2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
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>
28 static int name_matches(char *found
, char *required
, char *homehost
, int require_homehost
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
36 if (strcmp(found
, required
)==0)
38 sep
= strchr(found
, ':');
42 if (strncmp(found
, "any:", 4) == 0 ||
43 (homehost
&& strcmp(homehost
, "any") == 0) ||
45 (homehost
&& strlen(homehost
) == l
&&
46 strncmp(found
, homehost
, l
) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep
+1, required
) == 0)
54 static int is_member_busy(char *metadata_version
)
56 /* check if the given member array is active */
57 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
58 struct mdstat_ent
*ent
;
61 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
62 if (ent
->metadata_version
== NULL
)
64 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
66 if (!is_subarray(&ent
->metadata_version
[9]))
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
79 static int ident_matches(struct mddev_ident
*ident
,
80 struct mdinfo
*content
,
81 struct supertype
*tst
,
82 char *homehost
, int require_homehost
,
83 char *update
, char *devname
)
86 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
87 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0 &&
88 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
90 pr_err("%s has wrong uuid.\n", devname
);
93 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
94 name_matches(content
->name
, ident
->name
, homehost
, require_homehost
)==0) {
96 pr_err("%s has wrong name.\n", devname
);
99 if (ident
->super_minor
!= UnSet
&&
100 ident
->super_minor
!= content
->array
.md_minor
) {
102 pr_err("%s has wrong super-minor.\n",
106 if (ident
->level
!= UnSet
&&
107 ident
->level
!= content
->array
.level
) {
109 pr_err("%s has wrong raid level.\n",
113 if (ident
->raid_disks
!= UnSet
&&
114 content
->array
.raid_disks
!= 0 && /* metadata doesn't know how many to expect */
115 ident
->raid_disks
!= content
->array
.raid_disks
) {
117 pr_err("%s requires wrong number of drives.\n",
121 if (ident
->member
&& ident
->member
[0]) {
122 /* content->text_version must match */
123 char *s
= strchr(content
->text_version
+1, '/');
126 pr_err("%s is not a container and one is required.\n",
129 } else if (strcmp(ident
->member
, s
+1) != 0) {
131 pr_err("skipping wrong member %s is %s\n",
132 content
->text_version
, devname
);
139 static int select_devices(struct mddev_dev
*devlist
,
140 struct mddev_ident
*ident
,
141 struct supertype
**stp
,
142 struct mdinfo
**contentp
,
144 int inargv
, int auto_assem
)
146 struct mddev_dev
*tmpdev
;
148 struct supertype
*st
= *stp
;
149 struct mdinfo
*content
= NULL
;
150 int report_mismatch
= ((inargv
&& c
->verbose
>= 0) || c
->verbose
> 0);
151 struct domainlist
*domains
= NULL
;
154 tmpdev
= devlist
; num_devs
= 0;
160 tmpdev
->disposition
= 0;
161 tmpdev
= tmpdev
->next
;
164 /* first walk the list of devices to find a consistent set
165 * that match the criterea, if that is possible.
166 * We flag the ones we like with 'used'.
168 for (tmpdev
= devlist
;
170 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
171 char *devname
= tmpdev
->devname
;
173 struct supertype
*tst
;
174 struct dev_policy
*pol
= NULL
;
175 int found_container
= 0;
177 if (tmpdev
->used
> 1)
180 if (ident
->container
) {
181 if (ident
->container
[0] == '/' &&
182 !same_dev(ident
->container
, devname
)) {
184 pr_err("%s is not the container required (%s)\n",
185 devname
, ident
->container
);
188 } else if (ident
->devices
&&
189 !match_oneof(ident
->devices
, devname
)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
195 pr_err("%s is not one of %s\n", devname
, ident
->devices
);
201 dfd
= dev_open(devname
, O_RDONLY
);
204 pr_err("cannot open device %s: %s\n",
205 devname
, strerror(errno
));
207 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
209 } else if (must_be_container(dfd
)) {
211 /* already found some components, this cannot
215 pr_err("%s is a container, but we are looking for components\n",
218 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
220 pr_err("not a recognisable container: %s\n",
223 } else if (!tst
->ss
->load_container
||
224 tst
->ss
->load_container(tst
, dfd
, NULL
)) {
226 pr_err("no correct container type: %s\n",
229 } else if (auto_assem
&&
230 !conf_test_metadata(tst
->ss
->name
,
231 (pol
= devid_policy(rdev
)),
232 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
234 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
235 devname
, tst
->ss
->name
);
240 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
242 pr_err("no recogniseable superblock on %s\n",
245 } else if ((tst
->ignore_hw_compat
= 0),
246 tst
->ss
->load_super(tst
, dfd
,
247 report_mismatch
? devname
: NULL
)) {
249 pr_err("no RAID superblock on %s\n",
252 } else if (tst
->ss
->compare_super
== NULL
) {
254 pr_err("Cannot assemble %s metadata on %s\n",
255 tst
->ss
->name
, devname
);
257 } else if (auto_assem
&& st
== NULL
&&
258 !conf_test_metadata(tst
->ss
->name
,
259 (pol
= devid_policy(rdev
)),
260 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
262 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
263 devname
, tst
->ss
->name
);
267 if (dfd
>= 0) close(dfd
);
268 if (tmpdev
->used
== 2) {
269 if (auto_assem
|| !inargv
)
270 /* Ignore unrecognised devices during auto-assembly */
272 if (ident
->name
[0] ||
273 ident
->super_minor
!= UnSet
)
274 /* Ignore unrecognised device if looking for
278 pr_err("%s has no superblock - assembly aborted\n",
281 st
->ss
->free_super(st
);
282 dev_policy_free(pol
);
283 domain_free(domains
);
285 tst
->ss
->free_super(tst
);
289 if (found_container
) {
290 /* tmpdev is a container. We need to be either
291 * looking for a member, or auto-assembling
293 /* should be safe to try an exclusive open now, we
294 * have rejected anything that some other mdadm might
297 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
300 pr_err("%s is busy - skipping\n", devname
);
305 if (ident
->container
&& ident
->container
[0] != '/') {
310 tst
->ss
->getinfo_super(tst
, content
, NULL
);
312 if (!parse_uuid(ident
->container
, uuid
) ||
313 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
315 pr_err("%s has wrong UUID to be required container\n",
320 /* It is worth looking inside this container.
323 pr_err("looking in container %s\n",
326 for (content
= tst
->ss
->container_content(tst
, NULL
);
328 content
= content
->next
) {
330 if (!ident_matches(ident
, content
, tst
,
331 c
->homehost
, c
->require_homehost
,
333 report_mismatch
? devname
: NULL
))
334 /* message already printed */;
335 else if (is_member_busy(content
->text_version
)) {
337 pr_err("member %s in %s is already assembled\n",
338 content
->text_version
,
340 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
341 /* do not assemble arrays with unsupported configurations */
342 pr_err("Cannot activate member %s in %s.\n",
343 content
->text_version
,
350 goto loop
; /* empty container */
353 st
= tst
; tst
= NULL
;
354 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
355 pr_err("%s is a container, but is not only device given: confused and aborting\n",
357 st
->ss
->free_super(st
);
358 dev_policy_free(pol
);
359 domain_free(domains
);
363 pr_err("found match on member %s in %s\n",
364 content
->text_version
, devname
);
366 /* make sure we finished the loop */
371 tst
->ss
->getinfo_super(tst
, content
, NULL
);
373 if (!ident_matches(ident
, content
, tst
,
374 c
->homehost
, c
->require_homehost
,
376 report_mismatch
? devname
: NULL
))
380 /* Never auto-assemble things that conflict
381 * with mdadm.conf in some way
383 struct mddev_ident
*match
;
386 match
= conf_match(tst
, content
, devname
,
387 report_mismatch
? c
->verbose
: -1,
389 if (!match
&& rv
== 2)
391 if (match
&& match
->devname
&&
392 strcasecmp(match
->devname
, "<ignore>") == 0) {
394 pr_err("%s is a member of an explicitly ignored array\n",
398 if (match
&& !ident_matches(match
, content
, tst
,
399 c
->homehost
, c
->require_homehost
,
401 report_mismatch
? devname
: NULL
))
402 /* Array exists in mdadm.conf but some
403 * details don't match, so reject it
408 /* should be safe to try an exclusive open now, we
409 * have rejected anything that some other mdadm might
412 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
415 pr_err("%s is busy - skipping\n", devname
);
422 if (st
->minor_version
== -1)
423 st
->minor_version
= tst
->minor_version
;
425 if (memcmp(content
->uuid
, uuid_zero
,
426 sizeof(int[4])) == 0) {
427 /* this is a floating spare. It cannot define
428 * an array unless there are no more arrays of
429 * this type to be found. It can be included
430 * in an array of this type though.
436 if (st
->ss
!= tst
->ss
||
437 st
->minor_version
!= tst
->minor_version
||
438 st
->ss
->compare_super(st
, tst
, 1) != 0) {
439 /* Some mismatch. If exactly one array matches this host,
440 * we can resolve on that one.
441 * Or, if we are auto assembling, we just ignore the second
447 int first
= st
->ss
->match_home(st
, c
->homehost
);
448 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
450 (first
== 1 || last
== 1)) {
451 /* We can do something */
452 if (first
) {/* just ignore this one */
454 pr_err("%s misses out due to wrong homehost\n",
457 } else { /* reject all those sofar */
458 struct mddev_dev
*td
;
460 pr_err("%s overrides previous devices due to good homehost\n",
462 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
470 pr_err("superblock on %s doesn't match others - assembly aborted\n",
472 tst
->ss
->free_super(tst
);
473 st
->ss
->free_super(st
);
474 dev_policy_free(pol
);
475 domain_free(domains
);
481 /* Collect domain information from members only */
482 if (tmpdev
&& tmpdev
->used
== 1) {
484 pol
= devid_policy(rdev
);
485 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
487 dev_policy_free(pol
);
490 tst
->ss
->free_super(tst
);
493 /* Check if we found some imsm spares but no members */
496 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
498 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
499 if (tmpdev
->used
!= 3)
505 /* we need sb from one of the spares */
506 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
508 st
->ss
->load_super(st
, dfd
, NULL
))
515 /* Now reject spares that don't match domains of identified members */
516 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
517 if (tmpdev
->used
!= 3)
519 if (!stat_is_blkdev(tmpdev
->devname
, &rdev
)) {
522 struct dev_policy
*pol
= devid_policy(rdev
);
523 int dt
= domain_test(domains
, pol
, NULL
);
524 if (inargv
&& dt
!= 0)
525 /* take this spare as domains match
526 * if there are any */
528 else if (!inargv
&& dt
== 1)
529 /* device wasn't explicitly listed, so need
530 * explicit domain match - which we have */
533 /* if domains don't match mark as unused */
535 dev_policy_free(pol
);
538 domain_free(domains
);
540 if (st
&& st
->sb
&& content
== *contentp
)
541 st
->ss
->getinfo_super(st
, content
, NULL
);
549 int uptodate
; /* set once we decide that this device is as
550 * recent as everything else in the array.
552 int included
; /* set if the device is already in the array
553 * due to a previous '-I'
558 static int load_devices(struct devs
*devices
, char *devmap
,
559 struct mddev_ident
*ident
, struct supertype
**stp
,
560 struct mddev_dev
*devlist
, struct context
*c
,
561 struct mdinfo
*content
,
562 int mdfd
, char *mddev
,
563 int *most_recentp
, int *bestcntp
, int **bestp
,
566 struct mddev_dev
*tmpdev
;
570 int most_recent
= -1;
573 struct supertype
*st
= *stp
;
575 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
576 char *devname
= tmpdev
->devname
;
578 struct supertype
*tst
;
583 if (tmpdev
->used
!= 1)
585 /* looks like a good enough match to update the super block if needed */
587 /* prepare useful information in info structures */
592 if (strcmp(c
->update
, "uuid") == 0 && !ident
->uuid_set
)
593 random_uuid((__u8
*)ident
->uuid
);
595 if (strcmp(c
->update
, "ppl") == 0 &&
596 ident
->bitmap_fd
>= 0) {
597 pr_err("PPL is not compatible with bitmap\n");
604 dfd
= dev_open(devname
,
605 tmpdev
->disposition
== 'I'
606 ? O_RDWR
: (O_RDWR
|O_EXCL
));
609 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
610 pr_err("cannot re-read metadata from %s - aborting\n",
617 tst
->ss
->free_super(tst
);
622 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
624 memcpy(content
->uuid
, ident
->uuid
, 16);
625 strcpy(content
->name
, ident
->name
);
626 content
->array
.md_minor
= minor(stb2
.st_rdev
);
628 if (strcmp(c
->update
, "byteorder") == 0)
630 else if (strcmp(c
->update
, "home-cluster") == 0) {
631 tst
->cluster_name
= c
->homecluster
;
632 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
633 } else if (strcmp(c
->update
, "nodes") == 0) {
634 tst
->nodes
= c
->nodes
;
635 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
636 } else if (strcmp(c
->update
, "revert-reshape") == 0 &&
638 err
= tst
->ss
->update_super(tst
, content
,
639 "revert-reshape-nobackup",
644 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
650 pr_err("--update=%s not understood for %s metadata\n",
651 c
->update
, tst
->ss
->name
);
652 tst
->ss
->free_super(tst
);
661 if (strcmp(c
->update
, "uuid")==0 &&
664 memcpy(ident
->uuid
, content
->uuid
, 16);
666 if (tst
->ss
->store_super(tst
, dfd
))
667 pr_err("Could not re-write superblock on %s.\n",
670 if (strcmp(c
->update
, "uuid")==0 &&
671 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
672 if (bitmap_update_uuid(ident
->bitmap_fd
,
674 tst
->ss
->swapuuid
) != 0)
675 pr_err("Could not update uuid on external bitmap.\n");
680 dfd
= dev_open(devname
,
681 tmpdev
->disposition
== 'I'
682 ? O_RDWR
: (O_RDWR
|O_EXCL
));
685 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
686 pr_err("cannot re-read metadata from %s - aborting\n",
693 tst
->ss
->free_super(tst
);
698 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
705 pr_err("%s is identified as a member of %s, slot %d%s.\n",
706 devname
, mddev
, content
->disk
.raid_disk
,
707 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
708 devices
[devcnt
].devname
= devname
;
709 devices
[devcnt
].uptodate
= 0;
710 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
711 devices
[devcnt
].i
= *content
;
712 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
713 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
715 disk_state
= devices
[devcnt
].i
.disk
.state
& ~((1<<MD_DISK_FAILFAST
) |
716 (1<<MD_DISK_WRITEMOSTLY
));
717 if (disk_state
== ((1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
))) {
718 if (most_recent
< 0 ||
719 devices
[devcnt
].i
.events
720 > devices
[most_recent
].i
.events
) {
721 struct supertype
*tmp
= tst
;
724 most_recent
= devcnt
;
727 tst
->ss
->free_super(tst
);
730 if (content
->array
.level
== LEVEL_MULTIPATH
)
731 /* with multipath, the raid_disk from the superblock is meaningless */
734 i
= devices
[devcnt
].i
.disk
.raid_disk
;
735 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
736 if (nextspare
< content
->array
.raid_disks
*2)
737 nextspare
= content
->array
.raid_disks
*2;
740 /* i is raid_disk - double it so there is room for
743 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
745 if (i
>= content
->array
.raid_disks
*2 &&
751 int newbestcnt
= i
+10;
752 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
754 for (c
=0; c
< newbestcnt
; c
++)
756 newbest
[c
] = best
[c
];
761 bestcnt
= newbestcnt
;
764 devices
[best
[i
]].i
.events
==
765 devices
[devcnt
].i
.events
&&
766 (devices
[best
[i
]].i
.disk
.minor
!=
767 devices
[devcnt
].i
.disk
.minor
) &&
769 content
->array
.level
!= LEVEL_MULTIPATH
) {
770 /* two different devices with identical superblock.
771 * Could be a mis-detection caused by overlapping
772 * partitions. fail-safe.
774 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
775 " If they are really different, please --zero the superblock on one\n"
776 " If they are the same or overlap, please remove one from %s.\n",
777 devices
[best
[i
]].devname
, devname
,
778 inargv
? "the list" :
779 "the\n DEVICE list in mdadm.conf"
787 if (best
[i
] == -1 || (devices
[best
[i
]].i
.events
788 < devices
[devcnt
].i
.events
))
790 else if (st
->ss
== &super_imsm
)
795 if (most_recent
>= 0)
796 *most_recentp
= most_recent
;
803 static int force_array(struct mdinfo
*content
,
804 struct devs
*devices
,
805 int *best
, int bestcnt
, char *avail
,
807 struct supertype
*st
,
811 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
812 content
->array
.layout
, 1,
814 (content
->reshape_active
&& content
->delta_disks
> 0 &&
815 !enough(content
->array
.level
, (content
->array
.raid_disks
816 - content
->delta_disks
),
817 content
->new_layout
, 1, avail
))) {
818 /* Choose the newest best drive which is
819 * not up-to-date, update the superblock
823 struct supertype
*tst
;
824 unsigned long long current_events
;
825 int chosen_drive
= -1;
829 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
834 if (devices
[j
].uptodate
)
836 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
838 if (!devices
[j
].i
.reshape_active
||
839 devices
[j
].i
.delta_disks
<= 0)
841 /* When increasing number of devices, an
842 * added device also appears to be
843 * recovering. It is safe to include it
844 * as long as it won't be a source of
846 * For now, just allow for last data
847 * devices in RAID4 or last devices in RAID4/5/6.
849 delta
= devices
[j
].i
.delta_disks
;
850 if (devices
[j
].i
.array
.level
>= 4 &&
851 devices
[j
].i
.array
.level
<= 6 &&
852 i
/2 >= content
->array
.raid_disks
- delta
)
854 else if (devices
[j
].i
.array
.level
== 4 &&
855 i
/2 >= content
->array
.raid_disks
- delta
- 1)
859 } else if (devices
[j
].i
.reshape_active
!=
860 content
->reshape_active
||
861 (devices
[j
].i
.reshape_active
&&
862 devices
[j
].i
.reshape_progress
!=
863 content
->reshape_progress
))
864 /* Here, it may be a source of data. If two
865 * devices claim different progresses, it
866 * means that reshape boundaries differ for
867 * their own devices. Kernel will only treat
868 * the first one as reshape progress and
869 * go on. It may cause disaster, so avoid it.
872 if (chosen_drive
< 0 ||
874 > devices
[chosen_drive
].i
.events
)
877 if (chosen_drive
< 0)
879 current_events
= devices
[chosen_drive
].i
.events
;
882 pr_err("forcing event count in %s(%d) from %d up to %d\n",
883 devices
[chosen_drive
].devname
,
884 devices
[chosen_drive
].i
.disk
.raid_disk
,
885 (int)(devices
[chosen_drive
].i
.events
),
886 (int)(devices
[most_recent
].i
.events
));
887 fd
= dev_open(devices
[chosen_drive
].devname
,
888 devices
[chosen_drive
].included
? O_RDWR
891 pr_err("Couldn't open %s for write - not updating\n",
892 devices
[chosen_drive
].devname
);
893 devices
[chosen_drive
].i
.events
= 0;
897 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
899 pr_err("RAID superblock disappeared from %s - not updating.\n",
900 devices
[chosen_drive
].devname
);
901 devices
[chosen_drive
].i
.events
= 0;
904 content
->events
= devices
[most_recent
].i
.events
;
905 tst
->ss
->update_super(tst
, content
, "force-one",
906 devices
[chosen_drive
].devname
, c
->verbose
,
909 if (tst
->ss
->store_super(tst
, fd
)) {
911 pr_err("Could not re-write superblock on %s\n",
912 devices
[chosen_drive
].devname
);
913 devices
[chosen_drive
].i
.events
= 0;
914 tst
->ss
->free_super(tst
);
918 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
919 devices
[chosen_drive
].uptodate
= 1;
920 avail
[chosen_drive
] = 1;
922 tst
->ss
->free_super(tst
);
923 /* If there are any other drives of the same vintage,
924 * add them in as well. We can't lose and we might gain
927 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
931 !devices
[j
].uptodate
&&
932 devices
[j
].i
.recovery_start
== MaxSector
&&
933 devices
[j
].i
.events
== current_events
&&
934 ((!devices
[j
].i
.reshape_active
&&
935 !content
->reshape_active
) ||
936 (devices
[j
].i
.reshape_active
==
937 content
->reshape_active
&&
938 devices
[j
].i
.reshape_progress
==
939 content
->reshape_progress
))) {
948 static int start_array(int mdfd
,
950 struct mdinfo
*content
,
951 struct supertype
*st
,
952 struct mddev_ident
*ident
,
953 int *best
, int bestcnt
,
955 struct devs
*devices
,
957 unsigned int sparecnt
,
958 unsigned int rebuilding_cnt
,
959 unsigned int journalcnt
,
961 int clean
, char *avail
,
962 int start_partial_ok
,
969 unsigned int req_cnt
;
971 if (content
->journal_device_required
&& (content
->journal_clean
== 0)) {
973 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
976 pr_err("Journal is missing or stale, starting array read only.\n");
980 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
983 rv
= set_array_info(mdfd
, st
, content
);
985 pr_err("failed to set array info for %s: %s\n",
986 mddev
, strerror(errno
));
989 if (ident
->bitmap_fd
>= 0) {
990 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
991 pr_err("SET_BITMAP_FILE failed.\n");
994 } else if (ident
->bitmap_file
) {
995 /* From config file */
996 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
998 pr_err("Could not open bitmap file %s\n",
1002 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1003 pr_err("Failed to set bitmapfile for %s\n", mddev
);
1010 /* First, add the raid disks, but add the chosen one last */
1011 for (i
= 0; i
<= bestcnt
; i
++) {
1015 if (j
== chosen_drive
)
1020 if (j
>= 0 && !devices
[j
].included
) {
1023 dfd
= dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1025 remove_partitions(dfd
);
1028 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1031 pr_err("failed to add %s to %s: %s\n",
1032 devices
[j
].devname
, mddev
,
1034 if (errno
== EINVAL
&& content
->array
.level
== 0 &&
1035 content
->array
.layout
!= 0) {
1036 cont_err("Possibly your kernel doesn't support RAID0 layouts.\n");
1037 cont_err("Please upgrade.\n");
1039 if (i
< content
->array
.raid_disks
* 2 ||
1044 } else if (c
->verbose
> 0) {
1045 pr_err("added %s to %s as %d%s%s\n",
1046 devices
[j
].devname
, mddev
,
1047 devices
[j
].i
.disk
.raid_disk
,
1048 devices
[j
].uptodate
?"":
1049 " (possibly out of date)",
1050 (devices
[j
].i
.disk
.state
&
1051 (1<<MD_DISK_REPLACEMENT
)) ?
1054 } else if (j
>= 0) {
1056 pr_err("%s is already in %s as %d\n",
1057 devices
[j
].devname
, mddev
,
1058 devices
[j
].i
.disk
.raid_disk
);
1059 } else if (c
->verbose
> 0 &&
1060 i
< content
->array
.raid_disks
* 2 && (i
& 1) == 0)
1061 pr_err("no uptodate device for slot %d of %s\n",
1065 if (content
->array
.level
== LEVEL_CONTAINER
) {
1066 if (c
->verbose
>= 0) {
1067 pr_err("Container %s has been assembled with %d drive%s",
1068 mddev
, okcnt
+ sparecnt
+ journalcnt
,
1069 okcnt
+ sparecnt
+ journalcnt
== 1 ? "" : "s");
1070 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1071 fprintf(stderr
, " (out of %d)\n",
1072 content
->array
.raid_disks
);
1074 fprintf(stderr
, "\n");
1075 sysfs_rules_apply(mddev
, content
);
1079 if (st
->ss
->validate_container
) {
1080 struct mdinfo
*devices_list
;
1081 struct mdinfo
*info_devices
;
1084 devices_list
= NULL
;
1085 info_devices
= xmalloc(sizeof(struct mdinfo
) *
1086 (okcnt
+ sparecnt
));
1087 for (count
= 0; count
< okcnt
+ sparecnt
; count
++) {
1088 info_devices
[count
] = devices
[count
].i
;
1089 info_devices
[count
].next
= devices_list
;
1090 devices_list
= &info_devices
[count
];
1092 if (st
->ss
->validate_container(devices_list
))
1093 pr_err("Mismatch detected!\n");
1097 st
->ss
->free_super(st
);
1098 sysfs_uevent(content
, "change");
1099 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1100 /* Was partial, is still partial, so signal an error
1101 * to ensure we don't retry */
1106 /* Get number of in-sync devices according to the superblock.
1107 * We must have this number to start the array without -s or -R
1109 req_cnt
= content
->array
.working_disks
;
1111 if (c
->runstop
== 1 ||
1113 (enough(content
->array
.level
, content
->array
.raid_disks
,
1114 content
->array
.layout
, clean
, avail
) &&
1115 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)))) {
1116 /* This array is good-to-go.
1117 * If a reshape is in progress then we might need to
1118 * continue monitoring it. In that case we start
1119 * it read-only and let the grow code make it writable.
1123 if (content
->reshape_active
&&
1124 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1125 content
->delta_disks
<= 0) {
1126 if (!c
->backup_file
) {
1127 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1129 pr_err("Please provided one with \"--backup-file=...\"\n");
1131 strcmp(c
->update
, "revert-reshape") == 0)
1132 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1135 rv
= sysfs_set_str(content
, NULL
,
1136 "array_state", "readonly");
1138 rv
= Grow_continue(mdfd
, st
, content
,
1141 } else if (c
->readonly
&&
1142 sysfs_attribute_available(content
, NULL
,
1144 rv
= sysfs_set_str(content
, NULL
,
1145 "array_state", "readonly");
1147 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1148 reopen_mddev(mdfd
); /* drop O_EXCL */
1150 sysfs_rules_apply(mddev
, content
);
1151 if (c
->verbose
>= 0) {
1152 pr_err("%s has been started with %d drive%s",
1153 mddev
, okcnt
, okcnt
==1?"":"s");
1154 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1155 fprintf(stderr
, " (out of %d)",
1156 content
->array
.raid_disks
);
1158 fprintf(stderr
, "%s %d rebuilding",
1159 sparecnt
?",":" and",
1162 fprintf(stderr
, " and %d spare%s",
1164 sparecnt
== 1 ? "" : "s");
1165 if (content
->journal_clean
)
1166 fprintf(stderr
, " and %d journal",
1168 fprintf(stderr
, ".\n");
1170 if (content
->reshape_active
&&
1171 content
->array
.level
>= 4 &&
1172 content
->array
.level
<= 6) {
1173 /* might need to increase the size
1174 * of the stripe cache - default is 256
1176 int chunk_size
= content
->array
.chunk_size
;
1178 if (content
->reshape_active
&&
1179 content
->new_chunk
> chunk_size
)
1180 chunk_size
= content
->new_chunk
;
1181 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1184 sra
= sysfs_read(mdfd
, NULL
, 0);
1186 sysfs_set_num(sra
, NULL
,
1187 "stripe_cache_size",
1188 (4 * chunk_size
/ 4096) + 1);
1192 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1193 /* If any devices did not get added
1194 * because the kernel rejected them based
1195 * on event count, try adding them
1196 * again providing the action policy is
1197 * 're-add' or greater. The bitmap
1198 * might allow them to be included, or
1199 * they will become spares.
1201 for (i
= 0; i
< bestcnt
; i
++) {
1203 if (j
>= 0 && !devices
[j
].uptodate
) {
1204 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1206 rv
= add_disk(mdfd
, st
, content
,
1208 if (rv
== 0 && c
->verbose
>= 0)
1209 pr_err("%s has been re-added.\n",
1210 devices
[j
].devname
);
1214 if (content
->array
.level
== 6 &&
1215 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1219 sra
= sysfs_read(mdfd
, NULL
, 0);
1221 sysfs_set_str(sra
, NULL
,
1222 "sync_action", "repair");
1227 pr_err("failed to RUN_ARRAY %s: %s\n", mddev
, strerror(errno
));
1228 if (errno
== 524 /* ENOTSUP */ &&
1229 content
->array
.level
== 0 && content
->array
.layout
== 0)
1230 cont_err("Please use --update=layout-original or --update=layout-alternate\n");
1232 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1233 content
->array
.layout
, 1, avail
))
1234 pr_err("Not enough devices to start the array.\n");
1235 else if (!enough(content
->array
.level
,
1236 content
->array
.raid_disks
,
1237 content
->array
.layout
, clean
, avail
))
1238 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1242 if (c
->runstop
== -1) {
1243 pr_err("%s assembled from %d drive%s",
1244 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1245 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1246 fprintf(stderr
, " (out of %d)",
1247 content
->array
.raid_disks
);
1248 fprintf(stderr
, ", but not started.\n");
1251 if (c
->verbose
>= -1) {
1252 pr_err("%s assembled from %d drive%s",
1253 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1255 fprintf(stderr
, "%s %d rebuilding",
1256 sparecnt
? "," : " and", rebuilding_cnt
);
1258 fprintf(stderr
, " and %d spare%s", sparecnt
,
1259 sparecnt
== 1 ? "" : "s");
1260 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1261 content
->array
.layout
, 1, avail
))
1262 fprintf(stderr
, " - not enough to start the array.\n");
1263 else if (!enough(content
->array
.level
,
1264 content
->array
.raid_disks
,
1265 content
->array
.layout
, clean
, avail
))
1266 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1268 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1269 fprintf(stderr
, " - need all %d to start it",
1272 fprintf(stderr
, " - need %d to start", req_cnt
);
1273 fprintf(stderr
, " (use --run to insist).\n");
1279 int Assemble(struct supertype
*st
, char *mddev
,
1280 struct mddev_ident
*ident
,
1281 struct mddev_dev
*devlist
,
1285 * The task of Assemble is to find a collection of
1286 * devices that should (according to their superblocks)
1287 * form an array, and to give this collection to the MD driver.
1288 * In Linux-2.4 and later, this involves submitting a
1289 * SET_ARRAY_INFO ioctl with no arg - to prepare
1290 * the array - and then submit a number of
1291 * ADD_NEW_DISK ioctls to add disks into
1292 * the array. Finally RUN_ARRAY might
1293 * be submitted to start the array.
1295 * Much of the work of Assemble is in finding and/or
1296 * checking the disks to make sure they look right.
1298 * If mddev is not set, then scan must be set and we
1299 * read through the config file for dev+uuid mapping
1300 * We recurse, setting mddev, for each device that
1302 * - has a valid uuid (or any uuid if !uuidset)
1304 * If mddev is set, we try to determine state of md.
1305 * check version - must be at least 0.90.0
1306 * check kernel version. must be at least 2.4.
1307 * If not, we can possibly fall back on START_ARRAY
1308 * Try to GET_ARRAY_INFO.
1309 * If possible, give up
1310 * If not, try to STOP_ARRAY just to make sure
1312 * If !uuidset and scan, look in conf-file for uuid
1313 * If not found, give up
1314 * If !devlist and scan and uuidset, get list of devs from conf-file
1317 * Check superblock - discard if bad
1318 * Check uuid (set if we don't have one) - discard if no match
1319 * Check superblock similarity if we have a superblock - discard if different
1320 * Record events, devicenum
1321 * This should give us a list of devices for the array
1322 * We should collect the most recent event number
1324 * Count disks with recent enough event count
1325 * While force && !enough disks
1326 * Choose newest rejected disks, update event count
1327 * mark clean and rewrite superblock
1330 * foreach device with recent events : ADD_NEW_DISK
1331 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1333 * Check the device numbers in superblock are right
1334 * update superblock if any changes
1341 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1342 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0 &&
1343 (ident
->container
== NULL
|| ident
->member
== NULL
));
1344 struct devs
*devices
= NULL
;
1346 int *best
= NULL
; /* indexed by raid_disk */
1349 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1350 int journal_clean
= 0;
1353 int most_recent
= 0;
1357 int start_partial_ok
= (c
->runstop
>= 0) &&
1358 (c
->force
|| devlist
==NULL
|| auto_assem
);
1360 struct mddev_dev
*tmpdev
;
1362 struct mdinfo
*content
= NULL
;
1363 struct mdinfo
*pre_exist
= NULL
;
1366 char chosen_name
[1024];
1367 struct map_ent
*map
= NULL
;
1371 * If any subdevs are listed, then any that don't
1372 * match ident are discarded. Remainder must all match and
1374 * If no subdevs, then we scan all devices in the config file, but
1375 * there must be something in the identity
1379 ident
->uuid_set
== 0 &&
1380 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1381 ident
->name
[0] == 0 &&
1382 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1383 ident
->devices
== NULL
) {
1384 pr_err("No identity information available for %s - cannot assemble.\n",
1385 mddev
? mddev
: "further assembly");
1389 if (devlist
== NULL
)
1390 devlist
= conf_get_devs();
1395 /* We come back here when doing auto-assembly and attempting some
1396 * set of devices failed. Those are now marked as ->used==2 and
1397 * we ignore them and try again
1399 if (!st
&& ident
->st
)
1402 pr_err("looking for devices for %s\n",
1403 mddev
? mddev
: "further assembly");
1407 st
->ignore_hw_compat
= 1;
1408 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1409 inargv
, auto_assem
);
1413 if (!st
|| !st
->sb
|| !content
)
1416 /* We have a full set of devices - we now need to find the
1418 * However there is a risk that we are racing with "mdadm -I"
1419 * and the array is already partially assembled - we will have
1420 * rejected any devices already in this address.
1421 * So we take a lock on the map file - to prevent further races -
1422 * and look for the uuid in there. If found and the array is
1423 * active, we abort. If found and the array is not active
1424 * we commit to that md device and add all the contained devices
1425 * to our list. We flag them so that we don't try to re-add,
1426 * but can remove if they turn out to not be wanted.
1429 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1430 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1433 mp
= map_by_uuid(&map
, content
->uuid
);
1436 /* array already exists. */
1437 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1438 if (pre_exist
->array
.level
!= UnSet
) {
1439 pr_err("Found some drive for an array that is already active: %s\n",
1441 pr_err("giving up.\n");
1444 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1445 /* We want to add this device to our list,
1446 * but it could already be there if "mdadm -I"
1447 * started *after* we checked for O_EXCL.
1448 * If we add it to the top of the list
1449 * it will be preferred over later copies.
1451 struct mddev_dev
*newdev
;
1452 char *devname
= map_dev(dv
->disk
.major
,
1457 newdev
= xmalloc(sizeof(*newdev
));
1458 newdev
->devname
= devname
;
1459 newdev
->disposition
= 'I';
1461 newdev
->next
= devlist
;
1465 strcpy(chosen_name
, mp
->path
);
1466 if (c
->verbose
> 0 || mddev
== NULL
||
1467 strcmp(mddev
, chosen_name
) != 0)
1468 pr_err("Merging with already-assembled %s\n",
1470 mdfd
= open_dev_excl(mp
->devnm
);
1472 int trustworthy
= FOREIGN
;
1473 name
= content
->name
;
1474 switch (st
->ss
->match_home(st
, c
->homehost
)
1475 ?: st
->ss
->match_home(st
, "any")) {
1477 trustworthy
= LOCAL
;
1478 name
= strchr(content
->name
, ':');
1482 name
= content
->name
;
1485 if (mddev
&& map_by_name(&map
, mddev
) != NULL
) {
1486 pr_err("Cannot create device with %s because is in use\n", mddev
);
1490 /* If the array is listed in mdadm.conf or on
1491 * command line, then we trust the name
1492 * even if the array doesn't look local
1494 trustworthy
= LOCAL
;
1497 content
->array
.level
== LEVEL_CONTAINER
) {
1498 name
= content
->text_version
;
1499 trustworthy
= METADATA
;
1502 if (name
[0] && trustworthy
!= LOCAL
&&
1503 ! c
->require_homehost
&&
1504 conf_name_is_free(name
))
1505 trustworthy
= LOCAL
;
1507 if (trustworthy
== LOCAL
&&
1509 /* Ignore 'host:' prefix of name */
1510 name
= strchr(name
, ':')+1;
1512 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1516 st
->ss
->free_super(st
);
1521 mddev
= chosen_name
;
1522 if (pre_exist
== NULL
) {
1523 if (mddev_busy(fd2devnm(mdfd
))) {
1524 pr_err("%s already active, cannot restart it!\n",
1526 for (tmpdev
= devlist
;
1527 tmpdev
&& tmpdev
->used
!= 1;
1528 tmpdev
= tmpdev
->next
)
1530 if (tmpdev
&& auto_assem
)
1531 pr_err("%s needed for %s...\n",
1532 mddev
, tmpdev
->devname
);
1535 st
->ss
->free_super(st
);
1540 /* just incase it was started but has no content */
1541 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1544 if (content
!= &info
) {
1545 /* This is a member of a container. Try starting the array. */
1547 err
= assemble_container_content(st
, mdfd
, content
, c
,
1553 /* Ok, no bad inconsistancy, we can try updating etc */
1554 devices
= xcalloc(num_devs
, sizeof(*devices
));
1555 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1556 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1557 c
, content
, mdfd
, mddev
,
1558 &most_recent
, &bestcnt
, &best
, inargv
);
1562 * devices is already freed in load_devices, so set devices
1563 * to NULL to avoid double free devices.
1570 pr_err("no devices found for %s\n",
1573 st
->ss
->free_super(st
);
1578 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1579 st
->minor_version
= 90;
1581 st
->ss
->getinfo_super(st
, content
, NULL
);
1582 clean
= content
->array
.state
& 1;
1584 /* now we have some devices that might be suitable.
1587 avail
= xcalloc(content
->array
.raid_disks
, 1);
1593 for (i
=0; i
< bestcnt
; i
++) {
1595 int event_margin
= 1; /* always allow a difference of '1'
1596 * like the kernel does
1598 if (j
< 0) continue;
1599 /* note: we ignore error flags in multipath arrays
1600 * as they don't make sense
1602 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1603 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1604 if (content
->journal_device_required
)
1606 else /* unexpected journal, mark as faulty */
1607 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1608 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1609 if (!(devices
[j
].i
.disk
.state
1610 & (1<<MD_DISK_FAULTY
))) {
1611 devices
[j
].uptodate
= 1;
1617 /* If this device thinks that 'most_recent' has failed, then
1618 * we must reject this device.
1620 if (j
!= most_recent
&& !c
->force
&&
1621 content
->array
.raid_disks
> 0 &&
1622 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1623 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1624 if (c
->verbose
> -1)
1625 pr_err("ignoring %s as it reports %s as failed\n",
1626 devices
[j
].devname
, devices
[most_recent
].devname
);
1630 /* Require event counter to be same as, or just less than,
1631 * most recent. If it is bigger, it must be a stray spare and
1632 * should be ignored.
1634 if (devices
[j
].i
.events
+event_margin
>=
1635 devices
[most_recent
].i
.events
&&
1636 devices
[j
].i
.events
<=
1637 devices
[most_recent
].i
.events
1639 devices
[j
].uptodate
= 1;
1640 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1642 if (i
< content
->array
.raid_disks
* 2) {
1643 if (devices
[j
].i
.recovery_start
== MaxSector
||
1644 (content
->reshape_active
&&
1645 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1653 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1659 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1660 avail
, most_recent
, st
, c
);
1665 /* Now we want to look at the superblock which the kernel will base things on
1666 * and compare the devices that we think are working with the devices that the
1667 * superblock thinks are working.
1668 * If there are differences and --force is given, then update this chosen
1672 st
->ss
->free_super(st
);
1673 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1679 if (!devices
[j
].uptodate
)
1681 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1684 if ((fd
=dev_open(devices
[j
].devname
,
1685 devices
[j
].included
? O_RDONLY
1686 : (O_RDONLY
|O_EXCL
)))< 0) {
1687 pr_err("Cannot open %s: %s\n",
1688 devices
[j
].devname
, strerror(errno
));
1691 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1693 pr_err("RAID superblock has disappeared from %s\n",
1694 devices
[j
].devname
);
1699 if (st
->sb
== NULL
) {
1700 pr_err("No suitable drives found for %s\n", mddev
);
1703 st
->ss
->getinfo_super(st
, content
, NULL
);
1704 if (sysfs_init(content
, mdfd
, NULL
)) {
1705 pr_err("Unable to initialize sysfs\n");
1709 /* after reload context, store journal_clean in context */
1710 content
->journal_clean
= journal_clean
;
1711 for (i
=0; i
<bestcnt
; i
++) {
1713 unsigned int desired_state
;
1717 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1718 desired_state
= (1<<MD_DISK_JOURNAL
);
1719 else if (i
>= content
->array
.raid_disks
* 2)
1722 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1724 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1726 desired_state
|= devices
[j
].i
.disk
.state
& ((1<<MD_DISK_FAILFAST
) |
1727 (1<<MD_DISK_WRITEMOSTLY
));
1729 if (!devices
[j
].uptodate
)
1732 devices
[j
].i
.disk
.state
= desired_state
;
1733 if (!(devices
[j
].i
.array
.state
& 1))
1736 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1737 c
->verbose
, 0, NULL
)) {
1739 if (c
->verbose
>= 0)
1740 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1741 j
, mddev
, devices
[j
].devname
);
1744 if (c
->verbose
>= -1)
1745 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1746 i
, mddev
, devices
[j
].devname
);
1750 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1751 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1756 if (c
->force
&& !clean
&&
1757 !enough(content
->array
.level
, content
->array
.raid_disks
,
1758 content
->array
.layout
, clean
,
1760 change
+= st
->ss
->update_super(st
, content
, "force-array",
1761 devices
[chosen_drive
].devname
, c
->verbose
,
1769 fd
= dev_open(devices
[chosen_drive
].devname
,
1770 devices
[chosen_drive
].included
?
1771 O_RDWR
: (O_RDWR
|O_EXCL
));
1773 pr_err("Could not open %s for write - cannot Assemble array.\n",
1774 devices
[chosen_drive
].devname
);
1777 if (st
->ss
->store_super(st
, fd
)) {
1779 pr_err("Could not re-write superblock on %s\n",
1780 devices
[chosen_drive
].devname
);
1783 if (c
->verbose
>= 0)
1784 pr_err("Marking array %s as 'clean'\n",
1789 /* If we are in the middle of a reshape we may need to restore saved data
1790 * that was moved aside due to the reshape overwriting live data
1791 * The code of doing this lives in Grow.c
1793 if (content
->reshape_active
&&
1794 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1796 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1798 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1800 if (!c
->backup_file
)
1801 c
->backup_file
= locate_backup(content
->sys_name
);
1802 enable_fds(bestcnt
/2);
1803 for (i
= 0; i
< bestcnt
/2; i
++) {
1806 fdlist
[i
] = dev_open(devices
[j
].devname
,
1808 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1809 if (fdlist
[i
] < 0) {
1810 pr_err("Could not open %s for write - cannot Assemble array.\n",
1811 devices
[j
].devname
);
1819 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1820 err
= st
->ss
->recover_backup(st
, content
);
1822 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1823 c
->backup_file
, c
->verbose
> 0);
1824 if (err
&& c
->invalid_backup
) {
1826 pr_err("continuing without restoring backup\n");
1832 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1836 pr_err("Failed to restore critical section for reshape, sorry.\n");
1837 if (c
->backup_file
== NULL
)
1838 cont_err("Possibly you needed to specify the --backup-file\n");
1843 /* Almost ready to actually *do* something */
1844 /* First, fill in the map, so that udev can find our name
1845 * as soon as we become active.
1847 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1848 content
->array
.major_version
= 1;
1849 content
->array
.minor_version
= 0;
1850 strcpy(content
->text_version
, "1.0");
1853 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1854 content
->uuid
, chosen_name
);
1856 rv
= start_array(mdfd
, mddev
, content
,
1857 st
, ident
, best
, bestcnt
,
1858 chosen_drive
, devices
, okcnt
, sparecnt
,
1859 rebuilding_cnt
, journalcnt
,
1861 clean
, avail
, start_partial_ok
,
1864 if (rv
== 1 && !pre_exist
)
1865 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1870 wait_for(chosen_name
, mdfd
);
1874 /* There is a nasty race with 'mdadm --monitor'.
1875 * If it opens this device before we close it,
1876 * it gets an incomplete open on which IO
1877 * doesn't work and the capacity is
1879 * If we reopen (to check for layered devices)
1880 * before --monitor closes, we loose.
1882 * So: wait upto 1 second for there to be
1883 * a non-zero capacity.
1885 while (usecs
< 1000) {
1886 mdfd
= open(mddev
, O_RDONLY
);
1888 unsigned long long size
;
1889 if (get_dev_size(mdfd
, NULL
, &size
) &&
1898 } else if (mdfd
>= 0)
1901 /* '2' means 'OK, but not started yet' */
1906 return rv
== 2 ? 0 : rv
;
1909 int assemble_container_content(struct supertype
*st
, int mdfd
,
1910 struct mdinfo
*content
, struct context
*c
,
1911 char *chosen_name
, int *result
)
1913 struct mdinfo
*dev
, *sra
, *dev2
;
1914 int working
= 0, preexist
= 0;
1921 if (sysfs_init(content
, mdfd
, NULL
)) {
1922 pr_err("Unable to initialize sysfs\n");
1926 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1927 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1928 if (content
->array
.major_version
== -1 &&
1929 content
->array
.minor_version
== -2 &&
1931 content
->text_version
[0] == '/')
1932 content
->text_version
[0] = '-';
1933 if (sysfs_set_array(content
, 9003) != 0) {
1939 /* There are two types of reshape: container wide or sub-array specific
1940 * Check if metadata requests blocking container wide reshapes
1942 start_reshape
= (content
->reshape_active
&&
1943 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1944 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1946 /* Block subarray here if it is under reshape now
1947 * Do not allow for any changes in this array
1949 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1950 block_subarray(content
);
1952 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1953 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1954 if (dev2
->disk
.major
== dev
->disk
.major
&&
1955 dev2
->disk
.minor
== dev
->disk
.minor
)
1959 /* Don't want this one any more */
1960 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1962 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1966 sysfs_set_str(sra
, dev2
, "state", "remove");
1968 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1969 avail
= xcalloc(content
->array
.raid_disks
, 1);
1970 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1971 if (dev
->disk
.raid_disk
>= 0)
1972 avail
[dev
->disk
.raid_disk
] = 1;
1973 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1974 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1975 content
->reshape_active
)
1979 } else if (errno
== EEXIST
)
1983 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1985 return 1;/* Nothing new, don't try to start */
1987 map_update(NULL
, fd2devnm(mdfd
), content
->text_version
,
1988 content
->uuid
, chosen_name
);
1990 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
1991 st
->ss
->validate_ppl
) {
1992 content
->array
.state
|= 1;
1995 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
2000 ret
= st
->ss
->validate_ppl(st
, content
, dev
);
2010 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
2012 content
->array
.state
&= ~1;
2013 avail
[dev
->disk
.raid_disk
] = 0;
2017 /* have --force - overwrite the invalid ppl */
2018 devpath
= map_dev(dev
->disk
.major
, dev
->disk
.minor
, 0);
2019 dfd
= dev_open(devpath
, O_RDWR
);
2021 pr_err("Failed to open %s\n", devpath
);
2026 err
= st
->ss
->write_init_ppl(st
, content
, dfd
);
2037 } else if (c
->force
) {
2038 /* Set the array as 'clean' so that we can proceed with starting
2039 * it even if we don't have all devices. Mdmon doesn't care
2040 * if the dirty flag is set in metadata, it will start managing
2042 * This is really important for raid456 (RWH case), other levels
2043 * are started anyway.
2045 content
->array
.state
|= 1;
2048 if (enough(content
->array
.level
, content
->array
.raid_disks
,
2049 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
2050 if (c
->export
&& result
)
2052 else if (c
->verbose
>= 0) {
2053 pr_err("%s assembled with %d device%s",
2054 chosen_name
, preexist
+ working
,
2055 preexist
+ working
== 1 ? "":"s");
2057 fprintf(stderr
, " (%d new)", working
);
2058 fprintf(stderr
, " but not started\n");
2065 if ((working
+ preexist
+ expansion
) < content
->array
.working_disks
) {
2066 if (c
->runstop
<= 0) {
2067 if (c
->export
&& result
)
2068 *result
|= INCR_UNSAFE
;
2069 else if (c
->verbose
>= 0) {
2070 pr_err("%s assembled with %d device%s",
2071 chosen_name
, preexist
+ working
,
2072 preexist
+ working
== 1 ? "":"s");
2074 fprintf(stderr
, " (%d new)", working
);
2075 fprintf(stderr
, " but not safe to start\n");
2077 pr_err("Consider --run to start array as degraded.\n");
2080 } else if (content
->array
.level
>= 4 &&
2081 content
->array
.level
<= 6 &&
2082 content
->resync_start
!= MaxSector
&&
2084 /* Don't inform the kernel that the array is not
2085 * clean and requires resync.
2087 content
->resync_start
= MaxSector
;
2088 err
= sysfs_set_num(content
, NULL
, "resync_start",
2092 pr_err("%s array state forced to clean. It may cause data corruption.\n",
2098 * Before activating the array, perform extra steps required
2099 * to configure the internal write-intent bitmap.
2101 if (content
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
&&
2103 st
->ss
->set_bitmap(st
, content
);
2105 if (start_reshape
) {
2106 int spare
= content
->array
.raid_disks
+ expansion
;
2107 if (restore_backup(st
, content
,
2109 spare
, &c
->backup_file
, c
->verbose
) == 1)
2112 if (content
->reshape_progress
== 0) {
2113 /* If reshape progress is 0 - we are assembling the
2114 * array that was stopped, before reshape has started.
2115 * Array needs to be started as active, Grow_continue()
2116 * will start the reshape.
2118 sysfs_set_num(content
, NULL
, "reshape_position",
2120 err
= sysfs_set_str(content
, NULL
,
2121 "array_state", "active");
2122 sysfs_set_num(content
, NULL
, "reshape_position", 0);
2124 err
= sysfs_set_str(content
, NULL
,
2125 "array_state", "readonly");
2131 if (st
->ss
->external
) {
2132 if (!mdmon_running(st
->container_devnm
))
2133 start_mdmon(st
->container_devnm
);
2134 ping_monitor(st
->container_devnm
);
2135 if (mdmon_running(st
->container_devnm
) &&
2136 st
->update_tail
== NULL
)
2137 st
->update_tail
= &st
->updates
;
2140 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2141 0, c
->freeze_reshape
);
2142 } else switch(content
->array
.level
) {
2144 case LEVEL_MULTIPATH
:
2146 err
= sysfs_set_str(content
, NULL
, "array_state",
2147 c
->readonly
? "readonly" : "active");
2150 err
= sysfs_set_str(content
, NULL
, "array_state",
2152 /* start mdmon if needed. */
2154 if (!mdmon_running(st
->container_devnm
))
2155 start_mdmon(st
->container_devnm
);
2156 ping_monitor(st
->container_devnm
);
2161 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2163 /* Block subarray here if it is not reshaped now
2164 * It has be blocked a little later to allow mdmon to switch in
2167 if (st
->ss
->external
&& content
->recovery_blocked
&&
2169 block_subarray(content
);
2171 if (c
->export
&& result
) {
2175 *result
|= INCR_YES
;
2176 } else if (c
->verbose
>= 0) {
2178 pr_err("array %s now has %d device%s",
2179 chosen_name
, working
+ preexist
,
2180 working
+ preexist
== 1 ? "":"s");
2182 sysfs_rules_apply(chosen_name
, content
);
2183 pr_err("Started %s with %d device%s",
2184 chosen_name
, working
+ preexist
,
2185 working
+ preexist
== 1 ? "":"s");
2188 fprintf(stderr
, " (%d new)", working
);
2190 fprintf(stderr
, " ( + %d for expansion)",
2192 fprintf(stderr
, "\n");
2195 wait_for(chosen_name
, mdfd
);
2197 /* FIXME should have an O_EXCL and wait for read-auto */