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 round_size_and_verify(unsigned long long *size
, int chunk
)
34 *size
&= ~(unsigned long long)(chunk
- 1);
36 pr_err("Size cannot be smaller than chunk.\n");
42 static int default_layout(struct supertype
*st
, int level
, int verbose
)
46 if (st
&& st
->ss
->default_geometry
)
47 st
->ss
->default_geometry(st
, &level
, &layout
, NULL
);
51 default: /* no layout */
55 layout
= 0x102; /* near=2, far=1 */
57 pr_err("layout defaults to n2\n");
61 layout
= map_name(r5layout
, "default");
63 pr_err("layout defaults to %s\n", map_num(r5layout
, layout
));
66 layout
= map_name(faultylayout
, "default");
69 pr_err("layout defaults to %s\n", map_num(faultylayout
, layout
));
76 int Create(struct supertype
*st
, char *mddev
,
77 char *name
, int *uuid
,
78 int subdevs
, struct mddev_dev
*devlist
,
80 struct context
*c
, unsigned long long data_offset
)
83 * Create a new raid array.
85 * First check that necessary details are available
86 * (i.e. level, raid-disks)
88 * Then check each disk to see what might be on it
89 * and report anything interesting.
91 * If anything looks odd, and runstop not set,
94 * SET_ARRAY_INFO and ADD_NEW_DISK, and
95 * if runstop==run, or raiddisks disks were used,
99 unsigned long long minsize
= 0, maxsize
= 0;
100 char *mindisc
= NULL
;
101 char *maxdisc
= NULL
;
102 int dnum
, raid_disk_num
;
103 struct mddev_dev
*dv
;
105 int fail
= 0, warn
= 0;
106 int first_missing
= subdevs
* 2;
107 int second_missing
= subdevs
* 2;
108 int missing_disks
= 0;
109 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
114 int have_container
= 0;
115 int container_fd
= -1;
117 unsigned long long bitmapsize
;
118 struct mdinfo info
, *infos
;
120 int do_default_layout
= 0;
121 int do_default_chunk
= 0;
122 unsigned long safe_mode_delay
= 0;
123 char chosen_name
[1024];
124 struct map_ent
*map
= NULL
;
125 unsigned long long newsize
;
126 mdu_array_info_t inf
;
128 int major_num
= BITMAP_MAJOR_HI
;
129 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "clustered") == 0) {
130 major_num
= BITMAP_MAJOR_CLUSTERED
;
132 pr_err("At least 2 nodes are needed for cluster-md\n");
137 memset(&info
, 0, sizeof(info
));
138 if (s
->level
== UnSet
&& st
&& st
->ss
->default_geometry
)
139 st
->ss
->default_geometry(st
, &s
->level
, NULL
, NULL
);
140 if (s
->level
== UnSet
) {
141 pr_err("a RAID level is needed to create an array.\n");
144 if (s
->raiddisks
< 4 && s
->level
== 6) {
145 pr_err("at least 4 raid-devices needed for level 6\n");
148 if (s
->raiddisks
> 256 && s
->level
== 6) {
149 pr_err("no more than 256 raid-devices supported for level 6\n");
152 if (s
->raiddisks
< 2 && s
->level
>= 4) {
153 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
156 if (s
->level
<= 0 && s
->sparedisks
) {
157 pr_err("This level does not support spare devices\n");
161 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
162 /* If given a single device, it might be a container, and we can
163 * extract a device list from there
167 memset(&inf
, 0, sizeof(inf
));
168 fd
= open(devlist
->devname
, O_RDONLY
);
170 md_get_array_info(fd
, &inf
) == 0 && inf
.raid_disks
== 0) {
171 /* yep, looks like a container */
173 rv
= st
->ss
->load_container(st
, fd
,
178 st
= super_by_fd(fd
, NULL
);
179 if (st
&& !(rv
= st
->ss
->
180 load_container(st
, fd
,
186 if (have_container
) {
187 subdevs
= s
->raiddisks
;
188 first_missing
= subdevs
* 2;
189 second_missing
= subdevs
* 2;
190 insert_point
= subdevs
* 2;
196 if (st
&& st
->ss
->external
&& s
->sparedisks
) {
197 pr_err("This metadata type does not support spare disks at create time\n");
200 if (subdevs
> s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
201 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, s
->raiddisks
+s
->sparedisks
);
204 if (!have_container
&& subdevs
< s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
205 pr_err("You haven't given enough devices (real or missing) to create this array\n");
208 if (s
->bitmap_file
&& s
->level
<= 0) {
209 pr_err("bitmaps not meaningful with level %s\n",
210 map_num(pers
, s
->level
)?:"given");
214 /* now set some defaults */
216 if (s
->layout
== UnSet
) {
217 do_default_layout
= 1;
218 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
222 /* check layout fits in array*/
223 if ((s
->layout
&255) * ((s
->layout
>>8)&255) > s
->raiddisks
) {
224 pr_err("that layout requires at least %d devices\n",
225 (s
->layout
&255) * ((s
->layout
>>8)&255));
235 if (s
->chunk
== 0 || s
->chunk
== UnSet
) {
237 do_default_chunk
= 1;
238 /* chunk will be set later */
242 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
243 if (get_linux_version() < 2006016 && s
->chunk
== 0) {
246 pr_err("chunk size defaults to 64K\n");
251 case LEVEL_MULTIPATH
:
252 case LEVEL_CONTAINER
:
256 pr_err("chunk size ignored for this level\n");
260 pr_err("unknown level %d\n", s
->level
);
264 if (s
->size
== MAX_SIZE
)
265 /* use '0' to mean 'max' now... */
267 if (s
->size
&& s
->chunk
&& s
->chunk
!= UnSet
)
268 if (round_size_and_verify(&s
->size
, s
->chunk
))
271 newsize
= s
->size
* 2;
272 if (st
&& ! st
->ss
->validate_geometry(st
, s
->level
, s
->layout
, s
->raiddisks
,
273 &s
->chunk
, s
->size
*2,
275 &newsize
, s
->consistency_policy
,
279 if (s
->chunk
&& s
->chunk
!= UnSet
) {
280 newsize
&= ~(unsigned long long)(s
->chunk
*2 - 1);
281 if (do_default_chunk
) {
282 /* default chunk was just set */
284 pr_err("chunk size defaults to %dK\n", s
->chunk
);
285 if (round_size_and_verify(&s
->size
, s
->chunk
))
287 do_default_chunk
= 0;
292 s
->size
= newsize
/ 2;
294 /* If this is ever reshaped to RAID5, we will
295 * need a chunksize. So round it off a bit
296 * now just to be safe
298 s
->size
&= ~(64ULL-1);
300 if (s
->size
&& c
->verbose
> 0)
301 pr_err("setting size to %lluK\n", s
->size
);
304 /* now look at the subdevs */
305 info
.array
.active_disks
= 0;
306 info
.array
.working_disks
= 0;
308 for (dv
= devlist
; dv
; dv
= dv
->next
)
309 if (data_offset
== VARIABLE_OFFSET
)
310 dv
->data_offset
= INVALID_SECTORS
;
312 dv
->data_offset
= data_offset
;
314 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
315 char *dname
= dv
->devname
;
316 unsigned long long freesize
;
320 if (strcasecmp(dname
, "missing") == 0) {
321 if (first_missing
> dnum
)
322 first_missing
= dnum
;
323 if (second_missing
> dnum
&& dnum
> first_missing
)
324 second_missing
= dnum
;
328 if (data_offset
== VARIABLE_OFFSET
) {
329 doff
= strchr(dname
, ':');
332 dv
->data_offset
= parse_size(doff
);
334 dv
->data_offset
= INVALID_SECTORS
;
336 dv
->data_offset
= data_offset
;
338 dfd
= open(dname
, O_RDONLY
);
340 pr_err("cannot open %s: %s\n",
341 dname
, strerror(errno
));
344 if (!fstat_is_blkdev(dfd
, dname
, NULL
)) {
349 info
.array
.working_disks
++;
350 if (dnum
< s
->raiddisks
&& dv
->disposition
!= 'j')
351 info
.array
.active_disks
++;
353 struct createinfo
*ci
= conf_get_create_info();
358 /* Need to choose a default metadata, which is different
359 * depending on geometry of array.
362 char *name
= "default";
363 for(i
= 0; !st
&& superlist
[i
]; i
++) {
364 st
= superlist
[i
]->match_metadata_desc(name
);
367 if (do_default_layout
)
368 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
369 switch (st
->ss
->validate_geometry(
370 st
, s
->level
, s
->layout
, s
->raiddisks
,
371 &s
->chunk
, s
->size
*2,
372 dv
->data_offset
, dname
,
373 &freesize
, s
->consistency_policy
,
375 case -1: /* Not valid, message printed, and not
376 * worth checking any further */
379 case 0: /* Geometry not valid */
382 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
384 case 1: /* All happy */
390 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
392 pr_err("cannot open %s: %s\n",
393 dname
, strerror(errno
));
396 pr_err("device %s not suitable for any style of array\n",
400 if (st
->ss
!= &super0
||
401 st
->minor_version
!= 90)
404 if (do_default_layout
)
405 s
->layout
= default_layout(st
, s
->level
, 0);
406 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
408 &s
->chunk
, s
->size
*2,
411 s
->consistency_policy
,
414 pr_err("%s is not suitable for this array.\n",
421 if (dv
->disposition
== 'j')
422 goto skip_size_check
; /* skip write journal for size check */
424 freesize
/= 2; /* convert to K */
425 if (s
->chunk
&& s
->chunk
!= UnSet
) {
426 /* round to chunk size */
427 freesize
= freesize
& ~(s
->chunk
-1);
428 if (do_default_chunk
) {
429 /* default chunk was just set */
431 pr_err("chunk size defaults to %dK\n", s
->chunk
);
432 if (round_size_and_verify(&s
->size
, s
->chunk
))
434 do_default_chunk
= 0;
438 pr_err("no free space left on %s\n", dname
);
443 if (s
->size
&& freesize
< s
->size
) {
444 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
445 dname
, freesize
, s
->size
);
449 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
453 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
458 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
459 int fd
= open(dname
, O_RDONLY
);
461 pr_err("Cannot open %s: %s\n",
462 dname
, strerror(errno
));
466 warn
|= check_ext2(fd
, dname
);
467 warn
|= check_reiser(fd
, dname
);
468 warn
|= check_raid(fd
, dname
);
469 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
470 st
->minor_version
>= 1)
471 /* metadata at front */
472 warn
|= check_partitions(fd
, dname
, 0, 0);
473 else if (s
->level
== 1 || s
->level
== LEVEL_CONTAINER
||
474 (s
->level
== 0 && s
->raiddisks
== 1))
475 /* partitions could be meaningful */
476 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
478 /* partitions cannot be meaningful */
479 warn
|= check_partitions(fd
, dname
, 0, 0);
480 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
481 st
->minor_version
>= 1 &&
484 (warn
& 1024) == 0) {
486 pr_err("Note: this array has metadata at the start and\n"
487 " may not be suitable as a boot device. If you plan to\n"
488 " store '/boot' on this device please ensure that\n"
489 " your boot-loader understands md/v1.x metadata, or use\n"
490 " --metadata=0.90\n");
495 if (missing_disks
== dnum
&& !have_container
) {
496 pr_err("Subdevs can't be all missing\n");
499 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
500 pr_err("Too many devices: %s metadata only supports %d\n",
501 st
->ss
->name
, st
->max_devs
);
505 info
.array
.working_disks
= s
->raiddisks
;
507 pr_err("create aborted\n");
511 if (mindisc
== NULL
&& !have_container
) {
512 pr_err("no size and no drives given - aborting create.\n");
515 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
||
516 s
->level
== LEVEL_FAULTY
|| st
->ss
->external
) {
517 /* size is meaningful */
518 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
520 &s
->chunk
, minsize
*2,
523 s
->consistency_policy
, 0)) {
524 pr_err("devices too large for RAID level %d\n", s
->level
);
529 /* If this is ever reshaped to RAID5, we will
530 * need a chunksize. So round it off a bit
531 * now just to be safe
533 s
->size
&= ~(64ULL-1);
535 pr_err("size set to %lluK\n", s
->size
);
539 if (!s
->bitmap_file
&&
541 st
->ss
->add_internal_bitmap
&&
542 (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
543 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) &&
544 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
546 pr_err("automatically enabling write-intent bitmap on large array\n");
547 s
->bitmap_file
= "internal";
549 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "none") == 0)
550 s
->bitmap_file
= NULL
;
552 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
553 !st
->ss
->write_init_ppl
) {
554 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
558 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
559 if (c
->runstop
!= 1 || c
->verbose
>= 0)
560 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
565 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
566 if (c
->runstop
!= 1 || c
->verbose
>= 0)
567 pr_err("%s unable to enumerate platform support\n"
568 " array may not be compatible with hardware/firmware\n",
572 st
->nodes
= c
->nodes
;
573 st
->cluster_name
= c
->homecluster
;
576 if (c
->runstop
!= 1) {
577 if (!ask("Continue creating array? ")) {
578 pr_err("create aborted.\n");
583 pr_err("creation continuing despite oddities due to --run\n");
587 /* If this is raid4/5, we want to configure the last active slot
588 * as missing, so that a reconstruct happens (faster than re-parity)
589 * FIX: Can we do this for raid6 as well?
591 if (st
->ss
->external
== 0 && s
->assume_clean
== 0 &&
592 c
->force
== 0 && first_missing
>= s
->raiddisks
) {
596 insert_point
= s
->raiddisks
-1;
598 info
.array
.active_disks
--;
605 /* For raid6, if creating with 1 missing drive, make a good drive
606 * into a spare, else the create will fail
608 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
609 st
->ss
->external
== 0 &&
610 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
611 insert_point
= s
->raiddisks
- 1;
612 if (insert_point
== first_missing
)
615 info
.array
.active_disks
--;
619 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
620 pr_err("This level does not support missing devices\n");
624 /* We need to create the device */
626 mdfd
= create_mddev(mddev
, name
, c
->autof
, LOCAL
, chosen_name
, 1);
631 /* verify if chosen_name is not in use,
632 * it could be in conflict with already existing device
633 * e.g. container, array
635 if (strncmp(chosen_name
, "/dev/md/", 8) == 0 &&
636 map_by_name(&map
, chosen_name
+8) != NULL
) {
637 pr_err("Array name %s is in use already.\n",
646 memset(&inf
, 0, sizeof(inf
));
647 md_get_array_info(mdfd
, &inf
);
648 if (inf
.working_disks
!= 0) {
649 pr_err("another array by this name is already running.\n");
653 /* Ok, lets try some ioctls */
655 info
.array
.level
= s
->level
;
656 info
.array
.size
= s
->size
;
657 info
.array
.raid_disks
= s
->raiddisks
;
658 /* The kernel should *know* what md_minor we are dealing
659 * with, but it chooses to trust me instead. Sigh
661 info
.array
.md_minor
= 0;
662 if (fstat_is_blkdev(mdfd
, mddev
, &rdev
))
663 info
.array
.md_minor
= minor(rdev
);
664 info
.array
.not_persistent
= 0;
666 if (((s
->level
== 4 || s
->level
== 5) &&
667 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
)) ||
668 (s
->level
== 6 && (insert_point
< s
->raiddisks
||
669 second_missing
< s
->raiddisks
)) ||
670 (s
->level
<= 0) || s
->assume_clean
) {
671 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
672 info
.resync_start
= MaxSector
;
674 info
.array
.state
= 0; /* not clean, but no errors */
675 info
.resync_start
= 0;
677 if (s
->level
== 10) {
678 /* for raid10, the bitmap size is the capacity of the array,
679 * which is array.size * raid_disks / ncopies;
680 * .. but convert to sectors.
682 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
683 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
684 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
686 bitmapsize
= s
->size
* 2;
688 /* There is lots of redundancy in these disk counts,
689 * raid_disks is the most meaningful value
690 * it describes the geometry of the array
692 * nr_disks is total number of used slots.
693 * it should be raid_disks+spare_disks
694 * spare_disks is the number of extra disks present
696 * active_disks is the number of working disks in
697 * active slots. (With raid_disks)
698 * working_disks is the total number of working disks,
700 * failed_disks is the number of disks marked failed
702 * Ideally, the kernel would keep these (except raid_disks)
703 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
704 * So for now, we assume that all raid and spare
705 * devices will be given.
707 info
.array
.spare_disks
=s
->sparedisks
;
708 info
.array
.failed_disks
=missing_disks
;
709 info
.array
.nr_disks
= info
.array
.working_disks
710 + info
.array
.failed_disks
;
711 info
.array
.layout
= s
->layout
;
712 info
.array
.chunk_size
= s
->chunk
*1024;
714 if (name
== NULL
|| *name
== 0) {
715 /* base name on mddev */
721 * /dev/md/home -> home
722 * /dev/mdhome -> home
724 /* FIXME compare this with rules in create_mddev */
725 name
= strrchr(mddev
, '/');
728 if (strncmp(name
, "md_", 3) == 0 &&
729 strlen(name
) > 3 && (name
-mddev
) == 5 /* /dev/ */)
731 else if (strncmp(name
, "md", 2) == 0 &&
732 strlen(name
) > 2 && isdigit(name
[2]) &&
733 (name
-mddev
) == 5 /* /dev/ */)
737 if (!st
->ss
->init_super(st
, &info
.array
, s
, name
, c
->homehost
, uuid
,
741 total_slots
= info
.array
.nr_disks
;
742 st
->ss
->getinfo_super(st
, &info
, NULL
);
743 if (sysfs_init(&info
, mdfd
, NULL
)) {
744 pr_err("unable to initialize sysfs\n");
748 if (did_default
&& c
->verbose
>= 0) {
749 if (is_subarray(info
.text_version
)) {
754 strncpy(devnm
, info
.text_version
+1, 32);
756 ep
= strchr(devnm
, '/');
760 mdi
= sysfs_read(-1, devnm
, GET_VERSION
);
762 pr_err("Creating array inside %s container %s\n",
763 mdi
?mdi
->text_version
:"managed", devnm
);
766 pr_err("Defaulting to version %s metadata\n", info
.text_version
);
769 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
770 info
.uuid
, chosen_name
);
771 /* Keep map locked until devices have been added to array
772 * to stop another mdadm from finding and using those devices.
775 if (s
->bitmap_file
&& (strcmp(s
->bitmap_file
, "internal") == 0 ||
776 strcmp(s
->bitmap_file
, "clustered") == 0)) {
777 if (!st
->ss
->add_internal_bitmap
) {
778 pr_err("internal bitmaps not supported with %s metadata\n",
782 if (st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
783 c
->delay
, s
->write_behind
,
784 bitmapsize
, 1, major_num
)) {
785 pr_err("Given bitmap chunk size not supported.\n");
788 s
->bitmap_file
= NULL
;
791 if (sysfs_init(&info
, mdfd
, NULL
)) {
792 pr_err("unable to initialize sysfs\n");
796 if (st
->ss
->external
&& st
->container_devnm
[0]) {
799 /* When creating a member, we need to be careful
800 * to negotiate with mdmon properly.
801 * If it is already running, we cannot write to
802 * the devices and must ask it to do that part.
803 * If it isn't running, we write to the devices,
805 * We hold an exclusive open on the container
806 * device to make sure mdmon doesn't exit after
807 * we checked that it is running.
809 * For now, fail if it is already running.
811 container_fd
= open_dev_excl(st
->container_devnm
);
812 if (container_fd
< 0) {
813 pr_err("Cannot get exclusive open on container - weird.\n");
816 if (mdmon_running(st
->container_devnm
)) {
818 pr_err("reusing mdmon for %s.\n",
819 st
->container_devnm
);
820 st
->update_tail
= &st
->updates
;
824 rv
= set_array_info(mdfd
, st
, &info
);
826 pr_err("failed to set array info for %s: %s\n",
827 mddev
, strerror(errno
));
831 if (s
->bitmap_file
) {
834 st
->ss
->uuid_from_super(st
, uuid
);
835 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
836 c
->delay
, s
->write_behind
,
841 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
843 pr_err("weird: %s cannot be opened\n",
847 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
848 pr_err("Cannot set bitmap file for %s: %s\n",
849 mddev
, strerror(errno
));
854 infos
= xmalloc(sizeof(*infos
) * total_slots
);
855 enable_fds(total_slots
);
856 for (pass
= 1; pass
<= 2; pass
++) {
857 struct mddev_dev
*moved_disk
= NULL
; /* the disk that was moved out of the insert point */
859 for (dnum
= 0, raid_disk_num
= 0, dv
= devlist
; dv
;
860 dv
= (dv
->next
) ? (dv
->next
) : moved_disk
, dnum
++) {
862 struct mdinfo
*inf
= &infos
[dnum
];
864 if (dnum
>= total_slots
)
866 if (dnum
== insert_point
) {
871 if (strcasecmp(dv
->devname
, "missing") == 0) {
877 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
878 /* repeatedly use the container */
885 inf
->disk
.number
= dnum
;
886 inf
->disk
.raid_disk
= raid_disk_num
++;
888 if (dv
->disposition
== 'j') {
889 inf
->disk
.raid_disk
= MD_DISK_ROLE_JOURNAL
;
890 inf
->disk
.state
= (1<<MD_DISK_JOURNAL
);
892 } else if (inf
->disk
.raid_disk
< s
->raiddisks
)
893 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
898 if (dv
->writemostly
== FlagSet
)
899 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
900 if (dv
->failfast
== FlagSet
)
901 inf
->disk
.state
|= (1<<MD_DISK_FAILFAST
);
906 if (st
->ss
->external
&&
907 st
->container_devnm
[0])
908 fd
= open(dv
->devname
, O_RDWR
);
910 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
913 pr_err("failed to open %s after earlier success - aborting\n",
917 if (!fstat_is_blkdev(fd
, dv
->devname
, &rdev
))
919 inf
->disk
.major
= major(rdev
);
920 inf
->disk
.minor
= minor(rdev
);
923 remove_partitions(fd
);
924 if (st
->ss
->add_to_super(st
, &inf
->disk
,
927 ioctl(mdfd
, STOP_ARRAY
, NULL
);
930 st
->ss
->getinfo_super(st
, inf
, NULL
);
931 safe_mode_delay
= inf
->safe_mode_delay
;
933 if (have_container
&& c
->verbose
> 0)
934 pr_err("Using %s for device %d\n",
935 map_dev(inf
->disk
.major
,
939 if (!have_container
) {
940 /* getinfo_super might have lost these ... */
941 inf
->disk
.major
= major(rdev
);
942 inf
->disk
.minor
= minor(rdev
);
948 rv
= add_disk(mdfd
, st
, &info
, inf
);
951 pr_err("ADD_NEW_DISK for %s failed: %s\n",
952 dv
->devname
, strerror(errno
));
957 if (!have_container
&&
958 dv
== moved_disk
&& dnum
!= insert_point
) break;
961 struct mdinfo info_new
;
962 struct map_ent
*me
= NULL
;
964 /* check to see if the uuid has changed due to these
965 * metadata changes, and if so update the member array
966 * and container uuid. Note ->write_init_super clears
967 * the subarray cursor such that ->getinfo_super once
968 * again returns container info.
970 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
971 if (st
->ss
->external
&& s
->level
!= LEVEL_CONTAINER
&&
972 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
973 map_update(&map
, fd2devnm(mdfd
),
974 info_new
.text_version
,
975 info_new
.uuid
, chosen_name
);
976 me
= map_by_devnm(&map
, st
->container_devnm
);
979 if (st
->ss
->write_init_super(st
)) {
980 st
->ss
->free_super(st
);
984 /* update parent container uuid */
986 char *path
= xstrdup(me
->path
);
988 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
989 map_update(&map
, st
->container_devnm
,
990 info_new
.text_version
,
991 info_new
.uuid
, path
);
995 flush_metadata_updates(st
);
996 st
->ss
->free_super(st
);
1002 if (s
->level
== LEVEL_CONTAINER
) {
1003 /* No need to start. But we should signal udev to
1005 sysfs_uevent(&info
, "change");
1006 if (c
->verbose
>= 0)
1007 pr_err("container %s prepared.\n", mddev
);
1008 wait_for(chosen_name
, mdfd
);
1009 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
1010 if (st
->ss
->external
) {
1014 case LEVEL_MULTIPATH
:
1016 err
= sysfs_set_str(&info
, NULL
, "array_state",
1023 err
= sysfs_set_str(&info
, NULL
, "array_state",
1027 sysfs_set_safemode(&info
, safe_mode_delay
);
1029 pr_err("failed to activate array.\n");
1030 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1033 } else if (c
->readonly
&&
1034 sysfs_attribute_available(
1035 &info
, NULL
, "array_state")) {
1036 if (sysfs_set_str(&info
, NULL
,
1037 "array_state", "readonly") < 0) {
1038 pr_err("Failed to start array: %s\n",
1040 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1044 /* param is not actually used */
1046 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1047 pr_err("RUN_ARRAY failed: %s\n",
1049 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1050 cont_err("Problem may be that chunk size is not a power of 2\n");
1052 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1055 /* if start_ro module parameter is set, array is
1056 * auto-read-only, which is bad as the resync won't
1057 * start. So lets make it read-write now.
1059 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1061 if (c
->verbose
>= 0)
1062 pr_err("array %s started.\n", mddev
);
1063 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1065 start_mdmon(st
->container_devnm
);
1067 ping_monitor(st
->container_devnm
);
1068 close(container_fd
);
1070 wait_for(chosen_name
, mdfd
);
1072 pr_err("not starting array - not enough devices.\n");
1075 /* Give udev a moment to process the Change event caused
1086 map_remove(&map
, fd2devnm(mdfd
));