]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
46badc9d0d7d094d242af3f0a1430dcbd7992939
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
5 * Copyright (C) 2006-2009 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 require_homehost
,
43 /* Add this device to an array, creating the array if necessary
44 * and starting the array if sensible or - if runstop>0 - if possible.
46 * This has several steps:
48 * 1/ Check if device is permitted by mdadm.conf, reject if not.
49 * 2/ Find metadata, reject if none appropriate (check
50 * version/name from args)
51 * 3/ Check if there is a match in mdadm.conf
52 * 3a/ if not, check for homehost match. If no match, assemble as
54 * 4/ Determine device number.
55 * - If in mdadm.conf with std name, use that
56 * - UUID in /var/run/mdadm.map use that
57 * - If name is suggestive, use that. unless in use with different uuid.
58 * - Choose a free, high number.
59 * - Use a partitioned device unless strong suggestion not to.
61 * Don't choose partitioned for containers.
62 * 5/ Find out if array already exists
64 * - choose a name, from mdadm.conf or 'name' field in array.
68 * - check one drive in array to make sure metadata is a reasonably
69 * close match. Reject if not (e.g. different type)
71 * 6/ Make sure /var/run/mdadm.map contains this array.
72 * 7/ Is there enough devices to possibly start the array?
73 * For a container, this means running Incremental_container.
74 * 7a/ if not, finish with success.
76 * - read all metadata and arrange devices like -A does
77 * - if number of OK devices match expected, or -R and there are enough,
78 * start the array (auto-readonly).
82 struct mddev_ident_s
*array_list
, *match
;
83 char chosen_name
[1024];
85 struct map_ent
*mp
, *map
= NULL
;
89 int trustworthy
= FOREIGN
;
91 mdu_array_info_t ainf
;
93 struct createinfo
*ci
= conf_get_create_info();
96 /* 1/ Check if device is permitted by mdadm.conf */
98 if (!conf_test_dev(devname
)) {
101 ": %s not permitted by mdadm.conf.\n",
106 /* 2/ Find metadata, reject if none appropriate (check
107 * version/name from args) */
109 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
112 fprintf(stderr
, Name
": cannot open %s: %s.\n",
113 devname
, strerror(errno
));
116 if (fstat(dfd
, &stb
) < 0) {
118 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
119 devname
, strerror(errno
));
123 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
125 fprintf(stderr
, Name
": %s is not a block device.\n",
131 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
134 ": no recognisable superblock on %s.\n",
139 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
141 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
148 memset(&info
, 0, sizeof(info
));
149 st
->ss
->getinfo_super(st
, &info
);
150 /* 3/ Check if there is a match in mdadm.conf */
152 array_list
= conf_get_ident(NULL
);
154 for (; array_list
; array_list
= array_list
->next
) {
155 if (array_list
->uuid_set
&&
156 same_uuid(array_list
->uuid
, info
.uuid
, st
->ss
->swapuuid
)
158 if (verbose
>= 2 && array_list
->devname
)
160 ": UUID differs from %s.\n",
161 array_list
->devname
);
164 if (array_list
->name
[0] &&
165 strcasecmp(array_list
->name
, info
.name
) != 0) {
166 if (verbose
>= 2 && array_list
->devname
)
168 ": Name differs from %s.\n",
169 array_list
->devname
);
172 if (array_list
->devices
&&
173 !match_oneof(array_list
->devices
, devname
)) {
174 if (verbose
>= 2 && array_list
->devname
)
176 ": Not a listed device for %s.\n",
177 array_list
->devname
);
180 if (array_list
->super_minor
!= UnSet
&&
181 array_list
->super_minor
!= info
.array
.md_minor
) {
182 if (verbose
>= 2 && array_list
->devname
)
184 ": Different super-minor to %s.\n",
185 array_list
->devname
);
188 if (!array_list
->uuid_set
&&
189 !array_list
->name
[0] &&
190 !array_list
->devices
&&
191 array_list
->super_minor
== UnSet
) {
192 if (verbose
>= 2 && array_list
->devname
)
194 ": %s doesn't have any identifying information.\n",
195 array_list
->devname
);
198 /* FIXME, should I check raid_disks and level too?? */
202 if (match
->devname
&& array_list
->devname
)
204 ": we match both %s and %s - cannot decide which to use.\n",
205 match
->devname
, array_list
->devname
);
208 ": multiple lines in mdadm.conf match\n");
215 if (match
&& match
->devname
216 && strcasecmp(match
->devname
, "<ignore>") == 0) {
218 fprintf(stderr
, Name
": array containing %s is explicitly"
219 " ignored by mdadm.conf\n",
224 /* 3a/ if not, check for homehost match. If no match, continue
225 * but don't trust the 'name' in the array. Thus a 'random' minor
226 * number will be assigned, and the device name will be based
230 else if (st
->ss
->match_home(st
, homehost
) == 1)
232 else if (st
->ss
->match_home(st
, "any") == 1)
233 trustworthy
= LOCAL_ANY
;
235 trustworthy
= FOREIGN
;
238 if (!match
&& !conf_test_metadata(st
->ss
->name
,
239 (trustworthy
== LOCAL
))) {
242 ": %s has metadata type %s for which "
243 "auto-assembly is disabled\n",
244 devname
, st
->ss
->name
);
247 if (trustworthy
== LOCAL_ANY
)
250 /* There are three possible sources for 'autof': command line,
251 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
252 * ARRAY takes precedence, then command line, then
255 if (match
&& match
->autof
)
256 autof
= match
->autof
;
260 if (st
->ss
->container_content
&& st
->loaded_container
) {
261 if ((runstop
> 0 && info
.container_enough
>= 0) ||
262 info
.container_enough
> 0)
266 fprintf(stderr
, Name
": not enough devices to start the container\n");
270 /* This is a pre-built container array, so we do something
273 return Incremental_container(st
, devname
, verbose
, runstop
,
277 name_to_use
= info
.name
;
278 if (name_to_use
[0] == 0 &&
279 info
.array
.level
== LEVEL_CONTAINER
&&
280 trustworthy
== LOCAL
) {
281 name_to_use
= info
.text_version
;
282 trustworthy
= METADATA
;
284 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
285 ! require_homehost
&&
286 conf_name_is_free(name_to_use
))
289 /* strip "hostname:" prefix from name if we have decided
290 * to treat it as LOCAL
292 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
293 name_to_use
= strchr(name_to_use
, ':')+1;
295 /* 4/ Check if array exists.
298 fprintf(stderr
, Name
": failed to get exclusive lock on "
300 mp
= map_by_uuid(&map
, info
.uuid
);
302 mdfd
= open_dev(mp
->devnum
);
310 /* Couldn't find an existing array, maybe make a new one */
311 mdfd
= create_mddev(match
? match
->devname
: NULL
,
312 name_to_use
, autof
, trustworthy
, chosen_name
);
317 sysfs_init(&info
, mdfd
, 0);
319 if (set_array_info(mdfd
, st
, &info
) != 0) {
320 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
321 chosen_name
, strerror(errno
));
327 dinfo
.disk
.major
= major(stb
.st_rdev
);
328 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
329 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
330 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
331 devname
, chosen_name
, strerror(errno
));
332 ioctl(mdfd
, STOP_ARRAY
, 0);
336 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), GET_DEVS
);
337 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
338 /* It really should be 'none' - must be old buggy
339 * kernel, and mdadm -I may not be able to complete.
342 ioctl(mdfd
, STOP_ARRAY
, NULL
);
344 ": You have an old buggy kernel which cannot support\n"
345 " --incremental reliably. Aborting.\n");
350 info
.array
.working_disks
= 1;
352 /* 6/ Make sure /var/run/mdadm.map contains this array. */
353 map_update(&map
, fd2devnum(mdfd
),
355 info
.uuid
, chosen_name
);
358 /* - check one drive in array to make sure metadata is a reasonably */
359 /* close match. Reject if not (e.g. different type) */
360 /* - add the device */
365 struct supertype
*st2
;
366 struct mdinfo info2
, *d
;
369 strcpy(chosen_name
, mp
->path
);
371 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
373 /* It is generally not OK to add non-spare drives to a
374 * running array as they are probably missing because
375 * they failed. However if runstop is 1, then the
376 * array was possibly started early and our best be is
377 * to add this anyway. It would probably be good to
378 * allow explicit policy statement about this..
379 * This doesn't apply to containers as the 'non-spare'
380 * flag has a different meaning. The test has to happen
381 * at the device level there
383 if (!st
->ss
->external
384 && (info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
386 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
388 ": not adding %s to active array (without --run) %s\n",
389 devname
, chosen_name
);
394 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), (GET_DEVS
| GET_STATE
));
399 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
400 sra
->devs
->disk
.minor
);
401 dfd2
= dev_open(dn
, O_RDONLY
);
403 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
404 st
->ss
->compare_super(st
, st2
) != 0) {
406 ": metadata mismatch between %s and "
408 devname
, chosen_name
);
414 memset(&info2
, 0, sizeof(info2
));
415 st2
->ss
->getinfo_super(st2
, &info2
);
416 st2
->ss
->free_super(st2
);
417 if (info
.array
.level
!= info2
.array
.level
||
418 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
419 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
421 ": unexpected difference between %s and %s.\n",
422 chosen_name
, devname
);
427 info2
.disk
.major
= major(stb
.st_rdev
);
428 info2
.disk
.minor
= minor(stb
.st_rdev
);
429 /* add disk needs to know about containers */
430 if (st
->ss
->external
)
431 sra
->array
.level
= LEVEL_CONTAINER
;
432 err
= add_disk(mdfd
, st
, sra
, &info2
);
433 if (err
< 0 && errno
== EBUSY
) {
434 /* could be another device present with the same
435 * disk.number. Find and reject any such
437 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
438 info
.events
, verbose
, chosen_name
);
439 err
= add_disk(mdfd
, st
, sra
, &info2
);
442 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
443 devname
, chosen_name
, strerror(errno
));
447 info
.array
.working_disks
= 0;
448 for (d
= sra
->devs
; d
; d
=d
->next
)
449 info
.array
.working_disks
++;
453 /* 7/ Is there enough devices to possibly start the array? */
454 /* 7a/ if not, finish with success. */
455 if (info
.array
.level
== LEVEL_CONTAINER
) {
456 /* Try to assemble within the container */
458 sysfs_uevent(&info
, "change");
461 ": container %s now has %d devices\n",
462 chosen_name
, info
.array
.working_disks
);
463 wait_for(chosen_name
, mdfd
);
465 rv
= Incremental(chosen_name
, verbose
, runstop
,
466 NULL
, homehost
, require_homehost
, autof
);
468 /* Don't fail the whole -I if a subarray didn't
469 * have enough devices to start yet
475 active_disks
= count_active(st
, mdfd
, &avail
, &info
);
476 if (enough(info
.array
.level
, info
.array
.raid_disks
,
477 info
.array
.layout
, info
.array
.state
& 1,
478 avail
, active_disks
) == 0) {
482 ": %s attached to %s, not enough to start (%d).\n",
483 devname
, chosen_name
, active_disks
);
491 /* - if number of OK devices match expected, or -R and there */
493 /* + add any bitmap file */
494 /* + start the array (auto-readonly). */
496 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
499 ": %s attached to %s which is already active.\n",
500 devname
, chosen_name
);
507 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
509 /* Let's try to start it */
510 if (match
&& match
->bitmap_file
) {
511 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
514 ": Could not open bitmap file %s.\n",
519 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
522 ": Failed to set bitmapfile for %s.\n",
529 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), 0);
530 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
531 && trustworthy
!= FOREIGN
)
532 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
534 rv
= sysfs_set_str(sra
, NULL
,
535 "array_state", "read-auto");
539 ": %s attached to %s, which has been started.\n",
540 devname
, chosen_name
);
542 wait_for(chosen_name
, mdfd
);
545 ": %s attached to %s, but failed to start: %s.\n",
546 devname
, chosen_name
, strerror(errno
));
552 ": %s attached to %s, not enough to start safely.\n",
553 devname
, chosen_name
);
560 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
561 int number
, __u64 events
, int verbose
,
564 /* Find a device attached to this array with a disk.number of number
565 * and events less than the passed events, and remove the device.
570 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
571 return; /* not safe to remove from active arrays
572 * without thinking more */
574 for (d
= sra
->devs
; d
; d
= d
->next
) {
578 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
579 dfd
= dev_open(dn
, O_RDONLY
);
582 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
586 st
->ss
->getinfo_super(st
, &info
);
587 st
->ss
->free_super(st
);
590 if (info
.disk
.number
!= number
||
591 info
.events
>= events
)
594 if (d
->disk
.raid_disk
> -1)
595 sysfs_set_str(sra
, d
, "slot", "none");
596 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
599 ": removing old device %s from %s\n",
600 d
->sys_name
+4, array_name
);
604 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
605 struct mdinfo
*bestinfo
)
607 /* count how many devices in sra think they are active */
609 int cnt
= 0, cnt1
= 0;
610 __u64 max_events
= 0;
611 struct mdinfo
*sra
= sysfs_read(mdfd
, -1, GET_DEVS
| GET_STATE
);
617 for (d
= sra
->devs
; d
; d
= d
->next
) {
623 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
624 dfd
= dev_open(dn
, O_RDONLY
);
627 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
631 st
->ss
->getinfo_super(st
, &info
);
633 avail
= malloc(info
.array
.raid_disks
);
635 fprintf(stderr
, Name
": out of memory.\n");
638 memset(avail
, 0, info
.array
.raid_disks
);
642 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
646 max_events
= info
.events
;
647 avail
[info
.disk
.raid_disk
] = 2;
648 st
->ss
->getinfo_super(st
, bestinfo
);
649 } else if (info
.events
== max_events
) {
651 avail
[info
.disk
.raid_disk
] = 2;
652 } else if (info
.events
== max_events
-1) {
654 avail
[info
.disk
.raid_disk
] = 1;
655 } else if (info
.events
< max_events
- 1)
657 else if (info
.events
== max_events
+1) {
661 max_events
= info
.events
;
662 for (i
=0; i
<info
.array
.raid_disks
; i
++)
665 avail
[info
.disk
.raid_disk
] = 2;
666 st
->ss
->getinfo_super(st
, bestinfo
);
667 } else { /* info.events much bigger */
669 memset(avail
, 0, info
.disk
.raid_disk
);
670 max_events
= info
.events
;
671 st
->ss
->getinfo_super(st
, bestinfo
);
674 st
->ss
->free_super(st
);
679 int IncrementalScan(int verbose
)
681 /* look at every device listed in the 'map' file.
682 * If one is found that is not running then:
683 * look in mdadm.conf for bitmap file.
684 * if one exists, but array has none, add it.
685 * try to start array in auto-readonly mode
687 struct map_ent
*mapl
= NULL
;
689 mddev_ident_t devs
, mddev
;
693 devs
= conf_get_ident(NULL
);
695 for (me
= mapl
; me
; me
= me
->next
) {
696 mdu_array_info_t array
;
697 mdu_bitmap_file_t bmf
;
699 int mdfd
= open_dev(me
->devnum
);
703 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
708 /* Ok, we can try this one. Maybe it needs a bitmap */
709 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
710 if (mddev
->devname
&& me
->path
711 && devname_matches(mddev
->devname
, me
->path
))
713 if (mddev
&& mddev
->bitmap_file
) {
715 * Note: early kernels will wrongly fail this, so it
719 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
720 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
722 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
730 ": Added bitmap %s to %s\n",
731 mddev
->bitmap_file
, me
->path
);
732 else if (errno
!= EEXIST
)
734 ": Failed to add bitmap to %s: %s\n",
735 me
->path
, strerror(errno
));
738 sra
= sysfs_read(mdfd
, 0, 0);
740 if (sysfs_set_str(sra
, NULL
,
741 "array_state", "read-auto") == 0) {
744 ": started array %s\n",
745 me
->path
?: devnum2devname(me
->devnum
));
748 ": failed to start array %s: %s\n",
749 me
->path
?: devnum2devname(me
->devnum
),
758 static char *container2devname(char *devname
)
762 if (devname
[0] == '/') {
763 int fd
= open(devname
, O_RDONLY
);
765 mdname
= devnum2devname(fd2devnum(fd
));
770 struct map_ent
*mp
, *map
= NULL
;
772 if (!parse_uuid(devname
, uuid
))
774 mp
= map_by_uuid(&map
, uuid
);
776 mdname
= devnum2devname(mp
->devnum
);
783 int Incremental_container(struct supertype
*st
, char *devname
, int verbose
,
784 int runstop
, int autof
, int trustworthy
)
786 /* Collect the contents of this container and for each
787 * array, choose a device name and assemble the array.
790 struct mdinfo
*list
= st
->ss
->container_content(st
);
792 struct map_ent
*map
= NULL
;
795 fprintf(stderr
, Name
": failed to get exclusive lock on "
798 for (ra
= list
; ra
; ra
= ra
->next
) {
800 char chosen_name
[1024];
802 struct mddev_ident_s
*match
= NULL
;
804 mp
= map_by_uuid(&map
, ra
->uuid
);
807 mdfd
= open_dev(mp
->devnum
);
809 strcpy(chosen_name
, mp
->path
);
811 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
814 /* Check in mdadm.conf for container == devname and
815 * member == ra->text_version after second slash.
817 char *sub
= strchr(ra
->text_version
+1, '/');
818 struct mddev_ident_s
*array_list
;
821 array_list
= conf_get_ident(NULL
);
824 for(; array_list
; array_list
= array_list
->next
) {
826 if (array_list
->member
== NULL
||
827 array_list
->container
== NULL
)
829 if (strcmp(array_list
->member
, sub
) != 0)
831 if (array_list
->uuid_set
&&
832 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
834 dn
= container2devname(array_list
->container
);
837 if (strncmp(dn
, ra
->text_version
+1,
839 ra
->text_version
[strlen(dn
)+1] != '/') {
844 /* we have a match */
847 fprintf(stderr
, Name
": match found for member %s\n",
852 if (match
&& match
->devname
&&
853 strcasecmp(match
->devname
, "<ignore>") == 0) {
855 fprintf(stderr
, Name
": array %s/%s is "
856 "explicitly ignored by mdadm.conf\n",
857 match
->container
, match
->member
);
863 mdfd
= create_mddev(match
? match
->devname
: NULL
,
871 fprintf(stderr
, Name
": failed to open %s: %s.\n",
872 chosen_name
, strerror(errno
));
876 assemble_container_content(st
, mdfd
, ra
, runstop
,
877 chosen_name
, verbose
);
884 * IncrementalRemove - Attempt to see if the passed in device belongs to any
885 * raid arrays, and if so first fail (if needed) and then remove the device.
887 * @devname - The device we want to remove
889 * Note: the device name must be a kernel name like "sda", so
890 * that we can find it in /proc/mdstat
892 int IncrementalRemove(char *devname
, int verbose
)
896 struct mdstat_ent
*ent
;
897 struct mddev_dev_s devlist
;
899 if (strchr(devname
, '/')) {
900 fprintf(stderr
, Name
": incremental removal requires a "
901 "kernel device name, not a file: %s\n", devname
);
904 ent
= mdstat_by_component(devname
);
906 fprintf(stderr
, Name
": %s does not appear to be a component "
907 "of any array\n", devname
);
910 mdfd
= open_dev(ent
->devnum
);
912 fprintf(stderr
, Name
": Cannot open array %s!!\n", ent
->dev
);
915 memset(&devlist
, 0, sizeof(devlist
));
916 devlist
.devname
= devname
;
917 devlist
.disposition
= 'f';
918 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);
919 devlist
.disposition
= 'r';
920 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);