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 /* 28 * Stand-alone ZFS file reader. 29 */ 30 31 #include <stdbool.h> 32 #include <sys/endian.h> 33 #include <sys/stat.h> 34 #include <sys/stdint.h> 35 #include <sys/list.h> 36 #include <sys/zfs_bootenv.h> 37 #include <machine/_inttypes.h> 38 39 #include "zfsimpl.h" 40 #include "zfssubr.c" 41 42 #ifdef HAS_ZSTD_ZFS 43 extern int zstd_init(void); 44 #endif 45 46 struct zfsmount { 47 char *path; 48 const spa_t *spa; 49 objset_phys_t objset; 50 uint64_t rootobj; 51 STAILQ_ENTRY(zfsmount) next; 52 }; 53 54 typedef STAILQ_HEAD(zfs_mnt_list, zfsmount) zfs_mnt_list_t; 55 static zfs_mnt_list_t zfsmount = STAILQ_HEAD_INITIALIZER(zfsmount); 56 57 /* 58 * The indirect_child_t represents the vdev that we will read from, when we 59 * need to read all copies of the data (e.g. for scrub or reconstruction). 60 * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror), 61 * ic_vdev is the same as is_vdev. However, for mirror top-level vdevs, 62 * ic_vdev is a child of the mirror. 63 */ 64 typedef struct indirect_child { 65 void *ic_data; 66 vdev_t *ic_vdev; 67 } indirect_child_t; 68 69 /* 70 * The indirect_split_t represents one mapped segment of an i/o to the 71 * indirect vdev. For non-split (contiguously-mapped) blocks, there will be 72 * only one indirect_split_t, with is_split_offset==0 and is_size==io_size. 73 * For split blocks, there will be several of these. 74 */ 75 typedef struct indirect_split { 76 list_node_t is_node; /* link on iv_splits */ 77 78 /* 79 * is_split_offset is the offset into the i/o. 80 * This is the sum of the previous splits' is_size's. 81 */ 82 uint64_t is_split_offset; 83 84 vdev_t *is_vdev; /* top-level vdev */ 85 uint64_t is_target_offset; /* offset on is_vdev */ 86 uint64_t is_size; 87 int is_children; /* number of entries in is_child[] */ 88 89 /* 90 * is_good_child is the child that we are currently using to 91 * attempt reconstruction. 92 */ 93 int is_good_child; 94 95 indirect_child_t is_child[1]; /* variable-length */ 96 } indirect_split_t; 97 98 /* 99 * The indirect_vsd_t is associated with each i/o to the indirect vdev. 100 * It is the "Vdev-Specific Data" in the zio_t's io_vsd. 101 */ 102 typedef struct indirect_vsd { 103 boolean_t iv_split_block; 104 boolean_t iv_reconstruct; 105 106 list_t iv_splits; /* list of indirect_split_t's */ 107 } indirect_vsd_t; 108 109 /* 110 * List of all vdevs, chained through v_alllink. 111 */ 112 static vdev_list_t zfs_vdevs; 113 114 /* 115 * List of supported read-incompatible ZFS features. Do not add here features 116 * marked as ZFEATURE_FLAG_READONLY_COMPAT, they are irrelevant for read-only! 117 */ 118 static const char *features_for_read[] = { 119 "com.datto:bookmark_v2", 120 "com.datto:encryption", 121 "com.delphix:bookmark_written", 122 "com.delphix:device_removal", 123 "com.delphix:embedded_data", 124 "com.delphix:extensible_dataset", 125 "com.delphix:head_errlog", 126 "com.delphix:hole_birth", 127 "com.joyent:multi_vdev_crash_dump", 128 "com.klarasystems:vdev_zaps_v2", 129 "org.freebsd:zstd_compress", 130 "org.illumos:lz4_compress", 131 "org.illumos:sha512", 132 "org.illumos:skein", 133 "org.open-zfs:large_blocks", 134 "org.openzfs:blake3", 135 "org.zfsonlinux:large_dnode", 136 NULL 137 }; 138 139 /* 140 * List of all pools, chained through spa_link. 141 */ 142 static spa_list_t zfs_pools; 143 144 static const dnode_phys_t *dnode_cache_obj; 145 static uint64_t dnode_cache_bn; 146 static char *dnode_cache_buf; 147 148 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf); 149 static int zfs_get_root(const spa_t *spa, uint64_t *objid); 150 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result); 151 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, 152 const char *name, uint64_t integer_size, uint64_t num_integers, 153 void *value); 154 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t, 155 dnode_phys_t *); 156 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *, 157 size_t); 158 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t, 159 size_t); 160 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t); 161 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *, 162 uint64_t); 163 vdev_indirect_mapping_entry_phys_t * 164 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t, 165 uint64_t, uint64_t *); 166 167 static void 168 zfs_init(void) 169 { 170 STAILQ_INIT(&zfs_vdevs); 171 STAILQ_INIT(&zfs_pools); 172 173 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE); 174 175 zfs_init_crc(); 176 #ifdef HAS_ZSTD_ZFS 177 zstd_init(); 178 #endif 179 } 180 181 static int 182 nvlist_check_features_for_read(nvlist_t *nvl) 183 { 184 nvlist_t *features = NULL; 185 nvs_data_t *data; 186 nvp_header_t *nvp; 187 nv_string_t *nvp_name; 188 int rc; 189 190 rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ, 191 DATA_TYPE_NVLIST, NULL, &features, NULL); 192 switch (rc) { 193 case 0: 194 break; /* Continue with checks */ 195 196 case ENOENT: 197 return (0); /* All features are disabled */ 198 199 default: 200 return (rc); /* Error while reading nvlist */ 201 } 202 203 data = (nvs_data_t *)features->nv_data; 204 nvp = &data->nvl_pair; /* first pair in nvlist */ 205 206 while (nvp->encoded_size != 0 && nvp->decoded_size != 0) { 207 int i, found; 208 209 nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp)); 210 found = 0; 211 212 for (i = 0; features_for_read[i] != NULL; i++) { 213 if (memcmp(nvp_name->nv_data, features_for_read[i], 214 nvp_name->nv_size) == 0) { 215 found = 1; 216 break; 217 } 218 } 219 220 if (!found) { 221 printf("ZFS: unsupported feature: %.*s\n", 222 nvp_name->nv_size, nvp_name->nv_data); 223 rc = EIO; 224 } 225 nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size); 226 } 227 nvlist_destroy(features); 228 229 return (rc); 230 } 231 232 static int 233 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf, 234 off_t offset, size_t size) 235 { 236 size_t psize; 237 int rc; 238 239 if (vdev->v_phys_read == NULL) 240 return (ENOTSUP); 241 242 if (bp) { 243 psize = BP_GET_PSIZE(bp); 244 } else { 245 psize = size; 246 } 247 248 rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize); 249 if (rc == 0) { 250 if (bp != NULL) 251 rc = zio_checksum_verify(vdev->v_spa, bp, buf); 252 } 253 254 return (rc); 255 } 256 257 static int 258 vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size) 259 { 260 if (vdev->v_phys_write == NULL) 261 return (ENOTSUP); 262 263 return (vdev->v_phys_write(vdev, offset, buf, size)); 264 } 265 266 typedef struct remap_segment { 267 vdev_t *rs_vd; 268 uint64_t rs_offset; 269 uint64_t rs_asize; 270 uint64_t rs_split_offset; 271 list_node_t rs_node; 272 } remap_segment_t; 273 274 static remap_segment_t * 275 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset) 276 { 277 remap_segment_t *rs = malloc(sizeof (remap_segment_t)); 278 279 if (rs != NULL) { 280 rs->rs_vd = vd; 281 rs->rs_offset = offset; 282 rs->rs_asize = asize; 283 rs->rs_split_offset = split_offset; 284 } 285 286 return (rs); 287 } 288 289 vdev_indirect_mapping_t * 290 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os, 291 uint64_t mapping_object) 292 { 293 vdev_indirect_mapping_t *vim; 294 vdev_indirect_mapping_phys_t *vim_phys; 295 int rc; 296 297 vim = calloc(1, sizeof (*vim)); 298 if (vim == NULL) 299 return (NULL); 300 301 vim->vim_dn = calloc(1, sizeof (*vim->vim_dn)); 302 if (vim->vim_dn == NULL) { 303 free(vim); 304 return (NULL); 305 } 306 307 rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn); 308 if (rc != 0) { 309 free(vim->vim_dn); 310 free(vim); 311 return (NULL); 312 } 313 314 vim->vim_spa = spa; 315 vim->vim_phys = malloc(sizeof (*vim->vim_phys)); 316 if (vim->vim_phys == NULL) { 317 free(vim->vim_dn); 318 free(vim); 319 return (NULL); 320 } 321 322 vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn); 323 *vim->vim_phys = *vim_phys; 324 325 vim->vim_objset = os; 326 vim->vim_object = mapping_object; 327 vim->vim_entries = NULL; 328 329 vim->vim_havecounts = 330 (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0); 331 332 return (vim); 333 } 334 335 /* 336 * Compare an offset with an indirect mapping entry; there are three 337 * possible scenarios: 338 * 339 * 1. The offset is "less than" the mapping entry; meaning the 340 * offset is less than the source offset of the mapping entry. In 341 * this case, there is no overlap between the offset and the 342 * mapping entry and -1 will be returned. 343 * 344 * 2. The offset is "greater than" the mapping entry; meaning the 345 * offset is greater than the mapping entry's source offset plus 346 * the entry's size. In this case, there is no overlap between 347 * the offset and the mapping entry and 1 will be returned. 348 * 349 * NOTE: If the offset is actually equal to the entry's offset 350 * plus size, this is considered to be "greater" than the entry, 351 * and this case applies (i.e. 1 will be returned). Thus, the 352 * entry's "range" can be considered to be inclusive at its 353 * start, but exclusive at its end: e.g. [src, src + size). 354 * 355 * 3. The last case to consider is if the offset actually falls 356 * within the mapping entry's range. If this is the case, the 357 * offset is considered to be "equal to" the mapping entry and 358 * 0 will be returned. 359 * 360 * NOTE: If the offset is equal to the entry's source offset, 361 * this case applies and 0 will be returned. If the offset is 362 * equal to the entry's source plus its size, this case does 363 * *not* apply (see "NOTE" above for scenario 2), and 1 will be 364 * returned. 365 */ 366 static int 367 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem) 368 { 369 const uint64_t *key = v_key; 370 const vdev_indirect_mapping_entry_phys_t *array_elem = 371 v_array_elem; 372 uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem); 373 374 if (*key < src_offset) { 375 return (-1); 376 } else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) { 377 return (0); 378 } else { 379 return (1); 380 } 381 } 382 383 /* 384 * Return array entry. 385 */ 386 static vdev_indirect_mapping_entry_phys_t * 387 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index) 388 { 389 uint64_t size; 390 off_t offset = 0; 391 int rc; 392 393 if (vim->vim_phys->vimp_num_entries == 0) 394 return (NULL); 395 396 if (vim->vim_entries == NULL) { 397 uint64_t bsize; 398 399 bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 400 size = vim->vim_phys->vimp_num_entries * 401 sizeof (*vim->vim_entries); 402 if (size > bsize) { 403 size = bsize / sizeof (*vim->vim_entries); 404 size *= sizeof (*vim->vim_entries); 405 } 406 vim->vim_entries = malloc(size); 407 if (vim->vim_entries == NULL) 408 return (NULL); 409 vim->vim_num_entries = size / sizeof (*vim->vim_entries); 410 offset = index * sizeof (*vim->vim_entries); 411 } 412 413 /* We have data in vim_entries */ 414 if (offset == 0) { 415 if (index >= vim->vim_entry_offset && 416 index <= vim->vim_entry_offset + vim->vim_num_entries) { 417 index -= vim->vim_entry_offset; 418 return (&vim->vim_entries[index]); 419 } 420 offset = index * sizeof (*vim->vim_entries); 421 } 422 423 vim->vim_entry_offset = index; 424 size = vim->vim_num_entries * sizeof (*vim->vim_entries); 425 rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries, 426 size); 427 if (rc != 0) { 428 /* Read error, invalidate vim_entries. */ 429 free(vim->vim_entries); 430 vim->vim_entries = NULL; 431 return (NULL); 432 } 433 index -= vim->vim_entry_offset; 434 return (&vim->vim_entries[index]); 435 } 436 437 /* 438 * Returns the mapping entry for the given offset. 439 * 440 * It's possible that the given offset will not be in the mapping table 441 * (i.e. no mapping entries contain this offset), in which case, the 442 * return value depends on the "next_if_missing" parameter. 443 * 444 * If the offset is not found in the table and "next_if_missing" is 445 * B_FALSE, then NULL will always be returned. The behavior is intended 446 * to allow consumers to get the entry corresponding to the offset 447 * parameter, iff the offset overlaps with an entry in the table. 448 * 449 * If the offset is not found in the table and "next_if_missing" is 450 * B_TRUE, then the entry nearest to the given offset will be returned, 451 * such that the entry's source offset is greater than the offset 452 * passed in (i.e. the "next" mapping entry in the table is returned, if 453 * the offset is missing from the table). If there are no entries whose 454 * source offset is greater than the passed in offset, NULL is returned. 455 */ 456 static vdev_indirect_mapping_entry_phys_t * 457 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim, 458 uint64_t offset) 459 { 460 ASSERT(vim->vim_phys->vimp_num_entries > 0); 461 462 vdev_indirect_mapping_entry_phys_t *entry; 463 464 uint64_t last = vim->vim_phys->vimp_num_entries - 1; 465 uint64_t base = 0; 466 467 /* 468 * We don't define these inside of the while loop because we use 469 * their value in the case that offset isn't in the mapping. 470 */ 471 uint64_t mid; 472 int result; 473 474 while (last >= base) { 475 mid = base + ((last - base) >> 1); 476 477 entry = vdev_indirect_mapping_entry(vim, mid); 478 if (entry == NULL) 479 break; 480 result = dva_mapping_overlap_compare(&offset, entry); 481 482 if (result == 0) { 483 break; 484 } else if (result < 0) { 485 last = mid - 1; 486 } else { 487 base = mid + 1; 488 } 489 } 490 return (entry); 491 } 492 493 /* 494 * Given an indirect vdev and an extent on that vdev, it duplicates the 495 * physical entries of the indirect mapping that correspond to the extent 496 * to a new array and returns a pointer to it. In addition, copied_entries 497 * is populated with the number of mapping entries that were duplicated. 498 * 499 * Finally, since we are doing an allocation, it is up to the caller to 500 * free the array allocated in this function. 501 */ 502 vdev_indirect_mapping_entry_phys_t * 503 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset, 504 uint64_t asize, uint64_t *copied_entries) 505 { 506 vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL; 507 vdev_indirect_mapping_t *vim = vd->v_mapping; 508 uint64_t entries = 0; 509 510 vdev_indirect_mapping_entry_phys_t *first_mapping = 511 vdev_indirect_mapping_entry_for_offset(vim, offset); 512 ASSERT3P(first_mapping, !=, NULL); 513 514 vdev_indirect_mapping_entry_phys_t *m = first_mapping; 515 while (asize > 0) { 516 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst); 517 uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m); 518 uint64_t inner_size = MIN(asize, size - inner_offset); 519 520 offset += inner_size; 521 asize -= inner_size; 522 entries++; 523 m++; 524 } 525 526 size_t copy_length = entries * sizeof (*first_mapping); 527 duplicate_mappings = malloc(copy_length); 528 if (duplicate_mappings != NULL) 529 bcopy(first_mapping, duplicate_mappings, copy_length); 530 else 531 entries = 0; 532 533 *copied_entries = entries; 534 535 return (duplicate_mappings); 536 } 537 538 static vdev_t * 539 vdev_lookup_top(spa_t *spa, uint64_t vdev) 540 { 541 vdev_t *rvd; 542 vdev_list_t *vlist; 543 544 vlist = &spa->spa_root_vdev->v_children; 545 STAILQ_FOREACH(rvd, vlist, v_childlink) 546 if (rvd->v_id == vdev) 547 break; 548 549 return (rvd); 550 } 551 552 /* 553 * This is a callback for vdev_indirect_remap() which allocates an 554 * indirect_split_t for each split segment and adds it to iv_splits. 555 */ 556 static void 557 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset, 558 uint64_t size, void *arg) 559 { 560 int n = 1; 561 zio_t *zio = arg; 562 indirect_vsd_t *iv = zio->io_vsd; 563 564 if (vd->v_read == vdev_indirect_read) 565 return; 566 567 if (vd->v_read == vdev_mirror_read) 568 n = vd->v_nchildren; 569 570 indirect_split_t *is = 571 malloc(offsetof(indirect_split_t, is_child[n])); 572 if (is == NULL) { 573 zio->io_error = ENOMEM; 574 return; 575 } 576 bzero(is, offsetof(indirect_split_t, is_child[n])); 577 578 is->is_children = n; 579 is->is_size = size; 580 is->is_split_offset = split_offset; 581 is->is_target_offset = offset; 582 is->is_vdev = vd; 583 584 /* 585 * Note that we only consider multiple copies of the data for 586 * *mirror* vdevs. We don't for "replacing" or "spare" vdevs, even 587 * though they use the same ops as mirror, because there's only one 588 * "good" copy under the replacing/spare. 589 */ 590 if (vd->v_read == vdev_mirror_read) { 591 int i = 0; 592 vdev_t *kid; 593 594 STAILQ_FOREACH(kid, &vd->v_children, v_childlink) { 595 is->is_child[i++].ic_vdev = kid; 596 } 597 } else { 598 is->is_child[0].ic_vdev = vd; 599 } 600 601 list_insert_tail(&iv->iv_splits, is); 602 } 603 604 static void 605 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg) 606 { 607 list_t stack; 608 spa_t *spa = vd->v_spa; 609 zio_t *zio = arg; 610 remap_segment_t *rs; 611 612 list_create(&stack, sizeof (remap_segment_t), 613 offsetof(remap_segment_t, rs_node)); 614 615 rs = rs_alloc(vd, offset, asize, 0); 616 if (rs == NULL) { 617 printf("vdev_indirect_remap: out of memory.\n"); 618 zio->io_error = ENOMEM; 619 } 620 for (; rs != NULL; rs = list_remove_head(&stack)) { 621 vdev_t *v = rs->rs_vd; 622 uint64_t num_entries = 0; 623 /* vdev_indirect_mapping_t *vim = v->v_mapping; */ 624 vdev_indirect_mapping_entry_phys_t *mapping = 625 vdev_indirect_mapping_duplicate_adjacent_entries(v, 626 rs->rs_offset, rs->rs_asize, &num_entries); 627 628 if (num_entries == 0) 629 zio->io_error = ENOMEM; 630 631 for (uint64_t i = 0; i < num_entries; i++) { 632 vdev_indirect_mapping_entry_phys_t *m = &mapping[i]; 633 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst); 634 uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst); 635 uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst); 636 uint64_t inner_offset = rs->rs_offset - 637 DVA_MAPPING_GET_SRC_OFFSET(m); 638 uint64_t inner_size = 639 MIN(rs->rs_asize, size - inner_offset); 640 vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev); 641 642 if (dst_v->v_read == vdev_indirect_read) { 643 remap_segment_t *o; 644 645 o = rs_alloc(dst_v, dst_offset + inner_offset, 646 inner_size, rs->rs_split_offset); 647 if (o == NULL) { 648 printf("vdev_indirect_remap: " 649 "out of memory.\n"); 650 zio->io_error = ENOMEM; 651 break; 652 } 653 654 list_insert_head(&stack, o); 655 } 656 vdev_indirect_gather_splits(rs->rs_split_offset, dst_v, 657 dst_offset + inner_offset, 658 inner_size, arg); 659 660 /* 661 * vdev_indirect_gather_splits can have memory 662 * allocation error, we can not recover from it. 663 */ 664 if (zio->io_error != 0) 665 break; 666 rs->rs_offset += inner_size; 667 rs->rs_asize -= inner_size; 668 rs->rs_split_offset += inner_size; 669 } 670 671 free(mapping); 672 free(rs); 673 if (zio->io_error != 0) 674 break; 675 } 676 677 list_destroy(&stack); 678 } 679 680 static void 681 vdev_indirect_map_free(zio_t *zio) 682 { 683 indirect_vsd_t *iv = zio->io_vsd; 684 indirect_split_t *is; 685 686 while ((is = list_head(&iv->iv_splits)) != NULL) { 687 for (int c = 0; c < is->is_children; c++) { 688 indirect_child_t *ic = &is->is_child[c]; 689 free(ic->ic_data); 690 } 691 list_remove(&iv->iv_splits, is); 692 free(is); 693 } 694 free(iv); 695 } 696 697 static int 698 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 699 off_t offset, size_t bytes) 700 { 701 zio_t zio; 702 spa_t *spa = vdev->v_spa; 703 indirect_vsd_t *iv; 704 indirect_split_t *first; 705 int rc = EIO; 706 707 iv = calloc(1, sizeof(*iv)); 708 if (iv == NULL) 709 return (ENOMEM); 710 711 list_create(&iv->iv_splits, 712 sizeof (indirect_split_t), offsetof(indirect_split_t, is_node)); 713 714 bzero(&zio, sizeof(zio)); 715 zio.io_spa = spa; 716 zio.io_bp = (blkptr_t *)bp; 717 zio.io_data = buf; 718 zio.io_size = bytes; 719 zio.io_offset = offset; 720 zio.io_vd = vdev; 721 zio.io_vsd = iv; 722 723 if (vdev->v_mapping == NULL) { 724 vdev_indirect_config_t *vic; 725 726 vic = &vdev->vdev_indirect_config; 727 vdev->v_mapping = vdev_indirect_mapping_open(spa, 728 spa->spa_mos, vic->vic_mapping_object); 729 } 730 731 vdev_indirect_remap(vdev, offset, bytes, &zio); 732 if (zio.io_error != 0) 733 return (zio.io_error); 734 735 first = list_head(&iv->iv_splits); 736 if (first->is_size == zio.io_size) { 737 /* 738 * This is not a split block; we are pointing to the entire 739 * data, which will checksum the same as the original data. 740 * Pass the BP down so that the child i/o can verify the 741 * checksum, and try a different location if available 742 * (e.g. on a mirror). 743 * 744 * While this special case could be handled the same as the 745 * general (split block) case, doing it this way ensures 746 * that the vast majority of blocks on indirect vdevs 747 * (which are not split) are handled identically to blocks 748 * on non-indirect vdevs. This allows us to be less strict 749 * about performance in the general (but rare) case. 750 */ 751 rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp, 752 zio.io_data, first->is_target_offset, bytes); 753 } else { 754 iv->iv_split_block = B_TRUE; 755 /* 756 * Read one copy of each split segment, from the 757 * top-level vdev. Since we don't know the 758 * checksum of each split individually, the child 759 * zio can't ensure that we get the right data. 760 * E.g. if it's a mirror, it will just read from a 761 * random (healthy) leaf vdev. We have to verify 762 * the checksum in vdev_indirect_io_done(). 763 */ 764 for (indirect_split_t *is = list_head(&iv->iv_splits); 765 is != NULL; is = list_next(&iv->iv_splits, is)) { 766 char *ptr = zio.io_data; 767 768 rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp, 769 ptr + is->is_split_offset, is->is_target_offset, 770 is->is_size); 771 } 772 if (zio_checksum_verify(spa, zio.io_bp, zio.io_data)) 773 rc = ECKSUM; 774 else 775 rc = 0; 776 } 777 778 vdev_indirect_map_free(&zio); 779 if (rc == 0) 780 rc = zio.io_error; 781 782 return (rc); 783 } 784 785 static int 786 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 787 off_t offset, size_t bytes) 788 { 789 790 return (vdev_read_phys(vdev, bp, buf, 791 offset + VDEV_LABEL_START_SIZE, bytes)); 792 } 793 794 static int 795 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused, 796 void *buf __unused, off_t offset __unused, size_t bytes __unused) 797 { 798 799 return (ENOTSUP); 800 } 801 802 static int 803 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 804 off_t offset, size_t bytes) 805 { 806 vdev_t *kid; 807 int rc; 808 809 rc = EIO; 810 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 811 if (kid->v_state != VDEV_STATE_HEALTHY) 812 continue; 813 rc = kid->v_read(kid, bp, buf, offset, bytes); 814 if (!rc) 815 return (0); 816 } 817 818 return (rc); 819 } 820 821 static int 822 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf, 823 off_t offset, size_t bytes) 824 { 825 vdev_t *kid; 826 827 /* 828 * Here we should have two kids: 829 * First one which is the one we are replacing and we can trust 830 * only this one to have valid data, but it might not be present. 831 * Second one is that one we are replacing with. It is most likely 832 * healthy, but we can't trust it has needed data, so we won't use it. 833 */ 834 kid = STAILQ_FIRST(&vdev->v_children); 835 if (kid == NULL) 836 return (EIO); 837 if (kid->v_state != VDEV_STATE_HEALTHY) 838 return (EIO); 839 return (kid->v_read(kid, bp, buf, offset, bytes)); 840 } 841 842 static vdev_t * 843 vdev_find(uint64_t guid) 844 { 845 vdev_t *vdev; 846 847 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink) 848 if (vdev->v_guid == guid) 849 return (vdev); 850 851 return (0); 852 } 853 854 static vdev_t * 855 vdev_create(uint64_t guid, vdev_read_t *_read) 856 { 857 vdev_t *vdev; 858 vdev_indirect_config_t *vic; 859 860 vdev = calloc(1, sizeof(vdev_t)); 861 if (vdev != NULL) { 862 STAILQ_INIT(&vdev->v_children); 863 vdev->v_guid = guid; 864 vdev->v_read = _read; 865 866 /* 867 * root vdev has no read function, we use this fact to 868 * skip setting up data we do not need for root vdev. 869 * We only point root vdev from spa. 870 */ 871 if (_read != NULL) { 872 vic = &vdev->vdev_indirect_config; 873 vic->vic_prev_indirect_vdev = UINT64_MAX; 874 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink); 875 } 876 } 877 878 return (vdev); 879 } 880 881 static void 882 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist) 883 { 884 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present; 885 uint64_t is_log; 886 887 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0; 888 is_log = 0; 889 (void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL, 890 &is_offline, NULL); 891 (void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL, 892 &is_removed, NULL); 893 (void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL, 894 &is_faulted, NULL); 895 (void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 896 NULL, &is_degraded, NULL); 897 (void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 898 NULL, &isnt_present, NULL); 899 (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL, 900 &is_log, NULL); 901 902 if (is_offline != 0) 903 vdev->v_state = VDEV_STATE_OFFLINE; 904 else if (is_removed != 0) 905 vdev->v_state = VDEV_STATE_REMOVED; 906 else if (is_faulted != 0) 907 vdev->v_state = VDEV_STATE_FAULTED; 908 else if (is_degraded != 0) 909 vdev->v_state = VDEV_STATE_DEGRADED; 910 else if (isnt_present != 0) 911 vdev->v_state = VDEV_STATE_CANT_OPEN; 912 913 vdev->v_islog = is_log != 0; 914 } 915 916 static int 917 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp) 918 { 919 uint64_t id, ashift, asize, nparity; 920 const char *path; 921 const char *type; 922 int len, pathlen; 923 char *name; 924 vdev_t *vdev; 925 926 if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id, 927 NULL) || 928 nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL, 929 &type, &len)) { 930 return (ENOENT); 931 } 932 933 if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 && 934 memcmp(type, VDEV_TYPE_DISK, len) != 0 && 935 #ifdef ZFS_TEST 936 memcmp(type, VDEV_TYPE_FILE, len) != 0 && 937 #endif 938 memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 && 939 memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 && 940 memcmp(type, VDEV_TYPE_REPLACING, len) != 0 && 941 memcmp(type, VDEV_TYPE_HOLE, len) != 0) { 942 printf("ZFS: can only boot from disk, mirror, raidz1, " 943 "raidz2 and raidz3 vdevs, got: %.*s\n", len, type); 944 return (EIO); 945 } 946 947 if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0) 948 vdev = vdev_create(guid, vdev_mirror_read); 949 else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) 950 vdev = vdev_create(guid, vdev_raidz_read); 951 else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0) 952 vdev = vdev_create(guid, vdev_replacing_read); 953 else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) { 954 vdev_indirect_config_t *vic; 955 956 vdev = vdev_create(guid, vdev_indirect_read); 957 if (vdev != NULL) { 958 vdev->v_state = VDEV_STATE_HEALTHY; 959 vic = &vdev->vdev_indirect_config; 960 961 nvlist_find(nvlist, 962 ZPOOL_CONFIG_INDIRECT_OBJECT, 963 DATA_TYPE_UINT64, 964 NULL, &vic->vic_mapping_object, NULL); 965 nvlist_find(nvlist, 966 ZPOOL_CONFIG_INDIRECT_BIRTHS, 967 DATA_TYPE_UINT64, 968 NULL, &vic->vic_births_object, NULL); 969 nvlist_find(nvlist, 970 ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 971 DATA_TYPE_UINT64, 972 NULL, &vic->vic_prev_indirect_vdev, NULL); 973 } 974 } else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) { 975 vdev = vdev_create(guid, vdev_missing_read); 976 } else { 977 vdev = vdev_create(guid, vdev_disk_read); 978 } 979 980 if (vdev == NULL) 981 return (ENOMEM); 982 983 vdev_set_initial_state(vdev, nvlist); 984 vdev->v_id = id; 985 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT, 986 DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0) 987 vdev->v_ashift = ashift; 988 989 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE, 990 DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) { 991 vdev->v_psize = asize + 992 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 993 } 994 995 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY, 996 DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0) 997 vdev->v_nparity = nparity; 998 999 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH, 1000 DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) { 1001 char prefix[] = "/dev/"; 1002 1003 len = strlen(prefix); 1004 if (len < pathlen && memcmp(path, prefix, len) == 0) { 1005 path += len; 1006 pathlen -= len; 1007 } 1008 name = malloc(pathlen + 1); 1009 bcopy(path, name, pathlen); 1010 name[pathlen] = '\0'; 1011 vdev->v_name = name; 1012 } else { 1013 name = NULL; 1014 if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) { 1015 if (vdev->v_nparity < 1 || 1016 vdev->v_nparity > 3) { 1017 printf("ZFS: invalid raidz parity: %d\n", 1018 vdev->v_nparity); 1019 return (EIO); 1020 } 1021 (void) asprintf(&name, "%.*s%d-%" PRIu64, len, type, 1022 vdev->v_nparity, id); 1023 } else { 1024 (void) asprintf(&name, "%.*s-%" PRIu64, len, type, id); 1025 } 1026 vdev->v_name = name; 1027 } 1028 *vdevp = vdev; 1029 return (0); 1030 } 1031 1032 /* 1033 * Find slot for vdev. We return either NULL to signal to use 1034 * STAILQ_INSERT_HEAD, or we return link element to be used with 1035 * STAILQ_INSERT_AFTER. 1036 */ 1037 static vdev_t * 1038 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev) 1039 { 1040 vdev_t *v, *previous; 1041 1042 if (STAILQ_EMPTY(&top_vdev->v_children)) 1043 return (NULL); 1044 1045 previous = NULL; 1046 STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) { 1047 if (v->v_id > vdev->v_id) 1048 return (previous); 1049 1050 if (v->v_id == vdev->v_id) 1051 return (v); 1052 1053 if (v->v_id < vdev->v_id) 1054 previous = v; 1055 } 1056 return (previous); 1057 } 1058 1059 static size_t 1060 vdev_child_count(vdev_t *vdev) 1061 { 1062 vdev_t *v; 1063 size_t count; 1064 1065 count = 0; 1066 STAILQ_FOREACH(v, &vdev->v_children, v_childlink) { 1067 count++; 1068 } 1069 return (count); 1070 } 1071 1072 /* 1073 * Insert vdev into top_vdev children list. List is ordered by v_id. 1074 */ 1075 static void 1076 vdev_insert(vdev_t *top_vdev, vdev_t *vdev) 1077 { 1078 vdev_t *previous; 1079 size_t count; 1080 1081 /* 1082 * The top level vdev can appear in random order, depending how 1083 * the firmware is presenting the disk devices. 1084 * However, we will insert vdev to create list ordered by v_id, 1085 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER 1086 * as STAILQ does not have insert before. 1087 */ 1088 previous = vdev_find_previous(top_vdev, vdev); 1089 1090 if (previous == NULL) { 1091 STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink); 1092 } else if (previous->v_id == vdev->v_id) { 1093 /* 1094 * This vdev was configured from label config, 1095 * do not insert duplicate. 1096 */ 1097 return; 1098 } else { 1099 STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev, 1100 v_childlink); 1101 } 1102 1103 count = vdev_child_count(top_vdev); 1104 if (top_vdev->v_nchildren < count) 1105 top_vdev->v_nchildren = count; 1106 } 1107 1108 static int 1109 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, uint64_t txg, 1110 const nvlist_t *nvlist) 1111 { 1112 vdev_t *top_vdev, *vdev; 1113 nvlist_t **kids = NULL; 1114 int rc, nkids; 1115 1116 /* Get top vdev. */ 1117 top_vdev = vdev_find(top_guid); 1118 if (top_vdev == NULL) { 1119 rc = vdev_init(top_guid, nvlist, &top_vdev); 1120 if (rc != 0) 1121 return (rc); 1122 top_vdev->v_spa = spa; 1123 top_vdev->v_top = top_vdev; 1124 top_vdev->v_txg = txg; 1125 vdev_insert(spa->spa_root_vdev, top_vdev); 1126 } 1127 1128 /* Add children if there are any. */ 1129 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1130 &nkids, &kids, NULL); 1131 if (rc == 0) { 1132 for (int i = 0; i < nkids; i++) { 1133 uint64_t guid; 1134 1135 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, 1136 DATA_TYPE_UINT64, NULL, &guid, NULL); 1137 if (rc != 0) 1138 goto done; 1139 1140 rc = vdev_init(guid, kids[i], &vdev); 1141 if (rc != 0) 1142 goto done; 1143 1144 vdev->v_spa = spa; 1145 vdev->v_top = top_vdev; 1146 vdev_insert(top_vdev, vdev); 1147 } 1148 } else { 1149 /* 1150 * When there are no children, nvlist_find() does return 1151 * error, reset it because leaf devices have no children. 1152 */ 1153 rc = 0; 1154 } 1155 done: 1156 if (kids != NULL) { 1157 for (int i = 0; i < nkids; i++) 1158 nvlist_destroy(kids[i]); 1159 free(kids); 1160 } 1161 1162 return (rc); 1163 } 1164 1165 static int 1166 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist) 1167 { 1168 uint64_t pool_guid, top_guid, txg; 1169 nvlist_t *vdevs; 1170 int rc; 1171 1172 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 1173 NULL, &pool_guid, NULL) || 1174 nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64, 1175 NULL, &top_guid, NULL) || 1176 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64, 1177 NULL, &txg, NULL) != 0 || 1178 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1179 NULL, &vdevs, NULL)) { 1180 printf("ZFS: can't find vdev details\n"); 1181 return (ENOENT); 1182 } 1183 1184 rc = vdev_from_nvlist(spa, top_guid, txg, vdevs); 1185 nvlist_destroy(vdevs); 1186 return (rc); 1187 } 1188 1189 static void 1190 vdev_set_state(vdev_t *vdev) 1191 { 1192 vdev_t *kid; 1193 int good_kids; 1194 int bad_kids; 1195 1196 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1197 vdev_set_state(kid); 1198 } 1199 1200 /* 1201 * A mirror or raidz is healthy if all its kids are healthy. A 1202 * mirror is degraded if any of its kids is healthy; a raidz 1203 * is degraded if at most nparity kids are offline. 1204 */ 1205 if (STAILQ_FIRST(&vdev->v_children)) { 1206 good_kids = 0; 1207 bad_kids = 0; 1208 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1209 if (kid->v_state == VDEV_STATE_HEALTHY) 1210 good_kids++; 1211 else 1212 bad_kids++; 1213 } 1214 if (bad_kids == 0) { 1215 vdev->v_state = VDEV_STATE_HEALTHY; 1216 } else { 1217 if (vdev->v_read == vdev_mirror_read) { 1218 if (good_kids) { 1219 vdev->v_state = VDEV_STATE_DEGRADED; 1220 } else { 1221 vdev->v_state = VDEV_STATE_OFFLINE; 1222 } 1223 } else if (vdev->v_read == vdev_raidz_read) { 1224 if (bad_kids > vdev->v_nparity) { 1225 vdev->v_state = VDEV_STATE_OFFLINE; 1226 } else { 1227 vdev->v_state = VDEV_STATE_DEGRADED; 1228 } 1229 } 1230 } 1231 } 1232 } 1233 1234 static int 1235 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist) 1236 { 1237 vdev_t *vdev; 1238 nvlist_t **kids = NULL; 1239 int rc, nkids; 1240 1241 /* Update top vdev. */ 1242 vdev = vdev_find(top_guid); 1243 if (vdev != NULL) 1244 vdev_set_initial_state(vdev, nvlist); 1245 1246 /* Update children if there are any. */ 1247 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1248 &nkids, &kids, NULL); 1249 if (rc == 0) { 1250 for (int i = 0; i < nkids; i++) { 1251 uint64_t guid; 1252 1253 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, 1254 DATA_TYPE_UINT64, NULL, &guid, NULL); 1255 if (rc != 0) 1256 break; 1257 1258 vdev = vdev_find(guid); 1259 if (vdev != NULL) 1260 vdev_set_initial_state(vdev, kids[i]); 1261 } 1262 } else { 1263 rc = 0; 1264 } 1265 if (kids != NULL) { 1266 for (int i = 0; i < nkids; i++) 1267 nvlist_destroy(kids[i]); 1268 free(kids); 1269 } 1270 1271 return (rc); 1272 } 1273 1274 /* 1275 * Shall not be called on root vdev, that is not linked into zfs_vdevs. 1276 * See comment in vdev_create(). 1277 */ 1278 static void 1279 vdev_free(struct vdev *vdev) 1280 { 1281 struct vdev *kid, *safe; 1282 1283 STAILQ_FOREACH_SAFE(kid, &vdev->v_children, v_childlink, safe) 1284 vdev_free(kid); 1285 STAILQ_REMOVE(&zfs_vdevs, vdev, vdev, v_alllink); 1286 free(vdev); 1287 } 1288 1289 static int 1290 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist) 1291 { 1292 uint64_t pool_guid, vdev_children; 1293 nvlist_t *vdevs = NULL, **kids = NULL; 1294 int rc, nkids; 1295 1296 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 1297 NULL, &pool_guid, NULL) || 1298 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64, 1299 NULL, &vdev_children, NULL) || 1300 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1301 NULL, &vdevs, NULL)) { 1302 printf("ZFS: can't find vdev details\n"); 1303 return (ENOENT); 1304 } 1305 1306 /* Wrong guid?! */ 1307 if (spa->spa_guid != pool_guid) { 1308 nvlist_destroy(vdevs); 1309 return (EINVAL); 1310 } 1311 1312 spa->spa_root_vdev->v_nchildren = vdev_children; 1313 1314 rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY, 1315 &nkids, &kids, NULL); 1316 nvlist_destroy(vdevs); 1317 1318 /* 1319 * MOS config has at least one child for root vdev. 1320 */ 1321 if (rc != 0) 1322 return (rc); 1323 1324 for (int i = 0; i < nkids; i++) { 1325 uint64_t guid; 1326 vdev_t *vdev; 1327 1328 rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64, 1329 NULL, &guid, NULL); 1330 if (rc != 0) 1331 break; 1332 vdev = vdev_find(guid); 1333 /* 1334 * Top level vdev is missing, create it. 1335 * XXXGL: how can this happen? 1336 */ 1337 if (vdev == NULL) 1338 rc = vdev_from_nvlist(spa, guid, 0, kids[i]); 1339 else 1340 rc = vdev_update_from_nvlist(guid, kids[i]); 1341 if (rc != 0) 1342 break; 1343 } 1344 if (kids != NULL) { 1345 for (int i = 0; i < nkids; i++) 1346 nvlist_destroy(kids[i]); 1347 free(kids); 1348 } 1349 1350 /* 1351 * Re-evaluate top-level vdev state. 1352 */ 1353 vdev_set_state(spa->spa_root_vdev); 1354 1355 return (rc); 1356 } 1357 1358 static spa_t * 1359 spa_find_by_guid(uint64_t guid) 1360 { 1361 spa_t *spa; 1362 1363 STAILQ_FOREACH(spa, &zfs_pools, spa_link) 1364 if (spa->spa_guid == guid) 1365 return (spa); 1366 1367 return (NULL); 1368 } 1369 1370 static spa_t * 1371 spa_find_by_name(const char *name) 1372 { 1373 spa_t *spa; 1374 1375 STAILQ_FOREACH(spa, &zfs_pools, spa_link) 1376 if (strcmp(spa->spa_name, name) == 0) 1377 return (spa); 1378 1379 return (NULL); 1380 } 1381 1382 static spa_t * 1383 spa_create(uint64_t guid, const char *name) 1384 { 1385 spa_t *spa; 1386 1387 if ((spa = calloc(1, sizeof(spa_t))) == NULL) 1388 return (NULL); 1389 if ((spa->spa_name = strdup(name)) == NULL) { 1390 free(spa); 1391 return (NULL); 1392 } 1393 spa->spa_uberblock = &spa->spa_uberblock_master; 1394 spa->spa_mos = &spa->spa_mos_master; 1395 spa->spa_guid = guid; 1396 spa->spa_root_vdev = vdev_create(guid, NULL); 1397 if (spa->spa_root_vdev == NULL) { 1398 free(spa->spa_name); 1399 free(spa); 1400 return (NULL); 1401 } 1402 spa->spa_root_vdev->v_name = spa->spa_name; 1403 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link); 1404 1405 return (spa); 1406 } 1407 1408 static const char * 1409 state_name(vdev_state_t state) 1410 { 1411 static const char *names[] = { 1412 "UNKNOWN", 1413 "CLOSED", 1414 "OFFLINE", 1415 "REMOVED", 1416 "CANT_OPEN", 1417 "FAULTED", 1418 "DEGRADED", 1419 "ONLINE" 1420 }; 1421 return (names[state]); 1422 } 1423 1424 #ifdef BOOT2 1425 1426 #define pager_printf printf 1427 1428 #else 1429 1430 static int 1431 pager_printf(const char *fmt, ...) 1432 { 1433 char line[80]; 1434 va_list args; 1435 1436 va_start(args, fmt); 1437 vsnprintf(line, sizeof(line), fmt, args); 1438 va_end(args); 1439 return (pager_output(line)); 1440 } 1441 1442 #endif 1443 1444 #define STATUS_FORMAT " %s %s\n" 1445 1446 static int 1447 print_state(int indent, const char *name, vdev_state_t state) 1448 { 1449 int i; 1450 char buf[512]; 1451 1452 buf[0] = 0; 1453 for (i = 0; i < indent; i++) 1454 strcat(buf, " "); 1455 strcat(buf, name); 1456 return (pager_printf(STATUS_FORMAT, buf, state_name(state))); 1457 } 1458 1459 static int 1460 vdev_status(vdev_t *vdev, int indent) 1461 { 1462 vdev_t *kid; 1463 int ret; 1464 1465 if (vdev->v_islog) { 1466 (void) pager_output(" logs\n"); 1467 indent++; 1468 } 1469 1470 ret = print_state(indent, vdev->v_name, vdev->v_state); 1471 if (ret != 0) 1472 return (ret); 1473 1474 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1475 ret = vdev_status(kid, indent + 1); 1476 if (ret != 0) 1477 return (ret); 1478 } 1479 return (ret); 1480 } 1481 1482 static int 1483 spa_status(spa_t *spa) 1484 { 1485 static char bootfs[ZFS_MAXNAMELEN]; 1486 uint64_t rootid; 1487 vdev_list_t *vlist; 1488 vdev_t *vdev; 1489 int good_kids, bad_kids, degraded_kids, ret; 1490 vdev_state_t state; 1491 1492 ret = pager_printf(" pool: %s\n", spa->spa_name); 1493 if (ret != 0) 1494 return (ret); 1495 1496 if (zfs_get_root(spa, &rootid) == 0 && 1497 zfs_rlookup(spa, rootid, bootfs) == 0) { 1498 if (bootfs[0] == '\0') 1499 ret = pager_printf("bootfs: %s\n", spa->spa_name); 1500 else 1501 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name, 1502 bootfs); 1503 if (ret != 0) 1504 return (ret); 1505 } 1506 ret = pager_printf("config:\n\n"); 1507 if (ret != 0) 1508 return (ret); 1509 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE"); 1510 if (ret != 0) 1511 return (ret); 1512 1513 good_kids = 0; 1514 degraded_kids = 0; 1515 bad_kids = 0; 1516 vlist = &spa->spa_root_vdev->v_children; 1517 STAILQ_FOREACH(vdev, vlist, v_childlink) { 1518 if (vdev->v_state == VDEV_STATE_HEALTHY) 1519 good_kids++; 1520 else if (vdev->v_state == VDEV_STATE_DEGRADED) 1521 degraded_kids++; 1522 else 1523 bad_kids++; 1524 } 1525 1526 state = VDEV_STATE_CLOSED; 1527 if (good_kids > 0 && (degraded_kids + bad_kids) == 0) 1528 state = VDEV_STATE_HEALTHY; 1529 else if ((good_kids + degraded_kids) > 0) 1530 state = VDEV_STATE_DEGRADED; 1531 1532 ret = print_state(0, spa->spa_name, state); 1533 if (ret != 0) 1534 return (ret); 1535 1536 STAILQ_FOREACH(vdev, vlist, v_childlink) { 1537 ret = vdev_status(vdev, 1); 1538 if (ret != 0) 1539 return (ret); 1540 } 1541 return (ret); 1542 } 1543 1544 static int 1545 spa_all_status(void) 1546 { 1547 spa_t *spa; 1548 int first = 1, ret = 0; 1549 1550 STAILQ_FOREACH(spa, &zfs_pools, spa_link) { 1551 if (!first) { 1552 ret = pager_printf("\n"); 1553 if (ret != 0) 1554 return (ret); 1555 } 1556 first = 0; 1557 ret = spa_status(spa); 1558 if (ret != 0) 1559 return (ret); 1560 } 1561 return (ret); 1562 } 1563 1564 static uint64_t 1565 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 1566 { 1567 uint64_t label_offset; 1568 1569 if (l < VDEV_LABELS / 2) 1570 label_offset = 0; 1571 else 1572 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t); 1573 1574 return (offset + l * sizeof (vdev_label_t) + label_offset); 1575 } 1576 1577 static int 1578 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2) 1579 { 1580 unsigned int seq1 = 0; 1581 unsigned int seq2 = 0; 1582 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg); 1583 1584 if (cmp != 0) 1585 return (cmp); 1586 1587 cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp); 1588 if (cmp != 0) 1589 return (cmp); 1590 1591 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1)) 1592 seq1 = MMP_SEQ(ub1); 1593 1594 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2)) 1595 seq2 = MMP_SEQ(ub2); 1596 1597 return (AVL_CMP(seq1, seq2)); 1598 } 1599 1600 static int 1601 uberblock_verify(uberblock_t *ub) 1602 { 1603 if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) { 1604 byteswap_uint64_array(ub, sizeof (uberblock_t)); 1605 } 1606 1607 if (ub->ub_magic != UBERBLOCK_MAGIC || 1608 !SPA_VERSION_IS_SUPPORTED(ub->ub_version)) 1609 return (EINVAL); 1610 1611 return (0); 1612 } 1613 1614 static int 1615 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset, 1616 size_t size) 1617 { 1618 blkptr_t bp; 1619 off_t off; 1620 1621 off = vdev_label_offset(vd->v_psize, l, offset); 1622 1623 BP_ZERO(&bp); 1624 BP_SET_LSIZE(&bp, size); 1625 BP_SET_PSIZE(&bp, size); 1626 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 1627 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF); 1628 DVA_SET_OFFSET(BP_IDENTITY(&bp), off); 1629 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0); 1630 1631 return (vdev_read_phys(vd, &bp, buf, off, size)); 1632 } 1633 1634 /* 1635 * We do need to be sure we write to correct location. 1636 * Our vdev label does consist of 4 fields: 1637 * pad1 (8k), reserved. 1638 * bootenv (8k), checksummed, previously reserved, may contian garbage. 1639 * vdev_phys (112k), checksummed 1640 * uberblock ring (128k), checksummed. 1641 * 1642 * Since bootenv area may contain garbage, we can not reliably read it, as 1643 * we can get checksum errors. 1644 * Next best thing is vdev_phys - it is just after bootenv. It still may 1645 * be corrupted, but in such case we will miss this one write. 1646 */ 1647 static int 1648 vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset) 1649 { 1650 uint64_t off, o_phys; 1651 void *buf; 1652 size_t size = VDEV_PHYS_SIZE; 1653 int rc; 1654 1655 o_phys = offsetof(vdev_label_t, vl_vdev_phys); 1656 off = vdev_label_offset(vd->v_psize, l, o_phys); 1657 1658 /* off should be 8K from bootenv */ 1659 if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off) 1660 return (EINVAL); 1661 1662 buf = malloc(size); 1663 if (buf == NULL) 1664 return (ENOMEM); 1665 1666 /* Read vdev_phys */ 1667 rc = vdev_label_read(vd, l, buf, o_phys, size); 1668 free(buf); 1669 return (rc); 1670 } 1671 1672 static int 1673 vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset) 1674 { 1675 zio_checksum_info_t *ci; 1676 zio_cksum_t cksum; 1677 off_t off; 1678 size_t size = VDEV_PAD_SIZE; 1679 int rc; 1680 1681 if (vd->v_phys_write == NULL) 1682 return (ENOTSUP); 1683 1684 off = vdev_label_offset(vd->v_psize, l, offset); 1685 1686 rc = vdev_label_write_validate(vd, l, offset); 1687 if (rc != 0) { 1688 return (rc); 1689 } 1690 1691 ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL]; 1692 be->vbe_zbt.zec_magic = ZEC_MAGIC; 1693 zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off); 1694 ci->ci_func[0](be, size, NULL, &cksum); 1695 be->vbe_zbt.zec_cksum = cksum; 1696 1697 return (vdev_write_phys(vd, be, off, size)); 1698 } 1699 1700 static int 1701 vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be) 1702 { 1703 vdev_t *kid; 1704 int rv = 0, err; 1705 1706 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1707 if (kid->v_state != VDEV_STATE_HEALTHY) 1708 continue; 1709 err = vdev_write_bootenv_impl(kid, be); 1710 if (err != 0) 1711 rv = err; 1712 } 1713 1714 /* 1715 * Non-leaf vdevs do not have v_phys_write. 1716 */ 1717 if (vdev->v_phys_write == NULL) 1718 return (rv); 1719 1720 for (int l = 0; l < VDEV_LABELS; l++) { 1721 err = vdev_label_write(vdev, l, be, 1722 offsetof(vdev_label_t, vl_be)); 1723 if (err != 0) { 1724 printf("failed to write bootenv to %s label %d: %d\n", 1725 vdev->v_name ? vdev->v_name : "unknown", l, err); 1726 rv = err; 1727 } 1728 } 1729 return (rv); 1730 } 1731 1732 int 1733 vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl) 1734 { 1735 vdev_boot_envblock_t *be; 1736 nvlist_t nv, *nvp; 1737 uint64_t version; 1738 int rv; 1739 1740 if (nvl->nv_size > sizeof(be->vbe_bootenv)) 1741 return (E2BIG); 1742 1743 version = VB_RAW; 1744 nvp = vdev_read_bootenv(vdev); 1745 if (nvp != NULL) { 1746 nvlist_find(nvp, BOOTENV_VERSION, DATA_TYPE_UINT64, NULL, 1747 &version, NULL); 1748 nvlist_destroy(nvp); 1749 } 1750 1751 be = calloc(1, sizeof(*be)); 1752 if (be == NULL) 1753 return (ENOMEM); 1754 1755 be->vbe_version = version; 1756 switch (version) { 1757 case VB_RAW: 1758 /* 1759 * If there is no envmap, we will just wipe bootenv. 1760 */ 1761 nvlist_find(nvl, GRUB_ENVMAP, DATA_TYPE_STRING, NULL, 1762 be->vbe_bootenv, NULL); 1763 rv = 0; 1764 break; 1765 1766 case VB_NVLIST: 1767 nv.nv_header = nvl->nv_header; 1768 nv.nv_asize = nvl->nv_asize; 1769 nv.nv_size = nvl->nv_size; 1770 1771 bcopy(&nv.nv_header, be->vbe_bootenv, sizeof(nv.nv_header)); 1772 nv.nv_data = be->vbe_bootenv + sizeof(nvs_header_t); 1773 bcopy(nvl->nv_data, nv.nv_data, nv.nv_size); 1774 rv = nvlist_export(&nv); 1775 break; 1776 1777 default: 1778 rv = EINVAL; 1779 break; 1780 } 1781 1782 if (rv == 0) { 1783 be->vbe_version = htobe64(be->vbe_version); 1784 rv = vdev_write_bootenv_impl(vdev, be); 1785 } 1786 free(be); 1787 return (rv); 1788 } 1789 1790 /* 1791 * Read the bootenv area from pool label, return the nvlist from it. 1792 * We return from first successful read. 1793 */ 1794 nvlist_t * 1795 vdev_read_bootenv(vdev_t *vdev) 1796 { 1797 vdev_t *kid; 1798 nvlist_t *benv; 1799 vdev_boot_envblock_t *be; 1800 char *command; 1801 bool ok; 1802 int rv; 1803 1804 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) { 1805 if (kid->v_state != VDEV_STATE_HEALTHY) 1806 continue; 1807 1808 benv = vdev_read_bootenv(kid); 1809 if (benv != NULL) 1810 return (benv); 1811 } 1812 1813 be = malloc(sizeof (*be)); 1814 if (be == NULL) 1815 return (NULL); 1816 1817 rv = 0; 1818 for (int l = 0; l < VDEV_LABELS; l++) { 1819 rv = vdev_label_read(vdev, l, be, 1820 offsetof(vdev_label_t, vl_be), 1821 sizeof (*be)); 1822 if (rv == 0) 1823 break; 1824 } 1825 if (rv != 0) { 1826 free(be); 1827 return (NULL); 1828 } 1829 1830 be->vbe_version = be64toh(be->vbe_version); 1831 switch (be->vbe_version) { 1832 case VB_RAW: 1833 /* 1834 * we have textual data in vbe_bootenv, create nvlist 1835 * with key "envmap". 1836 */ 1837 benv = nvlist_create(NV_UNIQUE_NAME); 1838 if (benv != NULL) { 1839 if (*be->vbe_bootenv == '\0') { 1840 nvlist_add_uint64(benv, BOOTENV_VERSION, 1841 VB_NVLIST); 1842 break; 1843 } 1844 nvlist_add_uint64(benv, BOOTENV_VERSION, VB_RAW); 1845 be->vbe_bootenv[sizeof (be->vbe_bootenv) - 1] = '\0'; 1846 nvlist_add_string(benv, GRUB_ENVMAP, be->vbe_bootenv); 1847 } 1848 break; 1849 1850 case VB_NVLIST: 1851 benv = nvlist_import(be->vbe_bootenv, sizeof(be->vbe_bootenv)); 1852 break; 1853 1854 default: 1855 command = (char *)be; 1856 ok = false; 1857 1858 /* Check for legacy zfsbootcfg command string */ 1859 for (int i = 0; command[i] != '\0'; i++) { 1860 if (iscntrl(command[i])) { 1861 ok = false; 1862 break; 1863 } else { 1864 ok = true; 1865 } 1866 } 1867 benv = nvlist_create(NV_UNIQUE_NAME); 1868 if (benv != NULL) { 1869 if (ok) 1870 nvlist_add_string(benv, FREEBSD_BOOTONCE, 1871 command); 1872 else 1873 nvlist_add_uint64(benv, BOOTENV_VERSION, 1874 VB_NVLIST); 1875 } 1876 break; 1877 } 1878 free(be); 1879 return (benv); 1880 } 1881 1882 static uint64_t 1883 vdev_get_label_asize(nvlist_t *nvl) 1884 { 1885 nvlist_t *vdevs; 1886 uint64_t asize; 1887 const char *type; 1888 int len; 1889 1890 asize = 0; 1891 /* Get vdev tree */ 1892 if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, 1893 NULL, &vdevs, NULL) != 0) 1894 return (asize); 1895 1896 /* 1897 * Get vdev type. We will calculate asize for raidz, mirror and disk. 1898 * For raidz, the asize is raw size of all children. 1899 */ 1900 if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, 1901 NULL, &type, &len) != 0) 1902 goto done; 1903 1904 if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 && 1905 memcmp(type, VDEV_TYPE_DISK, len) != 0 && 1906 memcmp(type, VDEV_TYPE_RAIDZ, len) != 0) 1907 goto done; 1908 1909 if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64, 1910 NULL, &asize, NULL) != 0) 1911 goto done; 1912 1913 if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) { 1914 nvlist_t **kids; 1915 int nkids; 1916 1917 if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, 1918 DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) { 1919 asize = 0; 1920 goto done; 1921 } 1922 1923 asize /= nkids; 1924 for (int i = 0; i < nkids; i++) 1925 nvlist_destroy(kids[i]); 1926 free(kids); 1927 } 1928 1929 asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 1930 done: 1931 nvlist_destroy(vdevs); 1932 return (asize); 1933 } 1934 1935 static nvlist_t * 1936 vdev_label_read_config(vdev_t *vd, uint64_t txg) 1937 { 1938 vdev_phys_t *label; 1939 uint64_t best_txg = 0; 1940 uint64_t label_txg = 0; 1941 uint64_t asize; 1942 nvlist_t *nvl = NULL, *tmp; 1943 int error; 1944 1945 label = malloc(sizeof (vdev_phys_t)); 1946 if (label == NULL) 1947 return (NULL); 1948 1949 for (int l = 0; l < VDEV_LABELS; l++) { 1950 if (vdev_label_read(vd, l, label, 1951 offsetof(vdev_label_t, vl_vdev_phys), 1952 sizeof (vdev_phys_t))) 1953 continue; 1954 1955 tmp = nvlist_import(label->vp_nvlist, 1956 sizeof(label->vp_nvlist)); 1957 if (tmp == NULL) 1958 continue; 1959 1960 error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG, 1961 DATA_TYPE_UINT64, NULL, &label_txg, NULL); 1962 if (error != 0 || label_txg == 0) { 1963 nvlist_destroy(nvl); 1964 nvl = tmp; 1965 goto done; 1966 } 1967 1968 if (label_txg <= txg && label_txg > best_txg) { 1969 best_txg = label_txg; 1970 nvlist_destroy(nvl); 1971 nvl = tmp; 1972 tmp = NULL; 1973 1974 /* 1975 * Use asize from pool config. We need this 1976 * because we can get bad value from BIOS. 1977 */ 1978 asize = vdev_get_label_asize(nvl); 1979 if (asize != 0) { 1980 vd->v_psize = asize; 1981 } 1982 } 1983 nvlist_destroy(tmp); 1984 } 1985 1986 if (best_txg == 0) { 1987 nvlist_destroy(nvl); 1988 nvl = NULL; 1989 } 1990 done: 1991 free(label); 1992 return (nvl); 1993 } 1994 1995 static void 1996 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub) 1997 { 1998 uberblock_t *buf; 1999 2000 buf = malloc(VDEV_UBERBLOCK_SIZE(vd)); 2001 if (buf == NULL) 2002 return; 2003 2004 for (int l = 0; l < VDEV_LABELS; l++) { 2005 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 2006 if (vdev_label_read(vd, l, buf, 2007 VDEV_UBERBLOCK_OFFSET(vd, n), 2008 VDEV_UBERBLOCK_SIZE(vd))) 2009 continue; 2010 if (uberblock_verify(buf) != 0) 2011 continue; 2012 2013 if (vdev_uberblock_compare(buf, ub) > 0) 2014 *ub = *buf; 2015 } 2016 } 2017 free(buf); 2018 } 2019 2020 static int 2021 vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv, 2022 spa_t **spap) 2023 { 2024 vdev_t vtmp; 2025 spa_t *spa; 2026 vdev_t *vdev; 2027 nvlist_t *nvl; 2028 uint64_t val; 2029 uint64_t guid, pool_guid, top_guid, txg; 2030 const char *pool_name; 2031 int rc, namelen; 2032 2033 /* 2034 * Load the vdev label and figure out which 2035 * uberblock is most current. 2036 */ 2037 memset(&vtmp, 0, sizeof(vtmp)); 2038 vtmp.v_phys_read = _read; 2039 vtmp.v_phys_write = _write; 2040 vtmp.v_priv = priv; 2041 vtmp.v_psize = P2ALIGN(ldi_get_size(priv), 2042 (uint64_t)sizeof (vdev_label_t)); 2043 2044 /* Test for minimum device size. */ 2045 if (vtmp.v_psize < SPA_MINDEVSIZE) 2046 return (EIO); 2047 2048 nvl = vdev_label_read_config(&vtmp, UINT64_MAX); 2049 if (nvl == NULL) 2050 return (EIO); 2051 2052 if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64, 2053 NULL, &val, NULL) != 0) { 2054 nvlist_destroy(nvl); 2055 return (EIO); 2056 } 2057 2058 if (!SPA_VERSION_IS_SUPPORTED(val)) { 2059 printf("ZFS: unsupported ZFS version %u (should be %u)\n", 2060 (unsigned)val, (unsigned)SPA_VERSION); 2061 nvlist_destroy(nvl); 2062 return (EIO); 2063 } 2064 2065 /* Check ZFS features for read */ 2066 rc = nvlist_check_features_for_read(nvl); 2067 if (rc != 0) { 2068 nvlist_destroy(nvl); 2069 return (EIO); 2070 } 2071 2072 if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64, 2073 NULL, &val, NULL) != 0) { 2074 nvlist_destroy(nvl); 2075 return (EIO); 2076 } 2077 2078 if (val == POOL_STATE_DESTROYED) { 2079 /* We don't boot only from destroyed pools. */ 2080 nvlist_destroy(nvl); 2081 return (EIO); 2082 } 2083 2084 if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64, 2085 NULL, &txg, NULL) != 0 || 2086 nvlist_find(nvl, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64, 2087 NULL, &top_guid, NULL) != 0 || 2088 nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64, 2089 NULL, &pool_guid, NULL) != 0 || 2090 nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING, 2091 NULL, &pool_name, &namelen) != 0 || 2092 nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64, 2093 NULL, &guid, NULL) != 0) { 2094 /* 2095 * Cache and spare devices end up here - just ignore 2096 * them. 2097 */ 2098 nvlist_destroy(nvl); 2099 return (EIO); 2100 } 2101 2102 /* 2103 * Create the pool if this is the first time we've seen it. 2104 */ 2105 spa = spa_find_by_guid(pool_guid); 2106 if (spa == NULL) { 2107 char *name; 2108 2109 name = malloc(namelen + 1); 2110 if (name == NULL) { 2111 nvlist_destroy(nvl); 2112 return (ENOMEM); 2113 } 2114 bcopy(pool_name, name, namelen); 2115 name[namelen] = '\0'; 2116 spa = spa_create(pool_guid, name); 2117 free(name); 2118 if (spa == NULL) { 2119 nvlist_destroy(nvl); 2120 return (ENOMEM); 2121 } 2122 } else { 2123 struct vdev *kid; 2124 2125 STAILQ_FOREACH(kid, &spa->spa_root_vdev->v_children, 2126 v_childlink) 2127 if (kid->v_guid == top_guid && kid->v_txg < txg) { 2128 printf("ZFS: pool %s vdev %s ignoring stale " 2129 "label from txg 0x%jx, using 0x%jx@0x%jx\n", 2130 spa->spa_name, kid->v_name, 2131 kid->v_txg, guid, txg); 2132 STAILQ_REMOVE(&spa->spa_root_vdev->v_children, 2133 kid, vdev, v_childlink); 2134 vdev_free(kid); 2135 break; 2136 } 2137 } 2138 2139 /* 2140 * Get the vdev tree and create our in-core copy of it. 2141 * If we already have a vdev with this guid, this must 2142 * be some kind of alias (overlapping slices, dangerously dedicated 2143 * disks etc). 2144 */ 2145 vdev = vdev_find(guid); 2146 /* Has this vdev already been inited? */ 2147 if (vdev && vdev->v_phys_read) { 2148 nvlist_destroy(nvl); 2149 return (EIO); 2150 } 2151 2152 rc = vdev_init_from_label(spa, nvl); 2153 nvlist_destroy(nvl); 2154 if (rc != 0) 2155 return (rc); 2156 2157 /* 2158 * We should already have created an incomplete vdev for this 2159 * vdev. Find it and initialise it with our read proc. 2160 */ 2161 vdev = vdev_find(guid); 2162 if (vdev != NULL) { 2163 vdev->v_phys_read = _read; 2164 vdev->v_phys_write = _write; 2165 vdev->v_priv = priv; 2166 vdev->v_psize = vtmp.v_psize; 2167 /* 2168 * If no other state is set, mark vdev healthy. 2169 */ 2170 if (vdev->v_state == VDEV_STATE_UNKNOWN) 2171 vdev->v_state = VDEV_STATE_HEALTHY; 2172 } else { 2173 printf("ZFS: inconsistent nvlist contents\n"); 2174 return (EIO); 2175 } 2176 2177 if (vdev->v_islog) 2178 spa->spa_with_log = vdev->v_islog; 2179 2180 /* 2181 * Re-evaluate top-level vdev state. 2182 */ 2183 vdev_set_state(vdev->v_top); 2184 2185 /* 2186 * Ok, we are happy with the pool so far. Lets find 2187 * the best uberblock and then we can actually access 2188 * the contents of the pool. 2189 */ 2190 vdev_uberblock_load(vdev, spa->spa_uberblock); 2191 2192 if (spap != NULL) 2193 *spap = spa; 2194 return (0); 2195 } 2196 2197 static int 2198 ilog2(int n) 2199 { 2200 int v; 2201 2202 for (v = 0; v < 32; v++) 2203 if (n == (1 << v)) 2204 return (v); 2205 return (-1); 2206 } 2207 2208 static int 2209 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf) 2210 { 2211 blkptr_t gbh_bp; 2212 zio_gbh_phys_t zio_gb; 2213 char *pbuf; 2214 int i; 2215 2216 /* Artificial BP for gang block header. */ 2217 gbh_bp = *bp; 2218 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); 2219 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); 2220 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER); 2221 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF); 2222 for (i = 0; i < SPA_DVAS_PER_BP; i++) 2223 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0); 2224 2225 /* Read gang header block using the artificial BP. */ 2226 if (zio_read(spa, &gbh_bp, &zio_gb)) 2227 return (EIO); 2228 2229 pbuf = buf; 2230 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) { 2231 blkptr_t *gbp = &zio_gb.zg_blkptr[i]; 2232 2233 if (BP_IS_HOLE(gbp)) 2234 continue; 2235 if (zio_read(spa, gbp, pbuf)) 2236 return (EIO); 2237 pbuf += BP_GET_PSIZE(gbp); 2238 } 2239 2240 if (zio_checksum_verify(spa, bp, buf)) 2241 return (EIO); 2242 return (0); 2243 } 2244 2245 static int 2246 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) 2247 { 2248 int cpfunc = BP_GET_COMPRESS(bp); 2249 uint64_t align, size; 2250 void *pbuf; 2251 int i, error; 2252 2253 /* 2254 * Process data embedded in block pointer 2255 */ 2256 if (BP_IS_EMBEDDED(bp)) { 2257 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); 2258 2259 size = BPE_GET_PSIZE(bp); 2260 ASSERT(size <= BPE_PAYLOAD_SIZE); 2261 2262 if (cpfunc != ZIO_COMPRESS_OFF) 2263 pbuf = malloc(size); 2264 else 2265 pbuf = buf; 2266 2267 if (pbuf == NULL) 2268 return (ENOMEM); 2269 2270 decode_embedded_bp_compressed(bp, pbuf); 2271 error = 0; 2272 2273 if (cpfunc != ZIO_COMPRESS_OFF) { 2274 error = zio_decompress_data(cpfunc, pbuf, 2275 size, buf, BP_GET_LSIZE(bp)); 2276 free(pbuf); 2277 } 2278 if (error != 0) 2279 printf("ZFS: i/o error - unable to decompress " 2280 "block pointer data, error %d\n", error); 2281 return (error); 2282 } 2283 2284 error = EIO; 2285 2286 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 2287 const dva_t *dva = &bp->blk_dva[i]; 2288 vdev_t *vdev; 2289 vdev_list_t *vlist; 2290 uint64_t vdevid; 2291 off_t offset; 2292 2293 if (!dva->dva_word[0] && !dva->dva_word[1]) 2294 continue; 2295 2296 vdevid = DVA_GET_VDEV(dva); 2297 offset = DVA_GET_OFFSET(dva); 2298 vlist = &spa->spa_root_vdev->v_children; 2299 STAILQ_FOREACH(vdev, vlist, v_childlink) { 2300 if (vdev->v_id == vdevid) 2301 break; 2302 } 2303 if (!vdev || !vdev->v_read) 2304 continue; 2305 2306 size = BP_GET_PSIZE(bp); 2307 if (vdev->v_read == vdev_raidz_read) { 2308 align = 1ULL << vdev->v_ashift; 2309 if (P2PHASE(size, align) != 0) 2310 size = P2ROUNDUP(size, align); 2311 } 2312 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF) 2313 pbuf = malloc(size); 2314 else 2315 pbuf = buf; 2316 2317 if (pbuf == NULL) { 2318 error = ENOMEM; 2319 break; 2320 } 2321 2322 if (DVA_GET_GANG(dva)) 2323 error = zio_read_gang(spa, bp, pbuf); 2324 else 2325 error = vdev->v_read(vdev, bp, pbuf, offset, size); 2326 if (error == 0) { 2327 if (cpfunc != ZIO_COMPRESS_OFF) 2328 error = zio_decompress_data(cpfunc, pbuf, 2329 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp)); 2330 else if (size != BP_GET_PSIZE(bp)) 2331 bcopy(pbuf, buf, BP_GET_PSIZE(bp)); 2332 } else { 2333 printf("zio_read error: %d\n", error); 2334 } 2335 if (buf != pbuf) 2336 free(pbuf); 2337 if (error == 0) 2338 break; 2339 } 2340 if (error != 0) 2341 printf("ZFS: i/o error - all block copies unavailable\n"); 2342 2343 return (error); 2344 } 2345 2346 static int 2347 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, 2348 void *buf, size_t buflen) 2349 { 2350 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT; 2351 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2352 int nlevels = dnode->dn_nlevels; 2353 int i, rc; 2354 2355 if (bsize > SPA_MAXBLOCKSIZE) { 2356 printf("ZFS: I/O error - blocks larger than %llu are not " 2357 "supported\n", SPA_MAXBLOCKSIZE); 2358 return (EIO); 2359 } 2360 2361 /* 2362 * Handle odd block sizes, mirrors dmu_read_impl(). Data can't exist 2363 * past the first block, so we'll clip the read to the portion of the 2364 * buffer within bsize and zero out the remainder. 2365 */ 2366 if (dnode->dn_maxblkid == 0) { 2367 size_t newbuflen; 2368 2369 newbuflen = offset > bsize ? 0 : MIN(buflen, bsize - offset); 2370 bzero((char *)buf + newbuflen, buflen - newbuflen); 2371 buflen = newbuflen; 2372 } 2373 2374 /* 2375 * Note: bsize may not be a power of two here so we need to do an 2376 * actual divide rather than a bitshift. 2377 */ 2378 while (buflen > 0) { 2379 uint64_t bn = offset / bsize; 2380 int boff = offset % bsize; 2381 int ibn; 2382 const blkptr_t *indbp; 2383 blkptr_t bp; 2384 2385 if (bn > dnode->dn_maxblkid) 2386 return (EIO); 2387 2388 if (dnode == dnode_cache_obj && bn == dnode_cache_bn) 2389 goto cached; 2390 2391 indbp = dnode->dn_blkptr; 2392 for (i = 0; i < nlevels; i++) { 2393 /* 2394 * Copy the bp from the indirect array so that 2395 * we can re-use the scratch buffer for multi-level 2396 * objects. 2397 */ 2398 ibn = bn >> ((nlevels - i - 1) * ibshift); 2399 ibn &= ((1 << ibshift) - 1); 2400 bp = indbp[ibn]; 2401 if (BP_IS_HOLE(&bp)) { 2402 memset(dnode_cache_buf, 0, bsize); 2403 break; 2404 } 2405 rc = zio_read(spa, &bp, dnode_cache_buf); 2406 if (rc) 2407 return (rc); 2408 indbp = (const blkptr_t *) dnode_cache_buf; 2409 } 2410 dnode_cache_obj = dnode; 2411 dnode_cache_bn = bn; 2412 cached: 2413 2414 /* 2415 * The buffer contains our data block. Copy what we 2416 * need from it and loop. 2417 */ 2418 i = bsize - boff; 2419 if (i > buflen) i = buflen; 2420 memcpy(buf, &dnode_cache_buf[boff], i); 2421 buf = ((char *)buf) + i; 2422 offset += i; 2423 buflen -= i; 2424 } 2425 2426 return (0); 2427 } 2428 2429 /* 2430 * Lookup a value in a microzap directory. 2431 */ 2432 static int 2433 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name, 2434 uint64_t *value) 2435 { 2436 const mzap_ent_phys_t *mze; 2437 int chunks, i; 2438 2439 /* 2440 * Microzap objects use exactly one block. Read the whole 2441 * thing. 2442 */ 2443 chunks = size / MZAP_ENT_LEN - 1; 2444 for (i = 0; i < chunks; i++) { 2445 mze = &mz->mz_chunk[i]; 2446 if (strcmp(mze->mze_name, name) == 0) { 2447 *value = mze->mze_value; 2448 return (0); 2449 } 2450 } 2451 2452 return (ENOENT); 2453 } 2454 2455 /* 2456 * Compare a name with a zap leaf entry. Return non-zero if the name 2457 * matches. 2458 */ 2459 static int 2460 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, 2461 const char *name) 2462 { 2463 size_t namelen; 2464 const zap_leaf_chunk_t *nc; 2465 const char *p; 2466 2467 namelen = zc->l_entry.le_name_numints; 2468 2469 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk); 2470 p = name; 2471 while (namelen > 0) { 2472 size_t len; 2473 2474 len = namelen; 2475 if (len > ZAP_LEAF_ARRAY_BYTES) 2476 len = ZAP_LEAF_ARRAY_BYTES; 2477 if (memcmp(p, nc->l_array.la_array, len)) 2478 return (0); 2479 p += len; 2480 namelen -= len; 2481 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next); 2482 } 2483 2484 return (1); 2485 } 2486 2487 /* 2488 * Extract a uint64_t value from a zap leaf entry. 2489 */ 2490 static uint64_t 2491 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc) 2492 { 2493 const zap_leaf_chunk_t *vc; 2494 int i; 2495 uint64_t value; 2496 const uint8_t *p; 2497 2498 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk); 2499 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) { 2500 value = (value << 8) | p[i]; 2501 } 2502 2503 return (value); 2504 } 2505 2506 static void 2507 stv(int len, void *addr, uint64_t value) 2508 { 2509 switch (len) { 2510 case 1: 2511 *(uint8_t *)addr = value; 2512 return; 2513 case 2: 2514 *(uint16_t *)addr = value; 2515 return; 2516 case 4: 2517 *(uint32_t *)addr = value; 2518 return; 2519 case 8: 2520 *(uint64_t *)addr = value; 2521 return; 2522 } 2523 } 2524 2525 /* 2526 * Extract a array from a zap leaf entry. 2527 */ 2528 static void 2529 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, 2530 uint64_t integer_size, uint64_t num_integers, void *buf) 2531 { 2532 uint64_t array_int_len = zc->l_entry.le_value_intlen; 2533 uint64_t value = 0; 2534 uint64_t *u64 = buf; 2535 char *p = buf; 2536 int len = MIN(zc->l_entry.le_value_numints, num_integers); 2537 int chunk = zc->l_entry.le_value_chunk; 2538 int byten = 0; 2539 2540 if (integer_size == 8 && len == 1) { 2541 *u64 = fzap_leaf_value(zl, zc); 2542 return; 2543 } 2544 2545 while (len > 0) { 2546 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array; 2547 int i; 2548 2549 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl)); 2550 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) { 2551 value = (value << 8) | la->la_array[i]; 2552 byten++; 2553 if (byten == array_int_len) { 2554 stv(integer_size, p, value); 2555 byten = 0; 2556 len--; 2557 if (len == 0) 2558 return; 2559 p += integer_size; 2560 } 2561 } 2562 chunk = la->la_next; 2563 } 2564 } 2565 2566 static int 2567 fzap_check_size(uint64_t integer_size, uint64_t num_integers) 2568 { 2569 2570 switch (integer_size) { 2571 case 1: 2572 case 2: 2573 case 4: 2574 case 8: 2575 break; 2576 default: 2577 return (EINVAL); 2578 } 2579 2580 if (integer_size * num_integers > ZAP_MAXVALUELEN) 2581 return (E2BIG); 2582 2583 return (0); 2584 } 2585 2586 static void 2587 zap_leaf_free(zap_leaf_t *leaf) 2588 { 2589 free(leaf->l_phys); 2590 free(leaf); 2591 } 2592 2593 static int 2594 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp) 2595 { 2596 int bs = FZAP_BLOCK_SHIFT(zap); 2597 int err; 2598 2599 *lp = malloc(sizeof(**lp)); 2600 if (*lp == NULL) 2601 return (ENOMEM); 2602 2603 (*lp)->l_bs = bs; 2604 (*lp)->l_phys = malloc(1 << bs); 2605 2606 if ((*lp)->l_phys == NULL) { 2607 free(*lp); 2608 return (ENOMEM); 2609 } 2610 err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys, 2611 1 << bs); 2612 if (err != 0) { 2613 zap_leaf_free(*lp); 2614 } 2615 return (err); 2616 } 2617 2618 static int 2619 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, 2620 uint64_t *valp) 2621 { 2622 int bs = FZAP_BLOCK_SHIFT(zap); 2623 uint64_t blk = idx >> (bs - 3); 2624 uint64_t off = idx & ((1 << (bs - 3)) - 1); 2625 uint64_t *buf; 2626 int rc; 2627 2628 buf = malloc(1 << zap->zap_block_shift); 2629 if (buf == NULL) 2630 return (ENOMEM); 2631 rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs, 2632 buf, 1 << zap->zap_block_shift); 2633 if (rc == 0) 2634 *valp = buf[off]; 2635 free(buf); 2636 return (rc); 2637 } 2638 2639 static int 2640 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp) 2641 { 2642 if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) { 2643 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 2644 return (0); 2645 } else { 2646 return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl, 2647 idx, valp)); 2648 } 2649 } 2650 2651 #define ZAP_HASH_IDX(hash, n) (((n) == 0) ? 0 : ((hash) >> (64 - (n)))) 2652 static int 2653 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp) 2654 { 2655 uint64_t idx, blk; 2656 int err; 2657 2658 idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift); 2659 err = zap_idx_to_blk(zap, idx, &blk); 2660 if (err != 0) 2661 return (err); 2662 return (zap_get_leaf_byblk(zap, blk, lp)); 2663 } 2664 2665 #define CHAIN_END 0xffff /* end of the chunk chain */ 2666 #define LEAF_HASH(l, h) \ 2667 ((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \ 2668 ((h) >> \ 2669 (64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len))) 2670 #define LEAF_HASH_ENTPTR(l, h) (&(l)->l_phys->l_hash[LEAF_HASH(l, h)]) 2671 2672 static int 2673 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name, 2674 uint64_t integer_size, uint64_t num_integers, void *value) 2675 { 2676 int rc; 2677 uint16_t *chunkp; 2678 struct zap_leaf_entry *le; 2679 2680 /* 2681 * Make sure this chunk matches our hash. 2682 */ 2683 if (zl->l_phys->l_hdr.lh_prefix_len > 0 && 2684 zl->l_phys->l_hdr.lh_prefix != 2685 hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len)) 2686 return (EIO); 2687 2688 rc = ENOENT; 2689 for (chunkp = LEAF_HASH_ENTPTR(zl, hash); 2690 *chunkp != CHAIN_END; chunkp = &le->le_next) { 2691 zap_leaf_chunk_t *zc; 2692 uint16_t chunk = *chunkp; 2693 2694 le = ZAP_LEAF_ENTRY(zl, chunk); 2695 if (le->le_hash != hash) 2696 continue; 2697 zc = &ZAP_LEAF_CHUNK(zl, chunk); 2698 if (fzap_name_equal(zl, zc, name)) { 2699 if (zc->l_entry.le_value_intlen > integer_size) { 2700 rc = EINVAL; 2701 } else { 2702 fzap_leaf_array(zl, zc, integer_size, 2703 num_integers, value); 2704 rc = 0; 2705 } 2706 break; 2707 } 2708 } 2709 return (rc); 2710 } 2711 2712 /* 2713 * Lookup a value in a fatzap directory. 2714 */ 2715 static int 2716 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2717 const char *name, uint64_t integer_size, uint64_t num_integers, 2718 void *value) 2719 { 2720 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2721 fat_zap_t z; 2722 zap_leaf_t *zl; 2723 uint64_t hash; 2724 int rc; 2725 2726 if (zh->zap_magic != ZAP_MAGIC) 2727 return (EIO); 2728 2729 if ((rc = fzap_check_size(integer_size, num_integers)) != 0) { 2730 return (rc); 2731 } 2732 2733 z.zap_block_shift = ilog2(bsize); 2734 z.zap_phys = zh; 2735 z.zap_spa = spa; 2736 z.zap_dnode = dnode; 2737 2738 hash = zap_hash(zh->zap_salt, name); 2739 rc = zap_deref_leaf(&z, hash, &zl); 2740 if (rc != 0) 2741 return (rc); 2742 2743 rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value); 2744 2745 zap_leaf_free(zl); 2746 return (rc); 2747 } 2748 2749 /* 2750 * Lookup a name in a zap object and return its value as a uint64_t. 2751 */ 2752 static int 2753 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, 2754 uint64_t integer_size, uint64_t num_integers, void *value) 2755 { 2756 int rc; 2757 zap_phys_t *zap; 2758 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2759 2760 zap = malloc(size); 2761 if (zap == NULL) 2762 return (ENOMEM); 2763 2764 rc = dnode_read(spa, dnode, 0, zap, size); 2765 if (rc) 2766 goto done; 2767 2768 switch (zap->zap_block_type) { 2769 case ZBT_MICRO: 2770 rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value); 2771 break; 2772 case ZBT_HEADER: 2773 rc = fzap_lookup(spa, dnode, zap, name, integer_size, 2774 num_integers, value); 2775 break; 2776 default: 2777 printf("ZFS: invalid zap_type=%" PRIx64 "\n", 2778 zap->zap_block_type); 2779 rc = EIO; 2780 } 2781 done: 2782 free(zap); 2783 return (rc); 2784 } 2785 2786 /* 2787 * List a microzap directory. 2788 */ 2789 static int 2790 mzap_list(const mzap_phys_t *mz, size_t size, 2791 int (*callback)(const char *, uint64_t)) 2792 { 2793 const mzap_ent_phys_t *mze; 2794 int chunks, i, rc; 2795 2796 /* 2797 * Microzap objects use exactly one block. Read the whole 2798 * thing. 2799 */ 2800 rc = 0; 2801 chunks = size / MZAP_ENT_LEN - 1; 2802 for (i = 0; i < chunks; i++) { 2803 mze = &mz->mz_chunk[i]; 2804 if (mze->mze_name[0]) { 2805 rc = callback(mze->mze_name, mze->mze_value); 2806 if (rc != 0) 2807 break; 2808 } 2809 } 2810 2811 return (rc); 2812 } 2813 2814 /* 2815 * List a fatzap directory. 2816 */ 2817 static int 2818 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2819 int (*callback)(const char *, uint64_t)) 2820 { 2821 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2822 fat_zap_t z; 2823 uint64_t i; 2824 int j, rc; 2825 2826 if (zh->zap_magic != ZAP_MAGIC) 2827 return (EIO); 2828 2829 z.zap_block_shift = ilog2(bsize); 2830 z.zap_phys = zh; 2831 2832 /* 2833 * This assumes that the leaf blocks start at block 1. The 2834 * documentation isn't exactly clear on this. 2835 */ 2836 zap_leaf_t zl; 2837 zl.l_bs = z.zap_block_shift; 2838 zl.l_phys = malloc(bsize); 2839 if (zl.l_phys == NULL) 2840 return (ENOMEM); 2841 2842 for (i = 0; i < zh->zap_num_leafs; i++) { 2843 off_t off = ((off_t)(i + 1)) << zl.l_bs; 2844 char name[256], *p; 2845 uint64_t value; 2846 2847 if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) { 2848 free(zl.l_phys); 2849 return (EIO); 2850 } 2851 2852 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) { 2853 zap_leaf_chunk_t *zc, *nc; 2854 int namelen; 2855 2856 zc = &ZAP_LEAF_CHUNK(&zl, j); 2857 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY) 2858 continue; 2859 namelen = zc->l_entry.le_name_numints; 2860 if (namelen > sizeof(name)) 2861 namelen = sizeof(name); 2862 2863 /* 2864 * Paste the name back together. 2865 */ 2866 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk); 2867 p = name; 2868 while (namelen > 0) { 2869 int len; 2870 len = namelen; 2871 if (len > ZAP_LEAF_ARRAY_BYTES) 2872 len = ZAP_LEAF_ARRAY_BYTES; 2873 memcpy(p, nc->l_array.la_array, len); 2874 p += len; 2875 namelen -= len; 2876 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next); 2877 } 2878 2879 /* 2880 * Assume the first eight bytes of the value are 2881 * a uint64_t. 2882 */ 2883 value = fzap_leaf_value(&zl, zc); 2884 2885 /* printf("%s 0x%jx\n", name, (uintmax_t)value); */ 2886 rc = callback((const char *)name, value); 2887 if (rc != 0) { 2888 free(zl.l_phys); 2889 return (rc); 2890 } 2891 } 2892 } 2893 2894 free(zl.l_phys); 2895 return (0); 2896 } 2897 2898 static int zfs_printf(const char *name, uint64_t value __unused) 2899 { 2900 2901 printf("%s\n", name); 2902 2903 return (0); 2904 } 2905 2906 /* 2907 * List a zap directory. 2908 */ 2909 static int 2910 zap_list(const spa_t *spa, const dnode_phys_t *dnode) 2911 { 2912 zap_phys_t *zap; 2913 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2914 int rc; 2915 2916 zap = malloc(size); 2917 if (zap == NULL) 2918 return (ENOMEM); 2919 2920 rc = dnode_read(spa, dnode, 0, zap, size); 2921 if (rc == 0) { 2922 if (zap->zap_block_type == ZBT_MICRO) 2923 rc = mzap_list((const mzap_phys_t *)zap, size, 2924 zfs_printf); 2925 else 2926 rc = fzap_list(spa, dnode, zap, zfs_printf); 2927 } 2928 free(zap); 2929 return (rc); 2930 } 2931 2932 static int 2933 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, 2934 dnode_phys_t *dnode) 2935 { 2936 off_t offset; 2937 2938 offset = objnum * sizeof(dnode_phys_t); 2939 return dnode_read(spa, &os->os_meta_dnode, offset, 2940 dnode, sizeof(dnode_phys_t)); 2941 } 2942 2943 /* 2944 * Lookup a name in a microzap directory. 2945 */ 2946 static int 2947 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value) 2948 { 2949 const mzap_ent_phys_t *mze; 2950 int chunks, i; 2951 2952 /* 2953 * Microzap objects use exactly one block. Read the whole 2954 * thing. 2955 */ 2956 chunks = size / MZAP_ENT_LEN - 1; 2957 for (i = 0; i < chunks; i++) { 2958 mze = &mz->mz_chunk[i]; 2959 if (value == mze->mze_value) { 2960 strcpy(name, mze->mze_name); 2961 return (0); 2962 } 2963 } 2964 2965 return (ENOENT); 2966 } 2967 2968 static void 2969 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name) 2970 { 2971 size_t namelen; 2972 const zap_leaf_chunk_t *nc; 2973 char *p; 2974 2975 namelen = zc->l_entry.le_name_numints; 2976 2977 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk); 2978 p = name; 2979 while (namelen > 0) { 2980 size_t len; 2981 len = namelen; 2982 if (len > ZAP_LEAF_ARRAY_BYTES) 2983 len = ZAP_LEAF_ARRAY_BYTES; 2984 memcpy(p, nc->l_array.la_array, len); 2985 p += len; 2986 namelen -= len; 2987 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next); 2988 } 2989 2990 *p = '\0'; 2991 } 2992 2993 static int 2994 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh, 2995 char *name, uint64_t value) 2996 { 2997 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 2998 fat_zap_t z; 2999 uint64_t i; 3000 int j, rc; 3001 3002 if (zh->zap_magic != ZAP_MAGIC) 3003 return (EIO); 3004 3005 z.zap_block_shift = ilog2(bsize); 3006 z.zap_phys = zh; 3007 3008 /* 3009 * This assumes that the leaf blocks start at block 1. The 3010 * documentation isn't exactly clear on this. 3011 */ 3012 zap_leaf_t zl; 3013 zl.l_bs = z.zap_block_shift; 3014 zl.l_phys = malloc(bsize); 3015 if (zl.l_phys == NULL) 3016 return (ENOMEM); 3017 3018 for (i = 0; i < zh->zap_num_leafs; i++) { 3019 off_t off = ((off_t)(i + 1)) << zl.l_bs; 3020 3021 rc = dnode_read(spa, dnode, off, zl.l_phys, bsize); 3022 if (rc != 0) 3023 goto done; 3024 3025 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) { 3026 zap_leaf_chunk_t *zc; 3027 3028 zc = &ZAP_LEAF_CHUNK(&zl, j); 3029 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY) 3030 continue; 3031 if (zc->l_entry.le_value_intlen != 8 || 3032 zc->l_entry.le_value_numints != 1) 3033 continue; 3034 3035 if (fzap_leaf_value(&zl, zc) == value) { 3036 fzap_name_copy(&zl, zc, name); 3037 goto done; 3038 } 3039 } 3040 } 3041 3042 rc = ENOENT; 3043 done: 3044 free(zl.l_phys); 3045 return (rc); 3046 } 3047 3048 static int 3049 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, 3050 uint64_t value) 3051 { 3052 zap_phys_t *zap; 3053 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 3054 int rc; 3055 3056 zap = malloc(size); 3057 if (zap == NULL) 3058 return (ENOMEM); 3059 3060 rc = dnode_read(spa, dnode, 0, zap, size); 3061 if (rc == 0) { 3062 if (zap->zap_block_type == ZBT_MICRO) 3063 rc = mzap_rlookup((const mzap_phys_t *)zap, size, 3064 name, value); 3065 else 3066 rc = fzap_rlookup(spa, dnode, zap, name, value); 3067 } 3068 free(zap); 3069 return (rc); 3070 } 3071 3072 static int 3073 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result) 3074 { 3075 char name[256]; 3076 char component[256]; 3077 uint64_t dir_obj, parent_obj, child_dir_zapobj; 3078 dnode_phys_t child_dir_zap, snapnames_zap, dataset, dir, parent; 3079 dsl_dir_phys_t *dd; 3080 dsl_dataset_phys_t *ds; 3081 char *p; 3082 int len; 3083 boolean_t issnap = B_FALSE; 3084 3085 p = &name[sizeof(name) - 1]; 3086 *p = '\0'; 3087 3088 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3089 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3090 return (EIO); 3091 } 3092 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3093 dir_obj = ds->ds_dir_obj; 3094 if (ds->ds_snapnames_zapobj == 0) 3095 issnap = B_TRUE; 3096 3097 for (;;) { 3098 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0) 3099 return (EIO); 3100 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3101 3102 /* Actual loop condition. */ 3103 parent_obj = dd->dd_parent_obj; 3104 if (parent_obj == 0) 3105 break; 3106 3107 if (objset_get_dnode(spa, spa->spa_mos, parent_obj, 3108 &parent) != 0) 3109 return (EIO); 3110 dd = (dsl_dir_phys_t *)&parent.dn_bonus; 3111 if (issnap == B_TRUE) { 3112 /* 3113 * The dataset we are looking up is a snapshot 3114 * the dir_obj is the parent already, we don't want 3115 * the grandparent just yet. Reset to the parent. 3116 */ 3117 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3118 /* Lookup the dataset to get the snapname ZAP */ 3119 if (objset_get_dnode(spa, spa->spa_mos, 3120 dd->dd_head_dataset_obj, &dataset)) 3121 return (EIO); 3122 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3123 if (objset_get_dnode(spa, spa->spa_mos, 3124 ds->ds_snapnames_zapobj, &snapnames_zap) != 0) 3125 return (EIO); 3126 /* Get the name of the snapshot */ 3127 if (zap_rlookup(spa, &snapnames_zap, component, 3128 objnum) != 0) 3129 return (EIO); 3130 len = strlen(component); 3131 p -= len; 3132 memcpy(p, component, len); 3133 --p; 3134 *p = '@'; 3135 issnap = B_FALSE; 3136 continue; 3137 } 3138 3139 child_dir_zapobj = dd->dd_child_dir_zapobj; 3140 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3141 &child_dir_zap) != 0) 3142 return (EIO); 3143 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0) 3144 return (EIO); 3145 3146 len = strlen(component); 3147 p -= len; 3148 memcpy(p, component, len); 3149 --p; 3150 *p = '/'; 3151 3152 /* Actual loop iteration. */ 3153 dir_obj = parent_obj; 3154 } 3155 3156 if (*p != '\0') 3157 ++p; 3158 strcpy(result, p); 3159 3160 return (0); 3161 } 3162 3163 static int 3164 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum) 3165 { 3166 char element[256]; 3167 uint64_t dir_obj, child_dir_zapobj; 3168 dnode_phys_t child_dir_zap, snapnames_zap, dir, dataset; 3169 dsl_dir_phys_t *dd; 3170 dsl_dataset_phys_t *ds; 3171 const char *p, *q; 3172 boolean_t issnap = B_FALSE; 3173 3174 if (objset_get_dnode(spa, spa->spa_mos, 3175 DMU_POOL_DIRECTORY_OBJECT, &dir)) 3176 return (EIO); 3177 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj), 3178 1, &dir_obj)) 3179 return (EIO); 3180 3181 p = name; 3182 for (;;) { 3183 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) 3184 return (EIO); 3185 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3186 3187 while (*p == '/') 3188 p++; 3189 /* Actual loop condition #1. */ 3190 if (*p == '\0') 3191 break; 3192 3193 q = strchr(p, '/'); 3194 if (q) { 3195 memcpy(element, p, q - p); 3196 element[q - p] = '\0'; 3197 p = q + 1; 3198 } else { 3199 strcpy(element, p); 3200 p += strlen(p); 3201 } 3202 3203 if (issnap == B_TRUE) { 3204 if (objset_get_dnode(spa, spa->spa_mos, 3205 dd->dd_head_dataset_obj, &dataset)) 3206 return (EIO); 3207 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3208 if (objset_get_dnode(spa, spa->spa_mos, 3209 ds->ds_snapnames_zapobj, &snapnames_zap) != 0) 3210 return (EIO); 3211 /* Actual loop condition #2. */ 3212 if (zap_lookup(spa, &snapnames_zap, element, 3213 sizeof (dir_obj), 1, &dir_obj) != 0) 3214 return (ENOENT); 3215 *objnum = dir_obj; 3216 return (0); 3217 } else if ((q = strchr(element, '@')) != NULL) { 3218 issnap = B_TRUE; 3219 element[q - element] = '\0'; 3220 p = q + 1; 3221 } 3222 child_dir_zapobj = dd->dd_child_dir_zapobj; 3223 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3224 &child_dir_zap) != 0) 3225 return (EIO); 3226 3227 /* Actual loop condition #2. */ 3228 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj), 3229 1, &dir_obj) != 0) 3230 return (ENOENT); 3231 } 3232 3233 *objnum = dd->dd_head_dataset_obj; 3234 return (0); 3235 } 3236 3237 #ifndef BOOT2 3238 static int 3239 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/) 3240 { 3241 uint64_t dir_obj, child_dir_zapobj; 3242 dnode_phys_t child_dir_zap, dir, dataset; 3243 dsl_dataset_phys_t *ds; 3244 dsl_dir_phys_t *dd; 3245 3246 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3247 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3248 return (EIO); 3249 } 3250 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3251 dir_obj = ds->ds_dir_obj; 3252 3253 if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) { 3254 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj); 3255 return (EIO); 3256 } 3257 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3258 3259 child_dir_zapobj = dd->dd_child_dir_zapobj; 3260 if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3261 &child_dir_zap) != 0) { 3262 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj); 3263 return (EIO); 3264 } 3265 3266 return (zap_list(spa, &child_dir_zap) != 0); 3267 } 3268 3269 int 3270 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, 3271 int (*callback)(const char *, uint64_t)) 3272 { 3273 uint64_t dir_obj, child_dir_zapobj; 3274 dnode_phys_t child_dir_zap, dir, dataset; 3275 dsl_dataset_phys_t *ds; 3276 dsl_dir_phys_t *dd; 3277 zap_phys_t *zap; 3278 size_t size; 3279 int err; 3280 3281 err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset); 3282 if (err != 0) { 3283 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3284 return (err); 3285 } 3286 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3287 dir_obj = ds->ds_dir_obj; 3288 3289 err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir); 3290 if (err != 0) { 3291 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj); 3292 return (err); 3293 } 3294 dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3295 3296 child_dir_zapobj = dd->dd_child_dir_zapobj; 3297 err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj, 3298 &child_dir_zap); 3299 if (err != 0) { 3300 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj); 3301 return (err); 3302 } 3303 3304 size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT; 3305 zap = malloc(size); 3306 if (zap != NULL) { 3307 err = dnode_read(spa, &child_dir_zap, 0, zap, size); 3308 if (err != 0) 3309 goto done; 3310 3311 if (zap->zap_block_type == ZBT_MICRO) 3312 err = mzap_list((const mzap_phys_t *)zap, size, 3313 callback); 3314 else 3315 err = fzap_list(spa, &child_dir_zap, zap, callback); 3316 } else { 3317 err = ENOMEM; 3318 } 3319 done: 3320 free(zap); 3321 return (err); 3322 } 3323 #endif 3324 3325 /* 3326 * Find the object set given the object number of its dataset object 3327 * and return its details in *objset 3328 */ 3329 static int 3330 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset) 3331 { 3332 dnode_phys_t dataset; 3333 dsl_dataset_phys_t *ds; 3334 3335 if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) { 3336 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum); 3337 return (EIO); 3338 } 3339 3340 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus; 3341 if (zio_read(spa, &ds->ds_bp, objset)) { 3342 printf("ZFS: can't read object set for dataset %ju\n", 3343 (uintmax_t)objnum); 3344 return (EIO); 3345 } 3346 3347 return (0); 3348 } 3349 3350 /* 3351 * Find the object set pointed to by the BOOTFS property or the root 3352 * dataset if there is none and return its details in *objset 3353 */ 3354 static int 3355 zfs_get_root(const spa_t *spa, uint64_t *objid) 3356 { 3357 dnode_phys_t dir, propdir; 3358 uint64_t props, bootfs, root; 3359 3360 *objid = 0; 3361 3362 /* 3363 * Start with the MOS directory object. 3364 */ 3365 if (objset_get_dnode(spa, spa->spa_mos, 3366 DMU_POOL_DIRECTORY_OBJECT, &dir)) { 3367 printf("ZFS: can't read MOS object directory\n"); 3368 return (EIO); 3369 } 3370 3371 /* 3372 * Lookup the pool_props and see if we can find a bootfs. 3373 */ 3374 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, 3375 sizeof(props), 1, &props) == 0 && 3376 objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 && 3377 zap_lookup(spa, &propdir, "bootfs", 3378 sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) { 3379 *objid = bootfs; 3380 return (0); 3381 } 3382 /* 3383 * Lookup the root dataset directory 3384 */ 3385 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, 3386 sizeof(root), 1, &root) || 3387 objset_get_dnode(spa, spa->spa_mos, root, &dir)) { 3388 printf("ZFS: can't find root dsl_dir\n"); 3389 return (EIO); 3390 } 3391 3392 /* 3393 * Use the information from the dataset directory's bonus buffer 3394 * to find the dataset object and from that the object set itself. 3395 */ 3396 dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus; 3397 *objid = dd->dd_head_dataset_obj; 3398 return (0); 3399 } 3400 3401 static int 3402 zfs_mount_impl(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount) 3403 { 3404 3405 mount->spa = spa; 3406 3407 /* 3408 * Find the root object set if not explicitly provided 3409 */ 3410 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) { 3411 printf("ZFS: can't find root filesystem\n"); 3412 return (EIO); 3413 } 3414 3415 if (zfs_mount_dataset(spa, rootobj, &mount->objset)) { 3416 printf("ZFS: can't open root filesystem\n"); 3417 return (EIO); 3418 } 3419 3420 mount->rootobj = rootobj; 3421 3422 return (0); 3423 } 3424 3425 /* 3426 * callback function for feature name checks. 3427 */ 3428 static int 3429 check_feature(const char *name, uint64_t value) 3430 { 3431 int i; 3432 3433 if (value == 0) 3434 return (0); 3435 if (name[0] == '\0') 3436 return (0); 3437 3438 for (i = 0; features_for_read[i] != NULL; i++) { 3439 if (strcmp(name, features_for_read[i]) == 0) 3440 return (0); 3441 } 3442 printf("ZFS: unsupported feature: %s\n", name); 3443 return (EIO); 3444 } 3445 3446 /* 3447 * Checks whether the MOS features that are active are supported. 3448 */ 3449 static int 3450 check_mos_features(const spa_t *spa) 3451 { 3452 dnode_phys_t dir; 3453 zap_phys_t *zap; 3454 uint64_t objnum; 3455 size_t size; 3456 int rc; 3457 3458 if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY, 3459 &dir)) != 0) 3460 return (rc); 3461 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ, 3462 sizeof (objnum), 1, &objnum)) != 0) { 3463 /* 3464 * It is older pool without features. As we have already 3465 * tested the label, just return without raising the error. 3466 */ 3467 return (0); 3468 } 3469 3470 if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0) 3471 return (rc); 3472 3473 if (dir.dn_type != DMU_OTN_ZAP_METADATA) 3474 return (EIO); 3475 3476 size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT; 3477 zap = malloc(size); 3478 if (zap == NULL) 3479 return (ENOMEM); 3480 3481 if (dnode_read(spa, &dir, 0, zap, size)) { 3482 free(zap); 3483 return (EIO); 3484 } 3485 3486 if (zap->zap_block_type == ZBT_MICRO) 3487 rc = mzap_list((const mzap_phys_t *)zap, size, check_feature); 3488 else 3489 rc = fzap_list(spa, &dir, zap, check_feature); 3490 3491 free(zap); 3492 return (rc); 3493 } 3494 3495 static int 3496 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 3497 { 3498 dnode_phys_t dir; 3499 size_t size; 3500 int rc; 3501 char *nv; 3502 3503 *value = NULL; 3504 if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0) 3505 return (rc); 3506 if (dir.dn_type != DMU_OT_PACKED_NVLIST && 3507 dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) { 3508 return (EIO); 3509 } 3510 3511 if (dir.dn_bonuslen != sizeof (uint64_t)) 3512 return (EIO); 3513 3514 size = *(uint64_t *)DN_BONUS(&dir); 3515 nv = malloc(size); 3516 if (nv == NULL) 3517 return (ENOMEM); 3518 3519 rc = dnode_read(spa, &dir, 0, nv, size); 3520 if (rc != 0) { 3521 free(nv); 3522 nv = NULL; 3523 return (rc); 3524 } 3525 *value = nvlist_import(nv, size); 3526 free(nv); 3527 return (rc); 3528 } 3529 3530 static int 3531 zfs_spa_init(spa_t *spa) 3532 { 3533 struct uberblock checkpoint; 3534 dnode_phys_t dir; 3535 uint64_t config_object; 3536 nvlist_t *nvlist; 3537 int rc; 3538 3539 if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) { 3540 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name); 3541 return (EIO); 3542 } 3543 if (spa->spa_mos->os_type != DMU_OST_META) { 3544 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name); 3545 return (EIO); 3546 } 3547 3548 if (objset_get_dnode(spa, &spa->spa_mos_master, 3549 DMU_POOL_DIRECTORY_OBJECT, &dir)) { 3550 printf("ZFS: failed to read pool %s directory object\n", 3551 spa->spa_name); 3552 return (EIO); 3553 } 3554 /* this is allowed to fail, older pools do not have salt */ 3555 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1, 3556 sizeof (spa->spa_cksum_salt.zcs_bytes), 3557 spa->spa_cksum_salt.zcs_bytes); 3558 3559 rc = check_mos_features(spa); 3560 if (rc != 0) { 3561 printf("ZFS: pool %s is not supported\n", spa->spa_name); 3562 return (rc); 3563 } 3564 3565 rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG, 3566 sizeof (config_object), 1, &config_object); 3567 if (rc != 0) { 3568 printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG); 3569 return (EIO); 3570 } 3571 rc = load_nvlist(spa, config_object, &nvlist); 3572 if (rc != 0) { 3573 printf("ZFS: failed to load pool %s nvlist\n", spa->spa_name); 3574 return (rc); 3575 } 3576 3577 rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT, 3578 sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t), 3579 &checkpoint); 3580 if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) { 3581 memcpy(&spa->spa_uberblock_checkpoint, &checkpoint, 3582 sizeof(checkpoint)); 3583 if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp, 3584 &spa->spa_mos_checkpoint)) { 3585 printf("ZFS: can not read checkpoint data.\n"); 3586 return (EIO); 3587 } 3588 } 3589 3590 /* 3591 * Update vdevs from MOS config. Note, we do skip encoding bytes 3592 * here. See also vdev_label_read_config(). 3593 */ 3594 rc = vdev_init_from_nvlist(spa, nvlist); 3595 nvlist_destroy(nvlist); 3596 return (rc); 3597 } 3598 3599 static int 3600 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb) 3601 { 3602 3603 if (dn->dn_bonustype != DMU_OT_SA) { 3604 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus; 3605 3606 sb->st_mode = zp->zp_mode; 3607 sb->st_uid = zp->zp_uid; 3608 sb->st_gid = zp->zp_gid; 3609 sb->st_size = zp->zp_size; 3610 } else { 3611 sa_hdr_phys_t *sahdrp; 3612 int hdrsize; 3613 size_t size = 0; 3614 void *buf = NULL; 3615 3616 if (dn->dn_bonuslen != 0) 3617 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn); 3618 else { 3619 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) { 3620 blkptr_t *bp = DN_SPILL_BLKPTR(dn); 3621 int error; 3622 3623 size = BP_GET_LSIZE(bp); 3624 buf = malloc(size); 3625 if (buf == NULL) 3626 error = ENOMEM; 3627 else 3628 error = zio_read(spa, bp, buf); 3629 3630 if (error != 0) { 3631 free(buf); 3632 return (error); 3633 } 3634 sahdrp = buf; 3635 } else { 3636 return (EIO); 3637 } 3638 } 3639 hdrsize = SA_HDR_SIZE(sahdrp); 3640 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize + 3641 SA_MODE_OFFSET); 3642 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize + 3643 SA_UID_OFFSET); 3644 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize + 3645 SA_GID_OFFSET); 3646 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize + 3647 SA_SIZE_OFFSET); 3648 free(buf); 3649 } 3650 3651 return (0); 3652 } 3653 3654 static int 3655 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize) 3656 { 3657 int rc = 0; 3658 3659 if (dn->dn_bonustype == DMU_OT_SA) { 3660 sa_hdr_phys_t *sahdrp = NULL; 3661 size_t size = 0; 3662 void *buf = NULL; 3663 int hdrsize; 3664 char *p; 3665 3666 if (dn->dn_bonuslen != 0) { 3667 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn); 3668 } else { 3669 blkptr_t *bp; 3670 3671 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0) 3672 return (EIO); 3673 bp = DN_SPILL_BLKPTR(dn); 3674 3675 size = BP_GET_LSIZE(bp); 3676 buf = malloc(size); 3677 if (buf == NULL) 3678 rc = ENOMEM; 3679 else 3680 rc = zio_read(spa, bp, buf); 3681 if (rc != 0) { 3682 free(buf); 3683 return (rc); 3684 } 3685 sahdrp = buf; 3686 } 3687 hdrsize = SA_HDR_SIZE(sahdrp); 3688 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET); 3689 memcpy(path, p, psize); 3690 free(buf); 3691 return (0); 3692 } 3693 /* 3694 * Second test is purely to silence bogus compiler 3695 * warning about accessing past the end of dn_bonus. 3696 */ 3697 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen && 3698 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) { 3699 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize); 3700 } else { 3701 rc = dnode_read(spa, dn, 0, path, psize); 3702 } 3703 return (rc); 3704 } 3705 3706 struct obj_list { 3707 uint64_t objnum; 3708 STAILQ_ENTRY(obj_list) entry; 3709 }; 3710 3711 /* 3712 * Lookup a file and return its dnode. 3713 */ 3714 static int 3715 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode) 3716 { 3717 int rc; 3718 uint64_t objnum; 3719 const spa_t *spa; 3720 dnode_phys_t dn; 3721 const char *p, *q; 3722 char element[256]; 3723 char path[1024]; 3724 int symlinks_followed = 0; 3725 struct stat sb; 3726 struct obj_list *entry, *tentry; 3727 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache); 3728 3729 spa = mount->spa; 3730 if (mount->objset.os_type != DMU_OST_ZFS) { 3731 printf("ZFS: unexpected object set type %ju\n", 3732 (uintmax_t)mount->objset.os_type); 3733 return (EIO); 3734 } 3735 3736 if ((entry = malloc(sizeof(struct obj_list))) == NULL) 3737 return (ENOMEM); 3738 3739 /* 3740 * Get the root directory dnode. 3741 */ 3742 rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn); 3743 if (rc) { 3744 free(entry); 3745 return (rc); 3746 } 3747 3748 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum); 3749 if (rc) { 3750 free(entry); 3751 return (rc); 3752 } 3753 entry->objnum = objnum; 3754 STAILQ_INSERT_HEAD(&on_cache, entry, entry); 3755 3756 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3757 if (rc != 0) 3758 goto done; 3759 3760 p = upath; 3761 while (p && *p) { 3762 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3763 if (rc != 0) 3764 goto done; 3765 3766 while (*p == '/') 3767 p++; 3768 if (*p == '\0') 3769 break; 3770 q = p; 3771 while (*q != '\0' && *q != '/') 3772 q++; 3773 3774 /* skip dot */ 3775 if (p + 1 == q && p[0] == '.') { 3776 p++; 3777 continue; 3778 } 3779 /* double dot */ 3780 if (p + 2 == q && p[0] == '.' && p[1] == '.') { 3781 p += 2; 3782 if (STAILQ_FIRST(&on_cache) == 3783 STAILQ_LAST(&on_cache, obj_list, entry)) { 3784 rc = ENOENT; 3785 goto done; 3786 } 3787 entry = STAILQ_FIRST(&on_cache); 3788 STAILQ_REMOVE_HEAD(&on_cache, entry); 3789 free(entry); 3790 objnum = (STAILQ_FIRST(&on_cache))->objnum; 3791 continue; 3792 } 3793 if (q - p + 1 > sizeof(element)) { 3794 rc = ENAMETOOLONG; 3795 goto done; 3796 } 3797 memcpy(element, p, q - p); 3798 element[q - p] = 0; 3799 p = q; 3800 3801 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0) 3802 goto done; 3803 if (!S_ISDIR(sb.st_mode)) { 3804 rc = ENOTDIR; 3805 goto done; 3806 } 3807 3808 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum); 3809 if (rc) 3810 goto done; 3811 objnum = ZFS_DIRENT_OBJ(objnum); 3812 3813 if ((entry = malloc(sizeof(struct obj_list))) == NULL) { 3814 rc = ENOMEM; 3815 goto done; 3816 } 3817 entry->objnum = objnum; 3818 STAILQ_INSERT_HEAD(&on_cache, entry, entry); 3819 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn); 3820 if (rc) 3821 goto done; 3822 3823 /* 3824 * Check for symlink. 3825 */ 3826 rc = zfs_dnode_stat(spa, &dn, &sb); 3827 if (rc) 3828 goto done; 3829 if (S_ISLNK(sb.st_mode)) { 3830 if (symlinks_followed > 10) { 3831 rc = EMLINK; 3832 goto done; 3833 } 3834 symlinks_followed++; 3835 3836 /* 3837 * Read the link value and copy the tail of our 3838 * current path onto the end. 3839 */ 3840 if (sb.st_size + strlen(p) + 1 > sizeof(path)) { 3841 rc = ENAMETOOLONG; 3842 goto done; 3843 } 3844 strcpy(&path[sb.st_size], p); 3845 3846 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size); 3847 if (rc != 0) 3848 goto done; 3849 3850 /* 3851 * Restart with the new path, starting either at 3852 * the root or at the parent depending whether or 3853 * not the link is relative. 3854 */ 3855 p = path; 3856 if (*p == '/') { 3857 while (STAILQ_FIRST(&on_cache) != 3858 STAILQ_LAST(&on_cache, obj_list, entry)) { 3859 entry = STAILQ_FIRST(&on_cache); 3860 STAILQ_REMOVE_HEAD(&on_cache, entry); 3861 free(entry); 3862 } 3863 } else { 3864 entry = STAILQ_FIRST(&on_cache); 3865 STAILQ_REMOVE_HEAD(&on_cache, entry); 3866 free(entry); 3867 } 3868 objnum = (STAILQ_FIRST(&on_cache))->objnum; 3869 } 3870 } 3871 3872 *dnode = dn; 3873 done: 3874 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry) 3875 free(entry); 3876 return (rc); 3877 } 3878 3879 /* 3880 * Return either a cached copy of the bootenv, or read each of the vdev children 3881 * looking for the bootenv. Cache what's found and return the results. Returns 0 3882 * when benvp is filled in, and some errno when not. 3883 */ 3884 static int 3885 zfs_get_bootenv_spa(spa_t *spa, nvlist_t **benvp) 3886 { 3887 vdev_t *vd; 3888 nvlist_t *benv = NULL; 3889 3890 if (spa->spa_bootenv == NULL) { 3891 STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children, 3892 v_childlink) { 3893 benv = vdev_read_bootenv(vd); 3894 3895 if (benv != NULL) 3896 break; 3897 } 3898 spa->spa_bootenv = benv; 3899 } 3900 benv = spa->spa_bootenv; 3901 3902 if (benv == NULL) 3903 return (ENOENT); 3904 3905 *benvp = benv; 3906 return (0); 3907 } 3908 3909 /* 3910 * Store nvlist to pool label bootenv area. Also updates cached pointer in spa. 3911 */ 3912 static int 3913 zfs_set_bootenv_spa(spa_t *spa, nvlist_t *benv) 3914 { 3915 vdev_t *vd; 3916 3917 STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children, v_childlink) { 3918 vdev_write_bootenv(vd, benv); 3919 } 3920 3921 spa->spa_bootenv = benv; 3922 return (0); 3923 } 3924 3925 /* 3926 * Get bootonce value by key. The bootonce <key, value> pair is removed from the 3927 * bootenv nvlist and the remaining nvlist is committed back to disk. This process 3928 * the bootonce flag since we've reached the point in the boot that we've 'used' 3929 * the BE. For chained boot scenarios, we may reach this point multiple times (but 3930 * only remove it and return 0 the first time). 3931 */ 3932 static int 3933 zfs_get_bootonce_spa(spa_t *spa, const char *key, char *buf, size_t size) 3934 { 3935 nvlist_t *benv; 3936 char *result = NULL; 3937 int result_size, rv; 3938 3939 if ((rv = zfs_get_bootenv_spa(spa, &benv)) != 0) 3940 return (rv); 3941 3942 if ((rv = nvlist_find(benv, key, DATA_TYPE_STRING, NULL, 3943 &result, &result_size)) == 0) { 3944 if (result_size == 0) { 3945 /* ignore empty string */ 3946 rv = ENOENT; 3947 } else if (buf != NULL) { 3948 size = MIN((size_t)result_size + 1, size); 3949 strlcpy(buf, result, size); 3950 } 3951 (void)nvlist_remove(benv, key, DATA_TYPE_STRING); 3952 (void)zfs_set_bootenv_spa(spa, benv); 3953 } 3954 3955 return (rv); 3956 } 3957