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;
252 * If we have a valid config but cannot read any of these fields, then
253 * it means we have a half-initialized label. In vdev_label_init()
254 * we write a label with txg == 0 so that we can identify the device
255 * in case the user refers to the same disk later on. If we fail to
256 * create the pool, we'll be left with a label in this state
257 * which should not be considered part of a valid pool.
259 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
261 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
263 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
265 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
266 &txg) != 0 || txg == 0) {
271 * First, see if we know about this pool. If not, then add it to the
272 * list of known pools.
274 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
275 if (pe->pe_guid == pool_guid)
280 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
283 pe->pe_guid = pool_guid;
284 pe->pe_next = pl->pools;
289 * Second, see if we know about this toplevel vdev. Add it if its
292 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
293 if (ve->ve_guid == top_guid)
298 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) {
321 ce->ce_config = fnvlist_dup(config);
322 ce->ce_next = ve->ve_configs;
327 * At this point we've successfully added our config to the list of
328 * known configs. The last thing to do is add the vdev guid -> path
329 * mappings so that we can fix up the configuration as necessary before
332 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
335 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
340 ne->ne_guid = vdev_guid;
341 ne->ne_next = pl->names;
348 * Returns true if the named pool matches the given GUID.
351 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
357 if (zpool_open_silent(hdl, name, &zhp) != 0)
365 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
370 *isactive = (theguid == guid);
375 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
378 zfs_cmd_t zc = { 0 };
379 int err, dstbuf_size;
381 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
384 dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
386 if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
387 zcmd_free_nvlists(&zc);
391 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
392 &zc)) != 0 && errno == ENOMEM) {
393 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
394 zcmd_free_nvlists(&zc);
400 zcmd_free_nvlists(&zc);
404 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
405 zcmd_free_nvlists(&zc);
409 zcmd_free_nvlists(&zc);
414 * Determine if the vdev id is a hole in the namespace.
417 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
419 for (int c = 0; c < holes; c++) {
421 /* Top-level is a hole */
422 if (hole_array[c] == id)
429 * Convert our list of pools into the definitive set of configurations. We
430 * start by picking the best config for each toplevel vdev. Once that's done,
431 * we assemble the toplevel vdevs into a full config for the pool. We make a
432 * pass to fix up any incorrect paths, and then add it to the main list to
433 * return to the user.
436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
442 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
443 nvlist_t **spares, **l2cache;
444 uint_t i, nspares, nl2cache;
445 boolean_t config_seen;
447 char *name, *hostname = NULL;
450 nvlist_t **child = NULL;
452 uint64_t *hole_array, max_id;
457 boolean_t found_one = B_FALSE;
458 boolean_t valid_top_config = B_FALSE;
460 if (nvlist_alloc(&ret, 0, 0) != 0)
463 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
464 uint64_t id, max_txg = 0;
466 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
468 config_seen = B_FALSE;
471 * Iterate over all toplevel vdevs. Grab the pool configuration
472 * from the first one we find, and then go through the rest and
473 * add them as necessary to the 'vdevs' member of the config.
475 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
478 * Determine the best configuration for this vdev by
479 * selecting the config with the latest transaction
483 for (ce = ve->ve_configs; ce != NULL;
486 if (ce->ce_txg > best_txg) {
488 best_txg = ce->ce_txg;
493 * We rely on the fact that the max txg for the
494 * pool will contain the most up-to-date information
495 * about the valid top-levels in the vdev namespace.
497 if (best_txg > max_txg) {
498 (void) nvlist_remove(config,
499 ZPOOL_CONFIG_VDEV_CHILDREN,
501 (void) nvlist_remove(config,
502 ZPOOL_CONFIG_HOLE_ARRAY,
503 DATA_TYPE_UINT64_ARRAY);
509 valid_top_config = B_FALSE;
511 if (nvlist_lookup_uint64(tmp,
512 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
513 verify(nvlist_add_uint64(config,
514 ZPOOL_CONFIG_VDEV_CHILDREN,
516 valid_top_config = B_TRUE;
519 if (nvlist_lookup_uint64_array(tmp,
520 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
522 verify(nvlist_add_uint64_array(config,
523 ZPOOL_CONFIG_HOLE_ARRAY,
524 hole_array, holes) == 0);
530 * Copy the relevant pieces of data to the pool
536 * comment (if available)
538 * hostid (if available)
539 * hostname (if available)
541 uint64_t state, version;
542 char *comment = NULL;
544 version = fnvlist_lookup_uint64(tmp,
545 ZPOOL_CONFIG_VERSION);
546 fnvlist_add_uint64(config,
547 ZPOOL_CONFIG_VERSION, version);
548 guid = fnvlist_lookup_uint64(tmp,
549 ZPOOL_CONFIG_POOL_GUID);
550 fnvlist_add_uint64(config,
551 ZPOOL_CONFIG_POOL_GUID, guid);
552 name = fnvlist_lookup_string(tmp,
553 ZPOOL_CONFIG_POOL_NAME);
554 fnvlist_add_string(config,
555 ZPOOL_CONFIG_POOL_NAME, name);
557 if (nvlist_lookup_string(tmp,
558 ZPOOL_CONFIG_COMMENT, &comment) == 0)
559 fnvlist_add_string(config,
560 ZPOOL_CONFIG_COMMENT, comment);
562 state = fnvlist_lookup_uint64(tmp,
563 ZPOOL_CONFIG_POOL_STATE);
564 fnvlist_add_uint64(config,
565 ZPOOL_CONFIG_POOL_STATE, state);
568 if (nvlist_lookup_uint64(tmp,
569 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
570 fnvlist_add_uint64(config,
571 ZPOOL_CONFIG_HOSTID, hostid);
572 hostname = fnvlist_lookup_string(tmp,
573 ZPOOL_CONFIG_HOSTNAME);
574 fnvlist_add_string(config,
575 ZPOOL_CONFIG_HOSTNAME, hostname);
578 config_seen = B_TRUE;
582 * Add this top-level vdev to the child array.
584 verify(nvlist_lookup_nvlist(tmp,
585 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
586 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
589 if (id >= children) {
592 newchild = zfs_alloc(hdl, (id + 1) *
593 sizeof (nvlist_t *));
594 if (newchild == NULL)
597 for (c = 0; c < children; c++)
598 newchild[c] = child[c];
604 if (nvlist_dup(nvtop, &child[id], 0) != 0)
610 * If we have information about all the top-levels then
611 * clean up the nvlist which we've constructed. This
612 * means removing any extraneous devices that are
613 * beyond the valid range or adding devices to the end
614 * of our array which appear to be missing.
616 if (valid_top_config) {
617 if (max_id < children) {
618 for (c = max_id; c < children; c++)
619 nvlist_free(child[c]);
621 } else if (max_id > children) {
624 newchild = zfs_alloc(hdl, (max_id) *
625 sizeof (nvlist_t *));
626 if (newchild == NULL)
629 for (c = 0; c < children; c++)
630 newchild[c] = child[c];
638 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
642 * The vdev namespace may contain holes as a result of
643 * device removal. We must add them back into the vdev
644 * tree before we process any missing devices.
647 ASSERT(valid_top_config);
649 for (c = 0; c < children; c++) {
652 if (child[c] != NULL ||
653 !vdev_is_hole(hole_array, holes, c))
656 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
661 * Holes in the namespace are treated as
662 * "hole" top-level vdevs and have a
663 * special flag set on them.
665 if (nvlist_add_string(holey,
667 VDEV_TYPE_HOLE) != 0 ||
668 nvlist_add_uint64(holey,
669 ZPOOL_CONFIG_ID, c) != 0 ||
670 nvlist_add_uint64(holey,
671 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
680 * Look for any missing top-level vdevs. If this is the case,
681 * create a faked up 'missing' vdev as a placeholder. We cannot
682 * simply compress the child array, because the kernel performs
683 * certain checks to make sure the vdev IDs match their location
684 * in the configuration.
686 for (c = 0; c < children; c++) {
687 if (child[c] == NULL) {
689 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
692 if (nvlist_add_string(missing,
694 VDEV_TYPE_MISSING) != 0 ||
695 nvlist_add_uint64(missing,
696 ZPOOL_CONFIG_ID, c) != 0 ||
697 nvlist_add_uint64(missing,
698 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
699 nvlist_free(missing);
707 * Put all of this pool's top-level vdevs into a root vdev.
709 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
711 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
712 VDEV_TYPE_ROOT) != 0 ||
713 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
714 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
715 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
716 child, children) != 0) {
721 for (c = 0; c < children; c++)
722 nvlist_free(child[c]);
728 * Go through and fix up any paths and/or devids based on our
729 * known list of vdev GUID -> path mappings.
731 if (fix_paths(nvroot, pl->names) != 0) {
737 * Add the root vdev to this pool's configuration.
739 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
747 * zdb uses this path to report on active pools that were
748 * imported or created using -R.
754 * Determine if this pool is currently active, in which case we
755 * can't actually import it.
757 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
759 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
762 if (pool_active(hdl, name, guid, &isactive) != 0)
771 if (policy != NULL) {
772 if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
777 if ((nvl = refresh_config(hdl, config)) == NULL) {
787 * Go through and update the paths for spares, now that we have
790 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
792 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
793 &spares, &nspares) == 0) {
794 for (i = 0; i < nspares; i++) {
795 if (fix_paths(spares[i], pl->names) != 0)
801 * Update the paths for l2cache devices.
803 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
804 &l2cache, &nl2cache) == 0) {
805 for (i = 0; i < nl2cache; i++) {
806 if (fix_paths(l2cache[i], pl->names) != 0)
812 * Restore the original information read from the actual label.
814 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
816 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
819 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
821 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
827 * Add this pool to the list of configs.
829 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
831 if (nvlist_add_nvlist(ret, name, config) != 0)
847 (void) no_memory(hdl);
851 for (c = 0; c < children; c++)
852 nvlist_free(child[c]);
859 * Return the offset of the given label.
862 label_offset(uint64_t size, int l)
864 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
865 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
866 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
870 * Given a file descriptor, read the label information and return an nvlist
871 * describing the configuration, if there is one.
872 * Return 0 on success, or -1 on failure
875 zpool_read_label(int fd, nvlist_t **config)
877 struct stat64 statbuf;
880 uint64_t state, txg, size;
884 if (fstat64(fd, &statbuf) == -1)
886 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
888 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
891 for (l = 0; l < VDEV_LABELS; l++) {
892 if (pread64(fd, label, sizeof (vdev_label_t),
893 label_offset(size, l)) != sizeof (vdev_label_t))
896 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
897 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
900 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
901 &state) != 0 || state > POOL_STATE_L2CACHE) {
902 nvlist_free(*config);
906 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
907 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
908 &txg) != 0 || txg == 0)) {
909 nvlist_free(*config);
924 * Given a file descriptor, read the label information and return an nvlist
925 * describing the configuration, if there is one.
926 * returns the number of valid labels found
927 * If a label is found, returns it via config. The caller is responsible for
931 zpool_read_all_labels(int fd, nvlist_t **config)
933 struct stat64 statbuf;
934 struct aiocb aiocbs[VDEV_LABELS];
935 struct aiocb *aiocbps[VDEV_LABELS];
938 uint64_t state, txg, size;
943 if (fstat64(fd, &statbuf) == -1)
945 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
947 if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
950 memset(aiocbs, 0, sizeof(aiocbs));
951 for (l = 0; l < VDEV_LABELS; l++) {
952 aiocbs[l].aio_fildes = fd;
953 aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
954 aiocbs[l].aio_buf = &labels[l];
955 aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
956 aiocbs[l].aio_lio_opcode = LIO_READ;
957 aiocbps[l] = &aiocbs[l];
960 if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
961 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
962 for (l = 0; l < VDEV_LABELS; l++) {
964 int r = aio_error(&aiocbs[l]);
966 (void)aio_return(&aiocbs[l]);
973 for (l = 0; l < VDEV_LABELS; l++) {
974 nvlist_t *temp = NULL;
976 if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
979 if (nvlist_unpack(labels[l].vp_nvlist,
980 sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
983 if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
984 &state) != 0 || state > POOL_STATE_L2CACHE) {
990 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
991 (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
992 &txg) != 0 || txg == 0)) {
1007 typedef struct rdsk_node {
1010 libzfs_handle_t *rn_hdl;
1011 nvlist_t *rn_config;
1014 boolean_t rn_nozpool;
1018 slice_cache_compare(const void *arg1, const void *arg2)
1020 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1021 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1022 char *nm1slice, *nm2slice;
1026 * slices zero and two are the most likely to provide results,
1027 * so put those first
1029 nm1slice = strstr(nm1, "s0");
1030 nm2slice = strstr(nm2, "s0");
1031 if (nm1slice && !nm2slice) {
1034 if (!nm1slice && nm2slice) {
1037 nm1slice = strstr(nm1, "s2");
1038 nm2slice = strstr(nm2, "s2");
1039 if (nm1slice && !nm2slice) {
1042 if (!nm1slice && nm2slice) {
1046 rv = strcmp(nm1, nm2);
1049 return (rv > 0 ? 1 : -1);
1054 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1055 diskaddr_t size, uint_t blksz)
1057 rdsk_node_t tmpnode;
1059 char sname[MAXNAMELEN];
1061 tmpnode.rn_name = &sname[0];
1062 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1065 * protect against division by zero for disk labels that
1066 * contain a bogus sector size
1070 /* too small to contain a zpool? */
1071 if ((size < (SPA_MINDEVSIZE / blksz)) &&
1072 (node = avl_find(r, &tmpnode, NULL)))
1073 node->rn_nozpool = B_TRUE;
1075 #endif /* illumos */
1078 nozpool_all_slices(avl_tree_t *r, const char *sname)
1081 char diskname[MAXNAMELEN];
1085 (void) strncpy(diskname, sname, MAXNAMELEN);
1086 if (((ptr = strrchr(diskname, 's')) == NULL) &&
1087 ((ptr = strrchr(diskname, 'p')) == NULL))
1091 for (i = 0; i < NDKMAP; i++)
1092 check_one_slice(r, diskname, i, 0, 1);
1094 for (i = 0; i <= FD_NUMPART; i++)
1095 check_one_slice(r, diskname, i, 0, 1);
1096 #endif /* illumos */
1101 check_slices(avl_tree_t *r, int fd, const char *sname)
1103 struct extvtoc vtoc;
1105 char diskname[MAXNAMELEN];
1109 (void) strncpy(diskname, sname, MAXNAMELEN);
1110 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1114 if (read_extvtoc(fd, &vtoc) >= 0) {
1115 for (i = 0; i < NDKMAP; i++)
1116 check_one_slice(r, diskname, i,
1117 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1118 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1120 * on x86 we'll still have leftover links that point
1121 * to slices s[9-15], so use NDKMAP instead
1123 for (i = 0; i < NDKMAP; i++)
1124 check_one_slice(r, diskname, i,
1125 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1126 /* nodes p[1-4] are never used with EFI labels */
1128 for (i = 1; i <= FD_NUMPART; i++)
1129 check_one_slice(r, diskname, i, 0, 1);
1133 #endif /* illumos */
1136 zpool_open_func(void *arg)
1138 rdsk_node_t *rn = arg;
1139 struct stat64 statbuf;
1145 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1146 /* symlink to a device that's no longer there */
1147 if (errno == ENOENT)
1148 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1152 * Ignore failed stats. We only want regular
1153 * files, character devs and block devs.
1155 if (fstat64(fd, &statbuf) != 0 ||
1156 (!S_ISREG(statbuf.st_mode) &&
1157 !S_ISCHR(statbuf.st_mode) &&
1158 !S_ISBLK(statbuf.st_mode))) {
1162 /* this file is too small to hold a zpool */
1164 if (S_ISREG(statbuf.st_mode) &&
1165 statbuf.st_size < SPA_MINDEVSIZE) {
1168 } else if (!S_ISREG(statbuf.st_mode)) {
1170 * Try to read the disk label first so we don't have to
1171 * open a bunch of minor nodes that can't have a zpool.
1173 check_slices(rn->rn_avl, fd, rn->rn_name);
1175 #else /* !illumos */
1176 if (statbuf.st_size < SPA_MINDEVSIZE) {
1180 #endif /* illumos */
1182 if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1184 (void) no_memory(rn->rn_hdl);
1189 rn->rn_config = config;
1193 * Given a file descriptor, clear (zero) the label information.
1196 zpool_clear_label(int fd)
1198 struct stat64 statbuf;
1200 vdev_label_t *label;
1203 if (fstat64(fd, &statbuf) == -1)
1205 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1207 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1210 for (l = 0; l < VDEV_LABELS; l++) {
1211 if (pwrite64(fd, label, sizeof (vdev_label_t),
1212 label_offset(size, l)) != sizeof (vdev_label_t)) {
1223 * Given a list of directories to search, find all pools stored on disk. This
1224 * includes partial pools which are not available to import. If no args are
1225 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1226 * poolname or guid (but not both) are provided by the caller when trying
1227 * to import a specific pool.
1230 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1232 int i, dirs = iarg->paths;
1233 struct dirent64 *dp;
1234 char path[MAXPATHLEN];
1235 char *end, **dir = iarg->path;
1237 nvlist_t *ret = NULL;
1238 static char *default_dir = "/dev";
1239 pool_list_t pools = { 0 };
1240 pool_entry_t *pe, *penext;
1241 vdev_entry_t *ve, *venext;
1242 config_entry_t *ce, *cenext;
1243 name_entry_t *ne, *nenext;
1244 avl_tree_t slice_cache;
1254 * Go through and read the label configuration information from every
1255 * possible device, organizing the information according to pool GUID
1256 * and toplevel GUID.
1258 for (i = 0; i < dirs; i++) {
1260 char rdsk[MAXPATHLEN];
1262 boolean_t config_failed = B_FALSE;
1265 /* use realpath to normalize the path */
1266 if (realpath(dir[i], path) == 0) {
1267 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1268 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1271 end = &path[strlen(path)];
1274 pathleft = &path[sizeof (path)] - end;
1278 * Using raw devices instead of block devices when we're
1279 * reading the labels skips a bunch of slow operations during
1280 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1282 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1283 (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1286 (void) strlcpy(rdsk, path, sizeof (rdsk));
1288 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1289 (dirp = fdopendir(dfd)) == NULL) {
1292 zfs_error_aux(hdl, strerror(errno));
1293 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1294 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1299 avl_create(&slice_cache, slice_cache_compare,
1300 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1302 if (strcmp(rdsk, "/dev/") == 0) {
1306 struct gprovider *pp;
1308 errno = geom_gettree(&mesh);
1310 zfs_error_aux(hdl, strerror(errno));
1311 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1312 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1316 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1317 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1318 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1319 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1320 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1321 slice->rn_avl = &slice_cache;
1322 slice->rn_dfd = dfd;
1323 slice->rn_hdl = hdl;
1324 slice->rn_nozpool = B_FALSE;
1325 avl_add(&slice_cache, slice);
1330 geom_deletetree(&mesh);
1335 * This is not MT-safe, but we have no MT consumers of libzfs
1337 while ((dp = readdir64(dirp)) != NULL) {
1338 const char *name = dp->d_name;
1339 if (name[0] == '.' &&
1340 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1343 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1344 slice->rn_name = zfs_strdup(hdl, name);
1345 slice->rn_avl = &slice_cache;
1346 slice->rn_dfd = dfd;
1347 slice->rn_hdl = hdl;
1348 slice->rn_nozpool = B_FALSE;
1349 avl_add(&slice_cache, slice);
1353 * create a thread pool to do all of this in parallel;
1354 * rn_nozpool is not protected, so this is racy in that
1355 * multiple tasks could decide that the same slice can
1356 * not hold a zpool, which is benign. Also choose
1357 * double the number of processors; we hold a lot of
1358 * locks in the kernel, so going beyond this doesn't
1361 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1363 for (slice = avl_first(&slice_cache); slice;
1364 (slice = avl_walk(&slice_cache, slice,
1366 (void) tpool_dispatch(t, zpool_open_func, slice);
1371 while ((slice = avl_destroy_nodes(&slice_cache,
1372 &cookie)) != NULL) {
1373 if (slice->rn_config != NULL && !config_failed) {
1374 nvlist_t *config = slice->rn_config;
1375 boolean_t matched = B_TRUE;
1377 if (iarg->poolname != NULL) {
1380 matched = nvlist_lookup_string(config,
1381 ZPOOL_CONFIG_POOL_NAME,
1383 strcmp(iarg->poolname, pname) == 0;
1384 } else if (iarg->guid != 0) {
1387 matched = nvlist_lookup_uint64(config,
1388 ZPOOL_CONFIG_POOL_GUID,
1390 iarg->guid == this_guid;
1394 * use the non-raw path for the config
1396 (void) strlcpy(end, slice->rn_name,
1398 if (add_config(hdl, &pools, path,
1400 config_failed = B_TRUE;
1402 nvlist_free(config);
1404 free(slice->rn_name);
1407 avl_destroy(&slice_cache);
1409 (void) closedir(dirp);
1415 ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1418 for (pe = pools.pools; pe != NULL; pe = penext) {
1419 penext = pe->pe_next;
1420 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1421 venext = ve->ve_next;
1422 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1423 cenext = ce->ce_next;
1424 nvlist_free(ce->ce_config);
1432 for (ne = pools.names; ne != NULL; ne = nenext) {
1433 nenext = ne->ne_next;
1442 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1444 importargs_t iarg = { 0 };
1449 return (zpool_find_import_impl(hdl, &iarg));
1453 * Given a cache file, return the contents as a list of importable pools.
1454 * poolname or guid (but not both) are provided by the caller when trying
1455 * to import a specific pool.
1458 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1459 char *poolname, uint64_t guid)
1463 struct stat64 statbuf;
1464 nvlist_t *raw, *src, *dst;
1471 verify(poolname == NULL || guid == 0);
1473 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1474 zfs_error_aux(hdl, "%s", strerror(errno));
1475 (void) zfs_error(hdl, EZFS_BADCACHE,
1476 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1480 if (fstat64(fd, &statbuf) != 0) {
1481 zfs_error_aux(hdl, "%s", strerror(errno));
1483 (void) zfs_error(hdl, EZFS_BADCACHE,
1484 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1488 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1493 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1496 (void) zfs_error(hdl, EZFS_BADCACHE,
1497 dgettext(TEXT_DOMAIN,
1498 "failed to read cache file contents"));
1504 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1506 (void) zfs_error(hdl, EZFS_BADCACHE,
1507 dgettext(TEXT_DOMAIN,
1508 "invalid or corrupt cache file contents"));
1515 * Go through and get the current state of the pools and refresh their
1518 if (nvlist_alloc(&pools, 0, 0) != 0) {
1519 (void) no_memory(hdl);
1525 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1526 src = fnvpair_value_nvlist(elem);
1528 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1529 if (poolname != NULL && strcmp(poolname, name) != 0)
1532 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1533 if (guid != 0 && guid != this_guid)
1536 if (pool_active(hdl, name, this_guid, &active) != 0) {
1545 if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1547 (void) no_memory(hdl);
1553 if ((dst = refresh_config(hdl, src)) == NULL) {
1559 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1560 (void) no_memory(hdl);
1574 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1576 importargs_t *import = data;
1579 if (import->poolname != NULL) {
1582 verify(nvlist_lookup_string(zhp->zpool_config,
1583 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1584 if (strcmp(pool_name, import->poolname) == 0)
1589 verify(nvlist_lookup_uint64(zhp->zpool_config,
1590 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1591 if (pool_guid == import->guid)
1600 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1602 nvlist_t *pools = NULL;
1604 verify(import->poolname == NULL || import->guid == 0);
1607 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1609 if (import->cachefile != NULL)
1610 pools = zpool_find_import_cached(hdl, import->cachefile,
1611 import->poolname, import->guid);
1613 pools = zpool_find_import_impl(hdl, import);
1619 pool_match(nvlist_t *cfg, char *tgt)
1621 uint64_t v, guid = strtoull(tgt, NULL, 0);
1625 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1628 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1629 return (strcmp(s, tgt) == 0);
1635 zpool_tryimport(libzfs_handle_t *hdl, char *target, nvlist_t **configp,
1639 nvlist_t *match = NULL;
1640 nvlist_t *config = NULL;
1643 char *targetdup = strdup(target);
1647 if ((sepp = strpbrk(targetdup, "/@")) != NULL) {
1651 pools = zpool_search_import(hdl, args);
1653 if (pools != NULL) {
1654 nvpair_t *elem = NULL;
1655 while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1656 VERIFY0(nvpair_value_nvlist(elem, &config));
1657 if (pool_match(config, targetdup)) {
1659 if (match != NULL) {
1660 /* multiple matches found */
1686 find_guid(nvlist_t *nv, uint64_t guid)
1692 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1696 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1697 &child, &children) == 0) {
1698 for (c = 0; c < children; c++)
1699 if (find_guid(child[c], guid))
1706 typedef struct aux_cbdata {
1707 const char *cb_type;
1709 zpool_handle_t *cb_zhp;
1713 find_aux(zpool_handle_t *zhp, void *data)
1715 aux_cbdata_t *cbp = data;
1721 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1724 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1725 &list, &count) == 0) {
1726 for (i = 0; i < count; i++) {
1727 verify(nvlist_lookup_uint64(list[i],
1728 ZPOOL_CONFIG_GUID, &guid) == 0);
1729 if (guid == cbp->cb_guid) {
1741 * Determines if the pool is in use. If so, it returns true and the state of
1742 * the pool as well as the name of the pool. Both strings are allocated and
1743 * must be freed by the caller.
1746 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1752 uint64_t guid, vdev_guid;
1753 zpool_handle_t *zhp;
1754 nvlist_t *pool_config;
1755 uint64_t stateval, isspare;
1756 aux_cbdata_t cb = { 0 };
1761 if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1762 (void) no_memory(hdl);
1769 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1771 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1774 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1775 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1777 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1782 case POOL_STATE_EXPORTED:
1784 * A pool with an exported state may in fact be imported
1785 * read-only, so check the in-core state to see if it's
1786 * active and imported read-only. If it is, set
1787 * its state to active.
1789 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1790 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1791 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1792 stateval = POOL_STATE_ACTIVE;
1795 * All we needed the zpool handle for is the
1796 * readonly prop check.
1804 case POOL_STATE_ACTIVE:
1806 * For an active pool, we have to determine if it's really part
1807 * of a currently active pool (in which case the pool will exist
1808 * and the guid will be the same), or whether it's part of an
1809 * active pool that was disconnected without being explicitly
1812 if (pool_active(hdl, name, guid, &isactive) != 0) {
1813 nvlist_free(config);
1819 * Because the device may have been removed while
1820 * offlined, we only report it as active if the vdev is
1821 * still present in the config. Otherwise, pretend like
1824 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1825 (pool_config = zpool_get_config(zhp, NULL))
1829 verify(nvlist_lookup_nvlist(pool_config,
1830 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1831 ret = find_guid(nvroot, vdev_guid);
1837 * If this is an active spare within another pool, we
1838 * treat it like an unused hot spare. This allows the
1839 * user to create a pool with a hot spare that currently
1840 * in use within another pool. Since we return B_TRUE,
1841 * libdiskmgt will continue to prevent generic consumers
1842 * from using the device.
1844 if (ret && nvlist_lookup_uint64(config,
1845 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1846 stateval = POOL_STATE_SPARE;
1851 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1856 case POOL_STATE_SPARE:
1858 * For a hot spare, it can be either definitively in use, or
1859 * potentially active. To determine if it's in use, we iterate
1860 * over all pools in the system and search for one with a spare
1861 * with a matching guid.
1863 * Due to the shared nature of spares, we don't actually report
1864 * the potentially active case as in use. This means the user
1865 * can freely create pools on the hot spares of exported pools,
1866 * but to do otherwise makes the resulting code complicated, and
1867 * we end up having to deal with this case anyway.
1870 cb.cb_guid = vdev_guid;
1871 cb.cb_type = ZPOOL_CONFIG_SPARES;
1872 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1873 name = (char *)zpool_get_name(cb.cb_zhp);
1880 case POOL_STATE_L2CACHE:
1883 * Check if any pool is currently using this l2cache device.
1886 cb.cb_guid = vdev_guid;
1887 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1888 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1889 name = (char *)zpool_get_name(cb.cb_zhp);
1902 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1904 zpool_close(cb.cb_zhp);
1905 nvlist_free(config);
1908 *state = (pool_state_t)stateval;
1912 zpool_close(cb.cb_zhp);
1914 nvlist_free(config);