2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
5 * Copyright (C) 2006-2012 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
,
48 int Incremental(char *devname
, struct context
*c
,
51 /* Add this device to an array, creating the array if necessary
52 * and starting the array if sensible or - if runstop>0 - if possible.
54 * This has several steps:
56 * 1/ Check if device is permitted by mdadm.conf, reject if not.
57 * 2/ Find metadata, reject if none appropriate (check
58 * version/name from args)
59 * 3/ Check if there is a match in mdadm.conf
60 * 3a/ if not, check for homehost match. If no match, assemble as
62 * 4/ Determine device number.
63 * - If in mdadm.conf with std name, use that
64 * - UUID in /var/run/mdadm.map use that
65 * - If name is suggestive, use that. unless in use with different uuid.
66 * - Choose a free, high number.
67 * - Use a partitioned device unless strong suggestion not to.
69 * Don't choose partitioned for containers.
70 * 5/ Find out if array already exists
72 * - choose a name, from mdadm.conf or 'name' field in array.
76 * - check one drive in array to make sure metadata is a reasonably
77 * close match. Reject if not (e.g. different type)
79 * 6/ Make sure /var/run/mdadm.map contains this array.
80 * 7/ Is there enough devices to possibly start the array?
81 * For a container, this means running Incremental_container.
82 * 7a/ if not, finish with success.
84 * - read all metadata and arrange devices like -A does
85 * - if number of OK devices match expected, or -R and there are enough,
86 * start the array (auto-readonly).
89 struct mdinfo info
, dinfo
;
90 struct mdinfo
*sra
= NULL
, *d
;
91 struct mddev_ident
*match
;
92 char chosen_name
[1024];
94 struct map_ent
*mp
, *map
= NULL
;
95 int dfd
= -1, mdfd
= -1;
100 mdu_array_info_t ainf
;
101 struct dev_policy
*policy
= NULL
;
102 struct map_ent target_array
;
105 struct createinfo
*ci
= conf_get_create_info();
107 if (stat(devname
, &stb
) < 0) {
109 pr_err("stat failed for %s: %s.\n",
110 devname
, strerror(errno
));
113 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
115 pr_err("%s is not a block device.\n",
119 dfd
= dev_open(devname
, O_RDONLY
);
122 pr_err("cannot open %s: %s.\n",
123 devname
, strerror(errno
));
126 /* If the device is a container, we do something very different */
127 if (must_be_container(dfd
)) {
129 st
= super_by_fd(dfd
, NULL
);
130 if (st
&& st
->ss
->load_container
)
131 rv
= st
->ss
->load_container(st
, dfd
, NULL
);
134 if (!rv
&& st
->ss
->container_content
) {
136 pr_err("failed to get "
137 "exclusive lock on mapfile\n");
138 rv
= Incremental_container(st
, devname
, c
);
143 pr_err("%s is not part of an md array.\n",
148 /* 1/ Check if device is permitted by mdadm.conf */
150 if (!conf_test_dev(devname
)) {
152 pr_err("%s not permitted by mdadm.conf.\n",
157 /* 2/ Find metadata, reject if none appropriate (check
158 * version/name from args) */
160 if (fstat(dfd
, &stb
) < 0) {
162 pr_err("fstat failed for %s: %s.\n",
163 devname
, strerror(errno
));
166 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
168 pr_err("%s is not a block device.\n",
173 dinfo
.disk
.major
= major(stb
.st_rdev
);
174 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
176 policy
= disk_policy(&dinfo
);
177 have_target
= policy_check_path(&dinfo
, &target_array
);
179 if (st
== NULL
&& (st
= guess_super(dfd
)) == NULL
) {
181 pr_err("no recognisable superblock on %s.\n",
183 rv
= try_spare(devname
, &dfd
, policy
,
184 have_target
? &target_array
: NULL
,
188 if (st
->ss
->compare_super
== NULL
||
189 st
->ss
->load_super(st
, dfd
, NULL
)) {
191 pr_err("no RAID superblock on %s.\n",
193 rv
= try_spare(devname
, &dfd
, policy
,
194 have_target
? &target_array
: NULL
,
199 close (dfd
); dfd
= -1;
201 st
->ss
->getinfo_super(st
, &info
, NULL
);
203 /* 3/ Check if there is a match in mdadm.conf */
204 match
= conf_match(st
, &info
, devname
, c
->verbose
, &rv
);
205 if (!match
&& rv
== 2)
208 if (match
&& match
->devname
209 && strcasecmp(match
->devname
, "<ignore>") == 0) {
211 pr_err("array containing %s is explicitly"
212 " ignored by mdadm.conf\n",
217 /* 3a/ if not, check for homehost match. If no match, continue
218 * but don't trust the 'name' in the array. Thus a 'random' minor
219 * number will be assigned, and the device name will be based
223 else if (st
->ss
->match_home(st
, c
->homehost
) == 1)
225 else if (st
->ss
->match_home(st
, "any") == 1)
226 trustworthy
= LOCAL_ANY
;
228 trustworthy
= FOREIGN
;
230 if (!match
&& !conf_test_metadata(st
->ss
->name
, policy
,
231 (trustworthy
== LOCAL
))) {
233 pr_err("%s has metadata type %s for which "
234 "auto-assembly is disabled\n",
235 devname
, st
->ss
->name
);
238 if (trustworthy
== LOCAL_ANY
)
241 /* There are three possible sources for 'autof': command line,
242 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
243 * ARRAY takes precedence, then command line, then
246 if (match
&& match
->autof
)
247 c
->autof
= match
->autof
;
249 c
->autof
= ci
->autof
;
251 name_to_use
= info
.name
;
252 if (name_to_use
[0] == 0 &&
253 info
.array
.level
== LEVEL_CONTAINER
) {
254 name_to_use
= info
.text_version
;
255 trustworthy
= METADATA
;
257 if (name_to_use
[0] && trustworthy
!= LOCAL
&&
258 ! c
->require_homehost
&&
259 conf_name_is_free(name_to_use
))
262 /* strip "hostname:" prefix from name if we have decided
263 * to treat it as LOCAL
265 if (trustworthy
== LOCAL
&& strchr(name_to_use
, ':') != NULL
)
266 name_to_use
= strchr(name_to_use
, ':')+1;
268 /* 4/ Check if array exists.
271 pr_err("failed to get exclusive lock on "
273 /* Now check we can get O_EXCL. If not, probably "mdadm -A" has
276 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
279 pr_err("cannot reopen %s: %s.\n",
280 devname
, strerror(errno
));
283 /* Cannot hold it open while we add the device to the array,
284 * so we must release the O_EXCL and depend on the map_lock()
289 mp
= map_by_uuid(&map
, info
.uuid
);
291 mdfd
= open_dev(mp
->devnum
);
297 /* Couldn't find an existing array, maybe make a new one */
298 mdfd
= create_mddev(match
? match
->devname
: NULL
,
299 name_to_use
, c
->autof
, trustworthy
, chosen_name
);
304 sysfs_init(&info
, mdfd
, 0);
306 if (set_array_info(mdfd
, st
, &info
) != 0) {
307 pr_err("failed to set array info for %s: %s\n",
308 chosen_name
, strerror(errno
));
314 dinfo
.disk
.major
= major(stb
.st_rdev
);
315 dinfo
.disk
.minor
= minor(stb
.st_rdev
);
316 if (add_disk(mdfd
, st
, &info
, &dinfo
) != 0) {
317 pr_err("failed to add %s to %s: %s.\n",
318 devname
, chosen_name
, strerror(errno
));
319 ioctl(mdfd
, STOP_ARRAY
, 0);
323 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
324 GET_OFFSET
| GET_SIZE
));
326 if (!sra
|| !sra
->devs
|| sra
->devs
->disk
.raid_disk
>= 0) {
327 /* It really should be 'none' - must be old buggy
328 * kernel, and mdadm -I may not be able to complete.
331 ioctl(mdfd
, STOP_ARRAY
, NULL
);
332 pr_err("You have an old buggy kernel which cannot support\n"
333 " --incremental reliably. Aborting.\n");
337 info
.array
.working_disks
= 1;
338 /* 6/ Make sure /var/run/mdadm.map contains this array. */
339 map_update(&map
, fd2devnum(mdfd
),
341 info
.uuid
, chosen_name
);
344 /* - check one drive in array to make sure metadata is a reasonably */
345 /* close match. Reject if not (e.g. different type) */
346 /* - add the device */
350 struct supertype
*st2
;
351 struct mdinfo info2
, *d
;
353 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
354 GET_OFFSET
| GET_SIZE
));
357 strcpy(chosen_name
, mp
->path
);
359 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
361 /* It is generally not OK to add non-spare drives to a
362 * running array as they are probably missing because
363 * they failed. However if runstop is 1, then the
364 * array was possibly started early and our best bet is
365 * to add this anyway.
366 * Also if action policy is re-add or better we allow
368 * This doesn't apply to containers as the 'non-spare'
369 * flag has a different meaning. The test has to happen
370 * at the device level there
372 if (!st
->ss
->external
373 && (info
.disk
.state
& (1<<MD_DISK_SYNC
)) != 0
374 && ! policy_action_allows(policy
, st
->ss
->name
,
377 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ainf
) == 0) {
378 pr_err("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 pr_err("unable to open %s\n", devname
);
398 if (st2
->ss
->load_super(st2
, dfd2
, NULL
) ||
399 st
->ss
->compare_super(st
, st2
) != 0) {
400 pr_err("metadata mismatch between %s and "
402 devname
, chosen_name
);
408 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
409 st2
->ss
->free_super(st2
);
410 if (info
.array
.level
!= info2
.array
.level
||
411 memcmp(info
.uuid
, info2
.uuid
, 16) != 0 ||
412 info
.array
.raid_disks
!= info2
.array
.raid_disks
) {
413 pr_err("unexpected difference between %s and %s.\n",
414 chosen_name
, devname
);
419 info
.disk
.major
= major(stb
.st_rdev
);
420 info
.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
, &info
);
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
, c
->verbose
, chosen_name
);
431 err
= add_disk(mdfd
, st
, sra
, &info
);
434 pr_err("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");
452 pr_err("container %s now has %d device%s\n",
453 chosen_name
, info
.array
.working_disks
,
454 info
.array
.working_disks
== 1?"":"s");
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
, c
);
466 /* Don't fail the whole -I if a subarray didn't
467 * have enough devices to start yet
470 /* after spare is added, ping monitor for external metadata
471 * so that it can eg. try to rebuild degraded array */
472 if (st
->ss
->external
)
473 ping_monitor_by_id(devnum
);
477 /* We have added something to the array, so need to re-read the
478 * state. Eventually this state should be kept up-to-date as
482 sra
= sysfs_read(mdfd
, -1, (GET_DEVS
| GET_STATE
|
483 GET_OFFSET
| GET_SIZE
));
484 active_disks
= count_active(st
, sra
, mdfd
, &avail
, &info
);
485 if (enough(info
.array
.level
, info
.array
.raid_disks
,
486 info
.array
.layout
, info
.array
.state
& 1,
489 pr_err("%s attached to %s, not enough to start (%d).\n",
490 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) {
503 pr_err("%s attached to %s which is already active.\n",
504 devname
, chosen_name
);
510 if (c
->runstop
> 0 || active_disks
>= info
.array
.working_disks
) {
512 /* Let's try to start it */
514 if (info
.reshape_active
&& !(info
.reshape_active
& RESHAPE_NO_BACKUP
)) {
516 ": %s: This array is being reshaped and cannot be started\n"
517 " by --incremental. Please use --assemble\n",
521 if (match
&& match
->bitmap_file
) {
522 int bmfd
= open(match
->bitmap_file
, O_RDWR
);
524 pr_err("Could not open bitmap file %s.\n",
528 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
530 pr_err("Failed to set bitmapfile for %s.\n",
536 /* Need to remove from the array any devices which
537 * 'count_active' discerned were too old or inappropriate
539 for (d
= sra
? sra
->devs
: NULL
; d
; d
= d
->next
)
540 if (d
->disk
.state
& (1<<MD_DISK_REMOVED
))
541 remove_disk(mdfd
, st
, sra
, d
);
543 if ((sra
== NULL
|| active_disks
>= info
.array
.working_disks
)
544 && trustworthy
!= FOREIGN
)
545 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
547 rv
= sysfs_set_str(sra
, NULL
,
548 "array_state", "read-auto");
551 pr_err("%s attached to %s, which has been started.\n",
552 devname
, chosen_name
);
554 wait_for(chosen_name
, mdfd
);
555 /* We just started the array, so some devices
556 * might have been evicted from the array
557 * because their event counts were too old.
558 * If the action=re-add policy is in-force for
559 * those devices we should re-add them now.
561 for (dsk
= sra
->devs
; dsk
; dsk
= dsk
->next
) {
562 if (disk_action_allows(dsk
, st
->ss
->name
, act_re_add
) &&
563 add_disk(mdfd
, st
, sra
, dsk
) == 0)
564 pr_err("%s re-added to %s\n",
565 dsk
->sys_name
, chosen_name
);
568 pr_err("%s attached to %s, but failed to start: %s.\n",
569 devname
, chosen_name
, strerror(errno
));
574 pr_err("%s attached to %s, not enough to start safely.\n",
575 devname
, chosen_name
);
585 dev_policy_free(policy
);
594 static void find_reject(int mdfd
, struct supertype
*st
, struct mdinfo
*sra
,
595 int number
, __u64 events
, int verbose
,
598 /* Find a device attached to this array with a disk.number of number
599 * and events less than the passed events, and remove the device.
604 if (ioctl(mdfd
, GET_ARRAY_INFO
, &ra
) == 0)
605 return; /* not safe to remove from active arrays
606 * without thinking more */
608 for (d
= sra
->devs
; d
; d
= d
->next
) {
612 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
613 dfd
= dev_open(dn
, O_RDONLY
);
616 if (st
->ss
->load_super(st
, dfd
, NULL
)) {
620 st
->ss
->getinfo_super(st
, &info
, NULL
);
621 st
->ss
->free_super(st
);
624 if (info
.disk
.number
!= number
||
625 info
.events
>= events
)
628 if (d
->disk
.raid_disk
> -1)
629 sysfs_set_str(sra
, d
, "slot", "none");
630 if (sysfs_set_str(sra
, d
, "state", "remove") == 0)
632 pr_err("removing old device %s from %s\n",
633 d
->sys_name
+4, array_name
);
637 static int count_active(struct supertype
*st
, struct mdinfo
*sra
,
638 int mdfd
, char **availp
,
639 struct mdinfo
*bestinfo
)
641 /* count how many devices in sra think they are active */
644 __u64 max_events
= 0;
656 for (d
= sra
->devs
; d
; d
= d
->next
)
658 for (d
= sra
->devs
, devnum
= 0 ; d
; d
= d
->next
, devnum
++) {
664 sprintf(dn
, "%d:%d", d
->disk
.major
, d
->disk
.minor
);
665 dfd
= dev_open(dn
, O_RDONLY
);
668 ok
= st
->ss
->load_super(st
, dfd
, NULL
);
672 info
.array
.raid_disks
= raid_disks
;
673 st
->ss
->getinfo_super(st
, &info
, devmap
+ raid_disks
* devnum
);
675 raid_disks
= info
.array
.raid_disks
;
676 avail
= xcalloc(raid_disks
, 1);
679 best
= xcalloc(raid_disks
, sizeof(int));
680 devmap
= xcalloc(raid_disks
, numdevs
);
682 st
->ss
->getinfo_super(st
, &info
, devmap
);
685 if (info
.disk
.state
& (1<<MD_DISK_SYNC
))
689 max_events
= info
.events
;
690 avail
[info
.disk
.raid_disk
] = 2;
691 best
[info
.disk
.raid_disk
] = devnum
;
692 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
693 } else if (info
.events
== max_events
) {
694 avail
[info
.disk
.raid_disk
] = 2;
695 best
[info
.disk
.raid_disk
] = devnum
;
696 } else if (info
.events
== max_events
-1) {
697 if (avail
[info
.disk
.raid_disk
] == 0) {
698 avail
[info
.disk
.raid_disk
] = 1;
699 best
[info
.disk
.raid_disk
] = devnum
;
701 } else if (info
.events
< max_events
- 1)
703 else if (info
.events
== max_events
+1) {
705 max_events
= info
.events
;
706 for (i
= 0; i
< raid_disks
; i
++)
709 avail
[info
.disk
.raid_disk
] = 2;
710 best
[info
.disk
.raid_disk
] = devnum
;
711 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
712 } else { /* info.events much bigger */
713 memset(avail
, 0, raid_disks
);
714 max_events
= info
.events
;
715 avail
[info
.disk
.raid_disk
] = 2;
716 best
[info
.disk
.raid_disk
] = devnum
;
717 st
->ss
->getinfo_super(st
, bestinfo
, NULL
);
720 st
->ss
->free_super(st
);
724 /* We need to reject any device that thinks the best device is
725 * failed or missing */
726 for (b
= 0; b
< raid_disks
; b
++)
730 for (i
= 0 ; i
< raid_disks
; i
++) {
731 if (i
!= b
&& avail
[i
])
732 if (devmap
[raid_disks
* best
[i
] + b
] == 0) {
733 /* This device thinks 'b' is failed -
736 for (d
=sra
->devs
; devnum
; d
= d
->next
)
738 d
->disk
.state
|= (1 << MD_DISK_REMOVED
);
749 /* test if container has degraded member(s) */
750 static int container_members_max_degradation(struct map_ent
*map
, struct map_ent
*me
)
752 mdu_array_info_t array
;
754 int max_degraded
= 0;
756 for(; map
; map
= map
->next
) {
757 if (!is_subarray(map
->metadata
) ||
758 devname2devnum(map
->metadata
+1) != me
->devnum
)
760 afd
= open_dev(map
->devnum
);
763 /* most accurate information regarding array degradation */
764 if (ioctl(afd
, GET_ARRAY_INFO
, &array
) >= 0) {
765 int degraded
= array
.raid_disks
- array
.active_disks
-
767 if (degraded
> max_degraded
)
768 max_degraded
= degraded
;
772 return (max_degraded
);
775 static int array_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
776 struct map_ent
*target
, int bare
,
777 struct supertype
*st
, int verbose
)
779 /* This device doesn't have any md metadata
780 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
781 * If 'st' is not set, then we only know that some metadata allows this,
782 * others possibly don't.
783 * So look for a container or array to attach the device to.
784 * Prefer 'target' if that is set and the array is found.
786 * If st is set, then only arrays of that type are considered
787 * Return 0 on success, or some exit code on failure, probably 1.
791 struct map_ent
*mp
, *map
= NULL
;
792 struct mdinfo
*chosen
= NULL
;
795 if (fstat(dfd
, &stb
) != 0)
799 * Now we need to find a suitable array to add this to.
800 * We only accept arrays that:
802 * - are in the same domains as the device
803 * - are of an size for which the device will be useful
804 * and we choose the one that is the most degraded
807 if (map_lock(&map
)) {
808 pr_err("failed to get exclusive lock on "
812 for (mp
= map
; mp
; mp
= mp
->next
) {
813 struct supertype
*st2
;
814 struct domainlist
*dl
= NULL
;
816 unsigned long long devsize
;
817 unsigned long long component_size
= 0;
819 if (is_subarray(mp
->metadata
))
822 st2
= st
->ss
->match_metadata_desc(mp
->metadata
);
824 (st
->minor_version
>= 0 &&
825 st
->minor_version
!= st2
->minor_version
)) {
827 pr_err("not adding %s to %s as metadata type doesn't match\n",
834 sra
= sysfs_read(-1, mp
->devnum
,
835 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
836 GET_DEGRADED
|GET_COMPONENT
|GET_VERSION
);
838 /* Probably a container - no degraded info */
839 sra
= sysfs_read(-1, mp
->devnum
,
840 GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
|
841 GET_COMPONENT
|GET_VERSION
);
843 sra
->array
.failed_disks
= -1;
850 for(i
= 0; !st2
&& superlist
[i
]; i
++)
851 st2
= superlist
[i
]->match_metadata_desc(
855 pr_err("not adding %s to %s"
856 " as metadata not recognised.\n",
860 /* Need to double check the 'act_spare' permissions applies
863 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
865 if (!bare
&& !policy_action_allows(pol
, st2
->ss
->name
,
866 act_spare_same_slot
))
870 /* update number of failed disks for mostly degraded
871 * container member */
872 if (sra
->array
.failed_disks
== -1)
873 sra
->array
.failed_disks
= container_members_max_degradation(map
, mp
);
875 get_dev_size(dfd
, NULL
, &devsize
);
876 if (sra
->component_size
== 0) {
877 /* true for containers, here we must read superblock
878 * to obtain minimum spare size */
879 struct supertype
*st3
= dup_super(st2
);
880 int mdfd
= open_dev(mp
->devnum
);
885 if (st3
->ss
->load_container
&&
886 !st3
->ss
->load_container(st3
, mdfd
, mp
->path
)) {
887 component_size
= st3
->ss
->min_acceptable_spare_size(st3
);
888 st3
->ss
->free_super(st3
);
893 if ((sra
->component_size
> 0 &&
894 st2
->ss
->avail_size(st2
, devsize
,
896 ? sra
->devs
->data_offset
898 < sra
->component_size
)
900 (sra
->component_size
== 0 && devsize
< component_size
)) {
902 pr_err("not adding %s to %s as it is too small\n",
906 /* test against target.
907 * If 'target' is set and 'bare' is false, we only accept
908 * arrays/containers that match 'target'.
909 * If 'target' is set and 'bare' is true, we prefer the
910 * array which matches 'target'.
911 * target is considered only if we deal with degraded array
913 if (target
&& policy_action_allows(pol
, st2
->ss
->name
,
914 act_spare_same_slot
)) {
915 if (strcmp(target
->metadata
, mp
->metadata
) == 0 &&
916 memcmp(target
->uuid
, mp
->uuid
,
917 sizeof(target
->uuid
)) == 0 &&
918 sra
->array
.failed_disks
> 0) {
919 /* This is our target!! */
924 /* skip to end so we don't check any more */
934 dl
= domain_from_array(sra
, st2
->ss
->name
);
935 if (domain_test(dl
, pol
, st2
->ss
->name
) != 1) {
936 /* domain test fails */
938 pr_err("not adding %s to %s as"
939 " it is not in a compatible domain\n",
944 /* all tests passed, OK to add to this array */
948 } else if (chosen
->array
.failed_disks
< sra
->array
.failed_disks
) {
962 /* add current device to chosen array as a spare */
963 int mdfd
= open_dev(devname2devnum(chosen
->sys_name
));
965 struct mddev_dev devlist
;
969 devlist
.writemostly
= 0;
970 devlist
.devname
= devname
;
971 sprintf(devname
, "%d:%d", major(stb
.st_rdev
),
973 devlist
.disposition
= 'a';
976 rv
= Manage_subdevs(chosen
->sys_name
, mdfd
, &devlist
,
982 pr_err("added %s as spare for %s\n",
983 devname
, chosen
->sys_name
);
985 pr_err("failed to add %s as spare for %s\n",
986 devname
, chosen
->sys_name
);
994 static int partition_try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
995 struct supertype
*st
, int verbose
)
997 /* we know that at least one partition virtual-metadata is
998 * allowed to incorporate spares like this device. We need to
999 * find a suitable device to copy partition information from.
1001 * Getting a list of all disk (not partition) devices is
1002 * slightly non-trivial. We could look at /sys/block, but
1003 * that is theoretically due to be removed. Maybe best to use
1004 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1005 * we depend on this directory of 'path' info. But that fails
1006 * to find loop devices and probably others. Maybe don't
1007 * worry about that, they aren't the real target.
1009 * So: check things in /dev/disk/by-path to see if they are in
1010 * a compatible domain, then load the partition table and see
1011 * if it is OK for the new device, and choose the largest
1012 * partition table that fits.
1016 char *chosen
= NULL
;
1017 unsigned long long chosen_size
= 0;
1018 struct supertype
*chosen_st
= NULL
;
1021 dir
= opendir("/dev/disk/by-path");
1024 while ((de
= readdir(dir
)) != NULL
) {
1026 struct dev_policy
*pol2
= NULL
;
1027 struct domainlist
*domlist
= NULL
;
1030 struct supertype
*st2
= NULL
;
1031 char *devname
= NULL
;
1032 unsigned long long devsectors
;
1034 if (de
->d_ino
== 0 ||
1035 de
->d_name
[0] == '.' ||
1036 (de
->d_type
!= DT_LNK
&& de
->d_type
!= DT_UNKNOWN
))
1039 ep
= de
->d_name
+ strlen(de
->d_name
);
1040 while (ep
> de
->d_name
&&
1043 if (ep
> de
->d_name
+ 5 &&
1044 strncmp(ep
-5, "-part", 5) == 0)
1045 /* This is a partition - skip it */
1048 pol2
= path_policy(de
->d_name
, type_disk
);
1050 domain_merge(&domlist
, pol2
, st
? st
->ss
->name
: NULL
);
1051 if (domain_test(domlist
, pol
, st
? st
->ss
->name
: NULL
) != 1)
1052 /* new device is incompatible with this device. */
1055 domain_free(domlist
);
1058 if (asprintf(&devname
, "/dev/disk/by-path/%s", de
->d_name
) != 1) {
1062 fd
= open(devname
, O_RDONLY
);
1065 if (get_dev_size(fd
, devname
, &devsectors
) == 0)
1070 st2
= dup_super(st
);
1072 st2
= guess_super_type(fd
, guess_partitions
);
1074 st2
->ss
->load_super(st2
, fd
, NULL
) < 0)
1078 /* Check domain policy again, this time referring to metadata */
1079 domain_merge(&domlist
, pol2
, st2
->ss
->name
);
1080 if (domain_test(domlist
, pol
, st2
->ss
->name
) != 1)
1081 /* Incompatible devices for this metadata type */
1083 if (!policy_action_allows(pol
, st2
->ss
->name
, act_spare
))
1084 /* Some partition types allow sparing, but not
1090 st2
->ss
->getinfo_super(st2
, &info
, NULL
);
1091 if (info
.component_size
> devsectors
)
1092 /* This partitioning doesn't fit in the device */
1095 /* This is an acceptable device to copy partition
1096 * metadata from. We could just stop here, but I
1097 * think I want to keep looking incase a larger
1098 * metadata which makes better use of the device can
1101 if (chosen
== NULL
||
1102 chosen_size
< info
.component_size
) {
1103 chosen_size
= info
.component_size
;
1108 chosen_st
->ss
->free_super(chosen_st
);
1117 domain_free(domlist
);
1118 dev_policy_free(pol2
);
1120 st2
->ss
->free_super(st2
);
1132 /* 'chosen' is the best device we can find. Let's write its
1133 * metadata to devname dfd is read-only so don't use that
1135 fd
= open(devname
, O_RDWR
);
1137 chosen_st
->ss
->store_super(chosen_st
, fd
);
1141 chosen_st
->ss
->free_super(chosen_st
);
1146 static int is_bare(int dfd
)
1148 unsigned long long size
= 0;
1149 char bufpad
[4096 + 4096];
1150 char *buf
= (char*)(((long)bufpad
+ 4096) & ~4095);
1152 if (lseek(dfd
, 0, SEEK_SET
) != 0 ||
1153 read(dfd
, buf
, 4096) != 4096)
1156 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1158 if (memcmp(buf
, buf
+1, 4095) != 0)
1161 /* OK, first 4K appear blank, try the end. */
1162 get_dev_size(dfd
, NULL
, &size
);
1163 if (lseek(dfd
, size
-4096, SEEK_SET
) < 0 ||
1164 read(dfd
, buf
, 4096) != 4096)
1167 if (buf
[0] != '\0' && buf
[0] != '\x5a' && buf
[0] != '\xff')
1169 if (memcmp(buf
, buf
+1, 4095) != 0)
1175 /* adding a spare to a regular array is quite different from adding one to
1176 * a set-of-partitions virtual array.
1177 * This function determines which is worth trying and tries as appropriate.
1178 * Arrays are given priority over partitions.
1180 static int try_spare(char *devname
, int *dfdp
, struct dev_policy
*pol
,
1181 struct map_ent
*target
,
1182 struct supertype
*st
, int verbose
)
1187 int partitions_ok
= 0;
1191 /* Can only add a spare if device has at least one domain */
1192 if (pol_find(pol
, pol_domain
) == NULL
)
1194 /* And only if some action allows spares */
1195 if (!policy_action_allows(pol
, st
?st
->ss
->name
:NULL
, act_spare
))
1198 /* Now check if the device is bare.
1199 * bare devices can always be added as a spare
1200 * non-bare devices can only be added if spare-same-slot is permitted,
1201 * and this device is replacing a previous device - in which case 'target'
1204 if (!is_bare(dfd
)) {
1205 /* Must have a target and allow same_slot */
1206 /* Later - may allow force_spare without target */
1208 !policy_action_allows(pol
, st
?st
->ss
->name
:NULL
,
1209 act_spare_same_slot
)) {
1211 pr_err("%s is not bare, so not "
1212 "considering as a spare\n",
1220 /* It might be OK to add this device to an array - need to see
1221 * what arrays might be candidates.
1224 /* just try try 'array' or 'partition' based on this metadata */
1225 if (st
->ss
->add_to_super
)
1226 return array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1229 return partition_try_spare(devname
, dfdp
, pol
,
1232 /* No metadata was specified or found so options are open.
1233 * Check for whether any array metadata, or any partition metadata
1234 * might allow adding the spare. This check is just help to avoid
1235 * a more costly scan of all arrays when we can be sure that will
1238 for (i
= 0; (!arrays_ok
|| !partitions_ok
) && superlist
[i
] ; i
++) {
1239 if (superlist
[i
]->add_to_super
&& !arrays_ok
&&
1240 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1242 if (superlist
[i
]->add_to_super
== NULL
&& !partitions_ok
&&
1243 policy_action_allows(pol
, superlist
[i
]->name
, act_spare
))
1248 rv
= array_try_spare(devname
, dfdp
, pol
, target
, bare
,
1250 if (rv
!= 0 && partitions_ok
)
1251 rv
= partition_try_spare(devname
, dfdp
, pol
, st
, verbose
);
1255 int IncrementalScan(int verbose
)
1257 /* look at every device listed in the 'map' file.
1258 * If one is found that is not running then:
1259 * look in mdadm.conf for bitmap file.
1260 * if one exists, but array has none, add it.
1261 * try to start array in auto-readonly mode
1263 struct map_ent
*mapl
= NULL
;
1265 struct mddev_ident
*devs
, *mddev
;
1269 devs
= conf_get_ident(NULL
);
1271 for (me
= mapl
; me
; me
= me
->next
) {
1272 mdu_array_info_t array
;
1273 mdu_bitmap_file_t bmf
;
1275 int mdfd
= open_dev(me
->devnum
);
1279 if (ioctl(mdfd
, GET_ARRAY_INFO
, &array
) == 0 ||
1284 /* Ok, we can try this one. Maybe it needs a bitmap */
1285 for (mddev
= devs
; mddev
; mddev
= mddev
->next
)
1286 if (mddev
->devname
&& me
->path
1287 && devname_matches(mddev
->devname
, me
->path
))
1289 if (mddev
&& mddev
->bitmap_file
) {
1291 * Note: early kernels will wrongly fail this, so it
1295 if (ioctl(mdfd
, GET_ARRAY_INFO
, &bmf
) < 0) {
1296 int bmfd
= open(mddev
->bitmap_file
, O_RDWR
);
1298 added
= ioctl(mdfd
, SET_BITMAP_FILE
,
1305 pr_err("Added bitmap %s to %s\n",
1306 mddev
->bitmap_file
, me
->path
);
1307 else if (errno
!= EEXIST
)
1308 pr_err("Failed to add bitmap to %s: %s\n",
1309 me
->path
, strerror(errno
));
1312 /* FIXME check for reshape_active and consider not
1315 sra
= sysfs_read(mdfd
, 0, 0);
1317 if (sysfs_set_str(sra
, NULL
,
1318 "array_state", "read-auto") == 0) {
1320 pr_err("started array %s\n",
1321 me
->path
?: devnum2devname(me
->devnum
));
1323 pr_err("failed to start array %s: %s\n",
1324 me
->path
?: devnum2devname(me
->devnum
),
1334 static char *container2devname(char *devname
)
1336 char *mdname
= NULL
;
1338 if (devname
[0] == '/') {
1339 int fd
= open(devname
, O_RDONLY
);
1341 mdname
= devnum2devname(fd2devnum(fd
));
1346 struct map_ent
*mp
, *map
= NULL
;
1348 if (!parse_uuid(devname
, uuid
))
1350 mp
= map_by_uuid(&map
, uuid
);
1352 mdname
= devnum2devname(mp
->devnum
);
1359 static int Incremental_container(struct supertype
*st
, char *devname
,
1362 /* Collect the contents of this container and for each
1363 * array, choose a device name and assemble the array.
1366 struct mdinfo
*list
;
1368 struct map_ent
*map
= NULL
;
1371 struct mddev_ident
*match
;
1373 struct domainlist
*domains
;
1374 struct map_ent
*smp
;
1380 st
->ss
->getinfo_super(st
, &info
, NULL
);
1382 if ((c
->runstop
> 0 && info
.container_enough
>= 0) ||
1383 info
.container_enough
> 0)
1387 pr_err("not enough devices to start the container\n");
1391 match
= conf_match(st
, &info
, devname
, c
->verbose
, &rv
);
1392 if (match
== NULL
&& rv
== 2)
1395 /* Need to compute 'trustworthy' */
1397 trustworthy
= LOCAL
;
1398 else if (st
->ss
->match_home(st
, c
->homehost
) == 1)
1399 trustworthy
= LOCAL
;
1400 else if (st
->ss
->match_home(st
, "any") == 1)
1401 trustworthy
= LOCAL
;
1403 trustworthy
= FOREIGN
;
1405 list
= st
->ss
->container_content(st
, NULL
);
1406 /* when nothing to activate - quit */
1409 for (ra
= list
; ra
; ra
= ra
->next
) {
1411 char chosen_name
[1024];
1413 struct mddev_ident
*match
= NULL
;
1416 /* do not activate arrays blocked by metadata handler */
1417 if (ra
->array
.state
& (1 << MD_SB_BLOCK_VOLUME
)) {
1418 pr_err("Cannot activate array %s in %s.\n",
1419 ra
->text_version
, devname
);
1423 mp
= map_by_uuid(&map
, ra
->uuid
);
1426 mdfd
= open_dev(mp
->devnum
);
1428 strcpy(chosen_name
, mp
->path
);
1430 strcpy(chosen_name
, devnum2devname(mp
->devnum
));
1433 /* Check in mdadm.conf for container == devname and
1434 * member == ra->text_version after second slash.
1436 char *sub
= strchr(ra
->text_version
+1, '/');
1437 struct mddev_ident
*array_list
;
1440 array_list
= conf_get_ident(NULL
);
1443 for(; array_list
; array_list
= array_list
->next
) {
1445 if (array_list
->member
== NULL
||
1446 array_list
->container
== NULL
)
1448 if (strcmp(array_list
->member
, sub
) != 0)
1450 if (array_list
->uuid_set
&&
1451 !same_uuid(ra
->uuid
, array_list
->uuid
, st
->ss
->swapuuid
))
1453 dn
= container2devname(array_list
->container
);
1456 if (strncmp(dn
, ra
->text_version
+1,
1458 ra
->text_version
[strlen(dn
)+1] != '/') {
1463 /* we have a match */
1466 pr_err("match found for member %s\n",
1467 array_list
->member
);
1471 if (match
&& match
->devname
&&
1472 strcasecmp(match
->devname
, "<ignore>") == 0) {
1474 pr_err("array %s/%s is "
1475 "explicitly ignored by mdadm.conf\n",
1476 match
->container
, match
->member
);
1480 trustworthy
= LOCAL
;
1482 mdfd
= create_mddev(match
? match
->devname
: NULL
,
1490 pr_err("failed to open %s: %s.\n",
1491 chosen_name
, strerror(errno
));
1495 assemble_container_content(st
, mdfd
, ra
, c
,
1500 /* don't move spares to container with volume being activated
1501 when all volumes are blocked */
1502 if (ra_all
== ra_blocked
)
1505 /* Now move all suitable spares from spare container */
1506 domains
= domain_from_array(list
, st
->ss
->name
);
1507 memcpy(suuid
, uuid_zero
, sizeof(int[4]));
1509 (smp
= map_by_uuid(&map
, suuid
)) != NULL
&&
1510 (sfd
= open(smp
->path
, O_RDONLY
)) >= 0) {
1511 /* spare container found */
1512 struct supertype
*sst
=
1513 super_imsm
.match_metadata_desc("imsm");
1514 struct mdinfo
*sinfo
;
1515 unsigned long long min_size
= 0;
1516 if (st
->ss
->min_acceptable_spare_size
)
1517 min_size
= st
->ss
->min_acceptable_spare_size(st
);
1518 if (!sst
->ss
->load_container(sst
, sfd
, NULL
)) {
1520 sinfo
= container_choose_spares(sst
, min_size
,
1523 sst
->ss
->free_super(sst
);
1526 struct mdinfo
*disks
= sinfo
->devs
;
1528 /* move spare from spare
1529 * container to currently
1535 makedev(disks
->disk
.major
,
1536 disks
->disk
.minor
)))
1538 disks
= disks
->next
;
1541 pr_err("Added %d spare%s to %s\n",
1542 count
, count
>1?"s":"", devname
);
1548 domain_free(domains
);
1553 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1554 * raid arrays, and if so first fail (if needed) and then remove the device.
1556 * @devname - The device we want to remove
1557 * @id_path - name as found in /dev/disk/by-path for this device
1559 * Note: the device name must be a kernel name like "sda", so
1560 * that we can find it in /proc/mdstat
1562 int IncrementalRemove(char *devname
, char *id_path
, int verbose
)
1566 struct mdstat_ent
*ent
;
1567 struct mddev_dev devlist
;
1570 dprintf(Name
": incremental removal without --path <id_path> "
1571 "lacks the possibility to re-add new device in this "
1574 if (strchr(devname
, '/')) {
1575 pr_err("incremental removal requires a "
1576 "kernel device name, not a file: %s\n", devname
);
1579 ent
= mdstat_by_component(devname
);
1581 pr_err("%s does not appear to be a component "
1582 "of any array\n", devname
);
1585 mdfd
= open_dev(ent
->devnum
);
1587 pr_err("Cannot open array %s!!\n", ent
->dev
);
1593 struct map_ent
*map
= NULL
, *me
;
1594 me
= map_by_devnum(&map
, ent
->devnum
);
1596 policy_save_path(id_path
, me
);
1600 memset(&devlist
, 0, sizeof(devlist
));
1601 devlist
.devname
= devname
;
1602 devlist
.disposition
= 'f';
1603 /* for a container, we must fail each member array */
1604 if (ent
->metadata_version
&&
1605 strncmp(ent
->metadata_version
, "external:", 9) == 0) {
1606 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
1607 struct mdstat_ent
*memb
;
1608 for (memb
= mdstat
; memb
; memb
= memb
->next
)
1609 if (is_container_member(memb
, ent
->dev
)) {
1610 int subfd
= open_dev(memb
->devnum
);
1612 Manage_subdevs(memb
->dev
, subfd
,
1613 &devlist
, verbose
, 0,
1618 free_mdstat(mdstat
);
1620 Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
, 0);
1621 devlist
.disposition
= 'r';
1622 rv
= Manage_subdevs(ent
->dev
, mdfd
, &devlist
, verbose
, 0, NULL
, 0);