4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 * Pool import support functions.
28 * To import a pool, we rely on reading the configuration information from the
29 * ZFS label of each device. If we successfully read the label, then we
30 * organize the configuration information in the following hierarchy:
32 * pool guid -> toplevel vdev guid -> label txg
34 * Duplicate entries matching this same tuple will be discarded. Once we have
35 * examined every device, we pick the best label txg config for each toplevel
36 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
37 * update any paths that have changed. Finally, we attempt to import the pool
38 * using our derived config, and record the results.
53 #include <sys/dktp/fdisk.h>
54 #include <sys/efi_partition.h>
55 #include <thread_pool.h>
57 #include <sys/vdev_impl.h>
60 #include "libzfs_impl.h"
63 * Intermediate structures used to gather configuration information.
65 typedef struct config_entry {
68 struct config_entry *ce_next;
71 typedef struct vdev_entry {
73 config_entry_t *ve_configs;
74 struct vdev_entry *ve_next;
77 typedef struct pool_entry {
79 vdev_entry_t *pe_vdevs;
80 struct pool_entry *pe_next;
83 typedef struct name_entry {
86 struct name_entry *ne_next;
89 typedef struct pool_list {
95 get_devid(const char *path)
101 if ((fd = open(path, O_RDONLY)) < 0)
106 if (devid_get(fd, &devid) == 0) {
107 if (devid_get_minor_name(fd, &minor) == 0)
108 ret = devid_str_encode(devid, minor);
110 devid_str_free(minor);
120 * Go through and fix up any path and/or devid information for the given vdev
124 fix_paths(nvlist_t *nv, name_entry_t *names)
129 name_entry_t *ne, *best;
133 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
134 &child, &children) == 0) {
135 for (c = 0; c < children; c++)
136 if (fix_paths(child[c], names) != 0)
142 * This is a leaf (file or disk) vdev. In either case, go through
143 * the name list and see if we find a matching guid. If so, replace
144 * the path and see if we can calculate a new devid.
146 * There may be multiple names associated with a particular guid, in
147 * which case we have overlapping slices or multiple paths to the same
148 * disk. If this is the case, then we want to pick the path that is
149 * the most similar to the original, where "most similar" is the number
150 * of matching characters starting from the end of the path. This will
151 * preserve slice numbers even if the disks have been reorganized, and
152 * will also catch preferred disk names if multiple paths exist.
154 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
155 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
160 for (ne = names; ne != NULL; ne = ne->ne_next) {
161 if (ne->ne_guid == guid) {
162 const char *src, *dst;
170 src = ne->ne_name + strlen(ne->ne_name) - 1;
171 dst = path + strlen(path) - 1;
172 for (count = 0; src >= ne->ne_name && dst >= path;
173 src--, dst--, count++)
178 * At this point, 'count' is the number of characters
179 * matched from the end.
181 if (count > matched || best == NULL) {
191 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
194 if ((devid = get_devid(best->ne_name)) == NULL) {
195 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
197 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
199 devid_str_free(devid);
206 * Add the given configuration to the list of known devices.
209 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
212 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
219 * If this is a hot spare not currently in use or level 2 cache
220 * device, add it to the list of names to translate, but don't do
223 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
225 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
226 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
227 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
230 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
234 ne->ne_guid = vdev_guid;
235 ne->ne_next = pl->names;
241 * If we have a valid config but cannot read any of these fields, then
242 * it means we have a half-initialized label. In vdev_label_init()
243 * we write a label with txg == 0 so that we can identify the device
244 * in case the user refers to the same disk later on. If we fail to
245 * create the pool, we'll be left with a label in this state
246 * which should not be considered part of a valid pool.
248 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
250 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
252 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
254 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
255 &txg) != 0 || txg == 0) {
261 * First, see if we know about this pool. If not, then add it to the
262 * list of known pools.
264 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
265 if (pe->pe_guid == pool_guid)
270 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
274 pe->pe_guid = pool_guid;
275 pe->pe_next = pl->pools;
280 * Second, see if we know about this toplevel vdev. Add it if its
283 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
284 if (ve->ve_guid == top_guid)
289 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
293 ve->ve_guid = top_guid;
294 ve->ve_next = pe->pe_vdevs;
299 * Third, see if we have a config with a matching transaction group. If
300 * so, then we do nothing. Otherwise, add it to the list of known
303 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
304 if (ce->ce_txg == txg)
309 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
314 ce->ce_config = config;
315 ce->ce_next = ve->ve_configs;
322 * At this point we've successfully added our config to the list of
323 * known configs. The last thing to do is add the vdev guid -> path
324 * mappings so that we can fix up the configuration as necessary before
327 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
330 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
335 ne->ne_guid = vdev_guid;
336 ne->ne_next = pl->names;
343 * Returns true if the named pool matches the given GUID.
346 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
352 if (zpool_open_silent(hdl, name, &zhp) != 0)
360 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
365 *isactive = (theguid == guid);
370 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
373 zfs_cmd_t zc = { 0 };
376 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
379 if (zcmd_alloc_dst_nvlist(hdl, &zc,
380 zc.zc_nvlist_conf_size * 2) != 0) {
381 zcmd_free_nvlists(&zc);
385 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
386 &zc)) != 0 && errno == ENOMEM) {
387 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
388 zcmd_free_nvlists(&zc);
394 zcmd_free_nvlists(&zc);
398 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
399 zcmd_free_nvlists(&zc);
403 zcmd_free_nvlists(&zc);
408 * Determine if the vdev id is a hole in the namespace.
411 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
413 for (int c = 0; c < holes; c++) {
415 /* Top-level is a hole */
416 if (hole_array[c] == id)
423 * Convert our list of pools into the definitive set of configurations. We
424 * start by picking the best config for each toplevel vdev. Once that's done,
425 * we assemble the toplevel vdevs into a full config for the pool. We make a
426 * pass to fix up any incorrect paths, and then add it to the main list to
427 * return to the user.
430 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
435 nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
436 nvlist_t **spares, **l2cache;
437 uint_t i, nspares, nl2cache;
438 boolean_t config_seen;
440 char *name, *hostname;
441 uint64_t version, guid;
443 nvlist_t **child = NULL;
445 uint64_t *hole_array, max_id;
450 boolean_t found_one = B_FALSE;
451 boolean_t valid_top_config = B_FALSE;
453 if (nvlist_alloc(&ret, 0, 0) != 0)
456 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
457 uint64_t id, max_txg = 0;
459 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
461 config_seen = B_FALSE;
464 * Iterate over all toplevel vdevs. Grab the pool configuration
465 * from the first one we find, and then go through the rest and
466 * add them as necessary to the 'vdevs' member of the config.
468 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
471 * Determine the best configuration for this vdev by
472 * selecting the config with the latest transaction
476 for (ce = ve->ve_configs; ce != NULL;
479 if (ce->ce_txg > best_txg) {
481 best_txg = ce->ce_txg;
486 * We rely on the fact that the max txg for the
487 * pool will contain the most up-to-date information
488 * about the valid top-levels in the vdev namespace.
490 if (best_txg > max_txg) {
491 (void) nvlist_remove(config,
492 ZPOOL_CONFIG_VDEV_CHILDREN,
494 (void) nvlist_remove(config,
495 ZPOOL_CONFIG_HOLE_ARRAY,
496 DATA_TYPE_UINT64_ARRAY);
502 valid_top_config = B_FALSE;
504 if (nvlist_lookup_uint64(tmp,
505 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
506 verify(nvlist_add_uint64(config,
507 ZPOOL_CONFIG_VDEV_CHILDREN,
509 valid_top_config = B_TRUE;
512 if (nvlist_lookup_uint64_array(tmp,
513 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
515 verify(nvlist_add_uint64_array(config,
516 ZPOOL_CONFIG_HOLE_ARRAY,
517 hole_array, holes) == 0);
523 * Copy the relevant pieces of data to the pool
530 * hostid (if available)
531 * hostname (if available)
535 verify(nvlist_lookup_uint64(tmp,
536 ZPOOL_CONFIG_VERSION, &version) == 0);
537 if (nvlist_add_uint64(config,
538 ZPOOL_CONFIG_VERSION, version) != 0)
540 verify(nvlist_lookup_uint64(tmp,
541 ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
542 if (nvlist_add_uint64(config,
543 ZPOOL_CONFIG_POOL_GUID, guid) != 0)
545 verify(nvlist_lookup_string(tmp,
546 ZPOOL_CONFIG_POOL_NAME, &name) == 0);
547 if (nvlist_add_string(config,
548 ZPOOL_CONFIG_POOL_NAME, name) != 0)
550 verify(nvlist_lookup_uint64(tmp,
551 ZPOOL_CONFIG_POOL_STATE, &state) == 0);
552 if (nvlist_add_uint64(config,
553 ZPOOL_CONFIG_POOL_STATE, state) != 0)
556 if (nvlist_lookup_uint64(tmp,
557 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
558 if (nvlist_add_uint64(config,
559 ZPOOL_CONFIG_HOSTID, hostid) != 0)
561 verify(nvlist_lookup_string(tmp,
562 ZPOOL_CONFIG_HOSTNAME,
564 if (nvlist_add_string(config,
565 ZPOOL_CONFIG_HOSTNAME,
570 config_seen = B_TRUE;
574 * Add this top-level vdev to the child array.
576 verify(nvlist_lookup_nvlist(tmp,
577 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
578 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
581 if (id >= children) {
584 newchild = zfs_alloc(hdl, (id + 1) *
585 sizeof (nvlist_t *));
586 if (newchild == NULL)
589 for (c = 0; c < children; c++)
590 newchild[c] = child[c];
596 if (nvlist_dup(nvtop, &child[id], 0) != 0)
602 * If we have information about all the top-levels then
603 * clean up the nvlist which we've constructed. This
604 * means removing any extraneous devices that are
605 * beyond the valid range or adding devices to the end
606 * of our array which appear to be missing.
608 if (valid_top_config) {
609 if (max_id < children) {
610 for (c = max_id; c < children; c++)
611 nvlist_free(child[c]);
613 } else if (max_id > children) {
616 newchild = zfs_alloc(hdl, (max_id) *
617 sizeof (nvlist_t *));
618 if (newchild == NULL)
621 for (c = 0; c < children; c++)
622 newchild[c] = child[c];
630 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
634 * The vdev namespace may contain holes as a result of
635 * device removal. We must add them back into the vdev
636 * tree before we process any missing devices.
639 ASSERT(valid_top_config);
641 for (c = 0; c < children; c++) {
644 if (child[c] != NULL ||
645 !vdev_is_hole(hole_array, holes, c))
648 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
653 * Holes in the namespace are treated as
654 * "hole" top-level vdevs and have a
655 * special flag set on them.
657 if (nvlist_add_string(holey,
659 VDEV_TYPE_HOLE) != 0 ||
660 nvlist_add_uint64(holey,
661 ZPOOL_CONFIG_ID, c) != 0 ||
662 nvlist_add_uint64(holey,
663 ZPOOL_CONFIG_GUID, 0ULL) != 0)
670 * Look for any missing top-level vdevs. If this is the case,
671 * create a faked up 'missing' vdev as a placeholder. We cannot
672 * simply compress the child array, because the kernel performs
673 * certain checks to make sure the vdev IDs match their location
674 * in the configuration.
676 for (c = 0; c < children; c++) {
677 if (child[c] == NULL) {
679 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
682 if (nvlist_add_string(missing,
684 VDEV_TYPE_MISSING) != 0 ||
685 nvlist_add_uint64(missing,
686 ZPOOL_CONFIG_ID, c) != 0 ||
687 nvlist_add_uint64(missing,
688 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
689 nvlist_free(missing);
697 * Put all of this pool's top-level vdevs into a root vdev.
699 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
701 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
702 VDEV_TYPE_ROOT) != 0 ||
703 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
704 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
705 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
706 child, children) != 0) {
711 for (c = 0; c < children; c++)
712 nvlist_free(child[c]);
718 * Go through and fix up any paths and/or devids based on our
719 * known list of vdev GUID -> path mappings.
721 if (fix_paths(nvroot, pl->names) != 0) {
727 * Add the root vdev to this pool's configuration.
729 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
737 * zdb uses this path to report on active pools that were
738 * imported or created using -R.
744 * Determine if this pool is currently active, in which case we
745 * can't actually import it.
747 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
749 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
752 if (pool_active(hdl, name, guid, &isactive) != 0)
761 if ((nvl = refresh_config(hdl, config)) == NULL) {
771 * Go through and update the paths for spares, now that we have
774 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
776 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
777 &spares, &nspares) == 0) {
778 for (i = 0; i < nspares; i++) {
779 if (fix_paths(spares[i], pl->names) != 0)
785 * Update the paths for l2cache devices.
787 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
788 &l2cache, &nl2cache) == 0) {
789 for (i = 0; i < nl2cache; i++) {
790 if (fix_paths(l2cache[i], pl->names) != 0)
796 * Restore the original information read from the actual label.
798 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
800 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
803 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
805 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
811 * Add this pool to the list of configs.
813 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
815 if (nvlist_add_nvlist(ret, name, config) != 0)
831 (void) no_memory(hdl);
835 for (c = 0; c < children; c++)
836 nvlist_free(child[c]);
843 * Return the offset of the given label.
846 label_offset(uint64_t size, int l)
848 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
849 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
850 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
854 * Given a file descriptor, read the label information and return an nvlist
855 * describing the configuration, if there is one.
858 zpool_read_label(int fd, nvlist_t **config)
860 struct stat64 statbuf;
863 uint64_t state, txg, size;
867 if (fstat64(fd, &statbuf) == -1)
869 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
871 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
874 for (l = 0; l < VDEV_LABELS; l++) {
875 if (pread64(fd, label, sizeof (vdev_label_t),
876 label_offset(size, l)) != sizeof (vdev_label_t))
879 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
880 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
883 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
884 &state) != 0 || state > POOL_STATE_L2CACHE) {
885 nvlist_free(*config);
889 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
890 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
891 &txg) != 0 || txg == 0)) {
892 nvlist_free(*config);
905 typedef struct rdsk_node {
908 libzfs_handle_t *rn_hdl;
912 boolean_t rn_nozpool;
916 slice_cache_compare(const void *arg1, const void *arg2)
918 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
919 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
920 char *nm1slice, *nm2slice;
924 * slices zero and two are the most likely to provide results,
927 nm1slice = strstr(nm1, "s0");
928 nm2slice = strstr(nm2, "s0");
929 if (nm1slice && !nm2slice) {
932 if (!nm1slice && nm2slice) {
935 nm1slice = strstr(nm1, "s2");
936 nm2slice = strstr(nm2, "s2");
937 if (nm1slice && !nm2slice) {
940 if (!nm1slice && nm2slice) {
944 rv = strcmp(nm1, nm2);
947 return (rv > 0 ? 1 : -1);
951 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
952 diskaddr_t size, uint_t blksz)
956 char sname[MAXNAMELEN];
958 tmpnode.rn_name = &sname[0];
959 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
962 * protect against division by zero for disk labels that
963 * contain a bogus sector size
967 /* too small to contain a zpool? */
968 if ((size < (SPA_MINDEVSIZE / blksz)) &&
969 (node = avl_find(r, &tmpnode, NULL)))
970 node->rn_nozpool = B_TRUE;
974 nozpool_all_slices(avl_tree_t *r, const char *sname)
976 char diskname[MAXNAMELEN];
980 (void) strncpy(diskname, sname, MAXNAMELEN);
981 if (((ptr = strrchr(diskname, 's')) == NULL) &&
982 ((ptr = strrchr(diskname, 'p')) == NULL))
986 for (i = 0; i < NDKMAP; i++)
987 check_one_slice(r, diskname, i, 0, 1);
989 for (i = 0; i <= FD_NUMPART; i++)
990 check_one_slice(r, diskname, i, 0, 1);
994 check_slices(avl_tree_t *r, int fd, const char *sname)
998 char diskname[MAXNAMELEN];
1002 (void) strncpy(diskname, sname, MAXNAMELEN);
1003 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1007 if (read_extvtoc(fd, &vtoc) >= 0) {
1008 for (i = 0; i < NDKMAP; i++)
1009 check_one_slice(r, diskname, i,
1010 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1011 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1013 * on x86 we'll still have leftover links that point
1014 * to slices s[9-15], so use NDKMAP instead
1016 for (i = 0; i < NDKMAP; i++)
1017 check_one_slice(r, diskname, i,
1018 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1019 /* nodes p[1-4] are never used with EFI labels */
1021 for (i = 1; i <= FD_NUMPART; i++)
1022 check_one_slice(r, diskname, i, 0, 1);
1028 zpool_open_func(void *arg)
1030 rdsk_node_t *rn = arg;
1031 struct stat64 statbuf;
1037 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1038 /* symlink to a device that's no longer there */
1039 if (errno == ENOENT)
1040 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1044 * Ignore failed stats. We only want regular
1045 * files, character devs and block devs.
1047 if (fstat64(fd, &statbuf) != 0 ||
1048 (!S_ISREG(statbuf.st_mode) &&
1049 !S_ISCHR(statbuf.st_mode) &&
1050 !S_ISBLK(statbuf.st_mode))) {
1054 /* this file is too small to hold a zpool */
1055 if (S_ISREG(statbuf.st_mode) &&
1056 statbuf.st_size < SPA_MINDEVSIZE) {
1059 } else if (!S_ISREG(statbuf.st_mode)) {
1061 * Try to read the disk label first so we don't have to
1062 * open a bunch of minor nodes that can't have a zpool.
1064 check_slices(rn->rn_avl, fd, rn->rn_name);
1067 if ((zpool_read_label(fd, &config)) != 0) {
1069 (void) no_memory(rn->rn_hdl);
1075 rn->rn_config = config;
1076 if (config != NULL) {
1077 assert(rn->rn_nozpool == B_FALSE);
1082 * Given a file descriptor, clear (zero) the label information. This function
1083 * is currently only used in the appliance stack as part of the ZFS sysevent
1087 zpool_clear_label(int fd)
1089 struct stat64 statbuf;
1091 vdev_label_t *label;
1094 if (fstat64(fd, &statbuf) == -1)
1096 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1098 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1101 for (l = 0; l < VDEV_LABELS; l++) {
1102 if (pwrite64(fd, label, sizeof (vdev_label_t),
1103 label_offset(size, l)) != sizeof (vdev_label_t))
1112 * Given a list of directories to search, find all pools stored on disk. This
1113 * includes partial pools which are not available to import. If no args are
1114 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1115 * poolname or guid (but not both) are provided by the caller when trying
1116 * to import a specific pool.
1119 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1121 int i, dirs = iarg->paths;
1123 struct dirent64 *dp;
1124 char path[MAXPATHLEN];
1125 char *end, **dir = iarg->path;
1127 nvlist_t *ret = NULL;
1128 static char *default_dir = "/dev/dsk";
1129 pool_list_t pools = { 0 };
1130 pool_entry_t *pe, *penext;
1131 vdev_entry_t *ve, *venext;
1132 config_entry_t *ce, *cenext;
1133 name_entry_t *ne, *nenext;
1134 avl_tree_t slice_cache;
1144 * Go through and read the label configuration information from every
1145 * possible device, organizing the information according to pool GUID
1146 * and toplevel GUID.
1148 for (i = 0; i < dirs; i++) {
1153 /* use realpath to normalize the path */
1154 if (realpath(dir[i], path) == 0) {
1155 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1156 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1159 end = &path[strlen(path)];
1162 pathleft = &path[sizeof (path)] - end;
1165 * Using raw devices instead of block devices when we're
1166 * reading the labels skips a bunch of slow operations during
1167 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1169 if (strcmp(path, "/dev/dsk/") == 0)
1170 rdsk = "/dev/rdsk/";
1174 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1175 (dirp = fdopendir(dfd)) == NULL) {
1176 zfs_error_aux(hdl, strerror(errno));
1177 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1178 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1183 avl_create(&slice_cache, slice_cache_compare,
1184 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1186 * This is not MT-safe, but we have no MT consumers of libzfs
1188 while ((dp = readdir64(dirp)) != NULL) {
1189 const char *name = dp->d_name;
1190 if (name[0] == '.' &&
1191 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1194 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1195 slice->rn_name = zfs_strdup(hdl, name);
1196 slice->rn_avl = &slice_cache;
1197 slice->rn_dfd = dfd;
1198 slice->rn_hdl = hdl;
1199 slice->rn_nozpool = B_FALSE;
1200 avl_add(&slice_cache, slice);
1203 * create a thread pool to do all of this in parallel;
1204 * rn_nozpool is not protected, so this is racy in that
1205 * multiple tasks could decide that the same slice can
1206 * not hold a zpool, which is benign. Also choose
1207 * double the number of processors; we hold a lot of
1208 * locks in the kernel, so going beyond this doesn't
1211 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1213 for (slice = avl_first(&slice_cache); slice;
1214 (slice = avl_walk(&slice_cache, slice,
1216 (void) tpool_dispatch(t, zpool_open_func, slice);
1221 while ((slice = avl_destroy_nodes(&slice_cache,
1222 &cookie)) != NULL) {
1223 if (slice->rn_config != NULL) {
1224 nvlist_t *config = slice->rn_config;
1225 boolean_t matched = B_TRUE;
1227 if (iarg->poolname != NULL) {
1230 matched = nvlist_lookup_string(config,
1231 ZPOOL_CONFIG_POOL_NAME,
1233 strcmp(iarg->poolname, pname) == 0;
1234 } else if (iarg->guid != 0) {
1237 matched = nvlist_lookup_uint64(config,
1238 ZPOOL_CONFIG_POOL_GUID,
1240 iarg->guid == this_guid;
1243 nvlist_free(config);
1247 /* use the non-raw path for the config */
1248 (void) strlcpy(end, slice->rn_name, pathleft);
1249 if (add_config(hdl, &pools, path, config) != 0)
1252 free(slice->rn_name);
1255 avl_destroy(&slice_cache);
1257 (void) closedir(dirp);
1261 ret = get_configs(hdl, &pools, iarg->can_be_active);
1264 for (pe = pools.pools; pe != NULL; pe = penext) {
1265 penext = pe->pe_next;
1266 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1267 venext = ve->ve_next;
1268 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1269 cenext = ce->ce_next;
1271 nvlist_free(ce->ce_config);
1279 for (ne = pools.names; ne != NULL; ne = nenext) {
1280 nenext = ne->ne_next;
1287 (void) closedir(dirp);
1293 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1295 importargs_t iarg = { 0 };
1300 return (zpool_find_import_impl(hdl, &iarg));
1304 * Given a cache file, return the contents as a list of importable pools.
1305 * poolname or guid (but not both) are provided by the caller when trying
1306 * to import a specific pool.
1309 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1310 char *poolname, uint64_t guid)
1314 struct stat64 statbuf;
1315 nvlist_t *raw, *src, *dst;
1322 verify(poolname == NULL || guid == 0);
1324 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1325 zfs_error_aux(hdl, "%s", strerror(errno));
1326 (void) zfs_error(hdl, EZFS_BADCACHE,
1327 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1331 if (fstat64(fd, &statbuf) != 0) {
1332 zfs_error_aux(hdl, "%s", strerror(errno));
1334 (void) zfs_error(hdl, EZFS_BADCACHE,
1335 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1339 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1344 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1347 (void) zfs_error(hdl, EZFS_BADCACHE,
1348 dgettext(TEXT_DOMAIN,
1349 "failed to read cache file contents"));
1355 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1357 (void) zfs_error(hdl, EZFS_BADCACHE,
1358 dgettext(TEXT_DOMAIN,
1359 "invalid or corrupt cache file contents"));
1366 * Go through and get the current state of the pools and refresh their
1369 if (nvlist_alloc(&pools, 0, 0) != 0) {
1370 (void) no_memory(hdl);
1376 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1377 verify(nvpair_value_nvlist(elem, &src) == 0);
1379 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1381 if (poolname != NULL && strcmp(poolname, name) != 0)
1384 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1387 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1389 if (guid != this_guid)
1393 if (pool_active(hdl, name, this_guid, &active) != 0) {
1402 if ((dst = refresh_config(hdl, src)) == NULL) {
1408 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1409 (void) no_memory(hdl);
1423 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1425 importargs_t *import = data;
1428 if (import->poolname != NULL) {
1431 verify(nvlist_lookup_string(zhp->zpool_config,
1432 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1433 if (strcmp(pool_name, import->poolname) == 0)
1438 verify(nvlist_lookup_uint64(zhp->zpool_config,
1439 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1440 if (pool_guid == import->guid)
1449 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1451 verify(import->poolname == NULL || import->guid == 0);
1454 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1456 if (import->cachefile != NULL)
1457 return (zpool_find_import_cached(hdl, import->cachefile,
1458 import->poolname, import->guid));
1460 return (zpool_find_import_impl(hdl, import));
1464 find_guid(nvlist_t *nv, uint64_t guid)
1470 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1474 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1475 &child, &children) == 0) {
1476 for (c = 0; c < children; c++)
1477 if (find_guid(child[c], guid))
1484 typedef struct aux_cbdata {
1485 const char *cb_type;
1487 zpool_handle_t *cb_zhp;
1491 find_aux(zpool_handle_t *zhp, void *data)
1493 aux_cbdata_t *cbp = data;
1499 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1502 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1503 &list, &count) == 0) {
1504 for (i = 0; i < count; i++) {
1505 verify(nvlist_lookup_uint64(list[i],
1506 ZPOOL_CONFIG_GUID, &guid) == 0);
1507 if (guid == cbp->cb_guid) {
1519 * Determines if the pool is in use. If so, it returns true and the state of
1520 * the pool as well as the name of the pool. Both strings are allocated and
1521 * must be freed by the caller.
1524 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1530 uint64_t guid, vdev_guid;
1531 zpool_handle_t *zhp;
1532 nvlist_t *pool_config;
1533 uint64_t stateval, isspare;
1534 aux_cbdata_t cb = { 0 };
1539 if (zpool_read_label(fd, &config) != 0) {
1540 (void) no_memory(hdl);
1547 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1549 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1552 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1553 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1555 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1560 case POOL_STATE_EXPORTED:
1562 * A pool with an exported state may in fact be imported
1563 * read-only, so check the in-core state to see if it's
1564 * active and imported read-only. If it is, set
1565 * its state to active.
1567 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1568 (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1569 zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1570 stateval = POOL_STATE_ACTIVE;
1575 case POOL_STATE_ACTIVE:
1577 * For an active pool, we have to determine if it's really part
1578 * of a currently active pool (in which case the pool will exist
1579 * and the guid will be the same), or whether it's part of an
1580 * active pool that was disconnected without being explicitly
1583 if (pool_active(hdl, name, guid, &isactive) != 0) {
1584 nvlist_free(config);
1590 * Because the device may have been removed while
1591 * offlined, we only report it as active if the vdev is
1592 * still present in the config. Otherwise, pretend like
1595 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1596 (pool_config = zpool_get_config(zhp, NULL))
1600 verify(nvlist_lookup_nvlist(pool_config,
1601 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1602 ret = find_guid(nvroot, vdev_guid);
1608 * If this is an active spare within another pool, we
1609 * treat it like an unused hot spare. This allows the
1610 * user to create a pool with a hot spare that currently
1611 * in use within another pool. Since we return B_TRUE,
1612 * libdiskmgt will continue to prevent generic consumers
1613 * from using the device.
1615 if (ret && nvlist_lookup_uint64(config,
1616 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1617 stateval = POOL_STATE_SPARE;
1622 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1627 case POOL_STATE_SPARE:
1629 * For a hot spare, it can be either definitively in use, or
1630 * potentially active. To determine if it's in use, we iterate
1631 * over all pools in the system and search for one with a spare
1632 * with a matching guid.
1634 * Due to the shared nature of spares, we don't actually report
1635 * the potentially active case as in use. This means the user
1636 * can freely create pools on the hot spares of exported pools,
1637 * but to do otherwise makes the resulting code complicated, and
1638 * we end up having to deal with this case anyway.
1641 cb.cb_guid = vdev_guid;
1642 cb.cb_type = ZPOOL_CONFIG_SPARES;
1643 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1644 name = (char *)zpool_get_name(cb.cb_zhp);
1651 case POOL_STATE_L2CACHE:
1654 * Check if any pool is currently using this l2cache device.
1657 cb.cb_guid = vdev_guid;
1658 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1659 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1660 name = (char *)zpool_get_name(cb.cb_zhp);
1673 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1675 zpool_close(cb.cb_zhp);
1676 nvlist_free(config);
1679 *state = (pool_state_t)stateval;
1683 zpool_close(cb.cb_zhp);
1685 nvlist_free(config);