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
;
92 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(dfd
, &stb
) != 0 ||
329 (stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
331 pr_err("%s is not a block device\n",
336 info
.array
.working_disks
++;
337 if (dnum
< s
->raiddisks
&& dv
->disposition
!= 'j')
338 info
.array
.active_disks
++;
340 struct createinfo
*ci
= conf_get_create_info();
345 /* Need to choose a default metadata, which is different
346 * depending on geometry of array.
349 char *name
= "default";
350 for(i
= 0; !st
&& superlist
[i
]; i
++) {
351 st
= superlist
[i
]->match_metadata_desc(name
);
354 if (do_default_layout
)
355 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
356 switch (st
->ss
->validate_geometry(
357 st
, s
->level
, s
->layout
, s
->raiddisks
,
358 &s
->chunk
, s
->size
*2,
359 dv
->data_offset
, dname
,
360 &freesize
, s
->consistency_policy
,
362 case -1: /* Not valid, message printed, and not
363 * worth checking any further */
366 case 0: /* Geometry not valid */
369 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
371 case 1: /* All happy */
377 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
379 pr_err("cannot open %s: %s\n",
380 dname
, strerror(errno
));
383 pr_err("device %s not suitable for any style of array\n",
387 if (st
->ss
!= &super0
||
388 st
->minor_version
!= 90)
391 if (do_default_layout
)
392 s
->layout
= default_layout(st
, s
->level
, 0);
393 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
395 &s
->chunk
, s
->size
*2,
398 s
->consistency_policy
,
401 pr_err("%s is not suitable for this array.\n",
408 if (dv
->disposition
== 'j')
409 goto skip_size_check
; /* skip write journal for size check */
411 freesize
/= 2; /* convert to K */
412 if (s
->chunk
&& s
->chunk
!= UnSet
) {
413 /* round to chunk size */
414 freesize
= freesize
& ~(s
->chunk
-1);
415 if (do_default_chunk
) {
416 /* default chunk was just set */
418 pr_err("chunk size defaults to %dK\n", s
->chunk
);
419 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
420 do_default_chunk
= 0;
424 pr_err("no free space left on %s\n", dname
);
429 if (s
->size
&& freesize
< s
->size
) {
430 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
431 dname
, freesize
, s
->size
);
435 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
439 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
444 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
445 int fd
= open(dname
, O_RDONLY
);
447 pr_err("Cannot open %s: %s\n",
448 dname
, strerror(errno
));
452 warn
|= check_ext2(fd
, dname
);
453 warn
|= check_reiser(fd
, dname
);
454 warn
|= check_raid(fd
, dname
);
455 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
456 st
->minor_version
>= 1)
457 /* metadata at front */
458 warn
|= check_partitions(fd
, dname
, 0, 0);
459 else if (s
->level
== 1 || s
->level
== LEVEL_CONTAINER
||
460 (s
->level
== 0 && s
->raiddisks
== 1))
461 /* partitions could be meaningful */
462 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
464 /* partitions cannot be meaningful */
465 warn
|= check_partitions(fd
, dname
, 0, 0);
466 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
467 st
->minor_version
>= 1 &&
470 (warn
& 1024) == 0) {
472 pr_err("Note: this array has metadata at the start and\n"
473 " may not be suitable as a boot device. If you plan to\n"
474 " store '/boot' on this device please ensure that\n"
475 " your boot-loader understands md/v1.x metadata, or use\n"
476 " --metadata=0.90\n");
481 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
482 pr_err("Too many devices: %s metadata only supports %d\n",
483 st
->ss
->name
, st
->max_devs
);
487 info
.array
.working_disks
= s
->raiddisks
;
489 pr_err("create aborted\n");
493 if (mindisc
== NULL
&& !have_container
) {
494 pr_err("no size and no drives given - aborting create.\n");
497 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
||
498 s
->level
== LEVEL_FAULTY
|| st
->ss
->external
) {
499 /* size is meaningful */
500 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
502 &s
->chunk
, minsize
*2,
505 s
->consistency_policy
, 0)) {
506 pr_err("devices too large for RAID level %d\n", s
->level
);
511 /* If this is ever reshaped to RAID5, we will
512 * need a chunksize. So round it off a bit
513 * now just to be safe
515 s
->size
&= ~(64ULL-1);
517 pr_err("size set to %lluK\n", s
->size
);
521 if (!s
->bitmap_file
&&
523 st
->ss
->add_internal_bitmap
&&
524 (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
525 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) &&
526 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
528 pr_err("automatically enabling write-intent bitmap on large array\n");
529 s
->bitmap_file
= "internal";
531 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "none") == 0)
532 s
->bitmap_file
= NULL
;
534 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
535 !st
->ss
->write_init_ppl
) {
536 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
540 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
541 if (c
->runstop
!= 1 || c
->verbose
>= 0)
542 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
547 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
548 if (c
->runstop
!= 1 || c
->verbose
>= 0)
549 pr_err("%s unable to enumerate platform support\n"
550 " array may not be compatible with hardware/firmware\n",
554 st
->nodes
= c
->nodes
;
555 st
->cluster_name
= c
->homecluster
;
558 if (c
->runstop
!= 1) {
559 if (!ask("Continue creating array? ")) {
560 pr_err("create aborted.\n");
565 pr_err("creation continuing despite oddities due to --run\n");
569 /* If this is raid4/5, we want to configure the last active slot
570 * as missing, so that a reconstruct happens (faster than re-parity)
571 * FIX: Can we do this for raid6 as well?
573 if (st
->ss
->external
== 0 && s
->assume_clean
== 0 &&
574 c
->force
== 0 && first_missing
>= s
->raiddisks
) {
578 insert_point
= s
->raiddisks
-1;
580 info
.array
.active_disks
--;
587 /* For raid6, if creating with 1 missing drive, make a good drive
588 * into a spare, else the create will fail
590 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
591 st
->ss
->external
== 0 &&
592 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
593 insert_point
= s
->raiddisks
- 1;
594 if (insert_point
== first_missing
)
597 info
.array
.active_disks
--;
601 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
602 pr_err("This level does not support missing devices\n");
606 /* We need to create the device */
608 mdfd
= create_mddev(mddev
, name
, c
->autof
, LOCAL
, chosen_name
, 1);
613 /* verify if chosen_name is not in use,
614 * it could be in conflict with already existing device
615 * e.g. container, array
617 if (strncmp(chosen_name
, "/dev/md/", 8) == 0 &&
618 map_by_name(&map
, chosen_name
+8) != NULL
) {
619 pr_err("Array name %s is in use already.\n",
628 memset(&inf
, 0, sizeof(inf
));
629 md_get_array_info(mdfd
, &inf
);
630 if (inf
.working_disks
!= 0) {
631 pr_err("another array by this name is already running.\n");
635 /* Ok, lets try some ioctls */
637 info
.array
.level
= s
->level
;
638 info
.array
.size
= s
->size
;
639 info
.array
.raid_disks
= s
->raiddisks
;
640 /* The kernel should *know* what md_minor we are dealing
641 * with, but it chooses to trust me instead. Sigh
643 info
.array
.md_minor
= 0;
644 if (fstat(mdfd
, &stb
) == 0)
645 info
.array
.md_minor
= minor(stb
.st_rdev
);
646 info
.array
.not_persistent
= 0;
648 if (((s
->level
== 4 || s
->level
== 5) &&
649 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
)) ||
650 (s
->level
== 6 && (insert_point
< s
->raiddisks
||
651 second_missing
< s
->raiddisks
)) ||
652 (s
->level
<= 0) || s
->assume_clean
) {
653 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
654 info
.resync_start
= MaxSector
;
656 info
.array
.state
= 0; /* not clean, but no errors */
657 info
.resync_start
= 0;
659 if (s
->level
== 10) {
660 /* for raid10, the bitmap size is the capacity of the array,
661 * which is array.size * raid_disks / ncopies;
662 * .. but convert to sectors.
664 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
665 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
666 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
668 bitmapsize
= s
->size
* 2;
670 /* There is lots of redundancy in these disk counts,
671 * raid_disks is the most meaningful value
672 * it describes the geometry of the array
674 * nr_disks is total number of used slots.
675 * it should be raid_disks+spare_disks
676 * spare_disks is the number of extra disks present
678 * active_disks is the number of working disks in
679 * active slots. (With raid_disks)
680 * working_disks is the total number of working disks,
682 * failed_disks is the number of disks marked failed
684 * Ideally, the kernel would keep these (except raid_disks)
685 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
686 * So for now, we assume that all raid and spare
687 * devices will be given.
689 info
.array
.spare_disks
=s
->sparedisks
;
690 info
.array
.failed_disks
=missing_disks
;
691 info
.array
.nr_disks
= info
.array
.working_disks
692 + info
.array
.failed_disks
;
693 info
.array
.layout
= s
->layout
;
694 info
.array
.chunk_size
= s
->chunk
*1024;
696 if (name
== NULL
|| *name
== 0) {
697 /* base name on mddev */
703 * /dev/md/home -> home
704 * /dev/mdhome -> home
706 /* FIXME compare this with rules in create_mddev */
707 name
= strrchr(mddev
, '/');
710 if (strncmp(name
, "md_", 3) == 0 &&
711 strlen(name
) > 3 && (name
-mddev
) == 5 /* /dev/ */)
713 else if (strncmp(name
, "md", 2) == 0 &&
714 strlen(name
) > 2 && isdigit(name
[2]) &&
715 (name
-mddev
) == 5 /* /dev/ */)
719 if (!st
->ss
->init_super(st
, &info
.array
, s
, name
, c
->homehost
, uuid
,
723 total_slots
= info
.array
.nr_disks
;
724 st
->ss
->getinfo_super(st
, &info
, NULL
);
725 if (sysfs_init(&info
, mdfd
, NULL
)) {
726 pr_err("unable to initialize sysfs\n");
730 if (did_default
&& c
->verbose
>= 0) {
731 if (is_subarray(info
.text_version
)) {
736 strncpy(devnm
, info
.text_version
+1, 32);
738 ep
= strchr(devnm
, '/');
742 mdi
= sysfs_read(-1, devnm
, GET_VERSION
);
744 pr_err("Creating array inside %s container %s\n",
745 mdi
?mdi
->text_version
:"managed", devnm
);
748 pr_err("Defaulting to version %s metadata\n", info
.text_version
);
751 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
752 info
.uuid
, chosen_name
);
753 /* Keep map locked until devices have been added to array
754 * to stop another mdadm from finding and using those devices.
757 if (s
->bitmap_file
&& (strcmp(s
->bitmap_file
, "internal") == 0 ||
758 strcmp(s
->bitmap_file
, "clustered") == 0)) {
759 if (!st
->ss
->add_internal_bitmap
) {
760 pr_err("internal bitmaps not supported with %s metadata\n",
764 if (st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
765 c
->delay
, s
->write_behind
,
766 bitmapsize
, 1, major_num
)) {
767 pr_err("Given bitmap chunk size not supported.\n");
770 s
->bitmap_file
= NULL
;
773 if (sysfs_init(&info
, mdfd
, NULL
)) {
774 pr_err("unable to initialize sysfs\n");
778 if (st
->ss
->external
&& st
->container_devnm
[0]) {
781 /* When creating a member, we need to be careful
782 * to negotiate with mdmon properly.
783 * If it is already running, we cannot write to
784 * the devices and must ask it to do that part.
785 * If it isn't running, we write to the devices,
787 * We hold an exclusive open on the container
788 * device to make sure mdmon doesn't exit after
789 * we checked that it is running.
791 * For now, fail if it is already running.
793 container_fd
= open_dev_excl(st
->container_devnm
);
794 if (container_fd
< 0) {
795 pr_err("Cannot get exclusive open on container - weird.\n");
798 if (mdmon_running(st
->container_devnm
)) {
800 pr_err("reusing mdmon for %s.\n",
801 st
->container_devnm
);
802 st
->update_tail
= &st
->updates
;
806 rv
= set_array_info(mdfd
, st
, &info
);
808 pr_err("failed to set array info for %s: %s\n",
809 mddev
, strerror(errno
));
813 if (s
->bitmap_file
) {
816 st
->ss
->uuid_from_super(st
, uuid
);
817 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
818 c
->delay
, s
->write_behind
,
823 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
825 pr_err("weird: %s cannot be openned\n",
829 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
830 pr_err("Cannot set bitmap file for %s: %s\n",
831 mddev
, strerror(errno
));
836 infos
= xmalloc(sizeof(*infos
) * total_slots
);
837 enable_fds(total_slots
);
838 for (pass
= 1; pass
<= 2; pass
++) {
839 struct mddev_dev
*moved_disk
= NULL
; /* the disk that was moved out of the insert point */
841 for (dnum
= 0, raid_disk_num
= 0, dv
= devlist
; dv
;
842 dv
= (dv
->next
) ? (dv
->next
) : moved_disk
, dnum
++) {
845 struct mdinfo
*inf
= &infos
[dnum
];
847 if (dnum
>= total_slots
)
849 if (dnum
== insert_point
) {
854 if (strcasecmp(dv
->devname
, "missing") == 0) {
860 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
861 /* repeatedly use the container */
868 inf
->disk
.number
= dnum
;
869 inf
->disk
.raid_disk
= raid_disk_num
++;
871 if (dv
->disposition
== 'j') {
872 inf
->disk
.raid_disk
= MD_DISK_ROLE_JOURNAL
;
873 inf
->disk
.state
= (1<<MD_DISK_JOURNAL
);
875 } else if (inf
->disk
.raid_disk
< s
->raiddisks
)
876 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
881 if (dv
->writemostly
== FlagSet
)
882 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
883 if (dv
->failfast
== FlagSet
)
884 inf
->disk
.state
|= (1<<MD_DISK_FAILFAST
);
889 if (st
->ss
->external
&&
890 st
->container_devnm
[0])
891 fd
= open(dv
->devname
, O_RDWR
);
893 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
896 pr_err("failed to open %s after earlier success - aborting\n",
901 inf
->disk
.major
= major(stb2
.st_rdev
);
902 inf
->disk
.minor
= minor(stb2
.st_rdev
);
905 remove_partitions(fd
);
906 if (st
->ss
->add_to_super(st
, &inf
->disk
,
909 ioctl(mdfd
, STOP_ARRAY
, NULL
);
912 st
->ss
->getinfo_super(st
, inf
, NULL
);
913 safe_mode_delay
= inf
->safe_mode_delay
;
915 if (have_container
&& c
->verbose
> 0)
916 pr_err("Using %s for device %d\n",
917 map_dev(inf
->disk
.major
,
921 if (!have_container
) {
922 /* getinfo_super might have lost these ... */
923 inf
->disk
.major
= major(stb2
.st_rdev
);
924 inf
->disk
.minor
= minor(stb2
.st_rdev
);
930 rv
= add_disk(mdfd
, st
, &info
, inf
);
933 pr_err("ADD_NEW_DISK for %s failed: %s\n",
934 dv
->devname
, strerror(errno
));
939 if (!have_container
&&
940 dv
== moved_disk
&& dnum
!= insert_point
) break;
943 struct mdinfo info_new
;
944 struct map_ent
*me
= NULL
;
946 /* check to see if the uuid has changed due to these
947 * metadata changes, and if so update the member array
948 * and container uuid. Note ->write_init_super clears
949 * the subarray cursor such that ->getinfo_super once
950 * again returns container info.
952 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
953 if (st
->ss
->external
&& s
->level
!= LEVEL_CONTAINER
&&
954 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
955 map_update(&map
, fd2devnm(mdfd
),
956 info_new
.text_version
,
957 info_new
.uuid
, chosen_name
);
958 me
= map_by_devnm(&map
, st
->container_devnm
);
961 if (st
->ss
->write_init_super(st
)) {
962 st
->ss
->free_super(st
);
966 /* update parent container uuid */
968 char *path
= xstrdup(me
->path
);
970 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
971 map_update(&map
, st
->container_devnm
,
972 info_new
.text_version
,
973 info_new
.uuid
, path
);
977 flush_metadata_updates(st
);
978 st
->ss
->free_super(st
);
984 if (s
->level
== LEVEL_CONTAINER
) {
985 /* No need to start. But we should signal udev to
987 sysfs_uevent(&info
, "change");
989 pr_err("container %s prepared.\n", mddev
);
990 wait_for(chosen_name
, mdfd
);
991 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
992 if (st
->ss
->external
) {
996 case LEVEL_MULTIPATH
:
998 err
= sysfs_set_str(&info
, NULL
, "array_state",
1005 err
= sysfs_set_str(&info
, NULL
, "array_state",
1009 sysfs_set_safemode(&info
, safe_mode_delay
);
1011 pr_err("failed to activate array.\n");
1012 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1015 } else if (c
->readonly
&&
1016 sysfs_attribute_available(
1017 &info
, NULL
, "array_state")) {
1018 if (sysfs_set_str(&info
, NULL
,
1019 "array_state", "readonly") < 0) {
1020 pr_err("Failed to start array: %s\n",
1022 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1026 /* param is not actually used */
1028 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1029 pr_err("RUN_ARRAY failed: %s\n",
1031 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1032 cont_err("Problem may be that chunk size is not a power of 2\n");
1034 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1037 /* if start_ro module parameter is set, array is
1038 * auto-read-only, which is bad as the resync won't
1039 * start. So lets make it read-write now.
1041 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1043 if (c
->verbose
>= 0)
1044 pr_err("array %s started.\n", mddev
);
1045 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1047 start_mdmon(st
->container_devnm
);
1049 ping_monitor(st
->container_devnm
);
1050 close(container_fd
);
1052 wait_for(chosen_name
, mdfd
);
1054 pr_err("not starting array - not enough devices.\n");
1057 /* Give udev a moment to process the Change event caused
1068 map_remove(&map
, fd2devnm(mdfd
));