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 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 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 * 128 * Each leaf device label also contains the following: 129 * 130 * top_guid Unique ID for top-level vdev in which this is contained 131 * guid Unique ID for the leaf vdev 132 * 133 * The 'vs' configuration follows the format described in 'spa_config.c'. 134 */ 135 136 #include <sys/zfs_context.h> 137 #include <sys/spa.h> 138 #include <sys/spa_impl.h> 139 #include <sys/dmu.h> 140 #include <sys/zap.h> 141 #include <sys/vdev.h> 142 #include <sys/vdev_impl.h> 143 #include <sys/uberblock_impl.h> 144 #include <sys/metaslab.h> 145 #include <sys/zio.h> 146 #include <sys/fs/zfs.h> 147 148 /* 149 * Basic routines to read and write from a vdev label. 150 * Used throughout the rest of this file. 151 */ 152 uint64_t 153 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 154 { 155 ASSERT(offset < sizeof (vdev_label_t)); 156 157 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 158 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 159 } 160 161 static void 162 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 163 uint64_t size, zio_done_func_t *done, void *private) 164 { 165 ASSERT(vd->vdev_children == 0); 166 167 zio_nowait(zio_read_phys(zio, vd, 168 vdev_label_offset(vd->vdev_psize, l, offset), 169 size, buf, ZIO_CHECKSUM_LABEL, done, private, 170 ZIO_PRIORITY_SYNC_READ, 171 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE)); 172 } 173 174 static void 175 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 176 uint64_t size, zio_done_func_t *done, void *private) 177 { 178 ASSERT(vd->vdev_children == 0); 179 180 zio_nowait(zio_write_phys(zio, vd, 181 vdev_label_offset(vd->vdev_psize, l, offset), 182 size, buf, ZIO_CHECKSUM_LABEL, done, private, 183 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL)); 184 } 185 186 /* 187 * Generate the nvlist representing this vdev's config. 188 */ 189 nvlist_t * 190 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 191 boolean_t isspare) 192 { 193 nvlist_t *nv = NULL; 194 195 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 196 197 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 198 vd->vdev_ops->vdev_op_type) == 0); 199 if (!isspare) 200 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 201 == 0); 202 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 203 204 if (vd->vdev_path != NULL) 205 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 206 vd->vdev_path) == 0); 207 208 if (vd->vdev_devid != NULL) 209 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 210 vd->vdev_devid) == 0); 211 212 if (vd->vdev_physpath != NULL) 213 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 214 vd->vdev_physpath) == 0); 215 216 if (vd->vdev_nparity != 0) { 217 ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 218 VDEV_TYPE_RAIDZ) == 0); 219 220 /* 221 * Make sure someone hasn't managed to sneak a fancy new vdev 222 * into a crufty old storage pool. 223 */ 224 ASSERT(vd->vdev_nparity == 1 || 225 (vd->vdev_nparity == 2 && 226 spa_version(spa) >= ZFS_VERSION_RAID6)); 227 228 /* 229 * Note that we'll add the nparity tag even on storage pools 230 * that only support a single parity device -- older software 231 * will just ignore it. 232 */ 233 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 234 vd->vdev_nparity) == 0); 235 } 236 237 if (vd->vdev_wholedisk != -1ULL) 238 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 239 vd->vdev_wholedisk) == 0); 240 241 if (vd->vdev_not_present) 242 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 243 244 if (vd->vdev_isspare) 245 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 246 247 if (!isspare && vd == vd->vdev_top) { 248 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 249 vd->vdev_ms_array) == 0); 250 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 251 vd->vdev_ms_shift) == 0); 252 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 253 vd->vdev_ashift) == 0); 254 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 255 vd->vdev_asize) == 0); 256 } 257 258 if (vd->vdev_dtl.smo_object != 0) 259 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 260 vd->vdev_dtl.smo_object) == 0); 261 262 if (getstats) { 263 vdev_stat_t vs; 264 vdev_get_stats(vd, &vs); 265 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 266 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 267 } 268 269 if (!vd->vdev_ops->vdev_op_leaf) { 270 nvlist_t **child; 271 int c; 272 273 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 274 KM_SLEEP); 275 276 for (c = 0; c < vd->vdev_children; c++) 277 child[c] = vdev_config_generate(spa, vd->vdev_child[c], 278 getstats, isspare); 279 280 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 281 child, vd->vdev_children) == 0); 282 283 for (c = 0; c < vd->vdev_children; c++) 284 nvlist_free(child[c]); 285 286 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 287 288 } else { 289 if (vd->vdev_offline && !vd->vdev_tmpoffline) 290 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 291 B_TRUE) == 0); 292 if (vd->vdev_faulted) 293 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 294 B_TRUE) == 0); 295 if (vd->vdev_degraded) 296 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 297 B_TRUE) == 0); 298 if (vd->vdev_removed) 299 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 300 B_TRUE) == 0); 301 if (vd->vdev_unspare) 302 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 303 B_TRUE) == 0); 304 } 305 306 return (nv); 307 } 308 309 nvlist_t * 310 vdev_label_read_config(vdev_t *vd) 311 { 312 spa_t *spa = vd->vdev_spa; 313 nvlist_t *config = NULL; 314 vdev_phys_t *vp; 315 zio_t *zio; 316 int l; 317 318 ASSERT(spa_config_held(spa, RW_READER)); 319 320 if (vdev_is_dead(vd)) 321 return (NULL); 322 323 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 324 325 for (l = 0; l < VDEV_LABELS; l++) { 326 327 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | 328 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD); 329 330 vdev_label_read(zio, vd, l, vp, 331 offsetof(vdev_label_t, vl_vdev_phys), 332 sizeof (vdev_phys_t), NULL, NULL); 333 334 if (zio_wait(zio) == 0 && 335 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 336 &config, 0) == 0) 337 break; 338 339 if (config != NULL) { 340 nvlist_free(config); 341 config = NULL; 342 } 343 } 344 345 zio_buf_free(vp, sizeof (vdev_phys_t)); 346 347 return (config); 348 } 349 350 /* 351 * Determine if a device is in use. The 'spare_guid' parameter will be filled 352 * in with the device guid if this spare is active elsewhere on the system. 353 */ 354 static boolean_t 355 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 356 uint64_t *spare_guid) 357 { 358 spa_t *spa = vd->vdev_spa; 359 uint64_t state, pool_guid, device_guid, txg, spare_pool; 360 uint64_t vdtxg = 0; 361 nvlist_t *label; 362 363 if (spare_guid) 364 *spare_guid = 0ULL; 365 366 /* 367 * Read the label, if any, and perform some basic sanity checks. 368 */ 369 if ((label = vdev_label_read_config(vd)) == NULL) 370 return (B_FALSE); 371 372 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 373 &vdtxg); 374 375 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 376 &state) != 0 || 377 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 378 &device_guid) != 0) { 379 nvlist_free(label); 380 return (B_FALSE); 381 } 382 383 if (state != POOL_STATE_SPARE && 384 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 385 &pool_guid) != 0 || 386 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 387 &txg) != 0)) { 388 nvlist_free(label); 389 return (B_FALSE); 390 } 391 392 nvlist_free(label); 393 394 /* 395 * Check to see if this device indeed belongs to the pool it claims to 396 * be a part of. The only way this is allowed is if the device is a hot 397 * spare (which we check for later on). 398 */ 399 if (state != POOL_STATE_SPARE && 400 !spa_guid_exists(pool_guid, device_guid) && 401 !spa_spare_exists(device_guid, NULL)) 402 return (B_FALSE); 403 404 /* 405 * If the transaction group is zero, then this an initialized (but 406 * unused) label. This is only an error if the create transaction 407 * on-disk is the same as the one we're using now, in which case the 408 * user has attempted to add the same vdev multiple times in the same 409 * transaction. 410 */ 411 if (state != POOL_STATE_SPARE && txg == 0 && vdtxg == crtxg) 412 return (B_TRUE); 413 414 /* 415 * Check to see if this is a spare device. We do an explicit check for 416 * spa_has_spare() here because it may be on our pending list of spares 417 * to add. 418 */ 419 if (spa_spare_exists(device_guid, &spare_pool) || 420 spa_has_spare(spa, device_guid)) { 421 if (spare_guid) 422 *spare_guid = device_guid; 423 424 switch (reason) { 425 case VDEV_LABEL_CREATE: 426 return (B_TRUE); 427 428 case VDEV_LABEL_REPLACE: 429 return (!spa_has_spare(spa, device_guid) || 430 spare_pool != 0ULL); 431 432 case VDEV_LABEL_SPARE: 433 return (spa_has_spare(spa, device_guid)); 434 } 435 } 436 437 /* 438 * If the device is marked ACTIVE, then this device is in use by another 439 * pool on the system. 440 */ 441 return (state == POOL_STATE_ACTIVE); 442 } 443 444 /* 445 * Initialize a vdev label. We check to make sure each leaf device is not in 446 * use, and writable. We put down an initial label which we will later 447 * overwrite with a complete label. Note that it's important to do this 448 * sequentially, not in parallel, so that we catch cases of multiple use of the 449 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 450 * itself. 451 */ 452 int 453 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 454 { 455 spa_t *spa = vd->vdev_spa; 456 nvlist_t *label; 457 vdev_phys_t *vp; 458 vdev_boot_header_t *vb; 459 uberblock_t *ub; 460 zio_t *zio; 461 int l, c, n; 462 char *buf; 463 size_t buflen; 464 int error; 465 uint64_t spare_guid; 466 467 ASSERT(spa_config_held(spa, RW_WRITER)); 468 469 for (c = 0; c < vd->vdev_children; c++) 470 if ((error = vdev_label_init(vd->vdev_child[c], 471 crtxg, reason)) != 0) 472 return (error); 473 474 if (!vd->vdev_ops->vdev_op_leaf) 475 return (0); 476 477 /* 478 * Dead vdevs cannot be initialized. 479 */ 480 if (vdev_is_dead(vd)) 481 return (EIO); 482 483 /* 484 * Determine if the vdev is in use. 485 */ 486 if (reason != VDEV_LABEL_REMOVE && 487 vdev_inuse(vd, crtxg, reason, &spare_guid)) 488 return (EBUSY); 489 490 ASSERT(reason != VDEV_LABEL_REMOVE || 491 vdev_inuse(vd, crtxg, reason, NULL)); 492 493 /* 494 * If this is a request to add or replace a spare that is in use 495 * elsewhere on the system, then we must update the guid (which was 496 * initialized to a random value) to reflect the actual GUID (which is 497 * shared between multiple pools). 498 */ 499 if (reason != VDEV_LABEL_REMOVE && spare_guid != 0ULL) { 500 vdev_t *pvd = vd->vdev_parent; 501 502 for (; pvd != NULL; pvd = pvd->vdev_parent) { 503 pvd->vdev_guid_sum -= vd->vdev_guid; 504 pvd->vdev_guid_sum += spare_guid; 505 } 506 507 vd->vdev_guid = vd->vdev_guid_sum = spare_guid; 508 509 /* 510 * If this is a replacement, then we want to fallthrough to the 511 * rest of the code. If we're adding a spare, then it's already 512 * labeled appropriately and we can just return. 513 */ 514 if (reason == VDEV_LABEL_SPARE) 515 return (0); 516 ASSERT(reason == VDEV_LABEL_REPLACE); 517 } 518 519 /* 520 * Initialize its label. 521 */ 522 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 523 bzero(vp, sizeof (vdev_phys_t)); 524 525 /* 526 * Generate a label describing the pool and our top-level vdev. 527 * We mark it as being from txg 0 to indicate that it's not 528 * really part of an active pool just yet. The labels will 529 * be written again with a meaningful txg by spa_sync(). 530 */ 531 if (reason == VDEV_LABEL_SPARE || 532 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 533 /* 534 * For inactive hot spares, we generate a special label that 535 * identifies as a mutually shared hot spare. We write the 536 * label if we are adding a hot spare, or if we are removing an 537 * active hot spare (in which case we want to revert the 538 * labels). 539 */ 540 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 541 542 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 543 spa_version(spa)) == 0); 544 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 545 POOL_STATE_SPARE) == 0); 546 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 547 vd->vdev_guid) == 0); 548 } else { 549 label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 550 551 /* 552 * Add our creation time. This allows us to detect multiple 553 * vdev uses as described above, and automatically expires if we 554 * fail. 555 */ 556 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 557 crtxg) == 0); 558 } 559 560 buf = vp->vp_nvlist; 561 buflen = sizeof (vp->vp_nvlist); 562 563 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 564 if (error != 0) { 565 nvlist_free(label); 566 zio_buf_free(vp, sizeof (vdev_phys_t)); 567 /* EFAULT means nvlist_pack ran out of room */ 568 return (error == EFAULT ? ENAMETOOLONG : EINVAL); 569 } 570 571 /* 572 * Initialize boot block header. 573 */ 574 vb = zio_buf_alloc(sizeof (vdev_boot_header_t)); 575 bzero(vb, sizeof (vdev_boot_header_t)); 576 vb->vb_magic = VDEV_BOOT_MAGIC; 577 vb->vb_version = VDEV_BOOT_VERSION; 578 vb->vb_offset = VDEV_BOOT_OFFSET; 579 vb->vb_size = VDEV_BOOT_SIZE; 580 581 /* 582 * Initialize uberblock template. 583 */ 584 ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 585 bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 586 *ub = spa->spa_uberblock; 587 ub->ub_txg = 0; 588 589 /* 590 * Write everything in parallel. 591 */ 592 zio = zio_root(spa, NULL, NULL, 593 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 594 595 for (l = 0; l < VDEV_LABELS; l++) { 596 597 vdev_label_write(zio, vd, l, vp, 598 offsetof(vdev_label_t, vl_vdev_phys), 599 sizeof (vdev_phys_t), NULL, NULL); 600 601 vdev_label_write(zio, vd, l, vb, 602 offsetof(vdev_label_t, vl_boot_header), 603 sizeof (vdev_boot_header_t), NULL, NULL); 604 605 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 606 vdev_label_write(zio, vd, l, ub, 607 VDEV_UBERBLOCK_OFFSET(vd, n), 608 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL); 609 } 610 } 611 612 error = zio_wait(zio); 613 614 nvlist_free(label); 615 zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 616 zio_buf_free(vb, sizeof (vdev_boot_header_t)); 617 zio_buf_free(vp, sizeof (vdev_phys_t)); 618 619 /* 620 * If this vdev hasn't been previously identified as a spare, then we 621 * mark it as such only if a) we are labeling it as a spare, or b) it 622 * exists as a spare elsewhere in the system. 623 */ 624 if (error == 0 && !vd->vdev_isspare && 625 (reason == VDEV_LABEL_SPARE || 626 spa_spare_exists(vd->vdev_guid, NULL))) 627 spa_spare_add(vd); 628 629 return (error); 630 } 631 632 /* 633 * ========================================================================== 634 * uberblock load/sync 635 * ========================================================================== 636 */ 637 638 /* 639 * Consider the following situation: txg is safely synced to disk. We've 640 * written the first uberblock for txg + 1, and then we lose power. When we 641 * come back up, we fail to see the uberblock for txg + 1 because, say, 642 * it was on a mirrored device and the replica to which we wrote txg + 1 643 * is now offline. If we then make some changes and sync txg + 1, and then 644 * the missing replica comes back, then for a new seconds we'll have two 645 * conflicting uberblocks on disk with the same txg. The solution is simple: 646 * among uberblocks with equal txg, choose the one with the latest timestamp. 647 */ 648 static int 649 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 650 { 651 if (ub1->ub_txg < ub2->ub_txg) 652 return (-1); 653 if (ub1->ub_txg > ub2->ub_txg) 654 return (1); 655 656 if (ub1->ub_timestamp < ub2->ub_timestamp) 657 return (-1); 658 if (ub1->ub_timestamp > ub2->ub_timestamp) 659 return (1); 660 661 return (0); 662 } 663 664 static void 665 vdev_uberblock_load_done(zio_t *zio) 666 { 667 uberblock_t *ub = zio->io_data; 668 uberblock_t *ubbest = zio->io_private; 669 spa_t *spa = zio->io_spa; 670 671 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 672 673 if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 674 mutex_enter(&spa->spa_uberblock_lock); 675 if (vdev_uberblock_compare(ub, ubbest) > 0) 676 *ubbest = *ub; 677 mutex_exit(&spa->spa_uberblock_lock); 678 } 679 680 zio_buf_free(zio->io_data, zio->io_size); 681 } 682 683 void 684 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 685 { 686 int l, c, n; 687 688 for (c = 0; c < vd->vdev_children; c++) 689 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 690 691 if (!vd->vdev_ops->vdev_op_leaf) 692 return; 693 694 if (vdev_is_dead(vd)) 695 return; 696 697 for (l = 0; l < VDEV_LABELS; l++) { 698 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 699 vdev_label_read(zio, vd, l, 700 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 701 VDEV_UBERBLOCK_OFFSET(vd, n), 702 VDEV_UBERBLOCK_SIZE(vd), 703 vdev_uberblock_load_done, ubbest); 704 } 705 } 706 } 707 708 /* 709 * Write the uberblock to both labels of all leaves of the specified vdev. 710 * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 711 */ 712 static void 713 vdev_uberblock_sync_done(zio_t *zio) 714 { 715 uint64_t *good_writes = zio->io_root->io_private; 716 717 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 718 atomic_add_64(good_writes, 1); 719 } 720 721 static void 722 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, uint64_t txg) 723 { 724 int l, c, n; 725 726 for (c = 0; c < vd->vdev_children; c++) 727 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], txg); 728 729 if (!vd->vdev_ops->vdev_op_leaf) 730 return; 731 732 if (vdev_is_dead(vd)) 733 return; 734 735 n = txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 736 737 ASSERT(ub->ub_txg == txg); 738 739 for (l = 0; l < VDEV_LABELS; l++) 740 vdev_label_write(zio, vd, l, ub, 741 VDEV_UBERBLOCK_OFFSET(vd, n), 742 VDEV_UBERBLOCK_SIZE(vd), 743 vdev_uberblock_sync_done, NULL); 744 745 dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg); 746 } 747 748 static int 749 vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *vd, uint64_t txg) 750 { 751 uberblock_t *ubbuf; 752 size_t size = vd->vdev_top ? VDEV_UBERBLOCK_SIZE(vd) : SPA_MAXBLOCKSIZE; 753 uint64_t *good_writes; 754 zio_t *zio; 755 int error; 756 757 ubbuf = zio_buf_alloc(size); 758 bzero(ubbuf, size); 759 *ubbuf = *ub; 760 761 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 762 763 zio = zio_root(spa, NULL, good_writes, 764 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 765 766 vdev_uberblock_sync(zio, ubbuf, vd, txg); 767 768 error = zio_wait(zio); 769 770 if (error && *good_writes != 0) { 771 dprintf("partial success: good_writes = %llu\n", *good_writes); 772 error = 0; 773 } 774 775 /* 776 * It's possible to have no good writes and no error if every vdev is in 777 * the CANT_OPEN state. 778 */ 779 if (*good_writes == 0 && error == 0) 780 error = EIO; 781 782 kmem_free(good_writes, sizeof (uint64_t)); 783 zio_buf_free(ubbuf, size); 784 785 return (error); 786 } 787 788 /* 789 * Sync out an individual vdev. 790 */ 791 static void 792 vdev_sync_label_done(zio_t *zio) 793 { 794 uint64_t *good_writes = zio->io_root->io_private; 795 796 if (zio->io_error == 0) 797 atomic_add_64(good_writes, 1); 798 } 799 800 static void 801 vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg) 802 { 803 nvlist_t *label; 804 vdev_phys_t *vp; 805 char *buf; 806 size_t buflen; 807 int c; 808 809 for (c = 0; c < vd->vdev_children; c++) 810 vdev_sync_label(zio, vd->vdev_child[c], l, txg); 811 812 if (!vd->vdev_ops->vdev_op_leaf) 813 return; 814 815 if (vdev_is_dead(vd)) 816 return; 817 818 /* 819 * Generate a label describing the top-level config to which we belong. 820 */ 821 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 822 823 vp = zio_buf_alloc(sizeof (vdev_phys_t)); 824 bzero(vp, sizeof (vdev_phys_t)); 825 826 buf = vp->vp_nvlist; 827 buflen = sizeof (vp->vp_nvlist); 828 829 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) 830 vdev_label_write(zio, vd, l, vp, 831 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 832 vdev_sync_label_done, NULL); 833 834 zio_buf_free(vp, sizeof (vdev_phys_t)); 835 nvlist_free(label); 836 837 dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg); 838 } 839 840 static int 841 vdev_sync_labels(vdev_t *vd, int l, uint64_t txg) 842 { 843 uint64_t *good_writes; 844 zio_t *zio; 845 int error; 846 847 ASSERT(vd == vd->vdev_top); 848 849 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 850 851 zio = zio_root(vd->vdev_spa, NULL, good_writes, 852 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 853 854 /* 855 * Recursively kick off writes to all labels. 856 */ 857 vdev_sync_label(zio, vd, l, txg); 858 859 error = zio_wait(zio); 860 861 if (error && *good_writes != 0) { 862 dprintf("partial success: good_writes = %llu\n", *good_writes); 863 error = 0; 864 } 865 866 if (*good_writes == 0 && error == 0) 867 error = ENODEV; 868 869 kmem_free(good_writes, sizeof (uint64_t)); 870 871 return (error); 872 } 873 874 /* 875 * Sync the entire vdev configuration. 876 * 877 * The order of operations is carefully crafted to ensure that 878 * if the system panics or loses power at any time, the state on disk 879 * is still transactionally consistent. The in-line comments below 880 * describe the failure semantics at each stage. 881 * 882 * Moreover, it is designed to be idempotent: if spa_sync_labels() fails 883 * at any time, you can just call it again, and it will resume its work. 884 */ 885 int 886 vdev_config_sync(vdev_t *uvd, uint64_t txg) 887 { 888 spa_t *spa = uvd->vdev_spa; 889 uberblock_t *ub = &spa->spa_uberblock; 890 vdev_t *rvd = spa->spa_root_vdev; 891 vdev_t *vd; 892 zio_t *zio; 893 int l, error; 894 895 ASSERT(ub->ub_txg <= txg); 896 897 /* 898 * If this isn't a resync due to I/O errors, and nothing changed 899 * in this transaction group, and the vdev configuration hasn't changed, 900 * then there's nothing to do. 901 */ 902 if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE && 903 list_is_empty(&spa->spa_dirty_list)) { 904 dprintf("nothing to sync in %s in txg %llu\n", 905 spa_name(spa), txg); 906 return (0); 907 } 908 909 if (txg > spa_freeze_txg(spa)) 910 return (0); 911 912 ASSERT(txg <= spa->spa_final_txg); 913 914 dprintf("syncing %s txg %llu\n", spa_name(spa), txg); 915 916 /* 917 * Flush the write cache of every disk that's been written to 918 * in this transaction group. This ensures that all blocks 919 * written in this txg will be committed to stable storage 920 * before any uberblock that references them. 921 */ 922 zio = zio_root(spa, NULL, NULL, 923 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 924 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 925 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) { 926 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 927 NULL, NULL, ZIO_PRIORITY_NOW, 928 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 929 } 930 (void) zio_wait(zio); 931 932 /* 933 * Sync out the even labels (L0, L2) for every dirty vdev. If the 934 * system dies in the middle of this process, that's OK: all of the 935 * even labels that made it to disk will be newer than any uberblock, 936 * and will therefore be considered invalid. The odd labels (L1, L3), 937 * which have not yet been touched, will still be valid. 938 */ 939 for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 940 vd = list_next(&spa->spa_dirty_list, vd)) { 941 for (l = 0; l < VDEV_LABELS; l++) { 942 if (l & 1) 943 continue; 944 if ((error = vdev_sync_labels(vd, l, txg)) != 0) 945 return (error); 946 } 947 } 948 949 /* 950 * Flush the new labels to disk. This ensures that all even-label 951 * updates are committed to stable storage before the uberblock update. 952 */ 953 zio = zio_root(spa, NULL, NULL, 954 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 955 for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 956 vd = list_next(&spa->spa_dirty_list, vd)) { 957 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 958 NULL, NULL, ZIO_PRIORITY_NOW, 959 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 960 } 961 (void) zio_wait(zio); 962 963 /* 964 * Sync the uberblocks to all vdevs in the tree specified by uvd. 965 * If the system dies in the middle of this step, there are two cases 966 * to consider, and the on-disk state is consistent either way: 967 * 968 * (1) If none of the new uberblocks made it to disk, then the 969 * previous uberblock will be the newest, and the odd labels 970 * (which had not yet been touched) will be valid with respect 971 * to that uberblock. 972 * 973 * (2) If one or more new uberblocks made it to disk, then they 974 * will be the newest, and the even labels (which had all 975 * been successfully committed) will be valid with respect 976 * to the new uberblocks. 977 */ 978 if ((error = vdev_uberblock_sync_tree(spa, ub, uvd, txg)) != 0) 979 return (error); 980 981 /* 982 * Flush the uberblocks to disk. This ensures that the odd labels 983 * are no longer needed (because the new uberblocks and the even 984 * labels are safely on disk), so it is safe to overwrite them. 985 */ 986 (void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE, 987 NULL, NULL, ZIO_PRIORITY_NOW, 988 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 989 990 /* 991 * Sync out odd labels for every dirty vdev. If the system dies 992 * in the middle of this process, the even labels and the new 993 * uberblocks will suffice to open the pool. The next time 994 * the pool is opened, the first thing we'll do -- before any 995 * user data is modified -- is mark every vdev dirty so that 996 * all labels will be brought up to date. 997 */ 998 for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 999 vd = list_next(&spa->spa_dirty_list, vd)) { 1000 for (l = 0; l < VDEV_LABELS; l++) { 1001 if ((l & 1) == 0) 1002 continue; 1003 if ((error = vdev_sync_labels(vd, l, txg)) != 0) 1004 return (error); 1005 } 1006 } 1007 1008 /* 1009 * Flush the new labels to disk. This ensures that all odd-label 1010 * updates are committed to stable storage before the next 1011 * transaction group begins. 1012 */ 1013 zio = zio_root(spa, NULL, NULL, 1014 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 1015 for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1016 vd = list_next(&spa->spa_dirty_list, vd)) { 1017 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 1018 NULL, NULL, ZIO_PRIORITY_NOW, 1019 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 1020 } 1021 (void) zio_wait(zio); 1022 1023 return (0); 1024 } 1025