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, 2016 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) {
243 ne->ne_guid = vdev_guid;
244 ne->ne_next = pl->names;
250 * If we have a valid config but cannot read any of these fields, then
251 * it means we have a half-initialized label. In vdev_label_init()
252 * we write a label with txg == 0 so that we can identify the device
253 * in case the user refers to the same disk later on. If we fail to
254 * create the pool, we'll be left with a label in this state
255 * which should not be considered part of a valid pool.
257 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
259 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
261 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
263 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
264 &txg) != 0 || txg == 0) {
270 * First, see if we know about this pool. If not, then add it to the
271 * list of known pools.
273 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
274 if (pe->pe_guid == pool_guid)
279 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) {
302 ve->ve_guid = top_guid;
303 ve->ve_next = pe->pe_vdevs;
308 * Third, see if we have a config with a matching transaction group. If
309 * so, then we do nothing. Otherwise, add it to the list of known
312 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
313 if (ce->ce_txg == txg)
318 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
323 ce->ce_config = config;
324 ce->ce_next = ve->ve_configs;
331 * At this point we've successfully added our config to the list of
332 * known configs. The last thing to do is add the vdev guid -> path
333 * mappings so that we can fix up the configuration as necessary before
336 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
339 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
344 ne->ne_guid = vdev_guid;
345 ne->ne_next = pl->names;
352 * Returns true if the named pool matches the given GUID.
355 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
361 if (zpool_open_silent(hdl, name, &zhp) != 0)
369 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
374 *isactive = (theguid == guid);
379 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
382 zfs_cmd_t zc = { 0 };
383 int err, dstbuf_size;
385 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
388 dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
390 if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
391 zcmd_free_nvlists(&zc);
395 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
396 &zc)) != 0 && errno == ENOMEM) {
397 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
398 zcmd_free_nvlists(&zc);
404 zcmd_free_nvlists(&zc);
408 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
409 zcmd_free_nvlists(&zc);
413 zcmd_free_nvlists(&zc);
418 * Determine if the vdev id is a hole in the namespace.
421 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
423 for (int c = 0; c < holes; c++) {
425 /* Top-level is a hole */
426 if (hole_array[c] == id)
433 * Convert our list of pools into the definitive set of configurations. We
434 * start by picking the best config for each toplevel vdev. Once that's done,
435 * we assemble the toplevel vdevs into a full config for the pool. We make a
436 * pass to fix up any incorrect paths, and then add it to the main list to
437 * return to the user.
440 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
445 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
446 nvlist_t **spares, **l2cache;
447 uint_t i, nspares, nl2cache;
448 boolean_t config_seen;
450 char *name, *hostname = NULL;
453 nvlist_t **child = NULL;
455 uint64_t *hole_array, max_id;
460 boolean_t found_one = B_FALSE;
461 boolean_t valid_top_config = B_FALSE;
463 if (nvlist_alloc(&ret, 0, 0) != 0)
466 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
467 uint64_t id, max_txg = 0;
469 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
471 config_seen = B_FALSE;
474 * Iterate over all toplevel vdevs. Grab the pool configuration
475 * from the first one we find, and then go through the rest and
476 * add them as necessary to the 'vdevs' member of the config.
478 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
481 * Determine the best configuration for this vdev by
482 * selecting the config with the latest transaction
486 for (ce = ve->ve_configs; ce != NULL;
489 if (ce->ce_txg > best_txg) {
491 best_txg = ce->ce_txg;
496 * We rely on the fact that the max txg for the
497 * pool will contain the most up-to-date information
498 * about the valid top-levels in the vdev namespace.
500 if (best_txg > max_txg) {
501 (void) nvlist_remove(config,
502 ZPOOL_CONFIG_VDEV_CHILDREN,
504 (void) nvlist_remove(config,
505 ZPOOL_CONFIG_HOLE_ARRAY,
506 DATA_TYPE_UINT64_ARRAY);
512 valid_top_config = B_FALSE;
514 if (nvlist_lookup_uint64(tmp,
515 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
516 verify(nvlist_add_uint64(config,
517 ZPOOL_CONFIG_VDEV_CHILDREN,
519 valid_top_config = B_TRUE;
522 if (nvlist_lookup_uint64_array(tmp,
523 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
525 verify(nvlist_add_uint64_array(config,
526 ZPOOL_CONFIG_HOLE_ARRAY,
527 hole_array, holes) == 0);
533 * Copy the relevant pieces of data to the pool
539 * comment (if available)
541 * hostid (if available)
542 * hostname (if available)
544 uint64_t state, version;
545 char *comment = NULL;
547 version = fnvlist_lookup_uint64(tmp,
548 ZPOOL_CONFIG_VERSION);
549 fnvlist_add_uint64(config,
550 ZPOOL_CONFIG_VERSION, version);
551 guid = fnvlist_lookup_uint64(tmp,
552 ZPOOL_CONFIG_POOL_GUID);
553 fnvlist_add_uint64(config,
554 ZPOOL_CONFIG_POOL_GUID, guid);
555 name = fnvlist_lookup_string(tmp,
556 ZPOOL_CONFIG_POOL_NAME);
557 fnvlist_add_string(config,
558 ZPOOL_CONFIG_POOL_NAME, name);
560 if (nvlist_lookup_string(tmp,
561 ZPOOL_CONFIG_COMMENT, &comment) == 0)
562 fnvlist_add_string(config,
563 ZPOOL_CONFIG_COMMENT, comment);
565 state = fnvlist_lookup_uint64(tmp,
566 ZPOOL_CONFIG_POOL_STATE);
567 fnvlist_add_uint64(config,
568 ZPOOL_CONFIG_POOL_STATE, state);
571 if (nvlist_lookup_uint64(tmp,
572 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
573 fnvlist_add_uint64(config,
574 ZPOOL_CONFIG_HOSTID, hostid);
575 hostname = fnvlist_lookup_string(tmp,
576 ZPOOL_CONFIG_HOSTNAME);
577 fnvlist_add_string(config,
578 ZPOOL_CONFIG_HOSTNAME, hostname);
581 config_seen = B_TRUE;
585 * Add this top-level vdev to the child array.
587 verify(nvlist_lookup_nvlist(tmp,
588 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
589 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
592 if (id >= children) {
595 newchild = zfs_alloc(hdl, (id + 1) *
596 sizeof (nvlist_t *));
597 if (newchild == NULL)
600 for (c = 0; c < children; c++)
601 newchild[c] = child[c];
607 if (nvlist_dup(nvtop, &child[id], 0) != 0)
613 * If we have information about all the top-levels then
614 * clean up the nvlist which we've constructed. This
615 * means removing any extraneous devices that are
616 * beyond the valid range or adding devices to the end
617 * of our array which appear to be missing.
619 if (valid_top_config) {
620 if (max_id < children) {
621 for (c = max_id; c < children; c++)
622 nvlist_free(child[c]);
624 } else if (max_id > children) {
627 newchild = zfs_alloc(hdl, (max_id) *
628 sizeof (nvlist_t *));
629 if (newchild == NULL)
632 for (c = 0; c < children; c++)
633 newchild[c] = child[c];
641 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
645 * The vdev namespace may contain holes as a result of
646 * device removal. We must add them back into the vdev
647 * tree before we process any missing devices.
650 ASSERT(valid_top_config);
652 for (c = 0; c < children; c++) {
655 if (child[c] != NULL ||
656 !vdev_is_hole(hole_array, holes, c))
659 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
664 * Holes in the namespace are treated as
665 * "hole" top-level vdevs and have a
666 * special flag set on them.
668 if (nvlist_add_string(holey,
670 VDEV_TYPE_HOLE) != 0 ||
671 nvlist_add_uint64(holey,
672 ZPOOL_CONFIG_ID, c) != 0 ||
673 nvlist_add_uint64(holey,
674 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
683 * Look for any missing top-level vdevs. If this is the case,
684 * create a faked up 'missing' vdev as a placeholder. We cannot
685 * simply compress the child array, because the kernel performs
686 * certain checks to make sure the vdev IDs match their location
687 * in the configuration.
689 for (c = 0; c < children; c++) {
690 if (child[c] == NULL) {
692 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
695 if (nvlist_add_string(missing,
697 VDEV_TYPE_MISSING) != 0 ||
698 nvlist_add_uint64(missing,
699 ZPOOL_CONFIG_ID, c) != 0 ||
700 nvlist_add_uint64(missing,
701 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
702 nvlist_free(missing);
710 * Put all of this pool's top-level vdevs into a root vdev.
712 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
714 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
715 VDEV_TYPE_ROOT) != 0 ||
716 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
717 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
718 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
719 child, children) != 0) {
724 for (c = 0; c < children; c++)
725 nvlist_free(child[c]);
731 * Go through and fix up any paths and/or devids based on our
732 * known list of vdev GUID -> path mappings.
734 if (fix_paths(nvroot, pl->names) != 0) {
740 * Add the root vdev to this pool's configuration.
742 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
750 * zdb uses this path to report on active pools that were
751 * imported or created using -R.
757 * Determine if this pool is currently active, in which case we
758 * can't actually import it.
760 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
762 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
765 if (pool_active(hdl, name, guid, &isactive) != 0)
774 if ((nvl = refresh_config(hdl, config)) == NULL) {
784 * Go through and update the paths for spares, now that we have
787 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
789 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
790 &spares, &nspares) == 0) {
791 for (i = 0; i < nspares; i++) {
792 if (fix_paths(spares[i], pl->names) != 0)
798 * Update the paths for l2cache devices.
800 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
801 &l2cache, &nl2cache) == 0) {
802 for (i = 0; i < nl2cache; i++) {
803 if (fix_paths(l2cache[i], pl->names) != 0)
809 * Restore the original information read from the actual label.
811 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
813 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
816 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
818 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
824 * Add this pool to the list of configs.
826 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
828 if (nvlist_add_nvlist(ret, name, config) != 0)
844 (void) no_memory(hdl);
848 for (c = 0; c < children; c++)
849 nvlist_free(child[c]);
856 * Return the offset of the given label.
859 label_offset(uint64_t size, int l)
861 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
862 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
863 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
867 * Given a file descriptor, read the label information and return an nvlist
868 * describing the configuration, if there is one.
869 * Return 0 on success, or -1 on failure
872 zpool_read_label(int fd, nvlist_t **config)
874 struct stat64 statbuf;
877 uint64_t state, txg, size;
881 if (fstat64(fd, &statbuf) == -1)
883 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
885 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
888 for (l = 0; l < VDEV_LABELS; l++) {
889 if (pread64(fd, label, sizeof (vdev_label_t),
890 label_offset(size, l)) != sizeof (vdev_label_t))
893 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
894 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
897 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
898 &state) != 0 || state > POOL_STATE_L2CACHE) {
899 nvlist_free(*config);
903 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
904 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
905 &txg) != 0 || txg == 0)) {
906 nvlist_free(*config);
920 * Given a file descriptor, read the label information and return an nvlist
921 * describing the configuration, if there is one.
922 * returns the number of valid labels found
923 * If a label is found, returns it via config. The caller is responsible for
927 zpool_read_all_labels(int fd, nvlist_t **config)
929 struct stat64 statbuf;
930 struct aiocb aiocbs[VDEV_LABELS];
931 struct aiocb *aiocbps[VDEV_LABELS];
934 uint64_t state, txg, size;
939 if (fstat64(fd, &statbuf) == -1)
941 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
943 if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
946 memset(aiocbs, 0, sizeof(aiocbs));
947 for (l = 0; l < VDEV_LABELS; l++) {
948 aiocbs[l].aio_fildes = fd;
949 aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
950 aiocbs[l].aio_buf = &labels[l];
951 aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
952 aiocbs[l].aio_lio_opcode = LIO_READ;
953 aiocbps[l] = &aiocbs[l];
956 if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
957 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
958 for (l = 0; l < VDEV_LABELS; l++) {
960 int r = aio_error(&aiocbs[l]);
962 (void)aio_return(&aiocbs[l]);
969 for (l = 0; l < VDEV_LABELS; l++) {
970 nvlist_t *temp = NULL;
972 if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
975 if (nvlist_unpack(labels[l].vp_nvlist,
976 sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
979 if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
980 &state) != 0 || state > POOL_STATE_L2CACHE) {
986 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
987 (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
988 &txg) != 0 || txg == 0)) {
1003 typedef struct rdsk_node {
1006 libzfs_handle_t *rn_hdl;
1007 nvlist_t *rn_config;
1010 boolean_t rn_nozpool;
1014 slice_cache_compare(const void *arg1, const void *arg2)
1016 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1017 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1018 char *nm1slice, *nm2slice;
1022 * slices zero and two are the most likely to provide results,
1023 * so put those first
1025 nm1slice = strstr(nm1, "s0");
1026 nm2slice = strstr(nm2, "s0");
1027 if (nm1slice && !nm2slice) {
1030 if (!nm1slice && nm2slice) {
1033 nm1slice = strstr(nm1, "s2");
1034 nm2slice = strstr(nm2, "s2");
1035 if (nm1slice && !nm2slice) {
1038 if (!nm1slice && nm2slice) {
1042 rv = strcmp(nm1, nm2);
1045 return (rv > 0 ? 1 : -1);
1050 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1051 diskaddr_t size, uint_t blksz)
1053 rdsk_node_t tmpnode;
1055 char sname[MAXNAMELEN];
1057 tmpnode.rn_name = &sname[0];
1058 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1061 * protect against division by zero for disk labels that
1062 * contain a bogus sector size
1066 /* too small to contain a zpool? */
1067 if ((size < (SPA_MINDEVSIZE / blksz)) &&
1068 (node = avl_find(r, &tmpnode, NULL)))
1069 node->rn_nozpool = B_TRUE;
1071 #endif /* illumos */
1074 nozpool_all_slices(avl_tree_t *r, const char *sname)
1077 char diskname[MAXNAMELEN];
1081 (void) strncpy(diskname, sname, MAXNAMELEN);
1082 if (((ptr = strrchr(diskname, 's')) == NULL) &&
1083 ((ptr = strrchr(diskname, 'p')) == NULL))
1087 for (i = 0; i < NDKMAP; i++)
1088 check_one_slice(r, diskname, i, 0, 1);
1090 for (i = 0; i <= FD_NUMPART; i++)
1091 check_one_slice(r, diskname, i, 0, 1);
1092 #endif /* illumos */
1097 check_slices(avl_tree_t *r, int fd, const char *sname)
1099 struct extvtoc vtoc;
1101 char diskname[MAXNAMELEN];
1105 (void) strncpy(diskname, sname, MAXNAMELEN);
1106 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1110 if (read_extvtoc(fd, &vtoc) >= 0) {
1111 for (i = 0; i < NDKMAP; i++)
1112 check_one_slice(r, diskname, i,
1113 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1114 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1116 * on x86 we'll still have leftover links that point
1117 * to slices s[9-15], so use NDKMAP instead
1119 for (i = 0; i < NDKMAP; i++)
1120 check_one_slice(r, diskname, i,
1121 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1122 /* nodes p[1-4] are never used with EFI labels */
1124 for (i = 1; i <= FD_NUMPART; i++)
1125 check_one_slice(r, diskname, i, 0, 1);
1129 #endif /* illumos */
1132 zpool_open_func(void *arg)
1134 rdsk_node_t *rn = arg;
1135 struct stat64 statbuf;
1141 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1142 /* symlink to a device that's no longer there */
1143 if (errno == ENOENT)
1144 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1148 * Ignore failed stats. We only want regular
1149 * files, character devs and block devs.
1151 if (fstat64(fd, &statbuf) != 0 ||
1152 (!S_ISREG(statbuf.st_mode) &&
1153 !S_ISCHR(statbuf.st_mode) &&
1154 !S_ISBLK(statbuf.st_mode))) {
1158 /* this file is too small to hold a zpool */
1160 if (S_ISREG(statbuf.st_mode) &&
1161 statbuf.st_size < SPA_MINDEVSIZE) {
1164 } else if (!S_ISREG(statbuf.st_mode)) {
1166 * Try to read the disk label first so we don't have to
1167 * open a bunch of minor nodes that can't have a zpool.
1169 check_slices(rn->rn_avl, fd, rn->rn_name);
1171 #else /* !illumos */
1172 if (statbuf.st_size < SPA_MINDEVSIZE) {
1176 #endif /* illumos */
1178 if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1180 (void) no_memory(rn->rn_hdl);
1185 rn->rn_config = config;
1189 * Given a file descriptor, clear (zero) the label information.
1192 zpool_clear_label(int fd)
1194 struct stat64 statbuf;
1196 vdev_label_t *label;
1199 if (fstat64(fd, &statbuf) == -1)
1201 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1203 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1206 for (l = 0; l < VDEV_LABELS; l++) {
1207 if (pwrite64(fd, label, sizeof (vdev_label_t),
1208 label_offset(size, l)) != sizeof (vdev_label_t)) {
1219 * Given a list of directories to search, find all pools stored on disk. This
1220 * includes partial pools which are not available to import. If no args are
1221 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1222 * poolname or guid (but not both) are provided by the caller when trying
1223 * to import a specific pool.
1226 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1228 int i, dirs = iarg->paths;
1229 struct dirent64 *dp;
1230 char path[MAXPATHLEN];
1231 char *end, **dir = iarg->path;
1233 nvlist_t *ret = NULL;
1234 static char *default_dir = "/dev";
1235 pool_list_t pools = { 0 };
1236 pool_entry_t *pe, *penext;
1237 vdev_entry_t *ve, *venext;
1238 config_entry_t *ce, *cenext;
1239 name_entry_t *ne, *nenext;
1240 avl_tree_t slice_cache;
1250 * Go through and read the label configuration information from every
1251 * possible device, organizing the information according to pool GUID
1252 * and toplevel GUID.
1254 for (i = 0; i < dirs; i++) {
1256 char rdsk[MAXPATHLEN];
1258 boolean_t config_failed = B_FALSE;
1261 /* use realpath to normalize the path */
1262 if (realpath(dir[i], path) == 0) {
1263 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1264 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1267 end = &path[strlen(path)];
1270 pathleft = &path[sizeof (path)] - end;
1274 * Using raw devices instead of block devices when we're
1275 * reading the labels skips a bunch of slow operations during
1276 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1278 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1279 (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1282 (void) strlcpy(rdsk, path, sizeof (rdsk));
1284 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1285 (dirp = fdopendir(dfd)) == NULL) {
1288 zfs_error_aux(hdl, strerror(errno));
1289 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1290 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1295 avl_create(&slice_cache, slice_cache_compare,
1296 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1298 if (strcmp(rdsk, "/dev/") == 0) {
1302 struct gprovider *pp;
1304 errno = geom_gettree(&mesh);
1306 zfs_error_aux(hdl, strerror(errno));
1307 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1308 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1312 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1313 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1314 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1315 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1316 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1317 slice->rn_avl = &slice_cache;
1318 slice->rn_dfd = dfd;
1319 slice->rn_hdl = hdl;
1320 slice->rn_nozpool = B_FALSE;
1321 avl_add(&slice_cache, slice);
1326 geom_deletetree(&mesh);
1331 * This is not MT-safe, but we have no MT consumers of libzfs
1333 while ((dp = readdir64(dirp)) != NULL) {
1334 const char *name = dp->d_name;
1335 if (name[0] == '.' &&
1336 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1339 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1340 slice->rn_name = zfs_strdup(hdl, name);
1341 slice->rn_avl = &slice_cache;
1342 slice->rn_dfd = dfd;
1343 slice->rn_hdl = hdl;
1344 slice->rn_nozpool = B_FALSE;
1345 avl_add(&slice_cache, slice);
1349 * create a thread pool to do all of this in parallel;
1350 * rn_nozpool is not protected, so this is racy in that
1351 * multiple tasks could decide that the same slice can
1352 * not hold a zpool, which is benign. Also choose
1353 * double the number of processors; we hold a lot of
1354 * locks in the kernel, so going beyond this doesn't
1357 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1359 for (slice = avl_first(&slice_cache); slice;
1360 (slice = avl_walk(&slice_cache, slice,
1362 (void) tpool_dispatch(t, zpool_open_func, slice);
1367 while ((slice = avl_destroy_nodes(&slice_cache,
1368 &cookie)) != NULL) {
1369 if (slice->rn_config != NULL && !config_failed) {
1370 nvlist_t *config = slice->rn_config;
1371 boolean_t matched = B_TRUE;
1373 if (iarg->poolname != NULL) {
1376 matched = nvlist_lookup_string(config,
1377 ZPOOL_CONFIG_POOL_NAME,
1379 strcmp(iarg->poolname, pname) == 0;
1380 } else if (iarg->guid != 0) {
1383 matched = nvlist_lookup_uint64(config,
1384 ZPOOL_CONFIG_POOL_GUID,
1386 iarg->guid == this_guid;
1389 nvlist_free(config);
1392 * use the non-raw path for the config
1394 (void) strlcpy(end, slice->rn_name,
1396 if (add_config(hdl, &pools, path,
1398 config_failed = B_TRUE;
1401 free(slice->rn_name);
1404 avl_destroy(&slice_cache);
1406 (void) closedir(dirp);
1412 ret = get_configs(hdl, &pools, iarg->can_be_active);
1415 for (pe = pools.pools; pe != NULL; pe = penext) {
1416 penext = pe->pe_next;
1417 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1418 venext = ve->ve_next;
1419 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1420 cenext = ce->ce_next;
1421 nvlist_free(ce->ce_config);
1429 for (ne = pools.names; ne != NULL; ne = nenext) {
1430 nenext = ne->ne_next;
1439 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1441 importargs_t iarg = { 0 };
1446 return (zpool_find_import_impl(hdl, &iarg));
1450 * Given a cache file, return the contents as a list of importable pools.
1451 * poolname or guid (but not both) are provided by the caller when trying
1452 * to import a specific pool.
1455 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1456 char *poolname, uint64_t guid)
1460 struct stat64 statbuf;
1461 nvlist_t *raw, *src, *dst;
1468 verify(poolname == NULL || guid == 0);
1470 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1471 zfs_error_aux(hdl, "%s", strerror(errno));
1472 (void) zfs_error(hdl, EZFS_BADCACHE,
1473 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1477 if (fstat64(fd, &statbuf) != 0) {
1478 zfs_error_aux(hdl, "%s", strerror(errno));
1480 (void) zfs_error(hdl, EZFS_BADCACHE,
1481 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1485 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1490 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1493 (void) zfs_error(hdl, EZFS_BADCACHE,
1494 dgettext(TEXT_DOMAIN,
1495 "failed to read cache file contents"));
1501 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1503 (void) zfs_error(hdl, EZFS_BADCACHE,
1504 dgettext(TEXT_DOMAIN,
1505 "invalid or corrupt cache file contents"));
1512 * Go through and get the current state of the pools and refresh their
1515 if (nvlist_alloc(&pools, 0, 0) != 0) {
1516 (void) no_memory(hdl);
1522 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1523 src = fnvpair_value_nvlist(elem);
1525 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1526 if (poolname != NULL && strcmp(poolname, name) != 0)
1529 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1530 if (guid != 0 && guid != this_guid)
1533 if (pool_active(hdl, name, this_guid, &active) != 0) {
1542 if ((dst = refresh_config(hdl, src)) == NULL) {
1548 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1549 (void) no_memory(hdl);
1563 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1565 importargs_t *import = data;
1568 if (import->poolname != NULL) {
1571 verify(nvlist_lookup_string(zhp->zpool_config,
1572 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1573 if (strcmp(pool_name, import->poolname) == 0)
1578 verify(nvlist_lookup_uint64(zhp->zpool_config,
1579 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1580 if (pool_guid == import->guid)
1589 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1591 verify(import->poolname == NULL || import->guid == 0);
1594 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1596 if (import->cachefile != NULL)
1597 return (zpool_find_import_cached(hdl, import->cachefile,
1598 import->poolname, import->guid));
1600 return (zpool_find_import_impl(hdl, import));
1604 find_guid(nvlist_t *nv, uint64_t guid)
1610 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1614 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1615 &child, &children) == 0) {
1616 for (c = 0; c < children; c++)
1617 if (find_guid(child[c], guid))
1624 typedef struct aux_cbdata {
1625 const char *cb_type;
1627 zpool_handle_t *cb_zhp;
1631 find_aux(zpool_handle_t *zhp, void *data)
1633 aux_cbdata_t *cbp = data;
1639 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1642 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1643 &list, &count) == 0) {
1644 for (i = 0; i < count; i++) {
1645 verify(nvlist_lookup_uint64(list[i],
1646 ZPOOL_CONFIG_GUID, &guid) == 0);
1647 if (guid == cbp->cb_guid) {
1659 * Determines if the pool is in use. If so, it returns true and the state of
1660 * the pool as well as the name of the pool. Both strings are allocated and
1661 * must be freed by the caller.
1664 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1670 uint64_t guid, vdev_guid;
1671 zpool_handle_t *zhp;
1672 nvlist_t *pool_config;
1673 uint64_t stateval, isspare;
1674 aux_cbdata_t cb = { 0 };
1679 if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1680 (void) no_memory(hdl);
1687 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1689 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1692 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1693 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1695 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1700 case POOL_STATE_EXPORTED:
1702 * A pool with an exported state may in fact be imported
1703 * read-only, so check the in-core state to see if it's
1704 * active and imported read-only. If it is, set
1705 * its state to active.
1707 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1708 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1709 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1710 stateval = POOL_STATE_ACTIVE;
1713 * All we needed the zpool handle for is the
1714 * readonly prop check.
1722 case POOL_STATE_ACTIVE:
1724 * For an active pool, we have to determine if it's really part
1725 * of a currently active pool (in which case the pool will exist
1726 * and the guid will be the same), or whether it's part of an
1727 * active pool that was disconnected without being explicitly
1730 if (pool_active(hdl, name, guid, &isactive) != 0) {
1731 nvlist_free(config);
1737 * Because the device may have been removed while
1738 * offlined, we only report it as active if the vdev is
1739 * still present in the config. Otherwise, pretend like
1742 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1743 (pool_config = zpool_get_config(zhp, NULL))
1747 verify(nvlist_lookup_nvlist(pool_config,
1748 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1749 ret = find_guid(nvroot, vdev_guid);
1755 * If this is an active spare within another pool, we
1756 * treat it like an unused hot spare. This allows the
1757 * user to create a pool with a hot spare that currently
1758 * in use within another pool. Since we return B_TRUE,
1759 * libdiskmgt will continue to prevent generic consumers
1760 * from using the device.
1762 if (ret && nvlist_lookup_uint64(config,
1763 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1764 stateval = POOL_STATE_SPARE;
1769 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1774 case POOL_STATE_SPARE:
1776 * For a hot spare, it can be either definitively in use, or
1777 * potentially active. To determine if it's in use, we iterate
1778 * over all pools in the system and search for one with a spare
1779 * with a matching guid.
1781 * Due to the shared nature of spares, we don't actually report
1782 * the potentially active case as in use. This means the user
1783 * can freely create pools on the hot spares of exported pools,
1784 * but to do otherwise makes the resulting code complicated, and
1785 * we end up having to deal with this case anyway.
1788 cb.cb_guid = vdev_guid;
1789 cb.cb_type = ZPOOL_CONFIG_SPARES;
1790 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1791 name = (char *)zpool_get_name(cb.cb_zhp);
1798 case POOL_STATE_L2CACHE:
1801 * Check if any pool is currently using this l2cache device.
1804 cb.cb_guid = vdev_guid;
1805 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1806 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1807 name = (char *)zpool_get_name(cb.cb_zhp);
1820 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1822 zpool_close(cb.cb_zhp);
1823 nvlist_free(config);
1826 *state = (pool_state_t)stateval;
1830 zpool_close(cb.cb_zhp);
1832 nvlist_free(config);