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 ADATA SSD S396 3", 8192},
110 {"ATA APPLE SSD SM128E", 8192},
111 {"ATA APPLE SSD SM256E", 8192},
112 {"ATA APPLE SSD SM512E", 8192},
113 {"ATA APPLE SSD SM768E", 8192},
114 {"ATA C400-MTFDDAC064M", 8192},
115 {"ATA C400-MTFDDAC128M", 8192},
116 {"ATA C400-MTFDDAC256M", 8192},
117 {"ATA C400-MTFDDAC512M", 8192},
118 {"ATA Corsair Force 3 ", 8192},
119 {"ATA Corsair Force GS", 8192},
120 {"ATA INTEL SSDSA2CT04", 8192},
121 {"ATA INTEL SSDSA2BZ10", 8192},
122 {"ATA INTEL SSDSA2BZ20", 8192},
123 {"ATA INTEL SSDSA2BZ30", 8192},
124 {"ATA INTEL SSDSA2CW04", 8192},
125 {"ATA INTEL SSDSA2CW08", 8192},
126 {"ATA INTEL SSDSA2CW12", 8192},
127 {"ATA INTEL SSDSA2CW16", 8192},
128 {"ATA INTEL SSDSA2CW30", 8192},
129 {"ATA INTEL SSDSA2CW60", 8192},
130 {"ATA INTEL SSDSC2CT06", 8192},
131 {"ATA INTEL SSDSC2CT12", 8192},
132 {"ATA INTEL SSDSC2CT18", 8192},
133 {"ATA INTEL SSDSC2CT24", 8192},
134 {"ATA INTEL SSDSC2CW06", 8192},
135 {"ATA INTEL SSDSC2CW12", 8192},
136 {"ATA INTEL SSDSC2CW18", 8192},
137 {"ATA INTEL SSDSC2CW24", 8192},
138 {"ATA INTEL SSDSC2CW48", 8192},
139 {"ATA KINGSTON SH100S3", 8192},
140 {"ATA KINGSTON SH103S3", 8192},
141 {"ATA M4-CT064M4SSD2 ", 8192},
142 {"ATA M4-CT128M4SSD2 ", 8192},
143 {"ATA M4-CT256M4SSD2 ", 8192},
144 {"ATA M4-CT512M4SSD2 ", 8192},
145 {"ATA OCZ-AGILITY2 ", 8192},
146 {"ATA OCZ-AGILITY3 ", 8192},
147 {"ATA OCZ-VERTEX2 3.5 ", 8192},
148 {"ATA OCZ-VERTEX3 ", 8192},
149 {"ATA OCZ-VERTEX3 LT ", 8192},
150 {"ATA OCZ-VERTEX3 MI ", 8192},
151 {"ATA OCZ-VERTEX4 ", 8192},
152 {"ATA SAMSUNG MZ7WD120", 8192},
153 {"ATA SAMSUNG MZ7WD240", 8192},
154 {"ATA SAMSUNG MZ7WD480", 8192},
155 {"ATA SAMSUNG MZ7WD960", 8192},
156 {"ATA SAMSUNG SSD 830 ", 8192},
157 {"ATA Samsung SSD 840 ", 8192},
158 {"ATA SanDisk SSD U100", 8192},
159 {"ATA TOSHIBA THNSNH06", 8192},
160 {"ATA TOSHIBA THNSNH12", 8192},
161 {"ATA TOSHIBA THNSNH25", 8192},
162 {"ATA TOSHIBA THNSNH51", 8192},
163 {"ATA APPLE SSD TS064C", 4096},
164 {"ATA APPLE SSD TS128C", 4096},
165 {"ATA APPLE SSD TS256C", 4096},
166 {"ATA APPLE SSD TS512C", 4096},
167 {"ATA INTEL SSDSA2M040", 4096},
168 {"ATA INTEL SSDSA2M080", 4096},
169 {"ATA INTEL SSDSA2M160", 4096},
170 {"ATA INTEL SSDSC2MH12", 4096},
171 {"ATA INTEL SSDSC2MH25", 4096},
172 {"ATA OCZ CORE_SSD ", 4096},
173 {"ATA OCZ-VERTEX ", 4096},
174 {"ATA SAMSUNG MCCOE32G", 4096},
175 {"ATA SAMSUNG MCCOE64G", 4096},
176 {"ATA SAMSUNG SSD PM80", 4096},
177 /* Flash drives optimized for 4KB IOs on larger pages */
178 {"ATA INTEL SSDSC2BA10", 4096},
179 {"ATA INTEL SSDSC2BA20", 4096},
180 {"ATA INTEL SSDSC2BA40", 4096},
181 {"ATA INTEL SSDSC2BA80", 4096},
182 {"ATA INTEL SSDSC2BB08", 4096},
183 {"ATA INTEL SSDSC2BB12", 4096},
184 {"ATA INTEL SSDSC2BB16", 4096},
185 {"ATA INTEL SSDSC2BB24", 4096},
186 {"ATA INTEL SSDSC2BB30", 4096},
187 {"ATA INTEL SSDSC2BB40", 4096},
188 {"ATA INTEL SSDSC2BB48", 4096},
189 {"ATA INTEL SSDSC2BB60", 4096},
190 {"ATA INTEL SSDSC2BB80", 4096},
191 {"ATA INTEL SSDSC2BW24", 4096},
192 {"ATA INTEL SSDSC2BP24", 4096},
193 {"ATA INTEL SSDSC2BP48", 4096},
194 {"NA SmrtStorSDLKAE9W", 4096},
195 /* Imported from Open Solaris */
196 {"ATA MARVELL SD88SA02", 4096},
197 /* Advanced format Hard drives */
198 {"ATA Hitachi HDS5C303", 4096},
199 {"ATA SAMSUNG HD204UI ", 4096},
200 {"ATA ST2000DL004 HD20", 4096},
201 {"ATA WDC WD10EARS-00M", 4096},
202 {"ATA WDC WD10EARS-00S", 4096},
203 {"ATA WDC WD10EARS-00Z", 4096},
204 {"ATA WDC WD15EARS-00M", 4096},
205 {"ATA WDC WD15EARS-00S", 4096},
206 {"ATA WDC WD15EARS-00Z", 4096},
207 {"ATA WDC WD20EARS-00M", 4096},
208 {"ATA WDC WD20EARS-00S", 4096},
209 {"ATA WDC WD20EARS-00Z", 4096},
210 {"ATA WDC WD1600BEVT-0", 4096},
211 {"ATA WDC WD2500BEVT-0", 4096},
212 {"ATA WDC WD3200BEVT-0", 4096},
213 {"ATA WDC WD5000BEVT-0", 4096},
214 /* Virtual disks: Assume zvols with default volblocksize */
216 {"ATA QEMU HARDDISK ", 8192},
217 {"IET VIRTUAL-DISK ", 8192},
218 {"OI COMSTAR ", 8192},
219 {"SUN COMSTAR ", 8192},
220 {"NETAPP LUN ", 8192},
224 static const int vdev_disk_database_size =
225 sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
227 #define INQ_REPLY_LEN 96
228 #define INQ_CMD_LEN 6
231 check_sector_size_database(char *path, int *sector_size)
233 unsigned char inq_buff[INQ_REPLY_LEN];
234 unsigned char sense_buffer[32];
235 unsigned char inq_cmd_blk[INQ_CMD_LEN] =
236 {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
242 /* Prepare INQUIRY command */
243 memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
244 io_hdr.interface_id = 'S';
245 io_hdr.cmd_len = sizeof (inq_cmd_blk);
246 io_hdr.mx_sb_len = sizeof (sense_buffer);
247 io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
248 io_hdr.dxfer_len = INQ_REPLY_LEN;
249 io_hdr.dxferp = inq_buff;
250 io_hdr.cmdp = inq_cmd_blk;
251 io_hdr.sbp = sense_buffer;
252 io_hdr.timeout = 10; /* 10 milliseconds is ample time */
254 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
257 error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
264 if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
267 for (i = 0; i < vdev_disk_database_size; i++) {
268 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
271 *sector_size = vdev_disk_database[i].sector_size;
280 vdev_error(const char *fmt, ...)
285 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
287 (void) fprintf(stderr, gettext("use '-f' to override "
288 "the following errors:\n"));
290 (void) fprintf(stderr, gettext("the following errors "
291 "must be manually repaired:\n"));
296 (void) vfprintf(stderr, fmt, ap);
301 * Check that a file is valid. All we can do in this case is check that it's
302 * not in use by another pool, and not in use by swap.
305 check_file(const char *file, boolean_t force, boolean_t isspare)
313 if ((fd = open(file, O_RDONLY)) < 0)
316 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
320 case POOL_STATE_ACTIVE:
321 desc = gettext("active");
324 case POOL_STATE_EXPORTED:
325 desc = gettext("exported");
328 case POOL_STATE_POTENTIALLY_ACTIVE:
329 desc = gettext("potentially active");
333 desc = gettext("unknown");
338 * Allow hot spares to be shared between pools.
340 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)
379 /* No valid type detected device is safe to use */
380 value = blkid_get_tag_value(cache, "TYPE", path);
385 * If libblkid detects a ZFS device, we check the device
386 * using check_file() to see if it's safe. The one safe
387 * case is a spare device shared between multiple pools.
389 if (strcmp(value, "zfs_member") == 0) {
390 err = check_file(path, force, isspare);
396 vdev_error(gettext("%s contains a filesystem of "
397 "type '%s'\n"), path, value);
403 err = check_file(path, force, isspare);
404 #endif /* HAVE_LIBBLKID */
410 * Validate a whole disk. Iterate over all slices on the disk and make sure
411 * that none is in use by calling check_slice().
414 check_disk(const char *path, blkid_cache cache, int force,
415 boolean_t isspare, boolean_t iswholedisk)
418 char slice_path[MAXPATHLEN];
422 /* This is not a wholedisk we only check the given partition */
424 return (check_slice(path, cache, force, isspare));
427 * When the device is a whole disk try to read the efi partition
428 * label. If this is successful we safely check the all of the
429 * partitions. However, when it fails it may simply be because
430 * the disk is partitioned via the MBR. Since we currently can
431 * not easily decode the MBR return a failure and prompt to the
432 * user to use force option since we cannot check the partitions.
434 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
439 if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
445 vdev_error(gettext("%s does not contain an EFI "
446 "label but it may contain partition\n"
447 "information in the MBR.\n"), path);
453 * The primary efi partition label is damaged however the secondary
454 * label at the end of the device is intact. Rather than use this
455 * label we should play it safe and treat this as a non efi device.
457 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
462 /* Partitions will no be created using the backup */
465 vdev_error(gettext("%s contains a corrupt primary "
466 "EFI label.\n"), path);
471 for (i = 0; i < vtoc->efi_nparts; i++) {
473 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
474 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
477 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
478 (void) snprintf(slice_path, sizeof (slice_path),
479 "%s%s%d", path, "-part", i+1);
481 (void) snprintf(slice_path, sizeof (slice_path),
482 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
485 err = check_slice(slice_path, cache, force, isspare);
497 check_device(const char *path, boolean_t force,
498 boolean_t isspare, boolean_t iswholedisk)
500 static blkid_cache cache = NULL;
504 * There is no easy way to add a correct blkid_put_cache() call,
505 * memory will be reclaimed when the command exits.
510 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
515 if ((err = blkid_probe_all(cache)) != 0) {
516 blkid_put_cache(cache);
521 #endif /* HAVE_LIBBLKID */
523 return (check_disk(path, cache, force, isspare, iswholedisk));
527 * By "whole disk" we mean an entire physical disk (something we can
528 * label, toggle the write cache on, etc.) as opposed to the full
529 * capacity of a pseudo-device such as lofi or did. We act as if we
530 * are labeling the disk, which should be a pretty good test of whether
531 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
535 is_whole_disk(const char *path)
537 struct dk_gpt *label;
540 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
542 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
552 * This may be a shorthand device path or it could be total gibberish.
553 * Check to see if it is a known device available in zfs_vdev_paths.
554 * As part of this check, see if we've been given an entire disk
555 * (minus the slice number).
558 is_shorthand_path(const char *arg, char *path,
559 struct stat64 *statbuf, boolean_t *wholedisk)
563 error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
565 *wholedisk = is_whole_disk(path);
566 if (*wholedisk || (stat64(path, statbuf) == 0))
570 strlcpy(path, arg, sizeof (path));
571 memset(statbuf, 0, sizeof (*statbuf));
572 *wholedisk = B_FALSE;
578 * Determine if the given path is a hot spare within the given configuration.
579 * If no configuration is given we rely solely on the label.
582 is_spare(nvlist_t *config, const char *path)
588 uint64_t guid, spareguid;
594 if ((fd = open(path, O_RDONLY)) < 0)
597 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
599 state != POOL_STATE_SPARE ||
600 zpool_read_label(fd, &label, NULL) != 0) {
611 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
614 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
616 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
617 &spares, &nspares) == 0) {
618 for (i = 0; i < nspares; i++) {
619 verify(nvlist_lookup_uint64(spares[i],
620 ZPOOL_CONFIG_GUID, &spareguid) == 0);
621 if (spareguid == guid)
630 * Create a leaf vdev. Determine if this is a file or a device. If it's a
631 * device, fill in the device id to make a complete nvlist. Valid forms for a
634 * /dev/xxx Complete disk path
635 * /xxx Full path to file
636 * xxx Shorthand for <zfs_vdev_paths>/xxx
639 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
641 char path[MAXPATHLEN];
642 struct stat64 statbuf;
643 nvlist_t *vdev = NULL;
645 boolean_t wholedisk = B_FALSE;
650 * Determine what type of vdev this is, and put the full path into
651 * 'path'. We detect whether this is a device of file afterwards by
652 * checking the st_mode of the file.
656 * Complete device or file path. Exact type is determined by
657 * examining the file descriptor afterwards. Symbolic links
658 * are resolved to their real paths for the is_whole_disk()
659 * and S_ISBLK/S_ISREG type checks. However, we are careful
660 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
661 * can leverage udev's persistent device labels.
663 if (realpath(arg, path) == NULL) {
664 (void) fprintf(stderr,
665 gettext("cannot resolve path '%s'\n"), arg);
669 wholedisk = is_whole_disk(path);
670 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
671 (void) fprintf(stderr,
672 gettext("cannot open '%s': %s\n"),
673 path, strerror(errno));
677 /* After is_whole_disk() check restore original passed path */
678 strlcpy(path, arg, MAXPATHLEN);
680 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
683 * If we got ENOENT, then the user gave us
684 * gibberish, so try to direct them with a
685 * reasonable error message. Otherwise,
686 * regurgitate strerror() since it's the best we
690 (void) fprintf(stderr,
691 gettext("cannot open '%s': no such "
692 "device in %s\n"), arg, DISK_ROOT);
693 (void) fprintf(stderr,
694 gettext("must be a full path or "
695 "shorthand device name\n"));
698 (void) fprintf(stderr,
699 gettext("cannot open '%s': %s\n"),
700 path, strerror(errno));
707 * Determine whether this is a device or a file.
709 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
710 type = VDEV_TYPE_DISK;
711 } else if (S_ISREG(statbuf.st_mode)) {
712 type = VDEV_TYPE_FILE;
714 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
715 "block device or regular file\n"), path);
720 * Finally, we have the complete device or file, and we know that it is
721 * acceptable to use. Construct the nvlist to describe this vdev. All
722 * vdevs have a 'path' element, and devices also have a 'devid' element.
724 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
725 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
726 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
727 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
728 if (strcmp(type, VDEV_TYPE_DISK) == 0)
729 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
730 (uint64_t)wholedisk) == 0);
733 * Override defaults if custom properties are provided.
738 if (nvlist_lookup_string(props,
739 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
740 zfs_nicestrtonum(NULL, value, &ashift);
744 * If the device is known to incorrectly report its physical sector
745 * size explicitly provide the known correct value.
750 if (check_sector_size_database(path, §or_size) == B_TRUE)
751 ashift = highbit64(sector_size) - 1;
755 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
761 * Go through and verify the replication level of the pool is consistent.
762 * Performs the following checks:
764 * For the new spec, verifies that devices in mirrors and raidz are the
767 * If the current configuration already has inconsistent replication
768 * levels, ignore any other potential problems in the new spec.
770 * Otherwise, make sure that the current spec (if there is one) and the new
771 * spec have consistent replication levels.
773 typedef struct replication_level {
775 uint64_t zprl_children;
776 uint64_t zprl_parity;
777 } replication_level_t;
779 #define ZPOOL_FUZZ (16 * 1024 * 1024)
782 * Given a list of toplevel vdevs, return the current replication level. If
783 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
784 * an error message will be displayed for each self-inconsistent vdev.
786 static replication_level_t *
787 get_replication(nvlist_t *nvroot, boolean_t fatal)
795 replication_level_t lastrep = { 0 }, rep, *ret;
796 boolean_t dontreport;
798 ret = safe_malloc(sizeof (replication_level_t));
800 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
801 &top, &toplevels) == 0);
803 lastrep.zprl_type = NULL;
804 for (t = 0; t < toplevels; t++) {
805 uint64_t is_log = B_FALSE;
810 * For separate logs we ignore the top level vdev replication
813 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
817 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
819 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
820 &child, &children) != 0) {
822 * This is a 'file' or 'disk' vdev.
824 rep.zprl_type = type;
825 rep.zprl_children = 1;
831 * This is a mirror or RAID-Z vdev. Go through and make
832 * sure the contents are all the same (files vs. disks),
833 * keeping track of the number of elements in the
836 * We also check that the size of each vdev (if it can
837 * be determined) is the same.
839 rep.zprl_type = type;
840 rep.zprl_children = 0;
842 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
843 verify(nvlist_lookup_uint64(nv,
844 ZPOOL_CONFIG_NPARITY,
845 &rep.zprl_parity) == 0);
846 assert(rep.zprl_parity != 0);
852 * The 'dontreport' variable indicates that we've
853 * already reported an error for this spec, so don't
854 * bother doing it again.
859 for (c = 0; c < children; c++) {
860 nvlist_t *cnv = child[c];
862 struct stat64 statbuf;
863 uint64_t size = -1ULL;
869 verify(nvlist_lookup_string(cnv,
870 ZPOOL_CONFIG_TYPE, &childtype) == 0);
873 * If this is a replacing or spare vdev, then
874 * get the real first child of the vdev.
876 if (strcmp(childtype,
877 VDEV_TYPE_REPLACING) == 0 ||
878 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
882 verify(nvlist_lookup_nvlist_array(cnv,
883 ZPOOL_CONFIG_CHILDREN, &rchild,
885 assert(rchildren == 2);
888 verify(nvlist_lookup_string(cnv,
893 verify(nvlist_lookup_string(cnv,
894 ZPOOL_CONFIG_PATH, &path) == 0);
897 * If we have a raidz/mirror that combines disks
898 * with files, report it as an error.
900 if (!dontreport && type != NULL &&
901 strcmp(type, childtype) != 0) {
907 "mismatched replication "
908 "level: %s contains both "
909 "files and devices\n"),
917 * According to stat(2), the value of 'st_size'
918 * is undefined for block devices and character
919 * devices. But there is no effective way to
920 * determine the real size in userland.
922 * Instead, we'll take advantage of an
923 * implementation detail of spec_size(). If the
924 * device is currently open, then we (should)
925 * return a valid size.
927 * If we still don't get a valid size (indicated
928 * by a size of 0 or MAXOFFSET_T), then ignore
929 * this device altogether.
931 if ((fd = open(path, O_RDONLY)) >= 0) {
932 err = fstat64(fd, &statbuf);
935 err = stat64(path, &statbuf);
939 statbuf.st_size == 0 ||
940 statbuf.st_size == MAXOFFSET_T)
943 size = statbuf.st_size;
946 * Also make sure that devices and
947 * slices have a consistent size. If
948 * they differ by a significant amount
949 * (~16MB) then report an error.
952 (vdev_size != -1ULL &&
953 (labs(size - vdev_size) >
960 "%s contains devices of "
961 "different sizes\n"),
974 * At this point, we have the replication of the last toplevel
975 * vdev in 'rep'. Compare it to 'lastrep' to see if its
978 if (lastrep.zprl_type != NULL) {
979 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
985 "mismatched replication level: "
986 "both %s and %s vdevs are "
988 lastrep.zprl_type, rep.zprl_type);
991 } else if (lastrep.zprl_parity != rep.zprl_parity) {
997 "mismatched replication level: "
998 "both %llu and %llu device parity "
999 "%s vdevs are present\n"),
1000 lastrep.zprl_parity,
1005 } else if (lastrep.zprl_children != rep.zprl_children) {
1011 "mismatched replication level: "
1012 "both %llu-way and %llu-way %s "
1013 "vdevs are present\n"),
1014 lastrep.zprl_children,
1031 * Check the replication level of the vdev spec against the current pool. Calls
1032 * get_replication() to make sure the new spec is self-consistent. If the pool
1033 * has a consistent replication level, then we ignore any errors. Otherwise,
1034 * report any difference between the two.
1037 check_replication(nvlist_t *config, nvlist_t *newroot)
1041 replication_level_t *current = NULL, *new;
1045 * If we have a current pool configuration, check to see if it's
1046 * self-consistent. If not, simply return success.
1048 if (config != NULL) {
1051 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1053 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1057 * for spares there may be no children, and therefore no
1058 * replication level to check
1060 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1061 &child, &children) != 0) || (children == 0)) {
1067 * If all we have is logs then there's no replication level to check.
1069 if (num_logs(newroot) == children) {
1075 * Get the replication level of the new vdev spec, reporting any
1076 * inconsistencies found.
1078 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1084 * Check to see if the new vdev spec matches the replication level of
1088 if (current != NULL) {
1089 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1091 "mismatched replication level: pool uses %s "
1092 "and new vdev is %s\n"),
1093 current->zprl_type, new->zprl_type);
1095 } else if (current->zprl_parity != new->zprl_parity) {
1097 "mismatched replication level: pool uses %llu "
1098 "device parity and new vdev uses %llu\n"),
1099 current->zprl_parity, new->zprl_parity);
1101 } else if (current->zprl_children != new->zprl_children) {
1103 "mismatched replication level: pool uses %llu-way "
1104 "%s and new vdev uses %llu-way %s\n"),
1105 current->zprl_children, current->zprl_type,
1106 new->zprl_children, new->zprl_type);
1112 if (current != NULL)
1119 zero_label(char *path)
1121 const int size = 4096;
1125 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1126 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1127 path, strerror(errno));
1131 memset(buf, 0, size);
1132 err = write(fd, buf, size);
1133 (void) fdatasync(fd);
1137 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1138 "of '%s': %s\n"), size, path, strerror(errno));
1143 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1144 "of '%s'\n"), err, size, path);
1152 * Go through and find any whole disks in the vdev specification, labelling them
1153 * as appropriate. When constructing the vdev spec, we were unable to open this
1154 * device in order to provide a devid. Now that we have labelled the disk and
1155 * know that slice 0 is valid, we can construct the devid now.
1157 * If the disk was already labeled with an EFI label, we will have gotten the
1158 * devid already (because we were able to open the whole disk). Otherwise, we
1159 * need to get the devid after we label the disk.
1162 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1167 char devpath[MAXPATHLEN];
1168 char udevpath[MAXPATHLEN];
1170 struct stat64 statbuf;
1171 int is_exclusive = 0;
1175 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1177 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1178 &child, &children) != 0) {
1180 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1184 * We have a disk device. If this is a whole disk write
1185 * out the efi partition table, otherwise write zero's to
1186 * the first 4k of the partition. This is to ensure that
1187 * libblkid will not misidentify the partition due to a
1188 * magic value left by the previous filesystem.
1190 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1191 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1195 (void) zero_label(path);
1199 if (realpath(path, devpath) == NULL) {
1201 (void) fprintf(stderr,
1202 gettext("cannot resolve path '%s'\n"), path);
1207 * Remove any previously existing symlink from a udev path to
1208 * the device before labeling the disk. This makes
1209 * zpool_label_disk_wait() truly wait for the new link to show
1210 * up instead of returning if it finds an old link still in
1211 * place. Otherwise there is a window between when udev
1212 * deletes and recreates the link during which access attempts
1213 * will fail with ENOENT.
1215 strncpy(udevpath, path, MAXPATHLEN);
1216 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1218 fd = open(devpath, O_RDWR|O_EXCL);
1227 * If the partition exists, contains a valid spare label,
1228 * and is opened exclusively there is no need to partition
1229 * it. Hot spares have already been partitioned and are
1230 * held open exclusively by the kernel as a safety measure.
1232 * If the provided path is for a /dev/disk/ device its
1233 * symbolic link will be removed, partition table created,
1234 * and then block until udev creates the new link.
1236 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1237 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1239 ret = lstat64(udevpath, &statbuf);
1240 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1241 (void) unlink(udevpath);
1244 if (zpool_label_disk(g_zfs, zhp,
1245 strrchr(devpath, '/') + 1) == -1)
1248 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1250 (void) fprintf(stderr, gettext("cannot "
1251 "resolve path '%s': %d\n"), udevpath, ret);
1255 (void) zero_label(udevpath);
1259 * Update the path to refer to the partition. The presence of
1260 * the 'whole_disk' field indicates to the CLI that we should
1261 * chop off the partition number when displaying the device in
1264 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1269 for (c = 0; c < children; c++)
1270 if ((ret = make_disks(zhp, child[c])) != 0)
1273 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1274 &child, &children) == 0)
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_L2CACHE,
1280 &child, &children) == 0)
1281 for (c = 0; c < children; c++)
1282 if ((ret = make_disks(zhp, child[c])) != 0)
1289 * Go through and find any devices that are in use. We rely on libdiskmgt for
1290 * the majority of this task.
1293 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1294 boolean_t replacing, boolean_t isspare)
1300 char buf[MAXPATHLEN];
1301 uint64_t wholedisk = B_FALSE;
1303 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1305 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1306 &child, &children) != 0) {
1308 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1309 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1310 verify(!nvlist_lookup_uint64(nv,
1311 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1314 * As a generic check, we look to see if this is a replace of a
1315 * hot spare within the same pool. If so, we allow it
1316 * regardless of what libblkid or zpool_in_use() says.
1319 (void) strlcpy(buf, path, sizeof (buf));
1321 ret = zfs_append_partition(buf, sizeof (buf));
1326 if (is_spare(config, buf))
1330 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1331 ret = check_device(path, force, isspare, wholedisk);
1333 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1334 ret = check_file(path, force, isspare);
1339 for (c = 0; c < children; c++)
1340 if ((ret = check_in_use(config, child[c], force,
1341 replacing, B_FALSE)) != 0)
1344 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1345 &child, &children) == 0)
1346 for (c = 0; c < children; c++)
1347 if ((ret = check_in_use(config, child[c], force,
1348 replacing, B_TRUE)) != 0)
1351 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1352 &child, &children) == 0)
1353 for (c = 0; c < children; c++)
1354 if ((ret = check_in_use(config, child[c], force,
1355 replacing, B_FALSE)) != 0)
1362 is_grouping(const char *type, int *mindev, int *maxdev)
1364 if (strncmp(type, "raidz", 5) == 0) {
1365 const char *p = type + 5;
1371 } else if (*p == '0') {
1372 return (NULL); /* no zero prefixes allowed */
1375 nparity = strtol(p, &end, 10);
1376 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1382 *mindev = nparity + 1;
1385 return (VDEV_TYPE_RAIDZ);
1391 if (strcmp(type, "mirror") == 0) {
1394 return (VDEV_TYPE_MIRROR);
1397 if (strcmp(type, "spare") == 0) {
1400 return (VDEV_TYPE_SPARE);
1403 if (strcmp(type, "log") == 0) {
1406 return (VDEV_TYPE_LOG);
1409 if (strcmp(type, "cache") == 0) {
1412 return (VDEV_TYPE_L2CACHE);
1419 * Construct a syntactically valid vdev specification,
1420 * and ensure that all devices and files exist and can be opened.
1421 * Note: we don't bother freeing anything in the error paths
1422 * because the program is just going to exit anyway.
1425 construct_spec(nvlist_t *props, int argc, char **argv)
1427 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1428 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1431 boolean_t seen_logs;
1441 seen_logs = B_FALSE;
1447 * If it's a mirror or raidz, the subsequent arguments are
1448 * its leaves -- until we encounter the next mirror or raidz.
1450 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1451 nvlist_t **child = NULL;
1452 int c, children = 0;
1454 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1455 if (spares != NULL) {
1456 (void) fprintf(stderr,
1457 gettext("invalid vdev "
1458 "specification: 'spare' can be "
1459 "specified only once\n"));
1465 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1467 (void) fprintf(stderr,
1468 gettext("invalid vdev "
1469 "specification: 'log' can be "
1470 "specified only once\n"));
1478 * A log is not a real grouping device.
1479 * We just set is_log and continue.
1484 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1485 if (l2cache != NULL) {
1486 (void) fprintf(stderr,
1487 gettext("invalid vdev "
1488 "specification: 'cache' can be "
1489 "specified only once\n"));
1496 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1497 (void) fprintf(stderr,
1498 gettext("invalid vdev "
1499 "specification: unsupported 'log' "
1500 "device: %s\n"), type);
1506 for (c = 1; c < argc; c++) {
1507 if (is_grouping(argv[c], NULL, NULL) != NULL)
1510 child = realloc(child,
1511 children * sizeof (nvlist_t *));
1514 if ((nv = make_leaf_vdev(props, argv[c],
1517 child[children - 1] = nv;
1520 if (children < mindev) {
1521 (void) fprintf(stderr, gettext("invalid vdev "
1522 "specification: %s requires at least %d "
1523 "devices\n"), argv[0], mindev);
1527 if (children > maxdev) {
1528 (void) fprintf(stderr, gettext("invalid vdev "
1529 "specification: %s supports no more than "
1530 "%d devices\n"), argv[0], maxdev);
1537 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1541 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1543 nl2cache = children;
1546 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1548 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1550 verify(nvlist_add_uint64(nv,
1551 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1552 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1553 verify(nvlist_add_uint64(nv,
1554 ZPOOL_CONFIG_NPARITY,
1557 verify(nvlist_add_nvlist_array(nv,
1558 ZPOOL_CONFIG_CHILDREN, child,
1561 for (c = 0; c < children; c++)
1562 nvlist_free(child[c]);
1567 * We have a device. Pass off to make_leaf_vdev() to
1568 * construct the appropriate nvlist describing the vdev.
1570 if ((nv = make_leaf_vdev(props, argv[0],
1580 top = realloc(top, toplevels * sizeof (nvlist_t *));
1583 top[toplevels - 1] = nv;
1586 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1587 (void) fprintf(stderr, gettext("invalid vdev "
1588 "specification: at least one toplevel vdev must be "
1593 if (seen_logs && nlogs == 0) {
1594 (void) fprintf(stderr, gettext("invalid vdev specification: "
1595 "log requires at least 1 device\n"));
1600 * Finally, create nvroot and add all top-level vdevs to it.
1602 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1603 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1604 VDEV_TYPE_ROOT) == 0);
1605 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1606 top, toplevels) == 0);
1608 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1609 spares, nspares) == 0);
1611 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1612 l2cache, nl2cache) == 0);
1614 for (t = 0; t < toplevels; t++)
1615 nvlist_free(top[t]);
1616 for (t = 0; t < nspares; t++)
1617 nvlist_free(spares[t]);
1618 for (t = 0; t < nl2cache; t++)
1619 nvlist_free(l2cache[t]);
1630 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1631 splitflags_t flags, int argc, char **argv)
1633 nvlist_t *newroot = NULL, **child;
1637 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1638 (void) fprintf(stderr, gettext("Unable to build a "
1639 "pool from the specified devices\n"));
1643 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1644 nvlist_free(newroot);
1648 /* avoid any tricks in the spec */
1649 verify(nvlist_lookup_nvlist_array(newroot,
1650 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1651 for (c = 0; c < children; c++) {
1656 verify(nvlist_lookup_string(child[c],
1657 ZPOOL_CONFIG_PATH, &path) == 0);
1658 if ((type = is_grouping(path, &min, &max)) != NULL) {
1659 (void) fprintf(stderr, gettext("Cannot use "
1660 "'%s' as a device for splitting\n"), type);
1661 nvlist_free(newroot);
1667 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1668 if (newroot != NULL)
1669 nvlist_free(newroot);
1677 * Get and validate the contents of the given vdev specification. This ensures
1678 * that the nvlist returned is well-formed, that all the devices exist, and that
1679 * they are not currently in use by any other known consumer. The 'poolconfig'
1680 * parameter is the current configuration of the pool when adding devices
1681 * existing pool, and is used to perform additional checks, such as changing the
1682 * replication level of the pool. It can be 'NULL' to indicate that this is a
1683 * new pool. The 'force' flag controls whether devices should be forcefully
1684 * added, even if they appear in use.
1687 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1688 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1691 nvlist_t *poolconfig = NULL;
1695 * Construct the vdev specification. If this is successful, we know
1696 * that we have a valid specification, and that all devices can be
1699 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1702 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1706 * Validate each device to make sure that its not shared with another
1707 * subsystem. We do this even if 'force' is set, because there are some
1708 * uses (such as a dedicated dump device) that even '-f' cannot
1711 if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1712 nvlist_free(newroot);
1717 * Check the replication level of the given vdevs and report any errors
1718 * found. We include the existing pool spec, if any, as we need to
1719 * catch changes against the existing replication level.
1721 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1722 nvlist_free(newroot);
1727 * Run through the vdev specification and label any whole disks found.
1729 if (!dryrun && make_disks(zhp, newroot) != 0) {
1730 nvlist_free(newroot);