vdev_label.c revision 308915
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) 2012, 2015 by Delphix. All rights reserved. 25 */ 26 27/* 28 * Virtual Device Labels 29 * --------------------- 30 * 31 * The vdev label serves several distinct purposes: 32 * 33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its 34 * identity within the pool. 35 * 36 * 2. Verify that all the devices given in a configuration are present 37 * within the pool. 38 * 39 * 3. Determine the uberblock for the pool. 40 * 41 * 4. In case of an import operation, determine the configuration of the 42 * toplevel vdev of which it is a part. 43 * 44 * 5. If an import operation cannot find all the devices in the pool, 45 * provide enough information to the administrator to determine which 46 * devices are missing. 47 * 48 * It is important to note that while the kernel is responsible for writing the 49 * label, it only consumes the information in the first three cases. The 50 * latter information is only consumed in userland when determining the 51 * configuration to import a pool. 52 * 53 * 54 * Label Organization 55 * ------------------ 56 * 57 * Before describing the contents of the label, it's important to understand how 58 * the labels are written and updated with respect to the uberblock. 59 * 60 * When the pool configuration is altered, either because it was newly created 61 * or a device was added, we want to update all the labels such that we can deal 62 * with fatal failure at any point. To this end, each disk has two labels which 63 * are updated before and after the uberblock is synced. Assuming we have 64 * labels and an uberblock with the following transaction groups: 65 * 66 * L1 UB L2 67 * +------+ +------+ +------+ 68 * | | | | | | 69 * | t10 | | t10 | | t10 | 70 * | | | | | | 71 * +------+ +------+ +------+ 72 * 73 * In this stable state, the labels and the uberblock were all updated within 74 * the same transaction group (10). Each label is mirrored and checksummed, so 75 * that we can detect when we fail partway through writing the label. 76 * 77 * In order to identify which labels are valid, the labels are written in the 78 * following manner: 79 * 80 * 1. For each vdev, update 'L1' to the new label 81 * 2. Update the uberblock 82 * 3. For each vdev, update 'L2' to the new label 83 * 84 * Given arbitrary failure, we can determine the correct label to use based on 85 * the transaction group. If we fail after updating L1 but before updating the 86 * UB, we will notice that L1's transaction group is greater than the uberblock, 87 * so L2 must be valid. If we fail after writing the uberblock but before 88 * writing L2, we will notice that L2's transaction group is less than L1, and 89 * therefore L1 is valid. 90 * 91 * Another added complexity is that not every label is updated when the config 92 * is synced. If we add a single device, we do not want to have to re-write 93 * every label for every device in the pool. This means that both L1 and L2 may 94 * be older than the pool uberblock, because the necessary information is stored 95 * on another vdev. 96 * 97 * 98 * On-disk Format 99 * -------------- 100 * 101 * The vdev label consists of two distinct parts, and is wrapped within the 102 * vdev_label_t structure. The label includes 8k of padding to permit legacy 103 * VTOC disk labels, but is otherwise ignored. 104 * 105 * The first half of the label is a packed nvlist which contains pool wide 106 * properties, per-vdev properties, and configuration information. It is 107 * described in more detail below. 108 * 109 * The latter half of the label consists of a redundant array of uberblocks. 110 * These uberblocks are updated whenever a transaction group is committed, 111 * or when the configuration is updated. When a pool is loaded, we scan each 112 * vdev for the 'best' uberblock. 113 * 114 * 115 * Configuration Information 116 * ------------------------- 117 * 118 * The nvlist describing the pool and vdev contains the following elements: 119 * 120 * version ZFS on-disk version 121 * name Pool name 122 * state Pool state 123 * txg Transaction group in which this label was written 124 * pool_guid Unique identifier for this pool 125 * vdev_tree An nvlist describing vdev tree. 126 * features_for_read 127 * An nvlist of the features necessary for reading the MOS. 128 * 129 * Each leaf device label also contains the following: 130 * 131 * top_guid Unique ID for top-level vdev in which this is contained 132 * guid Unique ID for the leaf vdev 133 * 134 * The 'vs' configuration follows the format described in 'spa_config.c'. 135 */ 136 137#include <sys/zfs_context.h> 138#include <sys/spa.h> 139#include <sys/spa_impl.h> 140#include <sys/dmu.h> 141#include <sys/zap.h> 142#include <sys/vdev.h> 143#include <sys/vdev_impl.h> 144#include <sys/uberblock_impl.h> 145#include <sys/metaslab.h> 146#include <sys/zio.h> 147#include <sys/dsl_scan.h> 148#include <sys/trim_map.h> 149#include <sys/fs/zfs.h> 150 151static boolean_t vdev_trim_on_init = B_TRUE; 152SYSCTL_DECL(_vfs_zfs_vdev); 153SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW, 154 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation"); 155 156/* 157 * Basic routines to read and write from a vdev label. 158 * Used throughout the rest of this file. 159 */ 160uint64_t 161vdev_label_offset(uint64_t psize, int l, uint64_t offset) 162{ 163 ASSERT(offset < sizeof (vdev_label_t)); 164 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 165 166 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 167 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 168} 169 170/* 171 * Returns back the vdev label associated with the passed in offset. 172 */ 173int 174vdev_label_number(uint64_t psize, uint64_t offset) 175{ 176 int l; 177 178 if (offset >= psize - VDEV_LABEL_END_SIZE) { 179 offset -= psize - VDEV_LABEL_END_SIZE; 180 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 181 } 182 l = offset / sizeof (vdev_label_t); 183 return (l < VDEV_LABELS ? l : -1); 184} 185 186static void 187vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 188 uint64_t size, zio_done_func_t *done, void *private, int flags) 189{ 190 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 191 SCL_STATE_ALL); 192 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 193 194 zio_nowait(zio_read_phys(zio, vd, 195 vdev_label_offset(vd->vdev_psize, l, offset), 196 size, buf, ZIO_CHECKSUM_LABEL, done, private, 197 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 198} 199 200static void 201vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 202 uint64_t size, zio_done_func_t *done, void *private, int flags) 203{ 204 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 205 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 206 (SCL_CONFIG | SCL_STATE) && 207 dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 208 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 209 210 zio_nowait(zio_write_phys(zio, vd, 211 vdev_label_offset(vd->vdev_psize, l, offset), 212 size, buf, ZIO_CHECKSUM_LABEL, done, private, 213 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 214} 215 216/* 217 * Generate the nvlist representing this vdev's config. 218 */ 219nvlist_t * 220vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 221 vdev_config_flag_t flags) 222{ 223 nvlist_t *nv = NULL; 224 225 nv = fnvlist_alloc(); 226 227 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 228 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 229 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 230 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 231 232 if (vd->vdev_path != NULL) 233 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 234 235 if (vd->vdev_devid != NULL) 236 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 237 238 if (vd->vdev_physpath != NULL) 239 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 240 vd->vdev_physpath); 241 242 if (vd->vdev_fru != NULL) 243 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 244 245 if (vd->vdev_nparity != 0) { 246 ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 247 VDEV_TYPE_RAIDZ) == 0); 248 249 /* 250 * Make sure someone hasn't managed to sneak a fancy new vdev 251 * into a crufty old storage pool. 252 */ 253 ASSERT(vd->vdev_nparity == 1 || 254 (vd->vdev_nparity <= 2 && 255 spa_version(spa) >= SPA_VERSION_RAIDZ2) || 256 (vd->vdev_nparity <= 3 && 257 spa_version(spa) >= SPA_VERSION_RAIDZ3)); 258 259 /* 260 * Note that we'll add the nparity tag even on storage pools 261 * that only support a single parity device -- older software 262 * will just ignore it. 263 */ 264 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 265 } 266 267 if (vd->vdev_wholedisk != -1ULL) 268 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 269 vd->vdev_wholedisk); 270 271 if (vd->vdev_not_present) 272 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 273 274 if (vd->vdev_isspare) 275 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 276 277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 278 vd == vd->vdev_top) { 279 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 280 vd->vdev_ms_array); 281 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 282 vd->vdev_ms_shift); 283 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 284 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 285 vd->vdev_asize); 286 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 287 if (vd->vdev_removing) 288 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 289 vd->vdev_removing); 290 } 291 292 if (vd->vdev_dtl_sm != NULL) { 293 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 294 space_map_object(vd->vdev_dtl_sm)); 295 } 296 297 if (vd->vdev_crtxg) 298 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 299 300 if (getstats) { 301 vdev_stat_t vs; 302 pool_scan_stat_t ps; 303 304 vdev_get_stats(vd, &vs); 305 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 306 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 307 308 /* provide either current or previous scan information */ 309 if (spa_scan_get_stats(spa, &ps) == 0) { 310 fnvlist_add_uint64_array(nv, 311 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 312 sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 313 } 314 } 315 316 if (!vd->vdev_ops->vdev_op_leaf) { 317 nvlist_t **child; 318 int c, idx; 319 320 ASSERT(!vd->vdev_ishole); 321 322 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 323 KM_SLEEP); 324 325 for (c = 0, idx = 0; c < vd->vdev_children; c++) { 326 vdev_t *cvd = vd->vdev_child[c]; 327 328 /* 329 * If we're generating an nvlist of removing 330 * vdevs then skip over any device which is 331 * not being removed. 332 */ 333 if ((flags & VDEV_CONFIG_REMOVING) && 334 !cvd->vdev_removing) 335 continue; 336 337 child[idx++] = vdev_config_generate(spa, cvd, 338 getstats, flags); 339 } 340 341 if (idx) { 342 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 343 child, idx); 344 } 345 346 for (c = 0; c < idx; c++) 347 nvlist_free(child[c]); 348 349 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 350 351 } else { 352 const char *aux = NULL; 353 354 if (vd->vdev_offline && !vd->vdev_tmpoffline) 355 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 356 if (vd->vdev_resilver_txg != 0) 357 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 358 vd->vdev_resilver_txg); 359 if (vd->vdev_faulted) 360 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 361 if (vd->vdev_degraded) 362 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 363 if (vd->vdev_removed) 364 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 365 if (vd->vdev_unspare) 366 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 367 if (vd->vdev_ishole) 368 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 369 370 switch (vd->vdev_stat.vs_aux) { 371 case VDEV_AUX_ERR_EXCEEDED: 372 aux = "err_exceeded"; 373 break; 374 375 case VDEV_AUX_EXTERNAL: 376 aux = "external"; 377 break; 378 } 379 380 if (aux != NULL) 381 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 382 383 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 384 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 385 vd->vdev_orig_guid); 386 } 387 } 388 389 return (nv); 390} 391 392/* 393 * Generate a view of the top-level vdevs. If we currently have holes 394 * in the namespace, then generate an array which contains a list of holey 395 * vdevs. Additionally, add the number of top-level children that currently 396 * exist. 397 */ 398void 399vdev_top_config_generate(spa_t *spa, nvlist_t *config) 400{ 401 vdev_t *rvd = spa->spa_root_vdev; 402 uint64_t *array; 403 uint_t c, idx; 404 405 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 406 407 for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 408 vdev_t *tvd = rvd->vdev_child[c]; 409 410 if (tvd->vdev_ishole) 411 array[idx++] = c; 412 } 413 414 if (idx) { 415 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 416 array, idx) == 0); 417 } 418 419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 420 rvd->vdev_children) == 0); 421 422 kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 423} 424 425/* 426 * Returns the configuration from the label of the given vdev. For vdevs 427 * which don't have a txg value stored on their label (i.e. spares/cache) 428 * or have not been completely initialized (txg = 0) just return 429 * the configuration from the first valid label we find. Otherwise, 430 * find the most up-to-date label that does not exceed the specified 431 * 'txg' value. 432 */ 433nvlist_t * 434vdev_label_read_config(vdev_t *vd, uint64_t txg) 435{ 436 spa_t *spa = vd->vdev_spa; 437 nvlist_t *config = NULL; 438 vdev_phys_t *vp; 439 zio_t *zio; 440 uint64_t best_txg = 0; 441 int error = 0; 442 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 443 ZIO_FLAG_SPECULATIVE; 444 445 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 446 447 if (!vdev_readable(vd)) 448 return (NULL); 449 450 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 451 452retry: 453 for (int l = 0; l < VDEV_LABELS; l++) { 454 nvlist_t *label = NULL; 455 456 zio = zio_root(spa, NULL, NULL, flags); 457 458 vdev_label_read(zio, vd, l, vp, 459 offsetof(vdev_label_t, vl_vdev_phys), 460 sizeof (vdev_phys_t), NULL, NULL, flags); 461 462 if (zio_wait(zio) == 0 && 463 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 464 &label, 0) == 0) { 465 uint64_t label_txg = 0; 466 467 /* 468 * Auxiliary vdevs won't have txg values in their 469 * labels and newly added vdevs may not have been 470 * completely initialized so just return the 471 * configuration from the first valid label we 472 * encounter. 473 */ 474 error = nvlist_lookup_uint64(label, 475 ZPOOL_CONFIG_POOL_TXG, &label_txg); 476 if ((error || label_txg == 0) && !config) { 477 config = label; 478 break; 479 } else if (label_txg <= txg && label_txg > best_txg) { 480 best_txg = label_txg; 481 nvlist_free(config); 482 config = fnvlist_dup(label); 483 } 484 } 485 486 if (label != NULL) { 487 nvlist_free(label); 488 label = NULL; 489 } 490 } 491 492 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 493 flags |= ZIO_FLAG_TRYHARD; 494 goto retry; 495 } 496 497 zio_buf_free(vp, sizeof (vdev_phys_t)); 498 499 return (config); 500} 501 502/* 503 * Determine if a device is in use. The 'spare_guid' parameter will be filled 504 * in with the device guid if this spare is active elsewhere on the system. 505 */ 506static boolean_t 507vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 508 uint64_t *spare_guid, uint64_t *l2cache_guid) 509{ 510 spa_t *spa = vd->vdev_spa; 511 uint64_t state, pool_guid, device_guid, txg, spare_pool; 512 uint64_t vdtxg = 0; 513 nvlist_t *label; 514 515 if (spare_guid) 516 *spare_guid = 0ULL; 517 if (l2cache_guid) 518 *l2cache_guid = 0ULL; 519 520 /* 521 * Read the label, if any, and perform some basic sanity checks. 522 */ 523 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 524 return (B_FALSE); 525 526 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 527 &vdtxg); 528 529 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 530 &state) != 0 || 531 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 532 &device_guid) != 0) { 533 nvlist_free(label); 534 return (B_FALSE); 535 } 536 537 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 538 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 539 &pool_guid) != 0 || 540 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 541 &txg) != 0)) { 542 nvlist_free(label); 543 return (B_FALSE); 544 } 545 546 nvlist_free(label); 547 548 /* 549 * Check to see if this device indeed belongs to the pool it claims to 550 * be a part of. The only way this is allowed is if the device is a hot 551 * spare (which we check for later on). 552 */ 553 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 554 !spa_guid_exists(pool_guid, device_guid) && 555 !spa_spare_exists(device_guid, NULL, NULL) && 556 !spa_l2cache_exists(device_guid, NULL)) 557 return (B_FALSE); 558 559 /* 560 * If the transaction group is zero, then this an initialized (but 561 * unused) label. This is only an error if the create transaction 562 * on-disk is the same as the one we're using now, in which case the 563 * user has attempted to add the same vdev multiple times in the same 564 * transaction. 565 */ 566 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 567 txg == 0 && vdtxg == crtxg) 568 return (B_TRUE); 569 570 /* 571 * Check to see if this is a spare device. We do an explicit check for 572 * spa_has_spare() here because it may be on our pending list of spares 573 * to add. We also check if it is an l2cache device. 574 */ 575 if (spa_spare_exists(device_guid, &spare_pool, NULL) || 576 spa_has_spare(spa, device_guid)) { 577 if (spare_guid) 578 *spare_guid = device_guid; 579 580 switch (reason) { 581 case VDEV_LABEL_CREATE: 582 case VDEV_LABEL_L2CACHE: 583 return (B_TRUE); 584 585 case VDEV_LABEL_REPLACE: 586 return (!spa_has_spare(spa, device_guid) || 587 spare_pool != 0ULL); 588 589 case VDEV_LABEL_SPARE: 590 return (spa_has_spare(spa, device_guid)); 591 } 592 } 593 594 /* 595 * Check to see if this is an l2cache device. 596 */ 597 if (spa_l2cache_exists(device_guid, NULL)) 598 return (B_TRUE); 599 600 /* 601 * We can't rely on a pool's state if it's been imported 602 * read-only. Instead we look to see if the pools is marked 603 * read-only in the namespace and set the state to active. 604 */ 605 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 606 (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 607 spa_mode(spa) == FREAD) 608 state = POOL_STATE_ACTIVE; 609 610 /* 611 * If the device is marked ACTIVE, then this device is in use by another 612 * pool on the system. 613 */ 614 return (state == POOL_STATE_ACTIVE); 615} 616 617/* 618 * Initialize a vdev label. We check to make sure each leaf device is not in 619 * use, and writable. We put down an initial label which we will later 620 * overwrite with a complete label. Note that it's important to do this 621 * sequentially, not in parallel, so that we catch cases of multiple use of the 622 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 623 * itself. 624 */ 625int 626vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 627{ 628 spa_t *spa = vd->vdev_spa; 629 nvlist_t *label; 630 vdev_phys_t *vp; 631 char *pad2; 632 uberblock_t *ub; 633 zio_t *zio; 634 char *buf; 635 size_t buflen; 636 int error; 637 uint64_t spare_guid, l2cache_guid; 638 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 639 640 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 641 642 for (int c = 0; c < vd->vdev_children; c++) 643 if ((error = vdev_label_init(vd->vdev_child[c], 644 crtxg, reason)) != 0) 645 return (error); 646 647 /* Track the creation time for this vdev */ 648 vd->vdev_crtxg = crtxg; 649 650 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 651 return (0); 652 653 /* 654 * Dead vdevs cannot be initialized. 655 */ 656 if (vdev_is_dead(vd)) 657 return (SET_ERROR(EIO)); 658 659 /* 660 * Determine if the vdev is in use. 661 */ 662 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 663 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 664 return (SET_ERROR(EBUSY)); 665 666 /* 667 * If this is a request to add or replace a spare or l2cache device 668 * that is in use elsewhere on the system, then we must update the 669 * guid (which was initialized to a random value) to reflect the 670 * actual GUID (which is shared between multiple pools). 671 */ 672 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 673 spare_guid != 0ULL) { 674 uint64_t guid_delta = spare_guid - vd->vdev_guid; 675 676 vd->vdev_guid += guid_delta; 677 678 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 679 pvd->vdev_guid_sum += guid_delta; 680 681 /* 682 * If this is a replacement, then we want to fallthrough to the 683 * rest of the code. If we're adding a spare, then it's already 684 * labeled appropriately and we can just return. 685 */ 686 if (reason == VDEV_LABEL_SPARE) 687 return (0); 688 ASSERT(reason == VDEV_LABEL_REPLACE || 689 reason == VDEV_LABEL_SPLIT); 690 } 691 692 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 693 l2cache_guid != 0ULL) { 694 uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 695 696 vd->vdev_guid += guid_delta; 697 698 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 699 pvd->vdev_guid_sum += guid_delta; 700 701 /* 702 * If this is a replacement, then we want to fallthrough to the 703 * rest of the code. If we're adding an l2cache, then it's 704 * already labeled appropriately and we can just return. 705 */ 706 if (reason == VDEV_LABEL_L2CACHE) 707 return (0); 708 ASSERT(reason == VDEV_LABEL_REPLACE); 709 } 710 711 /* 712 * TRIM the whole thing so that we start with a clean slate. 713 * It's just an optimization, so we don't care if it fails. 714 * Don't TRIM if removing so that we don't interfere with zpool 715 * disaster recovery. 716 */ 717 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim && 718 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE || 719 reason == VDEV_LABEL_L2CACHE)) 720 zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize)); 721 722 /* 723 * Initialize its label. 724 */ 725 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 726 bzero(vp, sizeof (vdev_phys_t)); 727 728 /* 729 * Generate a label describing the pool and our top-level vdev. 730 * We mark it as being from txg 0 to indicate that it's not 731 * really part of an active pool just yet. The labels will 732 * be written again with a meaningful txg by spa_sync(). 733 */ 734 if (reason == VDEV_LABEL_SPARE || 735 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 736 /* 737 * For inactive hot spares, we generate a special label that 738 * identifies as a mutually shared hot spare. We write the 739 * label if we are adding a hot spare, or if we are removing an 740 * active hot spare (in which case we want to revert the 741 * labels). 742 */ 743 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 744 745 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 746 spa_version(spa)) == 0); 747 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 748 POOL_STATE_SPARE) == 0); 749 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 750 vd->vdev_guid) == 0); 751 } else if (reason == VDEV_LABEL_L2CACHE || 752 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 753 /* 754 * For level 2 ARC devices, add a special label. 755 */ 756 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 757 758 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 759 spa_version(spa)) == 0); 760 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 761 POOL_STATE_L2CACHE) == 0); 762 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 763 vd->vdev_guid) == 0); 764 } else { 765 uint64_t txg = 0ULL; 766 767 if (reason == VDEV_LABEL_SPLIT) 768 txg = spa->spa_uberblock.ub_txg; 769 label = spa_config_generate(spa, vd, txg, B_FALSE); 770 771 /* 772 * Add our creation time. This allows us to detect multiple 773 * vdev uses as described above, and automatically expires if we 774 * fail. 775 */ 776 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 777 crtxg) == 0); 778 } 779 780 buf = vp->vp_nvlist; 781 buflen = sizeof (vp->vp_nvlist); 782 783 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 784 if (error != 0) { 785 nvlist_free(label); 786 zio_buf_free(vp, sizeof (vdev_phys_t)); 787 /* EFAULT means nvlist_pack ran out of room */ 788 return (error == EFAULT ? ENAMETOOLONG : EINVAL); 789 } 790 791 /* 792 * Initialize uberblock template. 793 */ 794 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 795 bzero(ub, VDEV_UBERBLOCK_RING); 796 *ub = spa->spa_uberblock; 797 ub->ub_txg = 0; 798 799 /* Initialize the 2nd padding area. */ 800 pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 801 bzero(pad2, VDEV_PAD_SIZE); 802 803 /* 804 * Write everything in parallel. 805 */ 806retry: 807 zio = zio_root(spa, NULL, NULL, flags); 808 809 for (int l = 0; l < VDEV_LABELS; l++) { 810 811 vdev_label_write(zio, vd, l, vp, 812 offsetof(vdev_label_t, vl_vdev_phys), 813 sizeof (vdev_phys_t), NULL, NULL, flags); 814 815 /* 816 * Skip the 1st padding area. 817 * Zero out the 2nd padding area where it might have 818 * left over data from previous filesystem format. 819 */ 820 vdev_label_write(zio, vd, l, pad2, 821 offsetof(vdev_label_t, vl_pad2), 822 VDEV_PAD_SIZE, NULL, NULL, flags); 823 824 vdev_label_write(zio, vd, l, ub, 825 offsetof(vdev_label_t, vl_uberblock), 826 VDEV_UBERBLOCK_RING, NULL, NULL, flags); 827 } 828 829 error = zio_wait(zio); 830 831 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 832 flags |= ZIO_FLAG_TRYHARD; 833 goto retry; 834 } 835 836 nvlist_free(label); 837 zio_buf_free(pad2, VDEV_PAD_SIZE); 838 zio_buf_free(ub, VDEV_UBERBLOCK_RING); 839 zio_buf_free(vp, sizeof (vdev_phys_t)); 840 841 /* 842 * If this vdev hasn't been previously identified as a spare, then we 843 * mark it as such only if a) we are labeling it as a spare, or b) it 844 * exists as a spare elsewhere in the system. Do the same for 845 * level 2 ARC devices. 846 */ 847 if (error == 0 && !vd->vdev_isspare && 848 (reason == VDEV_LABEL_SPARE || 849 spa_spare_exists(vd->vdev_guid, NULL, NULL))) 850 spa_spare_add(vd); 851 852 if (error == 0 && !vd->vdev_isl2cache && 853 (reason == VDEV_LABEL_L2CACHE || 854 spa_l2cache_exists(vd->vdev_guid, NULL))) 855 spa_l2cache_add(vd); 856 857 return (error); 858} 859 860int 861vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size) 862{ 863 spa_t *spa = vd->vdev_spa; 864 zio_t *zio; 865 char *pad2; 866 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 867 int error; 868 869 if (size > VDEV_PAD_SIZE) 870 return (EINVAL); 871 872 if (!vd->vdev_ops->vdev_op_leaf) 873 return (ENODEV); 874 if (vdev_is_dead(vd)) 875 return (ENXIO); 876 877 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 878 879 pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 880 bzero(pad2, VDEV_PAD_SIZE); 881 memcpy(pad2, buf, size); 882 883retry: 884 zio = zio_root(spa, NULL, NULL, flags); 885 vdev_label_write(zio, vd, 0, pad2, 886 offsetof(vdev_label_t, vl_pad2), 887 VDEV_PAD_SIZE, NULL, NULL, flags); 888 error = zio_wait(zio); 889 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 890 flags |= ZIO_FLAG_TRYHARD; 891 goto retry; 892 } 893 894 zio_buf_free(pad2, VDEV_PAD_SIZE); 895 return (error); 896} 897 898/* 899 * ========================================================================== 900 * uberblock load/sync 901 * ========================================================================== 902 */ 903 904/* 905 * Consider the following situation: txg is safely synced to disk. We've 906 * written the first uberblock for txg + 1, and then we lose power. When we 907 * come back up, we fail to see the uberblock for txg + 1 because, say, 908 * it was on a mirrored device and the replica to which we wrote txg + 1 909 * is now offline. If we then make some changes and sync txg + 1, and then 910 * the missing replica comes back, then for a few seconds we'll have two 911 * conflicting uberblocks on disk with the same txg. The solution is simple: 912 * among uberblocks with equal txg, choose the one with the latest timestamp. 913 */ 914static int 915vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 916{ 917 if (ub1->ub_txg < ub2->ub_txg) 918 return (-1); 919 if (ub1->ub_txg > ub2->ub_txg) 920 return (1); 921 922 if (ub1->ub_timestamp < ub2->ub_timestamp) 923 return (-1); 924 if (ub1->ub_timestamp > ub2->ub_timestamp) 925 return (1); 926 927 return (0); 928} 929 930struct ubl_cbdata { 931 uberblock_t *ubl_ubbest; /* Best uberblock */ 932 vdev_t *ubl_vd; /* vdev associated with the above */ 933}; 934 935static void 936vdev_uberblock_load_done(zio_t *zio) 937{ 938 vdev_t *vd = zio->io_vd; 939 spa_t *spa = zio->io_spa; 940 zio_t *rio = zio->io_private; 941 uberblock_t *ub = zio->io_data; 942 struct ubl_cbdata *cbp = rio->io_private; 943 944 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 945 946 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 947 mutex_enter(&rio->io_lock); 948 if (ub->ub_txg <= spa->spa_load_max_txg && 949 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 950 /* 951 * Keep track of the vdev in which this uberblock 952 * was found. We will use this information later 953 * to obtain the config nvlist associated with 954 * this uberblock. 955 */ 956 *cbp->ubl_ubbest = *ub; 957 cbp->ubl_vd = vd; 958 } 959 mutex_exit(&rio->io_lock); 960 } 961 962 zio_buf_free(zio->io_data, zio->io_size); 963} 964 965static void 966vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 967 struct ubl_cbdata *cbp) 968{ 969 for (int c = 0; c < vd->vdev_children; c++) 970 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 971 972 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 973 for (int l = 0; l < VDEV_LABELS; l++) { 974 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 975 vdev_label_read(zio, vd, l, 976 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 977 VDEV_UBERBLOCK_OFFSET(vd, n), 978 VDEV_UBERBLOCK_SIZE(vd), 979 vdev_uberblock_load_done, zio, flags); 980 } 981 } 982 } 983} 984 985/* 986 * Reads the 'best' uberblock from disk along with its associated 987 * configuration. First, we read the uberblock array of each label of each 988 * vdev, keeping track of the uberblock with the highest txg in each array. 989 * Then, we read the configuration from the same vdev as the best uberblock. 990 */ 991void 992vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 993{ 994 zio_t *zio; 995 spa_t *spa = rvd->vdev_spa; 996 struct ubl_cbdata cb; 997 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 998 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 999 1000 ASSERT(ub); 1001 ASSERT(config); 1002 1003 bzero(ub, sizeof (uberblock_t)); 1004 *config = NULL; 1005 1006 cb.ubl_ubbest = ub; 1007 cb.ubl_vd = NULL; 1008 1009 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1010 zio = zio_root(spa, NULL, &cb, flags); 1011 vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1012 (void) zio_wait(zio); 1013 1014 /* 1015 * It's possible that the best uberblock was discovered on a label 1016 * that has a configuration which was written in a future txg. 1017 * Search all labels on this vdev to find the configuration that 1018 * matches the txg for our uberblock. 1019 */ 1020 if (cb.ubl_vd != NULL) 1021 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 1022 spa_config_exit(spa, SCL_ALL, FTAG); 1023} 1024 1025/* 1026 * On success, increment root zio's count of good writes. 1027 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1028 */ 1029static void 1030vdev_uberblock_sync_done(zio_t *zio) 1031{ 1032 uint64_t *good_writes = zio->io_private; 1033 1034 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1035 atomic_inc_64(good_writes); 1036} 1037 1038/* 1039 * Write the uberblock to all labels of all leaves of the specified vdev. 1040 */ 1041static void 1042vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 1043{ 1044 uberblock_t *ubbuf; 1045 int n; 1046 1047 for (int c = 0; c < vd->vdev_children; c++) 1048 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 1049 1050 if (!vd->vdev_ops->vdev_op_leaf) 1051 return; 1052 1053 if (!vdev_writeable(vd)) 1054 return; 1055 1056 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 1057 1058 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 1059 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 1060 *ubbuf = *ub; 1061 1062 for (int l = 0; l < VDEV_LABELS; l++) 1063 vdev_label_write(zio, vd, l, ubbuf, 1064 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1065 vdev_uberblock_sync_done, zio->io_private, 1066 flags | ZIO_FLAG_DONT_PROPAGATE); 1067 1068 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 1069} 1070 1071/* Sync the uberblocks to all vdevs in svd[] */ 1072int 1073vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1074{ 1075 spa_t *spa = svd[0]->vdev_spa; 1076 zio_t *zio; 1077 uint64_t good_writes = 0; 1078 1079 zio = zio_root(spa, NULL, &good_writes, flags); 1080 1081 for (int v = 0; v < svdcount; v++) 1082 vdev_uberblock_sync(zio, ub, svd[v], flags); 1083 1084 (void) zio_wait(zio); 1085 1086 /* 1087 * Flush the uberblocks to disk. This ensures that the odd labels 1088 * are no longer needed (because the new uberblocks and the even 1089 * labels are safely on disk), so it is safe to overwrite them. 1090 */ 1091 zio = zio_root(spa, NULL, NULL, flags); 1092 1093 for (int v = 0; v < svdcount; v++) 1094 zio_flush(zio, svd[v]); 1095 1096 (void) zio_wait(zio); 1097 1098 return (good_writes >= 1 ? 0 : EIO); 1099} 1100 1101/* 1102 * On success, increment the count of good writes for our top-level vdev. 1103 */ 1104static void 1105vdev_label_sync_done(zio_t *zio) 1106{ 1107 uint64_t *good_writes = zio->io_private; 1108 1109 if (zio->io_error == 0) 1110 atomic_inc_64(good_writes); 1111} 1112 1113/* 1114 * If there weren't enough good writes, indicate failure to the parent. 1115 */ 1116static void 1117vdev_label_sync_top_done(zio_t *zio) 1118{ 1119 uint64_t *good_writes = zio->io_private; 1120 1121 if (*good_writes == 0) 1122 zio->io_error = SET_ERROR(EIO); 1123 1124 kmem_free(good_writes, sizeof (uint64_t)); 1125} 1126 1127/* 1128 * We ignore errors for log and cache devices, simply free the private data. 1129 */ 1130static void 1131vdev_label_sync_ignore_done(zio_t *zio) 1132{ 1133 kmem_free(zio->io_private, sizeof (uint64_t)); 1134} 1135 1136/* 1137 * Write all even or odd labels to all leaves of the specified vdev. 1138 */ 1139static void 1140vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1141{ 1142 nvlist_t *label; 1143 vdev_phys_t *vp; 1144 char *buf; 1145 size_t buflen; 1146 1147 for (int c = 0; c < vd->vdev_children; c++) 1148 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1149 1150 if (!vd->vdev_ops->vdev_op_leaf) 1151 return; 1152 1153 if (!vdev_writeable(vd)) 1154 return; 1155 1156 /* 1157 * Generate a label describing the top-level config to which we belong. 1158 */ 1159 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1160 1161 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1162 bzero(vp, sizeof (vdev_phys_t)); 1163 1164 buf = vp->vp_nvlist; 1165 buflen = sizeof (vp->vp_nvlist); 1166 1167 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1168 for (; l < VDEV_LABELS; l += 2) { 1169 vdev_label_write(zio, vd, l, vp, 1170 offsetof(vdev_label_t, vl_vdev_phys), 1171 sizeof (vdev_phys_t), 1172 vdev_label_sync_done, zio->io_private, 1173 flags | ZIO_FLAG_DONT_PROPAGATE); 1174 } 1175 } 1176 1177 zio_buf_free(vp, sizeof (vdev_phys_t)); 1178 nvlist_free(label); 1179} 1180 1181int 1182vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1183{ 1184 list_t *dl = &spa->spa_config_dirty_list; 1185 vdev_t *vd; 1186 zio_t *zio; 1187 int error; 1188 1189 /* 1190 * Write the new labels to disk. 1191 */ 1192 zio = zio_root(spa, NULL, NULL, flags); 1193 1194 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1195 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1196 KM_SLEEP); 1197 1198 ASSERT(!vd->vdev_ishole); 1199 1200 zio_t *vio = zio_null(zio, spa, NULL, 1201 (vd->vdev_islog || vd->vdev_aux != NULL) ? 1202 vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1203 good_writes, flags); 1204 vdev_label_sync(vio, vd, l, txg, flags); 1205 zio_nowait(vio); 1206 } 1207 1208 error = zio_wait(zio); 1209 1210 /* 1211 * Flush the new labels to disk. 1212 */ 1213 zio = zio_root(spa, NULL, NULL, flags); 1214 1215 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1216 zio_flush(zio, vd); 1217 1218 (void) zio_wait(zio); 1219 1220 return (error); 1221} 1222 1223/* 1224 * Sync the uberblock and any changes to the vdev configuration. 1225 * 1226 * The order of operations is carefully crafted to ensure that 1227 * if the system panics or loses power at any time, the state on disk 1228 * is still transactionally consistent. The in-line comments below 1229 * describe the failure semantics at each stage. 1230 * 1231 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1232 * at any time, you can just call it again, and it will resume its work. 1233 */ 1234int 1235vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1236{ 1237 spa_t *spa = svd[0]->vdev_spa; 1238 uberblock_t *ub = &spa->spa_uberblock; 1239 vdev_t *vd; 1240 zio_t *zio; 1241 int error = 0; 1242 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1243 1244retry: 1245 /* 1246 * Normally, we don't want to try too hard to write every label and 1247 * uberblock. If there is a flaky disk, we don't want the rest of the 1248 * sync process to block while we retry. But if we can't write a 1249 * single label out, we should retry with ZIO_FLAG_TRYHARD before 1250 * bailing out and declaring the pool faulted. 1251 */ 1252 if (error != 0) { 1253 if ((flags & ZIO_FLAG_TRYHARD) != 0) 1254 return (error); 1255 flags |= ZIO_FLAG_TRYHARD; 1256 } 1257 1258 ASSERT(ub->ub_txg <= txg); 1259 1260 /* 1261 * If this isn't a resync due to I/O errors, 1262 * and nothing changed in this transaction group, 1263 * and the vdev configuration hasn't changed, 1264 * then there's nothing to do. 1265 */ 1266 if (ub->ub_txg < txg && 1267 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1268 list_is_empty(&spa->spa_config_dirty_list)) 1269 return (0); 1270 1271 if (txg > spa_freeze_txg(spa)) 1272 return (0); 1273 1274 ASSERT(txg <= spa->spa_final_txg); 1275 1276 /* 1277 * Flush the write cache of every disk that's been written to 1278 * in this transaction group. This ensures that all blocks 1279 * written in this txg will be committed to stable storage 1280 * before any uberblock that references them. 1281 */ 1282 zio = zio_root(spa, NULL, NULL, flags); 1283 1284 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 1285 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1286 zio_flush(zio, vd); 1287 1288 (void) zio_wait(zio); 1289 1290 /* 1291 * Sync out the even labels (L0, L2) for every dirty vdev. If the 1292 * system dies in the middle of this process, that's OK: all of the 1293 * even labels that made it to disk will be newer than any uberblock, 1294 * and will therefore be considered invalid. The odd labels (L1, L3), 1295 * which have not yet been touched, will still be valid. We flush 1296 * the new labels to disk to ensure that all even-label updates 1297 * are committed to stable storage before the uberblock update. 1298 */ 1299 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1300 goto retry; 1301 1302 /* 1303 * Sync the uberblocks to all vdevs in svd[]. 1304 * If the system dies in the middle of this step, there are two cases 1305 * to consider, and the on-disk state is consistent either way: 1306 * 1307 * (1) If none of the new uberblocks made it to disk, then the 1308 * previous uberblock will be the newest, and the odd labels 1309 * (which had not yet been touched) will be valid with respect 1310 * to that uberblock. 1311 * 1312 * (2) If one or more new uberblocks made it to disk, then they 1313 * will be the newest, and the even labels (which had all 1314 * been successfully committed) will be valid with respect 1315 * to the new uberblocks. 1316 */ 1317 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1318 goto retry; 1319 1320 /* 1321 * Sync out odd labels for every dirty vdev. If the system dies 1322 * in the middle of this process, the even labels and the new 1323 * uberblocks will suffice to open the pool. The next time 1324 * the pool is opened, the first thing we'll do -- before any 1325 * user data is modified -- is mark every vdev dirty so that 1326 * all labels will be brought up to date. We flush the new labels 1327 * to disk to ensure that all odd-label updates are committed to 1328 * stable storage before the next transaction group begins. 1329 */ 1330 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) 1331 goto retry;; 1332 1333 trim_thread_wakeup(spa); 1334 1335 return (0); 1336} 1337