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, 2020 by Delphix. All rights reserved. 25 * Copyright (c) 2017, Intel Corporation. 26 */ 27 28/* 29 * Virtual Device Labels 30 * --------------------- 31 * 32 * The vdev label serves several distinct purposes: 33 * 34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its 35 * identity within the pool. 36 * 37 * 2. Verify that all the devices given in a configuration are present 38 * within the pool. 39 * 40 * 3. Determine the uberblock for the pool. 41 * 42 * 4. In case of an import operation, determine the configuration of the 43 * toplevel vdev of which it is a part. 44 * 45 * 5. If an import operation cannot find all the devices in the pool, 46 * provide enough information to the administrator to determine which 47 * devices are missing. 48 * 49 * It is important to note that while the kernel is responsible for writing the 50 * label, it only consumes the information in the first three cases. The 51 * latter information is only consumed in userland when determining the 52 * configuration to import a pool. 53 * 54 * 55 * Label Organization 56 * ------------------ 57 * 58 * Before describing the contents of the label, it's important to understand how 59 * the labels are written and updated with respect to the uberblock. 60 * 61 * When the pool configuration is altered, either because it was newly created 62 * or a device was added, we want to update all the labels such that we can deal 63 * with fatal failure at any point. To this end, each disk has two labels which 64 * are updated before and after the uberblock is synced. Assuming we have 65 * labels and an uberblock with the following transaction groups: 66 * 67 * L1 UB L2 68 * +------+ +------+ +------+ 69 * | | | | | | 70 * | t10 | | t10 | | t10 | 71 * | | | | | | 72 * +------+ +------+ +------+ 73 * 74 * In this stable state, the labels and the uberblock were all updated within 75 * the same transaction group (10). Each label is mirrored and checksummed, so 76 * that we can detect when we fail partway through writing the label. 77 * 78 * In order to identify which labels are valid, the labels are written in the 79 * following manner: 80 * 81 * 1. For each vdev, update 'L1' to the new label 82 * 2. Update the uberblock 83 * 3. For each vdev, update 'L2' to the new label 84 * 85 * Given arbitrary failure, we can determine the correct label to use based on 86 * the transaction group. If we fail after updating L1 but before updating the 87 * UB, we will notice that L1's transaction group is greater than the uberblock, 88 * so L2 must be valid. If we fail after writing the uberblock but before 89 * writing L2, we will notice that L2's transaction group is less than L1, and 90 * therefore L1 is valid. 91 * 92 * Another added complexity is that not every label is updated when the config 93 * is synced. If we add a single device, we do not want to have to re-write 94 * every label for every device in the pool. This means that both L1 and L2 may 95 * be older than the pool uberblock, because the necessary information is stored 96 * on another vdev. 97 * 98 * 99 * On-disk Format 100 * -------------- 101 * 102 * The vdev label consists of two distinct parts, and is wrapped within the 103 * vdev_label_t structure. The label includes 8k of padding to permit legacy 104 * VTOC disk labels, but is otherwise ignored. 105 * 106 * The first half of the label is a packed nvlist which contains pool wide 107 * properties, per-vdev properties, and configuration information. It is 108 * described in more detail below. 109 * 110 * The latter half of the label consists of a redundant array of uberblocks. 111 * These uberblocks are updated whenever a transaction group is committed, 112 * or when the configuration is updated. When a pool is loaded, we scan each 113 * vdev for the 'best' uberblock. 114 * 115 * 116 * Configuration Information 117 * ------------------------- 118 * 119 * The nvlist describing the pool and vdev contains the following elements: 120 * 121 * version ZFS on-disk version 122 * name Pool name 123 * state Pool state 124 * txg Transaction group in which this label was written 125 * pool_guid Unique identifier for this pool 126 * vdev_tree An nvlist describing vdev tree. 127 * features_for_read 128 * An nvlist of the features necessary for reading the MOS. 129 * 130 * Each leaf device label also contains the following: 131 * 132 * top_guid Unique ID for top-level vdev in which this is contained 133 * guid Unique ID for the leaf vdev 134 * 135 * The 'vs' configuration follows the format described in 'spa_config.c'. 136 */ 137 138#include <sys/zfs_context.h> 139#include <sys/spa.h> 140#include <sys/spa_impl.h> 141#include <sys/dmu.h> 142#include <sys/zap.h> 143#include <sys/vdev.h> 144#include <sys/vdev_impl.h> 145#include <sys/vdev_draid.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#include <sys/byteorder.h> 154#include <sys/zfs_bootenv.h> 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, abd_t *buf, uint64_t offset, 188 uint64_t size, zio_done_func_t *done, void *private, int flags) 189{ 190 ASSERT( 191 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 192 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 193 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 194 195 zio_nowait(zio_read_phys(zio, vd, 196 vdev_label_offset(vd->vdev_psize, l, offset), 197 size, buf, ZIO_CHECKSUM_LABEL, done, private, 198 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 199} 200 201void 202vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 203 uint64_t size, zio_done_func_t *done, void *private, int flags) 204{ 205 ASSERT( 206 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 207 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 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 stats 218 */ 219void 220vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv) 221{ 222 nvlist_t *nvx; 223 vdev_stat_t *vs; 224 vdev_stat_ex_t *vsx; 225 226 vs = kmem_alloc(sizeof (*vs), KM_SLEEP); 227 vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP); 228 229 vdev_get_stats_ex(vd, vs, vsx); 230 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 231 (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t)); 232 233 /* 234 * Add extended stats into a special extended stats nvlist. This keeps 235 * all the extended stats nicely grouped together. The extended stats 236 * nvlist is then added to the main nvlist. 237 */ 238 nvx = fnvlist_alloc(); 239 240 /* ZIOs in flight to disk */ 241 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE, 242 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]); 243 244 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE, 245 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]); 246 247 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE, 248 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]); 249 250 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE, 251 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]); 252 253 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE, 254 vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]); 255 256 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE, 257 vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]); 258 259 /* ZIOs pending */ 260 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE, 261 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]); 262 263 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE, 264 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]); 265 266 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE, 267 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]); 268 269 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE, 270 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]); 271 272 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE, 273 vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]); 274 275 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE, 276 vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]); 277 278 /* Histograms */ 279 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO, 280 vsx->vsx_total_histo[ZIO_TYPE_READ], 281 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ])); 282 283 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO, 284 vsx->vsx_total_histo[ZIO_TYPE_WRITE], 285 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE])); 286 287 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO, 288 vsx->vsx_disk_histo[ZIO_TYPE_READ], 289 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ])); 290 291 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO, 292 vsx->vsx_disk_histo[ZIO_TYPE_WRITE], 293 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE])); 294 295 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO, 296 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ], 297 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ])); 298 299 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO, 300 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE], 301 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE])); 302 303 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO, 304 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ], 305 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ])); 306 307 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO, 308 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE], 309 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE])); 310 311 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO, 312 vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB], 313 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB])); 314 315 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO, 316 vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM], 317 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM])); 318 319 /* Request sizes */ 320 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO, 321 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ], 322 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ])); 323 324 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO, 325 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE], 326 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE])); 327 328 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO, 329 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ], 330 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ])); 331 332 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO, 333 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE], 334 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE])); 335 336 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO, 337 vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB], 338 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB])); 339 340 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO, 341 vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM], 342 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM])); 343 344 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO, 345 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ], 346 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ])); 347 348 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO, 349 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE], 350 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE])); 351 352 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO, 353 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ], 354 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ])); 355 356 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO, 357 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE], 358 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE])); 359 360 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO, 361 vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB], 362 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB])); 363 364 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO, 365 vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM], 366 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM])); 367 368 /* IO delays */ 369 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios); 370 371 /* Add extended stats nvlist to main nvlist */ 372 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx); 373 374 fnvlist_free(nvx); 375 kmem_free(vs, sizeof (*vs)); 376 kmem_free(vsx, sizeof (*vsx)); 377} 378 379static void 380root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 381{ 382 spa_t *spa = vd->vdev_spa; 383 384 if (vd != spa->spa_root_vdev) 385 return; 386 387 /* provide either current or previous scan information */ 388 pool_scan_stat_t ps; 389 if (spa_scan_get_stats(spa, &ps) == 0) { 390 fnvlist_add_uint64_array(nvl, 391 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 392 sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 393 } 394 395 pool_removal_stat_t prs; 396 if (spa_removal_get_stats(spa, &prs) == 0) { 397 fnvlist_add_uint64_array(nvl, 398 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs, 399 sizeof (prs) / sizeof (uint64_t)); 400 } 401 402 pool_checkpoint_stat_t pcs; 403 if (spa_checkpoint_get_stats(spa, &pcs) == 0) { 404 fnvlist_add_uint64_array(nvl, 405 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs, 406 sizeof (pcs) / sizeof (uint64_t)); 407 } 408} 409 410static void 411top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 412{ 413 if (vd == vd->vdev_top) { 414 vdev_rebuild_stat_t vrs; 415 if (vdev_rebuild_get_stats(vd, &vrs) == 0) { 416 fnvlist_add_uint64_array(nvl, 417 ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs, 418 sizeof (vrs) / sizeof (uint64_t)); 419 } 420 } 421} 422 423/* 424 * Generate the nvlist representing this vdev's config. 425 */ 426nvlist_t * 427vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 428 vdev_config_flag_t flags) 429{ 430 nvlist_t *nv = NULL; 431 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 432 433 nv = fnvlist_alloc(); 434 435 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 436 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 437 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 438 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 439 440 if (vd->vdev_path != NULL) 441 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 442 443 if (vd->vdev_devid != NULL) 444 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 445 446 if (vd->vdev_physpath != NULL) 447 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 448 vd->vdev_physpath); 449 450 if (vd->vdev_enc_sysfs_path != NULL) 451 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, 452 vd->vdev_enc_sysfs_path); 453 454 if (vd->vdev_fru != NULL) 455 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 456 457 if (vd->vdev_ops->vdev_op_config_generate != NULL) 458 vd->vdev_ops->vdev_op_config_generate(vd, nv); 459 460 if (vd->vdev_wholedisk != -1ULL) { 461 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 462 vd->vdev_wholedisk); 463 } 464 465 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING)) 466 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 467 468 if (vd->vdev_isspare) 469 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 470 471 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 472 vd == vd->vdev_top) { 473 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 474 vd->vdev_ms_array); 475 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 476 vd->vdev_ms_shift); 477 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 478 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 479 vd->vdev_asize); 480 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 481 if (vd->vdev_removing) { 482 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 483 vd->vdev_removing); 484 } 485 486 /* zpool command expects alloc class data */ 487 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) { 488 const char *bias = NULL; 489 490 switch (vd->vdev_alloc_bias) { 491 case VDEV_BIAS_LOG: 492 bias = VDEV_ALLOC_BIAS_LOG; 493 break; 494 case VDEV_BIAS_SPECIAL: 495 bias = VDEV_ALLOC_BIAS_SPECIAL; 496 break; 497 case VDEV_BIAS_DEDUP: 498 bias = VDEV_ALLOC_BIAS_DEDUP; 499 break; 500 default: 501 ASSERT3U(vd->vdev_alloc_bias, ==, 502 VDEV_BIAS_NONE); 503 } 504 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 505 bias); 506 } 507 } 508 509 if (vd->vdev_dtl_sm != NULL) { 510 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 511 space_map_object(vd->vdev_dtl_sm)); 512 } 513 514 if (vic->vic_mapping_object != 0) { 515 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 516 vic->vic_mapping_object); 517 } 518 519 if (vic->vic_births_object != 0) { 520 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 521 vic->vic_births_object); 522 } 523 524 if (vic->vic_prev_indirect_vdev != UINT64_MAX) { 525 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 526 vic->vic_prev_indirect_vdev); 527 } 528 529 if (vd->vdev_crtxg) 530 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 531 532 if (vd->vdev_expansion_time) 533 fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME, 534 vd->vdev_expansion_time); 535 536 if (flags & VDEV_CONFIG_MOS) { 537 if (vd->vdev_leaf_zap != 0) { 538 ASSERT(vd->vdev_ops->vdev_op_leaf); 539 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 540 vd->vdev_leaf_zap); 541 } 542 543 if (vd->vdev_top_zap != 0) { 544 ASSERT(vd == vd->vdev_top); 545 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 546 vd->vdev_top_zap); 547 } 548 549 if (vd->vdev_resilver_deferred) { 550 ASSERT(vd->vdev_ops->vdev_op_leaf); 551 ASSERT(spa->spa_resilver_deferred); 552 fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER); 553 } 554 } 555 556 if (getstats) { 557 vdev_config_generate_stats(vd, nv); 558 559 root_vdev_actions_getprogress(vd, nv); 560 top_vdev_actions_getprogress(vd, nv); 561 562 /* 563 * Note: this can be called from open context 564 * (spa_get_stats()), so we need the rwlock to prevent 565 * the mapping from being changed by condensing. 566 */ 567 rw_enter(&vd->vdev_indirect_rwlock, RW_READER); 568 if (vd->vdev_indirect_mapping != NULL) { 569 ASSERT(vd->vdev_indirect_births != NULL); 570 vdev_indirect_mapping_t *vim = 571 vd->vdev_indirect_mapping; 572 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 573 vdev_indirect_mapping_size(vim)); 574 } 575 rw_exit(&vd->vdev_indirect_rwlock); 576 if (vd->vdev_mg != NULL && 577 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) { 578 /* 579 * Compute approximately how much memory would be used 580 * for the indirect mapping if this device were to 581 * be removed. 582 * 583 * Note: If the frag metric is invalid, then not 584 * enough metaslabs have been converted to have 585 * histograms. 586 */ 587 uint64_t seg_count = 0; 588 uint64_t to_alloc = vd->vdev_stat.vs_alloc; 589 590 /* 591 * There are the same number of allocated segments 592 * as free segments, so we will have at least one 593 * entry per free segment. However, small free 594 * segments (smaller than vdev_removal_max_span) 595 * will be combined with adjacent allocated segments 596 * as a single mapping. 597 */ 598 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 599 if (i + 1 < highbit64(vdev_removal_max_span) 600 - 1) { 601 to_alloc += 602 vd->vdev_mg->mg_histogram[i] << 603 (i + 1); 604 } else { 605 seg_count += 606 vd->vdev_mg->mg_histogram[i]; 607 } 608 } 609 610 /* 611 * The maximum length of a mapping is 612 * zfs_remove_max_segment, so we need at least one entry 613 * per zfs_remove_max_segment of allocated data. 614 */ 615 seg_count += to_alloc / spa_remove_max_segment(spa); 616 617 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 618 seg_count * 619 sizeof (vdev_indirect_mapping_entry_phys_t)); 620 } 621 } 622 623 if (!vd->vdev_ops->vdev_op_leaf) { 624 nvlist_t **child; 625 int c, idx; 626 627 ASSERT(!vd->vdev_ishole); 628 629 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 630 KM_SLEEP); 631 632 for (c = 0, idx = 0; c < vd->vdev_children; c++) { 633 vdev_t *cvd = vd->vdev_child[c]; 634 635 /* 636 * If we're generating an nvlist of removing 637 * vdevs then skip over any device which is 638 * not being removed. 639 */ 640 if ((flags & VDEV_CONFIG_REMOVING) && 641 !cvd->vdev_removing) 642 continue; 643 644 child[idx++] = vdev_config_generate(spa, cvd, 645 getstats, flags); 646 } 647 648 if (idx) { 649 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 650 child, idx); 651 } 652 653 for (c = 0; c < idx; c++) 654 nvlist_free(child[c]); 655 656 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 657 658 } else { 659 const char *aux = NULL; 660 661 if (vd->vdev_offline && !vd->vdev_tmpoffline) 662 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 663 if (vd->vdev_resilver_txg != 0) 664 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 665 vd->vdev_resilver_txg); 666 if (vd->vdev_rebuild_txg != 0) 667 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG, 668 vd->vdev_rebuild_txg); 669 if (vd->vdev_faulted) 670 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 671 if (vd->vdev_degraded) 672 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 673 if (vd->vdev_removed) 674 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 675 if (vd->vdev_unspare) 676 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 677 if (vd->vdev_ishole) 678 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 679 680 /* Set the reason why we're FAULTED/DEGRADED. */ 681 switch (vd->vdev_stat.vs_aux) { 682 case VDEV_AUX_ERR_EXCEEDED: 683 aux = "err_exceeded"; 684 break; 685 686 case VDEV_AUX_EXTERNAL: 687 aux = "external"; 688 break; 689 } 690 691 if (aux != NULL && !vd->vdev_tmpoffline) { 692 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 693 } else { 694 /* 695 * We're healthy - clear any previous AUX_STATE values. 696 */ 697 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE)) 698 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE); 699 } 700 701 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 702 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 703 vd->vdev_orig_guid); 704 } 705 } 706 707 return (nv); 708} 709 710/* 711 * Generate a view of the top-level vdevs. If we currently have holes 712 * in the namespace, then generate an array which contains a list of holey 713 * vdevs. Additionally, add the number of top-level children that currently 714 * exist. 715 */ 716void 717vdev_top_config_generate(spa_t *spa, nvlist_t *config) 718{ 719 vdev_t *rvd = spa->spa_root_vdev; 720 uint64_t *array; 721 uint_t c, idx; 722 723 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 724 725 for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 726 vdev_t *tvd = rvd->vdev_child[c]; 727 728 if (tvd->vdev_ishole) { 729 array[idx++] = c; 730 } 731 } 732 733 if (idx) { 734 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 735 array, idx) == 0); 736 } 737 738 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 739 rvd->vdev_children) == 0); 740 741 kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 742} 743 744/* 745 * Returns the configuration from the label of the given vdev. For vdevs 746 * which don't have a txg value stored on their label (i.e. spares/cache) 747 * or have not been completely initialized (txg = 0) just return 748 * the configuration from the first valid label we find. Otherwise, 749 * find the most up-to-date label that does not exceed the specified 750 * 'txg' value. 751 */ 752nvlist_t * 753vdev_label_read_config(vdev_t *vd, uint64_t txg) 754{ 755 spa_t *spa = vd->vdev_spa; 756 nvlist_t *config = NULL; 757 vdev_phys_t *vp[VDEV_LABELS]; 758 abd_t *vp_abd[VDEV_LABELS]; 759 zio_t *zio[VDEV_LABELS]; 760 uint64_t best_txg = 0; 761 uint64_t label_txg = 0; 762 int error = 0; 763 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 764 ZIO_FLAG_SPECULATIVE; 765 766 ASSERT(vd->vdev_validate_thread == curthread || 767 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 768 769 if (!vdev_readable(vd)) 770 return (NULL); 771 772 /* 773 * The label for a dRAID distributed spare is not stored on disk. 774 * Instead it is generated when needed which allows us to bypass 775 * the pipeline when reading the config from the label. 776 */ 777 if (vd->vdev_ops == &vdev_draid_spare_ops) 778 return (vdev_draid_read_config_spare(vd)); 779 780 for (int l = 0; l < VDEV_LABELS; l++) { 781 vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 782 vp[l] = abd_to_buf(vp_abd[l]); 783 } 784 785retry: 786 for (int l = 0; l < VDEV_LABELS; l++) { 787 zio[l] = zio_root(spa, NULL, NULL, flags); 788 789 vdev_label_read(zio[l], vd, l, vp_abd[l], 790 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 791 NULL, NULL, flags); 792 } 793 for (int l = 0; l < VDEV_LABELS; l++) { 794 nvlist_t *label = NULL; 795 796 if (zio_wait(zio[l]) == 0 && 797 nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist), 798 &label, 0) == 0) { 799 /* 800 * Auxiliary vdevs won't have txg values in their 801 * labels and newly added vdevs may not have been 802 * completely initialized so just return the 803 * configuration from the first valid label we 804 * encounter. 805 */ 806 error = nvlist_lookup_uint64(label, 807 ZPOOL_CONFIG_POOL_TXG, &label_txg); 808 if ((error || label_txg == 0) && !config) { 809 config = label; 810 for (l++; l < VDEV_LABELS; l++) 811 zio_wait(zio[l]); 812 break; 813 } else if (label_txg <= txg && label_txg > best_txg) { 814 best_txg = label_txg; 815 nvlist_free(config); 816 config = fnvlist_dup(label); 817 } 818 } 819 820 if (label != NULL) { 821 nvlist_free(label); 822 label = NULL; 823 } 824 } 825 826 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 827 flags |= ZIO_FLAG_TRYHARD; 828 goto retry; 829 } 830 831 /* 832 * We found a valid label but it didn't pass txg restrictions. 833 */ 834 if (config == NULL && label_txg != 0) { 835 vdev_dbgmsg(vd, "label discarded as txg is too large " 836 "(%llu > %llu)", (u_longlong_t)label_txg, 837 (u_longlong_t)txg); 838 } 839 840 for (int l = 0; l < VDEV_LABELS; l++) { 841 abd_free(vp_abd[l]); 842 } 843 844 return (config); 845} 846 847/* 848 * Determine if a device is in use. The 'spare_guid' parameter will be filled 849 * in with the device guid if this spare is active elsewhere on the system. 850 */ 851static boolean_t 852vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 853 uint64_t *spare_guid, uint64_t *l2cache_guid) 854{ 855 spa_t *spa = vd->vdev_spa; 856 uint64_t state, pool_guid, device_guid, txg, spare_pool; 857 uint64_t vdtxg = 0; 858 nvlist_t *label; 859 860 if (spare_guid) 861 *spare_guid = 0ULL; 862 if (l2cache_guid) 863 *l2cache_guid = 0ULL; 864 865 /* 866 * Read the label, if any, and perform some basic sanity checks. 867 */ 868 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 869 return (B_FALSE); 870 871 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 872 &vdtxg); 873 874 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 875 &state) != 0 || 876 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 877 &device_guid) != 0) { 878 nvlist_free(label); 879 return (B_FALSE); 880 } 881 882 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 883 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 884 &pool_guid) != 0 || 885 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 886 &txg) != 0)) { 887 nvlist_free(label); 888 return (B_FALSE); 889 } 890 891 nvlist_free(label); 892 893 /* 894 * Check to see if this device indeed belongs to the pool it claims to 895 * be a part of. The only way this is allowed is if the device is a hot 896 * spare (which we check for later on). 897 */ 898 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 899 !spa_guid_exists(pool_guid, device_guid) && 900 !spa_spare_exists(device_guid, NULL, NULL) && 901 !spa_l2cache_exists(device_guid, NULL)) 902 return (B_FALSE); 903 904 /* 905 * If the transaction group is zero, then this an initialized (but 906 * unused) label. This is only an error if the create transaction 907 * on-disk is the same as the one we're using now, in which case the 908 * user has attempted to add the same vdev multiple times in the same 909 * transaction. 910 */ 911 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 912 txg == 0 && vdtxg == crtxg) 913 return (B_TRUE); 914 915 /* 916 * Check to see if this is a spare device. We do an explicit check for 917 * spa_has_spare() here because it may be on our pending list of spares 918 * to add. We also check if it is an l2cache device. 919 */ 920 if (spa_spare_exists(device_guid, &spare_pool, NULL) || 921 spa_has_spare(spa, device_guid)) { 922 if (spare_guid) 923 *spare_guid = device_guid; 924 925 switch (reason) { 926 case VDEV_LABEL_CREATE: 927 case VDEV_LABEL_L2CACHE: 928 return (B_TRUE); 929 930 case VDEV_LABEL_REPLACE: 931 return (!spa_has_spare(spa, device_guid) || 932 spare_pool != 0ULL); 933 934 case VDEV_LABEL_SPARE: 935 return (spa_has_spare(spa, device_guid)); 936 default: 937 break; 938 } 939 } 940 941 /* 942 * Check to see if this is an l2cache device. 943 */ 944 if (spa_l2cache_exists(device_guid, NULL)) 945 return (B_TRUE); 946 947 /* 948 * We can't rely on a pool's state if it's been imported 949 * read-only. Instead we look to see if the pools is marked 950 * read-only in the namespace and set the state to active. 951 */ 952 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 953 (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 954 spa_mode(spa) == SPA_MODE_READ) 955 state = POOL_STATE_ACTIVE; 956 957 /* 958 * If the device is marked ACTIVE, then this device is in use by another 959 * pool on the system. 960 */ 961 return (state == POOL_STATE_ACTIVE); 962} 963 964/* 965 * Initialize a vdev label. We check to make sure each leaf device is not in 966 * use, and writable. We put down an initial label which we will later 967 * overwrite with a complete label. Note that it's important to do this 968 * sequentially, not in parallel, so that we catch cases of multiple use of the 969 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 970 * itself. 971 */ 972int 973vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 974{ 975 spa_t *spa = vd->vdev_spa; 976 nvlist_t *label; 977 vdev_phys_t *vp; 978 abd_t *vp_abd; 979 abd_t *bootenv; 980 uberblock_t *ub; 981 abd_t *ub_abd; 982 zio_t *zio; 983 char *buf; 984 size_t buflen; 985 int error; 986 uint64_t spare_guid = 0, l2cache_guid = 0; 987 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 988 989 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 990 991 for (int c = 0; c < vd->vdev_children; c++) 992 if ((error = vdev_label_init(vd->vdev_child[c], 993 crtxg, reason)) != 0) 994 return (error); 995 996 /* Track the creation time for this vdev */ 997 vd->vdev_crtxg = crtxg; 998 999 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 1000 return (0); 1001 1002 /* 1003 * Dead vdevs cannot be initialized. 1004 */ 1005 if (vdev_is_dead(vd)) 1006 return (SET_ERROR(EIO)); 1007 1008 /* 1009 * Determine if the vdev is in use. 1010 */ 1011 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 1012 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 1013 return (SET_ERROR(EBUSY)); 1014 1015 /* 1016 * If this is a request to add or replace a spare or l2cache device 1017 * that is in use elsewhere on the system, then we must update the 1018 * guid (which was initialized to a random value) to reflect the 1019 * actual GUID (which is shared between multiple pools). 1020 */ 1021 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 1022 spare_guid != 0ULL) { 1023 uint64_t guid_delta = spare_guid - vd->vdev_guid; 1024 1025 vd->vdev_guid += guid_delta; 1026 1027 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1028 pvd->vdev_guid_sum += guid_delta; 1029 1030 /* 1031 * If this is a replacement, then we want to fallthrough to the 1032 * rest of the code. If we're adding a spare, then it's already 1033 * labeled appropriately and we can just return. 1034 */ 1035 if (reason == VDEV_LABEL_SPARE) 1036 return (0); 1037 ASSERT(reason == VDEV_LABEL_REPLACE || 1038 reason == VDEV_LABEL_SPLIT); 1039 } 1040 1041 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 1042 l2cache_guid != 0ULL) { 1043 uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 1044 1045 vd->vdev_guid += guid_delta; 1046 1047 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1048 pvd->vdev_guid_sum += guid_delta; 1049 1050 /* 1051 * If this is a replacement, then we want to fallthrough to the 1052 * rest of the code. If we're adding an l2cache, then it's 1053 * already labeled appropriately and we can just return. 1054 */ 1055 if (reason == VDEV_LABEL_L2CACHE) 1056 return (0); 1057 ASSERT(reason == VDEV_LABEL_REPLACE); 1058 } 1059 1060 /* 1061 * Initialize its label. 1062 */ 1063 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1064 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1065 vp = abd_to_buf(vp_abd); 1066 1067 /* 1068 * Generate a label describing the pool and our top-level vdev. 1069 * We mark it as being from txg 0 to indicate that it's not 1070 * really part of an active pool just yet. The labels will 1071 * be written again with a meaningful txg by spa_sync(). 1072 */ 1073 if (reason == VDEV_LABEL_SPARE || 1074 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 1075 /* 1076 * For inactive hot spares, we generate a special label that 1077 * identifies as a mutually shared hot spare. We write the 1078 * label if we are adding a hot spare, or if we are removing an 1079 * active hot spare (in which case we want to revert the 1080 * labels). 1081 */ 1082 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1083 1084 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 1085 spa_version(spa)) == 0); 1086 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1087 POOL_STATE_SPARE) == 0); 1088 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 1089 vd->vdev_guid) == 0); 1090 } else if (reason == VDEV_LABEL_L2CACHE || 1091 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 1092 /* 1093 * For level 2 ARC devices, add a special label. 1094 */ 1095 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1096 1097 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 1098 spa_version(spa)) == 0); 1099 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1100 POOL_STATE_L2CACHE) == 0); 1101 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 1102 vd->vdev_guid) == 0); 1103 } else { 1104 uint64_t txg = 0ULL; 1105 1106 if (reason == VDEV_LABEL_SPLIT) 1107 txg = spa->spa_uberblock.ub_txg; 1108 label = spa_config_generate(spa, vd, txg, B_FALSE); 1109 1110 /* 1111 * Add our creation time. This allows us to detect multiple 1112 * vdev uses as described above, and automatically expires if we 1113 * fail. 1114 */ 1115 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 1116 crtxg) == 0); 1117 } 1118 1119 buf = vp->vp_nvlist; 1120 buflen = sizeof (vp->vp_nvlist); 1121 1122 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 1123 if (error != 0) { 1124 nvlist_free(label); 1125 abd_free(vp_abd); 1126 /* EFAULT means nvlist_pack ran out of room */ 1127 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL)); 1128 } 1129 1130 /* 1131 * Initialize uberblock template. 1132 */ 1133 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE); 1134 abd_zero(ub_abd, VDEV_UBERBLOCK_RING); 1135 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t)); 1136 ub = abd_to_buf(ub_abd); 1137 ub->ub_txg = 0; 1138 1139 /* Initialize the 2nd padding area. */ 1140 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 1141 abd_zero(bootenv, VDEV_PAD_SIZE); 1142 1143 /* 1144 * Write everything in parallel. 1145 */ 1146retry: 1147 zio = zio_root(spa, NULL, NULL, flags); 1148 1149 for (int l = 0; l < VDEV_LABELS; l++) { 1150 1151 vdev_label_write(zio, vd, l, vp_abd, 1152 offsetof(vdev_label_t, vl_vdev_phys), 1153 sizeof (vdev_phys_t), NULL, NULL, flags); 1154 1155 /* 1156 * Skip the 1st padding area. 1157 * Zero out the 2nd padding area where it might have 1158 * left over data from previous filesystem format. 1159 */ 1160 vdev_label_write(zio, vd, l, bootenv, 1161 offsetof(vdev_label_t, vl_be), 1162 VDEV_PAD_SIZE, NULL, NULL, flags); 1163 1164 vdev_label_write(zio, vd, l, ub_abd, 1165 offsetof(vdev_label_t, vl_uberblock), 1166 VDEV_UBERBLOCK_RING, NULL, NULL, flags); 1167 } 1168 1169 error = zio_wait(zio); 1170 1171 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 1172 flags |= ZIO_FLAG_TRYHARD; 1173 goto retry; 1174 } 1175 1176 nvlist_free(label); 1177 abd_free(bootenv); 1178 abd_free(ub_abd); 1179 abd_free(vp_abd); 1180 1181 /* 1182 * If this vdev hasn't been previously identified as a spare, then we 1183 * mark it as such only if a) we are labeling it as a spare, or b) it 1184 * exists as a spare elsewhere in the system. Do the same for 1185 * level 2 ARC devices. 1186 */ 1187 if (error == 0 && !vd->vdev_isspare && 1188 (reason == VDEV_LABEL_SPARE || 1189 spa_spare_exists(vd->vdev_guid, NULL, NULL))) 1190 spa_spare_add(vd); 1191 1192 if (error == 0 && !vd->vdev_isl2cache && 1193 (reason == VDEV_LABEL_L2CACHE || 1194 spa_l2cache_exists(vd->vdev_guid, NULL))) 1195 spa_l2cache_add(vd); 1196 1197 return (error); 1198} 1199 1200/* 1201 * Done callback for vdev_label_read_bootenv_impl. If this is the first 1202 * callback to finish, store our abd in the callback pointer. Otherwise, we 1203 * just free our abd and return. 1204 */ 1205static void 1206vdev_label_read_bootenv_done(zio_t *zio) 1207{ 1208 zio_t *rio = zio->io_private; 1209 abd_t **cbp = rio->io_private; 1210 1211 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE); 1212 1213 if (zio->io_error == 0) { 1214 mutex_enter(&rio->io_lock); 1215 if (*cbp == NULL) { 1216 /* Will free this buffer in vdev_label_read_bootenv. */ 1217 *cbp = zio->io_abd; 1218 } else { 1219 abd_free(zio->io_abd); 1220 } 1221 mutex_exit(&rio->io_lock); 1222 } else { 1223 abd_free(zio->io_abd); 1224 } 1225} 1226 1227static void 1228vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags) 1229{ 1230 for (int c = 0; c < vd->vdev_children; c++) 1231 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags); 1232 1233 /* 1234 * We just use the first label that has a correct checksum; the 1235 * bootloader should have rewritten them all to be the same on boot, 1236 * and any changes we made since boot have been the same across all 1237 * labels. 1238 */ 1239 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1240 for (int l = 0; l < VDEV_LABELS; l++) { 1241 vdev_label_read(zio, vd, l, 1242 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE), 1243 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE, 1244 vdev_label_read_bootenv_done, zio, flags); 1245 } 1246 } 1247} 1248 1249int 1250vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv) 1251{ 1252 nvlist_t *config; 1253 spa_t *spa = rvd->vdev_spa; 1254 abd_t *abd = NULL; 1255 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1256 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1257 1258 ASSERT(bootenv); 1259 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1260 1261 zio_t *zio = zio_root(spa, NULL, &abd, flags); 1262 vdev_label_read_bootenv_impl(zio, rvd, flags); 1263 int err = zio_wait(zio); 1264 1265 if (abd != NULL) { 1266 char *buf; 1267 vdev_boot_envblock_t *vbe = abd_to_buf(abd); 1268 1269 vbe->vbe_version = ntohll(vbe->vbe_version); 1270 switch (vbe->vbe_version) { 1271 case VB_RAW: 1272 /* 1273 * if we have textual data in vbe_bootenv, create nvlist 1274 * with key "envmap". 1275 */ 1276 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW); 1277 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0'; 1278 fnvlist_add_string(bootenv, GRUB_ENVMAP, 1279 vbe->vbe_bootenv); 1280 break; 1281 1282 case VB_NVLIST: 1283 err = nvlist_unpack(vbe->vbe_bootenv, 1284 sizeof (vbe->vbe_bootenv), &config, 0); 1285 if (err == 0) { 1286 fnvlist_merge(bootenv, config); 1287 nvlist_free(config); 1288 break; 1289 } 1290 /* FALLTHROUGH */ 1291 default: 1292 /* Check for FreeBSD zfs bootonce command string */ 1293 buf = abd_to_buf(abd); 1294 if (*buf == '\0') { 1295 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, 1296 VB_NVLIST); 1297 break; 1298 } 1299 fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf); 1300 } 1301 1302 /* 1303 * abd was allocated in vdev_label_read_bootenv_impl() 1304 */ 1305 abd_free(abd); 1306 /* 1307 * If we managed to read any successfully, 1308 * return success. 1309 */ 1310 return (0); 1311 } 1312 return (err); 1313} 1314 1315int 1316vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env) 1317{ 1318 zio_t *zio; 1319 spa_t *spa = vd->vdev_spa; 1320 vdev_boot_envblock_t *bootenv; 1321 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1322 int error; 1323 size_t nvsize; 1324 char *nvbuf; 1325 1326 error = nvlist_size(env, &nvsize, NV_ENCODE_XDR); 1327 if (error != 0) 1328 return (SET_ERROR(error)); 1329 1330 if (nvsize >= sizeof (bootenv->vbe_bootenv)) { 1331 return (SET_ERROR(E2BIG)); 1332 } 1333 1334 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1335 1336 error = ENXIO; 1337 for (int c = 0; c < vd->vdev_children; c++) { 1338 int child_err; 1339 1340 child_err = vdev_label_write_bootenv(vd->vdev_child[c], env); 1341 /* 1342 * As long as any of the disks managed to write all of their 1343 * labels successfully, return success. 1344 */ 1345 if (child_err == 0) 1346 error = child_err; 1347 } 1348 1349 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) || 1350 !vdev_writeable(vd)) { 1351 return (error); 1352 } 1353 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE); 1354 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 1355 abd_zero(abd, VDEV_PAD_SIZE); 1356 1357 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE); 1358 nvbuf = bootenv->vbe_bootenv; 1359 nvsize = sizeof (bootenv->vbe_bootenv); 1360 1361 bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION); 1362 switch (bootenv->vbe_version) { 1363 case VB_RAW: 1364 if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) { 1365 (void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize); 1366 } 1367 error = 0; 1368 break; 1369 1370 case VB_NVLIST: 1371 error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR, 1372 KM_SLEEP); 1373 break; 1374 1375 default: 1376 error = EINVAL; 1377 break; 1378 } 1379 1380 if (error == 0) { 1381 bootenv->vbe_version = htonll(bootenv->vbe_version); 1382 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE); 1383 } else { 1384 abd_free(abd); 1385 return (SET_ERROR(error)); 1386 } 1387 1388retry: 1389 zio = zio_root(spa, NULL, NULL, flags); 1390 for (int l = 0; l < VDEV_LABELS; l++) { 1391 vdev_label_write(zio, vd, l, abd, 1392 offsetof(vdev_label_t, vl_be), 1393 VDEV_PAD_SIZE, NULL, NULL, flags); 1394 } 1395 1396 error = zio_wait(zio); 1397 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 1398 flags |= ZIO_FLAG_TRYHARD; 1399 goto retry; 1400 } 1401 1402 abd_free(abd); 1403 return (error); 1404} 1405 1406/* 1407 * ========================================================================== 1408 * uberblock load/sync 1409 * ========================================================================== 1410 */ 1411 1412/* 1413 * Consider the following situation: txg is safely synced to disk. We've 1414 * written the first uberblock for txg + 1, and then we lose power. When we 1415 * come back up, we fail to see the uberblock for txg + 1 because, say, 1416 * it was on a mirrored device and the replica to which we wrote txg + 1 1417 * is now offline. If we then make some changes and sync txg + 1, and then 1418 * the missing replica comes back, then for a few seconds we'll have two 1419 * conflicting uberblocks on disk with the same txg. The solution is simple: 1420 * among uberblocks with equal txg, choose the one with the latest timestamp. 1421 */ 1422static int 1423vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2) 1424{ 1425 int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg); 1426 1427 if (likely(cmp)) 1428 return (cmp); 1429 1430 cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp); 1431 if (likely(cmp)) 1432 return (cmp); 1433 1434 /* 1435 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware 1436 * ZFS, e.g. OpenZFS >= 0.7. 1437 * 1438 * If one ub has MMP and the other does not, they were written by 1439 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as 1440 * a 0 value. 1441 * 1442 * Since timestamp and txg are the same if we get this far, either is 1443 * acceptable for importing the pool. 1444 */ 1445 unsigned int seq1 = 0; 1446 unsigned int seq2 = 0; 1447 1448 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1)) 1449 seq1 = MMP_SEQ(ub1); 1450 1451 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2)) 1452 seq2 = MMP_SEQ(ub2); 1453 1454 return (TREE_CMP(seq1, seq2)); 1455} 1456 1457struct ubl_cbdata { 1458 uberblock_t *ubl_ubbest; /* Best uberblock */ 1459 vdev_t *ubl_vd; /* vdev associated with the above */ 1460}; 1461 1462static void 1463vdev_uberblock_load_done(zio_t *zio) 1464{ 1465 vdev_t *vd = zio->io_vd; 1466 spa_t *spa = zio->io_spa; 1467 zio_t *rio = zio->io_private; 1468 uberblock_t *ub = abd_to_buf(zio->io_abd); 1469 struct ubl_cbdata *cbp = rio->io_private; 1470 1471 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 1472 1473 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 1474 mutex_enter(&rio->io_lock); 1475 if (ub->ub_txg <= spa->spa_load_max_txg && 1476 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 1477 /* 1478 * Keep track of the vdev in which this uberblock 1479 * was found. We will use this information later 1480 * to obtain the config nvlist associated with 1481 * this uberblock. 1482 */ 1483 *cbp->ubl_ubbest = *ub; 1484 cbp->ubl_vd = vd; 1485 } 1486 mutex_exit(&rio->io_lock); 1487 } 1488 1489 abd_free(zio->io_abd); 1490} 1491 1492static void 1493vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 1494 struct ubl_cbdata *cbp) 1495{ 1496 for (int c = 0; c < vd->vdev_children; c++) 1497 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 1498 1499 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) && 1500 vd->vdev_ops != &vdev_draid_spare_ops) { 1501 for (int l = 0; l < VDEV_LABELS; l++) { 1502 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1503 vdev_label_read(zio, vd, l, 1504 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), 1505 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n), 1506 VDEV_UBERBLOCK_SIZE(vd), 1507 vdev_uberblock_load_done, zio, flags); 1508 } 1509 } 1510 } 1511} 1512 1513/* 1514 * Reads the 'best' uberblock from disk along with its associated 1515 * configuration. First, we read the uberblock array of each label of each 1516 * vdev, keeping track of the uberblock with the highest txg in each array. 1517 * Then, we read the configuration from the same vdev as the best uberblock. 1518 */ 1519void 1520vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 1521{ 1522 zio_t *zio; 1523 spa_t *spa = rvd->vdev_spa; 1524 struct ubl_cbdata cb; 1525 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1526 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1527 1528 ASSERT(ub); 1529 ASSERT(config); 1530 1531 bzero(ub, sizeof (uberblock_t)); 1532 *config = NULL; 1533 1534 cb.ubl_ubbest = ub; 1535 cb.ubl_vd = NULL; 1536 1537 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1538 zio = zio_root(spa, NULL, &cb, flags); 1539 vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1540 (void) zio_wait(zio); 1541 1542 /* 1543 * It's possible that the best uberblock was discovered on a label 1544 * that has a configuration which was written in a future txg. 1545 * Search all labels on this vdev to find the configuration that 1546 * matches the txg for our uberblock. 1547 */ 1548 if (cb.ubl_vd != NULL) { 1549 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. " 1550 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg); 1551 1552 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 1553 if (*config == NULL && spa->spa_extreme_rewind) { 1554 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. " 1555 "Trying again without txg restrictions."); 1556 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX); 1557 } 1558 if (*config == NULL) { 1559 vdev_dbgmsg(cb.ubl_vd, "failed to read label config"); 1560 } 1561 } 1562 spa_config_exit(spa, SCL_ALL, FTAG); 1563} 1564 1565/* 1566 * For use when a leaf vdev is expanded. 1567 * The location of labels 2 and 3 changed, and at the new location the 1568 * uberblock rings are either empty or contain garbage. The sync will write 1569 * new configs there because the vdev is dirty, but expansion also needs the 1570 * uberblock rings copied. Read them from label 0 which did not move. 1571 * 1572 * Since the point is to populate labels {2,3} with valid uberblocks, 1573 * we zero uberblocks we fail to read or which are not valid. 1574 */ 1575 1576static void 1577vdev_copy_uberblocks(vdev_t *vd) 1578{ 1579 abd_t *ub_abd; 1580 zio_t *write_zio; 1581 int locks = (SCL_L2ARC | SCL_ZIO); 1582 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1583 ZIO_FLAG_SPECULATIVE; 1584 1585 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) == 1586 SCL_STATE); 1587 ASSERT(vd->vdev_ops->vdev_op_leaf); 1588 1589 /* 1590 * No uberblocks are stored on distributed spares, they may be 1591 * safely skipped when expanding a leaf vdev. 1592 */ 1593 if (vd->vdev_ops == &vdev_draid_spare_ops) 1594 return; 1595 1596 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER); 1597 1598 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1599 1600 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags); 1601 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1602 const int src_label = 0; 1603 zio_t *zio; 1604 1605 zio = zio_root(vd->vdev_spa, NULL, NULL, flags); 1606 vdev_label_read(zio, vd, src_label, ub_abd, 1607 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1608 NULL, NULL, flags); 1609 1610 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd))) 1611 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1612 1613 for (int l = 2; l < VDEV_LABELS; l++) 1614 vdev_label_write(write_zio, vd, l, ub_abd, 1615 VDEV_UBERBLOCK_OFFSET(vd, n), 1616 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, 1617 flags | ZIO_FLAG_DONT_PROPAGATE); 1618 } 1619 (void) zio_wait(write_zio); 1620 1621 spa_config_exit(vd->vdev_spa, locks, FTAG); 1622 1623 abd_free(ub_abd); 1624} 1625 1626/* 1627 * On success, increment root zio's count of good writes. 1628 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1629 */ 1630static void 1631vdev_uberblock_sync_done(zio_t *zio) 1632{ 1633 uint64_t *good_writes = zio->io_private; 1634 1635 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1636 atomic_inc_64(good_writes); 1637} 1638 1639/* 1640 * Write the uberblock to all labels of all leaves of the specified vdev. 1641 */ 1642static void 1643vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes, 1644 uberblock_t *ub, vdev_t *vd, int flags) 1645{ 1646 for (uint64_t c = 0; c < vd->vdev_children; c++) { 1647 vdev_uberblock_sync(zio, good_writes, 1648 ub, vd->vdev_child[c], flags); 1649 } 1650 1651 if (!vd->vdev_ops->vdev_op_leaf) 1652 return; 1653 1654 if (!vdev_writeable(vd)) 1655 return; 1656 1657 /* 1658 * There's no need to write uberblocks to a distributed spare, they 1659 * are already stored on all the leaves of the parent dRAID. For 1660 * this same reason vdev_uberblock_load_impl() skips distributed 1661 * spares when reading uberblocks. 1662 */ 1663 if (vd->vdev_ops == &vdev_draid_spare_ops) 1664 return; 1665 1666 /* If the vdev was expanded, need to copy uberblock rings. */ 1667 if (vd->vdev_state == VDEV_STATE_HEALTHY && 1668 vd->vdev_copy_uberblocks == B_TRUE) { 1669 vdev_copy_uberblocks(vd); 1670 vd->vdev_copy_uberblocks = B_FALSE; 1671 } 1672 1673 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0; 1674 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m); 1675 1676 /* Copy the uberblock_t into the ABD */ 1677 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1678 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1679 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t)); 1680 1681 for (int l = 0; l < VDEV_LABELS; l++) 1682 vdev_label_write(zio, vd, l, ub_abd, 1683 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1684 vdev_uberblock_sync_done, good_writes, 1685 flags | ZIO_FLAG_DONT_PROPAGATE); 1686 1687 abd_free(ub_abd); 1688} 1689 1690/* Sync the uberblocks to all vdevs in svd[] */ 1691static int 1692vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1693{ 1694 spa_t *spa = svd[0]->vdev_spa; 1695 zio_t *zio; 1696 uint64_t good_writes = 0; 1697 1698 zio = zio_root(spa, NULL, NULL, flags); 1699 1700 for (int v = 0; v < svdcount; v++) 1701 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags); 1702 1703 (void) zio_wait(zio); 1704 1705 /* 1706 * Flush the uberblocks to disk. This ensures that the odd labels 1707 * are no longer needed (because the new uberblocks and the even 1708 * labels are safely on disk), so it is safe to overwrite them. 1709 */ 1710 zio = zio_root(spa, NULL, NULL, flags); 1711 1712 for (int v = 0; v < svdcount; v++) { 1713 if (vdev_writeable(svd[v])) { 1714 zio_flush(zio, svd[v]); 1715 } 1716 } 1717 1718 (void) zio_wait(zio); 1719 1720 return (good_writes >= 1 ? 0 : EIO); 1721} 1722 1723/* 1724 * On success, increment the count of good writes for our top-level vdev. 1725 */ 1726static void 1727vdev_label_sync_done(zio_t *zio) 1728{ 1729 uint64_t *good_writes = zio->io_private; 1730 1731 if (zio->io_error == 0) 1732 atomic_inc_64(good_writes); 1733} 1734 1735/* 1736 * If there weren't enough good writes, indicate failure to the parent. 1737 */ 1738static void 1739vdev_label_sync_top_done(zio_t *zio) 1740{ 1741 uint64_t *good_writes = zio->io_private; 1742 1743 if (*good_writes == 0) 1744 zio->io_error = SET_ERROR(EIO); 1745 1746 kmem_free(good_writes, sizeof (uint64_t)); 1747} 1748 1749/* 1750 * We ignore errors for log and cache devices, simply free the private data. 1751 */ 1752static void 1753vdev_label_sync_ignore_done(zio_t *zio) 1754{ 1755 kmem_free(zio->io_private, sizeof (uint64_t)); 1756} 1757 1758/* 1759 * Write all even or odd labels to all leaves of the specified vdev. 1760 */ 1761static void 1762vdev_label_sync(zio_t *zio, uint64_t *good_writes, 1763 vdev_t *vd, int l, uint64_t txg, int flags) 1764{ 1765 nvlist_t *label; 1766 vdev_phys_t *vp; 1767 abd_t *vp_abd; 1768 char *buf; 1769 size_t buflen; 1770 1771 for (int c = 0; c < vd->vdev_children; c++) { 1772 vdev_label_sync(zio, good_writes, 1773 vd->vdev_child[c], l, txg, flags); 1774 } 1775 1776 if (!vd->vdev_ops->vdev_op_leaf) 1777 return; 1778 1779 if (!vdev_writeable(vd)) 1780 return; 1781 1782 /* 1783 * The top-level config never needs to be written to a distributed 1784 * spare. When read vdev_dspare_label_read_config() will generate 1785 * the config for the vdev_label_read_config(). 1786 */ 1787 if (vd->vdev_ops == &vdev_draid_spare_ops) 1788 return; 1789 1790 /* 1791 * Generate a label describing the top-level config to which we belong. 1792 */ 1793 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1794 1795 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1796 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1797 vp = abd_to_buf(vp_abd); 1798 1799 buf = vp->vp_nvlist; 1800 buflen = sizeof (vp->vp_nvlist); 1801 1802 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) { 1803 for (; l < VDEV_LABELS; l += 2) { 1804 vdev_label_write(zio, vd, l, vp_abd, 1805 offsetof(vdev_label_t, vl_vdev_phys), 1806 sizeof (vdev_phys_t), 1807 vdev_label_sync_done, good_writes, 1808 flags | ZIO_FLAG_DONT_PROPAGATE); 1809 } 1810 } 1811 1812 abd_free(vp_abd); 1813 nvlist_free(label); 1814} 1815 1816static int 1817vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1818{ 1819 list_t *dl = &spa->spa_config_dirty_list; 1820 vdev_t *vd; 1821 zio_t *zio; 1822 int error; 1823 1824 /* 1825 * Write the new labels to disk. 1826 */ 1827 zio = zio_root(spa, NULL, NULL, flags); 1828 1829 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1830 uint64_t *good_writes; 1831 1832 ASSERT(!vd->vdev_ishole); 1833 1834 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 1835 zio_t *vio = zio_null(zio, spa, NULL, 1836 (vd->vdev_islog || vd->vdev_aux != NULL) ? 1837 vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1838 good_writes, flags); 1839 vdev_label_sync(vio, good_writes, vd, l, txg, flags); 1840 zio_nowait(vio); 1841 } 1842 1843 error = zio_wait(zio); 1844 1845 /* 1846 * Flush the new labels to disk. 1847 */ 1848 zio = zio_root(spa, NULL, NULL, flags); 1849 1850 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1851 zio_flush(zio, vd); 1852 1853 (void) zio_wait(zio); 1854 1855 return (error); 1856} 1857 1858/* 1859 * Sync the uberblock and any changes to the vdev configuration. 1860 * 1861 * The order of operations is carefully crafted to ensure that 1862 * if the system panics or loses power at any time, the state on disk 1863 * is still transactionally consistent. The in-line comments below 1864 * describe the failure semantics at each stage. 1865 * 1866 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1867 * at any time, you can just call it again, and it will resume its work. 1868 */ 1869int 1870vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1871{ 1872 spa_t *spa = svd[0]->vdev_spa; 1873 uberblock_t *ub = &spa->spa_uberblock; 1874 int error = 0; 1875 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1876 1877 ASSERT(svdcount != 0); 1878retry: 1879 /* 1880 * Normally, we don't want to try too hard to write every label and 1881 * uberblock. If there is a flaky disk, we don't want the rest of the 1882 * sync process to block while we retry. But if we can't write a 1883 * single label out, we should retry with ZIO_FLAG_TRYHARD before 1884 * bailing out and declaring the pool faulted. 1885 */ 1886 if (error != 0) { 1887 if ((flags & ZIO_FLAG_TRYHARD) != 0) 1888 return (error); 1889 flags |= ZIO_FLAG_TRYHARD; 1890 } 1891 1892 ASSERT(ub->ub_txg <= txg); 1893 1894 /* 1895 * If this isn't a resync due to I/O errors, 1896 * and nothing changed in this transaction group, 1897 * and the vdev configuration hasn't changed, 1898 * then there's nothing to do. 1899 */ 1900 if (ub->ub_txg < txg) { 1901 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev, 1902 txg, spa->spa_mmp.mmp_delay); 1903 1904 if (!changed && list_is_empty(&spa->spa_config_dirty_list)) 1905 return (0); 1906 } 1907 1908 if (txg > spa_freeze_txg(spa)) 1909 return (0); 1910 1911 ASSERT(txg <= spa->spa_final_txg); 1912 1913 /* 1914 * Flush the write cache of every disk that's been written to 1915 * in this transaction group. This ensures that all blocks 1916 * written in this txg will be committed to stable storage 1917 * before any uberblock that references them. 1918 */ 1919 zio_t *zio = zio_root(spa, NULL, NULL, flags); 1920 1921 for (vdev_t *vd = 1922 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL; 1923 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1924 zio_flush(zio, vd); 1925 1926 (void) zio_wait(zio); 1927 1928 /* 1929 * Sync out the even labels (L0, L2) for every dirty vdev. If the 1930 * system dies in the middle of this process, that's OK: all of the 1931 * even labels that made it to disk will be newer than any uberblock, 1932 * and will therefore be considered invalid. The odd labels (L1, L3), 1933 * which have not yet been touched, will still be valid. We flush 1934 * the new labels to disk to ensure that all even-label updates 1935 * are committed to stable storage before the uberblock update. 1936 */ 1937 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) { 1938 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1939 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1940 "for pool '%s' when syncing out the even labels " 1941 "of dirty vdevs", error, spa_name(spa)); 1942 } 1943 goto retry; 1944 } 1945 1946 /* 1947 * Sync the uberblocks to all vdevs in svd[]. 1948 * If the system dies in the middle of this step, there are two cases 1949 * to consider, and the on-disk state is consistent either way: 1950 * 1951 * (1) If none of the new uberblocks made it to disk, then the 1952 * previous uberblock will be the newest, and the odd labels 1953 * (which had not yet been touched) will be valid with respect 1954 * to that uberblock. 1955 * 1956 * (2) If one or more new uberblocks made it to disk, then they 1957 * will be the newest, and the even labels (which had all 1958 * been successfully committed) will be valid with respect 1959 * to the new uberblocks. 1960 */ 1961 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) { 1962 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1963 zfs_dbgmsg("vdev_uberblock_sync_list() returned error " 1964 "%d for pool '%s'", error, spa_name(spa)); 1965 } 1966 goto retry; 1967 } 1968 1969 if (spa_multihost(spa)) 1970 mmp_update_uberblock(spa, ub); 1971 1972 /* 1973 * Sync out odd labels for every dirty vdev. If the system dies 1974 * in the middle of this process, the even labels and the new 1975 * uberblocks will suffice to open the pool. The next time 1976 * the pool is opened, the first thing we'll do -- before any 1977 * user data is modified -- is mark every vdev dirty so that 1978 * all labels will be brought up to date. We flush the new labels 1979 * to disk to ensure that all odd-label updates are committed to 1980 * stable storage before the next transaction group begins. 1981 */ 1982 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) { 1983 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1984 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1985 "for pool '%s' when syncing out the odd labels of " 1986 "dirty vdevs", error, spa_name(spa)); 1987 } 1988 goto retry; 1989 } 1990 1991 return (0); 1992} 1993