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
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
35 strncpy(fnd
, found
, 32);
37 if (strcmp(found
, required
)==0)
40 int l
= strlen(homehost
);
41 if (l
< 32 && fnd
[l
] == ':' &&
42 strcmp(fnd
+l
+1, required
)==0)
48 static int is_member_busy(char *metadata_version
)
50 /* check if the given member array is active */
51 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
52 struct mdstat_ent
*ent
;
55 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
56 if (ent
->metadata_version
== NULL
)
58 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
60 if (!is_subarray(&ent
->metadata_version
[9]))
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
73 static int ident_matches(struct mddev_ident
*ident
,
74 struct mdinfo
*content
,
75 struct supertype
*tst
,
77 char *update
, char *devname
)
80 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
81 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0 &&
82 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
84 pr_err("%s has wrong uuid.\n", devname
);
87 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
88 name_matches(content
->name
, ident
->name
, homehost
)==0) {
90 pr_err("%s has wrong name.\n", devname
);
93 if (ident
->super_minor
!= UnSet
&&
94 ident
->super_minor
!= content
->array
.md_minor
) {
96 pr_err("%s has wrong super-minor.\n",
100 if (ident
->level
!= UnSet
&&
101 ident
->level
!= content
->array
.level
) {
103 pr_err("%s has wrong raid level.\n",
107 if (ident
->raid_disks
!= UnSet
&&
108 content
->array
.raid_disks
!= 0 && /* metadata doesn't know how many to expect */
109 ident
->raid_disks
!= content
->array
.raid_disks
) {
111 pr_err("%s requires wrong number of drives.\n",
115 if (ident
->member
&& ident
->member
[0]) {
116 /* content->text_version must match */
117 char *s
= strchr(content
->text_version
+1, '/');
120 pr_err("%s is not a container and one is required.\n",
123 } else if (strcmp(ident
->member
, s
+1) != 0) {
125 pr_err("skipping wrong member %s is %s\n",
126 content
->text_version
, devname
);
133 static int select_devices(struct mddev_dev
*devlist
,
134 struct mddev_ident
*ident
,
135 struct supertype
**stp
,
136 struct mdinfo
**contentp
,
138 int inargv
, int auto_assem
)
140 struct mddev_dev
*tmpdev
;
142 struct supertype
*st
= *stp
;
143 struct mdinfo
*content
= NULL
;
144 int report_mismatch
= ((inargv
&& c
->verbose
>= 0) || c
->verbose
> 0);
145 struct domainlist
*domains
= NULL
;
147 tmpdev
= devlist
; num_devs
= 0;
153 tmpdev
->disposition
= 0;
154 tmpdev
= tmpdev
->next
;
157 /* first walk the list of devices to find a consistent set
158 * that match the criterea, if that is possible.
159 * We flag the ones we like with 'used'.
161 for (tmpdev
= devlist
;
163 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
164 char *devname
= tmpdev
->devname
;
167 struct supertype
*tst
;
168 struct dev_policy
*pol
= NULL
;
169 int found_container
= 0;
171 if (tmpdev
->used
> 1)
174 if (ident
->container
) {
175 if (ident
->container
[0] == '/' &&
176 !same_dev(ident
->container
, devname
)) {
178 pr_err("%s is not the container required (%s)\n",
179 devname
, ident
->container
);
182 } else if (ident
->devices
&&
183 !match_oneof(ident
->devices
, devname
)) {
184 /* Note that we ignore the "device=" identifier if a
185 * "container=" is given. Checking both is unnecessarily
189 pr_err("%s is not one of %s\n", devname
, ident
->devices
);
195 dfd
= dev_open(devname
, O_RDONLY
);
198 pr_err("cannot open device %s: %s\n",
199 devname
, strerror(errno
));
201 } else if (fstat(dfd
, &stb
)< 0) {
203 pr_err("fstat failed for %s: %s\n",
204 devname
, strerror(errno
));
206 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
207 pr_err("%s is not a block device.\n",
210 } else if (must_be_container(dfd
)) {
212 /* already found some components, this cannot
216 pr_err("%s is a container, but we are looking for components\n",
219 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
220 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
222 pr_err("not a recognisable container: %s\n",
226 } else if (!tst
->ss
->load_container
227 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
229 pr_err("no correct container type: %s\n",
232 } else if (auto_assem
&&
233 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
234 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
236 pr_err("%s has metadata type %s for which "
237 "auto-assembly is disabled\n",
238 devname
, tst
->ss
->name
);
243 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
245 pr_err("no recogniseable superblock on %s\n",
248 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
250 pr_err("no RAID superblock on %s\n",
253 } else if (tst
->ss
->compare_super
== NULL
) {
255 pr_err("Cannot assemble %s metadata on %s\n",
256 tst
->ss
->name
, devname
);
258 } else if (auto_assem
&& st
== NULL
&&
259 !conf_test_metadata(tst
->ss
->name
, (pol
= devid_policy(stb
.st_rdev
)),
260 tst
->ss
->match_home(tst
, c
->homehost
) == 1)) {
262 pr_err("%s has metadata type %s for which "
263 "auto-assembly is disabled\n",
264 devname
, tst
->ss
->name
);
268 if (dfd
>= 0) close(dfd
);
269 if (tmpdev
->used
== 2) {
270 if (auto_assem
|| !inargv
)
271 /* Ignore unrecognised devices during auto-assembly */
273 if (ident
->uuid_set
|| ident
->name
[0] ||
274 ident
->super_minor
!= UnSet
)
275 /* Ignore unrecognised device if looking for
279 pr_err("%s has no superblock - assembly aborted\n",
282 st
->ss
->free_super(st
);
283 dev_policy_free(pol
);
284 domain_free(domains
);
288 if (found_container
) {
289 /* tmpdev is a container. We need to be either
290 * looking for a member, or auto-assembling
292 /* should be safe to try an exclusive open now, we
293 * have rejected anything that some other mdadm might
296 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
299 pr_err("%s is busy - skipping\n", devname
);
304 if (ident
->container
&& ident
->container
[0] != '/') {
309 tst
->ss
->getinfo_super(tst
, content
, NULL
);
311 if (!parse_uuid(ident
->container
, uuid
) ||
312 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
314 pr_err("%s has wrong UUID to be required container\n",
319 /* It is worth looking inside this container.
322 pr_err("looking in container %s\n",
325 for (content
= tst
->ss
->container_content(tst
, NULL
);
327 content
= content
->next
) {
329 if (!ident_matches(ident
, content
, tst
,
330 c
->homehost
, c
->update
,
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 "
354 "only device given: confused and aborting\n",
356 st
->ss
->free_super(st
);
357 dev_policy_free(pol
);
358 domain_free(domains
);
362 pr_err("found match on member %s in %s\n",
363 content
->text_version
, devname
);
365 /* make sure we finished the loop */
370 struct mddev_ident
*match
;
373 tst
->ss
->getinfo_super(tst
, content
, NULL
);
375 if (!ident_matches(ident
, content
, tst
,
376 c
->homehost
, c
->update
,
377 report_mismatch
? devname
: NULL
))
380 match
= conf_match(tst
, content
, devname
,
381 report_mismatch
? c
->verbose
: -1,
383 if (!match
&& rv
== 2)
385 if (match
&& match
->devname
&&
386 strcasecmp(match
->devname
, "<ignore>") == 0) {
388 pr_err("%s is a member of an explicitly ignored array\n",
392 if (match
&& !ident_matches(match
, content
, tst
,
393 c
->homehost
, c
->update
,
394 report_mismatch
? devname
: NULL
))
395 /* Array exists in mdadm.conf but some
396 * details don't match, so reject it
400 /* should be safe to try an exclusive open now, we
401 * have rejected anything that some other mdadm might
404 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
407 pr_err("%s is busy - skipping\n", devname
);
414 if (st
->minor_version
== -1)
415 st
->minor_version
= tst
->minor_version
;
417 if (memcmp(content
->uuid
, uuid_zero
,
418 sizeof(int[4])) == 0) {
419 /* this is a floating spare. It cannot define
420 * an array unless there are no more arrays of
421 * this type to be found. It can be included
422 * in an array of this type though.
428 if (st
->ss
!= tst
->ss
||
429 st
->minor_version
!= tst
->minor_version
||
430 st
->ss
->compare_super(st
, tst
) != 0) {
431 /* Some mismatch. If exactly one array matches this host,
432 * we can resolve on that one.
433 * Or, if we are auto assembling, we just ignore the second
439 int first
= st
->ss
->match_home(st
, c
->homehost
);
440 int last
= tst
->ss
->match_home(tst
, c
->homehost
);
442 (first
== 1 || last
== 1)) {
443 /* We can do something */
444 if (first
) {/* just ignore this one */
446 pr_err("%s misses out due to wrong homehost\n",
449 } else { /* reject all those sofar */
450 struct mddev_dev
*td
;
452 pr_err("%s overrides previous devices due to good homehost\n",
454 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
462 pr_err("superblock on %s doesn't match others - assembly aborted\n",
464 tst
->ss
->free_super(tst
);
465 st
->ss
->free_super(st
);
466 dev_policy_free(pol
);
467 domain_free(domains
);
473 /* Collect domain information from members only */
474 if (tmpdev
&& tmpdev
->used
== 1) {
476 pol
= devid_policy(stb
.st_rdev
);
477 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
479 dev_policy_free(pol
);
482 tst
->ss
->free_super(tst
);
485 /* Check if we found some imsm spares but no members */
488 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
490 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
491 if (tmpdev
->used
!= 3)
497 /* we need sb from one of the spares */
498 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
500 st
->ss
->load_super(st
, dfd
, NULL
))
507 /* Now reject spares that don't match domains of identified members */
508 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
510 if (tmpdev
->used
!= 3)
512 if (stat(tmpdev
->devname
, &stb
)< 0) {
513 pr_err("fstat failed for %s: %s\n",
514 tmpdev
->devname
, strerror(errno
));
517 struct dev_policy
*pol
= devid_policy(stb
.st_rdev
);
518 int dt
= domain_test(domains
, pol
, NULL
);
519 if (inargv
&& dt
!= 0)
520 /* take this spare as domains match
521 * if there are any */
523 else if (!inargv
&& dt
== 1)
524 /* device wasn't explicitly listed, so need
525 * explicit domain match - which we have */
528 /* if domains don't match mark as unused */
530 dev_policy_free(pol
);
533 domain_free(domains
);
535 if (st
&& st
->sb
&& content
== *contentp
)
536 st
->ss
->getinfo_super(st
, content
, NULL
);
544 int uptodate
; /* set once we decide that this device is as
545 * recent as everything else in the array.
547 int included
; /* set if the device is already in the array
548 * due to a previous '-I'
553 static int load_devices(struct devs
*devices
, char *devmap
,
554 struct mddev_ident
*ident
, struct supertype
**stp
,
555 struct mddev_dev
*devlist
, struct context
*c
,
556 struct mdinfo
*content
,
557 int mdfd
, char *mddev
,
558 int *most_recentp
, int *bestcntp
, int **bestp
,
561 struct mddev_dev
*tmpdev
;
567 int most_recent
= -1;
570 struct supertype
*st
= *stp
;
572 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) {
573 char *devname
= tmpdev
->devname
;
577 if (tmpdev
->used
!= 1)
579 /* looks like a good enough match to update the super block if needed */
583 /* prepare useful information in info structures */
585 struct supertype
*tst
;
589 if (strcmp(c
->update
, "uuid")==0 &&
592 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
593 read(rfd
, ident
->uuid
, 16) != 16) {
594 *(__u32
*)(ident
->uuid
) = random();
595 *(__u32
*)(ident
->uuid
+1) = random();
596 *(__u32
*)(ident
->uuid
+2) = random();
597 *(__u32
*)(ident
->uuid
+3) = random();
599 if (rfd
>= 0) close(rfd
);
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",
617 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
619 memcpy(content
->uuid
, ident
->uuid
, 16);
620 strcpy(content
->name
, ident
->name
);
621 content
->array
.md_minor
= minor(stb2
.st_rdev
);
623 if (strcmp(c
->update
, "byteorder") == 0)
626 err
= tst
->ss
->update_super(tst
, content
, c
->update
,
632 pr_err("--update=%s not understood"
633 " for %s metadata\n",
634 c
->update
, tst
->ss
->name
);
635 tst
->ss
->free_super(tst
);
644 if (strcmp(c
->update
, "uuid")==0 &&
647 memcpy(ident
->uuid
, content
->uuid
, 16);
649 if (tst
->ss
->store_super(tst
, dfd
))
650 pr_err("Could not re-write superblock on %s.\n",
654 if (strcmp(c
->update
, "uuid")==0 &&
655 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
656 if (bitmap_update_uuid(ident
->bitmap_fd
,
658 tst
->ss
->swapuuid
) != 0)
659 pr_err("Could not update uuid on external bitmap.\n");
663 tst
->ss
->free_super(tst
);
667 struct supertype
*tst
= dup_super(st
);
669 dfd
= dev_open(devname
,
670 tmpdev
->disposition
== 'I'
671 ? O_RDWR
: (O_RDWR
|O_EXCL
));
673 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
674 pr_err("cannot re-read metadata from %s - aborting\n",
684 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
685 tst
->ss
->free_super(tst
);
692 pr_err("%s is identified as a member of %s, slot %d%s.\n",
693 devname
, mddev
, content
->disk
.raid_disk
,
694 (content
->disk
.state
& (1<<MD_DISK_REPLACEMENT
)) ? " replacement":"");
695 devices
[devcnt
].devname
= devname
;
696 devices
[devcnt
].uptodate
= 0;
697 devices
[devcnt
].included
= (tmpdev
->disposition
== 'I');
698 devices
[devcnt
].i
= *content
;
699 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
700 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
702 if (devices
[devcnt
].i
.disk
.state
== 6) {
703 if (most_recent
< 0 ||
704 devices
[devcnt
].i
.events
705 > devices
[most_recent
].i
.events
)
706 most_recent
= devcnt
;
709 if (content
->array
.level
== LEVEL_MULTIPATH
)
710 /* with multipath, the raid_disk from the superblock is meaningless */
713 i
= devices
[devcnt
].i
.disk
.raid_disk
;
715 if (nextspare
< content
->array
.raid_disks
*2)
716 nextspare
= content
->array
.raid_disks
*2;
719 /* i is raid_disk - double it so there is room for
722 if (devices
[devcnt
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))
724 if (i
>= content
->array
.raid_disks
*2 &&
730 int newbestcnt
= i
+10;
731 int *newbest
= xmalloc(sizeof(int)*newbestcnt
);
733 for (c
=0; c
< newbestcnt
; c
++)
735 newbest
[c
] = best
[c
];
740 bestcnt
= newbestcnt
;
743 devices
[best
[i
]].i
.events
744 == devices
[devcnt
].i
.events
745 && (devices
[best
[i
]].i
.disk
.minor
746 != devices
[devcnt
].i
.disk
.minor
)
748 && content
->array
.level
!= LEVEL_MULTIPATH
) {
749 /* two different devices with identical superblock.
750 * Could be a mis-detection caused by overlapping
751 * partitions. fail-safe.
753 pr_err("WARNING %s and %s appear"
754 " to have very similar superblocks.\n"
755 " If they are really different, "
756 "please --zero the superblock on one\n"
757 " If they are the same or overlap,"
758 " please remove one from %s.\n",
759 devices
[best
[i
]].devname
, devname
,
760 inargv
? "the list" :
761 "the\n DEVICE list in mdadm.conf"
770 || (devices
[best
[i
]].i
.events
771 < devices
[devcnt
].i
.events
))
776 if (most_recent
>= 0)
777 *most_recentp
= most_recent
;
784 static int force_array(struct mdinfo
*content
,
785 struct devs
*devices
,
786 int *best
, int bestcnt
, char *avail
,
788 struct supertype
*st
,
792 while (!enough(content
->array
.level
, content
->array
.raid_disks
,
793 content
->array
.layout
, 1,
796 (content
->reshape_active
&& content
->delta_disks
> 0 &&
797 !enough(content
->array
.level
, (content
->array
.raid_disks
798 - content
->delta_disks
),
799 content
->new_layout
, 1,
802 /* Choose the newest best drive which is
803 * not up-to-date, update the superblock
807 struct supertype
*tst
;
808 unsigned long long current_events
;
809 int chosen_drive
= -1;
813 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
817 !devices
[j
].uptodate
&&
818 devices
[j
].i
.recovery_start
== MaxSector
&&
821 > devices
[chosen_drive
].i
.events
))
824 if (chosen_drive
< 0)
826 current_events
= devices
[chosen_drive
].i
.events
;
829 pr_err("forcing event count in %s(%d) from %d upto %d\n",
830 devices
[chosen_drive
].devname
,
831 devices
[chosen_drive
].i
.disk
.raid_disk
,
832 (int)(devices
[chosen_drive
].i
.events
),
833 (int)(devices
[most_recent
].i
.events
));
834 fd
= dev_open(devices
[chosen_drive
].devname
,
835 devices
[chosen_drive
].included
? O_RDWR
838 pr_err("Couldn't open %s for write - not updating\n",
839 devices
[chosen_drive
].devname
);
840 devices
[chosen_drive
].i
.events
= 0;
844 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
846 pr_err("RAID superblock disappeared from %s - not updating.\n",
847 devices
[chosen_drive
].devname
);
848 devices
[chosen_drive
].i
.events
= 0;
851 content
->events
= devices
[most_recent
].i
.events
;
852 tst
->ss
->update_super(tst
, content
, "force-one",
853 devices
[chosen_drive
].devname
, c
->verbose
,
856 if (tst
->ss
->store_super(tst
, fd
)) {
858 pr_err("Could not re-write superblock on %s\n",
859 devices
[chosen_drive
].devname
);
860 devices
[chosen_drive
].i
.events
= 0;
861 tst
->ss
->free_super(tst
);
865 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
866 devices
[chosen_drive
].uptodate
= 1;
867 avail
[chosen_drive
] = 1;
869 tst
->ss
->free_super(tst
);
871 /* If there are any other drives of the same vintage,
872 * add them in as well. We can't lose and we might gain
875 i
< content
->array
.raid_disks
* 2 && i
< bestcnt
;
879 !devices
[j
].uptodate
&&
880 devices
[j
].i
.recovery_start
== MaxSector
&&
881 devices
[j
].i
.events
== current_events
) {
890 static int start_array(int mdfd
,
892 struct mdinfo
*content
,
893 struct supertype
*st
,
894 struct mddev_ident
*ident
,
895 int *best
, int bestcnt
,
897 struct devs
*devices
,
899 unsigned int sparecnt
,
900 unsigned int rebuilding_cnt
,
902 int clean
, char *avail
,
903 int start_partial_ok
,
910 unsigned int req_cnt
;
912 rv
= set_array_info(mdfd
, st
, content
);
914 pr_err("failed to set array info for %s: %s\n",
915 mddev
, strerror(errno
));
918 if (ident
->bitmap_fd
>= 0) {
919 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
920 pr_err("SET_BITMAP_FILE failed.\n");
923 } else if (ident
->bitmap_file
) {
924 /* From config file */
925 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
927 pr_err("Could not open bitmap file %s\n",
931 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
932 pr_err("Failed to set bitmapfile for %s\n", mddev
);
939 /* First, add the raid disks, but add the chosen one last */
940 for (i
=0; i
<= bestcnt
; i
++) {
944 if (j
== chosen_drive
)
949 if (j
>= 0 && !devices
[j
].included
) {
950 int dfd
= dev_open(devices
[j
].devname
,
953 remove_partitions(dfd
);
956 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
959 pr_err("failed to add "
964 if (i
< content
->array
.raid_disks
* 2
969 } else if (c
->verbose
> 0)
970 pr_err("added %s to %s as %d%s%s\n",
971 devices
[j
].devname
, mddev
,
972 devices
[j
].i
.disk
.raid_disk
,
973 devices
[j
].uptodate
?"":
974 " (possibly out of date)",
975 (devices
[j
].i
.disk
.state
& (1<<MD_DISK_REPLACEMENT
))?" replacement":"");
978 pr_err("%s is already in %s as %d\n",
979 devices
[j
].devname
, mddev
,
980 devices
[j
].i
.disk
.raid_disk
);
981 } else if (c
->verbose
> 0 && i
< content
->array
.raid_disks
*2
983 pr_err("no uptodate device for slot %d of %s\n",
987 if (content
->array
.level
== LEVEL_CONTAINER
) {
988 if (c
->verbose
>= 0) {
989 pr_err("Container %s has been "
990 "assembled with %d drive%s",
991 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
992 if (okcnt
< (unsigned)content
->array
.raid_disks
)
993 fprintf(stderr
, " (out of %d)",
994 content
->array
.raid_disks
);
995 fprintf(stderr
, "\n");
997 st
->ss
->free_super(st
);
998 sysfs_uevent(content
, "change");
999 if (err_ok
&& okcnt
< (unsigned)content
->array
.raid_disks
)
1000 /* Was partial, is still partial, so signal an error
1001 * to ensure we don't retry */
1006 /* Get number of in-sync devices according to the superblock.
1007 * We must have this number to start the array without -s or -R
1009 req_cnt
= content
->array
.working_disks
;
1011 if (c
->runstop
== 1 ||
1013 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1014 content
->array
.layout
, clean
, avail
) &&
1015 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1017 /* This array is good-to-go.
1018 * If a reshape is in progress then we might need to
1019 * continue monitoring it. In that case we start
1020 * it read-only and let the grow code make it writable.
1024 if (content
->reshape_active
&&
1025 !(content
->reshape_active
& RESHAPE_NO_BACKUP
) &&
1026 content
->delta_disks
<= 0) {
1027 if (!c
->backup_file
) {
1028 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1030 pr_err("Please provided one with \"--backup-file=...\"\n");
1032 strcmp(c
->update
, "revert-reshape") == 0)
1033 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1036 rv
= sysfs_set_str(content
, NULL
,
1037 "array_state", "readonly");
1039 rv
= Grow_continue(mdfd
, st
, content
,
1042 } else if (c
->readonly
&&
1043 sysfs_attribute_available(
1044 content
, NULL
, "array_state")) {
1045 rv
= sysfs_set_str(content
, NULL
,
1046 "array_state", "readonly");
1049 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1050 reopen_mddev(mdfd
); /* drop O_EXCL */
1052 if (c
->verbose
>= 0) {
1053 pr_err("%s has been started with %d drive%s",
1054 mddev
, okcnt
, okcnt
==1?"":"s");
1055 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1056 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1058 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1060 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1061 fprintf(stderr
, ".\n");
1063 if (content
->reshape_active
&&
1064 content
->array
.level
>= 4 &&
1065 content
->array
.level
<= 6) {
1066 /* might need to increase the size
1067 * of the stripe cache - default is 256
1069 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1070 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1072 sysfs_set_num(sra
, NULL
,
1073 "stripe_cache_size",
1074 (4 * content
->array
.chunk_size
/ 4096) + 1);
1078 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1079 /* If any devices did not get added
1080 * because the kernel rejected them based
1081 * on event count, try adding them
1082 * again providing the action policy is
1083 * 're-add' or greater. The bitmap
1084 * might allow them to be included, or
1085 * they will become spares.
1087 for (i
= 0; i
< bestcnt
; i
++) {
1089 if (j
>= 0 && !devices
[j
].uptodate
) {
1090 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1092 rv
= add_disk(mdfd
, st
, content
,
1094 if (rv
== 0 && c
->verbose
>= 0)
1095 pr_err("%s has been re-added.\n",
1096 devices
[j
].devname
);
1100 if (content
->array
.level
== 6 &&
1101 okcnt
+ 1 == (unsigned)content
->array
.raid_disks
&&
1103 struct mdinfo
*sra
= sysfs_read(mdfd
, NULL
, 0);
1105 sysfs_set_str(sra
, NULL
,
1106 "sync_action", "repair");
1111 pr_err("failed to RUN_ARRAY %s: %s\n",
1112 mddev
, strerror(errno
));
1114 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1115 content
->array
.layout
, 1, avail
))
1116 pr_err("Not enough devices to "
1117 "start the array.\n");
1118 else if (!enough(content
->array
.level
,
1119 content
->array
.raid_disks
,
1120 content
->array
.layout
, clean
,
1122 pr_err("Not enough devices to "
1123 "start the array while not clean "
1124 "- consider --force.\n");
1128 if (c
->runstop
== -1) {
1129 pr_err("%s assembled from %d drive%s",
1130 mddev
, okcnt
, okcnt
==1?"":"s");
1131 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1132 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1133 fprintf(stderr
, ", but not started.\n");
1136 if (c
->verbose
>= -1) {
1137 pr_err("%s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1139 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1141 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1142 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1143 content
->array
.layout
, 1, avail
))
1144 fprintf(stderr
, " - not enough to start the array.\n");
1145 else if (!enough(content
->array
.level
,
1146 content
->array
.raid_disks
,
1147 content
->array
.layout
, clean
,
1149 fprintf(stderr
, " - not enough to start the "
1150 "array while not clean - consider "
1153 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1154 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1156 fprintf(stderr
, " - need %d to start", req_cnt
);
1157 fprintf(stderr
, " (use --run to insist).\n");
1163 int Assemble(struct supertype
*st
, char *mddev
,
1164 struct mddev_ident
*ident
,
1165 struct mddev_dev
*devlist
,
1169 * The task of Assemble is to find a collection of
1170 * devices that should (according to their superblocks)
1171 * form an array, and to give this collection to the MD driver.
1172 * In Linux-2.4 and later, this involves submitting a
1173 * SET_ARRAY_INFO ioctl with no arg - to prepare
1174 * the array - and then submit a number of
1175 * ADD_NEW_DISK ioctls to add disks into
1176 * the array. Finally RUN_ARRAY might
1177 * be submitted to start the array.
1179 * Much of the work of Assemble is in finding and/or
1180 * checking the disks to make sure they look right.
1182 * If mddev is not set, then scan must be set and we
1183 * read through the config file for dev+uuid mapping
1184 * We recurse, setting mddev, for each device that
1186 * - has a valid uuid (or any uuid if !uuidset)
1188 * If mddev is set, we try to determine state of md.
1189 * check version - must be at least 0.90.0
1190 * check kernel version. must be at least 2.4.
1191 * If not, we can possibly fall back on START_ARRAY
1192 * Try to GET_ARRAY_INFO.
1193 * If possible, give up
1194 * If not, try to STOP_ARRAY just to make sure
1196 * If !uuidset and scan, look in conf-file for uuid
1197 * If not found, give up
1198 * If !devlist and scan and uuidset, get list of devs from conf-file
1201 * Check superblock - discard if bad
1202 * Check uuid (set if we don't have one) - discard if no match
1203 * Check superblock similarity if we have a superblock - discard if different
1204 * Record events, devicenum
1205 * This should give us a list of devices for the array
1206 * We should collect the most recent event number
1208 * Count disks with recent enough event count
1209 * While force && !enough disks
1210 * Choose newest rejected disks, update event count
1211 * mark clean and rewrite superblock
1214 * foreach device with recent events : ADD_NEW_DISK
1215 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1217 * Check the device numbers in superblock are right
1218 * update superblock if any changes
1225 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
1226 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
1227 && (ident
->container
== NULL
|| ident
->member
== NULL
));
1228 struct devs
*devices
;
1230 int *best
= NULL
; /* indexed by raid_disk */
1233 unsigned int okcnt
, sparecnt
, rebuilding_cnt
, replcnt
;
1236 int most_recent
= 0;
1240 int start_partial_ok
= (c
->runstop
>= 0) &&
1241 (c
->force
|| devlist
==NULL
|| auto_assem
);
1243 struct mddev_dev
*tmpdev
;
1245 struct mdinfo
*content
= NULL
;
1246 struct mdinfo
*pre_exist
= NULL
;
1249 char chosen_name
[1024];
1250 struct map_ent
*map
= NULL
;
1254 * If any subdevs are listed, then any that don't
1255 * match ident are discarded. Remainder must all match and
1257 * If no subdevs, then we scan all devices in the config file, but
1258 * there must be something in the identity
1262 ident
->uuid_set
== 0 &&
1263 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
1264 ident
->name
[0] == 0 &&
1265 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
1266 ident
->devices
== NULL
) {
1267 pr_err("No identity information available for %s - cannot assemble.\n",
1268 mddev
? mddev
: "further assembly");
1272 if (devlist
== NULL
)
1273 devlist
= conf_get_devs();
1278 /* We come back here when doing auto-assembly and attempting some
1279 * set of devices failed. Those are now marked as ->used==2 and
1280 * we ignore them and try again
1282 if (!st
&& ident
->st
)
1285 pr_err("looking for devices for %s\n",
1286 mddev
? mddev
: "further assembly");
1290 st
->ignore_hw_compat
= 1;
1291 num_devs
= select_devices(devlist
, ident
, &st
, &content
, c
,
1292 inargv
, auto_assem
);
1296 if (!st
|| !st
->sb
|| !content
)
1299 /* We have a full set of devices - we now need to find the
1301 * However there is a risk that we are racing with "mdadm -I"
1302 * and the array is already partially assembled - we will have
1303 * rejected any devices already in this address.
1304 * So we take a lock on the map file - to prevent further races -
1305 * and look for the uuid in there. If found and the array is
1306 * active, we abort. If found and the array is not active
1307 * we commit to that md device and add all the contained devices
1308 * to our list. We flag them so that we don't try to re-add,
1309 * but can remove if they turn out to not be wanted.
1312 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1313 mp
= map_by_uuid(&map
, content
->uuid
);
1316 /* array already exists. */
1317 pre_exist
= sysfs_read(-1, mp
->devnm
, GET_LEVEL
|GET_DEVS
);
1318 if (pre_exist
->array
.level
!= UnSet
) {
1319 pr_err("Found some drive for an array that is already active: %s\n",
1321 pr_err("giving up.\n");
1324 for (dv
= pre_exist
->devs
; dv
; dv
= dv
->next
) {
1325 /* We want to add this device to our list,
1326 * but it could already be there if "mdadm -I"
1327 * started *after* we checked for O_EXCL.
1328 * If we add it to the top of the list
1329 * it will be preferred over later copies.
1331 struct mddev_dev
*newdev
;
1332 char *devname
= map_dev(dv
->disk
.major
,
1337 newdev
= xmalloc(sizeof(*newdev
));
1338 newdev
->devname
= devname
;
1339 newdev
->disposition
= 'I';
1341 newdev
->next
= devlist
;
1345 strcpy(chosen_name
, mp
->path
);
1346 if (c
->verbose
> 0 || mddev
== NULL
||
1347 strcmp(mddev
, chosen_name
) != 0)
1348 pr_err("Merging with already-assembled %s\n",
1350 mdfd
= open_dev_excl(mp
->devnm
);
1352 int trustworthy
= FOREIGN
;
1353 name
= content
->name
;
1354 switch (st
->ss
->match_home(st
, c
->homehost
)
1355 ?: st
->ss
->match_home(st
, "any")) {
1357 trustworthy
= LOCAL
;
1358 name
= strchr(content
->name
, ':');
1362 name
= content
->name
;
1366 /* If the array is listed in mdadm.conf or on
1367 * command line, then we trust the name
1368 * even if the array doesn't look local
1370 trustworthy
= LOCAL
;
1373 content
->array
.level
== LEVEL_CONTAINER
) {
1374 name
= content
->text_version
;
1375 trustworthy
= METADATA
;
1378 if (name
[0] && trustworthy
!= LOCAL
&&
1379 ! c
->require_homehost
&&
1380 conf_name_is_free(name
))
1381 trustworthy
= LOCAL
;
1383 if (trustworthy
== LOCAL
&&
1385 /* Ignore 'host:' prefix of name */
1386 name
= strchr(name
, ':')+1;
1388 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
1392 st
->ss
->free_super(st
);
1397 mddev
= chosen_name
;
1398 if (get_linux_version() < 2004000 ||
1399 md_get_version(mdfd
) < 9000) {
1400 pr_err("Assemble requires Linux 2.4 or later, and\n"
1401 " md driver version 0.90.0 or later.\n"
1402 " Upgrade your kernel or try --build\n");
1406 if (pre_exist
== NULL
) {
1407 if (mddev_busy(fd2devnm(mdfd
))) {
1408 pr_err("%s already active, cannot restart it!\n",
1410 for (tmpdev
= devlist
;
1411 tmpdev
&& tmpdev
->used
!= 1;
1412 tmpdev
= tmpdev
->next
)
1414 if (tmpdev
&& auto_assem
)
1415 pr_err("%s needed for %s...\n",
1416 mddev
, tmpdev
->devname
);
1419 st
->ss
->free_super(st
);
1424 /* just incase it was started but has no content */
1425 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1429 if (content
!= &info
) {
1430 /* This is a member of a container. Try starting the array. */
1432 err
= assemble_container_content(st
, mdfd
, content
, c
,
1438 /* Ok, no bad inconsistancy, we can try updating etc */
1439 devices
= xcalloc(num_devs
, sizeof(*devices
));
1440 devmap
= xcalloc(num_devs
, content
->array
.raid_disks
);
1441 devcnt
= load_devices(devices
, devmap
, ident
, &st
, devlist
,
1442 c
, content
, mdfd
, mddev
,
1443 &most_recent
, &bestcnt
, &best
, inargv
);
1448 pr_err("no devices found for %s\n",
1451 st
->ss
->free_super(st
);
1458 if (c
->update
&& strcmp(c
->update
, "byteorder")==0)
1459 st
->minor_version
= 90;
1461 st
->ss
->getinfo_super(st
, content
, NULL
);
1462 clean
= content
->array
.state
& 1;
1464 /* now we have some devices that might be suitable.
1467 avail
= xcalloc(content
->array
.raid_disks
, 1);
1472 for (i
=0; i
< bestcnt
; i
++) {
1474 int event_margin
= 1; /* always allow a difference of '1'
1475 * like the kernel does
1477 if (j
< 0) continue;
1478 /* note: we ignore error flags in multipath arrays
1479 * as they don't make sense
1481 if (content
->array
.level
!= LEVEL_MULTIPATH
)
1482 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
1483 if (!(devices
[j
].i
.disk
.state
1484 & (1<<MD_DISK_FAULTY
))) {
1485 devices
[j
].uptodate
= 1;
1490 /* If this device thinks that 'most_recent' has failed, then
1491 * we must reject this device.
1493 if (j
!= most_recent
&& !c
->force
&&
1494 content
->array
.raid_disks
> 0 &&
1495 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
1496 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
1497 if (c
->verbose
> -1)
1498 pr_err("ignoring %s as it reports %s as failed\n",
1499 devices
[j
].devname
, devices
[most_recent
].devname
);
1503 /* Require event counter to be same as, or just less than,
1504 * most recent. If it is bigger, it must be a stray spare and
1505 * should be ignored.
1507 if (devices
[j
].i
.events
+event_margin
>=
1508 devices
[most_recent
].i
.events
&&
1509 devices
[j
].i
.events
<=
1510 devices
[most_recent
].i
.events
1512 devices
[j
].uptodate
= 1;
1513 if (i
< content
->array
.raid_disks
* 2) {
1514 if (devices
[j
].i
.recovery_start
== MaxSector
||
1515 (content
->reshape_active
&&
1516 ((i
>= content
->array
.raid_disks
- content
->delta_disks
) ||
1517 (i
>= content
->array
.raid_disks
- content
->delta_disks
- 1
1518 && content
->array
.level
== 4)))) {
1532 int force_ok
= force_array(content
, devices
, best
, bestcnt
,
1533 avail
, most_recent
, st
, c
);
1538 /* Now we want to look at the superblock which the kernel will base things on
1539 * and compare the devices that we think are working with the devices that the
1540 * superblock thinks are working.
1541 * If there are differences and --force is given, then update this chosen
1545 st
->ss
->free_super(st
);
1546 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
+=2) {
1552 if (!devices
[j
].uptodate
)
1554 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1557 if ((fd
=dev_open(devices
[j
].devname
,
1558 devices
[j
].included
? O_RDONLY
1559 : (O_RDONLY
|O_EXCL
)))< 0) {
1560 pr_err("Cannot open %s: %s\n",
1561 devices
[j
].devname
, strerror(errno
));
1566 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1568 pr_err("RAID superblock has disappeared from %s\n",
1569 devices
[j
].devname
);
1576 if (st
->sb
== NULL
) {
1577 pr_err("No suitable drives found for %s\n", mddev
);
1582 st
->ss
->getinfo_super(st
, content
, NULL
);
1584 sysfs_init(content
, mdfd
, NULL
);
1586 for (i
=0; i
<bestcnt
; i
++) {
1588 unsigned int desired_state
;
1590 if (i
>= content
->array
.raid_disks
* 2)
1593 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_REPLACEMENT
);
1595 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1599 if (!devices
[j
].uptodate
)
1602 devices
[j
].i
.disk
.state
= desired_state
;
1603 if (!(devices
[j
].i
.array
.state
& 1))
1606 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1607 c
->verbose
, 0, NULL
)) {
1609 if (c
->verbose
>= 0)
1610 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1611 j
, mddev
, devices
[j
].devname
);
1614 if (c
->verbose
>= -1)
1615 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1616 i
, mddev
, devices
[j
].devname
);
1620 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1621 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1626 if (c
->force
&& !clean
&&
1627 !enough(content
->array
.level
, content
->array
.raid_disks
,
1628 content
->array
.layout
, clean
,
1630 change
+= st
->ss
->update_super(st
, content
, "force-array",
1631 devices
[chosen_drive
].devname
, c
->verbose
,
1639 fd
= dev_open(devices
[chosen_drive
].devname
,
1640 devices
[chosen_drive
].included
?
1641 O_RDWR
: (O_RDWR
|O_EXCL
));
1643 pr_err("Could not open %s for write - cannot Assemble array.\n",
1644 devices
[chosen_drive
].devname
);
1649 if (st
->ss
->store_super(st
, fd
)) {
1651 pr_err("Could not re-write superblock on %s\n",
1652 devices
[chosen_drive
].devname
);
1657 if (c
->verbose
>= 0)
1658 pr_err("Marking array %s as 'clean'\n",
1663 /* If we are in the middle of a reshape we may need to restore saved data
1664 * that was moved aside due to the reshape overwriting live data
1665 * The code of doing this lives in Grow.c
1668 if (content
->reshape_active
&&
1669 !(content
->reshape_active
& RESHAPE_NO_BACKUP
)) {
1671 int *fdlist
= xmalloc(sizeof(int)* bestcnt
);
1673 pr_err(":%s has an active reshape - checking "
1674 "if critical section needs to be restored\n",
1676 enable_fds(bestcnt
/2);
1677 for (i
= 0; i
< bestcnt
/2; i
++) {
1680 fdlist
[i
] = dev_open(devices
[j
].devname
,
1682 ? O_RDWR
: (O_RDWR
|O_EXCL
));
1683 if (fdlist
[i
] < 0) {
1684 pr_err("Could not open %s for write - cannot Assemble array.\n",
1685 devices
[j
].devname
);
1693 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1694 err
= st
->ss
->recover_backup(st
, content
);
1696 err
= Grow_restart(st
, content
, fdlist
, bestcnt
/2,
1697 c
->backup_file
, c
->verbose
> 0);
1698 if (err
&& c
->invalid_backup
) {
1701 " without restoring backup\n");
1707 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1711 pr_err("Failed to restore critical section for reshape, sorry.\n");
1712 if (c
->backup_file
== NULL
)
1713 cont_err("Possibly you needed to specify the --backup-file\n");
1721 /* Almost ready to actually *do* something */
1722 /* First, fill in the map, so that udev can find our name
1723 * as soon as we become active.
1725 if (c
->update
&& strcmp(c
->update
, "metadata")==0) {
1726 content
->array
.major_version
= 1;
1727 content
->array
.minor_version
= 0;
1728 strcpy(content
->text_version
, "1.0");
1731 map_update(&map
, fd2devnm(mdfd
), content
->text_version
,
1732 content
->uuid
, chosen_name
);
1734 rv
= start_array(mdfd
, mddev
, content
,
1735 st
, ident
, best
, bestcnt
,
1736 chosen_drive
, devices
, okcnt
, sparecnt
,
1739 clean
, avail
, start_partial_ok
,
1742 if (rv
== 1 && !pre_exist
)
1743 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1747 wait_for(chosen_name
, mdfd
);
1751 /* There is a nasty race with 'mdadm --monitor'.
1752 * If it opens this device before we close it,
1753 * it gets an incomplete open on which IO
1754 * doesn't work and the capacity is
1756 * If we reopen (to check for layered devices)
1757 * before --monitor closes, we loose.
1759 * So: wait upto 1 second for there to be
1760 * a non-zero capacity.
1762 while (usecs
< 1000) {
1763 mdfd
= open(mddev
, O_RDONLY
);
1765 unsigned long long size
;
1766 if (get_dev_size(mdfd
, NULL
, &size
) &&
1778 /* '2' means 'OK, but not started yet' */
1779 return rv
== 2 ? 0 : rv
;
1783 int assemble_container_content(struct supertype
*st
, int mdfd
,
1784 struct mdinfo
*content
, struct context
*c
,
1785 char *chosen_name
, int *result
)
1787 struct mdinfo
*dev
, *sra
;
1788 int working
= 0, preexist
= 0;
1790 struct map_ent
*map
= NULL
;
1796 sysfs_init(content
, mdfd
, NULL
);
1798 sra
= sysfs_read(mdfd
, NULL
, GET_VERSION
);
1799 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0) {
1800 if (content
->array
.major_version
== -1 &&
1801 content
->array
.minor_version
== -2 &&
1803 content
->text_version
[0] == '/')
1804 content
->text_version
[0] = '-';
1805 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1812 /* There are two types of reshape: container wide or sub-array specific
1813 * Check if metadata requests blocking container wide reshapes
1815 start_reshape
= (content
->reshape_active
&&
1816 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1817 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1819 /* Block subarray here if it is under reshape now
1820 * Do not allow for any changes in this array
1822 if (st
->ss
->external
&& content
->recovery_blocked
&& start_reshape
)
1823 block_subarray(content
);
1827 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1828 avail
= xcalloc(content
->array
.raid_disks
, 1);
1829 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1830 if (dev
->disk
.raid_disk
>= 0)
1831 avail
[dev
->disk
.raid_disk
] = 1;
1832 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1833 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1834 content
->reshape_active
)
1838 } else if (errno
== EEXIST
)
1841 if (working
+ expansion
== 0 && c
->runstop
<= 0) {
1843 return 1;/* Nothing new, don't try to start */
1845 map_update(&map
, fd2devnm(mdfd
),
1846 content
->text_version
,
1847 content
->uuid
, chosen_name
);
1850 if (enough(content
->array
.level
, content
->array
.raid_disks
,
1851 content
->array
.layout
, content
->array
.state
& 1, avail
) == 0) {
1852 if (c
->export
&& result
)
1854 else if (c
->verbose
>= 0) {
1855 pr_err("%s assembled with %d device%s",
1856 chosen_name
, preexist
+ working
,
1857 preexist
+ working
== 1 ? "":"s");
1859 fprintf(stderr
, " (%d new)", working
);
1860 fprintf(stderr
, " but not started\n");
1867 if (c
->runstop
<= 0 &&
1868 (working
+ preexist
+ expansion
) <
1869 content
->array
.working_disks
) {
1870 if (c
->export
&& result
)
1871 *result
|= INCR_UNSAFE
;
1872 else if (c
->verbose
>= 0) {
1873 pr_err("%s assembled with %d device%s",
1874 chosen_name
, preexist
+ working
,
1875 preexist
+ working
== 1 ? "":"s");
1877 fprintf(stderr
, " (%d new)", working
);
1878 fprintf(stderr
, " but not safe to start\n");
1884 if (start_reshape
) {
1885 int spare
= content
->array
.raid_disks
+ expansion
;
1886 if (restore_backup(st
, content
,
1888 spare
, c
->backup_file
, c
->verbose
) == 1)
1891 err
= sysfs_set_str(content
, NULL
,
1892 "array_state", "readonly");
1896 if (st
->ss
->external
) {
1897 if (!mdmon_running(st
->container_devnm
))
1898 start_mdmon(st
->container_devnm
);
1899 ping_monitor(st
->container_devnm
);
1900 if (mdmon_running(st
->container_devnm
) &&
1901 st
->update_tail
== NULL
)
1902 st
->update_tail
= &st
->updates
;
1905 err
= Grow_continue(mdfd
, st
, content
, c
->backup_file
,
1907 } else switch(content
->array
.level
) {
1909 case LEVEL_MULTIPATH
:
1911 err
= sysfs_set_str(content
, NULL
, "array_state",
1912 c
->readonly
? "readonly" : "active");
1915 err
= sysfs_set_str(content
, NULL
, "array_state",
1917 /* start mdmon if needed. */
1919 if (!mdmon_running(st
->container_devnm
))
1920 start_mdmon(st
->container_devnm
);
1921 ping_monitor(st
->container_devnm
);
1926 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1928 /* Block subarray here if it is not reshaped now
1929 * It has be blocked a little later to allow mdmon to switch in
1932 if (st
->ss
->external
&& content
->recovery_blocked
&&
1934 block_subarray(content
);
1936 if (c
->export
&& result
) {
1940 *result
|= INCR_YES
;
1941 } else if (c
->verbose
>= 0) {
1943 pr_err("array %s now has %d device%s",
1944 chosen_name
, working
+ preexist
,
1945 working
+ preexist
== 1 ? "":"s");
1947 pr_err("Started %s with %d device%s",
1948 chosen_name
, working
+ preexist
,
1949 working
+ preexist
== 1 ? "":"s");
1951 fprintf(stderr
, " (%d new)", working
);
1953 fprintf(stderr
, " ( + %d for expansion)",
1955 fprintf(stderr
, "\n");
1958 wait_for(chosen_name
, mdfd
);
1960 /* FIXME should have an O_EXCL and wait for read-auto */