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 * Copyright (c) 2017, Intel Corporation. 26 * Copyright 2019 Joyent, Inc. 27 */ 28 29 /* 30 * Virtual Device Labels 31 * --------------------- 32 * 33 * The vdev label serves several distinct purposes: 34 * 35 * 1. Uniquely identify this device as part of a ZFS pool and confirm its 36 * identity within the pool. 37 * 38 * 2. Verify that all the devices given in a configuration are present 39 * within the pool. 40 * 41 * 3. Determine the uberblock for the pool. 42 * 43 * 4. In case of an import operation, determine the configuration of the 44 * toplevel vdev of which it is a part. 45 * 46 * 5. If an import operation cannot find all the devices in the pool, 47 * provide enough information to the administrator to determine which 48 * devices are missing. 49 * 50 * It is important to note that while the kernel is responsible for writing the 51 * label, it only consumes the information in the first three cases. The 52 * latter information is only consumed in userland when determining the 53 * configuration to import a pool. 54 * 55 * 56 * Label Organization 57 * ------------------ 58 * 59 * Before describing the contents of the label, it's important to understand how 60 * the labels are written and updated with respect to the uberblock. 61 * 62 * When the pool configuration is altered, either because it was newly created 63 * or a device was added, we want to update all the labels such that we can deal 64 * with fatal failure at any point. To this end, each disk has two labels which 65 * are updated before and after the uberblock is synced. Assuming we have 66 * labels and an uberblock with the following transaction groups: 67 * 68 * L1 UB L2 69 * +------+ +------+ +------+ 70 * | | | | | | 71 * | t10 | | t10 | | t10 | 72 * | | | | | | 73 * +------+ +------+ +------+ 74 * 75 * In this stable state, the labels and the uberblock were all updated within 76 * the same transaction group (10). Each label is mirrored and checksummed, so 77 * that we can detect when we fail partway through writing the label. 78 * 79 * In order to identify which labels are valid, the labels are written in the 80 * following manner: 81 * 82 * 1. For each vdev, update 'L1' to the new label 83 * 2. Update the uberblock 84 * 3. For each vdev, update 'L2' to the new label 85 * 86 * Given arbitrary failure, we can determine the correct label to use based on 87 * the transaction group. If we fail after updating L1 but before updating the 88 * UB, we will notice that L1's transaction group is greater than the uberblock, 89 * so L2 must be valid. If we fail after writing the uberblock but before 90 * writing L2, we will notice that L2's transaction group is less than L1, and 91 * therefore L1 is valid. 92 * 93 * Another added complexity is that not every label is updated when the config 94 * is synced. If we add a single device, we do not want to have to re-write 95 * every label for every device in the pool. This means that both L1 and L2 may 96 * be older than the pool uberblock, because the necessary information is stored 97 * on another vdev. 98 * 99 * 100 * On-disk Format 101 * -------------- 102 * 103 * The vdev label consists of two distinct parts, and is wrapped within the 104 * vdev_label_t structure. The label includes 8k of padding to permit legacy 105 * VTOC disk labels, but is otherwise ignored. 106 * 107 * The first half of the label is a packed nvlist which contains pool wide 108 * properties, per-vdev properties, and configuration information. It is 109 * described in more detail below. 110 * 111 * The latter half of the label consists of a redundant array of uberblocks. 112 * These uberblocks are updated whenever a transaction group is committed, 113 * or when the configuration is updated. When a pool is loaded, we scan each 114 * vdev for the 'best' uberblock. 115 * 116 * 117 * Configuration Information 118 * ------------------------- 119 * 120 * The nvlist describing the pool and vdev contains the following elements: 121 * 122 * version ZFS on-disk version 123 * name Pool name 124 * state Pool state 125 * txg Transaction group in which this label was written 126 * pool_guid Unique identifier for this pool 127 * vdev_tree An nvlist describing vdev tree. 128 * features_for_read 129 * An nvlist of the features necessary for reading the MOS. 130 * 131 * Each leaf device label also contains the following: 132 * 133 * top_guid Unique ID for top-level vdev in which this is contained 134 * guid Unique ID for the leaf vdev 135 * 136 * The 'vs' configuration follows the format described in 'spa_config.c'. 137 */ 138 139 #include <sys/zfs_context.h> 140 #include <sys/spa.h> 141 #include <sys/spa_impl.h> 142 #include <sys/dmu.h> 143 #include <sys/zap.h> 144 #include <sys/vdev.h> 145 #include <sys/vdev_impl.h> 146 #include <sys/uberblock_impl.h> 147 #include <sys/metaslab.h> 148 #include <sys/metaslab_impl.h> 149 #include <sys/zio.h> 150 #include <sys/dsl_scan.h> 151 #include <sys/abd.h> 152 #include <sys/fs/zfs.h> 153 154 /* 155 * Basic routines to read and write from a vdev label. 156 * Used throughout the rest of this file. 157 */ 158 uint64_t 159 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 160 { 161 ASSERT(offset < sizeof (vdev_label_t)); 162 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 163 164 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 165 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 166 } 167 168 /* 169 * Returns back the vdev label associated with the passed in offset. 170 */ 171 int 172 vdev_label_number(uint64_t psize, uint64_t offset) 173 { 174 int l; 175 176 if (offset >= psize - VDEV_LABEL_END_SIZE) { 177 offset -= psize - VDEV_LABEL_END_SIZE; 178 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 179 } 180 l = offset / sizeof (vdev_label_t); 181 return (l < VDEV_LABELS ? l : -1); 182 } 183 184 static void 185 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 186 uint64_t size, zio_done_func_t *done, void *private, int flags) 187 { 188 ASSERT( 189 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 190 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 191 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 192 193 zio_nowait(zio_read_phys(zio, vd, 194 vdev_label_offset(vd->vdev_psize, l, offset), 195 size, buf, ZIO_CHECKSUM_LABEL, done, private, 196 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 197 } 198 199 void 200 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 201 uint64_t size, zio_done_func_t *done, void *private, int flags) 202 { 203 ASSERT( 204 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 205 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 206 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 207 208 zio_nowait(zio_write_phys(zio, vd, 209 vdev_label_offset(vd->vdev_psize, l, offset), 210 size, buf, ZIO_CHECKSUM_LABEL, done, private, 211 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 212 } 213 214 static void 215 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 216 { 217 spa_t *spa = vd->vdev_spa; 218 219 if (vd != spa->spa_root_vdev) 220 return; 221 222 /* provide either current or previous scan information */ 223 pool_scan_stat_t ps; 224 if (spa_scan_get_stats(spa, &ps) == 0) { 225 fnvlist_add_uint64_array(nvl, 226 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 227 sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 228 } 229 230 pool_removal_stat_t prs; 231 if (spa_removal_get_stats(spa, &prs) == 0) { 232 fnvlist_add_uint64_array(nvl, 233 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs, 234 sizeof (prs) / sizeof (uint64_t)); 235 } 236 237 pool_checkpoint_stat_t pcs; 238 if (spa_checkpoint_get_stats(spa, &pcs) == 0) { 239 fnvlist_add_uint64_array(nvl, 240 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs, 241 sizeof (pcs) / sizeof (uint64_t)); 242 } 243 } 244 245 /* 246 * Generate the nvlist representing this vdev's config. 247 */ 248 nvlist_t * 249 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 250 vdev_config_flag_t flags) 251 { 252 nvlist_t *nv = NULL; 253 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 254 255 nv = fnvlist_alloc(); 256 257 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 258 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 259 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 260 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 261 262 if (vd->vdev_path != NULL) 263 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 264 265 if (vd->vdev_devid != NULL) 266 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 267 268 if (vd->vdev_physpath != NULL) 269 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 270 vd->vdev_physpath); 271 272 if (vd->vdev_fru != NULL) 273 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 274 275 if (vd->vdev_nparity != 0) { 276 ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 277 VDEV_TYPE_RAIDZ) == 0); 278 279 /* 280 * Make sure someone hasn't managed to sneak a fancy new vdev 281 * into a crufty old storage pool. 282 */ 283 ASSERT(vd->vdev_nparity == 1 || 284 (vd->vdev_nparity <= 2 && 285 spa_version(spa) >= SPA_VERSION_RAIDZ2) || 286 (vd->vdev_nparity <= 3 && 287 spa_version(spa) >= SPA_VERSION_RAIDZ3)); 288 289 /* 290 * Note that we'll add the nparity tag even on storage pools 291 * that only support a single parity device -- older software 292 * will just ignore it. 293 */ 294 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 295 } 296 297 if (vd->vdev_wholedisk != -1ULL) 298 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 299 vd->vdev_wholedisk); 300 301 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING)) 302 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 303 304 if (vd->vdev_isspare) 305 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 306 307 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 308 vd == vd->vdev_top) { 309 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 310 vd->vdev_ms_array); 311 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 312 vd->vdev_ms_shift); 313 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 314 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 315 vd->vdev_asize); 316 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 317 if (vd->vdev_removing) { 318 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 319 vd->vdev_removing); 320 } 321 322 /* zpool command expects alloc class data */ 323 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) { 324 const char *bias = NULL; 325 326 switch (vd->vdev_alloc_bias) { 327 case VDEV_BIAS_LOG: 328 bias = VDEV_ALLOC_BIAS_LOG; 329 break; 330 case VDEV_BIAS_SPECIAL: 331 bias = VDEV_ALLOC_BIAS_SPECIAL; 332 break; 333 case VDEV_BIAS_DEDUP: 334 bias = VDEV_ALLOC_BIAS_DEDUP; 335 break; 336 default: 337 ASSERT3U(vd->vdev_alloc_bias, ==, 338 VDEV_BIAS_NONE); 339 } 340 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 341 bias); 342 } 343 } 344 345 if (vd->vdev_dtl_sm != NULL) { 346 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 347 space_map_object(vd->vdev_dtl_sm)); 348 } 349 350 if (vic->vic_mapping_object != 0) { 351 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 352 vic->vic_mapping_object); 353 } 354 355 if (vic->vic_births_object != 0) { 356 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 357 vic->vic_births_object); 358 } 359 360 if (vic->vic_prev_indirect_vdev != UINT64_MAX) { 361 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 362 vic->vic_prev_indirect_vdev); 363 } 364 365 if (vd->vdev_crtxg) 366 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 367 368 if (flags & VDEV_CONFIG_MOS) { 369 if (vd->vdev_leaf_zap != 0) { 370 ASSERT(vd->vdev_ops->vdev_op_leaf); 371 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 372 vd->vdev_leaf_zap); 373 } 374 375 if (vd->vdev_top_zap != 0) { 376 ASSERT(vd == vd->vdev_top); 377 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 378 vd->vdev_top_zap); 379 } 380 } 381 382 if (getstats) { 383 vdev_stat_t vs; 384 385 vdev_get_stats(vd, &vs); 386 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 387 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 388 389 root_vdev_actions_getprogress(vd, nv); 390 391 /* 392 * Note: this can be called from open context 393 * (spa_get_stats()), so we need the rwlock to prevent 394 * the mapping from being changed by condensing. 395 */ 396 rw_enter(&vd->vdev_indirect_rwlock, RW_READER); 397 if (vd->vdev_indirect_mapping != NULL) { 398 ASSERT(vd->vdev_indirect_births != NULL); 399 vdev_indirect_mapping_t *vim = 400 vd->vdev_indirect_mapping; 401 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 402 vdev_indirect_mapping_size(vim)); 403 } 404 rw_exit(&vd->vdev_indirect_rwlock); 405 if (vd->vdev_mg != NULL && 406 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) { 407 /* 408 * Compute approximately how much memory would be used 409 * for the indirect mapping if this device were to 410 * be removed. 411 * 412 * Note: If the frag metric is invalid, then not 413 * enough metaslabs have been converted to have 414 * histograms. 415 */ 416 uint64_t seg_count = 0; 417 uint64_t to_alloc = vd->vdev_stat.vs_alloc; 418 419 /* 420 * There are the same number of allocated segments 421 * as free segments, so we will have at least one 422 * entry per free segment. However, small free 423 * segments (smaller than vdev_removal_max_span) 424 * will be combined with adjacent allocated segments 425 * as a single mapping. 426 */ 427 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 428 if (1ULL << (i + 1) < vdev_removal_max_span) { 429 to_alloc += 430 vd->vdev_mg->mg_histogram[i] << 431 i + 1; 432 } else { 433 seg_count += 434 vd->vdev_mg->mg_histogram[i]; 435 } 436 } 437 438 /* 439 * The maximum length of a mapping is 440 * zfs_remove_max_segment, so we need at least one entry 441 * per zfs_remove_max_segment of allocated data. 442 */ 443 seg_count += to_alloc / zfs_remove_max_segment; 444 445 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 446 seg_count * 447 sizeof (vdev_indirect_mapping_entry_phys_t)); 448 } 449 } 450 451 if (!vd->vdev_ops->vdev_op_leaf) { 452 nvlist_t **child; 453 int c, idx; 454 455 ASSERT(!vd->vdev_ishole); 456 457 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 458 KM_SLEEP); 459 460 for (c = 0, idx = 0; c < vd->vdev_children; c++) { 461 vdev_t *cvd = vd->vdev_child[c]; 462 463 /* 464 * If we're generating an nvlist of removing 465 * vdevs then skip over any device which is 466 * not being removed. 467 */ 468 if ((flags & VDEV_CONFIG_REMOVING) && 469 !cvd->vdev_removing) 470 continue; 471 472 child[idx++] = vdev_config_generate(spa, cvd, 473 getstats, flags); 474 } 475 476 if (idx) { 477 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 478 child, idx); 479 } 480 481 for (c = 0; c < idx; c++) 482 nvlist_free(child[c]); 483 484 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 485 486 } else { 487 const char *aux = NULL; 488 489 if (vd->vdev_offline && !vd->vdev_tmpoffline) 490 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 491 if (vd->vdev_resilver_txg != 0) 492 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 493 vd->vdev_resilver_txg); 494 if (vd->vdev_faulted) 495 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 496 if (vd->vdev_degraded) 497 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 498 if (vd->vdev_removed) 499 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 500 if (vd->vdev_unspare) 501 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 502 if (vd->vdev_ishole) 503 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 504 505 switch (vd->vdev_stat.vs_aux) { 506 case VDEV_AUX_ERR_EXCEEDED: 507 aux = "err_exceeded"; 508 break; 509 510 case VDEV_AUX_EXTERNAL: 511 aux = "external"; 512 break; 513 } 514 515 if (aux != NULL) 516 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 517 518 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 519 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 520 vd->vdev_orig_guid); 521 } 522 } 523 524 return (nv); 525 } 526 527 /* 528 * Generate a view of the top-level vdevs. If we currently have holes 529 * in the namespace, then generate an array which contains a list of holey 530 * vdevs. Additionally, add the number of top-level children that currently 531 * exist. 532 */ 533 void 534 vdev_top_config_generate(spa_t *spa, nvlist_t *config) 535 { 536 vdev_t *rvd = spa->spa_root_vdev; 537 uint64_t *array; 538 uint_t c, idx; 539 540 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 541 542 for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 543 vdev_t *tvd = rvd->vdev_child[c]; 544 545 if (tvd->vdev_ishole) { 546 array[idx++] = c; 547 } 548 } 549 550 if (idx) { 551 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 552 array, idx) == 0); 553 } 554 555 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 556 rvd->vdev_children) == 0); 557 558 kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 559 } 560 561 /* 562 * Returns the configuration from the label of the given vdev. For vdevs 563 * which don't have a txg value stored on their label (i.e. spares/cache) 564 * or have not been completely initialized (txg = 0) just return 565 * the configuration from the first valid label we find. Otherwise, 566 * find the most up-to-date label that does not exceed the specified 567 * 'txg' value. 568 */ 569 nvlist_t * 570 vdev_label_read_config(vdev_t *vd, uint64_t txg) 571 { 572 spa_t *spa = vd->vdev_spa; 573 nvlist_t *config = NULL; 574 vdev_phys_t *vp; 575 abd_t *vp_abd; 576 zio_t *zio; 577 uint64_t best_txg = 0; 578 uint64_t label_txg = 0; 579 int error = 0; 580 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 581 ZIO_FLAG_SPECULATIVE; 582 583 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 584 585 if (!vdev_readable(vd)) 586 return (NULL); 587 588 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 589 vp = abd_to_buf(vp_abd); 590 591 retry: 592 for (int l = 0; l < VDEV_LABELS; l++) { 593 nvlist_t *label = NULL; 594 595 zio = zio_root(spa, NULL, NULL, flags); 596 597 vdev_label_read(zio, vd, l, vp_abd, 598 offsetof(vdev_label_t, vl_vdev_phys), 599 sizeof (vdev_phys_t), NULL, NULL, flags); 600 601 if (zio_wait(zio) == 0 && 602 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 603 &label, 0) == 0) { 604 /* 605 * Auxiliary vdevs won't have txg values in their 606 * labels and newly added vdevs may not have been 607 * completely initialized so just return the 608 * configuration from the first valid label we 609 * encounter. 610 */ 611 error = nvlist_lookup_uint64(label, 612 ZPOOL_CONFIG_POOL_TXG, &label_txg); 613 if ((error || label_txg == 0) && !config) { 614 config = label; 615 break; 616 } else if (label_txg <= txg && label_txg > best_txg) { 617 best_txg = label_txg; 618 nvlist_free(config); 619 config = fnvlist_dup(label); 620 } 621 } 622 623 if (label != NULL) { 624 nvlist_free(label); 625 label = NULL; 626 } 627 } 628 629 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 630 flags |= ZIO_FLAG_TRYHARD; 631 goto retry; 632 } 633 634 /* 635 * We found a valid label but it didn't pass txg restrictions. 636 */ 637 if (config == NULL && label_txg != 0) { 638 vdev_dbgmsg(vd, "label discarded as txg is too large " 639 "(%llu > %llu)", (u_longlong_t)label_txg, 640 (u_longlong_t)txg); 641 } 642 643 abd_free(vp_abd); 644 645 return (config); 646 } 647 648 /* 649 * Determine if a device is in use. The 'spare_guid' parameter will be filled 650 * in with the device guid if this spare is active elsewhere on the system. 651 */ 652 static boolean_t 653 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 654 uint64_t *spare_guid, uint64_t *l2cache_guid) 655 { 656 spa_t *spa = vd->vdev_spa; 657 uint64_t state, pool_guid, device_guid, txg, spare_pool; 658 uint64_t vdtxg = 0; 659 nvlist_t *label; 660 661 if (spare_guid) 662 *spare_guid = 0ULL; 663 if (l2cache_guid) 664 *l2cache_guid = 0ULL; 665 666 /* 667 * Read the label, if any, and perform some basic sanity checks. 668 */ 669 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 670 return (B_FALSE); 671 672 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 673 &vdtxg); 674 675 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 676 &state) != 0 || 677 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 678 &device_guid) != 0) { 679 nvlist_free(label); 680 return (B_FALSE); 681 } 682 683 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 684 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 685 &pool_guid) != 0 || 686 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 687 &txg) != 0)) { 688 nvlist_free(label); 689 return (B_FALSE); 690 } 691 692 nvlist_free(label); 693 694 /* 695 * Check to see if this device indeed belongs to the pool it claims to 696 * be a part of. The only way this is allowed is if the device is a hot 697 * spare (which we check for later on). 698 */ 699 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 700 !spa_guid_exists(pool_guid, device_guid) && 701 !spa_spare_exists(device_guid, NULL, NULL) && 702 !spa_l2cache_exists(device_guid, NULL)) 703 return (B_FALSE); 704 705 /* 706 * If the transaction group is zero, then this an initialized (but 707 * unused) label. This is only an error if the create transaction 708 * on-disk is the same as the one we're using now, in which case the 709 * user has attempted to add the same vdev multiple times in the same 710 * transaction. 711 */ 712 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 713 txg == 0 && vdtxg == crtxg) 714 return (B_TRUE); 715 716 /* 717 * Check to see if this is a spare device. We do an explicit check for 718 * spa_has_spare() here because it may be on our pending list of spares 719 * to add. We also check if it is an l2cache device. 720 */ 721 if (spa_spare_exists(device_guid, &spare_pool, NULL) || 722 spa_has_spare(spa, device_guid)) { 723 if (spare_guid) 724 *spare_guid = device_guid; 725 726 switch (reason) { 727 case VDEV_LABEL_CREATE: 728 case VDEV_LABEL_L2CACHE: 729 return (B_TRUE); 730 731 case VDEV_LABEL_REPLACE: 732 return (!spa_has_spare(spa, device_guid) || 733 spare_pool != 0ULL); 734 735 case VDEV_LABEL_SPARE: 736 return (spa_has_spare(spa, device_guid)); 737 } 738 } 739 740 /* 741 * Check to see if this is an l2cache device. 742 */ 743 if (spa_l2cache_exists(device_guid, NULL)) 744 return (B_TRUE); 745 746 /* 747 * We can't rely on a pool's state if it's been imported 748 * read-only. Instead we look to see if the pools is marked 749 * read-only in the namespace and set the state to active. 750 */ 751 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 752 (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 753 spa_mode(spa) == FREAD) 754 state = POOL_STATE_ACTIVE; 755 756 /* 757 * If the device is marked ACTIVE, then this device is in use by another 758 * pool on the system. 759 */ 760 return (state == POOL_STATE_ACTIVE); 761 } 762 763 /* 764 * Initialize a vdev label. We check to make sure each leaf device is not in 765 * use, and writable. We put down an initial label which we will later 766 * overwrite with a complete label. Note that it's important to do this 767 * sequentially, not in parallel, so that we catch cases of multiple use of the 768 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 769 * itself. 770 */ 771 int 772 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 773 { 774 spa_t *spa = vd->vdev_spa; 775 nvlist_t *label; 776 vdev_phys_t *vp; 777 abd_t *vp_abd; 778 abd_t *pad2; 779 uberblock_t *ub; 780 abd_t *ub_abd; 781 zio_t *zio; 782 char *buf; 783 size_t buflen; 784 int error; 785 uint64_t spare_guid, l2cache_guid; 786 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 787 788 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 789 790 for (int c = 0; c < vd->vdev_children; c++) 791 if ((error = vdev_label_init(vd->vdev_child[c], 792 crtxg, reason)) != 0) 793 return (error); 794 795 /* Track the creation time for this vdev */ 796 vd->vdev_crtxg = crtxg; 797 798 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 799 return (0); 800 801 /* 802 * Dead vdevs cannot be initialized. 803 */ 804 if (vdev_is_dead(vd)) 805 return (SET_ERROR(EIO)); 806 807 /* 808 * Determine if the vdev is in use. 809 */ 810 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 811 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 812 return (SET_ERROR(EBUSY)); 813 814 /* 815 * If this is a request to add or replace a spare or l2cache device 816 * that is in use elsewhere on the system, then we must update the 817 * guid (which was initialized to a random value) to reflect the 818 * actual GUID (which is shared between multiple pools). 819 */ 820 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 821 spare_guid != 0ULL) { 822 uint64_t guid_delta = spare_guid - vd->vdev_guid; 823 824 vd->vdev_guid += guid_delta; 825 826 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 827 pvd->vdev_guid_sum += guid_delta; 828 829 /* 830 * If this is a replacement, then we want to fallthrough to the 831 * rest of the code. If we're adding a spare, then it's already 832 * labeled appropriately and we can just return. 833 */ 834 if (reason == VDEV_LABEL_SPARE) 835 return (0); 836 ASSERT(reason == VDEV_LABEL_REPLACE || 837 reason == VDEV_LABEL_SPLIT); 838 } 839 840 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 841 l2cache_guid != 0ULL) { 842 uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 843 844 vd->vdev_guid += guid_delta; 845 846 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 847 pvd->vdev_guid_sum += guid_delta; 848 849 /* 850 * If this is a replacement, then we want to fallthrough to the 851 * rest of the code. If we're adding an l2cache, then it's 852 * already labeled appropriately and we can just return. 853 */ 854 if (reason == VDEV_LABEL_L2CACHE) 855 return (0); 856 ASSERT(reason == VDEV_LABEL_REPLACE); 857 } 858 859 /* 860 * Initialize its label. 861 */ 862 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 863 abd_zero(vp_abd, sizeof (vdev_phys_t)); 864 vp = abd_to_buf(vp_abd); 865 866 /* 867 * Generate a label describing the pool and our top-level vdev. 868 * We mark it as being from txg 0 to indicate that it's not 869 * really part of an active pool just yet. The labels will 870 * be written again with a meaningful txg by spa_sync(). 871 */ 872 if (reason == VDEV_LABEL_SPARE || 873 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 874 /* 875 * For inactive hot spares, we generate a special label that 876 * identifies as a mutually shared hot spare. We write the 877 * label if we are adding a hot spare, or if we are removing an 878 * active hot spare (in which case we want to revert the 879 * labels). 880 */ 881 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 882 883 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 884 spa_version(spa)) == 0); 885 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 886 POOL_STATE_SPARE) == 0); 887 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 888 vd->vdev_guid) == 0); 889 } else if (reason == VDEV_LABEL_L2CACHE || 890 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 891 /* 892 * For level 2 ARC devices, add a special label. 893 */ 894 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 895 896 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 897 spa_version(spa)) == 0); 898 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 899 POOL_STATE_L2CACHE) == 0); 900 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 901 vd->vdev_guid) == 0); 902 } else { 903 uint64_t txg = 0ULL; 904 905 if (reason == VDEV_LABEL_SPLIT) 906 txg = spa->spa_uberblock.ub_txg; 907 label = spa_config_generate(spa, vd, txg, B_FALSE); 908 909 /* 910 * Add our creation time. This allows us to detect multiple 911 * vdev uses as described above, and automatically expires if we 912 * fail. 913 */ 914 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 915 crtxg) == 0); 916 } 917 918 buf = vp->vp_nvlist; 919 buflen = sizeof (vp->vp_nvlist); 920 921 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 922 if (error != 0) { 923 nvlist_free(label); 924 abd_free(vp_abd); 925 /* EFAULT means nvlist_pack ran out of room */ 926 return (error == EFAULT ? ENAMETOOLONG : EINVAL); 927 } 928 929 /* 930 * Initialize uberblock template. 931 */ 932 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE); 933 abd_zero(ub_abd, VDEV_UBERBLOCK_RING); 934 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t)); 935 ub = abd_to_buf(ub_abd); 936 ub->ub_txg = 0; 937 938 /* Initialize the 2nd padding area. */ 939 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 940 abd_zero(pad2, VDEV_PAD_SIZE); 941 942 /* 943 * Write everything in parallel. 944 */ 945 retry: 946 zio = zio_root(spa, NULL, NULL, flags); 947 948 for (int l = 0; l < VDEV_LABELS; l++) { 949 950 vdev_label_write(zio, vd, l, vp_abd, 951 offsetof(vdev_label_t, vl_vdev_phys), 952 sizeof (vdev_phys_t), NULL, NULL, flags); 953 954 /* 955 * Skip the 1st padding area. 956 * Zero out the 2nd padding area where it might have 957 * left over data from previous filesystem format. 958 */ 959 vdev_label_write(zio, vd, l, pad2, 960 offsetof(vdev_label_t, vl_pad2), 961 VDEV_PAD_SIZE, NULL, NULL, flags); 962 963 vdev_label_write(zio, vd, l, ub_abd, 964 offsetof(vdev_label_t, vl_uberblock), 965 VDEV_UBERBLOCK_RING, NULL, NULL, flags); 966 } 967 968 error = zio_wait(zio); 969 970 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 971 flags |= ZIO_FLAG_TRYHARD; 972 goto retry; 973 } 974 975 nvlist_free(label); 976 abd_free(pad2); 977 abd_free(ub_abd); 978 abd_free(vp_abd); 979 980 /* 981 * If this vdev hasn't been previously identified as a spare, then we 982 * mark it as such only if a) we are labeling it as a spare, or b) it 983 * exists as a spare elsewhere in the system. Do the same for 984 * level 2 ARC devices. 985 */ 986 if (error == 0 && !vd->vdev_isspare && 987 (reason == VDEV_LABEL_SPARE || 988 spa_spare_exists(vd->vdev_guid, NULL, NULL))) 989 spa_spare_add(vd); 990 991 if (error == 0 && !vd->vdev_isl2cache && 992 (reason == VDEV_LABEL_L2CACHE || 993 spa_l2cache_exists(vd->vdev_guid, NULL))) 994 spa_l2cache_add(vd); 995 996 return (error); 997 } 998 999 /* 1000 * ========================================================================== 1001 * uberblock load/sync 1002 * ========================================================================== 1003 */ 1004 1005 /* 1006 * Consider the following situation: txg is safely synced to disk. We've 1007 * written the first uberblock for txg + 1, and then we lose power. When we 1008 * come back up, we fail to see the uberblock for txg + 1 because, say, 1009 * it was on a mirrored device and the replica to which we wrote txg + 1 1010 * is now offline. If we then make some changes and sync txg + 1, and then 1011 * the missing replica comes back, then for a few seconds we'll have two 1012 * conflicting uberblocks on disk with the same txg. The solution is simple: 1013 * among uberblocks with equal txg, choose the one with the latest timestamp. 1014 */ 1015 static int 1016 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 1017 { 1018 if (ub1->ub_txg < ub2->ub_txg) 1019 return (-1); 1020 if (ub1->ub_txg > ub2->ub_txg) 1021 return (1); 1022 1023 if (ub1->ub_timestamp < ub2->ub_timestamp) 1024 return (-1); 1025 if (ub1->ub_timestamp > ub2->ub_timestamp) 1026 return (1); 1027 1028 return (0); 1029 } 1030 1031 struct ubl_cbdata { 1032 uberblock_t *ubl_ubbest; /* Best uberblock */ 1033 vdev_t *ubl_vd; /* vdev associated with the above */ 1034 }; 1035 1036 static void 1037 vdev_uberblock_load_done(zio_t *zio) 1038 { 1039 vdev_t *vd = zio->io_vd; 1040 spa_t *spa = zio->io_spa; 1041 zio_t *rio = zio->io_private; 1042 uberblock_t *ub = abd_to_buf(zio->io_abd); 1043 struct ubl_cbdata *cbp = rio->io_private; 1044 1045 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 1046 1047 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 1048 mutex_enter(&rio->io_lock); 1049 if (ub->ub_txg <= spa->spa_load_max_txg && 1050 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 1051 /* 1052 * Keep track of the vdev in which this uberblock 1053 * was found. We will use this information later 1054 * to obtain the config nvlist associated with 1055 * this uberblock. 1056 */ 1057 *cbp->ubl_ubbest = *ub; 1058 cbp->ubl_vd = vd; 1059 } 1060 mutex_exit(&rio->io_lock); 1061 } 1062 1063 abd_free(zio->io_abd); 1064 } 1065 1066 static void 1067 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 1068 struct ubl_cbdata *cbp) 1069 { 1070 for (int c = 0; c < vd->vdev_children; c++) 1071 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 1072 1073 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1074 for (int l = 0; l < VDEV_LABELS; l++) { 1075 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1076 vdev_label_read(zio, vd, l, 1077 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), 1078 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n), 1079 VDEV_UBERBLOCK_SIZE(vd), 1080 vdev_uberblock_load_done, zio, flags); 1081 } 1082 } 1083 } 1084 } 1085 1086 /* 1087 * Reads the 'best' uberblock from disk along with its associated 1088 * configuration. First, we read the uberblock array of each label of each 1089 * vdev, keeping track of the uberblock with the highest txg in each array. 1090 * Then, we read the configuration from the same vdev as the best uberblock. 1091 */ 1092 void 1093 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 1094 { 1095 zio_t *zio; 1096 spa_t *spa = rvd->vdev_spa; 1097 struct ubl_cbdata cb; 1098 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1099 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1100 1101 ASSERT(ub); 1102 ASSERT(config); 1103 1104 bzero(ub, sizeof (uberblock_t)); 1105 *config = NULL; 1106 1107 cb.ubl_ubbest = ub; 1108 cb.ubl_vd = NULL; 1109 1110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1111 zio = zio_root(spa, NULL, &cb, flags); 1112 vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1113 (void) zio_wait(zio); 1114 1115 /* 1116 * It's possible that the best uberblock was discovered on a label 1117 * that has a configuration which was written in a future txg. 1118 * Search all labels on this vdev to find the configuration that 1119 * matches the txg for our uberblock. 1120 */ 1121 if (cb.ubl_vd != NULL) { 1122 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. " 1123 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg); 1124 1125 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 1126 if (*config == NULL && spa->spa_extreme_rewind) { 1127 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. " 1128 "Trying again without txg restrictions."); 1129 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX); 1130 } 1131 if (*config == NULL) { 1132 vdev_dbgmsg(cb.ubl_vd, "failed to read label config"); 1133 } 1134 } 1135 spa_config_exit(spa, SCL_ALL, FTAG); 1136 } 1137 1138 /* 1139 * On success, increment root zio's count of good writes. 1140 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1141 */ 1142 static void 1143 vdev_uberblock_sync_done(zio_t *zio) 1144 { 1145 uint64_t *good_writes = zio->io_private; 1146 1147 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1148 atomic_inc_64(good_writes); 1149 } 1150 1151 /* 1152 * Write the uberblock to all labels of all leaves of the specified vdev. 1153 */ 1154 static void 1155 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes, 1156 uberblock_t *ub, vdev_t *vd, int flags) 1157 { 1158 for (uint64_t c = 0; c < vd->vdev_children; c++) { 1159 vdev_uberblock_sync(zio, good_writes, 1160 ub, vd->vdev_child[c], flags); 1161 } 1162 1163 if (!vd->vdev_ops->vdev_op_leaf) 1164 return; 1165 1166 if (!vdev_writeable(vd)) 1167 return; 1168 1169 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0; 1170 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m); 1171 1172 /* Copy the uberblock_t into the ABD */ 1173 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1174 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1175 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t)); 1176 1177 for (int l = 0; l < VDEV_LABELS; l++) 1178 vdev_label_write(zio, vd, l, ub_abd, 1179 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1180 vdev_uberblock_sync_done, good_writes, 1181 flags | ZIO_FLAG_DONT_PROPAGATE); 1182 1183 abd_free(ub_abd); 1184 } 1185 1186 /* Sync the uberblocks to all vdevs in svd[] */ 1187 int 1188 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1189 { 1190 spa_t *spa = svd[0]->vdev_spa; 1191 zio_t *zio; 1192 uint64_t good_writes = 0; 1193 1194 zio = zio_root(spa, NULL, NULL, flags); 1195 1196 for (int v = 0; v < svdcount; v++) 1197 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags); 1198 1199 (void) zio_wait(zio); 1200 1201 /* 1202 * Flush the uberblocks to disk. This ensures that the odd labels 1203 * are no longer needed (because the new uberblocks and the even 1204 * labels are safely on disk), so it is safe to overwrite them. 1205 */ 1206 zio = zio_root(spa, NULL, NULL, flags); 1207 1208 for (int v = 0; v < svdcount; v++) { 1209 if (vdev_writeable(svd[v])) { 1210 zio_flush(zio, svd[v]); 1211 } 1212 } 1213 1214 (void) zio_wait(zio); 1215 1216 return (good_writes >= 1 ? 0 : EIO); 1217 } 1218 1219 /* 1220 * On success, increment the count of good writes for our top-level vdev. 1221 */ 1222 static void 1223 vdev_label_sync_done(zio_t *zio) 1224 { 1225 uint64_t *good_writes = zio->io_private; 1226 1227 if (zio->io_error == 0) 1228 atomic_inc_64(good_writes); 1229 } 1230 1231 /* 1232 * If there weren't enough good writes, indicate failure to the parent. 1233 */ 1234 static void 1235 vdev_label_sync_top_done(zio_t *zio) 1236 { 1237 uint64_t *good_writes = zio->io_private; 1238 1239 if (*good_writes == 0) 1240 zio->io_error = SET_ERROR(EIO); 1241 1242 kmem_free(good_writes, sizeof (uint64_t)); 1243 } 1244 1245 /* 1246 * We ignore errors for log and cache devices, simply free the private data. 1247 */ 1248 static void 1249 vdev_label_sync_ignore_done(zio_t *zio) 1250 { 1251 kmem_free(zio->io_private, sizeof (uint64_t)); 1252 } 1253 1254 /* 1255 * Write all even or odd labels to all leaves of the specified vdev. 1256 */ 1257 static void 1258 vdev_label_sync(zio_t *zio, uint64_t *good_writes, 1259 vdev_t *vd, int l, uint64_t txg, int flags) 1260 { 1261 nvlist_t *label; 1262 vdev_phys_t *vp; 1263 abd_t *vp_abd; 1264 char *buf; 1265 size_t buflen; 1266 1267 for (int c = 0; c < vd->vdev_children; c++) { 1268 vdev_label_sync(zio, good_writes, 1269 vd->vdev_child[c], l, txg, flags); 1270 } 1271 1272 if (!vd->vdev_ops->vdev_op_leaf) 1273 return; 1274 1275 if (!vdev_writeable(vd)) 1276 return; 1277 1278 /* 1279 * Generate a label describing the top-level config to which we belong. 1280 */ 1281 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1282 1283 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1284 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1285 vp = abd_to_buf(vp_abd); 1286 1287 buf = vp->vp_nvlist; 1288 buflen = sizeof (vp->vp_nvlist); 1289 1290 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1291 for (; l < VDEV_LABELS; l += 2) { 1292 vdev_label_write(zio, vd, l, vp_abd, 1293 offsetof(vdev_label_t, vl_vdev_phys), 1294 sizeof (vdev_phys_t), 1295 vdev_label_sync_done, good_writes, 1296 flags | ZIO_FLAG_DONT_PROPAGATE); 1297 } 1298 } 1299 1300 abd_free(vp_abd); 1301 nvlist_free(label); 1302 } 1303 1304 int 1305 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1306 { 1307 list_t *dl = &spa->spa_config_dirty_list; 1308 vdev_t *vd; 1309 zio_t *zio; 1310 int error; 1311 1312 /* 1313 * Write the new labels to disk. 1314 */ 1315 zio = zio_root(spa, NULL, NULL, flags); 1316 1317 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1318 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1319 KM_SLEEP); 1320 1321 ASSERT(!vd->vdev_ishole); 1322 1323 zio_t *vio = zio_null(zio, spa, NULL, 1324 (vd->vdev_islog || vd->vdev_aux != NULL) ? 1325 vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1326 good_writes, flags); 1327 vdev_label_sync(vio, good_writes, vd, l, txg, flags); 1328 zio_nowait(vio); 1329 } 1330 1331 error = zio_wait(zio); 1332 1333 /* 1334 * Flush the new labels to disk. 1335 */ 1336 zio = zio_root(spa, NULL, NULL, flags); 1337 1338 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1339 zio_flush(zio, vd); 1340 1341 (void) zio_wait(zio); 1342 1343 return (error); 1344 } 1345 1346 /* 1347 * Sync the uberblock and any changes to the vdev configuration. 1348 * 1349 * The order of operations is carefully crafted to ensure that 1350 * if the system panics or loses power at any time, the state on disk 1351 * is still transactionally consistent. The in-line comments below 1352 * describe the failure semantics at each stage. 1353 * 1354 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1355 * at any time, you can just call it again, and it will resume its work. 1356 */ 1357 int 1358 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1359 { 1360 spa_t *spa = svd[0]->vdev_spa; 1361 uberblock_t *ub = &spa->spa_uberblock; 1362 int error = 0; 1363 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1364 1365 ASSERT(svdcount != 0); 1366 retry: 1367 /* 1368 * Normally, we don't want to try too hard to write every label and 1369 * uberblock. If there is a flaky disk, we don't want the rest of the 1370 * sync process to block while we retry. But if we can't write a 1371 * single label out, we should retry with ZIO_FLAG_TRYHARD before 1372 * bailing out and declaring the pool faulted. 1373 */ 1374 if (error != 0) { 1375 if ((flags & ZIO_FLAG_TRYHARD) != 0) 1376 return (error); 1377 flags |= ZIO_FLAG_TRYHARD; 1378 } 1379 1380 ASSERT(ub->ub_txg <= txg); 1381 1382 /* 1383 * If this isn't a resync due to I/O errors, 1384 * and nothing changed in this transaction group, 1385 * and the vdev configuration hasn't changed, 1386 * then there's nothing to do. 1387 */ 1388 if (ub->ub_txg < txg) { 1389 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev, 1390 txg, spa->spa_mmp.mmp_delay); 1391 1392 if (!changed && list_is_empty(&spa->spa_config_dirty_list)) 1393 return (0); 1394 } 1395 1396 if (txg > spa_freeze_txg(spa)) 1397 return (0); 1398 1399 ASSERT(txg <= spa->spa_final_txg); 1400 1401 /* 1402 * Flush the write cache of every disk that's been written to 1403 * in this transaction group. This ensures that all blocks 1404 * written in this txg will be committed to stable storage 1405 * before any uberblock that references them. 1406 */ 1407 zio_t *zio = zio_root(spa, NULL, NULL, flags); 1408 1409 for (vdev_t *vd = 1410 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL; 1411 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1412 zio_flush(zio, vd); 1413 1414 (void) zio_wait(zio); 1415 1416 /* 1417 * Sync out the even labels (L0, L2) for every dirty vdev. If the 1418 * system dies in the middle of this process, that's OK: all of the 1419 * even labels that made it to disk will be newer than any uberblock, 1420 * and will therefore be considered invalid. The odd labels (L1, L3), 1421 * which have not yet been touched, will still be valid. We flush 1422 * the new labels to disk to ensure that all even-label updates 1423 * are committed to stable storage before the uberblock update. 1424 */ 1425 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) { 1426 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1427 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1428 "for pool '%s' when syncing out the even labels " 1429 "of dirty vdevs", error, spa_name(spa)); 1430 } 1431 goto retry; 1432 } 1433 1434 /* 1435 * Sync the uberblocks to all vdevs in svd[]. 1436 * If the system dies in the middle of this step, there are two cases 1437 * to consider, and the on-disk state is consistent either way: 1438 * 1439 * (1) If none of the new uberblocks made it to disk, then the 1440 * previous uberblock will be the newest, and the odd labels 1441 * (which had not yet been touched) will be valid with respect 1442 * to that uberblock. 1443 * 1444 * (2) If one or more new uberblocks made it to disk, then they 1445 * will be the newest, and the even labels (which had all 1446 * been successfully committed) will be valid with respect 1447 * to the new uberblocks. 1448 */ 1449 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) { 1450 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1451 zfs_dbgmsg("vdev_uberblock_sync_list() returned error " 1452 "%d for pool '%s'", error, spa_name(spa)); 1453 } 1454 goto retry; 1455 } 1456 1457 if (spa_multihost(spa)) 1458 mmp_update_uberblock(spa, ub); 1459 1460 /* 1461 * Sync out odd labels for every dirty vdev. If the system dies 1462 * in the middle of this process, the even labels and the new 1463 * uberblocks will suffice to open the pool. The next time 1464 * the pool is opened, the first thing we'll do -- before any 1465 * user data is modified -- is mark every vdev dirty so that 1466 * all labels will be brought up to date. We flush the new labels 1467 * to disk to ensure that all odd-label updates are committed to 1468 * stable storage before the next transaction group begins. 1469 */ 1470 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) { 1471 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1472 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1473 "for pool '%s' when syncing out the odd labels of " 1474 "dirty vdevs", error, spa_name(spa)); 1475 } 1476 goto retry; 1477 } 1478 1479 return (0); 1480 } 1481