4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * Pool import support functions.
29 * To import a pool, we rely on reading the configuration information from the
30 * ZFS label of each device. If we successfully read the label, then we
31 * organize the configuration information in the following hierarchy:
33 * pool guid -> toplevel vdev guid -> label txg
35 * Duplicate entries matching this same tuple will be discarded. Once we have
36 * examined every device, we pick the best label txg config for each toplevel
37 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
38 * update any paths that have changed. Finally, we attempt to import the pool
39 * using our derived config, and record the results.
54 #include <sys/dktp/fdisk.h>
55 #include <sys/efi_partition.h>
56 #include <thread_pool.h>
58 #include <sys/vdev_impl.h>
61 #include "libzfs_impl.h"
64 * Intermediate structures used to gather configuration information.
66 typedef struct config_entry {
69 struct config_entry *ce_next;
72 typedef struct vdev_entry {
74 config_entry_t *ve_configs;
75 struct vdev_entry *ve_next;
78 typedef struct pool_entry {
80 vdev_entry_t *pe_vdevs;
81 struct pool_entry *pe_next;
84 typedef struct name_entry {
87 struct name_entry *ne_next;
90 typedef struct pool_list {
96 get_devid(const char *path)
102 if ((fd = open(path, O_RDONLY)) < 0)
107 if (devid_get(fd, &devid) == 0) {
108 if (devid_get_minor_name(fd, &minor) == 0)
109 ret = devid_str_encode(devid, minor);
111 devid_str_free(minor);
121 * Go through and fix up any path and/or devid information for the given vdev
125 fix_paths(nvlist_t *nv, name_entry_t *names)
130 name_entry_t *ne, *best;
134 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
135 &child, &children) == 0) {
136 for (c = 0; c < children; c++)
137 if (fix_paths(child[c], names) != 0)
143 * This is a leaf (file or disk) vdev. In either case, go through
144 * the name list and see if we find a matching guid. If so, replace
145 * the path and see if we can calculate a new devid.
147 * There may be multiple names associated with a particular guid, in
148 * which case we have overlapping slices or multiple paths to the same
149 * disk. If this is the case, then we want to pick the path that is
150 * the most similar to the original, where "most similar" is the number
151 * of matching characters starting from the end of the path. This will
152 * preserve slice numbers even if the disks have been reorganized, and
153 * will also catch preferred disk names if multiple paths exist.
155 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
156 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
161 for (ne = names; ne != NULL; ne = ne->ne_next) {
162 if (ne->ne_guid == guid) {
163 const char *src, *dst;
171 src = ne->ne_name + strlen(ne->ne_name) - 1;
172 dst = path + strlen(path) - 1;
173 for (count = 0; src >= ne->ne_name && dst >= path;
174 src--, dst--, count++)
179 * At this point, 'count' is the number of characters
180 * matched from the end.
182 if (count > matched || best == NULL) {
192 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
195 if ((devid = get_devid(best->ne_name)) == NULL) {
196 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
198 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
200 devid_str_free(devid);
207 * Add the given configuration to the list of known devices.
210 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
213 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
220 * If this is a hot spare not currently in use or level 2 cache
221 * device, add it to the list of names to translate, but don't do
224 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
226 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
227 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
228 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
231 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
235 ne->ne_guid = vdev_guid;
236 ne->ne_next = pl->names;
242 * If we have a valid config but cannot read any of these fields, then
243 * it means we have a half-initialized label. In vdev_label_init()
244 * we write a label with txg == 0 so that we can identify the device
245 * in case the user refers to the same disk later on. If we fail to
246 * create the pool, we'll be left with a label in this state
247 * which should not be considered part of a valid pool.
249 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
251 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
253 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
255 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
256 &txg) != 0 || txg == 0) {
262 * First, see if we know about this pool. If not, then add it to the
263 * list of known pools.
265 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
266 if (pe->pe_guid == pool_guid)
271 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
275 pe->pe_guid = pool_guid;
276 pe->pe_next = pl->pools;
281 * Second, see if we know about this toplevel vdev. Add it if its
284 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
285 if (ve->ve_guid == top_guid)
290 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
294 ve->ve_guid = top_guid;
295 ve->ve_next = pe->pe_vdevs;
300 * Third, see if we have a config with a matching transaction group. If
301 * so, then we do nothing. Otherwise, add it to the list of known
304 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
305 if (ce->ce_txg == txg)
310 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
315 ce->ce_config = config;
316 ce->ce_next = ve->ve_configs;
323 * At this point we've successfully added our config to the list of
324 * known configs. The last thing to do is add the vdev guid -> path
325 * mappings so that we can fix up the configuration as necessary before
328 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
331 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
336 ne->ne_guid = vdev_guid;
337 ne->ne_next = pl->names;
344 * Returns true if the named pool matches the given GUID.
347 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
353 if (zpool_open_silent(hdl, name, &zhp) != 0)
361 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
366 *isactive = (theguid == guid);
371 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
374 zfs_cmd_t zc = { 0 };
377 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
380 if (zcmd_alloc_dst_nvlist(hdl, &zc,
381 zc.zc_nvlist_conf_size * 2) != 0) {
382 zcmd_free_nvlists(&zc);
386 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
387 &zc)) != 0 && errno == ENOMEM) {
388 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
389 zcmd_free_nvlists(&zc);
395 zcmd_free_nvlists(&zc);
399 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
400 zcmd_free_nvlists(&zc);
404 zcmd_free_nvlists(&zc);
409 * Determine if the vdev id is a hole in the namespace.
412 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
414 for (int c = 0; c < holes; c++) {
416 /* Top-level is a hole */
417 if (hole_array[c] == id)
424 * Convert our list of pools into the definitive set of configurations. We
425 * start by picking the best config for each toplevel vdev. Once that's done,
426 * we assemble the toplevel vdevs into a full config for the pool. We make a
427 * pass to fix up any incorrect paths, and then add it to the main list to
428 * return to the user.
431 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
436 nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
437 nvlist_t **spares, **l2cache;
438 uint_t i, nspares, nl2cache;
439 boolean_t config_seen;
441 char *name, *hostname;
442 uint64_t version, guid;
444 nvlist_t **child = NULL;
446 uint64_t *hole_array, max_id;
451 boolean_t found_one = B_FALSE;
452 boolean_t valid_top_config = B_FALSE;
454 if (nvlist_alloc(&ret, 0, 0) != 0)
457 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
458 uint64_t id, max_txg = 0;
460 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
462 config_seen = B_FALSE;
465 * Iterate over all toplevel vdevs. Grab the pool configuration
466 * from the first one we find, and then go through the rest and
467 * add them as necessary to the 'vdevs' member of the config.
469 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
472 * Determine the best configuration for this vdev by
473 * selecting the config with the latest transaction
477 for (ce = ve->ve_configs; ce != NULL;
480 if (ce->ce_txg > best_txg) {
482 best_txg = ce->ce_txg;
487 * We rely on the fact that the max txg for the
488 * pool will contain the most up-to-date information
489 * about the valid top-levels in the vdev namespace.
491 if (best_txg > max_txg) {
492 (void) nvlist_remove(config,
493 ZPOOL_CONFIG_VDEV_CHILDREN,
495 (void) nvlist_remove(config,
496 ZPOOL_CONFIG_HOLE_ARRAY,
497 DATA_TYPE_UINT64_ARRAY);
503 valid_top_config = B_FALSE;
505 if (nvlist_lookup_uint64(tmp,
506 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
507 verify(nvlist_add_uint64(config,
508 ZPOOL_CONFIG_VDEV_CHILDREN,
510 valid_top_config = B_TRUE;
513 if (nvlist_lookup_uint64_array(tmp,
514 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
516 verify(nvlist_add_uint64_array(config,
517 ZPOOL_CONFIG_HOLE_ARRAY,
518 hole_array, holes) == 0);
524 * Copy the relevant pieces of data to the pool
531 * hostid (if available)
532 * hostname (if available)
536 verify(nvlist_lookup_uint64(tmp,
537 ZPOOL_CONFIG_VERSION, &version) == 0);
538 if (nvlist_add_uint64(config,
539 ZPOOL_CONFIG_VERSION, version) != 0)
541 verify(nvlist_lookup_uint64(tmp,
542 ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
543 if (nvlist_add_uint64(config,
544 ZPOOL_CONFIG_POOL_GUID, guid) != 0)
546 verify(nvlist_lookup_string(tmp,
547 ZPOOL_CONFIG_POOL_NAME, &name) == 0);
548 if (nvlist_add_string(config,
549 ZPOOL_CONFIG_POOL_NAME, name) != 0)
551 verify(nvlist_lookup_uint64(tmp,
552 ZPOOL_CONFIG_POOL_STATE, &state) == 0);
553 if (nvlist_add_uint64(config,
554 ZPOOL_CONFIG_POOL_STATE, state) != 0)
557 if (nvlist_lookup_uint64(tmp,
558 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
559 if (nvlist_add_uint64(config,
560 ZPOOL_CONFIG_HOSTID, hostid) != 0)
562 verify(nvlist_lookup_string(tmp,
563 ZPOOL_CONFIG_HOSTNAME,
565 if (nvlist_add_string(config,
566 ZPOOL_CONFIG_HOSTNAME,
571 config_seen = B_TRUE;
575 * Add this top-level vdev to the child array.
577 verify(nvlist_lookup_nvlist(tmp,
578 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
579 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
582 if (id >= children) {
585 newchild = zfs_alloc(hdl, (id + 1) *
586 sizeof (nvlist_t *));
587 if (newchild == NULL)
590 for (c = 0; c < children; c++)
591 newchild[c] = child[c];
597 if (nvlist_dup(nvtop, &child[id], 0) != 0)
603 * If we have information about all the top-levels then
604 * clean up the nvlist which we've constructed. This
605 * means removing any extraneous devices that are
606 * beyond the valid range or adding devices to the end
607 * of our array which appear to be missing.
609 if (valid_top_config) {
610 if (max_id < children) {
611 for (c = max_id; c < children; c++)
612 nvlist_free(child[c]);
614 } else if (max_id > children) {
617 newchild = zfs_alloc(hdl, (max_id) *
618 sizeof (nvlist_t *));
619 if (newchild == NULL)
622 for (c = 0; c < children; c++)
623 newchild[c] = child[c];
631 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
635 * The vdev namespace may contain holes as a result of
636 * device removal. We must add them back into the vdev
637 * tree before we process any missing devices.
640 ASSERT(valid_top_config);
642 for (c = 0; c < children; c++) {
645 if (child[c] != NULL ||
646 !vdev_is_hole(hole_array, holes, c))
649 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
654 * Holes in the namespace are treated as
655 * "hole" top-level vdevs and have a
656 * special flag set on them.
658 if (nvlist_add_string(holey,
660 VDEV_TYPE_HOLE) != 0 ||
661 nvlist_add_uint64(holey,
662 ZPOOL_CONFIG_ID, c) != 0 ||
663 nvlist_add_uint64(holey,
664 ZPOOL_CONFIG_GUID, 0ULL) != 0)
671 * Look for any missing top-level vdevs. If this is the case,
672 * create a faked up 'missing' vdev as a placeholder. We cannot
673 * simply compress the child array, because the kernel performs
674 * certain checks to make sure the vdev IDs match their location
675 * in the configuration.
677 for (c = 0; c < children; c++) {
678 if (child[c] == NULL) {
680 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
683 if (nvlist_add_string(missing,
685 VDEV_TYPE_MISSING) != 0 ||
686 nvlist_add_uint64(missing,
687 ZPOOL_CONFIG_ID, c) != 0 ||
688 nvlist_add_uint64(missing,
689 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
690 nvlist_free(missing);
698 * Put all of this pool's top-level vdevs into a root vdev.
700 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
702 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
703 VDEV_TYPE_ROOT) != 0 ||
704 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
705 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
706 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
707 child, children) != 0) {
712 for (c = 0; c < children; c++)
713 nvlist_free(child[c]);
719 * Go through and fix up any paths and/or devids based on our
720 * known list of vdev GUID -> path mappings.
722 if (fix_paths(nvroot, pl->names) != 0) {
728 * Add the root vdev to this pool's configuration.
730 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
738 * zdb uses this path to report on active pools that were
739 * imported or created using -R.
745 * Determine if this pool is currently active, in which case we
746 * can't actually import it.
748 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
750 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
753 if (pool_active(hdl, name, guid, &isactive) != 0)
762 if ((nvl = refresh_config(hdl, config)) == NULL) {
772 * Go through and update the paths for spares, now that we have
775 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
777 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
778 &spares, &nspares) == 0) {
779 for (i = 0; i < nspares; i++) {
780 if (fix_paths(spares[i], pl->names) != 0)
786 * Update the paths for l2cache devices.
788 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
789 &l2cache, &nl2cache) == 0) {
790 for (i = 0; i < nl2cache; i++) {
791 if (fix_paths(l2cache[i], pl->names) != 0)
797 * Restore the original information read from the actual label.
799 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
801 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
804 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
806 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
812 * Add this pool to the list of configs.
814 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
816 if (nvlist_add_nvlist(ret, name, config) != 0)
832 (void) no_memory(hdl);
836 for (c = 0; c < children; c++)
837 nvlist_free(child[c]);
844 * Return the offset of the given label.
847 label_offset(uint64_t size, int l)
849 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
850 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
851 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
855 * Given a file descriptor, read the label information and return an nvlist
856 * describing the configuration, if there is one.
859 zpool_read_label(int fd, nvlist_t **config)
861 struct stat64 statbuf;
864 uint64_t state, txg, size;
868 if (fstat64(fd, &statbuf) == -1)
870 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
872 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
875 for (l = 0; l < VDEV_LABELS; l++) {
876 if (pread64(fd, label, sizeof (vdev_label_t),
877 label_offset(size, l)) != sizeof (vdev_label_t))
880 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
881 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
884 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
885 &state) != 0 || state > POOL_STATE_L2CACHE) {
886 nvlist_free(*config);
890 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
891 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
892 &txg) != 0 || txg == 0)) {
893 nvlist_free(*config);
906 typedef struct rdsk_node {
909 libzfs_handle_t *rn_hdl;
913 boolean_t rn_nozpool;
917 slice_cache_compare(const void *arg1, const void *arg2)
919 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
920 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
921 char *nm1slice, *nm2slice;
925 * slices zero and two are the most likely to provide results,
928 nm1slice = strstr(nm1, "s0");
929 nm2slice = strstr(nm2, "s0");
930 if (nm1slice && !nm2slice) {
933 if (!nm1slice && nm2slice) {
936 nm1slice = strstr(nm1, "s2");
937 nm2slice = strstr(nm2, "s2");
938 if (nm1slice && !nm2slice) {
941 if (!nm1slice && nm2slice) {
945 rv = strcmp(nm1, nm2);
948 return (rv > 0 ? 1 : -1);
952 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
953 diskaddr_t size, uint_t blksz)
957 char sname[MAXNAMELEN];
959 tmpnode.rn_name = &sname[0];
960 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
963 * protect against division by zero for disk labels that
964 * contain a bogus sector size
968 /* too small to contain a zpool? */
969 if ((size < (SPA_MINDEVSIZE / blksz)) &&
970 (node = avl_find(r, &tmpnode, NULL)))
971 node->rn_nozpool = B_TRUE;
975 nozpool_all_slices(avl_tree_t *r, const char *sname)
977 char diskname[MAXNAMELEN];
981 (void) strncpy(diskname, sname, MAXNAMELEN);
982 if (((ptr = strrchr(diskname, 's')) == NULL) &&
983 ((ptr = strrchr(diskname, 'p')) == NULL))
987 for (i = 0; i < NDKMAP; i++)
988 check_one_slice(r, diskname, i, 0, 1);
990 for (i = 0; i <= FD_NUMPART; i++)
991 check_one_slice(r, diskname, i, 0, 1);
995 check_slices(avl_tree_t *r, int fd, const char *sname)
999 char diskname[MAXNAMELEN];
1003 (void) strncpy(diskname, sname, MAXNAMELEN);
1004 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1008 if (read_extvtoc(fd, &vtoc) >= 0) {
1009 for (i = 0; i < NDKMAP; i++)
1010 check_one_slice(r, diskname, i,
1011 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1012 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1014 * on x86 we'll still have leftover links that point
1015 * to slices s[9-15], so use NDKMAP instead
1017 for (i = 0; i < NDKMAP; i++)
1018 check_one_slice(r, diskname, i,
1019 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1020 /* nodes p[1-4] are never used with EFI labels */
1022 for (i = 1; i <= FD_NUMPART; i++)
1023 check_one_slice(r, diskname, i, 0, 1);
1029 zpool_open_func(void *arg)
1031 rdsk_node_t *rn = arg;
1032 struct stat64 statbuf;
1038 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1039 /* symlink to a device that's no longer there */
1040 if (errno == ENOENT)
1041 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1045 * Ignore failed stats. We only want regular
1046 * files, character devs and block devs.
1048 if (fstat64(fd, &statbuf) != 0 ||
1049 (!S_ISREG(statbuf.st_mode) &&
1050 !S_ISCHR(statbuf.st_mode) &&
1051 !S_ISBLK(statbuf.st_mode))) {
1055 /* this file is too small to hold a zpool */
1056 if (S_ISREG(statbuf.st_mode) &&
1057 statbuf.st_size < SPA_MINDEVSIZE) {
1060 } else if (!S_ISREG(statbuf.st_mode)) {
1062 * Try to read the disk label first so we don't have to
1063 * open a bunch of minor nodes that can't have a zpool.
1065 check_slices(rn->rn_avl, fd, rn->rn_name);
1068 if ((zpool_read_label(fd, &config)) != 0) {
1070 (void) no_memory(rn->rn_hdl);
1076 rn->rn_config = config;
1077 if (config != NULL) {
1078 assert(rn->rn_nozpool == B_FALSE);
1083 * Given a file descriptor, clear (zero) the label information. This function
1084 * is currently only used in the appliance stack as part of the ZFS sysevent
1088 zpool_clear_label(int fd)
1090 struct stat64 statbuf;
1092 vdev_label_t *label;
1095 if (fstat64(fd, &statbuf) == -1)
1097 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1099 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1102 for (l = 0; l < VDEV_LABELS; l++) {
1103 if (pwrite64(fd, label, sizeof (vdev_label_t),
1104 label_offset(size, l)) != sizeof (vdev_label_t))
1113 * Given a list of directories to search, find all pools stored on disk. This
1114 * includes partial pools which are not available to import. If no args are
1115 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1116 * poolname or guid (but not both) are provided by the caller when trying
1117 * to import a specific pool.
1120 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1122 int i, dirs = iarg->paths;
1124 struct dirent64 *dp;
1125 char path[MAXPATHLEN];
1126 char *end, **dir = iarg->path;
1128 nvlist_t *ret = NULL;
1129 static char *default_dir = "/dev/dsk";
1130 pool_list_t pools = { 0 };
1131 pool_entry_t *pe, *penext;
1132 vdev_entry_t *ve, *venext;
1133 config_entry_t *ce, *cenext;
1134 name_entry_t *ne, *nenext;
1135 avl_tree_t slice_cache;
1145 * Go through and read the label configuration information from every
1146 * possible device, organizing the information according to pool GUID
1147 * and toplevel GUID.
1149 for (i = 0; i < dirs; i++) {
1154 /* use realpath to normalize the path */
1155 if (realpath(dir[i], path) == 0) {
1156 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1157 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1160 end = &path[strlen(path)];
1163 pathleft = &path[sizeof (path)] - end;
1166 * Using raw devices instead of block devices when we're
1167 * reading the labels skips a bunch of slow operations during
1168 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1170 if (strcmp(path, "/dev/dsk/") == 0)
1171 rdsk = "/dev/rdsk/";
1175 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1176 (dirp = fdopendir(dfd)) == NULL) {
1177 zfs_error_aux(hdl, strerror(errno));
1178 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1179 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1184 avl_create(&slice_cache, slice_cache_compare,
1185 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1187 * This is not MT-safe, but we have no MT consumers of libzfs
1189 while ((dp = readdir64(dirp)) != NULL) {
1190 const char *name = dp->d_name;
1191 if (name[0] == '.' &&
1192 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1195 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1196 slice->rn_name = zfs_strdup(hdl, name);
1197 slice->rn_avl = &slice_cache;
1198 slice->rn_dfd = dfd;
1199 slice->rn_hdl = hdl;
1200 slice->rn_nozpool = B_FALSE;
1201 avl_add(&slice_cache, slice);
1204 * create a thread pool to do all of this in parallel;
1205 * rn_nozpool is not protected, so this is racy in that
1206 * multiple tasks could decide that the same slice can
1207 * not hold a zpool, which is benign. Also choose
1208 * double the number of processors; we hold a lot of
1209 * locks in the kernel, so going beyond this doesn't
1212 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1214 for (slice = avl_first(&slice_cache); slice;
1215 (slice = avl_walk(&slice_cache, slice,
1217 (void) tpool_dispatch(t, zpool_open_func, slice);
1222 while ((slice = avl_destroy_nodes(&slice_cache,
1223 &cookie)) != NULL) {
1224 if (slice->rn_config != NULL) {
1225 nvlist_t *config = slice->rn_config;
1226 boolean_t matched = B_TRUE;
1228 if (iarg->poolname != NULL) {
1231 matched = nvlist_lookup_string(config,
1232 ZPOOL_CONFIG_POOL_NAME,
1234 strcmp(iarg->poolname, pname) == 0;
1235 } else if (iarg->guid != 0) {
1238 matched = nvlist_lookup_uint64(config,
1239 ZPOOL_CONFIG_POOL_GUID,
1241 iarg->guid == this_guid;
1244 nvlist_free(config);
1248 /* use the non-raw path for the config */
1249 (void) strlcpy(end, slice->rn_name, pathleft);
1250 if (add_config(hdl, &pools, path, config) != 0)
1253 free(slice->rn_name);
1256 avl_destroy(&slice_cache);
1258 (void) closedir(dirp);
1262 ret = get_configs(hdl, &pools, iarg->can_be_active);
1265 for (pe = pools.pools; pe != NULL; pe = penext) {
1266 penext = pe->pe_next;
1267 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1268 venext = ve->ve_next;
1269 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1270 cenext = ce->ce_next;
1272 nvlist_free(ce->ce_config);
1280 for (ne = pools.names; ne != NULL; ne = nenext) {
1281 nenext = ne->ne_next;
1288 (void) closedir(dirp);
1294 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1296 importargs_t iarg = { 0 };
1301 return (zpool_find_import_impl(hdl, &iarg));
1305 * Given a cache file, return the contents as a list of importable pools.
1306 * poolname or guid (but not both) are provided by the caller when trying
1307 * to import a specific pool.
1310 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1311 char *poolname, uint64_t guid)
1315 struct stat64 statbuf;
1316 nvlist_t *raw, *src, *dst;
1323 verify(poolname == NULL || guid == 0);
1325 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1326 zfs_error_aux(hdl, "%s", strerror(errno));
1327 (void) zfs_error(hdl, EZFS_BADCACHE,
1328 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1332 if (fstat64(fd, &statbuf) != 0) {
1333 zfs_error_aux(hdl, "%s", strerror(errno));
1335 (void) zfs_error(hdl, EZFS_BADCACHE,
1336 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1340 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1345 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1348 (void) zfs_error(hdl, EZFS_BADCACHE,
1349 dgettext(TEXT_DOMAIN,
1350 "failed to read cache file contents"));
1356 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1358 (void) zfs_error(hdl, EZFS_BADCACHE,
1359 dgettext(TEXT_DOMAIN,
1360 "invalid or corrupt cache file contents"));
1367 * Go through and get the current state of the pools and refresh their
1370 if (nvlist_alloc(&pools, 0, 0) != 0) {
1371 (void) no_memory(hdl);
1377 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1378 verify(nvpair_value_nvlist(elem, &src) == 0);
1380 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1382 if (poolname != NULL && strcmp(poolname, name) != 0)
1385 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1388 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1390 if (guid != this_guid)
1394 if (pool_active(hdl, name, this_guid, &active) != 0) {
1403 if ((dst = refresh_config(hdl, src)) == NULL) {
1409 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1410 (void) no_memory(hdl);
1424 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1426 importargs_t *import = data;
1429 if (import->poolname != NULL) {
1432 verify(nvlist_lookup_string(zhp->zpool_config,
1433 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1434 if (strcmp(pool_name, import->poolname) == 0)
1439 verify(nvlist_lookup_uint64(zhp->zpool_config,
1440 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1441 if (pool_guid == import->guid)
1450 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1452 verify(import->poolname == NULL || import->guid == 0);
1455 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1457 if (import->cachefile != NULL)
1458 return (zpool_find_import_cached(hdl, import->cachefile,
1459 import->poolname, import->guid));
1461 return (zpool_find_import_impl(hdl, import));
1465 find_guid(nvlist_t *nv, uint64_t guid)
1471 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1475 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1476 &child, &children) == 0) {
1477 for (c = 0; c < children; c++)
1478 if (find_guid(child[c], guid))
1485 typedef struct aux_cbdata {
1486 const char *cb_type;
1488 zpool_handle_t *cb_zhp;
1492 find_aux(zpool_handle_t *zhp, void *data)
1494 aux_cbdata_t *cbp = data;
1500 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1503 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1504 &list, &count) == 0) {
1505 for (i = 0; i < count; i++) {
1506 verify(nvlist_lookup_uint64(list[i],
1507 ZPOOL_CONFIG_GUID, &guid) == 0);
1508 if (guid == cbp->cb_guid) {
1520 * Determines if the pool is in use. If so, it returns true and the state of
1521 * the pool as well as the name of the pool. Both strings are allocated and
1522 * must be freed by the caller.
1525 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1531 uint64_t guid, vdev_guid;
1532 zpool_handle_t *zhp;
1533 nvlist_t *pool_config;
1534 uint64_t stateval, isspare;
1535 aux_cbdata_t cb = { 0 };
1540 if (zpool_read_label(fd, &config) != 0) {
1541 (void) no_memory(hdl);
1548 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1550 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1553 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1554 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1556 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1561 case POOL_STATE_EXPORTED:
1565 case POOL_STATE_ACTIVE:
1567 * For an active pool, we have to determine if it's really part
1568 * of a currently active pool (in which case the pool will exist
1569 * and the guid will be the same), or whether it's part of an
1570 * active pool that was disconnected without being explicitly
1573 if (pool_active(hdl, name, guid, &isactive) != 0) {
1574 nvlist_free(config);
1580 * Because the device may have been removed while
1581 * offlined, we only report it as active if the vdev is
1582 * still present in the config. Otherwise, pretend like
1585 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1586 (pool_config = zpool_get_config(zhp, NULL))
1590 verify(nvlist_lookup_nvlist(pool_config,
1591 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1592 ret = find_guid(nvroot, vdev_guid);
1598 * If this is an active spare within another pool, we
1599 * treat it like an unused hot spare. This allows the
1600 * user to create a pool with a hot spare that currently
1601 * in use within another pool. Since we return B_TRUE,
1602 * libdiskmgt will continue to prevent generic consumers
1603 * from using the device.
1605 if (ret && nvlist_lookup_uint64(config,
1606 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1607 stateval = POOL_STATE_SPARE;
1612 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1617 case POOL_STATE_SPARE:
1619 * For a hot spare, it can be either definitively in use, or
1620 * potentially active. To determine if it's in use, we iterate
1621 * over all pools in the system and search for one with a spare
1622 * with a matching guid.
1624 * Due to the shared nature of spares, we don't actually report
1625 * the potentially active case as in use. This means the user
1626 * can freely create pools on the hot spares of exported pools,
1627 * but to do otherwise makes the resulting code complicated, and
1628 * we end up having to deal with this case anyway.
1631 cb.cb_guid = vdev_guid;
1632 cb.cb_type = ZPOOL_CONFIG_SPARES;
1633 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1634 name = (char *)zpool_get_name(cb.cb_zhp);
1641 case POOL_STATE_L2CACHE:
1644 * Check if any pool is currently using this l2cache device.
1647 cb.cb_guid = vdev_guid;
1648 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1649 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1650 name = (char *)zpool_get_name(cb.cb_zhp);
1663 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1665 zpool_close(cb.cb_zhp);
1666 nvlist_free(config);
1669 *state = (pool_state_t)stateval;
1673 zpool_close(cb.cb_zhp);
1675 nvlist_free(config);