]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
5 * Copyright (C) 2006 Neil Brown <neilb@suse.de>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Email: <neilb@suse.de>
27 * QVB Post Office, NSW 1230
33 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
35 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
36 int number
, __u64 events
, int verbose
,
39 int Incremental(char *devname
, int verbose
, int runstop
,
40 struct supertype
*st
, char *homehost
, int autof
)
42 /* Add this device to an array, creating the array if necessary
43 * and starting the array if sensible or - if runstop>0 - if possible.
45 * This has several steps:
47 * 1/ Check if device is permitted by mdadm.conf, reject if not.
48 * 2/ Find metadata, reject if none appropriate (check
49 * version/name from args)
50 * 3/ Check if there is a match in mdadm.conf
51 * 3a/ if not, check for homehost match. If no match, assemble as
53 * 4/ Determine device number.
54 * - If in mdadm.conf with std name, use that
55 * - UUID in /var/run/mdadm.map use that
56 * - If name is suggestive, use that. unless in use with different uuid.
57 * - Choose a free, high number.
58 * - Use a partitioned device unless strong suggestion not to.
60 * Don't choose partitioned for containers.
61 * 5/ Find out if array already exists
63 * - choose a name, from mdadm.conf or 'name' field in array.
67 * - check one drive in array to make sure metadata is a reasonably
68 * close match. Reject if not (e.g. different type)
70 * 6/ Make sure /var/run/mdadm.map contains this array.
71 * 7/ Is there enough devices to possibly start the array?
72 * For a container, this means running Incremental_container.
73 * 7a/ if not, finish with success.
75 * - read all metadata and arrange devices like -A does
76 * - if number of OK devices match expected, or -R and there are enough,
77 * start the array (auto-readonly).
81 struct mddev_ident_s
*array_list
, *match
;
82 char chosen_name
[1024];
84 struct map_ent
*mp
, *map
= NULL
;
88 int trustworthy
= FOREIGN
;
90 mdu_array_info_t ainf
;
92 struct createinfo
*ci
= conf_get_create_info();
95 /* 1/ Check if device is permitted by mdadm.conf */
97 if (!conf_test_dev(devname
)) {
100 ": %s not permitted by mdadm.conf.\n",
105 /* 2/ Find metadata, reject if none appropriate (check
106 * version/name from args) */
108 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
111 fprintf(stderr
, Name
": cannot open %s: %s.\n",
112 devname
, strerror(errno
));
115 if (fstat(dfd
, &stb
) < 0) {
117 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
118 devname
, strerror(errno
));
122 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
124 fprintf(stderr
, Name
": %s is not a block device.\n",
130 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
133 ": no recognisable superblock on %s.\n",
138 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
140 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
147 memset(&info
, 0, sizeof(info
));
148 st
->ss
->getinfo_super(st
, &info
);
149 /* 3/ Check if there is a match in mdadm.conf */
151 array_list
= conf_get_ident(NULL
);
153 for (; array_list
; array_list
= array_list
->next
) {
154 if (array_list
->uuid_set
&&
155 same_uuid(array_list
->uuid
, info
.uuid
, st
->ss
->swapuuid
)
157 if (verbose
>= 2 && array_list
->devname
)
159 ": UUID differs from %s.\n",
160 array_list
->devname
);
163 if (array_list
->name
[0] &&
164 strcasecmp(array_list
->name
, info
.name
) != 0) {
165 if (verbose
>= 2 && array_list
->devname
)
167 ": Name differs from %s.\n",
168 array_list
->devname
);
171 if (array_list
->devices
&&
172 !match_oneof(array_list
->devices
, devname
)) {
173 if (verbose
>= 2 && array_list
->devname
)
175 ": Not a listed device for %s.\n",
176 array_list
->devname
);
179 if (array_list
->super_minor
!= UnSet
&&
180 array_list
->super_minor
!= info
.array
.md_minor
) {
181 if (verbose
>= 2 && array_list
->devname
)
183 ": Different super-minor to %s.\n",
184 array_list
->devname
);
187 if (!array_list
->uuid_set
&&
188 !array_list
->name
[0] &&
189 !array_list
->devices
&&
190 array_list
->super_minor
== UnSet
) {
191 if (verbose
>= 2 && array_list
->devname
)
193 ": %s doesn't have any identifying information.\n",
194 array_list
->devname
);
197 /* FIXME, should I check raid_disks and level too?? */
201 if (match
->devname
&& array_list
->devname
)
203 ": we match both %s and %s - cannot decide which to use.\n",
204 match
->devname
, array_list
->devname
);
207 ": multiple lines in mdadm.conf match\n");
214 if (match
&& match
->devname
215 && strcasecmp(match
->devname
, "<ignore>") == 0) {
217 fprintf(stderr
, Name
": array containing %s is explicitly"
218 " ignored by mdadm.conf\n",
223 if (!match
&& !conf_test_metadata(st
->ss
->name
)) {
226 ": %s has metadata type %s for which "
227 "auto-assembly is disabled\n",
228 devname
, st
->ss
->name
);
232 /* 3a/ if not, check for homehost match. If no match, continue
233 * but don't trust the 'name' in the array. Thus a 'random' minor
234 * number will be assigned, and the device name will be based
238 else if ((homehost
== NULL
||
239 st
->ss
->match_home(st
, homehost
) != 1) &&
240 st
->ss
->match_home(st
, "any") != 1)
241 trustworthy
= FOREIGN
;
245 /* There are three possible sources for 'autof': command line,
246 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
247 * ARRAY takes precedence, then command line, then
250 if (match
&& match
->autof
)
251 autof
= match
->autof
;
255 if (st
->ss
->container_content
&& st
->loaded_container
) {
256 /* This is a pre-built container array, so we do something
259 return Incremental_container(st
, devname
, verbose
, runstop
,
262 name_to_use
= strchr(info
.name
, ':');
266 name_to_use
= info
.name
;
268 if ((!name_to_use
|| name_to_use
[0] == 0) &&
269 info
.array
.level
== LEVEL_CONTAINER
&&
270 trustworthy
== LOCAL
) {
271 name_to_use
= info
.text_version
;
272 trustworthy
= METADATA
;
275 /* 4/ Check if array exists.
278 mp
= map_by_uuid(&map
, info
.uuid
);
280 mdfd
= open_mddev(mp
->path
, 0);
281 if (mdfd
< 0 && mddev_busy(mp
->devnum
)) {
282 /* maybe udev hasn't created it yet. */
284 sprintf(buf
, "%d:%d", dev2major(mp
->devnum
),
285 dev2minor(mp
->devnum
));
286 mdfd
= dev_open(buf
, O_RDWR
);
295 /* Couldn't find an existing array, maybe make a new one */
296 mdfd
= create_mddev(match
? match
->devname
: NULL
,
297 name_to_use
, autof
, trustworthy
, chosen_name
);
302 sysfs_init(&info
, mdfd
, 0);
304 if (set_array_info(mdfd
, st
, &info
) != 0) {
305 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
306 chosen_name
, strerror(errno
));
312 dinfo
.disk
.major
= major(stb
.st_rdev
);
313 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
314 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
315 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
316 devname
, chosen_name
, strerror(errno
));
317 ioctl(mdfd
, STOP_ARRAY
, 0);
321 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), GET_DEVS
);
322 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
323 /* It really should be 'none' - must be old buggy
324 * kernel, and mdadm -I may not be able to complete.
327 ioctl(mdfd
, STOP_ARRAY
, NULL
);
329 ": You have an old buggy kernel which cannot support\n"
330 " --incremental reliably. Aborting.\n");
335 info
.array
.working_disks
= 1;
337 /* 6/ Make sure /var/run/mdadm.map contains this array. */
338 map_update(&map
, fd2devnum(mdfd
),
340 info
.uuid
, chosen_name
);
343 /* - check one drive in array to make sure metadata is a reasonably */
344 /* close match. Reject if not (e.g. different type) */
345 /* - add the device */
350 struct supertype
*st2
;
351 struct mdinfo info2
, *d
;
353 strcpy(chosen_name
, mp
->path
);
355 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), (GET_DEVS
| GET_STATE
));
358 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
359 sra
->devs
->disk
.minor
);
360 dfd2
= dev_open(dn
, O_RDONLY
);
362 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
363 st
->ss
->compare_super(st
, st2
) != 0) {
365 ": metadata mismatch between %s and "
367 devname
, chosen_name
);
373 memset(&info2
, 0, sizeof(info2
));
374 st2
->ss
->getinfo_super(st2
, &info2
);
375 st2
->ss
->free_super(st2
);
376 if (info
.array
.level
!= info2
.array
.level
||
377 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
378 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
380 ": unexpected difference between %s and %s.\n",
381 chosen_name
, devname
);
386 info2
.disk
.major
= major(stb
.st_rdev
);
387 info2
.disk
.minor
= minor(stb
.st_rdev
);
388 /* add disk needs to know about containers */
389 if (st
->ss
->external
)
390 sra
->array
.level
= LEVEL_CONTAINER
;
391 err
= add_disk(mdfd
, st2
, sra
, &info2
);
392 if (err
< 0 && errno
== EBUSY
) {
393 /* could be another device present with the same
394 * disk.number. Find and reject any such
396 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
397 info
.events
, verbose
, chosen_name
);
398 err
= add_disk(mdfd
, st2
, sra
, &info2
);
401 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
402 devname
, chosen_name
, strerror(errno
));
406 info
.array
.working_disks
= 0;
407 for (d
= sra
->devs
; d
; d
=d
->next
)
408 info
.array
.working_disks
++;
412 /* 7/ Is there enough devices to possibly start the array? */
413 /* 7a/ if not, finish with success. */
414 if (info
.array
.level
== LEVEL_CONTAINER
) {
415 /* Try to assemble within the container */
417 sysfs_uevent(&info
, "change");
420 ": container %s now has %d devices\n",
421 chosen_name
, info
.array
.working_disks
);
422 wait_for(chosen_name
, mdfd
);
425 return 0; /* don't try to assemble */
426 rv
= Incremental(chosen_name
, verbose
, runstop
,
427 NULL
, homehost
, autof
);
429 /* Don't fail the whole -I if a subarray didn't
430 * have enough devices to start yet
436 active_disks
= count_active(st
, mdfd
, &avail
, &info
);
437 if (enough(info
.array
.level
, info
.array
.raid_disks
,
438 info
.array
.layout
, info
.array
.state
& 1,
439 avail
, active_disks
) == 0 ||
440 (runstop
< 0 && active_disks
< info
.array
.raid_disks
)) {
444 ": %s attached to %s, not enough to start (%d).\n",
445 devname
, chosen_name
, active_disks
);
453 /* - if number of OK devices match expected, or -R and there */
455 /* + add any bitmap file */
456 /* + start the array (auto-readonly). */
458 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
461 ": %s attached to %s which is already active.\n",
462 devname
, chosen_name
);
469 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
471 /* Let's try to start it */
472 if (match
&& match
->bitmap_file
) {
473 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
476 ": Could not open bitmap file %s.\n",
481 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
484 ": Failed to set bitmapfile for %s.\n",
491 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), 0);
492 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
493 && trustworthy
!= FOREIGN
)
494 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
496 rv
= sysfs_set_str(sra
, NULL
,
497 "array_state", "read-auto");
501 ": %s attached to %s, which has been started.\n",
502 devname
, chosen_name
);
504 wait_for(chosen_name
, mdfd
);
507 ": %s attached to %s, but failed to start: %s.\n",
508 devname
, chosen_name
, strerror(errno
));
514 ": %s attached to %s, not enough to start safely.\n",
515 devname
, chosen_name
);
522 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
523 int number
, __u64 events
, int verbose
,
526 /* Find a device attached to this array with a disk.number of number
527 * and events less than the passed events, and remove the device.
532 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
533 return; /* not safe to remove from active arrays
534 * without thinking more */
536 for (d
= sra
->devs
; d
; d
= d
->next
) {
540 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
541 dfd
= dev_open(dn
, O_RDONLY
);
544 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
548 st
->ss
->getinfo_super(st
, &info
);
549 st
->ss
->free_super(st
);
552 if (info
.disk
.number
!= number
||
553 info
.events
>= events
)
556 if (d
->disk
.raid_disk
> -1)
557 sysfs_set_str(sra
, d
, "slot", "none");
558 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
561 ": removing old device %s from %s\n",
562 d
->sys_name
+4, array_name
);
566 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
567 struct mdinfo
*bestinfo
)
569 /* count how many devices in sra think they are active */
571 int cnt
= 0, cnt1
= 0;
572 __u64 max_events
= 0;
573 struct mdinfo
*sra
= sysfs_read(mdfd
, -1, GET_DEVS
| GET_STATE
);
576 for (d
= sra
->devs
; d
; d
= d
->next
) {
582 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
583 dfd
= dev_open(dn
, O_RDONLY
);
586 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
590 st
->ss
->getinfo_super(st
, &info
);
592 avail
= malloc(info
.array
.raid_disks
);
594 fprintf(stderr
, Name
": out of memory.\n");
597 memset(avail
, 0, info
.array
.raid_disks
);
601 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
605 max_events
= info
.events
;
606 avail
[info
.disk
.raid_disk
] = 2;
607 st
->ss
->getinfo_super(st
, bestinfo
);
608 } else if (info
.events
== max_events
) {
610 avail
[info
.disk
.raid_disk
] = 2;
611 } else if (info
.events
== max_events
-1) {
613 avail
[info
.disk
.raid_disk
] = 1;
614 } else if (info
.events
< max_events
- 1)
616 else if (info
.events
== max_events
+1) {
620 max_events
= info
.events
;
621 for (i
=0; i
<info
.array
.raid_disks
; i
++)
624 avail
[info
.disk
.raid_disk
] = 2;
625 st
->ss
->getinfo_super(st
, bestinfo
);
626 } else { /* info.events much bigger */
628 memset(avail
, 0, info
.disk
.raid_disk
);
629 max_events
= info
.events
;
630 st
->ss
->getinfo_super(st
, bestinfo
);
633 st
->ss
->free_super(st
);
638 int IncrementalScan(int verbose
)
640 /* look at every device listed in the 'map' file.
641 * If one is found that is not running then:
642 * look in mdadm.conf for bitmap file.
643 * if one exists, but array has none, add it.
644 * try to start array in auto-readonly mode
646 struct map_ent
*mapl
= NULL
;
648 mddev_ident_t devs
, mddev
;
652 devs
= conf_get_ident(NULL
);
654 for (me
= mapl
; me
; me
= me
->next
) {
655 mdu_array_info_t array
;
656 mdu_bitmap_file_t bmf
;
658 int mdfd
= open_mddev(me
->path
, 0);
662 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
667 /* Ok, we can try this one. Maybe it needs a bitmap */
668 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
670 && strcmp(mddev
->devname
, me
->path
) == 0)
672 if (mddev
&& mddev
->bitmap_file
) {
674 * Note: early kernels will wrongly fail this, so it
678 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
679 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
681 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
689 ": Added bitmap %s to %s\n",
690 mddev
->bitmap_file
, me
->path
);
691 else if (errno
!= EEXIST
)
693 ": Failed to add bitmap to %s: %s\n",
694 me
->path
, strerror(errno
));
697 sra
= sysfs_read(mdfd
, 0, 0);
699 if (sysfs_set_str(sra
, NULL
,
700 "array_state", "read-auto") == 0) {
703 ": started array %s\n",
707 ": failed to start array %s: %s\n",
708 me
->path
, strerror(errno
));
716 static char *container2devname(char *devname
)
720 if (devname
[0] == '/') {
721 int fd
= open(devname
, O_RDONLY
);
723 mdname
= devnum2devname(fd2devnum(fd
));
728 struct map_ent
*mp
, *map
= NULL
;
730 if (!parse_uuid(devname
, uuid
))
732 mp
= map_by_uuid(&map
, uuid
);
734 mdname
= devnum2devname(mp
->devnum
);
741 int Incremental_container(struct supertype
*st
, char *devname
, int verbose
,
742 int runstop
, int autof
, int trustworthy
)
744 /* Collect the contents of this container and for each
745 * array, choose a device name and assemble the array.
748 struct mdinfo
*list
= st
->ss
->container_content(st
);
750 struct map_ent
*map
= NULL
;
754 for (ra
= list
; ra
; ra
= ra
->next
) {
756 char chosen_name
[1024];
758 struct mddev_ident_s
*match
= NULL
;
760 mp
= map_by_uuid(&map
, ra
->uuid
);
763 mdfd
= open_dev(mp
->devnum
);
764 strcpy(chosen_name
, mp
->path
);
767 /* Check in mdadm.conf for container == devname and
768 * member == ra->text_version after second slash.
770 char *sub
= strchr(ra
->text_version
+1, '/');
771 struct mddev_ident_s
*array_list
;
774 array_list
= conf_get_ident(NULL
);
777 for(; array_list
; array_list
= array_list
->next
) {
779 if (array_list
->member
== NULL
||
780 array_list
->container
== NULL
)
782 if (strcmp(array_list
->member
, sub
) != 0)
784 if (array_list
->uuid_set
&&
785 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
787 dn
= container2devname(array_list
->container
);
790 if (strncmp(dn
, ra
->text_version
+1,
792 ra
->text_version
[strlen(dn
)+1] != '/') {
797 /* we have a match */
800 fprintf(stderr
, Name
": match found for member %s\n",
805 if (match
&& match
->devname
&&
806 strcasecmp(match
->devname
, "<ignore>") == 0) {
808 fprintf(stderr
, Name
": array %s/%s is "
809 "explicitly ignored by mdadm.conf\n",
810 match
->container
, match
->member
);
814 mdfd
= create_mddev(match
? match
->devname
: NULL
,
822 fprintf(stderr
, Name
": failed to open %s: %s.\n",
823 chosen_name
, strerror(errno
));
827 assemble_container_content(st
, mdfd
, ra
, runstop
,
828 chosen_name
, verbose
);