]>
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 &&
82 memcmp(content
->uuid
, uuid_zero
, sizeof(int[4])) != 0) {
84 fprintf(stderr
, Name
": %s has wrong uuid.\n",
88 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
89 name_matches(content
->name
, ident
->name
, homehost
)==0) {
91 fprintf(stderr
, Name
": %s has wrong name.\n",
95 if (ident
->super_minor
!= UnSet
&&
96 ident
->super_minor
!= content
->array
.md_minor
) {
98 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
102 if (ident
->level
!= UnSet
&&
103 ident
->level
!= content
->array
.level
) {
105 fprintf(stderr
, Name
": %s has wrong raid level.\n",
109 if (ident
->raid_disks
!= UnSet
&&
110 ident
->raid_disks
!= content
->array
.raid_disks
) {
112 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
116 if (ident
->member
&& ident
->member
[0]) {
117 /* content->text_version must match */
118 char *s
= strchr(content
->text_version
+1, '/');
121 fprintf(stderr
, Name
": %s is not a container and one is required.\n",
124 } else if (strcmp(ident
->member
, s
+1) != 0) {
126 fprintf(stderr
, Name
": skipping wrong member %s is %s\n",
127 content
->text_version
, devname
);
135 int Assemble(struct supertype
*st
, char *mddev
,
136 struct mddev_ident
*ident
,
137 struct mddev_dev
*devlist
,
138 char *backup_file
, int invalid_backup
,
139 int readonly
, int runstop
,
140 char *update
, char *homehost
, int require_homehost
,
141 int verbose
, int force
)
144 * The task of Assemble is to find a collection of
145 * devices that should (according to their superblocks)
146 * form an array, and to give this collection to the MD driver.
147 * In Linux-2.4 and later, this involves submitting a
148 * SET_ARRAY_INFO ioctl with no arg - to prepare
149 * the array - and then submit a number of
150 * ADD_NEW_DISK ioctls to add disks into
151 * the array. Finally RUN_ARRAY might
152 * be submitted to start the array.
154 * Much of the work of Assemble is in finding and/or
155 * checking the disks to make sure they look right.
157 * If mddev is not set, then scan must be set and we
158 * read through the config file for dev+uuid mapping
159 * We recurse, setting mddev, for each device that
161 * - has a valid uuid (or any uuid if !uuidset)
163 * If mddev is set, we try to determine state of md.
164 * check version - must be at least 0.90.0
165 * check kernel version. must be at least 2.4.
166 * If not, we can possibly fall back on START_ARRAY
167 * Try to GET_ARRAY_INFO.
168 * If possible, give up
169 * If not, try to STOP_ARRAY just to make sure
171 * If !uuidset and scan, look in conf-file for uuid
172 * If not found, give up
173 * If !devlist and scan and uuidset, get list of devs from conf-file
176 * Check superblock - discard if bad
177 * Check uuid (set if we don't have one) - discard if no match
178 * Check superblock similarity if we have a superblock - discard if different
179 * Record events, devicenum
180 * This should give us a list of devices for the array
181 * We should collect the most recent event number
183 * Count disks with recent enough event count
184 * While force && !enough disks
185 * Choose newest rejected disks, update event count
186 * mark clean and rewrite superblock
189 * foreach device with recent events : ADD_NEW_DISK
190 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
192 * Check the device numbers in superblock are right
193 * update superblock if any changes
199 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
200 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
201 && (ident
->container
== NULL
|| ident
->member
== NULL
));
203 int vers
= vers
; /* Keep gcc quite - it really is initialised */
206 int uptodate
; /* set once we decide that this device is as
207 * recent as everything else in the array.
212 int *best
= NULL
; /* indexed by raid_disk */
215 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
216 unsigned int req_cnt
;
222 int report_missmatch
;
224 int start_partial_ok
= (runstop
>= 0) &&
225 (force
|| devlist
==NULL
|| auto_assem
);
226 unsigned int num_devs
;
227 struct mddev_dev
*tmpdev
;
229 struct mdinfo
*content
= NULL
;
234 char chosen_name
[1024];
235 struct domainlist
*domains
= NULL
;
237 if (get_linux_version() < 2004000)
241 * If any subdevs are listed, then any that don't
242 * match ident are discarded. Remainder must all match and
244 * If no subdevs, then we scan all devices in the config file, but
245 * there must be something in the identity
249 ident
->uuid_set
== 0 &&
250 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
251 ident
->name
[0] == 0 &&
252 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
253 ident
->devices
== NULL
) {
254 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
255 mddev
? mddev
: "further assembly");
260 devlist
= conf_get_devs();
264 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
266 /* We come back here when doing auto-assembly and attempting some
267 * set of devices failed. Those are now marked as ->used==2 and
268 * we ignore them and try again
271 tmpdev
= devlist
; num_devs
= 0;
277 tmpdev
= tmpdev
->next
;
280 if (!st
&& ident
->st
) st
= ident
->st
;
283 fprintf(stderr
, Name
": looking for devices for %s\n",
284 mddev
? mddev
: "further assembly");
286 /* first walk the list of devices to find a consistent set
287 * that match the criterea, if that is possible.
288 * We flag the ones we like with 'used'.
290 for (tmpdev
= devlist
;
292 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
293 char *devname
= tmpdev
->devname
;
296 struct supertype
*tst
= dup_super(st
);
297 struct dev_policy
*pol
= NULL
;
298 int found_container
= 0;
300 if (tmpdev
->used
> 1) continue;
302 if (ident
->devices
&&
303 !match_oneof(ident
->devices
, devname
)) {
304 if (report_missmatch
)
305 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
309 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
311 if (report_missmatch
)
312 fprintf(stderr
, Name
": cannot open device %s: %s\n",
313 devname
, strerror(errno
));
315 } else if (fstat(dfd
, &stb
)< 0) {
317 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
318 devname
, strerror(errno
));
320 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
321 fprintf(stderr
, Name
": %s is not a block device.\n",
324 } else if (must_be_container(dfd
)) {
326 /* already found some components, this cannot
329 if (report_missmatch
)
330 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
333 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
334 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
335 if (report_missmatch
)
336 fprintf(stderr
, Name
": not a recognisable container: %s\n",
340 } else if (!tst
->ss
->load_container
341 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
342 if (report_missmatch
)
343 fprintf(stderr
, Name
": no correct container type: %s\n",
346 } else if (auto_assem
&&
347 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
348 tst
->ss
->match_home(tst
, homehost
) == 1)) {
349 if (report_missmatch
)
350 fprintf(stderr
, Name
": %s has metadata type %s for which "
351 "auto-assembly is disabled\n",
352 devname
, tst
->ss
->name
);
357 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
358 if (report_missmatch
)
359 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
362 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
363 if (report_missmatch
)
364 fprintf(stderr
, Name
": no RAID superblock on %s\n",
367 } else if (tst
->ss
->compare_super
== NULL
) {
368 if (report_missmatch
)
369 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
370 tst
->ss
->name
, devname
);
372 } else if (auto_assem
&& st
== NULL
&&
373 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
374 tst
->ss
->match_home(tst
, homehost
) == 1)) {
375 if (report_missmatch
)
376 fprintf(stderr
, Name
": %s has metadata type %s for which "
377 "auto-assembly is disabled\n",
378 devname
, tst
->ss
->name
);
382 if (dfd
>= 0) close(dfd
);
383 if (tmpdev
->used
== 2) {
384 if (auto_assem
|| !inargv
)
385 /* Ignore unrecognised devices during auto-assembly */
387 if (ident
->uuid_set
|| ident
->name
[0] ||
388 ident
->super_minor
!= UnSet
)
389 /* Ignore unrecognised device if looking for
394 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
397 st
->ss
->free_super(st
);
398 dev_policy_free(pol
);
399 domain_free(domains
);
403 if (found_container
) {
404 /* tmpdev is a container. We need to be either
405 * looking for a member, or auto-assembling
408 if (ident
->container
) {
409 if (ident
->container
[0] == '/' &&
410 !same_dev(ident
->container
, devname
)) {
411 if (report_missmatch
)
412 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
413 devname
, ident
->container
);
416 if (ident
->container
[0] != '/') {
421 memset(content
, 0, sizeof(*content
));
422 tst
->ss
->getinfo_super(tst
, content
, NULL
);
424 if (!parse_uuid(ident
->container
, uuid
) ||
425 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
426 if (report_missmatch
)
427 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
433 /* It is worth looking inside this container.
436 fprintf(stderr
, Name
": looking in container %s\n",
439 for (content
= tst
->ss
->container_content(tst
, NULL
);
441 content
= content
->next
) {
443 /* do not assemble arrays that might have bad blocks */
444 if (content
->array
.state
& (1<<MD_SB_BBM_ERRORS
)) {
445 fprintf(stderr
, Name
": BBM log found in metadata. "
446 "Cannot activate array(s).\n");
450 if (!ident_matches(ident
, content
, tst
,
452 report_missmatch
? devname
: NULL
))
453 /* message already printed */;
454 else if (is_member_busy(content
->text_version
)) {
455 if (report_missmatch
)
456 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
457 content
->text_version
,
464 goto loop
; /* empty container */
467 st
= tst
; tst
= NULL
;
468 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
469 fprintf(stderr
, Name
": %s is a container, but is not "
470 "only device given: confused and aborting\n",
472 st
->ss
->free_super(st
);
473 dev_policy_free(pol
);
474 domain_free(domains
);
478 fprintf(stderr
, Name
": found match on member %s in %s\n",
479 content
->text_version
, devname
);
481 /* make sure we finished the loop */
487 memset(content
, 0, sizeof(*content
));
488 tst
->ss
->getinfo_super(tst
, content
, NULL
);
490 if (!ident_matches(ident
, content
, tst
,
492 report_missmatch
? devname
: NULL
))
497 if (st
->minor_version
== -1)
498 st
->minor_version
= tst
->minor_version
;
500 if (memcmp(content
->uuid
, uuid_zero
,
501 sizeof(int[4])) == 0) {
502 /* this is a floating spare. It cannot define
503 * an array unless there are no more arrays of
504 * this type to be found. It can be included
505 * in an array of this type though.
511 if (st
->ss
!= tst
->ss
||
512 st
->minor_version
!= tst
->minor_version
||
513 st
->ss
->compare_super(st
, tst
) != 0) {
514 /* Some mismatch. If exactly one array matches this host,
515 * we can resolve on that one.
516 * Or, if we are auto assembling, we just ignore the second
522 int first
= st
->ss
->match_home(st
, homehost
);
523 int last
= tst
->ss
->match_home(tst
, homehost
);
525 (first
== 1 || last
== 1)) {
526 /* We can do something */
527 if (first
) {/* just ignore this one */
528 if (report_missmatch
)
529 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
532 } else { /* reject all those sofar */
533 struct mddev_dev
*td
;
534 if (report_missmatch
)
535 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
537 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
545 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
547 tst
->ss
->free_super(tst
);
548 st
->ss
->free_super(st
);
549 dev_policy_free(pol
);
550 domain_free(domains
);
556 /* Collect domain information from members only */
557 if (tmpdev
&& tmpdev
->used
== 1) {
559 pol
= devnum_policy(stb
.st_rdev
);
560 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
562 dev_policy_free(pol
);
565 tst
->ss
->free_super(tst
);
568 /* Check if we found some imsm spares but no members */
571 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
573 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
574 if (tmpdev
->used
!= 3)
580 /* we need sb from one of the spares */
581 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
583 st
->ss
->load_super(st
, dfd
, NULL
))
590 /* Now reject spares that don't match domains of identified members */
591 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
593 if (tmpdev
->used
!= 3)
595 if (stat(tmpdev
->devname
, &stb
)< 0) {
596 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
597 tmpdev
->devname
, strerror(errno
));
600 struct dev_policy
*pol
= devnum_policy(stb
.st_rdev
);
601 int dt
= domain_test(domains
, pol
, NULL
);
602 if (inargv
&& dt
!= 0)
603 /* take this spare as domains match
604 * if there are any */
606 else if (!inargv
&& dt
== 1)
607 /* device wasn't explicitly listed, so need
608 * explicit domain match - which we have */
611 /* if domains don't match mark as unused */
613 dev_policy_free(pol
);
616 domain_free(domains
);
618 if (!st
|| !st
->sb
|| !content
)
621 /* Now need to open the array device. Use create_mddev */
622 if (content
== &info
)
623 st
->ss
->getinfo_super(st
, content
, NULL
);
625 trustworthy
= FOREIGN
;
626 name
= content
->name
;
627 switch (st
->ss
->match_home(st
, homehost
)
628 ?: st
->ss
->match_home(st
, "any")) {
631 name
= strchr(content
->name
, ':');
635 name
= content
->name
;
639 /* If the array is listed in mdadm.conf or on
640 * command line, then we trust the name
641 * even if the array doesn't look local
646 content
->array
.level
== LEVEL_CONTAINER
) {
647 name
= content
->text_version
;
648 trustworthy
= METADATA
;
651 if (name
[0] && trustworthy
!= LOCAL
&&
652 ! require_homehost
&&
653 conf_name_is_free(name
))
656 if (trustworthy
== LOCAL
&&
658 /* Ignore 'host:' prefix of name */
659 name
= strchr(name
, ':')+1;
661 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
664 st
->ss
->free_super(st
);
670 vers
= md_get_version(mdfd
);
672 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
673 " Upgrade your kernel or try --build\n");
677 if (mddev_busy(fd2devnum(mdfd
))) {
678 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
680 for (tmpdev
= devlist
;
681 tmpdev
&& tmpdev
->used
!= 1;
682 tmpdev
= tmpdev
->next
)
684 if (tmpdev
&& auto_assem
)
685 fprintf(stderr
, Name
": %s needed for %s...\n",
686 mddev
, tmpdev
->devname
);
689 st
->ss
->free_super(st
);
694 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
697 if (content
!= &info
) {
698 /* This is a member of a container. Try starting the array. */
700 err
= assemble_container_content(st
, mdfd
, content
, runstop
,
701 chosen_name
, verbose
,
707 /* Ok, no bad inconsistancy, we can try updating etc */
709 content
->update_private
= NULL
;
710 devices
= malloc(num_devs
* sizeof(*devices
));
711 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
712 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
713 char *devname
= tmpdev
->devname
;
715 /* looks like a good enough match to update the super block if needed */
719 /* prepare useful information in info structures */
721 struct supertype
*tst
;
725 if (strcmp(update
, "uuid")==0 &&
728 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
729 read(rfd
, ident
->uuid
, 16) != 16) {
730 *(__u32
*)(ident
->uuid
) = random();
731 *(__u32
*)(ident
->uuid
+1) = random();
732 *(__u32
*)(ident
->uuid
+2) = random();
733 *(__u32
*)(ident
->uuid
+3) = random();
735 if (rfd
>= 0) close(rfd
);
737 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
740 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
741 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
750 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
752 memcpy(content
->uuid
, ident
->uuid
, 16);
753 strcpy(content
->name
, ident
->name
);
754 content
->array
.md_minor
= minor(stb2
.st_rdev
);
756 if (strcmp(update
, "byteorder") == 0)
759 err
= tst
->ss
->update_super(tst
, content
, update
,
765 Name
": --update=%s not understood"
766 " for %s metadata\n",
767 update
, tst
->ss
->name
);
768 tst
->ss
->free_super(tst
);
776 if (strcmp(update
, "uuid")==0 &&
779 memcpy(ident
->uuid
, content
->uuid
, 16);
781 if (tst
->ss
->store_super(tst
, dfd
))
782 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
786 if (strcmp(update
, "uuid")==0 &&
787 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
788 if (bitmap_update_uuid(ident
->bitmap_fd
,
790 tst
->ss
->swapuuid
) != 0)
791 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
795 tst
->ss
->free_super(tst
);
799 struct supertype
*tst
= dup_super(st
);
801 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
803 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
804 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
813 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
814 tst
->ss
->free_super(tst
);
821 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
822 devname
, mddev
, content
->disk
.raid_disk
);
823 devices
[devcnt
].devname
= devname
;
824 devices
[devcnt
].uptodate
= 0;
825 devices
[devcnt
].i
= *content
;
826 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
827 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
828 if (most_recent
< devcnt
) {
829 if (devices
[devcnt
].i
.events
830 > devices
[most_recent
].i
.events
)
831 most_recent
= devcnt
;
833 if (content
->array
.level
== LEVEL_MULTIPATH
)
834 /* with multipath, the raid_disk from the superblock is meaningless */
837 i
= devices
[devcnt
].i
.disk
.raid_disk
;
839 if (nextspare
< content
->array
.raid_disks
)
840 nextspare
= content
->array
.raid_disks
;
843 if (i
>= content
->array
.raid_disks
&&
849 int newbestcnt
= i
+10;
850 int *newbest
= malloc(sizeof(int)*newbestcnt
);
852 for (c
=0; c
< newbestcnt
; c
++)
854 newbest
[c
] = best
[c
];
859 bestcnt
= newbestcnt
;
862 devices
[best
[i
]].i
.events
863 == devices
[devcnt
].i
.events
864 && (devices
[best
[i
]].i
.disk
.minor
865 != devices
[devcnt
].i
.disk
.minor
)
867 && content
->array
.level
!= LEVEL_MULTIPATH
) {
868 /* two different devices with identical superblock.
869 * Could be a mis-detection caused by overlapping
870 * partitions. fail-safe.
872 fprintf(stderr
, Name
": WARNING %s and %s appear"
873 " to have very similar superblocks.\n"
874 " If they are really different, "
875 "please --zero the superblock on one\n"
876 " If they are the same or overlap,"
877 " please remove one from %s.\n",
878 devices
[best
[i
]].devname
, devname
,
879 inargv
? "the list" :
880 "the\n DEVICE list in mdadm.conf"
888 || (devices
[best
[i
]].i
.events
889 < devices
[devcnt
].i
.events
))
894 free(content
->update_private
);
895 content
->update_private
= NULL
;
898 fprintf(stderr
, Name
": no devices found for %s\n",
901 st
->ss
->free_super(st
);
908 if (update
&& strcmp(update
, "byteorder")==0)
909 st
->minor_version
= 90;
911 st
->ss
->getinfo_super(st
, content
, NULL
);
912 clean
= content
->array
.state
& 1;
914 /* now we have some devices that might be suitable.
917 avail
= malloc(content
->array
.raid_disks
);
918 memset(avail
, 0, content
->array
.raid_disks
);
922 for (i
=0; i
< bestcnt
; i
++) {
924 int event_margin
= 1; /* always allow a difference of '1'
925 * like the kernel does
928 /* note: we ignore error flags in multipath arrays
929 * as they don't make sense
931 if (content
->array
.level
!= LEVEL_MULTIPATH
)
932 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
933 if (!(devices
[j
].i
.disk
.state
934 & (1<<MD_DISK_FAULTY
))) {
935 devices
[j
].uptodate
= 1;
940 /* If this devices thinks that 'most_recent' has failed, then
941 * we must reject this device.
943 if (j
!= most_recent
&&
944 content
->array
.raid_disks
> 0 &&
945 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
946 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
948 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
949 devices
[j
].devname
, devices
[most_recent
].devname
);
953 if (devices
[j
].i
.events
+event_margin
>=
954 devices
[most_recent
].i
.events
) {
955 devices
[j
].uptodate
= 1;
956 if (i
< content
->array
.raid_disks
) {
957 if (devices
[j
].i
.recovery_start
== MaxSector
||
958 (content
->reshape_active
&&
959 j
>= content
->array
.raid_disks
- content
->delta_disks
)) {
969 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
970 content
->array
.layout
, 1,
972 /* Choose the newest best drive which is
973 * not up-to-date, update the superblock
977 struct supertype
*tst
;
978 unsigned long long current_events
;
980 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
983 !devices
[j
].uptodate
&&
984 devices
[j
].i
.recovery_start
== MaxSector
&&
987 > devices
[chosen_drive
].i
.events
))
990 if (chosen_drive
< 0)
992 current_events
= devices
[chosen_drive
].i
.events
;
995 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
996 devices
[chosen_drive
].devname
,
997 devices
[chosen_drive
].i
.disk
.raid_disk
,
998 (int)(devices
[chosen_drive
].i
.events
),
999 (int)(devices
[most_recent
].i
.events
));
1000 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1002 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
1003 devices
[chosen_drive
].devname
);
1004 devices
[chosen_drive
].i
.events
= 0;
1007 tst
= dup_super(st
);
1008 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
1010 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
1011 devices
[chosen_drive
].devname
);
1012 devices
[chosen_drive
].i
.events
= 0;
1015 content
->events
= devices
[most_recent
].i
.events
;
1016 tst
->ss
->update_super(tst
, content
, "force-one",
1017 devices
[chosen_drive
].devname
, verbose
,
1020 if (tst
->ss
->store_super(tst
, fd
)) {
1022 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1023 devices
[chosen_drive
].devname
);
1024 devices
[chosen_drive
].i
.events
= 0;
1025 tst
->ss
->free_super(tst
);
1029 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
1030 devices
[chosen_drive
].uptodate
= 1;
1031 avail
[chosen_drive
] = 1;
1033 tst
->ss
->free_super(tst
);
1035 /* If there are any other drives of the same vintage,
1036 * add them in as well. We can't lose and we might gain
1038 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
1041 !devices
[j
].uptodate
&&
1042 devices
[j
].i
.events
== current_events
) {
1049 /* Now we want to look at the superblock which the kernel will base things on
1050 * and compare the devices that we think are working with the devices that the
1051 * superblock thinks are working.
1052 * If there are differences and --force is given, then update this chosen
1056 st
->ss
->free_super(st
);
1057 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
1063 if (!devices
[j
].uptodate
)
1066 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
1067 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1068 devices
[j
].devname
, strerror(errno
));
1073 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1075 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
1076 devices
[j
].devname
);
1083 if (st
->sb
== NULL
) {
1084 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
1089 st
->ss
->getinfo_super(st
, content
, NULL
);
1091 sysfs_init(content
, mdfd
, 0);
1093 for (i
=0; i
<bestcnt
; i
++) {
1095 unsigned int desired_state
;
1097 if (i
< content
->array
.raid_disks
)
1098 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1104 if (!devices
[j
].uptodate
)
1107 devices
[j
].i
.disk
.state
= desired_state
;
1108 if (!(devices
[j
].i
.array
.state
& 1))
1111 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1112 verbose
, 0, NULL
)) {
1115 fprintf(stderr
, Name
": "
1116 "clearing FAULTY flag for device %d in %s for %s\n",
1117 j
, mddev
, devices
[j
].devname
);
1121 fprintf(stderr
, Name
": "
1122 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1123 i
, mddev
, devices
[j
].devname
);
1127 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1128 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1133 if (force
&& !clean
&&
1134 !enough(content
->array
.level
, content
->array
.raid_disks
,
1135 content
->array
.layout
, clean
,
1137 change
+= st
->ss
->update_super(st
, content
, "force-array",
1138 devices
[chosen_drive
].devname
, verbose
,
1145 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1147 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1148 devices
[chosen_drive
].devname
);
1153 if (st
->ss
->store_super(st
, fd
)) {
1155 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1156 devices
[chosen_drive
].devname
);
1164 /* If we are in the middle of a reshape we may need to restore saved data
1165 * that was moved aside due to the reshape overwriting live data
1166 * The code of doing this lives in Grow.c
1169 if (content
->reshape_active
) {
1171 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1173 fprintf(stderr
, Name
":%s has an active reshape - checking "
1174 "if critical section needs to be restored\n",
1176 for (i
=0; i
<bestcnt
; i
++) {
1179 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1180 if (fdlist
[i
] < 0) {
1181 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1182 devices
[j
].devname
);
1190 err
= Grow_restart(st
, content
, fdlist
, bestcnt
,
1191 backup_file
, verbose
> 0);
1192 if (err
&& invalid_backup
) {
1194 fprintf(stderr
, Name
": continuing"
1195 " without restoring backup\n");
1201 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1204 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1205 if (backup_file
== NULL
)
1206 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1213 /* count number of in-sync devices according to the superblock.
1214 * We must have this number to start the array without -s or -R
1216 req_cnt
= content
->array
.working_disks
;
1218 /* Almost ready to actually *do* something */
1222 /* First, fill in the map, so that udev can find our name
1223 * as soon as we become active.
1225 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1226 content
->uuid
, chosen_name
);
1228 rv
= set_array_info(mdfd
, st
, content
);
1230 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1231 mddev
, strerror(errno
));
1232 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1237 if (ident
->bitmap_fd
>= 0) {
1238 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1239 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1240 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1245 } else if (ident
->bitmap_file
) {
1246 /* From config file */
1247 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1249 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1250 ident
->bitmap_file
);
1251 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1256 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1257 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1259 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1267 /* First, add the raid disks, but add the chosen one last */
1268 for (i
=0; i
<= bestcnt
; i
++) {
1272 if (j
== chosen_drive
)
1277 if (j
>= 0 /* && devices[j].uptodate */) {
1278 int dfd
= dev_open(devices
[j
].devname
,
1281 remove_partitions(dfd
);
1284 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1287 fprintf(stderr
, Name
": failed to add "
1292 if (i
< content
->array
.raid_disks
1297 } else if (verbose
> 0)
1298 fprintf(stderr
, Name
": added %s "
1300 devices
[j
].devname
, mddev
,
1301 devices
[j
].i
.disk
.raid_disk
);
1302 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1303 fprintf(stderr
, Name
": no uptodate device for "
1308 if (content
->array
.level
== LEVEL_CONTAINER
) {
1310 fprintf(stderr
, Name
": Container %s has been "
1311 "assembled with %d drive%s",
1312 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1313 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1314 fprintf(stderr
, " (out of %d)",
1315 content
->array
.raid_disks
);
1316 fprintf(stderr
, "\n");
1318 st
->ss
->free_super(st
);
1319 sysfs_uevent(content
, "change");
1320 wait_for(chosen_name
, mdfd
);
1328 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1329 content
->array
.layout
, clean
, avail
, okcnt
) &&
1330 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1332 /* This array is good-to-go.
1333 * If a reshape is in progress then we might need to
1334 * continue monitoring it. In that case we start
1335 * it read-only and let the grow code make it writable.
1339 if (content
->reshape_active
&&
1340 content
->delta_disks
<= 0)
1341 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1344 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1347 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1348 mddev
, okcnt
, okcnt
==1?"":"s");
1349 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1350 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1352 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1354 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1355 fprintf(stderr
, ".\n");
1357 if (content
->reshape_active
&&
1358 content
->array
.level
>= 4 &&
1359 content
->array
.level
<= 6) {
1360 /* might need to increase the size
1361 * of the stripe cache - default is 256
1363 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1364 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1366 sysfs_set_num(sra
, NULL
,
1367 "stripe_cache_size",
1368 (4 * content
->array
.chunk_size
/ 4096) + 1);
1371 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1372 /* If any devices did not get added
1373 * because the kernel rejected them based
1374 * on event count, try adding them
1375 * again providing the action policy is
1376 * 're-add' or greater. The bitmap
1377 * might allow them to be included, or
1378 * they will become spares.
1380 for (i
= 0; i
<= bestcnt
; i
++) {
1382 if (j
>= 0 && !devices
[j
].uptodate
) {
1383 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1385 rv
= add_disk(mdfd
, st
, content
,
1387 if (rv
== 0 && verbose
>= 0)
1389 Name
": %s has been re-added.\n",
1390 devices
[j
].devname
);
1394 wait_for(mddev
, mdfd
);
1398 /* There is a nasty race with 'mdadm --monitor'.
1399 * If it opens this device before we close it,
1400 * it gets an incomplete open on which IO
1401 * doesn't work and the capacity is
1403 * If we reopen (to check for layered devices)
1404 * before --monitor closes, we loose.
1406 * So: wait upto 1 second for there to be
1407 * a non-zero capacity.
1409 while (usecs
< 1000) {
1410 mdfd
= open(mddev
, O_RDONLY
);
1412 unsigned long long size
;
1413 if (get_dev_size(mdfd
, NULL
, &size
) &&
1425 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1426 mddev
, strerror(errno
));
1428 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1429 content
->array
.layout
, 1, avail
, okcnt
))
1430 fprintf(stderr
, Name
": Not enough devices to "
1431 "start the array.\n");
1432 else if (!enough(content
->array
.level
,
1433 content
->array
.raid_disks
,
1434 content
->array
.layout
, clean
,
1436 fprintf(stderr
, Name
": Not enough devices to "
1437 "start the array while not clean "
1438 "- consider --force.\n");
1441 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1446 if (runstop
== -1) {
1447 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1448 mddev
, okcnt
, okcnt
==1?"":"s");
1449 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1450 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1451 fprintf(stderr
, ", but not started.\n");
1456 if (verbose
>= -1) {
1457 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1459 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1461 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1462 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1463 content
->array
.layout
, 1, avail
, okcnt
))
1464 fprintf(stderr
, " - not enough to start the array.\n");
1465 else if (!enough(content
->array
.level
,
1466 content
->array
.raid_disks
,
1467 content
->array
.layout
, clean
,
1469 fprintf(stderr
, " - not enough to start the "
1470 "array while not clean - consider "
1473 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1474 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1476 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1477 fprintf(stderr
, " (use --run to insist).\n");
1481 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1486 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1487 * been updated to point to the current locations of devices.
1488 * so we can just start the array
1491 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1492 devices
[chosen_drive
].i
.disk
.minor
);
1493 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1494 fprintf(stderr
, Name
": Cannot start array: %s\n",
1505 int assemble_container_content(struct supertype
*st
, int mdfd
,
1506 struct mdinfo
*content
, int runstop
,
1507 char *chosen_name
, int verbose
,
1510 struct mdinfo
*dev
, *sra
;
1511 int working
= 0, preexist
= 0;
1513 struct map_ent
*map
= NULL
;
1515 sysfs_init(content
, mdfd
, 0);
1517 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1518 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1519 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0)
1522 if (content
->reshape_active
)
1523 block_subarray(content
);
1528 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1529 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1530 if (dev
->disk
.raid_disk
>= content
->array
.raid_disks
&&
1531 content
->reshape_active
)
1535 } else if (errno
== EEXIST
)
1538 return 1;/* Nothing new, don't try to start */
1540 map_update(&map
, fd2devnum(mdfd
),
1541 content
->text_version
,
1542 content
->uuid
, chosen_name
);
1545 (working
+ preexist
+ expansion
) >=
1546 content
->array
.working_disks
) {
1549 if (content
->reshape_active
) {
1550 int spare
= content
->array
.raid_disks
+ expansion
;
1552 int *fdlist
= malloc(sizeof(int) *
1553 (working
+ expansion
1554 + content
->array
.raid_disks
));
1555 for (i
=0; i
<spare
; i
++)
1557 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1560 sprintf(buf
, "%d:%d",
1563 fd
= dev_open(buf
, O_RDWR
);
1565 if (dev
->disk
.raid_disk
>= 0)
1566 fdlist
[dev
->disk
.raid_disk
] = fd
;
1568 fdlist
[spare
++] = fd
;
1570 err
= Grow_restart(st
, content
, fdlist
, spare
,
1571 backup_file
, verbose
> 0);
1574 if (fdlist
[spare
] >= 0)
1575 close(fdlist
[spare
]);
1578 fprintf(stderr
, Name
": Failed to restore critical"
1579 " section for reshape - sorry.\n");
1581 fprintf(stderr
, Name
": Possibly you need"
1582 " to specify a --backup-file\n");
1586 err
= Grow_continue(mdfd
, st
, content
, backup_file
);
1587 } else switch(content
->array
.level
) {
1589 case LEVEL_MULTIPATH
:
1591 err
= sysfs_set_str(content
, NULL
, "array_state",
1595 err
= sysfs_set_str(content
, NULL
, "array_state",
1597 /* start mdmon if needed. */
1599 if (!mdmon_running(st
->container_dev
))
1600 start_mdmon(st
->container_dev
);
1601 ping_monitor_by_id(st
->container_dev
);
1606 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1609 fprintf(stderr
, Name
1610 ": array %s now has %d devices",
1611 chosen_name
, working
+ preexist
);
1613 fprintf(stderr
, Name
1614 ": Started %s with %d devices",
1615 chosen_name
, working
+ preexist
);
1617 fprintf(stderr
, " (%d new)", working
);
1619 fprintf(stderr
, " ( + %d for expansion)",
1621 fprintf(stderr
, "\n");
1624 wait_for(chosen_name
, mdfd
);
1626 /* FIXME should have an O_EXCL and wait for read-auto */
1629 fprintf(stderr
, Name
1630 ": %s assembled with %d devices but "
1632 chosen_name
, working
);