]>
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
,
137 char *backup_file
, int invalid_backup
,
138 int readonly
, int runstop
,
139 char *update
, char *homehost
, int require_homehost
,
140 int verbose
, int force
)
143 * The task of Assemble is to find a collection of
144 * devices that should (according to their superblocks)
145 * form an array, and to give this collection to the MD driver.
146 * In Linux-2.4 and later, this involves submitting a
147 * SET_ARRAY_INFO ioctl with no arg - to prepare
148 * the array - and then submit a number of
149 * ADD_NEW_DISK ioctls to add disks into
150 * the array. Finally RUN_ARRAY might
151 * be submitted to start the array.
153 * Much of the work of Assemble is in finding and/or
154 * checking the disks to make sure they look right.
156 * If mddev is not set, then scan must be set and we
157 * read through the config file for dev+uuid mapping
158 * We recurse, setting mddev, for each device that
160 * - has a valid uuid (or any uuid if !uuidset)
162 * If mddev is set, we try to determine state of md.
163 * check version - must be at least 0.90.0
164 * check kernel version. must be at least 2.4.
165 * If not, we can possibly fall back on START_ARRAY
166 * Try to GET_ARRAY_INFO.
167 * If possible, give up
168 * If not, try to STOP_ARRAY just to make sure
170 * If !uuidset and scan, look in conf-file for uuid
171 * If not found, give up
172 * If !devlist and scan and uuidset, get list of devs from conf-file
175 * Check superblock - discard if bad
176 * Check uuid (set if we don't have one) - discard if no match
177 * Check superblock similarity if we have a superblock - discard if different
178 * Record events, devicenum
179 * This should give us a list of devices for the array
180 * We should collect the most recent event number
182 * Count disks with recent enough event count
183 * While force && !enough disks
184 * Choose newest rejected disks, update event count
185 * mark clean and rewrite superblock
188 * foreach device with recent events : ADD_NEW_DISK
189 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
191 * Check the device numbers in superblock are right
192 * update superblock if any changes
198 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
199 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
200 && (ident
->container
== NULL
|| ident
->member
== NULL
));
202 int vers
= vers
; /* Keep gcc quite - it really is initialised */
205 int uptodate
; /* set once we decide that this device is as
206 * recent as everything else in the array.
211 int *best
= NULL
; /* indexed by raid_disk */
214 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
215 unsigned int req_cnt
;
221 int report_missmatch
;
223 int start_partial_ok
= (runstop
>= 0) &&
224 (force
|| devlist
==NULL
|| auto_assem
);
225 unsigned int num_devs
;
226 struct mddev_dev
*tmpdev
;
228 struct mdinfo
*content
= NULL
;
233 char chosen_name
[1024];
235 if (get_linux_version() < 2004000)
239 * If any subdevs are listed, then any that don't
240 * match ident are discarded. Remainder must all match and
242 * If no subdevs, then we scan all devices in the config file, but
243 * there must be something in the identity
247 ident
->uuid_set
== 0 &&
248 (ident
->super_minor
< 0 || ident
->super_minor
== UnSet
) &&
249 ident
->name
[0] == 0 &&
250 (ident
->container
== NULL
|| ident
->member
== NULL
) &&
251 ident
->devices
== NULL
) {
252 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
253 mddev
? mddev
: "further assembly");
258 devlist
= conf_get_devs();
262 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
264 /* We come back here when doing auto-assembly and attempting some
265 * set of devices failed. Those are now marked as ->used==2 and
266 * we ignore them and try again
269 tmpdev
= devlist
; num_devs
= 0;
275 tmpdev
= tmpdev
->next
;
278 if (!st
&& ident
->st
) st
= ident
->st
;
281 fprintf(stderr
, Name
": looking for devices for %s\n",
282 mddev
? mddev
: "further assembly");
284 /* first walk the list of devices to find a consistent set
285 * that match the criterea, if that is possible.
286 * We flag the ones we like with 'used'.
288 for (tmpdev
= devlist
;
290 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
291 char *devname
= tmpdev
->devname
;
294 struct supertype
*tst
= dup_super(st
);
295 struct dev_policy
*pol
= NULL
;
296 int found_container
= 0;
298 if (tmpdev
->used
> 1) continue;
300 if (ident
->devices
&&
301 !match_oneof(ident
->devices
, devname
)) {
302 if (report_missmatch
)
303 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
307 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
309 if (report_missmatch
)
310 fprintf(stderr
, Name
": cannot open device %s: %s\n",
311 devname
, strerror(errno
));
313 } else if (fstat(dfd
, &stb
)< 0) {
315 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
316 devname
, strerror(errno
));
318 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
319 fprintf(stderr
, Name
": %s is not a block device.\n",
322 } else if (must_be_container(dfd
)) {
324 /* already found some components, this cannot
327 if (report_missmatch
)
328 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
331 } if (!tst
&& (tst
= super_by_fd(dfd
, NULL
)) == NULL
) {
332 if (report_missmatch
)
333 fprintf(stderr
, Name
": not a recognisable container: %s\n",
336 } else if (!tst
->ss
->load_container
337 || tst
->ss
->load_container(tst
, dfd
, NULL
)) {
338 if (report_missmatch
)
339 fprintf(stderr
, Name
": no correct container type: %s\n",
342 } else if (auto_assem
&&
343 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
344 tst
->ss
->match_home(tst
, homehost
) == 1)) {
345 if (report_missmatch
)
346 fprintf(stderr
, Name
": %s has metadata type %s for which "
347 "auto-assembly is disabled\n",
348 devname
, tst
->ss
->name
);
353 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
354 if (report_missmatch
)
355 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
358 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
359 if (report_missmatch
)
360 fprintf(stderr
, Name
": no RAID superblock on %s\n",
363 } else if (tst
->ss
->compare_super
== NULL
) {
364 if (report_missmatch
)
365 fprintf(stderr
, Name
": Cannot assemble %s metadata on %s\n",
366 tst
->ss
->name
, devname
);
368 } else if (auto_assem
&& st
== NULL
&&
369 !conf_test_metadata(tst
->ss
->name
, (pol
= devnum_policy(stb
.st_rdev
)),
370 tst
->ss
->match_home(tst
, homehost
) == 1)) {
371 if (report_missmatch
)
372 fprintf(stderr
, Name
": %s has metadata type %s for which "
373 "auto-assembly is disabled\n",
374 devname
, tst
->ss
->name
);
378 if (dfd
>= 0) close(dfd
);
379 if (tmpdev
->used
== 2) {
381 /* Ignore unrecognised devices during auto-assembly */
383 if (ident
->uuid_set
|| ident
->name
[0] ||
384 ident
->super_minor
!= UnSet
)
385 /* Ignore unrecognised device if looking for
390 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
393 st
->ss
->free_super(st
);
394 dev_policy_free(pol
);
398 if (found_container
) {
399 /* tmpdev is a container. We need to be either
400 * looking for a member, or auto-assembling
403 if (ident
->container
) {
404 if (ident
->container
[0] == '/' &&
405 !same_dev(ident
->container
, devname
)) {
406 if (report_missmatch
)
407 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
408 devname
, ident
->container
);
411 if (ident
->container
[0] != '/') {
416 memset(content
, 0, sizeof(*content
));
417 tst
->ss
->getinfo_super(tst
, content
, NULL
);
419 if (!parse_uuid(ident
->container
, uuid
) ||
420 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
421 if (report_missmatch
)
422 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
428 /* It is worth looking inside this container.
431 fprintf(stderr
, Name
": looking in container %s\n",
434 for (content
= tst
->ss
->container_content(tst
, NULL
);
436 content
= content
->next
) {
438 if (!ident_matches(ident
, content
, tst
,
440 report_missmatch
? devname
: NULL
))
441 /* message already printed */;
442 else if (is_member_busy(content
->text_version
)) {
443 if (report_missmatch
)
444 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
445 content
->text_version
,
452 goto loop
; /* empty container */
455 st
= tst
; tst
= NULL
;
456 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
457 fprintf(stderr
, Name
": %s is a container, but is not "
458 "only device given: confused and aborting\n",
460 st
->ss
->free_super(st
);
461 dev_policy_free(pol
);
465 fprintf(stderr
, Name
": found match on member %s in %s\n",
466 content
->text_version
, devname
);
468 /* make sure we finished the loop */
474 memset(content
, 0, sizeof(*content
));
475 tst
->ss
->getinfo_super(tst
, content
, NULL
);
477 if (!ident_matches(ident
, content
, tst
,
479 report_missmatch
? devname
: NULL
))
484 if (st
->minor_version
== -1)
485 st
->minor_version
= tst
->minor_version
;
486 if (st
->ss
!= tst
->ss
||
487 st
->minor_version
!= tst
->minor_version
||
488 st
->ss
->compare_super(st
, tst
) != 0) {
489 /* Some mismatch. If exactly one array matches this host,
490 * we can resolve on that one.
491 * Or, if we are auto assembling, we just ignore the second
497 int first
= st
->ss
->match_home(st
, homehost
);
498 int last
= tst
->ss
->match_home(tst
, homehost
);
500 (first
== 1 || last
== 1)) {
501 /* We can do something */
502 if (first
) {/* just ignore this one */
503 if (report_missmatch
)
504 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
507 } else { /* reject all those sofar */
508 struct mddev_dev
*td
;
509 if (report_missmatch
)
510 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
512 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
520 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
522 tst
->ss
->free_super(tst
);
523 st
->ss
->free_super(st
);
524 dev_policy_free(pol
);
530 dev_policy_free(pol
);
533 tst
->ss
->free_super(tst
);
536 if (!st
|| !st
->sb
|| !content
)
539 /* Now need to open the array device. Use create_mddev */
540 if (content
== &info
)
541 st
->ss
->getinfo_super(st
, content
, NULL
);
543 trustworthy
= FOREIGN
;
544 name
= content
->name
;
545 switch (st
->ss
->match_home(st
, homehost
)
546 ?: st
->ss
->match_home(st
, "any")) {
549 name
= strchr(content
->name
, ':');
553 name
= content
->name
;
557 /* If the array is listed in mdadm.conf or on
558 * command line, then we trust the name
559 * even if the array doesn't look local
564 content
->array
.level
== LEVEL_CONTAINER
) {
565 name
= content
->text_version
;
566 trustworthy
= METADATA
;
569 if (name
[0] && trustworthy
!= LOCAL
&&
570 ! require_homehost
&&
571 conf_name_is_free(name
))
574 if (trustworthy
== LOCAL
&&
576 /* Ignore 'host:' prefix of name */
577 name
= strchr(name
, ':')+1;
579 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
582 st
->ss
->free_super(st
);
588 vers
= md_get_version(mdfd
);
590 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
591 " Upgrade your kernel or try --build\n");
595 if (mddev_busy(fd2devnum(mdfd
))) {
596 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
598 for (tmpdev
= devlist
;
599 tmpdev
&& tmpdev
->used
!= 1;
600 tmpdev
= tmpdev
->next
)
602 if (tmpdev
&& auto_assem
)
603 fprintf(stderr
, Name
": %s needed for %s...\n",
604 mddev
, tmpdev
->devname
);
607 st
->ss
->free_super(st
);
612 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
615 if (content
!= &info
) {
616 /* This is a member of a container. Try starting the array. */
617 return assemble_container_content(st
, mdfd
, content
, runstop
,
618 chosen_name
, verbose
);
621 /* Ok, no bad inconsistancy, we can try updating etc */
623 content
->update_private
= NULL
;
624 devices
= malloc(num_devs
* sizeof(*devices
));
625 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
626 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
627 char *devname
= tmpdev
->devname
;
629 /* looks like a good enough match to update the super block if needed */
633 /* prepare useful information in info structures */
635 struct supertype
*tst
;
639 if (strcmp(update
, "uuid")==0 &&
642 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
643 read(rfd
, ident
->uuid
, 16) != 16) {
644 *(__u32
*)(ident
->uuid
) = random();
645 *(__u32
*)(ident
->uuid
+1) = random();
646 *(__u32
*)(ident
->uuid
+2) = random();
647 *(__u32
*)(ident
->uuid
+3) = random();
649 if (rfd
>= 0) close(rfd
);
651 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
654 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
655 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
664 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
666 memcpy(content
->uuid
, ident
->uuid
, 16);
667 strcpy(content
->name
, ident
->name
);
668 content
->array
.md_minor
= minor(stb2
.st_rdev
);
670 if (strcmp(update
, "byteorder") == 0)
673 err
= tst
->ss
->update_super(tst
, content
, update
,
679 Name
": --update=%s not understood"
680 " for %s metadata\n",
681 update
, tst
->ss
->name
);
682 tst
->ss
->free_super(tst
);
690 if (strcmp(update
, "uuid")==0 &&
693 memcpy(ident
->uuid
, content
->uuid
, 16);
695 if (tst
->ss
->store_super(tst
, dfd
))
696 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
700 if (strcmp(update
, "uuid")==0 &&
701 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
702 if (bitmap_update_uuid(ident
->bitmap_fd
,
704 tst
->ss
->swapuuid
) != 0)
705 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
709 tst
->ss
->free_super(tst
);
713 struct supertype
*tst
= dup_super(st
);
715 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
717 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
718 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
727 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
728 tst
->ss
->free_super(tst
);
735 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
736 devname
, mddev
, content
->disk
.raid_disk
);
737 devices
[devcnt
].devname
= devname
;
738 devices
[devcnt
].uptodate
= 0;
739 devices
[devcnt
].i
= *content
;
740 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
741 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
742 if (most_recent
< devcnt
) {
743 if (devices
[devcnt
].i
.events
744 > devices
[most_recent
].i
.events
)
745 most_recent
= devcnt
;
747 if (content
->array
.level
== LEVEL_MULTIPATH
)
748 /* with multipath, the raid_disk from the superblock is meaningless */
751 i
= devices
[devcnt
].i
.disk
.raid_disk
;
753 if (nextspare
< content
->array
.raid_disks
)
754 nextspare
= content
->array
.raid_disks
;
757 if (i
>= content
->array
.raid_disks
&&
763 int newbestcnt
= i
+10;
764 int *newbest
= malloc(sizeof(int)*newbestcnt
);
766 for (c
=0; c
< newbestcnt
; c
++)
768 newbest
[c
] = best
[c
];
773 bestcnt
= newbestcnt
;
776 devices
[best
[i
]].i
.events
777 == devices
[devcnt
].i
.events
778 && (devices
[best
[i
]].i
.disk
.minor
779 != devices
[devcnt
].i
.disk
.minor
)
781 && content
->array
.level
!= LEVEL_MULTIPATH
) {
782 /* two different devices with identical superblock.
783 * Could be a mis-detection caused by overlapping
784 * partitions. fail-safe.
786 fprintf(stderr
, Name
": WARNING %s and %s appear"
787 " to have very similar superblocks.\n"
788 " If they are really different, "
789 "please --zero the superblock on one\n"
790 " If they are the same or overlap,"
791 " please remove one from %s.\n",
792 devices
[best
[i
]].devname
, devname
,
793 inargv
? "the list" :
794 "the\n DEVICE list in mdadm.conf"
802 || (devices
[best
[i
]].i
.events
803 < devices
[devcnt
].i
.events
))
808 free(content
->update_private
);
809 content
->update_private
= NULL
;
812 fprintf(stderr
, Name
": no devices found for %s\n",
815 st
->ss
->free_super(st
);
822 if (update
&& strcmp(update
, "byteorder")==0)
823 st
->minor_version
= 90;
825 st
->ss
->getinfo_super(st
, content
, NULL
);
826 clean
= content
->array
.state
& 1;
828 /* now we have some devices that might be suitable.
831 avail
= malloc(content
->array
.raid_disks
);
832 memset(avail
, 0, content
->array
.raid_disks
);
836 for (i
=0; i
< bestcnt
; i
++) {
838 int event_margin
= 1; /* always allow a difference of '1'
839 * like the kernel does
842 /* note: we ignore error flags in multipath arrays
843 * as they don't make sense
845 if (content
->array
.level
!= LEVEL_MULTIPATH
)
846 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
847 if (!(devices
[j
].i
.disk
.state
848 & (1<<MD_DISK_FAULTY
)))
852 /* If this devices thinks that 'most_recent' has failed, then
853 * we must reject this device.
855 if (j
!= most_recent
&&
856 content
->array
.raid_disks
> 0 &&
857 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
858 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
860 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
861 devices
[j
].devname
, devices
[most_recent
].devname
);
865 if (devices
[j
].i
.events
+event_margin
>=
866 devices
[most_recent
].i
.events
) {
867 devices
[j
].uptodate
= 1;
868 if (i
< content
->array
.raid_disks
) {
869 if (devices
[j
].i
.recovery_start
== MaxSector
||
870 (content
->reshape_active
&&
871 j
>= content
->array
.raid_disks
- content
->delta_disks
)) {
881 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
882 content
->array
.layout
, 1,
884 /* Choose the newest best drive which is
885 * not up-to-date, update the superblock
889 struct supertype
*tst
;
890 unsigned long long current_events
;
892 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
895 !devices
[j
].uptodate
&&
896 devices
[j
].i
.recovery_start
== MaxSector
&&
899 > devices
[chosen_drive
].i
.events
))
902 if (chosen_drive
< 0)
904 current_events
= devices
[chosen_drive
].i
.events
;
907 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
908 devices
[chosen_drive
].devname
,
909 devices
[chosen_drive
].i
.disk
.raid_disk
,
910 (int)(devices
[chosen_drive
].i
.events
),
911 (int)(devices
[most_recent
].i
.events
));
912 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
914 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
915 devices
[chosen_drive
].devname
);
916 devices
[chosen_drive
].i
.events
= 0;
920 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
922 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
923 devices
[chosen_drive
].devname
);
924 devices
[chosen_drive
].i
.events
= 0;
927 content
->events
= devices
[most_recent
].i
.events
;
928 tst
->ss
->update_super(tst
, content
, "force-one",
929 devices
[chosen_drive
].devname
, verbose
,
932 if (tst
->ss
->store_super(tst
, fd
)) {
934 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
935 devices
[chosen_drive
].devname
);
936 devices
[chosen_drive
].i
.events
= 0;
937 tst
->ss
->free_super(tst
);
941 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
942 devices
[chosen_drive
].uptodate
= 1;
943 avail
[chosen_drive
] = 1;
945 tst
->ss
->free_super(tst
);
947 /* If there are any other drives of the same vintage,
948 * add them in as well. We can't lose and we might gain
950 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
953 !devices
[j
].uptodate
&&
954 devices
[j
].i
.events
== current_events
) {
961 /* Now we want to look at the superblock which the kernel will base things on
962 * and compare the devices that we think are working with the devices that the
963 * superblock thinks are working.
964 * If there are differences and --force is given, then update this chosen
968 st
->ss
->free_super(st
);
969 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
975 if (!devices
[j
].uptodate
)
978 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
979 fprintf(stderr
, Name
": Cannot open %s: %s\n",
980 devices
[j
].devname
, strerror(errno
));
985 if (st
->ss
->load_super(st
,fd
, NULL
)) {
987 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
995 if (st
->sb
== NULL
) {
996 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
1001 st
->ss
->getinfo_super(st
, content
, NULL
);
1003 sysfs_init(content
, mdfd
, 0);
1005 for (i
=0; i
<bestcnt
; i
++) {
1007 unsigned int desired_state
;
1009 if (i
< content
->array
.raid_disks
)
1010 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1016 if (!devices
[j
].uptodate
)
1019 devices
[j
].i
.disk
.state
= desired_state
;
1020 if (!(devices
[j
].i
.array
.state
& 1))
1023 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1024 verbose
, 0, NULL
)) {
1027 fprintf(stderr
, Name
": "
1028 "clearing FAULTY flag for device %d in %s for %s\n",
1029 j
, mddev
, devices
[j
].devname
);
1033 fprintf(stderr
, Name
": "
1034 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1035 i
, mddev
, devices
[j
].devname
);
1039 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1040 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1045 if (force
&& !clean
&&
1046 !enough(content
->array
.level
, content
->array
.raid_disks
,
1047 content
->array
.layout
, clean
,
1049 change
+= st
->ss
->update_super(st
, content
, "force-array",
1050 devices
[chosen_drive
].devname
, verbose
,
1057 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1059 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1060 devices
[chosen_drive
].devname
);
1065 if (st
->ss
->store_super(st
, fd
)) {
1067 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1068 devices
[chosen_drive
].devname
);
1076 /* If we are in the middle of a reshape we may need to restore saved data
1077 * that was moved aside due to the reshape overwriting live data
1078 * The code of doing this lives in Grow.c
1081 if (content
->reshape_active
) {
1083 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1085 fprintf(stderr
, Name
":%s has an active reshape - checking "
1086 "if critical section needs to be restored\n",
1088 for (i
=0; i
<bestcnt
; i
++) {
1091 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1092 if (fdlist
[i
] < 0) {
1093 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1094 devices
[j
].devname
);
1102 err
= Grow_restart(st
, content
, fdlist
, bestcnt
,
1103 backup_file
, verbose
> 0);
1104 if (err
&& invalid_backup
) {
1106 fprintf(stderr
, Name
": continuing"
1107 " without restoring backup\n");
1113 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1116 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1117 if (backup_file
== NULL
)
1118 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1125 /* count number of in-sync devices according to the superblock.
1126 * We must have this number to start the array without -s or -R
1128 req_cnt
= content
->array
.working_disks
;
1130 /* Almost ready to actually *do* something */
1134 /* First, fill in the map, so that udev can find our name
1135 * as soon as we become active.
1137 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1138 content
->uuid
, chosen_name
);
1140 rv
= set_array_info(mdfd
, st
, content
);
1142 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1143 mddev
, strerror(errno
));
1144 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1149 if (ident
->bitmap_fd
>= 0) {
1150 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1151 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1152 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1157 } else if (ident
->bitmap_file
) {
1158 /* From config file */
1159 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1161 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1162 ident
->bitmap_file
);
1163 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1168 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1169 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1171 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1179 /* First, add the raid disks, but add the chosen one last */
1180 for (i
=0; i
<= bestcnt
; i
++) {
1184 if (j
== chosen_drive
)
1189 if (j
>= 0 /* && devices[j].uptodate */) {
1190 int dfd
= dev_open(devices
[j
].devname
,
1193 remove_partitions(dfd
);
1196 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1199 fprintf(stderr
, Name
": failed to add "
1204 if (i
< content
->array
.raid_disks
1209 } else if (verbose
> 0)
1210 fprintf(stderr
, Name
": added %s "
1212 devices
[j
].devname
, mddev
,
1213 devices
[j
].i
.disk
.raid_disk
);
1214 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1215 fprintf(stderr
, Name
": no uptodate device for "
1220 if (content
->array
.level
== LEVEL_CONTAINER
) {
1222 fprintf(stderr
, Name
": Container %s has been "
1223 "assembled with %d drive%s",
1224 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1225 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1226 fprintf(stderr
, " (out of %d)",
1227 content
->array
.raid_disks
);
1228 fprintf(stderr
, "\n");
1230 sysfs_uevent(content
, "change");
1231 wait_for(chosen_name
, mdfd
);
1239 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1240 content
->array
.layout
, clean
, avail
, okcnt
) &&
1241 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1243 /* This array is good-to-go.
1244 * If a reshape is in progress then we might need to
1245 * continue monitoring it. In that case we start
1246 * it read-only and let the grow code make it writable.
1250 if (content
->reshape_active
&&
1251 content
->delta_disks
<= 0)
1252 rv
= Grow_continue(mdfd
, st
, content
, backup_file
);
1255 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1258 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1259 mddev
, okcnt
, okcnt
==1?"":"s");
1260 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1261 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1263 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1265 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1266 fprintf(stderr
, ".\n");
1268 if (content
->reshape_active
&&
1269 content
->array
.level
>= 4 &&
1270 content
->array
.level
<= 6) {
1271 /* might need to increase the size
1272 * of the stripe cache - default is 256
1274 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1275 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1277 sysfs_set_num(sra
, NULL
,
1278 "stripe_cache_size",
1279 (4 * content
->array
.chunk_size
/ 4096) + 1);
1282 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1283 /* If any devices did not get added
1284 * because the kernel rejected them based
1285 * on event count, try adding them
1286 * again providing the action policy is
1287 * 're-add' or greater. The bitmap
1288 * might allow them to be included, or
1289 * they will become spares.
1291 for (i
= 0; i
<= bestcnt
; i
++) {
1293 if (j
>= 0 && !devices
[j
].uptodate
) {
1294 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1296 rv
= add_disk(mdfd
, st
, content
,
1298 if (rv
== 0 && verbose
>= 0)
1300 Name
": %s has been re-added.\n",
1301 devices
[j
].devname
);
1305 wait_for(mddev
, mdfd
);
1309 /* There is a nasty race with 'mdadm --monitor'.
1310 * If it opens this device before we close it,
1311 * it gets an incomplete open on which IO
1312 * doesn't work and the capacity is
1314 * If we reopen (to check for layered devices)
1315 * before --monitor closes, we loose.
1317 * So: wait upto 1 second for there to be
1318 * a non-zero capacity.
1320 while (usecs
< 1000) {
1321 mdfd
= open(mddev
, O_RDONLY
);
1323 unsigned long long size
;
1324 if (get_dev_size(mdfd
, NULL
, &size
) &&
1336 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1337 mddev
, strerror(errno
));
1339 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1340 content
->array
.layout
, 1, avail
, okcnt
))
1341 fprintf(stderr
, Name
": Not enough devices to "
1342 "start the array.\n");
1343 else if (!enough(content
->array
.level
,
1344 content
->array
.raid_disks
,
1345 content
->array
.layout
, clean
,
1347 fprintf(stderr
, Name
": Not enough devices to "
1348 "start the array while not clean "
1349 "- consider --force.\n");
1352 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1357 if (runstop
== -1) {
1358 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1359 mddev
, okcnt
, okcnt
==1?"":"s");
1360 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1361 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1362 fprintf(stderr
, ", but not started.\n");
1367 if (verbose
>= -1) {
1368 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1370 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1372 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1373 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1374 content
->array
.layout
, 1, avail
, okcnt
))
1375 fprintf(stderr
, " - not enough to start the array.\n");
1376 else if (!enough(content
->array
.level
,
1377 content
->array
.raid_disks
,
1378 content
->array
.layout
, clean
,
1380 fprintf(stderr
, " - not enough to start the "
1381 "array while not clean - consider "
1384 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1385 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1387 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1388 fprintf(stderr
, " (use --run to insist).\n");
1392 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1397 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1398 * been updated to point to the current locations of devices.
1399 * so we can just start the array
1402 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1403 devices
[chosen_drive
].i
.disk
.minor
);
1404 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1405 fprintf(stderr
, Name
": Cannot start array: %s\n",
1416 int assemble_container_content(struct supertype
*st
, int mdfd
,
1417 struct mdinfo
*content
, int runstop
,
1418 char *chosen_name
, int verbose
)
1420 struct mdinfo
*dev
, *sra
;
1421 int working
= 0, preexist
= 0;
1422 struct map_ent
*map
= NULL
;
1424 sysfs_init(content
, mdfd
, 0);
1426 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1427 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1428 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1435 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1436 if (sysfs_add_disk(content
, dev
, 1) == 0)
1438 else if (errno
== EEXIST
)
1442 return 1;/* Nothing new, don't try to start */
1445 map_update(&map
, fd2devnum(mdfd
),
1446 content
->text_version
,
1447 content
->uuid
, chosen_name
);
1450 (working
+ preexist
) >= content
->array
.working_disks
) {
1453 switch(content
->array
.level
) {
1455 case LEVEL_MULTIPATH
:
1457 err
= sysfs_set_str(content
, NULL
, "array_state",
1461 err
= sysfs_set_str(content
, NULL
, "array_state",
1463 /* start mdmon if needed. */
1465 if (!mdmon_running(st
->container_dev
))
1466 start_mdmon(st
->container_dev
);
1467 ping_monitor(devnum2devname(st
->container_dev
));
1472 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1475 fprintf(stderr
, Name
1476 ": array %s now has %d devices",
1477 chosen_name
, working
+ preexist
);
1479 fprintf(stderr
, Name
1480 ": Started %s with %d devices",
1481 chosen_name
, working
+ preexist
);
1483 fprintf(stderr
, " (%d new)", working
);
1484 fprintf(stderr
, "\n");
1487 wait_for(chosen_name
, mdfd
);
1490 /* FIXME should have an O_EXCL and wait for read-auto */
1493 fprintf(stderr
, Name
1494 ": %s assembled with %d devices but "
1496 chosen_name
, working
);