]>
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
->
190 if (have_container
) {
192 first_missing
= subdevs
* 2;
193 second_missing
= subdevs
* 2;
194 insert_point
= subdevs
* 2;
200 if (st
&& st
->ss
->external
&& sparedisks
) {
202 Name
": This metadata type does not support "
203 "spare disks are create time\n");
206 if (subdevs
> raiddisks
+sparedisks
) {
207 fprintf(stderr
, Name
": You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, raiddisks
+sparedisks
);
210 if (!have_container
&& subdevs
< raiddisks
+sparedisks
) {
211 fprintf(stderr
, Name
": You haven't given enough devices (real or missing) to create this array\n");
214 if (bitmap_file
&& level
<= 0) {
215 fprintf(stderr
, Name
": bitmaps not meaningful with level %s\n",
216 map_num(pers
, level
)?:"given");
220 /* now set some defaults */
223 if (layout
== UnSet
) {
224 do_default_layout
= 1;
225 layout
= default_layout(st
, level
, verbose
);
229 /* check layout fits in array*/
230 if ((layout
&255) * ((layout
>>8)&255) > raiddisks
) {
231 fprintf(stderr
, Name
": that layout requires at least %d devices\n",
232 (layout
&255) * ((layout
>>8)&255));
242 case LEVEL_LINEAR
: /* linear */
246 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
251 case LEVEL_MULTIPATH
:
252 case LEVEL_CONTAINER
:
256 fprintf(stderr
, Name
": chunk size ignored for this level\n");
260 fprintf(stderr
, Name
": unknown level %d\n", level
);
265 size
&= ~(unsigned long long)(chunk
- 1);
267 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
268 chunk
, size
*2, NULL
, &newsize
, verbose
>=0))
272 if (size
&& verbose
> 0)
273 fprintf(stderr
, Name
": setting size to %lluK\n",
274 (unsigned long long)size
);
277 /* now look at the subdevs */
278 info
.array
.active_disks
= 0;
279 info
.array
.working_disks
= 0;
281 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
282 char *dname
= dv
->devname
;
283 unsigned long long freesize
;
284 if (strcasecmp(dname
, "missing")==0) {
285 if (first_missing
> dnum
)
286 first_missing
= dnum
;
287 if (second_missing
> dnum
&& dnum
> first_missing
)
288 second_missing
= dnum
;
292 info
.array
.working_disks
++;
293 if (dnum
< raiddisks
)
294 info
.array
.active_disks
++;
296 struct createinfo
*ci
= conf_get_create_info();
301 /* Need to choose a default metadata, which is different
302 * depending on geometry of array.
305 char *name
= "default";
306 for(i
=0; !st
&& superlist
[i
]; i
++) {
307 st
= superlist
[i
]->match_metadata_desc(name
);
308 if (do_default_layout
)
309 layout
= default_layout(st
, level
, verbose
);
310 if (st
&& !st
->ss
->validate_geometry
311 (st
, level
, layout
, raiddisks
,
312 chunk
, size
*2, dname
, &freesize
,
318 fprintf(stderr
, Name
": device %s not suitable "
319 "for any style of array\n",
323 if (st
->ss
!= &super0
||
324 st
->minor_version
!= 90)
327 if (do_default_layout
)
328 layout
= default_layout(st
, level
, verbose
);
329 if (!st
->ss
->validate_geometry(st
, level
, layout
,
331 chunk
, size
*2, dname
,
336 Name
": %s is not suitable for "
344 freesize
/= 2; /* convert to K */
346 /* round to chunk size */
347 freesize
= freesize
& ~(chunk
-1);
350 if (size
&& freesize
< size
) {
351 fprintf(stderr
, Name
": %s is smaller than given size."
352 " %lluK < %lluK + metadata\n",
353 dname
, freesize
, size
);
357 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
361 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
365 if (runstop
!= 1 || verbose
>= 0) {
366 int fd
= open(dname
, O_RDONLY
);
368 fprintf(stderr
, Name
": Cannot open %s: %s\n",
369 dname
, strerror(errno
));
373 warn
|= check_ext2(fd
, dname
);
374 warn
|= check_reiser(fd
, dname
);
375 warn
|= check_raid(fd
, dname
);
380 info
.array
.working_disks
= raiddisks
;
382 fprintf(stderr
, Name
": create aborted\n");
386 if (mindisc
== NULL
&& !have_container
) {
387 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
390 if (level
> 0 || level
== LEVEL_MULTIPATH
391 || level
== LEVEL_FAULTY
392 || st
->ss
->external
) {
393 /* size is meaningful */
394 if (!st
->ss
->validate_geometry(st
, level
, layout
,
398 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
403 fprintf(stderr
, Name
": size set to %lluK\n", size
);
406 if (!have_container
&& level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
407 if (runstop
!= 1 || verbose
>= 0)
408 fprintf(stderr
, Name
": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
413 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1) != 0) {
414 if (runstop
!= 1 || verbose
>= 0)
415 fprintf(stderr
, Name
": %s unable to enumerate platform support\n"
416 " array may not be compatible with hardware/firmware\n",
423 if (!ask("Continue creating array? ")) {
424 fprintf(stderr
, Name
": create aborted.\n");
429 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
433 /* If this is raid4/5, we want to configure the last active slot
434 * as missing, so that a reconstruct happens (faster than re-parity)
435 * FIX: Can we do this for raid6 as well?
437 if (st
->ss
->external
== 0 &&
438 assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
442 insert_point
= raiddisks
-1;
444 info
.array
.active_disks
--;
451 /* For raid6, if creating with 1 missing drive, make a good drive
452 * into a spare, else the create will fail
454 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
455 st
->ss
->external
== 0 &&
456 second_missing
>= raiddisks
&& level
== 6) {
457 insert_point
= raiddisks
- 1;
458 if (insert_point
== first_missing
)
461 info
.array
.active_disks
--;
465 if (level
<= 0 && first_missing
< subdevs
* 2) {
467 Name
": This level does not support missing devices\n");
471 /* We need to create the device */
473 mdfd
= create_mddev(mddev
, name
, autof
, LOCAL
, chosen_name
);
478 vers
= md_get_version(mdfd
);
480 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
483 mdu_array_info_t inf
;
484 memset(&inf
, 0, sizeof(inf
));
485 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
486 if (inf
.working_disks
!= 0) {
487 fprintf(stderr
, Name
": another array by this name"
488 " is already running.\n");
493 /* Ok, lets try some ioctls */
495 info
.array
.level
= level
;
496 info
.array
.size
= size
;
497 info
.array
.raid_disks
= raiddisks
;
498 /* The kernel should *know* what md_minor we are dealing
499 * with, but it chooses to trust me instead. Sigh
501 info
.array
.md_minor
= 0;
502 if (fstat(mdfd
, &stb
)==0)
503 info
.array
.md_minor
= minor(stb
.st_rdev
);
504 info
.array
.not_persistent
= 0;
506 if ( ( (level
== 4 || level
== 5) &&
507 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
509 ( level
== 6 && (insert_point
< raiddisks
510 || second_missing
< raiddisks
))
516 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
517 info
.resync_start
= ~0ULL;
519 info
.array
.state
= 0; /* not clean, but no errors */
520 info
.resync_start
= 0;
523 /* for raid10, the bitmap size is the capacity of the array,
524 * which is array.size * raid_disks / ncopies;
525 * .. but convert to sectors.
527 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
528 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
529 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
531 bitmapsize
= (unsigned long long)size
* 2;
533 /* There is lots of redundancy in these disk counts,
534 * raid_disks is the most meaningful value
535 * it describes the geometry of the array
537 * nr_disks is total number of used slots.
538 * it should be raid_disks+spare_disks
539 * spare_disks is the number of extra disks present
541 * active_disks is the number of working disks in
542 * active slots. (With raid_disks)
543 * working_disks is the total number of working disks,
545 * failed_disks is the number of disks marked failed
547 * Ideally, the kernel would keep these (except raid_disks)
548 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
549 * So for now, we assume that all raid and spare
550 * devices will be given.
552 info
.array
.spare_disks
=sparedisks
;
553 info
.array
.failed_disks
=missing_disks
;
554 info
.array
.nr_disks
= info
.array
.working_disks
555 + info
.array
.failed_disks
;
556 info
.array
.layout
= layout
;
557 info
.array
.chunk_size
= chunk
*1024;
559 if (name
== NULL
|| *name
== 0) {
560 /* base name on mddev */
565 * /dev/md/home -> home
566 * /dev/mdhome -> home
568 /* FIXME compare this with rules in create_mddev */
569 name
= strrchr(mddev
, '/');
572 if (strncmp(name
, "md_d", 4)==0 &&
575 (name
-mddev
) == 5 /* /dev/ */)
577 else if (strncmp(name
, "md", 2)==0 &&
580 (name
-mddev
) == 5 /* /dev/ */)
584 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
587 total_slots
= info
.array
.nr_disks
;
588 sysfs_init(&info
, mdfd
, 0);
589 st
->ss
->getinfo_super(st
, &info
);
591 if (did_default
&& verbose
>= 0) {
592 if (is_subarray(info
.text_version
)) {
593 int dnum
= devname2devnum(info
.text_version
+1);
595 int mdp
= get_mdp_major();
598 path
= map_dev(MD_MAJOR
, dnum
, 1);
600 path
= map_dev(mdp
, (-1-dnum
)<< 6, 1);
602 mdi
= sysfs_read(-1, dnum
, GET_VERSION
);
604 fprintf(stderr
, Name
": Creating array inside "
606 mdi
?mdi
->text_version
:"managed", path
);
609 fprintf(stderr
, Name
": Defaulting to version"
610 " %s metadata\n", info
.text_version
);
613 map_update(&map
, fd2devnum(mdfd
), info
.text_version
,
614 info
.uuid
, chosen_name
);
617 if (bitmap_file
&& vers
< 9003) {
618 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
620 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
621 " between different architectured. Consider upgrading the Linux kernel.\n");
625 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
626 if ((vers
%100) < 2) {
627 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
630 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
632 bitmapsize
, 1, major_num
)) {
633 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
640 sysfs_init(&info
, mdfd
, 0);
642 if (st
->ss
->external
&& st
->subarray
[0]) {
645 /* When creating a member, we need to be careful
646 * to negotiate with mdmon properly.
647 * If it is already running, we cannot write to
648 * the devices and must ask it to do that part.
649 * If it isn't running, we write to the devices,
651 * We hold an exclusive open on the container
652 * device to make sure mdmon doesn't exit after
653 * we checked that it is running.
655 * For now, fail if it is already running.
657 container_fd
= open_dev_excl(st
->container_dev
);
658 if (container_fd
< 0) {
659 fprintf(stderr
, Name
": Cannot get exclusive "
660 "open on container - weird.\n");
663 if (mdmon_running(st
->container_dev
)) {
665 fprintf(stderr
, Name
": reusing mdmon "
667 devnum2devname(st
->container_dev
));
668 st
->update_tail
= &st
->updates
;
672 rv
= set_array_info(mdfd
, st
, &info
);
674 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
675 mddev
, strerror(errno
));
682 st
->ss
->uuid_from_super(st
, uuid
);
683 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
689 bitmap_fd
= open(bitmap_file
, O_RDWR
);
691 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
695 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
696 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
697 mddev
, strerror(errno
));
702 infos
= malloc(sizeof(*infos
) * total_slots
);
704 for (pass
=1; pass
<=2 ; pass
++) {
705 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
707 for (dnum
=0, dv
= devlist
; dv
;
708 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
711 struct mdinfo
*inf
= &infos
[dnum
];
713 if (dnum
>= total_slots
)
715 if (dnum
== insert_point
) {
719 if (strcasecmp(dv
->devname
, "missing")==0)
723 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
724 /* repeatedly use the container */
731 inf
->disk
.number
= dnum
;
732 inf
->disk
.raid_disk
= dnum
;
733 if (inf
->disk
.raid_disk
< raiddisks
)
734 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
739 if (dv
->writemostly
== 1)
740 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
745 if (st
->ss
->external
&& st
->subarray
[0])
746 fd
= open(dv
->devname
, O_RDWR
);
748 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
751 fprintf(stderr
, Name
": failed to open %s "
752 "after earlier success - aborting\n",
757 inf
->disk
.major
= major(stb
.st_rdev
);
758 inf
->disk
.minor
= minor(stb
.st_rdev
);
761 remove_partitions(fd
);
762 if (st
->ss
->add_to_super(st
, &inf
->disk
,
765 st
->ss
->getinfo_super(st
, inf
);
766 safe_mode_delay
= inf
->safe_mode_delay
;
768 if (have_container
&& verbose
> 0)
769 fprintf(stderr
, Name
": Using %s for device %d\n",
770 map_dev(inf
->disk
.major
,
774 if (!have_container
) {
775 /* getinfo_super might have lost these ... */
776 inf
->disk
.major
= major(stb
.st_rdev
);
777 inf
->disk
.minor
= minor(stb
.st_rdev
);
784 rv
= add_disk(mdfd
, st
, &info
, inf
);
788 Name
": ADD_NEW_DISK for %s "
790 dv
->devname
, strerror(errno
));
791 st
->ss
->free_super(st
);
796 if (!have_container
&&
797 dv
== moved_disk
&& dnum
!= insert_point
) break;
800 st
->ss
->write_init_super(st
);
801 flush_metadata_updates(st
);
805 st
->ss
->free_super(st
);
807 if (level
== LEVEL_CONTAINER
) {
808 /* No need to start. But we should signal udev to
810 sysfs_uevent(&info
, "change");
812 fprintf(stderr
, Name
": container %s prepared.\n", mddev
);
813 wait_for(chosen_name
, mdfd
);
814 } else if (runstop
== 1 || subdevs
>= raiddisks
) {
815 if (st
->ss
->external
) {
818 case LEVEL_MULTIPATH
:
820 sysfs_set_str(&info
, NULL
, "array_state",
825 sysfs_set_str(&info
, NULL
, "array_state",
829 sysfs_set_safemode(&info
, safe_mode_delay
);
831 /* param is not actually used */
833 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
834 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
836 Manage_runstop(mddev
, mdfd
, -1, 0);
841 fprintf(stderr
, Name
": array %s started.\n", mddev
);
842 if (st
->ss
->external
&& st
->subarray
[0]) {
844 start_mdmon(st
->container_dev
);
846 ping_monitor(devnum2devname(st
->container_dev
));
849 wait_for(chosen_name
, mdfd
);
851 fprintf(stderr
, Name
": not starting array - not enough devices.\n");