]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Create.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
30 static int default_layout(struct supertype
*st
, int level
, int verbose
)
34 if (st
&& st
->ss
->default_layout
)
35 layout
= st
->ss
->default_layout(level
);
39 default: /* no layout */
43 layout
= 0x102; /* near=2, far=1 */
46 Name
": layout defaults to n1\n");
50 layout
= map_name(r5layout
, "default");
53 Name
": layout defaults to %s\n", map_num(r5layout
, layout
));
56 layout
= map_name(faultylayout
, "default");
60 Name
": layout defaults to %s\n", map_num(faultylayout
, layout
));
68 int Create(struct supertype
*st
, char *mddev
,
69 int chunk
, int level
, int layout
, unsigned long long size
, int raiddisks
, int sparedisks
,
70 char *name
, char *homehost
, int *uuid
,
71 int subdevs
, mddev_dev_t devlist
,
72 int runstop
, int verbose
, int force
, int assume_clean
,
73 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
, int autof
)
76 * Create a new raid array.
78 * First check that necessary details are available
79 * (i.e. level, raid-disks)
81 * Then check each disk to see what might be on it
82 * and report anything interesting.
84 * If anything looks odd, and runstop not set,
87 * SET_ARRAY_INFO and ADD_NEW_DISK, and
88 * if runstop==run, or raiddisks disks were used,
92 unsigned long long minsize
=0, maxsize
=0;
99 int first_missing
= subdevs
* 2;
100 int second_missing
= subdevs
* 2;
101 int missing_disks
= 0;
102 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
108 int have_container
= 0;
109 int container_fd
= -1;
111 unsigned long long bitmapsize
;
112 struct mdinfo info
, *infos
;
114 int do_default_layout
= 0;
115 unsigned long safe_mode_delay
= 0;
116 char chosen_name
[1024];
117 struct map_ent
*map
= NULL
;
118 unsigned long long newsize
;
120 int major_num
= BITMAP_MAJOR_HI
;
122 memset(&info
, 0, sizeof(info
));
124 if (level
== UnSet
) {
125 /* "ddf" and "imsm" metadata only supports one level - should possibly
126 * push this into metadata handler??
128 if (st
&& (st
->ss
== &super_ddf
|| st
->ss
== &super_imsm
))
129 level
= LEVEL_CONTAINER
;
132 if (level
== UnSet
) {
134 Name
": a RAID level is needed to create an array.\n");
137 if (raiddisks
< 4 && level
== 6) {
139 Name
": at least 4 raid-devices needed for level 6\n");
142 if (raiddisks
> 256 && level
== 6) {
144 Name
": no more than 256 raid-devices supported for level 6\n");
147 if (raiddisks
< 2 && level
>= 4) {
149 Name
": at least 2 raid-devices needed for level 4 or 5\n");
152 if (level
<= 0 && sparedisks
) {
154 Name
": This level does not support spare devices\n");
158 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
159 /* If given a single device, it might be a container, and we can
160 * extract a device list from there
162 mdu_array_info_t inf
;
165 memset(&inf
, 0, sizeof(inf
));
166 fd
= open(devlist
->devname
, O_RDONLY
);
168 ioctl(fd
, GET_ARRAY_INFO
, &inf
) == 0 &&
169 inf
.raid_disks
== 0) {
170 /* yep, looks like a container */
172 rv
= st
->ss
->load_super(st
, fd
,
177 st
= guess_super(fd
);
178 if (st
&& !(rv
= st
->ss
->
185 if (have_container
) {
187 first_missing
= subdevs
* 2;
188 second_missing
= subdevs
* 2;
189 insert_point
= subdevs
* 2;
195 if (st
&& st
->ss
->external
&& sparedisks
) {
197 Name
": This metadata type does not support "
198 "spare disks at create time\n");
201 if (subdevs
> raiddisks
+sparedisks
) {
202 fprintf(stderr
, Name
": You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, raiddisks
+sparedisks
);
205 if (!have_container
&& subdevs
< raiddisks
+sparedisks
) {
206 fprintf(stderr
, Name
": You haven't given enough devices (real or missing) to create this array\n");
209 if (bitmap_file
&& level
<= 0) {
210 fprintf(stderr
, Name
": bitmaps not meaningful with level %s\n",
211 map_num(pers
, level
)?:"given");
215 /* now set some defaults */
218 if (layout
== UnSet
) {
219 do_default_layout
= 1;
220 layout
= default_layout(st
, level
, verbose
);
224 /* check layout fits in array*/
225 if ((layout
&255) * ((layout
>>8)&255) > raiddisks
) {
226 fprintf(stderr
, Name
": that layout requires at least %d devices\n",
227 (layout
&255) * ((layout
>>8)&255));
240 fprintf(stderr
, Name
": chunk size defaults to 512K\n");
244 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
245 if (get_linux_version() < 2006016 && chunk
== 0) {
248 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
253 case LEVEL_MULTIPATH
:
254 case LEVEL_CONTAINER
:
258 fprintf(stderr
, Name
": chunk size ignored for this level\n");
262 fprintf(stderr
, Name
": unknown level %d\n", level
);
267 size
&= ~(unsigned long long)(chunk
- 1);
269 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
270 chunk
, size
*2, NULL
, &newsize
, verbose
>=0))
274 if (size
&& verbose
> 0)
275 fprintf(stderr
, Name
": setting size to %lluK\n",
276 (unsigned long long)size
);
279 /* now look at the subdevs */
280 info
.array
.active_disks
= 0;
281 info
.array
.working_disks
= 0;
283 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
284 char *dname
= dv
->devname
;
285 unsigned long long freesize
;
286 if (strcasecmp(dname
, "missing")==0) {
287 if (first_missing
> dnum
)
288 first_missing
= dnum
;
289 if (second_missing
> dnum
&& dnum
> first_missing
)
290 second_missing
= dnum
;
294 info
.array
.working_disks
++;
295 if (dnum
< raiddisks
)
296 info
.array
.active_disks
++;
298 struct createinfo
*ci
= conf_get_create_info();
303 /* Need to choose a default metadata, which is different
304 * depending on geometry of array.
307 char *name
= "default";
308 for(i
=0; !st
&& superlist
[i
]; i
++) {
309 st
= superlist
[i
]->match_metadata_desc(name
);
310 if (do_default_layout
)
311 layout
= default_layout(st
, level
, verbose
);
312 if (st
&& !st
->ss
->validate_geometry
313 (st
, level
, layout
, raiddisks
,
314 chunk
, size
*2, dname
, &freesize
,
320 fprintf(stderr
, Name
": device %s not suitable "
321 "for any style of array\n",
325 if (st
->ss
!= &super0
||
326 st
->minor_version
!= 90)
329 if (do_default_layout
)
330 layout
= default_layout(st
, level
, verbose
);
331 if (!st
->ss
->validate_geometry(st
, level
, layout
,
333 chunk
, size
*2, dname
,
338 Name
": %s is not suitable for "
346 freesize
/= 2; /* convert to K */
348 /* round to chunk size */
349 freesize
= freesize
& ~(chunk
-1);
352 if (size
&& freesize
< size
) {
353 fprintf(stderr
, Name
": %s is smaller than given size."
354 " %lluK < %lluK + metadata\n",
355 dname
, freesize
, size
);
359 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
363 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
367 if (runstop
!= 1 || verbose
>= 0) {
368 int fd
= open(dname
, O_RDONLY
);
370 fprintf(stderr
, Name
": Cannot open %s: %s\n",
371 dname
, strerror(errno
));
375 warn
|= check_ext2(fd
, dname
);
376 warn
|= check_reiser(fd
, dname
);
377 warn
|= check_raid(fd
, dname
);
378 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
379 st
->minor_version
>= 1)
380 /* metadata at front */
381 warn
|= check_partitions(fd
, dname
, 0);
382 else if (level
== 1 || level
== LEVEL_CONTAINER
)
383 /* partitions could be meaningful */
384 warn
|= check_partitions(fd
, dname
, freesize
*2);
386 /* partitions cannot be meaningful */
387 warn
|= check_partitions(fd
, dname
, 0);
388 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
389 st
->minor_version
>= 1 &&
392 (warn
& 1024) == 0) {
394 fprintf(stderr
, Name
": Note: this array has metadata at the start and\n"
395 " may not be suitable as a boot device. If you plan to\n"
396 " store '/boot' on this device please ensure that\n"
397 " your boot-loader understands md/v1.x metadata, or use\n"
398 " --metadata=0.90\n");
404 info
.array
.working_disks
= raiddisks
;
406 fprintf(stderr
, Name
": create aborted\n");
410 if (mindisc
== NULL
&& !have_container
) {
411 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
414 if (level
> 0 || level
== LEVEL_MULTIPATH
415 || level
== LEVEL_FAULTY
416 || st
->ss
->external
) {
417 /* size is meaningful */
418 if (!st
->ss
->validate_geometry(st
, level
, layout
,
422 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
427 fprintf(stderr
, Name
": size set to %lluK\n", size
);
430 if (!have_container
&& level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
431 if (runstop
!= 1 || verbose
>= 0)
432 fprintf(stderr
, Name
": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
437 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1) != 0) {
438 if (runstop
!= 1 || verbose
>= 0)
439 fprintf(stderr
, Name
": %s unable to enumerate platform support\n"
440 " array may not be compatible with hardware/firmware\n",
447 if (!ask("Continue creating array? ")) {
448 fprintf(stderr
, Name
": create aborted.\n");
453 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
457 /* If this is raid4/5, we want to configure the last active slot
458 * as missing, so that a reconstruct happens (faster than re-parity)
459 * FIX: Can we do this for raid6 as well?
461 if (st
->ss
->external
== 0 &&
462 assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
466 insert_point
= raiddisks
-1;
468 info
.array
.active_disks
--;
475 /* For raid6, if creating with 1 missing drive, make a good drive
476 * into a spare, else the create will fail
478 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
479 st
->ss
->external
== 0 &&
480 second_missing
>= raiddisks
&& level
== 6) {
481 insert_point
= raiddisks
- 1;
482 if (insert_point
== first_missing
)
485 info
.array
.active_disks
--;
489 if (level
<= 0 && first_missing
< subdevs
* 2) {
491 Name
": This level does not support missing devices\n");
495 /* We need to create the device */
497 mdfd
= create_mddev(mddev
, name
, autof
, LOCAL
, chosen_name
);
502 vers
= md_get_version(mdfd
);
504 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
507 mdu_array_info_t inf
;
508 memset(&inf
, 0, sizeof(inf
));
509 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
510 if (inf
.working_disks
!= 0) {
511 fprintf(stderr
, Name
": another array by this name"
512 " is already running.\n");
517 /* Ok, lets try some ioctls */
519 info
.array
.level
= level
;
520 info
.array
.size
= size
;
521 info
.array
.raid_disks
= raiddisks
;
522 /* The kernel should *know* what md_minor we are dealing
523 * with, but it chooses to trust me instead. Sigh
525 info
.array
.md_minor
= 0;
526 if (fstat(mdfd
, &stb
)==0)
527 info
.array
.md_minor
= minor(stb
.st_rdev
);
528 info
.array
.not_persistent
= 0;
530 if ( ( (level
== 4 || level
== 5) &&
531 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
533 ( level
== 6 && (insert_point
< raiddisks
534 || second_missing
< raiddisks
))
540 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
541 info
.resync_start
= MaxSector
;
543 info
.array
.state
= 0; /* not clean, but no errors */
544 info
.resync_start
= 0;
547 /* for raid10, the bitmap size is the capacity of the array,
548 * which is array.size * raid_disks / ncopies;
549 * .. but convert to sectors.
551 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
552 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
553 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
555 bitmapsize
= (unsigned long long)size
* 2;
557 /* There is lots of redundancy in these disk counts,
558 * raid_disks is the most meaningful value
559 * it describes the geometry of the array
561 * nr_disks is total number of used slots.
562 * it should be raid_disks+spare_disks
563 * spare_disks is the number of extra disks present
565 * active_disks is the number of working disks in
566 * active slots. (With raid_disks)
567 * working_disks is the total number of working disks,
569 * failed_disks is the number of disks marked failed
571 * Ideally, the kernel would keep these (except raid_disks)
572 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
573 * So for now, we assume that all raid and spare
574 * devices will be given.
576 info
.array
.spare_disks
=sparedisks
;
577 info
.array
.failed_disks
=missing_disks
;
578 info
.array
.nr_disks
= info
.array
.working_disks
579 + info
.array
.failed_disks
;
580 info
.array
.layout
= layout
;
581 info
.array
.chunk_size
= chunk
*1024;
583 if (name
== NULL
|| *name
== 0) {
584 /* base name on mddev */
589 * /dev/md/home -> home
590 * /dev/mdhome -> home
592 /* FIXME compare this with rules in create_mddev */
593 name
= strrchr(mddev
, '/');
596 if (strncmp(name
, "md_d", 4)==0 &&
599 (name
-mddev
) == 5 /* /dev/ */)
601 else if (strncmp(name
, "md", 2)==0 &&
604 (name
-mddev
) == 5 /* /dev/ */)
608 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
611 total_slots
= info
.array
.nr_disks
;
612 sysfs_init(&info
, mdfd
, 0);
613 st
->ss
->getinfo_super(st
, &info
);
615 if (did_default
&& verbose
>= 0) {
616 if (is_subarray(info
.text_version
)) {
617 int dnum
= devname2devnum(info
.text_version
+1);
619 int mdp
= get_mdp_major();
622 path
= map_dev(MD_MAJOR
, dnum
, 1);
624 path
= map_dev(mdp
, (-1-dnum
)<< 6, 1);
626 mdi
= sysfs_read(-1, dnum
, GET_VERSION
);
628 fprintf(stderr
, Name
": Creating array inside "
630 mdi
?mdi
->text_version
:"managed", path
);
633 fprintf(stderr
, Name
": Defaulting to version"
634 " %s metadata\n", info
.text_version
);
637 map_update(&map
, fd2devnum(mdfd
), info
.text_version
,
638 info
.uuid
, chosen_name
);
641 if (bitmap_file
&& vers
< 9003) {
642 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
644 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
645 " between different architectured. Consider upgrading the Linux kernel.\n");
649 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
650 if ((vers
%100) < 2) {
651 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
654 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
656 bitmapsize
, 1, major_num
)) {
657 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
664 sysfs_init(&info
, mdfd
, 0);
666 if (st
->ss
->external
&& st
->subarray
[0]) {
669 /* When creating a member, we need to be careful
670 * to negotiate with mdmon properly.
671 * If it is already running, we cannot write to
672 * the devices and must ask it to do that part.
673 * If it isn't running, we write to the devices,
675 * We hold an exclusive open on the container
676 * device to make sure mdmon doesn't exit after
677 * we checked that it is running.
679 * For now, fail if it is already running.
681 container_fd
= open_dev_excl(st
->container_dev
);
682 if (container_fd
< 0) {
683 fprintf(stderr
, Name
": Cannot get exclusive "
684 "open on container - weird.\n");
687 if (mdmon_running(st
->container_dev
)) {
689 fprintf(stderr
, Name
": reusing mdmon "
691 devnum2devname(st
->container_dev
));
692 st
->update_tail
= &st
->updates
;
696 rv
= set_array_info(mdfd
, st
, &info
);
698 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
699 mddev
, strerror(errno
));
706 st
->ss
->uuid_from_super(st
, uuid
);
707 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
713 bitmap_fd
= open(bitmap_file
, O_RDWR
);
715 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
719 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
720 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
721 mddev
, strerror(errno
));
726 infos
= malloc(sizeof(*infos
) * total_slots
);
728 for (pass
=1; pass
<=2 ; pass
++) {
729 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
731 for (dnum
=0, dv
= devlist
; dv
;
732 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
735 struct mdinfo
*inf
= &infos
[dnum
];
737 if (dnum
>= total_slots
)
739 if (dnum
== insert_point
) {
743 if (strcasecmp(dv
->devname
, "missing")==0)
747 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
748 /* repeatedly use the container */
755 inf
->disk
.number
= dnum
;
756 inf
->disk
.raid_disk
= dnum
;
757 if (inf
->disk
.raid_disk
< raiddisks
)
758 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
763 if (dv
->writemostly
== 1)
764 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
769 if (st
->ss
->external
&& st
->subarray
[0])
770 fd
= open(dv
->devname
, O_RDWR
);
772 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
775 fprintf(stderr
, Name
": failed to open %s "
776 "after earlier success - aborting\n",
781 inf
->disk
.major
= major(stb
.st_rdev
);
782 inf
->disk
.minor
= minor(stb
.st_rdev
);
785 remove_partitions(fd
);
786 if (st
->ss
->add_to_super(st
, &inf
->disk
,
789 st
->ss
->getinfo_super(st
, inf
);
790 safe_mode_delay
= inf
->safe_mode_delay
;
792 if (have_container
&& verbose
> 0)
793 fprintf(stderr
, Name
": Using %s for device %d\n",
794 map_dev(inf
->disk
.major
,
798 if (!have_container
) {
799 /* getinfo_super might have lost these ... */
800 inf
->disk
.major
= major(stb
.st_rdev
);
801 inf
->disk
.minor
= minor(stb
.st_rdev
);
808 rv
= add_disk(mdfd
, st
, &info
, inf
);
812 Name
": ADD_NEW_DISK for %s "
814 dv
->devname
, strerror(errno
));
815 st
->ss
->free_super(st
);
820 if (!have_container
&&
821 dv
== moved_disk
&& dnum
!= insert_point
) break;
824 struct mdinfo info_new
;
825 struct map_ent
*me
= NULL
;
827 /* check to see if the uuid has changed due to these
828 * metadata changes, and if so update the member array
829 * and container uuid. Note ->write_init_super clears
830 * the subarray cursor such that ->getinfo_super once
831 * again returns container info.
834 st
->ss
->getinfo_super(st
, &info_new
);
835 if (st
->ss
->external
&& level
!= LEVEL_CONTAINER
&&
836 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
837 map_update(&map
, fd2devnum(mdfd
),
838 info_new
.text_version
,
839 info_new
.uuid
, chosen_name
);
840 me
= map_by_devnum(&map
, st
->container_dev
);
843 st
->ss
->write_init_super(st
);
845 /* update parent container uuid */
847 char *path
= strdup(me
->path
);
849 st
->ss
->getinfo_super(st
, &info_new
);
850 map_update(&map
, st
->container_dev
,
851 info_new
.text_version
,
852 info_new
.uuid
, path
);
857 flush_metadata_updates(st
);
861 st
->ss
->free_super(st
);
863 if (level
== LEVEL_CONTAINER
) {
864 /* No need to start. But we should signal udev to
866 sysfs_uevent(&info
, "change");
868 fprintf(stderr
, Name
": container %s prepared.\n", mddev
);
869 wait_for(chosen_name
, mdfd
);
870 } else if (runstop
== 1 || subdevs
>= raiddisks
) {
871 if (st
->ss
->external
) {
874 case LEVEL_MULTIPATH
:
876 sysfs_set_str(&info
, NULL
, "array_state",
881 sysfs_set_str(&info
, NULL
, "array_state",
885 sysfs_set_safemode(&info
, safe_mode_delay
);
887 /* param is not actually used */
889 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
890 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
892 Manage_runstop(mddev
, mdfd
, -1, 0);
897 fprintf(stderr
, Name
": array %s started.\n", mddev
);
898 if (st
->ss
->external
&& st
->subarray
[0]) {
900 start_mdmon(st
->container_dev
);
902 ping_monitor(devnum2devname(st
->container_dev
));
905 wait_for(chosen_name
, mdfd
);
907 fprintf(stderr
, Name
": not starting array - not enough devices.\n");