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>
70 #include <scsi/scsi.h>
75 #include <sys/efi_partition.h>
78 #include <sys/mntent.h>
79 #include <uuid/uuid.h>
81 #include <blkid/blkid.h>
83 #define blkid_cache void *
84 #endif /* HAVE_LIBBLKID */
86 #include "zpool_util.h"
87 #include <sys/zfs_context.h>
90 * For any given vdev specification, we can have multiple errors. The
91 * vdev_error() function keeps track of whether we have seen an error yet, and
92 * prints out a header if its the first error we've seen.
97 typedef struct vdev_disk_db_entry
101 } vdev_disk_db_entry_t;
104 * Database of block devices that lie about physical sector sizes. The
105 * identification string must be precisely 24 characters to avoid false
108 static vdev_disk_db_entry_t vdev_disk_database[] = {
109 {"ATA Corsair Force 3 ", 8192},
110 {"ATA INTEL SSDSA2CT04", 8192},
111 {"ATA INTEL SSDSA2CW16", 8192},
112 {"ATA INTEL SSDSC2CT18", 8192},
113 {"ATA INTEL SSDSC2CW12", 8192},
114 {"ATA KINGSTON SH100S3", 8192},
115 {"ATA M4-CT064M4SSD2 ", 8192},
116 {"ATA M4-CT128M4SSD2 ", 8192},
117 {"ATA M4-CT256M4SSD2 ", 8192},
118 {"ATA M4-CT512M4SSD2 ", 8192},
119 {"ATA OCZ-AGILITY2 ", 8192},
120 {"ATA OCZ-VERTEX2 3.5 ", 8192},
121 {"ATA OCZ-VERTEX3 ", 8192},
122 {"ATA OCZ-VERTEX3 LT ", 8192},
123 {"ATA OCZ-VERTEX3 MI ", 8192},
124 {"ATA SAMSUNG SSD 830 ", 8192},
125 {"ATA Samsung SSD 840 ", 8192},
126 {"ATA INTEL SSDSA2M040", 4096},
127 {"ATA INTEL SSDSA2M080", 4096},
128 {"ATA INTEL SSDSA2M160", 4096},
129 /* Imported from Open Solaris*/
130 {"ATA MARVELL SD88SA02", 4096},
131 /* Advanced format Hard drives */
132 {"ATA Hitachi HDS5C303", 4096},
133 {"ATA SAMSUNG HD204UI ", 4096},
134 {"ATA ST2000DL004 HD20", 4096},
135 {"ATA WDC WD10EARS-00M", 4096},
136 {"ATA WDC WD10EARS-00S", 4096},
137 {"ATA WDC WD10EARS-00Z", 4096},
138 {"ATA WDC WD15EARS-00M", 4096},
139 {"ATA WDC WD15EARS-00S", 4096},
140 {"ATA WDC WD15EARS-00Z", 4096},
141 {"ATA WDC WD20EARS-00M", 4096},
142 {"ATA WDC WD20EARS-00S", 4096},
143 {"ATA WDC WD20EARS-00Z", 4096},
144 /* Virtual disks: Assume zvols with default volblocksize */
146 {"ATA QEMU HARDDISK ", 8192},
147 {"IET VIRTUAL-DISK ", 8192},
148 {"OI COMSTAR ", 8192},
152 static const int vdev_disk_database_size =
153 sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
155 #define INQ_REPLY_LEN 96
156 #define INQ_CMD_LEN 6
159 check_sector_size_database(char *path, int *sector_size)
161 unsigned char inq_buff[INQ_REPLY_LEN];
162 unsigned char sense_buffer[32];
163 unsigned char inq_cmd_blk[INQ_CMD_LEN] =
164 {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
170 /* Prepare INQUIRY command */
171 memset(&io_hdr, 0, sizeof(sg_io_hdr_t));
172 io_hdr.interface_id = 'S';
173 io_hdr.cmd_len = sizeof(inq_cmd_blk);
174 io_hdr.mx_sb_len = sizeof(sense_buffer);
175 io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
176 io_hdr.dxfer_len = INQ_REPLY_LEN;
177 io_hdr.dxferp = inq_buff;
178 io_hdr.cmdp = inq_cmd_blk;
179 io_hdr.sbp = sense_buffer;
180 io_hdr.timeout = 10; /* 10 milliseconds is ample time */
182 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
185 error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
192 if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
195 for (i = 0; i < vdev_disk_database_size; i++) {
196 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
199 *sector_size = vdev_disk_database[i].sector_size;
208 vdev_error(const char *fmt, ...)
213 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
215 (void) fprintf(stderr, gettext("use '-f' to override "
216 "the following errors:\n"));
218 (void) fprintf(stderr, gettext("the following errors "
219 "must be manually repaired:\n"));
224 (void) vfprintf(stderr, fmt, ap);
229 * Check that a file is valid. All we can do in this case is check that it's
230 * not in use by another pool, and not in use by swap.
233 check_file(const char *file, boolean_t force, boolean_t isspare)
241 if ((fd = open(file, O_RDONLY)) < 0)
244 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
248 case POOL_STATE_ACTIVE:
249 desc = gettext("active");
252 case POOL_STATE_EXPORTED:
253 desc = gettext("exported");
256 case POOL_STATE_POTENTIALLY_ACTIVE:
257 desc = gettext("potentially active");
261 desc = gettext("unknown");
266 * Allow hot spares to be shared between pools.
268 if (state == POOL_STATE_SPARE && isspare)
271 if (state == POOL_STATE_ACTIVE ||
272 state == POOL_STATE_SPARE || !force) {
274 case POOL_STATE_SPARE:
275 vdev_error(gettext("%s is reserved as a hot "
276 "spare for pool %s\n"), file, name);
279 vdev_error(gettext("%s is part of %s pool "
280 "'%s'\n"), file, desc, name);
296 (void) fprintf(stderr, gettext("warning: device in use checking "
297 "failed: %s\n"), strerror(err));
301 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
307 /* No valid type detected device is safe to use */
308 value = blkid_get_tag_value(cache, "TYPE", path);
313 * If libblkid detects a ZFS device, we check the device
314 * using check_file() to see if it's safe. The one safe
315 * case is a spare device shared between multiple pools.
317 if (strcmp(value, "zfs") == 0) {
318 err = check_file(path, force, isspare);
324 vdev_error(gettext("%s contains a filesystem of "
325 "type '%s'\n"), path, value);
331 err = check_file(path, force, isspare);
332 #endif /* HAVE_LIBBLKID */
338 * Validate a whole disk. Iterate over all slices on the disk and make sure
339 * that none is in use by calling check_slice().
342 check_disk(const char *path, blkid_cache cache, int force,
343 boolean_t isspare, boolean_t iswholedisk)
346 char slice_path[MAXPATHLEN];
350 /* This is not a wholedisk we only check the given partition */
352 return check_slice(path, cache, force, isspare);
355 * When the device is a whole disk try to read the efi partition
356 * label. If this is successful we safely check the all of the
357 * partitions. However, when it fails it may simply be because
358 * the disk is partitioned via the MBR. Since we currently can
359 * not easily decode the MBR return a failure and prompt to the
360 * user to use force option since we cannot check the partitions.
362 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
367 if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
373 vdev_error(gettext("%s does not contain an EFI "
374 "label but it may contain partition\n"
375 "information in the MBR.\n"), path);
381 * The primary efi partition label is damaged however the secondary
382 * label at the end of the device is intact. Rather than use this
383 * label we should play it safe and treat this as a non efi device.
385 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
390 /* Partitions will no be created using the backup */
393 vdev_error(gettext("%s contains a corrupt primary "
394 "EFI label.\n"), path);
399 for (i = 0; i < vtoc->efi_nparts; i++) {
401 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
402 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
405 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
406 (void) snprintf(slice_path, sizeof (slice_path),
407 "%s%s%d", path, "-part", i+1);
409 (void) snprintf(slice_path, sizeof (slice_path),
410 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
413 err = check_slice(slice_path, cache, force, isspare);
425 check_device(const char *path, boolean_t force,
426 boolean_t isspare, boolean_t iswholedisk)
428 static blkid_cache cache = NULL;
432 * There is no easy way to add a correct blkid_put_cache() call,
433 * memory will be reclaimed when the command exits.
438 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
443 if ((err = blkid_probe_all(cache)) != 0) {
444 blkid_put_cache(cache);
449 #endif /* HAVE_LIBBLKID */
451 return check_disk(path, cache, force, isspare, iswholedisk);
455 * By "whole disk" we mean an entire physical disk (something we can
456 * label, toggle the write cache on, etc.) as opposed to the full
457 * capacity of a pseudo-device such as lofi or did. We act as if we
458 * are labeling the disk, which should be a pretty good test of whether
459 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
463 is_whole_disk(const char *path)
465 struct dk_gpt *label;
468 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
470 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
480 * This may be a shorthand device path or it could be total gibberish.
481 * Check to see if it is a known device available in zfs_vdev_paths.
482 * As part of this check, see if we've been given an entire disk
483 * (minus the slice number).
486 is_shorthand_path(const char *arg, char *path,
487 struct stat64 *statbuf, boolean_t *wholedisk)
491 error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
493 *wholedisk = is_whole_disk(path);
494 if (*wholedisk || (stat64(path, statbuf) == 0))
498 strlcpy(path, arg, sizeof(path));
499 memset(statbuf, 0, sizeof(*statbuf));
500 *wholedisk = B_FALSE;
506 * Determine if the given path is a hot spare within the given configuration.
507 * If no configuration is given we rely solely on the label.
510 is_spare(nvlist_t *config, const char *path)
516 uint64_t guid, spareguid;
522 if ((fd = open(path, O_RDONLY)) < 0)
525 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
527 state != POOL_STATE_SPARE ||
528 zpool_read_label(fd, &label) != 0) {
539 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
542 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
544 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
545 &spares, &nspares) == 0) {
546 for (i = 0; i < nspares; i++) {
547 verify(nvlist_lookup_uint64(spares[i],
548 ZPOOL_CONFIG_GUID, &spareguid) == 0);
549 if (spareguid == guid)
558 * Create a leaf vdev. Determine if this is a file or a device. If it's a
559 * device, fill in the device id to make a complete nvlist. Valid forms for a
562 * /dev/xxx Complete disk path
563 * /xxx Full path to file
564 * xxx Shorthand for <zfs_vdev_paths>/xxx
567 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
569 char path[MAXPATHLEN];
570 struct stat64 statbuf;
571 nvlist_t *vdev = NULL;
573 boolean_t wholedisk = B_FALSE;
578 * Determine what type of vdev this is, and put the full path into
579 * 'path'. We detect whether this is a device of file afterwards by
580 * checking the st_mode of the file.
584 * Complete device or file path. Exact type is determined by
585 * examining the file descriptor afterwards. Symbolic links
586 * are resolved to their real paths for the is_whole_disk()
587 * and S_ISBLK/S_ISREG type checks. However, we are careful
588 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
589 * can leverage udev's persistent device labels.
591 if (realpath(arg, path) == NULL) {
592 (void) fprintf(stderr,
593 gettext("cannot resolve path '%s'\n"), arg);
597 wholedisk = is_whole_disk(path);
598 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
599 (void) fprintf(stderr,
600 gettext("cannot open '%s': %s\n"),
601 path, strerror(errno));
605 /* After is_whole_disk() check restore original passed path */
606 strlcpy(path, arg, MAXPATHLEN);
608 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
611 * If we got ENOENT, then the user gave us
612 * gibberish, so try to direct them with a
613 * reasonable error message. Otherwise,
614 * regurgitate strerror() since it's the best we
618 (void) fprintf(stderr,
619 gettext("cannot open '%s': no such "
620 "device in %s\n"), arg, DISK_ROOT);
621 (void) fprintf(stderr,
622 gettext("must be a full path or "
623 "shorthand device name\n"));
626 (void) fprintf(stderr,
627 gettext("cannot open '%s': %s\n"),
628 path, strerror(errno));
635 * Determine whether this is a device or a file.
637 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
638 type = VDEV_TYPE_DISK;
639 } else if (S_ISREG(statbuf.st_mode)) {
640 type = VDEV_TYPE_FILE;
642 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
643 "block device or regular file\n"), path);
648 * Finally, we have the complete device or file, and we know that it is
649 * acceptable to use. Construct the nvlist to describe this vdev. All
650 * vdevs have a 'path' element, and devices also have a 'devid' element.
652 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
653 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
654 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
655 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
656 if (strcmp(type, VDEV_TYPE_DISK) == 0)
657 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
658 (uint64_t)wholedisk) == 0);
661 * Override defaults if custom properties are provided.
666 if (nvlist_lookup_string(props,
667 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
668 zfs_nicestrtonum(NULL, value, &ashift);
672 * If the device is known to incorrectly report its physical sector
673 * size explicitly provide the known correct value.
678 if (check_sector_size_database(path, §or_size) == B_TRUE)
679 ashift = highbit(sector_size) - 1;
683 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
689 * Go through and verify the replication level of the pool is consistent.
690 * Performs the following checks:
692 * For the new spec, verifies that devices in mirrors and raidz are the
695 * If the current configuration already has inconsistent replication
696 * levels, ignore any other potential problems in the new spec.
698 * Otherwise, make sure that the current spec (if there is one) and the new
699 * spec have consistent replication levels.
701 typedef struct replication_level {
703 uint64_t zprl_children;
704 uint64_t zprl_parity;
705 } replication_level_t;
707 #define ZPOOL_FUZZ (16 * 1024 * 1024)
710 * Given a list of toplevel vdevs, return the current replication level. If
711 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
712 * an error message will be displayed for each self-inconsistent vdev.
714 static replication_level_t *
715 get_replication(nvlist_t *nvroot, boolean_t fatal)
723 replication_level_t lastrep = { 0 }, rep, *ret;
724 boolean_t dontreport;
726 ret = safe_malloc(sizeof (replication_level_t));
728 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
729 &top, &toplevels) == 0);
731 lastrep.zprl_type = NULL;
732 for (t = 0; t < toplevels; t++) {
733 uint64_t is_log = B_FALSE;
738 * For separate logs we ignore the top level vdev replication
741 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
745 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
747 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
748 &child, &children) != 0) {
750 * This is a 'file' or 'disk' vdev.
752 rep.zprl_type = type;
753 rep.zprl_children = 1;
759 * This is a mirror or RAID-Z vdev. Go through and make
760 * sure the contents are all the same (files vs. disks),
761 * keeping track of the number of elements in the
764 * We also check that the size of each vdev (if it can
765 * be determined) is the same.
767 rep.zprl_type = type;
768 rep.zprl_children = 0;
770 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
771 verify(nvlist_lookup_uint64(nv,
772 ZPOOL_CONFIG_NPARITY,
773 &rep.zprl_parity) == 0);
774 assert(rep.zprl_parity != 0);
780 * The 'dontreport' variable indicates that we've
781 * already reported an error for this spec, so don't
782 * bother doing it again.
787 for (c = 0; c < children; c++) {
788 nvlist_t *cnv = child[c];
790 struct stat64 statbuf;
791 uint64_t size = -1ULL;
797 verify(nvlist_lookup_string(cnv,
798 ZPOOL_CONFIG_TYPE, &childtype) == 0);
801 * If this is a replacing or spare vdev, then
802 * get the real first child of the vdev.
804 if (strcmp(childtype,
805 VDEV_TYPE_REPLACING) == 0 ||
806 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
810 verify(nvlist_lookup_nvlist_array(cnv,
811 ZPOOL_CONFIG_CHILDREN, &rchild,
813 assert(rchildren == 2);
816 verify(nvlist_lookup_string(cnv,
821 verify(nvlist_lookup_string(cnv,
822 ZPOOL_CONFIG_PATH, &path) == 0);
825 * If we have a raidz/mirror that combines disks
826 * with files, report it as an error.
828 if (!dontreport && type != NULL &&
829 strcmp(type, childtype) != 0) {
835 "mismatched replication "
836 "level: %s contains both "
837 "files and devices\n"),
845 * According to stat(2), the value of 'st_size'
846 * is undefined for block devices and character
847 * devices. But there is no effective way to
848 * determine the real size in userland.
850 * Instead, we'll take advantage of an
851 * implementation detail of spec_size(). If the
852 * device is currently open, then we (should)
853 * return a valid size.
855 * If we still don't get a valid size (indicated
856 * by a size of 0 or MAXOFFSET_T), then ignore
857 * this device altogether.
859 if ((fd = open(path, O_RDONLY)) >= 0) {
860 err = fstat64(fd, &statbuf);
863 err = stat64(path, &statbuf);
867 statbuf.st_size == 0 ||
868 statbuf.st_size == MAXOFFSET_T)
871 size = statbuf.st_size;
874 * Also make sure that devices and
875 * slices have a consistent size. If
876 * they differ by a significant amount
877 * (~16MB) then report an error.
880 (vdev_size != -1ULL &&
881 (labs(size - vdev_size) >
888 "%s contains devices of "
889 "different sizes\n"),
902 * At this point, we have the replication of the last toplevel
903 * vdev in 'rep'. Compare it to 'lastrep' to see if its
906 if (lastrep.zprl_type != NULL) {
907 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
913 "mismatched replication level: "
914 "both %s and %s vdevs are "
916 lastrep.zprl_type, rep.zprl_type);
919 } else if (lastrep.zprl_parity != rep.zprl_parity) {
925 "mismatched replication level: "
926 "both %llu and %llu device parity "
927 "%s vdevs are present\n"),
933 } else if (lastrep.zprl_children != rep.zprl_children) {
939 "mismatched replication level: "
940 "both %llu-way and %llu-way %s "
941 "vdevs are present\n"),
942 lastrep.zprl_children,
959 * Check the replication level of the vdev spec against the current pool. Calls
960 * get_replication() to make sure the new spec is self-consistent. If the pool
961 * has a consistent replication level, then we ignore any errors. Otherwise,
962 * report any difference between the two.
965 check_replication(nvlist_t *config, nvlist_t *newroot)
969 replication_level_t *current = NULL, *new;
973 * If we have a current pool configuration, check to see if it's
974 * self-consistent. If not, simply return success.
976 if (config != NULL) {
979 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
981 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
985 * for spares there may be no children, and therefore no
986 * replication level to check
988 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
989 &child, &children) != 0) || (children == 0)) {
995 * If all we have is logs then there's no replication level to check.
997 if (num_logs(newroot) == children) {
1003 * Get the replication level of the new vdev spec, reporting any
1004 * inconsistencies found.
1006 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1012 * Check to see if the new vdev spec matches the replication level of
1016 if (current != NULL) {
1017 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1019 "mismatched replication level: pool uses %s "
1020 "and new vdev is %s\n"),
1021 current->zprl_type, new->zprl_type);
1023 } else if (current->zprl_parity != new->zprl_parity) {
1025 "mismatched replication level: pool uses %llu "
1026 "device parity and new vdev uses %llu\n"),
1027 current->zprl_parity, new->zprl_parity);
1029 } else if (current->zprl_children != new->zprl_children) {
1031 "mismatched replication level: pool uses %llu-way "
1032 "%s and new vdev uses %llu-way %s\n"),
1033 current->zprl_children, current->zprl_type,
1034 new->zprl_children, new->zprl_type);
1040 if (current != NULL)
1047 zero_label(char *path)
1049 const int size = 4096;
1053 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1054 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1055 path, strerror(errno));
1059 memset(buf, 0, size);
1060 err = write(fd, buf, size);
1061 (void) fdatasync(fd);
1065 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1066 "of '%s': %s\n"), size, path, strerror(errno));
1071 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1072 "of '%s'\n"), err, size, path);
1080 * Go through and find any whole disks in the vdev specification, labelling them
1081 * as appropriate. When constructing the vdev spec, we were unable to open this
1082 * device in order to provide a devid. Now that we have labelled the disk and
1083 * know that slice 0 is valid, we can construct the devid now.
1085 * If the disk was already labeled with an EFI label, we will have gotten the
1086 * devid already (because we were able to open the whole disk). Otherwise, we
1087 * need to get the devid after we label the disk.
1090 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1095 char devpath[MAXPATHLEN];
1096 char udevpath[MAXPATHLEN];
1098 struct stat64 statbuf;
1099 int is_exclusive = 0;
1103 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1105 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1106 &child, &children) != 0) {
1108 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1112 * We have a disk device. If this is a whole disk write
1113 * out the efi partition table, otherwise write zero's to
1114 * the first 4k of the partition. This is to ensure that
1115 * libblkid will not misidentify the partition due to a
1116 * magic value left by the previous filesystem.
1118 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1119 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1123 (void) zero_label(path);
1127 if (realpath(path, devpath) == NULL) {
1129 (void) fprintf(stderr,
1130 gettext("cannot resolve path '%s'\n"), path);
1135 * Remove any previously existing symlink from a udev path to
1136 * the device before labeling the disk. This makes
1137 * zpool_label_disk_wait() truly wait for the new link to show
1138 * up instead of returning if it finds an old link still in
1139 * place. Otherwise there is a window between when udev
1140 * deletes and recreates the link during which access attempts
1141 * will fail with ENOENT.
1143 strncpy(udevpath, path, MAXPATHLEN);
1144 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1146 fd = open(devpath, O_RDWR|O_EXCL);
1155 * If the partition exists, contains a valid spare label,
1156 * and is opened exclusively there is no need to partition
1157 * it. Hot spares have already been partitioned and are
1158 * held open exclusively by the kernel as a safety measure.
1160 * If the provided path is for a /dev/disk/ device its
1161 * symbolic link will be removed, partition table created,
1162 * and then block until udev creates the new link.
1164 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1165 ret = strncmp(udevpath,UDISK_ROOT,strlen(UDISK_ROOT));
1167 ret = lstat64(udevpath, &statbuf);
1168 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1169 (void) unlink(udevpath);
1172 if (zpool_label_disk(g_zfs, zhp,
1173 strrchr(devpath, '/') + 1) == -1)
1176 ret = zpool_label_disk_wait(udevpath, 1000);
1178 (void) fprintf(stderr, gettext("cannot "
1179 "resolve path '%s': %d\n"), udevpath, ret);
1183 (void) zero_label(udevpath);
1187 * Update the path to refer to the partition. The presence of
1188 * the 'whole_disk' field indicates to the CLI that we should
1189 * chop off the partition number when displaying the device in
1192 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1197 for (c = 0; c < children; c++)
1198 if ((ret = make_disks(zhp, child[c])) != 0)
1201 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1202 &child, &children) == 0)
1203 for (c = 0; c < children; c++)
1204 if ((ret = make_disks(zhp, child[c])) != 0)
1207 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1208 &child, &children) == 0)
1209 for (c = 0; c < children; c++)
1210 if ((ret = make_disks(zhp, child[c])) != 0)
1217 * Go through and find any devices that are in use. We rely on libdiskmgt for
1218 * the majority of this task.
1221 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1222 boolean_t replacing, boolean_t isspare)
1228 char buf[MAXPATHLEN];
1229 uint64_t wholedisk = B_FALSE;
1231 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1233 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1234 &child, &children) != 0) {
1236 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1237 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1238 verify(!nvlist_lookup_uint64(nv,
1239 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1242 * As a generic check, we look to see if this is a replace of a
1243 * hot spare within the same pool. If so, we allow it
1244 * regardless of what libblkid or zpool_in_use() says.
1247 (void) strlcpy(buf, path, sizeof (buf));
1249 ret = zfs_append_partition(buf, sizeof (buf));
1254 if (is_spare(config, buf))
1258 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1259 ret = check_device(path, force, isspare, wholedisk);
1261 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1262 ret = check_file(path, force, isspare);
1267 for (c = 0; c < children; c++)
1268 if ((ret = check_in_use(config, child[c], force,
1269 replacing, B_FALSE)) != 0)
1272 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1273 &child, &children) == 0)
1274 for (c = 0; c < children; c++)
1275 if ((ret = check_in_use(config, child[c], force,
1276 replacing, B_TRUE)) != 0)
1279 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1280 &child, &children) == 0)
1281 for (c = 0; c < children; c++)
1282 if ((ret = check_in_use(config, child[c], force,
1283 replacing, B_FALSE)) != 0)
1290 is_grouping(const char *type, int *mindev, int *maxdev)
1292 if (strncmp(type, "raidz", 5) == 0) {
1293 const char *p = type + 5;
1299 } else if (*p == '0') {
1300 return (NULL); /* no zero prefixes allowed */
1303 nparity = strtol(p, &end, 10);
1304 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1310 *mindev = nparity + 1;
1313 return (VDEV_TYPE_RAIDZ);
1319 if (strcmp(type, "mirror") == 0) {
1322 return (VDEV_TYPE_MIRROR);
1325 if (strcmp(type, "spare") == 0) {
1328 return (VDEV_TYPE_SPARE);
1331 if (strcmp(type, "log") == 0) {
1334 return (VDEV_TYPE_LOG);
1337 if (strcmp(type, "cache") == 0) {
1340 return (VDEV_TYPE_L2CACHE);
1347 * Construct a syntactically valid vdev specification,
1348 * and ensure that all devices and files exist and can be opened.
1349 * Note: we don't bother freeing anything in the error paths
1350 * because the program is just going to exit anyway.
1353 construct_spec(nvlist_t *props, int argc, char **argv)
1355 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1356 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1359 boolean_t seen_logs;
1369 seen_logs = B_FALSE;
1375 * If it's a mirror or raidz, the subsequent arguments are
1376 * its leaves -- until we encounter the next mirror or raidz.
1378 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1379 nvlist_t **child = NULL;
1380 int c, children = 0;
1382 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1383 if (spares != NULL) {
1384 (void) fprintf(stderr,
1385 gettext("invalid vdev "
1386 "specification: 'spare' can be "
1387 "specified only once\n"));
1393 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1395 (void) fprintf(stderr,
1396 gettext("invalid vdev "
1397 "specification: 'log' can be "
1398 "specified only once\n"));
1406 * A log is not a real grouping device.
1407 * We just set is_log and continue.
1412 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1413 if (l2cache != NULL) {
1414 (void) fprintf(stderr,
1415 gettext("invalid vdev "
1416 "specification: 'cache' can be "
1417 "specified only once\n"));
1424 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1425 (void) fprintf(stderr,
1426 gettext("invalid vdev "
1427 "specification: unsupported 'log' "
1428 "device: %s\n"), type);
1434 for (c = 1; c < argc; c++) {
1435 if (is_grouping(argv[c], NULL, NULL) != NULL)
1438 child = realloc(child,
1439 children * sizeof (nvlist_t *));
1442 if ((nv = make_leaf_vdev(props, argv[c], B_FALSE))
1445 child[children - 1] = nv;
1448 if (children < mindev) {
1449 (void) fprintf(stderr, gettext("invalid vdev "
1450 "specification: %s requires at least %d "
1451 "devices\n"), argv[0], mindev);
1455 if (children > maxdev) {
1456 (void) fprintf(stderr, gettext("invalid vdev "
1457 "specification: %s supports no more than "
1458 "%d devices\n"), argv[0], maxdev);
1465 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1469 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1471 nl2cache = children;
1474 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1476 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1478 verify(nvlist_add_uint64(nv,
1479 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1480 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1481 verify(nvlist_add_uint64(nv,
1482 ZPOOL_CONFIG_NPARITY,
1485 verify(nvlist_add_nvlist_array(nv,
1486 ZPOOL_CONFIG_CHILDREN, child,
1489 for (c = 0; c < children; c++)
1490 nvlist_free(child[c]);
1495 * We have a device. Pass off to make_leaf_vdev() to
1496 * construct the appropriate nvlist describing the vdev.
1498 if ((nv = make_leaf_vdev(props, argv[0], is_log)) == NULL)
1507 top = realloc(top, toplevels * sizeof (nvlist_t *));
1510 top[toplevels - 1] = nv;
1513 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1514 (void) fprintf(stderr, gettext("invalid vdev "
1515 "specification: at least one toplevel vdev must be "
1520 if (seen_logs && nlogs == 0) {
1521 (void) fprintf(stderr, gettext("invalid vdev specification: "
1522 "log requires at least 1 device\n"));
1527 * Finally, create nvroot and add all top-level vdevs to it.
1529 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1530 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1531 VDEV_TYPE_ROOT) == 0);
1532 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1533 top, toplevels) == 0);
1535 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1536 spares, nspares) == 0);
1538 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1539 l2cache, nl2cache) == 0);
1541 for (t = 0; t < toplevels; t++)
1542 nvlist_free(top[t]);
1543 for (t = 0; t < nspares; t++)
1544 nvlist_free(spares[t]);
1545 for (t = 0; t < nl2cache; t++)
1546 nvlist_free(l2cache[t]);
1557 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1558 splitflags_t flags, int argc, char **argv)
1560 nvlist_t *newroot = NULL, **child;
1564 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1565 (void) fprintf(stderr, gettext("Unable to build a "
1566 "pool from the specified devices\n"));
1570 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1571 nvlist_free(newroot);
1575 /* avoid any tricks in the spec */
1576 verify(nvlist_lookup_nvlist_array(newroot,
1577 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1578 for (c = 0; c < children; c++) {
1583 verify(nvlist_lookup_string(child[c],
1584 ZPOOL_CONFIG_PATH, &path) == 0);
1585 if ((type = is_grouping(path, &min, &max)) != NULL) {
1586 (void) fprintf(stderr, gettext("Cannot use "
1587 "'%s' as a device for splitting\n"), type);
1588 nvlist_free(newroot);
1594 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1595 if (newroot != NULL)
1596 nvlist_free(newroot);
1604 * Get and validate the contents of the given vdev specification. This ensures
1605 * that the nvlist returned is well-formed, that all the devices exist, and that
1606 * they are not currently in use by any other known consumer. The 'poolconfig'
1607 * parameter is the current configuration of the pool when adding devices
1608 * existing pool, and is used to perform additional checks, such as changing the
1609 * replication level of the pool. It can be 'NULL' to indicate that this is a
1610 * new pool. The 'force' flag controls whether devices should be forcefully
1611 * added, even if they appear in use.
1614 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1615 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1618 nvlist_t *poolconfig = NULL;
1622 * Construct the vdev specification. If this is successful, we know
1623 * that we have a valid specification, and that all devices can be
1626 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1629 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1633 * Validate each device to make sure that its not shared with another
1634 * subsystem. We do this even if 'force' is set, because there are some
1635 * uses (such as a dedicated dump device) that even '-f' cannot
1638 if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1639 nvlist_free(newroot);
1644 * Check the replication level of the given vdevs and report any errors
1645 * found. We include the existing pool spec, if any, as we need to
1646 * catch changes against the existing replication level.
1648 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1649 nvlist_free(newroot);
1654 * Run through the vdev specification and label any whole disks found.
1656 if (!dryrun && make_disks(zhp, newroot) != 0) {
1657 nvlist_free(newroot);