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
, struct mdinfo
*sra
,
36 int mdfd
, char **availp
,
38 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
39 int number
, __u64 events
, int verbose
,
41 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
42 struct map_ent
*target
,
43 struct supertype
*st
, int verbose
);
45 static int Incremental_container(struct supertype
*st
, char *devname
,
47 int verbose
, int runstop
, int autof
,
50 int Incremental(char *devname
, int verbose
, int runstop
,
51 struct supertype
*st
, char *homehost
, int require_homehost
,
52 int autof
, int freeze_reshape
)
54 /* Add this device to an array, creating the array if necessary
55 * and starting the array if sensible or - if runstop>0 - if possible.
57 * This has several steps:
59 * 1/ Check if device is permitted by mdadm.conf, reject if not.
60 * 2/ Find metadata, reject if none appropriate (check
61 * version/name from args)
62 * 3/ Check if there is a match in mdadm.conf
63 * 3a/ if not, check for homehost match. If no match, assemble as
65 * 4/ Determine device number.
66 * - If in mdadm.conf with std name, use that
67 * - UUID in /var/run/mdadm.map use that
68 * - If name is suggestive, use that. unless in use with different uuid.
69 * - Choose a free, high number.
70 * - Use a partitioned device unless strong suggestion not to.
72 * Don't choose partitioned for containers.
73 * 5/ Find out if array already exists
75 * - choose a name, from mdadm.conf or 'name' field in array.
79 * - check one drive in array to make sure metadata is a reasonably
80 * close match. Reject if not (e.g. different type)
82 * 6/ Make sure /var/run/mdadm.map contains this array.
83 * 7/ Is there enough devices to possibly start the array?
84 * For a container, this means running Incremental_container.
85 * 7a/ if not, finish with success.
87 * - read all metadata and arrange devices like -A does
88 * - if number of OK devices match expected, or -R and there are enough,
89 * start the array (auto-readonly).
92 struct mdinfo info
, dinfo
;
93 struct mdinfo
*sra
= NULL
, *d
;
94 struct mddev_ident
*match
;
95 char chosen_name
[1024];
97 struct map_ent
*mp
, *map
= NULL
;
98 int dfd
= -1, mdfd
= -1;
103 mdu_array_info_t ainf
;
104 struct dev_policy
*policy
= NULL
;
105 struct map_ent target_array
;
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 (must_be_container(dfd
)) {
132 st
= super_by_fd(dfd
, NULL
);
133 if (st
&& st
->ss
->load_container
)
134 rv
= st
->ss
->load_container(st
, dfd
, NULL
);
137 if (!rv
&& st
->ss
->container_content
) {
139 fprintf(stderr
, Name
": failed to get "
140 "exclusive lock on mapfile\n");
141 rv
= Incremental_container(st
, devname
, homehost
,
142 verbose
, runstop
, autof
,
148 fprintf(stderr
, Name
": %s is not part of an md array.\n",
153 /* 1/ Check if device is permitted by mdadm.conf */
155 if (!conf_test_dev(devname
)) {
158 ": %s not permitted by mdadm.conf.\n",
163 /* 2/ Find metadata, reject if none appropriate (check
164 * version/name from args) */
166 if (fstat(dfd
, &stb
) < 0) {
168 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
169 devname
, strerror(errno
));
172 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
174 fprintf(stderr
, Name
": %s is not a block device.\n",
179 dinfo
.disk
.major
= major(stb
.st_rdev
);
180 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
182 policy
= disk_policy(&dinfo
);
183 have_target
= policy_check_path(&dinfo
, &target_array
);
185 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
188 ": no recognisable superblock on %s.\n",
190 rv
= try_spare(devname
, &dfd
, policy
,
191 have_target
? &target_array
: NULL
,
195 if (st
->ss
->compare_super
== NULL
||
196 st
->ss
->load_super(st
, dfd
, NULL
)) {
198 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
200 rv
= try_spare(devname
, &dfd
, policy
,
201 have_target
? &target_array
: NULL
,
206 close (dfd
); dfd
= -1;
208 st
->ss
->getinfo_super(st
, &info
, NULL
);
210 /* 3/ Check if there is a match in mdadm.conf */
211 match
= conf_match(st
, &info
, devname
, verbose
, &rv
);
212 if (!match
&& rv
== 2)
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
, policy
,
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 name_to_use
= info
.name
;
261 if (name_to_use
[0] == 0 &&
262 info
.array
.level
== LEVEL_CONTAINER
&&
263 trustworthy
== LOCAL
) {
264 name_to_use
= info
.text_version
;
265 trustworthy
= METADATA
;
267 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
268 ! require_homehost
&&
269 conf_name_is_free(name_to_use
))
272 /* strip "hostname:" prefix from name if we have decided
273 * to treat it as LOCAL
275 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
276 name_to_use
= strchr(name_to_use
, ':')+1;
278 /* 4/ Check if array exists.
281 fprintf(stderr
, Name
": failed to get exclusive lock on "
283 mp
= map_by_uuid(&map
, info
.uuid
);
285 mdfd
= open_dev(mp
->devnum
);
291 /* Couldn't find an existing array, maybe make a new one */
292 mdfd
= create_mddev(match
? match
->devname
: NULL
,
293 name_to_use
, autof
, trustworthy
, chosen_name
);
298 sysfs_init(&info
, mdfd
, 0);
300 if (set_array_info(mdfd
, st
, &info
) != 0) {
301 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
302 chosen_name
, strerror(errno
));
308 dinfo
.disk
.major
= major(stb
.st_rdev
);
309 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
310 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
311 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
312 devname
, chosen_name
, strerror(errno
));
313 ioctl(mdfd
, STOP_ARRAY
, 0);
317 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
318 GET_OFFSET
| GET_SIZE
));
320 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
321 /* It really should be 'none' - must be old buggy
322 * kernel, and mdadm -I may not be able to complete.
325 ioctl(mdfd
, STOP_ARRAY
, NULL
);
327 ": You have an old buggy kernel which cannot support\n"
328 " --incremental reliably. Aborting.\n");
333 info
.array
.working_disks
= 1;
334 /* 6/ Make sure /var/run/mdadm.map contains this array. */
335 map_update(&map
, fd2devnum(mdfd
),
337 info
.uuid
, chosen_name
);
340 /* - check one drive in array to make sure metadata is a reasonably */
341 /* close match. Reject if not (e.g. different type) */
342 /* - add the device */
346 struct supertype
*st2
;
347 struct mdinfo info2
, *d
;
349 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
350 GET_OFFSET
| GET_SIZE
));
353 strcpy(chosen_name
, mp
->path
);
355 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
357 /* It is generally not OK to add non-spare drives to a
358 * running array as they are probably missing because
359 * they failed. However if runstop is 1, then the
360 * array was possibly started early and our best bet is
361 * to add this anyway.
362 * Also if action policy is re-add or better we allow
364 * This doesn't apply to containers as the 'non-spare'
365 * flag has a different meaning. The test has to happen
366 * at the device level there
368 if (!st
->ss
->external
369 && (info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
370 && ! policy_action_allows(policy
, st
->ss
->name
,
373 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
375 ": not adding %s to active array (without --run) %s\n",
376 devname
, chosen_name
);
386 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
387 sra
->devs
->disk
.minor
);
388 dfd2
= dev_open(dn
, O_RDONLY
);
391 ": unable to open %s\n", devname
);
396 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
397 st
->ss
->compare_super(st
, st2
) != 0) {
399 ": metadata mismatch between %s and "
401 devname
, chosen_name
);
407 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
408 st2
->ss
->free_super(st2
);
409 if (info
.array
.level
!= info2
.array
.level
||
410 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
411 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
413 ": unexpected difference between %s and %s.\n",
414 chosen_name
, devname
);
419 info2
.disk
.major
= major(stb
.st_rdev
);
420 info2
.disk
.minor
= minor(stb
.st_rdev
);
421 /* add disk needs to know about containers */
422 if (st
->ss
->external
)
423 sra
->array
.level
= LEVEL_CONTAINER
;
424 err
= add_disk(mdfd
, st
, sra
, &info2
);
425 if (err
< 0 && errno
== EBUSY
) {
426 /* could be another device present with the same
427 * disk.number. Find and reject any such
429 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
430 info
.events
, verbose
, chosen_name
);
431 err
= add_disk(mdfd
, st
, sra
, &info2
);
434 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
435 devname
, chosen_name
, strerror(errno
));
439 info
.array
.working_disks
= 0;
440 for (d
= sra
->devs
; d
; d
=d
->next
)
441 info
.array
.working_disks
++;
445 /* 7/ Is there enough devices to possibly start the array? */
446 /* 7a/ if not, finish with success. */
447 if (info
.array
.level
== LEVEL_CONTAINER
) {
448 int devnum
= devnum
; /* defined and used iff ->external */
449 /* Try to assemble within the container */
450 sysfs_uevent(sra
, "change");
453 ": container %s now has %d devices\n",
454 chosen_name
, info
.array
.working_disks
);
455 wait_for(chosen_name
, mdfd
);
456 if (st
->ss
->external
)
457 devnum
= fd2devnum(mdfd
);
458 if (st
->ss
->load_container
)
459 rv
= st
->ss
->load_container(st
, mdfd
, NULL
);
463 rv
= Incremental_container(st
, chosen_name
, homehost
,
464 verbose
, runstop
, autof
,
468 /* Don't fail the whole -I if a subarray didn't
469 * have enough devices to start yet
472 /* after spare is added, ping monitor for external metadata
473 * so that it can eg. try to rebuild degraded array */
474 if (st
->ss
->external
)
475 ping_monitor_by_id(devnum
);
479 /* We have added something to the array, so need to re-read the
480 * state. Eventually this state should be kept up-to-date as
484 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
485 GET_OFFSET
| GET_SIZE
));
486 active_disks
= count_active(st
, sra
, mdfd
, &avail
, &info
);
487 if (enough(info
.array
.level
, info
.array
.raid_disks
,
488 info
.array
.layout
, info
.array
.state
& 1,
489 avail
, active_disks
) == 0) {
492 ": %s attached to %s, not enough to start (%d).\n",
493 devname
, chosen_name
, active_disks
);
499 /* - if number of OK devices match expected, or -R and there */
501 /* + add any bitmap file */
502 /* + start the array (auto-readonly). */
504 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
507 ": %s attached to %s which is already active.\n",
508 devname
, chosen_name
);
514 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
516 /* Let's try to start it */
517 if (match
&& match
->bitmap_file
) {
518 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
521 ": Could not open bitmap file %s.\n",
525 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
528 ": Failed to set bitmapfile for %s.\n",
534 /* Need to remove from the array any devices which
535 * 'count_active' discerned were too old or inappropriate
537 for (d
= sra
? sra
->devs
: NULL
; d
; d
= d
->next
)
538 if (d
->disk
.state
& (1<<MD_DISK_REMOVED
))
539 remove_disk(mdfd
, st
, sra
, d
);
541 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
542 && trustworthy
!= FOREIGN
)
543 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
545 rv
= sysfs_set_str(sra
, NULL
,
546 "array_state", "read-auto");
550 ": %s attached to %s, which has been started.\n",
551 devname
, chosen_name
);
553 wait_for(chosen_name
, mdfd
);
554 /* We just started the array, so some devices
555 * might have been evicted from the array
556 * because their event counts were too old.
557 * If the action=re-add policy is in-force for
558 * those devices we should re-add them now.
560 for (dsk
= sra
->devs
; dsk
; dsk
= dsk
->next
) {
561 if (disk_action_allows(dsk
, st
->ss
->name
, act_re_add
) &&
562 add_disk(mdfd
, st
, sra
, dsk
) == 0)
564 ": %s re-added to %s\n",
565 dsk
->sys_name
, chosen_name
);
569 ": %s attached to %s, but failed to start: %s.\n",
570 devname
, chosen_name
, strerror(errno
));
576 ": %s attached to %s, not enough to start safely.\n",
577 devname
, chosen_name
);
587 dev_policy_free(policy
);
596 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
597 int number
, __u64 events
, int verbose
,
600 /* Find a device attached to this array with a disk.number of number
601 * and events less than the passed events, and remove the device.
606 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
607 return; /* not safe to remove from active arrays
608 * without thinking more */
610 for (d
= sra
->devs
; d
; d
= d
->next
) {
614 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
615 dfd
= dev_open(dn
, O_RDONLY
);
618 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
622 st
->ss
->getinfo_super(st
, &info
, NULL
);
623 st
->ss
->free_super(st
);
626 if (info
.disk
.number
!= number
||
627 info
.events
>= events
)
630 if (d
->disk
.raid_disk
> -1)
631 sysfs_set_str(sra
, d
, "slot", "none");
632 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
635 ": removing old device %s from %s\n",
636 d
->sys_name
+4, array_name
);
640 static int count_active(struct supertype
*st
, struct mdinfo
*sra
,
641 int mdfd
, char **availp
,
642 struct mdinfo
*bestinfo
)
644 /* count how many devices in sra think they are active */
647 __u64 max_events
= 0;
659 for (d
= sra
->devs
; d
; d
= d
->next
)
661 for (d
= sra
->devs
, devnum
=0 ; d
; d
= d
->next
, devnum
++) {
667 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
668 dfd
= dev_open(dn
, O_RDONLY
);
671 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
675 info
.array
.raid_disks
= raid_disks
;
676 st
->ss
->getinfo_super(st
, &info
, devmap
+ raid_disks
* devnum
);
678 raid_disks
= info
.array
.raid_disks
;
679 avail
= calloc(raid_disks
, 1);
681 fprintf(stderr
, Name
": out of memory.\n");
686 best
= calloc(raid_disks
, sizeof(int));
687 devmap
= calloc(raid_disks
* numdevs
, 1);
689 st
->ss
->getinfo_super(st
, &info
, devmap
);
692 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
696 max_events
= info
.events
;
697 avail
[info
.disk
.raid_disk
] = 2;
698 best
[info
.disk
.raid_disk
] = devnum
;
699 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
700 } else if (info
.events
== max_events
) {
701 avail
[info
.disk
.raid_disk
] = 2;
702 best
[info
.disk
.raid_disk
] = devnum
;
703 } else if (info
.events
== max_events
-1) {
704 if (avail
[info
.disk
.raid_disk
] == 0) {
705 avail
[info
.disk
.raid_disk
] = 1;
706 best
[info
.disk
.raid_disk
] = devnum
;
708 } else if (info
.events
< max_events
- 1)
710 else if (info
.events
== max_events
+1) {
712 max_events
= info
.events
;
713 for (i
=0; i
< raid_disks
; i
++)
716 avail
[info
.disk
.raid_disk
] = 2;
717 best
[info
.disk
.raid_disk
] = devnum
;
718 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
719 } else { /* info.events much bigger */
720 memset(avail
, 0, raid_disks
);
721 max_events
= info
.events
;
722 avail
[info
.disk
.raid_disk
] = 2;
723 best
[info
.disk
.raid_disk
] = devnum
;
724 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
727 st
->ss
->free_super(st
);
731 /* We need to reject any device that thinks the best device is
732 * failed or missing */
733 for (b
= 0; b
< raid_disks
; b
++)
737 for (i
= 0 ; i
< raid_disks
; i
++) {
738 if (i
!= b
&& avail
[i
])
739 if (devmap
[raid_disks
* best
[i
] + b
] == 0) {
740 /* This device thinks 'b' is failed -
743 for (d
=sra
->devs
; devnum
; d
= d
->next
)
745 d
->disk
.state
|= (1 << MD_DISK_REMOVED
);
756 /* test if container has degraded member(s) */
757 static int container_members_max_degradation(struct map_ent
*map
, struct map_ent
*me
)
759 mdu_array_info_t array
;
761 int max_degraded
= 0;
763 for(; map
; map
= map
->next
) {
764 if (!is_subarray(map
->metadata
) ||
765 devname2devnum(map
->metadata
+1) != me
->devnum
)
767 afd
= open_dev(map
->devnum
);
770 /* most accurate information regarding array degradation */
771 if (ioctl(afd
, GET_ARRAY_INFO
, &array
) >= 0) {
772 int degraded
= array
.raid_disks
- array
.active_disks
-
774 if (degraded
> max_degraded
)
775 max_degraded
= degraded
;
779 return (max_degraded
);
782 static int array_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
783 struct map_ent
*target
, int bare
,
784 struct supertype
*st
, int verbose
)
786 /* This device doesn't have any md metadata
787 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
788 * If 'st' is not set, then we only know that some metadata allows this,
789 * others possibly don't.
790 * So look for a container or array to attach the device to.
791 * Prefer 'target' if that is set and the array is found.
793 * If st is set, then only arrays of that type are considered
794 * Return 0 on success, or some exit code on failure, probably 1.
798 struct map_ent
*mp
, *map
= NULL
;
799 struct mdinfo
*chosen
= NULL
;
802 if (fstat(dfd
, &stb
) != 0)
806 * Now we need to find a suitable array to add this to.
807 * We only accept arrays that:
809 * - are in the same domains as the device
810 * - are of an size for which the device will be useful
811 * and we choose the one that is the most degraded
814 if (map_lock(&map
)) {
815 fprintf(stderr
, Name
": failed to get exclusive lock on "
819 for (mp
= map
; mp
; mp
= mp
->next
) {
820 struct supertype
*st2
;
821 struct domainlist
*dl
= NULL
;
823 unsigned long long devsize
;
824 unsigned long long component_size
= 0;
826 if (is_subarray(mp
->metadata
))
829 st2
= st
->ss
->match_metadata_desc(mp
->metadata
);
831 (st
->minor_version
>= 0 &&
832 st
->minor_version
!= st2
->minor_version
)) {
834 fprintf(stderr
, Name
": not adding %s to %s as metadata type doesn't match\n",
841 sra
= sysfs_read(-1, mp
->devnum
,
842 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
843 GET_DEGRADED
|GET_COMPONENT
|GET_VERSION
);
845 /* Probably a container - no degraded info */
846 sra
= sysfs_read(-1, mp
->devnum
,
847 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
848 GET_COMPONENT
|GET_VERSION
);
850 sra
->array
.failed_disks
= -1;
857 for(i
=0; !st2
&& superlist
[i
]; i
++)
858 st2
= superlist
[i
]->match_metadata_desc(
862 fprintf(stderr
, Name
": not adding %s to %s"
863 " as metadata not recognised.\n",
867 /* Need to double check the 'act_spare' permissions applies
870 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
872 if (!bare
&& !policy_action_allows(pol
, st2
->ss
->name
,
873 act_spare_same_slot
))
877 /* update number of failed disks for mostly degraded
878 * container member */
879 if (sra
->array
.failed_disks
== -1)
880 sra
->array
.failed_disks
= container_members_max_degradation(map
, mp
);
882 get_dev_size(dfd
, NULL
, &devsize
);
883 if (sra
->component_size
== 0) {
884 /* true for containers, here we must read superblock
885 * to obtain minimum spare size */
886 struct supertype
*st3
= dup_super(st2
);
887 int mdfd
= open_dev(mp
->devnum
);
892 if (st3
->ss
->load_container
&&
893 !st3
->ss
->load_container(st3
, mdfd
, mp
->path
)) {
894 component_size
= st3
->ss
->min_acceptable_spare_size(st3
);
895 st3
->ss
->free_super(st3
);
900 if ((sra
->component_size
> 0 &&
901 st2
->ss
->avail_size(st2
, devsize
) < sra
->component_size
)
903 (sra
->component_size
== 0 && devsize
< component_size
)) {
905 fprintf(stderr
, Name
": not adding %s to %s as it is too small\n",
909 /* test against target.
910 * If 'target' is set and 'bare' is false, we only accept
911 * arrays/containers that match 'target'.
912 * If 'target' is set and 'bare' is true, we prefer the
913 * array which matches 'target'.
914 * target is considered only if we deal with degraded array
916 if (target
&& policy_action_allows(pol
, st2
->ss
->name
,
917 act_spare_same_slot
)) {
918 if (strcmp(target
->metadata
, mp
->metadata
) == 0 &&
919 memcmp(target
->uuid
, mp
->uuid
,
920 sizeof(target
->uuid
)) == 0 &&
921 sra
->array
.failed_disks
> 0) {
922 /* This is our target!! */
927 /* skip to end so we don't check any more */
937 dl
= domain_from_array(sra
, st2
->ss
->name
);
938 if (domain_test(dl
, pol
, st2
->ss
->name
) != 1) {
939 /* domain test fails */
941 fprintf(stderr
, Name
": not adding %s to %s as"
942 " it is not in a compatible domain\n",
947 /* all tests passed, OK to add to this array */
951 } else if (chosen
->array
.failed_disks
< sra
->array
.failed_disks
) {
965 /* add current device to chosen array as a spare */
966 int mdfd
= open_dev(devname2devnum(chosen
->sys_name
));
968 struct mddev_dev devlist
;
973 devlist
.writemostly
= 0;
974 devlist
.devname
= devname
;
975 sprintf(devname
, "%d:%d", major(stb
.st_rdev
),
977 devlist
.disposition
= 'a';
980 rv
= Manage_subdevs(chosen
->sys_name
, mdfd
, &devlist
,
986 fprintf(stderr
, Name
": added %s as spare for %s\n",
987 devname
, chosen
->sys_name
);
989 fprintf(stderr
, Name
": failed to add %s as spare for %s\n",
990 devname
, chosen
->sys_name
);
998 static int partition_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
999 struct supertype
*st
, int verbose
)
1001 /* we know that at least one partition virtual-metadata is
1002 * allowed to incorporate spares like this device. We need to
1003 * find a suitable device to copy partition information from.
1005 * Getting a list of all disk (not partition) devices is
1006 * slightly non-trivial. We could look at /sys/block, but
1007 * that is theoretically due to be removed. Maybe best to use
1008 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1009 * we depend on this directory of 'path' info. But that fails
1010 * to find loop devices and probably others. Maybe don't
1011 * worry about that, they aren't the real target.
1013 * So: check things in /dev/disk/by-path to see if they are in
1014 * a compatible domain, then load the partition table and see
1015 * if it is OK for the new device, and choose the largest
1016 * partition table that fits.
1020 char *chosen
= NULL
;
1021 unsigned long long chosen_size
= 0;
1022 struct supertype
*chosen_st
= NULL
;
1025 dir
= opendir("/dev/disk/by-path");
1028 while ((de
= readdir(dir
)) != NULL
) {
1030 struct dev_policy
*pol2
= NULL
;
1031 struct domainlist
*domlist
= NULL
;
1034 struct supertype
*st2
= NULL
;
1035 char *devname
= NULL
;
1036 unsigned long long devsectors
;
1038 if (de
->d_ino
== 0 ||
1039 de
->d_name
[0] == '.' ||
1040 (de
->d_type
!= DT_LNK
&& de
->d_type
!= DT_UNKNOWN
))
1043 ep
= de
->d_name
+ strlen(de
->d_name
);
1044 while (ep
> de
->d_name
&&
1047 if (ep
> de
->d_name
+ 5 &&
1048 strncmp(ep
-5, "-part", 5) == 0)
1049 /* This is a partition - skip it */
1052 pol2
= path_policy(de
->d_name
, type_disk
);
1054 domain_merge(&domlist
, pol2
, st
? st
->ss
->name
: NULL
);
1055 if (domain_test(domlist
, pol
, st
? st
->ss
->name
: NULL
) != 1)
1056 /* new device is incompatible with this device. */
1059 domain_free(domlist
);
1062 if (asprintf(&devname
, "/dev/disk/by-path/%s", de
->d_name
) != 1) {
1066 fd
= open(devname
, O_RDONLY
);
1069 if (get_dev_size(fd
, devname
, &devsectors
) == 0)
1074 st2
= dup_super(st
);
1076 st2
= guess_super_type(fd
, guess_partitions
);
1078 st2
->ss
->load_super(st2
, fd
, NULL
) < 0)
1082 /* Check domain policy again, this time referring to metadata */
1083 domain_merge(&domlist
, pol2
, st2
->ss
->name
);
1084 if (domain_test(domlist
, pol
, st2
->ss
->name
) != 1)
1085 /* Incompatible devices for this metadata type */
1087 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
1088 /* Some partition types allow sparing, but not
1094 st2
->ss
->getinfo_super(st2
, &info
, NULL
);
1095 if (info
.component_size
> devsectors
)
1096 /* This partitioning doesn't fit in the device */
1099 /* This is an acceptable device to copy partition
1100 * metadata from. We could just stop here, but I
1101 * think I want to keep looking incase a larger
1102 * metadata which makes better use of the device can
1105 if (chosen
== NULL
||
1106 chosen_size
< info
.component_size
) {
1107 chosen_size
= info
.component_size
;
1112 chosen_st
->ss
->free_super(chosen_st
);
1121 domain_free(domlist
);
1122 dev_policy_free(pol2
);
1124 st2
->ss
->free_super(st2
);
1136 /* 'chosen' is the best device we can find. Let's write its
1137 * metadata to devname dfd is read-only so don't use that
1139 fd
= open(devname
, O_RDWR
);
1141 chosen_st
->ss
->store_super(chosen_st
, fd
);
1145 chosen_st
->ss
->free_super(chosen_st
);
1150 static int is_bare(int dfd
)
1152 unsigned long long size
= 0;
1153 char bufpad
[4096 + 4096];
1154 char *buf
= (char*)(((long)bufpad
+ 4096) & ~4095);
1156 if (lseek(dfd
, 0, SEEK_SET
) != 0 ||
1157 read(dfd
, buf
, 4096) != 4096)
1160 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1162 if (memcmp(buf
, buf
+1, 4095) != 0)
1165 /* OK, first 4K appear blank, try the end. */
1166 get_dev_size(dfd
, NULL
, &size
);
1167 if (lseek(dfd
, size
-4096, SEEK_SET
) < 0 ||
1168 read(dfd
, buf
, 4096) != 4096)
1171 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1173 if (memcmp(buf
, buf
+1, 4095) != 0)
1179 /* adding a spare to a regular array is quite different from adding one to
1180 * a set-of-partitions virtual array.
1181 * This function determines which is worth trying and tries as appropriate.
1182 * Arrays are given priority over partitions.
1184 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1185 struct map_ent
*target
,
1186 struct supertype
*st
, int verbose
)
1191 int partitions_ok
= 0;
1195 /* Can only add a spare if device has at least one domain */
1196 if (pol_find(pol
, pol_domain
) == NULL
)
1198 /* And only if some action allows spares */
1199 if (!policy_action_allows(pol
, st
?st
->ss
->name
:NULL
, act_spare
))
1202 /* Now check if the device is bare.
1203 * bare devices can always be added as a spare
1204 * non-bare devices can only be added if spare-same-slot is permitted,
1205 * and this device is replacing a previous device - in which case 'target'
1208 if (!is_bare(dfd
)) {
1209 /* Must have a target and allow same_slot */
1210 /* Later - may allow force_spare without target */
1212 !policy_action_allows(pol
, st
?st
->ss
->name
:NULL
,
1213 act_spare_same_slot
)) {
1215 fprintf(stderr
, Name
": %s is not bare, so not "
1216 "considering as a spare\n",
1224 /* It might be OK to add this device to an array - need to see
1225 * what arrays might be candidates.
1228 /* just try try 'array' or 'partition' based on this metadata */
1229 if (st
->ss
->add_to_super
)
1230 return array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1233 return partition_try_spare(devname
, dfdp
, pol
,
1236 /* No metadata was specified or found so options are open.
1237 * Check for whether any array metadata, or any partition metadata
1238 * might allow adding the spare. This check is just help to avoid
1239 * a more costly scan of all arrays when we can be sure that will
1242 for (i
= 0; (!arrays_ok
|| !partitions_ok
) && superlist
[i
] ; i
++) {
1243 if (superlist
[i
]->add_to_super
&& !arrays_ok
&&
1244 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1246 if (superlist
[i
]->add_to_super
== NULL
&& !partitions_ok
&&
1247 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1252 rv
= array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1254 if (rv
!= 0 && partitions_ok
)
1255 rv
= partition_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1259 int IncrementalScan(int verbose
)
1261 /* look at every device listed in the 'map' file.
1262 * If one is found that is not running then:
1263 * look in mdadm.conf for bitmap file.
1264 * if one exists, but array has none, add it.
1265 * try to start array in auto-readonly mode
1267 struct map_ent
*mapl
= NULL
;
1269 struct mddev_ident
*devs
, *mddev
;
1273 devs
= conf_get_ident(NULL
);
1275 for (me
= mapl
; me
; me
= me
->next
) {
1276 mdu_array_info_t array
;
1277 mdu_bitmap_file_t bmf
;
1279 int mdfd
= open_dev(me
->devnum
);
1283 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
1288 /* Ok, we can try this one. Maybe it needs a bitmap */
1289 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
1290 if (mddev
->devname
&& me
->path
1291 && devname_matches(mddev
->devname
, me
->path
))
1293 if (mddev
&& mddev
->bitmap_file
) {
1295 * Note: early kernels will wrongly fail this, so it
1299 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
1300 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
1302 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
1309 fprintf(stderr
, Name
1310 ": Added bitmap %s to %s\n",
1311 mddev
->bitmap_file
, me
->path
);
1312 else if (errno
!= EEXIST
)
1313 fprintf(stderr
, Name
1314 ": Failed to add bitmap to %s: %s\n",
1315 me
->path
, strerror(errno
));
1318 sra
= sysfs_read(mdfd
, 0, 0);
1320 if (sysfs_set_str(sra
, NULL
,
1321 "array_state", "read-auto") == 0) {
1323 fprintf(stderr
, Name
1324 ": started array %s\n",
1325 me
->path
?: devnum2devname(me
->devnum
));
1327 fprintf(stderr
, Name
1328 ": failed to start array %s: %s\n",
1329 me
->path
?: devnum2devname(me
->devnum
),
1339 static char *container2devname(char *devname
)
1341 char *mdname
= NULL
;
1343 if (devname
[0] == '/') {
1344 int fd
= open(devname
, O_RDONLY
);
1346 mdname
= devnum2devname(fd2devnum(fd
));
1351 struct map_ent
*mp
, *map
= NULL
;
1353 if (!parse_uuid(devname
, uuid
))
1355 mp
= map_by_uuid(&map
, uuid
);
1357 mdname
= devnum2devname(mp
->devnum
);
1364 static int Incremental_container(struct supertype
*st
, char *devname
,
1365 char *homehost
, int verbose
,
1366 int runstop
, int autof
, int freeze_reshape
)
1368 /* Collect the contents of this container and for each
1369 * array, choose a device name and assemble the array.
1372 struct mdinfo
*list
;
1374 struct map_ent
*map
= NULL
;
1377 struct mddev_ident
*match
;
1379 struct domainlist
*domains
;
1380 struct map_ent
*smp
;
1386 st
->ss
->getinfo_super(st
, &info
, NULL
);
1388 if ((runstop
> 0 && info
.container_enough
>= 0) ||
1389 info
.container_enough
> 0)
1393 fprintf(stderr
, Name
": not enough devices to start the container\n");
1397 match
= conf_match(st
, &info
, devname
, verbose
, &rv
);
1398 if (match
== NULL
&& rv
== 2)
1401 /* Need to compute 'trustworthy' */
1403 trustworthy
= LOCAL
;
1404 else if (st
->ss
->match_home(st
, homehost
) == 1)
1405 trustworthy
= LOCAL
;
1406 else if (st
->ss
->match_home(st
, "any") == 1)
1407 trustworthy
= LOCAL
;
1409 trustworthy
= FOREIGN
;
1411 list
= st
->ss
->container_content(st
, NULL
);
1412 /* when nothing to activate - quit */
1415 for (ra
= list
; ra
; ra
= ra
->next
) {
1417 char chosen_name
[1024];
1419 struct mddev_ident
*match
= NULL
;
1422 /* do not activate arrays blocked by metadata handler */
1423 if (ra
->array
.state
& (1 << MD_SB_BLOCK_VOLUME
)) {
1424 fprintf(stderr
, Name
": Cannot activate array %s in %s.\n",
1425 ra
->text_version
, devname
);
1429 mp
= map_by_uuid(&map
, ra
->uuid
);
1432 mdfd
= open_dev(mp
->devnum
);
1434 strcpy(chosen_name
, mp
->path
);
1436 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
1439 /* Check in mdadm.conf for container == devname and
1440 * member == ra->text_version after second slash.
1442 char *sub
= strchr(ra
->text_version
+1, '/');
1443 struct mddev_ident
*array_list
;
1446 array_list
= conf_get_ident(NULL
);
1449 for(; array_list
; array_list
= array_list
->next
) {
1451 if (array_list
->member
== NULL
||
1452 array_list
->container
== NULL
)
1454 if (strcmp(array_list
->member
, sub
) != 0)
1456 if (array_list
->uuid_set
&&
1457 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
1459 dn
= container2devname(array_list
->container
);
1462 if (strncmp(dn
, ra
->text_version
+1,
1464 ra
->text_version
[strlen(dn
)+1] != '/') {
1469 /* we have a match */
1472 fprintf(stderr
, Name
": match found for member %s\n",
1473 array_list
->member
);
1477 if (match
&& match
->devname
&&
1478 strcasecmp(match
->devname
, "<ignore>") == 0) {
1480 fprintf(stderr
, Name
": array %s/%s is "
1481 "explicitly ignored by mdadm.conf\n",
1482 match
->container
, match
->member
);
1486 trustworthy
= LOCAL
;
1488 mdfd
= create_mddev(match
? match
->devname
: NULL
,
1496 fprintf(stderr
, Name
": failed to open %s: %s.\n",
1497 chosen_name
, strerror(errno
));
1501 assemble_container_content(st
, mdfd
, ra
, runstop
,
1502 chosen_name
, verbose
, NULL
,
1507 /* don't move spares to container with volume being activated
1508 when all volumes are blocked */
1509 if (ra_all
== ra_blocked
)
1512 /* Now move all suitable spares from spare container */
1513 domains
= domain_from_array(list
, st
->ss
->name
);
1514 memcpy(suuid
, uuid_zero
, sizeof(int[4]));
1516 (smp
= map_by_uuid(&map
, suuid
)) != NULL
&&
1517 (sfd
= open(smp
->path
, O_RDONLY
)) >= 0) {
1518 /* spare container found */
1519 struct supertype
*sst
=
1520 super_imsm
.match_metadata_desc("imsm");
1521 struct mdinfo
*sinfo
;
1522 unsigned long long min_size
= 0;
1523 if (st
->ss
->min_acceptable_spare_size
)
1524 min_size
= st
->ss
->min_acceptable_spare_size(st
);
1525 if (!sst
->ss
->load_container(sst
, sfd
, NULL
)) {
1527 sinfo
= container_choose_spares(sst
, min_size
,
1530 sst
->ss
->free_super(sst
);
1533 struct mdinfo
*disks
= sinfo
->devs
;
1535 /* move spare from spare
1536 * container to currently
1542 makedev(disks
->disk
.major
,
1543 disks
->disk
.minor
)))
1545 disks
= disks
->next
;
1548 fprintf(stderr
, Name
1549 ": Added %d spare%s to %s\n",
1550 count
, count
>1?"s":"", devname
);
1556 domain_free(domains
);
1561 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1562 * raid arrays, and if so first fail (if needed) and then remove the device.
1564 * @devname - The device we want to remove
1565 * @id_path - name as found in /dev/disk/by-path for this device
1567 * Note: the device name must be a kernel name like "sda", so
1568 * that we can find it in /proc/mdstat
1570 int IncrementalRemove(char *devname
, char *id_path
, int verbose
)
1574 struct mdstat_ent
*ent
;
1575 struct mddev_dev devlist
;
1578 dprintf(Name
": incremental removal without --path <id_path> "
1579 "lacks the possibility to re-add new device in this "
1582 if (strchr(devname
, '/')) {
1583 fprintf(stderr
, Name
": incremental removal requires a "
1584 "kernel device name, not a file: %s\n", devname
);
1587 ent
= mdstat_by_component(devname
);
1589 fprintf(stderr
, Name
": %s does not appear to be a component "
1590 "of any array\n", devname
);
1593 mdfd
= open_dev(ent
->devnum
);
1595 fprintf(stderr
, Name
": Cannot open array %s!!\n", ent
->dev
);
1601 struct map_ent
*map
= NULL
, *me
;
1602 me
= map_by_devnum(&map
, ent
->devnum
);
1604 policy_save_path(id_path
, me
);
1608 memset(&devlist
, 0, sizeof(devlist
));
1609 devlist
.devname
= devname
;
1610 devlist
.disposition
= 'f';
1611 /* for a container, we must fail each member array */
1612 if (ent
->metadata_version
&&
1613 strncmp(ent
->metadata_version
, "external:", 9) == 0) {
1614 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1615 struct mdstat_ent
*memb
;
1616 for (memb
= mdstat
; memb
; memb
= memb
->next
)
1617 if (is_container_member(memb
, ent
->dev
)) {
1618 int subfd
= open_dev(memb
->devnum
);
1620 Manage_subdevs(memb
->dev
, subfd
,
1621 &devlist
, verbose
, 0,
1626 free_mdstat(mdstat
);
1628 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
, 0);
1629 devlist
.disposition
= 'r';
1630 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
, 0);