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.
52 #include <thread_pool.h>
55 #include <sys/vdev_impl.h>
58 #include "libzfs_impl.h"
61 * Intermediate structures used to gather configuration information.
63 typedef struct config_entry {
66 struct config_entry *ce_next;
69 typedef struct vdev_entry {
71 config_entry_t *ve_configs;
72 struct vdev_entry *ve_next;
75 typedef struct pool_entry {
77 vdev_entry_t *pe_vdevs;
78 struct pool_entry *pe_next;
81 typedef struct name_entry {
84 struct name_entry *ne_next;
87 typedef struct pool_list {
93 get_devid(const char *path)
99 if ((fd = open(path, O_RDONLY)) < 0)
104 if (devid_get(fd, &devid) == 0) {
105 if (devid_get_minor_name(fd, &minor) == 0)
106 ret = devid_str_encode(devid, minor);
108 devid_str_free(minor);
118 * Go through and fix up any path and/or devid information for the given vdev
122 fix_paths(nvlist_t *nv, name_entry_t *names)
127 name_entry_t *ne, *best;
131 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
132 &child, &children) == 0) {
133 for (c = 0; c < children; c++)
134 if (fix_paths(child[c], names) != 0)
140 * This is a leaf (file or disk) vdev. In either case, go through
141 * the name list and see if we find a matching guid. If so, replace
142 * the path and see if we can calculate a new devid.
144 * There may be multiple names associated with a particular guid, in
145 * which case we have overlapping slices or multiple paths to the same
146 * disk. If this is the case, then we want to pick the path that is
147 * the most similar to the original, where "most similar" is the number
148 * of matching characters starting from the end of the path. This will
149 * preserve slice numbers even if the disks have been reorganized, and
150 * will also catch preferred disk names if multiple paths exist.
152 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
153 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
158 for (ne = names; ne != NULL; ne = ne->ne_next) {
159 if (ne->ne_guid == guid) {
160 const char *src, *dst;
168 src = ne->ne_name + strlen(ne->ne_name) - 1;
169 dst = path + strlen(path) - 1;
170 for (count = 0; src >= ne->ne_name && dst >= path;
171 src--, dst--, count++)
176 * At this point, 'count' is the number of characters
177 * matched from the end.
179 if (count > matched || best == NULL) {
189 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
192 if ((devid = get_devid(best->ne_name)) == NULL) {
193 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
195 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
197 devid_str_free(devid);
204 * Add the given configuration to the list of known devices.
207 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
210 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
217 * If this is a hot spare not currently in use or level 2 cache
218 * device, add it to the list of names to translate, but don't do
221 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
223 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
224 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
225 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
228 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
232 ne->ne_guid = vdev_guid;
233 ne->ne_next = pl->names;
239 * If we have a valid config but cannot read any of these fields, then
240 * it means we have a half-initialized label. In vdev_label_init()
241 * we write a label with txg == 0 so that we can identify the device
242 * in case the user refers to the same disk later on. If we fail to
243 * create the pool, we'll be left with a label in this state
244 * which should not be considered part of a valid pool.
246 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
248 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
250 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
252 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
253 &txg) != 0 || txg == 0) {
259 * First, see if we know about this pool. If not, then add it to the
260 * list of known pools.
262 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
263 if (pe->pe_guid == pool_guid)
268 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
272 pe->pe_guid = pool_guid;
273 pe->pe_next = pl->pools;
278 * Second, see if we know about this toplevel vdev. Add it if its
281 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
282 if (ve->ve_guid == top_guid)
287 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
291 ve->ve_guid = top_guid;
292 ve->ve_next = pe->pe_vdevs;
297 * Third, see if we have a config with a matching transaction group. If
298 * so, then we do nothing. Otherwise, add it to the list of known
301 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
302 if (ce->ce_txg == txg)
307 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
312 ce->ce_config = config;
313 ce->ce_next = ve->ve_configs;
320 * At this point we've successfully added our config to the list of
321 * known configs. The last thing to do is add the vdev guid -> path
322 * mappings so that we can fix up the configuration as necessary before
325 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
328 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
333 ne->ne_guid = vdev_guid;
334 ne->ne_next = pl->names;
341 * Returns true if the named pool matches the given GUID.
344 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
350 if (zpool_open_silent(hdl, name, &zhp) != 0)
358 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
363 *isactive = (theguid == guid);
368 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
371 zfs_cmd_t zc = { 0 };
374 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
377 if (zcmd_alloc_dst_nvlist(hdl, &zc,
378 zc.zc_nvlist_conf_size * 2) != 0) {
379 zcmd_free_nvlists(&zc);
383 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
384 &zc)) != 0 && errno == ENOMEM) {
385 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
386 zcmd_free_nvlists(&zc);
392 zcmd_free_nvlists(&zc);
396 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
397 zcmd_free_nvlists(&zc);
401 zcmd_free_nvlists(&zc);
406 * Determine if the vdev id is a hole in the namespace.
409 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
411 for (int c = 0; c < holes; c++) {
413 /* Top-level is a hole */
414 if (hole_array[c] == id)
421 * Convert our list of pools into the definitive set of configurations. We
422 * start by picking the best config for each toplevel vdev. Once that's done,
423 * we assemble the toplevel vdevs into a full config for the pool. We make a
424 * pass to fix up any incorrect paths, and then add it to the main list to
425 * return to the user.
428 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
433 nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
434 nvlist_t **spares, **l2cache;
435 uint_t i, nspares, nl2cache;
436 boolean_t config_seen;
438 char *name, *hostname;
439 uint64_t version, guid;
441 nvlist_t **child = NULL;
443 uint64_t *hole_array, max_id;
448 boolean_t found_one = B_FALSE;
449 boolean_t valid_top_config = B_FALSE;
451 if (nvlist_alloc(&ret, 0, 0) != 0)
454 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
455 uint64_t id, max_txg = 0;
457 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
459 config_seen = B_FALSE;
462 * Iterate over all toplevel vdevs. Grab the pool configuration
463 * from the first one we find, and then go through the rest and
464 * add them as necessary to the 'vdevs' member of the config.
466 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
469 * Determine the best configuration for this vdev by
470 * selecting the config with the latest transaction
474 for (ce = ve->ve_configs; ce != NULL;
477 if (ce->ce_txg > best_txg) {
479 best_txg = ce->ce_txg;
484 * We rely on the fact that the max txg for the
485 * pool will contain the most up-to-date information
486 * about the valid top-levels in the vdev namespace.
488 if (best_txg > max_txg) {
489 (void) nvlist_remove(config,
490 ZPOOL_CONFIG_VDEV_CHILDREN,
492 (void) nvlist_remove(config,
493 ZPOOL_CONFIG_HOLE_ARRAY,
494 DATA_TYPE_UINT64_ARRAY);
500 valid_top_config = B_FALSE;
502 if (nvlist_lookup_uint64(tmp,
503 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
504 verify(nvlist_add_uint64(config,
505 ZPOOL_CONFIG_VDEV_CHILDREN,
507 valid_top_config = B_TRUE;
510 if (nvlist_lookup_uint64_array(tmp,
511 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
513 verify(nvlist_add_uint64_array(config,
514 ZPOOL_CONFIG_HOLE_ARRAY,
515 hole_array, holes) == 0);
521 * Copy the relevant pieces of data to the pool
528 * hostid (if available)
529 * hostname (if available)
533 verify(nvlist_lookup_uint64(tmp,
534 ZPOOL_CONFIG_VERSION, &version) == 0);
535 if (nvlist_add_uint64(config,
536 ZPOOL_CONFIG_VERSION, version) != 0)
538 verify(nvlist_lookup_uint64(tmp,
539 ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
540 if (nvlist_add_uint64(config,
541 ZPOOL_CONFIG_POOL_GUID, guid) != 0)
543 verify(nvlist_lookup_string(tmp,
544 ZPOOL_CONFIG_POOL_NAME, &name) == 0);
545 if (nvlist_add_string(config,
546 ZPOOL_CONFIG_POOL_NAME, name) != 0)
548 verify(nvlist_lookup_uint64(tmp,
549 ZPOOL_CONFIG_POOL_STATE, &state) == 0);
550 if (nvlist_add_uint64(config,
551 ZPOOL_CONFIG_POOL_STATE, state) != 0)
554 if (nvlist_lookup_uint64(tmp,
555 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
556 if (nvlist_add_uint64(config,
557 ZPOOL_CONFIG_HOSTID, hostid) != 0)
559 verify(nvlist_lookup_string(tmp,
560 ZPOOL_CONFIG_HOSTNAME,
562 if (nvlist_add_string(config,
563 ZPOOL_CONFIG_HOSTNAME,
568 config_seen = B_TRUE;
572 * Add this top-level vdev to the child array.
574 verify(nvlist_lookup_nvlist(tmp,
575 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
576 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
579 if (id >= children) {
582 newchild = zfs_alloc(hdl, (id + 1) *
583 sizeof (nvlist_t *));
584 if (newchild == NULL)
587 for (c = 0; c < children; c++)
588 newchild[c] = child[c];
594 if (nvlist_dup(nvtop, &child[id], 0) != 0)
600 * If we have information about all the top-levels then
601 * clean up the nvlist which we've constructed. This
602 * means removing any extraneous devices that are
603 * beyond the valid range or adding devices to the end
604 * of our array which appear to be missing.
606 if (valid_top_config) {
607 if (max_id < children) {
608 for (c = max_id; c < children; c++)
609 nvlist_free(child[c]);
611 } else if (max_id > children) {
614 newchild = zfs_alloc(hdl, (max_id) *
615 sizeof (nvlist_t *));
616 if (newchild == NULL)
619 for (c = 0; c < children; c++)
620 newchild[c] = child[c];
628 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
632 * The vdev namespace may contain holes as a result of
633 * device removal. We must add them back into the vdev
634 * tree before we process any missing devices.
637 ASSERT(valid_top_config);
639 for (c = 0; c < children; c++) {
642 if (child[c] != NULL ||
643 !vdev_is_hole(hole_array, holes, c))
646 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
651 * Holes in the namespace are treated as
652 * "hole" top-level vdevs and have a
653 * special flag set on them.
655 if (nvlist_add_string(holey,
657 VDEV_TYPE_HOLE) != 0 ||
658 nvlist_add_uint64(holey,
659 ZPOOL_CONFIG_ID, c) != 0 ||
660 nvlist_add_uint64(holey,
661 ZPOOL_CONFIG_GUID, 0ULL) != 0)
668 * Look for any missing top-level vdevs. If this is the case,
669 * create a faked up 'missing' vdev as a placeholder. We cannot
670 * simply compress the child array, because the kernel performs
671 * certain checks to make sure the vdev IDs match their location
672 * in the configuration.
674 for (c = 0; c < children; c++) {
675 if (child[c] == NULL) {
677 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
680 if (nvlist_add_string(missing,
682 VDEV_TYPE_MISSING) != 0 ||
683 nvlist_add_uint64(missing,
684 ZPOOL_CONFIG_ID, c) != 0 ||
685 nvlist_add_uint64(missing,
686 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
687 nvlist_free(missing);
695 * Put all of this pool's top-level vdevs into a root vdev.
697 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
699 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
700 VDEV_TYPE_ROOT) != 0 ||
701 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
702 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
703 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
704 child, children) != 0) {
709 for (c = 0; c < children; c++)
710 nvlist_free(child[c]);
716 * Go through and fix up any paths and/or devids based on our
717 * known list of vdev GUID -> path mappings.
719 if (fix_paths(nvroot, pl->names) != 0) {
725 * Add the root vdev to this pool's configuration.
727 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
735 * zdb uses this path to report on active pools that were
736 * imported or created using -R.
742 * Determine if this pool is currently active, in which case we
743 * can't actually import it.
745 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
747 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
750 if (pool_active(hdl, name, guid, &isactive) != 0)
759 if ((nvl = refresh_config(hdl, config)) == NULL) {
769 * Go through and update the paths for spares, now that we have
772 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
774 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
775 &spares, &nspares) == 0) {
776 for (i = 0; i < nspares; i++) {
777 if (fix_paths(spares[i], pl->names) != 0)
783 * Update the paths for l2cache devices.
785 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
786 &l2cache, &nl2cache) == 0) {
787 for (i = 0; i < nl2cache; i++) {
788 if (fix_paths(l2cache[i], pl->names) != 0)
794 * Restore the original information read from the actual label.
796 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
798 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
801 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
803 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
809 * Add this pool to the list of configs.
811 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
813 if (nvlist_add_nvlist(ret, name, config) != 0)
829 (void) no_memory(hdl);
833 for (c = 0; c < children; c++)
834 nvlist_free(child[c]);
841 * Return the offset of the given label.
844 label_offset(uint64_t size, int l)
846 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
847 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
848 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
852 * Given a file descriptor, read the label information and return an nvlist
853 * describing the configuration, if there is one.
856 zpool_read_label(int fd, nvlist_t **config)
858 struct stat64 statbuf;
861 uint64_t state, txg, size;
865 if (fstat64(fd, &statbuf) == -1)
867 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
869 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
872 for (l = 0; l < VDEV_LABELS; l++) {
873 if (pread64(fd, label, sizeof (vdev_label_t),
874 label_offset(size, l)) != sizeof (vdev_label_t))
877 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
878 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
881 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
882 &state) != 0 || state > POOL_STATE_L2CACHE) {
883 nvlist_free(*config);
887 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
888 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
889 &txg) != 0 || txg == 0)) {
890 nvlist_free(*config);
903 typedef struct rdsk_node {
906 libzfs_handle_t *rn_hdl;
910 boolean_t rn_nozpool;
914 slice_cache_compare(const void *arg1, const void *arg2)
916 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
917 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
918 char *nm1slice, *nm2slice;
922 * slices zero and two are the most likely to provide results,
925 nm1slice = strstr(nm1, "s0");
926 nm2slice = strstr(nm2, "s0");
927 if (nm1slice && !nm2slice) {
930 if (!nm1slice && nm2slice) {
933 nm1slice = strstr(nm1, "s2");
934 nm2slice = strstr(nm2, "s2");
935 if (nm1slice && !nm2slice) {
938 if (!nm1slice && nm2slice) {
942 rv = strcmp(nm1, nm2);
945 return (rv > 0 ? 1 : -1);
950 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
951 diskaddr_t size, uint_t blksz)
955 char sname[MAXNAMELEN];
957 tmpnode.rn_name = &sname[0];
958 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
961 * protect against division by zero for disk labels that
962 * contain a bogus sector size
966 /* too small to contain a zpool? */
967 if ((size < (SPA_MINDEVSIZE / blksz)) &&
968 (node = avl_find(r, &tmpnode, NULL)))
969 node->rn_nozpool = B_TRUE;
974 nozpool_all_slices(avl_tree_t *r, const char *sname)
977 char diskname[MAXNAMELEN];
981 (void) strncpy(diskname, sname, MAXNAMELEN);
982 if (((ptr = strrchr(diskname, 's')) == NULL) &&
983 ((ptr = strrchr(diskname, 'p')) == NULL))
987 for (i = 0; i < NDKMAP; i++)
988 check_one_slice(r, diskname, i, 0, 1);
990 for (i = 0; i <= FD_NUMPART; i++)
991 check_one_slice(r, diskname, i, 0, 1);
996 check_slices(avl_tree_t *r, int fd, const char *sname)
1001 char diskname[MAXNAMELEN];
1005 (void) strncpy(diskname, sname, MAXNAMELEN);
1006 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1010 if (read_extvtoc(fd, &vtoc) >= 0) {
1011 for (i = 0; i < NDKMAP; i++)
1012 check_one_slice(r, diskname, i,
1013 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1014 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1016 * on x86 we'll still have leftover links that point
1017 * to slices s[9-15], so use NDKMAP instead
1019 for (i = 0; i < NDKMAP; i++)
1020 check_one_slice(r, diskname, i,
1021 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1022 /* nodes p[1-4] are never used with EFI labels */
1024 for (i = 1; i <= FD_NUMPART; i++)
1025 check_one_slice(r, diskname, i, 0, 1);
1032 zpool_open_func(void *arg)
1034 rdsk_node_t *rn = arg;
1035 struct stat64 statbuf;
1041 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1042 /* symlink to a device that's no longer there */
1043 if (errno == ENOENT)
1044 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1048 * Ignore failed stats. We only want regular
1049 * files, character devs and block devs.
1051 if (fstat64(fd, &statbuf) != 0 ||
1052 (!S_ISREG(statbuf.st_mode) &&
1053 !S_ISCHR(statbuf.st_mode) &&
1054 !S_ISBLK(statbuf.st_mode))) {
1058 /* this file is too small to hold a zpool */
1059 if (S_ISREG(statbuf.st_mode) &&
1060 statbuf.st_size < SPA_MINDEVSIZE) {
1063 } else if (!S_ISREG(statbuf.st_mode)) {
1065 * Try to read the disk label first so we don't have to
1066 * open a bunch of minor nodes that can't have a zpool.
1068 check_slices(rn->rn_avl, fd, rn->rn_name);
1071 if ((zpool_read_label(fd, &config)) != 0) {
1073 (void) no_memory(rn->rn_hdl);
1079 rn->rn_config = config;
1080 if (config != NULL) {
1081 assert(rn->rn_nozpool == B_FALSE);
1086 * Given a file descriptor, clear (zero) the label information. This function
1087 * is used in the appliance stack as part of the ZFS sysevent module and
1088 * to implement the "zpool labelclear" command.
1091 zpool_clear_label(int fd)
1093 struct stat64 statbuf;
1095 vdev_label_t *label;
1098 if (fstat64(fd, &statbuf) == -1)
1100 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1102 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1105 for (l = 0; l < VDEV_LABELS; l++) {
1106 if (pwrite64(fd, label, sizeof (vdev_label_t),
1107 label_offset(size, l)) != sizeof (vdev_label_t))
1116 * Given a list of directories to search, find all pools stored on disk. This
1117 * includes partial pools which are not available to import. If no args are
1118 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1119 * poolname or guid (but not both) are provided by the caller when trying
1120 * to import a specific pool.
1123 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1125 int i, dirs = iarg->paths;
1127 struct dirent64 *dp;
1128 char path[MAXPATHLEN];
1129 char *end, **dir = iarg->path;
1131 nvlist_t *ret = NULL;
1132 static char *default_dir = "/dev/dsk";
1133 pool_list_t pools = { 0 };
1134 pool_entry_t *pe, *penext;
1135 vdev_entry_t *ve, *venext;
1136 config_entry_t *ce, *cenext;
1137 name_entry_t *ne, *nenext;
1138 avl_tree_t slice_cache;
1148 * Go through and read the label configuration information from every
1149 * possible device, organizing the information according to pool GUID
1150 * and toplevel GUID.
1152 for (i = 0; i < dirs; i++) {
1157 /* use realpath to normalize the path */
1158 if (realpath(dir[i], path) == 0) {
1159 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1160 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1163 end = &path[strlen(path)];
1166 pathleft = &path[sizeof (path)] - end;
1169 * Using raw devices instead of block devices when we're
1170 * reading the labels skips a bunch of slow operations during
1171 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1173 if (strcmp(path, "/dev/dsk/") == 0)
1178 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1179 (dirp = fdopendir(dfd)) == NULL) {
1180 zfs_error_aux(hdl, strerror(errno));
1181 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1182 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1187 avl_create(&slice_cache, slice_cache_compare,
1188 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1190 if (strcmp(rdsk, "/dev/") == 0) {
1194 struct gprovider *pp;
1196 errno = geom_gettree(&mesh);
1198 zfs_error_aux(hdl, strerror(errno));
1199 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1200 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1204 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1205 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1206 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1207 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1208 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1209 slice->rn_avl = &slice_cache;
1210 slice->rn_dfd = dfd;
1211 slice->rn_hdl = hdl;
1212 slice->rn_nozpool = B_FALSE;
1213 avl_add(&slice_cache, slice);
1218 geom_deletetree(&mesh);
1223 * This is not MT-safe, but we have no MT consumers of libzfs
1225 while ((dp = readdir64(dirp)) != NULL) {
1226 const char *name = dp->d_name;
1227 if (name[0] == '.' &&
1228 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1231 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1232 slice->rn_name = zfs_strdup(hdl, name);
1233 slice->rn_avl = &slice_cache;
1234 slice->rn_dfd = dfd;
1235 slice->rn_hdl = hdl;
1236 slice->rn_nozpool = B_FALSE;
1237 avl_add(&slice_cache, slice);
1241 * create a thread pool to do all of this in parallel;
1242 * rn_nozpool is not protected, so this is racy in that
1243 * multiple tasks could decide that the same slice can
1244 * not hold a zpool, which is benign. Also choose
1245 * double the number of processors; we hold a lot of
1246 * locks in the kernel, so going beyond this doesn't
1249 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1251 for (slice = avl_first(&slice_cache); slice;
1252 (slice = avl_walk(&slice_cache, slice,
1254 (void) tpool_dispatch(t, zpool_open_func, slice);
1259 while ((slice = avl_destroy_nodes(&slice_cache,
1260 &cookie)) != NULL) {
1261 if (slice->rn_config != NULL) {
1262 nvlist_t *config = slice->rn_config;
1263 boolean_t matched = B_TRUE;
1265 if (iarg->poolname != NULL) {
1268 matched = nvlist_lookup_string(config,
1269 ZPOOL_CONFIG_POOL_NAME,
1271 strcmp(iarg->poolname, pname) == 0;
1272 } else if (iarg->guid != 0) {
1275 matched = nvlist_lookup_uint64(config,
1276 ZPOOL_CONFIG_POOL_GUID,
1278 iarg->guid == this_guid;
1281 nvlist_free(config);
1285 /* use the non-raw path for the config */
1286 (void) strlcpy(end, slice->rn_name, pathleft);
1287 if (add_config(hdl, &pools, path, config) != 0)
1290 free(slice->rn_name);
1293 avl_destroy(&slice_cache);
1295 (void) closedir(dirp);
1299 ret = get_configs(hdl, &pools, iarg->can_be_active);
1302 for (pe = pools.pools; pe != NULL; pe = penext) {
1303 penext = pe->pe_next;
1304 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1305 venext = ve->ve_next;
1306 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1307 cenext = ce->ce_next;
1309 nvlist_free(ce->ce_config);
1317 for (ne = pools.names; ne != NULL; ne = nenext) {
1318 nenext = ne->ne_next;
1325 (void) closedir(dirp);
1331 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1333 importargs_t iarg = { 0 };
1338 return (zpool_find_import_impl(hdl, &iarg));
1342 * Given a cache file, return the contents as a list of importable pools.
1343 * poolname or guid (but not both) are provided by the caller when trying
1344 * to import a specific pool.
1347 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1348 char *poolname, uint64_t guid)
1352 struct stat64 statbuf;
1353 nvlist_t *raw, *src, *dst;
1360 verify(poolname == NULL || guid == 0);
1362 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1363 zfs_error_aux(hdl, "%s", strerror(errno));
1364 (void) zfs_error(hdl, EZFS_BADCACHE,
1365 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1369 if (fstat64(fd, &statbuf) != 0) {
1370 zfs_error_aux(hdl, "%s", strerror(errno));
1372 (void) zfs_error(hdl, EZFS_BADCACHE,
1373 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1377 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1382 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1385 (void) zfs_error(hdl, EZFS_BADCACHE,
1386 dgettext(TEXT_DOMAIN,
1387 "failed to read cache file contents"));
1393 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1395 (void) zfs_error(hdl, EZFS_BADCACHE,
1396 dgettext(TEXT_DOMAIN,
1397 "invalid or corrupt cache file contents"));
1404 * Go through and get the current state of the pools and refresh their
1407 if (nvlist_alloc(&pools, 0, 0) != 0) {
1408 (void) no_memory(hdl);
1414 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1415 verify(nvpair_value_nvlist(elem, &src) == 0);
1417 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1419 if (poolname != NULL && strcmp(poolname, name) != 0)
1422 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1425 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1427 if (guid != this_guid)
1431 if (pool_active(hdl, name, this_guid, &active) != 0) {
1440 if ((dst = refresh_config(hdl, src)) == NULL) {
1446 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1447 (void) no_memory(hdl);
1461 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1463 importargs_t *import = data;
1466 if (import->poolname != NULL) {
1469 verify(nvlist_lookup_string(zhp->zpool_config,
1470 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1471 if (strcmp(pool_name, import->poolname) == 0)
1476 verify(nvlist_lookup_uint64(zhp->zpool_config,
1477 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1478 if (pool_guid == import->guid)
1487 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1489 verify(import->poolname == NULL || import->guid == 0);
1492 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1494 if (import->cachefile != NULL)
1495 return (zpool_find_import_cached(hdl, import->cachefile,
1496 import->poolname, import->guid));
1498 return (zpool_find_import_impl(hdl, import));
1502 find_guid(nvlist_t *nv, uint64_t guid)
1508 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1512 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1513 &child, &children) == 0) {
1514 for (c = 0; c < children; c++)
1515 if (find_guid(child[c], guid))
1522 typedef struct aux_cbdata {
1523 const char *cb_type;
1525 zpool_handle_t *cb_zhp;
1529 find_aux(zpool_handle_t *zhp, void *data)
1531 aux_cbdata_t *cbp = data;
1537 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1540 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1541 &list, &count) == 0) {
1542 for (i = 0; i < count; i++) {
1543 verify(nvlist_lookup_uint64(list[i],
1544 ZPOOL_CONFIG_GUID, &guid) == 0);
1545 if (guid == cbp->cb_guid) {
1557 * Determines if the pool is in use. If so, it returns true and the state of
1558 * the pool as well as the name of the pool. Both strings are allocated and
1559 * must be freed by the caller.
1562 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1568 uint64_t guid, vdev_guid;
1569 zpool_handle_t *zhp;
1570 nvlist_t *pool_config;
1571 uint64_t stateval, isspare;
1572 aux_cbdata_t cb = { 0 };
1577 if (zpool_read_label(fd, &config) != 0) {
1578 (void) no_memory(hdl);
1585 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1587 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1590 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1591 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1593 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1598 case POOL_STATE_EXPORTED:
1600 * A pool with an exported state may in fact be imported
1601 * read-only, so check the in-core state to see if it's
1602 * active and imported read-only. If it is, set
1603 * its state to active.
1605 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1606 (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1607 zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1608 stateval = POOL_STATE_ACTIVE;
1613 case POOL_STATE_ACTIVE:
1615 * For an active pool, we have to determine if it's really part
1616 * of a currently active pool (in which case the pool will exist
1617 * and the guid will be the same), or whether it's part of an
1618 * active pool that was disconnected without being explicitly
1621 if (pool_active(hdl, name, guid, &isactive) != 0) {
1622 nvlist_free(config);
1628 * Because the device may have been removed while
1629 * offlined, we only report it as active if the vdev is
1630 * still present in the config. Otherwise, pretend like
1633 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1634 (pool_config = zpool_get_config(zhp, NULL))
1638 verify(nvlist_lookup_nvlist(pool_config,
1639 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1640 ret = find_guid(nvroot, vdev_guid);
1646 * If this is an active spare within another pool, we
1647 * treat it like an unused hot spare. This allows the
1648 * user to create a pool with a hot spare that currently
1649 * in use within another pool. Since we return B_TRUE,
1650 * libdiskmgt will continue to prevent generic consumers
1651 * from using the device.
1653 if (ret && nvlist_lookup_uint64(config,
1654 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1655 stateval = POOL_STATE_SPARE;
1660 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1665 case POOL_STATE_SPARE:
1667 * For a hot spare, it can be either definitively in use, or
1668 * potentially active. To determine if it's in use, we iterate
1669 * over all pools in the system and search for one with a spare
1670 * with a matching guid.
1672 * Due to the shared nature of spares, we don't actually report
1673 * the potentially active case as in use. This means the user
1674 * can freely create pools on the hot spares of exported pools,
1675 * but to do otherwise makes the resulting code complicated, and
1676 * we end up having to deal with this case anyway.
1679 cb.cb_guid = vdev_guid;
1680 cb.cb_type = ZPOOL_CONFIG_SPARES;
1681 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1682 name = (char *)zpool_get_name(cb.cb_zhp);
1689 case POOL_STATE_L2CACHE:
1692 * Check if any pool is currently using this l2cache device.
1695 cb.cb_guid = vdev_guid;
1696 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1697 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1698 name = (char *)zpool_get_name(cb.cb_zhp);
1711 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1713 zpool_close(cb.cb_zhp);
1714 nvlist_free(config);
1717 *state = (pool_state_t)stateval;
1721 zpool_close(cb.cb_zhp);
1723 nvlist_free(config);