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 2015 Nexenta Systems, Inc. All rights reserved.
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
28 * Pool import support functions.
30 * To import a pool, we rely on reading the configuration information from the
31 * ZFS label of each device. If we successfully read the label, then we
32 * organize the configuration information in the following hierarchy:
34 * pool guid -> toplevel vdev guid -> label txg
36 * Duplicate entries matching this same tuple will be discarded. Once we have
37 * examined every device, we pick the best label txg config for each toplevel
38 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
39 * update any paths that have changed. Finally, we attempt to import the pool
40 * using our derived config, and record the results.
55 #include <sys/dktp/fdisk.h>
56 #include <sys/efi_partition.h>
58 #include <sys/vdev_impl.h>
60 #include <blkid/blkid.h>
64 #include "libzfs_impl.h"
67 * Intermediate structures used to gather configuration information.
69 typedef struct config_entry {
72 struct config_entry *ce_next;
75 typedef struct vdev_entry {
77 config_entry_t *ve_configs;
78 struct vdev_entry *ve_next;
81 typedef struct pool_entry {
83 vdev_entry_t *pe_vdevs;
84 struct pool_entry *pe_next;
87 typedef struct name_entry {
91 uint64_t ne_num_labels;
92 struct name_entry *ne_next;
95 typedef struct pool_list {
101 get_devid(const char *path)
107 if ((fd = open(path, O_RDONLY)) < 0)
112 if (devid_get(fd, &devid) == 0) {
113 if (devid_get_minor_name(fd, &minor) == 0)
114 ret = devid_str_encode(devid, minor);
116 devid_str_free(minor);
126 * Go through and fix up any path and/or devid information for the given vdev
130 fix_paths(nvlist_t *nv, name_entry_t *names)
135 name_entry_t *ne, *best;
138 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
139 &child, &children) == 0) {
140 for (c = 0; c < children; c++)
141 if (fix_paths(child[c], names) != 0)
147 * This is a leaf (file or disk) vdev. In either case, go through
148 * the name list and see if we find a matching guid. If so, replace
149 * the path and see if we can calculate a new devid.
151 * There may be multiple names associated with a particular guid, in
152 * which case we have overlapping partitions or multiple paths to the
153 * same disk. In this case we prefer to use the path name which
154 * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we
155 * use the lowest order device which corresponds to the first match
156 * while traversing the ZPOOL_IMPORT_PATH search path.
158 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
159 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
163 for (ne = names; ne != NULL; ne = ne->ne_next) {
164 if (ne->ne_guid == guid) {
171 if ((strlen(path) == strlen(ne->ne_name)) &&
172 strncmp(path, ne->ne_name, strlen(path)) == 0) {
182 /* Prefer paths with move vdev labels. */
183 if (ne->ne_num_labels > best->ne_num_labels) {
188 /* Prefer paths earlier in the search order. */
189 if (best->ne_num_labels == best->ne_num_labels &&
190 ne->ne_order < best->ne_order) {
200 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
203 if ((devid = get_devid(best->ne_name)) == NULL) {
204 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
206 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
207 devid_str_free(devid);
210 devid_str_free(devid);
217 * Add the given configuration to the list of known devices.
220 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
221 int order, int num_labels, nvlist_t *config)
223 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
230 * If this is a hot spare not currently in use or level 2 cache
231 * device, add it to the list of names to translate, but don't do
234 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
236 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
237 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
238 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
241 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
245 ne->ne_guid = vdev_guid;
246 ne->ne_order = order;
247 ne->ne_num_labels = num_labels;
248 ne->ne_next = pl->names;
254 * If we have a valid config but cannot read any of these fields, then
255 * it means we have a half-initialized label. In vdev_label_init()
256 * we write a label with txg == 0 so that we can identify the device
257 * in case the user refers to the same disk later on. If we fail to
258 * create the pool, we'll be left with a label in this state
259 * which should not be considered part of a valid pool.
261 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
263 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
265 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
267 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
268 &txg) != 0 || txg == 0) {
274 * First, see if we know about this pool. If not, then add it to the
275 * list of known pools.
277 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
278 if (pe->pe_guid == pool_guid)
283 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
287 pe->pe_guid = pool_guid;
288 pe->pe_next = pl->pools;
293 * Second, see if we know about this toplevel vdev. Add it if its
296 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
297 if (ve->ve_guid == top_guid)
302 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
306 ve->ve_guid = top_guid;
307 ve->ve_next = pe->pe_vdevs;
312 * Third, see if we have a config with a matching transaction group. If
313 * so, then we do nothing. Otherwise, add it to the list of known
316 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
317 if (ce->ce_txg == txg)
322 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
327 ce->ce_config = config;
328 ce->ce_next = ve->ve_configs;
335 * At this point we've successfully added our config to the list of
336 * known configs. The last thing to do is add the vdev guid -> path
337 * mappings so that we can fix up the configuration as necessary before
340 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
343 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
348 ne->ne_guid = vdev_guid;
349 ne->ne_order = order;
350 ne->ne_num_labels = num_labels;
351 ne->ne_next = pl->names;
358 * Returns true if the named pool matches the given GUID.
361 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
367 if (zpool_open_silent(hdl, name, &zhp) != 0)
375 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
380 *isactive = (theguid == guid);
385 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
388 zfs_cmd_t zc = {"\0"};
391 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
394 if (zcmd_alloc_dst_nvlist(hdl, &zc,
395 zc.zc_nvlist_conf_size * 2) != 0) {
396 zcmd_free_nvlists(&zc);
400 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
401 &zc)) != 0 && errno == ENOMEM) {
402 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
403 zcmd_free_nvlists(&zc);
409 zcmd_free_nvlists(&zc);
413 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
414 zcmd_free_nvlists(&zc);
418 zcmd_free_nvlists(&zc);
423 * Determine if the vdev id is a hole in the namespace.
426 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
430 for (c = 0; c < holes; c++) {
432 /* Top-level is a hole */
433 if (hole_array[c] == id)
440 * Convert our list of pools into the definitive set of configurations. We
441 * start by picking the best config for each toplevel vdev. Once that's done,
442 * we assemble the toplevel vdevs into a full config for the pool. We make a
443 * pass to fix up any incorrect paths, and then add it to the main list to
444 * return to the user.
447 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
452 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
453 nvlist_t **spares, **l2cache;
454 uint_t i, nspares, nl2cache;
455 boolean_t config_seen;
457 char *name, *hostname = NULL;
460 nvlist_t **child = NULL;
462 uint64_t *hole_array, max_id;
467 boolean_t valid_top_config = B_FALSE;
469 if (nvlist_alloc(&ret, 0, 0) != 0)
472 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
473 uint64_t id, max_txg = 0;
475 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
477 config_seen = B_FALSE;
480 * Iterate over all toplevel vdevs. Grab the pool configuration
481 * from the first one we find, and then go through the rest and
482 * add them as necessary to the 'vdevs' member of the config.
484 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
487 * Determine the best configuration for this vdev by
488 * selecting the config with the latest transaction
492 for (ce = ve->ve_configs; ce != NULL;
495 if (ce->ce_txg > best_txg) {
497 best_txg = ce->ce_txg;
502 * We rely on the fact that the max txg for the
503 * pool will contain the most up-to-date information
504 * about the valid top-levels in the vdev namespace.
506 if (best_txg > max_txg) {
507 (void) nvlist_remove(config,
508 ZPOOL_CONFIG_VDEV_CHILDREN,
510 (void) nvlist_remove(config,
511 ZPOOL_CONFIG_HOLE_ARRAY,
512 DATA_TYPE_UINT64_ARRAY);
518 valid_top_config = B_FALSE;
520 if (nvlist_lookup_uint64(tmp,
521 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
522 verify(nvlist_add_uint64(config,
523 ZPOOL_CONFIG_VDEV_CHILDREN,
525 valid_top_config = B_TRUE;
528 if (nvlist_lookup_uint64_array(tmp,
529 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
531 verify(nvlist_add_uint64_array(config,
532 ZPOOL_CONFIG_HOLE_ARRAY,
533 hole_array, holes) == 0);
539 * Copy the relevant pieces of data to the pool
545 * comment (if available)
547 * hostid (if available)
548 * hostname (if available)
550 uint64_t state, version;
551 char *comment = NULL;
553 version = fnvlist_lookup_uint64(tmp,
554 ZPOOL_CONFIG_VERSION);
555 fnvlist_add_uint64(config,
556 ZPOOL_CONFIG_VERSION, version);
557 guid = fnvlist_lookup_uint64(tmp,
558 ZPOOL_CONFIG_POOL_GUID);
559 fnvlist_add_uint64(config,
560 ZPOOL_CONFIG_POOL_GUID, guid);
561 name = fnvlist_lookup_string(tmp,
562 ZPOOL_CONFIG_POOL_NAME);
563 fnvlist_add_string(config,
564 ZPOOL_CONFIG_POOL_NAME, name);
566 if (nvlist_lookup_string(tmp,
567 ZPOOL_CONFIG_COMMENT, &comment) == 0)
568 fnvlist_add_string(config,
569 ZPOOL_CONFIG_COMMENT, comment);
571 state = fnvlist_lookup_uint64(tmp,
572 ZPOOL_CONFIG_POOL_STATE);
573 fnvlist_add_uint64(config,
574 ZPOOL_CONFIG_POOL_STATE, state);
577 if (nvlist_lookup_uint64(tmp,
578 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
579 fnvlist_add_uint64(config,
580 ZPOOL_CONFIG_HOSTID, hostid);
581 hostname = fnvlist_lookup_string(tmp,
582 ZPOOL_CONFIG_HOSTNAME);
583 fnvlist_add_string(config,
584 ZPOOL_CONFIG_HOSTNAME, hostname);
587 config_seen = B_TRUE;
591 * Add this top-level vdev to the child array.
593 verify(nvlist_lookup_nvlist(tmp,
594 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
595 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
598 if (id >= children) {
601 newchild = zfs_alloc(hdl, (id + 1) *
602 sizeof (nvlist_t *));
603 if (newchild == NULL)
606 for (c = 0; c < children; c++)
607 newchild[c] = child[c];
613 if (nvlist_dup(nvtop, &child[id], 0) != 0)
619 * If we have information about all the top-levels then
620 * clean up the nvlist which we've constructed. This
621 * means removing any extraneous devices that are
622 * beyond the valid range or adding devices to the end
623 * of our array which appear to be missing.
625 if (valid_top_config) {
626 if (max_id < children) {
627 for (c = max_id; c < children; c++)
628 nvlist_free(child[c]);
630 } else if (max_id > children) {
633 newchild = zfs_alloc(hdl, (max_id) *
634 sizeof (nvlist_t *));
635 if (newchild == NULL)
638 for (c = 0; c < children; c++)
639 newchild[c] = child[c];
647 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
651 * The vdev namespace may contain holes as a result of
652 * device removal. We must add them back into the vdev
653 * tree before we process any missing devices.
656 ASSERT(valid_top_config);
658 for (c = 0; c < children; c++) {
661 if (child[c] != NULL ||
662 !vdev_is_hole(hole_array, holes, c))
665 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
670 * Holes in the namespace are treated as
671 * "hole" top-level vdevs and have a
672 * special flag set on them.
674 if (nvlist_add_string(holey,
676 VDEV_TYPE_HOLE) != 0 ||
677 nvlist_add_uint64(holey,
678 ZPOOL_CONFIG_ID, c) != 0 ||
679 nvlist_add_uint64(holey,
680 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
689 * Look for any missing top-level vdevs. If this is the case,
690 * create a faked up 'missing' vdev as a placeholder. We cannot
691 * simply compress the child array, because the kernel performs
692 * certain checks to make sure the vdev IDs match their location
693 * in the configuration.
695 for (c = 0; c < children; c++) {
696 if (child[c] == NULL) {
698 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
701 if (nvlist_add_string(missing,
703 VDEV_TYPE_MISSING) != 0 ||
704 nvlist_add_uint64(missing,
705 ZPOOL_CONFIG_ID, c) != 0 ||
706 nvlist_add_uint64(missing,
707 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
708 nvlist_free(missing);
716 * Put all of this pool's top-level vdevs into a root vdev.
718 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
720 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
721 VDEV_TYPE_ROOT) != 0 ||
722 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
723 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
724 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
725 child, children) != 0) {
730 for (c = 0; c < children; c++)
731 nvlist_free(child[c]);
737 * Go through and fix up any paths and/or devids based on our
738 * known list of vdev GUID -> path mappings.
740 if (fix_paths(nvroot, pl->names) != 0) {
746 * Add the root vdev to this pool's configuration.
748 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
756 * zdb uses this path to report on active pools that were
757 * imported or created using -R.
763 * Determine if this pool is currently active, in which case we
764 * can't actually import it.
766 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
768 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
771 if (pool_active(hdl, name, guid, &isactive) != 0)
780 if ((nvl = refresh_config(hdl, config)) == NULL) {
790 * Go through and update the paths for spares, now that we have
793 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
795 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
796 &spares, &nspares) == 0) {
797 for (i = 0; i < nspares; i++) {
798 if (fix_paths(spares[i], pl->names) != 0)
804 * Update the paths for l2cache devices.
806 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
807 &l2cache, &nl2cache) == 0) {
808 for (i = 0; i < nl2cache; i++) {
809 if (fix_paths(l2cache[i], pl->names) != 0)
815 * Restore the original information read from the actual label.
817 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
819 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
822 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
824 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
830 * Add this pool to the list of configs.
832 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
834 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. The number of valid
869 * labels found will be returned in num_labels when non-NULL.
872 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
874 struct stat64 statbuf;
877 nvlist_t *expected_config = NULL;
878 uint64_t expected_guid = 0, size;
882 if (fstat64_blk(fd, &statbuf) == -1)
884 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
886 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
889 for (l = 0; l < VDEV_LABELS; l++) {
890 uint64_t state, guid, txg;
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_GUID,
901 &guid) != 0 || guid == 0) {
902 nvlist_free(*config);
906 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
907 &state) != 0 || state > POOL_STATE_L2CACHE) {
908 nvlist_free(*config);
912 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
913 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
914 &txg) != 0 || txg == 0)) {
915 nvlist_free(*config);
920 if (expected_guid == guid)
923 nvlist_free(*config);
925 expected_config = *config;
926 expected_guid = guid;
931 if (num_labels != NULL)
935 *config = expected_config;
940 typedef struct rdsk_node {
944 libzfs_handle_t *rn_hdl;
948 boolean_t rn_nozpool;
952 slice_cache_compare(const void *arg1, const void *arg2)
954 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
955 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
956 char *nm1slice, *nm2slice;
960 * partitions one and three (slices zero and two) are the most
961 * likely to provide results, so put those first
963 nm1slice = strstr(nm1, "part1");
964 nm2slice = strstr(nm2, "part1");
965 if (nm1slice && !nm2slice) {
968 if (!nm1slice && nm2slice) {
971 nm1slice = strstr(nm1, "part3");
972 nm2slice = strstr(nm2, "part3");
973 if (nm1slice && !nm2slice) {
976 if (!nm1slice && nm2slice) {
980 rv = strcmp(nm1, nm2);
983 return (rv > 0 ? 1 : -1);
988 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
989 diskaddr_t size, uint_t blksz)
993 char sname[MAXNAMELEN];
995 tmpnode.rn_name = &sname[0];
996 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
998 /* too small to contain a zpool? */
999 if ((size < (SPA_MINDEVSIZE / blksz)) &&
1000 (node = avl_find(r, &tmpnode, NULL)))
1001 node->rn_nozpool = B_TRUE;
1006 nozpool_all_slices(avl_tree_t *r, const char *sname)
1009 char diskname[MAXNAMELEN];
1013 (void) strncpy(diskname, sname, MAXNAMELEN);
1014 if (((ptr = strrchr(diskname, 's')) == NULL) &&
1015 ((ptr = strrchr(diskname, 'p')) == NULL))
1019 for (i = 0; i < NDKMAP; i++)
1020 check_one_slice(r, diskname, i, 0, 1);
1022 for (i = 0; i <= FD_NUMPART; i++)
1023 check_one_slice(r, diskname, i, 0, 1);
1028 check_slices(avl_tree_t *r, int fd, const char *sname)
1031 struct extvtoc vtoc;
1033 char diskname[MAXNAMELEN];
1037 (void) strncpy(diskname, sname, MAXNAMELEN);
1038 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1042 if (read_extvtoc(fd, &vtoc) >= 0) {
1043 for (i = 0; i < NDKMAP; i++)
1044 check_one_slice(r, diskname, i,
1045 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1046 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1048 * on x86 we'll still have leftover links that point
1049 * to slices s[9-15], so use NDKMAP instead
1051 for (i = 0; i < NDKMAP; i++)
1052 check_one_slice(r, diskname, i,
1053 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1054 /* nodes p[1-4] are never used with EFI labels */
1056 for (i = 1; i <= FD_NUMPART; i++)
1057 check_one_slice(r, diskname, i, 0, 1);
1064 zpool_open_func(void *arg)
1066 rdsk_node_t *rn = arg;
1067 struct stat64 statbuf;
1076 * Skip devices with well known prefixes there can be side effects
1077 * when opening devices which need to be avoided.
1079 * core - Symlink to /proc/kcore
1080 * fd* - Floppy interface.
1081 * fuse - Fuse control device.
1082 * hpet - High Precision Event Timer
1083 * lp* - Printer interface.
1084 * parport* - Parallel port interface.
1085 * ppp - Generic PPP driver.
1086 * random - Random device
1087 * rtc - Real Time Clock
1088 * tty* - Generic serial interface.
1089 * urandom - Random device.
1090 * usbmon* - USB IO monitor.
1091 * vcs* - Virtual console memory.
1092 * watchdog - Watchdog must be closed in a special way.
1094 if ((strncmp(rn->rn_name, "core", 4) == 0) ||
1095 (strncmp(rn->rn_name, "fd", 2) == 0) ||
1096 (strncmp(rn->rn_name, "fuse", 4) == 0) ||
1097 (strncmp(rn->rn_name, "hpet", 4) == 0) ||
1098 (strncmp(rn->rn_name, "lp", 2) == 0) ||
1099 (strncmp(rn->rn_name, "parport", 7) == 0) ||
1100 (strncmp(rn->rn_name, "ppp", 3) == 0) ||
1101 (strncmp(rn->rn_name, "random", 6) == 0) ||
1102 (strncmp(rn->rn_name, "rtc", 3) == 0) ||
1103 (strncmp(rn->rn_name, "tty", 3) == 0) ||
1104 (strncmp(rn->rn_name, "urandom", 7) == 0) ||
1105 (strncmp(rn->rn_name, "usbmon", 6) == 0) ||
1106 (strncmp(rn->rn_name, "vcs", 3) == 0) ||
1107 (strncmp(rn->rn_name, "watchdog", 8) == 0))
1111 * Ignore failed stats. We only want regular files and block devices.
1113 if (fstatat64(rn->rn_dfd, rn->rn_name, &statbuf, 0) != 0 ||
1114 (!S_ISREG(statbuf.st_mode) && !S_ISBLK(statbuf.st_mode)))
1117 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1118 /* symlink to a device that's no longer there */
1119 if (errno == ENOENT)
1120 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1124 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1125 /* symlink to a device that's no longer there */
1126 if (errno == ENOENT)
1127 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1131 * Ignore failed stats. We only want regular
1132 * files, character devs and block devs.
1134 if (fstat64(fd, &statbuf) != 0 ||
1135 (!S_ISREG(statbuf.st_mode) &&
1136 !S_ISCHR(statbuf.st_mode) &&
1137 !S_ISBLK(statbuf.st_mode))) {
1142 /* this file is too small to hold a zpool */
1143 if (S_ISREG(statbuf.st_mode) &&
1144 statbuf.st_size < SPA_MINDEVSIZE) {
1147 } else if (!S_ISREG(statbuf.st_mode)) {
1149 * Try to read the disk label first so we don't have to
1150 * open a bunch of minor nodes that can't have a zpool.
1152 check_slices(rn->rn_avl, fd, rn->rn_name);
1155 if ((zpool_read_label(fd, &config, &num_labels)) != 0) {
1157 (void) no_memory(rn->rn_hdl);
1161 if (num_labels == 0) {
1163 nvlist_free(config);
1169 rn->rn_config = config;
1170 rn->rn_num_labels = num_labels;
1171 if (config != NULL) {
1172 assert(rn->rn_nozpool == B_FALSE);
1177 * Given a file descriptor, clear (zero) the label information. This function
1178 * is used in the appliance stack as part of the ZFS sysevent module and
1179 * to implement the "zpool labelclear" command.
1182 zpool_clear_label(int fd)
1184 struct stat64 statbuf;
1186 vdev_label_t *label;
1189 if (fstat64_blk(fd, &statbuf) == -1)
1191 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1193 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1196 for (l = 0; l < VDEV_LABELS; l++) {
1197 if (pwrite64(fd, label, sizeof (vdev_label_t),
1198 label_offset(size, l)) != sizeof (vdev_label_t)) {
1208 #ifdef HAVE_LIBBLKID
1210 * Use libblkid to quickly search for zfs devices
1213 zpool_find_import_blkid(libzfs_handle_t *hdl, pool_list_t *pools)
1216 blkid_dev_iterate iter;
1218 const char *devname;
1220 int fd, err, num_labels;
1222 err = blkid_get_cache(&cache, NULL);
1224 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1225 dgettext(TEXT_DOMAIN, "blkid_get_cache() %d"), err);
1229 err = blkid_probe_all(cache);
1231 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1232 dgettext(TEXT_DOMAIN, "blkid_probe_all() %d"), err);
1236 iter = blkid_dev_iterate_begin(cache);
1238 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1239 dgettext(TEXT_DOMAIN, "blkid_dev_iterate_begin()"));
1243 err = blkid_dev_set_search(iter, "TYPE", "zfs_member");
1245 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1246 dgettext(TEXT_DOMAIN, "blkid_dev_set_search() %d"), err);
1250 while (blkid_dev_next(iter, &dev) == 0) {
1251 devname = blkid_dev_devname(dev);
1252 if ((fd = open64(devname, O_RDONLY)) < 0)
1255 err = zpool_read_label(fd, &config, &num_labels);
1259 (void) no_memory(hdl);
1263 if (config != NULL) {
1264 err = add_config(hdl, pools, devname, 0,
1265 num_labels, config);
1272 blkid_dev_iterate_end(iter);
1274 blkid_put_cache(cache);
1278 #endif /* HAVE_LIBBLKID */
1281 zpool_default_import_path[DEFAULT_IMPORT_PATH_SIZE] = {
1282 "/dev/disk/by-vdev", /* Custom rules, use first if they exist */
1283 "/dev/mapper", /* Use multipath devices before components */
1284 "/dev/disk/by-uuid", /* Single unique entry and persistent */
1285 "/dev/disk/by-id", /* May be multiple entries and persistent */
1286 "/dev/disk/by-path", /* Encodes physical location and persistent */
1287 "/dev/disk/by-label", /* Custom persistent labels */
1288 "/dev" /* UNSAFE device names will change */
1292 * Given a list of directories to search, find all pools stored on disk. This
1293 * includes partial pools which are not available to import. If no args are
1294 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1295 * poolname or guid (but not both) are provided by the caller when trying
1296 * to import a specific pool.
1299 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1301 int i, dirs = iarg->paths;
1302 struct dirent64 *dp;
1303 char path[MAXPATHLEN];
1304 char *end, **dir = iarg->path;
1306 nvlist_t *ret = NULL;
1307 pool_list_t pools = { 0 };
1308 pool_entry_t *pe, *penext;
1309 vdev_entry_t *ve, *venext;
1310 config_entry_t *ce, *cenext;
1311 name_entry_t *ne, *nenext;
1312 avl_tree_t slice_cache;
1316 verify(iarg->poolname == NULL || iarg->guid == 0);
1319 #ifdef HAVE_LIBBLKID
1320 /* Use libblkid to scan all device for their type */
1321 if (zpool_find_import_blkid(hdl, &pools) == 0)
1324 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1325 dgettext(TEXT_DOMAIN, "blkid failure falling back "
1326 "to manual probing"));
1327 #endif /* HAVE_LIBBLKID */
1329 dir = zpool_default_import_path;
1330 dirs = DEFAULT_IMPORT_PATH_SIZE;
1334 * Go through and read the label configuration information from every
1335 * possible device, organizing the information according to pool GUID
1336 * and toplevel GUID.
1338 for (i = 0; i < dirs; i++) {
1342 boolean_t config_failed = B_FALSE;
1345 /* use realpath to normalize the path */
1346 if (realpath(dir[i], path) == 0) {
1348 /* it is safe to skip missing search paths */
1349 if (errno == ENOENT)
1352 zfs_error_aux(hdl, strerror(errno));
1353 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1354 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1357 end = &path[strlen(path)];
1360 pathleft = &path[sizeof (path)] - end;
1363 * Using raw devices instead of block devices when we're
1364 * reading the labels skips a bunch of slow operations during
1365 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1367 if (strcmp(path, "/dev/dsk/") == 0)
1368 rdsk = "/dev/rdsk/";
1372 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1373 (dirp = fdopendir(dfd)) == NULL) {
1376 zfs_error_aux(hdl, strerror(errno));
1377 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1378 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1383 avl_create(&slice_cache, slice_cache_compare,
1384 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1387 * This is not MT-safe, but we have no MT consumers of libzfs
1389 while ((dp = readdir64(dirp)) != NULL) {
1390 const char *name = dp->d_name;
1391 if (name[0] == '.' &&
1392 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1395 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1396 slice->rn_name = zfs_strdup(hdl, name);
1397 slice->rn_avl = &slice_cache;
1398 slice->rn_dfd = dfd;
1399 slice->rn_hdl = hdl;
1400 slice->rn_nozpool = B_FALSE;
1401 avl_add(&slice_cache, slice);
1404 * create a thread pool to do all of this in parallel;
1405 * rn_nozpool is not protected, so this is racy in that
1406 * multiple tasks could decide that the same slice can
1407 * not hold a zpool, which is benign. Also choose
1408 * double the number of processors; we hold a lot of
1409 * locks in the kernel, so going beyond this doesn't
1413 t = taskq_create("z_import", 2 * boot_ncpus, defclsyspri,
1414 2 * boot_ncpus, INT_MAX, TASKQ_PREPOPULATE);
1415 for (slice = avl_first(&slice_cache); slice;
1416 (slice = avl_walk(&slice_cache, slice,
1418 (void) taskq_dispatch(t, zpool_open_func, slice,
1425 while ((slice = avl_destroy_nodes(&slice_cache,
1426 &cookie)) != NULL) {
1427 if (slice->rn_config != NULL && !config_failed) {
1428 nvlist_t *config = slice->rn_config;
1429 boolean_t matched = B_TRUE;
1431 if (iarg->poolname != NULL) {
1434 matched = nvlist_lookup_string(config,
1435 ZPOOL_CONFIG_POOL_NAME,
1437 strcmp(iarg->poolname, pname) == 0;
1438 } else if (iarg->guid != 0) {
1441 matched = nvlist_lookup_uint64(config,
1442 ZPOOL_CONFIG_POOL_GUID,
1444 iarg->guid == this_guid;
1447 nvlist_free(config);
1450 * use the non-raw path for the config
1452 (void) strlcpy(end, slice->rn_name,
1454 if (add_config(hdl, &pools, path, i+1,
1455 slice->rn_num_labels, config) != 0)
1456 config_failed = B_TRUE;
1459 free(slice->rn_name);
1462 avl_destroy(&slice_cache);
1464 (void) closedir(dirp);
1470 #ifdef HAVE_LIBBLKID
1473 ret = get_configs(hdl, &pools, iarg->can_be_active);
1476 for (pe = pools.pools; pe != NULL; pe = penext) {
1477 penext = pe->pe_next;
1478 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1479 venext = ve->ve_next;
1480 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1481 cenext = ce->ce_next;
1483 nvlist_free(ce->ce_config);
1491 for (ne = pools.names; ne != NULL; ne = nenext) {
1492 nenext = ne->ne_next;
1501 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1503 importargs_t iarg = { 0 };
1508 return (zpool_find_import_impl(hdl, &iarg));
1512 * Given a cache file, return the contents as a list of importable pools.
1513 * poolname or guid (but not both) are provided by the caller when trying
1514 * to import a specific pool.
1517 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1518 char *poolname, uint64_t guid)
1522 struct stat64 statbuf;
1523 nvlist_t *raw, *src, *dst;
1530 verify(poolname == NULL || guid == 0);
1532 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1533 zfs_error_aux(hdl, "%s", strerror(errno));
1534 (void) zfs_error(hdl, EZFS_BADCACHE,
1535 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1539 if (fstat64(fd, &statbuf) != 0) {
1540 zfs_error_aux(hdl, "%s", strerror(errno));
1542 (void) zfs_error(hdl, EZFS_BADCACHE,
1543 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1547 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1552 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1555 (void) zfs_error(hdl, EZFS_BADCACHE,
1556 dgettext(TEXT_DOMAIN,
1557 "failed to read cache file contents"));
1563 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1565 (void) zfs_error(hdl, EZFS_BADCACHE,
1566 dgettext(TEXT_DOMAIN,
1567 "invalid or corrupt cache file contents"));
1574 * Go through and get the current state of the pools and refresh their
1577 if (nvlist_alloc(&pools, 0, 0) != 0) {
1578 (void) no_memory(hdl);
1584 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1585 src = fnvpair_value_nvlist(elem);
1587 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1588 if (poolname != NULL && strcmp(poolname, name) != 0)
1591 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1592 if (guid != 0 && guid != this_guid)
1595 if (pool_active(hdl, name, this_guid, &active) != 0) {
1604 if ((dst = refresh_config(hdl, src)) == NULL) {
1610 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1611 (void) no_memory(hdl);
1625 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1627 importargs_t *import = data;
1630 if (import->poolname != NULL) {
1633 verify(nvlist_lookup_string(zhp->zpool_config,
1634 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1635 if (strcmp(pool_name, import->poolname) == 0)
1640 verify(nvlist_lookup_uint64(zhp->zpool_config,
1641 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1642 if (pool_guid == import->guid)
1651 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1653 verify(import->poolname == NULL || import->guid == 0);
1656 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1658 if (import->cachefile != NULL)
1659 return (zpool_find_import_cached(hdl, import->cachefile,
1660 import->poolname, import->guid));
1662 return (zpool_find_import_impl(hdl, import));
1666 find_guid(nvlist_t *nv, uint64_t guid)
1672 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1676 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1677 &child, &children) == 0) {
1678 for (c = 0; c < children; c++)
1679 if (find_guid(child[c], guid))
1686 typedef struct aux_cbdata {
1687 const char *cb_type;
1689 zpool_handle_t *cb_zhp;
1693 find_aux(zpool_handle_t *zhp, void *data)
1695 aux_cbdata_t *cbp = data;
1701 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1704 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1705 &list, &count) == 0) {
1706 for (i = 0; i < count; i++) {
1707 verify(nvlist_lookup_uint64(list[i],
1708 ZPOOL_CONFIG_GUID, &guid) == 0);
1709 if (guid == cbp->cb_guid) {
1721 * Determines if the pool is in use. If so, it returns true and the state of
1722 * the pool as well as the name of the pool. Both strings are allocated and
1723 * must be freed by the caller.
1726 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1732 uint64_t guid, vdev_guid;
1733 zpool_handle_t *zhp;
1734 nvlist_t *pool_config;
1735 uint64_t stateval, isspare;
1736 aux_cbdata_t cb = { 0 };
1741 if (zpool_read_label(fd, &config, NULL) != 0) {
1742 (void) no_memory(hdl);
1749 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1751 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1754 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1755 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1757 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1762 case POOL_STATE_EXPORTED:
1764 * A pool with an exported state may in fact be imported
1765 * read-only, so check the in-core state to see if it's
1766 * active and imported read-only. If it is, set
1767 * its state to active.
1769 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1770 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1771 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1772 stateval = POOL_STATE_ACTIVE;
1775 * All we needed the zpool handle for is the
1776 * readonly prop check.
1784 case POOL_STATE_ACTIVE:
1786 * For an active pool, we have to determine if it's really part
1787 * of a currently active pool (in which case the pool will exist
1788 * and the guid will be the same), or whether it's part of an
1789 * active pool that was disconnected without being explicitly
1792 if (pool_active(hdl, name, guid, &isactive) != 0) {
1793 nvlist_free(config);
1799 * Because the device may have been removed while
1800 * offlined, we only report it as active if the vdev is
1801 * still present in the config. Otherwise, pretend like
1804 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1805 (pool_config = zpool_get_config(zhp, NULL))
1809 verify(nvlist_lookup_nvlist(pool_config,
1810 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1811 ret = find_guid(nvroot, vdev_guid);
1817 * If this is an active spare within another pool, we
1818 * treat it like an unused hot spare. This allows the
1819 * user to create a pool with a hot spare that currently
1820 * in use within another pool. Since we return B_TRUE,
1821 * libdiskmgt will continue to prevent generic consumers
1822 * from using the device.
1824 if (ret && nvlist_lookup_uint64(config,
1825 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1826 stateval = POOL_STATE_SPARE;
1831 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1836 case POOL_STATE_SPARE:
1838 * For a hot spare, it can be either definitively in use, or
1839 * potentially active. To determine if it's in use, we iterate
1840 * over all pools in the system and search for one with a spare
1841 * with a matching guid.
1843 * Due to the shared nature of spares, we don't actually report
1844 * the potentially active case as in use. This means the user
1845 * can freely create pools on the hot spares of exported pools,
1846 * but to do otherwise makes the resulting code complicated, and
1847 * we end up having to deal with this case anyway.
1850 cb.cb_guid = vdev_guid;
1851 cb.cb_type = ZPOOL_CONFIG_SPARES;
1852 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1853 name = (char *)zpool_get_name(cb.cb_zhp);
1860 case POOL_STATE_L2CACHE:
1863 * Check if any pool is currently using this l2cache device.
1866 cb.cb_guid = vdev_guid;
1867 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1868 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1869 name = (char *)zpool_get_name(cb.cb_zhp);
1882 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1884 zpool_close(cb.cb_zhp);
1885 nvlist_free(config);
1888 *state = (pool_state_t)stateval;
1892 zpool_close(cb.cb_zhp);
1894 nvlist_free(config);