]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Create.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
35 int Create(struct supertype
*st
, char *mddev
,
36 int chunk
, int level
, int layout
, unsigned long long size
, int raiddisks
, int sparedisks
,
37 char *name
, char *homehost
, int *uuid
,
38 int subdevs
, mddev_dev_t devlist
,
39 int runstop
, int verbose
, int force
, int assume_clean
,
40 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
, int autof
)
43 * Create a new raid array.
45 * First check that necessary details are available
46 * (i.e. level, raid-disks)
48 * Then check each disk to see what might be on it
49 * and report anything interesting.
51 * If anything looks odd, and runstop not set,
54 * SET_ARRAY_INFO and ADD_NEW_DISK, and
55 * if runstop==run, or raiddisks disks were used,
59 unsigned long long minsize
=0, maxsize
=0;
66 int first_missing
= subdevs
* 2;
67 int second_missing
= subdevs
* 2;
68 int missing_disks
= 0;
69 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
75 int have_container
= 0;
76 int container_fd
= -1;
78 unsigned long long bitmapsize
;
79 struct mdinfo info
, *infos
;
81 unsigned long safe_mode_delay
= 0;
82 char chosen_name
[1024];
83 struct map_ent
*map
= NULL
;
84 unsigned long long newsize
;
86 int major_num
= BITMAP_MAJOR_HI
;
88 memset(&info
, 0, sizeof(info
));
91 /* "ddf" and "imsm" metadata only supports one level - should possibly
92 * push this into metadata handler??
94 if (st
&& (st
->ss
== &super_ddf
|| st
->ss
== &super_imsm
))
95 level
= LEVEL_CONTAINER
;
100 Name
": a RAID level is needed to create an array.\n");
103 if (raiddisks
< 4 && level
== 6) {
105 Name
": at least 4 raid-devices needed for level 6\n");
108 if (raiddisks
> 256 && level
== 6) {
110 Name
": no more than 256 raid-devices supported for level 6\n");
113 if (raiddisks
< 2 && level
>= 4) {
115 Name
": at least 2 raid-devices needed for level 4 or 5\n");
118 if (level
<= 0 && sparedisks
) {
120 Name
": This level does not support spare devices\n");
124 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
125 /* If given a single device, it might be a container, and we can
126 * extract a device list from there
128 mdu_array_info_t inf
;
131 memset(&inf
, 0, sizeof(inf
));
132 fd
= open(devlist
->devname
, O_RDONLY
);
134 ioctl(fd
, GET_ARRAY_INFO
, &inf
) == 0 &&
135 inf
.raid_disks
== 0) {
136 /* yep, looks like a container */
138 rv
= st
->ss
->load_super(st
, fd
,
143 st
= guess_super(fd
);
144 if (st
&& !(rv
= st
->ss
->
157 if (st
&& st
->ss
->external
&& sparedisks
) {
159 Name
": This metadata type does not support "
160 "spare disks are create time\n");
163 if (subdevs
> raiddisks
+sparedisks
) {
164 fprintf(stderr
, Name
": You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, raiddisks
+sparedisks
);
167 if (!have_container
&& subdevs
< raiddisks
+sparedisks
) {
168 fprintf(stderr
, Name
": You haven't given enough devices (real or missing) to create this array\n");
171 if (bitmap_file
&& level
<= 0) {
172 fprintf(stderr
, Name
": bitmaps not meaningful with level %s\n",
173 map_num(pers
, level
)?:"given");
177 /* now set some defaults */
180 default: /* no layout */
184 layout
= 0x102; /* near=2, far=1 */
187 Name
": layout defaults to n1\n");
191 layout
= map_name(r5layout
, "default");
194 Name
": layout defaults to %s\n", map_num(r5layout
, layout
));
197 layout
= map_name(faultylayout
, "default");
201 Name
": layout defaults to %s\n", map_num(faultylayout
, layout
));
206 /* check layout fits in array*/
207 if ((layout
&255) * ((layout
>>8)&255) > raiddisks
) {
208 fprintf(stderr
, Name
": that layout requires at least %d devices\n",
209 (layout
&255) * ((layout
>>8)&255));
219 case LEVEL_LINEAR
: /* linear */
223 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
228 case LEVEL_MULTIPATH
:
229 case LEVEL_CONTAINER
:
233 fprintf(stderr
, Name
": chunk size ignored for this level\n");
237 fprintf(stderr
, Name
": unknown level %d\n", level
);
242 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
243 chunk
, size
*2, NULL
, &newsize
, verbose
>=0))
247 if (size
&& verbose
> 0)
248 fprintf(stderr
, Name
": setting size to %lluK\n",
249 (unsigned long long)size
);
252 /* now look at the subdevs */
253 info
.array
.active_disks
= 0;
254 info
.array
.working_disks
= 0;
256 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
257 char *dname
= dv
->devname
;
258 unsigned long long freesize
;
259 if (strcasecmp(dname
, "missing")==0) {
260 if (first_missing
> dnum
)
261 first_missing
= dnum
;
262 if (second_missing
> dnum
&& dnum
> first_missing
)
263 second_missing
= dnum
;
267 info
.array
.working_disks
++;
268 if (dnum
< raiddisks
)
269 info
.array
.active_disks
++;
271 struct createinfo
*ci
= conf_get_create_info();
276 /* Need to choose a default metadata, which is different
277 * depending on geometry of array.
280 char *name
= "default";
281 for(i
=0; !st
&& superlist
[i
]; i
++) {
282 st
= superlist
[i
]->match_metadata_desc(name
);
283 if (st
&& !st
->ss
->validate_geometry
284 (st
, level
, layout
, raiddisks
,
285 chunk
, size
*2, dname
, &freesize
,
291 fprintf(stderr
, Name
": device %s not suitable "
292 "for any style of array\n",
296 if (st
->ss
!= &super0
||
297 st
->minor_version
!= 90)
300 if (!st
->ss
->validate_geometry(st
, level
, layout
,
302 chunk
, size
*2, dname
,
307 Name
": %s is not suitable for "
315 freesize
/= 2; /* convert to K */
317 /* round to chunk size */
318 freesize
= freesize
& ~(chunk
-1);
321 if (size
&& freesize
< size
) {
322 fprintf(stderr
, Name
": %s is smaller than given size."
323 " %lluK < %lluK + metadata\n",
324 dname
, freesize
, size
);
328 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
332 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
336 if (runstop
!= 1 || verbose
>= 0) {
337 int fd
= open(dname
, O_RDONLY
);
339 fprintf(stderr
, Name
": Cannot open %s: %s\n",
340 dname
, strerror(errno
));
344 warn
|= check_ext2(fd
, dname
);
345 warn
|= check_reiser(fd
, dname
);
346 warn
|= check_raid(fd
, dname
);
351 info
.array
.working_disks
= raiddisks
;
353 fprintf(stderr
, Name
": create aborted\n");
357 if (mindisc
== NULL
&& !have_container
) {
358 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
361 if (level
> 0 || level
== LEVEL_MULTIPATH
362 || level
== LEVEL_FAULTY
363 || st
->ss
->external
) {
364 /* size is meaningful */
365 if (!st
->ss
->validate_geometry(st
, level
, layout
,
369 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
374 fprintf(stderr
, Name
": size set to %lluK\n", size
);
377 if (!have_container
&& level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
378 if (runstop
!= 1 || verbose
>= 0)
379 fprintf(stderr
, Name
": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
386 if (!ask("Continue creating array? ")) {
387 fprintf(stderr
, Name
": create aborted.\n");
392 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
396 /* If this is raid4/5, we want to configure the last active slot
397 * as missing, so that a reconstruct happens (faster than re-parity)
398 * FIX: Can we do this for raid6 as well?
400 if (st
->ss
->external
== 0 &&
401 assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
405 insert_point
= raiddisks
-1;
407 info
.array
.active_disks
--;
414 /* For raid6, if creating with 1 missing drive, make a good drive
415 * into a spare, else the create will fail
417 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
418 st
->ss
->external
== 0 &&
419 second_missing
>= raiddisks
&& level
== 6) {
420 insert_point
= raiddisks
- 1;
421 if (insert_point
== first_missing
)
424 info
.array
.active_disks
--;
428 if (level
<= 0 && first_missing
< subdevs
* 2) {
430 Name
": This level does not support missing devices\n");
434 /* We need to create the device */
436 mdfd
= create_mddev(mddev
, name
, autof
, LOCAL
, chosen_name
);
441 vers
= md_get_version(mdfd
);
443 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
446 mdu_array_info_t inf
;
447 memset(&inf
, 0, sizeof(inf
));
448 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
449 if (inf
.working_disks
!= 0) {
450 fprintf(stderr
, Name
": another array by this name"
451 " is already running.\n");
456 /* Ok, lets try some ioctls */
458 info
.array
.level
= level
;
459 info
.array
.size
= size
;
460 info
.array
.raid_disks
= raiddisks
;
461 /* The kernel should *know* what md_minor we are dealing
462 * with, but it chooses to trust me instead. Sigh
464 info
.array
.md_minor
= 0;
465 if (fstat(mdfd
, &stb
)==0)
466 info
.array
.md_minor
= minor(stb
.st_rdev
);
467 info
.array
.not_persistent
= 0;
469 if ( ( (level
== 4 || level
== 5) &&
470 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
472 ( level
== 6 && (insert_point
< raiddisks
473 || second_missing
< raiddisks
))
479 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
480 info
.resync_start
= ~0ULL;
482 info
.array
.state
= 0; /* not clean, but no errors */
483 info
.resync_start
= 0;
486 /* for raid10, the bitmap size is the capacity of the array,
487 * which is array.size * raid_disks / ncopies;
488 * .. but convert to sectors.
490 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
491 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
492 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
494 bitmapsize
= (unsigned long long)size
* 2;
496 /* There is lots of redundancy in these disk counts,
497 * raid_disks is the most meaningful value
498 * it describes the geometry of the array
500 * nr_disks is total number of used slots.
501 * it should be raid_disks+spare_disks
502 * spare_disks is the number of extra disks present
504 * active_disks is the number of working disks in
505 * active slots. (With raid_disks)
506 * working_disks is the total number of working disks,
508 * failed_disks is the number of disks marked failed
510 * Ideally, the kernel would keep these (except raid_disks)
511 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
512 * So for now, we assume that all raid and spare
513 * devices will be given.
515 info
.array
.spare_disks
=sparedisks
;
516 info
.array
.failed_disks
=missing_disks
;
517 info
.array
.nr_disks
= info
.array
.working_disks
518 + info
.array
.failed_disks
;
519 info
.array
.layout
= layout
;
520 info
.array
.chunk_size
= chunk
*1024;
522 if (name
== NULL
|| *name
== 0) {
523 /* base name on mddev */
528 * /dev/md/home -> home
529 * /dev/mdhome -> home
531 /* FIXME compare this with rules in create_mddev */
532 name
= strrchr(mddev
, '/');
535 if (strncmp(name
, "md_d", 4)==0 &&
538 (name
-mddev
) == 5 /* /dev/ */)
540 else if (strncmp(name
, "md", 2)==0 &&
543 (name
-mddev
) == 5 /* /dev/ */)
547 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
550 total_slots
= info
.array
.nr_disks
;
551 sysfs_init(&info
, mdfd
, 0);
552 st
->ss
->getinfo_super(st
, &info
);
554 if (did_default
&& verbose
>= 0) {
555 if (is_subarray(info
.text_version
)) {
556 int dnum
= devname2devnum(info
.text_version
+1);
558 int mdp
= get_mdp_major();
561 path
= map_dev(MD_MAJOR
, dnum
, 1);
563 path
= map_dev(mdp
, (-1-dnum
)<< 6, 1);
565 mdi
= sysfs_read(-1, dnum
, GET_VERSION
);
567 fprintf(stderr
, Name
": Creating array inside "
569 mdi
?mdi
->text_version
:"managed", path
);
572 fprintf(stderr
, Name
": Defaulting to version"
573 " %s metadata\n", info
.text_version
);
576 map_update(&map
, fd2devnum(mdfd
), info
.text_version
,
577 info
.uuid
, chosen_name
);
580 if (bitmap_file
&& vers
< 9003) {
581 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
583 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
584 " between different architectured. Consider upgrading the Linux kernel.\n");
588 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
589 if ((vers
%100) < 2) {
590 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
593 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
595 bitmapsize
, 1, major_num
)) {
596 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
603 sysfs_init(&info
, mdfd
, 0);
605 if (st
->ss
->external
&& st
->subarray
[0]) {
608 /* When creating a member, we need to be careful
609 * to negotiate with mdmon properly.
610 * If it is already running, we cannot write to
611 * the devices and must ask it to do that part.
612 * If it isn't running, we write to the devices,
614 * We hold an exclusive open on the container
615 * device to make sure mdmon doesn't exit after
616 * we checked that it is running.
618 * For now, fail if it is already running.
620 container_fd
= open_dev_excl(st
->container_dev
);
621 if (container_fd
< 0) {
622 fprintf(stderr
, Name
": Cannot get exclusive "
623 "open on container - weird.\n");
626 if (mdmon_running(st
->container_dev
)) {
628 fprintf(stderr
, Name
": reusing mdmon "
630 devnum2devname(st
->container_dev
));
631 st
->update_tail
= &st
->updates
;
635 rv
= set_array_info(mdfd
, st
, &info
);
637 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
638 mddev
, strerror(errno
));
645 st
->ss
->uuid_from_super(st
, uuid
);
646 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
652 bitmap_fd
= open(bitmap_file
, O_RDWR
);
654 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
658 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
659 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
660 mddev
, strerror(errno
));
665 infos
= malloc(sizeof(*infos
) * total_slots
);
667 for (pass
=1; pass
<=2 ; pass
++) {
668 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
670 for (dnum
=0, dv
= devlist
; dv
;
671 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
674 struct mdinfo
*inf
= &infos
[dnum
];
676 if (dnum
>= total_slots
)
678 if (dnum
== insert_point
) {
682 if (strcasecmp(dv
->devname
, "missing")==0)
686 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
687 /* repeatedly use the container */
694 inf
->disk
.number
= dnum
;
695 inf
->disk
.raid_disk
= dnum
;
696 if (inf
->disk
.raid_disk
< raiddisks
)
697 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
702 if (dv
->writemostly
== 1)
703 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
708 if (st
->ss
->external
&& st
->subarray
[0])
709 fd
= open(dv
->devname
, O_RDWR
);
711 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
714 fprintf(stderr
, Name
": failed to open %s "
715 "after earlier success - aborting\n",
720 inf
->disk
.major
= major(stb
.st_rdev
);
721 inf
->disk
.minor
= minor(stb
.st_rdev
);
724 remove_partitions(fd
);
725 if (st
->ss
->add_to_super(st
, &inf
->disk
,
728 st
->ss
->getinfo_super(st
, inf
);
729 safe_mode_delay
= inf
->safe_mode_delay
;
731 if (have_container
&& verbose
> 0)
732 fprintf(stderr
, Name
": Using %s for device %d\n",
733 map_dev(inf
->disk
.major
,
737 if (!have_container
) {
738 /* getinfo_super might have lost these ... */
739 inf
->disk
.major
= major(stb
.st_rdev
);
740 inf
->disk
.minor
= minor(stb
.st_rdev
);
747 rv
= add_disk(mdfd
, st
, &info
, inf
);
751 Name
": ADD_NEW_DISK for %s "
753 dv
->devname
, strerror(errno
));
754 st
->ss
->free_super(st
);
759 if (!have_container
&&
760 dv
== moved_disk
&& dnum
!= insert_point
) break;
763 st
->ss
->write_init_super(st
);
764 flush_metadata_updates(st
);
768 st
->ss
->free_super(st
);
770 if (level
== LEVEL_CONTAINER
) {
771 /* No need to start. But we should signal udev to
773 sysfs_uevent(&info
, "change");
775 fprintf(stderr
, Name
": container %s prepared.\n", mddev
);
776 } else if (runstop
== 1 || subdevs
>= raiddisks
) {
777 if (st
->ss
->external
) {
780 case LEVEL_MULTIPATH
:
782 sysfs_set_str(&info
, NULL
, "array_state",
787 sysfs_set_str(&info
, NULL
, "array_state",
791 sysfs_set_safemode(&info
, safe_mode_delay
);
793 /* param is not actually used */
795 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
796 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
798 Manage_runstop(mddev
, mdfd
, -1, 0);
803 fprintf(stderr
, Name
": array %s started.\n", mddev
);
804 if (st
->ss
->external
&& st
->subarray
[0]) {
806 start_mdmon(st
->container_dev
);
808 ping_monitor(devnum2devname(st
->container_dev
));
811 wait_for(chosen_name
);
813 fprintf(stderr
, Name
": not starting array - not enough devices.\n");