]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Create.c
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;
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
;
118 memset(&info
, 0, sizeof(info
));
119 if (s
->level
== UnSet
&& st
&& st
->ss
->default_geometry
)
120 st
->ss
->default_geometry(st
, &s
->level
, NULL
, NULL
);
121 if (s
->level
== UnSet
) {
122 pr_err("a RAID level is needed to create an array.\n");
125 if (s
->raiddisks
< 4 && s
->level
== 6) {
126 pr_err("at least 4 raid-devices needed for level 6\n");
129 if (s
->raiddisks
> 256 && s
->level
== 6) {
130 pr_err("no more than 256 raid-devices supported for level 6\n");
133 if (s
->raiddisks
< 2 && s
->level
>= 4) {
134 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
137 if (s
->level
<= 0 && s
->sparedisks
) {
138 pr_err("This level does not support spare devices\n");
142 if (subdevs
== 1 && strcmp(devlist
->devname
, "missing") != 0) {
143 /* If given a single device, it might be a container, and we can
144 * extract a device list from there
146 mdu_array_info_t inf
;
149 memset(&inf
, 0, sizeof(inf
));
150 fd
= open(devlist
->devname
, O_RDONLY
);
152 ioctl(fd
, GET_ARRAY_INFO
, &inf
) == 0 &&
153 inf
.raid_disks
== 0) {
154 /* yep, looks like a container */
156 rv
= st
->ss
->load_container(st
, fd
,
161 st
= super_by_fd(fd
, NULL
);
162 if (st
&& !(rv
= st
->ss
->
163 load_container(st
, fd
,
169 if (have_container
) {
170 subdevs
= s
->raiddisks
;
171 first_missing
= subdevs
* 2;
172 second_missing
= subdevs
* 2;
173 insert_point
= subdevs
* 2;
179 if (st
&& st
->ss
->external
&& s
->sparedisks
) {
180 pr_err("This metadata type does not support spare disks at create time\n");
183 if (subdevs
> s
->raiddisks
+s
->sparedisks
) {
184 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, s
->raiddisks
+s
->sparedisks
);
187 if (!have_container
&& subdevs
< s
->raiddisks
+s
->sparedisks
) {
188 pr_err("You haven't given enough devices (real or missing) to create this array\n");
191 if (s
->bitmap_file
&& s
->level
<= 0) {
192 pr_err("bitmaps not meaningful with level %s\n",
193 map_num(pers
, s
->level
)?:"given");
197 /* now set some defaults */
199 if (s
->layout
== UnSet
) {
200 do_default_layout
= 1;
201 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
205 /* check layout fits in array*/
206 if ((s
->layout
&255) * ((s
->layout
>>8)&255) > s
->raiddisks
) {
207 pr_err("that layout requires at least %d devices\n",
208 (s
->layout
&255) * ((s
->layout
>>8)&255));
218 if (s
->chunk
== 0 || s
->chunk
== UnSet
) {
220 do_default_chunk
= 1;
221 /* chunk will be set later */
225 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
226 if (get_linux_version() < 2006016 && s
->chunk
== 0) {
229 pr_err("chunk size defaults to 64K\n");
234 case LEVEL_MULTIPATH
:
235 case LEVEL_CONTAINER
:
239 pr_err("chunk size ignored for this level\n");
243 pr_err("unknown level %d\n", s
->level
);
246 if (s
->size
== MAX_SIZE
)
247 /* use '0' to mean 'max' now... */
249 if (s
->size
&& s
->chunk
&& s
->chunk
!= UnSet
)
250 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
251 newsize
= s
->size
* 2;
252 if (st
&& ! st
->ss
->validate_geometry(st
, s
->level
, s
->layout
, s
->raiddisks
,
253 &s
->chunk
, s
->size
*2,
255 &newsize
, c
->verbose
>=0))
258 if (s
->chunk
&& s
->chunk
!= UnSet
) {
259 newsize
&= ~(unsigned long long)(s
->chunk
*2 - 1);
260 if (do_default_chunk
) {
261 /* default chunk was just set */
263 pr_err("chunk size defaults to %dK\n", s
->chunk
);
264 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
265 do_default_chunk
= 0;
270 s
->size
= newsize
/ 2;
272 /* If this is ever reshaped to RAID5, we will
273 * need a chunksize. So round it off a bit
274 * now just to be safe
276 s
->size
&= ~(64ULL-1);
278 if (s
->size
&& c
->verbose
> 0)
279 pr_err("setting size to %lluK\n", s
->size
);
282 /* now look at the subdevs */
283 info
.array
.active_disks
= 0;
284 info
.array
.working_disks
= 0;
286 for (dv
= devlist
; dv
; dv
= dv
->next
)
287 if (data_offset
== VARIABLE_OFFSET
)
288 dv
->data_offset
= INVALID_SECTORS
;
290 dv
->data_offset
= data_offset
;
292 for (dv
=devlist
; dv
&& !have_container
; dv
=dv
->next
, dnum
++) {
293 char *dname
= dv
->devname
;
294 unsigned long long freesize
;
298 if (strcasecmp(dname
, "missing")==0) {
299 if (first_missing
> dnum
)
300 first_missing
= dnum
;
301 if (second_missing
> dnum
&& dnum
> first_missing
)
302 second_missing
= dnum
;
306 if (data_offset
== VARIABLE_OFFSET
) {
307 doff
= strchr(dname
, ':');
310 dv
->data_offset
= parse_size(doff
);
312 dv
->data_offset
= INVALID_SECTORS
;
314 dv
->data_offset
= data_offset
;
316 dfd
= open(dname
, O_RDONLY
);
318 pr_err("cannot open %s: %s\n",
319 dname
, strerror(errno
));
322 if (fstat(dfd
, &stb
) != 0 ||
323 (stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
325 pr_err("%s is not a block device\n",
330 info
.array
.working_disks
++;
331 if (dnum
< s
->raiddisks
)
332 info
.array
.active_disks
++;
334 struct createinfo
*ci
= conf_get_create_info();
339 /* Need to choose a default metadata, which is different
340 * depending on geometry of array.
343 char *name
= "default";
344 for(i
=0; !st
&& superlist
[i
]; i
++) {
345 st
= superlist
[i
]->match_metadata_desc(name
);
348 if (do_default_layout
)
349 s
->layout
= default_layout(st
, s
->level
, c
->verbose
);
350 switch (st
->ss
->validate_geometry(
351 st
, s
->level
, s
->layout
, s
->raiddisks
,
352 &s
->chunk
, s
->size
*2,
353 dv
->data_offset
, dname
,
354 &freesize
, c
->verbose
> 0)) {
355 case -1: /* Not valid, message printed, and not
356 * worth checking any further */
359 case 0: /* Geometry not valid */
362 s
->chunk
= do_default_chunk
? UnSet
: s
->chunk
;
364 case 1: /* All happy */
370 int dfd
= open(dname
, O_RDONLY
|O_EXCL
);
372 pr_err("cannot open %s: %s\n",
373 dname
, strerror(errno
));
376 pr_err("device %s not suitable for any style of array\n",
380 if (st
->ss
!= &super0
||
381 st
->minor_version
!= 90)
384 if (do_default_layout
)
385 s
->layout
= default_layout(st
, s
->level
, 0);
386 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
388 &s
->chunk
, s
->size
*2,
393 pr_err("%s is not suitable for this array.\n",
400 freesize
/= 2; /* convert to K */
401 if (s
->chunk
&& s
->chunk
!= UnSet
) {
402 /* round to chunk size */
403 freesize
= freesize
& ~(s
->chunk
-1);
404 if (do_default_chunk
) {
405 /* default chunk was just set */
407 pr_err("chunk size defaults to %dK\n", s
->chunk
);
408 s
->size
&= ~(unsigned long long)(s
->chunk
- 1);
409 do_default_chunk
= 0;
413 pr_err("no free space left on %s\n", dname
);
418 if (s
->size
&& freesize
< s
->size
) {
419 pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
420 dname
, freesize
, s
->size
);
424 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
428 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
432 if (c
->runstop
!= 1 || c
->verbose
>= 0) {
433 int fd
= open(dname
, O_RDONLY
);
435 pr_err("Cannot open %s: %s\n",
436 dname
, strerror(errno
));
440 warn
|= check_ext2(fd
, dname
);
441 warn
|= check_reiser(fd
, dname
);
442 warn
|= check_raid(fd
, dname
);
443 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
444 st
->minor_version
>= 1)
445 /* metadata at front */
446 warn
|= check_partitions(fd
, dname
, 0, 0);
447 else if (s
->level
== 1 || s
->level
== LEVEL_CONTAINER
448 || (s
->level
== 0 && s
->raiddisks
== 1))
449 /* partitions could be meaningful */
450 warn
|= check_partitions(fd
, dname
, freesize
*2, s
->size
*2);
452 /* partitions cannot be meaningful */
453 warn
|= check_partitions(fd
, dname
, 0, 0);
454 if (strcmp(st
->ss
->name
, "1.x") == 0 &&
455 st
->minor_version
>= 1 &&
458 (warn
& 1024) == 0) {
460 pr_err("Note: this array has metadata at the start and\n"
461 " may not be suitable as a boot device. If you plan to\n"
462 " store '/boot' on this device please ensure that\n"
463 " your boot-loader understands md/v1.x metadata, or use\n"
464 " --metadata=0.90\n");
469 if (s
->raiddisks
+ s
->sparedisks
> st
->max_devs
) {
470 pr_err("Too many devices: %s metadata only supports %d\n",
471 st
->ss
->name
, st
->max_devs
);
475 info
.array
.working_disks
= s
->raiddisks
;
477 pr_err("create aborted\n");
481 if (mindisc
== NULL
&& !have_container
) {
482 pr_err("no size and no drives given - aborting create.\n");
485 if (s
->level
> 0 || s
->level
== LEVEL_MULTIPATH
486 || s
->level
== LEVEL_FAULTY
487 || st
->ss
->external
) {
488 /* size is meaningful */
489 if (!st
->ss
->validate_geometry(st
, s
->level
, s
->layout
,
491 &s
->chunk
, minsize
*2,
494 pr_err("devices too large for RAID level %d\n", s
->level
);
499 /* If this is ever reshaped to RAID5, we will
500 * need a chunksize. So round it off a bit
501 * now just to be safe
503 s
->size
&= ~(64ULL-1);
505 pr_err("size set to %lluK\n", s
->size
);
509 if (!s
->bitmap_file
&&
511 st
->ss
->add_internal_bitmap
&&
512 (s
->write_behind
|| s
->size
> 100*1024*1024ULL)) {
514 pr_err("automatically enabling write-intent bitmap on large array\n");
515 s
->bitmap_file
= "internal";
517 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "none") == 0)
518 s
->bitmap_file
= NULL
;
520 if (!have_container
&& s
->level
> 0 && ((maxsize
-s
->size
)*100 > maxsize
)) {
521 if (c
->runstop
!= 1 || c
->verbose
>= 0)
522 pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
527 if (st
->ss
->detail_platform
&& st
->ss
->detail_platform(0, 1, NULL
) != 0) {
528 if (c
->runstop
!= 1 || c
->verbose
>= 0)
529 pr_err("%s unable to enumerate platform support\n"
530 " array may not be compatible with hardware/firmware\n",
536 if (c
->runstop
!= 1) {
537 if (!ask("Continue creating array? ")) {
538 pr_err("create aborted.\n");
543 pr_err("creation continuing despite oddities due to --run\n");
547 /* If this is raid4/5, we want to configure the last active slot
548 * as missing, so that a reconstruct happens (faster than re-parity)
549 * FIX: Can we do this for raid6 as well?
551 if (st
->ss
->external
== 0 &&
552 s
->assume_clean
==0 && c
->force
== 0 && first_missing
>= s
->raiddisks
) {
553 switch ( s
->level
) {
556 insert_point
= s
->raiddisks
-1;
558 info
.array
.active_disks
--;
565 /* For raid6, if creating with 1 missing drive, make a good drive
566 * into a spare, else the create will fail
568 if (s
->assume_clean
== 0 && c
->force
== 0 && first_missing
< s
->raiddisks
&&
569 st
->ss
->external
== 0 &&
570 second_missing
>= s
->raiddisks
&& s
->level
== 6) {
571 insert_point
= s
->raiddisks
- 1;
572 if (insert_point
== first_missing
)
575 info
.array
.active_disks
--;
579 if (s
->level
<= 0 && first_missing
< subdevs
* 2) {
580 pr_err("This level does not support missing devices\n");
584 /* We need to create the device */
586 mdfd
= create_mddev(mddev
, name
, c
->autof
, LOCAL
, chosen_name
);
591 /* verify if chosen_name is not in use,
592 * it could be in conflict with already existing device
593 * e.g. container, array
595 if (strncmp(chosen_name
, "/dev/md/", 8) == 0
596 && map_by_name(&map
, chosen_name
+8) != NULL
) {
597 pr_err("Array name %s is in use already.\n",
605 vers
= md_get_version(mdfd
);
607 pr_err("Create requires md driver version 0.90.0 or later\n");
610 mdu_array_info_t inf
;
611 memset(&inf
, 0, sizeof(inf
));
612 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
613 if (inf
.working_disks
!= 0) {
614 pr_err("another array by this name is already running.\n");
619 /* Ok, lets try some ioctls */
621 info
.array
.level
= s
->level
;
622 info
.array
.size
= s
->size
;
623 info
.array
.raid_disks
= s
->raiddisks
;
624 /* The kernel should *know* what md_minor we are dealing
625 * with, but it chooses to trust me instead. Sigh
627 info
.array
.md_minor
= 0;
628 if (fstat(mdfd
, &stb
)==0)
629 info
.array
.md_minor
= minor(stb
.st_rdev
);
630 info
.array
.not_persistent
= 0;
632 if ( ( (s
->level
== 4 || s
->level
== 5) &&
633 (insert_point
< s
->raiddisks
|| first_missing
< s
->raiddisks
) )
635 ( s
->level
== 6 && (insert_point
< s
->raiddisks
636 || second_missing
< s
->raiddisks
))
642 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
643 info
.resync_start
= MaxSector
;
645 info
.array
.state
= 0; /* not clean, but no errors */
646 info
.resync_start
= 0;
648 if (s
->level
== 10) {
649 /* for raid10, the bitmap size is the capacity of the array,
650 * which is array.size * raid_disks / ncopies;
651 * .. but convert to sectors.
653 int ncopies
= ((s
->layout
>>8) & 255) * (s
->layout
& 255);
654 bitmapsize
= s
->size
* s
->raiddisks
/ ncopies
* 2;
655 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
657 bitmapsize
= s
->size
* 2;
659 /* There is lots of redundancy in these disk counts,
660 * raid_disks is the most meaningful value
661 * it describes the geometry of the array
663 * nr_disks is total number of used slots.
664 * it should be raid_disks+spare_disks
665 * spare_disks is the number of extra disks present
667 * active_disks is the number of working disks in
668 * active slots. (With raid_disks)
669 * working_disks is the total number of working disks,
671 * failed_disks is the number of disks marked failed
673 * Ideally, the kernel would keep these (except raid_disks)
674 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
675 * So for now, we assume that all raid and spare
676 * devices will be given.
678 info
.array
.spare_disks
=s
->sparedisks
;
679 info
.array
.failed_disks
=missing_disks
;
680 info
.array
.nr_disks
= info
.array
.working_disks
681 + info
.array
.failed_disks
;
682 info
.array
.layout
= s
->layout
;
683 info
.array
.chunk_size
= s
->chunk
*1024;
685 if (name
== NULL
|| *name
== 0) {
686 /* base name on mddev */
692 * /dev/md/home -> home
693 * /dev/mdhome -> home
695 /* FIXME compare this with rules in create_mddev */
696 name
= strrchr(mddev
, '/');
699 if (strncmp(name
, "md_", 3)==0 &&
701 (name
-mddev
) == 5 /* /dev/ */)
703 else if (strncmp(name
, "md", 2)==0 &&
706 (name
-mddev
) == 5 /* /dev/ */)
710 if (!st
->ss
->init_super(st
, &info
.array
, s
->size
, name
, c
->homehost
, uuid
,
714 total_slots
= info
.array
.nr_disks
;
715 st
->ss
->getinfo_super(st
, &info
, NULL
);
716 sysfs_init(&info
, mdfd
, NULL
);
718 if (did_default
&& c
->verbose
>= 0) {
719 if (is_subarray(info
.text_version
)) {
724 strncpy(devnm
, info
.text_version
+1, 32);
726 ep
= strchr(devnm
, '/');
730 mdi
= sysfs_read(-1, devnm
, GET_VERSION
);
732 pr_err("Creating array inside %s container %s\n",
733 mdi
?mdi
->text_version
:"managed", devnm
);
736 pr_err("Defaulting to version %s metadata\n", info
.text_version
);
739 map_update(&map
, fd2devnm(mdfd
), info
.text_version
,
740 info
.uuid
, chosen_name
);
741 /* Keep map locked until devices have been added to array
742 * to stop another mdadm from finding and using those devices.
745 if (s
->bitmap_file
&& vers
< 9003) {
746 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
748 pr_err("Warning - bitmaps created on this kernel are not portable\n"
749 " between different architectured. Consider upgrading the Linux kernel.\n");
753 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "internal")==0) {
754 if ((vers
%100) < 2) {
755 pr_err("internal bitmaps not supported by this kernel.\n");
758 if (!st
->ss
->add_internal_bitmap
) {
759 pr_err("internal bitmaps not supported with %s metadata\n",
763 if (!st
->ss
->add_internal_bitmap(st
, &s
->bitmap_chunk
,
764 c
->delay
, s
->write_behind
,
765 bitmapsize
, 1, major_num
)) {
766 pr_err("Given bitmap chunk size not supported.\n");
769 s
->bitmap_file
= NULL
;
772 sysfs_init(&info
, mdfd
, NULL
);
774 if (st
->ss
->external
&& st
->container_devnm
[0]) {
777 /* When creating a member, we need to be careful
778 * to negotiate with mdmon properly.
779 * If it is already running, we cannot write to
780 * the devices and must ask it to do that part.
781 * If it isn't running, we write to the devices,
783 * We hold an exclusive open on the container
784 * device to make sure mdmon doesn't exit after
785 * we checked that it is running.
787 * For now, fail if it is already running.
789 container_fd
= open_dev_excl(st
->container_devnm
);
790 if (container_fd
< 0) {
791 pr_err("Cannot get exclusive open on container - weird.\n");
794 if (mdmon_running(st
->container_devnm
)) {
796 pr_err("reusing mdmon for %s.\n",
797 st
->container_devnm
);
798 st
->update_tail
= &st
->updates
;
802 rv
= set_array_info(mdfd
, st
, &info
);
804 pr_err("failed to set array info for %s: %s\n",
805 mddev
, strerror(errno
));
809 if (s
->bitmap_file
) {
812 st
->ss
->uuid_from_super(st
, uuid
);
813 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
814 c
->delay
, s
->write_behind
,
819 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
821 pr_err("weird: %s cannot be openned\n",
825 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
826 pr_err("Cannot set bitmap file for %s: %s\n",
827 mddev
, strerror(errno
));
832 infos
= xmalloc(sizeof(*infos
) * total_slots
);
833 enable_fds(total_slots
);
834 for (pass
=1; pass
<=2 ; pass
++) {
835 struct mddev_dev
*moved_disk
= NULL
; /* the disk that was moved out of the insert point */
837 for (dnum
=0, dv
= devlist
; dv
;
838 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
841 struct mdinfo
*inf
= &infos
[dnum
];
843 if (dnum
>= total_slots
)
845 if (dnum
== insert_point
) {
849 if (strcasecmp(dv
->devname
, "missing")==0)
853 if (have_container
&& dnum
< info
.array
.raid_disks
- 1)
854 /* repeatedly use the container */
861 inf
->disk
.number
= dnum
;
862 inf
->disk
.raid_disk
= dnum
;
863 if (inf
->disk
.raid_disk
< s
->raiddisks
)
864 inf
->disk
.state
= (1<<MD_DISK_ACTIVE
) |
869 if (dv
->writemostly
== 1)
870 inf
->disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
875 if (st
->ss
->external
&&
876 st
->container_devnm
[0])
877 fd
= open(dv
->devname
, O_RDWR
);
879 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
);
882 pr_err("failed to open %s after earlier success - aborting\n",
887 inf
->disk
.major
= major(stb
.st_rdev
);
888 inf
->disk
.minor
= minor(stb
.st_rdev
);
891 remove_partitions(fd
);
892 if (st
->ss
->add_to_super(st
, &inf
->disk
,
895 ioctl(mdfd
, STOP_ARRAY
, NULL
);
898 st
->ss
->getinfo_super(st
, inf
, NULL
);
899 safe_mode_delay
= inf
->safe_mode_delay
;
901 if (have_container
&& c
->verbose
> 0)
902 pr_err("Using %s for device %d\n",
903 map_dev(inf
->disk
.major
,
907 if (!have_container
) {
908 /* getinfo_super might have lost these ... */
909 inf
->disk
.major
= major(stb
.st_rdev
);
910 inf
->disk
.minor
= minor(stb
.st_rdev
);
916 rv
= add_disk(mdfd
, st
, &info
, inf
);
919 pr_err("ADD_NEW_DISK for %s failed: %s\n",
920 dv
->devname
, strerror(errno
));
925 if (!have_container
&&
926 dv
== moved_disk
&& dnum
!= insert_point
) break;
929 struct mdinfo info_new
;
930 struct map_ent
*me
= NULL
;
932 /* check to see if the uuid has changed due to these
933 * metadata changes, and if so update the member array
934 * and container uuid. Note ->write_init_super clears
935 * the subarray cursor such that ->getinfo_super once
936 * again returns container info.
938 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
939 if (st
->ss
->external
&& s
->level
!= LEVEL_CONTAINER
&&
940 !same_uuid(info_new
.uuid
, info
.uuid
, 0)) {
941 map_update(&map
, fd2devnm(mdfd
),
942 info_new
.text_version
,
943 info_new
.uuid
, chosen_name
);
944 me
= map_by_devnm(&map
, st
->container_devnm
);
947 if (st
->ss
->write_init_super(st
)) {
948 st
->ss
->free_super(st
);
952 /* update parent container uuid */
954 char *path
= xstrdup(me
->path
);
956 st
->ss
->getinfo_super(st
, &info_new
, NULL
);
957 map_update(&map
, st
->container_devnm
,
958 info_new
.text_version
,
959 info_new
.uuid
, path
);
963 flush_metadata_updates(st
);
964 st
->ss
->free_super(st
);
970 if (s
->level
== LEVEL_CONTAINER
) {
971 /* No need to start. But we should signal udev to
973 sysfs_uevent(&info
, "change");
975 pr_err("container %s prepared.\n", mddev
);
976 wait_for(chosen_name
, mdfd
);
977 } else if (c
->runstop
== 1 || subdevs
>= s
->raiddisks
) {
978 if (st
->ss
->external
) {
982 case LEVEL_MULTIPATH
:
984 err
= sysfs_set_str(&info
, NULL
, "array_state",
991 err
= sysfs_set_str(&info
, NULL
, "array_state",
995 sysfs_set_safemode(&info
, safe_mode_delay
);
997 pr_err("failed to activate array.\n");
998 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1001 } else if (c
->readonly
&&
1002 sysfs_attribute_available(
1003 &info
, NULL
, "array_state")) {
1004 if (sysfs_set_str(&info
, NULL
,
1005 "array_state", "readonly") < 0) {
1006 pr_err("Failed to start array: %s\n",
1008 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1012 /* param is not actually used */
1014 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
1015 pr_err("RUN_ARRAY failed: %s\n",
1017 if (info
.array
.chunk_size
& (info
.array
.chunk_size
-1)) {
1018 cont_err("Problem may be that chunk size is not a power of 2\n");
1020 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1023 /* if start_ro module parameter is set, array is
1024 * auto-read-only, which is bad as the resync won't
1025 * start. So lets make it read-write now.
1027 ioctl(mdfd
, RESTART_ARRAY_RW
, NULL
);
1029 if (c
->verbose
>= 0)
1030 pr_err("array %s started.\n", mddev
);
1031 if (st
->ss
->external
&& st
->container_devnm
[0]) {
1033 start_mdmon(st
->container_devnm
);
1035 ping_monitor(st
->container_devnm
);
1036 close(container_fd
);
1038 wait_for(chosen_name
, mdfd
);
1040 pr_err("not starting array - not enough devices.\n");
1048 map_remove(&map
, fd2devnm(mdfd
));