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, 2017 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.
57 #include <thread_pool.h>
60 #include <sys/vdev_impl.h>
63 #include "libzfs_impl.h"
66 * Intermediate structures used to gather configuration information.
68 typedef struct config_entry {
71 struct config_entry *ce_next;
74 typedef struct vdev_entry {
76 config_entry_t *ve_configs;
77 struct vdev_entry *ve_next;
80 typedef struct pool_entry {
82 vdev_entry_t *pe_vdevs;
83 struct pool_entry *pe_next;
86 typedef struct name_entry {
89 struct name_entry *ne_next;
92 typedef struct pool_list {
98 get_devid(const char *path)
105 if ((fd = open(path, O_RDONLY)) < 0)
110 if (devid_get(fd, &devid) == 0) {
111 if (devid_get_minor_name(fd, &minor) == 0)
112 ret = devid_str_encode(devid, minor);
114 devid_str_free(minor);
127 * Go through and fix up any path and/or devid information for the given vdev
131 fix_paths(nvlist_t *nv, name_entry_t *names)
136 name_entry_t *ne, *best;
140 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
141 &child, &children) == 0) {
142 for (c = 0; c < children; c++)
143 if (fix_paths(child[c], names) != 0)
149 * This is a leaf (file or disk) vdev. In either case, go through
150 * the name list and see if we find a matching guid. If so, replace
151 * the path and see if we can calculate a new devid.
153 * There may be multiple names associated with a particular guid, in
154 * which case we have overlapping slices or multiple paths to the same
155 * disk. If this is the case, then we want to pick the path that is
156 * the most similar to the original, where "most similar" is the number
157 * of matching characters starting from the end of the path. This will
158 * preserve slice numbers even if the disks have been reorganized, and
159 * will also catch preferred disk names if multiple paths exist.
161 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
162 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
167 for (ne = names; ne != NULL; ne = ne->ne_next) {
168 if (ne->ne_guid == guid) {
169 const char *src, *dst;
177 src = ne->ne_name + strlen(ne->ne_name) - 1;
178 dst = path + strlen(path) - 1;
179 for (count = 0; src >= ne->ne_name && dst >= path;
180 src--, dst--, count++)
185 * At this point, 'count' is the number of characters
186 * matched from the end.
188 if (count > matched || best == NULL) {
198 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
201 if ((devid = get_devid(best->ne_name)) == NULL) {
202 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
204 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
205 devid_str_free(devid);
208 devid_str_free(devid);
215 * Add the given configuration to the list of known devices.
218 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
221 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
228 * If this is a hot spare not currently in use or level 2 cache
229 * device, add it to the list of names to translate, but don't do
232 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
234 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
235 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
236 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
239 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
244 ne->ne_guid = vdev_guid;
245 ne->ne_next = pl->names;
253 * If we have a valid config but cannot read any of these fields, then
254 * it means we have a half-initialized label. In vdev_label_init()
255 * we write a label with txg == 0 so that we can identify the device
256 * in case the user refers to the same disk later on. If we fail to
257 * create the pool, we'll be left with a label in this state
258 * which should not be considered part of a valid pool.
260 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
262 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
264 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
266 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
267 &txg) != 0 || txg == 0) {
273 * First, see if we know about this pool. If not, then add it to the
274 * list of known pools.
276 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
277 if (pe->pe_guid == pool_guid)
282 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
286 pe->pe_guid = pool_guid;
287 pe->pe_next = pl->pools;
292 * Second, see if we know about this toplevel vdev. Add it if its
295 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
296 if (ve->ve_guid == top_guid)
301 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
305 ve->ve_guid = top_guid;
306 ve->ve_next = pe->pe_vdevs;
311 * Third, see if we have a config with a matching transaction group. If
312 * so, then we do nothing. Otherwise, add it to the list of known
315 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
316 if (ce->ce_txg == txg)
321 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
326 ce->ce_config = config;
327 ce->ce_next = ve->ve_configs;
334 * At this point we've successfully added our config to the list of
335 * known configs. The last thing to do is add the vdev guid -> path
336 * mappings so that we can fix up the configuration as necessary before
339 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
342 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
347 ne->ne_guid = vdev_guid;
348 ne->ne_next = pl->names;
355 * Returns true if the named pool matches the given GUID.
358 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
364 if (zpool_open_silent(hdl, name, &zhp) != 0)
372 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
377 *isactive = (theguid == guid);
382 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
385 zfs_cmd_t zc = { 0 };
386 int err, dstbuf_size;
388 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
391 dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
393 if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
394 zcmd_free_nvlists(&zc);
398 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
399 &zc)) != 0 && errno == ENOMEM) {
400 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
401 zcmd_free_nvlists(&zc);
407 zcmd_free_nvlists(&zc);
411 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
412 zcmd_free_nvlists(&zc);
416 zcmd_free_nvlists(&zc);
421 * Determine if the vdev id is a hole in the namespace.
424 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
426 for (int c = 0; c < holes; c++) {
428 /* Top-level is a hole */
429 if (hole_array[c] == id)
436 * Convert our list of pools into the definitive set of configurations. We
437 * start by picking the best config for each toplevel vdev. Once that's done,
438 * we assemble the toplevel vdevs into a full config for the pool. We make a
439 * pass to fix up any incorrect paths, and then add it to the main list to
440 * return to the user.
443 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
449 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
450 nvlist_t **spares, **l2cache;
451 uint_t i, nspares, nl2cache;
452 boolean_t config_seen;
454 char *name, *hostname = NULL;
457 nvlist_t **child = NULL;
459 uint64_t *hole_array, max_id;
464 boolean_t found_one = B_FALSE;
465 boolean_t valid_top_config = B_FALSE;
467 if (nvlist_alloc(&ret, 0, 0) != 0)
470 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
471 uint64_t id, max_txg = 0;
473 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
475 config_seen = B_FALSE;
478 * Iterate over all toplevel vdevs. Grab the pool configuration
479 * from the first one we find, and then go through the rest and
480 * add them as necessary to the 'vdevs' member of the config.
482 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
485 * Determine the best configuration for this vdev by
486 * selecting the config with the latest transaction
490 for (ce = ve->ve_configs; ce != NULL;
493 if (ce->ce_txg > best_txg) {
495 best_txg = ce->ce_txg;
500 * We rely on the fact that the max txg for the
501 * pool will contain the most up-to-date information
502 * about the valid top-levels in the vdev namespace.
504 if (best_txg > max_txg) {
505 (void) nvlist_remove(config,
506 ZPOOL_CONFIG_VDEV_CHILDREN,
508 (void) nvlist_remove(config,
509 ZPOOL_CONFIG_HOLE_ARRAY,
510 DATA_TYPE_UINT64_ARRAY);
516 valid_top_config = B_FALSE;
518 if (nvlist_lookup_uint64(tmp,
519 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
520 verify(nvlist_add_uint64(config,
521 ZPOOL_CONFIG_VDEV_CHILDREN,
523 valid_top_config = B_TRUE;
526 if (nvlist_lookup_uint64_array(tmp,
527 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
529 verify(nvlist_add_uint64_array(config,
530 ZPOOL_CONFIG_HOLE_ARRAY,
531 hole_array, holes) == 0);
537 * Copy the relevant pieces of data to the pool
543 * comment (if available)
545 * hostid (if available)
546 * hostname (if available)
548 uint64_t state, version;
549 char *comment = NULL;
551 version = fnvlist_lookup_uint64(tmp,
552 ZPOOL_CONFIG_VERSION);
553 fnvlist_add_uint64(config,
554 ZPOOL_CONFIG_VERSION, version);
555 guid = fnvlist_lookup_uint64(tmp,
556 ZPOOL_CONFIG_POOL_GUID);
557 fnvlist_add_uint64(config,
558 ZPOOL_CONFIG_POOL_GUID, guid);
559 name = fnvlist_lookup_string(tmp,
560 ZPOOL_CONFIG_POOL_NAME);
561 fnvlist_add_string(config,
562 ZPOOL_CONFIG_POOL_NAME, name);
564 if (nvlist_lookup_string(tmp,
565 ZPOOL_CONFIG_COMMENT, &comment) == 0)
566 fnvlist_add_string(config,
567 ZPOOL_CONFIG_COMMENT, comment);
569 state = fnvlist_lookup_uint64(tmp,
570 ZPOOL_CONFIG_POOL_STATE);
571 fnvlist_add_uint64(config,
572 ZPOOL_CONFIG_POOL_STATE, state);
575 if (nvlist_lookup_uint64(tmp,
576 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
577 fnvlist_add_uint64(config,
578 ZPOOL_CONFIG_HOSTID, hostid);
579 hostname = fnvlist_lookup_string(tmp,
580 ZPOOL_CONFIG_HOSTNAME);
581 fnvlist_add_string(config,
582 ZPOOL_CONFIG_HOSTNAME, hostname);
585 config_seen = B_TRUE;
589 * Add this top-level vdev to the child array.
591 verify(nvlist_lookup_nvlist(tmp,
592 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
593 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
596 if (id >= children) {
599 newchild = zfs_alloc(hdl, (id + 1) *
600 sizeof (nvlist_t *));
601 if (newchild == NULL)
604 for (c = 0; c < children; c++)
605 newchild[c] = child[c];
611 if (nvlist_dup(nvtop, &child[id], 0) != 0)
617 * If we have information about all the top-levels then
618 * clean up the nvlist which we've constructed. This
619 * means removing any extraneous devices that are
620 * beyond the valid range or adding devices to the end
621 * of our array which appear to be missing.
623 if (valid_top_config) {
624 if (max_id < children) {
625 for (c = max_id; c < children; c++)
626 nvlist_free(child[c]);
628 } else if (max_id > children) {
631 newchild = zfs_alloc(hdl, (max_id) *
632 sizeof (nvlist_t *));
633 if (newchild == NULL)
636 for (c = 0; c < children; c++)
637 newchild[c] = child[c];
645 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
649 * The vdev namespace may contain holes as a result of
650 * device removal. We must add them back into the vdev
651 * tree before we process any missing devices.
654 ASSERT(valid_top_config);
656 for (c = 0; c < children; c++) {
659 if (child[c] != NULL ||
660 !vdev_is_hole(hole_array, holes, c))
663 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
668 * Holes in the namespace are treated as
669 * "hole" top-level vdevs and have a
670 * special flag set on them.
672 if (nvlist_add_string(holey,
674 VDEV_TYPE_HOLE) != 0 ||
675 nvlist_add_uint64(holey,
676 ZPOOL_CONFIG_ID, c) != 0 ||
677 nvlist_add_uint64(holey,
678 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
687 * Look for any missing top-level vdevs. If this is the case,
688 * create a faked up 'missing' vdev as a placeholder. We cannot
689 * simply compress the child array, because the kernel performs
690 * certain checks to make sure the vdev IDs match their location
691 * in the configuration.
693 for (c = 0; c < children; c++) {
694 if (child[c] == NULL) {
696 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
699 if (nvlist_add_string(missing,
701 VDEV_TYPE_MISSING) != 0 ||
702 nvlist_add_uint64(missing,
703 ZPOOL_CONFIG_ID, c) != 0 ||
704 nvlist_add_uint64(missing,
705 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
706 nvlist_free(missing);
714 * Put all of this pool's top-level vdevs into a root vdev.
716 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
718 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
719 VDEV_TYPE_ROOT) != 0 ||
720 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
721 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
722 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
723 child, children) != 0) {
728 for (c = 0; c < children; c++)
729 nvlist_free(child[c]);
735 * Go through and fix up any paths and/or devids based on our
736 * known list of vdev GUID -> path mappings.
738 if (fix_paths(nvroot, pl->names) != 0) {
744 * Add the root vdev to this pool's configuration.
746 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
754 * zdb uses this path to report on active pools that were
755 * imported or created using -R.
761 * Determine if this pool is currently active, in which case we
762 * can't actually import it.
764 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
766 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
769 if (pool_active(hdl, name, guid, &isactive) != 0)
778 if (policy != NULL) {
779 if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
784 if ((nvl = refresh_config(hdl, config)) == NULL) {
794 * Go through and update the paths for spares, now that we have
797 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
799 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
800 &spares, &nspares) == 0) {
801 for (i = 0; i < nspares; i++) {
802 if (fix_paths(spares[i], pl->names) != 0)
808 * Update the paths for l2cache devices.
810 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
811 &l2cache, &nl2cache) == 0) {
812 for (i = 0; i < nl2cache; i++) {
813 if (fix_paths(l2cache[i], pl->names) != 0)
819 * Restore the original information read from the actual label.
821 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
823 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
826 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
828 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
834 * Add this pool to the list of configs.
836 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
838 if (nvlist_add_nvlist(ret, name, config) != 0)
854 (void) no_memory(hdl);
858 for (c = 0; c < children; c++)
859 nvlist_free(child[c]);
866 * Return the offset of the given label.
869 label_offset(uint64_t size, int l)
871 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
872 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
873 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
877 * Given a file descriptor, read the label information and return an nvlist
878 * describing the configuration, if there is one.
879 * Return 0 on success, or -1 on failure
882 zpool_read_label(int fd, nvlist_t **config)
884 struct stat64 statbuf;
887 uint64_t state, txg, size;
891 if (fstat64(fd, &statbuf) == -1)
893 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
895 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
898 for (l = 0; l < VDEV_LABELS; l++) {
899 if (pread64(fd, label, sizeof (vdev_label_t),
900 label_offset(size, l)) != sizeof (vdev_label_t))
903 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
904 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
907 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
908 &state) != 0 || state > POOL_STATE_L2CACHE) {
909 nvlist_free(*config);
913 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
914 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
915 &txg) != 0 || txg == 0)) {
916 nvlist_free(*config);
931 * Given a file descriptor, read the label information and return an nvlist
932 * describing the configuration, if there is one.
933 * returns the number of valid labels found
934 * If a label is found, returns it via config. The caller is responsible for
938 zpool_read_all_labels(int fd, nvlist_t **config)
940 struct stat64 statbuf;
941 struct aiocb aiocbs[VDEV_LABELS];
942 struct aiocb *aiocbps[VDEV_LABELS];
945 uint64_t state, txg, size;
950 if (fstat64(fd, &statbuf) == -1)
952 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
954 if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
957 memset(aiocbs, 0, sizeof(aiocbs));
958 for (l = 0; l < VDEV_LABELS; l++) {
959 aiocbs[l].aio_fildes = fd;
960 aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
961 aiocbs[l].aio_buf = &labels[l];
962 aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
963 aiocbs[l].aio_lio_opcode = LIO_READ;
964 aiocbps[l] = &aiocbs[l];
967 if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
968 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
969 for (l = 0; l < VDEV_LABELS; l++) {
971 int r = aio_error(&aiocbs[l]);
973 (void)aio_return(&aiocbs[l]);
980 for (l = 0; l < VDEV_LABELS; l++) {
981 nvlist_t *temp = NULL;
983 if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
986 if (nvlist_unpack(labels[l].vp_nvlist,
987 sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
990 if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
991 &state) != 0 || state > POOL_STATE_L2CACHE) {
997 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
998 (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
999 &txg) != 0 || txg == 0)) {
1014 typedef struct rdsk_node {
1017 libzfs_handle_t *rn_hdl;
1018 nvlist_t *rn_config;
1021 boolean_t rn_nozpool;
1025 slice_cache_compare(const void *arg1, const void *arg2)
1027 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1028 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1029 char *nm1slice, *nm2slice;
1033 * slices zero and two are the most likely to provide results,
1034 * so put those first
1036 nm1slice = strstr(nm1, "s0");
1037 nm2slice = strstr(nm2, "s0");
1038 if (nm1slice && !nm2slice) {
1041 if (!nm1slice && nm2slice) {
1044 nm1slice = strstr(nm1, "s2");
1045 nm2slice = strstr(nm2, "s2");
1046 if (nm1slice && !nm2slice) {
1049 if (!nm1slice && nm2slice) {
1053 rv = strcmp(nm1, nm2);
1056 return (rv > 0 ? 1 : -1);
1061 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1062 diskaddr_t size, uint_t blksz)
1064 rdsk_node_t tmpnode;
1066 char sname[MAXNAMELEN];
1068 tmpnode.rn_name = &sname[0];
1069 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1072 * protect against division by zero for disk labels that
1073 * contain a bogus sector size
1077 /* too small to contain a zpool? */
1078 if ((size < (SPA_MINDEVSIZE / blksz)) &&
1079 (node = avl_find(r, &tmpnode, NULL)))
1080 node->rn_nozpool = B_TRUE;
1082 #endif /* illumos */
1085 nozpool_all_slices(avl_tree_t *r, const char *sname)
1088 char diskname[MAXNAMELEN];
1092 (void) strncpy(diskname, sname, MAXNAMELEN);
1093 if (((ptr = strrchr(diskname, 's')) == NULL) &&
1094 ((ptr = strrchr(diskname, 'p')) == NULL))
1098 for (i = 0; i < NDKMAP; i++)
1099 check_one_slice(r, diskname, i, 0, 1);
1101 for (i = 0; i <= FD_NUMPART; i++)
1102 check_one_slice(r, diskname, i, 0, 1);
1103 #endif /* illumos */
1108 check_slices(avl_tree_t *r, int fd, const char *sname)
1110 struct extvtoc vtoc;
1112 char diskname[MAXNAMELEN];
1116 (void) strncpy(diskname, sname, MAXNAMELEN);
1117 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1121 if (read_extvtoc(fd, &vtoc) >= 0) {
1122 for (i = 0; i < NDKMAP; i++)
1123 check_one_slice(r, diskname, i,
1124 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1125 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1127 * on x86 we'll still have leftover links that point
1128 * to slices s[9-15], so use NDKMAP instead
1130 for (i = 0; i < NDKMAP; i++)
1131 check_one_slice(r, diskname, i,
1132 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1133 /* nodes p[1-4] are never used with EFI labels */
1135 for (i = 1; i <= FD_NUMPART; i++)
1136 check_one_slice(r, diskname, i, 0, 1);
1140 #endif /* illumos */
1143 zpool_open_func(void *arg)
1145 rdsk_node_t *rn = arg;
1146 struct stat64 statbuf;
1152 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1153 /* symlink to a device that's no longer there */
1154 if (errno == ENOENT)
1155 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1159 * Ignore failed stats. We only want regular
1160 * files, character devs and block devs.
1162 if (fstat64(fd, &statbuf) != 0 ||
1163 (!S_ISREG(statbuf.st_mode) &&
1164 !S_ISCHR(statbuf.st_mode) &&
1165 !S_ISBLK(statbuf.st_mode))) {
1169 /* this file is too small to hold a zpool */
1171 if (S_ISREG(statbuf.st_mode) &&
1172 statbuf.st_size < SPA_MINDEVSIZE) {
1175 } else if (!S_ISREG(statbuf.st_mode)) {
1177 * Try to read the disk label first so we don't have to
1178 * open a bunch of minor nodes that can't have a zpool.
1180 check_slices(rn->rn_avl, fd, rn->rn_name);
1182 #else /* !illumos */
1183 if (statbuf.st_size < SPA_MINDEVSIZE) {
1187 #endif /* illumos */
1189 if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1191 (void) no_memory(rn->rn_hdl);
1196 rn->rn_config = config;
1200 * Given a file descriptor, clear (zero) the label information.
1203 zpool_clear_label(int fd)
1205 struct stat64 statbuf;
1207 vdev_label_t *label;
1210 if (fstat64(fd, &statbuf) == -1)
1212 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1214 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1217 for (l = 0; l < VDEV_LABELS; l++) {
1218 if (pwrite64(fd, label, sizeof (vdev_label_t),
1219 label_offset(size, l)) != sizeof (vdev_label_t)) {
1230 * Given a list of directories to search, find all pools stored on disk. This
1231 * includes partial pools which are not available to import. If no args are
1232 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1233 * poolname or guid (but not both) are provided by the caller when trying
1234 * to import a specific pool.
1237 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1239 int i, dirs = iarg->paths;
1240 struct dirent64 *dp;
1241 char path[MAXPATHLEN];
1242 char *end, **dir = iarg->path;
1244 nvlist_t *ret = NULL;
1245 static char *default_dir = "/dev";
1246 pool_list_t pools = { 0 };
1247 pool_entry_t *pe, *penext;
1248 vdev_entry_t *ve, *venext;
1249 config_entry_t *ce, *cenext;
1250 name_entry_t *ne, *nenext;
1251 avl_tree_t slice_cache;
1261 * Go through and read the label configuration information from every
1262 * possible device, organizing the information according to pool GUID
1263 * and toplevel GUID.
1265 for (i = 0; i < dirs; i++) {
1267 char rdsk[MAXPATHLEN];
1269 boolean_t config_failed = B_FALSE;
1272 /* use realpath to normalize the path */
1273 if (realpath(dir[i], path) == 0) {
1274 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1275 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1278 end = &path[strlen(path)];
1281 pathleft = &path[sizeof (path)] - end;
1285 * Using raw devices instead of block devices when we're
1286 * reading the labels skips a bunch of slow operations during
1287 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1289 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1290 (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1293 (void) strlcpy(rdsk, path, sizeof (rdsk));
1295 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1296 (dirp = fdopendir(dfd)) == NULL) {
1299 zfs_error_aux(hdl, strerror(errno));
1300 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1301 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1306 avl_create(&slice_cache, slice_cache_compare,
1307 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1309 if (strcmp(rdsk, "/dev/") == 0) {
1313 struct gprovider *pp;
1315 errno = geom_gettree(&mesh);
1317 zfs_error_aux(hdl, strerror(errno));
1318 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1319 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1323 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1324 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1325 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1326 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1327 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1328 slice->rn_avl = &slice_cache;
1329 slice->rn_dfd = dfd;
1330 slice->rn_hdl = hdl;
1331 slice->rn_nozpool = B_FALSE;
1332 avl_add(&slice_cache, slice);
1337 geom_deletetree(&mesh);
1342 * This is not MT-safe, but we have no MT consumers of libzfs
1344 while ((dp = readdir64(dirp)) != NULL) {
1345 const char *name = dp->d_name;
1346 if (name[0] == '.' &&
1347 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1350 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1351 slice->rn_name = zfs_strdup(hdl, name);
1352 slice->rn_avl = &slice_cache;
1353 slice->rn_dfd = dfd;
1354 slice->rn_hdl = hdl;
1355 slice->rn_nozpool = B_FALSE;
1356 avl_add(&slice_cache, slice);
1360 * create a thread pool to do all of this in parallel;
1361 * rn_nozpool is not protected, so this is racy in that
1362 * multiple tasks could decide that the same slice can
1363 * not hold a zpool, which is benign. Also choose
1364 * double the number of processors; we hold a lot of
1365 * locks in the kernel, so going beyond this doesn't
1368 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1370 for (slice = avl_first(&slice_cache); slice;
1371 (slice = avl_walk(&slice_cache, slice,
1373 (void) tpool_dispatch(t, zpool_open_func, slice);
1378 while ((slice = avl_destroy_nodes(&slice_cache,
1379 &cookie)) != NULL) {
1380 if (slice->rn_config != NULL && !config_failed) {
1381 nvlist_t *config = slice->rn_config;
1382 boolean_t matched = B_TRUE;
1384 if (iarg->poolname != NULL) {
1387 matched = nvlist_lookup_string(config,
1388 ZPOOL_CONFIG_POOL_NAME,
1390 strcmp(iarg->poolname, pname) == 0;
1391 } else if (iarg->guid != 0) {
1394 matched = nvlist_lookup_uint64(config,
1395 ZPOOL_CONFIG_POOL_GUID,
1397 iarg->guid == this_guid;
1400 nvlist_free(config);
1403 * use the non-raw path for the config
1405 (void) strlcpy(end, slice->rn_name,
1407 if (add_config(hdl, &pools, path,
1409 config_failed = B_TRUE;
1412 free(slice->rn_name);
1415 avl_destroy(&slice_cache);
1417 (void) closedir(dirp);
1423 ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1426 for (pe = pools.pools; pe != NULL; pe = penext) {
1427 penext = pe->pe_next;
1428 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1429 venext = ve->ve_next;
1430 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1431 cenext = ce->ce_next;
1432 nvlist_free(ce->ce_config);
1440 for (ne = pools.names; ne != NULL; ne = nenext) {
1441 nenext = ne->ne_next;
1450 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1452 importargs_t iarg = { 0 };
1457 return (zpool_find_import_impl(hdl, &iarg));
1461 * Given a cache file, return the contents as a list of importable pools.
1462 * poolname or guid (but not both) are provided by the caller when trying
1463 * to import a specific pool.
1466 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1467 char *poolname, uint64_t guid)
1471 struct stat64 statbuf;
1472 nvlist_t *raw, *src, *dst;
1479 verify(poolname == NULL || guid == 0);
1481 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1482 zfs_error_aux(hdl, "%s", strerror(errno));
1483 (void) zfs_error(hdl, EZFS_BADCACHE,
1484 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1488 if (fstat64(fd, &statbuf) != 0) {
1489 zfs_error_aux(hdl, "%s", strerror(errno));
1491 (void) zfs_error(hdl, EZFS_BADCACHE,
1492 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1496 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1501 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1504 (void) zfs_error(hdl, EZFS_BADCACHE,
1505 dgettext(TEXT_DOMAIN,
1506 "failed to read cache file contents"));
1512 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1514 (void) zfs_error(hdl, EZFS_BADCACHE,
1515 dgettext(TEXT_DOMAIN,
1516 "invalid or corrupt cache file contents"));
1523 * Go through and get the current state of the pools and refresh their
1526 if (nvlist_alloc(&pools, 0, 0) != 0) {
1527 (void) no_memory(hdl);
1533 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1534 src = fnvpair_value_nvlist(elem);
1536 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1537 if (poolname != NULL && strcmp(poolname, name) != 0)
1540 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1541 if (guid != 0 && guid != this_guid)
1544 if (pool_active(hdl, name, this_guid, &active) != 0) {
1553 if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1555 (void) no_memory(hdl);
1561 if ((dst = refresh_config(hdl, src)) == NULL) {
1567 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1568 (void) no_memory(hdl);
1582 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1584 importargs_t *import = data;
1587 if (import->poolname != NULL) {
1590 verify(nvlist_lookup_string(zhp->zpool_config,
1591 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1592 if (strcmp(pool_name, import->poolname) == 0)
1597 verify(nvlist_lookup_uint64(zhp->zpool_config,
1598 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1599 if (pool_guid == import->guid)
1608 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1610 verify(import->poolname == NULL || import->guid == 0);
1613 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1615 if (import->cachefile != NULL)
1616 return (zpool_find_import_cached(hdl, import->cachefile,
1617 import->poolname, import->guid));
1619 return (zpool_find_import_impl(hdl, import));
1623 find_guid(nvlist_t *nv, uint64_t guid)
1629 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1633 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1634 &child, &children) == 0) {
1635 for (c = 0; c < children; c++)
1636 if (find_guid(child[c], guid))
1643 typedef struct aux_cbdata {
1644 const char *cb_type;
1646 zpool_handle_t *cb_zhp;
1650 find_aux(zpool_handle_t *zhp, void *data)
1652 aux_cbdata_t *cbp = data;
1658 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1661 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1662 &list, &count) == 0) {
1663 for (i = 0; i < count; i++) {
1664 verify(nvlist_lookup_uint64(list[i],
1665 ZPOOL_CONFIG_GUID, &guid) == 0);
1666 if (guid == cbp->cb_guid) {
1678 * Determines if the pool is in use. If so, it returns true and the state of
1679 * the pool as well as the name of the pool. Both strings are allocated and
1680 * must be freed by the caller.
1683 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1689 uint64_t guid, vdev_guid;
1690 zpool_handle_t *zhp;
1691 nvlist_t *pool_config;
1692 uint64_t stateval, isspare;
1693 aux_cbdata_t cb = { 0 };
1698 if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1699 (void) no_memory(hdl);
1706 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1708 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1711 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1712 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1714 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1719 case POOL_STATE_EXPORTED:
1721 * A pool with an exported state may in fact be imported
1722 * read-only, so check the in-core state to see if it's
1723 * active and imported read-only. If it is, set
1724 * its state to active.
1726 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1727 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1728 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1729 stateval = POOL_STATE_ACTIVE;
1732 * All we needed the zpool handle for is the
1733 * readonly prop check.
1741 case POOL_STATE_ACTIVE:
1743 * For an active pool, we have to determine if it's really part
1744 * of a currently active pool (in which case the pool will exist
1745 * and the guid will be the same), or whether it's part of an
1746 * active pool that was disconnected without being explicitly
1749 if (pool_active(hdl, name, guid, &isactive) != 0) {
1750 nvlist_free(config);
1756 * Because the device may have been removed while
1757 * offlined, we only report it as active if the vdev is
1758 * still present in the config. Otherwise, pretend like
1761 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1762 (pool_config = zpool_get_config(zhp, NULL))
1766 verify(nvlist_lookup_nvlist(pool_config,
1767 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1768 ret = find_guid(nvroot, vdev_guid);
1774 * If this is an active spare within another pool, we
1775 * treat it like an unused hot spare. This allows the
1776 * user to create a pool with a hot spare that currently
1777 * in use within another pool. Since we return B_TRUE,
1778 * libdiskmgt will continue to prevent generic consumers
1779 * from using the device.
1781 if (ret && nvlist_lookup_uint64(config,
1782 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1783 stateval = POOL_STATE_SPARE;
1788 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1793 case POOL_STATE_SPARE:
1795 * For a hot spare, it can be either definitively in use, or
1796 * potentially active. To determine if it's in use, we iterate
1797 * over all pools in the system and search for one with a spare
1798 * with a matching guid.
1800 * Due to the shared nature of spares, we don't actually report
1801 * the potentially active case as in use. This means the user
1802 * can freely create pools on the hot spares of exported pools,
1803 * but to do otherwise makes the resulting code complicated, and
1804 * we end up having to deal with this case anyway.
1807 cb.cb_guid = vdev_guid;
1808 cb.cb_type = ZPOOL_CONFIG_SPARES;
1809 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1810 name = (char *)zpool_get_name(cb.cb_zhp);
1817 case POOL_STATE_L2CACHE:
1820 * Check if any pool is currently using this l2cache device.
1823 cb.cb_guid = vdev_guid;
1824 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1825 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1826 name = (char *)zpool_get_name(cb.cb_zhp);
1839 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1841 zpool_close(cb.cb_zhp);
1842 nvlist_free(config);
1845 *state = (pool_state_t)stateval;
1849 zpool_close(cb.cb_zhp);
1851 nvlist_free(config);