2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
31 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
32 #error no endian defined
37 int restore_backup(struct supertype
*st
,
38 struct mdinfo
*content
,
48 int disk_count
= next_spare
+ working_disks
;
49 char *backup_file
= *backup_filep
;
51 dprintf("Called restore_backup()\n");
52 fdlist
= xmalloc(sizeof(int) * disk_count
);
54 enable_fds(next_spare
);
55 for (i
= 0; i
< next_spare
; i
++)
57 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
61 sprintf(buf
, "%d:%d", dev
->disk
.major
, dev
->disk
.minor
);
62 fd
= dev_open(buf
, O_RDWR
);
64 if (dev
->disk
.raid_disk
>= 0)
65 fdlist
[dev
->disk
.raid_disk
] = fd
;
67 fdlist
[next_spare
++] = fd
;
71 backup_file
= locate_backup(content
->sys_name
);
72 *backup_filep
= backup_file
;
75 if (st
->ss
->external
&& st
->ss
->recover_backup
)
76 err
= st
->ss
->recover_backup(st
, content
);
78 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
79 backup_file
, verbose
> 0);
81 while (next_spare
> 0) {
83 if (fdlist
[next_spare
] >= 0)
84 close(fdlist
[next_spare
]);
88 pr_err("Failed to restore critical section for reshape - sorry.\n");
90 pr_err("Possibly you need to specify a --backup-file\n");
94 dprintf("restore_backup() returns status OK.\n");
98 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
100 /* Add a device to an active array.
101 * Currently, just extend a linear array.
102 * This requires writing a new superblock on the
103 * new device, calling the kernel to add the device,
104 * and if that succeeds, update the superblock on
106 * This means that we need to *find* all other devices.
113 struct supertype
*st
= NULL
;
114 char *subarray
= NULL
;
116 if (md_get_array_info(fd
, &info
.array
) < 0) {
117 pr_err("cannot get array info for %s\n", devname
);
121 if (info
.array
.level
!= -1) {
122 pr_err("can only add devices to linear arrays\n");
126 st
= super_by_fd(fd
, &subarray
);
128 pr_err("cannot handle arrays with superblock version %d\n",
129 info
.array
.major_version
);
134 pr_err("Cannot grow linear sub-arrays yet\n");
140 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
142 pr_err("cannot open %s\n", newdev
);
146 if (!fstat_is_blkdev(nfd
, newdev
, &rdev
)) {
151 /* now check out all the devices and make sure we can read the
153 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
154 mdu_disk_info_t disk
;
157 st
->ss
->free_super(st
);
160 if (md_get_disk_info(fd
, &disk
) < 0) {
161 pr_err("cannot get device detail for device %d\n", d
);
166 dv
= map_dev(disk
.major
, disk
.minor
, 1);
168 pr_err("cannot find device file for device %d\n", d
);
173 fd2
= dev_open(dv
, O_RDWR
);
175 pr_err("cannot open device file %s\n", dv
);
181 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
182 pr_err("cannot find super block on %s\n", dv
);
190 /* Ok, looks good. Lets update the superblock and write it out to
194 info
.disk
.number
= d
;
195 info
.disk
.major
= major(rdev
);
196 info
.disk
.minor
= minor(rdev
);
197 info
.disk
.raid_disk
= d
;
198 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
199 if (st
->ss
->update_super(st
, &info
, UOPT_SPEC_LINEAR_GROW_NEW
, newdev
,
201 pr_err("Preparing new metadata failed on %s\n", newdev
);
206 if (st
->ss
->store_super(st
, nfd
)) {
207 pr_err("Cannot store new superblock on %s\n", newdev
);
213 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
214 pr_err("Cannot add new disk to this array\n");
217 /* Well, that seems to have worked.
218 * Now go through and update all superblocks
221 if (md_get_array_info(fd
, &info
.array
) < 0) {
222 pr_err("cannot get array info for %s\n", devname
);
227 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
228 mdu_disk_info_t disk
;
232 if (md_get_disk_info(fd
, &disk
) < 0) {
233 pr_err("cannot get device detail for device %d\n", d
);
236 dv
= map_dev(disk
.major
, disk
.minor
, 1);
238 pr_err("cannot find device file for device %d\n", d
);
241 fd2
= dev_open(dv
, O_RDWR
);
243 pr_err("cannot open device file %s\n", dv
);
246 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
247 pr_err("cannot find super block on %s\n", dv
);
252 info
.array
.raid_disks
= nd
+1;
253 info
.array
.nr_disks
= nd
+1;
254 info
.array
.active_disks
= nd
+1;
255 info
.array
.working_disks
= nd
+1;
257 if (st
->ss
->update_super(st
, &info
, UOPT_SPEC_LINEAR_GROW_UPDATE
, dv
,
259 pr_err("Updating metadata failed on %s\n", dv
);
264 if (st
->ss
->store_super(st
, fd2
)) {
265 pr_err("Cannot store new superblock on %s\n", dv
);
275 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
286 mdu_bitmap_file_t bmf
;
287 mdu_array_info_t array
;
288 struct supertype
*st
;
289 char *subarray
= NULL
;
290 int major
= BITMAP_MAJOR_HI
;
291 unsigned long long bitmapsize
, array_size
;
295 * We only ever get called if s->bitmap_file is != NULL, so this check
296 * is just here to quiet down static code checkers.
301 if (strcmp(s
->bitmap_file
, "clustered") == 0)
302 major
= BITMAP_MAJOR_CLUSTERED
;
304 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
306 pr_err("Memory allocation failure.\n");
308 pr_err("bitmaps not supported by this kernel.\n");
311 if (bmf
.pathname
[0]) {
312 if (str_is_none(s
->bitmap_file
) == true) {
313 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
314 pr_err("failed to remove bitmap %s\n",
320 pr_err("%s already has a bitmap (%s)\n", devname
, bmf
.pathname
);
323 if (md_get_array_info(fd
, &array
) != 0) {
324 pr_err("cannot get array status for %s\n", devname
);
327 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
328 if (str_is_none(s
->bitmap_file
) == true) {
329 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
330 if (md_set_array_info(fd
, &array
) != 0) {
331 if (array
.state
& (1 << MD_SB_CLUSTERED
))
332 pr_err("failed to remove clustered bitmap.\n");
334 pr_err("failed to remove internal bitmap.\n");
339 pr_err("bitmap already present on %s\n", devname
);
343 if (str_is_none(s
->bitmap_file
) == true) {
344 pr_err("no bitmap found on %s\n", devname
);
347 if (array
.level
<= 0) {
348 pr_err("Bitmaps not meaningful with level %s\n",
349 map_num(pers
, array
.level
)?:"of this array");
352 bitmapsize
= array
.size
;
354 if (get_dev_size(fd
, NULL
, &array_size
) &&
355 array_size
> (0x7fffffffULL
<< 9)) {
356 /* Array is big enough that we cannot trust array.size
357 * try other approaches
359 bitmapsize
= get_component_size(fd
);
361 if (bitmapsize
== 0) {
362 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
366 if (array
.level
== 10) {
369 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
370 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
372 if (strcmp(s
->bitmap_file
, "clustered") == 0 &&
373 !is_near_layout_10(array
.layout
)) {
374 pr_err("only near layout is supported with clustered raid10\n");
379 st
= super_by_fd(fd
, &subarray
);
381 pr_err("Cannot understand version %d.%d\n",
382 array
.major_version
, array
.minor_version
);
386 pr_err("Cannot add bitmaps to sub-arrays yet\n");
392 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
394 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
395 pr_err("Cannot add bitmap to array with PPL\n");
403 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
404 strcmp(s
->bitmap_file
, "clustered") == 0) {
407 int offset_setable
= 0;
408 if (st
->ss
->add_internal_bitmap
== NULL
) {
409 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
412 st
->nodes
= c
->nodes
;
413 st
->cluster_name
= c
->homecluster
;
414 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
417 for (d
= 0; d
< st
->max_devs
; d
++) {
418 mdu_disk_info_t disk
;
423 if (md_get_disk_info(fd
, &disk
) < 0)
425 if (disk
.major
== 0 && disk
.minor
== 0)
427 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
429 dv
= map_dev(disk
.major
, disk
.minor
, 1);
432 if ((disk
.state
& (1 << MD_DISK_WRITEMOSTLY
)) &&
433 (strcmp(s
->bitmap_file
, "clustered") == 0)) {
434 pr_err("%s disks marked write-mostly are not supported with clustered bitmap\n",devname
);
438 fd2
= dev_open(dv
, O_RDWR
);
441 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
443 rv
= st
->ss
->add_internal_bitmap(
444 st
, &s
->bitmap_chunk
, c
->delay
,
445 s
->write_behind
, bitmapsize
,
446 offset_setable
, major
);
448 st
->ss
->write_bitmap(st
, fd2
,
451 pr_err("failed to create internal bitmap - chunksize problem.\n");
454 pr_err("failed to load super-block.\n");
462 if (offset_setable
) {
463 st
->ss
->getinfo_super(st
, mdi
, NULL
);
464 if (sysfs_init(mdi
, fd
, NULL
)) {
465 pr_err("failed to initialize sysfs.\n");
468 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
472 if (strcmp(s
->bitmap_file
, "clustered") == 0)
473 array
.state
|= (1 << MD_SB_CLUSTERED
);
474 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
475 rv
= md_set_array_info(fd
, &array
);
479 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
480 pr_err("failed to set internal bitmap.\n");
487 int max_devs
= st
->max_devs
;
489 /* try to load a superblock */
490 for (d
= 0; d
< max_devs
; d
++) {
491 mdu_disk_info_t disk
;
495 if (md_get_disk_info(fd
, &disk
) < 0)
497 if ((disk
.major
==0 && disk
.minor
== 0) ||
498 (disk
.state
& (1 << MD_DISK_REMOVED
)))
500 dv
= map_dev(disk
.major
, disk
.minor
, 1);
503 fd2
= dev_open(dv
, O_RDONLY
);
505 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
507 st
->ss
->uuid_from_super(st
, uuid
);
514 pr_err("cannot find UUID for array!\n");
517 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
518 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
519 bitmapsize
, major
)) {
522 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
524 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
527 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
530 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
531 pr_err("Cannot set bitmap file for %s: %s\n",
532 devname
, strerror(err
));
540 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
542 struct supertype
*st
;
545 char *subarray
= NULL
;
547 char container_dev
[PATH_MAX
];
548 char buf
[SYSFS_MAX_BUF_SIZE
];
550 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
551 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
552 pr_err("Operation not supported for consistency policy %s\n",
553 map_num_s(consistency_policies
, s
->consistency_policy
));
557 st
= super_by_fd(fd
, &subarray
);
561 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
568 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
569 !st
->ss
->write_init_ppl
) {
570 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
575 if (sra
->array
.level
!= 5) {
576 pr_err("Operation not supported for array level %d\n",
582 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
583 pr_err("Consistency policy is already %s\n",
584 map_num_s(consistency_policies
, s
->consistency_policy
));
587 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
588 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
589 pr_err("Current consistency policy is %s, cannot change to %s\n",
590 map_num_s(consistency_policies
, sra
->consistency_policy
),
591 map_num_s(consistency_policies
, s
->consistency_policy
));
596 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
597 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, sizeof(buf
)) <= 0) {
600 } else if (strcmp(buf
, "reshape\n") == 0) {
601 pr_err("PPL cannot be enabled when reshape is in progress\n");
608 enum update_opt update
;
610 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
613 update
= UOPT_NO_PPL
;
615 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
617 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
623 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
630 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
632 pr_err("Failed to open %s\n", container_dev
);
637 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
641 pr_err("Cannot read superblock for %s\n",
646 mdi
= st
->ss
->container_content(st
, subarray
);
651 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
655 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
656 dfd
= dev_open(devpath
, O_RDWR
);
658 pr_err("Failed to open %s\n", devpath
);
664 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
666 pr_err("Failed to load super-block.\n");
671 ret
= st
->ss
->update_super(st
, sra
, UOPT_PPL
,
673 c
->verbose
, 0, NULL
);
676 st
->ss
->free_super(st
);
679 st
->ss
->getinfo_super(st
, &info
, NULL
);
682 ret
|= sysfs_set_num(sra
, sd
, "ppl_sector",
684 ret
|= sysfs_set_num(sra
, sd
, "ppl_size",
688 pr_err("Failed to set PPL attributes for %s\n",
691 st
->ss
->free_super(st
);
695 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
697 pr_err("Failed to write PPL\n");
702 st
->ss
->free_super(st
);
709 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
710 map_num_s(consistency_policies
,
711 s
->consistency_policy
));
713 pr_err("Failed to change array consistency policy\n");
725 * When reshaping an array we might need to backup some data.
726 * This is written to all spares with a 'super_block' describing it.
727 * The superblock goes 4K from the end of the used space on the
729 * It if written after the backup is complete.
730 * It has the following structure.
733 static struct mdp_backup_super
{
734 char magic
[16]; /* md_backup_data-1 or -2 */
737 /* start/sizes in 512byte sectors */
738 __u64 devstart
; /* address on backup device/file of data */
741 __u32 sb_csum
; /* csum of preceeding bytes. */
743 __u64 devstart2
; /* offset in to data of second section */
746 __u32 sb_csum2
; /* csum of preceeding bytes. */
748 } __attribute__((aligned(512))) bsb
, bsb2
;
750 static __u32
bsb_csum(char *buf
, int len
)
754 for (i
= 0; i
< len
; i
++)
755 csum
= (csum
<<3) + buf
[0];
756 return __cpu_to_le32(csum
);
759 static int check_idle(struct supertype
*st
)
761 /* Check that all member arrays for this container, or the
762 * container of this array, are idle
764 char *container
= (st
->container_devnm
[0]
765 ? st
->container_devnm
: st
->devnm
);
766 struct mdstat_ent
*ent
, *e
;
769 ent
= mdstat_read(0, 0);
770 for (e
= ent
; e
; e
= e
->next
) {
771 if (!is_container_member(e
, container
))
773 /* frozen array is not idle*/
774 if (e
->percent
>= 0 || e
->metadata_version
[9] == '-') {
783 static int freeze_container(struct supertype
*st
)
785 char *container
= (st
->container_devnm
[0]
786 ? st
->container_devnm
: st
->devnm
);
791 if (block_monitor(container
, 1)) {
792 pr_err("failed to freeze container\n");
799 static void unfreeze_container(struct supertype
*st
)
801 char *container
= (st
->container_devnm
[0]
802 ? st
->container_devnm
: st
->devnm
);
804 unblock_monitor(container
, 1);
807 static int freeze(struct supertype
*st
)
809 /* Try to freeze resync/rebuild on this array/container.
810 * Return -1 if the array is busy,
811 * return -2 container cannot be frozen,
812 * return 0 if this kernel doesn't support 'frozen'
813 * return 1 if it worked.
815 if (st
->ss
->external
)
816 return freeze_container(st
);
818 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
820 char buf
[SYSFS_MAX_BUF_SIZE
];
824 /* Need to clear any 'read-auto' status */
825 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, sizeof(buf
)) > 0 &&
826 strncmp(buf
, "read-auto", 9) == 0)
827 sysfs_set_str(sra
, NULL
, "array_state", "clean");
829 err
= sysfs_freeze_array(sra
);
835 static void unfreeze(struct supertype
*st
)
837 if (st
->ss
->external
)
838 return unfreeze_container(st
);
840 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
841 char buf
[SYSFS_MAX_BUF_SIZE
];
844 sysfs_get_str(sra
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0 &&
845 strcmp(buf
, "frozen\n") == 0)
846 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
851 static void wait_reshape(struct mdinfo
*sra
)
853 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
854 char action
[SYSFS_MAX_BUF_SIZE
];
859 while (sysfs_fd_get_str(fd
, action
, sizeof(action
)) > 0 &&
860 strncmp(action
, "reshape", 7) == 0)
861 sysfs_wait(fd
, NULL
);
865 static int reshape_super(struct supertype
*st
, unsigned long long size
,
866 int level
, int layout
, int chunksize
, int raid_disks
,
867 int delta_disks
, char *backup_file
, char *dev
,
868 int direction
, int verbose
)
870 /* nothing extra to check in the native case */
871 if (!st
->ss
->external
)
873 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
874 pr_err("%s metadata does not support reshape\n",
879 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
880 raid_disks
, delta_disks
, backup_file
, dev
,
884 static void sync_metadata(struct supertype
*st
)
886 if (st
->ss
->external
) {
887 if (st
->update_tail
) {
888 flush_metadata_updates(st
);
889 st
->update_tail
= &st
->updates
;
891 st
->ss
->sync_metadata(st
);
895 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
897 /* when dealing with external metadata subarrays we need to be
898 * prepared to handle EAGAIN. The kernel may need to wait for
899 * mdmon to mark the array active so the kernel can handle
900 * allocations/writeback when preparing the reshape action
901 * (md_allow_write()). We temporarily disable safe_mode_delay
902 * to close a race with the array_state going clean before the
903 * next write to raid_disks / stripe_cache_size
905 char safe
[SYSFS_MAX_BUF_SIZE
];
908 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
910 (strcmp(name
, "raid_disks") != 0 &&
911 strcmp(name
, "stripe_cache_size") != 0))
912 return sysfs_set_num(sra
, NULL
, name
, n
);
914 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
917 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
918 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
919 if (rc
< 0 && errno
== EAGAIN
) {
920 ping_monitor(container
);
921 /* if we get EAGAIN here then the monitor is not active
924 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
926 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
930 int start_reshape(struct mdinfo
*sra
, int already_running
,
931 int before_data_disks
, int data_disks
, struct supertype
*st
)
934 unsigned long long sync_max_to_set
;
936 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
937 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
938 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
939 sra
->reshape_progress
);
940 if (before_data_disks
<= data_disks
)
941 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
943 sync_max_to_set
= (sra
->component_size
* data_disks
944 - sra
->reshape_progress
) / data_disks
;
946 if (!already_running
)
947 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
949 if (st
->ss
->external
)
950 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
952 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_max", "max");
954 if (!already_running
&& err
== 0) {
957 err
= sysfs_set_str(sra
, NULL
, "sync_action",
960 sleep_for(1, 0, true);
961 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
966 void abort_reshape(struct mdinfo
*sra
)
968 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
970 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
971 * suspend_hi to decrease as well as increase.")
972 * you could only increase suspend_{lo,hi} unless the region they
973 * covered was empty. So to reset to 0, you need to push suspend_lo
974 * up past suspend_hi first. So to maximize the chance of mdadm
975 * working on all kernels, we want to keep doing that.
977 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
978 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
979 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
980 sysfs_set_num(sra
, NULL
, "sync_min", 0);
981 // It isn't safe to reset sync_max as we aren't monitoring.
982 // Array really should be stopped at this point.
985 int remove_disks_for_takeover(struct supertype
*st
,
990 struct mdinfo
*remaining
;
993 if (st
->ss
->external
) {
995 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
997 * containter_content returns list of arrays in container
998 * If arrays->next is not NULL it means that there are
999 * 2 arrays in container and operation should be blocked
1006 pr_err("Error. Cannot perform operation on %s- for this operation "
1007 "it MUST be single array in container\n", st
->devnm
);
1013 if (sra
->array
.level
== 10)
1014 nr_of_copies
= layout
& 0xff;
1015 else if (sra
->array
.level
== 1)
1016 nr_of_copies
= sra
->array
.raid_disks
;
1020 remaining
= sra
->devs
;
1022 /* for each 'copy', select one device and remove from the list. */
1023 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1024 struct mdinfo
**diskp
;
1027 /* Find a working device to keep */
1028 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1029 struct mdinfo
*disk
= *diskp
;
1031 if (disk
->disk
.raid_disk
< slot
)
1033 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1035 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1037 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1039 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1042 /* We have found a good disk to use! */
1043 *diskp
= disk
->next
;
1044 disk
->next
= sra
->devs
;
1053 if (slot
< sra
->array
.raid_disks
) {
1054 /* didn't find all slots */
1060 sra
->devs
= remaining
;
1064 /* Remove all 'remaining' devices from the array */
1066 struct mdinfo
*sd
= remaining
;
1067 remaining
= sd
->next
;
1069 sysfs_set_str(sra
, sd
, "state", "faulty");
1070 sysfs_set_str(sra
, sd
, "slot", STR_COMMON_NONE
);
1071 /* for external metadata disks should be removed in mdmon */
1072 if (!st
->ss
->external
)
1073 sysfs_set_str(sra
, sd
, "state", "remove");
1074 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1075 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1076 sd
->next
= sra
->devs
;
1082 void reshape_free_fdlist(int *fdlist
,
1083 unsigned long long *offsets
,
1088 for (i
= 0; i
< size
; i
++)
1096 int reshape_prepare_fdlist(char *devname
,
1100 unsigned long blocks
,
1103 unsigned long long *offsets
)
1108 enable_fds(nrdisks
);
1109 for (d
= 0; d
<= nrdisks
; d
++)
1112 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1113 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1115 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1116 sd
->disk
.raid_disk
< raid_disks
) {
1117 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1118 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1119 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1120 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1121 pr_err("%s: cannot open component %s\n",
1122 devname
, dn
? dn
: "-unknown-");
1126 } else if (backup_file
== NULL
) {
1128 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1129 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1130 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1131 if (fdlist
[d
] < 0) {
1132 pr_err("%s: cannot open component %s\n",
1133 devname
, dn
? dn
: "-unknown-");
1144 int reshape_open_backup_file(char *backup_file
,
1149 unsigned long long *offsets
,
1153 /* Return 1 on success, 0 on any form of failure */
1154 /* need to check backup file is large enough */
1160 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1164 pr_err("%s: cannot create backup file %s: %s\n",
1165 devname
, backup_file
, strerror(errno
));
1168 /* Guard against backup file being on array device.
1169 * If array is partitioned or if LVM etc is in the
1170 * way this will not notice, but it is better than
1173 fstat(*fdlist
, &stb
);
1176 if (stb
.st_rdev
== dev
) {
1177 pr_err("backup file must NOT be on the array being reshaped.\n");
1182 memset(buf
, 0, 512);
1183 for (i
=0; i
< blocks
+ 8 ; i
++) {
1184 if (write(*fdlist
, buf
, 512) != 512) {
1185 pr_err("%s: cannot create backup file %s: %s\n",
1186 devname
, backup_file
, strerror(errno
));
1190 if (fsync(*fdlist
) != 0) {
1191 pr_err("%s: cannot create backup file %s: %s\n",
1192 devname
, backup_file
, strerror(errno
));
1196 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1197 char *bu
= make_backup(sys_name
);
1198 if (symlink(backup_file
, bu
))
1199 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1207 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1208 unsigned int ndata
, unsigned int odata
)
1210 unsigned long a
, b
, blocks
;
1211 /* So how much do we need to backup.
1212 * We need an amount of data which is both a whole number of
1213 * old stripes and a whole number of new stripes.
1214 * So LCM for (chunksize*datadisks).
1216 a
= (ochunk
/512) * odata
;
1217 b
= (nchunk
/512) * ndata
;
1220 /* LCM == product / GCD */
1221 blocks
= (unsigned long)(ochunk
/512) * (unsigned long)(nchunk
/512) *
1227 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1229 /* Based on the current array state in info->array and
1230 * the changes in info->new_* etc, determine:
1231 * - whether the change is possible
1232 * - Intermediate level/raid_disks/layout
1233 * - whether a restriping reshape is needed
1234 * - number of sectors in minimum change unit. This
1235 * will cover a whole number of stripes in 'before' and
1238 * Return message if the change should be rejected
1239 * NULL if the change can be achieved
1241 * This can be called as part of starting a reshape, or
1242 * when assembling an array that is undergoing reshape.
1244 int near
, far
, offset
, copies
;
1246 int old_chunk
, new_chunk
;
1247 /* delta_parity records change in number of devices
1248 * caused by level change
1250 int delta_parity
= 0;
1252 memset(re
, 0, sizeof(*re
));
1254 /* If a new level not explicitly given, we assume no-change */
1255 if (info
->new_level
== UnSet
)
1256 info
->new_level
= info
->array
.level
;
1258 if (info
->new_chunk
)
1259 switch (info
->new_level
) {
1265 /* chunk size is meaningful, must divide component_size
1268 if (info
->component_size
% (info
->new_chunk
/512)) {
1269 unsigned long long shrink
= info
->component_size
;
1270 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1271 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1272 info
->new_chunk
/1024, info
->component_size
/2);
1273 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1275 pr_err("will shrink the array so the given chunk size would work.\n");
1280 return "chunk size not meaningful for this level";
1283 info
->new_chunk
= info
->array
.chunk_size
;
1285 switch (info
->array
.level
) {
1287 return "No reshape is possibly for this RAID level";
1289 if (info
->delta_disks
!= UnSet
)
1290 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1292 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1294 /* RAID1 can convert to RAID1 with different disks, or
1295 * raid5 with 2 disks, or
1298 if (info
->new_level
> 1 && (info
->component_size
& 7))
1299 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1300 if (info
->new_level
== 0) {
1301 if (info
->delta_disks
!= UnSet
&&
1302 info
->delta_disks
!= 0)
1303 return "Cannot change number of disks with RAID1->RAID0 conversion";
1305 re
->before
.data_disks
= 1;
1306 re
->after
.data_disks
= 1;
1309 if (info
->new_level
== 1) {
1310 if (info
->delta_disks
== UnSet
)
1311 /* Don't know what to do */
1312 return "no change requested for Growing RAID1";
1316 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1317 return "Can only convert a 2-device array to RAID5";
1318 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1320 re
->before
.data_disks
= 1;
1321 if (info
->delta_disks
!= UnSet
&&
1322 info
->delta_disks
!= 0)
1323 re
->after
.data_disks
= 1 + info
->delta_disks
;
1325 re
->after
.data_disks
= 1;
1326 if (re
->after
.data_disks
< 1)
1327 return "Number of disks too small for RAID5";
1329 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1330 info
->array
.chunk_size
= 65536;
1333 /* Could do some multi-stage conversions, but leave that to
1336 return "Impossibly level change request for RAID1";
1339 /* RAID10 can be converted from near mode to
1340 * RAID0 by removing some devices.
1341 * It can also be reshaped if the kernel supports
1344 switch (info
->new_level
) {
1346 if ((info
->array
.layout
& ~0xff) != 0x100)
1347 return "Cannot Grow RAID10 with far/offset layout";
1349 * number of devices must be multiple of
1352 if (info
->array
.raid_disks
%
1353 (info
->array
.layout
& 0xff))
1354 return "RAID10 layout too complex for Grow operation";
1356 new_disks
= (info
->array
.raid_disks
/
1357 (info
->array
.layout
& 0xff));
1358 if (info
->delta_disks
== UnSet
)
1359 info
->delta_disks
= (new_disks
1360 - info
->array
.raid_disks
);
1362 if (info
->delta_disks
!=
1363 new_disks
- info
->array
.raid_disks
)
1364 return "New number of raid-devices impossible for RAID10";
1365 if (info
->new_chunk
&&
1366 info
->new_chunk
!= info
->array
.chunk_size
)
1367 return "Cannot change chunk-size with RAID10 Grow";
1371 re
->before
.data_disks
= new_disks
;
1372 re
->after
.data_disks
= re
->before
.data_disks
;
1376 near
= info
->array
.layout
& 0xff;
1377 far
= (info
->array
.layout
>> 8) & 0xff;
1378 offset
= info
->array
.layout
& 0x10000;
1379 if (far
> 1 && !offset
)
1380 return "Cannot reshape RAID10 in far-mode";
1381 copies
= near
* far
;
1383 old_chunk
= info
->array
.chunk_size
* far
;
1385 if (info
->new_layout
== UnSet
)
1386 info
->new_layout
= info
->array
.layout
;
1388 near
= info
->new_layout
& 0xff;
1389 far
= (info
->new_layout
>> 8) & 0xff;
1390 offset
= info
->new_layout
& 0x10000;
1391 if (far
> 1 && !offset
)
1392 return "Cannot reshape RAID10 to far-mode";
1393 if (near
* far
!= copies
)
1394 return "Cannot change number of copies when reshaping RAID10";
1396 if (info
->delta_disks
== UnSet
)
1397 info
->delta_disks
= 0;
1398 new_disks
= (info
->array
.raid_disks
+
1401 new_chunk
= info
->new_chunk
* far
;
1404 re
->before
.layout
= info
->array
.layout
;
1405 re
->before
.data_disks
= info
->array
.raid_disks
;
1406 re
->after
.layout
= info
->new_layout
;
1407 re
->after
.data_disks
= new_disks
;
1408 /* For RAID10 we don't do backup but do allow reshape,
1409 * so set backup_blocks to INVALID_SECTORS rather than
1411 * And there is no need to synchronise stripes on both
1412 * 'old' and 'new'. So the important
1413 * number is the minimum data_offset difference
1414 * which is the larger of (offset copies * chunk).
1416 re
->backup_blocks
= INVALID_SECTORS
;
1417 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1418 if (new_disks
< re
->before
.data_disks
&&
1419 info
->space_after
< re
->min_offset_change
)
1420 /* Reduce component size by one chunk */
1421 re
->new_size
= (info
->component_size
-
1422 re
->min_offset_change
);
1424 re
->new_size
= info
->component_size
;
1425 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1429 return "RAID10 can only be changed to RAID0";
1432 /* RAID0 can be converted to RAID10, or to RAID456 */
1433 if (info
->new_level
== 10) {
1434 if (info
->new_layout
== UnSet
&&
1435 info
->delta_disks
== UnSet
) {
1436 /* Assume near=2 layout */
1437 info
->new_layout
= 0x102;
1438 info
->delta_disks
= info
->array
.raid_disks
;
1440 if (info
->new_layout
== UnSet
) {
1441 int copies
= 1 + (info
->delta_disks
1442 / info
->array
.raid_disks
);
1443 if (info
->array
.raid_disks
* (copies
-1) !=
1445 return "Impossible number of devices for RAID0->RAID10";
1446 info
->new_layout
= 0x100 + copies
;
1448 if (info
->delta_disks
== UnSet
) {
1449 int copies
= info
->new_layout
& 0xff;
1450 if (info
->new_layout
!= 0x100 + copies
)
1451 return "New layout impossible for RAID0->RAID10";;
1452 info
->delta_disks
= (copies
- 1) *
1453 info
->array
.raid_disks
;
1455 if (info
->new_chunk
&&
1456 info
->new_chunk
!= info
->array
.chunk_size
)
1457 return "Cannot change chunk-size with RAID0->RAID10";
1460 re
->before
.data_disks
= (info
->array
.raid_disks
+
1462 re
->after
.data_disks
= re
->before
.data_disks
;
1463 re
->before
.layout
= info
->new_layout
;
1467 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1468 * a raid4 style layout of the final level.
1470 switch (info
->new_level
) {
1475 re
->before
.layout
= 0;
1480 re
->before
.layout
= ALGORITHM_PARITY_N
;
1481 if (info
->new_layout
== UnSet
)
1482 info
->new_layout
= map_name(r5layout
, "default");
1487 re
->before
.layout
= ALGORITHM_PARITY_N
;
1488 if (info
->new_layout
== UnSet
)
1489 info
->new_layout
= map_name(r6layout
, "default");
1492 return "Impossible level change requested";
1494 re
->before
.data_disks
= info
->array
.raid_disks
;
1495 /* determining 'after' layout happens outside this 'switch' */
1499 info
->array
.layout
= ALGORITHM_PARITY_N
;
1501 switch (info
->new_level
) {
1505 re
->level
= info
->array
.level
;
1506 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1507 re
->before
.layout
= info
->array
.layout
;
1511 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1512 re
->before
.layout
= info
->array
.layout
;
1517 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1518 switch (info
->array
.layout
) {
1519 case ALGORITHM_LEFT_ASYMMETRIC
:
1520 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1522 case ALGORITHM_RIGHT_ASYMMETRIC
:
1523 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1525 case ALGORITHM_LEFT_SYMMETRIC
:
1526 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1528 case ALGORITHM_RIGHT_SYMMETRIC
:
1529 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1531 case ALGORITHM_PARITY_0
:
1532 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1534 case ALGORITHM_PARITY_N
:
1535 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1538 return "Cannot convert an array with this layout";
1542 if (info
->array
.raid_disks
!= 2)
1543 return "Can only convert a 2-device array to RAID1";
1544 if (info
->delta_disks
!= UnSet
&&
1545 info
->delta_disks
!= 0)
1546 return "Cannot set raid_disk when converting RAID5->RAID1";
1548 info
->new_chunk
= 0;
1551 return "Impossible level change requested";
1555 switch (info
->new_level
) {
1561 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1562 re
->before
.layout
= info
->array
.layout
;
1565 return "Impossible level change requested";
1570 /* If we reached here then it looks like a re-stripe is
1571 * happening. We have determined the intermediate level
1572 * and initial raid_disks/layout and stored these in 're'.
1574 * We need to deduce the final layout that can be atomically
1575 * converted to the end state.
1577 switch (info
->new_level
) {
1579 /* We can only get to RAID0 from RAID4 or RAID5
1580 * with appropriate layout and one extra device
1582 if (re
->level
!= 4 && re
->level
!= 5)
1583 return "Cannot covert to RAID0 from this level";
1585 switch (re
->level
) {
1587 re
->before
.layout
= 0;
1588 re
->after
.layout
= 0;
1591 re
->after
.layout
= ALGORITHM_PARITY_N
;
1597 /* We can only get to RAID4 from RAID5 */
1598 if (re
->level
!= 4 && re
->level
!= 5)
1599 return "Cannot convert to RAID4 from this level";
1601 switch (re
->level
) {
1603 re
->after
.layout
= 0;
1606 re
->after
.layout
= ALGORITHM_PARITY_N
;
1612 /* We get to RAID5 from RAID5 or RAID6 */
1613 if (re
->level
!= 5 && re
->level
!= 6)
1614 return "Cannot convert to RAID5 from this level";
1616 switch (re
->level
) {
1618 if (info
->new_layout
== UnSet
)
1619 re
->after
.layout
= re
->before
.layout
;
1621 re
->after
.layout
= info
->new_layout
;
1624 if (info
->new_layout
== UnSet
)
1625 info
->new_layout
= re
->before
.layout
;
1627 /* after.layout needs to be raid6 version of new_layout */
1628 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1629 re
->after
.layout
= ALGORITHM_PARITY_N
;
1632 char *ls
= map_num(r5layout
, info
->new_layout
);
1635 /* Current RAID6 layout has a RAID5
1638 strcat(strcpy(layout
, ls
), "-6");
1639 l
= map_name(r6layout
, layout
);
1641 return "Cannot find RAID6 layout to convert to";
1643 /* Current RAID6 has no equivalent.
1644 * If it is already a '-6' layout we
1645 * can leave it unchanged, else we must
1648 ls
= map_num(r6layout
,
1651 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1652 return "Please specify new layout";
1653 l
= info
->new_layout
;
1655 re
->after
.layout
= l
;
1661 /* We must already be at level 6 */
1663 return "Impossible level change";
1664 if (info
->new_layout
== UnSet
)
1665 re
->after
.layout
= info
->array
.layout
;
1667 re
->after
.layout
= info
->new_layout
;
1670 return "Impossible level change requested";
1672 if (info
->delta_disks
== UnSet
)
1673 info
->delta_disks
= delta_parity
;
1675 re
->after
.data_disks
=
1676 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1678 switch (re
->level
) {
1690 /* So we have a restripe operation, we need to calculate the number
1691 * of blocks per reshape operation.
1693 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1694 if (info
->new_chunk
== 0)
1695 info
->new_chunk
= info
->array
.chunk_size
;
1696 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1697 re
->after
.layout
== re
->before
.layout
&&
1698 info
->new_chunk
== info
->array
.chunk_size
) {
1699 /* Nothing to change, can change level immediately. */
1700 re
->level
= info
->new_level
;
1701 re
->backup_blocks
= 0;
1704 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1705 /* chunk and layout changes make no difference */
1706 re
->level
= info
->new_level
;
1707 re
->backup_blocks
= 0;
1711 re
->backup_blocks
= compute_backup_blocks(
1712 info
->new_chunk
, info
->array
.chunk_size
,
1713 re
->after
.data_disks
, re
->before
.data_disks
);
1714 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1716 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1720 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1723 struct mdinfo
*info
;
1727 if ((st
== NULL
) || (sra
== NULL
))
1730 if (text_version
== NULL
)
1731 text_version
= sra
->text_version
;
1732 subarray
= strchr(text_version
+ 1, '/')+1;
1733 info
= st
->ss
->container_content(st
, subarray
);
1735 unsigned long long current_size
= 0;
1736 unsigned long long new_size
= info
->custom_array_size
/2;
1738 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1739 new_size
> current_size
) {
1740 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1742 dprintf("Error: Cannot set array size");
1745 dprintf("Array size changed");
1747 dprintf_cont(" from %llu to %llu.\n",
1748 current_size
, new_size
);
1752 dprintf("Error: set_array_size(): info pointer in NULL\n");
1757 static int reshape_array(char *container
, int fd
, char *devname
,
1758 struct supertype
*st
, struct mdinfo
*info
,
1759 int force
, struct mddev_dev
*devlist
,
1760 unsigned long long data_offset
,
1761 char *backup_file
, int verbose
, int forked
,
1762 int restart
, int freeze_reshape
);
1763 static int reshape_container(char *container
, char *devname
,
1765 struct supertype
*st
,
1766 struct mdinfo
*info
,
1768 char *backup_file
, int verbose
,
1769 int forked
, int restart
, int freeze_reshape
);
1772 * prepare_external_reshape() - prepares update on external metadata if supported.
1773 * @devname: Device name.
1774 * @subarray: Subarray.
1776 * @container: Container.
1777 * @cfd: Container file descriptor.
1779 * Function checks that the requested reshape is supported on external metadata,
1780 * and performs an initial check that the container holds the pre-requisite
1781 * spare devices (mdmon owns final validation).
1783 * Return: 0 on success, else 1
1785 static int prepare_external_reshape(char *devname
, char *subarray
,
1786 struct supertype
*st
, char *container
,
1789 struct mdinfo
*cc
= NULL
;
1790 struct mdinfo
*content
= NULL
;
1792 if (st
->ss
->load_container(st
, cfd
, NULL
)) {
1793 pr_err("Cannot read superblock for %s\n", devname
);
1797 if (!st
->ss
->container_content
)
1800 cc
= st
->ss
->container_content(st
, subarray
);
1801 for (content
= cc
; content
; content
= content
->next
) {
1803 * check if reshape is allowed based on metadata
1804 * indications stored in content.array.status
1806 if (is_bit_set(&content
->array
.state
, MD_SB_BLOCK_VOLUME
) ||
1807 is_bit_set(&content
->array
.state
, MD_SB_BLOCK_CONTAINER_RESHAPE
)) {
1808 pr_err("Cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1809 devname
, container
);
1812 if (content
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
1813 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1816 if (content
->consistency_policy
== CONSISTENCY_POLICY_BITMAP
) {
1817 pr_err("Operation not supported when write-intent bitmap consistency policy is enabled\n");
1822 if (mdmon_running(container
))
1823 st
->update_tail
= &st
->updates
;
1830 int Grow_reshape(char *devname
, int fd
,
1831 struct mddev_dev
*devlist
,
1832 struct context
*c
, struct shape
*s
)
1834 /* Make some changes in the shape of an array.
1835 * The kernel must support the change.
1837 * There are three different changes. Each can trigger
1838 * a resync or recovery so we freeze that until we have
1839 * requested everything (if kernel supports freezing - 2.6.30).
1841 * - change size (i.e. component_size)
1843 * - change layout/chunksize/ndisks
1845 * The last can require a reshape. It is different on different
1846 * levels so we need to check the level before actioning it.
1847 * Some times the level change needs to be requested after the
1848 * reshape (e.g. raid6->raid5, raid5->raid0)
1851 struct mdu_array_info_s array
;
1853 struct supertype
*st
;
1854 char *subarray
= NULL
;
1858 char *container
= NULL
;
1861 struct mddev_dev
*dv
;
1865 struct mdinfo
*sra
= NULL
;
1867 if (md_get_array_info(fd
, &array
) < 0) {
1868 pr_err("%s is not an active md array - aborting\n",
1872 if (s
->level
!= UnSet
&& s
->chunk
) {
1873 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1877 if (s
->data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1878 (array
.level
< 4 || array
.level
> 6)) {
1879 pr_err("--grow --data-offset not yet supported\n");
1884 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1885 pr_err("cannot change component size at the same time as other changes.\n"
1886 " Change size first, then check data is intact before making other changes.\n");
1890 if (array
.level
> 1 && s
->size
> 1 &&
1891 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1892 pr_err("component size must be larger than chunk size.\n");
1896 st
= super_by_fd(fd
, &subarray
);
1898 pr_err("Unable to determine metadata format for %s\n", devname
);
1901 if (s
->raiddisks
> st
->max_devs
) {
1902 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1905 if (s
->level
== 0 && (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) &&
1906 !(array
.state
& (1 << MD_SB_CLUSTERED
)) && !st
->ss
->external
) {
1907 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
1908 if (md_set_array_info(fd
, &array
) != 0) {
1909 pr_err("failed to remove internal bitmap.\n");
1914 if (st
->ss
->external
) {
1916 container
= st
->container_devnm
;
1917 cfd
= open_dev_excl(st
->container_devnm
);
1919 container
= st
->devnm
;
1921 cfd
= open_dev_excl(st
->devnm
);
1925 pr_err("Unable to open container for %s\n", devname
);
1930 rv
= prepare_external_reshape(devname
, subarray
, st
,
1938 if (s
->raiddisks
&& subarray
) {
1939 pr_err("--raid-devices operation can be performed on a container only\n");
1947 for (dv
= devlist
; dv
; dv
= dv
->next
)
1949 if (s
->raiddisks
> array
.raid_disks
&&
1950 array
.spare_disks
+ added_disks
<
1951 (s
->raiddisks
- array
.raid_disks
) &&
1953 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1954 " Use --force to over-ride this check.\n",
1955 s
->raiddisks
- array
.raid_disks
,
1956 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1957 array
.spare_disks
+ added_disks
);
1961 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1962 GET_STATE
| GET_VERSION
);
1964 if (st
->ss
->external
&& subarray
== NULL
) {
1965 array
.level
= LEVEL_CONTAINER
;
1966 sra
->array
.level
= LEVEL_CONTAINER
;
1969 pr_err("failed to read sysfs parameters for %s\n",
1973 frozen
= freeze(st
);
1975 /* freeze() already spewed the reason */
1978 } else if (frozen
< 0) {
1979 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1984 /* ========= set size =============== */
1986 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1987 unsigned long long orig_size
= get_component_size(fd
)/2;
1988 unsigned long long min_csize
;
1990 int raid0_takeover
= 0;
1993 orig_size
= (unsigned) array
.size
;
1995 if (orig_size
== 0) {
1996 pr_err("Cannot set device size in this type of array.\n");
2001 if (array
.level
== 0) {
2002 pr_err("Component size change is not supported for RAID0\n");
2007 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
2008 devname
, APPLY_METADATA_CHANGES
,
2014 if (st
->ss
->external
) {
2015 /* metadata can have size limitation
2016 * update size value according to metadata information
2018 struct mdinfo
*sizeinfo
=
2019 st
->ss
->container_content(st
, subarray
);
2021 unsigned long long new_size
=
2022 sizeinfo
->custom_array_size
/2;
2023 int data_disks
= get_data_disks(
2024 sizeinfo
->array
.level
,
2025 sizeinfo
->array
.layout
,
2026 sizeinfo
->array
.raid_disks
);
2027 new_size
/= data_disks
;
2028 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
2029 orig_size
, new_size
,
2030 new_size
* data_disks
);
2032 sysfs_free(sizeinfo
);
2036 /* Update the size of each member device in case
2037 * they have been resized. This will never reduce
2038 * below the current used-size. The "size" attribute
2039 * understands '0' to mean 'max'.
2042 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2043 sysfs_set_num(sra
, mdi
, "size",
2044 s
->size
== MAX_SIZE
? 0 : s
->size
);
2045 if (array
.not_persistent
== 0 &&
2046 array
.major_version
== 0 &&
2047 get_linux_version() < 3001000) {
2048 /* Dangerous to allow size to exceed 2TB */
2049 unsigned long long csize
;
2050 if (sysfs_get_ll(sra
, mdi
, "size",
2052 if (csize
>= 2ULL*1024*1024*1024)
2053 csize
= 2ULL*1024*1024*1024;
2054 if ((min_csize
== 0 ||
2055 (min_csize
> csize
)))
2060 if (min_csize
&& s
->size
> min_csize
) {
2061 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2063 goto size_change_error
;
2065 if (min_csize
&& s
->size
== MAX_SIZE
) {
2066 /* Don't let the kernel choose a size - it will get
2069 pr_err("Limited v0.90 array to 2TB per device\n");
2070 s
->size
= min_csize
;
2072 if (st
->ss
->external
) {
2073 if (sra
->array
.level
== 0) {
2074 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2077 /* get array parameters after takeover
2078 * to change one parameter at time only
2080 rv
= md_get_array_info(fd
, &array
);
2083 /* make sure mdmon is
2084 * aware of the new level */
2085 if (!mdmon_running(st
->container_devnm
))
2086 start_mdmon(st
->container_devnm
);
2087 ping_monitor(container
);
2088 if (mdmon_running(st
->container_devnm
) == false) {
2089 pr_err("No mdmon found. Grow cannot continue.\n");
2094 if (s
->size
== MAX_SIZE
)
2096 array
.size
= s
->size
;
2097 if (s
->size
& ~INT32_MAX
) {
2098 /* got truncated to 32bit, write to
2099 * component_size instead
2101 rv
= sysfs_set_num(sra
, NULL
, "component_size", s
->size
);
2103 rv
= md_set_array_info(fd
, &array
);
2105 /* manage array size when it is managed externally
2107 if ((rv
== 0) && st
->ss
->external
)
2108 rv
= set_array_size(st
, sra
, sra
->text_version
);
2111 if (raid0_takeover
) {
2112 /* do not recync non-existing parity,
2113 * we will drop it anyway
2115 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2116 /* go back to raid0, drop parity disk
2118 sysfs_set_str(sra
, NULL
, "level", "raid0");
2119 md_get_array_info(fd
, &array
);
2126 /* restore metadata */
2127 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2128 UnSet
, NULL
, devname
,
2129 ROLLBACK_METADATA_CHANGES
,
2132 pr_err("Cannot set device size for %s: %s\n",
2133 devname
, strerror(err
));
2135 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2136 cont_err("Bitmap must be removed before size can be changed\n");
2140 if (s
->assume_clean
) {
2141 /* This will fail on kernels older than 3.0 unless
2142 * a backport has been arranged.
2145 sysfs_set_str(sra
, NULL
, "resync_start", STR_COMMON_NONE
) < 0)
2146 pr_err("--assume-clean not supported with --grow on this kernel\n");
2148 md_get_array_info(fd
, &array
);
2149 s
->size
= get_component_size(fd
)/2;
2151 s
->size
= array
.size
;
2152 if (c
->verbose
>= 0) {
2153 if (s
->size
== orig_size
)
2154 pr_err("component size of %s unchanged at %lluK\n",
2157 pr_err("component size of %s has been set to %lluK\n",
2161 } else if (!is_container(array
.level
)) {
2162 s
->size
= get_component_size(fd
)/2;
2164 s
->size
= array
.size
;
2167 /* See if there is anything else to do */
2168 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2169 (s
->layout_str
== NULL
) &&
2170 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2171 s
->data_offset
== INVALID_SECTORS
&&
2172 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2173 /* Nothing more to do */
2174 if (!changed
&& c
->verbose
>= 0)
2175 pr_err("%s: no change requested\n", devname
);
2179 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2180 * current implementation assumes that following conditions must be met:
2183 * - near_copies == 2
2185 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2186 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2187 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2190 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2192 dprintf("Array cannot be reshaped\n");
2198 /* Make sure mdmon has seen the device removal
2199 * and updated metadata before we continue with
2203 ping_monitor(container
);
2206 memset(&info
, 0, sizeof(info
));
2208 if (sysfs_init(&info
, fd
, NULL
)) {
2209 pr_err("failed to initialize sysfs.\n");
2213 strcpy(info
.text_version
, sra
->text_version
);
2214 info
.component_size
= s
->size
*2;
2215 info
.new_level
= s
->level
;
2216 info
.new_chunk
= s
->chunk
* 1024;
2217 if (is_container(info
.array
.level
)) {
2218 info
.delta_disks
= UnSet
;
2219 info
.array
.raid_disks
= s
->raiddisks
;
2220 } else if (s
->raiddisks
)
2221 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2223 info
.delta_disks
= UnSet
;
2224 if (s
->layout_str
== NULL
) {
2225 info
.new_layout
= UnSet
;
2226 if (info
.array
.level
== 6 &&
2227 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2228 info
.array
.layout
>= 16) {
2229 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2230 cont_err("during the reshape, please specify --layout=preserve\n");
2231 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2235 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2236 strcmp(s
->layout_str
, "normalize") == 0) {
2237 /* If we have a -6 RAID6 layout, remove the '-6'. */
2238 info
.new_layout
= UnSet
;
2239 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2241 strcpy(l
, map_num_s(r6layout
, info
.array
.layout
));
2242 h
= strrchr(l
, '-');
2243 if (h
&& strcmp(h
, "-6") == 0) {
2245 info
.new_layout
= map_name(r6layout
, l
);
2248 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2252 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2253 /* This means that a non-standard RAID6 layout
2256 * - When reshape a RAID6 (e.g. adding a device)
2257 * which is in a non-standard layout, it is OK
2258 * to preserve that layout.
2259 * - When converting a RAID5 to RAID6, leave it in
2260 * the XXX-6 layout, don't re-layout.
2262 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2263 info
.new_layout
= info
.array
.layout
;
2264 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2266 strcpy(l
, map_num_s(r5layout
, info
.array
.layout
));
2268 info
.new_layout
= map_name(r6layout
, l
);
2270 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2275 int l
= info
.new_level
;
2277 l
= info
.array
.level
;
2280 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2283 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2286 info
.new_layout
= parse_layout_10(s
->layout_str
);
2289 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2292 pr_err("layout not meaningful with this level\n");
2296 if (info
.new_layout
== UnSet
) {
2297 pr_err("layout %s not understood for this level\n",
2304 if (array
.level
== LEVEL_FAULTY
) {
2305 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2306 pr_err("cannot change level of Faulty device\n");
2310 pr_err("cannot set chunksize of Faulty device\n");
2313 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2314 pr_err("cannot set raid_disks of Faulty device\n");
2317 if (s
->layout_str
) {
2318 if (md_get_array_info(fd
, &array
) != 0) {
2319 dprintf("Cannot get array information.\n");
2322 array
.layout
= info
.new_layout
;
2323 if (md_set_array_info(fd
, &array
) != 0) {
2324 pr_err("failed to set new layout\n");
2326 } else if (c
->verbose
>= 0)
2327 printf("layout for %s set to %d\n",
2328 devname
, array
.layout
);
2330 } else if (is_container(array
.level
)) {
2331 /* This change is to be applied to every array in the
2332 * container. This is only needed when the metadata imposes
2333 * restraints of the various arrays in the container.
2334 * Currently we only know that IMSM requires all arrays
2335 * to have the same number of devices so changing the
2336 * number of devices (On-Line Capacity Expansion) must be
2337 * performed at the level of the container
2340 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2341 c
->force
, c
->backup_file
, c
->verbose
,
2345 /* get spare devices from external metadata
2347 if (st
->ss
->external
) {
2348 struct mdinfo
*info2
;
2350 info2
= st
->ss
->container_content(st
, subarray
);
2352 info
.array
.spare_disks
=
2353 info2
->array
.spare_disks
;
2358 /* Impose these changes on a single array. First
2359 * check that the metadata is OK with the change. */
2361 if (reshape_super(st
, 0, info
.new_level
,
2362 info
.new_layout
, info
.new_chunk
,
2363 info
.array
.raid_disks
, info
.delta_disks
,
2364 c
->backup_file
, devname
,
2365 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2370 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2371 devlist
, s
->data_offset
, c
->backup_file
,
2372 c
->verbose
, 0, 0, 0);
2382 /* verify_reshape_position()
2383 * Function checks if reshape position in metadata is not farther
2384 * than position in md.
2386 * 0 : not valid sysfs entry
2387 * it can be caused by not started reshape, it should be started
2388 * by reshape array or raid0 array is before takeover
2389 * -1 : error, reshape position is obviously wrong
2390 * 1 : success, reshape progress correct or updated
2392 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2395 char buf
[SYSFS_MAX_BUF_SIZE
];
2398 /* read sync_max, failure can mean raid0 array */
2399 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, sizeof(buf
));
2403 unsigned long long position
= strtoull(buf
, &ep
, 0);
2405 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2406 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2407 position
*= get_data_disks(level
,
2409 info
->array
.raid_disks
);
2410 if (info
->reshape_progress
< position
) {
2411 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2412 info
->reshape_progress
, position
);
2413 info
->reshape_progress
= position
;
2415 } else if (info
->reshape_progress
> position
) {
2416 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2417 position
, info
->reshape_progress
);
2420 dprintf("Reshape position in md and metadata are the same;");
2424 } else if (rv
== 0) {
2425 /* for valid sysfs entry, 0-length content
2426 * should be indicated as error
2434 static unsigned long long choose_offset(unsigned long long lo
,
2435 unsigned long long hi
,
2436 unsigned long long min
,
2437 unsigned long long max
)
2439 /* Choose a new offset between hi and lo.
2440 * It must be between min and max, but
2441 * we would prefer something near the middle of hi/lo, and also
2442 * prefer to be aligned to a big power of 2.
2444 * So we start with the middle, then for each bit,
2445 * starting at '1' and increasing, if it is set, we either
2446 * add it or subtract it if possible, preferring the option
2447 * which is furthest from the boundary.
2449 * We stop once we get a 1MB alignment. As units are in sectors,
2450 * 1MB = 2*1024 sectors.
2452 unsigned long long choice
= (lo
+ hi
) / 2;
2453 unsigned long long bit
= 1;
2455 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2456 unsigned long long bigger
, smaller
;
2457 if (! (bit
& choice
))
2459 bigger
= choice
+ bit
;
2460 smaller
= choice
- bit
;
2461 if (bigger
> max
&& smaller
< min
)
2465 else if (smaller
< min
)
2467 else if (hi
- bigger
> smaller
- lo
)
2475 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2476 char *devname
, int delta_disks
,
2477 unsigned long long data_offset
,
2478 unsigned long long min
,
2484 unsigned long long before
, after
;
2486 /* Need to find min space before and after so same is used
2489 before
= UINT64_MAX
;
2491 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2495 struct supertype
*st2
;
2496 struct mdinfo info2
;
2498 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2500 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2501 dfd
= dev_open(dn
, O_RDONLY
);
2503 pr_err("%s: cannot open component %s\n",
2504 devname
, dn
? dn
: "-unknown-");
2507 st2
= dup_super(st
);
2508 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2512 pr_err("%s: cannot get superblock from %s\n",
2516 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2517 st2
->ss
->free_super(st2
);
2519 if (info2
.space_before
== 0 &&
2520 info2
.space_after
== 0) {
2521 /* Metadata doesn't support data_offset changes */
2523 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2527 if (before
> info2
.space_before
)
2528 before
= info2
.space_before
;
2529 if (after
> info2
.space_after
)
2530 after
= info2
.space_after
;
2532 if (data_offset
!= INVALID_SECTORS
) {
2534 if (info2
.data_offset
== data_offset
) {
2535 pr_err("%s: already has that data_offset\n",
2539 if (data_offset
< info2
.data_offset
)
2543 } else if ((data_offset
<= info2
.data_offset
&&
2545 (data_offset
>= info2
.data_offset
&&
2547 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2553 if (before
== UINT64_MAX
)
2554 /* impossible really, there must be no devices */
2557 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2558 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2559 unsigned long long new_data_offset
;
2561 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2563 if (delta_disks
< 0) {
2564 /* Don't need any space as array is shrinking
2565 * just move data_offset up by min
2567 if (data_offset
== INVALID_SECTORS
)
2568 new_data_offset
= sd
->data_offset
+ min
;
2570 if (data_offset
< sd
->data_offset
+ min
) {
2571 pr_err("--data-offset too small for %s\n",
2575 new_data_offset
= data_offset
;
2577 } else if (delta_disks
> 0) {
2578 /* need space before */
2582 pr_err("Insufficient head-space for reshape on %s\n",
2586 if (data_offset
== INVALID_SECTORS
)
2587 new_data_offset
= sd
->data_offset
- min
;
2589 if (data_offset
> sd
->data_offset
- min
) {
2590 pr_err("--data-offset too large for %s\n",
2594 new_data_offset
= data_offset
;
2598 /* can move up or down. If 'data_offset'
2599 * was set we would have already decided,
2600 * so just choose direction with most space.
2607 sysfs_set_str(sra
, NULL
, "reshape_direction",
2608 dir
== 1 ? "backwards" : "forwards");
2610 /* Increase data offset */
2614 pr_err("Insufficient tail-space for reshape on %s\n",
2618 if (data_offset
!= INVALID_SECTORS
&&
2619 data_offset
< sd
->data_offset
+ min
) {
2620 pr_err("--data-offset too small on %s\n",
2624 if (data_offset
!= INVALID_SECTORS
)
2625 new_data_offset
= data_offset
;
2627 new_data_offset
= choose_offset(sd
->data_offset
,
2628 sd
->data_offset
+ after
,
2629 sd
->data_offset
+ min
,
2630 sd
->data_offset
+ after
);
2632 /* Decrease data offset */
2636 pr_err("insufficient head-room on %s\n",
2640 if (data_offset
!= INVALID_SECTORS
&&
2641 data_offset
> sd
->data_offset
- min
) {
2642 pr_err("--data-offset too large on %s\n",
2646 if (data_offset
!= INVALID_SECTORS
)
2647 new_data_offset
= data_offset
;
2649 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2651 sd
->data_offset
- before
,
2652 sd
->data_offset
- min
);
2655 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2656 if (err
< 0 && errno
== E2BIG
) {
2657 /* try again after increasing data size to max */
2658 err
= sysfs_set_num(sra
, sd
, "size", 0);
2659 if (err
< 0 && errno
== EINVAL
&&
2660 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2661 /* some kernels have a bug where you cannot
2662 * use '0' on spare devices. */
2663 sysfs_set_num(sra
, sd
, "size",
2664 (sra
->component_size
+ after
)/2);
2666 err
= sysfs_set_num(sra
, sd
, "new_offset",
2670 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2671 pr_err("data-offset is too big for %s\n", dn
);
2674 if (sd
== sra
->devs
&&
2675 (errno
== ENOENT
|| errno
== E2BIG
))
2676 /* Early kernel, no 'new_offset' file,
2677 * or kernel doesn't like us.
2678 * For RAID5/6 this is not fatal
2681 pr_err("Cannot set new_offset for %s\n", dn
);
2689 /* Just use a backup file */
2693 static int raid10_reshape(char *container
, int fd
, char *devname
,
2694 struct supertype
*st
, struct mdinfo
*info
,
2695 struct reshape
*reshape
,
2696 unsigned long long data_offset
,
2697 int force
, int verbose
)
2699 /* Changing raid_disks, layout, chunksize or possibly
2700 * just data_offset for a RAID10.
2701 * We must always change data_offset. We change by at least
2702 * ->min_offset_change which is the largest of the old and new
2704 * If raid_disks is increasing, then data_offset must decrease
2705 * by at least this copy size.
2706 * If raid_disks is unchanged, data_offset must increase or
2707 * decrease by at least min_offset_change but preferably by much more.
2708 * We choose half of the available space.
2709 * If raid_disks is decreasing, data_offset must increase by
2710 * at least min_offset_change. To allow of this, component_size
2711 * must be decreased by the same amount.
2713 * So we calculate the required minimum and direction, possibly
2714 * reduce the component_size, then iterate through the devices
2715 * and set the new_data_offset.
2716 * If that all works, we set chunk_size, layout, raid_disks, and start
2720 unsigned long long min
;
2723 sra
= sysfs_read(fd
, NULL
,
2724 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2727 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2730 min
= reshape
->min_offset_change
;
2732 if (info
->delta_disks
)
2733 sysfs_set_str(sra
, NULL
, "reshape_direction",
2734 info
->delta_disks
< 0 ? "backwards" : "forwards");
2735 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2736 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2737 (sra
->component_size
- min
)/2);
2739 pr_err("cannot reduce component size\n");
2743 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2744 data_offset
, min
, 0);
2746 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2747 cont_err("supported on this kernel\n");
2753 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2755 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2756 reshape
->after
.layout
) < 0)
2759 sysfs_set_num(sra
, NULL
, "raid_disks",
2760 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2762 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2765 pr_err("Cannot set array shape for %s\n",
2768 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2769 cont_err(" Bitmap must be removed before shape can be changed\n");
2779 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2781 struct mdinfo
*sra
, *sd
;
2782 /* Initialisation to silence compiler warning */
2783 unsigned long long min_space_before
= 0, min_space_after
= 0;
2786 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2789 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2792 struct supertype
*st2
;
2793 struct mdinfo info2
;
2795 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2797 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2798 dfd
= dev_open(dn
, O_RDONLY
);
2801 st2
= dup_super(st
);
2802 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2808 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2809 st2
->ss
->free_super(st2
);
2812 min_space_before
> info2
.space_before
)
2813 min_space_before
= info2
.space_before
;
2815 min_space_after
> info2
.space_after
)
2816 min_space_after
= info2
.space_after
;
2819 if (sd
== NULL
&& !first
) {
2820 info
->space_after
= min_space_after
;
2821 info
->space_before
= min_space_before
;
2826 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2827 struct mdinfo
*info
,
2828 int disks
, unsigned long long blocks
)
2830 /* Check that the internal stripe cache is
2831 * large enough, or it won't work.
2832 * It must hold at least 4 stripes of the larger
2835 unsigned long cache
;
2836 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2837 cache
*= 4; /* 4 stripes minimum */
2838 cache
/= 512; /* convert to sectors */
2839 /* make sure there is room for 'blocks' with a bit to spare */
2840 if (cache
< 16 + blocks
/ disks
)
2841 cache
= 16 + blocks
/ disks
;
2842 cache
/= (4096/512); /* Convert from sectors to pages */
2844 if (sra
->cache_size
< cache
)
2845 subarray_set_num(container
, sra
, "stripe_cache_size",
2849 static int impose_reshape(struct mdinfo
*sra
,
2850 struct mdinfo
*info
,
2851 struct supertype
*st
,
2854 char *devname
, char *container
,
2855 struct reshape
*reshape
)
2857 struct mdu_array_info_s array
;
2859 sra
->new_chunk
= info
->new_chunk
;
2862 /* for external metadata checkpoint saved by mdmon can be lost
2863 * or missed /due to e.g. crash/. Check if md is not during
2864 * restart farther than metadata points to.
2865 * If so, this means metadata information is obsolete.
2867 if (st
->ss
->external
)
2868 verify_reshape_position(info
, reshape
->level
);
2869 sra
->reshape_progress
= info
->reshape_progress
;
2871 sra
->reshape_progress
= 0;
2872 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2873 /* start from the end of the new array */
2874 sra
->reshape_progress
= (sra
->component_size
2875 * reshape
->after
.data_disks
);
2878 md_get_array_info(fd
, &array
);
2879 if (info
->array
.chunk_size
== info
->new_chunk
&&
2880 reshape
->before
.layout
== reshape
->after
.layout
&&
2881 st
->ss
->external
== 0) {
2882 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2883 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2884 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2887 pr_err("Cannot set device shape for %s: %s\n",
2888 devname
, strerror(errno
));
2891 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2892 cont_err("Bitmap must be removed before shape can be changed\n");
2896 } else if (!restart
) {
2897 /* set them all just in case some old 'new_*' value
2898 * persists from some earlier problem.
2901 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2903 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2904 reshape
->after
.layout
) < 0)
2906 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2907 reshape
->after
.data_disks
+
2908 reshape
->parity
) < 0)
2911 pr_err("Cannot set device shape for %s\n", devname
);
2914 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2915 cont_err("Bitmap must be removed before shape can be changed\n");
2924 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2927 struct mdu_array_info_s array
;
2930 if (sysfs_init(&info
, fd
, NULL
)) {
2931 pr_err("failed to initialize sysfs.\n");
2935 md_get_array_info(fd
, &array
);
2936 if (level
== 0 && is_level456(array
.level
)) {
2937 /* To convert to RAID0 we need to fail and
2938 * remove any non-data devices. */
2941 int data_disks
= array
.raid_disks
- 1;
2942 if (array
.level
== 6)
2944 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2946 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2948 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2949 /* First remove any spares so no recovery starts */
2950 for (d
= 0, found
= 0;
2951 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2952 mdu_disk_info_t disk
;
2954 if (md_get_disk_info(fd
, &disk
) < 0)
2956 if (disk
.major
== 0 && disk
.minor
== 0)
2959 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2960 disk
.raid_disk
< data_disks
)
2963 ioctl(fd
, HOT_REMOVE_DISK
,
2964 makedev(disk
.major
, disk
.minor
));
2966 /* Now fail anything left */
2967 md_get_array_info(fd
, &array
);
2968 for (d
= 0, found
= 0;
2969 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2970 mdu_disk_info_t disk
;
2972 if (md_get_disk_info(fd
, &disk
) < 0)
2974 if (disk
.major
== 0 && disk
.minor
== 0)
2977 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2978 disk
.raid_disk
< data_disks
)
2981 ioctl(fd
, SET_DISK_FAULTY
,
2982 makedev(disk
.major
, disk
.minor
));
2983 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2986 c
= map_num(pers
, level
);
2988 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2991 pr_err("%s: could not set level to %s\n",
2994 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2995 cont_err("Bitmap must be removed before level can be changed\n");
2999 pr_err("level of %s changed to %s\n", devname
, c
);
3005 static void catch_term(int sig
)
3010 static int reshape_array(char *container
, int fd
, char *devname
,
3011 struct supertype
*st
, struct mdinfo
*info
,
3012 int force
, struct mddev_dev
*devlist
,
3013 unsigned long long data_offset
,
3014 char *backup_file
, int verbose
, int forked
,
3015 int restart
, int freeze_reshape
)
3017 struct reshape reshape
;
3020 int orig_level
= UnSet
;
3024 struct mdu_array_info_s array
;
3027 struct mddev_dev
*dv
;
3031 unsigned long long *offsets
= NULL
;
3035 unsigned long blocks
;
3036 unsigned long long array_size
;
3038 struct mdinfo
*sra
= NULL
;
3039 char buf
[SYSFS_MAX_BUF_SIZE
];
3041 /* when reshaping a RAID0, the component_size might be zero.
3042 * So try to fix that up.
3044 if (md_get_array_info(fd
, &array
) != 0) {
3045 dprintf("Cannot get array information.\n");
3048 if (st
->update_tail
== NULL
)
3049 st
->update_tail
= &st
->updates
;
3050 if (array
.level
== 0 && info
->component_size
== 0) {
3051 get_dev_size(fd
, NULL
, &array_size
);
3052 info
->component_size
= array_size
/ array
.raid_disks
;
3055 if (array
.level
== 10)
3056 /* Need space_after info */
3057 get_space_after(fd
, st
, info
);
3059 if (info
->reshape_active
) {
3060 int new_level
= info
->new_level
;
3061 info
->new_level
= UnSet
;
3062 if (info
->delta_disks
> 0)
3063 info
->array
.raid_disks
-= info
->delta_disks
;
3064 msg
= analyse_change(devname
, info
, &reshape
);
3065 info
->new_level
= new_level
;
3066 if (info
->delta_disks
> 0)
3067 info
->array
.raid_disks
+= info
->delta_disks
;
3069 /* Make sure the array isn't read-only */
3070 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3072 msg
= analyse_change(devname
, info
, &reshape
);
3074 /* if msg == "", error has already been printed */
3076 pr_err("%s\n", msg
);
3079 if (restart
&& (reshape
.level
!= info
->array
.level
||
3080 reshape
.before
.layout
!= info
->array
.layout
||
3081 reshape
.before
.data_disks
+ reshape
.parity
!=
3082 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3083 pr_err("reshape info is not in native format - cannot continue.\n");
3087 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3088 !((sysfs_get_str(info
, NULL
, "sync_action",
3089 buf
, sizeof(buf
)) > 0) &&
3090 (strncmp(buf
, "reshape", 7) == 0))) {
3091 /* When reshape is restarted from '0', very begin of array
3092 * it is possible that for external metadata reshape and array
3093 * configuration doesn't happen.
3094 * Check if md has the same opinion, and reshape is restarted
3095 * from 0. If so, this is regular reshape start after reshape
3096 * switch in metadata to next array only.
3098 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3099 (info
->reshape_progress
== 0))
3104 * reshape already started. just skip to monitoring
3107 if (reshape
.backup_blocks
== 0)
3109 if (restart
& RESHAPE_NO_BACKUP
)
3112 /* Need 'sra' down at 'started:' */
3113 sra
= sysfs_read(fd
, NULL
,
3114 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3115 GET_CHUNK
|GET_CACHE
);
3117 pr_err("%s: Cannot get array details from sysfs\n",
3123 backup_file
= locate_backup(sra
->sys_name
);
3127 /* The container is frozen but the array may not be.
3128 * So freeze the array so spares don't get put to the wrong use
3129 * FIXME there should probably be a cleaner separation between
3130 * freeze_array and freeze_container.
3132 sysfs_freeze_array(info
);
3133 /* Check we have enough spares to not be degraded */
3135 for (dv
= devlist
; dv
; dv
=dv
->next
)
3137 spares_needed
= max(reshape
.before
.data_disks
,
3138 reshape
.after
.data_disks
) +
3139 reshape
.parity
- array
.raid_disks
;
3141 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3142 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3143 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3144 " Use --force to over-ride this check.\n",
3146 spares_needed
== 1 ? "" : "s",
3147 info
->array
.spare_disks
+ added_disks
);
3150 /* Check we have enough spares to not fail */
3151 spares_needed
= max(reshape
.before
.data_disks
,
3152 reshape
.after
.data_disks
)
3154 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3155 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3156 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3157 spares_needed
, spares_needed
== 1 ? "" : "s",
3158 info
->array
.spare_disks
+ added_disks
);
3162 if (reshape
.level
!= array
.level
) {
3163 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3166 info
->new_layout
= UnSet
; /* after level change,
3167 * layout is meaningless */
3168 orig_level
= array
.level
;
3169 sysfs_freeze_array(info
);
3171 if (reshape
.level
> 0 && st
->ss
->external
) {
3172 /* make sure mdmon is aware of the new level */
3173 if (mdmon_running(container
))
3174 flush_mdmon(container
);
3176 if (!mdmon_running(container
))
3177 start_mdmon(container
);
3178 ping_monitor(container
);
3179 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3180 st
->update_tail
= &st
->updates
;
3183 /* ->reshape_super might have chosen some spares from the
3184 * container that it wants to be part of the new array.
3185 * We can collect them with ->container_content and give
3186 * them to the kernel.
3188 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3189 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3190 struct mdinfo
*info2
=
3191 st
->ss
->container_content(st
, subarray
);
3195 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3196 pr_err("unable to initialize sysfs for %s\n",
3201 /* When increasing number of devices, we need to set
3202 * new raid_disks before adding these, or they might
3205 if (reshape
.backup_blocks
&&
3206 reshape
.after
.data_disks
>
3207 reshape
.before
.data_disks
)
3208 subarray_set_num(container
, info2
, "raid_disks",
3209 reshape
.after
.data_disks
+
3211 for (d
= info2
->devs
; d
; d
= d
->next
) {
3212 if (d
->disk
.state
== 0 &&
3213 d
->disk
.raid_disk
>= 0) {
3214 /* This is a spare that wants to
3215 * be part of the array.
3217 add_disk(fd
, st
, info2
, d
);
3223 /* We might have been given some devices to add to the
3224 * array. Now that the array has been changed to the right
3225 * level and frozen, we can safely add them.
3228 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, UOPT_UNDEFINED
, 0))
3232 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3233 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3234 if (reshape
.backup_blocks
== 0) {
3235 /* No restriping needed, but we might need to impose
3236 * some more changes: layout, raid_disks, chunk_size
3238 /* read current array info */
3239 if (md_get_array_info(fd
, &array
) != 0) {
3240 dprintf("Cannot get array information.\n");
3243 /* compare current array info with new values and if
3244 * it is different update them to new */
3245 if (info
->new_layout
!= UnSet
&&
3246 info
->new_layout
!= array
.layout
) {
3247 array
.layout
= info
->new_layout
;
3248 if (md_set_array_info(fd
, &array
) != 0) {
3249 pr_err("failed to set new layout\n");
3251 } else if (verbose
>= 0)
3252 printf("layout for %s set to %d\n",
3253 devname
, array
.layout
);
3255 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3257 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3258 array
.raid_disks
+= info
->delta_disks
;
3259 if (md_set_array_info(fd
, &array
) != 0) {
3260 pr_err("failed to set raid disks\n");
3262 } else if (verbose
>= 0) {
3263 printf("raid_disks for %s set to %d\n",
3264 devname
, array
.raid_disks
);
3267 if (info
->new_chunk
!= 0 &&
3268 info
->new_chunk
!= array
.chunk_size
) {
3269 if (sysfs_set_num(info
, NULL
,
3270 "chunk_size", info
->new_chunk
) != 0) {
3271 pr_err("failed to set chunk size\n");
3273 } else if (verbose
>= 0)
3274 printf("chunk size for %s set to %d\n",
3275 devname
, info
->new_chunk
);
3282 * There are three possibilities.
3283 * 1/ The array will shrink.
3284 * We need to ensure the reshape will pause before reaching
3285 * the 'critical section'. We also need to fork and wait for
3286 * that to happen. When it does we
3287 * suspend/backup/complete/unfreeze
3289 * 2/ The array will not change size.
3290 * This requires that we keep a backup of a sliding window
3291 * so that we can restore data after a crash. So we need
3292 * to fork and monitor progress.
3293 * In future we will allow the data_offset to change, so
3294 * a sliding backup becomes unnecessary.
3296 * 3/ The array will grow. This is relatively easy.
3297 * However the kernel's restripe routines will cheerfully
3298 * overwrite some early data before it is safe. So we
3299 * need to make a backup of the early parts of the array
3300 * and be ready to restore it if rebuild aborts very early.
3301 * For externally managed metadata, we still need a forked
3302 * child to monitor the reshape and suspend IO over the region
3303 * that is being reshaped.
3305 * We backup data by writing it to one spare, or to a
3306 * file which was given on command line.
3308 * In each case, we first make sure that storage is available
3309 * for the required backup.
3311 * - request the shape change.
3312 * - fork to handle backup etc.
3314 /* Check that we can hold all the data */
3315 get_dev_size(fd
, NULL
, &array_size
);
3316 if (reshape
.new_size
< (array_size
/512)) {
3317 pr_err("this change will reduce the size of the array.\n"
3318 " use --grow --array-size first to truncate array.\n"
3319 " e.g. mdadm --grow %s --array-size %llu\n",
3320 devname
, reshape
.new_size
/2);
3324 if (array
.level
== 10) {
3325 /* Reshaping RAID10 does not require any data backup by
3326 * user-space. Instead it requires that the data_offset
3327 * is changed to avoid the need for backup.
3328 * So this is handled very separately
3331 /* Nothing to do. */
3333 return raid10_reshape(container
, fd
, devname
, st
, info
,
3334 &reshape
, data_offset
, force
, verbose
);
3336 sra
= sysfs_read(fd
, NULL
,
3337 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3340 pr_err("%s: Cannot get array details from sysfs\n",
3346 switch(set_new_data_offset(sra
, st
, devname
,
3347 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3349 reshape
.min_offset_change
, 1)) {
3353 /* Updated data_offset, so it's easy now */
3354 update_cache_size(container
, sra
, info
,
3355 min(reshape
.before
.data_disks
,
3356 reshape
.after
.data_disks
),
3357 reshape
.backup_blocks
);
3359 /* Right, everything seems fine. Let's kick things off.
3363 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3364 devname
, container
, &reshape
) < 0)
3366 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3368 if (errno
!= EINVAL
) {
3369 pr_err("Failed to initiate reshape!\n");
3372 /* revert data_offset and try the old way */
3373 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3374 sysfs_set_num(sra
, sd
, "new_offset",
3376 sysfs_set_str(sra
, NULL
, "reshape_direction",
3381 if (info
->new_level
== reshape
.level
)
3383 /* need to adjust level when reshape completes */
3385 case -1: /* ignore error, but don't wait */
3387 default: /* parent */
3396 fd
= open_dev(sra
->sys_name
);
3398 impose_level(fd
, info
->new_level
, devname
, verbose
);
3400 case 1: /* Couldn't set data_offset, try the old way */
3401 if (data_offset
!= INVALID_SECTORS
) {
3402 pr_err("Cannot update data_offset on this array\n");
3409 /* Decide how many blocks (sectors) for a reshape
3410 * unit. The number we have so far is just a minimum
3412 blocks
= reshape
.backup_blocks
;
3413 if (reshape
.before
.data_disks
==
3414 reshape
.after
.data_disks
) {
3415 /* Make 'blocks' bigger for better throughput, but
3416 * not so big that we reject it below.
3417 * Try for 16 megabytes
3419 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3422 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3424 if (blocks
>= sra
->component_size
/2) {
3425 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3429 /* Now we need to open all these devices so we can read/write.
3431 nrdisks
= max(reshape
.before
.data_disks
,
3432 reshape
.after
.data_disks
) + reshape
.parity
3433 + sra
->array
.spare_disks
;
3434 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3435 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3437 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3438 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3439 backup_file
, fdlist
, offsets
);
3443 if ((st
->ss
->manage_reshape
== NULL
) ||
3444 (st
->ss
->recover_backup
== NULL
)) {
3445 if (backup_file
== NULL
) {
3446 if (reshape
.after
.data_disks
<=
3447 reshape
.before
.data_disks
) {
3448 pr_err("%s: Cannot grow - need backup-file\n",
3450 pr_err(" Please provide one with \"--backup=...\"\n");
3452 } else if (d
== odisks
) {
3453 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3457 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3459 fdlist
+d
, offsets
+d
,
3460 sra
->sys_name
, restart
)) {
3467 update_cache_size(container
, sra
, info
,
3468 min(reshape
.before
.data_disks
,
3469 reshape
.after
.data_disks
), blocks
);
3471 /* Right, everything seems fine. Let's kick things off.
3472 * If only changing raid_disks, use ioctl, else use
3477 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3478 devname
, container
, &reshape
) < 0)
3481 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3482 reshape
.after
.data_disks
, st
);
3484 pr_err("Cannot %s reshape for %s\n",
3485 restart
? "continue" : "start", devname
);
3489 sysfs_set_str(sra
, NULL
, "array_state", "active");
3490 if (freeze_reshape
) {
3494 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3495 sra
->reshape_progress
);
3500 if (continue_via_systemd(container
?: sra
->sys_name
,
3501 GROW_SERVICE
, NULL
)) {
3508 /* Now we just need to kick off the reshape and watch, while
3509 * handling backups of the data...
3510 * This is all done by a forked background process.
3512 switch(forked
? 0 : fork()) {
3514 pr_err("Cannot run child to monitor reshape: %s\n",
3528 /* Close unused file descriptor in the forked process */
3531 /* If another array on the same devices is busy, the
3532 * reshape will wait for them. This would mean that
3533 * the first section that we suspend will stay suspended
3534 * for a long time. So check on that possibility
3535 * by looking for "DELAYED" in /proc/mdstat, and if found,
3539 struct mdstat_ent
*mds
, *m
;
3541 mds
= mdstat_read(1, 0);
3542 for (m
= mds
; m
; m
= m
->next
)
3543 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3544 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3547 /* Haven't started the reshape thread
3554 if (delayed
== 1 && get_linux_version() < 3007000) {
3555 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3556 " You might experience problems until other reshapes complete.\n");
3560 mdstat_wait(30 - (delayed
-1) * 25);
3563 if (check_env("MDADM_GROW_VERIFY"))
3564 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3567 mlockall(MCL_FUTURE
);
3569 if (signal_s(SIGTERM
, catch_term
) == SIG_ERR
)
3572 if (st
->ss
->external
) {
3573 /* metadata handler takes it from here */
3574 done
= st
->ss
->manage_reshape(
3575 fd
, sra
, &reshape
, st
, blocks
,
3576 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3579 done
= child_monitor(
3580 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3581 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3586 if (backup_file
&& done
) {
3588 bul
= make_backup(sra
->sys_name
);
3591 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3599 unlink(backup_file
);
3606 if (!st
->ss
->external
&&
3607 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3608 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3610 /* no need to wait for the reshape to finish as
3611 * there is nothing more to do.
3618 if (st
->ss
->external
) {
3619 /* Re-load the metadata as much could have changed */
3620 int cfd
= open_dev(st
->container_devnm
);
3622 flush_mdmon(container
);
3623 st
->ss
->free_super(st
);
3624 st
->ss
->load_container(st
, cfd
, container
);
3629 /* set new array size if required customer_array_size is used
3632 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3633 info
->custom_array_size
)
3634 set_array_size(st
, info
, info
->text_version
);
3636 if (info
->new_level
!= reshape
.level
) {
3638 fd
= open(devname
, O_RDONLY
);
3639 impose_level(fd
, info
->new_level
, devname
, verbose
);
3641 if (info
->new_level
== 0)
3642 st
->update_tail
= NULL
;
3654 if (orig_level
!= UnSet
&& sra
) {
3655 c
= map_num(pers
, orig_level
);
3656 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3657 pr_err("aborting level change\n");
3665 /* mdfd handle is passed to be closed in child process (after fork).
3667 int reshape_container(char *container
, char *devname
,
3669 struct supertype
*st
,
3670 struct mdinfo
*info
,
3672 char *backup_file
, int verbose
,
3673 int forked
, int restart
, int freeze_reshape
)
3675 struct mdinfo
*cc
= NULL
;
3677 char last_devnm
[32] = "";
3679 /* component_size is not meaningful for a container,
3680 * so pass '0' meaning 'no change'
3683 reshape_super(st
, 0, info
->new_level
,
3684 info
->new_layout
, info
->new_chunk
,
3685 info
->array
.raid_disks
, info
->delta_disks
,
3686 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3694 /* ping monitor to be sure that update is on disk
3696 ping_monitor(container
);
3698 if (!forked
&& !freeze_reshape
)
3699 if (continue_via_systemd(container
, GROW_SERVICE
, NULL
))
3702 switch (forked
? 0 : fork()) {
3703 case -1: /* error */
3704 perror("Cannot fork to complete reshape\n");
3707 default: /* parent */
3708 if (!freeze_reshape
)
3709 printf("%s: multi-array reshape continues in background\n", Name
);
3717 /* close unused handle in child process
3723 /* For each member array with reshape_active,
3724 * we need to perform the reshape.
3725 * We pick the first array that needs reshaping and
3726 * reshape it. reshape_array() will re-read the metadata
3727 * so the next time through a different array should be
3728 * ready for reshape.
3729 * It is possible that the 'different' array will not
3730 * be assembled yet. In that case we simple exit.
3731 * When it is assembled, the mdadm which assembles it
3732 * will take over the reshape.
3734 struct mdinfo
*content
;
3736 struct mdstat_ent
*mdstat
;
3742 cc
= st
->ss
->container_content(st
, NULL
);
3744 for (content
= cc
; content
; content
= content
->next
) {
3746 if (!content
->reshape_active
)
3749 subarray
= strchr(content
->text_version
+1, '/')+1;
3750 mdstat
= mdstat_by_subdev(subarray
, container
);
3753 if (mdstat
->active
== 0) {
3754 pr_err("Skipping inactive array %s.\n",
3756 free_mdstat(mdstat
);
3765 devid
= devnm2devid(mdstat
->devnm
);
3766 adev
= map_dev(major(devid
), minor(devid
), 0);
3768 adev
= content
->text_version
;
3770 fd
= open_dev(mdstat
->devnm
);
3772 pr_err("Device %s cannot be opened for reshape.\n",
3777 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3778 /* Do not allow for multiple reshape_array() calls for
3780 * It can happen when reshape_array() returns without
3781 * error, when reshape is not finished (wrong reshape
3782 * starting/continuation conditions). Mdmon doesn't
3783 * switch to next array in container and reentry
3784 * conditions for the same array occur.
3785 * This is possibly interim until the behaviour of
3786 * reshape_array is resolved().
3788 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3792 strcpy(last_devnm
, mdstat
->devnm
);
3794 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3795 pr_err("Unable to initialize sysfs for %s\n",
3801 if (mdmon_running(container
))
3802 flush_mdmon(container
);
3804 rv
= reshape_array(container
, fd
, adev
, st
,
3805 content
, force
, NULL
, INVALID_SECTORS
,
3806 backup_file
, verbose
, 1, restart
,
3810 if (freeze_reshape
) {
3819 if (mdmon_running(container
))
3820 flush_mdmon(container
);
3829 * We run a child process in the background which performs the following
3831 * - wait for resync to reach a certain point
3832 * - suspend io to the following section
3833 * - backup that section
3834 * - allow resync to proceed further
3836 * - discard the backup.
3838 * When are combined in slightly different ways in the three cases.
3840 * - suspend/backup/allow/wait/resume/discard
3842 * - allow/wait/suspend/backup/allow/wait/resume/discard
3844 * - wait/resume/discard/suspend/backup/allow
3846 * suspend/backup/allow always come together
3847 * wait/resume/discard do too.
3848 * For the same-size case we have two backups to improve flow.
3852 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3853 unsigned long long backup_point
,
3854 unsigned long long wait_point
,
3855 unsigned long long *suspend_point
,
3856 unsigned long long *reshape_completed
, int *frozen
)
3858 /* This function is called repeatedly by the reshape manager.
3859 * It determines how much progress can safely be made and allows
3861 * - 'info' identifies the array and particularly records in
3862 * ->reshape_progress the metadata's knowledge of progress
3863 * This is a sector offset from the start of the array
3864 * of the next array block to be relocated. This number
3865 * may increase from 0 or decrease from array_size, depending
3866 * on the type of reshape that is happening.
3867 * Note that in contrast, 'sync_completed' is a block count of the
3868 * reshape so far. It gives the distance between the start point
3869 * (head or tail of device) and the next place that data will be
3870 * written. It always increases.
3871 * - 'reshape' is the structure created by analyse_change
3872 * - 'backup_point' shows how much the metadata manager has backed-up
3873 * data. For reshapes with increasing progress, it is the next address
3874 * to be backed up, previous addresses have been backed-up. For
3875 * decreasing progress, it is the earliest address that has been
3876 * backed up - later address are also backed up.
3877 * So addresses between reshape_progress and backup_point are
3878 * backed up providing those are in the 'correct' order.
3879 * - 'wait_point' is an array address. When reshape_completed
3880 * passes this point, progress_reshape should return. It might
3881 * return earlier if it determines that ->reshape_progress needs
3882 * to be updated or further backup is needed.
3883 * - suspend_point is maintained by progress_reshape and the caller
3884 * should not touch it except to initialise to zero.
3885 * It is an array address and it only increases in 2.6.37 and earlier.
3886 * This makes it difficult to handle reducing reshapes with
3887 * external metadata.
3888 * However: it is similar to backup_point in that it records the
3889 * other end of a suspended region from reshape_progress.
3890 * it is moved to extend the region that is safe to backup and/or
3892 * - reshape_completed is read from sysfs and returned. The caller
3893 * should copy this into ->reshape_progress when it has reason to
3894 * believe that the metadata knows this, and any backup outside this
3898 * 1 if more data from backup_point - but only as far as suspend_point,
3899 * should be backed up
3900 * 0 if things are progressing smoothly
3901 * -1 if the reshape is finished because it is all done,
3902 * -2 if the reshape is finished due to an error.
3905 int advancing
= (reshape
->after
.data_disks
3906 >= reshape
->before
.data_disks
);
3907 unsigned long long need_backup
; /* All data between start of array and
3908 * here will at some point need to
3911 unsigned long long read_offset
, write_offset
;
3912 unsigned long long write_range
;
3913 unsigned long long max_progress
, target
, completed
;
3914 unsigned long long array_size
= (info
->component_size
3915 * reshape
->before
.data_disks
);
3917 char buf
[SYSFS_MAX_BUF_SIZE
];
3919 /* First, we unsuspend any region that is now known to be safe.
3920 * If suspend_point is on the 'wrong' side of reshape_progress, then
3921 * we don't have or need suspension at the moment. This is true for
3922 * native metadata when we don't need to back-up.
3925 if (info
->reshape_progress
<= *suspend_point
)
3926 sysfs_set_num(info
, NULL
, "suspend_lo",
3927 info
->reshape_progress
);
3929 /* Note: this won't work in 2.6.37 and before.
3930 * Something somewhere should make sure we don't need it!
3932 if (info
->reshape_progress
>= *suspend_point
)
3933 sysfs_set_num(info
, NULL
, "suspend_hi",
3934 info
->reshape_progress
);
3937 /* Now work out how far it is safe to progress.
3938 * If the read_offset for ->reshape_progress is less than
3939 * 'blocks' beyond the write_offset, we can only progress as far
3941 * Otherwise we can progress until the write_offset for the new location
3942 * reaches (within 'blocks' of) the read_offset at the current location.
3943 * However that region must be suspended unless we are using native
3945 * If we need to suspend more, we limit it to 128M per device, which is
3946 * rather arbitrary and should be some time-based calculation.
3948 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3949 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3950 write_range
= info
->new_chunk
/512;
3951 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3952 need_backup
= array_size
;
3954 need_backup
= reshape
->backup_blocks
;
3956 if (read_offset
< write_offset
+ write_range
)
3957 max_progress
= backup_point
;
3960 read_offset
* reshape
->after
.data_disks
;
3962 if (read_offset
> write_offset
- write_range
)
3963 /* Can only progress as far as has been backed up,
3964 * which must be suspended */
3965 max_progress
= backup_point
;
3966 else if (info
->reshape_progress
<= need_backup
)
3967 max_progress
= backup_point
;
3969 if (info
->array
.major_version
>= 0)
3970 /* Can progress until backup is needed */
3971 max_progress
= need_backup
;
3973 /* Can progress until metadata update is required */
3975 read_offset
* reshape
->after
.data_disks
;
3976 /* but data must be suspended */
3977 if (max_progress
< *suspend_point
)
3978 max_progress
= *suspend_point
;
3983 /* We know it is safe to progress to 'max_progress' providing
3984 * it is suspended or we are using native metadata.
3985 * Consider extending suspend_point 128M per device if it
3986 * is less than 64M per device beyond reshape_progress.
3987 * But always do a multiple of 'blocks'
3988 * FIXME this is too big - it takes to long to complete
3991 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3992 reshape
->after
.data_disks
);
3993 target
/= reshape
->backup_blocks
;
3996 target
*= reshape
->backup_blocks
;
3998 /* For externally managed metadata we always need to suspend IO to
3999 * the area being reshaped so we regularly push suspend_point forward.
4000 * For native metadata we only need the suspend if we are going to do
4004 if ((need_backup
> info
->reshape_progress
||
4005 info
->array
.major_version
< 0) &&
4006 *suspend_point
< info
->reshape_progress
+ target
) {
4007 if (need_backup
< *suspend_point
+ 2 * target
)
4008 *suspend_point
= need_backup
;
4009 else if (*suspend_point
+ 2 * target
< array_size
)
4010 *suspend_point
+= 2 * target
;
4012 *suspend_point
= array_size
;
4013 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4014 if (max_progress
> *suspend_point
)
4015 max_progress
= *suspend_point
;
4018 if (info
->array
.major_version
>= 0) {
4019 /* Only need to suspend when about to backup */
4020 if (info
->reshape_progress
< need_backup
* 2 &&
4021 *suspend_point
> 0) {
4023 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4024 sysfs_set_num(info
, NULL
, "suspend_hi",
4028 /* Need to suspend continually */
4029 if (info
->reshape_progress
< *suspend_point
)
4030 *suspend_point
= info
->reshape_progress
;
4031 if (*suspend_point
+ target
< info
->reshape_progress
)
4032 /* No need to move suspend region yet */;
4034 if (*suspend_point
>= 2 * target
)
4035 *suspend_point
-= 2 * target
;
4038 sysfs_set_num(info
, NULL
, "suspend_lo",
4041 if (max_progress
< *suspend_point
)
4042 max_progress
= *suspend_point
;
4046 /* now set sync_max to allow that progress. sync_max, like
4047 * sync_completed is a count of sectors written per device, so
4048 * we find the difference between max_progress and the start point,
4049 * and divide that by after.data_disks to get a sync_max
4051 * At the same time we convert wait_point to a similar number
4052 * for comparing against sync_completed.
4054 /* scale down max_progress to per_disk */
4055 max_progress
/= reshape
->after
.data_disks
;
4057 * Round to chunk size as some kernels give an erroneously
4060 max_progress
/= info
->new_chunk
/512;
4061 max_progress
*= info
->new_chunk
/512;
4062 /* And round to old chunk size as the kernel wants that */
4063 max_progress
/= info
->array
.chunk_size
/512;
4064 max_progress
*= info
->array
.chunk_size
/512;
4065 /* Limit progress to the whole device */
4066 if (max_progress
> info
->component_size
)
4067 max_progress
= info
->component_size
;
4068 wait_point
/= reshape
->after
.data_disks
;
4070 /* switch from 'device offset' to 'processed block count' */
4071 max_progress
= info
->component_size
- max_progress
;
4072 wait_point
= info
->component_size
- wait_point
;
4076 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4078 /* Now wait. If we have already reached the point that we were
4079 * asked to wait to, don't wait at all, else wait for any change.
4080 * We need to select on 'sync_completed' as that is the place that
4081 * notifications happen, but we are really interested in
4082 * 'reshape_position'
4084 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4086 goto check_progress
;
4088 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4089 goto check_progress
;
4091 while (completed
< max_progress
&& completed
< wait_point
) {
4092 /* Check that sync_action is still 'reshape' to avoid
4093 * waiting forever on a dead array
4095 char action
[SYSFS_MAX_BUF_SIZE
];
4096 if (sysfs_get_str(info
, NULL
, "sync_action", action
, sizeof(action
)) <= 0 ||
4097 strncmp(action
, "reshape", 7) != 0)
4099 /* Some kernels reset 'sync_completed' to zero
4100 * before setting 'sync_action' to 'idle'.
4101 * So we need these extra tests.
4103 if (completed
== 0 && advancing
&&
4104 strncmp(action
, "idle", 4) == 0 &&
4105 info
->reshape_progress
> 0)
4107 if (completed
== 0 && !advancing
&&
4108 strncmp(action
, "idle", 4) == 0 &&
4109 info
->reshape_progress
<
4110 (info
->component_size
* reshape
->after
.data_disks
))
4112 sysfs_wait(fd
, NULL
);
4113 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4114 goto check_progress
;
4116 /* Some kernels reset 'sync_completed' to zero,
4117 * we need to have real point we are in md.
4118 * So in that case, read 'reshape_position' from sysfs.
4120 if (completed
== 0) {
4121 unsigned long long reshapep
;
4122 char action
[SYSFS_MAX_BUF_SIZE
];
4123 if (sysfs_get_str(info
, NULL
, "sync_action", action
, sizeof(action
)) > 0 &&
4124 strncmp(action
, "idle", 4) == 0 &&
4125 sysfs_get_ll(info
, NULL
,
4126 "reshape_position", &reshapep
) == 0)
4127 *reshape_completed
= reshapep
;
4129 /* some kernels can give an incorrectly high
4130 * 'completed' number, so round down */
4131 completed
/= (info
->new_chunk
/512);
4132 completed
*= (info
->new_chunk
/512);
4133 /* Convert 'completed' back in to a 'progress' number */
4134 completed
*= reshape
->after
.data_disks
;
4136 completed
= (info
->component_size
4137 * reshape
->after
.data_disks
4139 *reshape_completed
= completed
;
4144 /* We return the need_backup flag. Caller will decide
4145 * how much - a multiple of ->backup_blocks up to *suspend_point
4148 return need_backup
> info
->reshape_progress
;
4150 return need_backup
>= info
->reshape_progress
;
4153 /* if we couldn't read a number from sync_completed, then
4154 * either the reshape did complete, or it aborted.
4155 * We can tell which by checking for 'none' in reshape_position.
4156 * If it did abort, then it might immediately restart if it
4157 * it was just a device failure that leaves us degraded but
4160 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0 ||
4161 str_is_none(buf
) == false) {
4162 /* The abort might only be temporary. Wait up to 10
4163 * seconds for fd to contain a valid number again.
4167 unsigned long long new_sync_max
;
4168 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4169 if (sysfs_wait(fd
, &wait
) != 1)
4171 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4173 /* all good again */
4175 /* If "sync_max" is no longer max_progress
4176 * we need to freeze things
4178 sysfs_get_ll(info
, NULL
, "sync_max",
4180 *frozen
= (new_sync_max
!= max_progress
);
4182 case -2: /* read error - abort */
4189 return rv
; /* abort */
4191 /* Maybe racing with array shutdown - check state */
4194 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4196 strncmp(buf
, "inactive", 8) == 0 ||
4197 strncmp(buf
, "clear",5) == 0)
4198 return -2; /* abort */
4199 return -1; /* complete */
4203 /* FIXME return status is never checked */
4204 static int grow_backup(struct mdinfo
*sra
,
4205 unsigned long long offset
, /* per device */
4206 unsigned long stripes
, /* per device, in old chunks */
4207 int *sources
, unsigned long long *offsets
,
4208 int disks
, int chunk
, int level
, int layout
,
4209 int dests
, int *destfd
, unsigned long long *destoffsets
,
4210 int part
, int *degraded
,
4213 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4214 * to storage 'destfd' (offset 'destoffsets'), after first
4215 * suspending IO. Then allow resync to continue
4216 * over the suspended section.
4217 * Use part 'part' of the backup-super-block.
4222 unsigned long long ll
;
4224 //printf("offset %llu\n", offset);
4230 /* Check that array hasn't become degraded, else we might backup the wrong data */
4231 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4232 return -1; /* FIXME this error is ignored */
4233 new_degraded
= (int)ll
;
4234 if (new_degraded
!= *degraded
) {
4235 /* check each device to ensure it is still working */
4237 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4238 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4240 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4241 char sbuf
[SYSFS_MAX_BUF_SIZE
];
4243 if (sysfs_get_str(sra
, sd
, "state",
4244 sbuf
, sizeof(sbuf
)) < 0 ||
4245 strstr(sbuf
, "faulty") ||
4246 strstr(sbuf
, "in_sync") == NULL
) {
4247 /* this device is dead */
4248 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4249 if (sd
->disk
.raid_disk
>= 0 &&
4250 sources
[sd
->disk
.raid_disk
] >= 0) {
4251 close(sources
[sd
->disk
.raid_disk
]);
4252 sources
[sd
->disk
.raid_disk
] = -1;
4257 *degraded
= new_degraded
;
4260 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4261 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4263 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4264 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4267 bsb
.magic
[15] = '2';
4268 for (i
= 0; i
< dests
; i
++)
4270 lseek64(destfd
[i
], destoffsets
[i
] +
4271 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4273 lseek64(destfd
[i
], destoffsets
[i
], 0);
4275 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4276 dests
, destfd
, offset
* 512 * odata
,
4277 stripes
* chunk
* odata
, buf
);
4281 bsb
.mtime
= __cpu_to_le64(time(0));
4282 for (i
= 0; i
< dests
; i
++) {
4283 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4285 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4286 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4287 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4288 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4289 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4292 if ((unsigned long long)lseek64(destfd
[i
],
4293 destoffsets
[i
] - 4096, 0) !=
4294 destoffsets
[i
] - 4096)
4296 if (write(destfd
[i
], &bsb
, 512) != 512)
4298 if (destoffsets
[i
] > 4096) {
4299 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4300 destoffsets
[i
]+stripes
*chunk
*odata
)
4302 if (write(destfd
[i
], &bsb
, 512) != 512)
4312 /* in 2.6.30, the value reported by sync_completed can be
4313 * less that it should be by one stripe.
4314 * This only happens when reshape hits sync_max and pauses.
4315 * So allow wait_backup to either extent sync_max further
4316 * than strictly necessary, or return before the
4317 * sync has got quite as far as we would really like.
4318 * This is what 'blocks2' is for.
4319 * The various caller give appropriate values so that
4322 /* FIXME return value is often ignored */
4323 static int forget_backup(int dests
, int *destfd
,
4324 unsigned long long *destoffsets
,
4328 * Erase backup 'part' (which is 0 or 1)
4334 bsb
.arraystart2
= __cpu_to_le64(0);
4335 bsb
.length2
= __cpu_to_le64(0);
4337 bsb
.arraystart
= __cpu_to_le64(0);
4338 bsb
.length
= __cpu_to_le64(0);
4340 bsb
.mtime
= __cpu_to_le64(time(0));
4342 for (i
= 0; i
< dests
; i
++) {
4343 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4344 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4345 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4346 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4347 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4348 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4349 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4350 destoffsets
[i
]-4096)
4352 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4359 static void fail(char *msg
)
4362 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4363 rv
|= (write(2, "\n", 1) != 1);
4367 static char *abuf
, *bbuf
;
4368 static unsigned long long abuflen
;
4369 static void validate(int afd
, int bfd
, unsigned long long offset
)
4371 /* check that the data in the backup against the array.
4372 * This is only used for regression testing and should not
4373 * be used while the array is active
4377 lseek64(bfd
, offset
- 4096, 0);
4378 if (read(bfd
, &bsb2
, 512) != 512)
4379 fail("cannot read bsb");
4380 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4381 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4382 fail("first csum bad");
4383 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4384 fail("magic is bad");
4385 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4386 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4387 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4388 fail("second csum bad");
4390 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4391 fail("devstart is wrong");
4394 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4396 if (abuflen
< len
) {
4400 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4401 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4403 /* just stop validating on mem-alloc failure */
4408 lseek64(bfd
, offset
, 0);
4409 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4410 //printf("len %llu\n", len);
4411 fail("read first backup failed");
4413 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4414 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4415 fail("read first from array failed");
4416 if (memcmp(bbuf
, abuf
, len
) != 0) {
4419 printf("offset=%llu len=%llu\n",
4420 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4421 for (i
=0; i
<len
; i
++)
4422 if (bbuf
[i
] != abuf
[i
]) {
4423 printf("first diff byte %d\n", i
);
4427 fail("data1 compare failed");
4431 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4433 if (abuflen
< len
) {
4437 abuf
= xmalloc(abuflen
);
4438 bbuf
= xmalloc(abuflen
);
4441 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4442 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4443 fail("read second backup failed");
4444 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4445 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4446 fail("read second from array failed");
4447 if (memcmp(bbuf
, abuf
, len
) != 0)
4448 fail("data2 compare failed");
4452 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4453 struct supertype
*st
, unsigned long blocks
,
4454 int *fds
, unsigned long long *offsets
,
4455 int dests
, int *destfd
, unsigned long long *destoffsets
)
4457 /* Monitor a reshape where backup is being performed using
4458 * 'native' mechanism - either to a backup file, or
4459 * to some space in a spare.
4463 unsigned long long speed
;
4464 unsigned long long suspend_point
, array_size
;
4465 unsigned long long backup_point
, wait_point
;
4466 unsigned long long reshape_completed
;
4468 int increasing
= reshape
->after
.data_disks
>=
4469 reshape
->before
.data_disks
;
4470 int part
= 0; /* The next part of the backup area to fill. It
4471 * may already be full, so we need to check */
4472 int level
= reshape
->level
;
4473 int layout
= reshape
->before
.layout
;
4474 int data
= reshape
->before
.data_disks
;
4475 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4476 int chunk
= sra
->array
.chunk_size
;
4478 unsigned long stripes
;
4482 /* set up the backup-super-block. This requires the
4483 * uuid from the array.
4485 /* Find a superblock */
4486 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4490 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4492 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4493 devfd
= dev_open(dn
, O_RDONLY
);
4496 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4502 pr_err("Cannot find a superblock\n");
4506 memset(&bsb
, 0, 512);
4507 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4508 st
->ss
->uuid_from_super(st
, uuid
);
4509 memcpy(bsb
.set_uuid
, uuid
, 16);
4510 bsb
.mtime
= __cpu_to_le64(time(0));
4511 bsb
.devstart2
= blocks
;
4513 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4514 reshape
->before
.data_disks
;
4516 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4517 /* Don't start the 'reshape' */
4519 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4520 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4521 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4525 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4526 backup_point
= sra
->reshape_progress
;
4529 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4530 backup_point
= reshape
->backup_blocks
;
4531 suspend_point
= array_size
;
4537 /* Want to return as soon the oldest backup slot can
4538 * be released as that allows us to start backing up
4539 * some more, providing suspend_point has been
4540 * advanced, which it should have.
4543 wait_point
= array_size
;
4544 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4545 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4546 __le64_to_cpu(bsb
.length
));
4547 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4548 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4549 __le64_to_cpu(bsb
.length2
));
4552 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4553 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4554 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4555 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4558 reshape_completed
= sra
->reshape_progress
;
4559 rv
= progress_reshape(sra
, reshape
,
4560 backup_point
, wait_point
,
4561 &suspend_point
, &reshape_completed
,
4563 /* external metadata would need to ping_monitor here */
4564 sra
->reshape_progress
= reshape_completed
;
4566 /* Clear any backup region that is before 'here' */
4568 if (__le64_to_cpu(bsb
.length
) > 0 &&
4569 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4570 __le64_to_cpu(bsb
.length
)))
4571 forget_backup(dests
, destfd
,
4573 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4574 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4575 __le64_to_cpu(bsb
.length2
)))
4576 forget_backup(dests
, destfd
,
4579 if (__le64_to_cpu(bsb
.length
) > 0 &&
4580 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4581 forget_backup(dests
, destfd
,
4583 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4584 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4585 forget_backup(dests
, destfd
,
4595 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4596 /* No longer need to monitor this reshape */
4597 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4603 unsigned long long offset
;
4604 unsigned long actual_stripes
;
4605 /* Need to backup some data.
4606 * If 'part' is not used and the desired
4607 * backup size is suspended, do a backup,
4608 * then consider the next part.
4610 /* Check that 'part' is unused */
4611 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4613 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4616 offset
= backup_point
/ data
;
4617 actual_stripes
= stripes
;
4619 if (offset
+ actual_stripes
* (chunk
/512) >
4620 sra
->component_size
)
4621 actual_stripes
= ((sra
->component_size
- offset
)
4623 if (offset
+ actual_stripes
* (chunk
/512) >
4627 if (offset
< actual_stripes
* (chunk
/512))
4628 actual_stripes
= offset
/ (chunk
/512);
4629 offset
-= actual_stripes
* (chunk
/512);
4630 if (offset
< suspend_point
/data
)
4633 if (actual_stripes
== 0)
4635 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4636 disks
, chunk
, level
, layout
, dests
, destfd
,
4637 destoffsets
, part
, °raded
, buf
);
4638 validate(afd
, destfd
[0], destoffsets
[0]);
4639 /* record where 'part' is up to */
4642 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4644 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4648 /* FIXME maybe call progress_reshape one more time instead */
4649 /* remove any remaining suspension */
4650 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4651 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4652 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4653 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4655 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4656 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4662 * If any spare contains md_back_data-1 which is recent wrt mtime,
4663 * write that data into the array and update the super blocks with
4664 * the new reshape_progress
4666 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4667 int cnt
, char *backup_file
, int verbose
)
4671 unsigned long long *offsets
;
4672 unsigned long long nstripe
, ostripe
;
4675 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4676 if (info
->array
.level
== 6)
4677 odata
--; /* number of data disks */
4678 ndata
= info
->array
.raid_disks
- 1;
4679 if (info
->new_level
== 6)
4682 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4684 if (info
->delta_disks
<= 0)
4685 /* Didn't grow, so the backup file must have
4689 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4690 struct mdinfo dinfo
;
4693 char *devname
, namebuf
[20];
4694 unsigned long long lo
, hi
;
4696 /* This was a spare and may have some saved data on it.
4697 * Load the superblock, find and load the
4698 * backup_super_block.
4699 * If either fail, go on to next device.
4700 * If the backup contains no new info, just return
4701 * else restore data and update all superblocks
4703 if (i
== old_disks
-1) {
4704 fd
= open(backup_file
, O_RDONLY
);
4706 pr_err("backup file %s inaccessible: %s\n",
4707 backup_file
, strerror(errno
));
4710 devname
= backup_file
;
4715 if (st
->ss
->load_super(st
, fd
, NULL
))
4718 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4719 st
->ss
->free_super(st
);
4722 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4724 pr_err("Cannot seek on device %d\n", i
);
4725 continue; /* Cannot seek */
4727 sprintf(namebuf
, "device-%d", i
);
4730 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4732 pr_err("Cannot read from %s\n", devname
);
4733 continue; /* Cannot read */
4735 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4736 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4738 pr_err("No backup metadata on %s\n", devname
);
4741 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4743 pr_err("Bad backup-metadata checksum on %s\n",
4745 continue; /* bad checksum */
4747 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4748 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4750 pr_err("Bad backup-metadata checksum2 on %s\n",
4752 continue; /* Bad second checksum */
4754 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4756 pr_err("Wrong uuid on backup-metadata on %s\n",
4758 continue; /* Wrong uuid */
4762 * array utime and backup-mtime should be updated at
4763 * much the same time, but it seems that sometimes
4764 * they aren't... So allow considerable flexability in
4765 * matching, and allow this test to be overridden by
4766 * an environment variable.
4768 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4769 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4770 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4771 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4772 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4773 (unsigned long)info
->array
.utime
);
4775 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4776 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4777 continue; /* time stamp is too bad */
4781 if (bsb
.magic
[15] == '1') {
4782 if (bsb
.length
== 0)
4784 if (info
->delta_disks
>= 0) {
4785 /* reshape_progress is increasing */
4786 if (__le64_to_cpu(bsb
.arraystart
)
4787 + __le64_to_cpu(bsb
.length
)
4788 < info
->reshape_progress
) {
4791 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4792 continue; /* No new data here */
4795 /* reshape_progress is decreasing */
4796 if (__le64_to_cpu(bsb
.arraystart
) >=
4797 info
->reshape_progress
)
4798 goto nonew
; /* No new data here */
4801 if (bsb
.length
== 0 && bsb
.length2
== 0)
4803 if (info
->delta_disks
>= 0) {
4804 /* reshape_progress is increasing */
4805 if ((__le64_to_cpu(bsb
.arraystart
)
4806 + __le64_to_cpu(bsb
.length
)
4807 < info
->reshape_progress
) &&
4808 (__le64_to_cpu(bsb
.arraystart2
)
4809 + __le64_to_cpu(bsb
.length2
)
4810 < info
->reshape_progress
))
4811 goto nonew
; /* No new data here */
4813 /* reshape_progress is decreasing */
4814 if (__le64_to_cpu(bsb
.arraystart
) >=
4815 info
->reshape_progress
&&
4816 __le64_to_cpu(bsb
.arraystart2
) >=
4817 info
->reshape_progress
)
4818 goto nonew
; /* No new data here */
4821 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4824 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4826 continue; /* Cannot seek */
4828 /* There should be a duplicate backup superblock 4k before here */
4829 if (lseek64(fd
, -4096, 1) < 0 ||
4830 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4831 goto second_fail
; /* Cannot find leading superblock */
4832 if (bsb
.magic
[15] == '1')
4833 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4835 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4836 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4837 goto second_fail
; /* Cannot find leading superblock */
4839 /* Now need the data offsets for all devices. */
4840 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4841 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4844 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4845 /* FIXME should be this be an error */
4847 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4848 st
->ss
->free_super(st
);
4849 offsets
[j
] = dinfo
.data_offset
* 512;
4851 printf("%s: restoring critical section\n", Name
);
4853 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4854 info
->new_chunk
, info
->new_level
,
4855 info
->new_layout
, fd
,
4856 __le64_to_cpu(bsb
.devstart
)*512,
4857 __le64_to_cpu(bsb
.arraystart
)*512,
4858 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4859 /* didn't succeed, so giveup */
4861 pr_err("Error restoring backup from %s\n",
4867 if (bsb
.magic
[15] == '2' &&
4868 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4869 info
->new_chunk
, info
->new_level
,
4870 info
->new_layout
, fd
,
4871 __le64_to_cpu(bsb
.devstart
)*512 +
4872 __le64_to_cpu(bsb
.devstart2
)*512,
4873 __le64_to_cpu(bsb
.arraystart2
)*512,
4874 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4875 /* didn't succeed, so giveup */
4877 pr_err("Error restoring second backup from %s\n",
4885 /* Ok, so the data is restored. Let's update those superblocks. */
4889 lo
= __le64_to_cpu(bsb
.arraystart
);
4890 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4892 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4893 unsigned long long lo1
, hi1
;
4894 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4895 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4899 } else if (lo
< lo1
)
4904 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4905 info
->reshape_progress
> hi
))
4906 /* backup does not affect reshape_progress*/ ;
4907 else if (info
->delta_disks
>= 0) {
4908 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4909 __le64_to_cpu(bsb
.length
);
4910 if (bsb
.magic
[15] == '2') {
4911 unsigned long long p2
;
4913 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4914 __le64_to_cpu(bsb
.length2
);
4915 if (p2
> info
->reshape_progress
)
4916 info
->reshape_progress
= p2
;
4919 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4920 if (bsb
.magic
[15] == '2') {
4921 unsigned long long p2
;
4923 p2
= __le64_to_cpu(bsb
.arraystart2
);
4924 if (p2
< info
->reshape_progress
)
4925 info
->reshape_progress
= p2
;
4928 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4931 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4933 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4934 dinfo
.reshape_progress
= info
->reshape_progress
;
4935 st
->ss
->update_super(st
, &dinfo
,
4936 UOPT_SPEC__RESHAPE_PROGRESS
,
4938 st
->ss
->store_super(st
, fdlist
[j
]);
4939 st
->ss
->free_super(st
);
4943 /* Didn't find any backup data, try to see if any
4946 if (info
->delta_disks
< 0) {
4947 /* When shrinking, the critical section is at the end.
4948 * So see if we are before the critical section.
4950 unsigned long long first_block
;
4951 nstripe
= ostripe
= 0;
4953 while (ostripe
>= nstripe
) {
4954 ostripe
+= info
->array
.chunk_size
/ 512;
4955 first_block
= ostripe
* odata
;
4956 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4957 (info
->new_chunk
/512);
4960 if (info
->reshape_progress
>= first_block
)
4963 if (info
->delta_disks
> 0) {
4964 /* See if we are beyond the critical section. */
4965 unsigned long long last_block
;
4966 nstripe
= ostripe
= 0;
4968 while (nstripe
>= ostripe
) {
4969 nstripe
+= info
->new_chunk
/ 512;
4970 last_block
= nstripe
* ndata
;
4971 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4972 (info
->array
.chunk_size
/512);
4975 if (info
->reshape_progress
>= last_block
)
4978 /* needed to recover critical section! */
4980 pr_err("Failed to find backup of critical section\n");
4984 int Grow_continue_command(char *devname
, int fd
,
4985 char *backup_file
, int verbose
)
4988 struct supertype
*st
= NULL
;
4989 struct mdinfo
*content
= NULL
;
4990 struct mdinfo array
;
4991 char *subarray
= NULL
;
4992 struct mdinfo
*cc
= NULL
;
4993 struct mdstat_ent
*mdstat
= NULL
;
4997 dprintf("Grow continue from command line called for %s\n", devname
);
4999 st
= super_by_fd(fd
, &subarray
);
5000 if (!st
|| !st
->ss
) {
5001 pr_err("Unable to determine metadata format for %s\n", devname
);
5004 dprintf("Grow continue is run for ");
5005 if (st
->ss
->external
== 0) {
5008 dprintf_cont("native array (%s)\n", devname
);
5009 if (md_get_array_info(fd
, &array
.array
) < 0) {
5010 pr_err("%s is not an active md array - aborting\n",
5013 goto Grow_continue_command_exit
;
5016 sysfs_init(content
, fd
, NULL
);
5017 /* Need to load a superblock.
5018 * FIXME we should really get what we need from
5022 for (d
= 0; d
< MAX_DISKS
; d
++) {
5023 mdu_disk_info_t disk
;
5027 if (md_get_disk_info(fd
, &disk
) < 0)
5029 if (disk
.major
== 0 && disk
.minor
== 0)
5031 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5033 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5036 fd2
= dev_open(dv
, O_RDONLY
);
5039 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5045 if (d
== MAX_DISKS
) {
5046 pr_err("Unable to load metadata for %s\n",
5049 goto Grow_continue_command_exit
;
5051 st
->ss
->getinfo_super(st
, content
, NULL
);
5052 if (!content
->reshape_active
)
5053 sleep_for(3, 0, true);
5056 } while (cnt
-- > 0);
5061 dprintf_cont("subarray (%s)\n", subarray
);
5062 container
= st
->container_devnm
;
5063 cfd
= open_dev_excl(st
->container_devnm
);
5065 container
= st
->devnm
;
5067 cfd
= open_dev_excl(st
->devnm
);
5068 dprintf_cont("container (%s)\n", container
);
5072 pr_err("Unable to open container for %s\n", devname
);
5074 goto Grow_continue_command_exit
;
5077 /* find in container array under reshape
5079 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5081 pr_err("Cannot read superblock for %s\n", devname
);
5083 goto Grow_continue_command_exit
;
5086 cc
= st
->ss
->container_content(st
, subarray
);
5087 for (content
= cc
; content
; content
= content
->next
) {
5089 int allow_reshape
= 1;
5091 if (content
->reshape_active
== 0)
5093 /* The decision about array or container wide
5094 * reshape is taken in Grow_continue based
5095 * content->reshape_active state, therefore we
5096 * need to check_reshape based on
5097 * reshape_active and subarray name
5099 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5101 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5102 (content
->array
.state
5103 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5106 if (!allow_reshape
) {
5107 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5108 devname
, container
);
5110 goto Grow_continue_command_exit
;
5113 array_name
= strchr(content
->text_version
+1, '/')+1;
5114 mdstat
= mdstat_by_subdev(array_name
, container
);
5117 if (mdstat
->active
== 0) {
5118 pr_err("Skipping inactive array %s.\n",
5120 free_mdstat(mdstat
);
5127 pr_err("Unable to determine reshaped array for %s\n", devname
);
5129 goto Grow_continue_command_exit
;
5131 fd2
= open_dev(mdstat
->devnm
);
5133 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5135 goto Grow_continue_command_exit
;
5138 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5139 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5143 goto Grow_continue_command_exit
;
5148 /* start mdmon in case it is not running
5150 if (!mdmon_running(container
))
5151 start_mdmon(container
);
5152 ping_monitor(container
);
5154 if (mdmon_running(container
) == false) {
5155 pr_err("No mdmon found. Grow cannot continue.\n");
5157 goto Grow_continue_command_exit
;
5161 /* verify that array under reshape is started from
5164 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5166 goto Grow_continue_command_exit
;
5171 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5173 Grow_continue_command_exit
:
5176 st
->ss
->free_super(st
);
5177 free_mdstat(mdstat
);
5184 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5185 char *backup_file
, int forked
, int freeze_reshape
)
5189 if (!info
->reshape_active
)
5192 if (st
->ss
->external
) {
5193 int cfd
= open_dev(st
->container_devnm
);
5198 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5200 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5201 st
, info
, 0, backup_file
, 0,
5202 forked
, 1 | info
->reshape_active
,
5205 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5206 NULL
, INVALID_SECTORS
, backup_file
,
5207 0, forked
, 1 | info
->reshape_active
,
5213 char *make_backup(char *name
)
5215 char *base
= "backup_file-";
5219 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5220 fname
= xmalloc(len
);
5221 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5225 char *locate_backup(char *name
)
5227 char *fl
= make_backup(name
);
5230 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))