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.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2016 Intel Corporation.
29 * Functions to convert between a list of vdevs and an nvlist representing the
30 * configuration. Each entry in the list can be one of:
33 * disk=(path=..., devid=...)
42 * While the underlying implementation supports it, group vdevs cannot contain
43 * other group vdevs. All userland verification of devices is contained within
44 * this file. If successful, the nvlist returned can be passed directly to the
45 * kernel; we've done as much verification as possible in userland.
47 * Hot spares are a special case, and passed down as an array of disk vdevs, at
48 * the same level as the root of the vdev tree.
50 * The only function exported by this file is 'make_root_vdev'. The
51 * function performs several passes:
53 * 1. Construct the vdev specification. Performs syntax validation and
54 * makes sure each device is valid.
55 * 2. Check for devices in use. Using libblkid to make sure that no
56 * devices are also in use. Some can be overridden using the 'force'
57 * flag, others cannot.
58 * 3. Check for replication errors if the 'force' flag is not specified.
59 * validates that the replication level is consistent across the
61 * 4. Call libzfs to label any whole disks with an EFI label.
70 #include <libnvpair.h>
72 #include <scsi/scsi.h>
77 #include <sys/efi_partition.h>
80 #include <sys/mntent.h>
81 #include <uuid/uuid.h>
82 #include <blkid/blkid.h>
83 #include "zpool_util.h"
84 #include <sys/zfs_context.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.
94 typedef struct vdev_disk_db_entry
98 } vdev_disk_db_entry_t;
101 * Database of block devices that lie about physical sector sizes. The
102 * identification string must be precisely 24 characters to avoid false
105 static vdev_disk_db_entry_t vdev_disk_database[] = {
106 {"ATA ADATA SSD S396 3", 8192},
107 {"ATA APPLE SSD SM128E", 8192},
108 {"ATA APPLE SSD SM256E", 8192},
109 {"ATA APPLE SSD SM512E", 8192},
110 {"ATA APPLE SSD SM768E", 8192},
111 {"ATA C400-MTFDDAC064M", 8192},
112 {"ATA C400-MTFDDAC128M", 8192},
113 {"ATA C400-MTFDDAC256M", 8192},
114 {"ATA C400-MTFDDAC512M", 8192},
115 {"ATA Corsair Force 3 ", 8192},
116 {"ATA Corsair Force GS", 8192},
117 {"ATA INTEL SSDSA2CT04", 8192},
118 {"ATA INTEL SSDSA2BZ10", 8192},
119 {"ATA INTEL SSDSA2BZ20", 8192},
120 {"ATA INTEL SSDSA2BZ30", 8192},
121 {"ATA INTEL SSDSA2CW04", 8192},
122 {"ATA INTEL SSDSA2CW08", 8192},
123 {"ATA INTEL SSDSA2CW12", 8192},
124 {"ATA INTEL SSDSA2CW16", 8192},
125 {"ATA INTEL SSDSA2CW30", 8192},
126 {"ATA INTEL SSDSA2CW60", 8192},
127 {"ATA INTEL SSDSC2CT06", 8192},
128 {"ATA INTEL SSDSC2CT12", 8192},
129 {"ATA INTEL SSDSC2CT18", 8192},
130 {"ATA INTEL SSDSC2CT24", 8192},
131 {"ATA INTEL SSDSC2CW06", 8192},
132 {"ATA INTEL SSDSC2CW12", 8192},
133 {"ATA INTEL SSDSC2CW18", 8192},
134 {"ATA INTEL SSDSC2CW24", 8192},
135 {"ATA INTEL SSDSC2CW48", 8192},
136 {"ATA KINGSTON SH100S3", 8192},
137 {"ATA KINGSTON SH103S3", 8192},
138 {"ATA M4-CT064M4SSD2 ", 8192},
139 {"ATA M4-CT128M4SSD2 ", 8192},
140 {"ATA M4-CT256M4SSD2 ", 8192},
141 {"ATA M4-CT512M4SSD2 ", 8192},
142 {"ATA OCZ-AGILITY2 ", 8192},
143 {"ATA OCZ-AGILITY3 ", 8192},
144 {"ATA OCZ-VERTEX2 3.5 ", 8192},
145 {"ATA OCZ-VERTEX3 ", 8192},
146 {"ATA OCZ-VERTEX3 LT ", 8192},
147 {"ATA OCZ-VERTEX3 MI ", 8192},
148 {"ATA OCZ-VERTEX4 ", 8192},
149 {"ATA SAMSUNG MZ7WD120", 8192},
150 {"ATA SAMSUNG MZ7WD240", 8192},
151 {"ATA SAMSUNG MZ7WD480", 8192},
152 {"ATA SAMSUNG MZ7WD960", 8192},
153 {"ATA SAMSUNG SSD 830 ", 8192},
154 {"ATA Samsung SSD 840 ", 8192},
155 {"ATA SanDisk SSD U100", 8192},
156 {"ATA TOSHIBA THNSNH06", 8192},
157 {"ATA TOSHIBA THNSNH12", 8192},
158 {"ATA TOSHIBA THNSNH25", 8192},
159 {"ATA TOSHIBA THNSNH51", 8192},
160 {"ATA APPLE SSD TS064C", 4096},
161 {"ATA APPLE SSD TS128C", 4096},
162 {"ATA APPLE SSD TS256C", 4096},
163 {"ATA APPLE SSD TS512C", 4096},
164 {"ATA INTEL SSDSA2M040", 4096},
165 {"ATA INTEL SSDSA2M080", 4096},
166 {"ATA INTEL SSDSA2M160", 4096},
167 {"ATA INTEL SSDSC2MH12", 4096},
168 {"ATA INTEL SSDSC2MH25", 4096},
169 {"ATA OCZ CORE_SSD ", 4096},
170 {"ATA OCZ-VERTEX ", 4096},
171 {"ATA SAMSUNG MCCOE32G", 4096},
172 {"ATA SAMSUNG MCCOE64G", 4096},
173 {"ATA SAMSUNG SSD PM80", 4096},
174 /* Flash drives optimized for 4KB IOs on larger pages */
175 {"ATA INTEL SSDSC2BA10", 4096},
176 {"ATA INTEL SSDSC2BA20", 4096},
177 {"ATA INTEL SSDSC2BA40", 4096},
178 {"ATA INTEL SSDSC2BA80", 4096},
179 {"ATA INTEL SSDSC2BB08", 4096},
180 {"ATA INTEL SSDSC2BB12", 4096},
181 {"ATA INTEL SSDSC2BB16", 4096},
182 {"ATA INTEL SSDSC2BB24", 4096},
183 {"ATA INTEL SSDSC2BB30", 4096},
184 {"ATA INTEL SSDSC2BB40", 4096},
185 {"ATA INTEL SSDSC2BB48", 4096},
186 {"ATA INTEL SSDSC2BB60", 4096},
187 {"ATA INTEL SSDSC2BB80", 4096},
188 {"ATA INTEL SSDSC2BW24", 4096},
189 {"ATA INTEL SSDSC2BP24", 4096},
190 {"ATA INTEL SSDSC2BP48", 4096},
191 {"NA SmrtStorSDLKAE9W", 4096},
192 /* Imported from Open Solaris */
193 {"ATA MARVELL SD88SA02", 4096},
194 /* Advanced format Hard drives */
195 {"ATA Hitachi HDS5C303", 4096},
196 {"ATA SAMSUNG HD204UI ", 4096},
197 {"ATA ST2000DL004 HD20", 4096},
198 {"ATA WDC WD10EARS-00M", 4096},
199 {"ATA WDC WD10EARS-00S", 4096},
200 {"ATA WDC WD10EARS-00Z", 4096},
201 {"ATA WDC WD15EARS-00M", 4096},
202 {"ATA WDC WD15EARS-00S", 4096},
203 {"ATA WDC WD15EARS-00Z", 4096},
204 {"ATA WDC WD20EARS-00M", 4096},
205 {"ATA WDC WD20EARS-00S", 4096},
206 {"ATA WDC WD20EARS-00Z", 4096},
207 {"ATA WDC WD1600BEVT-0", 4096},
208 {"ATA WDC WD2500BEVT-0", 4096},
209 {"ATA WDC WD3200BEVT-0", 4096},
210 {"ATA WDC WD5000BEVT-0", 4096},
211 /* Virtual disks: Assume zvols with default volblocksize */
213 {"ATA QEMU HARDDISK ", 8192},
214 {"IET VIRTUAL-DISK ", 8192},
215 {"OI COMSTAR ", 8192},
216 {"SUN COMSTAR ", 8192},
217 {"NETAPP LUN ", 8192},
221 static const int vdev_disk_database_size =
222 sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
224 #define INQ_REPLY_LEN 96
225 #define INQ_CMD_LEN 6
228 check_sector_size_database(char *path, int *sector_size)
230 unsigned char inq_buff[INQ_REPLY_LEN];
231 unsigned char sense_buffer[32];
232 unsigned char inq_cmd_blk[INQ_CMD_LEN] =
233 {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
239 /* Prepare INQUIRY command */
240 memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
241 io_hdr.interface_id = 'S';
242 io_hdr.cmd_len = sizeof (inq_cmd_blk);
243 io_hdr.mx_sb_len = sizeof (sense_buffer);
244 io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
245 io_hdr.dxfer_len = INQ_REPLY_LEN;
246 io_hdr.dxferp = inq_buff;
247 io_hdr.cmdp = inq_cmd_blk;
248 io_hdr.sbp = sense_buffer;
249 io_hdr.timeout = 10; /* 10 milliseconds is ample time */
251 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
254 error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
261 if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
264 for (i = 0; i < vdev_disk_database_size; i++) {
265 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
268 *sector_size = vdev_disk_database[i].sector_size;
277 vdev_error(const char *fmt, ...)
282 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
284 (void) fprintf(stderr, gettext("use '-f' to override "
285 "the following errors:\n"));
287 (void) fprintf(stderr, gettext("the following errors "
288 "must be manually repaired:\n"));
293 (void) vfprintf(stderr, fmt, ap);
298 * Check that a file is valid. All we can do in this case is check that it's
299 * not in use by another pool, and not in use by swap.
302 check_file(const char *file, boolean_t force, boolean_t isspare)
310 if ((fd = open(file, O_RDONLY)) < 0)
313 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
317 case POOL_STATE_ACTIVE:
318 desc = gettext("active");
321 case POOL_STATE_EXPORTED:
322 desc = gettext("exported");
325 case POOL_STATE_POTENTIALLY_ACTIVE:
326 desc = gettext("potentially active");
330 desc = gettext("unknown");
335 * Allow hot spares to be shared between pools.
337 if (state == POOL_STATE_SPARE && isspare) {
343 if (state == POOL_STATE_ACTIVE ||
344 state == POOL_STATE_SPARE || !force) {
346 case POOL_STATE_SPARE:
347 vdev_error(gettext("%s is reserved as a hot "
348 "spare for pool %s\n"), file, name);
351 vdev_error(gettext("%s is part of %s pool "
352 "'%s'\n"), file, desc, name);
368 (void) fprintf(stderr, gettext("warning: device in use checking "
369 "failed: %s\n"), strerror(err));
373 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
378 /* No valid type detected device is safe to use */
379 value = blkid_get_tag_value(cache, "TYPE", path);
384 * If libblkid detects a ZFS device, we check the device
385 * using check_file() to see if it's safe. The one safe
386 * case is a spare device shared between multiple pools.
388 if (strcmp(value, "zfs_member") == 0) {
389 err = check_file(path, force, isspare);
395 vdev_error(gettext("%s contains a filesystem of "
396 "type '%s'\n"), path, value);
406 * Validate that a disk including all partitions are safe to use.
408 * For EFI labeled disks this can done relatively easily with the libefi
409 * library. The partition numbers are extracted from the label and used
410 * to generate the expected /dev/ paths. Each partition can then be
411 * checked for conflicts.
413 * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
414 * but due to the lack of a readily available libraries this scanning is
415 * not implemented. Instead only the device path as given is checked.
418 check_disk(const char *path, blkid_cache cache, int force,
419 boolean_t isspare, boolean_t iswholedisk)
422 char slice_path[MAXPATHLEN];
427 return (check_slice(path, cache, force, isspare));
429 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
435 * Expected to fail for non-EFI labled disks. Just check the device
436 * as given and do not attempt to detect and scan partitions.
438 err = efi_alloc_and_read(fd, &vtoc);
441 return (check_slice(path, cache, force, isspare));
445 * The primary efi partition label is damaged however the secondary
446 * label at the end of the device is intact. Rather than use this
447 * label we should play it safe and treat this as a non efi device.
449 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
454 /* Partitions will now be created using the backup */
457 vdev_error(gettext("%s contains a corrupt primary "
458 "EFI label.\n"), path);
463 for (i = 0; i < vtoc->efi_nparts; i++) {
465 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
466 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
469 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
470 (void) snprintf(slice_path, sizeof (slice_path),
471 "%s%s%d", path, "-part", i+1);
473 (void) snprintf(slice_path, sizeof (slice_path),
474 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
477 err = check_slice(slice_path, cache, force, isspare);
489 check_device(const char *path, boolean_t force,
490 boolean_t isspare, boolean_t iswholedisk)
495 error = blkid_get_cache(&cache, NULL);
501 error = check_disk(path, cache, force, isspare, iswholedisk);
502 blkid_put_cache(cache);
508 * By "whole disk" we mean an entire physical disk (something we can
509 * label, toggle the write cache on, etc.) as opposed to the full
510 * capacity of a pseudo-device such as lofi or did. We act as if we
511 * are labeling the disk, which should be a pretty good test of whether
512 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
516 is_whole_disk(const char *path)
518 struct dk_gpt *label;
521 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
523 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
533 * This may be a shorthand device path or it could be total gibberish.
534 * Check to see if it is a known device available in zfs_vdev_paths.
535 * As part of this check, see if we've been given an entire disk
536 * (minus the slice number).
539 is_shorthand_path(const char *arg, char *path, size_t path_size,
540 struct stat64 *statbuf, boolean_t *wholedisk)
544 error = zfs_resolve_shortname(arg, path, path_size);
546 *wholedisk = is_whole_disk(path);
547 if (*wholedisk || (stat64(path, statbuf) == 0))
551 strlcpy(path, arg, path_size);
552 memset(statbuf, 0, sizeof (*statbuf));
553 *wholedisk = B_FALSE;
559 * Determine if the given path is a hot spare within the given configuration.
560 * If no configuration is given we rely solely on the label.
563 is_spare(nvlist_t *config, const char *path)
569 uint64_t guid, spareguid;
575 if ((fd = open(path, O_RDONLY)) < 0)
578 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
580 state != POOL_STATE_SPARE ||
581 zpool_read_label(fd, &label, NULL) != 0) {
589 if (config == NULL) {
594 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
597 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
599 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
600 &spares, &nspares) == 0) {
601 for (i = 0; i < nspares; i++) {
602 verify(nvlist_lookup_uint64(spares[i],
603 ZPOOL_CONFIG_GUID, &spareguid) == 0);
604 if (spareguid == guid)
613 * Create a leaf vdev. Determine if this is a file or a device. If it's a
614 * device, fill in the device id to make a complete nvlist. Valid forms for a
617 * /dev/xxx Complete disk path
618 * /xxx Full path to file
619 * xxx Shorthand for <zfs_vdev_paths>/xxx
622 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
624 char path[MAXPATHLEN];
625 struct stat64 statbuf;
626 nvlist_t *vdev = NULL;
628 boolean_t wholedisk = B_FALSE;
633 * Determine what type of vdev this is, and put the full path into
634 * 'path'. We detect whether this is a device of file afterwards by
635 * checking the st_mode of the file.
639 * Complete device or file path. Exact type is determined by
640 * examining the file descriptor afterwards. Symbolic links
641 * are resolved to their real paths for the is_whole_disk()
642 * and S_ISBLK/S_ISREG type checks. However, we are careful
643 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
644 * can leverage udev's persistent device labels.
646 if (realpath(arg, path) == NULL) {
647 (void) fprintf(stderr,
648 gettext("cannot resolve path '%s'\n"), arg);
652 wholedisk = is_whole_disk(path);
653 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
654 (void) fprintf(stderr,
655 gettext("cannot open '%s': %s\n"),
656 path, strerror(errno));
660 /* After is_whole_disk() check restore original passed path */
661 strlcpy(path, arg, sizeof (path));
663 err = is_shorthand_path(arg, path, sizeof (path),
664 &statbuf, &wholedisk);
667 * If we got ENOENT, then the user gave us
668 * gibberish, so try to direct them with a
669 * reasonable error message. Otherwise,
670 * regurgitate strerror() since it's the best we
674 (void) fprintf(stderr,
675 gettext("cannot open '%s': no such "
676 "device in %s\n"), arg, DISK_ROOT);
677 (void) fprintf(stderr,
678 gettext("must be a full path or "
679 "shorthand device name\n"));
682 (void) fprintf(stderr,
683 gettext("cannot open '%s': %s\n"),
684 path, strerror(errno));
691 * Determine whether this is a device or a file.
693 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
694 type = VDEV_TYPE_DISK;
695 } else if (S_ISREG(statbuf.st_mode)) {
696 type = VDEV_TYPE_FILE;
698 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
699 "block device or regular file\n"), path);
704 * Finally, we have the complete device or file, and we know that it is
705 * acceptable to use. Construct the nvlist to describe this vdev. All
706 * vdevs have a 'path' element, and devices also have a 'devid' element.
708 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
709 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
710 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
711 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
712 if (strcmp(type, VDEV_TYPE_DISK) == 0)
713 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
714 (uint64_t)wholedisk) == 0);
717 * Override defaults if custom properties are provided.
722 if (nvlist_lookup_string(props,
723 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
724 zfs_nicestrtonum(NULL, value, &ashift);
728 * If the device is known to incorrectly report its physical sector
729 * size explicitly provide the known correct value.
734 if (check_sector_size_database(path, §or_size) == B_TRUE)
735 ashift = highbit64(sector_size) - 1;
739 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
745 * Go through and verify the replication level of the pool is consistent.
746 * Performs the following checks:
748 * For the new spec, verifies that devices in mirrors and raidz are the
751 * If the current configuration already has inconsistent replication
752 * levels, ignore any other potential problems in the new spec.
754 * Otherwise, make sure that the current spec (if there is one) and the new
755 * spec have consistent replication levels.
757 typedef struct replication_level {
759 uint64_t zprl_children;
760 uint64_t zprl_parity;
761 } replication_level_t;
763 #define ZPOOL_FUZZ (16 * 1024 * 1024)
766 * Given a list of toplevel vdevs, return the current replication level. If
767 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
768 * an error message will be displayed for each self-inconsistent vdev.
770 static replication_level_t *
771 get_replication(nvlist_t *nvroot, boolean_t fatal)
779 replication_level_t lastrep = { 0 }, rep, *ret;
780 boolean_t dontreport;
782 ret = safe_malloc(sizeof (replication_level_t));
784 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
785 &top, &toplevels) == 0);
787 lastrep.zprl_type = NULL;
788 for (t = 0; t < toplevels; t++) {
789 uint64_t is_log = B_FALSE;
794 * For separate logs we ignore the top level vdev replication
797 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
801 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
803 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
804 &child, &children) != 0) {
806 * This is a 'file' or 'disk' vdev.
808 rep.zprl_type = type;
809 rep.zprl_children = 1;
815 * This is a mirror or RAID-Z vdev. Go through and make
816 * sure the contents are all the same (files vs. disks),
817 * keeping track of the number of elements in the
820 * We also check that the size of each vdev (if it can
821 * be determined) is the same.
823 rep.zprl_type = type;
824 rep.zprl_children = 0;
826 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
827 verify(nvlist_lookup_uint64(nv,
828 ZPOOL_CONFIG_NPARITY,
829 &rep.zprl_parity) == 0);
830 assert(rep.zprl_parity != 0);
836 * The 'dontreport' variable indicates that we've
837 * already reported an error for this spec, so don't
838 * bother doing it again.
843 for (c = 0; c < children; c++) {
844 nvlist_t *cnv = child[c];
846 struct stat64 statbuf;
847 uint64_t size = -1ULL;
853 verify(nvlist_lookup_string(cnv,
854 ZPOOL_CONFIG_TYPE, &childtype) == 0);
857 * If this is a replacing or spare vdev, then
858 * get the real first child of the vdev.
860 if (strcmp(childtype,
861 VDEV_TYPE_REPLACING) == 0 ||
862 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
866 verify(nvlist_lookup_nvlist_array(cnv,
867 ZPOOL_CONFIG_CHILDREN, &rchild,
869 assert(rchildren == 2);
872 verify(nvlist_lookup_string(cnv,
877 verify(nvlist_lookup_string(cnv,
878 ZPOOL_CONFIG_PATH, &path) == 0);
881 * If we have a raidz/mirror that combines disks
882 * with files, report it as an error.
884 if (!dontreport && type != NULL &&
885 strcmp(type, childtype) != 0) {
891 "mismatched replication "
892 "level: %s contains both "
893 "files and devices\n"),
901 * According to stat(2), the value of 'st_size'
902 * is undefined for block devices and character
903 * devices. But there is no effective way to
904 * determine the real size in userland.
906 * Instead, we'll take advantage of an
907 * implementation detail of spec_size(). If the
908 * device is currently open, then we (should)
909 * return a valid size.
911 * If we still don't get a valid size (indicated
912 * by a size of 0 or MAXOFFSET_T), then ignore
913 * this device altogether.
915 if ((fd = open(path, O_RDONLY)) >= 0) {
916 err = fstat64(fd, &statbuf);
919 err = stat64(path, &statbuf);
923 statbuf.st_size == 0 ||
924 statbuf.st_size == MAXOFFSET_T)
927 size = statbuf.st_size;
930 * Also make sure that devices and
931 * slices have a consistent size. If
932 * they differ by a significant amount
933 * (~16MB) then report an error.
936 (vdev_size != -1ULL &&
937 (labs(size - vdev_size) >
944 "%s contains devices of "
945 "different sizes\n"),
958 * At this point, we have the replication of the last toplevel
959 * vdev in 'rep'. Compare it to 'lastrep' to see if its
962 if (lastrep.zprl_type != NULL) {
963 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
969 "mismatched replication level: "
970 "both %s and %s vdevs are "
972 lastrep.zprl_type, rep.zprl_type);
975 } else if (lastrep.zprl_parity != rep.zprl_parity) {
981 "mismatched replication level: "
982 "both %llu and %llu device parity "
983 "%s vdevs are present\n"),
989 } else if (lastrep.zprl_children != rep.zprl_children) {
995 "mismatched replication level: "
996 "both %llu-way and %llu-way %s "
997 "vdevs are present\n"),
998 lastrep.zprl_children,
1015 * Check the replication level of the vdev spec against the current pool. Calls
1016 * get_replication() to make sure the new spec is self-consistent. If the pool
1017 * has a consistent replication level, then we ignore any errors. Otherwise,
1018 * report any difference between the two.
1021 check_replication(nvlist_t *config, nvlist_t *newroot)
1025 replication_level_t *current = NULL, *new;
1029 * If we have a current pool configuration, check to see if it's
1030 * self-consistent. If not, simply return success.
1032 if (config != NULL) {
1035 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1037 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1041 * for spares there may be no children, and therefore no
1042 * replication level to check
1044 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1045 &child, &children) != 0) || (children == 0)) {
1051 * If all we have is logs then there's no replication level to check.
1053 if (num_logs(newroot) == children) {
1059 * Get the replication level of the new vdev spec, reporting any
1060 * inconsistencies found.
1062 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1068 * Check to see if the new vdev spec matches the replication level of
1072 if (current != NULL) {
1073 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1075 "mismatched replication level: pool uses %s "
1076 "and new vdev is %s\n"),
1077 current->zprl_type, new->zprl_type);
1079 } else if (current->zprl_parity != new->zprl_parity) {
1081 "mismatched replication level: pool uses %llu "
1082 "device parity and new vdev uses %llu\n"),
1083 current->zprl_parity, new->zprl_parity);
1085 } else if (current->zprl_children != new->zprl_children) {
1087 "mismatched replication level: pool uses %llu-way "
1088 "%s and new vdev uses %llu-way %s\n"),
1089 current->zprl_children, current->zprl_type,
1090 new->zprl_children, new->zprl_type);
1096 if (current != NULL)
1103 zero_label(char *path)
1105 const int size = 4096;
1109 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1110 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1111 path, strerror(errno));
1115 memset(buf, 0, size);
1116 err = write(fd, buf, size);
1117 (void) fdatasync(fd);
1121 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1122 "of '%s': %s\n"), size, path, strerror(errno));
1127 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1128 "of '%s'\n"), err, size, path);
1136 * Go through and find any whole disks in the vdev specification, labelling them
1137 * as appropriate. When constructing the vdev spec, we were unable to open this
1138 * device in order to provide a devid. Now that we have labelled the disk and
1139 * know that slice 0 is valid, we can construct the devid now.
1141 * If the disk was already labeled with an EFI label, we will have gotten the
1142 * devid already (because we were able to open the whole disk). Otherwise, we
1143 * need to get the devid after we label the disk.
1146 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1151 char devpath[MAXPATHLEN];
1152 char udevpath[MAXPATHLEN];
1154 struct stat64 statbuf;
1155 int is_exclusive = 0;
1159 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1161 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1162 &child, &children) != 0) {
1164 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1168 * We have a disk device. If this is a whole disk write
1169 * out the efi partition table, otherwise write zero's to
1170 * the first 4k of the partition. This is to ensure that
1171 * libblkid will not misidentify the partition due to a
1172 * magic value left by the previous filesystem.
1174 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1175 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1180 * Update device id string for mpath nodes (Linux only)
1182 if (is_mpath_whole_disk(path))
1183 update_vdev_config_dev_strs(nv);
1185 (void) zero_label(path);
1189 if (realpath(path, devpath) == NULL) {
1191 (void) fprintf(stderr,
1192 gettext("cannot resolve path '%s'\n"), path);
1197 * Remove any previously existing symlink from a udev path to
1198 * the device before labeling the disk. This ensures that
1199 * only newly created links are used. Otherwise there is a
1200 * window between when udev deletes and recreates the link
1201 * during which access attempts will fail with ENOENT.
1203 strlcpy(udevpath, path, MAXPATHLEN);
1204 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1206 fd = open(devpath, O_RDWR|O_EXCL);
1215 * If the partition exists, contains a valid spare label,
1216 * and is opened exclusively there is no need to partition
1217 * it. Hot spares have already been partitioned and are
1218 * held open exclusively by the kernel as a safety measure.
1220 * If the provided path is for a /dev/disk/ device its
1221 * symbolic link will be removed, partition table created,
1222 * and then block until udev creates the new link.
1224 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1225 char *devnode = strrchr(devpath, '/') + 1;
1227 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1229 ret = lstat64(udevpath, &statbuf);
1230 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1231 (void) unlink(udevpath);
1235 * When labeling a pool the raw device node name
1236 * is provided as it appears under /dev/.
1238 if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1242 * Wait for udev to signal the device is available
1243 * by the provided path.
1245 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1247 (void) fprintf(stderr,
1248 gettext("missing link: %s was "
1249 "partitioned but %s is missing\n"),
1254 ret = zero_label(udevpath);
1260 * Update the path to refer to the partition. The presence of
1261 * the 'whole_disk' field indicates to the CLI that we should
1262 * chop off the partition number when displaying the device in
1265 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1268 * Update device id strings for whole disks (Linux only)
1270 update_vdev_config_dev_strs(nv);
1275 for (c = 0; c < children; c++)
1276 if ((ret = make_disks(zhp, child[c])) != 0)
1279 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1280 &child, &children) == 0)
1281 for (c = 0; c < children; c++)
1282 if ((ret = make_disks(zhp, child[c])) != 0)
1285 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1286 &child, &children) == 0)
1287 for (c = 0; c < children; c++)
1288 if ((ret = make_disks(zhp, child[c])) != 0)
1295 * Go through and find any devices that are in use. We rely on libdiskmgt for
1296 * the majority of this task.
1299 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1300 boolean_t replacing, boolean_t isspare)
1306 char buf[MAXPATHLEN];
1307 uint64_t wholedisk = B_FALSE;
1308 boolean_t anyinuse = B_FALSE;
1310 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1312 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1313 &child, &children) != 0) {
1315 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1316 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1317 verify(!nvlist_lookup_uint64(nv,
1318 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1321 * As a generic check, we look to see if this is a replace of a
1322 * hot spare within the same pool. If so, we allow it
1323 * regardless of what libblkid or zpool_in_use() says.
1326 (void) strlcpy(buf, path, sizeof (buf));
1328 ret = zfs_append_partition(buf, sizeof (buf));
1333 if (is_spare(config, buf))
1337 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1338 ret = check_device(path, force, isspare, wholedisk);
1340 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1341 ret = check_file(path, force, isspare);
1346 for (c = 0; c < children; c++)
1347 if (is_device_in_use(config, child[c], force, replacing,
1351 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1352 &child, &children) == 0)
1353 for (c = 0; c < children; c++)
1354 if (is_device_in_use(config, child[c], force, replacing,
1358 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1359 &child, &children) == 0)
1360 for (c = 0; c < children; c++)
1361 if (is_device_in_use(config, child[c], force, replacing,
1369 is_grouping(const char *type, int *mindev, int *maxdev)
1371 if (strncmp(type, "raidz", 5) == 0) {
1372 const char *p = type + 5;
1378 } else if (*p == '0') {
1379 return (NULL); /* no zero prefixes allowed */
1382 nparity = strtol(p, &end, 10);
1383 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1389 *mindev = nparity + 1;
1392 return (VDEV_TYPE_RAIDZ);
1398 if (strcmp(type, "mirror") == 0) {
1401 return (VDEV_TYPE_MIRROR);
1404 if (strcmp(type, "spare") == 0) {
1407 return (VDEV_TYPE_SPARE);
1410 if (strcmp(type, "log") == 0) {
1413 return (VDEV_TYPE_LOG);
1416 if (strcmp(type, "cache") == 0) {
1419 return (VDEV_TYPE_L2CACHE);
1426 * Construct a syntactically valid vdev specification,
1427 * and ensure that all devices and files exist and can be opened.
1428 * Note: we don't bother freeing anything in the error paths
1429 * because the program is just going to exit anyway.
1432 construct_spec(nvlist_t *props, int argc, char **argv)
1434 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1435 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1438 boolean_t seen_logs;
1448 seen_logs = B_FALSE;
1455 * If it's a mirror or raidz, the subsequent arguments are
1456 * its leaves -- until we encounter the next mirror or raidz.
1458 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1459 nvlist_t **child = NULL;
1460 int c, children = 0;
1462 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1463 if (spares != NULL) {
1464 (void) fprintf(stderr,
1465 gettext("invalid vdev "
1466 "specification: 'spare' can be "
1467 "specified only once\n"));
1473 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1475 (void) fprintf(stderr,
1476 gettext("invalid vdev "
1477 "specification: 'log' can be "
1478 "specified only once\n"));
1486 * A log is not a real grouping device.
1487 * We just set is_log and continue.
1492 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1493 if (l2cache != NULL) {
1494 (void) fprintf(stderr,
1495 gettext("invalid vdev "
1496 "specification: 'cache' can be "
1497 "specified only once\n"));
1504 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1505 (void) fprintf(stderr,
1506 gettext("invalid vdev "
1507 "specification: unsupported 'log' "
1508 "device: %s\n"), type);
1514 for (c = 1; c < argc; c++) {
1515 if (is_grouping(argv[c], NULL, NULL) != NULL)
1518 child = realloc(child,
1519 children * sizeof (nvlist_t *));
1522 if ((nv = make_leaf_vdev(props, argv[c],
1523 B_FALSE)) == NULL) {
1524 for (c = 0; c < children - 1; c++)
1525 nvlist_free(child[c]);
1530 child[children - 1] = nv;
1533 if (children < mindev) {
1534 (void) fprintf(stderr, gettext("invalid vdev "
1535 "specification: %s requires at least %d "
1536 "devices\n"), argv[0], mindev);
1537 for (c = 0; c < children; c++)
1538 nvlist_free(child[c]);
1543 if (children > maxdev) {
1544 (void) fprintf(stderr, gettext("invalid vdev "
1545 "specification: %s supports no more than "
1546 "%d devices\n"), argv[0], maxdev);
1547 for (c = 0; c < children; c++)
1548 nvlist_free(child[c]);
1556 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1560 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1562 nl2cache = children;
1565 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1567 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1569 verify(nvlist_add_uint64(nv,
1570 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1571 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1572 verify(nvlist_add_uint64(nv,
1573 ZPOOL_CONFIG_NPARITY,
1576 verify(nvlist_add_nvlist_array(nv,
1577 ZPOOL_CONFIG_CHILDREN, child,
1580 for (c = 0; c < children; c++)
1581 nvlist_free(child[c]);
1586 * We have a device. Pass off to make_leaf_vdev() to
1587 * construct the appropriate nvlist describing the vdev.
1589 if ((nv = make_leaf_vdev(props, argv[0],
1600 top = realloc(top, toplevels * sizeof (nvlist_t *));
1603 top[toplevels - 1] = nv;
1606 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1607 (void) fprintf(stderr, gettext("invalid vdev "
1608 "specification: at least one toplevel vdev must be "
1613 if (seen_logs && nlogs == 0) {
1614 (void) fprintf(stderr, gettext("invalid vdev specification: "
1615 "log requires at least 1 device\n"));
1620 * Finally, create nvroot and add all top-level vdevs to it.
1622 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1623 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1624 VDEV_TYPE_ROOT) == 0);
1625 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1626 top, toplevels) == 0);
1628 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1629 spares, nspares) == 0);
1631 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1632 l2cache, nl2cache) == 0);
1635 for (t = 0; t < toplevels; t++)
1636 nvlist_free(top[t]);
1637 for (t = 0; t < nspares; t++)
1638 nvlist_free(spares[t]);
1639 for (t = 0; t < nl2cache; t++)
1640 nvlist_free(l2cache[t]);
1650 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1651 splitflags_t flags, int argc, char **argv)
1653 nvlist_t *newroot = NULL, **child;
1657 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1658 (void) fprintf(stderr, gettext("Unable to build a "
1659 "pool from the specified devices\n"));
1663 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1664 nvlist_free(newroot);
1668 /* avoid any tricks in the spec */
1669 verify(nvlist_lookup_nvlist_array(newroot,
1670 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1671 for (c = 0; c < children; c++) {
1676 verify(nvlist_lookup_string(child[c],
1677 ZPOOL_CONFIG_PATH, &path) == 0);
1678 if ((type = is_grouping(path, &min, &max)) != NULL) {
1679 (void) fprintf(stderr, gettext("Cannot use "
1680 "'%s' as a device for splitting\n"), type);
1681 nvlist_free(newroot);
1687 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1688 nvlist_free(newroot);
1696 * Get and validate the contents of the given vdev specification. This ensures
1697 * that the nvlist returned is well-formed, that all the devices exist, and that
1698 * they are not currently in use by any other known consumer. The 'poolconfig'
1699 * parameter is the current configuration of the pool when adding devices
1700 * existing pool, and is used to perform additional checks, such as changing the
1701 * replication level of the pool. It can be 'NULL' to indicate that this is a
1702 * new pool. The 'force' flag controls whether devices should be forcefully
1703 * added, even if they appear in use.
1706 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1707 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1710 nvlist_t *poolconfig = NULL;
1714 * Construct the vdev specification. If this is successful, we know
1715 * that we have a valid specification, and that all devices can be
1718 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1721 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1722 nvlist_free(newroot);
1727 * Validate each device to make sure that its not shared with another
1728 * subsystem. We do this even if 'force' is set, because there are some
1729 * uses (such as a dedicated dump device) that even '-f' cannot
1732 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1733 nvlist_free(newroot);
1738 * Check the replication level of the given vdevs and report any errors
1739 * found. We include the existing pool spec, if any, as we need to
1740 * catch changes against the existing replication level.
1742 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1743 nvlist_free(newroot);
1748 * Run through the vdev specification and label any whole disks found.
1750 if (!dryrun && make_disks(zhp, newroot) != 0) {
1751 nvlist_free(newroot);