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.
38 * List of all vdevs, chained through v_alllink.
40 static vdev_list_t zfs_vdevs;
43 * List of all pools, chained through spa_link.
45 static spa_list_t zfs_pools;
47 static uint64_t zfs_crc64_table[256];
48 static const dnode_phys_t *dnode_cache_obj = 0;
49 static uint64_t dnode_cache_bn;
50 static char *dnode_cache_buf;
51 static char *zap_scratch;
52 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
54 #define TEMP_SIZE (1024 * 1024)
56 static int zio_read(spa_t *spa, const blkptr_t *bp, void *buf);
61 STAILQ_INIT(&zfs_vdevs);
62 STAILQ_INIT(&zfs_pools);
64 zfs_temp_buf = malloc(TEMP_SIZE);
65 zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
66 zfs_temp_ptr = zfs_temp_buf;
67 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
68 zap_scratch = malloc(SPA_MAXBLOCKSIZE);
74 zfs_alloc_temp(size_t sz)
78 if (zfs_temp_ptr + sz > zfs_temp_end) {
79 printf("ZFS: out of temporary buffer space\n");
92 zfs_temp_ptr = zfs_temp_buf;
96 xdr_int(const unsigned char **xdr, int *ip)
98 *ip = ((*xdr)[0] << 24)
107 xdr_u_int(const unsigned char **xdr, u_int *ip)
109 *ip = ((*xdr)[0] << 24)
118 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
124 *lp = (((uint64_t) hi) << 32) | lo;
129 nvlist_find(const unsigned char *nvlist, const char *name, int type,
130 int* elementsp, void *valuep)
132 const unsigned char *p, *pair;
134 int encoded_size, decoded_size;
141 xdr_int(&p, &encoded_size);
142 xdr_int(&p, &decoded_size);
143 while (encoded_size && decoded_size) {
144 int namelen, pairtype, elements;
145 const char *pairname;
147 xdr_int(&p, &namelen);
148 pairname = (const char*) p;
149 p += roundup(namelen, 4);
150 xdr_int(&p, &pairtype);
152 if (!memcmp(name, pairname, namelen) && type == pairtype) {
153 xdr_int(&p, &elements);
155 *elementsp = elements;
156 if (type == DATA_TYPE_UINT64) {
157 xdr_uint64_t(&p, (uint64_t *) valuep);
159 } else if (type == DATA_TYPE_STRING) {
162 (*(const char**) valuep) = (const char*) p;
164 } else if (type == DATA_TYPE_NVLIST
165 || type == DATA_TYPE_NVLIST_ARRAY) {
166 (*(const unsigned char**) valuep) =
167 (const unsigned char*) p;
174 * Not the pair we are looking for, skip to the next one.
176 p = pair + encoded_size;
180 xdr_int(&p, &encoded_size);
181 xdr_int(&p, &decoded_size);
188 * Return the next nvlist in an nvlist array.
190 static const unsigned char *
191 nvlist_next(const unsigned char *nvlist)
193 const unsigned char *p, *pair;
195 int encoded_size, decoded_size;
202 xdr_int(&p, &encoded_size);
203 xdr_int(&p, &decoded_size);
204 while (encoded_size && decoded_size) {
205 p = pair + encoded_size;
208 xdr_int(&p, &encoded_size);
209 xdr_int(&p, &decoded_size);
217 static const unsigned char *
218 nvlist_print(const unsigned char *nvlist, unsigned int indent)
220 static const char* typenames[] = {
231 "DATA_TYPE_BYTE_ARRAY",
232 "DATA_TYPE_INT16_ARRAY",
233 "DATA_TYPE_UINT16_ARRAY",
234 "DATA_TYPE_INT32_ARRAY",
235 "DATA_TYPE_UINT32_ARRAY",
236 "DATA_TYPE_INT64_ARRAY",
237 "DATA_TYPE_UINT64_ARRAY",
238 "DATA_TYPE_STRING_ARRAY",
241 "DATA_TYPE_NVLIST_ARRAY",
242 "DATA_TYPE_BOOLEAN_VALUE",
245 "DATA_TYPE_BOOLEAN_ARRAY",
246 "DATA_TYPE_INT8_ARRAY",
247 "DATA_TYPE_UINT8_ARRAY"
251 const unsigned char *p, *pair;
253 int encoded_size, decoded_size;
260 xdr_int(&p, &encoded_size);
261 xdr_int(&p, &decoded_size);
262 while (encoded_size && decoded_size) {
263 int namelen, pairtype, elements;
264 const char *pairname;
266 xdr_int(&p, &namelen);
267 pairname = (const char*) p;
268 p += roundup(namelen, 4);
269 xdr_int(&p, &pairtype);
271 for (i = 0; i < indent; i++)
273 printf("%s %s", typenames[pairtype], pairname);
275 xdr_int(&p, &elements);
277 case DATA_TYPE_UINT64: {
279 xdr_uint64_t(&p, &val);
280 printf(" = 0x%llx\n", val);
284 case DATA_TYPE_STRING: {
287 printf(" = \"%s\"\n", p);
291 case DATA_TYPE_NVLIST:
293 nvlist_print(p, indent + 1);
296 case DATA_TYPE_NVLIST_ARRAY:
297 for (j = 0; j < elements; j++) {
299 p = nvlist_print(p, indent + 1);
300 if (j != elements - 1) {
301 for (i = 0; i < indent; i++)
303 printf("%s %s", typenames[pairtype], pairname);
312 p = pair + encoded_size;
315 xdr_int(&p, &encoded_size);
316 xdr_int(&p, &decoded_size);
325 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
326 off_t offset, size_t size)
331 if (!vdev->v_phys_read)
335 psize = BP_GET_PSIZE(bp);
340 /*printf("ZFS: reading %d bytes at 0x%llx to %p\n", psize, offset, buf);*/
341 rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
344 if (bp && zio_checksum_error(bp, buf))
351 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
352 off_t offset, size_t bytes)
355 return (vdev_read_phys(vdev, bp, buf,
356 offset + VDEV_LABEL_START_SIZE, bytes));
361 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
362 off_t offset, size_t bytes)
368 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
369 if (kid->v_state != VDEV_STATE_HEALTHY)
371 rc = kid->v_read(kid, bp, buf, offset, bytes);
380 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
381 off_t offset, size_t bytes)
386 * Here we should have two kids:
387 * First one which is the one we are replacing and we can trust
388 * only this one to have valid data, but it might not be present.
389 * Second one is that one we are replacing with. It is most likely
390 * healthy, but we can't trust it has needed data, so we won't use it.
392 kid = STAILQ_FIRST(&vdev->v_children);
395 if (kid->v_state != VDEV_STATE_HEALTHY)
397 return (kid->v_read(kid, bp, buf, offset, bytes));
401 vdev_find(uint64_t guid)
405 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
406 if (vdev->v_guid == guid)
413 vdev_create(uint64_t guid, vdev_read_t *read)
417 vdev = malloc(sizeof(vdev_t));
418 memset(vdev, 0, sizeof(vdev_t));
419 STAILQ_INIT(&vdev->v_children);
421 vdev->v_state = VDEV_STATE_OFFLINE;
423 vdev->v_phys_read = 0;
424 vdev->v_read_priv = 0;
425 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
431 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t **vdevp, int is_newer)
434 uint64_t guid, id, ashift, nparity;
438 const unsigned char *kids;
439 int nkids, i, is_new;
440 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
442 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID,
443 DATA_TYPE_UINT64, 0, &guid)
444 || nvlist_find(nvlist, ZPOOL_CONFIG_ID,
445 DATA_TYPE_UINT64, 0, &id)
446 || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE,
447 DATA_TYPE_STRING, 0, &type)) {
448 printf("ZFS: can't find vdev details\n");
452 if (strcmp(type, VDEV_TYPE_MIRROR)
453 && strcmp(type, VDEV_TYPE_DISK)
454 && strcmp(type, VDEV_TYPE_RAIDZ)
455 && strcmp(type, VDEV_TYPE_REPLACING)) {
456 printf("ZFS: can only boot from disk, mirror or raidz vdevs\n");
460 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
462 nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, 0,
464 nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, 0,
466 nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, 0,
468 nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 0,
470 nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 0,
473 vdev = vdev_find(guid);
477 if (!strcmp(type, VDEV_TYPE_MIRROR))
478 vdev = vdev_create(guid, vdev_mirror_read);
479 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
480 vdev = vdev_create(guid, vdev_raidz_read);
481 else if (!strcmp(type, VDEV_TYPE_REPLACING))
482 vdev = vdev_create(guid, vdev_replacing_read);
484 vdev = vdev_create(guid, vdev_disk_read);
487 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
488 DATA_TYPE_UINT64, 0, &ashift) == 0)
489 vdev->v_ashift = ashift;
492 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
493 DATA_TYPE_UINT64, 0, &nparity) == 0)
494 vdev->v_nparity = nparity;
497 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
498 DATA_TYPE_STRING, 0, &path) == 0) {
499 if (strncmp(path, "/dev/", 5) == 0)
501 vdev->v_name = strdup(path);
503 if (!strcmp(type, "raidz")) {
504 if (vdev->v_nparity == 1)
505 vdev->v_name = "raidz1";
507 vdev->v_name = "raidz2";
509 vdev->v_name = strdup(type);
516 if (is_new || is_newer) {
518 * This is either new vdev or we've already seen this vdev,
519 * but from an older vdev label, so let's refresh its state
520 * from the newer label.
523 vdev->v_state = VDEV_STATE_OFFLINE;
525 vdev->v_state = VDEV_STATE_REMOVED;
527 vdev->v_state = VDEV_STATE_FAULTED;
528 else if (is_degraded)
529 vdev->v_state = VDEV_STATE_DEGRADED;
530 else if (isnt_present)
531 vdev->v_state = VDEV_STATE_CANT_OPEN;
533 vdev->v_state = VDEV_STATE_HEALTHY;
536 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN,
537 DATA_TYPE_NVLIST_ARRAY, &nkids, &kids);
539 * Its ok if we don't have any kids.
542 vdev->v_nchildren = nkids;
543 for (i = 0; i < nkids; i++) {
544 rc = vdev_init_from_nvlist(kids, &kid, is_newer);
548 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
550 kids = nvlist_next(kids);
553 vdev->v_nchildren = 0;
562 vdev_set_state(vdev_t *vdev)
569 * A mirror or raidz is healthy if all its kids are healthy. A
570 * mirror is degraded if any of its kids is healthy; a raidz
571 * is degraded if at most nparity kids are offline.
573 if (STAILQ_FIRST(&vdev->v_children)) {
576 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
577 if (kid->v_state == VDEV_STATE_HEALTHY)
583 vdev->v_state = VDEV_STATE_HEALTHY;
585 if (vdev->v_read == vdev_mirror_read) {
587 vdev->v_state = VDEV_STATE_DEGRADED;
589 vdev->v_state = VDEV_STATE_OFFLINE;
591 } else if (vdev->v_read == vdev_raidz_read) {
592 if (bad_kids > vdev->v_nparity) {
593 vdev->v_state = VDEV_STATE_OFFLINE;
595 vdev->v_state = VDEV_STATE_DEGRADED;
603 spa_find_by_guid(uint64_t guid)
607 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
608 if (spa->spa_guid == guid)
617 spa_find_by_name(const char *name)
621 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
622 if (!strcmp(spa->spa_name, name))
631 spa_create(uint64_t guid)
635 spa = malloc(sizeof(spa_t));
636 memset(spa, 0, sizeof(spa_t));
637 STAILQ_INIT(&spa->spa_vdevs);
638 spa->spa_guid = guid;
639 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
645 state_name(vdev_state_t state)
647 static const char* names[] = {
662 #define pager_printf printf
667 pager_printf(const char *fmt, ...)
673 vsprintf(line, fmt, args);
680 #define STATUS_FORMAT " %s %s\n"
683 print_state(int indent, const char *name, vdev_state_t state)
689 for (i = 0; i < indent; i++)
692 pager_printf(STATUS_FORMAT, buf, state_name(state));
697 vdev_status(vdev_t *vdev, int indent)
700 print_state(indent, vdev->v_name, vdev->v_state);
702 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
703 vdev_status(kid, indent + 1);
708 spa_status(spa_t *spa)
711 int good_kids, bad_kids, degraded_kids;
714 pager_printf(" pool: %s\n", spa->spa_name);
715 pager_printf("config:\n\n");
716 pager_printf(STATUS_FORMAT, "NAME", "STATE");
721 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
722 if (vdev->v_state == VDEV_STATE_HEALTHY)
724 else if (vdev->v_state == VDEV_STATE_DEGRADED)
730 state = VDEV_STATE_CLOSED;
731 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
732 state = VDEV_STATE_HEALTHY;
733 else if ((good_kids + degraded_kids) > 0)
734 state = VDEV_STATE_DEGRADED;
736 print_state(0, spa->spa_name, state);
737 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
738 vdev_status(vdev, 1);
748 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
757 vdev_probe(vdev_phys_read_t *read, void *read_priv, spa_t **spap)
760 vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
762 vdev_t *vdev, *top_vdev, *pool_vdev;
765 const unsigned char *nvlist;
768 uint64_t pool_txg, pool_guid;
770 const char *pool_name;
771 const unsigned char *vdevs;
774 const struct uberblock *up;
777 * Load the vdev label and figure out which
778 * uberblock is most current.
780 memset(&vtmp, 0, sizeof(vtmp));
781 vtmp.v_phys_read = read;
782 vtmp.v_read_priv = read_priv;
783 off = offsetof(vdev_label_t, vl_vdev_phys);
785 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
786 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
787 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
788 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
789 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
790 if (vdev_read_phys(&vtmp, &bp, vdev_label, off, 0))
793 if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR) {
797 nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
799 if (nvlist_find(nvlist,
800 ZPOOL_CONFIG_VERSION,
801 DATA_TYPE_UINT64, 0, &val)) {
805 if (val > SPA_VERSION) {
806 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
807 (unsigned) val, (unsigned) SPA_VERSION);
811 if (nvlist_find(nvlist,
812 ZPOOL_CONFIG_POOL_STATE,
813 DATA_TYPE_UINT64, 0, &val)) {
818 if (val != POOL_STATE_ACTIVE) {
820 * Don't print a message here. If we happen to reboot
821 * while where is an exported pool around, we don't
822 * need a cascade of confusing messages during boot.
824 /*printf("ZFS: pool is not active\n");*/
829 if (nvlist_find(nvlist,
830 ZPOOL_CONFIG_POOL_TXG,
831 DATA_TYPE_UINT64, 0, &pool_txg)
832 || nvlist_find(nvlist,
833 ZPOOL_CONFIG_POOL_GUID,
834 DATA_TYPE_UINT64, 0, &pool_guid)
835 || nvlist_find(nvlist,
836 ZPOOL_CONFIG_POOL_NAME,
837 DATA_TYPE_STRING, 0, &pool_name)) {
839 * Cache and spare devices end up here - just ignore
842 /*printf("ZFS: can't find pool details\n");*/
847 (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, 0,
853 * Create the pool if this is the first time we've seen it.
855 spa = spa_find_by_guid(pool_guid);
857 spa = spa_create(pool_guid);
858 spa->spa_name = strdup(pool_name);
860 if (pool_txg > spa->spa_txg) {
861 spa->spa_txg = pool_txg;
867 * Get the vdev tree and create our in-core copy of it.
868 * If we already have a vdev with this guid, this must
869 * be some kind of alias (overlapping slices, dangerously dedicated
872 if (nvlist_find(nvlist,
874 DATA_TYPE_UINT64, 0, &guid)) {
877 vdev = vdev_find(guid);
878 if (vdev && vdev->v_phys_read) /* Has this vdev already been inited? */
881 if (nvlist_find(nvlist,
882 ZPOOL_CONFIG_VDEV_TREE,
883 DATA_TYPE_NVLIST, 0, &vdevs)) {
887 rc = vdev_init_from_nvlist(vdevs, &top_vdev, is_newer);
892 * Add the toplevel vdev to the pool if its not already there.
894 STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
895 if (top_vdev == pool_vdev)
897 if (!pool_vdev && top_vdev)
898 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
901 * We should already have created an incomplete vdev for this
902 * vdev. Find it and initialise it with our read proc.
904 vdev = vdev_find(guid);
906 vdev->v_phys_read = read;
907 vdev->v_read_priv = read_priv;
909 printf("ZFS: inconsistent nvlist contents\n");
914 * Re-evaluate top-level vdev state.
916 vdev_set_state(top_vdev);
919 * Ok, we are happy with the pool so far. Lets find
920 * the best uberblock and then we can actually access
921 * the contents of the pool.
924 i < VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT;
926 off = offsetof(vdev_label_t, vl_uberblock);
927 off += i << UBERBLOCK_SHIFT;
929 DVA_SET_OFFSET(&bp.blk_dva[0], off);
930 BP_SET_LSIZE(&bp, 1 << UBERBLOCK_SHIFT);
931 BP_SET_PSIZE(&bp, 1 << UBERBLOCK_SHIFT);
932 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
933 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
934 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
935 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
938 up = (const struct uberblock *) upbuf;
939 if (up->ub_magic != UBERBLOCK_MAGIC)
941 if (up->ub_txg < spa->spa_txg)
943 if (up->ub_txg > spa->spa_uberblock.ub_txg) {
944 spa->spa_uberblock = *up;
945 } else if (up->ub_txg == spa->spa_uberblock.ub_txg) {
946 if (up->ub_timestamp > spa->spa_uberblock.ub_timestamp)
947 spa->spa_uberblock = *up;
961 for (v = 0; v < 32; v++)
968 zio_read_gang(spa_t *spa, const blkptr_t *bp, const dva_t *dva, void *buf)
970 zio_gbh_phys_t zio_gb;
976 vdevid = DVA_GET_VDEV(dva);
977 offset = DVA_GET_OFFSET(dva);
978 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink)
979 if (vdev->v_id == vdevid)
981 if (!vdev || !vdev->v_read)
983 if (vdev->v_read(vdev, NULL, &zio_gb, offset, SPA_GANGBLOCKSIZE))
986 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
987 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
991 if (zio_read(spa, gbp, buf))
993 buf = (char*)buf + BP_GET_PSIZE(gbp);
1000 zio_read(spa_t *spa, const blkptr_t *bp, void *buf)
1002 int cpfunc = BP_GET_COMPRESS(bp);
1003 size_t lsize = BP_GET_LSIZE(bp);
1004 size_t psize = BP_GET_PSIZE(bp);
1009 if (cpfunc != ZIO_COMPRESS_OFF)
1010 pbuf = zfs_alloc_temp(psize);
1014 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1015 const dva_t *dva = &bp->blk_dva[i];
1020 if (!dva->dva_word[0] && !dva->dva_word[1])
1023 if (DVA_GET_GANG(dva)) {
1024 if (zio_read_gang(spa, bp, dva, buf))
1027 vdevid = DVA_GET_VDEV(dva);
1028 offset = DVA_GET_OFFSET(dva);
1029 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink)
1030 if (vdev->v_id == vdevid)
1032 if (!vdev || !vdev->v_read) {
1035 if (vdev->v_read(vdev, bp, pbuf, offset, psize))
1038 if (cpfunc != ZIO_COMPRESS_OFF) {
1039 if (zio_decompress_data(cpfunc, pbuf, psize,
1047 printf("ZFS: i/o error - all block copies unavailable\n");
1053 dnode_read(spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1055 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1056 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1057 int nlevels = dnode->dn_nlevels;
1061 * Note: bsize may not be a power of two here so we need to do an
1062 * actual divide rather than a bitshift.
1064 while (buflen > 0) {
1065 uint64_t bn = offset / bsize;
1066 int boff = offset % bsize;
1068 const blkptr_t *indbp;
1071 if (bn > dnode->dn_maxblkid)
1074 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1077 indbp = dnode->dn_blkptr;
1078 for (i = 0; i < nlevels; i++) {
1080 * Copy the bp from the indirect array so that
1081 * we can re-use the scratch buffer for multi-level
1084 ibn = bn >> ((nlevels - i - 1) * ibshift);
1085 ibn &= ((1 << ibshift) - 1);
1087 rc = zio_read(spa, &bp, dnode_cache_buf);
1090 indbp = (const blkptr_t *) dnode_cache_buf;
1092 dnode_cache_obj = dnode;
1093 dnode_cache_bn = bn;
1097 * The buffer contains our data block. Copy what we
1098 * need from it and loop.
1101 if (i > buflen) i = buflen;
1102 memcpy(buf, &dnode_cache_buf[boff], i);
1103 buf = ((char*) buf) + i;
1112 * Lookup a value in a microzap directory. Assumes that the zap
1113 * scratch buffer contains the directory contents.
1116 mzap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1118 const mzap_phys_t *mz;
1119 const mzap_ent_phys_t *mze;
1124 * Microzap objects use exactly one block. Read the whole
1127 size = dnode->dn_datablkszsec * 512;
1129 mz = (const mzap_phys_t *) zap_scratch;
1130 chunks = size / MZAP_ENT_LEN - 1;
1132 for (i = 0; i < chunks; i++) {
1133 mze = &mz->mz_chunk[i];
1134 if (!strcmp(mze->mze_name, name)) {
1135 *value = mze->mze_value;
1144 * Compare a name with a zap leaf entry. Return non-zero if the name
1148 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1151 const zap_leaf_chunk_t *nc;
1154 namelen = zc->l_entry.le_name_length;
1156 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1158 while (namelen > 0) {
1161 if (len > ZAP_LEAF_ARRAY_BYTES)
1162 len = ZAP_LEAF_ARRAY_BYTES;
1163 if (memcmp(p, nc->l_array.la_array, len))
1167 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1174 * Extract a uint64_t value from a zap leaf entry.
1177 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1179 const zap_leaf_chunk_t *vc;
1184 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1185 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1186 value = (value << 8) | p[i];
1193 * Lookup a value in a fatzap directory. Assumes that the zap scratch
1194 * buffer contains the directory header.
1197 fzap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1199 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1200 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1206 if (zh.zap_magic != ZAP_MAGIC)
1209 z.zap_block_shift = ilog2(bsize);
1210 z.zap_phys = (zap_phys_t *) zap_scratch;
1213 * Figure out where the pointer table is and read it in if necessary.
1215 if (zh.zap_ptrtbl.zt_blk) {
1216 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1217 zap_scratch, bsize);
1220 ptrtbl = (uint64_t *) zap_scratch;
1222 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1225 hash = zap_hash(zh.zap_salt, name);
1228 zl.l_bs = z.zap_block_shift;
1230 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1231 zap_leaf_chunk_t *zc;
1233 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1237 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1240 * Make sure this chunk matches our hash.
1242 if (zl.l_phys->l_hdr.lh_prefix_len > 0
1243 && zl.l_phys->l_hdr.lh_prefix
1244 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1248 * Hash within the chunk to find our entry.
1250 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1251 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1252 h = zl.l_phys->l_hash[h];
1255 zc = &ZAP_LEAF_CHUNK(&zl, h);
1256 while (zc->l_entry.le_hash != hash) {
1257 if (zc->l_entry.le_next == 0xffff) {
1261 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1263 if (fzap_name_equal(&zl, zc, name)) {
1264 *value = fzap_leaf_value(&zl, zc);
1272 * Lookup a name in a zap object and return its value as a uint64_t.
1275 zap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1279 size_t size = dnode->dn_datablkszsec * 512;
1281 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1285 zap_type = *(uint64_t *) zap_scratch;
1286 if (zap_type == ZBT_MICRO)
1287 return mzap_lookup(spa, dnode, name, value);
1289 return fzap_lookup(spa, dnode, name, value);
1295 * List a microzap directory. Assumes that the zap scratch buffer contains
1296 * the directory contents.
1299 mzap_list(spa_t *spa, const dnode_phys_t *dnode)
1301 const mzap_phys_t *mz;
1302 const mzap_ent_phys_t *mze;
1307 * Microzap objects use exactly one block. Read the whole
1310 size = dnode->dn_datablkszsec * 512;
1311 mz = (const mzap_phys_t *) zap_scratch;
1312 chunks = size / MZAP_ENT_LEN - 1;
1314 for (i = 0; i < chunks; i++) {
1315 mze = &mz->mz_chunk[i];
1316 if (mze->mze_name[0])
1317 //printf("%-32s 0x%llx\n", mze->mze_name, mze->mze_value);
1318 printf("%s\n", mze->mze_name);
1325 * List a fatzap directory. Assumes that the zap scratch buffer contains
1326 * the directory header.
1329 fzap_list(spa_t *spa, const dnode_phys_t *dnode)
1331 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1332 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1336 if (zh.zap_magic != ZAP_MAGIC)
1339 z.zap_block_shift = ilog2(bsize);
1340 z.zap_phys = (zap_phys_t *) zap_scratch;
1343 * This assumes that the leaf blocks start at block 1. The
1344 * documentation isn't exactly clear on this.
1347 zl.l_bs = z.zap_block_shift;
1348 for (i = 0; i < zh.zap_num_leafs; i++) {
1349 off_t off = (i + 1) << zl.l_bs;
1353 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1356 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1358 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1359 zap_leaf_chunk_t *zc, *nc;
1362 zc = &ZAP_LEAF_CHUNK(&zl, j);
1363 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1365 namelen = zc->l_entry.le_name_length;
1366 if (namelen > sizeof(name))
1367 namelen = sizeof(name);
1370 * Paste the name back together.
1372 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1374 while (namelen > 0) {
1377 if (len > ZAP_LEAF_ARRAY_BYTES)
1378 len = ZAP_LEAF_ARRAY_BYTES;
1379 memcpy(p, nc->l_array.la_array, len);
1382 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1386 * Assume the first eight bytes of the value are
1389 value = fzap_leaf_value(&zl, zc);
1391 printf("%s 0x%llx\n", name, value);
1399 * List a zap directory.
1402 zap_list(spa_t *spa, const dnode_phys_t *dnode)
1405 size_t size = dnode->dn_datablkszsec * 512;
1407 if (dnode_read(spa, dnode, 0, zap_scratch, size))
1410 zap_type = *(uint64_t *) zap_scratch;
1411 if (zap_type == ZBT_MICRO)
1412 return mzap_list(spa, dnode);
1414 return fzap_list(spa, dnode);
1420 objset_get_dnode(spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1424 offset = objnum * sizeof(dnode_phys_t);
1425 return dnode_read(spa, &os->os_meta_dnode, offset,
1426 dnode, sizeof(dnode_phys_t));
1430 * Find the object set given the object number of its dataset object
1431 * and return its details in *objset
1434 zfs_mount_dataset(spa_t *spa, uint64_t objnum, objset_phys_t *objset)
1436 dnode_phys_t dataset;
1437 dsl_dataset_phys_t *ds;
1439 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1440 printf("ZFS: can't find dataset %llu\n", objnum);
1444 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1445 if (zio_read(spa, &ds->ds_bp, objset)) {
1446 printf("ZFS: can't read object set for dataset %llu\n", objnum);
1454 * Find the object set pointed to by the BOOTFS property or the root
1455 * dataset if there is none and return its details in *objset
1458 zfs_mount_root(spa_t *spa, objset_phys_t *objset)
1460 dnode_phys_t dir, propdir;
1461 uint64_t props, bootfs, root;
1464 * Start with the MOS directory object.
1466 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
1467 printf("ZFS: can't read MOS object directory\n");
1472 * Lookup the pool_props and see if we can find a bootfs.
1474 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
1475 && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
1476 && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
1478 return zfs_mount_dataset(spa, bootfs, objset);
1481 * Lookup the root dataset directory
1483 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
1484 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
1485 printf("ZFS: can't find root dsl_dir\n");
1490 * Use the information from the dataset directory's bonus buffer
1491 * to find the dataset object and from that the object set itself.
1493 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
1494 return zfs_mount_dataset(spa, dd->dd_head_dataset_obj, objset);
1498 zfs_mount_pool(spa_t *spa)
1501 * Find the MOS and work our way in from there.
1503 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
1504 printf("ZFS: can't read MOS\n");
1509 * Find the root object set
1511 if (zfs_mount_root(spa, &spa->spa_root_objset)) {
1512 printf("Can't find root filesystem - giving up\n");
1520 * Lookup a file and return its dnode.
1523 zfs_lookup(spa_t *spa, const char *upath, dnode_phys_t *dnode)
1526 uint64_t objnum, rootnum, parentnum;
1528 const znode_phys_t *zp = (const znode_phys_t *) dn.dn_bonus;
1532 int symlinks_followed = 0;
1534 if (spa->spa_root_objset.os_type != DMU_OST_ZFS) {
1535 printf("ZFS: unexpected object set type %llu\n",
1536 spa->spa_root_objset.os_type);
1541 * Get the root directory dnode.
1543 rc = objset_get_dnode(spa, &spa->spa_root_objset, MASTER_NODE_OBJ, &dn);
1547 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &rootnum);
1551 rc = objset_get_dnode(spa, &spa->spa_root_objset, rootnum, &dn);
1564 memcpy(element, p, q - p);
1572 if ((zp->zp_mode >> 12) != 0x4) {
1577 rc = zap_lookup(spa, &dn, element, &objnum);
1580 objnum = ZFS_DIRENT_OBJ(objnum);
1582 rc = objset_get_dnode(spa, &spa->spa_root_objset, objnum, &dn);
1587 * Check for symlink.
1589 if ((zp->zp_mode >> 12) == 0xa) {
1590 if (symlinks_followed > 10)
1592 symlinks_followed++;
1595 * Read the link value and copy the tail of our
1596 * current path onto the end.
1599 strcpy(&path[zp->zp_size], p);
1601 path[zp->zp_size] = 0;
1602 if (zp->zp_size + sizeof(znode_phys_t) <= dn.dn_bonuslen) {
1603 memcpy(path, &dn.dn_bonus[sizeof(znode_phys_t)],
1606 rc = dnode_read(spa, &dn, 0, path, zp->zp_size);
1612 * Restart with the new path, starting either at
1613 * the root or at the parent depending whether or
1614 * not the link is relative.
1621 objset_get_dnode(spa, &spa->spa_root_objset, objnum, &dn);