]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
33 static int name_matches(char *found
, char *required
, char *homehost
)
35 /* See if the name found matches the required name, possibly
36 * prefixed with 'homehost'
40 strncpy(fnd
, found
, 32);
42 if (strcmp(found
, required
)==0)
45 int l
= strlen(homehost
);
46 if (l
< 32 && fnd
[l
] == ':' &&
47 strcmp(fnd
+l
+1, required
)==0)
53 static int is_member_busy(char *metadata_version
)
55 /* check if the given member array is active */
56 struct mdstat_ent
*mdstat
= mdstat_read(1, 0);
57 struct mdstat_ent
*ent
;
60 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
61 if (ent
->metadata_version
== NULL
)
63 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
65 if (!is_subarray(&ent
->metadata_version
[9]))
67 /* Skip first char - it can be '/' or '-' */
68 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
78 int Assemble(struct supertype
*st
, char *mddev
,
80 mddev_dev_t devlist
, char *backup_file
,
81 int readonly
, int runstop
,
82 char *update
, char *homehost
, int require_homehost
,
83 int verbose
, int force
)
86 * The task of Assemble is to find a collection of
87 * devices that should (according to their superblocks)
88 * form an array, and to give this collection to the MD driver.
89 * In Linux-2.4 and later, this involves submitting a
90 * SET_ARRAY_INFO ioctl with no arg - to prepare
91 * the array - and then submit a number of
92 * ADD_NEW_DISK ioctls to add disks into
93 * the array. Finally RUN_ARRAY might
94 * be submitted to start the array.
96 * Much of the work of Assemble is in finding and/or
97 * checking the disks to make sure they look right.
99 * If mddev is not set, then scan must be set and we
100 * read through the config file for dev+uuid mapping
101 * We recurse, setting mddev, for each device that
103 * - has a valid uuid (or any uuid if !uuidset)
105 * If mddev is set, we try to determine state of md.
106 * check version - must be at least 0.90.0
107 * check kernel version. must be at least 2.4.
108 * If not, we can possibly fall back on START_ARRAY
109 * Try to GET_ARRAY_INFO.
110 * If possible, give up
111 * If not, try to STOP_ARRAY just to make sure
113 * If !uuidset and scan, look in conf-file for uuid
114 * If not found, give up
115 * If !devlist and scan and uuidset, get list of devs from conf-file
118 * Check superblock - discard if bad
119 * Check uuid (set if we don't have one) - discard if no match
120 * Check superblock similarity if we have a superblock - discard if different
121 * Record events, devicenum
122 * This should give us a list of devices for the array
123 * We should collect the most recent event number
125 * Count disks with recent enough event count
126 * While force && !enough disks
127 * Choose newest rejected disks, update event count
128 * mark clean and rewrite superblock
131 * foreach device with recent events : ADD_NEW_DISK
132 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
134 * Check the device numbers in superblock are right
135 * update superblock if any changes
141 int auto_assem
= (mddev
== NULL
&& !ident
->uuid_set
&&
142 ident
->super_minor
== UnSet
&& ident
->name
[0] == 0
143 && (ident
->container
== NULL
|| ident
->member
== NULL
));
145 int vers
= vers
; /* Keep gcc quite - it really is initialised */
148 int uptodate
; /* set once we decide that this device is as
149 * recent as everything else in the array.
153 int *best
= NULL
; /* indexed by raid_disk */
154 unsigned int bestcnt
= 0;
156 unsigned int okcnt
, sparecnt
;
157 unsigned int req_cnt
;
163 int report_missmatch
;
165 int start_partial_ok
= (runstop
>= 0) &&
166 (force
|| devlist
==NULL
|| auto_assem
);
167 unsigned int num_devs
;
170 struct mdinfo
*content
= NULL
;
175 char chosen_name
[1024];
177 if (get_linux_version() < 2004000)
181 * If any subdevs are listed, then any that don't
182 * match ident are discarded. Remainder must all match and
184 * If no subdevs, then we scan all devices in the config file, but
185 * there must be something in the identity
189 ident
->uuid_set
== 0 &&
190 ident
->super_minor
< 0 &&
191 ident
->devices
== NULL
) {
192 fprintf(stderr
, Name
": No identity information available for %s - cannot assemble.\n",
193 mddev
? mddev
: "further assembly");
198 devlist
= conf_get_devs();
202 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
204 /* We come back here when doing auto-assembly and attempting some
205 * set of devices failed. Those are now marked as ->used==2 and
206 * we ignore them and try again
209 tmpdev
= devlist
; num_devs
= 0;
215 tmpdev
= tmpdev
->next
;
217 devices
= malloc(num_devs
* sizeof(*devices
));
219 if (!st
&& ident
->st
) st
= ident
->st
;
222 fprintf(stderr
, Name
": looking for devices for %s\n",
223 mddev
? mddev
: "further assembly");
225 /* first walk the list of devices to find a consistent set
226 * that match the criterea, if that is possible.
227 * We flag the ones we like with 'used'.
229 for (tmpdev
= devlist
;
231 tmpdev
= tmpdev
->next
) {
232 char *devname
= tmpdev
->devname
;
235 struct supertype
*tst
= dup_super(st
);
237 if (tmpdev
->used
> 1) continue;
239 if (ident
->devices
&&
240 !match_oneof(ident
->devices
, devname
)) {
241 if (report_missmatch
)
242 fprintf(stderr
, Name
": %s is not one of %s\n", devname
, ident
->devices
);
246 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
248 if (report_missmatch
)
249 fprintf(stderr
, Name
": cannot open device %s: %s\n",
250 devname
, strerror(errno
));
252 } else if (fstat(dfd
, &stb
)< 0) {
254 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
255 devname
, strerror(errno
));
257 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
258 fprintf(stderr
, Name
": %s is not a block device.\n",
261 } else if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
262 if (report_missmatch
)
263 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
266 } else if (auto_assem
&& st
== NULL
&&
267 !conf_test_metadata(tst
->ss
->name
)) {
268 if (report_missmatch
)
269 fprintf(stderr
, Name
": %s has metadata type %s for which "
270 "auto-assembly is disabled\n",
271 devname
, tst
->ss
->name
);
273 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
274 if (report_missmatch
)
275 fprintf( stderr
, Name
": no RAID superblock on %s\n",
279 memset(content
, 0, sizeof(*content
));
280 tst
->ss
->getinfo_super(tst
, content
);
282 if (dfd
>= 0) close(dfd
);
284 if (tst
&& tst
->sb
&& tst
->ss
->container_content
285 && tst
->loaded_container
) {
286 /* tmpdev is a container. We need to be either
287 * looking for a member, or auto-assembling
290 /* already found some components, this cannot
293 if (report_missmatch
)
294 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
299 if (ident
->container
) {
300 if (ident
->container
[0] == '/' &&
301 !same_dev(ident
->container
, devname
)) {
302 if (report_missmatch
)
303 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
304 devname
, ident
->container
);
307 if (ident
->container
[0] != '/') {
310 if (!parse_uuid(ident
->container
, uuid
) ||
311 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
312 if (report_missmatch
)
313 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
319 /* It is worth looking inside this container.
323 content
= tmpdev
->content
;
325 content
= tst
->ss
->container_content(tst
);
327 tmpdev
->content
= content
->next
;
328 if (tmpdev
->content
== NULL
)
331 } else if (ident
->container
|| ident
->member
) {
332 /* No chance of this matching if we don't have
334 if (report_missmatch
)
335 fprintf(stderr
, Name
"%s is not a container, and one is required.\n",
340 if (ident
->uuid_set
&& (!update
|| strcmp(update
, "uuid")!= 0) &&
342 same_uuid(content
->uuid
, ident
->uuid
, tst
->ss
->swapuuid
)==0)) {
343 if (report_missmatch
)
344 fprintf(stderr
, Name
": %s has wrong uuid.\n",
348 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
350 name_matches(content
->name
, ident
->name
, homehost
)==0)) {
351 if (report_missmatch
)
352 fprintf(stderr
, Name
": %s has wrong name.\n",
356 if (ident
->super_minor
!= UnSet
&&
358 ident
->super_minor
!= content
->array
.md_minor
)) {
359 if (report_missmatch
)
360 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
364 if (ident
->level
!= UnSet
&&
366 ident
->level
!= content
->array
.level
)) {
367 if (report_missmatch
)
368 fprintf(stderr
, Name
": %s has wrong raid level.\n",
372 if (ident
->raid_disks
!= UnSet
&&
374 ident
->raid_disks
!= content
->array
.raid_disks
)) {
375 if (report_missmatch
)
376 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
381 if (tst
== NULL
|| tst
->sb
== NULL
)
384 /* If we are this far, then we are nearly commited to this device.
385 * If the super_block doesn't exist, or doesn't match others,
386 * then we probably cannot continue
387 * However if one of the arrays is for the homehost, and
388 * the other isn't that can disambiguate.
391 if (!tst
|| !tst
->sb
) {
392 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
395 st
->ss
->free_super(st
);
399 if (tst
&& tst
->sb
&& tst
->ss
->container_content
400 && tst
->loaded_container
) {
401 /* we have the one container we need, don't keep
402 * looking. If the chosen member is active, skip.
404 if (is_member_busy(content
->text_version
)) {
405 if (report_missmatch
)
406 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
407 content
->text_version
,
409 tst
->ss
->free_super(tst
);
416 st
= tst
; tst
= NULL
;
417 if (!auto_assem
&& tmpdev
->next
!= NULL
) {
418 fprintf(stderr
, Name
": %s is a container, but is not "
419 "only device given: confused and aborting\n",
421 st
->ss
->free_super(st
);
428 if (st
->minor_version
== -1)
429 st
->minor_version
= tst
->minor_version
;
430 if (st
->ss
!= tst
->ss
||
431 st
->minor_version
!= tst
->minor_version
||
432 st
->ss
->compare_super(st
, tst
) != 0) {
433 /* Some mismatch. If exactly one array matches this host,
434 * we can resolve on that one.
435 * Or, if we are auto assembling, we just ignore the second
441 int first
= st
->ss
->match_home(st
, homehost
);
442 int last
= tst
->ss
->match_home(tst
, homehost
);
444 (first
== 1 || last
== 1)) {
445 /* We can do something */
446 if (first
) {/* just ignore this one */
447 if (report_missmatch
)
448 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
451 } else { /* reject all those sofar */
453 if (report_missmatch
)
454 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
456 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
464 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
466 tst
->ss
->free_super(tst
);
467 st
->ss
->free_super(st
);
477 tst
->ss
->free_super(tst
);
480 if (!st
|| !st
->sb
|| !content
)
483 /* Now need to open the array device. Use create_mddev */
484 if (content
== &info
)
485 st
->ss
->getinfo_super(st
, content
);
487 trustworthy
= FOREIGN
;
488 name
= content
->name
;
489 switch (st
->ss
->match_home(st
, homehost
)
490 ?: st
->ss
->match_home(st
, "any")) {
493 name
= strchr(content
->name
, ':');
497 name
= content
->name
;
501 /* If the array is listed in mdadm.conf or on
502 * command line, then we trust the name
503 * even if the array doesn't look local
508 content
->array
.level
== LEVEL_CONTAINER
) {
509 name
= content
->text_version
;
510 trustworthy
= METADATA
;
513 if (name
[0] && trustworthy
!= LOCAL
&&
514 ! require_homehost
&&
515 conf_name_is_free(name
))
518 if (trustworthy
== LOCAL
&&
520 /* Ignore 'host:' prefix of name */
521 name
= strchr(name
, ':')+1;
523 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
526 st
->ss
->free_super(st
);
533 vers
= md_get_version(mdfd
);
535 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
536 " Upgrade your kernel or try --build\n");
540 if (mddev_busy(fd2devnum(mdfd
))) {
541 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
543 for (tmpdev
= devlist
;
544 tmpdev
&& tmpdev
->used
!= 1;
545 tmpdev
= tmpdev
->next
)
547 if (tmpdev
&& auto_assem
)
548 fprintf(stderr
, Name
": %s needed for %s...\n",
549 mddev
, tmpdev
->devname
);
552 st
->ss
->free_super(st
);
558 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
561 if (content
!= &info
) {
562 /* This is a member of a container. Try starting the array. */
563 return assemble_container_content(st
, mdfd
, content
, runstop
,
564 chosen_name
, verbose
);
567 /* Ok, no bad inconsistancy, we can try updating etc */
569 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
570 char *devname
= tmpdev
->devname
;
572 /* looks like a good enough match to update the super block if needed */
576 /* prepare useful information in info structures */
578 struct supertype
*tst
;
581 if (strcmp(update
, "uuid")==0 &&
584 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
585 read(rfd
, ident
->uuid
, 16) != 16) {
586 *(__u32
*)(ident
->uuid
) = random();
587 *(__u32
*)(ident
->uuid
+1) = random();
588 *(__u32
*)(ident
->uuid
+2) = random();
589 *(__u32
*)(ident
->uuid
+3) = random();
591 if (rfd
>= 0) close(rfd
);
593 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
595 remove_partitions(dfd
);
598 tst
->ss
->load_super(tst
, dfd
, NULL
);
599 tst
->ss
->getinfo_super(tst
, content
);
601 memcpy(content
->uuid
, ident
->uuid
, 16);
602 strcpy(content
->name
, ident
->name
);
603 content
->array
.md_minor
= minor(stb2
.st_rdev
);
605 tst
->ss
->update_super(tst
, content
, update
,
607 ident
->uuid_set
, homehost
);
608 if (strcmp(update
, "uuid")==0 &&
611 memcpy(ident
->uuid
, content
->uuid
, 16);
614 fprintf(stderr
, Name
": Cannot open %s for superblock update\n",
616 else if (tst
->ss
->store_super(tst
, dfd
))
617 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
622 if (strcmp(update
, "uuid")==0 &&
623 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
624 if (bitmap_update_uuid(ident
->bitmap_fd
,
626 tst
->ss
->swapuuid
) != 0)
627 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
631 tst
->ss
->free_super(tst
);
635 struct supertype
*tst
= dup_super(st
);
637 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
639 remove_partitions(dfd
);
641 tst
->ss
->load_super(tst
, dfd
, NULL
);
642 tst
->ss
->getinfo_super(tst
, content
);
643 tst
->ss
->free_super(tst
);
650 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
651 devname
, mddev
, content
->disk
.raid_disk
);
652 devices
[devcnt
].devname
= devname
;
653 devices
[devcnt
].uptodate
= 0;
654 devices
[devcnt
].i
= *content
;
655 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
656 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
657 if (most_recent
< devcnt
) {
658 if (devices
[devcnt
].i
.events
659 > devices
[most_recent
].i
.events
)
660 most_recent
= devcnt
;
662 if (content
->array
.level
== -4)
663 /* with multipath, the raid_disk from the superblock is meaningless */
666 i
= devices
[devcnt
].i
.disk
.raid_disk
;
668 if (nextspare
< content
->array
.raid_disks
)
669 nextspare
= content
->array
.raid_disks
;
672 if (i
>= content
->array
.raid_disks
&&
678 unsigned int newbestcnt
= i
+10;
679 int *newbest
= malloc(sizeof(int)*newbestcnt
);
681 for (c
=0; c
< newbestcnt
; c
++)
683 newbest
[c
] = best
[c
];
688 bestcnt
= newbestcnt
;
691 devices
[best
[i
]].i
.events
692 == devices
[devcnt
].i
.events
693 && (devices
[best
[i
]].i
.disk
.minor
694 != devices
[devcnt
].i
.disk
.minor
)
696 && content
->array
.level
!= LEVEL_MULTIPATH
) {
697 /* two different devices with identical superblock.
698 * Could be a mis-detection caused by overlapping
699 * partitions. fail-safe.
701 fprintf(stderr
, Name
": WARNING %s and %s appear"
702 " to have very similar superblocks.\n"
703 " If they are really different, "
704 "please --zero the superblock on one\n"
705 " If they are the same or overlap,"
706 " please remove one from %s.\n",
707 devices
[best
[i
]].devname
, devname
,
708 inargv
? "the list" :
709 "the\n DEVICE list in mdadm.conf"
715 || (devices
[best
[i
]].i
.events
716 < devices
[devcnt
].i
.events
))
723 fprintf(stderr
, Name
": no devices found for %s\n",
726 st
->ss
->free_super(st
);
731 if (update
&& strcmp(update
, "byteorder")==0)
732 st
->minor_version
= 90;
734 st
->ss
->getinfo_super(st
, content
);
735 clean
= content
->array
.state
& 1;
737 /* now we have some devices that might be suitable.
740 avail
= malloc(content
->array
.raid_disks
);
741 memset(avail
, 0, content
->array
.raid_disks
);
744 for (i
=0; i
< bestcnt
;i
++) {
746 int event_margin
= 1; /* always allow a difference of '1'
747 * like the kernel does
750 /* note: we ignore error flags in multipath arrays
751 * as they don't make sense
753 if (content
->array
.level
!= -4)
754 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_SYNC
))) {
755 if (!(devices
[j
].i
.disk
.state
756 & (1<<MD_DISK_FAULTY
)))
760 if (devices
[j
].i
.events
+event_margin
>=
761 devices
[most_recent
].i
.events
) {
762 devices
[j
].uptodate
= 1;
763 if (i
< content
->array
.raid_disks
) {
770 while (force
&& !enough(content
->array
.level
, content
->array
.raid_disks
,
771 content
->array
.layout
, 1,
773 /* Choose the newest best drive which is
774 * not up-to-date, update the superblock
778 struct supertype
*tst
;
779 long long current_events
;
781 for (i
=0; i
<content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
784 !devices
[j
].uptodate
&&
787 > devices
[chosen_drive
].i
.events
))
790 if (chosen_drive
< 0)
792 current_events
= devices
[chosen_drive
].i
.events
;
795 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
796 devices
[chosen_drive
].devname
,
797 devices
[chosen_drive
].i
.disk
.raid_disk
,
798 (int)(devices
[chosen_drive
].i
.events
),
799 (int)(devices
[most_recent
].i
.events
));
800 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
802 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
803 devices
[chosen_drive
].devname
);
804 devices
[chosen_drive
].i
.events
= 0;
808 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
810 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
811 devices
[chosen_drive
].devname
);
812 devices
[chosen_drive
].i
.events
= 0;
815 content
->events
= devices
[most_recent
].i
.events
;
816 tst
->ss
->update_super(tst
, content
, "force-one",
817 devices
[chosen_drive
].devname
, verbose
,
820 if (tst
->ss
->store_super(tst
, fd
)) {
822 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
823 devices
[chosen_drive
].devname
);
824 devices
[chosen_drive
].i
.events
= 0;
825 tst
->ss
->free_super(tst
);
829 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
830 devices
[chosen_drive
].uptodate
= 1;
831 avail
[chosen_drive
] = 1;
833 tst
->ss
->free_super(tst
);
835 /* If there are any other drives of the same vintage,
836 * add them in as well. We can't lose and we might gain
838 for (i
=0; i
<content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
841 !devices
[j
].uptodate
&&
842 devices
[j
].i
.events
== current_events
) {
849 /* Now we want to look at the superblock which the kernel will base things on
850 * and compare the devices that we think are working with the devices that the
851 * superblock thinks are working.
852 * If there are differences and --force is given, then update this chosen
856 st
->ss
->free_super(st
);
857 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
863 if (!devices
[j
].uptodate
)
866 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
867 fprintf(stderr
, Name
": Cannot open %s: %s\n",
868 devices
[j
].devname
, strerror(errno
));
872 if (st
->ss
->load_super(st
,fd
, NULL
)) {
874 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
881 if (st
->sb
== NULL
) {
882 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
886 st
->ss
->getinfo_super(st
, content
);
888 sysfs_init(content
, mdfd
, 0);
890 for (i
=0; i
<bestcnt
; i
++) {
892 unsigned int desired_state
;
894 if (i
< content
->array
.raid_disks
)
895 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
901 if (!devices
[j
].uptodate
)
904 devices
[j
].i
.disk
.state
= desired_state
;
905 if (!(devices
[j
].i
.array
.state
& 1))
908 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
912 fprintf(stderr
, Name
": "
913 "clearing FAULTY flag for device %d in %s for %s\n",
914 j
, mddev
, devices
[j
].devname
);
918 fprintf(stderr
, Name
": "
919 "device %d in %s has wrong state in superblock, but %s seems ok\n",
920 i
, mddev
, devices
[j
].devname
);
924 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
925 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
930 if (force
&& !clean
&&
931 !enough(content
->array
.level
, content
->array
.raid_disks
,
932 content
->array
.layout
, clean
,
934 change
+= st
->ss
->update_super(st
, content
, "force-array",
935 devices
[chosen_drive
].devname
, verbose
,
942 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
944 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
945 devices
[chosen_drive
].devname
);
949 if (st
->ss
->store_super(st
, fd
)) {
951 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
952 devices
[chosen_drive
].devname
);
959 /* If we are in the middle of a reshape we may need to restore saved data
960 * that was moved aside due to the reshape overwriting live data
961 * The code of doing this lives in Grow.c
964 if (content
->reshape_active
) {
966 int *fdlist
= malloc(sizeof(int)* bestcnt
);
967 for (i
=0; i
<bestcnt
; i
++) {
970 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
972 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
981 err
= Grow_restart(st
, content
, fdlist
, bestcnt
, backup_file
);
984 if (fdlist
[i
]>=0) close(fdlist
[i
]);
987 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
993 /* count number of in-sync devices according to the superblock.
994 * We must have this number to start the array without -s or -R
996 req_cnt
= content
->array
.working_disks
;
998 /* Almost ready to actually *do* something */
1002 /* First, fill in the map, so that udev can find our name
1003 * as soon as we become active.
1005 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1006 content
->uuid
, chosen_name
);
1008 rv
= set_array_info(mdfd
, st
, content
);
1010 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1011 mddev
, strerror(errno
));
1015 if (ident
->bitmap_fd
>= 0) {
1016 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1017 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1021 } else if (ident
->bitmap_file
) {
1022 /* From config file */
1023 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1025 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1026 ident
->bitmap_file
);
1030 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1031 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1039 /* First, add the raid disks, but add the chosen one last */
1040 for (i
=0; i
<= bestcnt
; i
++) {
1044 if (j
== chosen_drive
)
1049 if (j
>= 0 /* && devices[j].uptodate */) {
1050 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1053 fprintf(stderr
, Name
": failed to add "
1058 if (i
< content
->array
.raid_disks
1063 } else if (verbose
> 0)
1064 fprintf(stderr
, Name
": added %s "
1066 devices
[j
].devname
, mddev
,
1067 devices
[j
].i
.disk
.raid_disk
);
1068 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1069 fprintf(stderr
, Name
": no uptodate device for "
1074 if (content
->array
.level
== LEVEL_CONTAINER
) {
1076 fprintf(stderr
, Name
": Container %s has been "
1077 "assembled with %d drive%s",
1078 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1079 if (okcnt
< content
->array
.raid_disks
)
1080 fprintf(stderr
, " (out of %d)",
1081 content
->array
.raid_disks
);
1082 fprintf(stderr
, "\n");
1084 sysfs_uevent(content
, "change");
1085 wait_for(chosen_name
, mdfd
);
1092 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1093 content
->array
.layout
, clean
, avail
, okcnt
) &&
1094 (okcnt
>= req_cnt
|| start_partial_ok
)
1096 if (ioctl(mdfd
, RUN_ARRAY
, NULL
)==0) {
1098 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1099 mddev
, okcnt
, okcnt
==1?"":"s");
1100 if (okcnt
< content
->array
.raid_disks
)
1101 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1103 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1104 fprintf(stderr
, ".\n");
1106 if (content
->reshape_active
&&
1107 content
->array
.level
>= 4 &&
1108 content
->array
.level
<= 6) {
1109 /* might need to increase the size
1110 * of the stripe cache - default is 256
1112 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1113 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1115 sysfs_set_num(sra
, NULL
,
1116 "stripe_cache_size",
1117 (4 * content
->array
.chunk_size
/ 4096) + 1);
1120 wait_for(mddev
, mdfd
);
1124 /* There is a nasty race with 'mdadm --monitor'.
1125 * If it opens this device before we close it,
1126 * it gets an incomplete open on which IO
1127 * doesn't work and the capacity is
1129 * If we reopen (to check for layered devices)
1130 * before --monitor closes, we loose.
1132 * So: wait upto 1 second for there to be
1133 * a non-zero capacity.
1135 while (usecs
< 1000) {
1136 mdfd
= open(mddev
, O_RDONLY
);
1138 unsigned long long size
;
1139 if (get_dev_size(mdfd
, NULL
, &size
) &&
1150 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1151 mddev
, strerror(errno
));
1153 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1154 content
->array
.layout
, 1, avail
, okcnt
))
1155 fprintf(stderr
, Name
": Not enough devices to "
1156 "start the array.\n");
1157 else if (!enough(content
->array
.level
,
1158 content
->array
.raid_disks
,
1159 content
->array
.layout
, clean
,
1161 fprintf(stderr
, Name
": Not enough devices to "
1162 "start the array while not clean "
1163 "- consider --force.\n");
1166 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1170 if (runstop
== -1) {
1171 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1172 mddev
, okcnt
, okcnt
==1?"":"s");
1173 if (okcnt
!= content
->array
.raid_disks
)
1174 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1175 fprintf(stderr
, ", but not started.\n");
1179 if (verbose
>= -1) {
1180 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1182 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1183 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1184 content
->array
.layout
, 1, avail
, okcnt
))
1185 fprintf(stderr
, " - not enough to start the array.\n");
1186 else if (!enough(content
->array
.level
,
1187 content
->array
.raid_disks
,
1188 content
->array
.layout
, clean
,
1190 fprintf(stderr
, " - not enough to start the "
1191 "array while not clean - consider "
1194 if (req_cnt
== content
->array
.raid_disks
)
1195 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1197 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1198 fprintf(stderr
, " (use --run to insist).\n");
1202 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1206 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1207 * been updated to point to the current locations of devices.
1208 * so we can just start the array
1211 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1212 devices
[chosen_drive
].i
.disk
.minor
);
1213 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1214 fprintf(stderr
, Name
": Cannot start array: %s\n",
1224 int assemble_container_content(struct supertype
*st
, int mdfd
,
1225 struct mdinfo
*content
, int runstop
,
1226 char *chosen_name
, int verbose
)
1228 struct mdinfo
*dev
, *sra
;
1229 int working
= 0, preexist
= 0;
1230 struct map_ent
*map
= NULL
;
1232 sysfs_init(content
, mdfd
, 0);
1234 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1235 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1236 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1243 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1244 if (sysfs_add_disk(content
, dev
, 1) == 0)
1246 else if (errno
== EEXIST
)
1250 return 1;/* Nothing new, don't try to start */
1253 map_update(&map
, fd2devnum(mdfd
),
1254 content
->text_version
,
1255 content
->uuid
, chosen_name
);
1258 (working
+ preexist
) >= content
->array
.working_disks
) {
1261 switch(content
->array
.level
) {
1263 case LEVEL_MULTIPATH
:
1265 err
= sysfs_set_str(content
, NULL
, "array_state",
1269 err
= sysfs_set_str(content
, NULL
, "array_state",
1271 /* start mdmon if needed. */
1273 if (!mdmon_running(st
->container_dev
))
1274 start_mdmon(st
->container_dev
);
1275 ping_monitor(devnum2devname(st
->container_dev
));
1280 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1283 fprintf(stderr
, Name
1284 ": array %s now has %d devices",
1285 chosen_name
, working
+ preexist
);
1287 fprintf(stderr
, Name
1288 ": Started %s with %d devices",
1289 chosen_name
, working
+ preexist
);
1291 fprintf(stderr
, " (%d new)", working
);
1292 fprintf(stderr
, "\n");
1295 wait_for(chosen_name
, mdfd
);
1298 /* FIXME should have an O_EXCL and wait for read-auto */
1301 fprintf(stderr
, Name
1302 ": %s assembled with %d devices but "
1304 chosen_name
, working
);