2 * Copyright (c) 2007 Doug Rabson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
31 * Stand-alone ZFS file reader.
34 #include <sys/endian.h>
36 #include <sys/stdint.h>
48 static struct zfsmount zfsmount __unused;
51 * The indirect_child_t represents the vdev that we will read from, when we
52 * need to read all copies of the data (e.g. for scrub or reconstruction).
53 * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
54 * ic_vdev is the same as is_vdev. However, for mirror top-level vdevs,
55 * ic_vdev is a child of the mirror.
57 typedef struct indirect_child {
63 * The indirect_split_t represents one mapped segment of an i/o to the
64 * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
65 * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
66 * For split blocks, there will be several of these.
68 typedef struct indirect_split {
69 list_node_t is_node; /* link on iv_splits */
72 * is_split_offset is the offset into the i/o.
73 * This is the sum of the previous splits' is_size's.
75 uint64_t is_split_offset;
77 vdev_t *is_vdev; /* top-level vdev */
78 uint64_t is_target_offset; /* offset on is_vdev */
80 int is_children; /* number of entries in is_child[] */
83 * is_good_child is the child that we are currently using to
84 * attempt reconstruction.
88 indirect_child_t is_child[1]; /* variable-length */
92 * The indirect_vsd_t is associated with each i/o to the indirect vdev.
93 * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
95 typedef struct indirect_vsd {
96 boolean_t iv_split_block;
97 boolean_t iv_reconstruct;
99 list_t iv_splits; /* list of indirect_split_t's */
103 * List of all vdevs, chained through v_alllink.
105 static vdev_list_t zfs_vdevs;
108 * List of ZFS features supported for read
110 static const char *features_for_read[] = {
111 "org.illumos:lz4_compress",
112 "com.delphix:hole_birth",
113 "com.delphix:extensible_dataset",
114 "com.delphix:embedded_data",
115 "org.open-zfs:large_blocks",
116 "org.illumos:sha512",
118 "org.zfsonlinux:large_dnode",
119 "com.joyent:multi_vdev_crash_dump",
120 "com.delphix:spacemap_histogram",
121 "com.delphix:zpool_checkpoint",
122 "com.delphix:spacemap_v2",
123 "com.datto:encryption",
124 "org.zfsonlinux:allocation_classes",
125 "com.datto:resilver_defer",
126 "com.delphix:device_removal",
127 "com.delphix:obsolete_counts",
128 "com.intel:allocation_classes",
133 * List of all pools, chained through spa_link.
135 static spa_list_t zfs_pools;
137 static const dnode_phys_t *dnode_cache_obj;
138 static uint64_t dnode_cache_bn;
139 static char *dnode_cache_buf;
140 static char *zap_scratch;
141 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
143 #define TEMP_SIZE (1024 * 1024)
145 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
146 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
147 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
148 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
149 const char *name, uint64_t integer_size, uint64_t num_integers,
151 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
153 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
155 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
157 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
158 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
160 vdev_indirect_mapping_entry_phys_t *
161 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
162 uint64_t, uint64_t *);
167 STAILQ_INIT(&zfs_vdevs);
168 STAILQ_INIT(&zfs_pools);
170 zfs_temp_buf = malloc(TEMP_SIZE);
171 zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
172 zfs_temp_ptr = zfs_temp_buf;
173 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
174 zap_scratch = malloc(SPA_MAXBLOCKSIZE);
180 zfs_alloc(size_t size)
184 if (zfs_temp_ptr + size > zfs_temp_end) {
185 panic("ZFS: out of temporary buffer space");
188 zfs_temp_ptr += size;
194 zfs_free(void *ptr, size_t size)
197 zfs_temp_ptr -= size;
198 if (zfs_temp_ptr != ptr) {
199 panic("ZFS: zfs_alloc()/zfs_free() mismatch");
204 xdr_int(const unsigned char **xdr, int *ip)
212 xdr_u_int(const unsigned char **xdr, u_int *ip)
220 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
226 *lp = (((uint64_t) hi) << 32) | lo;
231 nvlist_find(const unsigned char *nvlist, const char *name, int type,
232 int *elementsp, void *valuep)
234 const unsigned char *p, *pair;
236 int encoded_size, decoded_size;
243 xdr_int(&p, &encoded_size);
244 xdr_int(&p, &decoded_size);
245 while (encoded_size && decoded_size) {
246 int namelen, pairtype, elements;
247 const char *pairname;
249 xdr_int(&p, &namelen);
250 pairname = (const char*) p;
251 p += roundup(namelen, 4);
252 xdr_int(&p, &pairtype);
254 if (!memcmp(name, pairname, namelen) && type == pairtype) {
255 xdr_int(&p, &elements);
257 *elementsp = elements;
258 if (type == DATA_TYPE_UINT64) {
259 xdr_uint64_t(&p, (uint64_t *) valuep);
261 } else if (type == DATA_TYPE_STRING) {
264 (*(const char**) valuep) = (const char*) p;
266 } else if (type == DATA_TYPE_NVLIST
267 || type == DATA_TYPE_NVLIST_ARRAY) {
268 (*(const unsigned char**) valuep) =
269 (const unsigned char*) p;
276 * Not the pair we are looking for, skip to the next one.
278 p = pair + encoded_size;
282 xdr_int(&p, &encoded_size);
283 xdr_int(&p, &decoded_size);
290 nvlist_check_features_for_read(const unsigned char *nvlist)
292 const unsigned char *p, *pair;
294 int encoded_size, decoded_size;
304 xdr_int(&p, &encoded_size);
305 xdr_int(&p, &decoded_size);
306 while (encoded_size && decoded_size) {
307 int namelen, pairtype;
308 const char *pairname;
313 xdr_int(&p, &namelen);
314 pairname = (const char*) p;
315 p += roundup(namelen, 4);
316 xdr_int(&p, &pairtype);
318 for (i = 0; features_for_read[i] != NULL; i++) {
319 if (!memcmp(pairname, features_for_read[i], namelen)) {
326 printf("ZFS: unsupported feature: %s\n", pairname);
330 p = pair + encoded_size;
333 xdr_int(&p, &encoded_size);
334 xdr_int(&p, &decoded_size);
341 * Return the next nvlist in an nvlist array.
343 static const unsigned char *
344 nvlist_next(const unsigned char *nvlist)
346 const unsigned char *p, *pair;
348 int encoded_size, decoded_size;
355 xdr_int(&p, &encoded_size);
356 xdr_int(&p, &decoded_size);
357 while (encoded_size && decoded_size) {
358 p = pair + encoded_size;
361 xdr_int(&p, &encoded_size);
362 xdr_int(&p, &decoded_size);
370 static const unsigned char *
371 nvlist_print(const unsigned char *nvlist, unsigned int indent)
373 static const char* typenames[] = {
384 "DATA_TYPE_BYTE_ARRAY",
385 "DATA_TYPE_INT16_ARRAY",
386 "DATA_TYPE_UINT16_ARRAY",
387 "DATA_TYPE_INT32_ARRAY",
388 "DATA_TYPE_UINT32_ARRAY",
389 "DATA_TYPE_INT64_ARRAY",
390 "DATA_TYPE_UINT64_ARRAY",
391 "DATA_TYPE_STRING_ARRAY",
394 "DATA_TYPE_NVLIST_ARRAY",
395 "DATA_TYPE_BOOLEAN_VALUE",
398 "DATA_TYPE_BOOLEAN_ARRAY",
399 "DATA_TYPE_INT8_ARRAY",
400 "DATA_TYPE_UINT8_ARRAY"
404 const unsigned char *p, *pair;
406 int encoded_size, decoded_size;
413 xdr_int(&p, &encoded_size);
414 xdr_int(&p, &decoded_size);
415 while (encoded_size && decoded_size) {
416 int namelen, pairtype, elements;
417 const char *pairname;
419 xdr_int(&p, &namelen);
420 pairname = (const char*) p;
421 p += roundup(namelen, 4);
422 xdr_int(&p, &pairtype);
424 for (i = 0; i < indent; i++)
426 printf("%s %s", typenames[pairtype], pairname);
428 xdr_int(&p, &elements);
430 case DATA_TYPE_UINT64: {
432 xdr_uint64_t(&p, &val);
433 printf(" = 0x%jx\n", (uintmax_t)val);
437 case DATA_TYPE_STRING: {
440 printf(" = \"%s\"\n", p);
444 case DATA_TYPE_NVLIST:
446 nvlist_print(p, indent + 1);
449 case DATA_TYPE_NVLIST_ARRAY:
450 for (j = 0; j < elements; j++) {
452 p = nvlist_print(p, indent + 1);
453 if (j != elements - 1) {
454 for (i = 0; i < indent; i++)
456 printf("%s %s", typenames[pairtype], pairname);
465 p = pair + encoded_size;
468 xdr_int(&p, &encoded_size);
469 xdr_int(&p, &decoded_size);
478 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
479 off_t offset, size_t size)
484 if (!vdev->v_phys_read)
488 psize = BP_GET_PSIZE(bp);
493 /*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
494 rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
498 return (zio_checksum_verify(vdev->spa, bp, buf));
503 typedef struct remap_segment {
507 uint64_t rs_split_offset;
511 static remap_segment_t *
512 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
514 remap_segment_t *rs = malloc(sizeof (remap_segment_t));
518 rs->rs_offset = offset;
519 rs->rs_asize = asize;
520 rs->rs_split_offset = split_offset;
526 vdev_indirect_mapping_t *
527 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
528 uint64_t mapping_object)
530 vdev_indirect_mapping_t *vim;
531 vdev_indirect_mapping_phys_t *vim_phys;
534 vim = calloc(1, sizeof (*vim));
538 vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
539 if (vim->vim_dn == NULL) {
544 rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
552 vim->vim_phys = malloc(sizeof (*vim->vim_phys));
553 if (vim->vim_phys == NULL) {
559 vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
560 *vim->vim_phys = *vim_phys;
562 vim->vim_objset = os;
563 vim->vim_object = mapping_object;
564 vim->vim_entries = NULL;
566 vim->vim_havecounts =
567 (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
572 * Compare an offset with an indirect mapping entry; there are three
573 * possible scenarios:
575 * 1. The offset is "less than" the mapping entry; meaning the
576 * offset is less than the source offset of the mapping entry. In
577 * this case, there is no overlap between the offset and the
578 * mapping entry and -1 will be returned.
580 * 2. The offset is "greater than" the mapping entry; meaning the
581 * offset is greater than the mapping entry's source offset plus
582 * the entry's size. In this case, there is no overlap between
583 * the offset and the mapping entry and 1 will be returned.
585 * NOTE: If the offset is actually equal to the entry's offset
586 * plus size, this is considered to be "greater" than the entry,
587 * and this case applies (i.e. 1 will be returned). Thus, the
588 * entry's "range" can be considered to be inclusive at its
589 * start, but exclusive at its end: e.g. [src, src + size).
591 * 3. The last case to consider is if the offset actually falls
592 * within the mapping entry's range. If this is the case, the
593 * offset is considered to be "equal to" the mapping entry and
594 * 0 will be returned.
596 * NOTE: If the offset is equal to the entry's source offset,
597 * this case applies and 0 will be returned. If the offset is
598 * equal to the entry's source plus its size, this case does
599 * *not* apply (see "NOTE" above for scenario 2), and 1 will be
603 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
605 const uint64_t *key = v_key;
606 const vdev_indirect_mapping_entry_phys_t *array_elem =
608 uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
610 if (*key < src_offset) {
612 } else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
620 * Return array entry.
622 static vdev_indirect_mapping_entry_phys_t *
623 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
629 if (vim->vim_phys->vimp_num_entries == 0)
632 if (vim->vim_entries == NULL) {
635 bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
636 size = vim->vim_phys->vimp_num_entries *
637 sizeof (*vim->vim_entries);
639 size = bsize / sizeof (*vim->vim_entries);
640 size *= sizeof (*vim->vim_entries);
642 vim->vim_entries = malloc(size);
643 if (vim->vim_entries == NULL)
645 vim->vim_num_entries = size / sizeof (*vim->vim_entries);
646 offset = index * sizeof (*vim->vim_entries);
649 /* We have data in vim_entries */
651 if (index >= vim->vim_entry_offset &&
652 index <= vim->vim_entry_offset + vim->vim_num_entries) {
653 index -= vim->vim_entry_offset;
654 return (&vim->vim_entries[index]);
656 offset = index * sizeof (*vim->vim_entries);
659 vim->vim_entry_offset = index;
660 size = vim->vim_num_entries * sizeof (*vim->vim_entries);
661 rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
664 /* Read error, invalidate vim_entries. */
665 free(vim->vim_entries);
666 vim->vim_entries = NULL;
669 index -= vim->vim_entry_offset;
670 return (&vim->vim_entries[index]);
674 * Returns the mapping entry for the given offset.
676 * It's possible that the given offset will not be in the mapping table
677 * (i.e. no mapping entries contain this offset), in which case, the
678 * return value value depends on the "next_if_missing" parameter.
680 * If the offset is not found in the table and "next_if_missing" is
681 * B_FALSE, then NULL will always be returned. The behavior is intended
682 * to allow consumers to get the entry corresponding to the offset
683 * parameter, iff the offset overlaps with an entry in the table.
685 * If the offset is not found in the table and "next_if_missing" is
686 * B_TRUE, then the entry nearest to the given offset will be returned,
687 * such that the entry's source offset is greater than the offset
688 * passed in (i.e. the "next" mapping entry in the table is returned, if
689 * the offset is missing from the table). If there are no entries whose
690 * source offset is greater than the passed in offset, NULL is returned.
692 static vdev_indirect_mapping_entry_phys_t *
693 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
696 ASSERT(vim->vim_phys->vimp_num_entries > 0);
698 vdev_indirect_mapping_entry_phys_t *entry;
700 uint64_t last = vim->vim_phys->vimp_num_entries - 1;
704 * We don't define these inside of the while loop because we use
705 * their value in the case that offset isn't in the mapping.
710 while (last >= base) {
711 mid = base + ((last - base) >> 1);
713 entry = vdev_indirect_mapping_entry(vim, mid);
716 result = dva_mapping_overlap_compare(&offset, entry);
720 } else if (result < 0) {
730 * Given an indirect vdev and an extent on that vdev, it duplicates the
731 * physical entries of the indirect mapping that correspond to the extent
732 * to a new array and returns a pointer to it. In addition, copied_entries
733 * is populated with the number of mapping entries that were duplicated.
735 * Finally, since we are doing an allocation, it is up to the caller to
736 * free the array allocated in this function.
738 vdev_indirect_mapping_entry_phys_t *
739 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
740 uint64_t asize, uint64_t *copied_entries)
742 vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
743 vdev_indirect_mapping_t *vim = vd->v_mapping;
744 uint64_t entries = 0;
746 vdev_indirect_mapping_entry_phys_t *first_mapping =
747 vdev_indirect_mapping_entry_for_offset(vim, offset);
748 ASSERT3P(first_mapping, !=, NULL);
750 vdev_indirect_mapping_entry_phys_t *m = first_mapping;
752 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
753 uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
754 uint64_t inner_size = MIN(asize, size - inner_offset);
756 offset += inner_size;
762 size_t copy_length = entries * sizeof (*first_mapping);
763 duplicate_mappings = malloc(copy_length);
764 if (duplicate_mappings != NULL)
765 bcopy(first_mapping, duplicate_mappings, copy_length);
769 *copied_entries = entries;
771 return (duplicate_mappings);
775 vdev_lookup_top(spa_t *spa, uint64_t vdev)
779 STAILQ_FOREACH(rvd, &spa->spa_vdevs, v_childlink)
780 if (rvd->v_id == vdev)
787 * This is a callback for vdev_indirect_remap() which allocates an
788 * indirect_split_t for each split segment and adds it to iv_splits.
791 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
792 uint64_t size, void *arg)
796 indirect_vsd_t *iv = zio->io_vsd;
798 if (vd->v_read == vdev_indirect_read)
801 if (vd->v_read == vdev_mirror_read)
804 indirect_split_t *is =
805 malloc(offsetof(indirect_split_t, is_child[n]));
807 zio->io_error = ENOMEM;
810 bzero(is, offsetof(indirect_split_t, is_child[n]));
814 is->is_split_offset = split_offset;
815 is->is_target_offset = offset;
819 * Note that we only consider multiple copies of the data for
820 * *mirror* vdevs. We don't for "replacing" or "spare" vdevs, even
821 * though they use the same ops as mirror, because there's only one
822 * "good" copy under the replacing/spare.
824 if (vd->v_read == vdev_mirror_read) {
828 STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
829 is->is_child[i++].ic_vdev = kid;
832 is->is_child[0].ic_vdev = vd;
835 list_insert_tail(&iv->iv_splits, is);
839 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
842 spa_t *spa = vd->spa;
846 list_create(&stack, sizeof (remap_segment_t),
847 offsetof(remap_segment_t, rs_node));
849 rs = rs_alloc(vd, offset, asize, 0);
851 printf("vdev_indirect_remap: out of memory.\n");
852 zio->io_error = ENOMEM;
854 for ( ; rs != NULL; rs = list_remove_head(&stack)) {
855 vdev_t *v = rs->rs_vd;
856 uint64_t num_entries = 0;
857 /* vdev_indirect_mapping_t *vim = v->v_mapping; */
858 vdev_indirect_mapping_entry_phys_t *mapping =
859 vdev_indirect_mapping_duplicate_adjacent_entries(v,
860 rs->rs_offset, rs->rs_asize, &num_entries);
862 if (num_entries == 0)
863 zio->io_error = ENOMEM;
865 for (uint64_t i = 0; i < num_entries; i++) {
866 vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
867 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
868 uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
869 uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
870 uint64_t inner_offset = rs->rs_offset -
871 DVA_MAPPING_GET_SRC_OFFSET(m);
872 uint64_t inner_size =
873 MIN(rs->rs_asize, size - inner_offset);
874 vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
876 if (dst_v->v_read == vdev_indirect_read) {
879 o = rs_alloc(dst_v, dst_offset + inner_offset,
880 inner_size, rs->rs_split_offset);
882 printf("vdev_indirect_remap: "
884 zio->io_error = ENOMEM;
888 list_insert_head(&stack, o);
890 vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
891 dst_offset + inner_offset,
895 * vdev_indirect_gather_splits can have memory
896 * allocation error, we can not recover from it.
898 if (zio->io_error != 0)
901 rs->rs_offset += inner_size;
902 rs->rs_asize -= inner_size;
903 rs->rs_split_offset += inner_size;
908 if (zio->io_error != 0)
912 list_destroy(&stack);
916 vdev_indirect_map_free(zio_t *zio)
918 indirect_vsd_t *iv = zio->io_vsd;
919 indirect_split_t *is;
921 while ((is = list_head(&iv->iv_splits)) != NULL) {
922 for (int c = 0; c < is->is_children; c++) {
923 indirect_child_t *ic = &is->is_child[c];
926 list_remove(&iv->iv_splits, is);
933 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
934 off_t offset, size_t bytes)
937 spa_t *spa = vdev->spa;
938 indirect_vsd_t *iv = malloc(sizeof (*iv));
939 indirect_split_t *first;
944 bzero(iv, sizeof (*iv));
946 list_create(&iv->iv_splits,
947 sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
950 zio.io_bp = (blkptr_t *)bp;
953 zio.io_offset = offset;
957 if (vdev->v_mapping == NULL) {
958 vdev_indirect_config_t *vic;
960 vic = &vdev->vdev_indirect_config;
961 vdev->v_mapping = vdev_indirect_mapping_open(spa,
962 &spa->spa_mos, vic->vic_mapping_object);
965 vdev_indirect_remap(vdev, offset, bytes, &zio);
966 if (zio.io_error != 0)
967 return (zio.io_error);
969 first = list_head(&iv->iv_splits);
970 if (first->is_size == zio.io_size) {
972 * This is not a split block; we are pointing to the entire
973 * data, which will checksum the same as the original data.
974 * Pass the BP down so that the child i/o can verify the
975 * checksum, and try a different location if available
976 * (e.g. on a mirror).
978 * While this special case could be handled the same as the
979 * general (split block) case, doing it this way ensures
980 * that the vast majority of blocks on indirect vdevs
981 * (which are not split) are handled identically to blocks
982 * on non-indirect vdevs. This allows us to be less strict
983 * about performance in the general (but rare) case.
985 rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
986 zio.io_data, first->is_target_offset, bytes);
988 iv->iv_split_block = B_TRUE;
990 * Read one copy of each split segment, from the
991 * top-level vdev. Since we don't know the
992 * checksum of each split individually, the child
993 * zio can't ensure that we get the right data.
994 * E.g. if it's a mirror, it will just read from a
995 * random (healthy) leaf vdev. We have to verify
996 * the checksum in vdev_indirect_io_done().
998 for (indirect_split_t *is = list_head(&iv->iv_splits);
999 is != NULL; is = list_next(&iv->iv_splits, is)) {
1000 char *ptr = zio.io_data;
1002 rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
1003 ptr + is->is_split_offset, is->is_target_offset,
1006 if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
1012 vdev_indirect_map_free(&zio);
1020 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1021 off_t offset, size_t bytes)
1024 return (vdev_read_phys(vdev, bp, buf,
1025 offset + VDEV_LABEL_START_SIZE, bytes));
1030 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1031 off_t offset, size_t bytes)
1037 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1038 if (kid->v_state != VDEV_STATE_HEALTHY)
1040 rc = kid->v_read(kid, bp, buf, offset, bytes);
1049 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1050 off_t offset, size_t bytes)
1055 * Here we should have two kids:
1056 * First one which is the one we are replacing and we can trust
1057 * only this one to have valid data, but it might not be present.
1058 * Second one is that one we are replacing with. It is most likely
1059 * healthy, but we can't trust it has needed data, so we won't use it.
1061 kid = STAILQ_FIRST(&vdev->v_children);
1064 if (kid->v_state != VDEV_STATE_HEALTHY)
1066 return (kid->v_read(kid, bp, buf, offset, bytes));
1070 vdev_find(uint64_t guid)
1074 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
1075 if (vdev->v_guid == guid)
1082 vdev_create(uint64_t guid, vdev_read_t *_read)
1085 vdev_indirect_config_t *vic;
1087 vdev = malloc(sizeof(vdev_t));
1088 memset(vdev, 0, sizeof(vdev_t));
1089 STAILQ_INIT(&vdev->v_children);
1090 vdev->v_guid = guid;
1091 vdev->v_state = VDEV_STATE_OFFLINE;
1092 vdev->v_read = _read;
1094 vic = &vdev->vdev_indirect_config;
1095 vic->vic_prev_indirect_vdev = UINT64_MAX;
1096 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
1102 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
1103 vdev_t **vdevp, int is_newer)
1106 uint64_t guid, id, ashift, asize, nparity;
1110 const unsigned char *kids;
1111 int nkids, i, is_new;
1112 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
1115 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1117 || nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id)
1118 || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1120 printf("ZFS: can't find vdev details\n");
1124 if (strcmp(type, VDEV_TYPE_MIRROR)
1125 && strcmp(type, VDEV_TYPE_DISK)
1127 && strcmp(type, VDEV_TYPE_FILE)
1129 && strcmp(type, VDEV_TYPE_RAIDZ)
1130 && strcmp(type, VDEV_TYPE_INDIRECT)
1131 && strcmp(type, VDEV_TYPE_REPLACING)) {
1132 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
1136 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
1139 nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
1141 nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
1143 nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
1145 nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
1147 nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
1149 nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
1152 vdev = vdev_find(guid);
1156 if (!strcmp(type, VDEV_TYPE_MIRROR))
1157 vdev = vdev_create(guid, vdev_mirror_read);
1158 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
1159 vdev = vdev_create(guid, vdev_raidz_read);
1160 else if (!strcmp(type, VDEV_TYPE_REPLACING))
1161 vdev = vdev_create(guid, vdev_replacing_read);
1162 else if (!strcmp(type, VDEV_TYPE_INDIRECT)) {
1163 vdev_indirect_config_t *vic;
1165 vdev = vdev_create(guid, vdev_indirect_read);
1166 vdev->v_state = VDEV_STATE_HEALTHY;
1167 vic = &vdev->vdev_indirect_config;
1170 ZPOOL_CONFIG_INDIRECT_OBJECT, DATA_TYPE_UINT64,
1171 NULL, &vic->vic_mapping_object);
1173 ZPOOL_CONFIG_INDIRECT_BIRTHS, DATA_TYPE_UINT64,
1174 NULL, &vic->vic_births_object);
1176 ZPOOL_CONFIG_PREV_INDIRECT_VDEV, DATA_TYPE_UINT64,
1177 NULL, &vic->vic_prev_indirect_vdev);
1179 vdev = vdev_create(guid, vdev_disk_read);
1182 vdev->v_top = pvdev != NULL ? pvdev : vdev;
1183 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
1184 DATA_TYPE_UINT64, NULL, &ashift) == 0) {
1185 vdev->v_ashift = ashift;
1189 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1190 DATA_TYPE_UINT64, NULL, &asize) == 0) {
1191 vdev->v_psize = asize +
1192 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1194 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
1195 DATA_TYPE_UINT64, NULL, &nparity) == 0) {
1196 vdev->v_nparity = nparity;
1198 vdev->v_nparity = 0;
1200 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1201 DATA_TYPE_STRING, NULL, &path) == 0) {
1202 if (strncmp(path, "/dev/", 5) == 0)
1204 vdev->v_name = strdup(path);
1208 if (!strcmp(type, "raidz")) {
1209 if (vdev->v_nparity < 1 ||
1210 vdev->v_nparity > 3) {
1211 printf("ZFS: can only boot from disk, "
1212 "mirror, raidz1, raidz2 and raidz3 "
1216 asprintf(&name, "%s%d-%jd", type,
1217 vdev->v_nparity, id);
1219 asprintf(&name, "%s-%jd", type, id);
1223 vdev->v_name = name;
1225 vdev->v_islog = is_log == 1;
1230 if (is_new || is_newer) {
1232 * This is either new vdev or we've already seen this vdev,
1233 * but from an older vdev label, so let's refresh its state
1234 * from the newer label.
1237 vdev->v_state = VDEV_STATE_OFFLINE;
1238 else if (is_removed)
1239 vdev->v_state = VDEV_STATE_REMOVED;
1240 else if (is_faulted)
1241 vdev->v_state = VDEV_STATE_FAULTED;
1242 else if (is_degraded)
1243 vdev->v_state = VDEV_STATE_DEGRADED;
1244 else if (isnt_present)
1245 vdev->v_state = VDEV_STATE_CANT_OPEN;
1248 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1251 * Its ok if we don't have any kids.
1254 vdev->v_nchildren = nkids;
1255 for (i = 0; i < nkids; i++) {
1256 rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
1260 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
1262 kids = nvlist_next(kids);
1265 vdev->v_nchildren = 0;
1274 vdev_set_state(vdev_t *vdev)
1281 * A mirror or raidz is healthy if all its kids are healthy. A
1282 * mirror is degraded if any of its kids is healthy; a raidz
1283 * is degraded if at most nparity kids are offline.
1285 if (STAILQ_FIRST(&vdev->v_children)) {
1288 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1289 if (kid->v_state == VDEV_STATE_HEALTHY)
1294 if (bad_kids == 0) {
1295 vdev->v_state = VDEV_STATE_HEALTHY;
1297 if (vdev->v_read == vdev_mirror_read) {
1299 vdev->v_state = VDEV_STATE_DEGRADED;
1301 vdev->v_state = VDEV_STATE_OFFLINE;
1303 } else if (vdev->v_read == vdev_raidz_read) {
1304 if (bad_kids > vdev->v_nparity) {
1305 vdev->v_state = VDEV_STATE_OFFLINE;
1307 vdev->v_state = VDEV_STATE_DEGRADED;
1315 spa_find_by_guid(uint64_t guid)
1319 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1320 if (spa->spa_guid == guid)
1327 spa_find_by_name(const char *name)
1331 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1332 if (!strcmp(spa->spa_name, name))
1340 spa_get_primary(void)
1343 return (STAILQ_FIRST(&zfs_pools));
1347 spa_get_primary_vdev(const spa_t *spa)
1353 spa = spa_get_primary();
1356 vdev = STAILQ_FIRST(&spa->spa_vdevs);
1359 for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
1360 kid = STAILQ_FIRST(&vdev->v_children))
1367 spa_create(uint64_t guid, const char *name)
1371 if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1373 if ((spa->spa_name = strdup(name)) == NULL) {
1377 STAILQ_INIT(&spa->spa_vdevs);
1378 spa->spa_guid = guid;
1379 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1385 state_name(vdev_state_t state)
1387 static const char* names[] = {
1397 return names[state];
1402 #define pager_printf printf
1407 pager_printf(const char *fmt, ...)
1412 va_start(args, fmt);
1413 vsprintf(line, fmt, args);
1416 return (pager_output(line));
1421 #define STATUS_FORMAT " %s %s\n"
1424 print_state(int indent, const char *name, vdev_state_t state)
1430 for (i = 0; i < indent; i++)
1434 return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1438 vdev_status(vdev_t *vdev, int indent)
1443 if (vdev->v_islog) {
1444 (void)pager_output(" logs\n");
1448 ret = print_state(indent, vdev->v_name, vdev->v_state);
1452 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1453 ret = vdev_status(kid, indent + 1);
1461 spa_status(spa_t *spa)
1463 static char bootfs[ZFS_MAXNAMELEN];
1466 int good_kids, bad_kids, degraded_kids, ret;
1469 ret = pager_printf(" pool: %s\n", spa->spa_name);
1473 if (zfs_get_root(spa, &rootid) == 0 &&
1474 zfs_rlookup(spa, rootid, bootfs) == 0) {
1475 if (bootfs[0] == '\0')
1476 ret = pager_printf("bootfs: %s\n", spa->spa_name);
1478 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1483 ret = pager_printf("config:\n\n");
1486 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1493 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1494 if (vdev->v_state == VDEV_STATE_HEALTHY)
1496 else if (vdev->v_state == VDEV_STATE_DEGRADED)
1502 state = VDEV_STATE_CLOSED;
1503 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1504 state = VDEV_STATE_HEALTHY;
1505 else if ((good_kids + degraded_kids) > 0)
1506 state = VDEV_STATE_DEGRADED;
1508 ret = print_state(0, spa->spa_name, state);
1511 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1512 ret = vdev_status(vdev, 1);
1520 spa_all_status(void)
1523 int first = 1, ret = 0;
1525 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1527 ret = pager_printf("\n");
1532 ret = spa_status(spa);
1540 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1542 uint64_t label_offset;
1544 if (l < VDEV_LABELS / 2)
1547 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1549 return (offset + l * sizeof (vdev_label_t) + label_offset);
1553 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1555 unsigned int seq1 = 0;
1556 unsigned int seq2 = 0;
1557 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1562 cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1566 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1567 seq1 = MMP_SEQ(ub1);
1569 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1570 seq2 = MMP_SEQ(ub2);
1572 return (AVL_CMP(seq1, seq2));
1576 uberblock_verify(uberblock_t *ub)
1578 if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1579 byteswap_uint64_array(ub, sizeof (uberblock_t));
1582 if (ub->ub_magic != UBERBLOCK_MAGIC ||
1583 !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1590 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1596 off = vdev_label_offset(vd->v_psize, l, offset);
1599 BP_SET_LSIZE(&bp, size);
1600 BP_SET_PSIZE(&bp, size);
1601 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1602 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1603 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1604 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1606 return (vdev_read_phys(vd, &bp, buf, off, size));
1609 static unsigned char *
1610 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1613 uint64_t best_txg = 0;
1614 uint64_t label_txg = 0;
1620 label = malloc(sizeof (vdev_phys_t));
1624 nvl_size = VDEV_PHYS_SIZE - sizeof (zio_eck_t) - 4;
1625 nvl = malloc(nvl_size);
1629 for (int l = 0; l < VDEV_LABELS; l++) {
1630 const unsigned char *nvlist;
1632 if (vdev_label_read(vd, l, label,
1633 offsetof(vdev_label_t, vl_vdev_phys),
1634 sizeof (vdev_phys_t)))
1637 if (label->vp_nvlist[0] != NV_ENCODE_XDR)
1640 nvlist = (const unsigned char *) label->vp_nvlist + 4;
1641 error = nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
1642 DATA_TYPE_UINT64, NULL, &label_txg);
1643 if (error != 0 || label_txg == 0) {
1644 memcpy(nvl, nvlist, nvl_size);
1648 if (label_txg <= txg && label_txg > best_txg) {
1649 best_txg = label_txg;
1650 memcpy(nvl, nvlist, nvl_size);
1653 * Use asize from pool config. We need this
1654 * because we can get bad value from BIOS.
1656 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1657 DATA_TYPE_UINT64, NULL, &asize) == 0) {
1658 vd->v_psize = asize +
1659 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1664 if (best_txg == 0) {
1674 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1678 buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1682 for (int l = 0; l < VDEV_LABELS; l++) {
1683 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1684 if (vdev_label_read(vd, l, buf,
1685 VDEV_UBERBLOCK_OFFSET(vd, n),
1686 VDEV_UBERBLOCK_SIZE(vd)))
1688 if (uberblock_verify(buf) != 0)
1691 if (vdev_uberblock_compare(buf, ub) > 0)
1699 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1703 vdev_t *vdev, *top_vdev, *pool_vdev;
1704 unsigned char *nvlist;
1707 uint64_t pool_txg, pool_guid;
1708 const char *pool_name;
1709 const unsigned char *vdevs;
1710 const unsigned char *features;
1714 * Load the vdev label and figure out which
1715 * uberblock is most current.
1717 memset(&vtmp, 0, sizeof(vtmp));
1718 vtmp.v_phys_read = _read;
1719 vtmp.v_read_priv = read_priv;
1720 vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1721 (uint64_t)sizeof (vdev_label_t));
1723 /* Test for minimum device size. */
1724 if (vtmp.v_psize < SPA_MINDEVSIZE)
1727 nvlist = vdev_label_read_config(&vtmp, UINT64_MAX);
1731 if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1737 if (!SPA_VERSION_IS_SUPPORTED(val)) {
1738 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1739 (unsigned) val, (unsigned) SPA_VERSION);
1744 /* Check ZFS features for read */
1745 if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1746 DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1747 nvlist_check_features_for_read(features) != 0) {
1752 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1758 if (val == POOL_STATE_DESTROYED) {
1759 /* We don't boot only from destroyed pools. */
1764 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1765 NULL, &pool_txg) != 0 ||
1766 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1767 NULL, &pool_guid) != 0 ||
1768 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1769 NULL, &pool_name) != 0) {
1771 * Cache and spare devices end up here - just ignore
1779 * Create the pool if this is the first time we've seen it.
1781 spa = spa_find_by_guid(pool_guid);
1783 spa = spa_create(pool_guid, pool_name);
1789 if (pool_txg > spa->spa_txg) {
1790 spa->spa_txg = pool_txg;
1797 * Get the vdev tree and create our in-core copy of it.
1798 * If we already have a vdev with this guid, this must
1799 * be some kind of alias (overlapping slices, dangerously dedicated
1802 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1803 NULL, &guid) != 0) {
1807 vdev = vdev_find(guid);
1808 /* Has this vdev already been inited? */
1809 if (vdev && vdev->v_phys_read) {
1814 if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1820 rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1826 * Add the toplevel vdev to the pool if its not already there.
1828 STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1829 if (top_vdev == pool_vdev)
1832 if (!pool_vdev && top_vdev) {
1833 top_vdev->spa = spa;
1834 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1838 * We should already have created an incomplete vdev for this
1839 * vdev. Find it and initialise it with our read proc.
1841 vdev = vdev_find(guid);
1843 vdev->v_phys_read = _read;
1844 vdev->v_read_priv = read_priv;
1845 vdev->v_state = VDEV_STATE_HEALTHY;
1846 vdev->v_psize = vtmp.v_psize;
1848 printf("ZFS: inconsistent nvlist contents\n");
1853 spa->spa_with_log = vdev->v_islog;
1856 * Re-evaluate top-level vdev state.
1858 vdev_set_state(top_vdev);
1861 * Ok, we are happy with the pool so far. Lets find
1862 * the best uberblock and then we can actually access
1863 * the contents of the pool.
1865 vdev_uberblock_load(vdev, &spa->spa_uberblock);
1878 for (v = 0; v < 32; v++)
1885 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1888 zio_gbh_phys_t zio_gb;
1892 /* Artificial BP for gang block header. */
1894 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1895 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1896 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1897 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1898 for (i = 0; i < SPA_DVAS_PER_BP; i++)
1899 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1901 /* Read gang header block using the artificial BP. */
1902 if (zio_read(spa, &gbh_bp, &zio_gb))
1906 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1907 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1909 if (BP_IS_HOLE(gbp))
1911 if (zio_read(spa, gbp, pbuf))
1913 pbuf += BP_GET_PSIZE(gbp);
1916 if (zio_checksum_verify(spa, bp, buf))
1922 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1924 int cpfunc = BP_GET_COMPRESS(bp);
1925 uint64_t align, size;
1930 * Process data embedded in block pointer
1932 if (BP_IS_EMBEDDED(bp)) {
1933 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1935 size = BPE_GET_PSIZE(bp);
1936 ASSERT(size <= BPE_PAYLOAD_SIZE);
1938 if (cpfunc != ZIO_COMPRESS_OFF)
1939 pbuf = zfs_alloc(size);
1943 decode_embedded_bp_compressed(bp, pbuf);
1946 if (cpfunc != ZIO_COMPRESS_OFF) {
1947 error = zio_decompress_data(cpfunc, pbuf,
1948 size, buf, BP_GET_LSIZE(bp));
1949 zfs_free(pbuf, size);
1952 printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1959 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1960 const dva_t *dva = &bp->blk_dva[i];
1965 if (!dva->dva_word[0] && !dva->dva_word[1])
1968 vdevid = DVA_GET_VDEV(dva);
1969 offset = DVA_GET_OFFSET(dva);
1970 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1971 if (vdev->v_id == vdevid)
1974 if (!vdev || !vdev->v_read)
1977 size = BP_GET_PSIZE(bp);
1978 if (vdev->v_read == vdev_raidz_read) {
1979 align = 1ULL << vdev->v_top->v_ashift;
1980 if (P2PHASE(size, align) != 0)
1981 size = P2ROUNDUP(size, align);
1983 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1984 pbuf = zfs_alloc(size);
1988 if (DVA_GET_GANG(dva))
1989 error = zio_read_gang(spa, bp, pbuf);
1991 error = vdev->v_read(vdev, bp, pbuf, offset, size);
1993 if (cpfunc != ZIO_COMPRESS_OFF)
1994 error = zio_decompress_data(cpfunc, pbuf,
1995 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1996 else if (size != BP_GET_PSIZE(bp))
1997 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2000 zfs_free(pbuf, size);
2005 printf("ZFS: i/o error - all block copies unavailable\n");
2010 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
2012 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2013 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2014 int nlevels = dnode->dn_nlevels;
2017 if (bsize > SPA_MAXBLOCKSIZE) {
2018 printf("ZFS: I/O error - blocks larger than %llu are not "
2019 "supported\n", SPA_MAXBLOCKSIZE);
2024 * Note: bsize may not be a power of two here so we need to do an
2025 * actual divide rather than a bitshift.
2027 while (buflen > 0) {
2028 uint64_t bn = offset / bsize;
2029 int boff = offset % bsize;
2031 const blkptr_t *indbp;
2034 if (bn > dnode->dn_maxblkid)
2037 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2040 indbp = dnode->dn_blkptr;
2041 for (i = 0; i < nlevels; i++) {
2043 * Copy the bp from the indirect array so that
2044 * we can re-use the scratch buffer for multi-level
2047 ibn = bn >> ((nlevels - i - 1) * ibshift);
2048 ibn &= ((1 << ibshift) - 1);
2050 if (BP_IS_HOLE(&bp)) {
2051 memset(dnode_cache_buf, 0, bsize);
2054 rc = zio_read(spa, &bp, dnode_cache_buf);
2057 indbp = (const blkptr_t *) dnode_cache_buf;
2059 dnode_cache_obj = dnode;
2060 dnode_cache_bn = bn;
2064 * The buffer contains our data block. Copy what we
2065 * need from it and loop.
2068 if (i > buflen) i = buflen;
2069 memcpy(buf, &dnode_cache_buf[boff], i);
2070 buf = ((char*) buf) + i;
2079 * Lookup a value in a microzap directory. Assumes that the zap
2080 * scratch buffer contains the directory contents.
2083 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
2085 const mzap_phys_t *mz;
2086 const mzap_ent_phys_t *mze;
2091 * Microzap objects use exactly one block. Read the whole
2094 size = dnode->dn_datablkszsec * 512;
2096 mz = (const mzap_phys_t *) zap_scratch;
2097 chunks = size / MZAP_ENT_LEN - 1;
2099 for (i = 0; i < chunks; i++) {
2100 mze = &mz->mz_chunk[i];
2101 if (!strcmp(mze->mze_name, name)) {
2102 *value = mze->mze_value;
2111 * Compare a name with a zap leaf entry. Return non-zero if the name
2115 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
2118 const zap_leaf_chunk_t *nc;
2121 namelen = zc->l_entry.le_name_numints;
2123 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2125 while (namelen > 0) {
2128 if (len > ZAP_LEAF_ARRAY_BYTES)
2129 len = ZAP_LEAF_ARRAY_BYTES;
2130 if (memcmp(p, nc->l_array.la_array, len))
2134 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2141 * Extract a uint64_t value from a zap leaf entry.
2144 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2146 const zap_leaf_chunk_t *vc;
2151 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2152 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2153 value = (value << 8) | p[i];
2160 stv(int len, void *addr, uint64_t value)
2164 *(uint8_t *)addr = value;
2167 *(uint16_t *)addr = value;
2170 *(uint32_t *)addr = value;
2173 *(uint64_t *)addr = value;
2179 * Extract a array from a zap leaf entry.
2182 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2183 uint64_t integer_size, uint64_t num_integers, void *buf)
2185 uint64_t array_int_len = zc->l_entry.le_value_intlen;
2187 uint64_t *u64 = buf;
2189 int len = MIN(zc->l_entry.le_value_numints, num_integers);
2190 int chunk = zc->l_entry.le_value_chunk;
2193 if (integer_size == 8 && len == 1) {
2194 *u64 = fzap_leaf_value(zl, zc);
2199 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2202 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2203 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2204 value = (value << 8) | la->la_array[i];
2206 if (byten == array_int_len) {
2207 stv(integer_size, p, value);
2215 chunk = la->la_next;
2220 * Lookup a value in a fatzap directory. Assumes that the zap scratch
2221 * buffer contains the directory header.
2224 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2225 uint64_t integer_size, uint64_t num_integers, void *value)
2227 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2228 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2234 if (zh.zap_magic != ZAP_MAGIC)
2237 z.zap_block_shift = ilog2(bsize);
2238 z.zap_phys = (zap_phys_t *) zap_scratch;
2241 * Figure out where the pointer table is and read it in if necessary.
2243 if (zh.zap_ptrtbl.zt_blk) {
2244 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
2245 zap_scratch, bsize);
2248 ptrtbl = (uint64_t *) zap_scratch;
2250 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
2253 hash = zap_hash(zh.zap_salt, name);
2256 zl.l_bs = z.zap_block_shift;
2258 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
2259 zap_leaf_chunk_t *zc;
2261 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
2265 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2268 * Make sure this chunk matches our hash.
2270 if (zl.l_phys->l_hdr.lh_prefix_len > 0
2271 && zl.l_phys->l_hdr.lh_prefix
2272 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
2276 * Hash within the chunk to find our entry.
2278 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
2279 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
2280 h = zl.l_phys->l_hash[h];
2283 zc = &ZAP_LEAF_CHUNK(&zl, h);
2284 while (zc->l_entry.le_hash != hash) {
2285 if (zc->l_entry.le_next == 0xffff)
2287 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
2289 if (fzap_name_equal(&zl, zc, name)) {
2290 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
2291 integer_size * num_integers)
2293 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
2301 * Lookup a name in a zap object and return its value as a uint64_t.
2304 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2305 uint64_t integer_size, uint64_t num_integers, void *value)
2309 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2311 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2315 zap_type = *(uint64_t *) zap_scratch;
2316 if (zap_type == ZBT_MICRO)
2317 return mzap_lookup(dnode, name, value);
2318 else if (zap_type == ZBT_HEADER) {
2319 return fzap_lookup(spa, dnode, name, integer_size,
2320 num_integers, value);
2322 printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
2327 * List a microzap directory. Assumes that the zap scratch buffer contains
2328 * the directory contents.
2331 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2333 const mzap_phys_t *mz;
2334 const mzap_ent_phys_t *mze;
2339 * Microzap objects use exactly one block. Read the whole
2342 size = dnode->dn_datablkszsec * 512;
2343 mz = (const mzap_phys_t *) zap_scratch;
2344 chunks = size / MZAP_ENT_LEN - 1;
2346 for (i = 0; i < chunks; i++) {
2347 mze = &mz->mz_chunk[i];
2348 if (mze->mze_name[0]) {
2349 rc = callback(mze->mze_name, mze->mze_value);
2359 * List a fatzap directory. Assumes that the zap scratch buffer contains
2360 * the directory header.
2363 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2365 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2366 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2370 if (zh.zap_magic != ZAP_MAGIC)
2373 z.zap_block_shift = ilog2(bsize);
2374 z.zap_phys = (zap_phys_t *) zap_scratch;
2377 * This assumes that the leaf blocks start at block 1. The
2378 * documentation isn't exactly clear on this.
2381 zl.l_bs = z.zap_block_shift;
2382 for (i = 0; i < zh.zap_num_leafs; i++) {
2383 off_t off = (i + 1) << zl.l_bs;
2387 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2390 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2392 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2393 zap_leaf_chunk_t *zc, *nc;
2396 zc = &ZAP_LEAF_CHUNK(&zl, j);
2397 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2399 namelen = zc->l_entry.le_name_numints;
2400 if (namelen > sizeof(name))
2401 namelen = sizeof(name);
2404 * Paste the name back together.
2406 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2408 while (namelen > 0) {
2411 if (len > ZAP_LEAF_ARRAY_BYTES)
2412 len = ZAP_LEAF_ARRAY_BYTES;
2413 memcpy(p, nc->l_array.la_array, len);
2416 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2420 * Assume the first eight bytes of the value are
2423 value = fzap_leaf_value(&zl, zc);
2425 //printf("%s 0x%jx\n", name, (uintmax_t)value);
2426 rc = callback((const char *)name, value);
2435 static int zfs_printf(const char *name, uint64_t value __unused)
2438 printf("%s\n", name);
2444 * List a zap directory.
2447 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2450 size_t size = dnode->dn_datablkszsec * 512;
2452 if (dnode_read(spa, dnode, 0, zap_scratch, size))
2455 zap_type = *(uint64_t *) zap_scratch;
2456 if (zap_type == ZBT_MICRO)
2457 return mzap_list(dnode, zfs_printf);
2459 return fzap_list(spa, dnode, zfs_printf);
2463 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
2467 offset = objnum * sizeof(dnode_phys_t);
2468 return dnode_read(spa, &os->os_meta_dnode, offset,
2469 dnode, sizeof(dnode_phys_t));
2473 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2475 const mzap_phys_t *mz;
2476 const mzap_ent_phys_t *mze;
2481 * Microzap objects use exactly one block. Read the whole
2484 size = dnode->dn_datablkszsec * 512;
2486 mz = (const mzap_phys_t *) zap_scratch;
2487 chunks = size / MZAP_ENT_LEN - 1;
2489 for (i = 0; i < chunks; i++) {
2490 mze = &mz->mz_chunk[i];
2491 if (value == mze->mze_value) {
2492 strcpy(name, mze->mze_name);
2501 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2504 const zap_leaf_chunk_t *nc;
2507 namelen = zc->l_entry.le_name_numints;
2509 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2511 while (namelen > 0) {
2514 if (len > ZAP_LEAF_ARRAY_BYTES)
2515 len = ZAP_LEAF_ARRAY_BYTES;
2516 memcpy(p, nc->l_array.la_array, len);
2519 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2526 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2528 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2529 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2533 if (zh.zap_magic != ZAP_MAGIC)
2536 z.zap_block_shift = ilog2(bsize);
2537 z.zap_phys = (zap_phys_t *) zap_scratch;
2540 * This assumes that the leaf blocks start at block 1. The
2541 * documentation isn't exactly clear on this.
2544 zl.l_bs = z.zap_block_shift;
2545 for (i = 0; i < zh.zap_num_leafs; i++) {
2546 off_t off = (i + 1) << zl.l_bs;
2548 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2551 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2553 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2554 zap_leaf_chunk_t *zc;
2556 zc = &ZAP_LEAF_CHUNK(&zl, j);
2557 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2559 if (zc->l_entry.le_value_intlen != 8 ||
2560 zc->l_entry.le_value_numints != 1)
2563 if (fzap_leaf_value(&zl, zc) == value) {
2564 fzap_name_copy(&zl, zc, name);
2574 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2578 size_t size = dnode->dn_datablkszsec * 512;
2580 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2584 zap_type = *(uint64_t *) zap_scratch;
2585 if (zap_type == ZBT_MICRO)
2586 return mzap_rlookup(spa, dnode, name, value);
2588 return fzap_rlookup(spa, dnode, name, value);
2592 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2595 char component[256];
2596 uint64_t dir_obj, parent_obj, child_dir_zapobj;
2597 dnode_phys_t child_dir_zap, dataset, dir, parent;
2599 dsl_dataset_phys_t *ds;
2603 p = &name[sizeof(name) - 1];
2606 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2607 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2610 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2611 dir_obj = ds->ds_dir_obj;
2614 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
2616 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2618 /* Actual loop condition. */
2619 parent_obj = dd->dd_parent_obj;
2620 if (parent_obj == 0)
2623 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
2625 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2626 child_dir_zapobj = dd->dd_child_dir_zapobj;
2627 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2629 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2632 len = strlen(component);
2634 memcpy(p, component, len);
2638 /* Actual loop iteration. */
2639 dir_obj = parent_obj;
2650 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2653 uint64_t dir_obj, child_dir_zapobj;
2654 dnode_phys_t child_dir_zap, dir;
2658 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
2660 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2666 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
2668 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2672 /* Actual loop condition #1. */
2678 memcpy(element, p, q - p);
2679 element[q - p] = '\0';
2686 child_dir_zapobj = dd->dd_child_dir_zapobj;
2687 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2690 /* Actual loop condition #2. */
2691 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2696 *objnum = dd->dd_head_dataset_obj;
2702 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2704 uint64_t dir_obj, child_dir_zapobj;
2705 dnode_phys_t child_dir_zap, dir, dataset;
2706 dsl_dataset_phys_t *ds;
2709 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2710 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2713 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2714 dir_obj = ds->ds_dir_obj;
2716 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2717 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2720 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2722 child_dir_zapobj = dd->dd_child_dir_zapobj;
2723 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2724 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2728 return (zap_list(spa, &child_dir_zap) != 0);
2732 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2734 uint64_t dir_obj, child_dir_zapobj, zap_type;
2735 dnode_phys_t child_dir_zap, dir, dataset;
2736 dsl_dataset_phys_t *ds;
2740 err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2742 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2745 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2746 dir_obj = ds->ds_dir_obj;
2748 err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2750 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2753 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2755 child_dir_zapobj = dd->dd_child_dir_zapobj;
2756 err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2758 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2762 err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
2766 zap_type = *(uint64_t *) zap_scratch;
2767 if (zap_type == ZBT_MICRO)
2768 return mzap_list(&child_dir_zap, callback);
2770 return fzap_list(spa, &child_dir_zap, callback);
2775 * Find the object set given the object number of its dataset object
2776 * and return its details in *objset
2779 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2781 dnode_phys_t dataset;
2782 dsl_dataset_phys_t *ds;
2784 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2785 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2789 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2790 if (zio_read(spa, &ds->ds_bp, objset)) {
2791 printf("ZFS: can't read object set for dataset %ju\n",
2800 * Find the object set pointed to by the BOOTFS property or the root
2801 * dataset if there is none and return its details in *objset
2804 zfs_get_root(const spa_t *spa, uint64_t *objid)
2806 dnode_phys_t dir, propdir;
2807 uint64_t props, bootfs, root;
2812 * Start with the MOS directory object.
2814 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2815 printf("ZFS: can't read MOS object directory\n");
2820 * Lookup the pool_props and see if we can find a bootfs.
2822 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, sizeof (props), 1, &props) == 0
2823 && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2824 && zap_lookup(spa, &propdir, "bootfs", sizeof (bootfs), 1, &bootfs) == 0
2831 * Lookup the root dataset directory
2833 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
2834 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2835 printf("ZFS: can't find root dsl_dir\n");
2840 * Use the information from the dataset directory's bonus buffer
2841 * to find the dataset object and from that the object set itself.
2843 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2844 *objid = dd->dd_head_dataset_obj;
2849 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
2855 * Find the root object set if not explicitly provided
2857 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2858 printf("ZFS: can't find root filesystem\n");
2862 if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
2863 printf("ZFS: can't open root filesystem\n");
2867 mount->rootobj = rootobj;
2873 * callback function for feature name checks.
2876 check_feature(const char *name, uint64_t value)
2882 if (name[0] == '\0')
2885 for (i = 0; features_for_read[i] != NULL; i++) {
2886 if (strcmp(name, features_for_read[i]) == 0)
2889 printf("ZFS: unsupported feature: %s\n", name);
2894 * Checks whether the MOS features that are active are supported.
2897 check_mos_features(const spa_t *spa)
2900 uint64_t objnum, zap_type;
2904 if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2907 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2908 sizeof (objnum), 1, &objnum)) != 0) {
2910 * It is older pool without features. As we have already
2911 * tested the label, just return without raising the error.
2916 if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2919 if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2922 size = dir.dn_datablkszsec * 512;
2923 if (dnode_read(spa, &dir, 0, zap_scratch, size))
2926 zap_type = *(uint64_t *) zap_scratch;
2927 if (zap_type == ZBT_MICRO)
2928 rc = mzap_list(&dir, check_feature);
2930 rc = fzap_list(spa, &dir, check_feature);
2936 load_nvlist(spa_t *spa, uint64_t obj, unsigned char **value)
2944 if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
2946 if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
2947 dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
2951 if (dir.dn_bonuslen != sizeof (uint64_t))
2954 size = *(uint64_t *)DN_BONUS(&dir);
2959 rc = dnode_read(spa, &dir, 0, nv, size);
2970 zfs_spa_init(spa_t *spa)
2973 uint64_t config_object;
2974 unsigned char *nvlist;
2976 const unsigned char *nv;
2979 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2980 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2983 if (spa->spa_mos.os_type != DMU_OST_META) {
2984 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2988 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
2990 printf("ZFS: failed to read pool %s directory object\n",
2994 /* this is allowed to fail, older pools do not have salt */
2995 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
2996 sizeof (spa->spa_cksum_salt.zcs_bytes),
2997 spa->spa_cksum_salt.zcs_bytes);
2999 rc = check_mos_features(spa);
3001 printf("ZFS: pool %s is not supported\n", spa->spa_name);
3005 rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3006 sizeof (config_object), 1, &config_object);
3008 printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3011 rc = load_nvlist(spa, config_object, &nvlist);
3015 /* Update vdevs from MOS config. */
3016 if (nvlist_find(nvlist + 4, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
3022 if (nvlist_find(nv, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
3024 printf("ZFS: can't find vdev details\n");
3028 if (strcmp(type, VDEV_TYPE_ROOT) != 0) {
3033 rc = nvlist_find(nv, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
3038 for (int i = 0; i < nkids; i++) {
3039 vdev_t *vd, *prev, *kid = NULL;
3040 rc = vdev_init_from_nvlist(nv, NULL, &kid, 0);
3042 printf("vdev_init_from_nvlist: %d\n", rc);
3047 STAILQ_FOREACH(vd, &spa->spa_vdevs, v_childlink) {
3048 /* Already present? */
3049 if (kid->v_id == vd->v_id) {
3053 if (vd->v_id > kid->v_id) {
3055 STAILQ_INSERT_HEAD(&spa->spa_vdevs,
3058 STAILQ_INSERT_AFTER(&spa->spa_vdevs,
3059 prev, kid, v_childlink);
3067 STAILQ_INSERT_TAIL(&spa->spa_vdevs, kid, v_childlink);
3068 nv = nvlist_next(nv);
3077 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3080 if (dn->dn_bonustype != DMU_OT_SA) {
3081 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3083 sb->st_mode = zp->zp_mode;
3084 sb->st_uid = zp->zp_uid;
3085 sb->st_gid = zp->zp_gid;
3086 sb->st_size = zp->zp_size;
3088 sa_hdr_phys_t *sahdrp;
3093 if (dn->dn_bonuslen != 0)
3094 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3096 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3097 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3100 size = BP_GET_LSIZE(bp);
3101 buf = zfs_alloc(size);
3102 error = zio_read(spa, bp, buf);
3104 zfs_free(buf, size);
3112 hdrsize = SA_HDR_SIZE(sahdrp);
3113 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3115 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3117 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3119 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3122 zfs_free(buf, size);
3129 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3133 if (dn->dn_bonustype == DMU_OT_SA) {
3134 sa_hdr_phys_t *sahdrp = NULL;
3140 if (dn->dn_bonuslen != 0)
3141 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3145 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3147 bp = DN_SPILL_BLKPTR(dn);
3149 size = BP_GET_LSIZE(bp);
3150 buf = zfs_alloc(size);
3151 rc = zio_read(spa, bp, buf);
3153 zfs_free(buf, size);
3158 hdrsize = SA_HDR_SIZE(sahdrp);
3159 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3160 memcpy(path, p, psize);
3162 zfs_free(buf, size);
3166 * Second test is purely to silence bogus compiler
3167 * warning about accessing past the end of dn_bonus.
3169 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3170 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3171 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3173 rc = dnode_read(spa, dn, 0, path, psize);
3180 STAILQ_ENTRY(obj_list) entry;
3184 * Lookup a file and return its dnode.
3187 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3196 int symlinks_followed = 0;
3198 struct obj_list *entry, *tentry;
3199 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3202 if (mount->objset.os_type != DMU_OST_ZFS) {
3203 printf("ZFS: unexpected object set type %ju\n",
3204 (uintmax_t)mount->objset.os_type);
3208 if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3212 * Get the root directory dnode.
3214 rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3220 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof (objnum), 1, &objnum);
3225 entry->objnum = objnum;
3226 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3228 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3234 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3243 while (*q != '\0' && *q != '/')
3247 if (p + 1 == q && p[0] == '.') {
3252 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3254 if (STAILQ_FIRST(&on_cache) ==
3255 STAILQ_LAST(&on_cache, obj_list, entry)) {
3259 entry = STAILQ_FIRST(&on_cache);
3260 STAILQ_REMOVE_HEAD(&on_cache, entry);
3262 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3265 if (q - p + 1 > sizeof(element)) {
3269 memcpy(element, p, q - p);
3273 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3275 if (!S_ISDIR(sb.st_mode)) {
3280 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3283 objnum = ZFS_DIRENT_OBJ(objnum);
3285 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3289 entry->objnum = objnum;
3290 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3291 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3296 * Check for symlink.
3298 rc = zfs_dnode_stat(spa, &dn, &sb);
3301 if (S_ISLNK(sb.st_mode)) {
3302 if (symlinks_followed > 10) {
3306 symlinks_followed++;
3309 * Read the link value and copy the tail of our
3310 * current path onto the end.
3312 if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3316 strcpy(&path[sb.st_size], p);
3318 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3323 * Restart with the new path, starting either at
3324 * the root or at the parent depending whether or
3325 * not the link is relative.
3329 while (STAILQ_FIRST(&on_cache) !=
3330 STAILQ_LAST(&on_cache, obj_list, entry)) {
3331 entry = STAILQ_FIRST(&on_cache);
3332 STAILQ_REMOVE_HEAD(&on_cache, entry);
3336 entry = STAILQ_FIRST(&on_cache);
3337 STAILQ_REMOVE_HEAD(&on_cache, entry);
3340 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3346 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)