1 /*- 2 * Copyright (c) 2007 Doug Rabson 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 /* 31 * Stand-alone ZFS file reader. 32 */ 33 34 #include <stdbool.h> 35 #include <sys/endian.h> 36 #include <sys/stat.h> 37 #include <sys/stdint.h> 38 #include <sys/list.h> 39 #include <sys/zfs_bootenv.h> 40 #include <machine/_inttypes.h> 41 42 #include "zfsimpl.h" 43 #include "zfssubr.c" 44 45 #ifdef HAS_ZSTD_ZFS 46 extern int zstd_init(void); 47 #endif 48 49 struct zfsmount { 50 char *path; 51 const spa_t *spa; 52 objset_phys_t objset; 53 uint64_t rootobj; 54 STAILQ_ENTRY(zfsmount) next; 55 }; 56 57 typedef STAILQ_HEAD(zfs_mnt_list, zfsmount) zfs_mnt_list_t; 58 static zfs_mnt_list_t zfsmount = STAILQ_HEAD_INITIALIZER(zfsmount); 59 60 /* 61 * The indirect_child_t represents the vdev that we will read from, when we 62 * need to read all copies of the data (e.g. for scrub or reconstruction). 63 * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror), 64 * ic_vdev is the same as is_vdev. However, for mirror top-level vdevs, 65 * ic_vdev is a child of the mirror. 66 */ 67 typedef struct indirect_child { 68 void *ic_data; 69 vdev_t *ic_vdev; 70 } indirect_child_t; 71 72 /* 73 * The indirect_split_t represents one mapped segment of an i/o to the 74 * indirect vdev. For non-split (contiguously-mapped) blocks, there will be 75 * only one indirect_split_t, with is_split_offset==0 and is_size==io_size. 76 * For split blocks, there will be several of these. 77 */ 78 typedef struct indirect_split { 79 list_node_t is_node; /* link on iv_splits */ 80 81 /* 82 * is_split_offset is the offset into the i/o. 83 * This is the sum of the previous splits' is_size's. 84 */ 85 uint64_t is_split_offset; 86 87 vdev_t *is_vdev; /* top-level vdev */ 88 uint64_t is_target_offset; /* offset on is_vdev */ 89 uint64_t is_size; 90 int is_children; /* number of entries in is_child[] */ 91 92 /* 93 * is_good_child is the child that we are currently using to 94 * attempt reconstruction. 95 */ 96 int is_good_child; 97 98 indirect_child_t is_child[1]; /* variable-length */ 99 } indirect_split_t; 100 101 /* 102 * The indirect_vsd_t is associated with each i/o to the indirect vdev. 103 * It is the "Vdev-Specific Data" in the zio_t's io_vsd. 104 */ 105 typedef struct indirect_vsd { 106 boolean_t iv_split_block; 107 boolean_t iv_reconstruct; 108 109 list_t iv_splits; /* list of indirect_split_t's */ 110 } indirect_vsd_t; 111 112 /* 113 * List of all vdevs, chained through v_alllink. 114 */ 115 static vdev_list_t zfs_vdevs; 116 117 /* 118 * List of ZFS features supported for read 119 */ 120 static const char *features_for_read[] = { 121 "com.datto:bookmark_v2", 122 "com.datto:encryption", 123 "com.datto:resilver_defer", 124 "com.delphix:bookmark_written", 125 "com.delphix:device_removal", 126 "com.delphix:embedded_data", 127 "com.delphix:extensible_dataset", 128 "com.delphix:head_errlog", 129 "com.delphix:hole_birth", 130 "com.delphix:obsolete_counts", 131 "com.delphix:spacemap_histogram", 132 "com.delphix:spacemap_v2", 133 "com.delphix:zpool_checkpoint", 134 "com.intel:allocation_classes", 135 "com.joyent:multi_vdev_crash_dump", 136 "com.klarasystems:vdev_zaps_v2", 137 "org.freebsd:zstd_compress", 138 "org.illumos:lz4_compress", 139 "org.illumos:sha512", 140 "org.illumos:skein", 141 "org.open-zfs:large_blocks", 142 "org.openzfs:blake3", 143 "org.zfsonlinux:allocation_classes", 144 "org.zfsonlinux:large_dnode", 145 NULL 146 }; 147 148 /* 149 * List of all pools, chained through spa_link. 150 */ 151 static spa_list_t zfs_pools; 152 153 static const dnode_phys_t *dnode_cache_obj; 154 static uint64_t dnode_cache_bn; 155 static char *dnode_cache_buf; 156 157 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf); 158 static int zfs_get_root(const spa_t *spa, uint64_t *objid); 159 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result); 160 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, 161 const char *name, uint64_t integer_size, uint64_t num_integers, 162 void *value); 163 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t, 164 dnode_phys_t *); 165 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *, 166 size_t); 167 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t, 168 size_t); 169 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t); 170 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *, 171 uint64_t); 172 vdev_indirect_mapping_entry_phys_t * 173 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t, 174 uint64_t, uint64_t *); 175 176 static void 177 zfs_init(void) 178 { 179 STAILQ_INIT(&zfs_vdevs); 180 STAILQ_INIT(&zfs_pools); 181 182 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE); 183 184 zfs_init_crc(); 185 #ifdef HAS_ZSTD_ZFS 186 zstd_init(); 187 #endif 188 } 189 190 static int 191 nvlist_check_features_for_read(nvlist_t *nvl) 192 { 193 nvlist_t *features = NULL; 194 nvs_data_t *data; 195 nvp_header_t *nvp; 196 nv_string_t *nvp_name; 197 int rc; 198 199 rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ, 200 DATA_TYPE_NVLIST, NULL, &features, NULL); 201 switch (rc) { 202 case 0: 203 break; /* Continue with checks */ 204 205 case ENOENT: 206 return (0); /* All features are disabled */ 207 208 default: 209 return (rc); /* Error while reading nvlist */ 210 } 211 212 data = (nvs_data_t *)features->nv_data; 213 nvp = &data->nvl_pair; /* first pair in nvlist */ 214 215 while (nvp->encoded_size != 0 && nvp->decoded_size != 0) { 216 int i, found; 217 218 nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp)); 219 found = 0; 220 221 for (i = 0; features_for_read[i] != NULL; i++) { 222 if (memcmp(nvp_name->nv_data, features_for_read[i], 223 nvp_name->nv_size) == 0) { 224 found = 1; 225 break; 226 } 227 } 228 229 if (!found) { 230 printf("ZFS: unsupported feature: %.*s\n", 231 nvp_name->nv_size, nvp_name->nv_data); 232 rc = EIO; 233 } 234 nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size); 235 } 236 nvlist_destroy(features); 237 238 return (rc); 239 } 240 241 static int 242 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf, 243 off_t offset, size_t size) 244 { 245 size_t psize; 246 int rc; 247 248 if (vdev->v_phys_read == NULL) 249 return (ENOTSUP); 250 251 if (bp) { 252 psize = BP_GET_PSIZE(bp); 253 } else { 254 psize = size; 255 } 256 257 rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize); 258 if (rc == 0) { 259 if (bp != NULL) 260 rc = zio_checksum_verify(vdev->v_spa, bp, buf); 261 } 262 263 return (rc); 264 } 265 266 static int 267 vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size) 268 { 269 if (vdev->v_phys_write == NULL) 270 return (ENOTSUP); 271 272 return (vdev->v_phys_write(vdev, offset, buf, size)); 273 } 274 275 typedef struct remap_segment { 276 vdev_t *rs_vd; 277 uint64_t rs_offset; 278 uint64_t rs_asize; 279 uint64_t rs_split_offset; 280 list_node_t rs_node; 281 } remap_segment_t; 282 283 static remap_segment_t * 284 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset) 285 { 286 remap_segment_t *rs = malloc(sizeof (remap_segment_t)); 287 288 if (rs != NULL) { 289 rs->rs_vd = vd; 290 rs->rs_offset = offset; 291 rs->rs_asize = asize; 292 rs->rs_split_offset = split_offset; 293 } 294 295 return (rs); 296 } 297 298 vdev_indirect_mapping_t * 299 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os, 300 uint64_t mapping_object) 301 { 302 vdev_indirect_mapping_t *vim; 303 vdev_indirect_mapping_phys_t *vim_phys; 304 int rc; 305 306 vim = calloc(1, sizeof (*vim)); 307 if (vim == NULL) 308 return (NULL); 309 310 vim->vim_dn = calloc(1, sizeof (*vim->vim_dn)); 311 if (vim->vim_dn == NULL) { 312 free(vim); 313 return (NULL); 314 } 315 316 rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn); 317 if (rc != 0) { 318 free(vim->vim_dn); 319 free(vim); 320 return (NULL); 321 } 322 323 vim->vim_spa = spa; 324 vim->vim_phys = malloc(sizeof (*vim->vim_phys)); 325 if (vim->vim_phys == NULL) { 326 free(vim->vim_dn); 327 free(vim); 328 return (NULL); 329 } 330 331 vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn); 332 *vim->vim_phys = *vim_phys; 333 334 vim->vim_objset = os; 335 vim->vim_object = mapping_object; 336 vim->vim_entries = NULL; 337 338 vim->vim_havecounts = 339 (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0); 340 341 return (vim); 342 } 343 344 /* 345 * Compare an offset with an indirect mapping entry; there are three 346 * possible scenarios: 347 * 348 * 1. The offset is "less than" the mapping entry; meaning the 349 * offset is less than the source offset of the mapping entry. In 350 * this case, there is no overlap between the offset and the 351 * mapping entry and -1 will be returned. 352 * 353 * 2. The offset is "greater than" the mapping entry; meaning the 354 * offset is greater than the mapping entry's source offset plus 355 * the entry's size. In this case, there is no overlap between 356 * the offset and the mapping entry and 1 will be returned. 357 * 358 * NOTE: If the offset is actually equal to the entry's offset 359 * plus size, this is considered to be "greater" than the entry, 360 * and this case applies (i.e. 1 will be returned). Thus, the 361 * entry's "range" can be considered to be inclusive at its 362 * start, but exclusive at its end: e.g. [src, src + size). 363 * 364 * 3. The last case to consider is if the offset actually falls 365 * within the mapping entry's range. If this is the case, the 366 * offset is considered to be "equal to" the mapping entry and 367 * 0 will be returned. 368 * 369 * NOTE: If the offset is equal to the entry's source offset, 370 * this case applies and 0 will be returned. If the offset is 371 * equal to the entry's source plus its size, this case does 372 * *not* apply (see "NOTE" above for scenario 2), and 1 will be 373 * returned. 374 */ 375 static int 376 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem) 377 { 378 const uint64_t *key = v_key; 379 const vdev_indirect_mapping_entry_phys_t *array_elem = 380 v_array_elem; 381 uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem); 382 383 if (*key < src_offset) { 384 return (-1); 385 } else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) { 386 return (0); 387 } else { 388 return (1); 389 } 390 } 391 392 /* 393 * Return array entry. 394 */ 395 static vdev_indirect_mapping_entry_phys_t * 396 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index) 397 { 398 uint64_t size; 399 off_t offset = 0; 400 int rc; 401 402 if (vim->vim_phys->vimp_num_entries == 0) 403 return (NULL); 404 405 if (vim->vim_entries == NULL) { 406 uint64_t bsize; 407 408 bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 409 size = vim->vim_phys->vimp_num_entries * 410 sizeof (*vim->vim_entries); 411 if (size > bsize) { 412 size = bsize / sizeof (*vim->vim_entries); 413 size *= sizeof (*vim->vim_entries); 414 } 415 vim->vim_entries = malloc(size); 416 if (vim->vim_entries == NULL) 417 return (NULL); 418 vim->vim_num_entries = size / sizeof (*vim->vim_entries); 419 offset = index * sizeof (*vim->vim_entries); 420 } 421 422 /* We have data in vim_entries */ 423 if (offset == 0) { 424 if (index >= vim->vim_entry_offset && 425 index <= vim->vim_entry_offset + vim->vim_num_entries) { 426 index -= vim->vim_entry_offset; 427 return (&vim->vim_entries[index]); 428 } 429 offset = index * sizeof (*vim->vim_entries); 430 } 431 432 vim->vim_entry_offset = index; 433 size = vim->vim_num_entries * sizeof (*vim->vim_entries); 434 rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries, 435 size); 436 if (rc != 0) { 437 /* Read error, invalidate vim_entries. */ 438 free(vim->vim_entries); 439 vim->vim_entries = NULL; 440 return (NULL); 441 } 442 index -= vim->vim_entry_offset; 443 return (&vim->vim_entries[index]); 444 } 445 446 /* 447 * Returns the mapping entry for the given offset. 448 * 449 * It's possible that the given offset will not be in the mapping table 450 * (i.e. no mapping entries contain this offset), in which case, the 451 * return value depends on the "next_if_missing" parameter. 452 * 453 * If the offset is not found in the table and "next_if_missing" is 454 * B_FALSE, then NULL will always be returned. The behavior is intended 455 * to allow consumers to get the entry corresponding to the offset 456 * parameter, iff the offset overlaps with an entry in the table. 457 * 458 * If the offset is not found in the table and "next_if_missing" is 459 * B_TRUE, then the entry nearest to the given offset will be returned, 460 * such that the entry's source offset is greater than the offset 461 * passed in (i.e. the "next" mapping entry in the table is returned, if 462 * the offset is missing from the table). If there are no entries whose 463 * source offset is greater than the passed in offset, NULL is returned. 464 */ 465 static vdev_indirect_mapping_entry_phys_t * 466 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim, 467 uint64_t offset) 468 { 469 ASSERT(vim->vim_phys->vimp_num_entries > 0); 470 471 vdev_indirect_mapping_entry_phys_t *entry; 472 473 uint64_t last = vim->vim_phys->vimp_num_entries - 1; 474 uint64_t base = 0; 475 476 /* 477 * We don't define these inside of the while loop because we use 478 * their value in the case that offset isn't in the mapping. 479 */ 480 uint64_t mid; 481 int result; 482 483 while (last >= base) { 484 mid = base + ((last - base) >> 1); 485 486 entry = vdev_indirect_mapping_entry(vim, mid); 487 if (entry == NULL) 488 break; 489 result = dva_mapping_overlap_compare(&offset, entry); 490 491 if (result == 0) { 492 break; 493 } else if (result < 0) { 494 last = mid - 1; 495 } else { 496 base = mid + 1; 497 } 498 } 499 return (entry); 500 } 501 502 /* 503 * Given an indirect vdev and an extent on that vdev, it duplicates the 504 * physical entries of the indirect mapping that correspond to the extent 505 * to a new array and returns a pointer to it. In addition, copied_entries 506 * is populated with the number of mapping entries that were duplicated. 507 * 508 * Finally, since we are doing an allocation, it is up to the caller to 509 * free the array allocated in this function. 510 */ 511 vdev_indirect_mapping_entry_phys_t * 512 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset, 513 uint64_t asize, uint64_t *copied_entries) 514 { 515 vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL; 516 vdev_indirect_mapping_t *vim = vd->v_mapping; 517 uint64_t entries = 0; 518 519 vdev_indirect_mapping_entry_phys_t *first_mapping = 520 vdev_indirect_mapping_entry_for_offset(vim, offset); 521 ASSERT3P(first_mapping, !=, NULL); 522 523 vdev_indirect_mapping_entry_phys_t *m = first_mapping; 524 while (asize > 0) { 525 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst); 526 uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m); 527 uint64_t inner_size = MIN(asize, size - inner_offset); 528 529 offset += inner_size; 530 asize -= inner_size; 531 entries++; 532 m++; 533 } 534 535 size_t copy_length = entries * sizeof (*first_mapping); 536 duplicate_mappings = malloc(copy_length); 537 if (duplicate_mappings != NULL) 538 bcopy(first_mapping, duplicate_mappings, copy_length); 539 else 540 entries = 0; 541 542 *copied_entries = entries; 543 544 return (duplicate_mappings); 545 } 546 547 static vdev_t * 548 vdev_lookup_top(spa_t *spa, uint64_t vdev) 549 { 550 vdev_t *rvd; 551 vdev_list_t *vlist; 552 553 vlist = &spa->spa_root_vdev->v_children; 554 STAILQ_FOREACH(rvd, vlist, v_childlink) 555 if (rvd->v_id == vdev) 556 break; 557 558 return (rvd); 559 } 560 561 /* 562 * This is a callback for vdev_indirect_remap() which allocates an 563 * indirect_split_t for each split segment and adds it to iv_splits. 564 */ 565 static void 566 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset, 567 uint64_t size, void *arg) 568 { 569 int n = 1; 570 zio_t *zio = arg; 571 indirect_vsd_t *iv = zio->io_vsd; 572 573 if (vd->v_read == vdev_indirect_read) 574 return; 575 576 if (vd->v_read == vdev_mirror_read) 577 n = vd->v_nchildren; 578 579 indirect_split_t *is = 580 malloc(offsetof(indirect_split_t, is_child[n])); 581 if (is == NULL) { 582 zio->io_error = ENOMEM; 583 return; 584 } 585 bzero(is, offsetof(indirect_split_t, is_child[n])); 586 587 is->is_children = n; 588 is->is_size = size; 589 is->is_split_offset = split_offset; 590 is->is_target_offset = offset; 591 is->is_vdev = vd; 592 593 /* 594 * Note that we only consider multiple copies of the data for 595 * *mirror* vdevs. We don't for "replacing" or "spare" vdevs, even 596 * though they use the same ops as mirror, because there's only one 597 * "good" copy under the replacing/spare. 598 */ 599 if (vd->v_read == vdev_mirror_read) { 600 int i = 0; 601 vdev_t *kid; 602 603 STAILQ_FOREACH(kid, &vd->v_children, v_childlink) { 604 is->is_child[i++].ic_vdev = kid; 605 } 606 } else { 607 is->is_child[0].ic_vdev = vd; 608 } 609 610 list_insert_tail(&iv->iv_splits, is); 611 } 612 613 static void 614 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg) 615 { 616 list_t stack; 617 spa_t *spa = vd->v_spa; 618 zio_t *zio = arg; 619 remap_segment_t *rs; 620 621 list_create(&stack, sizeof (remap_segment_t), 622 offsetof(remap_segment_t, rs_node)); 623 624 rs = rs_alloc(vd, offset, asize, 0); 625 if (rs == NULL) { 626 printf("vdev_indirect_remap: out of memory.\n"); 627 zio->io_error = ENOMEM; 628 } 629 for (; rs != NULL; rs = list_remove_head(&stack)) { 630 vdev_t *v = rs->rs_vd; 631 uint64_t num_entries = 0; 632 /* vdev_indirect_mapping_t *vim = v->v_mapping; */ 633 vdev_indirect_mapping_entry_phys_t *mapping = 634 vdev_indirect_mapping_duplicate_adjacent_entries(v, 635 rs->rs_offset, rs->rs_asize, &num_entries); 636 637 if (num_entries == 0) 638 zio->io_error = ENOMEM; 639 640 for (uint64_t i = 0; i < num_entries; i++) { 641 vdev_indirect_mapping_entry_phys_t *m = &mapping[i]; 642 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst); 643 uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst); 644 uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst); 645 uint64_t inner_offset = rs->rs_offset - 646 DVA_MAPPING_GET_SRC_OFFSET(m); 647 uint64_t inner_size = 648 MIN(rs->rs_asize, size - inner_offset); 649 vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev); 650 651 if (dst_v->v_read == vdev_indirect_read) { 652 remap_segment_t *o; 653 654 o = rs_alloc(dst_v, dst_offset + inner_offset, 655 inner_size, rs->rs_split_offset); 656 if (o == NULL) { 657 printf("vdev_indirect_remap: " 658 "out of memory.\n"); 659 zio->io_error = ENOMEM; 660 break; 661 } 662 663 list_insert_head(&stack, o); 664 } 665 vdev_indirect_gather_splits(rs->rs_split_offset, dst_v, 666 dst_offset + inner_offset, 667 inner_size, arg); 668 669 /* 670 * vdev_indirect_gather_splits can have memory 671 * allocation error, we can not recover from it. 672 */ 673 if (zio->io_error != 0) 674 break; 675 rs->rs_offset += inner_size; 676 rs->rs_asize -= inner_size; 677 rs->rs_split_offset += inner_size; 678 } 679 680 free(mapping); 681 free(rs); 682 if (zio->io_error != 0) 683 break; 684 } 685 686 list_destroy(&stack); 687 } 688 689 static void 690 vdev_indirect_map_free(zio_t *zio) 691 { 692 indirect_vsd_t *iv = zio->io_vsd; 693 indirect_split_t *is; 694 695 while ((is = list_head(&iv->iv_splits)) != NULL) { 696 for (int c = 0; c < is->is_children; c++) { 697 indirect_child_t *ic = &is->is_child[c]; 698 free(ic->ic_data); 699 } 700 list_remove(&iv->iv_splits, is); 701 free(is); 702 } 703 free(iv); 704 } 705 706 static int 707 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 708 off_t offset, size_t bytes) 709 { 710 zio_t zio; 711 spa_t *spa = vdev->v_spa; 712 indirect_vsd_t *iv; 713 indirect_split_t *first; 714 int rc = EIO; 715 716 iv = calloc(1, sizeof(*iv)); 717 if (iv == NULL) 718 return (ENOMEM); 719 720 list_create(&iv->iv_splits, 721 sizeof (indirect_split_t), offsetof(indirect_split_t, is_node)); 722 723 bzero(&zio, sizeof(zio)); 724 zio.io_spa = spa; 725 zio.io_bp = (blkptr_t *)bp; 726 zio.io_data = buf; 727 zio.io_size = bytes; 728 zio.io_offset = offset; 729 zio.io_vd = vdev; 730 zio.io_vsd = iv; 731 732 if (vdev->v_mapping == NULL) { 733 vdev_indirect_config_t *vic; 734 735 vic = &vdev->vdev_indirect_config; 736 vdev->v_mapping = vdev_indirect_mapping_open(spa, 737 spa->spa_mos, vic->vic_mapping_object); 738 } 739 740 vdev_indirect_remap(vdev, offset, bytes, &zio); 741 if (zio.io_error != 0) 742 return (zio.io_error); 743 744 first = list_head(&iv->iv_splits); 745 if (first->is_size == zio.io_size) { 746 /* 747 * This is not a split block; we are pointing to the entire 748 * data, which will checksum the same as the original data. 749 * Pass the BP down so that the child i/o can verify the 750 * checksum, and try a different location if available 751 * (e.g. on a mirror). 752 * 753 * While this special case could be handled the same as the 754 * general (split block) case, doing it this way ensures 755 * that the vast majority of blocks on indirect vdevs 756 * (which are not split) are handled identically to blocks 757 * on non-indirect vdevs. This allows us to be less strict 758 * about performance in the general (but rare) case. 759 */ 760 rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp, 761 zio.io_data, first->is_target_offset, bytes); 762 } else { 763 iv->iv_split_block = B_TRUE; 764 /* 765 * Read one copy of each split segment, from the 766 * top-level vdev. Since we don't know the 767 * checksum of each split individually, the child 768 * zio can't ensure that we get the right data. 769 * E.g. if it's a mirror, it will just read from a 770 * random (healthy) leaf vdev. We have to verify 771 * the checksum in vdev_indirect_io_done(). 772 */ 773 for (indirect_split_t *is = list_head(&iv->iv_splits); 774 is != NULL; is = list_next(&iv->iv_splits, is)) { 775 char *ptr = zio.io_data; 776 777 rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp, 778 ptr + is->is_split_offset, is->is_target_offset, 779 is->is_size); 780 } 781 if (zio_checksum_verify(spa, zio.io_bp, zio.io_data)) 782 rc = ECKSUM; 783 else 784 rc = 0; 785 } 786 787 vdev_indirect_map_free(&zio); 788 if (rc == 0) 789 rc = zio.io_error; 790 791 return (rc); 792 } 793 794 static int 795 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 796 off_t offset, size_t bytes) 797 { 798 799 return (vdev_read_phys(vdev, bp, buf, 800 offset + VDEV_LABEL_START_SIZE, bytes)); 801 } 802 803 static int 804 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused, 805 void *buf __unused, off_t offset __unused, size_t bytes __unused) 806 { 807 808 return (ENOTSUP); 809 } 810 811 static int 812 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 813 off_t offset, size_t bytes) 814 { 815 vdev_t *kid; 816 int rc; 817 818 rc = EIO; 819 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 820 if (kid->v_state != VDEV_STATE_HEALTHY) 821 continue; 822 rc = kid->v_read(kid, bp, buf, offset, bytes); 823 if (!rc) 824 return (0); 825 } 826 827 return (rc); 828 } 829 830 static int 831 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 832 off_t offset, size_t bytes) 833 { 834 vdev_t *kid; 835 836 /* 837 * Here we should have two kids: 838 * First one which is the one we are replacing and we can trust 839 * only this one to have valid data, but it might not be present. 840 * Second one is that one we are replacing with. It is most likely 841 * healthy, but we can't trust it has needed data, so we won't use it. 842 */ 843 kid = STAILQ_FIRST(&vdev->v_children); 844 if (kid == NULL) 845 return (EIO); 846 if (kid->v_state != VDEV_STATE_HEALTHY) 847 return (EIO); 848 return (kid->v_read(kid, bp, buf, offset, bytes)); 849 } 850 851 static vdev_t * 852 vdev_find(uint64_t guid) 853 { 854 vdev_t *vdev; 855 856 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink) 857 if (vdev->v_guid == guid) 858 return (vdev); 859 860 return (0); 861 } 862 863 static vdev_t * 864 vdev_create(uint64_t guid, vdev_read_t *_read) 865 { 866 vdev_t *vdev; 867 vdev_indirect_config_t *vic; 868 869 vdev = calloc(1, sizeof(vdev_t)); 870 if (vdev != NULL) { 871 STAILQ_INIT(&vdev->v_children); 872 vdev->v_guid = guid; 873 vdev->v_read = _read; 874 875 /* 876 * root vdev has no read function, we use this fact to 877 * skip setting up data we do not need for root vdev. 878 * We only point root vdev from spa. 879 */ 880 if (_read != NULL) { 881 vic = &vdev->vdev_indirect_config; 882 vic->vic_prev_indirect_vdev = UINT64_MAX; 883 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink); 884 } 885 } 886 887 return (vdev); 888 } 889 890 static void 891 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist) 892 { 893 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present; 894 uint64_t is_log; 895 896 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0; 897 is_log = 0; 898 (void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL, 899 &is_offline, NULL); 900 (void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL, 901 &is_removed, NULL); 902 (void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL, 903 &is_faulted, NULL); 904 (void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 905 NULL, &is_degraded, NULL); 906 (void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 907 NULL, &isnt_present, NULL); 908 (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL, 909 &is_log, NULL); 910 911 if (is_offline != 0) 912 vdev->v_state = VDEV_STATE_OFFLINE; 913 else if (is_removed != 0) 914 vdev->v_state = VDEV_STATE_REMOVED; 915 else if (is_faulted != 0) 916 vdev->v_state = VDEV_STATE_FAULTED; 917 else if (is_degraded != 0) 918 vdev->v_state = VDEV_STATE_DEGRADED; 919 else if (isnt_present != 0) 920 vdev->v_state = VDEV_STATE_CANT_OPEN; 921 922 vdev->v_islog = is_log != 0; 923 } 924 925 static int 926 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp) 927 { 928 uint64_t id, ashift, asize, nparity; 929 const char *path; 930 const char *type; 931 int len, pathlen; 932 char *name; 933 vdev_t *vdev; 934 935 if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id, 936 NULL) || 937 nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL, 938 &type, &len)) { 939 return (ENOENT); 940 } 941 942 if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 && 943 memcmp(type, VDEV_TYPE_DISK, len) != 0 && 944 #ifdef ZFS_TEST 945 memcmp(type, VDEV_TYPE_FILE, len) != 0 && 946 #endif 947 memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 && 948 memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 && 949 memcmp(type, VDEV_TYPE_REPLACING, len) != 0 && 950 memcmp(type, VDEV_TYPE_HOLE, len) != 0) { 951 printf("ZFS: can only boot from disk, mirror, raidz1, " 952 "raidz2 and raidz3 vdevs, got: %.*s\n", len, type); 953 return (EIO); 954 } 955 956 if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0) 957 vdev = vdev_create(guid, vdev_mirror_read); 958 else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) 959 vdev = vdev_create(guid, vdev_raidz_read); 960 else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0) 961 vdev = vdev_create(guid, vdev_replacing_read); 962 else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) { 963 vdev_indirect_config_t *vic; 964 965 vdev = vdev_create(guid, vdev_indirect_read); 966 if (vdev != NULL) { 967 vdev->v_state = VDEV_STATE_HEALTHY; 968 vic = &vdev->vdev_indirect_config; 969 970 nvlist_find(nvlist, 971 ZPOOL_CONFIG_INDIRECT_OBJECT, 972 DATA_TYPE_UINT64, 973 NULL, &vic->vic_mapping_object, NULL); 974 nvlist_find(nvlist, 975 ZPOOL_CONFIG_INDIRECT_BIRTHS, 976 DATA_TYPE_UINT64, 977 NULL, &vic->vic_births_object, NULL); 978 nvlist_find(nvlist, 979 ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 980 DATA_TYPE_UINT64, 981 NULL, &vic->vic_prev_indirect_vdev, NULL); 982 } 983 } else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) { 984 vdev = vdev_create(guid, vdev_missing_read); 985 } else { 986 vdev = vdev_create(guid, vdev_disk_read); 987 } 988 989 if (vdev == NULL) 990 return (ENOMEM); 991 992 vdev_set_initial_state(vdev, nvlist); 993 vdev->v_id = id; 994 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT, 995 DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0) 996 vdev->v_ashift = ashift; 997 998 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE, 999 DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) { 1000 vdev->v_psize = asize + 1001 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 1002 } 1003 1004 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY, 1005 DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0) 1006 vdev->v_nparity = nparity; 1007 1008 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH, 1009 DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) { 1010 char prefix[] = "/dev/"; 1011 1012 len = strlen(prefix); 1013 if (len < pathlen && memcmp(path, prefix, len) == 0) { 1014 path += len; 1015 pathlen -= len; 1016 } 1017 name = malloc(pathlen + 1); 1018 bcopy(path, name, pathlen); 1019 name[pathlen] = '\0'; 1020 vdev->v_name = name; 1021 } else { 1022 name = NULL; 1023 if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) { 1024 if (vdev->v_nparity < 1 || 1025 vdev->v_nparity > 3) { 1026 printf("ZFS: invalid raidz parity: %d\n", 1027 vdev->v_nparity); 1028 return (EIO); 1029 } 1030 (void) asprintf(&name, "%.*s%d-%" PRIu64, len, type, 1031 vdev->v_nparity, id); 1032 } else { 1033 (void) asprintf(&name, "%.*s-%" PRIu64, len, type, id); 1034 } 1035 vdev->v_name = name; 1036 } 1037 *vdevp = vdev; 1038 return (0); 1039 } 1040 1041 /* 1042 * Find slot for vdev. We return either NULL to signal to use 1043 * STAILQ_INSERT_HEAD, or we return link element to be used with 1044 * STAILQ_INSERT_AFTER. 1045 */ 1046 static vdev_t * 1047 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev) 1048 { 1049 vdev_t *v, *previous; 1050 1051 if (STAILQ_EMPTY(&top_vdev->v_children)) 1052 return (NULL); 1053 1054 previous = NULL; 1055 STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) { 1056 if (v->v_id > vdev->v_id) 1057 return (previous); 1058 1059 if (v->v_id == vdev->v_id) 1060 return (v); 1061 1062 if (v->v_id < vdev->v_id) 1063 previous = v; 1064 } 1065 return (previous); 1066 } 1067 1068 static size_t 1069 vdev_child_count(vdev_t *vdev) 1070 { 1071 vdev_t *v; 1072 size_t count; 1073 1074 count = 0; 1075 STAILQ_FOREACH(v, &vdev->v_children, v_childlink) { 1076 count++; 1077 } 1078 return (count); 1079 } 1080 1081 /* 1082 * Insert vdev into top_vdev children list. List is ordered by v_id. 1083 */ 1084 static void 1085 vdev_insert(vdev_t *top_vdev, vdev_t *vdev) 1086 { 1087 vdev_t *previous; 1088 size_t count; 1089 1090 /* 1091 * The top level vdev can appear in random order, depending how 1092 * the firmware is presenting the disk devices. 1093 * However, we will insert vdev to create list ordered by v_id, 1094 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER 1095 * as STAILQ does not have insert before. 1096 */ 1097 previous = vdev_find_previous(top_vdev, vdev); 1098 1099 if (previous == NULL) { 1100 STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink); 1101 } else if (previous->v_id == vdev->v_id) { 1102 /* 1103 * This vdev was configured from label config, 1104 * do not insert duplicate. 1105 */ 1106 return; 1107 } else { 1108 STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev, 1109 v_childlink); 1110 } 1111 1112 count = vdev_child_count(top_vdev); 1113 if (top_vdev->v_nchildren < count) 1114 top_vdev->v_nchildren = count; 1115 } 1116 1117 static int 1118 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist) 1119 { 1120 vdev_t *top_vdev, *vdev; 1121 nvlist_t **kids = NULL; 1122 int rc, nkids; 1123 1124 /* Get top vdev. */ 1125 top_vdev = vdev_find(top_guid); 1126 if (top_vdev == NULL) { 1127 rc = vdev_init(top_guid, nvlist, &top_vdev); 1128 if (rc != 0) 1129 return (rc); 1130 top_vdev->v_spa = spa; 1131 top_vdev->v_top = top_vdev; 1132 vdev_insert(spa->spa_root_vdev, top_vdev); 1133 } 1134 1135 /* Add children if there are any. */ 1136 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1137 &nkids, &kids, NULL); 1138 if (rc == 0) { 1139 for (int i = 0; i < nkids; i++) { 1140 uint64_t guid; 1141 1142 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, 1143 DATA_TYPE_UINT64, NULL, &guid, NULL); 1144 if (rc != 0) 1145 goto done; 1146 1147 rc = vdev_init(guid, kids[i], &vdev); 1148 if (rc != 0) 1149 goto done; 1150 1151 vdev->v_spa = spa; 1152 vdev->v_top = top_vdev; 1153 vdev_insert(top_vdev, vdev); 1154 } 1155 } else { 1156 /* 1157 * When there are no children, nvlist_find() does return 1158 * error, reset it because leaf devices have no children. 1159 */ 1160 rc = 0; 1161 } 1162 done: 1163 if (kids != NULL) { 1164 for (int i = 0; i < nkids; i++) 1165 nvlist_destroy(kids[i]); 1166 free(kids); 1167 } 1168 1169 return (rc); 1170 } 1171 1172 static int 1173 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist) 1174 { 1175 uint64_t pool_guid, top_guid; 1176 nvlist_t *vdevs; 1177 int rc; 1178 1179 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 1180 NULL, &pool_guid, NULL) || 1181 nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64, 1182 NULL, &top_guid, NULL) || 1183 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1184 NULL, &vdevs, NULL)) { 1185 printf("ZFS: can't find vdev details\n"); 1186 return (ENOENT); 1187 } 1188 1189 rc = vdev_from_nvlist(spa, top_guid, vdevs); 1190 nvlist_destroy(vdevs); 1191 return (rc); 1192 } 1193 1194 static void 1195 vdev_set_state(vdev_t *vdev) 1196 { 1197 vdev_t *kid; 1198 int good_kids; 1199 int bad_kids; 1200 1201 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1202 vdev_set_state(kid); 1203 } 1204 1205 /* 1206 * A mirror or raidz is healthy if all its kids are healthy. A 1207 * mirror is degraded if any of its kids is healthy; a raidz 1208 * is degraded if at most nparity kids are offline. 1209 */ 1210 if (STAILQ_FIRST(&vdev->v_children)) { 1211 good_kids = 0; 1212 bad_kids = 0; 1213 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1214 if (kid->v_state == VDEV_STATE_HEALTHY) 1215 good_kids++; 1216 else 1217 bad_kids++; 1218 } 1219 if (bad_kids == 0) { 1220 vdev->v_state = VDEV_STATE_HEALTHY; 1221 } else { 1222 if (vdev->v_read == vdev_mirror_read) { 1223 if (good_kids) { 1224 vdev->v_state = VDEV_STATE_DEGRADED; 1225 } else { 1226 vdev->v_state = VDEV_STATE_OFFLINE; 1227 } 1228 } else if (vdev->v_read == vdev_raidz_read) { 1229 if (bad_kids > vdev->v_nparity) { 1230 vdev->v_state = VDEV_STATE_OFFLINE; 1231 } else { 1232 vdev->v_state = VDEV_STATE_DEGRADED; 1233 } 1234 } 1235 } 1236 } 1237 } 1238 1239 static int 1240 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist) 1241 { 1242 vdev_t *vdev; 1243 nvlist_t **kids = NULL; 1244 int rc, nkids; 1245 1246 /* Update top vdev. */ 1247 vdev = vdev_find(top_guid); 1248 if (vdev != NULL) 1249 vdev_set_initial_state(vdev, nvlist); 1250 1251 /* Update children if there are any. */ 1252 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1253 &nkids, &kids, NULL); 1254 if (rc == 0) { 1255 for (int i = 0; i < nkids; i++) { 1256 uint64_t guid; 1257 1258 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, 1259 DATA_TYPE_UINT64, NULL, &guid, NULL); 1260 if (rc != 0) 1261 break; 1262 1263 vdev = vdev_find(guid); 1264 if (vdev != NULL) 1265 vdev_set_initial_state(vdev, kids[i]); 1266 } 1267 } else { 1268 rc = 0; 1269 } 1270 if (kids != NULL) { 1271 for (int i = 0; i < nkids; i++) 1272 nvlist_destroy(kids[i]); 1273 free(kids); 1274 } 1275 1276 return (rc); 1277 } 1278 1279 static int 1280 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist) 1281 { 1282 uint64_t pool_guid, vdev_children; 1283 nvlist_t *vdevs = NULL, **kids = NULL; 1284 int rc, nkids; 1285 1286 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 1287 NULL, &pool_guid, NULL) || 1288 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64, 1289 NULL, &vdev_children, NULL) || 1290 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1291 NULL, &vdevs, NULL)) { 1292 printf("ZFS: can't find vdev details\n"); 1293 return (ENOENT); 1294 } 1295 1296 /* Wrong guid?! */ 1297 if (spa->spa_guid != pool_guid) { 1298 nvlist_destroy(vdevs); 1299 return (EINVAL); 1300 } 1301 1302 spa->spa_root_vdev->v_nchildren = vdev_children; 1303 1304 rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1305 &nkids, &kids, NULL); 1306 nvlist_destroy(vdevs); 1307 1308 /* 1309 * MOS config has at least one child for root vdev. 1310 */ 1311 if (rc != 0) 1312 return (rc); 1313 1314 for (int i = 0; i < nkids; i++) { 1315 uint64_t guid; 1316 vdev_t *vdev; 1317 1318 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64, 1319 NULL, &guid, NULL); 1320 if (rc != 0) 1321 break; 1322 vdev = vdev_find(guid); 1323 /* 1324 * Top level vdev is missing, create it. 1325 */ 1326 if (vdev == NULL) 1327 rc = vdev_from_nvlist(spa, guid, kids[i]); 1328 else 1329 rc = vdev_update_from_nvlist(guid, kids[i]); 1330 if (rc != 0) 1331 break; 1332 } 1333 if (kids != NULL) { 1334 for (int i = 0; i < nkids; i++) 1335 nvlist_destroy(kids[i]); 1336 free(kids); 1337 } 1338 1339 /* 1340 * Re-evaluate top-level vdev state. 1341 */ 1342 vdev_set_state(spa->spa_root_vdev); 1343 1344 return (rc); 1345 } 1346 1347 static spa_t * 1348 spa_find_by_guid(uint64_t guid) 1349 { 1350 spa_t *spa; 1351 1352 STAILQ_FOREACH(spa, &zfs_pools, spa_link) 1353 if (spa->spa_guid == guid) 1354 return (spa); 1355 1356 return (NULL); 1357 } 1358 1359 static spa_t * 1360 spa_find_by_name(const char *name) 1361 { 1362 spa_t *spa; 1363 1364 STAILQ_FOREACH(spa, &zfs_pools, spa_link) 1365 if (strcmp(spa->spa_name, name) == 0) 1366 return (spa); 1367 1368 return (NULL); 1369 } 1370 1371 static spa_t * 1372 spa_create(uint64_t guid, const char *name) 1373 { 1374 spa_t *spa; 1375 1376 if ((spa = calloc(1, sizeof(spa_t))) == NULL) 1377 return (NULL); 1378 if ((spa->spa_name = strdup(name)) == NULL) { 1379 free(spa); 1380 return (NULL); 1381 } 1382 spa->spa_uberblock = &spa->spa_uberblock_master; 1383 spa->spa_mos = &spa->spa_mos_master; 1384 spa->spa_guid = guid; 1385 spa->spa_root_vdev = vdev_create(guid, NULL); 1386 if (spa->spa_root_vdev == NULL) { 1387 free(spa->spa_name); 1388 free(spa); 1389 return (NULL); 1390 } 1391 spa->spa_root_vdev->v_name = strdup("root"); 1392 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link); 1393 1394 return (spa); 1395 } 1396 1397 static const char * 1398 state_name(vdev_state_t state) 1399 { 1400 static const char *names[] = { 1401 "UNKNOWN", 1402 "CLOSED", 1403 "OFFLINE", 1404 "REMOVED", 1405 "CANT_OPEN", 1406 "FAULTED", 1407 "DEGRADED", 1408 "ONLINE" 1409 }; 1410 return (names[state]); 1411 } 1412 1413 #ifdef BOOT2 1414 1415 #define pager_printf printf 1416 1417 #else 1418 1419 static int 1420 pager_printf(const char *fmt, ...) 1421 { 1422 char line[80]; 1423 va_list args; 1424 1425 va_start(args, fmt); 1426 vsnprintf(line, sizeof(line), fmt, args); 1427 va_end(args); 1428 return (pager_output(line)); 1429 } 1430 1431 #endif 1432 1433 #define STATUS_FORMAT " %s %s\n" 1434 1435 static int 1436 print_state(int indent, const char *name, vdev_state_t state) 1437 { 1438 int i; 1439 char buf[512]; 1440 1441 buf[0] = 0; 1442 for (i = 0; i < indent; i++) 1443 strcat(buf, " "); 1444 strcat(buf, name); 1445 return (pager_printf(STATUS_FORMAT, buf, state_name(state))); 1446 } 1447 1448 static int 1449 vdev_status(vdev_t *vdev, int indent) 1450 { 1451 vdev_t *kid; 1452 int ret; 1453 1454 if (vdev->v_islog) { 1455 (void) pager_output(" logs\n"); 1456 indent++; 1457 } 1458 1459 ret = print_state(indent, vdev->v_name, vdev->v_state); 1460 if (ret != 0) 1461 return (ret); 1462 1463 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1464 ret = vdev_status(kid, indent + 1); 1465 if (ret != 0) 1466 return (ret); 1467 } 1468 return (ret); 1469 } 1470 1471 static int 1472 spa_status(spa_t *spa) 1473 { 1474 static char bootfs[ZFS_MAXNAMELEN]; 1475 uint64_t rootid; 1476 vdev_list_t *vlist; 1477 vdev_t *vdev; 1478 int good_kids, bad_kids, degraded_kids, ret; 1479 vdev_state_t state; 1480 1481 ret = pager_printf(" pool: %s\n", spa->spa_name); 1482 if (ret != 0) 1483 return (ret); 1484 1485 if (zfs_get_root(spa, &rootid) == 0 && 1486 zfs_rlookup(spa, rootid, bootfs) == 0) { 1487 if (bootfs[0] == '\0') 1488 ret = pager_printf("bootfs: %s\n", spa->spa_name); 1489 else 1490 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name, 1491 bootfs); 1492 if (ret != 0) 1493 return (ret); 1494 } 1495 ret = pager_printf("config:\n\n"); 1496 if (ret != 0) 1497 return (ret); 1498 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE"); 1499 if (ret != 0) 1500 return (ret); 1501 1502 good_kids = 0; 1503 degraded_kids = 0; 1504 bad_kids = 0; 1505 vlist = &spa->spa_root_vdev->v_children; 1506 STAILQ_FOREACH(vdev, vlist, v_childlink) { 1507 if (vdev->v_state == VDEV_STATE_HEALTHY) 1508 good_kids++; 1509 else if (vdev->v_state == VDEV_STATE_DEGRADED) 1510 degraded_kids++; 1511 else 1512 bad_kids++; 1513 } 1514 1515 state = VDEV_STATE_CLOSED; 1516 if (good_kids > 0 && (degraded_kids + bad_kids) == 0) 1517 state = VDEV_STATE_HEALTHY; 1518 else if ((good_kids + degraded_kids) > 0) 1519 state = VDEV_STATE_DEGRADED; 1520 1521 ret = print_state(0, spa->spa_name, state); 1522 if (ret != 0) 1523 return (ret); 1524 1525 STAILQ_FOREACH(vdev, vlist, v_childlink) { 1526 ret = vdev_status(vdev, 1); 1527 if (ret != 0) 1528 return (ret); 1529 } 1530 return (ret); 1531 } 1532 1533 static int 1534 spa_all_status(void) 1535 { 1536 spa_t *spa; 1537 int first = 1, ret = 0; 1538 1539 STAILQ_FOREACH(spa, &zfs_pools, spa_link) { 1540 if (!first) { 1541 ret = pager_printf("\n"); 1542 if (ret != 0) 1543 return (ret); 1544 } 1545 first = 0; 1546 ret = spa_status(spa); 1547 if (ret != 0) 1548 return (ret); 1549 } 1550 return (ret); 1551 } 1552 1553 static uint64_t 1554 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 1555 { 1556 uint64_t label_offset; 1557 1558 if (l < VDEV_LABELS / 2) 1559 label_offset = 0; 1560 else 1561 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t); 1562 1563 return (offset + l * sizeof (vdev_label_t) + label_offset); 1564 } 1565 1566 static int 1567 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2) 1568 { 1569 unsigned int seq1 = 0; 1570 unsigned int seq2 = 0; 1571 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg); 1572 1573 if (cmp != 0) 1574 return (cmp); 1575 1576 cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp); 1577 if (cmp != 0) 1578 return (cmp); 1579 1580 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1)) 1581 seq1 = MMP_SEQ(ub1); 1582 1583 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2)) 1584 seq2 = MMP_SEQ(ub2); 1585 1586 return (AVL_CMP(seq1, seq2)); 1587 } 1588 1589 static int 1590 uberblock_verify(uberblock_t *ub) 1591 { 1592 if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) { 1593 byteswap_uint64_array(ub, sizeof (uberblock_t)); 1594 } 1595 1596 if (ub->ub_magic != UBERBLOCK_MAGIC || 1597 !SPA_VERSION_IS_SUPPORTED(ub->ub_version)) 1598 return (EINVAL); 1599 1600 return (0); 1601 } 1602 1603 static int 1604 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset, 1605 size_t size) 1606 { 1607 blkptr_t bp; 1608 off_t off; 1609 1610 off = vdev_label_offset(vd->v_psize, l, offset); 1611 1612 BP_ZERO(&bp); 1613 BP_SET_LSIZE(&bp, size); 1614 BP_SET_PSIZE(&bp, size); 1615 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 1616 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF); 1617 DVA_SET_OFFSET(BP_IDENTITY(&bp), off); 1618 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0); 1619 1620 return (vdev_read_phys(vd, &bp, buf, off, size)); 1621 } 1622 1623 /* 1624 * We do need to be sure we write to correct location. 1625 * Our vdev label does consist of 4 fields: 1626 * pad1 (8k), reserved. 1627 * bootenv (8k), checksummed, previously reserved, may contian garbage. 1628 * vdev_phys (112k), checksummed 1629 * uberblock ring (128k), checksummed. 1630 * 1631 * Since bootenv area may contain garbage, we can not reliably read it, as 1632 * we can get checksum errors. 1633 * Next best thing is vdev_phys - it is just after bootenv. It still may 1634 * be corrupted, but in such case we will miss this one write. 1635 */ 1636 static int 1637 vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset) 1638 { 1639 uint64_t off, o_phys; 1640 void *buf; 1641 size_t size = VDEV_PHYS_SIZE; 1642 int rc; 1643 1644 o_phys = offsetof(vdev_label_t, vl_vdev_phys); 1645 off = vdev_label_offset(vd->v_psize, l, o_phys); 1646 1647 /* off should be 8K from bootenv */ 1648 if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off) 1649 return (EINVAL); 1650 1651 buf = malloc(size); 1652 if (buf == NULL) 1653 return (ENOMEM); 1654 1655 /* Read vdev_phys */ 1656 rc = vdev_label_read(vd, l, buf, o_phys, size); 1657 free(buf); 1658 return (rc); 1659 } 1660 1661 static int 1662 vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset) 1663 { 1664 zio_checksum_info_t *ci; 1665 zio_cksum_t cksum; 1666 off_t off; 1667 size_t size = VDEV_PAD_SIZE; 1668 int rc; 1669 1670 if (vd->v_phys_write == NULL) 1671 return (ENOTSUP); 1672 1673 off = vdev_label_offset(vd->v_psize, l, offset); 1674 1675 rc = vdev_label_write_validate(vd, l, offset); 1676 if (rc != 0) { 1677 return (rc); 1678 } 1679 1680 ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL]; 1681 be->vbe_zbt.zec_magic = ZEC_MAGIC; 1682 zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off); 1683 ci->ci_func[0](be, size, NULL, &cksum); 1684 be->vbe_zbt.zec_cksum = cksum; 1685 1686 return (vdev_write_phys(vd, be, off, size)); 1687 } 1688 1689 static int 1690 vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be) 1691 { 1692 vdev_t *kid; 1693 int rv = 0, rc; 1694 1695 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1696 if (kid->v_state != VDEV_STATE_HEALTHY) 1697 continue; 1698 rc = vdev_write_bootenv_impl(kid, be); 1699 if (rv == 0) 1700 rv = rc; 1701 } 1702 1703 /* 1704 * Non-leaf vdevs do not have v_phys_write. 1705 */ 1706 if (vdev->v_phys_write == NULL) 1707 return (rv); 1708 1709 for (int l = 0; l < VDEV_LABELS; l++) { 1710 rc = vdev_label_write(vdev, l, be, 1711 offsetof(vdev_label_t, vl_be)); 1712 if (rc != 0) { 1713 printf("failed to write bootenv to %s label %d: %d\n", 1714 vdev->v_name ? vdev->v_name : "unknown", l, rc); 1715 rv = rc; 1716 } 1717 } 1718 return (rv); 1719 } 1720 1721 int 1722 vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl) 1723 { 1724 vdev_boot_envblock_t *be; 1725 nvlist_t nv, *nvp; 1726 uint64_t version; 1727 int rv; 1728 1729 if (nvl->nv_size > sizeof(be->vbe_bootenv)) 1730 return (E2BIG); 1731 1732 version = VB_RAW; 1733 nvp = vdev_read_bootenv(vdev); 1734 if (nvp != NULL) { 1735 nvlist_find(nvp, BOOTENV_VERSION, DATA_TYPE_UINT64, NULL, 1736 &version, NULL); 1737 nvlist_destroy(nvp); 1738 } 1739 1740 be = calloc(1, sizeof(*be)); 1741 if (be == NULL) 1742 return (ENOMEM); 1743 1744 be->vbe_version = version; 1745 switch (version) { 1746 case VB_RAW: 1747 /* 1748 * If there is no envmap, we will just wipe bootenv. 1749 */ 1750 nvlist_find(nvl, GRUB_ENVMAP, DATA_TYPE_STRING, NULL, 1751 be->vbe_bootenv, NULL); 1752 rv = 0; 1753 break; 1754 1755 case VB_NVLIST: 1756 nv.nv_header = nvl->nv_header; 1757 nv.nv_asize = nvl->nv_asize; 1758 nv.nv_size = nvl->nv_size; 1759 1760 bcopy(&nv.nv_header, be->vbe_bootenv, sizeof(nv.nv_header)); 1761 nv.nv_data = be->vbe_bootenv + sizeof(nvs_header_t); 1762 bcopy(nvl->nv_data, nv.nv_data, nv.nv_size); 1763 rv = nvlist_export(&nv); 1764 break; 1765 1766 default: 1767 rv = EINVAL; 1768 break; 1769 } 1770 1771 if (rv == 0) { 1772 be->vbe_version = htobe64(be->vbe_version); 1773 rv = vdev_write_bootenv_impl(vdev, be); 1774 } 1775 free(be); 1776 return (rv); 1777 } 1778 1779 /* 1780 * Read the bootenv area from pool label, return the nvlist from it. 1781 * We return from first successful read. 1782 */ 1783 nvlist_t * 1784 vdev_read_bootenv(vdev_t *vdev) 1785 { 1786 vdev_t *kid; 1787 nvlist_t *benv; 1788 vdev_boot_envblock_t *be; 1789 char *command; 1790 bool ok; 1791 int rv; 1792 1793 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1794 if (kid->v_state != VDEV_STATE_HEALTHY) 1795 continue; 1796 1797 benv = vdev_read_bootenv(kid); 1798 if (benv != NULL) 1799 return (benv); 1800 } 1801 1802 be = malloc(sizeof (*be)); 1803 if (be == NULL) 1804 return (NULL); 1805 1806 rv = 0; 1807 for (int l = 0; l < VDEV_LABELS; l++) { 1808 rv = vdev_label_read(vdev, l, be, 1809 offsetof(vdev_label_t, vl_be), 1810 sizeof (*be)); 1811 if (rv == 0) 1812 break; 1813 } 1814 if (rv != 0) { 1815 free(be); 1816 return (NULL); 1817 } 1818 1819 be->vbe_version = be64toh(be->vbe_version); 1820 switch (be->vbe_version) { 1821 case VB_RAW: 1822 /* 1823 * we have textual data in vbe_bootenv, create nvlist 1824 * with key "envmap". 1825 */ 1826 benv = nvlist_create(NV_UNIQUE_NAME); 1827 if (benv != NULL) { 1828 if (*be->vbe_bootenv == '\0') { 1829 nvlist_add_uint64(benv, BOOTENV_VERSION, 1830 VB_NVLIST); 1831 break; 1832 } 1833 nvlist_add_uint64(benv, BOOTENV_VERSION, VB_RAW); 1834 be->vbe_bootenv[sizeof (be->vbe_bootenv) - 1] = '\0'; 1835 nvlist_add_string(benv, GRUB_ENVMAP, be->vbe_bootenv); 1836 } 1837 break; 1838 1839 case VB_NVLIST: 1840 benv = nvlist_import(be->vbe_bootenv, sizeof(be->vbe_bootenv)); 1841 break; 1842 1843 default: 1844 command = (char *)be; 1845 ok = false; 1846 1847 /* Check for legacy zfsbootcfg command string */ 1848 for (int i = 0; command[i] != '\0'; i++) { 1849 if (iscntrl(command[i])) { 1850 ok = false; 1851 break; 1852 } else { 1853 ok = true; 1854 } 1855 } 1856 benv = nvlist_create(NV_UNIQUE_NAME); 1857 if (benv != NULL) { 1858 if (ok) 1859 nvlist_add_string(benv, FREEBSD_BOOTONCE, 1860 command); 1861 else 1862 nvlist_add_uint64(benv, BOOTENV_VERSION, 1863 VB_NVLIST); 1864 } 1865 break; 1866 } 1867 free(be); 1868 return (benv); 1869 } 1870 1871 static uint64_t 1872 vdev_get_label_asize(nvlist_t *nvl) 1873 { 1874 nvlist_t *vdevs; 1875 uint64_t asize; 1876 const char *type; 1877 int len; 1878 1879 asize = 0; 1880 /* Get vdev tree */ 1881 if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1882 NULL, &vdevs, NULL) != 0) 1883 return (asize); 1884 1885 /* 1886 * Get vdev type. We will calculate asize for raidz, mirror and disk. 1887 * For raidz, the asize is raw size of all children. 1888 */ 1889 if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, 1890 NULL, &type, &len) != 0) 1891 goto done; 1892 1893 if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 && 1894 memcmp(type, VDEV_TYPE_DISK, len) != 0 && 1895 memcmp(type, VDEV_TYPE_RAIDZ, len) != 0) 1896 goto done; 1897 1898 if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64, 1899 NULL, &asize, NULL) != 0) 1900 goto done; 1901 1902 if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) { 1903 nvlist_t **kids; 1904 int nkids; 1905 1906 if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, 1907 DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) { 1908 asize = 0; 1909 goto done; 1910 } 1911 1912 asize /= nkids; 1913 for (int i = 0; i < nkids; i++) 1914 nvlist_destroy(kids[i]); 1915 free(kids); 1916 } 1917 1918 asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 1919 done: 1920 nvlist_destroy(vdevs); 1921 return (asize); 1922 } 1923 1924 static nvlist_t * 1925 vdev_label_read_config(vdev_t *vd, uint64_t txg) 1926 { 1927 vdev_phys_t *label; 1928 uint64_t best_txg = 0; 1929 uint64_t label_txg = 0; 1930 uint64_t asize; 1931 nvlist_t *nvl = NULL, *tmp; 1932 int error; 1933 1934 label = malloc(sizeof (vdev_phys_t)); 1935 if (label == NULL) 1936 return (NULL); 1937 1938 for (int l = 0; l < VDEV_LABELS; l++) { 1939 if (vdev_label_read(vd, l, label, 1940 offsetof(vdev_label_t, vl_vdev_phys), 1941 sizeof (vdev_phys_t))) 1942 continue; 1943 1944 tmp = nvlist_import(label->vp_nvlist, 1945 sizeof(label->vp_nvlist)); 1946 if (tmp == NULL) 1947 continue; 1948 1949 error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG, 1950 DATA_TYPE_UINT64, NULL, &label_txg, NULL); 1951 if (error != 0 || label_txg == 0) { 1952 nvlist_destroy(nvl); 1953 nvl = tmp; 1954 goto done; 1955 } 1956 1957 if (label_txg <= txg && label_txg > best_txg) { 1958 best_txg = label_txg; 1959 nvlist_destroy(nvl); 1960 nvl = tmp; 1961 tmp = NULL; 1962 1963 /* 1964 * Use asize from pool config. We need this 1965 * because we can get bad value from BIOS. 1966 */ 1967 asize = vdev_get_label_asize(nvl); 1968 if (asize != 0) { 1969 vd->v_psize = asize; 1970 } 1971 } 1972 nvlist_destroy(tmp); 1973 } 1974 1975 if (best_txg == 0) { 1976 nvlist_destroy(nvl); 1977 nvl = NULL; 1978 } 1979 done: 1980 free(label); 1981 return (nvl); 1982 } 1983 1984 static void 1985 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub) 1986 { 1987 uberblock_t *buf; 1988 1989 buf = malloc(VDEV_UBERBLOCK_SIZE(vd)); 1990 if (buf == NULL) 1991 return; 1992 1993 for (int l = 0; l < VDEV_LABELS; l++) { 1994 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1995 if (vdev_label_read(vd, l, buf, 1996 VDEV_UBERBLOCK_OFFSET(vd, n), 1997 VDEV_UBERBLOCK_SIZE(vd))) 1998 continue; 1999 if (uberblock_verify(buf) != 0) 2000 continue; 2001 2002 if (vdev_uberblock_compare(buf, ub) > 0) 2003 *ub = *buf; 2004 } 2005 } 2006 free(buf); 2007 } 2008 2009 static int 2010 vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv, 2011 spa_t **spap) 2012 { 2013 vdev_t vtmp; 2014 spa_t *spa; 2015 vdev_t *vdev; 2016 nvlist_t *nvl; 2017 uint64_t val; 2018 uint64_t guid, vdev_children; 2019 uint64_t pool_txg, pool_guid; 2020 const char *pool_name; 2021 int rc, namelen; 2022 2023 /* 2024 * Load the vdev label and figure out which 2025 * uberblock is most current. 2026 */ 2027 memset(&vtmp, 0, sizeof(vtmp)); 2028 vtmp.v_phys_read = _read; 2029 vtmp.v_phys_write = _write; 2030 vtmp.v_priv = priv; 2031 vtmp.v_psize = P2ALIGN(ldi_get_size(priv), 2032 (uint64_t)sizeof (vdev_label_t)); 2033 2034 /* Test for minimum device size. */ 2035 if (vtmp.v_psize < SPA_MINDEVSIZE) 2036 return (EIO); 2037 2038 nvl = vdev_label_read_config(&vtmp, UINT64_MAX); 2039 if (nvl == NULL) 2040 return (EIO); 2041 2042 if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64, 2043 NULL, &val, NULL) != 0) { 2044 nvlist_destroy(nvl); 2045 return (EIO); 2046 } 2047 2048 if (!SPA_VERSION_IS_SUPPORTED(val)) { 2049 printf("ZFS: unsupported ZFS version %u (should be %u)\n", 2050 (unsigned)val, (unsigned)SPA_VERSION); 2051 nvlist_destroy(nvl); 2052 return (EIO); 2053 } 2054 2055 /* Check ZFS features for read */ 2056 rc = nvlist_check_features_for_read(nvl); 2057 if (rc != 0) { 2058 nvlist_destroy(nvl); 2059 return (EIO); 2060 } 2061 2062 if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64, 2063 NULL, &val, NULL) != 0) { 2064 nvlist_destroy(nvl); 2065 return (EIO); 2066 } 2067 2068 if (val == POOL_STATE_DESTROYED) { 2069 /* We don't boot only from destroyed pools. */ 2070 nvlist_destroy(nvl); 2071 return (EIO); 2072 } 2073 2074 if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64, 2075 NULL, &pool_txg, NULL) != 0 || 2076 nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 2077 NULL, &pool_guid, NULL) != 0 || 2078 nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING, 2079 NULL, &pool_name, &namelen) != 0) { 2080 /* 2081 * Cache and spare devices end up here - just ignore 2082 * them. 2083 */ 2084 nvlist_destroy(nvl); 2085 return (EIO); 2086 } 2087 2088 /* 2089 * Create the pool if this is the first time we've seen it. 2090 */ 2091 spa = spa_find_by_guid(pool_guid); 2092 if (spa == NULL) { 2093 char *name; 2094 2095 nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN, 2096 DATA_TYPE_UINT64, NULL, &vdev_children, NULL); 2097 name = malloc(namelen + 1); 2098 if (name == NULL) { 2099 nvlist_destroy(nvl); 2100 return (ENOMEM); 2101 } 2102 bcopy(pool_name, name, namelen); 2103 name[namelen] = '\0'; 2104 spa = spa_create(pool_guid, name); 2105 free(name); 2106 if (spa == NULL) { 2107 nvlist_destroy(nvl); 2108 return (ENOMEM); 2109 } 2110 spa->spa_root_vdev->v_nchildren = vdev_children; 2111 } 2112 if (pool_txg > spa->spa_txg) 2113 spa->spa_txg = pool_txg; 2114 2115 /* 2116 * Get the vdev tree and create our in-core copy of it. 2117 * If we already have a vdev with this guid, this must 2118 * be some kind of alias (overlapping slices, dangerously dedicated 2119 * disks etc). 2120 */ 2121 if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64, 2122 NULL, &guid, NULL) != 0) { 2123 nvlist_destroy(nvl); 2124 return (EIO); 2125 } 2126 vdev = vdev_find(guid); 2127 /* Has this vdev already been inited? */ 2128 if (vdev && vdev->v_phys_read) { 2129 nvlist_destroy(nvl); 2130 return (EIO); 2131 } 2132 2133 rc = vdev_init_from_label(spa, nvl); 2134 nvlist_destroy(nvl); 2135 if (rc != 0) 2136 return (rc); 2137 2138 /* 2139 * We should already have created an incomplete vdev for this 2140 * vdev. Find it and initialise it with our read proc. 2141 */ 2142 vdev = vdev_find(guid); 2143 if (vdev != NULL) { 2144 vdev->v_phys_read = _read; 2145 vdev->v_phys_write = _write; 2146 vdev->v_priv = priv; 2147 vdev->v_psize = vtmp.v_psize; 2148 /* 2149 * If no other state is set, mark vdev healthy. 2150 */ 2151 if (vdev->v_state == VDEV_STATE_UNKNOWN) 2152 vdev->v_state = VDEV_STATE_HEALTHY; 2153 } else { 2154 printf("ZFS: inconsistent nvlist contents\n"); 2155 return (EIO); 2156 } 2157 2158 if (vdev->v_islog) 2159 spa->spa_with_log = vdev->v_islog; 2160 2161 /* 2162 * Re-evaluate top-level vdev state. 2163 */ 2164 vdev_set_state(vdev->v_top); 2165 2166 /* 2167 * Ok, we are happy with the pool so far. Lets find 2168 * the best uberblock and then we can actually access 2169 * the contents of the pool. 2170 */ 2171 vdev_uberblock_load(vdev, spa->spa_uberblock); 2172 2173 if (spap != NULL) 2174 *spap = spa; 2175 return (0); 2176 } 2177 2178 static int 2179 ilog2(int n) 2180 { 2181 int v; 2182 2183 for (v = 0; v < 32; v++) 2184 if (n == (1 << v)) 2185 return (v); 2186 return (-1); 2187 } 2188 2189 static int 2190 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf) 2191 { 2192 blkptr_t gbh_bp; 2193 zio_gbh_phys_t zio_gb; 2194 char *pbuf; 2195 int i; 2196 2197 /* Artificial BP for gang block header. */ 2198 gbh_bp = *bp; 2199 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); 2200 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); 2201 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER); 2202 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF); 2203 for (i = 0; i < SPA_DVAS_PER_BP; i++) 2204 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0); 2205 2206 /* Read gang header block using the artificial BP. */ 2207 if (zio_read(spa, &gbh_bp, &zio_gb)) 2208 return (EIO); 2209 2210 pbuf = buf; 2211 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) { 2212 blkptr_t *gbp = &zio_gb.zg_blkptr[i]; 2213 2214 if (BP_IS_HOLE(gbp)) 2215 continue; 2216 if (zio_read(spa, gbp, pbuf)) 2217 return (EIO); 2218 pbuf += BP_GET_PSIZE(gbp); 2219 } 2220 2221 if (zio_checksum_verify(spa, bp, buf)) 2222 return (EIO); 2223 return (0); 2224 } 2225 2226 static int 2227 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) 2228 { 2229 int cpfunc = BP_GET_COMPRESS(bp); 2230 uint64_t align, size; 2231 void *pbuf; 2232 int i, error; 2233 2234 /* 2235 * Process data embedded in block pointer 2236 */ 2237 if (BP_IS_EMBEDDED(bp)) { 2238 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); 2239 2240 size = BPE_GET_PSIZE(bp); 2241 ASSERT(size <= BPE_PAYLOAD_SIZE); 2242 2243 if (cpfunc != ZIO_COMPRESS_OFF) 2244 pbuf = malloc(size); 2245 else 2246 pbuf = buf; 2247 2248 if (pbuf == NULL) 2249 return (ENOMEM); 2250 2251 decode_embedded_bp_compressed(bp, pbuf); 2252 error = 0; 2253 2254 if (cpfunc != ZIO_COMPRESS_OFF) { 2255 error = zio_decompress_data(cpfunc, pbuf, 2256 size, buf, BP_GET_LSIZE(bp)); 2257 free(pbuf); 2258 } 2259 if (error != 0) 2260 printf("ZFS: i/o error - unable to decompress " 2261 "block pointer data, error %d\n", error); 2262 return (error); 2263 } 2264 2265 error = EIO; 2266 2267 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 2268 const dva_t *dva = &bp->blk_dva[i]; 2269 vdev_t *vdev; 2270 vdev_list_t *vlist; 2271 uint64_t vdevid; 2272 off_t offset; 2273 2274 if (!dva->dva_word[0] && !dva->dva_word[1]) 2275 continue; 2276 2277 vdevid = DVA_GET_VDEV(dva); 2278 offset = DVA_GET_OFFSET(dva); 2279 vlist = &spa->spa_root_vdev->v_children; 2280 STAILQ_FOREACH(vdev, vlist, v_childlink) { 2281 if (vdev->v_id == vdevid) 2282 break; 2283 } 2284 if (!vdev || !vdev->v_read) 2285 continue; 2286 2287 size = BP_GET_PSIZE(bp); 2288 if (vdev->v_read == vdev_raidz_read) { 2289 align = 1ULL << vdev->v_ashift; 2290 if (P2PHASE(size, align) != 0) 2291 size = P2ROUNDUP(size, align); 2292 } 2293 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF) 2294 pbuf = malloc(size); 2295 else 2296 pbuf = buf; 2297 2298 if (pbuf == NULL) { 2299 error = ENOMEM; 2300 break; 2301 } 2302 2303 if (DVA_GET_GANG(dva)) 2304 error = zio_read_gang(spa, bp, pbuf); 2305 else 2306 error = vdev->v_read(vdev, bp, pbuf, offset, size); 2307 if (error == 0) { 2308 if (cpfunc != ZIO_COMPRESS_OFF) 2309 error = zio_decompress_data(cpfunc, pbuf, 2310 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp)); 2311 else if (size != BP_GET_PSIZE(bp)) 2312 bcopy(pbuf, buf, BP_GET_PSIZE(bp)); 2313 } else { 2314 printf("zio_read error: %d\n", error); 2315 } 2316 if (buf != pbuf) 2317 free(pbuf); 2318 if (error == 0) 2319 break; 2320 } 2321 if (error != 0) 2322 printf("ZFS: i/o error - all block copies unavailable\n"); 2323 2324 return (error); 2325 } 2326 2327 static int 2328 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, 2329 void *buf, size_t buflen) 2330 { 2331 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT; 2332 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2333 int nlevels = dnode->dn_nlevels; 2334 int i, rc; 2335 2336 if (bsize > SPA_MAXBLOCKSIZE) { 2337 printf("ZFS: I/O error - blocks larger than %llu are not " 2338 "supported\n", SPA_MAXBLOCKSIZE); 2339 return (EIO); 2340 } 2341 2342 /* 2343 * Handle odd block sizes, mirrors dmu_read_impl(). Data can't exist 2344 * past the first block, so we'll clip the read to the portion of the 2345 * buffer within bsize and zero out the remainder. 2346 */ 2347 if (dnode->dn_maxblkid == 0) { 2348 size_t newbuflen; 2349 2350 newbuflen = offset > bsize ? 0 : MIN(buflen, bsize - offset); 2351 bzero((char *)buf + newbuflen, buflen - newbuflen); 2352 buflen = newbuflen; 2353 } 2354 2355 /* 2356 * Note: bsize may not be a power of two here so we need to do an 2357 * actual divide rather than a bitshift. 2358 */ 2359 while (buflen > 0) { 2360 uint64_t bn = offset / bsize; 2361 int boff = offset % bsize; 2362 int ibn; 2363 const blkptr_t *indbp; 2364 blkptr_t bp; 2365 2366 if (bn > dnode->dn_maxblkid) 2367 return (EIO); 2368 2369 if (dnode == dnode_cache_obj && bn == dnode_cache_bn) 2370 goto cached; 2371 2372 indbp = dnode->dn_blkptr; 2373 for (i = 0; i < nlevels; i++) { 2374 /* 2375 * Copy the bp from the indirect array so that 2376 * we can re-use the scratch buffer for multi-level 2377 * objects. 2378 */ 2379 ibn = bn >> ((nlevels - i - 1) * ibshift); 2380 ibn &= ((1 << ibshift) - 1); 2381 bp = indbp[ibn]; 2382 if (BP_IS_HOLE(&bp)) { 2383 memset(dnode_cache_buf, 0, bsize); 2384 break; 2385 } 2386 rc = zio_read(spa, &bp, dnode_cache_buf); 2387 if (rc) 2388 return (rc); 2389 indbp = (const blkptr_t *) dnode_cache_buf; 2390 } 2391 dnode_cache_obj = dnode; 2392 dnode_cache_bn = bn; 2393 cached: 2394 2395 /* 2396 * The buffer contains our data block. Copy what we 2397 * need from it and loop. 2398 */ 2399 i = bsize - boff; 2400 if (i > buflen) i = buflen; 2401 memcpy(buf, &dnode_cache_buf[boff], i); 2402 buf = ((char *)buf) + i; 2403 offset += i; 2404 buflen -= i; 2405 } 2406 2407 return (0); 2408 } 2409 2410 /* 2411 * Lookup a value in a microzap directory. 2412 */ 2413 static int 2414 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name, 2415 uint64_t *value) 2416 { 2417 const mzap_ent_phys_t *mze; 2418 int chunks, i; 2419 2420 /* 2421 * Microzap objects use exactly one block. Read the whole 2422 * thing. 2423 */ 2424 chunks = size / MZAP_ENT_LEN - 1; 2425 for (i = 0; i < chunks; i++) { 2426 mze = &mz->mz_chunk[i]; 2427 if (strcmp(mze->mze_name, name) == 0) { 2428 *value = mze->mze_value; 2429 return (0); 2430 } 2431 } 2432 2433 return (ENOENT); 2434 } 2435 2436 /* 2437 * Compare a name with a zap leaf entry. Return non-zero if the name 2438 * matches. 2439 */ 2440 static int 2441 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, 2442 const char *name) 2443 { 2444 size_t namelen; 2445 const zap_leaf_chunk_t *nc; 2446 const char *p; 2447 2448 namelen = zc->l_entry.le_name_numints; 2449 2450 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk); 2451 p = name; 2452 while (namelen > 0) { 2453 size_t len; 2454 2455 len = namelen; 2456 if (len > ZAP_LEAF_ARRAY_BYTES) 2457 len = ZAP_LEAF_ARRAY_BYTES; 2458 if (memcmp(p, nc->l_array.la_array, len)) 2459 return (0); 2460 p += len; 2461 namelen -= len; 2462 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next); 2463 } 2464 2465 return (1); 2466 } 2467 2468 /* 2469 * Extract a uint64_t value from a zap leaf entry. 2470 */ 2471 static uint64_t 2472 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc) 2473 { 2474 const zap_leaf_chunk_t *vc; 2475 int i; 2476 uint64_t value; 2477 const uint8_t *p; 2478 2479 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk); 2480 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) { 2481 value = (value << 8) | p[i]; 2482 } 2483 2484 return (value); 2485 } 2486 2487 static void 2488 stv(int len, void *addr, uint64_t value) 2489 { 2490 switch (len) { 2491 case 1: 2492 *(uint8_t *)addr = value; 2493 return; 2494 case 2: 2495 *(uint16_t *)addr = value; 2496 return; 2497 case 4: 2498 *(uint32_t *)addr = value; 2499 return; 2500 case 8: 2501 *(uint64_t *)addr = value; 2502 return; 2503 } 2504 } 2505 2506 /* 2507 * Extract a array from a zap leaf entry. 2508 */ 2509 static void 2510 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, 2511 uint64_t integer_size, uint64_t num_integers, void *buf) 2512 { 2513 uint64_t array_int_len = zc->l_entry.le_value_intlen; 2514 uint64_t value = 0; 2515 uint64_t *u64 = buf; 2516 char *p = buf; 2517 int len = MIN(zc->l_entry.le_value_numints, num_integers); 2518 int chunk = zc->l_entry.le_value_chunk; 2519 int byten = 0; 2520 2521 if (integer_size == 8 && len == 1) { 2522 *u64 = fzap_leaf_value(zl, zc); 2523 return; 2524 } 2525 2526 while (len > 0) { 2527 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array; 2528 int i; 2529 2530 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl)); 2531 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) { 2532 value = (value << 8) | la->la_array[i]; 2533 byten++; 2534 if (byten == array_int_len) { 2535 stv(integer_size, p, value); 2536 byten = 0; 2537 len--; 2538 if (len == 0) 2539 return; 2540 p += integer_size; 2541 } 2542 } 2543 chunk = la->la_next; 2544 } 2545 } 2546 2547 static int 2548 fzap_check_size(uint64_t integer_size, uint64_t num_integers) 2549 { 2550 2551 switch (integer_size) { 2552 case 1: 2553 case 2: 2554 case 4: 2555 case 8: 2556 break; 2557 default: 2558 return (EINVAL); 2559 } 2560 2561 if (integer_size * num_integers > ZAP_MAXVALUELEN) 2562 return (E2BIG); 2563 2564 return (0); 2565 } 2566 2567 static void 2568 zap_leaf_free(zap_leaf_t *leaf) 2569 { 2570 free(leaf->l_phys); 2571 free(leaf); 2572 } 2573 2574 static int 2575 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp) 2576 { 2577 int bs = FZAP_BLOCK_SHIFT(zap); 2578 int err; 2579 2580 *lp = malloc(sizeof(**lp)); 2581 if (*lp == NULL) 2582 return (ENOMEM); 2583 2584 (*lp)->l_bs = bs; 2585 (*lp)->l_phys = malloc(1 << bs); 2586 2587 if ((*lp)->l_phys == NULL) { 2588 free(*lp); 2589 return (ENOMEM); 2590 } 2591 err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys, 2592 1 << bs); 2593 if (err != 0) { 2594 zap_leaf_free(*lp); 2595 } 2596 return (err); 2597 } 2598 2599 static int 2600 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, 2601 uint64_t *valp) 2602 { 2603 int bs = FZAP_BLOCK_SHIFT(zap); 2604 uint64_t blk = idx >> (bs - 3); 2605 uint64_t off = idx & ((1 << (bs - 3)) - 1); 2606 uint64_t *buf; 2607 int rc; 2608 2609 buf = malloc(1 << zap->zap_block_shift); 2610 if (buf == NULL) 2611 return (ENOMEM); 2612 rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs, 2613 buf, 1 << zap->zap_block_shift); 2614 if (rc == 0) 2615 *valp = buf[off]; 2616 free(buf); 2617 return (rc); 2618 } 2619 2620 static int 2621 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp) 2622 { 2623 if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) { 2624 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 2625 return (0); 2626 } else { 2627 return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl, 2628 idx, valp)); 2629 } 2630 } 2631 2632 #define ZAP_HASH_IDX(hash, n) (((n) == 0) ? 0 : ((hash) >> (64 - (n)))) 2633 static int 2634 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp) 2635 { 2636 uint64_t idx, blk; 2637 int err; 2638 2639 idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift); 2640 err = zap_idx_to_blk(zap, idx, &blk); 2641 if (err != 0) 2642 return (err); 2643 return (zap_get_leaf_byblk(zap, blk, lp)); 2644 } 2645 2646 #define CHAIN_END 0xffff /* end of the chunk chain */ 2647 #define LEAF_HASH(l, h) \ 2648 ((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \ 2649 ((h) >> \ 2650 (64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len))) 2651 #define LEAF_HASH_ENTPTR(l, h) (&(l)->l_phys->l_hash[LEAF_HASH(l, h)]) 2652 2653 static int 2654 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name, 2655 uint64_t integer_size, uint64_t num_integers, void *value) 2656 { 2657 int rc; 2658 uint16_t *chunkp; 2659 struct zap_leaf_entry *le; 2660 2661 /* 2662 * Make sure this chunk matches our hash. 2663 */ 2664 if (zl->l_phys->l_hdr.lh_prefix_len > 0 && 2665 zl->l_phys->l_hdr.lh_prefix != 2666 hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len)) 2667 return (EIO); 2668 2669 rc = ENOENT; 2670 for (chunkp = LEAF_HASH_ENTPTR(zl, hash); 2671 *chunkp != CHAIN_END; chunkp = &le->le_next) { 2672 zap_leaf_chunk_t *zc; 2673 uint16_t chunk = *chunkp; 2674 2675 le = ZAP_LEAF_ENTRY(zl, chunk); 2676 if (le->le_hash != hash) 2677 continue; 2678 zc = &ZAP_LEAF_CHUNK(zl, chunk); 2679 if (fzap_name_equal(zl, zc, name)) { 2680 if (zc->l_entry.le_value_intlen > integer_size) { 2681 rc = EINVAL; 2682 } else { 2683 fzap_leaf_array(zl, zc, integer_size, 2684 num_integers, value); 2685 rc = 0; 2686 } 2687 break; 2688 } 2689 } 2690 return (rc); 2691 } 2692 2693 /* 2694 * Lookup a value in a fatzap directory. 2695 */ 2696 static int 2697 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2698 const char *name, uint64_t integer_size, uint64_t num_integers, 2699 void *value) 2700 { 2701 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2702 fat_zap_t z; 2703 zap_leaf_t *zl; 2704 uint64_t hash; 2705 int rc; 2706 2707 if (zh->zap_magic != ZAP_MAGIC) 2708 return (EIO); 2709 2710 if ((rc = fzap_check_size(integer_size, num_integers)) != 0) { 2711 return (rc); 2712 } 2713 2714 z.zap_block_shift = ilog2(bsize); 2715 z.zap_phys = zh; 2716 z.zap_spa = spa; 2717 z.zap_dnode = dnode; 2718 2719 hash = zap_hash(zh->zap_salt, name); 2720 rc = zap_deref_leaf(&z, hash, &zl); 2721 if (rc != 0) 2722 return (rc); 2723 2724 rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value); 2725 2726 zap_leaf_free(zl); 2727 return (rc); 2728 } 2729 2730 /* 2731 * Lookup a name in a zap object and return its value as a uint64_t. 2732 */ 2733 static int 2734 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, 2735 uint64_t integer_size, uint64_t num_integers, void *value) 2736 { 2737 int rc; 2738 zap_phys_t *zap; 2739 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2740 2741 zap = malloc(size); 2742 if (zap == NULL) 2743 return (ENOMEM); 2744 2745 rc = dnode_read(spa, dnode, 0, zap, size); 2746 if (rc) 2747 goto done; 2748 2749 switch (zap->zap_block_type) { 2750 case ZBT_MICRO: 2751 rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value); 2752 break; 2753 case ZBT_HEADER: 2754 rc = fzap_lookup(spa, dnode, zap, name, integer_size, 2755 num_integers, value); 2756 break; 2757 default: 2758 printf("ZFS: invalid zap_type=%" PRIx64 "\n", 2759 zap->zap_block_type); 2760 rc = EIO; 2761 } 2762 done: 2763 free(zap); 2764 return (rc); 2765 } 2766 2767 /* 2768 * List a microzap directory. 2769 */ 2770 static int 2771 mzap_list(const mzap_phys_t *mz, size_t size, 2772 int (*callback)(const char *, uint64_t)) 2773 { 2774 const mzap_ent_phys_t *mze; 2775 int chunks, i, rc; 2776 2777 /* 2778 * Microzap objects use exactly one block. Read the whole 2779 * thing. 2780 */ 2781 rc = 0; 2782 chunks = size / MZAP_ENT_LEN - 1; 2783 for (i = 0; i < chunks; i++) { 2784 mze = &mz->mz_chunk[i]; 2785 if (mze->mze_name[0]) { 2786 rc = callback(mze->mze_name, mze->mze_value); 2787 if (rc != 0) 2788 break; 2789 } 2790 } 2791 2792 return (rc); 2793 } 2794 2795 /* 2796 * List a fatzap directory. 2797 */ 2798 static int 2799 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2800 int (*callback)(const char *, uint64_t)) 2801 { 2802 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2803 fat_zap_t z; 2804 uint64_t i; 2805 int j, rc; 2806 2807 if (zh->zap_magic != ZAP_MAGIC) 2808 return (EIO); 2809 2810 z.zap_block_shift = ilog2(bsize); 2811 z.zap_phys = zh; 2812 2813 /* 2814 * This assumes that the leaf blocks start at block 1. The 2815 * documentation isn't exactly clear on this. 2816 */ 2817 zap_leaf_t zl; 2818 zl.l_bs = z.zap_block_shift; 2819 zl.l_phys = malloc(bsize); 2820 if (zl.l_phys == NULL) 2821 return (ENOMEM); 2822 2823 for (i = 0; i < zh->zap_num_leafs; i++) { 2824 off_t off = ((off_t)(i + 1)) << zl.l_bs; 2825 char name[256], *p; 2826 uint64_t value; 2827 2828 if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) { 2829 free(zl.l_phys); 2830 return (EIO); 2831 } 2832 2833 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) { 2834 zap_leaf_chunk_t *zc, *nc; 2835 int namelen; 2836 2837 zc = &ZAP_LEAF_CHUNK(&zl, j); 2838 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY) 2839 continue; 2840 namelen = zc->l_entry.le_name_numints; 2841 if (namelen > sizeof(name)) 2842 namelen = sizeof(name); 2843 2844 /* 2845 * Paste the name back together. 2846 */ 2847 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk); 2848 p = name; 2849 while (namelen > 0) { 2850 int len; 2851 len = namelen; 2852 if (len > ZAP_LEAF_ARRAY_BYTES) 2853 len = ZAP_LEAF_ARRAY_BYTES; 2854 memcpy(p, nc->l_array.la_array, len); 2855 p += len; 2856 namelen -= len; 2857 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next); 2858 } 2859 2860 /* 2861 * Assume the first eight bytes of the value are 2862 * a uint64_t. 2863 */ 2864 value = fzap_leaf_value(&zl, zc); 2865 2866 /* printf("%s 0x%jx\n", name, (uintmax_t)value); */ 2867 rc = callback((const char *)name, value); 2868 if (rc != 0) { 2869 free(zl.l_phys); 2870 return (rc); 2871 } 2872 } 2873 } 2874 2875 free(zl.l_phys); 2876 return (0); 2877 } 2878 2879 static int zfs_printf(const char *name, uint64_t value __unused) 2880 { 2881 2882 printf("%s\n", name); 2883 2884 return (0); 2885 } 2886 2887 /* 2888 * List a zap directory. 2889 */ 2890 static int 2891 zap_list(const spa_t *spa, const dnode_phys_t *dnode) 2892 { 2893 zap_phys_t *zap; 2894 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2895 int rc; 2896 2897 zap = malloc(size); 2898 if (zap == NULL) 2899 return (ENOMEM); 2900 2901 rc = dnode_read(spa, dnode, 0, zap, size); 2902 if (rc == 0) { 2903 if (zap->zap_block_type == ZBT_MICRO) 2904 rc = mzap_list((const mzap_phys_t *)zap, size, 2905 zfs_printf); 2906 else 2907 rc = fzap_list(spa, dnode, zap, zfs_printf); 2908 } 2909 free(zap); 2910 return (rc); 2911 } 2912 2913 static int 2914 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, 2915 dnode_phys_t *dnode) 2916 { 2917 off_t offset; 2918 2919 offset = objnum * sizeof(dnode_phys_t); 2920 return dnode_read(spa, &os->os_meta_dnode, offset, 2921 dnode, sizeof(dnode_phys_t)); 2922 } 2923 2924 /* 2925 * Lookup a name in a microzap directory. 2926 */ 2927 static int 2928 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value) 2929 { 2930 const mzap_ent_phys_t *mze; 2931 int chunks, i; 2932 2933 /* 2934 * Microzap objects use exactly one block. Read the whole 2935 * thing. 2936 */ 2937 chunks = size / MZAP_ENT_LEN - 1; 2938 for (i = 0; i < chunks; i++) { 2939 mze = &mz->mz_chunk[i]; 2940 if (value == mze->mze_value) { 2941 strcpy(name, mze->mze_name); 2942 return (0); 2943 } 2944 } 2945 2946 return (ENOENT); 2947 } 2948 2949 static void 2950 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name) 2951 { 2952 size_t namelen; 2953 const zap_leaf_chunk_t *nc; 2954 char *p; 2955 2956 namelen = zc->l_entry.le_name_numints; 2957 2958 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk); 2959 p = name; 2960 while (namelen > 0) { 2961 size_t len; 2962 len = namelen; 2963 if (len > ZAP_LEAF_ARRAY_BYTES) 2964 len = ZAP_LEAF_ARRAY_BYTES; 2965 memcpy(p, nc->l_array.la_array, len); 2966 p += len; 2967 namelen -= len; 2968 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next); 2969 } 2970 2971 *p = '\0'; 2972 } 2973 2974 static int 2975 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2976 char *name, uint64_t value) 2977 { 2978 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2979 fat_zap_t z; 2980 uint64_t i; 2981 int j, rc; 2982 2983 if (zh->zap_magic != ZAP_MAGIC) 2984 return (EIO); 2985 2986 z.zap_block_shift = ilog2(bsize); 2987 z.zap_phys = zh; 2988 2989 /* 2990 * This assumes that the leaf blocks start at block 1. The 2991 * documentation isn't exactly clear on this. 2992 */ 2993 zap_leaf_t zl; 2994 zl.l_bs = z.zap_block_shift; 2995 zl.l_phys = malloc(bsize); 2996 if (zl.l_phys == NULL) 2997 return (ENOMEM); 2998 2999 for (i = 0; i < zh->zap_num_leafs; i++) { 3000 off_t off = ((off_t)(i + 1)) << zl.l_bs; 3001 3002 rc = dnode_read(spa, dnode, off, zl.l_phys, bsize); 3003 if (rc != 0) 3004 goto done; 3005 3006 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) { 3007 zap_leaf_chunk_t *zc; 3008 3009 zc = &ZAP_LEAF_CHUNK(&zl, j); 3010 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY) 3011 continue; 3012 if (zc->l_entry.le_value_intlen != 8 || 3013 zc->l_entry.le_value_numints != 1) 3014 continue; 3015 3016 if (fzap_leaf_value(&zl, zc) == value) { 3017 fzap_name_copy(&zl, zc, name); 3018 goto done; 3019 } 3020 } 3021 } 3022 3023 rc = ENOENT; 3024 done: 3025 free(zl.l_phys); 3026 return (rc); 3027 } 3028 3029 static int 3030 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, 3031 uint64_t value) 3032 { 3033 zap_phys_t *zap; 3034 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 3035 int rc; 3036 3037 zap = malloc(size); 3038 if (zap == NULL) 3039 return (ENOMEM); 3040 3041 rc = dnode_read(spa, dnode, 0, zap, size); 3042 if (rc == 0) { 3043 if (zap->zap_block_type == ZBT_MICRO) 3044 rc = mzap_rlookup((const mzap_phys_t *)zap, size, 3045 name, value); 3046 else 3047 rc = fzap_rlookup(spa, dnode, zap, name, value); 3048 } 3049 free(zap); 3050 return (rc); 3051 } 3052 3053 static int 3054 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result) 3055 { 3056 char name[256]; 3057 char component[256]; 3058 uint64_t dir_obj, parent_obj, child_dir_zapobj; 3059 dnode_phys_t child_dir_zap, snapnames_zap, dataset, dir, parent; 3060 dsl_dir_phys_t *dd; 3061 dsl_dataset_phys_t *ds; 3062 char *p; 3063 int len; 3064 boolean_t issnap = B_FALSE; 3065 3066 p = &name[sizeof(name) - 1]; 3067 *p = '\0'; 3068 3069 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3070 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3071 return (EIO); 3072 } 3073 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3074 dir_obj = ds->ds_dir_obj; 3075 if (ds->ds_snapnames_zapobj == 0) 3076 issnap = B_TRUE; 3077 3078 for (;;) { 3079 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0) 3080 return (EIO); 3081 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3082 3083 /* Actual loop condition. */ 3084 parent_obj = dd->dd_parent_obj; 3085 if (parent_obj == 0) 3086 break; 3087 3088 if (objset_get_dnode(spa, spa->spa_mos, parent_obj, 3089 &parent) != 0) 3090 return (EIO); 3091 dd = (dsl_dir_phys_t *)&parent.dn_bonus; 3092 if (issnap == B_TRUE) { 3093 /* 3094 * The dataset we are looking up is a snapshot 3095 * the dir_obj is the parent already, we don't want 3096 * the grandparent just yet. Reset to the parent. 3097 */ 3098 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3099 /* Lookup the dataset to get the snapname ZAP */ 3100 if (objset_get_dnode(spa, spa->spa_mos, 3101 dd->dd_head_dataset_obj, &dataset)) 3102 return (EIO); 3103 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3104 if (objset_get_dnode(spa, spa->spa_mos, 3105 ds->ds_snapnames_zapobj, &snapnames_zap) != 0) 3106 return (EIO); 3107 /* Get the name of the snapshot */ 3108 if (zap_rlookup(spa, &snapnames_zap, component, 3109 objnum) != 0) 3110 return (EIO); 3111 len = strlen(component); 3112 p -= len; 3113 memcpy(p, component, len); 3114 --p; 3115 *p = '@'; 3116 issnap = B_FALSE; 3117 continue; 3118 } 3119 3120 child_dir_zapobj = dd->dd_child_dir_zapobj; 3121 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3122 &child_dir_zap) != 0) 3123 return (EIO); 3124 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0) 3125 return (EIO); 3126 3127 len = strlen(component); 3128 p -= len; 3129 memcpy(p, component, len); 3130 --p; 3131 *p = '/'; 3132 3133 /* Actual loop iteration. */ 3134 dir_obj = parent_obj; 3135 } 3136 3137 if (*p != '\0') 3138 ++p; 3139 strcpy(result, p); 3140 3141 return (0); 3142 } 3143 3144 static int 3145 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum) 3146 { 3147 char element[256]; 3148 uint64_t dir_obj, child_dir_zapobj; 3149 dnode_phys_t child_dir_zap, snapnames_zap, dir, dataset; 3150 dsl_dir_phys_t *dd; 3151 dsl_dataset_phys_t *ds; 3152 const char *p, *q; 3153 boolean_t issnap = B_FALSE; 3154 3155 if (objset_get_dnode(spa, spa->spa_mos, 3156 DMU_POOL_DIRECTORY_OBJECT, &dir)) 3157 return (EIO); 3158 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj), 3159 1, &dir_obj)) 3160 return (EIO); 3161 3162 p = name; 3163 for (;;) { 3164 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) 3165 return (EIO); 3166 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3167 3168 while (*p == '/') 3169 p++; 3170 /* Actual loop condition #1. */ 3171 if (*p == '\0') 3172 break; 3173 3174 q = strchr(p, '/'); 3175 if (q) { 3176 memcpy(element, p, q - p); 3177 element[q - p] = '\0'; 3178 p = q + 1; 3179 } else { 3180 strcpy(element, p); 3181 p += strlen(p); 3182 } 3183 3184 if (issnap == B_TRUE) { 3185 if (objset_get_dnode(spa, spa->spa_mos, 3186 dd->dd_head_dataset_obj, &dataset)) 3187 return (EIO); 3188 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3189 if (objset_get_dnode(spa, spa->spa_mos, 3190 ds->ds_snapnames_zapobj, &snapnames_zap) != 0) 3191 return (EIO); 3192 /* Actual loop condition #2. */ 3193 if (zap_lookup(spa, &snapnames_zap, element, 3194 sizeof (dir_obj), 1, &dir_obj) != 0) 3195 return (ENOENT); 3196 *objnum = dir_obj; 3197 return (0); 3198 } else if ((q = strchr(element, '@')) != NULL) { 3199 issnap = B_TRUE; 3200 element[q - element] = '\0'; 3201 p = q + 1; 3202 } 3203 child_dir_zapobj = dd->dd_child_dir_zapobj; 3204 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3205 &child_dir_zap) != 0) 3206 return (EIO); 3207 3208 /* Actual loop condition #2. */ 3209 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj), 3210 1, &dir_obj) != 0) 3211 return (ENOENT); 3212 } 3213 3214 *objnum = dd->dd_head_dataset_obj; 3215 return (0); 3216 } 3217 3218 #ifndef BOOT2 3219 static int 3220 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/) 3221 { 3222 uint64_t dir_obj, child_dir_zapobj; 3223 dnode_phys_t child_dir_zap, dir, dataset; 3224 dsl_dataset_phys_t *ds; 3225 dsl_dir_phys_t *dd; 3226 3227 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3228 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3229 return (EIO); 3230 } 3231 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3232 dir_obj = ds->ds_dir_obj; 3233 3234 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) { 3235 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj); 3236 return (EIO); 3237 } 3238 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3239 3240 child_dir_zapobj = dd->dd_child_dir_zapobj; 3241 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3242 &child_dir_zap) != 0) { 3243 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj); 3244 return (EIO); 3245 } 3246 3247 return (zap_list(spa, &child_dir_zap) != 0); 3248 } 3249 3250 int 3251 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, 3252 int (*callback)(const char *, uint64_t)) 3253 { 3254 uint64_t dir_obj, child_dir_zapobj; 3255 dnode_phys_t child_dir_zap, dir, dataset; 3256 dsl_dataset_phys_t *ds; 3257 dsl_dir_phys_t *dd; 3258 zap_phys_t *zap; 3259 size_t size; 3260 int err; 3261 3262 err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset); 3263 if (err != 0) { 3264 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3265 return (err); 3266 } 3267 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3268 dir_obj = ds->ds_dir_obj; 3269 3270 err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir); 3271 if (err != 0) { 3272 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj); 3273 return (err); 3274 } 3275 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3276 3277 child_dir_zapobj = dd->dd_child_dir_zapobj; 3278 err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3279 &child_dir_zap); 3280 if (err != 0) { 3281 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj); 3282 return (err); 3283 } 3284 3285 size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT; 3286 zap = malloc(size); 3287 if (zap != NULL) { 3288 err = dnode_read(spa, &child_dir_zap, 0, zap, size); 3289 if (err != 0) 3290 goto done; 3291 3292 if (zap->zap_block_type == ZBT_MICRO) 3293 err = mzap_list((const mzap_phys_t *)zap, size, 3294 callback); 3295 else 3296 err = fzap_list(spa, &child_dir_zap, zap, callback); 3297 } else { 3298 err = ENOMEM; 3299 } 3300 done: 3301 free(zap); 3302 return (err); 3303 } 3304 #endif 3305 3306 /* 3307 * Find the object set given the object number of its dataset object 3308 * and return its details in *objset 3309 */ 3310 static int 3311 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset) 3312 { 3313 dnode_phys_t dataset; 3314 dsl_dataset_phys_t *ds; 3315 3316 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3317 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3318 return (EIO); 3319 } 3320 3321 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3322 if (zio_read(spa, &ds->ds_bp, objset)) { 3323 printf("ZFS: can't read object set for dataset %ju\n", 3324 (uintmax_t)objnum); 3325 return (EIO); 3326 } 3327 3328 return (0); 3329 } 3330 3331 /* 3332 * Find the object set pointed to by the BOOTFS property or the root 3333 * dataset if there is none and return its details in *objset 3334 */ 3335 static int 3336 zfs_get_root(const spa_t *spa, uint64_t *objid) 3337 { 3338 dnode_phys_t dir, propdir; 3339 uint64_t props, bootfs, root; 3340 3341 *objid = 0; 3342 3343 /* 3344 * Start with the MOS directory object. 3345 */ 3346 if (objset_get_dnode(spa, spa->spa_mos, 3347 DMU_POOL_DIRECTORY_OBJECT, &dir)) { 3348 printf("ZFS: can't read MOS object directory\n"); 3349 return (EIO); 3350 } 3351 3352 /* 3353 * Lookup the pool_props and see if we can find a bootfs. 3354 */ 3355 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, 3356 sizeof(props), 1, &props) == 0 && 3357 objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 && 3358 zap_lookup(spa, &propdir, "bootfs", 3359 sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) { 3360 *objid = bootfs; 3361 return (0); 3362 } 3363 /* 3364 * Lookup the root dataset directory 3365 */ 3366 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, 3367 sizeof(root), 1, &root) || 3368 objset_get_dnode(spa, spa->spa_mos, root, &dir)) { 3369 printf("ZFS: can't find root dsl_dir\n"); 3370 return (EIO); 3371 } 3372 3373 /* 3374 * Use the information from the dataset directory's bonus buffer 3375 * to find the dataset object and from that the object set itself. 3376 */ 3377 dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3378 *objid = dd->dd_head_dataset_obj; 3379 return (0); 3380 } 3381 3382 static int 3383 zfs_mount_impl(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount) 3384 { 3385 3386 mount->spa = spa; 3387 3388 /* 3389 * Find the root object set if not explicitly provided 3390 */ 3391 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) { 3392 printf("ZFS: can't find root filesystem\n"); 3393 return (EIO); 3394 } 3395 3396 if (zfs_mount_dataset(spa, rootobj, &mount->objset)) { 3397 printf("ZFS: can't open root filesystem\n"); 3398 return (EIO); 3399 } 3400 3401 mount->rootobj = rootobj; 3402 3403 return (0); 3404 } 3405 3406 /* 3407 * callback function for feature name checks. 3408 */ 3409 static int 3410 check_feature(const char *name, uint64_t value) 3411 { 3412 int i; 3413 3414 if (value == 0) 3415 return (0); 3416 if (name[0] == '\0') 3417 return (0); 3418 3419 for (i = 0; features_for_read[i] != NULL; i++) { 3420 if (strcmp(name, features_for_read[i]) == 0) 3421 return (0); 3422 } 3423 printf("ZFS: unsupported feature: %s\n", name); 3424 return (EIO); 3425 } 3426 3427 /* 3428 * Checks whether the MOS features that are active are supported. 3429 */ 3430 static int 3431 check_mos_features(const spa_t *spa) 3432 { 3433 dnode_phys_t dir; 3434 zap_phys_t *zap; 3435 uint64_t objnum; 3436 size_t size; 3437 int rc; 3438 3439 if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY, 3440 &dir)) != 0) 3441 return (rc); 3442 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ, 3443 sizeof (objnum), 1, &objnum)) != 0) { 3444 /* 3445 * It is older pool without features. As we have already 3446 * tested the label, just return without raising the error. 3447 */ 3448 return (0); 3449 } 3450 3451 if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0) 3452 return (rc); 3453 3454 if (dir.dn_type != DMU_OTN_ZAP_METADATA) 3455 return (EIO); 3456 3457 size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT; 3458 zap = malloc(size); 3459 if (zap == NULL) 3460 return (ENOMEM); 3461 3462 if (dnode_read(spa, &dir, 0, zap, size)) { 3463 free(zap); 3464 return (EIO); 3465 } 3466 3467 if (zap->zap_block_type == ZBT_MICRO) 3468 rc = mzap_list((const mzap_phys_t *)zap, size, check_feature); 3469 else 3470 rc = fzap_list(spa, &dir, zap, check_feature); 3471 3472 free(zap); 3473 return (rc); 3474 } 3475 3476 static int 3477 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 3478 { 3479 dnode_phys_t dir; 3480 size_t size; 3481 int rc; 3482 char *nv; 3483 3484 *value = NULL; 3485 if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0) 3486 return (rc); 3487 if (dir.dn_type != DMU_OT_PACKED_NVLIST && 3488 dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) { 3489 return (EIO); 3490 } 3491 3492 if (dir.dn_bonuslen != sizeof (uint64_t)) 3493 return (EIO); 3494 3495 size = *(uint64_t *)DN_BONUS(&dir); 3496 nv = malloc(size); 3497 if (nv == NULL) 3498 return (ENOMEM); 3499 3500 rc = dnode_read(spa, &dir, 0, nv, size); 3501 if (rc != 0) { 3502 free(nv); 3503 nv = NULL; 3504 return (rc); 3505 } 3506 *value = nvlist_import(nv, size); 3507 free(nv); 3508 return (rc); 3509 } 3510 3511 static int 3512 zfs_spa_init(spa_t *spa) 3513 { 3514 struct uberblock checkpoint; 3515 dnode_phys_t dir; 3516 uint64_t config_object; 3517 nvlist_t *nvlist; 3518 int rc; 3519 3520 if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) { 3521 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name); 3522 return (EIO); 3523 } 3524 if (spa->spa_mos->os_type != DMU_OST_META) { 3525 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name); 3526 return (EIO); 3527 } 3528 3529 if (objset_get_dnode(spa, &spa->spa_mos_master, 3530 DMU_POOL_DIRECTORY_OBJECT, &dir)) { 3531 printf("ZFS: failed to read pool %s directory object\n", 3532 spa->spa_name); 3533 return (EIO); 3534 } 3535 /* this is allowed to fail, older pools do not have salt */ 3536 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1, 3537 sizeof (spa->spa_cksum_salt.zcs_bytes), 3538 spa->spa_cksum_salt.zcs_bytes); 3539 3540 rc = check_mos_features(spa); 3541 if (rc != 0) { 3542 printf("ZFS: pool %s is not supported\n", spa->spa_name); 3543 return (rc); 3544 } 3545 3546 rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG, 3547 sizeof (config_object), 1, &config_object); 3548 if (rc != 0) { 3549 printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG); 3550 return (EIO); 3551 } 3552 rc = load_nvlist(spa, config_object, &nvlist); 3553 if (rc != 0) 3554 return (rc); 3555 3556 rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT, 3557 sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t), 3558 &checkpoint); 3559 if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) { 3560 memcpy(&spa->spa_uberblock_checkpoint, &checkpoint, 3561 sizeof(checkpoint)); 3562 if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp, 3563 &spa->spa_mos_checkpoint)) { 3564 printf("ZFS: can not read checkpoint data.\n"); 3565 return (EIO); 3566 } 3567 } 3568 3569 /* 3570 * Update vdevs from MOS config. Note, we do skip encoding bytes 3571 * here. See also vdev_label_read_config(). 3572 */ 3573 rc = vdev_init_from_nvlist(spa, nvlist); 3574 nvlist_destroy(nvlist); 3575 return (rc); 3576 } 3577 3578 static int 3579 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb) 3580 { 3581 3582 if (dn->dn_bonustype != DMU_OT_SA) { 3583 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus; 3584 3585 sb->st_mode = zp->zp_mode; 3586 sb->st_uid = zp->zp_uid; 3587 sb->st_gid = zp->zp_gid; 3588 sb->st_size = zp->zp_size; 3589 } else { 3590 sa_hdr_phys_t *sahdrp; 3591 int hdrsize; 3592 size_t size = 0; 3593 void *buf = NULL; 3594 3595 if (dn->dn_bonuslen != 0) 3596 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn); 3597 else { 3598 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) { 3599 blkptr_t *bp = DN_SPILL_BLKPTR(dn); 3600 int error; 3601 3602 size = BP_GET_LSIZE(bp); 3603 buf = malloc(size); 3604 if (buf == NULL) 3605 error = ENOMEM; 3606 else 3607 error = zio_read(spa, bp, buf); 3608 3609 if (error != 0) { 3610 free(buf); 3611 return (error); 3612 } 3613 sahdrp = buf; 3614 } else { 3615 return (EIO); 3616 } 3617 } 3618 hdrsize = SA_HDR_SIZE(sahdrp); 3619 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize + 3620 SA_MODE_OFFSET); 3621 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize + 3622 SA_UID_OFFSET); 3623 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize + 3624 SA_GID_OFFSET); 3625 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize + 3626 SA_SIZE_OFFSET); 3627 free(buf); 3628 } 3629 3630 return (0); 3631 } 3632 3633 static int 3634 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize) 3635 { 3636 int rc = 0; 3637 3638 if (dn->dn_bonustype == DMU_OT_SA) { 3639 sa_hdr_phys_t *sahdrp = NULL; 3640 size_t size = 0; 3641 void *buf = NULL; 3642 int hdrsize; 3643 char *p; 3644 3645 if (dn->dn_bonuslen != 0) { 3646 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn); 3647 } else { 3648 blkptr_t *bp; 3649 3650 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0) 3651 return (EIO); 3652 bp = DN_SPILL_BLKPTR(dn); 3653 3654 size = BP_GET_LSIZE(bp); 3655 buf = malloc(size); 3656 if (buf == NULL) 3657 rc = ENOMEM; 3658 else 3659 rc = zio_read(spa, bp, buf); 3660 if (rc != 0) { 3661 free(buf); 3662 return (rc); 3663 } 3664 sahdrp = buf; 3665 } 3666 hdrsize = SA_HDR_SIZE(sahdrp); 3667 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET); 3668 memcpy(path, p, psize); 3669 free(buf); 3670 return (0); 3671 } 3672 /* 3673 * Second test is purely to silence bogus compiler 3674 * warning about accessing past the end of dn_bonus. 3675 */ 3676 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen && 3677 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) { 3678 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize); 3679 } else { 3680 rc = dnode_read(spa, dn, 0, path, psize); 3681 } 3682 return (rc); 3683 } 3684 3685 struct obj_list { 3686 uint64_t objnum; 3687 STAILQ_ENTRY(obj_list) entry; 3688 }; 3689 3690 /* 3691 * Lookup a file and return its dnode. 3692 */ 3693 static int 3694 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode) 3695 { 3696 int rc; 3697 uint64_t objnum; 3698 const spa_t *spa; 3699 dnode_phys_t dn; 3700 const char *p, *q; 3701 char element[256]; 3702 char path[1024]; 3703 int symlinks_followed = 0; 3704 struct stat sb; 3705 struct obj_list *entry, *tentry; 3706 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache); 3707 3708 spa = mount->spa; 3709 if (mount->objset.os_type != DMU_OST_ZFS) { 3710 printf("ZFS: unexpected object set type %ju\n", 3711 (uintmax_t)mount->objset.os_type); 3712 return (EIO); 3713 } 3714 3715 if ((entry = malloc(sizeof(struct obj_list))) == NULL) 3716 return (ENOMEM); 3717 3718 /* 3719 * Get the root directory dnode. 3720 */ 3721 rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn); 3722 if (rc) { 3723 free(entry); 3724 return (rc); 3725 } 3726 3727 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum); 3728 if (rc) { 3729 free(entry); 3730 return (rc); 3731 } 3732 entry->objnum = objnum; 3733 STAILQ_INSERT_HEAD(&on_cache, entry, entry); 3734 3735 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3736 if (rc != 0) 3737 goto done; 3738 3739 p = upath; 3740 while (p && *p) { 3741 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3742 if (rc != 0) 3743 goto done; 3744 3745 while (*p == '/') 3746 p++; 3747 if (*p == '\0') 3748 break; 3749 q = p; 3750 while (*q != '\0' && *q != '/') 3751 q++; 3752 3753 /* skip dot */ 3754 if (p + 1 == q && p[0] == '.') { 3755 p++; 3756 continue; 3757 } 3758 /* double dot */ 3759 if (p + 2 == q && p[0] == '.' && p[1] == '.') { 3760 p += 2; 3761 if (STAILQ_FIRST(&on_cache) == 3762 STAILQ_LAST(&on_cache, obj_list, entry)) { 3763 rc = ENOENT; 3764 goto done; 3765 } 3766 entry = STAILQ_FIRST(&on_cache); 3767 STAILQ_REMOVE_HEAD(&on_cache, entry); 3768 free(entry); 3769 objnum = (STAILQ_FIRST(&on_cache))->objnum; 3770 continue; 3771 } 3772 if (q - p + 1 > sizeof(element)) { 3773 rc = ENAMETOOLONG; 3774 goto done; 3775 } 3776 memcpy(element, p, q - p); 3777 element[q - p] = 0; 3778 p = q; 3779 3780 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0) 3781 goto done; 3782 if (!S_ISDIR(sb.st_mode)) { 3783 rc = ENOTDIR; 3784 goto done; 3785 } 3786 3787 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum); 3788 if (rc) 3789 goto done; 3790 objnum = ZFS_DIRENT_OBJ(objnum); 3791 3792 if ((entry = malloc(sizeof(struct obj_list))) == NULL) { 3793 rc = ENOMEM; 3794 goto done; 3795 } 3796 entry->objnum = objnum; 3797 STAILQ_INSERT_HEAD(&on_cache, entry, entry); 3798 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3799 if (rc) 3800 goto done; 3801 3802 /* 3803 * Check for symlink. 3804 */ 3805 rc = zfs_dnode_stat(spa, &dn, &sb); 3806 if (rc) 3807 goto done; 3808 if (S_ISLNK(sb.st_mode)) { 3809 if (symlinks_followed > 10) { 3810 rc = EMLINK; 3811 goto done; 3812 } 3813 symlinks_followed++; 3814 3815 /* 3816 * Read the link value and copy the tail of our 3817 * current path onto the end. 3818 */ 3819 if (sb.st_size + strlen(p) + 1 > sizeof(path)) { 3820 rc = ENAMETOOLONG; 3821 goto done; 3822 } 3823 strcpy(&path[sb.st_size], p); 3824 3825 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size); 3826 if (rc != 0) 3827 goto done; 3828 3829 /* 3830 * Restart with the new path, starting either at 3831 * the root or at the parent depending whether or 3832 * not the link is relative. 3833 */ 3834 p = path; 3835 if (*p == '/') { 3836 while (STAILQ_FIRST(&on_cache) != 3837 STAILQ_LAST(&on_cache, obj_list, entry)) { 3838 entry = STAILQ_FIRST(&on_cache); 3839 STAILQ_REMOVE_HEAD(&on_cache, entry); 3840 free(entry); 3841 } 3842 } else { 3843 entry = STAILQ_FIRST(&on_cache); 3844 STAILQ_REMOVE_HEAD(&on_cache, entry); 3845 free(entry); 3846 } 3847 objnum = (STAILQ_FIRST(&on_cache))->objnum; 3848 } 3849 } 3850 3851 *dnode = dn; 3852 done: 3853 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry) 3854 free(entry); 3855 return (rc); 3856 } 3857 3858 /* 3859 * Return either a cached copy of the bootenv, or read each of the vdev children 3860 * looking for the bootenv. Cache what's found and return the results. Returns 0 3861 * when benvp is filled in, and some errno when not. 3862 */ 3863 static int 3864 zfs_get_bootenv_spa(spa_t *spa, nvlist_t **benvp) 3865 { 3866 vdev_t *vd; 3867 nvlist_t *benv = NULL; 3868 3869 if (spa->spa_bootenv == NULL) { 3870 STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children, 3871 v_childlink) { 3872 benv = vdev_read_bootenv(vd); 3873 3874 if (benv != NULL) 3875 break; 3876 } 3877 spa->spa_bootenv = benv; 3878 } 3879 benv = spa->spa_bootenv; 3880 3881 if (benv == NULL) 3882 return (ENOENT); 3883 3884 *benvp = benv; 3885 return (0); 3886 } 3887 3888 /* 3889 * Store nvlist to pool label bootenv area. Also updates cached pointer in spa. 3890 */ 3891 static int 3892 zfs_set_bootenv_spa(spa_t *spa, nvlist_t *benv) 3893 { 3894 vdev_t *vd; 3895 3896 STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children, v_childlink) { 3897 vdev_write_bootenv(vd, benv); 3898 } 3899 3900 spa->spa_bootenv = benv; 3901 return (0); 3902 } 3903 3904 /* 3905 * Get bootonce value by key. The bootonce <key, value> pair is removed from the 3906 * bootenv nvlist and the remaining nvlist is committed back to disk. This process 3907 * the bootonce flag since we've reached the point in the boot that we've 'used' 3908 * the BE. For chained boot scenarios, we may reach this point multiple times (but 3909 * only remove it and return 0 the first time). 3910 */ 3911 static int 3912 zfs_get_bootonce_spa(spa_t *spa, const char *key, char *buf, size_t size) 3913 { 3914 nvlist_t *benv; 3915 char *result = NULL; 3916 int result_size, rv; 3917 3918 if ((rv = zfs_get_bootenv_spa(spa, &benv)) != 0) 3919 return (rv); 3920 3921 if ((rv = nvlist_find(benv, key, DATA_TYPE_STRING, NULL, 3922 &result, &result_size)) == 0) { 3923 if (result_size == 0) { 3924 /* ignore empty string */ 3925 rv = ENOENT; 3926 } else if (buf != NULL) { 3927 size = MIN((size_t)result_size + 1, size); 3928 strlcpy(buf, result, size); 3929 } 3930 (void)nvlist_remove(benv, key, DATA_TYPE_STRING); 3931 (void)zfs_set_bootenv_spa(spa, benv); 3932 } 3933 3934 return (rv); 3935 } 3936