]>
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",
115 if (ident
->member
&& ident
->member
[0]) {
116 /* content->text_version must match */
117 char *s
= strchr(content
->text_version
+1, '/');
120 fprintf(stderr
, Name
": %s is not a container and one is required.\n",
123 } else if (strcmp(ident
->member
, s
+1) != 0) {
125 fprintf(stderr
, Name
": skipping wrong member %s is %s\n",
126 content
->text_version
, devname
);
134 int Assemble(struct supertype
*st
, char *mddev
,
135 struct mddev_ident
*ident
,
136 struct mddev_dev
*devlist
, char *backup_file
,
137 int readonly
, int runstop
,
138 char *update
, char *homehost
, int require_homehost
,
139 int verbose
, int force
)
142 * The task of Assemble is to find a collection of
143 * devices that should (according to their superblocks)
144 * form an array, and to give this collection to the MD driver.
145 * In Linux-2.4 and later, this involves submitting a
146 * SET_ARRAY_INFO ioctl with no arg - to prepare
147 * the array - and then submit a number of
148 * ADD_NEW_DISK ioctls to add disks into
149 * the array. Finally RUN_ARRAY might
150 * be submitted to start the array.
152 * Much of the work of Assemble is in finding and/or
153 * checking the disks to make sure they look right.
155 * If mddev is not set, then scan must be set and we
156 * read through the config file for dev+uuid mapping
157 * We recurse, setting mddev, for each device that
159 * - has a valid uuid (or any uuid if !uuidset)
161 * If mddev is set, we try to determine state of md.
162 * check version - must be at least 0.90.0
163 * check kernel version. must be at least 2.4.
164 * If not, we can possibly fall back on START_ARRAY
165 * Try to GET_ARRAY_INFO.
166 * If possible, give up
167 * If not, try to STOP_ARRAY just to make sure
169 * If !uuidset and scan, look in conf-file for uuid
170 * If not found, give up
171 * If !devlist and scan and uuidset, get list of devs from conf-file
174 * Check superblock - discard if bad
175 * Check uuid (set if we don't have one) - discard if no match
176 * Check superblock similarity if we have a superblock - discard if different
177 * Record events, devicenum
178 * This should give us a list of devices for the array
179 * We should collect the most recent event number
181 * Count disks with recent enough event count
182 * While force && !enough disks
183 * Choose newest rejected disks, update event count
184 * mark clean and rewrite superblock
187 * foreach device with recent events : ADD_NEW_DISK
188 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
190 * Check the device numbers in superblock are right
191 * update superblock if any changes
197 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
198 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
199 && (ident
->container
== NULL
|| ident
->member
== NULL
));
201 int vers
= vers
; /* Keep gcc quite - it really is initialised */
204 int uptodate
; /* set once we decide that this device is as
205 * recent as everything else in the array.
210 int *best
= NULL
; /* indexed by raid_disk */
213 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
214 unsigned int req_cnt
;
220 int report_missmatch
;
222 int start_partial_ok
= (runstop
>= 0) &&
223 (force
|| devlist
==NULL
|| auto_assem
);
224 unsigned int num_devs
;
225 struct mddev_dev
*tmpdev
;
227 struct mdinfo
*content
= NULL
;
232 char chosen_name
[1024];
234 if (get_linux_version() < 2004000)
238 * If any subdevs are listed, then any that don't
239 * match ident are discarded. Remainder must all match and
241 * If no subdevs, then we scan all devices in the config file, but
242 * there must be something in the identity
246 ident
->uuid_set
== 0 &&
247 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
248 ident
->name
[0] == 0 &&
249 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
250 ident
->devices
== NULL
) {
251 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
252 mddev
? mddev
: "further assembly");
257 devlist
= conf_get_devs();
261 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
263 /* We come back here when doing auto-assembly and attempting some
264 * set of devices failed. Those are now marked as ->used==2 and
265 * we ignore them and try again
268 tmpdev
= devlist
; num_devs
= 0;
274 tmpdev
= tmpdev
->next
;
277 if (!st
&& ident
->st
) st
= ident
->st
;
280 fprintf(stderr
, Name
": looking for devices for %s\n",
281 mddev
? mddev
: "further assembly");
283 /* first walk the list of devices to find a consistent set
284 * that match the criterea, if that is possible.
285 * We flag the ones we like with 'used'.
287 for (tmpdev
= devlist
;
289 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
290 char *devname
= tmpdev
->devname
;
293 struct supertype
*tst
= dup_super(st
);
294 struct dev_policy
*pol
= NULL
;
295 int found_container
= 0;
297 if (tmpdev
->used
> 1) continue;
299 if (ident
->devices
&&
300 !match_oneof(ident
->devices
, devname
)) {
301 if (report_missmatch
)
302 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
306 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
308 if (report_missmatch
)
309 fprintf(stderr
, Name
": cannot open device %s: %s\n",
310 devname
, strerror(errno
));
312 } else if (fstat(dfd
, &stb
)< 0) {
314 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
315 devname
, strerror(errno
));
317 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
318 fprintf(stderr
, Name
": %s is not a block device.\n",
321 } else if (must_be_container(dfd
)) {
323 /* already found some components, this cannot
326 if (report_missmatch
)
327 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
330 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
331 if (report_missmatch
)
332 fprintf(stderr
, Name
": not a recognisable container: %s\n",
335 } else if (tst
->ss
->load_container(tst
, dfd
, NULL
)) {
336 if (report_missmatch
)
337 fprintf(stderr
, Name
": no correct container type: %s\n",
340 } else if (auto_assem
&&
341 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
342 tst
->ss
->match_home(tst
, homehost
) == 1)) {
343 if (report_missmatch
)
344 fprintf(stderr
, Name
": %s has metadata type %s for which "
345 "auto-assembly is disabled\n",
346 devname
, tst
->ss
->name
);
351 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
352 if (report_missmatch
)
353 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
356 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
357 if (report_missmatch
)
358 fprintf(stderr
, Name
": no RAID superblock on %s\n",
361 } else if (tst
->ss
->compare_super
== NULL
) {
362 if (report_missmatch
)
363 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
364 tst
->ss
->name
, devname
);
366 } else if (auto_assem
&& st
== NULL
&&
367 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
368 tst
->ss
->match_home(tst
, homehost
) == 1)) {
369 if (report_missmatch
)
370 fprintf(stderr
, Name
": %s has metadata type %s for which "
371 "auto-assembly is disabled\n",
372 devname
, tst
->ss
->name
);
376 if (dfd
>= 0) close(dfd
);
377 if (tmpdev
->used
== 2) {
379 /* Ignore unrecognised devices during auto-assembly */
381 if (ident
->uuid_set
|| ident
->name
[0] ||
382 ident
->super_minor
!= UnSet
)
383 /* Ignore unrecognised device if looking for
388 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
391 st
->ss
->free_super(st
);
392 dev_policy_free(pol
);
396 if (found_container
) {
397 /* tmpdev is a container. We need to be either
398 * looking for a member, or auto-assembling
401 if (ident
->container
) {
402 if (ident
->container
[0] == '/' &&
403 !same_dev(ident
->container
, devname
)) {
404 if (report_missmatch
)
405 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
406 devname
, ident
->container
);
409 if (ident
->container
[0] != '/') {
412 if (!parse_uuid(ident
->container
, uuid
) ||
413 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
414 if (report_missmatch
)
415 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
421 /* It is worth looking inside this container.
424 fprintf(stderr
, Name
": looking in container %s\n",
427 for (content
= tst
->ss
->container_content(tst
, NULL
);
429 content
= content
->next
) {
431 if (!ident_matches(ident
, content
, tst
,
433 report_missmatch
? devname
: NULL
))
434 /* message already printed */;
435 else if (is_member_busy(content
->text_version
)) {
436 if (report_missmatch
)
437 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
438 content
->text_version
,
445 goto loop
; /* empty container */
448 st
= tst
; tst
= NULL
;
449 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
450 fprintf(stderr
, Name
": %s is a container, but is not "
451 "only device given: confused and aborting\n",
453 st
->ss
->free_super(st
);
454 dev_policy_free(pol
);
458 fprintf(stderr
, Name
": found match on member %s in %s\n",
459 content
->text_version
, devname
);
461 /* make sure we finished the loop */
467 memset(content
, 0, sizeof(*content
));
468 tst
->ss
->getinfo_super(tst
, content
, NULL
);
470 if (!ident_matches(ident
, content
, tst
,
472 report_missmatch
? devname
: NULL
))
477 if (st
->minor_version
== -1)
478 st
->minor_version
= tst
->minor_version
;
479 if (st
->ss
!= tst
->ss
||
480 st
->minor_version
!= tst
->minor_version
||
481 st
->ss
->compare_super(st
, tst
) != 0) {
482 /* Some mismatch. If exactly one array matches this host,
483 * we can resolve on that one.
484 * Or, if we are auto assembling, we just ignore the second
490 int first
= st
->ss
->match_home(st
, homehost
);
491 int last
= tst
->ss
->match_home(tst
, homehost
);
493 (first
== 1 || last
== 1)) {
494 /* We can do something */
495 if (first
) {/* just ignore this one */
496 if (report_missmatch
)
497 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
500 } else { /* reject all those sofar */
501 struct mddev_dev
*td
;
502 if (report_missmatch
)
503 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
505 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
513 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
515 tst
->ss
->free_super(tst
);
516 st
->ss
->free_super(st
);
517 dev_policy_free(pol
);
523 dev_policy_free(pol
);
526 tst
->ss
->free_super(tst
);
529 if (!st
|| !st
->sb
|| !content
)
532 /* Now need to open the array device. Use create_mddev */
533 if (content
== &info
)
534 st
->ss
->getinfo_super(st
, content
, NULL
);
536 trustworthy
= FOREIGN
;
537 name
= content
->name
;
538 switch (st
->ss
->match_home(st
, homehost
)
539 ?: st
->ss
->match_home(st
, "any")) {
542 name
= strchr(content
->name
, ':');
546 name
= content
->name
;
550 /* If the array is listed in mdadm.conf or on
551 * command line, then we trust the name
552 * even if the array doesn't look local
557 content
->array
.level
== LEVEL_CONTAINER
) {
558 name
= content
->text_version
;
559 trustworthy
= METADATA
;
562 if (name
[0] && trustworthy
!= LOCAL
&&
563 ! require_homehost
&&
564 conf_name_is_free(name
))
567 if (trustworthy
== LOCAL
&&
569 /* Ignore 'host:' prefix of name */
570 name
= strchr(name
, ':')+1;
572 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
575 st
->ss
->free_super(st
);
581 vers
= md_get_version(mdfd
);
583 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
584 " Upgrade your kernel or try --build\n");
588 if (mddev_busy(fd2devnum(mdfd
))) {
589 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
591 for (tmpdev
= devlist
;
592 tmpdev
&& tmpdev
->used
!= 1;
593 tmpdev
= tmpdev
->next
)
595 if (tmpdev
&& auto_assem
)
596 fprintf(stderr
, Name
": %s needed for %s...\n",
597 mddev
, tmpdev
->devname
);
600 st
->ss
->free_super(st
);
605 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
608 if (content
!= &info
) {
609 /* This is a member of a container. Try starting the array. */
610 return assemble_container_content(st
, mdfd
, content
, runstop
,
611 chosen_name
, verbose
);
614 /* Ok, no bad inconsistancy, we can try updating etc */
616 content
->update_private
= NULL
;
617 devices
= malloc(num_devs
* sizeof(*devices
));
618 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
619 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
620 char *devname
= tmpdev
->devname
;
622 /* looks like a good enough match to update the super block if needed */
626 /* prepare useful information in info structures */
628 struct supertype
*tst
;
632 if (strcmp(update
, "uuid")==0 &&
635 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
636 read(rfd
, ident
->uuid
, 16) != 16) {
637 *(__u32
*)(ident
->uuid
) = random();
638 *(__u32
*)(ident
->uuid
+1) = random();
639 *(__u32
*)(ident
->uuid
+2) = random();
640 *(__u32
*)(ident
->uuid
+3) = random();
642 if (rfd
>= 0) close(rfd
);
644 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
646 remove_partitions(dfd
);
649 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
650 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
659 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
661 memcpy(content
->uuid
, ident
->uuid
, 16);
662 strcpy(content
->name
, ident
->name
);
663 content
->array
.md_minor
= minor(stb2
.st_rdev
);
665 if (strcmp(update
, "byteorder") == 0)
668 err
= tst
->ss
->update_super(tst
, content
, update
,
674 Name
": --update=%s not understood"
675 " for %s metadata\n",
676 update
, tst
->ss
->name
);
677 tst
->ss
->free_super(tst
);
685 if (strcmp(update
, "uuid")==0 &&
688 memcpy(ident
->uuid
, content
->uuid
, 16);
690 if (tst
->ss
->store_super(tst
, dfd
))
691 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
695 if (strcmp(update
, "uuid")==0 &&
696 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
697 if (bitmap_update_uuid(ident
->bitmap_fd
,
699 tst
->ss
->swapuuid
) != 0)
700 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
704 tst
->ss
->free_super(tst
);
708 struct supertype
*tst
= dup_super(st
);
710 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
712 remove_partitions(dfd
);
714 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
715 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
724 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
725 tst
->ss
->free_super(tst
);
732 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
733 devname
, mddev
, content
->disk
.raid_disk
);
734 devices
[devcnt
].devname
= devname
;
735 devices
[devcnt
].uptodate
= 0;
736 devices
[devcnt
].i
= *content
;
737 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
738 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
739 if (most_recent
< devcnt
) {
740 if (devices
[devcnt
].i
.events
741 > devices
[most_recent
].i
.events
)
742 most_recent
= devcnt
;
744 if (content
->array
.level
== LEVEL_MULTIPATH
)
745 /* with multipath, the raid_disk from the superblock is meaningless */
748 i
= devices
[devcnt
].i
.disk
.raid_disk
;
750 if (nextspare
< content
->array
.raid_disks
)
751 nextspare
= content
->array
.raid_disks
;
754 if (i
>= content
->array
.raid_disks
&&
760 int newbestcnt
= i
+10;
761 int *newbest
= malloc(sizeof(int)*newbestcnt
);
763 for (c
=0; c
< newbestcnt
; c
++)
765 newbest
[c
] = best
[c
];
770 bestcnt
= newbestcnt
;
773 devices
[best
[i
]].i
.events
774 == devices
[devcnt
].i
.events
775 && (devices
[best
[i
]].i
.disk
.minor
776 != devices
[devcnt
].i
.disk
.minor
)
778 && content
->array
.level
!= LEVEL_MULTIPATH
) {
779 /* two different devices with identical superblock.
780 * Could be a mis-detection caused by overlapping
781 * partitions. fail-safe.
783 fprintf(stderr
, Name
": WARNING %s and %s appear"
784 " to have very similar superblocks.\n"
785 " If they are really different, "
786 "please --zero the superblock on one\n"
787 " If they are the same or overlap,"
788 " please remove one from %s.\n",
789 devices
[best
[i
]].devname
, devname
,
790 inargv
? "the list" :
791 "the\n DEVICE list in mdadm.conf"
799 || (devices
[best
[i
]].i
.events
800 < devices
[devcnt
].i
.events
))
805 free(content
->update_private
);
806 content
->update_private
= NULL
;
809 fprintf(stderr
, Name
": no devices found for %s\n",
812 st
->ss
->free_super(st
);
819 if (update
&& strcmp(update
, "byteorder")==0)
820 st
->minor_version
= 90;
822 st
->ss
->getinfo_super(st
, content
, NULL
);
823 clean
= content
->array
.state
& 1;
825 /* now we have some devices that might be suitable.
828 avail
= malloc(content
->array
.raid_disks
);
829 memset(avail
, 0, content
->array
.raid_disks
);
833 for (i
=0; i
< bestcnt
; i
++) {
835 int event_margin
= 1; /* always allow a difference of '1'
836 * like the kernel does
839 /* note: we ignore error flags in multipath arrays
840 * as they don't make sense
842 if (content
->array
.level
!= LEVEL_MULTIPATH
)
843 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
844 if (!(devices
[j
].i
.disk
.state
845 & (1<<MD_DISK_FAULTY
)))
849 /* If this devices thinks that 'most_recent' has failed, then
850 * we must reject this device.
852 if (j
!= most_recent
&&
853 content
->array
.raid_disks
> 0 &&
854 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
855 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
857 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
858 devices
[j
].devname
, devices
[most_recent
].devname
);
862 if (devices
[j
].i
.events
+event_margin
>=
863 devices
[most_recent
].i
.events
) {
864 devices
[j
].uptodate
= 1;
865 if (i
< content
->array
.raid_disks
) {
866 if (devices
[j
].i
.recovery_start
== MaxSector
) {
876 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
877 content
->array
.layout
, 1,
879 /* Choose the newest best drive which is
880 * not up-to-date, update the superblock
884 struct supertype
*tst
;
885 unsigned long long current_events
;
887 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
890 !devices
[j
].uptodate
&&
891 devices
[j
].i
.recovery_start
== MaxSector
&&
894 > devices
[chosen_drive
].i
.events
))
897 if (chosen_drive
< 0)
899 current_events
= devices
[chosen_drive
].i
.events
;
902 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
903 devices
[chosen_drive
].devname
,
904 devices
[chosen_drive
].i
.disk
.raid_disk
,
905 (int)(devices
[chosen_drive
].i
.events
),
906 (int)(devices
[most_recent
].i
.events
));
907 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
909 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
910 devices
[chosen_drive
].devname
);
911 devices
[chosen_drive
].i
.events
= 0;
915 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
917 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
918 devices
[chosen_drive
].devname
);
919 devices
[chosen_drive
].i
.events
= 0;
922 content
->events
= devices
[most_recent
].i
.events
;
923 tst
->ss
->update_super(tst
, content
, "force-one",
924 devices
[chosen_drive
].devname
, verbose
,
927 if (tst
->ss
->store_super(tst
, fd
)) {
929 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
930 devices
[chosen_drive
].devname
);
931 devices
[chosen_drive
].i
.events
= 0;
932 tst
->ss
->free_super(tst
);
936 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
937 devices
[chosen_drive
].uptodate
= 1;
938 avail
[chosen_drive
] = 1;
940 tst
->ss
->free_super(tst
);
942 /* If there are any other drives of the same vintage,
943 * add them in as well. We can't lose and we might gain
945 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
948 !devices
[j
].uptodate
&&
949 devices
[j
].i
.events
== current_events
) {
956 /* Now we want to look at the superblock which the kernel will base things on
957 * and compare the devices that we think are working with the devices that the
958 * superblock thinks are working.
959 * If there are differences and --force is given, then update this chosen
963 st
->ss
->free_super(st
);
964 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
970 if (!devices
[j
].uptodate
)
973 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
974 fprintf(stderr
, Name
": Cannot open %s: %s\n",
975 devices
[j
].devname
, strerror(errno
));
980 if (st
->ss
->load_super(st
,fd
, NULL
)) {
982 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
990 if (st
->sb
== NULL
) {
991 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
996 st
->ss
->getinfo_super(st
, content
, NULL
);
998 sysfs_init(content
, mdfd
, 0);
1000 for (i
=0; i
<bestcnt
; i
++) {
1002 unsigned int desired_state
;
1004 if (i
< content
->array
.raid_disks
)
1005 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1011 if (!devices
[j
].uptodate
)
1014 devices
[j
].i
.disk
.state
= desired_state
;
1015 if (!(devices
[j
].i
.array
.state
& 1))
1018 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1019 verbose
, 0, NULL
)) {
1022 fprintf(stderr
, Name
": "
1023 "clearing FAULTY flag for device %d in %s for %s\n",
1024 j
, mddev
, devices
[j
].devname
);
1028 fprintf(stderr
, Name
": "
1029 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1030 i
, mddev
, devices
[j
].devname
);
1034 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1035 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1040 if (force
&& !clean
&&
1041 !enough(content
->array
.level
, content
->array
.raid_disks
,
1042 content
->array
.layout
, clean
,
1044 change
+= st
->ss
->update_super(st
, content
, "force-array",
1045 devices
[chosen_drive
].devname
, verbose
,
1052 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1054 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1055 devices
[chosen_drive
].devname
);
1060 if (st
->ss
->store_super(st
, fd
)) {
1062 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1063 devices
[chosen_drive
].devname
);
1071 /* If we are in the middle of a reshape we may need to restore saved data
1072 * that was moved aside due to the reshape overwriting live data
1073 * The code of doing this lives in Grow.c
1076 if (content
->reshape_active
) {
1078 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1080 fprintf(stderr
, Name
":%s has an active reshape - checking "
1081 "if critical section needs to be restored\n",
1083 for (i
=0; i
<bestcnt
; i
++) {
1086 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1087 if (fdlist
[i
] < 0) {
1088 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1089 devices
[j
].devname
);
1097 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
, verbose
> 0);
1100 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1103 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1104 if (backup_file
== NULL
)
1105 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1112 /* count number of in-sync devices according to the superblock.
1113 * We must have this number to start the array without -s or -R
1115 req_cnt
= content
->array
.working_disks
;
1117 /* Almost ready to actually *do* something */
1121 /* First, fill in the map, so that udev can find our name
1122 * as soon as we become active.
1124 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1125 content
->uuid
, chosen_name
);
1127 rv
= set_array_info(mdfd
, st
, content
);
1129 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1130 mddev
, strerror(errno
));
1131 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1136 if (ident
->bitmap_fd
>= 0) {
1137 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1138 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1139 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1144 } else if (ident
->bitmap_file
) {
1145 /* From config file */
1146 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1148 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1149 ident
->bitmap_file
);
1150 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1155 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1156 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1158 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1166 /* First, add the raid disks, but add the chosen one last */
1167 for (i
=0; i
<= bestcnt
; i
++) {
1171 if (j
== chosen_drive
)
1176 if (j
>= 0 /* && devices[j].uptodate */) {
1177 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1180 fprintf(stderr
, Name
": failed to add "
1185 if (i
< content
->array
.raid_disks
1190 } else if (verbose
> 0)
1191 fprintf(stderr
, Name
": added %s "
1193 devices
[j
].devname
, mddev
,
1194 devices
[j
].i
.disk
.raid_disk
);
1195 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1196 fprintf(stderr
, Name
": no uptodate device for "
1201 if (content
->array
.level
== LEVEL_CONTAINER
) {
1203 fprintf(stderr
, Name
": Container %s has been "
1204 "assembled with %d drive%s",
1205 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1206 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1207 fprintf(stderr
, " (out of %d)",
1208 content
->array
.raid_disks
);
1209 fprintf(stderr
, "\n");
1211 sysfs_uevent(content
, "change");
1212 wait_for(chosen_name
, mdfd
);
1220 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1221 content
->array
.layout
, clean
, avail
, okcnt
) &&
1222 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1224 /* This array is good-to-go.
1225 * If a reshape is in progress then we might need to
1226 * continue monitoring it. In that case we start
1227 * it read-only and let the grow code make it writable.
1231 if (content
->reshape_active
&&
1232 content
->delta_disks
<= 0)
1233 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1236 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1239 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1240 mddev
, okcnt
, okcnt
==1?"":"s");
1241 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1242 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1244 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1246 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1247 fprintf(stderr
, ".\n");
1249 if (content
->reshape_active
&&
1250 content
->array
.level
>= 4 &&
1251 content
->array
.level
<= 6) {
1252 /* might need to increase the size
1253 * of the stripe cache - default is 256
1255 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1256 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1258 sysfs_set_num(sra
, NULL
,
1259 "stripe_cache_size",
1260 (4 * content
->array
.chunk_size
/ 4096) + 1);
1263 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1264 /* If any devices did not get added
1265 * because the kernel rejected them based
1266 * on event count, try adding them
1267 * again providing the action policy is
1268 * 're-add' or greater. The bitmap
1269 * might allow them to be included, or
1270 * they will become spares.
1272 for (i
= 0; i
<= bestcnt
; i
++) {
1274 if (j
>= 0 && !devices
[j
].uptodate
) {
1275 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1277 rv
= add_disk(mdfd
, st
, content
,
1279 if (rv
== 0 && verbose
>= 0)
1281 Name
": %s has been re-added.\n",
1282 devices
[j
].devname
);
1286 wait_for(mddev
, mdfd
);
1290 /* There is a nasty race with 'mdadm --monitor'.
1291 * If it opens this device before we close it,
1292 * it gets an incomplete open on which IO
1293 * doesn't work and the capacity is
1295 * If we reopen (to check for layered devices)
1296 * before --monitor closes, we loose.
1298 * So: wait upto 1 second for there to be
1299 * a non-zero capacity.
1301 while (usecs
< 1000) {
1302 mdfd
= open(mddev
, O_RDONLY
);
1304 unsigned long long size
;
1305 if (get_dev_size(mdfd
, NULL
, &size
) &&
1317 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1318 mddev
, strerror(errno
));
1320 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1321 content
->array
.layout
, 1, avail
, okcnt
))
1322 fprintf(stderr
, Name
": Not enough devices to "
1323 "start the array.\n");
1324 else if (!enough(content
->array
.level
,
1325 content
->array
.raid_disks
,
1326 content
->array
.layout
, clean
,
1328 fprintf(stderr
, Name
": Not enough devices to "
1329 "start the array while not clean "
1330 "- consider --force.\n");
1333 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1338 if (runstop
== -1) {
1339 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1340 mddev
, okcnt
, okcnt
==1?"":"s");
1341 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1342 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1343 fprintf(stderr
, ", but not started.\n");
1348 if (verbose
>= -1) {
1349 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1351 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1353 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1354 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1355 content
->array
.layout
, 1, avail
, okcnt
))
1356 fprintf(stderr
, " - not enough to start the array.\n");
1357 else if (!enough(content
->array
.level
,
1358 content
->array
.raid_disks
,
1359 content
->array
.layout
, clean
,
1361 fprintf(stderr
, " - not enough to start the "
1362 "array while not clean - consider "
1365 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1366 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1368 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1369 fprintf(stderr
, " (use --run to insist).\n");
1373 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1378 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1379 * been updated to point to the current locations of devices.
1380 * so we can just start the array
1383 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1384 devices
[chosen_drive
].i
.disk
.minor
);
1385 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1386 fprintf(stderr
, Name
": Cannot start array: %s\n",
1397 int assemble_container_content(struct supertype
*st
, int mdfd
,
1398 struct mdinfo
*content
, int runstop
,
1399 char *chosen_name
, int verbose
)
1401 struct mdinfo
*dev
, *sra
;
1402 int working
= 0, preexist
= 0;
1403 struct map_ent
*map
= NULL
;
1405 sysfs_init(content
, mdfd
, 0);
1407 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1408 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1409 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1416 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1417 if (sysfs_add_disk(content
, dev
, 1) == 0)
1419 else if (errno
== EEXIST
)
1423 return 1;/* Nothing new, don't try to start */
1426 map_update(&map
, fd2devnum(mdfd
),
1427 content
->text_version
,
1428 content
->uuid
, chosen_name
);
1431 (working
+ preexist
) >= content
->array
.working_disks
) {
1434 switch(content
->array
.level
) {
1436 case LEVEL_MULTIPATH
:
1438 err
= sysfs_set_str(content
, NULL
, "array_state",
1442 err
= sysfs_set_str(content
, NULL
, "array_state",
1444 /* start mdmon if needed. */
1446 if (!mdmon_running(st
->container_dev
))
1447 start_mdmon(st
->container_dev
);
1448 ping_monitor(devnum2devname(st
->container_dev
));
1453 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1456 fprintf(stderr
, Name
1457 ": array %s now has %d devices",
1458 chosen_name
, working
+ preexist
);
1460 fprintf(stderr
, Name
1461 ": Started %s with %d devices",
1462 chosen_name
, working
+ preexist
);
1464 fprintf(stderr
, " (%d new)", working
);
1465 fprintf(stderr
, "\n");
1468 wait_for(chosen_name
, mdfd
);
1471 /* FIXME should have an O_EXCL and wait for read-auto */
1474 fprintf(stderr
, Name
1475 ": %s assembled with %d devices but "
1477 chosen_name
, working
);