]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
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",
394 content
= tst
->ss
->container_content(tst
, NULL
);
398 goto loop
; /* empty container */
400 if (content
->next
== NULL
)
403 if (!ident_matches(ident
, content
, tst
,
405 report_missmatch
? devname
: NULL
)) {
406 content
= content
->next
;
410 /* we have the one container we need, don't keep
411 * looking. If the chosen member is active, skip.
413 if (is_member_busy(content
->text_version
)) {
414 if (report_missmatch
)
415 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
416 content
->text_version
,
419 content
= content
->next
;
422 tst
->ss
->free_super(tst
);
426 dev_policy_free(pol
);
429 if (ident
->member
&& ident
->member
[0]) {
430 char *s
= strchr(content
->text_version
+1, '/');
432 fprintf(stderr
, Name
": badly formatted version: %s\n",
433 content
->text_version
);
436 if (strcmp(ident
->member
, s
+1) != 0) {
437 if (report_missmatch
)
439 Name
": skipping wrong member %s\n",
440 content
->text_version
);
444 st
= tst
; tst
= NULL
;
445 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
446 fprintf(stderr
, Name
": %s is a container, but is not "
447 "only device given: confused and aborting\n",
449 st
->ss
->free_super(st
);
450 dev_policy_free(pol
);
454 fprintf(stderr
, Name
": found match on member %s in %s\n",
455 content
->text_version
, devname
);
459 if (ident
->container
|| ident
->member
) {
460 /* No chance of this matching if we don't have
462 if (report_missmatch
)
463 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
468 if (!ident_matches(ident
, content
, tst
,
470 report_missmatch
? devname
: NULL
))
475 if (st
->minor_version
== -1)
476 st
->minor_version
= tst
->minor_version
;
477 if (st
->ss
!= tst
->ss
||
478 st
->minor_version
!= tst
->minor_version
||
479 st
->ss
->compare_super(st
, tst
) != 0) {
480 /* Some mismatch. If exactly one array matches this host,
481 * we can resolve on that one.
482 * Or, if we are auto assembling, we just ignore the second
488 int first
= st
->ss
->match_home(st
, homehost
);
489 int last
= tst
->ss
->match_home(tst
, homehost
);
491 (first
== 1 || last
== 1)) {
492 /* We can do something */
493 if (first
) {/* just ignore this one */
494 if (report_missmatch
)
495 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
498 } else { /* reject all those sofar */
499 struct mddev_dev
*td
;
500 if (report_missmatch
)
501 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
503 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
511 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
513 tst
->ss
->free_super(tst
);
514 st
->ss
->free_super(st
);
515 dev_policy_free(pol
);
522 dev_policy_free(pol
);
525 tst
->ss
->free_super(tst
);
528 if (!st
|| !st
->sb
|| !content
)
531 /* Now need to open the array device. Use create_mddev */
532 if (content
== &info
)
533 st
->ss
->getinfo_super(st
, content
, NULL
);
535 trustworthy
= FOREIGN
;
536 name
= content
->name
;
537 switch (st
->ss
->match_home(st
, homehost
)
538 ?: st
->ss
->match_home(st
, "any")) {
541 name
= strchr(content
->name
, ':');
545 name
= content
->name
;
549 /* If the array is listed in mdadm.conf or on
550 * command line, then we trust the name
551 * even if the array doesn't look local
556 content
->array
.level
== LEVEL_CONTAINER
) {
557 name
= content
->text_version
;
558 trustworthy
= METADATA
;
561 if (name
[0] && trustworthy
!= LOCAL
&&
562 ! require_homehost
&&
563 conf_name_is_free(name
))
566 if (trustworthy
== LOCAL
&&
568 /* Ignore 'host:' prefix of name */
569 name
= strchr(name
, ':')+1;
571 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
574 st
->ss
->free_super(st
);
580 vers
= md_get_version(mdfd
);
582 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
583 " Upgrade your kernel or try --build\n");
587 if (mddev_busy(fd2devnum(mdfd
))) {
588 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
590 for (tmpdev
= devlist
;
591 tmpdev
&& tmpdev
->used
!= 1;
592 tmpdev
= tmpdev
->next
)
594 if (tmpdev
&& auto_assem
)
595 fprintf(stderr
, Name
": %s needed for %s...\n",
596 mddev
, tmpdev
->devname
);
599 st
->ss
->free_super(st
);
604 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
607 if (content
!= &info
) {
608 /* This is a member of a container. Try starting the array. */
609 return assemble_container_content(st
, mdfd
, content
, runstop
,
610 chosen_name
, verbose
);
613 /* Ok, no bad inconsistancy, we can try updating etc */
615 content
->update_private
= NULL
;
616 devices
= malloc(num_devs
* sizeof(*devices
));
617 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
618 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
619 char *devname
= tmpdev
->devname
;
621 /* looks like a good enough match to update the super block if needed */
625 /* prepare useful information in info structures */
627 struct supertype
*tst
;
631 if (strcmp(update
, "uuid")==0 &&
634 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
635 read(rfd
, ident
->uuid
, 16) != 16) {
636 *(__u32
*)(ident
->uuid
) = random();
637 *(__u32
*)(ident
->uuid
+1) = random();
638 *(__u32
*)(ident
->uuid
+2) = random();
639 *(__u32
*)(ident
->uuid
+3) = random();
641 if (rfd
>= 0) close(rfd
);
643 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
645 remove_partitions(dfd
);
648 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
649 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
658 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
660 memcpy(content
->uuid
, ident
->uuid
, 16);
661 strcpy(content
->name
, ident
->name
);
662 content
->array
.md_minor
= minor(stb2
.st_rdev
);
664 if (strcmp(update
, "byteorder") == 0)
667 err
= tst
->ss
->update_super(tst
, content
, update
,
673 Name
": --update=%s not understood"
674 " for %s metadata\n",
675 update
, tst
->ss
->name
);
676 tst
->ss
->free_super(tst
);
684 if (strcmp(update
, "uuid")==0 &&
687 memcpy(ident
->uuid
, content
->uuid
, 16);
689 if (tst
->ss
->store_super(tst
, dfd
))
690 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
694 if (strcmp(update
, "uuid")==0 &&
695 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
696 if (bitmap_update_uuid(ident
->bitmap_fd
,
698 tst
->ss
->swapuuid
) != 0)
699 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
703 tst
->ss
->free_super(tst
);
707 struct supertype
*tst
= dup_super(st
);
709 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
711 remove_partitions(dfd
);
713 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
714 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
723 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
724 tst
->ss
->free_super(tst
);
731 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
732 devname
, mddev
, content
->disk
.raid_disk
);
733 devices
[devcnt
].devname
= devname
;
734 devices
[devcnt
].uptodate
= 0;
735 devices
[devcnt
].i
= *content
;
736 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
737 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
738 if (most_recent
< devcnt
) {
739 if (devices
[devcnt
].i
.events
740 > devices
[most_recent
].i
.events
)
741 most_recent
= devcnt
;
743 if (content
->array
.level
== LEVEL_MULTIPATH
)
744 /* with multipath, the raid_disk from the superblock is meaningless */
747 i
= devices
[devcnt
].i
.disk
.raid_disk
;
749 if (nextspare
< content
->array
.raid_disks
)
750 nextspare
= content
->array
.raid_disks
;
753 if (i
>= content
->array
.raid_disks
&&
759 int newbestcnt
= i
+10;
760 int *newbest
= malloc(sizeof(int)*newbestcnt
);
762 for (c
=0; c
< newbestcnt
; c
++)
764 newbest
[c
] = best
[c
];
769 bestcnt
= newbestcnt
;
772 devices
[best
[i
]].i
.events
773 == devices
[devcnt
].i
.events
774 && (devices
[best
[i
]].i
.disk
.minor
775 != devices
[devcnt
].i
.disk
.minor
)
777 && content
->array
.level
!= LEVEL_MULTIPATH
) {
778 /* two different devices with identical superblock.
779 * Could be a mis-detection caused by overlapping
780 * partitions. fail-safe.
782 fprintf(stderr
, Name
": WARNING %s and %s appear"
783 " to have very similar superblocks.\n"
784 " If they are really different, "
785 "please --zero the superblock on one\n"
786 " If they are the same or overlap,"
787 " please remove one from %s.\n",
788 devices
[best
[i
]].devname
, devname
,
789 inargv
? "the list" :
790 "the\n DEVICE list in mdadm.conf"
798 || (devices
[best
[i
]].i
.events
799 < devices
[devcnt
].i
.events
))
804 free(content
->update_private
);
805 content
->update_private
= NULL
;
808 fprintf(stderr
, Name
": no devices found for %s\n",
811 st
->ss
->free_super(st
);
818 if (update
&& strcmp(update
, "byteorder")==0)
819 st
->minor_version
= 90;
821 st
->ss
->getinfo_super(st
, content
, NULL
);
822 clean
= content
->array
.state
& 1;
824 /* now we have some devices that might be suitable.
827 avail
= malloc(content
->array
.raid_disks
);
828 memset(avail
, 0, content
->array
.raid_disks
);
832 for (i
=0; i
< bestcnt
; i
++) {
834 int event_margin
= 1; /* always allow a difference of '1'
835 * like the kernel does
838 /* note: we ignore error flags in multipath arrays
839 * as they don't make sense
841 if (content
->array
.level
!= LEVEL_MULTIPATH
)
842 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
843 if (!(devices
[j
].i
.disk
.state
844 & (1<<MD_DISK_FAULTY
)))
848 /* If this devices thinks that 'most_recent' has failed, then
849 * we must reject this device.
851 if (j
!= most_recent
&&
852 content
->array
.raid_disks
> 0 &&
853 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
854 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
856 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
857 devices
[j
].devname
, devices
[most_recent
].devname
);
861 if (devices
[j
].i
.events
+event_margin
>=
862 devices
[most_recent
].i
.events
) {
863 devices
[j
].uptodate
= 1;
864 if (i
< content
->array
.raid_disks
) {
865 if (devices
[j
].i
.recovery_start
== MaxSector
) {
875 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
876 content
->array
.layout
, 1,
878 /* Choose the newest best drive which is
879 * not up-to-date, update the superblock
883 struct supertype
*tst
;
884 unsigned long long current_events
;
886 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
889 !devices
[j
].uptodate
&&
890 devices
[j
].i
.recovery_start
== MaxSector
&&
893 > devices
[chosen_drive
].i
.events
))
896 if (chosen_drive
< 0)
898 current_events
= devices
[chosen_drive
].i
.events
;
901 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
902 devices
[chosen_drive
].devname
,
903 devices
[chosen_drive
].i
.disk
.raid_disk
,
904 (int)(devices
[chosen_drive
].i
.events
),
905 (int)(devices
[most_recent
].i
.events
));
906 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
908 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
909 devices
[chosen_drive
].devname
);
910 devices
[chosen_drive
].i
.events
= 0;
914 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
916 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
917 devices
[chosen_drive
].devname
);
918 devices
[chosen_drive
].i
.events
= 0;
921 content
->events
= devices
[most_recent
].i
.events
;
922 tst
->ss
->update_super(tst
, content
, "force-one",
923 devices
[chosen_drive
].devname
, verbose
,
926 if (tst
->ss
->store_super(tst
, fd
)) {
928 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
929 devices
[chosen_drive
].devname
);
930 devices
[chosen_drive
].i
.events
= 0;
931 tst
->ss
->free_super(tst
);
935 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
936 devices
[chosen_drive
].uptodate
= 1;
937 avail
[chosen_drive
] = 1;
939 tst
->ss
->free_super(tst
);
941 /* If there are any other drives of the same vintage,
942 * add them in as well. We can't lose and we might gain
944 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
947 !devices
[j
].uptodate
&&
948 devices
[j
].i
.events
== current_events
) {
955 /* Now we want to look at the superblock which the kernel will base things on
956 * and compare the devices that we think are working with the devices that the
957 * superblock thinks are working.
958 * If there are differences and --force is given, then update this chosen
962 st
->ss
->free_super(st
);
963 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
969 if (!devices
[j
].uptodate
)
972 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
973 fprintf(stderr
, Name
": Cannot open %s: %s\n",
974 devices
[j
].devname
, strerror(errno
));
979 if (st
->ss
->load_super(st
,fd
, NULL
)) {
981 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
989 if (st
->sb
== NULL
) {
990 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
995 st
->ss
->getinfo_super(st
, content
, NULL
);
997 sysfs_init(content
, mdfd
, 0);
999 for (i
=0; i
<bestcnt
; i
++) {
1001 unsigned int desired_state
;
1003 if (i
< content
->array
.raid_disks
)
1004 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1010 if (!devices
[j
].uptodate
)
1013 devices
[j
].i
.disk
.state
= desired_state
;
1014 if (!(devices
[j
].i
.array
.state
& 1))
1017 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1018 verbose
, 0, NULL
)) {
1021 fprintf(stderr
, Name
": "
1022 "clearing FAULTY flag for device %d in %s for %s\n",
1023 j
, mddev
, devices
[j
].devname
);
1027 fprintf(stderr
, Name
": "
1028 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1029 i
, mddev
, devices
[j
].devname
);
1033 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1034 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1039 if (force
&& !clean
&&
1040 !enough(content
->array
.level
, content
->array
.raid_disks
,
1041 content
->array
.layout
, clean
,
1043 change
+= st
->ss
->update_super(st
, content
, "force-array",
1044 devices
[chosen_drive
].devname
, verbose
,
1051 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1053 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1054 devices
[chosen_drive
].devname
);
1059 if (st
->ss
->store_super(st
, fd
)) {
1061 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1062 devices
[chosen_drive
].devname
);
1070 /* If we are in the middle of a reshape we may need to restore saved data
1071 * that was moved aside due to the reshape overwriting live data
1072 * The code of doing this lives in Grow.c
1075 if (content
->reshape_active
) {
1077 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1079 fprintf(stderr
, Name
":%s has an active reshape - checking "
1080 "if critical section needs to be restored\n",
1082 for (i
=0; i
<bestcnt
; i
++) {
1085 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1086 if (fdlist
[i
] < 0) {
1087 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1088 devices
[j
].devname
);
1096 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1099 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1102 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1103 if (backup_file
== NULL
)
1104 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1111 /* count number of in-sync devices according to the superblock.
1112 * We must have this number to start the array without -s or -R
1114 req_cnt
= content
->array
.working_disks
;
1116 /* Almost ready to actually *do* something */
1120 /* First, fill in the map, so that udev can find our name
1121 * as soon as we become active.
1123 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1124 content
->uuid
, chosen_name
);
1126 rv
= set_array_info(mdfd
, st
, content
);
1128 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1129 mddev
, strerror(errno
));
1130 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1135 if (ident
->bitmap_fd
>= 0) {
1136 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1137 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1138 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1143 } else if (ident
->bitmap_file
) {
1144 /* From config file */
1145 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1147 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1148 ident
->bitmap_file
);
1149 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1154 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1155 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1157 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1165 /* First, add the raid disks, but add the chosen one last */
1166 for (i
=0; i
<= bestcnt
; i
++) {
1170 if (j
== chosen_drive
)
1175 if (j
>= 0 /* && devices[j].uptodate */) {
1176 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1179 fprintf(stderr
, Name
": failed to add "
1184 if (i
< content
->array
.raid_disks
1189 } else if (verbose
> 0)
1190 fprintf(stderr
, Name
": added %s "
1192 devices
[j
].devname
, mddev
,
1193 devices
[j
].i
.disk
.raid_disk
);
1194 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1195 fprintf(stderr
, Name
": no uptodate device for "
1200 if (content
->array
.level
== LEVEL_CONTAINER
) {
1202 fprintf(stderr
, Name
": Container %s has been "
1203 "assembled with %d drive%s",
1204 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1205 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1206 fprintf(stderr
, " (out of %d)",
1207 content
->array
.raid_disks
);
1208 fprintf(stderr
, "\n");
1210 sysfs_uevent(content
, "change");
1211 wait_for(chosen_name
, mdfd
);
1219 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1220 content
->array
.layout
, clean
, avail
, okcnt
) &&
1221 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1223 /* This array is good-to-go.
1224 * If a reshape is in progress then we might need to
1225 * continue monitoring it. In that case we start
1226 * it read-only and let the grow code make it writable.
1230 if (content
->reshape_active
&&
1231 content
->delta_disks
<= 0)
1232 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1235 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1238 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1239 mddev
, okcnt
, okcnt
==1?"":"s");
1240 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1241 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1243 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1245 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1246 fprintf(stderr
, ".\n");
1248 if (content
->reshape_active
&&
1249 content
->array
.level
>= 4 &&
1250 content
->array
.level
<= 6) {
1251 /* might need to increase the size
1252 * of the stripe cache - default is 256
1254 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1255 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1257 sysfs_set_num(sra
, NULL
,
1258 "stripe_cache_size",
1259 (4 * content
->array
.chunk_size
/ 4096) + 1);
1262 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1263 /* If any devices did not get added
1264 * because the kernel rejected them based
1265 * on event count, try adding them
1266 * again providing the action policy is
1267 * 're-add' or greater. The bitmap
1268 * might allow them to be included, or
1269 * they will become spares.
1271 for (i
= 0; i
<= bestcnt
; i
++) {
1273 if (j
>= 0 && !devices
[j
].uptodate
) {
1274 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1276 rv
= add_disk(mdfd
, st
, content
,
1278 if (rv
== 0 && verbose
>= 0)
1280 Name
": %s has been re-added.\n",
1281 devices
[j
].devname
);
1285 wait_for(mddev
, mdfd
);
1289 /* There is a nasty race with 'mdadm --monitor'.
1290 * If it opens this device before we close it,
1291 * it gets an incomplete open on which IO
1292 * doesn't work and the capacity is
1294 * If we reopen (to check for layered devices)
1295 * before --monitor closes, we loose.
1297 * So: wait upto 1 second for there to be
1298 * a non-zero capacity.
1300 while (usecs
< 1000) {
1301 mdfd
= open(mddev
, O_RDONLY
);
1303 unsigned long long size
;
1304 if (get_dev_size(mdfd
, NULL
, &size
) &&
1316 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1317 mddev
, strerror(errno
));
1319 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1320 content
->array
.layout
, 1, avail
, okcnt
))
1321 fprintf(stderr
, Name
": Not enough devices to "
1322 "start the array.\n");
1323 else if (!enough(content
->array
.level
,
1324 content
->array
.raid_disks
,
1325 content
->array
.layout
, clean
,
1327 fprintf(stderr
, Name
": Not enough devices to "
1328 "start the array while not clean "
1329 "- consider --force.\n");
1332 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1337 if (runstop
== -1) {
1338 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1339 mddev
, okcnt
, okcnt
==1?"":"s");
1340 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1341 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1342 fprintf(stderr
, ", but not started.\n");
1347 if (verbose
>= -1) {
1348 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1350 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1352 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1353 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1354 content
->array
.layout
, 1, avail
, okcnt
))
1355 fprintf(stderr
, " - not enough to start the array.\n");
1356 else if (!enough(content
->array
.level
,
1357 content
->array
.raid_disks
,
1358 content
->array
.layout
, clean
,
1360 fprintf(stderr
, " - not enough to start the "
1361 "array while not clean - consider "
1364 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1365 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1367 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1368 fprintf(stderr
, " (use --run to insist).\n");
1372 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1377 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1378 * been updated to point to the current locations of devices.
1379 * so we can just start the array
1382 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1383 devices
[chosen_drive
].i
.disk
.minor
);
1384 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1385 fprintf(stderr
, Name
": Cannot start array: %s\n",
1396 int assemble_container_content(struct supertype
*st
, int mdfd
,
1397 struct mdinfo
*content
, int runstop
,
1398 char *chosen_name
, int verbose
)
1400 struct mdinfo
*dev
, *sra
;
1401 int working
= 0, preexist
= 0;
1402 struct map_ent
*map
= NULL
;
1404 sysfs_init(content
, mdfd
, 0);
1406 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1407 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1408 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1415 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1416 if (sysfs_add_disk(content
, dev
, 1) == 0)
1418 else if (errno
== EEXIST
)
1422 return 1;/* Nothing new, don't try to start */
1425 map_update(&map
, fd2devnum(mdfd
),
1426 content
->text_version
,
1427 content
->uuid
, chosen_name
);
1430 (working
+ preexist
) >= content
->array
.working_disks
) {
1433 switch(content
->array
.level
) {
1435 case LEVEL_MULTIPATH
:
1437 err
= sysfs_set_str(content
, NULL
, "array_state",
1441 err
= sysfs_set_str(content
, NULL
, "array_state",
1443 /* start mdmon if needed. */
1445 if (!mdmon_running(st
->container_dev
))
1446 start_mdmon(st
->container_dev
);
1447 ping_monitor(devnum2devname(st
->container_dev
));
1452 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1455 fprintf(stderr
, Name
1456 ": array %s now has %d devices",
1457 chosen_name
, working
+ preexist
);
1459 fprintf(stderr
, Name
1460 ": Started %s with %d devices",
1461 chosen_name
, working
+ preexist
);
1463 fprintf(stderr
, " (%d new)", working
);
1464 fprintf(stderr
, "\n");
1467 wait_for(chosen_name
, mdfd
);
1470 /* FIXME should have an O_EXCL and wait for read-auto */
1473 fprintf(stderr
, Name
1474 ": %s assembled with %d devices but "
1476 chosen_name
, working
);