2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 static int name_matches(char *found
, char *required
, char *homehost
, int require_homehost
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
36 if (strcmp(found
, required
)==0)
38 sep
= strchr(found
, ':');
42 if (strncmp(found
, "any:", 4) == 0 ||
43 (homehost
&& strcmp(homehost
, "any") == 0) ||
45 (homehost
&& strlen(homehost
) == l
&&
46 strncmp(found
, homehost
, l
) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep
+1, required
) == 0)
54 static int is_member_busy(char *metadata_version
)
56 /* check if the given member array is active */
57 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
58 struct mdstat_ent
*ent
;
61 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
62 if (ent
->metadata_version
== NULL
)
64 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
66 if (!is_subarray(&ent
->metadata_version
[9]))
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
79 static int ident_matches(struct mddev_ident
*ident
,
80 struct mdinfo
*content
,
81 struct supertype
*tst
,
82 char *homehost
, int require_homehost
,
83 char *update
, char *devname
)
86 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
87 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0 &&
88 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
90 pr_err("%s has wrong uuid.\n", devname
);
93 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
94 name_matches(content
->name
, ident
->name
, homehost
, require_homehost
)==0) {
96 pr_err("%s has wrong name.\n", devname
);
99 if (ident
->super_minor
!= UnSet
&&
100 ident
->super_minor
!= content
->array
.md_minor
) {
102 pr_err("%s has wrong super-minor.\n",
106 if (ident
->level
!= UnSet
&&
107 ident
->level
!= content
->array
.level
) {
109 pr_err("%s has wrong raid level.\n",
113 if (ident
->raid_disks
!= UnSet
&&
114 content
->array
.raid_disks
!= 0 && /* metadata doesn't know how many to expect */
115 ident
->raid_disks
!= content
->array
.raid_disks
) {
117 pr_err("%s requires wrong number of drives.\n",
121 if (ident
->member
&& ident
->member
[0]) {
122 /* content->text_version must match */
123 char *s
= strchr(content
->text_version
+1, '/');
126 pr_err("%s is not a container and one is required.\n",
129 } else if (strcmp(ident
->member
, s
+1) != 0) {
131 pr_err("skipping wrong member %s is %s\n",
132 content
->text_version
, devname
);
139 static int select_devices(struct mddev_dev
*devlist
,
140 struct mddev_ident
*ident
,
141 struct supertype
**stp
,
142 struct mdinfo
**contentp
,
144 int inargv
, int auto_assem
)
146 struct mddev_dev
*tmpdev
;
148 struct supertype
*st
= *stp
;
149 struct mdinfo
*content
= NULL
;
150 int report_mismatch
= ((inargv
&& c
->verbose
>= 0) || c
->verbose
> 0);
151 struct domainlist
*domains
= NULL
;
154 tmpdev
= devlist
; num_devs
= 0;
160 tmpdev
->disposition
= 0;
161 tmpdev
= tmpdev
->next
;
164 /* first walk the list of devices to find a consistent set
165 * that match the criterea, if that is possible.
166 * We flag the ones we like with 'used'.
168 for (tmpdev
= devlist
;
170 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
171 char *devname
= tmpdev
->devname
;
173 struct supertype
*tst
;
174 struct dev_policy
*pol
= NULL
;
175 int found_container
= 0;
177 if (tmpdev
->used
> 1)
180 if (ident
->container
) {
181 if (ident
->container
[0] == '/' &&
182 !same_dev(ident
->container
, devname
)) {
184 pr_err("%s is not the container required (%s)\n",
185 devname
, ident
->container
);
188 } else if (ident
->devices
&&
189 !match_oneof(ident
->devices
, devname
)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
195 pr_err("%s is not one of %s\n", devname
, ident
->devices
);
201 dfd
= dev_open(devname
, O_RDONLY
);
204 pr_err("cannot open device %s: %s\n",
205 devname
, strerror(errno
));
207 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
209 } else if (must_be_container(dfd
)) {
211 /* already found some components, this cannot
215 pr_err("%s is a container, but we are looking for components\n",
218 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
220 pr_err("not a recognisable container: %s\n",
223 } else if (!tst
->ss
->load_container
||
224 tst
->ss
->load_container(tst
, dfd
, NULL
)) {
226 pr_err("no correct container type: %s\n",
229 } else if (auto_assem
&&
230 !conf_test_metadata(tst
->ss
->name
,
231 (pol
= devid_policy(rdev
)),
232 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
234 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
235 devname
, tst
->ss
->name
);
240 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
242 pr_err("no recogniseable superblock on %s\n",
245 } else if ((tst
->ignore_hw_compat
= 0),
246 tst
->ss
->load_super(tst
, dfd
,
247 report_mismatch
? devname
: NULL
)) {
249 pr_err("no RAID superblock on %s\n",
252 } else if (tst
->ss
->compare_super
== NULL
) {
254 pr_err("Cannot assemble %s metadata on %s\n",
255 tst
->ss
->name
, devname
);
257 } else if (auto_assem
&& st
== NULL
&&
258 !conf_test_metadata(tst
->ss
->name
,
259 (pol
= devid_policy(rdev
)),
260 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
262 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
263 devname
, tst
->ss
->name
);
267 if (dfd
>= 0) close(dfd
);
268 if (tmpdev
->used
== 2) {
269 if (auto_assem
|| !inargv
)
270 /* Ignore unrecognised devices during auto-assembly */
272 if (ident
->uuid_set
|| ident
->name
[0] ||
273 ident
->super_minor
!= UnSet
)
274 /* Ignore unrecognised device if looking for
278 pr_err("%s has no superblock - assembly aborted\n",
281 st
->ss
->free_super(st
);
282 dev_policy_free(pol
);
283 domain_free(domains
);
287 if (found_container
) {
288 /* tmpdev is a container. We need to be either
289 * looking for a member, or auto-assembling
291 /* should be safe to try an exclusive open now, we
292 * have rejected anything that some other mdadm might
295 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
298 pr_err("%s is busy - skipping\n", devname
);
303 if (ident
->container
&& ident
->container
[0] != '/') {
308 tst
->ss
->getinfo_super(tst
, content
, NULL
);
310 if (!parse_uuid(ident
->container
, uuid
) ||
311 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
313 pr_err("%s has wrong UUID to be required container\n",
318 /* It is worth looking inside this container.
321 pr_err("looking in container %s\n",
324 for (content
= tst
->ss
->container_content(tst
, NULL
);
326 content
= content
->next
) {
328 if (!ident_matches(ident
, content
, tst
,
329 c
->homehost
, c
->require_homehost
,
331 report_mismatch
? devname
: NULL
))
332 /* message already printed */;
333 else if (is_member_busy(content
->text_version
)) {
335 pr_err("member %s in %s is already assembled\n",
336 content
->text_version
,
338 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
339 /* do not assemble arrays with unsupported configurations */
340 pr_err("Cannot activate member %s in %s.\n",
341 content
->text_version
,
348 goto loop
; /* empty container */
351 st
= tst
; tst
= NULL
;
352 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
353 pr_err("%s is a container, but is not only device given: confused and aborting\n",
355 st
->ss
->free_super(st
);
356 dev_policy_free(pol
);
357 domain_free(domains
);
361 pr_err("found match on member %s in %s\n",
362 content
->text_version
, devname
);
364 /* make sure we finished the loop */
369 tst
->ss
->getinfo_super(tst
, content
, NULL
);
371 if (!ident_matches(ident
, content
, tst
,
372 c
->homehost
, c
->require_homehost
,
374 report_mismatch
? devname
: NULL
))
378 /* Never auto-assemble things that conflict
379 * with mdadm.conf in some way
381 struct mddev_ident
*match
;
384 match
= conf_match(tst
, content
, devname
,
385 report_mismatch
? c
->verbose
: -1,
387 if (!match
&& rv
== 2)
389 if (match
&& match
->devname
&&
390 strcasecmp(match
->devname
, "<ignore>") == 0) {
392 pr_err("%s is a member of an explicitly ignored array\n",
396 if (match
&& !ident_matches(match
, content
, tst
,
397 c
->homehost
, c
->require_homehost
,
399 report_mismatch
? devname
: NULL
))
400 /* Array exists in mdadm.conf but some
401 * details don't match, so reject it
406 /* should be safe to try an exclusive open now, we
407 * have rejected anything that some other mdadm might
410 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
413 pr_err("%s is busy - skipping\n", devname
);
420 if (st
->minor_version
== -1)
421 st
->minor_version
= tst
->minor_version
;
423 if (memcmp(content
->uuid
, uuid_zero
,
424 sizeof(int[4])) == 0) {
425 /* this is a floating spare. It cannot define
426 * an array unless there are no more arrays of
427 * this type to be found. It can be included
428 * in an array of this type though.
434 if (st
->ss
!= tst
->ss
||
435 st
->minor_version
!= tst
->minor_version
||
436 st
->ss
->compare_super(st
, tst
) != 0) {
437 /* Some mismatch. If exactly one array matches this host,
438 * we can resolve on that one.
439 * Or, if we are auto assembling, we just ignore the second
445 int first
= st
->ss
->match_home(st
, c
->homehost
);
446 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
448 (first
== 1 || last
== 1)) {
449 /* We can do something */
450 if (first
) {/* just ignore this one */
452 pr_err("%s misses out due to wrong homehost\n",
455 } else { /* reject all those sofar */
456 struct mddev_dev
*td
;
458 pr_err("%s overrides previous devices due to good homehost\n",
460 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
468 pr_err("superblock on %s doesn't match others - assembly aborted\n",
470 tst
->ss
->free_super(tst
);
471 st
->ss
->free_super(st
);
472 dev_policy_free(pol
);
473 domain_free(domains
);
479 /* Collect domain information from members only */
480 if (tmpdev
&& tmpdev
->used
== 1) {
482 pol
= devid_policy(rdev
);
483 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
485 dev_policy_free(pol
);
488 tst
->ss
->free_super(tst
);
491 /* Check if we found some imsm spares but no members */
494 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
496 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
497 if (tmpdev
->used
!= 3)
503 /* we need sb from one of the spares */
504 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
506 st
->ss
->load_super(st
, dfd
, NULL
))
513 /* Now reject spares that don't match domains of identified members */
514 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
515 if (tmpdev
->used
!= 3)
517 if (!stat_is_blkdev(tmpdev
->devname
, &rdev
)) {
520 struct dev_policy
*pol
= devid_policy(rdev
);
521 int dt
= domain_test(domains
, pol
, NULL
);
522 if (inargv
&& dt
!= 0)
523 /* take this spare as domains match
524 * if there are any */
526 else if (!inargv
&& dt
== 1)
527 /* device wasn't explicitly listed, so need
528 * explicit domain match - which we have */
531 /* if domains don't match mark as unused */
533 dev_policy_free(pol
);
536 domain_free(domains
);
538 if (st
&& st
->sb
&& content
== *contentp
)
539 st
->ss
->getinfo_super(st
, content
, NULL
);
547 int uptodate
; /* set once we decide that this device is as
548 * recent as everything else in the array.
550 int included
; /* set if the device is already in the array
551 * due to a previous '-I'
556 static int load_devices(struct devs
*devices
, char *devmap
,
557 struct mddev_ident
*ident
, struct supertype
**stp
,
558 struct mddev_dev
*devlist
, struct context
*c
,
559 struct mdinfo
*content
,
560 int mdfd
, char *mddev
,
561 int *most_recentp
, int *bestcntp
, int **bestp
,
564 struct mddev_dev
*tmpdev
;
568 int most_recent
= -1;
571 struct supertype
*st
= *stp
;
573 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
574 char *devname
= tmpdev
->devname
;
576 struct supertype
*tst
;
580 if (tmpdev
->used
!= 1)
582 /* looks like a good enough match to update the super block if needed */
584 /* prepare useful information in info structures */
589 if (strcmp(c
->update
, "uuid") == 0 && !ident
->uuid_set
)
590 random_uuid((__u8
*)ident
->uuid
);
592 if (strcmp(c
->update
, "ppl") == 0 &&
593 ident
->bitmap_fd
>= 0) {
594 pr_err("PPL is not compatible with bitmap\n");
598 dfd
= dev_open(devname
,
599 tmpdev
->disposition
== 'I'
600 ? O_RDWR
: (O_RDWR
|O_EXCL
));
603 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
604 pr_err("cannot re-read metadata from %s - aborting\n",
611 tst
->ss
->free_super(tst
);
616 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
618 memcpy(content
->uuid
, ident
->uuid
, 16);
619 strcpy(content
->name
, ident
->name
);
620 content
->array
.md_minor
= minor(stb2
.st_rdev
);
622 if (strcmp(c
->update
, "byteorder") == 0)
624 else if (strcmp(c
->update
, "home-cluster") == 0) {
625 tst
->cluster_name
= c
->homecluster
;
626 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
627 } else if (strcmp(c
->update
, "nodes") == 0) {
628 tst
->nodes
= c
->nodes
;
629 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
630 } else if (strcmp(c
->update
, "revert-reshape") == 0 &&
632 err
= tst
->ss
->update_super(tst
, content
,
633 "revert-reshape-nobackup",
638 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
644 pr_err("--update=%s not understood for %s metadata\n",
645 c
->update
, tst
->ss
->name
);
646 tst
->ss
->free_super(tst
);
655 if (strcmp(c
->update
, "uuid")==0 &&
658 memcpy(ident
->uuid
, content
->uuid
, 16);
660 if (tst
->ss
->store_super(tst
, dfd
))
661 pr_err("Could not re-write superblock on %s.\n",
664 if (strcmp(c
->update
, "uuid")==0 &&
665 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
666 if (bitmap_update_uuid(ident
->bitmap_fd
,
668 tst
->ss
->swapuuid
) != 0)
669 pr_err("Could not update uuid on external bitmap.\n");
674 dfd
= dev_open(devname
,
675 tmpdev
->disposition
== 'I'
676 ? O_RDWR
: (O_RDWR
|O_EXCL
));
679 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
680 pr_err("cannot re-read metadata from %s - aborting\n",
687 tst
->ss
->free_super(tst
);
692 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
699 pr_err("%s is identified as a member of %s, slot %d%s.\n",
700 devname
, mddev
, content
->disk
.raid_disk
,
701 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
702 devices
[devcnt
].devname
= devname
;
703 devices
[devcnt
].uptodate
= 0;
704 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
705 devices
[devcnt
].i
= *content
;
706 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
707 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
709 if (devices
[devcnt
].i
.disk
.state
== 6) {
710 if (most_recent
< 0 ||
711 devices
[devcnt
].i
.events
712 > devices
[most_recent
].i
.events
) {
713 struct supertype
*tmp
= tst
;
716 most_recent
= devcnt
;
719 tst
->ss
->free_super(tst
);
722 if (content
->array
.level
== LEVEL_MULTIPATH
)
723 /* with multipath, the raid_disk from the superblock is meaningless */
726 i
= devices
[devcnt
].i
.disk
.raid_disk
;
727 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
728 if (nextspare
< content
->array
.raid_disks
*2)
729 nextspare
= content
->array
.raid_disks
*2;
732 /* i is raid_disk - double it so there is room for
735 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
737 if (i
>= content
->array
.raid_disks
*2 &&
743 int newbestcnt
= i
+10;
744 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
746 for (c
=0; c
< newbestcnt
; c
++)
748 newbest
[c
] = best
[c
];
753 bestcnt
= newbestcnt
;
756 devices
[best
[i
]].i
.events
==
757 devices
[devcnt
].i
.events
&&
758 (devices
[best
[i
]].i
.disk
.minor
!=
759 devices
[devcnt
].i
.disk
.minor
) &&
761 content
->array
.level
!= LEVEL_MULTIPATH
) {
762 /* two different devices with identical superblock.
763 * Could be a mis-detection caused by overlapping
764 * partitions. fail-safe.
766 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
767 " If they are really different, please --zero the superblock on one\n"
768 " If they are the same or overlap, please remove one from %s.\n",
769 devices
[best
[i
]].devname
, devname
,
770 inargv
? "the list" :
771 "the\n DEVICE list in mdadm.conf"
779 if (best
[i
] == -1 || (devices
[best
[i
]].i
.events
780 < devices
[devcnt
].i
.events
))
782 else if (st
->ss
== &super_imsm
)
787 if (most_recent
>= 0)
788 *most_recentp
= most_recent
;
795 static int force_array(struct mdinfo
*content
,
796 struct devs
*devices
,
797 int *best
, int bestcnt
, char *avail
,
799 struct supertype
*st
,
803 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
804 content
->array
.layout
, 1,
806 (content
->reshape_active
&& content
->delta_disks
> 0 &&
807 !enough(content
->array
.level
, (content
->array
.raid_disks
808 - content
->delta_disks
),
809 content
->new_layout
, 1, avail
))) {
810 /* Choose the newest best drive which is
811 * not up-to-date, update the superblock
815 struct supertype
*tst
;
816 unsigned long long current_events
;
817 int chosen_drive
= -1;
821 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
826 if (devices
[j
].uptodate
)
828 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
830 if (!devices
[j
].i
.reshape_active
||
831 devices
[j
].i
.delta_disks
<= 0)
833 /* When increasing number of devices, an
834 * added device also appears to be
835 * recovering. It is safe to include it
836 * as long as it won't be a source of
838 * For now, just allow for last data
839 * devices in RAID4 or last devices in RAID4/5/6.
841 delta
= devices
[j
].i
.delta_disks
;
842 if (devices
[j
].i
.array
.level
>= 4 &&
843 devices
[j
].i
.array
.level
<= 6 &&
844 i
/2 >= content
->array
.raid_disks
- delta
)
846 else if (devices
[j
].i
.array
.level
== 4 &&
847 i
/2 >= content
->array
.raid_disks
- delta
- 1)
851 } else if (devices
[j
].i
.reshape_active
!=
852 content
->reshape_active
||
853 (devices
[j
].i
.reshape_active
&&
854 devices
[j
].i
.reshape_progress
!=
855 content
->reshape_progress
))
856 /* Here, it may be a source of data. If two
857 * devices claim different progresses, it
858 * means that reshape boundaries differ for
859 * their own devices. Kernel will only treat
860 * the first one as reshape progress and
861 * go on. It may cause disaster, so avoid it.
864 if (chosen_drive
< 0 ||
866 > devices
[chosen_drive
].i
.events
)
869 if (chosen_drive
< 0)
871 current_events
= devices
[chosen_drive
].i
.events
;
874 pr_err("forcing event count in %s(%d) from %d upto %d\n",
875 devices
[chosen_drive
].devname
,
876 devices
[chosen_drive
].i
.disk
.raid_disk
,
877 (int)(devices
[chosen_drive
].i
.events
),
878 (int)(devices
[most_recent
].i
.events
));
879 fd
= dev_open(devices
[chosen_drive
].devname
,
880 devices
[chosen_drive
].included
? O_RDWR
883 pr_err("Couldn't open %s for write - not updating\n",
884 devices
[chosen_drive
].devname
);
885 devices
[chosen_drive
].i
.events
= 0;
889 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
891 pr_err("RAID superblock disappeared from %s - not updating.\n",
892 devices
[chosen_drive
].devname
);
893 devices
[chosen_drive
].i
.events
= 0;
896 content
->events
= devices
[most_recent
].i
.events
;
897 tst
->ss
->update_super(tst
, content
, "force-one",
898 devices
[chosen_drive
].devname
, c
->verbose
,
901 if (tst
->ss
->store_super(tst
, fd
)) {
903 pr_err("Could not re-write superblock on %s\n",
904 devices
[chosen_drive
].devname
);
905 devices
[chosen_drive
].i
.events
= 0;
906 tst
->ss
->free_super(tst
);
910 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
911 devices
[chosen_drive
].uptodate
= 1;
912 avail
[chosen_drive
] = 1;
914 tst
->ss
->free_super(tst
);
915 /* If there are any other drives of the same vintage,
916 * add them in as well. We can't lose and we might gain
919 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
923 !devices
[j
].uptodate
&&
924 devices
[j
].i
.recovery_start
== MaxSector
&&
925 devices
[j
].i
.events
== current_events
&&
926 ((!devices
[j
].i
.reshape_active
&&
927 !content
->reshape_active
) ||
928 (devices
[j
].i
.reshape_active
==
929 content
->reshape_active
&&
930 devices
[j
].i
.reshape_progress
==
931 content
->reshape_progress
))) {
940 static int start_array(int mdfd
,
942 struct mdinfo
*content
,
943 struct supertype
*st
,
944 struct mddev_ident
*ident
,
945 int *best
, int bestcnt
,
947 struct devs
*devices
,
949 unsigned int sparecnt
,
950 unsigned int rebuilding_cnt
,
951 unsigned int journalcnt
,
953 int clean
, char *avail
,
954 int start_partial_ok
,
961 unsigned int req_cnt
;
963 if (content
->journal_device_required
&& (content
->journal_clean
== 0)) {
965 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
968 pr_err("Journal is missing or stale, starting array read only.\n");
972 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
975 rv
= set_array_info(mdfd
, st
, content
);
977 pr_err("failed to set array info for %s: %s\n",
978 mddev
, strerror(errno
));
981 if (ident
->bitmap_fd
>= 0) {
982 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
983 pr_err("SET_BITMAP_FILE failed.\n");
986 } else if (ident
->bitmap_file
) {
987 /* From config file */
988 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
990 pr_err("Could not open bitmap file %s\n",
994 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
995 pr_err("Failed to set bitmapfile for %s\n", mddev
);
1002 /* First, add the raid disks, but add the chosen one last */
1003 for (i
= 0; i
<= bestcnt
; i
++) {
1007 if (j
== chosen_drive
)
1012 if (j
>= 0 && !devices
[j
].included
) {
1015 dfd
= dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1017 remove_partitions(dfd
);
1020 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1023 pr_err("failed to add %s to %s: %s\n",
1024 devices
[j
].devname
, mddev
,
1026 if (i
< content
->array
.raid_disks
* 2 ||
1031 } else if (c
->verbose
> 0) {
1032 pr_err("added %s to %s as %d%s%s\n",
1033 devices
[j
].devname
, mddev
,
1034 devices
[j
].i
.disk
.raid_disk
,
1035 devices
[j
].uptodate
?"":
1036 " (possibly out of date)",
1037 (devices
[j
].i
.disk
.state
&
1038 (1<<MD_DISK_REPLACEMENT
)) ?
1041 } else if (j
>= 0) {
1043 pr_err("%s is already in %s as %d\n",
1044 devices
[j
].devname
, mddev
,
1045 devices
[j
].i
.disk
.raid_disk
);
1046 } else if (c
->verbose
> 0 &&
1047 i
< content
->array
.raid_disks
* 2 && (i
& 1) == 0)
1048 pr_err("no uptodate device for slot %d of %s\n",
1052 if (content
->array
.level
== LEVEL_CONTAINER
) {
1053 if (c
->verbose
>= 0) {
1054 pr_err("Container %s has been assembled with %d drive%s",
1055 mddev
, okcnt
+ sparecnt
+ journalcnt
,
1056 okcnt
+ sparecnt
+ journalcnt
== 1 ? "" : "s");
1057 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1058 fprintf(stderr
, " (out of %d)",
1059 content
->array
.raid_disks
);
1060 fprintf(stderr
, "\n");
1063 if (st
->ss
->validate_container
) {
1064 struct mdinfo
*devices_list
;
1065 struct mdinfo
*info_devices
;
1068 devices_list
= NULL
;
1069 info_devices
= xmalloc(sizeof(struct mdinfo
) *
1070 (okcnt
+ sparecnt
));
1071 for (count
= 0; count
< okcnt
+ sparecnt
; count
++) {
1072 info_devices
[count
] = devices
[count
].i
;
1073 info_devices
[count
].next
= devices_list
;
1074 devices_list
= &info_devices
[count
];
1076 if (st
->ss
->validate_container(devices_list
))
1077 pr_err("Mismatch detected!\n");
1081 st
->ss
->free_super(st
);
1082 sysfs_uevent(content
, "change");
1083 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1084 /* Was partial, is still partial, so signal an error
1085 * to ensure we don't retry */
1090 /* Get number of in-sync devices according to the superblock.
1091 * We must have this number to start the array without -s or -R
1093 req_cnt
= content
->array
.working_disks
;
1095 if (c
->runstop
== 1 ||
1097 (enough(content
->array
.level
, content
->array
.raid_disks
,
1098 content
->array
.layout
, clean
, avail
) &&
1099 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)))) {
1100 /* This array is good-to-go.
1101 * If a reshape is in progress then we might need to
1102 * continue monitoring it. In that case we start
1103 * it read-only and let the grow code make it writable.
1107 if (content
->reshape_active
&&
1108 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1109 content
->delta_disks
<= 0) {
1110 if (!c
->backup_file
) {
1111 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1113 pr_err("Please provided one with \"--backup-file=...\"\n");
1115 strcmp(c
->update
, "revert-reshape") == 0)
1116 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1119 rv
= sysfs_set_str(content
, NULL
,
1120 "array_state", "readonly");
1122 rv
= Grow_continue(mdfd
, st
, content
,
1125 } else if (c
->readonly
&&
1126 sysfs_attribute_available(content
, NULL
,
1128 rv
= sysfs_set_str(content
, NULL
,
1129 "array_state", "readonly");
1131 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1132 reopen_mddev(mdfd
); /* drop O_EXCL */
1134 if (c
->verbose
>= 0) {
1135 pr_err("%s has been started with %d drive%s",
1136 mddev
, okcnt
, okcnt
==1?"":"s");
1137 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1138 fprintf(stderr
, " (out of %d)",
1139 content
->array
.raid_disks
);
1141 fprintf(stderr
, "%s %d rebuilding",
1142 sparecnt
?",":" and",
1145 fprintf(stderr
, " and %d spare%s",
1147 sparecnt
== 1 ? "" : "s");
1148 if (content
->journal_clean
)
1149 fprintf(stderr
, " and %d journal",
1151 fprintf(stderr
, ".\n");
1153 if (content
->reshape_active
&&
1154 content
->array
.level
>= 4 &&
1155 content
->array
.level
<= 6) {
1156 /* might need to increase the size
1157 * of the stripe cache - default is 256
1159 int chunk_size
= content
->array
.chunk_size
;
1161 if (content
->reshape_active
&&
1162 content
->new_chunk
> chunk_size
)
1163 chunk_size
= content
->new_chunk
;
1164 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1167 sra
= sysfs_read(mdfd
, NULL
, 0);
1169 sysfs_set_num(sra
, NULL
,
1170 "stripe_cache_size",
1171 (4 * chunk_size
/ 4096) + 1);
1175 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1176 /* If any devices did not get added
1177 * because the kernel rejected them based
1178 * on event count, try adding them
1179 * again providing the action policy is
1180 * 're-add' or greater. The bitmap
1181 * might allow them to be included, or
1182 * they will become spares.
1184 for (i
= 0; i
< bestcnt
; i
++) {
1186 if (j
>= 0 && !devices
[j
].uptodate
) {
1187 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1189 rv
= add_disk(mdfd
, st
, content
,
1191 if (rv
== 0 && c
->verbose
>= 0)
1192 pr_err("%s has been re-added.\n",
1193 devices
[j
].devname
);
1197 if (content
->array
.level
== 6 &&
1198 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1202 sra
= sysfs_read(mdfd
, NULL
, 0);
1204 sysfs_set_str(sra
, NULL
,
1205 "sync_action", "repair");
1210 pr_err("failed to RUN_ARRAY %s: %s\n", mddev
, strerror(errno
));
1212 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1213 content
->array
.layout
, 1, avail
))
1214 pr_err("Not enough devices to start the array.\n");
1215 else if (!enough(content
->array
.level
,
1216 content
->array
.raid_disks
,
1217 content
->array
.layout
, clean
, avail
))
1218 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1222 if (c
->runstop
== -1) {
1223 pr_err("%s assembled from %d drive%s",
1224 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1225 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1226 fprintf(stderr
, " (out of %d)",
1227 content
->array
.raid_disks
);
1228 fprintf(stderr
, ", but not started.\n");
1231 if (c
->verbose
>= -1) {
1232 pr_err("%s assembled from %d drive%s",
1233 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1235 fprintf(stderr
, "%s %d rebuilding",
1236 sparecnt
? "," : " and", rebuilding_cnt
);
1238 fprintf(stderr
, " and %d spare%s", sparecnt
,
1239 sparecnt
== 1 ? "" : "s");
1240 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1241 content
->array
.layout
, 1, avail
))
1242 fprintf(stderr
, " - not enough to start the array.\n");
1243 else if (!enough(content
->array
.level
,
1244 content
->array
.raid_disks
,
1245 content
->array
.layout
, clean
, avail
))
1246 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1248 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1249 fprintf(stderr
, " - need all %d to start it",
1252 fprintf(stderr
, " - need %d to start", req_cnt
);
1253 fprintf(stderr
, " (use --run to insist).\n");
1259 int Assemble(struct supertype
*st
, char *mddev
,
1260 struct mddev_ident
*ident
,
1261 struct mddev_dev
*devlist
,
1265 * The task of Assemble is to find a collection of
1266 * devices that should (according to their superblocks)
1267 * form an array, and to give this collection to the MD driver.
1268 * In Linux-2.4 and later, this involves submitting a
1269 * SET_ARRAY_INFO ioctl with no arg - to prepare
1270 * the array - and then submit a number of
1271 * ADD_NEW_DISK ioctls to add disks into
1272 * the array. Finally RUN_ARRAY might
1273 * be submitted to start the array.
1275 * Much of the work of Assemble is in finding and/or
1276 * checking the disks to make sure they look right.
1278 * If mddev is not set, then scan must be set and we
1279 * read through the config file for dev+uuid mapping
1280 * We recurse, setting mddev, for each device that
1282 * - has a valid uuid (or any uuid if !uuidset)
1284 * If mddev is set, we try to determine state of md.
1285 * check version - must be at least 0.90.0
1286 * check kernel version. must be at least 2.4.
1287 * If not, we can possibly fall back on START_ARRAY
1288 * Try to GET_ARRAY_INFO.
1289 * If possible, give up
1290 * If not, try to STOP_ARRAY just to make sure
1292 * If !uuidset and scan, look in conf-file for uuid
1293 * If not found, give up
1294 * If !devlist and scan and uuidset, get list of devs from conf-file
1297 * Check superblock - discard if bad
1298 * Check uuid (set if we don't have one) - discard if no match
1299 * Check superblock similarity if we have a superblock - discard if different
1300 * Record events, devicenum
1301 * This should give us a list of devices for the array
1302 * We should collect the most recent event number
1304 * Count disks with recent enough event count
1305 * While force && !enough disks
1306 * Choose newest rejected disks, update event count
1307 * mark clean and rewrite superblock
1310 * foreach device with recent events : ADD_NEW_DISK
1311 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1313 * Check the device numbers in superblock are right
1314 * update superblock if any changes
1321 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1322 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0 &&
1323 (ident
->container
== NULL
|| ident
->member
== NULL
));
1324 struct devs
*devices
= NULL
;
1326 int *best
= NULL
; /* indexed by raid_disk */
1329 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1330 int journal_clean
= 0;
1333 int most_recent
= 0;
1337 int start_partial_ok
= (c
->runstop
>= 0) &&
1338 (c
->force
|| devlist
==NULL
|| auto_assem
);
1340 struct mddev_dev
*tmpdev
;
1342 struct mdinfo
*content
= NULL
;
1343 struct mdinfo
*pre_exist
= NULL
;
1346 char chosen_name
[1024];
1347 struct map_ent
*map
= NULL
;
1350 struct mdp_superblock_1
*sb
;
1351 bitmap_super_t
*bms
;
1354 * If any subdevs are listed, then any that don't
1355 * match ident are discarded. Remainder must all match and
1357 * If no subdevs, then we scan all devices in the config file, but
1358 * there must be something in the identity
1362 ident
->uuid_set
== 0 &&
1363 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1364 ident
->name
[0] == 0 &&
1365 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1366 ident
->devices
== NULL
) {
1367 pr_err("No identity information available for %s - cannot assemble.\n",
1368 mddev
? mddev
: "further assembly");
1372 if (devlist
== NULL
)
1373 devlist
= conf_get_devs();
1378 /* We come back here when doing auto-assembly and attempting some
1379 * set of devices failed. Those are now marked as ->used==2 and
1380 * we ignore them and try again
1384 * if come back try_again is called, then need to unlock first,
1385 * and lock again since the metadate is re-read.
1387 cluster_release_dlmlock();
1388 if (!st
&& ident
->st
)
1391 pr_err("looking for devices for %s\n",
1392 mddev
? mddev
: "further assembly");
1396 st
->ignore_hw_compat
= 1;
1397 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1398 inargv
, auto_assem
);
1402 if (!st
|| !st
->sb
|| !content
)
1406 bms
= (bitmap_super_t
*)(((char*)sb
) + 4096);
1407 if (sb
&& bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
1408 locked
= cluster_get_dlmlock();
1413 /* We have a full set of devices - we now need to find the
1415 * However there is a risk that we are racing with "mdadm -I"
1416 * and the array is already partially assembled - we will have
1417 * rejected any devices already in this address.
1418 * So we take a lock on the map file - to prevent further races -
1419 * and look for the uuid in there. If found and the array is
1420 * active, we abort. If found and the array is not active
1421 * we commit to that md device and add all the contained devices
1422 * to our list. We flag them so that we don't try to re-add,
1423 * but can remove if they turn out to not be wanted.
1426 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1427 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1430 mp
= map_by_uuid(&map
, content
->uuid
);
1433 /* array already exists. */
1434 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1435 if (pre_exist
->array
.level
!= UnSet
) {
1436 pr_err("Found some drive for an array that is already active: %s\n",
1438 pr_err("giving up.\n");
1441 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1442 /* We want to add this device to our list,
1443 * but it could already be there if "mdadm -I"
1444 * started *after* we checked for O_EXCL.
1445 * If we add it to the top of the list
1446 * it will be preferred over later copies.
1448 struct mddev_dev
*newdev
;
1449 char *devname
= map_dev(dv
->disk
.major
,
1454 newdev
= xmalloc(sizeof(*newdev
));
1455 newdev
->devname
= devname
;
1456 newdev
->disposition
= 'I';
1458 newdev
->next
= devlist
;
1462 strcpy(chosen_name
, mp
->path
);
1463 if (c
->verbose
> 0 || mddev
== NULL
||
1464 strcmp(mddev
, chosen_name
) != 0)
1465 pr_err("Merging with already-assembled %s\n",
1467 mdfd
= open_dev_excl(mp
->devnm
);
1469 int trustworthy
= FOREIGN
;
1470 name
= content
->name
;
1471 switch (st
->ss
->match_home(st
, c
->homehost
)
1472 ?: st
->ss
->match_home(st
, "any")) {
1474 trustworthy
= LOCAL
;
1475 name
= strchr(content
->name
, ':');
1479 name
= content
->name
;
1483 /* If the array is listed in mdadm.conf or on
1484 * command line, then we trust the name
1485 * even if the array doesn't look local
1487 trustworthy
= LOCAL
;
1490 content
->array
.level
== LEVEL_CONTAINER
) {
1491 name
= content
->text_version
;
1492 trustworthy
= METADATA
;
1495 if (name
[0] && trustworthy
!= LOCAL
&&
1496 ! c
->require_homehost
&&
1497 conf_name_is_free(name
))
1498 trustworthy
= LOCAL
;
1500 if (trustworthy
== LOCAL
&&
1502 /* Ignore 'host:' prefix of name */
1503 name
= strchr(name
, ':')+1;
1505 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1509 st
->ss
->free_super(st
);
1514 mddev
= chosen_name
;
1515 if (pre_exist
== NULL
) {
1516 if (mddev_busy(fd2devnm(mdfd
))) {
1517 pr_err("%s already active, cannot restart it!\n",
1519 for (tmpdev
= devlist
;
1520 tmpdev
&& tmpdev
->used
!= 1;
1521 tmpdev
= tmpdev
->next
)
1523 if (tmpdev
&& auto_assem
)
1524 pr_err("%s needed for %s...\n",
1525 mddev
, tmpdev
->devname
);
1528 st
->ss
->free_super(st
);
1533 /* just incase it was started but has no content */
1534 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1537 if (content
!= &info
) {
1538 /* This is a member of a container. Try starting the array. */
1540 err
= assemble_container_content(st
, mdfd
, content
, c
,
1544 cluster_release_dlmlock();
1548 /* Ok, no bad inconsistancy, we can try updating etc */
1549 devices
= xcalloc(num_devs
, sizeof(*devices
));
1550 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1551 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1552 c
, content
, mdfd
, mddev
,
1553 &most_recent
, &bestcnt
, &best
, inargv
);
1560 pr_err("no devices found for %s\n",
1563 st
->ss
->free_super(st
);
1568 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1569 st
->minor_version
= 90;
1571 st
->ss
->getinfo_super(st
, content
, NULL
);
1572 clean
= content
->array
.state
& 1;
1574 /* now we have some devices that might be suitable.
1577 avail
= xcalloc(content
->array
.raid_disks
, 1);
1583 for (i
=0; i
< bestcnt
; i
++) {
1585 int event_margin
= 1; /* always allow a difference of '1'
1586 * like the kernel does
1588 if (j
< 0) continue;
1589 /* note: we ignore error flags in multipath arrays
1590 * as they don't make sense
1592 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1593 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1594 if (content
->journal_device_required
)
1596 else /* unexpected journal, mark as faulty */
1597 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1598 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1599 if (!(devices
[j
].i
.disk
.state
1600 & (1<<MD_DISK_FAULTY
))) {
1601 devices
[j
].uptodate
= 1;
1607 /* If this device thinks that 'most_recent' has failed, then
1608 * we must reject this device.
1610 if (j
!= most_recent
&& !c
->force
&&
1611 content
->array
.raid_disks
> 0 &&
1612 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1613 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1614 if (c
->verbose
> -1)
1615 pr_err("ignoring %s as it reports %s as failed\n",
1616 devices
[j
].devname
, devices
[most_recent
].devname
);
1620 /* Require event counter to be same as, or just less than,
1621 * most recent. If it is bigger, it must be a stray spare and
1622 * should be ignored.
1624 if (devices
[j
].i
.events
+event_margin
>=
1625 devices
[most_recent
].i
.events
&&
1626 devices
[j
].i
.events
<=
1627 devices
[most_recent
].i
.events
1629 devices
[j
].uptodate
= 1;
1630 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1632 if (i
< content
->array
.raid_disks
* 2) {
1633 if (devices
[j
].i
.recovery_start
== MaxSector
||
1634 (content
->reshape_active
&&
1635 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1643 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1649 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1650 avail
, most_recent
, st
, c
);
1655 /* Now we want to look at the superblock which the kernel will base things on
1656 * and compare the devices that we think are working with the devices that the
1657 * superblock thinks are working.
1658 * If there are differences and --force is given, then update this chosen
1662 st
->ss
->free_super(st
);
1663 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1669 if (!devices
[j
].uptodate
)
1671 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1674 if ((fd
=dev_open(devices
[j
].devname
,
1675 devices
[j
].included
? O_RDONLY
1676 : (O_RDONLY
|O_EXCL
)))< 0) {
1677 pr_err("Cannot open %s: %s\n",
1678 devices
[j
].devname
, strerror(errno
));
1681 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1683 pr_err("RAID superblock has disappeared from %s\n",
1684 devices
[j
].devname
);
1689 if (st
->sb
== NULL
) {
1690 pr_err("No suitable drives found for %s\n", mddev
);
1693 st
->ss
->getinfo_super(st
, content
, NULL
);
1694 if (sysfs_init(content
, mdfd
, NULL
)) {
1695 pr_err("Unable to initialize sysfs\n");
1699 /* after reload context, store journal_clean in context */
1700 content
->journal_clean
= journal_clean
;
1701 for (i
=0; i
<bestcnt
; i
++) {
1703 unsigned int desired_state
;
1707 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1708 desired_state
= (1<<MD_DISK_JOURNAL
);
1709 else if (i
>= content
->array
.raid_disks
* 2)
1712 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1714 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1716 if (!devices
[j
].uptodate
)
1719 devices
[j
].i
.disk
.state
= desired_state
;
1720 if (!(devices
[j
].i
.array
.state
& 1))
1723 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1724 c
->verbose
, 0, NULL
)) {
1726 if (c
->verbose
>= 0)
1727 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1728 j
, mddev
, devices
[j
].devname
);
1731 if (c
->verbose
>= -1)
1732 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1733 i
, mddev
, devices
[j
].devname
);
1737 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1738 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1743 if (c
->force
&& !clean
&&
1744 !enough(content
->array
.level
, content
->array
.raid_disks
,
1745 content
->array
.layout
, clean
,
1747 change
+= st
->ss
->update_super(st
, content
, "force-array",
1748 devices
[chosen_drive
].devname
, c
->verbose
,
1756 fd
= dev_open(devices
[chosen_drive
].devname
,
1757 devices
[chosen_drive
].included
?
1758 O_RDWR
: (O_RDWR
|O_EXCL
));
1760 pr_err("Could not open %s for write - cannot Assemble array.\n",
1761 devices
[chosen_drive
].devname
);
1764 if (st
->ss
->store_super(st
, fd
)) {
1766 pr_err("Could not re-write superblock on %s\n",
1767 devices
[chosen_drive
].devname
);
1770 if (c
->verbose
>= 0)
1771 pr_err("Marking array %s as 'clean'\n",
1776 /* If we are in the middle of a reshape we may need to restore saved data
1777 * that was moved aside due to the reshape overwriting live data
1778 * The code of doing this lives in Grow.c
1780 if (content
->reshape_active
&&
1781 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1783 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1785 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1787 if (!c
->backup_file
)
1788 c
->backup_file
= locate_backup(content
->sys_name
);
1789 enable_fds(bestcnt
/2);
1790 for (i
= 0; i
< bestcnt
/2; i
++) {
1793 fdlist
[i
] = dev_open(devices
[j
].devname
,
1795 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1796 if (fdlist
[i
] < 0) {
1797 pr_err("Could not open %s for write - cannot Assemble array.\n",
1798 devices
[j
].devname
);
1806 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1807 err
= st
->ss
->recover_backup(st
, content
);
1809 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1810 c
->backup_file
, c
->verbose
> 0);
1811 if (err
&& c
->invalid_backup
) {
1813 pr_err("continuing without restoring backup\n");
1819 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1823 pr_err("Failed to restore critical section for reshape, sorry.\n");
1824 if (c
->backup_file
== NULL
)
1825 cont_err("Possibly you needed to specify the --backup-file\n");
1830 /* Almost ready to actually *do* something */
1831 /* First, fill in the map, so that udev can find our name
1832 * as soon as we become active.
1834 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1835 content
->array
.major_version
= 1;
1836 content
->array
.minor_version
= 0;
1837 strcpy(content
->text_version
, "1.0");
1840 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1841 content
->uuid
, chosen_name
);
1843 rv
= start_array(mdfd
, mddev
, content
,
1844 st
, ident
, best
, bestcnt
,
1845 chosen_drive
, devices
, okcnt
, sparecnt
,
1846 rebuilding_cnt
, journalcnt
,
1848 clean
, avail
, start_partial_ok
,
1851 if (rv
== 1 && !pre_exist
)
1852 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1857 wait_for(chosen_name
, mdfd
);
1861 /* There is a nasty race with 'mdadm --monitor'.
1862 * If it opens this device before we close it,
1863 * it gets an incomplete open on which IO
1864 * doesn't work and the capacity is
1866 * If we reopen (to check for layered devices)
1867 * before --monitor closes, we loose.
1869 * So: wait upto 1 second for there to be
1870 * a non-zero capacity.
1872 while (usecs
< 1000) {
1873 mdfd
= open(mddev
, O_RDONLY
);
1875 unsigned long long size
;
1876 if (get_dev_size(mdfd
, NULL
, &size
) &&
1885 } else if (mdfd
>= 0)
1888 /* '2' means 'OK, but not started yet' */
1890 cluster_release_dlmlock();
1895 return rv
== 2 ? 0 : rv
;
1898 int assemble_container_content(struct supertype
*st
, int mdfd
,
1899 struct mdinfo
*content
, struct context
*c
,
1900 char *chosen_name
, int *result
)
1902 struct mdinfo
*dev
, *sra
, *dev2
;
1903 int working
= 0, preexist
= 0;
1910 if (sysfs_init(content
, mdfd
, NULL
)) {
1911 pr_err("Unable to initialize sysfs\n");
1915 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1916 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1917 if (content
->array
.major_version
== -1 &&
1918 content
->array
.minor_version
== -2 &&
1920 content
->text_version
[0] == '/')
1921 content
->text_version
[0] = '-';
1922 if (sysfs_set_array(content
, 9003) != 0) {
1928 /* There are two types of reshape: container wide or sub-array specific
1929 * Check if metadata requests blocking container wide reshapes
1931 start_reshape
= (content
->reshape_active
&&
1932 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1933 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1935 /* Block subarray here if it is under reshape now
1936 * Do not allow for any changes in this array
1938 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1939 block_subarray(content
);
1941 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1942 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1943 if (dev2
->disk
.major
== dev
->disk
.major
&&
1944 dev2
->disk
.minor
== dev
->disk
.minor
)
1948 /* Don't want this one any more */
1949 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1951 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1955 sysfs_set_str(sra
, dev2
, "state", "remove");
1957 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1958 avail
= xcalloc(content
->array
.raid_disks
, 1);
1959 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1960 if (dev
->disk
.raid_disk
>= 0)
1961 avail
[dev
->disk
.raid_disk
] = 1;
1962 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1963 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1964 content
->reshape_active
)
1968 } else if (errno
== EEXIST
)
1972 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1974 return 1;/* Nothing new, don't try to start */
1976 map_update(NULL
, fd2devnm(mdfd
), content
->text_version
,
1977 content
->uuid
, chosen_name
);
1979 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
1980 st
->ss
->validate_ppl
) {
1981 content
->array
.state
|= 1;
1984 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1989 ret
= st
->ss
->validate_ppl(st
, content
, dev
);
1999 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
2001 content
->array
.state
&= ~1;
2002 avail
[dev
->disk
.raid_disk
] = 0;
2006 /* have --force - overwrite the invalid ppl */
2007 devpath
= map_dev(dev
->disk
.major
, dev
->disk
.minor
, 0);
2008 dfd
= dev_open(devpath
, O_RDWR
);
2010 pr_err("Failed to open %s\n", devpath
);
2015 err
= st
->ss
->write_init_ppl(st
, content
, dfd
);
2028 if (enough(content
->array
.level
, content
->array
.raid_disks
,
2029 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
2030 if (c
->export
&& result
)
2032 else if (c
->verbose
>= 0) {
2033 pr_err("%s assembled with %d device%s",
2034 chosen_name
, preexist
+ working
,
2035 preexist
+ working
== 1 ? "":"s");
2037 fprintf(stderr
, " (%d new)", working
);
2038 fprintf(stderr
, " but not started\n");
2045 if (c
->runstop
<= 0 &&
2046 (working
+ preexist
+ expansion
) <
2047 content
->array
.working_disks
) {
2048 if (c
->export
&& result
)
2049 *result
|= INCR_UNSAFE
;
2050 else if (c
->verbose
>= 0) {
2051 pr_err("%s assembled with %d device%s",
2052 chosen_name
, preexist
+ working
,
2053 preexist
+ working
== 1 ? "":"s");
2055 fprintf(stderr
, " (%d new)", working
);
2056 fprintf(stderr
, " but not safe to start\n");
2062 if (start_reshape
) {
2063 int spare
= content
->array
.raid_disks
+ expansion
;
2064 if (restore_backup(st
, content
,
2066 spare
, &c
->backup_file
, c
->verbose
) == 1)
2069 err
= sysfs_set_str(content
, NULL
,
2070 "array_state", "readonly");
2074 if (st
->ss
->external
) {
2075 if (!mdmon_running(st
->container_devnm
))
2076 start_mdmon(st
->container_devnm
);
2077 ping_monitor(st
->container_devnm
);
2078 if (mdmon_running(st
->container_devnm
) &&
2079 st
->update_tail
== NULL
)
2080 st
->update_tail
= &st
->updates
;
2083 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2084 0, c
->freeze_reshape
);
2085 } else switch(content
->array
.level
) {
2087 case LEVEL_MULTIPATH
:
2089 err
= sysfs_set_str(content
, NULL
, "array_state",
2090 c
->readonly
? "readonly" : "active");
2093 err
= sysfs_set_str(content
, NULL
, "array_state",
2095 /* start mdmon if needed. */
2097 if (!mdmon_running(st
->container_devnm
))
2098 start_mdmon(st
->container_devnm
);
2099 ping_monitor(st
->container_devnm
);
2104 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2106 /* Block subarray here if it is not reshaped now
2107 * It has be blocked a little later to allow mdmon to switch in
2110 if (st
->ss
->external
&& content
->recovery_blocked
&&
2112 block_subarray(content
);
2114 if (c
->export
&& result
) {
2118 *result
|= INCR_YES
;
2119 } else if (c
->verbose
>= 0) {
2121 pr_err("array %s now has %d device%s",
2122 chosen_name
, working
+ preexist
,
2123 working
+ preexist
== 1 ? "":"s");
2125 pr_err("Started %s with %d device%s",
2126 chosen_name
, working
+ preexist
,
2127 working
+ preexist
== 1 ? "":"s");
2129 fprintf(stderr
, " (%d new)", working
);
2131 fprintf(stderr
, " ( + %d for expansion)",
2133 fprintf(stderr
, "\n");
2136 wait_for(chosen_name
, mdfd
);
2138 /* FIXME should have an O_EXCL and wait for read-auto */