4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2016 Intel Corporation.
29 * Functions to convert between a list of vdevs and an nvlist representing the
30 * configuration. Each entry in the list can be one of:
33 * disk=(path=..., devid=...)
42 * While the underlying implementation supports it, group vdevs cannot contain
43 * other group vdevs. All userland verification of devices is contained within
44 * this file. If successful, the nvlist returned can be passed directly to the
45 * kernel; we've done as much verification as possible in userland.
47 * Hot spares are a special case, and passed down as an array of disk vdevs, at
48 * the same level as the root of the vdev tree.
50 * The only function exported by this file is 'make_root_vdev'. The
51 * function performs several passes:
53 * 1. Construct the vdev specification. Performs syntax validation and
54 * makes sure each device is valid.
55 * 2. Check for devices in use. Using libblkid to make sure that no
56 * devices are also in use. Some can be overridden using the 'force'
57 * flag, others cannot.
58 * 3. Check for replication errors if the 'force' flag is not specified.
59 * validates that the replication level is consistent across the
61 * 4. Call libzfs to label any whole disks with an EFI label.
70 #include <libnvpair.h>
72 #include <scsi/scsi.h>
77 #include <sys/efi_partition.h>
80 #include <sys/mntent.h>
81 #include <uuid/uuid.h>
82 #include <blkid/blkid.h>
83 #include "zpool_util.h"
84 #include <sys/zfs_context.h>
87 * For any given vdev specification, we can have multiple errors. The
88 * vdev_error() function keeps track of whether we have seen an error yet, and
89 * prints out a header if its the first error we've seen.
94 typedef struct vdev_disk_db_entry
98 } vdev_disk_db_entry_t;
101 * Database of block devices that lie about physical sector sizes. The
102 * identification string must be precisely 24 characters to avoid false
105 static vdev_disk_db_entry_t vdev_disk_database[] = {
106 {"ATA ADATA SSD S396 3", 8192},
107 {"ATA APPLE SSD SM128E", 8192},
108 {"ATA APPLE SSD SM256E", 8192},
109 {"ATA APPLE SSD SM512E", 8192},
110 {"ATA APPLE SSD SM768E", 8192},
111 {"ATA C400-MTFDDAC064M", 8192},
112 {"ATA C400-MTFDDAC128M", 8192},
113 {"ATA C400-MTFDDAC256M", 8192},
114 {"ATA C400-MTFDDAC512M", 8192},
115 {"ATA Corsair Force 3 ", 8192},
116 {"ATA Corsair Force GS", 8192},
117 {"ATA INTEL SSDSA2CT04", 8192},
118 {"ATA INTEL SSDSA2BZ10", 8192},
119 {"ATA INTEL SSDSA2BZ20", 8192},
120 {"ATA INTEL SSDSA2BZ30", 8192},
121 {"ATA INTEL SSDSA2CW04", 8192},
122 {"ATA INTEL SSDSA2CW08", 8192},
123 {"ATA INTEL SSDSA2CW12", 8192},
124 {"ATA INTEL SSDSA2CW16", 8192},
125 {"ATA INTEL SSDSA2CW30", 8192},
126 {"ATA INTEL SSDSA2CW60", 8192},
127 {"ATA INTEL SSDSC2CT06", 8192},
128 {"ATA INTEL SSDSC2CT12", 8192},
129 {"ATA INTEL SSDSC2CT18", 8192},
130 {"ATA INTEL SSDSC2CT24", 8192},
131 {"ATA INTEL SSDSC2CW06", 8192},
132 {"ATA INTEL SSDSC2CW12", 8192},
133 {"ATA INTEL SSDSC2CW18", 8192},
134 {"ATA INTEL SSDSC2CW24", 8192},
135 {"ATA INTEL SSDSC2CW48", 8192},
136 {"ATA KINGSTON SH100S3", 8192},
137 {"ATA KINGSTON SH103S3", 8192},
138 {"ATA M4-CT064M4SSD2 ", 8192},
139 {"ATA M4-CT128M4SSD2 ", 8192},
140 {"ATA M4-CT256M4SSD2 ", 8192},
141 {"ATA M4-CT512M4SSD2 ", 8192},
142 {"ATA OCZ-AGILITY2 ", 8192},
143 {"ATA OCZ-AGILITY3 ", 8192},
144 {"ATA OCZ-VERTEX2 3.5 ", 8192},
145 {"ATA OCZ-VERTEX3 ", 8192},
146 {"ATA OCZ-VERTEX3 LT ", 8192},
147 {"ATA OCZ-VERTEX3 MI ", 8192},
148 {"ATA OCZ-VERTEX4 ", 8192},
149 {"ATA SAMSUNG MZ7WD120", 8192},
150 {"ATA SAMSUNG MZ7WD240", 8192},
151 {"ATA SAMSUNG MZ7WD480", 8192},
152 {"ATA SAMSUNG MZ7WD960", 8192},
153 {"ATA SAMSUNG SSD 830 ", 8192},
154 {"ATA Samsung SSD 840 ", 8192},
155 {"ATA SanDisk SSD U100", 8192},
156 {"ATA TOSHIBA THNSNH06", 8192},
157 {"ATA TOSHIBA THNSNH12", 8192},
158 {"ATA TOSHIBA THNSNH25", 8192},
159 {"ATA TOSHIBA THNSNH51", 8192},
160 {"ATA APPLE SSD TS064C", 4096},
161 {"ATA APPLE SSD TS128C", 4096},
162 {"ATA APPLE SSD TS256C", 4096},
163 {"ATA APPLE SSD TS512C", 4096},
164 {"ATA INTEL SSDSA2M040", 4096},
165 {"ATA INTEL SSDSA2M080", 4096},
166 {"ATA INTEL SSDSA2M160", 4096},
167 {"ATA INTEL SSDSC2MH12", 4096},
168 {"ATA INTEL SSDSC2MH25", 4096},
169 {"ATA OCZ CORE_SSD ", 4096},
170 {"ATA OCZ-VERTEX ", 4096},
171 {"ATA SAMSUNG MCCOE32G", 4096},
172 {"ATA SAMSUNG MCCOE64G", 4096},
173 {"ATA SAMSUNG SSD PM80", 4096},
174 /* Flash drives optimized for 4KB IOs on larger pages */
175 {"ATA INTEL SSDSC2BA10", 4096},
176 {"ATA INTEL SSDSC2BA20", 4096},
177 {"ATA INTEL SSDSC2BA40", 4096},
178 {"ATA INTEL SSDSC2BA80", 4096},
179 {"ATA INTEL SSDSC2BB08", 4096},
180 {"ATA INTEL SSDSC2BB12", 4096},
181 {"ATA INTEL SSDSC2BB16", 4096},
182 {"ATA INTEL SSDSC2BB24", 4096},
183 {"ATA INTEL SSDSC2BB30", 4096},
184 {"ATA INTEL SSDSC2BB40", 4096},
185 {"ATA INTEL SSDSC2BB48", 4096},
186 {"ATA INTEL SSDSC2BB60", 4096},
187 {"ATA INTEL SSDSC2BB80", 4096},
188 {"ATA INTEL SSDSC2BW24", 4096},
189 {"ATA INTEL SSDSC2BP24", 4096},
190 {"ATA INTEL SSDSC2BP48", 4096},
191 {"NA SmrtStorSDLKAE9W", 4096},
192 /* Imported from Open Solaris */
193 {"ATA MARVELL SD88SA02", 4096},
194 /* Advanced format Hard drives */
195 {"ATA Hitachi HDS5C303", 4096},
196 {"ATA SAMSUNG HD204UI ", 4096},
197 {"ATA ST2000DL004 HD20", 4096},
198 {"ATA WDC WD10EARS-00M", 4096},
199 {"ATA WDC WD10EARS-00S", 4096},
200 {"ATA WDC WD10EARS-00Z", 4096},
201 {"ATA WDC WD15EARS-00M", 4096},
202 {"ATA WDC WD15EARS-00S", 4096},
203 {"ATA WDC WD15EARS-00Z", 4096},
204 {"ATA WDC WD20EARS-00M", 4096},
205 {"ATA WDC WD20EARS-00S", 4096},
206 {"ATA WDC WD20EARS-00Z", 4096},
207 {"ATA WDC WD1600BEVT-0", 4096},
208 {"ATA WDC WD2500BEVT-0", 4096},
209 {"ATA WDC WD3200BEVT-0", 4096},
210 {"ATA WDC WD5000BEVT-0", 4096},
211 /* Virtual disks: Assume zvols with default volblocksize */
213 {"ATA QEMU HARDDISK ", 8192},
214 {"IET VIRTUAL-DISK ", 8192},
215 {"OI COMSTAR ", 8192},
216 {"SUN COMSTAR ", 8192},
217 {"NETAPP LUN ", 8192},
221 static const int vdev_disk_database_size =
222 sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
224 #define INQ_REPLY_LEN 96
225 #define INQ_CMD_LEN 6
228 check_sector_size_database(char *path, int *sector_size)
230 unsigned char inq_buff[INQ_REPLY_LEN];
231 unsigned char sense_buffer[32];
232 unsigned char inq_cmd_blk[INQ_CMD_LEN] =
233 {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
239 /* Prepare INQUIRY command */
240 memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
241 io_hdr.interface_id = 'S';
242 io_hdr.cmd_len = sizeof (inq_cmd_blk);
243 io_hdr.mx_sb_len = sizeof (sense_buffer);
244 io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
245 io_hdr.dxfer_len = INQ_REPLY_LEN;
246 io_hdr.dxferp = inq_buff;
247 io_hdr.cmdp = inq_cmd_blk;
248 io_hdr.sbp = sense_buffer;
249 io_hdr.timeout = 10; /* 10 milliseconds is ample time */
251 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
254 error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
261 if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
264 for (i = 0; i < vdev_disk_database_size; i++) {
265 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
268 *sector_size = vdev_disk_database[i].sector_size;
277 vdev_error(const char *fmt, ...)
282 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
284 (void) fprintf(stderr, gettext("use '-f' to override "
285 "the following errors:\n"));
287 (void) fprintf(stderr, gettext("the following errors "
288 "must be manually repaired:\n"));
293 (void) vfprintf(stderr, fmt, ap);
298 * Check that a file is valid. All we can do in this case is check that it's
299 * not in use by another pool, and not in use by swap.
302 check_file(const char *file, boolean_t force, boolean_t isspare)
310 if ((fd = open(file, O_RDONLY)) < 0)
313 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
317 case POOL_STATE_ACTIVE:
318 desc = gettext("active");
321 case POOL_STATE_EXPORTED:
322 desc = gettext("exported");
325 case POOL_STATE_POTENTIALLY_ACTIVE:
326 desc = gettext("potentially active");
330 desc = gettext("unknown");
335 * Allow hot spares to be shared between pools.
337 if (state == POOL_STATE_SPARE && isspare)
340 if (state == POOL_STATE_ACTIVE ||
341 state == POOL_STATE_SPARE || !force) {
343 case POOL_STATE_SPARE:
344 vdev_error(gettext("%s is reserved as a hot "
345 "spare for pool %s\n"), file, name);
348 vdev_error(gettext("%s is part of %s pool "
349 "'%s'\n"), file, desc, name);
365 (void) fprintf(stderr, gettext("warning: device in use checking "
366 "failed: %s\n"), strerror(err));
370 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
375 /* No valid type detected device is safe to use */
376 value = blkid_get_tag_value(cache, "TYPE", path);
381 * If libblkid detects a ZFS device, we check the device
382 * using check_file() to see if it's safe. The one safe
383 * case is a spare device shared between multiple pools.
385 if (strcmp(value, "zfs_member") == 0) {
386 err = check_file(path, force, isspare);
392 vdev_error(gettext("%s contains a filesystem of "
393 "type '%s'\n"), path, value);
403 * Validate that a disk including all partitions are safe to use.
405 * For EFI labeled disks this can done relatively easily with the libefi
406 * library. The partition numbers are extracted from the label and used
407 * to generate the expected /dev/ paths. Each partition can then be
408 * checked for conflicts.
410 * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
411 * but due to the lack of a readily available libraries this scanning is
412 * not implemented. Instead only the device path as given is checked.
415 check_disk(const char *path, blkid_cache cache, int force,
416 boolean_t isspare, boolean_t iswholedisk)
419 char slice_path[MAXPATHLEN];
424 return (check_slice(path, cache, force, isspare));
426 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
432 * Expected to fail for non-EFI labled disks. Just check the device
433 * as given and do not attempt to detect and scan partitions.
435 err = efi_alloc_and_read(fd, &vtoc);
438 return (check_slice(path, cache, force, isspare));
442 * The primary efi partition label is damaged however the secondary
443 * label at the end of the device is intact. Rather than use this
444 * label we should play it safe and treat this as a non efi device.
446 if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
451 /* Partitions will now be created using the backup */
454 vdev_error(gettext("%s contains a corrupt primary "
455 "EFI label.\n"), path);
460 for (i = 0; i < vtoc->efi_nparts; i++) {
462 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
463 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
466 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
467 (void) snprintf(slice_path, sizeof (slice_path),
468 "%s%s%d", path, "-part", i+1);
470 (void) snprintf(slice_path, sizeof (slice_path),
471 "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
474 err = check_slice(slice_path, cache, force, isspare);
486 check_device(const char *path, boolean_t force,
487 boolean_t isspare, boolean_t iswholedisk)
489 static blkid_cache cache = NULL;
492 * There is no easy way to add a correct blkid_put_cache() call,
493 * memory will be reclaimed when the command exits.
498 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
503 if ((err = blkid_probe_all(cache)) != 0) {
504 blkid_put_cache(cache);
510 return (check_disk(path, cache, force, isspare, iswholedisk));
514 * By "whole disk" we mean an entire physical disk (something we can
515 * label, toggle the write cache on, etc.) as opposed to the full
516 * capacity of a pseudo-device such as lofi or did. We act as if we
517 * are labeling the disk, which should be a pretty good test of whether
518 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
522 is_whole_disk(const char *path)
524 struct dk_gpt *label;
527 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
529 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
539 * This may be a shorthand device path or it could be total gibberish.
540 * Check to see if it is a known device available in zfs_vdev_paths.
541 * As part of this check, see if we've been given an entire disk
542 * (minus the slice number).
545 is_shorthand_path(const char *arg, char *path,
546 struct stat64 *statbuf, boolean_t *wholedisk)
550 error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
552 *wholedisk = is_whole_disk(path);
553 if (*wholedisk || (stat64(path, statbuf) == 0))
557 strlcpy(path, arg, sizeof (path));
558 memset(statbuf, 0, sizeof (*statbuf));
559 *wholedisk = B_FALSE;
565 * Determine if the given path is a hot spare within the given configuration.
566 * If no configuration is given we rely solely on the label.
569 is_spare(nvlist_t *config, const char *path)
575 uint64_t guid, spareguid;
581 if ((fd = open(path, O_RDONLY)) < 0)
584 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
586 state != POOL_STATE_SPARE ||
587 zpool_read_label(fd, &label, NULL) != 0) {
598 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
601 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
603 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
604 &spares, &nspares) == 0) {
605 for (i = 0; i < nspares; i++) {
606 verify(nvlist_lookup_uint64(spares[i],
607 ZPOOL_CONFIG_GUID, &spareguid) == 0);
608 if (spareguid == guid)
617 * Create a leaf vdev. Determine if this is a file or a device. If it's a
618 * device, fill in the device id to make a complete nvlist. Valid forms for a
621 * /dev/xxx Complete disk path
622 * /xxx Full path to file
623 * xxx Shorthand for <zfs_vdev_paths>/xxx
626 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
628 char path[MAXPATHLEN];
629 struct stat64 statbuf;
630 nvlist_t *vdev = NULL;
632 boolean_t wholedisk = B_FALSE;
637 * Determine what type of vdev this is, and put the full path into
638 * 'path'. We detect whether this is a device of file afterwards by
639 * checking the st_mode of the file.
643 * Complete device or file path. Exact type is determined by
644 * examining the file descriptor afterwards. Symbolic links
645 * are resolved to their real paths for the is_whole_disk()
646 * and S_ISBLK/S_ISREG type checks. However, we are careful
647 * to store the given path as ZPOOL_CONFIG_PATH to ensure we
648 * can leverage udev's persistent device labels.
650 if (realpath(arg, path) == NULL) {
651 (void) fprintf(stderr,
652 gettext("cannot resolve path '%s'\n"), arg);
656 wholedisk = is_whole_disk(path);
657 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
658 (void) fprintf(stderr,
659 gettext("cannot open '%s': %s\n"),
660 path, strerror(errno));
664 /* After is_whole_disk() check restore original passed path */
665 strlcpy(path, arg, MAXPATHLEN);
667 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
670 * If we got ENOENT, then the user gave us
671 * gibberish, so try to direct them with a
672 * reasonable error message. Otherwise,
673 * regurgitate strerror() since it's the best we
677 (void) fprintf(stderr,
678 gettext("cannot open '%s': no such "
679 "device in %s\n"), arg, DISK_ROOT);
680 (void) fprintf(stderr,
681 gettext("must be a full path or "
682 "shorthand device name\n"));
685 (void) fprintf(stderr,
686 gettext("cannot open '%s': %s\n"),
687 path, strerror(errno));
694 * Determine whether this is a device or a file.
696 if (wholedisk || S_ISBLK(statbuf.st_mode)) {
697 type = VDEV_TYPE_DISK;
698 } else if (S_ISREG(statbuf.st_mode)) {
699 type = VDEV_TYPE_FILE;
701 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
702 "block device or regular file\n"), path);
707 * Finally, we have the complete device or file, and we know that it is
708 * acceptable to use. Construct the nvlist to describe this vdev. All
709 * vdevs have a 'path' element, and devices also have a 'devid' element.
711 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
712 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
713 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
714 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
715 if (strcmp(type, VDEV_TYPE_DISK) == 0)
716 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
717 (uint64_t)wholedisk) == 0);
720 * Override defaults if custom properties are provided.
725 if (nvlist_lookup_string(props,
726 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
727 zfs_nicestrtonum(NULL, value, &ashift);
731 * If the device is known to incorrectly report its physical sector
732 * size explicitly provide the known correct value.
737 if (check_sector_size_database(path, §or_size) == B_TRUE)
738 ashift = highbit64(sector_size) - 1;
742 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
748 * Go through and verify the replication level of the pool is consistent.
749 * Performs the following checks:
751 * For the new spec, verifies that devices in mirrors and raidz are the
754 * If the current configuration already has inconsistent replication
755 * levels, ignore any other potential problems in the new spec.
757 * Otherwise, make sure that the current spec (if there is one) and the new
758 * spec have consistent replication levels.
760 typedef struct replication_level {
762 uint64_t zprl_children;
763 uint64_t zprl_parity;
764 } replication_level_t;
766 #define ZPOOL_FUZZ (16 * 1024 * 1024)
769 * Given a list of toplevel vdevs, return the current replication level. If
770 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
771 * an error message will be displayed for each self-inconsistent vdev.
773 static replication_level_t *
774 get_replication(nvlist_t *nvroot, boolean_t fatal)
782 replication_level_t lastrep = { 0 }, rep, *ret;
783 boolean_t dontreport;
785 ret = safe_malloc(sizeof (replication_level_t));
787 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
788 &top, &toplevels) == 0);
790 lastrep.zprl_type = NULL;
791 for (t = 0; t < toplevels; t++) {
792 uint64_t is_log = B_FALSE;
797 * For separate logs we ignore the top level vdev replication
800 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
804 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
806 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
807 &child, &children) != 0) {
809 * This is a 'file' or 'disk' vdev.
811 rep.zprl_type = type;
812 rep.zprl_children = 1;
818 * This is a mirror or RAID-Z vdev. Go through and make
819 * sure the contents are all the same (files vs. disks),
820 * keeping track of the number of elements in the
823 * We also check that the size of each vdev (if it can
824 * be determined) is the same.
826 rep.zprl_type = type;
827 rep.zprl_children = 0;
829 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
830 verify(nvlist_lookup_uint64(nv,
831 ZPOOL_CONFIG_NPARITY,
832 &rep.zprl_parity) == 0);
833 assert(rep.zprl_parity != 0);
839 * The 'dontreport' variable indicates that we've
840 * already reported an error for this spec, so don't
841 * bother doing it again.
846 for (c = 0; c < children; c++) {
847 nvlist_t *cnv = child[c];
849 struct stat64 statbuf;
850 uint64_t size = -1ULL;
856 verify(nvlist_lookup_string(cnv,
857 ZPOOL_CONFIG_TYPE, &childtype) == 0);
860 * If this is a replacing or spare vdev, then
861 * get the real first child of the vdev.
863 if (strcmp(childtype,
864 VDEV_TYPE_REPLACING) == 0 ||
865 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
869 verify(nvlist_lookup_nvlist_array(cnv,
870 ZPOOL_CONFIG_CHILDREN, &rchild,
872 assert(rchildren == 2);
875 verify(nvlist_lookup_string(cnv,
880 verify(nvlist_lookup_string(cnv,
881 ZPOOL_CONFIG_PATH, &path) == 0);
884 * If we have a raidz/mirror that combines disks
885 * with files, report it as an error.
887 if (!dontreport && type != NULL &&
888 strcmp(type, childtype) != 0) {
894 "mismatched replication "
895 "level: %s contains both "
896 "files and devices\n"),
904 * According to stat(2), the value of 'st_size'
905 * is undefined for block devices and character
906 * devices. But there is no effective way to
907 * determine the real size in userland.
909 * Instead, we'll take advantage of an
910 * implementation detail of spec_size(). If the
911 * device is currently open, then we (should)
912 * return a valid size.
914 * If we still don't get a valid size (indicated
915 * by a size of 0 or MAXOFFSET_T), then ignore
916 * this device altogether.
918 if ((fd = open(path, O_RDONLY)) >= 0) {
919 err = fstat64(fd, &statbuf);
922 err = stat64(path, &statbuf);
926 statbuf.st_size == 0 ||
927 statbuf.st_size == MAXOFFSET_T)
930 size = statbuf.st_size;
933 * Also make sure that devices and
934 * slices have a consistent size. If
935 * they differ by a significant amount
936 * (~16MB) then report an error.
939 (vdev_size != -1ULL &&
940 (labs(size - vdev_size) >
947 "%s contains devices of "
948 "different sizes\n"),
961 * At this point, we have the replication of the last toplevel
962 * vdev in 'rep'. Compare it to 'lastrep' to see if its
965 if (lastrep.zprl_type != NULL) {
966 if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
972 "mismatched replication level: "
973 "both %s and %s vdevs are "
975 lastrep.zprl_type, rep.zprl_type);
978 } else if (lastrep.zprl_parity != rep.zprl_parity) {
984 "mismatched replication level: "
985 "both %llu and %llu device parity "
986 "%s vdevs are present\n"),
992 } else if (lastrep.zprl_children != rep.zprl_children) {
998 "mismatched replication level: "
999 "both %llu-way and %llu-way %s "
1000 "vdevs are present\n"),
1001 lastrep.zprl_children,
1018 * Check the replication level of the vdev spec against the current pool. Calls
1019 * get_replication() to make sure the new spec is self-consistent. If the pool
1020 * has a consistent replication level, then we ignore any errors. Otherwise,
1021 * report any difference between the two.
1024 check_replication(nvlist_t *config, nvlist_t *newroot)
1028 replication_level_t *current = NULL, *new;
1032 * If we have a current pool configuration, check to see if it's
1033 * self-consistent. If not, simply return success.
1035 if (config != NULL) {
1038 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1040 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1044 * for spares there may be no children, and therefore no
1045 * replication level to check
1047 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1048 &child, &children) != 0) || (children == 0)) {
1054 * If all we have is logs then there's no replication level to check.
1056 if (num_logs(newroot) == children) {
1062 * Get the replication level of the new vdev spec, reporting any
1063 * inconsistencies found.
1065 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1071 * Check to see if the new vdev spec matches the replication level of
1075 if (current != NULL) {
1076 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1078 "mismatched replication level: pool uses %s "
1079 "and new vdev is %s\n"),
1080 current->zprl_type, new->zprl_type);
1082 } else if (current->zprl_parity != new->zprl_parity) {
1084 "mismatched replication level: pool uses %llu "
1085 "device parity and new vdev uses %llu\n"),
1086 current->zprl_parity, new->zprl_parity);
1088 } else if (current->zprl_children != new->zprl_children) {
1090 "mismatched replication level: pool uses %llu-way "
1091 "%s and new vdev uses %llu-way %s\n"),
1092 current->zprl_children, current->zprl_type,
1093 new->zprl_children, new->zprl_type);
1099 if (current != NULL)
1106 zero_label(char *path)
1108 const int size = 4096;
1112 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1113 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1114 path, strerror(errno));
1118 memset(buf, 0, size);
1119 err = write(fd, buf, size);
1120 (void) fdatasync(fd);
1124 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1125 "of '%s': %s\n"), size, path, strerror(errno));
1130 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1131 "of '%s'\n"), err, size, path);
1139 * Go through and find any whole disks in the vdev specification, labelling them
1140 * as appropriate. When constructing the vdev spec, we were unable to open this
1141 * device in order to provide a devid. Now that we have labelled the disk and
1142 * know that slice 0 is valid, we can construct the devid now.
1144 * If the disk was already labeled with an EFI label, we will have gotten the
1145 * devid already (because we were able to open the whole disk). Otherwise, we
1146 * need to get the devid after we label the disk.
1149 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1154 char devpath[MAXPATHLEN];
1155 char udevpath[MAXPATHLEN];
1157 struct stat64 statbuf;
1158 int is_exclusive = 0;
1162 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1164 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1165 &child, &children) != 0) {
1167 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1171 * We have a disk device. If this is a whole disk write
1172 * out the efi partition table, otherwise write zero's to
1173 * the first 4k of the partition. This is to ensure that
1174 * libblkid will not misidentify the partition due to a
1175 * magic value left by the previous filesystem.
1177 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1178 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1183 * Update device id string for mpath nodes (Linux only)
1185 if (is_mpath_whole_disk(path))
1186 update_vdev_config_dev_strs(nv);
1188 (void) zero_label(path);
1192 if (realpath(path, devpath) == NULL) {
1194 (void) fprintf(stderr,
1195 gettext("cannot resolve path '%s'\n"), path);
1200 * Remove any previously existing symlink from a udev path to
1201 * the device before labeling the disk. This ensures that
1202 * only newly created links are used. Otherwise there is a
1203 * window between when udev deletes and recreates the link
1204 * during which access attempts will fail with ENOENT.
1206 strncpy(udevpath, path, MAXPATHLEN);
1207 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1209 fd = open(devpath, O_RDWR|O_EXCL);
1218 * If the partition exists, contains a valid spare label,
1219 * and is opened exclusively there is no need to partition
1220 * it. Hot spares have already been partitioned and are
1221 * held open exclusively by the kernel as a safety measure.
1223 * If the provided path is for a /dev/disk/ device its
1224 * symbolic link will be removed, partition table created,
1225 * and then block until udev creates the new link.
1227 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1228 char *devnode = strrchr(devpath, '/') + 1;
1230 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1232 ret = lstat64(udevpath, &statbuf);
1233 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1234 (void) unlink(udevpath);
1238 * When labeling a pool the raw device node name
1239 * is provided as it appears under /dev/.
1241 if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1245 * Wait for udev to signal the device is available
1246 * by the provided path.
1248 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1250 (void) fprintf(stderr,
1251 gettext("missing link: %s was "
1252 "partitioned but %s is missing\n"),
1257 ret = zero_label(udevpath);
1263 * Update the path to refer to the partition. The presence of
1264 * the 'whole_disk' field indicates to the CLI that we should
1265 * chop off the partition number when displaying the device in
1268 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1271 * Update device id strings for whole disks (Linux only)
1273 update_vdev_config_dev_strs(nv);
1278 for (c = 0; c < children; c++)
1279 if ((ret = make_disks(zhp, child[c])) != 0)
1282 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1283 &child, &children) == 0)
1284 for (c = 0; c < children; c++)
1285 if ((ret = make_disks(zhp, child[c])) != 0)
1288 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1289 &child, &children) == 0)
1290 for (c = 0; c < children; c++)
1291 if ((ret = make_disks(zhp, child[c])) != 0)
1298 * Go through and find any devices that are in use. We rely on libdiskmgt for
1299 * the majority of this task.
1302 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1303 boolean_t replacing, boolean_t isspare)
1309 char buf[MAXPATHLEN];
1310 uint64_t wholedisk = B_FALSE;
1311 boolean_t anyinuse = B_FALSE;
1313 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1315 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1316 &child, &children) != 0) {
1318 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1319 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1320 verify(!nvlist_lookup_uint64(nv,
1321 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1324 * As a generic check, we look to see if this is a replace of a
1325 * hot spare within the same pool. If so, we allow it
1326 * regardless of what libblkid or zpool_in_use() says.
1329 (void) strlcpy(buf, path, sizeof (buf));
1331 ret = zfs_append_partition(buf, sizeof (buf));
1336 if (is_spare(config, buf))
1340 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1341 ret = check_device(path, force, isspare, wholedisk);
1343 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1344 ret = check_file(path, force, isspare);
1349 for (c = 0; c < children; c++)
1350 if (is_device_in_use(config, child[c], force, replacing,
1354 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1355 &child, &children) == 0)
1356 for (c = 0; c < children; c++)
1357 if (is_device_in_use(config, child[c], force, replacing,
1361 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1362 &child, &children) == 0)
1363 for (c = 0; c < children; c++)
1364 if (is_device_in_use(config, child[c], force, replacing,
1372 is_grouping(const char *type, int *mindev, int *maxdev)
1374 if (strncmp(type, "raidz", 5) == 0) {
1375 const char *p = type + 5;
1381 } else if (*p == '0') {
1382 return (NULL); /* no zero prefixes allowed */
1385 nparity = strtol(p, &end, 10);
1386 if (errno != 0 || nparity < 1 || nparity >= 255 ||
1392 *mindev = nparity + 1;
1395 return (VDEV_TYPE_RAIDZ);
1401 if (strcmp(type, "mirror") == 0) {
1404 return (VDEV_TYPE_MIRROR);
1407 if (strcmp(type, "spare") == 0) {
1410 return (VDEV_TYPE_SPARE);
1413 if (strcmp(type, "log") == 0) {
1416 return (VDEV_TYPE_LOG);
1419 if (strcmp(type, "cache") == 0) {
1422 return (VDEV_TYPE_L2CACHE);
1429 * Construct a syntactically valid vdev specification,
1430 * and ensure that all devices and files exist and can be opened.
1431 * Note: we don't bother freeing anything in the error paths
1432 * because the program is just going to exit anyway.
1435 construct_spec(nvlist_t *props, int argc, char **argv)
1437 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1438 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1441 boolean_t seen_logs;
1451 seen_logs = B_FALSE;
1457 * If it's a mirror or raidz, the subsequent arguments are
1458 * its leaves -- until we encounter the next mirror or raidz.
1460 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1461 nvlist_t **child = NULL;
1462 int c, children = 0;
1464 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1465 if (spares != NULL) {
1466 (void) fprintf(stderr,
1467 gettext("invalid vdev "
1468 "specification: 'spare' can be "
1469 "specified only once\n"));
1475 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1477 (void) fprintf(stderr,
1478 gettext("invalid vdev "
1479 "specification: 'log' can be "
1480 "specified only once\n"));
1488 * A log is not a real grouping device.
1489 * We just set is_log and continue.
1494 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1495 if (l2cache != NULL) {
1496 (void) fprintf(stderr,
1497 gettext("invalid vdev "
1498 "specification: 'cache' can be "
1499 "specified only once\n"));
1506 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1507 (void) fprintf(stderr,
1508 gettext("invalid vdev "
1509 "specification: unsupported 'log' "
1510 "device: %s\n"), type);
1516 for (c = 1; c < argc; c++) {
1517 if (is_grouping(argv[c], NULL, NULL) != NULL)
1520 child = realloc(child,
1521 children * sizeof (nvlist_t *));
1524 if ((nv = make_leaf_vdev(props, argv[c],
1527 child[children - 1] = nv;
1530 if (children < mindev) {
1531 (void) fprintf(stderr, gettext("invalid vdev "
1532 "specification: %s requires at least %d "
1533 "devices\n"), argv[0], mindev);
1537 if (children > maxdev) {
1538 (void) fprintf(stderr, gettext("invalid vdev "
1539 "specification: %s supports no more than "
1540 "%d devices\n"), argv[0], maxdev);
1547 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1551 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1553 nl2cache = children;
1556 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1558 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1560 verify(nvlist_add_uint64(nv,
1561 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1562 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1563 verify(nvlist_add_uint64(nv,
1564 ZPOOL_CONFIG_NPARITY,
1567 verify(nvlist_add_nvlist_array(nv,
1568 ZPOOL_CONFIG_CHILDREN, child,
1571 for (c = 0; c < children; c++)
1572 nvlist_free(child[c]);
1577 * We have a device. Pass off to make_leaf_vdev() to
1578 * construct the appropriate nvlist describing the vdev.
1580 if ((nv = make_leaf_vdev(props, argv[0],
1590 top = realloc(top, toplevels * sizeof (nvlist_t *));
1593 top[toplevels - 1] = nv;
1596 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1597 (void) fprintf(stderr, gettext("invalid vdev "
1598 "specification: at least one toplevel vdev must be "
1603 if (seen_logs && nlogs == 0) {
1604 (void) fprintf(stderr, gettext("invalid vdev specification: "
1605 "log requires at least 1 device\n"));
1610 * Finally, create nvroot and add all top-level vdevs to it.
1612 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1613 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1614 VDEV_TYPE_ROOT) == 0);
1615 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1616 top, toplevels) == 0);
1618 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1619 spares, nspares) == 0);
1621 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1622 l2cache, nl2cache) == 0);
1624 for (t = 0; t < toplevels; t++)
1625 nvlist_free(top[t]);
1626 for (t = 0; t < nspares; t++)
1627 nvlist_free(spares[t]);
1628 for (t = 0; t < nl2cache; t++)
1629 nvlist_free(l2cache[t]);
1640 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1641 splitflags_t flags, int argc, char **argv)
1643 nvlist_t *newroot = NULL, **child;
1647 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1648 (void) fprintf(stderr, gettext("Unable to build a "
1649 "pool from the specified devices\n"));
1653 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1654 nvlist_free(newroot);
1658 /* avoid any tricks in the spec */
1659 verify(nvlist_lookup_nvlist_array(newroot,
1660 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1661 for (c = 0; c < children; c++) {
1666 verify(nvlist_lookup_string(child[c],
1667 ZPOOL_CONFIG_PATH, &path) == 0);
1668 if ((type = is_grouping(path, &min, &max)) != NULL) {
1669 (void) fprintf(stderr, gettext("Cannot use "
1670 "'%s' as a device for splitting\n"), type);
1671 nvlist_free(newroot);
1677 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1678 nvlist_free(newroot);
1686 * Get and validate the contents of the given vdev specification. This ensures
1687 * that the nvlist returned is well-formed, that all the devices exist, and that
1688 * they are not currently in use by any other known consumer. The 'poolconfig'
1689 * parameter is the current configuration of the pool when adding devices
1690 * existing pool, and is used to perform additional checks, such as changing the
1691 * replication level of the pool. It can be 'NULL' to indicate that this is a
1692 * new pool. The 'force' flag controls whether devices should be forcefully
1693 * added, even if they appear in use.
1696 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1697 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1700 nvlist_t *poolconfig = NULL;
1704 * Construct the vdev specification. If this is successful, we know
1705 * that we have a valid specification, and that all devices can be
1708 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1711 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1712 nvlist_free(newroot);
1717 * Validate each device to make sure that its not shared with another
1718 * subsystem. We do this even if 'force' is set, because there are some
1719 * uses (such as a dedicated dump device) that even '-f' cannot
1722 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1723 nvlist_free(newroot);
1728 * Check the replication level of the given vdevs and report any errors
1729 * found. We include the existing pool spec, if any, as we need to
1730 * catch changes against the existing replication level.
1732 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1733 nvlist_free(newroot);
1738 * Run through the vdev specification and label any whole disks found.
1740 if (!dryrun && make_disks(zhp, newroot) != 0) {
1741 nvlist_free(newroot);