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.
35 #include <sys/stdint.h>
47 static struct zfsmount zfsmount __unused;
50 * The indirect_child_t represents the vdev that we will read from, when we
51 * need to read all copies of the data (e.g. for scrub or reconstruction).
52 * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
53 * ic_vdev is the same as is_vdev. However, for mirror top-level vdevs,
54 * ic_vdev is a child of the mirror.
56 typedef struct indirect_child {
62 * The indirect_split_t represents one mapped segment of an i/o to the
63 * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
64 * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
65 * For split blocks, there will be several of these.
67 typedef struct indirect_split {
68 list_node_t is_node; /* link on iv_splits */
71 * is_split_offset is the offset into the i/o.
72 * This is the sum of the previous splits' is_size's.
74 uint64_t is_split_offset;
76 vdev_t *is_vdev; /* top-level vdev */
77 uint64_t is_target_offset; /* offset on is_vdev */
79 int is_children; /* number of entries in is_child[] */
82 * is_good_child is the child that we are currently using to
83 * attempt reconstruction.
87 indirect_child_t is_child[1]; /* variable-length */
91 * The indirect_vsd_t is associated with each i/o to the indirect vdev.
92 * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
94 typedef struct indirect_vsd {
95 boolean_t iv_split_block;
96 boolean_t iv_reconstruct;
98 list_t iv_splits; /* list of indirect_split_t's */
102 * List of all vdevs, chained through v_alllink.
104 static vdev_list_t zfs_vdevs;
107 * List of ZFS features supported for read
109 static const char *features_for_read[] = {
110 "org.illumos:lz4_compress",
111 "com.delphix:hole_birth",
112 "com.delphix:extensible_dataset",
113 "com.delphix:embedded_data",
114 "org.open-zfs:large_blocks",
115 "org.illumos:sha512",
117 "org.zfsonlinux:large_dnode",
118 "com.joyent:multi_vdev_crash_dump",
119 "com.delphix:device_removal",
120 "com.delphix:obsolete_counts",
125 * List of all pools, chained through spa_link.
127 static spa_list_t zfs_pools;
129 static const dnode_phys_t *dnode_cache_obj;
130 static uint64_t dnode_cache_bn;
131 static char *dnode_cache_buf;
132 static char *zap_scratch;
133 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
135 #define TEMP_SIZE (1024 * 1024)
137 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
138 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
139 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
140 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
141 const char *name, uint64_t integer_size, uint64_t num_integers,
143 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
145 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
147 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
149 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
150 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
152 vdev_indirect_mapping_entry_phys_t *
153 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
154 uint64_t, uint64_t *);
159 STAILQ_INIT(&zfs_vdevs);
160 STAILQ_INIT(&zfs_pools);
162 zfs_temp_buf = malloc(TEMP_SIZE);
163 zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
164 zfs_temp_ptr = zfs_temp_buf;
165 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
166 zap_scratch = malloc(SPA_MAXBLOCKSIZE);
172 zfs_alloc(size_t size)
176 if (zfs_temp_ptr + size > zfs_temp_end) {
177 printf("ZFS: out of temporary buffer space\n");
181 zfs_temp_ptr += size;
187 zfs_free(void *ptr, size_t size)
190 zfs_temp_ptr -= size;
191 if (zfs_temp_ptr != ptr) {
192 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
198 xdr_int(const unsigned char **xdr, int *ip)
200 *ip = ((*xdr)[0] << 24)
209 xdr_u_int(const unsigned char **xdr, u_int *ip)
211 *ip = ((*xdr)[0] << 24)
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;
845 list_create(&stack, sizeof (remap_segment_t),
846 offsetof(remap_segment_t, rs_node));
848 for (remap_segment_t *rs = rs_alloc(vd, offset, asize, 0);
849 rs != NULL; rs = list_remove_head(&stack)) {
850 vdev_t *v = rs->rs_vd;
851 uint64_t num_entries = 0;
852 /* vdev_indirect_mapping_t *vim = v->v_mapping; */
853 vdev_indirect_mapping_entry_phys_t *mapping =
854 vdev_indirect_mapping_duplicate_adjacent_entries(v,
855 rs->rs_offset, rs->rs_asize, &num_entries);
857 for (uint64_t i = 0; i < num_entries; i++) {
858 vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
859 uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
860 uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
861 uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
862 uint64_t inner_offset = rs->rs_offset -
863 DVA_MAPPING_GET_SRC_OFFSET(m);
864 uint64_t inner_size =
865 MIN(rs->rs_asize, size - inner_offset);
866 vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
868 if (dst_v->v_read == vdev_indirect_read) {
869 list_insert_head(&stack,
870 rs_alloc(dst_v, dst_offset + inner_offset,
871 inner_size, rs->rs_split_offset));
873 vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
874 dst_offset + inner_offset,
878 * vdev_indirect_gather_splits can have memory
879 * allocation error, we can not recover from it.
881 if (zio->io_error != 0)
884 rs->rs_offset += inner_size;
885 rs->rs_asize -= inner_size;
886 rs->rs_split_offset += inner_size;
891 if (zio->io_error != 0)
895 list_destroy(&stack);
899 vdev_indirect_map_free(zio_t *zio)
901 indirect_vsd_t *iv = zio->io_vsd;
902 indirect_split_t *is;
904 while ((is = list_head(&iv->iv_splits)) != NULL) {
905 for (int c = 0; c < is->is_children; c++) {
906 indirect_child_t *ic = &is->is_child[c];
909 list_remove(&iv->iv_splits, is);
916 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
917 off_t offset, size_t bytes)
920 spa_t *spa = vdev->spa;
921 indirect_vsd_t *iv = malloc(sizeof (*iv));
922 indirect_split_t *first;
927 bzero(iv, sizeof (*iv));
929 list_create(&iv->iv_splits,
930 sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
933 zio.io_bp = (blkptr_t *)bp;
936 zio.io_offset = offset;
940 if (vdev->v_mapping == NULL) {
941 vdev_indirect_config_t *vic;
943 vic = &vdev->vdev_indirect_config;
944 vdev->v_mapping = vdev_indirect_mapping_open(spa,
945 &spa->spa_mos, vic->vic_mapping_object);
948 vdev_indirect_remap(vdev, offset, bytes, &zio);
949 if (zio.io_error != 0)
950 return (zio.io_error);
952 first = list_head(&iv->iv_splits);
953 if (first->is_size == zio.io_size) {
955 * This is not a split block; we are pointing to the entire
956 * data, which will checksum the same as the original data.
957 * Pass the BP down so that the child i/o can verify the
958 * checksum, and try a different location if available
959 * (e.g. on a mirror).
961 * While this special case could be handled the same as the
962 * general (split block) case, doing it this way ensures
963 * that the vast majority of blocks on indirect vdevs
964 * (which are not split) are handled identically to blocks
965 * on non-indirect vdevs. This allows us to be less strict
966 * about performance in the general (but rare) case.
968 rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
969 zio.io_data, first->is_target_offset, bytes);
971 iv->iv_split_block = B_TRUE;
973 * Read one copy of each split segment, from the
974 * top-level vdev. Since we don't know the
975 * checksum of each split individually, the child
976 * zio can't ensure that we get the right data.
977 * E.g. if it's a mirror, it will just read from a
978 * random (healthy) leaf vdev. We have to verify
979 * the checksum in vdev_indirect_io_done().
981 for (indirect_split_t *is = list_head(&iv->iv_splits);
982 is != NULL; is = list_next(&iv->iv_splits, is)) {
983 char *ptr = zio.io_data;
985 rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
986 ptr + is->is_split_offset, is->is_target_offset,
989 if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
995 vdev_indirect_map_free(&zio);
1003 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1004 off_t offset, size_t bytes)
1007 return (vdev_read_phys(vdev, bp, buf,
1008 offset + VDEV_LABEL_START_SIZE, bytes));
1013 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1014 off_t offset, size_t bytes)
1020 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1021 if (kid->v_state != VDEV_STATE_HEALTHY)
1023 rc = kid->v_read(kid, bp, buf, offset, bytes);
1032 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1033 off_t offset, size_t bytes)
1038 * Here we should have two kids:
1039 * First one which is the one we are replacing and we can trust
1040 * only this one to have valid data, but it might not be present.
1041 * Second one is that one we are replacing with. It is most likely
1042 * healthy, but we can't trust it has needed data, so we won't use it.
1044 kid = STAILQ_FIRST(&vdev->v_children);
1047 if (kid->v_state != VDEV_STATE_HEALTHY)
1049 return (kid->v_read(kid, bp, buf, offset, bytes));
1053 vdev_find(uint64_t guid)
1057 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
1058 if (vdev->v_guid == guid)
1065 vdev_create(uint64_t guid, vdev_read_t *_read)
1068 vdev_indirect_config_t *vic;
1070 vdev = malloc(sizeof(vdev_t));
1071 memset(vdev, 0, sizeof(vdev_t));
1072 STAILQ_INIT(&vdev->v_children);
1073 vdev->v_guid = guid;
1074 vdev->v_state = VDEV_STATE_OFFLINE;
1075 vdev->v_read = _read;
1077 vic = &vdev->vdev_indirect_config;
1078 vic->vic_prev_indirect_vdev = UINT64_MAX;
1079 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
1085 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
1086 vdev_t **vdevp, int is_newer)
1089 uint64_t guid, id, ashift, nparity;
1093 const unsigned char *kids;
1094 int nkids, i, is_new;
1095 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
1097 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1099 || nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id)
1100 || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1102 printf("ZFS: can't find vdev details\n");
1106 if (strcmp(type, VDEV_TYPE_MIRROR)
1107 && strcmp(type, VDEV_TYPE_DISK)
1109 && strcmp(type, VDEV_TYPE_FILE)
1111 && strcmp(type, VDEV_TYPE_RAIDZ)
1112 && strcmp(type, VDEV_TYPE_INDIRECT)
1113 && strcmp(type, VDEV_TYPE_REPLACING)) {
1114 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
1118 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
1120 nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
1122 nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
1124 nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
1126 nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
1128 nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
1131 vdev = vdev_find(guid);
1135 if (!strcmp(type, VDEV_TYPE_MIRROR))
1136 vdev = vdev_create(guid, vdev_mirror_read);
1137 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
1138 vdev = vdev_create(guid, vdev_raidz_read);
1139 else if (!strcmp(type, VDEV_TYPE_REPLACING))
1140 vdev = vdev_create(guid, vdev_replacing_read);
1141 else if (!strcmp(type, VDEV_TYPE_INDIRECT)) {
1142 vdev_indirect_config_t *vic;
1144 vdev = vdev_create(guid, vdev_indirect_read);
1145 vdev->v_state = VDEV_STATE_HEALTHY;
1146 vic = &vdev->vdev_indirect_config;
1149 ZPOOL_CONFIG_INDIRECT_OBJECT, DATA_TYPE_UINT64,
1150 NULL, &vic->vic_mapping_object);
1152 ZPOOL_CONFIG_INDIRECT_BIRTHS, DATA_TYPE_UINT64,
1153 NULL, &vic->vic_births_object);
1155 ZPOOL_CONFIG_PREV_INDIRECT_VDEV, DATA_TYPE_UINT64,
1156 NULL, &vic->vic_prev_indirect_vdev);
1158 vdev = vdev_create(guid, vdev_disk_read);
1161 vdev->v_top = pvdev != NULL ? pvdev : vdev;
1162 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
1163 DATA_TYPE_UINT64, NULL, &ashift) == 0) {
1164 vdev->v_ashift = ashift;
1168 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
1169 DATA_TYPE_UINT64, NULL, &nparity) == 0) {
1170 vdev->v_nparity = nparity;
1172 vdev->v_nparity = 0;
1174 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1175 DATA_TYPE_STRING, NULL, &path) == 0) {
1176 if (strncmp(path, "/dev/", 5) == 0)
1178 vdev->v_name = strdup(path);
1182 if (!strcmp(type, "raidz")) {
1183 if (vdev->v_nparity < 1 ||
1184 vdev->v_nparity > 3) {
1185 printf("ZFS: can only boot from disk, "
1186 "mirror, raidz1, raidz2 and raidz3 "
1190 asprintf(&name, "%s%d-%jd", type,
1191 vdev->v_nparity, id);
1193 asprintf(&name, "%s-%jd", type, id);
1197 vdev->v_name = name;
1203 if (is_new || is_newer) {
1205 * This is either new vdev or we've already seen this vdev,
1206 * but from an older vdev label, so let's refresh its state
1207 * from the newer label.
1210 vdev->v_state = VDEV_STATE_OFFLINE;
1211 else if (is_removed)
1212 vdev->v_state = VDEV_STATE_REMOVED;
1213 else if (is_faulted)
1214 vdev->v_state = VDEV_STATE_FAULTED;
1215 else if (is_degraded)
1216 vdev->v_state = VDEV_STATE_DEGRADED;
1217 else if (isnt_present)
1218 vdev->v_state = VDEV_STATE_CANT_OPEN;
1221 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1224 * Its ok if we don't have any kids.
1227 vdev->v_nchildren = nkids;
1228 for (i = 0; i < nkids; i++) {
1229 rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
1233 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
1235 kids = nvlist_next(kids);
1238 vdev->v_nchildren = 0;
1247 vdev_set_state(vdev_t *vdev)
1254 * A mirror or raidz is healthy if all its kids are healthy. A
1255 * mirror is degraded if any of its kids is healthy; a raidz
1256 * is degraded if at most nparity kids are offline.
1258 if (STAILQ_FIRST(&vdev->v_children)) {
1261 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1262 if (kid->v_state == VDEV_STATE_HEALTHY)
1267 if (bad_kids == 0) {
1268 vdev->v_state = VDEV_STATE_HEALTHY;
1270 if (vdev->v_read == vdev_mirror_read) {
1272 vdev->v_state = VDEV_STATE_DEGRADED;
1274 vdev->v_state = VDEV_STATE_OFFLINE;
1276 } else if (vdev->v_read == vdev_raidz_read) {
1277 if (bad_kids > vdev->v_nparity) {
1278 vdev->v_state = VDEV_STATE_OFFLINE;
1280 vdev->v_state = VDEV_STATE_DEGRADED;
1288 spa_find_by_guid(uint64_t guid)
1292 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1293 if (spa->spa_guid == guid)
1300 spa_find_by_name(const char *name)
1304 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1305 if (!strcmp(spa->spa_name, name))
1313 spa_get_primary(void)
1316 return (STAILQ_FIRST(&zfs_pools));
1320 spa_get_primary_vdev(const spa_t *spa)
1326 spa = spa_get_primary();
1329 vdev = STAILQ_FIRST(&spa->spa_vdevs);
1332 for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
1333 kid = STAILQ_FIRST(&vdev->v_children))
1340 spa_create(uint64_t guid, const char *name)
1344 if ((spa = malloc(sizeof(spa_t))) == NULL)
1346 memset(spa, 0, sizeof(spa_t));
1347 if ((spa->spa_name = strdup(name)) == NULL) {
1351 STAILQ_INIT(&spa->spa_vdevs);
1352 spa->spa_guid = guid;
1353 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1359 state_name(vdev_state_t state)
1361 static const char* names[] = {
1371 return names[state];
1376 #define pager_printf printf
1381 pager_printf(const char *fmt, ...)
1386 va_start(args, fmt);
1387 vsprintf(line, fmt, args);
1390 return (pager_output(line));
1395 #define STATUS_FORMAT " %s %s\n"
1398 print_state(int indent, const char *name, vdev_state_t state)
1404 for (i = 0; i < indent; i++)
1408 return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1412 vdev_status(vdev_t *vdev, int indent)
1416 ret = print_state(indent, vdev->v_name, vdev->v_state);
1420 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1421 ret = vdev_status(kid, indent + 1);
1429 spa_status(spa_t *spa)
1431 static char bootfs[ZFS_MAXNAMELEN];
1434 int good_kids, bad_kids, degraded_kids, ret;
1437 ret = pager_printf(" pool: %s\n", spa->spa_name);
1441 if (zfs_get_root(spa, &rootid) == 0 &&
1442 zfs_rlookup(spa, rootid, bootfs) == 0) {
1443 if (bootfs[0] == '\0')
1444 ret = pager_printf("bootfs: %s\n", spa->spa_name);
1446 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1451 ret = pager_printf("config:\n\n");
1454 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1461 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1462 if (vdev->v_state == VDEV_STATE_HEALTHY)
1464 else if (vdev->v_state == VDEV_STATE_DEGRADED)
1470 state = VDEV_STATE_CLOSED;
1471 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1472 state = VDEV_STATE_HEALTHY;
1473 else if ((good_kids + degraded_kids) > 0)
1474 state = VDEV_STATE_DEGRADED;
1476 ret = print_state(0, spa->spa_name, state);
1479 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1480 ret = vdev_status(vdev, 1);
1488 spa_all_status(void)
1491 int first = 1, ret = 0;
1493 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1495 ret = pager_printf("\n");
1500 ret = spa_status(spa);
1508 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1510 uint64_t label_offset;
1512 if (l < VDEV_LABELS / 2)
1515 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1517 return (offset + l * sizeof (vdev_label_t) + label_offset);
1521 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1524 vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
1525 vdev_phys_t *tmp_label;
1527 vdev_t *vdev, *top_vdev, *pool_vdev;
1530 const unsigned char *nvlist = NULL;
1533 uint64_t best_txg = 0;
1534 uint64_t pool_txg, pool_guid;
1536 const char *pool_name;
1537 const unsigned char *vdevs;
1538 const unsigned char *features;
1539 int i, l, rc, is_newer;
1541 const struct uberblock *up;
1544 * Load the vdev label and figure out which
1545 * uberblock is most current.
1547 memset(&vtmp, 0, sizeof(vtmp));
1548 vtmp.v_phys_read = _read;
1549 vtmp.v_read_priv = read_priv;
1550 psize = P2ALIGN(ldi_get_size(read_priv),
1551 (uint64_t)sizeof (vdev_label_t));
1553 /* Test for minimum pool size. */
1554 if (psize < SPA_MINDEVSIZE)
1557 tmp_label = zfs_alloc(sizeof(vdev_phys_t));
1559 for (l = 0; l < VDEV_LABELS; l++) {
1560 off = vdev_label_offset(psize, l,
1561 offsetof(vdev_label_t, vl_vdev_phys));
1564 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
1565 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
1566 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1567 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1568 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1569 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1571 if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
1574 if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
1577 nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
1578 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
1579 DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
1582 if (best_txg <= pool_txg) {
1583 best_txg = pool_txg;
1584 memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
1588 zfs_free(tmp_label, sizeof (vdev_phys_t));
1593 if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
1596 nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1598 if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1603 if (!SPA_VERSION_IS_SUPPORTED(val)) {
1604 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1605 (unsigned) val, (unsigned) SPA_VERSION);
1609 /* Check ZFS features for read */
1610 if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1611 DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1612 nvlist_check_features_for_read(features) != 0) {
1616 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1621 if (val == POOL_STATE_DESTROYED) {
1622 /* We don't boot only from destroyed pools. */
1626 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1627 NULL, &pool_txg) != 0 ||
1628 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1629 NULL, &pool_guid) != 0 ||
1630 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1631 NULL, &pool_name) != 0) {
1633 * Cache and spare devices end up here - just ignore
1636 /*printf("ZFS: can't find pool details\n");*/
1640 if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1641 NULL, &val) == 0 && val != 0) {
1646 * Create the pool if this is the first time we've seen it.
1648 spa = spa_find_by_guid(pool_guid);
1650 spa = spa_create(pool_guid, pool_name);
1654 if (pool_txg > spa->spa_txg) {
1655 spa->spa_txg = pool_txg;
1662 * Get the vdev tree and create our in-core copy of it.
1663 * If we already have a vdev with this guid, this must
1664 * be some kind of alias (overlapping slices, dangerously dedicated
1667 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1668 NULL, &guid) != 0) {
1671 vdev = vdev_find(guid);
1672 if (vdev && vdev->v_phys_read) /* Has this vdev already been inited? */
1675 if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1680 rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1685 * Add the toplevel vdev to the pool if its not already there.
1687 STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1688 if (top_vdev == pool_vdev)
1690 if (!pool_vdev && top_vdev) {
1691 top_vdev->spa = spa;
1692 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1696 * We should already have created an incomplete vdev for this
1697 * vdev. Find it and initialise it with our read proc.
1699 vdev = vdev_find(guid);
1701 vdev->v_phys_read = _read;
1702 vdev->v_read_priv = read_priv;
1703 vdev->v_state = VDEV_STATE_HEALTHY;
1705 printf("ZFS: inconsistent nvlist contents\n");
1710 * Re-evaluate top-level vdev state.
1712 vdev_set_state(top_vdev);
1715 * Ok, we are happy with the pool so far. Lets find
1716 * the best uberblock and then we can actually access
1717 * the contents of the pool.
1719 upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1720 up = (const struct uberblock *)upbuf;
1721 for (l = 0; l < VDEV_LABELS; l++) {
1722 for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1723 off = vdev_label_offset(psize, l,
1724 VDEV_UBERBLOCK_OFFSET(vdev, i));
1726 DVA_SET_OFFSET(&bp.blk_dva[0], off);
1727 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1728 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1729 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1730 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1731 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1733 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
1736 if (up->ub_magic != UBERBLOCK_MAGIC)
1738 if (up->ub_txg < spa->spa_txg)
1740 if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1741 (up->ub_txg == spa->spa_uberblock.ub_txg &&
1743 spa->spa_uberblock.ub_timestamp)) {
1744 spa->spa_uberblock = *up;
1748 zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1761 for (v = 0; v < 32; v++)
1768 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1771 zio_gbh_phys_t zio_gb;
1775 /* Artificial BP for gang block header. */
1777 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1778 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1779 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1780 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1781 for (i = 0; i < SPA_DVAS_PER_BP; i++)
1782 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1784 /* Read gang header block using the artificial BP. */
1785 if (zio_read(spa, &gbh_bp, &zio_gb))
1789 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1790 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1792 if (BP_IS_HOLE(gbp))
1794 if (zio_read(spa, gbp, pbuf))
1796 pbuf += BP_GET_PSIZE(gbp);
1799 if (zio_checksum_verify(spa, bp, buf))
1805 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1807 int cpfunc = BP_GET_COMPRESS(bp);
1808 uint64_t align, size;
1813 * Process data embedded in block pointer
1815 if (BP_IS_EMBEDDED(bp)) {
1816 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1818 size = BPE_GET_PSIZE(bp);
1819 ASSERT(size <= BPE_PAYLOAD_SIZE);
1821 if (cpfunc != ZIO_COMPRESS_OFF)
1822 pbuf = zfs_alloc(size);
1826 decode_embedded_bp_compressed(bp, pbuf);
1829 if (cpfunc != ZIO_COMPRESS_OFF) {
1830 error = zio_decompress_data(cpfunc, pbuf,
1831 size, buf, BP_GET_LSIZE(bp));
1832 zfs_free(pbuf, size);
1835 printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1842 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1843 const dva_t *dva = &bp->blk_dva[i];
1848 if (!dva->dva_word[0] && !dva->dva_word[1])
1851 vdevid = DVA_GET_VDEV(dva);
1852 offset = DVA_GET_OFFSET(dva);
1853 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1854 if (vdev->v_id == vdevid)
1857 if (!vdev || !vdev->v_read)
1860 size = BP_GET_PSIZE(bp);
1861 if (vdev->v_read == vdev_raidz_read) {
1862 align = 1ULL << vdev->v_top->v_ashift;
1863 if (P2PHASE(size, align) != 0)
1864 size = P2ROUNDUP(size, align);
1866 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1867 pbuf = zfs_alloc(size);
1871 if (DVA_GET_GANG(dva))
1872 error = zio_read_gang(spa, bp, pbuf);
1874 error = vdev->v_read(vdev, bp, pbuf, offset, size);
1876 if (cpfunc != ZIO_COMPRESS_OFF)
1877 error = zio_decompress_data(cpfunc, pbuf,
1878 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1879 else if (size != BP_GET_PSIZE(bp))
1880 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1883 zfs_free(pbuf, size);
1888 printf("ZFS: i/o error - all block copies unavailable\n");
1893 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1895 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1896 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1897 int nlevels = dnode->dn_nlevels;
1900 if (bsize > SPA_MAXBLOCKSIZE) {
1901 printf("ZFS: I/O error - blocks larger than %llu are not "
1902 "supported\n", SPA_MAXBLOCKSIZE);
1907 * Note: bsize may not be a power of two here so we need to do an
1908 * actual divide rather than a bitshift.
1910 while (buflen > 0) {
1911 uint64_t bn = offset / bsize;
1912 int boff = offset % bsize;
1914 const blkptr_t *indbp;
1917 if (bn > dnode->dn_maxblkid)
1920 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1923 indbp = dnode->dn_blkptr;
1924 for (i = 0; i < nlevels; i++) {
1926 * Copy the bp from the indirect array so that
1927 * we can re-use the scratch buffer for multi-level
1930 ibn = bn >> ((nlevels - i - 1) * ibshift);
1931 ibn &= ((1 << ibshift) - 1);
1933 if (BP_IS_HOLE(&bp)) {
1934 memset(dnode_cache_buf, 0, bsize);
1937 rc = zio_read(spa, &bp, dnode_cache_buf);
1940 indbp = (const blkptr_t *) dnode_cache_buf;
1942 dnode_cache_obj = dnode;
1943 dnode_cache_bn = bn;
1947 * The buffer contains our data block. Copy what we
1948 * need from it and loop.
1951 if (i > buflen) i = buflen;
1952 memcpy(buf, &dnode_cache_buf[boff], i);
1953 buf = ((char*) buf) + i;
1962 * Lookup a value in a microzap directory. Assumes that the zap
1963 * scratch buffer contains the directory contents.
1966 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1968 const mzap_phys_t *mz;
1969 const mzap_ent_phys_t *mze;
1974 * Microzap objects use exactly one block. Read the whole
1977 size = dnode->dn_datablkszsec * 512;
1979 mz = (const mzap_phys_t *) zap_scratch;
1980 chunks = size / MZAP_ENT_LEN - 1;
1982 for (i = 0; i < chunks; i++) {
1983 mze = &mz->mz_chunk[i];
1984 if (!strcmp(mze->mze_name, name)) {
1985 *value = mze->mze_value;
1994 * Compare a name with a zap leaf entry. Return non-zero if the name
1998 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
2001 const zap_leaf_chunk_t *nc;
2004 namelen = zc->l_entry.le_name_numints;
2006 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2008 while (namelen > 0) {
2011 if (len > ZAP_LEAF_ARRAY_BYTES)
2012 len = ZAP_LEAF_ARRAY_BYTES;
2013 if (memcmp(p, nc->l_array.la_array, len))
2017 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2024 * Extract a uint64_t value from a zap leaf entry.
2027 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2029 const zap_leaf_chunk_t *vc;
2034 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2035 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2036 value = (value << 8) | p[i];
2043 stv(int len, void *addr, uint64_t value)
2047 *(uint8_t *)addr = value;
2050 *(uint16_t *)addr = value;
2053 *(uint32_t *)addr = value;
2056 *(uint64_t *)addr = value;
2062 * Extract a array from a zap leaf entry.
2065 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2066 uint64_t integer_size, uint64_t num_integers, void *buf)
2068 uint64_t array_int_len = zc->l_entry.le_value_intlen;
2070 uint64_t *u64 = buf;
2072 int len = MIN(zc->l_entry.le_value_numints, num_integers);
2073 int chunk = zc->l_entry.le_value_chunk;
2076 if (integer_size == 8 && len == 1) {
2077 *u64 = fzap_leaf_value(zl, zc);
2082 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2085 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2086 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2087 value = (value << 8) | la->la_array[i];
2089 if (byten == array_int_len) {
2090 stv(integer_size, p, value);
2098 chunk = la->la_next;
2103 * Lookup a value in a fatzap directory. Assumes that the zap scratch
2104 * buffer contains the directory header.
2107 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2108 uint64_t integer_size, uint64_t num_integers, void *value)
2110 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2111 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2117 if (zh.zap_magic != ZAP_MAGIC)
2120 z.zap_block_shift = ilog2(bsize);
2121 z.zap_phys = (zap_phys_t *) zap_scratch;
2124 * Figure out where the pointer table is and read it in if necessary.
2126 if (zh.zap_ptrtbl.zt_blk) {
2127 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
2128 zap_scratch, bsize);
2131 ptrtbl = (uint64_t *) zap_scratch;
2133 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
2136 hash = zap_hash(zh.zap_salt, name);
2139 zl.l_bs = z.zap_block_shift;
2141 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
2142 zap_leaf_chunk_t *zc;
2144 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
2148 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2151 * Make sure this chunk matches our hash.
2153 if (zl.l_phys->l_hdr.lh_prefix_len > 0
2154 && zl.l_phys->l_hdr.lh_prefix
2155 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
2159 * Hash within the chunk to find our entry.
2161 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
2162 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
2163 h = zl.l_phys->l_hash[h];
2166 zc = &ZAP_LEAF_CHUNK(&zl, h);
2167 while (zc->l_entry.le_hash != hash) {
2168 if (zc->l_entry.le_next == 0xffff) {
2172 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
2174 if (fzap_name_equal(&zl, zc, name)) {
2175 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
2176 integer_size * num_integers)
2178 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
2186 * Lookup a name in a zap object and return its value as a uint64_t.
2189 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2190 uint64_t integer_size, uint64_t num_integers, void *value)
2194 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2196 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2200 zap_type = *(uint64_t *) zap_scratch;
2201 if (zap_type == ZBT_MICRO)
2202 return mzap_lookup(dnode, name, value);
2203 else if (zap_type == ZBT_HEADER) {
2204 return fzap_lookup(spa, dnode, name, integer_size,
2205 num_integers, value);
2207 printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
2212 * List a microzap directory. Assumes that the zap scratch buffer contains
2213 * the directory contents.
2216 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2218 const mzap_phys_t *mz;
2219 const mzap_ent_phys_t *mze;
2224 * Microzap objects use exactly one block. Read the whole
2227 size = dnode->dn_datablkszsec * 512;
2228 mz = (const mzap_phys_t *) zap_scratch;
2229 chunks = size / MZAP_ENT_LEN - 1;
2231 for (i = 0; i < chunks; i++) {
2232 mze = &mz->mz_chunk[i];
2233 if (mze->mze_name[0]) {
2234 rc = callback(mze->mze_name, mze->mze_value);
2244 * List a fatzap directory. Assumes that the zap scratch buffer contains
2245 * the directory header.
2248 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2250 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2251 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2255 if (zh.zap_magic != ZAP_MAGIC)
2258 z.zap_block_shift = ilog2(bsize);
2259 z.zap_phys = (zap_phys_t *) zap_scratch;
2262 * This assumes that the leaf blocks start at block 1. The
2263 * documentation isn't exactly clear on this.
2266 zl.l_bs = z.zap_block_shift;
2267 for (i = 0; i < zh.zap_num_leafs; i++) {
2268 off_t off = (i + 1) << zl.l_bs;
2272 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2275 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2277 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2278 zap_leaf_chunk_t *zc, *nc;
2281 zc = &ZAP_LEAF_CHUNK(&zl, j);
2282 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2284 namelen = zc->l_entry.le_name_numints;
2285 if (namelen > sizeof(name))
2286 namelen = sizeof(name);
2289 * Paste the name back together.
2291 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2293 while (namelen > 0) {
2296 if (len > ZAP_LEAF_ARRAY_BYTES)
2297 len = ZAP_LEAF_ARRAY_BYTES;
2298 memcpy(p, nc->l_array.la_array, len);
2301 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2305 * Assume the first eight bytes of the value are
2308 value = fzap_leaf_value(&zl, zc);
2310 //printf("%s 0x%jx\n", name, (uintmax_t)value);
2311 rc = callback((const char *)name, value);
2320 static int zfs_printf(const char *name, uint64_t value __unused)
2323 printf("%s\n", name);
2329 * List a zap directory.
2332 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2335 size_t size = dnode->dn_datablkszsec * 512;
2337 if (dnode_read(spa, dnode, 0, zap_scratch, size))
2340 zap_type = *(uint64_t *) zap_scratch;
2341 if (zap_type == ZBT_MICRO)
2342 return mzap_list(dnode, zfs_printf);
2344 return fzap_list(spa, dnode, zfs_printf);
2348 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
2352 offset = objnum * sizeof(dnode_phys_t);
2353 return dnode_read(spa, &os->os_meta_dnode, offset,
2354 dnode, sizeof(dnode_phys_t));
2358 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2360 const mzap_phys_t *mz;
2361 const mzap_ent_phys_t *mze;
2366 * Microzap objects use exactly one block. Read the whole
2369 size = dnode->dn_datablkszsec * 512;
2371 mz = (const mzap_phys_t *) zap_scratch;
2372 chunks = size / MZAP_ENT_LEN - 1;
2374 for (i = 0; i < chunks; i++) {
2375 mze = &mz->mz_chunk[i];
2376 if (value == mze->mze_value) {
2377 strcpy(name, mze->mze_name);
2386 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2389 const zap_leaf_chunk_t *nc;
2392 namelen = zc->l_entry.le_name_numints;
2394 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2396 while (namelen > 0) {
2399 if (len > ZAP_LEAF_ARRAY_BYTES)
2400 len = ZAP_LEAF_ARRAY_BYTES;
2401 memcpy(p, nc->l_array.la_array, len);
2404 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2411 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2413 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2414 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
2418 if (zh.zap_magic != ZAP_MAGIC)
2421 z.zap_block_shift = ilog2(bsize);
2422 z.zap_phys = (zap_phys_t *) zap_scratch;
2425 * This assumes that the leaf blocks start at block 1. The
2426 * documentation isn't exactly clear on this.
2429 zl.l_bs = z.zap_block_shift;
2430 for (i = 0; i < zh.zap_num_leafs; i++) {
2431 off_t off = (i + 1) << zl.l_bs;
2433 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2436 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
2438 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2439 zap_leaf_chunk_t *zc;
2441 zc = &ZAP_LEAF_CHUNK(&zl, j);
2442 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2444 if (zc->l_entry.le_value_intlen != 8 ||
2445 zc->l_entry.le_value_numints != 1)
2448 if (fzap_leaf_value(&zl, zc) == value) {
2449 fzap_name_copy(&zl, zc, name);
2459 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
2463 size_t size = dnode->dn_datablkszsec * 512;
2465 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2469 zap_type = *(uint64_t *) zap_scratch;
2470 if (zap_type == ZBT_MICRO)
2471 return mzap_rlookup(spa, dnode, name, value);
2473 return fzap_rlookup(spa, dnode, name, value);
2477 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2480 char component[256];
2481 uint64_t dir_obj, parent_obj, child_dir_zapobj;
2482 dnode_phys_t child_dir_zap, dataset, dir, parent;
2484 dsl_dataset_phys_t *ds;
2488 p = &name[sizeof(name) - 1];
2491 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2492 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2495 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2496 dir_obj = ds->ds_dir_obj;
2499 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
2501 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2503 /* Actual loop condition. */
2504 parent_obj = dd->dd_parent_obj;
2505 if (parent_obj == 0)
2508 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
2510 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2511 child_dir_zapobj = dd->dd_child_dir_zapobj;
2512 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2514 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2517 len = strlen(component);
2519 memcpy(p, component, len);
2523 /* Actual loop iteration. */
2524 dir_obj = parent_obj;
2535 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2538 uint64_t dir_obj, child_dir_zapobj;
2539 dnode_phys_t child_dir_zap, dir;
2543 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
2545 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2551 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
2553 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2557 /* Actual loop condition #1. */
2563 memcpy(element, p, q - p);
2564 element[q - p] = '\0';
2571 child_dir_zapobj = dd->dd_child_dir_zapobj;
2572 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2575 /* Actual loop condition #2. */
2576 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2581 *objnum = dd->dd_head_dataset_obj;
2587 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2589 uint64_t dir_obj, child_dir_zapobj;
2590 dnode_phys_t child_dir_zap, dir, dataset;
2591 dsl_dataset_phys_t *ds;
2594 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2595 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2598 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2599 dir_obj = ds->ds_dir_obj;
2601 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2602 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2605 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2607 child_dir_zapobj = dd->dd_child_dir_zapobj;
2608 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2609 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2613 return (zap_list(spa, &child_dir_zap) != 0);
2617 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2619 uint64_t dir_obj, child_dir_zapobj, zap_type;
2620 dnode_phys_t child_dir_zap, dir, dataset;
2621 dsl_dataset_phys_t *ds;
2625 err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2627 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2630 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2631 dir_obj = ds->ds_dir_obj;
2633 err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2635 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2638 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2640 child_dir_zapobj = dd->dd_child_dir_zapobj;
2641 err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2643 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2647 err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
2651 zap_type = *(uint64_t *) zap_scratch;
2652 if (zap_type == ZBT_MICRO)
2653 return mzap_list(&child_dir_zap, callback);
2655 return fzap_list(spa, &child_dir_zap, callback);
2660 * Find the object set given the object number of its dataset object
2661 * and return its details in *objset
2664 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2666 dnode_phys_t dataset;
2667 dsl_dataset_phys_t *ds;
2669 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2670 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2674 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2675 if (zio_read(spa, &ds->ds_bp, objset)) {
2676 printf("ZFS: can't read object set for dataset %ju\n",
2685 * Find the object set pointed to by the BOOTFS property or the root
2686 * dataset if there is none and return its details in *objset
2689 zfs_get_root(const spa_t *spa, uint64_t *objid)
2691 dnode_phys_t dir, propdir;
2692 uint64_t props, bootfs, root;
2697 * Start with the MOS directory object.
2699 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2700 printf("ZFS: can't read MOS object directory\n");
2705 * Lookup the pool_props and see if we can find a bootfs.
2707 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, sizeof (props), 1, &props) == 0
2708 && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2709 && zap_lookup(spa, &propdir, "bootfs", sizeof (bootfs), 1, &bootfs) == 0
2716 * Lookup the root dataset directory
2718 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
2719 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2720 printf("ZFS: can't find root dsl_dir\n");
2725 * Use the information from the dataset directory's bonus buffer
2726 * to find the dataset object and from that the object set itself.
2728 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2729 *objid = dd->dd_head_dataset_obj;
2734 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
2740 * Find the root object set if not explicitly provided
2742 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2743 printf("ZFS: can't find root filesystem\n");
2747 if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
2748 printf("ZFS: can't open root filesystem\n");
2752 mount->rootobj = rootobj;
2758 * callback function for feature name checks.
2761 check_feature(const char *name, uint64_t value)
2767 if (name[0] == '\0')
2770 for (i = 0; features_for_read[i] != NULL; i++) {
2771 if (strcmp(name, features_for_read[i]) == 0)
2774 printf("ZFS: unsupported feature: %s\n", name);
2779 * Checks whether the MOS features that are active are supported.
2782 check_mos_features(const spa_t *spa)
2785 uint64_t objnum, zap_type;
2789 if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2792 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2793 sizeof (objnum), 1, &objnum)) != 0) {
2795 * It is older pool without features. As we have already
2796 * tested the label, just return without raising the error.
2801 if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2804 if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2807 size = dir.dn_datablkszsec * 512;
2808 if (dnode_read(spa, &dir, 0, zap_scratch, size))
2811 zap_type = *(uint64_t *) zap_scratch;
2812 if (zap_type == ZBT_MICRO)
2813 rc = mzap_list(&dir, check_feature);
2815 rc = fzap_list(spa, &dir, check_feature);
2821 load_nvlist(spa_t *spa, uint64_t obj, unsigned char **value)
2829 if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
2831 if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
2832 dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
2836 if (dir.dn_bonuslen != sizeof (uint64_t))
2839 size = *(uint64_t *)DN_BONUS(&dir);
2844 rc = dnode_read(spa, &dir, 0, nv, size);
2855 zfs_spa_init(spa_t *spa)
2858 uint64_t config_object;
2859 unsigned char *nvlist;
2861 const unsigned char *nv;
2864 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2865 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2868 if (spa->spa_mos.os_type != DMU_OST_META) {
2869 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2873 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
2875 printf("ZFS: failed to read pool %s directory object\n",
2879 /* this is allowed to fail, older pools do not have salt */
2880 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
2881 sizeof (spa->spa_cksum_salt.zcs_bytes),
2882 spa->spa_cksum_salt.zcs_bytes);
2884 rc = check_mos_features(spa);
2886 printf("ZFS: pool %s is not supported\n", spa->spa_name);
2890 rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
2891 sizeof (config_object), 1, &config_object);
2893 printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
2896 rc = load_nvlist(spa, config_object, &nvlist);
2900 /* Update vdevs from MOS config. */
2901 if (nvlist_find(nvlist + 4, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
2907 if (nvlist_find(nv, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
2909 printf("ZFS: can't find vdev details\n");
2913 if (strcmp(type, VDEV_TYPE_ROOT) != 0) {
2918 rc = nvlist_find(nv, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
2923 for (int i = 0; i < nkids; i++) {
2924 vdev_t *vd, *prev, *kid = NULL;
2925 rc = vdev_init_from_nvlist(nv, NULL, &kid, 0);
2927 printf("vdev_init_from_nvlist: %d\n", rc);
2932 STAILQ_FOREACH(vd, &spa->spa_vdevs, v_childlink) {
2933 /* Already present? */
2934 if (kid->v_id == vd->v_id) {
2938 if (vd->v_id > kid->v_id) {
2940 STAILQ_INSERT_HEAD(&spa->spa_vdevs,
2943 STAILQ_INSERT_AFTER(&spa->spa_vdevs,
2944 prev, kid, v_childlink);
2952 STAILQ_INSERT_TAIL(&spa->spa_vdevs, kid, v_childlink);
2953 nv = nvlist_next(nv);
2962 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2965 if (dn->dn_bonustype != DMU_OT_SA) {
2966 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2968 sb->st_mode = zp->zp_mode;
2969 sb->st_uid = zp->zp_uid;
2970 sb->st_gid = zp->zp_gid;
2971 sb->st_size = zp->zp_size;
2973 sa_hdr_phys_t *sahdrp;
2978 if (dn->dn_bonuslen != 0)
2979 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2981 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2982 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2985 size = BP_GET_LSIZE(bp);
2986 buf = zfs_alloc(size);
2987 error = zio_read(spa, bp, buf);
2989 zfs_free(buf, size);
2997 hdrsize = SA_HDR_SIZE(sahdrp);
2998 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3000 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3002 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3004 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3007 zfs_free(buf, size);
3014 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3018 if (dn->dn_bonustype == DMU_OT_SA) {
3019 sa_hdr_phys_t *sahdrp = NULL;
3025 if (dn->dn_bonuslen != 0)
3026 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3030 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3032 bp = DN_SPILL_BLKPTR(dn);
3034 size = BP_GET_LSIZE(bp);
3035 buf = zfs_alloc(size);
3036 rc = zio_read(spa, bp, buf);
3038 zfs_free(buf, size);
3043 hdrsize = SA_HDR_SIZE(sahdrp);
3044 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3045 memcpy(path, p, psize);
3047 zfs_free(buf, size);
3051 * Second test is purely to silence bogus compiler
3052 * warning about accessing past the end of dn_bonus.
3054 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3055 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3056 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3058 rc = dnode_read(spa, dn, 0, path, psize);
3065 STAILQ_ENTRY(obj_list) entry;
3069 * Lookup a file and return its dnode.
3072 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3081 int symlinks_followed = 0;
3083 struct obj_list *entry, *tentry;
3084 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3087 if (mount->objset.os_type != DMU_OST_ZFS) {
3088 printf("ZFS: unexpected object set type %ju\n",
3089 (uintmax_t)mount->objset.os_type);
3093 if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3097 * Get the root directory dnode.
3099 rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3105 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof (objnum), 1, &objnum);
3110 entry->objnum = objnum;
3111 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3113 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3119 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3128 while (*q != '\0' && *q != '/')
3132 if (p + 1 == q && p[0] == '.') {
3137 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3139 if (STAILQ_FIRST(&on_cache) ==
3140 STAILQ_LAST(&on_cache, obj_list, entry)) {
3144 entry = STAILQ_FIRST(&on_cache);
3145 STAILQ_REMOVE_HEAD(&on_cache, entry);
3147 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3150 if (q - p + 1 > sizeof(element)) {
3154 memcpy(element, p, q - p);
3158 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3160 if (!S_ISDIR(sb.st_mode)) {
3165 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3168 objnum = ZFS_DIRENT_OBJ(objnum);
3170 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3174 entry->objnum = objnum;
3175 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3176 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3181 * Check for symlink.
3183 rc = zfs_dnode_stat(spa, &dn, &sb);
3186 if (S_ISLNK(sb.st_mode)) {
3187 if (symlinks_followed > 10) {
3191 symlinks_followed++;
3194 * Read the link value and copy the tail of our
3195 * current path onto the end.
3197 if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3201 strcpy(&path[sb.st_size], p);
3203 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3208 * Restart with the new path, starting either at
3209 * the root or at the parent depending whether or
3210 * not the link is relative.
3214 while (STAILQ_FIRST(&on_cache) !=
3215 STAILQ_LAST(&on_cache, obj_list, entry)) {
3216 entry = STAILQ_FIRST(&on_cache);
3217 STAILQ_REMOVE_HEAD(&on_cache, entry);
3221 entry = STAILQ_FIRST(&on_cache);
3222 STAILQ_REMOVE_HEAD(&on_cache, entry);
3225 objnum = (STAILQ_FIRST(&on_cache))->objnum;
3231 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)