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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 RackTop Systems.
26 * Copyright 2016 Nexenta Systems, Inc.
30 * Pool import support functions.
32 * To import a pool, we rely on reading the configuration information from the
33 * ZFS label of each device. If we successfully read the label, then we
34 * organize the configuration information in the following hierarchy:
36 * pool guid -> toplevel vdev guid -> label txg
38 * Duplicate entries matching this same tuple will be discarded. Once we have
39 * examined every device, we pick the best label txg config for each toplevel
40 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
41 * update any paths that have changed. Finally, we attempt to import the pool
42 * using our derived config, and record the results.
56 #include <thread_pool.h>
59 #include <sys/vdev_impl.h>
62 #include "libzfs_impl.h"
65 * Intermediate structures used to gather configuration information.
67 typedef struct config_entry {
70 struct config_entry *ce_next;
73 typedef struct vdev_entry {
75 config_entry_t *ve_configs;
76 struct vdev_entry *ve_next;
79 typedef struct pool_entry {
81 vdev_entry_t *pe_vdevs;
82 struct pool_entry *pe_next;
85 typedef struct name_entry {
88 struct name_entry *ne_next;
91 typedef struct pool_list {
97 get_devid(const char *path)
104 if ((fd = open(path, O_RDONLY)) < 0)
109 if (devid_get(fd, &devid) == 0) {
110 if (devid_get_minor_name(fd, &minor) == 0)
111 ret = devid_str_encode(devid, minor);
113 devid_str_free(minor);
126 * Go through and fix up any path and/or devid information for the given vdev
130 fix_paths(nvlist_t *nv, name_entry_t *names)
135 name_entry_t *ne, *best;
139 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
140 &child, &children) == 0) {
141 for (c = 0; c < children; c++)
142 if (fix_paths(child[c], names) != 0)
148 * This is a leaf (file or disk) vdev. In either case, go through
149 * the name list and see if we find a matching guid. If so, replace
150 * the path and see if we can calculate a new devid.
152 * There may be multiple names associated with a particular guid, in
153 * which case we have overlapping slices or multiple paths to the same
154 * disk. If this is the case, then we want to pick the path that is
155 * the most similar to the original, where "most similar" is the number
156 * of matching characters starting from the end of the path. This will
157 * preserve slice numbers even if the disks have been reorganized, and
158 * will also catch preferred disk names if multiple paths exist.
160 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
161 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
166 for (ne = names; ne != NULL; ne = ne->ne_next) {
167 if (ne->ne_guid == guid) {
168 const char *src, *dst;
176 src = ne->ne_name + strlen(ne->ne_name) - 1;
177 dst = path + strlen(path) - 1;
178 for (count = 0; src >= ne->ne_name && dst >= path;
179 src--, dst--, count++)
184 * At this point, 'count' is the number of characters
185 * matched from the end.
187 if (count > matched || best == NULL) {
197 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
200 if ((devid = get_devid(best->ne_name)) == NULL) {
201 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
203 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
204 devid_str_free(devid);
207 devid_str_free(devid);
214 * Add the given configuration to the list of known devices.
217 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
220 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
227 * If this is a hot spare not currently in use or level 2 cache
228 * device, add it to the list of names to translate, but don't do
231 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
233 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
234 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
235 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
238 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
242 ne->ne_guid = vdev_guid;
243 ne->ne_next = pl->names;
249 * If we have a valid config but cannot read any of these fields, then
250 * it means we have a half-initialized label. In vdev_label_init()
251 * we write a label with txg == 0 so that we can identify the device
252 * in case the user refers to the same disk later on. If we fail to
253 * create the pool, we'll be left with a label in this state
254 * which should not be considered part of a valid pool.
256 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
258 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
260 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
262 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
263 &txg) != 0 || txg == 0) {
269 * First, see if we know about this pool. If not, then add it to the
270 * list of known pools.
272 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
273 if (pe->pe_guid == pool_guid)
278 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
282 pe->pe_guid = pool_guid;
283 pe->pe_next = pl->pools;
288 * Second, see if we know about this toplevel vdev. Add it if its
291 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
292 if (ve->ve_guid == top_guid)
297 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
301 ve->ve_guid = top_guid;
302 ve->ve_next = pe->pe_vdevs;
307 * Third, see if we have a config with a matching transaction group. If
308 * so, then we do nothing. Otherwise, add it to the list of known
311 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
312 if (ce->ce_txg == txg)
317 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
322 ce->ce_config = config;
323 ce->ce_next = ve->ve_configs;
330 * At this point we've successfully added our config to the list of
331 * known configs. The last thing to do is add the vdev guid -> path
332 * mappings so that we can fix up the configuration as necessary before
335 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
338 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
343 ne->ne_guid = vdev_guid;
344 ne->ne_next = pl->names;
351 * Returns true if the named pool matches the given GUID.
354 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
360 if (zpool_open_silent(hdl, name, &zhp) != 0)
368 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
373 *isactive = (theguid == guid);
378 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
381 zfs_cmd_t zc = { 0 };
384 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
387 if (zcmd_alloc_dst_nvlist(hdl, &zc,
388 zc.zc_nvlist_conf_size * 2) != 0) {
389 zcmd_free_nvlists(&zc);
393 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
394 &zc)) != 0 && errno == ENOMEM) {
395 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
396 zcmd_free_nvlists(&zc);
402 zcmd_free_nvlists(&zc);
406 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
407 zcmd_free_nvlists(&zc);
411 zcmd_free_nvlists(&zc);
416 * Determine if the vdev id is a hole in the namespace.
419 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
421 for (int c = 0; c < holes; c++) {
423 /* Top-level is a hole */
424 if (hole_array[c] == id)
431 * Convert our list of pools into the definitive set of configurations. We
432 * start by picking the best config for each toplevel vdev. Once that's done,
433 * we assemble the toplevel vdevs into a full config for the pool. We make a
434 * pass to fix up any incorrect paths, and then add it to the main list to
435 * return to the user.
438 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
443 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
444 nvlist_t **spares, **l2cache;
445 uint_t i, nspares, nl2cache;
446 boolean_t config_seen;
448 char *name, *hostname = NULL;
451 nvlist_t **child = NULL;
453 uint64_t *hole_array, max_id;
458 boolean_t found_one = B_FALSE;
459 boolean_t valid_top_config = B_FALSE;
461 if (nvlist_alloc(&ret, 0, 0) != 0)
464 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
465 uint64_t id, max_txg = 0;
467 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
469 config_seen = B_FALSE;
472 * Iterate over all toplevel vdevs. Grab the pool configuration
473 * from the first one we find, and then go through the rest and
474 * add them as necessary to the 'vdevs' member of the config.
476 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
479 * Determine the best configuration for this vdev by
480 * selecting the config with the latest transaction
484 for (ce = ve->ve_configs; ce != NULL;
487 if (ce->ce_txg > best_txg) {
489 best_txg = ce->ce_txg;
494 * We rely on the fact that the max txg for the
495 * pool will contain the most up-to-date information
496 * about the valid top-levels in the vdev namespace.
498 if (best_txg > max_txg) {
499 (void) nvlist_remove(config,
500 ZPOOL_CONFIG_VDEV_CHILDREN,
502 (void) nvlist_remove(config,
503 ZPOOL_CONFIG_HOLE_ARRAY,
504 DATA_TYPE_UINT64_ARRAY);
510 valid_top_config = B_FALSE;
512 if (nvlist_lookup_uint64(tmp,
513 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
514 verify(nvlist_add_uint64(config,
515 ZPOOL_CONFIG_VDEV_CHILDREN,
517 valid_top_config = B_TRUE;
520 if (nvlist_lookup_uint64_array(tmp,
521 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
523 verify(nvlist_add_uint64_array(config,
524 ZPOOL_CONFIG_HOLE_ARRAY,
525 hole_array, holes) == 0);
531 * Copy the relevant pieces of data to the pool
537 * comment (if available)
539 * hostid (if available)
540 * hostname (if available)
542 uint64_t state, version;
543 char *comment = NULL;
545 version = fnvlist_lookup_uint64(tmp,
546 ZPOOL_CONFIG_VERSION);
547 fnvlist_add_uint64(config,
548 ZPOOL_CONFIG_VERSION, version);
549 guid = fnvlist_lookup_uint64(tmp,
550 ZPOOL_CONFIG_POOL_GUID);
551 fnvlist_add_uint64(config,
552 ZPOOL_CONFIG_POOL_GUID, guid);
553 name = fnvlist_lookup_string(tmp,
554 ZPOOL_CONFIG_POOL_NAME);
555 fnvlist_add_string(config,
556 ZPOOL_CONFIG_POOL_NAME, name);
558 if (nvlist_lookup_string(tmp,
559 ZPOOL_CONFIG_COMMENT, &comment) == 0)
560 fnvlist_add_string(config,
561 ZPOOL_CONFIG_COMMENT, comment);
563 state = fnvlist_lookup_uint64(tmp,
564 ZPOOL_CONFIG_POOL_STATE);
565 fnvlist_add_uint64(config,
566 ZPOOL_CONFIG_POOL_STATE, state);
569 if (nvlist_lookup_uint64(tmp,
570 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
571 fnvlist_add_uint64(config,
572 ZPOOL_CONFIG_HOSTID, hostid);
573 hostname = fnvlist_lookup_string(tmp,
574 ZPOOL_CONFIG_HOSTNAME);
575 fnvlist_add_string(config,
576 ZPOOL_CONFIG_HOSTNAME, hostname);
579 config_seen = B_TRUE;
583 * Add this top-level vdev to the child array.
585 verify(nvlist_lookup_nvlist(tmp,
586 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
587 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
590 if (id >= children) {
593 newchild = zfs_alloc(hdl, (id + 1) *
594 sizeof (nvlist_t *));
595 if (newchild == NULL)
598 for (c = 0; c < children; c++)
599 newchild[c] = child[c];
605 if (nvlist_dup(nvtop, &child[id], 0) != 0)
611 * If we have information about all the top-levels then
612 * clean up the nvlist which we've constructed. This
613 * means removing any extraneous devices that are
614 * beyond the valid range or adding devices to the end
615 * of our array which appear to be missing.
617 if (valid_top_config) {
618 if (max_id < children) {
619 for (c = max_id; c < children; c++)
620 nvlist_free(child[c]);
622 } else if (max_id > children) {
625 newchild = zfs_alloc(hdl, (max_id) *
626 sizeof (nvlist_t *));
627 if (newchild == NULL)
630 for (c = 0; c < children; c++)
631 newchild[c] = child[c];
639 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
643 * The vdev namespace may contain holes as a result of
644 * device removal. We must add them back into the vdev
645 * tree before we process any missing devices.
648 ASSERT(valid_top_config);
650 for (c = 0; c < children; c++) {
653 if (child[c] != NULL ||
654 !vdev_is_hole(hole_array, holes, c))
657 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
662 * Holes in the namespace are treated as
663 * "hole" top-level vdevs and have a
664 * special flag set on them.
666 if (nvlist_add_string(holey,
668 VDEV_TYPE_HOLE) != 0 ||
669 nvlist_add_uint64(holey,
670 ZPOOL_CONFIG_ID, c) != 0 ||
671 nvlist_add_uint64(holey,
672 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
681 * Look for any missing top-level vdevs. If this is the case,
682 * create a faked up 'missing' vdev as a placeholder. We cannot
683 * simply compress the child array, because the kernel performs
684 * certain checks to make sure the vdev IDs match their location
685 * in the configuration.
687 for (c = 0; c < children; c++) {
688 if (child[c] == NULL) {
690 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
693 if (nvlist_add_string(missing,
695 VDEV_TYPE_MISSING) != 0 ||
696 nvlist_add_uint64(missing,
697 ZPOOL_CONFIG_ID, c) != 0 ||
698 nvlist_add_uint64(missing,
699 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
700 nvlist_free(missing);
708 * Put all of this pool's top-level vdevs into a root vdev.
710 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
712 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
713 VDEV_TYPE_ROOT) != 0 ||
714 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
715 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
716 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
717 child, children) != 0) {
722 for (c = 0; c < children; c++)
723 nvlist_free(child[c]);
729 * Go through and fix up any paths and/or devids based on our
730 * known list of vdev GUID -> path mappings.
732 if (fix_paths(nvroot, pl->names) != 0) {
738 * Add the root vdev to this pool's configuration.
740 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
748 * zdb uses this path to report on active pools that were
749 * imported or created using -R.
755 * Determine if this pool is currently active, in which case we
756 * can't actually import it.
758 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
760 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
763 if (pool_active(hdl, name, guid, &isactive) != 0)
772 if ((nvl = refresh_config(hdl, config)) == NULL) {
782 * Go through and update the paths for spares, now that we have
785 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
787 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
788 &spares, &nspares) == 0) {
789 for (i = 0; i < nspares; i++) {
790 if (fix_paths(spares[i], pl->names) != 0)
796 * Update the paths for l2cache devices.
798 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
799 &l2cache, &nl2cache) == 0) {
800 for (i = 0; i < nl2cache; i++) {
801 if (fix_paths(l2cache[i], pl->names) != 0)
807 * Restore the original information read from the actual label.
809 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
811 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
814 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
816 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
822 * Add this pool to the list of configs.
824 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
826 if (nvlist_add_nvlist(ret, name, config) != 0)
842 (void) no_memory(hdl);
846 for (c = 0; c < children; c++)
847 nvlist_free(child[c]);
854 * Return the offset of the given label.
857 label_offset(uint64_t size, int l)
859 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
860 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
861 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
865 * Given a file descriptor, read the label information and return an nvlist
866 * describing the configuration, if there is one.
867 * Return 0 on success, or -1 on failure
870 zpool_read_label(int fd, nvlist_t **config)
872 struct stat64 statbuf;
875 uint64_t state, txg, size;
879 if (fstat64(fd, &statbuf) == -1)
881 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
883 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
886 for (l = 0; l < VDEV_LABELS; l++) {
887 if (pread64(fd, label, sizeof (vdev_label_t),
888 label_offset(size, l)) != sizeof (vdev_label_t))
891 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
892 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
895 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
896 &state) != 0 || state > POOL_STATE_L2CACHE) {
897 nvlist_free(*config);
901 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
902 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
903 &txg) != 0 || txg == 0)) {
904 nvlist_free(*config);
918 typedef struct rdsk_node {
921 libzfs_handle_t *rn_hdl;
925 boolean_t rn_nozpool;
929 slice_cache_compare(const void *arg1, const void *arg2)
931 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
932 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
933 char *nm1slice, *nm2slice;
937 * slices zero and two are the most likely to provide results,
940 nm1slice = strstr(nm1, "s0");
941 nm2slice = strstr(nm2, "s0");
942 if (nm1slice && !nm2slice) {
945 if (!nm1slice && nm2slice) {
948 nm1slice = strstr(nm1, "s2");
949 nm2slice = strstr(nm2, "s2");
950 if (nm1slice && !nm2slice) {
953 if (!nm1slice && nm2slice) {
957 rv = strcmp(nm1, nm2);
960 return (rv > 0 ? 1 : -1);
965 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
966 diskaddr_t size, uint_t blksz)
970 char sname[MAXNAMELEN];
972 tmpnode.rn_name = &sname[0];
973 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
976 * protect against division by zero for disk labels that
977 * contain a bogus sector size
981 /* too small to contain a zpool? */
982 if ((size < (SPA_MINDEVSIZE / blksz)) &&
983 (node = avl_find(r, &tmpnode, NULL)))
984 node->rn_nozpool = B_TRUE;
989 nozpool_all_slices(avl_tree_t *r, const char *sname)
992 char diskname[MAXNAMELEN];
996 (void) strncpy(diskname, sname, MAXNAMELEN);
997 if (((ptr = strrchr(diskname, 's')) == NULL) &&
998 ((ptr = strrchr(diskname, 'p')) == NULL))
1002 for (i = 0; i < NDKMAP; i++)
1003 check_one_slice(r, diskname, i, 0, 1);
1005 for (i = 0; i <= FD_NUMPART; i++)
1006 check_one_slice(r, diskname, i, 0, 1);
1007 #endif /* illumos */
1012 check_slices(avl_tree_t *r, int fd, const char *sname)
1014 struct extvtoc vtoc;
1016 char diskname[MAXNAMELEN];
1020 (void) strncpy(diskname, sname, MAXNAMELEN);
1021 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1025 if (read_extvtoc(fd, &vtoc) >= 0) {
1026 for (i = 0; i < NDKMAP; i++)
1027 check_one_slice(r, diskname, i,
1028 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1029 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1031 * on x86 we'll still have leftover links that point
1032 * to slices s[9-15], so use NDKMAP instead
1034 for (i = 0; i < NDKMAP; i++)
1035 check_one_slice(r, diskname, i,
1036 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1037 /* nodes p[1-4] are never used with EFI labels */
1039 for (i = 1; i <= FD_NUMPART; i++)
1040 check_one_slice(r, diskname, i, 0, 1);
1044 #endif /* illumos */
1047 zpool_open_func(void *arg)
1049 rdsk_node_t *rn = arg;
1050 struct stat64 statbuf;
1056 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1057 /* symlink to a device that's no longer there */
1058 if (errno == ENOENT)
1059 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1063 * Ignore failed stats. We only want regular
1064 * files, character devs and block devs.
1066 if (fstat64(fd, &statbuf) != 0 ||
1067 (!S_ISREG(statbuf.st_mode) &&
1068 !S_ISCHR(statbuf.st_mode) &&
1069 !S_ISBLK(statbuf.st_mode))) {
1073 /* this file is too small to hold a zpool */
1075 if (S_ISREG(statbuf.st_mode) &&
1076 statbuf.st_size < SPA_MINDEVSIZE) {
1079 } else if (!S_ISREG(statbuf.st_mode)) {
1081 * Try to read the disk label first so we don't have to
1082 * open a bunch of minor nodes that can't have a zpool.
1084 check_slices(rn->rn_avl, fd, rn->rn_name);
1086 #else /* !illumos */
1087 if (statbuf.st_size < SPA_MINDEVSIZE) {
1091 #endif /* illumos */
1093 if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1095 (void) no_memory(rn->rn_hdl);
1100 rn->rn_config = config;
1104 * Given a file descriptor, clear (zero) the label information.
1107 zpool_clear_label(int fd)
1109 struct stat64 statbuf;
1111 vdev_label_t *label;
1114 if (fstat64(fd, &statbuf) == -1)
1116 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1118 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1121 for (l = 0; l < VDEV_LABELS; l++) {
1122 if (pwrite64(fd, label, sizeof (vdev_label_t),
1123 label_offset(size, l)) != sizeof (vdev_label_t)) {
1134 * Given a list of directories to search, find all pools stored on disk. This
1135 * includes partial pools which are not available to import. If no args are
1136 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1137 * poolname or guid (but not both) are provided by the caller when trying
1138 * to import a specific pool.
1141 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1143 int i, dirs = iarg->paths;
1144 struct dirent64 *dp;
1145 char path[MAXPATHLEN];
1146 char *end, **dir = iarg->path;
1148 nvlist_t *ret = NULL;
1149 static char *default_dir = "/dev";
1150 pool_list_t pools = { 0 };
1151 pool_entry_t *pe, *penext;
1152 vdev_entry_t *ve, *venext;
1153 config_entry_t *ce, *cenext;
1154 name_entry_t *ne, *nenext;
1155 avl_tree_t slice_cache;
1165 * Go through and read the label configuration information from every
1166 * possible device, organizing the information according to pool GUID
1167 * and toplevel GUID.
1169 for (i = 0; i < dirs; i++) {
1171 char rdsk[MAXPATHLEN];
1173 boolean_t config_failed = B_FALSE;
1176 /* use realpath to normalize the path */
1177 if (realpath(dir[i], path) == 0) {
1178 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1179 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1182 end = &path[strlen(path)];
1185 pathleft = &path[sizeof (path)] - end;
1189 * Using raw devices instead of block devices when we're
1190 * reading the labels skips a bunch of slow operations during
1191 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1193 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1194 (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1197 (void) strlcpy(rdsk, path, sizeof (rdsk));
1199 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1200 (dirp = fdopendir(dfd)) == NULL) {
1203 zfs_error_aux(hdl, strerror(errno));
1204 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1205 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1210 avl_create(&slice_cache, slice_cache_compare,
1211 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1213 if (strcmp(rdsk, "/dev/") == 0) {
1217 struct gprovider *pp;
1219 errno = geom_gettree(&mesh);
1221 zfs_error_aux(hdl, strerror(errno));
1222 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1223 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1227 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1228 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1229 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1230 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1231 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1232 slice->rn_avl = &slice_cache;
1233 slice->rn_dfd = dfd;
1234 slice->rn_hdl = hdl;
1235 slice->rn_nozpool = B_FALSE;
1236 avl_add(&slice_cache, slice);
1241 geom_deletetree(&mesh);
1246 * This is not MT-safe, but we have no MT consumers of libzfs
1248 while ((dp = readdir64(dirp)) != NULL) {
1249 const char *name = dp->d_name;
1250 if (name[0] == '.' &&
1251 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1254 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1255 slice->rn_name = zfs_strdup(hdl, name);
1256 slice->rn_avl = &slice_cache;
1257 slice->rn_dfd = dfd;
1258 slice->rn_hdl = hdl;
1259 slice->rn_nozpool = B_FALSE;
1260 avl_add(&slice_cache, slice);
1264 * create a thread pool to do all of this in parallel;
1265 * rn_nozpool is not protected, so this is racy in that
1266 * multiple tasks could decide that the same slice can
1267 * not hold a zpool, which is benign. Also choose
1268 * double the number of processors; we hold a lot of
1269 * locks in the kernel, so going beyond this doesn't
1272 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1274 for (slice = avl_first(&slice_cache); slice;
1275 (slice = avl_walk(&slice_cache, slice,
1277 (void) tpool_dispatch(t, zpool_open_func, slice);
1282 while ((slice = avl_destroy_nodes(&slice_cache,
1283 &cookie)) != NULL) {
1284 if (slice->rn_config != NULL && !config_failed) {
1285 nvlist_t *config = slice->rn_config;
1286 boolean_t matched = B_TRUE;
1288 if (iarg->poolname != NULL) {
1291 matched = nvlist_lookup_string(config,
1292 ZPOOL_CONFIG_POOL_NAME,
1294 strcmp(iarg->poolname, pname) == 0;
1295 } else if (iarg->guid != 0) {
1298 matched = nvlist_lookup_uint64(config,
1299 ZPOOL_CONFIG_POOL_GUID,
1301 iarg->guid == this_guid;
1304 nvlist_free(config);
1307 * use the non-raw path for the config
1309 (void) strlcpy(end, slice->rn_name,
1311 if (add_config(hdl, &pools, path,
1313 config_failed = B_TRUE;
1316 free(slice->rn_name);
1319 avl_destroy(&slice_cache);
1321 (void) closedir(dirp);
1327 ret = get_configs(hdl, &pools, iarg->can_be_active);
1330 for (pe = pools.pools; pe != NULL; pe = penext) {
1331 penext = pe->pe_next;
1332 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1333 venext = ve->ve_next;
1334 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1335 cenext = ce->ce_next;
1336 nvlist_free(ce->ce_config);
1344 for (ne = pools.names; ne != NULL; ne = nenext) {
1345 nenext = ne->ne_next;
1354 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1356 importargs_t iarg = { 0 };
1361 return (zpool_find_import_impl(hdl, &iarg));
1365 * Given a cache file, return the contents as a list of importable pools.
1366 * poolname or guid (but not both) are provided by the caller when trying
1367 * to import a specific pool.
1370 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1371 char *poolname, uint64_t guid)
1375 struct stat64 statbuf;
1376 nvlist_t *raw, *src, *dst;
1383 verify(poolname == NULL || guid == 0);
1385 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1386 zfs_error_aux(hdl, "%s", strerror(errno));
1387 (void) zfs_error(hdl, EZFS_BADCACHE,
1388 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1392 if (fstat64(fd, &statbuf) != 0) {
1393 zfs_error_aux(hdl, "%s", strerror(errno));
1395 (void) zfs_error(hdl, EZFS_BADCACHE,
1396 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1400 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1405 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1408 (void) zfs_error(hdl, EZFS_BADCACHE,
1409 dgettext(TEXT_DOMAIN,
1410 "failed to read cache file contents"));
1416 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1418 (void) zfs_error(hdl, EZFS_BADCACHE,
1419 dgettext(TEXT_DOMAIN,
1420 "invalid or corrupt cache file contents"));
1427 * Go through and get the current state of the pools and refresh their
1430 if (nvlist_alloc(&pools, 0, 0) != 0) {
1431 (void) no_memory(hdl);
1437 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1438 src = fnvpair_value_nvlist(elem);
1440 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1441 if (poolname != NULL && strcmp(poolname, name) != 0)
1444 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1445 if (guid != 0 && guid != this_guid)
1448 if (pool_active(hdl, name, this_guid, &active) != 0) {
1457 if ((dst = refresh_config(hdl, src)) == NULL) {
1463 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1464 (void) no_memory(hdl);
1478 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1480 importargs_t *import = data;
1483 if (import->poolname != NULL) {
1486 verify(nvlist_lookup_string(zhp->zpool_config,
1487 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1488 if (strcmp(pool_name, import->poolname) == 0)
1493 verify(nvlist_lookup_uint64(zhp->zpool_config,
1494 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1495 if (pool_guid == import->guid)
1504 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1506 verify(import->poolname == NULL || import->guid == 0);
1509 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1511 if (import->cachefile != NULL)
1512 return (zpool_find_import_cached(hdl, import->cachefile,
1513 import->poolname, import->guid));
1515 return (zpool_find_import_impl(hdl, import));
1519 find_guid(nvlist_t *nv, uint64_t guid)
1525 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1529 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1530 &child, &children) == 0) {
1531 for (c = 0; c < children; c++)
1532 if (find_guid(child[c], guid))
1539 typedef struct aux_cbdata {
1540 const char *cb_type;
1542 zpool_handle_t *cb_zhp;
1546 find_aux(zpool_handle_t *zhp, void *data)
1548 aux_cbdata_t *cbp = data;
1554 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1557 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1558 &list, &count) == 0) {
1559 for (i = 0; i < count; i++) {
1560 verify(nvlist_lookup_uint64(list[i],
1561 ZPOOL_CONFIG_GUID, &guid) == 0);
1562 if (guid == cbp->cb_guid) {
1574 * Determines if the pool is in use. If so, it returns true and the state of
1575 * the pool as well as the name of the pool. Both strings are allocated and
1576 * must be freed by the caller.
1579 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1585 uint64_t guid, vdev_guid;
1586 zpool_handle_t *zhp;
1587 nvlist_t *pool_config;
1588 uint64_t stateval, isspare;
1589 aux_cbdata_t cb = { 0 };
1594 if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1595 (void) no_memory(hdl);
1602 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1604 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1607 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1608 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1610 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1615 case POOL_STATE_EXPORTED:
1617 * A pool with an exported state may in fact be imported
1618 * read-only, so check the in-core state to see if it's
1619 * active and imported read-only. If it is, set
1620 * its state to active.
1622 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1623 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1624 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1625 stateval = POOL_STATE_ACTIVE;
1628 * All we needed the zpool handle for is the
1629 * readonly prop check.
1637 case POOL_STATE_ACTIVE:
1639 * For an active pool, we have to determine if it's really part
1640 * of a currently active pool (in which case the pool will exist
1641 * and the guid will be the same), or whether it's part of an
1642 * active pool that was disconnected without being explicitly
1645 if (pool_active(hdl, name, guid, &isactive) != 0) {
1646 nvlist_free(config);
1652 * Because the device may have been removed while
1653 * offlined, we only report it as active if the vdev is
1654 * still present in the config. Otherwise, pretend like
1657 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1658 (pool_config = zpool_get_config(zhp, NULL))
1662 verify(nvlist_lookup_nvlist(pool_config,
1663 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1664 ret = find_guid(nvroot, vdev_guid);
1670 * If this is an active spare within another pool, we
1671 * treat it like an unused hot spare. This allows the
1672 * user to create a pool with a hot spare that currently
1673 * in use within another pool. Since we return B_TRUE,
1674 * libdiskmgt will continue to prevent generic consumers
1675 * from using the device.
1677 if (ret && nvlist_lookup_uint64(config,
1678 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1679 stateval = POOL_STATE_SPARE;
1684 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1689 case POOL_STATE_SPARE:
1691 * For a hot spare, it can be either definitively in use, or
1692 * potentially active. To determine if it's in use, we iterate
1693 * over all pools in the system and search for one with a spare
1694 * with a matching guid.
1696 * Due to the shared nature of spares, we don't actually report
1697 * the potentially active case as in use. This means the user
1698 * can freely create pools on the hot spares of exported pools,
1699 * but to do otherwise makes the resulting code complicated, and
1700 * we end up having to deal with this case anyway.
1703 cb.cb_guid = vdev_guid;
1704 cb.cb_type = ZPOOL_CONFIG_SPARES;
1705 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1706 name = (char *)zpool_get_name(cb.cb_zhp);
1713 case POOL_STATE_L2CACHE:
1716 * Check if any pool is currently using this l2cache device.
1719 cb.cb_guid = vdev_guid;
1720 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1721 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1722 name = (char *)zpool_get_name(cb.cb_zhp);
1735 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1737 zpool_close(cb.cb_zhp);
1738 nvlist_free(config);
1741 *state = (pool_state_t)stateval;
1745 zpool_close(cb.cb_zhp);
1747 nvlist_free(config);