2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 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_geometry
)
35 st
->ss
->default_geometry(st
, &level
, &layout
, NULL
);
39 default: /* no layout */
43 layout
= 0x102; /* near=2, far=1 */
45 pr_err("layout defaults to n2\n");
49 layout
= map_name(r5layout
, "default");
51 pr_err("layout defaults to %s\n", map_num(r5layout
, layout
));
54 layout
= map_name(faultylayout
, "default");
57 pr_err("layout defaults to %s\n", map_num(faultylayout
, layout
));
64 int Create(struct supertype
*st
, char *mddev
,
65 char *name
, int *uuid
,
66 int subdevs
, struct mddev_dev
*devlist
,
68 struct context
*c
, unsigned long long data_offset
)
71 * Create a new raid array.
73 * First check that necessary details are available
74 * (i.e. level, raid-disks)
76 * Then check each disk to see what might be on it
77 * and report anything interesting.
79 * If anything looks odd, and runstop not set,
82 * SET_ARRAY_INFO and ADD_NEW_DISK, and
83 * if runstop==run, or raiddisks disks were used,
87 unsigned long long minsize
= 0, maxsize
= 0;
90 int dnum
, raid_disk_num
;
93 int fail
= 0, warn
= 0;
94 int first_missing
= subdevs
* 2;
95 int second_missing
= subdevs
* 2;
96 int missing_disks
= 0;
97 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
102 int have_container
= 0;
103 int container_fd
= -1;
105 unsigned long long bitmapsize
;
106 struct mdinfo info
, *infos
;
108 int do_default_layout
= 0;
109 int do_default_chunk
= 0;
110 unsigned long safe_mode_delay
= 0;
111 char chosen_name
[1024];
112 struct map_ent
*map
= NULL
;
113 unsigned long long newsize
;
114 mdu_array_info_t inf
;
116 int major_num
= BITMAP_MAJOR_HI
;
117 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "clustered") == 0) {
118 major_num
= BITMAP_MAJOR_CLUSTERED
;
120 pr_err("At least 2 nodes are needed for cluster-md\n");
125 memset(&info
, 0, sizeof(info
));
126 if (s
->level
== UnSet
&& st
&& st
->ss
->default_geometry
)
127 st
->ss
->default_geometry(st
, &s
->level
, NULL
, NULL
);
128 if (s
->level
== UnSet
) {
129 pr_err("a RAID level is needed to create an array.\n");
132 if (s
->raiddisks
< 4 && s
->level
== 6) {
133 pr_err("at least 4 raid-devices needed for level 6\n");
136 if (s
->raiddisks
> 256 && s
->level
== 6) {
137 pr_err("no more than 256 raid-devices supported for level 6\n");
140 if (s
->raiddisks
< 2 && s
->level
>= 4) {
141 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
144 if (s
->level
<= 0 && s
->sparedisks
) {
145 pr_err("This level does not support spare devices\n");
149 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
150 /* If given a single device, it might be a container, and we can
151 * extract a device list from there
155 memset(&inf
, 0, sizeof(inf
));
156 fd
= open(devlist
->devname
, O_RDONLY
);
158 md_get_array_info(fd
, &inf
) == 0 && inf
.raid_disks
== 0) {
159 /* yep, looks like a container */
161 rv
= st
->ss
->load_container(st
, fd
,
166 st
= super_by_fd(fd
, NULL
);
167 if (st
&& !(rv
= st
->ss
->
168 load_container(st
, fd
,
174 if (have_container
) {
175 subdevs
= s
->raiddisks
;
176 first_missing
= subdevs
* 2;
177 second_missing
= subdevs
* 2;
178 insert_point
= subdevs
* 2;
184 if (st
&& st
->ss
->external
&& s
->sparedisks
) {
185 pr_err("This metadata type does not support spare disks at create time\n");
188 if (subdevs
> s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
189 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, s
->raiddisks
+s
->sparedisks
);
192 if (!have_container
&& subdevs
< s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
193 pr_err("You haven't given enough devices (real or missing) to create this array\n");
196 if (s
->bitmap_file
&& s
->level
<= 0) {
197 pr_err("bitmaps not meaningful with level %s\n",
198 map_num(pers
, s
->level
)?:"given");
202 /* now set some defaults */
204 if (s
->layout
== UnSet
) {
205 do_default_layout
= 1;
206 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
210 /* check layout fits in array*/
211 if ((s
->layout
&255) * ((s
->layout
>>8)&255) > s
->raiddisks
) {
212 pr_err("that layout requires at least %d devices\n",
213 (s
->layout
&255) * ((s
->layout
>>8)&255));
223 if (s
->chunk
== 0 || s
->chunk
== UnSet
) {
225 do_default_chunk
= 1;
226 /* chunk will be set later */
230 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
231 if (get_linux_version() < 2006016 && s
->chunk
== 0) {
234 pr_err("chunk size defaults to 64K\n");
239 case LEVEL_MULTIPATH
:
240 case LEVEL_CONTAINER
:
244 pr_err("chunk size ignored for this level\n");
248 pr_err("unknown level %d\n", s
->level
);
251 if (s
->size
== MAX_SIZE
)
252 /* use '0' to mean 'max' now... */
254 if (s
->size
&& s
->chunk
&& s
->chunk
!= UnSet
)
255 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
256 newsize
= s
->size
* 2;
257 if (st
&& ! st
->ss
->validate_geometry(st
, s
->level
, s
->layout
, s
->raiddisks
,
258 &s
->chunk
, s
->size
*2,
260 &newsize
, s
->consistency_policy
,
264 if (s
->chunk
&& s
->chunk
!= UnSet
) {
265 newsize
&= ~(unsigned long long)(s
->chunk
*2 - 1);
266 if (do_default_chunk
) {
267 /* default chunk was just set */
269 pr_err("chunk size defaults to %dK\n", s
->chunk
);
270 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
271 do_default_chunk
= 0;
276 s
->size
= newsize
/ 2;
278 /* If this is ever reshaped to RAID5, we will
279 * need a chunksize. So round it off a bit
280 * now just to be safe
282 s
->size
&= ~(64ULL-1);
284 if (s
->size
&& c
->verbose
> 0)
285 pr_err("setting size to %lluK\n", s
->size
);
288 /* now look at the subdevs */
289 info
.array
.active_disks
= 0;
290 info
.array
.working_disks
= 0;
292 for (dv
= devlist
; dv
; dv
= dv
->next
)
293 if (data_offset
== VARIABLE_OFFSET
)
294 dv
->data_offset
= INVALID_SECTORS
;
296 dv
->data_offset
= data_offset
;
298 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
299 char *dname
= dv
->devname
;
300 unsigned long long freesize
;
304 if (strcasecmp(dname
, "missing") == 0) {
305 if (first_missing
> dnum
)
306 first_missing
= dnum
;
307 if (second_missing
> dnum
&& dnum
> first_missing
)
308 second_missing
= dnum
;
312 if (data_offset
== VARIABLE_OFFSET
) {
313 doff
= strchr(dname
, ':');
316 dv
->data_offset
= parse_size(doff
);
318 dv
->data_offset
= INVALID_SECTORS
;
320 dv
->data_offset
= data_offset
;
322 dfd
= open(dname
, O_RDONLY
);
324 pr_err("cannot open %s: %s\n",
325 dname
, strerror(errno
));
328 if (!fstat_is_blkdev(dfd
, dname
, NULL
)) {
333 info
.array
.working_disks
++;
334 if (dnum
< s
->raiddisks
&& dv
->disposition
!= 'j')
335 info
.array
.active_disks
++;
337 struct createinfo
*ci
= conf_get_create_info();
342 /* Need to choose a default metadata, which is different
343 * depending on geometry of array.
346 char *name
= "default";
347 for(i
= 0; !st
&& superlist
[i
]; i
++) {
348 st
= superlist
[i
]->match_metadata_desc(name
);
351 if (do_default_layout
)
352 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
353 switch (st
->ss
->validate_geometry(
354 st
, s
->level
, s
->layout
, s
->raiddisks
,
355 &s
->chunk
, s
->size
*2,
356 dv
->data_offset
, dname
,
357 &freesize
, s
->consistency_policy
,
359 case -1: /* Not valid, message printed, and not
360 * worth checking any further */
363 case 0: /* Geometry not valid */
366 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
368 case 1: /* All happy */
374 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
376 pr_err("cannot open %s: %s\n",
377 dname
, strerror(errno
));
380 pr_err("device %s not suitable for any style of array\n",
384 if (st
->ss
!= &super0
||
385 st
->minor_version
!= 90)
388 if (do_default_layout
)
389 s
->layout
= default_layout(st
, s
->level
, 0);
390 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
392 &s
->chunk
, s
->size
*2,
395 s
->consistency_policy
,
398 pr_err("%s is not suitable for this array.\n",
405 if (dv
->disposition
== 'j')
406 goto skip_size_check
; /* skip write journal for size check */
408 freesize
/= 2; /* convert to K */
409 if (s
->chunk
&& s
->chunk
!= UnSet
) {
410 /* round to chunk size */
411 freesize
= freesize
& ~(s
->chunk
-1);
412 if (do_default_chunk
) {
413 /* default chunk was just set */
415 pr_err("chunk size defaults to %dK\n", s
->chunk
);
416 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
417 do_default_chunk
= 0;
421 pr_err("no free space left on %s\n", dname
);
426 if (s
->size
&& freesize
< s
->size
) {
427 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
428 dname
, freesize
, s
->size
);
432 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
436 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
441 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
442 int fd
= open(dname
, O_RDONLY
);
444 pr_err("Cannot open %s: %s\n",
445 dname
, strerror(errno
));
449 warn
|= check_ext2(fd
, dname
);
450 warn
|= check_reiser(fd
, dname
);
451 warn
|= check_raid(fd
, dname
);
452 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
453 st
->minor_version
>= 1)
454 /* metadata at front */
455 warn
|= check_partitions(fd
, dname
, 0, 0);
456 else if (s
->level
== 1 || s
->level
== LEVEL_CONTAINER
||
457 (s
->level
== 0 && s
->raiddisks
== 1))
458 /* partitions could be meaningful */
459 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
461 /* partitions cannot be meaningful */
462 warn
|= check_partitions(fd
, dname
, 0, 0);
463 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
464 st
->minor_version
>= 1 &&
467 (warn
& 1024) == 0) {
469 pr_err("Note: this array has metadata at the start and\n"
470 " may not be suitable as a boot device. If you plan to\n"
471 " store '/boot' on this device please ensure that\n"
472 " your boot-loader understands md/v1.x metadata, or use\n"
473 " --metadata=0.90\n");
478 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
479 pr_err("Too many devices: %s metadata only supports %d\n",
480 st
->ss
->name
, st
->max_devs
);
484 info
.array
.working_disks
= s
->raiddisks
;
486 pr_err("create aborted\n");
490 if (mindisc
== NULL
&& !have_container
) {
491 pr_err("no size and no drives given - aborting create.\n");
494 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
||
495 s
->level
== LEVEL_FAULTY
|| st
->ss
->external
) {
496 /* size is meaningful */
497 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
499 &s
->chunk
, minsize
*2,
502 s
->consistency_policy
, 0)) {
503 pr_err("devices too large for RAID level %d\n", s
->level
);
508 /* If this is ever reshaped to RAID5, we will
509 * need a chunksize. So round it off a bit
510 * now just to be safe
512 s
->size
&= ~(64ULL-1);
514 pr_err("size set to %lluK\n", s
->size
);
518 if (!s
->bitmap_file
&&
520 st
->ss
->add_internal_bitmap
&&
521 (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
522 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) &&
523 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
525 pr_err("automatically enabling write-intent bitmap on large array\n");
526 s
->bitmap_file
= "internal";
528 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "none") == 0)
529 s
->bitmap_file
= NULL
;
531 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
532 !st
->ss
->write_init_ppl
) {
533 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
537 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
538 if (c
->runstop
!= 1 || c
->verbose
>= 0)
539 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
544 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
545 if (c
->runstop
!= 1 || c
->verbose
>= 0)
546 pr_err("%s unable to enumerate platform support\n"
547 " array may not be compatible with hardware/firmware\n",
551 st
->nodes
= c
->nodes
;
552 st
->cluster_name
= c
->homecluster
;
555 if (c
->runstop
!= 1) {
556 if (!ask("Continue creating array? ")) {
557 pr_err("create aborted.\n");
562 pr_err("creation continuing despite oddities due to --run\n");
566 /* If this is raid4/5, we want to configure the last active slot
567 * as missing, so that a reconstruct happens (faster than re-parity)
568 * FIX: Can we do this for raid6 as well?
570 if (st
->ss
->external
== 0 && s
->assume_clean
== 0 &&
571 c
->force
== 0 && first_missing
>= s
->raiddisks
) {
575 insert_point
= s
->raiddisks
-1;
577 info
.array
.active_disks
--;
584 /* For raid6, if creating with 1 missing drive, make a good drive
585 * into a spare, else the create will fail
587 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
588 st
->ss
->external
== 0 &&
589 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
590 insert_point
= s
->raiddisks
- 1;
591 if (insert_point
== first_missing
)
594 info
.array
.active_disks
--;
598 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
599 pr_err("This level does not support missing devices\n");
603 /* We need to create the device */
605 mdfd
= create_mddev(mddev
, name
, c
->autof
, LOCAL
, chosen_name
, 1);
610 /* verify if chosen_name is not in use,
611 * it could be in conflict with already existing device
612 * e.g. container, array
614 if (strncmp(chosen_name
, "/dev/md/", 8) == 0 &&
615 map_by_name(&map
, chosen_name
+8) != NULL
) {
616 pr_err("Array name %s is in use already.\n",
625 memset(&inf
, 0, sizeof(inf
));
626 md_get_array_info(mdfd
, &inf
);
627 if (inf
.working_disks
!= 0) {
628 pr_err("another array by this name is already running.\n");
632 /* Ok, lets try some ioctls */
634 info
.array
.level
= s
->level
;
635 info
.array
.size
= s
->size
;
636 info
.array
.raid_disks
= s
->raiddisks
;
637 /* The kernel should *know* what md_minor we are dealing
638 * with, but it chooses to trust me instead. Sigh
640 info
.array
.md_minor
= 0;
641 if (fstat_is_blkdev(mdfd
, mddev
, &rdev
))
642 info
.array
.md_minor
= minor(rdev
);
643 info
.array
.not_persistent
= 0;
645 if (((s
->level
== 4 || s
->level
== 5) &&
646 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
)) ||
647 (s
->level
== 6 && (insert_point
< s
->raiddisks
||
648 second_missing
< s
->raiddisks
)) ||
649 (s
->level
<= 0) || s
->assume_clean
) {
650 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
651 info
.resync_start
= MaxSector
;
653 info
.array
.state
= 0; /* not clean, but no errors */
654 info
.resync_start
= 0;
656 if (s
->level
== 10) {
657 /* for raid10, the bitmap size is the capacity of the array,
658 * which is array.size * raid_disks / ncopies;
659 * .. but convert to sectors.
661 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
662 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
663 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
665 bitmapsize
= s
->size
* 2;
667 /* There is lots of redundancy in these disk counts,
668 * raid_disks is the most meaningful value
669 * it describes the geometry of the array
671 * nr_disks is total number of used slots.
672 * it should be raid_disks+spare_disks
673 * spare_disks is the number of extra disks present
675 * active_disks is the number of working disks in
676 * active slots. (With raid_disks)
677 * working_disks is the total number of working disks,
679 * failed_disks is the number of disks marked failed
681 * Ideally, the kernel would keep these (except raid_disks)
682 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
683 * So for now, we assume that all raid and spare
684 * devices will be given.
686 info
.array
.spare_disks
=s
->sparedisks
;
687 info
.array
.failed_disks
=missing_disks
;
688 info
.array
.nr_disks
= info
.array
.working_disks
689 + info
.array
.failed_disks
;
690 info
.array
.layout
= s
->layout
;
691 info
.array
.chunk_size
= s
->chunk
*1024;
693 if (name
== NULL
|| *name
== 0) {
694 /* base name on mddev */
700 * /dev/md/home -> home
701 * /dev/mdhome -> home
703 /* FIXME compare this with rules in create_mddev */
704 name
= strrchr(mddev
, '/');
707 if (strncmp(name
, "md_", 3) == 0 &&
708 strlen(name
) > 3 && (name
-mddev
) == 5 /* /dev/ */)
710 else if (strncmp(name
, "md", 2) == 0 &&
711 strlen(name
) > 2 && isdigit(name
[2]) &&
712 (name
-mddev
) == 5 /* /dev/ */)
716 if (!st
->ss
->init_super(st
, &info
.array
, s
, name
, c
->homehost
, uuid
,
720 total_slots
= info
.array
.nr_disks
;
721 st
->ss
->getinfo_super(st
, &info
, NULL
);
722 if (sysfs_init(&info
, mdfd
, NULL
)) {
723 pr_err("unable to initialize sysfs\n");
727 if (did_default
&& c
->verbose
>= 0) {
728 if (is_subarray(info
.text_version
)) {
733 strncpy(devnm
, info
.text_version
+1, 32);
735 ep
= strchr(devnm
, '/');
739 mdi
= sysfs_read(-1, devnm
, GET_VERSION
);
741 pr_err("Creating array inside %s container %s\n",
742 mdi
?mdi
->text_version
:"managed", devnm
);
745 pr_err("Defaulting to version %s metadata\n", info
.text_version
);
748 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
749 info
.uuid
, chosen_name
);
750 /* Keep map locked until devices have been added to array
751 * to stop another mdadm from finding and using those devices.
754 if (s
->bitmap_file
&& (strcmp(s
->bitmap_file
, "internal") == 0 ||
755 strcmp(s
->bitmap_file
, "clustered") == 0)) {
756 if (!st
->ss
->add_internal_bitmap
) {
757 pr_err("internal bitmaps not supported with %s metadata\n",
761 if (st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
762 c
->delay
, s
->write_behind
,
763 bitmapsize
, 1, major_num
)) {
764 pr_err("Given bitmap chunk size not supported.\n");
767 s
->bitmap_file
= NULL
;
770 if (sysfs_init(&info
, mdfd
, NULL
)) {
771 pr_err("unable to initialize sysfs\n");
775 if (st
->ss
->external
&& st
->container_devnm
[0]) {
778 /* When creating a member, we need to be careful
779 * to negotiate with mdmon properly.
780 * If it is already running, we cannot write to
781 * the devices and must ask it to do that part.
782 * If it isn't running, we write to the devices,
784 * We hold an exclusive open on the container
785 * device to make sure mdmon doesn't exit after
786 * we checked that it is running.
788 * For now, fail if it is already running.
790 container_fd
= open_dev_excl(st
->container_devnm
);
791 if (container_fd
< 0) {
792 pr_err("Cannot get exclusive open on container - weird.\n");
795 if (mdmon_running(st
->container_devnm
)) {
797 pr_err("reusing mdmon for %s.\n",
798 st
->container_devnm
);
799 st
->update_tail
= &st
->updates
;
803 rv
= set_array_info(mdfd
, st
, &info
);
805 pr_err("failed to set array info for %s: %s\n",
806 mddev
, strerror(errno
));
810 if (s
->bitmap_file
) {
813 st
->ss
->uuid_from_super(st
, uuid
);
814 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
815 c
->delay
, s
->write_behind
,
820 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
822 pr_err("weird: %s cannot be openned\n",
826 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
827 pr_err("Cannot set bitmap file for %s: %s\n",
828 mddev
, strerror(errno
));
833 infos
= xmalloc(sizeof(*infos
) * total_slots
);
834 enable_fds(total_slots
);
835 for (pass
= 1; pass
<= 2; pass
++) {
836 struct mddev_dev
*moved_disk
= NULL
; /* the disk that was moved out of the insert point */
838 for (dnum
= 0, raid_disk_num
= 0, dv
= devlist
; dv
;
839 dv
= (dv
->next
) ? (dv
->next
) : moved_disk
, dnum
++) {
841 struct mdinfo
*inf
= &infos
[dnum
];
843 if (dnum
>= total_slots
)
845 if (dnum
== insert_point
) {
850 if (strcasecmp(dv
->devname
, "missing") == 0) {
856 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
857 /* repeatedly use the container */
864 inf
->disk
.number
= dnum
;
865 inf
->disk
.raid_disk
= raid_disk_num
++;
867 if (dv
->disposition
== 'j') {
868 inf
->disk
.raid_disk
= MD_DISK_ROLE_JOURNAL
;
869 inf
->disk
.state
= (1<<MD_DISK_JOURNAL
);
871 } else if (inf
->disk
.raid_disk
< s
->raiddisks
)
872 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
877 if (dv
->writemostly
== FlagSet
)
878 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
879 if (dv
->failfast
== FlagSet
)
880 inf
->disk
.state
|= (1<<MD_DISK_FAILFAST
);
885 if (st
->ss
->external
&&
886 st
->container_devnm
[0])
887 fd
= open(dv
->devname
, O_RDWR
);
889 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
892 pr_err("failed to open %s after earlier success - aborting\n",
896 if (!fstat_is_blkdev(fd
, dv
->devname
, &rdev
))
898 inf
->disk
.major
= major(rdev
);
899 inf
->disk
.minor
= minor(rdev
);
902 remove_partitions(fd
);
903 if (st
->ss
->add_to_super(st
, &inf
->disk
,
906 ioctl(mdfd
, STOP_ARRAY
, NULL
);
909 st
->ss
->getinfo_super(st
, inf
, NULL
);
910 safe_mode_delay
= inf
->safe_mode_delay
;
912 if (have_container
&& c
->verbose
> 0)
913 pr_err("Using %s for device %d\n",
914 map_dev(inf
->disk
.major
,
918 if (!have_container
) {
919 /* getinfo_super might have lost these ... */
920 inf
->disk
.major
= major(rdev
);
921 inf
->disk
.minor
= minor(rdev
);
927 rv
= add_disk(mdfd
, st
, &info
, inf
);
930 pr_err("ADD_NEW_DISK for %s failed: %s\n",
931 dv
->devname
, strerror(errno
));
936 if (!have_container
&&
937 dv
== moved_disk
&& dnum
!= insert_point
) break;
940 struct mdinfo info_new
;
941 struct map_ent
*me
= NULL
;
943 /* check to see if the uuid has changed due to these
944 * metadata changes, and if so update the member array
945 * and container uuid. Note ->write_init_super clears
946 * the subarray cursor such that ->getinfo_super once
947 * again returns container info.
949 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
950 if (st
->ss
->external
&& s
->level
!= LEVEL_CONTAINER
&&
951 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
952 map_update(&map
, fd2devnm(mdfd
),
953 info_new
.text_version
,
954 info_new
.uuid
, chosen_name
);
955 me
= map_by_devnm(&map
, st
->container_devnm
);
958 if (st
->ss
->write_init_super(st
)) {
959 st
->ss
->free_super(st
);
963 /* update parent container uuid */
965 char *path
= xstrdup(me
->path
);
967 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
968 map_update(&map
, st
->container_devnm
,
969 info_new
.text_version
,
970 info_new
.uuid
, path
);
974 flush_metadata_updates(st
);
975 st
->ss
->free_super(st
);
981 if (s
->level
== LEVEL_CONTAINER
) {
982 /* No need to start. But we should signal udev to
984 sysfs_uevent(&info
, "change");
986 pr_err("container %s prepared.\n", mddev
);
987 wait_for(chosen_name
, mdfd
);
988 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
989 if (st
->ss
->external
) {
993 case LEVEL_MULTIPATH
:
995 err
= sysfs_set_str(&info
, NULL
, "array_state",
1002 err
= sysfs_set_str(&info
, NULL
, "array_state",
1006 sysfs_set_safemode(&info
, safe_mode_delay
);
1008 pr_err("failed to activate array.\n");
1009 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1012 } else if (c
->readonly
&&
1013 sysfs_attribute_available(
1014 &info
, NULL
, "array_state")) {
1015 if (sysfs_set_str(&info
, NULL
,
1016 "array_state", "readonly") < 0) {
1017 pr_err("Failed to start array: %s\n",
1019 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1023 /* param is not actually used */
1025 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1026 pr_err("RUN_ARRAY failed: %s\n",
1028 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1029 cont_err("Problem may be that chunk size is not a power of 2\n");
1031 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1034 /* if start_ro module parameter is set, array is
1035 * auto-read-only, which is bad as the resync won't
1036 * start. So lets make it read-write now.
1038 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1040 if (c
->verbose
>= 0)
1041 pr_err("array %s started.\n", mddev
);
1042 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1044 start_mdmon(st
->container_devnm
);
1046 ping_monitor(st
->container_devnm
);
1047 close(container_fd
);
1049 wait_for(chosen_name
, mdfd
);
1051 pr_err("not starting array - not enough devices.\n");
1054 /* Give udev a moment to process the Change event caused
1065 map_remove(&map
, fd2devnm(mdfd
));