1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (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 */ 32 33 #include <sys/zfs_context.h> 34 #include <sys/fm/fs/zfs.h> 35 #include <sys/spa.h> 36 #include <sys/spa_impl.h> 37 #include <sys/bpobj.h> 38 #include <sys/dmu.h> 39 #include <sys/dmu_tx.h> 40 #include <sys/dsl_dir.h> 41 #include <sys/vdev_impl.h> 42 #include <sys/vdev_rebuild.h> 43 #include <sys/vdev_draid.h> 44 #include <sys/uberblock_impl.h> 45 #include <sys/metaslab.h> 46 #include <sys/metaslab_impl.h> 47 #include <sys/space_map.h> 48 #include <sys/space_reftree.h> 49 #include <sys/zio.h> 50 #include <sys/zap.h> 51 #include <sys/fs/zfs.h> 52 #include <sys/arc.h> 53 #include <sys/zil.h> 54 #include <sys/dsl_scan.h> 55 #include <sys/vdev_raidz.h> 56 #include <sys/abd.h> 57 #include <sys/vdev_initialize.h> 58 #include <sys/vdev_trim.h> 59 #include <sys/zvol.h> 60 #include <sys/zfs_ratelimit.h> 61 62 /* 63 * One metaslab from each (normal-class) vdev is used by the ZIL. These are 64 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are 65 * part of the spa_embedded_log_class. The metaslab with the most free space 66 * in each vdev is selected for this purpose when the pool is opened (or a 67 * vdev is added). See vdev_metaslab_init(). 68 * 69 * Log blocks can be allocated from the following locations. Each one is tried 70 * in order until the allocation succeeds: 71 * 1. dedicated log vdevs, aka "slog" (spa_log_class) 72 * 2. embedded slog metaslabs (spa_embedded_log_class) 73 * 3. other metaslabs in normal vdevs (spa_normal_class) 74 * 75 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer 76 * than this number of metaslabs in the vdev. This ensures that we don't set 77 * aside an unreasonable amount of space for the ZIL. If set to less than 78 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced 79 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab. 80 */ 81 int zfs_embedded_slog_min_ms = 64; 82 83 /* default target for number of metaslabs per top-level vdev */ 84 int zfs_vdev_default_ms_count = 200; 85 86 /* minimum number of metaslabs per top-level vdev */ 87 int zfs_vdev_min_ms_count = 16; 88 89 /* practical upper limit of total metaslabs per top-level vdev */ 90 int zfs_vdev_ms_count_limit = 1ULL << 17; 91 92 /* lower limit for metaslab size (512M) */ 93 int zfs_vdev_default_ms_shift = 29; 94 95 /* upper limit for metaslab size (16G) */ 96 int zfs_vdev_max_ms_shift = 34; 97 98 int vdev_validate_skip = B_FALSE; 99 100 /* 101 * Since the DTL space map of a vdev is not expected to have a lot of 102 * entries, we default its block size to 4K. 103 */ 104 int zfs_vdev_dtl_sm_blksz = (1 << 12); 105 106 /* 107 * Rate limit slow IO (delay) events to this many per second. 108 */ 109 unsigned int zfs_slow_io_events_per_second = 20; 110 111 /* 112 * Rate limit checksum events after this many checksum errors per second. 113 */ 114 unsigned int zfs_checksum_events_per_second = 20; 115 116 /* 117 * Ignore errors during scrub/resilver. Allows to work around resilver 118 * upon import when there are pool errors. 119 */ 120 int zfs_scan_ignore_errors = 0; 121 122 /* 123 * vdev-wide space maps that have lots of entries written to them at 124 * the end of each transaction can benefit from a higher I/O bandwidth 125 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 126 */ 127 int zfs_vdev_standard_sm_blksz = (1 << 17); 128 129 /* 130 * Tunable parameter for debugging or performance analysis. Setting this 131 * will cause pool corruption on power loss if a volatile out-of-order 132 * write cache is enabled. 133 */ 134 int zfs_nocacheflush = 0; 135 136 uint64_t zfs_vdev_max_auto_ashift = ASHIFT_MAX; 137 uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN; 138 139 /*PRINTFLIKE2*/ 140 void 141 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 142 { 143 va_list adx; 144 char buf[256]; 145 146 va_start(adx, fmt); 147 (void) vsnprintf(buf, sizeof (buf), fmt, adx); 148 va_end(adx); 149 150 if (vd->vdev_path != NULL) { 151 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 152 vd->vdev_path, buf); 153 } else { 154 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 155 vd->vdev_ops->vdev_op_type, 156 (u_longlong_t)vd->vdev_id, 157 (u_longlong_t)vd->vdev_guid, buf); 158 } 159 } 160 161 void 162 vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 163 { 164 char state[20]; 165 166 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 167 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 168 vd->vdev_ops->vdev_op_type); 169 return; 170 } 171 172 switch (vd->vdev_state) { 173 case VDEV_STATE_UNKNOWN: 174 (void) snprintf(state, sizeof (state), "unknown"); 175 break; 176 case VDEV_STATE_CLOSED: 177 (void) snprintf(state, sizeof (state), "closed"); 178 break; 179 case VDEV_STATE_OFFLINE: 180 (void) snprintf(state, sizeof (state), "offline"); 181 break; 182 case VDEV_STATE_REMOVED: 183 (void) snprintf(state, sizeof (state), "removed"); 184 break; 185 case VDEV_STATE_CANT_OPEN: 186 (void) snprintf(state, sizeof (state), "can't open"); 187 break; 188 case VDEV_STATE_FAULTED: 189 (void) snprintf(state, sizeof (state), "faulted"); 190 break; 191 case VDEV_STATE_DEGRADED: 192 (void) snprintf(state, sizeof (state), "degraded"); 193 break; 194 case VDEV_STATE_HEALTHY: 195 (void) snprintf(state, sizeof (state), "healthy"); 196 break; 197 default: 198 (void) snprintf(state, sizeof (state), "<state %u>", 199 (uint_t)vd->vdev_state); 200 } 201 202 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 203 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 204 vd->vdev_islog ? " (log)" : "", 205 (u_longlong_t)vd->vdev_guid, 206 vd->vdev_path ? vd->vdev_path : "N/A", state); 207 208 for (uint64_t i = 0; i < vd->vdev_children; i++) 209 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 210 } 211 212 /* 213 * Virtual device management. 214 */ 215 216 static vdev_ops_t *vdev_ops_table[] = { 217 &vdev_root_ops, 218 &vdev_raidz_ops, 219 &vdev_draid_ops, 220 &vdev_draid_spare_ops, 221 &vdev_mirror_ops, 222 &vdev_replacing_ops, 223 &vdev_spare_ops, 224 &vdev_disk_ops, 225 &vdev_file_ops, 226 &vdev_missing_ops, 227 &vdev_hole_ops, 228 &vdev_indirect_ops, 229 NULL 230 }; 231 232 /* 233 * Given a vdev type, return the appropriate ops vector. 234 */ 235 static vdev_ops_t * 236 vdev_getops(const char *type) 237 { 238 vdev_ops_t *ops, **opspp; 239 240 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 241 if (strcmp(ops->vdev_op_type, type) == 0) 242 break; 243 244 return (ops); 245 } 246 247 /* 248 * Given a vdev and a metaslab class, find which metaslab group we're 249 * interested in. All vdevs may belong to two different metaslab classes. 250 * Dedicated slog devices use only the primary metaslab group, rather than a 251 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL. 252 */ 253 metaslab_group_t * 254 vdev_get_mg(vdev_t *vd, metaslab_class_t *mc) 255 { 256 if (mc == spa_embedded_log_class(vd->vdev_spa) && 257 vd->vdev_log_mg != NULL) 258 return (vd->vdev_log_mg); 259 else 260 return (vd->vdev_mg); 261 } 262 263 /* ARGSUSED */ 264 void 265 vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 266 range_seg64_t *physical_rs, range_seg64_t *remain_rs) 267 { 268 physical_rs->rs_start = logical_rs->rs_start; 269 physical_rs->rs_end = logical_rs->rs_end; 270 } 271 272 /* 273 * Derive the enumerated allocation bias from string input. 274 * String origin is either the per-vdev zap or zpool(8). 275 */ 276 static vdev_alloc_bias_t 277 vdev_derive_alloc_bias(const char *bias) 278 { 279 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 280 281 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 282 alloc_bias = VDEV_BIAS_LOG; 283 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 284 alloc_bias = VDEV_BIAS_SPECIAL; 285 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 286 alloc_bias = VDEV_BIAS_DEDUP; 287 288 return (alloc_bias); 289 } 290 291 /* 292 * Default asize function: return the MAX of psize with the asize of 293 * all children. This is what's used by anything other than RAID-Z. 294 */ 295 uint64_t 296 vdev_default_asize(vdev_t *vd, uint64_t psize) 297 { 298 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 299 uint64_t csize; 300 301 for (int c = 0; c < vd->vdev_children; c++) { 302 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 303 asize = MAX(asize, csize); 304 } 305 306 return (asize); 307 } 308 309 uint64_t 310 vdev_default_min_asize(vdev_t *vd) 311 { 312 return (vd->vdev_min_asize); 313 } 314 315 /* 316 * Get the minimum allocatable size. We define the allocatable size as 317 * the vdev's asize rounded to the nearest metaslab. This allows us to 318 * replace or attach devices which don't have the same physical size but 319 * can still satisfy the same number of allocations. 320 */ 321 uint64_t 322 vdev_get_min_asize(vdev_t *vd) 323 { 324 vdev_t *pvd = vd->vdev_parent; 325 326 /* 327 * If our parent is NULL (inactive spare or cache) or is the root, 328 * just return our own asize. 329 */ 330 if (pvd == NULL) 331 return (vd->vdev_asize); 332 333 /* 334 * The top-level vdev just returns the allocatable size rounded 335 * to the nearest metaslab. 336 */ 337 if (vd == vd->vdev_top) 338 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 339 340 return (pvd->vdev_ops->vdev_op_min_asize(pvd)); 341 } 342 343 void 344 vdev_set_min_asize(vdev_t *vd) 345 { 346 vd->vdev_min_asize = vdev_get_min_asize(vd); 347 348 for (int c = 0; c < vd->vdev_children; c++) 349 vdev_set_min_asize(vd->vdev_child[c]); 350 } 351 352 /* 353 * Get the minimal allocation size for the top-level vdev. 354 */ 355 uint64_t 356 vdev_get_min_alloc(vdev_t *vd) 357 { 358 uint64_t min_alloc = 1ULL << vd->vdev_ashift; 359 360 if (vd->vdev_ops->vdev_op_min_alloc != NULL) 361 min_alloc = vd->vdev_ops->vdev_op_min_alloc(vd); 362 363 return (min_alloc); 364 } 365 366 /* 367 * Get the parity level for a top-level vdev. 368 */ 369 uint64_t 370 vdev_get_nparity(vdev_t *vd) 371 { 372 uint64_t nparity = 0; 373 374 if (vd->vdev_ops->vdev_op_nparity != NULL) 375 nparity = vd->vdev_ops->vdev_op_nparity(vd); 376 377 return (nparity); 378 } 379 380 /* 381 * Get the number of data disks for a top-level vdev. 382 */ 383 uint64_t 384 vdev_get_ndisks(vdev_t *vd) 385 { 386 uint64_t ndisks = 1; 387 388 if (vd->vdev_ops->vdev_op_ndisks != NULL) 389 ndisks = vd->vdev_ops->vdev_op_ndisks(vd); 390 391 return (ndisks); 392 } 393 394 vdev_t * 395 vdev_lookup_top(spa_t *spa, uint64_t vdev) 396 { 397 vdev_t *rvd = spa->spa_root_vdev; 398 399 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 400 401 if (vdev < rvd->vdev_children) { 402 ASSERT(rvd->vdev_child[vdev] != NULL); 403 return (rvd->vdev_child[vdev]); 404 } 405 406 return (NULL); 407 } 408 409 vdev_t * 410 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 411 { 412 vdev_t *mvd; 413 414 if (vd->vdev_guid == guid) 415 return (vd); 416 417 for (int c = 0; c < vd->vdev_children; c++) 418 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 419 NULL) 420 return (mvd); 421 422 return (NULL); 423 } 424 425 static int 426 vdev_count_leaves_impl(vdev_t *vd) 427 { 428 int n = 0; 429 430 if (vd->vdev_ops->vdev_op_leaf) 431 return (1); 432 433 for (int c = 0; c < vd->vdev_children; c++) 434 n += vdev_count_leaves_impl(vd->vdev_child[c]); 435 436 return (n); 437 } 438 439 int 440 vdev_count_leaves(spa_t *spa) 441 { 442 int rc; 443 444 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 445 rc = vdev_count_leaves_impl(spa->spa_root_vdev); 446 spa_config_exit(spa, SCL_VDEV, FTAG); 447 448 return (rc); 449 } 450 451 void 452 vdev_add_child(vdev_t *pvd, vdev_t *cvd) 453 { 454 size_t oldsize, newsize; 455 uint64_t id = cvd->vdev_id; 456 vdev_t **newchild; 457 458 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 459 ASSERT(cvd->vdev_parent == NULL); 460 461 cvd->vdev_parent = pvd; 462 463 if (pvd == NULL) 464 return; 465 466 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 467 468 oldsize = pvd->vdev_children * sizeof (vdev_t *); 469 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 470 newsize = pvd->vdev_children * sizeof (vdev_t *); 471 472 newchild = kmem_alloc(newsize, KM_SLEEP); 473 if (pvd->vdev_child != NULL) { 474 bcopy(pvd->vdev_child, newchild, oldsize); 475 kmem_free(pvd->vdev_child, oldsize); 476 } 477 478 pvd->vdev_child = newchild; 479 pvd->vdev_child[id] = cvd; 480 481 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 482 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 483 484 /* 485 * Walk up all ancestors to update guid sum. 486 */ 487 for (; pvd != NULL; pvd = pvd->vdev_parent) 488 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 489 490 if (cvd->vdev_ops->vdev_op_leaf) { 491 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 492 cvd->vdev_spa->spa_leaf_list_gen++; 493 } 494 } 495 496 void 497 vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 498 { 499 int c; 500 uint_t id = cvd->vdev_id; 501 502 ASSERT(cvd->vdev_parent == pvd); 503 504 if (pvd == NULL) 505 return; 506 507 ASSERT(id < pvd->vdev_children); 508 ASSERT(pvd->vdev_child[id] == cvd); 509 510 pvd->vdev_child[id] = NULL; 511 cvd->vdev_parent = NULL; 512 513 for (c = 0; c < pvd->vdev_children; c++) 514 if (pvd->vdev_child[c]) 515 break; 516 517 if (c == pvd->vdev_children) { 518 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 519 pvd->vdev_child = NULL; 520 pvd->vdev_children = 0; 521 } 522 523 if (cvd->vdev_ops->vdev_op_leaf) { 524 spa_t *spa = cvd->vdev_spa; 525 list_remove(&spa->spa_leaf_list, cvd); 526 spa->spa_leaf_list_gen++; 527 } 528 529 /* 530 * Walk up all ancestors to update guid sum. 531 */ 532 for (; pvd != NULL; pvd = pvd->vdev_parent) 533 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 534 } 535 536 /* 537 * Remove any holes in the child array. 538 */ 539 void 540 vdev_compact_children(vdev_t *pvd) 541 { 542 vdev_t **newchild, *cvd; 543 int oldc = pvd->vdev_children; 544 int newc; 545 546 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 547 548 if (oldc == 0) 549 return; 550 551 for (int c = newc = 0; c < oldc; c++) 552 if (pvd->vdev_child[c]) 553 newc++; 554 555 if (newc > 0) { 556 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); 557 558 for (int c = newc = 0; c < oldc; c++) { 559 if ((cvd = pvd->vdev_child[c]) != NULL) { 560 newchild[newc] = cvd; 561 cvd->vdev_id = newc++; 562 } 563 } 564 } else { 565 newchild = NULL; 566 } 567 568 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 569 pvd->vdev_child = newchild; 570 pvd->vdev_children = newc; 571 } 572 573 /* 574 * Allocate and minimally initialize a vdev_t. 575 */ 576 vdev_t * 577 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 578 { 579 vdev_t *vd; 580 vdev_indirect_config_t *vic; 581 582 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 583 vic = &vd->vdev_indirect_config; 584 585 if (spa->spa_root_vdev == NULL) { 586 ASSERT(ops == &vdev_root_ops); 587 spa->spa_root_vdev = vd; 588 spa->spa_load_guid = spa_generate_guid(NULL); 589 } 590 591 if (guid == 0 && ops != &vdev_hole_ops) { 592 if (spa->spa_root_vdev == vd) { 593 /* 594 * The root vdev's guid will also be the pool guid, 595 * which must be unique among all pools. 596 */ 597 guid = spa_generate_guid(NULL); 598 } else { 599 /* 600 * Any other vdev's guid must be unique within the pool. 601 */ 602 guid = spa_generate_guid(spa); 603 } 604 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 605 } 606 607 vd->vdev_spa = spa; 608 vd->vdev_id = id; 609 vd->vdev_guid = guid; 610 vd->vdev_guid_sum = guid; 611 vd->vdev_ops = ops; 612 vd->vdev_state = VDEV_STATE_CLOSED; 613 vd->vdev_ishole = (ops == &vdev_hole_ops); 614 vic->vic_prev_indirect_vdev = UINT64_MAX; 615 616 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 617 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 618 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 619 0, 0); 620 621 /* 622 * Initialize rate limit structs for events. We rate limit ZIO delay 623 * and checksum events so that we don't overwhelm ZED with thousands 624 * of events when a disk is acting up. 625 */ 626 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second, 627 1); 628 zfs_ratelimit_init(&vd->vdev_checksum_rl, 629 &zfs_checksum_events_per_second, 1); 630 631 list_link_init(&vd->vdev_config_dirty_node); 632 list_link_init(&vd->vdev_state_dirty_node); 633 list_link_init(&vd->vdev_initialize_node); 634 list_link_init(&vd->vdev_leaf_node); 635 list_link_init(&vd->vdev_trim_node); 636 637 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); 638 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 639 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 640 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 641 642 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 643 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 644 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 645 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 646 647 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 648 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 649 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 650 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 651 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 652 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 653 654 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL); 655 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL); 656 657 for (int t = 0; t < DTL_TYPES; t++) { 658 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 659 0); 660 } 661 662 txg_list_create(&vd->vdev_ms_list, spa, 663 offsetof(struct metaslab, ms_txg_node)); 664 txg_list_create(&vd->vdev_dtl_list, spa, 665 offsetof(struct vdev, vdev_dtl_node)); 666 vd->vdev_stat.vs_timestamp = gethrtime(); 667 vdev_queue_init(vd); 668 vdev_cache_init(vd); 669 670 return (vd); 671 } 672 673 /* 674 * Allocate a new vdev. The 'alloctype' is used to control whether we are 675 * creating a new vdev or loading an existing one - the behavior is slightly 676 * different for each case. 677 */ 678 int 679 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 680 int alloctype) 681 { 682 vdev_ops_t *ops; 683 char *type; 684 uint64_t guid = 0, islog; 685 vdev_t *vd; 686 vdev_indirect_config_t *vic; 687 char *tmp = NULL; 688 int rc; 689 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 690 boolean_t top_level = (parent && !parent->vdev_parent); 691 692 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 693 694 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 695 return (SET_ERROR(EINVAL)); 696 697 if ((ops = vdev_getops(type)) == NULL) 698 return (SET_ERROR(EINVAL)); 699 700 /* 701 * If this is a load, get the vdev guid from the nvlist. 702 * Otherwise, vdev_alloc_common() will generate one for us. 703 */ 704 if (alloctype == VDEV_ALLOC_LOAD) { 705 uint64_t label_id; 706 707 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 708 label_id != id) 709 return (SET_ERROR(EINVAL)); 710 711 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 712 return (SET_ERROR(EINVAL)); 713 } else if (alloctype == VDEV_ALLOC_SPARE) { 714 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 715 return (SET_ERROR(EINVAL)); 716 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 717 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 718 return (SET_ERROR(EINVAL)); 719 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 720 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 721 return (SET_ERROR(EINVAL)); 722 } 723 724 /* 725 * The first allocated vdev must be of type 'root'. 726 */ 727 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 728 return (SET_ERROR(EINVAL)); 729 730 /* 731 * Determine whether we're a log vdev. 732 */ 733 islog = 0; 734 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 735 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 736 return (SET_ERROR(ENOTSUP)); 737 738 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 739 return (SET_ERROR(ENOTSUP)); 740 741 if (top_level && alloctype == VDEV_ALLOC_ADD) { 742 char *bias; 743 744 /* 745 * If creating a top-level vdev, check for allocation 746 * classes input. 747 */ 748 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 749 &bias) == 0) { 750 alloc_bias = vdev_derive_alloc_bias(bias); 751 752 /* spa_vdev_add() expects feature to be enabled */ 753 if (spa->spa_load_state != SPA_LOAD_CREATE && 754 !spa_feature_is_enabled(spa, 755 SPA_FEATURE_ALLOCATION_CLASSES)) { 756 return (SET_ERROR(ENOTSUP)); 757 } 758 } 759 760 /* spa_vdev_add() expects feature to be enabled */ 761 if (ops == &vdev_draid_ops && 762 spa->spa_load_state != SPA_LOAD_CREATE && 763 !spa_feature_is_enabled(spa, SPA_FEATURE_DRAID)) { 764 return (SET_ERROR(ENOTSUP)); 765 } 766 } 767 768 /* 769 * Initialize the vdev specific data. This is done before calling 770 * vdev_alloc_common() since it may fail and this simplifies the 771 * error reporting and cleanup code paths. 772 */ 773 void *tsd = NULL; 774 if (ops->vdev_op_init != NULL) { 775 rc = ops->vdev_op_init(spa, nv, &tsd); 776 if (rc != 0) { 777 return (rc); 778 } 779 } 780 781 vd = vdev_alloc_common(spa, id, guid, ops); 782 vd->vdev_tsd = tsd; 783 vd->vdev_islog = islog; 784 785 if (top_level && alloc_bias != VDEV_BIAS_NONE) 786 vd->vdev_alloc_bias = alloc_bias; 787 788 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 789 vd->vdev_path = spa_strdup(vd->vdev_path); 790 791 /* 792 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a 793 * fault on a vdev and want it to persist across imports (like with 794 * zpool offline -f). 795 */ 796 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp); 797 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) { 798 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; 799 vd->vdev_faulted = 1; 800 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 801 } 802 803 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 804 vd->vdev_devid = spa_strdup(vd->vdev_devid); 805 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 806 &vd->vdev_physpath) == 0) 807 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 808 809 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, 810 &vd->vdev_enc_sysfs_path) == 0) 811 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); 812 813 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 814 vd->vdev_fru = spa_strdup(vd->vdev_fru); 815 816 /* 817 * Set the whole_disk property. If it's not specified, leave the value 818 * as -1. 819 */ 820 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 821 &vd->vdev_wholedisk) != 0) 822 vd->vdev_wholedisk = -1ULL; 823 824 vic = &vd->vdev_indirect_config; 825 826 ASSERT0(vic->vic_mapping_object); 827 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 828 &vic->vic_mapping_object); 829 ASSERT0(vic->vic_births_object); 830 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 831 &vic->vic_births_object); 832 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 833 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 834 &vic->vic_prev_indirect_vdev); 835 836 /* 837 * Look for the 'not present' flag. This will only be set if the device 838 * was not present at the time of import. 839 */ 840 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 841 &vd->vdev_not_present); 842 843 /* 844 * Get the alignment requirement. 845 */ 846 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 847 848 /* 849 * Retrieve the vdev creation time. 850 */ 851 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 852 &vd->vdev_crtxg); 853 854 /* 855 * If we're a top-level vdev, try to load the allocation parameters. 856 */ 857 if (top_level && 858 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 859 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 860 &vd->vdev_ms_array); 861 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 862 &vd->vdev_ms_shift); 863 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 864 &vd->vdev_asize); 865 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 866 &vd->vdev_removing); 867 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 868 &vd->vdev_top_zap); 869 } else { 870 ASSERT0(vd->vdev_top_zap); 871 } 872 873 if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 874 ASSERT(alloctype == VDEV_ALLOC_LOAD || 875 alloctype == VDEV_ALLOC_ADD || 876 alloctype == VDEV_ALLOC_SPLIT || 877 alloctype == VDEV_ALLOC_ROOTPOOL); 878 /* Note: metaslab_group_create() is now deferred */ 879 } 880 881 if (vd->vdev_ops->vdev_op_leaf && 882 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 883 (void) nvlist_lookup_uint64(nv, 884 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 885 } else { 886 ASSERT0(vd->vdev_leaf_zap); 887 } 888 889 /* 890 * If we're a leaf vdev, try to load the DTL object and other state. 891 */ 892 893 if (vd->vdev_ops->vdev_op_leaf && 894 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 895 alloctype == VDEV_ALLOC_ROOTPOOL)) { 896 if (alloctype == VDEV_ALLOC_LOAD) { 897 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 898 &vd->vdev_dtl_object); 899 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 900 &vd->vdev_unspare); 901 } 902 903 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 904 uint64_t spare = 0; 905 906 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 907 &spare) == 0 && spare) 908 spa_spare_add(vd); 909 } 910 911 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 912 &vd->vdev_offline); 913 914 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 915 &vd->vdev_resilver_txg); 916 917 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG, 918 &vd->vdev_rebuild_txg); 919 920 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 921 vdev_defer_resilver(vd); 922 923 /* 924 * In general, when importing a pool we want to ignore the 925 * persistent fault state, as the diagnosis made on another 926 * system may not be valid in the current context. The only 927 * exception is if we forced a vdev to a persistently faulted 928 * state with 'zpool offline -f'. The persistent fault will 929 * remain across imports until cleared. 930 * 931 * Local vdevs will remain in the faulted state. 932 */ 933 if (spa_load_state(spa) == SPA_LOAD_OPEN || 934 spa_load_state(spa) == SPA_LOAD_IMPORT) { 935 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 936 &vd->vdev_faulted); 937 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 938 &vd->vdev_degraded); 939 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 940 &vd->vdev_removed); 941 942 if (vd->vdev_faulted || vd->vdev_degraded) { 943 char *aux; 944 945 vd->vdev_label_aux = 946 VDEV_AUX_ERR_EXCEEDED; 947 if (nvlist_lookup_string(nv, 948 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 949 strcmp(aux, "external") == 0) 950 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 951 else 952 vd->vdev_faulted = 0ULL; 953 } 954 } 955 } 956 957 /* 958 * Add ourselves to the parent's list of children. 959 */ 960 vdev_add_child(parent, vd); 961 962 *vdp = vd; 963 964 return (0); 965 } 966 967 void 968 vdev_free(vdev_t *vd) 969 { 970 spa_t *spa = vd->vdev_spa; 971 972 ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 973 ASSERT3P(vd->vdev_trim_thread, ==, NULL); 974 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 975 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL); 976 977 /* 978 * Scan queues are normally destroyed at the end of a scan. If the 979 * queue exists here, that implies the vdev is being removed while 980 * the scan is still running. 981 */ 982 if (vd->vdev_scan_io_queue != NULL) { 983 mutex_enter(&vd->vdev_scan_io_queue_lock); 984 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 985 vd->vdev_scan_io_queue = NULL; 986 mutex_exit(&vd->vdev_scan_io_queue_lock); 987 } 988 989 /* 990 * vdev_free() implies closing the vdev first. This is simpler than 991 * trying to ensure complicated semantics for all callers. 992 */ 993 vdev_close(vd); 994 995 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 996 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 997 998 /* 999 * Free all children. 1000 */ 1001 for (int c = 0; c < vd->vdev_children; c++) 1002 vdev_free(vd->vdev_child[c]); 1003 1004 ASSERT(vd->vdev_child == NULL); 1005 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 1006 1007 if (vd->vdev_ops->vdev_op_fini != NULL) 1008 vd->vdev_ops->vdev_op_fini(vd); 1009 1010 /* 1011 * Discard allocation state. 1012 */ 1013 if (vd->vdev_mg != NULL) { 1014 vdev_metaslab_fini(vd); 1015 metaslab_group_destroy(vd->vdev_mg); 1016 vd->vdev_mg = NULL; 1017 } 1018 if (vd->vdev_log_mg != NULL) { 1019 ASSERT0(vd->vdev_ms_count); 1020 metaslab_group_destroy(vd->vdev_log_mg); 1021 vd->vdev_log_mg = NULL; 1022 } 1023 1024 ASSERT0(vd->vdev_stat.vs_space); 1025 ASSERT0(vd->vdev_stat.vs_dspace); 1026 ASSERT0(vd->vdev_stat.vs_alloc); 1027 1028 /* 1029 * Remove this vdev from its parent's child list. 1030 */ 1031 vdev_remove_child(vd->vdev_parent, vd); 1032 1033 ASSERT(vd->vdev_parent == NULL); 1034 ASSERT(!list_link_active(&vd->vdev_leaf_node)); 1035 1036 /* 1037 * Clean up vdev structure. 1038 */ 1039 vdev_queue_fini(vd); 1040 vdev_cache_fini(vd); 1041 1042 if (vd->vdev_path) 1043 spa_strfree(vd->vdev_path); 1044 if (vd->vdev_devid) 1045 spa_strfree(vd->vdev_devid); 1046 if (vd->vdev_physpath) 1047 spa_strfree(vd->vdev_physpath); 1048 1049 if (vd->vdev_enc_sysfs_path) 1050 spa_strfree(vd->vdev_enc_sysfs_path); 1051 1052 if (vd->vdev_fru) 1053 spa_strfree(vd->vdev_fru); 1054 1055 if (vd->vdev_isspare) 1056 spa_spare_remove(vd); 1057 if (vd->vdev_isl2cache) 1058 spa_l2cache_remove(vd); 1059 1060 txg_list_destroy(&vd->vdev_ms_list); 1061 txg_list_destroy(&vd->vdev_dtl_list); 1062 1063 mutex_enter(&vd->vdev_dtl_lock); 1064 space_map_close(vd->vdev_dtl_sm); 1065 for (int t = 0; t < DTL_TYPES; t++) { 1066 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 1067 range_tree_destroy(vd->vdev_dtl[t]); 1068 } 1069 mutex_exit(&vd->vdev_dtl_lock); 1070 1071 EQUIV(vd->vdev_indirect_births != NULL, 1072 vd->vdev_indirect_mapping != NULL); 1073 if (vd->vdev_indirect_births != NULL) { 1074 vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 1075 vdev_indirect_births_close(vd->vdev_indirect_births); 1076 } 1077 1078 if (vd->vdev_obsolete_sm != NULL) { 1079 ASSERT(vd->vdev_removing || 1080 vd->vdev_ops == &vdev_indirect_ops); 1081 space_map_close(vd->vdev_obsolete_sm); 1082 vd->vdev_obsolete_sm = NULL; 1083 } 1084 range_tree_destroy(vd->vdev_obsolete_segments); 1085 rw_destroy(&vd->vdev_indirect_rwlock); 1086 mutex_destroy(&vd->vdev_obsolete_lock); 1087 1088 mutex_destroy(&vd->vdev_dtl_lock); 1089 mutex_destroy(&vd->vdev_stat_lock); 1090 mutex_destroy(&vd->vdev_probe_lock); 1091 mutex_destroy(&vd->vdev_scan_io_queue_lock); 1092 1093 mutex_destroy(&vd->vdev_initialize_lock); 1094 mutex_destroy(&vd->vdev_initialize_io_lock); 1095 cv_destroy(&vd->vdev_initialize_io_cv); 1096 cv_destroy(&vd->vdev_initialize_cv); 1097 1098 mutex_destroy(&vd->vdev_trim_lock); 1099 mutex_destroy(&vd->vdev_autotrim_lock); 1100 mutex_destroy(&vd->vdev_trim_io_lock); 1101 cv_destroy(&vd->vdev_trim_cv); 1102 cv_destroy(&vd->vdev_autotrim_cv); 1103 cv_destroy(&vd->vdev_trim_io_cv); 1104 1105 mutex_destroy(&vd->vdev_rebuild_lock); 1106 cv_destroy(&vd->vdev_rebuild_cv); 1107 1108 zfs_ratelimit_fini(&vd->vdev_delay_rl); 1109 zfs_ratelimit_fini(&vd->vdev_checksum_rl); 1110 1111 if (vd == spa->spa_root_vdev) 1112 spa->spa_root_vdev = NULL; 1113 1114 kmem_free(vd, sizeof (vdev_t)); 1115 } 1116 1117 /* 1118 * Transfer top-level vdev state from svd to tvd. 1119 */ 1120 static void 1121 vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 1122 { 1123 spa_t *spa = svd->vdev_spa; 1124 metaslab_t *msp; 1125 vdev_t *vd; 1126 int t; 1127 1128 ASSERT(tvd == tvd->vdev_top); 1129 1130 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; 1131 tvd->vdev_ms_array = svd->vdev_ms_array; 1132 tvd->vdev_ms_shift = svd->vdev_ms_shift; 1133 tvd->vdev_ms_count = svd->vdev_ms_count; 1134 tvd->vdev_top_zap = svd->vdev_top_zap; 1135 1136 svd->vdev_ms_array = 0; 1137 svd->vdev_ms_shift = 0; 1138 svd->vdev_ms_count = 0; 1139 svd->vdev_top_zap = 0; 1140 1141 if (tvd->vdev_mg) 1142 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 1143 if (tvd->vdev_log_mg) 1144 ASSERT3P(tvd->vdev_log_mg, ==, svd->vdev_log_mg); 1145 tvd->vdev_mg = svd->vdev_mg; 1146 tvd->vdev_log_mg = svd->vdev_log_mg; 1147 tvd->vdev_ms = svd->vdev_ms; 1148 1149 svd->vdev_mg = NULL; 1150 svd->vdev_log_mg = NULL; 1151 svd->vdev_ms = NULL; 1152 1153 if (tvd->vdev_mg != NULL) 1154 tvd->vdev_mg->mg_vd = tvd; 1155 if (tvd->vdev_log_mg != NULL) 1156 tvd->vdev_log_mg->mg_vd = tvd; 1157 1158 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 1159 svd->vdev_checkpoint_sm = NULL; 1160 1161 tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 1162 svd->vdev_alloc_bias = VDEV_BIAS_NONE; 1163 1164 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 1165 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 1166 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 1167 1168 svd->vdev_stat.vs_alloc = 0; 1169 svd->vdev_stat.vs_space = 0; 1170 svd->vdev_stat.vs_dspace = 0; 1171 1172 /* 1173 * State which may be set on a top-level vdev that's in the 1174 * process of being removed. 1175 */ 1176 ASSERT0(tvd->vdev_indirect_config.vic_births_object); 1177 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 1178 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 1179 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 1180 ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 1181 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 1182 ASSERT0(tvd->vdev_removing); 1183 ASSERT0(tvd->vdev_rebuilding); 1184 tvd->vdev_removing = svd->vdev_removing; 1185 tvd->vdev_rebuilding = svd->vdev_rebuilding; 1186 tvd->vdev_rebuild_config = svd->vdev_rebuild_config; 1187 tvd->vdev_indirect_config = svd->vdev_indirect_config; 1188 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 1189 tvd->vdev_indirect_births = svd->vdev_indirect_births; 1190 range_tree_swap(&svd->vdev_obsolete_segments, 1191 &tvd->vdev_obsolete_segments); 1192 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 1193 svd->vdev_indirect_config.vic_mapping_object = 0; 1194 svd->vdev_indirect_config.vic_births_object = 0; 1195 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 1196 svd->vdev_indirect_mapping = NULL; 1197 svd->vdev_indirect_births = NULL; 1198 svd->vdev_obsolete_sm = NULL; 1199 svd->vdev_removing = 0; 1200 svd->vdev_rebuilding = 0; 1201 1202 for (t = 0; t < TXG_SIZE; t++) { 1203 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1204 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1205 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1206 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1207 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1208 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1209 } 1210 1211 if (list_link_active(&svd->vdev_config_dirty_node)) { 1212 vdev_config_clean(svd); 1213 vdev_config_dirty(tvd); 1214 } 1215 1216 if (list_link_active(&svd->vdev_state_dirty_node)) { 1217 vdev_state_clean(svd); 1218 vdev_state_dirty(tvd); 1219 } 1220 1221 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 1222 svd->vdev_deflate_ratio = 0; 1223 1224 tvd->vdev_islog = svd->vdev_islog; 1225 svd->vdev_islog = 0; 1226 1227 dsl_scan_io_queue_vdev_xfer(svd, tvd); 1228 } 1229 1230 static void 1231 vdev_top_update(vdev_t *tvd, vdev_t *vd) 1232 { 1233 if (vd == NULL) 1234 return; 1235 1236 vd->vdev_top = tvd; 1237 1238 for (int c = 0; c < vd->vdev_children; c++) 1239 vdev_top_update(tvd, vd->vdev_child[c]); 1240 } 1241 1242 /* 1243 * Add a mirror/replacing vdev above an existing vdev. There is no need to 1244 * call .vdev_op_init() since mirror/replacing vdevs do not have private state. 1245 */ 1246 vdev_t * 1247 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1248 { 1249 spa_t *spa = cvd->vdev_spa; 1250 vdev_t *pvd = cvd->vdev_parent; 1251 vdev_t *mvd; 1252 1253 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1254 1255 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1256 1257 mvd->vdev_asize = cvd->vdev_asize; 1258 mvd->vdev_min_asize = cvd->vdev_min_asize; 1259 mvd->vdev_max_asize = cvd->vdev_max_asize; 1260 mvd->vdev_psize = cvd->vdev_psize; 1261 mvd->vdev_ashift = cvd->vdev_ashift; 1262 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift; 1263 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift; 1264 mvd->vdev_state = cvd->vdev_state; 1265 mvd->vdev_crtxg = cvd->vdev_crtxg; 1266 1267 vdev_remove_child(pvd, cvd); 1268 vdev_add_child(pvd, mvd); 1269 cvd->vdev_id = mvd->vdev_children; 1270 vdev_add_child(mvd, cvd); 1271 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1272 1273 if (mvd == mvd->vdev_top) 1274 vdev_top_transfer(cvd, mvd); 1275 1276 return (mvd); 1277 } 1278 1279 /* 1280 * Remove a 1-way mirror/replacing vdev from the tree. 1281 */ 1282 void 1283 vdev_remove_parent(vdev_t *cvd) 1284 { 1285 vdev_t *mvd = cvd->vdev_parent; 1286 vdev_t *pvd = mvd->vdev_parent; 1287 1288 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1289 1290 ASSERT(mvd->vdev_children == 1); 1291 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 1292 mvd->vdev_ops == &vdev_replacing_ops || 1293 mvd->vdev_ops == &vdev_spare_ops); 1294 cvd->vdev_ashift = mvd->vdev_ashift; 1295 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift; 1296 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift; 1297 vdev_remove_child(mvd, cvd); 1298 vdev_remove_child(pvd, mvd); 1299 1300 /* 1301 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1302 * Otherwise, we could have detached an offline device, and when we 1303 * go to import the pool we'll think we have two top-level vdevs, 1304 * instead of a different version of the same top-level vdev. 1305 */ 1306 if (mvd->vdev_top == mvd) { 1307 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 1308 cvd->vdev_orig_guid = cvd->vdev_guid; 1309 cvd->vdev_guid += guid_delta; 1310 cvd->vdev_guid_sum += guid_delta; 1311 1312 /* 1313 * If pool not set for autoexpand, we need to also preserve 1314 * mvd's asize to prevent automatic expansion of cvd. 1315 * Otherwise if we are adjusting the mirror by attaching and 1316 * detaching children of non-uniform sizes, the mirror could 1317 * autoexpand, unexpectedly requiring larger devices to 1318 * re-establish the mirror. 1319 */ 1320 if (!cvd->vdev_spa->spa_autoexpand) 1321 cvd->vdev_asize = mvd->vdev_asize; 1322 } 1323 cvd->vdev_id = mvd->vdev_id; 1324 vdev_add_child(pvd, cvd); 1325 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1326 1327 if (cvd == cvd->vdev_top) 1328 vdev_top_transfer(mvd, cvd); 1329 1330 ASSERT(mvd->vdev_children == 0); 1331 vdev_free(mvd); 1332 } 1333 1334 void 1335 vdev_metaslab_group_create(vdev_t *vd) 1336 { 1337 spa_t *spa = vd->vdev_spa; 1338 1339 /* 1340 * metaslab_group_create was delayed until allocation bias was available 1341 */ 1342 if (vd->vdev_mg == NULL) { 1343 metaslab_class_t *mc; 1344 1345 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1346 vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1347 1348 ASSERT3U(vd->vdev_islog, ==, 1349 (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1350 1351 switch (vd->vdev_alloc_bias) { 1352 case VDEV_BIAS_LOG: 1353 mc = spa_log_class(spa); 1354 break; 1355 case VDEV_BIAS_SPECIAL: 1356 mc = spa_special_class(spa); 1357 break; 1358 case VDEV_BIAS_DEDUP: 1359 mc = spa_dedup_class(spa); 1360 break; 1361 default: 1362 mc = spa_normal_class(spa); 1363 } 1364 1365 vd->vdev_mg = metaslab_group_create(mc, vd, 1366 spa->spa_alloc_count); 1367 1368 if (!vd->vdev_islog) { 1369 vd->vdev_log_mg = metaslab_group_create( 1370 spa_embedded_log_class(spa), vd, 1); 1371 } 1372 1373 /* 1374 * The spa ashift min/max only apply for the normal metaslab 1375 * class. Class destination is late binding so ashift boundry 1376 * setting had to wait until now. 1377 */ 1378 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1379 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1380 if (vd->vdev_ashift > spa->spa_max_ashift) 1381 spa->spa_max_ashift = vd->vdev_ashift; 1382 if (vd->vdev_ashift < spa->spa_min_ashift) 1383 spa->spa_min_ashift = vd->vdev_ashift; 1384 1385 uint64_t min_alloc = vdev_get_min_alloc(vd); 1386 if (min_alloc < spa->spa_min_alloc) 1387 spa->spa_min_alloc = min_alloc; 1388 } 1389 } 1390 } 1391 1392 int 1393 vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1394 { 1395 spa_t *spa = vd->vdev_spa; 1396 uint64_t oldc = vd->vdev_ms_count; 1397 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1398 metaslab_t **mspp; 1399 int error; 1400 boolean_t expanding = (oldc != 0); 1401 1402 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1403 1404 /* 1405 * This vdev is not being allocated from yet or is a hole. 1406 */ 1407 if (vd->vdev_ms_shift == 0) 1408 return (0); 1409 1410 ASSERT(!vd->vdev_ishole); 1411 1412 ASSERT(oldc <= newc); 1413 1414 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1415 1416 if (expanding) { 1417 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1418 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1419 } 1420 1421 vd->vdev_ms = mspp; 1422 vd->vdev_ms_count = newc; 1423 1424 for (uint64_t m = oldc; m < newc; m++) { 1425 uint64_t object = 0; 1426 /* 1427 * vdev_ms_array may be 0 if we are creating the "fake" 1428 * metaslabs for an indirect vdev for zdb's leak detection. 1429 * See zdb_leak_init(). 1430 */ 1431 if (txg == 0 && vd->vdev_ms_array != 0) { 1432 error = dmu_read(spa->spa_meta_objset, 1433 vd->vdev_ms_array, 1434 m * sizeof (uint64_t), sizeof (uint64_t), &object, 1435 DMU_READ_PREFETCH); 1436 if (error != 0) { 1437 vdev_dbgmsg(vd, "unable to read the metaslab " 1438 "array [error=%d]", error); 1439 return (error); 1440 } 1441 } 1442 1443 error = metaslab_init(vd->vdev_mg, m, object, txg, 1444 &(vd->vdev_ms[m])); 1445 if (error != 0) { 1446 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 1447 error); 1448 return (error); 1449 } 1450 } 1451 1452 /* 1453 * Find the emptiest metaslab on the vdev and mark it for use for 1454 * embedded slog by moving it from the regular to the log metaslab 1455 * group. 1456 */ 1457 if (vd->vdev_mg->mg_class == spa_normal_class(spa) && 1458 vd->vdev_ms_count > zfs_embedded_slog_min_ms && 1459 avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) { 1460 uint64_t slog_msid = 0; 1461 uint64_t smallest = UINT64_MAX; 1462 1463 /* 1464 * Note, we only search the new metaslabs, because the old 1465 * (pre-existing) ones may be active (e.g. have non-empty 1466 * range_tree's), and we don't move them to the new 1467 * metaslab_t. 1468 */ 1469 for (uint64_t m = oldc; m < newc; m++) { 1470 uint64_t alloc = 1471 space_map_allocated(vd->vdev_ms[m]->ms_sm); 1472 if (alloc < smallest) { 1473 slog_msid = m; 1474 smallest = alloc; 1475 } 1476 } 1477 metaslab_t *slog_ms = vd->vdev_ms[slog_msid]; 1478 /* 1479 * The metaslab was marked as dirty at the end of 1480 * metaslab_init(). Remove it from the dirty list so that we 1481 * can uninitialize and reinitialize it to the new class. 1482 */ 1483 if (txg != 0) { 1484 (void) txg_list_remove_this(&vd->vdev_ms_list, 1485 slog_ms, txg); 1486 } 1487 uint64_t sm_obj = space_map_object(slog_ms->ms_sm); 1488 metaslab_fini(slog_ms); 1489 VERIFY0(metaslab_init(vd->vdev_log_mg, slog_msid, sm_obj, txg, 1490 &vd->vdev_ms[slog_msid])); 1491 } 1492 1493 if (txg == 0) 1494 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1495 1496 /* 1497 * If the vdev is being removed we don't activate 1498 * the metaslabs since we want to ensure that no new 1499 * allocations are performed on this device. 1500 */ 1501 if (!expanding && !vd->vdev_removing) { 1502 metaslab_group_activate(vd->vdev_mg); 1503 if (vd->vdev_log_mg != NULL) 1504 metaslab_group_activate(vd->vdev_log_mg); 1505 } 1506 1507 if (txg == 0) 1508 spa_config_exit(spa, SCL_ALLOC, FTAG); 1509 1510 /* 1511 * Regardless whether this vdev was just added or it is being 1512 * expanded, the metaslab count has changed. Recalculate the 1513 * block limit. 1514 */ 1515 spa_log_sm_set_blocklimit(spa); 1516 1517 return (0); 1518 } 1519 1520 void 1521 vdev_metaslab_fini(vdev_t *vd) 1522 { 1523 if (vd->vdev_checkpoint_sm != NULL) { 1524 ASSERT(spa_feature_is_active(vd->vdev_spa, 1525 SPA_FEATURE_POOL_CHECKPOINT)); 1526 space_map_close(vd->vdev_checkpoint_sm); 1527 /* 1528 * Even though we close the space map, we need to set its 1529 * pointer to NULL. The reason is that vdev_metaslab_fini() 1530 * may be called multiple times for certain operations 1531 * (i.e. when destroying a pool) so we need to ensure that 1532 * this clause never executes twice. This logic is similar 1533 * to the one used for the vdev_ms clause below. 1534 */ 1535 vd->vdev_checkpoint_sm = NULL; 1536 } 1537 1538 if (vd->vdev_ms != NULL) { 1539 metaslab_group_t *mg = vd->vdev_mg; 1540 1541 metaslab_group_passivate(mg); 1542 if (vd->vdev_log_mg != NULL) { 1543 ASSERT(!vd->vdev_islog); 1544 metaslab_group_passivate(vd->vdev_log_mg); 1545 } 1546 1547 uint64_t count = vd->vdev_ms_count; 1548 for (uint64_t m = 0; m < count; m++) { 1549 metaslab_t *msp = vd->vdev_ms[m]; 1550 if (msp != NULL) 1551 metaslab_fini(msp); 1552 } 1553 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1554 vd->vdev_ms = NULL; 1555 vd->vdev_ms_count = 0; 1556 1557 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 1558 ASSERT0(mg->mg_histogram[i]); 1559 if (vd->vdev_log_mg != NULL) 1560 ASSERT0(vd->vdev_log_mg->mg_histogram[i]); 1561 } 1562 } 1563 ASSERT0(vd->vdev_ms_count); 1564 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); 1565 } 1566 1567 typedef struct vdev_probe_stats { 1568 boolean_t vps_readable; 1569 boolean_t vps_writeable; 1570 int vps_flags; 1571 } vdev_probe_stats_t; 1572 1573 static void 1574 vdev_probe_done(zio_t *zio) 1575 { 1576 spa_t *spa = zio->io_spa; 1577 vdev_t *vd = zio->io_vd; 1578 vdev_probe_stats_t *vps = zio->io_private; 1579 1580 ASSERT(vd->vdev_probe_zio != NULL); 1581 1582 if (zio->io_type == ZIO_TYPE_READ) { 1583 if (zio->io_error == 0) 1584 vps->vps_readable = 1; 1585 if (zio->io_error == 0 && spa_writeable(spa)) { 1586 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1587 zio->io_offset, zio->io_size, zio->io_abd, 1588 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1589 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1590 } else { 1591 abd_free(zio->io_abd); 1592 } 1593 } else if (zio->io_type == ZIO_TYPE_WRITE) { 1594 if (zio->io_error == 0) 1595 vps->vps_writeable = 1; 1596 abd_free(zio->io_abd); 1597 } else if (zio->io_type == ZIO_TYPE_NULL) { 1598 zio_t *pio; 1599 zio_link_t *zl; 1600 1601 vd->vdev_cant_read |= !vps->vps_readable; 1602 vd->vdev_cant_write |= !vps->vps_writeable; 1603 1604 if (vdev_readable(vd) && 1605 (vdev_writeable(vd) || !spa_writeable(spa))) { 1606 zio->io_error = 0; 1607 } else { 1608 ASSERT(zio->io_error != 0); 1609 vdev_dbgmsg(vd, "failed probe"); 1610 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1611 spa, vd, NULL, NULL, 0); 1612 zio->io_error = SET_ERROR(ENXIO); 1613 } 1614 1615 mutex_enter(&vd->vdev_probe_lock); 1616 ASSERT(vd->vdev_probe_zio == zio); 1617 vd->vdev_probe_zio = NULL; 1618 mutex_exit(&vd->vdev_probe_lock); 1619 1620 zl = NULL; 1621 while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1622 if (!vdev_accessible(vd, pio)) 1623 pio->io_error = SET_ERROR(ENXIO); 1624 1625 kmem_free(vps, sizeof (*vps)); 1626 } 1627 } 1628 1629 /* 1630 * Determine whether this device is accessible. 1631 * 1632 * Read and write to several known locations: the pad regions of each 1633 * vdev label but the first, which we leave alone in case it contains 1634 * a VTOC. 1635 */ 1636 zio_t * 1637 vdev_probe(vdev_t *vd, zio_t *zio) 1638 { 1639 spa_t *spa = vd->vdev_spa; 1640 vdev_probe_stats_t *vps = NULL; 1641 zio_t *pio; 1642 1643 ASSERT(vd->vdev_ops->vdev_op_leaf); 1644 1645 /* 1646 * Don't probe the probe. 1647 */ 1648 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1649 return (NULL); 1650 1651 /* 1652 * To prevent 'probe storms' when a device fails, we create 1653 * just one probe i/o at a time. All zios that want to probe 1654 * this vdev will become parents of the probe io. 1655 */ 1656 mutex_enter(&vd->vdev_probe_lock); 1657 1658 if ((pio = vd->vdev_probe_zio) == NULL) { 1659 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1660 1661 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1662 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1663 ZIO_FLAG_TRYHARD; 1664 1665 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1666 /* 1667 * vdev_cant_read and vdev_cant_write can only 1668 * transition from TRUE to FALSE when we have the 1669 * SCL_ZIO lock as writer; otherwise they can only 1670 * transition from FALSE to TRUE. This ensures that 1671 * any zio looking at these values can assume that 1672 * failures persist for the life of the I/O. That's 1673 * important because when a device has intermittent 1674 * connectivity problems, we want to ensure that 1675 * they're ascribed to the device (ENXIO) and not 1676 * the zio (EIO). 1677 * 1678 * Since we hold SCL_ZIO as writer here, clear both 1679 * values so the probe can reevaluate from first 1680 * principles. 1681 */ 1682 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1683 vd->vdev_cant_read = B_FALSE; 1684 vd->vdev_cant_write = B_FALSE; 1685 } 1686 1687 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1688 vdev_probe_done, vps, 1689 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1690 1691 /* 1692 * We can't change the vdev state in this context, so we 1693 * kick off an async task to do it on our behalf. 1694 */ 1695 if (zio != NULL) { 1696 vd->vdev_probe_wanted = B_TRUE; 1697 spa_async_request(spa, SPA_ASYNC_PROBE); 1698 } 1699 } 1700 1701 if (zio != NULL) 1702 zio_add_child(zio, pio); 1703 1704 mutex_exit(&vd->vdev_probe_lock); 1705 1706 if (vps == NULL) { 1707 ASSERT(zio != NULL); 1708 return (NULL); 1709 } 1710 1711 for (int l = 1; l < VDEV_LABELS; l++) { 1712 zio_nowait(zio_read_phys(pio, vd, 1713 vdev_label_offset(vd->vdev_psize, l, 1714 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE, 1715 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1716 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1717 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1718 } 1719 1720 if (zio == NULL) 1721 return (pio); 1722 1723 zio_nowait(pio); 1724 return (NULL); 1725 } 1726 1727 static void 1728 vdev_load_child(void *arg) 1729 { 1730 vdev_t *vd = arg; 1731 1732 vd->vdev_load_error = vdev_load(vd); 1733 } 1734 1735 static void 1736 vdev_open_child(void *arg) 1737 { 1738 vdev_t *vd = arg; 1739 1740 vd->vdev_open_thread = curthread; 1741 vd->vdev_open_error = vdev_open(vd); 1742 vd->vdev_open_thread = NULL; 1743 } 1744 1745 static boolean_t 1746 vdev_uses_zvols(vdev_t *vd) 1747 { 1748 #ifdef _KERNEL 1749 if (zvol_is_zvol(vd->vdev_path)) 1750 return (B_TRUE); 1751 #endif 1752 1753 for (int c = 0; c < vd->vdev_children; c++) 1754 if (vdev_uses_zvols(vd->vdev_child[c])) 1755 return (B_TRUE); 1756 1757 return (B_FALSE); 1758 } 1759 1760 /* 1761 * Returns B_TRUE if the passed child should be opened. 1762 */ 1763 static boolean_t 1764 vdev_default_open_children_func(vdev_t *vd) 1765 { 1766 return (B_TRUE); 1767 } 1768 1769 /* 1770 * Open the requested child vdevs. If any of the leaf vdevs are using 1771 * a ZFS volume then do the opens in a single thread. This avoids a 1772 * deadlock when the current thread is holding the spa_namespace_lock. 1773 */ 1774 static void 1775 vdev_open_children_impl(vdev_t *vd, vdev_open_children_func_t *open_func) 1776 { 1777 int children = vd->vdev_children; 1778 1779 taskq_t *tq = taskq_create("vdev_open", children, minclsyspri, 1780 children, children, TASKQ_PREPOPULATE); 1781 vd->vdev_nonrot = B_TRUE; 1782 1783 for (int c = 0; c < children; c++) { 1784 vdev_t *cvd = vd->vdev_child[c]; 1785 1786 if (open_func(cvd) == B_FALSE) 1787 continue; 1788 1789 if (tq == NULL || vdev_uses_zvols(vd)) { 1790 cvd->vdev_open_error = vdev_open(cvd); 1791 } else { 1792 VERIFY(taskq_dispatch(tq, vdev_open_child, 1793 cvd, TQ_SLEEP) != TASKQID_INVALID); 1794 } 1795 1796 vd->vdev_nonrot &= cvd->vdev_nonrot; 1797 } 1798 1799 if (tq != NULL) { 1800 taskq_wait(tq); 1801 taskq_destroy(tq); 1802 } 1803 } 1804 1805 /* 1806 * Open all child vdevs. 1807 */ 1808 void 1809 vdev_open_children(vdev_t *vd) 1810 { 1811 vdev_open_children_impl(vd, vdev_default_open_children_func); 1812 } 1813 1814 /* 1815 * Conditionally open a subset of child vdevs. 1816 */ 1817 void 1818 vdev_open_children_subset(vdev_t *vd, vdev_open_children_func_t *open_func) 1819 { 1820 vdev_open_children_impl(vd, open_func); 1821 } 1822 1823 /* 1824 * Compute the raidz-deflation ratio. Note, we hard-code 1825 * in 128k (1 << 17) because it is the "typical" blocksize. 1826 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 1827 * otherwise it would inconsistently account for existing bp's. 1828 */ 1829 static void 1830 vdev_set_deflate_ratio(vdev_t *vd) 1831 { 1832 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 1833 vd->vdev_deflate_ratio = (1 << 17) / 1834 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 1835 } 1836 } 1837 1838 /* 1839 * Maximize performance by inflating the configured ashift for top level 1840 * vdevs to be as close to the physical ashift as possible while maintaining 1841 * administrator defined limits and ensuring it doesn't go below the 1842 * logical ashift. 1843 */ 1844 static void 1845 vdev_ashift_optimize(vdev_t *vd) 1846 { 1847 ASSERT(vd == vd->vdev_top); 1848 1849 if (vd->vdev_ashift < vd->vdev_physical_ashift) { 1850 vd->vdev_ashift = MIN( 1851 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift), 1852 MAX(zfs_vdev_min_auto_ashift, 1853 vd->vdev_physical_ashift)); 1854 } else { 1855 /* 1856 * If the logical and physical ashifts are the same, then 1857 * we ensure that the top-level vdev's ashift is not smaller 1858 * than our minimum ashift value. For the unusual case 1859 * where logical ashift > physical ashift, we can't cap 1860 * the calculated ashift based on max ashift as that 1861 * would cause failures. 1862 * We still check if we need to increase it to match 1863 * the min ashift. 1864 */ 1865 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift, 1866 vd->vdev_ashift); 1867 } 1868 } 1869 1870 /* 1871 * Prepare a virtual device for access. 1872 */ 1873 int 1874 vdev_open(vdev_t *vd) 1875 { 1876 spa_t *spa = vd->vdev_spa; 1877 int error; 1878 uint64_t osize = 0; 1879 uint64_t max_osize = 0; 1880 uint64_t asize, max_asize, psize; 1881 uint64_t logical_ashift = 0; 1882 uint64_t physical_ashift = 0; 1883 1884 ASSERT(vd->vdev_open_thread == curthread || 1885 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1886 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1887 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1888 vd->vdev_state == VDEV_STATE_OFFLINE); 1889 1890 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1891 vd->vdev_cant_read = B_FALSE; 1892 vd->vdev_cant_write = B_FALSE; 1893 vd->vdev_min_asize = vdev_get_min_asize(vd); 1894 1895 /* 1896 * If this vdev is not removed, check its fault status. If it's 1897 * faulted, bail out of the open. 1898 */ 1899 if (!vd->vdev_removed && vd->vdev_faulted) { 1900 ASSERT(vd->vdev_children == 0); 1901 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1902 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1903 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1904 vd->vdev_label_aux); 1905 return (SET_ERROR(ENXIO)); 1906 } else if (vd->vdev_offline) { 1907 ASSERT(vd->vdev_children == 0); 1908 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1909 return (SET_ERROR(ENXIO)); 1910 } 1911 1912 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, 1913 &logical_ashift, &physical_ashift); 1914 /* 1915 * Physical volume size should never be larger than its max size, unless 1916 * the disk has shrunk while we were reading it or the device is buggy 1917 * or damaged: either way it's not safe for use, bail out of the open. 1918 */ 1919 if (osize > max_osize) { 1920 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1921 VDEV_AUX_OPEN_FAILED); 1922 return (SET_ERROR(ENXIO)); 1923 } 1924 1925 /* 1926 * Reset the vdev_reopening flag so that we actually close 1927 * the vdev on error. 1928 */ 1929 vd->vdev_reopening = B_FALSE; 1930 if (zio_injection_enabled && error == 0) 1931 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO)); 1932 1933 if (error) { 1934 if (vd->vdev_removed && 1935 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1936 vd->vdev_removed = B_FALSE; 1937 1938 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1939 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1940 vd->vdev_stat.vs_aux); 1941 } else { 1942 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1943 vd->vdev_stat.vs_aux); 1944 } 1945 return (error); 1946 } 1947 1948 vd->vdev_removed = B_FALSE; 1949 1950 /* 1951 * Recheck the faulted flag now that we have confirmed that 1952 * the vdev is accessible. If we're faulted, bail. 1953 */ 1954 if (vd->vdev_faulted) { 1955 ASSERT(vd->vdev_children == 0); 1956 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1957 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1958 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1959 vd->vdev_label_aux); 1960 return (SET_ERROR(ENXIO)); 1961 } 1962 1963 if (vd->vdev_degraded) { 1964 ASSERT(vd->vdev_children == 0); 1965 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1966 VDEV_AUX_ERR_EXCEEDED); 1967 } else { 1968 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1969 } 1970 1971 /* 1972 * For hole or missing vdevs we just return success. 1973 */ 1974 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1975 return (0); 1976 1977 for (int c = 0; c < vd->vdev_children; c++) { 1978 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1979 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1980 VDEV_AUX_NONE); 1981 break; 1982 } 1983 } 1984 1985 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1986 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1987 1988 if (vd->vdev_children == 0) { 1989 if (osize < SPA_MINDEVSIZE) { 1990 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1991 VDEV_AUX_TOO_SMALL); 1992 return (SET_ERROR(EOVERFLOW)); 1993 } 1994 psize = osize; 1995 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1996 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1997 VDEV_LABEL_END_SIZE); 1998 } else { 1999 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 2000 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 2001 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2002 VDEV_AUX_TOO_SMALL); 2003 return (SET_ERROR(EOVERFLOW)); 2004 } 2005 psize = 0; 2006 asize = osize; 2007 max_asize = max_osize; 2008 } 2009 2010 /* 2011 * If the vdev was expanded, record this so that we can re-create the 2012 * uberblock rings in labels {2,3}, during the next sync. 2013 */ 2014 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0)) 2015 vd->vdev_copy_uberblocks = B_TRUE; 2016 2017 vd->vdev_psize = psize; 2018 2019 /* 2020 * Make sure the allocatable size hasn't shrunk too much. 2021 */ 2022 if (asize < vd->vdev_min_asize) { 2023 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2024 VDEV_AUX_BAD_LABEL); 2025 return (SET_ERROR(EINVAL)); 2026 } 2027 2028 /* 2029 * We can always set the logical/physical ashift members since 2030 * their values are only used to calculate the vdev_ashift when 2031 * the device is first added to the config. These values should 2032 * not be used for anything else since they may change whenever 2033 * the device is reopened and we don't store them in the label. 2034 */ 2035 vd->vdev_physical_ashift = 2036 MAX(physical_ashift, vd->vdev_physical_ashift); 2037 vd->vdev_logical_ashift = MAX(logical_ashift, 2038 vd->vdev_logical_ashift); 2039 2040 if (vd->vdev_asize == 0) { 2041 /* 2042 * This is the first-ever open, so use the computed values. 2043 * For compatibility, a different ashift can be requested. 2044 */ 2045 vd->vdev_asize = asize; 2046 vd->vdev_max_asize = max_asize; 2047 2048 /* 2049 * If the vdev_ashift was not overriden at creation time, 2050 * then set it the logical ashift and optimize the ashift. 2051 */ 2052 if (vd->vdev_ashift == 0) { 2053 vd->vdev_ashift = vd->vdev_logical_ashift; 2054 2055 if (vd->vdev_logical_ashift > ASHIFT_MAX) { 2056 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2057 VDEV_AUX_ASHIFT_TOO_BIG); 2058 return (SET_ERROR(EDOM)); 2059 } 2060 2061 if (vd->vdev_top == vd) { 2062 vdev_ashift_optimize(vd); 2063 } 2064 } 2065 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 2066 vd->vdev_ashift > ASHIFT_MAX)) { 2067 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2068 VDEV_AUX_BAD_ASHIFT); 2069 return (SET_ERROR(EDOM)); 2070 } 2071 } else { 2072 /* 2073 * Make sure the alignment required hasn't increased. 2074 */ 2075 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift && 2076 vd->vdev_ops->vdev_op_leaf) { 2077 (void) zfs_ereport_post( 2078 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 2079 spa, vd, NULL, NULL, 0); 2080 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2081 VDEV_AUX_BAD_LABEL); 2082 return (SET_ERROR(EDOM)); 2083 } 2084 vd->vdev_max_asize = max_asize; 2085 } 2086 2087 /* 2088 * If all children are healthy we update asize if either: 2089 * The asize has increased, due to a device expansion caused by dynamic 2090 * LUN growth or vdev replacement, and automatic expansion is enabled; 2091 * making the additional space available. 2092 * 2093 * The asize has decreased, due to a device shrink usually caused by a 2094 * vdev replace with a smaller device. This ensures that calculations 2095 * based of max_asize and asize e.g. esize are always valid. It's safe 2096 * to do this as we've already validated that asize is greater than 2097 * vdev_min_asize. 2098 */ 2099 if (vd->vdev_state == VDEV_STATE_HEALTHY && 2100 ((asize > vd->vdev_asize && 2101 (vd->vdev_expanding || spa->spa_autoexpand)) || 2102 (asize < vd->vdev_asize))) 2103 vd->vdev_asize = asize; 2104 2105 vdev_set_min_asize(vd); 2106 2107 /* 2108 * Ensure we can issue some IO before declaring the 2109 * vdev open for business. 2110 */ 2111 if (vd->vdev_ops->vdev_op_leaf && 2112 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 2113 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 2114 VDEV_AUX_ERR_EXCEEDED); 2115 return (error); 2116 } 2117 2118 /* 2119 * Track the the minimum allocation size. 2120 */ 2121 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 2122 vd->vdev_islog == 0 && vd->vdev_aux == NULL) { 2123 uint64_t min_alloc = vdev_get_min_alloc(vd); 2124 if (min_alloc < spa->spa_min_alloc) 2125 spa->spa_min_alloc = min_alloc; 2126 } 2127 2128 /* 2129 * If this is a leaf vdev, assess whether a resilver is needed. 2130 * But don't do this if we are doing a reopen for a scrub, since 2131 * this would just restart the scrub we are already doing. 2132 */ 2133 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) 2134 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); 2135 2136 return (0); 2137 } 2138 2139 static void 2140 vdev_validate_child(void *arg) 2141 { 2142 vdev_t *vd = arg; 2143 2144 vd->vdev_validate_thread = curthread; 2145 vd->vdev_validate_error = vdev_validate(vd); 2146 vd->vdev_validate_thread = NULL; 2147 } 2148 2149 /* 2150 * Called once the vdevs are all opened, this routine validates the label 2151 * contents. This needs to be done before vdev_load() so that we don't 2152 * inadvertently do repair I/Os to the wrong device. 2153 * 2154 * This function will only return failure if one of the vdevs indicates that it 2155 * has since been destroyed or exported. This is only possible if 2156 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 2157 * will be updated but the function will return 0. 2158 */ 2159 int 2160 vdev_validate(vdev_t *vd) 2161 { 2162 spa_t *spa = vd->vdev_spa; 2163 taskq_t *tq = NULL; 2164 nvlist_t *label; 2165 uint64_t guid = 0, aux_guid = 0, top_guid; 2166 uint64_t state; 2167 nvlist_t *nvl; 2168 uint64_t txg; 2169 int children = vd->vdev_children; 2170 2171 if (vdev_validate_skip) 2172 return (0); 2173 2174 if (children > 0) { 2175 tq = taskq_create("vdev_validate", children, minclsyspri, 2176 children, children, TASKQ_PREPOPULATE); 2177 } 2178 2179 for (uint64_t c = 0; c < children; c++) { 2180 vdev_t *cvd = vd->vdev_child[c]; 2181 2182 if (tq == NULL || vdev_uses_zvols(cvd)) { 2183 vdev_validate_child(cvd); 2184 } else { 2185 VERIFY(taskq_dispatch(tq, vdev_validate_child, cvd, 2186 TQ_SLEEP) != TASKQID_INVALID); 2187 } 2188 } 2189 if (tq != NULL) { 2190 taskq_wait(tq); 2191 taskq_destroy(tq); 2192 } 2193 for (int c = 0; c < children; c++) { 2194 int error = vd->vdev_child[c]->vdev_validate_error; 2195 2196 if (error != 0) 2197 return (SET_ERROR(EBADF)); 2198 } 2199 2200 2201 /* 2202 * If the device has already failed, or was marked offline, don't do 2203 * any further validation. Otherwise, label I/O will fail and we will 2204 * overwrite the previous state. 2205 */ 2206 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 2207 return (0); 2208 2209 /* 2210 * If we are performing an extreme rewind, we allow for a label that 2211 * was modified at a point after the current txg. 2212 * If config lock is not held do not check for the txg. spa_sync could 2213 * be updating the vdev's label before updating spa_last_synced_txg. 2214 */ 2215 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 2216 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 2217 txg = UINT64_MAX; 2218 else 2219 txg = spa_last_synced_txg(spa); 2220 2221 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 2222 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2223 VDEV_AUX_BAD_LABEL); 2224 vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 2225 "txg %llu", (u_longlong_t)txg); 2226 return (0); 2227 } 2228 2229 /* 2230 * Determine if this vdev has been split off into another 2231 * pool. If so, then refuse to open it. 2232 */ 2233 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 2234 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 2235 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2236 VDEV_AUX_SPLIT_POOL); 2237 nvlist_free(label); 2238 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 2239 return (0); 2240 } 2241 2242 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 2243 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2244 VDEV_AUX_CORRUPT_DATA); 2245 nvlist_free(label); 2246 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2247 ZPOOL_CONFIG_POOL_GUID); 2248 return (0); 2249 } 2250 2251 /* 2252 * If config is not trusted then ignore the spa guid check. This is 2253 * necessary because if the machine crashed during a re-guid the new 2254 * guid might have been written to all of the vdev labels, but not the 2255 * cached config. The check will be performed again once we have the 2256 * trusted config from the MOS. 2257 */ 2258 if (spa->spa_trust_config && guid != spa_guid(spa)) { 2259 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2260 VDEV_AUX_CORRUPT_DATA); 2261 nvlist_free(label); 2262 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 2263 "match config (%llu != %llu)", (u_longlong_t)guid, 2264 (u_longlong_t)spa_guid(spa)); 2265 return (0); 2266 } 2267 2268 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 2269 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 2270 &aux_guid) != 0) 2271 aux_guid = 0; 2272 2273 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 2274 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2275 VDEV_AUX_CORRUPT_DATA); 2276 nvlist_free(label); 2277 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2278 ZPOOL_CONFIG_GUID); 2279 return (0); 2280 } 2281 2282 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 2283 != 0) { 2284 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2285 VDEV_AUX_CORRUPT_DATA); 2286 nvlist_free(label); 2287 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2288 ZPOOL_CONFIG_TOP_GUID); 2289 return (0); 2290 } 2291 2292 /* 2293 * If this vdev just became a top-level vdev because its sibling was 2294 * detached, it will have adopted the parent's vdev guid -- but the 2295 * label may or may not be on disk yet. Fortunately, either version 2296 * of the label will have the same top guid, so if we're a top-level 2297 * vdev, we can safely compare to that instead. 2298 * However, if the config comes from a cachefile that failed to update 2299 * after the detach, a top-level vdev will appear as a non top-level 2300 * vdev in the config. Also relax the constraints if we perform an 2301 * extreme rewind. 2302 * 2303 * If we split this vdev off instead, then we also check the 2304 * original pool's guid. We don't want to consider the vdev 2305 * corrupt if it is partway through a split operation. 2306 */ 2307 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 2308 boolean_t mismatch = B_FALSE; 2309 if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 2310 if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 2311 mismatch = B_TRUE; 2312 } else { 2313 if (vd->vdev_guid != top_guid && 2314 vd->vdev_top->vdev_guid != guid) 2315 mismatch = B_TRUE; 2316 } 2317 2318 if (mismatch) { 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: config guid " 2323 "doesn't match label guid"); 2324 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 2325 (u_longlong_t)vd->vdev_guid, 2326 (u_longlong_t)vd->vdev_top->vdev_guid); 2327 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 2328 "aux_guid %llu", (u_longlong_t)guid, 2329 (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 2330 return (0); 2331 } 2332 } 2333 2334 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 2335 &state) != 0) { 2336 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2337 VDEV_AUX_CORRUPT_DATA); 2338 nvlist_free(label); 2339 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 2340 ZPOOL_CONFIG_POOL_STATE); 2341 return (0); 2342 } 2343 2344 nvlist_free(label); 2345 2346 /* 2347 * If this is a verbatim import, no need to check the 2348 * state of the pool. 2349 */ 2350 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 2351 spa_load_state(spa) == SPA_LOAD_OPEN && 2352 state != POOL_STATE_ACTIVE) { 2353 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 2354 "for spa %s", (u_longlong_t)state, spa->spa_name); 2355 return (SET_ERROR(EBADF)); 2356 } 2357 2358 /* 2359 * If we were able to open and validate a vdev that was 2360 * previously marked permanently unavailable, clear that state 2361 * now. 2362 */ 2363 if (vd->vdev_not_present) 2364 vd->vdev_not_present = 0; 2365 2366 return (0); 2367 } 2368 2369 static void 2370 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 2371 { 2372 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 2373 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 2374 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 2375 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 2376 dvd->vdev_path, svd->vdev_path); 2377 spa_strfree(dvd->vdev_path); 2378 dvd->vdev_path = spa_strdup(svd->vdev_path); 2379 } 2380 } else if (svd->vdev_path != NULL) { 2381 dvd->vdev_path = spa_strdup(svd->vdev_path); 2382 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 2383 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 2384 } 2385 } 2386 2387 /* 2388 * Recursively copy vdev paths from one vdev to another. Source and destination 2389 * vdev trees must have same geometry otherwise return error. Intended to copy 2390 * paths from userland config into MOS config. 2391 */ 2392 int 2393 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 2394 { 2395 if ((svd->vdev_ops == &vdev_missing_ops) || 2396 (svd->vdev_ishole && dvd->vdev_ishole) || 2397 (dvd->vdev_ops == &vdev_indirect_ops)) 2398 return (0); 2399 2400 if (svd->vdev_ops != dvd->vdev_ops) { 2401 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 2402 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 2403 return (SET_ERROR(EINVAL)); 2404 } 2405 2406 if (svd->vdev_guid != dvd->vdev_guid) { 2407 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 2408 "%llu)", (u_longlong_t)svd->vdev_guid, 2409 (u_longlong_t)dvd->vdev_guid); 2410 return (SET_ERROR(EINVAL)); 2411 } 2412 2413 if (svd->vdev_children != dvd->vdev_children) { 2414 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 2415 "%llu != %llu", (u_longlong_t)svd->vdev_children, 2416 (u_longlong_t)dvd->vdev_children); 2417 return (SET_ERROR(EINVAL)); 2418 } 2419 2420 for (uint64_t i = 0; i < svd->vdev_children; i++) { 2421 int error = vdev_copy_path_strict(svd->vdev_child[i], 2422 dvd->vdev_child[i]); 2423 if (error != 0) 2424 return (error); 2425 } 2426 2427 if (svd->vdev_ops->vdev_op_leaf) 2428 vdev_copy_path_impl(svd, dvd); 2429 2430 return (0); 2431 } 2432 2433 static void 2434 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 2435 { 2436 ASSERT(stvd->vdev_top == stvd); 2437 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 2438 2439 for (uint64_t i = 0; i < dvd->vdev_children; i++) { 2440 vdev_copy_path_search(stvd, dvd->vdev_child[i]); 2441 } 2442 2443 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 2444 return; 2445 2446 /* 2447 * The idea here is that while a vdev can shift positions within 2448 * a top vdev (when replacing, attaching mirror, etc.) it cannot 2449 * step outside of it. 2450 */ 2451 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 2452 2453 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 2454 return; 2455 2456 ASSERT(vd->vdev_ops->vdev_op_leaf); 2457 2458 vdev_copy_path_impl(vd, dvd); 2459 } 2460 2461 /* 2462 * Recursively copy vdev paths from one root vdev to another. Source and 2463 * destination vdev trees may differ in geometry. For each destination leaf 2464 * vdev, search a vdev with the same guid and top vdev id in the source. 2465 * Intended to copy paths from userland config into MOS config. 2466 */ 2467 void 2468 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 2469 { 2470 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 2471 ASSERT(srvd->vdev_ops == &vdev_root_ops); 2472 ASSERT(drvd->vdev_ops == &vdev_root_ops); 2473 2474 for (uint64_t i = 0; i < children; i++) { 2475 vdev_copy_path_search(srvd->vdev_child[i], 2476 drvd->vdev_child[i]); 2477 } 2478 } 2479 2480 /* 2481 * Close a virtual device. 2482 */ 2483 void 2484 vdev_close(vdev_t *vd) 2485 { 2486 vdev_t *pvd = vd->vdev_parent; 2487 spa_t *spa __maybe_unused = vd->vdev_spa; 2488 2489 ASSERT(vd != NULL); 2490 ASSERT(vd->vdev_open_thread == curthread || 2491 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2492 2493 /* 2494 * If our parent is reopening, then we are as well, unless we are 2495 * going offline. 2496 */ 2497 if (pvd != NULL && pvd->vdev_reopening) 2498 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2499 2500 vd->vdev_ops->vdev_op_close(vd); 2501 2502 vdev_cache_purge(vd); 2503 2504 /* 2505 * We record the previous state before we close it, so that if we are 2506 * doing a reopen(), we don't generate FMA ereports if we notice that 2507 * it's still faulted. 2508 */ 2509 vd->vdev_prevstate = vd->vdev_state; 2510 2511 if (vd->vdev_offline) 2512 vd->vdev_state = VDEV_STATE_OFFLINE; 2513 else 2514 vd->vdev_state = VDEV_STATE_CLOSED; 2515 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2516 } 2517 2518 void 2519 vdev_hold(vdev_t *vd) 2520 { 2521 spa_t *spa = vd->vdev_spa; 2522 2523 ASSERT(spa_is_root(spa)); 2524 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2525 return; 2526 2527 for (int c = 0; c < vd->vdev_children; c++) 2528 vdev_hold(vd->vdev_child[c]); 2529 2530 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_hold != NULL) 2531 vd->vdev_ops->vdev_op_hold(vd); 2532 } 2533 2534 void 2535 vdev_rele(vdev_t *vd) 2536 { 2537 ASSERT(spa_is_root(vd->vdev_spa)); 2538 for (int c = 0; c < vd->vdev_children; c++) 2539 vdev_rele(vd->vdev_child[c]); 2540 2541 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_rele != NULL) 2542 vd->vdev_ops->vdev_op_rele(vd); 2543 } 2544 2545 /* 2546 * Reopen all interior vdevs and any unopened leaves. We don't actually 2547 * reopen leaf vdevs which had previously been opened as they might deadlock 2548 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2549 * If the leaf has never been opened then open it, as usual. 2550 */ 2551 void 2552 vdev_reopen(vdev_t *vd) 2553 { 2554 spa_t *spa = vd->vdev_spa; 2555 2556 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2557 2558 /* set the reopening flag unless we're taking the vdev offline */ 2559 vd->vdev_reopening = !vd->vdev_offline; 2560 vdev_close(vd); 2561 (void) vdev_open(vd); 2562 2563 /* 2564 * Call vdev_validate() here to make sure we have the same device. 2565 * Otherwise, a device with an invalid label could be successfully 2566 * opened in response to vdev_reopen(). 2567 */ 2568 if (vd->vdev_aux) { 2569 (void) vdev_validate_aux(vd); 2570 if (vdev_readable(vd) && vdev_writeable(vd) && 2571 vd->vdev_aux == &spa->spa_l2cache) { 2572 /* 2573 * In case the vdev is present we should evict all ARC 2574 * buffers and pointers to log blocks and reclaim their 2575 * space before restoring its contents to L2ARC. 2576 */ 2577 if (l2arc_vdev_present(vd)) { 2578 l2arc_rebuild_vdev(vd, B_TRUE); 2579 } else { 2580 l2arc_add_vdev(spa, vd); 2581 } 2582 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); 2583 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); 2584 } 2585 } else { 2586 (void) vdev_validate(vd); 2587 } 2588 2589 /* 2590 * Reassess parent vdev's health. 2591 */ 2592 vdev_propagate_state(vd); 2593 } 2594 2595 int 2596 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2597 { 2598 int error; 2599 2600 /* 2601 * Normally, partial opens (e.g. of a mirror) are allowed. 2602 * For a create, however, we want to fail the request if 2603 * there are any components we can't open. 2604 */ 2605 error = vdev_open(vd); 2606 2607 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2608 vdev_close(vd); 2609 return (error ? error : SET_ERROR(ENXIO)); 2610 } 2611 2612 /* 2613 * Recursively load DTLs and initialize all labels. 2614 */ 2615 if ((error = vdev_dtl_load(vd)) != 0 || 2616 (error = vdev_label_init(vd, txg, isreplacing ? 2617 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2618 vdev_close(vd); 2619 return (error); 2620 } 2621 2622 return (0); 2623 } 2624 2625 void 2626 vdev_metaslab_set_size(vdev_t *vd) 2627 { 2628 uint64_t asize = vd->vdev_asize; 2629 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2630 uint64_t ms_shift; 2631 2632 /* 2633 * There are two dimensions to the metaslab sizing calculation: 2634 * the size of the metaslab and the count of metaslabs per vdev. 2635 * 2636 * The default values used below are a good balance between memory 2637 * usage (larger metaslab size means more memory needed for loaded 2638 * metaslabs; more metaslabs means more memory needed for the 2639 * metaslab_t structs), metaslab load time (larger metaslabs take 2640 * longer to load), and metaslab sync time (more metaslabs means 2641 * more time spent syncing all of them). 2642 * 2643 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2644 * The range of the dimensions are as follows: 2645 * 2646 * 2^29 <= ms_size <= 2^34 2647 * 16 <= ms_count <= 131,072 2648 * 2649 * On the lower end of vdev sizes, we aim for metaslabs sizes of 2650 * at least 512MB (2^29) to minimize fragmentation effects when 2651 * testing with smaller devices. However, the count constraint 2652 * of at least 16 metaslabs will override this minimum size goal. 2653 * 2654 * On the upper end of vdev sizes, we aim for a maximum metaslab 2655 * size of 16GB. However, we will cap the total count to 2^17 2656 * metaslabs to keep our memory footprint in check and let the 2657 * metaslab size grow from there if that limit is hit. 2658 * 2659 * The net effect of applying above constrains is summarized below. 2660 * 2661 * vdev size metaslab count 2662 * --------------|----------------- 2663 * < 8GB ~16 2664 * 8GB - 100GB one per 512MB 2665 * 100GB - 3TB ~200 2666 * 3TB - 2PB one per 16GB 2667 * > 2PB ~131,072 2668 * -------------------------------- 2669 * 2670 * Finally, note that all of the above calculate the initial 2671 * number of metaslabs. Expanding a top-level vdev will result 2672 * in additional metaslabs being allocated making it possible 2673 * to exceed the zfs_vdev_ms_count_limit. 2674 */ 2675 2676 if (ms_count < zfs_vdev_min_ms_count) 2677 ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2678 else if (ms_count > zfs_vdev_default_ms_count) 2679 ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2680 else 2681 ms_shift = zfs_vdev_default_ms_shift; 2682 2683 if (ms_shift < SPA_MAXBLOCKSHIFT) { 2684 ms_shift = SPA_MAXBLOCKSHIFT; 2685 } else if (ms_shift > zfs_vdev_max_ms_shift) { 2686 ms_shift = zfs_vdev_max_ms_shift; 2687 /* cap the total count to constrain memory footprint */ 2688 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2689 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 2690 } 2691 2692 vd->vdev_ms_shift = ms_shift; 2693 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2694 } 2695 2696 void 2697 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2698 { 2699 ASSERT(vd == vd->vdev_top); 2700 /* indirect vdevs don't have metaslabs or dtls */ 2701 ASSERT(vdev_is_concrete(vd) || flags == 0); 2702 ASSERT(ISP2(flags)); 2703 ASSERT(spa_writeable(vd->vdev_spa)); 2704 2705 if (flags & VDD_METASLAB) 2706 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2707 2708 if (flags & VDD_DTL) 2709 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2710 2711 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2712 } 2713 2714 void 2715 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 2716 { 2717 for (int c = 0; c < vd->vdev_children; c++) 2718 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 2719 2720 if (vd->vdev_ops->vdev_op_leaf) 2721 vdev_dirty(vd->vdev_top, flags, vd, txg); 2722 } 2723 2724 /* 2725 * DTLs. 2726 * 2727 * A vdev's DTL (dirty time log) is the set of transaction groups for which 2728 * the vdev has less than perfect replication. There are four kinds of DTL: 2729 * 2730 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 2731 * 2732 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 2733 * 2734 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 2735 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 2736 * txgs that was scrubbed. 2737 * 2738 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 2739 * persistent errors or just some device being offline. 2740 * Unlike the other three, the DTL_OUTAGE map is not generally 2741 * maintained; it's only computed when needed, typically to 2742 * determine whether a device can be detached. 2743 * 2744 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 2745 * either has the data or it doesn't. 2746 * 2747 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 2748 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 2749 * if any child is less than fully replicated, then so is its parent. 2750 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 2751 * comprising only those txgs which appear in 'maxfaults' or more children; 2752 * those are the txgs we don't have enough replication to read. For example, 2753 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 2754 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 2755 * two child DTL_MISSING maps. 2756 * 2757 * It should be clear from the above that to compute the DTLs and outage maps 2758 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 2759 * Therefore, that is all we keep on disk. When loading the pool, or after 2760 * a configuration change, we generate all other DTLs from first principles. 2761 */ 2762 void 2763 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2764 { 2765 range_tree_t *rt = vd->vdev_dtl[t]; 2766 2767 ASSERT(t < DTL_TYPES); 2768 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2769 ASSERT(spa_writeable(vd->vdev_spa)); 2770 2771 mutex_enter(&vd->vdev_dtl_lock); 2772 if (!range_tree_contains(rt, txg, size)) 2773 range_tree_add(rt, txg, size); 2774 mutex_exit(&vd->vdev_dtl_lock); 2775 } 2776 2777 boolean_t 2778 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2779 { 2780 range_tree_t *rt = vd->vdev_dtl[t]; 2781 boolean_t dirty = B_FALSE; 2782 2783 ASSERT(t < DTL_TYPES); 2784 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2785 2786 /* 2787 * While we are loading the pool, the DTLs have not been loaded yet. 2788 * This isn't a problem but it can result in devices being tried 2789 * which are known to not have the data. In which case, the import 2790 * is relying on the checksum to ensure that we get the right data. 2791 * Note that while importing we are only reading the MOS, which is 2792 * always checksummed. 2793 */ 2794 mutex_enter(&vd->vdev_dtl_lock); 2795 if (!range_tree_is_empty(rt)) 2796 dirty = range_tree_contains(rt, txg, size); 2797 mutex_exit(&vd->vdev_dtl_lock); 2798 2799 return (dirty); 2800 } 2801 2802 boolean_t 2803 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 2804 { 2805 range_tree_t *rt = vd->vdev_dtl[t]; 2806 boolean_t empty; 2807 2808 mutex_enter(&vd->vdev_dtl_lock); 2809 empty = range_tree_is_empty(rt); 2810 mutex_exit(&vd->vdev_dtl_lock); 2811 2812 return (empty); 2813 } 2814 2815 /* 2816 * Check if the txg falls within the range which must be 2817 * resilvered. DVAs outside this range can always be skipped. 2818 */ 2819 boolean_t 2820 vdev_default_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, 2821 uint64_t phys_birth) 2822 { 2823 /* Set by sequential resilver. */ 2824 if (phys_birth == TXG_UNKNOWN) 2825 return (B_TRUE); 2826 2827 return (vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1)); 2828 } 2829 2830 /* 2831 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered. 2832 */ 2833 boolean_t 2834 vdev_dtl_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, 2835 uint64_t phys_birth) 2836 { 2837 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2838 2839 if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2840 vd->vdev_ops->vdev_op_leaf) 2841 return (B_TRUE); 2842 2843 return (vd->vdev_ops->vdev_op_need_resilver(vd, dva, psize, 2844 phys_birth)); 2845 } 2846 2847 /* 2848 * Returns the lowest txg in the DTL range. 2849 */ 2850 static uint64_t 2851 vdev_dtl_min(vdev_t *vd) 2852 { 2853 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2854 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2855 ASSERT0(vd->vdev_children); 2856 2857 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); 2858 } 2859 2860 /* 2861 * Returns the highest txg in the DTL. 2862 */ 2863 static uint64_t 2864 vdev_dtl_max(vdev_t *vd) 2865 { 2866 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2867 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2868 ASSERT0(vd->vdev_children); 2869 2870 return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); 2871 } 2872 2873 /* 2874 * Determine if a resilvering vdev should remove any DTL entries from 2875 * its range. If the vdev was resilvering for the entire duration of the 2876 * scan then it should excise that range from its DTLs. Otherwise, this 2877 * vdev is considered partially resilvered and should leave its DTL 2878 * entries intact. The comment in vdev_dtl_reassess() describes how we 2879 * excise the DTLs. 2880 */ 2881 static boolean_t 2882 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done) 2883 { 2884 ASSERT0(vd->vdev_children); 2885 2886 if (vd->vdev_state < VDEV_STATE_DEGRADED) 2887 return (B_FALSE); 2888 2889 if (vd->vdev_resilver_deferred) 2890 return (B_FALSE); 2891 2892 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2893 return (B_TRUE); 2894 2895 if (rebuild_done) { 2896 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; 2897 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; 2898 2899 /* Rebuild not initiated by attach */ 2900 if (vd->vdev_rebuild_txg == 0) 2901 return (B_TRUE); 2902 2903 /* 2904 * When a rebuild completes without error then all missing data 2905 * up to the rebuild max txg has been reconstructed and the DTL 2906 * is eligible for excision. 2907 */ 2908 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE && 2909 vdev_dtl_max(vd) <= vrp->vrp_max_txg) { 2910 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd)); 2911 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg); 2912 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg); 2913 return (B_TRUE); 2914 } 2915 } else { 2916 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan; 2917 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys; 2918 2919 /* Resilver not initiated by attach */ 2920 if (vd->vdev_resilver_txg == 0) 2921 return (B_TRUE); 2922 2923 /* 2924 * When a resilver is initiated the scan will assign the 2925 * scn_max_txg value to the highest txg value that exists 2926 * in all DTLs. If this device's max DTL is not part of this 2927 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg] 2928 * then it is not eligible for excision. 2929 */ 2930 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2931 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd)); 2932 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg); 2933 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg); 2934 return (B_TRUE); 2935 } 2936 } 2937 2938 return (B_FALSE); 2939 } 2940 2941 /* 2942 * Reassess DTLs after a config change or scrub completion. If txg == 0 no 2943 * write operations will be issued to the pool. 2944 */ 2945 void 2946 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, 2947 boolean_t scrub_done, boolean_t rebuild_done) 2948 { 2949 spa_t *spa = vd->vdev_spa; 2950 avl_tree_t reftree; 2951 int minref; 2952 2953 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2954 2955 for (int c = 0; c < vd->vdev_children; c++) 2956 vdev_dtl_reassess(vd->vdev_child[c], txg, 2957 scrub_txg, scrub_done, rebuild_done); 2958 2959 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 2960 return; 2961 2962 if (vd->vdev_ops->vdev_op_leaf) { 2963 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2964 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; 2965 boolean_t check_excise = B_FALSE; 2966 boolean_t wasempty = B_TRUE; 2967 2968 mutex_enter(&vd->vdev_dtl_lock); 2969 2970 /* 2971 * If requested, pretend the scan or rebuild completed cleanly. 2972 */ 2973 if (zfs_scan_ignore_errors) { 2974 if (scn != NULL) 2975 scn->scn_phys.scn_errors = 0; 2976 if (vr != NULL) 2977 vr->vr_rebuild_phys.vrp_errors = 0; 2978 } 2979 2980 if (scrub_txg != 0 && 2981 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 2982 wasempty = B_FALSE; 2983 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d " 2984 "dtl:%llu/%llu errors:%llu", 2985 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg, 2986 (u_longlong_t)scrub_txg, spa->spa_scrub_started, 2987 (u_longlong_t)vdev_dtl_min(vd), 2988 (u_longlong_t)vdev_dtl_max(vd), 2989 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0)); 2990 } 2991 2992 /* 2993 * If we've completed a scrub/resilver or a rebuild cleanly 2994 * then determine if this vdev should remove any DTLs. We 2995 * only want to excise regions on vdevs that were available 2996 * during the entire duration of this scan. 2997 */ 2998 if (rebuild_done && 2999 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) { 3000 check_excise = B_TRUE; 3001 } else { 3002 if (spa->spa_scrub_started || 3003 (scn != NULL && scn->scn_phys.scn_errors == 0)) { 3004 check_excise = B_TRUE; 3005 } 3006 } 3007 3008 if (scrub_txg && check_excise && 3009 vdev_dtl_should_excise(vd, rebuild_done)) { 3010 /* 3011 * We completed a scrub, resilver or rebuild up to 3012 * scrub_txg. If we did it without rebooting, then 3013 * the scrub dtl will be valid, so excise the old 3014 * region and fold in the scrub dtl. Otherwise, 3015 * leave the dtl as-is if there was an error. 3016 * 3017 * There's little trick here: to excise the beginning 3018 * of the DTL_MISSING map, we put it into a reference 3019 * tree and then add a segment with refcnt -1 that 3020 * covers the range [0, scrub_txg). This means 3021 * that each txg in that range has refcnt -1 or 0. 3022 * We then add DTL_SCRUB with a refcnt of 2, so that 3023 * entries in the range [0, scrub_txg) will have a 3024 * positive refcnt -- either 1 or 2. We then convert 3025 * the reference tree into the new DTL_MISSING map. 3026 */ 3027 space_reftree_create(&reftree); 3028 space_reftree_add_map(&reftree, 3029 vd->vdev_dtl[DTL_MISSING], 1); 3030 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 3031 space_reftree_add_map(&reftree, 3032 vd->vdev_dtl[DTL_SCRUB], 2); 3033 space_reftree_generate_map(&reftree, 3034 vd->vdev_dtl[DTL_MISSING], 1); 3035 space_reftree_destroy(&reftree); 3036 3037 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 3038 zfs_dbgmsg("update DTL_MISSING:%llu/%llu", 3039 (u_longlong_t)vdev_dtl_min(vd), 3040 (u_longlong_t)vdev_dtl_max(vd)); 3041 } else if (!wasempty) { 3042 zfs_dbgmsg("DTL_MISSING is now empty"); 3043 } 3044 } 3045 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 3046 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 3047 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 3048 if (scrub_done) 3049 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 3050 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 3051 if (!vdev_readable(vd)) 3052 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 3053 else 3054 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 3055 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 3056 3057 /* 3058 * If the vdev was resilvering or rebuilding and no longer 3059 * has any DTLs then reset the appropriate flag and dirty 3060 * the top level so that we persist the change. 3061 */ 3062 if (txg != 0 && 3063 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 3064 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) { 3065 if (vd->vdev_rebuild_txg != 0) { 3066 vd->vdev_rebuild_txg = 0; 3067 vdev_config_dirty(vd->vdev_top); 3068 } else if (vd->vdev_resilver_txg != 0) { 3069 vd->vdev_resilver_txg = 0; 3070 vdev_config_dirty(vd->vdev_top); 3071 } 3072 } 3073 3074 mutex_exit(&vd->vdev_dtl_lock); 3075 3076 if (txg != 0) 3077 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 3078 return; 3079 } 3080 3081 mutex_enter(&vd->vdev_dtl_lock); 3082 for (int t = 0; t < DTL_TYPES; t++) { 3083 /* account for child's outage in parent's missing map */ 3084 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 3085 if (t == DTL_SCRUB) 3086 continue; /* leaf vdevs only */ 3087 if (t == DTL_PARTIAL) 3088 minref = 1; /* i.e. non-zero */ 3089 else if (vdev_get_nparity(vd) != 0) 3090 minref = vdev_get_nparity(vd) + 1; /* RAID-Z, dRAID */ 3091 else 3092 minref = vd->vdev_children; /* any kind of mirror */ 3093 space_reftree_create(&reftree); 3094 for (int c = 0; c < vd->vdev_children; c++) { 3095 vdev_t *cvd = vd->vdev_child[c]; 3096 mutex_enter(&cvd->vdev_dtl_lock); 3097 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 3098 mutex_exit(&cvd->vdev_dtl_lock); 3099 } 3100 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 3101 space_reftree_destroy(&reftree); 3102 } 3103 mutex_exit(&vd->vdev_dtl_lock); 3104 } 3105 3106 int 3107 vdev_dtl_load(vdev_t *vd) 3108 { 3109 spa_t *spa = vd->vdev_spa; 3110 objset_t *mos = spa->spa_meta_objset; 3111 range_tree_t *rt; 3112 int error = 0; 3113 3114 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 3115 ASSERT(vdev_is_concrete(vd)); 3116 3117 error = space_map_open(&vd->vdev_dtl_sm, mos, 3118 vd->vdev_dtl_object, 0, -1ULL, 0); 3119 if (error) 3120 return (error); 3121 ASSERT(vd->vdev_dtl_sm != NULL); 3122 3123 rt = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 3124 error = space_map_load(vd->vdev_dtl_sm, rt, SM_ALLOC); 3125 if (error == 0) { 3126 mutex_enter(&vd->vdev_dtl_lock); 3127 range_tree_walk(rt, range_tree_add, 3128 vd->vdev_dtl[DTL_MISSING]); 3129 mutex_exit(&vd->vdev_dtl_lock); 3130 } 3131 3132 range_tree_vacate(rt, NULL, NULL); 3133 range_tree_destroy(rt); 3134 3135 return (error); 3136 } 3137 3138 for (int c = 0; c < vd->vdev_children; c++) { 3139 error = vdev_dtl_load(vd->vdev_child[c]); 3140 if (error != 0) 3141 break; 3142 } 3143 3144 return (error); 3145 } 3146 3147 static void 3148 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 3149 { 3150 spa_t *spa = vd->vdev_spa; 3151 objset_t *mos = spa->spa_meta_objset; 3152 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 3153 const char *string; 3154 3155 ASSERT(alloc_bias != VDEV_BIAS_NONE); 3156 3157 string = 3158 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 3159 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 3160 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 3161 3162 ASSERT(string != NULL); 3163 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 3164 1, strlen(string) + 1, string, tx)); 3165 3166 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 3167 spa_activate_allocation_classes(spa, tx); 3168 } 3169 } 3170 3171 void 3172 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 3173 { 3174 spa_t *spa = vd->vdev_spa; 3175 3176 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 3177 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 3178 zapobj, tx)); 3179 } 3180 3181 uint64_t 3182 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 3183 { 3184 spa_t *spa = vd->vdev_spa; 3185 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 3186 DMU_OT_NONE, 0, tx); 3187 3188 ASSERT(zap != 0); 3189 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 3190 zap, tx)); 3191 3192 return (zap); 3193 } 3194 3195 void 3196 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 3197 { 3198 if (vd->vdev_ops != &vdev_hole_ops && 3199 vd->vdev_ops != &vdev_missing_ops && 3200 vd->vdev_ops != &vdev_root_ops && 3201 !vd->vdev_top->vdev_removing) { 3202 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 3203 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 3204 } 3205 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 3206 vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 3207 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 3208 vdev_zap_allocation_data(vd, tx); 3209 } 3210 } 3211 3212 for (uint64_t i = 0; i < vd->vdev_children; i++) { 3213 vdev_construct_zaps(vd->vdev_child[i], tx); 3214 } 3215 } 3216 3217 static void 3218 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 3219 { 3220 spa_t *spa = vd->vdev_spa; 3221 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 3222 objset_t *mos = spa->spa_meta_objset; 3223 range_tree_t *rtsync; 3224 dmu_tx_t *tx; 3225 uint64_t object = space_map_object(vd->vdev_dtl_sm); 3226 3227 ASSERT(vdev_is_concrete(vd)); 3228 ASSERT(vd->vdev_ops->vdev_op_leaf); 3229 3230 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3231 3232 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 3233 mutex_enter(&vd->vdev_dtl_lock); 3234 space_map_free(vd->vdev_dtl_sm, tx); 3235 space_map_close(vd->vdev_dtl_sm); 3236 vd->vdev_dtl_sm = NULL; 3237 mutex_exit(&vd->vdev_dtl_lock); 3238 3239 /* 3240 * We only destroy the leaf ZAP for detached leaves or for 3241 * removed log devices. Removed data devices handle leaf ZAP 3242 * cleanup later, once cancellation is no longer possible. 3243 */ 3244 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 3245 vd->vdev_top->vdev_islog)) { 3246 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 3247 vd->vdev_leaf_zap = 0; 3248 } 3249 3250 dmu_tx_commit(tx); 3251 return; 3252 } 3253 3254 if (vd->vdev_dtl_sm == NULL) { 3255 uint64_t new_object; 3256 3257 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); 3258 VERIFY3U(new_object, !=, 0); 3259 3260 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 3261 0, -1ULL, 0)); 3262 ASSERT(vd->vdev_dtl_sm != NULL); 3263 } 3264 3265 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 3266 3267 mutex_enter(&vd->vdev_dtl_lock); 3268 range_tree_walk(rt, range_tree_add, rtsync); 3269 mutex_exit(&vd->vdev_dtl_lock); 3270 3271 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); 3272 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 3273 range_tree_vacate(rtsync, NULL, NULL); 3274 3275 range_tree_destroy(rtsync); 3276 3277 /* 3278 * If the object for the space map has changed then dirty 3279 * the top level so that we update the config. 3280 */ 3281 if (object != space_map_object(vd->vdev_dtl_sm)) { 3282 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 3283 "new object %llu", (u_longlong_t)txg, spa_name(spa), 3284 (u_longlong_t)object, 3285 (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 3286 vdev_config_dirty(vd->vdev_top); 3287 } 3288 3289 dmu_tx_commit(tx); 3290 } 3291 3292 /* 3293 * Determine whether the specified vdev can be offlined/detached/removed 3294 * without losing data. 3295 */ 3296 boolean_t 3297 vdev_dtl_required(vdev_t *vd) 3298 { 3299 spa_t *spa = vd->vdev_spa; 3300 vdev_t *tvd = vd->vdev_top; 3301 uint8_t cant_read = vd->vdev_cant_read; 3302 boolean_t required; 3303 3304 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3305 3306 if (vd == spa->spa_root_vdev || vd == tvd) 3307 return (B_TRUE); 3308 3309 /* 3310 * Temporarily mark the device as unreadable, and then determine 3311 * whether this results in any DTL outages in the top-level vdev. 3312 * If not, we can safely offline/detach/remove the device. 3313 */ 3314 vd->vdev_cant_read = B_TRUE; 3315 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); 3316 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 3317 vd->vdev_cant_read = cant_read; 3318 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); 3319 3320 if (!required && zio_injection_enabled) { 3321 required = !!zio_handle_device_injection(vd, NULL, 3322 SET_ERROR(ECHILD)); 3323 } 3324 3325 return (required); 3326 } 3327 3328 /* 3329 * Determine if resilver is needed, and if so the txg range. 3330 */ 3331 boolean_t 3332 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 3333 { 3334 boolean_t needed = B_FALSE; 3335 uint64_t thismin = UINT64_MAX; 3336 uint64_t thismax = 0; 3337 3338 if (vd->vdev_children == 0) { 3339 mutex_enter(&vd->vdev_dtl_lock); 3340 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 3341 vdev_writeable(vd)) { 3342 3343 thismin = vdev_dtl_min(vd); 3344 thismax = vdev_dtl_max(vd); 3345 needed = B_TRUE; 3346 } 3347 mutex_exit(&vd->vdev_dtl_lock); 3348 } else { 3349 for (int c = 0; c < vd->vdev_children; c++) { 3350 vdev_t *cvd = vd->vdev_child[c]; 3351 uint64_t cmin, cmax; 3352 3353 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 3354 thismin = MIN(thismin, cmin); 3355 thismax = MAX(thismax, cmax); 3356 needed = B_TRUE; 3357 } 3358 } 3359 } 3360 3361 if (needed && minp) { 3362 *minp = thismin; 3363 *maxp = thismax; 3364 } 3365 return (needed); 3366 } 3367 3368 /* 3369 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj 3370 * will contain either the checkpoint spacemap object or zero if none exists. 3371 * All other errors are returned to the caller. 3372 */ 3373 int 3374 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj) 3375 { 3376 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 3377 3378 if (vd->vdev_top_zap == 0) { 3379 *sm_obj = 0; 3380 return (0); 3381 } 3382 3383 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 3384 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj); 3385 if (error == ENOENT) { 3386 *sm_obj = 0; 3387 error = 0; 3388 } 3389 3390 return (error); 3391 } 3392 3393 int 3394 vdev_load(vdev_t *vd) 3395 { 3396 int children = vd->vdev_children; 3397 int error = 0; 3398 taskq_t *tq = NULL; 3399 3400 /* 3401 * It's only worthwhile to use the taskq for the root vdev, because the 3402 * slow part is metaslab_init, and that only happens for top-level 3403 * vdevs. 3404 */ 3405 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_children > 0) { 3406 tq = taskq_create("vdev_load", children, minclsyspri, 3407 children, children, TASKQ_PREPOPULATE); 3408 } 3409 3410 /* 3411 * Recursively load all children. 3412 */ 3413 for (int c = 0; c < vd->vdev_children; c++) { 3414 vdev_t *cvd = vd->vdev_child[c]; 3415 3416 if (tq == NULL || vdev_uses_zvols(cvd)) { 3417 cvd->vdev_load_error = vdev_load(cvd); 3418 } else { 3419 VERIFY(taskq_dispatch(tq, vdev_load_child, 3420 cvd, TQ_SLEEP) != TASKQID_INVALID); 3421 } 3422 } 3423 3424 if (tq != NULL) { 3425 taskq_wait(tq); 3426 taskq_destroy(tq); 3427 } 3428 3429 for (int c = 0; c < vd->vdev_children; c++) { 3430 int error = vd->vdev_child[c]->vdev_load_error; 3431 3432 if (error != 0) 3433 return (error); 3434 } 3435 3436 vdev_set_deflate_ratio(vd); 3437 3438 /* 3439 * On spa_load path, grab the allocation bias from our zap 3440 */ 3441 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 3442 spa_t *spa = vd->vdev_spa; 3443 char bias_str[64]; 3444 3445 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 3446 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 3447 bias_str); 3448 if (error == 0) { 3449 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 3450 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 3451 } else if (error != ENOENT) { 3452 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3453 VDEV_AUX_CORRUPT_DATA); 3454 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) " 3455 "failed [error=%d]", vd->vdev_top_zap, error); 3456 return (error); 3457 } 3458 } 3459 3460 /* 3461 * Load any rebuild state from the top-level vdev zap. 3462 */ 3463 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 3464 error = vdev_rebuild_load(vd); 3465 if (error && error != ENOTSUP) { 3466 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3467 VDEV_AUX_CORRUPT_DATA); 3468 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load " 3469 "failed [error=%d]", error); 3470 return (error); 3471 } 3472 } 3473 3474 /* 3475 * If this is a top-level vdev, initialize its metaslabs. 3476 */ 3477 if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 3478 vdev_metaslab_group_create(vd); 3479 3480 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 3481 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3482 VDEV_AUX_CORRUPT_DATA); 3483 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 3484 "asize=%llu", (u_longlong_t)vd->vdev_ashift, 3485 (u_longlong_t)vd->vdev_asize); 3486 return (SET_ERROR(ENXIO)); 3487 } 3488 3489 error = vdev_metaslab_init(vd, 0); 3490 if (error != 0) { 3491 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 3492 "[error=%d]", error); 3493 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3494 VDEV_AUX_CORRUPT_DATA); 3495 return (error); 3496 } 3497 3498 uint64_t checkpoint_sm_obj; 3499 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj); 3500 if (error == 0 && checkpoint_sm_obj != 0) { 3501 objset_t *mos = spa_meta_objset(vd->vdev_spa); 3502 ASSERT(vd->vdev_asize != 0); 3503 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 3504 3505 error = space_map_open(&vd->vdev_checkpoint_sm, 3506 mos, checkpoint_sm_obj, 0, vd->vdev_asize, 3507 vd->vdev_ashift); 3508 if (error != 0) { 3509 vdev_dbgmsg(vd, "vdev_load: space_map_open " 3510 "failed for checkpoint spacemap (obj %llu) " 3511 "[error=%d]", 3512 (u_longlong_t)checkpoint_sm_obj, error); 3513 return (error); 3514 } 3515 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 3516 3517 /* 3518 * Since the checkpoint_sm contains free entries 3519 * exclusively we can use space_map_allocated() to 3520 * indicate the cumulative checkpointed space that 3521 * has been freed. 3522 */ 3523 vd->vdev_stat.vs_checkpoint_space = 3524 -space_map_allocated(vd->vdev_checkpoint_sm); 3525 vd->vdev_spa->spa_checkpoint_info.sci_dspace += 3526 vd->vdev_stat.vs_checkpoint_space; 3527 } else if (error != 0) { 3528 vdev_dbgmsg(vd, "vdev_load: failed to retrieve " 3529 "checkpoint space map object from vdev ZAP " 3530 "[error=%d]", error); 3531 return (error); 3532 } 3533 } 3534 3535 /* 3536 * If this is a leaf vdev, load its DTL. 3537 */ 3538 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 3539 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3540 VDEV_AUX_CORRUPT_DATA); 3541 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 3542 "[error=%d]", error); 3543 return (error); 3544 } 3545 3546 uint64_t obsolete_sm_object; 3547 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object); 3548 if (error == 0 && obsolete_sm_object != 0) { 3549 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3550 ASSERT(vd->vdev_asize != 0); 3551 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 3552 3553 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 3554 obsolete_sm_object, 0, vd->vdev_asize, 0))) { 3555 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3556 VDEV_AUX_CORRUPT_DATA); 3557 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 3558 "obsolete spacemap (obj %llu) [error=%d]", 3559 (u_longlong_t)obsolete_sm_object, error); 3560 return (error); 3561 } 3562 } else if (error != 0) { 3563 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete " 3564 "space map object from vdev ZAP [error=%d]", error); 3565 return (error); 3566 } 3567 3568 return (0); 3569 } 3570 3571 /* 3572 * The special vdev case is used for hot spares and l2cache devices. Its 3573 * sole purpose it to set the vdev state for the associated vdev. To do this, 3574 * we make sure that we can open the underlying device, then try to read the 3575 * label, and make sure that the label is sane and that it hasn't been 3576 * repurposed to another pool. 3577 */ 3578 int 3579 vdev_validate_aux(vdev_t *vd) 3580 { 3581 nvlist_t *label; 3582 uint64_t guid, version; 3583 uint64_t state; 3584 3585 if (!vdev_readable(vd)) 3586 return (0); 3587 3588 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 3589 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3590 VDEV_AUX_CORRUPT_DATA); 3591 return (-1); 3592 } 3593 3594 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3595 !SPA_VERSION_IS_SUPPORTED(version) || 3596 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 3597 guid != vd->vdev_guid || 3598 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 3599 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3600 VDEV_AUX_CORRUPT_DATA); 3601 nvlist_free(label); 3602 return (-1); 3603 } 3604 3605 /* 3606 * We don't actually check the pool state here. If it's in fact in 3607 * use by another pool, we update this fact on the fly when requested. 3608 */ 3609 nvlist_free(label); 3610 return (0); 3611 } 3612 3613 static void 3614 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) 3615 { 3616 objset_t *mos = spa_meta_objset(vd->vdev_spa); 3617 3618 if (vd->vdev_top_zap == 0) 3619 return; 3620 3621 uint64_t object = 0; 3622 int err = zap_lookup(mos, vd->vdev_top_zap, 3623 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); 3624 if (err == ENOENT) 3625 return; 3626 VERIFY0(err); 3627 3628 VERIFY0(dmu_object_free(mos, object, tx)); 3629 VERIFY0(zap_remove(mos, vd->vdev_top_zap, 3630 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); 3631 } 3632 3633 /* 3634 * Free the objects used to store this vdev's spacemaps, and the array 3635 * that points to them. 3636 */ 3637 void 3638 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 3639 { 3640 if (vd->vdev_ms_array == 0) 3641 return; 3642 3643 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3644 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 3645 size_t array_bytes = array_count * sizeof (uint64_t); 3646 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 3647 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 3648 array_bytes, smobj_array, 0)); 3649 3650 for (uint64_t i = 0; i < array_count; i++) { 3651 uint64_t smobj = smobj_array[i]; 3652 if (smobj == 0) 3653 continue; 3654 3655 space_map_free_obj(mos, smobj, tx); 3656 } 3657 3658 kmem_free(smobj_array, array_bytes); 3659 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3660 vdev_destroy_ms_flush_data(vd, tx); 3661 vd->vdev_ms_array = 0; 3662 } 3663 3664 static void 3665 vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 3666 { 3667 spa_t *spa = vd->vdev_spa; 3668 3669 ASSERT(vd->vdev_islog); 3670 ASSERT(vd == vd->vdev_top); 3671 ASSERT3U(txg, ==, spa_syncing_txg(spa)); 3672 3673 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3674 3675 vdev_destroy_spacemaps(vd, tx); 3676 if (vd->vdev_top_zap != 0) { 3677 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3678 vd->vdev_top_zap = 0; 3679 } 3680 3681 dmu_tx_commit(tx); 3682 } 3683 3684 void 3685 vdev_sync_done(vdev_t *vd, uint64_t txg) 3686 { 3687 metaslab_t *msp; 3688 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3689 3690 ASSERT(vdev_is_concrete(vd)); 3691 3692 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3693 != NULL) 3694 metaslab_sync_done(msp, txg); 3695 3696 if (reassess) { 3697 metaslab_sync_reassess(vd->vdev_mg); 3698 if (vd->vdev_log_mg != NULL) 3699 metaslab_sync_reassess(vd->vdev_log_mg); 3700 } 3701 } 3702 3703 void 3704 vdev_sync(vdev_t *vd, uint64_t txg) 3705 { 3706 spa_t *spa = vd->vdev_spa; 3707 vdev_t *lvd; 3708 metaslab_t *msp; 3709 3710 ASSERT3U(txg, ==, spa->spa_syncing_txg); 3711 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3712 if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 3713 ASSERT(vd->vdev_removing || 3714 vd->vdev_ops == &vdev_indirect_ops); 3715 3716 vdev_indirect_sync_obsolete(vd, tx); 3717 3718 /* 3719 * If the vdev is indirect, it can't have dirty 3720 * metaslabs or DTLs. 3721 */ 3722 if (vd->vdev_ops == &vdev_indirect_ops) { 3723 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 3724 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3725 dmu_tx_commit(tx); 3726 return; 3727 } 3728 } 3729 3730 ASSERT(vdev_is_concrete(vd)); 3731 3732 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 3733 !vd->vdev_removing) { 3734 ASSERT(vd == vd->vdev_top); 3735 ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3736 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3737 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3738 ASSERT(vd->vdev_ms_array != 0); 3739 vdev_config_dirty(vd); 3740 } 3741 3742 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3743 metaslab_sync(msp, txg); 3744 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3745 } 3746 3747 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3748 vdev_dtl_sync(lvd, txg); 3749 3750 /* 3751 * If this is an empty log device being removed, destroy the 3752 * metadata associated with it. 3753 */ 3754 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 3755 vdev_remove_empty_log(vd, txg); 3756 3757 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3758 dmu_tx_commit(tx); 3759 } 3760 3761 uint64_t 3762 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3763 { 3764 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3765 } 3766 3767 /* 3768 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 3769 * not be opened, and no I/O is attempted. 3770 */ 3771 int 3772 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3773 { 3774 vdev_t *vd, *tvd; 3775 3776 spa_vdev_state_enter(spa, SCL_NONE); 3777 3778 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3779 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3780 3781 if (!vd->vdev_ops->vdev_op_leaf) 3782 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3783 3784 tvd = vd->vdev_top; 3785 3786 /* 3787 * If user did a 'zpool offline -f' then make the fault persist across 3788 * reboots. 3789 */ 3790 if (aux == VDEV_AUX_EXTERNAL_PERSIST) { 3791 /* 3792 * There are two kinds of forced faults: temporary and 3793 * persistent. Temporary faults go away at pool import, while 3794 * persistent faults stay set. Both types of faults can be 3795 * cleared with a zpool clear. 3796 * 3797 * We tell if a vdev is persistently faulted by looking at the 3798 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at 3799 * import then it's a persistent fault. Otherwise, it's 3800 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external" 3801 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This 3802 * tells vdev_config_generate() (which gets run later) to set 3803 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist. 3804 */ 3805 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; 3806 vd->vdev_tmpoffline = B_FALSE; 3807 aux = VDEV_AUX_EXTERNAL; 3808 } else { 3809 vd->vdev_tmpoffline = B_TRUE; 3810 } 3811 3812 /* 3813 * We don't directly use the aux state here, but if we do a 3814 * vdev_reopen(), we need this value to be present to remember why we 3815 * were faulted. 3816 */ 3817 vd->vdev_label_aux = aux; 3818 3819 /* 3820 * Faulted state takes precedence over degraded. 3821 */ 3822 vd->vdev_delayed_close = B_FALSE; 3823 vd->vdev_faulted = 1ULL; 3824 vd->vdev_degraded = 0ULL; 3825 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 3826 3827 /* 3828 * If this device has the only valid copy of the data, then 3829 * back off and simply mark the vdev as degraded instead. 3830 */ 3831 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 3832 vd->vdev_degraded = 1ULL; 3833 vd->vdev_faulted = 0ULL; 3834 3835 /* 3836 * If we reopen the device and it's not dead, only then do we 3837 * mark it degraded. 3838 */ 3839 vdev_reopen(tvd); 3840 3841 if (vdev_readable(vd)) 3842 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 3843 } 3844 3845 return (spa_vdev_state_exit(spa, vd, 0)); 3846 } 3847 3848 /* 3849 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 3850 * user that something is wrong. The vdev continues to operate as normal as far 3851 * as I/O is concerned. 3852 */ 3853 int 3854 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3855 { 3856 vdev_t *vd; 3857 3858 spa_vdev_state_enter(spa, SCL_NONE); 3859 3860 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3861 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3862 3863 if (!vd->vdev_ops->vdev_op_leaf) 3864 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3865 3866 /* 3867 * If the vdev is already faulted, then don't do anything. 3868 */ 3869 if (vd->vdev_faulted || vd->vdev_degraded) 3870 return (spa_vdev_state_exit(spa, NULL, 0)); 3871 3872 vd->vdev_degraded = 1ULL; 3873 if (!vdev_is_dead(vd)) 3874 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3875 aux); 3876 3877 return (spa_vdev_state_exit(spa, vd, 0)); 3878 } 3879 3880 /* 3881 * Online the given vdev. 3882 * 3883 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3884 * spare device should be detached when the device finishes resilvering. 3885 * Second, the online should be treated like a 'test' online case, so no FMA 3886 * events are generated if the device fails to open. 3887 */ 3888 int 3889 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 3890 { 3891 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 3892 boolean_t wasoffline; 3893 vdev_state_t oldstate; 3894 3895 spa_vdev_state_enter(spa, SCL_NONE); 3896 3897 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3898 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3899 3900 if (!vd->vdev_ops->vdev_op_leaf) 3901 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3902 3903 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 3904 oldstate = vd->vdev_state; 3905 3906 tvd = vd->vdev_top; 3907 vd->vdev_offline = B_FALSE; 3908 vd->vdev_tmpoffline = B_FALSE; 3909 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3910 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3911 3912 /* XXX - L2ARC 1.0 does not support expansion */ 3913 if (!vd->vdev_aux) { 3914 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3915 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) || 3916 spa->spa_autoexpand); 3917 vd->vdev_expansion_time = gethrestime_sec(); 3918 } 3919 3920 vdev_reopen(tvd); 3921 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 3922 3923 if (!vd->vdev_aux) { 3924 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3925 pvd->vdev_expanding = B_FALSE; 3926 } 3927 3928 if (newstate) 3929 *newstate = vd->vdev_state; 3930 if ((flags & ZFS_ONLINE_UNSPARE) && 3931 !vdev_is_dead(vd) && vd->vdev_parent && 3932 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3933 vd->vdev_parent->vdev_child[0] == vd) 3934 vd->vdev_unspare = B_TRUE; 3935 3936 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3937 3938 /* XXX - L2ARC 1.0 does not support expansion */ 3939 if (vd->vdev_aux) 3940 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3941 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3942 } 3943 3944 /* Restart initializing if necessary */ 3945 mutex_enter(&vd->vdev_initialize_lock); 3946 if (vdev_writeable(vd) && 3947 vd->vdev_initialize_thread == NULL && 3948 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3949 (void) vdev_initialize(vd); 3950 } 3951 mutex_exit(&vd->vdev_initialize_lock); 3952 3953 /* 3954 * Restart trimming if necessary. We do not restart trimming for cache 3955 * devices here. This is triggered by l2arc_rebuild_vdev() 3956 * asynchronously for the whole device or in l2arc_evict() as it evicts 3957 * space for upcoming writes. 3958 */ 3959 mutex_enter(&vd->vdev_trim_lock); 3960 if (vdev_writeable(vd) && !vd->vdev_isl2cache && 3961 vd->vdev_trim_thread == NULL && 3962 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 3963 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 3964 vd->vdev_trim_secure); 3965 } 3966 mutex_exit(&vd->vdev_trim_lock); 3967 3968 if (wasoffline || 3969 (oldstate < VDEV_STATE_DEGRADED && 3970 vd->vdev_state >= VDEV_STATE_DEGRADED)) 3971 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 3972 3973 return (spa_vdev_state_exit(spa, vd, 0)); 3974 } 3975 3976 static int 3977 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3978 { 3979 vdev_t *vd, *tvd; 3980 int error = 0; 3981 uint64_t generation; 3982 metaslab_group_t *mg; 3983 3984 top: 3985 spa_vdev_state_enter(spa, SCL_ALLOC); 3986 3987 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3988 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); 3989 3990 if (!vd->vdev_ops->vdev_op_leaf) 3991 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); 3992 3993 if (vd->vdev_ops == &vdev_draid_spare_ops) 3994 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3995 3996 tvd = vd->vdev_top; 3997 mg = tvd->vdev_mg; 3998 generation = spa->spa_config_generation + 1; 3999 4000 /* 4001 * If the device isn't already offline, try to offline it. 4002 */ 4003 if (!vd->vdev_offline) { 4004 /* 4005 * If this device has the only valid copy of some data, 4006 * don't allow it to be offlined. Log devices are always 4007 * expendable. 4008 */ 4009 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 4010 vdev_dtl_required(vd)) 4011 return (spa_vdev_state_exit(spa, NULL, 4012 SET_ERROR(EBUSY))); 4013 4014 /* 4015 * If the top-level is a slog and it has had allocations 4016 * then proceed. We check that the vdev's metaslab group 4017 * is not NULL since it's possible that we may have just 4018 * added this vdev but not yet initialized its metaslabs. 4019 */ 4020 if (tvd->vdev_islog && mg != NULL) { 4021 /* 4022 * Prevent any future allocations. 4023 */ 4024 ASSERT3P(tvd->vdev_log_mg, ==, NULL); 4025 metaslab_group_passivate(mg); 4026 (void) spa_vdev_state_exit(spa, vd, 0); 4027 4028 error = spa_reset_logs(spa); 4029 4030 /* 4031 * If the log device was successfully reset but has 4032 * checkpointed data, do not offline it. 4033 */ 4034 if (error == 0 && 4035 tvd->vdev_checkpoint_sm != NULL) { 4036 ASSERT3U(space_map_allocated( 4037 tvd->vdev_checkpoint_sm), !=, 0); 4038 error = ZFS_ERR_CHECKPOINT_EXISTS; 4039 } 4040 4041 spa_vdev_state_enter(spa, SCL_ALLOC); 4042 4043 /* 4044 * Check to see if the config has changed. 4045 */ 4046 if (error || generation != spa->spa_config_generation) { 4047 metaslab_group_activate(mg); 4048 if (error) 4049 return (spa_vdev_state_exit(spa, 4050 vd, error)); 4051 (void) spa_vdev_state_exit(spa, vd, 0); 4052 goto top; 4053 } 4054 ASSERT0(tvd->vdev_stat.vs_alloc); 4055 } 4056 4057 /* 4058 * Offline this device and reopen its top-level vdev. 4059 * If the top-level vdev is a log device then just offline 4060 * it. Otherwise, if this action results in the top-level 4061 * vdev becoming unusable, undo it and fail the request. 4062 */ 4063 vd->vdev_offline = B_TRUE; 4064 vdev_reopen(tvd); 4065 4066 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 4067 vdev_is_dead(tvd)) { 4068 vd->vdev_offline = B_FALSE; 4069 vdev_reopen(tvd); 4070 return (spa_vdev_state_exit(spa, NULL, 4071 SET_ERROR(EBUSY))); 4072 } 4073 4074 /* 4075 * Add the device back into the metaslab rotor so that 4076 * once we online the device it's open for business. 4077 */ 4078 if (tvd->vdev_islog && mg != NULL) 4079 metaslab_group_activate(mg); 4080 } 4081 4082 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 4083 4084 return (spa_vdev_state_exit(spa, vd, 0)); 4085 } 4086 4087 int 4088 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 4089 { 4090 int error; 4091 4092 mutex_enter(&spa->spa_vdev_top_lock); 4093 error = vdev_offline_locked(spa, guid, flags); 4094 mutex_exit(&spa->spa_vdev_top_lock); 4095 4096 return (error); 4097 } 4098 4099 /* 4100 * Clear the error counts associated with this vdev. Unlike vdev_online() and 4101 * vdev_offline(), we assume the spa config is locked. We also clear all 4102 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 4103 */ 4104 void 4105 vdev_clear(spa_t *spa, vdev_t *vd) 4106 { 4107 vdev_t *rvd = spa->spa_root_vdev; 4108 4109 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 4110 4111 if (vd == NULL) 4112 vd = rvd; 4113 4114 vd->vdev_stat.vs_read_errors = 0; 4115 vd->vdev_stat.vs_write_errors = 0; 4116 vd->vdev_stat.vs_checksum_errors = 0; 4117 vd->vdev_stat.vs_slow_ios = 0; 4118 4119 for (int c = 0; c < vd->vdev_children; c++) 4120 vdev_clear(spa, vd->vdev_child[c]); 4121 4122 /* 4123 * It makes no sense to "clear" an indirect vdev. 4124 */ 4125 if (!vdev_is_concrete(vd)) 4126 return; 4127 4128 /* 4129 * If we're in the FAULTED state or have experienced failed I/O, then 4130 * clear the persistent state and attempt to reopen the device. We 4131 * also mark the vdev config dirty, so that the new faulted state is 4132 * written out to disk. 4133 */ 4134 if (vd->vdev_faulted || vd->vdev_degraded || 4135 !vdev_readable(vd) || !vdev_writeable(vd)) { 4136 /* 4137 * When reopening in response to a clear event, it may be due to 4138 * a fmadm repair request. In this case, if the device is 4139 * still broken, we want to still post the ereport again. 4140 */ 4141 vd->vdev_forcefault = B_TRUE; 4142 4143 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 4144 vd->vdev_cant_read = B_FALSE; 4145 vd->vdev_cant_write = B_FALSE; 4146 vd->vdev_stat.vs_aux = 0; 4147 4148 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 4149 4150 vd->vdev_forcefault = B_FALSE; 4151 4152 if (vd != rvd && vdev_writeable(vd->vdev_top)) 4153 vdev_state_dirty(vd->vdev_top); 4154 4155 /* If a resilver isn't required, check if vdevs can be culled */ 4156 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && 4157 !dsl_scan_resilvering(spa->spa_dsl_pool) && 4158 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) 4159 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 4160 4161 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 4162 } 4163 4164 /* 4165 * When clearing a FMA-diagnosed fault, we always want to 4166 * unspare the device, as we assume that the original spare was 4167 * done in response to the FMA fault. 4168 */ 4169 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 4170 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 4171 vd->vdev_parent->vdev_child[0] == vd) 4172 vd->vdev_unspare = B_TRUE; 4173 4174 /* Clear recent error events cache (i.e. duplicate events tracking) */ 4175 zfs_ereport_clear(spa, vd); 4176 } 4177 4178 boolean_t 4179 vdev_is_dead(vdev_t *vd) 4180 { 4181 /* 4182 * Holes and missing devices are always considered "dead". 4183 * This simplifies the code since we don't have to check for 4184 * these types of devices in the various code paths. 4185 * Instead we rely on the fact that we skip over dead devices 4186 * before issuing I/O to them. 4187 */ 4188 return (vd->vdev_state < VDEV_STATE_DEGRADED || 4189 vd->vdev_ops == &vdev_hole_ops || 4190 vd->vdev_ops == &vdev_missing_ops); 4191 } 4192 4193 boolean_t 4194 vdev_readable(vdev_t *vd) 4195 { 4196 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 4197 } 4198 4199 boolean_t 4200 vdev_writeable(vdev_t *vd) 4201 { 4202 return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 4203 vdev_is_concrete(vd)); 4204 } 4205 4206 boolean_t 4207 vdev_allocatable(vdev_t *vd) 4208 { 4209 uint64_t state = vd->vdev_state; 4210 4211 /* 4212 * We currently allow allocations from vdevs which may be in the 4213 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 4214 * fails to reopen then we'll catch it later when we're holding 4215 * the proper locks. Note that we have to get the vdev state 4216 * in a local variable because although it changes atomically, 4217 * we're asking two separate questions about it. 4218 */ 4219 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 4220 !vd->vdev_cant_write && vdev_is_concrete(vd) && 4221 vd->vdev_mg->mg_initialized); 4222 } 4223 4224 boolean_t 4225 vdev_accessible(vdev_t *vd, zio_t *zio) 4226 { 4227 ASSERT(zio->io_vd == vd); 4228 4229 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 4230 return (B_FALSE); 4231 4232 if (zio->io_type == ZIO_TYPE_READ) 4233 return (!vd->vdev_cant_read); 4234 4235 if (zio->io_type == ZIO_TYPE_WRITE) 4236 return (!vd->vdev_cant_write); 4237 4238 return (B_TRUE); 4239 } 4240 4241 static void 4242 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) 4243 { 4244 /* 4245 * Exclude the dRAID spare when aggregating to avoid double counting 4246 * the ops and bytes. These IOs are counted by the physical leaves. 4247 */ 4248 if (cvd->vdev_ops == &vdev_draid_spare_ops) 4249 return; 4250 4251 for (int t = 0; t < VS_ZIO_TYPES; t++) { 4252 vs->vs_ops[t] += cvs->vs_ops[t]; 4253 vs->vs_bytes[t] += cvs->vs_bytes[t]; 4254 } 4255 4256 cvs->vs_scan_removing = cvd->vdev_removing; 4257 } 4258 4259 /* 4260 * Get extended stats 4261 */ 4262 static void 4263 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) 4264 { 4265 int t, b; 4266 for (t = 0; t < ZIO_TYPES; t++) { 4267 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) 4268 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; 4269 4270 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { 4271 vsx->vsx_total_histo[t][b] += 4272 cvsx->vsx_total_histo[t][b]; 4273 } 4274 } 4275 4276 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { 4277 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { 4278 vsx->vsx_queue_histo[t][b] += 4279 cvsx->vsx_queue_histo[t][b]; 4280 } 4281 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; 4282 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; 4283 4284 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) 4285 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; 4286 4287 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) 4288 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; 4289 } 4290 4291 } 4292 4293 boolean_t 4294 vdev_is_spacemap_addressable(vdev_t *vd) 4295 { 4296 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 4297 return (B_TRUE); 4298 4299 /* 4300 * If double-word space map entries are not enabled we assume 4301 * 47 bits of the space map entry are dedicated to the entry's 4302 * offset (see SM_OFFSET_BITS in space_map.h). We then use that 4303 * to calculate the maximum address that can be described by a 4304 * space map entry for the given device. 4305 */ 4306 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 4307 4308 if (shift >= 63) /* detect potential overflow */ 4309 return (B_TRUE); 4310 4311 return (vd->vdev_asize < (1ULL << shift)); 4312 } 4313 4314 /* 4315 * Get statistics for the given vdev. 4316 */ 4317 static void 4318 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 4319 { 4320 int t; 4321 /* 4322 * If we're getting stats on the root vdev, aggregate the I/O counts 4323 * over all top-level vdevs (i.e. the direct children of the root). 4324 */ 4325 if (!vd->vdev_ops->vdev_op_leaf) { 4326 if (vs) { 4327 memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); 4328 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); 4329 } 4330 if (vsx) 4331 memset(vsx, 0, sizeof (*vsx)); 4332 4333 for (int c = 0; c < vd->vdev_children; c++) { 4334 vdev_t *cvd = vd->vdev_child[c]; 4335 vdev_stat_t *cvs = &cvd->vdev_stat; 4336 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; 4337 4338 vdev_get_stats_ex_impl(cvd, cvs, cvsx); 4339 if (vs) 4340 vdev_get_child_stat(cvd, vs, cvs); 4341 if (vsx) 4342 vdev_get_child_stat_ex(cvd, vsx, cvsx); 4343 } 4344 } else { 4345 /* 4346 * We're a leaf. Just copy our ZIO active queue stats in. The 4347 * other leaf stats are updated in vdev_stat_update(). 4348 */ 4349 if (!vsx) 4350 return; 4351 4352 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); 4353 4354 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { 4355 vsx->vsx_active_queue[t] = 4356 vd->vdev_queue.vq_class[t].vqc_active; 4357 vsx->vsx_pend_queue[t] = avl_numnodes( 4358 &vd->vdev_queue.vq_class[t].vqc_queued_tree); 4359 } 4360 } 4361 } 4362 4363 void 4364 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 4365 { 4366 vdev_t *tvd = vd->vdev_top; 4367 mutex_enter(&vd->vdev_stat_lock); 4368 if (vs) { 4369 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 4370 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 4371 vs->vs_state = vd->vdev_state; 4372 vs->vs_rsize = vdev_get_min_asize(vd); 4373 4374 if (vd->vdev_ops->vdev_op_leaf) { 4375 vs->vs_rsize += VDEV_LABEL_START_SIZE + 4376 VDEV_LABEL_END_SIZE; 4377 /* 4378 * Report initializing progress. Since we don't 4379 * have the initializing locks held, this is only 4380 * an estimate (although a fairly accurate one). 4381 */ 4382 vs->vs_initialize_bytes_done = 4383 vd->vdev_initialize_bytes_done; 4384 vs->vs_initialize_bytes_est = 4385 vd->vdev_initialize_bytes_est; 4386 vs->vs_initialize_state = vd->vdev_initialize_state; 4387 vs->vs_initialize_action_time = 4388 vd->vdev_initialize_action_time; 4389 4390 /* 4391 * Report manual TRIM progress. Since we don't have 4392 * the manual TRIM locks held, this is only an 4393 * estimate (although fairly accurate one). 4394 */ 4395 vs->vs_trim_notsup = !vd->vdev_has_trim; 4396 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 4397 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 4398 vs->vs_trim_state = vd->vdev_trim_state; 4399 vs->vs_trim_action_time = vd->vdev_trim_action_time; 4400 4401 /* Set when there is a deferred resilver. */ 4402 vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 4403 } 4404 4405 /* 4406 * Report expandable space on top-level, non-auxiliary devices 4407 * only. The expandable space is reported in terms of metaslab 4408 * sized units since that determines how much space the pool 4409 * can expand. 4410 */ 4411 if (vd->vdev_aux == NULL && tvd != NULL) { 4412 vs->vs_esize = P2ALIGN( 4413 vd->vdev_max_asize - vd->vdev_asize, 4414 1ULL << tvd->vdev_ms_shift); 4415 } 4416 4417 vs->vs_configured_ashift = vd->vdev_top != NULL 4418 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift; 4419 vs->vs_logical_ashift = vd->vdev_logical_ashift; 4420 vs->vs_physical_ashift = vd->vdev_physical_ashift; 4421 4422 /* 4423 * Report fragmentation and rebuild progress for top-level, 4424 * non-auxiliary, concrete devices. 4425 */ 4426 if (vd->vdev_aux == NULL && vd == vd->vdev_top && 4427 vdev_is_concrete(vd)) { 4428 /* 4429 * The vdev fragmentation rating doesn't take into 4430 * account the embedded slog metaslab (vdev_log_mg). 4431 * Since it's only one metaslab, it would have a tiny 4432 * impact on the overall fragmentation. 4433 */ 4434 vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 4435 vd->vdev_mg->mg_fragmentation : 0; 4436 } 4437 } 4438 4439 vdev_get_stats_ex_impl(vd, vs, vsx); 4440 mutex_exit(&vd->vdev_stat_lock); 4441 } 4442 4443 void 4444 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 4445 { 4446 return (vdev_get_stats_ex(vd, vs, NULL)); 4447 } 4448 4449 void 4450 vdev_clear_stats(vdev_t *vd) 4451 { 4452 mutex_enter(&vd->vdev_stat_lock); 4453 vd->vdev_stat.vs_space = 0; 4454 vd->vdev_stat.vs_dspace = 0; 4455 vd->vdev_stat.vs_alloc = 0; 4456 mutex_exit(&vd->vdev_stat_lock); 4457 } 4458 4459 void 4460 vdev_scan_stat_init(vdev_t *vd) 4461 { 4462 vdev_stat_t *vs = &vd->vdev_stat; 4463 4464 for (int c = 0; c < vd->vdev_children; c++) 4465 vdev_scan_stat_init(vd->vdev_child[c]); 4466 4467 mutex_enter(&vd->vdev_stat_lock); 4468 vs->vs_scan_processed = 0; 4469 mutex_exit(&vd->vdev_stat_lock); 4470 } 4471 4472 void 4473 vdev_stat_update(zio_t *zio, uint64_t psize) 4474 { 4475 spa_t *spa = zio->io_spa; 4476 vdev_t *rvd = spa->spa_root_vdev; 4477 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 4478 vdev_t *pvd; 4479 uint64_t txg = zio->io_txg; 4480 vdev_stat_t *vs = &vd->vdev_stat; 4481 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; 4482 zio_type_t type = zio->io_type; 4483 int flags = zio->io_flags; 4484 4485 /* 4486 * If this i/o is a gang leader, it didn't do any actual work. 4487 */ 4488 if (zio->io_gang_tree) 4489 return; 4490 4491 if (zio->io_error == 0) { 4492 /* 4493 * If this is a root i/o, don't count it -- we've already 4494 * counted the top-level vdevs, and vdev_get_stats() will 4495 * aggregate them when asked. This reduces contention on 4496 * the root vdev_stat_lock and implicitly handles blocks 4497 * that compress away to holes, for which there is no i/o. 4498 * (Holes never create vdev children, so all the counters 4499 * remain zero, which is what we want.) 4500 * 4501 * Note: this only applies to successful i/o (io_error == 0) 4502 * because unlike i/o counts, errors are not additive. 4503 * When reading a ditto block, for example, failure of 4504 * one top-level vdev does not imply a root-level error. 4505 */ 4506 if (vd == rvd) 4507 return; 4508 4509 ASSERT(vd == zio->io_vd); 4510 4511 if (flags & ZIO_FLAG_IO_BYPASS) 4512 return; 4513 4514 mutex_enter(&vd->vdev_stat_lock); 4515 4516 if (flags & ZIO_FLAG_IO_REPAIR) { 4517 /* 4518 * Repair is the result of a resilver issued by the 4519 * scan thread (spa_sync). 4520 */ 4521 if (flags & ZIO_FLAG_SCAN_THREAD) { 4522 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 4523 dsl_scan_phys_t *scn_phys = &scn->scn_phys; 4524 uint64_t *processed = &scn_phys->scn_processed; 4525 4526 if (vd->vdev_ops->vdev_op_leaf) 4527 atomic_add_64(processed, psize); 4528 vs->vs_scan_processed += psize; 4529 } 4530 4531 /* 4532 * Repair is the result of a rebuild issued by the 4533 * rebuild thread (vdev_rebuild_thread). To avoid 4534 * double counting repaired bytes the virtual dRAID 4535 * spare vdev is excluded from the processed bytes. 4536 */ 4537 if (zio->io_priority == ZIO_PRIORITY_REBUILD) { 4538 vdev_t *tvd = vd->vdev_top; 4539 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config; 4540 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; 4541 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt; 4542 4543 if (vd->vdev_ops->vdev_op_leaf && 4544 vd->vdev_ops != &vdev_draid_spare_ops) { 4545 atomic_add_64(rebuilt, psize); 4546 } 4547 vs->vs_rebuild_processed += psize; 4548 } 4549 4550 if (flags & ZIO_FLAG_SELF_HEAL) 4551 vs->vs_self_healed += psize; 4552 } 4553 4554 /* 4555 * The bytes/ops/histograms are recorded at the leaf level and 4556 * aggregated into the higher level vdevs in vdev_get_stats(). 4557 */ 4558 if (vd->vdev_ops->vdev_op_leaf && 4559 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { 4560 zio_type_t vs_type = type; 4561 zio_priority_t priority = zio->io_priority; 4562 4563 /* 4564 * TRIM ops and bytes are reported to user space as 4565 * ZIO_TYPE_IOCTL. This is done to preserve the 4566 * vdev_stat_t structure layout for user space. 4567 */ 4568 if (type == ZIO_TYPE_TRIM) 4569 vs_type = ZIO_TYPE_IOCTL; 4570 4571 /* 4572 * Solely for the purposes of 'zpool iostat -lqrw' 4573 * reporting use the priority to catagorize the IO. 4574 * Only the following are reported to user space: 4575 * 4576 * ZIO_PRIORITY_SYNC_READ, 4577 * ZIO_PRIORITY_SYNC_WRITE, 4578 * ZIO_PRIORITY_ASYNC_READ, 4579 * ZIO_PRIORITY_ASYNC_WRITE, 4580 * ZIO_PRIORITY_SCRUB, 4581 * ZIO_PRIORITY_TRIM. 4582 */ 4583 if (priority == ZIO_PRIORITY_REBUILD) { 4584 priority = ((type == ZIO_TYPE_WRITE) ? 4585 ZIO_PRIORITY_ASYNC_WRITE : 4586 ZIO_PRIORITY_SCRUB); 4587 } else if (priority == ZIO_PRIORITY_INITIALIZING) { 4588 ASSERT3U(type, ==, ZIO_TYPE_WRITE); 4589 priority = ZIO_PRIORITY_ASYNC_WRITE; 4590 } else if (priority == ZIO_PRIORITY_REMOVAL) { 4591 priority = ((type == ZIO_TYPE_WRITE) ? 4592 ZIO_PRIORITY_ASYNC_WRITE : 4593 ZIO_PRIORITY_ASYNC_READ); 4594 } 4595 4596 vs->vs_ops[vs_type]++; 4597 vs->vs_bytes[vs_type] += psize; 4598 4599 if (flags & ZIO_FLAG_DELEGATED) { 4600 vsx->vsx_agg_histo[priority] 4601 [RQ_HISTO(zio->io_size)]++; 4602 } else { 4603 vsx->vsx_ind_histo[priority] 4604 [RQ_HISTO(zio->io_size)]++; 4605 } 4606 4607 if (zio->io_delta && zio->io_delay) { 4608 vsx->vsx_queue_histo[priority] 4609 [L_HISTO(zio->io_delta - zio->io_delay)]++; 4610 vsx->vsx_disk_histo[type] 4611 [L_HISTO(zio->io_delay)]++; 4612 vsx->vsx_total_histo[type] 4613 [L_HISTO(zio->io_delta)]++; 4614 } 4615 } 4616 4617 mutex_exit(&vd->vdev_stat_lock); 4618 return; 4619 } 4620 4621 if (flags & ZIO_FLAG_SPECULATIVE) 4622 return; 4623 4624 /* 4625 * If this is an I/O error that is going to be retried, then ignore the 4626 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 4627 * hard errors, when in reality they can happen for any number of 4628 * innocuous reasons (bus resets, MPxIO link failure, etc). 4629 */ 4630 if (zio->io_error == EIO && 4631 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 4632 return; 4633 4634 /* 4635 * Intent logs writes won't propagate their error to the root 4636 * I/O so don't mark these types of failures as pool-level 4637 * errors. 4638 */ 4639 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 4640 return; 4641 4642 if (type == ZIO_TYPE_WRITE && txg != 0 && 4643 (!(flags & ZIO_FLAG_IO_REPAIR) || 4644 (flags & ZIO_FLAG_SCAN_THREAD) || 4645 spa->spa_claiming)) { 4646 /* 4647 * This is either a normal write (not a repair), or it's 4648 * a repair induced by the scrub thread, or it's a repair 4649 * made by zil_claim() during spa_load() in the first txg. 4650 * In the normal case, we commit the DTL change in the same 4651 * txg as the block was born. In the scrub-induced repair 4652 * case, we know that scrubs run in first-pass syncing context, 4653 * so we commit the DTL change in spa_syncing_txg(spa). 4654 * In the zil_claim() case, we commit in spa_first_txg(spa). 4655 * 4656 * We currently do not make DTL entries for failed spontaneous 4657 * self-healing writes triggered by normal (non-scrubbing) 4658 * reads, because we have no transactional context in which to 4659 * do so -- and it's not clear that it'd be desirable anyway. 4660 */ 4661 if (vd->vdev_ops->vdev_op_leaf) { 4662 uint64_t commit_txg = txg; 4663 if (flags & ZIO_FLAG_SCAN_THREAD) { 4664 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 4665 ASSERT(spa_sync_pass(spa) == 1); 4666 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 4667 commit_txg = spa_syncing_txg(spa); 4668 } else if (spa->spa_claiming) { 4669 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 4670 commit_txg = spa_first_txg(spa); 4671 } 4672 ASSERT(commit_txg >= spa_syncing_txg(spa)); 4673 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 4674 return; 4675 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 4676 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 4677 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 4678 } 4679 if (vd != rvd) 4680 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 4681 } 4682 } 4683 4684 int64_t 4685 vdev_deflated_space(vdev_t *vd, int64_t space) 4686 { 4687 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 4688 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 4689 4690 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 4691 } 4692 4693 /* 4694 * Update the in-core space usage stats for this vdev, its metaslab class, 4695 * and the root vdev. 4696 */ 4697 void 4698 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 4699 int64_t space_delta) 4700 { 4701 int64_t dspace_delta; 4702 spa_t *spa = vd->vdev_spa; 4703 vdev_t *rvd = spa->spa_root_vdev; 4704 4705 ASSERT(vd == vd->vdev_top); 4706 4707 /* 4708 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 4709 * factor. We must calculate this here and not at the root vdev 4710 * because the root vdev's psize-to-asize is simply the max of its 4711 * children's, thus not accurate enough for us. 4712 */ 4713 dspace_delta = vdev_deflated_space(vd, space_delta); 4714 4715 mutex_enter(&vd->vdev_stat_lock); 4716 /* ensure we won't underflow */ 4717 if (alloc_delta < 0) { 4718 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 4719 } 4720 4721 vd->vdev_stat.vs_alloc += alloc_delta; 4722 vd->vdev_stat.vs_space += space_delta; 4723 vd->vdev_stat.vs_dspace += dspace_delta; 4724 mutex_exit(&vd->vdev_stat_lock); 4725 4726 /* every class but log contributes to root space stats */ 4727 if (vd->vdev_mg != NULL && !vd->vdev_islog) { 4728 ASSERT(!vd->vdev_isl2cache); 4729 mutex_enter(&rvd->vdev_stat_lock); 4730 rvd->vdev_stat.vs_alloc += alloc_delta; 4731 rvd->vdev_stat.vs_space += space_delta; 4732 rvd->vdev_stat.vs_dspace += dspace_delta; 4733 mutex_exit(&rvd->vdev_stat_lock); 4734 } 4735 /* Note: metaslab_class_space_update moved to metaslab_space_update */ 4736 } 4737 4738 /* 4739 * Mark a top-level vdev's config as dirty, placing it on the dirty list 4740 * so that it will be written out next time the vdev configuration is synced. 4741 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 4742 */ 4743 void 4744 vdev_config_dirty(vdev_t *vd) 4745 { 4746 spa_t *spa = vd->vdev_spa; 4747 vdev_t *rvd = spa->spa_root_vdev; 4748 int c; 4749 4750 ASSERT(spa_writeable(spa)); 4751 4752 /* 4753 * If this is an aux vdev (as with l2cache and spare devices), then we 4754 * update the vdev config manually and set the sync flag. 4755 */ 4756 if (vd->vdev_aux != NULL) { 4757 spa_aux_vdev_t *sav = vd->vdev_aux; 4758 nvlist_t **aux; 4759 uint_t naux; 4760 4761 for (c = 0; c < sav->sav_count; c++) { 4762 if (sav->sav_vdevs[c] == vd) 4763 break; 4764 } 4765 4766 if (c == sav->sav_count) { 4767 /* 4768 * We're being removed. There's nothing more to do. 4769 */ 4770 ASSERT(sav->sav_sync == B_TRUE); 4771 return; 4772 } 4773 4774 sav->sav_sync = B_TRUE; 4775 4776 if (nvlist_lookup_nvlist_array(sav->sav_config, 4777 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 4778 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 4779 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 4780 } 4781 4782 ASSERT(c < naux); 4783 4784 /* 4785 * Setting the nvlist in the middle if the array is a little 4786 * sketchy, but it will work. 4787 */ 4788 nvlist_free(aux[c]); 4789 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 4790 4791 return; 4792 } 4793 4794 /* 4795 * The dirty list is protected by the SCL_CONFIG lock. The caller 4796 * must either hold SCL_CONFIG as writer, or must be the sync thread 4797 * (which holds SCL_CONFIG as reader). There's only one sync thread, 4798 * so this is sufficient to ensure mutual exclusion. 4799 */ 4800 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4801 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4802 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4803 4804 if (vd == rvd) { 4805 for (c = 0; c < rvd->vdev_children; c++) 4806 vdev_config_dirty(rvd->vdev_child[c]); 4807 } else { 4808 ASSERT(vd == vd->vdev_top); 4809 4810 if (!list_link_active(&vd->vdev_config_dirty_node) && 4811 vdev_is_concrete(vd)) { 4812 list_insert_head(&spa->spa_config_dirty_list, vd); 4813 } 4814 } 4815 } 4816 4817 void 4818 vdev_config_clean(vdev_t *vd) 4819 { 4820 spa_t *spa = vd->vdev_spa; 4821 4822 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4823 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4824 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4825 4826 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4827 list_remove(&spa->spa_config_dirty_list, vd); 4828 } 4829 4830 /* 4831 * Mark a top-level vdev's state as dirty, so that the next pass of 4832 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4833 * the state changes from larger config changes because they require 4834 * much less locking, and are often needed for administrative actions. 4835 */ 4836 void 4837 vdev_state_dirty(vdev_t *vd) 4838 { 4839 spa_t *spa = vd->vdev_spa; 4840 4841 ASSERT(spa_writeable(spa)); 4842 ASSERT(vd == vd->vdev_top); 4843 4844 /* 4845 * The state list is protected by the SCL_STATE lock. The caller 4846 * must either hold SCL_STATE as writer, or must be the sync thread 4847 * (which holds SCL_STATE as reader). There's only one sync thread, 4848 * so this is sufficient to ensure mutual exclusion. 4849 */ 4850 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4851 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4852 spa_config_held(spa, SCL_STATE, RW_READER))); 4853 4854 if (!list_link_active(&vd->vdev_state_dirty_node) && 4855 vdev_is_concrete(vd)) 4856 list_insert_head(&spa->spa_state_dirty_list, vd); 4857 } 4858 4859 void 4860 vdev_state_clean(vdev_t *vd) 4861 { 4862 spa_t *spa = vd->vdev_spa; 4863 4864 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4865 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4866 spa_config_held(spa, SCL_STATE, RW_READER))); 4867 4868 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 4869 list_remove(&spa->spa_state_dirty_list, vd); 4870 } 4871 4872 /* 4873 * Propagate vdev state up from children to parent. 4874 */ 4875 void 4876 vdev_propagate_state(vdev_t *vd) 4877 { 4878 spa_t *spa = vd->vdev_spa; 4879 vdev_t *rvd = spa->spa_root_vdev; 4880 int degraded = 0, faulted = 0; 4881 int corrupted = 0; 4882 vdev_t *child; 4883 4884 if (vd->vdev_children > 0) { 4885 for (int c = 0; c < vd->vdev_children; c++) { 4886 child = vd->vdev_child[c]; 4887 4888 /* 4889 * Don't factor holes or indirect vdevs into the 4890 * decision. 4891 */ 4892 if (!vdev_is_concrete(child)) 4893 continue; 4894 4895 if (!vdev_readable(child) || 4896 (!vdev_writeable(child) && spa_writeable(spa))) { 4897 /* 4898 * Root special: if there is a top-level log 4899 * device, treat the root vdev as if it were 4900 * degraded. 4901 */ 4902 if (child->vdev_islog && vd == rvd) 4903 degraded++; 4904 else 4905 faulted++; 4906 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 4907 degraded++; 4908 } 4909 4910 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 4911 corrupted++; 4912 } 4913 4914 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 4915 4916 /* 4917 * Root special: if there is a top-level vdev that cannot be 4918 * opened due to corrupted metadata, then propagate the root 4919 * vdev's aux state as 'corrupt' rather than 'insufficient 4920 * replicas'. 4921 */ 4922 if (corrupted && vd == rvd && 4923 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 4924 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 4925 VDEV_AUX_CORRUPT_DATA); 4926 } 4927 4928 if (vd->vdev_parent) 4929 vdev_propagate_state(vd->vdev_parent); 4930 } 4931 4932 /* 4933 * Set a vdev's state. If this is during an open, we don't update the parent 4934 * state, because we're in the process of opening children depth-first. 4935 * Otherwise, we propagate the change to the parent. 4936 * 4937 * If this routine places a device in a faulted state, an appropriate ereport is 4938 * generated. 4939 */ 4940 void 4941 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4942 { 4943 uint64_t save_state; 4944 spa_t *spa = vd->vdev_spa; 4945 4946 if (state == vd->vdev_state) { 4947 /* 4948 * Since vdev_offline() code path is already in an offline 4949 * state we can miss a statechange event to OFFLINE. Check 4950 * the previous state to catch this condition. 4951 */ 4952 if (vd->vdev_ops->vdev_op_leaf && 4953 (state == VDEV_STATE_OFFLINE) && 4954 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { 4955 /* post an offline state change */ 4956 zfs_post_state_change(spa, vd, vd->vdev_prevstate); 4957 } 4958 vd->vdev_stat.vs_aux = aux; 4959 return; 4960 } 4961 4962 save_state = vd->vdev_state; 4963 4964 vd->vdev_state = state; 4965 vd->vdev_stat.vs_aux = aux; 4966 4967 /* 4968 * If we are setting the vdev state to anything but an open state, then 4969 * always close the underlying device unless the device has requested 4970 * a delayed close (i.e. we're about to remove or fault the device). 4971 * Otherwise, we keep accessible but invalid devices open forever. 4972 * We don't call vdev_close() itself, because that implies some extra 4973 * checks (offline, etc) that we don't want here. This is limited to 4974 * leaf devices, because otherwise closing the device will affect other 4975 * children. 4976 */ 4977 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 4978 vd->vdev_ops->vdev_op_leaf) 4979 vd->vdev_ops->vdev_op_close(vd); 4980 4981 if (vd->vdev_removed && 4982 state == VDEV_STATE_CANT_OPEN && 4983 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 4984 /* 4985 * If the previous state is set to VDEV_STATE_REMOVED, then this 4986 * device was previously marked removed and someone attempted to 4987 * reopen it. If this failed due to a nonexistent device, then 4988 * keep the device in the REMOVED state. We also let this be if 4989 * it is one of our special test online cases, which is only 4990 * attempting to online the device and shouldn't generate an FMA 4991 * fault. 4992 */ 4993 vd->vdev_state = VDEV_STATE_REMOVED; 4994 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 4995 } else if (state == VDEV_STATE_REMOVED) { 4996 vd->vdev_removed = B_TRUE; 4997 } else if (state == VDEV_STATE_CANT_OPEN) { 4998 /* 4999 * If we fail to open a vdev during an import or recovery, we 5000 * mark it as "not available", which signifies that it was 5001 * never there to begin with. Failure to open such a device 5002 * is not considered an error. 5003 */ 5004 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 5005 spa_load_state(spa) == SPA_LOAD_RECOVER) && 5006 vd->vdev_ops->vdev_op_leaf) 5007 vd->vdev_not_present = 1; 5008 5009 /* 5010 * Post the appropriate ereport. If the 'prevstate' field is 5011 * set to something other than VDEV_STATE_UNKNOWN, it indicates 5012 * that this is part of a vdev_reopen(). In this case, we don't 5013 * want to post the ereport if the device was already in the 5014 * CANT_OPEN state beforehand. 5015 * 5016 * If the 'checkremove' flag is set, then this is an attempt to 5017 * online the device in response to an insertion event. If we 5018 * hit this case, then we have detected an insertion event for a 5019 * faulted or offline device that wasn't in the removed state. 5020 * In this scenario, we don't post an ereport because we are 5021 * about to replace the device, or attempt an online with 5022 * vdev_forcefault, which will generate the fault for us. 5023 */ 5024 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 5025 !vd->vdev_not_present && !vd->vdev_checkremove && 5026 vd != spa->spa_root_vdev) { 5027 const char *class; 5028 5029 switch (aux) { 5030 case VDEV_AUX_OPEN_FAILED: 5031 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 5032 break; 5033 case VDEV_AUX_CORRUPT_DATA: 5034 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 5035 break; 5036 case VDEV_AUX_NO_REPLICAS: 5037 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 5038 break; 5039 case VDEV_AUX_BAD_GUID_SUM: 5040 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 5041 break; 5042 case VDEV_AUX_TOO_SMALL: 5043 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 5044 break; 5045 case VDEV_AUX_BAD_LABEL: 5046 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 5047 break; 5048 case VDEV_AUX_BAD_ASHIFT: 5049 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 5050 break; 5051 default: 5052 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 5053 } 5054 5055 (void) zfs_ereport_post(class, spa, vd, NULL, NULL, 5056 save_state); 5057 } 5058 5059 /* Erase any notion of persistent removed state */ 5060 vd->vdev_removed = B_FALSE; 5061 } else { 5062 vd->vdev_removed = B_FALSE; 5063 } 5064 5065 /* 5066 * Notify ZED of any significant state-change on a leaf vdev. 5067 * 5068 */ 5069 if (vd->vdev_ops->vdev_op_leaf) { 5070 /* preserve original state from a vdev_reopen() */ 5071 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && 5072 (vd->vdev_prevstate != vd->vdev_state) && 5073 (save_state <= VDEV_STATE_CLOSED)) 5074 save_state = vd->vdev_prevstate; 5075 5076 /* filter out state change due to initial vdev_open */ 5077 if (save_state > VDEV_STATE_CLOSED) 5078 zfs_post_state_change(spa, vd, save_state); 5079 } 5080 5081 if (!isopen && vd->vdev_parent) 5082 vdev_propagate_state(vd->vdev_parent); 5083 } 5084 5085 boolean_t 5086 vdev_children_are_offline(vdev_t *vd) 5087 { 5088 ASSERT(!vd->vdev_ops->vdev_op_leaf); 5089 5090 for (uint64_t i = 0; i < vd->vdev_children; i++) { 5091 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 5092 return (B_FALSE); 5093 } 5094 5095 return (B_TRUE); 5096 } 5097 5098 /* 5099 * Check the vdev configuration to ensure that it's capable of supporting 5100 * a root pool. We do not support partial configuration. 5101 */ 5102 boolean_t 5103 vdev_is_bootable(vdev_t *vd) 5104 { 5105 if (!vd->vdev_ops->vdev_op_leaf) { 5106 const char *vdev_type = vd->vdev_ops->vdev_op_type; 5107 5108 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) 5109 return (B_FALSE); 5110 } 5111 5112 for (int c = 0; c < vd->vdev_children; c++) { 5113 if (!vdev_is_bootable(vd->vdev_child[c])) 5114 return (B_FALSE); 5115 } 5116 return (B_TRUE); 5117 } 5118 5119 boolean_t 5120 vdev_is_concrete(vdev_t *vd) 5121 { 5122 vdev_ops_t *ops = vd->vdev_ops; 5123 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 5124 ops == &vdev_missing_ops || ops == &vdev_root_ops) { 5125 return (B_FALSE); 5126 } else { 5127 return (B_TRUE); 5128 } 5129 } 5130 5131 /* 5132 * Determine if a log device has valid content. If the vdev was 5133 * removed or faulted in the MOS config then we know that 5134 * the content on the log device has already been written to the pool. 5135 */ 5136 boolean_t 5137 vdev_log_state_valid(vdev_t *vd) 5138 { 5139 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 5140 !vd->vdev_removed) 5141 return (B_TRUE); 5142 5143 for (int c = 0; c < vd->vdev_children; c++) 5144 if (vdev_log_state_valid(vd->vdev_child[c])) 5145 return (B_TRUE); 5146 5147 return (B_FALSE); 5148 } 5149 5150 /* 5151 * Expand a vdev if possible. 5152 */ 5153 void 5154 vdev_expand(vdev_t *vd, uint64_t txg) 5155 { 5156 ASSERT(vd->vdev_top == vd); 5157 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5158 ASSERT(vdev_is_concrete(vd)); 5159 5160 vdev_set_deflate_ratio(vd); 5161 5162 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 5163 vdev_is_concrete(vd)) { 5164 vdev_metaslab_group_create(vd); 5165 VERIFY(vdev_metaslab_init(vd, txg) == 0); 5166 vdev_config_dirty(vd); 5167 } 5168 } 5169 5170 /* 5171 * Split a vdev. 5172 */ 5173 void 5174 vdev_split(vdev_t *vd) 5175 { 5176 vdev_t *cvd, *pvd = vd->vdev_parent; 5177 5178 vdev_remove_child(pvd, vd); 5179 vdev_compact_children(pvd); 5180 5181 cvd = pvd->vdev_child[0]; 5182 if (pvd->vdev_children == 1) { 5183 vdev_remove_parent(cvd); 5184 cvd->vdev_splitting = B_TRUE; 5185 } 5186 vdev_propagate_state(cvd); 5187 } 5188 5189 void 5190 vdev_deadman(vdev_t *vd, char *tag) 5191 { 5192 for (int c = 0; c < vd->vdev_children; c++) { 5193 vdev_t *cvd = vd->vdev_child[c]; 5194 5195 vdev_deadman(cvd, tag); 5196 } 5197 5198 if (vd->vdev_ops->vdev_op_leaf) { 5199 vdev_queue_t *vq = &vd->vdev_queue; 5200 5201 mutex_enter(&vq->vq_lock); 5202 if (avl_numnodes(&vq->vq_active_tree) > 0) { 5203 spa_t *spa = vd->vdev_spa; 5204 zio_t *fio; 5205 uint64_t delta; 5206 5207 zfs_dbgmsg("slow vdev: %s has %d active IOs", 5208 vd->vdev_path, avl_numnodes(&vq->vq_active_tree)); 5209 5210 /* 5211 * Look at the head of all the pending queues, 5212 * if any I/O has been outstanding for longer than 5213 * the spa_deadman_synctime invoke the deadman logic. 5214 */ 5215 fio = avl_first(&vq->vq_active_tree); 5216 delta = gethrtime() - fio->io_timestamp; 5217 if (delta > spa_deadman_synctime(spa)) 5218 zio_deadman(fio, tag); 5219 } 5220 mutex_exit(&vq->vq_lock); 5221 } 5222 } 5223 5224 void 5225 vdev_defer_resilver(vdev_t *vd) 5226 { 5227 ASSERT(vd->vdev_ops->vdev_op_leaf); 5228 5229 vd->vdev_resilver_deferred = B_TRUE; 5230 vd->vdev_spa->spa_resilver_deferred = B_TRUE; 5231 } 5232 5233 /* 5234 * Clears the resilver deferred flag on all leaf devs under vd. Returns 5235 * B_TRUE if we have devices that need to be resilvered and are available to 5236 * accept resilver I/Os. 5237 */ 5238 boolean_t 5239 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) 5240 { 5241 boolean_t resilver_needed = B_FALSE; 5242 spa_t *spa = vd->vdev_spa; 5243 5244 for (int c = 0; c < vd->vdev_children; c++) { 5245 vdev_t *cvd = vd->vdev_child[c]; 5246 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); 5247 } 5248 5249 if (vd == spa->spa_root_vdev && 5250 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { 5251 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); 5252 vdev_config_dirty(vd); 5253 spa->spa_resilver_deferred = B_FALSE; 5254 return (resilver_needed); 5255 } 5256 5257 if (!vdev_is_concrete(vd) || vd->vdev_aux || 5258 !vd->vdev_ops->vdev_op_leaf) 5259 return (resilver_needed); 5260 5261 vd->vdev_resilver_deferred = B_FALSE; 5262 5263 return (!vdev_is_dead(vd) && !vd->vdev_offline && 5264 vdev_resilver_needed(vd, NULL, NULL)); 5265 } 5266 5267 boolean_t 5268 vdev_xlate_is_empty(range_seg64_t *rs) 5269 { 5270 return (rs->rs_start == rs->rs_end); 5271 } 5272 5273 /* 5274 * Translate a logical range to the first contiguous physical range for the 5275 * specified vdev_t. This function is initially called with a leaf vdev and 5276 * will walk each parent vdev until it reaches a top-level vdev. Once the 5277 * top-level is reached the physical range is initialized and the recursive 5278 * function begins to unwind. As it unwinds it calls the parent's vdev 5279 * specific translation function to do the real conversion. 5280 */ 5281 void 5282 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 5283 range_seg64_t *physical_rs, range_seg64_t *remain_rs) 5284 { 5285 /* 5286 * Walk up the vdev tree 5287 */ 5288 if (vd != vd->vdev_top) { 5289 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs, 5290 remain_rs); 5291 } else { 5292 /* 5293 * We've reached the top-level vdev, initialize the physical 5294 * range to the logical range and set an empty remaining 5295 * range then start to unwind. 5296 */ 5297 physical_rs->rs_start = logical_rs->rs_start; 5298 physical_rs->rs_end = logical_rs->rs_end; 5299 5300 remain_rs->rs_start = logical_rs->rs_start; 5301 remain_rs->rs_end = logical_rs->rs_start; 5302 5303 return; 5304 } 5305 5306 vdev_t *pvd = vd->vdev_parent; 5307 ASSERT3P(pvd, !=, NULL); 5308 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 5309 5310 /* 5311 * As this recursive function unwinds, translate the logical 5312 * range into its physical and any remaining components by calling 5313 * the vdev specific translate function. 5314 */ 5315 range_seg64_t intermediate = { 0 }; 5316 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate, remain_rs); 5317 5318 physical_rs->rs_start = intermediate.rs_start; 5319 physical_rs->rs_end = intermediate.rs_end; 5320 } 5321 5322 void 5323 vdev_xlate_walk(vdev_t *vd, const range_seg64_t *logical_rs, 5324 vdev_xlate_func_t *func, void *arg) 5325 { 5326 range_seg64_t iter_rs = *logical_rs; 5327 range_seg64_t physical_rs; 5328 range_seg64_t remain_rs; 5329 5330 while (!vdev_xlate_is_empty(&iter_rs)) { 5331 5332 vdev_xlate(vd, &iter_rs, &physical_rs, &remain_rs); 5333 5334 /* 5335 * With raidz and dRAID, it's possible that the logical range 5336 * does not live on this leaf vdev. Only when there is a non- 5337 * zero physical size call the provided function. 5338 */ 5339 if (!vdev_xlate_is_empty(&physical_rs)) 5340 func(arg, &physical_rs); 5341 5342 iter_rs = remain_rs; 5343 } 5344 } 5345 5346 /* 5347 * Look at the vdev tree and determine whether any devices are currently being 5348 * replaced. 5349 */ 5350 boolean_t 5351 vdev_replace_in_progress(vdev_t *vdev) 5352 { 5353 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0); 5354 5355 if (vdev->vdev_ops == &vdev_replacing_ops) 5356 return (B_TRUE); 5357 5358 /* 5359 * A 'spare' vdev indicates that we have a replace in progress, unless 5360 * it has exactly two children, and the second, the hot spare, has 5361 * finished being resilvered. 5362 */ 5363 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 || 5364 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING))) 5365 return (B_TRUE); 5366 5367 for (int i = 0; i < vdev->vdev_children; i++) { 5368 if (vdev_replace_in_progress(vdev->vdev_child[i])) 5369 return (B_TRUE); 5370 } 5371 5372 return (B_FALSE); 5373 } 5374 5375 EXPORT_SYMBOL(vdev_fault); 5376 EXPORT_SYMBOL(vdev_degrade); 5377 EXPORT_SYMBOL(vdev_online); 5378 EXPORT_SYMBOL(vdev_offline); 5379 EXPORT_SYMBOL(vdev_clear); 5380 5381 /* BEGIN CSTYLED */ 5382 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW, 5383 "Target number of metaslabs per top-level vdev"); 5384 5385 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW, 5386 "Default limit for metaslab size"); 5387 5388 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW, 5389 "Minimum number of metaslabs per top-level vdev"); 5390 5391 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW, 5392 "Practical upper limit of total metaslabs per top-level vdev"); 5393 5394 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW, 5395 "Rate limit slow IO (delay) events to this many per second"); 5396 5397 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW, 5398 "Rate limit checksum events to this many checksum errors per second " 5399 "(do not set below zed threshold)."); 5400 5401 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW, 5402 "Ignore errors during resilver/scrub"); 5403 5404 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW, 5405 "Bypass vdev_validate()"); 5406 5407 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW, 5408 "Disable cache flushes"); 5409 5410 ZFS_MODULE_PARAM(zfs, zfs_, embedded_slog_min_ms, INT, ZMOD_RW, 5411 "Minimum number of metaslabs required to dedicate one for log blocks"); 5412 5413 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift, 5414 param_set_min_auto_ashift, param_get_ulong, ZMOD_RW, 5415 "Minimum ashift used when creating new top-level vdevs"); 5416 5417 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift, 5418 param_set_max_auto_ashift, param_get_ulong, ZMOD_RW, 5419 "Maximum ashift used when optimizing for logical -> physical sector " 5420 "size on new top-level vdevs"); 5421 /* END CSTYLED */ 5422