1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2021 by Delphix. All rights reserved. 25 * Copyright 2017 Nexenta Systems, Inc. 26 * Copyright (c) 2014 Integros [integros.com] 27 * Copyright 2016 Toomas Soome <tsoome@me.com> 28 * Copyright 2017 Joyent, Inc. 29 * Copyright (c) 2017, Intel Corporation. 30 * Copyright (c) 2019, Datto Inc. All rights reserved. 31 * Copyright (c) 2021, Klara Inc. 32 * Copyright [2021] Hewlett Packard Enterprise Development LP 33 */ 34 35 #include <sys/zfs_context.h> 36 #include <sys/fm/fs/zfs.h> 37 #include <sys/spa.h> 38 #include <sys/spa_impl.h> 39 #include <sys/bpobj.h> 40 #include <sys/dmu.h> 41 #include <sys/dmu_tx.h> 42 #include <sys/dsl_dir.h> 43 #include <sys/vdev_impl.h> 44 #include <sys/vdev_rebuild.h> 45 #include <sys/vdev_draid.h> 46 #include <sys/uberblock_impl.h> 47 #include <sys/metaslab.h> 48 #include <sys/metaslab_impl.h> 49 #include <sys/space_map.h> 50 #include <sys/space_reftree.h> 51 #include <sys/zio.h> 52 #include <sys/zap.h> 53 #include <sys/fs/zfs.h> 54 #include <sys/arc.h> 55 #include <sys/zil.h> 56 #include <sys/dsl_scan.h> 57 #include <sys/vdev_raidz.h> 58 #include <sys/abd.h> 59 #include <sys/vdev_initialize.h> 60 #include <sys/vdev_trim.h> 61 #include <sys/zvol.h> 62 #include <sys/zfs_ratelimit.h> 63 #include "zfs_prop.h" 64 65 /* 66 * One metaslab from each (normal-class) vdev is used by the ZIL. These are 67 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are 68 * part of the spa_embedded_log_class. The metaslab with the most free space 69 * in each vdev is selected for this purpose when the pool is opened (or a 70 * vdev is added). See vdev_metaslab_init(). 71 * 72 * Log blocks can be allocated from the following locations. Each one is tried 73 * in order until the allocation succeeds: 74 * 1. dedicated log vdevs, aka "slog" (spa_log_class) 75 * 2. embedded slog metaslabs (spa_embedded_log_class) 76 * 3. other metaslabs in normal vdevs (spa_normal_class) 77 * 78 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer 79 * than this number of metaslabs in the vdev. This ensures that we don't set 80 * aside an unreasonable amount of space for the ZIL. If set to less than 81 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced 82 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab. 83 */ 84 static uint_t zfs_embedded_slog_min_ms = 64; 85 86 /* default target for number of metaslabs per top-level vdev */ 87 static uint_t zfs_vdev_default_ms_count = 200; 88 89 /* minimum number of metaslabs per top-level vdev */ 90 static uint_t zfs_vdev_min_ms_count = 16; 91 92 /* practical upper limit of total metaslabs per top-level vdev */ 93 static uint_t zfs_vdev_ms_count_limit = 1ULL << 17; 94 95 /* lower limit for metaslab size (512M) */ 96 static uint_t zfs_vdev_default_ms_shift = 29; 97 98 /* upper limit for metaslab size (16G) */ 99 static const uint_t zfs_vdev_max_ms_shift = 34; 100 101 int vdev_validate_skip = B_FALSE; 102 103 /* 104 * Since the DTL space map of a vdev is not expected to have a lot of 105 * entries, we default its block size to 4K. 106 */ 107 int zfs_vdev_dtl_sm_blksz = (1 << 12); 108 109 /* 110 * Rate limit slow IO (delay) events to this many per second. 111 */ 112 static unsigned int zfs_slow_io_events_per_second = 20; 113 114 /* 115 * Rate limit checksum events after this many checksum errors per second. 116 */ 117 static unsigned int zfs_checksum_events_per_second = 20; 118 119 /* 120 * Ignore errors during scrub/resilver. Allows to work around resilver 121 * upon import when there are pool errors. 122 */ 123 static int zfs_scan_ignore_errors = 0; 124 125 /* 126 * vdev-wide space maps that have lots of entries written to them at 127 * the end of each transaction can benefit from a higher I/O bandwidth 128 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 129 */ 130 int zfs_vdev_standard_sm_blksz = (1 << 17); 131 132 /* 133 * Tunable parameter for debugging or performance analysis. Setting this 134 * will cause pool corruption on power loss if a volatile out-of-order 135 * write cache is enabled. 136 */ 137 int zfs_nocacheflush = 0; 138 139 /* 140 * Maximum and minimum ashift values that can be automatically set based on 141 * vdev's physical ashift (disk's physical sector size). While ASHIFT_MAX 142 * is higher than the maximum value, it is intentionally limited here to not 143 * excessively impact pool space efficiency. Higher ashift values may still 144 * be forced by vdev logical ashift or by user via ashift property, but won't 145 * be set automatically as a performance optimization. 146 */ 147 uint64_t zfs_vdev_max_auto_ashift = 14; 148 uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN; 149 150 void 151 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 152 { 153 va_list adx; 154 char buf[256]; 155 156 va_start(adx, fmt); 157 (void) vsnprintf(buf, sizeof (buf), fmt, adx); 158 va_end(adx); 159 160 if (vd->vdev_path != NULL) { 161 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 162 vd->vdev_path, buf); 163 } else { 164 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 165 vd->vdev_ops->vdev_op_type, 166 (u_longlong_t)vd->vdev_id, 167 (u_longlong_t)vd->vdev_guid, buf); 168 } 169 } 170 171 void 172 vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 173 { 174 char state[20]; 175 176 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 177 zfs_dbgmsg("%*svdev %llu: %s", indent, "", 178 (u_longlong_t)vd->vdev_id, 179 vd->vdev_ops->vdev_op_type); 180 return; 181 } 182 183 switch (vd->vdev_state) { 184 case VDEV_STATE_UNKNOWN: 185 (void) snprintf(state, sizeof (state), "unknown"); 186 break; 187 case VDEV_STATE_CLOSED: 188 (void) snprintf(state, sizeof (state), "closed"); 189 break; 190 case VDEV_STATE_OFFLINE: 191 (void) snprintf(state, sizeof (state), "offline"); 192 break; 193 case VDEV_STATE_REMOVED: 194 (void) snprintf(state, sizeof (state), "removed"); 195 break; 196 case VDEV_STATE_CANT_OPEN: 197 (void) snprintf(state, sizeof (state), "can't open"); 198 break; 199 case VDEV_STATE_FAULTED: 200 (void) snprintf(state, sizeof (state), "faulted"); 201 break; 202 case VDEV_STATE_DEGRADED: 203 (void) snprintf(state, sizeof (state), "degraded"); 204 break; 205 case VDEV_STATE_HEALTHY: 206 (void) snprintf(state, sizeof (state), "healthy"); 207 break; 208 default: 209 (void) snprintf(state, sizeof (state), "<state %u>", 210 (uint_t)vd->vdev_state); 211 } 212 213 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 214 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 215 vd->vdev_islog ? " (log)" : "", 216 (u_longlong_t)vd->vdev_guid, 217 vd->vdev_path ? vd->vdev_path : "N/A", state); 218 219 for (uint64_t i = 0; i < vd->vdev_children; i++) 220 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 221 } 222 223 /* 224 * Virtual device management. 225 */ 226 227 static vdev_ops_t *const vdev_ops_table[] = { 228 &vdev_root_ops, 229 &vdev_raidz_ops, 230 &vdev_draid_ops, 231 &vdev_draid_spare_ops, 232 &vdev_mirror_ops, 233 &vdev_replacing_ops, 234 &vdev_spare_ops, 235 &vdev_disk_ops, 236 &vdev_file_ops, 237 &vdev_missing_ops, 238 &vdev_hole_ops, 239 &vdev_indirect_ops, 240 NULL 241 }; 242 243 /* 244 * Given a vdev type, return the appropriate ops vector. 245 */ 246 static vdev_ops_t * 247 vdev_getops(const char *type) 248 { 249 vdev_ops_t *ops, *const *opspp; 250 251 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 252 if (strcmp(ops->vdev_op_type, type) == 0) 253 break; 254 255 return (ops); 256 } 257 258 /* 259 * Given a vdev and a metaslab class, find which metaslab group we're 260 * interested in. All vdevs may belong to two different metaslab classes. 261 * Dedicated slog devices use only the primary metaslab group, rather than a 262 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL. 263 */ 264 metaslab_group_t * 265 vdev_get_mg(vdev_t *vd, metaslab_class_t *mc) 266 { 267 if (mc == spa_embedded_log_class(vd->vdev_spa) && 268 vd->vdev_log_mg != NULL) 269 return (vd->vdev_log_mg); 270 else 271 return (vd->vdev_mg); 272 } 273 274 void 275 vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 276 range_seg64_t *physical_rs, range_seg64_t *remain_rs) 277 { 278 (void) vd, (void) remain_rs; 279 280 physical_rs->rs_start = logical_rs->rs_start; 281 physical_rs->rs_end = logical_rs->rs_end; 282 } 283 284 /* 285 * Derive the enumerated allocation bias from string input. 286 * String origin is either the per-vdev zap or zpool(8). 287 */ 288 static vdev_alloc_bias_t 289 vdev_derive_alloc_bias(const char *bias) 290 { 291 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 292 293 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 294 alloc_bias = VDEV_BIAS_LOG; 295 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 296 alloc_bias = VDEV_BIAS_SPECIAL; 297 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 298 alloc_bias = VDEV_BIAS_DEDUP; 299 300 return (alloc_bias); 301 } 302 303 /* 304 * Default asize function: return the MAX of psize with the asize of 305 * all children. This is what's used by anything other than RAID-Z. 306 */ 307 uint64_t 308 vdev_default_asize(vdev_t *vd, uint64_t psize) 309 { 310 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 311 uint64_t csize; 312 313 for (int c = 0; c < vd->vdev_children; c++) { 314 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 315 asize = MAX(asize, csize); 316 } 317 318 return (asize); 319 } 320 321 uint64_t 322 vdev_default_min_asize(vdev_t *vd) 323 { 324 return (vd->vdev_min_asize); 325 } 326 327 /* 328 * Get the minimum allocatable size. We define the allocatable size as 329 * the vdev's asize rounded to the nearest metaslab. This allows us to 330 * replace or attach devices which don't have the same physical size but 331 * can still satisfy the same number of allocations. 332 */ 333 uint64_t 334 vdev_get_min_asize(vdev_t *vd) 335 { 336 vdev_t *pvd = vd->vdev_parent; 337 338 /* 339 * If our parent is NULL (inactive spare or cache) or is the root, 340 * just return our own asize. 341 */ 342 if (pvd == NULL) 343 return (vd->vdev_asize); 344 345 /* 346 * The top-level vdev just returns the allocatable size rounded 347 * to the nearest metaslab. 348 */ 349 if (vd == vd->vdev_top) 350 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 351 352 return (pvd->vdev_ops->vdev_op_min_asize(pvd)); 353 } 354 355 void 356 vdev_set_min_asize(vdev_t *vd) 357 { 358 vd->vdev_min_asize = vdev_get_min_asize(vd); 359 360 for (int c = 0; c < vd->vdev_children; c++) 361 vdev_set_min_asize(vd->vdev_child[c]); 362 } 363 364 /* 365 * Get the minimal allocation size for the top-level vdev. 366 */ 367 uint64_t 368 vdev_get_min_alloc(vdev_t *vd) 369 { 370 uint64_t min_alloc = 1ULL << vd->vdev_ashift; 371 372 if (vd->vdev_ops->vdev_op_min_alloc != NULL) 373 min_alloc = vd->vdev_ops->vdev_op_min_alloc(vd); 374 375 return (min_alloc); 376 } 377 378 /* 379 * Get the parity level for a top-level vdev. 380 */ 381 uint64_t 382 vdev_get_nparity(vdev_t *vd) 383 { 384 uint64_t nparity = 0; 385 386 if (vd->vdev_ops->vdev_op_nparity != NULL) 387 nparity = vd->vdev_ops->vdev_op_nparity(vd); 388 389 return (nparity); 390 } 391 392 /* 393 * Get the number of data disks for a top-level vdev. 394 */ 395 uint64_t 396 vdev_get_ndisks(vdev_t *vd) 397 { 398 uint64_t ndisks = 1; 399 400 if (vd->vdev_ops->vdev_op_ndisks != NULL) 401 ndisks = vd->vdev_ops->vdev_op_ndisks(vd); 402 403 return (ndisks); 404 } 405 406 vdev_t * 407 vdev_lookup_top(spa_t *spa, uint64_t vdev) 408 { 409 vdev_t *rvd = spa->spa_root_vdev; 410 411 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 412 413 if (vdev < rvd->vdev_children) { 414 ASSERT(rvd->vdev_child[vdev] != NULL); 415 return (rvd->vdev_child[vdev]); 416 } 417 418 return (NULL); 419 } 420 421 vdev_t * 422 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 423 { 424 vdev_t *mvd; 425 426 if (vd->vdev_guid == guid) 427 return (vd); 428 429 for (int c = 0; c < vd->vdev_children; c++) 430 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 431 NULL) 432 return (mvd); 433 434 return (NULL); 435 } 436 437 static int 438 vdev_count_leaves_impl(vdev_t *vd) 439 { 440 int n = 0; 441 442 if (vd->vdev_ops->vdev_op_leaf) 443 return (1); 444 445 for (int c = 0; c < vd->vdev_children; c++) 446 n += vdev_count_leaves_impl(vd->vdev_child[c]); 447 448 return (n); 449 } 450 451 int 452 vdev_count_leaves(spa_t *spa) 453 { 454 int rc; 455 456 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 457 rc = vdev_count_leaves_impl(spa->spa_root_vdev); 458 spa_config_exit(spa, SCL_VDEV, FTAG); 459 460 return (rc); 461 } 462 463 void 464 vdev_add_child(vdev_t *pvd, vdev_t *cvd) 465 { 466 size_t oldsize, newsize; 467 uint64_t id = cvd->vdev_id; 468 vdev_t **newchild; 469 470 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 471 ASSERT(cvd->vdev_parent == NULL); 472 473 cvd->vdev_parent = pvd; 474 475 if (pvd == NULL) 476 return; 477 478 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 479 480 oldsize = pvd->vdev_children * sizeof (vdev_t *); 481 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 482 newsize = pvd->vdev_children * sizeof (vdev_t *); 483 484 newchild = kmem_alloc(newsize, KM_SLEEP); 485 if (pvd->vdev_child != NULL) { 486 memcpy(newchild, pvd->vdev_child, oldsize); 487 kmem_free(pvd->vdev_child, oldsize); 488 } 489 490 pvd->vdev_child = newchild; 491 pvd->vdev_child[id] = cvd; 492 493 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 494 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 495 496 /* 497 * Walk up all ancestors to update guid sum. 498 */ 499 for (; pvd != NULL; pvd = pvd->vdev_parent) 500 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 501 502 if (cvd->vdev_ops->vdev_op_leaf) { 503 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 504 cvd->vdev_spa->spa_leaf_list_gen++; 505 } 506 } 507 508 void 509 vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 510 { 511 int c; 512 uint_t id = cvd->vdev_id; 513 514 ASSERT(cvd->vdev_parent == pvd); 515 516 if (pvd == NULL) 517 return; 518 519 ASSERT(id < pvd->vdev_children); 520 ASSERT(pvd->vdev_child[id] == cvd); 521 522 pvd->vdev_child[id] = NULL; 523 cvd->vdev_parent = NULL; 524 525 for (c = 0; c < pvd->vdev_children; c++) 526 if (pvd->vdev_child[c]) 527 break; 528 529 if (c == pvd->vdev_children) { 530 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 531 pvd->vdev_child = NULL; 532 pvd->vdev_children = 0; 533 } 534 535 if (cvd->vdev_ops->vdev_op_leaf) { 536 spa_t *spa = cvd->vdev_spa; 537 list_remove(&spa->spa_leaf_list, cvd); 538 spa->spa_leaf_list_gen++; 539 } 540 541 /* 542 * Walk up all ancestors to update guid sum. 543 */ 544 for (; pvd != NULL; pvd = pvd->vdev_parent) 545 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 546 } 547 548 /* 549 * Remove any holes in the child array. 550 */ 551 void 552 vdev_compact_children(vdev_t *pvd) 553 { 554 vdev_t **newchild, *cvd; 555 int oldc = pvd->vdev_children; 556 int newc; 557 558 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 559 560 if (oldc == 0) 561 return; 562 563 for (int c = newc = 0; c < oldc; c++) 564 if (pvd->vdev_child[c]) 565 newc++; 566 567 if (newc > 0) { 568 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); 569 570 for (int c = newc = 0; c < oldc; c++) { 571 if ((cvd = pvd->vdev_child[c]) != NULL) { 572 newchild[newc] = cvd; 573 cvd->vdev_id = newc++; 574 } 575 } 576 } else { 577 newchild = NULL; 578 } 579 580 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 581 pvd->vdev_child = newchild; 582 pvd->vdev_children = newc; 583 } 584 585 /* 586 * Allocate and minimally initialize a vdev_t. 587 */ 588 vdev_t * 589 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 590 { 591 vdev_t *vd; 592 vdev_indirect_config_t *vic; 593 594 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 595 vic = &vd->vdev_indirect_config; 596 597 if (spa->spa_root_vdev == NULL) { 598 ASSERT(ops == &vdev_root_ops); 599 spa->spa_root_vdev = vd; 600 spa->spa_load_guid = spa_generate_guid(NULL); 601 } 602 603 if (guid == 0 && ops != &vdev_hole_ops) { 604 if (spa->spa_root_vdev == vd) { 605 /* 606 * The root vdev's guid will also be the pool guid, 607 * which must be unique among all pools. 608 */ 609 guid = spa_generate_guid(NULL); 610 } else { 611 /* 612 * Any other vdev's guid must be unique within the pool. 613 */ 614 guid = spa_generate_guid(spa); 615 } 616 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 617 } 618 619 vd->vdev_spa = spa; 620 vd->vdev_id = id; 621 vd->vdev_guid = guid; 622 vd->vdev_guid_sum = guid; 623 vd->vdev_ops = ops; 624 vd->vdev_state = VDEV_STATE_CLOSED; 625 vd->vdev_ishole = (ops == &vdev_hole_ops); 626 vic->vic_prev_indirect_vdev = UINT64_MAX; 627 628 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 629 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 630 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 631 0, 0); 632 633 /* 634 * Initialize rate limit structs for events. We rate limit ZIO delay 635 * and checksum events so that we don't overwhelm ZED with thousands 636 * of events when a disk is acting up. 637 */ 638 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second, 639 1); 640 zfs_ratelimit_init(&vd->vdev_deadman_rl, &zfs_slow_io_events_per_second, 641 1); 642 zfs_ratelimit_init(&vd->vdev_checksum_rl, 643 &zfs_checksum_events_per_second, 1); 644 645 list_link_init(&vd->vdev_config_dirty_node); 646 list_link_init(&vd->vdev_state_dirty_node); 647 list_link_init(&vd->vdev_initialize_node); 648 list_link_init(&vd->vdev_leaf_node); 649 list_link_init(&vd->vdev_trim_node); 650 651 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); 652 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 653 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 654 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 655 656 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 657 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 658 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 659 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 660 661 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 662 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 663 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 664 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 665 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 666 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 667 668 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL); 669 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL); 670 671 for (int t = 0; t < DTL_TYPES; t++) { 672 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 673 0); 674 } 675 676 txg_list_create(&vd->vdev_ms_list, spa, 677 offsetof(struct metaslab, ms_txg_node)); 678 txg_list_create(&vd->vdev_dtl_list, spa, 679 offsetof(struct vdev, vdev_dtl_node)); 680 vd->vdev_stat.vs_timestamp = gethrtime(); 681 vdev_queue_init(vd); 682 vdev_cache_init(vd); 683 684 return (vd); 685 } 686 687 /* 688 * Allocate a new vdev. The 'alloctype' is used to control whether we are 689 * creating a new vdev or loading an existing one - the behavior is slightly 690 * different for each case. 691 */ 692 int 693 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 694 int alloctype) 695 { 696 vdev_ops_t *ops; 697 char *type; 698 uint64_t guid = 0, islog; 699 vdev_t *vd; 700 vdev_indirect_config_t *vic; 701 char *tmp = NULL; 702 int rc; 703 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 704 boolean_t top_level = (parent && !parent->vdev_parent); 705 706 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 707 708 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 709 return (SET_ERROR(EINVAL)); 710 711 if ((ops = vdev_getops(type)) == NULL) 712 return (SET_ERROR(EINVAL)); 713 714 /* 715 * If this is a load, get the vdev guid from the nvlist. 716 * Otherwise, vdev_alloc_common() will generate one for us. 717 */ 718 if (alloctype == VDEV_ALLOC_LOAD) { 719 uint64_t label_id; 720 721 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 722 label_id != id) 723 return (SET_ERROR(EINVAL)); 724 725 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 726 return (SET_ERROR(EINVAL)); 727 } else if (alloctype == VDEV_ALLOC_SPARE) { 728 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 729 return (SET_ERROR(EINVAL)); 730 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 731 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 732 return (SET_ERROR(EINVAL)); 733 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 734 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 735 return (SET_ERROR(EINVAL)); 736 } 737 738 /* 739 * The first allocated vdev must be of type 'root'. 740 */ 741 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 742 return (SET_ERROR(EINVAL)); 743 744 /* 745 * Determine whether we're a log vdev. 746 */ 747 islog = 0; 748 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 749 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 750 return (SET_ERROR(ENOTSUP)); 751 752 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 753 return (SET_ERROR(ENOTSUP)); 754 755 if (top_level && alloctype == VDEV_ALLOC_ADD) { 756 char *bias; 757 758 /* 759 * If creating a top-level vdev, check for allocation 760 * classes input. 761 */ 762 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 763 &bias) == 0) { 764 alloc_bias = vdev_derive_alloc_bias(bias); 765 766 /* spa_vdev_add() expects feature to be enabled */ 767 if (spa->spa_load_state != SPA_LOAD_CREATE && 768 !spa_feature_is_enabled(spa, 769 SPA_FEATURE_ALLOCATION_CLASSES)) { 770 return (SET_ERROR(ENOTSUP)); 771 } 772 } 773 774 /* spa_vdev_add() expects feature to be enabled */ 775 if (ops == &vdev_draid_ops && 776 spa->spa_load_state != SPA_LOAD_CREATE && 777 !spa_feature_is_enabled(spa, SPA_FEATURE_DRAID)) { 778 return (SET_ERROR(ENOTSUP)); 779 } 780 } 781 782 /* 783 * Initialize the vdev specific data. This is done before calling 784 * vdev_alloc_common() since it may fail and this simplifies the 785 * error reporting and cleanup code paths. 786 */ 787 void *tsd = NULL; 788 if (ops->vdev_op_init != NULL) { 789 rc = ops->vdev_op_init(spa, nv, &tsd); 790 if (rc != 0) { 791 return (rc); 792 } 793 } 794 795 vd = vdev_alloc_common(spa, id, guid, ops); 796 vd->vdev_tsd = tsd; 797 vd->vdev_islog = islog; 798 799 if (top_level && alloc_bias != VDEV_BIAS_NONE) 800 vd->vdev_alloc_bias = alloc_bias; 801 802 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 803 vd->vdev_path = spa_strdup(vd->vdev_path); 804 805 /* 806 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a 807 * fault on a vdev and want it to persist across imports (like with 808 * zpool offline -f). 809 */ 810 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp); 811 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) { 812 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; 813 vd->vdev_faulted = 1; 814 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 815 } 816 817 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 818 vd->vdev_devid = spa_strdup(vd->vdev_devid); 819 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 820 &vd->vdev_physpath) == 0) 821 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 822 823 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, 824 &vd->vdev_enc_sysfs_path) == 0) 825 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); 826 827 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 828 vd->vdev_fru = spa_strdup(vd->vdev_fru); 829 830 /* 831 * Set the whole_disk property. If it's not specified, leave the value 832 * as -1. 833 */ 834 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 835 &vd->vdev_wholedisk) != 0) 836 vd->vdev_wholedisk = -1ULL; 837 838 vic = &vd->vdev_indirect_config; 839 840 ASSERT0(vic->vic_mapping_object); 841 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 842 &vic->vic_mapping_object); 843 ASSERT0(vic->vic_births_object); 844 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 845 &vic->vic_births_object); 846 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 847 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 848 &vic->vic_prev_indirect_vdev); 849 850 /* 851 * Look for the 'not present' flag. This will only be set if the device 852 * was not present at the time of import. 853 */ 854 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 855 &vd->vdev_not_present); 856 857 /* 858 * Get the alignment requirement. 859 */ 860 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 861 862 /* 863 * Retrieve the vdev creation time. 864 */ 865 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 866 &vd->vdev_crtxg); 867 868 /* 869 * If we're a top-level vdev, try to load the allocation parameters. 870 */ 871 if (top_level && 872 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 873 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 874 &vd->vdev_ms_array); 875 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 876 &vd->vdev_ms_shift); 877 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 878 &vd->vdev_asize); 879 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NONALLOCATING, 880 &vd->vdev_noalloc); 881 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 882 &vd->vdev_removing); 883 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 884 &vd->vdev_top_zap); 885 } else { 886 ASSERT0(vd->vdev_top_zap); 887 } 888 889 if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 890 ASSERT(alloctype == VDEV_ALLOC_LOAD || 891 alloctype == VDEV_ALLOC_ADD || 892 alloctype == VDEV_ALLOC_SPLIT || 893 alloctype == VDEV_ALLOC_ROOTPOOL); 894 /* Note: metaslab_group_create() is now deferred */ 895 } 896 897 if (vd->vdev_ops->vdev_op_leaf && 898 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 899 (void) nvlist_lookup_uint64(nv, 900 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 901 } else { 902 ASSERT0(vd->vdev_leaf_zap); 903 } 904 905 /* 906 * If we're a leaf vdev, try to load the DTL object and other state. 907 */ 908 909 if (vd->vdev_ops->vdev_op_leaf && 910 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 911 alloctype == VDEV_ALLOC_ROOTPOOL)) { 912 if (alloctype == VDEV_ALLOC_LOAD) { 913 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 914 &vd->vdev_dtl_object); 915 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 916 &vd->vdev_unspare); 917 } 918 919 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 920 uint64_t spare = 0; 921 922 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 923 &spare) == 0 && spare) 924 spa_spare_add(vd); 925 } 926 927 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 928 &vd->vdev_offline); 929 930 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 931 &vd->vdev_resilver_txg); 932 933 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG, 934 &vd->vdev_rebuild_txg); 935 936 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 937 vdev_defer_resilver(vd); 938 939 /* 940 * In general, when importing a pool we want to ignore the 941 * persistent fault state, as the diagnosis made on another 942 * system may not be valid in the current context. The only 943 * exception is if we forced a vdev to a persistently faulted 944 * state with 'zpool offline -f'. The persistent fault will 945 * remain across imports until cleared. 946 * 947 * Local vdevs will remain in the faulted state. 948 */ 949 if (spa_load_state(spa) == SPA_LOAD_OPEN || 950 spa_load_state(spa) == SPA_LOAD_IMPORT) { 951 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 952 &vd->vdev_faulted); 953 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 954 &vd->vdev_degraded); 955 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 956 &vd->vdev_removed); 957 958 if (vd->vdev_faulted || vd->vdev_degraded) { 959 char *aux; 960 961 vd->vdev_label_aux = 962 VDEV_AUX_ERR_EXCEEDED; 963 if (nvlist_lookup_string(nv, 964 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 965 strcmp(aux, "external") == 0) 966 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 967 else 968 vd->vdev_faulted = 0ULL; 969 } 970 } 971 } 972 973 /* 974 * Add ourselves to the parent's list of children. 975 */ 976 vdev_add_child(parent, vd); 977 978 *vdp = vd; 979 980 return (0); 981 } 982 983 void 984 vdev_free(vdev_t *vd) 985 { 986 spa_t *spa = vd->vdev_spa; 987 988 ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 989 ASSERT3P(vd->vdev_trim_thread, ==, NULL); 990 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 991 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL); 992 993 /* 994 * Scan queues are normally destroyed at the end of a scan. If the 995 * queue exists here, that implies the vdev is being removed while 996 * the scan is still running. 997 */ 998 if (vd->vdev_scan_io_queue != NULL) { 999 mutex_enter(&vd->vdev_scan_io_queue_lock); 1000 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 1001 vd->vdev_scan_io_queue = NULL; 1002 mutex_exit(&vd->vdev_scan_io_queue_lock); 1003 } 1004 1005 /* 1006 * vdev_free() implies closing the vdev first. This is simpler than 1007 * trying to ensure complicated semantics for all callers. 1008 */ 1009 vdev_close(vd); 1010 1011 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 1012 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 1013 1014 /* 1015 * Free all children. 1016 */ 1017 for (int c = 0; c < vd->vdev_children; c++) 1018 vdev_free(vd->vdev_child[c]); 1019 1020 ASSERT(vd->vdev_child == NULL); 1021 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 1022 1023 if (vd->vdev_ops->vdev_op_fini != NULL) 1024 vd->vdev_ops->vdev_op_fini(vd); 1025 1026 /* 1027 * Discard allocation state. 1028 */ 1029 if (vd->vdev_mg != NULL) { 1030 vdev_metaslab_fini(vd); 1031 metaslab_group_destroy(vd->vdev_mg); 1032 vd->vdev_mg = NULL; 1033 } 1034 if (vd->vdev_log_mg != NULL) { 1035 ASSERT0(vd->vdev_ms_count); 1036 metaslab_group_destroy(vd->vdev_log_mg); 1037 vd->vdev_log_mg = NULL; 1038 } 1039 1040 ASSERT0(vd->vdev_stat.vs_space); 1041 ASSERT0(vd->vdev_stat.vs_dspace); 1042 ASSERT0(vd->vdev_stat.vs_alloc); 1043 1044 /* 1045 * Remove this vdev from its parent's child list. 1046 */ 1047 vdev_remove_child(vd->vdev_parent, vd); 1048 1049 ASSERT(vd->vdev_parent == NULL); 1050 ASSERT(!list_link_active(&vd->vdev_leaf_node)); 1051 1052 /* 1053 * Clean up vdev structure. 1054 */ 1055 vdev_queue_fini(vd); 1056 vdev_cache_fini(vd); 1057 1058 if (vd->vdev_path) 1059 spa_strfree(vd->vdev_path); 1060 if (vd->vdev_devid) 1061 spa_strfree(vd->vdev_devid); 1062 if (vd->vdev_physpath) 1063 spa_strfree(vd->vdev_physpath); 1064 1065 if (vd->vdev_enc_sysfs_path) 1066 spa_strfree(vd->vdev_enc_sysfs_path); 1067 1068 if (vd->vdev_fru) 1069 spa_strfree(vd->vdev_fru); 1070 1071 if (vd->vdev_isspare) 1072 spa_spare_remove(vd); 1073 if (vd->vdev_isl2cache) 1074 spa_l2cache_remove(vd); 1075 1076 txg_list_destroy(&vd->vdev_ms_list); 1077 txg_list_destroy(&vd->vdev_dtl_list); 1078 1079 mutex_enter(&vd->vdev_dtl_lock); 1080 space_map_close(vd->vdev_dtl_sm); 1081 for (int t = 0; t < DTL_TYPES; t++) { 1082 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 1083 range_tree_destroy(vd->vdev_dtl[t]); 1084 } 1085 mutex_exit(&vd->vdev_dtl_lock); 1086 1087 EQUIV(vd->vdev_indirect_births != NULL, 1088 vd->vdev_indirect_mapping != NULL); 1089 if (vd->vdev_indirect_births != NULL) { 1090 vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 1091 vdev_indirect_births_close(vd->vdev_indirect_births); 1092 } 1093 1094 if (vd->vdev_obsolete_sm != NULL) { 1095 ASSERT(vd->vdev_removing || 1096 vd->vdev_ops == &vdev_indirect_ops); 1097 space_map_close(vd->vdev_obsolete_sm); 1098 vd->vdev_obsolete_sm = NULL; 1099 } 1100 range_tree_destroy(vd->vdev_obsolete_segments); 1101 rw_destroy(&vd->vdev_indirect_rwlock); 1102 mutex_destroy(&vd->vdev_obsolete_lock); 1103 1104 mutex_destroy(&vd->vdev_dtl_lock); 1105 mutex_destroy(&vd->vdev_stat_lock); 1106 mutex_destroy(&vd->vdev_probe_lock); 1107 mutex_destroy(&vd->vdev_scan_io_queue_lock); 1108 1109 mutex_destroy(&vd->vdev_initialize_lock); 1110 mutex_destroy(&vd->vdev_initialize_io_lock); 1111 cv_destroy(&vd->vdev_initialize_io_cv); 1112 cv_destroy(&vd->vdev_initialize_cv); 1113 1114 mutex_destroy(&vd->vdev_trim_lock); 1115 mutex_destroy(&vd->vdev_autotrim_lock); 1116 mutex_destroy(&vd->vdev_trim_io_lock); 1117 cv_destroy(&vd->vdev_trim_cv); 1118 cv_destroy(&vd->vdev_autotrim_cv); 1119 cv_destroy(&vd->vdev_trim_io_cv); 1120 1121 mutex_destroy(&vd->vdev_rebuild_lock); 1122 cv_destroy(&vd->vdev_rebuild_cv); 1123 1124 zfs_ratelimit_fini(&vd->vdev_delay_rl); 1125 zfs_ratelimit_fini(&vd->vdev_deadman_rl); 1126 zfs_ratelimit_fini(&vd->vdev_checksum_rl); 1127 1128 if (vd == spa->spa_root_vdev) 1129 spa->spa_root_vdev = NULL; 1130 1131 kmem_free(vd, sizeof (vdev_t)); 1132 } 1133 1134 /* 1135 * Transfer top-level vdev state from svd to tvd. 1136 */ 1137 static void 1138 vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 1139 { 1140 spa_t *spa = svd->vdev_spa; 1141 metaslab_t *msp; 1142 vdev_t *vd; 1143 int t; 1144 1145 ASSERT(tvd == tvd->vdev_top); 1146 1147 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; 1148 tvd->vdev_ms_array = svd->vdev_ms_array; 1149 tvd->vdev_ms_shift = svd->vdev_ms_shift; 1150 tvd->vdev_ms_count = svd->vdev_ms_count; 1151 tvd->vdev_top_zap = svd->vdev_top_zap; 1152 1153 svd->vdev_ms_array = 0; 1154 svd->vdev_ms_shift = 0; 1155 svd->vdev_ms_count = 0; 1156 svd->vdev_top_zap = 0; 1157 1158 if (tvd->vdev_mg) 1159 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 1160 if (tvd->vdev_log_mg) 1161 ASSERT3P(tvd->vdev_log_mg, ==, svd->vdev_log_mg); 1162 tvd->vdev_mg = svd->vdev_mg; 1163 tvd->vdev_log_mg = svd->vdev_log_mg; 1164 tvd->vdev_ms = svd->vdev_ms; 1165 1166 svd->vdev_mg = NULL; 1167 svd->vdev_log_mg = NULL; 1168 svd->vdev_ms = NULL; 1169 1170 if (tvd->vdev_mg != NULL) 1171 tvd->vdev_mg->mg_vd = tvd; 1172 if (tvd->vdev_log_mg != NULL) 1173 tvd->vdev_log_mg->mg_vd = tvd; 1174 1175 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 1176 svd->vdev_checkpoint_sm = NULL; 1177 1178 tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 1179 svd->vdev_alloc_bias = VDEV_BIAS_NONE; 1180 1181 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 1182 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 1183 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 1184 1185 svd->vdev_stat.vs_alloc = 0; 1186 svd->vdev_stat.vs_space = 0; 1187 svd->vdev_stat.vs_dspace = 0; 1188 1189 /* 1190 * State which may be set on a top-level vdev that's in the 1191 * process of being removed. 1192 */ 1193 ASSERT0(tvd->vdev_indirect_config.vic_births_object); 1194 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 1195 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 1196 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 1197 ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 1198 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 1199 ASSERT0(tvd->vdev_noalloc); 1200 ASSERT0(tvd->vdev_removing); 1201 ASSERT0(tvd->vdev_rebuilding); 1202 tvd->vdev_noalloc = svd->vdev_noalloc; 1203 tvd->vdev_removing = svd->vdev_removing; 1204 tvd->vdev_rebuilding = svd->vdev_rebuilding; 1205 tvd->vdev_rebuild_config = svd->vdev_rebuild_config; 1206 tvd->vdev_indirect_config = svd->vdev_indirect_config; 1207 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 1208 tvd->vdev_indirect_births = svd->vdev_indirect_births; 1209 range_tree_swap(&svd->vdev_obsolete_segments, 1210 &tvd->vdev_obsolete_segments); 1211 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 1212 svd->vdev_indirect_config.vic_mapping_object = 0; 1213 svd->vdev_indirect_config.vic_births_object = 0; 1214 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 1215 svd->vdev_indirect_mapping = NULL; 1216 svd->vdev_indirect_births = NULL; 1217 svd->vdev_obsolete_sm = NULL; 1218 svd->vdev_noalloc = 0; 1219 svd->vdev_removing = 0; 1220 svd->vdev_rebuilding = 0; 1221 1222 for (t = 0; t < TXG_SIZE; t++) { 1223 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1224 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1225 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1226 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1227 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1228 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1229 } 1230 1231 if (list_link_active(&svd->vdev_config_dirty_node)) { 1232 vdev_config_clean(svd); 1233 vdev_config_dirty(tvd); 1234 } 1235 1236 if (list_link_active(&svd->vdev_state_dirty_node)) { 1237 vdev_state_clean(svd); 1238 vdev_state_dirty(tvd); 1239 } 1240 1241 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 1242 svd->vdev_deflate_ratio = 0; 1243 1244 tvd->vdev_islog = svd->vdev_islog; 1245 svd->vdev_islog = 0; 1246 1247 dsl_scan_io_queue_vdev_xfer(svd, tvd); 1248 } 1249 1250 static void 1251 vdev_top_update(vdev_t *tvd, vdev_t *vd) 1252 { 1253 if (vd == NULL) 1254 return; 1255 1256 vd->vdev_top = tvd; 1257 1258 for (int c = 0; c < vd->vdev_children; c++) 1259 vdev_top_update(tvd, vd->vdev_child[c]); 1260 } 1261 1262 /* 1263 * Add a mirror/replacing vdev above an existing vdev. There is no need to 1264 * call .vdev_op_init() since mirror/replacing vdevs do not have private state. 1265 */ 1266 vdev_t * 1267 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1268 { 1269 spa_t *spa = cvd->vdev_spa; 1270 vdev_t *pvd = cvd->vdev_parent; 1271 vdev_t *mvd; 1272 1273 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1274 1275 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1276 1277 mvd->vdev_asize = cvd->vdev_asize; 1278 mvd->vdev_min_asize = cvd->vdev_min_asize; 1279 mvd->vdev_max_asize = cvd->vdev_max_asize; 1280 mvd->vdev_psize = cvd->vdev_psize; 1281 mvd->vdev_ashift = cvd->vdev_ashift; 1282 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift; 1283 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift; 1284 mvd->vdev_state = cvd->vdev_state; 1285 mvd->vdev_crtxg = cvd->vdev_crtxg; 1286 1287 vdev_remove_child(pvd, cvd); 1288 vdev_add_child(pvd, mvd); 1289 cvd->vdev_id = mvd->vdev_children; 1290 vdev_add_child(mvd, cvd); 1291 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1292 1293 if (mvd == mvd->vdev_top) 1294 vdev_top_transfer(cvd, mvd); 1295 1296 return (mvd); 1297 } 1298 1299 /* 1300 * Remove a 1-way mirror/replacing vdev from the tree. 1301 */ 1302 void 1303 vdev_remove_parent(vdev_t *cvd) 1304 { 1305 vdev_t *mvd = cvd->vdev_parent; 1306 vdev_t *pvd = mvd->vdev_parent; 1307 1308 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1309 1310 ASSERT(mvd->vdev_children == 1); 1311 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 1312 mvd->vdev_ops == &vdev_replacing_ops || 1313 mvd->vdev_ops == &vdev_spare_ops); 1314 cvd->vdev_ashift = mvd->vdev_ashift; 1315 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift; 1316 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift; 1317 vdev_remove_child(mvd, cvd); 1318 vdev_remove_child(pvd, mvd); 1319 1320 /* 1321 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1322 * Otherwise, we could have detached an offline device, and when we 1323 * go to import the pool we'll think we have two top-level vdevs, 1324 * instead of a different version of the same top-level vdev. 1325 */ 1326 if (mvd->vdev_top == mvd) { 1327 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 1328 cvd->vdev_orig_guid = cvd->vdev_guid; 1329 cvd->vdev_guid += guid_delta; 1330 cvd->vdev_guid_sum += guid_delta; 1331 1332 /* 1333 * If pool not set for autoexpand, we need to also preserve 1334 * mvd's asize to prevent automatic expansion of cvd. 1335 * Otherwise if we are adjusting the mirror by attaching and 1336 * detaching children of non-uniform sizes, the mirror could 1337 * autoexpand, unexpectedly requiring larger devices to 1338 * re-establish the mirror. 1339 */ 1340 if (!cvd->vdev_spa->spa_autoexpand) 1341 cvd->vdev_asize = mvd->vdev_asize; 1342 } 1343 cvd->vdev_id = mvd->vdev_id; 1344 vdev_add_child(pvd, cvd); 1345 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1346 1347 if (cvd == cvd->vdev_top) 1348 vdev_top_transfer(mvd, cvd); 1349 1350 ASSERT(mvd->vdev_children == 0); 1351 vdev_free(mvd); 1352 } 1353 1354 void 1355 vdev_metaslab_group_create(vdev_t *vd) 1356 { 1357 spa_t *spa = vd->vdev_spa; 1358 1359 /* 1360 * metaslab_group_create was delayed until allocation bias was available 1361 */ 1362 if (vd->vdev_mg == NULL) { 1363 metaslab_class_t *mc; 1364 1365 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1366 vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1367 1368 ASSERT3U(vd->vdev_islog, ==, 1369 (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1370 1371 switch (vd->vdev_alloc_bias) { 1372 case VDEV_BIAS_LOG: 1373 mc = spa_log_class(spa); 1374 break; 1375 case VDEV_BIAS_SPECIAL: 1376 mc = spa_special_class(spa); 1377 break; 1378 case VDEV_BIAS_DEDUP: 1379 mc = spa_dedup_class(spa); 1380 break; 1381 default: 1382 mc = spa_normal_class(spa); 1383 } 1384 1385 vd->vdev_mg = metaslab_group_create(mc, vd, 1386 spa->spa_alloc_count); 1387 1388 if (!vd->vdev_islog) { 1389 vd->vdev_log_mg = metaslab_group_create( 1390 spa_embedded_log_class(spa), vd, 1); 1391 } 1392 1393 /* 1394 * The spa ashift min/max only apply for the normal metaslab 1395 * class. Class destination is late binding so ashift boundary 1396 * setting had to wait until now. 1397 */ 1398 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1399 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1400 if (vd->vdev_ashift > spa->spa_max_ashift) 1401 spa->spa_max_ashift = vd->vdev_ashift; 1402 if (vd->vdev_ashift < spa->spa_min_ashift) 1403 spa->spa_min_ashift = vd->vdev_ashift; 1404 1405 uint64_t min_alloc = vdev_get_min_alloc(vd); 1406 if (min_alloc < spa->spa_min_alloc) 1407 spa->spa_min_alloc = min_alloc; 1408 } 1409 } 1410 } 1411 1412 int 1413 vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1414 { 1415 spa_t *spa = vd->vdev_spa; 1416 uint64_t oldc = vd->vdev_ms_count; 1417 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1418 metaslab_t **mspp; 1419 int error; 1420 boolean_t expanding = (oldc != 0); 1421 1422 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1423 1424 /* 1425 * This vdev is not being allocated from yet or is a hole. 1426 */ 1427 if (vd->vdev_ms_shift == 0) 1428 return (0); 1429 1430 ASSERT(!vd->vdev_ishole); 1431 1432 ASSERT(oldc <= newc); 1433 1434 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1435 1436 if (expanding) { 1437 memcpy(mspp, vd->vdev_ms, oldc * sizeof (*mspp)); 1438 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1439 } 1440 1441 vd->vdev_ms = mspp; 1442 vd->vdev_ms_count = newc; 1443 1444 for (uint64_t m = oldc; m < newc; m++) { 1445 uint64_t object = 0; 1446 /* 1447 * vdev_ms_array may be 0 if we are creating the "fake" 1448 * metaslabs for an indirect vdev for zdb's leak detection. 1449 * See zdb_leak_init(). 1450 */ 1451 if (txg == 0 && vd->vdev_ms_array != 0) { 1452 error = dmu_read(spa->spa_meta_objset, 1453 vd->vdev_ms_array, 1454 m * sizeof (uint64_t), sizeof (uint64_t), &object, 1455 DMU_READ_PREFETCH); 1456 if (error != 0) { 1457 vdev_dbgmsg(vd, "unable to read the metaslab " 1458 "array [error=%d]", error); 1459 return (error); 1460 } 1461 } 1462 1463 error = metaslab_init(vd->vdev_mg, m, object, txg, 1464 &(vd->vdev_ms[m])); 1465 if (error != 0) { 1466 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 1467 error); 1468 return (error); 1469 } 1470 } 1471 1472 /* 1473 * Find the emptiest metaslab on the vdev and mark it for use for 1474 * embedded slog by moving it from the regular to the log metaslab 1475 * group. 1476 */ 1477 if (vd->vdev_mg->mg_class == spa_normal_class(spa) && 1478 vd->vdev_ms_count > zfs_embedded_slog_min_ms && 1479 avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) { 1480 uint64_t slog_msid = 0; 1481 uint64_t smallest = UINT64_MAX; 1482 1483 /* 1484 * Note, we only search the new metaslabs, because the old 1485 * (pre-existing) ones may be active (e.g. have non-empty 1486 * range_tree's), and we don't move them to the new 1487 * metaslab_t. 1488 */ 1489 for (uint64_t m = oldc; m < newc; m++) { 1490 uint64_t alloc = 1491 space_map_allocated(vd->vdev_ms[m]->ms_sm); 1492 if (alloc < smallest) { 1493 slog_msid = m; 1494 smallest = alloc; 1495 } 1496 } 1497 metaslab_t *slog_ms = vd->vdev_ms[slog_msid]; 1498 /* 1499 * The metaslab was marked as dirty at the end of 1500 * metaslab_init(). Remove it from the dirty list so that we 1501 * can uninitialize and reinitialize it to the new class. 1502 */ 1503 if (txg != 0) { 1504 (void) txg_list_remove_this(&vd->vdev_ms_list, 1505 slog_ms, txg); 1506 } 1507 uint64_t sm_obj = space_map_object(slog_ms->ms_sm); 1508 metaslab_fini(slog_ms); 1509 VERIFY0(metaslab_init(vd->vdev_log_mg, slog_msid, sm_obj, txg, 1510 &vd->vdev_ms[slog_msid])); 1511 } 1512 1513 if (txg == 0) 1514 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1515 1516 /* 1517 * If the vdev is marked as non-allocating then don't 1518 * activate the metaslabs since we want to ensure that 1519 * no allocations are performed on this device. 1520 */ 1521 if (vd->vdev_noalloc) { 1522 /* track non-allocating vdev space */ 1523 spa->spa_nonallocating_dspace += spa_deflate(spa) ? 1524 vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space; 1525 } else if (!expanding) { 1526 metaslab_group_activate(vd->vdev_mg); 1527 if (vd->vdev_log_mg != NULL) 1528 metaslab_group_activate(vd->vdev_log_mg); 1529 } 1530 1531 if (txg == 0) 1532 spa_config_exit(spa, SCL_ALLOC, FTAG); 1533 1534 return (0); 1535 } 1536 1537 void 1538 vdev_metaslab_fini(vdev_t *vd) 1539 { 1540 if (vd->vdev_checkpoint_sm != NULL) { 1541 ASSERT(spa_feature_is_active(vd->vdev_spa, 1542 SPA_FEATURE_POOL_CHECKPOINT)); 1543 space_map_close(vd->vdev_checkpoint_sm); 1544 /* 1545 * Even though we close the space map, we need to set its 1546 * pointer to NULL. The reason is that vdev_metaslab_fini() 1547 * may be called multiple times for certain operations 1548 * (i.e. when destroying a pool) so we need to ensure that 1549 * this clause never executes twice. This logic is similar 1550 * to the one used for the vdev_ms clause below. 1551 */ 1552 vd->vdev_checkpoint_sm = NULL; 1553 } 1554 1555 if (vd->vdev_ms != NULL) { 1556 metaslab_group_t *mg = vd->vdev_mg; 1557 1558 metaslab_group_passivate(mg); 1559 if (vd->vdev_log_mg != NULL) { 1560 ASSERT(!vd->vdev_islog); 1561 metaslab_group_passivate(vd->vdev_log_mg); 1562 } 1563 1564 uint64_t count = vd->vdev_ms_count; 1565 for (uint64_t m = 0; m < count; m++) { 1566 metaslab_t *msp = vd->vdev_ms[m]; 1567 if (msp != NULL) 1568 metaslab_fini(msp); 1569 } 1570 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1571 vd->vdev_ms = NULL; 1572 vd->vdev_ms_count = 0; 1573 1574 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 1575 ASSERT0(mg->mg_histogram[i]); 1576 if (vd->vdev_log_mg != NULL) 1577 ASSERT0(vd->vdev_log_mg->mg_histogram[i]); 1578 } 1579 } 1580 ASSERT0(vd->vdev_ms_count); 1581 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); 1582 } 1583 1584 typedef struct vdev_probe_stats { 1585 boolean_t vps_readable; 1586 boolean_t vps_writeable; 1587 int vps_flags; 1588 } vdev_probe_stats_t; 1589 1590 static void 1591 vdev_probe_done(zio_t *zio) 1592 { 1593 spa_t *spa = zio->io_spa; 1594 vdev_t *vd = zio->io_vd; 1595 vdev_probe_stats_t *vps = zio->io_private; 1596 1597 ASSERT(vd->vdev_probe_zio != NULL); 1598 1599 if (zio->io_type == ZIO_TYPE_READ) { 1600 if (zio->io_error == 0) 1601 vps->vps_readable = 1; 1602 if (zio->io_error == 0 && spa_writeable(spa)) { 1603 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1604 zio->io_offset, zio->io_size, zio->io_abd, 1605 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1606 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1607 } else { 1608 abd_free(zio->io_abd); 1609 } 1610 } else if (zio->io_type == ZIO_TYPE_WRITE) { 1611 if (zio->io_error == 0) 1612 vps->vps_writeable = 1; 1613 abd_free(zio->io_abd); 1614 } else if (zio->io_type == ZIO_TYPE_NULL) { 1615 zio_t *pio; 1616 zio_link_t *zl; 1617 1618 vd->vdev_cant_read |= !vps->vps_readable; 1619 vd->vdev_cant_write |= !vps->vps_writeable; 1620 1621 if (vdev_readable(vd) && 1622 (vdev_writeable(vd) || !spa_writeable(spa))) { 1623 zio->io_error = 0; 1624 } else { 1625 ASSERT(zio->io_error != 0); 1626 vdev_dbgmsg(vd, "failed probe"); 1627 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1628 spa, vd, NULL, NULL, 0); 1629 zio->io_error = SET_ERROR(ENXIO); 1630 } 1631 1632 mutex_enter(&vd->vdev_probe_lock); 1633 ASSERT(vd->vdev_probe_zio == zio); 1634 vd->vdev_probe_zio = NULL; 1635 mutex_exit(&vd->vdev_probe_lock); 1636 1637 zl = NULL; 1638 while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1639 if (!vdev_accessible(vd, pio)) 1640 pio->io_error = SET_ERROR(ENXIO); 1641 1642 kmem_free(vps, sizeof (*vps)); 1643 } 1644 } 1645 1646 /* 1647 * Determine whether this device is accessible. 1648 * 1649 * Read and write to several known locations: the pad regions of each 1650 * vdev label but the first, which we leave alone in case it contains 1651 * a VTOC. 1652 */ 1653 zio_t * 1654 vdev_probe(vdev_t *vd, zio_t *zio) 1655 { 1656 spa_t *spa = vd->vdev_spa; 1657 vdev_probe_stats_t *vps = NULL; 1658 zio_t *pio; 1659 1660 ASSERT(vd->vdev_ops->vdev_op_leaf); 1661 1662 /* 1663 * Don't probe the probe. 1664 */ 1665 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1666 return (NULL); 1667 1668 /* 1669 * To prevent 'probe storms' when a device fails, we create 1670 * just one probe i/o at a time. All zios that want to probe 1671 * this vdev will become parents of the probe io. 1672 */ 1673 mutex_enter(&vd->vdev_probe_lock); 1674 1675 if ((pio = vd->vdev_probe_zio) == NULL) { 1676 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1677 1678 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1679 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1680 ZIO_FLAG_TRYHARD; 1681 1682 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1683 /* 1684 * vdev_cant_read and vdev_cant_write can only 1685 * transition from TRUE to FALSE when we have the 1686 * SCL_ZIO lock as writer; otherwise they can only 1687 * transition from FALSE to TRUE. This ensures that 1688 * any zio looking at these values can assume that 1689 * failures persist for the life of the I/O. That's 1690 * important because when a device has intermittent 1691 * connectivity problems, we want to ensure that 1692 * they're ascribed to the device (ENXIO) and not 1693 * the zio (EIO). 1694 * 1695 * Since we hold SCL_ZIO as writer here, clear both 1696 * values so the probe can reevaluate from first 1697 * principles. 1698 */ 1699 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1700 vd->vdev_cant_read = B_FALSE; 1701 vd->vdev_cant_write = B_FALSE; 1702 } 1703 1704 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1705 vdev_probe_done, vps, 1706 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1707 1708 /* 1709 * We can't change the vdev state in this context, so we 1710 * kick off an async task to do it on our behalf. 1711 */ 1712 if (zio != NULL) { 1713 vd->vdev_probe_wanted = B_TRUE; 1714 spa_async_request(spa, SPA_ASYNC_PROBE); 1715 } 1716 } 1717 1718 if (zio != NULL) 1719 zio_add_child(zio, pio); 1720 1721 mutex_exit(&vd->vdev_probe_lock); 1722 1723 if (vps == NULL) { 1724 ASSERT(zio != NULL); 1725 return (NULL); 1726 } 1727 1728 for (int l = 1; l < VDEV_LABELS; l++) { 1729 zio_nowait(zio_read_phys(pio, vd, 1730 vdev_label_offset(vd->vdev_psize, l, 1731 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE, 1732 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1733 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1734 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1735 } 1736 1737 if (zio == NULL) 1738 return (pio); 1739 1740 zio_nowait(pio); 1741 return (NULL); 1742 } 1743 1744 static void 1745 vdev_load_child(void *arg) 1746 { 1747 vdev_t *vd = arg; 1748 1749 vd->vdev_load_error = vdev_load(vd); 1750 } 1751 1752 static void 1753 vdev_open_child(void *arg) 1754 { 1755 vdev_t *vd = arg; 1756 1757 vd->vdev_open_thread = curthread; 1758 vd->vdev_open_error = vdev_open(vd); 1759 vd->vdev_open_thread = NULL; 1760 } 1761 1762 static boolean_t 1763 vdev_uses_zvols(vdev_t *vd) 1764 { 1765 #ifdef _KERNEL 1766 if (zvol_is_zvol(vd->vdev_path)) 1767 return (B_TRUE); 1768 #endif 1769 1770 for (int c = 0; c < vd->vdev_children; c++) 1771 if (vdev_uses_zvols(vd->vdev_child[c])) 1772 return (B_TRUE); 1773 1774 return (B_FALSE); 1775 } 1776 1777 /* 1778 * Returns B_TRUE if the passed child should be opened. 1779 */ 1780 static boolean_t 1781 vdev_default_open_children_func(vdev_t *vd) 1782 { 1783 (void) vd; 1784 return (B_TRUE); 1785 } 1786 1787 /* 1788 * Open the requested child vdevs. If any of the leaf vdevs are using 1789 * a ZFS volume then do the opens in a single thread. This avoids a 1790 * deadlock when the current thread is holding the spa_namespace_lock. 1791 */ 1792 static void 1793 vdev_open_children_impl(vdev_t *vd, vdev_open_children_func_t *open_func) 1794 { 1795 int children = vd->vdev_children; 1796 1797 taskq_t *tq = taskq_create("vdev_open", children, minclsyspri, 1798 children, children, TASKQ_PREPOPULATE); 1799 vd->vdev_nonrot = B_TRUE; 1800 1801 for (int c = 0; c < children; c++) { 1802 vdev_t *cvd = vd->vdev_child[c]; 1803 1804 if (open_func(cvd) == B_FALSE) 1805 continue; 1806 1807 if (tq == NULL || vdev_uses_zvols(vd)) { 1808 cvd->vdev_open_error = vdev_open(cvd); 1809 } else { 1810 VERIFY(taskq_dispatch(tq, vdev_open_child, 1811 cvd, TQ_SLEEP) != TASKQID_INVALID); 1812 } 1813 1814 vd->vdev_nonrot &= cvd->vdev_nonrot; 1815 } 1816 1817 if (tq != NULL) { 1818 taskq_wait(tq); 1819 taskq_destroy(tq); 1820 } 1821 } 1822 1823 /* 1824 * Open all child vdevs. 1825 */ 1826 void 1827 vdev_open_children(vdev_t *vd) 1828 { 1829 vdev_open_children_impl(vd, vdev_default_open_children_func); 1830 } 1831 1832 /* 1833 * Conditionally open a subset of child vdevs. 1834 */ 1835 void 1836 vdev_open_children_subset(vdev_t *vd, vdev_open_children_func_t *open_func) 1837 { 1838 vdev_open_children_impl(vd, open_func); 1839 } 1840 1841 /* 1842 * Compute the raidz-deflation ratio. Note, we hard-code 1843 * in 128k (1 << 17) because it is the "typical" blocksize. 1844 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 1845 * otherwise it would inconsistently account for existing bp's. 1846 */ 1847 static void 1848 vdev_set_deflate_ratio(vdev_t *vd) 1849 { 1850 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 1851 vd->vdev_deflate_ratio = (1 << 17) / 1852 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 1853 } 1854 } 1855 1856 /* 1857 * Choose the best of two ashifts, preferring one between logical ashift 1858 * (absolute minimum) and administrator defined maximum, otherwise take 1859 * the biggest of the two. 1860 */ 1861 uint64_t 1862 vdev_best_ashift(uint64_t logical, uint64_t a, uint64_t b) 1863 { 1864 if (a > logical && a <= zfs_vdev_max_auto_ashift) { 1865 if (b <= logical || b > zfs_vdev_max_auto_ashift) 1866 return (a); 1867 else 1868 return (MAX(a, b)); 1869 } else if (b <= logical || b > zfs_vdev_max_auto_ashift) 1870 return (MAX(a, b)); 1871 return (b); 1872 } 1873 1874 /* 1875 * Maximize performance by inflating the configured ashift for top level 1876 * vdevs to be as close to the physical ashift as possible while maintaining 1877 * administrator defined limits and ensuring it doesn't go below the 1878 * logical ashift. 1879 */ 1880 static void 1881 vdev_ashift_optimize(vdev_t *vd) 1882 { 1883 ASSERT(vd == vd->vdev_top); 1884 1885 if (vd->vdev_ashift < vd->vdev_physical_ashift && 1886 vd->vdev_physical_ashift <= zfs_vdev_max_auto_ashift) { 1887 vd->vdev_ashift = MIN( 1888 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift), 1889 MAX(zfs_vdev_min_auto_ashift, 1890 vd->vdev_physical_ashift)); 1891 } else { 1892 /* 1893 * If the logical and physical ashifts are the same, then 1894 * we ensure that the top-level vdev's ashift is not smaller 1895 * than our minimum ashift value. For the unusual case 1896 * where logical ashift > physical ashift, we can't cap 1897 * the calculated ashift based on max ashift as that 1898 * would cause failures. 1899 * We still check if we need to increase it to match 1900 * the min ashift. 1901 */ 1902 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift, 1903 vd->vdev_ashift); 1904 } 1905 } 1906 1907 /* 1908 * Prepare a virtual device for access. 1909 */ 1910 int 1911 vdev_open(vdev_t *vd) 1912 { 1913 spa_t *spa = vd->vdev_spa; 1914 int error; 1915 uint64_t osize = 0; 1916 uint64_t max_osize = 0; 1917 uint64_t asize, max_asize, psize; 1918 uint64_t logical_ashift = 0; 1919 uint64_t physical_ashift = 0; 1920 1921 ASSERT(vd->vdev_open_thread == curthread || 1922 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1923 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1924 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1925 vd->vdev_state == VDEV_STATE_OFFLINE); 1926 1927 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1928 vd->vdev_cant_read = B_FALSE; 1929 vd->vdev_cant_write = B_FALSE; 1930 vd->vdev_min_asize = vdev_get_min_asize(vd); 1931 1932 /* 1933 * If this vdev is not removed, check its fault status. If it's 1934 * faulted, bail out of the open. 1935 */ 1936 if (!vd->vdev_removed && vd->vdev_faulted) { 1937 ASSERT(vd->vdev_children == 0); 1938 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1939 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1940 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1941 vd->vdev_label_aux); 1942 return (SET_ERROR(ENXIO)); 1943 } else if (vd->vdev_offline) { 1944 ASSERT(vd->vdev_children == 0); 1945 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1946 return (SET_ERROR(ENXIO)); 1947 } 1948 1949 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, 1950 &logical_ashift, &physical_ashift); 1951 1952 /* Keep the device in removed state if unplugged */ 1953 if (error == ENOENT && vd->vdev_removed) { 1954 vdev_set_state(vd, B_TRUE, VDEV_STATE_REMOVED, 1955 VDEV_AUX_NONE); 1956 return (error); 1957 } 1958 1959 /* 1960 * Physical volume size should never be larger than its max size, unless 1961 * the disk has shrunk while we were reading it or the device is buggy 1962 * or damaged: either way it's not safe for use, bail out of the open. 1963 */ 1964 if (osize > max_osize) { 1965 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1966 VDEV_AUX_OPEN_FAILED); 1967 return (SET_ERROR(ENXIO)); 1968 } 1969 1970 /* 1971 * Reset the vdev_reopening flag so that we actually close 1972 * the vdev on error. 1973 */ 1974 vd->vdev_reopening = B_FALSE; 1975 if (zio_injection_enabled && error == 0) 1976 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO)); 1977 1978 if (error) { 1979 if (vd->vdev_removed && 1980 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1981 vd->vdev_removed = B_FALSE; 1982 1983 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1984 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1985 vd->vdev_stat.vs_aux); 1986 } else { 1987 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1988 vd->vdev_stat.vs_aux); 1989 } 1990 return (error); 1991 } 1992 1993 vd->vdev_removed = B_FALSE; 1994 1995 /* 1996 * Recheck the faulted flag now that we have confirmed that 1997 * the vdev is accessible. If we're faulted, bail. 1998 */ 1999 if (vd->vdev_faulted) { 2000 ASSERT(vd->vdev_children == 0); 2001 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 2002 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 2003 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 2004 vd->vdev_label_aux); 2005 return (SET_ERROR(ENXIO)); 2006 } 2007 2008 if (vd->vdev_degraded) { 2009 ASSERT(vd->vdev_children == 0); 2010 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 2011 VDEV_AUX_ERR_EXCEEDED); 2012 } else { 2013 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 2014 } 2015 2016 /* 2017 * For hole or missing vdevs we just return success. 2018 */ 2019 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 2020 return (0); 2021 2022 for (int c = 0; c < vd->vdev_children; c++) { 2023 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 2024 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 2025 VDEV_AUX_NONE); 2026 break; 2027 } 2028 } 2029 2030 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 2031 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 2032 2033 if (vd->vdev_children == 0) { 2034 if (osize < SPA_MINDEVSIZE) { 2035 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2036 VDEV_AUX_TOO_SMALL); 2037 return (SET_ERROR(EOVERFLOW)); 2038 } 2039 psize = osize; 2040 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 2041 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 2042 VDEV_LABEL_END_SIZE); 2043 } else { 2044 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 2045 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 2046 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2047 VDEV_AUX_TOO_SMALL); 2048 return (SET_ERROR(EOVERFLOW)); 2049 } 2050 psize = 0; 2051 asize = osize; 2052 max_asize = max_osize; 2053 } 2054 2055 /* 2056 * If the vdev was expanded, record this so that we can re-create the 2057 * uberblock rings in labels {2,3}, during the next sync. 2058 */ 2059 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0)) 2060 vd->vdev_copy_uberblocks = B_TRUE; 2061 2062 vd->vdev_psize = psize; 2063 2064 /* 2065 * Make sure the allocatable size hasn't shrunk too much. 2066 */ 2067 if (asize < vd->vdev_min_asize) { 2068 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2069 VDEV_AUX_BAD_LABEL); 2070 return (SET_ERROR(EINVAL)); 2071 } 2072 2073 /* 2074 * We can always set the logical/physical ashift members since 2075 * their values are only used to calculate the vdev_ashift when 2076 * the device is first added to the config. These values should 2077 * not be used for anything else since they may change whenever 2078 * the device is reopened and we don't store them in the label. 2079 */ 2080 vd->vdev_physical_ashift = 2081 MAX(physical_ashift, vd->vdev_physical_ashift); 2082 vd->vdev_logical_ashift = MAX(logical_ashift, 2083 vd->vdev_logical_ashift); 2084 2085 if (vd->vdev_asize == 0) { 2086 /* 2087 * This is the first-ever open, so use the computed values. 2088 * For compatibility, a different ashift can be requested. 2089 */ 2090 vd->vdev_asize = asize; 2091 vd->vdev_max_asize = max_asize; 2092 2093 /* 2094 * If the vdev_ashift was not overridden at creation time, 2095 * then set it the logical ashift and optimize the ashift. 2096 */ 2097 if (vd->vdev_ashift == 0) { 2098 vd->vdev_ashift = vd->vdev_logical_ashift; 2099 2100 if (vd->vdev_logical_ashift > ASHIFT_MAX) { 2101 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2102 VDEV_AUX_ASHIFT_TOO_BIG); 2103 return (SET_ERROR(EDOM)); 2104 } 2105 2106 if (vd->vdev_top == vd) { 2107 vdev_ashift_optimize(vd); 2108 } 2109 } 2110 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 2111 vd->vdev_ashift > ASHIFT_MAX)) { 2112 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2113 VDEV_AUX_BAD_ASHIFT); 2114 return (SET_ERROR(EDOM)); 2115 } 2116 } else { 2117 /* 2118 * Make sure the alignment required hasn't increased. 2119 */ 2120 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift && 2121 vd->vdev_ops->vdev_op_leaf) { 2122 (void) zfs_ereport_post( 2123 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 2124 spa, vd, NULL, NULL, 0); 2125 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2126 VDEV_AUX_BAD_LABEL); 2127 return (SET_ERROR(EDOM)); 2128 } 2129 vd->vdev_max_asize = max_asize; 2130 } 2131 2132 /* 2133 * If all children are healthy we update asize if either: 2134 * The asize has increased, due to a device expansion caused by dynamic 2135 * LUN growth or vdev replacement, and automatic expansion is enabled; 2136 * making the additional space available. 2137 * 2138 * The asize has decreased, due to a device shrink usually caused by a 2139 * vdev replace with a smaller device. This ensures that calculations 2140 * based of max_asize and asize e.g. esize are always valid. It's safe 2141 * to do this as we've already validated that asize is greater than 2142 * vdev_min_asize. 2143 */ 2144 if (vd->vdev_state == VDEV_STATE_HEALTHY && 2145 ((asize > vd->vdev_asize && 2146 (vd->vdev_expanding || spa->spa_autoexpand)) || 2147 (asize < vd->vdev_asize))) 2148 vd->vdev_asize = asize; 2149 2150 vdev_set_min_asize(vd); 2151 2152 /* 2153 * Ensure we can issue some IO before declaring the 2154 * vdev open for business. 2155 */ 2156 if (vd->vdev_ops->vdev_op_leaf && 2157 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 2158 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 2159 VDEV_AUX_ERR_EXCEEDED); 2160 return (error); 2161 } 2162 2163 /* 2164 * Track the minimum allocation size. 2165 */ 2166 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 2167 vd->vdev_islog == 0 && vd->vdev_aux == NULL) { 2168 uint64_t min_alloc = vdev_get_min_alloc(vd); 2169 if (min_alloc < spa->spa_min_alloc) 2170 spa->spa_min_alloc = min_alloc; 2171 } 2172 2173 /* 2174 * If this is a leaf vdev, assess whether a resilver is needed. 2175 * But don't do this if we are doing a reopen for a scrub, since 2176 * this would just restart the scrub we are already doing. 2177 */ 2178 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) 2179 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); 2180 2181 return (0); 2182 } 2183 2184 static void 2185 vdev_validate_child(void *arg) 2186 { 2187 vdev_t *vd = arg; 2188 2189 vd->vdev_validate_thread = curthread; 2190 vd->vdev_validate_error = vdev_validate(vd); 2191 vd->vdev_validate_thread = NULL; 2192 } 2193 2194 /* 2195 * Called once the vdevs are all opened, this routine validates the label 2196 * contents. This needs to be done before vdev_load() so that we don't 2197 * inadvertently do repair I/Os to the wrong device. 2198 * 2199 * This function will only return failure if one of the vdevs indicates that it 2200 * has since been destroyed or exported. This is only possible if 2201 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 2202 * will be updated but the function will return 0. 2203 */ 2204 int 2205 vdev_validate(vdev_t *vd) 2206 { 2207 spa_t *spa = vd->vdev_spa; 2208 taskq_t *tq = NULL; 2209 nvlist_t *label; 2210 uint64_t guid = 0, aux_guid = 0, top_guid; 2211 uint64_t state; 2212 nvlist_t *nvl; 2213 uint64_t txg; 2214 int children = vd->vdev_children; 2215 2216 if (vdev_validate_skip) 2217 return (0); 2218 2219 if (children > 0) { 2220 tq = taskq_create("vdev_validate", children, minclsyspri, 2221 children, children, TASKQ_PREPOPULATE); 2222 } 2223 2224 for (uint64_t c = 0; c < children; c++) { 2225 vdev_t *cvd = vd->vdev_child[c]; 2226 2227 if (tq == NULL || vdev_uses_zvols(cvd)) { 2228 vdev_validate_child(cvd); 2229 } else { 2230 VERIFY(taskq_dispatch(tq, vdev_validate_child, cvd, 2231 TQ_SLEEP) != TASKQID_INVALID); 2232 } 2233 } 2234 if (tq != NULL) { 2235 taskq_wait(tq); 2236 taskq_destroy(tq); 2237 } 2238 for (int c = 0; c < children; c++) { 2239 int error = vd->vdev_child[c]->vdev_validate_error; 2240 2241 if (error != 0) 2242 return (SET_ERROR(EBADF)); 2243 } 2244 2245 2246 /* 2247 * If the device has already failed, or was marked offline, don't do 2248 * any further validation. Otherwise, label I/O will fail and we will 2249 * overwrite the previous state. 2250 */ 2251 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 2252 return (0); 2253 2254 /* 2255 * If we are performing an extreme rewind, we allow for a label that 2256 * was modified at a point after the current txg. 2257 * If config lock is not held do not check for the txg. spa_sync could 2258 * be updating the vdev's label before updating spa_last_synced_txg. 2259 */ 2260 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 2261 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 2262 txg = UINT64_MAX; 2263 else 2264 txg = spa_last_synced_txg(spa); 2265 2266 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 2267 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2268 VDEV_AUX_BAD_LABEL); 2269 vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 2270 "txg %llu", (u_longlong_t)txg); 2271 return (0); 2272 } 2273 2274 /* 2275 * Determine if this vdev has been split off into another 2276 * pool. If so, then refuse to open it. 2277 */ 2278 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 2279 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 2280 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2281 VDEV_AUX_SPLIT_POOL); 2282 nvlist_free(label); 2283 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 2284 return (0); 2285 } 2286 2287 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 2288 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2289 VDEV_AUX_CORRUPT_DATA); 2290 nvlist_free(label); 2291 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2292 ZPOOL_CONFIG_POOL_GUID); 2293 return (0); 2294 } 2295 2296 /* 2297 * If config is not trusted then ignore the spa guid check. This is 2298 * necessary because if the machine crashed during a re-guid the new 2299 * guid might have been written to all of the vdev labels, but not the 2300 * cached config. The check will be performed again once we have the 2301 * trusted config from the MOS. 2302 */ 2303 if (spa->spa_trust_config && guid != spa_guid(spa)) { 2304 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2305 VDEV_AUX_CORRUPT_DATA); 2306 nvlist_free(label); 2307 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 2308 "match config (%llu != %llu)", (u_longlong_t)guid, 2309 (u_longlong_t)spa_guid(spa)); 2310 return (0); 2311 } 2312 2313 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 2314 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 2315 &aux_guid) != 0) 2316 aux_guid = 0; 2317 2318 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 2319 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2320 VDEV_AUX_CORRUPT_DATA); 2321 nvlist_free(label); 2322 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2323 ZPOOL_CONFIG_GUID); 2324 return (0); 2325 } 2326 2327 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 2328 != 0) { 2329 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2330 VDEV_AUX_CORRUPT_DATA); 2331 nvlist_free(label); 2332 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2333 ZPOOL_CONFIG_TOP_GUID); 2334 return (0); 2335 } 2336 2337 /* 2338 * If this vdev just became a top-level vdev because its sibling was 2339 * detached, it will have adopted the parent's vdev guid -- but the 2340 * label may or may not be on disk yet. Fortunately, either version 2341 * of the label will have the same top guid, so if we're a top-level 2342 * vdev, we can safely compare to that instead. 2343 * However, if the config comes from a cachefile that failed to update 2344 * after the detach, a top-level vdev will appear as a non top-level 2345 * vdev in the config. Also relax the constraints if we perform an 2346 * extreme rewind. 2347 * 2348 * If we split this vdev off instead, then we also check the 2349 * original pool's guid. We don't want to consider the vdev 2350 * corrupt if it is partway through a split operation. 2351 */ 2352 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 2353 boolean_t mismatch = B_FALSE; 2354 if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 2355 if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 2356 mismatch = B_TRUE; 2357 } else { 2358 if (vd->vdev_guid != top_guid && 2359 vd->vdev_top->vdev_guid != guid) 2360 mismatch = B_TRUE; 2361 } 2362 2363 if (mismatch) { 2364 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2365 VDEV_AUX_CORRUPT_DATA); 2366 nvlist_free(label); 2367 vdev_dbgmsg(vd, "vdev_validate: config guid " 2368 "doesn't match label guid"); 2369 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 2370 (u_longlong_t)vd->vdev_guid, 2371 (u_longlong_t)vd->vdev_top->vdev_guid); 2372 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 2373 "aux_guid %llu", (u_longlong_t)guid, 2374 (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 2375 return (0); 2376 } 2377 } 2378 2379 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 2380 &state) != 0) { 2381 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2382 VDEV_AUX_CORRUPT_DATA); 2383 nvlist_free(label); 2384 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2385 ZPOOL_CONFIG_POOL_STATE); 2386 return (0); 2387 } 2388 2389 nvlist_free(label); 2390 2391 /* 2392 * If this is a verbatim import, no need to check the 2393 * state of the pool. 2394 */ 2395 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 2396 spa_load_state(spa) == SPA_LOAD_OPEN && 2397 state != POOL_STATE_ACTIVE) { 2398 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 2399 "for spa %s", (u_longlong_t)state, spa->spa_name); 2400 return (SET_ERROR(EBADF)); 2401 } 2402 2403 /* 2404 * If we were able to open and validate a vdev that was 2405 * previously marked permanently unavailable, clear that state 2406 * now. 2407 */ 2408 if (vd->vdev_not_present) 2409 vd->vdev_not_present = 0; 2410 2411 return (0); 2412 } 2413 2414 static void 2415 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 2416 { 2417 char *old, *new; 2418 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 2419 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 2420 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 2421 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 2422 dvd->vdev_path, svd->vdev_path); 2423 spa_strfree(dvd->vdev_path); 2424 dvd->vdev_path = spa_strdup(svd->vdev_path); 2425 } 2426 } else if (svd->vdev_path != NULL) { 2427 dvd->vdev_path = spa_strdup(svd->vdev_path); 2428 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 2429 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 2430 } 2431 2432 /* 2433 * Our enclosure sysfs path may have changed between imports 2434 */ 2435 old = dvd->vdev_enc_sysfs_path; 2436 new = svd->vdev_enc_sysfs_path; 2437 if ((old != NULL && new == NULL) || 2438 (old == NULL && new != NULL) || 2439 ((old != NULL && new != NULL) && strcmp(new, old) != 0)) { 2440 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path " 2441 "changed from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 2442 old, new); 2443 2444 if (dvd->vdev_enc_sysfs_path) 2445 spa_strfree(dvd->vdev_enc_sysfs_path); 2446 2447 if (svd->vdev_enc_sysfs_path) { 2448 dvd->vdev_enc_sysfs_path = spa_strdup( 2449 svd->vdev_enc_sysfs_path); 2450 } else { 2451 dvd->vdev_enc_sysfs_path = NULL; 2452 } 2453 } 2454 } 2455 2456 /* 2457 * Recursively copy vdev paths from one vdev to another. Source and destination 2458 * vdev trees must have same geometry otherwise return error. Intended to copy 2459 * paths from userland config into MOS config. 2460 */ 2461 int 2462 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 2463 { 2464 if ((svd->vdev_ops == &vdev_missing_ops) || 2465 (svd->vdev_ishole && dvd->vdev_ishole) || 2466 (dvd->vdev_ops == &vdev_indirect_ops)) 2467 return (0); 2468 2469 if (svd->vdev_ops != dvd->vdev_ops) { 2470 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 2471 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 2472 return (SET_ERROR(EINVAL)); 2473 } 2474 2475 if (svd->vdev_guid != dvd->vdev_guid) { 2476 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 2477 "%llu)", (u_longlong_t)svd->vdev_guid, 2478 (u_longlong_t)dvd->vdev_guid); 2479 return (SET_ERROR(EINVAL)); 2480 } 2481 2482 if (svd->vdev_children != dvd->vdev_children) { 2483 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 2484 "%llu != %llu", (u_longlong_t)svd->vdev_children, 2485 (u_longlong_t)dvd->vdev_children); 2486 return (SET_ERROR(EINVAL)); 2487 } 2488 2489 for (uint64_t i = 0; i < svd->vdev_children; i++) { 2490 int error = vdev_copy_path_strict(svd->vdev_child[i], 2491 dvd->vdev_child[i]); 2492 if (error != 0) 2493 return (error); 2494 } 2495 2496 if (svd->vdev_ops->vdev_op_leaf) 2497 vdev_copy_path_impl(svd, dvd); 2498 2499 return (0); 2500 } 2501 2502 static void 2503 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 2504 { 2505 ASSERT(stvd->vdev_top == stvd); 2506 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 2507 2508 for (uint64_t i = 0; i < dvd->vdev_children; i++) { 2509 vdev_copy_path_search(stvd, dvd->vdev_child[i]); 2510 } 2511 2512 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 2513 return; 2514 2515 /* 2516 * The idea here is that while a vdev can shift positions within 2517 * a top vdev (when replacing, attaching mirror, etc.) it cannot 2518 * step outside of it. 2519 */ 2520 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 2521 2522 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 2523 return; 2524 2525 ASSERT(vd->vdev_ops->vdev_op_leaf); 2526 2527 vdev_copy_path_impl(vd, dvd); 2528 } 2529 2530 /* 2531 * Recursively copy vdev paths from one root vdev to another. Source and 2532 * destination vdev trees may differ in geometry. For each destination leaf 2533 * vdev, search a vdev with the same guid and top vdev id in the source. 2534 * Intended to copy paths from userland config into MOS config. 2535 */ 2536 void 2537 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 2538 { 2539 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 2540 ASSERT(srvd->vdev_ops == &vdev_root_ops); 2541 ASSERT(drvd->vdev_ops == &vdev_root_ops); 2542 2543 for (uint64_t i = 0; i < children; i++) { 2544 vdev_copy_path_search(srvd->vdev_child[i], 2545 drvd->vdev_child[i]); 2546 } 2547 } 2548 2549 /* 2550 * Close a virtual device. 2551 */ 2552 void 2553 vdev_close(vdev_t *vd) 2554 { 2555 vdev_t *pvd = vd->vdev_parent; 2556 spa_t *spa __maybe_unused = vd->vdev_spa; 2557 2558 ASSERT(vd != NULL); 2559 ASSERT(vd->vdev_open_thread == curthread || 2560 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2561 2562 /* 2563 * If our parent is reopening, then we are as well, unless we are 2564 * going offline. 2565 */ 2566 if (pvd != NULL && pvd->vdev_reopening) 2567 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2568 2569 vd->vdev_ops->vdev_op_close(vd); 2570 2571 vdev_cache_purge(vd); 2572 2573 /* 2574 * We record the previous state before we close it, so that if we are 2575 * doing a reopen(), we don't generate FMA ereports if we notice that 2576 * it's still faulted. 2577 */ 2578 vd->vdev_prevstate = vd->vdev_state; 2579 2580 if (vd->vdev_offline) 2581 vd->vdev_state = VDEV_STATE_OFFLINE; 2582 else 2583 vd->vdev_state = VDEV_STATE_CLOSED; 2584 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2585 } 2586 2587 void 2588 vdev_hold(vdev_t *vd) 2589 { 2590 spa_t *spa = vd->vdev_spa; 2591 2592 ASSERT(spa_is_root(spa)); 2593 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2594 return; 2595 2596 for (int c = 0; c < vd->vdev_children; c++) 2597 vdev_hold(vd->vdev_child[c]); 2598 2599 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_hold != NULL) 2600 vd->vdev_ops->vdev_op_hold(vd); 2601 } 2602 2603 void 2604 vdev_rele(vdev_t *vd) 2605 { 2606 ASSERT(spa_is_root(vd->vdev_spa)); 2607 for (int c = 0; c < vd->vdev_children; c++) 2608 vdev_rele(vd->vdev_child[c]); 2609 2610 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_rele != NULL) 2611 vd->vdev_ops->vdev_op_rele(vd); 2612 } 2613 2614 /* 2615 * Reopen all interior vdevs and any unopened leaves. We don't actually 2616 * reopen leaf vdevs which had previously been opened as they might deadlock 2617 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2618 * If the leaf has never been opened then open it, as usual. 2619 */ 2620 void 2621 vdev_reopen(vdev_t *vd) 2622 { 2623 spa_t *spa = vd->vdev_spa; 2624 2625 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2626 2627 /* set the reopening flag unless we're taking the vdev offline */ 2628 vd->vdev_reopening = !vd->vdev_offline; 2629 vdev_close(vd); 2630 (void) vdev_open(vd); 2631 2632 /* 2633 * Call vdev_validate() here to make sure we have the same device. 2634 * Otherwise, a device with an invalid label could be successfully 2635 * opened in response to vdev_reopen(). 2636 */ 2637 if (vd->vdev_aux) { 2638 (void) vdev_validate_aux(vd); 2639 if (vdev_readable(vd) && vdev_writeable(vd) && 2640 vd->vdev_aux == &spa->spa_l2cache) { 2641 /* 2642 * In case the vdev is present we should evict all ARC 2643 * buffers and pointers to log blocks and reclaim their 2644 * space before restoring its contents to L2ARC. 2645 */ 2646 if (l2arc_vdev_present(vd)) { 2647 l2arc_rebuild_vdev(vd, B_TRUE); 2648 } else { 2649 l2arc_add_vdev(spa, vd); 2650 } 2651 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); 2652 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); 2653 } 2654 } else { 2655 (void) vdev_validate(vd); 2656 } 2657 2658 /* 2659 * Reassess parent vdev's health. 2660 */ 2661 vdev_propagate_state(vd); 2662 } 2663 2664 int 2665 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2666 { 2667 int error; 2668 2669 /* 2670 * Normally, partial opens (e.g. of a mirror) are allowed. 2671 * For a create, however, we want to fail the request if 2672 * there are any components we can't open. 2673 */ 2674 error = vdev_open(vd); 2675 2676 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2677 vdev_close(vd); 2678 return (error ? error : SET_ERROR(ENXIO)); 2679 } 2680 2681 /* 2682 * Recursively load DTLs and initialize all labels. 2683 */ 2684 if ((error = vdev_dtl_load(vd)) != 0 || 2685 (error = vdev_label_init(vd, txg, isreplacing ? 2686 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2687 vdev_close(vd); 2688 return (error); 2689 } 2690 2691 return (0); 2692 } 2693 2694 void 2695 vdev_metaslab_set_size(vdev_t *vd) 2696 { 2697 uint64_t asize = vd->vdev_asize; 2698 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2699 uint64_t ms_shift; 2700 2701 /* 2702 * There are two dimensions to the metaslab sizing calculation: 2703 * the size of the metaslab and the count of metaslabs per vdev. 2704 * 2705 * The default values used below are a good balance between memory 2706 * usage (larger metaslab size means more memory needed for loaded 2707 * metaslabs; more metaslabs means more memory needed for the 2708 * metaslab_t structs), metaslab load time (larger metaslabs take 2709 * longer to load), and metaslab sync time (more metaslabs means 2710 * more time spent syncing all of them). 2711 * 2712 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2713 * The range of the dimensions are as follows: 2714 * 2715 * 2^29 <= ms_size <= 2^34 2716 * 16 <= ms_count <= 131,072 2717 * 2718 * On the lower end of vdev sizes, we aim for metaslabs sizes of 2719 * at least 512MB (2^29) to minimize fragmentation effects when 2720 * testing with smaller devices. However, the count constraint 2721 * of at least 16 metaslabs will override this minimum size goal. 2722 * 2723 * On the upper end of vdev sizes, we aim for a maximum metaslab 2724 * size of 16GB. However, we will cap the total count to 2^17 2725 * metaslabs to keep our memory footprint in check and let the 2726 * metaslab size grow from there if that limit is hit. 2727 * 2728 * The net effect of applying above constrains is summarized below. 2729 * 2730 * vdev size metaslab count 2731 * --------------|----------------- 2732 * < 8GB ~16 2733 * 8GB - 100GB one per 512MB 2734 * 100GB - 3TB ~200 2735 * 3TB - 2PB one per 16GB 2736 * > 2PB ~131,072 2737 * -------------------------------- 2738 * 2739 * Finally, note that all of the above calculate the initial 2740 * number of metaslabs. Expanding a top-level vdev will result 2741 * in additional metaslabs being allocated making it possible 2742 * to exceed the zfs_vdev_ms_count_limit. 2743 */ 2744 2745 if (ms_count < zfs_vdev_min_ms_count) 2746 ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2747 else if (ms_count > zfs_vdev_default_ms_count) 2748 ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2749 else 2750 ms_shift = zfs_vdev_default_ms_shift; 2751 2752 if (ms_shift < SPA_MAXBLOCKSHIFT) { 2753 ms_shift = SPA_MAXBLOCKSHIFT; 2754 } else if (ms_shift > zfs_vdev_max_ms_shift) { 2755 ms_shift = zfs_vdev_max_ms_shift; 2756 /* cap the total count to constrain memory footprint */ 2757 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2758 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 2759 } 2760 2761 vd->vdev_ms_shift = ms_shift; 2762 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2763 } 2764 2765 void 2766 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2767 { 2768 ASSERT(vd == vd->vdev_top); 2769 /* indirect vdevs don't have metaslabs or dtls */ 2770 ASSERT(vdev_is_concrete(vd) || flags == 0); 2771 ASSERT(ISP2(flags)); 2772 ASSERT(spa_writeable(vd->vdev_spa)); 2773 2774 if (flags & VDD_METASLAB) 2775 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2776 2777 if (flags & VDD_DTL) 2778 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2779 2780 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2781 } 2782 2783 void 2784 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 2785 { 2786 for (int c = 0; c < vd->vdev_children; c++) 2787 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 2788 2789 if (vd->vdev_ops->vdev_op_leaf) 2790 vdev_dirty(vd->vdev_top, flags, vd, txg); 2791 } 2792 2793 /* 2794 * DTLs. 2795 * 2796 * A vdev's DTL (dirty time log) is the set of transaction groups for which 2797 * the vdev has less than perfect replication. There are four kinds of DTL: 2798 * 2799 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 2800 * 2801 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 2802 * 2803 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 2804 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 2805 * txgs that was scrubbed. 2806 * 2807 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 2808 * persistent errors or just some device being offline. 2809 * Unlike the other three, the DTL_OUTAGE map is not generally 2810 * maintained; it's only computed when needed, typically to 2811 * determine whether a device can be detached. 2812 * 2813 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 2814 * either has the data or it doesn't. 2815 * 2816 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 2817 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 2818 * if any child is less than fully replicated, then so is its parent. 2819 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 2820 * comprising only those txgs which appear in 'maxfaults' or more children; 2821 * those are the txgs we don't have enough replication to read. For example, 2822 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 2823 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 2824 * two child DTL_MISSING maps. 2825 * 2826 * It should be clear from the above that to compute the DTLs and outage maps 2827 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 2828 * Therefore, that is all we keep on disk. When loading the pool, or after 2829 * a configuration change, we generate all other DTLs from first principles. 2830 */ 2831 void 2832 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2833 { 2834 range_tree_t *rt = vd->vdev_dtl[t]; 2835 2836 ASSERT(t < DTL_TYPES); 2837 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2838 ASSERT(spa_writeable(vd->vdev_spa)); 2839 2840 mutex_enter(&vd->vdev_dtl_lock); 2841 if (!range_tree_contains(rt, txg, size)) 2842 range_tree_add(rt, txg, size); 2843 mutex_exit(&vd->vdev_dtl_lock); 2844 } 2845 2846 boolean_t 2847 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2848 { 2849 range_tree_t *rt = vd->vdev_dtl[t]; 2850 boolean_t dirty = B_FALSE; 2851 2852 ASSERT(t < DTL_TYPES); 2853 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2854 2855 /* 2856 * While we are loading the pool, the DTLs have not been loaded yet. 2857 * This isn't a problem but it can result in devices being tried 2858 * which are known to not have the data. In which case, the import 2859 * is relying on the checksum to ensure that we get the right data. 2860 * Note that while importing we are only reading the MOS, which is 2861 * always checksummed. 2862 */ 2863 mutex_enter(&vd->vdev_dtl_lock); 2864 if (!range_tree_is_empty(rt)) 2865 dirty = range_tree_contains(rt, txg, size); 2866 mutex_exit(&vd->vdev_dtl_lock); 2867 2868 return (dirty); 2869 } 2870 2871 boolean_t 2872 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 2873 { 2874 range_tree_t *rt = vd->vdev_dtl[t]; 2875 boolean_t empty; 2876 2877 mutex_enter(&vd->vdev_dtl_lock); 2878 empty = range_tree_is_empty(rt); 2879 mutex_exit(&vd->vdev_dtl_lock); 2880 2881 return (empty); 2882 } 2883 2884 /* 2885 * Check if the txg falls within the range which must be 2886 * resilvered. DVAs outside this range can always be skipped. 2887 */ 2888 boolean_t 2889 vdev_default_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, 2890 uint64_t phys_birth) 2891 { 2892 (void) dva, (void) psize; 2893 2894 /* Set by sequential resilver. */ 2895 if (phys_birth == TXG_UNKNOWN) 2896 return (B_TRUE); 2897 2898 return (vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1)); 2899 } 2900 2901 /* 2902 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered. 2903 */ 2904 boolean_t 2905 vdev_dtl_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, 2906 uint64_t phys_birth) 2907 { 2908 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2909 2910 if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2911 vd->vdev_ops->vdev_op_leaf) 2912 return (B_TRUE); 2913 2914 return (vd->vdev_ops->vdev_op_need_resilver(vd, dva, psize, 2915 phys_birth)); 2916 } 2917 2918 /* 2919 * Returns the lowest txg in the DTL range. 2920 */ 2921 static uint64_t 2922 vdev_dtl_min(vdev_t *vd) 2923 { 2924 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2925 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2926 ASSERT0(vd->vdev_children); 2927 2928 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); 2929 } 2930 2931 /* 2932 * Returns the highest txg in the DTL. 2933 */ 2934 static uint64_t 2935 vdev_dtl_max(vdev_t *vd) 2936 { 2937 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2938 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2939 ASSERT0(vd->vdev_children); 2940 2941 return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); 2942 } 2943 2944 /* 2945 * Determine if a resilvering vdev should remove any DTL entries from 2946 * its range. If the vdev was resilvering for the entire duration of the 2947 * scan then it should excise that range from its DTLs. Otherwise, this 2948 * vdev is considered partially resilvered and should leave its DTL 2949 * entries intact. The comment in vdev_dtl_reassess() describes how we 2950 * excise the DTLs. 2951 */ 2952 static boolean_t 2953 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done) 2954 { 2955 ASSERT0(vd->vdev_children); 2956 2957 if (vd->vdev_state < VDEV_STATE_DEGRADED) 2958 return (B_FALSE); 2959 2960 if (vd->vdev_resilver_deferred) 2961 return (B_FALSE); 2962 2963 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2964 return (B_TRUE); 2965 2966 if (rebuild_done) { 2967 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; 2968 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; 2969 2970 /* Rebuild not initiated by attach */ 2971 if (vd->vdev_rebuild_txg == 0) 2972 return (B_TRUE); 2973 2974 /* 2975 * When a rebuild completes without error then all missing data 2976 * up to the rebuild max txg has been reconstructed and the DTL 2977 * is eligible for excision. 2978 */ 2979 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE && 2980 vdev_dtl_max(vd) <= vrp->vrp_max_txg) { 2981 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd)); 2982 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg); 2983 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg); 2984 return (B_TRUE); 2985 } 2986 } else { 2987 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan; 2988 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys; 2989 2990 /* Resilver not initiated by attach */ 2991 if (vd->vdev_resilver_txg == 0) 2992 return (B_TRUE); 2993 2994 /* 2995 * When a resilver is initiated the scan will assign the 2996 * scn_max_txg value to the highest txg value that exists 2997 * in all DTLs. If this device's max DTL is not part of this 2998 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg] 2999 * then it is not eligible for excision. 3000 */ 3001 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 3002 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd)); 3003 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg); 3004 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg); 3005 return (B_TRUE); 3006 } 3007 } 3008 3009 return (B_FALSE); 3010 } 3011 3012 /* 3013 * Reassess DTLs after a config change or scrub completion. If txg == 0 no 3014 * write operations will be issued to the pool. 3015 */ 3016 void 3017 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, 3018 boolean_t scrub_done, boolean_t rebuild_done) 3019 { 3020 spa_t *spa = vd->vdev_spa; 3021 avl_tree_t reftree; 3022 int minref; 3023 3024 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3025 3026 for (int c = 0; c < vd->vdev_children; c++) 3027 vdev_dtl_reassess(vd->vdev_child[c], txg, 3028 scrub_txg, scrub_done, rebuild_done); 3029 3030 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 3031 return; 3032 3033 if (vd->vdev_ops->vdev_op_leaf) { 3034 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 3035 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; 3036 boolean_t check_excise = B_FALSE; 3037 boolean_t wasempty = B_TRUE; 3038 3039 mutex_enter(&vd->vdev_dtl_lock); 3040 3041 /* 3042 * If requested, pretend the scan or rebuild completed cleanly. 3043 */ 3044 if (zfs_scan_ignore_errors) { 3045 if (scn != NULL) 3046 scn->scn_phys.scn_errors = 0; 3047 if (vr != NULL) 3048 vr->vr_rebuild_phys.vrp_errors = 0; 3049 } 3050 3051 if (scrub_txg != 0 && 3052 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 3053 wasempty = B_FALSE; 3054 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d " 3055 "dtl:%llu/%llu errors:%llu", 3056 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg, 3057 (u_longlong_t)scrub_txg, spa->spa_scrub_started, 3058 (u_longlong_t)vdev_dtl_min(vd), 3059 (u_longlong_t)vdev_dtl_max(vd), 3060 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0)); 3061 } 3062 3063 /* 3064 * If we've completed a scrub/resilver or a rebuild cleanly 3065 * then determine if this vdev should remove any DTLs. We 3066 * only want to excise regions on vdevs that were available 3067 * during the entire duration of this scan. 3068 */ 3069 if (rebuild_done && 3070 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) { 3071 check_excise = B_TRUE; 3072 } else { 3073 if (spa->spa_scrub_started || 3074 (scn != NULL && scn->scn_phys.scn_errors == 0)) { 3075 check_excise = B_TRUE; 3076 } 3077 } 3078 3079 if (scrub_txg && check_excise && 3080 vdev_dtl_should_excise(vd, rebuild_done)) { 3081 /* 3082 * We completed a scrub, resilver or rebuild up to 3083 * scrub_txg. If we did it without rebooting, then 3084 * the scrub dtl will be valid, so excise the old 3085 * region and fold in the scrub dtl. Otherwise, 3086 * leave the dtl as-is if there was an error. 3087 * 3088 * There's little trick here: to excise the beginning 3089 * of the DTL_MISSING map, we put it into a reference 3090 * tree and then add a segment with refcnt -1 that 3091 * covers the range [0, scrub_txg). This means 3092 * that each txg in that range has refcnt -1 or 0. 3093 * We then add DTL_SCRUB with a refcnt of 2, so that 3094 * entries in the range [0, scrub_txg) will have a 3095 * positive refcnt -- either 1 or 2. We then convert 3096 * the reference tree into the new DTL_MISSING map. 3097 */ 3098 space_reftree_create(&reftree); 3099 space_reftree_add_map(&reftree, 3100 vd->vdev_dtl[DTL_MISSING], 1); 3101 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 3102 space_reftree_add_map(&reftree, 3103 vd->vdev_dtl[DTL_SCRUB], 2); 3104 space_reftree_generate_map(&reftree, 3105 vd->vdev_dtl[DTL_MISSING], 1); 3106 space_reftree_destroy(&reftree); 3107 3108 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 3109 zfs_dbgmsg("update DTL_MISSING:%llu/%llu", 3110 (u_longlong_t)vdev_dtl_min(vd), 3111 (u_longlong_t)vdev_dtl_max(vd)); 3112 } else if (!wasempty) { 3113 zfs_dbgmsg("DTL_MISSING is now empty"); 3114 } 3115 } 3116 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 3117 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 3118 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 3119 if (scrub_done) 3120 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 3121 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 3122 if (!vdev_readable(vd)) 3123 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 3124 else 3125 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 3126 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 3127 3128 /* 3129 * If the vdev was resilvering or rebuilding and no longer 3130 * has any DTLs then reset the appropriate flag and dirty 3131 * the top level so that we persist the change. 3132 */ 3133 if (txg != 0 && 3134 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 3135 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) { 3136 if (vd->vdev_rebuild_txg != 0) { 3137 vd->vdev_rebuild_txg = 0; 3138 vdev_config_dirty(vd->vdev_top); 3139 } else if (vd->vdev_resilver_txg != 0) { 3140 vd->vdev_resilver_txg = 0; 3141 vdev_config_dirty(vd->vdev_top); 3142 } 3143 } 3144 3145 mutex_exit(&vd->vdev_dtl_lock); 3146 3147 if (txg != 0) 3148 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 3149 return; 3150 } 3151 3152 mutex_enter(&vd->vdev_dtl_lock); 3153 for (int t = 0; t < DTL_TYPES; t++) { 3154 /* account for child's outage in parent's missing map */ 3155 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 3156 if (t == DTL_SCRUB) 3157 continue; /* leaf vdevs only */ 3158 if (t == DTL_PARTIAL) 3159 minref = 1; /* i.e. non-zero */ 3160 else if (vdev_get_nparity(vd) != 0) 3161 minref = vdev_get_nparity(vd) + 1; /* RAID-Z, dRAID */ 3162 else 3163 minref = vd->vdev_children; /* any kind of mirror */ 3164 space_reftree_create(&reftree); 3165 for (int c = 0; c < vd->vdev_children; c++) { 3166 vdev_t *cvd = vd->vdev_child[c]; 3167 mutex_enter(&cvd->vdev_dtl_lock); 3168 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 3169 mutex_exit(&cvd->vdev_dtl_lock); 3170 } 3171 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 3172 space_reftree_destroy(&reftree); 3173 } 3174 mutex_exit(&vd->vdev_dtl_lock); 3175 } 3176 3177 /* 3178 * Iterate over all the vdevs except spare, and post kobj events 3179 */ 3180 void 3181 vdev_post_kobj_evt(vdev_t *vd) 3182 { 3183 if (vd->vdev_ops->vdev_op_kobj_evt_post && 3184 vd->vdev_kobj_flag == B_FALSE) { 3185 vd->vdev_kobj_flag = B_TRUE; 3186 vd->vdev_ops->vdev_op_kobj_evt_post(vd); 3187 } 3188 3189 for (int c = 0; c < vd->vdev_children; c++) 3190 vdev_post_kobj_evt(vd->vdev_child[c]); 3191 } 3192 3193 /* 3194 * Iterate over all the vdevs except spare, and clear kobj events 3195 */ 3196 void 3197 vdev_clear_kobj_evt(vdev_t *vd) 3198 { 3199 vd->vdev_kobj_flag = B_FALSE; 3200 3201 for (int c = 0; c < vd->vdev_children; c++) 3202 vdev_clear_kobj_evt(vd->vdev_child[c]); 3203 } 3204 3205 int 3206 vdev_dtl_load(vdev_t *vd) 3207 { 3208 spa_t *spa = vd->vdev_spa; 3209 objset_t *mos = spa->spa_meta_objset; 3210 range_tree_t *rt; 3211 int error = 0; 3212 3213 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 3214 ASSERT(vdev_is_concrete(vd)); 3215 3216 /* 3217 * If the dtl cannot be sync'd there is no need to open it. 3218 */ 3219 if (spa->spa_mode == SPA_MODE_READ && !spa->spa_read_spacemaps) 3220 return (0); 3221 3222 error = space_map_open(&vd->vdev_dtl_sm, mos, 3223 vd->vdev_dtl_object, 0, -1ULL, 0); 3224 if (error) 3225 return (error); 3226 ASSERT(vd->vdev_dtl_sm != NULL); 3227 3228 rt = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 3229 error = space_map_load(vd->vdev_dtl_sm, rt, SM_ALLOC); 3230 if (error == 0) { 3231 mutex_enter(&vd->vdev_dtl_lock); 3232 range_tree_walk(rt, range_tree_add, 3233 vd->vdev_dtl[DTL_MISSING]); 3234 mutex_exit(&vd->vdev_dtl_lock); 3235 } 3236 3237 range_tree_vacate(rt, NULL, NULL); 3238 range_tree_destroy(rt); 3239 3240 return (error); 3241 } 3242 3243 for (int c = 0; c < vd->vdev_children; c++) { 3244 error = vdev_dtl_load(vd->vdev_child[c]); 3245 if (error != 0) 3246 break; 3247 } 3248 3249 return (error); 3250 } 3251 3252 static void 3253 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 3254 { 3255 spa_t *spa = vd->vdev_spa; 3256 objset_t *mos = spa->spa_meta_objset; 3257 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 3258 const char *string; 3259 3260 ASSERT(alloc_bias != VDEV_BIAS_NONE); 3261 3262 string = 3263 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 3264 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 3265 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 3266 3267 ASSERT(string != NULL); 3268 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 3269 1, strlen(string) + 1, string, tx)); 3270 3271 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 3272 spa_activate_allocation_classes(spa, tx); 3273 } 3274 } 3275 3276 void 3277 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 3278 { 3279 spa_t *spa = vd->vdev_spa; 3280 3281 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 3282 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 3283 zapobj, tx)); 3284 } 3285 3286 uint64_t 3287 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 3288 { 3289 spa_t *spa = vd->vdev_spa; 3290 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 3291 DMU_OT_NONE, 0, tx); 3292 3293 ASSERT(zap != 0); 3294 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 3295 zap, tx)); 3296 3297 return (zap); 3298 } 3299 3300 void 3301 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 3302 { 3303 if (vd->vdev_ops != &vdev_hole_ops && 3304 vd->vdev_ops != &vdev_missing_ops && 3305 vd->vdev_ops != &vdev_root_ops && 3306 !vd->vdev_top->vdev_removing) { 3307 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 3308 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 3309 } 3310 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 3311 vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 3312 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 3313 vdev_zap_allocation_data(vd, tx); 3314 } 3315 } 3316 3317 for (uint64_t i = 0; i < vd->vdev_children; i++) { 3318 vdev_construct_zaps(vd->vdev_child[i], tx); 3319 } 3320 } 3321 3322 static void 3323 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 3324 { 3325 spa_t *spa = vd->vdev_spa; 3326 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 3327 objset_t *mos = spa->spa_meta_objset; 3328 range_tree_t *rtsync; 3329 dmu_tx_t *tx; 3330 uint64_t object = space_map_object(vd->vdev_dtl_sm); 3331 3332 ASSERT(vdev_is_concrete(vd)); 3333 ASSERT(vd->vdev_ops->vdev_op_leaf); 3334 3335 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3336 3337 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 3338 mutex_enter(&vd->vdev_dtl_lock); 3339 space_map_free(vd->vdev_dtl_sm, tx); 3340 space_map_close(vd->vdev_dtl_sm); 3341 vd->vdev_dtl_sm = NULL; 3342 mutex_exit(&vd->vdev_dtl_lock); 3343 3344 /* 3345 * We only destroy the leaf ZAP for detached leaves or for 3346 * removed log devices. Removed data devices handle leaf ZAP 3347 * cleanup later, once cancellation is no longer possible. 3348 */ 3349 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 3350 vd->vdev_top->vdev_islog)) { 3351 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 3352 vd->vdev_leaf_zap = 0; 3353 } 3354 3355 dmu_tx_commit(tx); 3356 return; 3357 } 3358 3359 if (vd->vdev_dtl_sm == NULL) { 3360 uint64_t new_object; 3361 3362 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); 3363 VERIFY3U(new_object, !=, 0); 3364 3365 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 3366 0, -1ULL, 0)); 3367 ASSERT(vd->vdev_dtl_sm != NULL); 3368 } 3369 3370 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 3371 3372 mutex_enter(&vd->vdev_dtl_lock); 3373 range_tree_walk(rt, range_tree_add, rtsync); 3374 mutex_exit(&vd->vdev_dtl_lock); 3375 3376 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); 3377 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 3378 range_tree_vacate(rtsync, NULL, NULL); 3379 3380 range_tree_destroy(rtsync); 3381 3382 /* 3383 * If the object for the space map has changed then dirty 3384 * the top level so that we update the config. 3385 */ 3386 if (object != space_map_object(vd->vdev_dtl_sm)) { 3387 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 3388 "new object %llu", (u_longlong_t)txg, spa_name(spa), 3389 (u_longlong_t)object, 3390 (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 3391 vdev_config_dirty(vd->vdev_top); 3392 } 3393 3394 dmu_tx_commit(tx); 3395 } 3396 3397 /* 3398 * Determine whether the specified vdev can be offlined/detached/removed 3399 * without losing data. 3400 */ 3401 boolean_t 3402 vdev_dtl_required(vdev_t *vd) 3403 { 3404 spa_t *spa = vd->vdev_spa; 3405 vdev_t *tvd = vd->vdev_top; 3406 uint8_t cant_read = vd->vdev_cant_read; 3407 boolean_t required; 3408 3409 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3410 3411 if (vd == spa->spa_root_vdev || vd == tvd) 3412 return (B_TRUE); 3413 3414 /* 3415 * Temporarily mark the device as unreadable, and then determine 3416 * whether this results in any DTL outages in the top-level vdev. 3417 * If not, we can safely offline/detach/remove the device. 3418 */ 3419 vd->vdev_cant_read = B_TRUE; 3420 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); 3421 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 3422 vd->vdev_cant_read = cant_read; 3423 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); 3424 3425 if (!required && zio_injection_enabled) { 3426 required = !!zio_handle_device_injection(vd, NULL, 3427 SET_ERROR(ECHILD)); 3428 } 3429 3430 return (required); 3431 } 3432 3433 /* 3434 * Determine if resilver is needed, and if so the txg range. 3435 */ 3436 boolean_t 3437 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 3438 { 3439 boolean_t needed = B_FALSE; 3440 uint64_t thismin = UINT64_MAX; 3441 uint64_t thismax = 0; 3442 3443 if (vd->vdev_children == 0) { 3444 mutex_enter(&vd->vdev_dtl_lock); 3445 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 3446 vdev_writeable(vd)) { 3447 3448 thismin = vdev_dtl_min(vd); 3449 thismax = vdev_dtl_max(vd); 3450 needed = B_TRUE; 3451 } 3452 mutex_exit(&vd->vdev_dtl_lock); 3453 } else { 3454 for (int c = 0; c < vd->vdev_children; c++) { 3455 vdev_t *cvd = vd->vdev_child[c]; 3456 uint64_t cmin, cmax; 3457 3458 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 3459 thismin = MIN(thismin, cmin); 3460 thismax = MAX(thismax, cmax); 3461 needed = B_TRUE; 3462 } 3463 } 3464 } 3465 3466 if (needed && minp) { 3467 *minp = thismin; 3468 *maxp = thismax; 3469 } 3470 return (needed); 3471 } 3472 3473 /* 3474 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj 3475 * will contain either the checkpoint spacemap object or zero if none exists. 3476 * All other errors are returned to the caller. 3477 */ 3478 int 3479 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj) 3480 { 3481 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 3482 3483 if (vd->vdev_top_zap == 0) { 3484 *sm_obj = 0; 3485 return (0); 3486 } 3487 3488 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 3489 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj); 3490 if (error == ENOENT) { 3491 *sm_obj = 0; 3492 error = 0; 3493 } 3494 3495 return (error); 3496 } 3497 3498 int 3499 vdev_load(vdev_t *vd) 3500 { 3501 int children = vd->vdev_children; 3502 int error = 0; 3503 taskq_t *tq = NULL; 3504 3505 /* 3506 * It's only worthwhile to use the taskq for the root vdev, because the 3507 * slow part is metaslab_init, and that only happens for top-level 3508 * vdevs. 3509 */ 3510 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_children > 0) { 3511 tq = taskq_create("vdev_load", children, minclsyspri, 3512 children, children, TASKQ_PREPOPULATE); 3513 } 3514 3515 /* 3516 * Recursively load all children. 3517 */ 3518 for (int c = 0; c < vd->vdev_children; c++) { 3519 vdev_t *cvd = vd->vdev_child[c]; 3520 3521 if (tq == NULL || vdev_uses_zvols(cvd)) { 3522 cvd->vdev_load_error = vdev_load(cvd); 3523 } else { 3524 VERIFY(taskq_dispatch(tq, vdev_load_child, 3525 cvd, TQ_SLEEP) != TASKQID_INVALID); 3526 } 3527 } 3528 3529 if (tq != NULL) { 3530 taskq_wait(tq); 3531 taskq_destroy(tq); 3532 } 3533 3534 for (int c = 0; c < vd->vdev_children; c++) { 3535 int error = vd->vdev_child[c]->vdev_load_error; 3536 3537 if (error != 0) 3538 return (error); 3539 } 3540 3541 vdev_set_deflate_ratio(vd); 3542 3543 /* 3544 * On spa_load path, grab the allocation bias from our zap 3545 */ 3546 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 3547 spa_t *spa = vd->vdev_spa; 3548 char bias_str[64]; 3549 3550 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 3551 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 3552 bias_str); 3553 if (error == 0) { 3554 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 3555 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 3556 } else if (error != ENOENT) { 3557 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3558 VDEV_AUX_CORRUPT_DATA); 3559 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) " 3560 "failed [error=%d]", 3561 (u_longlong_t)vd->vdev_top_zap, error); 3562 return (error); 3563 } 3564 } 3565 3566 /* 3567 * Load any rebuild state from the top-level vdev zap. 3568 */ 3569 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 3570 error = vdev_rebuild_load(vd); 3571 if (error && error != ENOTSUP) { 3572 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3573 VDEV_AUX_CORRUPT_DATA); 3574 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load " 3575 "failed [error=%d]", error); 3576 return (error); 3577 } 3578 } 3579 3580 /* 3581 * If this is a top-level vdev, initialize its metaslabs. 3582 */ 3583 if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 3584 vdev_metaslab_group_create(vd); 3585 3586 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 3587 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3588 VDEV_AUX_CORRUPT_DATA); 3589 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 3590 "asize=%llu", (u_longlong_t)vd->vdev_ashift, 3591 (u_longlong_t)vd->vdev_asize); 3592 return (SET_ERROR(ENXIO)); 3593 } 3594 3595 error = vdev_metaslab_init(vd, 0); 3596 if (error != 0) { 3597 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 3598 "[error=%d]", error); 3599 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3600 VDEV_AUX_CORRUPT_DATA); 3601 return (error); 3602 } 3603 3604 uint64_t checkpoint_sm_obj; 3605 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj); 3606 if (error == 0 && checkpoint_sm_obj != 0) { 3607 objset_t *mos = spa_meta_objset(vd->vdev_spa); 3608 ASSERT(vd->vdev_asize != 0); 3609 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 3610 3611 error = space_map_open(&vd->vdev_checkpoint_sm, 3612 mos, checkpoint_sm_obj, 0, vd->vdev_asize, 3613 vd->vdev_ashift); 3614 if (error != 0) { 3615 vdev_dbgmsg(vd, "vdev_load: space_map_open " 3616 "failed for checkpoint spacemap (obj %llu) " 3617 "[error=%d]", 3618 (u_longlong_t)checkpoint_sm_obj, error); 3619 return (error); 3620 } 3621 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 3622 3623 /* 3624 * Since the checkpoint_sm contains free entries 3625 * exclusively we can use space_map_allocated() to 3626 * indicate the cumulative checkpointed space that 3627 * has been freed. 3628 */ 3629 vd->vdev_stat.vs_checkpoint_space = 3630 -space_map_allocated(vd->vdev_checkpoint_sm); 3631 vd->vdev_spa->spa_checkpoint_info.sci_dspace += 3632 vd->vdev_stat.vs_checkpoint_space; 3633 } else if (error != 0) { 3634 vdev_dbgmsg(vd, "vdev_load: failed to retrieve " 3635 "checkpoint space map object from vdev ZAP " 3636 "[error=%d]", error); 3637 return (error); 3638 } 3639 } 3640 3641 /* 3642 * If this is a leaf vdev, load its DTL. 3643 */ 3644 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 3645 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3646 VDEV_AUX_CORRUPT_DATA); 3647 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 3648 "[error=%d]", error); 3649 return (error); 3650 } 3651 3652 uint64_t obsolete_sm_object; 3653 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object); 3654 if (error == 0 && obsolete_sm_object != 0) { 3655 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3656 ASSERT(vd->vdev_asize != 0); 3657 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 3658 3659 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 3660 obsolete_sm_object, 0, vd->vdev_asize, 0))) { 3661 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3662 VDEV_AUX_CORRUPT_DATA); 3663 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 3664 "obsolete spacemap (obj %llu) [error=%d]", 3665 (u_longlong_t)obsolete_sm_object, error); 3666 return (error); 3667 } 3668 } else if (error != 0) { 3669 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete " 3670 "space map object from vdev ZAP [error=%d]", error); 3671 return (error); 3672 } 3673 3674 return (0); 3675 } 3676 3677 /* 3678 * The special vdev case is used for hot spares and l2cache devices. Its 3679 * sole purpose it to set the vdev state for the associated vdev. To do this, 3680 * we make sure that we can open the underlying device, then try to read the 3681 * label, and make sure that the label is sane and that it hasn't been 3682 * repurposed to another pool. 3683 */ 3684 int 3685 vdev_validate_aux(vdev_t *vd) 3686 { 3687 nvlist_t *label; 3688 uint64_t guid, version; 3689 uint64_t state; 3690 3691 if (!vdev_readable(vd)) 3692 return (0); 3693 3694 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 3695 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3696 VDEV_AUX_CORRUPT_DATA); 3697 return (-1); 3698 } 3699 3700 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3701 !SPA_VERSION_IS_SUPPORTED(version) || 3702 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 3703 guid != vd->vdev_guid || 3704 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 3705 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3706 VDEV_AUX_CORRUPT_DATA); 3707 nvlist_free(label); 3708 return (-1); 3709 } 3710 3711 /* 3712 * We don't actually check the pool state here. If it's in fact in 3713 * use by another pool, we update this fact on the fly when requested. 3714 */ 3715 nvlist_free(label); 3716 return (0); 3717 } 3718 3719 static void 3720 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) 3721 { 3722 objset_t *mos = spa_meta_objset(vd->vdev_spa); 3723 3724 if (vd->vdev_top_zap == 0) 3725 return; 3726 3727 uint64_t object = 0; 3728 int err = zap_lookup(mos, vd->vdev_top_zap, 3729 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); 3730 if (err == ENOENT) 3731 return; 3732 VERIFY0(err); 3733 3734 VERIFY0(dmu_object_free(mos, object, tx)); 3735 VERIFY0(zap_remove(mos, vd->vdev_top_zap, 3736 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); 3737 } 3738 3739 /* 3740 * Free the objects used to store this vdev's spacemaps, and the array 3741 * that points to them. 3742 */ 3743 void 3744 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 3745 { 3746 if (vd->vdev_ms_array == 0) 3747 return; 3748 3749 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3750 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 3751 size_t array_bytes = array_count * sizeof (uint64_t); 3752 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 3753 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 3754 array_bytes, smobj_array, 0)); 3755 3756 for (uint64_t i = 0; i < array_count; i++) { 3757 uint64_t smobj = smobj_array[i]; 3758 if (smobj == 0) 3759 continue; 3760 3761 space_map_free_obj(mos, smobj, tx); 3762 } 3763 3764 kmem_free(smobj_array, array_bytes); 3765 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3766 vdev_destroy_ms_flush_data(vd, tx); 3767 vd->vdev_ms_array = 0; 3768 } 3769 3770 static void 3771 vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 3772 { 3773 spa_t *spa = vd->vdev_spa; 3774 3775 ASSERT(vd->vdev_islog); 3776 ASSERT(vd == vd->vdev_top); 3777 ASSERT3U(txg, ==, spa_syncing_txg(spa)); 3778 3779 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3780 3781 vdev_destroy_spacemaps(vd, tx); 3782 if (vd->vdev_top_zap != 0) { 3783 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3784 vd->vdev_top_zap = 0; 3785 } 3786 3787 dmu_tx_commit(tx); 3788 } 3789 3790 void 3791 vdev_sync_done(vdev_t *vd, uint64_t txg) 3792 { 3793 metaslab_t *msp; 3794 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3795 3796 ASSERT(vdev_is_concrete(vd)); 3797 3798 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3799 != NULL) 3800 metaslab_sync_done(msp, txg); 3801 3802 if (reassess) { 3803 metaslab_sync_reassess(vd->vdev_mg); 3804 if (vd->vdev_log_mg != NULL) 3805 metaslab_sync_reassess(vd->vdev_log_mg); 3806 } 3807 } 3808 3809 void 3810 vdev_sync(vdev_t *vd, uint64_t txg) 3811 { 3812 spa_t *spa = vd->vdev_spa; 3813 vdev_t *lvd; 3814 metaslab_t *msp; 3815 3816 ASSERT3U(txg, ==, spa->spa_syncing_txg); 3817 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3818 if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 3819 ASSERT(vd->vdev_removing || 3820 vd->vdev_ops == &vdev_indirect_ops); 3821 3822 vdev_indirect_sync_obsolete(vd, tx); 3823 3824 /* 3825 * If the vdev is indirect, it can't have dirty 3826 * metaslabs or DTLs. 3827 */ 3828 if (vd->vdev_ops == &vdev_indirect_ops) { 3829 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 3830 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3831 dmu_tx_commit(tx); 3832 return; 3833 } 3834 } 3835 3836 ASSERT(vdev_is_concrete(vd)); 3837 3838 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 3839 !vd->vdev_removing) { 3840 ASSERT(vd == vd->vdev_top); 3841 ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3842 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3843 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3844 ASSERT(vd->vdev_ms_array != 0); 3845 vdev_config_dirty(vd); 3846 } 3847 3848 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3849 metaslab_sync(msp, txg); 3850 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3851 } 3852 3853 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3854 vdev_dtl_sync(lvd, txg); 3855 3856 /* 3857 * If this is an empty log device being removed, destroy the 3858 * metadata associated with it. 3859 */ 3860 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 3861 vdev_remove_empty_log(vd, txg); 3862 3863 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3864 dmu_tx_commit(tx); 3865 } 3866 3867 uint64_t 3868 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3869 { 3870 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3871 } 3872 3873 /* 3874 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 3875 * not be opened, and no I/O is attempted. 3876 */ 3877 int 3878 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3879 { 3880 vdev_t *vd, *tvd; 3881 3882 spa_vdev_state_enter(spa, SCL_NONE); 3883 3884 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3885 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3886 3887 if (!vd->vdev_ops->vdev_op_leaf) 3888 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3889 3890 tvd = vd->vdev_top; 3891 3892 /* 3893 * If user did a 'zpool offline -f' then make the fault persist across 3894 * reboots. 3895 */ 3896 if (aux == VDEV_AUX_EXTERNAL_PERSIST) { 3897 /* 3898 * There are two kinds of forced faults: temporary and 3899 * persistent. Temporary faults go away at pool import, while 3900 * persistent faults stay set. Both types of faults can be 3901 * cleared with a zpool clear. 3902 * 3903 * We tell if a vdev is persistently faulted by looking at the 3904 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at 3905 * import then it's a persistent fault. Otherwise, it's 3906 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external" 3907 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This 3908 * tells vdev_config_generate() (which gets run later) to set 3909 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist. 3910 */ 3911 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; 3912 vd->vdev_tmpoffline = B_FALSE; 3913 aux = VDEV_AUX_EXTERNAL; 3914 } else { 3915 vd->vdev_tmpoffline = B_TRUE; 3916 } 3917 3918 /* 3919 * We don't directly use the aux state here, but if we do a 3920 * vdev_reopen(), we need this value to be present to remember why we 3921 * were faulted. 3922 */ 3923 vd->vdev_label_aux = aux; 3924 3925 /* 3926 * Faulted state takes precedence over degraded. 3927 */ 3928 vd->vdev_delayed_close = B_FALSE; 3929 vd->vdev_faulted = 1ULL; 3930 vd->vdev_degraded = 0ULL; 3931 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 3932 3933 /* 3934 * If this device has the only valid copy of the data, then 3935 * back off and simply mark the vdev as degraded instead. 3936 */ 3937 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 3938 vd->vdev_degraded = 1ULL; 3939 vd->vdev_faulted = 0ULL; 3940 3941 /* 3942 * If we reopen the device and it's not dead, only then do we 3943 * mark it degraded. 3944 */ 3945 vdev_reopen(tvd); 3946 3947 if (vdev_readable(vd)) 3948 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 3949 } 3950 3951 return (spa_vdev_state_exit(spa, vd, 0)); 3952 } 3953 3954 /* 3955 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 3956 * user that something is wrong. The vdev continues to operate as normal as far 3957 * as I/O is concerned. 3958 */ 3959 int 3960 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3961 { 3962 vdev_t *vd; 3963 3964 spa_vdev_state_enter(spa, SCL_NONE); 3965 3966 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3967 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3968 3969 if (!vd->vdev_ops->vdev_op_leaf) 3970 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3971 3972 /* 3973 * If the vdev is already faulted, then don't do anything. 3974 */ 3975 if (vd->vdev_faulted || vd->vdev_degraded) 3976 return (spa_vdev_state_exit(spa, NULL, 0)); 3977 3978 vd->vdev_degraded = 1ULL; 3979 if (!vdev_is_dead(vd)) 3980 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3981 aux); 3982 3983 return (spa_vdev_state_exit(spa, vd, 0)); 3984 } 3985 3986 int 3987 vdev_remove_wanted(spa_t *spa, uint64_t guid) 3988 { 3989 vdev_t *vd; 3990 3991 spa_vdev_state_enter(spa, SCL_NONE); 3992 3993 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3994 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3995 3996 /* 3997 * If the vdev is already removed, then don't do anything. 3998 */ 3999 if (vd->vdev_removed) 4000 return (spa_vdev_state_exit(spa, NULL, 0)); 4001 4002 vd->vdev_remove_wanted = B_TRUE; 4003 spa_async_request(spa, SPA_ASYNC_REMOVE); 4004 4005 return (spa_vdev_state_exit(spa, vd, 0)); 4006 } 4007 4008 4009 /* 4010 * Online the given vdev. 4011 * 4012 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 4013 * spare device should be detached when the device finishes resilvering. 4014 * Second, the online should be treated like a 'test' online case, so no FMA 4015 * events are generated if the device fails to open. 4016 */ 4017 int 4018 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 4019 { 4020 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 4021 boolean_t wasoffline; 4022 vdev_state_t oldstate; 4023 4024 spa_vdev_state_enter(spa, SCL_NONE); 4025 4026 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 4027 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 4028 4029 if (!vd->vdev_ops->vdev_op_leaf) 4030 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 4031 4032 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 4033 oldstate = vd->vdev_state; 4034 4035 tvd = vd->vdev_top; 4036 vd->vdev_offline = B_FALSE; 4037 vd->vdev_tmpoffline = B_FALSE; 4038 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 4039 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 4040 4041 /* XXX - L2ARC 1.0 does not support expansion */ 4042 if (!vd->vdev_aux) { 4043 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 4044 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) || 4045 spa->spa_autoexpand); 4046 vd->vdev_expansion_time = gethrestime_sec(); 4047 } 4048 4049 vdev_reopen(tvd); 4050 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 4051 4052 if (!vd->vdev_aux) { 4053 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 4054 pvd->vdev_expanding = B_FALSE; 4055 } 4056 4057 if (newstate) 4058 *newstate = vd->vdev_state; 4059 if ((flags & ZFS_ONLINE_UNSPARE) && 4060 !vdev_is_dead(vd) && vd->vdev_parent && 4061 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 4062 vd->vdev_parent->vdev_child[0] == vd) 4063 vd->vdev_unspare = B_TRUE; 4064 4065 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 4066 4067 /* XXX - L2ARC 1.0 does not support expansion */ 4068 if (vd->vdev_aux) 4069 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 4070 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 4071 } 4072 4073 /* Restart initializing if necessary */ 4074 mutex_enter(&vd->vdev_initialize_lock); 4075 if (vdev_writeable(vd) && 4076 vd->vdev_initialize_thread == NULL && 4077 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 4078 (void) vdev_initialize(vd); 4079 } 4080 mutex_exit(&vd->vdev_initialize_lock); 4081 4082 /* 4083 * Restart trimming if necessary. We do not restart trimming for cache 4084 * devices here. This is triggered by l2arc_rebuild_vdev() 4085 * asynchronously for the whole device or in l2arc_evict() as it evicts 4086 * space for upcoming writes. 4087 */ 4088 mutex_enter(&vd->vdev_trim_lock); 4089 if (vdev_writeable(vd) && !vd->vdev_isl2cache && 4090 vd->vdev_trim_thread == NULL && 4091 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 4092 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 4093 vd->vdev_trim_secure); 4094 } 4095 mutex_exit(&vd->vdev_trim_lock); 4096 4097 if (wasoffline || 4098 (oldstate < VDEV_STATE_DEGRADED && 4099 vd->vdev_state >= VDEV_STATE_DEGRADED)) 4100 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 4101 4102 return (spa_vdev_state_exit(spa, vd, 0)); 4103 } 4104 4105 static int 4106 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 4107 { 4108 vdev_t *vd, *tvd; 4109 int error = 0; 4110 uint64_t generation; 4111 metaslab_group_t *mg; 4112 4113 top: 4114 spa_vdev_state_enter(spa, SCL_ALLOC); 4115 4116 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 4117 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 4118 4119 if (!vd->vdev_ops->vdev_op_leaf) 4120 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 4121 4122 if (vd->vdev_ops == &vdev_draid_spare_ops) 4123 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 4124 4125 tvd = vd->vdev_top; 4126 mg = tvd->vdev_mg; 4127 generation = spa->spa_config_generation + 1; 4128 4129 /* 4130 * If the device isn't already offline, try to offline it. 4131 */ 4132 if (!vd->vdev_offline) { 4133 /* 4134 * If this device has the only valid copy of some data, 4135 * don't allow it to be offlined. Log devices are always 4136 * expendable. 4137 */ 4138 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 4139 vdev_dtl_required(vd)) 4140 return (spa_vdev_state_exit(spa, NULL, 4141 SET_ERROR(EBUSY))); 4142 4143 /* 4144 * If the top-level is a slog and it has had allocations 4145 * then proceed. We check that the vdev's metaslab group 4146 * is not NULL since it's possible that we may have just 4147 * added this vdev but not yet initialized its metaslabs. 4148 */ 4149 if (tvd->vdev_islog && mg != NULL) { 4150 /* 4151 * Prevent any future allocations. 4152 */ 4153 ASSERT3P(tvd->vdev_log_mg, ==, NULL); 4154 metaslab_group_passivate(mg); 4155 (void) spa_vdev_state_exit(spa, vd, 0); 4156 4157 error = spa_reset_logs(spa); 4158 4159 /* 4160 * If the log device was successfully reset but has 4161 * checkpointed data, do not offline it. 4162 */ 4163 if (error == 0 && 4164 tvd->vdev_checkpoint_sm != NULL) { 4165 ASSERT3U(space_map_allocated( 4166 tvd->vdev_checkpoint_sm), !=, 0); 4167 error = ZFS_ERR_CHECKPOINT_EXISTS; 4168 } 4169 4170 spa_vdev_state_enter(spa, SCL_ALLOC); 4171 4172 /* 4173 * Check to see if the config has changed. 4174 */ 4175 if (error || generation != spa->spa_config_generation) { 4176 metaslab_group_activate(mg); 4177 if (error) 4178 return (spa_vdev_state_exit(spa, 4179 vd, error)); 4180 (void) spa_vdev_state_exit(spa, vd, 0); 4181 goto top; 4182 } 4183 ASSERT0(tvd->vdev_stat.vs_alloc); 4184 } 4185 4186 /* 4187 * Offline this device and reopen its top-level vdev. 4188 * If the top-level vdev is a log device then just offline 4189 * it. Otherwise, if this action results in the top-level 4190 * vdev becoming unusable, undo it and fail the request. 4191 */ 4192 vd->vdev_offline = B_TRUE; 4193 vdev_reopen(tvd); 4194 4195 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 4196 vdev_is_dead(tvd)) { 4197 vd->vdev_offline = B_FALSE; 4198 vdev_reopen(tvd); 4199 return (spa_vdev_state_exit(spa, NULL, 4200 SET_ERROR(EBUSY))); 4201 } 4202 4203 /* 4204 * Add the device back into the metaslab rotor so that 4205 * once we online the device it's open for business. 4206 */ 4207 if (tvd->vdev_islog && mg != NULL) 4208 metaslab_group_activate(mg); 4209 } 4210 4211 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 4212 4213 return (spa_vdev_state_exit(spa, vd, 0)); 4214 } 4215 4216 int 4217 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 4218 { 4219 int error; 4220 4221 mutex_enter(&spa->spa_vdev_top_lock); 4222 error = vdev_offline_locked(spa, guid, flags); 4223 mutex_exit(&spa->spa_vdev_top_lock); 4224 4225 return (error); 4226 } 4227 4228 /* 4229 * Clear the error counts associated with this vdev. Unlike vdev_online() and 4230 * vdev_offline(), we assume the spa config is locked. We also clear all 4231 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 4232 */ 4233 void 4234 vdev_clear(spa_t *spa, vdev_t *vd) 4235 { 4236 vdev_t *rvd = spa->spa_root_vdev; 4237 4238 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 4239 4240 if (vd == NULL) 4241 vd = rvd; 4242 4243 vd->vdev_stat.vs_read_errors = 0; 4244 vd->vdev_stat.vs_write_errors = 0; 4245 vd->vdev_stat.vs_checksum_errors = 0; 4246 vd->vdev_stat.vs_slow_ios = 0; 4247 4248 for (int c = 0; c < vd->vdev_children; c++) 4249 vdev_clear(spa, vd->vdev_child[c]); 4250 4251 /* 4252 * It makes no sense to "clear" an indirect vdev. 4253 */ 4254 if (!vdev_is_concrete(vd)) 4255 return; 4256 4257 /* 4258 * If we're in the FAULTED state or have experienced failed I/O, then 4259 * clear the persistent state and attempt to reopen the device. We 4260 * also mark the vdev config dirty, so that the new faulted state is 4261 * written out to disk. 4262 */ 4263 if (vd->vdev_faulted || vd->vdev_degraded || 4264 !vdev_readable(vd) || !vdev_writeable(vd)) { 4265 /* 4266 * When reopening in response to a clear event, it may be due to 4267 * a fmadm repair request. In this case, if the device is 4268 * still broken, we want to still post the ereport again. 4269 */ 4270 vd->vdev_forcefault = B_TRUE; 4271 4272 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 4273 vd->vdev_cant_read = B_FALSE; 4274 vd->vdev_cant_write = B_FALSE; 4275 vd->vdev_stat.vs_aux = 0; 4276 4277 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 4278 4279 vd->vdev_forcefault = B_FALSE; 4280 4281 if (vd != rvd && vdev_writeable(vd->vdev_top)) 4282 vdev_state_dirty(vd->vdev_top); 4283 4284 /* If a resilver isn't required, check if vdevs can be culled */ 4285 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && 4286 !dsl_scan_resilvering(spa->spa_dsl_pool) && 4287 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) 4288 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 4289 4290 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 4291 } 4292 4293 /* 4294 * When clearing a FMA-diagnosed fault, we always want to 4295 * unspare the device, as we assume that the original spare was 4296 * done in response to the FMA fault. 4297 */ 4298 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 4299 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 4300 vd->vdev_parent->vdev_child[0] == vd) 4301 vd->vdev_unspare = B_TRUE; 4302 4303 /* Clear recent error events cache (i.e. duplicate events tracking) */ 4304 zfs_ereport_clear(spa, vd); 4305 } 4306 4307 boolean_t 4308 vdev_is_dead(vdev_t *vd) 4309 { 4310 /* 4311 * Holes and missing devices are always considered "dead". 4312 * This simplifies the code since we don't have to check for 4313 * these types of devices in the various code paths. 4314 * Instead we rely on the fact that we skip over dead devices 4315 * before issuing I/O to them. 4316 */ 4317 return (vd->vdev_state < VDEV_STATE_DEGRADED || 4318 vd->vdev_ops == &vdev_hole_ops || 4319 vd->vdev_ops == &vdev_missing_ops); 4320 } 4321 4322 boolean_t 4323 vdev_readable(vdev_t *vd) 4324 { 4325 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 4326 } 4327 4328 boolean_t 4329 vdev_writeable(vdev_t *vd) 4330 { 4331 return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 4332 vdev_is_concrete(vd)); 4333 } 4334 4335 boolean_t 4336 vdev_allocatable(vdev_t *vd) 4337 { 4338 uint64_t state = vd->vdev_state; 4339 4340 /* 4341 * We currently allow allocations from vdevs which may be in the 4342 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 4343 * fails to reopen then we'll catch it later when we're holding 4344 * the proper locks. Note that we have to get the vdev state 4345 * in a local variable because although it changes atomically, 4346 * we're asking two separate questions about it. 4347 */ 4348 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 4349 !vd->vdev_cant_write && vdev_is_concrete(vd) && 4350 vd->vdev_mg->mg_initialized); 4351 } 4352 4353 boolean_t 4354 vdev_accessible(vdev_t *vd, zio_t *zio) 4355 { 4356 ASSERT(zio->io_vd == vd); 4357 4358 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 4359 return (B_FALSE); 4360 4361 if (zio->io_type == ZIO_TYPE_READ) 4362 return (!vd->vdev_cant_read); 4363 4364 if (zio->io_type == ZIO_TYPE_WRITE) 4365 return (!vd->vdev_cant_write); 4366 4367 return (B_TRUE); 4368 } 4369 4370 static void 4371 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) 4372 { 4373 /* 4374 * Exclude the dRAID spare when aggregating to avoid double counting 4375 * the ops and bytes. These IOs are counted by the physical leaves. 4376 */ 4377 if (cvd->vdev_ops == &vdev_draid_spare_ops) 4378 return; 4379 4380 for (int t = 0; t < VS_ZIO_TYPES; t++) { 4381 vs->vs_ops[t] += cvs->vs_ops[t]; 4382 vs->vs_bytes[t] += cvs->vs_bytes[t]; 4383 } 4384 4385 cvs->vs_scan_removing = cvd->vdev_removing; 4386 } 4387 4388 /* 4389 * Get extended stats 4390 */ 4391 static void 4392 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) 4393 { 4394 (void) cvd; 4395 4396 int t, b; 4397 for (t = 0; t < ZIO_TYPES; t++) { 4398 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) 4399 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; 4400 4401 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { 4402 vsx->vsx_total_histo[t][b] += 4403 cvsx->vsx_total_histo[t][b]; 4404 } 4405 } 4406 4407 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { 4408 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { 4409 vsx->vsx_queue_histo[t][b] += 4410 cvsx->vsx_queue_histo[t][b]; 4411 } 4412 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; 4413 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; 4414 4415 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) 4416 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; 4417 4418 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) 4419 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; 4420 } 4421 4422 } 4423 4424 boolean_t 4425 vdev_is_spacemap_addressable(vdev_t *vd) 4426 { 4427 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 4428 return (B_TRUE); 4429 4430 /* 4431 * If double-word space map entries are not enabled we assume 4432 * 47 bits of the space map entry are dedicated to the entry's 4433 * offset (see SM_OFFSET_BITS in space_map.h). We then use that 4434 * to calculate the maximum address that can be described by a 4435 * space map entry for the given device. 4436 */ 4437 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 4438 4439 if (shift >= 63) /* detect potential overflow */ 4440 return (B_TRUE); 4441 4442 return (vd->vdev_asize < (1ULL << shift)); 4443 } 4444 4445 /* 4446 * Get statistics for the given vdev. 4447 */ 4448 static void 4449 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 4450 { 4451 int t; 4452 /* 4453 * If we're getting stats on the root vdev, aggregate the I/O counts 4454 * over all top-level vdevs (i.e. the direct children of the root). 4455 */ 4456 if (!vd->vdev_ops->vdev_op_leaf) { 4457 if (vs) { 4458 memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); 4459 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); 4460 } 4461 if (vsx) 4462 memset(vsx, 0, sizeof (*vsx)); 4463 4464 for (int c = 0; c < vd->vdev_children; c++) { 4465 vdev_t *cvd = vd->vdev_child[c]; 4466 vdev_stat_t *cvs = &cvd->vdev_stat; 4467 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; 4468 4469 vdev_get_stats_ex_impl(cvd, cvs, cvsx); 4470 if (vs) 4471 vdev_get_child_stat(cvd, vs, cvs); 4472 if (vsx) 4473 vdev_get_child_stat_ex(cvd, vsx, cvsx); 4474 } 4475 } else { 4476 /* 4477 * We're a leaf. Just copy our ZIO active queue stats in. The 4478 * other leaf stats are updated in vdev_stat_update(). 4479 */ 4480 if (!vsx) 4481 return; 4482 4483 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); 4484 4485 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { 4486 vsx->vsx_active_queue[t] = 4487 vd->vdev_queue.vq_class[t].vqc_active; 4488 vsx->vsx_pend_queue[t] = avl_numnodes( 4489 &vd->vdev_queue.vq_class[t].vqc_queued_tree); 4490 } 4491 } 4492 } 4493 4494 void 4495 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 4496 { 4497 vdev_t *tvd = vd->vdev_top; 4498 mutex_enter(&vd->vdev_stat_lock); 4499 if (vs) { 4500 memcpy(vs, &vd->vdev_stat, sizeof (*vs)); 4501 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 4502 vs->vs_state = vd->vdev_state; 4503 vs->vs_rsize = vdev_get_min_asize(vd); 4504 4505 if (vd->vdev_ops->vdev_op_leaf) { 4506 vs->vs_pspace = vd->vdev_psize; 4507 vs->vs_rsize += VDEV_LABEL_START_SIZE + 4508 VDEV_LABEL_END_SIZE; 4509 /* 4510 * Report initializing progress. Since we don't 4511 * have the initializing locks held, this is only 4512 * an estimate (although a fairly accurate one). 4513 */ 4514 vs->vs_initialize_bytes_done = 4515 vd->vdev_initialize_bytes_done; 4516 vs->vs_initialize_bytes_est = 4517 vd->vdev_initialize_bytes_est; 4518 vs->vs_initialize_state = vd->vdev_initialize_state; 4519 vs->vs_initialize_action_time = 4520 vd->vdev_initialize_action_time; 4521 4522 /* 4523 * Report manual TRIM progress. Since we don't have 4524 * the manual TRIM locks held, this is only an 4525 * estimate (although fairly accurate one). 4526 */ 4527 vs->vs_trim_notsup = !vd->vdev_has_trim; 4528 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 4529 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 4530 vs->vs_trim_state = vd->vdev_trim_state; 4531 vs->vs_trim_action_time = vd->vdev_trim_action_time; 4532 4533 /* Set when there is a deferred resilver. */ 4534 vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 4535 } 4536 4537 /* 4538 * Report expandable space on top-level, non-auxiliary devices 4539 * only. The expandable space is reported in terms of metaslab 4540 * sized units since that determines how much space the pool 4541 * can expand. 4542 */ 4543 if (vd->vdev_aux == NULL && tvd != NULL) { 4544 vs->vs_esize = P2ALIGN( 4545 vd->vdev_max_asize - vd->vdev_asize, 4546 1ULL << tvd->vdev_ms_shift); 4547 } 4548 4549 vs->vs_configured_ashift = vd->vdev_top != NULL 4550 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift; 4551 vs->vs_logical_ashift = vd->vdev_logical_ashift; 4552 if (vd->vdev_physical_ashift <= ASHIFT_MAX) 4553 vs->vs_physical_ashift = vd->vdev_physical_ashift; 4554 else 4555 vs->vs_physical_ashift = 0; 4556 4557 /* 4558 * Report fragmentation and rebuild progress for top-level, 4559 * non-auxiliary, concrete devices. 4560 */ 4561 if (vd->vdev_aux == NULL && vd == vd->vdev_top && 4562 vdev_is_concrete(vd)) { 4563 /* 4564 * The vdev fragmentation rating doesn't take into 4565 * account the embedded slog metaslab (vdev_log_mg). 4566 * Since it's only one metaslab, it would have a tiny 4567 * impact on the overall fragmentation. 4568 */ 4569 vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 4570 vd->vdev_mg->mg_fragmentation : 0; 4571 } 4572 vs->vs_noalloc = MAX(vd->vdev_noalloc, 4573 tvd ? tvd->vdev_noalloc : 0); 4574 } 4575 4576 vdev_get_stats_ex_impl(vd, vs, vsx); 4577 mutex_exit(&vd->vdev_stat_lock); 4578 } 4579 4580 void 4581 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 4582 { 4583 return (vdev_get_stats_ex(vd, vs, NULL)); 4584 } 4585 4586 void 4587 vdev_clear_stats(vdev_t *vd) 4588 { 4589 mutex_enter(&vd->vdev_stat_lock); 4590 vd->vdev_stat.vs_space = 0; 4591 vd->vdev_stat.vs_dspace = 0; 4592 vd->vdev_stat.vs_alloc = 0; 4593 mutex_exit(&vd->vdev_stat_lock); 4594 } 4595 4596 void 4597 vdev_scan_stat_init(vdev_t *vd) 4598 { 4599 vdev_stat_t *vs = &vd->vdev_stat; 4600 4601 for (int c = 0; c < vd->vdev_children; c++) 4602 vdev_scan_stat_init(vd->vdev_child[c]); 4603 4604 mutex_enter(&vd->vdev_stat_lock); 4605 vs->vs_scan_processed = 0; 4606 mutex_exit(&vd->vdev_stat_lock); 4607 } 4608 4609 void 4610 vdev_stat_update(zio_t *zio, uint64_t psize) 4611 { 4612 spa_t *spa = zio->io_spa; 4613 vdev_t *rvd = spa->spa_root_vdev; 4614 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 4615 vdev_t *pvd; 4616 uint64_t txg = zio->io_txg; 4617 vdev_stat_t *vs = vd ? &vd->vdev_stat : NULL; 4618 vdev_stat_ex_t *vsx = vd ? &vd->vdev_stat_ex : NULL; 4619 zio_type_t type = zio->io_type; 4620 int flags = zio->io_flags; 4621 4622 /* 4623 * If this i/o is a gang leader, it didn't do any actual work. 4624 */ 4625 if (zio->io_gang_tree) 4626 return; 4627 4628 if (zio->io_error == 0) { 4629 /* 4630 * If this is a root i/o, don't count it -- we've already 4631 * counted the top-level vdevs, and vdev_get_stats() will 4632 * aggregate them when asked. This reduces contention on 4633 * the root vdev_stat_lock and implicitly handles blocks 4634 * that compress away to holes, for which there is no i/o. 4635 * (Holes never create vdev children, so all the counters 4636 * remain zero, which is what we want.) 4637 * 4638 * Note: this only applies to successful i/o (io_error == 0) 4639 * because unlike i/o counts, errors are not additive. 4640 * When reading a ditto block, for example, failure of 4641 * one top-level vdev does not imply a root-level error. 4642 */ 4643 if (vd == rvd) 4644 return; 4645 4646 ASSERT(vd == zio->io_vd); 4647 4648 if (flags & ZIO_FLAG_IO_BYPASS) 4649 return; 4650 4651 mutex_enter(&vd->vdev_stat_lock); 4652 4653 if (flags & ZIO_FLAG_IO_REPAIR) { 4654 /* 4655 * Repair is the result of a resilver issued by the 4656 * scan thread (spa_sync). 4657 */ 4658 if (flags & ZIO_FLAG_SCAN_THREAD) { 4659 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 4660 dsl_scan_phys_t *scn_phys = &scn->scn_phys; 4661 uint64_t *processed = &scn_phys->scn_processed; 4662 4663 if (vd->vdev_ops->vdev_op_leaf) 4664 atomic_add_64(processed, psize); 4665 vs->vs_scan_processed += psize; 4666 } 4667 4668 /* 4669 * Repair is the result of a rebuild issued by the 4670 * rebuild thread (vdev_rebuild_thread). To avoid 4671 * double counting repaired bytes the virtual dRAID 4672 * spare vdev is excluded from the processed bytes. 4673 */ 4674 if (zio->io_priority == ZIO_PRIORITY_REBUILD) { 4675 vdev_t *tvd = vd->vdev_top; 4676 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config; 4677 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; 4678 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt; 4679 4680 if (vd->vdev_ops->vdev_op_leaf && 4681 vd->vdev_ops != &vdev_draid_spare_ops) { 4682 atomic_add_64(rebuilt, psize); 4683 } 4684 vs->vs_rebuild_processed += psize; 4685 } 4686 4687 if (flags & ZIO_FLAG_SELF_HEAL) 4688 vs->vs_self_healed += psize; 4689 } 4690 4691 /* 4692 * The bytes/ops/histograms are recorded at the leaf level and 4693 * aggregated into the higher level vdevs in vdev_get_stats(). 4694 */ 4695 if (vd->vdev_ops->vdev_op_leaf && 4696 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { 4697 zio_type_t vs_type = type; 4698 zio_priority_t priority = zio->io_priority; 4699 4700 /* 4701 * TRIM ops and bytes are reported to user space as 4702 * ZIO_TYPE_IOCTL. This is done to preserve the 4703 * vdev_stat_t structure layout for user space. 4704 */ 4705 if (type == ZIO_TYPE_TRIM) 4706 vs_type = ZIO_TYPE_IOCTL; 4707 4708 /* 4709 * Solely for the purposes of 'zpool iostat -lqrw' 4710 * reporting use the priority to categorize the IO. 4711 * Only the following are reported to user space: 4712 * 4713 * ZIO_PRIORITY_SYNC_READ, 4714 * ZIO_PRIORITY_SYNC_WRITE, 4715 * ZIO_PRIORITY_ASYNC_READ, 4716 * ZIO_PRIORITY_ASYNC_WRITE, 4717 * ZIO_PRIORITY_SCRUB, 4718 * ZIO_PRIORITY_TRIM, 4719 * ZIO_PRIORITY_REBUILD. 4720 */ 4721 if (priority == ZIO_PRIORITY_INITIALIZING) { 4722 ASSERT3U(type, ==, ZIO_TYPE_WRITE); 4723 priority = ZIO_PRIORITY_ASYNC_WRITE; 4724 } else if (priority == ZIO_PRIORITY_REMOVAL) { 4725 priority = ((type == ZIO_TYPE_WRITE) ? 4726 ZIO_PRIORITY_ASYNC_WRITE : 4727 ZIO_PRIORITY_ASYNC_READ); 4728 } 4729 4730 vs->vs_ops[vs_type]++; 4731 vs->vs_bytes[vs_type] += psize; 4732 4733 if (flags & ZIO_FLAG_DELEGATED) { 4734 vsx->vsx_agg_histo[priority] 4735 [RQ_HISTO(zio->io_size)]++; 4736 } else { 4737 vsx->vsx_ind_histo[priority] 4738 [RQ_HISTO(zio->io_size)]++; 4739 } 4740 4741 if (zio->io_delta && zio->io_delay) { 4742 vsx->vsx_queue_histo[priority] 4743 [L_HISTO(zio->io_delta - zio->io_delay)]++; 4744 vsx->vsx_disk_histo[type] 4745 [L_HISTO(zio->io_delay)]++; 4746 vsx->vsx_total_histo[type] 4747 [L_HISTO(zio->io_delta)]++; 4748 } 4749 } 4750 4751 mutex_exit(&vd->vdev_stat_lock); 4752 return; 4753 } 4754 4755 if (flags & ZIO_FLAG_SPECULATIVE) 4756 return; 4757 4758 /* 4759 * If this is an I/O error that is going to be retried, then ignore the 4760 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 4761 * hard errors, when in reality they can happen for any number of 4762 * innocuous reasons (bus resets, MPxIO link failure, etc). 4763 */ 4764 if (zio->io_error == EIO && 4765 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 4766 return; 4767 4768 /* 4769 * Intent logs writes won't propagate their error to the root 4770 * I/O so don't mark these types of failures as pool-level 4771 * errors. 4772 */ 4773 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 4774 return; 4775 4776 if (type == ZIO_TYPE_WRITE && txg != 0 && 4777 (!(flags & ZIO_FLAG_IO_REPAIR) || 4778 (flags & ZIO_FLAG_SCAN_THREAD) || 4779 spa->spa_claiming)) { 4780 /* 4781 * This is either a normal write (not a repair), or it's 4782 * a repair induced by the scrub thread, or it's a repair 4783 * made by zil_claim() during spa_load() in the first txg. 4784 * In the normal case, we commit the DTL change in the same 4785 * txg as the block was born. In the scrub-induced repair 4786 * case, we know that scrubs run in first-pass syncing context, 4787 * so we commit the DTL change in spa_syncing_txg(spa). 4788 * In the zil_claim() case, we commit in spa_first_txg(spa). 4789 * 4790 * We currently do not make DTL entries for failed spontaneous 4791 * self-healing writes triggered by normal (non-scrubbing) 4792 * reads, because we have no transactional context in which to 4793 * do so -- and it's not clear that it'd be desirable anyway. 4794 */ 4795 if (vd->vdev_ops->vdev_op_leaf) { 4796 uint64_t commit_txg = txg; 4797 if (flags & ZIO_FLAG_SCAN_THREAD) { 4798 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 4799 ASSERT(spa_sync_pass(spa) == 1); 4800 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 4801 commit_txg = spa_syncing_txg(spa); 4802 } else if (spa->spa_claiming) { 4803 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 4804 commit_txg = spa_first_txg(spa); 4805 } 4806 ASSERT(commit_txg >= spa_syncing_txg(spa)); 4807 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 4808 return; 4809 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 4810 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 4811 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 4812 } 4813 if (vd != rvd) 4814 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 4815 } 4816 } 4817 4818 int64_t 4819 vdev_deflated_space(vdev_t *vd, int64_t space) 4820 { 4821 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 4822 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 4823 4824 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 4825 } 4826 4827 /* 4828 * Update the in-core space usage stats for this vdev, its metaslab class, 4829 * and the root vdev. 4830 */ 4831 void 4832 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 4833 int64_t space_delta) 4834 { 4835 (void) defer_delta; 4836 int64_t dspace_delta; 4837 spa_t *spa = vd->vdev_spa; 4838 vdev_t *rvd = spa->spa_root_vdev; 4839 4840 ASSERT(vd == vd->vdev_top); 4841 4842 /* 4843 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 4844 * factor. We must calculate this here and not at the root vdev 4845 * because the root vdev's psize-to-asize is simply the max of its 4846 * children's, thus not accurate enough for us. 4847 */ 4848 dspace_delta = vdev_deflated_space(vd, space_delta); 4849 4850 mutex_enter(&vd->vdev_stat_lock); 4851 /* ensure we won't underflow */ 4852 if (alloc_delta < 0) { 4853 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 4854 } 4855 4856 vd->vdev_stat.vs_alloc += alloc_delta; 4857 vd->vdev_stat.vs_space += space_delta; 4858 vd->vdev_stat.vs_dspace += dspace_delta; 4859 mutex_exit(&vd->vdev_stat_lock); 4860 4861 /* every class but log contributes to root space stats */ 4862 if (vd->vdev_mg != NULL && !vd->vdev_islog) { 4863 ASSERT(!vd->vdev_isl2cache); 4864 mutex_enter(&rvd->vdev_stat_lock); 4865 rvd->vdev_stat.vs_alloc += alloc_delta; 4866 rvd->vdev_stat.vs_space += space_delta; 4867 rvd->vdev_stat.vs_dspace += dspace_delta; 4868 mutex_exit(&rvd->vdev_stat_lock); 4869 } 4870 /* Note: metaslab_class_space_update moved to metaslab_space_update */ 4871 } 4872 4873 /* 4874 * Mark a top-level vdev's config as dirty, placing it on the dirty list 4875 * so that it will be written out next time the vdev configuration is synced. 4876 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 4877 */ 4878 void 4879 vdev_config_dirty(vdev_t *vd) 4880 { 4881 spa_t *spa = vd->vdev_spa; 4882 vdev_t *rvd = spa->spa_root_vdev; 4883 int c; 4884 4885 ASSERT(spa_writeable(spa)); 4886 4887 /* 4888 * If this is an aux vdev (as with l2cache and spare devices), then we 4889 * update the vdev config manually and set the sync flag. 4890 */ 4891 if (vd->vdev_aux != NULL) { 4892 spa_aux_vdev_t *sav = vd->vdev_aux; 4893 nvlist_t **aux; 4894 uint_t naux; 4895 4896 for (c = 0; c < sav->sav_count; c++) { 4897 if (sav->sav_vdevs[c] == vd) 4898 break; 4899 } 4900 4901 if (c == sav->sav_count) { 4902 /* 4903 * We're being removed. There's nothing more to do. 4904 */ 4905 ASSERT(sav->sav_sync == B_TRUE); 4906 return; 4907 } 4908 4909 sav->sav_sync = B_TRUE; 4910 4911 if (nvlist_lookup_nvlist_array(sav->sav_config, 4912 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 4913 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 4914 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 4915 } 4916 4917 ASSERT(c < naux); 4918 4919 /* 4920 * Setting the nvlist in the middle if the array is a little 4921 * sketchy, but it will work. 4922 */ 4923 nvlist_free(aux[c]); 4924 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 4925 4926 return; 4927 } 4928 4929 /* 4930 * The dirty list is protected by the SCL_CONFIG lock. The caller 4931 * must either hold SCL_CONFIG as writer, or must be the sync thread 4932 * (which holds SCL_CONFIG as reader). There's only one sync thread, 4933 * so this is sufficient to ensure mutual exclusion. 4934 */ 4935 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4936 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4937 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4938 4939 if (vd == rvd) { 4940 for (c = 0; c < rvd->vdev_children; c++) 4941 vdev_config_dirty(rvd->vdev_child[c]); 4942 } else { 4943 ASSERT(vd == vd->vdev_top); 4944 4945 if (!list_link_active(&vd->vdev_config_dirty_node) && 4946 vdev_is_concrete(vd)) { 4947 list_insert_head(&spa->spa_config_dirty_list, vd); 4948 } 4949 } 4950 } 4951 4952 void 4953 vdev_config_clean(vdev_t *vd) 4954 { 4955 spa_t *spa = vd->vdev_spa; 4956 4957 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4958 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4959 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4960 4961 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4962 list_remove(&spa->spa_config_dirty_list, vd); 4963 } 4964 4965 /* 4966 * Mark a top-level vdev's state as dirty, so that the next pass of 4967 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4968 * the state changes from larger config changes because they require 4969 * much less locking, and are often needed for administrative actions. 4970 */ 4971 void 4972 vdev_state_dirty(vdev_t *vd) 4973 { 4974 spa_t *spa = vd->vdev_spa; 4975 4976 ASSERT(spa_writeable(spa)); 4977 ASSERT(vd == vd->vdev_top); 4978 4979 /* 4980 * The state list is protected by the SCL_STATE lock. The caller 4981 * must either hold SCL_STATE as writer, or must be the sync thread 4982 * (which holds SCL_STATE as reader). There's only one sync thread, 4983 * so this is sufficient to ensure mutual exclusion. 4984 */ 4985 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4986 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4987 spa_config_held(spa, SCL_STATE, RW_READER))); 4988 4989 if (!list_link_active(&vd->vdev_state_dirty_node) && 4990 vdev_is_concrete(vd)) 4991 list_insert_head(&spa->spa_state_dirty_list, vd); 4992 } 4993 4994 void 4995 vdev_state_clean(vdev_t *vd) 4996 { 4997 spa_t *spa = vd->vdev_spa; 4998 4999 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 5000 (dsl_pool_sync_context(spa_get_dsl(spa)) && 5001 spa_config_held(spa, SCL_STATE, RW_READER))); 5002 5003 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 5004 list_remove(&spa->spa_state_dirty_list, vd); 5005 } 5006 5007 /* 5008 * Propagate vdev state up from children to parent. 5009 */ 5010 void 5011 vdev_propagate_state(vdev_t *vd) 5012 { 5013 spa_t *spa = vd->vdev_spa; 5014 vdev_t *rvd = spa->spa_root_vdev; 5015 int degraded = 0, faulted = 0; 5016 int corrupted = 0; 5017 vdev_t *child; 5018 5019 if (vd->vdev_children > 0) { 5020 for (int c = 0; c < vd->vdev_children; c++) { 5021 child = vd->vdev_child[c]; 5022 5023 /* 5024 * Don't factor holes or indirect vdevs into the 5025 * decision. 5026 */ 5027 if (!vdev_is_concrete(child)) 5028 continue; 5029 5030 if (!vdev_readable(child) || 5031 (!vdev_writeable(child) && spa_writeable(spa))) { 5032 /* 5033 * Root special: if there is a top-level log 5034 * device, treat the root vdev as if it were 5035 * degraded. 5036 */ 5037 if (child->vdev_islog && vd == rvd) 5038 degraded++; 5039 else 5040 faulted++; 5041 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 5042 degraded++; 5043 } 5044 5045 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 5046 corrupted++; 5047 } 5048 5049 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 5050 5051 /* 5052 * Root special: if there is a top-level vdev that cannot be 5053 * opened due to corrupted metadata, then propagate the root 5054 * vdev's aux state as 'corrupt' rather than 'insufficient 5055 * replicas'. 5056 */ 5057 if (corrupted && vd == rvd && 5058 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 5059 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 5060 VDEV_AUX_CORRUPT_DATA); 5061 } 5062 5063 if (vd->vdev_parent) 5064 vdev_propagate_state(vd->vdev_parent); 5065 } 5066 5067 /* 5068 * Set a vdev's state. If this is during an open, we don't update the parent 5069 * state, because we're in the process of opening children depth-first. 5070 * Otherwise, we propagate the change to the parent. 5071 * 5072 * If this routine places a device in a faulted state, an appropriate ereport is 5073 * generated. 5074 */ 5075 void 5076 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 5077 { 5078 uint64_t save_state; 5079 spa_t *spa = vd->vdev_spa; 5080 5081 if (state == vd->vdev_state) { 5082 /* 5083 * Since vdev_offline() code path is already in an offline 5084 * state we can miss a statechange event to OFFLINE. Check 5085 * the previous state to catch this condition. 5086 */ 5087 if (vd->vdev_ops->vdev_op_leaf && 5088 (state == VDEV_STATE_OFFLINE) && 5089 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { 5090 /* post an offline state change */ 5091 zfs_post_state_change(spa, vd, vd->vdev_prevstate); 5092 } 5093 vd->vdev_stat.vs_aux = aux; 5094 return; 5095 } 5096 5097 save_state = vd->vdev_state; 5098 5099 vd->vdev_state = state; 5100 vd->vdev_stat.vs_aux = aux; 5101 5102 /* 5103 * If we are setting the vdev state to anything but an open state, then 5104 * always close the underlying device unless the device has requested 5105 * a delayed close (i.e. we're about to remove or fault the device). 5106 * Otherwise, we keep accessible but invalid devices open forever. 5107 * We don't call vdev_close() itself, because that implies some extra 5108 * checks (offline, etc) that we don't want here. This is limited to 5109 * leaf devices, because otherwise closing the device will affect other 5110 * children. 5111 */ 5112 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 5113 vd->vdev_ops->vdev_op_leaf) 5114 vd->vdev_ops->vdev_op_close(vd); 5115 5116 if (vd->vdev_removed && 5117 state == VDEV_STATE_CANT_OPEN && 5118 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 5119 /* 5120 * If the previous state is set to VDEV_STATE_REMOVED, then this 5121 * device was previously marked removed and someone attempted to 5122 * reopen it. If this failed due to a nonexistent device, then 5123 * keep the device in the REMOVED state. We also let this be if 5124 * it is one of our special test online cases, which is only 5125 * attempting to online the device and shouldn't generate an FMA 5126 * fault. 5127 */ 5128 vd->vdev_state = VDEV_STATE_REMOVED; 5129 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 5130 } else if (state == VDEV_STATE_REMOVED) { 5131 vd->vdev_removed = B_TRUE; 5132 } else if (state == VDEV_STATE_CANT_OPEN) { 5133 /* 5134 * If we fail to open a vdev during an import or recovery, we 5135 * mark it as "not available", which signifies that it was 5136 * never there to begin with. Failure to open such a device 5137 * is not considered an error. 5138 */ 5139 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 5140 spa_load_state(spa) == SPA_LOAD_RECOVER) && 5141 vd->vdev_ops->vdev_op_leaf) 5142 vd->vdev_not_present = 1; 5143 5144 /* 5145 * Post the appropriate ereport. If the 'prevstate' field is 5146 * set to something other than VDEV_STATE_UNKNOWN, it indicates 5147 * that this is part of a vdev_reopen(). In this case, we don't 5148 * want to post the ereport if the device was already in the 5149 * CANT_OPEN state beforehand. 5150 * 5151 * If the 'checkremove' flag is set, then this is an attempt to 5152 * online the device in response to an insertion event. If we 5153 * hit this case, then we have detected an insertion event for a 5154 * faulted or offline device that wasn't in the removed state. 5155 * In this scenario, we don't post an ereport because we are 5156 * about to replace the device, or attempt an online with 5157 * vdev_forcefault, which will generate the fault for us. 5158 */ 5159 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 5160 !vd->vdev_not_present && !vd->vdev_checkremove && 5161 vd != spa->spa_root_vdev) { 5162 const char *class; 5163 5164 switch (aux) { 5165 case VDEV_AUX_OPEN_FAILED: 5166 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 5167 break; 5168 case VDEV_AUX_CORRUPT_DATA: 5169 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 5170 break; 5171 case VDEV_AUX_NO_REPLICAS: 5172 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 5173 break; 5174 case VDEV_AUX_BAD_GUID_SUM: 5175 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 5176 break; 5177 case VDEV_AUX_TOO_SMALL: 5178 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 5179 break; 5180 case VDEV_AUX_BAD_LABEL: 5181 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 5182 break; 5183 case VDEV_AUX_BAD_ASHIFT: 5184 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 5185 break; 5186 default: 5187 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 5188 } 5189 5190 (void) zfs_ereport_post(class, spa, vd, NULL, NULL, 5191 save_state); 5192 } 5193 5194 /* Erase any notion of persistent removed state */ 5195 vd->vdev_removed = B_FALSE; 5196 } else { 5197 vd->vdev_removed = B_FALSE; 5198 } 5199 5200 /* 5201 * Notify ZED of any significant state-change on a leaf vdev. 5202 * 5203 */ 5204 if (vd->vdev_ops->vdev_op_leaf) { 5205 /* preserve original state from a vdev_reopen() */ 5206 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && 5207 (vd->vdev_prevstate != vd->vdev_state) && 5208 (save_state <= VDEV_STATE_CLOSED)) 5209 save_state = vd->vdev_prevstate; 5210 5211 /* filter out state change due to initial vdev_open */ 5212 if (save_state > VDEV_STATE_CLOSED) 5213 zfs_post_state_change(spa, vd, save_state); 5214 } 5215 5216 if (!isopen && vd->vdev_parent) 5217 vdev_propagate_state(vd->vdev_parent); 5218 } 5219 5220 boolean_t 5221 vdev_children_are_offline(vdev_t *vd) 5222 { 5223 ASSERT(!vd->vdev_ops->vdev_op_leaf); 5224 5225 for (uint64_t i = 0; i < vd->vdev_children; i++) { 5226 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 5227 return (B_FALSE); 5228 } 5229 5230 return (B_TRUE); 5231 } 5232 5233 /* 5234 * Check the vdev configuration to ensure that it's capable of supporting 5235 * a root pool. We do not support partial configuration. 5236 */ 5237 boolean_t 5238 vdev_is_bootable(vdev_t *vd) 5239 { 5240 if (!vd->vdev_ops->vdev_op_leaf) { 5241 const char *vdev_type = vd->vdev_ops->vdev_op_type; 5242 5243 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) 5244 return (B_FALSE); 5245 } 5246 5247 for (int c = 0; c < vd->vdev_children; c++) { 5248 if (!vdev_is_bootable(vd->vdev_child[c])) 5249 return (B_FALSE); 5250 } 5251 return (B_TRUE); 5252 } 5253 5254 boolean_t 5255 vdev_is_concrete(vdev_t *vd) 5256 { 5257 vdev_ops_t *ops = vd->vdev_ops; 5258 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 5259 ops == &vdev_missing_ops || ops == &vdev_root_ops) { 5260 return (B_FALSE); 5261 } else { 5262 return (B_TRUE); 5263 } 5264 } 5265 5266 /* 5267 * Determine if a log device has valid content. If the vdev was 5268 * removed or faulted in the MOS config then we know that 5269 * the content on the log device has already been written to the pool. 5270 */ 5271 boolean_t 5272 vdev_log_state_valid(vdev_t *vd) 5273 { 5274 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 5275 !vd->vdev_removed) 5276 return (B_TRUE); 5277 5278 for (int c = 0; c < vd->vdev_children; c++) 5279 if (vdev_log_state_valid(vd->vdev_child[c])) 5280 return (B_TRUE); 5281 5282 return (B_FALSE); 5283 } 5284 5285 /* 5286 * Expand a vdev if possible. 5287 */ 5288 void 5289 vdev_expand(vdev_t *vd, uint64_t txg) 5290 { 5291 ASSERT(vd->vdev_top == vd); 5292 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5293 ASSERT(vdev_is_concrete(vd)); 5294 5295 vdev_set_deflate_ratio(vd); 5296 5297 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 5298 vdev_is_concrete(vd)) { 5299 vdev_metaslab_group_create(vd); 5300 VERIFY(vdev_metaslab_init(vd, txg) == 0); 5301 vdev_config_dirty(vd); 5302 } 5303 } 5304 5305 /* 5306 * Split a vdev. 5307 */ 5308 void 5309 vdev_split(vdev_t *vd) 5310 { 5311 vdev_t *cvd, *pvd = vd->vdev_parent; 5312 5313 vdev_remove_child(pvd, vd); 5314 vdev_compact_children(pvd); 5315 5316 cvd = pvd->vdev_child[0]; 5317 if (pvd->vdev_children == 1) { 5318 vdev_remove_parent(cvd); 5319 cvd->vdev_splitting = B_TRUE; 5320 } 5321 vdev_propagate_state(cvd); 5322 } 5323 5324 void 5325 vdev_deadman(vdev_t *vd, const char *tag) 5326 { 5327 for (int c = 0; c < vd->vdev_children; c++) { 5328 vdev_t *cvd = vd->vdev_child[c]; 5329 5330 vdev_deadman(cvd, tag); 5331 } 5332 5333 if (vd->vdev_ops->vdev_op_leaf) { 5334 vdev_queue_t *vq = &vd->vdev_queue; 5335 5336 mutex_enter(&vq->vq_lock); 5337 if (avl_numnodes(&vq->vq_active_tree) > 0) { 5338 spa_t *spa = vd->vdev_spa; 5339 zio_t *fio; 5340 uint64_t delta; 5341 5342 zfs_dbgmsg("slow vdev: %s has %lu active IOs", 5343 vd->vdev_path, avl_numnodes(&vq->vq_active_tree)); 5344 5345 /* 5346 * Look at the head of all the pending queues, 5347 * if any I/O has been outstanding for longer than 5348 * the spa_deadman_synctime invoke the deadman logic. 5349 */ 5350 fio = avl_first(&vq->vq_active_tree); 5351 delta = gethrtime() - fio->io_timestamp; 5352 if (delta > spa_deadman_synctime(spa)) 5353 zio_deadman(fio, tag); 5354 } 5355 mutex_exit(&vq->vq_lock); 5356 } 5357 } 5358 5359 void 5360 vdev_defer_resilver(vdev_t *vd) 5361 { 5362 ASSERT(vd->vdev_ops->vdev_op_leaf); 5363 5364 vd->vdev_resilver_deferred = B_TRUE; 5365 vd->vdev_spa->spa_resilver_deferred = B_TRUE; 5366 } 5367 5368 /* 5369 * Clears the resilver deferred flag on all leaf devs under vd. Returns 5370 * B_TRUE if we have devices that need to be resilvered and are available to 5371 * accept resilver I/Os. 5372 */ 5373 boolean_t 5374 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) 5375 { 5376 boolean_t resilver_needed = B_FALSE; 5377 spa_t *spa = vd->vdev_spa; 5378 5379 for (int c = 0; c < vd->vdev_children; c++) { 5380 vdev_t *cvd = vd->vdev_child[c]; 5381 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); 5382 } 5383 5384 if (vd == spa->spa_root_vdev && 5385 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { 5386 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); 5387 vdev_config_dirty(vd); 5388 spa->spa_resilver_deferred = B_FALSE; 5389 return (resilver_needed); 5390 } 5391 5392 if (!vdev_is_concrete(vd) || vd->vdev_aux || 5393 !vd->vdev_ops->vdev_op_leaf) 5394 return (resilver_needed); 5395 5396 vd->vdev_resilver_deferred = B_FALSE; 5397 5398 return (!vdev_is_dead(vd) && !vd->vdev_offline && 5399 vdev_resilver_needed(vd, NULL, NULL)); 5400 } 5401 5402 boolean_t 5403 vdev_xlate_is_empty(range_seg64_t *rs) 5404 { 5405 return (rs->rs_start == rs->rs_end); 5406 } 5407 5408 /* 5409 * Translate a logical range to the first contiguous physical range for the 5410 * specified vdev_t. This function is initially called with a leaf vdev and 5411 * will walk each parent vdev until it reaches a top-level vdev. Once the 5412 * top-level is reached the physical range is initialized and the recursive 5413 * function begins to unwind. As it unwinds it calls the parent's vdev 5414 * specific translation function to do the real conversion. 5415 */ 5416 void 5417 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 5418 range_seg64_t *physical_rs, range_seg64_t *remain_rs) 5419 { 5420 /* 5421 * Walk up the vdev tree 5422 */ 5423 if (vd != vd->vdev_top) { 5424 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs, 5425 remain_rs); 5426 } else { 5427 /* 5428 * We've reached the top-level vdev, initialize the physical 5429 * range to the logical range and set an empty remaining 5430 * range then start to unwind. 5431 */ 5432 physical_rs->rs_start = logical_rs->rs_start; 5433 physical_rs->rs_end = logical_rs->rs_end; 5434 5435 remain_rs->rs_start = logical_rs->rs_start; 5436 remain_rs->rs_end = logical_rs->rs_start; 5437 5438 return; 5439 } 5440 5441 vdev_t *pvd = vd->vdev_parent; 5442 ASSERT3P(pvd, !=, NULL); 5443 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 5444 5445 /* 5446 * As this recursive function unwinds, translate the logical 5447 * range into its physical and any remaining components by calling 5448 * the vdev specific translate function. 5449 */ 5450 range_seg64_t intermediate = { 0 }; 5451 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate, remain_rs); 5452 5453 physical_rs->rs_start = intermediate.rs_start; 5454 physical_rs->rs_end = intermediate.rs_end; 5455 } 5456 5457 void 5458 vdev_xlate_walk(vdev_t *vd, const range_seg64_t *logical_rs, 5459 vdev_xlate_func_t *func, void *arg) 5460 { 5461 range_seg64_t iter_rs = *logical_rs; 5462 range_seg64_t physical_rs; 5463 range_seg64_t remain_rs; 5464 5465 while (!vdev_xlate_is_empty(&iter_rs)) { 5466 5467 vdev_xlate(vd, &iter_rs, &physical_rs, &remain_rs); 5468 5469 /* 5470 * With raidz and dRAID, it's possible that the logical range 5471 * does not live on this leaf vdev. Only when there is a non- 5472 * zero physical size call the provided function. 5473 */ 5474 if (!vdev_xlate_is_empty(&physical_rs)) 5475 func(arg, &physical_rs); 5476 5477 iter_rs = remain_rs; 5478 } 5479 } 5480 5481 static char * 5482 vdev_name(vdev_t *vd, char *buf, int buflen) 5483 { 5484 if (vd->vdev_path == NULL) { 5485 if (strcmp(vd->vdev_ops->vdev_op_type, "root") == 0) { 5486 strlcpy(buf, vd->vdev_spa->spa_name, buflen); 5487 } else if (!vd->vdev_ops->vdev_op_leaf) { 5488 snprintf(buf, buflen, "%s-%llu", 5489 vd->vdev_ops->vdev_op_type, 5490 (u_longlong_t)vd->vdev_id); 5491 } 5492 } else { 5493 strlcpy(buf, vd->vdev_path, buflen); 5494 } 5495 return (buf); 5496 } 5497 5498 /* 5499 * Look at the vdev tree and determine whether any devices are currently being 5500 * replaced. 5501 */ 5502 boolean_t 5503 vdev_replace_in_progress(vdev_t *vdev) 5504 { 5505 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0); 5506 5507 if (vdev->vdev_ops == &vdev_replacing_ops) 5508 return (B_TRUE); 5509 5510 /* 5511 * A 'spare' vdev indicates that we have a replace in progress, unless 5512 * it has exactly two children, and the second, the hot spare, has 5513 * finished being resilvered. 5514 */ 5515 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 || 5516 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING))) 5517 return (B_TRUE); 5518 5519 for (int i = 0; i < vdev->vdev_children; i++) { 5520 if (vdev_replace_in_progress(vdev->vdev_child[i])) 5521 return (B_TRUE); 5522 } 5523 5524 return (B_FALSE); 5525 } 5526 5527 /* 5528 * Add a (source=src, propname=propval) list to an nvlist. 5529 */ 5530 static void 5531 vdev_prop_add_list(nvlist_t *nvl, const char *propname, char *strval, 5532 uint64_t intval, zprop_source_t src) 5533 { 5534 nvlist_t *propval; 5535 5536 propval = fnvlist_alloc(); 5537 fnvlist_add_uint64(propval, ZPROP_SOURCE, src); 5538 5539 if (strval != NULL) 5540 fnvlist_add_string(propval, ZPROP_VALUE, strval); 5541 else 5542 fnvlist_add_uint64(propval, ZPROP_VALUE, intval); 5543 5544 fnvlist_add_nvlist(nvl, propname, propval); 5545 nvlist_free(propval); 5546 } 5547 5548 static void 5549 vdev_props_set_sync(void *arg, dmu_tx_t *tx) 5550 { 5551 vdev_t *vd; 5552 nvlist_t *nvp = arg; 5553 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 5554 objset_t *mos = spa->spa_meta_objset; 5555 nvpair_t *elem = NULL; 5556 uint64_t vdev_guid; 5557 nvlist_t *nvprops; 5558 5559 vdev_guid = fnvlist_lookup_uint64(nvp, ZPOOL_VDEV_PROPS_SET_VDEV); 5560 nvprops = fnvlist_lookup_nvlist(nvp, ZPOOL_VDEV_PROPS_SET_PROPS); 5561 vd = spa_lookup_by_guid(spa, vdev_guid, B_TRUE); 5562 5563 /* this vdev could get removed while waiting for this sync task */ 5564 if (vd == NULL) 5565 return; 5566 5567 mutex_enter(&spa->spa_props_lock); 5568 5569 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) { 5570 uint64_t intval, objid = 0; 5571 char *strval; 5572 vdev_prop_t prop; 5573 const char *propname = nvpair_name(elem); 5574 zprop_type_t proptype; 5575 5576 /* 5577 * Set vdev property values in the vdev props mos object. 5578 */ 5579 if (vd->vdev_top_zap != 0) { 5580 objid = vd->vdev_top_zap; 5581 } else if (vd->vdev_leaf_zap != 0) { 5582 objid = vd->vdev_leaf_zap; 5583 } else { 5584 panic("vdev not top or leaf"); 5585 } 5586 5587 switch (prop = vdev_name_to_prop(propname)) { 5588 case VDEV_PROP_USERPROP: 5589 if (vdev_prop_user(propname)) { 5590 strval = fnvpair_value_string(elem); 5591 if (strlen(strval) == 0) { 5592 /* remove the property if value == "" */ 5593 (void) zap_remove(mos, objid, propname, 5594 tx); 5595 } else { 5596 VERIFY0(zap_update(mos, objid, propname, 5597 1, strlen(strval) + 1, strval, tx)); 5598 } 5599 spa_history_log_internal(spa, "vdev set", tx, 5600 "vdev_guid=%llu: %s=%s", 5601 (u_longlong_t)vdev_guid, nvpair_name(elem), 5602 strval); 5603 } 5604 break; 5605 default: 5606 /* normalize the property name */ 5607 propname = vdev_prop_to_name(prop); 5608 proptype = vdev_prop_get_type(prop); 5609 5610 if (nvpair_type(elem) == DATA_TYPE_STRING) { 5611 ASSERT(proptype == PROP_TYPE_STRING); 5612 strval = fnvpair_value_string(elem); 5613 VERIFY0(zap_update(mos, objid, propname, 5614 1, strlen(strval) + 1, strval, tx)); 5615 spa_history_log_internal(spa, "vdev set", tx, 5616 "vdev_guid=%llu: %s=%s", 5617 (u_longlong_t)vdev_guid, nvpair_name(elem), 5618 strval); 5619 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 5620 intval = fnvpair_value_uint64(elem); 5621 5622 if (proptype == PROP_TYPE_INDEX) { 5623 const char *unused; 5624 VERIFY0(vdev_prop_index_to_string( 5625 prop, intval, &unused)); 5626 } 5627 VERIFY0(zap_update(mos, objid, propname, 5628 sizeof (uint64_t), 1, &intval, tx)); 5629 spa_history_log_internal(spa, "vdev set", tx, 5630 "vdev_guid=%llu: %s=%lld", 5631 (u_longlong_t)vdev_guid, 5632 nvpair_name(elem), (longlong_t)intval); 5633 } else { 5634 panic("invalid vdev property type %u", 5635 nvpair_type(elem)); 5636 } 5637 } 5638 5639 } 5640 5641 mutex_exit(&spa->spa_props_lock); 5642 } 5643 5644 int 5645 vdev_prop_set(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl) 5646 { 5647 spa_t *spa = vd->vdev_spa; 5648 nvpair_t *elem = NULL; 5649 uint64_t vdev_guid; 5650 nvlist_t *nvprops; 5651 int error; 5652 5653 ASSERT(vd != NULL); 5654 5655 if (nvlist_lookup_uint64(innvl, ZPOOL_VDEV_PROPS_SET_VDEV, 5656 &vdev_guid) != 0) 5657 return (SET_ERROR(EINVAL)); 5658 5659 if (nvlist_lookup_nvlist(innvl, ZPOOL_VDEV_PROPS_SET_PROPS, 5660 &nvprops) != 0) 5661 return (SET_ERROR(EINVAL)); 5662 5663 if ((vd = spa_lookup_by_guid(spa, vdev_guid, B_TRUE)) == NULL) 5664 return (SET_ERROR(EINVAL)); 5665 5666 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) { 5667 char *propname = nvpair_name(elem); 5668 vdev_prop_t prop = vdev_name_to_prop(propname); 5669 uint64_t intval = 0; 5670 char *strval = NULL; 5671 5672 if (prop == VDEV_PROP_USERPROP && !vdev_prop_user(propname)) { 5673 error = EINVAL; 5674 goto end; 5675 } 5676 5677 if (vdev_prop_readonly(prop)) { 5678 error = EROFS; 5679 goto end; 5680 } 5681 5682 /* Special Processing */ 5683 switch (prop) { 5684 case VDEV_PROP_PATH: 5685 if (vd->vdev_path == NULL) { 5686 error = EROFS; 5687 break; 5688 } 5689 if (nvpair_value_string(elem, &strval) != 0) { 5690 error = EINVAL; 5691 break; 5692 } 5693 /* New path must start with /dev/ */ 5694 if (strncmp(strval, "/dev/", 5)) { 5695 error = EINVAL; 5696 break; 5697 } 5698 error = spa_vdev_setpath(spa, vdev_guid, strval); 5699 break; 5700 case VDEV_PROP_ALLOCATING: 5701 if (nvpair_value_uint64(elem, &intval) != 0) { 5702 error = EINVAL; 5703 break; 5704 } 5705 if (intval != vd->vdev_noalloc) 5706 break; 5707 if (intval == 0) 5708 error = spa_vdev_noalloc(spa, vdev_guid); 5709 else 5710 error = spa_vdev_alloc(spa, vdev_guid); 5711 break; 5712 default: 5713 /* Most processing is done in vdev_props_set_sync */ 5714 break; 5715 } 5716 end: 5717 if (error != 0) { 5718 intval = error; 5719 vdev_prop_add_list(outnvl, propname, strval, intval, 0); 5720 return (error); 5721 } 5722 } 5723 5724 return (dsl_sync_task(spa->spa_name, NULL, vdev_props_set_sync, 5725 innvl, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED)); 5726 } 5727 5728 int 5729 vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl) 5730 { 5731 spa_t *spa = vd->vdev_spa; 5732 objset_t *mos = spa->spa_meta_objset; 5733 int err = 0; 5734 uint64_t objid; 5735 uint64_t vdev_guid; 5736 nvpair_t *elem = NULL; 5737 nvlist_t *nvprops = NULL; 5738 uint64_t intval = 0; 5739 char *strval = NULL; 5740 const char *propname = NULL; 5741 vdev_prop_t prop; 5742 5743 ASSERT(vd != NULL); 5744 ASSERT(mos != NULL); 5745 5746 if (nvlist_lookup_uint64(innvl, ZPOOL_VDEV_PROPS_GET_VDEV, 5747 &vdev_guid) != 0) 5748 return (SET_ERROR(EINVAL)); 5749 5750 nvlist_lookup_nvlist(innvl, ZPOOL_VDEV_PROPS_GET_PROPS, &nvprops); 5751 5752 if (vd->vdev_top_zap != 0) { 5753 objid = vd->vdev_top_zap; 5754 } else if (vd->vdev_leaf_zap != 0) { 5755 objid = vd->vdev_leaf_zap; 5756 } else { 5757 return (SET_ERROR(EINVAL)); 5758 } 5759 ASSERT(objid != 0); 5760 5761 mutex_enter(&spa->spa_props_lock); 5762 5763 if (nvprops != NULL) { 5764 char namebuf[64] = { 0 }; 5765 5766 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) { 5767 intval = 0; 5768 strval = NULL; 5769 propname = nvpair_name(elem); 5770 prop = vdev_name_to_prop(propname); 5771 zprop_source_t src = ZPROP_SRC_DEFAULT; 5772 uint64_t integer_size, num_integers; 5773 5774 switch (prop) { 5775 /* Special Read-only Properties */ 5776 case VDEV_PROP_NAME: 5777 strval = vdev_name(vd, namebuf, 5778 sizeof (namebuf)); 5779 if (strval == NULL) 5780 continue; 5781 vdev_prop_add_list(outnvl, propname, strval, 0, 5782 ZPROP_SRC_NONE); 5783 continue; 5784 case VDEV_PROP_CAPACITY: 5785 /* percent used */ 5786 intval = (vd->vdev_stat.vs_dspace == 0) ? 0 : 5787 (vd->vdev_stat.vs_alloc * 100 / 5788 vd->vdev_stat.vs_dspace); 5789 vdev_prop_add_list(outnvl, propname, NULL, 5790 intval, ZPROP_SRC_NONE); 5791 continue; 5792 case VDEV_PROP_STATE: 5793 vdev_prop_add_list(outnvl, propname, NULL, 5794 vd->vdev_state, ZPROP_SRC_NONE); 5795 continue; 5796 case VDEV_PROP_GUID: 5797 vdev_prop_add_list(outnvl, propname, NULL, 5798 vd->vdev_guid, ZPROP_SRC_NONE); 5799 continue; 5800 case VDEV_PROP_ASIZE: 5801 vdev_prop_add_list(outnvl, propname, NULL, 5802 vd->vdev_asize, ZPROP_SRC_NONE); 5803 continue; 5804 case VDEV_PROP_PSIZE: 5805 vdev_prop_add_list(outnvl, propname, NULL, 5806 vd->vdev_psize, ZPROP_SRC_NONE); 5807 continue; 5808 case VDEV_PROP_ASHIFT: 5809 vdev_prop_add_list(outnvl, propname, NULL, 5810 vd->vdev_ashift, ZPROP_SRC_NONE); 5811 continue; 5812 case VDEV_PROP_SIZE: 5813 vdev_prop_add_list(outnvl, propname, NULL, 5814 vd->vdev_stat.vs_dspace, ZPROP_SRC_NONE); 5815 continue; 5816 case VDEV_PROP_FREE: 5817 vdev_prop_add_list(outnvl, propname, NULL, 5818 vd->vdev_stat.vs_dspace - 5819 vd->vdev_stat.vs_alloc, ZPROP_SRC_NONE); 5820 continue; 5821 case VDEV_PROP_ALLOCATED: 5822 vdev_prop_add_list(outnvl, propname, NULL, 5823 vd->vdev_stat.vs_alloc, ZPROP_SRC_NONE); 5824 continue; 5825 case VDEV_PROP_EXPANDSZ: 5826 vdev_prop_add_list(outnvl, propname, NULL, 5827 vd->vdev_stat.vs_esize, ZPROP_SRC_NONE); 5828 continue; 5829 case VDEV_PROP_FRAGMENTATION: 5830 vdev_prop_add_list(outnvl, propname, NULL, 5831 vd->vdev_stat.vs_fragmentation, 5832 ZPROP_SRC_NONE); 5833 continue; 5834 case VDEV_PROP_PARITY: 5835 vdev_prop_add_list(outnvl, propname, NULL, 5836 vdev_get_nparity(vd), ZPROP_SRC_NONE); 5837 continue; 5838 case VDEV_PROP_PATH: 5839 if (vd->vdev_path == NULL) 5840 continue; 5841 vdev_prop_add_list(outnvl, propname, 5842 vd->vdev_path, 0, ZPROP_SRC_NONE); 5843 continue; 5844 case VDEV_PROP_DEVID: 5845 if (vd->vdev_devid == NULL) 5846 continue; 5847 vdev_prop_add_list(outnvl, propname, 5848 vd->vdev_devid, 0, ZPROP_SRC_NONE); 5849 continue; 5850 case VDEV_PROP_PHYS_PATH: 5851 if (vd->vdev_physpath == NULL) 5852 continue; 5853 vdev_prop_add_list(outnvl, propname, 5854 vd->vdev_physpath, 0, ZPROP_SRC_NONE); 5855 continue; 5856 case VDEV_PROP_ENC_PATH: 5857 if (vd->vdev_enc_sysfs_path == NULL) 5858 continue; 5859 vdev_prop_add_list(outnvl, propname, 5860 vd->vdev_enc_sysfs_path, 0, ZPROP_SRC_NONE); 5861 continue; 5862 case VDEV_PROP_FRU: 5863 if (vd->vdev_fru == NULL) 5864 continue; 5865 vdev_prop_add_list(outnvl, propname, 5866 vd->vdev_fru, 0, ZPROP_SRC_NONE); 5867 continue; 5868 case VDEV_PROP_PARENT: 5869 if (vd->vdev_parent != NULL) { 5870 strval = vdev_name(vd->vdev_parent, 5871 namebuf, sizeof (namebuf)); 5872 vdev_prop_add_list(outnvl, propname, 5873 strval, 0, ZPROP_SRC_NONE); 5874 } 5875 continue; 5876 case VDEV_PROP_CHILDREN: 5877 if (vd->vdev_children > 0) 5878 strval = kmem_zalloc(ZAP_MAXVALUELEN, 5879 KM_SLEEP); 5880 for (uint64_t i = 0; i < vd->vdev_children; 5881 i++) { 5882 const char *vname; 5883 5884 vname = vdev_name(vd->vdev_child[i], 5885 namebuf, sizeof (namebuf)); 5886 if (vname == NULL) 5887 vname = "(unknown)"; 5888 if (strlen(strval) > 0) 5889 strlcat(strval, ",", 5890 ZAP_MAXVALUELEN); 5891 strlcat(strval, vname, ZAP_MAXVALUELEN); 5892 } 5893 if (strval != NULL) { 5894 vdev_prop_add_list(outnvl, propname, 5895 strval, 0, ZPROP_SRC_NONE); 5896 kmem_free(strval, ZAP_MAXVALUELEN); 5897 } 5898 continue; 5899 case VDEV_PROP_NUMCHILDREN: 5900 vdev_prop_add_list(outnvl, propname, NULL, 5901 vd->vdev_children, ZPROP_SRC_NONE); 5902 continue; 5903 case VDEV_PROP_READ_ERRORS: 5904 vdev_prop_add_list(outnvl, propname, NULL, 5905 vd->vdev_stat.vs_read_errors, 5906 ZPROP_SRC_NONE); 5907 continue; 5908 case VDEV_PROP_WRITE_ERRORS: 5909 vdev_prop_add_list(outnvl, propname, NULL, 5910 vd->vdev_stat.vs_write_errors, 5911 ZPROP_SRC_NONE); 5912 continue; 5913 case VDEV_PROP_CHECKSUM_ERRORS: 5914 vdev_prop_add_list(outnvl, propname, NULL, 5915 vd->vdev_stat.vs_checksum_errors, 5916 ZPROP_SRC_NONE); 5917 continue; 5918 case VDEV_PROP_INITIALIZE_ERRORS: 5919 vdev_prop_add_list(outnvl, propname, NULL, 5920 vd->vdev_stat.vs_initialize_errors, 5921 ZPROP_SRC_NONE); 5922 continue; 5923 case VDEV_PROP_OPS_NULL: 5924 vdev_prop_add_list(outnvl, propname, NULL, 5925 vd->vdev_stat.vs_ops[ZIO_TYPE_NULL], 5926 ZPROP_SRC_NONE); 5927 continue; 5928 case VDEV_PROP_OPS_READ: 5929 vdev_prop_add_list(outnvl, propname, NULL, 5930 vd->vdev_stat.vs_ops[ZIO_TYPE_READ], 5931 ZPROP_SRC_NONE); 5932 continue; 5933 case VDEV_PROP_OPS_WRITE: 5934 vdev_prop_add_list(outnvl, propname, NULL, 5935 vd->vdev_stat.vs_ops[ZIO_TYPE_WRITE], 5936 ZPROP_SRC_NONE); 5937 continue; 5938 case VDEV_PROP_OPS_FREE: 5939 vdev_prop_add_list(outnvl, propname, NULL, 5940 vd->vdev_stat.vs_ops[ZIO_TYPE_FREE], 5941 ZPROP_SRC_NONE); 5942 continue; 5943 case VDEV_PROP_OPS_CLAIM: 5944 vdev_prop_add_list(outnvl, propname, NULL, 5945 vd->vdev_stat.vs_ops[ZIO_TYPE_CLAIM], 5946 ZPROP_SRC_NONE); 5947 continue; 5948 case VDEV_PROP_OPS_TRIM: 5949 /* 5950 * TRIM ops and bytes are reported to user 5951 * space as ZIO_TYPE_IOCTL. This is done to 5952 * preserve the vdev_stat_t structure layout 5953 * for user space. 5954 */ 5955 vdev_prop_add_list(outnvl, propname, NULL, 5956 vd->vdev_stat.vs_ops[ZIO_TYPE_IOCTL], 5957 ZPROP_SRC_NONE); 5958 continue; 5959 case VDEV_PROP_BYTES_NULL: 5960 vdev_prop_add_list(outnvl, propname, NULL, 5961 vd->vdev_stat.vs_bytes[ZIO_TYPE_NULL], 5962 ZPROP_SRC_NONE); 5963 continue; 5964 case VDEV_PROP_BYTES_READ: 5965 vdev_prop_add_list(outnvl, propname, NULL, 5966 vd->vdev_stat.vs_bytes[ZIO_TYPE_READ], 5967 ZPROP_SRC_NONE); 5968 continue; 5969 case VDEV_PROP_BYTES_WRITE: 5970 vdev_prop_add_list(outnvl, propname, NULL, 5971 vd->vdev_stat.vs_bytes[ZIO_TYPE_WRITE], 5972 ZPROP_SRC_NONE); 5973 continue; 5974 case VDEV_PROP_BYTES_FREE: 5975 vdev_prop_add_list(outnvl, propname, NULL, 5976 vd->vdev_stat.vs_bytes[ZIO_TYPE_FREE], 5977 ZPROP_SRC_NONE); 5978 continue; 5979 case VDEV_PROP_BYTES_CLAIM: 5980 vdev_prop_add_list(outnvl, propname, NULL, 5981 vd->vdev_stat.vs_bytes[ZIO_TYPE_CLAIM], 5982 ZPROP_SRC_NONE); 5983 continue; 5984 case VDEV_PROP_BYTES_TRIM: 5985 /* 5986 * TRIM ops and bytes are reported to user 5987 * space as ZIO_TYPE_IOCTL. This is done to 5988 * preserve the vdev_stat_t structure layout 5989 * for user space. 5990 */ 5991 vdev_prop_add_list(outnvl, propname, NULL, 5992 vd->vdev_stat.vs_bytes[ZIO_TYPE_IOCTL], 5993 ZPROP_SRC_NONE); 5994 continue; 5995 case VDEV_PROP_REMOVING: 5996 vdev_prop_add_list(outnvl, propname, NULL, 5997 vd->vdev_removing, ZPROP_SRC_NONE); 5998 continue; 5999 /* Numeric Properites */ 6000 case VDEV_PROP_ALLOCATING: 6001 src = ZPROP_SRC_LOCAL; 6002 strval = NULL; 6003 6004 err = zap_lookup(mos, objid, nvpair_name(elem), 6005 sizeof (uint64_t), 1, &intval); 6006 if (err == ENOENT) { 6007 intval = 6008 vdev_prop_default_numeric(prop); 6009 err = 0; 6010 } else if (err) 6011 break; 6012 if (intval == vdev_prop_default_numeric(prop)) 6013 src = ZPROP_SRC_DEFAULT; 6014 6015 /* Leaf vdevs cannot have this property */ 6016 if (vd->vdev_mg == NULL && 6017 vd->vdev_top != NULL) { 6018 src = ZPROP_SRC_NONE; 6019 intval = ZPROP_BOOLEAN_NA; 6020 } 6021 6022 vdev_prop_add_list(outnvl, propname, strval, 6023 intval, src); 6024 break; 6025 /* Text Properties */ 6026 case VDEV_PROP_COMMENT: 6027 /* Exists in the ZAP below */ 6028 /* FALLTHRU */ 6029 case VDEV_PROP_USERPROP: 6030 /* User Properites */ 6031 src = ZPROP_SRC_LOCAL; 6032 6033 err = zap_length(mos, objid, nvpair_name(elem), 6034 &integer_size, &num_integers); 6035 if (err) 6036 break; 6037 6038 switch (integer_size) { 6039 case 8: 6040 /* User properties cannot be integers */ 6041 err = EINVAL; 6042 break; 6043 case 1: 6044 /* string property */ 6045 strval = kmem_alloc(num_integers, 6046 KM_SLEEP); 6047 err = zap_lookup(mos, objid, 6048 nvpair_name(elem), 1, 6049 num_integers, strval); 6050 if (err) { 6051 kmem_free(strval, 6052 num_integers); 6053 break; 6054 } 6055 vdev_prop_add_list(outnvl, propname, 6056 strval, 0, src); 6057 kmem_free(strval, num_integers); 6058 break; 6059 } 6060 break; 6061 default: 6062 err = ENOENT; 6063 break; 6064 } 6065 if (err) 6066 break; 6067 } 6068 } else { 6069 /* 6070 * Get all properties from the MOS vdev property object. 6071 */ 6072 zap_cursor_t zc; 6073 zap_attribute_t za; 6074 for (zap_cursor_init(&zc, mos, objid); 6075 (err = zap_cursor_retrieve(&zc, &za)) == 0; 6076 zap_cursor_advance(&zc)) { 6077 intval = 0; 6078 strval = NULL; 6079 zprop_source_t src = ZPROP_SRC_DEFAULT; 6080 propname = za.za_name; 6081 prop = vdev_name_to_prop(propname); 6082 6083 switch (za.za_integer_length) { 6084 case 8: 6085 /* We do not allow integer user properties */ 6086 /* This is likely an internal value */ 6087 break; 6088 case 1: 6089 /* string property */ 6090 strval = kmem_alloc(za.za_num_integers, 6091 KM_SLEEP); 6092 err = zap_lookup(mos, objid, za.za_name, 1, 6093 za.za_num_integers, strval); 6094 if (err) { 6095 kmem_free(strval, za.za_num_integers); 6096 break; 6097 } 6098 vdev_prop_add_list(outnvl, propname, strval, 0, 6099 src); 6100 kmem_free(strval, za.za_num_integers); 6101 break; 6102 6103 default: 6104 break; 6105 } 6106 } 6107 zap_cursor_fini(&zc); 6108 } 6109 6110 mutex_exit(&spa->spa_props_lock); 6111 if (err && err != ENOENT) { 6112 return (err); 6113 } 6114 6115 return (0); 6116 } 6117 6118 EXPORT_SYMBOL(vdev_fault); 6119 EXPORT_SYMBOL(vdev_degrade); 6120 EXPORT_SYMBOL(vdev_online); 6121 EXPORT_SYMBOL(vdev_offline); 6122 EXPORT_SYMBOL(vdev_clear); 6123 6124 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, UINT, ZMOD_RW, 6125 "Target number of metaslabs per top-level vdev"); 6126 6127 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, UINT, ZMOD_RW, 6128 "Default limit for metaslab size"); 6129 6130 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, UINT, ZMOD_RW, 6131 "Minimum number of metaslabs per top-level vdev"); 6132 6133 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, UINT, ZMOD_RW, 6134 "Practical upper limit of total metaslabs per top-level vdev"); 6135 6136 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW, 6137 "Rate limit slow IO (delay) events to this many per second"); 6138 6139 /* BEGIN CSTYLED */ 6140 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW, 6141 "Rate limit checksum events to this many checksum errors per second " 6142 "(do not set below ZED threshold)."); 6143 /* END CSTYLED */ 6144 6145 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW, 6146 "Ignore errors during resilver/scrub"); 6147 6148 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW, 6149 "Bypass vdev_validate()"); 6150 6151 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW, 6152 "Disable cache flushes"); 6153 6154 ZFS_MODULE_PARAM(zfs, zfs_, embedded_slog_min_ms, UINT, ZMOD_RW, 6155 "Minimum number of metaslabs required to dedicate one for log blocks"); 6156 6157 /* BEGIN CSTYLED */ 6158 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift, 6159 param_set_min_auto_ashift, param_get_ulong, ZMOD_RW, 6160 "Minimum ashift used when creating new top-level vdevs"); 6161 6162 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift, 6163 param_set_max_auto_ashift, param_get_ulong, ZMOD_RW, 6164 "Maximum ashift used when optimizing for logical -> physical sector " 6165 "size on new top-level vdevs"); 6166 /* END CSTYLED */ 6167