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.
28 * Functions to convert between a list of vdevs and an nvlist representing the
29 * configuration. Each entry in the list can be one of:
32 * disk=(path=..., devid=...)
41 * While the underlying implementation supports it, group vdevs cannot contain
42 * other group vdevs. All userland verification of devices is contained within
43 * this file. If successful, the nvlist returned can be passed directly to the
44 * kernel; we've done as much verification as possible in userland.
46 * Hot spares are a special case, and passed down as an array of disk vdevs, at
47 * the same level as the root of the vdev tree.
49 * The only function exported by this file is 'make_root_vdev'. The
50 * function performs several passes:
52 * 1. Construct the vdev specification. Performs syntax validation and
53 * makes sure each device is valid.
54 * 2. Check for devices in use. Using libblkid to make sure that no
55 * devices are also in use. Some can be overridden using the 'force'
56 * flag, others cannot.
57 * 3. Check for replication errors if the 'force' flag is not specified.
58 * validates that the replication level is consistent across the
60 * 4. Call libzfs to label any whole disks with an EFI label.
69 #include <libnvpair.h>
71 #include <scsi/scsi.h>
76 #include <sys/efi_partition.h>
79 #include <sys/mntent.h>
80 #include <uuid/uuid.h>
82 #include <blkid/blkid.h>
84 #define blkid_cache void *
85 #endif /* HAVE_LIBBLKID */
87 #include "zpool_util.h"
88 #include <sys/zfs_context.h>
91 * For any given vdev specification, we can have multiple errors. The
92 * vdev_error() function keeps track of whether we have seen an error yet, and
93 * prints out a header if its the first error we've seen.
98 typedef struct vdev_disk_db_entry
102 } vdev_disk_db_entry_t
;
105 * Database of block devices that lie about physical sector sizes. The
106 * identification string must be precisely 24 characters to avoid false
109 static vdev_disk_db_entry_t vdev_disk_database
[] = {
110 {"ATA ADATA SSD S396 3", 8192},
111 {"ATA APPLE SSD SM128E", 8192},
112 {"ATA APPLE SSD SM256E", 8192},
113 {"ATA APPLE SSD SM512E", 8192},
114 {"ATA APPLE SSD SM768E", 8192},
115 {"ATA C400-MTFDDAC064M", 8192},
116 {"ATA C400-MTFDDAC128M", 8192},
117 {"ATA C400-MTFDDAC256M", 8192},
118 {"ATA C400-MTFDDAC512M", 8192},
119 {"ATA Corsair Force 3 ", 8192},
120 {"ATA Corsair Force GS", 8192},
121 {"ATA INTEL SSDSA2CT04", 8192},
122 {"ATA INTEL SSDSA2BZ10", 8192},
123 {"ATA INTEL SSDSA2BZ20", 8192},
124 {"ATA INTEL SSDSA2BZ30", 8192},
125 {"ATA INTEL SSDSA2CW04", 8192},
126 {"ATA INTEL SSDSA2CW08", 8192},
127 {"ATA INTEL SSDSA2CW12", 8192},
128 {"ATA INTEL SSDSA2CW16", 8192},
129 {"ATA INTEL SSDSA2CW30", 8192},
130 {"ATA INTEL SSDSA2CW60", 8192},
131 {"ATA INTEL SSDSC2CT06", 8192},
132 {"ATA INTEL SSDSC2CT12", 8192},
133 {"ATA INTEL SSDSC2CT18", 8192},
134 {"ATA INTEL SSDSC2CT24", 8192},
135 {"ATA INTEL SSDSC2CW06", 8192},
136 {"ATA INTEL SSDSC2CW12", 8192},
137 {"ATA INTEL SSDSC2CW18", 8192},
138 {"ATA INTEL SSDSC2CW24", 8192},
139 {"ATA INTEL SSDSC2CW48", 8192},
140 {"ATA KINGSTON SH100S3", 8192},
141 {"ATA KINGSTON SH103S3", 8192},
142 {"ATA M4-CT064M4SSD2 ", 8192},
143 {"ATA M4-CT128M4SSD2 ", 8192},
144 {"ATA M4-CT256M4SSD2 ", 8192},
145 {"ATA M4-CT512M4SSD2 ", 8192},
146 {"ATA OCZ-AGILITY2 ", 8192},
147 {"ATA OCZ-AGILITY3 ", 8192},
148 {"ATA OCZ-VERTEX2 3.5 ", 8192},
149 {"ATA OCZ-VERTEX3 ", 8192},
150 {"ATA OCZ-VERTEX3 LT ", 8192},
151 {"ATA OCZ-VERTEX3 MI ", 8192},
152 {"ATA OCZ-VERTEX4 ", 8192},
153 {"ATA SAMSUNG MZ7WD120", 8192},
154 {"ATA SAMSUNG MZ7WD240", 8192},
155 {"ATA SAMSUNG MZ7WD480", 8192},
156 {"ATA SAMSUNG MZ7WD960", 8192},
157 {"ATA SAMSUNG SSD 830 ", 8192},
158 {"ATA Samsung SSD 840 ", 8192},
159 {"ATA SanDisk SSD U100", 8192},
160 {"ATA TOSHIBA THNSNH06", 8192},
161 {"ATA TOSHIBA THNSNH12", 8192},
162 {"ATA TOSHIBA THNSNH25", 8192},
163 {"ATA TOSHIBA THNSNH51", 8192},
164 {"ATA APPLE SSD TS064C", 4096},
165 {"ATA APPLE SSD TS128C", 4096},
166 {"ATA APPLE SSD TS256C", 4096},
167 {"ATA APPLE SSD TS512C", 4096},
168 {"ATA INTEL SSDSA2M040", 4096},
169 {"ATA INTEL SSDSA2M080", 4096},
170 {"ATA INTEL SSDSA2M160", 4096},
171 {"ATA INTEL SSDSC2MH12", 4096},
172 {"ATA INTEL SSDSC2MH25", 4096},
173 {"ATA OCZ CORE_SSD ", 4096},
174 {"ATA OCZ-VERTEX ", 4096},
175 {"ATA SAMSUNG MCCOE32G", 4096},
176 {"ATA SAMSUNG MCCOE64G", 4096},
177 {"ATA SAMSUNG SSD PM80", 4096},
178 /* Flash drives optimized for 4KB IOs on larger pages */
179 {"ATA INTEL SSDSC2BA10", 4096},
180 {"ATA INTEL SSDSC2BA20", 4096},
181 {"ATA INTEL SSDSC2BA40", 4096},
182 {"ATA INTEL SSDSC2BA80", 4096},
183 {"ATA INTEL SSDSC2BB08", 4096},
184 {"ATA INTEL SSDSC2BB12", 4096},
185 {"ATA INTEL SSDSC2BB16", 4096},
186 {"ATA INTEL SSDSC2BB24", 4096},
187 {"ATA INTEL SSDSC2BB30", 4096},
188 {"ATA INTEL SSDSC2BB40", 4096},
189 {"ATA INTEL SSDSC2BB48", 4096},
190 {"ATA INTEL SSDSC2BB60", 4096},
191 {"ATA INTEL SSDSC2BB80", 4096},
192 {"ATA INTEL SSDSC2BW24", 4096},
193 {"ATA INTEL SSDSC2BP24", 4096},
194 {"ATA INTEL SSDSC2BP48", 4096},
195 {"NA SmrtStorSDLKAE9W", 4096},
196 /* Imported from Open Solaris */
197 {"ATA MARVELL SD88SA02", 4096},
198 /* Advanced format Hard drives */
199 {"ATA Hitachi HDS5C303", 4096},
200 {"ATA SAMSUNG HD204UI ", 4096},
201 {"ATA ST2000DL004 HD20", 4096},
202 {"ATA WDC WD10EARS-00M", 4096},
203 {"ATA WDC WD10EARS-00S", 4096},
204 {"ATA WDC WD10EARS-00Z", 4096},
205 {"ATA WDC WD15EARS-00M", 4096},
206 {"ATA WDC WD15EARS-00S", 4096},
207 {"ATA WDC WD15EARS-00Z", 4096},
208 {"ATA WDC WD20EARS-00M", 4096},
209 {"ATA WDC WD20EARS-00S", 4096},
210 {"ATA WDC WD20EARS-00Z", 4096},
211 {"ATA WDC WD1600BEVT-0", 4096},
212 {"ATA WDC WD2500BEVT-0", 4096},
213 {"ATA WDC WD3200BEVT-0", 4096},
214 {"ATA WDC WD5000BEVT-0", 4096},
215 /* Virtual disks: Assume zvols with default volblocksize */
217 {"ATA QEMU HARDDISK ", 8192},
218 {"IET VIRTUAL-DISK ", 8192},
219 {"OI COMSTAR ", 8192},
220 {"SUN COMSTAR ", 8192},
221 {"NETAPP LUN ", 8192},
225 static const int vdev_disk_database_size
=
226 sizeof (vdev_disk_database
) / sizeof (vdev_disk_database
[0]);
228 #define INQ_REPLY_LEN 96
229 #define INQ_CMD_LEN 6
232 check_sector_size_database(char *path
, int *sector_size
)
234 unsigned char inq_buff
[INQ_REPLY_LEN
];
235 unsigned char sense_buffer
[32];
236 unsigned char inq_cmd_blk
[INQ_CMD_LEN
] =
237 {INQUIRY
, 0, 0, 0, INQ_REPLY_LEN
, 0};
243 /* Prepare INQUIRY command */
244 memset(&io_hdr
, 0, sizeof (sg_io_hdr_t
));
245 io_hdr
.interface_id
= 'S';
246 io_hdr
.cmd_len
= sizeof (inq_cmd_blk
);
247 io_hdr
.mx_sb_len
= sizeof (sense_buffer
);
248 io_hdr
.dxfer_direction
= SG_DXFER_FROM_DEV
;
249 io_hdr
.dxfer_len
= INQ_REPLY_LEN
;
250 io_hdr
.dxferp
= inq_buff
;
251 io_hdr
.cmdp
= inq_cmd_blk
;
252 io_hdr
.sbp
= sense_buffer
;
253 io_hdr
.timeout
= 10; /* 10 milliseconds is ample time */
255 if ((fd
= open(path
, O_RDONLY
|O_DIRECT
)) < 0)
258 error
= ioctl(fd
, SG_IO
, (unsigned long) &io_hdr
);
265 if ((io_hdr
.info
& SG_INFO_OK_MASK
) != SG_INFO_OK
)
268 for (i
= 0; i
< vdev_disk_database_size
; i
++) {
269 if (memcmp(inq_buff
+ 8, vdev_disk_database
[i
].id
, 24))
272 *sector_size
= vdev_disk_database
[i
].sector_size
;
281 vdev_error(const char *fmt
, ...)
286 (void) fprintf(stderr
, gettext("invalid vdev specification\n"));
288 (void) fprintf(stderr
, gettext("use '-f' to override "
289 "the following errors:\n"));
291 (void) fprintf(stderr
, gettext("the following errors "
292 "must be manually repaired:\n"));
297 (void) vfprintf(stderr
, fmt
, ap
);
302 * Check that a file is valid. All we can do in this case is check that it's
303 * not in use by another pool, and not in use by swap.
306 check_file(const char *file
, boolean_t force
, boolean_t isspare
)
314 if ((fd
= open(file
, O_RDONLY
)) < 0)
317 if (zpool_in_use(g_zfs
, fd
, &state
, &name
, &inuse
) == 0 && inuse
) {
321 case POOL_STATE_ACTIVE
:
322 desc
= gettext("active");
325 case POOL_STATE_EXPORTED
:
326 desc
= gettext("exported");
329 case POOL_STATE_POTENTIALLY_ACTIVE
:
330 desc
= gettext("potentially active");
334 desc
= gettext("unknown");
339 * Allow hot spares to be shared between pools.
341 if (state
== POOL_STATE_SPARE
&& isspare
)
344 if (state
== POOL_STATE_ACTIVE
||
345 state
== POOL_STATE_SPARE
|| !force
) {
347 case POOL_STATE_SPARE
:
348 vdev_error(gettext("%s is reserved as a hot "
349 "spare for pool %s\n"), file
, name
);
352 vdev_error(gettext("%s is part of %s pool "
353 "'%s'\n"), file
, desc
, name
);
369 (void) fprintf(stderr
, gettext("warning: device in use checking "
370 "failed: %s\n"), strerror(err
));
374 check_slice(const char *path
, blkid_cache cache
, int force
, boolean_t isspare
)
380 /* No valid type detected device is safe to use */
381 value
= blkid_get_tag_value(cache
, "TYPE", path
);
386 * If libblkid detects a ZFS device, we check the device
387 * using check_file() to see if it's safe. The one safe
388 * case is a spare device shared between multiple pools.
390 if (strcmp(value
, "zfs_member") == 0) {
391 err
= check_file(path
, force
, isspare
);
397 vdev_error(gettext("%s contains a filesystem of "
398 "type '%s'\n"), path
, value
);
404 err
= check_file(path
, force
, isspare
);
405 #endif /* HAVE_LIBBLKID */
411 * Validate a whole disk. Iterate over all slices on the disk and make sure
412 * that none is in use by calling check_slice().
415 check_disk(const char *path
, blkid_cache cache
, int force
,
416 boolean_t isspare
, boolean_t iswholedisk
)
419 char slice_path
[MAXPATHLEN
];
423 /* This is not a wholedisk we only check the given partition */
425 return (check_slice(path
, cache
, force
, isspare
));
428 * When the device is a whole disk try to read the efi partition
429 * label. If this is successful we safely check the all of the
430 * partitions. However, when it fails it may simply be because
431 * the disk is partitioned via the MBR. Since we currently can
432 * not easily decode the MBR return a failure and prompt to the
433 * user to use force option since we cannot check the partitions.
435 if ((fd
= open(path
, O_RDONLY
|O_DIRECT
)) < 0) {
440 if ((err
= efi_alloc_and_read(fd
, &vtoc
)) != 0) {
446 vdev_error(gettext("%s does not contain an EFI "
447 "label but it may contain partition\n"
448 "information in the MBR.\n"), path
);
454 * The primary efi partition label is damaged however the secondary
455 * label at the end of the device is intact. Rather than use this
456 * label we should play it safe and treat this as a non efi device.
458 if (vtoc
->efi_flags
& EFI_GPT_PRIMARY_CORRUPT
) {
463 /* Partitions will no be created using the backup */
466 vdev_error(gettext("%s contains a corrupt primary "
467 "EFI label.\n"), path
);
472 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
474 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
||
475 uuid_is_null((uchar_t
*)&vtoc
->efi_parts
[i
].p_guid
))
478 if (strncmp(path
, UDISK_ROOT
, strlen(UDISK_ROOT
)) == 0)
479 (void) snprintf(slice_path
, sizeof (slice_path
),
480 "%s%s%d", path
, "-part", i
+1);
482 (void) snprintf(slice_path
, sizeof (slice_path
),
483 "%s%s%d", path
, isdigit(path
[strlen(path
)-1]) ?
486 err
= check_slice(slice_path
, cache
, force
, isspare
);
498 check_device(const char *path
, boolean_t force
,
499 boolean_t isspare
, boolean_t iswholedisk
)
501 static blkid_cache cache
= NULL
;
505 * There is no easy way to add a correct blkid_put_cache() call,
506 * memory will be reclaimed when the command exits.
511 if ((err
= blkid_get_cache(&cache
, NULL
)) != 0) {
516 if ((err
= blkid_probe_all(cache
)) != 0) {
517 blkid_put_cache(cache
);
522 #endif /* HAVE_LIBBLKID */
524 return (check_disk(path
, cache
, force
, isspare
, iswholedisk
));
528 * By "whole disk" we mean an entire physical disk (something we can
529 * label, toggle the write cache on, etc.) as opposed to the full
530 * capacity of a pseudo-device such as lofi or did. We act as if we
531 * are labeling the disk, which should be a pretty good test of whether
532 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
536 is_whole_disk(const char *path
)
538 struct dk_gpt
*label
;
541 if ((fd
= open(path
, O_RDONLY
|O_DIRECT
)) < 0)
543 if (efi_alloc_and_init(fd
, EFI_NUMPAR
, &label
) != 0) {
553 * This may be a shorthand device path or it could be total gibberish.
554 * Check to see if it is a known device available in zfs_vdev_paths.
555 * As part of this check, see if we've been given an entire disk
556 * (minus the slice number).
559 is_shorthand_path(const char *arg
, char *path
,
560 struct stat64
*statbuf
, boolean_t
*wholedisk
)
564 error
= zfs_resolve_shortname(arg
, path
, MAXPATHLEN
);
566 *wholedisk
= is_whole_disk(path
);
567 if (*wholedisk
|| (stat64(path
, statbuf
) == 0))
571 strlcpy(path
, arg
, sizeof (path
));
572 memset(statbuf
, 0, sizeof (*statbuf
));
573 *wholedisk
= B_FALSE
;
579 * Determine if the given path is a hot spare within the given configuration.
580 * If no configuration is given we rely solely on the label.
583 is_spare(nvlist_t
*config
, const char *path
)
589 uint64_t guid
, spareguid
;
595 if ((fd
= open(path
, O_RDONLY
)) < 0)
598 if (zpool_in_use(g_zfs
, fd
, &state
, &name
, &inuse
) != 0 ||
600 state
!= POOL_STATE_SPARE
||
601 zpool_read_label(fd
, &label
, NULL
) != 0) {
612 verify(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) == 0);
615 verify(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
617 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
618 &spares
, &nspares
) == 0) {
619 for (i
= 0; i
< nspares
; i
++) {
620 verify(nvlist_lookup_uint64(spares
[i
],
621 ZPOOL_CONFIG_GUID
, &spareguid
) == 0);
622 if (spareguid
== guid
)
631 * Create a leaf vdev. Determine if this is a file or a device. If it's a
632 * device, fill in the device id to make a complete nvlist. Valid forms for a
635 * /dev/xxx Complete disk path
636 * /xxx Full path to file
637 * xxx Shorthand for <zfs_vdev_paths>/xxx
640 make_leaf_vdev(nvlist_t
*props
, const char *arg
, uint64_t is_log
)
642 char path
[MAXPATHLEN
];
643 struct stat64 statbuf
;
644 nvlist_t
*vdev
= NULL
;
646 boolean_t wholedisk
= B_FALSE
;
651 * Determine what type of vdev this is, and put the full path into
652 * 'path'. We detect whether this is a device of file afterwards by
653 * checking the st_mode of the file.
657 * Complete device or file path. Exact type is determined by
658 * examining the file descriptor afterwards. Symbolic links
659 * are resolved to their real paths for the is_whole_disk()
660 * and S_ISBLK/S_ISREG type checks. However, we are careful
661 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
662 * can leverage udev's persistent device labels.
664 if (realpath(arg
, path
) == NULL
) {
665 (void) fprintf(stderr
,
666 gettext("cannot resolve path '%s'\n"), arg
);
670 wholedisk
= is_whole_disk(path
);
671 if (!wholedisk
&& (stat64(path
, &statbuf
) != 0)) {
672 (void) fprintf(stderr
,
673 gettext("cannot open '%s': %s\n"),
674 path
, strerror(errno
));
678 /* After is_whole_disk() check restore original passed path */
679 strlcpy(path
, arg
, MAXPATHLEN
);
681 err
= is_shorthand_path(arg
, path
, &statbuf
, &wholedisk
);
684 * If we got ENOENT, then the user gave us
685 * gibberish, so try to direct them with a
686 * reasonable error message. Otherwise,
687 * regurgitate strerror() since it's the best we
691 (void) fprintf(stderr
,
692 gettext("cannot open '%s': no such "
693 "device in %s\n"), arg
, DISK_ROOT
);
694 (void) fprintf(stderr
,
695 gettext("must be a full path or "
696 "shorthand device name\n"));
699 (void) fprintf(stderr
,
700 gettext("cannot open '%s': %s\n"),
701 path
, strerror(errno
));
708 * Determine whether this is a device or a file.
710 if (wholedisk
|| S_ISBLK(statbuf
.st_mode
)) {
711 type
= VDEV_TYPE_DISK
;
712 } else if (S_ISREG(statbuf
.st_mode
)) {
713 type
= VDEV_TYPE_FILE
;
715 (void) fprintf(stderr
, gettext("cannot use '%s': must be a "
716 "block device or regular file\n"), path
);
721 * Finally, we have the complete device or file, and we know that it is
722 * acceptable to use. Construct the nvlist to describe this vdev. All
723 * vdevs have a 'path' element, and devices also have a 'devid' element.
725 verify(nvlist_alloc(&vdev
, NV_UNIQUE_NAME
, 0) == 0);
726 verify(nvlist_add_string(vdev
, ZPOOL_CONFIG_PATH
, path
) == 0);
727 verify(nvlist_add_string(vdev
, ZPOOL_CONFIG_TYPE
, type
) == 0);
728 verify(nvlist_add_uint64(vdev
, ZPOOL_CONFIG_IS_LOG
, is_log
) == 0);
729 if (strcmp(type
, VDEV_TYPE_DISK
) == 0)
730 verify(nvlist_add_uint64(vdev
, ZPOOL_CONFIG_WHOLE_DISK
,
731 (uint64_t)wholedisk
) == 0);
734 * Override defaults if custom properties are provided.
739 if (nvlist_lookup_string(props
,
740 zpool_prop_to_name(ZPOOL_PROP_ASHIFT
), &value
) == 0)
741 zfs_nicestrtonum(NULL
, value
, &ashift
);
745 * If the device is known to incorrectly report its physical sector
746 * size explicitly provide the known correct value.
751 if (check_sector_size_database(path
, §or_size
) == B_TRUE
)
752 ashift
= highbit64(sector_size
) - 1;
756 nvlist_add_uint64(vdev
, ZPOOL_CONFIG_ASHIFT
, ashift
);
762 * Go through and verify the replication level of the pool is consistent.
763 * Performs the following checks:
765 * For the new spec, verifies that devices in mirrors and raidz are the
768 * If the current configuration already has inconsistent replication
769 * levels, ignore any other potential problems in the new spec.
771 * Otherwise, make sure that the current spec (if there is one) and the new
772 * spec have consistent replication levels.
774 typedef struct replication_level
{
776 uint64_t zprl_children
;
777 uint64_t zprl_parity
;
778 } replication_level_t
;
780 #define ZPOOL_FUZZ (16 * 1024 * 1024)
783 * Given a list of toplevel vdevs, return the current replication level. If
784 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
785 * an error message will be displayed for each self-inconsistent vdev.
787 static replication_level_t
*
788 get_replication(nvlist_t
*nvroot
, boolean_t fatal
)
796 replication_level_t lastrep
= { 0 }, rep
, *ret
;
797 boolean_t dontreport
;
799 ret
= safe_malloc(sizeof (replication_level_t
));
801 verify(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
802 &top
, &toplevels
) == 0);
804 lastrep
.zprl_type
= NULL
;
805 for (t
= 0; t
< toplevels
; t
++) {
806 uint64_t is_log
= B_FALSE
;
811 * For separate logs we ignore the top level vdev replication
814 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &is_log
);
818 verify(nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
,
820 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
821 &child
, &children
) != 0) {
823 * This is a 'file' or 'disk' vdev.
825 rep
.zprl_type
= type
;
826 rep
.zprl_children
= 1;
832 * This is a mirror or RAID-Z vdev. Go through and make
833 * sure the contents are all the same (files vs. disks),
834 * keeping track of the number of elements in the
837 * We also check that the size of each vdev (if it can
838 * be determined) is the same.
840 rep
.zprl_type
= type
;
841 rep
.zprl_children
= 0;
843 if (strcmp(type
, VDEV_TYPE_RAIDZ
) == 0) {
844 verify(nvlist_lookup_uint64(nv
,
845 ZPOOL_CONFIG_NPARITY
,
846 &rep
.zprl_parity
) == 0);
847 assert(rep
.zprl_parity
!= 0);
853 * The 'dontreport' variable indicates that we've
854 * already reported an error for this spec, so don't
855 * bother doing it again.
860 for (c
= 0; c
< children
; c
++) {
861 nvlist_t
*cnv
= child
[c
];
863 struct stat64 statbuf
;
864 uint64_t size
= -1ULL;
870 verify(nvlist_lookup_string(cnv
,
871 ZPOOL_CONFIG_TYPE
, &childtype
) == 0);
874 * If this is a replacing or spare vdev, then
875 * get the real first child of the vdev.
877 if (strcmp(childtype
,
878 VDEV_TYPE_REPLACING
) == 0 ||
879 strcmp(childtype
, VDEV_TYPE_SPARE
) == 0) {
883 verify(nvlist_lookup_nvlist_array(cnv
,
884 ZPOOL_CONFIG_CHILDREN
, &rchild
,
886 assert(rchildren
== 2);
889 verify(nvlist_lookup_string(cnv
,
894 verify(nvlist_lookup_string(cnv
,
895 ZPOOL_CONFIG_PATH
, &path
) == 0);
898 * If we have a raidz/mirror that combines disks
899 * with files, report it as an error.
901 if (!dontreport
&& type
!= NULL
&&
902 strcmp(type
, childtype
) != 0) {
908 "mismatched replication "
909 "level: %s contains both "
910 "files and devices\n"),
918 * According to stat(2), the value of 'st_size'
919 * is undefined for block devices and character
920 * devices. But there is no effective way to
921 * determine the real size in userland.
923 * Instead, we'll take advantage of an
924 * implementation detail of spec_size(). If the
925 * device is currently open, then we (should)
926 * return a valid size.
928 * If we still don't get a valid size (indicated
929 * by a size of 0 or MAXOFFSET_T), then ignore
930 * this device altogether.
932 if ((fd
= open(path
, O_RDONLY
)) >= 0) {
933 err
= fstat64(fd
, &statbuf
);
936 err
= stat64(path
, &statbuf
);
940 statbuf
.st_size
== 0 ||
941 statbuf
.st_size
== MAXOFFSET_T
)
944 size
= statbuf
.st_size
;
947 * Also make sure that devices and
948 * slices have a consistent size. If
949 * they differ by a significant amount
950 * (~16MB) then report an error.
953 (vdev_size
!= -1ULL &&
954 (labs(size
- vdev_size
) >
961 "%s contains devices of "
962 "different sizes\n"),
975 * At this point, we have the replication of the last toplevel
976 * vdev in 'rep'. Compare it to 'lastrep' to see if its
979 if (lastrep
.zprl_type
!= NULL
) {
980 if (strcmp(lastrep
.zprl_type
, rep
.zprl_type
) != 0) {
986 "mismatched replication level: "
987 "both %s and %s vdevs are "
989 lastrep
.zprl_type
, rep
.zprl_type
);
992 } else if (lastrep
.zprl_parity
!= rep
.zprl_parity
) {
998 "mismatched replication level: "
999 "both %llu and %llu device parity "
1000 "%s vdevs are present\n"),
1001 lastrep
.zprl_parity
,
1006 } else if (lastrep
.zprl_children
!= rep
.zprl_children
) {
1012 "mismatched replication level: "
1013 "both %llu-way and %llu-way %s "
1014 "vdevs are present\n"),
1015 lastrep
.zprl_children
,
1032 * Check the replication level of the vdev spec against the current pool. Calls
1033 * get_replication() to make sure the new spec is self-consistent. If the pool
1034 * has a consistent replication level, then we ignore any errors. Otherwise,
1035 * report any difference between the two.
1038 check_replication(nvlist_t
*config
, nvlist_t
*newroot
)
1042 replication_level_t
*current
= NULL
, *new;
1046 * If we have a current pool configuration, check to see if it's
1047 * self-consistent. If not, simply return success.
1049 if (config
!= NULL
) {
1052 verify(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
1054 if ((current
= get_replication(nvroot
, B_FALSE
)) == NULL
)
1058 * for spares there may be no children, and therefore no
1059 * replication level to check
1061 if ((nvlist_lookup_nvlist_array(newroot
, ZPOOL_CONFIG_CHILDREN
,
1062 &child
, &children
) != 0) || (children
== 0)) {
1068 * If all we have is logs then there's no replication level to check.
1070 if (num_logs(newroot
) == children
) {
1076 * Get the replication level of the new vdev spec, reporting any
1077 * inconsistencies found.
1079 if ((new = get_replication(newroot
, B_TRUE
)) == NULL
) {
1085 * Check to see if the new vdev spec matches the replication level of
1089 if (current
!= NULL
) {
1090 if (strcmp(current
->zprl_type
, new->zprl_type
) != 0) {
1092 "mismatched replication level: pool uses %s "
1093 "and new vdev is %s\n"),
1094 current
->zprl_type
, new->zprl_type
);
1096 } else if (current
->zprl_parity
!= new->zprl_parity
) {
1098 "mismatched replication level: pool uses %llu "
1099 "device parity and new vdev uses %llu\n"),
1100 current
->zprl_parity
, new->zprl_parity
);
1102 } else if (current
->zprl_children
!= new->zprl_children
) {
1104 "mismatched replication level: pool uses %llu-way "
1105 "%s and new vdev uses %llu-way %s\n"),
1106 current
->zprl_children
, current
->zprl_type
,
1107 new->zprl_children
, new->zprl_type
);
1113 if (current
!= NULL
)
1120 zero_label(char *path
)
1122 const int size
= 4096;
1126 if ((fd
= open(path
, O_WRONLY
|O_EXCL
)) < 0) {
1127 (void) fprintf(stderr
, gettext("cannot open '%s': %s\n"),
1128 path
, strerror(errno
));
1132 memset(buf
, 0, size
);
1133 err
= write(fd
, buf
, size
);
1134 (void) fdatasync(fd
);
1138 (void) fprintf(stderr
, gettext("cannot zero first %d bytes "
1139 "of '%s': %s\n"), size
, path
, strerror(errno
));
1144 (void) fprintf(stderr
, gettext("could only zero %d/%d bytes "
1145 "of '%s'\n"), err
, size
, path
);
1153 * Go through and find any whole disks in the vdev specification, labelling them
1154 * as appropriate. When constructing the vdev spec, we were unable to open this
1155 * device in order to provide a devid. Now that we have labelled the disk and
1156 * know that slice 0 is valid, we can construct the devid now.
1158 * If the disk was already labeled with an EFI label, we will have gotten the
1159 * devid already (because we were able to open the whole disk). Otherwise, we
1160 * need to get the devid after we label the disk.
1163 make_disks(zpool_handle_t
*zhp
, nvlist_t
*nv
)
1168 char devpath
[MAXPATHLEN
];
1169 char udevpath
[MAXPATHLEN
];
1171 struct stat64 statbuf
;
1172 int is_exclusive
= 0;
1176 verify(nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) == 0);
1178 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1179 &child
, &children
) != 0) {
1181 if (strcmp(type
, VDEV_TYPE_DISK
) != 0)
1185 * We have a disk device. If this is a whole disk write
1186 * out the efi partition table, otherwise write zero's to
1187 * the first 4k of the partition. This is to ensure that
1188 * libblkid will not misidentify the partition due to a
1189 * magic value left by the previous filesystem.
1191 verify(!nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &path
));
1192 verify(!nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
1196 (void) zero_label(path
);
1200 if (realpath(path
, devpath
) == NULL
) {
1202 (void) fprintf(stderr
,
1203 gettext("cannot resolve path '%s'\n"), path
);
1208 * Remove any previously existing symlink from a udev path to
1209 * the device before labeling the disk. This makes
1210 * zpool_label_disk_wait() truly wait for the new link to show
1211 * up instead of returning if it finds an old link still in
1212 * place. Otherwise there is a window between when udev
1213 * deletes and recreates the link during which access attempts
1214 * will fail with ENOENT.
1216 strncpy(udevpath
, path
, MAXPATHLEN
);
1217 (void) zfs_append_partition(udevpath
, MAXPATHLEN
);
1219 fd
= open(devpath
, O_RDWR
|O_EXCL
);
1228 * If the partition exists, contains a valid spare label,
1229 * and is opened exclusively there is no need to partition
1230 * it. Hot spares have already been partitioned and are
1231 * held open exclusively by the kernel as a safety measure.
1233 * If the provided path is for a /dev/disk/ device its
1234 * symbolic link will be removed, partition table created,
1235 * and then block until udev creates the new link.
1237 if (!is_exclusive
|| !is_spare(NULL
, udevpath
)) {
1238 ret
= strncmp(udevpath
, UDISK_ROOT
, strlen(UDISK_ROOT
));
1240 ret
= lstat64(udevpath
, &statbuf
);
1241 if (ret
== 0 && S_ISLNK(statbuf
.st_mode
))
1242 (void) unlink(udevpath
);
1245 if (zpool_label_disk(g_zfs
, zhp
,
1246 strrchr(devpath
, '/') + 1) == -1)
1249 ret
= zpool_label_disk_wait(udevpath
, DISK_LABEL_WAIT
);
1251 (void) fprintf(stderr
, gettext("cannot "
1252 "resolve path '%s': %d\n"), udevpath
, ret
);
1256 (void) 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);
1270 for (c
= 0; c
< children
; c
++)
1271 if ((ret
= make_disks(zhp
, child
[c
])) != 0)
1274 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_SPARES
,
1275 &child
, &children
) == 0)
1276 for (c
= 0; c
< children
; c
++)
1277 if ((ret
= make_disks(zhp
, child
[c
])) != 0)
1280 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_L2CACHE
,
1281 &child
, &children
) == 0)
1282 for (c
= 0; c
< children
; c
++)
1283 if ((ret
= make_disks(zhp
, child
[c
])) != 0)
1290 * Go through and find any devices that are in use. We rely on libdiskmgt for
1291 * the majority of this task.
1294 is_device_in_use(nvlist_t
*config
, nvlist_t
*nv
, boolean_t force
,
1295 boolean_t replacing
, boolean_t isspare
)
1301 char buf
[MAXPATHLEN
];
1302 uint64_t wholedisk
= B_FALSE
;
1303 boolean_t anyinuse
= B_FALSE
;
1305 verify(nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) == 0);
1307 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1308 &child
, &children
) != 0) {
1310 verify(!nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &path
));
1311 if (strcmp(type
, VDEV_TYPE_DISK
) == 0)
1312 verify(!nvlist_lookup_uint64(nv
,
1313 ZPOOL_CONFIG_WHOLE_DISK
, &wholedisk
));
1316 * As a generic check, we look to see if this is a replace of a
1317 * hot spare within the same pool. If so, we allow it
1318 * regardless of what libblkid or zpool_in_use() says.
1321 (void) strlcpy(buf
, path
, sizeof (buf
));
1323 ret
= zfs_append_partition(buf
, sizeof (buf
));
1328 if (is_spare(config
, buf
))
1332 if (strcmp(type
, VDEV_TYPE_DISK
) == 0)
1333 ret
= check_device(path
, force
, isspare
, wholedisk
);
1335 else if (strcmp(type
, VDEV_TYPE_FILE
) == 0)
1336 ret
= check_file(path
, force
, isspare
);
1341 for (c
= 0; c
< children
; c
++)
1342 if (is_device_in_use(config
, child
[c
], force
, replacing
,
1346 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_SPARES
,
1347 &child
, &children
) == 0)
1348 for (c
= 0; c
< children
; c
++)
1349 if (is_device_in_use(config
, child
[c
], force
, replacing
,
1353 if (nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_L2CACHE
,
1354 &child
, &children
) == 0)
1355 for (c
= 0; c
< children
; c
++)
1356 if (is_device_in_use(config
, child
[c
], force
, replacing
,
1364 is_grouping(const char *type
, int *mindev
, int *maxdev
)
1366 if (strncmp(type
, "raidz", 5) == 0) {
1367 const char *p
= type
+ 5;
1373 } else if (*p
== '0') {
1374 return (NULL
); /* no zero prefixes allowed */
1377 nparity
= strtol(p
, &end
, 10);
1378 if (errno
!= 0 || nparity
< 1 || nparity
>= 255 ||
1384 *mindev
= nparity
+ 1;
1387 return (VDEV_TYPE_RAIDZ
);
1393 if (strcmp(type
, "mirror") == 0) {
1396 return (VDEV_TYPE_MIRROR
);
1399 if (strcmp(type
, "spare") == 0) {
1402 return (VDEV_TYPE_SPARE
);
1405 if (strcmp(type
, "log") == 0) {
1408 return (VDEV_TYPE_LOG
);
1411 if (strcmp(type
, "cache") == 0) {
1414 return (VDEV_TYPE_L2CACHE
);
1421 * Construct a syntactically valid vdev specification,
1422 * and ensure that all devices and files exist and can be opened.
1423 * Note: we don't bother freeing anything in the error paths
1424 * because the program is just going to exit anyway.
1427 construct_spec(nvlist_t
*props
, int argc
, char **argv
)
1429 nvlist_t
*nvroot
, *nv
, **top
, **spares
, **l2cache
;
1430 int t
, toplevels
, mindev
, maxdev
, nspares
, nlogs
, nl2cache
;
1433 boolean_t seen_logs
;
1443 seen_logs
= B_FALSE
;
1449 * If it's a mirror or raidz, the subsequent arguments are
1450 * its leaves -- until we encounter the next mirror or raidz.
1452 if ((type
= is_grouping(argv
[0], &mindev
, &maxdev
)) != NULL
) {
1453 nvlist_t
**child
= NULL
;
1454 int c
, children
= 0;
1456 if (strcmp(type
, VDEV_TYPE_SPARE
) == 0) {
1457 if (spares
!= NULL
) {
1458 (void) fprintf(stderr
,
1459 gettext("invalid vdev "
1460 "specification: 'spare' can be "
1461 "specified only once\n"));
1467 if (strcmp(type
, VDEV_TYPE_LOG
) == 0) {
1469 (void) fprintf(stderr
,
1470 gettext("invalid vdev "
1471 "specification: 'log' can be "
1472 "specified only once\n"));
1480 * A log is not a real grouping device.
1481 * We just set is_log and continue.
1486 if (strcmp(type
, VDEV_TYPE_L2CACHE
) == 0) {
1487 if (l2cache
!= NULL
) {
1488 (void) fprintf(stderr
,
1489 gettext("invalid vdev "
1490 "specification: 'cache' can be "
1491 "specified only once\n"));
1498 if (strcmp(type
, VDEV_TYPE_MIRROR
) != 0) {
1499 (void) fprintf(stderr
,
1500 gettext("invalid vdev "
1501 "specification: unsupported 'log' "
1502 "device: %s\n"), type
);
1508 for (c
= 1; c
< argc
; c
++) {
1509 if (is_grouping(argv
[c
], NULL
, NULL
) != NULL
)
1512 child
= realloc(child
,
1513 children
* sizeof (nvlist_t
*));
1516 if ((nv
= make_leaf_vdev(props
, argv
[c
],
1519 child
[children
- 1] = nv
;
1522 if (children
< mindev
) {
1523 (void) fprintf(stderr
, gettext("invalid vdev "
1524 "specification: %s requires at least %d "
1525 "devices\n"), argv
[0], mindev
);
1529 if (children
> maxdev
) {
1530 (void) fprintf(stderr
, gettext("invalid vdev "
1531 "specification: %s supports no more than "
1532 "%d devices\n"), argv
[0], maxdev
);
1539 if (strcmp(type
, VDEV_TYPE_SPARE
) == 0) {
1543 } else if (strcmp(type
, VDEV_TYPE_L2CACHE
) == 0) {
1545 nl2cache
= children
;
1548 verify(nvlist_alloc(&nv
, NV_UNIQUE_NAME
,
1550 verify(nvlist_add_string(nv
, ZPOOL_CONFIG_TYPE
,
1552 verify(nvlist_add_uint64(nv
,
1553 ZPOOL_CONFIG_IS_LOG
, is_log
) == 0);
1554 if (strcmp(type
, VDEV_TYPE_RAIDZ
) == 0) {
1555 verify(nvlist_add_uint64(nv
,
1556 ZPOOL_CONFIG_NPARITY
,
1559 verify(nvlist_add_nvlist_array(nv
,
1560 ZPOOL_CONFIG_CHILDREN
, child
,
1563 for (c
= 0; c
< children
; c
++)
1564 nvlist_free(child
[c
]);
1569 * We have a device. Pass off to make_leaf_vdev() to
1570 * construct the appropriate nvlist describing the vdev.
1572 if ((nv
= make_leaf_vdev(props
, argv
[0],
1582 top
= realloc(top
, toplevels
* sizeof (nvlist_t
*));
1585 top
[toplevels
- 1] = nv
;
1588 if (toplevels
== 0 && nspares
== 0 && nl2cache
== 0) {
1589 (void) fprintf(stderr
, gettext("invalid vdev "
1590 "specification: at least one toplevel vdev must be "
1595 if (seen_logs
&& nlogs
== 0) {
1596 (void) fprintf(stderr
, gettext("invalid vdev specification: "
1597 "log requires at least 1 device\n"));
1602 * Finally, create nvroot and add all top-level vdevs to it.
1604 verify(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, 0) == 0);
1605 verify(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
1606 VDEV_TYPE_ROOT
) == 0);
1607 verify(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
1608 top
, toplevels
) == 0);
1610 verify(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
1611 spares
, nspares
) == 0);
1613 verify(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
1614 l2cache
, nl2cache
) == 0);
1616 for (t
= 0; t
< toplevels
; t
++)
1617 nvlist_free(top
[t
]);
1618 for (t
= 0; t
< nspares
; t
++)
1619 nvlist_free(spares
[t
]);
1620 for (t
= 0; t
< nl2cache
; t
++)
1621 nvlist_free(l2cache
[t
]);
1632 split_mirror_vdev(zpool_handle_t
*zhp
, char *newname
, nvlist_t
*props
,
1633 splitflags_t flags
, int argc
, char **argv
)
1635 nvlist_t
*newroot
= NULL
, **child
;
1639 if ((newroot
= construct_spec(props
, argc
, argv
)) == NULL
) {
1640 (void) fprintf(stderr
, gettext("Unable to build a "
1641 "pool from the specified devices\n"));
1645 if (!flags
.dryrun
&& make_disks(zhp
, newroot
) != 0) {
1646 nvlist_free(newroot
);
1650 /* avoid any tricks in the spec */
1651 verify(nvlist_lookup_nvlist_array(newroot
,
1652 ZPOOL_CONFIG_CHILDREN
, &child
, &children
) == 0);
1653 for (c
= 0; c
< children
; c
++) {
1658 verify(nvlist_lookup_string(child
[c
],
1659 ZPOOL_CONFIG_PATH
, &path
) == 0);
1660 if ((type
= is_grouping(path
, &min
, &max
)) != NULL
) {
1661 (void) fprintf(stderr
, gettext("Cannot use "
1662 "'%s' as a device for splitting\n"), type
);
1663 nvlist_free(newroot
);
1669 if (zpool_vdev_split(zhp
, newname
, &newroot
, props
, flags
) != 0) {
1670 if (newroot
!= NULL
)
1671 nvlist_free(newroot
);
1679 * Get and validate the contents of the given vdev specification. This ensures
1680 * that the nvlist returned is well-formed, that all the devices exist, and that
1681 * they are not currently in use by any other known consumer. The 'poolconfig'
1682 * parameter is the current configuration of the pool when adding devices
1683 * existing pool, and is used to perform additional checks, such as changing the
1684 * replication level of the pool. It can be 'NULL' to indicate that this is a
1685 * new pool. The 'force' flag controls whether devices should be forcefully
1686 * added, even if they appear in use.
1689 make_root_vdev(zpool_handle_t
*zhp
, nvlist_t
*props
, int force
, int check_rep
,
1690 boolean_t replacing
, boolean_t dryrun
, int argc
, char **argv
)
1693 nvlist_t
*poolconfig
= NULL
;
1697 * Construct the vdev specification. If this is successful, we know
1698 * that we have a valid specification, and that all devices can be
1701 if ((newroot
= construct_spec(props
, argc
, argv
)) == NULL
)
1704 if (zhp
&& ((poolconfig
= zpool_get_config(zhp
, NULL
)) == NULL
)) {
1705 nvlist_free(newroot
);
1710 * Validate each device to make sure that its not shared with another
1711 * subsystem. We do this even if 'force' is set, because there are some
1712 * uses (such as a dedicated dump device) that even '-f' cannot
1715 if (is_device_in_use(poolconfig
, newroot
, force
, replacing
, B_FALSE
)) {
1716 nvlist_free(newroot
);
1721 * Check the replication level of the given vdevs and report any errors
1722 * found. We include the existing pool spec, if any, as we need to
1723 * catch changes against the existing replication level.
1725 if (check_rep
&& check_replication(poolconfig
, newroot
) != 0) {
1726 nvlist_free(newroot
);
1731 * Run through the vdev specification and label any whole disks found.
1733 if (!dryrun
&& make_disks(zhp
, newroot
) != 0) {
1734 nvlist_free(newroot
);