]>
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
, 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
);
49 static struct mddev_ident
*search_mdstat(struct supertype
*st
,
52 int verbose
, int *rvp
);
54 int Incremental(char *devname
, int verbose
, int runstop
,
55 struct supertype
*st
, char *homehost
, int require_homehost
,
58 /* Add this device to an array, creating the array if necessary
59 * and starting the array if sensible or - if runstop>0 - if possible.
61 * This has several steps:
63 * 1/ Check if device is permitted by mdadm.conf, reject if not.
64 * 2/ Find metadata, reject if none appropriate (check
65 * version/name from args)
66 * 3/ Check if there is a match in mdadm.conf
67 * 3a/ if not, check for homehost match. If no match, assemble as
69 * 4/ Determine device number.
70 * - If in mdadm.conf with std name, use that
71 * - UUID in /var/run/mdadm.map use that
72 * - If name is suggestive, use that. unless in use with different uuid.
73 * - Choose a free, high number.
74 * - Use a partitioned device unless strong suggestion not to.
76 * Don't choose partitioned for containers.
77 * 5/ Find out if array already exists
79 * - choose a name, from mdadm.conf or 'name' field in array.
83 * - check one drive in array to make sure metadata is a reasonably
84 * close match. Reject if not (e.g. different type)
86 * 6/ Make sure /var/run/mdadm.map contains this array.
87 * 7/ Is there enough devices to possibly start the array?
88 * For a container, this means running Incremental_container.
89 * 7a/ if not, finish with success.
91 * - read all metadata and arrange devices like -A does
92 * - if number of OK devices match expected, or -R and there are enough,
93 * start the array (auto-readonly).
96 struct mdinfo info
, dinfo
;
97 struct mdinfo
*sra
= NULL
, *d
;
98 struct mddev_ident
*match
;
99 char chosen_name
[1024];
101 struct map_ent
*mp
, *map
= NULL
;
102 int dfd
= -1, mdfd
= -1;
107 mdu_array_info_t ainf
;
108 struct dev_policy
*policy
= NULL
;
109 struct map_ent target_array
;
112 struct createinfo
*ci
= conf_get_create_info();
114 if (stat(devname
, &stb
) < 0) {
116 fprintf(stderr
, Name
": stat failed for %s: %s.\n",
117 devname
, strerror(errno
));
120 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
122 fprintf(stderr
, Name
": %s is not a block device.\n",
126 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
129 fprintf(stderr
, Name
": cannot open %s: %s.\n",
130 devname
, strerror(errno
));
133 /* If the device is a container, we do something very different */
134 if (must_be_container(dfd
)) {
136 st
= super_by_fd(dfd
, NULL
);
137 if (st
&& st
->ss
->load_container
)
138 rv
= st
->ss
->load_container(st
, dfd
, NULL
);
141 if (!rv
&& st
->ss
->container_content
)
142 return Incremental_container(st
, devname
, homehost
,
143 verbose
, runstop
, autof
);
145 fprintf(stderr
, Name
": %s is not part of an md array.\n",
150 /* 1/ Check if device is permitted by mdadm.conf */
152 if (!conf_test_dev(devname
)) {
155 ": %s not permitted by mdadm.conf.\n",
160 /* 2/ Find metadata, reject if none appropriate (check
161 * version/name from args) */
163 if (fstat(dfd
, &stb
) < 0) {
165 fprintf(stderr
, Name
": fstat failed for %s: %s.\n",
166 devname
, strerror(errno
));
169 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
171 fprintf(stderr
, Name
": %s is not a block device.\n",
176 dinfo
.disk
.major
= major(stb
.st_rdev
);
177 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
179 policy
= disk_policy(&dinfo
);
180 have_target
= policy_check_path(&dinfo
, &target_array
);
182 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
185 ": no recognisable superblock on %s.\n",
187 rv
= try_spare(devname
, &dfd
, policy
,
188 have_target
? &target_array
: NULL
,
192 if (st
->ss
->compare_super
== NULL
||
193 st
->ss
->load_super(st
, dfd
, NULL
)) {
195 fprintf(stderr
, Name
": no RAID superblock on %s.\n",
197 rv
= try_spare(devname
, &dfd
, policy
,
198 have_target
? &target_array
: NULL
,
203 close (dfd
); dfd
= -1;
205 memset(&info
, 0, sizeof(info
));
206 st
->ss
->getinfo_super(st
, &info
, NULL
);
208 /* 3/ Check if there is a match in mdadm.conf */
209 match
= search_mdstat(st
, &info
, devname
, verbose
, &rv
);
210 if (!match
&& rv
== 2)
213 if (match
&& match
->devname
214 && strcasecmp(match
->devname
, "<ignore>") == 0) {
216 fprintf(stderr
, Name
": array containing %s is explicitly"
217 " ignored by mdadm.conf\n",
222 /* 3a/ if not, check for homehost match. If no match, continue
223 * but don't trust the 'name' in the array. Thus a 'random' minor
224 * number will be assigned, and the device name will be based
228 else if (st
->ss
->match_home(st
, homehost
) == 1)
230 else if (st
->ss
->match_home(st
, "any") == 1)
231 trustworthy
= LOCAL_ANY
;
233 trustworthy
= FOREIGN
;
236 if (!match
&& !conf_test_metadata(st
->ss
->name
, policy
,
237 (trustworthy
== LOCAL
))) {
240 ": %s has metadata type %s for which "
241 "auto-assembly is disabled\n",
242 devname
, st
->ss
->name
);
245 if (trustworthy
== LOCAL_ANY
)
248 /* There are three possible sources for 'autof': command line,
249 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
250 * ARRAY takes precedence, then command line, then
253 if (match
&& match
->autof
)
254 autof
= match
->autof
;
258 name_to_use
= info
.name
;
259 if (name_to_use
[0] == 0 &&
260 info
.array
.level
== LEVEL_CONTAINER
&&
261 trustworthy
== LOCAL
) {
262 name_to_use
= info
.text_version
;
263 trustworthy
= METADATA
;
265 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
266 ! require_homehost
&&
267 conf_name_is_free(name_to_use
))
270 /* strip "hostname:" prefix from name if we have decided
271 * to treat it as LOCAL
273 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
274 name_to_use
= strchr(name_to_use
, ':')+1;
276 /* 4/ Check if array exists.
279 fprintf(stderr
, Name
": failed to get exclusive lock on "
281 mp
= map_by_uuid(&map
, info
.uuid
);
283 mdfd
= open_dev(mp
->devnum
);
289 /* Couldn't find an existing array, maybe make a new one */
290 mdfd
= create_mddev(match
? match
->devname
: NULL
,
291 name_to_use
, autof
, trustworthy
, chosen_name
);
296 sysfs_init(&info
, mdfd
, 0);
298 if (set_array_info(mdfd
, st
, &info
) != 0) {
299 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
300 chosen_name
, strerror(errno
));
306 dinfo
.disk
.major
= major(stb
.st_rdev
);
307 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
308 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
309 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
310 devname
, chosen_name
, strerror(errno
));
311 ioctl(mdfd
, STOP_ARRAY
, 0);
315 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
316 GET_OFFSET
| GET_SIZE
));
318 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
319 /* It really should be 'none' - must be old buggy
320 * kernel, and mdadm -I may not be able to complete.
323 ioctl(mdfd
, STOP_ARRAY
, NULL
);
325 ": You have an old buggy kernel which cannot support\n"
326 " --incremental reliably. Aborting.\n");
331 info
.array
.working_disks
= 1;
332 /* 6/ Make sure /var/run/mdadm.map contains this array. */
333 map_update(&map
, fd2devnum(mdfd
),
335 info
.uuid
, chosen_name
);
338 /* - check one drive in array to make sure metadata is a reasonably */
339 /* close match. Reject if not (e.g. different type) */
340 /* - add the device */
344 struct supertype
*st2
;
345 struct mdinfo info2
, *d
;
347 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
348 GET_OFFSET
| GET_SIZE
));
351 strcpy(chosen_name
, mp
->path
);
353 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
355 /* It is generally not OK to add non-spare drives to a
356 * running array as they are probably missing because
357 * they failed. However if runstop is 1, then the
358 * array was possibly started early and our best bet is
359 * to add this anyway.
360 * Also if action policy is re-add or better we allow
363 if ((info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
364 && ! policy_action_allows(policy
, st
->ss
->name
,
369 if (st
->ss
->external
) {
370 char *devname
= devnum2devname(fd2devnum(mdfd
));
372 active
= devname
&& is_container_active(devname
);
374 } else if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0)
378 ": not adding %s to active array (without --run) %s\n",
379 devname
, chosen_name
);
389 sprintf(dn
, "%d:%d", sra
->devs
->disk
.major
,
390 sra
->devs
->disk
.minor
);
391 dfd2
= dev_open(dn
, O_RDONLY
);
393 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
394 st
->ss
->compare_super(st
, st2
) != 0) {
396 ": metadata mismatch between %s and "
398 devname
, chosen_name
);
404 memset(&info2
, 0, sizeof(info2
));
405 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
406 st2
->ss
->free_super(st2
);
407 if (info
.array
.level
!= info2
.array
.level
||
408 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
409 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
411 ": unexpected difference between %s and %s.\n",
412 chosen_name
, devname
);
417 info2
.disk
.major
= major(stb
.st_rdev
);
418 info2
.disk
.minor
= minor(stb
.st_rdev
);
419 /* add disk needs to know about containers */
420 if (st
->ss
->external
)
421 sra
->array
.level
= LEVEL_CONTAINER
;
422 err
= add_disk(mdfd
, st
, sra
, &info2
);
423 if (err
< 0 && errno
== EBUSY
) {
424 /* could be another device present with the same
425 * disk.number. Find and reject any such
427 find_reject(mdfd
, st
, sra
, info
.disk
.number
,
428 info
.events
, verbose
, chosen_name
);
429 err
= add_disk(mdfd
, st
, sra
, &info2
);
432 fprintf(stderr
, Name
": failed to add %s to %s: %s.\n",
433 devname
, chosen_name
, strerror(errno
));
437 info
.array
.working_disks
= 0;
438 for (d
= sra
->devs
; d
; d
=d
->next
)
439 info
.array
.working_disks
++;
443 /* 7/ Is there enough devices to possibly start the array? */
444 /* 7a/ if not, finish with success. */
445 if (info
.array
.level
== LEVEL_CONTAINER
) {
446 char *devname
= NULL
;
447 /* Try to assemble within the container */
449 sysfs_uevent(&info
, "change");
452 ": container %s now has %d devices\n",
453 chosen_name
, info
.array
.working_disks
);
454 wait_for(chosen_name
, mdfd
);
455 if (st
->ss
->external
)
456 devname
= devnum2devname(fd2devnum(mdfd
));
459 rv
= Incremental(chosen_name
, verbose
, runstop
,
460 NULL
, homehost
, require_homehost
, autof
);
462 /* Don't fail the whole -I if a subarray didn't
463 * have enough devices to start yet
466 /* after spare is added, ping monitor for external metadata
467 * so that it can eg. try to rebuild degraded array */
468 if (st
->ss
->external
) {
469 ping_monitor(devname
);
475 /* We have added something to the array, so need to re-read the
476 * state. Eventually this state should be kept up-to-date as
480 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
481 GET_OFFSET
| GET_SIZE
));
482 active_disks
= count_active(st
, sra
, mdfd
, &avail
, &info
);
483 if (enough(info
.array
.level
, info
.array
.raid_disks
,
484 info
.array
.layout
, info
.array
.state
& 1,
485 avail
, active_disks
) == 0) {
488 ": %s attached to %s, not enough to start (%d).\n",
489 devname
, chosen_name
, active_disks
);
496 /* - if number of OK devices match expected, or -R and there */
498 /* + add any bitmap file */
499 /* + start the array (auto-readonly). */
501 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
504 ": %s attached to %s which is already active.\n",
505 devname
, chosen_name
);
512 if (runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
514 /* Let's try to start it */
515 if (match
&& match
->bitmap_file
) {
516 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
519 ": Could not open bitmap file %s.\n",
523 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
526 ": Failed to set bitmapfile for %s.\n",
532 /* Need to remove from the array any devices which
533 * 'count_active' discerned were too old or inappropriate
535 for (d
= sra
? sra
->devs
: NULL
; d
; d
= d
->next
)
536 if (d
->disk
.state
& (1<<MD_DISK_REMOVED
))
537 remove_disk(mdfd
, st
, sra
, d
);
539 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
540 && trustworthy
!= FOREIGN
)
541 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
543 rv
= sysfs_set_str(sra
, NULL
,
544 "array_state", "read-auto");
548 ": %s attached to %s, which has been started.\n",
549 devname
, chosen_name
);
551 wait_for(chosen_name
, mdfd
);
552 /* We just started the array, so some devices
553 * might have been evicted from the array
554 * because their event counts were too old.
555 * If the action=re-add policy is in-force for
556 * those devices we should re-add them now.
558 for (dsk
= sra
->devs
; dsk
; dsk
= dsk
->next
) {
559 if (disk_action_allows(dsk
, st
->ss
->name
, act_re_add
) &&
560 add_disk(mdfd
, st
, sra
, dsk
) == 0)
562 ": %s re-added to %s\n",
563 dsk
->sys_name
, chosen_name
);
567 ": %s attached to %s, but failed to start: %s.\n",
568 devname
, chosen_name
, strerror(errno
));
574 ": %s attached to %s, not enough to start safely.\n",
575 devname
, chosen_name
);
585 dev_policy_free(policy
);
591 static struct mddev_ident
*search_mdstat(struct supertype
*st
,
594 int verbose
, int *rvp
)
596 struct mddev_ident
*array_list
, *match
;
597 array_list
= conf_get_ident(NULL
);
599 for (; array_list
; array_list
= array_list
->next
) {
600 if (array_list
->uuid_set
&&
601 same_uuid(array_list
->uuid
, info
->uuid
, st
->ss
->swapuuid
)
603 if (verbose
>= 2 && array_list
->devname
)
605 ": UUID differs from %s.\n",
606 array_list
->devname
);
609 if (array_list
->name
[0] &&
610 strcasecmp(array_list
->name
, info
->name
) != 0) {
611 if (verbose
>= 2 && array_list
->devname
)
613 ": Name differs from %s.\n",
614 array_list
->devname
);
617 if (array_list
->devices
&&
618 !match_oneof(array_list
->devices
, devname
)) {
619 if (verbose
>= 2 && array_list
->devname
)
621 ": Not a listed device for %s.\n",
622 array_list
->devname
);
625 if (array_list
->super_minor
!= UnSet
&&
626 array_list
->super_minor
!= info
->array
.md_minor
) {
627 if (verbose
>= 2 && array_list
->devname
)
629 ": Different super-minor to %s.\n",
630 array_list
->devname
);
633 if (!array_list
->uuid_set
&&
634 !array_list
->name
[0] &&
635 !array_list
->devices
&&
636 array_list
->super_minor
== UnSet
) {
637 if (verbose
>= 2 && array_list
->devname
)
639 ": %s doesn't have any identifying information.\n",
640 array_list
->devname
);
643 /* FIXME, should I check raid_disks and level too?? */
647 if (match
->devname
&& array_list
->devname
)
649 ": we match both %s and %s - cannot decide which to use.\n",
650 match
->devname
, array_list
->devname
);
653 ": multiple lines in mdadm.conf match\n");
664 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
665 int number
, __u64 events
, int verbose
,
668 /* Find a device attached to this array with a disk.number of number
669 * and events less than the passed events, and remove the device.
674 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
675 return; /* not safe to remove from active arrays
676 * without thinking more */
678 for (d
= sra
->devs
; d
; d
= d
->next
) {
682 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
683 dfd
= dev_open(dn
, O_RDONLY
);
686 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
690 st
->ss
->getinfo_super(st
, &info
, NULL
);
691 st
->ss
->free_super(st
);
694 if (info
.disk
.number
!= number
||
695 info
.events
>= events
)
698 if (d
->disk
.raid_disk
> -1)
699 sysfs_set_str(sra
, d
, "slot", "none");
700 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
703 ": removing old device %s from %s\n",
704 d
->sys_name
+4, array_name
);
708 static int count_active(struct supertype
*st
, struct mdinfo
*sra
,
709 int mdfd
, char **availp
,
710 struct mdinfo
*bestinfo
)
712 /* count how many devices in sra think they are active */
715 __u64 max_events
= 0;
727 for (d
= sra
->devs
; d
; d
= d
->next
)
729 for (d
= sra
->devs
, devnum
=0 ; d
; d
= d
->next
, devnum
++) {
735 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
736 dfd
= dev_open(dn
, O_RDONLY
);
739 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
743 info
.array
.raid_disks
= raid_disks
;
744 st
->ss
->getinfo_super(st
, &info
, devmap
+ raid_disks
* devnum
);
746 raid_disks
= info
.array
.raid_disks
;
747 avail
= calloc(raid_disks
, 1);
749 fprintf(stderr
, Name
": out of memory.\n");
754 best
= calloc(raid_disks
, sizeof(int));
755 devmap
= calloc(raid_disks
* numdevs
, 1);
757 st
->ss
->getinfo_super(st
, &info
, devmap
);
760 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
764 max_events
= info
.events
;
765 avail
[info
.disk
.raid_disk
] = 2;
766 best
[info
.disk
.raid_disk
] = devnum
;
767 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
768 } else if (info
.events
== max_events
) {
769 avail
[info
.disk
.raid_disk
] = 2;
770 best
[info
.disk
.raid_disk
] = devnum
;
771 } else if (info
.events
== max_events
-1) {
772 if (avail
[info
.disk
.raid_disk
] == 0) {
773 avail
[info
.disk
.raid_disk
] = 1;
774 best
[info
.disk
.raid_disk
] = devnum
;
776 } else if (info
.events
< max_events
- 1)
778 else if (info
.events
== max_events
+1) {
780 max_events
= info
.events
;
781 for (i
=0; i
< raid_disks
; i
++)
784 avail
[info
.disk
.raid_disk
] = 2;
785 best
[info
.disk
.raid_disk
] = devnum
;
786 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
787 } else { /* info.events much bigger */
788 memset(avail
, 0, info
.disk
.raid_disk
);
789 max_events
= info
.events
;
790 avail
[info
.disk
.raid_disk
] = 2;
791 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
794 st
->ss
->free_super(st
);
798 /* We need to reject any device that thinks the best device is
799 * failed or missing */
800 for (b
= 0; b
< raid_disks
; b
++)
804 for (i
= 0 ; i
< raid_disks
; i
++) {
805 if (i
!= b
&& avail
[i
])
806 if (devmap
[raid_disks
* best
[i
] + b
] == 0) {
807 /* This device thinks 'b' is failed -
810 for (d
=sra
->devs
; devnum
; d
= d
->next
)
812 d
->disk
.state
|= (1 << MD_DISK_REMOVED
);
823 static int array_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
824 struct map_ent
*target
, int bare
,
825 struct supertype
*st
, int verbose
)
827 /* This device doesn't have any md metadata
828 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
829 * If 'st' is not set, then we only know that some metadata allows this,
830 * others possibly don't.
831 * So look for a container or array to attach the device to.
832 * Prefer 'target' if that is set and the array is found.
834 * If st is set, then only arrays of that type are considered
835 * Return 0 on success, or some exit code on failure, probably 1.
839 struct map_ent
*mp
, *map
= NULL
;
840 struct mdinfo
*chosen
= NULL
;
843 if (fstat(dfd
, &stb
) != 0)
847 * Now we need to find a suitable array to add this to.
848 * We only accept arrays that:
850 * - are in the same domains as the device
851 * - are of an size for which the device will be useful
852 * and we choose the one that is the most degraded
855 if (map_lock(&map
)) {
856 fprintf(stderr
, Name
": failed to get exclusive lock on "
860 for (mp
= map
; mp
; mp
= mp
->next
) {
861 struct supertype
*st2
;
862 struct domainlist
*dl
= NULL
;
864 unsigned long long devsize
;
866 if (is_subarray(mp
->metadata
))
869 st2
= st
->ss
->match_metadata_desc(mp
->metadata
);
871 (st
->minor_version
>= 0 &&
872 st
->minor_version
!= st2
->minor_version
)) {
874 fprintf(stderr
, Name
": not adding %s to %s as metadata type doesn't match\n",
881 sra
= sysfs_read(-1, mp
->devnum
,
882 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
883 GET_DEGRADED
|GET_COMPONENT
|GET_VERSION
);
885 /* Probably a container - no degraded info */
886 sra
= sysfs_read(-1, mp
->devnum
,
887 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
888 GET_COMPONENT
|GET_VERSION
);
890 sra
->array
.failed_disks
= 0;
897 for(i
=0; !st2
&& superlist
[i
]; i
++)
898 st2
= superlist
[i
]->match_metadata_desc(
902 fprintf(stderr
, Name
": not adding %s to %s"
903 " as metadata not recognised.\n",
907 /* Need to double check the 'act_spare' permissions applies
910 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
912 if (!bare
&& !policy_action_allows(pol
, st2
->ss
->name
,
913 act_spare_same_slot
))
917 get_dev_size(dfd
, NULL
, &devsize
);
918 if (st2
->ss
->avail_size(st2
, devsize
) < sra
->component_size
) {
920 fprintf(stderr
, Name
": not adding %s to %s as it is too small\n",
924 dl
= domain_from_array(sra
, st2
->ss
->name
);
925 if (!domain_test(dl
, pol
, st2
->ss
->name
)) {
926 /* domain test fails */
928 fprintf(stderr
, Name
": not adding %s to %s as it is not in a compatible domain\n",
933 /* test against target.
934 * If 'target' is set and 'bare' is false, we only accept
935 * arrays/containers that match 'target'.
936 * If 'target' is set and 'bare' is true, we prefer the
937 * array which matches 'target'.
940 if (strcmp(target
->metadata
, mp
->metadata
) == 0 &&
941 memcmp(target
->uuid
, mp
->uuid
,
942 sizeof(target
->uuid
)) == 0) {
943 /* This is our target!! */
948 /* skip to end so we don't check any more */
958 /* all tests passed, OK to add to this array */
962 } else if (chosen
->array
.failed_disks
< sra
->array
.failed_disks
) {
976 /* add current device to chosen array as a spare */
977 int mdfd
= open_dev(devname2devnum(chosen
->sys_name
));
979 struct mddev_dev devlist
;
984 devlist
.writemostly
= 0;
985 devlist
.devname
= devname
;
986 sprintf(devname
, "%d:%d", major(stb
.st_rdev
),
988 devlist
.disposition
= 'a';
991 rv
= Manage_subdevs(chosen
->sys_name
, mdfd
, &devlist
,
997 fprintf(stderr
, Name
": added %s as spare for %s\n",
998 devname
, chosen
->sys_name
);
1000 fprintf(stderr
, Name
": failed to add %s as spare for %s\n",
1001 devname
, chosen
->sys_name
);
1008 static int partition_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1009 struct supertype
*st
, int verbose
)
1011 /* we know that at least one partition virtual-metadata is
1012 * allowed to incorporate spares like this device. We need to
1013 * find a suitable device to copy partition information from.
1015 * Getting a list of all disk (not partition) devices is
1016 * slightly non-trivial. We could look at /sys/block, but
1017 * that is theoretically due to be removed. Maybe best to use
1018 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1019 * we depend on this directory of 'path' info. But that fails
1020 * to find loop devices and probably others. Maybe don't
1021 * worry about that, they aren't the real target.
1023 * So: check things in /dev/disk/by-path to see if they are in
1024 * a compatible domain, then load the partition table and see
1025 * if it is OK for the new device, and choose the largest
1026 * partition table that fits.
1030 char *chosen
= NULL
;
1031 unsigned long long chosen_size
;
1032 struct supertype
*chosen_st
= NULL
;
1035 dir
= opendir("/dev/disk/by-path");
1038 while ((de
= readdir(dir
)) != NULL
) {
1040 struct dev_policy
*pol2
= NULL
;
1041 struct domainlist
*domlist
= NULL
;
1044 struct supertype
*st2
= NULL
;
1045 char *devname
= NULL
;
1046 unsigned long long devsectors
;
1048 if (de
->d_ino
== 0 ||
1049 de
->d_name
[0] == '.' ||
1050 (de
->d_type
!= DT_LNK
&& de
->d_type
!= DT_UNKNOWN
))
1053 ep
= de
->d_name
+ strlen(de
->d_name
);
1054 while (ep
> de
->d_name
&&
1057 if (ep
> de
->d_name
+ 5 &&
1058 strncmp(ep
-5, "-part", 5) == 0)
1059 /* This is a partition - skip it */
1062 pol2
= path_policy(de
->d_name
, type_disk
);
1064 domain_merge(&domlist
, pol2
, st
? st
->ss
->name
: NULL
);
1065 if (domain_test(domlist
, pol
, st
? st
->ss
->name
: NULL
) == 0)
1066 /* new device is incompatible with this device. */
1069 domain_free(domlist
);
1072 asprintf(&devname
, "/dev/disk/by-path/%s", de
->d_name
);
1073 fd
= open(devname
, O_RDONLY
);
1076 if (get_dev_size(fd
, devname
, &devsectors
) == 0)
1081 st2
= dup_super(st
);
1083 st2
= guess_super_type(fd
, guess_partitions
);
1085 st2
->ss
->load_super(st2
, fd
, NULL
) < 0)
1089 /* Check domain policy again, this time referring to metadata */
1090 domain_merge(&domlist
, pol2
, st2
->ss
->name
);
1091 if (domain_test(domlist
, pol
, st2
->ss
->name
) == 0)
1092 /* Incompatible devices for this metadata type */
1094 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
1095 /* Some partition types allow sparing, but not
1101 st2
->ss
->getinfo_super(st2
, &info
, NULL
);
1102 if (info
.component_size
> devsectors
)
1103 /* This partitioning doesn't fit in the device */
1106 /* This is an acceptable device to copy partition
1107 * metadata from. We could just stop here, but I
1108 * think I want to keep looking incase a larger
1109 * metadata which makes better use of the device can
1112 if (chosen
== NULL
||
1113 chosen_size
< info
.component_size
) {
1114 chosen_size
= info
.component_size
;
1119 chosen_st
->ss
->free_super(chosen_st
);
1128 domain_free(domlist
);
1129 dev_policy_free(pol2
);
1131 st2
->ss
->free_super(st2
);
1141 /* 'chosen' is the best device we can find. Let's write its
1142 * metadata to devname dfd is read-only so don't use that
1144 fd
= open(devname
, O_RDWR
);
1146 chosen_st
->ss
->store_super(chosen_st
, fd
);
1150 chosen_st
->ss
->free_super(chosen_st
);
1155 static int is_bare(int dfd
)
1157 unsigned long long size
= 0;
1158 char bufpad
[4096 + 4096];
1159 char *buf
= (char*)(((long)bufpad
+ 4096) & ~4095);
1161 if (lseek(dfd
, 0, SEEK_SET
) != 0 ||
1162 read(dfd
, buf
, 4096) != 4096)
1165 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1167 if (memcmp(buf
, buf
+1, 4095) != 0)
1170 /* OK, first 4K appear blank, try the end. */
1171 get_dev_size(dfd
, NULL
, &size
);
1172 if (lseek(dfd
, size
-4096, SEEK_SET
) < 0 ||
1173 read(dfd
, buf
, 4096) != 4096)
1176 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1178 if (memcmp(buf
, buf
+1, 4095) != 0)
1184 /* adding a spare to a regular array is quite different from adding one to
1185 * a set-of-partitions virtual array.
1186 * This function determines which is worth trying and tries as appropriate.
1187 * Arrays are given priority over partitions.
1189 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1190 struct map_ent
*target
,
1191 struct supertype
*st
, int verbose
)
1196 int partitions_ok
= 0;
1200 /* Can only add a spare if device has at least one domain */
1201 if (pol_find(pol
, pol_domain
) == NULL
)
1203 /* And only if some action allows spares */
1204 if (!policy_action_allows(pol
, st
?st
->ss
->name
:NULL
, act_spare
))
1207 /* Now check if the device is bare.
1208 * bare devices can always be added as a spare
1209 * non-bare devices can only be added if spare-same-slot is permitted,
1210 * and this device is replacing a previous device - in which case 'target'
1213 if (!is_bare(dfd
)) {
1214 /* Must have a target and allow same_slot */
1215 /* Later - may allow force_spare without target */
1217 !policy_action_allows(pol
, st
?st
->ss
->name
:NULL
,
1218 act_spare_same_slot
)) {
1220 fprintf(stderr
, Name
": %s is not bare, so not "
1221 "considering as a spare\n",
1229 /* It might be OK to add this device to an array - need to see
1230 * what arrays might be candidates.
1233 /* just try try 'array' or 'partition' based on this metadata */
1234 if (st
->ss
->add_to_super
)
1235 return array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1238 return partition_try_spare(devname
, dfdp
, pol
,
1241 /* No metadata was specified or found so options are open.
1242 * Check for whether any array metadata, or any partition metadata
1243 * might allow adding the spare. This check is just help to avoid
1244 * a more costly scan of all arrays when we can be sure that will
1247 for (i
= 0; (!arrays_ok
|| !partitions_ok
) && superlist
[i
] ; i
++) {
1248 if (superlist
[i
]->add_to_super
&& !arrays_ok
&&
1249 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1251 if (superlist
[i
]->add_to_super
== NULL
&& !partitions_ok
&&
1252 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1257 rv
= array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1259 if (rv
!= 0 && partitions_ok
)
1260 rv
= partition_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1264 int IncrementalScan(int verbose
)
1266 /* look at every device listed in the 'map' file.
1267 * If one is found that is not running then:
1268 * look in mdadm.conf for bitmap file.
1269 * if one exists, but array has none, add it.
1270 * try to start array in auto-readonly mode
1272 struct map_ent
*mapl
= NULL
;
1274 struct mddev_ident
*devs
, *mddev
;
1278 devs
= conf_get_ident(NULL
);
1280 for (me
= mapl
; me
; me
= me
->next
) {
1281 mdu_array_info_t array
;
1282 mdu_bitmap_file_t bmf
;
1284 int mdfd
= open_dev(me
->devnum
);
1288 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
1293 /* Ok, we can try this one. Maybe it needs a bitmap */
1294 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
1295 if (mddev
->devname
&& me
->path
1296 && devname_matches(mddev
->devname
, me
->path
))
1298 if (mddev
&& mddev
->bitmap_file
) {
1300 * Note: early kernels will wrongly fail this, so it
1304 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
1305 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
1307 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
1314 fprintf(stderr
, Name
1315 ": Added bitmap %s to %s\n",
1316 mddev
->bitmap_file
, me
->path
);
1317 else if (errno
!= EEXIST
)
1318 fprintf(stderr
, Name
1319 ": Failed to add bitmap to %s: %s\n",
1320 me
->path
, strerror(errno
));
1323 sra
= sysfs_read(mdfd
, 0, 0);
1325 if (sysfs_set_str(sra
, NULL
,
1326 "array_state", "read-auto") == 0) {
1328 fprintf(stderr
, Name
1329 ": started array %s\n",
1330 me
->path
?: devnum2devname(me
->devnum
));
1332 fprintf(stderr
, Name
1333 ": failed to start array %s: %s\n",
1334 me
->path
?: devnum2devname(me
->devnum
),
1343 static char *container2devname(char *devname
)
1345 char *mdname
= NULL
;
1347 if (devname
[0] == '/') {
1348 int fd
= open(devname
, O_RDONLY
);
1350 mdname
= devnum2devname(fd2devnum(fd
));
1355 struct map_ent
*mp
, *map
= NULL
;
1357 if (!parse_uuid(devname
, uuid
))
1359 mp
= map_by_uuid(&map
, uuid
);
1361 mdname
= devnum2devname(mp
->devnum
);
1368 static int Incremental_container(struct supertype
*st
, char *devname
,
1369 char *homehost
, int verbose
,
1370 int runstop
, int autof
)
1372 /* Collect the contents of this container and for each
1373 * array, choose a device name and assemble the array.
1376 struct mdinfo
*list
;
1378 struct map_ent
*map
= NULL
;
1381 struct mddev_ident
*match
;
1384 memset(&info
, 0, sizeof(info
));
1385 st
->ss
->getinfo_super(st
, &info
, NULL
);
1387 if ((runstop
> 0 && info
.container_enough
>= 0) ||
1388 info
.container_enough
> 0)
1392 fprintf(stderr
, Name
": not enough devices to start the container\n");
1396 match
= search_mdstat(st
, &info
, devname
, verbose
, &rv
);
1397 if (match
== NULL
&& rv
== 2)
1400 /* Need to compute 'trustworthy' */
1402 trustworthy
= LOCAL
;
1403 else if (st
->ss
->match_home(st
, homehost
) == 1)
1404 trustworthy
= LOCAL
;
1405 else if (st
->ss
->match_home(st
, "any") == 1)
1406 trustworthy
= LOCAL
;
1408 trustworthy
= FOREIGN
;
1410 list
= st
->ss
->container_content(st
, NULL
);
1412 fprintf(stderr
, Name
": failed to get exclusive lock on "
1415 for (ra
= list
; ra
; ra
= ra
->next
) {
1417 char chosen_name
[1024];
1419 struct mddev_ident
*match
= NULL
;
1421 mp
= map_by_uuid(&map
, ra
->uuid
);
1424 mdfd
= open_dev(mp
->devnum
);
1426 strcpy(chosen_name
, mp
->path
);
1428 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
1431 /* Check in mdadm.conf for container == devname and
1432 * member == ra->text_version after second slash.
1434 char *sub
= strchr(ra
->text_version
+1, '/');
1435 struct mddev_ident
*array_list
;
1438 array_list
= conf_get_ident(NULL
);
1441 for(; array_list
; array_list
= array_list
->next
) {
1443 if (array_list
->member
== NULL
||
1444 array_list
->container
== NULL
)
1446 if (strcmp(array_list
->member
, sub
) != 0)
1448 if (array_list
->uuid_set
&&
1449 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
1451 dn
= container2devname(array_list
->container
);
1454 if (strncmp(dn
, ra
->text_version
+1,
1456 ra
->text_version
[strlen(dn
)+1] != '/') {
1461 /* we have a match */
1464 fprintf(stderr
, Name
": match found for member %s\n",
1465 array_list
->member
);
1469 if (match
&& match
->devname
&&
1470 strcasecmp(match
->devname
, "<ignore>") == 0) {
1472 fprintf(stderr
, Name
": array %s/%s is "
1473 "explicitly ignored by mdadm.conf\n",
1474 match
->container
, match
->member
);
1478 trustworthy
= LOCAL
;
1480 mdfd
= create_mddev(match
? match
->devname
: NULL
,
1488 fprintf(stderr
, Name
": failed to open %s: %s.\n",
1489 chosen_name
, strerror(errno
));
1493 assemble_container_content(st
, mdfd
, ra
, runstop
,
1494 chosen_name
, verbose
);
1501 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1502 * raid arrays, and if so first fail (if needed) and then remove the device.
1504 * @devname - The device we want to remove
1505 * @id_path - name as found in /dev/disk/by-path for this device
1507 * Note: the device name must be a kernel name like "sda", so
1508 * that we can find it in /proc/mdstat
1510 int IncrementalRemove(char *devname
, char *id_path
, int verbose
)
1514 struct mdstat_ent
*ent
;
1515 struct mddev_dev devlist
;
1518 dprintf(Name
": incremental removal without --path <id_path> "
1519 "lacks the possibility to re-add new device in this "
1522 if (strchr(devname
, '/')) {
1523 fprintf(stderr
, Name
": incremental removal requires a "
1524 "kernel device name, not a file: %s\n", devname
);
1527 ent
= mdstat_by_component(devname
);
1529 fprintf(stderr
, Name
": %s does not appear to be a component "
1530 "of any array\n", devname
);
1533 mdfd
= open_dev(ent
->devnum
);
1535 fprintf(stderr
, Name
": Cannot open array %s!!\n", ent
->dev
);
1541 struct map_ent
*map
= NULL
, *me
;
1542 me
= map_by_devnum(&map
, ent
->devnum
);
1544 policy_save_path(id_path
, me
);
1548 memset(&devlist
, 0, sizeof(devlist
));
1549 devlist
.devname
= devname
;
1550 devlist
.disposition
= 'f';
1551 /* for a container, we must fail each member array */
1552 if (ent
->metadata_version
&&
1553 strncmp(ent
->metadata_version
, "external:", 9) == 0) {
1554 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1555 struct mdstat_ent
*memb
;
1556 for (memb
= mdstat
; memb
; memb
= memb
->next
)
1557 if (is_container_member(memb
, ent
->dev
)) {
1558 int subfd
= open_dev(memb
->devnum
);
1560 Manage_subdevs(memb
->dev
, subfd
,
1561 &devlist
, verbose
, 0,
1566 free_mdstat(mdstat
);
1568 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
);
1569 devlist
.disposition
= 'r';
1570 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
);