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
;
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 */
103 int have_container
= 0;
104 int container_fd
= -1;
106 unsigned long long bitmapsize
;
107 struct mdinfo info
, *infos
;
109 int do_default_layout
= 0;
110 int do_default_chunk
= 0;
111 unsigned long safe_mode_delay
= 0;
112 char chosen_name
[1024];
113 struct map_ent
*map
= NULL
;
114 unsigned long long newsize
;
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
153 mdu_array_info_t inf
;
156 memset(&inf
, 0, sizeof(inf
));
157 fd
= open(devlist
->devname
, O_RDONLY
);
159 md_get_array_info(fd
, &inf
) == 0 && inf
.raid_disks
== 0) {
160 /* yep, looks like a container */
162 rv
= st
->ss
->load_container(st
, fd
,
167 st
= super_by_fd(fd
, NULL
);
168 if (st
&& !(rv
= st
->ss
->
169 load_container(st
, fd
,
175 if (have_container
) {
176 subdevs
= s
->raiddisks
;
177 first_missing
= subdevs
* 2;
178 second_missing
= subdevs
* 2;
179 insert_point
= subdevs
* 2;
185 if (st
&& st
->ss
->external
&& s
->sparedisks
) {
186 pr_err("This metadata type does not support spare disks at create time\n");
189 if (subdevs
> s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
190 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, s
->raiddisks
+s
->sparedisks
);
193 if (!have_container
&& subdevs
< s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
194 pr_err("You haven't given enough devices (real or missing) to create this array\n");
197 if (s
->bitmap_file
&& s
->level
<= 0) {
198 pr_err("bitmaps not meaningful with level %s\n",
199 map_num(pers
, s
->level
)?:"given");
203 /* now set some defaults */
205 if (s
->layout
== UnSet
) {
206 do_default_layout
= 1;
207 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
211 /* check layout fits in array*/
212 if ((s
->layout
&255) * ((s
->layout
>>8)&255) > s
->raiddisks
) {
213 pr_err("that layout requires at least %d devices\n",
214 (s
->layout
&255) * ((s
->layout
>>8)&255));
224 if (s
->chunk
== 0 || s
->chunk
== UnSet
) {
226 do_default_chunk
= 1;
227 /* chunk will be set later */
231 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
232 if (get_linux_version() < 2006016 && s
->chunk
== 0) {
235 pr_err("chunk size defaults to 64K\n");
240 case LEVEL_MULTIPATH
:
241 case LEVEL_CONTAINER
:
245 pr_err("chunk size ignored for this level\n");
249 pr_err("unknown level %d\n", s
->level
);
252 if (s
->size
== MAX_SIZE
)
253 /* use '0' to mean 'max' now... */
255 if (s
->size
&& s
->chunk
&& s
->chunk
!= UnSet
)
256 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
257 newsize
= s
->size
* 2;
258 if (st
&& ! st
->ss
->validate_geometry(st
, s
->level
, s
->layout
, s
->raiddisks
,
259 &s
->chunk
, s
->size
*2,
261 &newsize
, s
->consistency_policy
,
265 if (s
->chunk
&& s
->chunk
!= UnSet
) {
266 newsize
&= ~(unsigned long long)(s
->chunk
*2 - 1);
267 if (do_default_chunk
) {
268 /* default chunk was just set */
270 pr_err("chunk size defaults to %dK\n", s
->chunk
);
271 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
272 do_default_chunk
= 0;
277 s
->size
= newsize
/ 2;
279 /* If this is ever reshaped to RAID5, we will
280 * need a chunksize. So round it off a bit
281 * now just to be safe
283 s
->size
&= ~(64ULL-1);
285 if (s
->size
&& c
->verbose
> 0)
286 pr_err("setting size to %lluK\n", s
->size
);
289 /* now look at the subdevs */
290 info
.array
.active_disks
= 0;
291 info
.array
.working_disks
= 0;
293 for (dv
= devlist
; dv
; dv
= dv
->next
)
294 if (data_offset
== VARIABLE_OFFSET
)
295 dv
->data_offset
= INVALID_SECTORS
;
297 dv
->data_offset
= data_offset
;
299 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
300 char *dname
= dv
->devname
;
301 unsigned long long freesize
;
305 if (strcasecmp(dname
, "missing")==0) {
306 if (first_missing
> dnum
)
307 first_missing
= dnum
;
308 if (second_missing
> dnum
&& dnum
> first_missing
)
309 second_missing
= dnum
;
313 if (data_offset
== VARIABLE_OFFSET
) {
314 doff
= strchr(dname
, ':');
317 dv
->data_offset
= parse_size(doff
);
319 dv
->data_offset
= INVALID_SECTORS
;
321 dv
->data_offset
= data_offset
;
323 dfd
= open(dname
, O_RDONLY
);
325 pr_err("cannot open %s: %s\n",
326 dname
, strerror(errno
));
329 if (fstat(dfd
, &stb
) != 0 ||
330 (stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
332 pr_err("%s is not a block device\n",
337 info
.array
.working_disks
++;
338 if (dnum
< s
->raiddisks
&& dv
->disposition
!= 'j')
339 info
.array
.active_disks
++;
341 struct createinfo
*ci
= conf_get_create_info();
346 /* Need to choose a default metadata, which is different
347 * depending on geometry of array.
350 char *name
= "default";
351 for(i
=0; !st
&& superlist
[i
]; i
++) {
352 st
= superlist
[i
]->match_metadata_desc(name
);
355 if (do_default_layout
)
356 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
357 switch (st
->ss
->validate_geometry(
358 st
, s
->level
, s
->layout
, s
->raiddisks
,
359 &s
->chunk
, s
->size
*2,
360 dv
->data_offset
, dname
,
361 &freesize
, s
->consistency_policy
,
363 case -1: /* Not valid, message printed, and not
364 * worth checking any further */
367 case 0: /* Geometry not valid */
370 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
372 case 1: /* All happy */
378 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
380 pr_err("cannot open %s: %s\n",
381 dname
, strerror(errno
));
384 pr_err("device %s not suitable for any style of array\n",
388 if (st
->ss
!= &super0
||
389 st
->minor_version
!= 90)
392 if (do_default_layout
)
393 s
->layout
= default_layout(st
, s
->level
, 0);
394 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
396 &s
->chunk
, s
->size
*2,
399 s
->consistency_policy
,
402 pr_err("%s is not suitable for this array.\n",
409 if (dv
->disposition
== 'j')
410 goto skip_size_check
; /* skip write journal for size check */
412 freesize
/= 2; /* convert to K */
413 if (s
->chunk
&& s
->chunk
!= UnSet
) {
414 /* round to chunk size */
415 freesize
= freesize
& ~(s
->chunk
-1);
416 if (do_default_chunk
) {
417 /* default chunk was just set */
419 pr_err("chunk size defaults to %dK\n", s
->chunk
);
420 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
421 do_default_chunk
= 0;
425 pr_err("no free space left on %s\n", dname
);
430 if (s
->size
&& freesize
< s
->size
) {
431 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
432 dname
, freesize
, s
->size
);
436 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
440 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
445 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
446 int fd
= open(dname
, O_RDONLY
);
448 pr_err("Cannot open %s: %s\n",
449 dname
, strerror(errno
));
453 warn
|= check_ext2(fd
, dname
);
454 warn
|= check_reiser(fd
, dname
);
455 warn
|= check_raid(fd
, dname
);
456 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
457 st
->minor_version
>= 1)
458 /* metadata at front */
459 warn
|= check_partitions(fd
, dname
, 0, 0);
460 else if (s
->level
== 1 || s
->level
== LEVEL_CONTAINER
461 || (s
->level
== 0 && s
->raiddisks
== 1))
462 /* partitions could be meaningful */
463 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
465 /* partitions cannot be meaningful */
466 warn
|= check_partitions(fd
, dname
, 0, 0);
467 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
468 st
->minor_version
>= 1 &&
471 (warn
& 1024) == 0) {
473 pr_err("Note: this array has metadata at the start and\n"
474 " may not be suitable as a boot device. If you plan to\n"
475 " store '/boot' on this device please ensure that\n"
476 " your boot-loader understands md/v1.x metadata, or use\n"
477 " --metadata=0.90\n");
482 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
483 pr_err("Too many devices: %s metadata only supports %d\n",
484 st
->ss
->name
, st
->max_devs
);
488 info
.array
.working_disks
= s
->raiddisks
;
490 pr_err("create aborted\n");
494 if (mindisc
== NULL
&& !have_container
) {
495 pr_err("no size and no drives given - aborting create.\n");
498 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
499 || s
->level
== LEVEL_FAULTY
500 || st
->ss
->external
) {
501 /* size is meaningful */
502 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
504 &s
->chunk
, minsize
*2,
507 s
->consistency_policy
, 0)) {
508 pr_err("devices too large for RAID level %d\n", s
->level
);
513 /* If this is ever reshaped to RAID5, we will
514 * need a chunksize. So round it off a bit
515 * now just to be safe
517 s
->size
&= ~(64ULL-1);
519 pr_err("size set to %lluK\n", s
->size
);
523 if (!s
->bitmap_file
&&
525 st
->ss
->add_internal_bitmap
&&
526 (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
527 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) &&
528 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
530 pr_err("automatically enabling write-intent bitmap on large array\n");
531 s
->bitmap_file
= "internal";
533 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "none") == 0)
534 s
->bitmap_file
= NULL
;
536 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
537 !st
->ss
->write_init_ppl
) {
538 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
542 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
543 if (c
->runstop
!= 1 || c
->verbose
>= 0)
544 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
549 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
550 if (c
->runstop
!= 1 || c
->verbose
>= 0)
551 pr_err("%s unable to enumerate platform support\n"
552 " array may not be compatible with hardware/firmware\n",
556 st
->nodes
= c
->nodes
;
557 st
->cluster_name
= c
->homecluster
;
560 if (c
->runstop
!= 1) {
561 if (!ask("Continue creating array? ")) {
562 pr_err("create aborted.\n");
567 pr_err("creation continuing despite oddities due to --run\n");
571 /* If this is raid4/5, we want to configure the last active slot
572 * as missing, so that a reconstruct happens (faster than re-parity)
573 * FIX: Can we do this for raid6 as well?
575 if (st
->ss
->external
== 0 &&
576 s
->assume_clean
==0 && c
->force
== 0 && first_missing
>= s
->raiddisks
) {
577 switch ( s
->level
) {
580 insert_point
= s
->raiddisks
-1;
582 info
.array
.active_disks
--;
589 /* For raid6, if creating with 1 missing drive, make a good drive
590 * into a spare, else the create will fail
592 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
593 st
->ss
->external
== 0 &&
594 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
595 insert_point
= s
->raiddisks
- 1;
596 if (insert_point
== first_missing
)
599 info
.array
.active_disks
--;
603 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
604 pr_err("This level does not support missing devices\n");
608 /* We need to create the device */
610 mdfd
= create_mddev(mddev
, name
, c
->autof
, LOCAL
, chosen_name
);
615 /* verify if chosen_name is not in use,
616 * it could be in conflict with already existing device
617 * e.g. container, array
619 if (strncmp(chosen_name
, "/dev/md/", 8) == 0
620 && map_by_name(&map
, chosen_name
+8) != NULL
) {
621 pr_err("Array name %s is in use already.\n",
629 vers
= md_get_version(mdfd
);
631 pr_err("Create requires md driver version 0.90.0 or later\n");
634 mdu_array_info_t inf
;
635 memset(&inf
, 0, sizeof(inf
));
636 md_get_array_info(mdfd
, &inf
);
637 if (inf
.working_disks
!= 0) {
638 pr_err("another array by this name is already running.\n");
643 /* Ok, lets try some ioctls */
645 info
.array
.level
= s
->level
;
646 info
.array
.size
= s
->size
;
647 info
.array
.raid_disks
= s
->raiddisks
;
648 /* The kernel should *know* what md_minor we are dealing
649 * with, but it chooses to trust me instead. Sigh
651 info
.array
.md_minor
= 0;
652 if (fstat(mdfd
, &stb
)==0)
653 info
.array
.md_minor
= minor(stb
.st_rdev
);
654 info
.array
.not_persistent
= 0;
656 if ( ( (s
->level
== 4 || s
->level
== 5) &&
657 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
) )
659 ( s
->level
== 6 && (insert_point
< s
->raiddisks
660 || second_missing
< s
->raiddisks
))
666 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
667 info
.resync_start
= MaxSector
;
669 info
.array
.state
= 0; /* not clean, but no errors */
670 info
.resync_start
= 0;
672 if (s
->level
== 10) {
673 /* for raid10, the bitmap size is the capacity of the array,
674 * which is array.size * raid_disks / ncopies;
675 * .. but convert to sectors.
677 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
678 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
679 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
681 bitmapsize
= s
->size
* 2;
683 /* There is lots of redundancy in these disk counts,
684 * raid_disks is the most meaningful value
685 * it describes the geometry of the array
687 * nr_disks is total number of used slots.
688 * it should be raid_disks+spare_disks
689 * spare_disks is the number of extra disks present
691 * active_disks is the number of working disks in
692 * active slots. (With raid_disks)
693 * working_disks is the total number of working disks,
695 * failed_disks is the number of disks marked failed
697 * Ideally, the kernel would keep these (except raid_disks)
698 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
699 * So for now, we assume that all raid and spare
700 * devices will be given.
702 info
.array
.spare_disks
=s
->sparedisks
;
703 info
.array
.failed_disks
=missing_disks
;
704 info
.array
.nr_disks
= info
.array
.working_disks
705 + info
.array
.failed_disks
;
706 info
.array
.layout
= s
->layout
;
707 info
.array
.chunk_size
= s
->chunk
*1024;
709 if (name
== NULL
|| *name
== 0) {
710 /* base name on mddev */
716 * /dev/md/home -> home
717 * /dev/mdhome -> home
719 /* FIXME compare this with rules in create_mddev */
720 name
= strrchr(mddev
, '/');
723 if (strncmp(name
, "md_", 3)==0 &&
725 (name
-mddev
) == 5 /* /dev/ */)
727 else if (strncmp(name
, "md", 2)==0 &&
730 (name
-mddev
) == 5 /* /dev/ */)
734 if (!st
->ss
->init_super(st
, &info
.array
, s
, name
, c
->homehost
, uuid
,
738 total_slots
= info
.array
.nr_disks
;
739 st
->ss
->getinfo_super(st
, &info
, NULL
);
740 if (sysfs_init(&info
, mdfd
, NULL
)) {
741 pr_err("unable to initialize sysfs\n");
745 if (did_default
&& c
->verbose
>= 0) {
746 if (is_subarray(info
.text_version
)) {
751 strncpy(devnm
, info
.text_version
+1, 32);
753 ep
= strchr(devnm
, '/');
757 mdi
= sysfs_read(-1, devnm
, GET_VERSION
);
759 pr_err("Creating array inside %s container %s\n",
760 mdi
?mdi
->text_version
:"managed", devnm
);
763 pr_err("Defaulting to version %s metadata\n", info
.text_version
);
766 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
767 info
.uuid
, chosen_name
);
768 /* Keep map locked until devices have been added to array
769 * to stop another mdadm from finding and using those devices.
772 if (s
->bitmap_file
&& vers
< 9003) {
773 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
775 pr_err("Warning - bitmaps created on this kernel are not portable\n"
776 " between different architectured. Consider upgrading the Linux kernel.\n");
780 if (s
->bitmap_file
&& (strcmp(s
->bitmap_file
, "internal")==0 ||
781 strcmp(s
->bitmap_file
, "clustered")==0)) {
782 if ((vers
%100) < 2) {
783 pr_err("internal bitmaps not supported by this kernel.\n");
786 if (!st
->ss
->add_internal_bitmap
) {
787 pr_err("internal bitmaps not supported with %s metadata\n",
791 if (st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
792 c
->delay
, s
->write_behind
,
793 bitmapsize
, 1, major_num
)) {
794 pr_err("Given bitmap chunk size not supported.\n");
797 s
->bitmap_file
= NULL
;
800 if (sysfs_init(&info
, mdfd
, NULL
)) {
801 pr_err("unable to initialize sysfs\n");
805 if (st
->ss
->external
&& st
->container_devnm
[0]) {
808 /* When creating a member, we need to be careful
809 * to negotiate with mdmon properly.
810 * If it is already running, we cannot write to
811 * the devices and must ask it to do that part.
812 * If it isn't running, we write to the devices,
814 * We hold an exclusive open on the container
815 * device to make sure mdmon doesn't exit after
816 * we checked that it is running.
818 * For now, fail if it is already running.
820 container_fd
= open_dev_excl(st
->container_devnm
);
821 if (container_fd
< 0) {
822 pr_err("Cannot get exclusive open on container - weird.\n");
825 if (mdmon_running(st
->container_devnm
)) {
827 pr_err("reusing mdmon for %s.\n",
828 st
->container_devnm
);
829 st
->update_tail
= &st
->updates
;
833 rv
= set_array_info(mdfd
, st
, &info
);
835 pr_err("failed to set array info for %s: %s\n",
836 mddev
, strerror(errno
));
840 if (s
->bitmap_file
) {
843 st
->ss
->uuid_from_super(st
, uuid
);
844 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
845 c
->delay
, s
->write_behind
,
850 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
852 pr_err("weird: %s cannot be openned\n",
856 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
857 pr_err("Cannot set bitmap file for %s: %s\n",
858 mddev
, strerror(errno
));
863 infos
= xmalloc(sizeof(*infos
) * total_slots
);
864 enable_fds(total_slots
);
865 for (pass
=1; pass
<=2 ; pass
++) {
866 struct mddev_dev
*moved_disk
= NULL
; /* the disk that was moved out of the insert point */
868 for (dnum
=0, raid_disk_num
=0, dv
= devlist
; dv
;
869 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
872 struct mdinfo
*inf
= &infos
[dnum
];
874 if (dnum
>= total_slots
)
876 if (dnum
== insert_point
) {
881 if (strcasecmp(dv
->devname
, "missing")==0) {
887 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
888 /* repeatedly use the container */
895 inf
->disk
.number
= dnum
;
896 inf
->disk
.raid_disk
= raid_disk_num
++;
898 if (dv
->disposition
== 'j') {
899 inf
->disk
.raid_disk
= MD_DISK_ROLE_JOURNAL
;
900 inf
->disk
.state
= (1<<MD_DISK_JOURNAL
);
902 } else if (inf
->disk
.raid_disk
< s
->raiddisks
)
903 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
908 if (dv
->writemostly
== FlagSet
)
909 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
910 if (dv
->failfast
== FlagSet
)
911 inf
->disk
.state
|= (1<<MD_DISK_FAILFAST
);
916 if (st
->ss
->external
&&
917 st
->container_devnm
[0])
918 fd
= open(dv
->devname
, O_RDWR
);
920 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
923 pr_err("failed to open %s after earlier success - aborting\n",
928 inf
->disk
.major
= major(stb
.st_rdev
);
929 inf
->disk
.minor
= minor(stb
.st_rdev
);
932 remove_partitions(fd
);
933 if (st
->ss
->add_to_super(st
, &inf
->disk
,
936 ioctl(mdfd
, STOP_ARRAY
, NULL
);
939 st
->ss
->getinfo_super(st
, inf
, NULL
);
940 safe_mode_delay
= inf
->safe_mode_delay
;
942 if (have_container
&& c
->verbose
> 0)
943 pr_err("Using %s for device %d\n",
944 map_dev(inf
->disk
.major
,
948 if (!have_container
) {
949 /* getinfo_super might have lost these ... */
950 inf
->disk
.major
= major(stb
.st_rdev
);
951 inf
->disk
.minor
= minor(stb
.st_rdev
);
957 rv
= add_disk(mdfd
, st
, &info
, inf
);
960 pr_err("ADD_NEW_DISK for %s failed: %s\n",
961 dv
->devname
, strerror(errno
));
966 if (!have_container
&&
967 dv
== moved_disk
&& dnum
!= insert_point
) break;
970 struct mdinfo info_new
;
971 struct map_ent
*me
= NULL
;
973 /* check to see if the uuid has changed due to these
974 * metadata changes, and if so update the member array
975 * and container uuid. Note ->write_init_super clears
976 * the subarray cursor such that ->getinfo_super once
977 * again returns container info.
979 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
980 if (st
->ss
->external
&& s
->level
!= LEVEL_CONTAINER
&&
981 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
982 map_update(&map
, fd2devnm(mdfd
),
983 info_new
.text_version
,
984 info_new
.uuid
, chosen_name
);
985 me
= map_by_devnm(&map
, st
->container_devnm
);
988 if (st
->ss
->write_init_super(st
)) {
989 st
->ss
->free_super(st
);
993 /* update parent container uuid */
995 char *path
= xstrdup(me
->path
);
997 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
998 map_update(&map
, st
->container_devnm
,
999 info_new
.text_version
,
1000 info_new
.uuid
, path
);
1004 flush_metadata_updates(st
);
1005 st
->ss
->free_super(st
);
1011 if (s
->level
== LEVEL_CONTAINER
) {
1012 /* No need to start. But we should signal udev to
1014 sysfs_uevent(&info
, "change");
1015 if (c
->verbose
>= 0)
1016 pr_err("container %s prepared.\n", mddev
);
1017 wait_for(chosen_name
, mdfd
);
1018 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
1019 if (st
->ss
->external
) {
1023 case LEVEL_MULTIPATH
:
1025 err
= sysfs_set_str(&info
, NULL
, "array_state",
1032 err
= sysfs_set_str(&info
, NULL
, "array_state",
1036 sysfs_set_safemode(&info
, safe_mode_delay
);
1038 pr_err("failed to activate array.\n");
1039 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1042 } else if (c
->readonly
&&
1043 sysfs_attribute_available(
1044 &info
, NULL
, "array_state")) {
1045 if (sysfs_set_str(&info
, NULL
,
1046 "array_state", "readonly") < 0) {
1047 pr_err("Failed to start array: %s\n",
1049 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1053 /* param is not actually used */
1055 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1056 pr_err("RUN_ARRAY failed: %s\n",
1058 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1059 cont_err("Problem may be that chunk size is not a power of 2\n");
1061 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1064 /* if start_ro module parameter is set, array is
1065 * auto-read-only, which is bad as the resync won't
1066 * start. So lets make it read-write now.
1068 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1070 if (c
->verbose
>= 0)
1071 pr_err("array %s started.\n", mddev
);
1072 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1074 start_mdmon(st
->container_devnm
);
1076 ping_monitor(st
->container_devnm
);
1077 close(container_fd
);
1079 wait_for(chosen_name
, mdfd
);
1081 pr_err("not starting array - not enough devices.\n");
1089 map_remove(&map
, fd2devnm(mdfd
));