]>
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 tst
->ss
->getinfo_super(tst
, content
, NULL
);
423 if (!parse_uuid(ident
->container
, uuid
) ||
424 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
425 if (report_missmatch
)
426 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
432 /* It is worth looking inside this container.
435 fprintf(stderr
, Name
": looking in container %s\n",
438 for (content
= tst
->ss
->container_content(tst
, NULL
);
440 content
= content
->next
) {
442 /* do not assemble arrays that might have bad blocks */
443 if (content
->array
.state
& (1<<MD_SB_BBM_ERRORS
)) {
444 fprintf(stderr
, Name
": BBM log found in metadata. "
445 "Cannot activate array(s).\n");
449 if (!ident_matches(ident
, content
, tst
,
451 report_missmatch
? devname
: NULL
))
452 /* message already printed */;
453 else if (is_member_busy(content
->text_version
)) {
454 if (report_missmatch
)
455 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
456 content
->text_version
,
463 goto loop
; /* empty container */
466 st
= tst
; tst
= NULL
;
467 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
468 fprintf(stderr
, Name
": %s is a container, but is not "
469 "only device given: confused and aborting\n",
471 st
->ss
->free_super(st
);
472 dev_policy_free(pol
);
473 domain_free(domains
);
477 fprintf(stderr
, Name
": found match on member %s in %s\n",
478 content
->text_version
, devname
);
480 /* make sure we finished the loop */
486 tst
->ss
->getinfo_super(tst
, content
, NULL
);
488 if (!ident_matches(ident
, content
, tst
,
490 report_missmatch
? devname
: NULL
))
495 if (st
->minor_version
== -1)
496 st
->minor_version
= tst
->minor_version
;
498 if (memcmp(content
->uuid
, uuid_zero
,
499 sizeof(int[4])) == 0) {
500 /* this is a floating spare. It cannot define
501 * an array unless there are no more arrays of
502 * this type to be found. It can be included
503 * in an array of this type though.
509 if (st
->ss
!= tst
->ss
||
510 st
->minor_version
!= tst
->minor_version
||
511 st
->ss
->compare_super(st
, tst
) != 0) {
512 /* Some mismatch. If exactly one array matches this host,
513 * we can resolve on that one.
514 * Or, if we are auto assembling, we just ignore the second
520 int first
= st
->ss
->match_home(st
, homehost
);
521 int last
= tst
->ss
->match_home(tst
, homehost
);
523 (first
== 1 || last
== 1)) {
524 /* We can do something */
525 if (first
) {/* just ignore this one */
526 if (report_missmatch
)
527 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
530 } else { /* reject all those sofar */
531 struct mddev_dev
*td
;
532 if (report_missmatch
)
533 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
535 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
543 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
545 tst
->ss
->free_super(tst
);
546 st
->ss
->free_super(st
);
547 dev_policy_free(pol
);
548 domain_free(domains
);
554 /* Collect domain information from members only */
555 if (tmpdev
&& tmpdev
->used
== 1) {
557 pol
= devnum_policy(stb
.st_rdev
);
558 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
560 dev_policy_free(pol
);
563 tst
->ss
->free_super(tst
);
566 /* Check if we found some imsm spares but no members */
569 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
571 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
572 if (tmpdev
->used
!= 3)
578 /* we need sb from one of the spares */
579 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
581 st
->ss
->load_super(st
, dfd
, NULL
))
588 /* Now reject spares that don't match domains of identified members */
589 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
591 if (tmpdev
->used
!= 3)
593 if (stat(tmpdev
->devname
, &stb
)< 0) {
594 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
595 tmpdev
->devname
, strerror(errno
));
598 struct dev_policy
*pol
= devnum_policy(stb
.st_rdev
);
599 int dt
= domain_test(domains
, pol
, NULL
);
600 if (inargv
&& dt
!= 0)
601 /* take this spare as domains match
602 * if there are any */
604 else if (!inargv
&& dt
== 1)
605 /* device wasn't explicitly listed, so need
606 * explicit domain match - which we have */
609 /* if domains don't match mark as unused */
611 dev_policy_free(pol
);
614 domain_free(domains
);
616 if (!st
|| !st
->sb
|| !content
)
619 /* Now need to open the array device. Use create_mddev */
620 if (content
== &info
)
621 st
->ss
->getinfo_super(st
, content
, NULL
);
623 trustworthy
= FOREIGN
;
624 name
= content
->name
;
625 switch (st
->ss
->match_home(st
, homehost
)
626 ?: st
->ss
->match_home(st
, "any")) {
629 name
= strchr(content
->name
, ':');
633 name
= content
->name
;
637 /* If the array is listed in mdadm.conf or on
638 * command line, then we trust the name
639 * even if the array doesn't look local
644 content
->array
.level
== LEVEL_CONTAINER
) {
645 name
= content
->text_version
;
646 trustworthy
= METADATA
;
649 if (name
[0] && trustworthy
!= LOCAL
&&
650 ! require_homehost
&&
651 conf_name_is_free(name
))
654 if (trustworthy
== LOCAL
&&
656 /* Ignore 'host:' prefix of name */
657 name
= strchr(name
, ':')+1;
659 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
662 st
->ss
->free_super(st
);
668 vers
= md_get_version(mdfd
);
670 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
671 " Upgrade your kernel or try --build\n");
675 if (mddev_busy(fd2devnum(mdfd
))) {
676 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
678 for (tmpdev
= devlist
;
679 tmpdev
&& tmpdev
->used
!= 1;
680 tmpdev
= tmpdev
->next
)
682 if (tmpdev
&& auto_assem
)
683 fprintf(stderr
, Name
": %s needed for %s...\n",
684 mddev
, tmpdev
->devname
);
687 st
->ss
->free_super(st
);
692 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
695 if (content
!= &info
) {
696 /* This is a member of a container. Try starting the array. */
698 err
= assemble_container_content(st
, mdfd
, content
, runstop
,
699 chosen_name
, verbose
,
705 /* Ok, no bad inconsistancy, we can try updating etc */
707 content
->update_private
= NULL
;
708 devices
= malloc(num_devs
* sizeof(*devices
));
709 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
710 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
711 char *devname
= tmpdev
->devname
;
713 /* looks like a good enough match to update the super block if needed */
717 /* prepare useful information in info structures */
719 struct supertype
*tst
;
723 if (strcmp(update
, "uuid")==0 &&
726 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
727 read(rfd
, ident
->uuid
, 16) != 16) {
728 *(__u32
*)(ident
->uuid
) = random();
729 *(__u32
*)(ident
->uuid
+1) = random();
730 *(__u32
*)(ident
->uuid
+2) = random();
731 *(__u32
*)(ident
->uuid
+3) = random();
733 if (rfd
>= 0) close(rfd
);
735 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
738 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
739 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
748 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
750 memcpy(content
->uuid
, ident
->uuid
, 16);
751 strcpy(content
->name
, ident
->name
);
752 content
->array
.md_minor
= minor(stb2
.st_rdev
);
754 if (strcmp(update
, "byteorder") == 0)
757 err
= tst
->ss
->update_super(tst
, content
, update
,
763 Name
": --update=%s not understood"
764 " for %s metadata\n",
765 update
, tst
->ss
->name
);
766 tst
->ss
->free_super(tst
);
774 if (strcmp(update
, "uuid")==0 &&
777 memcpy(ident
->uuid
, content
->uuid
, 16);
779 if (tst
->ss
->store_super(tst
, dfd
))
780 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
784 if (strcmp(update
, "uuid")==0 &&
785 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
786 if (bitmap_update_uuid(ident
->bitmap_fd
,
788 tst
->ss
->swapuuid
) != 0)
789 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
793 tst
->ss
->free_super(tst
);
797 struct supertype
*tst
= dup_super(st
);
799 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
801 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
802 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
811 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
812 tst
->ss
->free_super(tst
);
819 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
820 devname
, mddev
, content
->disk
.raid_disk
);
821 devices
[devcnt
].devname
= devname
;
822 devices
[devcnt
].uptodate
= 0;
823 devices
[devcnt
].i
= *content
;
824 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
825 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
826 if (most_recent
< devcnt
) {
827 if (devices
[devcnt
].i
.events
828 > devices
[most_recent
].i
.events
)
829 most_recent
= devcnt
;
831 if (content
->array
.level
== LEVEL_MULTIPATH
)
832 /* with multipath, the raid_disk from the superblock is meaningless */
835 i
= devices
[devcnt
].i
.disk
.raid_disk
;
837 if (nextspare
< content
->array
.raid_disks
)
838 nextspare
= content
->array
.raid_disks
;
841 if (i
>= content
->array
.raid_disks
&&
847 int newbestcnt
= i
+10;
848 int *newbest
= malloc(sizeof(int)*newbestcnt
);
850 for (c
=0; c
< newbestcnt
; c
++)
852 newbest
[c
] = best
[c
];
857 bestcnt
= newbestcnt
;
860 devices
[best
[i
]].i
.events
861 == devices
[devcnt
].i
.events
862 && (devices
[best
[i
]].i
.disk
.minor
863 != devices
[devcnt
].i
.disk
.minor
)
865 && content
->array
.level
!= LEVEL_MULTIPATH
) {
866 /* two different devices with identical superblock.
867 * Could be a mis-detection caused by overlapping
868 * partitions. fail-safe.
870 fprintf(stderr
, Name
": WARNING %s and %s appear"
871 " to have very similar superblocks.\n"
872 " If they are really different, "
873 "please --zero the superblock on one\n"
874 " If they are the same or overlap,"
875 " please remove one from %s.\n",
876 devices
[best
[i
]].devname
, devname
,
877 inargv
? "the list" :
878 "the\n DEVICE list in mdadm.conf"
886 || (devices
[best
[i
]].i
.events
887 < devices
[devcnt
].i
.events
))
892 free(content
->update_private
);
893 content
->update_private
= NULL
;
896 fprintf(stderr
, Name
": no devices found for %s\n",
899 st
->ss
->free_super(st
);
906 if (update
&& strcmp(update
, "byteorder")==0)
907 st
->minor_version
= 90;
909 st
->ss
->getinfo_super(st
, content
, NULL
);
910 clean
= content
->array
.state
& 1;
912 /* now we have some devices that might be suitable.
915 avail
= malloc(content
->array
.raid_disks
);
916 memset(avail
, 0, content
->array
.raid_disks
);
920 for (i
=0; i
< bestcnt
; i
++) {
922 int event_margin
= 1; /* always allow a difference of '1'
923 * like the kernel does
926 /* note: we ignore error flags in multipath arrays
927 * as they don't make sense
929 if (content
->array
.level
!= LEVEL_MULTIPATH
)
930 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
931 if (!(devices
[j
].i
.disk
.state
932 & (1<<MD_DISK_FAULTY
))) {
933 devices
[j
].uptodate
= 1;
938 /* If this devices thinks that 'most_recent' has failed, then
939 * we must reject this device.
941 if (j
!= most_recent
&&
942 content
->array
.raid_disks
> 0 &&
943 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
944 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
946 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
947 devices
[j
].devname
, devices
[most_recent
].devname
);
951 if (devices
[j
].i
.events
+event_margin
>=
952 devices
[most_recent
].i
.events
) {
953 devices
[j
].uptodate
= 1;
954 if (i
< content
->array
.raid_disks
) {
955 if (devices
[j
].i
.recovery_start
== MaxSector
||
956 (content
->reshape_active
&&
957 j
>= content
->array
.raid_disks
- content
->delta_disks
)) {
967 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
968 content
->array
.layout
, 1,
970 /* Choose the newest best drive which is
971 * not up-to-date, update the superblock
975 struct supertype
*tst
;
976 unsigned long long current_events
;
978 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
981 !devices
[j
].uptodate
&&
982 devices
[j
].i
.recovery_start
== MaxSector
&&
985 > devices
[chosen_drive
].i
.events
))
988 if (chosen_drive
< 0)
990 current_events
= devices
[chosen_drive
].i
.events
;
993 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
994 devices
[chosen_drive
].devname
,
995 devices
[chosen_drive
].i
.disk
.raid_disk
,
996 (int)(devices
[chosen_drive
].i
.events
),
997 (int)(devices
[most_recent
].i
.events
));
998 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1000 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
1001 devices
[chosen_drive
].devname
);
1002 devices
[chosen_drive
].i
.events
= 0;
1005 tst
= dup_super(st
);
1006 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
1008 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
1009 devices
[chosen_drive
].devname
);
1010 devices
[chosen_drive
].i
.events
= 0;
1013 content
->events
= devices
[most_recent
].i
.events
;
1014 tst
->ss
->update_super(tst
, content
, "force-one",
1015 devices
[chosen_drive
].devname
, verbose
,
1018 if (tst
->ss
->store_super(tst
, fd
)) {
1020 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1021 devices
[chosen_drive
].devname
);
1022 devices
[chosen_drive
].i
.events
= 0;
1023 tst
->ss
->free_super(tst
);
1027 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
1028 devices
[chosen_drive
].uptodate
= 1;
1029 avail
[chosen_drive
] = 1;
1031 tst
->ss
->free_super(tst
);
1033 /* If there are any other drives of the same vintage,
1034 * add them in as well. We can't lose and we might gain
1036 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
1039 !devices
[j
].uptodate
&&
1040 devices
[j
].i
.events
== current_events
) {
1047 /* Now we want to look at the superblock which the kernel will base things on
1048 * and compare the devices that we think are working with the devices that the
1049 * superblock thinks are working.
1050 * If there are differences and --force is given, then update this chosen
1054 st
->ss
->free_super(st
);
1055 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
1061 if (!devices
[j
].uptodate
)
1063 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
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
);
1162 fprintf(stderr
, Name
": Marking array %s as 'clean'\n",
1167 /* If we are in the middle of a reshape we may need to restore saved data
1168 * that was moved aside due to the reshape overwriting live data
1169 * The code of doing this lives in Grow.c
1172 if (content
->reshape_active
) {
1174 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1176 fprintf(stderr
, Name
":%s has an active reshape - checking "
1177 "if critical section needs to be restored\n",
1179 for (i
=0; i
<bestcnt
; i
++) {
1182 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1183 if (fdlist
[i
] < 0) {
1184 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1185 devices
[j
].devname
);
1193 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1194 err
= st
->ss
->recover_backup(st
, content
);
1196 err
= Grow_restart(st
, content
, fdlist
, bestcnt
,
1197 backup_file
, verbose
> 0);
1198 if (err
&& invalid_backup
) {
1200 fprintf(stderr
, Name
": continuing"
1201 " without restoring backup\n");
1207 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1210 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1211 if (backup_file
== NULL
)
1212 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1219 /* count number of in-sync devices according to the superblock.
1220 * We must have this number to start the array without -s or -R
1222 req_cnt
= content
->array
.working_disks
;
1224 /* Almost ready to actually *do* something */
1228 /* First, fill in the map, so that udev can find our name
1229 * as soon as we become active.
1231 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1232 content
->uuid
, chosen_name
);
1234 rv
= set_array_info(mdfd
, st
, content
);
1236 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1237 mddev
, strerror(errno
));
1238 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1243 if (ident
->bitmap_fd
>= 0) {
1244 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1245 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1246 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1251 } else if (ident
->bitmap_file
) {
1252 /* From config file */
1253 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1255 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1256 ident
->bitmap_file
);
1257 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1262 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1263 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1265 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1273 /* First, add the raid disks, but add the chosen one last */
1274 for (i
=0; i
<= bestcnt
; i
++) {
1278 if (j
== chosen_drive
)
1283 if (j
>= 0 /* && devices[j].uptodate */) {
1284 int dfd
= dev_open(devices
[j
].devname
,
1287 remove_partitions(dfd
);
1290 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1293 fprintf(stderr
, Name
": failed to add "
1298 if (i
< content
->array
.raid_disks
1303 } else if (verbose
> 0)
1304 fprintf(stderr
, Name
": added %s "
1306 devices
[j
].devname
, mddev
,
1307 devices
[j
].i
.disk
.raid_disk
);
1308 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1309 fprintf(stderr
, Name
": no uptodate device for "
1314 if (content
->array
.level
== LEVEL_CONTAINER
) {
1316 fprintf(stderr
, Name
": Container %s has been "
1317 "assembled with %d drive%s",
1318 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1319 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1320 fprintf(stderr
, " (out of %d)",
1321 content
->array
.raid_disks
);
1322 fprintf(stderr
, "\n");
1324 st
->ss
->free_super(st
);
1325 sysfs_uevent(content
, "change");
1326 wait_for(chosen_name
, mdfd
);
1334 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1335 content
->array
.layout
, clean
, avail
, okcnt
) &&
1336 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1338 /* This array is good-to-go.
1339 * If a reshape is in progress then we might need to
1340 * continue monitoring it. In that case we start
1341 * it read-only and let the grow code make it writable.
1345 if (content
->reshape_active
&&
1346 content
->delta_disks
<= 0)
1347 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1350 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1353 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1354 mddev
, okcnt
, okcnt
==1?"":"s");
1355 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1356 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1358 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1360 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1361 fprintf(stderr
, ".\n");
1363 if (content
->reshape_active
&&
1364 content
->array
.level
>= 4 &&
1365 content
->array
.level
<= 6) {
1366 /* might need to increase the size
1367 * of the stripe cache - default is 256
1369 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1370 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1372 sysfs_set_num(sra
, NULL
,
1373 "stripe_cache_size",
1374 (4 * content
->array
.chunk_size
/ 4096) + 1);
1377 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1378 /* If any devices did not get added
1379 * because the kernel rejected them based
1380 * on event count, try adding them
1381 * again providing the action policy is
1382 * 're-add' or greater. The bitmap
1383 * might allow them to be included, or
1384 * they will become spares.
1386 for (i
= 0; i
<= bestcnt
; i
++) {
1388 if (j
>= 0 && !devices
[j
].uptodate
) {
1389 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1391 rv
= add_disk(mdfd
, st
, content
,
1393 if (rv
== 0 && verbose
>= 0)
1395 Name
": %s has been re-added.\n",
1396 devices
[j
].devname
);
1400 wait_for(mddev
, mdfd
);
1404 /* There is a nasty race with 'mdadm --monitor'.
1405 * If it opens this device before we close it,
1406 * it gets an incomplete open on which IO
1407 * doesn't work and the capacity is
1409 * If we reopen (to check for layered devices)
1410 * before --monitor closes, we loose.
1412 * So: wait upto 1 second for there to be
1413 * a non-zero capacity.
1415 while (usecs
< 1000) {
1416 mdfd
= open(mddev
, O_RDONLY
);
1418 unsigned long long size
;
1419 if (get_dev_size(mdfd
, NULL
, &size
) &&
1431 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1432 mddev
, strerror(errno
));
1434 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1435 content
->array
.layout
, 1, avail
, okcnt
))
1436 fprintf(stderr
, Name
": Not enough devices to "
1437 "start the array.\n");
1438 else if (!enough(content
->array
.level
,
1439 content
->array
.raid_disks
,
1440 content
->array
.layout
, clean
,
1442 fprintf(stderr
, Name
": Not enough devices to "
1443 "start the array while not clean "
1444 "- consider --force.\n");
1447 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1452 if (runstop
== -1) {
1453 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1454 mddev
, okcnt
, okcnt
==1?"":"s");
1455 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1456 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1457 fprintf(stderr
, ", but not started.\n");
1462 if (verbose
>= -1) {
1463 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1465 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1467 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1468 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1469 content
->array
.layout
, 1, avail
, okcnt
))
1470 fprintf(stderr
, " - not enough to start the array.\n");
1471 else if (!enough(content
->array
.level
,
1472 content
->array
.raid_disks
,
1473 content
->array
.layout
, clean
,
1475 fprintf(stderr
, " - not enough to start the "
1476 "array while not clean - consider "
1479 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1480 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1482 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1483 fprintf(stderr
, " (use --run to insist).\n");
1487 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1492 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1493 * been updated to point to the current locations of devices.
1494 * so we can just start the array
1497 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1498 devices
[chosen_drive
].i
.disk
.minor
);
1499 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1500 fprintf(stderr
, Name
": Cannot start array: %s\n",
1511 int assemble_container_content(struct supertype
*st
, int mdfd
,
1512 struct mdinfo
*content
, int runstop
,
1513 char *chosen_name
, int verbose
,
1516 struct mdinfo
*dev
, *sra
;
1517 int working
= 0, preexist
= 0;
1519 struct map_ent
*map
= NULL
;
1522 sysfs_init(content
, mdfd
, 0);
1524 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1525 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1526 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0)
1529 if (content
->reshape_active
)
1530 block_subarray(content
);
1534 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1535 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1536 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1537 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1538 content
->reshape_active
)
1542 } else if (errno
== EEXIST
)
1545 return 1;/* Nothing new, don't try to start */
1547 map_update(&map
, fd2devnum(mdfd
),
1548 content
->text_version
,
1549 content
->uuid
, chosen_name
);
1552 (working
+ preexist
+ expansion
) >=
1553 content
->array
.working_disks
) {
1556 if (content
->reshape_active
) {
1557 int spare
= content
->array
.raid_disks
+ expansion
;
1559 int *fdlist
= malloc(sizeof(int) *
1560 (working
+ expansion
1561 + content
->array
.raid_disks
));
1562 for (i
=0; i
<spare
; i
++)
1564 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
1567 sprintf(buf
, "%d:%d",
1570 fd
= dev_open(buf
, O_RDWR
);
1572 if (dev
->disk
.raid_disk
>= 0)
1573 fdlist
[dev
->disk
.raid_disk
] = fd
;
1575 fdlist
[spare
++] = fd
;
1577 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1578 err
= st
->ss
->recover_backup(st
, content
);
1580 err
= Grow_restart(st
, content
, fdlist
, spare
,
1581 backup_file
, verbose
> 0);
1584 if (fdlist
[spare
] >= 0)
1585 close(fdlist
[spare
]);
1588 fprintf(stderr
, Name
": Failed to restore critical"
1589 " section for reshape - sorry.\n");
1591 fprintf(stderr
, Name
": Possibly you need"
1592 " to specify a --backup-file\n");
1596 err
= Grow_continue(mdfd
, st
, content
, backup_file
);
1597 } else switch(content
->array
.level
) {
1599 case LEVEL_MULTIPATH
:
1601 err
= sysfs_set_str(content
, NULL
, "array_state",
1605 err
= sysfs_set_str(content
, NULL
, "array_state",
1607 /* start mdmon if needed. */
1609 if (!mdmon_running(st
->container_dev
))
1610 start_mdmon(st
->container_dev
);
1611 ping_monitor_by_id(st
->container_dev
);
1616 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1619 fprintf(stderr
, Name
1620 ": array %s now has %d devices",
1621 chosen_name
, working
+ preexist
);
1623 fprintf(stderr
, Name
1624 ": Started %s with %d devices",
1625 chosen_name
, working
+ preexist
);
1627 fprintf(stderr
, " (%d new)", working
);
1629 fprintf(stderr
, " ( + %d for expansion)",
1631 fprintf(stderr
, "\n");
1634 wait_for(chosen_name
, mdfd
);
1636 /* FIXME should have an O_EXCL and wait for read-auto */
1639 fprintf(stderr
, Name
1640 ": %s assembled with %d devices but "
1642 chosen_name
, working
);