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
);
285 tst
->ss
->free_super(tst
);
289 if (found_container
) {
290 /* tmpdev is a container. We need to be either
291 * looking for a member, or auto-assembling
293 /* should be safe to try an exclusive open now, we
294 * have rejected anything that some other mdadm might
297 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
300 pr_err("%s is busy - skipping\n", devname
);
305 if (ident
->container
&& ident
->container
[0] != '/') {
310 tst
->ss
->getinfo_super(tst
, content
, NULL
);
312 if (!parse_uuid(ident
->container
, uuid
) ||
313 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
315 pr_err("%s has wrong UUID to be required container\n",
320 /* It is worth looking inside this container.
323 pr_err("looking in container %s\n",
326 for (content
= tst
->ss
->container_content(tst
, NULL
);
328 content
= content
->next
) {
330 if (!ident_matches(ident
, content
, tst
,
331 c
->homehost
, c
->require_homehost
,
333 report_mismatch
? devname
: NULL
))
334 /* message already printed */;
335 else if (is_member_busy(content
->text_version
)) {
337 pr_err("member %s in %s is already assembled\n",
338 content
->text_version
,
340 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
341 /* do not assemble arrays with unsupported configurations */
342 pr_err("Cannot activate member %s in %s.\n",
343 content
->text_version
,
350 goto loop
; /* empty container */
353 st
= tst
; tst
= NULL
;
354 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
355 pr_err("%s is a container, but is not only device given: confused and aborting\n",
357 st
->ss
->free_super(st
);
358 dev_policy_free(pol
);
359 domain_free(domains
);
363 pr_err("found match on member %s in %s\n",
364 content
->text_version
, devname
);
366 /* make sure we finished the loop */
371 tst
->ss
->getinfo_super(tst
, content
, NULL
);
373 if (!ident_matches(ident
, content
, tst
,
374 c
->homehost
, c
->require_homehost
,
376 report_mismatch
? devname
: NULL
))
380 /* Never auto-assemble things that conflict
381 * with mdadm.conf in some way
383 struct mddev_ident
*match
;
386 match
= conf_match(tst
, content
, devname
,
387 report_mismatch
? c
->verbose
: -1,
389 if (!match
&& rv
== 2)
391 if (match
&& match
->devname
&&
392 strcasecmp(match
->devname
, "<ignore>") == 0) {
394 pr_err("%s is a member of an explicitly ignored array\n",
398 if (match
&& !ident_matches(match
, content
, tst
,
399 c
->homehost
, c
->require_homehost
,
401 report_mismatch
? devname
: NULL
))
402 /* Array exists in mdadm.conf but some
403 * details don't match, so reject it
408 /* should be safe to try an exclusive open now, we
409 * have rejected anything that some other mdadm might
412 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
415 pr_err("%s is busy - skipping\n", devname
);
422 if (st
->minor_version
== -1)
423 st
->minor_version
= tst
->minor_version
;
425 if (memcmp(content
->uuid
, uuid_zero
,
426 sizeof(int[4])) == 0) {
427 /* this is a floating spare. It cannot define
428 * an array unless there are no more arrays of
429 * this type to be found. It can be included
430 * in an array of this type though.
436 if (st
->ss
!= tst
->ss
||
437 st
->minor_version
!= tst
->minor_version
||
438 st
->ss
->compare_super(st
, tst
) != 0) {
439 /* Some mismatch. If exactly one array matches this host,
440 * we can resolve on that one.
441 * Or, if we are auto assembling, we just ignore the second
447 int first
= st
->ss
->match_home(st
, c
->homehost
);
448 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
450 (first
== 1 || last
== 1)) {
451 /* We can do something */
452 if (first
) {/* just ignore this one */
454 pr_err("%s misses out due to wrong homehost\n",
457 } else { /* reject all those sofar */
458 struct mddev_dev
*td
;
460 pr_err("%s overrides previous devices due to good homehost\n",
462 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
470 pr_err("superblock on %s doesn't match others - assembly aborted\n",
472 tst
->ss
->free_super(tst
);
473 st
->ss
->free_super(st
);
474 dev_policy_free(pol
);
475 domain_free(domains
);
481 /* Collect domain information from members only */
482 if (tmpdev
&& tmpdev
->used
== 1) {
484 pol
= devid_policy(rdev
);
485 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
487 dev_policy_free(pol
);
490 tst
->ss
->free_super(tst
);
493 /* Check if we found some imsm spares but no members */
496 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
498 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
499 if (tmpdev
->used
!= 3)
505 /* we need sb from one of the spares */
506 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
508 st
->ss
->load_super(st
, dfd
, NULL
))
515 /* Now reject spares that don't match domains of identified members */
516 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
517 if (tmpdev
->used
!= 3)
519 if (!stat_is_blkdev(tmpdev
->devname
, &rdev
)) {
522 struct dev_policy
*pol
= devid_policy(rdev
);
523 int dt
= domain_test(domains
, pol
, NULL
);
524 if (inargv
&& dt
!= 0)
525 /* take this spare as domains match
526 * if there are any */
528 else if (!inargv
&& dt
== 1)
529 /* device wasn't explicitly listed, so need
530 * explicit domain match - which we have */
533 /* if domains don't match mark as unused */
535 dev_policy_free(pol
);
538 domain_free(domains
);
540 if (st
&& st
->sb
&& content
== *contentp
)
541 st
->ss
->getinfo_super(st
, content
, NULL
);
549 int uptodate
; /* set once we decide that this device is as
550 * recent as everything else in the array.
552 int included
; /* set if the device is already in the array
553 * due to a previous '-I'
558 static int load_devices(struct devs
*devices
, char *devmap
,
559 struct mddev_ident
*ident
, struct supertype
**stp
,
560 struct mddev_dev
*devlist
, struct context
*c
,
561 struct mdinfo
*content
,
562 int mdfd
, char *mddev
,
563 int *most_recentp
, int *bestcntp
, int **bestp
,
566 struct mddev_dev
*tmpdev
;
570 int most_recent
= -1;
573 struct supertype
*st
= *stp
;
575 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
576 char *devname
= tmpdev
->devname
;
578 struct supertype
*tst
;
582 if (tmpdev
->used
!= 1)
584 /* looks like a good enough match to update the super block if needed */
586 /* prepare useful information in info structures */
591 if (strcmp(c
->update
, "uuid") == 0 && !ident
->uuid_set
)
592 random_uuid((__u8
*)ident
->uuid
);
594 if (strcmp(c
->update
, "ppl") == 0 &&
595 ident
->bitmap_fd
>= 0) {
596 pr_err("PPL is not compatible with bitmap\n");
603 dfd
= dev_open(devname
,
604 tmpdev
->disposition
== 'I'
605 ? O_RDWR
: (O_RDWR
|O_EXCL
));
608 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
609 pr_err("cannot re-read metadata from %s - aborting\n",
616 tst
->ss
->free_super(tst
);
621 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
623 memcpy(content
->uuid
, ident
->uuid
, 16);
624 strcpy(content
->name
, ident
->name
);
625 content
->array
.md_minor
= minor(stb2
.st_rdev
);
627 if (strcmp(c
->update
, "byteorder") == 0)
629 else if (strcmp(c
->update
, "home-cluster") == 0) {
630 tst
->cluster_name
= c
->homecluster
;
631 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
632 } else if (strcmp(c
->update
, "nodes") == 0) {
633 tst
->nodes
= c
->nodes
;
634 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
635 } else if (strcmp(c
->update
, "revert-reshape") == 0 &&
637 err
= tst
->ss
->update_super(tst
, content
,
638 "revert-reshape-nobackup",
643 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
649 pr_err("--update=%s not understood for %s metadata\n",
650 c
->update
, tst
->ss
->name
);
651 tst
->ss
->free_super(tst
);
660 if (strcmp(c
->update
, "uuid")==0 &&
663 memcpy(ident
->uuid
, content
->uuid
, 16);
665 if (tst
->ss
->store_super(tst
, dfd
))
666 pr_err("Could not re-write superblock on %s.\n",
669 if (strcmp(c
->update
, "uuid")==0 &&
670 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
671 if (bitmap_update_uuid(ident
->bitmap_fd
,
673 tst
->ss
->swapuuid
) != 0)
674 pr_err("Could not update uuid on external bitmap.\n");
679 dfd
= dev_open(devname
,
680 tmpdev
->disposition
== 'I'
681 ? O_RDWR
: (O_RDWR
|O_EXCL
));
684 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
685 pr_err("cannot re-read metadata from %s - aborting\n",
692 tst
->ss
->free_super(tst
);
697 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
704 pr_err("%s is identified as a member of %s, slot %d%s.\n",
705 devname
, mddev
, content
->disk
.raid_disk
,
706 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
707 devices
[devcnt
].devname
= devname
;
708 devices
[devcnt
].uptodate
= 0;
709 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
710 devices
[devcnt
].i
= *content
;
711 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
712 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
714 if (devices
[devcnt
].i
.disk
.state
== 6) {
715 if (most_recent
< 0 ||
716 devices
[devcnt
].i
.events
717 > devices
[most_recent
].i
.events
) {
718 struct supertype
*tmp
= tst
;
721 most_recent
= devcnt
;
724 tst
->ss
->free_super(tst
);
727 if (content
->array
.level
== LEVEL_MULTIPATH
)
728 /* with multipath, the raid_disk from the superblock is meaningless */
731 i
= devices
[devcnt
].i
.disk
.raid_disk
;
732 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
733 if (nextspare
< content
->array
.raid_disks
*2)
734 nextspare
= content
->array
.raid_disks
*2;
737 /* i is raid_disk - double it so there is room for
740 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
742 if (i
>= content
->array
.raid_disks
*2 &&
748 int newbestcnt
= i
+10;
749 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
751 for (c
=0; c
< newbestcnt
; c
++)
753 newbest
[c
] = best
[c
];
758 bestcnt
= newbestcnt
;
761 devices
[best
[i
]].i
.events
==
762 devices
[devcnt
].i
.events
&&
763 (devices
[best
[i
]].i
.disk
.minor
!=
764 devices
[devcnt
].i
.disk
.minor
) &&
766 content
->array
.level
!= LEVEL_MULTIPATH
) {
767 /* two different devices with identical superblock.
768 * Could be a mis-detection caused by overlapping
769 * partitions. fail-safe.
771 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
772 " If they are really different, please --zero the superblock on one\n"
773 " If they are the same or overlap, please remove one from %s.\n",
774 devices
[best
[i
]].devname
, devname
,
775 inargv
? "the list" :
776 "the\n DEVICE list in mdadm.conf"
784 if (best
[i
] == -1 || (devices
[best
[i
]].i
.events
785 < devices
[devcnt
].i
.events
))
787 else if (st
->ss
== &super_imsm
)
792 if (most_recent
>= 0)
793 *most_recentp
= most_recent
;
800 static int force_array(struct mdinfo
*content
,
801 struct devs
*devices
,
802 int *best
, int bestcnt
, char *avail
,
804 struct supertype
*st
,
808 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
809 content
->array
.layout
, 1,
811 (content
->reshape_active
&& content
->delta_disks
> 0 &&
812 !enough(content
->array
.level
, (content
->array
.raid_disks
813 - content
->delta_disks
),
814 content
->new_layout
, 1, avail
))) {
815 /* Choose the newest best drive which is
816 * not up-to-date, update the superblock
820 struct supertype
*tst
;
821 unsigned long long current_events
;
822 int chosen_drive
= -1;
826 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
831 if (devices
[j
].uptodate
)
833 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
835 if (!devices
[j
].i
.reshape_active
||
836 devices
[j
].i
.delta_disks
<= 0)
838 /* When increasing number of devices, an
839 * added device also appears to be
840 * recovering. It is safe to include it
841 * as long as it won't be a source of
843 * For now, just allow for last data
844 * devices in RAID4 or last devices in RAID4/5/6.
846 delta
= devices
[j
].i
.delta_disks
;
847 if (devices
[j
].i
.array
.level
>= 4 &&
848 devices
[j
].i
.array
.level
<= 6 &&
849 i
/2 >= content
->array
.raid_disks
- delta
)
851 else if (devices
[j
].i
.array
.level
== 4 &&
852 i
/2 >= content
->array
.raid_disks
- delta
- 1)
856 } else if (devices
[j
].i
.reshape_active
!=
857 content
->reshape_active
||
858 (devices
[j
].i
.reshape_active
&&
859 devices
[j
].i
.reshape_progress
!=
860 content
->reshape_progress
))
861 /* Here, it may be a source of data. If two
862 * devices claim different progresses, it
863 * means that reshape boundaries differ for
864 * their own devices. Kernel will only treat
865 * the first one as reshape progress and
866 * go on. It may cause disaster, so avoid it.
869 if (chosen_drive
< 0 ||
871 > devices
[chosen_drive
].i
.events
)
874 if (chosen_drive
< 0)
876 current_events
= devices
[chosen_drive
].i
.events
;
879 pr_err("forcing event count in %s(%d) from %d upto %d\n",
880 devices
[chosen_drive
].devname
,
881 devices
[chosen_drive
].i
.disk
.raid_disk
,
882 (int)(devices
[chosen_drive
].i
.events
),
883 (int)(devices
[most_recent
].i
.events
));
884 fd
= dev_open(devices
[chosen_drive
].devname
,
885 devices
[chosen_drive
].included
? O_RDWR
888 pr_err("Couldn't open %s for write - not updating\n",
889 devices
[chosen_drive
].devname
);
890 devices
[chosen_drive
].i
.events
= 0;
894 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
896 pr_err("RAID superblock disappeared from %s - not updating.\n",
897 devices
[chosen_drive
].devname
);
898 devices
[chosen_drive
].i
.events
= 0;
901 content
->events
= devices
[most_recent
].i
.events
;
902 tst
->ss
->update_super(tst
, content
, "force-one",
903 devices
[chosen_drive
].devname
, c
->verbose
,
906 if (tst
->ss
->store_super(tst
, fd
)) {
908 pr_err("Could not re-write superblock on %s\n",
909 devices
[chosen_drive
].devname
);
910 devices
[chosen_drive
].i
.events
= 0;
911 tst
->ss
->free_super(tst
);
915 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
916 devices
[chosen_drive
].uptodate
= 1;
917 avail
[chosen_drive
] = 1;
919 tst
->ss
->free_super(tst
);
920 /* If there are any other drives of the same vintage,
921 * add them in as well. We can't lose and we might gain
924 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
928 !devices
[j
].uptodate
&&
929 devices
[j
].i
.recovery_start
== MaxSector
&&
930 devices
[j
].i
.events
== current_events
&&
931 ((!devices
[j
].i
.reshape_active
&&
932 !content
->reshape_active
) ||
933 (devices
[j
].i
.reshape_active
==
934 content
->reshape_active
&&
935 devices
[j
].i
.reshape_progress
==
936 content
->reshape_progress
))) {
945 static int start_array(int mdfd
,
947 struct mdinfo
*content
,
948 struct supertype
*st
,
949 struct mddev_ident
*ident
,
950 int *best
, int bestcnt
,
952 struct devs
*devices
,
954 unsigned int sparecnt
,
955 unsigned int rebuilding_cnt
,
956 unsigned int journalcnt
,
958 int clean
, char *avail
,
959 int start_partial_ok
,
966 unsigned int req_cnt
;
968 if (content
->journal_device_required
&& (content
->journal_clean
== 0)) {
970 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
973 pr_err("Journal is missing or stale, starting array read only.\n");
977 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
980 rv
= set_array_info(mdfd
, st
, content
);
982 pr_err("failed to set array info for %s: %s\n",
983 mddev
, strerror(errno
));
986 if (ident
->bitmap_fd
>= 0) {
987 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
988 pr_err("SET_BITMAP_FILE failed.\n");
991 } else if (ident
->bitmap_file
) {
992 /* From config file */
993 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
995 pr_err("Could not open bitmap file %s\n",
999 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1000 pr_err("Failed to set bitmapfile for %s\n", mddev
);
1007 /* First, add the raid disks, but add the chosen one last */
1008 for (i
= 0; i
<= bestcnt
; i
++) {
1012 if (j
== chosen_drive
)
1017 if (j
>= 0 && !devices
[j
].included
) {
1020 dfd
= dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1022 remove_partitions(dfd
);
1025 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1028 pr_err("failed to add %s to %s: %s\n",
1029 devices
[j
].devname
, mddev
,
1031 if (i
< content
->array
.raid_disks
* 2 ||
1036 } else if (c
->verbose
> 0) {
1037 pr_err("added %s to %s as %d%s%s\n",
1038 devices
[j
].devname
, mddev
,
1039 devices
[j
].i
.disk
.raid_disk
,
1040 devices
[j
].uptodate
?"":
1041 " (possibly out of date)",
1042 (devices
[j
].i
.disk
.state
&
1043 (1<<MD_DISK_REPLACEMENT
)) ?
1046 } else if (j
>= 0) {
1048 pr_err("%s is already in %s as %d\n",
1049 devices
[j
].devname
, mddev
,
1050 devices
[j
].i
.disk
.raid_disk
);
1051 } else if (c
->verbose
> 0 &&
1052 i
< content
->array
.raid_disks
* 2 && (i
& 1) == 0)
1053 pr_err("no uptodate device for slot %d of %s\n",
1057 if (content
->array
.level
== LEVEL_CONTAINER
) {
1058 if (c
->verbose
>= 0) {
1059 pr_err("Container %s has been assembled with %d drive%s",
1060 mddev
, okcnt
+ sparecnt
+ journalcnt
,
1061 okcnt
+ sparecnt
+ journalcnt
== 1 ? "" : "s");
1062 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1063 fprintf(stderr
, " (out of %d)",
1064 content
->array
.raid_disks
);
1065 fprintf(stderr
, "\n");
1068 if (st
->ss
->validate_container
) {
1069 struct mdinfo
*devices_list
;
1070 struct mdinfo
*info_devices
;
1073 devices_list
= NULL
;
1074 info_devices
= xmalloc(sizeof(struct mdinfo
) *
1075 (okcnt
+ sparecnt
));
1076 for (count
= 0; count
< okcnt
+ sparecnt
; count
++) {
1077 info_devices
[count
] = devices
[count
].i
;
1078 info_devices
[count
].next
= devices_list
;
1079 devices_list
= &info_devices
[count
];
1081 if (st
->ss
->validate_container(devices_list
))
1082 pr_err("Mismatch detected!\n");
1086 st
->ss
->free_super(st
);
1087 sysfs_uevent(content
, "change");
1088 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1089 /* Was partial, is still partial, so signal an error
1090 * to ensure we don't retry */
1095 /* Get number of in-sync devices according to the superblock.
1096 * We must have this number to start the array without -s or -R
1098 req_cnt
= content
->array
.working_disks
;
1100 if (c
->runstop
== 1 ||
1102 (enough(content
->array
.level
, content
->array
.raid_disks
,
1103 content
->array
.layout
, clean
, avail
) &&
1104 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)))) {
1105 /* This array is good-to-go.
1106 * If a reshape is in progress then we might need to
1107 * continue monitoring it. In that case we start
1108 * it read-only and let the grow code make it writable.
1112 if (content
->reshape_active
&&
1113 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1114 content
->delta_disks
<= 0) {
1115 if (!c
->backup_file
) {
1116 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1118 pr_err("Please provided one with \"--backup-file=...\"\n");
1120 strcmp(c
->update
, "revert-reshape") == 0)
1121 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1124 rv
= sysfs_set_str(content
, NULL
,
1125 "array_state", "readonly");
1127 rv
= Grow_continue(mdfd
, st
, content
,
1130 } else if (c
->readonly
&&
1131 sysfs_attribute_available(content
, NULL
,
1133 rv
= sysfs_set_str(content
, NULL
,
1134 "array_state", "readonly");
1136 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1137 reopen_mddev(mdfd
); /* drop O_EXCL */
1139 if (c
->verbose
>= 0) {
1140 pr_err("%s has been started with %d drive%s",
1141 mddev
, okcnt
, okcnt
==1?"":"s");
1142 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1143 fprintf(stderr
, " (out of %d)",
1144 content
->array
.raid_disks
);
1146 fprintf(stderr
, "%s %d rebuilding",
1147 sparecnt
?",":" and",
1150 fprintf(stderr
, " and %d spare%s",
1152 sparecnt
== 1 ? "" : "s");
1153 if (content
->journal_clean
)
1154 fprintf(stderr
, " and %d journal",
1156 fprintf(stderr
, ".\n");
1158 if (content
->reshape_active
&&
1159 content
->array
.level
>= 4 &&
1160 content
->array
.level
<= 6) {
1161 /* might need to increase the size
1162 * of the stripe cache - default is 256
1164 int chunk_size
= content
->array
.chunk_size
;
1166 if (content
->reshape_active
&&
1167 content
->new_chunk
> chunk_size
)
1168 chunk_size
= content
->new_chunk
;
1169 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1172 sra
= sysfs_read(mdfd
, NULL
, 0);
1174 sysfs_set_num(sra
, NULL
,
1175 "stripe_cache_size",
1176 (4 * chunk_size
/ 4096) + 1);
1180 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1181 /* If any devices did not get added
1182 * because the kernel rejected them based
1183 * on event count, try adding them
1184 * again providing the action policy is
1185 * 're-add' or greater. The bitmap
1186 * might allow them to be included, or
1187 * they will become spares.
1189 for (i
= 0; i
< bestcnt
; i
++) {
1191 if (j
>= 0 && !devices
[j
].uptodate
) {
1192 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1194 rv
= add_disk(mdfd
, st
, content
,
1196 if (rv
== 0 && c
->verbose
>= 0)
1197 pr_err("%s has been re-added.\n",
1198 devices
[j
].devname
);
1202 if (content
->array
.level
== 6 &&
1203 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1207 sra
= sysfs_read(mdfd
, NULL
, 0);
1209 sysfs_set_str(sra
, NULL
,
1210 "sync_action", "repair");
1215 pr_err("failed to RUN_ARRAY %s: %s\n", mddev
, strerror(errno
));
1217 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1218 content
->array
.layout
, 1, avail
))
1219 pr_err("Not enough devices to start the array.\n");
1220 else if (!enough(content
->array
.level
,
1221 content
->array
.raid_disks
,
1222 content
->array
.layout
, clean
, avail
))
1223 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1227 if (c
->runstop
== -1) {
1228 pr_err("%s assembled from %d drive%s",
1229 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1230 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1231 fprintf(stderr
, " (out of %d)",
1232 content
->array
.raid_disks
);
1233 fprintf(stderr
, ", but not started.\n");
1236 if (c
->verbose
>= -1) {
1237 pr_err("%s assembled from %d drive%s",
1238 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1240 fprintf(stderr
, "%s %d rebuilding",
1241 sparecnt
? "," : " and", rebuilding_cnt
);
1243 fprintf(stderr
, " and %d spare%s", sparecnt
,
1244 sparecnt
== 1 ? "" : "s");
1245 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1246 content
->array
.layout
, 1, avail
))
1247 fprintf(stderr
, " - not enough to start the array.\n");
1248 else if (!enough(content
->array
.level
,
1249 content
->array
.raid_disks
,
1250 content
->array
.layout
, clean
, avail
))
1251 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1253 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1254 fprintf(stderr
, " - need all %d to start it",
1257 fprintf(stderr
, " - need %d to start", req_cnt
);
1258 fprintf(stderr
, " (use --run to insist).\n");
1264 int Assemble(struct supertype
*st
, char *mddev
,
1265 struct mddev_ident
*ident
,
1266 struct mddev_dev
*devlist
,
1270 * The task of Assemble is to find a collection of
1271 * devices that should (according to their superblocks)
1272 * form an array, and to give this collection to the MD driver.
1273 * In Linux-2.4 and later, this involves submitting a
1274 * SET_ARRAY_INFO ioctl with no arg - to prepare
1275 * the array - and then submit a number of
1276 * ADD_NEW_DISK ioctls to add disks into
1277 * the array. Finally RUN_ARRAY might
1278 * be submitted to start the array.
1280 * Much of the work of Assemble is in finding and/or
1281 * checking the disks to make sure they look right.
1283 * If mddev is not set, then scan must be set and we
1284 * read through the config file for dev+uuid mapping
1285 * We recurse, setting mddev, for each device that
1287 * - has a valid uuid (or any uuid if !uuidset)
1289 * If mddev is set, we try to determine state of md.
1290 * check version - must be at least 0.90.0
1291 * check kernel version. must be at least 2.4.
1292 * If not, we can possibly fall back on START_ARRAY
1293 * Try to GET_ARRAY_INFO.
1294 * If possible, give up
1295 * If not, try to STOP_ARRAY just to make sure
1297 * If !uuidset and scan, look in conf-file for uuid
1298 * If not found, give up
1299 * If !devlist and scan and uuidset, get list of devs from conf-file
1302 * Check superblock - discard if bad
1303 * Check uuid (set if we don't have one) - discard if no match
1304 * Check superblock similarity if we have a superblock - discard if different
1305 * Record events, devicenum
1306 * This should give us a list of devices for the array
1307 * We should collect the most recent event number
1309 * Count disks with recent enough event count
1310 * While force && !enough disks
1311 * Choose newest rejected disks, update event count
1312 * mark clean and rewrite superblock
1315 * foreach device with recent events : ADD_NEW_DISK
1316 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1318 * Check the device numbers in superblock are right
1319 * update superblock if any changes
1326 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1327 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0 &&
1328 (ident
->container
== NULL
|| ident
->member
== NULL
));
1329 struct devs
*devices
= NULL
;
1331 int *best
= NULL
; /* indexed by raid_disk */
1334 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1335 int journal_clean
= 0;
1338 int most_recent
= 0;
1342 int start_partial_ok
= (c
->runstop
>= 0) &&
1343 (c
->force
|| devlist
==NULL
|| auto_assem
);
1345 struct mddev_dev
*tmpdev
;
1347 struct mdinfo
*content
= NULL
;
1348 struct mdinfo
*pre_exist
= NULL
;
1351 char chosen_name
[1024];
1352 struct map_ent
*map
= NULL
;
1356 * If any subdevs are listed, then any that don't
1357 * match ident are discarded. Remainder must all match and
1359 * If no subdevs, then we scan all devices in the config file, but
1360 * there must be something in the identity
1364 ident
->uuid_set
== 0 &&
1365 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1366 ident
->name
[0] == 0 &&
1367 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1368 ident
->devices
== NULL
) {
1369 pr_err("No identity information available for %s - cannot assemble.\n",
1370 mddev
? mddev
: "further assembly");
1374 if (devlist
== NULL
)
1375 devlist
= conf_get_devs();
1380 /* We come back here when doing auto-assembly and attempting some
1381 * set of devices failed. Those are now marked as ->used==2 and
1382 * we ignore them and try again
1384 if (!st
&& ident
->st
)
1387 pr_err("looking for devices for %s\n",
1388 mddev
? mddev
: "further assembly");
1392 st
->ignore_hw_compat
= 1;
1393 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1394 inargv
, auto_assem
);
1398 if (!st
|| !st
->sb
|| !content
)
1401 /* We have a full set of devices - we now need to find the
1403 * However there is a risk that we are racing with "mdadm -I"
1404 * and the array is already partially assembled - we will have
1405 * rejected any devices already in this address.
1406 * So we take a lock on the map file - to prevent further races -
1407 * and look for the uuid in there. If found and the array is
1408 * active, we abort. If found and the array is not active
1409 * we commit to that md device and add all the contained devices
1410 * to our list. We flag them so that we don't try to re-add,
1411 * but can remove if they turn out to not be wanted.
1414 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1415 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1418 mp
= map_by_uuid(&map
, content
->uuid
);
1421 /* array already exists. */
1422 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1423 if (pre_exist
->array
.level
!= UnSet
) {
1424 pr_err("Found some drive for an array that is already active: %s\n",
1426 pr_err("giving up.\n");
1429 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1430 /* We want to add this device to our list,
1431 * but it could already be there if "mdadm -I"
1432 * started *after* we checked for O_EXCL.
1433 * If we add it to the top of the list
1434 * it will be preferred over later copies.
1436 struct mddev_dev
*newdev
;
1437 char *devname
= map_dev(dv
->disk
.major
,
1442 newdev
= xmalloc(sizeof(*newdev
));
1443 newdev
->devname
= devname
;
1444 newdev
->disposition
= 'I';
1446 newdev
->next
= devlist
;
1450 strcpy(chosen_name
, mp
->path
);
1451 if (c
->verbose
> 0 || mddev
== NULL
||
1452 strcmp(mddev
, chosen_name
) != 0)
1453 pr_err("Merging with already-assembled %s\n",
1455 mdfd
= open_dev_excl(mp
->devnm
);
1457 int trustworthy
= FOREIGN
;
1458 name
= content
->name
;
1459 switch (st
->ss
->match_home(st
, c
->homehost
)
1460 ?: st
->ss
->match_home(st
, "any")) {
1462 trustworthy
= LOCAL
;
1463 name
= strchr(content
->name
, ':');
1467 name
= content
->name
;
1471 /* If the array is listed in mdadm.conf or on
1472 * command line, then we trust the name
1473 * even if the array doesn't look local
1475 trustworthy
= LOCAL
;
1478 content
->array
.level
== LEVEL_CONTAINER
) {
1479 name
= content
->text_version
;
1480 trustworthy
= METADATA
;
1483 if (name
[0] && trustworthy
!= LOCAL
&&
1484 ! c
->require_homehost
&&
1485 conf_name_is_free(name
))
1486 trustworthy
= LOCAL
;
1488 if (trustworthy
== LOCAL
&&
1490 /* Ignore 'host:' prefix of name */
1491 name
= strchr(name
, ':')+1;
1493 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1497 st
->ss
->free_super(st
);
1502 mddev
= chosen_name
;
1503 if (pre_exist
== NULL
) {
1504 if (mddev_busy(fd2devnm(mdfd
))) {
1505 pr_err("%s already active, cannot restart it!\n",
1507 for (tmpdev
= devlist
;
1508 tmpdev
&& tmpdev
->used
!= 1;
1509 tmpdev
= tmpdev
->next
)
1511 if (tmpdev
&& auto_assem
)
1512 pr_err("%s needed for %s...\n",
1513 mddev
, tmpdev
->devname
);
1516 st
->ss
->free_super(st
);
1521 /* just incase it was started but has no content */
1522 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1525 if (content
!= &info
) {
1526 /* This is a member of a container. Try starting the array. */
1528 err
= assemble_container_content(st
, mdfd
, content
, c
,
1534 /* Ok, no bad inconsistancy, we can try updating etc */
1535 devices
= xcalloc(num_devs
, sizeof(*devices
));
1536 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1537 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1538 c
, content
, mdfd
, mddev
,
1539 &most_recent
, &bestcnt
, &best
, inargv
);
1546 pr_err("no devices found for %s\n",
1549 st
->ss
->free_super(st
);
1554 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1555 st
->minor_version
= 90;
1557 st
->ss
->getinfo_super(st
, content
, NULL
);
1558 clean
= content
->array
.state
& 1;
1560 /* now we have some devices that might be suitable.
1563 avail
= xcalloc(content
->array
.raid_disks
, 1);
1569 for (i
=0; i
< bestcnt
; i
++) {
1571 int event_margin
= 1; /* always allow a difference of '1'
1572 * like the kernel does
1574 if (j
< 0) continue;
1575 /* note: we ignore error flags in multipath arrays
1576 * as they don't make sense
1578 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1579 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1580 if (content
->journal_device_required
)
1582 else /* unexpected journal, mark as faulty */
1583 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1584 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1585 if (!(devices
[j
].i
.disk
.state
1586 & (1<<MD_DISK_FAULTY
))) {
1587 devices
[j
].uptodate
= 1;
1593 /* If this device thinks that 'most_recent' has failed, then
1594 * we must reject this device.
1596 if (j
!= most_recent
&& !c
->force
&&
1597 content
->array
.raid_disks
> 0 &&
1598 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1599 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1600 if (c
->verbose
> -1)
1601 pr_err("ignoring %s as it reports %s as failed\n",
1602 devices
[j
].devname
, devices
[most_recent
].devname
);
1606 /* Require event counter to be same as, or just less than,
1607 * most recent. If it is bigger, it must be a stray spare and
1608 * should be ignored.
1610 if (devices
[j
].i
.events
+event_margin
>=
1611 devices
[most_recent
].i
.events
&&
1612 devices
[j
].i
.events
<=
1613 devices
[most_recent
].i
.events
1615 devices
[j
].uptodate
= 1;
1616 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1618 if (i
< content
->array
.raid_disks
* 2) {
1619 if (devices
[j
].i
.recovery_start
== MaxSector
||
1620 (content
->reshape_active
&&
1621 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1629 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1635 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1636 avail
, most_recent
, st
, c
);
1641 /* Now we want to look at the superblock which the kernel will base things on
1642 * and compare the devices that we think are working with the devices that the
1643 * superblock thinks are working.
1644 * If there are differences and --force is given, then update this chosen
1648 st
->ss
->free_super(st
);
1649 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1655 if (!devices
[j
].uptodate
)
1657 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1660 if ((fd
=dev_open(devices
[j
].devname
,
1661 devices
[j
].included
? O_RDONLY
1662 : (O_RDONLY
|O_EXCL
)))< 0) {
1663 pr_err("Cannot open %s: %s\n",
1664 devices
[j
].devname
, strerror(errno
));
1667 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1669 pr_err("RAID superblock has disappeared from %s\n",
1670 devices
[j
].devname
);
1675 if (st
->sb
== NULL
) {
1676 pr_err("No suitable drives found for %s\n", mddev
);
1679 st
->ss
->getinfo_super(st
, content
, NULL
);
1680 if (sysfs_init(content
, mdfd
, NULL
)) {
1681 pr_err("Unable to initialize sysfs\n");
1685 /* after reload context, store journal_clean in context */
1686 content
->journal_clean
= journal_clean
;
1687 for (i
=0; i
<bestcnt
; i
++) {
1689 unsigned int desired_state
;
1693 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1694 desired_state
= (1<<MD_DISK_JOURNAL
);
1695 else if (i
>= content
->array
.raid_disks
* 2)
1698 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1700 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1702 if (!devices
[j
].uptodate
)
1705 devices
[j
].i
.disk
.state
= desired_state
;
1706 if (!(devices
[j
].i
.array
.state
& 1))
1709 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1710 c
->verbose
, 0, NULL
)) {
1712 if (c
->verbose
>= 0)
1713 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1714 j
, mddev
, devices
[j
].devname
);
1717 if (c
->verbose
>= -1)
1718 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1719 i
, mddev
, devices
[j
].devname
);
1723 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1724 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1729 if (c
->force
&& !clean
&&
1730 !enough(content
->array
.level
, content
->array
.raid_disks
,
1731 content
->array
.layout
, clean
,
1733 change
+= st
->ss
->update_super(st
, content
, "force-array",
1734 devices
[chosen_drive
].devname
, c
->verbose
,
1742 fd
= dev_open(devices
[chosen_drive
].devname
,
1743 devices
[chosen_drive
].included
?
1744 O_RDWR
: (O_RDWR
|O_EXCL
));
1746 pr_err("Could not open %s for write - cannot Assemble array.\n",
1747 devices
[chosen_drive
].devname
);
1750 if (st
->ss
->store_super(st
, fd
)) {
1752 pr_err("Could not re-write superblock on %s\n",
1753 devices
[chosen_drive
].devname
);
1756 if (c
->verbose
>= 0)
1757 pr_err("Marking array %s as 'clean'\n",
1762 /* If we are in the middle of a reshape we may need to restore saved data
1763 * that was moved aside due to the reshape overwriting live data
1764 * The code of doing this lives in Grow.c
1766 if (content
->reshape_active
&&
1767 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1769 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1771 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1773 if (!c
->backup_file
)
1774 c
->backup_file
= locate_backup(content
->sys_name
);
1775 enable_fds(bestcnt
/2);
1776 for (i
= 0; i
< bestcnt
/2; i
++) {
1779 fdlist
[i
] = dev_open(devices
[j
].devname
,
1781 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1782 if (fdlist
[i
] < 0) {
1783 pr_err("Could not open %s for write - cannot Assemble array.\n",
1784 devices
[j
].devname
);
1792 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1793 err
= st
->ss
->recover_backup(st
, content
);
1795 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1796 c
->backup_file
, c
->verbose
> 0);
1797 if (err
&& c
->invalid_backup
) {
1799 pr_err("continuing without restoring backup\n");
1805 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1809 pr_err("Failed to restore critical section for reshape, sorry.\n");
1810 if (c
->backup_file
== NULL
)
1811 cont_err("Possibly you needed to specify the --backup-file\n");
1816 /* Almost ready to actually *do* something */
1817 /* First, fill in the map, so that udev can find our name
1818 * as soon as we become active.
1820 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1821 content
->array
.major_version
= 1;
1822 content
->array
.minor_version
= 0;
1823 strcpy(content
->text_version
, "1.0");
1826 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1827 content
->uuid
, chosen_name
);
1829 rv
= start_array(mdfd
, mddev
, content
,
1830 st
, ident
, best
, bestcnt
,
1831 chosen_drive
, devices
, okcnt
, sparecnt
,
1832 rebuilding_cnt
, journalcnt
,
1834 clean
, avail
, start_partial_ok
,
1837 if (rv
== 1 && !pre_exist
)
1838 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1843 wait_for(chosen_name
, mdfd
);
1847 /* There is a nasty race with 'mdadm --monitor'.
1848 * If it opens this device before we close it,
1849 * it gets an incomplete open on which IO
1850 * doesn't work and the capacity is
1852 * If we reopen (to check for layered devices)
1853 * before --monitor closes, we loose.
1855 * So: wait upto 1 second for there to be
1856 * a non-zero capacity.
1858 while (usecs
< 1000) {
1859 mdfd
= open(mddev
, O_RDONLY
);
1861 unsigned long long size
;
1862 if (get_dev_size(mdfd
, NULL
, &size
) &&
1871 } else if (mdfd
>= 0)
1874 /* '2' means 'OK, but not started yet' */
1879 return rv
== 2 ? 0 : rv
;
1882 int assemble_container_content(struct supertype
*st
, int mdfd
,
1883 struct mdinfo
*content
, struct context
*c
,
1884 char *chosen_name
, int *result
)
1886 struct mdinfo
*dev
, *sra
, *dev2
;
1887 int working
= 0, preexist
= 0;
1894 if (sysfs_init(content
, mdfd
, NULL
)) {
1895 pr_err("Unable to initialize sysfs\n");
1899 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1900 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1901 if (content
->array
.major_version
== -1 &&
1902 content
->array
.minor_version
== -2 &&
1904 content
->text_version
[0] == '/')
1905 content
->text_version
[0] = '-';
1906 if (sysfs_set_array(content
, 9003) != 0) {
1912 /* There are two types of reshape: container wide or sub-array specific
1913 * Check if metadata requests blocking container wide reshapes
1915 start_reshape
= (content
->reshape_active
&&
1916 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1917 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1919 /* Block subarray here if it is under reshape now
1920 * Do not allow for any changes in this array
1922 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1923 block_subarray(content
);
1925 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1926 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1927 if (dev2
->disk
.major
== dev
->disk
.major
&&
1928 dev2
->disk
.minor
== dev
->disk
.minor
)
1932 /* Don't want this one any more */
1933 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1935 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1939 sysfs_set_str(sra
, dev2
, "state", "remove");
1941 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1942 avail
= xcalloc(content
->array
.raid_disks
, 1);
1943 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1944 if (dev
->disk
.raid_disk
>= 0)
1945 avail
[dev
->disk
.raid_disk
] = 1;
1946 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1947 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1948 content
->reshape_active
)
1952 } else if (errno
== EEXIST
)
1956 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1958 return 1;/* Nothing new, don't try to start */
1960 map_update(NULL
, fd2devnm(mdfd
), content
->text_version
,
1961 content
->uuid
, chosen_name
);
1963 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
1964 st
->ss
->validate_ppl
) {
1965 content
->array
.state
|= 1;
1968 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1973 ret
= st
->ss
->validate_ppl(st
, content
, dev
);
1983 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1985 content
->array
.state
&= ~1;
1986 avail
[dev
->disk
.raid_disk
] = 0;
1990 /* have --force - overwrite the invalid ppl */
1991 devpath
= map_dev(dev
->disk
.major
, dev
->disk
.minor
, 0);
1992 dfd
= dev_open(devpath
, O_RDWR
);
1994 pr_err("Failed to open %s\n", devpath
);
1999 err
= st
->ss
->write_init_ppl(st
, content
, dfd
);
2012 if (enough(content
->array
.level
, content
->array
.raid_disks
,
2013 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
2014 if (c
->export
&& result
)
2016 else if (c
->verbose
>= 0) {
2017 pr_err("%s assembled with %d device%s",
2018 chosen_name
, preexist
+ working
,
2019 preexist
+ working
== 1 ? "":"s");
2021 fprintf(stderr
, " (%d new)", working
);
2022 fprintf(stderr
, " but not started\n");
2029 if (c
->runstop
<= 0 &&
2030 (working
+ preexist
+ expansion
) <
2031 content
->array
.working_disks
) {
2032 if (c
->export
&& result
)
2033 *result
|= INCR_UNSAFE
;
2034 else if (c
->verbose
>= 0) {
2035 pr_err("%s assembled with %d device%s",
2036 chosen_name
, preexist
+ working
,
2037 preexist
+ working
== 1 ? "":"s");
2039 fprintf(stderr
, " (%d new)", working
);
2040 fprintf(stderr
, " but not safe to start\n");
2046 if (start_reshape
) {
2047 int spare
= content
->array
.raid_disks
+ expansion
;
2048 if (restore_backup(st
, content
,
2050 spare
, &c
->backup_file
, c
->verbose
) == 1)
2053 err
= sysfs_set_str(content
, NULL
,
2054 "array_state", "readonly");
2058 if (st
->ss
->external
) {
2059 if (!mdmon_running(st
->container_devnm
))
2060 start_mdmon(st
->container_devnm
);
2061 ping_monitor(st
->container_devnm
);
2062 if (mdmon_running(st
->container_devnm
) &&
2063 st
->update_tail
== NULL
)
2064 st
->update_tail
= &st
->updates
;
2067 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2068 0, c
->freeze_reshape
);
2069 } else switch(content
->array
.level
) {
2071 case LEVEL_MULTIPATH
:
2073 err
= sysfs_set_str(content
, NULL
, "array_state",
2074 c
->readonly
? "readonly" : "active");
2077 err
= sysfs_set_str(content
, NULL
, "array_state",
2079 /* start mdmon if needed. */
2081 if (!mdmon_running(st
->container_devnm
))
2082 start_mdmon(st
->container_devnm
);
2083 ping_monitor(st
->container_devnm
);
2088 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2090 /* Block subarray here if it is not reshaped now
2091 * It has be blocked a little later to allow mdmon to switch in
2094 if (st
->ss
->external
&& content
->recovery_blocked
&&
2096 block_subarray(content
);
2098 if (c
->export
&& result
) {
2102 *result
|= INCR_YES
;
2103 } else if (c
->verbose
>= 0) {
2105 pr_err("array %s now has %d device%s",
2106 chosen_name
, working
+ preexist
,
2107 working
+ preexist
== 1 ? "":"s");
2109 pr_err("Started %s with %d device%s",
2110 chosen_name
, working
+ preexist
,
2111 working
+ preexist
== 1 ? "":"s");
2113 fprintf(stderr
, " (%d new)", working
);
2115 fprintf(stderr
, " ( + %d for expansion)",
2117 fprintf(stderr
, "\n");
2120 wait_for(chosen_name
, mdfd
);
2122 /* FIXME should have an O_EXCL and wait for read-auto */