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 */ 160 uint64_t 161 vdev_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 */ 173 int 174 vdev_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 } 185 186 static void 187 vdev_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 } 200 201 void 202 vdev_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 */ 219 void 220 vdev_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 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE, 260 vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]); 261 262 /* ZIOs pending */ 263 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE, 264 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]); 265 266 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE, 267 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]); 268 269 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE, 270 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]); 271 272 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE, 273 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]); 274 275 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE, 276 vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]); 277 278 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE, 279 vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]); 280 281 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE, 282 vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]); 283 284 /* Histograms */ 285 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO, 286 vsx->vsx_total_histo[ZIO_TYPE_READ], 287 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ])); 288 289 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO, 290 vsx->vsx_total_histo[ZIO_TYPE_WRITE], 291 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE])); 292 293 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO, 294 vsx->vsx_disk_histo[ZIO_TYPE_READ], 295 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ])); 296 297 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO, 298 vsx->vsx_disk_histo[ZIO_TYPE_WRITE], 299 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE])); 300 301 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO, 302 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ], 303 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ])); 304 305 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO, 306 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE], 307 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE])); 308 309 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO, 310 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ], 311 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ])); 312 313 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO, 314 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE], 315 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE])); 316 317 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO, 318 vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB], 319 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB])); 320 321 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO, 322 vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM], 323 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM])); 324 325 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO, 326 vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD], 327 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD])); 328 329 /* Request sizes */ 330 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO, 331 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ], 332 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ])); 333 334 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO, 335 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE], 336 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE])); 337 338 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO, 339 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ], 340 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ])); 341 342 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO, 343 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE], 344 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE])); 345 346 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO, 347 vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB], 348 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB])); 349 350 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO, 351 vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM], 352 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM])); 353 354 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO, 355 vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD], 356 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD])); 357 358 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO, 359 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ], 360 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ])); 361 362 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO, 363 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE], 364 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE])); 365 366 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO, 367 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ], 368 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ])); 369 370 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO, 371 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE], 372 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE])); 373 374 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO, 375 vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB], 376 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB])); 377 378 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO, 379 vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM], 380 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM])); 381 382 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO, 383 vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD], 384 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD])); 385 386 /* IO delays */ 387 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios); 388 389 /* Add extended stats nvlist to main nvlist */ 390 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx); 391 392 fnvlist_free(nvx); 393 kmem_free(vs, sizeof (*vs)); 394 kmem_free(vsx, sizeof (*vsx)); 395 } 396 397 static void 398 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 399 { 400 spa_t *spa = vd->vdev_spa; 401 402 if (vd != spa->spa_root_vdev) 403 return; 404 405 /* provide either current or previous scan information */ 406 pool_scan_stat_t ps; 407 if (spa_scan_get_stats(spa, &ps) == 0) { 408 fnvlist_add_uint64_array(nvl, 409 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 410 sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 411 } 412 413 pool_removal_stat_t prs; 414 if (spa_removal_get_stats(spa, &prs) == 0) { 415 fnvlist_add_uint64_array(nvl, 416 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs, 417 sizeof (prs) / sizeof (uint64_t)); 418 } 419 420 pool_checkpoint_stat_t pcs; 421 if (spa_checkpoint_get_stats(spa, &pcs) == 0) { 422 fnvlist_add_uint64_array(nvl, 423 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs, 424 sizeof (pcs) / sizeof (uint64_t)); 425 } 426 } 427 428 static void 429 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 430 { 431 if (vd == vd->vdev_top) { 432 vdev_rebuild_stat_t vrs; 433 if (vdev_rebuild_get_stats(vd, &vrs) == 0) { 434 fnvlist_add_uint64_array(nvl, 435 ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs, 436 sizeof (vrs) / sizeof (uint64_t)); 437 } 438 } 439 } 440 441 /* 442 * Generate the nvlist representing this vdev's config. 443 */ 444 nvlist_t * 445 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 446 vdev_config_flag_t flags) 447 { 448 nvlist_t *nv = NULL; 449 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 450 451 nv = fnvlist_alloc(); 452 453 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 454 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 455 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 456 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 457 458 if (vd->vdev_path != NULL) 459 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 460 461 if (vd->vdev_devid != NULL) 462 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 463 464 if (vd->vdev_physpath != NULL) 465 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 466 vd->vdev_physpath); 467 468 if (vd->vdev_enc_sysfs_path != NULL) 469 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, 470 vd->vdev_enc_sysfs_path); 471 472 if (vd->vdev_fru != NULL) 473 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 474 475 if (vd->vdev_ops->vdev_op_config_generate != NULL) 476 vd->vdev_ops->vdev_op_config_generate(vd, nv); 477 478 if (vd->vdev_wholedisk != -1ULL) { 479 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 480 vd->vdev_wholedisk); 481 } 482 483 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING)) 484 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 485 486 if (vd->vdev_isspare) 487 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 488 489 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 490 vd == vd->vdev_top) { 491 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 492 vd->vdev_ms_array); 493 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 494 vd->vdev_ms_shift); 495 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 496 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 497 vd->vdev_asize); 498 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 499 if (vd->vdev_removing) { 500 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 501 vd->vdev_removing); 502 } 503 504 /* zpool command expects alloc class data */ 505 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) { 506 const char *bias = NULL; 507 508 switch (vd->vdev_alloc_bias) { 509 case VDEV_BIAS_LOG: 510 bias = VDEV_ALLOC_BIAS_LOG; 511 break; 512 case VDEV_BIAS_SPECIAL: 513 bias = VDEV_ALLOC_BIAS_SPECIAL; 514 break; 515 case VDEV_BIAS_DEDUP: 516 bias = VDEV_ALLOC_BIAS_DEDUP; 517 break; 518 default: 519 ASSERT3U(vd->vdev_alloc_bias, ==, 520 VDEV_BIAS_NONE); 521 } 522 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 523 bias); 524 } 525 } 526 527 if (vd->vdev_dtl_sm != NULL) { 528 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 529 space_map_object(vd->vdev_dtl_sm)); 530 } 531 532 if (vic->vic_mapping_object != 0) { 533 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 534 vic->vic_mapping_object); 535 } 536 537 if (vic->vic_births_object != 0) { 538 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 539 vic->vic_births_object); 540 } 541 542 if (vic->vic_prev_indirect_vdev != UINT64_MAX) { 543 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 544 vic->vic_prev_indirect_vdev); 545 } 546 547 if (vd->vdev_crtxg) 548 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 549 550 if (vd->vdev_expansion_time) 551 fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME, 552 vd->vdev_expansion_time); 553 554 if (flags & VDEV_CONFIG_MOS) { 555 if (vd->vdev_leaf_zap != 0) { 556 ASSERT(vd->vdev_ops->vdev_op_leaf); 557 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 558 vd->vdev_leaf_zap); 559 } 560 561 if (vd->vdev_top_zap != 0) { 562 ASSERT(vd == vd->vdev_top); 563 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 564 vd->vdev_top_zap); 565 } 566 567 if (vd->vdev_resilver_deferred) { 568 ASSERT(vd->vdev_ops->vdev_op_leaf); 569 ASSERT(spa->spa_resilver_deferred); 570 fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER); 571 } 572 } 573 574 if (getstats) { 575 vdev_config_generate_stats(vd, nv); 576 577 root_vdev_actions_getprogress(vd, nv); 578 top_vdev_actions_getprogress(vd, nv); 579 580 /* 581 * Note: this can be called from open context 582 * (spa_get_stats()), so we need the rwlock to prevent 583 * the mapping from being changed by condensing. 584 */ 585 rw_enter(&vd->vdev_indirect_rwlock, RW_READER); 586 if (vd->vdev_indirect_mapping != NULL) { 587 ASSERT(vd->vdev_indirect_births != NULL); 588 vdev_indirect_mapping_t *vim = 589 vd->vdev_indirect_mapping; 590 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 591 vdev_indirect_mapping_size(vim)); 592 } 593 rw_exit(&vd->vdev_indirect_rwlock); 594 if (vd->vdev_mg != NULL && 595 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) { 596 /* 597 * Compute approximately how much memory would be used 598 * for the indirect mapping if this device were to 599 * be removed. 600 * 601 * Note: If the frag metric is invalid, then not 602 * enough metaslabs have been converted to have 603 * histograms. 604 */ 605 uint64_t seg_count = 0; 606 uint64_t to_alloc = vd->vdev_stat.vs_alloc; 607 608 /* 609 * There are the same number of allocated segments 610 * as free segments, so we will have at least one 611 * entry per free segment. However, small free 612 * segments (smaller than vdev_removal_max_span) 613 * will be combined with adjacent allocated segments 614 * as a single mapping. 615 */ 616 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 617 if (i + 1 < highbit64(vdev_removal_max_span) 618 - 1) { 619 to_alloc += 620 vd->vdev_mg->mg_histogram[i] << 621 (i + 1); 622 } else { 623 seg_count += 624 vd->vdev_mg->mg_histogram[i]; 625 } 626 } 627 628 /* 629 * The maximum length of a mapping is 630 * zfs_remove_max_segment, so we need at least one entry 631 * per zfs_remove_max_segment of allocated data. 632 */ 633 seg_count += to_alloc / spa_remove_max_segment(spa); 634 635 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 636 seg_count * 637 sizeof (vdev_indirect_mapping_entry_phys_t)); 638 } 639 } 640 641 if (!vd->vdev_ops->vdev_op_leaf) { 642 nvlist_t **child; 643 int c, idx; 644 645 ASSERT(!vd->vdev_ishole); 646 647 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 648 KM_SLEEP); 649 650 for (c = 0, idx = 0; c < vd->vdev_children; c++) { 651 vdev_t *cvd = vd->vdev_child[c]; 652 653 /* 654 * If we're generating an nvlist of removing 655 * vdevs then skip over any device which is 656 * not being removed. 657 */ 658 if ((flags & VDEV_CONFIG_REMOVING) && 659 !cvd->vdev_removing) 660 continue; 661 662 child[idx++] = vdev_config_generate(spa, cvd, 663 getstats, flags); 664 } 665 666 if (idx) { 667 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 668 child, idx); 669 } 670 671 for (c = 0; c < idx; c++) 672 nvlist_free(child[c]); 673 674 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 675 676 } else { 677 const char *aux = NULL; 678 679 if (vd->vdev_offline && !vd->vdev_tmpoffline) 680 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 681 if (vd->vdev_resilver_txg != 0) 682 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 683 vd->vdev_resilver_txg); 684 if (vd->vdev_rebuild_txg != 0) 685 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG, 686 vd->vdev_rebuild_txg); 687 if (vd->vdev_faulted) 688 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 689 if (vd->vdev_degraded) 690 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 691 if (vd->vdev_removed) 692 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 693 if (vd->vdev_unspare) 694 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 695 if (vd->vdev_ishole) 696 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 697 698 /* Set the reason why we're FAULTED/DEGRADED. */ 699 switch (vd->vdev_stat.vs_aux) { 700 case VDEV_AUX_ERR_EXCEEDED: 701 aux = "err_exceeded"; 702 break; 703 704 case VDEV_AUX_EXTERNAL: 705 aux = "external"; 706 break; 707 } 708 709 if (aux != NULL && !vd->vdev_tmpoffline) { 710 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 711 } else { 712 /* 713 * We're healthy - clear any previous AUX_STATE values. 714 */ 715 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE)) 716 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE); 717 } 718 719 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 720 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 721 vd->vdev_orig_guid); 722 } 723 } 724 725 return (nv); 726 } 727 728 /* 729 * Generate a view of the top-level vdevs. If we currently have holes 730 * in the namespace, then generate an array which contains a list of holey 731 * vdevs. Additionally, add the number of top-level children that currently 732 * exist. 733 */ 734 void 735 vdev_top_config_generate(spa_t *spa, nvlist_t *config) 736 { 737 vdev_t *rvd = spa->spa_root_vdev; 738 uint64_t *array; 739 uint_t c, idx; 740 741 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 742 743 for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 744 vdev_t *tvd = rvd->vdev_child[c]; 745 746 if (tvd->vdev_ishole) { 747 array[idx++] = c; 748 } 749 } 750 751 if (idx) { 752 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 753 array, idx) == 0); 754 } 755 756 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 757 rvd->vdev_children) == 0); 758 759 kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 760 } 761 762 /* 763 * Returns the configuration from the label of the given vdev. For vdevs 764 * which don't have a txg value stored on their label (i.e. spares/cache) 765 * or have not been completely initialized (txg = 0) just return 766 * the configuration from the first valid label we find. Otherwise, 767 * find the most up-to-date label that does not exceed the specified 768 * 'txg' value. 769 */ 770 nvlist_t * 771 vdev_label_read_config(vdev_t *vd, uint64_t txg) 772 { 773 spa_t *spa = vd->vdev_spa; 774 nvlist_t *config = NULL; 775 vdev_phys_t *vp[VDEV_LABELS]; 776 abd_t *vp_abd[VDEV_LABELS]; 777 zio_t *zio[VDEV_LABELS]; 778 uint64_t best_txg = 0; 779 uint64_t label_txg = 0; 780 int error = 0; 781 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 782 ZIO_FLAG_SPECULATIVE; 783 784 ASSERT(vd->vdev_validate_thread == curthread || 785 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 786 787 if (!vdev_readable(vd)) 788 return (NULL); 789 790 /* 791 * The label for a dRAID distributed spare is not stored on disk. 792 * Instead it is generated when needed which allows us to bypass 793 * the pipeline when reading the config from the label. 794 */ 795 if (vd->vdev_ops == &vdev_draid_spare_ops) 796 return (vdev_draid_read_config_spare(vd)); 797 798 for (int l = 0; l < VDEV_LABELS; l++) { 799 vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 800 vp[l] = abd_to_buf(vp_abd[l]); 801 } 802 803 retry: 804 for (int l = 0; l < VDEV_LABELS; l++) { 805 zio[l] = zio_root(spa, NULL, NULL, flags); 806 807 vdev_label_read(zio[l], vd, l, vp_abd[l], 808 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 809 NULL, NULL, flags); 810 } 811 for (int l = 0; l < VDEV_LABELS; l++) { 812 nvlist_t *label = NULL; 813 814 if (zio_wait(zio[l]) == 0 && 815 nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist), 816 &label, 0) == 0) { 817 /* 818 * Auxiliary vdevs won't have txg values in their 819 * labels and newly added vdevs may not have been 820 * completely initialized so just return the 821 * configuration from the first valid label we 822 * encounter. 823 */ 824 error = nvlist_lookup_uint64(label, 825 ZPOOL_CONFIG_POOL_TXG, &label_txg); 826 if ((error || label_txg == 0) && !config) { 827 config = label; 828 for (l++; l < VDEV_LABELS; l++) 829 zio_wait(zio[l]); 830 break; 831 } else if (label_txg <= txg && label_txg > best_txg) { 832 best_txg = label_txg; 833 nvlist_free(config); 834 config = fnvlist_dup(label); 835 } 836 } 837 838 if (label != NULL) { 839 nvlist_free(label); 840 label = NULL; 841 } 842 } 843 844 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 845 flags |= ZIO_FLAG_TRYHARD; 846 goto retry; 847 } 848 849 /* 850 * We found a valid label but it didn't pass txg restrictions. 851 */ 852 if (config == NULL && label_txg != 0) { 853 vdev_dbgmsg(vd, "label discarded as txg is too large " 854 "(%llu > %llu)", (u_longlong_t)label_txg, 855 (u_longlong_t)txg); 856 } 857 858 for (int l = 0; l < VDEV_LABELS; l++) { 859 abd_free(vp_abd[l]); 860 } 861 862 return (config); 863 } 864 865 /* 866 * Determine if a device is in use. The 'spare_guid' parameter will be filled 867 * in with the device guid if this spare is active elsewhere on the system. 868 */ 869 static boolean_t 870 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 871 uint64_t *spare_guid, uint64_t *l2cache_guid) 872 { 873 spa_t *spa = vd->vdev_spa; 874 uint64_t state, pool_guid, device_guid, txg, spare_pool; 875 uint64_t vdtxg = 0; 876 nvlist_t *label; 877 878 if (spare_guid) 879 *spare_guid = 0ULL; 880 if (l2cache_guid) 881 *l2cache_guid = 0ULL; 882 883 /* 884 * Read the label, if any, and perform some basic sanity checks. 885 */ 886 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 887 return (B_FALSE); 888 889 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 890 &vdtxg); 891 892 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 893 &state) != 0 || 894 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 895 &device_guid) != 0) { 896 nvlist_free(label); 897 return (B_FALSE); 898 } 899 900 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 901 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 902 &pool_guid) != 0 || 903 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 904 &txg) != 0)) { 905 nvlist_free(label); 906 return (B_FALSE); 907 } 908 909 nvlist_free(label); 910 911 /* 912 * Check to see if this device indeed belongs to the pool it claims to 913 * be a part of. The only way this is allowed is if the device is a hot 914 * spare (which we check for later on). 915 */ 916 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 917 !spa_guid_exists(pool_guid, device_guid) && 918 !spa_spare_exists(device_guid, NULL, NULL) && 919 !spa_l2cache_exists(device_guid, NULL)) 920 return (B_FALSE); 921 922 /* 923 * If the transaction group is zero, then this an initialized (but 924 * unused) label. This is only an error if the create transaction 925 * on-disk is the same as the one we're using now, in which case the 926 * user has attempted to add the same vdev multiple times in the same 927 * transaction. 928 */ 929 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 930 txg == 0 && vdtxg == crtxg) 931 return (B_TRUE); 932 933 /* 934 * Check to see if this is a spare device. We do an explicit check for 935 * spa_has_spare() here because it may be on our pending list of spares 936 * to add. We also check if it is an l2cache device. 937 */ 938 if (spa_spare_exists(device_guid, &spare_pool, NULL) || 939 spa_has_spare(spa, device_guid)) { 940 if (spare_guid) 941 *spare_guid = device_guid; 942 943 switch (reason) { 944 case VDEV_LABEL_CREATE: 945 case VDEV_LABEL_L2CACHE: 946 return (B_TRUE); 947 948 case VDEV_LABEL_REPLACE: 949 return (!spa_has_spare(spa, device_guid) || 950 spare_pool != 0ULL); 951 952 case VDEV_LABEL_SPARE: 953 return (spa_has_spare(spa, device_guid)); 954 default: 955 break; 956 } 957 } 958 959 /* 960 * Check to see if this is an l2cache device. 961 */ 962 if (spa_l2cache_exists(device_guid, NULL)) 963 return (B_TRUE); 964 965 /* 966 * We can't rely on a pool's state if it's been imported 967 * read-only. Instead we look to see if the pools is marked 968 * read-only in the namespace and set the state to active. 969 */ 970 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 971 (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 972 spa_mode(spa) == SPA_MODE_READ) 973 state = POOL_STATE_ACTIVE; 974 975 /* 976 * If the device is marked ACTIVE, then this device is in use by another 977 * pool on the system. 978 */ 979 return (state == POOL_STATE_ACTIVE); 980 } 981 982 /* 983 * Initialize a vdev label. We check to make sure each leaf device is not in 984 * use, and writable. We put down an initial label which we will later 985 * overwrite with a complete label. Note that it's important to do this 986 * sequentially, not in parallel, so that we catch cases of multiple use of the 987 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 988 * itself. 989 */ 990 int 991 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 992 { 993 spa_t *spa = vd->vdev_spa; 994 nvlist_t *label; 995 vdev_phys_t *vp; 996 abd_t *vp_abd; 997 abd_t *bootenv; 998 uberblock_t *ub; 999 abd_t *ub_abd; 1000 zio_t *zio; 1001 char *buf; 1002 size_t buflen; 1003 int error; 1004 uint64_t spare_guid = 0, l2cache_guid = 0; 1005 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1006 1007 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1008 1009 for (int c = 0; c < vd->vdev_children; c++) 1010 if ((error = vdev_label_init(vd->vdev_child[c], 1011 crtxg, reason)) != 0) 1012 return (error); 1013 1014 /* Track the creation time for this vdev */ 1015 vd->vdev_crtxg = crtxg; 1016 1017 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 1018 return (0); 1019 1020 /* 1021 * Dead vdevs cannot be initialized. 1022 */ 1023 if (vdev_is_dead(vd)) 1024 return (SET_ERROR(EIO)); 1025 1026 /* 1027 * Determine if the vdev is in use. 1028 */ 1029 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 1030 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 1031 return (SET_ERROR(EBUSY)); 1032 1033 /* 1034 * If this is a request to add or replace a spare or l2cache device 1035 * that is in use elsewhere on the system, then we must update the 1036 * guid (which was initialized to a random value) to reflect the 1037 * actual GUID (which is shared between multiple pools). 1038 */ 1039 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 1040 spare_guid != 0ULL) { 1041 uint64_t guid_delta = spare_guid - vd->vdev_guid; 1042 1043 vd->vdev_guid += guid_delta; 1044 1045 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1046 pvd->vdev_guid_sum += guid_delta; 1047 1048 /* 1049 * If this is a replacement, then we want to fallthrough to the 1050 * rest of the code. If we're adding a spare, then it's already 1051 * labeled appropriately and we can just return. 1052 */ 1053 if (reason == VDEV_LABEL_SPARE) 1054 return (0); 1055 ASSERT(reason == VDEV_LABEL_REPLACE || 1056 reason == VDEV_LABEL_SPLIT); 1057 } 1058 1059 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 1060 l2cache_guid != 0ULL) { 1061 uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 1062 1063 vd->vdev_guid += guid_delta; 1064 1065 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1066 pvd->vdev_guid_sum += guid_delta; 1067 1068 /* 1069 * If this is a replacement, then we want to fallthrough to the 1070 * rest of the code. If we're adding an l2cache, then it's 1071 * already labeled appropriately and we can just return. 1072 */ 1073 if (reason == VDEV_LABEL_L2CACHE) 1074 return (0); 1075 ASSERT(reason == VDEV_LABEL_REPLACE); 1076 } 1077 1078 /* 1079 * Initialize its label. 1080 */ 1081 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1082 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1083 vp = abd_to_buf(vp_abd); 1084 1085 /* 1086 * Generate a label describing the pool and our top-level vdev. 1087 * We mark it as being from txg 0 to indicate that it's not 1088 * really part of an active pool just yet. The labels will 1089 * be written again with a meaningful txg by spa_sync(). 1090 */ 1091 if (reason == VDEV_LABEL_SPARE || 1092 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 1093 /* 1094 * For inactive hot spares, we generate a special label that 1095 * identifies as a mutually shared hot spare. We write the 1096 * label if we are adding a hot spare, or if we are removing an 1097 * active hot spare (in which case we want to revert the 1098 * labels). 1099 */ 1100 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1101 1102 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 1103 spa_version(spa)) == 0); 1104 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1105 POOL_STATE_SPARE) == 0); 1106 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 1107 vd->vdev_guid) == 0); 1108 } else if (reason == VDEV_LABEL_L2CACHE || 1109 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 1110 /* 1111 * For level 2 ARC devices, add a special label. 1112 */ 1113 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1114 1115 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 1116 spa_version(spa)) == 0); 1117 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1118 POOL_STATE_L2CACHE) == 0); 1119 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 1120 vd->vdev_guid) == 0); 1121 } else { 1122 uint64_t txg = 0ULL; 1123 1124 if (reason == VDEV_LABEL_SPLIT) 1125 txg = spa->spa_uberblock.ub_txg; 1126 label = spa_config_generate(spa, vd, txg, B_FALSE); 1127 1128 /* 1129 * Add our creation time. This allows us to detect multiple 1130 * vdev uses as described above, and automatically expires if we 1131 * fail. 1132 */ 1133 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 1134 crtxg) == 0); 1135 } 1136 1137 buf = vp->vp_nvlist; 1138 buflen = sizeof (vp->vp_nvlist); 1139 1140 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 1141 if (error != 0) { 1142 nvlist_free(label); 1143 abd_free(vp_abd); 1144 /* EFAULT means nvlist_pack ran out of room */ 1145 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL)); 1146 } 1147 1148 /* 1149 * Initialize uberblock template. 1150 */ 1151 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE); 1152 abd_zero(ub_abd, VDEV_UBERBLOCK_RING); 1153 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t)); 1154 ub = abd_to_buf(ub_abd); 1155 ub->ub_txg = 0; 1156 1157 /* Initialize the 2nd padding area. */ 1158 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 1159 abd_zero(bootenv, VDEV_PAD_SIZE); 1160 1161 /* 1162 * Write everything in parallel. 1163 */ 1164 retry: 1165 zio = zio_root(spa, NULL, NULL, flags); 1166 1167 for (int l = 0; l < VDEV_LABELS; l++) { 1168 1169 vdev_label_write(zio, vd, l, vp_abd, 1170 offsetof(vdev_label_t, vl_vdev_phys), 1171 sizeof (vdev_phys_t), NULL, NULL, flags); 1172 1173 /* 1174 * Skip the 1st padding area. 1175 * Zero out the 2nd padding area where it might have 1176 * left over data from previous filesystem format. 1177 */ 1178 vdev_label_write(zio, vd, l, bootenv, 1179 offsetof(vdev_label_t, vl_be), 1180 VDEV_PAD_SIZE, NULL, NULL, flags); 1181 1182 vdev_label_write(zio, vd, l, ub_abd, 1183 offsetof(vdev_label_t, vl_uberblock), 1184 VDEV_UBERBLOCK_RING, NULL, NULL, flags); 1185 } 1186 1187 error = zio_wait(zio); 1188 1189 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 1190 flags |= ZIO_FLAG_TRYHARD; 1191 goto retry; 1192 } 1193 1194 nvlist_free(label); 1195 abd_free(bootenv); 1196 abd_free(ub_abd); 1197 abd_free(vp_abd); 1198 1199 /* 1200 * If this vdev hasn't been previously identified as a spare, then we 1201 * mark it as such only if a) we are labeling it as a spare, or b) it 1202 * exists as a spare elsewhere in the system. Do the same for 1203 * level 2 ARC devices. 1204 */ 1205 if (error == 0 && !vd->vdev_isspare && 1206 (reason == VDEV_LABEL_SPARE || 1207 spa_spare_exists(vd->vdev_guid, NULL, NULL))) 1208 spa_spare_add(vd); 1209 1210 if (error == 0 && !vd->vdev_isl2cache && 1211 (reason == VDEV_LABEL_L2CACHE || 1212 spa_l2cache_exists(vd->vdev_guid, NULL))) 1213 spa_l2cache_add(vd); 1214 1215 return (error); 1216 } 1217 1218 /* 1219 * Done callback for vdev_label_read_bootenv_impl. If this is the first 1220 * callback to finish, store our abd in the callback pointer. Otherwise, we 1221 * just free our abd and return. 1222 */ 1223 static void 1224 vdev_label_read_bootenv_done(zio_t *zio) 1225 { 1226 zio_t *rio = zio->io_private; 1227 abd_t **cbp = rio->io_private; 1228 1229 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE); 1230 1231 if (zio->io_error == 0) { 1232 mutex_enter(&rio->io_lock); 1233 if (*cbp == NULL) { 1234 /* Will free this buffer in vdev_label_read_bootenv. */ 1235 *cbp = zio->io_abd; 1236 } else { 1237 abd_free(zio->io_abd); 1238 } 1239 mutex_exit(&rio->io_lock); 1240 } else { 1241 abd_free(zio->io_abd); 1242 } 1243 } 1244 1245 static void 1246 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags) 1247 { 1248 for (int c = 0; c < vd->vdev_children; c++) 1249 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags); 1250 1251 /* 1252 * We just use the first label that has a correct checksum; the 1253 * bootloader should have rewritten them all to be the same on boot, 1254 * and any changes we made since boot have been the same across all 1255 * labels. 1256 */ 1257 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1258 for (int l = 0; l < VDEV_LABELS; l++) { 1259 vdev_label_read(zio, vd, l, 1260 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE), 1261 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE, 1262 vdev_label_read_bootenv_done, zio, flags); 1263 } 1264 } 1265 } 1266 1267 int 1268 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv) 1269 { 1270 nvlist_t *config; 1271 spa_t *spa = rvd->vdev_spa; 1272 abd_t *abd = NULL; 1273 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1274 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1275 1276 ASSERT(bootenv); 1277 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1278 1279 zio_t *zio = zio_root(spa, NULL, &abd, flags); 1280 vdev_label_read_bootenv_impl(zio, rvd, flags); 1281 int err = zio_wait(zio); 1282 1283 if (abd != NULL) { 1284 char *buf; 1285 vdev_boot_envblock_t *vbe = abd_to_buf(abd); 1286 1287 vbe->vbe_version = ntohll(vbe->vbe_version); 1288 switch (vbe->vbe_version) { 1289 case VB_RAW: 1290 /* 1291 * if we have textual data in vbe_bootenv, create nvlist 1292 * with key "envmap". 1293 */ 1294 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW); 1295 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0'; 1296 fnvlist_add_string(bootenv, GRUB_ENVMAP, 1297 vbe->vbe_bootenv); 1298 break; 1299 1300 case VB_NVLIST: 1301 err = nvlist_unpack(vbe->vbe_bootenv, 1302 sizeof (vbe->vbe_bootenv), &config, 0); 1303 if (err == 0) { 1304 fnvlist_merge(bootenv, config); 1305 nvlist_free(config); 1306 break; 1307 } 1308 fallthrough; 1309 default: 1310 /* Check for FreeBSD zfs bootonce command string */ 1311 buf = abd_to_buf(abd); 1312 if (*buf == '\0') { 1313 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, 1314 VB_NVLIST); 1315 break; 1316 } 1317 fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf); 1318 } 1319 1320 /* 1321 * abd was allocated in vdev_label_read_bootenv_impl() 1322 */ 1323 abd_free(abd); 1324 /* 1325 * If we managed to read any successfully, 1326 * return success. 1327 */ 1328 return (0); 1329 } 1330 return (err); 1331 } 1332 1333 int 1334 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env) 1335 { 1336 zio_t *zio; 1337 spa_t *spa = vd->vdev_spa; 1338 vdev_boot_envblock_t *bootenv; 1339 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1340 int error; 1341 size_t nvsize; 1342 char *nvbuf; 1343 1344 error = nvlist_size(env, &nvsize, NV_ENCODE_XDR); 1345 if (error != 0) 1346 return (SET_ERROR(error)); 1347 1348 if (nvsize >= sizeof (bootenv->vbe_bootenv)) { 1349 return (SET_ERROR(E2BIG)); 1350 } 1351 1352 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1353 1354 error = ENXIO; 1355 for (int c = 0; c < vd->vdev_children; c++) { 1356 int child_err; 1357 1358 child_err = vdev_label_write_bootenv(vd->vdev_child[c], env); 1359 /* 1360 * As long as any of the disks managed to write all of their 1361 * labels successfully, return success. 1362 */ 1363 if (child_err == 0) 1364 error = child_err; 1365 } 1366 1367 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) || 1368 !vdev_writeable(vd)) { 1369 return (error); 1370 } 1371 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE); 1372 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 1373 abd_zero(abd, VDEV_PAD_SIZE); 1374 1375 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE); 1376 nvbuf = bootenv->vbe_bootenv; 1377 nvsize = sizeof (bootenv->vbe_bootenv); 1378 1379 bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION); 1380 switch (bootenv->vbe_version) { 1381 case VB_RAW: 1382 if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) { 1383 (void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize); 1384 } 1385 error = 0; 1386 break; 1387 1388 case VB_NVLIST: 1389 error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR, 1390 KM_SLEEP); 1391 break; 1392 1393 default: 1394 error = EINVAL; 1395 break; 1396 } 1397 1398 if (error == 0) { 1399 bootenv->vbe_version = htonll(bootenv->vbe_version); 1400 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE); 1401 } else { 1402 abd_free(abd); 1403 return (SET_ERROR(error)); 1404 } 1405 1406 retry: 1407 zio = zio_root(spa, NULL, NULL, flags); 1408 for (int l = 0; l < VDEV_LABELS; l++) { 1409 vdev_label_write(zio, vd, l, abd, 1410 offsetof(vdev_label_t, vl_be), 1411 VDEV_PAD_SIZE, NULL, NULL, flags); 1412 } 1413 1414 error = zio_wait(zio); 1415 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 1416 flags |= ZIO_FLAG_TRYHARD; 1417 goto retry; 1418 } 1419 1420 abd_free(abd); 1421 return (error); 1422 } 1423 1424 /* 1425 * ========================================================================== 1426 * uberblock load/sync 1427 * ========================================================================== 1428 */ 1429 1430 /* 1431 * Consider the following situation: txg is safely synced to disk. We've 1432 * written the first uberblock for txg + 1, and then we lose power. When we 1433 * come back up, we fail to see the uberblock for txg + 1 because, say, 1434 * it was on a mirrored device and the replica to which we wrote txg + 1 1435 * is now offline. If we then make some changes and sync txg + 1, and then 1436 * the missing replica comes back, then for a few seconds we'll have two 1437 * conflicting uberblocks on disk with the same txg. The solution is simple: 1438 * among uberblocks with equal txg, choose the one with the latest timestamp. 1439 */ 1440 static int 1441 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2) 1442 { 1443 int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg); 1444 1445 if (likely(cmp)) 1446 return (cmp); 1447 1448 cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp); 1449 if (likely(cmp)) 1450 return (cmp); 1451 1452 /* 1453 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware 1454 * ZFS, e.g. OpenZFS >= 0.7. 1455 * 1456 * If one ub has MMP and the other does not, they were written by 1457 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as 1458 * a 0 value. 1459 * 1460 * Since timestamp and txg are the same if we get this far, either is 1461 * acceptable for importing the pool. 1462 */ 1463 unsigned int seq1 = 0; 1464 unsigned int seq2 = 0; 1465 1466 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1)) 1467 seq1 = MMP_SEQ(ub1); 1468 1469 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2)) 1470 seq2 = MMP_SEQ(ub2); 1471 1472 return (TREE_CMP(seq1, seq2)); 1473 } 1474 1475 struct ubl_cbdata { 1476 uberblock_t *ubl_ubbest; /* Best uberblock */ 1477 vdev_t *ubl_vd; /* vdev associated with the above */ 1478 }; 1479 1480 static void 1481 vdev_uberblock_load_done(zio_t *zio) 1482 { 1483 vdev_t *vd = zio->io_vd; 1484 spa_t *spa = zio->io_spa; 1485 zio_t *rio = zio->io_private; 1486 uberblock_t *ub = abd_to_buf(zio->io_abd); 1487 struct ubl_cbdata *cbp = rio->io_private; 1488 1489 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 1490 1491 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 1492 mutex_enter(&rio->io_lock); 1493 if (ub->ub_txg <= spa->spa_load_max_txg && 1494 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 1495 /* 1496 * Keep track of the vdev in which this uberblock 1497 * was found. We will use this information later 1498 * to obtain the config nvlist associated with 1499 * this uberblock. 1500 */ 1501 *cbp->ubl_ubbest = *ub; 1502 cbp->ubl_vd = vd; 1503 } 1504 mutex_exit(&rio->io_lock); 1505 } 1506 1507 abd_free(zio->io_abd); 1508 } 1509 1510 static void 1511 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 1512 struct ubl_cbdata *cbp) 1513 { 1514 for (int c = 0; c < vd->vdev_children; c++) 1515 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 1516 1517 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) && 1518 vd->vdev_ops != &vdev_draid_spare_ops) { 1519 for (int l = 0; l < VDEV_LABELS; l++) { 1520 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1521 vdev_label_read(zio, vd, l, 1522 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), 1523 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n), 1524 VDEV_UBERBLOCK_SIZE(vd), 1525 vdev_uberblock_load_done, zio, flags); 1526 } 1527 } 1528 } 1529 } 1530 1531 /* 1532 * Reads the 'best' uberblock from disk along with its associated 1533 * configuration. First, we read the uberblock array of each label of each 1534 * vdev, keeping track of the uberblock with the highest txg in each array. 1535 * Then, we read the configuration from the same vdev as the best uberblock. 1536 */ 1537 void 1538 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 1539 { 1540 zio_t *zio; 1541 spa_t *spa = rvd->vdev_spa; 1542 struct ubl_cbdata cb; 1543 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1544 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1545 1546 ASSERT(ub); 1547 ASSERT(config); 1548 1549 bzero(ub, sizeof (uberblock_t)); 1550 *config = NULL; 1551 1552 cb.ubl_ubbest = ub; 1553 cb.ubl_vd = NULL; 1554 1555 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1556 zio = zio_root(spa, NULL, &cb, flags); 1557 vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1558 (void) zio_wait(zio); 1559 1560 /* 1561 * It's possible that the best uberblock was discovered on a label 1562 * that has a configuration which was written in a future txg. 1563 * Search all labels on this vdev to find the configuration that 1564 * matches the txg for our uberblock. 1565 */ 1566 if (cb.ubl_vd != NULL) { 1567 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. " 1568 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg); 1569 1570 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 1571 if (*config == NULL && spa->spa_extreme_rewind) { 1572 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. " 1573 "Trying again without txg restrictions."); 1574 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX); 1575 } 1576 if (*config == NULL) { 1577 vdev_dbgmsg(cb.ubl_vd, "failed to read label config"); 1578 } 1579 } 1580 spa_config_exit(spa, SCL_ALL, FTAG); 1581 } 1582 1583 /* 1584 * For use when a leaf vdev is expanded. 1585 * The location of labels 2 and 3 changed, and at the new location the 1586 * uberblock rings are either empty or contain garbage. The sync will write 1587 * new configs there because the vdev is dirty, but expansion also needs the 1588 * uberblock rings copied. Read them from label 0 which did not move. 1589 * 1590 * Since the point is to populate labels {2,3} with valid uberblocks, 1591 * we zero uberblocks we fail to read or which are not valid. 1592 */ 1593 1594 static void 1595 vdev_copy_uberblocks(vdev_t *vd) 1596 { 1597 abd_t *ub_abd; 1598 zio_t *write_zio; 1599 int locks = (SCL_L2ARC | SCL_ZIO); 1600 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1601 ZIO_FLAG_SPECULATIVE; 1602 1603 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) == 1604 SCL_STATE); 1605 ASSERT(vd->vdev_ops->vdev_op_leaf); 1606 1607 /* 1608 * No uberblocks are stored on distributed spares, they may be 1609 * safely skipped when expanding a leaf vdev. 1610 */ 1611 if (vd->vdev_ops == &vdev_draid_spare_ops) 1612 return; 1613 1614 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER); 1615 1616 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1617 1618 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags); 1619 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1620 const int src_label = 0; 1621 zio_t *zio; 1622 1623 zio = zio_root(vd->vdev_spa, NULL, NULL, flags); 1624 vdev_label_read(zio, vd, src_label, ub_abd, 1625 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1626 NULL, NULL, flags); 1627 1628 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd))) 1629 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1630 1631 for (int l = 2; l < VDEV_LABELS; l++) 1632 vdev_label_write(write_zio, vd, l, ub_abd, 1633 VDEV_UBERBLOCK_OFFSET(vd, n), 1634 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, 1635 flags | ZIO_FLAG_DONT_PROPAGATE); 1636 } 1637 (void) zio_wait(write_zio); 1638 1639 spa_config_exit(vd->vdev_spa, locks, FTAG); 1640 1641 abd_free(ub_abd); 1642 } 1643 1644 /* 1645 * On success, increment root zio's count of good writes. 1646 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1647 */ 1648 static void 1649 vdev_uberblock_sync_done(zio_t *zio) 1650 { 1651 uint64_t *good_writes = zio->io_private; 1652 1653 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1654 atomic_inc_64(good_writes); 1655 } 1656 1657 /* 1658 * Write the uberblock to all labels of all leaves of the specified vdev. 1659 */ 1660 static void 1661 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes, 1662 uberblock_t *ub, vdev_t *vd, int flags) 1663 { 1664 for (uint64_t c = 0; c < vd->vdev_children; c++) { 1665 vdev_uberblock_sync(zio, good_writes, 1666 ub, vd->vdev_child[c], flags); 1667 } 1668 1669 if (!vd->vdev_ops->vdev_op_leaf) 1670 return; 1671 1672 if (!vdev_writeable(vd)) 1673 return; 1674 1675 /* 1676 * There's no need to write uberblocks to a distributed spare, they 1677 * are already stored on all the leaves of the parent dRAID. For 1678 * this same reason vdev_uberblock_load_impl() skips distributed 1679 * spares when reading uberblocks. 1680 */ 1681 if (vd->vdev_ops == &vdev_draid_spare_ops) 1682 return; 1683 1684 /* If the vdev was expanded, need to copy uberblock rings. */ 1685 if (vd->vdev_state == VDEV_STATE_HEALTHY && 1686 vd->vdev_copy_uberblocks == B_TRUE) { 1687 vdev_copy_uberblocks(vd); 1688 vd->vdev_copy_uberblocks = B_FALSE; 1689 } 1690 1691 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0; 1692 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m); 1693 1694 /* Copy the uberblock_t into the ABD */ 1695 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1696 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1697 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t)); 1698 1699 for (int l = 0; l < VDEV_LABELS; l++) 1700 vdev_label_write(zio, vd, l, ub_abd, 1701 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1702 vdev_uberblock_sync_done, good_writes, 1703 flags | ZIO_FLAG_DONT_PROPAGATE); 1704 1705 abd_free(ub_abd); 1706 } 1707 1708 /* Sync the uberblocks to all vdevs in svd[] */ 1709 static int 1710 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1711 { 1712 spa_t *spa = svd[0]->vdev_spa; 1713 zio_t *zio; 1714 uint64_t good_writes = 0; 1715 1716 zio = zio_root(spa, NULL, NULL, flags); 1717 1718 for (int v = 0; v < svdcount; v++) 1719 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags); 1720 1721 (void) zio_wait(zio); 1722 1723 /* 1724 * Flush the uberblocks to disk. This ensures that the odd labels 1725 * are no longer needed (because the new uberblocks and the even 1726 * labels are safely on disk), so it is safe to overwrite them. 1727 */ 1728 zio = zio_root(spa, NULL, NULL, flags); 1729 1730 for (int v = 0; v < svdcount; v++) { 1731 if (vdev_writeable(svd[v])) { 1732 zio_flush(zio, svd[v]); 1733 } 1734 } 1735 1736 (void) zio_wait(zio); 1737 1738 return (good_writes >= 1 ? 0 : EIO); 1739 } 1740 1741 /* 1742 * On success, increment the count of good writes for our top-level vdev. 1743 */ 1744 static void 1745 vdev_label_sync_done(zio_t *zio) 1746 { 1747 uint64_t *good_writes = zio->io_private; 1748 1749 if (zio->io_error == 0) 1750 atomic_inc_64(good_writes); 1751 } 1752 1753 /* 1754 * If there weren't enough good writes, indicate failure to the parent. 1755 */ 1756 static void 1757 vdev_label_sync_top_done(zio_t *zio) 1758 { 1759 uint64_t *good_writes = zio->io_private; 1760 1761 if (*good_writes == 0) 1762 zio->io_error = SET_ERROR(EIO); 1763 1764 kmem_free(good_writes, sizeof (uint64_t)); 1765 } 1766 1767 /* 1768 * We ignore errors for log and cache devices, simply free the private data. 1769 */ 1770 static void 1771 vdev_label_sync_ignore_done(zio_t *zio) 1772 { 1773 kmem_free(zio->io_private, sizeof (uint64_t)); 1774 } 1775 1776 /* 1777 * Write all even or odd labels to all leaves of the specified vdev. 1778 */ 1779 static void 1780 vdev_label_sync(zio_t *zio, uint64_t *good_writes, 1781 vdev_t *vd, int l, uint64_t txg, int flags) 1782 { 1783 nvlist_t *label; 1784 vdev_phys_t *vp; 1785 abd_t *vp_abd; 1786 char *buf; 1787 size_t buflen; 1788 1789 for (int c = 0; c < vd->vdev_children; c++) { 1790 vdev_label_sync(zio, good_writes, 1791 vd->vdev_child[c], l, txg, flags); 1792 } 1793 1794 if (!vd->vdev_ops->vdev_op_leaf) 1795 return; 1796 1797 if (!vdev_writeable(vd)) 1798 return; 1799 1800 /* 1801 * The top-level config never needs to be written to a distributed 1802 * spare. When read vdev_dspare_label_read_config() will generate 1803 * the config for the vdev_label_read_config(). 1804 */ 1805 if (vd->vdev_ops == &vdev_draid_spare_ops) 1806 return; 1807 1808 /* 1809 * Generate a label describing the top-level config to which we belong. 1810 */ 1811 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1812 1813 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1814 abd_zero(vp_abd, sizeof (vdev_phys_t)); 1815 vp = abd_to_buf(vp_abd); 1816 1817 buf = vp->vp_nvlist; 1818 buflen = sizeof (vp->vp_nvlist); 1819 1820 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) { 1821 for (; l < VDEV_LABELS; l += 2) { 1822 vdev_label_write(zio, vd, l, vp_abd, 1823 offsetof(vdev_label_t, vl_vdev_phys), 1824 sizeof (vdev_phys_t), 1825 vdev_label_sync_done, good_writes, 1826 flags | ZIO_FLAG_DONT_PROPAGATE); 1827 } 1828 } 1829 1830 abd_free(vp_abd); 1831 nvlist_free(label); 1832 } 1833 1834 static int 1835 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1836 { 1837 list_t *dl = &spa->spa_config_dirty_list; 1838 vdev_t *vd; 1839 zio_t *zio; 1840 int error; 1841 1842 /* 1843 * Write the new labels to disk. 1844 */ 1845 zio = zio_root(spa, NULL, NULL, flags); 1846 1847 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1848 uint64_t *good_writes; 1849 1850 ASSERT(!vd->vdev_ishole); 1851 1852 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 1853 zio_t *vio = zio_null(zio, spa, NULL, 1854 (vd->vdev_islog || vd->vdev_aux != NULL) ? 1855 vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1856 good_writes, flags); 1857 vdev_label_sync(vio, good_writes, vd, l, txg, flags); 1858 zio_nowait(vio); 1859 } 1860 1861 error = zio_wait(zio); 1862 1863 /* 1864 * Flush the new labels to disk. 1865 */ 1866 zio = zio_root(spa, NULL, NULL, flags); 1867 1868 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1869 zio_flush(zio, vd); 1870 1871 (void) zio_wait(zio); 1872 1873 return (error); 1874 } 1875 1876 /* 1877 * Sync the uberblock and any changes to the vdev configuration. 1878 * 1879 * The order of operations is carefully crafted to ensure that 1880 * if the system panics or loses power at any time, the state on disk 1881 * is still transactionally consistent. The in-line comments below 1882 * describe the failure semantics at each stage. 1883 * 1884 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1885 * at any time, you can just call it again, and it will resume its work. 1886 */ 1887 int 1888 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1889 { 1890 spa_t *spa = svd[0]->vdev_spa; 1891 uberblock_t *ub = &spa->spa_uberblock; 1892 int error = 0; 1893 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1894 1895 ASSERT(svdcount != 0); 1896 retry: 1897 /* 1898 * Normally, we don't want to try too hard to write every label and 1899 * uberblock. If there is a flaky disk, we don't want the rest of the 1900 * sync process to block while we retry. But if we can't write a 1901 * single label out, we should retry with ZIO_FLAG_TRYHARD before 1902 * bailing out and declaring the pool faulted. 1903 */ 1904 if (error != 0) { 1905 if ((flags & ZIO_FLAG_TRYHARD) != 0) 1906 return (error); 1907 flags |= ZIO_FLAG_TRYHARD; 1908 } 1909 1910 ASSERT(ub->ub_txg <= txg); 1911 1912 /* 1913 * If this isn't a resync due to I/O errors, 1914 * and nothing changed in this transaction group, 1915 * and the vdev configuration hasn't changed, 1916 * then there's nothing to do. 1917 */ 1918 if (ub->ub_txg < txg) { 1919 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev, 1920 txg, spa->spa_mmp.mmp_delay); 1921 1922 if (!changed && list_is_empty(&spa->spa_config_dirty_list)) 1923 return (0); 1924 } 1925 1926 if (txg > spa_freeze_txg(spa)) 1927 return (0); 1928 1929 ASSERT(txg <= spa->spa_final_txg); 1930 1931 /* 1932 * Flush the write cache of every disk that's been written to 1933 * in this transaction group. This ensures that all blocks 1934 * written in this txg will be committed to stable storage 1935 * before any uberblock that references them. 1936 */ 1937 zio_t *zio = zio_root(spa, NULL, NULL, flags); 1938 1939 for (vdev_t *vd = 1940 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL; 1941 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1942 zio_flush(zio, vd); 1943 1944 (void) zio_wait(zio); 1945 1946 /* 1947 * Sync out the even labels (L0, L2) for every dirty vdev. If the 1948 * system dies in the middle of this process, that's OK: all of the 1949 * even labels that made it to disk will be newer than any uberblock, 1950 * and will therefore be considered invalid. The odd labels (L1, L3), 1951 * which have not yet been touched, will still be valid. We flush 1952 * the new labels to disk to ensure that all even-label updates 1953 * are committed to stable storage before the uberblock update. 1954 */ 1955 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) { 1956 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1957 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 1958 "for pool '%s' when syncing out the even labels " 1959 "of dirty vdevs", error, spa_name(spa)); 1960 } 1961 goto retry; 1962 } 1963 1964 /* 1965 * Sync the uberblocks to all vdevs in svd[]. 1966 * If the system dies in the middle of this step, there are two cases 1967 * to consider, and the on-disk state is consistent either way: 1968 * 1969 * (1) If none of the new uberblocks made it to disk, then the 1970 * previous uberblock will be the newest, and the odd labels 1971 * (which had not yet been touched) will be valid with respect 1972 * to that uberblock. 1973 * 1974 * (2) If one or more new uberblocks made it to disk, then they 1975 * will be the newest, and the even labels (which had all 1976 * been successfully committed) will be valid with respect 1977 * to the new uberblocks. 1978 */ 1979 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) { 1980 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 1981 zfs_dbgmsg("vdev_uberblock_sync_list() returned error " 1982 "%d for pool '%s'", error, spa_name(spa)); 1983 } 1984 goto retry; 1985 } 1986 1987 if (spa_multihost(spa)) 1988 mmp_update_uberblock(spa, ub); 1989 1990 /* 1991 * Sync out odd labels for every dirty vdev. If the system dies 1992 * in the middle of this process, the even labels and the new 1993 * uberblocks will suffice to open the pool. The next time 1994 * the pool is opened, the first thing we'll do -- before any 1995 * user data is modified -- is mark every vdev dirty so that 1996 * all labels will be brought up to date. We flush the new labels 1997 * to disk to ensure that all odd-label updates are committed to 1998 * stable storage before the next transaction group begins. 1999 */ 2000 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) { 2001 if ((flags & ZIO_FLAG_TRYHARD) != 0) { 2002 zfs_dbgmsg("vdev_label_sync_list() returned error %d " 2003 "for pool '%s' when syncing out the odd labels of " 2004 "dirty vdevs", error, spa_name(spa)); 2005 } 2006 goto retry; 2007 } 2008 2009 return (0); 2010 } 2011