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
) {
516 if (tmpdev
->used
!= 3)
518 if (stat(tmpdev
->devname
, &stb
)< 0) {
519 pr_err("fstat failed for %s: %s\n",
520 tmpdev
->devname
, strerror(errno
));
523 struct dev_policy
*pol
= devid_policy(stb
.st_rdev
);
524 int dt
= domain_test(domains
, pol
, NULL
);
525 if (inargv
&& dt
!= 0)
526 /* take this spare as domains match
527 * if there are any */
529 else if (!inargv
&& dt
== 1)
530 /* device wasn't explicitly listed, so need
531 * explicit domain match - which we have */
534 /* if domains don't match mark as unused */
536 dev_policy_free(pol
);
539 domain_free(domains
);
541 if (st
&& st
->sb
&& content
== *contentp
)
542 st
->ss
->getinfo_super(st
, content
, NULL
);
550 int uptodate
; /* set once we decide that this device is as
551 * recent as everything else in the array.
553 int included
; /* set if the device is already in the array
554 * due to a previous '-I'
559 static int load_devices(struct devs
*devices
, char *devmap
,
560 struct mddev_ident
*ident
, struct supertype
**stp
,
561 struct mddev_dev
*devlist
, struct context
*c
,
562 struct mdinfo
*content
,
563 int mdfd
, char *mddev
,
564 int *most_recentp
, int *bestcntp
, int **bestp
,
567 struct mddev_dev
*tmpdev
;
571 int most_recent
= -1;
574 struct supertype
*st
= *stp
;
576 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
577 char *devname
= tmpdev
->devname
;
579 struct supertype
*tst
;
583 if (tmpdev
->used
!= 1)
585 /* looks like a good enough match to update the super block if needed */
587 /* prepare useful information in info structures */
592 if (strcmp(c
->update
, "uuid") == 0 && !ident
->uuid_set
)
593 random_uuid((__u8
*)ident
->uuid
);
595 if (strcmp(c
->update
, "ppl") == 0 &&
596 ident
->bitmap_fd
>= 0) {
597 pr_err("PPL is not compatible with bitmap\n");
601 dfd
= dev_open(devname
,
602 tmpdev
->disposition
== 'I'
603 ? O_RDWR
: (O_RDWR
|O_EXCL
));
606 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
607 pr_err("cannot re-read metadata from %s - aborting\n",
614 tst
->ss
->free_super(tst
);
619 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
621 memcpy(content
->uuid
, ident
->uuid
, 16);
622 strcpy(content
->name
, ident
->name
);
623 content
->array
.md_minor
= minor(stb2
.st_rdev
);
625 if (strcmp(c
->update
, "byteorder") == 0)
627 else if (strcmp(c
->update
, "home-cluster") == 0) {
628 tst
->cluster_name
= c
->homecluster
;
629 err
= tst
->ss
->write_bitmap(tst
, dfd
, NameUpdate
);
630 } else if (strcmp(c
->update
, "nodes") == 0) {
631 tst
->nodes
= c
->nodes
;
632 err
= tst
->ss
->write_bitmap(tst
, dfd
, NodeNumUpdate
);
633 } else if (strcmp(c
->update
, "revert-reshape") == 0 &&
635 err
= tst
->ss
->update_super(tst
, content
,
636 "revert-reshape-nobackup",
641 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
647 pr_err("--update=%s not understood for %s metadata\n",
648 c
->update
, tst
->ss
->name
);
649 tst
->ss
->free_super(tst
);
658 if (strcmp(c
->update
, "uuid")==0 &&
661 memcpy(ident
->uuid
, content
->uuid
, 16);
663 if (tst
->ss
->store_super(tst
, dfd
))
664 pr_err("Could not re-write superblock on %s.\n",
667 if (strcmp(c
->update
, "uuid")==0 &&
668 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
669 if (bitmap_update_uuid(ident
->bitmap_fd
,
671 tst
->ss
->swapuuid
) != 0)
672 pr_err("Could not update uuid on external bitmap.\n");
677 dfd
= dev_open(devname
,
678 tmpdev
->disposition
== 'I'
679 ? O_RDWR
: (O_RDWR
|O_EXCL
));
682 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
683 pr_err("cannot re-read metadata from %s - aborting\n",
690 tst
->ss
->free_super(tst
);
695 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
702 pr_err("%s is identified as a member of %s, slot %d%s.\n",
703 devname
, mddev
, content
->disk
.raid_disk
,
704 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
705 devices
[devcnt
].devname
= devname
;
706 devices
[devcnt
].uptodate
= 0;
707 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
708 devices
[devcnt
].i
= *content
;
709 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
710 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
712 if (devices
[devcnt
].i
.disk
.state
== 6) {
713 if (most_recent
< 0 ||
714 devices
[devcnt
].i
.events
715 > devices
[most_recent
].i
.events
) {
716 struct supertype
*tmp
= tst
;
719 most_recent
= devcnt
;
722 tst
->ss
->free_super(tst
);
725 if (content
->array
.level
== LEVEL_MULTIPATH
)
726 /* with multipath, the raid_disk from the superblock is meaningless */
729 i
= devices
[devcnt
].i
.disk
.raid_disk
;
730 if (i
+1 == 0 || i
== MD_DISK_ROLE_JOURNAL
) {
731 if (nextspare
< content
->array
.raid_disks
*2)
732 nextspare
= content
->array
.raid_disks
*2;
735 /* i is raid_disk - double it so there is room for
738 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
740 if (i
>= content
->array
.raid_disks
*2 &&
746 int newbestcnt
= i
+10;
747 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
749 for (c
=0; c
< newbestcnt
; c
++)
751 newbest
[c
] = best
[c
];
756 bestcnt
= newbestcnt
;
759 devices
[best
[i
]].i
.events
760 == devices
[devcnt
].i
.events
761 && (devices
[best
[i
]].i
.disk
.minor
762 != devices
[devcnt
].i
.disk
.minor
)
764 && content
->array
.level
!= LEVEL_MULTIPATH
) {
765 /* two different devices with identical superblock.
766 * Could be a mis-detection caused by overlapping
767 * partitions. fail-safe.
769 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
770 " If they are really different, please --zero the superblock on one\n"
771 " If they are the same or overlap, please remove one from %s.\n",
772 devices
[best
[i
]].devname
, devname
,
773 inargv
? "the list" :
774 "the\n DEVICE list in mdadm.conf"
783 || (devices
[best
[i
]].i
.events
784 < devices
[devcnt
].i
.events
))
789 if (most_recent
>= 0)
790 *most_recentp
= most_recent
;
797 static int force_array(struct mdinfo
*content
,
798 struct devs
*devices
,
799 int *best
, int bestcnt
, char *avail
,
801 struct supertype
*st
,
805 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
806 content
->array
.layout
, 1,
809 (content
->reshape_active
&& content
->delta_disks
> 0 &&
810 !enough(content
->array
.level
, (content
->array
.raid_disks
811 - content
->delta_disks
),
812 content
->new_layout
, 1,
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)
857 if (chosen_drive
< 0 ||
859 > devices
[chosen_drive
].i
.events
)
862 if (chosen_drive
< 0)
864 current_events
= devices
[chosen_drive
].i
.events
;
867 pr_err("forcing event count in %s(%d) from %d upto %d\n",
868 devices
[chosen_drive
].devname
,
869 devices
[chosen_drive
].i
.disk
.raid_disk
,
870 (int)(devices
[chosen_drive
].i
.events
),
871 (int)(devices
[most_recent
].i
.events
));
872 fd
= dev_open(devices
[chosen_drive
].devname
,
873 devices
[chosen_drive
].included
? O_RDWR
876 pr_err("Couldn't open %s for write - not updating\n",
877 devices
[chosen_drive
].devname
);
878 devices
[chosen_drive
].i
.events
= 0;
882 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
884 pr_err("RAID superblock disappeared from %s - not updating.\n",
885 devices
[chosen_drive
].devname
);
886 devices
[chosen_drive
].i
.events
= 0;
889 content
->events
= devices
[most_recent
].i
.events
;
890 tst
->ss
->update_super(tst
, content
, "force-one",
891 devices
[chosen_drive
].devname
, c
->verbose
,
894 if (tst
->ss
->store_super(tst
, fd
)) {
896 pr_err("Could not re-write superblock on %s\n",
897 devices
[chosen_drive
].devname
);
898 devices
[chosen_drive
].i
.events
= 0;
899 tst
->ss
->free_super(tst
);
903 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
904 devices
[chosen_drive
].uptodate
= 1;
905 avail
[chosen_drive
] = 1;
907 tst
->ss
->free_super(tst
);
908 /* If there are any other drives of the same vintage,
909 * add them in as well. We can't lose and we might gain
912 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
916 !devices
[j
].uptodate
&&
917 devices
[j
].i
.recovery_start
== MaxSector
&&
918 devices
[j
].i
.events
== current_events
) {
927 static int start_array(int mdfd
,
929 struct mdinfo
*content
,
930 struct supertype
*st
,
931 struct mddev_ident
*ident
,
932 int *best
, int bestcnt
,
934 struct devs
*devices
,
936 unsigned int sparecnt
,
937 unsigned int rebuilding_cnt
,
938 unsigned int journalcnt
,
940 int clean
, char *avail
,
941 int start_partial_ok
,
948 unsigned int req_cnt
;
950 if (content
->journal_device_required
&& (content
->journal_clean
== 0)) {
952 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
955 pr_err("Journal is missing or stale, starting array read only.\n");
959 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
962 rv
= set_array_info(mdfd
, st
, content
);
964 pr_err("failed to set array info for %s: %s\n",
965 mddev
, strerror(errno
));
968 if (ident
->bitmap_fd
>= 0) {
969 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
970 pr_err("SET_BITMAP_FILE failed.\n");
973 } else if (ident
->bitmap_file
) {
974 /* From config file */
975 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
977 pr_err("Could not open bitmap file %s\n",
981 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
982 pr_err("Failed to set bitmapfile for %s\n", mddev
);
989 /* First, add the raid disks, but add the chosen one last */
990 for (i
= 0; i
<= bestcnt
; i
++) {
994 if (j
== chosen_drive
)
999 if (j
>= 0 && !devices
[j
].included
) {
1002 dfd
= dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1004 remove_partitions(dfd
);
1007 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1010 pr_err("failed to add %s to %s: %s\n",
1011 devices
[j
].devname
, mddev
,
1013 if (i
< content
->array
.raid_disks
* 2 ||
1018 } else if (c
->verbose
> 0) {
1019 pr_err("added %s to %s as %d%s%s\n",
1020 devices
[j
].devname
, mddev
,
1021 devices
[j
].i
.disk
.raid_disk
,
1022 devices
[j
].uptodate
?"":
1023 " (possibly out of date)",
1024 (devices
[j
].i
.disk
.state
&
1025 (1<<MD_DISK_REPLACEMENT
)) ?
1028 } else if (j
>= 0) {
1030 pr_err("%s is already in %s as %d\n",
1031 devices
[j
].devname
, mddev
,
1032 devices
[j
].i
.disk
.raid_disk
);
1033 } else if (c
->verbose
> 0 &&
1034 i
< content
->array
.raid_disks
* 2 && (i
& 1) == 0)
1035 pr_err("no uptodate device for slot %d of %s\n",
1039 if (content
->array
.level
== LEVEL_CONTAINER
) {
1040 if (c
->verbose
>= 0) {
1041 pr_err("Container %s has been assembled with %d drive%s",
1042 mddev
, okcnt
+ sparecnt
+ journalcnt
,
1043 okcnt
+ sparecnt
+ journalcnt
== 1 ? "" : "s");
1044 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1045 fprintf(stderr
, " (out of %d)",
1046 content
->array
.raid_disks
);
1047 fprintf(stderr
, "\n");
1050 if (st
->ss
->validate_container
) {
1051 struct mdinfo
*devices_list
;
1052 struct mdinfo
*info_devices
;
1055 devices_list
= NULL
;
1056 info_devices
= xmalloc(sizeof(struct mdinfo
) *
1057 (okcnt
+ sparecnt
));
1058 for (count
= 0; count
< okcnt
+ sparecnt
; count
++) {
1059 info_devices
[count
] = devices
[count
].i
;
1060 info_devices
[count
].next
= devices_list
;
1061 devices_list
= &info_devices
[count
];
1063 if (st
->ss
->validate_container(devices_list
))
1064 pr_err("Mismatch detected!\n");
1068 st
->ss
->free_super(st
);
1069 sysfs_uevent(content
, "change");
1070 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1071 /* Was partial, is still partial, so signal an error
1072 * to ensure we don't retry */
1077 /* Get number of in-sync devices according to the superblock.
1078 * We must have this number to start the array without -s or -R
1080 req_cnt
= content
->array
.working_disks
;
1082 if (c
->runstop
== 1 ||
1084 (enough(content
->array
.level
, content
->array
.raid_disks
,
1085 content
->array
.layout
, clean
, avail
) &&
1086 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)))) {
1087 /* This array is good-to-go.
1088 * If a reshape is in progress then we might need to
1089 * continue monitoring it. In that case we start
1090 * it read-only and let the grow code make it writable.
1094 if (content
->reshape_active
&&
1095 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1096 content
->delta_disks
<= 0) {
1097 if (!c
->backup_file
) {
1098 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1100 pr_err("Please provided one with \"--backup-file=...\"\n");
1102 strcmp(c
->update
, "revert-reshape") == 0)
1103 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1106 rv
= sysfs_set_str(content
, NULL
,
1107 "array_state", "readonly");
1109 rv
= Grow_continue(mdfd
, st
, content
,
1112 } else if (c
->readonly
&&
1113 sysfs_attribute_available(content
, NULL
,
1115 rv
= sysfs_set_str(content
, NULL
,
1116 "array_state", "readonly");
1118 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1119 reopen_mddev(mdfd
); /* drop O_EXCL */
1121 if (c
->verbose
>= 0) {
1122 pr_err("%s has been started with %d drive%s",
1123 mddev
, okcnt
, okcnt
==1?"":"s");
1124 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1125 fprintf(stderr
, " (out of %d)",
1126 content
->array
.raid_disks
);
1128 fprintf(stderr
, "%s %d rebuilding",
1129 sparecnt
?",":" and",
1132 fprintf(stderr
, " and %d spare%s",
1134 sparecnt
== 1 ? "" : "s");
1135 if (content
->journal_clean
)
1136 fprintf(stderr
, " and %d journal",
1138 fprintf(stderr
, ".\n");
1140 if (content
->reshape_active
&&
1141 content
->array
.level
>= 4 &&
1142 content
->array
.level
<= 6) {
1143 /* might need to increase the size
1144 * of the stripe cache - default is 256
1146 int chunk_size
= content
->array
.chunk_size
;
1148 if (content
->reshape_active
&&
1149 content
->new_chunk
> chunk_size
)
1150 chunk_size
= content
->new_chunk
;
1151 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1154 sra
= sysfs_read(mdfd
, NULL
, 0);
1156 sysfs_set_num(sra
, NULL
,
1157 "stripe_cache_size",
1158 (4 * chunk_size
/ 4096) + 1);
1162 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1163 /* If any devices did not get added
1164 * because the kernel rejected them based
1165 * on event count, try adding them
1166 * again providing the action policy is
1167 * 're-add' or greater. The bitmap
1168 * might allow them to be included, or
1169 * they will become spares.
1171 for (i
= 0; i
< bestcnt
; i
++) {
1173 if (j
>= 0 && !devices
[j
].uptodate
) {
1174 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1176 rv
= add_disk(mdfd
, st
, content
,
1178 if (rv
== 0 && c
->verbose
>= 0)
1179 pr_err("%s has been re-added.\n",
1180 devices
[j
].devname
);
1184 if (content
->array
.level
== 6 &&
1185 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1189 sra
= sysfs_read(mdfd
, NULL
, 0);
1191 sysfs_set_str(sra
, NULL
,
1192 "sync_action", "repair");
1197 pr_err("failed to RUN_ARRAY %s: %s\n", mddev
, strerror(errno
));
1199 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1200 content
->array
.layout
, 1, avail
))
1201 pr_err("Not enough devices to start the array.\n");
1202 else if (!enough(content
->array
.level
,
1203 content
->array
.raid_disks
,
1204 content
->array
.layout
, clean
, avail
))
1205 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1209 if (c
->runstop
== -1) {
1210 pr_err("%s assembled from %d drive%s",
1211 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1212 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1213 fprintf(stderr
, " (out of %d)",
1214 content
->array
.raid_disks
);
1215 fprintf(stderr
, ", but not started.\n");
1218 if (c
->verbose
>= -1) {
1219 pr_err("%s assembled from %d drive%s",
1220 mddev
, okcnt
, okcnt
== 1 ? "" : "s");
1222 fprintf(stderr
, "%s %d rebuilding",
1223 sparecnt
? "," : " and", rebuilding_cnt
);
1225 fprintf(stderr
, " and %d spare%s", sparecnt
,
1226 sparecnt
== 1 ? "" : "s");
1227 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1228 content
->array
.layout
, 1, avail
))
1229 fprintf(stderr
, " - not enough to start the array.\n");
1230 else if (!enough(content
->array
.level
,
1231 content
->array
.raid_disks
,
1232 content
->array
.layout
, clean
, avail
))
1233 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1235 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1236 fprintf(stderr
, " - need all %d to start it",
1239 fprintf(stderr
, " - need %d to start", req_cnt
);
1240 fprintf(stderr
, " (use --run to insist).\n");
1246 int Assemble(struct supertype
*st
, char *mddev
,
1247 struct mddev_ident
*ident
,
1248 struct mddev_dev
*devlist
,
1252 * The task of Assemble is to find a collection of
1253 * devices that should (according to their superblocks)
1254 * form an array, and to give this collection to the MD driver.
1255 * In Linux-2.4 and later, this involves submitting a
1256 * SET_ARRAY_INFO ioctl with no arg - to prepare
1257 * the array - and then submit a number of
1258 * ADD_NEW_DISK ioctls to add disks into
1259 * the array. Finally RUN_ARRAY might
1260 * be submitted to start the array.
1262 * Much of the work of Assemble is in finding and/or
1263 * checking the disks to make sure they look right.
1265 * If mddev is not set, then scan must be set and we
1266 * read through the config file for dev+uuid mapping
1267 * We recurse, setting mddev, for each device that
1269 * - has a valid uuid (or any uuid if !uuidset)
1271 * If mddev is set, we try to determine state of md.
1272 * check version - must be at least 0.90.0
1273 * check kernel version. must be at least 2.4.
1274 * If not, we can possibly fall back on START_ARRAY
1275 * Try to GET_ARRAY_INFO.
1276 * If possible, give up
1277 * If not, try to STOP_ARRAY just to make sure
1279 * If !uuidset and scan, look in conf-file for uuid
1280 * If not found, give up
1281 * If !devlist and scan and uuidset, get list of devs from conf-file
1284 * Check superblock - discard if bad
1285 * Check uuid (set if we don't have one) - discard if no match
1286 * Check superblock similarity if we have a superblock - discard if different
1287 * Record events, devicenum
1288 * This should give us a list of devices for the array
1289 * We should collect the most recent event number
1291 * Count disks with recent enough event count
1292 * While force && !enough disks
1293 * Choose newest rejected disks, update event count
1294 * mark clean and rewrite superblock
1297 * foreach device with recent events : ADD_NEW_DISK
1298 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1300 * Check the device numbers in superblock are right
1301 * update superblock if any changes
1308 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1309 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1310 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1311 struct devs
*devices
;
1313 int *best
= NULL
; /* indexed by raid_disk */
1316 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
, journalcnt
;
1317 int journal_clean
= 0;
1320 int most_recent
= 0;
1324 int start_partial_ok
= (c
->runstop
>= 0) &&
1325 (c
->force
|| devlist
==NULL
|| auto_assem
);
1327 struct mddev_dev
*tmpdev
;
1329 struct mdinfo
*content
= NULL
;
1330 struct mdinfo
*pre_exist
= NULL
;
1333 char chosen_name
[1024];
1334 struct map_ent
*map
= NULL
;
1338 * If any subdevs are listed, then any that don't
1339 * match ident are discarded. Remainder must all match and
1341 * If no subdevs, then we scan all devices in the config file, but
1342 * there must be something in the identity
1346 ident
->uuid_set
== 0 &&
1347 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1348 ident
->name
[0] == 0 &&
1349 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1350 ident
->devices
== NULL
) {
1351 pr_err("No identity information available for %s - cannot assemble.\n",
1352 mddev
? mddev
: "further assembly");
1356 if (devlist
== NULL
)
1357 devlist
= conf_get_devs();
1362 /* We come back here when doing auto-assembly and attempting some
1363 * set of devices failed. Those are now marked as ->used==2 and
1364 * we ignore them and try again
1366 if (!st
&& ident
->st
)
1369 pr_err("looking for devices for %s\n",
1370 mddev
? mddev
: "further assembly");
1374 st
->ignore_hw_compat
= 1;
1375 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1376 inargv
, auto_assem
);
1380 if (!st
|| !st
->sb
|| !content
)
1383 /* We have a full set of devices - we now need to find the
1385 * However there is a risk that we are racing with "mdadm -I"
1386 * and the array is already partially assembled - we will have
1387 * rejected any devices already in this address.
1388 * So we take a lock on the map file - to prevent further races -
1389 * and look for the uuid in there. If found and the array is
1390 * active, we abort. If found and the array is not active
1391 * we commit to that md device and add all the contained devices
1392 * to our list. We flag them so that we don't try to re-add,
1393 * but can remove if they turn out to not be wanted.
1396 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1397 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1400 mp
= map_by_uuid(&map
, content
->uuid
);
1403 /* array already exists. */
1404 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1405 if (pre_exist
->array
.level
!= UnSet
) {
1406 pr_err("Found some drive for an array that is already active: %s\n",
1408 pr_err("giving up.\n");
1411 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1412 /* We want to add this device to our list,
1413 * but it could already be there if "mdadm -I"
1414 * started *after* we checked for O_EXCL.
1415 * If we add it to the top of the list
1416 * it will be preferred over later copies.
1418 struct mddev_dev
*newdev
;
1419 char *devname
= map_dev(dv
->disk
.major
,
1424 newdev
= xmalloc(sizeof(*newdev
));
1425 newdev
->devname
= devname
;
1426 newdev
->disposition
= 'I';
1428 newdev
->next
= devlist
;
1432 strcpy(chosen_name
, mp
->path
);
1433 if (c
->verbose
> 0 || mddev
== NULL
||
1434 strcmp(mddev
, chosen_name
) != 0)
1435 pr_err("Merging with already-assembled %s\n",
1437 mdfd
= open_dev_excl(mp
->devnm
);
1439 int trustworthy
= FOREIGN
;
1440 name
= content
->name
;
1441 switch (st
->ss
->match_home(st
, c
->homehost
)
1442 ?: st
->ss
->match_home(st
, "any")) {
1444 trustworthy
= LOCAL
;
1445 name
= strchr(content
->name
, ':');
1449 name
= content
->name
;
1453 /* If the array is listed in mdadm.conf or on
1454 * command line, then we trust the name
1455 * even if the array doesn't look local
1457 trustworthy
= LOCAL
;
1460 content
->array
.level
== LEVEL_CONTAINER
) {
1461 name
= content
->text_version
;
1462 trustworthy
= METADATA
;
1465 if (name
[0] && trustworthy
!= LOCAL
&&
1466 ! c
->require_homehost
&&
1467 conf_name_is_free(name
))
1468 trustworthy
= LOCAL
;
1470 if (trustworthy
== LOCAL
&&
1472 /* Ignore 'host:' prefix of name */
1473 name
= strchr(name
, ':')+1;
1475 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1479 st
->ss
->free_super(st
);
1484 mddev
= chosen_name
;
1485 if (pre_exist
== NULL
) {
1486 if (mddev_busy(fd2devnm(mdfd
))) {
1487 pr_err("%s already active, cannot restart it!\n",
1489 for (tmpdev
= devlist
;
1490 tmpdev
&& tmpdev
->used
!= 1;
1491 tmpdev
= tmpdev
->next
)
1493 if (tmpdev
&& auto_assem
)
1494 pr_err("%s needed for %s...\n",
1495 mddev
, tmpdev
->devname
);
1498 st
->ss
->free_super(st
);
1503 /* just incase it was started but has no content */
1504 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1507 if (content
!= &info
) {
1508 /* This is a member of a container. Try starting the array. */
1510 err
= assemble_container_content(st
, mdfd
, content
, c
,
1516 /* Ok, no bad inconsistancy, we can try updating etc */
1517 devices
= xcalloc(num_devs
, sizeof(*devices
));
1518 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1519 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1520 c
, content
, mdfd
, mddev
,
1521 &most_recent
, &bestcnt
, &best
, inargv
);
1526 pr_err("no devices found for %s\n",
1529 st
->ss
->free_super(st
);
1536 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1537 st
->minor_version
= 90;
1539 st
->ss
->getinfo_super(st
, content
, NULL
);
1540 clean
= content
->array
.state
& 1;
1542 /* now we have some devices that might be suitable.
1545 avail
= xcalloc(content
->array
.raid_disks
, 1);
1551 for (i
=0; i
< bestcnt
; i
++) {
1553 int event_margin
= 1; /* always allow a difference of '1'
1554 * like the kernel does
1556 if (j
< 0) continue;
1557 /* note: we ignore error flags in multipath arrays
1558 * as they don't make sense
1560 if (content
->array
.level
!= LEVEL_MULTIPATH
) {
1561 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
)) {
1562 if (content
->journal_device_required
)
1564 else /* unexpected journal, mark as faulty */
1565 devices
[j
].i
.disk
.state
|= (1<<MD_DISK_FAULTY
);
1566 } else if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1567 if (!(devices
[j
].i
.disk
.state
1568 & (1<<MD_DISK_FAULTY
))) {
1569 devices
[j
].uptodate
= 1;
1575 /* If this device thinks that 'most_recent' has failed, then
1576 * we must reject this device.
1578 if (j
!= most_recent
&& !c
->force
&&
1579 content
->array
.raid_disks
> 0 &&
1580 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1581 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1582 if (c
->verbose
> -1)
1583 pr_err("ignoring %s as it reports %s as failed\n",
1584 devices
[j
].devname
, devices
[most_recent
].devname
);
1588 /* Require event counter to be same as, or just less than,
1589 * most recent. If it is bigger, it must be a stray spare and
1590 * should be ignored.
1592 if (devices
[j
].i
.events
+event_margin
>=
1593 devices
[most_recent
].i
.events
&&
1594 devices
[j
].i
.events
<=
1595 devices
[most_recent
].i
.events
1597 devices
[j
].uptodate
= 1;
1598 if (devices
[j
].i
.disk
.state
& (1<<MD_DISK_JOURNAL
))
1600 if (i
< content
->array
.raid_disks
* 2) {
1601 if (devices
[j
].i
.recovery_start
== MaxSector
||
1602 (content
->reshape_active
&&
1603 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1611 } else if (devices
[j
].i
.disk
.raid_disk
!= MD_DISK_ROLE_JOURNAL
)
1617 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1618 avail
, most_recent
, st
, c
);
1623 /* Now we want to look at the superblock which the kernel will base things on
1624 * and compare the devices that we think are working with the devices that the
1625 * superblock thinks are working.
1626 * If there are differences and --force is given, then update this chosen
1630 st
->ss
->free_super(st
);
1631 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1637 if (!devices
[j
].uptodate
)
1639 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1642 if ((fd
=dev_open(devices
[j
].devname
,
1643 devices
[j
].included
? O_RDONLY
1644 : (O_RDONLY
|O_EXCL
)))< 0) {
1645 pr_err("Cannot open %s: %s\n",
1646 devices
[j
].devname
, strerror(errno
));
1651 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1653 pr_err("RAID superblock has disappeared from %s\n",
1654 devices
[j
].devname
);
1661 if (st
->sb
== NULL
) {
1662 pr_err("No suitable drives found for %s\n", mddev
);
1667 st
->ss
->getinfo_super(st
, content
, NULL
);
1668 if (sysfs_init(content
, mdfd
, NULL
)) {
1669 pr_err("Unable to initialize sysfs\n");
1675 /* after reload context, store journal_clean in context */
1676 content
->journal_clean
= journal_clean
;
1677 for (i
=0; i
<bestcnt
; i
++) {
1679 unsigned int desired_state
;
1681 if (devices
[j
].i
.disk
.raid_disk
== MD_DISK_ROLE_JOURNAL
)
1682 desired_state
= (1<<MD_DISK_JOURNAL
);
1683 else if (i
>= content
->array
.raid_disks
* 2)
1686 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1688 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1692 if (!devices
[j
].uptodate
)
1695 devices
[j
].i
.disk
.state
= desired_state
;
1696 if (!(devices
[j
].i
.array
.state
& 1))
1699 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1700 c
->verbose
, 0, NULL
)) {
1702 if (c
->verbose
>= 0)
1703 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1704 j
, mddev
, devices
[j
].devname
);
1707 if (c
->verbose
>= -1)
1708 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1709 i
, mddev
, devices
[j
].devname
);
1713 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1714 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1719 if (c
->force
&& !clean
&&
1720 !enough(content
->array
.level
, content
->array
.raid_disks
,
1721 content
->array
.layout
, clean
,
1723 change
+= st
->ss
->update_super(st
, content
, "force-array",
1724 devices
[chosen_drive
].devname
, c
->verbose
,
1732 fd
= dev_open(devices
[chosen_drive
].devname
,
1733 devices
[chosen_drive
].included
?
1734 O_RDWR
: (O_RDWR
|O_EXCL
));
1736 pr_err("Could not open %s for write - cannot Assemble array.\n",
1737 devices
[chosen_drive
].devname
);
1742 if (st
->ss
->store_super(st
, fd
)) {
1744 pr_err("Could not re-write superblock on %s\n",
1745 devices
[chosen_drive
].devname
);
1750 if (c
->verbose
>= 0)
1751 pr_err("Marking array %s as 'clean'\n",
1756 /* If we are in the middle of a reshape we may need to restore saved data
1757 * that was moved aside due to the reshape overwriting live data
1758 * The code of doing this lives in Grow.c
1760 if (content
->reshape_active
&&
1761 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1763 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1765 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1767 if (!c
->backup_file
)
1768 c
->backup_file
= locate_backup(content
->sys_name
);
1769 enable_fds(bestcnt
/2);
1770 for (i
= 0; i
< bestcnt
/2; i
++) {
1773 fdlist
[i
] = dev_open(devices
[j
].devname
,
1775 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1776 if (fdlist
[i
] < 0) {
1777 pr_err("Could not open %s for write - cannot Assemble array.\n",
1778 devices
[j
].devname
);
1786 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1787 err
= st
->ss
->recover_backup(st
, content
);
1789 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1790 c
->backup_file
, c
->verbose
> 0);
1791 if (err
&& c
->invalid_backup
) {
1793 pr_err("continuing without restoring backup\n");
1799 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1803 pr_err("Failed to restore critical section for reshape, sorry.\n");
1804 if (c
->backup_file
== NULL
)
1805 cont_err("Possibly you needed to specify the --backup-file\n");
1812 /* Almost ready to actually *do* something */
1813 /* First, fill in the map, so that udev can find our name
1814 * as soon as we become active.
1816 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1817 content
->array
.major_version
= 1;
1818 content
->array
.minor_version
= 0;
1819 strcpy(content
->text_version
, "1.0");
1822 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1823 content
->uuid
, chosen_name
);
1825 rv
= start_array(mdfd
, mddev
, content
,
1826 st
, ident
, best
, bestcnt
,
1827 chosen_drive
, devices
, okcnt
, sparecnt
,
1828 rebuilding_cnt
, journalcnt
,
1830 clean
, avail
, start_partial_ok
,
1833 if (rv
== 1 && !pre_exist
)
1834 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1838 wait_for(chosen_name
, mdfd
);
1842 /* There is a nasty race with 'mdadm --monitor'.
1843 * If it opens this device before we close it,
1844 * it gets an incomplete open on which IO
1845 * doesn't work and the capacity is
1847 * If we reopen (to check for layered devices)
1848 * before --monitor closes, we loose.
1850 * So: wait upto 1 second for there to be
1851 * a non-zero capacity.
1853 while (usecs
< 1000) {
1854 mdfd
= open(mddev
, O_RDONLY
);
1856 unsigned long long size
;
1857 if (get_dev_size(mdfd
, NULL
, &size
) &&
1869 /* '2' means 'OK, but not started yet' */
1870 return rv
== 2 ? 0 : rv
;
1873 int assemble_container_content(struct supertype
*st
, int mdfd
,
1874 struct mdinfo
*content
, struct context
*c
,
1875 char *chosen_name
, int *result
)
1877 struct mdinfo
*dev
, *sra
, *dev2
;
1878 int working
= 0, preexist
= 0;
1885 if (sysfs_init(content
, mdfd
, NULL
)) {
1886 pr_err("Unable to initialize sysfs\n");
1890 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1891 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1892 if (content
->array
.major_version
== -1 &&
1893 content
->array
.minor_version
== -2 &&
1895 content
->text_version
[0] == '/')
1896 content
->text_version
[0] = '-';
1897 if (sysfs_set_array(content
, 9003) != 0) {
1903 /* There are two types of reshape: container wide or sub-array specific
1904 * Check if metadata requests blocking container wide reshapes
1906 start_reshape
= (content
->reshape_active
&&
1907 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1908 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1910 /* Block subarray here if it is under reshape now
1911 * Do not allow for any changes in this array
1913 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1914 block_subarray(content
);
1916 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1917 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1918 if (dev2
->disk
.major
== dev
->disk
.major
&&
1919 dev2
->disk
.minor
== dev
->disk
.minor
)
1923 /* Don't want this one any more */
1924 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1926 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1930 sysfs_set_str(sra
, dev2
, "state", "remove");
1932 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1933 avail
= xcalloc(content
->array
.raid_disks
, 1);
1934 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1935 if (dev
->disk
.raid_disk
>= 0)
1936 avail
[dev
->disk
.raid_disk
] = 1;
1937 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1938 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1939 content
->reshape_active
)
1943 } else if (errno
== EEXIST
)
1947 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1949 return 1;/* Nothing new, don't try to start */
1951 map_update(NULL
, fd2devnm(mdfd
), content
->text_version
,
1952 content
->uuid
, chosen_name
);
1954 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
1955 st
->ss
->validate_ppl
) {
1956 content
->array
.state
|= 1;
1959 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1964 ret
= st
->ss
->validate_ppl(st
, content
, dev
);
1974 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1976 content
->array
.state
&= ~1;
1977 avail
[dev
->disk
.raid_disk
] = 0;
1981 /* have --force - overwrite the invalid ppl */
1982 devpath
= map_dev(dev
->disk
.major
, dev
->disk
.minor
, 0);
1983 dfd
= dev_open(devpath
, O_RDWR
);
1985 pr_err("Failed to open %s\n", devpath
);
1990 err
= st
->ss
->write_init_ppl(st
, content
, dfd
);
2003 if (enough(content
->array
.level
, content
->array
.raid_disks
,
2004 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
2005 if (c
->export
&& result
)
2007 else if (c
->verbose
>= 0) {
2008 pr_err("%s assembled with %d device%s",
2009 chosen_name
, preexist
+ working
,
2010 preexist
+ working
== 1 ? "":"s");
2012 fprintf(stderr
, " (%d new)", working
);
2013 fprintf(stderr
, " but not started\n");
2020 if (c
->runstop
<= 0 &&
2021 (working
+ preexist
+ expansion
) <
2022 content
->array
.working_disks
) {
2023 if (c
->export
&& result
)
2024 *result
|= INCR_UNSAFE
;
2025 else if (c
->verbose
>= 0) {
2026 pr_err("%s assembled with %d device%s",
2027 chosen_name
, preexist
+ working
,
2028 preexist
+ working
== 1 ? "":"s");
2030 fprintf(stderr
, " (%d new)", working
);
2031 fprintf(stderr
, " but not safe to start\n");
2037 if (start_reshape
) {
2038 int spare
= content
->array
.raid_disks
+ expansion
;
2039 if (restore_backup(st
, content
,
2041 spare
, &c
->backup_file
, c
->verbose
) == 1)
2044 err
= sysfs_set_str(content
, NULL
,
2045 "array_state", "readonly");
2049 if (st
->ss
->external
) {
2050 if (!mdmon_running(st
->container_devnm
))
2051 start_mdmon(st
->container_devnm
);
2052 ping_monitor(st
->container_devnm
);
2053 if (mdmon_running(st
->container_devnm
) &&
2054 st
->update_tail
== NULL
)
2055 st
->update_tail
= &st
->updates
;
2058 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
2059 0, c
->freeze_reshape
);
2060 } else switch(content
->array
.level
) {
2062 case LEVEL_MULTIPATH
:
2064 err
= sysfs_set_str(content
, NULL
, "array_state",
2065 c
->readonly
? "readonly" : "active");
2068 err
= sysfs_set_str(content
, NULL
, "array_state",
2070 /* start mdmon if needed. */
2072 if (!mdmon_running(st
->container_devnm
))
2073 start_mdmon(st
->container_devnm
);
2074 ping_monitor(st
->container_devnm
);
2079 sysfs_set_safemode(content
, content
->safe_mode_delay
);
2081 /* Block subarray here if it is not reshaped now
2082 * It has be blocked a little later to allow mdmon to switch in
2085 if (st
->ss
->external
&& content
->recovery_blocked
&&
2087 block_subarray(content
);
2089 if (c
->export
&& result
) {
2093 *result
|= INCR_YES
;
2094 } else if (c
->verbose
>= 0) {
2096 pr_err("array %s now has %d device%s",
2097 chosen_name
, working
+ preexist
,
2098 working
+ preexist
== 1 ? "":"s");
2100 pr_err("Started %s with %d device%s",
2101 chosen_name
, working
+ preexist
,
2102 working
+ preexist
== 1 ? "":"s");
2104 fprintf(stderr
, " (%d new)", working
);
2106 fprintf(stderr
, " ( + %d for expansion)",
2108 fprintf(stderr
, "\n");
2111 wait_for(chosen_name
, mdfd
);
2113 /* FIXME should have an O_EXCL and wait for read-auto */