Print this page
3742 zfs comments need cleaner, more consistent style
Submitted by: Will Andrews <willa@spectralogic.com>
Submitted by: Alan Somers <alans@spectralogic.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@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>
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 39 #include <sys/zio.h>
40 40 #include <sys/zap.h>
41 41 #include <sys/fs/zfs.h>
42 42 #include <sys/arc.h>
43 43 #include <sys/zil.h>
44 44 #include <sys/dsl_scan.h>
45 45
46 46 /*
47 47 * Virtual device management.
48 48 */
49 49
50 50 static vdev_ops_t *vdev_ops_table[] = {
51 51 &vdev_root_ops,
52 52 &vdev_raidz_ops,
53 53 &vdev_mirror_ops,
54 54 &vdev_replacing_ops,
55 55 &vdev_spare_ops,
56 56 &vdev_disk_ops,
57 57 &vdev_file_ops,
58 58 &vdev_missing_ops,
59 59 &vdev_hole_ops,
60 60 NULL
61 61 };
62 62
63 63 /* maximum scrub/resilver I/O queue per leaf vdev */
64 64 int zfs_scrub_limit = 10;
65 65
66 66 /*
67 67 * Given a vdev type, return the appropriate ops vector.
68 68 */
69 69 static vdev_ops_t *
70 70 vdev_getops(const char *type)
71 71 {
72 72 vdev_ops_t *ops, **opspp;
73 73
74 74 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
75 75 if (strcmp(ops->vdev_op_type, type) == 0)
76 76 break;
77 77
78 78 return (ops);
79 79 }
80 80
81 81 /*
82 82 * Default asize function: return the MAX of psize with the asize of
83 83 * all children. This is what's used by anything other than RAID-Z.
84 84 */
85 85 uint64_t
86 86 vdev_default_asize(vdev_t *vd, uint64_t psize)
87 87 {
88 88 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
89 89 uint64_t csize;
90 90
91 91 for (int c = 0; c < vd->vdev_children; c++) {
92 92 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
93 93 asize = MAX(asize, csize);
94 94 }
95 95
96 96 return (asize);
97 97 }
98 98
99 99 /*
100 100 * Get the minimum allocatable size. We define the allocatable size as
101 101 * the vdev's asize rounded to the nearest metaslab. This allows us to
102 102 * replace or attach devices which don't have the same physical size but
103 103 * can still satisfy the same number of allocations.
104 104 */
105 105 uint64_t
106 106 vdev_get_min_asize(vdev_t *vd)
107 107 {
108 108 vdev_t *pvd = vd->vdev_parent;
109 109
110 110 /*
111 111 * If our parent is NULL (inactive spare or cache) or is the root,
112 112 * just return our own asize.
113 113 */
114 114 if (pvd == NULL)
115 115 return (vd->vdev_asize);
116 116
117 117 /*
118 118 * The top-level vdev just returns the allocatable size rounded
119 119 * to the nearest metaslab.
120 120 */
121 121 if (vd == vd->vdev_top)
122 122 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
123 123
124 124 /*
125 125 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
126 126 * so each child must provide at least 1/Nth of its asize.
127 127 */
128 128 if (pvd->vdev_ops == &vdev_raidz_ops)
129 129 return (pvd->vdev_min_asize / pvd->vdev_children);
130 130
131 131 return (pvd->vdev_min_asize);
132 132 }
133 133
134 134 void
135 135 vdev_set_min_asize(vdev_t *vd)
136 136 {
137 137 vd->vdev_min_asize = vdev_get_min_asize(vd);
138 138
139 139 for (int c = 0; c < vd->vdev_children; c++)
140 140 vdev_set_min_asize(vd->vdev_child[c]);
141 141 }
142 142
143 143 vdev_t *
144 144 vdev_lookup_top(spa_t *spa, uint64_t vdev)
145 145 {
146 146 vdev_t *rvd = spa->spa_root_vdev;
147 147
148 148 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
149 149
150 150 if (vdev < rvd->vdev_children) {
151 151 ASSERT(rvd->vdev_child[vdev] != NULL);
152 152 return (rvd->vdev_child[vdev]);
153 153 }
154 154
155 155 return (NULL);
156 156 }
157 157
158 158 vdev_t *
159 159 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
160 160 {
161 161 vdev_t *mvd;
162 162
163 163 if (vd->vdev_guid == guid)
164 164 return (vd);
165 165
166 166 for (int c = 0; c < vd->vdev_children; c++)
167 167 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
168 168 NULL)
169 169 return (mvd);
170 170
171 171 return (NULL);
172 172 }
173 173
174 174 void
175 175 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
176 176 {
177 177 size_t oldsize, newsize;
178 178 uint64_t id = cvd->vdev_id;
179 179 vdev_t **newchild;
180 180
181 181 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
182 182 ASSERT(cvd->vdev_parent == NULL);
183 183
184 184 cvd->vdev_parent = pvd;
185 185
186 186 if (pvd == NULL)
187 187 return;
188 188
189 189 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
190 190
191 191 oldsize = pvd->vdev_children * sizeof (vdev_t *);
192 192 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
193 193 newsize = pvd->vdev_children * sizeof (vdev_t *);
194 194
195 195 newchild = kmem_zalloc(newsize, KM_SLEEP);
196 196 if (pvd->vdev_child != NULL) {
197 197 bcopy(pvd->vdev_child, newchild, oldsize);
198 198 kmem_free(pvd->vdev_child, oldsize);
199 199 }
200 200
201 201 pvd->vdev_child = newchild;
202 202 pvd->vdev_child[id] = cvd;
203 203
204 204 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
205 205 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
206 206
207 207 /*
208 208 * Walk up all ancestors to update guid sum.
209 209 */
210 210 for (; pvd != NULL; pvd = pvd->vdev_parent)
211 211 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
212 212 }
213 213
214 214 void
215 215 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
216 216 {
217 217 int c;
218 218 uint_t id = cvd->vdev_id;
219 219
220 220 ASSERT(cvd->vdev_parent == pvd);
221 221
222 222 if (pvd == NULL)
223 223 return;
224 224
225 225 ASSERT(id < pvd->vdev_children);
226 226 ASSERT(pvd->vdev_child[id] == cvd);
227 227
228 228 pvd->vdev_child[id] = NULL;
229 229 cvd->vdev_parent = NULL;
230 230
231 231 for (c = 0; c < pvd->vdev_children; c++)
232 232 if (pvd->vdev_child[c])
233 233 break;
234 234
235 235 if (c == pvd->vdev_children) {
236 236 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
237 237 pvd->vdev_child = NULL;
238 238 pvd->vdev_children = 0;
239 239 }
240 240
241 241 /*
242 242 * Walk up all ancestors to update guid sum.
243 243 */
244 244 for (; pvd != NULL; pvd = pvd->vdev_parent)
245 245 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
246 246 }
247 247
248 248 /*
249 249 * Remove any holes in the child array.
250 250 */
251 251 void
252 252 vdev_compact_children(vdev_t *pvd)
253 253 {
254 254 vdev_t **newchild, *cvd;
255 255 int oldc = pvd->vdev_children;
256 256 int newc;
257 257
258 258 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
259 259
260 260 for (int c = newc = 0; c < oldc; c++)
261 261 if (pvd->vdev_child[c])
262 262 newc++;
263 263
264 264 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
265 265
266 266 for (int c = newc = 0; c < oldc; c++) {
267 267 if ((cvd = pvd->vdev_child[c]) != NULL) {
268 268 newchild[newc] = cvd;
269 269 cvd->vdev_id = newc++;
270 270 }
271 271 }
272 272
273 273 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
274 274 pvd->vdev_child = newchild;
275 275 pvd->vdev_children = newc;
276 276 }
277 277
278 278 /*
279 279 * Allocate and minimally initialize a vdev_t.
280 280 */
281 281 vdev_t *
282 282 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
283 283 {
284 284 vdev_t *vd;
285 285
286 286 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
287 287
288 288 if (spa->spa_root_vdev == NULL) {
289 289 ASSERT(ops == &vdev_root_ops);
290 290 spa->spa_root_vdev = vd;
291 291 spa->spa_load_guid = spa_generate_guid(NULL);
292 292 }
293 293
294 294 if (guid == 0 && ops != &vdev_hole_ops) {
295 295 if (spa->spa_root_vdev == vd) {
296 296 /*
297 297 * The root vdev's guid will also be the pool guid,
298 298 * which must be unique among all pools.
299 299 */
300 300 guid = spa_generate_guid(NULL);
301 301 } else {
302 302 /*
303 303 * Any other vdev's guid must be unique within the pool.
304 304 */
305 305 guid = spa_generate_guid(spa);
306 306 }
307 307 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
308 308 }
309 309
310 310 vd->vdev_spa = spa;
311 311 vd->vdev_id = id;
312 312 vd->vdev_guid = guid;
313 313 vd->vdev_guid_sum = guid;
314 314 vd->vdev_ops = ops;
315 315 vd->vdev_state = VDEV_STATE_CLOSED;
316 316 vd->vdev_ishole = (ops == &vdev_hole_ops);
317 317
318 318 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
319 319 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
320 320 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
321 321 for (int t = 0; t < DTL_TYPES; t++) {
322 322 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
323 323 &vd->vdev_dtl_lock);
324 324 }
325 325 txg_list_create(&vd->vdev_ms_list,
326 326 offsetof(struct metaslab, ms_txg_node));
327 327 txg_list_create(&vd->vdev_dtl_list,
328 328 offsetof(struct vdev, vdev_dtl_node));
329 329 vd->vdev_stat.vs_timestamp = gethrtime();
330 330 vdev_queue_init(vd);
331 331 vdev_cache_init(vd);
332 332
333 333 return (vd);
334 334 }
335 335
336 336 /*
337 337 * Allocate a new vdev. The 'alloctype' is used to control whether we are
338 338 * creating a new vdev or loading an existing one - the behavior is slightly
339 339 * different for each case.
340 340 */
341 341 int
342 342 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
343 343 int alloctype)
344 344 {
345 345 vdev_ops_t *ops;
346 346 char *type;
347 347 uint64_t guid = 0, islog, nparity;
348 348 vdev_t *vd;
349 349
350 350 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
351 351
352 352 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
353 353 return (SET_ERROR(EINVAL));
354 354
355 355 if ((ops = vdev_getops(type)) == NULL)
356 356 return (SET_ERROR(EINVAL));
357 357
358 358 /*
359 359 * If this is a load, get the vdev guid from the nvlist.
360 360 * Otherwise, vdev_alloc_common() will generate one for us.
361 361 */
362 362 if (alloctype == VDEV_ALLOC_LOAD) {
363 363 uint64_t label_id;
364 364
365 365 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
366 366 label_id != id)
367 367 return (SET_ERROR(EINVAL));
368 368
369 369 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
370 370 return (SET_ERROR(EINVAL));
371 371 } else if (alloctype == VDEV_ALLOC_SPARE) {
372 372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
373 373 return (SET_ERROR(EINVAL));
374 374 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
375 375 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
376 376 return (SET_ERROR(EINVAL));
377 377 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
378 378 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 379 return (SET_ERROR(EINVAL));
380 380 }
381 381
382 382 /*
383 383 * The first allocated vdev must be of type 'root'.
384 384 */
385 385 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
386 386 return (SET_ERROR(EINVAL));
387 387
388 388 /*
389 389 * Determine whether we're a log vdev.
390 390 */
391 391 islog = 0;
392 392 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
393 393 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
394 394 return (SET_ERROR(ENOTSUP));
395 395
396 396 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
397 397 return (SET_ERROR(ENOTSUP));
398 398
399 399 /*
400 400 * Set the nparity property for RAID-Z vdevs.
401 401 */
402 402 nparity = -1ULL;
403 403 if (ops == &vdev_raidz_ops) {
404 404 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
405 405 &nparity) == 0) {
406 406 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
407 407 return (SET_ERROR(EINVAL));
408 408 /*
409 409 * Previous versions could only support 1 or 2 parity
410 410 * device.
411 411 */
412 412 if (nparity > 1 &&
413 413 spa_version(spa) < SPA_VERSION_RAIDZ2)
414 414 return (SET_ERROR(ENOTSUP));
415 415 if (nparity > 2 &&
416 416 spa_version(spa) < SPA_VERSION_RAIDZ3)
417 417 return (SET_ERROR(ENOTSUP));
418 418 } else {
419 419 /*
420 420 * We require the parity to be specified for SPAs that
421 421 * support multiple parity levels.
422 422 */
423 423 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
424 424 return (SET_ERROR(EINVAL));
425 425 /*
426 426 * Otherwise, we default to 1 parity device for RAID-Z.
427 427 */
428 428 nparity = 1;
429 429 }
430 430 } else {
431 431 nparity = 0;
432 432 }
433 433 ASSERT(nparity != -1ULL);
434 434
435 435 vd = vdev_alloc_common(spa, id, guid, ops);
436 436
437 437 vd->vdev_islog = islog;
438 438 vd->vdev_nparity = nparity;
439 439
440 440 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
441 441 vd->vdev_path = spa_strdup(vd->vdev_path);
442 442 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
443 443 vd->vdev_devid = spa_strdup(vd->vdev_devid);
444 444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
445 445 &vd->vdev_physpath) == 0)
446 446 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
447 447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
448 448 vd->vdev_fru = spa_strdup(vd->vdev_fru);
449 449
450 450 /*
451 451 * Set the whole_disk property. If it's not specified, leave the value
452 452 * as -1.
453 453 */
454 454 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
455 455 &vd->vdev_wholedisk) != 0)
456 456 vd->vdev_wholedisk = -1ULL;
457 457
458 458 /*
459 459 * Look for the 'not present' flag. This will only be set if the device
460 460 * was not present at the time of import.
461 461 */
462 462 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
463 463 &vd->vdev_not_present);
464 464
465 465 /*
466 466 * Get the alignment requirement.
467 467 */
468 468 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
469 469
470 470 /*
471 471 * Retrieve the vdev creation time.
472 472 */
473 473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
474 474 &vd->vdev_crtxg);
475 475
476 476 /*
477 477 * If we're a top-level vdev, try to load the allocation parameters.
478 478 */
479 479 if (parent && !parent->vdev_parent &&
480 480 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
481 481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
482 482 &vd->vdev_ms_array);
483 483 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
484 484 &vd->vdev_ms_shift);
485 485 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
486 486 &vd->vdev_asize);
487 487 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
488 488 &vd->vdev_removing);
489 489 }
490 490
491 491 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
492 492 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
493 493 alloctype == VDEV_ALLOC_ADD ||
494 494 alloctype == VDEV_ALLOC_SPLIT ||
495 495 alloctype == VDEV_ALLOC_ROOTPOOL);
496 496 vd->vdev_mg = metaslab_group_create(islog ?
497 497 spa_log_class(spa) : spa_normal_class(spa), vd);
498 498 }
499 499
500 500 /*
501 501 * If we're a leaf vdev, try to load the DTL object and other state.
502 502 */
503 503 if (vd->vdev_ops->vdev_op_leaf &&
504 504 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
505 505 alloctype == VDEV_ALLOC_ROOTPOOL)) {
506 506 if (alloctype == VDEV_ALLOC_LOAD) {
507 507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
508 508 &vd->vdev_dtl_smo.smo_object);
509 509 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
510 510 &vd->vdev_unspare);
511 511 }
512 512
513 513 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
514 514 uint64_t spare = 0;
515 515
516 516 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
517 517 &spare) == 0 && spare)
518 518 spa_spare_add(vd);
519 519 }
520 520
521 521 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
522 522 &vd->vdev_offline);
523 523
524 524 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
525 525 &vd->vdev_resilvering);
526 526
527 527 /*
528 528 * When importing a pool, we want to ignore the persistent fault
529 529 * state, as the diagnosis made on another system may not be
530 530 * valid in the current context. Local vdevs will
531 531 * remain in the faulted state.
532 532 */
533 533 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
534 534 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
535 535 &vd->vdev_faulted);
536 536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
537 537 &vd->vdev_degraded);
538 538 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
539 539 &vd->vdev_removed);
540 540
541 541 if (vd->vdev_faulted || vd->vdev_degraded) {
542 542 char *aux;
543 543
544 544 vd->vdev_label_aux =
545 545 VDEV_AUX_ERR_EXCEEDED;
546 546 if (nvlist_lookup_string(nv,
547 547 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
548 548 strcmp(aux, "external") == 0)
549 549 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
550 550 }
551 551 }
552 552 }
553 553
554 554 /*
555 555 * Add ourselves to the parent's list of children.
556 556 */
557 557 vdev_add_child(parent, vd);
558 558
559 559 *vdp = vd;
560 560
561 561 return (0);
562 562 }
563 563
564 564 void
565 565 vdev_free(vdev_t *vd)
566 566 {
567 567 spa_t *spa = vd->vdev_spa;
568 568
569 569 /*
570 570 * vdev_free() implies closing the vdev first. This is simpler than
571 571 * trying to ensure complicated semantics for all callers.
572 572 */
573 573 vdev_close(vd);
574 574
575 575 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
576 576 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
577 577
578 578 /*
579 579 * Free all children.
580 580 */
581 581 for (int c = 0; c < vd->vdev_children; c++)
582 582 vdev_free(vd->vdev_child[c]);
583 583
584 584 ASSERT(vd->vdev_child == NULL);
585 585 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
586 586
587 587 /*
588 588 * Discard allocation state.
589 589 */
590 590 if (vd->vdev_mg != NULL) {
591 591 vdev_metaslab_fini(vd);
592 592 metaslab_group_destroy(vd->vdev_mg);
593 593 }
594 594
595 595 ASSERT0(vd->vdev_stat.vs_space);
596 596 ASSERT0(vd->vdev_stat.vs_dspace);
597 597 ASSERT0(vd->vdev_stat.vs_alloc);
598 598
599 599 /*
600 600 * Remove this vdev from its parent's child list.
601 601 */
602 602 vdev_remove_child(vd->vdev_parent, vd);
603 603
604 604 ASSERT(vd->vdev_parent == NULL);
605 605
606 606 /*
607 607 * Clean up vdev structure.
608 608 */
609 609 vdev_queue_fini(vd);
610 610 vdev_cache_fini(vd);
611 611
612 612 if (vd->vdev_path)
613 613 spa_strfree(vd->vdev_path);
614 614 if (vd->vdev_devid)
615 615 spa_strfree(vd->vdev_devid);
616 616 if (vd->vdev_physpath)
617 617 spa_strfree(vd->vdev_physpath);
618 618 if (vd->vdev_fru)
619 619 spa_strfree(vd->vdev_fru);
620 620
621 621 if (vd->vdev_isspare)
622 622 spa_spare_remove(vd);
623 623 if (vd->vdev_isl2cache)
624 624 spa_l2cache_remove(vd);
625 625
626 626 txg_list_destroy(&vd->vdev_ms_list);
627 627 txg_list_destroy(&vd->vdev_dtl_list);
628 628
629 629 mutex_enter(&vd->vdev_dtl_lock);
630 630 for (int t = 0; t < DTL_TYPES; t++) {
631 631 space_map_unload(&vd->vdev_dtl[t]);
632 632 space_map_destroy(&vd->vdev_dtl[t]);
633 633 }
634 634 mutex_exit(&vd->vdev_dtl_lock);
635 635
636 636 mutex_destroy(&vd->vdev_dtl_lock);
637 637 mutex_destroy(&vd->vdev_stat_lock);
638 638 mutex_destroy(&vd->vdev_probe_lock);
639 639
640 640 if (vd == spa->spa_root_vdev)
641 641 spa->spa_root_vdev = NULL;
642 642
643 643 kmem_free(vd, sizeof (vdev_t));
644 644 }
645 645
646 646 /*
647 647 * Transfer top-level vdev state from svd to tvd.
648 648 */
649 649 static void
650 650 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
651 651 {
652 652 spa_t *spa = svd->vdev_spa;
653 653 metaslab_t *msp;
654 654 vdev_t *vd;
655 655 int t;
656 656
657 657 ASSERT(tvd == tvd->vdev_top);
658 658
659 659 tvd->vdev_ms_array = svd->vdev_ms_array;
660 660 tvd->vdev_ms_shift = svd->vdev_ms_shift;
661 661 tvd->vdev_ms_count = svd->vdev_ms_count;
662 662
663 663 svd->vdev_ms_array = 0;
664 664 svd->vdev_ms_shift = 0;
665 665 svd->vdev_ms_count = 0;
666 666
667 667 if (tvd->vdev_mg)
668 668 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
669 669 tvd->vdev_mg = svd->vdev_mg;
670 670 tvd->vdev_ms = svd->vdev_ms;
671 671
672 672 svd->vdev_mg = NULL;
673 673 svd->vdev_ms = NULL;
674 674
675 675 if (tvd->vdev_mg != NULL)
676 676 tvd->vdev_mg->mg_vd = tvd;
677 677
678 678 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
679 679 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
680 680 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
681 681
682 682 svd->vdev_stat.vs_alloc = 0;
683 683 svd->vdev_stat.vs_space = 0;
684 684 svd->vdev_stat.vs_dspace = 0;
685 685
686 686 for (t = 0; t < TXG_SIZE; t++) {
687 687 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
688 688 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
689 689 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
690 690 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
691 691 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
692 692 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
693 693 }
694 694
695 695 if (list_link_active(&svd->vdev_config_dirty_node)) {
696 696 vdev_config_clean(svd);
697 697 vdev_config_dirty(tvd);
698 698 }
699 699
700 700 if (list_link_active(&svd->vdev_state_dirty_node)) {
701 701 vdev_state_clean(svd);
702 702 vdev_state_dirty(tvd);
703 703 }
704 704
705 705 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
706 706 svd->vdev_deflate_ratio = 0;
707 707
708 708 tvd->vdev_islog = svd->vdev_islog;
709 709 svd->vdev_islog = 0;
710 710 }
711 711
712 712 static void
713 713 vdev_top_update(vdev_t *tvd, vdev_t *vd)
714 714 {
715 715 if (vd == NULL)
716 716 return;
717 717
718 718 vd->vdev_top = tvd;
719 719
720 720 for (int c = 0; c < vd->vdev_children; c++)
721 721 vdev_top_update(tvd, vd->vdev_child[c]);
722 722 }
723 723
724 724 /*
725 725 * Add a mirror/replacing vdev above an existing vdev.
726 726 */
727 727 vdev_t *
728 728 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
729 729 {
730 730 spa_t *spa = cvd->vdev_spa;
731 731 vdev_t *pvd = cvd->vdev_parent;
732 732 vdev_t *mvd;
733 733
734 734 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
735 735
736 736 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
737 737
738 738 mvd->vdev_asize = cvd->vdev_asize;
739 739 mvd->vdev_min_asize = cvd->vdev_min_asize;
740 740 mvd->vdev_max_asize = cvd->vdev_max_asize;
741 741 mvd->vdev_ashift = cvd->vdev_ashift;
742 742 mvd->vdev_state = cvd->vdev_state;
743 743 mvd->vdev_crtxg = cvd->vdev_crtxg;
744 744
745 745 vdev_remove_child(pvd, cvd);
746 746 vdev_add_child(pvd, mvd);
747 747 cvd->vdev_id = mvd->vdev_children;
748 748 vdev_add_child(mvd, cvd);
749 749 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
750 750
751 751 if (mvd == mvd->vdev_top)
752 752 vdev_top_transfer(cvd, mvd);
753 753
754 754 return (mvd);
755 755 }
756 756
757 757 /*
758 758 * Remove a 1-way mirror/replacing vdev from the tree.
759 759 */
760 760 void
761 761 vdev_remove_parent(vdev_t *cvd)
762 762 {
763 763 vdev_t *mvd = cvd->vdev_parent;
764 764 vdev_t *pvd = mvd->vdev_parent;
765 765
766 766 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
767 767
768 768 ASSERT(mvd->vdev_children == 1);
769 769 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
770 770 mvd->vdev_ops == &vdev_replacing_ops ||
771 771 mvd->vdev_ops == &vdev_spare_ops);
772 772 cvd->vdev_ashift = mvd->vdev_ashift;
773 773
774 774 vdev_remove_child(mvd, cvd);
775 775 vdev_remove_child(pvd, mvd);
776 776
777 777 /*
778 778 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
779 779 * Otherwise, we could have detached an offline device, and when we
780 780 * go to import the pool we'll think we have two top-level vdevs,
781 781 * instead of a different version of the same top-level vdev.
782 782 */
783 783 if (mvd->vdev_top == mvd) {
784 784 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
785 785 cvd->vdev_orig_guid = cvd->vdev_guid;
786 786 cvd->vdev_guid += guid_delta;
787 787 cvd->vdev_guid_sum += guid_delta;
788 788 }
789 789 cvd->vdev_id = mvd->vdev_id;
790 790 vdev_add_child(pvd, cvd);
791 791 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
792 792
793 793 if (cvd == cvd->vdev_top)
794 794 vdev_top_transfer(mvd, cvd);
795 795
796 796 ASSERT(mvd->vdev_children == 0);
797 797 vdev_free(mvd);
798 798 }
799 799
800 800 int
801 801 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
802 802 {
803 803 spa_t *spa = vd->vdev_spa;
804 804 objset_t *mos = spa->spa_meta_objset;
805 805 uint64_t m;
806 806 uint64_t oldc = vd->vdev_ms_count;
807 807 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
808 808 metaslab_t **mspp;
809 809 int error;
810 810
811 811 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
812 812
813 813 /*
814 814 * This vdev is not being allocated from yet or is a hole.
815 815 */
816 816 if (vd->vdev_ms_shift == 0)
817 817 return (0);
818 818
819 819 ASSERT(!vd->vdev_ishole);
820 820
821 821 /*
822 822 * Compute the raidz-deflation ratio. Note, we hard-code
823 823 * in 128k (1 << 17) because it is the current "typical" blocksize.
824 824 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
825 825 * or we will inconsistently account for existing bp's.
826 826 */
827 827 vd->vdev_deflate_ratio = (1 << 17) /
828 828 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
829 829
830 830 ASSERT(oldc <= newc);
831 831
832 832 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
833 833
834 834 if (oldc != 0) {
835 835 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
836 836 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
837 837 }
838 838
839 839 vd->vdev_ms = mspp;
840 840 vd->vdev_ms_count = newc;
841 841
842 842 for (m = oldc; m < newc; m++) {
843 843 space_map_obj_t smo = { 0, 0, 0 };
844 844 if (txg == 0) {
845 845 uint64_t object = 0;
846 846 error = dmu_read(mos, vd->vdev_ms_array,
847 847 m * sizeof (uint64_t), sizeof (uint64_t), &object,
848 848 DMU_READ_PREFETCH);
849 849 if (error)
850 850 return (error);
851 851 if (object != 0) {
852 852 dmu_buf_t *db;
853 853 error = dmu_bonus_hold(mos, object, FTAG, &db);
854 854 if (error)
855 855 return (error);
856 856 ASSERT3U(db->db_size, >=, sizeof (smo));
857 857 bcopy(db->db_data, &smo, sizeof (smo));
858 858 ASSERT3U(smo.smo_object, ==, object);
859 859 dmu_buf_rele(db, FTAG);
860 860 }
861 861 }
862 862 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
863 863 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
864 864 }
865 865
866 866 if (txg == 0)
867 867 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
868 868
869 869 /*
870 870 * If the vdev is being removed we don't activate
871 871 * the metaslabs since we want to ensure that no new
872 872 * allocations are performed on this device.
873 873 */
874 874 if (oldc == 0 && !vd->vdev_removing)
875 875 metaslab_group_activate(vd->vdev_mg);
876 876
877 877 if (txg == 0)
878 878 spa_config_exit(spa, SCL_ALLOC, FTAG);
879 879
880 880 return (0);
881 881 }
882 882
883 883 void
884 884 vdev_metaslab_fini(vdev_t *vd)
885 885 {
886 886 uint64_t m;
887 887 uint64_t count = vd->vdev_ms_count;
888 888
889 889 if (vd->vdev_ms != NULL) {
890 890 metaslab_group_passivate(vd->vdev_mg);
891 891 for (m = 0; m < count; m++)
892 892 if (vd->vdev_ms[m] != NULL)
893 893 metaslab_fini(vd->vdev_ms[m]);
894 894 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
895 895 vd->vdev_ms = NULL;
896 896 }
897 897 }
898 898
899 899 typedef struct vdev_probe_stats {
900 900 boolean_t vps_readable;
901 901 boolean_t vps_writeable;
902 902 int vps_flags;
903 903 } vdev_probe_stats_t;
904 904
905 905 static void
906 906 vdev_probe_done(zio_t *zio)
907 907 {
908 908 spa_t *spa = zio->io_spa;
909 909 vdev_t *vd = zio->io_vd;
910 910 vdev_probe_stats_t *vps = zio->io_private;
911 911
912 912 ASSERT(vd->vdev_probe_zio != NULL);
913 913
914 914 if (zio->io_type == ZIO_TYPE_READ) {
915 915 if (zio->io_error == 0)
916 916 vps->vps_readable = 1;
917 917 if (zio->io_error == 0 && spa_writeable(spa)) {
918 918 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
919 919 zio->io_offset, zio->io_size, zio->io_data,
920 920 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
921 921 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
922 922 } else {
923 923 zio_buf_free(zio->io_data, zio->io_size);
924 924 }
925 925 } else if (zio->io_type == ZIO_TYPE_WRITE) {
926 926 if (zio->io_error == 0)
927 927 vps->vps_writeable = 1;
928 928 zio_buf_free(zio->io_data, zio->io_size);
929 929 } else if (zio->io_type == ZIO_TYPE_NULL) {
930 930 zio_t *pio;
931 931
932 932 vd->vdev_cant_read |= !vps->vps_readable;
933 933 vd->vdev_cant_write |= !vps->vps_writeable;
934 934
935 935 if (vdev_readable(vd) &&
936 936 (vdev_writeable(vd) || !spa_writeable(spa))) {
937 937 zio->io_error = 0;
938 938 } else {
939 939 ASSERT(zio->io_error != 0);
940 940 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
941 941 spa, vd, NULL, 0, 0);
942 942 zio->io_error = SET_ERROR(ENXIO);
943 943 }
944 944
945 945 mutex_enter(&vd->vdev_probe_lock);
946 946 ASSERT(vd->vdev_probe_zio == zio);
947 947 vd->vdev_probe_zio = NULL;
948 948 mutex_exit(&vd->vdev_probe_lock);
|
↓ open down ↓ |
948 lines elided |
↑ open up ↑ |
949 949
950 950 while ((pio = zio_walk_parents(zio)) != NULL)
951 951 if (!vdev_accessible(vd, pio))
952 952 pio->io_error = SET_ERROR(ENXIO);
953 953
954 954 kmem_free(vps, sizeof (*vps));
955 955 }
956 956 }
957 957
958 958 /*
959 - * Determine whether this device is accessible by reading and writing
960 - * to several known locations: the pad regions of each vdev label
961 - * but the first (which we leave alone in case it contains a VTOC).
959 + * Determine whether this device is accessible.
960 + *
961 + * Read and write to several known locations: the pad regions of each
962 + * vdev label but the first, which we leave alone in case it contains
963 + * a VTOC.
962 964 */
963 965 zio_t *
964 966 vdev_probe(vdev_t *vd, zio_t *zio)
965 967 {
966 968 spa_t *spa = vd->vdev_spa;
967 969 vdev_probe_stats_t *vps = NULL;
968 970 zio_t *pio;
969 971
970 972 ASSERT(vd->vdev_ops->vdev_op_leaf);
971 973
972 974 /*
973 975 * Don't probe the probe.
974 976 */
975 977 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
976 978 return (NULL);
977 979
978 980 /*
979 981 * To prevent 'probe storms' when a device fails, we create
980 982 * just one probe i/o at a time. All zios that want to probe
981 983 * this vdev will become parents of the probe io.
982 984 */
983 985 mutex_enter(&vd->vdev_probe_lock);
984 986
985 987 if ((pio = vd->vdev_probe_zio) == NULL) {
986 988 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
987 989
988 990 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
989 991 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
990 992 ZIO_FLAG_TRYHARD;
991 993
992 994 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
993 995 /*
994 996 * vdev_cant_read and vdev_cant_write can only
995 997 * transition from TRUE to FALSE when we have the
996 998 * SCL_ZIO lock as writer; otherwise they can only
997 999 * transition from FALSE to TRUE. This ensures that
998 1000 * any zio looking at these values can assume that
999 1001 * failures persist for the life of the I/O. That's
1000 1002 * important because when a device has intermittent
1001 1003 * connectivity problems, we want to ensure that
1002 1004 * they're ascribed to the device (ENXIO) and not
1003 1005 * the zio (EIO).
1004 1006 *
1005 1007 * Since we hold SCL_ZIO as writer here, clear both
1006 1008 * values so the probe can reevaluate from first
1007 1009 * principles.
1008 1010 */
1009 1011 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1010 1012 vd->vdev_cant_read = B_FALSE;
1011 1013 vd->vdev_cant_write = B_FALSE;
1012 1014 }
1013 1015
1014 1016 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1015 1017 vdev_probe_done, vps,
1016 1018 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1017 1019
1018 1020 /*
1019 1021 * We can't change the vdev state in this context, so we
1020 1022 * kick off an async task to do it on our behalf.
1021 1023 */
1022 1024 if (zio != NULL) {
1023 1025 vd->vdev_probe_wanted = B_TRUE;
1024 1026 spa_async_request(spa, SPA_ASYNC_PROBE);
1025 1027 }
1026 1028 }
1027 1029
1028 1030 if (zio != NULL)
1029 1031 zio_add_child(zio, pio);
1030 1032
1031 1033 mutex_exit(&vd->vdev_probe_lock);
1032 1034
1033 1035 if (vps == NULL) {
1034 1036 ASSERT(zio != NULL);
1035 1037 return (NULL);
1036 1038 }
1037 1039
1038 1040 for (int l = 1; l < VDEV_LABELS; l++) {
1039 1041 zio_nowait(zio_read_phys(pio, vd,
1040 1042 vdev_label_offset(vd->vdev_psize, l,
1041 1043 offsetof(vdev_label_t, vl_pad2)),
1042 1044 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1043 1045 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1044 1046 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1045 1047 }
1046 1048
1047 1049 if (zio == NULL)
1048 1050 return (pio);
1049 1051
1050 1052 zio_nowait(pio);
1051 1053 return (NULL);
1052 1054 }
1053 1055
1054 1056 static void
1055 1057 vdev_open_child(void *arg)
1056 1058 {
1057 1059 vdev_t *vd = arg;
1058 1060
1059 1061 vd->vdev_open_thread = curthread;
1060 1062 vd->vdev_open_error = vdev_open(vd);
1061 1063 vd->vdev_open_thread = NULL;
1062 1064 }
1063 1065
1064 1066 boolean_t
1065 1067 vdev_uses_zvols(vdev_t *vd)
1066 1068 {
1067 1069 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1068 1070 strlen(ZVOL_DIR)) == 0)
1069 1071 return (B_TRUE);
1070 1072 for (int c = 0; c < vd->vdev_children; c++)
1071 1073 if (vdev_uses_zvols(vd->vdev_child[c]))
1072 1074 return (B_TRUE);
1073 1075 return (B_FALSE);
1074 1076 }
1075 1077
1076 1078 void
1077 1079 vdev_open_children(vdev_t *vd)
1078 1080 {
1079 1081 taskq_t *tq;
1080 1082 int children = vd->vdev_children;
1081 1083
1082 1084 /*
1083 1085 * in order to handle pools on top of zvols, do the opens
1084 1086 * in a single thread so that the same thread holds the
1085 1087 * spa_namespace_lock
1086 1088 */
1087 1089 if (vdev_uses_zvols(vd)) {
1088 1090 for (int c = 0; c < children; c++)
1089 1091 vd->vdev_child[c]->vdev_open_error =
1090 1092 vdev_open(vd->vdev_child[c]);
1091 1093 return;
1092 1094 }
1093 1095 tq = taskq_create("vdev_open", children, minclsyspri,
1094 1096 children, children, TASKQ_PREPOPULATE);
1095 1097
1096 1098 for (int c = 0; c < children; c++)
1097 1099 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1098 1100 TQ_SLEEP) != NULL);
1099 1101
1100 1102 taskq_destroy(tq);
1101 1103 }
1102 1104
1103 1105 /*
1104 1106 * Prepare a virtual device for access.
1105 1107 */
1106 1108 int
1107 1109 vdev_open(vdev_t *vd)
1108 1110 {
1109 1111 spa_t *spa = vd->vdev_spa;
1110 1112 int error;
1111 1113 uint64_t osize = 0;
1112 1114 uint64_t max_osize = 0;
1113 1115 uint64_t asize, max_asize, psize;
1114 1116 uint64_t ashift = 0;
1115 1117
1116 1118 ASSERT(vd->vdev_open_thread == curthread ||
1117 1119 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1118 1120 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1119 1121 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1120 1122 vd->vdev_state == VDEV_STATE_OFFLINE);
1121 1123
1122 1124 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1123 1125 vd->vdev_cant_read = B_FALSE;
1124 1126 vd->vdev_cant_write = B_FALSE;
1125 1127 vd->vdev_min_asize = vdev_get_min_asize(vd);
1126 1128
1127 1129 /*
1128 1130 * If this vdev is not removed, check its fault status. If it's
1129 1131 * faulted, bail out of the open.
1130 1132 */
1131 1133 if (!vd->vdev_removed && vd->vdev_faulted) {
1132 1134 ASSERT(vd->vdev_children == 0);
1133 1135 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1134 1136 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1135 1137 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1136 1138 vd->vdev_label_aux);
1137 1139 return (SET_ERROR(ENXIO));
1138 1140 } else if (vd->vdev_offline) {
1139 1141 ASSERT(vd->vdev_children == 0);
1140 1142 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1141 1143 return (SET_ERROR(ENXIO));
1142 1144 }
1143 1145
1144 1146 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1145 1147
1146 1148 /*
1147 1149 * Reset the vdev_reopening flag so that we actually close
1148 1150 * the vdev on error.
1149 1151 */
1150 1152 vd->vdev_reopening = B_FALSE;
1151 1153 if (zio_injection_enabled && error == 0)
1152 1154 error = zio_handle_device_injection(vd, NULL, ENXIO);
1153 1155
1154 1156 if (error) {
1155 1157 if (vd->vdev_removed &&
1156 1158 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1157 1159 vd->vdev_removed = B_FALSE;
1158 1160
1159 1161 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1160 1162 vd->vdev_stat.vs_aux);
1161 1163 return (error);
1162 1164 }
1163 1165
1164 1166 vd->vdev_removed = B_FALSE;
1165 1167
1166 1168 /*
1167 1169 * Recheck the faulted flag now that we have confirmed that
1168 1170 * the vdev is accessible. If we're faulted, bail.
1169 1171 */
1170 1172 if (vd->vdev_faulted) {
1171 1173 ASSERT(vd->vdev_children == 0);
1172 1174 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1173 1175 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1174 1176 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1175 1177 vd->vdev_label_aux);
1176 1178 return (SET_ERROR(ENXIO));
1177 1179 }
1178 1180
1179 1181 if (vd->vdev_degraded) {
1180 1182 ASSERT(vd->vdev_children == 0);
1181 1183 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1182 1184 VDEV_AUX_ERR_EXCEEDED);
1183 1185 } else {
1184 1186 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1185 1187 }
1186 1188
1187 1189 /*
1188 1190 * For hole or missing vdevs we just return success.
1189 1191 */
1190 1192 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1191 1193 return (0);
1192 1194
1193 1195 for (int c = 0; c < vd->vdev_children; c++) {
1194 1196 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1195 1197 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1196 1198 VDEV_AUX_NONE);
1197 1199 break;
1198 1200 }
1199 1201 }
1200 1202
1201 1203 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1202 1204 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1203 1205
1204 1206 if (vd->vdev_children == 0) {
1205 1207 if (osize < SPA_MINDEVSIZE) {
1206 1208 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1207 1209 VDEV_AUX_TOO_SMALL);
1208 1210 return (SET_ERROR(EOVERFLOW));
1209 1211 }
1210 1212 psize = osize;
1211 1213 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1212 1214 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1213 1215 VDEV_LABEL_END_SIZE);
1214 1216 } else {
1215 1217 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1216 1218 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1217 1219 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1218 1220 VDEV_AUX_TOO_SMALL);
1219 1221 return (SET_ERROR(EOVERFLOW));
1220 1222 }
1221 1223 psize = 0;
1222 1224 asize = osize;
1223 1225 max_asize = max_osize;
1224 1226 }
1225 1227
1226 1228 vd->vdev_psize = psize;
1227 1229
1228 1230 /*
1229 1231 * Make sure the allocatable size hasn't shrunk.
1230 1232 */
1231 1233 if (asize < vd->vdev_min_asize) {
1232 1234 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1233 1235 VDEV_AUX_BAD_LABEL);
1234 1236 return (SET_ERROR(EINVAL));
1235 1237 }
1236 1238
1237 1239 if (vd->vdev_asize == 0) {
1238 1240 /*
1239 1241 * This is the first-ever open, so use the computed values.
1240 1242 * For testing purposes, a higher ashift can be requested.
1241 1243 */
1242 1244 vd->vdev_asize = asize;
1243 1245 vd->vdev_max_asize = max_asize;
1244 1246 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1245 1247 } else {
1246 1248 /*
1247 1249 * Detect if the alignment requirement has increased.
1248 1250 * We don't want to make the pool unavailable, just
1249 1251 * issue a warning instead.
1250 1252 */
1251 1253 if (ashift > vd->vdev_top->vdev_ashift &&
1252 1254 vd->vdev_ops->vdev_op_leaf) {
1253 1255 cmn_err(CE_WARN,
1254 1256 "Disk, '%s', has a block alignment that is "
1255 1257 "larger than the pool's alignment\n",
1256 1258 vd->vdev_path);
1257 1259 }
1258 1260 vd->vdev_max_asize = max_asize;
1259 1261 }
1260 1262
1261 1263 /*
1262 1264 * If all children are healthy and the asize has increased,
1263 1265 * then we've experienced dynamic LUN growth. If automatic
1264 1266 * expansion is enabled then use the additional space.
1265 1267 */
1266 1268 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1267 1269 (vd->vdev_expanding || spa->spa_autoexpand))
1268 1270 vd->vdev_asize = asize;
1269 1271
1270 1272 vdev_set_min_asize(vd);
1271 1273
1272 1274 /*
1273 1275 * Ensure we can issue some IO before declaring the
1274 1276 * vdev open for business.
1275 1277 */
1276 1278 if (vd->vdev_ops->vdev_op_leaf &&
1277 1279 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1278 1280 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1279 1281 VDEV_AUX_ERR_EXCEEDED);
1280 1282 return (error);
1281 1283 }
1282 1284
1283 1285 /*
1284 1286 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1285 1287 * resilver. But don't do this if we are doing a reopen for a scrub,
1286 1288 * since this would just restart the scrub we are already doing.
1287 1289 */
1288 1290 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1289 1291 vdev_resilver_needed(vd, NULL, NULL))
1290 1292 spa_async_request(spa, SPA_ASYNC_RESILVER);
1291 1293
1292 1294 return (0);
1293 1295 }
1294 1296
1295 1297 /*
1296 1298 * Called once the vdevs are all opened, this routine validates the label
1297 1299 * contents. This needs to be done before vdev_load() so that we don't
1298 1300 * inadvertently do repair I/Os to the wrong device.
1299 1301 *
1300 1302 * If 'strict' is false ignore the spa guid check. This is necessary because
1301 1303 * if the machine crashed during a re-guid the new guid might have been written
1302 1304 * to all of the vdev labels, but not the cached config. The strict check
1303 1305 * will be performed when the pool is opened again using the mos config.
1304 1306 *
1305 1307 * This function will only return failure if one of the vdevs indicates that it
1306 1308 * has since been destroyed or exported. This is only possible if
1307 1309 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1308 1310 * will be updated but the function will return 0.
1309 1311 */
1310 1312 int
1311 1313 vdev_validate(vdev_t *vd, boolean_t strict)
1312 1314 {
1313 1315 spa_t *spa = vd->vdev_spa;
1314 1316 nvlist_t *label;
1315 1317 uint64_t guid = 0, top_guid;
1316 1318 uint64_t state;
1317 1319
1318 1320 for (int c = 0; c < vd->vdev_children; c++)
1319 1321 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1320 1322 return (SET_ERROR(EBADF));
1321 1323
1322 1324 /*
1323 1325 * If the device has already failed, or was marked offline, don't do
1324 1326 * any further validation. Otherwise, label I/O will fail and we will
1325 1327 * overwrite the previous state.
1326 1328 */
1327 1329 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1328 1330 uint64_t aux_guid = 0;
1329 1331 nvlist_t *nvl;
1330 1332 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1331 1333 spa_last_synced_txg(spa) : -1ULL;
1332 1334
1333 1335 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1334 1336 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1335 1337 VDEV_AUX_BAD_LABEL);
1336 1338 return (0);
1337 1339 }
1338 1340
1339 1341 /*
1340 1342 * Determine if this vdev has been split off into another
1341 1343 * pool. If so, then refuse to open it.
1342 1344 */
1343 1345 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1344 1346 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1345 1347 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1346 1348 VDEV_AUX_SPLIT_POOL);
1347 1349 nvlist_free(label);
1348 1350 return (0);
1349 1351 }
1350 1352
1351 1353 if (strict && (nvlist_lookup_uint64(label,
1352 1354 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1353 1355 guid != spa_guid(spa))) {
1354 1356 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1355 1357 VDEV_AUX_CORRUPT_DATA);
1356 1358 nvlist_free(label);
1357 1359 return (0);
1358 1360 }
1359 1361
1360 1362 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1361 1363 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1362 1364 &aux_guid) != 0)
1363 1365 aux_guid = 0;
1364 1366
1365 1367 /*
1366 1368 * If this vdev just became a top-level vdev because its
1367 1369 * sibling was detached, it will have adopted the parent's
1368 1370 * vdev guid -- but the label may or may not be on disk yet.
1369 1371 * Fortunately, either version of the label will have the
1370 1372 * same top guid, so if we're a top-level vdev, we can
1371 1373 * safely compare to that instead.
1372 1374 *
1373 1375 * If we split this vdev off instead, then we also check the
1374 1376 * original pool's guid. We don't want to consider the vdev
1375 1377 * corrupt if it is partway through a split operation.
1376 1378 */
1377 1379 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1378 1380 &guid) != 0 ||
1379 1381 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1380 1382 &top_guid) != 0 ||
1381 1383 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1382 1384 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1383 1385 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1384 1386 VDEV_AUX_CORRUPT_DATA);
1385 1387 nvlist_free(label);
1386 1388 return (0);
1387 1389 }
1388 1390
1389 1391 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1390 1392 &state) != 0) {
1391 1393 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1392 1394 VDEV_AUX_CORRUPT_DATA);
1393 1395 nvlist_free(label);
1394 1396 return (0);
1395 1397 }
1396 1398
1397 1399 nvlist_free(label);
1398 1400
1399 1401 /*
1400 1402 * If this is a verbatim import, no need to check the
1401 1403 * state of the pool.
1402 1404 */
1403 1405 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1404 1406 spa_load_state(spa) == SPA_LOAD_OPEN &&
1405 1407 state != POOL_STATE_ACTIVE)
1406 1408 return (SET_ERROR(EBADF));
1407 1409
1408 1410 /*
1409 1411 * If we were able to open and validate a vdev that was
1410 1412 * previously marked permanently unavailable, clear that state
1411 1413 * now.
1412 1414 */
1413 1415 if (vd->vdev_not_present)
1414 1416 vd->vdev_not_present = 0;
1415 1417 }
1416 1418
1417 1419 return (0);
1418 1420 }
1419 1421
1420 1422 /*
1421 1423 * Close a virtual device.
1422 1424 */
1423 1425 void
1424 1426 vdev_close(vdev_t *vd)
1425 1427 {
1426 1428 spa_t *spa = vd->vdev_spa;
1427 1429 vdev_t *pvd = vd->vdev_parent;
1428 1430
1429 1431 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1430 1432
1431 1433 /*
1432 1434 * If our parent is reopening, then we are as well, unless we are
1433 1435 * going offline.
1434 1436 */
1435 1437 if (pvd != NULL && pvd->vdev_reopening)
1436 1438 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1437 1439
1438 1440 vd->vdev_ops->vdev_op_close(vd);
1439 1441
1440 1442 vdev_cache_purge(vd);
1441 1443
1442 1444 /*
1443 1445 * We record the previous state before we close it, so that if we are
1444 1446 * doing a reopen(), we don't generate FMA ereports if we notice that
1445 1447 * it's still faulted.
1446 1448 */
1447 1449 vd->vdev_prevstate = vd->vdev_state;
1448 1450
1449 1451 if (vd->vdev_offline)
1450 1452 vd->vdev_state = VDEV_STATE_OFFLINE;
1451 1453 else
1452 1454 vd->vdev_state = VDEV_STATE_CLOSED;
1453 1455 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1454 1456 }
1455 1457
1456 1458 void
1457 1459 vdev_hold(vdev_t *vd)
1458 1460 {
1459 1461 spa_t *spa = vd->vdev_spa;
1460 1462
1461 1463 ASSERT(spa_is_root(spa));
1462 1464 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1463 1465 return;
1464 1466
1465 1467 for (int c = 0; c < vd->vdev_children; c++)
1466 1468 vdev_hold(vd->vdev_child[c]);
1467 1469
1468 1470 if (vd->vdev_ops->vdev_op_leaf)
1469 1471 vd->vdev_ops->vdev_op_hold(vd);
1470 1472 }
1471 1473
1472 1474 void
1473 1475 vdev_rele(vdev_t *vd)
1474 1476 {
1475 1477 spa_t *spa = vd->vdev_spa;
1476 1478
1477 1479 ASSERT(spa_is_root(spa));
1478 1480 for (int c = 0; c < vd->vdev_children; c++)
1479 1481 vdev_rele(vd->vdev_child[c]);
1480 1482
1481 1483 if (vd->vdev_ops->vdev_op_leaf)
1482 1484 vd->vdev_ops->vdev_op_rele(vd);
1483 1485 }
1484 1486
1485 1487 /*
1486 1488 * Reopen all interior vdevs and any unopened leaves. We don't actually
1487 1489 * reopen leaf vdevs which had previously been opened as they might deadlock
1488 1490 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1489 1491 * If the leaf has never been opened then open it, as usual.
1490 1492 */
1491 1493 void
1492 1494 vdev_reopen(vdev_t *vd)
1493 1495 {
1494 1496 spa_t *spa = vd->vdev_spa;
1495 1497
1496 1498 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1497 1499
1498 1500 /* set the reopening flag unless we're taking the vdev offline */
1499 1501 vd->vdev_reopening = !vd->vdev_offline;
1500 1502 vdev_close(vd);
1501 1503 (void) vdev_open(vd);
1502 1504
1503 1505 /*
1504 1506 * Call vdev_validate() here to make sure we have the same device.
1505 1507 * Otherwise, a device with an invalid label could be successfully
1506 1508 * opened in response to vdev_reopen().
1507 1509 */
1508 1510 if (vd->vdev_aux) {
1509 1511 (void) vdev_validate_aux(vd);
1510 1512 if (vdev_readable(vd) && vdev_writeable(vd) &&
1511 1513 vd->vdev_aux == &spa->spa_l2cache &&
1512 1514 !l2arc_vdev_present(vd))
1513 1515 l2arc_add_vdev(spa, vd);
1514 1516 } else {
1515 1517 (void) vdev_validate(vd, B_TRUE);
1516 1518 }
1517 1519
1518 1520 /*
1519 1521 * Reassess parent vdev's health.
1520 1522 */
1521 1523 vdev_propagate_state(vd);
1522 1524 }
1523 1525
1524 1526 int
1525 1527 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1526 1528 {
1527 1529 int error;
1528 1530
1529 1531 /*
1530 1532 * Normally, partial opens (e.g. of a mirror) are allowed.
1531 1533 * For a create, however, we want to fail the request if
1532 1534 * there are any components we can't open.
1533 1535 */
1534 1536 error = vdev_open(vd);
1535 1537
1536 1538 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1537 1539 vdev_close(vd);
1538 1540 return (error ? error : ENXIO);
1539 1541 }
1540 1542
1541 1543 /*
1542 1544 * Recursively initialize all labels.
1543 1545 */
1544 1546 if ((error = vdev_label_init(vd, txg, isreplacing ?
1545 1547 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1546 1548 vdev_close(vd);
1547 1549 return (error);
1548 1550 }
1549 1551
1550 1552 return (0);
1551 1553 }
1552 1554
1553 1555 void
1554 1556 vdev_metaslab_set_size(vdev_t *vd)
1555 1557 {
1556 1558 /*
1557 1559 * Aim for roughly 200 metaslabs per vdev.
1558 1560 */
1559 1561 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1560 1562 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1561 1563 }
1562 1564
1563 1565 void
1564 1566 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1565 1567 {
1566 1568 ASSERT(vd == vd->vdev_top);
1567 1569 ASSERT(!vd->vdev_ishole);
1568 1570 ASSERT(ISP2(flags));
1569 1571 ASSERT(spa_writeable(vd->vdev_spa));
1570 1572
1571 1573 if (flags & VDD_METASLAB)
1572 1574 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1573 1575
1574 1576 if (flags & VDD_DTL)
1575 1577 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1576 1578
1577 1579 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1578 1580 }
1579 1581
1580 1582 /*
1581 1583 * DTLs.
1582 1584 *
1583 1585 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1584 1586 * the vdev has less than perfect replication. There are four kinds of DTL:
1585 1587 *
1586 1588 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1587 1589 *
1588 1590 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1589 1591 *
1590 1592 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1591 1593 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1592 1594 * txgs that was scrubbed.
1593 1595 *
1594 1596 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1595 1597 * persistent errors or just some device being offline.
1596 1598 * Unlike the other three, the DTL_OUTAGE map is not generally
1597 1599 * maintained; it's only computed when needed, typically to
1598 1600 * determine whether a device can be detached.
1599 1601 *
1600 1602 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1601 1603 * either has the data or it doesn't.
1602 1604 *
1603 1605 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1604 1606 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1605 1607 * if any child is less than fully replicated, then so is its parent.
1606 1608 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1607 1609 * comprising only those txgs which appear in 'maxfaults' or more children;
1608 1610 * those are the txgs we don't have enough replication to read. For example,
1609 1611 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1610 1612 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1611 1613 * two child DTL_MISSING maps.
1612 1614 *
1613 1615 * It should be clear from the above that to compute the DTLs and outage maps
1614 1616 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1615 1617 * Therefore, that is all we keep on disk. When loading the pool, or after
1616 1618 * a configuration change, we generate all other DTLs from first principles.
1617 1619 */
1618 1620 void
1619 1621 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1620 1622 {
1621 1623 space_map_t *sm = &vd->vdev_dtl[t];
1622 1624
1623 1625 ASSERT(t < DTL_TYPES);
1624 1626 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1625 1627 ASSERT(spa_writeable(vd->vdev_spa));
1626 1628
1627 1629 mutex_enter(sm->sm_lock);
1628 1630 if (!space_map_contains(sm, txg, size))
1629 1631 space_map_add(sm, txg, size);
1630 1632 mutex_exit(sm->sm_lock);
1631 1633 }
1632 1634
1633 1635 boolean_t
1634 1636 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1635 1637 {
1636 1638 space_map_t *sm = &vd->vdev_dtl[t];
1637 1639 boolean_t dirty = B_FALSE;
1638 1640
1639 1641 ASSERT(t < DTL_TYPES);
1640 1642 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1641 1643
1642 1644 mutex_enter(sm->sm_lock);
1643 1645 if (sm->sm_space != 0)
1644 1646 dirty = space_map_contains(sm, txg, size);
1645 1647 mutex_exit(sm->sm_lock);
1646 1648
1647 1649 return (dirty);
1648 1650 }
1649 1651
1650 1652 boolean_t
1651 1653 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1652 1654 {
1653 1655 space_map_t *sm = &vd->vdev_dtl[t];
1654 1656 boolean_t empty;
1655 1657
1656 1658 mutex_enter(sm->sm_lock);
1657 1659 empty = (sm->sm_space == 0);
1658 1660 mutex_exit(sm->sm_lock);
1659 1661
1660 1662 return (empty);
1661 1663 }
1662 1664
1663 1665 /*
1664 1666 * Reassess DTLs after a config change or scrub completion.
1665 1667 */
1666 1668 void
1667 1669 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1668 1670 {
1669 1671 spa_t *spa = vd->vdev_spa;
1670 1672 avl_tree_t reftree;
1671 1673 int minref;
1672 1674
1673 1675 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1674 1676
1675 1677 for (int c = 0; c < vd->vdev_children; c++)
1676 1678 vdev_dtl_reassess(vd->vdev_child[c], txg,
1677 1679 scrub_txg, scrub_done);
1678 1680
1679 1681 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1680 1682 return;
1681 1683
1682 1684 if (vd->vdev_ops->vdev_op_leaf) {
1683 1685 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1684 1686
1685 1687 mutex_enter(&vd->vdev_dtl_lock);
1686 1688 if (scrub_txg != 0 &&
1687 1689 (spa->spa_scrub_started ||
1688 1690 (scn && scn->scn_phys.scn_errors == 0))) {
1689 1691 /*
1690 1692 * We completed a scrub up to scrub_txg. If we
1691 1693 * did it without rebooting, then the scrub dtl
1692 1694 * will be valid, so excise the old region and
1693 1695 * fold in the scrub dtl. Otherwise, leave the
1694 1696 * dtl as-is if there was an error.
1695 1697 *
1696 1698 * There's little trick here: to excise the beginning
1697 1699 * of the DTL_MISSING map, we put it into a reference
1698 1700 * tree and then add a segment with refcnt -1 that
1699 1701 * covers the range [0, scrub_txg). This means
1700 1702 * that each txg in that range has refcnt -1 or 0.
1701 1703 * We then add DTL_SCRUB with a refcnt of 2, so that
1702 1704 * entries in the range [0, scrub_txg) will have a
1703 1705 * positive refcnt -- either 1 or 2. We then convert
1704 1706 * the reference tree into the new DTL_MISSING map.
1705 1707 */
1706 1708 space_map_ref_create(&reftree);
1707 1709 space_map_ref_add_map(&reftree,
1708 1710 &vd->vdev_dtl[DTL_MISSING], 1);
1709 1711 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1710 1712 space_map_ref_add_map(&reftree,
1711 1713 &vd->vdev_dtl[DTL_SCRUB], 2);
1712 1714 space_map_ref_generate_map(&reftree,
1713 1715 &vd->vdev_dtl[DTL_MISSING], 1);
1714 1716 space_map_ref_destroy(&reftree);
1715 1717 }
1716 1718 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1717 1719 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1718 1720 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1719 1721 if (scrub_done)
1720 1722 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1721 1723 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1722 1724 if (!vdev_readable(vd))
1723 1725 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1724 1726 else
1725 1727 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1726 1728 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1727 1729 mutex_exit(&vd->vdev_dtl_lock);
1728 1730
1729 1731 if (txg != 0)
1730 1732 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1731 1733 return;
1732 1734 }
1733 1735
1734 1736 mutex_enter(&vd->vdev_dtl_lock);
1735 1737 for (int t = 0; t < DTL_TYPES; t++) {
1736 1738 /* account for child's outage in parent's missing map */
1737 1739 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1738 1740 if (t == DTL_SCRUB)
1739 1741 continue; /* leaf vdevs only */
1740 1742 if (t == DTL_PARTIAL)
1741 1743 minref = 1; /* i.e. non-zero */
1742 1744 else if (vd->vdev_nparity != 0)
1743 1745 minref = vd->vdev_nparity + 1; /* RAID-Z */
1744 1746 else
1745 1747 minref = vd->vdev_children; /* any kind of mirror */
1746 1748 space_map_ref_create(&reftree);
1747 1749 for (int c = 0; c < vd->vdev_children; c++) {
1748 1750 vdev_t *cvd = vd->vdev_child[c];
1749 1751 mutex_enter(&cvd->vdev_dtl_lock);
1750 1752 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1751 1753 mutex_exit(&cvd->vdev_dtl_lock);
1752 1754 }
1753 1755 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1754 1756 space_map_ref_destroy(&reftree);
1755 1757 }
1756 1758 mutex_exit(&vd->vdev_dtl_lock);
1757 1759 }
1758 1760
1759 1761 static int
1760 1762 vdev_dtl_load(vdev_t *vd)
1761 1763 {
1762 1764 spa_t *spa = vd->vdev_spa;
1763 1765 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1764 1766 objset_t *mos = spa->spa_meta_objset;
1765 1767 dmu_buf_t *db;
1766 1768 int error;
1767 1769
1768 1770 ASSERT(vd->vdev_children == 0);
1769 1771
1770 1772 if (smo->smo_object == 0)
1771 1773 return (0);
1772 1774
1773 1775 ASSERT(!vd->vdev_ishole);
1774 1776
1775 1777 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1776 1778 return (error);
1777 1779
1778 1780 ASSERT3U(db->db_size, >=, sizeof (*smo));
1779 1781 bcopy(db->db_data, smo, sizeof (*smo));
1780 1782 dmu_buf_rele(db, FTAG);
1781 1783
1782 1784 mutex_enter(&vd->vdev_dtl_lock);
1783 1785 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1784 1786 NULL, SM_ALLOC, smo, mos);
1785 1787 mutex_exit(&vd->vdev_dtl_lock);
1786 1788
1787 1789 return (error);
1788 1790 }
1789 1791
1790 1792 void
1791 1793 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1792 1794 {
1793 1795 spa_t *spa = vd->vdev_spa;
1794 1796 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1795 1797 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1796 1798 objset_t *mos = spa->spa_meta_objset;
1797 1799 space_map_t smsync;
1798 1800 kmutex_t smlock;
1799 1801 dmu_buf_t *db;
1800 1802 dmu_tx_t *tx;
1801 1803
1802 1804 ASSERT(!vd->vdev_ishole);
1803 1805
1804 1806 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1805 1807
1806 1808 if (vd->vdev_detached) {
1807 1809 if (smo->smo_object != 0) {
1808 1810 int err = dmu_object_free(mos, smo->smo_object, tx);
1809 1811 ASSERT0(err);
1810 1812 smo->smo_object = 0;
1811 1813 }
1812 1814 dmu_tx_commit(tx);
1813 1815 return;
1814 1816 }
1815 1817
1816 1818 if (smo->smo_object == 0) {
1817 1819 ASSERT(smo->smo_objsize == 0);
1818 1820 ASSERT(smo->smo_alloc == 0);
1819 1821 smo->smo_object = dmu_object_alloc(mos,
1820 1822 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1821 1823 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1822 1824 ASSERT(smo->smo_object != 0);
1823 1825 vdev_config_dirty(vd->vdev_top);
1824 1826 }
1825 1827
1826 1828 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1827 1829
1828 1830 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1829 1831 &smlock);
1830 1832
1831 1833 mutex_enter(&smlock);
1832 1834
1833 1835 mutex_enter(&vd->vdev_dtl_lock);
1834 1836 space_map_walk(sm, space_map_add, &smsync);
1835 1837 mutex_exit(&vd->vdev_dtl_lock);
1836 1838
1837 1839 space_map_truncate(smo, mos, tx);
1838 1840 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1839 1841 space_map_vacate(&smsync, NULL, NULL);
1840 1842
1841 1843 space_map_destroy(&smsync);
1842 1844
1843 1845 mutex_exit(&smlock);
1844 1846 mutex_destroy(&smlock);
1845 1847
1846 1848 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1847 1849 dmu_buf_will_dirty(db, tx);
1848 1850 ASSERT3U(db->db_size, >=, sizeof (*smo));
1849 1851 bcopy(smo, db->db_data, sizeof (*smo));
1850 1852 dmu_buf_rele(db, FTAG);
1851 1853
1852 1854 dmu_tx_commit(tx);
1853 1855 }
1854 1856
1855 1857 /*
1856 1858 * Determine whether the specified vdev can be offlined/detached/removed
1857 1859 * without losing data.
1858 1860 */
1859 1861 boolean_t
1860 1862 vdev_dtl_required(vdev_t *vd)
1861 1863 {
1862 1864 spa_t *spa = vd->vdev_spa;
1863 1865 vdev_t *tvd = vd->vdev_top;
1864 1866 uint8_t cant_read = vd->vdev_cant_read;
1865 1867 boolean_t required;
1866 1868
1867 1869 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1868 1870
1869 1871 if (vd == spa->spa_root_vdev || vd == tvd)
1870 1872 return (B_TRUE);
1871 1873
1872 1874 /*
1873 1875 * Temporarily mark the device as unreadable, and then determine
1874 1876 * whether this results in any DTL outages in the top-level vdev.
1875 1877 * If not, we can safely offline/detach/remove the device.
1876 1878 */
1877 1879 vd->vdev_cant_read = B_TRUE;
1878 1880 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1879 1881 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1880 1882 vd->vdev_cant_read = cant_read;
1881 1883 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1882 1884
1883 1885 if (!required && zio_injection_enabled)
1884 1886 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1885 1887
1886 1888 return (required);
1887 1889 }
1888 1890
1889 1891 /*
1890 1892 * Determine if resilver is needed, and if so the txg range.
1891 1893 */
1892 1894 boolean_t
1893 1895 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1894 1896 {
1895 1897 boolean_t needed = B_FALSE;
1896 1898 uint64_t thismin = UINT64_MAX;
1897 1899 uint64_t thismax = 0;
1898 1900
1899 1901 if (vd->vdev_children == 0) {
1900 1902 mutex_enter(&vd->vdev_dtl_lock);
1901 1903 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1902 1904 vdev_writeable(vd)) {
1903 1905 space_seg_t *ss;
1904 1906
1905 1907 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1906 1908 thismin = ss->ss_start - 1;
1907 1909 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1908 1910 thismax = ss->ss_end;
1909 1911 needed = B_TRUE;
1910 1912 }
1911 1913 mutex_exit(&vd->vdev_dtl_lock);
1912 1914 } else {
1913 1915 for (int c = 0; c < vd->vdev_children; c++) {
1914 1916 vdev_t *cvd = vd->vdev_child[c];
1915 1917 uint64_t cmin, cmax;
1916 1918
1917 1919 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1918 1920 thismin = MIN(thismin, cmin);
1919 1921 thismax = MAX(thismax, cmax);
1920 1922 needed = B_TRUE;
1921 1923 }
1922 1924 }
1923 1925 }
1924 1926
1925 1927 if (needed && minp) {
1926 1928 *minp = thismin;
1927 1929 *maxp = thismax;
1928 1930 }
1929 1931 return (needed);
1930 1932 }
1931 1933
1932 1934 void
1933 1935 vdev_load(vdev_t *vd)
1934 1936 {
1935 1937 /*
1936 1938 * Recursively load all children.
1937 1939 */
1938 1940 for (int c = 0; c < vd->vdev_children; c++)
1939 1941 vdev_load(vd->vdev_child[c]);
1940 1942
1941 1943 /*
1942 1944 * If this is a top-level vdev, initialize its metaslabs.
1943 1945 */
1944 1946 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1945 1947 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1946 1948 vdev_metaslab_init(vd, 0) != 0))
1947 1949 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1948 1950 VDEV_AUX_CORRUPT_DATA);
1949 1951
1950 1952 /*
1951 1953 * If this is a leaf vdev, load its DTL.
1952 1954 */
1953 1955 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1954 1956 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1955 1957 VDEV_AUX_CORRUPT_DATA);
1956 1958 }
1957 1959
1958 1960 /*
1959 1961 * The special vdev case is used for hot spares and l2cache devices. Its
1960 1962 * sole purpose it to set the vdev state for the associated vdev. To do this,
1961 1963 * we make sure that we can open the underlying device, then try to read the
1962 1964 * label, and make sure that the label is sane and that it hasn't been
1963 1965 * repurposed to another pool.
1964 1966 */
1965 1967 int
1966 1968 vdev_validate_aux(vdev_t *vd)
1967 1969 {
1968 1970 nvlist_t *label;
1969 1971 uint64_t guid, version;
1970 1972 uint64_t state;
1971 1973
1972 1974 if (!vdev_readable(vd))
1973 1975 return (0);
1974 1976
1975 1977 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
1976 1978 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1977 1979 VDEV_AUX_CORRUPT_DATA);
1978 1980 return (-1);
1979 1981 }
1980 1982
1981 1983 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1982 1984 !SPA_VERSION_IS_SUPPORTED(version) ||
1983 1985 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1984 1986 guid != vd->vdev_guid ||
1985 1987 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1986 1988 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1987 1989 VDEV_AUX_CORRUPT_DATA);
1988 1990 nvlist_free(label);
1989 1991 return (-1);
1990 1992 }
1991 1993
1992 1994 /*
1993 1995 * We don't actually check the pool state here. If it's in fact in
1994 1996 * use by another pool, we update this fact on the fly when requested.
1995 1997 */
1996 1998 nvlist_free(label);
1997 1999 return (0);
1998 2000 }
1999 2001
2000 2002 void
2001 2003 vdev_remove(vdev_t *vd, uint64_t txg)
2002 2004 {
2003 2005 spa_t *spa = vd->vdev_spa;
2004 2006 objset_t *mos = spa->spa_meta_objset;
2005 2007 dmu_tx_t *tx;
2006 2008
2007 2009 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2008 2010
2009 2011 if (vd->vdev_dtl_smo.smo_object) {
2010 2012 ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2011 2013 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2012 2014 vd->vdev_dtl_smo.smo_object = 0;
2013 2015 }
2014 2016
2015 2017 if (vd->vdev_ms != NULL) {
2016 2018 for (int m = 0; m < vd->vdev_ms_count; m++) {
2017 2019 metaslab_t *msp = vd->vdev_ms[m];
2018 2020
2019 2021 if (msp == NULL || msp->ms_smo.smo_object == 0)
2020 2022 continue;
2021 2023
2022 2024 ASSERT0(msp->ms_smo.smo_alloc);
2023 2025 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2024 2026 msp->ms_smo.smo_object = 0;
2025 2027 }
2026 2028 }
2027 2029
2028 2030 if (vd->vdev_ms_array) {
2029 2031 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2030 2032 vd->vdev_ms_array = 0;
2031 2033 vd->vdev_ms_shift = 0;
2032 2034 }
2033 2035 dmu_tx_commit(tx);
2034 2036 }
2035 2037
2036 2038 void
2037 2039 vdev_sync_done(vdev_t *vd, uint64_t txg)
2038 2040 {
2039 2041 metaslab_t *msp;
2040 2042 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2041 2043
2042 2044 ASSERT(!vd->vdev_ishole);
2043 2045
2044 2046 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2045 2047 metaslab_sync_done(msp, txg);
2046 2048
2047 2049 if (reassess)
2048 2050 metaslab_sync_reassess(vd->vdev_mg);
2049 2051 }
2050 2052
2051 2053 void
2052 2054 vdev_sync(vdev_t *vd, uint64_t txg)
2053 2055 {
2054 2056 spa_t *spa = vd->vdev_spa;
2055 2057 vdev_t *lvd;
2056 2058 metaslab_t *msp;
2057 2059 dmu_tx_t *tx;
2058 2060
2059 2061 ASSERT(!vd->vdev_ishole);
2060 2062
2061 2063 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2062 2064 ASSERT(vd == vd->vdev_top);
2063 2065 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2064 2066 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2065 2067 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2066 2068 ASSERT(vd->vdev_ms_array != 0);
2067 2069 vdev_config_dirty(vd);
2068 2070 dmu_tx_commit(tx);
2069 2071 }
2070 2072
2071 2073 /*
2072 2074 * Remove the metadata associated with this vdev once it's empty.
2073 2075 */
2074 2076 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2075 2077 vdev_remove(vd, txg);
2076 2078
2077 2079 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2078 2080 metaslab_sync(msp, txg);
2079 2081 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2080 2082 }
2081 2083
2082 2084 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2083 2085 vdev_dtl_sync(lvd, txg);
2084 2086
2085 2087 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2086 2088 }
2087 2089
2088 2090 uint64_t
2089 2091 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2090 2092 {
2091 2093 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2092 2094 }
2093 2095
2094 2096 /*
2095 2097 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2096 2098 * not be opened, and no I/O is attempted.
2097 2099 */
2098 2100 int
2099 2101 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2100 2102 {
2101 2103 vdev_t *vd, *tvd;
2102 2104
2103 2105 spa_vdev_state_enter(spa, SCL_NONE);
2104 2106
2105 2107 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2106 2108 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2107 2109
2108 2110 if (!vd->vdev_ops->vdev_op_leaf)
2109 2111 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2110 2112
2111 2113 tvd = vd->vdev_top;
2112 2114
2113 2115 /*
2114 2116 * We don't directly use the aux state here, but if we do a
2115 2117 * vdev_reopen(), we need this value to be present to remember why we
2116 2118 * were faulted.
2117 2119 */
2118 2120 vd->vdev_label_aux = aux;
2119 2121
2120 2122 /*
2121 2123 * Faulted state takes precedence over degraded.
2122 2124 */
2123 2125 vd->vdev_delayed_close = B_FALSE;
2124 2126 vd->vdev_faulted = 1ULL;
2125 2127 vd->vdev_degraded = 0ULL;
2126 2128 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2127 2129
2128 2130 /*
2129 2131 * If this device has the only valid copy of the data, then
2130 2132 * back off and simply mark the vdev as degraded instead.
2131 2133 */
2132 2134 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2133 2135 vd->vdev_degraded = 1ULL;
2134 2136 vd->vdev_faulted = 0ULL;
2135 2137
2136 2138 /*
2137 2139 * If we reopen the device and it's not dead, only then do we
2138 2140 * mark it degraded.
2139 2141 */
2140 2142 vdev_reopen(tvd);
2141 2143
2142 2144 if (vdev_readable(vd))
2143 2145 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2144 2146 }
2145 2147
2146 2148 return (spa_vdev_state_exit(spa, vd, 0));
2147 2149 }
2148 2150
2149 2151 /*
2150 2152 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2151 2153 * user that something is wrong. The vdev continues to operate as normal as far
2152 2154 * as I/O is concerned.
2153 2155 */
2154 2156 int
2155 2157 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2156 2158 {
2157 2159 vdev_t *vd;
2158 2160
2159 2161 spa_vdev_state_enter(spa, SCL_NONE);
2160 2162
2161 2163 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2162 2164 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2163 2165
2164 2166 if (!vd->vdev_ops->vdev_op_leaf)
2165 2167 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2166 2168
2167 2169 /*
2168 2170 * If the vdev is already faulted, then don't do anything.
2169 2171 */
2170 2172 if (vd->vdev_faulted || vd->vdev_degraded)
2171 2173 return (spa_vdev_state_exit(spa, NULL, 0));
|
↓ open down ↓ |
1200 lines elided |
↑ open up ↑ |
2172 2174
2173 2175 vd->vdev_degraded = 1ULL;
2174 2176 if (!vdev_is_dead(vd))
2175 2177 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2176 2178 aux);
2177 2179
2178 2180 return (spa_vdev_state_exit(spa, vd, 0));
2179 2181 }
2180 2182
2181 2183 /*
2182 - * Online the given vdev. If 'unspare' is set, it implies two things. First,
2183 - * any attached spare device should be detached when the device finishes
2184 - * resilvering. Second, the online should be treated like a 'test' online case,
2185 - * so no FMA events are generated if the device fails to open.
2184 + * Online the given vdev.
2185 + *
2186 + * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2187 + * spare device should be detached when the device finishes resilvering.
2188 + * Second, the online should be treated like a 'test' online case, so no FMA
2189 + * events are generated if the device fails to open.
2186 2190 */
2187 2191 int
2188 2192 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2189 2193 {
2190 2194 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2191 2195
2192 2196 spa_vdev_state_enter(spa, SCL_NONE);
2193 2197
2194 2198 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2195 2199 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2196 2200
2197 2201 if (!vd->vdev_ops->vdev_op_leaf)
2198 2202 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2199 2203
2200 2204 tvd = vd->vdev_top;
2201 2205 vd->vdev_offline = B_FALSE;
2202 2206 vd->vdev_tmpoffline = B_FALSE;
2203 2207 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2204 2208 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2205 2209
2206 2210 /* XXX - L2ARC 1.0 does not support expansion */
2207 2211 if (!vd->vdev_aux) {
2208 2212 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2209 2213 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2210 2214 }
2211 2215
2212 2216 vdev_reopen(tvd);
2213 2217 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2214 2218
2215 2219 if (!vd->vdev_aux) {
2216 2220 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2217 2221 pvd->vdev_expanding = B_FALSE;
2218 2222 }
2219 2223
2220 2224 if (newstate)
2221 2225 *newstate = vd->vdev_state;
2222 2226 if ((flags & ZFS_ONLINE_UNSPARE) &&
2223 2227 !vdev_is_dead(vd) && vd->vdev_parent &&
2224 2228 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2225 2229 vd->vdev_parent->vdev_child[0] == vd)
2226 2230 vd->vdev_unspare = B_TRUE;
2227 2231
2228 2232 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2229 2233
2230 2234 /* XXX - L2ARC 1.0 does not support expansion */
2231 2235 if (vd->vdev_aux)
2232 2236 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2233 2237 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2234 2238 }
2235 2239 return (spa_vdev_state_exit(spa, vd, 0));
2236 2240 }
2237 2241
2238 2242 static int
2239 2243 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2240 2244 {
2241 2245 vdev_t *vd, *tvd;
2242 2246 int error = 0;
2243 2247 uint64_t generation;
2244 2248 metaslab_group_t *mg;
2245 2249
2246 2250 top:
2247 2251 spa_vdev_state_enter(spa, SCL_ALLOC);
2248 2252
2249 2253 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2250 2254 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2251 2255
2252 2256 if (!vd->vdev_ops->vdev_op_leaf)
2253 2257 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2254 2258
2255 2259 tvd = vd->vdev_top;
2256 2260 mg = tvd->vdev_mg;
2257 2261 generation = spa->spa_config_generation + 1;
2258 2262
2259 2263 /*
2260 2264 * If the device isn't already offline, try to offline it.
2261 2265 */
2262 2266 if (!vd->vdev_offline) {
2263 2267 /*
2264 2268 * If this device has the only valid copy of some data,
2265 2269 * don't allow it to be offlined. Log devices are always
2266 2270 * expendable.
2267 2271 */
2268 2272 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2269 2273 vdev_dtl_required(vd))
2270 2274 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2271 2275
2272 2276 /*
2273 2277 * If the top-level is a slog and it has had allocations
2274 2278 * then proceed. We check that the vdev's metaslab group
2275 2279 * is not NULL since it's possible that we may have just
2276 2280 * added this vdev but not yet initialized its metaslabs.
2277 2281 */
2278 2282 if (tvd->vdev_islog && mg != NULL) {
2279 2283 /*
2280 2284 * Prevent any future allocations.
2281 2285 */
2282 2286 metaslab_group_passivate(mg);
2283 2287 (void) spa_vdev_state_exit(spa, vd, 0);
2284 2288
2285 2289 error = spa_offline_log(spa);
2286 2290
2287 2291 spa_vdev_state_enter(spa, SCL_ALLOC);
2288 2292
2289 2293 /*
2290 2294 * Check to see if the config has changed.
2291 2295 */
2292 2296 if (error || generation != spa->spa_config_generation) {
2293 2297 metaslab_group_activate(mg);
2294 2298 if (error)
2295 2299 return (spa_vdev_state_exit(spa,
2296 2300 vd, error));
2297 2301 (void) spa_vdev_state_exit(spa, vd, 0);
2298 2302 goto top;
2299 2303 }
2300 2304 ASSERT0(tvd->vdev_stat.vs_alloc);
2301 2305 }
2302 2306
2303 2307 /*
2304 2308 * Offline this device and reopen its top-level vdev.
2305 2309 * If the top-level vdev is a log device then just offline
2306 2310 * it. Otherwise, if this action results in the top-level
2307 2311 * vdev becoming unusable, undo it and fail the request.
2308 2312 */
2309 2313 vd->vdev_offline = B_TRUE;
2310 2314 vdev_reopen(tvd);
2311 2315
2312 2316 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2313 2317 vdev_is_dead(tvd)) {
2314 2318 vd->vdev_offline = B_FALSE;
2315 2319 vdev_reopen(tvd);
2316 2320 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2317 2321 }
2318 2322
2319 2323 /*
2320 2324 * Add the device back into the metaslab rotor so that
2321 2325 * once we online the device it's open for business.
2322 2326 */
2323 2327 if (tvd->vdev_islog && mg != NULL)
2324 2328 metaslab_group_activate(mg);
2325 2329 }
2326 2330
2327 2331 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2328 2332
2329 2333 return (spa_vdev_state_exit(spa, vd, 0));
2330 2334 }
2331 2335
2332 2336 int
2333 2337 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2334 2338 {
2335 2339 int error;
2336 2340
2337 2341 mutex_enter(&spa->spa_vdev_top_lock);
2338 2342 error = vdev_offline_locked(spa, guid, flags);
2339 2343 mutex_exit(&spa->spa_vdev_top_lock);
2340 2344
2341 2345 return (error);
2342 2346 }
2343 2347
2344 2348 /*
2345 2349 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2346 2350 * vdev_offline(), we assume the spa config is locked. We also clear all
2347 2351 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2348 2352 */
2349 2353 void
2350 2354 vdev_clear(spa_t *spa, vdev_t *vd)
2351 2355 {
2352 2356 vdev_t *rvd = spa->spa_root_vdev;
2353 2357
2354 2358 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2355 2359
2356 2360 if (vd == NULL)
2357 2361 vd = rvd;
2358 2362
2359 2363 vd->vdev_stat.vs_read_errors = 0;
2360 2364 vd->vdev_stat.vs_write_errors = 0;
2361 2365 vd->vdev_stat.vs_checksum_errors = 0;
2362 2366
2363 2367 for (int c = 0; c < vd->vdev_children; c++)
2364 2368 vdev_clear(spa, vd->vdev_child[c]);
2365 2369
2366 2370 /*
2367 2371 * If we're in the FAULTED state or have experienced failed I/O, then
2368 2372 * clear the persistent state and attempt to reopen the device. We
2369 2373 * also mark the vdev config dirty, so that the new faulted state is
2370 2374 * written out to disk.
2371 2375 */
2372 2376 if (vd->vdev_faulted || vd->vdev_degraded ||
2373 2377 !vdev_readable(vd) || !vdev_writeable(vd)) {
2374 2378
2375 2379 /*
2376 2380 * When reopening in reponse to a clear event, it may be due to
2377 2381 * a fmadm repair request. In this case, if the device is
2378 2382 * still broken, we want to still post the ereport again.
2379 2383 */
2380 2384 vd->vdev_forcefault = B_TRUE;
2381 2385
2382 2386 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2383 2387 vd->vdev_cant_read = B_FALSE;
2384 2388 vd->vdev_cant_write = B_FALSE;
2385 2389
2386 2390 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2387 2391
2388 2392 vd->vdev_forcefault = B_FALSE;
2389 2393
2390 2394 if (vd != rvd && vdev_writeable(vd->vdev_top))
2391 2395 vdev_state_dirty(vd->vdev_top);
2392 2396
2393 2397 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2394 2398 spa_async_request(spa, SPA_ASYNC_RESILVER);
2395 2399
2396 2400 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2397 2401 }
2398 2402
2399 2403 /*
2400 2404 * When clearing a FMA-diagnosed fault, we always want to
2401 2405 * unspare the device, as we assume that the original spare was
2402 2406 * done in response to the FMA fault.
2403 2407 */
2404 2408 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2405 2409 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2406 2410 vd->vdev_parent->vdev_child[0] == vd)
2407 2411 vd->vdev_unspare = B_TRUE;
2408 2412 }
2409 2413
2410 2414 boolean_t
2411 2415 vdev_is_dead(vdev_t *vd)
2412 2416 {
2413 2417 /*
2414 2418 * Holes and missing devices are always considered "dead".
2415 2419 * This simplifies the code since we don't have to check for
2416 2420 * these types of devices in the various code paths.
2417 2421 * Instead we rely on the fact that we skip over dead devices
2418 2422 * before issuing I/O to them.
2419 2423 */
2420 2424 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2421 2425 vd->vdev_ops == &vdev_missing_ops);
2422 2426 }
2423 2427
2424 2428 boolean_t
2425 2429 vdev_readable(vdev_t *vd)
2426 2430 {
2427 2431 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2428 2432 }
2429 2433
2430 2434 boolean_t
2431 2435 vdev_writeable(vdev_t *vd)
2432 2436 {
2433 2437 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2434 2438 }
2435 2439
2436 2440 boolean_t
2437 2441 vdev_allocatable(vdev_t *vd)
2438 2442 {
2439 2443 uint64_t state = vd->vdev_state;
2440 2444
2441 2445 /*
2442 2446 * We currently allow allocations from vdevs which may be in the
2443 2447 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2444 2448 * fails to reopen then we'll catch it later when we're holding
2445 2449 * the proper locks. Note that we have to get the vdev state
2446 2450 * in a local variable because although it changes atomically,
2447 2451 * we're asking two separate questions about it.
2448 2452 */
2449 2453 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2450 2454 !vd->vdev_cant_write && !vd->vdev_ishole);
2451 2455 }
2452 2456
2453 2457 boolean_t
2454 2458 vdev_accessible(vdev_t *vd, zio_t *zio)
2455 2459 {
2456 2460 ASSERT(zio->io_vd == vd);
2457 2461
2458 2462 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2459 2463 return (B_FALSE);
2460 2464
2461 2465 if (zio->io_type == ZIO_TYPE_READ)
2462 2466 return (!vd->vdev_cant_read);
2463 2467
2464 2468 if (zio->io_type == ZIO_TYPE_WRITE)
2465 2469 return (!vd->vdev_cant_write);
2466 2470
2467 2471 return (B_TRUE);
2468 2472 }
2469 2473
2470 2474 /*
2471 2475 * Get statistics for the given vdev.
2472 2476 */
2473 2477 void
2474 2478 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2475 2479 {
2476 2480 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2477 2481
2478 2482 mutex_enter(&vd->vdev_stat_lock);
2479 2483 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2480 2484 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2481 2485 vs->vs_state = vd->vdev_state;
2482 2486 vs->vs_rsize = vdev_get_min_asize(vd);
2483 2487 if (vd->vdev_ops->vdev_op_leaf)
2484 2488 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2485 2489 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2486 2490 mutex_exit(&vd->vdev_stat_lock);
2487 2491
2488 2492 /*
2489 2493 * If we're getting stats on the root vdev, aggregate the I/O counts
2490 2494 * over all top-level vdevs (i.e. the direct children of the root).
2491 2495 */
2492 2496 if (vd == rvd) {
2493 2497 for (int c = 0; c < rvd->vdev_children; c++) {
2494 2498 vdev_t *cvd = rvd->vdev_child[c];
2495 2499 vdev_stat_t *cvs = &cvd->vdev_stat;
2496 2500
2497 2501 mutex_enter(&vd->vdev_stat_lock);
2498 2502 for (int t = 0; t < ZIO_TYPES; t++) {
2499 2503 vs->vs_ops[t] += cvs->vs_ops[t];
2500 2504 vs->vs_bytes[t] += cvs->vs_bytes[t];
2501 2505 }
2502 2506 cvs->vs_scan_removing = cvd->vdev_removing;
2503 2507 mutex_exit(&vd->vdev_stat_lock);
2504 2508 }
2505 2509 }
2506 2510 }
2507 2511
2508 2512 void
2509 2513 vdev_clear_stats(vdev_t *vd)
2510 2514 {
2511 2515 mutex_enter(&vd->vdev_stat_lock);
2512 2516 vd->vdev_stat.vs_space = 0;
2513 2517 vd->vdev_stat.vs_dspace = 0;
2514 2518 vd->vdev_stat.vs_alloc = 0;
2515 2519 mutex_exit(&vd->vdev_stat_lock);
2516 2520 }
2517 2521
2518 2522 void
2519 2523 vdev_scan_stat_init(vdev_t *vd)
2520 2524 {
2521 2525 vdev_stat_t *vs = &vd->vdev_stat;
2522 2526
2523 2527 for (int c = 0; c < vd->vdev_children; c++)
2524 2528 vdev_scan_stat_init(vd->vdev_child[c]);
2525 2529
2526 2530 mutex_enter(&vd->vdev_stat_lock);
2527 2531 vs->vs_scan_processed = 0;
2528 2532 mutex_exit(&vd->vdev_stat_lock);
2529 2533 }
2530 2534
2531 2535 void
2532 2536 vdev_stat_update(zio_t *zio, uint64_t psize)
2533 2537 {
2534 2538 spa_t *spa = zio->io_spa;
2535 2539 vdev_t *rvd = spa->spa_root_vdev;
2536 2540 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2537 2541 vdev_t *pvd;
2538 2542 uint64_t txg = zio->io_txg;
2539 2543 vdev_stat_t *vs = &vd->vdev_stat;
2540 2544 zio_type_t type = zio->io_type;
2541 2545 int flags = zio->io_flags;
2542 2546
2543 2547 /*
2544 2548 * If this i/o is a gang leader, it didn't do any actual work.
2545 2549 */
2546 2550 if (zio->io_gang_tree)
2547 2551 return;
2548 2552
2549 2553 if (zio->io_error == 0) {
2550 2554 /*
2551 2555 * If this is a root i/o, don't count it -- we've already
2552 2556 * counted the top-level vdevs, and vdev_get_stats() will
2553 2557 * aggregate them when asked. This reduces contention on
2554 2558 * the root vdev_stat_lock and implicitly handles blocks
2555 2559 * that compress away to holes, for which there is no i/o.
2556 2560 * (Holes never create vdev children, so all the counters
2557 2561 * remain zero, which is what we want.)
2558 2562 *
2559 2563 * Note: this only applies to successful i/o (io_error == 0)
2560 2564 * because unlike i/o counts, errors are not additive.
2561 2565 * When reading a ditto block, for example, failure of
2562 2566 * one top-level vdev does not imply a root-level error.
2563 2567 */
2564 2568 if (vd == rvd)
2565 2569 return;
2566 2570
2567 2571 ASSERT(vd == zio->io_vd);
2568 2572
2569 2573 if (flags & ZIO_FLAG_IO_BYPASS)
2570 2574 return;
2571 2575
2572 2576 mutex_enter(&vd->vdev_stat_lock);
2573 2577
2574 2578 if (flags & ZIO_FLAG_IO_REPAIR) {
2575 2579 if (flags & ZIO_FLAG_SCAN_THREAD) {
2576 2580 dsl_scan_phys_t *scn_phys =
2577 2581 &spa->spa_dsl_pool->dp_scan->scn_phys;
2578 2582 uint64_t *processed = &scn_phys->scn_processed;
2579 2583
2580 2584 /* XXX cleanup? */
2581 2585 if (vd->vdev_ops->vdev_op_leaf)
2582 2586 atomic_add_64(processed, psize);
2583 2587 vs->vs_scan_processed += psize;
2584 2588 }
2585 2589
2586 2590 if (flags & ZIO_FLAG_SELF_HEAL)
2587 2591 vs->vs_self_healed += psize;
2588 2592 }
2589 2593
2590 2594 vs->vs_ops[type]++;
2591 2595 vs->vs_bytes[type] += psize;
2592 2596
2593 2597 mutex_exit(&vd->vdev_stat_lock);
2594 2598 return;
2595 2599 }
2596 2600
2597 2601 if (flags & ZIO_FLAG_SPECULATIVE)
2598 2602 return;
2599 2603
2600 2604 /*
2601 2605 * If this is an I/O error that is going to be retried, then ignore the
2602 2606 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2603 2607 * hard errors, when in reality they can happen for any number of
2604 2608 * innocuous reasons (bus resets, MPxIO link failure, etc).
2605 2609 */
2606 2610 if (zio->io_error == EIO &&
2607 2611 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2608 2612 return;
2609 2613
2610 2614 /*
2611 2615 * Intent logs writes won't propagate their error to the root
2612 2616 * I/O so don't mark these types of failures as pool-level
2613 2617 * errors.
2614 2618 */
2615 2619 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2616 2620 return;
2617 2621
2618 2622 mutex_enter(&vd->vdev_stat_lock);
2619 2623 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2620 2624 if (zio->io_error == ECKSUM)
2621 2625 vs->vs_checksum_errors++;
2622 2626 else
2623 2627 vs->vs_read_errors++;
2624 2628 }
2625 2629 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2626 2630 vs->vs_write_errors++;
2627 2631 mutex_exit(&vd->vdev_stat_lock);
2628 2632
2629 2633 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2630 2634 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2631 2635 (flags & ZIO_FLAG_SCAN_THREAD) ||
2632 2636 spa->spa_claiming)) {
2633 2637 /*
2634 2638 * This is either a normal write (not a repair), or it's
2635 2639 * a repair induced by the scrub thread, or it's a repair
2636 2640 * made by zil_claim() during spa_load() in the first txg.
2637 2641 * In the normal case, we commit the DTL change in the same
2638 2642 * txg as the block was born. In the scrub-induced repair
2639 2643 * case, we know that scrubs run in first-pass syncing context,
2640 2644 * so we commit the DTL change in spa_syncing_txg(spa).
2641 2645 * In the zil_claim() case, we commit in spa_first_txg(spa).
2642 2646 *
2643 2647 * We currently do not make DTL entries for failed spontaneous
2644 2648 * self-healing writes triggered by normal (non-scrubbing)
2645 2649 * reads, because we have no transactional context in which to
2646 2650 * do so -- and it's not clear that it'd be desirable anyway.
2647 2651 */
2648 2652 if (vd->vdev_ops->vdev_op_leaf) {
2649 2653 uint64_t commit_txg = txg;
2650 2654 if (flags & ZIO_FLAG_SCAN_THREAD) {
2651 2655 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2652 2656 ASSERT(spa_sync_pass(spa) == 1);
2653 2657 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2654 2658 commit_txg = spa_syncing_txg(spa);
2655 2659 } else if (spa->spa_claiming) {
2656 2660 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2657 2661 commit_txg = spa_first_txg(spa);
2658 2662 }
2659 2663 ASSERT(commit_txg >= spa_syncing_txg(spa));
2660 2664 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2661 2665 return;
2662 2666 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2663 2667 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2664 2668 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2665 2669 }
2666 2670 if (vd != rvd)
2667 2671 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2668 2672 }
2669 2673 }
2670 2674
2671 2675 /*
2672 2676 * Update the in-core space usage stats for this vdev, its metaslab class,
2673 2677 * and the root vdev.
2674 2678 */
2675 2679 void
2676 2680 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2677 2681 int64_t space_delta)
2678 2682 {
2679 2683 int64_t dspace_delta = space_delta;
2680 2684 spa_t *spa = vd->vdev_spa;
2681 2685 vdev_t *rvd = spa->spa_root_vdev;
2682 2686 metaslab_group_t *mg = vd->vdev_mg;
2683 2687 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2684 2688
2685 2689 ASSERT(vd == vd->vdev_top);
2686 2690
2687 2691 /*
2688 2692 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2689 2693 * factor. We must calculate this here and not at the root vdev
2690 2694 * because the root vdev's psize-to-asize is simply the max of its
2691 2695 * childrens', thus not accurate enough for us.
2692 2696 */
2693 2697 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2694 2698 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2695 2699 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2696 2700 vd->vdev_deflate_ratio;
2697 2701
2698 2702 mutex_enter(&vd->vdev_stat_lock);
2699 2703 vd->vdev_stat.vs_alloc += alloc_delta;
2700 2704 vd->vdev_stat.vs_space += space_delta;
2701 2705 vd->vdev_stat.vs_dspace += dspace_delta;
2702 2706 mutex_exit(&vd->vdev_stat_lock);
2703 2707
2704 2708 if (mc == spa_normal_class(spa)) {
2705 2709 mutex_enter(&rvd->vdev_stat_lock);
2706 2710 rvd->vdev_stat.vs_alloc += alloc_delta;
2707 2711 rvd->vdev_stat.vs_space += space_delta;
2708 2712 rvd->vdev_stat.vs_dspace += dspace_delta;
2709 2713 mutex_exit(&rvd->vdev_stat_lock);
2710 2714 }
2711 2715
2712 2716 if (mc != NULL) {
2713 2717 ASSERT(rvd == vd->vdev_parent);
2714 2718 ASSERT(vd->vdev_ms_count != 0);
2715 2719
2716 2720 metaslab_class_space_update(mc,
2717 2721 alloc_delta, defer_delta, space_delta, dspace_delta);
2718 2722 }
2719 2723 }
2720 2724
2721 2725 /*
2722 2726 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2723 2727 * so that it will be written out next time the vdev configuration is synced.
2724 2728 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2725 2729 */
2726 2730 void
2727 2731 vdev_config_dirty(vdev_t *vd)
2728 2732 {
2729 2733 spa_t *spa = vd->vdev_spa;
2730 2734 vdev_t *rvd = spa->spa_root_vdev;
2731 2735 int c;
2732 2736
2733 2737 ASSERT(spa_writeable(spa));
2734 2738
2735 2739 /*
2736 2740 * If this is an aux vdev (as with l2cache and spare devices), then we
2737 2741 * update the vdev config manually and set the sync flag.
2738 2742 */
2739 2743 if (vd->vdev_aux != NULL) {
2740 2744 spa_aux_vdev_t *sav = vd->vdev_aux;
2741 2745 nvlist_t **aux;
2742 2746 uint_t naux;
2743 2747
2744 2748 for (c = 0; c < sav->sav_count; c++) {
2745 2749 if (sav->sav_vdevs[c] == vd)
2746 2750 break;
2747 2751 }
2748 2752
2749 2753 if (c == sav->sav_count) {
2750 2754 /*
2751 2755 * We're being removed. There's nothing more to do.
2752 2756 */
2753 2757 ASSERT(sav->sav_sync == B_TRUE);
2754 2758 return;
2755 2759 }
2756 2760
2757 2761 sav->sav_sync = B_TRUE;
2758 2762
2759 2763 if (nvlist_lookup_nvlist_array(sav->sav_config,
2760 2764 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2761 2765 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2762 2766 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2763 2767 }
2764 2768
2765 2769 ASSERT(c < naux);
2766 2770
2767 2771 /*
2768 2772 * Setting the nvlist in the middle if the array is a little
2769 2773 * sketchy, but it will work.
2770 2774 */
2771 2775 nvlist_free(aux[c]);
2772 2776 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2773 2777
2774 2778 return;
2775 2779 }
2776 2780
2777 2781 /*
2778 2782 * The dirty list is protected by the SCL_CONFIG lock. The caller
2779 2783 * must either hold SCL_CONFIG as writer, or must be the sync thread
2780 2784 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2781 2785 * so this is sufficient to ensure mutual exclusion.
2782 2786 */
2783 2787 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2784 2788 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2785 2789 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2786 2790
2787 2791 if (vd == rvd) {
2788 2792 for (c = 0; c < rvd->vdev_children; c++)
2789 2793 vdev_config_dirty(rvd->vdev_child[c]);
2790 2794 } else {
2791 2795 ASSERT(vd == vd->vdev_top);
2792 2796
2793 2797 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2794 2798 !vd->vdev_ishole)
2795 2799 list_insert_head(&spa->spa_config_dirty_list, vd);
2796 2800 }
2797 2801 }
2798 2802
2799 2803 void
2800 2804 vdev_config_clean(vdev_t *vd)
2801 2805 {
2802 2806 spa_t *spa = vd->vdev_spa;
2803 2807
2804 2808 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2805 2809 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2806 2810 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2807 2811
2808 2812 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2809 2813 list_remove(&spa->spa_config_dirty_list, vd);
2810 2814 }
2811 2815
2812 2816 /*
2813 2817 * Mark a top-level vdev's state as dirty, so that the next pass of
2814 2818 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2815 2819 * the state changes from larger config changes because they require
2816 2820 * much less locking, and are often needed for administrative actions.
2817 2821 */
2818 2822 void
2819 2823 vdev_state_dirty(vdev_t *vd)
2820 2824 {
2821 2825 spa_t *spa = vd->vdev_spa;
2822 2826
2823 2827 ASSERT(spa_writeable(spa));
2824 2828 ASSERT(vd == vd->vdev_top);
2825 2829
2826 2830 /*
2827 2831 * The state list is protected by the SCL_STATE lock. The caller
2828 2832 * must either hold SCL_STATE as writer, or must be the sync thread
2829 2833 * (which holds SCL_STATE as reader). There's only one sync thread,
2830 2834 * so this is sufficient to ensure mutual exclusion.
2831 2835 */
2832 2836 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2833 2837 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2834 2838 spa_config_held(spa, SCL_STATE, RW_READER)));
2835 2839
2836 2840 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2837 2841 list_insert_head(&spa->spa_state_dirty_list, vd);
2838 2842 }
2839 2843
2840 2844 void
2841 2845 vdev_state_clean(vdev_t *vd)
2842 2846 {
2843 2847 spa_t *spa = vd->vdev_spa;
2844 2848
2845 2849 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2846 2850 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2847 2851 spa_config_held(spa, SCL_STATE, RW_READER)));
2848 2852
2849 2853 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2850 2854 list_remove(&spa->spa_state_dirty_list, vd);
2851 2855 }
2852 2856
2853 2857 /*
2854 2858 * Propagate vdev state up from children to parent.
2855 2859 */
2856 2860 void
2857 2861 vdev_propagate_state(vdev_t *vd)
2858 2862 {
2859 2863 spa_t *spa = vd->vdev_spa;
2860 2864 vdev_t *rvd = spa->spa_root_vdev;
2861 2865 int degraded = 0, faulted = 0;
2862 2866 int corrupted = 0;
2863 2867 vdev_t *child;
2864 2868
2865 2869 if (vd->vdev_children > 0) {
2866 2870 for (int c = 0; c < vd->vdev_children; c++) {
2867 2871 child = vd->vdev_child[c];
2868 2872
2869 2873 /*
2870 2874 * Don't factor holes into the decision.
2871 2875 */
2872 2876 if (child->vdev_ishole)
2873 2877 continue;
2874 2878
2875 2879 if (!vdev_readable(child) ||
2876 2880 (!vdev_writeable(child) && spa_writeable(spa))) {
2877 2881 /*
2878 2882 * Root special: if there is a top-level log
2879 2883 * device, treat the root vdev as if it were
2880 2884 * degraded.
2881 2885 */
2882 2886 if (child->vdev_islog && vd == rvd)
2883 2887 degraded++;
2884 2888 else
2885 2889 faulted++;
2886 2890 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2887 2891 degraded++;
2888 2892 }
2889 2893
2890 2894 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2891 2895 corrupted++;
2892 2896 }
2893 2897
2894 2898 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2895 2899
2896 2900 /*
2897 2901 * Root special: if there is a top-level vdev that cannot be
2898 2902 * opened due to corrupted metadata, then propagate the root
2899 2903 * vdev's aux state as 'corrupt' rather than 'insufficient
2900 2904 * replicas'.
2901 2905 */
2902 2906 if (corrupted && vd == rvd &&
2903 2907 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2904 2908 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2905 2909 VDEV_AUX_CORRUPT_DATA);
2906 2910 }
2907 2911
2908 2912 if (vd->vdev_parent)
2909 2913 vdev_propagate_state(vd->vdev_parent);
2910 2914 }
2911 2915
2912 2916 /*
2913 2917 * Set a vdev's state. If this is during an open, we don't update the parent
2914 2918 * state, because we're in the process of opening children depth-first.
2915 2919 * Otherwise, we propagate the change to the parent.
2916 2920 *
2917 2921 * If this routine places a device in a faulted state, an appropriate ereport is
2918 2922 * generated.
2919 2923 */
2920 2924 void
2921 2925 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2922 2926 {
2923 2927 uint64_t save_state;
2924 2928 spa_t *spa = vd->vdev_spa;
2925 2929
2926 2930 if (state == vd->vdev_state) {
2927 2931 vd->vdev_stat.vs_aux = aux;
2928 2932 return;
2929 2933 }
2930 2934
2931 2935 save_state = vd->vdev_state;
2932 2936
2933 2937 vd->vdev_state = state;
2934 2938 vd->vdev_stat.vs_aux = aux;
2935 2939
2936 2940 /*
2937 2941 * If we are setting the vdev state to anything but an open state, then
2938 2942 * always close the underlying device unless the device has requested
2939 2943 * a delayed close (i.e. we're about to remove or fault the device).
2940 2944 * Otherwise, we keep accessible but invalid devices open forever.
2941 2945 * We don't call vdev_close() itself, because that implies some extra
2942 2946 * checks (offline, etc) that we don't want here. This is limited to
2943 2947 * leaf devices, because otherwise closing the device will affect other
2944 2948 * children.
2945 2949 */
2946 2950 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2947 2951 vd->vdev_ops->vdev_op_leaf)
2948 2952 vd->vdev_ops->vdev_op_close(vd);
2949 2953
2950 2954 /*
2951 2955 * If we have brought this vdev back into service, we need
2952 2956 * to notify fmd so that it can gracefully repair any outstanding
2953 2957 * cases due to a missing device. We do this in all cases, even those
2954 2958 * that probably don't correlate to a repaired fault. This is sure to
2955 2959 * catch all cases, and we let the zfs-retire agent sort it out. If
2956 2960 * this is a transient state it's OK, as the retire agent will
2957 2961 * double-check the state of the vdev before repairing it.
2958 2962 */
2959 2963 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2960 2964 vd->vdev_prevstate != state)
2961 2965 zfs_post_state_change(spa, vd);
2962 2966
2963 2967 if (vd->vdev_removed &&
2964 2968 state == VDEV_STATE_CANT_OPEN &&
2965 2969 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2966 2970 /*
2967 2971 * If the previous state is set to VDEV_STATE_REMOVED, then this
2968 2972 * device was previously marked removed and someone attempted to
2969 2973 * reopen it. If this failed due to a nonexistent device, then
2970 2974 * keep the device in the REMOVED state. We also let this be if
2971 2975 * it is one of our special test online cases, which is only
2972 2976 * attempting to online the device and shouldn't generate an FMA
2973 2977 * fault.
2974 2978 */
2975 2979 vd->vdev_state = VDEV_STATE_REMOVED;
2976 2980 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2977 2981 } else if (state == VDEV_STATE_REMOVED) {
2978 2982 vd->vdev_removed = B_TRUE;
2979 2983 } else if (state == VDEV_STATE_CANT_OPEN) {
2980 2984 /*
2981 2985 * If we fail to open a vdev during an import or recovery, we
2982 2986 * mark it as "not available", which signifies that it was
2983 2987 * never there to begin with. Failure to open such a device
2984 2988 * is not considered an error.
2985 2989 */
2986 2990 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2987 2991 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2988 2992 vd->vdev_ops->vdev_op_leaf)
2989 2993 vd->vdev_not_present = 1;
2990 2994
2991 2995 /*
2992 2996 * Post the appropriate ereport. If the 'prevstate' field is
2993 2997 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2994 2998 * that this is part of a vdev_reopen(). In this case, we don't
2995 2999 * want to post the ereport if the device was already in the
2996 3000 * CANT_OPEN state beforehand.
2997 3001 *
2998 3002 * If the 'checkremove' flag is set, then this is an attempt to
2999 3003 * online the device in response to an insertion event. If we
3000 3004 * hit this case, then we have detected an insertion event for a
3001 3005 * faulted or offline device that wasn't in the removed state.
3002 3006 * In this scenario, we don't post an ereport because we are
3003 3007 * about to replace the device, or attempt an online with
3004 3008 * vdev_forcefault, which will generate the fault for us.
3005 3009 */
3006 3010 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3007 3011 !vd->vdev_not_present && !vd->vdev_checkremove &&
3008 3012 vd != spa->spa_root_vdev) {
3009 3013 const char *class;
3010 3014
3011 3015 switch (aux) {
3012 3016 case VDEV_AUX_OPEN_FAILED:
3013 3017 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3014 3018 break;
3015 3019 case VDEV_AUX_CORRUPT_DATA:
3016 3020 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3017 3021 break;
3018 3022 case VDEV_AUX_NO_REPLICAS:
3019 3023 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3020 3024 break;
3021 3025 case VDEV_AUX_BAD_GUID_SUM:
3022 3026 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3023 3027 break;
3024 3028 case VDEV_AUX_TOO_SMALL:
3025 3029 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3026 3030 break;
3027 3031 case VDEV_AUX_BAD_LABEL:
3028 3032 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3029 3033 break;
3030 3034 default:
3031 3035 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3032 3036 }
3033 3037
3034 3038 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3035 3039 }
3036 3040
3037 3041 /* Erase any notion of persistent removed state */
3038 3042 vd->vdev_removed = B_FALSE;
3039 3043 } else {
3040 3044 vd->vdev_removed = B_FALSE;
3041 3045 }
3042 3046
3043 3047 if (!isopen && vd->vdev_parent)
3044 3048 vdev_propagate_state(vd->vdev_parent);
3045 3049 }
3046 3050
3047 3051 /*
3048 3052 * Check the vdev configuration to ensure that it's capable of supporting
3049 3053 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3050 3054 * In addition, only a single top-level vdev is allowed and none of the leaves
3051 3055 * can be wholedisks.
3052 3056 */
3053 3057 boolean_t
3054 3058 vdev_is_bootable(vdev_t *vd)
3055 3059 {
3056 3060 if (!vd->vdev_ops->vdev_op_leaf) {
3057 3061 char *vdev_type = vd->vdev_ops->vdev_op_type;
3058 3062
3059 3063 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3060 3064 vd->vdev_children > 1) {
3061 3065 return (B_FALSE);
3062 3066 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3063 3067 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3064 3068 return (B_FALSE);
3065 3069 }
3066 3070 } else if (vd->vdev_wholedisk == 1) {
3067 3071 return (B_FALSE);
3068 3072 }
3069 3073
3070 3074 for (int c = 0; c < vd->vdev_children; c++) {
3071 3075 if (!vdev_is_bootable(vd->vdev_child[c]))
3072 3076 return (B_FALSE);
3073 3077 }
3074 3078 return (B_TRUE);
3075 3079 }
3076 3080
3077 3081 /*
3078 3082 * Load the state from the original vdev tree (ovd) which
3079 3083 * we've retrieved from the MOS config object. If the original
3080 3084 * vdev was offline or faulted then we transfer that state to the
3081 3085 * device in the current vdev tree (nvd).
3082 3086 */
3083 3087 void
3084 3088 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3085 3089 {
3086 3090 spa_t *spa = nvd->vdev_spa;
3087 3091
3088 3092 ASSERT(nvd->vdev_top->vdev_islog);
3089 3093 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3090 3094 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3091 3095
3092 3096 for (int c = 0; c < nvd->vdev_children; c++)
3093 3097 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3094 3098
3095 3099 if (nvd->vdev_ops->vdev_op_leaf) {
3096 3100 /*
3097 3101 * Restore the persistent vdev state
3098 3102 */
3099 3103 nvd->vdev_offline = ovd->vdev_offline;
3100 3104 nvd->vdev_faulted = ovd->vdev_faulted;
3101 3105 nvd->vdev_degraded = ovd->vdev_degraded;
3102 3106 nvd->vdev_removed = ovd->vdev_removed;
3103 3107 }
3104 3108 }
3105 3109
3106 3110 /*
3107 3111 * Determine if a log device has valid content. If the vdev was
3108 3112 * removed or faulted in the MOS config then we know that
3109 3113 * the content on the log device has already been written to the pool.
3110 3114 */
3111 3115 boolean_t
3112 3116 vdev_log_state_valid(vdev_t *vd)
3113 3117 {
3114 3118 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3115 3119 !vd->vdev_removed)
3116 3120 return (B_TRUE);
3117 3121
3118 3122 for (int c = 0; c < vd->vdev_children; c++)
3119 3123 if (vdev_log_state_valid(vd->vdev_child[c]))
3120 3124 return (B_TRUE);
3121 3125
3122 3126 return (B_FALSE);
3123 3127 }
3124 3128
3125 3129 /*
3126 3130 * Expand a vdev if possible.
3127 3131 */
3128 3132 void
3129 3133 vdev_expand(vdev_t *vd, uint64_t txg)
3130 3134 {
3131 3135 ASSERT(vd->vdev_top == vd);
3132 3136 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3133 3137
3134 3138 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3135 3139 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3136 3140 vdev_config_dirty(vd);
3137 3141 }
3138 3142 }
3139 3143
3140 3144 /*
3141 3145 * Split a vdev.
3142 3146 */
3143 3147 void
3144 3148 vdev_split(vdev_t *vd)
3145 3149 {
3146 3150 vdev_t *cvd, *pvd = vd->vdev_parent;
3147 3151
3148 3152 vdev_remove_child(pvd, vd);
3149 3153 vdev_compact_children(pvd);
3150 3154
3151 3155 cvd = pvd->vdev_child[0];
3152 3156 if (pvd->vdev_children == 1) {
3153 3157 vdev_remove_parent(cvd);
3154 3158 cvd->vdev_splitting = B_TRUE;
3155 3159 }
3156 3160 vdev_propagate_state(cvd);
3157 3161 }
3158 3162
3159 3163 void
3160 3164 vdev_deadman(vdev_t *vd)
3161 3165 {
3162 3166 for (int c = 0; c < vd->vdev_children; c++) {
3163 3167 vdev_t *cvd = vd->vdev_child[c];
3164 3168
3165 3169 vdev_deadman(cvd);
3166 3170 }
3167 3171
3168 3172 if (vd->vdev_ops->vdev_op_leaf) {
3169 3173 vdev_queue_t *vq = &vd->vdev_queue;
3170 3174
3171 3175 mutex_enter(&vq->vq_lock);
3172 3176 if (avl_numnodes(&vq->vq_pending_tree) > 0) {
3173 3177 spa_t *spa = vd->vdev_spa;
3174 3178 zio_t *fio;
3175 3179 uint64_t delta;
3176 3180
3177 3181 /*
3178 3182 * Look at the head of all the pending queues,
3179 3183 * if any I/O has been outstanding for longer than
3180 3184 * the spa_deadman_synctime we panic the system.
3181 3185 */
3182 3186 fio = avl_first(&vq->vq_pending_tree);
3183 3187 delta = gethrtime() - fio->io_timestamp;
3184 3188 if (delta > spa_deadman_synctime(spa)) {
3185 3189 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3186 3190 "delta %lluns, last io %lluns",
3187 3191 fio->io_timestamp, delta,
3188 3192 vq->vq_io_complete_ts);
3189 3193 fm_panic("I/O to pool '%s' appears to be "
3190 3194 "hung.", spa_name(spa));
3191 3195 }
3192 3196 }
3193 3197 mutex_exit(&vq->vq_lock);
3194 3198 }
3195 3199 }
|
↓ open down ↓ |
1000 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX