]>
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) 2006-2009 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
35 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
37 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
38 int number
, __u64 events
, int verbose
,
40 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
41 struct supertype
*st
, int verbose
);
43 static int Incremental_container(struct supertype
*st
, char *devname
,
45 int verbose
, int runstop
, int autof
);
47 static struct mddev_ident_s
*search_mdstat(struct supertype
*st
,
50 int verbose
, int *rvp
);
52 int Incremental(char *devname
, int verbose
, int runstop
,
53 struct supertype
*st
, char *homehost
, int require_homehost
,
56 /* Add this device to an array, creating the array if necessary
57 * and starting the array if sensible or - if runstop>0 - if possible.
59 * This has several steps:
61 * 1/ Check if device is permitted by mdadm.conf, reject if not.
62 * 2/ Find metadata, reject if none appropriate (check
63 * version/name from args)
64 * 3/ Check if there is a match in mdadm.conf
65 * 3a/ if not, check for homehost match. If no match, assemble as
67 * 4/ Determine device number.
68 * - If in mdadm.conf with std name, use that
69 * - UUID in /var/run/mdadm.map use that
70 * - If name is suggestive, use that. unless in use with different uuid.
71 * - Choose a free, high number.
72 * - Use a partitioned device unless strong suggestion not to.
74 * Don't choose partitioned for containers.
75 * 5/ Find out if array already exists
77 * - choose a name, from mdadm.conf or 'name' field in array.
81 * - check one drive in array to make sure metadata is a reasonably
82 * close match. Reject if not (e.g. different type)
84 * 6/ Make sure /var/run/mdadm.map contains this array.
85 * 7/ Is there enough devices to possibly start the array?
86 * For a container, this means running Incremental_container.
87 * 7a/ if not, finish with success.
89 * - read all metadata and arrange devices like -A does
90 * - if number of OK devices match expected, or -R and there are enough,
91 * start the array (auto-readonly).
94 struct mdinfo info
, dinfo
;
95 struct mddev_ident_s
*match
;
96 char chosen_name
[1024];
98 struct map_ent
*mp
, *map
= NULL
;
99 int dfd
= -1, mdfd
= -1;
104 mdu_array_info_t ainf
;
105 struct dev_policy
*policy
= NULL
;
106 unsigned long long size
;
108 struct createinfo
*ci
= conf_get_create_info();
110 if (stat(devname
, &stb
) < 0) {
112 fprintf(stderr
, Name
": stat failed for %s: %s.\n",
113 devname
, strerror(errno
));
116 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
118 fprintf(stderr
, Name
": %s is not a block device.\n",
122 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
125 fprintf(stderr
, Name
": cannot open %s: %s.\n",
126 devname
, strerror(errno
));
129 /* If the device is a container, we do something very different */
130 if (get_dev_size(dfd
, devname
, &size
) == 0)
134 st
= super_by_fd(dfd
, NULL
);
136 rv
= st
->ss
->load_container(st
, dfd
, NULL
);
139 if (!rv
&& st
->ss
->container_content
)
140 return Incremental_container(st
, devname
, homehost
,
141 verbose
, runstop
, autof
);
143 fprintf(stderr
, Name
": %s is not part of an md array.\n",
148 /* 1/ Check if device is permitted by mdadm.conf */
150 if (!conf_test_dev(devname
)) {
153 ": %s not permitted by mdadm.conf.\n",
158 /* 2/ Find metadata, reject if none appropriate (check
159 * version/name from args) */
161 if (fstat(dfd
, &stb
) < 0) {
163 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
164 devname
, strerror(errno
));
167 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
169 fprintf(stderr
, Name
": %s is not a block device.\n",
174 dinfo
.disk
.major
= major(stb
.st_rdev
);
175 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
177 policy
= disk_policy(&dinfo
);
179 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
182 ": no recognisable superblock on %s.\n",
184 rv
= try_spare(devname
, &dfd
, policy
, st
, verbose
);
187 if (st
->ss
->compare_super
== NULL
||
188 st
->ss
->load_super(st
, dfd
, NULL
)) {
190 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
192 rv
= try_spare(devname
, &dfd
, policy
, st
, verbose
);
196 close (dfd
); dfd
= -1;
198 memset(&info
, 0, sizeof(info
));
199 st
->ss
->getinfo_super(st
, &info
, NULL
);
201 /* 3/ Check if there is a match in mdadm.conf */
202 match
= search_mdstat(st
, &info
, devname
, verbose
, &rv
);
203 if (!match
&& rv
== 2)
206 if (match
&& match
->devname
207 && strcasecmp(match
->devname
, "<ignore>") == 0) {
209 fprintf(stderr
, Name
": array containing %s is explicitly"
210 " ignored by mdadm.conf\n",
215 /* 3a/ if not, check for homehost match. If no match, continue
216 * but don't trust the 'name' in the array. Thus a 'random' minor
217 * number will be assigned, and the device name will be based
221 else if (st
->ss
->match_home(st
, homehost
) == 1)
223 else if (st
->ss
->match_home(st
, "any") == 1)
224 trustworthy
= LOCAL_ANY
;
226 trustworthy
= FOREIGN
;
229 if (!match
&& !conf_test_metadata(st
->ss
->name
, policy
,
230 (trustworthy
== LOCAL
))) {
233 ": %s has metadata type %s for which "
234 "auto-assembly is disabled\n",
235 devname
, st
->ss
->name
);
238 if (trustworthy
== LOCAL_ANY
)
241 /* There are three possible sources for 'autof': command line,
242 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
243 * ARRAY takes precedence, then command line, then
246 if (match
&& match
->autof
)
247 autof
= match
->autof
;
251 name_to_use
= info
.name
;
252 if (name_to_use
[0] == 0 &&
253 info
.array
.level
== LEVEL_CONTAINER
&&
254 trustworthy
== LOCAL
) {
255 name_to_use
= info
.text_version
;
256 trustworthy
= METADATA
;
258 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
259 ! require_homehost
&&
260 conf_name_is_free(name_to_use
))
263 /* strip "hostname:" prefix from name if we have decided
264 * to treat it as LOCAL
266 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
267 name_to_use
= strchr(name_to_use
, ':')+1;
269 /* 4/ Check if array exists.
272 fprintf(stderr
, Name
": failed to get exclusive lock on "
274 mp
= map_by_uuid(&map
, info
.uuid
);
276 mdfd
= open_dev(mp
->devnum
);
283 /* Couldn't find an existing array, maybe make a new one */
284 mdfd
= create_mddev(match
? match
->devname
: NULL
,
285 name_to_use
, autof
, trustworthy
, chosen_name
);
290 sysfs_init(&info
, mdfd
, 0);
292 if (set_array_info(mdfd
, st
, &info
) != 0) {
293 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
294 chosen_name
, strerror(errno
));
300 dinfo
.disk
.major
= major(stb
.st_rdev
);
301 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
302 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
303 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
304 devname
, chosen_name
, strerror(errno
));
305 ioctl(mdfd
, STOP_ARRAY
, 0);
309 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), GET_DEVS
);
310 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
311 /* It really should be 'none' - must be old buggy
312 * kernel, and mdadm -I may not be able to complete.
315 ioctl(mdfd
, STOP_ARRAY
, NULL
);
317 ": You have an old buggy kernel which cannot support\n"
318 " --incremental reliably. Aborting.\n");
323 info
.array
.working_disks
= 1;
325 /* 6/ Make sure /var/run/mdadm.map contains this array. */
326 map_update(&map
, fd2devnum(mdfd
),
328 info
.uuid
, chosen_name
);
331 /* - check one drive in array to make sure metadata is a reasonably */
332 /* close match. Reject if not (e.g. different type) */
333 /* - add the device */
338 struct supertype
*st2
;
339 struct mdinfo info2
, *d
;
342 strcpy(chosen_name
, mp
->path
);
344 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
346 /* It is generally not OK to add non-spare drives to a
347 * running array as they are probably missing because
348 * they failed. However if runstop is 1, then the
349 * array was possibly started early and our best bet is
350 * to add this anyway.
351 * Also if action policy is re-add or better we allow
354 if ((info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
355 && ! policy_action_allows(policy
, st
->ss
->name
,
360 if (st
->ss
->external
) {
361 char *devname
= devnum2devname(fd2devnum(mdfd
));
363 active
= devname
&& is_container_active(devname
);
365 } else if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0)
369 ": not adding %s to active array (without --run) %s\n",
370 devname
, chosen_name
);
375 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), (GET_DEVS
| GET_STATE
));
381 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
382 sra
->devs
->disk
.minor
);
383 dfd2
= dev_open(dn
, O_RDONLY
);
385 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
386 st
->ss
->compare_super(st
, st2
) != 0) {
388 ": metadata mismatch between %s and "
390 devname
, chosen_name
);
396 memset(&info2
, 0, sizeof(info2
));
397 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
398 st2
->ss
->free_super(st2
);
399 if (info
.array
.level
!= info2
.array
.level
||
400 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
401 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
403 ": unexpected difference between %s and %s.\n",
404 chosen_name
, devname
);
409 info2
.disk
.major
= major(stb
.st_rdev
);
410 info2
.disk
.minor
= minor(stb
.st_rdev
);
411 /* add disk needs to know about containers */
412 if (st
->ss
->external
)
413 sra
->array
.level
= LEVEL_CONTAINER
;
414 err
= add_disk(mdfd
, st
, sra
, &info2
);
415 if (err
< 0 && errno
== EBUSY
) {
416 /* could be another device present with the same
417 * disk.number. Find and reject any such
419 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
420 info
.events
, verbose
, chosen_name
);
421 err
= add_disk(mdfd
, st
, sra
, &info2
);
424 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
425 devname
, chosen_name
, strerror(errno
));
429 info
.array
.working_disks
= 0;
430 for (d
= sra
->devs
; d
; d
=d
->next
)
431 info
.array
.working_disks
++;
435 /* 7/ Is there enough devices to possibly start the array? */
436 /* 7a/ if not, finish with success. */
437 if (info
.array
.level
== LEVEL_CONTAINER
) {
438 /* Try to assemble within the container */
440 sysfs_uevent(&info
, "change");
443 ": container %s now has %d devices\n",
444 chosen_name
, info
.array
.working_disks
);
445 wait_for(chosen_name
, mdfd
);
447 rv
= Incremental(chosen_name
, verbose
, runstop
,
448 NULL
, homehost
, require_homehost
, autof
);
450 /* Don't fail the whole -I if a subarray didn't
451 * have enough devices to start yet
457 active_disks
= count_active(st
, mdfd
, &avail
, &info
);
458 if (enough(info
.array
.level
, info
.array
.raid_disks
,
459 info
.array
.layout
, info
.array
.state
& 1,
460 avail
, active_disks
) == 0) {
464 ": %s attached to %s, not enough to start (%d).\n",
465 devname
, chosen_name
, active_disks
);
473 /* - if number of OK devices match expected, or -R and there */
475 /* + add any bitmap file */
476 /* + start the array (auto-readonly). */
478 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
481 ": %s attached to %s which is already active.\n",
482 devname
, chosen_name
);
489 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
490 struct mdinfo
*sra
, *dsk
;
491 /* Let's try to start it */
492 if (match
&& match
->bitmap_file
) {
493 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
496 ": Could not open bitmap file %s.\n",
500 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
503 ": Failed to set bitmapfile for %s.\n",
509 /* GET_* needed so add_disk works below */
510 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
),
511 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
);
512 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
513 && trustworthy
!= FOREIGN
)
514 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
516 rv
= sysfs_set_str(sra
, NULL
,
517 "array_state", "read-auto");
521 ": %s attached to %s, which has been started.\n",
522 devname
, chosen_name
);
524 wait_for(chosen_name
, mdfd
);
525 /* We just started the array, so some devices
526 * might have been evicted from the array
527 * because their event counts were too old.
528 * If the action=re-add policy is in-force for
529 * those devices we should re-add them now.
531 for (dsk
= sra
->devs
; dsk
; dsk
= dsk
->next
) {
532 if (disk_action_allows(dsk
, st
->ss
->name
, act_re_add
) &&
533 add_disk(mdfd
, st
, sra
, dsk
) == 0)
535 ": %s re-added to %s\n",
536 dsk
->sys_name
, chosen_name
);
540 ": %s attached to %s, but failed to start: %s.\n",
541 devname
, chosen_name
, strerror(errno
));
547 ": %s attached to %s, not enough to start safely.\n",
548 devname
, chosen_name
);
557 dev_policy_free(policy
);
561 static struct mddev_ident_s
*search_mdstat(struct supertype
*st
,
564 int verbose
, int *rvp
)
566 struct mddev_ident_s
*array_list
, *match
;
567 array_list
= conf_get_ident(NULL
);
569 for (; array_list
; array_list
= array_list
->next
) {
570 if (array_list
->uuid_set
&&
571 same_uuid(array_list
->uuid
, info
->uuid
, st
->ss
->swapuuid
)
573 if (verbose
>= 2 && array_list
->devname
)
575 ": UUID differs from %s.\n",
576 array_list
->devname
);
579 if (array_list
->name
[0] &&
580 strcasecmp(array_list
->name
, info
->name
) != 0) {
581 if (verbose
>= 2 && array_list
->devname
)
583 ": Name differs from %s.\n",
584 array_list
->devname
);
587 if (array_list
->devices
&&
588 !match_oneof(array_list
->devices
, devname
)) {
589 if (verbose
>= 2 && array_list
->devname
)
591 ": Not a listed device for %s.\n",
592 array_list
->devname
);
595 if (array_list
->super_minor
!= UnSet
&&
596 array_list
->super_minor
!= info
->array
.md_minor
) {
597 if (verbose
>= 2 && array_list
->devname
)
599 ": Different super-minor to %s.\n",
600 array_list
->devname
);
603 if (!array_list
->uuid_set
&&
604 !array_list
->name
[0] &&
605 !array_list
->devices
&&
606 array_list
->super_minor
== UnSet
) {
607 if (verbose
>= 2 && array_list
->devname
)
609 ": %s doesn't have any identifying information.\n",
610 array_list
->devname
);
613 /* FIXME, should I check raid_disks and level too?? */
617 if (match
->devname
&& array_list
->devname
)
619 ": we match both %s and %s - cannot decide which to use.\n",
620 match
->devname
, array_list
->devname
);
623 ": multiple lines in mdadm.conf match\n");
634 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
635 int number
, __u64 events
, int verbose
,
638 /* Find a device attached to this array with a disk.number of number
639 * and events less than the passed events, and remove the device.
644 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
645 return; /* not safe to remove from active arrays
646 * without thinking more */
648 for (d
= sra
->devs
; d
; d
= d
->next
) {
652 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
653 dfd
= dev_open(dn
, O_RDONLY
);
656 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
660 st
->ss
->getinfo_super(st
, &info
, NULL
);
661 st
->ss
->free_super(st
);
664 if (info
.disk
.number
!= number
||
665 info
.events
>= events
)
668 if (d
->disk
.raid_disk
> -1)
669 sysfs_set_str(sra
, d
, "slot", "none");
670 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
673 ": removing old device %s from %s\n",
674 d
->sys_name
+4, array_name
);
678 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
679 struct mdinfo
*bestinfo
)
681 /* count how many devices in sra think they are active */
683 int cnt
= 0, cnt1
= 0;
684 __u64 max_events
= 0;
685 struct mdinfo
*sra
= sysfs_read(mdfd
, -1, GET_DEVS
| GET_STATE
);
691 for (d
= sra
->devs
; d
; d
= d
->next
) {
697 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
698 dfd
= dev_open(dn
, O_RDONLY
);
701 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
705 st
->ss
->getinfo_super(st
, &info
, NULL
);
707 avail
= malloc(info
.array
.raid_disks
);
709 fprintf(stderr
, Name
": out of memory.\n");
712 memset(avail
, 0, info
.array
.raid_disks
);
716 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
720 max_events
= info
.events
;
721 avail
[info
.disk
.raid_disk
] = 2;
722 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
723 } else if (info
.events
== max_events
) {
725 avail
[info
.disk
.raid_disk
] = 2;
726 } else if (info
.events
== max_events
-1) {
728 avail
[info
.disk
.raid_disk
] = 1;
729 } else if (info
.events
< max_events
- 1)
731 else if (info
.events
== max_events
+1) {
735 max_events
= info
.events
;
736 for (i
=0; i
<info
.array
.raid_disks
; i
++)
739 avail
[info
.disk
.raid_disk
] = 2;
740 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
741 } else { /* info.events much bigger */
743 memset(avail
, 0, info
.disk
.raid_disk
);
744 max_events
= info
.events
;
745 avail
[info
.disk
.raid_disk
] = 2;
746 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
749 st
->ss
->free_super(st
);
754 static int array_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
755 struct supertype
*st
, int verbose
)
757 /* This device doesn't have any md metadata
758 * If it is 'bare' and theh device policy allows 'spare' look for
759 * an array or container to attach it to.
760 * If st is set, then only arrays of that type are considered
761 * Return 0 on success, or some exit code on failure, probably 1.
765 struct map_ent
*mp
, *map
= NULL
;
766 struct mdinfo
*chosen
= NULL
;
769 if (fstat(dfd
, &stb
) != 0)
773 * Now we need to find a suitable array to add this to.
774 * We only accept arrays that:
776 * - are in the same domains as the device
777 * - are of an size for which the device will be useful
778 * and we choose the one that is the most degraded
781 if (map_lock(&map
)) {
782 fprintf(stderr
, Name
": failed to get exclusive lock on "
786 for (mp
= map
; mp
; mp
= mp
->next
) {
787 struct supertype
*st2
;
788 struct domainlist
*dl
= NULL
;
790 unsigned long long devsize
;
792 if (is_subarray(mp
->metadata
))
795 st2
= st
->ss
->match_metadata_desc(mp
->metadata
);
797 (st
->minor_version
>= 0 &&
798 st
->minor_version
!= st2
->minor_version
)) {
800 fprintf(stderr
, Name
": not adding %s to %s as metadata type doesn't match\n",
807 sra
= sysfs_read(-1, mp
->devnum
,
808 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
809 GET_DEGRADED
|GET_COMPONENT
|GET_VERSION
);
811 /* Probably a container - no degraded info */
812 sra
= sysfs_read(-1, mp
->devnum
,
813 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
814 GET_COMPONENT
|GET_VERSION
);
816 sra
->array
.failed_disks
= 0;
823 for(i
=0; !st2
&& superlist
[i
]; i
++)
824 st2
= superlist
[i
]->match_metadata_desc(
828 get_dev_size(dfd
, NULL
, &devsize
);
829 if (st2
->ss
->avail_size(st2
, devsize
) < sra
->component_size
) {
831 fprintf(stderr
, Name
": not adding %s to %s as it is too small\n",
835 dl
= domain_from_array(sra
, st2
->ss
->name
);
836 if (!domain_test(dl
, pol
, st2
->ss
->name
)) {
837 /* domain test fails */
839 fprintf(stderr
, Name
": not adding %s to %s as it is not in a compatible domain\n",
844 /* all tests passed, OK to add to this array */
848 } else if (chosen
->array
.failed_disks
< sra
->array
.failed_disks
) {
862 /* add current device to chosen array as a spare */
863 int mdfd
= open_dev(devname2devnum(chosen
->sys_name
));
865 struct mddev_dev_s devlist
;
870 devlist
.writemostly
= 0;
871 devlist
.devname
= devname
;
872 sprintf(devname
, "%d:%d", major(stb
.st_rdev
),
874 devlist
.disposition
= 'a';
877 rv
= Manage_subdevs(chosen
->sys_name
, mdfd
, &devlist
,
883 fprintf(stderr
, Name
": added %s as spare for %s\n",
884 devname
, chosen
->sys_name
);
886 fprintf(stderr
, Name
": failed to add %s as spare for %s\n",
887 devname
, chosen
->sys_name
);
894 static int partition_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
895 struct supertype
*st
, int verbose
)
897 /* we know that at least one partition virtual-metadata is
898 * allowed to incorporate spares like this device. We need to
899 * find a suitable device to copy partition information from.
901 * Getting a list of all disk (not partition) devices is
902 * slightly non-trivial. We could look at /sys/block, but
903 * that is theoretically due to be removed. Maybe best to use
904 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
905 * we depend on this directory of 'path' info. But that fails
906 * to find loop devices and probably others. Maybe don't
907 * worry about that, they aren't the real target.
909 * So: check things in /dev/disk/by-path to see if they are in
910 * a compatible domain, then load the partition table and see
911 * if it is OK for the new device, and choose the largest
912 * partition table that fits.
917 unsigned long long chosen_size
;
918 struct supertype
*chosen_st
= NULL
;
921 dir
= opendir("/dev/disk/by-path");
924 while ((de
= readdir(dir
)) != NULL
) {
926 struct dev_policy
*pol2
= NULL
;
927 struct domainlist
*domlist
= NULL
;
930 struct supertype
*st2
= NULL
;
931 char *devname
= NULL
;
932 unsigned long long devsectors
;
934 if (de
->d_ino
== 0 ||
935 de
->d_name
[0] == '.' ||
936 (de
->d_type
!= DT_LNK
&& de
->d_type
!= DT_UNKNOWN
))
939 ep
= de
->d_name
+ strlen(de
->d_name
);
940 while (ep
> de
->d_name
&&
943 if (ep
> de
->d_name
+ 5 &&
944 strncmp(ep
-5, "-part", 5) == 0)
945 /* This is a partition - skip it */
948 pol2
= path_policy(de
->d_name
, type_disk
);
950 domain_merge(&domlist
, pol2
, st
? st
->ss
->name
: NULL
);
951 if (domain_test(domlist
, pol
, st
? st
->ss
->name
: NULL
) == 0)
952 /* new device is incompatible with this device. */
955 domain_free(domlist
);
958 asprintf(&devname
, "/dev/disk/by-path/%s", de
->d_name
);
959 fd
= open(devname
, O_RDONLY
);
962 if (get_dev_size(fd
, devname
, &devsectors
) == 0)
969 st2
= guess_super_type(fd
, guess_partitions
);
971 st2
->ss
->load_super(st2
, fd
, NULL
) < 0)
975 /* Check domain policy again, this time referring to metadata */
976 domain_merge(&domlist
, pol2
, st2
->ss
->name
);
977 if (domain_test(domlist
, pol
, st2
->ss
->name
) == 0)
978 /* Incompatible devices for this metadata type */
982 st2
->ss
->getinfo_super(st2
, &info
, NULL
);
983 if (info
.component_size
> devsectors
)
984 /* This partitioning doesn't fit in the device */
987 /* This is an acceptable device to copy partition
988 * metadata from. We could just stop here, but I
989 * think I want to keep looking incase a larger
990 * metadata which makes better use of the device can
993 if (chosen
== NULL
||
994 chosen_size
< info
.component_size
) {
995 chosen_size
= info
.component_size
;
1000 chosen_st
->ss
->free_super(chosen_st
);
1009 domain_free(domlist
);
1010 dev_policy_free(pol2
);
1012 st2
->ss
->free_super(st2
);
1022 /* 'chosen' is the best device we can find. Let's write its
1023 * metadata to devname dfd is read-only so don't use that
1025 fd
= open(devname
, O_RDWR
);
1027 chosen_st
->ss
->store_super(chosen_st
, fd
);
1031 chosen_st
->ss
->free_super(chosen_st
);
1037 /* adding a spare to a regular array is quite different from adding one to
1038 * a set-of-partitions virtual array.
1039 * This function determines which is worth trying and tries as appropriate.
1040 * Arrays are given priority over partitions.
1042 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1043 struct supertype
*st
, int verbose
)
1048 int partitions_ok
= 0;
1049 char bufpad
[4096 + 4096];
1050 char *buf
= (char*)(((long)bufpad
+ 4096) & ~4095);
1053 /* Can only add a spare if device has at least one domains */
1054 if (pol_find(pol
, pol_domain
) == NULL
)
1056 /* And only if some action allows spares */
1057 if (!policy_action_allows(pol
, st
?st
->ss
->name
:NULL
, act_spare
))
1060 /* Now check if the device is bare - we don't add non-bare devices
1061 * yet even if action=-spare
1064 if (lseek(dfd
, 0, SEEK_SET
) != 0 ||
1065 read(dfd
, buf
, 4096) != 4096) {
1068 fprintf(stderr
, Name
": %s is not bare, so not considering as a spare\n",
1072 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1074 if (memcmp(buf
, buf
+1, 4095) != 0)
1077 /* OK, first 4K appear blank, try the end. */
1078 if (lseek(dfd
, -4096, SEEK_END
) < 0 ||
1079 read(dfd
, buf
, 4096) != 4096)
1082 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1084 if (memcmp(buf
, buf
+1, 4095) != 0)
1087 /* This device passes our test for 'is bare'.
1088 * Let's see what policy allows for such things.
1091 /* just try try 'array' or 'partition' based on this metadata */
1092 if (st
->ss
->add_to_super
)
1093 return array_try_spare(devname
, dfdp
, pol
,
1096 return partition_try_spare(devname
, dfdp
, pol
,
1099 /* Now see which metadata type support spare */
1100 for (i
= 0; (!arrays_ok
|| !partitions_ok
) && superlist
[i
] ; i
++) {
1101 if (superlist
[i
]->add_to_super
&& !arrays_ok
&&
1102 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1104 if (superlist
[i
]->add_to_super
== NULL
&& !partitions_ok
&&
1105 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1110 rv
= array_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1111 if (rv
== 0 && partitions_ok
)
1112 rv
= partition_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1116 int IncrementalScan(int verbose
)
1118 /* look at every device listed in the 'map' file.
1119 * If one is found that is not running then:
1120 * look in mdadm.conf for bitmap file.
1121 * if one exists, but array has none, add it.
1122 * try to start array in auto-readonly mode
1124 struct map_ent
*mapl
= NULL
;
1126 mddev_ident_t devs
, mddev
;
1130 devs
= conf_get_ident(NULL
);
1132 for (me
= mapl
; me
; me
= me
->next
) {
1133 mdu_array_info_t array
;
1134 mdu_bitmap_file_t bmf
;
1136 int mdfd
= open_dev(me
->devnum
);
1140 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
1145 /* Ok, we can try this one. Maybe it needs a bitmap */
1146 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
1147 if (mddev
->devname
&& me
->path
1148 && devname_matches(mddev
->devname
, me
->path
))
1150 if (mddev
&& mddev
->bitmap_file
) {
1152 * Note: early kernels will wrongly fail this, so it
1156 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
1157 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
1159 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
1166 fprintf(stderr
, Name
1167 ": Added bitmap %s to %s\n",
1168 mddev
->bitmap_file
, me
->path
);
1169 else if (errno
!= EEXIST
)
1170 fprintf(stderr
, Name
1171 ": Failed to add bitmap to %s: %s\n",
1172 me
->path
, strerror(errno
));
1175 sra
= sysfs_read(mdfd
, 0, 0);
1177 if (sysfs_set_str(sra
, NULL
,
1178 "array_state", "read-auto") == 0) {
1180 fprintf(stderr
, Name
1181 ": started array %s\n",
1182 me
->path
?: devnum2devname(me
->devnum
));
1184 fprintf(stderr
, Name
1185 ": failed to start array %s: %s\n",
1186 me
->path
?: devnum2devname(me
->devnum
),
1195 static char *container2devname(char *devname
)
1197 char *mdname
= NULL
;
1199 if (devname
[0] == '/') {
1200 int fd
= open(devname
, O_RDONLY
);
1202 mdname
= devnum2devname(fd2devnum(fd
));
1207 struct map_ent
*mp
, *map
= NULL
;
1209 if (!parse_uuid(devname
, uuid
))
1211 mp
= map_by_uuid(&map
, uuid
);
1213 mdname
= devnum2devname(mp
->devnum
);
1220 static int Incremental_container(struct supertype
*st
, char *devname
,
1221 char *homehost
, int verbose
,
1222 int runstop
, int autof
)
1224 /* Collect the contents of this container and for each
1225 * array, choose a device name and assemble the array.
1228 struct mdinfo
*list
;
1230 struct map_ent
*map
= NULL
;
1233 struct mddev_ident_s
*match
;
1236 memset(&info
, 0, sizeof(info
));
1237 st
->ss
->getinfo_super(st
, &info
, NULL
);
1239 if ((runstop
> 0 && info
.container_enough
>= 0) ||
1240 info
.container_enough
> 0)
1244 fprintf(stderr
, Name
": not enough devices to start the container\n");
1248 match
= search_mdstat(st
, &info
, devname
, verbose
, &rv
);
1249 if (match
== NULL
&& rv
== 2)
1252 /* Need to compute 'trustworthy' */
1254 trustworthy
= LOCAL
;
1255 else if (st
->ss
->match_home(st
, homehost
) == 1)
1256 trustworthy
= LOCAL
;
1257 else if (st
->ss
->match_home(st
, "any") == 1)
1258 trustworthy
= LOCAL
;
1260 trustworthy
= FOREIGN
;
1262 list
= st
->ss
->container_content(st
, NULL
);
1264 fprintf(stderr
, Name
": failed to get exclusive lock on "
1267 for (ra
= list
; ra
; ra
= ra
->next
) {
1269 char chosen_name
[1024];
1271 struct mddev_ident_s
*match
= NULL
;
1273 mp
= map_by_uuid(&map
, ra
->uuid
);
1276 mdfd
= open_dev(mp
->devnum
);
1278 strcpy(chosen_name
, mp
->path
);
1280 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
1283 /* Check in mdadm.conf for container == devname and
1284 * member == ra->text_version after second slash.
1286 char *sub
= strchr(ra
->text_version
+1, '/');
1287 struct mddev_ident_s
*array_list
;
1290 array_list
= conf_get_ident(NULL
);
1293 for(; array_list
; array_list
= array_list
->next
) {
1295 if (array_list
->member
== NULL
||
1296 array_list
->container
== NULL
)
1298 if (strcmp(array_list
->member
, sub
) != 0)
1300 if (array_list
->uuid_set
&&
1301 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
1303 dn
= container2devname(array_list
->container
);
1306 if (strncmp(dn
, ra
->text_version
+1,
1308 ra
->text_version
[strlen(dn
)+1] != '/') {
1313 /* we have a match */
1316 fprintf(stderr
, Name
": match found for member %s\n",
1317 array_list
->member
);
1321 if (match
&& match
->devname
&&
1322 strcasecmp(match
->devname
, "<ignore>") == 0) {
1324 fprintf(stderr
, Name
": array %s/%s is "
1325 "explicitly ignored by mdadm.conf\n",
1326 match
->container
, match
->member
);
1330 trustworthy
= LOCAL
;
1332 mdfd
= create_mddev(match
? match
->devname
: NULL
,
1340 fprintf(stderr
, Name
": failed to open %s: %s.\n",
1341 chosen_name
, strerror(errno
));
1345 assemble_container_content(st
, mdfd
, ra
, runstop
,
1346 chosen_name
, verbose
);
1353 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1354 * raid arrays, and if so first fail (if needed) and then remove the device.
1356 * @devname - The device we want to remove
1358 * Note: the device name must be a kernel name like "sda", so
1359 * that we can find it in /proc/mdstat
1361 int IncrementalRemove(char *devname
, int verbose
)
1365 struct mdstat_ent
*ent
;
1366 struct mddev_dev_s devlist
;
1368 if (strchr(devname
, '/')) {
1369 fprintf(stderr
, Name
": incremental removal requires a "
1370 "kernel device name, not a file: %s\n", devname
);
1373 ent
= mdstat_by_component(devname
);
1375 fprintf(stderr
, Name
": %s does not appear to be a component "
1376 "of any array\n", devname
);
1379 mdfd
= open_dev(ent
->devnum
);
1381 fprintf(stderr
, Name
": Cannot open array %s!!\n", ent
->dev
);
1384 memset(&devlist
, 0, sizeof(devlist
));
1385 devlist
.devname
= devname
;
1386 devlist
.disposition
= 'f';
1387 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);
1388 devlist
.disposition
= 'r';
1389 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);