]>
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
,
44 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;
102 int trustworthy
= FOREIGN
;
104 mdu_array_info_t ainf
;
105 struct dev_policy
*policy
= NULL
;
107 struct createinfo
*ci
= conf_get_create_info();
110 /* 1/ Check if device is permitted by mdadm.conf */
112 if (!conf_test_dev(devname
)) {
115 ": %s not permitted by mdadm.conf.\n",
120 /* 2/ Find metadata, reject if none appropriate (check
121 * version/name from args) */
123 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
126 fprintf(stderr
, Name
": cannot open %s: %s.\n",
127 devname
, strerror(errno
));
130 if (fstat(dfd
, &stb
) < 0) {
132 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
133 devname
, strerror(errno
));
136 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
138 fprintf(stderr
, Name
": %s is not a block device.\n",
143 dinfo
.disk
.major
= major(stb
.st_rdev
);
144 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
146 policy
= disk_policy(&dinfo
);
148 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
151 ": no recognisable superblock on %s.\n",
153 rv
= try_spare(devname
, &dfd
, policy
, st
, verbose
);
156 if (st
->ss
->compare_super
== NULL
||
157 st
->ss
->load_super(st
, dfd
, NULL
)) {
159 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
161 rv
= try_spare(devname
, &dfd
, policy
, st
, verbose
);
165 close (dfd
); dfd
= -1;
167 memset(&info
, 0, sizeof(info
));
168 st
->ss
->getinfo_super(st
, &info
, NULL
);
170 /* 3/ Check if there is a match in mdadm.conf */
171 match
= search_mdstat(st
, &info
, devname
, verbose
, &rv
);
172 if (!match
&& rv
== 2)
175 if (match
&& match
->devname
176 && strcasecmp(match
->devname
, "<ignore>") == 0) {
178 fprintf(stderr
, Name
": array containing %s is explicitly"
179 " ignored by mdadm.conf\n",
184 /* 3a/ if not, check for homehost match. If no match, continue
185 * but don't trust the 'name' in the array. Thus a 'random' minor
186 * number will be assigned, and the device name will be based
190 else if (st
->ss
->match_home(st
, homehost
) == 1)
192 else if (st
->ss
->match_home(st
, "any") == 1)
193 trustworthy
= LOCAL_ANY
;
195 trustworthy
= FOREIGN
;
198 if (!match
&& !conf_test_metadata(st
->ss
->name
, policy
,
199 (trustworthy
== LOCAL
))) {
202 ": %s has metadata type %s for which "
203 "auto-assembly is disabled\n",
204 devname
, st
->ss
->name
);
207 if (trustworthy
== LOCAL_ANY
)
210 /* There are three possible sources for 'autof': command line,
211 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
212 * ARRAY takes precedence, then command line, then
215 if (match
&& match
->autof
)
216 autof
= match
->autof
;
220 if (st
->ss
->container_content
&& st
->loaded_container
) {
221 if ((runstop
> 0 && info
.container_enough
>= 0) ||
222 info
.container_enough
> 0)
226 fprintf(stderr
, Name
": not enough devices to start the container\n");
231 /* This is a pre-built container array, so we do something
234 rv
= Incremental_container(st
, devname
, verbose
, runstop
,
239 name_to_use
= info
.name
;
240 if (name_to_use
[0] == 0 &&
241 info
.array
.level
== LEVEL_CONTAINER
&&
242 trustworthy
== LOCAL
) {
243 name_to_use
= info
.text_version
;
244 trustworthy
= METADATA
;
246 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
247 ! require_homehost
&&
248 conf_name_is_free(name_to_use
))
251 /* strip "hostname:" prefix from name if we have decided
252 * to treat it as LOCAL
254 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
255 name_to_use
= strchr(name_to_use
, ':')+1;
257 /* 4/ Check if array exists.
260 fprintf(stderr
, Name
": failed to get exclusive lock on "
262 mp
= map_by_uuid(&map
, info
.uuid
);
264 mdfd
= open_dev(mp
->devnum
);
271 /* Couldn't find an existing array, maybe make a new one */
272 mdfd
= create_mddev(match
? match
->devname
: NULL
,
273 name_to_use
, autof
, trustworthy
, chosen_name
);
278 sysfs_init(&info
, mdfd
, 0);
280 if (set_array_info(mdfd
, st
, &info
) != 0) {
281 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
282 chosen_name
, strerror(errno
));
288 dinfo
.disk
.major
= major(stb
.st_rdev
);
289 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
290 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
291 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
292 devname
, chosen_name
, strerror(errno
));
293 ioctl(mdfd
, STOP_ARRAY
, 0);
297 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), GET_DEVS
);
298 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
299 /* It really should be 'none' - must be old buggy
300 * kernel, and mdadm -I may not be able to complete.
303 ioctl(mdfd
, STOP_ARRAY
, NULL
);
305 ": You have an old buggy kernel which cannot support\n"
306 " --incremental reliably. Aborting.\n");
311 info
.array
.working_disks
= 1;
313 /* 6/ Make sure /var/run/mdadm.map contains this array. */
314 map_update(&map
, fd2devnum(mdfd
),
316 info
.uuid
, chosen_name
);
319 /* - check one drive in array to make sure metadata is a reasonably */
320 /* close match. Reject if not (e.g. different type) */
321 /* - add the device */
326 struct supertype
*st2
;
327 struct mdinfo info2
, *d
;
330 strcpy(chosen_name
, mp
->path
);
332 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
334 /* It is generally not OK to add non-spare drives to a
335 * running array as they are probably missing because
336 * they failed. However if runstop is 1, then the
337 * array was possibly started early and our best bet is
338 * to add this anyway.
339 * Also if action policy is re-add or better we allow
342 if ((info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
343 && ! policy_action_allows(policy
, st
->ss
->name
,
348 if (st
->ss
->external
) {
349 char *devname
= devnum2devname(fd2devnum(mdfd
));
351 active
= devname
&& is_container_active(devname
);
353 } else if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0)
357 ": not adding %s to active array (without --run) %s\n",
358 devname
, chosen_name
);
363 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
), (GET_DEVS
| GET_STATE
));
369 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
370 sra
->devs
->disk
.minor
);
371 dfd2
= dev_open(dn
, O_RDONLY
);
373 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
374 st
->ss
->compare_super(st
, st2
) != 0) {
376 ": metadata mismatch between %s and "
378 devname
, chosen_name
);
384 memset(&info2
, 0, sizeof(info2
));
385 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
386 st2
->ss
->free_super(st2
);
387 if (info
.array
.level
!= info2
.array
.level
||
388 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
389 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
391 ": unexpected difference between %s and %s.\n",
392 chosen_name
, devname
);
397 info2
.disk
.major
= major(stb
.st_rdev
);
398 info2
.disk
.minor
= minor(stb
.st_rdev
);
399 /* add disk needs to know about containers */
400 if (st
->ss
->external
)
401 sra
->array
.level
= LEVEL_CONTAINER
;
402 err
= add_disk(mdfd
, st
, sra
, &info2
);
403 if (err
< 0 && errno
== EBUSY
) {
404 /* could be another device present with the same
405 * disk.number. Find and reject any such
407 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
408 info
.events
, verbose
, chosen_name
);
409 err
= add_disk(mdfd
, st
, sra
, &info2
);
412 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
413 devname
, chosen_name
, strerror(errno
));
417 info
.array
.working_disks
= 0;
418 for (d
= sra
->devs
; d
; d
=d
->next
)
419 info
.array
.working_disks
++;
423 /* 7/ Is there enough devices to possibly start the array? */
424 /* 7a/ if not, finish with success. */
425 if (info
.array
.level
== LEVEL_CONTAINER
) {
426 /* Try to assemble within the container */
428 sysfs_uevent(&info
, "change");
431 ": container %s now has %d devices\n",
432 chosen_name
, info
.array
.working_disks
);
433 wait_for(chosen_name
, mdfd
);
435 rv
= Incremental(chosen_name
, verbose
, runstop
,
436 NULL
, homehost
, require_homehost
, autof
);
438 /* Don't fail the whole -I if a subarray didn't
439 * have enough devices to start yet
445 active_disks
= count_active(st
, mdfd
, &avail
, &info
);
446 if (enough(info
.array
.level
, info
.array
.raid_disks
,
447 info
.array
.layout
, info
.array
.state
& 1,
448 avail
, active_disks
) == 0) {
452 ": %s attached to %s, not enough to start (%d).\n",
453 devname
, chosen_name
, active_disks
);
461 /* - if number of OK devices match expected, or -R and there */
463 /* + add any bitmap file */
464 /* + start the array (auto-readonly). */
466 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
469 ": %s attached to %s which is already active.\n",
470 devname
, chosen_name
);
477 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
478 struct mdinfo
*sra
, *dsk
;
479 /* Let's try to start it */
480 if (match
&& match
->bitmap_file
) {
481 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
484 ": Could not open bitmap file %s.\n",
488 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
491 ": Failed to set bitmapfile for %s.\n",
497 /* GET_* needed so add_disk works below */
498 sra
= sysfs_read(mdfd
, fd2devnum(mdfd
),
499 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
);
500 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
501 && trustworthy
!= FOREIGN
)
502 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
504 rv
= sysfs_set_str(sra
, NULL
,
505 "array_state", "read-auto");
509 ": %s attached to %s, which has been started.\n",
510 devname
, chosen_name
);
512 wait_for(chosen_name
, mdfd
);
513 /* We just started the array, so some devices
514 * might have been evicted from the array
515 * because their event counts were too old.
516 * If the action=re-add policy is in-force for
517 * those devices we should re-add them now.
519 for (dsk
= sra
->devs
; dsk
; dsk
= dsk
->next
) {
520 if (disk_action_allows(dsk
, st
->ss
->name
, act_re_add
) &&
521 add_disk(mdfd
, st
, sra
, dsk
) == 0)
523 ": %s re-added to %s\n",
524 dsk
->sys_name
, chosen_name
);
528 ": %s attached to %s, but failed to start: %s.\n",
529 devname
, chosen_name
, strerror(errno
));
535 ": %s attached to %s, not enough to start safely.\n",
536 devname
, chosen_name
);
545 dev_policy_free(policy
);
549 static struct mddev_ident_s
*search_mdstat(struct supertype
*st
,
552 int verbose
, int *rvp
)
554 struct mddev_ident_s
*array_list
, *match
;
555 array_list
= conf_get_ident(NULL
);
557 for (; array_list
; array_list
= array_list
->next
) {
558 if (array_list
->uuid_set
&&
559 same_uuid(array_list
->uuid
, info
->uuid
, st
->ss
->swapuuid
)
561 if (verbose
>= 2 && array_list
->devname
)
563 ": UUID differs from %s.\n",
564 array_list
->devname
);
567 if (array_list
->name
[0] &&
568 strcasecmp(array_list
->name
, info
->name
) != 0) {
569 if (verbose
>= 2 && array_list
->devname
)
571 ": Name differs from %s.\n",
572 array_list
->devname
);
575 if (array_list
->devices
&&
576 !match_oneof(array_list
->devices
, devname
)) {
577 if (verbose
>= 2 && array_list
->devname
)
579 ": Not a listed device for %s.\n",
580 array_list
->devname
);
583 if (array_list
->super_minor
!= UnSet
&&
584 array_list
->super_minor
!= info
->array
.md_minor
) {
585 if (verbose
>= 2 && array_list
->devname
)
587 ": Different super-minor to %s.\n",
588 array_list
->devname
);
591 if (!array_list
->uuid_set
&&
592 !array_list
->name
[0] &&
593 !array_list
->devices
&&
594 array_list
->super_minor
== UnSet
) {
595 if (verbose
>= 2 && array_list
->devname
)
597 ": %s doesn't have any identifying information.\n",
598 array_list
->devname
);
601 /* FIXME, should I check raid_disks and level too?? */
605 if (match
->devname
&& array_list
->devname
)
607 ": we match both %s and %s - cannot decide which to use.\n",
608 match
->devname
, array_list
->devname
);
611 ": multiple lines in mdadm.conf match\n");
622 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
623 int number
, __u64 events
, int verbose
,
626 /* Find a device attached to this array with a disk.number of number
627 * and events less than the passed events, and remove the device.
632 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
633 return; /* not safe to remove from active arrays
634 * without thinking more */
636 for (d
= sra
->devs
; d
; d
= d
->next
) {
640 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
641 dfd
= dev_open(dn
, O_RDONLY
);
644 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
648 st
->ss
->getinfo_super(st
, &info
, NULL
);
649 st
->ss
->free_super(st
);
652 if (info
.disk
.number
!= number
||
653 info
.events
>= events
)
656 if (d
->disk
.raid_disk
> -1)
657 sysfs_set_str(sra
, d
, "slot", "none");
658 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
661 ": removing old device %s from %s\n",
662 d
->sys_name
+4, array_name
);
666 static int count_active(struct supertype
*st
, int mdfd
, char **availp
,
667 struct mdinfo
*bestinfo
)
669 /* count how many devices in sra think they are active */
671 int cnt
= 0, cnt1
= 0;
672 __u64 max_events
= 0;
673 struct mdinfo
*sra
= sysfs_read(mdfd
, -1, GET_DEVS
| GET_STATE
);
679 for (d
= sra
->devs
; d
; d
= d
->next
) {
685 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
686 dfd
= dev_open(dn
, O_RDONLY
);
689 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
693 st
->ss
->getinfo_super(st
, &info
, NULL
);
695 avail
= malloc(info
.array
.raid_disks
);
697 fprintf(stderr
, Name
": out of memory.\n");
700 memset(avail
, 0, info
.array
.raid_disks
);
704 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
708 max_events
= info
.events
;
709 avail
[info
.disk
.raid_disk
] = 2;
710 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
711 } else if (info
.events
== max_events
) {
713 avail
[info
.disk
.raid_disk
] = 2;
714 } else if (info
.events
== max_events
-1) {
716 avail
[info
.disk
.raid_disk
] = 1;
717 } else if (info
.events
< max_events
- 1)
719 else if (info
.events
== max_events
+1) {
723 max_events
= info
.events
;
724 for (i
=0; i
<info
.array
.raid_disks
; i
++)
727 avail
[info
.disk
.raid_disk
] = 2;
728 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
729 } else { /* info.events much bigger */
731 memset(avail
, 0, info
.disk
.raid_disk
);
732 max_events
= info
.events
;
733 avail
[info
.disk
.raid_disk
] = 2;
734 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
737 st
->ss
->free_super(st
);
742 static int array_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
743 struct supertype
*st
, int verbose
)
745 /* This device doesn't have any md metadata
746 * If it is 'bare' and theh device policy allows 'spare' look for
747 * an array or container to attach it to.
748 * If st is set, then only arrays of that type are considered
749 * Return 0 on success, or some exit code on failure, probably 1.
753 struct map_ent
*mp
, *map
= NULL
;
754 struct mdinfo
*chosen
= NULL
;
757 if (fstat(dfd
, &stb
) != 0)
761 * Now we need to find a suitable array to add this to.
762 * We only accept arrays that:
764 * - are in the same domains as the device
765 * - are of an size for which the device will be useful
766 * and we choose the one that is the most degraded
769 if (map_lock(&map
)) {
770 fprintf(stderr
, Name
": failed to get exclusive lock on "
774 for (mp
= map
; mp
; mp
= mp
->next
) {
775 struct supertype
*st2
;
776 struct domainlist
*dl
= NULL
;
778 unsigned long long devsize
;
780 if (is_subarray(mp
->metadata
))
783 st2
= st
->ss
->match_metadata_desc(mp
->metadata
);
785 (st
->minor_version
>= 0 &&
786 st
->minor_version
!= st2
->minor_version
)) {
788 fprintf(stderr
, Name
": not adding %s to %s as metadata type doesn't match\n",
795 sra
= sysfs_read(-1, mp
->devnum
,
796 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
797 GET_DEGRADED
|GET_COMPONENT
|GET_VERSION
);
799 /* Probably a container - no degraded info */
800 sra
= sysfs_read(-1, mp
->devnum
,
801 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
802 GET_COMPONENT
|GET_VERSION
);
804 sra
->array
.failed_disks
= 0;
811 for(i
=0; !st2
&& superlist
[i
]; i
++)
812 st2
= superlist
[i
]->match_metadata_desc(
816 get_dev_size(dfd
, NULL
, &devsize
);
817 if (st2
->ss
->avail_size(st2
, devsize
) < sra
->component_size
) {
819 fprintf(stderr
, Name
": not adding %s to %s as it is too small\n",
823 dl
= domain_from_array(sra
, st2
->ss
->name
);
824 if (!domain_test(dl
, pol
, st2
->ss
->name
)) {
825 /* domain test fails */
827 fprintf(stderr
, Name
": not adding %s to %s as it is not in a compatible domain\n",
832 /* all tests passed, OK to add to this array */
836 } else if (chosen
->array
.failed_disks
< sra
->array
.failed_disks
) {
850 /* add current device to chosen array as a spare */
851 int mdfd
= open_dev(devname2devnum(chosen
->sys_name
));
853 struct mddev_dev_s devlist
;
858 devlist
.writemostly
= 0;
859 devlist
.devname
= devname
;
860 sprintf(devname
, "%d:%d", major(stb
.st_rdev
),
862 devlist
.disposition
= 'a';
865 rv
= Manage_subdevs(chosen
->sys_name
, mdfd
, &devlist
,
871 fprintf(stderr
, Name
": added %s as spare for %s\n",
872 devname
, chosen
->sys_name
);
874 fprintf(stderr
, Name
": failed to add %s as spare for %s\n",
875 devname
, chosen
->sys_name
);
882 static int partition_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
883 struct supertype
*st
, int verbose
)
885 /* we know that at least one partition virtual-metadata is
886 * allowed to incorporate spares like this device. We need to
887 * find a suitable device to copy partition information from.
889 * Getting a list of all disk (not partition) devices is
890 * slightly non-trivial. We could look at /sys/block, but
891 * that is theoretically due to be removed. Maybe best to use
892 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
893 * we depend on this directory of 'path' info. But that fails
894 * to find loop devices and probably others. Maybe don't
895 * worry about that, they aren't the real target.
897 * So: check things in /dev/disk/by-path to see if they are in
898 * a compatible domain, then load the partition table and see
899 * if it is OK for the new device, and choose the largest
900 * partition table that fits.
905 unsigned long long chosen_size
;
906 struct supertype
*chosen_st
= NULL
;
909 dir
= opendir("/dev/disk/by-path");
912 while ((de
= readdir(dir
)) != NULL
) {
914 struct dev_policy
*pol2
= NULL
;
915 struct domainlist
*domlist
= NULL
;
918 struct supertype
*st2
= NULL
;
919 char *devname
= NULL
;
920 unsigned long long devsectors
;
922 if (de
->d_ino
== 0 ||
923 de
->d_name
[0] == '.' ||
924 (de
->d_type
!= DT_LNK
&& de
->d_type
!= DT_UNKNOWN
))
927 ep
= de
->d_name
+ strlen(de
->d_name
);
928 while (ep
> de
->d_name
&&
931 if (ep
> de
->d_name
+ 5 &&
932 strncmp(ep
-5, "-part", 5) == 0)
933 /* This is a partition - skip it */
936 pol2
= path_policy(de
->d_name
, type_disk
);
938 domain_merge(&domlist
, pol2
, st
? st
->ss
->name
: NULL
);
939 if (domain_test(domlist
, pol
, st
? st
->ss
->name
: NULL
) == 0)
940 /* new device is incompatible with this device. */
943 domain_free(domlist
);
946 asprintf(&devname
, "/dev/disk/by-path/%s", de
->d_name
);
947 fd
= open(devname
, O_RDONLY
);
950 if (get_dev_size(fd
, devname
, &devsectors
) == 0)
957 st2
= guess_super_type(fd
, guess_partitions
);
959 st2
->ss
->load_super(st2
, fd
, NULL
) < 0)
963 /* Check domain policy again, this time referring to metadata */
964 domain_merge(&domlist
, pol2
, st2
->ss
->name
);
965 if (domain_test(domlist
, pol
, st2
->ss
->name
) == 0)
966 /* Incompatible devices for this metadata type */
970 st2
->ss
->getinfo_super(st2
, &info
, NULL
);
971 if (info
.component_size
> devsectors
)
972 /* This partitioning doesn't fit in the device */
975 /* This is an acceptable device to copy partition
976 * metadata from. We could just stop here, but I
977 * think I want to keep looking incase a larger
978 * metadata which makes better use of the device can
981 if (chosen
== NULL
||
982 chosen_size
< info
.component_size
) {
983 chosen_size
= info
.component_size
;
988 chosen_st
->ss
->free_super(chosen_st
);
997 domain_free(domlist
);
998 dev_policy_free(pol2
);
1000 st2
->ss
->free_super(st2
);
1010 /* 'chosen' is the best device we can find. Let's write its
1011 * metadata to devname dfd is read-only so don't use that
1013 fd
= open(devname
, O_RDWR
);
1015 chosen_st
->ss
->store_super(chosen_st
, fd
);
1019 chosen_st
->ss
->free_super(chosen_st
);
1025 /* adding a spare to a regular array is quite different from adding one to
1026 * a set-of-partitions virtual array.
1027 * This function determines which is worth trying and tries as appropriate.
1028 * Arrays are given priority over partitions.
1030 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1031 struct supertype
*st
, int verbose
)
1036 int partitions_ok
= 0;
1037 char bufpad
[4096 + 4096];
1038 char *buf
= (char*)(((long)bufpad
+ 4096) & ~4095);
1041 /* Can only add a spare if device has at least one domains */
1042 if (pol_find(pol
, pol_domain
) == NULL
)
1044 /* And only if some action allows spares */
1045 if (!policy_action_allows(pol
, st
?st
->ss
->name
:NULL
, act_spare
))
1048 /* Now check if the device is bare - we don't add non-bare devices
1049 * yet even if action=-spare
1052 if (lseek(dfd
, 0, SEEK_SET
) != 0 ||
1053 read(dfd
, buf
, 4096) != 4096) {
1056 fprintf(stderr
, Name
": %s is not bare, so not considering as a spare\n",
1060 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1062 if (memcmp(buf
, buf
+1, 4095) != 0)
1065 /* OK, first 4K appear blank, try the end. */
1066 if (lseek(dfd
, -4096, SEEK_END
) < 0 ||
1067 read(dfd
, buf
, 4096) != 4096)
1070 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1072 if (memcmp(buf
, buf
+1, 4095) != 0)
1075 /* This device passes our test for 'is bare'.
1076 * Let's see what policy allows for such things.
1079 /* just try try 'array' or 'partition' based on this metadata */
1080 if (st
->ss
->add_to_super
)
1081 return array_try_spare(devname
, dfdp
, pol
,
1084 return partition_try_spare(devname
, dfdp
, pol
,
1087 /* Now see which metadata type support spare */
1088 for (i
= 0; (!arrays_ok
|| !partitions_ok
) && superlist
[i
] ; i
++) {
1089 if (superlist
[i
]->add_to_super
&& !arrays_ok
&&
1090 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1092 if (superlist
[i
]->add_to_super
== NULL
&& !partitions_ok
&&
1093 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1098 rv
= array_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1099 if (rv
== 0 && partitions_ok
)
1100 rv
= partition_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1104 int IncrementalScan(int verbose
)
1106 /* look at every device listed in the 'map' file.
1107 * If one is found that is not running then:
1108 * look in mdadm.conf for bitmap file.
1109 * if one exists, but array has none, add it.
1110 * try to start array in auto-readonly mode
1112 struct map_ent
*mapl
= NULL
;
1114 mddev_ident_t devs
, mddev
;
1118 devs
= conf_get_ident(NULL
);
1120 for (me
= mapl
; me
; me
= me
->next
) {
1121 mdu_array_info_t array
;
1122 mdu_bitmap_file_t bmf
;
1124 int mdfd
= open_dev(me
->devnum
);
1128 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
1133 /* Ok, we can try this one. Maybe it needs a bitmap */
1134 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
1135 if (mddev
->devname
&& me
->path
1136 && devname_matches(mddev
->devname
, me
->path
))
1138 if (mddev
&& mddev
->bitmap_file
) {
1140 * Note: early kernels will wrongly fail this, so it
1144 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
1145 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
1147 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
1154 fprintf(stderr
, Name
1155 ": Added bitmap %s to %s\n",
1156 mddev
->bitmap_file
, me
->path
);
1157 else if (errno
!= EEXIST
)
1158 fprintf(stderr
, Name
1159 ": Failed to add bitmap to %s: %s\n",
1160 me
->path
, strerror(errno
));
1163 sra
= sysfs_read(mdfd
, 0, 0);
1165 if (sysfs_set_str(sra
, NULL
,
1166 "array_state", "read-auto") == 0) {
1168 fprintf(stderr
, Name
1169 ": started array %s\n",
1170 me
->path
?: devnum2devname(me
->devnum
));
1172 fprintf(stderr
, Name
1173 ": failed to start array %s: %s\n",
1174 me
->path
?: devnum2devname(me
->devnum
),
1183 static char *container2devname(char *devname
)
1185 char *mdname
= NULL
;
1187 if (devname
[0] == '/') {
1188 int fd
= open(devname
, O_RDONLY
);
1190 mdname
= devnum2devname(fd2devnum(fd
));
1195 struct map_ent
*mp
, *map
= NULL
;
1197 if (!parse_uuid(devname
, uuid
))
1199 mp
= map_by_uuid(&map
, uuid
);
1201 mdname
= devnum2devname(mp
->devnum
);
1208 static int Incremental_container(struct supertype
*st
, char *devname
, int verbose
,
1209 int runstop
, int autof
, int trustworthy
)
1211 /* Collect the contents of this container and for each
1212 * array, choose a device name and assemble the array.
1215 struct mdinfo
*list
= st
->ss
->container_content(st
, NULL
);
1217 struct map_ent
*map
= NULL
;
1220 fprintf(stderr
, Name
": failed to get exclusive lock on "
1223 for (ra
= list
; ra
; ra
= ra
->next
) {
1225 char chosen_name
[1024];
1227 struct mddev_ident_s
*match
= NULL
;
1229 mp
= map_by_uuid(&map
, ra
->uuid
);
1232 mdfd
= open_dev(mp
->devnum
);
1234 strcpy(chosen_name
, mp
->path
);
1236 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
1239 /* Check in mdadm.conf for container == devname and
1240 * member == ra->text_version after second slash.
1242 char *sub
= strchr(ra
->text_version
+1, '/');
1243 struct mddev_ident_s
*array_list
;
1246 array_list
= conf_get_ident(NULL
);
1249 for(; array_list
; array_list
= array_list
->next
) {
1251 if (array_list
->member
== NULL
||
1252 array_list
->container
== NULL
)
1254 if (strcmp(array_list
->member
, sub
) != 0)
1256 if (array_list
->uuid_set
&&
1257 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
1259 dn
= container2devname(array_list
->container
);
1262 if (strncmp(dn
, ra
->text_version
+1,
1264 ra
->text_version
[strlen(dn
)+1] != '/') {
1269 /* we have a match */
1272 fprintf(stderr
, Name
": match found for member %s\n",
1273 array_list
->member
);
1277 if (match
&& match
->devname
&&
1278 strcasecmp(match
->devname
, "<ignore>") == 0) {
1280 fprintf(stderr
, Name
": array %s/%s is "
1281 "explicitly ignored by mdadm.conf\n",
1282 match
->container
, match
->member
);
1286 trustworthy
= LOCAL
;
1288 mdfd
= create_mddev(match
? match
->devname
: NULL
,
1296 fprintf(stderr
, Name
": failed to open %s: %s.\n",
1297 chosen_name
, strerror(errno
));
1301 assemble_container_content(st
, mdfd
, ra
, runstop
,
1302 chosen_name
, verbose
);
1309 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1310 * raid arrays, and if so first fail (if needed) and then remove the device.
1312 * @devname - The device we want to remove
1314 * Note: the device name must be a kernel name like "sda", so
1315 * that we can find it in /proc/mdstat
1317 int IncrementalRemove(char *devname
, int verbose
)
1321 struct mdstat_ent
*ent
;
1322 struct mddev_dev_s devlist
;
1324 if (strchr(devname
, '/')) {
1325 fprintf(stderr
, Name
": incremental removal requires a "
1326 "kernel device name, not a file: %s\n", devname
);
1329 ent
= mdstat_by_component(devname
);
1331 fprintf(stderr
, Name
": %s does not appear to be a component "
1332 "of any array\n", devname
);
1335 mdfd
= open_dev(ent
->devnum
);
1337 fprintf(stderr
, Name
": Cannot open array %s!!\n", ent
->dev
);
1340 memset(&devlist
, 0, sizeof(devlist
));
1341 devlist
.devname
= devname
;
1342 devlist
.disposition
= 'f';
1343 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);
1344 devlist
.disposition
= 'r';
1345 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0);