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, 2016 by Delphix. All rights reserved.
25 * Copyright 2015 RackTop Systems.
26 * Copyright (c) 2016, Intel Corporation.
30 * Pool import support functions.
32 * To import a pool, we rely on reading the configuration information from the
33 * ZFS label of each device. If we successfully read the label, then we
34 * organize the configuration information in the following hierarchy:
36 * pool guid -> toplevel vdev guid -> label txg
38 * Duplicate entries matching this same tuple will be discarded. Once we have
39 * examined every device, we pick the best label txg config for each toplevel
40 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
41 * update any paths that have changed. Finally, we attempt to import the pool
42 * using our derived config, and record the results.
61 #include <sys/dktp/fdisk.h>
62 #include <sys/efi_partition.h>
63 #include <sys/vdev_impl.h>
64 #include <blkid/blkid.h>
66 #include "libzfs_impl.h"
70 * Intermediate structures used to gather configuration information.
72 typedef struct config_entry {
75 struct config_entry *ce_next;
78 typedef struct vdev_entry {
80 config_entry_t *ve_configs;
81 struct vdev_entry *ve_next;
84 typedef struct pool_entry {
86 vdev_entry_t *pe_vdevs;
87 struct pool_entry *pe_next;
90 typedef struct name_entry {
94 uint64_t ne_num_labels;
95 struct name_entry *ne_next;
98 typedef struct pool_list {
103 #define DEV_BYID_PATH "/dev/disk/by-id/"
106 * Linux persistent device strings for vdev labels
108 * based on libudev for consistency with libudev disk add/remove events
112 typedef struct vdev_dev_strs {
114 char vds_devphys[128];
118 * Obtain the persistent device id string (describes what)
120 * used by ZED vdev matching for auto-{online,expand,replace}
123 zfs_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
125 struct udev_list_entry *entry;
127 char devbyid[MAXPATHLEN];
129 /* The bus based by-id path is preferred */
130 bus = udev_device_get_property_value(dev, "ID_BUS");
136 * For multipath nodes use the persistent uuid based identifier
138 * Example: /dev/disk/by-id/dm-uuid-mpath-35000c5006304de3f
140 dm_uuid = udev_device_get_property_value(dev, "DM_UUID");
141 if (dm_uuid != NULL) {
142 (void) snprintf(bufptr, buflen, "dm-uuid-%s", dm_uuid);
149 * locate the bus specific by-id link
151 (void) snprintf(devbyid, sizeof (devbyid), "%s%s-", DEV_BYID_PATH, bus);
152 entry = udev_device_get_devlinks_list_entry(dev);
153 while (entry != NULL) {
156 name = udev_list_entry_get_name(entry);
157 if (strncmp(name, devbyid, strlen(devbyid)) == 0) {
158 name += strlen(DEV_BYID_PATH);
159 (void) strlcpy(bufptr, name, buflen);
162 entry = udev_list_entry_get_next(entry);
169 * Obtain the persistent physical location string (describes where)
171 * used by ZED vdev matching for auto-{online,expand,replace}
174 zfs_device_get_physical(struct udev_device *dev, char *bufptr, size_t buflen)
176 const char *physpath = NULL;
179 * Normal disks use ID_PATH for their physical path. Device mapper
180 * devices are virtual and don't have a physical path. For them we
181 * use ID_VDEV instead, which is setup via the /etc/vdev_id.conf file.
182 * ID_VDEV provides a persistent path to a virtual device. If you
183 * don't have vdev_id.conf setup, you cannot use multipath autoreplace.
185 if (!((physpath = udev_device_get_property_value(dev, "ID_PATH")) &&
188 udev_device_get_property_value(dev, "ID_VDEV")) &&
194 (void) strlcpy(bufptr, physpath, buflen);
200 udev_is_mpath(struct udev_device *dev)
202 return udev_device_get_property_value(dev, "DM_UUID") &&
203 udev_device_get_property_value(dev, "MPATH_SBIN_PATH");
207 * A disk is considered a multipath whole disk when:
208 * DEVNAME key value has "dm-"
209 * DM_NAME key value has "mpath" prefix
211 * ID_PART_TABLE_TYPE key does not exist or is not gpt
214 udev_mpath_whole_disk(struct udev_device *dev)
216 const char *devname, *type, *uuid;
218 devname = udev_device_get_property_value(dev, "DEVNAME");
219 type = udev_device_get_property_value(dev, "ID_PART_TABLE_TYPE");
220 uuid = udev_device_get_property_value(dev, "DM_UUID");
222 if ((devname != NULL && strncmp(devname, "/dev/dm-", 8) == 0) &&
223 ((type == NULL) || (strcmp(type, "gpt") != 0)) &&
232 * Check if a disk is effectively a multipath whole disk
235 is_mpath_whole_disk(const char *path)
238 struct udev_device *dev = NULL;
239 char nodepath[MAXPATHLEN];
241 boolean_t wholedisk = B_FALSE;
243 if (realpath(path, nodepath) == NULL)
245 sysname = strrchr(nodepath, '/') + 1;
246 if (strncmp(sysname, "dm-", 3) != 0)
248 if ((udev = udev_new()) == NULL)
250 if ((dev = udev_device_new_from_subsystem_sysname(udev, "block",
252 udev_device_unref(dev);
256 wholedisk = udev_mpath_whole_disk(dev);
258 udev_device_unref(dev);
263 udev_device_is_ready(struct udev_device *dev)
265 #ifdef HAVE_LIBUDEV_UDEV_DEVICE_GET_IS_INITIALIZED
266 return (udev_device_get_is_initialized(dev));
268 /* wait for DEVLINKS property to be initialized */
269 return (udev_device_get_property_value(dev, "DEVLINKS") != NULL);
274 * Wait up to timeout_ms for udev to set up the device node. The device is
275 * considered ready when libudev determines it has been initialized, all of
276 * the device links have been verified to exist, and it has been allowed to
277 * settle. At this point the device the device can be accessed reliably.
278 * Depending on the complexity of the udev rules this process could take
282 zpool_label_disk_wait(char *path, int timeout_ms)
285 struct udev_device *dev = NULL;
286 char nodepath[MAXPATHLEN];
287 char *sysname = NULL;
291 hrtime_t start, settle;
293 if ((udev = udev_new()) == NULL)
300 if (sysname == NULL) {
301 if (realpath(path, nodepath) != NULL) {
302 sysname = strrchr(nodepath, '/') + 1;
304 (void) usleep(sleep_ms * MILLISEC);
309 dev = udev_device_new_from_subsystem_sysname(udev,
311 if ((dev != NULL) && udev_device_is_ready(dev)) {
312 struct udev_list_entry *links, *link;
315 links = udev_device_get_devlinks_list_entry(dev);
317 udev_list_entry_foreach(link, links) {
318 struct stat64 statbuf;
321 name = udev_list_entry_get_name(link);
323 if (stat64(name, &statbuf) == 0 && errno == 0)
333 settle = gethrtime();
334 } else if (NSEC2MSEC(gethrtime() - settle) >=
336 udev_device_unref(dev);
342 udev_device_unref(dev);
343 (void) usleep(sleep_ms * MILLISEC);
345 } while (NSEC2MSEC(gethrtime() - start) < timeout_ms);
354 * Encode the persistent devices strings
355 * used for the vdev disk label
358 encode_device_strings(const char *path, vdev_dev_strs_t *ds,
362 struct udev_device *dev = NULL;
363 char nodepath[MAXPATHLEN];
368 if ((udev = udev_new()) == NULL)
371 /* resolve path to a runtime device node instance */
372 if (realpath(path, nodepath) == NULL)
375 sysname = strrchr(nodepath, '/') + 1;
378 * Wait up to 3 seconds for udev to set up the device node context
382 dev = udev_device_new_from_subsystem_sysname(udev, "block",
386 if (udev_device_is_ready(dev))
387 break; /* udev ready */
389 udev_device_unref(dev);
392 if (NSEC2MSEC(gethrtime() - start) < 10)
393 (void) sched_yield(); /* yield/busy wait up to 10ms */
395 (void) usleep(10 * MILLISEC);
397 } while (NSEC2MSEC(gethrtime() - start) < (3 * MILLISEC));
403 * Only whole disks require extra device strings
405 if (!wholedisk && !udev_mpath_whole_disk(dev))
408 ret = zfs_device_get_devid(dev, ds->vds_devid, sizeof (ds->vds_devid));
412 /* physical location string (optional) */
413 if (zfs_device_get_physical(dev, ds->vds_devphys,
414 sizeof (ds->vds_devphys)) != 0) {
415 ds->vds_devphys[0] = '\0'; /* empty string --> not available */
419 udev_device_unref(dev);
427 * Update a leaf vdev's persistent device strings (Linux only)
429 * - only applies for a dedicated leaf vdev (aka whole disk)
430 * - updated during pool create|add|attach|import
431 * - used for matching device matching during auto-{online,expand,replace}
432 * - stored in a leaf disk config label (i.e. alongside 'path' NVP)
433 * - these strings are currently not used in kernel (i.e. for vdev_disk_open)
435 * single device node example:
436 * devid: 'scsi-MG03SCA300_350000494a8cb3d67-part1'
437 * phys_path: 'pci-0000:04:00.0-sas-0x50000394a8cb3d67-lun-0'
439 * multipath device node example:
440 * devid: 'dm-uuid-mpath-35000c5006304de3f'
442 * We also store the enclosure sysfs path for turning on enclosure LEDs
444 * vdev_enc_sysfs_path: '/sys/class/enclosure/11:0:1:0/SLOT 4'
447 update_vdev_config_dev_strs(nvlist_t *nv)
450 char *env, *type, *path;
451 uint64_t wholedisk = 0;
455 * For the benefit of legacy ZFS implementations, allow
456 * for opting out of devid strings in the vdev label.
459 * env ZFS_VDEV_DEVID_OPT_OUT=YES zpool import dozer
462 * Older ZFS on Linux implementations had issues when attempting to
463 * display pool config VDEV names if a "devid" NVP value is present
464 * in the pool's config.
466 * For example, a pool that originated on illumos platform would
467 * have a devid value in the config and "zpool status" would fail
468 * when listing the config.
470 * A pool can be stripped of any "devid" values on import or
471 * prevented from adding them on zpool create|add by setting
472 * ZFS_VDEV_DEVID_OPT_OUT.
474 env = getenv("ZFS_VDEV_DEVID_OPT_OUT");
475 if (env && (strtoul(env, NULL, 0) > 0 ||
476 !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2))) {
477 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
478 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_PHYS_PATH);
482 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0 ||
483 strcmp(type, VDEV_TYPE_DISK) != 0) {
486 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
491 * Update device string values in config nvlist
493 if (encode_device_strings(path, &vds, (boolean_t)wholedisk) == 0) {
494 (void) nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vds.vds_devid);
495 if (vds.vds_devphys[0] != '\0') {
496 (void) nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
500 /* Add enclosure sysfs path (if disk is in an enclosure) */
501 upath = zfs_get_underlying_path(path);
502 spath = zfs_get_enclosure_sysfs_path(upath);
504 nvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
507 nvlist_remove_all(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
512 /* clear out any stale entries */
513 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
514 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_PHYS_PATH);
515 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
521 is_mpath_whole_disk(const char *path)
527 * Wait up to timeout_ms for udev to set up the device node. The device is
528 * considered ready when the provided path have been verified to exist and
529 * it has been allowed to settle. At this point the device the device can
530 * be accessed reliably. Depending on the complexity of the udev rules thisi
531 * process could take several seconds.
534 zpool_label_disk_wait(char *path, int timeout_ms)
538 hrtime_t start, settle;
539 struct stat64 statbuf;
546 if ((stat64(path, &statbuf) == 0) && (errno == 0)) {
548 settle = gethrtime();
549 else if (NSEC2MSEC(gethrtime() - settle) >= settle_ms)
551 } else if (errno != ENOENT) {
555 usleep(sleep_ms * MILLISEC);
556 } while (NSEC2MSEC(gethrtime() - start) < timeout_ms);
562 update_vdev_config_dev_strs(nvlist_t *nv)
566 #endif /* HAVE_LIBUDEV */
569 * Go through and fix up any path and/or devid information for the given vdev
573 fix_paths(nvlist_t *nv, name_entry_t *names)
578 name_entry_t *ne, *best;
581 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
582 &child, &children) == 0) {
583 for (c = 0; c < children; c++)
584 if (fix_paths(child[c], names) != 0)
590 * This is a leaf (file or disk) vdev. In either case, go through
591 * the name list and see if we find a matching guid. If so, replace
592 * the path and see if we can calculate a new devid.
594 * There may be multiple names associated with a particular guid, in
595 * which case we have overlapping partitions or multiple paths to the
596 * same disk. In this case we prefer to use the path name which
597 * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we
598 * use the lowest order device which corresponds to the first match
599 * while traversing the ZPOOL_IMPORT_PATH search path.
601 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
602 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
606 for (ne = names; ne != NULL; ne = ne->ne_next) {
607 if (ne->ne_guid == guid) {
613 if ((strlen(path) == strlen(ne->ne_name)) &&
614 strncmp(path, ne->ne_name, strlen(path)) == 0) {
624 /* Prefer paths with move vdev labels. */
625 if (ne->ne_num_labels > best->ne_num_labels) {
630 /* Prefer paths earlier in the search order. */
631 if (ne->ne_num_labels == best->ne_num_labels &&
632 ne->ne_order < best->ne_order) {
642 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
645 /* Linux only - update ZPOOL_CONFIG_DEVID and ZPOOL_CONFIG_PHYS_PATH */
646 update_vdev_config_dev_strs(nv);
652 * Add the given configuration to the list of known devices.
655 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
656 int order, int num_labels, nvlist_t *config)
658 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
665 * If this is a hot spare not currently in use or level 2 cache
666 * device, add it to the list of names to translate, but don't do
669 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
671 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
672 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
673 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL) {
678 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
683 ne->ne_guid = vdev_guid;
684 ne->ne_order = order;
685 ne->ne_num_labels = num_labels;
686 ne->ne_next = pl->names;
693 * If we have a valid config but cannot read any of these fields, then
694 * it means we have a half-initialized label. In vdev_label_init()
695 * we write a label with txg == 0 so that we can identify the device
696 * in case the user refers to the same disk later on. If we fail to
697 * create the pool, we'll be left with a label in this state
698 * which should not be considered part of a valid pool.
700 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
702 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
704 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
706 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
707 &txg) != 0 || txg == 0) {
713 * First, see if we know about this pool. If not, then add it to the
714 * list of known pools.
716 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
717 if (pe->pe_guid == pool_guid)
722 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
726 pe->pe_guid = pool_guid;
727 pe->pe_next = pl->pools;
732 * Second, see if we know about this toplevel vdev. Add it if its
735 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
736 if (ve->ve_guid == top_guid)
741 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
745 ve->ve_guid = top_guid;
746 ve->ve_next = pe->pe_vdevs;
751 * Third, see if we have a config with a matching transaction group. If
752 * so, then we do nothing. Otherwise, add it to the list of known
755 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
756 if (ce->ce_txg == txg)
761 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
766 ce->ce_config = config;
767 ce->ce_next = ve->ve_configs;
774 * At this point we've successfully added our config to the list of
775 * known configs. The last thing to do is add the vdev guid -> path
776 * mappings so that we can fix up the configuration as necessary before
779 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
782 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
787 ne->ne_guid = vdev_guid;
788 ne->ne_order = order;
789 ne->ne_num_labels = num_labels;
790 ne->ne_next = pl->names;
797 * Returns true if the named pool matches the given GUID.
800 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
806 if (zpool_open_silent(hdl, name, &zhp) != 0)
814 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
819 *isactive = (theguid == guid);
824 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
827 zfs_cmd_t zc = {"\0"};
828 int err, dstbuf_size;
830 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
833 dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
835 if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
836 zcmd_free_nvlists(&zc);
840 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
841 &zc)) != 0 && errno == ENOMEM) {
842 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
843 zcmd_free_nvlists(&zc);
849 zcmd_free_nvlists(&zc);
853 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
854 zcmd_free_nvlists(&zc);
858 zcmd_free_nvlists(&zc);
863 * Determine if the vdev id is a hole in the namespace.
866 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
870 for (c = 0; c < holes; c++) {
872 /* Top-level is a hole */
873 if (hole_array[c] == id)
880 * Convert our list of pools into the definitive set of configurations. We
881 * start by picking the best config for each toplevel vdev. Once that's done,
882 * we assemble the toplevel vdevs into a full config for the pool. We make a
883 * pass to fix up any incorrect paths, and then add it to the main list to
884 * return to the user.
887 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
892 nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
893 nvlist_t **spares, **l2cache;
894 uint_t i, nspares, nl2cache;
895 boolean_t config_seen;
897 char *name, *hostname = NULL;
900 nvlist_t **child = NULL;
902 uint64_t *hole_array, max_id;
907 boolean_t valid_top_config = B_FALSE;
909 if (nvlist_alloc(&ret, 0, 0) != 0)
912 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
913 uint64_t id, max_txg = 0;
915 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
917 config_seen = B_FALSE;
920 * Iterate over all toplevel vdevs. Grab the pool configuration
921 * from the first one we find, and then go through the rest and
922 * add them as necessary to the 'vdevs' member of the config.
924 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
927 * Determine the best configuration for this vdev by
928 * selecting the config with the latest transaction
932 for (ce = ve->ve_configs; ce != NULL;
935 if (ce->ce_txg > best_txg) {
937 best_txg = ce->ce_txg;
942 * We rely on the fact that the max txg for the
943 * pool will contain the most up-to-date information
944 * about the valid top-levels in the vdev namespace.
946 if (best_txg > max_txg) {
947 (void) nvlist_remove(config,
948 ZPOOL_CONFIG_VDEV_CHILDREN,
950 (void) nvlist_remove(config,
951 ZPOOL_CONFIG_HOLE_ARRAY,
952 DATA_TYPE_UINT64_ARRAY);
958 valid_top_config = B_FALSE;
960 if (nvlist_lookup_uint64(tmp,
961 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
962 verify(nvlist_add_uint64(config,
963 ZPOOL_CONFIG_VDEV_CHILDREN,
965 valid_top_config = B_TRUE;
968 if (nvlist_lookup_uint64_array(tmp,
969 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
971 verify(nvlist_add_uint64_array(config,
972 ZPOOL_CONFIG_HOLE_ARRAY,
973 hole_array, holes) == 0);
979 * Copy the relevant pieces of data to the pool
985 * comment (if available)
987 * hostid (if available)
988 * hostname (if available)
990 uint64_t state, version;
991 char *comment = NULL;
993 version = fnvlist_lookup_uint64(tmp,
994 ZPOOL_CONFIG_VERSION);
995 fnvlist_add_uint64(config,
996 ZPOOL_CONFIG_VERSION, version);
997 guid = fnvlist_lookup_uint64(tmp,
998 ZPOOL_CONFIG_POOL_GUID);
999 fnvlist_add_uint64(config,
1000 ZPOOL_CONFIG_POOL_GUID, guid);
1001 name = fnvlist_lookup_string(tmp,
1002 ZPOOL_CONFIG_POOL_NAME);
1003 fnvlist_add_string(config,
1004 ZPOOL_CONFIG_POOL_NAME, name);
1006 if (nvlist_lookup_string(tmp,
1007 ZPOOL_CONFIG_COMMENT, &comment) == 0)
1008 fnvlist_add_string(config,
1009 ZPOOL_CONFIG_COMMENT, comment);
1011 state = fnvlist_lookup_uint64(tmp,
1012 ZPOOL_CONFIG_POOL_STATE);
1013 fnvlist_add_uint64(config,
1014 ZPOOL_CONFIG_POOL_STATE, state);
1017 if (nvlist_lookup_uint64(tmp,
1018 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1019 fnvlist_add_uint64(config,
1020 ZPOOL_CONFIG_HOSTID, hostid);
1021 hostname = fnvlist_lookup_string(tmp,
1022 ZPOOL_CONFIG_HOSTNAME);
1023 fnvlist_add_string(config,
1024 ZPOOL_CONFIG_HOSTNAME, hostname);
1027 config_seen = B_TRUE;
1031 * Add this top-level vdev to the child array.
1033 verify(nvlist_lookup_nvlist(tmp,
1034 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
1035 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
1038 if (id >= children) {
1039 nvlist_t **newchild;
1041 newchild = zfs_alloc(hdl, (id + 1) *
1042 sizeof (nvlist_t *));
1043 if (newchild == NULL)
1046 for (c = 0; c < children; c++)
1047 newchild[c] = child[c];
1053 if (nvlist_dup(nvtop, &child[id], 0) != 0)
1059 * If we have information about all the top-levels then
1060 * clean up the nvlist which we've constructed. This
1061 * means removing any extraneous devices that are
1062 * beyond the valid range or adding devices to the end
1063 * of our array which appear to be missing.
1065 if (valid_top_config) {
1066 if (max_id < children) {
1067 for (c = max_id; c < children; c++)
1068 nvlist_free(child[c]);
1070 } else if (max_id > children) {
1071 nvlist_t **newchild;
1073 newchild = zfs_alloc(hdl, (max_id) *
1074 sizeof (nvlist_t *));
1075 if (newchild == NULL)
1078 for (c = 0; c < children; c++)
1079 newchild[c] = child[c];
1087 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1091 * The vdev namespace may contain holes as a result of
1092 * device removal. We must add them back into the vdev
1093 * tree before we process any missing devices.
1096 ASSERT(valid_top_config);
1098 for (c = 0; c < children; c++) {
1101 if (child[c] != NULL ||
1102 !vdev_is_hole(hole_array, holes, c))
1105 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
1110 * Holes in the namespace are treated as
1111 * "hole" top-level vdevs and have a
1112 * special flag set on them.
1114 if (nvlist_add_string(holey,
1116 VDEV_TYPE_HOLE) != 0 ||
1117 nvlist_add_uint64(holey,
1118 ZPOOL_CONFIG_ID, c) != 0 ||
1119 nvlist_add_uint64(holey,
1120 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
1129 * Look for any missing top-level vdevs. If this is the case,
1130 * create a faked up 'missing' vdev as a placeholder. We cannot
1131 * simply compress the child array, because the kernel performs
1132 * certain checks to make sure the vdev IDs match their location
1133 * in the configuration.
1135 for (c = 0; c < children; c++) {
1136 if (child[c] == NULL) {
1138 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
1141 if (nvlist_add_string(missing,
1143 VDEV_TYPE_MISSING) != 0 ||
1144 nvlist_add_uint64(missing,
1145 ZPOOL_CONFIG_ID, c) != 0 ||
1146 nvlist_add_uint64(missing,
1147 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
1148 nvlist_free(missing);
1156 * Put all of this pool's top-level vdevs into a root vdev.
1158 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
1160 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1161 VDEV_TYPE_ROOT) != 0 ||
1162 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
1163 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
1164 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1165 child, children) != 0) {
1166 nvlist_free(nvroot);
1170 for (c = 0; c < children; c++)
1171 nvlist_free(child[c]);
1177 * Go through and fix up any paths and/or devids based on our
1178 * known list of vdev GUID -> path mappings.
1180 if (fix_paths(nvroot, pl->names) != 0) {
1181 nvlist_free(nvroot);
1186 * Add the root vdev to this pool's configuration.
1188 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1190 nvlist_free(nvroot);
1193 nvlist_free(nvroot);
1196 * zdb uses this path to report on active pools that were
1197 * imported or created using -R.
1203 * Determine if this pool is currently active, in which case we
1204 * can't actually import it.
1206 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1208 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1211 if (pool_active(hdl, name, guid, &isactive) != 0)
1215 nvlist_free(config);
1220 if ((nvl = refresh_config(hdl, config)) == NULL) {
1221 nvlist_free(config);
1226 nvlist_free(config);
1230 * Go through and update the paths for spares, now that we have
1233 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1235 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1236 &spares, &nspares) == 0) {
1237 for (i = 0; i < nspares; i++) {
1238 if (fix_paths(spares[i], pl->names) != 0)
1244 * Update the paths for l2cache devices.
1246 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1247 &l2cache, &nl2cache) == 0) {
1248 for (i = 0; i < nl2cache; i++) {
1249 if (fix_paths(l2cache[i], pl->names) != 0)
1255 * Restore the original information read from the actual label.
1257 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
1259 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
1262 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
1264 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
1270 * Add this pool to the list of configs.
1272 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1274 if (nvlist_add_nvlist(ret, name, config) != 0)
1277 nvlist_free(config);
1284 (void) no_memory(hdl);
1286 nvlist_free(config);
1288 for (c = 0; c < children; c++)
1289 nvlist_free(child[c]);
1296 * Return the offset of the given label.
1299 label_offset(uint64_t size, int l)
1301 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
1302 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
1303 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
1307 * Given a file descriptor, read the label information and return an nvlist
1308 * describing the configuration, if there is one. The number of valid
1309 * labels found will be returned in num_labels when non-NULL.
1312 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
1314 struct stat64 statbuf;
1316 vdev_label_t *label;
1317 nvlist_t *expected_config = NULL;
1318 uint64_t expected_guid = 0, size;
1323 if (fstat64_blk(fd, &statbuf) == -1)
1325 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1327 error = posix_memalign((void **)&label, PAGESIZE, sizeof (*label));
1331 for (l = 0; l < VDEV_LABELS; l++) {
1332 uint64_t state, guid, txg;
1334 if (pread64(fd, label, sizeof (vdev_label_t),
1335 label_offset(size, l)) != sizeof (vdev_label_t))
1338 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
1339 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
1342 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
1343 &guid) != 0 || guid == 0) {
1344 nvlist_free(*config);
1348 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
1349 &state) != 0 || state > POOL_STATE_L2CACHE) {
1350 nvlist_free(*config);
1354 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1355 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
1356 &txg) != 0 || txg == 0)) {
1357 nvlist_free(*config);
1361 if (expected_guid) {
1362 if (expected_guid == guid)
1365 nvlist_free(*config);
1367 expected_config = *config;
1368 expected_guid = guid;
1373 if (num_labels != NULL)
1374 *num_labels = count;
1377 *config = expected_config;
1382 typedef struct rdsk_node {
1383 char *rn_name; /* Full path to device */
1384 int rn_order; /* Preferred order (low to high) */
1385 int rn_num_labels; /* Number of valid labels */
1386 uint64_t rn_vdev_guid; /* Expected vdev guid when set */
1387 libzfs_handle_t *rn_hdl;
1388 nvlist_t *rn_config; /* Label config */
1392 boolean_t rn_labelpaths;
1396 * Sorted by vdev guid and full path to allow for multiple entries with
1397 * the same full path name. This is required because it's possible to
1398 * have multiple block devices with labels that refer to the same
1399 * ZPOOL_CONFIG_PATH yet have different vdev guids. In this case both
1400 * entries need to be added to the cache. Scenarios where this can occur
1401 * include overwritten pool labels, devices which are visible from multiple
1402 * hosts and multipath devices.
1405 slice_cache_compare(const void *arg1, const void *arg2)
1407 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1408 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1409 uint64_t guid1 = ((rdsk_node_t *)arg1)->rn_vdev_guid;
1410 uint64_t guid2 = ((rdsk_node_t *)arg2)->rn_vdev_guid;
1413 rv = AVL_CMP(guid1, guid2);
1417 return (AVL_ISIGN(strcmp(nm1, nm2)));
1421 is_watchdog_dev(char *dev)
1423 /* For 'watchdog' dev */
1424 if (strcmp(dev, "watchdog") == 0)
1427 /* For 'watchdog<digit><whatever> */
1428 if (strstr(dev, "watchdog") == dev && isdigit(dev[8]))
1435 label_paths_impl(libzfs_handle_t *hdl, nvlist_t *nvroot, uint64_t pool_guid,
1436 uint64_t vdev_guid, char **path, char **devid)
1444 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1445 &child, &children) == 0) {
1446 for (c = 0; c < children; c++) {
1447 error = label_paths_impl(hdl, child[c],
1448 pool_guid, vdev_guid, path, devid);
1458 error = nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_GUID, &guid);
1459 if ((error != 0) || (guid != vdev_guid))
1462 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_PATH, &val);
1466 error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_DEVID, &val);
1474 * Given a disk label fetch the ZPOOL_CONFIG_PATH and ZPOOL_CONFIG_DEVID
1475 * and store these strings as config_path and devid_path respectively.
1476 * The returned pointers are only valid as long as label remains valid.
1479 label_paths(libzfs_handle_t *hdl, nvlist_t *label, char **path, char **devid)
1488 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1489 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &pool_guid) ||
1490 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &vdev_guid))
1493 return (label_paths_impl(hdl, nvroot, pool_guid, vdev_guid, path,
1498 zpool_open_func(void *arg)
1500 rdsk_node_t *rn = arg;
1501 libzfs_handle_t *hdl = rn->rn_hdl;
1502 struct stat64 statbuf;
1504 char *bname, *dupname;
1505 uint64_t vdev_guid = 0;
1511 * Skip devices with well known prefixes there can be side effects
1512 * when opening devices which need to be avoided.
1514 * hpet - High Precision Event Timer
1515 * watchdog - Watchdog must be closed in a special way.
1517 dupname = zfs_strdup(hdl, rn->rn_name);
1518 bname = basename(dupname);
1519 error = ((strcmp(bname, "hpet") == 0) || is_watchdog_dev(bname));
1525 * Ignore failed stats. We only want regular files and block devices.
1527 if (stat64(rn->rn_name, &statbuf) != 0 ||
1528 (!S_ISREG(statbuf.st_mode) && !S_ISBLK(statbuf.st_mode)))
1532 * Preferentially open using O_DIRECT to bypass the block device
1533 * cache which may be stale for multipath devices. An EINVAL errno
1534 * indicates O_DIRECT is unsupported so fallback to just O_RDONLY.
1536 fd = open(rn->rn_name, O_RDONLY | O_DIRECT);
1537 if ((fd < 0) && (errno == EINVAL))
1538 fd = open(rn->rn_name, O_RDONLY);
1544 * This file is too small to hold a zpool
1546 if (S_ISREG(statbuf.st_mode) && statbuf.st_size < SPA_MINDEVSIZE) {
1551 error = zpool_read_label(fd, &config, &num_labels);
1557 if (num_labels == 0) {
1559 nvlist_free(config);
1564 * Check that the vdev is for the expected guid. Additional entries
1565 * are speculatively added based on the paths stored in the labels.
1566 * Entries with valid paths but incorrect guids must be removed.
1568 error = nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid);
1569 if (error || (rn->rn_vdev_guid && rn->rn_vdev_guid != vdev_guid)) {
1571 nvlist_free(config);
1577 rn->rn_config = config;
1578 rn->rn_num_labels = num_labels;
1581 * Add additional entries for paths described by this label.
1583 if (rn->rn_labelpaths) {
1590 if (label_paths(rn->rn_hdl, rn->rn_config, &path, &devid))
1594 * Allow devlinks to stabilize so all paths are available.
1596 zpool_label_disk_wait(rn->rn_name, DISK_LABEL_WAIT);
1599 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1600 slice->rn_name = zfs_strdup(hdl, path);
1601 slice->rn_vdev_guid = vdev_guid;
1602 slice->rn_avl = rn->rn_avl;
1603 slice->rn_hdl = hdl;
1604 slice->rn_order = IMPORT_ORDER_PREFERRED_1;
1605 slice->rn_labelpaths = B_FALSE;
1606 mutex_enter(rn->rn_lock);
1607 if (avl_find(rn->rn_avl, slice, &where)) {
1608 mutex_exit(rn->rn_lock);
1609 free(slice->rn_name);
1612 avl_insert(rn->rn_avl, slice, where);
1613 mutex_exit(rn->rn_lock);
1614 zpool_open_func(slice);
1618 if (devid != NULL) {
1619 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1620 error = asprintf(&slice->rn_name, "%s%s",
1621 DEV_BYID_PATH, devid);
1627 slice->rn_vdev_guid = vdev_guid;
1628 slice->rn_avl = rn->rn_avl;
1629 slice->rn_hdl = hdl;
1630 slice->rn_order = IMPORT_ORDER_PREFERRED_2;
1631 slice->rn_labelpaths = B_FALSE;
1632 mutex_enter(rn->rn_lock);
1633 if (avl_find(rn->rn_avl, slice, &where)) {
1634 mutex_exit(rn->rn_lock);
1635 free(slice->rn_name);
1638 avl_insert(rn->rn_avl, slice, where);
1639 mutex_exit(rn->rn_lock);
1640 zpool_open_func(slice);
1647 * Given a file descriptor, clear (zero) the label information. This function
1648 * is used in the appliance stack as part of the ZFS sysevent module and
1649 * to implement the "zpool labelclear" command.
1652 zpool_clear_label(int fd)
1654 struct stat64 statbuf;
1656 vdev_label_t *label;
1659 if (fstat64_blk(fd, &statbuf) == -1)
1661 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1663 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1666 for (l = 0; l < VDEV_LABELS; l++) {
1667 if (pwrite64(fd, label, sizeof (vdev_label_t),
1668 label_offset(size, l)) != sizeof (vdev_label_t)) {
1679 * Scan a list of directories for zfs devices.
1682 zpool_find_import_scan(libzfs_handle_t *hdl, kmutex_t *lock,
1683 avl_tree_t **slice_cache, char **dir, int dirs)
1690 *slice_cache = NULL;
1691 cache = zfs_alloc(hdl, sizeof (avl_tree_t));
1692 avl_create(cache, slice_cache_compare, sizeof (rdsk_node_t),
1693 offsetof(rdsk_node_t, rn_node));
1695 for (i = 0; i < dirs; i++) {
1696 char path[MAXPATHLEN];
1697 struct dirent64 *dp;
1700 if (realpath(dir[i], path) == NULL) {
1702 if (error == ENOENT)
1705 zfs_error_aux(hdl, strerror(error));
1706 (void) zfs_error_fmt(hdl, EZFS_BADPATH, dgettext(
1707 TEXT_DOMAIN, "cannot resolve path '%s'"), dir[i]);
1711 dirp = opendir(path);
1714 zfs_error_aux(hdl, strerror(error));
1715 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1716 dgettext(TEXT_DOMAIN, "cannot open '%s'"), path);
1720 while ((dp = readdir64(dirp)) != NULL) {
1721 const char *name = dp->d_name;
1722 if (name[0] == '.' &&
1723 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1726 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1727 error = asprintf(&slice->rn_name, "%s/%s", path, name);
1732 slice->rn_vdev_guid = 0;
1733 slice->rn_lock = lock;
1734 slice->rn_avl = cache;
1735 slice->rn_hdl = hdl;
1736 slice->rn_order = i + IMPORT_ORDER_SCAN_OFFSET;
1737 slice->rn_labelpaths = B_FALSE;
1739 avl_add(cache, slice);
1743 (void) closedir(dirp);
1746 *slice_cache = cache;
1751 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1752 free(slice->rn_name);
1761 * Use libblkid to quickly enumerate all known zfs devices.
1764 zpool_find_import_blkid(libzfs_handle_t *hdl, kmutex_t *lock,
1765 avl_tree_t **slice_cache)
1769 blkid_dev_iterate iter;
1774 *slice_cache = NULL;
1776 error = blkid_get_cache(&cache, NULL);
1780 error = blkid_probe_all_new(cache);
1782 blkid_put_cache(cache);
1786 iter = blkid_dev_iterate_begin(cache);
1788 blkid_put_cache(cache);
1792 error = blkid_dev_set_search(iter, "TYPE", "zfs_member");
1794 blkid_dev_iterate_end(iter);
1795 blkid_put_cache(cache);
1799 *slice_cache = zfs_alloc(hdl, sizeof (avl_tree_t));
1800 avl_create(*slice_cache, slice_cache_compare, sizeof (rdsk_node_t),
1801 offsetof(rdsk_node_t, rn_node));
1803 while (blkid_dev_next(iter, &dev) == 0) {
1804 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1805 slice->rn_name = zfs_strdup(hdl, blkid_dev_devname(dev));
1806 slice->rn_vdev_guid = 0;
1807 slice->rn_lock = lock;
1808 slice->rn_avl = *slice_cache;
1809 slice->rn_hdl = hdl;
1810 slice->rn_labelpaths = B_TRUE;
1812 error = zfs_path_order(slice->rn_name, &slice->rn_order);
1814 slice->rn_order += IMPORT_ORDER_SCAN_OFFSET;
1816 slice->rn_order = IMPORT_ORDER_DEFAULT;
1819 if (avl_find(*slice_cache, slice, &where)) {
1820 free(slice->rn_name);
1823 avl_insert(*slice_cache, slice, where);
1828 blkid_dev_iterate_end(iter);
1829 blkid_put_cache(cache);
1835 zpool_default_import_path[DEFAULT_IMPORT_PATH_SIZE] = {
1836 "/dev/disk/by-vdev", /* Custom rules, use first if they exist */
1837 "/dev/mapper", /* Use multipath devices before components */
1838 "/dev/disk/by-partlabel", /* Single unique entry set by user */
1839 "/dev/disk/by-partuuid", /* Generated partition uuid */
1840 "/dev/disk/by-label", /* Custom persistent labels */
1841 "/dev/disk/by-uuid", /* Single unique entry and persistent */
1842 "/dev/disk/by-id", /* May be multiple entries and persistent */
1843 "/dev/disk/by-path", /* Encodes physical location and persistent */
1844 "/dev" /* UNSAFE device names will change */
1848 * Given a list of directories to search, find all pools stored on disk. This
1849 * includes partial pools which are not available to import. If no args are
1850 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1851 * poolname or guid (but not both) are provided by the caller when trying
1852 * to import a specific pool.
1855 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1857 nvlist_t *ret = NULL;
1858 pool_list_t pools = { 0 };
1859 pool_entry_t *pe, *penext;
1860 vdev_entry_t *ve, *venext;
1861 config_entry_t *ce, *cenext;
1862 name_entry_t *ne, *nenext;
1869 verify(iarg->poolname == NULL || iarg->guid == 0);
1870 mutex_init(&lock, NULL, MUTEX_DEFAULT, NULL);
1873 * Locate pool member vdevs using libblkid or by directory scanning.
1874 * On success a newly allocated AVL tree which is populated with an
1875 * entry for each discovered vdev will be returned as the cache.
1876 * It's the callers responsibility to consume and destroy this tree.
1878 if (iarg->scan || iarg->paths != 0) {
1879 int dirs = iarg->paths;
1880 char **dir = iarg->path;
1883 dir = zpool_default_import_path;
1884 dirs = DEFAULT_IMPORT_PATH_SIZE;
1887 if (zpool_find_import_scan(hdl, &lock, &cache, dir, dirs) != 0)
1890 if (zpool_find_import_blkid(hdl, &lock, &cache) != 0)
1895 * Create a thread pool to parallelize the process of reading and
1896 * validating labels, a large number of threads can be used due to
1897 * minimal contention.
1899 t = taskq_create("z_import", 2 * boot_ncpus, defclsyspri,
1900 2 * boot_ncpus, INT_MAX, TASKQ_PREPOPULATE);
1902 for (slice = avl_first(cache); slice;
1903 (slice = avl_walk(cache, slice, AVL_AFTER)))
1904 (void) taskq_dispatch(t, zpool_open_func, slice, TQ_SLEEP);
1910 * Process the cache filtering out any entries which are not
1911 * for the specificed pool then adding matching label configs.
1914 while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) {
1915 if (slice->rn_config != NULL) {
1916 nvlist_t *config = slice->rn_config;
1917 boolean_t matched = B_TRUE;
1920 if (iarg->poolname != NULL) {
1923 matched = nvlist_lookup_string(config,
1924 ZPOOL_CONFIG_POOL_NAME, &pname) == 0 &&
1925 strcmp(iarg->poolname, pname) == 0;
1926 } else if (iarg->guid != 0) {
1929 matched = nvlist_lookup_uint64(config,
1930 ZPOOL_CONFIG_POOL_GUID, &this_guid) == 0 &&
1931 iarg->guid == this_guid;
1934 nvlist_free(config);
1937 * Verify all remaining entries can be opened
1938 * exclusively. This will prune all underlying
1939 * multipath devices which otherwise could
1940 * result in the vdev appearing as UNAVAIL.
1942 * Under zdb, this step isn't required and
1943 * would prevent a zdb -e of active pools with
1946 fd = open(slice->rn_name, O_RDONLY | O_EXCL);
1947 if (fd >= 0 || iarg->can_be_active) {
1950 add_config(hdl, &pools,
1951 slice->rn_name, slice->rn_order,
1952 slice->rn_num_labels, config);
1954 nvlist_free(config);
1958 free(slice->rn_name);
1963 mutex_destroy(&lock);
1965 ret = get_configs(hdl, &pools, iarg->can_be_active);
1967 for (pe = pools.pools; pe != NULL; pe = penext) {
1968 penext = pe->pe_next;
1969 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1970 venext = ve->ve_next;
1971 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1972 cenext = ce->ce_next;
1973 nvlist_free(ce->ce_config);
1981 for (ne = pools.names; ne != NULL; ne = nenext) {
1982 nenext = ne->ne_next;
1991 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1993 importargs_t iarg = { 0 };
1998 return (zpool_find_import_impl(hdl, &iarg));
2002 * Given a cache file, return the contents as a list of importable pools.
2003 * poolname or guid (but not both) are provided by the caller when trying
2004 * to import a specific pool.
2007 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
2008 char *poolname, uint64_t guid)
2012 struct stat64 statbuf;
2013 nvlist_t *raw, *src, *dst;
2020 verify(poolname == NULL || guid == 0);
2022 if ((fd = open(cachefile, O_RDONLY)) < 0) {
2023 zfs_error_aux(hdl, "%s", strerror(errno));
2024 (void) zfs_error(hdl, EZFS_BADCACHE,
2025 dgettext(TEXT_DOMAIN, "failed to open cache file"));
2029 if (fstat64(fd, &statbuf) != 0) {
2030 zfs_error_aux(hdl, "%s", strerror(errno));
2032 (void) zfs_error(hdl, EZFS_BADCACHE,
2033 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
2037 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
2042 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
2045 (void) zfs_error(hdl, EZFS_BADCACHE,
2046 dgettext(TEXT_DOMAIN,
2047 "failed to read cache file contents"));
2053 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
2055 (void) zfs_error(hdl, EZFS_BADCACHE,
2056 dgettext(TEXT_DOMAIN,
2057 "invalid or corrupt cache file contents"));
2064 * Go through and get the current state of the pools and refresh their
2067 if (nvlist_alloc(&pools, 0, 0) != 0) {
2068 (void) no_memory(hdl);
2074 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
2075 src = fnvpair_value_nvlist(elem);
2077 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
2078 if (poolname != NULL && strcmp(poolname, name) != 0)
2081 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
2082 if (guid != 0 && guid != this_guid)
2085 if (pool_active(hdl, name, this_guid, &active) != 0) {
2094 if ((dst = refresh_config(hdl, src)) == NULL) {
2100 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
2101 (void) no_memory(hdl);
2115 name_or_guid_exists(zpool_handle_t *zhp, void *data)
2117 importargs_t *import = data;
2120 if (import->poolname != NULL) {
2123 verify(nvlist_lookup_string(zhp->zpool_config,
2124 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
2125 if (strcmp(pool_name, import->poolname) == 0)
2130 verify(nvlist_lookup_uint64(zhp->zpool_config,
2131 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
2132 if (pool_guid == import->guid)
2141 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
2143 verify(import->poolname == NULL || import->guid == 0);
2146 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
2148 if (import->cachefile != NULL)
2149 return (zpool_find_import_cached(hdl, import->cachefile,
2150 import->poolname, import->guid));
2152 return (zpool_find_import_impl(hdl, import));
2156 find_guid(nvlist_t *nv, uint64_t guid)
2162 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
2166 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2167 &child, &children) == 0) {
2168 for (c = 0; c < children; c++)
2169 if (find_guid(child[c], guid))
2176 typedef struct aux_cbdata {
2177 const char *cb_type;
2179 zpool_handle_t *cb_zhp;
2183 find_aux(zpool_handle_t *zhp, void *data)
2185 aux_cbdata_t *cbp = data;
2191 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2194 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
2195 &list, &count) == 0) {
2196 for (i = 0; i < count; i++) {
2197 verify(nvlist_lookup_uint64(list[i],
2198 ZPOOL_CONFIG_GUID, &guid) == 0);
2199 if (guid == cbp->cb_guid) {
2211 * Determines if the pool is in use. If so, it returns true and the state of
2212 * the pool as well as the name of the pool. Both strings are allocated and
2213 * must be freed by the caller.
2216 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
2222 uint64_t guid, vdev_guid;
2223 zpool_handle_t *zhp;
2224 nvlist_t *pool_config;
2225 uint64_t stateval, isspare;
2226 aux_cbdata_t cb = { 0 };
2231 if (zpool_read_label(fd, &config, NULL) != 0) {
2232 (void) no_memory(hdl);
2239 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2241 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
2244 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
2245 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
2247 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2252 case POOL_STATE_EXPORTED:
2254 * A pool with an exported state may in fact be imported
2255 * read-only, so check the in-core state to see if it's
2256 * active and imported read-only. If it is, set
2257 * its state to active.
2259 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
2260 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
2261 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
2262 stateval = POOL_STATE_ACTIVE;
2265 * All we needed the zpool handle for is the
2266 * readonly prop check.
2274 case POOL_STATE_ACTIVE:
2276 * For an active pool, we have to determine if it's really part
2277 * of a currently active pool (in which case the pool will exist
2278 * and the guid will be the same), or whether it's part of an
2279 * active pool that was disconnected without being explicitly
2282 if (pool_active(hdl, name, guid, &isactive) != 0) {
2283 nvlist_free(config);
2289 * Because the device may have been removed while
2290 * offlined, we only report it as active if the vdev is
2291 * still present in the config. Otherwise, pretend like
2294 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
2295 (pool_config = zpool_get_config(zhp, NULL))
2299 verify(nvlist_lookup_nvlist(pool_config,
2300 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2301 ret = find_guid(nvroot, vdev_guid);
2307 * If this is an active spare within another pool, we
2308 * treat it like an unused hot spare. This allows the
2309 * user to create a pool with a hot spare that currently
2310 * in use within another pool. Since we return B_TRUE,
2311 * libdiskmgt will continue to prevent generic consumers
2312 * from using the device.
2314 if (ret && nvlist_lookup_uint64(config,
2315 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
2316 stateval = POOL_STATE_SPARE;
2321 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
2326 case POOL_STATE_SPARE:
2328 * For a hot spare, it can be either definitively in use, or
2329 * potentially active. To determine if it's in use, we iterate
2330 * over all pools in the system and search for one with a spare
2331 * with a matching guid.
2333 * Due to the shared nature of spares, we don't actually report
2334 * the potentially active case as in use. This means the user
2335 * can freely create pools on the hot spares of exported pools,
2336 * but to do otherwise makes the resulting code complicated, and
2337 * we end up having to deal with this case anyway.
2340 cb.cb_guid = vdev_guid;
2341 cb.cb_type = ZPOOL_CONFIG_SPARES;
2342 if (zpool_iter(hdl, find_aux, &cb) == 1) {
2343 name = (char *)zpool_get_name(cb.cb_zhp);
2350 case POOL_STATE_L2CACHE:
2353 * Check if any pool is currently using this l2cache device.
2356 cb.cb_guid = vdev_guid;
2357 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
2358 if (zpool_iter(hdl, find_aux, &cb) == 1) {
2359 name = (char *)zpool_get_name(cb.cb_zhp);
2372 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
2374 zpool_close(cb.cb_zhp);
2375 nvlist_free(config);
2378 *state = (pool_state_t)stateval;
2382 zpool_close(cb.cb_zhp);
2384 nvlist_free(config);