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>
81 #include <blkid/blkid.h>
82 #include "zpool_util.h"
83 #include <sys/zfs_context.h>
86 * For any given vdev specification, we can have multiple errors. The
87 * vdev_error() function keeps track of whether we have seen an error yet, and
88 * prints out a header if its the first error we've seen.
93 typedef struct vdev_disk_db_entry
97 } vdev_disk_db_entry_t;
100 * Database of block devices that lie about physical sector sizes. The
101 * identification string must be precisely 24 characters to avoid false
104 static vdev_disk_db_entry_t vdev_disk_database[] = {
105 {"ATA ADATA SSD S396 3", 8192},
106 {"ATA APPLE SSD SM128E", 8192},
107 {"ATA APPLE SSD SM256E", 8192},
108 {"ATA APPLE SSD SM512E", 8192},
109 {"ATA APPLE SSD SM768E", 8192},
110 {"ATA C400-MTFDDAC064M", 8192},
111 {"ATA C400-MTFDDAC128M", 8192},
112 {"ATA C400-MTFDDAC256M", 8192},
113 {"ATA C400-MTFDDAC512M", 8192},
114 {"ATA Corsair Force 3 ", 8192},
115 {"ATA Corsair Force GS", 8192},
116 {"ATA INTEL SSDSA2CT04", 8192},
117 {"ATA INTEL SSDSA2BZ10", 8192},
118 {"ATA INTEL SSDSA2BZ20", 8192},
119 {"ATA INTEL SSDSA2BZ30", 8192},
120 {"ATA INTEL SSDSA2CW04", 8192},
121 {"ATA INTEL SSDSA2CW08", 8192},
122 {"ATA INTEL SSDSA2CW12", 8192},
123 {"ATA INTEL SSDSA2CW16", 8192},
124 {"ATA INTEL SSDSA2CW30", 8192},
125 {"ATA INTEL SSDSA2CW60", 8192},
126 {"ATA INTEL SSDSC2CT06", 8192},
127 {"ATA INTEL SSDSC2CT12", 8192},
128 {"ATA INTEL SSDSC2CT18", 8192},
129 {"ATA INTEL SSDSC2CT24", 8192},
130 {"ATA INTEL SSDSC2CW06", 8192},
131 {"ATA INTEL SSDSC2CW12", 8192},
132 {"ATA INTEL SSDSC2CW18", 8192},
133 {"ATA INTEL SSDSC2CW24", 8192},
134 {"ATA INTEL SSDSC2CW48", 8192},
135 {"ATA KINGSTON SH100S3", 8192},
136 {"ATA KINGSTON SH103S3", 8192},
137 {"ATA M4-CT064M4SSD2 ", 8192},
138 {"ATA M4-CT128M4SSD2 ", 8192},
139 {"ATA M4-CT256M4SSD2 ", 8192},
140 {"ATA M4-CT512M4SSD2 ", 8192},
141 {"ATA OCZ-AGILITY2 ", 8192},
142 {"ATA OCZ-AGILITY3 ", 8192},
143 {"ATA OCZ-VERTEX2 3.5 ", 8192},
144 {"ATA OCZ-VERTEX3 ", 8192},
145 {"ATA OCZ-VERTEX3 LT ", 8192},
146 {"ATA OCZ-VERTEX3 MI ", 8192},
147 {"ATA OCZ-VERTEX4 ", 8192},
148 {"ATA SAMSUNG MZ7WD120", 8192},
149 {"ATA SAMSUNG MZ7WD240", 8192},
150 {"ATA SAMSUNG MZ7WD480", 8192},
151 {"ATA SAMSUNG MZ7WD960", 8192},
152 {"ATA SAMSUNG SSD 830 ", 8192},
153 {"ATA Samsung SSD 840 ", 8192},
154 {"ATA SanDisk SSD U100", 8192},
155 {"ATA TOSHIBA THNSNH06", 8192},
156 {"ATA TOSHIBA THNSNH12", 8192},
157 {"ATA TOSHIBA THNSNH25", 8192},
158 {"ATA TOSHIBA THNSNH51", 8192},
159 {"ATA APPLE SSD TS064C", 4096},
160 {"ATA APPLE SSD TS128C", 4096},
161 {"ATA APPLE SSD TS256C", 4096},
162 {"ATA APPLE SSD TS512C", 4096},
163 {"ATA INTEL SSDSA2M040", 4096},
164 {"ATA INTEL SSDSA2M080", 4096},
165 {"ATA INTEL SSDSA2M160", 4096},
166 {"ATA INTEL SSDSC2MH12", 4096},
167 {"ATA INTEL SSDSC2MH25", 4096},
168 {"ATA OCZ CORE_SSD ", 4096},
169 {"ATA OCZ-VERTEX ", 4096},
170 {"ATA SAMSUNG MCCOE32G", 4096},
171 {"ATA SAMSUNG MCCOE64G", 4096},
172 {"ATA SAMSUNG SSD PM80", 4096},
173 /* Flash drives optimized for 4KB IOs on larger pages */
174 {"ATA INTEL SSDSC2BA10", 4096},
175 {"ATA INTEL SSDSC2BA20", 4096},
176 {"ATA INTEL SSDSC2BA40", 4096},
177 {"ATA INTEL SSDSC2BA80", 4096},
178 {"ATA INTEL SSDSC2BB08", 4096},
179 {"ATA INTEL SSDSC2BB12", 4096},
180 {"ATA INTEL SSDSC2BB16", 4096},
181 {"ATA INTEL SSDSC2BB24", 4096},
182 {"ATA INTEL SSDSC2BB30", 4096},
183 {"ATA INTEL SSDSC2BB40", 4096},
184 {"ATA INTEL SSDSC2BB48", 4096},
185 {"ATA INTEL SSDSC2BB60", 4096},
186 {"ATA INTEL SSDSC2BB80", 4096},
187 {"ATA INTEL SSDSC2BW24", 4096},
188 {"ATA INTEL SSDSC2BP24", 4096},
189 {"ATA INTEL SSDSC2BP48", 4096},
190 {"NA SmrtStorSDLKAE9W", 4096},
191 /* Imported from Open Solaris */
192 {"ATA MARVELL SD88SA02", 4096},
193 /* Advanced format Hard drives */
194 {"ATA Hitachi HDS5C303", 4096},
195 {"ATA SAMSUNG HD204UI ", 4096},
196 {"ATA ST2000DL004 HD20", 4096},
197 {"ATA WDC WD10EARS-00M", 4096},
198 {"ATA WDC WD10EARS-00S", 4096},
199 {"ATA WDC WD10EARS-00Z", 4096},
200 {"ATA WDC WD15EARS-00M", 4096},
201 {"ATA WDC WD15EARS-00S", 4096},
202 {"ATA WDC WD15EARS-00Z", 4096},
203 {"ATA WDC WD20EARS-00M", 4096},
204 {"ATA WDC WD20EARS-00S", 4096},
205 {"ATA WDC WD20EARS-00Z", 4096},
206 {"ATA WDC WD1600BEVT-0", 4096},
207 {"ATA WDC WD2500BEVT-0", 4096},
208 {"ATA WDC WD3200BEVT-0", 4096},
209 {"ATA WDC WD5000BEVT-0", 4096},
210 /* Virtual disks: Assume zvols with default volblocksize */
212 {"ATA QEMU HARDDISK ", 8192},
213 {"IET VIRTUAL-DISK ", 8192},
214 {"OI COMSTAR ", 8192},
215 {"SUN COMSTAR ", 8192},
216 {"NETAPP LUN ", 8192},
220 static const int vdev_disk_database_size =
221 sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
223 #define INQ_REPLY_LEN 96
224 #define INQ_CMD_LEN 6
227 check_sector_size_database(char *path, int *sector_size)
229 unsigned char inq_buff[INQ_REPLY_LEN];
230 unsigned char sense_buffer[32];
231 unsigned char inq_cmd_blk[INQ_CMD_LEN] =
232 {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
238 /* Prepare INQUIRY command */
239 memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
240 io_hdr.interface_id = 'S';
241 io_hdr.cmd_len = sizeof (inq_cmd_blk);
242 io_hdr.mx_sb_len = sizeof (sense_buffer);
243 io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
244 io_hdr.dxfer_len = INQ_REPLY_LEN;
245 io_hdr.dxferp = inq_buff;
246 io_hdr.cmdp = inq_cmd_blk;
247 io_hdr.sbp = sense_buffer;
248 io_hdr.timeout = 10; /* 10 milliseconds is ample time */
250 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
253 error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
260 if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
263 for (i = 0; i < vdev_disk_database_size; i++) {
264 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
267 *sector_size = vdev_disk_database[i].sector_size;
276 vdev_error(const char *fmt, ...)
281 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
283 (void) fprintf(stderr, gettext("use '-f' to override "
284 "the following errors:\n"));
286 (void) fprintf(stderr, gettext("the following errors "
287 "must be manually repaired:\n"));
292 (void) vfprintf(stderr, fmt, ap);
297 * Check that a file is valid. All we can do in this case is check that it's
298 * not in use by another pool, and not in use by swap.
301 check_file(const char *file, boolean_t force, boolean_t isspare)
309 if ((fd = open(file, O_RDONLY)) < 0)
312 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
316 case POOL_STATE_ACTIVE:
317 desc = gettext("active");
320 case POOL_STATE_EXPORTED:
321 desc = gettext("exported");
324 case POOL_STATE_POTENTIALLY_ACTIVE:
325 desc = gettext("potentially active");
329 desc = gettext("unknown");
334 * Allow hot spares to be shared between pools.
336 if (state == POOL_STATE_SPARE && isspare)
339 if (state == POOL_STATE_ACTIVE ||
340 state == POOL_STATE_SPARE || !force) {
342 case POOL_STATE_SPARE:
343 vdev_error(gettext("%s is reserved as a hot "
344 "spare for pool %s\n"), file, name);
347 vdev_error(gettext("%s is part of %s pool "
348 "'%s'\n"), file, desc, name);
364 (void) fprintf(stderr, gettext("warning: device in use checking "
365 "failed: %s\n"), strerror(err));
369 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
374 /* No valid type detected device is safe to use */
375 value = blkid_get_tag_value(cache, "TYPE", path);
380 * If libblkid detects a ZFS device, we check the device
381 * using check_file() to see if it's safe. The one safe
382 * case is a spare device shared between multiple pools.
384 if (strcmp(value, "zfs_member") == 0) {
385 err = check_file(path, force, isspare);
391 vdev_error(gettext("%s contains a filesystem of "
392 "type '%s'\n"), path, value);
402 * Validate that a disk including all partitions are safe to use.
404 * For EFI labeled disks this can done relatively easily with the libefi
405 * library. The partition numbers are extracted from the label and used
406 * to generate the expected /dev/ paths. Each partition can then be
407 * checked for conflicts.
409 * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
410 * but due to the lack of a readily available libraries this scanning is
411 * not implemented. Instead only the device path as given is checked.
414 check_disk(const char *path, blkid_cache cache, int force,
415 boolean_t isspare, boolean_t iswholedisk)
418 char slice_path[MAXPATHLEN];
423 return (check_slice(path, cache, force, isspare));
425 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
431 * Expected to fail for non-EFI labled disks. Just check the device
432 * as given and do not attempt to detect and scan partitions.
434 err = efi_alloc_and_read(fd, &vtoc);
437 return (check_slice(path, cache, force, isspare));
441 * The primary efi partition label is damaged however the secondary
442 * label at the end of the device is intact. Rather than use this
443 * label we should play it safe and treat this as a non efi device.
445 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
450 /* Partitions will no be created using the backup */
453 vdev_error(gettext("%s contains a corrupt primary "
454 "EFI label.\n"), path);
459 for (i = 0; i < vtoc->efi_nparts; i++) {
461 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
462 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
465 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
466 (void) snprintf(slice_path, sizeof (slice_path),
467 "%s%s%d", path, "-part", i+1);
469 (void) snprintf(slice_path, sizeof (slice_path),
470 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
473 err = check_slice(slice_path, cache, force, isspare);
485 check_device(const char *path, boolean_t force,
486 boolean_t isspare, boolean_t iswholedisk)
488 static blkid_cache cache = NULL;
491 * There is no easy way to add a correct blkid_put_cache() call,
492 * memory will be reclaimed when the command exits.
497 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
502 if ((err = blkid_probe_all(cache)) != 0) {
503 blkid_put_cache(cache);
509 return (check_disk(path, cache, force, isspare, iswholedisk));
513 * By "whole disk" we mean an entire physical disk (something we can
514 * label, toggle the write cache on, etc.) as opposed to the full
515 * capacity of a pseudo-device such as lofi or did. We act as if we
516 * are labeling the disk, which should be a pretty good test of whether
517 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
521 is_whole_disk(const char *path)
523 struct dk_gpt *label;
526 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
528 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
538 * This may be a shorthand device path or it could be total gibberish.
539 * Check to see if it is a known device available in zfs_vdev_paths.
540 * As part of this check, see if we've been given an entire disk
541 * (minus the slice number).
544 is_shorthand_path(const char *arg, char *path,
545 struct stat64 *statbuf, boolean_t *wholedisk)
549 error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
551 *wholedisk = is_whole_disk(path);
552 if (*wholedisk || (stat64(path, statbuf) == 0))
556 strlcpy(path, arg, sizeof (path));
557 memset(statbuf, 0, sizeof (*statbuf));
558 *wholedisk = B_FALSE;
564 * Determine if the given path is a hot spare within the given configuration.
565 * If no configuration is given we rely solely on the label.
568 is_spare(nvlist_t *config, const char *path)
574 uint64_t guid, spareguid;
580 if ((fd = open(path, O_RDONLY)) < 0)
583 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
585 state != POOL_STATE_SPARE ||
586 zpool_read_label(fd, &label, NULL) != 0) {
597 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
600 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
602 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
603 &spares, &nspares) == 0) {
604 for (i = 0; i < nspares; i++) {
605 verify(nvlist_lookup_uint64(spares[i],
606 ZPOOL_CONFIG_GUID, &spareguid) == 0);
607 if (spareguid == guid)
616 * Create a leaf vdev. Determine if this is a file or a device. If it's a
617 * device, fill in the device id to make a complete nvlist. Valid forms for a
620 * /dev/xxx Complete disk path
621 * /xxx Full path to file
622 * xxx Shorthand for <zfs_vdev_paths>/xxx
625 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
627 char path[MAXPATHLEN];
628 struct stat64 statbuf;
629 nvlist_t *vdev = NULL;
631 boolean_t wholedisk = B_FALSE;
636 * Determine what type of vdev this is, and put the full path into
637 * 'path'. We detect whether this is a device of file afterwards by
638 * checking the st_mode of the file.
642 * Complete device or file path. Exact type is determined by
643 * examining the file descriptor afterwards. Symbolic links
644 * are resolved to their real paths for the is_whole_disk()
645 * and S_ISBLK/S_ISREG type checks. However, we are careful
646 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
647 * can leverage udev's persistent device labels.
649 if (realpath(arg, path) == NULL) {
650 (void) fprintf(stderr,
651 gettext("cannot resolve path '%s'\n"), arg);
655 wholedisk = is_whole_disk(path);
656 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
657 (void) fprintf(stderr,
658 gettext("cannot open '%s': %s\n"),
659 path, strerror(errno));
663 /* After is_whole_disk() check restore original passed path */
664 strlcpy(path, arg, MAXPATHLEN);
666 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
669 * If we got ENOENT, then the user gave us
670 * gibberish, so try to direct them with a
671 * reasonable error message. Otherwise,
672 * regurgitate strerror() since it's the best we
676 (void) fprintf(stderr,
677 gettext("cannot open '%s': no such "
678 "device in %s\n"), arg, DISK_ROOT);
679 (void) fprintf(stderr,
680 gettext("must be a full path or "
681 "shorthand device name\n"));
684 (void) fprintf(stderr,
685 gettext("cannot open '%s': %s\n"),
686 path, strerror(errno));
693 * Determine whether this is a device or a file.
695 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
696 type = VDEV_TYPE_DISK;
697 } else if (S_ISREG(statbuf.st_mode)) {
698 type = VDEV_TYPE_FILE;
700 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
701 "block device or regular file\n"), path);
706 * Finally, we have the complete device or file, and we know that it is
707 * acceptable to use. Construct the nvlist to describe this vdev. All
708 * vdevs have a 'path' element, and devices also have a 'devid' element.
710 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
711 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
712 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
713 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
714 if (strcmp(type, VDEV_TYPE_DISK) == 0)
715 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
716 (uint64_t)wholedisk) == 0);
719 * Override defaults if custom properties are provided.
724 if (nvlist_lookup_string(props,
725 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
726 zfs_nicestrtonum(NULL, value, &ashift);
730 * If the device is known to incorrectly report its physical sector
731 * size explicitly provide the known correct value.
736 if (check_sector_size_database(path, §or_size) == B_TRUE)
737 ashift = highbit64(sector_size) - 1;
741 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
747 * Go through and verify the replication level of the pool is consistent.
748 * Performs the following checks:
750 * For the new spec, verifies that devices in mirrors and raidz are the
753 * If the current configuration already has inconsistent replication
754 * levels, ignore any other potential problems in the new spec.
756 * Otherwise, make sure that the current spec (if there is one) and the new
757 * spec have consistent replication levels.
759 typedef struct replication_level {
761 uint64_t zprl_children;
762 uint64_t zprl_parity;
763 } replication_level_t;
765 #define ZPOOL_FUZZ (16 * 1024 * 1024)
768 * Given a list of toplevel vdevs, return the current replication level. If
769 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
770 * an error message will be displayed for each self-inconsistent vdev.
772 static replication_level_t *
773 get_replication(nvlist_t *nvroot, boolean_t fatal)
781 replication_level_t lastrep = { 0 }, rep, *ret;
782 boolean_t dontreport;
784 ret = safe_malloc(sizeof (replication_level_t));
786 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
787 &top, &toplevels) == 0);
789 lastrep.zprl_type = NULL;
790 for (t = 0; t < toplevels; t++) {
791 uint64_t is_log = B_FALSE;
796 * For separate logs we ignore the top level vdev replication
799 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
803 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
805 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
806 &child, &children) != 0) {
808 * This is a 'file' or 'disk' vdev.
810 rep.zprl_type = type;
811 rep.zprl_children = 1;
817 * This is a mirror or RAID-Z vdev. Go through and make
818 * sure the contents are all the same (files vs. disks),
819 * keeping track of the number of elements in the
822 * We also check that the size of each vdev (if it can
823 * be determined) is the same.
825 rep.zprl_type = type;
826 rep.zprl_children = 0;
828 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
829 verify(nvlist_lookup_uint64(nv,
830 ZPOOL_CONFIG_NPARITY,
831 &rep.zprl_parity) == 0);
832 assert(rep.zprl_parity != 0);
838 * The 'dontreport' variable indicates that we've
839 * already reported an error for this spec, so don't
840 * bother doing it again.
845 for (c = 0; c < children; c++) {
846 nvlist_t *cnv = child[c];
848 struct stat64 statbuf;
849 uint64_t size = -1ULL;
855 verify(nvlist_lookup_string(cnv,
856 ZPOOL_CONFIG_TYPE, &childtype) == 0);
859 * If this is a replacing or spare vdev, then
860 * get the real first child of the vdev.
862 if (strcmp(childtype,
863 VDEV_TYPE_REPLACING) == 0 ||
864 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
868 verify(nvlist_lookup_nvlist_array(cnv,
869 ZPOOL_CONFIG_CHILDREN, &rchild,
871 assert(rchildren == 2);
874 verify(nvlist_lookup_string(cnv,
879 verify(nvlist_lookup_string(cnv,
880 ZPOOL_CONFIG_PATH, &path) == 0);
883 * If we have a raidz/mirror that combines disks
884 * with files, report it as an error.
886 if (!dontreport && type != NULL &&
887 strcmp(type, childtype) != 0) {
893 "mismatched replication "
894 "level: %s contains both "
895 "files and devices\n"),
903 * According to stat(2), the value of 'st_size'
904 * is undefined for block devices and character
905 * devices. But there is no effective way to
906 * determine the real size in userland.
908 * Instead, we'll take advantage of an
909 * implementation detail of spec_size(). If the
910 * device is currently open, then we (should)
911 * return a valid size.
913 * If we still don't get a valid size (indicated
914 * by a size of 0 or MAXOFFSET_T), then ignore
915 * this device altogether.
917 if ((fd = open(path, O_RDONLY)) >= 0) {
918 err = fstat64(fd, &statbuf);
921 err = stat64(path, &statbuf);
925 statbuf.st_size == 0 ||
926 statbuf.st_size == MAXOFFSET_T)
929 size = statbuf.st_size;
932 * Also make sure that devices and
933 * slices have a consistent size. If
934 * they differ by a significant amount
935 * (~16MB) then report an error.
938 (vdev_size != -1ULL &&
939 (labs(size - vdev_size) >
946 "%s contains devices of "
947 "different sizes\n"),
960 * At this point, we have the replication of the last toplevel
961 * vdev in 'rep'. Compare it to 'lastrep' to see if its
964 if (lastrep.zprl_type != NULL) {
965 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
971 "mismatched replication level: "
972 "both %s and %s vdevs are "
974 lastrep.zprl_type, rep.zprl_type);
977 } else if (lastrep.zprl_parity != rep.zprl_parity) {
983 "mismatched replication level: "
984 "both %llu and %llu device parity "
985 "%s vdevs are present\n"),
991 } else if (lastrep.zprl_children != rep.zprl_children) {
997 "mismatched replication level: "
998 "both %llu-way and %llu-way %s "
999 "vdevs are present\n"),
1000 lastrep.zprl_children,
1017 * Check the replication level of the vdev spec against the current pool. Calls
1018 * get_replication() to make sure the new spec is self-consistent. If the pool
1019 * has a consistent replication level, then we ignore any errors. Otherwise,
1020 * report any difference between the two.
1023 check_replication(nvlist_t *config, nvlist_t *newroot)
1027 replication_level_t *current = NULL, *new;
1031 * If we have a current pool configuration, check to see if it's
1032 * self-consistent. If not, simply return success.
1034 if (config != NULL) {
1037 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1039 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1043 * for spares there may be no children, and therefore no
1044 * replication level to check
1046 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1047 &child, &children) != 0) || (children == 0)) {
1053 * If all we have is logs then there's no replication level to check.
1055 if (num_logs(newroot) == children) {
1061 * Get the replication level of the new vdev spec, reporting any
1062 * inconsistencies found.
1064 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1070 * Check to see if the new vdev spec matches the replication level of
1074 if (current != NULL) {
1075 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1077 "mismatched replication level: pool uses %s "
1078 "and new vdev is %s\n"),
1079 current->zprl_type, new->zprl_type);
1081 } else if (current->zprl_parity != new->zprl_parity) {
1083 "mismatched replication level: pool uses %llu "
1084 "device parity and new vdev uses %llu\n"),
1085 current->zprl_parity, new->zprl_parity);
1087 } else if (current->zprl_children != new->zprl_children) {
1089 "mismatched replication level: pool uses %llu-way "
1090 "%s and new vdev uses %llu-way %s\n"),
1091 current->zprl_children, current->zprl_type,
1092 new->zprl_children, new->zprl_type);
1098 if (current != NULL)
1105 zero_label(char *path)
1107 const int size = 4096;
1111 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1112 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1113 path, strerror(errno));
1117 memset(buf, 0, size);
1118 err = write(fd, buf, size);
1119 (void) fdatasync(fd);
1123 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1124 "of '%s': %s\n"), size, path, strerror(errno));
1129 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1130 "of '%s'\n"), err, size, path);
1138 * Go through and find any whole disks in the vdev specification, labelling them
1139 * as appropriate. When constructing the vdev spec, we were unable to open this
1140 * device in order to provide a devid. Now that we have labelled the disk and
1141 * know that slice 0 is valid, we can construct the devid now.
1143 * If the disk was already labeled with an EFI label, we will have gotten the
1144 * devid already (because we were able to open the whole disk). Otherwise, we
1145 * need to get the devid after we label the disk.
1148 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1153 char devpath[MAXPATHLEN];
1154 char udevpath[MAXPATHLEN];
1156 struct stat64 statbuf;
1157 int is_exclusive = 0;
1161 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1163 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1164 &child, &children) != 0) {
1166 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1170 * We have a disk device. If this is a whole disk write
1171 * out the efi partition table, otherwise write zero's to
1172 * the first 4k of the partition. This is to ensure that
1173 * libblkid will not misidentify the partition due to a
1174 * magic value left by the previous filesystem.
1176 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1177 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1181 (void) zero_label(path);
1185 if (realpath(path, devpath) == NULL) {
1187 (void) fprintf(stderr,
1188 gettext("cannot resolve path '%s'\n"), path);
1193 * Remove any previously existing symlink from a udev path to
1194 * the device before labeling the disk. This makes
1195 * zpool_label_disk_wait() truly wait for the new link to show
1196 * up instead of returning if it finds an old link still in
1197 * place. Otherwise there is a window between when udev
1198 * deletes and recreates the link during which access attempts
1199 * will fail with ENOENT.
1201 strncpy(udevpath, path, MAXPATHLEN);
1202 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1204 fd = open(devpath, O_RDWR|O_EXCL);
1213 * If the partition exists, contains a valid spare label,
1214 * and is opened exclusively there is no need to partition
1215 * it. Hot spares have already been partitioned and are
1216 * held open exclusively by the kernel as a safety measure.
1218 * If the provided path is for a /dev/disk/ device its
1219 * symbolic link will be removed, partition table created,
1220 * and then block until udev creates the new link.
1222 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1223 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1225 ret = lstat64(udevpath, &statbuf);
1226 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1227 (void) unlink(udevpath);
1230 if (zpool_label_disk(g_zfs, zhp,
1231 strrchr(devpath, '/') + 1) == -1)
1234 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1236 (void) fprintf(stderr, gettext("cannot "
1237 "resolve path '%s': %d\n"), udevpath, ret);
1241 (void) zero_label(udevpath);
1245 * Update the path to refer to the partition. The presence of
1246 * the 'whole_disk' field indicates to the CLI that we should
1247 * chop off the partition number when displaying the device in
1250 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1255 for (c = 0; c < children; c++)
1256 if ((ret = make_disks(zhp, child[c])) != 0)
1259 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1260 &child, &children) == 0)
1261 for (c = 0; c < children; c++)
1262 if ((ret = make_disks(zhp, child[c])) != 0)
1265 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1266 &child, &children) == 0)
1267 for (c = 0; c < children; c++)
1268 if ((ret = make_disks(zhp, child[c])) != 0)
1275 * Go through and find any devices that are in use. We rely on libdiskmgt for
1276 * the majority of this task.
1279 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1280 boolean_t replacing, boolean_t isspare)
1286 char buf[MAXPATHLEN];
1287 uint64_t wholedisk = B_FALSE;
1288 boolean_t anyinuse = B_FALSE;
1290 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1292 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1293 &child, &children) != 0) {
1295 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1296 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1297 verify(!nvlist_lookup_uint64(nv,
1298 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1301 * As a generic check, we look to see if this is a replace of a
1302 * hot spare within the same pool. If so, we allow it
1303 * regardless of what libblkid or zpool_in_use() says.
1306 (void) strlcpy(buf, path, sizeof (buf));
1308 ret = zfs_append_partition(buf, sizeof (buf));
1313 if (is_spare(config, buf))
1317 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1318 ret = check_device(path, force, isspare, wholedisk);
1320 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1321 ret = check_file(path, force, isspare);
1326 for (c = 0; c < children; c++)
1327 if (is_device_in_use(config, child[c], force, replacing,
1331 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1332 &child, &children) == 0)
1333 for (c = 0; c < children; c++)
1334 if (is_device_in_use(config, child[c], force, replacing,
1338 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1339 &child, &children) == 0)
1340 for (c = 0; c < children; c++)
1341 if (is_device_in_use(config, child[c], force, replacing,
1349 is_grouping(const char *type, int *mindev, int *maxdev)
1351 if (strncmp(type, "raidz", 5) == 0) {
1352 const char *p = type + 5;
1358 } else if (*p == '0') {
1359 return (NULL); /* no zero prefixes allowed */
1362 nparity = strtol(p, &end, 10);
1363 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1369 *mindev = nparity + 1;
1372 return (VDEV_TYPE_RAIDZ);
1378 if (strcmp(type, "mirror") == 0) {
1381 return (VDEV_TYPE_MIRROR);
1384 if (strcmp(type, "spare") == 0) {
1387 return (VDEV_TYPE_SPARE);
1390 if (strcmp(type, "log") == 0) {
1393 return (VDEV_TYPE_LOG);
1396 if (strcmp(type, "cache") == 0) {
1399 return (VDEV_TYPE_L2CACHE);
1406 * Construct a syntactically valid vdev specification,
1407 * and ensure that all devices and files exist and can be opened.
1408 * Note: we don't bother freeing anything in the error paths
1409 * because the program is just going to exit anyway.
1412 construct_spec(nvlist_t *props, int argc, char **argv)
1414 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1415 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1418 boolean_t seen_logs;
1428 seen_logs = B_FALSE;
1434 * If it's a mirror or raidz, the subsequent arguments are
1435 * its leaves -- until we encounter the next mirror or raidz.
1437 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1438 nvlist_t **child = NULL;
1439 int c, children = 0;
1441 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1442 if (spares != NULL) {
1443 (void) fprintf(stderr,
1444 gettext("invalid vdev "
1445 "specification: 'spare' can be "
1446 "specified only once\n"));
1452 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1454 (void) fprintf(stderr,
1455 gettext("invalid vdev "
1456 "specification: 'log' can be "
1457 "specified only once\n"));
1465 * A log is not a real grouping device.
1466 * We just set is_log and continue.
1471 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1472 if (l2cache != NULL) {
1473 (void) fprintf(stderr,
1474 gettext("invalid vdev "
1475 "specification: 'cache' can be "
1476 "specified only once\n"));
1483 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1484 (void) fprintf(stderr,
1485 gettext("invalid vdev "
1486 "specification: unsupported 'log' "
1487 "device: %s\n"), type);
1493 for (c = 1; c < argc; c++) {
1494 if (is_grouping(argv[c], NULL, NULL) != NULL)
1497 child = realloc(child,
1498 children * sizeof (nvlist_t *));
1501 if ((nv = make_leaf_vdev(props, argv[c],
1504 child[children - 1] = nv;
1507 if (children < mindev) {
1508 (void) fprintf(stderr, gettext("invalid vdev "
1509 "specification: %s requires at least %d "
1510 "devices\n"), argv[0], mindev);
1514 if (children > maxdev) {
1515 (void) fprintf(stderr, gettext("invalid vdev "
1516 "specification: %s supports no more than "
1517 "%d devices\n"), argv[0], maxdev);
1524 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1528 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1530 nl2cache = children;
1533 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1535 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1537 verify(nvlist_add_uint64(nv,
1538 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1539 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1540 verify(nvlist_add_uint64(nv,
1541 ZPOOL_CONFIG_NPARITY,
1544 verify(nvlist_add_nvlist_array(nv,
1545 ZPOOL_CONFIG_CHILDREN, child,
1548 for (c = 0; c < children; c++)
1549 nvlist_free(child[c]);
1554 * We have a device. Pass off to make_leaf_vdev() to
1555 * construct the appropriate nvlist describing the vdev.
1557 if ((nv = make_leaf_vdev(props, argv[0],
1567 top = realloc(top, toplevels * sizeof (nvlist_t *));
1570 top[toplevels - 1] = nv;
1573 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1574 (void) fprintf(stderr, gettext("invalid vdev "
1575 "specification: at least one toplevel vdev must be "
1580 if (seen_logs && nlogs == 0) {
1581 (void) fprintf(stderr, gettext("invalid vdev specification: "
1582 "log requires at least 1 device\n"));
1587 * Finally, create nvroot and add all top-level vdevs to it.
1589 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1590 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1591 VDEV_TYPE_ROOT) == 0);
1592 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1593 top, toplevels) == 0);
1595 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1596 spares, nspares) == 0);
1598 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1599 l2cache, nl2cache) == 0);
1601 for (t = 0; t < toplevels; t++)
1602 nvlist_free(top[t]);
1603 for (t = 0; t < nspares; t++)
1604 nvlist_free(spares[t]);
1605 for (t = 0; t < nl2cache; t++)
1606 nvlist_free(l2cache[t]);
1617 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1618 splitflags_t flags, int argc, char **argv)
1620 nvlist_t *newroot = NULL, **child;
1624 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1625 (void) fprintf(stderr, gettext("Unable to build a "
1626 "pool from the specified devices\n"));
1630 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1631 nvlist_free(newroot);
1635 /* avoid any tricks in the spec */
1636 verify(nvlist_lookup_nvlist_array(newroot,
1637 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1638 for (c = 0; c < children; c++) {
1643 verify(nvlist_lookup_string(child[c],
1644 ZPOOL_CONFIG_PATH, &path) == 0);
1645 if ((type = is_grouping(path, &min, &max)) != NULL) {
1646 (void) fprintf(stderr, gettext("Cannot use "
1647 "'%s' as a device for splitting\n"), type);
1648 nvlist_free(newroot);
1654 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1655 if (newroot != NULL)
1656 nvlist_free(newroot);
1664 * Get and validate the contents of the given vdev specification. This ensures
1665 * that the nvlist returned is well-formed, that all the devices exist, and that
1666 * they are not currently in use by any other known consumer. The 'poolconfig'
1667 * parameter is the current configuration of the pool when adding devices
1668 * existing pool, and is used to perform additional checks, such as changing the
1669 * replication level of the pool. It can be 'NULL' to indicate that this is a
1670 * new pool. The 'force' flag controls whether devices should be forcefully
1671 * added, even if they appear in use.
1674 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1675 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1678 nvlist_t *poolconfig = NULL;
1682 * Construct the vdev specification. If this is successful, we know
1683 * that we have a valid specification, and that all devices can be
1686 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1689 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1690 nvlist_free(newroot);
1695 * Validate each device to make sure that its not shared with another
1696 * subsystem. We do this even if 'force' is set, because there are some
1697 * uses (such as a dedicated dump device) that even '-f' cannot
1700 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1701 nvlist_free(newroot);
1706 * Check the replication level of the given vdevs and report any errors
1707 * found. We include the existing pool spec, if any, as we need to
1708 * catch changes against the existing replication level.
1710 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1711 nvlist_free(newroot);
1716 * Run through the vdev specification and label any whole disks found.
1718 if (!dryrun && make_disks(zhp, newroot) != 0) {
1719 nvlist_free(newroot);