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>
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
;
153 tmpdev
= devlist
; num_devs
= 0;
159 tmpdev
->disposition
= 0;
160 tmpdev
= tmpdev
->next
;
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
167 for (tmpdev
= devlist
;
169 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
170 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(dfd
, &stb
)< 0) {
209 pr_err("fstat failed for %s: %s\n",
210 devname
, strerror(errno
));
212 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
213 pr_err("%s is not a block device.\n",
216 } else if (must_be_container(dfd
)) {
218 /* already found some components, this cannot
222 pr_err("%s is a container, but we are looking for components\n",
225 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
226 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
228 pr_err("not a recognisable container: %s\n",
232 } else if (!tst
->ss
->load_container
233 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
235 pr_err("no correct container type: %s\n",
238 } else if (auto_assem
&&
239 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
240 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
242 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
243 devname
, tst
->ss
->name
);
248 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
250 pr_err("no recogniseable superblock on %s\n",
253 } else if ((tst
->ignore_hw_compat
= 0),
254 tst
->ss
->load_super(tst
, dfd
,
255 report_mismatch
? devname
: NULL
)) {
257 pr_err("no RAID superblock on %s\n",
260 } else if (tst
->ss
->compare_super
== NULL
) {
262 pr_err("Cannot assemble %s metadata on %s\n",
263 tst
->ss
->name
, devname
);
265 } else if (auto_assem
&& st
== NULL
&&
266 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
267 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
269 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
270 devname
, tst
->ss
->name
);
274 if (dfd
>= 0) close(dfd
);
275 if (tmpdev
->used
== 2) {
276 if (auto_assem
|| !inargv
)
277 /* Ignore unrecognised devices during auto-assembly */
279 if (ident
->uuid_set
|| ident
->name
[0] ||
280 ident
->super_minor
!= UnSet
)
281 /* Ignore unrecognised device if looking for
285 pr_err("%s has no superblock - assembly aborted\n",
288 st
->ss
->free_super(st
);
289 dev_policy_free(pol
);
290 domain_free(domains
);
294 if (found_container
) {
295 /* tmpdev is a container. We need to be either
296 * looking for a member, or auto-assembling
298 /* should be safe to try an exclusive open now, we
299 * have rejected anything that some other mdadm might
302 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
305 pr_err("%s is busy - skipping\n", devname
);
310 if (ident
->container
&& ident
->container
[0] != '/') {
315 tst
->ss
->getinfo_super(tst
, content
, NULL
);
317 if (!parse_uuid(ident
->container
, uuid
) ||
318 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
320 pr_err("%s has wrong UUID to be required container\n",
325 /* It is worth looking inside this container.
328 pr_err("looking in container %s\n",
331 for (content
= tst
->ss
->container_content(tst
, NULL
);
333 content
= content
->next
) {
335 if (!ident_matches(ident
, content
, tst
,
336 c
->homehost
, c
->require_homehost
,
338 report_mismatch
? devname
: NULL
))
339 /* message already printed */;
340 else if (is_member_busy(content
->text_version
)) {
342 pr_err("member %s in %s is already assembled\n",
343 content
->text_version
,
345 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
346 /* do not assemble arrays with unsupported configurations */
347 pr_err("Cannot activate member %s in %s.\n",
348 content
->text_version
,
355 goto loop
; /* empty container */
358 st
= tst
; tst
= NULL
;
359 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
360 pr_err("%s is a container, but is not only device given: confused and aborting\n",
362 st
->ss
->free_super(st
);
363 dev_policy_free(pol
);
364 domain_free(domains
);
368 pr_err("found match on member %s in %s\n",
369 content
->text_version
, devname
);
371 /* make sure we finished the loop */
376 tst
->ss
->getinfo_super(tst
, content
, NULL
);
378 if (!ident_matches(ident
, content
, tst
,
379 c
->homehost
, c
->require_homehost
,
381 report_mismatch
? devname
: NULL
))
385 /* Never auto-assemble things that conflict
386 * with mdadm.conf in some way
388 struct mddev_ident
*match
;
391 match
= conf_match(tst
, content
, devname
,
392 report_mismatch
? c
->verbose
: -1,
394 if (!match
&& rv
== 2)
396 if (match
&& match
->devname
&&
397 strcasecmp(match
->devname
, "<ignore>") == 0) {
399 pr_err("%s is a member of an explicitly ignored array\n",
403 if (match
&& !ident_matches(match
, content
, tst
,
404 c
->homehost
, c
->require_homehost
,
406 report_mismatch
? devname
: NULL
))
407 /* Array exists in mdadm.conf but some
408 * details don't match, so reject it
413 /* should be safe to try an exclusive open now, we
414 * have rejected anything that some other mdadm might
417 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
420 pr_err("%s is busy - skipping\n", devname
);
427 if (st
->minor_version
== -1)
428 st
->minor_version
= tst
->minor_version
;
430 if (memcmp(content
->uuid
, uuid_zero
,
431 sizeof(int[4])) == 0) {
432 /* this is a floating spare. It cannot define
433 * an array unless there are no more arrays of
434 * this type to be found. It can be included
435 * in an array of this type though.
441 if (st
->ss
!= tst
->ss
||
442 st
->minor_version
!= tst
->minor_version
||
443 st
->ss
->compare_super(st
, tst
) != 0) {
444 /* Some mismatch. If exactly one array matches this host,
445 * we can resolve on that one.
446 * Or, if we are auto assembling, we just ignore the second
452 int first
= st
->ss
->match_home(st
, c
->homehost
);
453 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
455 (first
== 1 || last
== 1)) {
456 /* We can do something */
457 if (first
) {/* just ignore this one */
459 pr_err("%s misses out due to wrong homehost\n",
462 } else { /* reject all those sofar */
463 struct mddev_dev
*td
;
465 pr_err("%s overrides previous devices due to good homehost\n",
467 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
475 pr_err("superblock on %s doesn't match others - assembly aborted\n",
477 tst
->ss
->free_super(tst
);
478 st
->ss
->free_super(st
);
479 dev_policy_free(pol
);
480 domain_free(domains
);
486 /* Collect domain information from members only */
487 if (tmpdev
&& tmpdev
->used
== 1) {
489 pol
= devid_policy(stb
.st_rdev
);
490 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
492 dev_policy_free(pol
);
495 tst
->ss
->free_super(tst
);
498 /* Check if we found some imsm spares but no members */
501 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
503 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
504 if (tmpdev
->used
!= 3)
510 /* we need sb from one of the spares */
511 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
513 st
->ss
->load_super(st
, dfd
, NULL
))
520 /* Now reject spares that don't match domains of identified members */
521 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
523 if (tmpdev
->used
!= 3)
525 if (stat(tmpdev
->devname
, &stb
)< 0) {
526 pr_err("fstat failed for %s: %s\n",
527 tmpdev
->devname
, strerror(errno
));
530 struct dev_policy
*pol
= devid_policy(stb
.st_rdev
);
531 int dt
= domain_test(domains
, pol
, NULL
);
532 if (inargv
&& dt
!= 0)
533 /* take this spare as domains match
534 * if there are any */
536 else if (!inargv
&& dt
== 1)
537 /* device wasn't explicitly listed, so need
538 * explicit domain match - which we have */
541 /* if domains don't match mark as unused */
543 dev_policy_free(pol
);
546 domain_free(domains
);
548 if (st
&& st
->sb
&& content
== *contentp
)
549 st
->ss
->getinfo_super(st
, content
, NULL
);
557 int uptodate
; /* set once we decide that this device is as
558 * recent as everything else in the array.
560 int included
; /* set if the device is already in the array
561 * due to a previous '-I'
566 static int load_devices(struct devs
*devices
, char *devmap
,
567 struct mddev_ident
*ident
, struct supertype
**stp
,
568 struct mddev_dev
*devlist
, struct context
*c
,
569 struct mdinfo
*content
,
570 int mdfd
, char *mddev
,
571 int *most_recentp
, int *bestcntp
, int **bestp
,
574 struct mddev_dev
*tmpdev
;
580 int most_recent
= -1;
583 struct supertype
*st
= *stp
;
585 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
586 char *devname
= tmpdev
->devname
;
588 struct supertype
*tst
;
592 if (tmpdev
->used
!= 1)
594 /* looks like a good enough match to update the super block if needed */
597 /* prepare useful information in info structures */
602 if (strcmp(c
->update
, "uuid")==0 &&
605 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
606 read(rfd
, ident
->uuid
, 16) != 16) {
607 *(__u32
*)(ident
->uuid
) = random();
608 *(__u32
*)(ident
->uuid
+1) = random();
609 *(__u32
*)(ident
->uuid
+2) = random();
610 *(__u32
*)(ident
->uuid
+3) = random();
612 if (rfd
>= 0) close(rfd
);
614 dfd
= dev_open(devname
,
615 tmpdev
->disposition
== 'I'
616 ? O_RDWR
: (O_RDWR
|O_EXCL
));
619 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
620 pr_err("cannot re-read metadata from %s - aborting\n",
627 tst
->ss
->free_super(tst
);
632 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
634 memcpy(content
->uuid
, ident
->uuid
, 16);
635 strcpy(content
->name
, ident
->name
);
636 content
->array
.md_minor
= minor(stb2
.st_rdev
);
638 if (strcmp(c
->update
, "byteorder") == 0)
640 else if (strcmp(c
->update
, "home-cluster") == 0) {
641 tst
->cluster_name
= c
->homecluster
;
642 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
643 } else if (strcmp(c
->update
, "nodes") == 0) {
644 tst
->nodes
= c
->nodes
;
645 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
647 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
653 pr_err("--update=%s not understood for %s metadata\n",
654 c
->update
, tst
->ss
->name
);
655 tst
->ss
->free_super(tst
);
664 if (strcmp(c
->update
, "uuid")==0 &&
667 memcpy(ident
->uuid
, content
->uuid
, 16);
669 if (tst
->ss
->store_super(tst
, dfd
))
670 pr_err("Could not re-write superblock on %s.\n",
673 if (strcmp(c
->update
, "uuid")==0 &&
674 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
675 if (bitmap_update_uuid(ident
->bitmap_fd
,
677 tst
->ss
->swapuuid
) != 0)
678 pr_err("Could not update uuid on external bitmap.\n");
685 dfd
= dev_open(devname
,
686 tmpdev
->disposition
== 'I'
687 ? O_RDWR
: (O_RDWR
|O_EXCL
));
690 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
691 pr_err("cannot re-read metadata from %s - aborting\n",
698 tst
->ss
->free_super(tst
);
703 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
710 pr_err("%s is identified as a member of %s, slot %d%s.\n",
711 devname
, mddev
, content
->disk
.raid_disk
,
712 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
713 devices
[devcnt
].devname
= devname
;
714 devices
[devcnt
].uptodate
= 0;
715 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
716 devices
[devcnt
].i
= *content
;
717 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
718 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
720 if (devices
[devcnt
].i
.disk
.state
== 6) {
721 if (most_recent
< 0 ||
722 devices
[devcnt
].i
.events
723 > devices
[most_recent
].i
.events
) {
724 struct supertype
*tmp
= tst
;
727 most_recent
= devcnt
;
730 tst
->ss
->free_super(tst
);
733 if (content
->array
.level
== LEVEL_MULTIPATH
)
734 /* with multipath, the raid_disk from the superblock is meaningless */
737 i
= devices
[devcnt
].i
.disk
.raid_disk
;
738 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
739 if (nextspare
< content
->array
.raid_disks
*2)
740 nextspare
= content
->array
.raid_disks
*2;
743 /* i is raid_disk - double it so there is room for
746 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
748 if (i
>= content
->array
.raid_disks
*2 &&
754 int newbestcnt
= i
+10;
755 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
757 for (c
=0; c
< newbestcnt
; c
++)
759 newbest
[c
] = best
[c
];
764 bestcnt
= newbestcnt
;
767 devices
[best
[i
]].i
.events
768 == devices
[devcnt
].i
.events
769 && (devices
[best
[i
]].i
.disk
.minor
770 != devices
[devcnt
].i
.disk
.minor
)
772 && content
->array
.level
!= LEVEL_MULTIPATH
) {
773 /* two different devices with identical superblock.
774 * Could be a mis-detection caused by overlapping
775 * partitions. fail-safe.
777 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
778 " If they are really different, please --zero the superblock on one\n"
779 " If they are the same or overlap, please remove one from %s.\n",
780 devices
[best
[i
]].devname
, devname
,
781 inargv
? "the list" :
782 "the\n DEVICE list in mdadm.conf"
791 || (devices
[best
[i
]].i
.events
792 < devices
[devcnt
].i
.events
))
797 if (most_recent
>= 0)
798 *most_recentp
= most_recent
;
805 static int force_array(struct mdinfo
*content
,
806 struct devs
*devices
,
807 int *best
, int bestcnt
, char *avail
,
809 struct supertype
*st
,
813 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
814 content
->array
.layout
, 1,
817 (content
->reshape_active
&& content
->delta_disks
> 0 &&
818 !enough(content
->array
.level
, (content
->array
.raid_disks
819 - content
->delta_disks
),
820 content
->new_layout
, 1,
823 /* Choose the newest best drive which is
824 * not up-to-date, update the superblock
828 struct supertype
*tst
;
829 unsigned long long current_events
;
830 int chosen_drive
= -1;
834 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
839 if (devices
[j
].uptodate
)
841 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
843 if (!devices
[j
].i
.reshape_active
||
844 devices
[j
].i
.delta_disks
<= 0)
846 /* When increasing number of devices, an
847 * added device also appears to be
848 * recovering. It is safe to include it
849 * as long as it won't be a source of
851 * For now, just allow for last data
852 * devices in RAID4 or last devices in RAID4/5/6.
854 delta
= devices
[j
].i
.delta_disks
;
855 if (devices
[j
].i
.array
.level
>= 4 &&
856 devices
[j
].i
.array
.level
<= 6 &&
857 i
/2 >= content
->array
.raid_disks
- delta
)
859 else if (devices
[j
].i
.array
.level
== 4 &&
860 i
/2 >= content
->array
.raid_disks
- delta
- 1)
865 if (chosen_drive
< 0 ||
867 > devices
[chosen_drive
].i
.events
)
870 if (chosen_drive
< 0)
872 current_events
= devices
[chosen_drive
].i
.events
;
875 pr_err("forcing event count in %s(%d) from %d upto %d\n",
876 devices
[chosen_drive
].devname
,
877 devices
[chosen_drive
].i
.disk
.raid_disk
,
878 (int)(devices
[chosen_drive
].i
.events
),
879 (int)(devices
[most_recent
].i
.events
));
880 fd
= dev_open(devices
[chosen_drive
].devname
,
881 devices
[chosen_drive
].included
? O_RDWR
884 pr_err("Couldn't open %s for write - not updating\n",
885 devices
[chosen_drive
].devname
);
886 devices
[chosen_drive
].i
.events
= 0;
890 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
892 pr_err("RAID superblock disappeared from %s - not updating.\n",
893 devices
[chosen_drive
].devname
);
894 devices
[chosen_drive
].i
.events
= 0;
897 content
->events
= devices
[most_recent
].i
.events
;
898 tst
->ss
->update_super(tst
, content
, "force-one",
899 devices
[chosen_drive
].devname
, c
->verbose
,
902 if (tst
->ss
->store_super(tst
, fd
)) {
904 pr_err("Could not re-write superblock on %s\n",
905 devices
[chosen_drive
].devname
);
906 devices
[chosen_drive
].i
.events
= 0;
907 tst
->ss
->free_super(tst
);
911 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
912 devices
[chosen_drive
].uptodate
= 1;
913 avail
[chosen_drive
] = 1;
915 tst
->ss
->free_super(tst
);
916 /* If there are any other drives of the same vintage,
917 * add them in as well. We can't lose and we might gain
920 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
924 !devices
[j
].uptodate
&&
925 devices
[j
].i
.recovery_start
== MaxSector
&&
926 devices
[j
].i
.events
== current_events
) {
935 static int start_array(int mdfd
,
937 struct mdinfo
*content
,
938 struct supertype
*st
,
939 struct mddev_ident
*ident
,
940 int *best
, int bestcnt
,
942 struct devs
*devices
,
944 unsigned int sparecnt
,
945 unsigned int rebuilding_cnt
,
946 unsigned int journalcnt
,
948 int clean
, char *avail
,
949 int start_partial_ok
,
958 unsigned int req_cnt
;
960 if (expect_journal
&& (journal_clean
== 0)) {
962 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
965 pr_err("Journal is missing or stale, starting array read only.\n");
969 rv
= set_array_info(mdfd
, st
, content
);
971 pr_err("failed to set array info for %s: %s\n",
972 mddev
, strerror(errno
));
975 if (ident
->bitmap_fd
>= 0) {
976 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
977 pr_err("SET_BITMAP_FILE failed.\n");
980 } else if (ident
->bitmap_file
) {
981 /* From config file */
982 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
984 pr_err("Could not open bitmap file %s\n",
988 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
989 pr_err("Failed to set bitmapfile for %s\n", mddev
);
996 /* First, add the raid disks, but add the chosen one last */
997 for (i
=0; i
<= bestcnt
; i
++) {
1001 if (j
== chosen_drive
)
1006 if (j
>= 0 && !devices
[j
].included
) {
1007 int dfd
= dev_open(devices
[j
].devname
,
1010 remove_partitions(dfd
);
1013 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1016 pr_err("failed to add %s to %s: %s\n",
1020 if (i
< content
->array
.raid_disks
* 2
1025 } else if (c
->verbose
> 0)
1026 pr_err("added %s to %s as %d%s%s\n",
1027 devices
[j
].devname
, mddev
,
1028 devices
[j
].i
.disk
.raid_disk
,
1029 devices
[j
].uptodate
?"":
1030 " (possibly out of date)",
1031 (devices
[j
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))?" replacement":"");
1032 } else if (j
>= 0) {
1034 pr_err("%s is already in %s as %d\n",
1035 devices
[j
].devname
, mddev
,
1036 devices
[j
].i
.disk
.raid_disk
);
1037 } else if (c
->verbose
> 0 && i
< content
->array
.raid_disks
*2
1039 pr_err("no uptodate device for slot %d of %s\n",
1043 if (content
->array
.level
== LEVEL_CONTAINER
) {
1044 if (c
->verbose
>= 0) {
1045 pr_err("Container %s has been assembled with %d drive%s",
1046 mddev
, okcnt
+sparecnt
+journalcnt
,
1047 okcnt
+sparecnt
+journalcnt
==1?"":"s");
1048 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1049 fprintf(stderr
, " (out of %d)",
1050 content
->array
.raid_disks
);
1051 fprintf(stderr
, "\n");
1054 if (st
->ss
->validate_container
) {
1055 struct mdinfo
*devices_list
;
1056 struct mdinfo
*info_devices
= xmalloc(sizeof(struct mdinfo
)*(okcnt
+sparecnt
));
1058 devices_list
= NULL
;
1059 for (count
= 0; count
< okcnt
+sparecnt
; count
++) {
1060 info_devices
[count
] = devices
[count
].i
;
1061 info_devices
[count
].next
= devices_list
;
1062 devices_list
= &info_devices
[count
];
1064 if (st
->ss
->validate_container(devices_list
))
1065 pr_err("Mismatch detected!\n");
1069 st
->ss
->free_super(st
);
1070 sysfs_uevent(content
, "change");
1071 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1072 /* Was partial, is still partial, so signal an error
1073 * to ensure we don't retry */
1078 /* Get number of in-sync devices according to the superblock.
1079 * We must have this number to start the array without -s or -R
1081 req_cnt
= content
->array
.working_disks
;
1083 if (c
->runstop
== 1 ||
1085 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1086 content
->array
.layout
, clean
, avail
) &&
1087 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1089 /* This array is good-to-go.
1090 * If a reshape is in progress then we might need to
1091 * continue monitoring it. In that case we start
1092 * it read-only and let the grow code make it writable.
1096 if (content
->reshape_active
&&
1097 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1098 content
->delta_disks
<= 0) {
1099 if (!c
->backup_file
) {
1100 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1102 pr_err("Please provided one with \"--backup-file=...\"\n");
1104 strcmp(c
->update
, "revert-reshape") == 0)
1105 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1108 rv
= sysfs_set_str(content
, NULL
,
1109 "array_state", "readonly");
1111 rv
= Grow_continue(mdfd
, st
, content
,
1114 } else if (c
->readonly
&&
1115 sysfs_attribute_available(
1116 content
, NULL
, "array_state")) {
1117 rv
= sysfs_set_str(content
, NULL
,
1118 "array_state", "readonly");
1121 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1122 reopen_mddev(mdfd
); /* drop O_EXCL */
1124 if (c
->verbose
>= 0) {
1125 pr_err("%s has been started with %d drive%s",
1126 mddev
, okcnt
, okcnt
==1?"":"s");
1127 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1128 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1130 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1132 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1134 fprintf(stderr
, " and %d journal", journalcnt
);
1135 fprintf(stderr
, ".\n");
1137 if (content
->reshape_active
&&
1138 content
->array
.level
>= 4 &&
1139 content
->array
.level
<= 6) {
1140 /* might need to increase the size
1141 * of the stripe cache - default is 256
1143 int chunk_size
= content
->array
.chunk_size
;
1144 if (content
->reshape_active
&&
1145 content
->new_chunk
> chunk_size
)
1146 chunk_size
= content
->new_chunk
;
1147 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1148 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1150 sysfs_set_num(sra
, NULL
,
1151 "stripe_cache_size",
1152 (4 * chunk_size
/ 4096) + 1);
1156 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1157 /* If any devices did not get added
1158 * because the kernel rejected them based
1159 * on event count, try adding them
1160 * again providing the action policy is
1161 * 're-add' or greater. The bitmap
1162 * might allow them to be included, or
1163 * they will become spares.
1165 for (i
= 0; i
< bestcnt
; i
++) {
1167 if (j
>= 0 && !devices
[j
].uptodate
) {
1168 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1170 rv
= add_disk(mdfd
, st
, content
,
1172 if (rv
== 0 && c
->verbose
>= 0)
1173 pr_err("%s has been re-added.\n",
1174 devices
[j
].devname
);
1178 if (content
->array
.level
== 6 &&
1179 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1181 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1183 sysfs_set_str(sra
, NULL
,
1184 "sync_action", "repair");
1189 pr_err("failed to RUN_ARRAY %s: %s\n",
1190 mddev
, strerror(errno
));
1192 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1193 content
->array
.layout
, 1, avail
))
1194 pr_err("Not enough devices to start the array.\n");
1195 else if (!enough(content
->array
.level
,
1196 content
->array
.raid_disks
,
1197 content
->array
.layout
, clean
,
1199 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1203 if (c
->runstop
== -1) {
1204 pr_err("%s assembled from %d drive%s",
1205 mddev
, okcnt
, okcnt
==1?"":"s");
1206 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1207 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1208 fprintf(stderr
, ", but not started.\n");
1211 if (c
->verbose
>= -1) {
1212 pr_err("%s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1214 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1216 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1217 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1218 content
->array
.layout
, 1, avail
))
1219 fprintf(stderr
, " - not enough to start the array.\n");
1220 else if (!enough(content
->array
.level
,
1221 content
->array
.raid_disks
,
1222 content
->array
.layout
, clean
,
1224 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1226 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1227 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1229 fprintf(stderr
, " - need %d to start", req_cnt
);
1230 fprintf(stderr
, " (use --run to insist).\n");
1236 int Assemble(struct supertype
*st
, char *mddev
,
1237 struct mddev_ident
*ident
,
1238 struct mddev_dev
*devlist
,
1242 * The task of Assemble is to find a collection of
1243 * devices that should (according to their superblocks)
1244 * form an array, and to give this collection to the MD driver.
1245 * In Linux-2.4 and later, this involves submitting a
1246 * SET_ARRAY_INFO ioctl with no arg - to prepare
1247 * the array - and then submit a number of
1248 * ADD_NEW_DISK ioctls to add disks into
1249 * the array. Finally RUN_ARRAY might
1250 * be submitted to start the array.
1252 * Much of the work of Assemble is in finding and/or
1253 * checking the disks to make sure they look right.
1255 * If mddev is not set, then scan must be set and we
1256 * read through the config file for dev+uuid mapping
1257 * We recurse, setting mddev, for each device that
1259 * - has a valid uuid (or any uuid if !uuidset)
1261 * If mddev is set, we try to determine state of md.
1262 * check version - must be at least 0.90.0
1263 * check kernel version. must be at least 2.4.
1264 * If not, we can possibly fall back on START_ARRAY
1265 * Try to GET_ARRAY_INFO.
1266 * If possible, give up
1267 * If not, try to STOP_ARRAY just to make sure
1269 * If !uuidset and scan, look in conf-file for uuid
1270 * If not found, give up
1271 * If !devlist and scan and uuidset, get list of devs from conf-file
1274 * Check superblock - discard if bad
1275 * Check uuid (set if we don't have one) - discard if no match
1276 * Check superblock similarity if we have a superblock - discard if different
1277 * Record events, devicenum
1278 * This should give us a list of devices for the array
1279 * We should collect the most recent event number
1281 * Count disks with recent enough event count
1282 * While force && !enough disks
1283 * Choose newest rejected disks, update event count
1284 * mark clean and rewrite superblock
1287 * foreach device with recent events : ADD_NEW_DISK
1288 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1290 * Check the device numbers in superblock are right
1291 * update superblock if any changes
1298 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1299 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1300 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1301 struct devs
*devices
;
1303 int *best
= NULL
; /* indexed by raid_disk */
1306 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1309 int most_recent
= 0;
1310 int expect_journal
= 0;
1311 int journal_clean
= 0;
1315 int start_partial_ok
= (c
->runstop
>= 0) &&
1316 (c
->force
|| devlist
==NULL
|| auto_assem
);
1318 struct mddev_dev
*tmpdev
;
1320 struct mdinfo
*content
= NULL
;
1321 struct mdinfo
*pre_exist
= NULL
;
1324 char chosen_name
[1024];
1325 struct map_ent
*map
= NULL
;
1329 * If any subdevs are listed, then any that don't
1330 * match ident are discarded. Remainder must all match and
1332 * If no subdevs, then we scan all devices in the config file, but
1333 * there must be something in the identity
1337 ident
->uuid_set
== 0 &&
1338 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1339 ident
->name
[0] == 0 &&
1340 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1341 ident
->devices
== NULL
) {
1342 pr_err("No identity information available for %s - cannot assemble.\n",
1343 mddev
? mddev
: "further assembly");
1347 if (devlist
== NULL
)
1348 devlist
= conf_get_devs();
1353 /* We come back here when doing auto-assembly and attempting some
1354 * set of devices failed. Those are now marked as ->used==2 and
1355 * we ignore them and try again
1357 if (!st
&& ident
->st
)
1360 pr_err("looking for devices for %s\n",
1361 mddev
? mddev
: "further assembly");
1365 st
->ignore_hw_compat
= 1;
1366 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1367 inargv
, auto_assem
);
1371 if (!st
|| !st
->sb
|| !content
)
1374 if (st
->ss
->require_journal
) {
1375 expect_journal
= st
->ss
->require_journal(st
);
1376 if (expect_journal
== 2) {
1377 pr_err("BUG: Superblock not loaded in Assemble.c:Assemble\n");
1382 /* We have a full set of devices - we now need to find the
1384 * However there is a risk that we are racing with "mdadm -I"
1385 * and the array is already partially assembled - we will have
1386 * rejected any devices already in this address.
1387 * So we take a lock on the map file - to prevent further races -
1388 * and look for the uuid in there. If found and the array is
1389 * active, we abort. If found and the array is not active
1390 * we commit to that md device and add all the contained devices
1391 * to our list. We flag them so that we don't try to re-add,
1392 * but can remove if they turn out to not be wanted.
1395 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1396 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1399 mp
= map_by_uuid(&map
, content
->uuid
);
1402 /* array already exists. */
1403 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1404 if (pre_exist
->array
.level
!= UnSet
) {
1405 pr_err("Found some drive for an array that is already active: %s\n",
1407 pr_err("giving up.\n");
1410 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1411 /* We want to add this device to our list,
1412 * but it could already be there if "mdadm -I"
1413 * started *after* we checked for O_EXCL.
1414 * If we add it to the top of the list
1415 * it will be preferred over later copies.
1417 struct mddev_dev
*newdev
;
1418 char *devname
= map_dev(dv
->disk
.major
,
1423 newdev
= xmalloc(sizeof(*newdev
));
1424 newdev
->devname
= devname
;
1425 newdev
->disposition
= 'I';
1427 newdev
->next
= devlist
;
1431 strcpy(chosen_name
, mp
->path
);
1432 if (c
->verbose
> 0 || mddev
== NULL
||
1433 strcmp(mddev
, chosen_name
) != 0)
1434 pr_err("Merging with already-assembled %s\n",
1436 mdfd
= open_dev_excl(mp
->devnm
);
1438 int trustworthy
= FOREIGN
;
1439 name
= content
->name
;
1440 switch (st
->ss
->match_home(st
, c
->homehost
)
1441 ?: st
->ss
->match_home(st
, "any")) {
1443 trustworthy
= LOCAL
;
1444 name
= strchr(content
->name
, ':');
1448 name
= content
->name
;
1452 /* If the array is listed in mdadm.conf or on
1453 * command line, then we trust the name
1454 * even if the array doesn't look local
1456 trustworthy
= LOCAL
;
1459 content
->array
.level
== LEVEL_CONTAINER
) {
1460 name
= content
->text_version
;
1461 trustworthy
= METADATA
;
1464 if (name
[0] && trustworthy
!= LOCAL
&&
1465 ! c
->require_homehost
&&
1466 conf_name_is_free(name
))
1467 trustworthy
= LOCAL
;
1469 if (trustworthy
== LOCAL
&&
1471 /* Ignore 'host:' prefix of name */
1472 name
= strchr(name
, ':')+1;
1474 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1478 st
->ss
->free_super(st
);
1483 mddev
= chosen_name
;
1484 if (get_linux_version() < 2004000 ||
1485 md_get_version(mdfd
) < 9000) {
1486 pr_err("Assemble requires Linux 2.4 or later, and\n"
1487 " md driver version 0.90.0 or later.\n"
1488 " Upgrade your kernel or try --build\n");
1492 if (pre_exist
== NULL
) {
1493 if (mddev_busy(fd2devnm(mdfd
))) {
1494 pr_err("%s already active, cannot restart it!\n",
1496 for (tmpdev
= devlist
;
1497 tmpdev
&& tmpdev
->used
!= 1;
1498 tmpdev
= tmpdev
->next
)
1500 if (tmpdev
&& auto_assem
)
1501 pr_err("%s needed for %s...\n",
1502 mddev
, tmpdev
->devname
);
1505 st
->ss
->free_super(st
);
1510 /* just incase it was started but has no content */
1511 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1515 if (content
!= &info
) {
1516 /* This is a member of a container. Try starting the array. */
1518 err
= assemble_container_content(st
, mdfd
, content
, c
,
1524 /* Ok, no bad inconsistancy, we can try updating etc */
1525 devices
= xcalloc(num_devs
, sizeof(*devices
));
1526 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1527 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1528 c
, content
, mdfd
, mddev
,
1529 &most_recent
, &bestcnt
, &best
, inargv
);
1534 pr_err("no devices found for %s\n",
1537 st
->ss
->free_super(st
);
1544 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1545 st
->minor_version
= 90;
1547 st
->ss
->getinfo_super(st
, content
, NULL
);
1548 clean
= content
->array
.state
& 1;
1550 /* now we have some devices that might be suitable.
1553 avail
= xcalloc(content
->array
.raid_disks
, 1);
1559 for (i
=0; i
< bestcnt
; i
++) {
1561 int event_margin
= 1; /* always allow a difference of '1'
1562 * like the kernel does
1564 if (j
< 0) continue;
1565 /* note: we ignore error flags in multipath arrays
1566 * as they don't make sense
1568 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1569 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1572 else /* unexpected journal, mark as faulty */
1573 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1574 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1575 if (!(devices
[j
].i
.disk
.state
1576 & (1<<MD_DISK_FAULTY
))) {
1577 devices
[j
].uptodate
= 1;
1583 /* If this device thinks that 'most_recent' has failed, then
1584 * we must reject this device.
1586 if (j
!= most_recent
&& !c
->force
&&
1587 content
->array
.raid_disks
> 0 &&
1588 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1589 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1590 if (c
->verbose
> -1)
1591 pr_err("ignoring %s as it reports %s as failed\n",
1592 devices
[j
].devname
, devices
[most_recent
].devname
);
1596 /* Require event counter to be same as, or just less than,
1597 * most recent. If it is bigger, it must be a stray spare and
1598 * should be ignored.
1600 if (devices
[j
].i
.events
+event_margin
>=
1601 devices
[most_recent
].i
.events
&&
1602 devices
[j
].i
.events
<=
1603 devices
[most_recent
].i
.events
1605 devices
[j
].uptodate
= 1;
1606 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1608 if (i
< content
->array
.raid_disks
* 2) {
1609 if (devices
[j
].i
.recovery_start
== MaxSector
||
1610 (content
->reshape_active
&&
1611 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1619 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1625 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1626 avail
, most_recent
, st
, c
);
1631 /* Now we want to look at the superblock which the kernel will base things on
1632 * and compare the devices that we think are working with the devices that the
1633 * superblock thinks are working.
1634 * If there are differences and --force is given, then update this chosen
1638 st
->ss
->free_super(st
);
1639 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1645 if (!devices
[j
].uptodate
)
1647 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1650 if ((fd
=dev_open(devices
[j
].devname
,
1651 devices
[j
].included
? O_RDONLY
1652 : (O_RDONLY
|O_EXCL
)))< 0) {
1653 pr_err("Cannot open %s: %s\n",
1654 devices
[j
].devname
, strerror(errno
));
1659 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1661 pr_err("RAID superblock has disappeared from %s\n",
1662 devices
[j
].devname
);
1669 if (st
->sb
== NULL
) {
1670 pr_err("No suitable drives found for %s\n", mddev
);
1675 st
->ss
->getinfo_super(st
, content
, NULL
);
1677 sysfs_init(content
, mdfd
, NULL
);
1679 for (i
=0; i
<bestcnt
; i
++) {
1681 unsigned int desired_state
;
1683 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1684 desired_state
= (1<<MD_DISK_JOURNAL
);
1685 else if (i
>= content
->array
.raid_disks
* 2)
1688 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1690 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1694 if (!devices
[j
].uptodate
)
1697 devices
[j
].i
.disk
.state
= desired_state
;
1698 if (!(devices
[j
].i
.array
.state
& 1))
1701 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1702 c
->verbose
, 0, NULL
)) {
1704 if (c
->verbose
>= 0)
1705 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1706 j
, mddev
, devices
[j
].devname
);
1709 if (c
->verbose
>= -1)
1710 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1711 i
, mddev
, devices
[j
].devname
);
1715 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1716 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1721 if (c
->force
&& !clean
&&
1722 !enough(content
->array
.level
, content
->array
.raid_disks
,
1723 content
->array
.layout
, clean
,
1725 change
+= st
->ss
->update_super(st
, content
, "force-array",
1726 devices
[chosen_drive
].devname
, c
->verbose
,
1734 fd
= dev_open(devices
[chosen_drive
].devname
,
1735 devices
[chosen_drive
].included
?
1736 O_RDWR
: (O_RDWR
|O_EXCL
));
1738 pr_err("Could not open %s for write - cannot Assemble array.\n",
1739 devices
[chosen_drive
].devname
);
1744 if (st
->ss
->store_super(st
, fd
)) {
1746 pr_err("Could not re-write superblock on %s\n",
1747 devices
[chosen_drive
].devname
);
1752 if (c
->verbose
>= 0)
1753 pr_err("Marking array %s as 'clean'\n",
1758 /* If we are in the middle of a reshape we may need to restore saved data
1759 * that was moved aside due to the reshape overwriting live data
1760 * The code of doing this lives in Grow.c
1763 if (content
->reshape_active
&&
1764 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1766 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1768 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1770 if (!c
->backup_file
)
1771 c
->backup_file
= locate_backup(content
->sys_name
);
1772 enable_fds(bestcnt
/2);
1773 for (i
= 0; i
< bestcnt
/2; i
++) {
1776 fdlist
[i
] = dev_open(devices
[j
].devname
,
1778 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1779 if (fdlist
[i
] < 0) {
1780 pr_err("Could not open %s for write - cannot Assemble array.\n",
1781 devices
[j
].devname
);
1789 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1790 err
= st
->ss
->recover_backup(st
, content
);
1792 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1793 c
->backup_file
, c
->verbose
> 0);
1794 if (err
&& c
->invalid_backup
) {
1796 pr_err("continuing without restoring backup\n");
1802 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1806 pr_err("Failed to restore critical section for reshape, sorry.\n");
1807 if (c
->backup_file
== NULL
)
1808 cont_err("Possibly you needed to specify the --backup-file\n");
1816 /* Almost ready to actually *do* something */
1817 /* First, fill in the map, so that udev can find our name
1818 * as soon as we become active.
1820 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1821 content
->array
.major_version
= 1;
1822 content
->array
.minor_version
= 0;
1823 strcpy(content
->text_version
, "1.0");
1826 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1827 content
->uuid
, chosen_name
);
1829 rv
= start_array(mdfd
, mddev
, content
,
1830 st
, ident
, best
, bestcnt
,
1831 chosen_drive
, devices
, okcnt
, sparecnt
,
1832 rebuilding_cnt
, journalcnt
,
1834 clean
, avail
, start_partial_ok
,
1836 was_forced
, expect_journal
, journal_clean
);
1837 if (rv
== 1 && !pre_exist
)
1838 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1842 wait_for(chosen_name
, mdfd
);
1846 /* There is a nasty race with 'mdadm --monitor'.
1847 * If it opens this device before we close it,
1848 * it gets an incomplete open on which IO
1849 * doesn't work and the capacity is
1851 * If we reopen (to check for layered devices)
1852 * before --monitor closes, we loose.
1854 * So: wait upto 1 second for there to be
1855 * a non-zero capacity.
1857 while (usecs
< 1000) {
1858 mdfd
= open(mddev
, O_RDONLY
);
1860 unsigned long long size
;
1861 if (get_dev_size(mdfd
, NULL
, &size
) &&
1873 /* '2' means 'OK, but not started yet' */
1874 return rv
== 2 ? 0 : rv
;
1878 int assemble_container_content(struct supertype
*st
, int mdfd
,
1879 struct mdinfo
*content
, struct context
*c
,
1880 char *chosen_name
, int *result
)
1882 struct mdinfo
*dev
, *sra
, *dev2
;
1883 int working
= 0, preexist
= 0;
1885 struct map_ent
*map
= NULL
;
1891 sysfs_init(content
, mdfd
, NULL
);
1893 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1894 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1895 if (content
->array
.major_version
== -1 &&
1896 content
->array
.minor_version
== -2 &&
1898 content
->text_version
[0] == '/')
1899 content
->text_version
[0] = '-';
1900 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1907 /* There are two types of reshape: container wide or sub-array specific
1908 * Check if metadata requests blocking container wide reshapes
1910 start_reshape
= (content
->reshape_active
&&
1911 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1912 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1914 /* Block subarray here if it is under reshape now
1915 * Do not allow for any changes in this array
1917 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1918 block_subarray(content
);
1920 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1921 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1922 if (dev2
->disk
.major
== dev
->disk
.major
&&
1923 dev2
->disk
.minor
== dev
->disk
.minor
)
1927 /* Don't want this one any more */
1928 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1930 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1934 sysfs_set_str(sra
, dev2
, "state", "remove");
1936 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1937 avail
= xcalloc(content
->array
.raid_disks
, 1);
1938 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1939 if (dev
->disk
.raid_disk
>= 0)
1940 avail
[dev
->disk
.raid_disk
] = 1;
1941 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1942 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1943 content
->reshape_active
)
1947 } else if (errno
== EEXIST
)
1951 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1953 return 1;/* Nothing new, don't try to start */
1955 map_update(&map
, fd2devnm(mdfd
),
1956 content
->text_version
,
1957 content
->uuid
, chosen_name
);
1960 if (enough(content
->array
.level
, content
->array
.raid_disks
,
1961 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
1962 if (c
->export
&& result
)
1964 else if (c
->verbose
>= 0) {
1965 pr_err("%s assembled with %d device%s",
1966 chosen_name
, preexist
+ working
,
1967 preexist
+ working
== 1 ? "":"s");
1969 fprintf(stderr
, " (%d new)", working
);
1970 fprintf(stderr
, " but not started\n");
1977 if (c
->runstop
<= 0 &&
1978 (working
+ preexist
+ expansion
) <
1979 content
->array
.working_disks
) {
1980 if (c
->export
&& result
)
1981 *result
|= INCR_UNSAFE
;
1982 else if (c
->verbose
>= 0) {
1983 pr_err("%s assembled with %d device%s",
1984 chosen_name
, preexist
+ working
,
1985 preexist
+ working
== 1 ? "":"s");
1987 fprintf(stderr
, " (%d new)", working
);
1988 fprintf(stderr
, " but not safe to start\n");
1994 if (start_reshape
) {
1995 int spare
= content
->array
.raid_disks
+ expansion
;
1996 if (restore_backup(st
, content
,
1998 spare
, &c
->backup_file
, c
->verbose
) == 1)
2001 err
= sysfs_set_str(content
, NULL
,
2002 "array_state", "readonly");
2006 if (st
->ss
->external
) {
2007 if (!mdmon_running(st
->container_devnm
))
2008 start_mdmon(st
->container_devnm
);
2009 ping_monitor(st
->container_devnm
);
2010 if (mdmon_running(st
->container_devnm
) &&
2011 st
->update_tail
== NULL
)
2012 st
->update_tail
= &st
->updates
;
2015 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2016 0, c
->freeze_reshape
);
2017 } else switch(content
->array
.level
) {
2019 case LEVEL_MULTIPATH
:
2021 err
= sysfs_set_str(content
, NULL
, "array_state",
2022 c
->readonly
? "readonly" : "active");
2025 err
= sysfs_set_str(content
, NULL
, "array_state",
2027 /* start mdmon if needed. */
2029 if (!mdmon_running(st
->container_devnm
))
2030 start_mdmon(st
->container_devnm
);
2031 ping_monitor(st
->container_devnm
);
2036 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2038 /* Block subarray here if it is not reshaped now
2039 * It has be blocked a little later to allow mdmon to switch in
2042 if (st
->ss
->external
&& content
->recovery_blocked
&&
2044 block_subarray(content
);
2046 if (c
->export
&& result
) {
2050 *result
|= INCR_YES
;
2051 } else if (c
->verbose
>= 0) {
2053 pr_err("array %s now has %d device%s",
2054 chosen_name
, working
+ preexist
,
2055 working
+ preexist
== 1 ? "":"s");
2057 pr_err("Started %s with %d device%s",
2058 chosen_name
, working
+ preexist
,
2059 working
+ preexist
== 1 ? "":"s");
2061 fprintf(stderr
, " (%d new)", working
);
2063 fprintf(stderr
, " ( + %d for expansion)",
2065 fprintf(stderr
, "\n");
2068 wait_for(chosen_name
, mdfd
);
2070 /* FIXME should have an O_EXCL and wait for read-auto */