]>
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 static int default_layout(struct supertype
*st
, int level
, int verbose
)
39 if (st
&& st
->ss
->default_layout
)
40 layout
= st
->ss
->default_layout(level
);
44 default: /* no layout */
48 layout
= 0x102; /* near=2, far=1 */
51 Name
": layout defaults to n1\n");
55 layout
= map_name(r5layout
, "default");
58 Name
": layout defaults to %s\n", map_num(r5layout
, layout
));
61 layout
= map_name(faultylayout
, "default");
65 Name
": layout defaults to %s\n", map_num(faultylayout
, layout
));
73 int Create(struct supertype
*st
, char *mddev
,
74 int chunk
, int level
, int layout
, unsigned long long size
, int raiddisks
, int sparedisks
,
75 char *name
, char *homehost
, int *uuid
,
76 int subdevs
, mddev_dev_t devlist
,
77 int runstop
, int verbose
, int force
, int assume_clean
,
78 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
, int autof
)
81 * Create a new raid array.
83 * First check that necessary details are available
84 * (i.e. level, raid-disks)
86 * Then check each disk to see what might be on it
87 * and report anything interesting.
89 * If anything looks odd, and runstop not set,
92 * SET_ARRAY_INFO and ADD_NEW_DISK, and
93 * if runstop==run, or raiddisks disks were used,
97 unsigned long long minsize
=0, maxsize
=0;
104 int first_missing
= subdevs
* 2;
105 int second_missing
= subdevs
* 2;
106 int missing_disks
= 0;
107 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
113 int have_container
= 0;
114 int container_fd
= -1;
116 unsigned long long bitmapsize
;
117 struct mdinfo info
, *infos
;
119 int do_default_layout
= 0;
120 unsigned long safe_mode_delay
= 0;
121 char chosen_name
[1024];
122 struct map_ent
*map
= NULL
;
123 unsigned long long newsize
;
125 int major_num
= BITMAP_MAJOR_HI
;
127 memset(&info
, 0, sizeof(info
));
129 if (level
== UnSet
) {
130 /* "ddf" and "imsm" metadata only supports one level - should possibly
131 * push this into metadata handler??
133 if (st
&& (st
->ss
== &super_ddf
|| st
->ss
== &super_imsm
))
134 level
= LEVEL_CONTAINER
;
137 if (level
== UnSet
) {
139 Name
": a RAID level is needed to create an array.\n");
142 if (raiddisks
< 4 && level
== 6) {
144 Name
": at least 4 raid-devices needed for level 6\n");
147 if (raiddisks
> 256 && level
== 6) {
149 Name
": no more than 256 raid-devices supported for level 6\n");
152 if (raiddisks
< 2 && level
>= 4) {
154 Name
": at least 2 raid-devices needed for level 4 or 5\n");
157 if (level
<= 0 && sparedisks
) {
159 Name
": This level does not support spare devices\n");
163 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
164 /* If given a single device, it might be a container, and we can
165 * extract a device list from there
167 mdu_array_info_t inf
;
170 memset(&inf
, 0, sizeof(inf
));
171 fd
= open(devlist
->devname
, O_RDONLY
);
173 ioctl(fd
, GET_ARRAY_INFO
, &inf
) == 0 &&
174 inf
.raid_disks
== 0) {
175 /* yep, looks like a container */
177 rv
= st
->ss
->load_super(st
, fd
,
182 st
= guess_super(fd
);
183 if (st
&& !(rv
= st
->ss
->
196 if (st
&& st
->ss
->external
&& sparedisks
) {
198 Name
": This metadata type does not support "
199 "spare disks are create time\n");
202 if (subdevs
> raiddisks
+sparedisks
) {
203 fprintf(stderr
, Name
": You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, raiddisks
+sparedisks
);
206 if (!have_container
&& subdevs
< raiddisks
+sparedisks
) {
207 fprintf(stderr
, Name
": You haven't given enough devices (real or missing) to create this array\n");
210 if (bitmap_file
&& level
<= 0) {
211 fprintf(stderr
, Name
": bitmaps not meaningful with level %s\n",
212 map_num(pers
, level
)?:"given");
216 /* now set some defaults */
219 if (layout
== UnSet
) {
220 do_default_layout
= 1;
221 layout
= default_layout(st
, level
, verbose
);
225 /* check layout fits in array*/
226 if ((layout
&255) * ((layout
>>8)&255) > raiddisks
) {
227 fprintf(stderr
, Name
": that layout requires at least %d devices\n",
228 (layout
&255) * ((layout
>>8)&255));
238 case LEVEL_LINEAR
: /* linear */
242 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
247 case LEVEL_MULTIPATH
:
248 case LEVEL_CONTAINER
:
252 fprintf(stderr
, Name
": chunk size ignored for this level\n");
256 fprintf(stderr
, Name
": unknown level %d\n", level
);
261 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
262 chunk
, size
*2, NULL
, &newsize
, verbose
>=0))
266 if (size
&& verbose
> 0)
267 fprintf(stderr
, Name
": setting size to %lluK\n",
268 (unsigned long long)size
);
271 /* now look at the subdevs */
272 info
.array
.active_disks
= 0;
273 info
.array
.working_disks
= 0;
275 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
276 char *dname
= dv
->devname
;
277 unsigned long long freesize
;
278 if (strcasecmp(dname
, "missing")==0) {
279 if (first_missing
> dnum
)
280 first_missing
= dnum
;
281 if (second_missing
> dnum
&& dnum
> first_missing
)
282 second_missing
= dnum
;
286 info
.array
.working_disks
++;
287 if (dnum
< raiddisks
)
288 info
.array
.active_disks
++;
290 struct createinfo
*ci
= conf_get_create_info();
295 /* Need to choose a default metadata, which is different
296 * depending on geometry of array.
299 char *name
= "default";
300 for(i
=0; !st
&& superlist
[i
]; i
++) {
301 st
= superlist
[i
]->match_metadata_desc(name
);
302 if (do_default_layout
)
303 layout
= default_layout(st
, level
, verbose
);
304 if (st
&& !st
->ss
->validate_geometry
305 (st
, level
, layout
, raiddisks
,
306 chunk
, size
*2, dname
, &freesize
,
312 fprintf(stderr
, Name
": device %s not suitable "
313 "for any style of array\n",
317 if (st
->ss
!= &super0
||
318 st
->minor_version
!= 90)
321 if (do_default_layout
)
322 layout
= default_layout(st
, level
, verbose
);
323 if (!st
->ss
->validate_geometry(st
, level
, layout
,
325 chunk
, size
*2, dname
,
330 Name
": %s is not suitable for "
338 freesize
/= 2; /* convert to K */
340 /* round to chunk size */
341 freesize
= freesize
& ~(chunk
-1);
344 if (size
&& freesize
< size
) {
345 fprintf(stderr
, Name
": %s is smaller than given size."
346 " %lluK < %lluK + metadata\n",
347 dname
, freesize
, size
);
351 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
355 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
359 if (runstop
!= 1 || verbose
>= 0) {
360 int fd
= open(dname
, O_RDONLY
);
362 fprintf(stderr
, Name
": Cannot open %s: %s\n",
363 dname
, strerror(errno
));
367 warn
|= check_ext2(fd
, dname
);
368 warn
|= check_reiser(fd
, dname
);
369 warn
|= check_raid(fd
, dname
);
374 info
.array
.working_disks
= raiddisks
;
376 fprintf(stderr
, Name
": create aborted\n");
380 if (mindisc
== NULL
&& !have_container
) {
381 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
384 if (level
> 0 || level
== LEVEL_MULTIPATH
385 || level
== LEVEL_FAULTY
386 || st
->ss
->external
) {
387 /* size is meaningful */
388 if (!st
->ss
->validate_geometry(st
, level
, layout
,
392 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
397 fprintf(stderr
, Name
": size set to %lluK\n", size
);
400 if (!have_container
&& level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
401 if (runstop
!= 1 || verbose
>= 0)
402 fprintf(stderr
, Name
": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
407 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1) != 0) {
408 if (runstop
!= 1 || verbose
>= 0)
409 fprintf(stderr
, Name
": %s unable to enumerate platform support\n"
410 " array may not be compatible with hardware/firmware\n",
417 if (!ask("Continue creating array? ")) {
418 fprintf(stderr
, Name
": create aborted.\n");
423 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
427 /* If this is raid4/5, we want to configure the last active slot
428 * as missing, so that a reconstruct happens (faster than re-parity)
429 * FIX: Can we do this for raid6 as well?
431 if (st
->ss
->external
== 0 &&
432 assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
436 insert_point
= raiddisks
-1;
438 info
.array
.active_disks
--;
445 /* For raid6, if creating with 1 missing drive, make a good drive
446 * into a spare, else the create will fail
448 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
449 st
->ss
->external
== 0 &&
450 second_missing
>= raiddisks
&& level
== 6) {
451 insert_point
= raiddisks
- 1;
452 if (insert_point
== first_missing
)
455 info
.array
.active_disks
--;
459 if (level
<= 0 && first_missing
< subdevs
* 2) {
461 Name
": This level does not support missing devices\n");
465 /* We need to create the device */
467 mdfd
= create_mddev(mddev
, name
, autof
, LOCAL
, chosen_name
);
472 vers
= md_get_version(mdfd
);
474 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
477 mdu_array_info_t inf
;
478 memset(&inf
, 0, sizeof(inf
));
479 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
480 if (inf
.working_disks
!= 0) {
481 fprintf(stderr
, Name
": another array by this name"
482 " is already running.\n");
487 /* Ok, lets try some ioctls */
489 info
.array
.level
= level
;
490 info
.array
.size
= size
;
491 info
.array
.raid_disks
= raiddisks
;
492 /* The kernel should *know* what md_minor we are dealing
493 * with, but it chooses to trust me instead. Sigh
495 info
.array
.md_minor
= 0;
496 if (fstat(mdfd
, &stb
)==0)
497 info
.array
.md_minor
= minor(stb
.st_rdev
);
498 info
.array
.not_persistent
= 0;
500 if ( ( (level
== 4 || level
== 5) &&
501 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
503 ( level
== 6 && (insert_point
< raiddisks
504 || second_missing
< raiddisks
))
510 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
511 info
.resync_start
= ~0ULL;
513 info
.array
.state
= 0; /* not clean, but no errors */
514 info
.resync_start
= 0;
517 /* for raid10, the bitmap size is the capacity of the array,
518 * which is array.size * raid_disks / ncopies;
519 * .. but convert to sectors.
521 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
522 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
523 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
525 bitmapsize
= (unsigned long long)size
* 2;
527 /* There is lots of redundancy in these disk counts,
528 * raid_disks is the most meaningful value
529 * it describes the geometry of the array
531 * nr_disks is total number of used slots.
532 * it should be raid_disks+spare_disks
533 * spare_disks is the number of extra disks present
535 * active_disks is the number of working disks in
536 * active slots. (With raid_disks)
537 * working_disks is the total number of working disks,
539 * failed_disks is the number of disks marked failed
541 * Ideally, the kernel would keep these (except raid_disks)
542 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
543 * So for now, we assume that all raid and spare
544 * devices will be given.
546 info
.array
.spare_disks
=sparedisks
;
547 info
.array
.failed_disks
=missing_disks
;
548 info
.array
.nr_disks
= info
.array
.working_disks
549 + info
.array
.failed_disks
;
550 info
.array
.layout
= layout
;
551 info
.array
.chunk_size
= chunk
*1024;
553 if (name
== NULL
|| *name
== 0) {
554 /* base name on mddev */
559 * /dev/md/home -> home
560 * /dev/mdhome -> home
562 /* FIXME compare this with rules in create_mddev */
563 name
= strrchr(mddev
, '/');
566 if (strncmp(name
, "md_d", 4)==0 &&
569 (name
-mddev
) == 5 /* /dev/ */)
571 else if (strncmp(name
, "md", 2)==0 &&
574 (name
-mddev
) == 5 /* /dev/ */)
578 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
581 total_slots
= info
.array
.nr_disks
;
582 sysfs_init(&info
, mdfd
, 0);
583 st
->ss
->getinfo_super(st
, &info
);
585 if (did_default
&& verbose
>= 0) {
586 if (is_subarray(info
.text_version
)) {
587 int dnum
= devname2devnum(info
.text_version
+1);
589 int mdp
= get_mdp_major();
592 path
= map_dev(MD_MAJOR
, dnum
, 1);
594 path
= map_dev(mdp
, (-1-dnum
)<< 6, 1);
596 mdi
= sysfs_read(-1, dnum
, GET_VERSION
);
598 fprintf(stderr
, Name
": Creating array inside "
600 mdi
?mdi
->text_version
:"managed", path
);
603 fprintf(stderr
, Name
": Defaulting to version"
604 " %s metadata\n", info
.text_version
);
607 map_update(&map
, fd2devnum(mdfd
), info
.text_version
,
608 info
.uuid
, chosen_name
);
611 if (bitmap_file
&& vers
< 9003) {
612 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
614 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
615 " between different architectured. Consider upgrading the Linux kernel.\n");
619 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
620 if ((vers
%100) < 2) {
621 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
624 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
626 bitmapsize
, 1, major_num
)) {
627 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
634 sysfs_init(&info
, mdfd
, 0);
636 if (st
->ss
->external
&& st
->subarray
[0]) {
639 /* When creating a member, we need to be careful
640 * to negotiate with mdmon properly.
641 * If it is already running, we cannot write to
642 * the devices and must ask it to do that part.
643 * If it isn't running, we write to the devices,
645 * We hold an exclusive open on the container
646 * device to make sure mdmon doesn't exit after
647 * we checked that it is running.
649 * For now, fail if it is already running.
651 container_fd
= open_dev_excl(st
->container_dev
);
652 if (container_fd
< 0) {
653 fprintf(stderr
, Name
": Cannot get exclusive "
654 "open on container - weird.\n");
657 if (mdmon_running(st
->container_dev
)) {
659 fprintf(stderr
, Name
": reusing mdmon "
661 devnum2devname(st
->container_dev
));
662 st
->update_tail
= &st
->updates
;
666 rv
= set_array_info(mdfd
, st
, &info
);
668 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
669 mddev
, strerror(errno
));
676 st
->ss
->uuid_from_super(st
, uuid
);
677 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
683 bitmap_fd
= open(bitmap_file
, O_RDWR
);
685 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
689 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
690 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
691 mddev
, strerror(errno
));
696 infos
= malloc(sizeof(*infos
) * total_slots
);
698 for (pass
=1; pass
<=2 ; pass
++) {
699 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
701 for (dnum
=0, dv
= devlist
; dv
;
702 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
705 struct mdinfo
*inf
= &infos
[dnum
];
707 if (dnum
>= total_slots
)
709 if (dnum
== insert_point
) {
713 if (strcasecmp(dv
->devname
, "missing")==0)
717 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
718 /* repeatedly use the container */
725 inf
->disk
.number
= dnum
;
726 inf
->disk
.raid_disk
= dnum
;
727 if (inf
->disk
.raid_disk
< raiddisks
)
728 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
733 if (dv
->writemostly
== 1)
734 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
739 if (st
->ss
->external
&& st
->subarray
[0])
740 fd
= open(dv
->devname
, O_RDWR
);
742 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
745 fprintf(stderr
, Name
": failed to open %s "
746 "after earlier success - aborting\n",
751 inf
->disk
.major
= major(stb
.st_rdev
);
752 inf
->disk
.minor
= minor(stb
.st_rdev
);
755 remove_partitions(fd
);
756 if (st
->ss
->add_to_super(st
, &inf
->disk
,
759 st
->ss
->getinfo_super(st
, inf
);
760 safe_mode_delay
= inf
->safe_mode_delay
;
762 if (have_container
&& verbose
> 0)
763 fprintf(stderr
, Name
": Using %s for device %d\n",
764 map_dev(inf
->disk
.major
,
768 if (!have_container
) {
769 /* getinfo_super might have lost these ... */
770 inf
->disk
.major
= major(stb
.st_rdev
);
771 inf
->disk
.minor
= minor(stb
.st_rdev
);
778 rv
= add_disk(mdfd
, st
, &info
, inf
);
782 Name
": ADD_NEW_DISK for %s "
784 dv
->devname
, strerror(errno
));
785 st
->ss
->free_super(st
);
790 if (!have_container
&&
791 dv
== moved_disk
&& dnum
!= insert_point
) break;
794 st
->ss
->write_init_super(st
);
795 flush_metadata_updates(st
);
799 st
->ss
->free_super(st
);
801 if (level
== LEVEL_CONTAINER
) {
802 /* No need to start. But we should signal udev to
804 sysfs_uevent(&info
, "change");
806 fprintf(stderr
, Name
": container %s prepared.\n", mddev
);
807 } else if (runstop
== 1 || subdevs
>= raiddisks
) {
808 if (st
->ss
->external
) {
811 case LEVEL_MULTIPATH
:
813 sysfs_set_str(&info
, NULL
, "array_state",
818 sysfs_set_str(&info
, NULL
, "array_state",
822 sysfs_set_safemode(&info
, safe_mode_delay
);
824 /* param is not actually used */
826 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
827 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
829 Manage_runstop(mddev
, mdfd
, -1, 0);
834 fprintf(stderr
, Name
": array %s started.\n", mddev
);
835 if (st
->ss
->external
&& st
->subarray
[0]) {
837 start_mdmon(st
->container_dev
);
839 ping_monitor(devnum2devname(st
->container_dev
));
842 wait_for(chosen_name
);
844 fprintf(stderr
, Name
": not starting array - not enough devices.\n");