 2759 ASSERT(spa_sync_pass(spa) == 1);
2741 2760 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2742 2761 commit_txg = spa_syncing_txg(spa);
2743 2762 } else if (spa->spa_claiming) {
2744 2763 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2745 2764 commit_txg = spa_first_txg(spa);
2746 2765 }
2747 2766 ASSERT(commit_txg >= spa_syncing_txg(spa));
2748 2767 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2749 2768 return;
2750 2769 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2751 2770 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2752 2771 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2753 2772 }
2754 2773 if (vd != rvd)
2755 2774 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2756 2775 }
2757 2776 }
2758 2777
2759 2778 /*
2760 2779 * Update the in-core space usage stats for this vdev, its metaslab class,
2761 2780 * and the root vdev.
2762 2781 */
2763 2782 void
2764 2783 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2765 2784 int64_t space_delta)
2766 2785 {
2767 2786 int64_t dspace_delta = space_delta;
2768 2787 spa_t *spa = vd->vdev_spa;
2769 2788 vdev_t *rvd = spa->spa_root_vdev;
2770 2789 metaslab_group_t *mg = vd->vdev_mg;
2771 2790 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2772 2791
2773 2792 ASSERT(vd == vd->vdev_top);
2774 2793
2775 2794 /*
2776 2795 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2777 2796 * factor. We must calculate this here and not at the root vdev
2778 2797 * because the root vdev's psize-to-asize is simply the max of its
2779 2798 * childrens', thus not accurate enough for us.
2780 2799 */
2781 2800 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2782 2801 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2783 2802 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2784 2803 vd->vdev_deflate_ratio;
2785 2804
2786 2805 mutex_enter(&vd->vdev_stat_lock);
2787 2806 vd->vdev_stat.vs_alloc += alloc_delta;
2788 2807 vd->vdev_stat.vs_space += space_delta;
2789 2808 vd->vdev_stat.vs_dspace += dspace_delta;
2790 2809 mutex_exit(&vd->vdev_stat_lock);
2791 2810
2792 2811 if (mc == spa_normal_class(spa)) {
2793 2812 mutex_enter(&rvd->vdev_stat_lock);
2794 2813 rvd->vdev_stat.vs_alloc += alloc_delta;
2795 2814 rvd->vdev_stat.vs_space += space_delta;
2796 2815 rvd->vdev_stat.vs_dspace += dspace_delta;
2797 2816 mutex_exit(&rvd->vdev_stat_lock);
2798 2817 }
2799 2818
2800 2819 if (mc != NULL) {
2801 2820 ASSERT(rvd == vd->vdev_parent);
2802 2821 ASSERT(vd->vdev_ms_count != 0);
2803 2822
2804 2823 metaslab_class_space_update(mc,
2805 2824 alloc_delta, defer_delta, space_delta, dspace_delta);
2806 2825 }
2807 2826 }
2808 2827
2809 2828 /*
2810 2829 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2811 2830 * so that it will be written out next time the vdev configuration is synced.
2812 2831 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2813 2832 */
2814 2833 void
2815 2834 vdev_config_dirty(vdev_t *vd)
2816 2835 {
2817 2836 spa_t *spa = vd->vdev_spa;
2818 2837 vdev_t *rvd = spa->spa_root_vdev;
2819 2838 int c;
2820 2839
2821 2840 ASSERT(spa_writeable(spa));
2822 2841
2823 2842 /*
2824 2843 * If this is an aux vdev (as with l2cache and spare devices), then we
2825 2844 * update the vdev config manually and set the sync flag.
2826 2845 */
2827 2846 if (vd->vdev_aux != NULL) {
2828 2847 spa_aux_vdev_t *sav = vd->vdev_aux;
2829 2848 nvlist_t **aux;
2830 2849 uint_t naux;
2831 2850
2832 2851 for (c = 0; c < sav->sav_count; c++) {
2833 2852 if (sav->sav_vdevs[c] == vd)
2834 2853 break;
2835 2854 }
2836 2855
2837 2856 if (c == sav->sav_count) {
2838 2857 /*
2839 2858 * We're being removed. There's nothing more to do.
2840 2859 */
2841 2860 ASSERT(sav->sav_sync == B_TRUE);
2842 2861 return;
2843 2862 }
2844 2863
2845 2864 sav->sav_sync = B_TRUE;
2846 2865
2847 2866 if (nvlist_lookup_nvlist_array(sav->sav_config,
2848 2867 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2849 2868 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2850 2869 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2851 2870 }
2852 2871
2853 2872 ASSERT(c < naux);
2854 2873
2855 2874 /*
2856 2875 * Setting the nvlist in the middle if the array is a little
2857 2876 * sketchy, but it will work.
2858 2877 */
2859 2878 nvlist_free(aux[c]);
2860 2879 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2861 2880
2862 2881 return;
2863 2882 }
2864 2883
2865 2884 /*
2866 2885 * The dirty list is protected by the SCL_CONFIG lock. The caller
2867 2886 * must either hold SCL_CONFIG as writer, or must be the sync thread
2868 2887 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2869 2888 * so this is sufficient to ensure mutual exclusion.
2870 2889 */
2871 2890 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2872 2891 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2873 2892 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2874 2893
2875 2894 if (vd == rvd) {
2876 2895 for (c = 0; c < rvd->vdev_children; c++)
2877 2896 vdev_config_dirty(rvd->vdev_child[c]);
2878 2897 } else {
2879 2898 ASSERT(vd == vd->vdev_top);
2880 2899
2881 2900 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2882 2901 !vd->vdev_ishole)
2883 2902 list_insert_head(&spa->spa_config_dirty_list, vd);
2884 2903 }
2885 2904 }
2886 2905
2887 2906 void
2888 2907 vdev_config_clean(vdev_t *vd)
2889 2908 {
2890 2909 spa_t *spa = vd->vdev_spa;
2891 2910
2892 2911 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2893 2912 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2894 2913 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2895 2914
2896 2915 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2897 2916 list_remove(&spa->spa_config_dirty_list, vd);
2898 2917 }
2899 2918
2900 2919 /*
2901 2920 * Mark a top-level vdev's state as dirty, so that the next pass of
2902 2921 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2903 2922 * the state changes from larger config changes because they require
2904 2923 * much less locking, and are often needed for administrative actions.
2905 2924 */
2906 2925 void
2907 2926 vdev_state_dirty(vdev_t *vd)
2908 2927 {
2909 2928 spa_t *spa = vd->vdev_spa;
2910 2929
2911 2930 ASSERT(spa_writeable(spa));
2912 2931 ASSERT(vd == vd->vdev_top);
2913 2932
2914 2933 /*
2915 2934 * The state list is protected by the SCL_STATE lock. The caller
2916 2935 * must either hold SCL_STATE as writer, or must be the sync thread
2917 2936 * (which holds SCL_STATE as reader). There's only one sync thread,
2918 2937 * so this is sufficient to ensure mutual exclusion.
2919 2938 */
2920 2939 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2921 2940 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2922 2941 spa_config_held(spa, SCL_STATE, RW_READER)));
2923 2942
2924 2943 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2925 2944 list_insert_head(&spa->spa_state_dirty_list, vd);
2926 2945 }
2927 2946
2928 2947 void
2929 2948 vdev_state_clean(vdev_t *vd)
2930 2949 {
2931 2950 spa_t *spa = vd->vdev_spa;
2932 2951
2933 2952 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2934 2953 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2935 2954 spa_config_held(spa, SCL_STATE, RW_READER)));
2936 2955
2937 2956 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2938 2957 list_remove(&spa->spa_state_dirty_list, vd);
2939 2958 }
2940 2959
2941 2960 /*
2942 2961 * Propagate vdev state up from children to parent.
2943 2962 */
2944 2963 void
2945 2964 vdev_propagate_state(vdev_t *vd)
2946 2965 {
2947 2966 spa_t *spa = vd->vdev_spa;
2948 2967 vdev_t *rvd = spa->spa_root_vdev;
2949 2968 int degraded = 0, faulted = 0;
2950 2969 int corrupted = 0;
2951 2970 vdev_t *child;
2952 2971
2953 2972 if (vd->vdev_children > 0) {
2954 2973 for (int c = 0; c < vd->vdev_children; c++) {
2955 2974 child = vd->vdev_child[c];
2956 2975
2957 2976 /*
2958 2977 * Don't factor holes into the decision.
2959 2978 */
2960 2979 if (child->vdev_ishole)
2961 2980 continue;
2962 2981
2963 2982 if (!vdev_readable(child) ||
2964 2983 (!vdev_writeable(child) && spa_writeable(spa))) {
2965 2984 /*
2966 2985 * Root special: if there is a top-level log
2967 2986 * device, treat the root vdev as if it were
2968 2987 * degraded.
2969 2988 */
2970 2989 if (child->vdev_islog && vd == rvd)
2971 2990 degraded++;
2972 2991 else
2973 2992 faulted++;
2974 2993 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2975 2994 degraded++;
2976 2995 }
2977 2996
2978 2997 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2979 2998 corrupted++;
2980 2999 }
2981 3000
2982 3001 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2983 3002
2984 3003 /*
2985 3004 * Root special: if there is a top-level vdev that cannot be
2986 3005 * opened due to corrupted metadata, then propagate the root
2987 3006 * vdev's aux state as 'corrupt' rather than 'insufficient
2988 3007 * replicas'.
2989 3008 */
2990 3009 if (corrupted && vd == rvd &&
2991 3010 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2992 3011 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2993 3012 VDEV_AUX_CORRUPT_DATA);
2994 3013 }
2995 3014
2996 3015 if (vd->vdev_parent)
2997 3016 vdev_propagate_state(vd->vdev_parent);
2998 3017 }
2999 3018
3000 3019 /*
3001 3020 * Set a vdev's state. If this is during an open, we don't update the parent
3002 3021 * state, because we're in the process of opening children depth-first.
3003 3022 * Otherwise, we propagate the change to the parent.
3004 3023 *
3005 3024 * If this routine places a device in a faulted state, an appropriate ereport is
3006 3025 * generated.
3007 3026 */
3008 3027 void
3009 3028 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3010 3029 {
3011 3030 uint64_t save_state;
3012 3031 spa_t *spa = vd->vdev_spa;
3013 3032
3014 3033 if (state == vd->vdev_state) {
3015 3034 vd->vdev_stat.vs_aux = aux;
3016 3035 return;
3017 3036 }
3018 3037
3019 3038 save_state = vd->vdev_state;
3020 3039
3021 3040 vd->vdev_state = state;
3022 3041 vd->vdev_stat.vs_aux = aux;
3023 3042
3024 3043 /*
3025 3044 * If we are setting the vdev state to anything but an open state, then
3026 3045 * always close the underlying device unless the device has requested
3027 3046 * a delayed close (i.e. we're about to remove or fault the device).
3028 3047 * Otherwise, we keep accessible but invalid devices open forever.
3029 3048 * We don't call vdev_close() itself, because that implies some extra
3030 3049 * checks (offline, etc) that we don't want here. This is limited to
3031 3050 * leaf devices, because otherwise closing the device will affect other
3032 3051 * children.
3033 3052 */
3034 3053 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3035 3054 vd->vdev_ops->vdev_op_leaf)
3036 3055 vd->vdev_ops->vdev_op_close(vd);
3037 3056
3038 3057 /*
3039 3058 * If we have brought this vdev back into service, we need
3040 3059 * to notify fmd so that it can gracefully repair any outstanding
3041 3060 * cases due to a missing device. We do this in all cases, even those
3042 3061 * that probably don't correlate to a repaired fault. This is sure to
3043 3062 * catch all cases, and we let the zfs-retire agent sort it out. If
3044 3063 * this is a transient state it's OK, as the retire agent will
3045 3064 * double-check the state of the vdev before repairing it.
3046 3065 */
3047 3066 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3048 3067 vd->vdev_prevstate != state)
3049 3068 zfs_post_state_change(spa, vd);
3050 3069
3051 3070 if (vd->vdev_removed &&
3052 3071 state == VDEV_STATE_CANT_OPEN &&
3053 3072 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3054 3073 /*
3055 3074 * If the previous state is set to VDEV_STATE_REMOVED, then this
3056 3075 * device was previously marked removed and someone attempted to
3057 3076 * reopen it. If this failed due to a nonexistent device, then
3058 3077 * keep the device in the REMOVED state. We also let this be if
3059 3078 * it is one of our special test online cases, which is only
3060 3079 * attempting to online the device and shouldn't generate an FMA
3061 3080 * fault.
3062 3081 */
3063 3082 vd->vdev_state = VDEV_STATE_REMOVED;
3064 3083 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3065 3084 } else if (state == VDEV_STATE_REMOVED) {
3066 3085 vd->vdev_removed = B_TRUE;
3067 3086 } else if (state == VDEV_STATE_CANT_OPEN) {
3068 3087 /*
3069 3088 * If we fail to open a vdev during an import or recovery, we
3070 3089 * mark it as "not available", which signifies that it was
3071 3090 * never there to begin with. Failure to open such a device
3072 3091 * is not considered an error.
3073 3092 */
3074 3093 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3075 3094 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3076 3095 vd->vdev_ops->vdev_op_leaf)
3077 3096 vd->vdev_not_present = 1;
3078 3097
3079 3098 /*
3080 3099 * Post the appropriate ereport. If the 'prevstate' field is
3081 3100 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3082 3101 * that this is part of a vdev_reopen(). In this case, we don't
3083 3102 * want to post the ereport if the device was already in the
3084 3103 * CANT_OPEN state beforehand.
3085 3104 *
3086 3105 * If the 'checkremove' flag is set, then this is an attempt to
3087 3106 * online the device in response to an insertion event. If we
3088 3107 * hit this case, then we have detected an insertion event for a
3089 3108 * faulted or offline device that wasn't in the removed state.
3090 3109 * In this scenario, we don't post an ereport because we are
3091 3110 * about to replace the device, or attempt an online with
3092 3111 * vdev_forcefault, which will generate the fault for us.
3093 3112 */
3094 3113 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3095 3114 !vd->vdev_not_present && !vd->vdev_checkremove &&
3096 3115 vd != spa->spa_root_vdev) {
3097 3116 const char *class;
3098 3117
3099 3118 switch (aux) {
3100 3119 case VDEV_AUX_OPEN_FAILED:
3101 3120 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3102 3121 break;
3103 3122 case VDEV_AUX_CORRUPT_DATA:
3104 3123 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3105 3124 break;
3106 3125 case VDEV_AUX_NO_REPLICAS:
3107 3126 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3108 3127 break;
3109 3128 case VDEV_AUX_BAD_GUID_SUM:
3110 3129 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3111 3130 break;
3112 3131 case VDEV_AUX_TOO_SMALL:
3113 3132 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3114 3133 break;
3115 3134 case VDEV_AUX_BAD_LABEL:
3116 3135 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3117 3136 break;
3118 3137 default:
3119 3138 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3120 3139 }
3121 3140
3122 3141 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3123 3142 }
3124 3143
3125 3144 /* Erase any notion of persistent removed state */
3126 3145 vd->vdev_removed = B_FALSE;
3127 3146 } else {
3128 3147 vd->vdev_removed = B_FALSE;
3129 3148 }
3130 3149
3131 3150 if (!isopen && vd->vdev_parent)
3132 3151 vdev_propagate_state(vd->vdev_parent);
3133 3152 }
3134 3153
3135 3154 /*
3136 3155 * Check the vdev configuration to ensure that it's capable of supporting
3137 3156 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3138 3157 * In addition, only a single top-level vdev is allowed and none of the leaves
3139 3158 * can be wholedisks.
3140 3159 */
3141 3160 boolean_t
3142 3161 vdev_is_bootable(vdev_t *vd)
3143 3162 {
3144 3163 if (!vd->vdev_ops->vdev_op_leaf) {
3145 3164 char *vdev_type = vd->vdev_ops->vdev_op_type;
3146 3165
3147 3166 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3148 3167 vd->vdev_children > 1) {
3149 3168 return (B_FALSE);
3150 3169 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3151 3170 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3152 3171 return (B_FALSE);
3153 3172 }
3154 3173 } else if (vd->vdev_wholedisk == 1) {
3155 3174 return (B_FALSE);
3156 3175 }
3157 3176
3158 3177 for (int c = 0; c < vd->vdev_children; c++) {
3159 3178 if (!vdev_is_bootable(vd->vdev_child[c]))
3160 3179 return (B_FALSE);
3161 3180 }
3162 3181 return (B_TRUE);
3163 3182 }
3164 3183
3165 3184 /*
3166 3185 * Load the state from the original vdev tree (ovd) which
3167 3186 * we've retrieved from the MOS config object. If the original
3168 3187 * vdev was offline or faulted then we transfer that state to the
3169 3188 * device in the current vdev tree (nvd).
3170 3189 */
3171 3190 void
3172 3191 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3173 3192 {
3174 3193 spa_t *spa = nvd->vdev_spa;
3175 3194
3176 3195 ASSERT(nvd->vdev_top->vdev_islog);
3177 3196 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3178 3197 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3179 3198
3180 3199 for (int c = 0; c < nvd->vdev_children; c++)
3181 3200 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3182 3201
3183 3202 if (nvd->vdev_ops->vdev_op_leaf) {
3184 3203 /*
3185 3204 * Restore the persistent vdev state
3186 3205 */
3187 3206 nvd->vdev_offline = ovd->vdev_offline;
3188 3207 nvd->vdev_faulted = ovd->vdev_faulted;
3189 3208 nvd->vdev_degraded = ovd->vdev_degraded;
3190 3209 nvd->vdev_removed = ovd->vdev_removed;
3191 3210 }
3192 3211 }
3193 3212
3194 3213 /*
3195 3214 * Determine if a log device has valid content. If the vdev was
3196 3215 * removed or faulted in the MOS config then we know that
3197 3216 * the content on the log device has already been written to the pool.
3198 3217 */
3199 3218 boolean_t
3200 3219 vdev_log_state_valid(vdev_t *vd)
3201 3220 {
3202 3221 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3203 3222 !vd->vdev_removed)
3204 3223 return (B_TRUE);
3205 3224
3206 3225 for (int c = 0; c < vd->vdev_children; c++)
3207 3226 if (vdev_log_state_valid(vd->vdev_child[c]))
3208 3227 return (B_TRUE);
3209 3228
3210 3229 return (B_FALSE);
3211 3230 }
3212 3231
3213 3232 /*
3214 3233 * Expand a vdev if possible.
3215 3234 */
3216 3235 void
3217 3236 vdev_expand(vdev_t *vd, uint64_t txg)
3218 3237 {
3219 3238 ASSERT(vd->vdev_top == vd);
3220 3239 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3221 3240
3222 3241 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3223 3242 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3224 3243 vdev_config_dirty(vd);
3225 3244 }
3226 3245 }
3227 3246
3228 3247 /*
3229 3248 * Split a vdev.
3230 3249 */
3231 3250 void
3232 3251 vdev_split(vdev_t *vd)
3233 3252 {
3234 3253 vdev_t *cvd, *pvd = vd->vdev_parent;
3235 3254
3236 3255 vdev_remove_child(pvd, vd);
3237 3256 vdev_compact_children(pvd);
3238 3257
3239 3258 cvd = pvd->vdev_child[0];
3240 3259 if (pvd->vdev_children == 1) {
3241 3260 vdev_remove_parent(cvd);
3242 3261 cvd->vdev_splitting = B_TRUE;
3243 3262 }
3244 3263 vdev_propagate_state(cvd);
3245 3264 }
3246 3265
3247 3266 void
3248 3267 vdev_deadman(vdev_t *vd)
3249 3268 {
3250 3269 for (int c = 0; c < vd->vdev_children; c++) {
3251 3270 vdev_t *cvd = vd->vdev_child[c];
3252 3271
3253 3272 vdev_deadman(cvd);
3254 3273 }
3255 3274
3256 3275 if (vd->vdev_ops->vdev_op_leaf) {
3257 3276 vdev_queue_t *vq = &vd->vdev_queue;
3258 3277
3259 3278 mutex_enter(&vq->vq_lock);
3260 3279 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3261 3280 spa_t *spa = vd->vdev_spa;
3262 3281 zio_t *fio;
3263 3282 uint64_t delta;
3264 3283
3265 3284 /*
3266 3285 * Look at the head of all the pending queues,
3267 3286 * if any I/O has been outstanding for longer than
3268 3287 * the spa_deadman_synctime we panic the system.
3269 3288 */
3270 3289 fio = avl_first(&vq->vq_active_tree);
3271 3290 delta = gethrtime() - fio->io_timestamp;
3272 3291 if (delta > spa_deadman_synctime(spa)) {
3273 3292 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3274 3293 "delta %lluns, last io %lluns",
3275 3294 fio->io_timestamp, delta,
3276 3295 vq->vq_io_complete_ts);
3277 3296 fm_panic("I/O to pool '%s' appears to be "
3278 3297 "hung.", spa_name(spa));
3279 3298 }
3280 3299 }
3281 3300 mutex_exit(&vq->vq_lock);
3282 3301 }
3283 3302 }
|
↓ open down ↓ |
1156 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX