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>
38 #include <machine/_inttypes.h>
49 static struct zfsmount zfsmount __unused;
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.
58 typedef struct indirect_child {
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.
69 typedef struct indirect_split {
70 list_node_t is_node; /* link on iv_splits */
73 * is_split_offset is the offset into the i/o.
74 * This is the sum of the previous splits' is_size's.
76 uint64_t is_split_offset;
78 vdev_t *is_vdev; /* top-level vdev */
79 uint64_t is_target_offset; /* offset on is_vdev */
81 int is_children; /* number of entries in is_child[] */
84 * is_good_child is the child that we are currently using to
85 * attempt reconstruction.
89 indirect_child_t is_child[1]; /* variable-length */
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.
96 typedef struct indirect_vsd {
97 boolean_t iv_split_block;
98 boolean_t iv_reconstruct;
100 list_t iv_splits; /* list of indirect_split_t's */
104 * List of all vdevs, chained through v_alllink.
106 static vdev_list_t zfs_vdevs;
109 * List of ZFS features supported for read
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",
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",
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);
497 rc = zio_checksum_verify(vdev->v_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);
573 * Compare an offset with an indirect mapping entry; there are three
574 * possible scenarios:
576 * 1. The offset is "less than" the mapping entry; meaning the
577 * offset is less than the source offset of the mapping entry. In
578 * this case, there is no overlap between the offset and the
579 * mapping entry and -1 will be returned.
581 * 2. The offset is "greater than" the mapping entry; meaning the
582 * offset is greater than the mapping entry's source offset plus
583 * the entry's size. In this case, there is no overlap between
584 * the offset and the mapping entry and 1 will be returned.
586 * NOTE: If the offset is actually equal to the entry's offset
587 * plus size, this is considered to be "greater" than the entry,
588 * and this case applies (i.e. 1 will be returned). Thus, the
589 * entry's "range" can be considered to be inclusive at its
590 * start, but exclusive at its end: e.g. [src, src + size).
592 * 3. The last case to consider is if the offset actually falls
593 * within the mapping entry's range. If this is the case, the
594 * offset is considered to be "equal to" the mapping entry and
595 * 0 will be returned.
597 * NOTE: If the offset is equal to the entry's source offset,
598 * this case applies and 0 will be returned. If the offset is
599 * equal to the entry's source plus its size, this case does
600 * *not* apply (see "NOTE" above for scenario 2), and 1 will be
604 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
606 const uint64_t *key = v_key;
607 const vdev_indirect_mapping_entry_phys_t *array_elem =
609 uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
611 if (*key < src_offset) {
613 } else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
621 * Return array entry.
623 static vdev_indirect_mapping_entry_phys_t *
624 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
630 if (vim->vim_phys->vimp_num_entries == 0)
633 if (vim->vim_entries == NULL) {
636 bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
637 size = vim->vim_phys->vimp_num_entries *
638 sizeof (*vim->vim_entries);
640 size = bsize / sizeof (*vim->vim_entries);
641 size *= sizeof (*vim->vim_entries);
643 vim->vim_entries = malloc(size);
644 if (vim->vim_entries == NULL)
646 vim->vim_num_entries = size / sizeof (*vim->vim_entries);
647 offset = index * sizeof (*vim->vim_entries);
650 /* We have data in vim_entries */
652 if (index >= vim->vim_entry_offset &&
653 index <= vim->vim_entry_offset + vim->vim_num_entries) {
654 index -= vim->vim_entry_offset;
655 return (&vim->vim_entries[index]);
657 offset = index * sizeof (*vim->vim_entries);
660 vim->vim_entry_offset = index;
661 size = vim->vim_num_entries * sizeof (*vim->vim_entries);
662 rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
665 /* Read error, invalidate vim_entries. */
666 free(vim->vim_entries);
667 vim->vim_entries = NULL;
670 index -= vim->vim_entry_offset;
671 return (&vim->vim_entries[index]);
675 * Returns the mapping entry for the given offset.
677 * It's possible that the given offset will not be in the mapping table
678 * (i.e. no mapping entries contain this offset), in which case, the
679 * return value value depends on the "next_if_missing" parameter.
681 * If the offset is not found in the table and "next_if_missing" is
682 * B_FALSE, then NULL will always be returned. The behavior is intended
683 * to allow consumers to get the entry corresponding to the offset
684 * parameter, iff the offset overlaps with an entry in the table.
686 * If the offset is not found in the table and "next_if_missing" is
687 * B_TRUE, then the entry nearest to the given offset will be returned,
688 * such that the entry's source offset is greater than the offset
689 * passed in (i.e. the "next" mapping entry in the table is returned, if
690 * the offset is missing from the table). If there are no entries whose
691 * source offset is greater than the passed in offset, NULL is returned.
693 static vdev_indirect_mapping_entry_phys_t *
694 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
697 ASSERT(vim->vim_phys->vimp_num_entries > 0);
699 vdev_indirect_mapping_entry_phys_t *entry;
701 uint64_t last = vim->vim_phys->vimp_num_entries - 1;
705 * We don't define these inside of the while loop because we use
706 * their value in the case that offset isn't in the mapping.
711 while (last >= base) {
712 mid = base + ((last - base) >> 1);
714 entry = vdev_indirect_mapping_entry(vim, mid);
717 result = dva_mapping_overlap_compare(&offset, entry);
721 } else if (result < 0) {
731 * Given an indirect vdev and an extent on that vdev, it duplicates the
732 * physical entries of the indirect mapping that correspond to the extent
733 * to a new array and returns a pointer to it. In addition, copied_entries
734 * is populated with the number of mapping entries that were duplicated.
736 * Finally, since we are doing an allocation, it is up to the caller to
737 * free the array allocated in this function.
739 vdev_indirect_mapping_entry_phys_t *
740 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
741 uint64_t asize, uint64_t *copied_entries)
743 vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
744 vdev_indirect_mapping_t *vim = vd->v_mapping;
745 uint64_t entries = 0;
747 vdev_indirect_mapping_entry_phys_t *first_mapping =
748 vdev_indirect_mapping_entry_for_offset(vim, offset);
749 ASSERT3P(first_mapping, !=, NULL);
751 vdev_indirect_mapping_entry_phys_t *m = first_mapping;
753 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
754 uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
755 uint64_t inner_size = MIN(asize, size - inner_offset);
757 offset += inner_size;
763 size_t copy_length = entries * sizeof (*first_mapping);
764 duplicate_mappings = malloc(copy_length);
765 if (duplicate_mappings != NULL)
766 bcopy(first_mapping, duplicate_mappings, copy_length);
770 *copied_entries = entries;
772 return (duplicate_mappings);
776 vdev_lookup_top(spa_t *spa, uint64_t vdev)
781 vlist = &spa->spa_root_vdev->v_children;
782 STAILQ_FOREACH(rvd, vlist, v_childlink)
783 if (rvd->v_id == vdev)
790 * This is a callback for vdev_indirect_remap() which allocates an
791 * indirect_split_t for each split segment and adds it to iv_splits.
794 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
795 uint64_t size, void *arg)
799 indirect_vsd_t *iv = zio->io_vsd;
801 if (vd->v_read == vdev_indirect_read)
804 if (vd->v_read == vdev_mirror_read)
807 indirect_split_t *is =
808 malloc(offsetof(indirect_split_t, is_child[n]));
810 zio->io_error = ENOMEM;
813 bzero(is, offsetof(indirect_split_t, is_child[n]));
817 is->is_split_offset = split_offset;
818 is->is_target_offset = offset;
822 * Note that we only consider multiple copies of the data for
823 * *mirror* vdevs. We don't for "replacing" or "spare" vdevs, even
824 * though they use the same ops as mirror, because there's only one
825 * "good" copy under the replacing/spare.
827 if (vd->v_read == vdev_mirror_read) {
831 STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
832 is->is_child[i++].ic_vdev = kid;
835 is->is_child[0].ic_vdev = vd;
838 list_insert_tail(&iv->iv_splits, is);
842 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
845 spa_t *spa = vd->v_spa;
849 list_create(&stack, sizeof (remap_segment_t),
850 offsetof(remap_segment_t, rs_node));
852 rs = rs_alloc(vd, offset, asize, 0);
854 printf("vdev_indirect_remap: out of memory.\n");
855 zio->io_error = ENOMEM;
857 for ( ; rs != NULL; rs = list_remove_head(&stack)) {
858 vdev_t *v = rs->rs_vd;
859 uint64_t num_entries = 0;
860 /* vdev_indirect_mapping_t *vim = v->v_mapping; */
861 vdev_indirect_mapping_entry_phys_t *mapping =
862 vdev_indirect_mapping_duplicate_adjacent_entries(v,
863 rs->rs_offset, rs->rs_asize, &num_entries);
865 if (num_entries == 0)
866 zio->io_error = ENOMEM;
868 for (uint64_t i = 0; i < num_entries; i++) {
869 vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
870 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
871 uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
872 uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
873 uint64_t inner_offset = rs->rs_offset -
874 DVA_MAPPING_GET_SRC_OFFSET(m);
875 uint64_t inner_size =
876 MIN(rs->rs_asize, size - inner_offset);
877 vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
879 if (dst_v->v_read == vdev_indirect_read) {
882 o = rs_alloc(dst_v, dst_offset + inner_offset,
883 inner_size, rs->rs_split_offset);
885 printf("vdev_indirect_remap: "
887 zio->io_error = ENOMEM;
891 list_insert_head(&stack, o);
893 vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
894 dst_offset + inner_offset,
898 * vdev_indirect_gather_splits can have memory
899 * allocation error, we can not recover from it.
901 if (zio->io_error != 0)
903 rs->rs_offset += inner_size;
904 rs->rs_asize -= inner_size;
905 rs->rs_split_offset += inner_size;
910 if (zio->io_error != 0)
914 list_destroy(&stack);
918 vdev_indirect_map_free(zio_t *zio)
920 indirect_vsd_t *iv = zio->io_vsd;
921 indirect_split_t *is;
923 while ((is = list_head(&iv->iv_splits)) != NULL) {
924 for (int c = 0; c < is->is_children; c++) {
925 indirect_child_t *ic = &is->is_child[c];
928 list_remove(&iv->iv_splits, is);
935 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
936 off_t offset, size_t bytes)
939 spa_t *spa = vdev->v_spa;
941 indirect_split_t *first;
944 iv = calloc(1, sizeof(*iv));
948 list_create(&iv->iv_splits,
949 sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
951 bzero(&zio, sizeof(zio));
953 zio.io_bp = (blkptr_t *)bp;
956 zio.io_offset = offset;
960 if (vdev->v_mapping == NULL) {
961 vdev_indirect_config_t *vic;
963 vic = &vdev->vdev_indirect_config;
964 vdev->v_mapping = vdev_indirect_mapping_open(spa,
965 &spa->spa_mos, vic->vic_mapping_object);
968 vdev_indirect_remap(vdev, offset, bytes, &zio);
969 if (zio.io_error != 0)
970 return (zio.io_error);
972 first = list_head(&iv->iv_splits);
973 if (first->is_size == zio.io_size) {
975 * This is not a split block; we are pointing to the entire
976 * data, which will checksum the same as the original data.
977 * Pass the BP down so that the child i/o can verify the
978 * checksum, and try a different location if available
979 * (e.g. on a mirror).
981 * While this special case could be handled the same as the
982 * general (split block) case, doing it this way ensures
983 * that the vast majority of blocks on indirect vdevs
984 * (which are not split) are handled identically to blocks
985 * on non-indirect vdevs. This allows us to be less strict
986 * about performance in the general (but rare) case.
988 rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
989 zio.io_data, first->is_target_offset, bytes);
991 iv->iv_split_block = B_TRUE;
993 * Read one copy of each split segment, from the
994 * top-level vdev. Since we don't know the
995 * checksum of each split individually, the child
996 * zio can't ensure that we get the right data.
997 * E.g. if it's a mirror, it will just read from a
998 * random (healthy) leaf vdev. We have to verify
999 * the checksum in vdev_indirect_io_done().
1001 for (indirect_split_t *is = list_head(&iv->iv_splits);
1002 is != NULL; is = list_next(&iv->iv_splits, is)) {
1003 char *ptr = zio.io_data;
1005 rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
1006 ptr + is->is_split_offset, is->is_target_offset,
1009 if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
1015 vdev_indirect_map_free(&zio);
1023 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1024 off_t offset, size_t bytes)
1027 return (vdev_read_phys(vdev, bp, buf,
1028 offset + VDEV_LABEL_START_SIZE, bytes));
1033 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1034 off_t offset, size_t bytes)
1040 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1041 if (kid->v_state != VDEV_STATE_HEALTHY)
1043 rc = kid->v_read(kid, bp, buf, offset, bytes);
1052 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1053 off_t offset, size_t bytes)
1058 * Here we should have two kids:
1059 * First one which is the one we are replacing and we can trust
1060 * only this one to have valid data, but it might not be present.
1061 * Second one is that one we are replacing with. It is most likely
1062 * healthy, but we can't trust it has needed data, so we won't use it.
1064 kid = STAILQ_FIRST(&vdev->v_children);
1067 if (kid->v_state != VDEV_STATE_HEALTHY)
1069 return (kid->v_read(kid, bp, buf, offset, bytes));
1073 vdev_find(uint64_t guid)
1077 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
1078 if (vdev->v_guid == guid)
1085 vdev_create(uint64_t guid, vdev_read_t *_read)
1088 vdev_indirect_config_t *vic;
1090 vdev = calloc(1, sizeof(vdev_t));
1092 STAILQ_INIT(&vdev->v_children);
1093 vdev->v_guid = guid;
1094 vdev->v_read = _read;
1097 * root vdev has no read function.
1098 * We only point root vdev from spa.
1100 if (_read != NULL) {
1101 vic = &vdev->vdev_indirect_config;
1102 vic->vic_prev_indirect_vdev = UINT64_MAX;
1103 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
1111 vdev_set_initial_state(vdev_t *vdev, const unsigned char *nvlist)
1113 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
1116 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
1118 (void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
1120 (void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
1122 (void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
1124 (void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
1125 NULL, &is_degraded);
1126 (void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
1127 NULL, &isnt_present);
1128 (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
1131 if (is_offline != 0)
1132 vdev->v_state = VDEV_STATE_OFFLINE;
1133 else if (is_removed != 0)
1134 vdev->v_state = VDEV_STATE_REMOVED;
1135 else if (is_faulted != 0)
1136 vdev->v_state = VDEV_STATE_FAULTED;
1137 else if (is_degraded != 0)
1138 vdev->v_state = VDEV_STATE_DEGRADED;
1139 else if (isnt_present != 0)
1140 vdev->v_state = VDEV_STATE_CANT_OPEN;
1142 vdev->v_islog = is_log == 1;
1146 vdev_init(uint64_t guid, const unsigned char *nvlist, vdev_t **vdevp)
1148 uint64_t id, ashift, asize, nparity;
1153 if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id) ||
1154 nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1159 if (strcmp(type, VDEV_TYPE_MIRROR)
1160 && strcmp(type, VDEV_TYPE_DISK)
1162 && strcmp(type, VDEV_TYPE_FILE)
1164 && strcmp(type, VDEV_TYPE_RAIDZ)
1165 && strcmp(type, VDEV_TYPE_INDIRECT)
1166 && strcmp(type, VDEV_TYPE_REPLACING)) {
1167 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
1171 if (strcmp(type, VDEV_TYPE_MIRROR) == 0)
1172 vdev = vdev_create(guid, vdev_mirror_read);
1173 else if (strcmp(type, VDEV_TYPE_RAIDZ) == 0)
1174 vdev = vdev_create(guid, vdev_raidz_read);
1175 else if (strcmp(type, VDEV_TYPE_REPLACING) == 0)
1176 vdev = vdev_create(guid, vdev_replacing_read);
1177 else if (strcmp(type, VDEV_TYPE_INDIRECT) == 0) {
1178 vdev_indirect_config_t *vic;
1180 vdev = vdev_create(guid, vdev_indirect_read);
1182 vdev->v_state = VDEV_STATE_HEALTHY;
1183 vic = &vdev->vdev_indirect_config;
1186 ZPOOL_CONFIG_INDIRECT_OBJECT,
1188 NULL, &vic->vic_mapping_object);
1190 ZPOOL_CONFIG_INDIRECT_BIRTHS,
1192 NULL, &vic->vic_births_object);
1194 ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
1196 NULL, &vic->vic_prev_indirect_vdev);
1199 vdev = vdev_create(guid, vdev_disk_read);
1205 vdev_set_initial_state(vdev, nvlist);
1207 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
1208 DATA_TYPE_UINT64, NULL, &ashift) == 0)
1209 vdev->v_ashift = ashift;
1211 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1212 DATA_TYPE_UINT64, NULL, &asize) == 0) {
1213 vdev->v_psize = asize +
1214 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1217 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
1218 DATA_TYPE_UINT64, NULL, &nparity) == 0)
1219 vdev->v_nparity = nparity;
1221 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1222 DATA_TYPE_STRING, NULL, &path) == 0) {
1223 if (strncmp(path, "/dev/", 5) == 0)
1225 vdev->v_name = strdup(path);
1230 if (strcmp(type, "raidz") == 0) {
1231 if (vdev->v_nparity < 1 ||
1232 vdev->v_nparity > 3) {
1233 printf("ZFS: can only boot from disk, "
1234 "mirror, raidz1, raidz2 and raidz3 "
1238 (void) asprintf(&name, "%s%d-%" PRIu64, type,
1239 vdev->v_nparity, id);
1241 (void) asprintf(&name, "%s-%" PRIu64, type, id);
1243 vdev->v_name = name;
1250 * Find slot for vdev. We return either NULL to signal to use
1251 * STAILQ_INSERT_HEAD, or we return link element to be used with
1252 * STAILQ_INSERT_AFTER.
1255 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1257 vdev_t *v, *previous;
1259 if (STAILQ_EMPTY(&top_vdev->v_children))
1263 STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1264 if (v->v_id > vdev->v_id)
1267 if (v->v_id == vdev->v_id)
1270 if (v->v_id < vdev->v_id)
1277 vdev_child_count(vdev_t *vdev)
1283 STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1290 * Insert vdev into top_vdev children list. List is ordered by v_id.
1293 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1299 * The top level vdev can appear in random order, depending how
1300 * the firmware is presenting the disk devices.
1301 * However, we will insert vdev to create list ordered by v_id,
1302 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1303 * as STAILQ does not have insert before.
1305 previous = vdev_find_previous(top_vdev, vdev);
1307 if (previous == NULL) {
1308 STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1309 count = vdev_child_count(top_vdev);
1310 if (top_vdev->v_nchildren < count)
1311 top_vdev->v_nchildren = count;
1315 if (previous->v_id == vdev->v_id)
1318 STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev, v_childlink);
1319 count = vdev_child_count(top_vdev);
1320 if (top_vdev->v_nchildren < count)
1321 top_vdev->v_nchildren = count;
1325 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const unsigned char *nvlist)
1327 vdev_t *top_vdev, *vdev;
1328 const unsigned char *kids;
1332 top_vdev = vdev_find(top_guid);
1333 if (top_vdev == NULL) {
1334 rc = vdev_init(top_guid, nvlist, &top_vdev);
1337 top_vdev->v_spa = spa;
1338 top_vdev->v_top = top_vdev;
1339 vdev_insert(spa->spa_root_vdev, top_vdev);
1342 /* Add children if there are any. */
1343 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1346 for (int i = 0; i < nkids; i++) {
1349 rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1350 DATA_TYPE_UINT64, NULL, &guid);
1353 rc = vdev_init(guid, kids, &vdev);
1358 vdev->v_top = top_vdev;
1359 vdev_insert(top_vdev, vdev);
1361 kids = nvlist_next(kids);
1371 vdev_init_from_label(spa_t *spa, const unsigned char *nvlist)
1373 uint64_t pool_guid, top_guid;
1374 const unsigned char *vdevs;
1376 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1377 NULL, &pool_guid) ||
1378 nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1380 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1382 printf("ZFS: can't find vdev details\n");
1386 return (vdev_from_nvlist(spa, top_guid, vdevs));
1390 vdev_set_state(vdev_t *vdev)
1396 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1397 vdev_set_state(kid);
1401 * A mirror or raidz is healthy if all its kids are healthy. A
1402 * mirror is degraded if any of its kids is healthy; a raidz
1403 * is degraded if at most nparity kids are offline.
1405 if (STAILQ_FIRST(&vdev->v_children)) {
1408 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1409 if (kid->v_state == VDEV_STATE_HEALTHY)
1414 if (bad_kids == 0) {
1415 vdev->v_state = VDEV_STATE_HEALTHY;
1417 if (vdev->v_read == vdev_mirror_read) {
1419 vdev->v_state = VDEV_STATE_DEGRADED;
1421 vdev->v_state = VDEV_STATE_OFFLINE;
1423 } else if (vdev->v_read == vdev_raidz_read) {
1424 if (bad_kids > vdev->v_nparity) {
1425 vdev->v_state = VDEV_STATE_OFFLINE;
1427 vdev->v_state = VDEV_STATE_DEGRADED;
1435 vdev_update_from_nvlist(uint64_t top_guid, const unsigned char *nvlist)
1438 const unsigned char *kids;
1441 /* Update top vdev. */
1442 vdev = vdev_find(top_guid);
1444 vdev_set_initial_state(vdev, nvlist);
1446 /* Update children if there are any. */
1447 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1450 for (int i = 0; i < nkids; i++) {
1453 rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1454 DATA_TYPE_UINT64, NULL, &guid);
1458 vdev = vdev_find(guid);
1460 vdev_set_initial_state(vdev, kids);
1462 kids = nvlist_next(kids);
1472 vdev_init_from_nvlist(spa_t *spa, const unsigned char *nvlist)
1474 uint64_t pool_guid, vdev_children;
1475 const unsigned char *vdevs, *kids;
1478 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1479 NULL, &pool_guid) ||
1480 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1481 NULL, &vdev_children) ||
1482 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1484 printf("ZFS: can't find vdev details\n");
1489 if (spa->spa_guid != pool_guid)
1492 spa->spa_root_vdev->v_nchildren = vdev_children;
1494 rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1498 * MOS config has at least one child for root vdev.
1503 for (int i = 0; i < nkids; i++) {
1507 rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1511 vdev = vdev_find(guid);
1513 * Top level vdev is missing, create it.
1516 rc = vdev_from_nvlist(spa, guid, kids);
1518 rc = vdev_update_from_nvlist(guid, kids);
1521 kids = nvlist_next(kids);
1525 * Re-evaluate top-level vdev state.
1527 vdev_set_state(spa->spa_root_vdev);
1533 spa_find_by_guid(uint64_t guid)
1537 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1538 if (spa->spa_guid == guid)
1545 spa_find_by_name(const char *name)
1549 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1550 if (!strcmp(spa->spa_name, name))
1558 spa_get_primary(void)
1561 return (STAILQ_FIRST(&zfs_pools));
1565 spa_get_primary_vdev(const spa_t *spa)
1571 spa = spa_get_primary();
1574 vdev = spa->spa_root_vdev;
1577 for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
1578 kid = STAILQ_FIRST(&vdev->v_children))
1585 spa_create(uint64_t guid, const char *name)
1589 if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1591 if ((spa->spa_name = strdup(name)) == NULL) {
1595 spa->spa_guid = guid;
1596 spa->spa_root_vdev = vdev_create(guid, NULL);
1597 if (spa->spa_root_vdev == NULL) {
1598 free(spa->spa_name);
1602 spa->spa_root_vdev->v_name = strdup("root");
1603 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1609 state_name(vdev_state_t state)
1611 static const char* names[] = {
1621 return names[state];
1626 #define pager_printf printf
1631 pager_printf(const char *fmt, ...)
1636 va_start(args, fmt);
1637 vsnprintf(line, sizeof(line), fmt, args);
1639 return (pager_output(line));
1644 #define STATUS_FORMAT " %s %s\n"
1647 print_state(int indent, const char *name, vdev_state_t state)
1653 for (i = 0; i < indent; i++)
1656 return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1660 vdev_status(vdev_t *vdev, int indent)
1665 if (vdev->v_islog) {
1666 (void)pager_output(" logs\n");
1670 ret = print_state(indent, vdev->v_name, vdev->v_state);
1674 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1675 ret = vdev_status(kid, indent + 1);
1683 spa_status(spa_t *spa)
1685 static char bootfs[ZFS_MAXNAMELEN];
1689 int good_kids, bad_kids, degraded_kids, ret;
1692 ret = pager_printf(" pool: %s\n", spa->spa_name);
1696 if (zfs_get_root(spa, &rootid) == 0 &&
1697 zfs_rlookup(spa, rootid, bootfs) == 0) {
1698 if (bootfs[0] == '\0')
1699 ret = pager_printf("bootfs: %s\n", spa->spa_name);
1701 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1706 ret = pager_printf("config:\n\n");
1709 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1716 vlist = &spa->spa_root_vdev->v_children;
1717 STAILQ_FOREACH(vdev, vlist, v_childlink) {
1718 if (vdev->v_state == VDEV_STATE_HEALTHY)
1720 else if (vdev->v_state == VDEV_STATE_DEGRADED)
1726 state = VDEV_STATE_CLOSED;
1727 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1728 state = VDEV_STATE_HEALTHY;
1729 else if ((good_kids + degraded_kids) > 0)
1730 state = VDEV_STATE_DEGRADED;
1732 ret = print_state(0, spa->spa_name, state);
1736 STAILQ_FOREACH(vdev, vlist, v_childlink) {
1737 ret = vdev_status(vdev, 1);
1745 spa_all_status(void)
1748 int first = 1, ret = 0;
1750 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1752 ret = pager_printf("\n");
1757 ret = spa_status(spa);
1765 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1767 uint64_t label_offset;
1769 if (l < VDEV_LABELS / 2)
1772 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1774 return (offset + l * sizeof (vdev_label_t) + label_offset);
1778 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1780 unsigned int seq1 = 0;
1781 unsigned int seq2 = 0;
1782 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1787 cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1791 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1792 seq1 = MMP_SEQ(ub1);
1794 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1795 seq2 = MMP_SEQ(ub2);
1797 return (AVL_CMP(seq1, seq2));
1801 uberblock_verify(uberblock_t *ub)
1803 if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1804 byteswap_uint64_array(ub, sizeof (uberblock_t));
1807 if (ub->ub_magic != UBERBLOCK_MAGIC ||
1808 !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1815 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1821 off = vdev_label_offset(vd->v_psize, l, offset);
1824 BP_SET_LSIZE(&bp, size);
1825 BP_SET_PSIZE(&bp, size);
1826 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1827 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1828 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1829 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1831 return (vdev_read_phys(vd, &bp, buf, off, size));
1834 static unsigned char *
1835 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1838 uint64_t best_txg = 0;
1839 uint64_t label_txg = 0;
1845 label = malloc(sizeof (vdev_phys_t));
1849 nvl_size = VDEV_PHYS_SIZE - sizeof (zio_eck_t) - 4;
1850 nvl = malloc(nvl_size);
1854 for (int l = 0; l < VDEV_LABELS; l++) {
1855 const unsigned char *nvlist;
1857 if (vdev_label_read(vd, l, label,
1858 offsetof(vdev_label_t, vl_vdev_phys),
1859 sizeof (vdev_phys_t)))
1862 if (label->vp_nvlist[0] != NV_ENCODE_XDR)
1865 nvlist = (const unsigned char *) label->vp_nvlist + 4;
1866 error = nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
1867 DATA_TYPE_UINT64, NULL, &label_txg);
1868 if (error != 0 || label_txg == 0) {
1869 memcpy(nvl, nvlist, nvl_size);
1873 if (label_txg <= txg && label_txg > best_txg) {
1874 best_txg = label_txg;
1875 memcpy(nvl, nvlist, nvl_size);
1878 * Use asize from pool config. We need this
1879 * because we can get bad value from BIOS.
1881 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1882 DATA_TYPE_UINT64, NULL, &asize) == 0) {
1883 vd->v_psize = asize +
1884 VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1889 if (best_txg == 0) {
1899 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1903 buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1907 for (int l = 0; l < VDEV_LABELS; l++) {
1908 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1909 if (vdev_label_read(vd, l, buf,
1910 VDEV_UBERBLOCK_OFFSET(vd, n),
1911 VDEV_UBERBLOCK_SIZE(vd)))
1913 if (uberblock_verify(buf) != 0)
1916 if (vdev_uberblock_compare(buf, ub) > 0)
1924 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1929 unsigned char *nvlist;
1931 uint64_t guid, vdev_children;
1932 uint64_t pool_txg, pool_guid;
1933 const char *pool_name;
1934 const unsigned char *features;
1938 * Load the vdev label and figure out which
1939 * uberblock is most current.
1941 memset(&vtmp, 0, sizeof(vtmp));
1942 vtmp.v_phys_read = _read;
1943 vtmp.v_read_priv = read_priv;
1944 vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1945 (uint64_t)sizeof (vdev_label_t));
1947 /* Test for minimum device size. */
1948 if (vtmp.v_psize < SPA_MINDEVSIZE)
1951 nvlist = vdev_label_read_config(&vtmp, UINT64_MAX);
1955 if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1961 if (!SPA_VERSION_IS_SUPPORTED(val)) {
1962 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1963 (unsigned) val, (unsigned) SPA_VERSION);
1968 /* Check ZFS features for read */
1969 if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1970 DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1971 nvlist_check_features_for_read(features) != 0) {
1976 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1982 if (val == POOL_STATE_DESTROYED) {
1983 /* We don't boot only from destroyed pools. */
1988 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1989 NULL, &pool_txg) != 0 ||
1990 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1991 NULL, &pool_guid) != 0 ||
1992 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1993 NULL, &pool_name) != 0) {
1995 * Cache and spare devices end up here - just ignore
2003 * Create the pool if this is the first time we've seen it.
2005 spa = spa_find_by_guid(pool_guid);
2007 nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN,
2008 DATA_TYPE_UINT64, NULL, &vdev_children);
2009 spa = spa_create(pool_guid, pool_name);
2014 spa->spa_root_vdev->v_nchildren = vdev_children;
2016 if (pool_txg > spa->spa_txg)
2017 spa->spa_txg = pool_txg;
2020 * Get the vdev tree and create our in-core copy of it.
2021 * If we already have a vdev with this guid, this must
2022 * be some kind of alias (overlapping slices, dangerously dedicated
2025 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
2026 NULL, &guid) != 0) {
2030 vdev = vdev_find(guid);
2031 /* Has this vdev already been inited? */
2032 if (vdev && vdev->v_phys_read) {
2037 rc = vdev_init_from_label(spa, nvlist);
2043 * We should already have created an incomplete vdev for this
2044 * vdev. Find it and initialise it with our read proc.
2046 vdev = vdev_find(guid);
2048 vdev->v_phys_read = _read;
2049 vdev->v_read_priv = read_priv;
2050 vdev->v_psize = vtmp.v_psize;
2052 * If no other state is set, mark vdev healthy.
2054 if (vdev->v_state == VDEV_STATE_UNKNOWN)
2055 vdev->v_state = VDEV_STATE_HEALTHY;
2057 printf("ZFS: inconsistent nvlist contents\n");
2062 spa->spa_with_log = vdev->v_islog;
2065 * Re-evaluate top-level vdev state.
2067 vdev_set_state(vdev->v_top);
2070 * Ok, we are happy with the pool so far. Lets find
2071 * the best uberblock and then we can actually access
2072 * the contents of the pool.
2074 vdev_uberblock_load(vdev, &spa->spa_uberblock);
2086 for (v = 0; v < 32; v++)
2093 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
2096 zio_gbh_phys_t zio_gb;
2100 /* Artificial BP for gang block header. */
2102 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2103 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2104 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
2105 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
2106 for (i = 0; i < SPA_DVAS_PER_BP; i++)
2107 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
2109 /* Read gang header block using the artificial BP. */
2110 if (zio_read(spa, &gbh_bp, &zio_gb))
2114 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
2115 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
2117 if (BP_IS_HOLE(gbp))
2119 if (zio_read(spa, gbp, pbuf))
2121 pbuf += BP_GET_PSIZE(gbp);
2124 if (zio_checksum_verify(spa, bp, buf))
2130 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
2132 int cpfunc = BP_GET_COMPRESS(bp);
2133 uint64_t align, size;
2138 * Process data embedded in block pointer
2140 if (BP_IS_EMBEDDED(bp)) {
2141 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
2143 size = BPE_GET_PSIZE(bp);
2144 ASSERT(size <= BPE_PAYLOAD_SIZE);
2146 if (cpfunc != ZIO_COMPRESS_OFF)
2147 pbuf = zfs_alloc(size);
2151 decode_embedded_bp_compressed(bp, pbuf);
2154 if (cpfunc != ZIO_COMPRESS_OFF) {
2155 error = zio_decompress_data(cpfunc, pbuf,
2156 size, buf, BP_GET_LSIZE(bp));
2157 zfs_free(pbuf, size);
2160 printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
2167 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
2168 const dva_t *dva = &bp->blk_dva[i];
2174 if (!dva->dva_word[0] && !dva->dva_word[1])
2177 vdevid = DVA_GET_VDEV(dva);
2178 offset = DVA_GET_OFFSET(dva);
2179 vlist = &spa->spa_root_vdev->v_children;
2180 STAILQ_FOREACH(vdev, vlist, v_childlink) {
2181 if (vdev->v_id == vdevid)
2184 if (!vdev || !vdev->v_read)
2187 size = BP_GET_PSIZE(bp);
2188 if (vdev->v_read == vdev_raidz_read) {
2189 align = 1ULL << vdev->v_ashift;
2190 if (P2PHASE(size, align) != 0)
2191 size = P2ROUNDUP(size, align);
2193 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2194 pbuf = zfs_alloc(size);
2198 if (DVA_GET_GANG(dva))
2199 error = zio_read_gang(spa, bp, pbuf);
2201 error = vdev->v_read(vdev, bp, pbuf, offset, size);
2203 if (cpfunc != ZIO_COMPRESS_OFF)
2204 error = zio_decompress_data(cpfunc, pbuf,
2205 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2206 else if (size != BP_GET_PSIZE(bp))
2207 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2210 zfs_free(pbuf, size);
2215 printf("ZFS: i/o error - all block copies unavailable\n");
2220 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
2222 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2223 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2224 int nlevels = dnode->dn_nlevels;
2227 if (bsize > SPA_MAXBLOCKSIZE) {
2228 printf("ZFS: I/O error - blocks larger than %llu are not "
2229 "supported\n", SPA_MAXBLOCKSIZE);
2234 * Note: bsize may not be a power of two here so we need to do an
2235 * actual divide rather than a bitshift.
2237 while (buflen > 0) {
2238 uint64_t bn = offset / bsize;
2239 int boff = offset % bsize;
2241 const blkptr_t *indbp;
2244 if (bn > dnode->dn_maxblkid)
2247 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2250 indbp = dnode->dn_blkptr;
2251 for (i = 0; i < nlevels; i++) {
2253 * Copy the bp from the indirect array so that
2254 * we can re-use the scratch buffer for multi-level
2257 ibn = bn >> ((nlevels - i - 1) * ibshift);
2258 ibn &= ((1 << ibshift) - 1);
2260 if (BP_IS_HOLE(&bp)) {
2261 memset(dnode_cache_buf, 0, bsize);
2264 rc = zio_read(spa, &bp, dnode_cache_buf);
2267 indbp = (const blkptr_t *) dnode_cache_buf;
2269 dnode_cache_obj = dnode;
2270 dnode_cache_bn = bn;
2274 * The buffer contains our data block. Copy what we
2275 * need from it and loop.
2278 if (i > buflen) i = buflen;
2279 memcpy(buf, &dnode_cache_buf[boff], i);
2280 buf = ((char *)buf) + i;
2289 * Lookup a value in a microzap directory. Assumes that the zap
2290 * scratch buffer contains the directory contents.
2293 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
2295 const mzap_phys_t *mz;
2296 const mzap_ent_phys_t *mze;
2301 * Microzap objects use exactly one block. Read the whole
2304 size = dnode->dn_datablkszsec * 512;
2306 mz = (const mzap_phys_t *) zap_scratch;
2307 chunks = size / MZAP_ENT_LEN - 1;
2309 for (i = 0; i < chunks; i++) {
2310 mze = &mz->mz_chunk[i];
2311 if (!strcmp(mze->mze_name, name)) {
2312 *value = mze->mze_value;
2321 * Compare a name with a zap leaf entry. Return non-zero if the name
2325 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
2328 const zap_leaf_chunk_t *nc;
2331 namelen = zc->l_entry.le_name_numints;
2333 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2335 while (namelen > 0) {
2338 if (len > ZAP_LEAF_ARRAY_BYTES)
2339 len = ZAP_LEAF_ARRAY_BYTES;
2340 if (memcmp(p, nc->l_array.la_array, len))
2344 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2351 * Extract a uint64_t value from a zap leaf entry.
2354 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2356 const zap_leaf_chunk_t *vc;
2361 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2362 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2363 value = (value << 8) | p[i];
2370 stv(int len, void *addr, uint64_t value)
2374 *(uint8_t *)addr = value;
2377 *(uint16_t *)addr = value;
2380 *(uint32_t *)addr = value;
2383 *(uint64_t *)addr = value;
2389 * Extract a array from a zap leaf entry.
2392 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2393 uint64_t integer_size, uint64_t num_integers, void *buf)
2395 uint64_t array_int_len = zc->l_entry.le_value_intlen;
2397 uint64_t *u64 = buf;
2399 int len = MIN(zc->l_entry.le_value_numints, num_integers);
2400 int chunk = zc->l_entry.le_value_chunk;
2403 if (integer_size == 8 && len == 1) {
2404 *u64 = fzap_leaf_value(zl, zc);
2409 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2412 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2413 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2414 value = (value << 8) | la->la_array[i];
2416 if (byten == array_int_len) {
2417 stv(integer_size, p, value);
2425 chunk = la->la_next;
2430 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2433 switch (integer_size) {
2443 if (integer_size * num_integers > ZAP_MAXVALUELEN)
2450 * Lookup a value in a fatzap directory. Assumes that the zap scratch
2451 * buffer contains the directory header.
2454 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2455 uint64_t integer_size, uint64_t num_integers, void *value)
2457 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2458 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2464 if (zh.zap_magic != ZAP_MAGIC)
2467 if ((rc = fzap_check_size(integer_size, num_integers)) != 0)
2470 z.zap_block_shift = ilog2(bsize);
2471 z.zap_phys = (zap_phys_t *) zap_scratch;
2474 * Figure out where the pointer table is and read it in if necessary.
2476 if (zh.zap_ptrtbl.zt_blk) {
2477 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
2478 zap_scratch, bsize);
2481 ptrtbl = (uint64_t *) zap_scratch;
2483 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
2486 hash = zap_hash(zh.zap_salt, name);
2489 zl.l_bs = z.zap_block_shift;
2491 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
2492 zap_leaf_chunk_t *zc;
2494 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
2498 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2501 * Make sure this chunk matches our hash.
2503 if (zl.l_phys->l_hdr.lh_prefix_len > 0
2504 && zl.l_phys->l_hdr.lh_prefix
2505 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
2509 * Hash within the chunk to find our entry.
2511 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
2512 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
2513 h = zl.l_phys->l_hash[h];
2516 zc = &ZAP_LEAF_CHUNK(&zl, h);
2517 while (zc->l_entry.le_hash != hash) {
2518 if (zc->l_entry.le_next == 0xffff) {
2522 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
2524 if (fzap_name_equal(&zl, zc, name)) {
2525 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
2526 integer_size * num_integers)
2528 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
2536 * Lookup a name in a zap object and return its value as a uint64_t.
2539 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2540 uint64_t integer_size, uint64_t num_integers, void *value)
2544 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2546 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2550 zap_type = *(uint64_t *) zap_scratch;
2551 if (zap_type == ZBT_MICRO)
2552 return mzap_lookup(dnode, name, value);
2553 else if (zap_type == ZBT_HEADER) {
2554 return fzap_lookup(spa, dnode, name, integer_size,
2555 num_integers, value);
2557 printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
2562 * List a microzap directory. Assumes that the zap scratch buffer contains
2563 * the directory contents.
2566 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2568 const mzap_phys_t *mz;
2569 const mzap_ent_phys_t *mze;
2574 * Microzap objects use exactly one block. Read the whole
2577 size = dnode->dn_datablkszsec * 512;
2578 mz = (const mzap_phys_t *) zap_scratch;
2579 chunks = size / MZAP_ENT_LEN - 1;
2581 for (i = 0; i < chunks; i++) {
2582 mze = &mz->mz_chunk[i];
2583 if (mze->mze_name[0]) {
2584 rc = callback(mze->mze_name, mze->mze_value);
2594 * List a fatzap directory. Assumes that the zap scratch buffer contains
2595 * the directory header.
2598 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2600 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2601 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2605 if (zh.zap_magic != ZAP_MAGIC)
2608 z.zap_block_shift = ilog2(bsize);
2609 z.zap_phys = (zap_phys_t *) zap_scratch;
2612 * This assumes that the leaf blocks start at block 1. The
2613 * documentation isn't exactly clear on this.
2616 zl.l_bs = z.zap_block_shift;
2617 for (i = 0; i < zh.zap_num_leafs; i++) {
2618 off_t off = ((off_t)(i + 1)) << zl.l_bs;
2622 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2625 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2627 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2628 zap_leaf_chunk_t *zc, *nc;
2631 zc = &ZAP_LEAF_CHUNK(&zl, j);
2632 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2634 namelen = zc->l_entry.le_name_numints;
2635 if (namelen > sizeof(name))
2636 namelen = sizeof(name);
2639 * Paste the name back together.
2641 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2643 while (namelen > 0) {
2646 if (len > ZAP_LEAF_ARRAY_BYTES)
2647 len = ZAP_LEAF_ARRAY_BYTES;
2648 memcpy(p, nc->l_array.la_array, len);
2651 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2655 * Assume the first eight bytes of the value are
2658 value = fzap_leaf_value(&zl, zc);
2660 //printf("%s 0x%jx\n", name, (uintmax_t)value);
2661 rc = callback((const char *)name, value);
2670 static int zfs_printf(const char *name, uint64_t value __unused)
2673 printf("%s\n", name);
2679 * List a zap directory.
2682 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2685 size_t size = dnode->dn_datablkszsec * 512;
2687 if (dnode_read(spa, dnode, 0, zap_scratch, size))
2690 zap_type = *(uint64_t *) zap_scratch;
2691 if (zap_type == ZBT_MICRO)
2692 return mzap_list(dnode, zfs_printf);
2694 return fzap_list(spa, dnode, zfs_printf);
2698 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
2702 offset = objnum * sizeof(dnode_phys_t);
2703 return dnode_read(spa, &os->os_meta_dnode, offset,
2704 dnode, sizeof(dnode_phys_t));
2708 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2710 const mzap_phys_t *mz;
2711 const mzap_ent_phys_t *mze;
2716 * Microzap objects use exactly one block. Read the whole
2719 size = dnode->dn_datablkszsec * 512;
2721 mz = (const mzap_phys_t *) zap_scratch;
2722 chunks = size / MZAP_ENT_LEN - 1;
2724 for (i = 0; i < chunks; i++) {
2725 mze = &mz->mz_chunk[i];
2726 if (value == mze->mze_value) {
2727 strcpy(name, mze->mze_name);
2736 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2739 const zap_leaf_chunk_t *nc;
2742 namelen = zc->l_entry.le_name_numints;
2744 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2746 while (namelen > 0) {
2749 if (len > ZAP_LEAF_ARRAY_BYTES)
2750 len = ZAP_LEAF_ARRAY_BYTES;
2751 memcpy(p, nc->l_array.la_array, len);
2754 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2761 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2763 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2764 zap_phys_t zh = *(zap_phys_t *)zap_scratch;
2768 if (zh.zap_magic != ZAP_MAGIC)
2771 z.zap_block_shift = ilog2(bsize);
2772 z.zap_phys = (zap_phys_t *) zap_scratch;
2775 * This assumes that the leaf blocks start at block 1. The
2776 * documentation isn't exactly clear on this.
2779 zl.l_bs = z.zap_block_shift;
2780 for (i = 0; i < zh.zap_num_leafs; i++) {
2781 off_t off = ((off_t)(i + 1)) << zl.l_bs;
2783 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2786 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2788 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2789 zap_leaf_chunk_t *zc;
2791 zc = &ZAP_LEAF_CHUNK(&zl, j);
2792 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2794 if (zc->l_entry.le_value_intlen != 8 ||
2795 zc->l_entry.le_value_numints != 1)
2798 if (fzap_leaf_value(&zl, zc) == value) {
2799 fzap_name_copy(&zl, zc, name);
2809 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2813 size_t size = dnode->dn_datablkszsec * 512;
2815 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2819 zap_type = *(uint64_t *) zap_scratch;
2820 if (zap_type == ZBT_MICRO)
2821 return mzap_rlookup(spa, dnode, name, value);
2823 return fzap_rlookup(spa, dnode, name, value);
2827 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2830 char component[256];
2831 uint64_t dir_obj, parent_obj, child_dir_zapobj;
2832 dnode_phys_t child_dir_zap, dataset, dir, parent;
2834 dsl_dataset_phys_t *ds;
2838 p = &name[sizeof(name) - 1];
2841 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2842 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2845 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2846 dir_obj = ds->ds_dir_obj;
2849 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
2851 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2853 /* Actual loop condition. */
2854 parent_obj = dd->dd_parent_obj;
2855 if (parent_obj == 0)
2858 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
2860 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2861 child_dir_zapobj = dd->dd_child_dir_zapobj;
2862 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2864 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2867 len = strlen(component);
2869 memcpy(p, component, len);
2873 /* Actual loop iteration. */
2874 dir_obj = parent_obj;
2885 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2888 uint64_t dir_obj, child_dir_zapobj;
2889 dnode_phys_t child_dir_zap, dir;
2893 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
2895 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2901 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
2903 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2907 /* Actual loop condition #1. */
2913 memcpy(element, p, q - p);
2914 element[q - p] = '\0';
2921 child_dir_zapobj = dd->dd_child_dir_zapobj;
2922 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2925 /* Actual loop condition #2. */
2926 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2931 *objnum = dd->dd_head_dataset_obj;
2937 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2939 uint64_t dir_obj, child_dir_zapobj;
2940 dnode_phys_t child_dir_zap, dir, dataset;
2941 dsl_dataset_phys_t *ds;
2944 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2945 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2948 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2949 dir_obj = ds->ds_dir_obj;
2951 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2952 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2955 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2957 child_dir_zapobj = dd->dd_child_dir_zapobj;
2958 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2959 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2963 return (zap_list(spa, &child_dir_zap) != 0);
2967 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2969 uint64_t dir_obj, child_dir_zapobj, zap_type;
2970 dnode_phys_t child_dir_zap, dir, dataset;
2971 dsl_dataset_phys_t *ds;
2975 err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2977 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2980 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2981 dir_obj = ds->ds_dir_obj;
2983 err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2985 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2988 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2990 child_dir_zapobj = dd->dd_child_dir_zapobj;
2991 err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2993 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2997 err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
3001 zap_type = *(uint64_t *) zap_scratch;
3002 if (zap_type == ZBT_MICRO)
3003 return mzap_list(&child_dir_zap, callback);
3005 return fzap_list(spa, &child_dir_zap, callback);
3010 * Find the object set given the object number of its dataset object
3011 * and return its details in *objset
3014 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
3016 dnode_phys_t dataset;
3017 dsl_dataset_phys_t *ds;
3019 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
3020 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3024 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
3025 if (zio_read(spa, &ds->ds_bp, objset)) {
3026 printf("ZFS: can't read object set for dataset %ju\n",
3035 * Find the object set pointed to by the BOOTFS property or the root
3036 * dataset if there is none and return its details in *objset
3039 zfs_get_root(const spa_t *spa, uint64_t *objid)
3041 dnode_phys_t dir, propdir;
3042 uint64_t props, bootfs, root;
3047 * Start with the MOS directory object.
3049 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3050 printf("ZFS: can't read MOS object directory\n");
3055 * Lookup the pool_props and see if we can find a bootfs.
3057 if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3058 sizeof(props), 1, &props) == 0 &&
3059 objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0 &&
3060 zap_lookup(spa, &propdir, "bootfs",
3061 sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3066 * Lookup the root dataset directory
3068 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
3069 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
3070 printf("ZFS: can't find root dsl_dir\n");
3075 * Use the information from the dataset directory's bonus buffer
3076 * to find the dataset object and from that the object set itself.
3078 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
3079 *objid = dd->dd_head_dataset_obj;
3084 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3090 * Find the root object set if not explicitly provided
3092 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3093 printf("ZFS: can't find root filesystem\n");
3097 if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3098 printf("ZFS: can't open root filesystem\n");
3102 mount->rootobj = rootobj;
3108 * callback function for feature name checks.
3111 check_feature(const char *name, uint64_t value)
3117 if (name[0] == '\0')
3120 for (i = 0; features_for_read[i] != NULL; i++) {
3121 if (strcmp(name, features_for_read[i]) == 0)
3124 printf("ZFS: unsupported feature: %s\n", name);
3129 * Checks whether the MOS features that are active are supported.
3132 check_mos_features(const spa_t *spa)
3135 uint64_t objnum, zap_type;
3139 if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3142 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3143 sizeof (objnum), 1, &objnum)) != 0) {
3145 * It is older pool without features. As we have already
3146 * tested the label, just return without raising the error.
3151 if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
3154 if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3157 size = dir.dn_datablkszsec * 512;
3158 if (dnode_read(spa, &dir, 0, zap_scratch, size))
3161 zap_type = *(uint64_t *) zap_scratch;
3162 if (zap_type == ZBT_MICRO)
3163 rc = mzap_list(&dir, check_feature);
3165 rc = fzap_list(spa, &dir, check_feature);
3171 load_nvlist(spa_t *spa, uint64_t obj, unsigned char **value)
3179 if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
3181 if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3182 dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3186 if (dir.dn_bonuslen != sizeof (uint64_t))
3189 size = *(uint64_t *)DN_BONUS(&dir);
3194 rc = dnode_read(spa, &dir, 0, nv, size);
3205 zfs_spa_init(spa_t *spa)
3208 uint64_t config_object;
3209 unsigned char *nvlist;
3212 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
3213 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3216 if (spa->spa_mos.os_type != DMU_OST_META) {
3217 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3221 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
3223 printf("ZFS: failed to read pool %s directory object\n",
3227 /* this is allowed to fail, older pools do not have salt */
3228 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3229 sizeof (spa->spa_cksum_salt.zcs_bytes),
3230 spa->spa_cksum_salt.zcs_bytes);
3232 rc = check_mos_features(spa);
3234 printf("ZFS: pool %s is not supported\n", spa->spa_name);
3238 rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3239 sizeof (config_object), 1, &config_object);
3241 printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3244 rc = load_nvlist(spa, config_object, &nvlist);
3248 /* Update vdevs from MOS config. */
3249 rc = vdev_init_from_nvlist(spa, nvlist + 4);
3255 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3258 if (dn->dn_bonustype != DMU_OT_SA) {
3259 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3261 sb->st_mode = zp->zp_mode;
3262 sb->st_uid = zp->zp_uid;
3263 sb->st_gid = zp->zp_gid;
3264 sb->st_size = zp->zp_size;
3266 sa_hdr_phys_t *sahdrp;
3271 if (dn->dn_bonuslen != 0)
3272 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3274 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3275 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3278 size = BP_GET_LSIZE(bp);
3279 buf = zfs_alloc(size);
3280 error = zio_read(spa, bp, buf);
3282 zfs_free(buf, size);
3290 hdrsize = SA_HDR_SIZE(sahdrp);
3291 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3293 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3295 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3297 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3300 zfs_free(buf, size);
3307 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3311 if (dn->dn_bonustype == DMU_OT_SA) {
3312 sa_hdr_phys_t *sahdrp = NULL;
3318 if (dn->dn_bonuslen != 0)
3319 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3323 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3325 bp = DN_SPILL_BLKPTR(dn);
3327 size = BP_GET_LSIZE(bp);
3328 buf = zfs_alloc(size);
3329 rc = zio_read(spa, bp, buf);
3331 zfs_free(buf, size);
3336 hdrsize = SA_HDR_SIZE(sahdrp);
3337 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3338 memcpy(path, p, psize);
3340 zfs_free(buf, size);
3344 * Second test is purely to silence bogus compiler
3345 * warning about accessing past the end of dn_bonus.
3347 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3348 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3349 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3351 rc = dnode_read(spa, dn, 0, path, psize);
3358 STAILQ_ENTRY(obj_list) entry;
3362 * Lookup a file and return its dnode.
3365 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3374 int symlinks_followed = 0;
3376 struct obj_list *entry, *tentry;
3377 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3380 if (mount->objset.os_type != DMU_OST_ZFS) {
3381 printf("ZFS: unexpected object set type %ju\n",
3382 (uintmax_t)mount->objset.os_type);
3386 if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3390 * Get the root directory dnode.
3392 rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3398 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof (objnum), 1, &objnum);
3403 entry->objnum = objnum;
3404 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3406 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3412 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3421 while (*q != '\0' && *q != '/')
3425 if (p + 1 == q && p[0] == '.') {
3430 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3432 if (STAILQ_FIRST(&on_cache) ==
3433 STAILQ_LAST(&on_cache, obj_list, entry)) {
3437 entry = STAILQ_FIRST(&on_cache);
3438 STAILQ_REMOVE_HEAD(&on_cache, entry);
3440 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3443 if (q - p + 1 > sizeof(element)) {
3447 memcpy(element, p, q - p);
3451 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3453 if (!S_ISDIR(sb.st_mode)) {
3458 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3461 objnum = ZFS_DIRENT_OBJ(objnum);
3463 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3467 entry->objnum = objnum;
3468 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3469 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3474 * Check for symlink.
3476 rc = zfs_dnode_stat(spa, &dn, &sb);
3479 if (S_ISLNK(sb.st_mode)) {
3480 if (symlinks_followed > 10) {
3484 symlinks_followed++;
3487 * Read the link value and copy the tail of our
3488 * current path onto the end.
3490 if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3494 strcpy(&path[sb.st_size], p);
3496 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3501 * Restart with the new path, starting either at
3502 * the root or at the parent depending whether or
3503 * not the link is relative.
3507 while (STAILQ_FIRST(&on_cache) !=
3508 STAILQ_LAST(&on_cache, obj_list, entry)) {
3509 entry = STAILQ_FIRST(&on_cache);
3510 STAILQ_REMOVE_HEAD(&on_cache, entry);
3514 entry = STAILQ_FIRST(&on_cache);
3515 STAILQ_REMOVE_HEAD(&on_cache, entry);
3518 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3524 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)