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, 2018 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/metaslab_impl.h> 147 #include <sys/zio.h> 148 #include <sys/dsl_scan.h> 149 #include <sys/abd.h> 150 #include <sys/fs/zfs.h> 151 152 /* 153 * Basic routines to read and write from a vdev label. 154 * Used throughout the rest of this file. 155 */ 156 uint64_t 157 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 158 { 159 ASSERT(offset < sizeof (vdev_label_t)); 160 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 161 162 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 163 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 164 } 165 166 /* 167 * Returns back the vdev label associated with the passed in offset. 168 */ 169 int 170 vdev_label_number(uint64_t psize, uint64_t offset) 171 { 172 int l; 173 174 if (offset >= psize - VDEV_LABEL_END_SIZE) { 175 offset -= psize - VDEV_LABEL_END_SIZE; 176 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 177 } 178 l = offset / sizeof (vdev_label_t); 179 return (l < VDEV_LABELS ? l : -1); 180 } 181 182 static void 183 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 184 uint64_t size, zio_done_func_t *done, void *private, int flags) 185 { 186 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 187 SCL_STATE_ALL); 188 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 189 190 zio_nowait(zio_read_phys(zio, vd, 191 vdev_label_offset(vd->vdev_psize, l, offset), 192 size, buf, ZIO_CHECKSUM_LABEL, done, private, 193 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 194 } 195 196 static void 197 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 198 uint64_t size, zio_done_func_t *done, void *private, int flags) 199 { 200 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 201 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 202 (SCL_CONFIG | SCL_STATE) && 203 dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 204 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 205 206 zio_nowait(zio_write_phys(zio, vd, 207 vdev_label_offset(vd->vdev_psize, l, offset), 208 size, buf, ZIO_CHECKSUM_LABEL, done, private, 209 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 210 } 211 212 static void 213 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 214 { 215 spa_t *spa = vd->vdev_spa; 216 217 if (vd != spa->spa_root_vdev) 218 return; 219 220 /* provide either current or previous scan information */ 221 pool_scan_stat_t ps; 222 if (spa_scan_get_stats(spa, &ps) == 0) { 223 fnvlist_add_uint64_array(nvl, 224 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 225 sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 226 } 227 228 pool_removal_stat_t prs; 229 if (spa_removal_get_stats(spa, &prs) == 0) { 230 fnvlist_add_uint64_array(nvl, 231 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs, 232 sizeof (prs) / sizeof (uint64_t)); 233 } 234 235 pool_checkpoint_stat_t pcs; 236 if (spa_checkpoint_get_stats(spa, &pcs) == 0) { 237 fnvlist_add_uint64_array(nvl, 238 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs, 239 sizeof (pcs) / sizeof (uint64_t)); 240 } 241 } 242 243 /* 244 * Generate the nvlist representing this vdev's config. 245 */ 246 nvlist_t * 247 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 248 vdev_config_flag_t flags) 249 { 250 nvlist_t *nv = NULL; 251 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 252 253 nv = fnvlist_alloc(); 254 255 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 256 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 257 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 258 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 259 260 if (vd->vdev_path != NULL) 261 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 262 263 if (vd->vdev_devid != NULL) 264 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 265 266 if (vd->vdev_physpath != NULL) 267 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 268 vd->vdev_physpath); 269 270 if (vd->vdev_fru != NULL) 271 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 272 273 if (vd->vdev_nparity != 0) { 274 ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 275 VDEV_TYPE_RAIDZ) == 0); 276 277 /* 278 * Make sure someone hasn't managed to sneak a fancy new vdev 279 * into a crufty old storage pool. 280 */ 281 ASSERT(vd->vdev_nparity == 1 || 282 (vd->vdev_nparity <= 2 && 283 spa_version(spa) >= SPA_VERSION_RAIDZ2) || 284 (vd->vdev_nparity <= 3 && 285 spa_version(spa) >= SPA_VERSION_RAIDZ3)); 286 287 /* 288 * Note that we'll add the nparity tag even on storage pools 289 * that only support a single parity device -- older software 290 * will just ignore it. 291 */ 292 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 293 } 294 295 if (vd->vdev_wholedisk != -1ULL) 296 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 297 vd->vdev_wholedisk); 298 299 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING)) 300 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 301 302 if (vd->vdev_isspare) 303 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 304 305 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 306 vd == vd->vdev_top) { 307 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 308 vd->vdev_ms_array); 309 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 310 vd->vdev_ms_shift); 311 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 312 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 313 vd->vdev_asize); 314 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 315 if (vd->vdev_removing) { 316 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 317 vd->vdev_removing); 318 } 319 } 320 321 if (vd->vdev_dtl_sm != NULL) { 322 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 323 space_map_object(vd->vdev_dtl_sm)); 324 } 325 326 if (vic->vic_mapping_object != 0) { 327 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 328 vic->vic_mapping_object); 329 } 330 331 if (vic->vic_births_object != 0) { 332 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 333 vic->vic_births_object); 334 } 335 336 if (vic->vic_prev_indirect_vdev != UINT64_MAX) { 337 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 338 vic->vic_prev_indirect_vdev); 339 } 340 341 if (vd->vdev_crtxg) 342 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 343 344 if (flags & VDEV_CONFIG_MOS) { 345 if (vd->vdev_leaf_zap != 0) { 346 ASSERT(vd->vdev_ops->vdev_op_leaf); 347 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 348 vd->vdev_leaf_zap); 349 } 350 351 if (vd->vdev_top_zap != 0) { 352 ASSERT(vd == vd->vdev_top); 353 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 354 vd->vdev_top_zap); 355 } 356 } 357 358 if (getstats) { 359 vdev_stat_t vs; 360 361 vdev_get_stats(vd, &vs); 362 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 363 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 364 365 root_vdev_actions_getprogress(vd, nv); 366 367 /* 368 * Note: this can be called from open context 369 * (spa_get_stats()), so we need the rwlock to prevent 370 * the mapping from being changed by condensing. 371 */ 372 rw_enter(&vd->vdev_indirect_rwlock, RW_READER); 373 if (vd->vdev_indirect_mapping != NULL) { 374 ASSERT(vd->vdev_indirect_births != NULL); 375 vdev_indirect_mapping_t *vim = 376 vd->vdev_indirect_mapping; 377 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 378 vdev_indirect_mapping_size(vim)); 379 } 380 rw_exit(&vd->vdev_indirect_rwlock); 381 if (vd->vdev_mg != NULL && 382 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) { 383 /* 384 * Compute approximately how much memory would be used 385 * for the indirect mapping if this device were to 386 * be removed. 387 * 388 * Note: If the frag metric is invalid, then not 389 * enough metaslabs have been converted to have 390 * histograms. 391 */ 392 uint64_t seg_count = 0; 393 394 /* 395 * There are the same number of allocated segments 396 * as free segments, so we will have at least one 397 * entry per free segment. 398 */ 399 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 400 seg_count += vd->vdev_mg->mg_histogram[i]; 401 } 402 403 /* 404 * The maximum length of a mapping is SPA_MAXBLOCKSIZE, 405 * so we need at least one entry per SPA_MAXBLOCKSIZE 406 * of allocated data. 407 */ 408 seg_count += vd->vdev_stat.vs_alloc / SPA_MAXBLOCKSIZE; 409 410 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 411 seg_count * 412 sizeof (vdev_indirect_mapping_entry_phys_t)); 413 } 414 } 415 416 if (!vd->vdev_ops->vdev_op_leaf) { 417 nvlist_t **child; 418 int c, idx; 419 420 ASSERT(!vd->vdev_ishole); 421 422 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 423 KM_SLEEP); 424 425 for (c = 0, idx = 0; c < vd->vdev_children; c++) { 426 vdev_t *cvd = vd->vdev_child[c]; 427 428 /* 429 * If we're generating an nvlist of removing 430 * vdevs then skip over any device which is 431 * not being removed. 432 */ 433 if ((flags & VDEV_CONFIG_REMOVING) && 434 !cvd->vdev_removing) 435 continue; 436 437 child[idx++] = vdev_config_generate(spa, cvd, 438 getstats, flags); 439 } 440 441 if (idx) { 442 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 443 child, idx); 444 } 445 446 for (c = 0; c < idx; c++) 447 nvlist_free(child[c]); 448 449 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 450 451 } else { 452 const char *aux = NULL; 453 454 if (vd->vdev_offline && !vd->vdev_tmpoffline) 455 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 456 if (vd->vdev_resilver_txg != 0) 457 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 458 vd->vdev_resilver_txg); 459 if (vd->vdev_faulted) 460 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 461 if (vd->vdev_degraded) 462 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 463 if (vd->vdev_removed) 464 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 465 if (vd->vdev_unspare) 466 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 467 if (vd->vdev_ishole) 468 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 469 470 switch (vd->vdev_stat.vs_aux) { 471 case VDEV_AUX_ERR_EXCEEDED: 472 aux = "err_exceeded"; 473 break; 474 475 case VDEV_AUX_EXTERNAL: 476 aux = "external"; 477 break; 478 } 479 480 if (aux != NULL) 481 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 482 483 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 484 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 485 vd->vdev_orig_guid); 486 } 487 } 488 489 return (nv); 490 } 491 492 /* 493 * Generate a view of the top-level vdevs. If we currently have holes 494 * in the namespace, then generate an array which contains a list of holey 495 * vdevs. Additionally, add the number of top-level children that currently 496 * exist. 497 */ 498 void 499 vdev_top_config_generate(spa_t *spa, nvlist_t *config) 500 { 501 vdev_t *rvd = spa->spa_root_vdev; 502 uint64_t *array; 503 uint_t c, idx; 504 505 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 506 507 for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 508 vdev_t *tvd = rvd->vdev_child[c]; 509 510 if (tvd->vdev_ishole) { 511 array[idx++] = c; 512 } 513 } 514 515 if (idx) { 516 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 517 array, idx) == 0); 518 } 519 520 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 521 rvd->vdev_children) == 0); 522 523 kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 524 } 525 526 /* 527 * Returns the configuration from the label of the given vdev. For vdevs 528 * which don't have a txg value stored on their label (i.e. spares/cache) 529 * or have not been completely initialized (txg = 0) just return 530 * the configuration from the first valid label we find. Otherwise, 531 * find the most up-to-date label that does not exceed the specified 532 * 'txg' value. 533 */ 534 nvlist_t * 535 vdev_label_read_config(vdev_t *vd, uint64_t txg) 536 { 537 spa_t *spa = vd->vdev_spa; 538 nvlist_t *config = NULL; 539 vdev_phys_t *vp; 540 abd_t *vp_abd; 541 zio_t *zio; 542 uint64_t best_txg = 0; 543 int error = 0; 544 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 545 ZIO_FLAG_SPECULATIVE; 546 547 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 548 549 if (!vdev_readable(vd)) 550 return (NULL); 551 552 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 553 vp = abd_to_buf(vp_abd); 554 555 retry: 556 for (int l = 0; l < VDEV_LABELS; l++) { 557 nvlist_t *label = NULL; 558 559 zio = zio_root(spa, NULL, NULL, flags); 560 561 vdev_label_read(zio, vd, l, vp_abd, 562 offsetof(vdev_label_t, vl_vdev_phys), 563 sizeof (vdev_phys_t), NULL, NULL, flags); 564 565 if (zio_wait(zio) == 0 && 566 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 567 &label, 0) == 0) { 568 uint64_t label_txg = 0; 569 570 /* 571 * Auxiliary vdevs won't have txg values in their 572 * labels and newly added vdevs may not have been 573 * completely initialized so just return the 574 * configuration from the first valid label we 575 * encounter. 576 */ 577 error = nvlist_lookup_uint64(label, 578 ZPOOL_CONFIG_POOL_TXG, &label_txg); 579 if ((error || label_txg == 0) && !config) { 580 config = label; 581 break; 582 } else if (label_txg <= txg && label_txg > best_txg) { 583 best_txg = label_txg; 584 nvlist_free(config); 585 config = fnvlist_dup(label); 586 } 587 } 588 589 if (label != NULL) { 590 nvlist_free(label); 591 label = NULL; 592 } 593 } 594 595 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 596 flags |= ZIO_FLAG_TRYHARD; 597 goto retry; 598 } 599 600 abd_free(vp_abd); 601 602 return (config); 603 } 604 605 /* 606 * Determine if a device is in use. The 'spare_guid' parameter will be filled 607 * in with the device guid if this spare is active elsewhere on the system. 608 */ 609 static boolean_t 610 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 611 uint64_t *spare_guid, uint64_t *l2cache_guid) 612 { 613 spa_t *spa = vd->vdev_spa; 614 uint64_t state, pool_guid, device_guid, txg, spare_pool; 615 uint64_t vdtxg = 0; 616 nvlist_t *label; 617 618 if (spare_guid) 619 *spare_guid = 0ULL; 620 if (l2cache_guid) 621 *l2cache_guid = 0ULL; 622 623 /* 624 * Read the label, if any, and perform some basic sanity checks. 625 */ 626 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 627 return (B_FALSE); 628 629 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 630 &vdtxg); 631 632 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 633 &state) != 0 || 634 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 635 &device_guid) != 0) { 636 nvlist_free(label); 637 return (B_FALSE); 638 } 639 640 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 641 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 642 &pool_guid) != 0 || 643 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 644 &txg) != 0)) { 645 nvlist_free(label); 646 return (B_FALSE); 647 } 648 649 nvlist_free(label); 650 651 /* 652 * Check to see if this device indeed belongs to the pool it claims to 653 * be a part of. The only way this is allowed is if the device is a hot 654 * spare (which we check for later on). 655 */ 656 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 657 !spa_guid_exists(pool_guid, device_guid) && 658 !spa_spare_exists(device_guid, NULL, NULL) && 659 !spa_l2cache_exists(device_guid, NULL)) 660 return (B_FALSE); 661 662 /* 663 * If the transaction group is zero, then this an initialized (but 664 * unused) label. This is only an error if the create transaction 665 * on-disk is the same as the one we're using now, in which case the 666 * user has attempted to add the same vdev multiple times in the same 667 * transaction. 668 */ 669 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 670 txg == 0 && vdtxg == crtxg) 671 return (B_TRUE); 672 673 /* 674 * Check to see if this is a spare device. We do an explicit check for 675 * spa_has_spare() here because it may be on our pending list of spares 676 * to add. We also check if it is an l2cache device. 677 */ 678 if (spa_spare_exists(device_guid, &spare_pool, NULL) || 679 spa_has_spare(spa, device_guid)) { 680 if (spare_guid) 681 *spare_guid = device_guid; 682 683 switch (reason) { 684 case VDEV_LABEL_CREATE: 685 case VDEV_LABEL_L2CACHE: 686 return (B_TRUE); 687 688 case VDEV_LABEL_REPLACE: 689 return (!spa_has_spare(spa, device_guid) || 690 spare_pool != 0ULL); 691 692 case VDEV_LABEL_SPARE: 693 return (spa_has_spare(spa, device_guid)); 694 } 695 } 696 697 /* 698 * Check to see if this is an l2cache device. 699 */ 700 if (spa_l2cache_exists(device_guid, NULL)) 701 return (B_TRUE); 702 703 /* 704 * We can't rely on a pool's state if it's been imported 705 * read-only. Instead we look to see if the pools is marked 706 * read-only in the namespace and set the state to active. 707 */ 708 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 709 (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 710 spa_mode(spa) == FREAD) 711 state = POOL_STATE_ACTIVE; 712 713 /* 714 * If the device is marked ACTIVE, then this device is in use by another 715 * pool on the system. 716 */ 717 return (state == POOL_STATE_ACTIVE); 718 } 719 720 /* 721 * Initialize a vdev label. We check to make sure each leaf device is not in 722 * use, and writable. We put down an initial label which we will later 723 * overwrite with a complete label. Note that it's important to do this 724 * sequentially, not in parallel, so that we catch cases of multiple use of the 725 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 726 * itself. 727 */ 728 int 729 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 730 { 731 spa_t *spa = vd->vdev_spa; 732 nvlist_t *label; 733 vdev_phys_t *vp; 734 abd_t *vp_abd; 735 abd_t *pad2; 736 uberblock_t *ub; 737 abd_t *ub_abd; 738 zio_t *zio; 739 char *buf; 740 size_t buflen; 741 int error; 742 uint64_t spare_guid, l2cache_guid; 743 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 744 745 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 746 747 for (int c = 0; c < vd->vdev_children; c++) 748 if ((error = vdev_label_init(vd->vdev_child[c], 749 crtxg, reason)) != 0) 750 return (error); 751 752 /* Track the creation time for this vdev */ 753 vd->vdev_crtxg = crtxg; 754 755 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 756 return (0); 757 758 /* 759 * Dead vdevs cannot be initialized. 760 */ 761 if (vdev_is_dead(vd)) 762 return (SET_ERROR(EIO)); 763 764 /* 765 * Determine if the vdev is in use. 766 */ 767 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 768 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 769 return (SET_ERROR(EBUSY)); 770 771 /* 772 * If this is a request to add or replace a spare or l2cache device 773 * that is in use elsewhere on the system, then we must update the 774 * guid (which was initialized to a random value) to reflect the 775 * actual GUID (which is shared between multiple pools). 776 */ 777 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 778 spare_guid != 0ULL) { 779 uint64_t guid_delta = spare_guid - vd->vdev_guid; 780 781 vd->vdev_guid += guid_delta; 782 783 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 784 pvd->vdev_guid_sum += guid_delta; 785 786 /* 787 * If this is a replacement, then we want to fallthrough to the 788 * rest of the code. If we're adding a spare, then it's already 789 * labeled appropriately and we can just return. 790 */ 791 if (reason == VDEV_LABEL_SPARE) 792 return (0); 793 ASSERT(reason == VDEV_LABEL_REPLACE || 794 reason == VDEV_LABEL_SPLIT); 795 } 796 797 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 798 l2cache_guid != 0ULL) { 799 uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 800 801 vd->vdev_guid += guid_delta; 802 803 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 804 pvd->vdev_guid_sum += guid_delta; 805 806 /* 807 * If this is a replacement, then we want to fallthrough to the 808 * rest of the code. If we're adding an l2cache, then it's 809 * already labeled appropriately and we can just return. 810 */ 811 if (reason == VDEV_LABEL_L2CACHE) 812 return (0); 813 ASSERT(reason == VDEV_LABEL_REPLACE); 814 } 815 816 /* 817 * Initialize its label. 818 */ 819 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 820 abd_zero(vp_abd, sizeof (vdev_phys_t)); 821 vp = abd_to_buf(vp_abd); 822 823 /* 824 * Generate a label describing the pool and our top-level vdev. 825 * We mark it as being from txg 0 to indicate that it's not 826 * really part of an active pool just yet. The labels will 827 * be written again with a meaningful txg by spa_sync(). 828 */ 829 if (reason == VDEV_LABEL_SPARE || 830 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 831 /* 832 * For inactive hot spares, we generate a special label that 833 * identifies as a mutually shared hot spare. We write the 834 * label if we are adding a hot spare, or if we are removing an 835 * active hot spare (in which case we want to revert the 836 * labels). 837 */ 838 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 839 840 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 841 spa_version(spa)) == 0); 842 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 843 POOL_STATE_SPARE) == 0); 844 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 845 vd->vdev_guid) == 0); 846 } else if (reason == VDEV_LABEL_L2CACHE || 847 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 848 /* 849 * For level 2 ARC devices, add a special label. 850 */ 851 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 852 853 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 854 spa_version(spa)) == 0); 855 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 856 POOL_STATE_L2CACHE) == 0); 857 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 858 vd->vdev_guid) == 0); 859 } else { 860 uint64_t txg = 0ULL; 861 862 if (reason == VDEV_LABEL_SPLIT) 863 txg = spa->spa_uberblock.ub_txg; 864 label = spa_config_generate(spa, vd, txg, B_FALSE); 865 866 /* 867 * Add our creation time. This allows us to detect multiple 868 * vdev uses as described above, and automatically expires if we 869 * fail. 870 */ 871 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 872 crtxg) == 0); 873 } 874 875 buf = vp->vp_nvlist; 876 buflen = sizeof (vp->vp_nvlist); 877 878 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 879 if (error != 0) { 880 nvlist_free(label); 881 abd_free(vp_abd); 882 /* EFAULT means nvlist_pack ran out of room */ 883 return (error == EFAULT ? ENAMETOOLONG : EINVAL); 884 } 885 886 /* 887 * Initialize uberblock template. 888 */ 889 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE); 890 abd_zero(ub_abd, VDEV_UBERBLOCK_RING); 891 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t)); 892 ub = abd_to_buf(ub_abd); 893 ub->ub_txg = 0; 894 895 /* Initialize the 2nd padding area. */ 896 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 897 abd_zero(pad2, VDEV_PAD_SIZE); 898 899 /* 900 * Write everything in parallel. 901 */ 902 retry: 903 zio = zio_root(spa, NULL, NULL, flags); 904 905 for (int l = 0; l < VDEV_LABELS; l++) { 906 907 vdev_label_write(zio, vd, l, vp_abd, 908 offsetof(vdev_label_t, vl_vdev_phys), 909 sizeof (vdev_phys_t), NULL, NULL, flags); 910 911 /* 912 * Skip the 1st padding area. 913 * Zero out the 2nd padding area where it might have 914 * left over data from previous filesystem format. 915 */ 916 vdev_label_write(zio, vd, l, pad2, 917 offsetof(vdev_label_t, vl_pad2), 918 VDEV_PAD_SIZE, NULL, NULL, flags); 919 920 vdev_label_write(zio, vd, l, ub_abd, 921 offsetof(vdev_label_t, vl_uberblock), 922 VDEV_UBERBLOCK_RING, NULL, NULL, flags); 923 } 924 925 error = zio_wait(zio); 926 927 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 928 flags |= ZIO_FLAG_TRYHARD; 929 goto retry; 930 } 931 932 nvlist_free(label); 933 abd_free(pad2); 934 abd_free(ub_abd); 935 abd_free(vp_abd); 936 937 /* 938 * If this vdev hasn't been previously identified as a spare, then we 939 * mark it as such only if a) we are labeling it as a spare, or b) it 940 * exists as a spare elsewhere in the system. Do the same for 941 * level 2 ARC devices. 942 */ 943 if (error == 0 && !vd->vdev_isspare && 944 (reason == VDEV_LABEL_SPARE || 945 spa_spare_exists(vd->vdev_guid, NULL, NULL))) 946 spa_spare_add(vd); 947 948 if (error == 0 && !vd->vdev_isl2cache && 949 (reason == VDEV_LABEL_L2CACHE || 950 spa_l2cache_exists(vd->vdev_guid, NULL))) 951 spa_l2cache_add(vd); 952 953 return (error); 954 } 955 956 /* 957 * ========================================================================== 958 * uberblock load/sync 959 * ========================================================================== 960 */ 961 962 /* 963 * Consider the following situation: txg is safely synced to disk. We've 964 * written the first uberblock for txg + 1, and then we lose power. When we 965 * come back up, we fail to see the uberblock for txg + 1 because, say, 966 * it was on a mirrored device and the replica to which we wrote txg + 1 967 * is now offline. If we then make some changes and sync txg + 1, and then 968 * the missing replica comes back, then for a few seconds we'll have two 969 * conflicting uberblocks on disk with the same txg. The solution is simple: 970 * among uberblocks with equal txg, choose the one with the latest timestamp. 971 */ 972 static int 973 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 974 { 975 if (ub1->ub_txg < ub2->ub_txg) 976 return (-1); 977 if (ub1->ub_txg > ub2->ub_txg) 978 return (1); 979 980 if (ub1->ub_timestamp < ub2->ub_timestamp) 981 return (-1); 982 if (ub1->ub_timestamp > ub2->ub_timestamp) 983 return (1); 984 985 return (0); 986 } 987 988 struct ubl_cbdata { 989 uberblock_t *ubl_ubbest; /* Best uberblock */ 990 vdev_t *ubl_vd; /* vdev associated with the above */ 991 }; 992 993 static void 994 vdev_uberblock_load_done(zio_t *zio) 995 { 996 vdev_t *vd = zio->io_vd; 997 spa_t *spa = zio->io_spa; 998 zio_t *rio = zio->io_private; 999 uberblock_t *ub = abd_to_buf(zio->io_abd); 1000 struct ubl_cbdata *cbp = rio->io_private; 1001 1002 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 1003 1004 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 1005 mutex_enter(&rio->io_lock); 1006 if (ub->ub_txg <= spa->spa_load_max_txg && 1007 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 1008 /* 1009 * Keep track of the vdev in which this uberblock 1010 * was found. We will use this information later 1011 * to obtain the config nvlist associated with 1012 * this uberblock. 1013 */ 1014 *cbp->ubl_ubbest = *ub; 1015 cbp->ubl_vd = vd; 1016 } 1017 mutex_exit(&rio->io_lock); 1018 } 1019 1020 abd_free(zio->io_abd); 1021 } 1022 1023 static void 1024 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 1025 struct ubl_cbdata *cbp) 1026 { 1027 for (int c = 0; c < vd->vdev_children; c++) 1028 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 1029 1030 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1031 for (int l = 0; l < VDEV_LABELS; l++) { 1032 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1033 vdev_label_read(zio, vd, l, 1034 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), 1035 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n), 1036 VDEV_UBERBLOCK_SIZE(vd), 1037 vdev_uberblock_load_done, zio, flags); 1038 } 1039 } 1040 } 1041 } 1042 1043 /* 1044 * Reads the 'best' uberblock from disk along with its associated 1045 * configuration. First, we read the uberblock array of each label of each 1046 * vdev, keeping track of the uberblock with the highest txg in each array. 1047 * Then, we read the configuration from the same vdev as the best uberblock. 1048 */ 1049 void 1050 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 1051 { 1052 zio_t *zio; 1053 spa_t *spa = rvd->vdev_spa; 1054 struct ubl_cbdata cb; 1055 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1056 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1057 1058 ASSERT(ub); 1059 ASSERT(config); 1060 1061 bzero(ub, sizeof (uberblock_t)); 1062 *config = NULL; 1063 1064 cb.ubl_ubbest = ub; 1065 cb.ubl_vd = NULL; 1066 1067 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1068 zio = zio_root(spa, NULL, &cb, flags); 1069 vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1070 (void) zio_wait(zio); 1071 1072 /* 1073 * It's possible that the best uberblock was discovered on a label 1074 * that has a configuration which was written in a future txg. 1075 * Search all labels on this vdev to find the configuration that 1076 * matches the txg for our uberblock. 1077 */ 1078 if (cb.ubl_vd != NULL) { 1079 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. " 1080 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg); 1081 1082 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 1083 if (*config == NULL && spa->spa_extreme_rewind) { 1084 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. " 1085 "Trying again without txg restrictions."); 1086 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX); 1087 } 1088 if (*config == NULL) { 1089 vdev_dbgmsg(cb.ubl_vd, "failed to read label config"); 1090 } 1091 } 1092 spa_config_exit(spa, SCL_ALL, FTAG); 1093 } 1094 1095 /* 1096 * On success, increment root zio's count of good writes. 1097 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1098 */ 1099 static void 1100 vdev_uberblock_sync_done(zio_t *zio) 1101 { 1102 uint64_t *good_writes = zio->io_private; 1103 1104 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1105 atomic_inc_64(good_writes); 1106 } 1107 1108 /* 1109 * Write the uberblock to all labels of all leaves of the specified vdev. 1110 */ 1111 static void 1112 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 1113 { 1114 for (uint64_t c = 0; c < vd->vdev_children; c++) 1115 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 1116 1117 if (!vd->vdev_ops->vdev_op_leaf) 1118 return; 1119 1120 if (!vdev_writeable(vd)) 1121 return; 1122 1123 int n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 1124 1125 /* Copy the uberblock_t into the ABD */ 1126 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1127 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1128 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t)); 1129 1130 for (int l = 0; l < VDEV_LABELS; l++) 1131 vdev_label_write(zio, vd, l, ub_abd, 1132 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1133 vdev_uberblock_sync_done, zio->io_private, 1134 flags | ZIO_FLAG_DONT_PROPAGATE); 1135 1136 abd_free(ub_abd); 1137 } 1138 1139 /* Sync the uberblocks to all vdevs in svd[] */ 1140 int 1141 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1142 { 1143 spa_t *spa = svd[0]->vdev_spa; 1144 zio_t *zio; 1145 uint64_t good_writes = 0; 1146 1147 zio = zio_root(spa, NULL, &good_writes, flags); 1148 1149 for (int v = 0; v < svdcount; v++) 1150 vdev_uberblock_sync(zio, ub, svd[v], flags); 1151 1152 (void) zio_wait(zio); 1153 1154 /* 1155 * Flush the uberblocks to disk. This ensures that the odd labels 1156 * are no longer needed (because the new uberblocks and the even 1157 * labels are safely on disk), so it is safe to overwrite them. 1158 */ 1159 zio = zio_root(spa, NULL, NULL, flags); 1160 1161 for (int v = 0; v < svdcount; v++) { 1162 if (vdev_writeable(svd[v])) { 1163 zio_flush(zio, svd[v]); 1164 } 1165 } 1166 1167 (void) zio_wait(zio); 1168 1169 return (good_writes >= 1 ? 0 : EIO); 1170 } 1171 1172 /* 1173 * On success, increment the count of good writes for our top-level vdev. 1174 */ 1175 static void 1176 vdev_label_sync_done(zio_t *zio) 1177 { 1178 uint64_t *good_writes = zio->io_private; 1179 1180 if (zio->io_error == 0) 1181 atomic_inc_64(good_writes); 1182 } 1183 1184 /* 1185 * If there weren't enough good writes, indicate failure to the parent. 1186 */ 1187 static void 1188 vdev_label_sync_top_done(zio_t *zio) 1189 { 1190 uint64_t *good_writes = zio->io_private; 1191 1192 if (*good_writes == 0) 1193 zio->io_error = SET_ERROR(EIO); 1194 1195 kmem_free(good_writes, sizeof (uint64_t)); 1196 } 1197 1198 /* 1199 * We ignore errors for log and cache devices, simply free the private data. 1200 */ 1201 static void 1202 vdev_label_sync_ignore_done(zio_t *zio) 1203 { 1204 kmem_free(zio->io_private, sizeof (uint64_t)); 1205 } 1206 1207 /* 1208 * Write all even or odd labels to all leaves of the specified vdev. 1209 */ 1210 static void 1211 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1212 { 1213 nvlist_t *label; 1214 vdev_phys_t *vp; 1215 abd_t *vp_abd; 1216 char *buf; 1217 size_t buflen; 1218 1219 for (int c = 0; c < vd->vdev_children; c++) 1220 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1221 1222 if (!vd->vdev_ops->vdev_op_leaf) 1223 return; 1224 1225 if (!vdev_writeable(vd)) 1226 return; 1227 1228 /* 1229 * Generate a label describing the top-level config to which we belong. 1230 */ 1231 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1232 1233 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1234 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1235 vp = abd_to_buf(vp_abd); 1236 1237 buf = vp->vp_nvlist; 1238 buflen = sizeof (vp->vp_nvlist); 1239 1240 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1241 for (; l < VDEV_LABELS; l += 2) { 1242 vdev_label_write(zio, vd, l, vp_abd, 1243 offsetof(vdev_label_t, vl_vdev_phys), 1244 sizeof (vdev_phys_t), 1245 vdev_label_sync_done, zio->io_private, 1246 flags | ZIO_FLAG_DONT_PROPAGATE); 1247 } 1248 } 1249 1250 abd_free(vp_abd); 1251 nvlist_free(label); 1252 } 1253 1254 int 1255 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1256 { 1257 list_t *dl = &spa->spa_config_dirty_list; 1258 vdev_t *vd; 1259 zio_t *zio; 1260 int error; 1261 1262 /* 1263 * Write the new labels to disk. 1264 */ 1265 zio = zio_root(spa, NULL, NULL, flags); 1266 1267 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1268 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1269 KM_SLEEP); 1270 1271 ASSERT(!vd->vdev_ishole); 1272 1273 zio_t *vio = zio_null(zio, spa, NULL, 1274 (vd->vdev_islog || vd->vdev_aux != NULL) ? 1275 vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1276 good_writes, flags); 1277 vdev_label_sync(vio, vd, l, txg, flags); 1278 zio_nowait(vio); 1279 } 1280 1281 error = zio_wait(zio); 1282 1283 /* 1284 * Flush the new labels to disk. 1285 */ 1286 zio = zio_root(spa, NULL, NULL, flags); 1287 1288 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1289 zio_flush(zio, vd); 1290 1291 (void) zio_wait(zio); 1292 1293 return (error); 1294 } 1295 1296 /* 1297 * Sync the uberblock and any changes to the vdev configuration. 1298 * 1299 * The order of operations is carefully crafted to ensure that 1300 * if the system panics or loses power at any time, the state on disk 1301 * is still transactionally consistent. The in-line comments below 1302 * describe the failure semantics at each stage. 1303 * 1304 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1305 * at any time, you can just call it again, and it will resume its work. 1306 */ 1307 int 1308 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1309 { 1310 spa_t *spa = svd[0]->vdev_spa; 1311 uberblock_t *ub = &spa->spa_uberblock; 1312 int error = 0; 1313 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1314 1315 ASSERT(svdcount != 0); 1316 retry: 1317 /* 1318 * Normally, we don't want to try too hard to write every label and 1319 * uberblock. If there is a flaky disk, we don't want the rest of the 1320 * sync process to block while we retry. But if we can't write a 1321 * single label out, we should retry with ZIO_FLAG_TRYHARD before 1322 * bailing out and declaring the pool faulted. 1323 */ 1324 if (error != 0) { 1325 if ((flags & ZIO_FLAG_TRYHARD) != 0) 1326 return (error); 1327 flags |= ZIO_FLAG_TRYHARD; 1328 } 1329 1330 ASSERT(ub->ub_txg <= txg); 1331 1332 /* 1333 * If this isn't a resync due to I/O errors, 1334 * and nothing changed in this transaction group, 1335 * and the vdev configuration hasn't changed, 1336 * then there's nothing to do. 1337 */ 1338 if (ub->ub_txg < txg && 1339 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1340 list_is_empty(&spa->spa_config_dirty_list)) 1341 return (0); 1342 1343 if (txg > spa_freeze_txg(spa)) 1344 return (0); 1345 1346 ASSERT(txg <= spa->spa_final_txg); 1347 1348 /* 1349 * Flush the write cache of every disk that's been written to 1350 * in this transaction group. This ensures that all blocks 1351 * written in this txg will be committed to stable storage 1352 * before any uberblock that references them. 1353 */ 1354 zio_t *zio = zio_root(spa, NULL, NULL, flags); 1355 1356 for (vdev_t *vd = 1357 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL; 1358 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1359 zio_flush(zio, vd); 1360 1361 (void) zio_wait(zio); 1362 1363 /* 1364 * Sync out the even labels (L0, L2) for every dirty vdev. If the 1365 * system dies in the middle of this process, that's OK: all of the 1366 * even labels that made it to disk will be newer than any uberblock, 1367 * and will therefore be considered invalid. The odd labels (L1, L3), 1368 * which have not yet been touched, will still be valid. We flush 1369 * the new labels to disk to ensure that all even-label updates 1370 * are committed to stable storage before the uberblock update. 1371 */ 1372 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) { 1373 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1374 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1375 "for pool '%s' when syncing out the even labels " 1376 "of dirty vdevs", error, spa_name(spa)); 1377 } 1378 goto retry; 1379 } 1380 1381 /* 1382 * Sync the uberblocks to all vdevs in svd[]. 1383 * If the system dies in the middle of this step, there are two cases 1384 * to consider, and the on-disk state is consistent either way: 1385 * 1386 * (1) If none of the new uberblocks made it to disk, then the 1387 * previous uberblock will be the newest, and the odd labels 1388 * (which had not yet been touched) will be valid with respect 1389 * to that uberblock. 1390 * 1391 * (2) If one or more new uberblocks made it to disk, then they 1392 * will be the newest, and the even labels (which had all 1393 * been successfully committed) will be valid with respect 1394 * to the new uberblocks. 1395 */ 1396 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) { 1397 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1398 zfs_dbgmsg("vdev_uberblock_sync_list() returned error " 1399 "%d for pool '%s'", error, spa_name(spa)); 1400 } 1401 goto retry; 1402 } 1403 1404 /* 1405 * Sync out odd labels for every dirty vdev. If the system dies 1406 * in the middle of this process, the even labels and the new 1407 * uberblocks will suffice to open the pool. The next time 1408 * the pool is opened, the first thing we'll do -- before any 1409 * user data is modified -- is mark every vdev dirty so that 1410 * all labels will be brought up to date. We flush the new labels 1411 * to disk to ensure that all odd-label updates are committed to 1412 * stable storage before the next transaction group begins. 1413 */ 1414 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) { 1415 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1416 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1417 "for pool '%s' when syncing out the odd labels of " 1418 "dirty vdevs", error, spa_name(spa)); 1419 } 1420 goto retry; 1421 } 1422 1423 return (0); 1424 } 1425