4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
27 * Functions to convert between a list of vdevs and an nvlist representing the
28 * configuration. Each entry in the list can be one of:
31 * disk=(path=..., devid=...)
40 * While the underlying implementation supports it, group vdevs cannot contain
41 * other group vdevs. All userland verification of devices is contained within
42 * this file. If successful, the nvlist returned can be passed directly to the
43 * kernel; we've done as much verification as possible in userland.
45 * Hot spares are a special case, and passed down as an array of disk vdevs, at
46 * the same level as the root of the vdev tree.
48 * The only function exported by this file is 'make_root_vdev'. The
49 * function performs several passes:
51 * 1. Construct the vdev specification. Performs syntax validation and
52 * makes sure each device is valid.
53 * 2. Check for devices in use. Using libblkid to make sure that no
54 * devices are also in use. Some can be overridden using the 'force'
55 * flag, others cannot.
56 * 3. Check for replication errors if the 'force' flag is not specified.
57 * validates that the replication level is consistent across the
59 * 4. Call libzfs to label any whole disks with an EFI label.
68 #include <libnvpair.h>
73 #include <sys/efi_partition.h>
76 #include <sys/mntent.h>
77 #include <uuid/uuid.h>
79 #include <blkid/blkid.h>
81 #define blkid_cache void *
82 #endif /* HAVE_LIBBLKID */
84 #include "zpool_util.h"
87 * For any given vdev specification, we can have multiple errors. The
88 * vdev_error() function keeps track of whether we have seen an error yet, and
89 * prints out a header if its the first error we've seen.
96 vdev_error(const char *fmt, ...)
101 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
103 (void) fprintf(stderr, gettext("use '-f' to override "
104 "the following errors:\n"));
106 (void) fprintf(stderr, gettext("the following errors "
107 "must be manually repaired:\n"));
112 (void) vfprintf(stderr, fmt, ap);
117 * Check that a file is valid. All we can do in this case is check that it's
118 * not in use by another pool, and not in use by swap.
121 check_file(const char *file, boolean_t force, boolean_t isspare)
129 if ((fd = open(file, O_RDONLY)) < 0)
132 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
136 case POOL_STATE_ACTIVE:
137 desc = gettext("active");
140 case POOL_STATE_EXPORTED:
141 desc = gettext("exported");
144 case POOL_STATE_POTENTIALLY_ACTIVE:
145 desc = gettext("potentially active");
149 desc = gettext("unknown");
154 * Allow hot spares to be shared between pools.
156 if (state == POOL_STATE_SPARE && isspare)
159 if (state == POOL_STATE_ACTIVE ||
160 state == POOL_STATE_SPARE || !force) {
162 case POOL_STATE_SPARE:
163 vdev_error(gettext("%s is reserved as a hot "
164 "spare for pool %s\n"), file, name);
167 vdev_error(gettext("%s is part of %s pool "
168 "'%s'\n"), file, desc, name);
184 (void) fprintf(stderr, gettext("warning: device in use checking "
185 "failed: %s\n"), strerror(err));
189 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
191 struct stat64 statbuf;
195 #endif /* HAVE_LIBBLKID */
197 if (stat64(path, &statbuf) != 0) {
198 vdev_error(gettext("cannot stat %s: %s\n"),
199 path, strerror(errno));
204 /* No valid type detected device is safe to use */
205 value = blkid_get_tag_value(cache, "TYPE", path);
210 * If libblkid detects a ZFS device, we check the device
211 * using check_file() to see if it's safe. The one safe
212 * case is a spare device shared between multiple pools.
214 if (strcmp(value, "zfs") == 0) {
215 err = check_file(path, force, isspare);
221 vdev_error(gettext("%s contains a filesystem of "
222 "type '%s'\n"), path, value);
228 err = check_file(path, force, isspare);
229 #endif /* HAVE_LIBBLKID */
235 * Validate a whole disk. Iterate over all slices on the disk and make sure
236 * that none is in use by calling check_slice().
239 check_disk(const char *path, blkid_cache cache, int force,
240 boolean_t isspare, boolean_t iswholedisk)
243 char slice_path[MAXPATHLEN];
247 /* This is not a wholedisk we only check the given partition */
249 return check_slice(path, cache, force, isspare);
252 * When the device is a whole disk try to read the efi partition
253 * label. If this is successful we safely check the all of the
254 * partitions. However, when it fails it may simply be because
255 * the disk is partitioned via the MBR. Since we currently can
256 * not easily decode the MBR return a failure and prompt to the
257 * user to use force option since we cannot check the partitions.
259 if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0) {
264 if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
270 vdev_error(gettext("%s does not contain an EFI "
271 "label but it may contain partition\n"
272 "information in the MBR.\n"), path);
278 * The primary efi partition label is damaged however the secondary
279 * label at the end of the device is intact. Rather than use this
280 * label we should play it safe and treat this as a non efi device.
282 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
287 /* Partitions will no be created using the backup */
290 vdev_error(gettext("%s contains a corrupt primary "
291 "EFI label.\n"), path);
296 for (i = 0; i < vtoc->efi_nparts; i++) {
298 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
299 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
302 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
303 (void) snprintf(slice_path, sizeof (slice_path),
304 "%s%s%d", path, "-part", i+1);
306 (void) snprintf(slice_path, sizeof (slice_path),
307 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
310 err = check_slice(slice_path, cache, force, isspare);
322 check_device(const char *path, boolean_t force,
323 boolean_t isspare, boolean_t iswholedisk)
325 static blkid_cache cache = NULL;
329 * There is no easy way to add a correct blkid_put_cache() call,
330 * memory will be reclaimed when the command exits.
335 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
340 if ((err = blkid_probe_all(cache)) != 0) {
341 blkid_put_cache(cache);
346 #endif /* HAVE_LIBBLKID */
348 return check_disk(path, cache, force, isspare, iswholedisk);
352 * By "whole disk" we mean an entire physical disk (something we can
353 * label, toggle the write cache on, etc.) as opposed to the full
354 * capacity of a pseudo-device such as lofi or did. We act as if we
355 * are labeling the disk, which should be a pretty good test of whether
356 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
360 is_whole_disk(const char *path)
362 struct dk_gpt *label;
365 if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0)
367 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
377 * This may be a shorthand device path or it could be total gibberish.
378 * Check to see if it's a known device in /dev/, /dev/disk/by-id,
379 * /dev/disk/by-label, /dev/disk/by-path, /dev/disk/by-uuid, or
380 * /dev/disk/zpool/. As part of this check, see if we've been given
381 * an entire disk (minus the slice number).
384 is_shorthand_path(const char *arg, char *path,
385 struct stat64 *statbuf, boolean_t *wholedisk)
387 char dirs[5][9] = {"by-id", "by-label", "by-path", "by-uuid", "zpool"};
391 (void) snprintf(path, MAXPATHLEN, "%s/%s", DISK_ROOT, arg);
392 *wholedisk = is_whole_disk(path);
393 err = stat64(path, statbuf);
394 if (*wholedisk || err == 0)
397 /* /dev/disk/<dirs>/<name> */
398 for (i = 0; i < 5; i++) {
399 (void) snprintf(path, MAXPATHLEN, "%s/%s/%s",
400 UDISK_ROOT, dirs[i], arg);
401 *wholedisk = is_whole_disk(path);
402 err = stat64(path, statbuf);
403 if (*wholedisk || err == 0)
407 strlcpy(path, arg, sizeof(path));
408 memset(statbuf, 0, sizeof(*statbuf));
409 *wholedisk = B_FALSE;
415 * Create a leaf vdev. Determine if this is a file or a device. If it's a
416 * device, fill in the device id to make a complete nvlist. Valid forms for a
419 * /dev/xxx Complete disk path
420 * /xxx Full path to file
421 * xxx Shorthand for /dev/disk/yyy/xxx
424 make_leaf_vdev(const char *arg, uint64_t is_log)
426 char path[MAXPATHLEN];
427 struct stat64 statbuf;
428 nvlist_t *vdev = NULL;
430 boolean_t wholedisk = B_FALSE;
434 * Determine what type of vdev this is, and put the full path into
435 * 'path'. We detect whether this is a device of file afterwards by
436 * checking the st_mode of the file.
440 * Complete device or file path. Exact type is determined by
441 * examining the file descriptor afterwards. Symbolic links
442 * are resolved to their real paths for the is_whole_disk()
443 * and S_ISBLK/S_ISREG type checks. However, we are careful
444 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
445 * can leverage udev's persistent device labels.
447 if (realpath(arg, path) == NULL) {
448 (void) fprintf(stderr,
449 gettext("cannot resolve path '%s'\n"), arg);
453 wholedisk = is_whole_disk(path);
454 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
455 (void) fprintf(stderr,
456 gettext("cannot open '%s': %s\n"),
457 path, strerror(errno));
461 /* After is_whole_disk() check restore original passed path */
462 strlcpy(path, arg, MAXPATHLEN);
464 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
467 * If we got ENOENT, then the user gave us
468 * gibberish, so try to direct them with a
469 * reasonable error message. Otherwise,
470 * regurgitate strerror() since it's the best we
474 (void) fprintf(stderr,
475 gettext("cannot open '%s': no such "
476 "device in %s\n"), arg, DISK_ROOT);
477 (void) fprintf(stderr,
478 gettext("must be a full path or "
479 "shorthand device name\n"));
482 (void) fprintf(stderr,
483 gettext("cannot open '%s': %s\n"),
484 path, strerror(errno));
491 * Determine whether this is a device or a file.
493 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
494 type = VDEV_TYPE_DISK;
495 } else if (S_ISREG(statbuf.st_mode)) {
496 type = VDEV_TYPE_FILE;
498 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
499 "block device or regular file\n"), path);
504 * Finally, we have the complete device or file, and we know that it is
505 * acceptable to use. Construct the nvlist to describe this vdev. All
506 * vdevs have a 'path' element, and devices also have a 'devid' element.
508 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
509 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
510 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
511 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
512 if (strcmp(type, VDEV_TYPE_DISK) == 0)
513 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
514 (uint64_t)wholedisk) == 0);
516 #if defined(__sun__) || defined(__sun)
518 * For a whole disk, defer getting its devid until after labeling it.
520 if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
522 * Get the devid for the device.
526 char *minor = NULL, *devid_str = NULL;
528 if ((fd = open(path, O_RDONLY|O_EXCL)) < 0) {
529 (void) fprintf(stderr, gettext("cannot open '%s': "
530 "%s\n"), path, strerror(errno));
535 if (devid_get(fd, &devid) == 0) {
536 if (devid_get_minor_name(fd, &minor) == 0 &&
537 (devid_str = devid_str_encode(devid, minor)) !=
539 verify(nvlist_add_string(vdev,
540 ZPOOL_CONFIG_DEVID, devid_str) == 0);
542 if (devid_str != NULL)
543 devid_str_free(devid_str);
545 devid_str_free(minor);
557 * Go through and verify the replication level of the pool is consistent.
558 * Performs the following checks:
560 * For the new spec, verifies that devices in mirrors and raidz are the
563 * If the current configuration already has inconsistent replication
564 * levels, ignore any other potential problems in the new spec.
566 * Otherwise, make sure that the current spec (if there is one) and the new
567 * spec have consistent replication levels.
569 typedef struct replication_level {
571 uint64_t zprl_children;
572 uint64_t zprl_parity;
573 } replication_level_t;
575 #define ZPOOL_FUZZ (16 * 1024 * 1024)
578 * Given a list of toplevel vdevs, return the current replication level. If
579 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
580 * an error message will be displayed for each self-inconsistent vdev.
582 static replication_level_t *
583 get_replication(nvlist_t *nvroot, boolean_t fatal)
591 replication_level_t lastrep = { 0 }, rep, *ret;
592 boolean_t dontreport;
594 ret = safe_malloc(sizeof (replication_level_t));
596 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
597 &top, &toplevels) == 0);
599 lastrep.zprl_type = NULL;
600 for (t = 0; t < toplevels; t++) {
601 uint64_t is_log = B_FALSE;
606 * For separate logs we ignore the top level vdev replication
609 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
613 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
615 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
616 &child, &children) != 0) {
618 * This is a 'file' or 'disk' vdev.
620 rep.zprl_type = type;
621 rep.zprl_children = 1;
627 * This is a mirror or RAID-Z vdev. Go through and make
628 * sure the contents are all the same (files vs. disks),
629 * keeping track of the number of elements in the
632 * We also check that the size of each vdev (if it can
633 * be determined) is the same.
635 rep.zprl_type = type;
636 rep.zprl_children = 0;
638 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
639 verify(nvlist_lookup_uint64(nv,
640 ZPOOL_CONFIG_NPARITY,
641 &rep.zprl_parity) == 0);
642 assert(rep.zprl_parity != 0);
648 * The 'dontreport' variable indicates that we've
649 * already reported an error for this spec, so don't
650 * bother doing it again.
655 for (c = 0; c < children; c++) {
656 nvlist_t *cnv = child[c];
658 struct stat64 statbuf;
659 uint64_t size = -1ULL;
665 verify(nvlist_lookup_string(cnv,
666 ZPOOL_CONFIG_TYPE, &childtype) == 0);
669 * If this is a replacing or spare vdev, then
670 * get the real first child of the vdev.
672 if (strcmp(childtype,
673 VDEV_TYPE_REPLACING) == 0 ||
674 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
678 verify(nvlist_lookup_nvlist_array(cnv,
679 ZPOOL_CONFIG_CHILDREN, &rchild,
681 assert(rchildren == 2);
684 verify(nvlist_lookup_string(cnv,
689 verify(nvlist_lookup_string(cnv,
690 ZPOOL_CONFIG_PATH, &path) == 0);
693 * If we have a raidz/mirror that combines disks
694 * with files, report it as an error.
696 if (!dontreport && type != NULL &&
697 strcmp(type, childtype) != 0) {
703 "mismatched replication "
704 "level: %s contains both "
705 "files and devices\n"),
713 * According to stat(2), the value of 'st_size'
714 * is undefined for block devices and character
715 * devices. But there is no effective way to
716 * determine the real size in userland.
718 * Instead, we'll take advantage of an
719 * implementation detail of spec_size(). If the
720 * device is currently open, then we (should)
721 * return a valid size.
723 * If we still don't get a valid size (indicated
724 * by a size of 0 or MAXOFFSET_T), then ignore
725 * this device altogether.
727 if ((fd = open(path, O_RDONLY)) >= 0) {
728 err = fstat64(fd, &statbuf);
731 err = stat64(path, &statbuf);
735 statbuf.st_size == 0 ||
736 statbuf.st_size == MAXOFFSET_T)
739 size = statbuf.st_size;
742 * Also make sure that devices and
743 * slices have a consistent size. If
744 * they differ by a significant amount
745 * (~16MB) then report an error.
748 (vdev_size != -1ULL &&
749 (labs(size - vdev_size) >
756 "%s contains devices of "
757 "different sizes\n"),
770 * At this point, we have the replication of the last toplevel
771 * vdev in 'rep'. Compare it to 'lastrep' to see if its
774 if (lastrep.zprl_type != NULL) {
775 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
781 "mismatched replication level: "
782 "both %s and %s vdevs are "
784 lastrep.zprl_type, rep.zprl_type);
787 } else if (lastrep.zprl_parity != rep.zprl_parity) {
793 "mismatched replication level: "
794 "both %llu and %llu device parity "
795 "%s vdevs are present\n"),
801 } else if (lastrep.zprl_children != rep.zprl_children) {
807 "mismatched replication level: "
808 "both %llu-way and %llu-way %s "
809 "vdevs are present\n"),
810 lastrep.zprl_children,
827 * Check the replication level of the vdev spec against the current pool. Calls
828 * get_replication() to make sure the new spec is self-consistent. If the pool
829 * has a consistent replication level, then we ignore any errors. Otherwise,
830 * report any difference between the two.
833 check_replication(nvlist_t *config, nvlist_t *newroot)
837 replication_level_t *current = NULL, *new;
841 * If we have a current pool configuration, check to see if it's
842 * self-consistent. If not, simply return success.
844 if (config != NULL) {
847 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
849 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
853 * for spares there may be no children, and therefore no
854 * replication level to check
856 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
857 &child, &children) != 0) || (children == 0)) {
863 * If all we have is logs then there's no replication level to check.
865 if (num_logs(newroot) == children) {
871 * Get the replication level of the new vdev spec, reporting any
872 * inconsistencies found.
874 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
880 * Check to see if the new vdev spec matches the replication level of
884 if (current != NULL) {
885 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
887 "mismatched replication level: pool uses %s "
888 "and new vdev is %s\n"),
889 current->zprl_type, new->zprl_type);
891 } else if (current->zprl_parity != new->zprl_parity) {
893 "mismatched replication level: pool uses %llu "
894 "device parity and new vdev uses %llu\n"),
895 current->zprl_parity, new->zprl_parity);
897 } else if (current->zprl_children != new->zprl_children) {
899 "mismatched replication level: pool uses %llu-way "
900 "%s and new vdev uses %llu-way %s\n"),
901 current->zprl_children, current->zprl_type,
902 new->zprl_children, new->zprl_type);
915 zero_label(char *path)
917 const int size = 4096;
921 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
922 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
923 path, strerror(errno));
927 memset(buf, 0, size);
928 err = write(fd, buf, size);
929 (void) fdatasync(fd);
933 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
934 "of '%s': %s\n"), size, path, strerror(errno));
939 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
940 "of '%s'\n"), err, size, path);
948 * Go through and find any whole disks in the vdev specification, labelling them
949 * as appropriate. When constructing the vdev spec, we were unable to open this
950 * device in order to provide a devid. Now that we have labelled the disk and
951 * know that slice 0 is valid, we can construct the devid now.
953 * If the disk was already labeled with an EFI label, we will have gotten the
954 * devid already (because we were able to open the whole disk). Otherwise, we
955 * need to get the devid after we label the disk.
958 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
962 char *type, *path, *diskname;
963 char buf[MAXPATHLEN];
967 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
969 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
970 &child, &children) != 0) {
972 if (strcmp(type, VDEV_TYPE_DISK) != 0)
976 * We have a disk device. If this is a whole disk write
977 * out the efi partition table, otherwise write zero's to
978 * the first 4k of the partition. This is to ensure that
979 * libblkid will not misidentify the partition due to a
980 * magic value left by the previous filesystem.
982 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
983 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
987 ret = zero_label(path);
991 if (realpath(path, buf) == NULL) {
993 (void) fprintf(stderr,
994 gettext("cannot resolve path '%s'\n"), path);
998 diskname = strrchr(buf, '/');
999 assert(diskname != NULL);
1001 if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
1005 * Now the we've labeled the disk and the partitions have
1006 * been created. We still need to wait for udev to create
1007 * the symlinks to those partitions. If we are accessing
1008 * the devices via a udev disk path, /dev/disk, then wait
1009 * for *-part# to be created. Otherwise just use the normal
1010 * syntax for devices in /dev.
1012 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
1013 (void) snprintf(buf, sizeof (buf),
1014 "%s%s%s", path, "-part", FIRST_SLICE);
1016 (void) snprintf(buf, sizeof (buf),
1017 "%s%s%s", path, isdigit(path[strlen(path)-1]) ?
1018 "p" : "", FIRST_SLICE);
1020 if ((ret = zpool_label_disk_wait(buf, 1000)) != 0) {
1021 (void) fprintf(stderr,
1022 gettext( "cannot resolve path '%s'\n"), buf);
1027 * Update the path to refer to FIRST_SLICE. The presence of
1028 * the 'whole_disk' field indicates to the CLI that we should
1029 * chop off the slice number when displaying the device in
1032 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
1034 /* Just in case this partition already existed. */
1035 (void) zero_label(buf);
1040 for (c = 0; c < children; c++)
1041 if ((ret = make_disks(zhp, child[c])) != 0)
1044 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1045 &child, &children) == 0)
1046 for (c = 0; c < children; c++)
1047 if ((ret = make_disks(zhp, child[c])) != 0)
1050 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1051 &child, &children) == 0)
1052 for (c = 0; c < children; c++)
1053 if ((ret = make_disks(zhp, child[c])) != 0)
1060 * Determine if the given path is a hot spare within the given configuration.
1063 is_spare(nvlist_t *config, const char *path)
1069 uint64_t guid, spareguid;
1075 if ((fd = open(path, O_RDONLY|O_EXCL)) < 0)
1078 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1080 state != POOL_STATE_SPARE ||
1081 zpool_read_label(fd, &label) != 0) {
1089 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1092 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1094 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1095 &spares, &nspares) == 0) {
1096 for (i = 0; i < nspares; i++) {
1097 verify(nvlist_lookup_uint64(spares[i],
1098 ZPOOL_CONFIG_GUID, &spareguid) == 0);
1099 if (spareguid == guid)
1108 * Go through and find any devices that are in use. We rely on libdiskmgt for
1109 * the majority of this task.
1112 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1113 boolean_t replacing, boolean_t isspare)
1119 char buf[MAXPATHLEN];
1120 uint64_t wholedisk = B_FALSE;
1122 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1124 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1125 &child, &children) != 0) {
1127 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1128 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1129 verify(!nvlist_lookup_uint64(nv,
1130 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1133 * As a generic check, we look to see if this is a replace of a
1134 * hot spare within the same pool. If so, we allow it
1135 * regardless of what libblkid or zpool_in_use() says.
1139 (void) snprintf(buf, sizeof (buf), "%ss0",
1142 (void) strlcpy(buf, path, sizeof (buf));
1144 if (is_spare(config, buf))
1148 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1149 ret = check_device(path, force, isspare, wholedisk);
1151 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1152 ret = check_file(path, force, isspare);
1157 for (c = 0; c < children; c++)
1158 if ((ret = check_in_use(config, child[c], force,
1159 replacing, B_FALSE)) != 0)
1162 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1163 &child, &children) == 0)
1164 for (c = 0; c < children; c++)
1165 if ((ret = check_in_use(config, child[c], force,
1166 replacing, B_TRUE)) != 0)
1169 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1170 &child, &children) == 0)
1171 for (c = 0; c < children; c++)
1172 if ((ret = check_in_use(config, child[c], force,
1173 replacing, B_FALSE)) != 0)
1180 is_grouping(const char *type, int *mindev, int *maxdev)
1182 if (strncmp(type, "raidz", 5) == 0) {
1183 const char *p = type + 5;
1189 } else if (*p == '0') {
1190 return (NULL); /* no zero prefixes allowed */
1193 nparity = strtol(p, &end, 10);
1194 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1200 *mindev = nparity + 1;
1203 return (VDEV_TYPE_RAIDZ);
1209 if (strcmp(type, "mirror") == 0) {
1212 return (VDEV_TYPE_MIRROR);
1215 if (strcmp(type, "spare") == 0) {
1218 return (VDEV_TYPE_SPARE);
1221 if (strcmp(type, "log") == 0) {
1224 return (VDEV_TYPE_LOG);
1227 if (strcmp(type, "cache") == 0) {
1230 return (VDEV_TYPE_L2CACHE);
1237 * Construct a syntactically valid vdev specification,
1238 * and ensure that all devices and files exist and can be opened.
1239 * Note: we don't bother freeing anything in the error paths
1240 * because the program is just going to exit anyway.
1243 construct_spec(int argc, char **argv)
1245 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1246 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1249 boolean_t seen_logs;
1259 seen_logs = B_FALSE;
1265 * If it's a mirror or raidz, the subsequent arguments are
1266 * its leaves -- until we encounter the next mirror or raidz.
1268 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1269 nvlist_t **child = NULL;
1270 int c, children = 0;
1272 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1273 if (spares != NULL) {
1274 (void) fprintf(stderr,
1275 gettext("invalid vdev "
1276 "specification: 'spare' can be "
1277 "specified only once\n"));
1283 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1285 (void) fprintf(stderr,
1286 gettext("invalid vdev "
1287 "specification: 'log' can be "
1288 "specified only once\n"));
1296 * A log is not a real grouping device.
1297 * We just set is_log and continue.
1302 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1303 if (l2cache != NULL) {
1304 (void) fprintf(stderr,
1305 gettext("invalid vdev "
1306 "specification: 'cache' can be "
1307 "specified only once\n"));
1314 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1315 (void) fprintf(stderr,
1316 gettext("invalid vdev "
1317 "specification: unsupported 'log' "
1318 "device: %s\n"), type);
1324 for (c = 1; c < argc; c++) {
1325 if (is_grouping(argv[c], NULL, NULL) != NULL)
1328 child = realloc(child,
1329 children * sizeof (nvlist_t *));
1332 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1335 child[children - 1] = nv;
1338 if (children < mindev) {
1339 (void) fprintf(stderr, gettext("invalid vdev "
1340 "specification: %s requires at least %d "
1341 "devices\n"), argv[0], mindev);
1345 if (children > maxdev) {
1346 (void) fprintf(stderr, gettext("invalid vdev "
1347 "specification: %s supports no more than "
1348 "%d devices\n"), argv[0], maxdev);
1355 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1359 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1361 nl2cache = children;
1364 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1366 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1368 verify(nvlist_add_uint64(nv,
1369 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1370 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1371 verify(nvlist_add_uint64(nv,
1372 ZPOOL_CONFIG_NPARITY,
1375 verify(nvlist_add_nvlist_array(nv,
1376 ZPOOL_CONFIG_CHILDREN, child,
1379 for (c = 0; c < children; c++)
1380 nvlist_free(child[c]);
1385 * We have a device. Pass off to make_leaf_vdev() to
1386 * construct the appropriate nvlist describing the vdev.
1388 if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1397 top = realloc(top, toplevels * sizeof (nvlist_t *));
1400 top[toplevels - 1] = nv;
1403 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1404 (void) fprintf(stderr, gettext("invalid vdev "
1405 "specification: at least one toplevel vdev must be "
1410 if (seen_logs && nlogs == 0) {
1411 (void) fprintf(stderr, gettext("invalid vdev specification: "
1412 "log requires at least 1 device\n"));
1417 * Finally, create nvroot and add all top-level vdevs to it.
1419 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1420 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1421 VDEV_TYPE_ROOT) == 0);
1422 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1423 top, toplevels) == 0);
1425 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1426 spares, nspares) == 0);
1428 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1429 l2cache, nl2cache) == 0);
1431 for (t = 0; t < toplevels; t++)
1432 nvlist_free(top[t]);
1433 for (t = 0; t < nspares; t++)
1434 nvlist_free(spares[t]);
1435 for (t = 0; t < nl2cache; t++)
1436 nvlist_free(l2cache[t]);
1447 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1448 splitflags_t flags, int argc, char **argv)
1450 nvlist_t *newroot = NULL, **child;
1454 if ((newroot = construct_spec(argc, argv)) == NULL) {
1455 (void) fprintf(stderr, gettext("Unable to build a "
1456 "pool from the specified devices\n"));
1460 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1461 nvlist_free(newroot);
1465 /* avoid any tricks in the spec */
1466 verify(nvlist_lookup_nvlist_array(newroot,
1467 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1468 for (c = 0; c < children; c++) {
1473 verify(nvlist_lookup_string(child[c],
1474 ZPOOL_CONFIG_PATH, &path) == 0);
1475 if ((type = is_grouping(path, &min, &max)) != NULL) {
1476 (void) fprintf(stderr, gettext("Cannot use "
1477 "'%s' as a device for splitting\n"), type);
1478 nvlist_free(newroot);
1484 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1485 if (newroot != NULL)
1486 nvlist_free(newroot);
1494 * Get and validate the contents of the given vdev specification. This ensures
1495 * that the nvlist returned is well-formed, that all the devices exist, and that
1496 * they are not currently in use by any other known consumer. The 'poolconfig'
1497 * parameter is the current configuration of the pool when adding devices
1498 * existing pool, and is used to perform additional checks, such as changing the
1499 * replication level of the pool. It can be 'NULL' to indicate that this is a
1500 * new pool. The 'force' flag controls whether devices should be forcefully
1501 * added, even if they appear in use.
1504 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1505 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1508 nvlist_t *poolconfig = NULL;
1512 * Construct the vdev specification. If this is successful, we know
1513 * that we have a valid specification, and that all devices can be
1516 if ((newroot = construct_spec(argc, argv)) == NULL)
1519 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1523 * Validate each device to make sure that its not shared with another
1524 * subsystem. We do this even if 'force' is set, because there are some
1525 * uses (such as a dedicated dump device) that even '-f' cannot
1528 if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1529 nvlist_free(newroot);
1534 * Check the replication level of the given vdevs and report any errors
1535 * found. We include the existing pool spec, if any, as we need to
1536 * catch changes against the existing replication level.
1538 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1539 nvlist_free(newroot);
1544 * Run through the vdev specification and label any whole disks found.
1546 if (!dryrun && make_disks(zhp, newroot) != 0) {
1547 nvlist_free(newroot);