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.
25 * Copyright 2015 RackTop Systems.
29 * Pool import support functions.
31 * To import a pool, we rely on reading the configuration information from the
32 * ZFS label of each device. If we successfully read the label, then we
33 * organize the configuration information in the following hierarchy:
35 * pool guid -> toplevel vdev guid -> label txg
37 * Duplicate entries matching this same tuple will be discarded. Once we have
38 * examined every device, we pick the best label txg config for each toplevel
39 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
40 * update any paths that have changed. Finally, we attempt to import the pool
41 * using our derived config, and record the results.
56 #include <sys/dktp/fdisk.h>
57 #include <sys/efi_partition.h>
59 #include <sys/vdev_impl.h>
61 #include <blkid/blkid.h>
65 #include "libzfs_impl.h"
68 * Intermediate structures used to gather configuration information.
70 typedef struct config_entry {
73 struct config_entry *ce_next;
76 typedef struct vdev_entry {
78 config_entry_t *ve_configs;
79 struct vdev_entry *ve_next;
82 typedef struct pool_entry {
84 vdev_entry_t *pe_vdevs;
85 struct pool_entry *pe_next;
88 typedef struct name_entry {
92 uint64_t ne_num_labels;
93 struct name_entry *ne_next;
96 typedef struct pool_list {
102 get_devid(const char *path)
108 if ((fd = open(path, O_RDONLY)) < 0)
113 if (devid_get(fd, &devid) == 0) {
114 if (devid_get_minor_name(fd, &minor) == 0)
115 ret = devid_str_encode(devid, minor);
117 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;
139 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
140 &child, &children) == 0) {
141 for (c = 0; c < children; c++)
142 if (fix_paths(child[c], names) != 0)
148 * This is a leaf (file or disk) vdev. In either case, go through
149 * the name list and see if we find a matching guid. If so, replace
150 * the path and see if we can calculate a new devid.
152 * There may be multiple names associated with a particular guid, in
153 * which case we have overlapping partitions or multiple paths to the
154 * same disk. In this case we prefer to use the path name which
155 * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we
156 * use the lowest order device which corresponds to the first match
157 * while traversing the ZPOOL_IMPORT_PATH search path.
159 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
160 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
164 for (ne = names; ne != NULL; ne = ne->ne_next) {
165 if (ne->ne_guid == guid) {
172 if ((strlen(path) == strlen(ne->ne_name)) &&
173 strncmp(path, ne->ne_name, strlen(path)) == 0) {
183 /* Prefer paths with move vdev labels. */
184 if (ne->ne_num_labels > best->ne_num_labels) {
189 /* Prefer paths earlier in the search order. */
190 if (best->ne_num_labels == best->ne_num_labels &&
191 ne->ne_order < best->ne_order) {
201 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
204 if ((devid = get_devid(best->ne_name)) == NULL) {
205 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
207 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
208 devid_str_free(devid);
211 devid_str_free(devid);
218 * Add the given configuration to the list of known devices.
221 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
222 int order, int num_labels, nvlist_t *config)
224 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
231 * If this is a hot spare not currently in use or level 2 cache
232 * device, add it to the list of names to translate, but don't do
235 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
237 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
238 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
239 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
242 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
246 ne->ne_guid = vdev_guid;
247 ne->ne_order = order;
248 ne->ne_num_labels = num_labels;
249 ne->ne_next = pl->names;
255 * If we have a valid config but cannot read any of these fields, then
256 * it means we have a half-initialized label. In vdev_label_init()
257 * we write a label with txg == 0 so that we can identify the device
258 * in case the user refers to the same disk later on. If we fail to
259 * create the pool, we'll be left with a label in this state
260 * which should not be considered part of a valid pool.
262 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
264 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
266 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
268 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
269 &txg) != 0 || txg == 0) {
275 * First, see if we know about this pool. If not, then add it to the
276 * list of known pools.
278 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
279 if (pe->pe_guid == pool_guid)
284 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
288 pe->pe_guid = pool_guid;
289 pe->pe_next = pl->pools;
294 * Second, see if we know about this toplevel vdev. Add it if its
297 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
298 if (ve->ve_guid == top_guid)
303 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
307 ve->ve_guid = top_guid;
308 ve->ve_next = pe->pe_vdevs;
313 * Third, see if we have a config with a matching transaction group. If
314 * so, then we do nothing. Otherwise, add it to the list of known
317 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
318 if (ce->ce_txg == txg)
323 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
328 ce->ce_config = config;
329 ce->ce_next = ve->ve_configs;
336 * At this point we've successfully added our config to the list of
337 * known configs. The last thing to do is add the vdev guid -> path
338 * mappings so that we can fix up the configuration as necessary before
341 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
344 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
349 ne->ne_guid = vdev_guid;
350 ne->ne_order = order;
351 ne->ne_num_labels = num_labels;
352 ne->ne_next = pl->names;
359 * Returns true if the named pool matches the given GUID.
362 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
368 if (zpool_open_silent(hdl, name, &zhp) != 0)
376 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
381 *isactive = (theguid == guid);
386 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
389 zfs_cmd_t zc = {"\0"};
392 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
395 if (zcmd_alloc_dst_nvlist(hdl, &zc,
396 zc.zc_nvlist_conf_size * 2) != 0) {
397 zcmd_free_nvlists(&zc);
401 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
402 &zc)) != 0 && errno == ENOMEM) {
403 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
404 zcmd_free_nvlists(&zc);
410 zcmd_free_nvlists(&zc);
414 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
415 zcmd_free_nvlists(&zc);
419 zcmd_free_nvlists(&zc);
424 * Determine if the vdev id is a hole in the namespace.
427 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
431 for (c = 0; c < holes; c++) {
433 /* Top-level is a hole */
434 if (hole_array[c] == id)
441 * Convert our list of pools into the definitive set of configurations. We
442 * start by picking the best config for each toplevel vdev. Once that's done,
443 * we assemble the toplevel vdevs into a full config for the pool. We make a
444 * pass to fix up any incorrect paths, and then add it to the main list to
445 * return to the user.
448 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
453 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
454 nvlist_t **spares, **l2cache;
455 uint_t i, nspares, nl2cache;
456 boolean_t config_seen;
458 char *name, *hostname = NULL;
461 nvlist_t **child = NULL;
463 uint64_t *hole_array, max_id;
468 boolean_t valid_top_config = B_FALSE;
470 if (nvlist_alloc(&ret, 0, 0) != 0)
473 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
474 uint64_t id, max_txg = 0;
476 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
478 config_seen = B_FALSE;
481 * Iterate over all toplevel vdevs. Grab the pool configuration
482 * from the first one we find, and then go through the rest and
483 * add them as necessary to the 'vdevs' member of the config.
485 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
488 * Determine the best configuration for this vdev by
489 * selecting the config with the latest transaction
493 for (ce = ve->ve_configs; ce != NULL;
496 if (ce->ce_txg > best_txg) {
498 best_txg = ce->ce_txg;
503 * We rely on the fact that the max txg for the
504 * pool will contain the most up-to-date information
505 * about the valid top-levels in the vdev namespace.
507 if (best_txg > max_txg) {
508 (void) nvlist_remove(config,
509 ZPOOL_CONFIG_VDEV_CHILDREN,
511 (void) nvlist_remove(config,
512 ZPOOL_CONFIG_HOLE_ARRAY,
513 DATA_TYPE_UINT64_ARRAY);
519 valid_top_config = B_FALSE;
521 if (nvlist_lookup_uint64(tmp,
522 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
523 verify(nvlist_add_uint64(config,
524 ZPOOL_CONFIG_VDEV_CHILDREN,
526 valid_top_config = B_TRUE;
529 if (nvlist_lookup_uint64_array(tmp,
530 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
532 verify(nvlist_add_uint64_array(config,
533 ZPOOL_CONFIG_HOLE_ARRAY,
534 hole_array, holes) == 0);
540 * Copy the relevant pieces of data to the pool
546 * comment (if available)
548 * hostid (if available)
549 * hostname (if available)
551 uint64_t state, version;
552 char *comment = NULL;
554 version = fnvlist_lookup_uint64(tmp,
555 ZPOOL_CONFIG_VERSION);
556 fnvlist_add_uint64(config,
557 ZPOOL_CONFIG_VERSION, version);
558 guid = fnvlist_lookup_uint64(tmp,
559 ZPOOL_CONFIG_POOL_GUID);
560 fnvlist_add_uint64(config,
561 ZPOOL_CONFIG_POOL_GUID, guid);
562 name = fnvlist_lookup_string(tmp,
563 ZPOOL_CONFIG_POOL_NAME);
564 fnvlist_add_string(config,
565 ZPOOL_CONFIG_POOL_NAME, name);
567 if (nvlist_lookup_string(tmp,
568 ZPOOL_CONFIG_COMMENT, &comment) == 0)
569 fnvlist_add_string(config,
570 ZPOOL_CONFIG_COMMENT, comment);
572 state = fnvlist_lookup_uint64(tmp,
573 ZPOOL_CONFIG_POOL_STATE);
574 fnvlist_add_uint64(config,
575 ZPOOL_CONFIG_POOL_STATE, state);
578 if (nvlist_lookup_uint64(tmp,
579 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
580 fnvlist_add_uint64(config,
581 ZPOOL_CONFIG_HOSTID, hostid);
582 hostname = fnvlist_lookup_string(tmp,
583 ZPOOL_CONFIG_HOSTNAME);
584 fnvlist_add_string(config,
585 ZPOOL_CONFIG_HOSTNAME, hostname);
588 config_seen = B_TRUE;
592 * Add this top-level vdev to the child array.
594 verify(nvlist_lookup_nvlist(tmp,
595 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
596 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
599 if (id >= children) {
602 newchild = zfs_alloc(hdl, (id + 1) *
603 sizeof (nvlist_t *));
604 if (newchild == NULL)
607 for (c = 0; c < children; c++)
608 newchild[c] = child[c];
614 if (nvlist_dup(nvtop, &child[id], 0) != 0)
620 * If we have information about all the top-levels then
621 * clean up the nvlist which we've constructed. This
622 * means removing any extraneous devices that are
623 * beyond the valid range or adding devices to the end
624 * of our array which appear to be missing.
626 if (valid_top_config) {
627 if (max_id < children) {
628 for (c = max_id; c < children; c++)
629 nvlist_free(child[c]);
631 } else if (max_id > children) {
634 newchild = zfs_alloc(hdl, (max_id) *
635 sizeof (nvlist_t *));
636 if (newchild == NULL)
639 for (c = 0; c < children; c++)
640 newchild[c] = child[c];
648 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
652 * The vdev namespace may contain holes as a result of
653 * device removal. We must add them back into the vdev
654 * tree before we process any missing devices.
657 ASSERT(valid_top_config);
659 for (c = 0; c < children; c++) {
662 if (child[c] != NULL ||
663 !vdev_is_hole(hole_array, holes, c))
666 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
671 * Holes in the namespace are treated as
672 * "hole" top-level vdevs and have a
673 * special flag set on them.
675 if (nvlist_add_string(holey,
677 VDEV_TYPE_HOLE) != 0 ||
678 nvlist_add_uint64(holey,
679 ZPOOL_CONFIG_ID, c) != 0 ||
680 nvlist_add_uint64(holey,
681 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
690 * Look for any missing top-level vdevs. If this is the case,
691 * create a faked up 'missing' vdev as a placeholder. We cannot
692 * simply compress the child array, because the kernel performs
693 * certain checks to make sure the vdev IDs match their location
694 * in the configuration.
696 for (c = 0; c < children; c++) {
697 if (child[c] == NULL) {
699 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
702 if (nvlist_add_string(missing,
704 VDEV_TYPE_MISSING) != 0 ||
705 nvlist_add_uint64(missing,
706 ZPOOL_CONFIG_ID, c) != 0 ||
707 nvlist_add_uint64(missing,
708 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
709 nvlist_free(missing);
717 * Put all of this pool's top-level vdevs into a root vdev.
719 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
721 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
722 VDEV_TYPE_ROOT) != 0 ||
723 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
724 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
725 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
726 child, children) != 0) {
731 for (c = 0; c < children; c++)
732 nvlist_free(child[c]);
738 * Go through and fix up any paths and/or devids based on our
739 * known list of vdev GUID -> path mappings.
741 if (fix_paths(nvroot, pl->names) != 0) {
747 * Add the root vdev to this pool's configuration.
749 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
757 * zdb uses this path to report on active pools that were
758 * imported or created using -R.
764 * Determine if this pool is currently active, in which case we
765 * can't actually import it.
767 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
769 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
772 if (pool_active(hdl, name, guid, &isactive) != 0)
781 if ((nvl = refresh_config(hdl, config)) == NULL) {
791 * Go through and update the paths for spares, now that we have
794 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
796 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
797 &spares, &nspares) == 0) {
798 for (i = 0; i < nspares; i++) {
799 if (fix_paths(spares[i], pl->names) != 0)
805 * Update the paths for l2cache devices.
807 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
808 &l2cache, &nl2cache) == 0) {
809 for (i = 0; i < nl2cache; i++) {
810 if (fix_paths(l2cache[i], pl->names) != 0)
816 * Restore the original information read from the actual label.
818 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
820 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
823 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
825 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
831 * Add this pool to the list of configs.
833 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
835 if (nvlist_add_nvlist(ret, name, config) != 0)
845 (void) no_memory(hdl);
849 for (c = 0; c < children; c++)
850 nvlist_free(child[c]);
857 * Return the offset of the given label.
860 label_offset(uint64_t size, int l)
862 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
863 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
864 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
868 * Given a file descriptor, read the label information and return an nvlist
869 * describing the configuration, if there is one. The number of valid
870 * labels found will be returned in num_labels when non-NULL.
873 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
875 struct stat64 statbuf;
878 nvlist_t *expected_config = NULL;
879 uint64_t expected_guid = 0, size;
883 if (fstat64_blk(fd, &statbuf) == -1)
885 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
887 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
890 for (l = 0; l < VDEV_LABELS; l++) {
891 uint64_t state, guid, txg;
893 if (pread64(fd, label, sizeof (vdev_label_t),
894 label_offset(size, l)) != sizeof (vdev_label_t))
897 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
898 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
901 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
902 &guid) != 0 || guid == 0) {
903 nvlist_free(*config);
907 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
908 &state) != 0 || state > POOL_STATE_L2CACHE) {
909 nvlist_free(*config);
913 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
914 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
915 &txg) != 0 || txg == 0)) {
916 nvlist_free(*config);
921 if (expected_guid == guid)
924 nvlist_free(*config);
926 expected_config = *config;
927 expected_guid = guid;
932 if (num_labels != NULL)
936 *config = expected_config;
941 typedef struct rdsk_node {
945 libzfs_handle_t *rn_hdl;
949 boolean_t rn_nozpool;
953 slice_cache_compare(const void *arg1, const void *arg2)
955 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
956 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
957 char *nm1slice, *nm2slice;
961 * partitions one and three (slices zero and two) are the most
962 * likely to provide results, so put those first
964 nm1slice = strstr(nm1, "part1");
965 nm2slice = strstr(nm2, "part1");
966 if (nm1slice && !nm2slice) {
969 if (!nm1slice && nm2slice) {
972 nm1slice = strstr(nm1, "part3");
973 nm2slice = strstr(nm2, "part3");
974 if (nm1slice && !nm2slice) {
977 if (!nm1slice && nm2slice) {
981 rv = strcmp(nm1, nm2);
984 return (rv > 0 ? 1 : -1);
989 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
990 diskaddr_t size, uint_t blksz)
994 char sname[MAXNAMELEN];
996 tmpnode.rn_name = &sname[0];
997 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
999 /* too small to contain a zpool? */
1000 if ((size < (SPA_MINDEVSIZE / blksz)) &&
1001 (node = avl_find(r, &tmpnode, NULL)))
1002 node->rn_nozpool = B_TRUE;
1007 nozpool_all_slices(avl_tree_t *r, const char *sname)
1010 char diskname[MAXNAMELEN];
1014 (void) strncpy(diskname, sname, MAXNAMELEN);
1015 if (((ptr = strrchr(diskname, 's')) == NULL) &&
1016 ((ptr = strrchr(diskname, 'p')) == NULL))
1020 for (i = 0; i < NDKMAP; i++)
1021 check_one_slice(r, diskname, i, 0, 1);
1023 for (i = 0; i <= FD_NUMPART; i++)
1024 check_one_slice(r, diskname, i, 0, 1);
1029 check_slices(avl_tree_t *r, int fd, const char *sname)
1032 struct extvtoc vtoc;
1034 char diskname[MAXNAMELEN];
1038 (void) strncpy(diskname, sname, MAXNAMELEN);
1039 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1043 if (read_extvtoc(fd, &vtoc) >= 0) {
1044 for (i = 0; i < NDKMAP; i++)
1045 check_one_slice(r, diskname, i,
1046 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1047 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1049 * on x86 we'll still have leftover links that point
1050 * to slices s[9-15], so use NDKMAP instead
1052 for (i = 0; i < NDKMAP; i++)
1053 check_one_slice(r, diskname, i,
1054 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1055 /* nodes p[1-4] are never used with EFI labels */
1057 for (i = 1; i <= FD_NUMPART; i++)
1058 check_one_slice(r, diskname, i, 0, 1);
1065 zpool_open_func(void *arg)
1067 rdsk_node_t *rn = arg;
1068 struct stat64 statbuf;
1077 * Skip devices with well known prefixes there can be side effects
1078 * when opening devices which need to be avoided.
1080 * core - Symlink to /proc/kcore
1081 * fd* - Floppy interface.
1082 * fuse - Fuse control device.
1083 * hpet - High Precision Event Timer
1084 * lp* - Printer interface.
1085 * parport* - Parallel port interface.
1086 * ppp - Generic PPP driver.
1087 * random - Random device
1088 * rtc - Real Time Clock
1089 * tty* - Generic serial interface.
1090 * urandom - Random device.
1091 * usbmon* - USB IO monitor.
1092 * vcs* - Virtual console memory.
1093 * watchdog - Watchdog must be closed in a special way.
1095 if ((strncmp(rn->rn_name, "core", 4) == 0) ||
1096 (strncmp(rn->rn_name, "fd", 2) == 0) ||
1097 (strncmp(rn->rn_name, "fuse", 4) == 0) ||
1098 (strncmp(rn->rn_name, "hpet", 4) == 0) ||
1099 (strncmp(rn->rn_name, "lp", 2) == 0) ||
1100 (strncmp(rn->rn_name, "parport", 7) == 0) ||
1101 (strncmp(rn->rn_name, "ppp", 3) == 0) ||
1102 (strncmp(rn->rn_name, "random", 6) == 0) ||
1103 (strncmp(rn->rn_name, "rtc", 3) == 0) ||
1104 (strncmp(rn->rn_name, "tty", 3) == 0) ||
1105 (strncmp(rn->rn_name, "urandom", 7) == 0) ||
1106 (strncmp(rn->rn_name, "usbmon", 6) == 0) ||
1107 (strncmp(rn->rn_name, "vcs", 3) == 0) ||
1108 (strncmp(rn->rn_name, "watchdog", 8) == 0))
1112 * Ignore failed stats. We only want regular files and block devices.
1114 if (fstatat64(rn->rn_dfd, rn->rn_name, &statbuf, 0) != 0 ||
1115 (!S_ISREG(statbuf.st_mode) && !S_ISBLK(statbuf.st_mode)))
1118 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1119 /* symlink to a device that's no longer there */
1120 if (errno == ENOENT)
1121 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1125 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1126 /* symlink to a device that's no longer there */
1127 if (errno == ENOENT)
1128 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1132 * Ignore failed stats. We only want regular
1133 * files, character devs and block devs.
1135 if (fstat64(fd, &statbuf) != 0 ||
1136 (!S_ISREG(statbuf.st_mode) &&
1137 !S_ISCHR(statbuf.st_mode) &&
1138 !S_ISBLK(statbuf.st_mode))) {
1143 /* this file is too small to hold a zpool */
1144 if (S_ISREG(statbuf.st_mode) &&
1145 statbuf.st_size < SPA_MINDEVSIZE) {
1148 } else if (!S_ISREG(statbuf.st_mode)) {
1150 * Try to read the disk label first so we don't have to
1151 * open a bunch of minor nodes that can't have a zpool.
1153 check_slices(rn->rn_avl, fd, rn->rn_name);
1156 if ((zpool_read_label(fd, &config, &num_labels)) != 0) {
1158 (void) no_memory(rn->rn_hdl);
1162 if (num_labels == 0) {
1164 nvlist_free(config);
1170 rn->rn_config = config;
1171 rn->rn_num_labels = num_labels;
1175 * Given a file descriptor, clear (zero) the label information. This function
1176 * is used in the appliance stack as part of the ZFS sysevent module and
1177 * to implement the "zpool labelclear" command.
1180 zpool_clear_label(int fd)
1182 struct stat64 statbuf;
1184 vdev_label_t *label;
1187 if (fstat64_blk(fd, &statbuf) == -1)
1189 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1191 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1194 for (l = 0; l < VDEV_LABELS; l++) {
1195 if (pwrite64(fd, label, sizeof (vdev_label_t),
1196 label_offset(size, l)) != sizeof (vdev_label_t)) {
1206 #ifdef HAVE_LIBBLKID
1208 * Use libblkid to quickly search for zfs devices
1211 zpool_find_import_blkid(libzfs_handle_t *hdl, pool_list_t *pools)
1214 blkid_dev_iterate iter;
1216 const char *devname;
1218 int fd, err, num_labels;
1220 err = blkid_get_cache(&cache, NULL);
1222 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1223 dgettext(TEXT_DOMAIN, "blkid_get_cache() %d"), err);
1227 err = blkid_probe_all(cache);
1229 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1230 dgettext(TEXT_DOMAIN, "blkid_probe_all() %d"), err);
1234 iter = blkid_dev_iterate_begin(cache);
1236 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1237 dgettext(TEXT_DOMAIN, "blkid_dev_iterate_begin()"));
1241 err = blkid_dev_set_search(iter, "TYPE", "zfs_member");
1243 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1244 dgettext(TEXT_DOMAIN, "blkid_dev_set_search() %d"), err);
1248 while (blkid_dev_next(iter, &dev) == 0) {
1249 devname = blkid_dev_devname(dev);
1250 if ((fd = open64(devname, O_RDONLY)) < 0)
1253 err = zpool_read_label(fd, &config, &num_labels);
1257 (void) no_memory(hdl);
1261 if (config != NULL) {
1262 err = add_config(hdl, pools, devname, 0,
1263 num_labels, config);
1270 blkid_dev_iterate_end(iter);
1272 blkid_put_cache(cache);
1276 #endif /* HAVE_LIBBLKID */
1279 zpool_default_import_path[DEFAULT_IMPORT_PATH_SIZE] = {
1280 "/dev/disk/by-vdev", /* Custom rules, use first if they exist */
1281 "/dev/mapper", /* Use multipath devices before components */
1282 "/dev/disk/by-uuid", /* Single unique entry and persistent */
1283 "/dev/disk/by-id", /* May be multiple entries and persistent */
1284 "/dev/disk/by-path", /* Encodes physical location and persistent */
1285 "/dev/disk/by-label", /* Custom persistent labels */
1286 "/dev" /* UNSAFE device names will change */
1290 * Given a list of directories to search, find all pools stored on disk. This
1291 * includes partial pools which are not available to import. If no args are
1292 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1293 * poolname or guid (but not both) are provided by the caller when trying
1294 * to import a specific pool.
1297 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1299 int i, dirs = iarg->paths;
1300 struct dirent64 *dp;
1301 char path[MAXPATHLEN];
1302 char *end, **dir = iarg->path;
1304 nvlist_t *ret = NULL;
1305 pool_list_t pools = { 0 };
1306 pool_entry_t *pe, *penext;
1307 vdev_entry_t *ve, *venext;
1308 config_entry_t *ce, *cenext;
1309 name_entry_t *ne, *nenext;
1310 avl_tree_t slice_cache;
1314 verify(iarg->poolname == NULL || iarg->guid == 0);
1317 #ifdef HAVE_LIBBLKID
1318 /* Use libblkid to scan all device for their type */
1319 if (zpool_find_import_blkid(hdl, &pools) == 0)
1322 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1323 dgettext(TEXT_DOMAIN, "blkid failure falling back "
1324 "to manual probing"));
1325 #endif /* HAVE_LIBBLKID */
1327 dir = zpool_default_import_path;
1328 dirs = DEFAULT_IMPORT_PATH_SIZE;
1332 * Go through and read the label configuration information from every
1333 * possible device, organizing the information according to pool GUID
1334 * and toplevel GUID.
1336 for (i = 0; i < dirs; i++) {
1340 boolean_t config_failed = B_FALSE;
1343 /* use realpath to normalize the path */
1344 if (realpath(dir[i], path) == 0) {
1346 /* it is safe to skip missing search paths */
1347 if (errno == ENOENT)
1350 zfs_error_aux(hdl, strerror(errno));
1351 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1352 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1355 end = &path[strlen(path)];
1358 pathleft = &path[sizeof (path)] - end;
1361 * Using raw devices instead of block devices when we're
1362 * reading the labels skips a bunch of slow operations during
1363 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1365 if (strcmp(path, "/dev/dsk/") == 0)
1366 rdsk = "/dev/rdsk/";
1370 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1371 (dirp = fdopendir(dfd)) == NULL) {
1374 zfs_error_aux(hdl, strerror(errno));
1375 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1376 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1381 avl_create(&slice_cache, slice_cache_compare,
1382 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1385 * This is not MT-safe, but we have no MT consumers of libzfs
1387 while ((dp = readdir64(dirp)) != NULL) {
1388 const char *name = dp->d_name;
1389 if (name[0] == '.' &&
1390 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1393 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1394 slice->rn_name = zfs_strdup(hdl, name);
1395 slice->rn_avl = &slice_cache;
1396 slice->rn_dfd = dfd;
1397 slice->rn_hdl = hdl;
1398 slice->rn_nozpool = B_FALSE;
1399 avl_add(&slice_cache, slice);
1402 * create a thread pool to do all of this in parallel;
1403 * rn_nozpool is not protected, so this is racy in that
1404 * multiple tasks could decide that the same slice can
1405 * not hold a zpool, which is benign. Also choose
1406 * double the number of processors; we hold a lot of
1407 * locks in the kernel, so going beyond this doesn't
1411 t = taskq_create("z_import", 2 * boot_ncpus, defclsyspri,
1412 2 * boot_ncpus, INT_MAX, TASKQ_PREPOPULATE);
1413 for (slice = avl_first(&slice_cache); slice;
1414 (slice = avl_walk(&slice_cache, slice,
1416 (void) taskq_dispatch(t, zpool_open_func, slice,
1423 while ((slice = avl_destroy_nodes(&slice_cache,
1424 &cookie)) != NULL) {
1425 if (slice->rn_config != NULL && !config_failed) {
1426 nvlist_t *config = slice->rn_config;
1427 boolean_t matched = B_TRUE;
1429 if (iarg->poolname != NULL) {
1432 matched = nvlist_lookup_string(config,
1433 ZPOOL_CONFIG_POOL_NAME,
1435 strcmp(iarg->poolname, pname) == 0;
1436 } else if (iarg->guid != 0) {
1439 matched = nvlist_lookup_uint64(config,
1440 ZPOOL_CONFIG_POOL_GUID,
1442 iarg->guid == this_guid;
1445 nvlist_free(config);
1448 * use the non-raw path for the config
1450 (void) strlcpy(end, slice->rn_name,
1452 if (add_config(hdl, &pools, path, i+1,
1453 slice->rn_num_labels, config) != 0)
1454 config_failed = B_TRUE;
1457 free(slice->rn_name);
1460 avl_destroy(&slice_cache);
1462 (void) closedir(dirp);
1468 #ifdef HAVE_LIBBLKID
1471 ret = get_configs(hdl, &pools, iarg->can_be_active);
1474 for (pe = pools.pools; pe != NULL; pe = penext) {
1475 penext = pe->pe_next;
1476 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1477 venext = ve->ve_next;
1478 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1479 cenext = ce->ce_next;
1481 nvlist_free(ce->ce_config);
1489 for (ne = pools.names; ne != NULL; ne = nenext) {
1490 nenext = ne->ne_next;
1499 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1501 importargs_t iarg = { 0 };
1506 return (zpool_find_import_impl(hdl, &iarg));
1510 * Given a cache file, return the contents as a list of importable pools.
1511 * poolname or guid (but not both) are provided by the caller when trying
1512 * to import a specific pool.
1515 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1516 char *poolname, uint64_t guid)
1520 struct stat64 statbuf;
1521 nvlist_t *raw, *src, *dst;
1528 verify(poolname == NULL || guid == 0);
1530 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1531 zfs_error_aux(hdl, "%s", strerror(errno));
1532 (void) zfs_error(hdl, EZFS_BADCACHE,
1533 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1537 if (fstat64(fd, &statbuf) != 0) {
1538 zfs_error_aux(hdl, "%s", strerror(errno));
1540 (void) zfs_error(hdl, EZFS_BADCACHE,
1541 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1545 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1550 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1553 (void) zfs_error(hdl, EZFS_BADCACHE,
1554 dgettext(TEXT_DOMAIN,
1555 "failed to read cache file contents"));
1561 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1563 (void) zfs_error(hdl, EZFS_BADCACHE,
1564 dgettext(TEXT_DOMAIN,
1565 "invalid or corrupt cache file contents"));
1572 * Go through and get the current state of the pools and refresh their
1575 if (nvlist_alloc(&pools, 0, 0) != 0) {
1576 (void) no_memory(hdl);
1582 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1583 src = fnvpair_value_nvlist(elem);
1585 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1586 if (poolname != NULL && strcmp(poolname, name) != 0)
1589 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1590 if (guid != 0 && guid != this_guid)
1593 if (pool_active(hdl, name, this_guid, &active) != 0) {
1602 if ((dst = refresh_config(hdl, src)) == NULL) {
1608 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1609 (void) no_memory(hdl);
1623 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1625 importargs_t *import = data;
1628 if (import->poolname != NULL) {
1631 verify(nvlist_lookup_string(zhp->zpool_config,
1632 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1633 if (strcmp(pool_name, import->poolname) == 0)
1638 verify(nvlist_lookup_uint64(zhp->zpool_config,
1639 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1640 if (pool_guid == import->guid)
1649 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1651 verify(import->poolname == NULL || import->guid == 0);
1654 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1656 if (import->cachefile != NULL)
1657 return (zpool_find_import_cached(hdl, import->cachefile,
1658 import->poolname, import->guid));
1660 return (zpool_find_import_impl(hdl, import));
1664 find_guid(nvlist_t *nv, uint64_t guid)
1670 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1674 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1675 &child, &children) == 0) {
1676 for (c = 0; c < children; c++)
1677 if (find_guid(child[c], guid))
1684 typedef struct aux_cbdata {
1685 const char *cb_type;
1687 zpool_handle_t *cb_zhp;
1691 find_aux(zpool_handle_t *zhp, void *data)
1693 aux_cbdata_t *cbp = data;
1699 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1702 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1703 &list, &count) == 0) {
1704 for (i = 0; i < count; i++) {
1705 verify(nvlist_lookup_uint64(list[i],
1706 ZPOOL_CONFIG_GUID, &guid) == 0);
1707 if (guid == cbp->cb_guid) {
1719 * Determines if the pool is in use. If so, it returns true and the state of
1720 * the pool as well as the name of the pool. Both strings are allocated and
1721 * must be freed by the caller.
1724 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1730 uint64_t guid, vdev_guid;
1731 zpool_handle_t *zhp;
1732 nvlist_t *pool_config;
1733 uint64_t stateval, isspare;
1734 aux_cbdata_t cb = { 0 };
1739 if (zpool_read_label(fd, &config, NULL) != 0) {
1740 (void) no_memory(hdl);
1747 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1749 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1752 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1753 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1755 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1760 case POOL_STATE_EXPORTED:
1762 * A pool with an exported state may in fact be imported
1763 * read-only, so check the in-core state to see if it's
1764 * active and imported read-only. If it is, set
1765 * its state to active.
1767 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1768 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1769 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1770 stateval = POOL_STATE_ACTIVE;
1773 * All we needed the zpool handle for is the
1774 * readonly prop check.
1782 case POOL_STATE_ACTIVE:
1784 * For an active pool, we have to determine if it's really part
1785 * of a currently active pool (in which case the pool will exist
1786 * and the guid will be the same), or whether it's part of an
1787 * active pool that was disconnected without being explicitly
1790 if (pool_active(hdl, name, guid, &isactive) != 0) {
1791 nvlist_free(config);
1797 * Because the device may have been removed while
1798 * offlined, we only report it as active if the vdev is
1799 * still present in the config. Otherwise, pretend like
1802 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1803 (pool_config = zpool_get_config(zhp, NULL))
1807 verify(nvlist_lookup_nvlist(pool_config,
1808 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1809 ret = find_guid(nvroot, vdev_guid);
1815 * If this is an active spare within another pool, we
1816 * treat it like an unused hot spare. This allows the
1817 * user to create a pool with a hot spare that currently
1818 * in use within another pool. Since we return B_TRUE,
1819 * libdiskmgt will continue to prevent generic consumers
1820 * from using the device.
1822 if (ret && nvlist_lookup_uint64(config,
1823 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1824 stateval = POOL_STATE_SPARE;
1829 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1834 case POOL_STATE_SPARE:
1836 * For a hot spare, it can be either definitively in use, or
1837 * potentially active. To determine if it's in use, we iterate
1838 * over all pools in the system and search for one with a spare
1839 * with a matching guid.
1841 * Due to the shared nature of spares, we don't actually report
1842 * the potentially active case as in use. This means the user
1843 * can freely create pools on the hot spares of exported pools,
1844 * but to do otherwise makes the resulting code complicated, and
1845 * we end up having to deal with this case anyway.
1848 cb.cb_guid = vdev_guid;
1849 cb.cb_type = ZPOOL_CONFIG_SPARES;
1850 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1851 name = (char *)zpool_get_name(cb.cb_zhp);
1858 case POOL_STATE_L2CACHE:
1861 * Check if any pool is currently using this l2cache device.
1864 cb.cb_guid = vdev_guid;
1865 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1866 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1867 name = (char *)zpool_get_name(cb.cb_zhp);
1880 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1882 zpool_close(cb.cb_zhp);
1883 nvlist_free(config);
1886 *state = (pool_state_t)stateval;
1890 zpool_close(cb.cb_zhp);
1892 nvlist_free(config);