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