2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 static int name_matches(char *found
, char *required
, char *homehost
, int require_homehost
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
36 if (strcmp(found
, required
)==0)
38 sep
= strchr(found
, ':');
42 if (strncmp(found
, "any:", 4) == 0 ||
43 (homehost
&& strcmp(homehost
, "any") == 0) ||
45 (homehost
&& strlen(homehost
) == l
&&
46 strncmp(found
, homehost
, l
) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep
+1, required
) == 0)
54 static int is_member_busy(char *metadata_version
)
56 /* check if the given member array is active */
57 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
58 struct mdstat_ent
*ent
;
61 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
62 if (ent
->metadata_version
== NULL
)
64 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
66 if (!is_subarray(&ent
->metadata_version
[9]))
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
79 static int ident_matches(struct mddev_ident
*ident
,
80 struct mdinfo
*content
,
81 struct supertype
*tst
,
82 char *homehost
, int require_homehost
,
83 char *update
, char *devname
)
86 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
87 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0 &&
88 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
90 pr_err("%s has wrong uuid.\n", devname
);
93 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
94 name_matches(content
->name
, ident
->name
, homehost
, require_homehost
)==0) {
96 pr_err("%s has wrong name.\n", devname
);
99 if (ident
->super_minor
!= UnSet
&&
100 ident
->super_minor
!= content
->array
.md_minor
) {
102 pr_err("%s has wrong super-minor.\n",
106 if (ident
->level
!= UnSet
&&
107 ident
->level
!= content
->array
.level
) {
109 pr_err("%s has wrong raid level.\n",
113 if (ident
->raid_disks
!= UnSet
&&
114 content
->array
.raid_disks
!= 0 && /* metadata doesn't know how many to expect */
115 ident
->raid_disks
!= content
->array
.raid_disks
) {
117 pr_err("%s requires wrong number of drives.\n",
121 if (ident
->member
&& ident
->member
[0]) {
122 /* content->text_version must match */
123 char *s
= strchr(content
->text_version
+1, '/');
126 pr_err("%s is not a container and one is required.\n",
129 } else if (strcmp(ident
->member
, s
+1) != 0) {
131 pr_err("skipping wrong member %s is %s\n",
132 content
->text_version
, devname
);
139 static int select_devices(struct mddev_dev
*devlist
,
140 struct mddev_ident
*ident
,
141 struct supertype
**stp
,
142 struct mdinfo
**contentp
,
144 int inargv
, int auto_assem
)
146 struct mddev_dev
*tmpdev
;
148 struct supertype
*st
= *stp
;
149 struct mdinfo
*content
= NULL
;
150 int report_mismatch
= ((inargv
&& c
->verbose
>= 0) || c
->verbose
> 0);
151 struct domainlist
*domains
= NULL
;
153 tmpdev
= devlist
; num_devs
= 0;
159 tmpdev
->disposition
= 0;
160 tmpdev
= tmpdev
->next
;
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
167 for (tmpdev
= devlist
;
169 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
170 char *devname
= tmpdev
->devname
;
173 struct supertype
*tst
;
174 struct dev_policy
*pol
= NULL
;
175 int found_container
= 0;
177 if (tmpdev
->used
> 1)
180 if (ident
->container
) {
181 if (ident
->container
[0] == '/' &&
182 !same_dev(ident
->container
, devname
)) {
184 pr_err("%s is not the container required (%s)\n",
185 devname
, ident
->container
);
188 } else if (ident
->devices
&&
189 !match_oneof(ident
->devices
, devname
)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
195 pr_err("%s is not one of %s\n", devname
, ident
->devices
);
201 dfd
= dev_open(devname
, O_RDONLY
);
204 pr_err("cannot open device %s: %s\n",
205 devname
, strerror(errno
));
207 } else if (fstat(dfd
, &stb
)< 0) {
209 pr_err("fstat failed for %s: %s\n",
210 devname
, strerror(errno
));
212 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
213 pr_err("%s is not a block device.\n",
216 } else if (must_be_container(dfd
)) {
218 /* already found some components, this cannot
222 pr_err("%s is a container, but we are looking for components\n",
225 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
226 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
228 pr_err("not a recognisable container: %s\n",
232 } else if (!tst
->ss
->load_container
233 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
235 pr_err("no correct container type: %s\n",
238 } else if (auto_assem
&&
239 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
240 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
242 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
243 devname
, tst
->ss
->name
);
248 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
250 pr_err("no recogniseable superblock on %s\n",
253 } else if ((tst
->ignore_hw_compat
= 0),
254 tst
->ss
->load_super(tst
, dfd
,
255 report_mismatch
? devname
: NULL
)) {
257 pr_err("no RAID superblock on %s\n",
260 } else if (tst
->ss
->compare_super
== NULL
) {
262 pr_err("Cannot assemble %s metadata on %s\n",
263 tst
->ss
->name
, devname
);
265 } else if (auto_assem
&& st
== NULL
&&
266 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
267 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
269 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
270 devname
, tst
->ss
->name
);
274 if (dfd
>= 0) close(dfd
);
275 if (tmpdev
->used
== 2) {
276 if (auto_assem
|| !inargv
)
277 /* Ignore unrecognised devices during auto-assembly */
279 if (ident
->uuid_set
|| ident
->name
[0] ||
280 ident
->super_minor
!= UnSet
)
281 /* Ignore unrecognised device if looking for
285 pr_err("%s has no superblock - assembly aborted\n",
288 st
->ss
->free_super(st
);
289 dev_policy_free(pol
);
290 domain_free(domains
);
294 if (found_container
) {
295 /* tmpdev is a container. We need to be either
296 * looking for a member, or auto-assembling
298 /* should be safe to try an exclusive open now, we
299 * have rejected anything that some other mdadm might
302 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
305 pr_err("%s is busy - skipping\n", devname
);
310 if (ident
->container
&& ident
->container
[0] != '/') {
315 tst
->ss
->getinfo_super(tst
, content
, NULL
);
317 if (!parse_uuid(ident
->container
, uuid
) ||
318 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
320 pr_err("%s has wrong UUID to be required container\n",
325 /* It is worth looking inside this container.
328 pr_err("looking in container %s\n",
331 for (content
= tst
->ss
->container_content(tst
, NULL
);
333 content
= content
->next
) {
335 if (!ident_matches(ident
, content
, tst
,
336 c
->homehost
, c
->require_homehost
,
338 report_mismatch
? devname
: NULL
))
339 /* message already printed */;
340 else if (is_member_busy(content
->text_version
)) {
342 pr_err("member %s in %s is already assembled\n",
343 content
->text_version
,
345 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
346 /* do not assemble arrays with unsupported configurations */
347 pr_err("Cannot activate member %s in %s.\n",
348 content
->text_version
,
355 goto loop
; /* empty container */
358 st
= tst
; tst
= NULL
;
359 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
360 pr_err("%s is a container, but is not only device given: confused and aborting\n",
362 st
->ss
->free_super(st
);
363 dev_policy_free(pol
);
364 domain_free(domains
);
368 pr_err("found match on member %s in %s\n",
369 content
->text_version
, devname
);
371 /* make sure we finished the loop */
376 tst
->ss
->getinfo_super(tst
, content
, NULL
);
378 if (!ident_matches(ident
, content
, tst
,
379 c
->homehost
, c
->require_homehost
,
381 report_mismatch
? devname
: NULL
))
385 /* Never auto-assemble things that conflict
386 * with mdadm.conf in some way
388 struct mddev_ident
*match
;
391 match
= conf_match(tst
, content
, devname
,
392 report_mismatch
? c
->verbose
: -1,
394 if (!match
&& rv
== 2)
396 if (match
&& match
->devname
&&
397 strcasecmp(match
->devname
, "<ignore>") == 0) {
399 pr_err("%s is a member of an explicitly ignored array\n",
403 if (match
&& !ident_matches(match
, content
, tst
,
404 c
->homehost
, c
->require_homehost
,
406 report_mismatch
? devname
: NULL
))
407 /* Array exists in mdadm.conf but some
408 * details don't match, so reject it
413 /* should be safe to try an exclusive open now, we
414 * have rejected anything that some other mdadm might
417 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
420 pr_err("%s is busy - skipping\n", devname
);
427 if (st
->minor_version
== -1)
428 st
->minor_version
= tst
->minor_version
;
430 if (memcmp(content
->uuid
, uuid_zero
,
431 sizeof(int[4])) == 0) {
432 /* this is a floating spare. It cannot define
433 * an array unless there are no more arrays of
434 * this type to be found. It can be included
435 * in an array of this type though.
441 if (st
->ss
!= tst
->ss
||
442 st
->minor_version
!= tst
->minor_version
||
443 st
->ss
->compare_super(st
, tst
) != 0) {
444 /* Some mismatch. If exactly one array matches this host,
445 * we can resolve on that one.
446 * Or, if we are auto assembling, we just ignore the second
452 int first
= st
->ss
->match_home(st
, c
->homehost
);
453 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
455 (first
== 1 || last
== 1)) {
456 /* We can do something */
457 if (first
) {/* just ignore this one */
459 pr_err("%s misses out due to wrong homehost\n",
462 } else { /* reject all those sofar */
463 struct mddev_dev
*td
;
465 pr_err("%s overrides previous devices due to good homehost\n",
467 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
475 pr_err("superblock on %s doesn't match others - assembly aborted\n",
477 tst
->ss
->free_super(tst
);
478 st
->ss
->free_super(st
);
479 dev_policy_free(pol
);
480 domain_free(domains
);
486 /* Collect domain information from members only */
487 if (tmpdev
&& tmpdev
->used
== 1) {
489 pol
= devid_policy(stb
.st_rdev
);
490 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
492 dev_policy_free(pol
);
495 tst
->ss
->free_super(tst
);
498 /* Check if we found some imsm spares but no members */
501 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
503 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
504 if (tmpdev
->used
!= 3)
510 /* we need sb from one of the spares */
511 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
513 st
->ss
->load_super(st
, dfd
, NULL
))
520 /* Now reject spares that don't match domains of identified members */
521 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
523 if (tmpdev
->used
!= 3)
525 if (stat(tmpdev
->devname
, &stb
)< 0) {
526 pr_err("fstat failed for %s: %s\n",
527 tmpdev
->devname
, strerror(errno
));
530 struct dev_policy
*pol
= devid_policy(stb
.st_rdev
);
531 int dt
= domain_test(domains
, pol
, NULL
);
532 if (inargv
&& dt
!= 0)
533 /* take this spare as domains match
534 * if there are any */
536 else if (!inargv
&& dt
== 1)
537 /* device wasn't explicitly listed, so need
538 * explicit domain match - which we have */
541 /* if domains don't match mark as unused */
543 dev_policy_free(pol
);
546 domain_free(domains
);
548 if (st
&& st
->sb
&& content
== *contentp
)
549 st
->ss
->getinfo_super(st
, content
, NULL
);
557 int uptodate
; /* set once we decide that this device is as
558 * recent as everything else in the array.
560 int included
; /* set if the device is already in the array
561 * due to a previous '-I'
566 static int load_devices(struct devs
*devices
, char *devmap
,
567 struct mddev_ident
*ident
, struct supertype
**stp
,
568 struct mddev_dev
*devlist
, struct context
*c
,
569 struct mdinfo
*content
,
570 int mdfd
, char *mddev
,
571 int *most_recentp
, int *bestcntp
, int **bestp
,
574 struct mddev_dev
*tmpdev
;
580 int most_recent
= -1;
583 struct supertype
*st
= *stp
;
585 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
586 char *devname
= tmpdev
->devname
;
588 struct supertype
*tst
;
592 if (tmpdev
->used
!= 1)
594 /* looks like a good enough match to update the super block if needed */
597 /* prepare useful information in info structures */
602 if (strcmp(c
->update
, "uuid")==0 &&
605 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
606 read(rfd
, ident
->uuid
, 16) != 16) {
607 *(__u32
*)(ident
->uuid
) = random();
608 *(__u32
*)(ident
->uuid
+1) = random();
609 *(__u32
*)(ident
->uuid
+2) = random();
610 *(__u32
*)(ident
->uuid
+3) = random();
612 if (rfd
>= 0) close(rfd
);
614 dfd
= dev_open(devname
,
615 tmpdev
->disposition
== 'I'
616 ? O_RDWR
: (O_RDWR
|O_EXCL
));
619 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
620 pr_err("cannot re-read metadata from %s - aborting\n",
627 tst
->ss
->free_super(tst
);
632 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
634 memcpy(content
->uuid
, ident
->uuid
, 16);
635 strcpy(content
->name
, ident
->name
);
636 content
->array
.md_minor
= minor(stb2
.st_rdev
);
638 if (strcmp(c
->update
, "byteorder") == 0)
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");
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
;
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"
785 || (devices
[best
[i
]].i
.events
786 < devices
[devcnt
].i
.events
))
791 if (most_recent
>= 0)
792 *most_recentp
= most_recent
;
799 static int force_array(struct mdinfo
*content
,
800 struct devs
*devices
,
801 int *best
, int bestcnt
, char *avail
,
803 struct supertype
*st
,
807 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
808 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,
817 /* Choose the newest best drive which is
818 * not up-to-date, update the superblock
822 struct supertype
*tst
;
823 unsigned long long current_events
;
824 int chosen_drive
= -1;
828 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
833 if (devices
[j
].uptodate
)
835 if (devices
[j
].i
.recovery_start
!= MaxSector
) {
837 if (!devices
[j
].i
.reshape_active
||
838 devices
[j
].i
.delta_disks
<= 0)
840 /* When increasing number of devices, an
841 * added device also appears to be
842 * recovering. It is safe to include it
843 * as long as it won't be a source of
845 * For now, just allow for last data
846 * devices in RAID4 or last devices in RAID4/5/6.
848 delta
= devices
[j
].i
.delta_disks
;
849 if (devices
[j
].i
.array
.level
>= 4 &&
850 devices
[j
].i
.array
.level
<= 6 &&
851 i
/2 >= content
->array
.raid_disks
- delta
)
853 else if (devices
[j
].i
.array
.level
== 4 &&
854 i
/2 >= content
->array
.raid_disks
- delta
- 1)
859 if (chosen_drive
< 0 ||
861 > devices
[chosen_drive
].i
.events
)
864 if (chosen_drive
< 0)
866 current_events
= devices
[chosen_drive
].i
.events
;
869 pr_err("forcing event count in %s(%d) from %d upto %d\n",
870 devices
[chosen_drive
].devname
,
871 devices
[chosen_drive
].i
.disk
.raid_disk
,
872 (int)(devices
[chosen_drive
].i
.events
),
873 (int)(devices
[most_recent
].i
.events
));
874 fd
= dev_open(devices
[chosen_drive
].devname
,
875 devices
[chosen_drive
].included
? O_RDWR
878 pr_err("Couldn't open %s for write - not updating\n",
879 devices
[chosen_drive
].devname
);
880 devices
[chosen_drive
].i
.events
= 0;
884 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
886 pr_err("RAID superblock disappeared from %s - not updating.\n",
887 devices
[chosen_drive
].devname
);
888 devices
[chosen_drive
].i
.events
= 0;
891 content
->events
= devices
[most_recent
].i
.events
;
892 tst
->ss
->update_super(tst
, content
, "force-one",
893 devices
[chosen_drive
].devname
, c
->verbose
,
896 if (tst
->ss
->store_super(tst
, fd
)) {
898 pr_err("Could not re-write superblock on %s\n",
899 devices
[chosen_drive
].devname
);
900 devices
[chosen_drive
].i
.events
= 0;
901 tst
->ss
->free_super(tst
);
905 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
906 devices
[chosen_drive
].uptodate
= 1;
907 avail
[chosen_drive
] = 1;
909 tst
->ss
->free_super(tst
);
911 /* If there are any other drives of the same vintage,
912 * add them in as well. We can't lose and we might gain
915 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
919 !devices
[j
].uptodate
&&
920 devices
[j
].i
.recovery_start
== MaxSector
&&
921 devices
[j
].i
.events
== current_events
) {
930 static int start_array(int mdfd
,
932 struct mdinfo
*content
,
933 struct supertype
*st
,
934 struct mddev_ident
*ident
,
935 int *best
, int bestcnt
,
937 struct devs
*devices
,
939 unsigned int sparecnt
,
940 unsigned int rebuilding_cnt
,
942 int clean
, char *avail
,
943 int start_partial_ok
,
950 unsigned int req_cnt
;
952 rv
= set_array_info(mdfd
, st
, content
);
954 pr_err("failed to set array info for %s: %s\n",
955 mddev
, strerror(errno
));
958 if (ident
->bitmap_fd
>= 0) {
959 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
960 pr_err("SET_BITMAP_FILE failed.\n");
963 } else if (ident
->bitmap_file
) {
964 /* From config file */
965 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
967 pr_err("Could not open bitmap file %s\n",
971 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
972 pr_err("Failed to set bitmapfile for %s\n", mddev
);
979 /* First, add the raid disks, but add the chosen one last */
980 for (i
=0; i
<= bestcnt
; i
++) {
984 if (j
== chosen_drive
)
989 if (j
>= 0 && !devices
[j
].included
) {
990 int dfd
= dev_open(devices
[j
].devname
,
993 remove_partitions(dfd
);
996 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
999 pr_err("failed to add %s to %s: %s\n",
1003 if (i
< content
->array
.raid_disks
* 2
1008 } else if (c
->verbose
> 0)
1009 pr_err("added %s to %s as %d%s%s\n",
1010 devices
[j
].devname
, mddev
,
1011 devices
[j
].i
.disk
.raid_disk
,
1012 devices
[j
].uptodate
?"":
1013 " (possibly out of date)",
1014 (devices
[j
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))?" replacement":"");
1015 } else if (j
>= 0) {
1017 pr_err("%s is already in %s as %d\n",
1018 devices
[j
].devname
, mddev
,
1019 devices
[j
].i
.disk
.raid_disk
);
1020 } else if (c
->verbose
> 0 && i
< content
->array
.raid_disks
*2
1022 pr_err("no uptodate device for slot %d of %s\n",
1026 if (content
->array
.level
== LEVEL_CONTAINER
) {
1027 if (c
->verbose
>= 0) {
1028 pr_err("Container %s has been assembled with %d drive%s",
1029 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1030 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1031 fprintf(stderr
, " (out of %d)",
1032 content
->array
.raid_disks
);
1033 fprintf(stderr
, "\n");
1036 if (st
->ss
->validate_container
) {
1037 struct mdinfo
*devices_list
;
1038 struct mdinfo
*info_devices
= xmalloc(sizeof(struct mdinfo
)*(okcnt
+sparecnt
));
1040 devices_list
= NULL
;
1041 for (count
= 0; count
< okcnt
+sparecnt
; count
++) {
1042 info_devices
[count
] = devices
[count
].i
;
1043 info_devices
[count
].next
= devices_list
;
1044 devices_list
= &info_devices
[count
];
1046 if (st
->ss
->validate_container(devices_list
))
1047 pr_err("Mismatch detected!\n");
1051 st
->ss
->free_super(st
);
1052 sysfs_uevent(content
, "change");
1053 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1054 /* Was partial, is still partial, so signal an error
1055 * to ensure we don't retry */
1060 /* Get number of in-sync devices according to the superblock.
1061 * We must have this number to start the array without -s or -R
1063 req_cnt
= content
->array
.working_disks
;
1065 if (c
->runstop
== 1 ||
1067 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1068 content
->array
.layout
, clean
, avail
) &&
1069 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1071 /* This array is good-to-go.
1072 * If a reshape is in progress then we might need to
1073 * continue monitoring it. In that case we start
1074 * it read-only and let the grow code make it writable.
1078 if (content
->reshape_active
&&
1079 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1080 content
->delta_disks
<= 0) {
1081 if (!c
->backup_file
) {
1082 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1084 pr_err("Please provided one with \"--backup-file=...\"\n");
1086 strcmp(c
->update
, "revert-reshape") == 0)
1087 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1090 rv
= sysfs_set_str(content
, NULL
,
1091 "array_state", "readonly");
1093 rv
= Grow_continue(mdfd
, st
, content
,
1096 } else if (c
->readonly
&&
1097 sysfs_attribute_available(
1098 content
, NULL
, "array_state")) {
1099 rv
= sysfs_set_str(content
, NULL
,
1100 "array_state", "readonly");
1103 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1104 reopen_mddev(mdfd
); /* drop O_EXCL */
1106 if (c
->verbose
>= 0) {
1107 pr_err("%s has been started with %d drive%s",
1108 mddev
, okcnt
, okcnt
==1?"":"s");
1109 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1110 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1112 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1114 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1115 fprintf(stderr
, ".\n");
1117 if (content
->reshape_active
&&
1118 content
->array
.level
>= 4 &&
1119 content
->array
.level
<= 6) {
1120 /* might need to increase the size
1121 * of the stripe cache - default is 256
1123 int chunk_size
= content
->array
.chunk_size
;
1124 if (content
->reshape_active
&&
1125 content
->new_chunk
> chunk_size
)
1126 chunk_size
= content
->new_chunk
;
1127 if (256 < 4 * ((chunk_size
+4065)/4096)) {
1128 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1130 sysfs_set_num(sra
, NULL
,
1131 "stripe_cache_size",
1132 (4 * chunk_size
/ 4096) + 1);
1136 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1137 /* If any devices did not get added
1138 * because the kernel rejected them based
1139 * on event count, try adding them
1140 * again providing the action policy is
1141 * 're-add' or greater. The bitmap
1142 * might allow them to be included, or
1143 * they will become spares.
1145 for (i
= 0; i
< bestcnt
; i
++) {
1147 if (j
>= 0 && !devices
[j
].uptodate
) {
1148 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1150 rv
= add_disk(mdfd
, st
, content
,
1152 if (rv
== 0 && c
->verbose
>= 0)
1153 pr_err("%s has been re-added.\n",
1154 devices
[j
].devname
);
1158 if (content
->array
.level
== 6 &&
1159 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1161 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1163 sysfs_set_str(sra
, NULL
,
1164 "sync_action", "repair");
1169 pr_err("failed to RUN_ARRAY %s: %s\n",
1170 mddev
, strerror(errno
));
1172 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1173 content
->array
.layout
, 1, avail
))
1174 pr_err("Not enough devices to start the array.\n");
1175 else if (!enough(content
->array
.level
,
1176 content
->array
.raid_disks
,
1177 content
->array
.layout
, clean
,
1179 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1183 if (c
->runstop
== -1) {
1184 pr_err("%s assembled from %d drive%s",
1185 mddev
, okcnt
, okcnt
==1?"":"s");
1186 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1187 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1188 fprintf(stderr
, ", but not started.\n");
1191 if (c
->verbose
>= -1) {
1192 pr_err("%s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1194 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1196 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1197 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1198 content
->array
.layout
, 1, avail
))
1199 fprintf(stderr
, " - not enough to start the array.\n");
1200 else if (!enough(content
->array
.level
,
1201 content
->array
.raid_disks
,
1202 content
->array
.layout
, clean
,
1204 fprintf(stderr
, " - not enough to start the array while not clean - consider --force.\n");
1206 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1207 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1209 fprintf(stderr
, " - need %d to start", req_cnt
);
1210 fprintf(stderr
, " (use --run to insist).\n");
1216 int Assemble(struct supertype
*st
, char *mddev
,
1217 struct mddev_ident
*ident
,
1218 struct mddev_dev
*devlist
,
1222 * The task of Assemble is to find a collection of
1223 * devices that should (according to their superblocks)
1224 * form an array, and to give this collection to the MD driver.
1225 * In Linux-2.4 and later, this involves submitting a
1226 * SET_ARRAY_INFO ioctl with no arg - to prepare
1227 * the array - and then submit a number of
1228 * ADD_NEW_DISK ioctls to add disks into
1229 * the array. Finally RUN_ARRAY might
1230 * be submitted to start the array.
1232 * Much of the work of Assemble is in finding and/or
1233 * checking the disks to make sure they look right.
1235 * If mddev is not set, then scan must be set and we
1236 * read through the config file for dev+uuid mapping
1237 * We recurse, setting mddev, for each device that
1239 * - has a valid uuid (or any uuid if !uuidset)
1241 * If mddev is set, we try to determine state of md.
1242 * check version - must be at least 0.90.0
1243 * check kernel version. must be at least 2.4.
1244 * If not, we can possibly fall back on START_ARRAY
1245 * Try to GET_ARRAY_INFO.
1246 * If possible, give up
1247 * If not, try to STOP_ARRAY just to make sure
1249 * If !uuidset and scan, look in conf-file for uuid
1250 * If not found, give up
1251 * If !devlist and scan and uuidset, get list of devs from conf-file
1254 * Check superblock - discard if bad
1255 * Check uuid (set if we don't have one) - discard if no match
1256 * Check superblock similarity if we have a superblock - discard if different
1257 * Record events, devicenum
1258 * This should give us a list of devices for the array
1259 * We should collect the most recent event number
1261 * Count disks with recent enough event count
1262 * While force && !enough disks
1263 * Choose newest rejected disks, update event count
1264 * mark clean and rewrite superblock
1267 * foreach device with recent events : ADD_NEW_DISK
1268 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1270 * Check the device numbers in superblock are right
1271 * update superblock if any changes
1278 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1279 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1280 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1281 struct devs
*devices
;
1283 int *best
= NULL
; /* indexed by raid_disk */
1286 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
;
1289 int most_recent
= 0;
1293 int start_partial_ok
= (c
->runstop
>= 0) &&
1294 (c
->force
|| devlist
==NULL
|| auto_assem
);
1296 struct mddev_dev
*tmpdev
;
1298 struct mdinfo
*content
= NULL
;
1299 struct mdinfo
*pre_exist
= NULL
;
1302 char chosen_name
[1024];
1303 struct map_ent
*map
= NULL
;
1307 * If any subdevs are listed, then any that don't
1308 * match ident are discarded. Remainder must all match and
1310 * If no subdevs, then we scan all devices in the config file, but
1311 * there must be something in the identity
1315 ident
->uuid_set
== 0 &&
1316 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1317 ident
->name
[0] == 0 &&
1318 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1319 ident
->devices
== NULL
) {
1320 pr_err("No identity information available for %s - cannot assemble.\n",
1321 mddev
? mddev
: "further assembly");
1325 if (devlist
== NULL
)
1326 devlist
= conf_get_devs();
1331 /* We come back here when doing auto-assembly and attempting some
1332 * set of devices failed. Those are now marked as ->used==2 and
1333 * we ignore them and try again
1335 if (!st
&& ident
->st
)
1338 pr_err("looking for devices for %s\n",
1339 mddev
? mddev
: "further assembly");
1343 st
->ignore_hw_compat
= 1;
1344 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1345 inargv
, auto_assem
);
1349 if (!st
|| !st
->sb
|| !content
)
1352 /* We have a full set of devices - we now need to find the
1354 * However there is a risk that we are racing with "mdadm -I"
1355 * and the array is already partially assembled - we will have
1356 * rejected any devices already in this address.
1357 * So we take a lock on the map file - to prevent further races -
1358 * and look for the uuid in there. If found and the array is
1359 * active, we abort. If found and the array is not active
1360 * we commit to that md device and add all the contained devices
1361 * to our list. We flag them so that we don't try to re-add,
1362 * but can remove if they turn out to not be wanted.
1365 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1366 if (c
->update
&& strcmp(c
->update
,"uuid") == 0)
1369 mp
= map_by_uuid(&map
, content
->uuid
);
1372 /* array already exists. */
1373 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1374 if (pre_exist
->array
.level
!= UnSet
) {
1375 pr_err("Found some drive for an array that is already active: %s\n",
1377 pr_err("giving up.\n");
1380 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1381 /* We want to add this device to our list,
1382 * but it could already be there if "mdadm -I"
1383 * started *after* we checked for O_EXCL.
1384 * If we add it to the top of the list
1385 * it will be preferred over later copies.
1387 struct mddev_dev
*newdev
;
1388 char *devname
= map_dev(dv
->disk
.major
,
1393 newdev
= xmalloc(sizeof(*newdev
));
1394 newdev
->devname
= devname
;
1395 newdev
->disposition
= 'I';
1397 newdev
->next
= devlist
;
1401 strcpy(chosen_name
, mp
->path
);
1402 if (c
->verbose
> 0 || mddev
== NULL
||
1403 strcmp(mddev
, chosen_name
) != 0)
1404 pr_err("Merging with already-assembled %s\n",
1406 mdfd
= open_dev_excl(mp
->devnm
);
1408 int trustworthy
= FOREIGN
;
1409 name
= content
->name
;
1410 switch (st
->ss
->match_home(st
, c
->homehost
)
1411 ?: st
->ss
->match_home(st
, "any")) {
1413 trustworthy
= LOCAL
;
1414 name
= strchr(content
->name
, ':');
1418 name
= content
->name
;
1422 /* If the array is listed in mdadm.conf or on
1423 * command line, then we trust the name
1424 * even if the array doesn't look local
1426 trustworthy
= LOCAL
;
1429 content
->array
.level
== LEVEL_CONTAINER
) {
1430 name
= content
->text_version
;
1431 trustworthy
= METADATA
;
1434 if (name
[0] && trustworthy
!= LOCAL
&&
1435 ! c
->require_homehost
&&
1436 conf_name_is_free(name
))
1437 trustworthy
= LOCAL
;
1439 if (trustworthy
== LOCAL
&&
1441 /* Ignore 'host:' prefix of name */
1442 name
= strchr(name
, ':')+1;
1444 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1448 st
->ss
->free_super(st
);
1453 mddev
= chosen_name
;
1454 if (get_linux_version() < 2004000 ||
1455 md_get_version(mdfd
) < 9000) {
1456 pr_err("Assemble requires Linux 2.4 or later, and\n"
1457 " md driver version 0.90.0 or later.\n"
1458 " Upgrade your kernel or try --build\n");
1462 if (pre_exist
== NULL
) {
1463 if (mddev_busy(fd2devnm(mdfd
))) {
1464 pr_err("%s already active, cannot restart it!\n",
1466 for (tmpdev
= devlist
;
1467 tmpdev
&& tmpdev
->used
!= 1;
1468 tmpdev
= tmpdev
->next
)
1470 if (tmpdev
&& auto_assem
)
1471 pr_err("%s needed for %s...\n",
1472 mddev
, tmpdev
->devname
);
1475 st
->ss
->free_super(st
);
1480 /* just incase it was started but has no content */
1481 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1485 if (content
!= &info
) {
1486 /* This is a member of a container. Try starting the array. */
1488 err
= assemble_container_content(st
, mdfd
, content
, c
,
1494 /* Ok, no bad inconsistancy, we can try updating etc */
1495 devices
= xcalloc(num_devs
, sizeof(*devices
));
1496 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1497 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1498 c
, content
, mdfd
, mddev
,
1499 &most_recent
, &bestcnt
, &best
, inargv
);
1504 pr_err("no devices found for %s\n",
1507 st
->ss
->free_super(st
);
1514 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1515 st
->minor_version
= 90;
1517 st
->ss
->getinfo_super(st
, content
, NULL
);
1518 clean
= content
->array
.state
& 1;
1520 /* now we have some devices that might be suitable.
1523 avail
= xcalloc(content
->array
.raid_disks
, 1);
1528 for (i
=0; i
< bestcnt
; i
++) {
1530 int event_margin
= 1; /* always allow a difference of '1'
1531 * like the kernel does
1533 if (j
< 0) continue;
1534 /* note: we ignore error flags in multipath arrays
1535 * as they don't make sense
1537 if (content
->array
.level
!= LEVEL_MULTIPATH
)
1538 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1539 if (!(devices
[j
].i
.disk
.state
1540 & (1<<MD_DISK_FAULTY
))) {
1541 devices
[j
].uptodate
= 1;
1546 /* If this device thinks that 'most_recent' has failed, then
1547 * we must reject this device.
1549 if (j
!= most_recent
&& !c
->force
&&
1550 content
->array
.raid_disks
> 0 &&
1551 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1552 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1553 if (c
->verbose
> -1)
1554 pr_err("ignoring %s as it reports %s as failed\n",
1555 devices
[j
].devname
, devices
[most_recent
].devname
);
1559 /* Require event counter to be same as, or just less than,
1560 * most recent. If it is bigger, it must be a stray spare and
1561 * should be ignored.
1563 if (devices
[j
].i
.events
+event_margin
>=
1564 devices
[most_recent
].i
.events
&&
1565 devices
[j
].i
.events
<=
1566 devices
[most_recent
].i
.events
1568 devices
[j
].uptodate
= 1;
1569 if (i
< content
->array
.raid_disks
* 2) {
1570 if (devices
[j
].i
.recovery_start
== MaxSector
||
1571 (content
->reshape_active
&&
1572 i
>= content
->array
.raid_disks
- content
->delta_disks
)) {
1586 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1587 avail
, most_recent
, st
, c
);
1592 /* Now we want to look at the superblock which the kernel will base things on
1593 * and compare the devices that we think are working with the devices that the
1594 * superblock thinks are working.
1595 * If there are differences and --force is given, then update this chosen
1599 st
->ss
->free_super(st
);
1600 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1606 if (!devices
[j
].uptodate
)
1608 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1611 if ((fd
=dev_open(devices
[j
].devname
,
1612 devices
[j
].included
? O_RDONLY
1613 : (O_RDONLY
|O_EXCL
)))< 0) {
1614 pr_err("Cannot open %s: %s\n",
1615 devices
[j
].devname
, strerror(errno
));
1620 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1622 pr_err("RAID superblock has disappeared from %s\n",
1623 devices
[j
].devname
);
1630 if (st
->sb
== NULL
) {
1631 pr_err("No suitable drives found for %s\n", mddev
);
1636 st
->ss
->getinfo_super(st
, content
, NULL
);
1638 sysfs_init(content
, mdfd
, NULL
);
1640 for (i
=0; i
<bestcnt
; i
++) {
1642 unsigned int desired_state
;
1644 if (i
>= content
->array
.raid_disks
* 2)
1647 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1649 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1653 if (!devices
[j
].uptodate
)
1656 devices
[j
].i
.disk
.state
= desired_state
;
1657 if (!(devices
[j
].i
.array
.state
& 1))
1660 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1661 c
->verbose
, 0, NULL
)) {
1663 if (c
->verbose
>= 0)
1664 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1665 j
, mddev
, devices
[j
].devname
);
1668 if (c
->verbose
>= -1)
1669 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1670 i
, mddev
, devices
[j
].devname
);
1674 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1675 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1680 if (c
->force
&& !clean
&&
1681 !enough(content
->array
.level
, content
->array
.raid_disks
,
1682 content
->array
.layout
, clean
,
1684 change
+= st
->ss
->update_super(st
, content
, "force-array",
1685 devices
[chosen_drive
].devname
, c
->verbose
,
1693 fd
= dev_open(devices
[chosen_drive
].devname
,
1694 devices
[chosen_drive
].included
?
1695 O_RDWR
: (O_RDWR
|O_EXCL
));
1697 pr_err("Could not open %s for write - cannot Assemble array.\n",
1698 devices
[chosen_drive
].devname
);
1703 if (st
->ss
->store_super(st
, fd
)) {
1705 pr_err("Could not re-write superblock on %s\n",
1706 devices
[chosen_drive
].devname
);
1711 if (c
->verbose
>= 0)
1712 pr_err("Marking array %s as 'clean'\n",
1717 /* If we are in the middle of a reshape we may need to restore saved data
1718 * that was moved aside due to the reshape overwriting live data
1719 * The code of doing this lives in Grow.c
1722 if (content
->reshape_active
&&
1723 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1725 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1727 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1729 if (!c
->backup_file
)
1730 c
->backup_file
= locate_backup(content
->sys_name
);
1731 enable_fds(bestcnt
/2);
1732 for (i
= 0; i
< bestcnt
/2; i
++) {
1735 fdlist
[i
] = dev_open(devices
[j
].devname
,
1737 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1738 if (fdlist
[i
] < 0) {
1739 pr_err("Could not open %s for write - cannot Assemble array.\n",
1740 devices
[j
].devname
);
1748 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1749 err
= st
->ss
->recover_backup(st
, content
);
1751 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1752 c
->backup_file
, c
->verbose
> 0);
1753 if (err
&& c
->invalid_backup
) {
1755 pr_err("continuing without restoring backup\n");
1761 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1765 pr_err("Failed to restore critical section for reshape, sorry.\n");
1766 if (c
->backup_file
== NULL
)
1767 cont_err("Possibly you needed to specify the --backup-file\n");
1775 /* Almost ready to actually *do* something */
1776 /* First, fill in the map, so that udev can find our name
1777 * as soon as we become active.
1779 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1780 content
->array
.major_version
= 1;
1781 content
->array
.minor_version
= 0;
1782 strcpy(content
->text_version
, "1.0");
1785 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1786 content
->uuid
, chosen_name
);
1788 rv
= start_array(mdfd
, mddev
, content
,
1789 st
, ident
, best
, bestcnt
,
1790 chosen_drive
, devices
, okcnt
, sparecnt
,
1793 clean
, avail
, start_partial_ok
,
1796 if (rv
== 1 && !pre_exist
)
1797 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1801 wait_for(chosen_name
, mdfd
);
1805 /* There is a nasty race with 'mdadm --monitor'.
1806 * If it opens this device before we close it,
1807 * it gets an incomplete open on which IO
1808 * doesn't work and the capacity is
1810 * If we reopen (to check for layered devices)
1811 * before --monitor closes, we loose.
1813 * So: wait upto 1 second for there to be
1814 * a non-zero capacity.
1816 while (usecs
< 1000) {
1817 mdfd
= open(mddev
, O_RDONLY
);
1819 unsigned long long size
;
1820 if (get_dev_size(mdfd
, NULL
, &size
) &&
1832 /* '2' means 'OK, but not started yet' */
1833 return rv
== 2 ? 0 : rv
;
1837 int assemble_container_content(struct supertype
*st
, int mdfd
,
1838 struct mdinfo
*content
, struct context
*c
,
1839 char *chosen_name
, int *result
)
1841 struct mdinfo
*dev
, *sra
, *dev2
;
1842 int working
= 0, preexist
= 0;
1844 struct map_ent
*map
= NULL
;
1850 sysfs_init(content
, mdfd
, NULL
);
1852 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
|GET_DEVS
);
1853 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1854 if (content
->array
.major_version
== -1 &&
1855 content
->array
.minor_version
== -2 &&
1857 content
->text_version
[0] == '/')
1858 content
->text_version
[0] = '-';
1859 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1866 /* There are two types of reshape: container wide or sub-array specific
1867 * Check if metadata requests blocking container wide reshapes
1869 start_reshape
= (content
->reshape_active
&&
1870 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1871 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1873 /* Block subarray here if it is under reshape now
1874 * Do not allow for any changes in this array
1876 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1877 block_subarray(content
);
1879 for (dev2
= sra
->devs
; dev2
; dev2
= dev2
->next
) {
1880 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1881 if (dev2
->disk
.major
== dev
->disk
.major
&&
1882 dev2
->disk
.minor
== dev
->disk
.minor
)
1886 /* Don't want this one any more */
1887 if (sysfs_set_str(sra
, dev2
, "slot", "none") < 0 &&
1889 pr_err("Cannot remove old device %s: not updating %s\n", dev2
->sys_name
, sra
->sys_name
);
1893 sysfs_set_str(sra
, dev2
, "state", "remove");
1895 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1896 avail
= xcalloc(content
->array
.raid_disks
, 1);
1897 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1898 if (dev
->disk
.raid_disk
>= 0)
1899 avail
[dev
->disk
.raid_disk
] = 1;
1900 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1901 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1902 content
->reshape_active
)
1906 } else if (errno
== EEXIST
)
1910 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1912 return 1;/* Nothing new, don't try to start */
1914 map_update(&map
, fd2devnm(mdfd
),
1915 content
->text_version
,
1916 content
->uuid
, chosen_name
);
1919 if (enough(content
->array
.level
, content
->array
.raid_disks
,
1920 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
1921 if (c
->export
&& result
)
1923 else if (c
->verbose
>= 0) {
1924 pr_err("%s assembled with %d device%s",
1925 chosen_name
, preexist
+ working
,
1926 preexist
+ working
== 1 ? "":"s");
1928 fprintf(stderr
, " (%d new)", working
);
1929 fprintf(stderr
, " but not started\n");
1936 if (c
->runstop
<= 0 &&
1937 (working
+ preexist
+ expansion
) <
1938 content
->array
.working_disks
) {
1939 if (c
->export
&& result
)
1940 *result
|= INCR_UNSAFE
;
1941 else if (c
->verbose
>= 0) {
1942 pr_err("%s assembled with %d device%s",
1943 chosen_name
, preexist
+ working
,
1944 preexist
+ working
== 1 ? "":"s");
1946 fprintf(stderr
, " (%d new)", working
);
1947 fprintf(stderr
, " but not safe to start\n");
1953 if (start_reshape
) {
1954 int spare
= content
->array
.raid_disks
+ expansion
;
1955 if (restore_backup(st
, content
,
1957 spare
, &c
->backup_file
, c
->verbose
) == 1)
1960 err
= sysfs_set_str(content
, NULL
,
1961 "array_state", "readonly");
1965 if (st
->ss
->external
) {
1966 if (!mdmon_running(st
->container_devnm
))
1967 start_mdmon(st
->container_devnm
);
1968 ping_monitor(st
->container_devnm
);
1969 if (mdmon_running(st
->container_devnm
) &&
1970 st
->update_tail
== NULL
)
1971 st
->update_tail
= &st
->updates
;
1974 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
1975 0, c
->freeze_reshape
);
1976 } else switch(content
->array
.level
) {
1978 case LEVEL_MULTIPATH
:
1980 err
= sysfs_set_str(content
, NULL
, "array_state",
1981 c
->readonly
? "readonly" : "active");
1984 err
= sysfs_set_str(content
, NULL
, "array_state",
1986 /* start mdmon if needed. */
1988 if (!mdmon_running(st
->container_devnm
))
1989 start_mdmon(st
->container_devnm
);
1990 ping_monitor(st
->container_devnm
);
1995 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1997 /* Block subarray here if it is not reshaped now
1998 * It has be blocked a little later to allow mdmon to switch in
2001 if (st
->ss
->external
&& content
->recovery_blocked
&&
2003 block_subarray(content
);
2005 if (c
->export
&& result
) {
2009 *result
|= INCR_YES
;
2010 } else if (c
->verbose
>= 0) {
2012 pr_err("array %s now has %d device%s",
2013 chosen_name
, working
+ preexist
,
2014 working
+ preexist
== 1 ? "":"s");
2016 pr_err("Started %s with %d device%s",
2017 chosen_name
, working
+ preexist
,
2018 working
+ preexist
== 1 ? "":"s");
2020 fprintf(stderr
, " (%d new)", working
);
2022 fprintf(stderr
, " ( + %d for expansion)",
2024 fprintf(stderr
, "\n");
2027 wait_for(chosen_name
, mdfd
);
2029 /* FIXME should have an O_EXCL and wait for read-auto */