]>
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 n2\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));
238 if (st
&& st
->ss
->default_chunk
)
239 chunk
= st
->ss
->default_chunk(st
);
241 chunk
= chunk
? : 512;
244 fprintf(stderr
, Name
": chunk size defaults to %dK\n", chunk
);
248 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
249 if (get_linux_version() < 2006016 && chunk
== 0) {
252 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
257 case LEVEL_MULTIPATH
:
258 case LEVEL_CONTAINER
:
262 fprintf(stderr
, Name
": chunk size ignored for this level\n");
266 fprintf(stderr
, Name
": unknown level %d\n", level
);
271 size
&= ~(unsigned long long)(chunk
- 1);
273 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
274 chunk
, size
*2, NULL
, &newsize
, verbose
>=0))
278 if (size
&& verbose
> 0)
279 fprintf(stderr
, Name
": setting size to %lluK\n",
280 (unsigned long long)size
);
283 /* now look at the subdevs */
284 info
.array
.active_disks
= 0;
285 info
.array
.working_disks
= 0;
287 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
288 char *dname
= dv
->devname
;
289 unsigned long long freesize
;
290 if (strcasecmp(dname
, "missing")==0) {
291 if (first_missing
> dnum
)
292 first_missing
= dnum
;
293 if (second_missing
> dnum
&& dnum
> first_missing
)
294 second_missing
= dnum
;
298 info
.array
.working_disks
++;
299 if (dnum
< raiddisks
)
300 info
.array
.active_disks
++;
302 struct createinfo
*ci
= conf_get_create_info();
307 /* Need to choose a default metadata, which is different
308 * depending on geometry of array.
311 char *name
= "default";
312 for(i
=0; !st
&& superlist
[i
]; i
++) {
313 st
= superlist
[i
]->match_metadata_desc(name
);
314 if (do_default_layout
)
315 layout
= default_layout(st
, level
, verbose
);
316 if (st
&& !st
->ss
->validate_geometry
317 (st
, level
, layout
, raiddisks
,
318 chunk
, size
*2, dname
, &freesize
,
324 fprintf(stderr
, Name
": device %s not suitable "
325 "for any style of array\n",
329 if (st
->ss
!= &super0
||
330 st
->minor_version
!= 90)
333 if (do_default_layout
)
334 layout
= default_layout(st
, level
, verbose
);
335 if (!st
->ss
->validate_geometry(st
, level
, layout
,
337 chunk
, size
*2, dname
,
342 Name
": %s is not suitable for "
350 freesize
/= 2; /* convert to K */
352 /* round to chunk size */
353 freesize
= freesize
& ~(chunk
-1);
356 if (size
&& freesize
< size
) {
357 fprintf(stderr
, Name
": %s is smaller than given size."
358 " %lluK < %lluK + metadata\n",
359 dname
, freesize
, size
);
363 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
367 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
371 if (runstop
!= 1 || verbose
>= 0) {
372 int fd
= open(dname
, O_RDONLY
);
374 fprintf(stderr
, Name
": Cannot open %s: %s\n",
375 dname
, strerror(errno
));
379 warn
|= check_ext2(fd
, dname
);
380 warn
|= check_reiser(fd
, dname
);
381 warn
|= check_raid(fd
, dname
);
382 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
383 st
->minor_version
>= 1)
384 /* metadata at front */
385 warn
|= check_partitions(fd
, dname
, 0);
386 else if (level
== 1 || level
== LEVEL_CONTAINER
)
387 /* partitions could be meaningful */
388 warn
|= check_partitions(fd
, dname
, freesize
*2);
390 /* partitions cannot be meaningful */
391 warn
|= check_partitions(fd
, dname
, 0);
392 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
393 st
->minor_version
>= 1 &&
396 (warn
& 1024) == 0) {
398 fprintf(stderr
, Name
": Note: this array has metadata at the start and\n"
399 " may not be suitable as a boot device. If you plan to\n"
400 " store '/boot' on this device please ensure that\n"
401 " your boot-loader understands md/v1.x metadata, or use\n"
402 " --metadata=0.90\n");
408 info
.array
.working_disks
= raiddisks
;
410 fprintf(stderr
, Name
": create aborted\n");
414 if (mindisc
== NULL
&& !have_container
) {
415 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
418 if (level
> 0 || level
== LEVEL_MULTIPATH
419 || level
== LEVEL_FAULTY
420 || st
->ss
->external
) {
421 /* size is meaningful */
422 if (!st
->ss
->validate_geometry(st
, level
, layout
,
426 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
431 fprintf(stderr
, Name
": size set to %lluK\n", size
);
434 if (!have_container
&& level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
435 if (runstop
!= 1 || verbose
>= 0)
436 fprintf(stderr
, Name
": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
441 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1) != 0) {
442 if (runstop
!= 1 || verbose
>= 0)
443 fprintf(stderr
, Name
": %s unable to enumerate platform support\n"
444 " array may not be compatible with hardware/firmware\n",
451 if (!ask("Continue creating array? ")) {
452 fprintf(stderr
, Name
": create aborted.\n");
457 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
461 /* If this is raid4/5, we want to configure the last active slot
462 * as missing, so that a reconstruct happens (faster than re-parity)
463 * FIX: Can we do this for raid6 as well?
465 if (st
->ss
->external
== 0 &&
466 assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
470 insert_point
= raiddisks
-1;
472 info
.array
.active_disks
--;
479 /* For raid6, if creating with 1 missing drive, make a good drive
480 * into a spare, else the create will fail
482 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
483 st
->ss
->external
== 0 &&
484 second_missing
>= raiddisks
&& level
== 6) {
485 insert_point
= raiddisks
- 1;
486 if (insert_point
== first_missing
)
489 info
.array
.active_disks
--;
493 if (level
<= 0 && first_missing
< subdevs
* 2) {
495 Name
": This level does not support missing devices\n");
499 /* We need to create the device */
501 mdfd
= create_mddev(mddev
, name
, autof
, LOCAL
, chosen_name
);
506 vers
= md_get_version(mdfd
);
508 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
511 mdu_array_info_t inf
;
512 memset(&inf
, 0, sizeof(inf
));
513 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
514 if (inf
.working_disks
!= 0) {
515 fprintf(stderr
, Name
": another array by this name"
516 " is already running.\n");
521 /* Ok, lets try some ioctls */
523 info
.array
.level
= level
;
524 info
.array
.size
= size
;
525 info
.array
.raid_disks
= raiddisks
;
526 /* The kernel should *know* what md_minor we are dealing
527 * with, but it chooses to trust me instead. Sigh
529 info
.array
.md_minor
= 0;
530 if (fstat(mdfd
, &stb
)==0)
531 info
.array
.md_minor
= minor(stb
.st_rdev
);
532 info
.array
.not_persistent
= 0;
534 if ( ( (level
== 4 || level
== 5) &&
535 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
537 ( level
== 6 && (insert_point
< raiddisks
538 || second_missing
< raiddisks
))
544 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
545 info
.resync_start
= MaxSector
;
547 info
.array
.state
= 0; /* not clean, but no errors */
548 info
.resync_start
= 0;
551 /* for raid10, the bitmap size is the capacity of the array,
552 * which is array.size * raid_disks / ncopies;
553 * .. but convert to sectors.
555 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
556 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
557 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
559 bitmapsize
= (unsigned long long)size
* 2;
561 /* There is lots of redundancy in these disk counts,
562 * raid_disks is the most meaningful value
563 * it describes the geometry of the array
565 * nr_disks is total number of used slots.
566 * it should be raid_disks+spare_disks
567 * spare_disks is the number of extra disks present
569 * active_disks is the number of working disks in
570 * active slots. (With raid_disks)
571 * working_disks is the total number of working disks,
573 * failed_disks is the number of disks marked failed
575 * Ideally, the kernel would keep these (except raid_disks)
576 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
577 * So for now, we assume that all raid and spare
578 * devices will be given.
580 info
.array
.spare_disks
=sparedisks
;
581 info
.array
.failed_disks
=missing_disks
;
582 info
.array
.nr_disks
= info
.array
.working_disks
583 + info
.array
.failed_disks
;
584 info
.array
.layout
= layout
;
585 info
.array
.chunk_size
= chunk
*1024;
587 if (name
== NULL
|| *name
== 0) {
588 /* base name on mddev */
593 * /dev/md/home -> home
594 * /dev/mdhome -> home
596 /* FIXME compare this with rules in create_mddev */
597 name
= strrchr(mddev
, '/');
600 if (strncmp(name
, "md_d", 4)==0 &&
603 (name
-mddev
) == 5 /* /dev/ */)
605 else if (strncmp(name
, "md", 2)==0 &&
608 (name
-mddev
) == 5 /* /dev/ */)
612 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
615 total_slots
= info
.array
.nr_disks
;
616 sysfs_init(&info
, mdfd
, 0);
617 st
->ss
->getinfo_super(st
, &info
, NULL
);
619 if (did_default
&& verbose
>= 0) {
620 if (is_subarray(info
.text_version
)) {
621 int dnum
= devname2devnum(info
.text_version
+1);
623 int mdp
= get_mdp_major();
626 path
= map_dev(MD_MAJOR
, dnum
, 1);
628 path
= map_dev(mdp
, (-1-dnum
)<< 6, 1);
630 mdi
= sysfs_read(-1, dnum
, GET_VERSION
);
632 fprintf(stderr
, Name
": Creating array inside "
634 mdi
?mdi
->text_version
:"managed", path
);
637 fprintf(stderr
, Name
": Defaulting to version"
638 " %s metadata\n", info
.text_version
);
641 map_update(&map
, fd2devnum(mdfd
), info
.text_version
,
642 info
.uuid
, chosen_name
);
645 if (bitmap_file
&& vers
< 9003) {
646 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
648 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
649 " between different architectured. Consider upgrading the Linux kernel.\n");
653 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
654 if ((vers
%100) < 2) {
655 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
658 if (!st
->ss
->add_internal_bitmap
) {
659 fprintf(stderr
, Name
": internal bitmaps not supported with %s metadata\n",
663 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
665 bitmapsize
, 1, major_num
)) {
666 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
673 sysfs_init(&info
, mdfd
, 0);
675 if (st
->ss
->external
&& st
->subarray
[0]) {
678 /* When creating a member, we need to be careful
679 * to negotiate with mdmon properly.
680 * If it is already running, we cannot write to
681 * the devices and must ask it to do that part.
682 * If it isn't running, we write to the devices,
684 * We hold an exclusive open on the container
685 * device to make sure mdmon doesn't exit after
686 * we checked that it is running.
688 * For now, fail if it is already running.
690 container_fd
= open_dev_excl(st
->container_dev
);
691 if (container_fd
< 0) {
692 fprintf(stderr
, Name
": Cannot get exclusive "
693 "open on container - weird.\n");
696 if (mdmon_running(st
->container_dev
)) {
698 fprintf(stderr
, Name
": reusing mdmon "
700 devnum2devname(st
->container_dev
));
701 st
->update_tail
= &st
->updates
;
705 rv
= set_array_info(mdfd
, st
, &info
);
707 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
708 mddev
, strerror(errno
));
715 st
->ss
->uuid_from_super(st
, uuid
);
716 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
722 bitmap_fd
= open(bitmap_file
, O_RDWR
);
724 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
728 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
729 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
730 mddev
, strerror(errno
));
735 infos
= malloc(sizeof(*infos
) * total_slots
);
737 for (pass
=1; pass
<=2 ; pass
++) {
738 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
740 for (dnum
=0, dv
= devlist
; dv
;
741 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
744 struct mdinfo
*inf
= &infos
[dnum
];
746 if (dnum
>= total_slots
)
748 if (dnum
== insert_point
) {
752 if (strcasecmp(dv
->devname
, "missing")==0)
756 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
757 /* repeatedly use the container */
764 inf
->disk
.number
= dnum
;
765 inf
->disk
.raid_disk
= dnum
;
766 if (inf
->disk
.raid_disk
< raiddisks
)
767 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
772 if (dv
->writemostly
== 1)
773 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
778 if (st
->ss
->external
&& st
->subarray
[0])
779 fd
= open(dv
->devname
, O_RDWR
);
781 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
784 fprintf(stderr
, Name
": failed to open %s "
785 "after earlier success - aborting\n",
790 inf
->disk
.major
= major(stb
.st_rdev
);
791 inf
->disk
.minor
= minor(stb
.st_rdev
);
794 remove_partitions(fd
);
795 if (st
->ss
->add_to_super(st
, &inf
->disk
,
797 ioctl(mdfd
, STOP_ARRAY
, NULL
);
800 st
->ss
->getinfo_super(st
, inf
, NULL
);
801 safe_mode_delay
= inf
->safe_mode_delay
;
803 if (have_container
&& verbose
> 0)
804 fprintf(stderr
, Name
": Using %s for device %d\n",
805 map_dev(inf
->disk
.major
,
809 if (!have_container
) {
810 /* getinfo_super might have lost these ... */
811 inf
->disk
.major
= major(stb
.st_rdev
);
812 inf
->disk
.minor
= minor(stb
.st_rdev
);
819 rv
= add_disk(mdfd
, st
, &info
, inf
);
823 Name
": ADD_NEW_DISK for %s "
825 dv
->devname
, strerror(errno
));
826 st
->ss
->free_super(st
);
831 if (!have_container
&&
832 dv
== moved_disk
&& dnum
!= insert_point
) break;
835 struct mdinfo info_new
;
836 struct map_ent
*me
= NULL
;
838 /* check to see if the uuid has changed due to these
839 * metadata changes, and if so update the member array
840 * and container uuid. Note ->write_init_super clears
841 * the subarray cursor such that ->getinfo_super once
842 * again returns container info.
845 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
846 if (st
->ss
->external
&& level
!= LEVEL_CONTAINER
&&
847 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
848 map_update(&map
, fd2devnum(mdfd
),
849 info_new
.text_version
,
850 info_new
.uuid
, chosen_name
);
851 me
= map_by_devnum(&map
, st
->container_dev
);
854 st
->ss
->write_init_super(st
);
856 /* update parent container uuid */
858 char *path
= strdup(me
->path
);
860 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
861 map_update(&map
, st
->container_dev
,
862 info_new
.text_version
,
863 info_new
.uuid
, path
);
868 flush_metadata_updates(st
);
872 st
->ss
->free_super(st
);
874 if (level
== LEVEL_CONTAINER
) {
875 /* No need to start. But we should signal udev to
877 sysfs_uevent(&info
, "change");
879 fprintf(stderr
, Name
": container %s prepared.\n", mddev
);
880 wait_for(chosen_name
, mdfd
);
881 } else if (runstop
== 1 || subdevs
>= raiddisks
) {
882 if (st
->ss
->external
) {
885 case LEVEL_MULTIPATH
:
887 sysfs_set_str(&info
, NULL
, "array_state",
892 sysfs_set_str(&info
, NULL
, "array_state",
896 sysfs_set_safemode(&info
, safe_mode_delay
);
898 /* param is not actually used */
900 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
901 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
903 ioctl(mdfd
, STOP_ARRAY
, NULL
);
908 fprintf(stderr
, Name
": array %s started.\n", mddev
);
909 if (st
->ss
->external
&& st
->subarray
[0]) {
911 start_mdmon(st
->container_dev
);
913 ping_monitor(devnum2devname(st
->container_dev
));
916 wait_for(chosen_name
, mdfd
);
918 fprintf(stderr
, Name
": not starting array - not enough devices.\n");
925 map_remove(&map
, fd2devnum(mdfd
));