]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
6ba78a8669c5919177514279397cdb29b4804dad
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 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(1, 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) {
83 fprintf(stderr
, Name
": %s has wrong uuid.\n",
87 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
88 name_matches(content
->name
, ident
->name
, homehost
)==0) {
90 fprintf(stderr
, Name
": %s has wrong name.\n",
94 if (ident
->super_minor
!= UnSet
&&
95 ident
->super_minor
!= content
->array
.md_minor
) {
97 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
101 if (ident
->level
!= UnSet
&&
102 ident
->level
!= content
->array
.level
) {
104 fprintf(stderr
, Name
": %s has wrong raid level.\n",
108 if (ident
->raid_disks
!= UnSet
&&
109 ident
->raid_disks
!= content
->array
.raid_disks
) {
111 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
119 int Assemble(struct supertype
*st
, char *mddev
,
120 struct mddev_ident
*ident
,
121 struct mddev_dev
*devlist
, char *backup_file
,
122 int readonly
, int runstop
,
123 char *update
, char *homehost
, int require_homehost
,
124 int verbose
, int force
)
127 * The task of Assemble is to find a collection of
128 * devices that should (according to their superblocks)
129 * form an array, and to give this collection to the MD driver.
130 * In Linux-2.4 and later, this involves submitting a
131 * SET_ARRAY_INFO ioctl with no arg - to prepare
132 * the array - and then submit a number of
133 * ADD_NEW_DISK ioctls to add disks into
134 * the array. Finally RUN_ARRAY might
135 * be submitted to start the array.
137 * Much of the work of Assemble is in finding and/or
138 * checking the disks to make sure they look right.
140 * If mddev is not set, then scan must be set and we
141 * read through the config file for dev+uuid mapping
142 * We recurse, setting mddev, for each device that
144 * - has a valid uuid (or any uuid if !uuidset)
146 * If mddev is set, we try to determine state of md.
147 * check version - must be at least 0.90.0
148 * check kernel version. must be at least 2.4.
149 * If not, we can possibly fall back on START_ARRAY
150 * Try to GET_ARRAY_INFO.
151 * If possible, give up
152 * If not, try to STOP_ARRAY just to make sure
154 * If !uuidset and scan, look in conf-file for uuid
155 * If not found, give up
156 * If !devlist and scan and uuidset, get list of devs from conf-file
159 * Check superblock - discard if bad
160 * Check uuid (set if we don't have one) - discard if no match
161 * Check superblock similarity if we have a superblock - discard if different
162 * Record events, devicenum
163 * This should give us a list of devices for the array
164 * We should collect the most recent event number
166 * Count disks with recent enough event count
167 * While force && !enough disks
168 * Choose newest rejected disks, update event count
169 * mark clean and rewrite superblock
172 * foreach device with recent events : ADD_NEW_DISK
173 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
175 * Check the device numbers in superblock are right
176 * update superblock if any changes
182 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
183 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
184 && (ident
->container
== NULL
|| ident
->member
== NULL
));
186 int vers
= vers
; /* Keep gcc quite - it really is initialised */
189 int uptodate
; /* set once we decide that this device is as
190 * recent as everything else in the array.
195 int *best
= NULL
; /* indexed by raid_disk */
198 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
199 unsigned int req_cnt
;
205 int report_missmatch
;
207 int start_partial_ok
= (runstop
>= 0) &&
208 (force
|| devlist
==NULL
|| auto_assem
);
209 unsigned int num_devs
;
210 struct mddev_dev
*tmpdev
;
212 struct mdinfo
*content
= NULL
;
217 char chosen_name
[1024];
219 if (get_linux_version() < 2004000)
223 * If any subdevs are listed, then any that don't
224 * match ident are discarded. Remainder must all match and
226 * If no subdevs, then we scan all devices in the config file, but
227 * there must be something in the identity
231 ident
->uuid_set
== 0 &&
232 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
233 ident
->name
[0] == 0 &&
234 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
235 ident
->devices
== NULL
) {
236 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
237 mddev
? mddev
: "further assembly");
242 devlist
= conf_get_devs();
246 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
248 /* We come back here when doing auto-assembly and attempting some
249 * set of devices failed. Those are now marked as ->used==2 and
250 * we ignore them and try again
253 tmpdev
= devlist
; num_devs
= 0;
259 tmpdev
= tmpdev
->next
;
262 if (!st
&& ident
->st
) st
= ident
->st
;
265 fprintf(stderr
, Name
": looking for devices for %s\n",
266 mddev
? mddev
: "further assembly");
268 /* first walk the list of devices to find a consistent set
269 * that match the criterea, if that is possible.
270 * We flag the ones we like with 'used'.
272 for (tmpdev
= devlist
;
274 tmpdev
= tmpdev
->next
) {
275 char *devname
= tmpdev
->devname
;
278 struct supertype
*tst
= dup_super(st
);
279 struct dev_policy
*pol
= NULL
;
281 if (tmpdev
->used
> 1) continue;
283 if (ident
->devices
&&
284 !match_oneof(ident
->devices
, devname
)) {
285 if (report_missmatch
)
286 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
290 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
292 if (report_missmatch
)
293 fprintf(stderr
, Name
": cannot open device %s: %s\n",
294 devname
, strerror(errno
));
296 } else if (fstat(dfd
, &stb
)< 0) {
298 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
299 devname
, strerror(errno
));
301 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
302 fprintf(stderr
, Name
": %s is not a block device.\n",
305 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
306 if (report_missmatch
)
307 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
310 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
311 if (report_missmatch
)
312 fprintf(stderr
, Name
": no RAID superblock on %s\n",
315 } else if (tst
->ss
->compare_super
== NULL
) {
316 if (report_missmatch
)
317 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
318 tst
->ss
->name
, devname
);
320 } else if (auto_assem
&& st
== NULL
&&
321 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
322 tst
->ss
->match_home(tst
, homehost
) == 1)) {
323 if (report_missmatch
)
324 fprintf(stderr
, Name
": %s has metadata type %s for which "
325 "auto-assembly is disabled\n",
326 devname
, tst
->ss
->name
);
330 memset(content
, 0, sizeof(*content
));
331 tst
->ss
->getinfo_super(tst
, content
, NULL
);
333 if (dfd
>= 0) close(dfd
);
334 if (tmpdev
->used
== 2) {
336 /* Ignore unrecognised devices during auto-assembly */
338 if (ident
->uuid_set
|| ident
->name
[0] ||
339 ident
->super_minor
!= UnSet
)
340 /* Ignore unrecognised device if looking for
345 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
348 st
->ss
->free_super(st
);
349 dev_policy_free(pol
);
353 if (tst
->ss
->container_content
354 && tst
->loaded_container
) {
355 /* tmpdev is a container. We need to be either
356 * looking for a member, or auto-assembling
359 /* already found some components, this cannot
362 if (report_missmatch
)
363 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
368 if (ident
->container
) {
369 if (ident
->container
[0] == '/' &&
370 !same_dev(ident
->container
, devname
)) {
371 if (report_missmatch
)
372 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
373 devname
, ident
->container
);
376 if (ident
->container
[0] != '/') {
379 if (!parse_uuid(ident
->container
, uuid
) ||
380 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
381 if (report_missmatch
)
382 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
388 /* It is worth looking inside this container.
391 fprintf(stderr
, Name
": looking in container %s\n",
395 content
= tmpdev
->content
;
397 content
= tst
->ss
->container_content(tst
, NULL
);
399 goto loop
; /* empty container */
401 tmpdev
->content
= content
->next
;
402 if (tmpdev
->content
== NULL
)
405 } else if (ident
->container
|| ident
->member
) {
406 /* No chance of this matching if we don't have
408 if (report_missmatch
)
409 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
414 if (!ident_matches(ident
, content
, tst
,
416 report_missmatch
? devname
: NULL
))
419 if (tst
->ss
->container_content
420 && tst
->loaded_container
) {
421 /* we have the one container we need, don't keep
422 * looking. If the chosen member is active, skip.
424 if (is_member_busy(content
->text_version
)) {
425 if (report_missmatch
)
426 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
427 content
->text_version
,
432 tst
->ss
->free_super(tst
);
437 dev_policy_free(pol
);
440 if (ident
->member
&& ident
->member
[0]) {
441 char *s
= strchr(content
->text_version
+1, '/');
443 fprintf(stderr
, Name
": badly formatted version: %s\n",
444 content
->text_version
);
447 if (strcmp(ident
->member
, s
+1) != 0) {
448 if (report_missmatch
)
450 Name
": skipping wrong member %s\n",
451 content
->text_version
);
455 st
= tst
; tst
= NULL
;
456 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
457 fprintf(stderr
, Name
": %s is a container, but is not "
458 "only device given: confused and aborting\n",
460 st
->ss
->free_super(st
);
461 dev_policy_free(pol
);
465 fprintf(stderr
, Name
": found match on member %s in %s\n",
466 content
->text_version
, devname
);
471 if (st
->minor_version
== -1)
472 st
->minor_version
= tst
->minor_version
;
473 if (st
->ss
!= tst
->ss
||
474 st
->minor_version
!= tst
->minor_version
||
475 st
->ss
->compare_super(st
, tst
) != 0) {
476 /* Some mismatch. If exactly one array matches this host,
477 * we can resolve on that one.
478 * Or, if we are auto assembling, we just ignore the second
484 int first
= st
->ss
->match_home(st
, homehost
);
485 int last
= tst
->ss
->match_home(tst
, homehost
);
487 (first
== 1 || last
== 1)) {
488 /* We can do something */
489 if (first
) {/* just ignore this one */
490 if (report_missmatch
)
491 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
494 } else { /* reject all those sofar */
495 struct mddev_dev
*td
;
496 if (report_missmatch
)
497 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
499 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
507 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
509 tst
->ss
->free_super(tst
);
510 st
->ss
->free_super(st
);
511 dev_policy_free(pol
);
518 dev_policy_free(pol
);
523 tst
->ss
->free_super(tst
);
526 if (!st
|| !st
->sb
|| !content
)
529 /* Now need to open the array device. Use create_mddev */
530 if (content
== &info
)
531 st
->ss
->getinfo_super(st
, content
, NULL
);
533 trustworthy
= FOREIGN
;
534 name
= content
->name
;
535 switch (st
->ss
->match_home(st
, homehost
)
536 ?: st
->ss
->match_home(st
, "any")) {
539 name
= strchr(content
->name
, ':');
543 name
= content
->name
;
547 /* If the array is listed in mdadm.conf or on
548 * command line, then we trust the name
549 * even if the array doesn't look local
554 content
->array
.level
== LEVEL_CONTAINER
) {
555 name
= content
->text_version
;
556 trustworthy
= METADATA
;
559 if (name
[0] && trustworthy
!= LOCAL
&&
560 ! require_homehost
&&
561 conf_name_is_free(name
))
564 if (trustworthy
== LOCAL
&&
566 /* Ignore 'host:' prefix of name */
567 name
= strchr(name
, ':')+1;
569 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
572 st
->ss
->free_super(st
);
578 vers
= md_get_version(mdfd
);
580 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
581 " Upgrade your kernel or try --build\n");
585 if (mddev_busy(fd2devnum(mdfd
))) {
586 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
588 for (tmpdev
= devlist
;
589 tmpdev
&& tmpdev
->used
!= 1;
590 tmpdev
= tmpdev
->next
)
592 if (tmpdev
&& auto_assem
)
593 fprintf(stderr
, Name
": %s needed for %s...\n",
594 mddev
, tmpdev
->devname
);
597 st
->ss
->free_super(st
);
602 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
605 if (content
!= &info
) {
606 /* This is a member of a container. Try starting the array. */
607 return assemble_container_content(st
, mdfd
, content
, runstop
,
608 chosen_name
, verbose
);
611 /* Ok, no bad inconsistancy, we can try updating etc */
613 content
->update_private
= NULL
;
614 devices
= malloc(num_devs
* sizeof(*devices
));
615 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
616 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
617 char *devname
= tmpdev
->devname
;
619 /* looks like a good enough match to update the super block if needed */
623 /* prepare useful information in info structures */
625 struct supertype
*tst
;
629 if (strcmp(update
, "uuid")==0 &&
632 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
633 read(rfd
, ident
->uuid
, 16) != 16) {
634 *(__u32
*)(ident
->uuid
) = random();
635 *(__u32
*)(ident
->uuid
+1) = random();
636 *(__u32
*)(ident
->uuid
+2) = random();
637 *(__u32
*)(ident
->uuid
+3) = random();
639 if (rfd
>= 0) close(rfd
);
641 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
643 remove_partitions(dfd
);
646 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
647 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
656 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
658 memcpy(content
->uuid
, ident
->uuid
, 16);
659 strcpy(content
->name
, ident
->name
);
660 content
->array
.md_minor
= minor(stb2
.st_rdev
);
662 if (strcmp(update
, "byteorder") == 0)
665 err
= tst
->ss
->update_super(tst
, content
, update
,
671 Name
": --update=%s not understood"
672 " for %s metadata\n",
673 update
, tst
->ss
->name
);
674 tst
->ss
->free_super(tst
);
682 if (strcmp(update
, "uuid")==0 &&
685 memcpy(ident
->uuid
, content
->uuid
, 16);
687 if (tst
->ss
->store_super(tst
, dfd
))
688 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
692 if (strcmp(update
, "uuid")==0 &&
693 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
694 if (bitmap_update_uuid(ident
->bitmap_fd
,
696 tst
->ss
->swapuuid
) != 0)
697 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
701 tst
->ss
->free_super(tst
);
705 struct supertype
*tst
= dup_super(st
);
707 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
709 remove_partitions(dfd
);
711 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
712 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
721 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
722 tst
->ss
->free_super(tst
);
729 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
730 devname
, mddev
, content
->disk
.raid_disk
);
731 devices
[devcnt
].devname
= devname
;
732 devices
[devcnt
].uptodate
= 0;
733 devices
[devcnt
].i
= *content
;
734 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
735 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
736 if (most_recent
< devcnt
) {
737 if (devices
[devcnt
].i
.events
738 > devices
[most_recent
].i
.events
)
739 most_recent
= devcnt
;
741 if (content
->array
.level
== LEVEL_MULTIPATH
)
742 /* with multipath, the raid_disk from the superblock is meaningless */
745 i
= devices
[devcnt
].i
.disk
.raid_disk
;
747 if (nextspare
< content
->array
.raid_disks
)
748 nextspare
= content
->array
.raid_disks
;
751 if (i
>= content
->array
.raid_disks
&&
757 int newbestcnt
= i
+10;
758 int *newbest
= malloc(sizeof(int)*newbestcnt
);
760 for (c
=0; c
< newbestcnt
; c
++)
762 newbest
[c
] = best
[c
];
767 bestcnt
= newbestcnt
;
770 devices
[best
[i
]].i
.events
771 == devices
[devcnt
].i
.events
772 && (devices
[best
[i
]].i
.disk
.minor
773 != devices
[devcnt
].i
.disk
.minor
)
775 && content
->array
.level
!= LEVEL_MULTIPATH
) {
776 /* two different devices with identical superblock.
777 * Could be a mis-detection caused by overlapping
778 * partitions. fail-safe.
780 fprintf(stderr
, Name
": WARNING %s and %s appear"
781 " to have very similar superblocks.\n"
782 " If they are really different, "
783 "please --zero the superblock on one\n"
784 " If they are the same or overlap,"
785 " please remove one from %s.\n",
786 devices
[best
[i
]].devname
, devname
,
787 inargv
? "the list" :
788 "the\n DEVICE list in mdadm.conf"
796 || (devices
[best
[i
]].i
.events
797 < devices
[devcnt
].i
.events
))
802 free(content
->update_private
);
803 content
->update_private
= NULL
;
806 fprintf(stderr
, Name
": no devices found for %s\n",
809 st
->ss
->free_super(st
);
816 if (update
&& strcmp(update
, "byteorder")==0)
817 st
->minor_version
= 90;
819 st
->ss
->getinfo_super(st
, content
, NULL
);
820 clean
= content
->array
.state
& 1;
822 /* now we have some devices that might be suitable.
825 avail
= malloc(content
->array
.raid_disks
);
826 memset(avail
, 0, content
->array
.raid_disks
);
830 for (i
=0; i
< bestcnt
; i
++) {
832 int event_margin
= 1; /* always allow a difference of '1'
833 * like the kernel does
836 /* note: we ignore error flags in multipath arrays
837 * as they don't make sense
839 if (content
->array
.level
!= LEVEL_MULTIPATH
)
840 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
841 if (!(devices
[j
].i
.disk
.state
842 & (1<<MD_DISK_FAULTY
)))
846 /* If this devices thinks that 'most_recent' has failed, then
847 * we must reject this device.
849 if (j
!= most_recent
&&
850 content
->array
.raid_disks
> 0 &&
851 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
852 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
854 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
855 devices
[j
].devname
, devices
[most_recent
].devname
);
859 if (devices
[j
].i
.events
+event_margin
>=
860 devices
[most_recent
].i
.events
) {
861 devices
[j
].uptodate
= 1;
862 if (i
< content
->array
.raid_disks
) {
863 if (devices
[j
].i
.recovery_start
== MaxSector
) {
873 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
874 content
->array
.layout
, 1,
876 /* Choose the newest best drive which is
877 * not up-to-date, update the superblock
881 struct supertype
*tst
;
882 unsigned long long current_events
;
884 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
887 !devices
[j
].uptodate
&&
888 devices
[j
].i
.recovery_start
== MaxSector
&&
891 > devices
[chosen_drive
].i
.events
))
894 if (chosen_drive
< 0)
896 current_events
= devices
[chosen_drive
].i
.events
;
899 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
900 devices
[chosen_drive
].devname
,
901 devices
[chosen_drive
].i
.disk
.raid_disk
,
902 (int)(devices
[chosen_drive
].i
.events
),
903 (int)(devices
[most_recent
].i
.events
));
904 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
906 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
907 devices
[chosen_drive
].devname
);
908 devices
[chosen_drive
].i
.events
= 0;
912 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
914 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
915 devices
[chosen_drive
].devname
);
916 devices
[chosen_drive
].i
.events
= 0;
919 content
->events
= devices
[most_recent
].i
.events
;
920 tst
->ss
->update_super(tst
, content
, "force-one",
921 devices
[chosen_drive
].devname
, verbose
,
924 if (tst
->ss
->store_super(tst
, fd
)) {
926 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
927 devices
[chosen_drive
].devname
);
928 devices
[chosen_drive
].i
.events
= 0;
929 tst
->ss
->free_super(tst
);
933 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
934 devices
[chosen_drive
].uptodate
= 1;
935 avail
[chosen_drive
] = 1;
937 tst
->ss
->free_super(tst
);
939 /* If there are any other drives of the same vintage,
940 * add them in as well. We can't lose and we might gain
942 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
945 !devices
[j
].uptodate
&&
946 devices
[j
].i
.events
== current_events
) {
953 /* Now we want to look at the superblock which the kernel will base things on
954 * and compare the devices that we think are working with the devices that the
955 * superblock thinks are working.
956 * If there are differences and --force is given, then update this chosen
960 st
->ss
->free_super(st
);
961 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
967 if (!devices
[j
].uptodate
)
970 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
971 fprintf(stderr
, Name
": Cannot open %s: %s\n",
972 devices
[j
].devname
, strerror(errno
));
977 if (st
->ss
->load_super(st
,fd
, NULL
)) {
979 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
987 if (st
->sb
== NULL
) {
988 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
993 st
->ss
->getinfo_super(st
, content
, NULL
);
995 sysfs_init(content
, mdfd
, 0);
997 for (i
=0; i
<bestcnt
; i
++) {
999 unsigned int desired_state
;
1001 if (i
< content
->array
.raid_disks
)
1002 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1008 if (!devices
[j
].uptodate
)
1011 devices
[j
].i
.disk
.state
= desired_state
;
1012 if (!(devices
[j
].i
.array
.state
& 1))
1015 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1016 verbose
, 0, NULL
)) {
1019 fprintf(stderr
, Name
": "
1020 "clearing FAULTY flag for device %d in %s for %s\n",
1021 j
, mddev
, devices
[j
].devname
);
1025 fprintf(stderr
, Name
": "
1026 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1027 i
, mddev
, devices
[j
].devname
);
1031 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1032 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1037 if (force
&& !clean
&&
1038 !enough(content
->array
.level
, content
->array
.raid_disks
,
1039 content
->array
.layout
, clean
,
1041 change
+= st
->ss
->update_super(st
, content
, "force-array",
1042 devices
[chosen_drive
].devname
, verbose
,
1049 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1051 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1052 devices
[chosen_drive
].devname
);
1057 if (st
->ss
->store_super(st
, fd
)) {
1059 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1060 devices
[chosen_drive
].devname
);
1068 /* If we are in the middle of a reshape we may need to restore saved data
1069 * that was moved aside due to the reshape overwriting live data
1070 * The code of doing this lives in Grow.c
1073 if (content
->reshape_active
) {
1075 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1077 fprintf(stderr
, Name
":%s has an active reshape - checking "
1078 "if critical section needs to be restored\n",
1080 for (i
=0; i
<bestcnt
; i
++) {
1083 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1084 if (fdlist
[i
] < 0) {
1085 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1086 devices
[j
].devname
);
1094 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1097 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1100 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1101 if (backup_file
== NULL
)
1102 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1109 /* count number of in-sync devices according to the superblock.
1110 * We must have this number to start the array without -s or -R
1112 req_cnt
= content
->array
.working_disks
;
1114 /* Almost ready to actually *do* something */
1118 /* First, fill in the map, so that udev can find our name
1119 * as soon as we become active.
1121 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1122 content
->uuid
, chosen_name
);
1124 rv
= set_array_info(mdfd
, st
, content
);
1126 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1127 mddev
, strerror(errno
));
1128 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1133 if (ident
->bitmap_fd
>= 0) {
1134 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1135 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1136 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1141 } else if (ident
->bitmap_file
) {
1142 /* From config file */
1143 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1145 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1146 ident
->bitmap_file
);
1147 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1152 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1153 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1155 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1163 /* First, add the raid disks, but add the chosen one last */
1164 for (i
=0; i
<= bestcnt
; i
++) {
1168 if (j
== chosen_drive
)
1173 if (j
>= 0 /* && devices[j].uptodate */) {
1174 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1177 fprintf(stderr
, Name
": failed to add "
1182 if (i
< content
->array
.raid_disks
1187 } else if (verbose
> 0)
1188 fprintf(stderr
, Name
": added %s "
1190 devices
[j
].devname
, mddev
,
1191 devices
[j
].i
.disk
.raid_disk
);
1192 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1193 fprintf(stderr
, Name
": no uptodate device for "
1198 if (content
->array
.level
== LEVEL_CONTAINER
) {
1200 fprintf(stderr
, Name
": Container %s has been "
1201 "assembled with %d drive%s",
1202 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1203 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1204 fprintf(stderr
, " (out of %d)",
1205 content
->array
.raid_disks
);
1206 fprintf(stderr
, "\n");
1208 sysfs_uevent(content
, "change");
1209 wait_for(chosen_name
, mdfd
);
1217 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1218 content
->array
.layout
, clean
, avail
, okcnt
) &&
1219 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1221 /* This array is good-to-go.
1222 * If a reshape is in progress then we might need to
1223 * continue monitoring it. In that case we start
1224 * it read-only and let the grow code make it writable.
1228 if (content
->reshape_active
&&
1229 content
->delta_disks
<= 0)
1230 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1233 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1236 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1237 mddev
, okcnt
, okcnt
==1?"":"s");
1238 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1239 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1241 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1243 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1244 fprintf(stderr
, ".\n");
1246 if (content
->reshape_active
&&
1247 content
->array
.level
>= 4 &&
1248 content
->array
.level
<= 6) {
1249 /* might need to increase the size
1250 * of the stripe cache - default is 256
1252 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1253 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1255 sysfs_set_num(sra
, NULL
,
1256 "stripe_cache_size",
1257 (4 * content
->array
.chunk_size
/ 4096) + 1);
1260 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1261 /* If any devices did not get added
1262 * because the kernel rejected them based
1263 * on event count, try adding them
1264 * again providing the action policy is
1265 * 're-add' or greater. The bitmap
1266 * might allow them to be included, or
1267 * they will become spares.
1269 for (i
= 0; i
<= bestcnt
; i
++) {
1271 if (j
>= 0 && !devices
[j
].uptodate
) {
1272 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1274 rv
= add_disk(mdfd
, st
, content
,
1276 if (rv
== 0 && verbose
>= 0)
1278 Name
": %s has been re-added.\n",
1279 devices
[j
].devname
);
1283 wait_for(mddev
, mdfd
);
1287 /* There is a nasty race with 'mdadm --monitor'.
1288 * If it opens this device before we close it,
1289 * it gets an incomplete open on which IO
1290 * doesn't work and the capacity is
1292 * If we reopen (to check for layered devices)
1293 * before --monitor closes, we loose.
1295 * So: wait upto 1 second for there to be
1296 * a non-zero capacity.
1298 while (usecs
< 1000) {
1299 mdfd
= open(mddev
, O_RDONLY
);
1301 unsigned long long size
;
1302 if (get_dev_size(mdfd
, NULL
, &size
) &&
1314 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1315 mddev
, strerror(errno
));
1317 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1318 content
->array
.layout
, 1, avail
, okcnt
))
1319 fprintf(stderr
, Name
": Not enough devices to "
1320 "start the array.\n");
1321 else if (!enough(content
->array
.level
,
1322 content
->array
.raid_disks
,
1323 content
->array
.layout
, clean
,
1325 fprintf(stderr
, Name
": Not enough devices to "
1326 "start the array while not clean "
1327 "- consider --force.\n");
1330 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1335 if (runstop
== -1) {
1336 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1337 mddev
, okcnt
, okcnt
==1?"":"s");
1338 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1339 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1340 fprintf(stderr
, ", but not started.\n");
1345 if (verbose
>= -1) {
1346 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1348 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1350 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1351 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1352 content
->array
.layout
, 1, avail
, okcnt
))
1353 fprintf(stderr
, " - not enough to start the array.\n");
1354 else if (!enough(content
->array
.level
,
1355 content
->array
.raid_disks
,
1356 content
->array
.layout
, clean
,
1358 fprintf(stderr
, " - not enough to start the "
1359 "array while not clean - consider "
1362 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1363 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1365 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1366 fprintf(stderr
, " (use --run to insist).\n");
1370 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1375 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1376 * been updated to point to the current locations of devices.
1377 * so we can just start the array
1380 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1381 devices
[chosen_drive
].i
.disk
.minor
);
1382 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1383 fprintf(stderr
, Name
": Cannot start array: %s\n",
1394 int assemble_container_content(struct supertype
*st
, int mdfd
,
1395 struct mdinfo
*content
, int runstop
,
1396 char *chosen_name
, int verbose
)
1398 struct mdinfo
*dev
, *sra
;
1399 int working
= 0, preexist
= 0;
1400 struct map_ent
*map
= NULL
;
1402 sysfs_init(content
, mdfd
, 0);
1404 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1405 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1406 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1413 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1414 if (sysfs_add_disk(content
, dev
, 1) == 0)
1416 else if (errno
== EEXIST
)
1420 return 1;/* Nothing new, don't try to start */
1423 map_update(&map
, fd2devnum(mdfd
),
1424 content
->text_version
,
1425 content
->uuid
, chosen_name
);
1428 (working
+ preexist
) >= content
->array
.working_disks
) {
1431 switch(content
->array
.level
) {
1433 case LEVEL_MULTIPATH
:
1435 err
= sysfs_set_str(content
, NULL
, "array_state",
1439 err
= sysfs_set_str(content
, NULL
, "array_state",
1441 /* start mdmon if needed. */
1443 if (!mdmon_running(st
->container_dev
))
1444 start_mdmon(st
->container_dev
);
1445 ping_monitor(devnum2devname(st
->container_dev
));
1450 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1453 fprintf(stderr
, Name
1454 ": array %s now has %d devices",
1455 chosen_name
, working
+ preexist
);
1457 fprintf(stderr
, Name
1458 ": Started %s with %d devices",
1459 chosen_name
, working
+ preexist
);
1461 fprintf(stderr
, " (%d new)", working
);
1462 fprintf(stderr
, "\n");
1465 wait_for(chosen_name
, mdfd
);
1468 /* FIXME should have an O_EXCL and wait for read-auto */
1471 fprintf(stderr
, Name
1472 ": %s assembled with %d devices but "
1474 chosen_name
, working
);