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
;
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
);
917 /* If there are any other drives of the same vintage,
918 * add them in as well. We can't lose and we might gain
921 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
925 !devices
[j
].uptodate
&&
926 devices
[j
].i
.recovery_start
== MaxSector
&&
927 devices
[j
].i
.events
== current_events
) {
936 static int start_array(int mdfd
,
938 struct mdinfo
*content
,
939 struct supertype
*st
,
940 struct mddev_ident
*ident
,
941 int *best
, int bestcnt
,
943 struct devs
*devices
,
945 unsigned int sparecnt
,
946 unsigned int rebuilding_cnt
,
948 int clean
, char *avail
,
949 int start_partial_ok
,
956 unsigned int req_cnt
;
958 rv
= set_array_info(mdfd
, st
, content
);
960 pr_err("failed to set array info for %s: %s\n",
961 mddev
, strerror(errno
));
964 if (ident
->bitmap_fd
>= 0) {
965 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
966 pr_err("SET_BITMAP_FILE failed.\n");
969 } else if (ident
->bitmap_file
) {
970 /* From config file */
971 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
973 pr_err("Could not open bitmap file %s\n",
977 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
978 pr_err("Failed to set bitmapfile for %s\n", mddev
);
985 /* First, add the raid disks, but add the chosen one last */
986 for (i
=0; i
<= bestcnt
; i
++) {
990 if (j
== chosen_drive
)
995 if (j
>= 0 && !devices
[j
].included
) {
996 int dfd
= dev_open(devices
[j
].devname
,
999 remove_partitions(dfd
);
1002 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1005 pr_err("failed to add %s to %s: %s\n",
1009 if (i
< content
->array
.raid_disks
* 2
1014 } else if (c
->verbose
> 0)
1015 pr_err("added %s to %s as %d%s%s\n",
1016 devices
[j
].devname
, mddev
,
1017 devices
[j
].i
.disk
.raid_disk
,
1018 devices
[j
].uptodate
?"":
1019 " (possibly out of date)",
1020 (devices
[j
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))?" replacement":"");
1021 } else if (j
>= 0) {
1023 pr_err("%s is already in %s as %d\n",
1024 devices
[j
].devname
, mddev
,
1025 devices
[j
].i
.disk
.raid_disk
);
1026 } else if (c
->verbose
> 0 && i
< content
->array
.raid_disks
*2
1028 pr_err("no uptodate device for slot %d of %s\n",
1032 if (content
->array
.level
== LEVEL_CONTAINER
) {
1033 if (c
->verbose
>= 0) {
1034 pr_err("Container %s has been assembled with %d drive%s",
1035 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1036 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1037 fprintf(stderr
, " (out of %d)",
1038 content
->array
.raid_disks
);
1039 fprintf(stderr
, "\n");
1042 if (st
->ss
->validate_container
) {
1043 struct mdinfo
*devices_list
;
1044 struct mdinfo
*info_devices
= xmalloc(sizeof(struct mdinfo
)*(okcnt
+sparecnt
));
1046 devices_list
= NULL
;
1047 for (count
= 0; count
< okcnt
+sparecnt
; count
++) {
1048 info_devices
[count
] = devices
[count
].i
;
1049 info_devices
[count
].next
= devices_list
;
1050 devices_list
= &info_devices
[count
];
1052 if (st
->ss
->validate_container(devices_list
))
1053 pr_err("Mismatch detected!\n");
1057 st
->ss
->free_super(st
);
1058 sysfs_uevent(content
, "change");
1059 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1060 /* Was partial, is still partial, so signal an error
1061 * to ensure we don't retry */
1066 /* Get number of in-sync devices according to the superblock.
1067 * We must have this number to start the array without -s or -R
1069 req_cnt
= content
->array
.working_disks
;
1071 if (c
->runstop
== 1 ||
1073 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1074 content
->array
.layout
, clean
, avail
) &&
1075 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1077 /* This array is good-to-go.
1078 * If a reshape is in progress then we might need to
1079 * continue monitoring it. In that case we start
1080 * it read-only and let the grow code make it writable.
1084 if (content
->reshape_active
&&
1085 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1086 content
->delta_disks
<= 0) {
1087 if (!c
->backup_file
) {
1088 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1090 pr_err("Please provided one with \"--backup-file=...\"\n");
1092 strcmp(c
->update
, "revert-reshape") == 0)
1093 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1096 rv
= sysfs_set_str(content
, NULL
,
1097 "array_state", "readonly");
1099 rv
= Grow_continue(mdfd
, st
, content
,
1102 } else if (c
->readonly
&&
1103 sysfs_attribute_available(
1104 content
, NULL
, "array_state")) {
1105 rv
= sysfs_set_str(content
, NULL
,
1106 "array_state", "readonly");
1109 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1110 reopen_mddev(mdfd
); /* drop O_EXCL */
1112 if (c
->verbose
>= 0) {
1113 pr_err("%s has been started with %d drive%s",
1114 mddev
, okcnt
, okcnt
==1?"":"s");
1115 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1116 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1118 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1120 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1121 fprintf(stderr
, ".\n");
1123 if (content
->reshape_active
&&
1124 content
->array
.level
>= 4 &&
1125 content
->array
.level
<= 6) {
1126 /* might need to increase the size
1127 * of the stripe cache - default is 256
1129 int chunk_size
= content
->array
.chunk_size
;
1130 if (content
->reshape_active
&&
1131 content
->new_chunk
> chunk_size
)
1132 chunk_size
= content
->new_chunk
;
1133 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1134 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1136 sysfs_set_num(sra
, NULL
,
1137 "stripe_cache_size",
1138 (4 * chunk_size
/ 4096) + 1);
1142 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1143 /* If any devices did not get added
1144 * because the kernel rejected them based
1145 * on event count, try adding them
1146 * again providing the action policy is
1147 * 're-add' or greater. The bitmap
1148 * might allow them to be included, or
1149 * they will become spares.
1151 for (i
= 0; i
< bestcnt
; i
++) {
1153 if (j
>= 0 && !devices
[j
].uptodate
) {
1154 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1156 rv
= add_disk(mdfd
, st
, content
,
1158 if (rv
== 0 && c
->verbose
>= 0)
1159 pr_err("%s has been re-added.\n",
1160 devices
[j
].devname
);
1164 if (content
->array
.level
== 6 &&
1165 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1167 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1169 sysfs_set_str(sra
, NULL
,
1170 "sync_action", "repair");
1175 pr_err("failed to RUN_ARRAY %s: %s\n",
1176 mddev
, strerror(errno
));
1178 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1179 content
->array
.layout
, 1, avail
))
1180 pr_err("Not enough devices to start the array.\n");
1181 else if (!enough(content
->array
.level
,
1182 content
->array
.raid_disks
,
1183 content
->array
.layout
, clean
,
1185 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1189 if (c
->runstop
== -1) {
1190 pr_err("%s assembled from %d drive%s",
1191 mddev
, okcnt
, okcnt
==1?"":"s");
1192 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1193 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1194 fprintf(stderr
, ", but not started.\n");
1197 if (c
->verbose
>= -1) {
1198 pr_err("%s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1200 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1202 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1203 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1204 content
->array
.layout
, 1, avail
))
1205 fprintf(stderr
, " - not enough to start the array.\n");
1206 else if (!enough(content
->array
.level
,
1207 content
->array
.raid_disks
,
1208 content
->array
.layout
, clean
,
1210 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1212 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1213 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1215 fprintf(stderr
, " - need %d to start", req_cnt
);
1216 fprintf(stderr
, " (use --run to insist).\n");
1222 int Assemble(struct supertype
*st
, char *mddev
,
1223 struct mddev_ident
*ident
,
1224 struct mddev_dev
*devlist
,
1228 * The task of Assemble is to find a collection of
1229 * devices that should (according to their superblocks)
1230 * form an array, and to give this collection to the MD driver.
1231 * In Linux-2.4 and later, this involves submitting a
1232 * SET_ARRAY_INFO ioctl with no arg - to prepare
1233 * the array - and then submit a number of
1234 * ADD_NEW_DISK ioctls to add disks into
1235 * the array. Finally RUN_ARRAY might
1236 * be submitted to start the array.
1238 * Much of the work of Assemble is in finding and/or
1239 * checking the disks to make sure they look right.
1241 * If mddev is not set, then scan must be set and we
1242 * read through the config file for dev+uuid mapping
1243 * We recurse, setting mddev, for each device that
1245 * - has a valid uuid (or any uuid if !uuidset)
1247 * If mddev is set, we try to determine state of md.
1248 * check version - must be at least 0.90.0
1249 * check kernel version. must be at least 2.4.
1250 * If not, we can possibly fall back on START_ARRAY
1251 * Try to GET_ARRAY_INFO.
1252 * If possible, give up
1253 * If not, try to STOP_ARRAY just to make sure
1255 * If !uuidset and scan, look in conf-file for uuid
1256 * If not found, give up
1257 * If !devlist and scan and uuidset, get list of devs from conf-file
1260 * Check superblock - discard if bad
1261 * Check uuid (set if we don't have one) - discard if no match
1262 * Check superblock similarity if we have a superblock - discard if different
1263 * Record events, devicenum
1264 * This should give us a list of devices for the array
1265 * We should collect the most recent event number
1267 * Count disks with recent enough event count
1268 * While force && !enough disks
1269 * Choose newest rejected disks, update event count
1270 * mark clean and rewrite superblock
1273 * foreach device with recent events : ADD_NEW_DISK
1274 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1276 * Check the device numbers in superblock are right
1277 * update superblock if any changes
1284 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1285 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1286 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1287 struct devs
*devices
;
1289 int *best
= NULL
; /* indexed by raid_disk */
1292 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
;
1295 int most_recent
= 0;
1299 int start_partial_ok
= (c
->runstop
>= 0) &&
1300 (c
->force
|| devlist
==NULL
|| auto_assem
);
1302 struct mddev_dev
*tmpdev
;
1304 struct mdinfo
*content
= NULL
;
1305 struct mdinfo
*pre_exist
= NULL
;
1308 char chosen_name
[1024];
1309 struct map_ent
*map
= NULL
;
1313 * If any subdevs are listed, then any that don't
1314 * match ident are discarded. Remainder must all match and
1316 * If no subdevs, then we scan all devices in the config file, but
1317 * there must be something in the identity
1321 ident
->uuid_set
== 0 &&
1322 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1323 ident
->name
[0] == 0 &&
1324 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1325 ident
->devices
== NULL
) {
1326 pr_err("No identity information available for %s - cannot assemble.\n",
1327 mddev
? mddev
: "further assembly");
1331 if (devlist
== NULL
)
1332 devlist
= conf_get_devs();
1337 /* We come back here when doing auto-assembly and attempting some
1338 * set of devices failed. Those are now marked as ->used==2 and
1339 * we ignore them and try again
1341 if (!st
&& ident
->st
)
1344 pr_err("looking for devices for %s\n",
1345 mddev
? mddev
: "further assembly");
1349 st
->ignore_hw_compat
= 1;
1350 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1351 inargv
, auto_assem
);
1355 if (!st
|| !st
->sb
|| !content
)
1358 /* We have a full set of devices - we now need to find the
1360 * However there is a risk that we are racing with "mdadm -I"
1361 * and the array is already partially assembled - we will have
1362 * rejected any devices already in this address.
1363 * So we take a lock on the map file - to prevent further races -
1364 * and look for the uuid in there. If found and the array is
1365 * active, we abort. If found and the array is not active
1366 * we commit to that md device and add all the contained devices
1367 * to our list. We flag them so that we don't try to re-add,
1368 * but can remove if they turn out to not be wanted.
1371 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1372 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1375 mp
= map_by_uuid(&map
, content
->uuid
);
1378 /* array already exists. */
1379 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1380 if (pre_exist
->array
.level
!= UnSet
) {
1381 pr_err("Found some drive for an array that is already active: %s\n",
1383 pr_err("giving up.\n");
1386 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1387 /* We want to add this device to our list,
1388 * but it could already be there if "mdadm -I"
1389 * started *after* we checked for O_EXCL.
1390 * If we add it to the top of the list
1391 * it will be preferred over later copies.
1393 struct mddev_dev
*newdev
;
1394 char *devname
= map_dev(dv
->disk
.major
,
1399 newdev
= xmalloc(sizeof(*newdev
));
1400 newdev
->devname
= devname
;
1401 newdev
->disposition
= 'I';
1403 newdev
->next
= devlist
;
1407 strcpy(chosen_name
, mp
->path
);
1408 if (c
->verbose
> 0 || mddev
== NULL
||
1409 strcmp(mddev
, chosen_name
) != 0)
1410 pr_err("Merging with already-assembled %s\n",
1412 mdfd
= open_dev_excl(mp
->devnm
);
1414 int trustworthy
= FOREIGN
;
1415 name
= content
->name
;
1416 switch (st
->ss
->match_home(st
, c
->homehost
)
1417 ?: st
->ss
->match_home(st
, "any")) {
1419 trustworthy
= LOCAL
;
1420 name
= strchr(content
->name
, ':');
1424 name
= content
->name
;
1428 /* If the array is listed in mdadm.conf or on
1429 * command line, then we trust the name
1430 * even if the array doesn't look local
1432 trustworthy
= LOCAL
;
1435 content
->array
.level
== LEVEL_CONTAINER
) {
1436 name
= content
->text_version
;
1437 trustworthy
= METADATA
;
1440 if (name
[0] && trustworthy
!= LOCAL
&&
1441 ! c
->require_homehost
&&
1442 conf_name_is_free(name
))
1443 trustworthy
= LOCAL
;
1445 if (trustworthy
== LOCAL
&&
1447 /* Ignore 'host:' prefix of name */
1448 name
= strchr(name
, ':')+1;
1450 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1454 st
->ss
->free_super(st
);
1459 mddev
= chosen_name
;
1460 if (get_linux_version() < 2004000 ||
1461 md_get_version(mdfd
) < 9000) {
1462 pr_err("Assemble requires Linux 2.4 or later, and\n"
1463 " md driver version 0.90.0 or later.\n"
1464 " Upgrade your kernel or try --build\n");
1468 if (pre_exist
== NULL
) {
1469 if (mddev_busy(fd2devnm(mdfd
))) {
1470 pr_err("%s already active, cannot restart it!\n",
1472 for (tmpdev
= devlist
;
1473 tmpdev
&& tmpdev
->used
!= 1;
1474 tmpdev
= tmpdev
->next
)
1476 if (tmpdev
&& auto_assem
)
1477 pr_err("%s needed for %s...\n",
1478 mddev
, tmpdev
->devname
);
1481 st
->ss
->free_super(st
);
1486 /* just incase it was started but has no content */
1487 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1491 if (content
!= &info
) {
1492 /* This is a member of a container. Try starting the array. */
1494 err
= assemble_container_content(st
, mdfd
, content
, c
,
1500 /* Ok, no bad inconsistancy, we can try updating etc */
1501 devices
= xcalloc(num_devs
, sizeof(*devices
));
1502 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1503 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1504 c
, content
, mdfd
, mddev
,
1505 &most_recent
, &bestcnt
, &best
, inargv
);
1510 pr_err("no devices found for %s\n",
1513 st
->ss
->free_super(st
);
1520 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1521 st
->minor_version
= 90;
1523 st
->ss
->getinfo_super(st
, content
, NULL
);
1524 clean
= content
->array
.state
& 1;
1526 /* now we have some devices that might be suitable.
1529 avail
= xcalloc(content
->array
.raid_disks
, 1);
1534 for (i
=0; i
< bestcnt
; i
++) {
1536 int event_margin
= 1; /* always allow a difference of '1'
1537 * like the kernel does
1539 if (j
< 0) continue;
1540 /* note: we ignore error flags in multipath arrays
1541 * as they don't make sense
1543 if (content
->array
.level
!= LEVEL_MULTIPATH
)
1544 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1545 if (!(devices
[j
].i
.disk
.state
1546 & (1<<MD_DISK_FAULTY
))) {
1547 devices
[j
].uptodate
= 1;
1552 /* If this device thinks that 'most_recent' has failed, then
1553 * we must reject this device.
1555 if (j
!= most_recent
&& !c
->force
&&
1556 content
->array
.raid_disks
> 0 &&
1557 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1558 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1559 if (c
->verbose
> -1)
1560 pr_err("ignoring %s as it reports %s as failed\n",
1561 devices
[j
].devname
, devices
[most_recent
].devname
);
1565 /* Require event counter to be same as, or just less than,
1566 * most recent. If it is bigger, it must be a stray spare and
1567 * should be ignored.
1569 if (devices
[j
].i
.events
+event_margin
>=
1570 devices
[most_recent
].i
.events
&&
1571 devices
[j
].i
.events
<=
1572 devices
[most_recent
].i
.events
1574 devices
[j
].uptodate
= 1;
1575 if (i
< content
->array
.raid_disks
* 2) {
1576 if (devices
[j
].i
.recovery_start
== MaxSector
||
1577 (content
->reshape_active
&&
1578 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1592 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1593 avail
, most_recent
, st
, c
);
1598 /* Now we want to look at the superblock which the kernel will base things on
1599 * and compare the devices that we think are working with the devices that the
1600 * superblock thinks are working.
1601 * If there are differences and --force is given, then update this chosen
1605 st
->ss
->free_super(st
);
1606 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1612 if (!devices
[j
].uptodate
)
1614 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1617 if ((fd
=dev_open(devices
[j
].devname
,
1618 devices
[j
].included
? O_RDONLY
1619 : (O_RDONLY
|O_EXCL
)))< 0) {
1620 pr_err("Cannot open %s: %s\n",
1621 devices
[j
].devname
, strerror(errno
));
1626 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1628 pr_err("RAID superblock has disappeared from %s\n",
1629 devices
[j
].devname
);
1636 if (st
->sb
== NULL
) {
1637 pr_err("No suitable drives found for %s\n", mddev
);
1642 st
->ss
->getinfo_super(st
, content
, NULL
);
1644 sysfs_init(content
, mdfd
, NULL
);
1646 for (i
=0; i
<bestcnt
; i
++) {
1648 unsigned int desired_state
;
1650 if (i
>= content
->array
.raid_disks
* 2)
1653 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1655 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1659 if (!devices
[j
].uptodate
)
1662 devices
[j
].i
.disk
.state
= desired_state
;
1663 if (!(devices
[j
].i
.array
.state
& 1))
1666 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1667 c
->verbose
, 0, NULL
)) {
1669 if (c
->verbose
>= 0)
1670 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1671 j
, mddev
, devices
[j
].devname
);
1674 if (c
->verbose
>= -1)
1675 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1676 i
, mddev
, devices
[j
].devname
);
1680 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1681 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1686 if (c
->force
&& !clean
&&
1687 !enough(content
->array
.level
, content
->array
.raid_disks
,
1688 content
->array
.layout
, clean
,
1690 change
+= st
->ss
->update_super(st
, content
, "force-array",
1691 devices
[chosen_drive
].devname
, c
->verbose
,
1699 fd
= dev_open(devices
[chosen_drive
].devname
,
1700 devices
[chosen_drive
].included
?
1701 O_RDWR
: (O_RDWR
|O_EXCL
));
1703 pr_err("Could not open %s for write - cannot Assemble array.\n",
1704 devices
[chosen_drive
].devname
);
1709 if (st
->ss
->store_super(st
, fd
)) {
1711 pr_err("Could not re-write superblock on %s\n",
1712 devices
[chosen_drive
].devname
);
1717 if (c
->verbose
>= 0)
1718 pr_err("Marking array %s as 'clean'\n",
1723 /* If we are in the middle of a reshape we may need to restore saved data
1724 * that was moved aside due to the reshape overwriting live data
1725 * The code of doing this lives in Grow.c
1728 if (content
->reshape_active
&&
1729 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1731 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1733 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1735 if (!c
->backup_file
)
1736 c
->backup_file
= locate_backup(content
->sys_name
);
1737 enable_fds(bestcnt
/2);
1738 for (i
= 0; i
< bestcnt
/2; i
++) {
1741 fdlist
[i
] = dev_open(devices
[j
].devname
,
1743 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1744 if (fdlist
[i
] < 0) {
1745 pr_err("Could not open %s for write - cannot Assemble array.\n",
1746 devices
[j
].devname
);
1754 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1755 err
= st
->ss
->recover_backup(st
, content
);
1757 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1758 c
->backup_file
, c
->verbose
> 0);
1759 if (err
&& c
->invalid_backup
) {
1761 pr_err("continuing without restoring backup\n");
1767 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1771 pr_err("Failed to restore critical section for reshape, sorry.\n");
1772 if (c
->backup_file
== NULL
)
1773 cont_err("Possibly you needed to specify the --backup-file\n");
1781 /* Almost ready to actually *do* something */
1782 /* First, fill in the map, so that udev can find our name
1783 * as soon as we become active.
1785 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1786 content
->array
.major_version
= 1;
1787 content
->array
.minor_version
= 0;
1788 strcpy(content
->text_version
, "1.0");
1791 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1792 content
->uuid
, chosen_name
);
1794 rv
= start_array(mdfd
, mddev
, content
,
1795 st
, ident
, best
, bestcnt
,
1796 chosen_drive
, devices
, okcnt
, sparecnt
,
1799 clean
, avail
, start_partial_ok
,
1802 if (rv
== 1 && !pre_exist
)
1803 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1807 wait_for(chosen_name
, mdfd
);
1811 /* There is a nasty race with 'mdadm --monitor'.
1812 * If it opens this device before we close it,
1813 * it gets an incomplete open on which IO
1814 * doesn't work and the capacity is
1816 * If we reopen (to check for layered devices)
1817 * before --monitor closes, we loose.
1819 * So: wait upto 1 second for there to be
1820 * a non-zero capacity.
1822 while (usecs
< 1000) {
1823 mdfd
= open(mddev
, O_RDONLY
);
1825 unsigned long long size
;
1826 if (get_dev_size(mdfd
, NULL
, &size
) &&
1838 /* '2' means 'OK, but not started yet' */
1839 return rv
== 2 ? 0 : rv
;
1843 int assemble_container_content(struct supertype
*st
, int mdfd
,
1844 struct mdinfo
*content
, struct context
*c
,
1845 char *chosen_name
, int *result
)
1847 struct mdinfo
*dev
, *sra
, *dev2
;
1848 int working
= 0, preexist
= 0;
1850 struct map_ent
*map
= NULL
;
1856 sysfs_init(content
, mdfd
, NULL
);
1858 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1859 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1860 if (content
->array
.major_version
== -1 &&
1861 content
->array
.minor_version
== -2 &&
1863 content
->text_version
[0] == '/')
1864 content
->text_version
[0] = '-';
1865 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1872 /* There are two types of reshape: container wide or sub-array specific
1873 * Check if metadata requests blocking container wide reshapes
1875 start_reshape
= (content
->reshape_active
&&
1876 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1877 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1879 /* Block subarray here if it is under reshape now
1880 * Do not allow for any changes in this array
1882 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1883 block_subarray(content
);
1885 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1886 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1887 if (dev2
->disk
.major
== dev
->disk
.major
&&
1888 dev2
->disk
.minor
== dev
->disk
.minor
)
1892 /* Don't want this one any more */
1893 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1895 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1899 sysfs_set_str(sra
, dev2
, "state", "remove");
1901 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1902 avail
= xcalloc(content
->array
.raid_disks
, 1);
1903 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1904 if (dev
->disk
.raid_disk
>= 0)
1905 avail
[dev
->disk
.raid_disk
] = 1;
1906 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1907 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1908 content
->reshape_active
)
1912 } else if (errno
== EEXIST
)
1916 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1918 return 1;/* Nothing new, don't try to start */
1920 map_update(&map
, fd2devnm(mdfd
),
1921 content
->text_version
,
1922 content
->uuid
, chosen_name
);
1925 if (enough(content
->array
.level
, content
->array
.raid_disks
,
1926 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
1927 if (c
->export
&& result
)
1929 else if (c
->verbose
>= 0) {
1930 pr_err("%s assembled with %d device%s",
1931 chosen_name
, preexist
+ working
,
1932 preexist
+ working
== 1 ? "":"s");
1934 fprintf(stderr
, " (%d new)", working
);
1935 fprintf(stderr
, " but not started\n");
1942 if (c
->runstop
<= 0 &&
1943 (working
+ preexist
+ expansion
) <
1944 content
->array
.working_disks
) {
1945 if (c
->export
&& result
)
1946 *result
|= INCR_UNSAFE
;
1947 else if (c
->verbose
>= 0) {
1948 pr_err("%s assembled with %d device%s",
1949 chosen_name
, preexist
+ working
,
1950 preexist
+ working
== 1 ? "":"s");
1952 fprintf(stderr
, " (%d new)", working
);
1953 fprintf(stderr
, " but not safe to start\n");
1959 if (start_reshape
) {
1960 int spare
= content
->array
.raid_disks
+ expansion
;
1961 if (restore_backup(st
, content
,
1963 spare
, &c
->backup_file
, c
->verbose
) == 1)
1966 err
= sysfs_set_str(content
, NULL
,
1967 "array_state", "readonly");
1971 if (st
->ss
->external
) {
1972 if (!mdmon_running(st
->container_devnm
))
1973 start_mdmon(st
->container_devnm
);
1974 ping_monitor(st
->container_devnm
);
1975 if (mdmon_running(st
->container_devnm
) &&
1976 st
->update_tail
== NULL
)
1977 st
->update_tail
= &st
->updates
;
1980 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
1981 0, c
->freeze_reshape
);
1982 } else switch(content
->array
.level
) {
1984 case LEVEL_MULTIPATH
:
1986 err
= sysfs_set_str(content
, NULL
, "array_state",
1987 c
->readonly
? "readonly" : "active");
1990 err
= sysfs_set_str(content
, NULL
, "array_state",
1992 /* start mdmon if needed. */
1994 if (!mdmon_running(st
->container_devnm
))
1995 start_mdmon(st
->container_devnm
);
1996 ping_monitor(st
->container_devnm
);
2001 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2003 /* Block subarray here if it is not reshaped now
2004 * It has be blocked a little later to allow mdmon to switch in
2007 if (st
->ss
->external
&& content
->recovery_blocked
&&
2009 block_subarray(content
);
2011 if (c
->export
&& result
) {
2015 *result
|= INCR_YES
;
2016 } else if (c
->verbose
>= 0) {
2018 pr_err("array %s now has %d device%s",
2019 chosen_name
, working
+ preexist
,
2020 working
+ preexist
== 1 ? "":"s");
2022 pr_err("Started %s with %d device%s",
2023 chosen_name
, working
+ preexist
,
2024 working
+ preexist
== 1 ? "":"s");
2026 fprintf(stderr
, " (%d new)", working
);
2028 fprintf(stderr
, " ( + %d for expansion)",
2030 fprintf(stderr
, "\n");
2033 wait_for(chosen_name
, mdfd
);
2035 /* FIXME should have an O_EXCL and wait for read-auto */