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, 2018 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 2019 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/uberblock_impl.h> 43 #include <sys/metaslab.h> 44 #include <sys/metaslab_impl.h> 45 #include <sys/space_map.h> 46 #include <sys/space_reftree.h> 47 #include <sys/zio.h> 48 #include <sys/zap.h> 49 #include <sys/fs/zfs.h> 50 #include <sys/arc.h> 51 #include <sys/zil.h> 52 #include <sys/dsl_scan.h> 53 #include <sys/abd.h> 54 #include <sys/vdev_initialize.h> 55 #include <sys/vdev_trim.h> 56 57 /* 58 * Virtual device management. 59 */ 60 61 static vdev_ops_t *vdev_ops_table[] = { 62 &vdev_root_ops, 63 &vdev_raidz_ops, 64 &vdev_mirror_ops, 65 &vdev_replacing_ops, 66 &vdev_spare_ops, 67 &vdev_disk_ops, 68 &vdev_file_ops, 69 &vdev_missing_ops, 70 &vdev_hole_ops, 71 &vdev_indirect_ops, 72 NULL 73 }; 74 75 /* maximum scrub/resilver I/O queue per leaf vdev */ 76 int zfs_scrub_limit = 10; 77 78 /* default target for number of metaslabs per top-level vdev */ 79 int zfs_vdev_default_ms_count = 200; 80 81 /* minimum number of metaslabs per top-level vdev */ 82 int zfs_vdev_min_ms_count = 16; 83 84 /* practical upper limit of total metaslabs per top-level vdev */ 85 int zfs_vdev_ms_count_limit = 1ULL << 17; 86 87 /* lower limit for metaslab size (512M) */ 88 int zfs_vdev_default_ms_shift = 29; 89 90 /* upper limit for metaslab size (16G) */ 91 int zfs_vdev_max_ms_shift = 34; 92 93 boolean_t vdev_validate_skip = B_FALSE; 94 95 /* 96 * Since the DTL space map of a vdev is not expected to have a lot of 97 * entries, we default its block size to 4K. 98 */ 99 int zfs_vdev_dtl_sm_blksz = (1 << 12); 100 101 /* 102 * vdev-wide space maps that have lots of entries written to them at 103 * the end of each transaction can benefit from a higher I/O bandwidth 104 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 105 */ 106 int zfs_vdev_standard_sm_blksz = (1 << 17); 107 108 int zfs_ashift_min; 109 110 /*PRINTFLIKE2*/ 111 void 112 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 113 { 114 va_list adx; 115 char buf[256]; 116 117 va_start(adx, fmt); 118 (void) vsnprintf(buf, sizeof (buf), fmt, adx); 119 va_end(adx); 120 121 if (vd->vdev_path != NULL) { 122 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 123 vd->vdev_path, buf); 124 } else { 125 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 126 vd->vdev_ops->vdev_op_type, 127 (u_longlong_t)vd->vdev_id, 128 (u_longlong_t)vd->vdev_guid, buf); 129 } 130 } 131 132 void 133 vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 134 { 135 char state[20]; 136 137 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 138 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 139 vd->vdev_ops->vdev_op_type); 140 return; 141 } 142 143 switch (vd->vdev_state) { 144 case VDEV_STATE_UNKNOWN: 145 (void) snprintf(state, sizeof (state), "unknown"); 146 break; 147 case VDEV_STATE_CLOSED: 148 (void) snprintf(state, sizeof (state), "closed"); 149 break; 150 case VDEV_STATE_OFFLINE: 151 (void) snprintf(state, sizeof (state), "offline"); 152 break; 153 case VDEV_STATE_REMOVED: 154 (void) snprintf(state, sizeof (state), "removed"); 155 break; 156 case VDEV_STATE_CANT_OPEN: 157 (void) snprintf(state, sizeof (state), "can't open"); 158 break; 159 case VDEV_STATE_FAULTED: 160 (void) snprintf(state, sizeof (state), "faulted"); 161 break; 162 case VDEV_STATE_DEGRADED: 163 (void) snprintf(state, sizeof (state), "degraded"); 164 break; 165 case VDEV_STATE_HEALTHY: 166 (void) snprintf(state, sizeof (state), "healthy"); 167 break; 168 default: 169 (void) snprintf(state, sizeof (state), "<state %u>", 170 (uint_t)vd->vdev_state); 171 } 172 173 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 174 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 175 vd->vdev_islog ? " (log)" : "", 176 (u_longlong_t)vd->vdev_guid, 177 vd->vdev_path ? vd->vdev_path : "N/A", state); 178 179 for (uint64_t i = 0; i < vd->vdev_children; i++) 180 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 181 } 182 183 /* 184 * Given a vdev type, return the appropriate ops vector. 185 */ 186 static vdev_ops_t * 187 vdev_getops(const char *type) 188 { 189 vdev_ops_t *ops, **opspp; 190 191 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 192 if (strcmp(ops->vdev_op_type, type) == 0) 193 break; 194 195 return (ops); 196 } 197 198 /* 199 * Derive the enumerated alloction bias from string input. 200 * String origin is either the per-vdev zap or zpool(1M). 201 */ 202 static vdev_alloc_bias_t 203 vdev_derive_alloc_bias(const char *bias) 204 { 205 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 206 207 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 208 alloc_bias = VDEV_BIAS_LOG; 209 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 210 alloc_bias = VDEV_BIAS_SPECIAL; 211 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 212 alloc_bias = VDEV_BIAS_DEDUP; 213 214 return (alloc_bias); 215 } 216 217 /* ARGSUSED */ 218 void 219 vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res) 220 { 221 res->rs_start = in->rs_start; 222 res->rs_end = in->rs_end; 223 } 224 225 /* 226 * Default asize function: return the MAX of psize with the asize of 227 * all children. This is what's used by anything other than RAID-Z. 228 */ 229 uint64_t 230 vdev_default_asize(vdev_t *vd, uint64_t psize) 231 { 232 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 233 uint64_t csize; 234 235 for (int c = 0; c < vd->vdev_children; c++) { 236 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 237 asize = MAX(asize, csize); 238 } 239 240 return (asize); 241 } 242 243 /* 244 * Get the minimum allocatable size. We define the allocatable size as 245 * the vdev's asize rounded to the nearest metaslab. This allows us to 246 * replace or attach devices which don't have the same physical size but 247 * can still satisfy the same number of allocations. 248 */ 249 uint64_t 250 vdev_get_min_asize(vdev_t *vd) 251 { 252 vdev_t *pvd = vd->vdev_parent; 253 254 /* 255 * If our parent is NULL (inactive spare or cache) or is the root, 256 * just return our own asize. 257 */ 258 if (pvd == NULL) 259 return (vd->vdev_asize); 260 261 /* 262 * The top-level vdev just returns the allocatable size rounded 263 * to the nearest metaslab. 264 */ 265 if (vd == vd->vdev_top) 266 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 267 268 /* 269 * The allocatable space for a raidz vdev is N * sizeof(smallest child), 270 * so each child must provide at least 1/Nth of its asize. 271 */ 272 if (pvd->vdev_ops == &vdev_raidz_ops) 273 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 274 pvd->vdev_children); 275 276 return (pvd->vdev_min_asize); 277 } 278 279 void 280 vdev_set_min_asize(vdev_t *vd) 281 { 282 vd->vdev_min_asize = vdev_get_min_asize(vd); 283 284 for (int c = 0; c < vd->vdev_children; c++) 285 vdev_set_min_asize(vd->vdev_child[c]); 286 } 287 288 vdev_t * 289 vdev_lookup_top(spa_t *spa, uint64_t vdev) 290 { 291 vdev_t *rvd = spa->spa_root_vdev; 292 293 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 294 295 if (vdev < rvd->vdev_children) { 296 ASSERT(rvd->vdev_child[vdev] != NULL); 297 return (rvd->vdev_child[vdev]); 298 } 299 300 return (NULL); 301 } 302 303 vdev_t * 304 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 305 { 306 vdev_t *mvd; 307 308 if (vd->vdev_guid == guid) 309 return (vd); 310 311 for (int c = 0; c < vd->vdev_children; c++) 312 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 313 NULL) 314 return (mvd); 315 316 return (NULL); 317 } 318 319 static int 320 vdev_count_leaves_impl(vdev_t *vd) 321 { 322 int n = 0; 323 324 if (vd->vdev_ops->vdev_op_leaf) 325 return (1); 326 327 for (int c = 0; c < vd->vdev_children; c++) 328 n += vdev_count_leaves_impl(vd->vdev_child[c]); 329 330 return (n); 331 } 332 333 int 334 vdev_count_leaves(spa_t *spa) 335 { 336 return (vdev_count_leaves_impl(spa->spa_root_vdev)); 337 } 338 339 void 340 vdev_add_child(vdev_t *pvd, vdev_t *cvd) 341 { 342 size_t oldsize, newsize; 343 uint64_t id = cvd->vdev_id; 344 vdev_t **newchild; 345 spa_t *spa = cvd->vdev_spa; 346 347 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 348 ASSERT(cvd->vdev_parent == NULL); 349 350 cvd->vdev_parent = pvd; 351 352 if (pvd == NULL) 353 return; 354 355 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 356 357 oldsize = pvd->vdev_children * sizeof (vdev_t *); 358 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 359 newsize = pvd->vdev_children * sizeof (vdev_t *); 360 361 newchild = kmem_zalloc(newsize, KM_SLEEP); 362 if (pvd->vdev_child != NULL) { 363 bcopy(pvd->vdev_child, newchild, oldsize); 364 kmem_free(pvd->vdev_child, oldsize); 365 } 366 367 pvd->vdev_child = newchild; 368 pvd->vdev_child[id] = cvd; 369 370 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 371 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 372 373 /* 374 * Walk up all ancestors to update guid sum. 375 */ 376 for (; pvd != NULL; pvd = pvd->vdev_parent) 377 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 378 379 if (cvd->vdev_ops->vdev_op_leaf) { 380 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 381 cvd->vdev_spa->spa_leaf_list_gen++; 382 } 383 } 384 385 void 386 vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 387 { 388 int c; 389 uint_t id = cvd->vdev_id; 390 391 ASSERT(cvd->vdev_parent == pvd); 392 393 if (pvd == NULL) 394 return; 395 396 ASSERT(id < pvd->vdev_children); 397 ASSERT(pvd->vdev_child[id] == cvd); 398 399 pvd->vdev_child[id] = NULL; 400 cvd->vdev_parent = NULL; 401 402 for (c = 0; c < pvd->vdev_children; c++) 403 if (pvd->vdev_child[c]) 404 break; 405 406 if (c == pvd->vdev_children) { 407 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 408 pvd->vdev_child = NULL; 409 pvd->vdev_children = 0; 410 } 411 412 if (cvd->vdev_ops->vdev_op_leaf) { 413 spa_t *spa = cvd->vdev_spa; 414 list_remove(&spa->spa_leaf_list, cvd); 415 spa->spa_leaf_list_gen++; 416 } 417 418 /* 419 * Walk up all ancestors to update guid sum. 420 */ 421 for (; pvd != NULL; pvd = pvd->vdev_parent) 422 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 423 } 424 425 /* 426 * Remove any holes in the child array. 427 */ 428 void 429 vdev_compact_children(vdev_t *pvd) 430 { 431 vdev_t **newchild, *cvd; 432 int oldc = pvd->vdev_children; 433 int newc; 434 435 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 436 437 for (int c = newc = 0; c < oldc; c++) 438 if (pvd->vdev_child[c]) 439 newc++; 440 441 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 442 443 for (int c = newc = 0; c < oldc; c++) { 444 if ((cvd = pvd->vdev_child[c]) != NULL) { 445 newchild[newc] = cvd; 446 cvd->vdev_id = newc++; 447 } 448 } 449 450 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 451 pvd->vdev_child = newchild; 452 pvd->vdev_children = newc; 453 } 454 455 /* 456 * Allocate and minimally initialize a vdev_t. 457 */ 458 vdev_t * 459 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 460 { 461 vdev_t *vd; 462 vdev_indirect_config_t *vic; 463 464 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 465 vic = &vd->vdev_indirect_config; 466 467 if (spa->spa_root_vdev == NULL) { 468 ASSERT(ops == &vdev_root_ops); 469 spa->spa_root_vdev = vd; 470 spa->spa_load_guid = spa_generate_guid(NULL); 471 } 472 473 if (guid == 0 && ops != &vdev_hole_ops) { 474 if (spa->spa_root_vdev == vd) { 475 /* 476 * The root vdev's guid will also be the pool guid, 477 * which must be unique among all pools. 478 */ 479 guid = spa_generate_guid(NULL); 480 } else { 481 /* 482 * Any other vdev's guid must be unique within the pool. 483 */ 484 guid = spa_generate_guid(spa); 485 } 486 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 487 } 488 489 vd->vdev_spa = spa; 490 vd->vdev_id = id; 491 vd->vdev_guid = guid; 492 vd->vdev_guid_sum = guid; 493 vd->vdev_ops = ops; 494 vd->vdev_state = VDEV_STATE_CLOSED; 495 vd->vdev_ishole = (ops == &vdev_hole_ops); 496 vic->vic_prev_indirect_vdev = UINT64_MAX; 497 498 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 499 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 500 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 501 0, 0); 502 503 list_link_init(&vd->vdev_initialize_node); 504 list_link_init(&vd->vdev_leaf_node); 505 list_link_init(&vd->vdev_trim_node); 506 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 507 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 508 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 509 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 510 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 511 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 512 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 513 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 514 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 515 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 516 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 517 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 518 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 519 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 520 521 for (int t = 0; t < DTL_TYPES; t++) { 522 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 523 0); 524 } 525 txg_list_create(&vd->vdev_ms_list, spa, 526 offsetof(struct metaslab, ms_txg_node)); 527 txg_list_create(&vd->vdev_dtl_list, spa, 528 offsetof(struct vdev, vdev_dtl_node)); 529 vd->vdev_stat.vs_timestamp = gethrtime(); 530 vdev_queue_init(vd); 531 vdev_cache_init(vd); 532 533 return (vd); 534 } 535 536 /* 537 * Allocate a new vdev. The 'alloctype' is used to control whether we are 538 * creating a new vdev or loading an existing one - the behavior is slightly 539 * different for each case. 540 */ 541 int 542 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 543 int alloctype) 544 { 545 vdev_ops_t *ops; 546 char *type; 547 uint64_t guid = 0, islog, nparity; 548 vdev_t *vd; 549 vdev_indirect_config_t *vic; 550 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 551 boolean_t top_level = (parent && !parent->vdev_parent); 552 553 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 554 555 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 556 return (SET_ERROR(EINVAL)); 557 558 if ((ops = vdev_getops(type)) == NULL) 559 return (SET_ERROR(EINVAL)); 560 561 /* 562 * If this is a load, get the vdev guid from the nvlist. 563 * Otherwise, vdev_alloc_common() will generate one for us. 564 */ 565 if (alloctype == VDEV_ALLOC_LOAD) { 566 uint64_t label_id; 567 568 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 569 label_id != id) 570 return (SET_ERROR(EINVAL)); 571 572 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 573 return (SET_ERROR(EINVAL)); 574 } else if (alloctype == VDEV_ALLOC_SPARE) { 575 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 576 return (SET_ERROR(EINVAL)); 577 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 578 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 579 return (SET_ERROR(EINVAL)); 580 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 581 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 582 return (SET_ERROR(EINVAL)); 583 } 584 585 /* 586 * The first allocated vdev must be of type 'root'. 587 */ 588 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 589 return (SET_ERROR(EINVAL)); 590 591 /* 592 * Determine whether we're a log vdev. 593 */ 594 islog = 0; 595 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 596 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 597 return (SET_ERROR(ENOTSUP)); 598 599 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 600 return (SET_ERROR(ENOTSUP)); 601 602 /* 603 * Set the nparity property for RAID-Z vdevs. 604 */ 605 nparity = -1ULL; 606 if (ops == &vdev_raidz_ops) { 607 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 608 &nparity) == 0) { 609 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 610 return (SET_ERROR(EINVAL)); 611 /* 612 * Previous versions could only support 1 or 2 parity 613 * device. 614 */ 615 if (nparity > 1 && 616 spa_version(spa) < SPA_VERSION_RAIDZ2) 617 return (SET_ERROR(ENOTSUP)); 618 if (nparity > 2 && 619 spa_version(spa) < SPA_VERSION_RAIDZ3) 620 return (SET_ERROR(ENOTSUP)); 621 } else { 622 /* 623 * We require the parity to be specified for SPAs that 624 * support multiple parity levels. 625 */ 626 if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 627 return (SET_ERROR(EINVAL)); 628 /* 629 * Otherwise, we default to 1 parity device for RAID-Z. 630 */ 631 nparity = 1; 632 } 633 } else { 634 nparity = 0; 635 } 636 ASSERT(nparity != -1ULL); 637 638 /* 639 * If creating a top-level vdev, check for allocation classes input 640 */ 641 if (top_level && alloctype == VDEV_ALLOC_ADD) { 642 char *bias; 643 644 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 645 &bias) == 0) { 646 alloc_bias = vdev_derive_alloc_bias(bias); 647 648 /* spa_vdev_add() expects feature to be enabled */ 649 if (alloc_bias != VDEV_BIAS_LOG && 650 spa->spa_load_state != SPA_LOAD_CREATE && 651 !spa_feature_is_enabled(spa, 652 SPA_FEATURE_ALLOCATION_CLASSES)) { 653 return (SET_ERROR(ENOTSUP)); 654 } 655 } 656 } 657 658 vd = vdev_alloc_common(spa, id, guid, ops); 659 vic = &vd->vdev_indirect_config; 660 661 vd->vdev_islog = islog; 662 vd->vdev_nparity = nparity; 663 if (top_level && alloc_bias != VDEV_BIAS_NONE) 664 vd->vdev_alloc_bias = alloc_bias; 665 666 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 667 vd->vdev_path = spa_strdup(vd->vdev_path); 668 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 669 vd->vdev_devid = spa_strdup(vd->vdev_devid); 670 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 671 &vd->vdev_physpath) == 0) 672 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 673 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 674 vd->vdev_fru = spa_strdup(vd->vdev_fru); 675 676 /* 677 * Set the whole_disk property. If it's not specified, leave the value 678 * as -1. 679 */ 680 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 681 &vd->vdev_wholedisk) != 0) 682 vd->vdev_wholedisk = -1ULL; 683 684 ASSERT0(vic->vic_mapping_object); 685 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 686 &vic->vic_mapping_object); 687 ASSERT0(vic->vic_births_object); 688 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 689 &vic->vic_births_object); 690 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 691 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 692 &vic->vic_prev_indirect_vdev); 693 694 /* 695 * Look for the 'not present' flag. This will only be set if the device 696 * was not present at the time of import. 697 */ 698 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 699 &vd->vdev_not_present); 700 701 /* 702 * Get the alignment requirement. 703 */ 704 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 705 706 /* 707 * Retrieve the vdev creation time. 708 */ 709 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 710 &vd->vdev_crtxg); 711 712 /* 713 * If we're a top-level vdev, try to load the allocation parameters. 714 */ 715 if (top_level && 716 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 717 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 718 &vd->vdev_ms_array); 719 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 720 &vd->vdev_ms_shift); 721 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 722 &vd->vdev_asize); 723 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 724 &vd->vdev_removing); 725 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 726 &vd->vdev_top_zap); 727 } else { 728 ASSERT0(vd->vdev_top_zap); 729 } 730 731 if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 732 ASSERT(alloctype == VDEV_ALLOC_LOAD || 733 alloctype == VDEV_ALLOC_ADD || 734 alloctype == VDEV_ALLOC_SPLIT || 735 alloctype == VDEV_ALLOC_ROOTPOOL); 736 /* Note: metaslab_group_create() is now deferred */ 737 } 738 739 if (vd->vdev_ops->vdev_op_leaf && 740 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 741 (void) nvlist_lookup_uint64(nv, 742 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 743 } else { 744 ASSERT0(vd->vdev_leaf_zap); 745 } 746 747 /* 748 * If we're a leaf vdev, try to load the DTL object and other state. 749 */ 750 751 if (vd->vdev_ops->vdev_op_leaf && 752 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 753 alloctype == VDEV_ALLOC_ROOTPOOL)) { 754 if (alloctype == VDEV_ALLOC_LOAD) { 755 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 756 &vd->vdev_dtl_object); 757 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 758 &vd->vdev_unspare); 759 } 760 761 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 762 uint64_t spare = 0; 763 764 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 765 &spare) == 0 && spare) 766 spa_spare_add(vd); 767 } 768 769 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 770 &vd->vdev_offline); 771 772 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 773 &vd->vdev_resilver_txg); 774 775 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 776 vdev_defer_resilver(vd); 777 778 /* 779 * When importing a pool, we want to ignore the persistent fault 780 * state, as the diagnosis made on another system may not be 781 * valid in the current context. Local vdevs will 782 * remain in the faulted state. 783 */ 784 if (spa_load_state(spa) == SPA_LOAD_OPEN) { 785 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 786 &vd->vdev_faulted); 787 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 788 &vd->vdev_degraded); 789 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 790 &vd->vdev_removed); 791 792 if (vd->vdev_faulted || vd->vdev_degraded) { 793 char *aux; 794 795 vd->vdev_label_aux = 796 VDEV_AUX_ERR_EXCEEDED; 797 if (nvlist_lookup_string(nv, 798 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 799 strcmp(aux, "external") == 0) 800 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 801 } 802 } 803 } 804 805 /* 806 * Add ourselves to the parent's list of children. 807 */ 808 vdev_add_child(parent, vd); 809 810 *vdp = vd; 811 812 return (0); 813 } 814 815 void 816 vdev_free(vdev_t *vd) 817 { 818 spa_t *spa = vd->vdev_spa; 819 820 ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 821 ASSERT3P(vd->vdev_trim_thread, ==, NULL); 822 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 823 824 /* 825 * Scan queues are normally destroyed at the end of a scan. If the 826 * queue exists here, that implies the vdev is being removed while 827 * the scan is still running. 828 */ 829 if (vd->vdev_scan_io_queue != NULL) { 830 mutex_enter(&vd->vdev_scan_io_queue_lock); 831 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 832 vd->vdev_scan_io_queue = NULL; 833 mutex_exit(&vd->vdev_scan_io_queue_lock); 834 } 835 836 /* 837 * vdev_free() implies closing the vdev first. This is simpler than 838 * trying to ensure complicated semantics for all callers. 839 */ 840 vdev_close(vd); 841 842 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 843 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 844 845 /* 846 * Free all children. 847 */ 848 for (int c = 0; c < vd->vdev_children; c++) 849 vdev_free(vd->vdev_child[c]); 850 851 ASSERT(vd->vdev_child == NULL); 852 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 853 854 /* 855 * Discard allocation state. 856 */ 857 if (vd->vdev_mg != NULL) { 858 vdev_metaslab_fini(vd); 859 metaslab_group_destroy(vd->vdev_mg); 860 vd->vdev_mg = NULL; 861 } 862 863 ASSERT0(vd->vdev_stat.vs_space); 864 ASSERT0(vd->vdev_stat.vs_dspace); 865 ASSERT0(vd->vdev_stat.vs_alloc); 866 867 /* 868 * Remove this vdev from its parent's child list. 869 */ 870 vdev_remove_child(vd->vdev_parent, vd); 871 872 ASSERT(vd->vdev_parent == NULL); 873 ASSERT(!list_link_active(&vd->vdev_leaf_node)); 874 875 /* 876 * Clean up vdev structure. 877 */ 878 vdev_queue_fini(vd); 879 vdev_cache_fini(vd); 880 881 if (vd->vdev_path) 882 spa_strfree(vd->vdev_path); 883 if (vd->vdev_devid) 884 spa_strfree(vd->vdev_devid); 885 if (vd->vdev_physpath) 886 spa_strfree(vd->vdev_physpath); 887 if (vd->vdev_fru) 888 spa_strfree(vd->vdev_fru); 889 890 if (vd->vdev_isspare) 891 spa_spare_remove(vd); 892 if (vd->vdev_isl2cache) 893 spa_l2cache_remove(vd); 894 895 txg_list_destroy(&vd->vdev_ms_list); 896 txg_list_destroy(&vd->vdev_dtl_list); 897 898 mutex_enter(&vd->vdev_dtl_lock); 899 space_map_close(vd->vdev_dtl_sm); 900 for (int t = 0; t < DTL_TYPES; t++) { 901 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 902 range_tree_destroy(vd->vdev_dtl[t]); 903 } 904 mutex_exit(&vd->vdev_dtl_lock); 905 906 EQUIV(vd->vdev_indirect_births != NULL, 907 vd->vdev_indirect_mapping != NULL); 908 if (vd->vdev_indirect_births != NULL) { 909 vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 910 vdev_indirect_births_close(vd->vdev_indirect_births); 911 } 912 913 if (vd->vdev_obsolete_sm != NULL) { 914 ASSERT(vd->vdev_removing || 915 vd->vdev_ops == &vdev_indirect_ops); 916 space_map_close(vd->vdev_obsolete_sm); 917 vd->vdev_obsolete_sm = NULL; 918 } 919 range_tree_destroy(vd->vdev_obsolete_segments); 920 rw_destroy(&vd->vdev_indirect_rwlock); 921 mutex_destroy(&vd->vdev_obsolete_lock); 922 923 mutex_destroy(&vd->vdev_dtl_lock); 924 mutex_destroy(&vd->vdev_stat_lock); 925 mutex_destroy(&vd->vdev_probe_lock); 926 mutex_destroy(&vd->vdev_scan_io_queue_lock); 927 mutex_destroy(&vd->vdev_initialize_lock); 928 mutex_destroy(&vd->vdev_initialize_io_lock); 929 cv_destroy(&vd->vdev_initialize_io_cv); 930 cv_destroy(&vd->vdev_initialize_cv); 931 mutex_destroy(&vd->vdev_trim_lock); 932 mutex_destroy(&vd->vdev_autotrim_lock); 933 mutex_destroy(&vd->vdev_trim_io_lock); 934 cv_destroy(&vd->vdev_trim_cv); 935 cv_destroy(&vd->vdev_autotrim_cv); 936 cv_destroy(&vd->vdev_trim_io_cv); 937 938 if (vd == spa->spa_root_vdev) 939 spa->spa_root_vdev = NULL; 940 941 kmem_free(vd, sizeof (vdev_t)); 942 } 943 944 /* 945 * Transfer top-level vdev state from svd to tvd. 946 */ 947 static void 948 vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 949 { 950 spa_t *spa = svd->vdev_spa; 951 metaslab_t *msp; 952 vdev_t *vd; 953 int t; 954 955 ASSERT(tvd == tvd->vdev_top); 956 957 tvd->vdev_ms_array = svd->vdev_ms_array; 958 tvd->vdev_ms_shift = svd->vdev_ms_shift; 959 tvd->vdev_ms_count = svd->vdev_ms_count; 960 tvd->vdev_top_zap = svd->vdev_top_zap; 961 962 svd->vdev_ms_array = 0; 963 svd->vdev_ms_shift = 0; 964 svd->vdev_ms_count = 0; 965 svd->vdev_top_zap = 0; 966 967 if (tvd->vdev_mg) 968 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 969 tvd->vdev_mg = svd->vdev_mg; 970 tvd->vdev_ms = svd->vdev_ms; 971 972 svd->vdev_mg = NULL; 973 svd->vdev_ms = NULL; 974 975 if (tvd->vdev_mg != NULL) 976 tvd->vdev_mg->mg_vd = tvd; 977 978 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 979 svd->vdev_checkpoint_sm = NULL; 980 981 tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 982 svd->vdev_alloc_bias = VDEV_BIAS_NONE; 983 984 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 985 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 986 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 987 988 svd->vdev_stat.vs_alloc = 0; 989 svd->vdev_stat.vs_space = 0; 990 svd->vdev_stat.vs_dspace = 0; 991 992 /* 993 * State which may be set on a top-level vdev that's in the 994 * process of being removed. 995 */ 996 ASSERT0(tvd->vdev_indirect_config.vic_births_object); 997 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 998 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 999 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 1000 ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 1001 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 1002 ASSERT0(tvd->vdev_removing); 1003 tvd->vdev_removing = svd->vdev_removing; 1004 tvd->vdev_indirect_config = svd->vdev_indirect_config; 1005 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 1006 tvd->vdev_indirect_births = svd->vdev_indirect_births; 1007 range_tree_swap(&svd->vdev_obsolete_segments, 1008 &tvd->vdev_obsolete_segments); 1009 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 1010 svd->vdev_indirect_config.vic_mapping_object = 0; 1011 svd->vdev_indirect_config.vic_births_object = 0; 1012 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 1013 svd->vdev_indirect_mapping = NULL; 1014 svd->vdev_indirect_births = NULL; 1015 svd->vdev_obsolete_sm = NULL; 1016 svd->vdev_removing = 0; 1017 1018 for (t = 0; t < TXG_SIZE; t++) { 1019 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1020 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1021 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1022 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1023 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1024 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1025 } 1026 1027 if (list_link_active(&svd->vdev_config_dirty_node)) { 1028 vdev_config_clean(svd); 1029 vdev_config_dirty(tvd); 1030 } 1031 1032 if (list_link_active(&svd->vdev_state_dirty_node)) { 1033 vdev_state_clean(svd); 1034 vdev_state_dirty(tvd); 1035 } 1036 1037 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 1038 svd->vdev_deflate_ratio = 0; 1039 1040 tvd->vdev_islog = svd->vdev_islog; 1041 svd->vdev_islog = 0; 1042 1043 dsl_scan_io_queue_vdev_xfer(svd, tvd); 1044 } 1045 1046 static void 1047 vdev_top_update(vdev_t *tvd, vdev_t *vd) 1048 { 1049 if (vd == NULL) 1050 return; 1051 1052 vd->vdev_top = tvd; 1053 1054 for (int c = 0; c < vd->vdev_children; c++) 1055 vdev_top_update(tvd, vd->vdev_child[c]); 1056 } 1057 1058 /* 1059 * Add a mirror/replacing vdev above an existing vdev. 1060 */ 1061 vdev_t * 1062 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1063 { 1064 spa_t *spa = cvd->vdev_spa; 1065 vdev_t *pvd = cvd->vdev_parent; 1066 vdev_t *mvd; 1067 1068 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1069 1070 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1071 1072 mvd->vdev_asize = cvd->vdev_asize; 1073 mvd->vdev_min_asize = cvd->vdev_min_asize; 1074 mvd->vdev_max_asize = cvd->vdev_max_asize; 1075 mvd->vdev_psize = cvd->vdev_psize; 1076 mvd->vdev_ashift = cvd->vdev_ashift; 1077 mvd->vdev_state = cvd->vdev_state; 1078 mvd->vdev_crtxg = cvd->vdev_crtxg; 1079 1080 vdev_remove_child(pvd, cvd); 1081 vdev_add_child(pvd, mvd); 1082 cvd->vdev_id = mvd->vdev_children; 1083 vdev_add_child(mvd, cvd); 1084 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1085 1086 if (mvd == mvd->vdev_top) 1087 vdev_top_transfer(cvd, mvd); 1088 1089 return (mvd); 1090 } 1091 1092 /* 1093 * Remove a 1-way mirror/replacing vdev from the tree. 1094 */ 1095 void 1096 vdev_remove_parent(vdev_t *cvd) 1097 { 1098 vdev_t *mvd = cvd->vdev_parent; 1099 vdev_t *pvd = mvd->vdev_parent; 1100 1101 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1102 1103 ASSERT(mvd->vdev_children == 1); 1104 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 1105 mvd->vdev_ops == &vdev_replacing_ops || 1106 mvd->vdev_ops == &vdev_spare_ops); 1107 cvd->vdev_ashift = mvd->vdev_ashift; 1108 1109 vdev_remove_child(mvd, cvd); 1110 vdev_remove_child(pvd, mvd); 1111 1112 /* 1113 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1114 * Otherwise, we could have detached an offline device, and when we 1115 * go to import the pool we'll think we have two top-level vdevs, 1116 * instead of a different version of the same top-level vdev. 1117 */ 1118 if (mvd->vdev_top == mvd) { 1119 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 1120 cvd->vdev_orig_guid = cvd->vdev_guid; 1121 cvd->vdev_guid += guid_delta; 1122 cvd->vdev_guid_sum += guid_delta; 1123 } 1124 cvd->vdev_id = mvd->vdev_id; 1125 vdev_add_child(pvd, cvd); 1126 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1127 1128 if (cvd == cvd->vdev_top) 1129 vdev_top_transfer(mvd, cvd); 1130 1131 ASSERT(mvd->vdev_children == 0); 1132 vdev_free(mvd); 1133 } 1134 1135 static void 1136 vdev_metaslab_group_create(vdev_t *vd) 1137 { 1138 spa_t *spa = vd->vdev_spa; 1139 1140 /* 1141 * metaslab_group_create was delayed until allocation bias was available 1142 */ 1143 if (vd->vdev_mg == NULL) { 1144 metaslab_class_t *mc; 1145 1146 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1147 vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1148 1149 ASSERT3U(vd->vdev_islog, ==, 1150 (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1151 1152 switch (vd->vdev_alloc_bias) { 1153 case VDEV_BIAS_LOG: 1154 mc = spa_log_class(spa); 1155 break; 1156 case VDEV_BIAS_SPECIAL: 1157 mc = spa_special_class(spa); 1158 break; 1159 case VDEV_BIAS_DEDUP: 1160 mc = spa_dedup_class(spa); 1161 break; 1162 default: 1163 mc = spa_normal_class(spa); 1164 } 1165 1166 vd->vdev_mg = metaslab_group_create(mc, vd, 1167 spa->spa_alloc_count); 1168 1169 /* 1170 * The spa ashift values currently only reflect the 1171 * general vdev classes. Class destination is late 1172 * binding so ashift checking had to wait until now 1173 */ 1174 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1175 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1176 if (vd->vdev_ashift > spa->spa_max_ashift) 1177 spa->spa_max_ashift = vd->vdev_ashift; 1178 if (vd->vdev_ashift < spa->spa_min_ashift) 1179 spa->spa_min_ashift = vd->vdev_ashift; 1180 } 1181 } 1182 } 1183 1184 int 1185 vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1186 { 1187 spa_t *spa = vd->vdev_spa; 1188 objset_t *mos = spa->spa_meta_objset; 1189 uint64_t m; 1190 uint64_t oldc = vd->vdev_ms_count; 1191 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1192 metaslab_t **mspp; 1193 int error; 1194 boolean_t expanding = (oldc != 0); 1195 1196 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1197 1198 /* 1199 * This vdev is not being allocated from yet or is a hole. 1200 */ 1201 if (vd->vdev_ms_shift == 0) 1202 return (0); 1203 1204 ASSERT(!vd->vdev_ishole); 1205 1206 ASSERT(oldc <= newc); 1207 1208 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1209 1210 if (expanding) { 1211 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1212 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1213 } 1214 1215 vd->vdev_ms = mspp; 1216 vd->vdev_ms_count = newc; 1217 for (m = oldc; m < newc; m++) { 1218 uint64_t object = 0; 1219 1220 /* 1221 * vdev_ms_array may be 0 if we are creating the "fake" 1222 * metaslabs for an indirect vdev for zdb's leak detection. 1223 * See zdb_leak_init(). 1224 */ 1225 if (txg == 0 && vd->vdev_ms_array != 0) { 1226 error = dmu_read(mos, vd->vdev_ms_array, 1227 m * sizeof (uint64_t), sizeof (uint64_t), &object, 1228 DMU_READ_PREFETCH); 1229 if (error != 0) { 1230 vdev_dbgmsg(vd, "unable to read the metaslab " 1231 "array [error=%d]", error); 1232 return (error); 1233 } 1234 } 1235 1236 #ifndef _KERNEL 1237 /* 1238 * To accomodate zdb_leak_init() fake indirect 1239 * metaslabs, we allocate a metaslab group for 1240 * indirect vdevs which normally don't have one. 1241 */ 1242 if (vd->vdev_mg == NULL) { 1243 ASSERT0(vdev_is_concrete(vd)); 1244 vdev_metaslab_group_create(vd); 1245 } 1246 #endif 1247 error = metaslab_init(vd->vdev_mg, m, object, txg, 1248 &(vd->vdev_ms[m])); 1249 if (error != 0) { 1250 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 1251 error); 1252 return (error); 1253 } 1254 } 1255 1256 if (txg == 0) 1257 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1258 1259 /* 1260 * If the vdev is being removed we don't activate 1261 * the metaslabs since we want to ensure that no new 1262 * allocations are performed on this device. 1263 */ 1264 if (!expanding && !vd->vdev_removing) { 1265 metaslab_group_activate(vd->vdev_mg); 1266 } 1267 1268 if (txg == 0) 1269 spa_config_exit(spa, SCL_ALLOC, FTAG); 1270 1271 /* 1272 * Regardless whether this vdev was just added or it is being 1273 * expanded, the metaslab count has changed. Recalculate the 1274 * block limit. 1275 */ 1276 spa_log_sm_set_blocklimit(spa); 1277 1278 return (0); 1279 } 1280 1281 void 1282 vdev_metaslab_fini(vdev_t *vd) 1283 { 1284 if (vd->vdev_checkpoint_sm != NULL) { 1285 ASSERT(spa_feature_is_active(vd->vdev_spa, 1286 SPA_FEATURE_POOL_CHECKPOINT)); 1287 space_map_close(vd->vdev_checkpoint_sm); 1288 /* 1289 * Even though we close the space map, we need to set its 1290 * pointer to NULL. The reason is that vdev_metaslab_fini() 1291 * may be called multiple times for certain operations 1292 * (i.e. when destroying a pool) so we need to ensure that 1293 * this clause never executes twice. This logic is similar 1294 * to the one used for the vdev_ms clause below. 1295 */ 1296 vd->vdev_checkpoint_sm = NULL; 1297 } 1298 1299 if (vd->vdev_ms != NULL) { 1300 metaslab_group_t *mg = vd->vdev_mg; 1301 metaslab_group_passivate(mg); 1302 1303 uint64_t count = vd->vdev_ms_count; 1304 for (uint64_t m = 0; m < count; m++) { 1305 metaslab_t *msp = vd->vdev_ms[m]; 1306 if (msp != NULL) 1307 metaslab_fini(msp); 1308 } 1309 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1310 vd->vdev_ms = NULL; 1311 1312 vd->vdev_ms_count = 0; 1313 1314 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 1315 ASSERT0(mg->mg_histogram[i]); 1316 } 1317 ASSERT0(vd->vdev_ms_count); 1318 } 1319 1320 typedef struct vdev_probe_stats { 1321 boolean_t vps_readable; 1322 boolean_t vps_writeable; 1323 int vps_flags; 1324 } vdev_probe_stats_t; 1325 1326 static void 1327 vdev_probe_done(zio_t *zio) 1328 { 1329 spa_t *spa = zio->io_spa; 1330 vdev_t *vd = zio->io_vd; 1331 vdev_probe_stats_t *vps = zio->io_private; 1332 1333 ASSERT(vd->vdev_probe_zio != NULL); 1334 1335 if (zio->io_type == ZIO_TYPE_READ) { 1336 if (zio->io_error == 0) 1337 vps->vps_readable = 1; 1338 if (zio->io_error == 0 && spa_writeable(spa)) { 1339 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1340 zio->io_offset, zio->io_size, zio->io_abd, 1341 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1342 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1343 } else { 1344 abd_free(zio->io_abd); 1345 } 1346 } else if (zio->io_type == ZIO_TYPE_WRITE) { 1347 if (zio->io_error == 0) 1348 vps->vps_writeable = 1; 1349 abd_free(zio->io_abd); 1350 } else if (zio->io_type == ZIO_TYPE_NULL) { 1351 zio_t *pio; 1352 1353 vd->vdev_cant_read |= !vps->vps_readable; 1354 vd->vdev_cant_write |= !vps->vps_writeable; 1355 1356 if (vdev_readable(vd) && 1357 (vdev_writeable(vd) || !spa_writeable(spa))) { 1358 zio->io_error = 0; 1359 } else { 1360 ASSERT(zio->io_error != 0); 1361 vdev_dbgmsg(vd, "failed probe"); 1362 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1363 spa, vd, NULL, NULL, 0, 0); 1364 zio->io_error = SET_ERROR(ENXIO); 1365 } 1366 1367 mutex_enter(&vd->vdev_probe_lock); 1368 ASSERT(vd->vdev_probe_zio == zio); 1369 vd->vdev_probe_zio = NULL; 1370 mutex_exit(&vd->vdev_probe_lock); 1371 1372 zio_link_t *zl = NULL; 1373 while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1374 if (!vdev_accessible(vd, pio)) 1375 pio->io_error = SET_ERROR(ENXIO); 1376 1377 kmem_free(vps, sizeof (*vps)); 1378 } 1379 } 1380 1381 /* 1382 * Determine whether this device is accessible. 1383 * 1384 * Read and write to several known locations: the pad regions of each 1385 * vdev label but the first, which we leave alone in case it contains 1386 * a VTOC. 1387 */ 1388 zio_t * 1389 vdev_probe(vdev_t *vd, zio_t *zio) 1390 { 1391 spa_t *spa = vd->vdev_spa; 1392 vdev_probe_stats_t *vps = NULL; 1393 zio_t *pio; 1394 1395 ASSERT(vd->vdev_ops->vdev_op_leaf); 1396 1397 /* 1398 * Don't probe the probe. 1399 */ 1400 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1401 return (NULL); 1402 1403 /* 1404 * To prevent 'probe storms' when a device fails, we create 1405 * just one probe i/o at a time. All zios that want to probe 1406 * this vdev will become parents of the probe io. 1407 */ 1408 mutex_enter(&vd->vdev_probe_lock); 1409 1410 if ((pio = vd->vdev_probe_zio) == NULL) { 1411 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1412 1413 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1414 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1415 ZIO_FLAG_TRYHARD; 1416 1417 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1418 /* 1419 * vdev_cant_read and vdev_cant_write can only 1420 * transition from TRUE to FALSE when we have the 1421 * SCL_ZIO lock as writer; otherwise they can only 1422 * transition from FALSE to TRUE. This ensures that 1423 * any zio looking at these values can assume that 1424 * failures persist for the life of the I/O. That's 1425 * important because when a device has intermittent 1426 * connectivity problems, we want to ensure that 1427 * they're ascribed to the device (ENXIO) and not 1428 * the zio (EIO). 1429 * 1430 * Since we hold SCL_ZIO as writer here, clear both 1431 * values so the probe can reevaluate from first 1432 * principles. 1433 */ 1434 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1435 vd->vdev_cant_read = B_FALSE; 1436 vd->vdev_cant_write = B_FALSE; 1437 } 1438 1439 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1440 vdev_probe_done, vps, 1441 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1442 1443 /* 1444 * We can't change the vdev state in this context, so we 1445 * kick off an async task to do it on our behalf. 1446 */ 1447 if (zio != NULL) { 1448 vd->vdev_probe_wanted = B_TRUE; 1449 spa_async_request(spa, SPA_ASYNC_PROBE); 1450 } 1451 } 1452 1453 if (zio != NULL) 1454 zio_add_child(zio, pio); 1455 1456 mutex_exit(&vd->vdev_probe_lock); 1457 1458 if (vps == NULL) { 1459 ASSERT(zio != NULL); 1460 return (NULL); 1461 } 1462 1463 for (int l = 1; l < VDEV_LABELS; l++) { 1464 zio_nowait(zio_read_phys(pio, vd, 1465 vdev_label_offset(vd->vdev_psize, l, 1466 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1467 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1468 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1469 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1470 } 1471 1472 if (zio == NULL) 1473 return (pio); 1474 1475 zio_nowait(pio); 1476 return (NULL); 1477 } 1478 1479 static void 1480 vdev_open_child(void *arg) 1481 { 1482 vdev_t *vd = arg; 1483 1484 vd->vdev_open_thread = curthread; 1485 vd->vdev_open_error = vdev_open(vd); 1486 vd->vdev_open_thread = NULL; 1487 } 1488 1489 boolean_t 1490 vdev_uses_zvols(vdev_t *vd) 1491 { 1492 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1493 strlen(ZVOL_DIR)) == 0) 1494 return (B_TRUE); 1495 for (int c = 0; c < vd->vdev_children; c++) 1496 if (vdev_uses_zvols(vd->vdev_child[c])) 1497 return (B_TRUE); 1498 return (B_FALSE); 1499 } 1500 1501 void 1502 vdev_open_children(vdev_t *vd) 1503 { 1504 taskq_t *tq; 1505 int children = vd->vdev_children; 1506 1507 /* 1508 * in order to handle pools on top of zvols, do the opens 1509 * in a single thread so that the same thread holds the 1510 * spa_namespace_lock 1511 */ 1512 if (vdev_uses_zvols(vd)) { 1513 retry_sync: 1514 for (int c = 0; c < children; c++) 1515 vd->vdev_child[c]->vdev_open_error = 1516 vdev_open(vd->vdev_child[c]); 1517 } else { 1518 tq = taskq_create("vdev_open", children, minclsyspri, 1519 children, children, TASKQ_PREPOPULATE); 1520 if (tq == NULL) 1521 goto retry_sync; 1522 1523 for (int c = 0; c < children; c++) 1524 VERIFY(taskq_dispatch(tq, vdev_open_child, 1525 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); 1526 1527 taskq_destroy(tq); 1528 } 1529 1530 vd->vdev_nonrot = B_TRUE; 1531 1532 for (int c = 0; c < children; c++) 1533 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; 1534 } 1535 1536 /* 1537 * Compute the raidz-deflation ratio. Note, we hard-code 1538 * in 128k (1 << 17) because it is the "typical" blocksize. 1539 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 1540 * otherwise it would inconsistently account for existing bp's. 1541 */ 1542 static void 1543 vdev_set_deflate_ratio(vdev_t *vd) 1544 { 1545 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 1546 vd->vdev_deflate_ratio = (1 << 17) / 1547 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 1548 } 1549 } 1550 1551 /* 1552 * Prepare a virtual device for access. 1553 */ 1554 int 1555 vdev_open(vdev_t *vd) 1556 { 1557 spa_t *spa = vd->vdev_spa; 1558 int error; 1559 uint64_t osize = 0; 1560 uint64_t max_osize = 0; 1561 uint64_t asize, max_asize, psize; 1562 uint64_t ashift = 0; 1563 1564 ASSERT(vd->vdev_open_thread == curthread || 1565 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1566 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1567 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1568 vd->vdev_state == VDEV_STATE_OFFLINE); 1569 1570 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1571 vd->vdev_cant_read = B_FALSE; 1572 vd->vdev_cant_write = B_FALSE; 1573 vd->vdev_min_asize = vdev_get_min_asize(vd); 1574 1575 /* 1576 * If this vdev is not removed, check its fault status. If it's 1577 * faulted, bail out of the open. 1578 */ 1579 if (!vd->vdev_removed && vd->vdev_faulted) { 1580 ASSERT(vd->vdev_children == 0); 1581 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1582 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1583 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1584 vd->vdev_label_aux); 1585 return (SET_ERROR(ENXIO)); 1586 } else if (vd->vdev_offline) { 1587 ASSERT(vd->vdev_children == 0); 1588 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1589 return (SET_ERROR(ENXIO)); 1590 } 1591 1592 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1593 1594 /* 1595 * Reset the vdev_reopening flag so that we actually close 1596 * the vdev on error. 1597 */ 1598 vd->vdev_reopening = B_FALSE; 1599 if (zio_injection_enabled && error == 0) 1600 error = zio_handle_device_injection(vd, NULL, ENXIO); 1601 1602 if (error) { 1603 if (vd->vdev_removed && 1604 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1605 vd->vdev_removed = B_FALSE; 1606 1607 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1608 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1609 vd->vdev_stat.vs_aux); 1610 } else { 1611 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1612 vd->vdev_stat.vs_aux); 1613 } 1614 return (error); 1615 } 1616 1617 vd->vdev_removed = B_FALSE; 1618 1619 /* 1620 * Recheck the faulted flag now that we have confirmed that 1621 * the vdev is accessible. If we're faulted, bail. 1622 */ 1623 if (vd->vdev_faulted) { 1624 ASSERT(vd->vdev_children == 0); 1625 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1626 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1627 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1628 vd->vdev_label_aux); 1629 return (SET_ERROR(ENXIO)); 1630 } 1631 1632 if (vd->vdev_degraded) { 1633 ASSERT(vd->vdev_children == 0); 1634 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1635 VDEV_AUX_ERR_EXCEEDED); 1636 } else { 1637 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1638 } 1639 1640 /* 1641 * For hole or missing vdevs we just return success. 1642 */ 1643 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1644 return (0); 1645 1646 for (int c = 0; c < vd->vdev_children; c++) { 1647 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1648 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1649 VDEV_AUX_NONE); 1650 break; 1651 } 1652 } 1653 1654 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1655 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1656 1657 if (vd->vdev_children == 0) { 1658 if (osize < SPA_MINDEVSIZE) { 1659 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1660 VDEV_AUX_TOO_SMALL); 1661 return (SET_ERROR(EOVERFLOW)); 1662 } 1663 psize = osize; 1664 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1665 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1666 VDEV_LABEL_END_SIZE); 1667 } else { 1668 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1669 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1670 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1671 VDEV_AUX_TOO_SMALL); 1672 return (SET_ERROR(EOVERFLOW)); 1673 } 1674 psize = 0; 1675 asize = osize; 1676 max_asize = max_osize; 1677 } 1678 1679 vd->vdev_psize = psize; 1680 1681 /* 1682 * Make sure the allocatable size hasn't shrunk too much. 1683 */ 1684 if (asize < vd->vdev_min_asize) { 1685 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1686 VDEV_AUX_BAD_LABEL); 1687 return (SET_ERROR(EINVAL)); 1688 } 1689 1690 if (vd->vdev_asize == 0) { 1691 /* 1692 * This is the first-ever open, so use the computed values. 1693 * For compatibility, a different ashift can be requested. 1694 */ 1695 vd->vdev_asize = asize; 1696 vd->vdev_max_asize = max_asize; 1697 if (vd->vdev_ashift == 0) { 1698 vd->vdev_ashift = ashift; /* use detected value */ 1699 } 1700 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 1701 vd->vdev_ashift > ASHIFT_MAX)) { 1702 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1703 VDEV_AUX_BAD_ASHIFT); 1704 return (SET_ERROR(EDOM)); 1705 } 1706 } else { 1707 /* 1708 * Detect if the alignment requirement has increased. 1709 * We don't want to make the pool unavailable, just 1710 * post an event instead. 1711 */ 1712 if (ashift > vd->vdev_top->vdev_ashift && 1713 vd->vdev_ops->vdev_op_leaf) { 1714 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 1715 spa, vd, NULL, NULL, 0, 0); 1716 } 1717 1718 vd->vdev_max_asize = max_asize; 1719 } 1720 1721 /* 1722 * If all children are healthy we update asize if either: 1723 * The asize has increased, due to a device expansion caused by dynamic 1724 * LUN growth or vdev replacement, and automatic expansion is enabled; 1725 * making the additional space available. 1726 * 1727 * The asize has decreased, due to a device shrink usually caused by a 1728 * vdev replace with a smaller device. This ensures that calculations 1729 * based of max_asize and asize e.g. esize are always valid. It's safe 1730 * to do this as we've already validated that asize is greater than 1731 * vdev_min_asize. 1732 */ 1733 if (vd->vdev_state == VDEV_STATE_HEALTHY && 1734 ((asize > vd->vdev_asize && 1735 (vd->vdev_expanding || spa->spa_autoexpand)) || 1736 (asize < vd->vdev_asize))) 1737 vd->vdev_asize = asize; 1738 1739 vdev_set_min_asize(vd); 1740 1741 /* 1742 * Ensure we can issue some IO before declaring the 1743 * vdev open for business. 1744 */ 1745 if (vd->vdev_ops->vdev_op_leaf && 1746 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1747 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1748 VDEV_AUX_ERR_EXCEEDED); 1749 return (error); 1750 } 1751 1752 /* 1753 * Track the min and max ashift values for normal data devices. 1754 * 1755 * DJB - TBD these should perhaps be tracked per allocation class 1756 * (e.g. spa_min_ashift is used to round up post compression buffers) 1757 */ 1758 if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1759 vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1760 vd->vdev_aux == NULL) { 1761 if (vd->vdev_ashift > spa->spa_max_ashift) 1762 spa->spa_max_ashift = vd->vdev_ashift; 1763 if (vd->vdev_ashift < spa->spa_min_ashift) 1764 spa->spa_min_ashift = vd->vdev_ashift; 1765 } 1766 1767 /* 1768 * If this is a leaf vdev, assess whether a resilver is needed. 1769 * But don't do this if we are doing a reopen for a scrub, since 1770 * this would just restart the scrub we are already doing. 1771 */ 1772 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) 1773 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); 1774 1775 return (0); 1776 } 1777 1778 /* 1779 * Called once the vdevs are all opened, this routine validates the label 1780 * contents. This needs to be done before vdev_load() so that we don't 1781 * inadvertently do repair I/Os to the wrong device. 1782 * 1783 * This function will only return failure if one of the vdevs indicates that it 1784 * has since been destroyed or exported. This is only possible if 1785 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1786 * will be updated but the function will return 0. 1787 */ 1788 int 1789 vdev_validate(vdev_t *vd) 1790 { 1791 spa_t *spa = vd->vdev_spa; 1792 nvlist_t *label; 1793 uint64_t guid = 0, aux_guid = 0, top_guid; 1794 uint64_t state; 1795 nvlist_t *nvl; 1796 uint64_t txg; 1797 1798 if (vdev_validate_skip) 1799 return (0); 1800 1801 for (uint64_t c = 0; c < vd->vdev_children; c++) 1802 if (vdev_validate(vd->vdev_child[c]) != 0) 1803 return (SET_ERROR(EBADF)); 1804 1805 /* 1806 * If the device has already failed, or was marked offline, don't do 1807 * any further validation. Otherwise, label I/O will fail and we will 1808 * overwrite the previous state. 1809 */ 1810 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 1811 return (0); 1812 1813 /* 1814 * If we are performing an extreme rewind, we allow for a label that 1815 * was modified at a point after the current txg. 1816 * If config lock is not held do not check for the txg. spa_sync could 1817 * be updating the vdev's label before updating spa_last_synced_txg. 1818 */ 1819 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1820 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 1821 txg = UINT64_MAX; 1822 else 1823 txg = spa_last_synced_txg(spa); 1824 1825 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1826 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1827 VDEV_AUX_BAD_LABEL); 1828 vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1829 "txg %llu", (u_longlong_t)txg); 1830 return (0); 1831 } 1832 1833 /* 1834 * Determine if this vdev has been split off into another 1835 * pool. If so, then refuse to open it. 1836 */ 1837 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1838 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1839 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1840 VDEV_AUX_SPLIT_POOL); 1841 nvlist_free(label); 1842 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 1843 return (0); 1844 } 1845 1846 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 1847 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1848 VDEV_AUX_CORRUPT_DATA); 1849 nvlist_free(label); 1850 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1851 ZPOOL_CONFIG_POOL_GUID); 1852 return (0); 1853 } 1854 1855 /* 1856 * If config is not trusted then ignore the spa guid check. This is 1857 * necessary because if the machine crashed during a re-guid the new 1858 * guid might have been written to all of the vdev labels, but not the 1859 * cached config. The check will be performed again once we have the 1860 * trusted config from the MOS. 1861 */ 1862 if (spa->spa_trust_config && guid != spa_guid(spa)) { 1863 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1864 VDEV_AUX_CORRUPT_DATA); 1865 nvlist_free(label); 1866 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 1867 "match config (%llu != %llu)", (u_longlong_t)guid, 1868 (u_longlong_t)spa_guid(spa)); 1869 return (0); 1870 } 1871 1872 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1873 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1874 &aux_guid) != 0) 1875 aux_guid = 0; 1876 1877 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 1878 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1879 VDEV_AUX_CORRUPT_DATA); 1880 nvlist_free(label); 1881 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1882 ZPOOL_CONFIG_GUID); 1883 return (0); 1884 } 1885 1886 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 1887 != 0) { 1888 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1889 VDEV_AUX_CORRUPT_DATA); 1890 nvlist_free(label); 1891 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1892 ZPOOL_CONFIG_TOP_GUID); 1893 return (0); 1894 } 1895 1896 /* 1897 * If this vdev just became a top-level vdev because its sibling was 1898 * detached, it will have adopted the parent's vdev guid -- but the 1899 * label may or may not be on disk yet. Fortunately, either version 1900 * of the label will have the same top guid, so if we're a top-level 1901 * vdev, we can safely compare to that instead. 1902 * However, if the config comes from a cachefile that failed to update 1903 * after the detach, a top-level vdev will appear as a non top-level 1904 * vdev in the config. Also relax the constraints if we perform an 1905 * extreme rewind. 1906 * 1907 * If we split this vdev off instead, then we also check the 1908 * original pool's guid. We don't want to consider the vdev 1909 * corrupt if it is partway through a split operation. 1910 */ 1911 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 1912 boolean_t mismatch = B_FALSE; 1913 if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 1914 if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 1915 mismatch = B_TRUE; 1916 } else { 1917 if (vd->vdev_guid != top_guid && 1918 vd->vdev_top->vdev_guid != guid) 1919 mismatch = B_TRUE; 1920 } 1921 1922 if (mismatch) { 1923 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1924 VDEV_AUX_CORRUPT_DATA); 1925 nvlist_free(label); 1926 vdev_dbgmsg(vd, "vdev_validate: config guid " 1927 "doesn't match label guid"); 1928 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 1929 (u_longlong_t)vd->vdev_guid, 1930 (u_longlong_t)vd->vdev_top->vdev_guid); 1931 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 1932 "aux_guid %llu", (u_longlong_t)guid, 1933 (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1934 return (0); 1935 } 1936 } 1937 1938 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1939 &state) != 0) { 1940 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1941 VDEV_AUX_CORRUPT_DATA); 1942 nvlist_free(label); 1943 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1944 ZPOOL_CONFIG_POOL_STATE); 1945 return (0); 1946 } 1947 1948 nvlist_free(label); 1949 1950 /* 1951 * If this is a verbatim import, no need to check the 1952 * state of the pool. 1953 */ 1954 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1955 spa_load_state(spa) == SPA_LOAD_OPEN && 1956 state != POOL_STATE_ACTIVE) { 1957 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 1958 "for spa %s", (u_longlong_t)state, spa->spa_name); 1959 return (SET_ERROR(EBADF)); 1960 } 1961 1962 /* 1963 * If we were able to open and validate a vdev that was 1964 * previously marked permanently unavailable, clear that state 1965 * now. 1966 */ 1967 if (vd->vdev_not_present) 1968 vd->vdev_not_present = 0; 1969 1970 return (0); 1971 } 1972 1973 static void 1974 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 1975 { 1976 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 1977 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 1978 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 1979 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 1980 dvd->vdev_path, svd->vdev_path); 1981 spa_strfree(dvd->vdev_path); 1982 dvd->vdev_path = spa_strdup(svd->vdev_path); 1983 } 1984 } else if (svd->vdev_path != NULL) { 1985 dvd->vdev_path = spa_strdup(svd->vdev_path); 1986 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 1987 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 1988 } 1989 } 1990 1991 /* 1992 * Recursively copy vdev paths from one vdev to another. Source and destination 1993 * vdev trees must have same geometry otherwise return error. Intended to copy 1994 * paths from userland config into MOS config. 1995 */ 1996 int 1997 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 1998 { 1999 if ((svd->vdev_ops == &vdev_missing_ops) || 2000 (svd->vdev_ishole && dvd->vdev_ishole) || 2001 (dvd->vdev_ops == &vdev_indirect_ops)) 2002 return (0); 2003 2004 if (svd->vdev_ops != dvd->vdev_ops) { 2005 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 2006 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 2007 return (SET_ERROR(EINVAL)); 2008 } 2009 2010 if (svd->vdev_guid != dvd->vdev_guid) { 2011 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 2012 "%llu)", (u_longlong_t)svd->vdev_guid, 2013 (u_longlong_t)dvd->vdev_guid); 2014 return (SET_ERROR(EINVAL)); 2015 } 2016 2017 if (svd->vdev_children != dvd->vdev_children) { 2018 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 2019 "%llu != %llu", (u_longlong_t)svd->vdev_children, 2020 (u_longlong_t)dvd->vdev_children); 2021 return (SET_ERROR(EINVAL)); 2022 } 2023 2024 for (uint64_t i = 0; i < svd->vdev_children; i++) { 2025 int error = vdev_copy_path_strict(svd->vdev_child[i], 2026 dvd->vdev_child[i]); 2027 if (error != 0) 2028 return (error); 2029 } 2030 2031 if (svd->vdev_ops->vdev_op_leaf) 2032 vdev_copy_path_impl(svd, dvd); 2033 2034 return (0); 2035 } 2036 2037 static void 2038 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 2039 { 2040 ASSERT(stvd->vdev_top == stvd); 2041 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 2042 2043 for (uint64_t i = 0; i < dvd->vdev_children; i++) { 2044 vdev_copy_path_search(stvd, dvd->vdev_child[i]); 2045 } 2046 2047 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 2048 return; 2049 2050 /* 2051 * The idea here is that while a vdev can shift positions within 2052 * a top vdev (when replacing, attaching mirror, etc.) it cannot 2053 * step outside of it. 2054 */ 2055 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 2056 2057 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 2058 return; 2059 2060 ASSERT(vd->vdev_ops->vdev_op_leaf); 2061 2062 vdev_copy_path_impl(vd, dvd); 2063 } 2064 2065 /* 2066 * Recursively copy vdev paths from one root vdev to another. Source and 2067 * destination vdev trees may differ in geometry. For each destination leaf 2068 * vdev, search a vdev with the same guid and top vdev id in the source. 2069 * Intended to copy paths from userland config into MOS config. 2070 */ 2071 void 2072 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 2073 { 2074 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 2075 ASSERT(srvd->vdev_ops == &vdev_root_ops); 2076 ASSERT(drvd->vdev_ops == &vdev_root_ops); 2077 2078 for (uint64_t i = 0; i < children; i++) { 2079 vdev_copy_path_search(srvd->vdev_child[i], 2080 drvd->vdev_child[i]); 2081 } 2082 } 2083 2084 /* 2085 * Close a virtual device. 2086 */ 2087 void 2088 vdev_close(vdev_t *vd) 2089 { 2090 spa_t *spa = vd->vdev_spa; 2091 vdev_t *pvd = vd->vdev_parent; 2092 2093 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2094 2095 /* 2096 * If our parent is reopening, then we are as well, unless we are 2097 * going offline. 2098 */ 2099 if (pvd != NULL && pvd->vdev_reopening) 2100 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2101 2102 vd->vdev_ops->vdev_op_close(vd); 2103 2104 vdev_cache_purge(vd); 2105 2106 /* 2107 * We record the previous state before we close it, so that if we are 2108 * doing a reopen(), we don't generate FMA ereports if we notice that 2109 * it's still faulted. 2110 */ 2111 vd->vdev_prevstate = vd->vdev_state; 2112 2113 if (vd->vdev_offline) 2114 vd->vdev_state = VDEV_STATE_OFFLINE; 2115 else 2116 vd->vdev_state = VDEV_STATE_CLOSED; 2117 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2118 } 2119 2120 void 2121 vdev_hold(vdev_t *vd) 2122 { 2123 spa_t *spa = vd->vdev_spa; 2124 2125 ASSERT(spa_is_root(spa)); 2126 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2127 return; 2128 2129 for (int c = 0; c < vd->vdev_children; c++) 2130 vdev_hold(vd->vdev_child[c]); 2131 2132 if (vd->vdev_ops->vdev_op_leaf) 2133 vd->vdev_ops->vdev_op_hold(vd); 2134 } 2135 2136 void 2137 vdev_rele(vdev_t *vd) 2138 { 2139 spa_t *spa = vd->vdev_spa; 2140 2141 ASSERT(spa_is_root(spa)); 2142 for (int c = 0; c < vd->vdev_children; c++) 2143 vdev_rele(vd->vdev_child[c]); 2144 2145 if (vd->vdev_ops->vdev_op_leaf) 2146 vd->vdev_ops->vdev_op_rele(vd); 2147 } 2148 2149 /* 2150 * Reopen all interior vdevs and any unopened leaves. We don't actually 2151 * reopen leaf vdevs which had previously been opened as they might deadlock 2152 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2153 * If the leaf has never been opened then open it, as usual. 2154 */ 2155 void 2156 vdev_reopen(vdev_t *vd) 2157 { 2158 spa_t *spa = vd->vdev_spa; 2159 2160 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2161 2162 /* set the reopening flag unless we're taking the vdev offline */ 2163 vd->vdev_reopening = !vd->vdev_offline; 2164 vdev_close(vd); 2165 (void) vdev_open(vd); 2166 2167 /* 2168 * Call vdev_validate() here to make sure we have the same device. 2169 * Otherwise, a device with an invalid label could be successfully 2170 * opened in response to vdev_reopen(). 2171 */ 2172 if (vd->vdev_aux) { 2173 (void) vdev_validate_aux(vd); 2174 if (vdev_readable(vd) && vdev_writeable(vd) && 2175 vd->vdev_aux == &spa->spa_l2cache && 2176 !l2arc_vdev_present(vd)) 2177 l2arc_add_vdev(spa, vd); 2178 } else { 2179 (void) vdev_validate(vd); 2180 } 2181 2182 /* 2183 * Reassess parent vdev's health. 2184 */ 2185 vdev_propagate_state(vd); 2186 } 2187 2188 int 2189 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2190 { 2191 int error; 2192 2193 /* 2194 * Normally, partial opens (e.g. of a mirror) are allowed. 2195 * For a create, however, we want to fail the request if 2196 * there are any components we can't open. 2197 */ 2198 error = vdev_open(vd); 2199 2200 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2201 vdev_close(vd); 2202 return (error ? error : ENXIO); 2203 } 2204 2205 /* 2206 * Recursively load DTLs and initialize all labels. 2207 */ 2208 if ((error = vdev_dtl_load(vd)) != 0 || 2209 (error = vdev_label_init(vd, txg, isreplacing ? 2210 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2211 vdev_close(vd); 2212 return (error); 2213 } 2214 2215 return (0); 2216 } 2217 2218 void 2219 vdev_metaslab_set_size(vdev_t *vd) 2220 { 2221 uint64_t asize = vd->vdev_asize; 2222 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2223 uint64_t ms_shift; 2224 2225 /* BEGIN CSTYLED */ 2226 /* 2227 * There are two dimensions to the metaslab sizing calculation: 2228 * the size of the metaslab and the count of metaslabs per vdev. 2229 * 2230 * The default values used below are a good balance between memory 2231 * usage (larger metaslab size means more memory needed for loaded 2232 * metaslabs; more metaslabs means more memory needed for the 2233 * metaslab_t structs), metaslab load time (larger metaslabs take 2234 * longer to load), and metaslab sync time (more metaslabs means 2235 * more time spent syncing all of them). 2236 * 2237 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2238 * The range of the dimensions are as follows: 2239 * 2240 * 2^29 <= ms_size <= 2^34 2241 * 16 <= ms_count <= 131,072 2242 * 2243 * On the lower end of vdev sizes, we aim for metaslabs sizes of 2244 * at least 512MB (2^29) to minimize fragmentation effects when 2245 * testing with smaller devices. However, the count constraint 2246 * of at least 16 metaslabs will override this minimum size goal. 2247 * 2248 * On the upper end of vdev sizes, we aim for a maximum metaslab 2249 * size of 16GB. However, we will cap the total count to 2^17 2250 * metaslabs to keep our memory footprint in check and let the 2251 * metaslab size grow from there if that limit is hit. 2252 * 2253 * The net effect of applying above constrains is summarized below. 2254 * 2255 * vdev size metaslab count 2256 * --------------|----------------- 2257 * < 8GB ~16 2258 * 8GB - 100GB one per 512MB 2259 * 100GB - 3TB ~200 2260 * 3TB - 2PB one per 16GB 2261 * > 2PB ~131,072 2262 * -------------------------------- 2263 * 2264 * Finally, note that all of the above calculate the initial 2265 * number of metaslabs. Expanding a top-level vdev will result 2266 * in additional metaslabs being allocated making it possible 2267 * to exceed the zfs_vdev_ms_count_limit. 2268 */ 2269 /* END CSTYLED */ 2270 2271 if (ms_count < zfs_vdev_min_ms_count) 2272 ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2273 else if (ms_count > zfs_vdev_default_ms_count) 2274 ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2275 else 2276 ms_shift = zfs_vdev_default_ms_shift; 2277 2278 if (ms_shift < SPA_MAXBLOCKSHIFT) { 2279 ms_shift = SPA_MAXBLOCKSHIFT; 2280 } else if (ms_shift > zfs_vdev_max_ms_shift) { 2281 ms_shift = zfs_vdev_max_ms_shift; 2282 /* cap the total count to constrain memory footprint */ 2283 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2284 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 2285 } 2286 2287 vd->vdev_ms_shift = ms_shift; 2288 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2289 } 2290 2291 void 2292 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2293 { 2294 ASSERT(vd == vd->vdev_top); 2295 /* indirect vdevs don't have metaslabs or dtls */ 2296 ASSERT(vdev_is_concrete(vd) || flags == 0); 2297 ASSERT(ISP2(flags)); 2298 ASSERT(spa_writeable(vd->vdev_spa)); 2299 2300 if (flags & VDD_METASLAB) 2301 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2302 2303 if (flags & VDD_DTL) 2304 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2305 2306 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2307 } 2308 2309 void 2310 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 2311 { 2312 for (int c = 0; c < vd->vdev_children; c++) 2313 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 2314 2315 if (vd->vdev_ops->vdev_op_leaf) 2316 vdev_dirty(vd->vdev_top, flags, vd, txg); 2317 } 2318 2319 /* 2320 * DTLs. 2321 * 2322 * A vdev's DTL (dirty time log) is the set of transaction groups for which 2323 * the vdev has less than perfect replication. There are four kinds of DTL: 2324 * 2325 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 2326 * 2327 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 2328 * 2329 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 2330 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 2331 * txgs that was scrubbed. 2332 * 2333 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 2334 * persistent errors or just some device being offline. 2335 * Unlike the other three, the DTL_OUTAGE map is not generally 2336 * maintained; it's only computed when needed, typically to 2337 * determine whether a device can be detached. 2338 * 2339 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 2340 * either has the data or it doesn't. 2341 * 2342 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 2343 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 2344 * if any child is less than fully replicated, then so is its parent. 2345 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 2346 * comprising only those txgs which appear in 'maxfaults' or more children; 2347 * those are the txgs we don't have enough replication to read. For example, 2348 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 2349 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 2350 * two child DTL_MISSING maps. 2351 * 2352 * It should be clear from the above that to compute the DTLs and outage maps 2353 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 2354 * Therefore, that is all we keep on disk. When loading the pool, or after 2355 * a configuration change, we generate all other DTLs from first principles. 2356 */ 2357 void 2358 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2359 { 2360 range_tree_t *rt = vd->vdev_dtl[t]; 2361 2362 ASSERT(t < DTL_TYPES); 2363 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2364 ASSERT(spa_writeable(vd->vdev_spa)); 2365 2366 mutex_enter(&vd->vdev_dtl_lock); 2367 if (!range_tree_contains(rt, txg, size)) 2368 range_tree_add(rt, txg, size); 2369 mutex_exit(&vd->vdev_dtl_lock); 2370 } 2371 2372 boolean_t 2373 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2374 { 2375 range_tree_t *rt = vd->vdev_dtl[t]; 2376 boolean_t dirty = B_FALSE; 2377 2378 ASSERT(t < DTL_TYPES); 2379 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2380 2381 /* 2382 * While we are loading the pool, the DTLs have not been loaded yet. 2383 * Ignore the DTLs and try all devices. This avoids a recursive 2384 * mutex enter on the vdev_dtl_lock, and also makes us try hard 2385 * when loading the pool (relying on the checksum to ensure that 2386 * we get the right data -- note that we while loading, we are 2387 * only reading the MOS, which is always checksummed). 2388 */ 2389 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 2390 return (B_FALSE); 2391 2392 mutex_enter(&vd->vdev_dtl_lock); 2393 if (!range_tree_is_empty(rt)) 2394 dirty = range_tree_contains(rt, txg, size); 2395 mutex_exit(&vd->vdev_dtl_lock); 2396 2397 return (dirty); 2398 } 2399 2400 boolean_t 2401 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 2402 { 2403 range_tree_t *rt = vd->vdev_dtl[t]; 2404 boolean_t empty; 2405 2406 mutex_enter(&vd->vdev_dtl_lock); 2407 empty = range_tree_is_empty(rt); 2408 mutex_exit(&vd->vdev_dtl_lock); 2409 2410 return (empty); 2411 } 2412 2413 /* 2414 * Returns B_TRUE if vdev determines offset needs to be resilvered. 2415 */ 2416 boolean_t 2417 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2418 { 2419 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2420 2421 if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2422 vd->vdev_ops->vdev_op_leaf) 2423 return (B_TRUE); 2424 2425 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2426 } 2427 2428 /* 2429 * Returns the lowest txg in the DTL range. 2430 */ 2431 static uint64_t 2432 vdev_dtl_min(vdev_t *vd) 2433 { 2434 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2435 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2436 ASSERT0(vd->vdev_children); 2437 2438 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); 2439 } 2440 2441 /* 2442 * Returns the highest txg in the DTL. 2443 */ 2444 static uint64_t 2445 vdev_dtl_max(vdev_t *vd) 2446 { 2447 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 2448 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2449 ASSERT0(vd->vdev_children); 2450 2451 return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); 2452 } 2453 2454 /* 2455 * Determine if a resilvering vdev should remove any DTL entries from 2456 * its range. If the vdev was resilvering for the entire duration of the 2457 * scan then it should excise that range from its DTLs. Otherwise, this 2458 * vdev is considered partially resilvered and should leave its DTL 2459 * entries intact. The comment in vdev_dtl_reassess() describes how we 2460 * excise the DTLs. 2461 */ 2462 static boolean_t 2463 vdev_dtl_should_excise(vdev_t *vd) 2464 { 2465 spa_t *spa = vd->vdev_spa; 2466 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2467 2468 ASSERT0(scn->scn_phys.scn_errors); 2469 ASSERT0(vd->vdev_children); 2470 2471 if (vd->vdev_state < VDEV_STATE_DEGRADED) 2472 return (B_FALSE); 2473 2474 if (vd->vdev_resilver_deferred) 2475 return (B_FALSE); 2476 2477 if (vd->vdev_resilver_txg == 0 || 2478 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2479 return (B_TRUE); 2480 2481 /* 2482 * When a resilver is initiated the scan will assign the scn_max_txg 2483 * value to the highest txg value that exists in all DTLs. If this 2484 * device's max DTL is not part of this scan (i.e. it is not in 2485 * the range (scn_min_txg, scn_max_txg] then it is not eligible 2486 * for excision. 2487 */ 2488 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2489 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2490 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2491 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2492 return (B_TRUE); 2493 } 2494 return (B_FALSE); 2495 } 2496 2497 /* 2498 * Reassess DTLs after a config change or scrub completion. 2499 */ 2500 void 2501 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2502 { 2503 spa_t *spa = vd->vdev_spa; 2504 avl_tree_t reftree; 2505 int minref; 2506 2507 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2508 2509 for (int c = 0; c < vd->vdev_children; c++) 2510 vdev_dtl_reassess(vd->vdev_child[c], txg, 2511 scrub_txg, scrub_done); 2512 2513 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 2514 return; 2515 2516 if (vd->vdev_ops->vdev_op_leaf) { 2517 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2518 2519 mutex_enter(&vd->vdev_dtl_lock); 2520 2521 /* 2522 * If we've completed a scan cleanly then determine 2523 * if this vdev should remove any DTLs. We only want to 2524 * excise regions on vdevs that were available during 2525 * the entire duration of this scan. 2526 */ 2527 if (scrub_txg != 0 && 2528 (spa->spa_scrub_started || 2529 (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2530 vdev_dtl_should_excise(vd)) { 2531 /* 2532 * We completed a scrub up to scrub_txg. If we 2533 * did it without rebooting, then the scrub dtl 2534 * will be valid, so excise the old region and 2535 * fold in the scrub dtl. Otherwise, leave the 2536 * dtl as-is if there was an error. 2537 * 2538 * There's little trick here: to excise the beginning 2539 * of the DTL_MISSING map, we put it into a reference 2540 * tree and then add a segment with refcnt -1 that 2541 * covers the range [0, scrub_txg). This means 2542 * that each txg in that range has refcnt -1 or 0. 2543 * We then add DTL_SCRUB with a refcnt of 2, so that 2544 * entries in the range [0, scrub_txg) will have a 2545 * positive refcnt -- either 1 or 2. We then convert 2546 * the reference tree into the new DTL_MISSING map. 2547 */ 2548 space_reftree_create(&reftree); 2549 space_reftree_add_map(&reftree, 2550 vd->vdev_dtl[DTL_MISSING], 1); 2551 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 2552 space_reftree_add_map(&reftree, 2553 vd->vdev_dtl[DTL_SCRUB], 2); 2554 space_reftree_generate_map(&reftree, 2555 vd->vdev_dtl[DTL_MISSING], 1); 2556 space_reftree_destroy(&reftree); 2557 } 2558 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 2559 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2560 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2561 if (scrub_done) 2562 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 2563 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 2564 if (!vdev_readable(vd)) 2565 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 2566 else 2567 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 2568 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2569 2570 /* 2571 * If the vdev was resilvering and no longer has any 2572 * DTLs then reset its resilvering flag. 2573 */ 2574 if (vd->vdev_resilver_txg != 0 && 2575 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2576 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2577 vd->vdev_resilver_txg = 0; 2578 2579 mutex_exit(&vd->vdev_dtl_lock); 2580 2581 if (txg != 0) 2582 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2583 return; 2584 } 2585 2586 mutex_enter(&vd->vdev_dtl_lock); 2587 for (int t = 0; t < DTL_TYPES; t++) { 2588 /* account for child's outage in parent's missing map */ 2589 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 2590 if (t == DTL_SCRUB) 2591 continue; /* leaf vdevs only */ 2592 if (t == DTL_PARTIAL) 2593 minref = 1; /* i.e. non-zero */ 2594 else if (vd->vdev_nparity != 0) 2595 minref = vd->vdev_nparity + 1; /* RAID-Z */ 2596 else 2597 minref = vd->vdev_children; /* any kind of mirror */ 2598 space_reftree_create(&reftree); 2599 for (int c = 0; c < vd->vdev_children; c++) { 2600 vdev_t *cvd = vd->vdev_child[c]; 2601 mutex_enter(&cvd->vdev_dtl_lock); 2602 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 2603 mutex_exit(&cvd->vdev_dtl_lock); 2604 } 2605 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 2606 space_reftree_destroy(&reftree); 2607 } 2608 mutex_exit(&vd->vdev_dtl_lock); 2609 } 2610 2611 int 2612 vdev_dtl_load(vdev_t *vd) 2613 { 2614 spa_t *spa = vd->vdev_spa; 2615 objset_t *mos = spa->spa_meta_objset; 2616 int error = 0; 2617 2618 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 2619 ASSERT(vdev_is_concrete(vd)); 2620 2621 error = space_map_open(&vd->vdev_dtl_sm, mos, 2622 vd->vdev_dtl_object, 0, -1ULL, 0); 2623 if (error) 2624 return (error); 2625 ASSERT(vd->vdev_dtl_sm != NULL); 2626 2627 mutex_enter(&vd->vdev_dtl_lock); 2628 error = space_map_load(vd->vdev_dtl_sm, 2629 vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 2630 mutex_exit(&vd->vdev_dtl_lock); 2631 2632 return (error); 2633 } 2634 2635 for (int c = 0; c < vd->vdev_children; c++) { 2636 error = vdev_dtl_load(vd->vdev_child[c]); 2637 if (error != 0) 2638 break; 2639 } 2640 2641 return (error); 2642 } 2643 2644 static void 2645 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2646 { 2647 spa_t *spa = vd->vdev_spa; 2648 objset_t *mos = spa->spa_meta_objset; 2649 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2650 const char *string; 2651 2652 ASSERT(alloc_bias != VDEV_BIAS_NONE); 2653 2654 string = 2655 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2656 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2657 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2658 2659 ASSERT(string != NULL); 2660 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2661 1, strlen(string) + 1, string, tx)); 2662 2663 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2664 spa_activate_allocation_classes(spa, tx); 2665 } 2666 } 2667 2668 void 2669 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2670 { 2671 spa_t *spa = vd->vdev_spa; 2672 2673 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2674 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2675 zapobj, tx)); 2676 } 2677 2678 uint64_t 2679 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2680 { 2681 spa_t *spa = vd->vdev_spa; 2682 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2683 DMU_OT_NONE, 0, tx); 2684 2685 ASSERT(zap != 0); 2686 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2687 zap, tx)); 2688 2689 return (zap); 2690 } 2691 2692 void 2693 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2694 { 2695 if (vd->vdev_ops != &vdev_hole_ops && 2696 vd->vdev_ops != &vdev_missing_ops && 2697 vd->vdev_ops != &vdev_root_ops && 2698 !vd->vdev_top->vdev_removing) { 2699 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2700 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2701 } 2702 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2703 vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2704 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2705 vdev_zap_allocation_data(vd, tx); 2706 } 2707 } 2708 2709 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2710 vdev_construct_zaps(vd->vdev_child[i], tx); 2711 } 2712 } 2713 2714 void 2715 vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2716 { 2717 spa_t *spa = vd->vdev_spa; 2718 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2719 objset_t *mos = spa->spa_meta_objset; 2720 range_tree_t *rtsync; 2721 dmu_tx_t *tx; 2722 uint64_t object = space_map_object(vd->vdev_dtl_sm); 2723 2724 ASSERT(vdev_is_concrete(vd)); 2725 ASSERT(vd->vdev_ops->vdev_op_leaf); 2726 2727 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2728 2729 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 2730 mutex_enter(&vd->vdev_dtl_lock); 2731 space_map_free(vd->vdev_dtl_sm, tx); 2732 space_map_close(vd->vdev_dtl_sm); 2733 vd->vdev_dtl_sm = NULL; 2734 mutex_exit(&vd->vdev_dtl_lock); 2735 2736 /* 2737 * We only destroy the leaf ZAP for detached leaves or for 2738 * removed log devices. Removed data devices handle leaf ZAP 2739 * cleanup later, once cancellation is no longer possible. 2740 */ 2741 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2742 vd->vdev_top->vdev_islog)) { 2743 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2744 vd->vdev_leaf_zap = 0; 2745 } 2746 2747 dmu_tx_commit(tx); 2748 return; 2749 } 2750 2751 if (vd->vdev_dtl_sm == NULL) { 2752 uint64_t new_object; 2753 2754 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); 2755 VERIFY3U(new_object, !=, 0); 2756 2757 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 2758 0, -1ULL, 0)); 2759 ASSERT(vd->vdev_dtl_sm != NULL); 2760 } 2761 2762 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 2763 2764 mutex_enter(&vd->vdev_dtl_lock); 2765 range_tree_walk(rt, range_tree_add, rtsync); 2766 mutex_exit(&vd->vdev_dtl_lock); 2767 2768 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); 2769 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 2770 range_tree_vacate(rtsync, NULL, NULL); 2771 2772 range_tree_destroy(rtsync); 2773 2774 /* 2775 * If the object for the space map has changed then dirty 2776 * the top level so that we update the config. 2777 */ 2778 if (object != space_map_object(vd->vdev_dtl_sm)) { 2779 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 2780 "new object %llu", (u_longlong_t)txg, spa_name(spa), 2781 (u_longlong_t)object, 2782 (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 2783 vdev_config_dirty(vd->vdev_top); 2784 } 2785 2786 dmu_tx_commit(tx); 2787 } 2788 2789 /* 2790 * Determine whether the specified vdev can be offlined/detached/removed 2791 * without losing data. 2792 */ 2793 boolean_t 2794 vdev_dtl_required(vdev_t *vd) 2795 { 2796 spa_t *spa = vd->vdev_spa; 2797 vdev_t *tvd = vd->vdev_top; 2798 uint8_t cant_read = vd->vdev_cant_read; 2799 boolean_t required; 2800 2801 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2802 2803 if (vd == spa->spa_root_vdev || vd == tvd) 2804 return (B_TRUE); 2805 2806 /* 2807 * Temporarily mark the device as unreadable, and then determine 2808 * whether this results in any DTL outages in the top-level vdev. 2809 * If not, we can safely offline/detach/remove the device. 2810 */ 2811 vd->vdev_cant_read = B_TRUE; 2812 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2813 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 2814 vd->vdev_cant_read = cant_read; 2815 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2816 2817 if (!required && zio_injection_enabled) 2818 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2819 2820 return (required); 2821 } 2822 2823 /* 2824 * Determine if resilver is needed, and if so the txg range. 2825 */ 2826 boolean_t 2827 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2828 { 2829 boolean_t needed = B_FALSE; 2830 uint64_t thismin = UINT64_MAX; 2831 uint64_t thismax = 0; 2832 2833 if (vd->vdev_children == 0) { 2834 mutex_enter(&vd->vdev_dtl_lock); 2835 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2836 vdev_writeable(vd)) { 2837 2838 thismin = vdev_dtl_min(vd); 2839 thismax = vdev_dtl_max(vd); 2840 needed = B_TRUE; 2841 } 2842 mutex_exit(&vd->vdev_dtl_lock); 2843 } else { 2844 for (int c = 0; c < vd->vdev_children; c++) { 2845 vdev_t *cvd = vd->vdev_child[c]; 2846 uint64_t cmin, cmax; 2847 2848 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2849 thismin = MIN(thismin, cmin); 2850 thismax = MAX(thismax, cmax); 2851 needed = B_TRUE; 2852 } 2853 } 2854 } 2855 2856 if (needed && minp) { 2857 *minp = thismin; 2858 *maxp = thismax; 2859 } 2860 return (needed); 2861 } 2862 2863 /* 2864 * Gets the checkpoint space map object from the vdev's ZAP. 2865 * Returns the spacemap object, or 0 if it wasn't in the ZAP 2866 * or the ZAP doesn't exist yet. 2867 */ 2868 int 2869 vdev_checkpoint_sm_object(vdev_t *vd) 2870 { 2871 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 2872 if (vd->vdev_top_zap == 0) { 2873 return (0); 2874 } 2875 2876 uint64_t sm_obj = 0; 2877 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 2878 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 2879 2880 ASSERT(err == 0 || err == ENOENT); 2881 2882 return (sm_obj); 2883 } 2884 2885 int 2886 vdev_load(vdev_t *vd) 2887 { 2888 int error = 0; 2889 /* 2890 * Recursively load all children. 2891 */ 2892 for (int c = 0; c < vd->vdev_children; c++) { 2893 error = vdev_load(vd->vdev_child[c]); 2894 if (error != 0) { 2895 return (error); 2896 } 2897 } 2898 2899 vdev_set_deflate_ratio(vd); 2900 2901 /* 2902 * On spa_load path, grab the allocation bias from our zap 2903 */ 2904 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2905 spa_t *spa = vd->vdev_spa; 2906 char bias_str[64]; 2907 2908 if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2909 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2910 bias_str) == 0) { 2911 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2912 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2913 } 2914 } 2915 2916 /* 2917 * If this is a top-level vdev, initialize its metaslabs. 2918 */ 2919 if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2920 vdev_metaslab_group_create(vd); 2921 2922 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 2923 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2924 VDEV_AUX_CORRUPT_DATA); 2925 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 2926 "asize=%llu", (u_longlong_t)vd->vdev_ashift, 2927 (u_longlong_t)vd->vdev_asize); 2928 return (SET_ERROR(ENXIO)); 2929 } 2930 2931 error = vdev_metaslab_init(vd, 0); 2932 if (error != 0) { 2933 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 2934 "[error=%d]", error); 2935 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2936 VDEV_AUX_CORRUPT_DATA); 2937 return (error); 2938 } 2939 2940 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 2941 if (checkpoint_sm_obj != 0) { 2942 objset_t *mos = spa_meta_objset(vd->vdev_spa); 2943 ASSERT(vd->vdev_asize != 0); 2944 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 2945 2946 error = space_map_open(&vd->vdev_checkpoint_sm, 2947 mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2948 vd->vdev_ashift); 2949 if (error != 0) { 2950 vdev_dbgmsg(vd, "vdev_load: space_map_open " 2951 "failed for checkpoint spacemap (obj %llu) " 2952 "[error=%d]", 2953 (u_longlong_t)checkpoint_sm_obj, error); 2954 return (error); 2955 } 2956 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 2957 2958 /* 2959 * Since the checkpoint_sm contains free entries 2960 * exclusively we can use space_map_allocated() to 2961 * indicate the cumulative checkpointed space that 2962 * has been freed. 2963 */ 2964 vd->vdev_stat.vs_checkpoint_space = 2965 -space_map_allocated(vd->vdev_checkpoint_sm); 2966 vd->vdev_spa->spa_checkpoint_info.sci_dspace += 2967 vd->vdev_stat.vs_checkpoint_space; 2968 } 2969 } 2970 2971 /* 2972 * If this is a leaf vdev, load its DTL. 2973 */ 2974 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2975 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2976 VDEV_AUX_CORRUPT_DATA); 2977 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 2978 "[error=%d]", error); 2979 return (error); 2980 } 2981 2982 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 2983 if (obsolete_sm_object != 0) { 2984 objset_t *mos = vd->vdev_spa->spa_meta_objset; 2985 ASSERT(vd->vdev_asize != 0); 2986 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 2987 2988 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 2989 obsolete_sm_object, 0, vd->vdev_asize, 0))) { 2990 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2991 VDEV_AUX_CORRUPT_DATA); 2992 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 2993 "obsolete spacemap (obj %llu) [error=%d]", 2994 (u_longlong_t)obsolete_sm_object, error); 2995 return (error); 2996 } 2997 } 2998 2999 return (0); 3000 } 3001 3002 /* 3003 * The special vdev case is used for hot spares and l2cache devices. Its 3004 * sole purpose it to set the vdev state for the associated vdev. To do this, 3005 * we make sure that we can open the underlying device, then try to read the 3006 * label, and make sure that the label is sane and that it hasn't been 3007 * repurposed to another pool. 3008 */ 3009 int 3010 vdev_validate_aux(vdev_t *vd) 3011 { 3012 nvlist_t *label; 3013 uint64_t guid, version; 3014 uint64_t state; 3015 3016 if (!vdev_readable(vd)) 3017 return (0); 3018 3019 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 3020 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3021 VDEV_AUX_CORRUPT_DATA); 3022 return (-1); 3023 } 3024 3025 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3026 !SPA_VERSION_IS_SUPPORTED(version) || 3027 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 3028 guid != vd->vdev_guid || 3029 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 3030 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 3031 VDEV_AUX_CORRUPT_DATA); 3032 nvlist_free(label); 3033 return (-1); 3034 } 3035 3036 /* 3037 * We don't actually check the pool state here. If it's in fact in 3038 * use by another pool, we update this fact on the fly when requested. 3039 */ 3040 nvlist_free(label); 3041 return (0); 3042 } 3043 3044 static void 3045 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) 3046 { 3047 objset_t *mos = spa_meta_objset(vd->vdev_spa); 3048 3049 if (vd->vdev_top_zap == 0) 3050 return; 3051 3052 uint64_t object = 0; 3053 int err = zap_lookup(mos, vd->vdev_top_zap, 3054 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); 3055 if (err == ENOENT) 3056 return; 3057 3058 VERIFY0(dmu_object_free(mos, object, tx)); 3059 VERIFY0(zap_remove(mos, vd->vdev_top_zap, 3060 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); 3061 } 3062 3063 /* 3064 * Free the objects used to store this vdev's spacemaps, and the array 3065 * that points to them. 3066 */ 3067 void 3068 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 3069 { 3070 if (vd->vdev_ms_array == 0) 3071 return; 3072 3073 objset_t *mos = vd->vdev_spa->spa_meta_objset; 3074 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 3075 size_t array_bytes = array_count * sizeof (uint64_t); 3076 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 3077 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 3078 array_bytes, smobj_array, 0)); 3079 3080 for (uint64_t i = 0; i < array_count; i++) { 3081 uint64_t smobj = smobj_array[i]; 3082 if (smobj == 0) 3083 continue; 3084 3085 space_map_free_obj(mos, smobj, tx); 3086 } 3087 3088 kmem_free(smobj_array, array_bytes); 3089 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3090 vdev_destroy_ms_flush_data(vd, tx); 3091 vd->vdev_ms_array = 0; 3092 } 3093 3094 static void 3095 vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 3096 { 3097 spa_t *spa = vd->vdev_spa; 3098 3099 ASSERT(vd->vdev_islog); 3100 ASSERT(vd == vd->vdev_top); 3101 ASSERT3U(txg, ==, spa_syncing_txg(spa)); 3102 3103 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3104 3105 vdev_destroy_spacemaps(vd, tx); 3106 if (vd->vdev_top_zap != 0) { 3107 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3108 vd->vdev_top_zap = 0; 3109 } 3110 3111 dmu_tx_commit(tx); 3112 } 3113 3114 void 3115 vdev_sync_done(vdev_t *vd, uint64_t txg) 3116 { 3117 metaslab_t *msp; 3118 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3119 3120 ASSERT(vdev_is_concrete(vd)); 3121 3122 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3123 != NULL) 3124 metaslab_sync_done(msp, txg); 3125 3126 if (reassess) 3127 metaslab_sync_reassess(vd->vdev_mg); 3128 } 3129 3130 void 3131 vdev_sync(vdev_t *vd, uint64_t txg) 3132 { 3133 spa_t *spa = vd->vdev_spa; 3134 vdev_t *lvd; 3135 metaslab_t *msp; 3136 3137 ASSERT3U(txg, ==, spa->spa_syncing_txg); 3138 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3139 if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 3140 ASSERT(vd->vdev_removing || 3141 vd->vdev_ops == &vdev_indirect_ops); 3142 3143 vdev_indirect_sync_obsolete(vd, tx); 3144 3145 /* 3146 * If the vdev is indirect, it can't have dirty 3147 * metaslabs or DTLs. 3148 */ 3149 if (vd->vdev_ops == &vdev_indirect_ops) { 3150 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 3151 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3152 dmu_tx_commit(tx); 3153 return; 3154 } 3155 } 3156 3157 ASSERT(vdev_is_concrete(vd)); 3158 3159 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 3160 !vd->vdev_removing) { 3161 ASSERT(vd == vd->vdev_top); 3162 ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3163 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3164 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3165 ASSERT(vd->vdev_ms_array != 0); 3166 vdev_config_dirty(vd); 3167 } 3168 3169 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3170 metaslab_sync(msp, txg); 3171 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3172 } 3173 3174 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3175 vdev_dtl_sync(lvd, txg); 3176 3177 /* 3178 * If this is an empty log device being removed, destroy the 3179 * metadata associated with it. 3180 */ 3181 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 3182 vdev_remove_empty_log(vd, txg); 3183 3184 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3185 dmu_tx_commit(tx); 3186 } 3187 3188 uint64_t 3189 vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3190 { 3191 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3192 } 3193 3194 /* 3195 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 3196 * not be opened, and no I/O is attempted. 3197 */ 3198 int 3199 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3200 { 3201 vdev_t *vd, *tvd; 3202 3203 spa_vdev_state_enter(spa, SCL_NONE); 3204 3205 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3206 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3207 3208 if (!vd->vdev_ops->vdev_op_leaf) 3209 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3210 3211 tvd = vd->vdev_top; 3212 3213 /* 3214 * We don't directly use the aux state here, but if we do a 3215 * vdev_reopen(), we need this value to be present to remember why we 3216 * were faulted. 3217 */ 3218 vd->vdev_label_aux = aux; 3219 3220 /* 3221 * Faulted state takes precedence over degraded. 3222 */ 3223 vd->vdev_delayed_close = B_FALSE; 3224 vd->vdev_faulted = 1ULL; 3225 vd->vdev_degraded = 0ULL; 3226 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 3227 3228 /* 3229 * If this device has the only valid copy of the data, then 3230 * back off and simply mark the vdev as degraded instead. 3231 */ 3232 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 3233 vd->vdev_degraded = 1ULL; 3234 vd->vdev_faulted = 0ULL; 3235 3236 /* 3237 * If we reopen the device and it's not dead, only then do we 3238 * mark it degraded. 3239 */ 3240 vdev_reopen(tvd); 3241 3242 if (vdev_readable(vd)) 3243 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 3244 } 3245 3246 return (spa_vdev_state_exit(spa, vd, 0)); 3247 } 3248 3249 /* 3250 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 3251 * user that something is wrong. The vdev continues to operate as normal as far 3252 * as I/O is concerned. 3253 */ 3254 int 3255 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3256 { 3257 vdev_t *vd; 3258 3259 spa_vdev_state_enter(spa, SCL_NONE); 3260 3261 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3262 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3263 3264 if (!vd->vdev_ops->vdev_op_leaf) 3265 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3266 3267 /* 3268 * If the vdev is already faulted, then don't do anything. 3269 */ 3270 if (vd->vdev_faulted || vd->vdev_degraded) 3271 return (spa_vdev_state_exit(spa, NULL, 0)); 3272 3273 vd->vdev_degraded = 1ULL; 3274 if (!vdev_is_dead(vd)) 3275 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3276 aux); 3277 3278 return (spa_vdev_state_exit(spa, vd, 0)); 3279 } 3280 3281 /* 3282 * Online the given vdev. 3283 * 3284 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3285 * spare device should be detached when the device finishes resilvering. 3286 * Second, the online should be treated like a 'test' online case, so no FMA 3287 * events are generated if the device fails to open. 3288 */ 3289 int 3290 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 3291 { 3292 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 3293 boolean_t wasoffline; 3294 vdev_state_t oldstate; 3295 3296 spa_vdev_state_enter(spa, SCL_NONE); 3297 3298 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3299 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3300 3301 if (!vd->vdev_ops->vdev_op_leaf) 3302 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3303 3304 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 3305 oldstate = vd->vdev_state; 3306 3307 tvd = vd->vdev_top; 3308 vd->vdev_offline = B_FALSE; 3309 vd->vdev_tmpoffline = B_FALSE; 3310 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3311 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3312 3313 /* XXX - L2ARC 1.0 does not support expansion */ 3314 if (!vd->vdev_aux) { 3315 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3316 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3317 } 3318 3319 vdev_reopen(tvd); 3320 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 3321 3322 if (!vd->vdev_aux) { 3323 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3324 pvd->vdev_expanding = B_FALSE; 3325 } 3326 3327 if (newstate) 3328 *newstate = vd->vdev_state; 3329 if ((flags & ZFS_ONLINE_UNSPARE) && 3330 !vdev_is_dead(vd) && vd->vdev_parent && 3331 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3332 vd->vdev_parent->vdev_child[0] == vd) 3333 vd->vdev_unspare = B_TRUE; 3334 3335 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3336 3337 /* XXX - L2ARC 1.0 does not support expansion */ 3338 if (vd->vdev_aux) 3339 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3340 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3341 } 3342 3343 /* Restart initializing if necessary */ 3344 mutex_enter(&vd->vdev_initialize_lock); 3345 if (vdev_writeable(vd) && 3346 vd->vdev_initialize_thread == NULL && 3347 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3348 (void) vdev_initialize(vd); 3349 } 3350 mutex_exit(&vd->vdev_initialize_lock); 3351 3352 /* Restart trimming if necessary */ 3353 mutex_enter(&vd->vdev_trim_lock); 3354 if (vdev_writeable(vd) && 3355 vd->vdev_trim_thread == NULL && 3356 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 3357 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 3358 vd->vdev_trim_secure); 3359 } 3360 mutex_exit(&vd->vdev_trim_lock); 3361 3362 if (wasoffline || 3363 (oldstate < VDEV_STATE_DEGRADED && 3364 vd->vdev_state >= VDEV_STATE_DEGRADED)) 3365 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 3366 3367 return (spa_vdev_state_exit(spa, vd, 0)); 3368 } 3369 3370 static int 3371 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3372 { 3373 vdev_t *vd, *tvd; 3374 int error = 0; 3375 uint64_t generation; 3376 metaslab_group_t *mg; 3377 3378 top: 3379 spa_vdev_state_enter(spa, SCL_ALLOC); 3380 3381 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3382 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3383 3384 if (!vd->vdev_ops->vdev_op_leaf) 3385 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3386 3387 tvd = vd->vdev_top; 3388 mg = tvd->vdev_mg; 3389 generation = spa->spa_config_generation + 1; 3390 3391 /* 3392 * If the device isn't already offline, try to offline it. 3393 */ 3394 if (!vd->vdev_offline) { 3395 /* 3396 * If this device has the only valid copy of some data, 3397 * don't allow it to be offlined. Log devices are always 3398 * expendable. 3399 */ 3400 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3401 vdev_dtl_required(vd)) 3402 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3403 3404 /* 3405 * If the top-level is a slog and it has had allocations 3406 * then proceed. We check that the vdev's metaslab group 3407 * is not NULL since it's possible that we may have just 3408 * added this vdev but not yet initialized its metaslabs. 3409 */ 3410 if (tvd->vdev_islog && mg != NULL) { 3411 /* 3412 * Prevent any future allocations. 3413 */ 3414 metaslab_group_passivate(mg); 3415 (void) spa_vdev_state_exit(spa, vd, 0); 3416 3417 error = spa_reset_logs(spa); 3418 3419 /* 3420 * If the log device was successfully reset but has 3421 * checkpointed data, do not offline it. 3422 */ 3423 if (error == 0 && 3424 tvd->vdev_checkpoint_sm != NULL) { 3425 error = ZFS_ERR_CHECKPOINT_EXISTS; 3426 } 3427 3428 spa_vdev_state_enter(spa, SCL_ALLOC); 3429 3430 /* 3431 * Check to see if the config has changed. 3432 */ 3433 if (error || generation != spa->spa_config_generation) { 3434 metaslab_group_activate(mg); 3435 if (error) 3436 return (spa_vdev_state_exit(spa, 3437 vd, error)); 3438 (void) spa_vdev_state_exit(spa, vd, 0); 3439 goto top; 3440 } 3441 ASSERT0(tvd->vdev_stat.vs_alloc); 3442 } 3443 3444 /* 3445 * Offline this device and reopen its top-level vdev. 3446 * If the top-level vdev is a log device then just offline 3447 * it. Otherwise, if this action results in the top-level 3448 * vdev becoming unusable, undo it and fail the request. 3449 */ 3450 vd->vdev_offline = B_TRUE; 3451 vdev_reopen(tvd); 3452 3453 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3454 vdev_is_dead(tvd)) { 3455 vd->vdev_offline = B_FALSE; 3456 vdev_reopen(tvd); 3457 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3458 } 3459 3460 /* 3461 * Add the device back into the metaslab rotor so that 3462 * once we online the device it's open for business. 3463 */ 3464 if (tvd->vdev_islog && mg != NULL) 3465 metaslab_group_activate(mg); 3466 } 3467 3468 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3469 3470 return (spa_vdev_state_exit(spa, vd, 0)); 3471 } 3472 3473 int 3474 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3475 { 3476 int error; 3477 3478 mutex_enter(&spa->spa_vdev_top_lock); 3479 error = vdev_offline_locked(spa, guid, flags); 3480 mutex_exit(&spa->spa_vdev_top_lock); 3481 3482 return (error); 3483 } 3484 3485 /* 3486 * Clear the error counts associated with this vdev. Unlike vdev_online() and 3487 * vdev_offline(), we assume the spa config is locked. We also clear all 3488 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3489 */ 3490 void 3491 vdev_clear(spa_t *spa, vdev_t *vd) 3492 { 3493 vdev_t *rvd = spa->spa_root_vdev; 3494 3495 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3496 3497 if (vd == NULL) 3498 vd = rvd; 3499 3500 vd->vdev_stat.vs_read_errors = 0; 3501 vd->vdev_stat.vs_write_errors = 0; 3502 vd->vdev_stat.vs_checksum_errors = 0; 3503 vd->vdev_stat.vs_slow_ios = 0; 3504 3505 for (int c = 0; c < vd->vdev_children; c++) 3506 vdev_clear(spa, vd->vdev_child[c]); 3507 3508 /* 3509 * It makes no sense to "clear" an indirect vdev. 3510 */ 3511 if (!vdev_is_concrete(vd)) 3512 return; 3513 3514 /* 3515 * If we're in the FAULTED state or have experienced failed I/O, then 3516 * clear the persistent state and attempt to reopen the device. We 3517 * also mark the vdev config dirty, so that the new faulted state is 3518 * written out to disk. 3519 */ 3520 if (vd->vdev_faulted || vd->vdev_degraded || 3521 !vdev_readable(vd) || !vdev_writeable(vd)) { 3522 3523 /* 3524 * When reopening in reponse to a clear event, it may be due to 3525 * a fmadm repair request. In this case, if the device is 3526 * still broken, we want to still post the ereport again. 3527 */ 3528 vd->vdev_forcefault = B_TRUE; 3529 3530 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3531 vd->vdev_cant_read = B_FALSE; 3532 vd->vdev_cant_write = B_FALSE; 3533 3534 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 3535 3536 vd->vdev_forcefault = B_FALSE; 3537 3538 if (vd != rvd && vdev_writeable(vd->vdev_top)) 3539 vdev_state_dirty(vd->vdev_top); 3540 3541 /* If a resilver isn't required, check if vdevs can be culled */ 3542 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && 3543 !dsl_scan_resilvering(spa->spa_dsl_pool) && 3544 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) 3545 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 3546 3547 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 3548 } 3549 3550 /* 3551 * When clearing a FMA-diagnosed fault, we always want to 3552 * unspare the device, as we assume that the original spare was 3553 * done in response to the FMA fault. 3554 */ 3555 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3556 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3557 vd->vdev_parent->vdev_child[0] == vd) 3558 vd->vdev_unspare = B_TRUE; 3559 } 3560 3561 boolean_t 3562 vdev_is_dead(vdev_t *vd) 3563 { 3564 /* 3565 * Holes and missing devices are always considered "dead". 3566 * This simplifies the code since we don't have to check for 3567 * these types of devices in the various code paths. 3568 * Instead we rely on the fact that we skip over dead devices 3569 * before issuing I/O to them. 3570 */ 3571 return (vd->vdev_state < VDEV_STATE_DEGRADED || 3572 vd->vdev_ops == &vdev_hole_ops || 3573 vd->vdev_ops == &vdev_missing_ops); 3574 } 3575 3576 boolean_t 3577 vdev_readable(vdev_t *vd) 3578 { 3579 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 3580 } 3581 3582 boolean_t 3583 vdev_writeable(vdev_t *vd) 3584 { 3585 return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 3586 vdev_is_concrete(vd)); 3587 } 3588 3589 boolean_t 3590 vdev_allocatable(vdev_t *vd) 3591 { 3592 uint64_t state = vd->vdev_state; 3593 3594 /* 3595 * We currently allow allocations from vdevs which may be in the 3596 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3597 * fails to reopen then we'll catch it later when we're holding 3598 * the proper locks. Note that we have to get the vdev state 3599 * in a local variable because although it changes atomically, 3600 * we're asking two separate questions about it. 3601 */ 3602 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 3603 !vd->vdev_cant_write && vdev_is_concrete(vd) && 3604 vd->vdev_mg->mg_initialized); 3605 } 3606 3607 boolean_t 3608 vdev_accessible(vdev_t *vd, zio_t *zio) 3609 { 3610 ASSERT(zio->io_vd == vd); 3611 3612 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3613 return (B_FALSE); 3614 3615 if (zio->io_type == ZIO_TYPE_READ) 3616 return (!vd->vdev_cant_read); 3617 3618 if (zio->io_type == ZIO_TYPE_WRITE) 3619 return (!vd->vdev_cant_write); 3620 3621 return (B_TRUE); 3622 } 3623 3624 static void 3625 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) 3626 { 3627 for (int t = 0; t < VS_ZIO_TYPES; t++) { 3628 vs->vs_ops[t] += cvs->vs_ops[t]; 3629 vs->vs_bytes[t] += cvs->vs_bytes[t]; 3630 } 3631 3632 cvs->vs_scan_removing = cvd->vdev_removing; 3633 } 3634 3635 /* 3636 * Get extended stats 3637 */ 3638 static void 3639 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) 3640 { 3641 int t, b; 3642 for (t = 0; t < ZIO_TYPES; t++) { 3643 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) 3644 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; 3645 3646 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { 3647 vsx->vsx_total_histo[t][b] += 3648 cvsx->vsx_total_histo[t][b]; 3649 } 3650 } 3651 3652 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { 3653 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { 3654 vsx->vsx_queue_histo[t][b] += 3655 cvsx->vsx_queue_histo[t][b]; 3656 } 3657 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; 3658 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; 3659 3660 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) 3661 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; 3662 3663 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) 3664 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; 3665 } 3666 3667 } 3668 3669 boolean_t 3670 vdev_is_spacemap_addressable(vdev_t *vd) 3671 { 3672 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3673 return (B_TRUE); 3674 3675 /* 3676 * If double-word space map entries are not enabled we assume 3677 * 47 bits of the space map entry are dedicated to the entry's 3678 * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3679 * to calculate the maximum address that can be described by a 3680 * space map entry for the given device. 3681 */ 3682 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 3683 3684 if (shift >= 63) /* detect potential overflow */ 3685 return (B_TRUE); 3686 3687 return (vd->vdev_asize < (1ULL << shift)); 3688 } 3689 3690 /* 3691 * Get statistics for the given vdev. 3692 */ 3693 static void 3694 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3695 { 3696 int t; 3697 /* 3698 * If we're getting stats on the root vdev, aggregate the I/O counts 3699 * over all top-level vdevs (i.e. the direct children of the root). 3700 */ 3701 if (!vd->vdev_ops->vdev_op_leaf) { 3702 if (vs) { 3703 memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); 3704 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); 3705 } 3706 if (vsx) 3707 memset(vsx, 0, sizeof (*vsx)); 3708 3709 for (int c = 0; c < vd->vdev_children; c++) { 3710 vdev_t *cvd = vd->vdev_child[c]; 3711 vdev_stat_t *cvs = &cvd->vdev_stat; 3712 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; 3713 3714 vdev_get_stats_ex_impl(cvd, cvs, cvsx); 3715 if (vs) 3716 vdev_get_child_stat(cvd, vs, cvs); 3717 if (vsx) 3718 vdev_get_child_stat_ex(cvd, vsx, cvsx); 3719 3720 } 3721 } else { 3722 /* 3723 * We're a leaf. Just copy our ZIO active queue stats in. The 3724 * other leaf stats are updated in vdev_stat_update(). 3725 */ 3726 if (!vsx) 3727 return; 3728 3729 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); 3730 3731 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { 3732 vsx->vsx_active_queue[t] = 3733 vd->vdev_queue.vq_class[t].vqc_active; 3734 vsx->vsx_pend_queue[t] = avl_numnodes( 3735 &vd->vdev_queue.vq_class[t].vqc_queued_tree); 3736 } 3737 } 3738 } 3739 3740 void 3741 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3742 { 3743 vdev_t *tvd = vd->vdev_top; 3744 mutex_enter(&vd->vdev_stat_lock); 3745 if (vs) { 3746 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3747 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3748 vs->vs_state = vd->vdev_state; 3749 vs->vs_rsize = vdev_get_min_asize(vd); 3750 if (vd->vdev_ops->vdev_op_leaf) { 3751 vs->vs_rsize += VDEV_LABEL_START_SIZE + 3752 VDEV_LABEL_END_SIZE; 3753 /* 3754 * Report initializing progress. Since we don't 3755 * have the initializing locks held, this is only 3756 * an estimate (although a fairly accurate one). 3757 */ 3758 vs->vs_initialize_bytes_done = 3759 vd->vdev_initialize_bytes_done; 3760 vs->vs_initialize_bytes_est = 3761 vd->vdev_initialize_bytes_est; 3762 vs->vs_initialize_state = vd->vdev_initialize_state; 3763 vs->vs_initialize_action_time = 3764 vd->vdev_initialize_action_time; 3765 3766 /* 3767 * Report manual TRIM progress. Since we don't have 3768 * the manual TRIM locks held, this is only an 3769 * estimate (although fairly accurate one). 3770 */ 3771 vs->vs_trim_notsup = !vd->vdev_has_trim; 3772 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 3773 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 3774 vs->vs_trim_state = vd->vdev_trim_state; 3775 vs->vs_trim_action_time = vd->vdev_trim_action_time; 3776 } 3777 /* 3778 * Report expandable space on top-level, non-auxiliary devices 3779 * only. The expandable space is reported in terms of metaslab 3780 * sized units since that determines how much space the pool 3781 * can expand. 3782 */ 3783 if (vd->vdev_aux == NULL && tvd != NULL) { 3784 vs->vs_esize = P2ALIGN( 3785 vd->vdev_max_asize - vd->vdev_asize, 3786 1ULL << tvd->vdev_ms_shift); 3787 } 3788 if (vd->vdev_aux == NULL && vd == vd->vdev_top && 3789 vdev_is_concrete(vd)) { 3790 vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3791 vd->vdev_mg->mg_fragmentation : 0; 3792 } 3793 if (vd->vdev_ops->vdev_op_leaf) 3794 vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 3795 } 3796 3797 vdev_get_stats_ex_impl(vd, vs, vsx); 3798 mutex_exit(&vd->vdev_stat_lock); 3799 } 3800 3801 void 3802 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3803 { 3804 return (vdev_get_stats_ex(vd, vs, NULL)); 3805 } 3806 3807 void 3808 vdev_clear_stats(vdev_t *vd) 3809 { 3810 mutex_enter(&vd->vdev_stat_lock); 3811 vd->vdev_stat.vs_space = 0; 3812 vd->vdev_stat.vs_dspace = 0; 3813 vd->vdev_stat.vs_alloc = 0; 3814 mutex_exit(&vd->vdev_stat_lock); 3815 } 3816 3817 void 3818 vdev_scan_stat_init(vdev_t *vd) 3819 { 3820 vdev_stat_t *vs = &vd->vdev_stat; 3821 3822 for (int c = 0; c < vd->vdev_children; c++) 3823 vdev_scan_stat_init(vd->vdev_child[c]); 3824 3825 mutex_enter(&vd->vdev_stat_lock); 3826 vs->vs_scan_processed = 0; 3827 mutex_exit(&vd->vdev_stat_lock); 3828 } 3829 3830 void 3831 vdev_stat_update(zio_t *zio, uint64_t psize) 3832 { 3833 spa_t *spa = zio->io_spa; 3834 vdev_t *rvd = spa->spa_root_vdev; 3835 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3836 vdev_t *pvd; 3837 uint64_t txg = zio->io_txg; 3838 vdev_stat_t *vs = &vd->vdev_stat; 3839 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; 3840 zio_type_t type = zio->io_type; 3841 int flags = zio->io_flags; 3842 3843 /* 3844 * If this i/o is a gang leader, it didn't do any actual work. 3845 */ 3846 if (zio->io_gang_tree) 3847 return; 3848 3849 if (zio->io_error == 0) { 3850 /* 3851 * If this is a root i/o, don't count it -- we've already 3852 * counted the top-level vdevs, and vdev_get_stats() will 3853 * aggregate them when asked. This reduces contention on 3854 * the root vdev_stat_lock and implicitly handles blocks 3855 * that compress away to holes, for which there is no i/o. 3856 * (Holes never create vdev children, so all the counters 3857 * remain zero, which is what we want.) 3858 * 3859 * Note: this only applies to successful i/o (io_error == 0) 3860 * because unlike i/o counts, errors are not additive. 3861 * When reading a ditto block, for example, failure of 3862 * one top-level vdev does not imply a root-level error. 3863 */ 3864 if (vd == rvd) 3865 return; 3866 3867 ASSERT(vd == zio->io_vd); 3868 3869 if (flags & ZIO_FLAG_IO_BYPASS) 3870 return; 3871 3872 mutex_enter(&vd->vdev_stat_lock); 3873 3874 if (flags & ZIO_FLAG_IO_REPAIR) { 3875 if (flags & ZIO_FLAG_SCAN_THREAD) { 3876 dsl_scan_phys_t *scn_phys = 3877 &spa->spa_dsl_pool->dp_scan->scn_phys; 3878 uint64_t *processed = &scn_phys->scn_processed; 3879 3880 /* XXX cleanup? */ 3881 if (vd->vdev_ops->vdev_op_leaf) 3882 atomic_add_64(processed, psize); 3883 vs->vs_scan_processed += psize; 3884 } 3885 3886 if (flags & ZIO_FLAG_SELF_HEAL) 3887 vs->vs_self_healed += psize; 3888 } 3889 3890 /* 3891 * The bytes/ops/histograms are recorded at the leaf level and 3892 * aggregated into the higher level vdevs in vdev_get_stats(). 3893 */ 3894 if (vd->vdev_ops->vdev_op_leaf && 3895 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { 3896 zio_type_t vs_type = type; 3897 3898 /* 3899 * TRIM ops and bytes are reported to user space as 3900 * ZIO_TYPE_IOCTL. This is done to preserve the 3901 * vdev_stat_t structure layout for user space. 3902 */ 3903 if (type == ZIO_TYPE_TRIM) 3904 vs_type = ZIO_TYPE_IOCTL; 3905 3906 vs->vs_ops[vs_type]++; 3907 vs->vs_bytes[vs_type] += psize; 3908 3909 if (flags & ZIO_FLAG_DELEGATED) { 3910 vsx->vsx_agg_histo[zio->io_priority] 3911 [RQ_HISTO(zio->io_size)]++; 3912 } else { 3913 vsx->vsx_ind_histo[zio->io_priority] 3914 [RQ_HISTO(zio->io_size)]++; 3915 } 3916 3917 if (zio->io_delta && zio->io_delay) { 3918 vsx->vsx_queue_histo[zio->io_priority] 3919 [L_HISTO(zio->io_delta - zio->io_delay)]++; 3920 vsx->vsx_disk_histo[type] 3921 [L_HISTO(zio->io_delay)]++; 3922 vsx->vsx_total_histo[type] 3923 [L_HISTO(zio->io_delta)]++; 3924 } 3925 } 3926 3927 mutex_exit(&vd->vdev_stat_lock); 3928 return; 3929 } 3930 3931 if (flags & ZIO_FLAG_SPECULATIVE) 3932 return; 3933 3934 /* 3935 * If this is an I/O error that is going to be retried, then ignore the 3936 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 3937 * hard errors, when in reality they can happen for any number of 3938 * innocuous reasons (bus resets, MPxIO link failure, etc). 3939 */ 3940 if (zio->io_error == EIO && 3941 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 3942 return; 3943 3944 /* 3945 * Intent logs writes won't propagate their error to the root 3946 * I/O so don't mark these types of failures as pool-level 3947 * errors. 3948 */ 3949 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 3950 return; 3951 3952 mutex_enter(&vd->vdev_stat_lock); 3953 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3954 if (zio->io_error == ECKSUM) 3955 vs->vs_checksum_errors++; 3956 else 3957 vs->vs_read_errors++; 3958 } 3959 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3960 vs->vs_write_errors++; 3961 mutex_exit(&vd->vdev_stat_lock); 3962 3963 if (spa->spa_load_state == SPA_LOAD_NONE && 3964 type == ZIO_TYPE_WRITE && txg != 0 && 3965 (!(flags & ZIO_FLAG_IO_REPAIR) || 3966 (flags & ZIO_FLAG_SCAN_THREAD) || 3967 spa->spa_claiming)) { 3968 /* 3969 * This is either a normal write (not a repair), or it's 3970 * a repair induced by the scrub thread, or it's a repair 3971 * made by zil_claim() during spa_load() in the first txg. 3972 * In the normal case, we commit the DTL change in the same 3973 * txg as the block was born. In the scrub-induced repair 3974 * case, we know that scrubs run in first-pass syncing context, 3975 * so we commit the DTL change in spa_syncing_txg(spa). 3976 * In the zil_claim() case, we commit in spa_first_txg(spa). 3977 * 3978 * We currently do not make DTL entries for failed spontaneous 3979 * self-healing writes triggered by normal (non-scrubbing) 3980 * reads, because we have no transactional context in which to 3981 * do so -- and it's not clear that it'd be desirable anyway. 3982 */ 3983 if (vd->vdev_ops->vdev_op_leaf) { 3984 uint64_t commit_txg = txg; 3985 if (flags & ZIO_FLAG_SCAN_THREAD) { 3986 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3987 ASSERT(spa_sync_pass(spa) == 1); 3988 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3989 commit_txg = spa_syncing_txg(spa); 3990 } else if (spa->spa_claiming) { 3991 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3992 commit_txg = spa_first_txg(spa); 3993 } 3994 ASSERT(commit_txg >= spa_syncing_txg(spa)); 3995 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3996 return; 3997 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3998 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 3999 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 4000 } 4001 if (vd != rvd) 4002 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 4003 } 4004 } 4005 4006 int64_t 4007 vdev_deflated_space(vdev_t *vd, int64_t space) 4008 { 4009 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 4010 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 4011 4012 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 4013 } 4014 4015 /* 4016 * Update the in-core space usage stats for this vdev, its metaslab class, 4017 * and the root vdev. 4018 */ 4019 void 4020 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 4021 int64_t space_delta) 4022 { 4023 int64_t dspace_delta; 4024 spa_t *spa = vd->vdev_spa; 4025 vdev_t *rvd = spa->spa_root_vdev; 4026 4027 ASSERT(vd == vd->vdev_top); 4028 4029 /* 4030 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 4031 * factor. We must calculate this here and not at the root vdev 4032 * because the root vdev's psize-to-asize is simply the max of its 4033 * childrens', thus not accurate enough for us. 4034 */ 4035 dspace_delta = vdev_deflated_space(vd, space_delta); 4036 4037 mutex_enter(&vd->vdev_stat_lock); 4038 /* ensure we won't underflow */ 4039 if (alloc_delta < 0) { 4040 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 4041 } 4042 4043 vd->vdev_stat.vs_alloc += alloc_delta; 4044 vd->vdev_stat.vs_space += space_delta; 4045 vd->vdev_stat.vs_dspace += dspace_delta; 4046 mutex_exit(&vd->vdev_stat_lock); 4047 4048 /* every class but log contributes to root space stats */ 4049 if (vd->vdev_mg != NULL && !vd->vdev_islog) { 4050 ASSERT(!vd->vdev_isl2cache); 4051 mutex_enter(&rvd->vdev_stat_lock); 4052 rvd->vdev_stat.vs_alloc += alloc_delta; 4053 rvd->vdev_stat.vs_space += space_delta; 4054 rvd->vdev_stat.vs_dspace += dspace_delta; 4055 mutex_exit(&rvd->vdev_stat_lock); 4056 } 4057 /* Note: metaslab_class_space_update moved to metaslab_space_update */ 4058 } 4059 4060 /* 4061 * Mark a top-level vdev's config as dirty, placing it on the dirty list 4062 * so that it will be written out next time the vdev configuration is synced. 4063 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 4064 */ 4065 void 4066 vdev_config_dirty(vdev_t *vd) 4067 { 4068 spa_t *spa = vd->vdev_spa; 4069 vdev_t *rvd = spa->spa_root_vdev; 4070 int c; 4071 4072 ASSERT(spa_writeable(spa)); 4073 4074 /* 4075 * If this is an aux vdev (as with l2cache and spare devices), then we 4076 * update the vdev config manually and set the sync flag. 4077 */ 4078 if (vd->vdev_aux != NULL) { 4079 spa_aux_vdev_t *sav = vd->vdev_aux; 4080 nvlist_t **aux; 4081 uint_t naux; 4082 4083 for (c = 0; c < sav->sav_count; c++) { 4084 if (sav->sav_vdevs[c] == vd) 4085 break; 4086 } 4087 4088 if (c == sav->sav_count) { 4089 /* 4090 * We're being removed. There's nothing more to do. 4091 */ 4092 ASSERT(sav->sav_sync == B_TRUE); 4093 return; 4094 } 4095 4096 sav->sav_sync = B_TRUE; 4097 4098 if (nvlist_lookup_nvlist_array(sav->sav_config, 4099 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 4100 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 4101 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 4102 } 4103 4104 ASSERT(c < naux); 4105 4106 /* 4107 * Setting the nvlist in the middle if the array is a little 4108 * sketchy, but it will work. 4109 */ 4110 nvlist_free(aux[c]); 4111 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 4112 4113 return; 4114 } 4115 4116 /* 4117 * The dirty list is protected by the SCL_CONFIG lock. The caller 4118 * must either hold SCL_CONFIG as writer, or must be the sync thread 4119 * (which holds SCL_CONFIG as reader). There's only one sync thread, 4120 * so this is sufficient to ensure mutual exclusion. 4121 */ 4122 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4123 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4124 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4125 4126 if (vd == rvd) { 4127 for (c = 0; c < rvd->vdev_children; c++) 4128 vdev_config_dirty(rvd->vdev_child[c]); 4129 } else { 4130 ASSERT(vd == vd->vdev_top); 4131 4132 if (!list_link_active(&vd->vdev_config_dirty_node) && 4133 vdev_is_concrete(vd)) { 4134 list_insert_head(&spa->spa_config_dirty_list, vd); 4135 } 4136 } 4137 } 4138 4139 void 4140 vdev_config_clean(vdev_t *vd) 4141 { 4142 spa_t *spa = vd->vdev_spa; 4143 4144 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4145 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4146 spa_config_held(spa, SCL_CONFIG, RW_READER))); 4147 4148 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4149 list_remove(&spa->spa_config_dirty_list, vd); 4150 } 4151 4152 /* 4153 * Mark a top-level vdev's state as dirty, so that the next pass of 4154 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4155 * the state changes from larger config changes because they require 4156 * much less locking, and are often needed for administrative actions. 4157 */ 4158 void 4159 vdev_state_dirty(vdev_t *vd) 4160 { 4161 spa_t *spa = vd->vdev_spa; 4162 4163 ASSERT(spa_writeable(spa)); 4164 ASSERT(vd == vd->vdev_top); 4165 4166 /* 4167 * The state list is protected by the SCL_STATE lock. The caller 4168 * must either hold SCL_STATE as writer, or must be the sync thread 4169 * (which holds SCL_STATE as reader). There's only one sync thread, 4170 * so this is sufficient to ensure mutual exclusion. 4171 */ 4172 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4173 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4174 spa_config_held(spa, SCL_STATE, RW_READER))); 4175 4176 if (!list_link_active(&vd->vdev_state_dirty_node) && 4177 vdev_is_concrete(vd)) 4178 list_insert_head(&spa->spa_state_dirty_list, vd); 4179 } 4180 4181 void 4182 vdev_state_clean(vdev_t *vd) 4183 { 4184 spa_t *spa = vd->vdev_spa; 4185 4186 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4187 (dsl_pool_sync_context(spa_get_dsl(spa)) && 4188 spa_config_held(spa, SCL_STATE, RW_READER))); 4189 4190 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 4191 list_remove(&spa->spa_state_dirty_list, vd); 4192 } 4193 4194 /* 4195 * Propagate vdev state up from children to parent. 4196 */ 4197 void 4198 vdev_propagate_state(vdev_t *vd) 4199 { 4200 spa_t *spa = vd->vdev_spa; 4201 vdev_t *rvd = spa->spa_root_vdev; 4202 int degraded = 0, faulted = 0; 4203 int corrupted = 0; 4204 vdev_t *child; 4205 4206 if (vd->vdev_children > 0) { 4207 for (int c = 0; c < vd->vdev_children; c++) { 4208 child = vd->vdev_child[c]; 4209 4210 /* 4211 * Don't factor holes or indirect vdevs into the 4212 * decision. 4213 */ 4214 if (!vdev_is_concrete(child)) 4215 continue; 4216 4217 if (!vdev_readable(child) || 4218 (!vdev_writeable(child) && spa_writeable(spa))) { 4219 /* 4220 * Root special: if there is a top-level log 4221 * device, treat the root vdev as if it were 4222 * degraded. 4223 */ 4224 if (child->vdev_islog && vd == rvd) 4225 degraded++; 4226 else 4227 faulted++; 4228 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 4229 degraded++; 4230 } 4231 4232 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 4233 corrupted++; 4234 } 4235 4236 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 4237 4238 /* 4239 * Root special: if there is a top-level vdev that cannot be 4240 * opened due to corrupted metadata, then propagate the root 4241 * vdev's aux state as 'corrupt' rather than 'insufficient 4242 * replicas'. 4243 */ 4244 if (corrupted && vd == rvd && 4245 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 4246 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 4247 VDEV_AUX_CORRUPT_DATA); 4248 } 4249 4250 if (vd->vdev_parent) 4251 vdev_propagate_state(vd->vdev_parent); 4252 } 4253 4254 /* 4255 * Set a vdev's state. If this is during an open, we don't update the parent 4256 * state, because we're in the process of opening children depth-first. 4257 * Otherwise, we propagate the change to the parent. 4258 * 4259 * If this routine places a device in a faulted state, an appropriate ereport is 4260 * generated. 4261 */ 4262 void 4263 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4264 { 4265 uint64_t save_state; 4266 spa_t *spa = vd->vdev_spa; 4267 4268 if (state == vd->vdev_state) { 4269 vd->vdev_stat.vs_aux = aux; 4270 return; 4271 } 4272 4273 save_state = vd->vdev_state; 4274 4275 vd->vdev_state = state; 4276 vd->vdev_stat.vs_aux = aux; 4277 4278 /* 4279 * If we are setting the vdev state to anything but an open state, then 4280 * always close the underlying device unless the device has requested 4281 * a delayed close (i.e. we're about to remove or fault the device). 4282 * Otherwise, we keep accessible but invalid devices open forever. 4283 * We don't call vdev_close() itself, because that implies some extra 4284 * checks (offline, etc) that we don't want here. This is limited to 4285 * leaf devices, because otherwise closing the device will affect other 4286 * children. 4287 */ 4288 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 4289 vd->vdev_ops->vdev_op_leaf) 4290 vd->vdev_ops->vdev_op_close(vd); 4291 4292 /* 4293 * If we have brought this vdev back into service, we need 4294 * to notify fmd so that it can gracefully repair any outstanding 4295 * cases due to a missing device. We do this in all cases, even those 4296 * that probably don't correlate to a repaired fault. This is sure to 4297 * catch all cases, and we let the zfs-retire agent sort it out. If 4298 * this is a transient state it's OK, as the retire agent will 4299 * double-check the state of the vdev before repairing it. 4300 */ 4301 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4302 vd->vdev_prevstate != state) 4303 zfs_post_state_change(spa, vd); 4304 4305 if (vd->vdev_removed && 4306 state == VDEV_STATE_CANT_OPEN && 4307 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 4308 /* 4309 * If the previous state is set to VDEV_STATE_REMOVED, then this 4310 * device was previously marked removed and someone attempted to 4311 * reopen it. If this failed due to a nonexistent device, then 4312 * keep the device in the REMOVED state. We also let this be if 4313 * it is one of our special test online cases, which is only 4314 * attempting to online the device and shouldn't generate an FMA 4315 * fault. 4316 */ 4317 vd->vdev_state = VDEV_STATE_REMOVED; 4318 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 4319 } else if (state == VDEV_STATE_REMOVED) { 4320 vd->vdev_removed = B_TRUE; 4321 } else if (state == VDEV_STATE_CANT_OPEN) { 4322 /* 4323 * If we fail to open a vdev during an import or recovery, we 4324 * mark it as "not available", which signifies that it was 4325 * never there to begin with. Failure to open such a device 4326 * is not considered an error. 4327 */ 4328 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4329 spa_load_state(spa) == SPA_LOAD_RECOVER) && 4330 vd->vdev_ops->vdev_op_leaf) 4331 vd->vdev_not_present = 1; 4332 4333 /* 4334 * Post the appropriate ereport. If the 'prevstate' field is 4335 * set to something other than VDEV_STATE_UNKNOWN, it indicates 4336 * that this is part of a vdev_reopen(). In this case, we don't 4337 * want to post the ereport if the device was already in the 4338 * CANT_OPEN state beforehand. 4339 * 4340 * If the 'checkremove' flag is set, then this is an attempt to 4341 * online the device in response to an insertion event. If we 4342 * hit this case, then we have detected an insertion event for a 4343 * faulted or offline device that wasn't in the removed state. 4344 * In this scenario, we don't post an ereport because we are 4345 * about to replace the device, or attempt an online with 4346 * vdev_forcefault, which will generate the fault for us. 4347 */ 4348 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 4349 !vd->vdev_not_present && !vd->vdev_checkremove && 4350 vd != spa->spa_root_vdev) { 4351 const char *class; 4352 4353 switch (aux) { 4354 case VDEV_AUX_OPEN_FAILED: 4355 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4356 break; 4357 case VDEV_AUX_CORRUPT_DATA: 4358 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4359 break; 4360 case VDEV_AUX_NO_REPLICAS: 4361 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4362 break; 4363 case VDEV_AUX_BAD_GUID_SUM: 4364 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4365 break; 4366 case VDEV_AUX_TOO_SMALL: 4367 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4368 break; 4369 case VDEV_AUX_BAD_LABEL: 4370 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4371 break; 4372 case VDEV_AUX_BAD_ASHIFT: 4373 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 4374 break; 4375 default: 4376 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4377 } 4378 4379 zfs_ereport_post(class, spa, vd, NULL, NULL, 4380 save_state, 0); 4381 } 4382 4383 /* Erase any notion of persistent removed state */ 4384 vd->vdev_removed = B_FALSE; 4385 } else { 4386 vd->vdev_removed = B_FALSE; 4387 } 4388 4389 if (!isopen && vd->vdev_parent) 4390 vdev_propagate_state(vd->vdev_parent); 4391 } 4392 4393 boolean_t 4394 vdev_children_are_offline(vdev_t *vd) 4395 { 4396 ASSERT(!vd->vdev_ops->vdev_op_leaf); 4397 4398 for (uint64_t i = 0; i < vd->vdev_children; i++) { 4399 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 4400 return (B_FALSE); 4401 } 4402 4403 return (B_TRUE); 4404 } 4405 4406 /* 4407 * Check the vdev configuration to ensure that it's capable of supporting 4408 * a root pool. We do not support partial configuration. 4409 * In addition, only a single top-level vdev is allowed. 4410 */ 4411 boolean_t 4412 vdev_is_bootable(vdev_t *vd) 4413 { 4414 if (!vd->vdev_ops->vdev_op_leaf) { 4415 char *vdev_type = vd->vdev_ops->vdev_op_type; 4416 4417 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 4418 vd->vdev_children > 1) { 4419 int non_indirect = 0; 4420 4421 for (int c = 0; c < vd->vdev_children; c++) { 4422 vdev_type = 4423 vd->vdev_child[c]->vdev_ops->vdev_op_type; 4424 if (strcmp(vdev_type, VDEV_TYPE_INDIRECT) != 0) 4425 non_indirect++; 4426 } 4427 /* 4428 * non_indirect > 1 means we have more than one 4429 * top-level vdev, so we stop here. 4430 */ 4431 if (non_indirect > 1) 4432 return (B_FALSE); 4433 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 4434 return (B_FALSE); 4435 } 4436 } 4437 4438 for (int c = 0; c < vd->vdev_children; c++) { 4439 if (!vdev_is_bootable(vd->vdev_child[c])) 4440 return (B_FALSE); 4441 } 4442 return (B_TRUE); 4443 } 4444 4445 boolean_t 4446 vdev_is_concrete(vdev_t *vd) 4447 { 4448 vdev_ops_t *ops = vd->vdev_ops; 4449 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 4450 ops == &vdev_missing_ops || ops == &vdev_root_ops) { 4451 return (B_FALSE); 4452 } else { 4453 return (B_TRUE); 4454 } 4455 } 4456 4457 /* 4458 * Determine if a log device has valid content. If the vdev was 4459 * removed or faulted in the MOS config then we know that 4460 * the content on the log device has already been written to the pool. 4461 */ 4462 boolean_t 4463 vdev_log_state_valid(vdev_t *vd) 4464 { 4465 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 4466 !vd->vdev_removed) 4467 return (B_TRUE); 4468 4469 for (int c = 0; c < vd->vdev_children; c++) 4470 if (vdev_log_state_valid(vd->vdev_child[c])) 4471 return (B_TRUE); 4472 4473 return (B_FALSE); 4474 } 4475 4476 /* 4477 * Expand a vdev if possible. 4478 */ 4479 void 4480 vdev_expand(vdev_t *vd, uint64_t txg) 4481 { 4482 ASSERT(vd->vdev_top == vd); 4483 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4484 ASSERT(vdev_is_concrete(vd)); 4485 4486 vdev_set_deflate_ratio(vd); 4487 4488 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4489 vdev_is_concrete(vd)) { 4490 vdev_metaslab_group_create(vd); 4491 VERIFY(vdev_metaslab_init(vd, txg) == 0); 4492 vdev_config_dirty(vd); 4493 } 4494 } 4495 4496 /* 4497 * Split a vdev. 4498 */ 4499 void 4500 vdev_split(vdev_t *vd) 4501 { 4502 vdev_t *cvd, *pvd = vd->vdev_parent; 4503 4504 vdev_remove_child(pvd, vd); 4505 vdev_compact_children(pvd); 4506 4507 cvd = pvd->vdev_child[0]; 4508 if (pvd->vdev_children == 1) { 4509 vdev_remove_parent(cvd); 4510 cvd->vdev_splitting = B_TRUE; 4511 } 4512 vdev_propagate_state(cvd); 4513 } 4514 4515 void 4516 vdev_deadman(vdev_t *vd) 4517 { 4518 for (int c = 0; c < vd->vdev_children; c++) { 4519 vdev_t *cvd = vd->vdev_child[c]; 4520 4521 vdev_deadman(cvd); 4522 } 4523 4524 if (vd->vdev_ops->vdev_op_leaf) { 4525 vdev_queue_t *vq = &vd->vdev_queue; 4526 4527 mutex_enter(&vq->vq_lock); 4528 if (avl_numnodes(&vq->vq_active_tree) > 0) { 4529 spa_t *spa = vd->vdev_spa; 4530 zio_t *fio; 4531 uint64_t delta; 4532 4533 /* 4534 * Look at the head of all the pending queues, 4535 * if any I/O has been outstanding for longer than 4536 * the spa_deadman_synctime we panic the system. 4537 */ 4538 fio = avl_first(&vq->vq_active_tree); 4539 delta = gethrtime() - fio->io_timestamp; 4540 if (delta > spa_deadman_synctime(spa)) { 4541 vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 4542 "%lluns, delta %lluns, last io %lluns", 4543 fio->io_timestamp, (u_longlong_t)delta, 4544 vq->vq_io_complete_ts); 4545 fm_panic("I/O to pool '%s' appears to be " 4546 "hung.", spa_name(spa)); 4547 } 4548 } 4549 mutex_exit(&vq->vq_lock); 4550 } 4551 } 4552 4553 void 4554 vdev_defer_resilver(vdev_t *vd) 4555 { 4556 ASSERT(vd->vdev_ops->vdev_op_leaf); 4557 4558 vd->vdev_resilver_deferred = B_TRUE; 4559 vd->vdev_spa->spa_resilver_deferred = B_TRUE; 4560 } 4561 4562 /* 4563 * Clears the resilver deferred flag on all leaf devs under vd. Returns 4564 * B_TRUE if we have devices that need to be resilvered and are available to 4565 * accept resilver I/Os. 4566 */ 4567 boolean_t 4568 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) 4569 { 4570 boolean_t resilver_needed = B_FALSE; 4571 spa_t *spa = vd->vdev_spa; 4572 4573 for (int c = 0; c < vd->vdev_children; c++) { 4574 vdev_t *cvd = vd->vdev_child[c]; 4575 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); 4576 } 4577 4578 if (vd == spa->spa_root_vdev && 4579 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { 4580 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); 4581 vdev_config_dirty(vd); 4582 spa->spa_resilver_deferred = B_FALSE; 4583 return (resilver_needed); 4584 } 4585 4586 if (!vdev_is_concrete(vd) || vd->vdev_aux || 4587 !vd->vdev_ops->vdev_op_leaf) 4588 return (resilver_needed); 4589 4590 vd->vdev_resilver_deferred = B_FALSE; 4591 4592 return (!vdev_is_dead(vd) && !vd->vdev_offline && 4593 vdev_resilver_needed(vd, NULL, NULL)); 4594 } 4595 4596 /* 4597 * Translate a logical range to the physical range for the specified vdev_t. 4598 * This function is initially called with a leaf vdev and will walk each 4599 * parent vdev until it reaches a top-level vdev. Once the top-level is 4600 * reached the physical range is initialized and the recursive function 4601 * begins to unwind. As it unwinds it calls the parent's vdev specific 4602 * translation function to do the real conversion. 4603 */ 4604 void 4605 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 4606 range_seg64_t *physical_rs) 4607 { 4608 /* 4609 * Walk up the vdev tree 4610 */ 4611 if (vd != vd->vdev_top) { 4612 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs); 4613 } else { 4614 /* 4615 * We've reached the top-level vdev, initialize the 4616 * physical range to the logical range and start to 4617 * unwind. 4618 */ 4619 physical_rs->rs_start = logical_rs->rs_start; 4620 physical_rs->rs_end = logical_rs->rs_end; 4621 return; 4622 } 4623 4624 vdev_t *pvd = vd->vdev_parent; 4625 ASSERT3P(pvd, !=, NULL); 4626 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 4627 4628 /* 4629 * As this recursive function unwinds, translate the logical 4630 * range into its physical components by calling the 4631 * vdev specific translate function. 4632 */ 4633 range_seg64_t intermediate = { 0 }; 4634 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate); 4635 4636 physical_rs->rs_start = intermediate.rs_start; 4637 physical_rs->rs_end = intermediate.rs_end; 4638 } 4639