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>
32 #include <sys/signalfd.h>
35 #ifndef FALLOC_FL_ZERO_RANGE
36 #define FALLOC_FL_ZERO_RANGE 16
39 static int round_size_and_verify(unsigned long long *size
, int chunk
)
43 *size
&= ~(unsigned long long)(chunk
- 1);
45 pr_err("Size cannot be smaller than chunk.\n");
52 * default_layout() - Get default layout for level.
53 * @st: metadata requested, could be NULL.
54 * @level: raid level requested.
55 * @verbose: verbose level.
57 * Try to ask metadata handler first, otherwise use global defaults.
59 * Return: Layout or &UnSet, return value meaning depends of level used.
61 int default_layout(struct supertype
*st
, int level
, int verbose
)
64 mapping_t
*layout_map
= NULL
;
65 char *layout_name
= NULL
;
67 if (st
&& st
->ss
->default_geometry
)
68 st
->ss
->default_geometry(st
, &level
, &layout
, NULL
);
74 default: /* no layout */
78 layout
= RAID0_ORIG_LAYOUT
;
81 layout
= 0x102; /* near=2, far=1 */
86 layout_map
= r5layout
;
89 layout_map
= faultylayout
;
94 layout
= map_name(layout_map
, "default");
95 layout_name
= map_num_s(layout_map
, layout
);
97 if (layout_name
&& verbose
> 0)
98 pr_err("layout defaults to %s\n", layout_name
);
103 static pid_t
write_zeroes_fork(int fd
, struct shape
*s
, struct supertype
*st
,
104 struct mddev_dev
*dv
)
107 const unsigned long long req_size
= 1 << 30;
108 unsigned long long offset_bytes
, size_bytes
, sz
;
113 size_bytes
= KIB_TO_BYTES(s
->size
);
116 * If size_bytes is zero, this is a zoned raid array where
117 * each disk is of a different size and uses its full
118 * disk. Thus zero the entire disk.
120 if (!size_bytes
&& !get_dev_size(fd
, dv
->devname
, &size_bytes
))
123 if (dv
->data_offset
!= INVALID_SECTORS
)
124 offset_bytes
= SEC_TO_BYTES(dv
->data_offset
);
126 offset_bytes
= SEC_TO_BYTES(st
->data_offset
);
128 pr_info("zeroing data from %lld to %lld on: %s\n",
129 offset_bytes
, size_bytes
, dv
->devname
);
133 pr_err("Could not fork to zero disks: %s\n", strerror(errno
));
135 } else if (pid
!= 0) {
139 sigemptyset(&sigset
);
140 sigaddset(&sigset
, SIGINT
);
141 sigprocmask(SIG_UNBLOCK
, &sigset
, NULL
);
145 * Split requests to the kernel into 1GB chunks seeing the
146 * fallocate() call is not interruptible and blocking a
147 * ctrl-c for several minutes is not desirable.
149 * 1GB is chosen as a compromise: the user may still have
150 * to wait several seconds if they ctrl-c on devices that
151 * zero slowly, but will reduce the number of requests
152 * required and thus the overhead on devices that perform
159 if (fallocate(fd
, FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
,
161 pr_err("zeroing %s failed: %s\n", dv
->devname
,
174 static int wait_for_zero_forks(int *zero_pids
, int count
)
176 int wstatus
, ret
= 0, i
, sfd
, wait_count
= 0;
177 struct signalfd_siginfo fdsi
;
178 bool interrupted
= false;
182 for (i
= 0; i
< count
; i
++)
188 sigemptyset(&sigset
);
189 sigaddset(&sigset
, SIGINT
);
190 sigaddset(&sigset
, SIGCHLD
);
191 sigprocmask(SIG_BLOCK
, &sigset
, NULL
);
193 sfd
= signalfd(-1, &sigset
, 0);
195 pr_err("Unable to create signalfd: %s\n", strerror(errno
));
200 s
= read(sfd
, &fdsi
, sizeof(fdsi
));
201 if (s
!= sizeof(fdsi
)) {
202 pr_err("Invalid signalfd read: %s\n", strerror(errno
));
207 if (fdsi
.ssi_signo
== SIGINT
) {
209 pr_info("Interrupting zeroing processes, please wait...\n");
211 } else if (fdsi
.ssi_signo
== SIGCHLD
) {
219 for (i
= 0; i
< count
; i
++) {
223 waitpid(zero_pids
[i
], &wstatus
, 0);
225 if (!WIFEXITED(wstatus
) || WEXITSTATUS(wstatus
))
230 pr_err("zeroing interrupted!\n");
235 pr_err("zeroing failed!\n");
237 pr_info("zeroing finished\n");
242 static int add_disk_to_super(int mdfd
, struct shape
*s
, struct context
*c
,
243 struct supertype
*st
, struct mddev_dev
*dv
,
244 struct mdinfo
*info
, int have_container
, int major_num
,
250 if (dv
->disposition
== 'j') {
251 info
->disk
.raid_disk
= MD_DISK_ROLE_JOURNAL
;
252 info
->disk
.state
= (1<<MD_DISK_JOURNAL
);
253 } else if (info
->disk
.raid_disk
< s
->raiddisks
) {
254 info
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
257 info
->disk
.state
= 0;
260 if (dv
->writemostly
== FlagSet
) {
261 if (major_num
== BITMAP_MAJOR_CLUSTERED
) {
262 pr_err("Can not set %s --write-mostly with a clustered bitmap\n",dv
->devname
);
265 info
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
270 if (dv
->failfast
== FlagSet
)
271 info
->disk
.state
|= (1<<MD_DISK_FAILFAST
);
273 if (have_container
) {
276 if (st
->ss
->external
&& st
->container_devnm
[0])
277 fd
= open(dv
->devname
, O_RDWR
);
279 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
282 pr_err("failed to open %s after earlier success - aborting\n",
286 if (!fstat_is_blkdev(fd
, dv
->devname
, &rdev
)) {
290 info
->disk
.major
= major(rdev
);
291 info
->disk
.minor
= minor(rdev
);
294 remove_partitions(fd
);
295 if (st
->ss
->add_to_super(st
, &info
->disk
, fd
, dv
->devname
,
297 ioctl(mdfd
, STOP_ARRAY
, NULL
);
301 st
->ss
->getinfo_super(st
, info
, NULL
);
303 if (fd
>= 0 && s
->write_zeroes
) {
304 *zero_pid
= write_zeroes_fork(fd
, s
, st
, dv
);
305 if (*zero_pid
<= 0) {
306 ioctl(mdfd
, STOP_ARRAY
, NULL
);
312 if (have_container
&& c
->verbose
> 0)
313 pr_err("Using %s for device %d\n",
314 map_dev(info
->disk
.major
, info
->disk
.minor
, 0),
317 if (!have_container
) {
318 /* getinfo_super might have lost these ... */
319 info
->disk
.major
= major(rdev
);
320 info
->disk
.minor
= minor(rdev
);
326 static int update_metadata(int mdfd
, struct shape
*s
, struct supertype
*st
,
327 struct map_ent
**map
, struct mdinfo
*info
,
330 struct mdinfo info_new
;
331 struct map_ent
*me
= NULL
;
333 /* check to see if the uuid has changed due to these
334 * metadata changes, and if so update the member array
335 * and container uuid. Note ->write_init_super clears
336 * the subarray cursor such that ->getinfo_super once
337 * again returns container info.
339 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
340 if (st
->ss
->external
&& !is_container(s
->level
) &&
341 !same_uuid(info_new
.uuid
, info
->uuid
, 0)) {
342 map_update(map
, fd2devnm(mdfd
),
343 info_new
.text_version
,
344 info_new
.uuid
, chosen_name
);
345 me
= map_by_devnm(map
, st
->container_devnm
);
348 if (st
->ss
->write_init_super(st
)) {
349 st
->ss
->free_super(st
);
354 * Before activating the array, perform extra steps
355 * required to configure the internal write-intent
358 if (info_new
.consistency_policy
== CONSISTENCY_POLICY_BITMAP
&&
359 st
->ss
->set_bitmap
&& st
->ss
->set_bitmap(st
, info
)) {
360 st
->ss
->free_super(st
);
364 /* update parent container uuid */
366 char *path
= xstrdup(me
->path
);
368 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
369 map_update(map
, st
->container_devnm
, info_new
.text_version
,
370 info_new
.uuid
, path
);
374 flush_metadata_updates(st
);
375 st
->ss
->free_super(st
);
380 static int add_disks(int mdfd
, struct mdinfo
*info
, struct shape
*s
,
381 struct context
*c
, struct supertype
*st
,
382 struct map_ent
**map
, struct mddev_dev
*devlist
,
383 int total_slots
, int have_container
, int insert_point
,
384 int major_num
, char *chosen_name
)
386 struct mddev_dev
*moved_disk
= NULL
;
387 int pass
, raid_disk_num
, dnum
;
388 int zero_pids
[total_slots
];
389 struct mddev_dev
*dv
;
390 struct mdinfo
*infos
;
391 sigset_t sigset
, orig_sigset
;
395 * Block SIGINT so the main thread will always wait for the
396 * zeroing processes when being interrupted. Otherwise the
397 * zeroing processes will finish their work in the background
398 * keeping the disk busy.
400 sigemptyset(&sigset
);
401 sigaddset(&sigset
, SIGINT
);
402 sigprocmask(SIG_BLOCK
, &sigset
, &orig_sigset
);
403 memset(zero_pids
, 0, sizeof(zero_pids
));
405 infos
= xmalloc(sizeof(*infos
) * total_slots
);
406 enable_fds(total_slots
);
407 for (pass
= 1; pass
<= 2; pass
++) {
408 for (dnum
= 0, raid_disk_num
= 0, dv
= devlist
; dv
;
409 dv
= (dv
->next
) ? (dv
->next
) : moved_disk
, dnum
++) {
410 if (dnum
>= total_slots
)
412 if (dnum
== insert_point
) {
417 if (strcasecmp(dv
->devname
, "missing") == 0) {
423 if (have_container
&& dnum
< total_slots
- 1)
424 /* repeatedly use the container */
430 infos
[dnum
].disk
.number
= dnum
;
431 infos
[dnum
].disk
.raid_disk
= raid_disk_num
++;
433 if (dv
->disposition
== 'j')
436 ret
= add_disk_to_super(mdfd
, s
, c
, st
, dv
,
437 &infos
[dnum
], have_container
,
438 major_num
, &zero_pids
[dnum
]);
444 infos
[dnum
].errors
= 0;
446 ret
= add_disk(mdfd
, st
, info
, &infos
[dnum
]);
448 pr_err("ADD_NEW_DISK for %s failed: %s\n",
449 dv
->devname
, strerror(errno
));
450 if (errno
== EINVAL
&&
451 info
->array
.level
== 0) {
452 pr_err("Possibly your kernel doesn't support RAID0 layouts.\n");
453 pr_err("Either upgrade, or use --layout=dangerous\n");
459 if (!have_container
&&
460 dv
== moved_disk
&& dnum
!= insert_point
) break;
464 ret
= wait_for_zero_forks(zero_pids
, total_slots
);
468 ret
= update_metadata(mdfd
, s
, st
, map
, info
,
477 wait_for_zero_forks(zero_pids
, total_slots
);
479 sigprocmask(SIG_SETMASK
, &orig_sigset
, NULL
);
483 int Create(struct supertype
*st
, struct mddev_ident
*ident
, int subdevs
,
484 struct mddev_dev
*devlist
, struct shape
*s
, struct context
*c
)
487 * Create a new raid array.
489 * First check that necessary details are available
490 * (i.e. level, raid-disks)
492 * Then check each disk to see what might be on it
493 * and report anything interesting.
495 * If anything looks odd, and runstop not set,
498 * SET_ARRAY_INFO and ADD_NEW_DISK, and
499 * if runstop==run, or raiddisks disks were used,
503 unsigned long long minsize
= 0, maxsize
= 0;
504 dev_policy_t
*custom_pols
= NULL
;
505 char *mindisc
= NULL
;
506 char *maxdisc
= NULL
;
507 char *name
= ident
->name
;
508 int *uuid
= ident
->uuid_set
== 1 ? ident
->uuid
: NULL
;
510 struct mddev_dev
*dv
;
512 int fail
= 0, warn
= 0;
513 int first_missing
= subdevs
* 2;
514 int second_missing
= subdevs
* 2;
515 int missing_disks
= 0;
516 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
520 int have_container
= 0;
521 int container_fd
= -1;
523 unsigned long long bitmapsize
;
526 int do_default_layout
= 0;
527 int do_default_chunk
= 0;
528 char chosen_name
[1024];
529 struct map_ent
*map
= NULL
;
530 unsigned long long newsize
;
531 mdu_array_info_t inf
;
533 int major_num
= BITMAP_MAJOR_HI
;
534 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "clustered") == 0) {
535 major_num
= BITMAP_MAJOR_CLUSTERED
;
537 pr_err("At least 2 nodes are needed for cluster-md\n");
542 memset(&info
, 0, sizeof(info
));
543 if (s
->level
== UnSet
&& st
&& st
->ss
->default_geometry
)
544 st
->ss
->default_geometry(st
, &s
->level
, NULL
, NULL
);
545 if (s
->level
== UnSet
) {
546 pr_err("a RAID level is needed to create an array.\n");
549 if (s
->raiddisks
< 4 && s
->level
== 6) {
550 pr_err("at least 4 raid-devices needed for level 6\n");
553 if (s
->raiddisks
> 256 && s
->level
== 6) {
554 pr_err("no more than 256 raid-devices supported for level 6\n");
557 if (s
->raiddisks
< 2 && s
->level
>= 4) {
558 pr_err("at least 2 raid-devices needed for level %d\n", s
->level
);
561 if (s
->level
<= 0 && s
->sparedisks
) {
562 pr_err("This level does not support spare devices\n");
566 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
567 /* If given a single device, it might be a container, and we can
568 * extract a device list from there
572 memset(&inf
, 0, sizeof(inf
));
573 fd
= open(devlist
->devname
, O_RDONLY
);
575 md_get_array_info(fd
, &inf
) == 0 && inf
.raid_disks
== 0) {
576 /* yep, looks like a container */
578 rv
= st
->ss
->load_container(st
, fd
,
583 st
= super_by_fd(fd
, NULL
);
584 if (st
&& !(rv
= st
->ss
->
585 load_container(st
, fd
,
591 if (have_container
) {
592 subdevs
= s
->raiddisks
;
593 first_missing
= subdevs
* 2;
594 second_missing
= subdevs
* 2;
595 insert_point
= subdevs
* 2;
597 if (mddev_test_and_add_drive_policies(st
, &custom_pols
, fd
, 1))
604 if (st
&& st
->ss
->external
&& s
->sparedisks
) {
605 pr_err("This metadata type does not support spare disks at create time\n");
608 if (subdevs
> s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
609 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, s
->raiddisks
+s
->sparedisks
);
612 if (!have_container
&& subdevs
< s
->raiddisks
+s
->sparedisks
+s
->journaldisks
) {
613 pr_err("You haven't given enough devices (real or missing) to create this array\n");
616 if (s
->bitmap_file
&& s
->level
<= 0) {
617 pr_err("bitmaps not meaningful with level %s\n",
618 map_num(pers
, s
->level
)?:"given");
622 /* now set some defaults */
624 if (s
->layout
== UnSet
) {
625 do_default_layout
= 1;
626 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
630 /* check layout fits in array*/
631 if ((s
->layout
&255) * ((s
->layout
>>8)&255) > s
->raiddisks
) {
632 pr_err("that layout requires at least %d devices\n",
633 (s
->layout
&255) * ((s
->layout
>>8)&255));
643 if (s
->chunk
== 0 || s
->chunk
== UnSet
) {
645 do_default_chunk
= 1;
646 /* chunk will be set later */
650 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
654 case LEVEL_MULTIPATH
:
655 case LEVEL_CONTAINER
:
657 pr_err("specifying chunk size is forbidden for this level\n");
662 pr_err("unknown level %d\n", s
->level
);
666 if (s
->size
== MAX_SIZE
)
667 /* use '0' to mean 'max' now... */
669 if (s
->size
&& s
->chunk
&& s
->chunk
!= UnSet
)
670 if (round_size_and_verify(&s
->size
, s
->chunk
))
673 newsize
= s
->size
* 2;
674 if (st
&& ! st
->ss
->validate_geometry(st
, s
->level
, s
->layout
, s
->raiddisks
,
675 &s
->chunk
, s
->size
*2,
676 s
->data_offset
, NULL
,
677 &newsize
, s
->consistency_policy
,
681 if (s
->chunk
&& s
->chunk
!= UnSet
) {
682 newsize
&= ~(unsigned long long)(s
->chunk
*2 - 1);
683 if (do_default_chunk
) {
684 /* default chunk was just set */
686 pr_err("chunk size defaults to %dK\n", s
->chunk
);
687 if (round_size_and_verify(&s
->size
, s
->chunk
))
689 do_default_chunk
= 0;
694 s
->size
= newsize
/ 2;
696 /* If this is ever reshaped to RAID5, we will
697 * need a chunksize. So round it off a bit
698 * now just to be safe
700 s
->size
&= ~(64ULL-1);
702 if (s
->size
&& c
->verbose
> 0)
703 pr_err("setting size to %lluK\n", s
->size
);
706 /* now look at the subdevs */
707 info
.array
.active_disks
= 0;
708 info
.array
.working_disks
= 0;
710 for (dv
= devlist
; dv
; dv
= dv
->next
)
711 if (s
->data_offset
== VARIABLE_OFFSET
)
712 dv
->data_offset
= INVALID_SECTORS
;
714 dv
->data_offset
= s
->data_offset
;
716 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
717 char *dname
= dv
->devname
;
718 unsigned long long freesize
;
722 if (strcasecmp(dname
, "missing") == 0) {
723 if (first_missing
> dnum
)
724 first_missing
= dnum
;
725 if (second_missing
> dnum
&& dnum
> first_missing
)
726 second_missing
= dnum
;
730 if (s
->data_offset
== VARIABLE_OFFSET
) {
731 doff
= strchr(dname
, ':');
734 dv
->data_offset
= parse_size(doff
);
736 dv
->data_offset
= INVALID_SECTORS
;
738 dv
->data_offset
= s
->data_offset
;
740 dfd
= open(dname
, O_RDONLY
);
742 pr_err("cannot open %s: %s\n",
743 dname
, strerror(errno
));
746 if (!fstat_is_blkdev(dfd
, dname
, NULL
)) {
751 info
.array
.working_disks
++;
752 if (dnum
< s
->raiddisks
&& dv
->disposition
!= 'j')
753 info
.array
.active_disks
++;
755 struct createinfo
*ci
= conf_get_create_info();
760 /* Need to choose a default metadata, which is different
761 * depending on geometry of array.
764 char *name
= "default";
765 for(i
= 0; !st
&& superlist
[i
]; i
++) {
766 st
= superlist
[i
]->match_metadata_desc(name
);
769 if (do_default_layout
)
770 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
771 switch (st
->ss
->validate_geometry(
772 st
, s
->level
, s
->layout
, s
->raiddisks
,
773 &s
->chunk
, s
->size
*2,
774 dv
->data_offset
, dname
,
775 &freesize
, s
->consistency_policy
,
777 case -1: /* Not valid, message printed, and not
778 * worth checking any further */
781 case 0: /* Geometry not valid */
784 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
786 case 1: /* All happy */
792 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
794 pr_err("cannot open %s: %s\n",
795 dname
, strerror(errno
));
798 pr_err("device %s not suitable for any style of array\n",
802 if (st
->ss
!= &super0
||
803 st
->minor_version
!= 90)
806 if (do_default_layout
)
807 s
->layout
= default_layout(st
, s
->level
, 0);
808 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
810 &s
->chunk
, s
->size
*2,
813 s
->consistency_policy
,
816 pr_err("%s is not suitable for this array.\n",
823 if (drive_test_and_add_policies(st
, &custom_pols
, dfd
, 1))
828 if (dv
->disposition
== 'j')
829 goto skip_size_check
; /* skip write journal for size check */
831 freesize
/= 2; /* convert to K */
832 if (s
->chunk
&& s
->chunk
!= UnSet
) {
833 /* round to chunk size */
834 freesize
= freesize
& ~(s
->chunk
-1);
835 if (do_default_chunk
) {
836 /* default chunk was just set */
838 pr_err("chunk size defaults to %dK\n", s
->chunk
);
839 if (round_size_and_verify(&s
->size
, s
->chunk
))
841 do_default_chunk
= 0;
845 pr_err("no free space left on %s\n", dname
);
850 if (s
->size
&& freesize
< s
->size
) {
851 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
852 dname
, freesize
, s
->size
);
856 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
860 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
865 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
866 int fd
= open(dname
, O_RDONLY
);
868 pr_err("Cannot open %s: %s\n",
869 dname
, strerror(errno
));
873 warn
|= check_ext2(fd
, dname
);
874 warn
|= check_reiser(fd
, dname
);
875 warn
|= check_raid(fd
, dname
);
876 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
877 st
->minor_version
>= 1)
878 /* metadata at front */
879 warn
|= check_partitions(fd
, dname
, 0, 0);
880 else if (s
->level
== 1 || is_container(s
->level
) ||
881 (s
->level
== 0 && s
->raiddisks
== 1))
882 /* partitions could be meaningful */
883 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
885 /* partitions cannot be meaningful */
886 warn
|= check_partitions(fd
, dname
, 0, 0);
887 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
888 st
->minor_version
>= 1 &&
891 (warn
& 1024) == 0) {
893 pr_err("Note: this array has metadata at the start and\n"
894 " may not be suitable as a boot device. If you plan to\n"
895 " store '/boot' on this device please ensure that\n"
896 " your boot-loader understands md/v1.x metadata, or use\n"
897 " --metadata=0.90\n");
903 if (missing_disks
== dnum
&& !have_container
) {
904 pr_err("Subdevs can't be all missing\n");
907 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
908 pr_err("Too many devices: %s metadata only supports %d\n",
909 st
->ss
->name
, st
->max_devs
);
913 info
.array
.working_disks
= s
->raiddisks
;
915 pr_err("create aborted\n");
919 if (mindisc
== NULL
&& !have_container
) {
920 pr_err("no size and no drives given - aborting create.\n");
923 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
||
924 s
->level
== LEVEL_FAULTY
|| st
->ss
->external
) {
925 /* size is meaningful */
926 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
928 &s
->chunk
, minsize
*2,
931 s
->consistency_policy
, 0)) {
932 pr_err("devices too large for RAID level %d\n", s
->level
);
937 /* If this is ever reshaped to RAID5, we will
938 * need a chunksize. So round it off a bit
939 * now just to be safe
941 s
->size
&= ~(64ULL-1);
943 pr_err("size set to %lluK\n", s
->size
);
947 if (!s
->bitmap_file
&&
950 st
->ss
->add_internal_bitmap
&&
951 s
->journaldisks
== 0 &&
952 (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
953 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) &&
954 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
956 pr_err("automatically enabling write-intent bitmap on large array\n");
957 s
->bitmap_file
= "internal";
959 if (s
->bitmap_file
&& str_is_none(s
->bitmap_file
) == true)
960 s
->bitmap_file
= NULL
;
962 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
963 !st
->ss
->write_init_ppl
) {
964 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
968 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
969 if (c
->runstop
!= 1 || c
->verbose
>= 0)
970 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
975 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
976 if (c
->runstop
!= 1 || c
->verbose
>= 0)
977 pr_err("%s unable to enumerate platform support\n"
978 " array may not be compatible with hardware/firmware\n",
982 st
->nodes
= c
->nodes
;
983 st
->cluster_name
= c
->homecluster
;
986 if (c
->runstop
!= 1) {
987 if (!ask("Continue creating array? ")) {
988 pr_err("create aborted.\n");
993 pr_err("creation continuing despite oddities due to --run\n");
997 /* If this is raid4/5, we want to configure the last active slot
998 * as missing, so that a reconstruct happens (faster than re-parity)
999 * FIX: Can we do this for raid6 as well?
1001 if (st
->ss
->external
== 0 && s
->assume_clean
== 0 &&
1002 c
->force
== 0 && first_missing
>= s
->raiddisks
) {
1006 insert_point
= s
->raiddisks
-1;
1008 info
.array
.active_disks
--;
1015 /* For raid6, if creating with 1 missing drive, make a good drive
1016 * into a spare, else the create will fail
1018 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
1019 st
->ss
->external
== 0 &&
1020 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
1021 insert_point
= s
->raiddisks
- 1;
1022 if (insert_point
== first_missing
)
1025 info
.array
.active_disks
--;
1029 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
1030 pr_err("This level does not support missing devices\n");
1034 /* We need to create the device */
1036 mdfd
= create_mddev(ident
->devname
, ident
->name
, c
->autof
, LOCAL
, chosen_name
, 1);
1041 /* verify if chosen_name is not in use,
1042 * it could be in conflict with already existing device
1043 * e.g. container, array
1045 if (strncmp(chosen_name
, DEV_MD_DIR
, DEV_MD_DIR_LEN
) == 0 &&
1046 map_by_name(&map
, chosen_name
+ DEV_MD_DIR_LEN
)) {
1047 pr_err("Array name %s is in use already.\n", chosen_name
);
1054 memset(&inf
, 0, sizeof(inf
));
1055 md_get_array_info(mdfd
, &inf
);
1056 if (inf
.working_disks
!= 0) {
1057 pr_err("another array by this name is already running.\n");
1061 /* Ok, lets try some ioctls */
1063 info
.array
.level
= s
->level
;
1064 info
.array
.size
= s
->size
;
1065 info
.array
.raid_disks
= s
->raiddisks
;
1066 /* The kernel should *know* what md_minor we are dealing
1067 * with, but it chooses to trust me instead. Sigh
1069 info
.array
.md_minor
= 0;
1070 if (fstat_is_blkdev(mdfd
, chosen_name
, &rdev
))
1071 info
.array
.md_minor
= minor(rdev
);
1072 info
.array
.not_persistent
= 0;
1074 if (((s
->level
== 4 || s
->level
== 5) &&
1075 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
)) ||
1076 (s
->level
== 6 && (insert_point
< s
->raiddisks
||
1077 second_missing
< s
->raiddisks
)) ||
1078 (s
->level
<= 0) || s
->assume_clean
) {
1079 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
1080 info
.resync_start
= MaxSector
;
1082 info
.array
.state
= 0; /* not clean, but no errors */
1083 info
.resync_start
= 0;
1085 if (s
->level
== 10) {
1086 /* for raid10, the bitmap size is the capacity of the array,
1087 * which is array.size * raid_disks / ncopies;
1088 * .. but convert to sectors.
1090 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
1091 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
1092 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
1094 bitmapsize
= s
->size
* 2;
1096 /* There is lots of redundancy in these disk counts,
1097 * raid_disks is the most meaningful value
1098 * it describes the geometry of the array
1100 * nr_disks is total number of used slots.
1101 * it should be raid_disks+spare_disks
1102 * spare_disks is the number of extra disks present
1104 * active_disks is the number of working disks in
1105 * active slots. (With raid_disks)
1106 * working_disks is the total number of working disks,
1108 * failed_disks is the number of disks marked failed
1110 * Ideally, the kernel would keep these (except raid_disks)
1111 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
1112 * So for now, we assume that all raid and spare
1113 * devices will be given.
1115 info
.array
.spare_disks
=s
->sparedisks
;
1116 info
.array
.failed_disks
=missing_disks
;
1117 info
.array
.nr_disks
= info
.array
.working_disks
1118 + info
.array
.failed_disks
;
1119 info
.array
.layout
= s
->layout
;
1120 info
.array
.chunk_size
= s
->chunk
*1024;
1123 /* base name on devname */
1126 * /dev/md_foo -> foo
1129 * /dev/md/home -> home
1130 * /dev/mdhome -> home
1132 /* FIXME compare this with rules in create_mddev */
1133 name
= strrchr(chosen_name
, '/');
1137 if (strncmp(name
, "md_", 3) == 0 &&
1138 strlen(name
) > 3 && (name
- chosen_name
) == 5 /* /dev/ */)
1140 else if (strncmp(name
, "md", 2) == 0 &&
1141 strlen(name
) > 2 && isdigit(name
[2]) &&
1142 (name
- chosen_name
) == 5 /* /dev/ */)
1146 if (!st
->ss
->init_super(st
, &info
.array
, s
, name
, c
->homehost
, uuid
,
1150 total_slots
= info
.array
.nr_disks
;
1151 st
->ss
->getinfo_super(st
, &info
, NULL
);
1152 if (sysfs_init(&info
, mdfd
, NULL
)) {
1153 pr_err("unable to initialize sysfs\n");
1158 if (is_subarray(info
.text_version
)) {
1159 char devnm
[MD_NAME_MAX
];
1162 sysfs_get_container_devnm(&info
, devnm
);
1164 mdi
= sysfs_read(-1, devnm
, GET_VERSION
| GET_DEVS
);
1166 pr_err("Cannot open sysfs for container %s\n", devnm
);
1170 if (sysfs_test_and_add_drive_policies(st
, &custom_pols
, mdi
, 1))
1173 if (c
->verbose
>= 0)
1174 pr_info("Creating array inside %s container /dev/%s\n",
1175 mdi
->text_version
, devnm
);
1178 } else if (c
->verbose
>= 0) {
1179 pr_info("Defaulting to version %s metadata\n", info
.text_version
);
1183 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
1184 info
.uuid
, chosen_name
);
1185 /* Keep map locked until devices have been added to array
1186 * to stop another mdadm from finding and using those devices.
1189 if (s
->bitmap_file
&& (strcmp(s
->bitmap_file
, "internal") == 0 ||
1190 strcmp(s
->bitmap_file
, "clustered") == 0)) {
1191 if (!st
->ss
->add_internal_bitmap
) {
1192 pr_err("internal bitmaps not supported with %s metadata\n",
1196 if (st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
1197 c
->delay
, s
->write_behind
,
1198 bitmapsize
, 1, major_num
)) {
1199 pr_err("Given bitmap chunk size not supported.\n");
1202 s
->bitmap_file
= NULL
;
1205 if (sysfs_init(&info
, mdfd
, NULL
)) {
1206 pr_err("unable to initialize sysfs\n");
1210 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1213 /* When creating a member, we need to be careful
1214 * to negotiate with mdmon properly.
1215 * If it is already running, we cannot write to
1216 * the devices and must ask it to do that part.
1217 * If it isn't running, we write to the devices,
1218 * and then start it.
1219 * We hold an exclusive open on the container
1220 * device to make sure mdmon doesn't exit after
1221 * we checked that it is running.
1223 * For now, fail if it is already running.
1225 container_fd
= open_dev_excl(st
->container_devnm
);
1226 if (container_fd
< 0) {
1227 pr_err("Cannot get exclusive open on container - weird.\n");
1230 if (mdmon_running(st
->container_devnm
)) {
1232 pr_err("reusing mdmon for %s.\n",
1233 st
->container_devnm
);
1234 st
->update_tail
= &st
->updates
;
1238 rv
= set_array_info(mdfd
, st
, &info
);
1240 pr_err("failed to set array info for %s: %s\n", chosen_name
, strerror(errno
));
1244 if (s
->bitmap_file
) {
1247 st
->ss
->uuid_from_super(st
, uuid
);
1248 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
1249 c
->delay
, s
->write_behind
,
1254 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
1255 if (bitmap_fd
< 0) {
1256 pr_err("weird: %s cannot be opened\n",
1260 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
1261 pr_err("Cannot set bitmap file for %s: %s\n", chosen_name
, strerror(errno
));
1266 if (add_disks(mdfd
, &info
, s
, c
, st
, &map
, devlist
, total_slots
,
1267 have_container
, insert_point
, major_num
, chosen_name
))
1272 if (is_container(s
->level
)) {
1273 /* No need to start. But we should signal udev to
1275 sysfs_uevent(&info
, "change");
1276 if (c
->verbose
>= 0)
1277 pr_err("container %s prepared.\n", chosen_name
);
1278 wait_for(chosen_name
, mdfd
);
1279 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
1280 if (st
->ss
->external
) {
1284 case LEVEL_MULTIPATH
:
1286 err
= sysfs_set_str(&info
, NULL
, "array_state",
1293 err
= sysfs_set_str(&info
, NULL
, "array_state",
1297 sysfs_set_safemode(&info
, info
.safe_mode_delay
);
1299 pr_err("failed to activate array.\n");
1300 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1303 } else if (c
->readonly
&&
1304 sysfs_attribute_available(
1305 &info
, NULL
, "array_state")) {
1306 if (sysfs_set_str(&info
, NULL
,
1307 "array_state", "readonly") < 0) {
1308 pr_err("Failed to start array: %s\n",
1310 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1314 /* param is not actually used */
1316 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1317 pr_err("RUN_ARRAY failed: %s\n",
1319 if (errno
== 524 /* ENOTSUP */ &&
1320 info
.array
.level
== 0)
1321 cont_err("Please use --layout=original or --layout=alternate\n");
1322 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1323 cont_err("Problem may be that chunk size is not a power of 2\n");
1325 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1328 /* if start_ro module parameter is set, array is
1329 * auto-read-only, which is bad as the resync won't
1330 * start. So lets make it read-write now.
1332 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1334 if (c
->verbose
>= 0)
1335 pr_info("array %s started.\n", chosen_name
);
1336 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1338 start_mdmon(st
->container_devnm
);
1340 ping_monitor(st
->container_devnm
);
1341 close(container_fd
);
1343 wait_for(chosen_name
, mdfd
);
1345 pr_err("not starting array - not enough devices.\n");
1349 sysfs_uevent(&info
, "change");
1350 dev_policy_free(custom_pols
);
1358 map_remove(&map
, fd2devnm(mdfd
));
1364 dev_policy_free(custom_pols
);